btrfs: make lseek and fiemap much more efficient
cover btrfs: make lseek and fiemap much more efficient
From: Filipe Manana <[email protected]>
We often get reports of fiemap and hole/data seeking (lseek) being too slow on btrfs, or even unusable in some cases due to being extremely slow.
Some recent reports for fiemap:
https://lore.kernel.org/linux-btrfs/[email protected]/ https://lore.kernel.org/linux-btrfs/[email protected]/
For lseek (LSF/MM from 2017):
https://lwn.net/Articles/718805/
Basically both are slow due to very high algorithmic complexity which scales badly with the number of extents in a file and the heigth of subvolume and extent b+trees.
Using Pavel's test case (first Link tag for fiemap), which uses files with many 4K extents and holes before and after each extent (kind of a worst case scenario), the speedup is of several orders of magnitude (for the 1G file, from ~225 seconds down to ~0.1 seconds).
Finally the new algorithm for fiemap also ends up solving a bug with the current algorithm. This happens because we are currently relying on extent maps to report extents, which can be merged, and this may cause us to report 2 different extents as a single one that is not shared but one of them is shared (or the other way around). More details on this on patches 9/10 and 10/10.
Patches 1/10 and 2/10 are for lseek, introducing some code that will later be used by fiemap too (patch 10/10). More details in the changelogs.
There are a few more things that can be done to speedup fiemap and lseek, but I'll leave those other optimizations I have in mind for some other time.
Filipe Manana (10):
btrfs: allow hole and data seeking to be interruptible
btrfs: make hole and data seeking a lot more efficient
btrfs: remove check for impossible block start for an extent map at fiemap
btrfs: remove zero length check when entering fiemap
btrfs: properly flush delalloc when entering fiemap
btrfs: allow fiemap to be interruptible
btrfs: rename btrfs_check_shared() to a more descriptive name
btrfs: speedup checking for extent sharedness during fiemap
btrfs: skip unnecessary extent buffer sharedness checks during fiemap
btrfs: make fiemap more efficient and accurate reporting extent sharedness
fs/btrfs/backref.c | 153 ++++++++- fs/btrfs/backref.h | 20 +- fs/btrfs/ctree.h | 22 +- fs/btrfs/extent-tree.c | 10 +- fs/btrfs/extent_io.c | 703 ++++++++++++++++++++++++++++------------- fs/btrfs/file.c | 439 +++++++++++++++++++++++-- fs/btrfs/inode.c | 146 ++------- 7 files changed, 1111 insertions(+), 382 deletions(-)
Signature · 1 line
2.35.1
PATCH 1/10 btrfs: allow hole and data seeking to be interruptible
From: Filipe Manana <[email protected]>
Doing hole or data seeking on a file with a very large number of extents can take a long time, and we have reports of it being too slow (such as at LSFMM from 2017, see the Link below). So make it interruptible.
--- fs/btrfs/file.c | 4 ++++ 1 file changed, 4 insertions(+)
fs/btrfs/file.c +4 -0
@@ -3652,6 +3652,10 @@ static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset, start = em->start + em->len;free_extent_map (em); em =NULL ; +if (fatal_signal_pending (current)) { + ret = -EINTR ; +break ; + }cond_resched (); }free_extent_map (em);
Signature · 1 line
2.35.1
Re: [PATCH 01/10] btrfs: allow hole and data seeking to be interruptible
Quoted 8 lines · On Thu, Sep 01, 2022 at 02:18:21PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
Doing hole or data seeking on a file with a very large number of extents can take a long time, and we have reports of it being too slow (such as at LSFMM from 2017, see the Link below). So make it interruptible.
Signed-off-by Filipe Manana <[email protected]>
Thanks,
Josef
Re: [PATCH 01/10] btrfs: allow hole and data seeking to be interruptible
In reply to [email protected]Quoted 8 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
Doing hole or data seeking on a file with a very large number of extents can take a long time, and we have reports of it being too slow (such as at LSFMM from 2017, see the Link below). So make it interruptible.
Signed-off-by Filipe Manana <[email protected]>
Thanks,
Qu
Quoted 19 lines
--- fs/btrfs/file.c | 4 ++++ 1 file changed, 4 insertions(+)fs/btrfs/file.c +4
-0@@ -3652,6 +3652,10 @@ static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset, start = em->start + em->len;free_extent_map (em); em =NULL ; +if (fatal_signal_pending (current)) { + ret = -EINTR ; +break ; + }cond_resched (); }free_extent_map (em);
PATCH 2/10 btrfs: make hole and data seeking a lot more efficient
In reply to [email protected]From: Filipe Manana <[email protected]>
The current implementation of hole and data seeking for llseek does not scale well in regards to the number of extents and the distance between the start offset and the next hole or extent. This is due to a very high algorithmic complexity. Often we also get reports of btrfs' hole and data seeking (llseek) being too slow, such as at 2017's LSFMM (see the Link tag at the bottom).
In order to better understand it, lets consider the case where the start offset is 0, we are seeking for a hole and the file size is 16G. Between file offset 0 and the first hole in the file there are 100K extents - this is common for large files, specially if we have compression enabled, since the maximum extent size is limited to 128K. The steps take by the main loop of the current algorithm are the following:
1) We start by calling btrfs_get_extent_fiemap(), for file offset 0, which calls btrfs_get_extent(). This will first lookup for an extent map in the inode's extent map tree (a red black tree). If the extent map is not loaded in memory, then it will do a lookup for the corresponding file extent item in the subvolume's b+tree, create an extent map based on the contents of the file extent item and then add the extent map to the extent map tree of the inode;
2) The second iteration calls btrfs_get_extent_fiemap() again, this time with a start offset matching the end offset of the previous extent. Again, btrfs_get_extent() will first search the extent map tree, and if it doesn't find an extent map there, it will again search in the b+tree of the subvolume for a matching file extent item, build an extent map based on the file extent item, and add the extent map to to the extent map tree of the inode;
3) This repeats over and over until we find the first hole (when seeking for holes) or until we find the first extent (when seeking for data).
If there no extent maps loaded in memory for each iteration, then on each iteration we do 1 extent map tree search, 1 b+tree search, plus 1 more extent map tree traversal to insert an extent map - plus we allocate memory for the extent map.
On each iteration we are growing the size of the extent map tree, making each future search slower, and also visiting the same b+tree leaves over and over again - taking into account with the default leaf size of 16K we can fit more than 200 file extent items in a leaf - so we can visit the same b+tree leaf 200+ times, on each visit walking down a path from the root to the leaf.
So it's easy to see that what we have now doesn't scale well. Also, it loads an extent map for every file extent item into memory, which is not efficient - we should add extents maps only when doing IO (writing or reading file data).
This change implements a new algorithm which scales much better, and works like this:
1) We iterate over the subvolume's b+tree, visiting each leaf that has file extent items once and only once;
2) For any file extent items found, that don't represent holes or prealloc extents, it will not search the extent map tree - there's no need at all for that - an extent map is just an in-memory representation of a file extent item;
3) When a hole is found, or a prealloc extent, it will check if there's delalloc for its range. For this it will search for EXTENT_DELALLOC bits in the inode's io tree and check the extent map tree - this is for accounting for unflushed delalloc and for flushed delalloc (the period between running delalloc and ordered extent completion), respectively. This is similar to what the current implementation does when it finds a hole or prealloc extent, but without creating extent maps and adding them to the extent map tree in case they are not loaded in memory;
4) It never allocates extent maps, or adds extent maps to the inode's extent map tree. This not only saves memory and time (from the tree insertions and allocations), but also eliminates the possibility of -ENOMEM due to allocating too many extent maps.
Part of this new code will also be used later for fiemap (which also suffers similar scalability problems).
The following test example can be used to quickly measure the efficiency before and after this patch:
$ cat test-seek-hole.sh
#!/bin/bash DEV=/dev/sdi
MNT=/mnt/sdi
mkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 16G file -> 131073 compressed extents. xfs_io -f -c "pwrite -S 0xab -b 1M 0 16G" $MNT/foobar
# Leave a 1M hole at file offset 15G.
xfs_io -c "fpunch 15G 1M" $MNT/foobar
# Unmount and mount again, so that we can test when there's no # metadata cached in memory.
umount $MNT
mount -o compress=lzo $DEV $MNT
# Test seeking for hole from offset 0 (hole is at offset 15G).
start=$(date +%s%N)
xfs_io -c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata not cached)" echo
start=$(date +%s%N)
xfs_io -c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata cached)" echo
umount $MNT
Before this change:
$ ./test-seek-hole.sh
(...)
Whence Result
HOLE 16106127360
Took 176 milliseconds to seek first hole (metadata not cached) Whence Result
HOLE 16106127360
Took 17 milliseconds to seek first hole (metadata cached)
After this change:
$ ./test-seek-hole.sh
(...)
Whence Result
HOLE 16106127360
Took 43 milliseconds to seek first hole (metadata not cached) Whence Result
HOLE 16106127360
Took 13 milliseconds to seek first hole (metadata cached)
That's about 4X faster when no metadata is cached and about 30% faster when all metadata is cached.
In practice the differences may often be significantly higher, either due to a higher number of extents in a file or because the subvolume's b+tree is much bigger than in this example, where we only have one file.
--- fs/btrfs/file.c | 437 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 406 insertions(+), 31 deletions(-)
fs/btrfs/file.c +406 -31
@@ -3601,22 +3601,281 @@ static long btrfs_fallocate(struct file *file, int mode,return ret; } +/* +* Helper for have_delalloc_in_range(). Find a subrange in a given range that +* has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so have_delalloc_in_range() keeps looping +* while it gets adjacent subranges, and merging them together. +*/ +static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +const u64 len = end +1 - start; +struct extent_map_tree *em_tree = &inode->extent_tree; +struct extent_map *em; +u64 em_end; +u64 delalloc_len; + +/* +* Search the io tree first for EXTENT_DELALLOC. If we find any, it +* means we have delalloc (dirty pages) for which writeback has not +* started yet. +*/ + *delalloc_start_ret = start; + delalloc_len =count_range_bits (&inode->io_tree, delalloc_start_ret, end, + len,EXTENT_DELALLOC ,1 ); +/* +* If delalloc was found then *delalloc_start_ret has a sector size +* aligned value (rounded down). +*/ +if (delalloc_len >0 ) + *delalloc_end_ret = *delalloc_start_ret + delalloc_len -1 ; + +/* +* Now also check if there's any extent map in the range that does not +* map to a hole or prealloc extent. We do this because: +* +* 1) When delalloc is flushed, the file range is locked, we clear the +* EXTENT_DELALLOC bit from the io tree and create an extent map for +* an allocated extent. So we might just have been called after +* delalloc is flushed and before the ordered extent completes and +* inserts the new file extent item in the subvolume's btree; +* +* 2) We may have an extent map created by flushing delalloc for a +* subrange that starts before the subrange we found marked with +* EXTENT_DELALLOC in the io tree. +*/ +read_lock (&em_tree->lock); + em =lookup_extent_mapping (em_tree, start, len); +read_unlock (&em_tree->lock); + +/* extent_map_end() returns a non-inclusive end offset. */ + em_end = em ?extent_map_end (em) :0 ; + +/* +* If we have a hole/prealloc extent map, check the next one if this one +* ends before our range's end. +*/ +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +struct extent_map *next_em; + +read_lock (&em_tree->lock); + next_em =lookup_extent_mapping (em_tree, em_end, len - em_end); +read_unlock (&em_tree->lock); + +free_extent_map (em); + em_end = next_em ?extent_map_end (next_em) :0 ; + em = next_em; + } + +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { +free_extent_map (em); + em =NULL ; + } + +/* +* No extent map or one for a hole or prealloc extent. Use the delalloc +* range we found in the io tree if we have one. +*/ +if (!em) +return (delalloc_len >0 ); + +/* +* We don't have any range as EXTENT_DELALLOC in the io tree, so the +* extent map is the only subrange representing delalloc. +*/ +if (delalloc_len ==0 ) { + *delalloc_start_ret = em->start; + *delalloc_end_ret =min (end, em_end -1 ); +free_extent_map (em); +return true; + } + +/* +* The extent map represents a delalloc range that starts before the +* delalloc range we found in the io tree. +*/ +if (em->start < *delalloc_start_ret) { + *delalloc_start_ret = em->start; +/* +* If the ranges are adjacent, return a combined range. +* Otherwise return the extent map's range. +*/ +if (em_end < *delalloc_start_ret) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; + } + +/* +* The extent map starts after the delalloc range we found in the io +* tree. If it's adjacent, return a combined range, otherwise return +* the range found in the io tree. +*/ +if (*delalloc_end_ret +1 == em->start) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; +} + +/* +* Check if there's delalloc in a given range. +* +* @inode: The inode. +* @start: The start offset of the range. It does not need to be +* sector size aligned. +* @end: The end offset (inclusive value) of the search range. +* It does not need to be sector size aligned. +* @delalloc_start_ret: Output argument, set to the start offset of the +* subrange found with delalloc (may not be sector size +* aligned). +* @delalloc_end_ret: Output argument, set to he end offset (inclusive value) +* of the subrange found with delalloc. +* +* Returns true if a subrange with delalloc is found within the given range, and +* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and +* end offsets of the subrange. +*/ +static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize); +u64 prev_delalloc_end =0 ; +bool ret = false; + +while (cur_offset < end) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =find_delalloc_subrange (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; + +if (prev_delalloc_end ==0 ) { +/* First subrange found. */ + *delalloc_start_ret =max (delalloc_start, start); + *delalloc_end_ret = delalloc_end; + ret = true; + }else if (delalloc_start == prev_delalloc_end +1 ) { +/* Subrange adjacent to the previous one, merge them. */ + *delalloc_end_ret = delalloc_end; + }else { +/* Subrange not adjacent to the previous one, exit. */ +break ; + } + + prev_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; +cond_resched (); + } + +return ret; +} + +/* +* Check if there's a hole or delalloc range in a range representing a hole (or +* prealloc extent) found in the inode's subvolume btree. +* +* @inode: The inode. +* @whence: Seek mode (SEEK_DATA or SEEK_HOLE). +* @start: Start offset of the hole region. It does not need to be sector +* size aligned. +* @end: End offset (inclusive value) of the hole region. It does not +* need to be sector size aligned. +* @start_ret: Return parameter, used to set the start of the subrange in the +* hole that matches the search criteria (seek mode), if such +* subrange is found (return value of the function is true). +* The value returned here may not be sector size aligned. +* +* Returns true if a subrange matching the given seek mode is found, and if one +* is found, it updates @start_ret with the start of the subrange. +*/ +static bool find_desired_extent_in_hole (struct btrfs_inode *inode,int whence, +u64 start,u64 end,u64 *start_ret) +{ +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, + &delalloc_end); +if (delalloc && whence ==SEEK_DATA ) { + *start_ret = delalloc_start; +return true; + } + +if (delalloc && whence ==SEEK_HOLE ) { +/* +* We found delalloc but it starts after out start offset. So we +* have a hole between our start offset and the delalloc start. +*/ +if (start < delalloc_start) { + *start_ret = start; +return true; + } +/* +* Delalloc range starts at our start offset. +* If the delalloc range's length is smaller than our range, +* then it means we have a hole that starts where the delalloc +* subrange ends. +*/ +if (delalloc_end < end) { + *start_ret = delalloc_end +1 ; +return true; + } + +/* There's delalloc for the whole range. */ +return false; + } + +if (!delalloc && whence ==SEEK_HOLE ) { + *start_ret = start; +return true; + } + +/* +* No delalloc in the range and we are seeking for data. The caller has +* to iterate to the next extent item in the subvolume btree. +*/ +return false; +} +static loff_t find_desired_extent (struct btrfs_inode *inode,loff_t offset,int whence) {struct btrfs_fs_info *fs_info = inode->root->fs_info; -struct extent_map *em =NULL ;struct extent_state *cached_state =NULL ; -loff_t i_size = inode->vfs_inode.i_size; +const loff_t i_size =i_size_read (&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct btrfs_path *path; +struct btrfs_key key; +u64 last_extent_end;u64 lockstart;u64 lockend;u64 start; -u64 len; -int ret =0 ; +int ret; +bool found = false;if (i_size ==0 || offset >= i_size)return -ENXIO ; +/* +* Quick path. If the inode has no prealloc extents and its number of +* bytes used matches its i_size, then it can not have holes. +*/ +if (whence ==SEEK_HOLE && + !(inode->flags &BTRFS_INODE_PREALLOC ) && +inode_get_bytes (&inode->vfs_inode) == i_size) +return i_size; +/* * offset can be negative, in this case we start finding DATA/HOLE from * the very start of the file.
@@ -3628,49 +3887,165 @@ static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,if (lockend <= lockstart) lockend = lockstart + fs_info->sectorsize; lockend--; - len = lockend - lockstart +1 ; + + path =btrfs_alloc_path (); +if (!path) +return -ENOMEM ; + path->reada =READA_FORWARD ; + + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = start; + + last_extent_end = lockstart;lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) { + goto out; + }else if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } +while (start < i_size) { - em =btrfs_get_extent_fiemap (inode, start, len); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - em =NULL ; -break ; +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *extent; +u64 extent_end; + +if (path->slots[0 ] >=btrfs_header_nritems (leaf)) { + ret =btrfs_next_leaf (root, path); +if (ret <0 ) + goto out; +else if (ret >0 ) +break ; + + leaf = path->nodes[0 ]; } -if (whence ==SEEK_HOLE && - (em->block_start ==EXTENT_MAP_HOLE || -test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) -break ; -else if (whence ==SEEK_DATA && - (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY )break ; - start = em->start + em->len; -free_extent_map (em); - em =NULL ; + extent_end =btrfs_file_extent_end (path); + +/* +* In the first iteration we may have a slot that points to an +* extent that ends before our start offset, so skip it. +*/ +if (extent_end <= start) { + path->slots[0 ]++; +continue ; + } + +/* We have an implicit hole, NO_HOLES feature is likely set. */ +if (last_extent_end < key.offset) { +u64 search_start = last_extent_end; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + key.offset -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the extent. +*/ + } + + extent =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + +if (btrfs_file_extent_disk_bytenr (leaf, extent) ==0 || +btrfs_file_extent_type (leaf, extent) == +BTRFS_FILE_EXTENT_PREALLOC ) { +/* +* Explicit hole or prealloc extent, search for delalloc. +* A prealloc extent is treated like a hole. +*/ +u64 search_start = key.offset; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + extent_end -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the next +* extent item. +*/ + }else { +/* +* Found a regular or inline extent. +* If we are seeking for data, adjust the start offset +* and stop, we're done. +*/ +if (whence ==SEEK_DATA ) { + start =max_t (u64, key.offset, offset); + found = true; +break ; + } +/* +* Else, we are seeking for a hole, check the next file +* extent item. +*/ + } + + start = extent_end; + last_extent_end = extent_end; + path->slots[0 ]++;if (fatal_signal_pending (current)) { ret = -EINTR ; -break ; + goto out; }cond_resched (); } -free_extent_map (em); + +/* We have an implicit hole from the last extent found up to i_size. */ +if (!found && start < i_size) { + found =find_desired_extent_in_hole (inode, whence, start, + i_size -1 , &start); +if (!found) + start = i_size; + } + +out:unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); -if (ret) { - offset = ret; - }else { -if (whence ==SEEK_DATA && start >= i_size) - offset = -ENXIO ; -else - offset =min_t (loff_t, start, i_size); - } +btrfs_free_path (path); + +if (ret <0 ) +return ret; + +if (whence ==SEEK_DATA && start >= i_size) +return -ENXIO ; -return offset; +return min_t (loff_t, start, i_size); }static loff_t btrfs_file_llseek (struct file *file,loff_t offset,int whence)
Signature · 1 line
2.35.1
Re: [PATCH 02/10] btrfs: make hole and data seeking a lot more efficient
Quoted 239 lines · On Thu, Sep 01, 2022 at 02:18:22PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current implementation of hole and data seeking for llseek does not scale well in regards to the number of extents and the distance between the start offset and the next hole or extent. This is due to a very high algorithmic complexity. Often we also get reports of btrfs' hole and data seeking (llseek) being too slow, such as at 2017's LSFMM (see the Link tag at the bottom).
In order to better understand it, lets consider the case where the start offset is 0, we are seeking for a hole and the file size is 16G. Between file offset 0 and the first hole in the file there are 100K extents - this is common for large files, specially if we have compression enabled, since the maximum extent size is limited to 128K. The steps take by the main loop of the current algorithm are the following:
1) We start by calling
btrfs_get_extent_fiemap(), for file offset 0, which callsbtrfs_get_extent(). This will first lookup for an extent map in the inode's extent map tree (a red black tree). If the extent map is not loaded in memory, then it will do a lookup for the corresponding file extent item in the subvolume's b+tree, create an extent map based on the contents of the file extent item and then add the extent map to the extent map tree of the inode;2) The second iteration calls
btrfs_get_extent_fiemap()again, this time with a start offset matching the end offset of the previous extent. Again,btrfs_get_extent()will first search the extent map tree, and if it doesn't find an extent map there, it will again search in the b+tree of the subvolume for a matching file extent item, build an extent map based on the file extent item, and add the extent map to to the extent map tree of the inode;3) This repeats over and over until we find the first hole (when seeking for holes) or until we find the first extent (when seeking for data).
If there no extent maps loaded in memory for each iteration, then on each iteration we do 1 extent map tree search, 1 b+tree search, plus 1 more extent map tree traversal to insert an extent map - plus we allocate memory for the extent map.
On each iteration we are growing the size of the extent map tree, making each future search slower, and also visiting the same b+tree leaves over and over again - taking into account with the default leaf size of 16K we can fit more than 200 file extent items in a leaf - so we can visit the same b+tree leaf 200+ times, on each visit walking down a path from the root to the leaf.
So it's easy to see that what we have now doesn't scale well. Also, it loads an extent map for every file extent item into memory, which is not efficient - we should add extents maps only when doing IO (writing or reading file data).
This change implements a new algorithm which scales much better, and works like this:
1) We iterate over the subvolume's b+tree, visiting each leaf that has file extent items once and only once;
2) For any file extent items found, that don't represent holes or prealloc extents, it will not search the extent map tree - there's no need at all for that - an extent map is just an in-memory representation of a file extent item;
3) When a hole is found, or a prealloc extent, it will check if there's delalloc for its range. For this it will search for
EXTENT_DELALLOCbits in the inode's io tree and check the extent map tree - this is for accounting for unflushed delalloc and for flushed delalloc (the period between running delalloc and ordered extent completion), respectively. This is similar to what the current implementation does when it finds a hole or prealloc extent, but without creating extent maps and adding them to the extent map tree in case they are not loaded in memory;4) It never allocates extent maps, or adds extent maps to the inode's extent map tree. This not only saves memory and time (from the tree insertions and allocations), but also eliminates the possibility of
-ENOMEMdue to allocating too many extent maps.Part of this new code will also be used later for fiemap (which also suffers similar scalability problems).
The following test example can be used to quickly measure the efficiency before and after this patch:
$ cat test-seek-hole.sh #!/bin/bashDEV=/dev/sdi
MNT=/mnt/sdimkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 16G file -> 131073 compressed extents.
xfs_io-f -c "pwrite -S0xab-b 1M 0 16G" $MNT/foobar# Leave a 1M hole at file offset 15G.
xfs_io-c "fpunch 15G 1M" $MNT/foobar# Unmount and mount again, so that we can test when there's no # metadata cached in memory.
umount $MNT
mount -o compress=lzo $DEV $MNT# Test seeking for hole from offset 0 (hole is at offset 15G).
start=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata not cached)" echostart=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata cached)" echoumount $MNT
Before this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 176 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 17 milliseconds to seek first hole (metadata cached)After this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 43 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 13 milliseconds to seek first hole (metadata cached)That's about 4X faster when no metadata is cached and about 30% faster when all metadata is cached.
In practice the differences may often be significantly higher, either due to a higher number of extents in a file or because the subvolume's b+tree is much bigger than in this example, where we only have one file.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/file.c | 437 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 406 insertions(+), 31 deletions(-)fs/btrfs/file.c +71
-0@@ -3601,22 +3601,281 @@ static long btrfs_fallocate(struct file *file, int mode,return ret; } +/* +* Helper for have_delalloc_in_range(). Find a subrange in a given range that +* has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so have_delalloc_in_range() keeps looping +* while it gets adjacent subranges, and merging them together. +*/ +static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +const u64 len = end +1 - start; +struct extent_map_tree *em_tree = &inode->extent_tree; +struct extent_map *em; +u64 em_end; +u64 delalloc_len; + +/* +* Search the io tree first for EXTENT_DELALLOC. If we find any, it +* means we have delalloc (dirty pages) for which writeback has not +* started yet. +*/ + *delalloc_start_ret = start; + delalloc_len =count_range_bits (&inode->io_tree, delalloc_start_ret, end, + len,EXTENT_DELALLOC ,1 ); +/* +* If delalloc was found then *delalloc_start_ret has a sector size +* aligned value (rounded down). +*/ +if (delalloc_len >0 ) + *delalloc_end_ret = *delalloc_start_ret + delalloc_len -1 ; + +/* +* Now also check if there's any extent map in the range that does not +* map to a hole or prealloc extent. We do this because: +* +* 1) When delalloc is flushed, the file range is locked, we clear the +* EXTENT_DELALLOC bit from the io tree and create an extent map for +* an allocated extent. So we might just have been called after +* delalloc is flushed and before the ordered extent completes and +* inserts the new file extent item in the subvolume's btree; +* +* 2) We may have an extent map created by flushing delalloc for a +* subrange that starts before the subrange we found marked with +* EXTENT_DELALLOC in the io tree. +*/ +read_lock (&em_tree->lock); + em =lookup_extent_mapping (em_tree, start, len); +read_unlock (&em_tree->lock); + +/* extent_map_end() returns a non-inclusive end offset. */ + em_end = em ?extent_map_end (em) :0 ; + +/* +* If we have a hole/prealloc extent map, check the next one if this one +* ends before our range's end. +*/ +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +struct extent_map *next_em; + +read_lock (&em_tree->lock); + next_em =lookup_extent_mapping (em_tree, em_end, len - em_end); +read_unlock (&em_tree->lock); + +free_extent_map (em); + em_end = next_em ?extent_map_end (next_em) :0 ; + em = next_em; + } + +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { +free_extent_map (em);
fs/btrfs/file.c static long btrfs_fallocate(struct file *file, int mode, · line 3684
Quoted 10 lines
fs/btrfs/file.c +10
-0+ em =NULL ; + } + +/* +* No extent map or one for a hole or prealloc extent. Use the delalloc +* range we found in the io tree if we have one. +*/ +if (!em) +return (delalloc_len >0 ); +
You can move this after the lookup, and then remove the if (em && parts above. Then all you need to do is in the second if statement return (delalloc_len > 0); Thanks,
Josef
Re: [PATCH 02/10] btrfs: make hole and data seeking a lot more efficient
Quoted 245 lines · Josef Bacik
Quoted 244 lines · On Thu, Sep 01, 2022 at 02:18:22PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current implementation of hole and data seeking for llseek does not scale well in regards to the number of extents and the distance between the start offset and the next hole or extent. This is due to a very high algorithmic complexity. Often we also get reports of btrfs' hole and data seeking (llseek) being too slow, such as at 2017's LSFMM (see the Link tag at the bottom).
In order to better understand it, lets consider the case where the start offset is 0, we are seeking for a hole and the file size is 16G. Between file offset 0 and the first hole in the file there are 100K extents - this is common for large files, specially if we have compression enabled, since the maximum extent size is limited to 128K. The steps take by the main loop of the current algorithm are the following:
1) We start by calling
btrfs_get_extent_fiemap(), for file offset 0, which callsbtrfs_get_extent(). This will first lookup for an extent map in the inode's extent map tree (a red black tree). If the extent map is not loaded in memory, then it will do a lookup for the corresponding file extent item in the subvolume's b+tree, create an extent map based on the contents of the file extent item and then add the extent map to the extent map tree of the inode;2) The second iteration calls
btrfs_get_extent_fiemap()again, this time with a start offset matching the end offset of the previous extent. Again,btrfs_get_extent()will first search the extent map tree, and if it doesn't find an extent map there, it will again search in the b+tree of the subvolume for a matching file extent item, build an extent map based on the file extent item, and add the extent map to to the extent map tree of the inode;3) This repeats over and over until we find the first hole (when seeking for holes) or until we find the first extent (when seeking for data).
If there no extent maps loaded in memory for each iteration, then on each iteration we do 1 extent map tree search, 1 b+tree search, plus 1 more extent map tree traversal to insert an extent map - plus we allocate memory for the extent map.
On each iteration we are growing the size of the extent map tree, making each future search slower, and also visiting the same b+tree leaves over and over again - taking into account with the default leaf size of 16K we can fit more than 200 file extent items in a leaf - so we can visit the same b+tree leaf 200+ times, on each visit walking down a path from the root to the leaf.
So it's easy to see that what we have now doesn't scale well. Also, it loads an extent map for every file extent item into memory, which is not efficient - we should add extents maps only when doing IO (writing or reading file data).
This change implements a new algorithm which scales much better, and works like this:
1) We iterate over the subvolume's b+tree, visiting each leaf that has file extent items once and only once;
2) For any file extent items found, that don't represent holes or prealloc extents, it will not search the extent map tree - there's no need at all for that - an extent map is just an in-memory representation of a file extent item;
3) When a hole is found, or a prealloc extent, it will check if there's delalloc for its range. For this it will search for
EXTENT_DELALLOCbits in the inode's io tree and check the extent map tree - this is for accounting for unflushed delalloc and for flushed delalloc (the period between running delalloc and ordered extent completion), respectively. This is similar to what the current implementation does when it finds a hole or prealloc extent, but without creating extent maps and adding them to the extent map tree in case they are not loaded in memory;4) It never allocates extent maps, or adds extent maps to the inode's extent map tree. This not only saves memory and time (from the tree insertions and allocations), but also eliminates the possibility of
-ENOMEMdue to allocating too many extent maps.Part of this new code will also be used later for fiemap (which also suffers similar scalability problems).
The following test example can be used to quickly measure the efficiency before and after this patch:
$ cat test-seek-hole.sh #!/bin/bashDEV=/dev/sdi
MNT=/mnt/sdimkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 16G file -> 131073 compressed extents.
xfs_io-f -c "pwrite -S0xab-b 1M 0 16G" $MNT/foobar# Leave a 1M hole at file offset 15G.
xfs_io-c "fpunch 15G 1M" $MNT/foobar# Unmount and mount again, so that we can test when there's no # metadata cached in memory.
umount $MNT
mount -o compress=lzo $DEV $MNT# Test seeking for hole from offset 0 (hole is at offset 15G).
start=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata not cached)" echostart=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata cached)" echoumount $MNT
Before this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 176 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 17 milliseconds to seek first hole (metadata cached)After this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 43 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 13 milliseconds to seek first hole (metadata cached)That's about 4X faster when no metadata is cached and about 30% faster when all metadata is cached.
In practice the differences may often be significantly higher, either due to a higher number of extents in a file or because the subvolume's b+tree is much bigger than in this example, where we only have one file.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/file.c | 437 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 406 insertions(+), 31 deletions(-)fs/btrfs/file.c +76
-0@@ -3601,22 +3601,281 @@ static long btrfs_fallocate(struct file *file, int mode,return ret; } +/* +* Helper for have_delalloc_in_range(). Find a subrange in a given range that +* has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so have_delalloc_in_range() keeps looping +* while it gets adjacent subranges, and merging them together. +*/ +static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +const u64 len = end +1 - start; +struct extent_map_tree *em_tree = &inode->extent_tree; +struct extent_map *em; +u64 em_end; +u64 delalloc_len; + +/* +* Search the io tree first for EXTENT_DELALLOC. If we find any, it +* means we have delalloc (dirty pages) for which writeback has not +* started yet. +*/ + *delalloc_start_ret = start; + delalloc_len =count_range_bits (&inode->io_tree, delalloc_start_ret, end, + len,EXTENT_DELALLOC ,1 ); +/* +* If delalloc was found then *delalloc_start_ret has a sector size +* aligned value (rounded down). +*/ +if (delalloc_len >0 ) + *delalloc_end_ret = *delalloc_start_ret + delalloc_len -1 ; + +/* +* Now also check if there's any extent map in the range that does not +* map to a hole or prealloc extent. We do this because: +* +* 1) When delalloc is flushed, the file range is locked, we clear the +* EXTENT_DELALLOC bit from the io tree and create an extent map for +* an allocated extent. So we might just have been called after +* delalloc is flushed and before the ordered extent completes and +* inserts the new file extent item in the subvolume's btree; +* +* 2) We may have an extent map created by flushing delalloc for a +* subrange that starts before the subrange we found marked with +* EXTENT_DELALLOC in the io tree. +*/ +read_lock (&em_tree->lock); + em =lookup_extent_mapping (em_tree, start, len); +read_unlock (&em_tree->lock); + +/* extent_map_end() returns a non-inclusive end offset. */ + em_end = em ?extent_map_end (em) :0 ; + +/* +* If we have a hole/prealloc extent map, check the next one if this one +* ends before our range's end. +*/ +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +struct extent_map *next_em; + +read_lock (&em_tree->lock); + next_em =lookup_extent_mapping (em_tree, em_end, len - em_end); +read_unlock (&em_tree->lock); + +free_extent_map (em); + em_end = next_em ?extent_map_end (next_em) :0 ; + em = next_em; + } + +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { +free_extent_map (em); + em =NULL ; + } + +/* +* No extent map or one for a hole or prealloc extent. Use the delalloc
Quoted 8 lines
fs/btrfs/file.c static long btrfs_fallocate(struct file *file, int mode, · line 3684
Quoted 5 lines
fs/btrfs/file.c +5
-0+* range we found in the io tree if we have one. +*/ +if (!em) +return (delalloc_len >0 ); +You can move this after the lookup, and then remove the if (em && parts above. Then all you need to do is in the second if statement return (
delalloc_len> 0);
Nop, it won't work by doing just that.
To move that if statement, it would require all the following changes, which to me it doesn't seem to provide any benefit, aesthetically or otherwise:
fs/btrfs/file.c +18 -18
@@ -3649,39 +3649,39 @@ static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end em =lookup_extent_mapping (em_tree, start, len);read_unlock (&em_tree->lock); +if (!em) +return (delalloc_len >0 ); +/* extent_map_end() returns a non-inclusive end offset. */ - em_end = em ?extent_map_end (em) :0 ; + em_end =extent_map_end (em);/* * If we have a hole/prealloc extent map, check the next one if this one * ends before our range's end. */ -if (em && (em->block_start ==EXTENT_MAP_HOLE || -test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +if (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) {struct extent_map *next_em; +free_extent_map (em); + +if (em_end >= end) +return (delalloc_len >0 ); +read_lock (&em_tree->lock); next_em =lookup_extent_mapping (em_tree, em_end, len - em_end);read_unlock (&em_tree->lock); -free_extent_map (em); - em_end = next_em ?extent_map_end (next_em) :0 ; - em = next_em; - } +if (!next_em || next_em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &next_em->flags)) { +free_extent_map (next_em); +return (delalloc_len >0 ); + } -if (em && (em->block_start ==EXTENT_MAP_HOLE || -test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { -free_extent_map (em); - em =NULL ; + em = next_em; + em_end =extent_map_end (em); } -/* -* No extent map or one for a hole or prealloc extent. Use the delalloc -* range we found in the io tree if we have one. -*/ -if (!em) -return (delalloc_len >0 ); -/* * We don't have any range as EXTENT_DELALLOC in the io tree, so the * extent map is the only subrange representing delalloc.
Quoted 3 lines
Thanks,
Josef
Josef Bacik
Reviewed-byRe: [PATCH 02/10] btrfs: make hole and data seeking a lot more efficient
Quoted 255 lines · Filipe Manana
Quoted 245 lines · Josef Bacik
Quoted 244 lines · On Thu, Sep 01, 2022 at 02:18:22PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current implementation of hole and data seeking for llseek does not scale well in regards to the number of extents and the distance between the start offset and the next hole or extent. This is due to a very high algorithmic complexity. Often we also get reports of btrfs' hole and data seeking (llseek) being too slow, such as at 2017's LSFMM (see the Link tag at the bottom).
In order to better understand it, lets consider the case where the start offset is 0, we are seeking for a hole and the file size is 16G. Between file offset 0 and the first hole in the file there are 100K extents - this is common for large files, specially if we have compression enabled, since the maximum extent size is limited to 128K. The steps take by the main loop of the current algorithm are the following:
1) We start by calling
btrfs_get_extent_fiemap(), for file offset 0, which callsbtrfs_get_extent(). This will first lookup for an extent map in the inode's extent map tree (a red black tree). If the extent map is not loaded in memory, then it will do a lookup for the corresponding file extent item in the subvolume's b+tree, create an extent map based on the contents of the file extent item and then add the extent map to the extent map tree of the inode;2) The second iteration calls
btrfs_get_extent_fiemap()again, this time with a start offset matching the end offset of the previous extent. Again,btrfs_get_extent()will first search the extent map tree, and if it doesn't find an extent map there, it will again search in the b+tree of the subvolume for a matching file extent item, build an extent map based on the file extent item, and add the extent map to to the extent map tree of the inode;3) This repeats over and over until we find the first hole (when seeking for holes) or until we find the first extent (when seeking for data).
If there no extent maps loaded in memory for each iteration, then on each iteration we do 1 extent map tree search, 1 b+tree search, plus 1 more extent map tree traversal to insert an extent map - plus we allocate memory for the extent map.
On each iteration we are growing the size of the extent map tree, making each future search slower, and also visiting the same b+tree leaves over and over again - taking into account with the default leaf size of 16K we can fit more than 200 file extent items in a leaf - so we can visit the same b+tree leaf 200+ times, on each visit walking down a path from the root to the leaf.
So it's easy to see that what we have now doesn't scale well. Also, it loads an extent map for every file extent item into memory, which is not efficient - we should add extents maps only when doing IO (writing or reading file data).
This change implements a new algorithm which scales much better, and works like this:
1) We iterate over the subvolume's b+tree, visiting each leaf that has file extent items once and only once;
2) For any file extent items found, that don't represent holes or prealloc extents, it will not search the extent map tree - there's no need at all for that - an extent map is just an in-memory representation of a file extent item;
3) When a hole is found, or a prealloc extent, it will check if there's delalloc for its range. For this it will search for
EXTENT_DELALLOCbits in the inode's io tree and check the extent map tree - this is for accounting for unflushed delalloc and for flushed delalloc (the period between running delalloc and ordered extent completion), respectively. This is similar to what the current implementation does when it finds a hole or prealloc extent, but without creating extent maps and adding them to the extent map tree in case they are not loaded in memory;4) It never allocates extent maps, or adds extent maps to the inode's extent map tree. This not only saves memory and time (from the tree insertions and allocations), but also eliminates the possibility of
-ENOMEMdue to allocating too many extent maps.Part of this new code will also be used later for fiemap (which also suffers similar scalability problems).
The following test example can be used to quickly measure the efficiency before and after this patch:
$ cat test-seek-hole.sh #!/bin/bashDEV=/dev/sdi
MNT=/mnt/sdimkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 16G file -> 131073 compressed extents.
xfs_io-f -c "pwrite -S0xab-b 1M 0 16G" $MNT/foobar# Leave a 1M hole at file offset 15G.
xfs_io-c "fpunch 15G 1M" $MNT/foobar# Unmount and mount again, so that we can test when there's no # metadata cached in memory.
umount $MNT
mount -o compress=lzo $DEV $MNT# Test seeking for hole from offset 0 (hole is at offset 15G).
start=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata not cached)" echostart=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata cached)" echoumount $MNT
Before this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 176 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 17 milliseconds to seek first hole (metadata cached)After this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 43 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 13 milliseconds to seek first hole (metadata cached)That's about 4X faster when no metadata is cached and about 30% faster when all metadata is cached.
In practice the differences may often be significantly higher, either due to a higher number of extents in a file or because the subvolume's b+tree is much bigger than in this example, where we only have one file.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/file.c | 437 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 406 insertions(+), 31 deletions(-)fs/btrfs/file.c +76
-0@@ -3601,22 +3601,281 @@ static long btrfs_fallocate(struct file *file, int mode,return ret; } +/* +* Helper for have_delalloc_in_range(). Find a subrange in a given range that +* has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so have_delalloc_in_range() keeps looping +* while it gets adjacent subranges, and merging them together. +*/ +static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +const u64 len = end +1 - start; +struct extent_map_tree *em_tree = &inode->extent_tree; +struct extent_map *em; +u64 em_end; +u64 delalloc_len; + +/* +* Search the io tree first for EXTENT_DELALLOC. If we find any, it +* means we have delalloc (dirty pages) for which writeback has not +* started yet. +*/ + *delalloc_start_ret = start; + delalloc_len =count_range_bits (&inode->io_tree, delalloc_start_ret, end, + len,EXTENT_DELALLOC ,1 ); +/* +* If delalloc was found then *delalloc_start_ret has a sector size +* aligned value (rounded down). +*/ +if (delalloc_len >0 ) + *delalloc_end_ret = *delalloc_start_ret + delalloc_len -1 ; + +/* +* Now also check if there's any extent map in the range that does not +* map to a hole or prealloc extent. We do this because: +* +* 1) When delalloc is flushed, the file range is locked, we clear the +* EXTENT_DELALLOC bit from the io tree and create an extent map for +* an allocated extent. So we might just have been called after +* delalloc is flushed and before the ordered extent completes and +* inserts the new file extent item in the subvolume's btree; +* +* 2) We may have an extent map created by flushing delalloc for a +* subrange that starts before the subrange we found marked with +* EXTENT_DELALLOC in the io tree. +*/ +read_lock (&em_tree->lock); + em =lookup_extent_mapping (em_tree, start, len); +read_unlock (&em_tree->lock); + +/* extent_map_end() returns a non-inclusive end offset. */ + em_end = em ?extent_map_end (em) :0 ; + +/* +* If we have a hole/prealloc extent map, check the next one if this one +* ends before our range's end. +*/ +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +struct extent_map *next_em; + +read_lock (&em_tree->lock); + next_em =lookup_extent_mapping (em_tree, em_end, len - em_end); +read_unlock (&em_tree->lock); + +free_extent_map (em); + em_end = next_em ?extent_map_end (next_em) :0 ; + em = next_em; + } + +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { +free_extent_map (em); + em =NULL ; + } + +/* +* No extent map or one for a hole or prealloc extent. Use the delalloc Quoted 8 lines
fs/btrfs/file.c static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end · line 3654
Quoted 5 lines
fs/btrfs/file.c +5
-0+* range we found in the io tree if we have one. +*/ +if (!em) +return (delalloc_len >0 ); +You can move this after the lookup, and then remove the if (em && parts above. Then all you need to do is in the second if statement return (
delalloc_len> 0);
Quoted 6 lines
Nop, it won't work by doing just that.
To move that if statement, it would require all the following changes, which to me it doesn't seem to provide any benefit, aesthetically or otherwise:
Ah yeah I see, in that case you can add
Thanks,
Josef
Re: [PATCH 02/10] btrfs: make hole and data seeking a lot more efficient
In reply to [email protected]Quoted 32 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current implementation of hole and data seeking for llseek does not scale well in regards to the number of extents and the distance between the start offset and the next hole or extent. This is due to a very high algorithmic complexity. Often we also get reports of btrfs' hole and data seeking (llseek) being too slow, such as at 2017's LSFMM (see the Link tag at the bottom).
In order to better understand it, lets consider the case where the start offset is 0, we are seeking for a hole and the file size is 16G. Between file offset 0 and the first hole in the file there are 100K extents - this is common for large files, specially if we have compression enabled, since the maximum extent size is limited to 128K. The steps take by the main loop of the current algorithm are the following:
1) We start by calling
btrfs_get_extent_fiemap(), for file offset 0, which callsbtrfs_get_extent(). This will first lookup for an extent map in the inode's extent map tree (a red black tree). If the extent map is not loaded in memory, then it will do a lookup for the corresponding file extent item in the subvolume's b+tree, create an extent map based on the contents of the file extent item and then add the extent map to the extent map tree of the inode;2) The second iteration calls
btrfs_get_extent_fiemap()again, this time with a start offset matching the end offset of the previous extent. Again,btrfs_get_extent()will first search the extent map tree, and if it doesn't find an extent map there, it will again search in the b+tree of the subvolume for a matching file extent item, build an extent map based on the file extent item, and add the extent map to to the extent map tree of the inode;
Quoted 7 lines
3) This repeats over and over until we find the first hole (when seeking for holes) or until we find the first extent (when seeking for data).
If there no extent maps loaded in memory for each iteration, then on each iteration we do 1 extent map tree search, 1 b+tree search, plus 1 more extent map tree traversal to insert an extent map - plus we allocate memory for the extent map.
I'm a little intereted in if we have other workload which are heavily relying on extent map tree search?
If so, would it make sense to load a batch of file extents items into extent map tree in one go?
Another thing not that related to the patchset is, since extent map doesn't really get freed up unless that inode is invicted/truncated, I'm wondering would it be a problem for heavily fragmented files to take up too much memory just for extent map tree?
Would we need a way to drop extent map in the future?
Quoted 24 lines
On each iteration we are growing the size of the extent map tree, making each future search slower, and also visiting the same b+tree leaves over and over again - taking into account with the default leaf size of 16K we can fit more than 200 file extent items in a leaf - so we can visit the same b+tree leaf 200+ times, on each visit walking down a path from the root to the leaf.
So it's easy to see that what we have now doesn't scale well. Also, it loads an extent map for every file extent item into memory, which is not efficient - we should add extents maps only when doing IO (writing or reading file data).
This change implements a new algorithm which scales much better, and works like this:
1) We iterate over the subvolume's b+tree, visiting each leaf that has
file extent items once and only once;2) For any file extent items found, that don't represent holes or prealloc extents, it will not search the extent map tree - there's no need at all for that - an extent map is just an in-memory representation of a file extent item;
Quoted 9 lines
3) When a hole is found, or a prealloc extent, it will check if there's delalloc for its range. For this it will search for
EXTENT_DELALLOCbits in the inode's io tree and check the extent map tree - this is for accounting for unflushed delalloc and for flushed delalloc (the period between running delalloc and ordered extent completion), respectively. This is similar to what the current implementation does when it finds a hole or prealloc extent, but without creating extent maps and adding them to the extent map tree in case they are not loaded in memory;
Would it be possible that, before we starting the subvolume tree search, just run all delalloc of that target inode and prevent new writes, so we can forget about the delalloc situation completely?
Quoted 365 lines
4) It never allocates extent maps, or adds extent maps to the inode's extent map tree. This not only saves memory and time (from the tree insertions and allocations), but also eliminates the possibility of
-ENOMEMdue to allocating too many extent maps.Part of this new code will also be used later for fiemap (which also suffers similar scalability problems).
The following test example can be used to quickly measure the efficiency before and after this patch:
$ cat test-seek-hole.sh #!/bin/bashDEV=/dev/sdi
MNT=/mnt/sdimkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 16G file -> 131073 compressed extents.
xfs_io-f -c "pwrite -S0xab-b 1M 0 16G" $MNT/foobar# Leave a 1M hole at file offset 15G.
xfs_io-c "fpunch 15G 1M" $MNT/foobar# Unmount and mount again, so that we can test when there's no # metadata cached in memory.
umount $MNT
mount -o compress=lzo $DEV $MNT# Test seeking for hole from offset 0 (hole is at offset 15G).
start=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata not cached)" echostart=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata cached)" echoumount $MNT
Before this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 176 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 17 milliseconds to seek first hole (metadata cached)After this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 43 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 13 milliseconds to seek first hole (metadata cached)That's about 4X faster when no metadata is cached and about 30% faster when all metadata is cached.
In practice the differences may often be significantly higher, either due to a higher number of extents in a file or because the subvolume's b+tree is much bigger than in this example, where we only have one file.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/file.c | 437 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 406 insertions(+), 31 deletions(-)fs/btrfs/file.c +254
-4@@ -3601,22 +3601,281 @@ static long btrfs_fallocate(struct file *file, int mode,return ret; } +/* +* Helper for have_delalloc_in_range(). Find a subrange in a given range that +* has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so have_delalloc_in_range() keeps looping +* while it gets adjacent subranges, and merging them together. +*/ +static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +const u64 len = end +1 - start; +struct extent_map_tree *em_tree = &inode->extent_tree; +struct extent_map *em; +u64 em_end; +u64 delalloc_len; + +/* +* Search the io tree first for EXTENT_DELALLOC. If we find any, it +* means we have delalloc (dirty pages) for which writeback has not +* started yet. +*/ + *delalloc_start_ret = start; + delalloc_len =count_range_bits (&inode->io_tree, delalloc_start_ret, end, + len,EXTENT_DELALLOC ,1 ); +/* +* If delalloc was found then *delalloc_start_ret has a sector size +* aligned value (rounded down). +*/ +if (delalloc_len >0 ) + *delalloc_end_ret = *delalloc_start_ret + delalloc_len -1 ; + +/* +* Now also check if there's any extent map in the range that does not +* map to a hole or prealloc extent. We do this because: +* +* 1) When delalloc is flushed, the file range is locked, we clear the +* EXTENT_DELALLOC bit from the io tree and create an extent map for +* an allocated extent. So we might just have been called after +* delalloc is flushed and before the ordered extent completes and +* inserts the new file extent item in the subvolume's btree; +* +* 2) We may have an extent map created by flushing delalloc for a +* subrange that starts before the subrange we found marked with +* EXTENT_DELALLOC in the io tree. +*/ +read_lock (&em_tree->lock); + em =lookup_extent_mapping (em_tree, start, len); +read_unlock (&em_tree->lock); + +/* extent_map_end() returns a non-inclusive end offset. */ + em_end = em ?extent_map_end (em) :0 ; + +/* +* If we have a hole/prealloc extent map, check the next one if this one +* ends before our range's end. +*/ +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +struct extent_map *next_em; + +read_lock (&em_tree->lock); + next_em =lookup_extent_mapping (em_tree, em_end, len - em_end); +read_unlock (&em_tree->lock); + +free_extent_map (em); + em_end = next_em ?extent_map_end (next_em) :0 ; + em = next_em; + } + +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { +free_extent_map (em); + em =NULL ; + } + +/* +* No extent map or one for a hole or prealloc extent. Use the delalloc +* range we found in the io tree if we have one. +*/ +if (!em) +return (delalloc_len >0 ); + +/* +* We don't have any range as EXTENT_DELALLOC in the io tree, so the +* extent map is the only subrange representing delalloc. +*/ +if (delalloc_len ==0 ) { + *delalloc_start_ret = em->start; + *delalloc_end_ret =min (end, em_end -1 ); +free_extent_map (em); +return true; + } + +/* +* The extent map represents a delalloc range that starts before the +* delalloc range we found in the io tree. +*/ +if (em->start < *delalloc_start_ret) { + *delalloc_start_ret = em->start; +/* +* If the ranges are adjacent, return a combined range. +* Otherwise return the extent map's range. +*/ +if (em_end < *delalloc_start_ret) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; + } + +/* +* The extent map starts after the delalloc range we found in the io +* tree. If it's adjacent, return a combined range, otherwise return +* the range found in the io tree. +*/ +if (*delalloc_end_ret +1 == em->start) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; +} + +/* +* Check if there's delalloc in a given range. +* +* @inode: The inode. +* @start: The start offset of the range. It does not need to be +* sector size aligned. +* @end: The end offset (inclusive value) of the search range. +* It does not need to be sector size aligned. +* @delalloc_start_ret: Output argument, set to the start offset of the +* subrange found with delalloc (may not be sector size +* aligned). +* @delalloc_end_ret: Output argument, set to he end offset (inclusive value) +* of the subrange found with delalloc. +* +* Returns true if a subrange with delalloc is found within the given range, and +* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and +* end offsets of the subrange. +*/ +static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize); +u64 prev_delalloc_end =0 ; +bool ret = false; + +while (cur_offset < end) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =find_delalloc_subrange (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; + +if (prev_delalloc_end ==0 ) { +/* First subrange found. */ + *delalloc_start_ret =max (delalloc_start, start); + *delalloc_end_ret = delalloc_end; + ret = true; + }else if (delalloc_start == prev_delalloc_end +1 ) { +/* Subrange adjacent to the previous one, merge them. */ + *delalloc_end_ret = delalloc_end; + }else { +/* Subrange not adjacent to the previous one, exit. */ +break ; + } + + prev_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; +cond_resched (); + } + +return ret; +} + +/* +* Check if there's a hole or delalloc range in a range representing a hole (or +* prealloc extent) found in the inode's subvolume btree. +* +* @inode: The inode. +* @whence: Seek mode (SEEK_DATA or SEEK_HOLE). +* @start: Start offset of the hole region. It does not need to be sector +* size aligned. +* @end: End offset (inclusive value) of the hole region. It does not +* need to be sector size aligned. +* @start_ret: Return parameter, used to set the start of the subrange in the +* hole that matches the search criteria (seek mode), if such +* subrange is found (return value of the function is true). +* The value returned here may not be sector size aligned. +* +* Returns true if a subrange matching the given seek mode is found, and if one +* is found, it updates @start_ret with the start of the subrange. +*/ +static bool find_desired_extent_in_hole (struct btrfs_inode *inode,int whence, +u64 start,u64 end,u64 *start_ret) +{ +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, + &delalloc_end); +if (delalloc && whence ==SEEK_DATA ) { + *start_ret = delalloc_start; +return true; + } + +if (delalloc && whence ==SEEK_HOLE ) { +/* +* We found delalloc but it starts after out start offset. So we +* have a hole between our start offset and the delalloc start. +*/ +if (start < delalloc_start) { + *start_ret = start; +return true; + } +/* +* Delalloc range starts at our start offset. +* If the delalloc range's length is smaller than our range, +* then it means we have a hole that starts where the delalloc +* subrange ends. +*/ +if (delalloc_end < end) { + *start_ret = delalloc_end +1 ; +return true; + } + +/* There's delalloc for the whole range. */ +return false; + } + +if (!delalloc && whence ==SEEK_HOLE ) { + *start_ret = start; +return true; + } + +/* +* No delalloc in the range and we are seeking for data. The caller has +* to iterate to the next extent item in the subvolume btree. +*/ +return false; +} +static loff_t find_desired_extent (struct btrfs_inode *inode,loff_t offset,int whence) {struct btrfs_fs_info *fs_info = inode->root->fs_info; -struct extent_map *em =NULL ;struct extent_state *cached_state =NULL ; -loff_t i_size = inode->vfs_inode.i_size; +const loff_t i_size =i_size_read (&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct btrfs_path *path; +struct btrfs_key key; +u64 last_extent_end;u64 lockstart;u64 lockend;u64 start; -u64 len; -int ret =0 ; +int ret; +bool found = false;if (i_size ==0 || offset >= i_size)return -ENXIO ;
fs/btrfs/file.c static long btrfs_fallocate(struct file *file, int mode, · line 3878
Quoted 10 lines
fs/btrfs/file.c +9
-0+/* +* Quick path. If the inode has no prealloc extents and its number of +* bytes used matches its i_size, then it can not have holes. +*/ +if (whence ==SEEK_HOLE && + !(inode->flags &BTRFS_INODE_PREALLOC ) && +inode_get_bytes (&inode->vfs_inode) == i_size) +return i_size; +
Would we need a counter part for all holes quick path?'
Thanks,
Qu
Quoted 196 lines
fs/btrfs/file.c +143
-27/* * offset can be negative, in this case we start finding DATA/HOLE from * the very start of the file. @@ -3628,49 +3887,165 @@ static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,if (lockend <= lockstart) lockend = lockstart + fs_info->sectorsize; lockend--; - len = lockend - lockstart +1 ; + + path =btrfs_alloc_path (); +if (!path) +return -ENOMEM ; + path->reada =READA_FORWARD ; + + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = start; + + last_extent_end = lockstart;lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) { + goto out; + }else if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } +while (start < i_size) { - em =btrfs_get_extent_fiemap (inode, start, len); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - em =NULL ; -break ; +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *extent; +u64 extent_end; + +if (path->slots[0 ] >=btrfs_header_nritems (leaf)) { + ret =btrfs_next_leaf (root, path); +if (ret <0 ) + goto out; +else if (ret >0 ) +break ; + + leaf = path->nodes[0 ]; } -if (whence ==SEEK_HOLE && - (em->block_start ==EXTENT_MAP_HOLE || -test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) -break ; -else if (whence ==SEEK_DATA && - (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY )break ; - start = em->start + em->len; -free_extent_map (em); - em =NULL ; + extent_end =btrfs_file_extent_end (path); + +/* +* In the first iteration we may have a slot that points to an +* extent that ends before our start offset, so skip it. +*/ +if (extent_end <= start) { + path->slots[0 ]++; +continue ; + } + +/* We have an implicit hole, NO_HOLES feature is likely set. */ +if (last_extent_end < key.offset) { +u64 search_start = last_extent_end; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + key.offset -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the extent. +*/ + } + + extent =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + +if (btrfs_file_extent_disk_bytenr (leaf, extent) ==0 || +btrfs_file_extent_type (leaf, extent) == +BTRFS_FILE_EXTENT_PREALLOC ) { +/* +* Explicit hole or prealloc extent, search for delalloc. +* A prealloc extent is treated like a hole. +*/ +u64 search_start = key.offset; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + extent_end -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the next +* extent item. +*/ + }else { +/* +* Found a regular or inline extent. +* If we are seeking for data, adjust the start offset +* and stop, we're done. +*/ +if (whence ==SEEK_DATA ) { + start =max_t (u64, key.offset, offset); + found = true; +break ; + } +/* +* Else, we are seeking for a hole, check the next file +* extent item. +*/ + } + + start = extent_end; + last_extent_end = extent_end; + path->slots[0 ]++;if (fatal_signal_pending (current)) { ret = -EINTR ; -break ; + goto out; }cond_resched (); } -free_extent_map (em); + +/* We have an implicit hole from the last extent found up to i_size. */ +if (!found && start < i_size) { + found =find_desired_extent_in_hole (inode, whence, start, + i_size -1 , &start); +if (!found) + start = i_size; + } + +out:unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); -if (ret) { - offset = ret; - }else { -if (whence ==SEEK_DATA && start >= i_size) - offset = -ENXIO ; -else - offset =min_t (loff_t, start, i_size); - } +btrfs_free_path (path); + +if (ret <0 ) +return ret; + +if (whence ==SEEK_DATA && start >= i_size) +return -ENXIO ; -return offset; +return min_t (loff_t, start, i_size); }static loff_t btrfs_file_llseek (struct file *file,loff_t offset,int whence)
Re: [PATCH 02/10] btrfs: make hole and data seeking a lot more efficient
Quoted 44 lines · Qu Wenruo
Quoted 32 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current implementation of hole and data seeking for llseek does not scale well in regards to the number of extents and the distance between the start offset and the next hole or extent. This is due to a very high algorithmic complexity. Often we also get reports of btrfs' hole and data seeking (llseek) being too slow, such as at 2017's LSFMM (see the Link tag at the bottom).
In order to better understand it, lets consider the case where the start offset is 0, we are seeking for a hole and the file size is 16G. Between file offset 0 and the first hole in the file there are 100K extents - this is common for large files, specially if we have compression enabled, since the maximum extent size is limited to 128K. The steps take by the main loop of the current algorithm are the following:
1) We start by calling
btrfs_get_extent_fiemap(), for file offset 0, which callsbtrfs_get_extent(). This will first lookup for an extent map in the inode's extent map tree (a red black tree). If the extent map is not loaded in memory, then it will do a lookup for the corresponding file extent item in the subvolume's b+tree, create an extent map based on the contents of the file extent item and then add the extent map to the extent map tree of the inode;2) The second iteration calls
btrfs_get_extent_fiemap()again, this time with a start offset matching the end offset of the previous extent. Again,btrfs_get_extent()will first search the extent map tree, and if it doesn't find an extent map there, it will again search in the b+tree of the subvolume for a matching file extent item, build an extent map based on the file extent item, and add the extent map to to the extent map tree of the inode;Quoted 7 lines
3) This repeats over and over until we find the first hole (when seeking for holes) or until we find the first extent (when seeking for data).
If there no extent maps loaded in memory for each iteration, then on each iteration we do 1 extent map tree search, 1 b+tree search, plus 1 more extent map tree traversal to insert an extent map - plus we allocate memory for the extent map.
Quoted 5 lines
I'm a little intereted in if we have other workload which are heavily relying on extent map tree search?
If so, would it make sense to load a batch of file extents items into extent map tree in one go?
It depends on the workload...
I don't recall any other case, plus in the case of lseek and fiemap we don't need
to load the extent maps at all.
Quoted 4 lines
Another thing not that related to the patchset is, since extent map doesn't really get freed up unless that inode is invicted/truncated, I'm
It also gets freed on page/folio release.
Quoted 4 lines
wondering would it be a problem for heavily fragmented files to take up too much memory just for extent map tree?
Would we need a way to drop extent map in the future?
Extent maps are necessary for an efficient fsync for example.
There's a problem with extent maps I want to address, but that's unrelated to lseek and fiemap,
it's quite a separate issue.
Quoted 35 lines
Quoted 25 lines
On each iteration we are growing the size of the extent map tree, making each future search slower, and also visiting the same b+tree leaves over and over again - taking into account with the default leaf size of 16K we can fit more than 200 file extent items in a leaf - so we can visit the same b+tree leaf 200+ times, on each visit walking down a path from the root to the leaf.
So it's easy to see that what we have now doesn't scale well. Also, it loads an extent map for every file extent item into memory, which is not efficient - we should add extents maps only when doing IO (writing or reading file data).
This change implements a new algorithm which scales much better, and works like this:
1) We iterate over the subvolume's b+tree, visiting each leaf that has
file extent items once and only once;2) For any file extent items found, that don't represent holes or prealloc extents, it will not search the extent map tree - there's no need at all for that - an extent map is just an in-memory representation of a file extent item;
Quoted 9 lines
3) When a hole is found, or a prealloc extent, it will check if there's delalloc for its range. For this it will search for
EXTENT_DELALLOCbits in the inode's io tree and check the extent map tree - this is for accounting for unflushed delalloc and for flushed delalloc (the period between running delalloc and ordered extent completion), respectively. This is similar to what the current implementation does when it finds a hole or prealloc extent, but without creating extent maps and adding them to the extent map tree in case they are not loaded in memory;
Quoted 3 lines
Would it be possible that, before we starting the subvolume tree search, just run all delalloc of that target inode and prevent new writes, so we can forget about the delalloc situation completely?
That would be a significant behavioural change.
ext4 and xfs don't seem to do it, and given that lseek is widely used (cp for example),
making such change would possibly result in people reporting extra latency. I understand your pov to reduce/simplify code, but from a user's perspective I don't see
a justification to flush delalloc and wait for IO (and ordered extents) to complete.
Quoted 367 lines
Quoted 367 lines
4) It never allocates extent maps, or adds extent maps to the inode's extent map tree. This not only saves memory and time (from the tree insertions and allocations), but also eliminates the possibility of
-ENOMEMdue to allocating too many extent maps.Part of this new code will also be used later for fiemap (which also suffers similar scalability problems).
The following test example can be used to quickly measure the efficiency before and after this patch:
$ cat test-seek-hole.sh #!/bin/bashDEV=/dev/sdi
MNT=/mnt/sdimkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 16G file -> 131073 compressed extents.
xfs_io-f -c "pwrite -S0xab-b 1M 0 16G" $MNT/foobar# Leave a 1M hole at file offset 15G.
xfs_io-c "fpunch 15G 1M" $MNT/foobar# Unmount and mount again, so that we can test when there's no # metadata cached in memory.
umount $MNT
mount -o compress=lzo $DEV $MNT# Test seeking for hole from offset 0 (hole is at offset 15G).
start=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata not cached)" echostart=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata cached)" echoumount $MNT
Before this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 176 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 17 milliseconds to seek first hole (metadata cached)After this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 43 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 13 milliseconds to seek first hole (metadata cached)That's about 4X faster when no metadata is cached and about 30% faster when all metadata is cached.
In practice the differences may often be significantly higher, either due to a higher number of extents in a file or because the subvolume's b+tree is much bigger than in this example, where we only have one file.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/file.c | 437 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 406 insertions(+), 31 deletions(-)fs/btrfs/file.c +254
-4@@ -3601,22 +3601,281 @@ static long btrfs_fallocate(struct file *file, int mode,return ret; } +/* +* Helper for have_delalloc_in_range(). Find a subrange in a given range that +* has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so have_delalloc_in_range() keeps looping +* while it gets adjacent subranges, and merging them together. +*/ +static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +const u64 len = end +1 - start; +struct extent_map_tree *em_tree = &inode->extent_tree; +struct extent_map *em; +u64 em_end; +u64 delalloc_len; + +/* +* Search the io tree first for EXTENT_DELALLOC. If we find any, it +* means we have delalloc (dirty pages) for which writeback has not +* started yet. +*/ + *delalloc_start_ret = start; + delalloc_len =count_range_bits (&inode->io_tree, delalloc_start_ret, end, + len,EXTENT_DELALLOC ,1 ); +/* +* If delalloc was found then *delalloc_start_ret has a sector size +* aligned value (rounded down). +*/ +if (delalloc_len >0 ) + *delalloc_end_ret = *delalloc_start_ret + delalloc_len -1 ; + +/* +* Now also check if there's any extent map in the range that does not +* map to a hole or prealloc extent. We do this because: +* +* 1) When delalloc is flushed, the file range is locked, we clear the +* EXTENT_DELALLOC bit from the io tree and create an extent map for +* an allocated extent. So we might just have been called after +* delalloc is flushed and before the ordered extent completes and +* inserts the new file extent item in the subvolume's btree; +* +* 2) We may have an extent map created by flushing delalloc for a +* subrange that starts before the subrange we found marked with +* EXTENT_DELALLOC in the io tree. +*/ +read_lock (&em_tree->lock); + em =lookup_extent_mapping (em_tree, start, len); +read_unlock (&em_tree->lock); + +/* extent_map_end() returns a non-inclusive end offset. */ + em_end = em ?extent_map_end (em) :0 ; + +/* +* If we have a hole/prealloc extent map, check the next one if this one +* ends before our range's end. +*/ +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +struct extent_map *next_em; + +read_lock (&em_tree->lock); + next_em =lookup_extent_mapping (em_tree, em_end, len - em_end); +read_unlock (&em_tree->lock); + +free_extent_map (em); + em_end = next_em ?extent_map_end (next_em) :0 ; + em = next_em; + } + +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { +free_extent_map (em); + em =NULL ; + } + +/* +* No extent map or one for a hole or prealloc extent. Use the delalloc +* range we found in the io tree if we have one. +*/ +if (!em) +return (delalloc_len >0 ); + +/* +* We don't have any range as EXTENT_DELALLOC in the io tree, so the +* extent map is the only subrange representing delalloc. +*/ +if (delalloc_len ==0 ) { + *delalloc_start_ret = em->start; + *delalloc_end_ret =min (end, em_end -1 ); +free_extent_map (em); +return true; + } + +/* +* The extent map represents a delalloc range that starts before the +* delalloc range we found in the io tree. +*/ +if (em->start < *delalloc_start_ret) { + *delalloc_start_ret = em->start; +/* +* If the ranges are adjacent, return a combined range. +* Otherwise return the extent map's range. +*/ +if (em_end < *delalloc_start_ret) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; + } + +/* +* The extent map starts after the delalloc range we found in the io +* tree. If it's adjacent, return a combined range, otherwise return +* the range found in the io tree. +*/ +if (*delalloc_end_ret +1 == em->start) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; +} + +/* +* Check if there's delalloc in a given range. +* +* @inode: The inode. +* @start: The start offset of the range. It does not need to be +* sector size aligned. +* @end: The end offset (inclusive value) of the search range. +* It does not need to be sector size aligned. +* @delalloc_start_ret: Output argument, set to the start offset of the +* subrange found with delalloc (may not be sector size +* aligned). +* @delalloc_end_ret: Output argument, set to he end offset (inclusive value) +* of the subrange found with delalloc. +* +* Returns true if a subrange with delalloc is found within the given range, and +* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and +* end offsets of the subrange. +*/ +static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize); +u64 prev_delalloc_end =0 ; +bool ret = false; + +while (cur_offset < end) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =find_delalloc_subrange (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; + +if (prev_delalloc_end ==0 ) { +/* First subrange found. */ + *delalloc_start_ret =max (delalloc_start, start); + *delalloc_end_ret = delalloc_end; + ret = true; + }else if (delalloc_start == prev_delalloc_end +1 ) { +/* Subrange adjacent to the previous one, merge them. */ + *delalloc_end_ret = delalloc_end; + }else { +/* Subrange not adjacent to the previous one, exit. */ +break ; + } + + prev_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; +cond_resched (); + } + +return ret; +} + +/* +* Check if there's a hole or delalloc range in a range representing a hole (or +* prealloc extent) found in the inode's subvolume btree. +* +* @inode: The inode. +* @whence: Seek mode (SEEK_DATA or SEEK_HOLE). +* @start: Start offset of the hole region. It does not need to be sector +* size aligned. +* @end: End offset (inclusive value) of the hole region. It does not +* need to be sector size aligned. +* @start_ret: Return parameter, used to set the start of the subrange in the +* hole that matches the search criteria (seek mode), if such +* subrange is found (return value of the function is true). +* The value returned here may not be sector size aligned. +* +* Returns true if a subrange matching the given seek mode is found, and if one +* is found, it updates @start_ret with the start of the subrange. +*/ +static bool find_desired_extent_in_hole (struct btrfs_inode *inode,int whence, +u64 start,u64 end,u64 *start_ret) +{ +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, + &delalloc_end); +if (delalloc && whence ==SEEK_DATA ) { + *start_ret = delalloc_start; +return true; + } + +if (delalloc && whence ==SEEK_HOLE ) { +/* +* We found delalloc but it starts after out start offset. So we +* have a hole between our start offset and the delalloc start. +*/ +if (start < delalloc_start) { + *start_ret = start; +return true; + } +/* +* Delalloc range starts at our start offset. +* If the delalloc range's length is smaller than our range, +* then it means we have a hole that starts where the delalloc +* subrange ends. +*/ +if (delalloc_end < end) { + *start_ret = delalloc_end +1 ; +return true; + } + +/* There's delalloc for the whole range. */ +return false; + } + +if (!delalloc && whence ==SEEK_HOLE ) { + *start_ret = start; +return true; + } + +/* +* No delalloc in the range and we are seeking for data. The caller has +* to iterate to the next extent item in the subvolume btree. +*/ +return false; +} +static loff_t find_desired_extent (struct btrfs_inode *inode,loff_t offset,int whence) {struct btrfs_fs_info *fs_info = inode->root->fs_info; -struct extent_map *em =NULL ;struct extent_state *cached_state =NULL ; -loff_t i_size = inode->vfs_inode.i_size; +const loff_t i_size =i_size_read (&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct btrfs_path *path; +struct btrfs_key key; +u64 last_extent_end;u64 lockstart;u64 lockend;u64 start; -u64 len; -int ret =0 ; +int ret; +bool found = false;if (i_size ==0 || offset >= i_size)return -ENXIO ;
Quoted 11 lines
fs/btrfs/file.c static long btrfs_fallocate(struct file *file, int mode, · line 3878
Quoted 9 lines
fs/btrfs/file.c +9
-0+/* +* Quick path. If the inode has no prealloc extents and its number of +* bytes used matches its i_size, then it can not have holes. +*/ +if (whence ==SEEK_HOLE && + !(inode->flags &BTRFS_INODE_PREALLOC ) && +inode_get_bytes (&inode->vfs_inode) == i_size) +return i_size; +Would we need a counter part for all holes quick path?'
I suppose you mean a file with an i_size > 0 and no extents at all. That is already fast... the first btrfs_search_slot() will not find any extent item and we finish very quickly.
Thanks.
Quoted 200 lines
Thanks,
QuQuoted 196 lines
fs/btrfs/file.c +143
-27/* * offset can be negative, in this case we start finding DATA/HOLE from * the very start of the file. @@ -3628,49 +3887,165 @@ static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,if (lockend <= lockstart) lockend = lockstart + fs_info->sectorsize; lockend--; - len = lockend - lockstart +1 ; + + path =btrfs_alloc_path (); +if (!path) +return -ENOMEM ; + path->reada =READA_FORWARD ; + + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = start; + + last_extent_end = lockstart;lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) { + goto out; + }else if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } +while (start < i_size) { - em =btrfs_get_extent_fiemap (inode, start, len); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - em =NULL ; -break ; +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *extent; +u64 extent_end; + +if (path->slots[0 ] >=btrfs_header_nritems (leaf)) { + ret =btrfs_next_leaf (root, path); +if (ret <0 ) + goto out; +else if (ret >0 ) +break ; + + leaf = path->nodes[0 ]; } -if (whence ==SEEK_HOLE && - (em->block_start ==EXTENT_MAP_HOLE || -test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) -break ; -else if (whence ==SEEK_DATA && - (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY )break ; - start = em->start + em->len; -free_extent_map (em); - em =NULL ; + extent_end =btrfs_file_extent_end (path); + +/* +* In the first iteration we may have a slot that points to an +* extent that ends before our start offset, so skip it. +*/ +if (extent_end <= start) { + path->slots[0 ]++; +continue ; + } + +/* We have an implicit hole, NO_HOLES feature is likely set. */ +if (last_extent_end < key.offset) { +u64 search_start = last_extent_end; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + key.offset -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the extent. +*/ + } + + extent =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + +if (btrfs_file_extent_disk_bytenr (leaf, extent) ==0 || +btrfs_file_extent_type (leaf, extent) == +BTRFS_FILE_EXTENT_PREALLOC ) { +/* +* Explicit hole or prealloc extent, search for delalloc. +* A prealloc extent is treated like a hole. +*/ +u64 search_start = key.offset; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + extent_end -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the next +* extent item. +*/ + }else { +/* +* Found a regular or inline extent. +* If we are seeking for data, adjust the start offset +* and stop, we're done. +*/ +if (whence ==SEEK_DATA ) { + start =max_t (u64, key.offset, offset); + found = true; +break ; + } +/* +* Else, we are seeking for a hole, check the next file +* extent item. +*/ + } + + start = extent_end; + last_extent_end = extent_end; + path->slots[0 ]++;if (fatal_signal_pending (current)) { ret = -EINTR ; -break ; + goto out; }cond_resched (); } -free_extent_map (em); + +/* We have an implicit hole from the last extent found up to i_size. */ +if (!found && start < i_size) { + found =find_desired_extent_in_hole (inode, whence, start, + i_size -1 , &start); +if (!found) + start = i_size; + } + +out:unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); -if (ret) { - offset = ret; - }else { -if (whence ==SEEK_DATA && start >= i_size) - offset = -ENXIO ; -else - offset =min_t (loff_t, start, i_size); - } +btrfs_free_path (path); + +if (ret <0 ) +return ret; + +if (whence ==SEEK_DATA && start >= i_size) +return -ENXIO ; -return offset; +return min_t (loff_t, start, i_size); }static loff_t btrfs_file_llseek (struct file *file,loff_t offset,int whence)
Re: [PATCH 02/10] btrfs: make hole and data seeking a lot more efficient
In reply to [email protected]Quoted 24 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current implementation of hole and data seeking for llseek does not scale well in regards to the number of extents and the distance between the start offset and the next hole or extent. This is due to a very high algorithmic complexity. Often we also get reports of btrfs' hole and data seeking (llseek) being too slow, such as at 2017's LSFMM (see the Link tag at the bottom).
In order to better understand it, lets consider the case where the start offset is 0, we are seeking for a hole and the file size is 16G. Between file offset 0 and the first hole in the file there are 100K extents - this is common for large files, specially if we have compression enabled, since the maximum extent size is limited to 128K. The steps take by the main loop of the current algorithm are the following:
1) We start by calling
btrfs_get_extent_fiemap(), for file offset 0, which callsbtrfs_get_extent(). This will first lookup for an extent map in the inode's extent map tree (a red black tree). If the extent map is not loaded in memory, then it will do a lookup for the corresponding file extent item in the subvolume's b+tree, create an extent map based on the contents of the file extent item and then add the extent map to the extent map tree of the inode;
Quoted 7 lines
2) The second iteration calls
btrfs_get_extent_fiemap()again, this time with a start offset matching the end offset of the previous extent. Again,btrfs_get_extent()will first search the extent map tree, and if it doesn't find an extent map there, it will again search in the b+tree of the subvolume for a matching file extent item, build an extent map based on the file extent item, and add the extent map to to the extent map tree of the inode;
One small question, unrelated to the whole series since the series mostly no longer utilize the extent map tree anymore.
I'm wondering if we should add all near by extent map in a batch, other than one by one, in btrfs_get_extent()?
(E.g. if we find what we need, and there are more file extents in the same leaf, we can just continue to add them all into the cache)
However I'm not 100% sure if it's good or not, as the extent map can take up memory space and no way to free unless truncation/evict. And we have very limited usage for extent map (other than read and log tree?)
Thanks,
Qu
Quoted 612 lines
3) This repeats over and over until we find the first hole (when seeking for holes) or until we find the first extent (when seeking for data).
If there no extent maps loaded in memory for each iteration, then on each iteration we do 1 extent map tree search, 1 b+tree search, plus 1 more extent map tree traversal to insert an extent map - plus we allocate memory for the extent map.
On each iteration we are growing the size of the extent map tree, making each future search slower, and also visiting the same b+tree leaves over and over again - taking into account with the default leaf size of 16K we can fit more than 200 file extent items in a leaf - so we can visit the same b+tree leaf 200+ times, on each visit walking down a path from the root to the leaf.
So it's easy to see that what we have now doesn't scale well. Also, it loads an extent map for every file extent item into memory, which is not efficient - we should add extents maps only when doing IO (writing or reading file data).
This change implements a new algorithm which scales much better, and works like this:
1) We iterate over the subvolume's b+tree, visiting each leaf that has
file extent items once and only once;2) For any file extent items found, that don't represent holes or prealloc extents, it will not search the extent map tree - there's no need at all for that - an extent map is just an in-memory representation of a file extent item;
3) When a hole is found, or a prealloc extent, it will check if there's delalloc for its range. For this it will search for
EXTENT_DELALLOCbits in the inode's io tree and check the extent map tree - this is for accounting for unflushed delalloc and for flushed delalloc (the period between running delalloc and ordered extent completion), respectively. This is similar to what the current implementation does when it finds a hole or prealloc extent, but without creating extent maps and adding them to the extent map tree in case they are not loaded in memory;4) It never allocates extent maps, or adds extent maps to the inode's extent map tree. This not only saves memory and time (from the tree insertions and allocations), but also eliminates the possibility of
-ENOMEMdue to allocating too many extent maps.Part of this new code will also be used later for fiemap (which also suffers similar scalability problems).
The following test example can be used to quickly measure the efficiency before and after this patch:
$ cat test-seek-hole.sh #!/bin/bashDEV=/dev/sdi
MNT=/mnt/sdimkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 16G file -> 131073 compressed extents.
xfs_io-f -c "pwrite -S0xab-b 1M 0 16G" $MNT/foobar# Leave a 1M hole at file offset 15G.
xfs_io-c "fpunch 15G 1M" $MNT/foobar# Unmount and mount again, so that we can test when there's no # metadata cached in memory.
umount $MNT
mount -o compress=lzo $DEV $MNT# Test seeking for hole from offset 0 (hole is at offset 15G).
start=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata not cached)" echostart=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata cached)" echoumount $MNT
Before this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 176 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 17 milliseconds to seek first hole (metadata cached)After this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 43 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 13 milliseconds to seek first hole (metadata cached)That's about 4X faster when no metadata is cached and about 30% faster when all metadata is cached.
In practice the differences may often be significantly higher, either due to a higher number of extents in a file or because the subvolume's b+tree is much bigger than in this example, where we only have one file.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/file.c | 437 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 406 insertions(+), 31 deletions(-)fs/btrfs/file.c +406
-31@@ -3601,22 +3601,281 @@ static long btrfs_fallocate(struct file *file, int mode,return ret; } +/* +* Helper for have_delalloc_in_range(). Find a subrange in a given range that +* has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so have_delalloc_in_range() keeps looping +* while it gets adjacent subranges, and merging them together. +*/ +static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +const u64 len = end +1 - start; +struct extent_map_tree *em_tree = &inode->extent_tree; +struct extent_map *em; +u64 em_end; +u64 delalloc_len; + +/* +* Search the io tree first for EXTENT_DELALLOC. If we find any, it +* means we have delalloc (dirty pages) for which writeback has not +* started yet. +*/ + *delalloc_start_ret = start; + delalloc_len =count_range_bits (&inode->io_tree, delalloc_start_ret, end, + len,EXTENT_DELALLOC ,1 ); +/* +* If delalloc was found then *delalloc_start_ret has a sector size +* aligned value (rounded down). +*/ +if (delalloc_len >0 ) + *delalloc_end_ret = *delalloc_start_ret + delalloc_len -1 ; + +/* +* Now also check if there's any extent map in the range that does not +* map to a hole or prealloc extent. We do this because: +* +* 1) When delalloc is flushed, the file range is locked, we clear the +* EXTENT_DELALLOC bit from the io tree and create an extent map for +* an allocated extent. So we might just have been called after +* delalloc is flushed and before the ordered extent completes and +* inserts the new file extent item in the subvolume's btree; +* +* 2) We may have an extent map created by flushing delalloc for a +* subrange that starts before the subrange we found marked with +* EXTENT_DELALLOC in the io tree. +*/ +read_lock (&em_tree->lock); + em =lookup_extent_mapping (em_tree, start, len); +read_unlock (&em_tree->lock); + +/* extent_map_end() returns a non-inclusive end offset. */ + em_end = em ?extent_map_end (em) :0 ; + +/* +* If we have a hole/prealloc extent map, check the next one if this one +* ends before our range's end. +*/ +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +struct extent_map *next_em; + +read_lock (&em_tree->lock); + next_em =lookup_extent_mapping (em_tree, em_end, len - em_end); +read_unlock (&em_tree->lock); + +free_extent_map (em); + em_end = next_em ?extent_map_end (next_em) :0 ; + em = next_em; + } + +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { +free_extent_map (em); + em =NULL ; + } + +/* +* No extent map or one for a hole or prealloc extent. Use the delalloc +* range we found in the io tree if we have one. +*/ +if (!em) +return (delalloc_len >0 ); + +/* +* We don't have any range as EXTENT_DELALLOC in the io tree, so the +* extent map is the only subrange representing delalloc. +*/ +if (delalloc_len ==0 ) { + *delalloc_start_ret = em->start; + *delalloc_end_ret =min (end, em_end -1 ); +free_extent_map (em); +return true; + } + +/* +* The extent map represents a delalloc range that starts before the +* delalloc range we found in the io tree. +*/ +if (em->start < *delalloc_start_ret) { + *delalloc_start_ret = em->start; +/* +* If the ranges are adjacent, return a combined range. +* Otherwise return the extent map's range. +*/ +if (em_end < *delalloc_start_ret) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; + } + +/* +* The extent map starts after the delalloc range we found in the io +* tree. If it's adjacent, return a combined range, otherwise return +* the range found in the io tree. +*/ +if (*delalloc_end_ret +1 == em->start) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; +} + +/* +* Check if there's delalloc in a given range. +* +* @inode: The inode. +* @start: The start offset of the range. It does not need to be +* sector size aligned. +* @end: The end offset (inclusive value) of the search range. +* It does not need to be sector size aligned. +* @delalloc_start_ret: Output argument, set to the start offset of the +* subrange found with delalloc (may not be sector size +* aligned). +* @delalloc_end_ret: Output argument, set to he end offset (inclusive value) +* of the subrange found with delalloc. +* +* Returns true if a subrange with delalloc is found within the given range, and +* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and +* end offsets of the subrange. +*/ +static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize); +u64 prev_delalloc_end =0 ; +bool ret = false; + +while (cur_offset < end) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =find_delalloc_subrange (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; + +if (prev_delalloc_end ==0 ) { +/* First subrange found. */ + *delalloc_start_ret =max (delalloc_start, start); + *delalloc_end_ret = delalloc_end; + ret = true; + }else if (delalloc_start == prev_delalloc_end +1 ) { +/* Subrange adjacent to the previous one, merge them. */ + *delalloc_end_ret = delalloc_end; + }else { +/* Subrange not adjacent to the previous one, exit. */ +break ; + } + + prev_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; +cond_resched (); + } + +return ret; +} + +/* +* Check if there's a hole or delalloc range in a range representing a hole (or +* prealloc extent) found in the inode's subvolume btree. +* +* @inode: The inode. +* @whence: Seek mode (SEEK_DATA or SEEK_HOLE). +* @start: Start offset of the hole region. It does not need to be sector +* size aligned. +* @end: End offset (inclusive value) of the hole region. It does not +* need to be sector size aligned. +* @start_ret: Return parameter, used to set the start of the subrange in the +* hole that matches the search criteria (seek mode), if such +* subrange is found (return value of the function is true). +* The value returned here may not be sector size aligned. +* +* Returns true if a subrange matching the given seek mode is found, and if one +* is found, it updates @start_ret with the start of the subrange. +*/ +static bool find_desired_extent_in_hole (struct btrfs_inode *inode,int whence, +u64 start,u64 end,u64 *start_ret) +{ +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, + &delalloc_end); +if (delalloc && whence ==SEEK_DATA ) { + *start_ret = delalloc_start; +return true; + } + +if (delalloc && whence ==SEEK_HOLE ) { +/* +* We found delalloc but it starts after out start offset. So we +* have a hole between our start offset and the delalloc start. +*/ +if (start < delalloc_start) { + *start_ret = start; +return true; + } +/* +* Delalloc range starts at our start offset. +* If the delalloc range's length is smaller than our range, +* then it means we have a hole that starts where the delalloc +* subrange ends. +*/ +if (delalloc_end < end) { + *start_ret = delalloc_end +1 ; +return true; + } + +/* There's delalloc for the whole range. */ +return false; + } + +if (!delalloc && whence ==SEEK_HOLE ) { + *start_ret = start; +return true; + } + +/* +* No delalloc in the range and we are seeking for data. The caller has +* to iterate to the next extent item in the subvolume btree. +*/ +return false; +} +static loff_t find_desired_extent (struct btrfs_inode *inode,loff_t offset,int whence) {struct btrfs_fs_info *fs_info = inode->root->fs_info; -struct extent_map *em =NULL ;struct extent_state *cached_state =NULL ; -loff_t i_size = inode->vfs_inode.i_size; +const loff_t i_size =i_size_read (&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct btrfs_path *path; +struct btrfs_key key; +u64 last_extent_end;u64 lockstart;u64 lockend;u64 start; -u64 len; -int ret =0 ; +int ret; +bool found = false;if (i_size ==0 || offset >= i_size)return -ENXIO ; +/* +* Quick path. If the inode has no prealloc extents and its number of +* bytes used matches its i_size, then it can not have holes. +*/ +if (whence ==SEEK_HOLE && + !(inode->flags &BTRFS_INODE_PREALLOC ) && +inode_get_bytes (&inode->vfs_inode) == i_size) +return i_size; +/* * offset can be negative, in this case we start finding DATA/HOLE from * the very start of the file. @@ -3628,49 +3887,165 @@ static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,if (lockend <= lockstart) lockend = lockstart + fs_info->sectorsize; lockend--; - len = lockend - lockstart +1 ; + + path =btrfs_alloc_path (); +if (!path) +return -ENOMEM ; + path->reada =READA_FORWARD ; + + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = start; + + last_extent_end = lockstart;lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) { + goto out; + }else if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } +while (start < i_size) { - em =btrfs_get_extent_fiemap (inode, start, len); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - em =NULL ; -break ; +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *extent; +u64 extent_end; + +if (path->slots[0 ] >=btrfs_header_nritems (leaf)) { + ret =btrfs_next_leaf (root, path); +if (ret <0 ) + goto out; +else if (ret >0 ) +break ; + + leaf = path->nodes[0 ]; } -if (whence ==SEEK_HOLE && - (em->block_start ==EXTENT_MAP_HOLE || -test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) -break ; -else if (whence ==SEEK_DATA && - (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY )break ; - start = em->start + em->len; -free_extent_map (em); - em =NULL ; + extent_end =btrfs_file_extent_end (path); + +/* +* In the first iteration we may have a slot that points to an +* extent that ends before our start offset, so skip it. +*/ +if (extent_end <= start) { + path->slots[0 ]++; +continue ; + } + +/* We have an implicit hole, NO_HOLES feature is likely set. */ +if (last_extent_end < key.offset) { +u64 search_start = last_extent_end; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + key.offset -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the extent. +*/ + } + + extent =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + +if (btrfs_file_extent_disk_bytenr (leaf, extent) ==0 || +btrfs_file_extent_type (leaf, extent) == +BTRFS_FILE_EXTENT_PREALLOC ) { +/* +* Explicit hole or prealloc extent, search for delalloc. +* A prealloc extent is treated like a hole. +*/ +u64 search_start = key.offset; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + extent_end -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the next +* extent item. +*/ + }else { +/* +* Found a regular or inline extent. +* If we are seeking for data, adjust the start offset +* and stop, we're done. +*/ +if (whence ==SEEK_DATA ) { + start =max_t (u64, key.offset, offset); + found = true; +break ; + } +/* +* Else, we are seeking for a hole, check the next file +* extent item. +*/ + } + + start = extent_end; + last_extent_end = extent_end; + path->slots[0 ]++;if (fatal_signal_pending (current)) { ret = -EINTR ; -break ; + goto out; }cond_resched (); } -free_extent_map (em); + +/* We have an implicit hole from the last extent found up to i_size. */ +if (!found && start < i_size) { + found =find_desired_extent_in_hole (inode, whence, start, + i_size -1 , &start); +if (!found) + start = i_size; + } + +out:unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); -if (ret) { - offset = ret; - }else { -if (whence ==SEEK_DATA && start >= i_size) - offset = -ENXIO ; -else - offset =min_t (loff_t, start, i_size); - } +btrfs_free_path (path); + +if (ret <0 ) +return ret; + +if (whence ==SEEK_DATA && start >= i_size) +return -ENXIO ; -return offset; +return min_t (loff_t, start, i_size); }static loff_t btrfs_file_llseek (struct file *file,loff_t offset,int whence)
Re: [PATCH 02/10] btrfs: make hole and data seeking a lot more efficient
Quoted 44 lines · Qu Wenruo
Quoted 24 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current implementation of hole and data seeking for llseek does not scale well in regards to the number of extents and the distance between the start offset and the next hole or extent. This is due to a very high algorithmic complexity. Often we also get reports of btrfs' hole and data seeking (llseek) being too slow, such as at 2017's LSFMM (see the Link tag at the bottom).
In order to better understand it, lets consider the case where the start offset is 0, we are seeking for a hole and the file size is 16G. Between file offset 0 and the first hole in the file there are 100K extents - this is common for large files, specially if we have compression enabled, since the maximum extent size is limited to 128K. The steps take by the main loop of the current algorithm are the following:
1) We start by calling
btrfs_get_extent_fiemap(), for file offset 0, which callsbtrfs_get_extent(). This will first lookup for an extent map in the inode's extent map tree (a red black tree). If the extent map is not loaded in memory, then it will do a lookup for the corresponding file extent item in the subvolume's b+tree, create an extent map based on the contents of the file extent item and then add the extent map to the extent map tree of the inode;Quoted 7 lines
2) The second iteration calls
btrfs_get_extent_fiemap()again, this time with a start offset matching the end offset of the previous extent. Again,btrfs_get_extent()will first search the extent map tree, and if it doesn't find an extent map there, it will again search in the b+tree of the subvolume for a matching file extent item, build an extent map based on the file extent item, and add the extent map to to the extent map tree of the inode;One small question, unrelated to the whole series since the series mostly no longer utilize the extent map tree anymore.
I'm wondering if we should add all near by extent map in a batch, other than one by one, in
btrfs_get_extent()?
(E.g. if we find what we need, and there are more file extents in the same leaf, we can just continue to add them all into the cache)
Quoted 4 lines
However I'm not 100% sure if it's good or not, as the extent map can take up memory space and no way to free unless truncation/evict. And we have very limited usage for extent map (other than read and log tree?)
This seems like basically the same thing you asked before at:
https://lore.kernel.org/linux-btrfs/[email protected]/T/#md0595933858b548804fb329b4ae216ae6d76fa9a
My answer is the same I gave you back then.
I don't see anywhere in all of btrfs where we could have a benefit from bulk loading extent maps.
If something needs to load and use a large number of extent maps in a short period, then it's better off checking and using
file extent items directly, and not create and use extent maps.
If you think there's some place that could benefit from it, then try it and measure it.
As this is all unrelated to lseek, fiemap and this patchset, I'd suggest moving this discussion elsewhere.
Thanks.
Quoted 616 lines
Thanks,
QuQuoted 612 lines
3) This repeats over and over until we find the first hole (when seeking for holes) or until we find the first extent (when seeking for data).
If there no extent maps loaded in memory for each iteration, then on each iteration we do 1 extent map tree search, 1 b+tree search, plus 1 more extent map tree traversal to insert an extent map - plus we allocate memory for the extent map.
On each iteration we are growing the size of the extent map tree, making each future search slower, and also visiting the same b+tree leaves over and over again - taking into account with the default leaf size of 16K we can fit more than 200 file extent items in a leaf - so we can visit the same b+tree leaf 200+ times, on each visit walking down a path from the root to the leaf.
So it's easy to see that what we have now doesn't scale well. Also, it loads an extent map for every file extent item into memory, which is not efficient - we should add extents maps only when doing IO (writing or reading file data).
This change implements a new algorithm which scales much better, and works like this:
1) We iterate over the subvolume's b+tree, visiting each leaf that has
file extent items once and only once;2) For any file extent items found, that don't represent holes or prealloc extents, it will not search the extent map tree - there's no need at all for that - an extent map is just an in-memory representation of a file extent item;
3) When a hole is found, or a prealloc extent, it will check if there's delalloc for its range. For this it will search for
EXTENT_DELALLOCbits in the inode's io tree and check the extent map tree - this is for accounting for unflushed delalloc and for flushed delalloc (the period between running delalloc and ordered extent completion), respectively. This is similar to what the current implementation does when it finds a hole or prealloc extent, but without creating extent maps and adding them to the extent map tree in case they are not loaded in memory;4) It never allocates extent maps, or adds extent maps to the inode's extent map tree. This not only saves memory and time (from the tree insertions and allocations), but also eliminates the possibility of
-ENOMEMdue to allocating too many extent maps.Part of this new code will also be used later for fiemap (which also suffers similar scalability problems).
The following test example can be used to quickly measure the efficiency before and after this patch:
$ cat test-seek-hole.sh #!/bin/bashDEV=/dev/sdi
MNT=/mnt/sdimkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 16G file -> 131073 compressed extents.
xfs_io-f -c "pwrite -S0xab-b 1M 0 16G" $MNT/foobar# Leave a 1M hole at file offset 15G.
xfs_io-c "fpunch 15G 1M" $MNT/foobar# Unmount and mount again, so that we can test when there's no # metadata cached in memory.
umount $MNT
mount -o compress=lzo $DEV $MNT# Test seeking for hole from offset 0 (hole is at offset 15G).
start=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata not cached)" echostart=$(date +%s%N)
xfs_io-c "seek -h 0" $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "Took $dur milliseconds to seek first hole (metadata cached)" echoumount $MNT
Before this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 176 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 17 milliseconds to seek first hole (metadata cached)After this change:
$ ./test-seek-hole.sh (...) Whence Result HOLE 16106127360 Took 43 milliseconds to seek first hole (metadata not cached)Whence Result
HOLE 16106127360
Took 13 milliseconds to seek first hole (metadata cached)That's about 4X faster when no metadata is cached and about 30% faster when all metadata is cached.
In practice the differences may often be significantly higher, either due to a higher number of extents in a file or because the subvolume's b+tree is much bigger than in this example, where we only have one file.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/file.c | 437 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 406 insertions(+), 31 deletions(-)fs/btrfs/file.c +406
-31@@ -3601,22 +3601,281 @@ static long btrfs_fallocate(struct file *file, int mode,return ret; } +/* +* Helper for have_delalloc_in_range(). Find a subrange in a given range that +* has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so have_delalloc_in_range() keeps looping +* while it gets adjacent subranges, and merging them together. +*/ +static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +const u64 len = end +1 - start; +struct extent_map_tree *em_tree = &inode->extent_tree; +struct extent_map *em; +u64 em_end; +u64 delalloc_len; + +/* +* Search the io tree first for EXTENT_DELALLOC. If we find any, it +* means we have delalloc (dirty pages) for which writeback has not +* started yet. +*/ + *delalloc_start_ret = start; + delalloc_len =count_range_bits (&inode->io_tree, delalloc_start_ret, end, + len,EXTENT_DELALLOC ,1 ); +/* +* If delalloc was found then *delalloc_start_ret has a sector size +* aligned value (rounded down). +*/ +if (delalloc_len >0 ) + *delalloc_end_ret = *delalloc_start_ret + delalloc_len -1 ; + +/* +* Now also check if there's any extent map in the range that does not +* map to a hole or prealloc extent. We do this because: +* +* 1) When delalloc is flushed, the file range is locked, we clear the +* EXTENT_DELALLOC bit from the io tree and create an extent map for +* an allocated extent. So we might just have been called after +* delalloc is flushed and before the ordered extent completes and +* inserts the new file extent item in the subvolume's btree; +* +* 2) We may have an extent map created by flushing delalloc for a +* subrange that starts before the subrange we found marked with +* EXTENT_DELALLOC in the io tree. +*/ +read_lock (&em_tree->lock); + em =lookup_extent_mapping (em_tree, start, len); +read_unlock (&em_tree->lock); + +/* extent_map_end() returns a non-inclusive end offset. */ + em_end = em ?extent_map_end (em) :0 ; + +/* +* If we have a hole/prealloc extent map, check the next one if this one +* ends before our range's end. +*/ +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) && em_end < end) { +struct extent_map *next_em; + +read_lock (&em_tree->lock); + next_em =lookup_extent_mapping (em_tree, em_end, len - em_end); +read_unlock (&em_tree->lock); + +free_extent_map (em); + em_end = next_em ?extent_map_end (next_em) :0 ; + em = next_em; + } + +if (em && (em->block_start ==EXTENT_MAP_HOLE || +test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) { +free_extent_map (em); + em =NULL ; + } + +/* +* No extent map or one for a hole or prealloc extent. Use the delalloc +* range we found in the io tree if we have one. +*/ +if (!em) +return (delalloc_len >0 ); + +/* +* We don't have any range as EXTENT_DELALLOC in the io tree, so the +* extent map is the only subrange representing delalloc. +*/ +if (delalloc_len ==0 ) { + *delalloc_start_ret = em->start; + *delalloc_end_ret =min (end, em_end -1 ); +free_extent_map (em); +return true; + } + +/* +* The extent map represents a delalloc range that starts before the +* delalloc range we found in the io tree. +*/ +if (em->start < *delalloc_start_ret) { + *delalloc_start_ret = em->start; +/* +* If the ranges are adjacent, return a combined range. +* Otherwise return the extent map's range. +*/ +if (em_end < *delalloc_start_ret) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; + } + +/* +* The extent map starts after the delalloc range we found in the io +* tree. If it's adjacent, return a combined range, otherwise return +* the range found in the io tree. +*/ +if (*delalloc_end_ret +1 == em->start) + *delalloc_end_ret =min (end, em_end -1 ); + +free_extent_map (em); +return true; +} + +/* +* Check if there's delalloc in a given range. +* +* @inode: The inode. +* @start: The start offset of the range. It does not need to be +* sector size aligned. +* @end: The end offset (inclusive value) of the search range. +* It does not need to be sector size aligned. +* @delalloc_start_ret: Output argument, set to the start offset of the +* subrange found with delalloc (may not be sector size +* aligned). +* @delalloc_end_ret: Output argument, set to he end offset (inclusive value) +* of the subrange found with delalloc. +* +* Returns true if a subrange with delalloc is found within the given range, and +* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and +* end offsets of the subrange. +*/ +static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) +{ +u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize); +u64 prev_delalloc_end =0 ; +bool ret = false; + +while (cur_offset < end) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =find_delalloc_subrange (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; + +if (prev_delalloc_end ==0 ) { +/* First subrange found. */ + *delalloc_start_ret =max (delalloc_start, start); + *delalloc_end_ret = delalloc_end; + ret = true; + }else if (delalloc_start == prev_delalloc_end +1 ) { +/* Subrange adjacent to the previous one, merge them. */ + *delalloc_end_ret = delalloc_end; + }else { +/* Subrange not adjacent to the previous one, exit. */ +break ; + } + + prev_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; +cond_resched (); + } + +return ret; +} + +/* +* Check if there's a hole or delalloc range in a range representing a hole (or +* prealloc extent) found in the inode's subvolume btree. +* +* @inode: The inode. +* @whence: Seek mode (SEEK_DATA or SEEK_HOLE). +* @start: Start offset of the hole region. It does not need to be sector +* size aligned. +* @end: End offset (inclusive value) of the hole region. It does not +* need to be sector size aligned. +* @start_ret: Return parameter, used to set the start of the subrange in the +* hole that matches the search criteria (seek mode), if such +* subrange is found (return value of the function is true). +* The value returned here may not be sector size aligned. +* +* Returns true if a subrange matching the given seek mode is found, and if one +* is found, it updates @start_ret with the start of the subrange. +*/ +static bool find_desired_extent_in_hole (struct btrfs_inode *inode,int whence, +u64 start,u64 end,u64 *start_ret) +{ +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, + &delalloc_end); +if (delalloc && whence ==SEEK_DATA ) { + *start_ret = delalloc_start; +return true; + } + +if (delalloc && whence ==SEEK_HOLE ) { +/* +* We found delalloc but it starts after out start offset. So we +* have a hole between our start offset and the delalloc start. +*/ +if (start < delalloc_start) { + *start_ret = start; +return true; + } +/* +* Delalloc range starts at our start offset. +* If the delalloc range's length is smaller than our range, +* then it means we have a hole that starts where the delalloc +* subrange ends. +*/ +if (delalloc_end < end) { + *start_ret = delalloc_end +1 ; +return true; + } + +/* There's delalloc for the whole range. */ +return false; + } + +if (!delalloc && whence ==SEEK_HOLE ) { + *start_ret = start; +return true; + } + +/* +* No delalloc in the range and we are seeking for data. The caller has +* to iterate to the next extent item in the subvolume btree. +*/ +return false; +} +static loff_t find_desired_extent (struct btrfs_inode *inode,loff_t offset,int whence) {struct btrfs_fs_info *fs_info = inode->root->fs_info; -struct extent_map *em =NULL ;struct extent_state *cached_state =NULL ; -loff_t i_size = inode->vfs_inode.i_size; +const loff_t i_size =i_size_read (&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct btrfs_path *path; +struct btrfs_key key; +u64 last_extent_end;u64 lockstart;u64 lockend;u64 start; -u64 len; -int ret =0 ; +int ret; +bool found = false;if (i_size ==0 || offset >= i_size)return -ENXIO ; +/* +* Quick path. If the inode has no prealloc extents and its number of +* bytes used matches its i_size, then it can not have holes. +*/ +if (whence ==SEEK_HOLE && + !(inode->flags &BTRFS_INODE_PREALLOC ) && +inode_get_bytes (&inode->vfs_inode) == i_size) +return i_size; +/* * offset can be negative, in this case we start finding DATA/HOLE from * the very start of the file. @@ -3628,49 +3887,165 @@ static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,if (lockend <= lockstart) lockend = lockstart + fs_info->sectorsize; lockend--; - len = lockend - lockstart +1 ; + + path =btrfs_alloc_path (); +if (!path) +return -ENOMEM ; + path->reada =READA_FORWARD ; + + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = start; + + last_extent_end = lockstart;lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) { + goto out; + }else if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } +while (start < i_size) { - em =btrfs_get_extent_fiemap (inode, start, len); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - em =NULL ; -break ; +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *extent; +u64 extent_end; + +if (path->slots[0 ] >=btrfs_header_nritems (leaf)) { + ret =btrfs_next_leaf (root, path); +if (ret <0 ) + goto out; +else if (ret >0 ) +break ; + + leaf = path->nodes[0 ]; } -if (whence ==SEEK_HOLE && - (em->block_start ==EXTENT_MAP_HOLE || -test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) -break ; -else if (whence ==SEEK_DATA && - (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags))) +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY )break ; - start = em->start + em->len; -free_extent_map (em); - em =NULL ; + extent_end =btrfs_file_extent_end (path); + +/* +* In the first iteration we may have a slot that points to an +* extent that ends before our start offset, so skip it. +*/ +if (extent_end <= start) { + path->slots[0 ]++; +continue ; + } + +/* We have an implicit hole, NO_HOLES feature is likely set. */ +if (last_extent_end < key.offset) { +u64 search_start = last_extent_end; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + key.offset -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the extent. +*/ + } + + extent =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + +if (btrfs_file_extent_disk_bytenr (leaf, extent) ==0 || +btrfs_file_extent_type (leaf, extent) == +BTRFS_FILE_EXTENT_PREALLOC ) { +/* +* Explicit hole or prealloc extent, search for delalloc. +* A prealloc extent is treated like a hole. +*/ +u64 search_start = key.offset; +u64 found_start; + +/* +* First iteration, @start matches @offset and it's +* within the hole. +*/ +if (start == offset) + search_start = offset; + + found =find_desired_extent_in_hole (inode, whence, + search_start, + extent_end -1 , + &found_start); +if (found) { + start = found_start; +break ; + } +/* +* Didn't find data or a hole (due to delalloc) in the +* implicit hole range, so need to analyze the next +* extent item. +*/ + }else { +/* +* Found a regular or inline extent. +* If we are seeking for data, adjust the start offset +* and stop, we're done. +*/ +if (whence ==SEEK_DATA ) { + start =max_t (u64, key.offset, offset); + found = true; +break ; + } +/* +* Else, we are seeking for a hole, check the next file +* extent item. +*/ + } + + start = extent_end; + last_extent_end = extent_end; + path->slots[0 ]++;if (fatal_signal_pending (current)) { ret = -EINTR ; -break ; + goto out; }cond_resched (); } -free_extent_map (em); + +/* We have an implicit hole from the last extent found up to i_size. */ +if (!found && start < i_size) { + found =find_desired_extent_in_hole (inode, whence, start, + i_size -1 , &start); +if (!found) + start = i_size; + } + +out:unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); -if (ret) { - offset = ret; - }else { -if (whence ==SEEK_DATA && start >= i_size) - offset = -ENXIO ; -else - offset =min_t (loff_t, start, i_size); - } +btrfs_free_path (path); + +if (ret <0 ) +return ret; + +if (whence ==SEEK_DATA && start >= i_size) +return -ENXIO ; -return offset; +return min_t (loff_t, start, i_size); }static loff_t btrfs_file_llseek (struct file *file,loff_t offset,int whence)
PATCH 3/10 btrfs: remove check for impossible block start for an extent map at fiemap
In reply to [email protected]From: Filipe Manana <[email protected]>
During fiemap we are testing if an extent map has a block start with a value of EXTENT_MAP_LAST_BYTE, but that is never set on an extent map, and never was according to git history. So remove that useless check.
--- fs/btrfs/extent_io.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)
fs/btrfs/extent_io.c +1 -4
@@ -5642,10 +5642,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,if (off >= max) end =1 ; -if (em->block_start ==EXTENT_MAP_LAST_BYTE ) { - end =1 ; - flags |=FIEMAP_EXTENT_LAST ; - }else if (em->block_start ==EXTENT_MAP_INLINE ) { +if (em->block_start ==EXTENT_MAP_INLINE ) { flags |= (FIEMAP_EXTENT_DATA_INLINE |FIEMAP_EXTENT_NOT_ALIGNED ); }else if (em->block_start ==EXTENT_MAP_DELALLOC ) {
Signature · 1 line
2.35.1
Re: [PATCH 03/10] btrfs: remove check for impossible block start for an extent map at fiemap
Quoted 7 lines · On Thu, Sep 01, 2022 at 02:18:23PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
During fiemap we are testing if an extent map has a block start with a value of
EXTENT_MAP_LAST_BYTE, but that is never set on an extent map, and never was according to git history. So remove that useless check.Signed-off-by Filipe Manana <[email protected]>
Thanks,
Josef
Re: [PATCH 03/10] btrfs: remove check for impossible block start for an extent map at fiemap
In reply to [email protected]Quoted 7 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
During fiemap we are testing if an extent map has a block start with a value of
EXTENT_MAP_LAST_BYTE, but that is never set on an extent map, and never was according to git history. So remove that useless check.Signed-off-by Filipe Manana <[email protected]>
Thanks,
Qu
Quoted 20 lines
--- fs/btrfs/extent_io.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)fs/btrfs/extent_io.c +1
-4@@ -5642,10 +5642,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,if (off >= max) end =1 ; -if (em->block_start ==EXTENT_MAP_LAST_BYTE ) { - end =1 ; - flags |=FIEMAP_EXTENT_LAST ; - }else if (em->block_start ==EXTENT_MAP_INLINE ) { +if (em->block_start ==EXTENT_MAP_INLINE ) { flags |= (FIEMAP_EXTENT_DATA_INLINE |FIEMAP_EXTENT_NOT_ALIGNED ); }else if (em->block_start ==EXTENT_MAP_DELALLOC ) {
PATCH 4/10 btrfs: remove zero length check when entering fiemap
In reply to [email protected]From: Filipe Manana <[email protected]>
There's no point to check for a 0 length at extent_fiemap(), as before calling it, we called fiemap_prep() at btrfs_fiemap(), which already checks for a zero length and returns the same -EINVAL error. So remove the pointless check.
--- fs/btrfs/extent_io.c | 3 --- 1 file changed, 3 deletions(-)
fs/btrfs/extent_io.c +0 -3
@@ -5526,9 +5526,6 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,u64 em_len =0 ;u64 em_end =0 ; -if (len ==0 ) -return -EINVAL ; - path =btrfs_alloc_path ();if (!path)return -ENOMEM ;
Signature · 1 line
2.35.1
Re: [PATCH 04/10] btrfs: remove zero length check when entering fiemap
Quoted 8 lines · On Thu, Sep 01, 2022 at 02:18:24PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
There's no point to check for a 0 length at
extent_fiemap(), as before calling it, we calledfiemap_prep()atbtrfs_fiemap(), which already checks for a zero length and returns the same-EINVALerror. So remove the pointless check.Signed-off-by Filipe Manana <[email protected]>
Thanks,
Josef
Re: [PATCH 04/10] btrfs: remove zero length check when entering fiemap
In reply to [email protected]Quoted 8 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
There's no point to check for a 0 length at
extent_fiemap(), as before calling it, we calledfiemap_prep()atbtrfs_fiemap(), which already checks for a zero length and returns the same-EINVALerror. So remove the pointless check.Signed-off-by Filipe Manana <[email protected]>
Thanks,
Qu
Quoted 18 lines
--- fs/btrfs/extent_io.c | 3 --- 1 file changed, 3 deletions(-)fs/btrfs/extent_io.c +0
-3@@ -5526,9 +5526,6 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,u64 em_len =0 ;u64 em_end =0 ; -if (len ==0 ) -return -EINVAL ; - path =btrfs_alloc_path ();if (!path)return -ENOMEM ;
PATCH 5/10 btrfs: properly flush delalloc when entering fiemap
In reply to [email protected]From: Filipe Manana <[email protected]>
If the flag FIEMAP_FLAG_SYNC is passed to fiemap, it means all delalloc should be flushed and writeback complete. We call the generic helper fiemap_prep() which does a filemap_write_and_wait() in case that flag is given, however that is not enough if we have compression. Because a single filemap_fdatawrite_range() only starts compression (in an async thread) and therefore returns before the compression is done and writeback is started.
So make btrfs_fiemap(), actually wait for all writeback to start and complete if FIEMAP_FLAG_SYNC is set. We start and wait for writeback on the whole possible file range, from 0 to LLONG_MAX, because that is what the generic code at fiemap_prep() does.
--- fs/btrfs/inode.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+)
fs/btrfs/inode.c +20 -0
@@ -8258,6 +8258,26 @@ static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,if (ret)return ret; +/* +* fiemap_prep() called filemap_write_and_wait() for the whole possible +* file range (0 to LLONG_MAX), but that is not enough if we have +* compression enabled. The first filemap_fdatawrite_range() only kicks +* in the compression of data (in an async thread) and will return +* before the compression is done and writeback is started. A second +* filemap_fdatawrite_range() is needed to wait for the compression to +* complete and writeback to start. Without this, our user is very +* likely to get stale results, because the extents and extent maps for +* delalloc regions are only allocated when writeback starts. +*/ +if (fieinfo->fi_flags &FIEMAP_FLAG_SYNC ) { + ret =btrfs_fdatawrite_range (inode,0 ,LLONG_MAX ); +if (ret) +return ret; + ret =filemap_fdatawait_range (inode->i_mapping,0 ,LLONG_MAX ); +if (ret) +return ret; + } +return extent_fiemap (BTRFS_I (inode), fieinfo, start, len); }
Signature · 1 line
2.35.1
Re: [PATCH 05/10] btrfs: properly flush delalloc when entering fiemap
Quoted 10 lines · On Thu, Sep 01, 2022 at 02:18:25PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
If the flag
FIEMAP_FLAG_SYNCis passed to fiemap, it means all delalloc should be flushed and writeback complete. We call the generic helperfiemap_prep()which does afilemap_write_and_wait()in case that flag is given, however that is not enough if we have compression. Because a singlefilemap_fdatawrite_range()only starts compression (in an async thread) and therefore returns before the compression is done and writeback is started.
Quoted 6 lines
So make
btrfs_fiemap(), actually wait for all writeback to start and complete ifFIEMAP_FLAG_SYNCis set. We start and wait for writeback on the whole possible file range, from 0 toLLONG_MAX, because that is what the generic code atfiemap_prep()does.Signed-off-by Filipe Manana <[email protected]>
Thanks,
Josef
Re: [PATCH 05/10] btrfs: properly flush delalloc when entering fiemap
In reply to [email protected]Quoted 10 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
If the flag
FIEMAP_FLAG_SYNCis passed to fiemap, it means all delalloc should be flushed and writeback complete. We call the generic helperfiemap_prep()which does afilemap_write_and_wait()in case that flag is given, however that is not enough if we have compression. Because a singlefilemap_fdatawrite_range()only starts compression (in an async thread) and therefore returns before the compression is done and writeback is started.
Quoted 6 lines
So make
btrfs_fiemap(), actually wait for all writeback to start and complete ifFIEMAP_FLAG_SYNCis set. We start and wait for writeback on the whole possible file range, from 0 toLLONG_MAX, because that is what the generic code atfiemap_prep()does.Signed-off-by Filipe Manana <[email protected]>
Thanks,
Qu
Quoted 35 lines
--- fs/btrfs/inode.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+)fs/btrfs/inode.c +20
-0@@ -8258,6 +8258,26 @@ static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,if (ret)return ret; +/* +* fiemap_prep() called filemap_write_and_wait() for the whole possible +* file range (0 to LLONG_MAX), but that is not enough if we have +* compression enabled. The first filemap_fdatawrite_range() only kicks +* in the compression of data (in an async thread) and will return +* before the compression is done and writeback is started. A second +* filemap_fdatawrite_range() is needed to wait for the compression to +* complete and writeback to start. Without this, our user is very +* likely to get stale results, because the extents and extent maps for +* delalloc regions are only allocated when writeback starts. +*/ +if (fieinfo->fi_flags &FIEMAP_FLAG_SYNC ) { + ret =btrfs_fdatawrite_range (inode,0 ,LLONG_MAX ); +if (ret) +return ret; + ret =filemap_fdatawait_range (inode->i_mapping,0 ,LLONG_MAX ); +if (ret) +return ret; + } +return extent_fiemap (BTRFS_I (inode), fieinfo, start, len); }
PATCH 6/10 btrfs: allow fiemap to be interruptible
In reply to [email protected]From: Filipe Manana <[email protected]>
Doing fiemap on a file with a very large number of extents can take a very long time, and we have reports of it being too slow (two recent examples in the Link tags below), so make it interruptible.
--- fs/btrfs/extent_io.c | 5 +++++ 1 file changed, 5 insertions(+)
fs/btrfs/extent_io.c +5 -0
@@ -5694,6 +5694,11 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, ret =0 ; goto out_free; } + +if (fatal_signal_pending (current)) { + ret = -EINTR ; + goto out_free; + } } out_free:if (!ret)
Signature · 1 line
2.35.1
Re: [PATCH 06/10] btrfs: allow fiemap to be interruptible
Quoted 9 lines · On Thu, Sep 01, 2022 at 02:18:26PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
Doing fiemap on a file with a very large number of extents can take a very long time, and we have reports of it being too slow (two recent examples in the Link tags below), so make it interruptible.
Thanks,
Josef
Re: [PATCH 06/10] btrfs: allow fiemap to be interruptible
In reply to [email protected]Quoted 9 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
Doing fiemap on a file with a very large number of extents can take a very long time, and we have reports of it being too slow (two recent examples in the Link tags below), so make it interruptible.
Just one small question unrelated to the patch itself.
Would it be possible that, introducing a new flag to skip SHARED flag check can further speed up the fiemap operation in btrfs?
Thanks,
Qu
Quoted 20 lines
--- fs/btrfs/extent_io.c | 5 +++++ 1 file changed, 5 insertions(+)fs/btrfs/extent_io.c +5
-0@@ -5694,6 +5694,11 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, ret =0 ; goto out_free; } + +if (fatal_signal_pending (current)) { + ret = -EINTR ; + goto out_free; + } } out_free:if (!ret)
Re: [PATCH 06/10] btrfs: allow fiemap to be interruptible
Quoted 14 lines · Qu Wenruo
Quoted 9 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
Doing fiemap on a file with a very large number of extents can take a very long time, and we have reports of it being too slow (two recent examples in the Link tags below), so make it interruptible.
Quoted 6 lines
Reviewed-by Qu Wenruo <[email protected]>Just one small question unrelated to the patch itself.
Would it be possible that, introducing a new flag to skip SHARED flag check can further speed up the fiemap operation in btrfs?
Maybe, but given that fiemap is not btrfs specific, that's an interface change that would have to be
discussed with all filesystem people.
Quoted 23 lines
Thanks,
QuQuoted 20 lines
--- fs/btrfs/extent_io.c | 5 +++++ 1 file changed, 5 insertions(+)fs/btrfs/extent_io.c +5
-0@@ -5694,6 +5694,11 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, ret =0 ; goto out_free; } + +if (fatal_signal_pending (current)) { + ret = -EINTR ; + goto out_free; + } } out_free:if (!ret)
PATCH 7/10 btrfs: rename btrfs_check_shared() to a more descriptive name
In reply to [email protected]From: Filipe Manana <[email protected]>
The function btrfs_check_shared() is supposed to be used to check if a data extent is shared, but its name is too generic, may easily cause confusion in the sense that it may be used for metadata extents.
So rename it to btrfs_is_data_extent_shared(), which will also make it less confusing after the next change that adds a backref lookup cache for the b+tree nodes that lead to the leaf that contains the file extent item that points to the target data extent.
--- fs/btrfs/backref.c | 8 ++++---- fs/btrfs/backref.h | 4 ++-- fs/btrfs/extent_io.c | 5 +++-- 3 files changed, 9 insertions(+), 8 deletions(-)
fs/btrfs/backref.c +4 -4
@@ -1512,7 +1512,7 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans, }/** -* Check if an extent is shared or not +* Check if a data extent is shared or not. * * @root: root inode belongs to * @inum: inode number of the inode whose extent we are checking
@@ -1520,7 +1520,7 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* @roots: list of roots this extent is shared among * @tmp: temporary list used for iteration * -* btrfs_check_shared uses the backref walking code but will short +* btrfs_is_data_extent_shared uses the backref walking code but will short * circuit as soon as it finds a root or inode that doesn't match the * one passed in. This provides a significant performance benefit for * callers (such as fiemap) which want to know whether the extent is
@@ -1531,8 +1531,8 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. */ -int btrfs_check_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp) +int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, +struct ulist *roots,struct ulist *tmp) {struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_trans_handle *trans;
fs/btrfs/backref.h +2 -2
@@ -62,8 +62,8 @@ int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, u64 start_off, struct btrfs_path *path, struct btrfs_inode_extref **ret_extref, u64 *found_off); -int btrfs_check_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp_ulist); +int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, +struct ulist *roots,struct ulist *tmp);int __init btrfs_prelim_ref_init (void );void __cold btrfs_prelim_ref_exit (void );
fs/btrfs/extent_io.c +3 -2
@@ -5656,8 +5656,9 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,* then we're just getting a count and we can skip the * lookup stuff. */ - ret =btrfs_check_shared (root,btrfs_ino (inode), - bytenr, roots, tmp_ulist); + ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), + bytenr, roots, + tmp_ulist);if (ret <0 ) goto out_free;if (ret)
Signature · 1 line
2.35.1
Re: [PATCH 07/10] btrfs: rename btrfs_check_shared() to a more descriptive name
Quoted 6 lines · On Thu, Sep 01, 2022 at 02:18:27PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
The function
btrfs_check_shared()is supposed to be used to check if a data extent is shared, but its name is too generic, may easily cause confusion in the sense that it may be used for metadata extents.
Quoted 6 lines
So rename it to
btrfs_is_data_extent_shared(), which will also make it less confusing after the next change that adds a backref lookup cache for the b+tree nodes that lead to the leaf that contains the file extent item that points to the target data extent.Signed-off-by Filipe Manana <[email protected]>
Thanks,
Josef
Re: [PATCH 07/10] btrfs: rename btrfs_check_shared() to a more descriptive name
In reply to [email protected]Quoted 6 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
The function
btrfs_check_shared()is supposed to be used to check if a data extent is shared, but its name is too generic, may easily cause confusion in the sense that it may be used for metadata extents.
Quoted 6 lines
So rename it to
btrfs_is_data_extent_shared(), which will also make it less confusing after the next change that adds a backref lookup cache for the b+tree nodes that lead to the leaf that contains the file extent item that points to the target data extent.Signed-off-by Filipe Manana <[email protected]>
Thanks,
Qu
Quoted 70 lines
--- fs/btrfs/backref.c | 8 ++++---- fs/btrfs/backref.h | 4 ++-- fs/btrfs/extent_io.c | 5 +++-- 3 files changed, 9 insertions(+), 8 deletions(-)fs/btrfs/backref.c +4
-4@@ -1512,7 +1512,7 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans, }/** -* Check if an extent is shared or not +* Check if a data extent is shared or not. * * @root: root inode belongs to * @inum: inode number of the inode whose extent we are checking @@ -1520,7 +1520,7 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* @roots: list of roots this extent is shared among * @tmp: temporary list used for iteration * -* btrfs_check_shared uses the backref walking code but will short +* btrfs_is_data_extent_shared uses the backref walking code but will short * circuit as soon as it finds a root or inode that doesn't match the * one passed in. This provides a significant performance benefit for * callers (such as fiemap) which want to know whether the extent is @@ -1531,8 +1531,8 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. */ -int btrfs_check_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp) +int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, +struct ulist *roots,struct ulist *tmp) {struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_trans_handle *trans;fs/btrfs/backref.h +2
-2@@ -62,8 +62,8 @@ int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, u64 start_off, struct btrfs_path *path, struct btrfs_inode_extref **ret_extref, u64 *found_off); -int btrfs_check_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp_ulist); +int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, +struct ulist *roots,struct ulist *tmp);int __init btrfs_prelim_ref_init (void );void __cold btrfs_prelim_ref_exit (void );fs/btrfs/extent_io.c +3
-2@@ -5656,8 +5656,9 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,* then we're just getting a count and we can skip the * lookup stuff. */ - ret =btrfs_check_shared (root,btrfs_ino (inode), - bytenr, roots, tmp_ulist); + ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), + bytenr, roots, + tmp_ulist);if (ret <0 ) goto out_free;if (ret)
PATCH 8/10 btrfs: speedup checking for extent sharedness during fiemap
In reply to [email protected]From: Filipe Manana <[email protected]>
One of the most expensive tasks performed during fiemap is to check if an extent is shared. This task has two major steps:
1) Check if the data extent is shared. This implies checking the extent item in the extent tree, checking delayed references, etc. If we find the data extent is directly shared, we terminate immediately;
2) If the data extent is not directly shared (its extent item has a refcount of 1), then it may be shared if we have snapshots that share subtrees of the inode's subvolume b+tree. So we check if the leaf containing the file extent item is shared, then its parent node, then the parent node of the parent node, etc, until we reach the root node or we find one of them is shared - in which case we stop immediately.
During fiemap we process the extents of a file from left to right, from file offset 0 to eof. This means that we iterate b+tree leaves from left to right, and has the implication that we keep repeating that second step above several times for the same b+tree path of the inode's subvolume b+tree.
For example, if we have two file extent items in leaf X, and the path to leaf X is A -> B -> C -> X, then when we try to determine if the data extent referenced by the first extent item is shared, we check if the data extent is shared - if it's not, then we check if leaf X is shared, if not, then we check if node C is shared, if not, then check if node B is shared, if not than check if node A is shared. When we move to the next file extent item, after determining the data extent is not shared, we repeat the checks for X, C, B and A - doing all the expensive searches in the extent tree, delayed refs, etc. If we have thousands of tile extents, then we keep repeating the sharedness checks for the same paths over and over.
On a file that has no shared extents or only a small portion, it's easy to see that this scales terribly with the number of extents in the file and the sizes of the extent and subvolume b+trees.
This change eliminates the repeated sharedness check on extent buffers by caching the results of the last path used. The results can be used as long as no snapshots were created since they were cached (for not shared extent buffers) or no roots were dropped since they were cached (for shared extent buffers). This greatly reduces the time spent by fiemap for files with thousands of extents and/or large extent and subvolume b+trees.
Example performance test:
$ cat fiemap-perf-test.sh
#!/bin/bash DEV=/dev/sdi
MNT=/mnt/sdi
mkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 40G gives 327680 128K file extents (due to compression).
xfs_io -f -c "pwrite -S 0xab -b 1M 0 40G" $MNT/foobar
umount $MNT
mount -o compress=lzo $DEV $MNT
start=$(date +%s%N)
filefrag $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "fiemap took $dur milliseconds (metadata not cached)"
start=$(date +%s%N)
filefrag $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "fiemap took $dur milliseconds (metadata cached)"
umount $MNTBefore this patch:
$ ./fiemap-perf-test.sh
(...)
/mnt/sdi/foobar: 327680 extents found
fiemap took 3597 milliseconds (metadata not cached)
/mnt/sdi/foobar: 327680 extents found
fiemap took 2107 milliseconds (metadata cached)After this patch:
$ ./fiemap-perf-test.sh
(...)
/mnt/sdi/foobar: 327680 extents found
fiemap took 1646 milliseconds (metadata not cached)
/mnt/sdi/foobar: 327680 extents found
fiemap took 698 milliseconds (metadata cached)That's about 2.2x faster when no metadata is cached, and about 3x faster when all metadata is cached. On a real filesystem with many other files, data, directories, etc, the b+trees will be 2 or 3 levels higher, therefore this optimization will have a higher impact.
Several reports of a slow fiemap show up often, the two Link tags below refer to two recent reports of such slowness. This patch, together with the next ones in the series, is meant to address that.
--- fs/btrfs/backref.c | 122 ++++++++++++++++++++++++++++++++++++++++- fs/btrfs/backref.h | 17 +++++- fs/btrfs/ctree.h | 18 ++++++ fs/btrfs/extent-tree.c | 10 +++- fs/btrfs/extent_io.c | 11 ++-- 5 files changed, 170 insertions(+), 8 deletions(-)
fs/btrfs/backref.c +121 -1
@@ -1511,6 +1511,105 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,return ret; } +/* +* The caller has joined a transaction or is holding a read lock on the +* fs_info->commit_root_sem semaphore, so no need to worry about the root's last +* snapshot field changing while updating or checking the cache. +*/ +static bool lookup_backref_shared_cache (struct btrfs_backref_shared_cache *cache, +struct btrfs_root *root, +u64 bytenr,int level,bool *is_shared) +{ +struct btrfs_backref_shared_cache_entry *entry; + +if (WARN_ON_ONCE (level >=BTRFS_MAX_LEVEL )) +return false; + +/* +* Level -1 is used for the data extent, which is not reliable to cache +* because its reference count can increase or decrease without us +* realizing. We cache results only for extent buffers that lead from +* the root node down to the leaf with the file extent item. +*/ +ASSERT (level >=0 ); + + entry = &cache->entries[level]; + +/* Unused cache entry or being used for some other extent buffer. */ +if (entry->bytenr != bytenr) +return false; + +/* +* We cached a false result, but the last snapshot generation of the +* root changed, so we now have a snapshot. Don't trust the result. +*/ +if (!entry->is_shared && + entry->gen !=btrfs_root_last_snapshot (&root->root_item)) +return false; + +/* +* If we cached a true result and the last generation used for dropping +* a root changed, we can not trust the result, because the dropped root +* could be a snapshot sharing this extent buffer. +*/ +if (entry->is_shared && + entry->gen !=btrfs_get_last_root_drop_gen (root->fs_info)) +return false; + + *is_shared = entry->is_shared; + +return true; +} + +/* +* The caller has joined a transaction or is holding a read lock on the +* fs_info->commit_root_sem semaphore, so no need to worry about the root's last +* snapshot field changing while updating or checking the cache. +*/ +static void store_backref_shared_cache (struct btrfs_backref_shared_cache *cache, +struct btrfs_root *root, +u64 bytenr,int level,bool is_shared) +{ +struct btrfs_backref_shared_cache_entry *entry; +u64 gen; + +if (WARN_ON_ONCE (level >=BTRFS_MAX_LEVEL )) +return ; + +/* +* Level -1 is used for the data extent, which is not reliable to cache +* because its reference count can increase or decrease without us +* realizing. We cache results only for extent buffers that lead from +* the root node down to the leaf with the file extent item. +*/ +ASSERT (level >=0 ); + +if (is_shared) + gen =btrfs_get_last_root_drop_gen (root->fs_info); +else + gen =btrfs_root_last_snapshot (&root->root_item); + + entry = &cache->entries[level]; + entry->bytenr = bytenr; + entry->is_shared = is_shared; + entry->gen = gen; + +/* +* If we found an extent buffer is shared, set the cache result for all +* extent buffers below it to true. As nodes in the path are COWed, +* their sharedness is moved to their children, and if a leaf is COWed, +* then the sharedness of a data extent becomes direct, the refcount of +* data extent is increased in the extent item at the extent tree. +*/ +if (is_shared) { +for (int i =0 ; i < level; i++) { + entry = &cache->entries[i]; + entry->is_shared = is_shared; + entry->gen = gen; + } + } +} +/** * Check if a data extent is shared or not. *
@@ -1519,6 +1618,7 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* @bytenr: logical bytenr of the extent we are checking * @roots: list of roots this extent is shared among * @tmp: temporary list used for iteration +* @cache: a backref lookup result cache * * btrfs_is_data_extent_shared uses the backref walking code but will short * circuit as soon as it finds a root or inode that doesn't match the
@@ -1532,7 +1632,8 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. */ int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp) +struct ulist *roots,struct ulist *tmp, +struct btrfs_backref_shared_cache *cache) {struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_trans_handle *trans;
@@ -1545,6 +1646,7 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr, .inum = inum, .share_count = 0, }; +int level;ulist_init (roots);ulist_init (tmp);
@@ -1561,22 +1663,40 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,btrfs_get_tree_mod_seq (fs_info, &elem); } +/* -1 means we are in the bytenr of the data extent. */ + level =-1 ;ULIST_ITER_INIT (&uiter);while (1 ) { +bool is_shared; +bool cached; + ret =find_parent_nodes (trans, fs_info, bytenr, elem.seq, tmp, roots,NULL , &shared, false);if (ret ==BACKREF_FOUND_SHARED ) {/* this is the only condition under which we return 1 */ ret =1 ; +if (level >=0 ) +store_backref_shared_cache (cache, root, bytenr, + level, true);break ; }if (ret <0 && ret != -ENOENT )break ; ret =0 ; +if (level >=0 ) +store_backref_shared_cache (cache, root, bytenr, + level, false); node =ulist_next (tmp, &uiter);if (!node)break ; bytenr = node->val; + level++; + cached =lookup_backref_shared_cache (cache, root, bytenr, level, + &is_shared); +if (cached) { + ret = is_shared ?1 :0 ; +break ; + } shared.share_count =0 ;cond_resched (); }
fs/btrfs/backref.h +16 -1
@@ -17,6 +17,20 @@ struct inode_fs_paths {struct btrfs_data_container *fspath; }; +struct btrfs_backref_shared_cache_entry { +u64 bytenr; +u64 gen; +bool is_shared; +}; + +struct btrfs_backref_shared_cache { +/* +* A path from a root to a leaf that has a file extent item pointing to +* a given data extent should never exceed the maximum b+tree heigth. +*/ +struct btrfs_backref_shared_cache_entry entries[BTRFS_MAX_LEVEL ]; +}; +typedef int (iterate_extent_inodes_t )(u64 inum,u64 offset,u64 root,void *ctx);
@@ -63,7 +77,8 @@ int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,struct btrfs_inode_extref **ret_extref, u64 *found_off);int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp); +struct ulist *roots,struct ulist *tmp, +struct btrfs_backref_shared_cache *cache);int __init btrfs_prelim_ref_init (void );void __cold btrfs_prelim_ref_exit (void );
fs/btrfs/ctree.h +18 -0
@@ -1095,6 +1095,13 @@ struct btrfs_fs_info {/* Updates are not protected by any lock */ struct btrfs_commit_stats commit_stats; +/* +* Last generation where we dropped a non-relocation root. +* Use btrfs_set_last_root_drop_gen() and btrfs_get_last_root_drop_gen() +* to change it and to read it, respectively. +*/ +u64 last_root_drop_gen; +/* * Annotations for transaction events (structures are empty when * compiled without lockdep).
@@ -1119,6 +1126,17 @@ struct btrfs_fs_info {#endif }; +static inline void btrfs_set_last_root_drop_gen (struct btrfs_fs_info *fs_info, +u64 gen) +{ +WRITE_ONCE (fs_info->last_root_drop_gen, gen); +} + +static inline u64 btrfs_get_last_root_drop_gen (const struct btrfs_fs_info *fs_info) +{ +return READ_ONCE (fs_info->last_root_drop_gen); +} +static inline struct btrfs_fs_info *btrfs_sb (struct super_block *sb) {return sb->s_fs_info;
fs/btrfs/extent-tree.c +9 -1
@@ -5635,6 +5635,8 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,*/ int btrfs_drop_snapshot (struct btrfs_root *root,int update_ref,int for_reloc) { +const bool is_reloc_root = (root->root_key.objectid == +BTRFS_TREE_RELOC_OBJECTID );struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_path *path;struct btrfs_trans_handle *trans;
@@ -5794,6 +5796,9 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) goto out_end_trans; } +if (!is_reloc_root) +btrfs_set_last_root_drop_gen (fs_info, trans->transid); +btrfs_end_transaction_throttle (trans);if (!for_reloc &&btrfs_need_cleaner_sleep (fs_info)) { btrfs_debug(fs_info,
@@ -5828,7 +5833,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) goto out_end_trans; } -if (root->root_key.objectid !=BTRFS_TREE_RELOC_OBJECTID ) { +if (!is_reloc_root) { ret =btrfs_find_root (tree_root, &root->root_key, path,NULL ,NULL );if (ret <0 ) {
@@ -5860,6 +5865,9 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)btrfs_put_root (root); root_dropped = true; out_end_trans: +if (!is_reloc_root) +btrfs_set_last_root_drop_gen (fs_info, trans->transid); +btrfs_end_transaction_throttle (trans); out_free:kfree (wc);
fs/btrfs/extent_io.c +6 -5
@@ -5519,6 +5519,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,struct btrfs_path *path;struct btrfs_root *root = inode->root;struct fiemap_cache cache = {0 }; +struct btrfs_backref_shared_cache *backref_cache;struct ulist *roots;struct ulist *tmp_ulist;int end =0 ;
@@ -5526,13 +5527,11 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,u64 em_len =0 ;u64 em_end =0 ; + backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); path =btrfs_alloc_path (); -if (!path) -return -ENOMEM ; - roots =ulist_alloc (GFP_KERNEL ); tmp_ulist =ulist_alloc (GFP_KERNEL ); -if (!roots || !tmp_ulist) { +if (!backref_cache || !path || !roots || !tmp_ulist) { ret = -ENOMEM ; goto out_free_ulist; }
@@ -5658,7 +5657,8 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,*/ ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), bytenr, roots, - tmp_ulist); + tmp_ulist, + backref_cache);if (ret <0 ) goto out_free;if (ret)
@@ -5710,6 +5710,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, &cached_state); out_free_ulist: +kfree (backref_cache);btrfs_free_path (path);ulist_free (roots);ulist_free (tmp_ulist);
Signature · 1 line
2.35.1
Re: [PATCH 08/10] btrfs: speedup checking for extent sharedness during fiemap
Quoted 98 lines · On Thu, Sep 01, 2022 at 02:18:28PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
One of the most expensive tasks performed during fiemap is to check if an extent is shared. This task has two major steps:
1) Check if the data extent is shared. This implies checking the extent item in the extent tree, checking delayed references, etc. If we find the data extent is directly shared, we terminate immediately;
2) If the data extent is not directly shared (its extent item has a refcount of 1), then it may be shared if we have snapshots that share subtrees of the inode's subvolume b+tree. So we check if the leaf containing the file extent item is shared, then its parent node, then the parent node of the parent node, etc, until we reach the root node or we find one of them is shared - in which case we stop immediately.
During fiemap we process the extents of a file from left to right, from file offset 0 to eof. This means that we iterate b+tree leaves from left to right, and has the implication that we keep repeating that second step above several times for the same b+tree path of the inode's subvolume b+tree.
For example, if we have two file extent items in leaf X, and the path to leaf X is A -> B -> C -> X, then when we try to determine if the data extent referenced by the first extent item is shared, we check if the data extent is shared - if it's not, then we check if leaf X is shared, if not, then we check if node C is shared, if not, then check if node B is shared, if not than check if node A is shared. When we move to the next file extent item, after determining the data extent is not shared, we repeat the checks for X, C, B and A - doing all the expensive searches in the extent tree, delayed refs, etc. If we have thousands of tile extents, then we keep repeating the sharedness checks for the same paths over and over.
On a file that has no shared extents or only a small portion, it's easy to see that this scales terribly with the number of extents in the file and the sizes of the extent and subvolume b+trees.
This change eliminates the repeated sharedness check on extent buffers by caching the results of the last path used. The results can be used as long as no snapshots were created since they were cached (for not shared extent buffers) or no roots were dropped since they were cached (for shared extent buffers). This greatly reduces the time spent by fiemap for files with thousands of extents and/or large extent and subvolume b+trees.
Example performance test:
$ cat fiemap-perf-test.sh #!/bin/bashDEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f $DEV mount -o compress=lzo $DEV $MNT # 40G gives 327680 128K file extents (due to compression). xfs_io -f -c "pwrite -S 0xab -b 1M 0 40G" $MNT/foobar umount $MNT mount -o compress=lzo $DEV $MNT start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata not cached)" start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata cached)" umount $MNTBefore this patch:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 3597 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 2107 milliseconds (metadata cached)After this patch:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 1646 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 698 milliseconds (metadata cached)That's about 2.2x faster when no metadata is cached, and about 3x faster when all metadata is cached. On a real filesystem with many other files, data, directories, etc, the b+trees will be 2 or 3 levels higher, therefore this optimization will have a higher impact.
Quoted 7 lines
Several reports of a slow fiemap show up often, the two Link tags below refer to two recent reports of such slowness. This patch, together with the next ones in the series, is meant to address that.
Thanks,
Josef
Re: [PATCH 08/10] btrfs: speedup checking for extent sharedness during fiemap
In reply to [email protected]Quoted 37 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
One of the most expensive tasks performed during fiemap is to check if an extent is shared. This task has two major steps:
1) Check if the data extent is shared. This implies checking the extent item in the extent tree, checking delayed references, etc. If we find the data extent is directly shared, we terminate immediately;
2) If the data extent is not directly shared (its extent item has a refcount of 1), then it may be shared if we have snapshots that share subtrees of the inode's subvolume b+tree. So we check if the leaf containing the file extent item is shared, then its parent node, then the parent node of the parent node, etc, until we reach the root node or we find one of them is shared - in which case we stop immediately.
During fiemap we process the extents of a file from left to right, from file offset 0 to eof. This means that we iterate b+tree leaves from left to right, and has the implication that we keep repeating that second step above several times for the same b+tree path of the inode's subvolume b+tree.
For example, if we have two file extent items in leaf X, and the path to leaf X is A -> B -> C -> X, then when we try to determine if the data extent referenced by the first extent item is shared, we check if the data extent is shared - if it's not, then we check if leaf X is shared, if not, then we check if node C is shared, if not, then check if node B is shared, if not than check if node A is shared. When we move to the next file extent item, after determining the data extent is not shared, we repeat the checks for X, C, B and A - doing all the expensive searches in the extent tree, delayed refs, etc. If we have thousands of tile extents, then we keep repeating the sharedness checks for the same paths over and over.
On a file that has no shared extents or only a small portion, it's easy to see that this scales terribly with the number of extents in the file and the sizes of the extent and subvolume b+trees.
Quoted 6 lines
This change eliminates the repeated sharedness check on extent buffers by caching the results of the last path used. The results can be used as long as no snapshots were created since they were cached (for not shared extent buffers) or no roots were dropped since they were cached (for shared extent buffers). This greatly reduces the time spent by fiemap for files with thousands of extents and/or large extent and subvolume b+trees.
This sounds pretty much like the existing btrfs_backref_cache is doing.
It stores a map to speedup the backref lookup.
But a quick search didn't hit things like btrfs_backref_edge() or btrfs_backref_cache().
Would it be possible to reuse the existing facility to do the same thing?
Thanks,
Qu
Quoted 406 lines
Example performance test:
$ cat fiemap-perf-test.sh #!/bin/bashDEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f $DEV mount -o compress=lzo $DEV $MNT # 40G gives 327680 128K file extents (due to compression). xfs_io -f -c "pwrite -S 0xab -b 1M 0 40G" $MNT/foobar umount $MNT mount -o compress=lzo $DEV $MNT start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata not cached)" start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata cached)" umount $MNTBefore this patch:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 3597 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 2107 milliseconds (metadata cached)After this patch:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 1646 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 698 milliseconds (metadata cached)That's about 2.2x faster when no metadata is cached, and about 3x faster when all metadata is cached. On a real filesystem with many other files, data, directories, etc, the b+trees will be 2 or 3 levels higher, therefore this optimization will have a higher impact.
Several reports of a slow fiemap show up often, the two Link tags below refer to two recent reports of such slowness. This patch, together with the next ones in the series, is meant to address that.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/backref.c | 122 ++++++++++++++++++++++++++++++++++++++++- fs/btrfs/backref.h | 17 +++++- fs/btrfs/ctree.h | 18 ++++++ fs/btrfs/extent-tree.c | 10 +++- fs/btrfs/extent_io.c | 11 ++-- 5 files changed, 170 insertions(+), 8 deletions(-)fs/btrfs/backref.c +121
-1@@ -1511,6 +1511,105 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,return ret; } +/* +* The caller has joined a transaction or is holding a read lock on the +* fs_info->commit_root_sem semaphore, so no need to worry about the root's last +* snapshot field changing while updating or checking the cache. +*/ +static bool lookup_backref_shared_cache (struct btrfs_backref_shared_cache *cache, +struct btrfs_root *root, +u64 bytenr,int level,bool *is_shared) +{ +struct btrfs_backref_shared_cache_entry *entry; + +if (WARN_ON_ONCE (level >=BTRFS_MAX_LEVEL )) +return false; + +/* +* Level -1 is used for the data extent, which is not reliable to cache +* because its reference count can increase or decrease without us +* realizing. We cache results only for extent buffers that lead from +* the root node down to the leaf with the file extent item. +*/ +ASSERT (level >=0 ); + + entry = &cache->entries[level]; + +/* Unused cache entry or being used for some other extent buffer. */ +if (entry->bytenr != bytenr) +return false; + +/* +* We cached a false result, but the last snapshot generation of the +* root changed, so we now have a snapshot. Don't trust the result. +*/ +if (!entry->is_shared && + entry->gen !=btrfs_root_last_snapshot (&root->root_item)) +return false; + +/* +* If we cached a true result and the last generation used for dropping +* a root changed, we can not trust the result, because the dropped root +* could be a snapshot sharing this extent buffer. +*/ +if (entry->is_shared && + entry->gen !=btrfs_get_last_root_drop_gen (root->fs_info)) +return false; + + *is_shared = entry->is_shared; + +return true; +} + +/* +* The caller has joined a transaction or is holding a read lock on the +* fs_info->commit_root_sem semaphore, so no need to worry about the root's last +* snapshot field changing while updating or checking the cache. +*/ +static void store_backref_shared_cache (struct btrfs_backref_shared_cache *cache, +struct btrfs_root *root, +u64 bytenr,int level,bool is_shared) +{ +struct btrfs_backref_shared_cache_entry *entry; +u64 gen; + +if (WARN_ON_ONCE (level >=BTRFS_MAX_LEVEL )) +return ; + +/* +* Level -1 is used for the data extent, which is not reliable to cache +* because its reference count can increase or decrease without us +* realizing. We cache results only for extent buffers that lead from +* the root node down to the leaf with the file extent item. +*/ +ASSERT (level >=0 ); + +if (is_shared) + gen =btrfs_get_last_root_drop_gen (root->fs_info); +else + gen =btrfs_root_last_snapshot (&root->root_item); + + entry = &cache->entries[level]; + entry->bytenr = bytenr; + entry->is_shared = is_shared; + entry->gen = gen; + +/* +* If we found an extent buffer is shared, set the cache result for all +* extent buffers below it to true. As nodes in the path are COWed, +* their sharedness is moved to their children, and if a leaf is COWed, +* then the sharedness of a data extent becomes direct, the refcount of +* data extent is increased in the extent item at the extent tree. +*/ +if (is_shared) { +for (int i =0 ; i < level; i++) { + entry = &cache->entries[i]; + entry->is_shared = is_shared; + entry->gen = gen; + } + } +} +/** * Check if a data extent is shared or not. * @@ -1519,6 +1618,7 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* @bytenr: logical bytenr of the extent we are checking * @roots: list of roots this extent is shared among * @tmp: temporary list used for iteration +* @cache: a backref lookup result cache * * btrfs_is_data_extent_shared uses the backref walking code but will short * circuit as soon as it finds a root or inode that doesn't match the @@ -1532,7 +1632,8 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. */ int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp) +struct ulist *roots,struct ulist *tmp, +struct btrfs_backref_shared_cache *cache) {struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_trans_handle *trans;@@ -1545,6 +1646,7 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr, .inum = inum, .share_count = 0, }; +int level;ulist_init (roots);ulist_init (tmp);@@ -1561,22 +1663,40 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,btrfs_get_tree_mod_seq (fs_info, &elem); } +/* -1 means we are in the bytenr of the data extent. */ + level =-1 ;ULIST_ITER_INIT (&uiter);while (1 ) { +bool is_shared; +bool cached; + ret =find_parent_nodes (trans, fs_info, bytenr, elem.seq, tmp, roots,NULL , &shared, false);if (ret ==BACKREF_FOUND_SHARED ) {/* this is the only condition under which we return 1 */ ret =1 ; +if (level >=0 ) +store_backref_shared_cache (cache, root, bytenr, + level, true);break ; }if (ret <0 && ret != -ENOENT )break ; ret =0 ; +if (level >=0 ) +store_backref_shared_cache (cache, root, bytenr, + level, false); node =ulist_next (tmp, &uiter);if (!node)break ; bytenr = node->val; + level++; + cached =lookup_backref_shared_cache (cache, root, bytenr, level, + &is_shared); +if (cached) { + ret = is_shared ?1 :0 ; +break ; + } shared.share_count =0 ;cond_resched (); }fs/btrfs/backref.h +16
-1@@ -17,6 +17,20 @@ struct inode_fs_paths {struct btrfs_data_container *fspath; }; +struct btrfs_backref_shared_cache_entry { +u64 bytenr; +u64 gen; +bool is_shared; +}; + +struct btrfs_backref_shared_cache { +/* +* A path from a root to a leaf that has a file extent item pointing to +* a given data extent should never exceed the maximum b+tree heigth. +*/ +struct btrfs_backref_shared_cache_entry entries[BTRFS_MAX_LEVEL ]; +}; +typedef int (iterate_extent_inodes_t )(u64 inum,u64 offset,u64 root,void *ctx);@@ -63,7 +77,8 @@ int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,struct btrfs_inode_extref **ret_extref, u64 *found_off);int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp); +struct ulist *roots,struct ulist *tmp, +struct btrfs_backref_shared_cache *cache);int __init btrfs_prelim_ref_init (void );void __cold btrfs_prelim_ref_exit (void );fs/btrfs/ctree.h +18
-0@@ -1095,6 +1095,13 @@ struct btrfs_fs_info {/* Updates are not protected by any lock */ struct btrfs_commit_stats commit_stats; +/* +* Last generation where we dropped a non-relocation root. +* Use btrfs_set_last_root_drop_gen() and btrfs_get_last_root_drop_gen() +* to change it and to read it, respectively. +*/ +u64 last_root_drop_gen; +/* * Annotations for transaction events (structures are empty when * compiled without lockdep). @@ -1119,6 +1126,17 @@ struct btrfs_fs_info {#endif }; +static inline void btrfs_set_last_root_drop_gen (struct btrfs_fs_info *fs_info, +u64 gen) +{ +WRITE_ONCE (fs_info->last_root_drop_gen, gen); +} + +static inline u64 btrfs_get_last_root_drop_gen (const struct btrfs_fs_info *fs_info) +{ +return READ_ONCE (fs_info->last_root_drop_gen); +} +static inline struct btrfs_fs_info *btrfs_sb (struct super_block *sb) {return sb->s_fs_info;fs/btrfs/extent-tree.c +9
-1@@ -5635,6 +5635,8 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,*/ int btrfs_drop_snapshot (struct btrfs_root *root,int update_ref,int for_reloc) { +const bool is_reloc_root = (root->root_key.objectid == +BTRFS_TREE_RELOC_OBJECTID );struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_path *path;struct btrfs_trans_handle *trans;@@ -5794,6 +5796,9 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) goto out_end_trans; } +if (!is_reloc_root) +btrfs_set_last_root_drop_gen (fs_info, trans->transid); +btrfs_end_transaction_throttle (trans);if (!for_reloc &&btrfs_need_cleaner_sleep (fs_info)) { btrfs_debug(fs_info,@@ -5828,7 +5833,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) goto out_end_trans; } -if (root->root_key.objectid !=BTRFS_TREE_RELOC_OBJECTID ) { +if (!is_reloc_root) { ret =btrfs_find_root (tree_root, &root->root_key, path,NULL ,NULL );if (ret <0 ) {@@ -5860,6 +5865,9 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)btrfs_put_root (root); root_dropped = true; out_end_trans: +if (!is_reloc_root) +btrfs_set_last_root_drop_gen (fs_info, trans->transid); +btrfs_end_transaction_throttle (trans); out_free:kfree (wc);fs/btrfs/extent_io.c +6
-5@@ -5519,6 +5519,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,struct btrfs_path *path;struct btrfs_root *root = inode->root;struct fiemap_cache cache = {0 }; +struct btrfs_backref_shared_cache *backref_cache;struct ulist *roots;struct ulist *tmp_ulist;int end =0 ;@@ -5526,13 +5527,11 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,u64 em_len =0 ;u64 em_end =0 ; + backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); path =btrfs_alloc_path (); -if (!path) -return -ENOMEM ; - roots =ulist_alloc (GFP_KERNEL ); tmp_ulist =ulist_alloc (GFP_KERNEL ); -if (!roots || !tmp_ulist) { +if (!backref_cache || !path || !roots || !tmp_ulist) { ret = -ENOMEM ; goto out_free_ulist; }@@ -5658,7 +5657,8 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,*/ ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), bytenr, roots, - tmp_ulist); + tmp_ulist, + backref_cache);if (ret <0 ) goto out_free;if (ret)@@ -5710,6 +5710,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, &cached_state); out_free_ulist: +kfree (backref_cache);btrfs_free_path (path);ulist_free (roots);ulist_free (tmp_ulist);
Re: [PATCH 08/10] btrfs: speedup checking for extent sharedness during fiemap
Quoted 48 lines · Qu Wenruo
Quoted 37 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
One of the most expensive tasks performed during fiemap is to check if an extent is shared. This task has two major steps:
1) Check if the data extent is shared. This implies checking the extent item in the extent tree, checking delayed references, etc. If we find the data extent is directly shared, we terminate immediately;
2) If the data extent is not directly shared (its extent item has a refcount of 1), then it may be shared if we have snapshots that share subtrees of the inode's subvolume b+tree. So we check if the leaf containing the file extent item is shared, then its parent node, then the parent node of the parent node, etc, until we reach the root node or we find one of them is shared - in which case we stop immediately.
During fiemap we process the extents of a file from left to right, from file offset 0 to eof. This means that we iterate b+tree leaves from left to right, and has the implication that we keep repeating that second step above several times for the same b+tree path of the inode's subvolume b+tree.
For example, if we have two file extent items in leaf X, and the path to leaf X is A -> B -> C -> X, then when we try to determine if the data extent referenced by the first extent item is shared, we check if the data extent is shared - if it's not, then we check if leaf X is shared, if not, then we check if node C is shared, if not, then check if node B is shared, if not than check if node A is shared. When we move to the next file extent item, after determining the data extent is not shared, we repeat the checks for X, C, B and A - doing all the expensive searches in the extent tree, delayed refs, etc. If we have thousands of tile extents, then we keep repeating the sharedness checks for the same paths over and over.
On a file that has no shared extents or only a small portion, it's easy to see that this scales terribly with the number of extents in the file and the sizes of the extent and subvolume b+trees.
Quoted 6 lines
This change eliminates the repeated sharedness check on extent buffers by caching the results of the last path used. The results can be used as long as no snapshots were created since they were cached (for not shared extent buffers) or no roots were dropped since they were cached (for shared extent buffers). This greatly reduces the time spent by fiemap for files with thousands of extents and/or large extent and subvolume b+trees.
Quoted 8 lines
This sounds pretty much like the existing
btrfs_backref_cacheis doing.It stores a map to speedup the backref lookup.
But a quick search didn't hit things like
btrfs_backref_edge()orbtrfs_backref_cache().Would it be possible to reuse the existing facility to do the same thing?
Nop, the existing facility is heavy and is meant to collect all backreferences for an extent.
It stores the backreferences in a rb tree, has to allocate memory for them, etc.
btrfs_check_shared() (now renamed to btrfs_is_data_extent_shared()), is a much simpler
thing than collecting backreferences - it just stops once it finds an extent is shared.
The cache I introduced is equally simple and the best fit for it, it has a fixed size, no need
to collect backreferences, allocate memory for them, add them to a red black tree, etc.
All it needs is a single path that doesn't change very often, and when it does it's just one
extent buffer at a time. Also, it allows to short circuit on 'not shared' results right at level 0,
which is very common and makes a huge difference.
Thanks.
Quoted 409 lines
Thanks,
QuQuoted 406 lines
Example performance test:
$ cat fiemap-perf-test.sh #!/bin/bashDEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f $DEV mount -o compress=lzo $DEV $MNT # 40G gives 327680 128K file extents (due to compression). xfs_io -f -c "pwrite -S 0xab -b 1M 0 40G" $MNT/foobar umount $MNT mount -o compress=lzo $DEV $MNT start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata not cached)" start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata cached)" umount $MNTBefore this patch:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 3597 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 2107 milliseconds (metadata cached)After this patch:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 1646 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 698 milliseconds (metadata cached)That's about 2.2x faster when no metadata is cached, and about 3x faster when all metadata is cached. On a real filesystem with many other files, data, directories, etc, the b+trees will be 2 or 3 levels higher, therefore this optimization will have a higher impact.
Several reports of a slow fiemap show up often, the two Link tags below refer to two recent reports of such slowness. This patch, together with the next ones in the series, is meant to address that.
Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/backref.c | 122 ++++++++++++++++++++++++++++++++++++++++- fs/btrfs/backref.h | 17 +++++- fs/btrfs/ctree.h | 18 ++++++ fs/btrfs/extent-tree.c | 10 +++- fs/btrfs/extent_io.c | 11 ++-- 5 files changed, 170 insertions(+), 8 deletions(-)fs/btrfs/backref.c +121
-1@@ -1511,6 +1511,105 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,return ret; } +/* +* The caller has joined a transaction or is holding a read lock on the +* fs_info->commit_root_sem semaphore, so no need to worry about the root's last +* snapshot field changing while updating or checking the cache. +*/ +static bool lookup_backref_shared_cache (struct btrfs_backref_shared_cache *cache, +struct btrfs_root *root, +u64 bytenr,int level,bool *is_shared) +{ +struct btrfs_backref_shared_cache_entry *entry; + +if (WARN_ON_ONCE (level >=BTRFS_MAX_LEVEL )) +return false; + +/* +* Level -1 is used for the data extent, which is not reliable to cache +* because its reference count can increase or decrease without us +* realizing. We cache results only for extent buffers that lead from +* the root node down to the leaf with the file extent item. +*/ +ASSERT (level >=0 ); + + entry = &cache->entries[level]; + +/* Unused cache entry or being used for some other extent buffer. */ +if (entry->bytenr != bytenr) +return false; + +/* +* We cached a false result, but the last snapshot generation of the +* root changed, so we now have a snapshot. Don't trust the result. +*/ +if (!entry->is_shared && + entry->gen !=btrfs_root_last_snapshot (&root->root_item)) +return false; + +/* +* If we cached a true result and the last generation used for dropping +* a root changed, we can not trust the result, because the dropped root +* could be a snapshot sharing this extent buffer. +*/ +if (entry->is_shared && + entry->gen !=btrfs_get_last_root_drop_gen (root->fs_info)) +return false; + + *is_shared = entry->is_shared; + +return true; +} + +/* +* The caller has joined a transaction or is holding a read lock on the +* fs_info->commit_root_sem semaphore, so no need to worry about the root's last +* snapshot field changing while updating or checking the cache. +*/ +static void store_backref_shared_cache (struct btrfs_backref_shared_cache *cache, +struct btrfs_root *root, +u64 bytenr,int level,bool is_shared) +{ +struct btrfs_backref_shared_cache_entry *entry; +u64 gen; + +if (WARN_ON_ONCE (level >=BTRFS_MAX_LEVEL )) +return ; + +/* +* Level -1 is used for the data extent, which is not reliable to cache +* because its reference count can increase or decrease without us +* realizing. We cache results only for extent buffers that lead from +* the root node down to the leaf with the file extent item. +*/ +ASSERT (level >=0 ); + +if (is_shared) + gen =btrfs_get_last_root_drop_gen (root->fs_info); +else + gen =btrfs_root_last_snapshot (&root->root_item); + + entry = &cache->entries[level]; + entry->bytenr = bytenr; + entry->is_shared = is_shared; + entry->gen = gen; + +/* +* If we found an extent buffer is shared, set the cache result for all +* extent buffers below it to true. As nodes in the path are COWed, +* their sharedness is moved to their children, and if a leaf is COWed, +* then the sharedness of a data extent becomes direct, the refcount of +* data extent is increased in the extent item at the extent tree. +*/ +if (is_shared) { +for (int i =0 ; i < level; i++) { + entry = &cache->entries[i]; + entry->is_shared = is_shared; + entry->gen = gen; + } + } +} +/** * Check if a data extent is shared or not. * @@ -1519,6 +1618,7 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* @bytenr: logical bytenr of the extent we are checking * @roots: list of roots this extent is shared among * @tmp: temporary list used for iteration +* @cache: a backref lookup result cache * * btrfs_is_data_extent_shared uses the backref walking code but will short * circuit as soon as it finds a root or inode that doesn't match the @@ -1532,7 +1632,8 @@ int btrfs_find_all_roots(struct btrfs_trans_handle *trans,* Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. */ int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp) +struct ulist *roots,struct ulist *tmp, +struct btrfs_backref_shared_cache *cache) {struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_trans_handle *trans;@@ -1545,6 +1646,7 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr, .inum = inum, .share_count = 0, }; +int level;ulist_init (roots);ulist_init (tmp);@@ -1561,22 +1663,40 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,btrfs_get_tree_mod_seq (fs_info, &elem); } +/* -1 means we are in the bytenr of the data extent. */ + level =-1 ;ULIST_ITER_INIT (&uiter);while (1 ) { +bool is_shared; +bool cached; + ret =find_parent_nodes (trans, fs_info, bytenr, elem.seq, tmp, roots,NULL , &shared, false);if (ret ==BACKREF_FOUND_SHARED ) {/* this is the only condition under which we return 1 */ ret =1 ; +if (level >=0 ) +store_backref_shared_cache (cache, root, bytenr, + level, true);break ; }if (ret <0 && ret != -ENOENT )break ; ret =0 ; +if (level >=0 ) +store_backref_shared_cache (cache, root, bytenr, + level, false); node =ulist_next (tmp, &uiter);if (!node)break ; bytenr = node->val; + level++; + cached =lookup_backref_shared_cache (cache, root, bytenr, level, + &is_shared); +if (cached) { + ret = is_shared ?1 :0 ; +break ; + } shared.share_count =0 ;cond_resched (); }fs/btrfs/backref.h +16
-1@@ -17,6 +17,20 @@ struct inode_fs_paths {struct btrfs_data_container *fspath; }; +struct btrfs_backref_shared_cache_entry { +u64 bytenr; +u64 gen; +bool is_shared; +}; + +struct btrfs_backref_shared_cache { +/* +* A path from a root to a leaf that has a file extent item pointing to +* a given data extent should never exceed the maximum b+tree heigth. +*/ +struct btrfs_backref_shared_cache_entry entries[BTRFS_MAX_LEVEL ]; +}; +typedef int (iterate_extent_inodes_t )(u64 inum,u64 offset,u64 root,void *ctx);@@ -63,7 +77,8 @@ int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,struct btrfs_inode_extref **ret_extref, u64 *found_off);int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, -struct ulist *roots,struct ulist *tmp); +struct ulist *roots,struct ulist *tmp, +struct btrfs_backref_shared_cache *cache);int __init btrfs_prelim_ref_init (void );void __cold btrfs_prelim_ref_exit (void );fs/btrfs/ctree.h +18
-0@@ -1095,6 +1095,13 @@ struct btrfs_fs_info {/* Updates are not protected by any lock */ struct btrfs_commit_stats commit_stats; +/* +* Last generation where we dropped a non-relocation root. +* Use btrfs_set_last_root_drop_gen() and btrfs_get_last_root_drop_gen() +* to change it and to read it, respectively. +*/ +u64 last_root_drop_gen; +/* * Annotations for transaction events (structures are empty when * compiled without lockdep). @@ -1119,6 +1126,17 @@ struct btrfs_fs_info {#endif }; +static inline void btrfs_set_last_root_drop_gen (struct btrfs_fs_info *fs_info, +u64 gen) +{ +WRITE_ONCE (fs_info->last_root_drop_gen, gen); +} + +static inline u64 btrfs_get_last_root_drop_gen (const struct btrfs_fs_info *fs_info) +{ +return READ_ONCE (fs_info->last_root_drop_gen); +} +static inline struct btrfs_fs_info *btrfs_sb (struct super_block *sb) {return sb->s_fs_info;fs/btrfs/extent-tree.c +9
-1@@ -5635,6 +5635,8 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,*/ int btrfs_drop_snapshot (struct btrfs_root *root,int update_ref,int for_reloc) { +const bool is_reloc_root = (root->root_key.objectid == +BTRFS_TREE_RELOC_OBJECTID );struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_path *path;struct btrfs_trans_handle *trans;@@ -5794,6 +5796,9 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) goto out_end_trans; } +if (!is_reloc_root) +btrfs_set_last_root_drop_gen (fs_info, trans->transid); +btrfs_end_transaction_throttle (trans);if (!for_reloc &&btrfs_need_cleaner_sleep (fs_info)) { btrfs_debug(fs_info,@@ -5828,7 +5833,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) goto out_end_trans; } -if (root->root_key.objectid !=BTRFS_TREE_RELOC_OBJECTID ) { +if (!is_reloc_root) { ret =btrfs_find_root (tree_root, &root->root_key, path,NULL ,NULL );if (ret <0 ) {@@ -5860,6 +5865,9 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)btrfs_put_root (root); root_dropped = true; out_end_trans: +if (!is_reloc_root) +btrfs_set_last_root_drop_gen (fs_info, trans->transid); +btrfs_end_transaction_throttle (trans); out_free:kfree (wc);fs/btrfs/extent_io.c +6
-5@@ -5519,6 +5519,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,struct btrfs_path *path;struct btrfs_root *root = inode->root;struct fiemap_cache cache = {0 }; +struct btrfs_backref_shared_cache *backref_cache;struct ulist *roots;struct ulist *tmp_ulist;int end =0 ;@@ -5526,13 +5527,11 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,u64 em_len =0 ;u64 em_end =0 ; + backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); path =btrfs_alloc_path (); -if (!path) -return -ENOMEM ; - roots =ulist_alloc (GFP_KERNEL ); tmp_ulist =ulist_alloc (GFP_KERNEL ); -if (!roots || !tmp_ulist) { +if (!backref_cache || !path || !roots || !tmp_ulist) { ret = -ENOMEM ; goto out_free_ulist; }@@ -5658,7 +5657,8 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,*/ ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), bytenr, roots, - tmp_ulist); + tmp_ulist, + backref_cache);if (ret <0 ) goto out_free;if (ret)@@ -5710,6 +5710,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, &cached_state); out_free_ulist: +kfree (backref_cache);btrfs_free_path (path);ulist_free (roots);ulist_free (tmp_ulist);
PATCH 9/10 btrfs: skip unnecessary extent buffer sharedness checks during fiemap
In reply to [email protected]From: Filipe Manana <[email protected]>
During fiemap, for each file extent we find, we must check if it's shared or not. The sharedness check starts by verifying if the extent is directly shared (its refcount in the extent tree is > 1), and if it is not directly shared, then we will check if every node in the subvolume b+tree leading from the root to the leaf that has the file extent item (in reverse order), is shared (through snapshots).
However this second step is not needed if our extent was created in a transaction more recent than the last transaction where a snapshot of the inode's root happened, because it can't be shared indirectly (through shared subtrees) without a snapshot created in a more recent transaction.
So grab the generation of the extent from the extent map and pass it to btrfs_is_data_extent_shared(), which will skip this second phase when the generation is more recent than the root's last snapshot value. Note that we skip this optimization if the extent map is the result of merging 2 or more extent maps, because in this case its generation is the maximum of the generations of all merged extent maps.
The fact the we use extent maps and they can be merged despite the underlying extents being distinct (different file extent items in the subvolume b+tree and different extent items in the extent b+tree), can result in some bugs when reporting shared extents. But this is a problem of the current implementation of fiemap relying on extent maps. One example where we get incorrect results is:
$ cat fiemap-bug.sh
#!/bin/bash DEV=/dev/sdj
MNT=/mnt/sdj
mkfs.btrfs -f $DEV
mount $DEV $MNT
# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents. xfs_io -f -c "pwrite -S 0xab 0 256K" \
-c "fsync" \
-c "pwrite -S 0xcd 256K 256K" \
-c "fsync" \
$MNT/foo
# Now clone only the second extent into another file. xfs_io -f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar
# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/foo
umount $MNT
Running the reproducer:
$ ./fiemap-bug.sh
wrote 262144/262144 bytes at offset 0
256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec)
wrote 262144/262144 bytes at offset 262144
256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec)
linked 262144/262144 bytes at offset 0
256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec) Filesystem type is: 9123683e
File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes)
ext: logical_offset: physical_offset: length: expected: flags:
0: 0.. 127: 3328.. 3455: 128: last,eof
/mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
This is z problem that existed before this change, and remains after this change, as it can't be easily fixed. The next patch in the series reworks fiemap to primarily use file extent items instead of extent maps (except for checking for delalloc ranges), with the goal of improving its scalability and performance, but it also ends up fixing this particular bug caused by extent map merging.
--- fs/btrfs/backref.c | 27 +++++++++++++++++++++------ fs/btrfs/backref.h | 1 + fs/btrfs/extent_io.c | 18 ++++++++++++++++-- 3 files changed, 38 insertions(+), 8 deletions(-)
fs/btrfs/backref.c +21 -6
@@ -1613,12 +1613,14 @@ static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache,/** * Check if a data extent is shared or not. * -* @root: root inode belongs to -* @inum: inode number of the inode whose extent we are checking -* @bytenr: logical bytenr of the extent we are checking -* @roots: list of roots this extent is shared among -* @tmp: temporary list used for iteration -* @cache: a backref lookup result cache +* @root: The root the inode belongs to. +* @inum: Number of the inode whose extent we are checking. +* @bytenr: Logical bytenr of the extent we are checking. +* @extent_gen: Generation of the extent (file extent item) or 0 if it is +* not known. +* @roots: List of roots this extent is shared among. +* @tmp: Temporary list used for iteration. +* @cache: A backref lookup result cache. * * btrfs_is_data_extent_shared uses the backref walking code but will short * circuit as soon as it finds a root or inode that doesn't match the
@@ -1632,6 +1634,7 @@ static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache,* Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. */ int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, +u64 extent_gen,struct ulist *roots,struct ulist *tmp,struct btrfs_backref_shared_cache *cache) {
@@ -1683,6 +1686,18 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,if (ret <0 && ret != -ENOENT )break ; ret =0 ; +/* +* If our data extent is not shared through reflinks and it was +* created in a generation after the last one used to create a +* snapshot of the inode's root, then it can not be shared +* indirectly through subtrees, as that can only happen with +* snapshots. In this case bail out, no need to check for the +* sharedness of extent buffers. +*/ +if (level ==-1 && + extent_gen >btrfs_root_last_snapshot (&root->root_item)) +break ; +if (level >=0 )store_backref_shared_cache (cache, root, bytenr, level, false);
fs/btrfs/backref.h +1 -0
@@ -77,6 +77,7 @@ int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,struct btrfs_inode_extref **ret_extref, u64 *found_off);int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, +u64 extent_gen,struct ulist *roots,struct ulist *tmp,struct btrfs_backref_shared_cache *cache);
fs/btrfs/extent_io.c +16 -2
@@ -5645,9 +5645,23 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, flags |= (FIEMAP_EXTENT_DELALLOC |FIEMAP_EXTENT_UNKNOWN ); }else if (fieinfo->fi_extents_max) { +u64 extent_gen;u64 bytenr = em->block_start - (em->start - em->orig_start); +/* +* If two extent maps are merged, then their generation +* is set to the maximum between their generations. +* Otherwise its generation matches the one we have in +* corresponding file extent item. If we have a merged +* extent map, don't use its generation to speedup the +* sharedness check below. +*/ +if (test_bit (EXTENT_FLAG_MERGED , &em->flags)) + extent_gen =0 ; +else + extent_gen = em->generation; +/* * As btrfs supports shared space, this information * can be exported to userspace tools via
@@ -5656,8 +5670,8 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,* lookup stuff. */ ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), - bytenr, roots, - tmp_ulist, + bytenr, extent_gen, + roots, tmp_ulist, backref_cache);if (ret <0 ) goto out_free;
Signature · 1 line
2.35.1
Re: [PATCH 09/10] btrfs: skip unnecessary extent buffer sharedness checks during fiemap
Quoted 77 lines · On Thu, Sep 01, 2022 at 02:18:29PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
During fiemap, for each file extent we find, we must check if it's shared or not. The sharedness check starts by verifying if the extent is directly shared (its refcount in the extent tree is > 1), and if it is not directly shared, then we will check if every node in the subvolume b+tree leading from the root to the leaf that has the file extent item (in reverse order), is shared (through snapshots).
However this second step is not needed if our extent was created in a transaction more recent than the last transaction where a snapshot of the inode's root happened, because it can't be shared indirectly (through shared subtrees) without a snapshot created in a more recent transaction.
So grab the generation of the extent from the extent map and pass it to
btrfs_is_data_extent_shared(), which will skip this second phase when the generation is more recent than the root's last snapshot value. Note that we skip this optimization if the extent map is the result of merging 2 or more extent maps, because in this case its generation is the maximum of the generations of all merged extent maps.The fact the we use extent maps and they can be merged despite the underlying extents being distinct (different file extent items in the subvolume b+tree and different extent items in the extent b+tree), can result in some bugs when reporting shared extents. But this is a problem of the current implementation of fiemap relying on extent maps. One example where we get incorrect results is:
$ cat fiemap-bug.sh #!/bin/bashDEV=/dev/sdj
MNT=/mnt/sdjmkfs.btrfs -f $DEV
mount $DEV $MNT# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents.xfs_io-f -c "pwrite -S0xab0 256K" \
-c "fsync" \
-c "pwrite -S0xcd256K 256K" \
-c "fsync" \
$MNT/foo# Now clone only the second extent into another file.
xfs_io-f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/fooumount $MNT
Running the reproducer:
$ ./fiemap-bug.sh wrote 262144/262144 bytes at offset 0 256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec) wrote 262144/262144 bytes at offset 262144 256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec) linked 262144/262144 bytes at offset 0 256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec)Filesystem type is: 9123683e File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes) ext: logical_offset: physical_offset: length: expected: flags: 0: 0.. 127: 3328.. 3455: 128: last,eof /mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
Quoted 8 lines
This is z problem that existed before this change, and remains after this change, as it can't be easily fixed. The next patch in the series reworks fiemap to primarily use file extent items instead of extent maps (except for checking for delalloc ranges), with the goal of improving its scalability and performance, but it also ends up fixing this particular bug caused by extent map merging.
Signed-off-by Filipe Manana <[email protected]>
Thanks,
Josef
Re: [PATCH 09/10] btrfs: skip unnecessary extent buffer sharedness checks during fiemap
In reply to [email protected]Quoted 9 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
During fiemap, for each file extent we find, we must check if it's shared or not. The sharedness check starts by verifying if the extent is directly shared (its refcount in the extent tree is > 1), and if it is not directly shared, then we will check if every node in the subvolume b+tree leading from the root to the leaf that has the file extent item (in reverse order), is shared (through snapshots).
Quoted 4 lines
However this second step is not needed if our extent was created in a transaction more recent than the last transaction where a snapshot of the inode's root happened, because it can't be shared indirectly (through shared subtrees) without a snapshot created in a more recent transaction.
This is a pretty awesome!
Quoted 7 lines
So grab the generation of the extent from the extent map and pass it to
btrfs_is_data_extent_shared(), which will skip this second phase when the generation is more recent than the root's last snapshot value. Note that we skip this optimization if the extent map is the result of merging 2 or more extent maps, because in this case its generation is the maximum of the generations of all merged extent maps.
And this pitfall also taken into consideration, even better.
Quoted 57 lines
The fact the we use extent maps and they can be merged despite the underlying extents being distinct (different file extent items in the subvolume b+tree and different extent items in the extent b+tree), can result in some bugs when reporting shared extents. But this is a problem of the current implementation of fiemap relying on extent maps. One example where we get incorrect results is:
$ cat fiemap-bug.sh #!/bin/bashDEV=/dev/sdj
MNT=/mnt/sdjmkfs.btrfs -f $DEV
mount $DEV $MNT# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents.xfs_io-f -c "pwrite -S0xab0 256K" \
-c "fsync" \
-c "pwrite -S0xcd256K 256K" \
-c "fsync" \
$MNT/foo# Now clone only the second extent into another file.
xfs_io-f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/fooumount $MNT
Running the reproducer:
$ ./fiemap-bug.sh wrote 262144/262144 bytes at offset 0 256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec) wrote 262144/262144 bytes at offset 262144 256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec) linked 262144/262144 bytes at offset 0 256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec)Filesystem type is: 9123683e File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes) ext: logical_offset: physical_offset: length: expected: flags: 0: 0.. 127: 3328.. 3455: 128: last,eof /mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
Quoted 8 lines
This is z problem that existed before this change, and remains after this change, as it can't be easily fixed. The next patch in the series reworks fiemap to primarily use file extent items instead of extent maps (except for checking for delalloc ranges), with the goal of improving its scalability and performance, but it also ends up fixing this particular bug caused by extent map merging.
Signed-off-by Filipe Manana <[email protected]>
Thanks,
Qu
Quoted 109 lines
--- fs/btrfs/backref.c | 27 +++++++++++++++++++++------ fs/btrfs/backref.h | 1 + fs/btrfs/extent_io.c | 18 ++++++++++++++++-- 3 files changed, 38 insertions(+), 8 deletions(-)fs/btrfs/backref.c +21
-6@@ -1613,12 +1613,14 @@ static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache,/** * Check if a data extent is shared or not. * -* @root: root inode belongs to -* @inum: inode number of the inode whose extent we are checking -* @bytenr: logical bytenr of the extent we are checking -* @roots: list of roots this extent is shared among -* @tmp: temporary list used for iteration -* @cache: a backref lookup result cache +* @root: The root the inode belongs to. +* @inum: Number of the inode whose extent we are checking. +* @bytenr: Logical bytenr of the extent we are checking. +* @extent_gen: Generation of the extent (file extent item) or 0 if it is +* not known. +* @roots: List of roots this extent is shared among. +* @tmp: Temporary list used for iteration. +* @cache: A backref lookup result cache. * * btrfs_is_data_extent_shared uses the backref walking code but will short * circuit as soon as it finds a root or inode that doesn't match the @@ -1632,6 +1634,7 @@ static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache,* Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. */ int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, +u64 extent_gen,struct ulist *roots,struct ulist *tmp,struct btrfs_backref_shared_cache *cache) {@@ -1683,6 +1686,18 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,if (ret <0 && ret != -ENOENT )break ; ret =0 ; +/* +* If our data extent is not shared through reflinks and it was +* created in a generation after the last one used to create a +* snapshot of the inode's root, then it can not be shared +* indirectly through subtrees, as that can only happen with +* snapshots. In this case bail out, no need to check for the +* sharedness of extent buffers. +*/ +if (level ==-1 && + extent_gen >btrfs_root_last_snapshot (&root->root_item)) +break ; +if (level >=0 )store_backref_shared_cache (cache, root, bytenr, level, false);fs/btrfs/backref.h +1
-0@@ -77,6 +77,7 @@ int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,struct btrfs_inode_extref **ret_extref, u64 *found_off);int btrfs_is_data_extent_shared (struct btrfs_root *root,u64 inum,u64 bytenr, +u64 extent_gen,struct ulist *roots,struct ulist *tmp,struct btrfs_backref_shared_cache *cache);fs/btrfs/extent_io.c +16
-2@@ -5645,9 +5645,23 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, flags |= (FIEMAP_EXTENT_DELALLOC |FIEMAP_EXTENT_UNKNOWN ); }else if (fieinfo->fi_extents_max) { +u64 extent_gen;u64 bytenr = em->block_start - (em->start - em->orig_start); +/* +* If two extent maps are merged, then their generation +* is set to the maximum between their generations. +* Otherwise its generation matches the one we have in +* corresponding file extent item. If we have a merged +* extent map, don't use its generation to speedup the +* sharedness check below. +*/ +if (test_bit (EXTENT_FLAG_MERGED , &em->flags)) + extent_gen =0 ; +else + extent_gen = em->generation; +/* * As btrfs supports shared space, this information * can be exported to userspace tools via @@ -5656,8 +5670,8 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,* lookup stuff. */ ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), - bytenr, roots, - tmp_ulist, + bytenr, extent_gen, + roots, tmp_ulist, backref_cache);if (ret <0 ) goto out_free;
PATCH 10/10 btrfs: make fiemap more efficient and accurate reporting extent sharedness
In reply to [email protected]From: Filipe Manana <[email protected]>
The current fiemap implementation does not scale very well with the number of extents a file has. This is both because the main algorithm to find out the extents has a high algorithmic complexity and because for each extent we have to check if it's shared. This second part, checking if an extent is shared, is significantly improved by the two previous patches in this patchset, while the first part is improved by this specific patch. Every now and then we get reports from users mentioning fiemap is too slow or even unusable for files with a very large number of extents, such as the two recent reports referred to by the Link tags at the bottom of this change log.
To understand why the part of finding which extents a file has is very inneficient, consider the example of doing a full ranged fiemap against a file that has over 100K extents (normal for example for a file with more than 10G of data and using compression, which limits the extent size to 128K). When we enter fiemap at extent_fiemap(), the following happens:
1) Before entering the main loop, we call get_extent_skip_holes() to get the first extent map. This leads us to btrfs_get_extent_fiemap(), which in turn calls btrfs_get_extent(), to find the first extent map that covers the file range [0, LLONG_MAX).
btrfs_get_extent() will first search the inode's extent map tree, to see if we have an extent map there that covers the range. If it does not find one, then it will search the inode's subvolume b+tree for a fitting file extent item. After finding the file extent item, it will allocate an extent map, fill it in with information extracted from the file extent item, and add it to the inode's extent map tree (which requires a search for insertion in the tree).
2) Then we enter the main loop at extent_fiemap(), emit the details of the extent, and call again get_extent_skip_holes(), with a start offset matching the end of the extent map we previously processed.
We end up at btrfs_get_extent() again, will search the extent map tree and then search the subvolume b+tree for a file extent item if we could not find an extent map in the extent tree. We allocate an extent map, fill it in with the details in the file extent item, and then insert it into the extent map tree (yet another search in this tree).
3) The second step is repeated over and over, until we have processed the whole file range. Each iteration ends at btrfs_get_extent(), which does a red black tree search on the extent map tree, then searches the subvolume b+tree, allocates an extent map and then does another search in the extent map tree in order to insert the extent map.
In the best scenario we have all the extent maps already in the extent tree, and so for each extent we do a single search on a red black tree, so we have a complexity of O(n log n).
In the worst scenario we don't have any extent map already loaded in the extent map tree, or have very few already there. In this case the complexity is much higher since we do:
- A red black tree search on the extent map tree, which has O(log n) complexity, initially very fast since the tree is empty or very small, but as we end up allocating extent maps and adding them to the tree when we don't find them there, each subsequent search on the tree gets slower, since it's getting bigger and bigger after each iteration.
- A search on the subvolume b+tree, also O(log n) complexity, but it has items for all inodes in the subvolume, not just items for our inode. Plus on a filesystem with concurrent operations on other inodes, we can block doing the search due to lock contention on b+tree nodes/leaves.
- Allocate an extent map - this can block, and can also fail if we are under serious memory pressure.
- Do another search on the extent maps red black tree, with the goal of inserting the extent map we just allocated. Again, after every iteration this tree is getting bigger by 1 element, so after many iterations the searches are slower and slower.
- We will not need the allocated extent map anymore, so it's pointless to add it to the extent map tree. It's just wasting time and memory.
In short we end up searching the extent map tree multiple times, on a tree that is growing bigger and bigger after each iteration. And besides that we visit the same leaf of the subvolume b+tree many times, since a leaf with the default size of 16K can easily have more than 200 file extent items.
This is very inneficient overall. This patch changes the algorithm to instead iterate over the subvolume b+tree, visiting each leaf only once, and only searching in the extent map tree for file ranges that have holes or prealloc extents, in order to figure out if we have delalloc there. It will never allocate an extent map and add it to the extent map tree. This is very similar to what was previously done for the lseek's hole and data seeking features.
Also, the current implementation relying on extent maps for figuring out which extents we have is not correct. This is because extent maps can be merged even if they represent different extents - we do this to minimize memory utilization and keep extent map trees smaller. For example if we have two extents that are contiguous on disk, once we load the two extent maps, they get merged into a single one - however if only one of the extents is shared, we end up reporting both as shared or both as not shared, which is incorrect.
This reproducer triggers that bug:
$ cat fiemap-bug.sh
#!/bin/bash DEV=/dev/sdj
MNT=/mnt/sdj
mkfs.btrfs -f $DEV
mount $DEV $MNT
# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents. xfs_io -f -c "pwrite -S 0xab 0 256K" \
-c "fsync" \
-c "pwrite -S 0xcd 256K 256K" \
-c "fsync" \
$MNT/foo
# Now clone only the second extent into another file. xfs_io -f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar
# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/foo
umount $MNT
Running the reproducer:
$ ./fiemap-bug.sh
wrote 262144/262144 bytes at offset 0
256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec)
wrote 262144/262144 bytes at offset 262144
256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec)
linked 262144/262144 bytes at offset 0
256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec) Filesystem type is: 9123683e
File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes)
ext: logical_offset: physical_offset: length: expected: flags:
0: 0.. 127: 3328.. 3455: 128: last,eof
/mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
This patch is part of a larger patchset that is comprised of the following patches:
btrfs: allow hole and data seeking to be interruptible
btrfs: make hole and data seeking a lot more efficient
btrfs: remove check for impossible block start for an extent map at fiemap
btrfs: remove zero length check when entering fiemap
btrfs: properly flush delalloc when entering fiemap
btrfs: allow fiemap to be interruptible
btrfs: rename btrfs_check_shared() to a more descriptive name
btrfs: speedup checking for extent sharedness during fiemap
btrfs: skip unnecessary extent buffer sharedness checks during fiemap
btrfs: make fiemap more efficient and accurate reporting extent sharedness
The patchset was tested on a machine running a non-debug kernel (Debian's default config) and compared the tests below on a branch without the patchset versus the same branch with the whole patchset applied.
The following test for a large compressed file without holes:
$ cat fiemap-perf-test.sh
#!/bin/bash DEV=/dev/sdi
MNT=/mnt/sdi
mkfs.btrfs -f $DEV
mount -o compress=lzo $DEV $MNT
# 40G gives 327680 128K file extents (due to compression).
xfs_io -f -c "pwrite -S 0xab -b 1M 0 20G" $MNT/foobar
umount $MNT
mount -o compress=lzo $DEV $MNT
start=$(date +%s%N)
filefrag $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "fiemap took $dur milliseconds (metadata not cached)"
start=$(date +%s%N)
filefrag $MNT/foobar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "fiemap took $dur milliseconds (metadata cached)"
umount $MNTBefore patchset:
$ ./fiemap-perf-test.sh
(...)
/mnt/sdi/foobar: 327680 extents found
fiemap took 3597 milliseconds (metadata not cached)
/mnt/sdi/foobar: 327680 extents found
fiemap took 2107 milliseconds (metadata cached)After patchset:
$ ./fiemap-perf-test.sh
(...)
/mnt/sdi/foobar: 327680 extents found
fiemap took 1214 milliseconds (metadata not cached)
/mnt/sdi/foobar: 327680 extents found
fiemap took 684 milliseconds (metadata cached)That's a speedup of about 3x for both cases (no metadata cached and all metadata cached).
The test provided by Pavel (first Link tag at the bottom), which uses files with a large number of holes, was also used to measure the gains, and it consists on a small C program and a shell script to invoke it. The C program is the following:
$ cat pavels-test.c
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <fcntl.h> #include <sys/stat.h>
#include <sys/time.h>
#include <sys/ioctl.h>
#include <linux/fs.h>
#include <linux/fiemap.h>
#define FILE_INTERVAL (1<<13) /* 8Kb */
long long interval(struct timeval t1, struct timeval t2)
{
long long val = 0;
val += (t2.tv_usec - t1.tv_usec);
val += (t2.tv_sec - t1.tv_sec) * 1000 * 1000;
return val;
}
int main(int argc, char **argv)
{
struct fiemap fiemap = {};
struct timeval t1, t2;
char data = 'a';
struct stat st;
int fd, off, file_size = FILE_INTERVAL;
if (argc != 3 && argc != 2) {
printf("usage: %s <path> [size]\n", argv[0]);
return 1;
}
if (argc == 3)
file_size = atoi(argv[2]);
if (file_size < FILE_INTERVAL)
file_size = FILE_INTERVAL;
file_size -= file_size % FILE_INTERVAL;
fd = open(argv[1], O_RDWR | O_CREAT | O_TRUNC, 0644);
if (fd < 0) {
perror("open");
return 1;
}
for (off = 0; off < file_size; off += FILE_INTERVAL) {
if (pwrite(fd, &data, 1, off) != 1) {
perror("pwrite");
close(fd);
return 1;
}
}
if (ftruncate(fd, file_size)) {
perror("ftruncate");
close(fd);
return 1;
}
if (fstat(fd, &st) < 0) {
perror("fstat");
close(fd);
return 1;
}
printf("size: %ld\n", st.st_size);
printf("actual size: %ld\n", st.st_blocks * 512);
fiemap.fm_length = FIEMAP_MAX_OFFSET;
gettimeofday(&t1, NULL);
if (ioctl(fd, FS_IOC_FIEMAP, &fiemap) < 0) {
perror("fiemap");
close(fd);
return 1;
}
gettimeofday(&t2, NULL);
printf("fiemap: fm_mapped_extents = %d\n",
fiemap.fm_mapped_extents);
printf("time = %lld us\n", interval(t1, t2));
close(fd);
return 0;
}
$ gcc -o pavels_test pavels_test.cAnd the wrapper shell script:
$ cat fiemap-pavels-test.sh
#!/bin/bash
DEV=/dev/sdi
MNT=/mnt/sdi
mkfs.btrfs -f -O no-holes $DEV
mount $DEV $MNT
echo
echo "*********** 256M ***********"
echo
./pavels-test $MNT/testfile $((1 << 28))
echo
./pavels-test $MNT/testfile $((1 << 28))
echo
echo "*********** 512M ***********"
echo
./pavels-test $MNT/testfile $((1 << 29))
echo
./pavels-test $MNT/testfile $((1 << 29))
echo
echo "*********** 1G ***********"
echo
./pavels-test $MNT/testfile $((1 << 30))
echo
./pavels-test $MNT/testfile $((1 << 30))
umount $MNTRunning his reproducer before applying the patchset:
*********** 256M ***********
size: 268435456
actual size: 134217728
fiemap: fm_mapped_extents = 32768
time = 4003133 us
size: 268435456
actual size: 134217728
fiemap: fm_mapped_extents = 32768
time = 4895330 us
*********** 512M ***********
size: 536870912
actual size: 268435456
fiemap: fm_mapped_extents = 65536
time = 30123675 us
size: 536870912
actual size: 268435456
fiemap: fm_mapped_extents = 65536
time = 33450934 us
*********** 1G ***********
size: 1073741824
actual size: 536870912
fiemap: fm_mapped_extents = 131072
time = 224924074 us
size: 1073741824
actual size: 536870912
fiemap: fm_mapped_extents = 131072
time = 217239242 usRunning it after applying the patchset:
*********** 256M ***********
size: 268435456
actual size: 134217728
fiemap: fm_mapped_extents = 32768
time = 29475 us
size: 268435456
actual size: 134217728
fiemap: fm_mapped_extents = 32768
time = 29307 us
*********** 512M ***********
size: 536870912
actual size: 268435456
fiemap: fm_mapped_extents = 65536
time = 58996 us
size: 536870912
actual size: 268435456
fiemap: fm_mapped_extents = 65536
time = 59115 us
*********** 1G ***********
size: 1073741824
actual size: 536870912
fiemap: fm_mapped_extents = 116251
time = 124141 us
size: 1073741824
actual size: 536870912
fiemap: fm_mapped_extents = 131072
time = 119387 usThe speedup is massive, both on the first fiemap call and on the second one as well, as his test creates files with many holes and small extents (every extent follows a hole and precedes another hole).
For the 256M file we go from 4 seconds down to 29 milliseconds in the first run, and then from 4.9 seconds down to 29 milliseconds again in the second run, a speedup of 138x and 169x, respectively.
For the 512M file we go from 30.1 seconds down to 59 milliseconds in the first run, and then from 33.5 seconds down to 59 milliseconds again in the second run, a speedup of 510x and 568x, respectively.
For the 1G file, we go from 225 seconds down to 124 milliseconds in the first run, and then from 217 seconds down to 119 milliseconds in the second run, a speedup of 1815x and 1824x, respectively.
--- fs/btrfs/ctree.h | 4 +- fs/btrfs/extent_io.c | 714 +++++++++++++++++++++++++++++-------------- fs/btrfs/file.c | 16 +- fs/btrfs/inode.c | 140 +-------- 4 files changed, 506 insertions(+), 368 deletions(-)
fs/btrfs/ctree.h +2 -2
@@ -3402,8 +3402,6 @@ unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio, u64 start, u64 end);int btrfs_check_data_csum (struct inode *inode,struct btrfs_bio *bbio,u32 bio_offset,struct page *page,u32 pgoff); -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len); noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,u64 *orig_start,u64 *orig_block_len,u64 *ram_bytes,bool strict);
@@ -3583,6 +3581,8 @@ int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end);int btrfs_check_nocow_lock (struct btrfs_inode *inode,loff_t pos,size_t *write_bytes);void btrfs_check_nocow_unlock (struct btrfs_inode *inode); +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret);/* tree-defrag.c */ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
fs/btrfs/extent_io.c +490 -224
791 lines not on this page · raw message
fs/btrfs/file.c +8 -8
@@ -3602,10 +3602,10 @@ static long btrfs_fallocate(struct file *file, int mode, }/* -* Helper for have_delalloc_in_range(). Find a subrange in a given range that -* has unflushed and/or flushing delalloc. There might be other adjacent -* subranges after the one it found, so have_delalloc_in_range() keeps looping -* while it gets adjacent subranges, and merging them together. +* Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range +* that has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so btrfs_find_delalloc_in_range() keeps +* looping while it gets adjacent subranges, and merging them together. */ static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end,u64 *delalloc_start_ret,u64 *delalloc_end_ret)
@@ -3740,8 +3740,8 @@ static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and * end offsets of the subrange. */ -static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, -u64 *delalloc_start_ret,u64 *delalloc_end_ret) +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) {u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize);u64 prev_delalloc_end =0 ;
@@ -3804,8 +3804,8 @@ static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,u64 delalloc_end;bool delalloc; - delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, - &delalloc_end); + delalloc =btrfs_find_delalloc_in_range (inode, start, end, + &delalloc_start, &delalloc_end);if (delalloc && whence ==SEEK_DATA ) { *start_ret = delalloc_start;return true;
fs/btrfs/inode.c +6 -134
154 lines not on this page · raw message
Signature · 1 line
2.35.1
Re: [PATCH 10/10] btrfs: make fiemap more efficient and accurate reporting extent sharedness
Quoted 1,474 lines · On Thu, Sep 01, 2022 at 02:18:30PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current fiemap implementation does not scale very well with the number of extents a file has. This is both because the main algorithm to find out the extents has a high algorithmic complexity and because for each extent we have to check if it's shared. This second part, checking if an extent is shared, is significantly improved by the two previous patches in this patchset, while the first part is improved by this specific patch. Every now and then we get reports from users mentioning fiemap is too slow or even unusable for files with a very large number of extents, such as the two recent reports referred to by the Link tags at the bottom of this change log.
To understand why the part of finding which extents a file has is very inneficient, consider the example of doing a full ranged fiemap against a file that has over 100K extents (normal for example for a file with more than 10G of data and using compression, which limits the extent size to 128K). When we enter fiemap at
extent_fiemap(), the following happens:1) Before entering the main loop, we call
get_extent_skip_holes()to get the first extent map. This leads us tobtrfs_get_extent_fiemap(), which in turn callsbtrfs_get_extent(), to find the first extent map that covers the file range [0,LLONG_MAX).
btrfs_get_extent()will first search the inode's extent map tree, to see if we have an extent map there that covers the range. If it does not find one, then it will search the inode's subvolume b+tree for a fitting file extent item. After finding the file extent item, it will allocate an extent map, fill it in with information extracted from the file extent item, and add it to the inode's extent map tree (which requires a search for insertion in the tree).2) Then we enter the main loop at
extent_fiemap(), emit the details of the extent, and call againget_extent_skip_holes(), with a start offset matching the end of the extent map we previously processed.We end up at
btrfs_get_extent()again, will search the extent map tree and then search the subvolume b+tree for a file extent item if we could not find an extent map in the extent tree. We allocate an extent map, fill it in with the details in the file extent item, and then insert it into the extent map tree (yet another search in this tree).3) The second step is repeated over and over, until we have processed the whole file range. Each iteration ends at
btrfs_get_extent(), which does a red black tree search on the extent map tree, then searches the subvolume b+tree, allocates an extent map and then does another search in the extent map tree in order to insert the extent map.In the best scenario we have all the extent maps already in the extent tree, and so for each extent we do a single search on a red black tree, so we have a complexity of O(n log n).
In the worst scenario we don't have any extent map already loaded in the extent map tree, or have very few already there. In this case the complexity is much higher since we do:
- A red black tree search on the extent map tree, which has O(log n) complexity, initially very fast since the tree is empty or very small, but as we end up allocating extent maps and adding them to the tree when we don't find them there, each subsequent search on the tree gets slower, since it's getting bigger and bigger after each iteration.
- A search on the subvolume b+tree, also O(log n) complexity, but it has items for all inodes in the subvolume, not just items for our inode. Plus on a filesystem with concurrent operations on other inodes, we can block doing the search due to lock contention on b+tree nodes/leaves.
- Allocate an extent map - this can block, and can also fail if we are under serious memory pressure.
- Do another search on the extent maps red black tree, with the goal of inserting the extent map we just allocated. Again, after every iteration this tree is getting bigger by 1 element, so after many iterations the searches are slower and slower.
- We will not need the allocated extent map anymore, so it's pointless to add it to the extent map tree. It's just wasting time and memory.
In short we end up searching the extent map tree multiple times, on a tree that is growing bigger and bigger after each iteration. And besides that we visit the same leaf of the subvolume b+tree many times, since a leaf with the default size of 16K can easily have more than 200 file extent items.
This is very inneficient overall. This patch changes the algorithm to instead iterate over the subvolume b+tree, visiting each leaf only once, and only searching in the extent map tree for file ranges that have holes or prealloc extents, in order to figure out if we have delalloc there. It will never allocate an extent map and add it to the extent map tree. This is very similar to what was previously done for the lseek's hole and data seeking features.
Also, the current implementation relying on extent maps for figuring out which extents we have is not correct. This is because extent maps can be merged even if they represent different extents - we do this to minimize memory utilization and keep extent map trees smaller. For example if we have two extents that are contiguous on disk, once we load the two extent maps, they get merged into a single one - however if only one of the extents is shared, we end up reporting both as shared or both as not shared, which is incorrect.
This reproducer triggers that bug:
$ cat fiemap-bug.sh #!/bin/bashDEV=/dev/sdj
MNT=/mnt/sdjmkfs.btrfs -f $DEV
mount $DEV $MNT# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents.xfs_io-f -c "pwrite -S0xab0 256K" \
-c "fsync" \
-c "pwrite -S0xcd256K 256K" \
-c "fsync" \
$MNT/foo# Now clone only the second extent into another file.
xfs_io-f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/fooumount $MNT
Running the reproducer:
$ ./fiemap-bug.sh wrote 262144/262144 bytes at offset 0 256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec) wrote 262144/262144 bytes at offset 262144 256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec) linked 262144/262144 bytes at offset 0 256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec)Filesystem type is: 9123683e File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes) ext: logical_offset: physical_offset: length: expected: flags: 0: 0.. 127: 3328.. 3455: 128: last,eof /mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
This patch is part of a larger patchset that is comprised of the following patches:
btrfs: allow hole and data seeking to be interruptible
btrfs: make hole and data seeking a lot more efficient
btrfs: remove check for impossible block start for an extent map at fiemap
btrfs: remove zero length check when entering fiemap
btrfs: properly flush delalloc when entering fiemap
btrfs: allow fiemap to be interruptible
btrfs: renamebtrfs_check_shared()to a more descriptive name
btrfs: speedup checking for extent sharedness during fiemap
btrfs: skip unnecessary extent buffer sharedness checks during fiemap
btrfs: make fiemap more efficient and accurate reporting extent sharednessThe patchset was tested on a machine running a non-debug kernel (Debian's default config) and compared the tests below on a branch without the patchset versus the same branch with the whole patchset applied.
The following test for a large compressed file without holes:
$ cat fiemap-perf-test.sh #!/bin/bashDEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f $DEV mount -o compress=lzo $DEV $MNT # 40G gives 327680 128K file extents (due to compression). xfs_io -f -c "pwrite -S 0xab -b 1M 0 20G" $MNT/foobar umount $MNT mount -o compress=lzo $DEV $MNT start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata not cached)" start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata cached)" umount $MNTBefore patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 3597 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 2107 milliseconds (metadata cached)After patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 1214 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 684 milliseconds (metadata cached)That's a speedup of about 3x for both cases (no metadata cached and all metadata cached).
The test provided by Pavel (first Link tag at the bottom), which uses files with a large number of holes, was also used to measure the gains, and it consists on a small C program and a shell script to invoke it. The C program is the following:
$ cat pavels-test.c #include <stdio.h> #include <unistd.h> #include <stdlib.h> #include <fcntl.h>#include <sys/stat.h> #include <sys/time.h> #include <sys/ioctl.h> #include <linux/fs.h> #include <linux/fiemap.h> #define FILE_INTERVAL (1<<13) /* 8Kb */ long long interval(struct timeval t1, struct timeval t2) { long long val = 0; val += (t2.tv_usec - t1.tv_usec); val += (t2.tv_sec - t1.tv_sec) * 1000 * 1000; return val; } int main(int argc, char **argv) { struct fiemap fiemap = {}; struct timeval t1, t2; char data = 'a'; struct stat st; int fd, off, file_size = FILE_INTERVAL; if (argc != 3 && argc != 2) { printf("usage: %s <path> [size]\n", argv[0]); return 1; } if (argc == 3) file_size = atoi(argv[2]); if (file_size < FILE_INTERVAL) file_size = FILE_INTERVAL; file_size -= file_size % FILE_INTERVAL; fd = open(argv[1], O_RDWR | O_CREAT | O_TRUNC, 0644); if (fd < 0) { perror("open"); return 1; } for (off = 0; off < file_size; off += FILE_INTERVAL) { if (pwrite(fd, &data, 1, off) != 1) { perror("pwrite"); close(fd); return 1; } } if (ftruncate(fd, file_size)) { perror("ftruncate"); close(fd); return 1; } if (fstat(fd, &st) < 0) { perror("fstat"); close(fd); return 1; } printf("size: %ld\n", st.st_size); printf("actual size: %ld\n", st.st_blocks * 512); fiemap.fm_length = FIEMAP_MAX_OFFSET; gettimeofday(&t1, NULL); if (ioctl(fd, FS_IOC_FIEMAP, &fiemap) < 0) { perror("fiemap"); close(fd); return 1; } gettimeofday(&t2, NULL); printf("fiemap: fm_mapped_extents = %d\n", fiemap.fm_mapped_extents); printf("time = %lld us\n", interval(t1, t2)); close(fd); return 0; } $ gcc -o pavels_test pavels_test.cAnd the wrapper shell script:
$ cat fiemap-pavels-test.sh#!/bin/bash DEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f -O no-holes $DEV mount $DEV $MNT echo echo "*********** 256M ***********" echo ./pavels-test $MNT/testfile $((1 << 28)) echo ./pavels-test $MNT/testfile $((1 << 28)) echo echo "*********** 512M ***********" echo ./pavels-test $MNT/testfile $((1 << 29)) echo ./pavels-test $MNT/testfile $((1 << 29)) echo echo "*********** 1G ***********" echo ./pavels-test $MNT/testfile $((1 << 30)) echo ./pavels-test $MNT/testfile $((1 << 30)) umount $MNTRunning his reproducer before applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4003133 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4895330 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 30123675 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 33450934 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 224924074 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 217239242 usRunning it after applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29475 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29307 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 58996 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 59115 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 116251 time = 124141 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 119387 usThe speedup is massive, both on the first fiemap call and on the second one as well, as his test creates files with many holes and small extents (every extent follows a hole and precedes another hole).
For the 256M file we go from 4 seconds down to 29 milliseconds in the first run, and then from 4.9 seconds down to 29 milliseconds again in the second run, a speedup of 138x and 169x, respectively.
For the 512M file we go from 30.1 seconds down to 59 milliseconds in the first run, and then from 33.5 seconds down to 59 milliseconds again in the second run, a speedup of 510x and 568x, respectively.
For the 1G file, we go from 225 seconds down to 124 milliseconds in the first run, and then from 217 seconds down to 119 milliseconds in the second run, a speedup of 1815x and 1824x, respectively.
Reported-by Pavel Tikhomirov <[email protected]>Reported-by Dominique MARTINET <[email protected]>Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/ctree.h | 4 +- fs/btrfs/extent_io.c | 714 +++++++++++++++++++++++++++++-------------- fs/btrfs/file.c | 16 +- fs/btrfs/inode.c | 140 +-------- 4 files changed, 506 insertions(+), 368 deletions(-)fs/btrfs/ctree.h +2
-2@@ -3402,8 +3402,6 @@ unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio, u64 start, u64 end);int btrfs_check_data_csum (struct inode *inode,struct btrfs_bio *bbio,u32 bio_offset,struct page *page,u32 pgoff); -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len); noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,u64 *orig_start,u64 *orig_block_len,u64 *ram_bytes,bool strict);@@ -3583,6 +3581,8 @@ int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end);int btrfs_check_nocow_lock (struct btrfs_inode *inode,loff_t pos,size_t *write_bytes);void btrfs_check_nocow_unlock (struct btrfs_inode *inode); +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret);/* tree-defrag.c */ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,fs/btrfs/extent_io.c +490
-224@@ -5353,42 +5353,6 @@ int try_release_extent_mapping(struct page *page, gfp_t mask)return try_release_extent_state (tree, page, mask); } -/* -* helper function for fiemap, which doesn't want to see any holes. -* This maps until we find something past 'last' -*/ -static struct extent_map *get_extent_skip_holes (struct btrfs_inode *inode, -u64 offset,u64 last) -{ -u64 sectorsize =btrfs_inode_sectorsize (inode); -struct extent_map *em; -u64 len; - -if (offset >= last) -return NULL ; - -while (1 ) { - len = last - offset; -if (len ==0 ) -break ; - len =ALIGN (len, sectorsize); - em =btrfs_get_extent_fiemap (inode, offset, len); -if (IS_ERR (em)) -return em; - -/* if this isn't a hole return it */ -if (em->block_start !=EXTENT_MAP_HOLE ) -return em; - -/* this is a hole, advance to the next extent */ - offset =extent_map_end (em); -free_extent_map (em); -if (offset >= last) -break ; - } -return NULL ; -} -/* * To cache previous fiemap extent * @@ -5418,6 +5382,9 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, {int ret =0 ; +/* Set at the end of extent_fiemap(). */ +ASSERT ((flags &FIEMAP_EXTENT_LAST ) ==0 ); +if (!cache->cached) goto assign;@@ -5446,11 +5413,10 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,*/ if (cache->offset + cache->len == offset && cache->phys + cache->len == phys && - (cache->flags & ~FIEMAP_EXTENT_LAST ) == - (flags & ~FIEMAP_EXTENT_LAST )) { + cache->flags == flags) { cache->len += len; cache->flags |= flags; - goto try_submit_last; +return 0 ; }/* Not mergeable, need to submit cached one */ @@ -5465,13 +5431,8 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, cache->phys = phys; cache->len = len; cache->flags = flags; -try_submit_last: -if (cache->flags &FIEMAP_EXTENT_LAST ) { - ret =fiemap_fill_next_extent (fieinfo, cache->offset, - cache->phys, cache->len, cache->flags); - cache->cached = false; - } -return ret; + +return 0 ; }/* @@ -5501,229 +5462,534 @@ static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,return ret; } -int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, -u64 start,u64 len) +static int fiemap_next_leaf_item (struct btrfs_inode *inode, +struct btrfs_path *path) { -int ret =0 ; -u64 off; -u64 max = start + len; -u32 flags =0 ; -u32 found_type; -u64 last; -u64 last_for_get_extent =0 ; -u64 disko =0 ; -u64 isize =i_size_read (&inode->vfs_inode); -struct btrfs_key found_key; -struct extent_map *em =NULL ; -struct extent_state *cached_state =NULL ; -struct btrfs_path *path; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; + + path->slots[0 ]++; +if (path->slots[0 ] <btrfs_header_nritems (path->nodes[0 ])) +return 0 ; + + ret =btrfs_next_leaf (inode->root, path); +if (ret !=0 ) +return ret; + +/* +* Don't bother with cloning if there are no more file extent items for +* our inode. +*/ +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid !=btrfs_ino (inode) || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; + +/* See the comment at fiemap_search_slot() about why we clone. */ + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot; + +return 0 ; +} + +/* +* Search for the first file extent item that starts at a given file offset or +* the one that starts immediately before that offset. +* Returns: 0 on success, < 0 on error, 1 if not found. +*/ +static int fiemap_search_slot (struct btrfs_inode *inode,struct btrfs_path *path, +u64 file_offset) +{ +const u64 ino =btrfs_ino (inode);struct btrfs_root *root = inode->root; -struct fiemap_cache cache = {0 }; -struct btrfs_backref_shared_cache *backref_cache; -struct ulist *roots; -struct ulist *tmp_ulist; -int end =0 ; -u64 em_start =0 ; -u64 em_len =0 ; -u64 em_end =0 ; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; - backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); - path =btrfs_alloc_path (); - roots =ulist_alloc (GFP_KERNEL ); - tmp_ulist =ulist_alloc (GFP_KERNEL ); -if (!backref_cache || !path || !roots || !tmp_ulist) { - ret = -ENOMEM ; - goto out_free_ulist; + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = file_offset; + + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) +return ret; + +if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } + +if (path->slots[0 ] >=btrfs_header_nritems (path->nodes[0 ])) { + ret =btrfs_next_leaf (root, path); +if (ret !=0 ) +return ret; + +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; }/* -* We can't initialize that to 'start' as this could miss extents due -* to extent item merging +* We clone the leaf and use it during fiemap. This is because while +* using the leaf we do expensive things like checking if an extent is +* shared, which can take a long time. In order to prevent blocking +* other tasks for too long, we use a clone of the leaf. We have locked +* the file range in the inode's io tree, so we know none of our file +* extent items can change. This way we avoid blocking other tasks that +* want to insert items for other inodes in the same leaf or b+tree +* rebalance operations (triggered for example when someone is trying +* to push items into this leaf when trying to insert an item in a +* neighbour leaf). +* We also need the private clone because holding a read lock on an +* extent buffer of the subvolume's b+tree will make lockdep unhappy +* when we call fiemap_fill_next_extent(), because that may cause a page +* fault when filling the user space buffer with fiemap data. */ - off =0 ; - start =round_down (start,btrfs_inode_sectorsize (inode)); - len =round_up (max,btrfs_inode_sectorsize (inode)) - start; + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot; + +return 0 ; +} + +/* +* Process a range which is a hole or a prealloc extent in the inode's subvolume +* btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc +* extent. The end offset (@end) is inclusive. +*/ +static int fiemap_process_hole (struct btrfs_inode *inode, +struct fiemap_extent_info *fieinfo, +struct fiemap_cache *cache, +struct btrfs_backref_shared_cache *backref_cache, +u64 disk_bytenr,u64 extent_offset, +u64 extent_gen, +struct ulist *roots,struct ulist *tmp_ulist, +u64 start,u64 end) +{ +const u64 i_size =i_size_read (&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +u64 cur_offset = start; +u64 last_delalloc_end =0 ; +u32 prealloc_flags =FIEMAP_EXTENT_UNWRITTEN ; +bool checked_extent_shared = false; +int ret;/* -* lookup the last file extent. We're not using i_size here -* because there might be preallocation past i_size +* There can be no delalloc past i_size, so don't waste time looking for +* it beyond i_size. */ - ret =btrfs_lookup_file_extent (NULL , root, path,btrfs_ino (inode),-1 , -0 ); -if (ret <0 ) { - goto out_free_ulist; - }else { -WARN_ON (!ret); -if (ret ==1 ) - ret =0 ; - } +while (cur_offset < end && cur_offset < i_size) { +u64 delalloc_start; +u64 delalloc_end; +u64 prealloc_start; +u64 prealloc_len =0 ; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; - path->slots[0 ]--; -btrfs_item_key_to_cpu (path->nodes[0 ], &found_key, path->slots[0 ]); - found_type = found_key.type; - -/* No extents, but there might be delalloc bits */ -if (found_key.objectid !=btrfs_ino (inode) || - found_type !=BTRFS_EXTENT_DATA_KEY ) { -/* have to trust i_size as the end */ - last = (u64)-1 ; - last_for_get_extent = isize; - }else {/* -* remember the start of the last extent. There are a -* bunch of different factors that go into the length of the -* extent, so its much less complex to remember where it started +* If this is a prealloc extent we have to report every section +* of it that has no delalloc. */ - last = found_key.offset; - last_for_get_extent = last +1 ; +if (disk_bytenr !=0 ) { +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = delalloc_start - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = delalloc_start - prealloc_start; + } + } + +if (prealloc_len >0 ) { +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + + checked_extent_shared = true; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + extent_offset += prealloc_len; + } + + ret =emit_fiemap_extent (fieinfo, cache, delalloc_start,0 , + delalloc_end +1 - delalloc_start, +FIEMAP_EXTENT_DELALLOC | +FIEMAP_EXTENT_UNKNOWN ); +if (ret) +return ret; + + last_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; + extent_offset += cur_offset - delalloc_start; +cond_resched (); + } + +/* +* Either we found no delalloc for the whole prealloc extent or we have +* a prealloc extent that spans i_size or starts at or after i_size. +*/ +if (disk_bytenr !=0 && last_delalloc_end < end) { +u64 prealloc_start; +u64 prealloc_len; + +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = end +1 - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = end +1 - prealloc_start; + } + +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + } + +return 0 ; +} + +static int fiemap_find_last_extent_offset (struct btrfs_inode *inode, +struct btrfs_path *path, +u64 *last_extent_end_ret) +{ +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct extent_buffer *leaf; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 disk_bytenr; +int ret; + +/* +* Lookup the last file extent. We're not using i_size here because +* there might be preallocation past i_size. +*/ + ret =btrfs_lookup_file_extent (NULL , root, path, ino, (u64)-1 ,0 ); +/* There can't be a file extent item at offset (u64)-1 */ +ASSERT (ret !=0 ); +if (ret <0 ) +return ret; + +/* +* For a non-existing key, btrfs_search_slot() always leaves us at a +* slot > 0, except if the btree is empty, which is impossible because +* at least it has the inode item for this inode and all the items for +* the root inode 256. +*/ +ASSERT (path->slots[0 ] >0 ); + path->slots[0 ]--; + leaf = path->nodes[0 ]; +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) { +/* No file extent items in the subvolume tree. */ + *last_extent_end_ret =0 ; +return 0 ; } -btrfs_release_path (path);/* -* we might have some extents allocated but more delalloc past those -* extents. so, we trust isize unless the start of the last extent is -* beyond isize +* For an inline extent, the disk_bytenr is where inline data starts at, +* so first check if we have an inline extent item before checking if we +* have an implicit hole (disk_bytenr == 0). */ -if (last < isize) { - last = (u64)-1 ; - last_for_get_extent = isize; + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); +if (btrfs_file_extent_type (leaf, ei) ==BTRFS_FILE_EXTENT_INLINE ) { + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; } -lock_extent_bits (&inode->io_tree, start, start + len -1 , - &cached_state); +/* +* Find the last file extent item that is not a hole (when NO_HOLES is +* not enabled). This should take at most 2 iterations in the worst +* case: we have one hole file extent item at slot 0 of a leaf and +* another hole file extent item as the last item in the previous leaf. +* This is because we merge file extent items that represent holes. +*/ + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +while (disk_bytenr ==0 ) { + ret =btrfs_previous_item (root, path, ino,BTRFS_EXTENT_DATA_KEY ); +if (ret <0 ) { +return ret; + }else if (ret >0 ) { +/* No file extent items that are not holes. */ + *last_extent_end_ret =0 ; +return 0 ; + } + leaf = path->nodes[0 ]; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); + } - em =get_extent_skip_holes (inode, start, last_for_get_extent); -if (!em) - goto out; -if (IS_ERR (em)) { - ret =PTR_ERR (em); + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; +} + +int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, +u64 start,u64 len) +{ +const u64 ino =btrfs_ino (inode); +struct extent_state *cached_state =NULL ; +struct btrfs_path *path; +struct btrfs_root *root = inode->root; +struct fiemap_cache cache = {0 }; +struct btrfs_backref_shared_cache *backref_cache; +struct ulist *roots; +struct ulist *tmp_ulist; +u64 last_extent_end; +u64 prev_extent_end; +u64 lockstart; +u64 lockend; +bool stopped = false; +int ret; + + backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); + path =btrfs_alloc_path (); + roots =ulist_alloc (GFP_KERNEL ); + tmp_ulist =ulist_alloc (GFP_KERNEL ); +if (!backref_cache || !path || !roots || !tmp_ulist) { + ret = -ENOMEM ; goto out; } -while (!end) { -u64 offset_in_extent =0 ; + lockstart =round_down (start,btrfs_inode_sectorsize (inode)); + lockend =round_up (start + len,btrfs_inode_sectorsize (inode)); + prev_extent_end = lockstart; -/* break if the extent we found is outside the range */ -if (em->start >= max ||extent_map_end (em) < off) -break ; +lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); -/* -* get_extent may return an extent that starts before our -* requested range. We have to make sure the ranges -* we return to fiemap always move forward and don't -* overlap, so adjust the offsets here -*/ - em_start =max (em->start, off); + ret =fiemap_find_last_extent_offset (inode, path, &last_extent_end); +if (ret <0 ) + goto out_unlock; +btrfs_release_path (path); + path->reada =READA_FORWARD ; + ret =fiemap_search_slot (inode, path, lockstart); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) {/* -* record the offset from the start of the extent -* for adjusting the disk offset below. Only do this if the -* extent isn't compressed since our in ram offset may be past -* what we have actually allocated on disk. +* No file extent item found, but we may have delalloc between +* the current offset and i_size. So check for that. */ -if (!test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - offset_in_extent = em_start - em->start; - em_end =extent_map_end (em); - em_len = em_end - em_start; - flags =0 ; -if (em->block_start <EXTENT_MAP_LAST_BYTE ) - disko = em->block_start + offset_in_extent; -else - disko =0 ; + ret =0 ; + goto check_eof_delalloc; + } + +while (prev_extent_end < lockend) { +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 extent_end; +u64 extent_len; +u64 extent_offset =0 ; +u64 extent_gen; +u64 disk_bytenr =0 ; +u64 flags =0 ; +int extent_type; +u8 compression; + +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +break ; + + extent_end =btrfs_file_extent_end (path);/* -* bump off for our next call to get_extent +* The first iteration can leave us at an extent item that ends +* before our range's start. Move to the next item. */ - off =extent_map_end (em); -if (off >= max) - end =1 ; - -if (em->block_start ==EXTENT_MAP_INLINE ) { - flags |= (FIEMAP_EXTENT_DATA_INLINE | -FIEMAP_EXTENT_NOT_ALIGNED ); - }else if (em->block_start ==EXTENT_MAP_DELALLOC ) { - flags |= (FIEMAP_EXTENT_DELALLOC | -FIEMAP_EXTENT_UNKNOWN ); - }else if (fieinfo->fi_extents_max) { -u64 extent_gen; -u64 bytenr = em->block_start - - (em->start - em->orig_start); +if (extent_end <= lockstart) + goto next_item; -/* -* If two extent maps are merged, then their generation -* is set to the maximum between their generations. -* Otherwise its generation matches the one we have in -* corresponding file extent item. If we have a merged -* extent map, don't use its generation to speedup the -* sharedness check below. -*/ -if (test_bit (EXTENT_FLAG_MERGED , &em->flags)) - extent_gen =0 ; -else - extent_gen = em->generation; +/* We have in implicit hole (NO_HOLES feature enabled). */ +if (prev_extent_end < key.offset) { +const u64 range_end =min (key.offset, lockend) -1 ; -/* -* As btrfs supports shared space, this information -* can be exported to userspace tools via -* flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 -* then we're just getting a count and we can skip the -* lookup stuff. -*/ - ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), - bytenr, extent_gen, - roots, tmp_ulist, - backref_cache); -if (ret <0 ) - goto out_free; -if (ret) - flags |=FIEMAP_EXTENT_SHARED ; - ret =0 ; - } -if (test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - flags |=FIEMAP_EXTENT_ENCODED ; -if (test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) - flags |=FIEMAP_EXTENT_UNWRITTEN ; + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + prev_extent_end, range_end); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; + } -free_extent_map (em); - em =NULL ; -if ((em_start >= last) || em_len == (u64)-1 || - (last == (u64)-1 && isize <= em_end)) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; +/* We've reached the end of the fiemap range, stop. */ +if (key.offset >= lockend) { + stopped = true; +break ; + } } -/* now scan forward to see if this is really the last extent. */ - em =get_extent_skip_holes (inode, off, last_for_get_extent); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - goto out; + extent_len = extent_end - key.offset; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + compression =btrfs_file_extent_compression (leaf, ei); + extent_type =btrfs_file_extent_type (leaf, ei); + extent_gen =btrfs_file_extent_generation (leaf, ei); + +if (extent_type !=BTRFS_FILE_EXTENT_INLINE ) { + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +if (compression ==BTRFS_COMPRESS_NONE ) + extent_offset =btrfs_file_extent_offset (leaf, ei); } -if (!em) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; + +if (compression !=BTRFS_COMPRESS_NONE ) + flags |=FIEMAP_EXTENT_ENCODED ; + +if (extent_type ==BTRFS_FILE_EXTENT_INLINE ) { + flags |=FIEMAP_EXTENT_DATA_INLINE ; + flags |=FIEMAP_EXTENT_NOT_ALIGNED ; + ret =emit_fiemap_extent (fieinfo, &cache, key.offset,0 , + extent_len, flags); + }else if (extent_type ==BTRFS_FILE_EXTENT_PREALLOC ) { + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache, + disk_bytenr, extent_offset, + extent_gen, roots, tmp_ulist, + key.offset, extent_end -1 ); + }else if (disk_bytenr ==0 ) { +/* We have an explicit hole. */ + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + key.offset, extent_end -1 ); + }else { +/* We have a regular extent. */ +if (fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (root, ino, + disk_bytenr, + extent_gen, + roots, + tmp_ulist, + backref_cache); +if (ret <0 ) + goto out_unlock; +else if (ret >0 ) + flags |=FIEMAP_EXTENT_SHARED ; + } + + ret =emit_fiemap_extent (fieinfo, &cache, key.offset, + disk_bytenr + extent_offset, + extent_len, flags); } - ret =emit_fiemap_extent (fieinfo, &cache, em_start, disko, - em_len, flags); -if (ret) { -if (ret ==1 ) - ret =0 ; - goto out_free; + +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; } + prev_extent_end = extent_end; +next_item:if (fatal_signal_pending (current)) { ret = -EINTR ; - goto out_free; + goto out_unlock; } + + ret =fiemap_next_leaf_item (inode, path); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* No more file extent items for this inode. */ +break ; + } +cond_resched (); } -out_free: -if (!ret) - ret =emit_last_fiemap_cache (fieinfo, &cache); -free_extent_map (em); -out: -unlock_extent_cached (&inode->io_tree, start, start + len -1 , - &cached_state); -out_free_ulist: +check_eof_delalloc: +/* +* Release (and free) the path before emitting any final entries to +* fiemap_fill_next_extent() to keep lockdep happy. This is because +* once we find no more file extent items exist, we may have a +* non-cloned leaf, and fiemap_fill_next_extent() can trigger page +* faults when copying data to the user space buffer. +*/ +btrfs_free_path (path); + path =NULL ; + +if (!stopped && prev_extent_end < lockend) { + ret =fiemap_process_hole (inode, fieinfo, &cache, backref_cache, +0 ,0 ,0 , roots, tmp_ulist, + prev_extent_end, lockend -1 ); +if (ret <0 ) + goto out_unlock; + prev_extent_end = lockend; + } + +if (cache.cached && cache.offset + cache.len >= last_extent_end) { +const u64 i_size =i_size_read (&inode->vfs_inode); + +if (prev_extent_end < i_size) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, + prev_extent_end, + i_size -1 , + &delalloc_start, + &delalloc_end); +if (!delalloc) + cache.flags |=FIEMAP_EXTENT_LAST ; + }else { + cache.flags |=FIEMAP_EXTENT_LAST ; + } + } + + ret =emit_last_fiemap_cache (fieinfo, &cache); + +out_unlock: +unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); +out:kfree (backref_cache);btrfs_free_path (path);ulist_free (roots);fs/btrfs/file.c +8
-8@@ -3602,10 +3602,10 @@ static long btrfs_fallocate(struct file *file, int mode, }/* -* Helper for have_delalloc_in_range(). Find a subrange in a given range that -* has unflushed and/or flushing delalloc. There might be other adjacent -* subranges after the one it found, so have_delalloc_in_range() keeps looping -* while it gets adjacent subranges, and merging them together. +* Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range +* that has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so btrfs_find_delalloc_in_range() keeps +* looping while it gets adjacent subranges, and merging them together. */ static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end,u64 *delalloc_start_ret,u64 *delalloc_end_ret)@@ -3740,8 +3740,8 @@ static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and * end offsets of the subrange. */ -static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, -u64 *delalloc_start_ret,u64 *delalloc_end_ret) +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) {u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize);u64 prev_delalloc_end =0 ;@@ -3804,8 +3804,8 @@ static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,u64 delalloc_end;bool delalloc; - delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, - &delalloc_end); + delalloc =btrfs_find_delalloc_in_range (inode, start, end, + &delalloc_start, &delalloc_end);if (delalloc && whence ==SEEK_DATA ) { *start_ret = delalloc_start;return true;fs/btrfs/inode.c +5
-130@@ -7064,133 +7064,6 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,return em; } -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len) -{ -struct extent_map *em; -struct extent_map *hole_em =NULL ; -u64 delalloc_start = start; -u64 end; -u64 delalloc_len; -u64 delalloc_end; -int err =0 ; - - em =btrfs_get_extent (inode,NULL ,0 , start, len); -if (IS_ERR (em)) -return em; -/* -* If our em maps to: -* - a hole or -* - a pre-alloc extent, -* there might actually be delalloc bytes behind it. -*/ -if (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) -return em; -else - hole_em = em; - -/* check to see if we've wrapped (len == -1 or similar) */ - end = start + len; -if (end < start) - end = (u64)-1 ; -else - end -=1 ; - - em =NULL ; - -/* ok, we didn't find anything, lets look for delalloc */ - delalloc_len =count_range_bits (&inode->io_tree, &delalloc_start, - end, len,EXTENT_DELALLOC ,1 ); - delalloc_end = delalloc_start + delalloc_len; -if (delalloc_end < delalloc_start) - delalloc_end = (u64)-1 ; - -/* -* We didn't find anything useful, return the original results from -* get_extent() -*/ -if (delalloc_start > end || delalloc_end <= start) { - em = hole_em; - hole_em =NULL ; - goto out; - } - -/* -* Adjust the delalloc_start to make sure it doesn't go backwards from -* the start they passed in -*/ - delalloc_start =max (start, delalloc_start); - delalloc_len = delalloc_end - delalloc_start; - -if (delalloc_len >0 ) { -u64 hole_start; -u64 hole_len; -const u64 hole_end =extent_map_end (hole_em); - - em =alloc_extent_map (); -if (!em) { - err = -ENOMEM ; - goto out; - } - -ASSERT (hole_em); -/* -* When btrfs_get_extent can't find anything it returns one -* huge hole -* -* Make sure what it found really fits our range, and adjust to -* make sure it is based on the start from the caller -*/ -if (hole_end <= start || hole_em->start > end) { -free_extent_map (hole_em); - hole_em =NULL ; - }else { - hole_start =max (hole_em->start, start); - hole_len = hole_end - hole_start; - } - -if (hole_em && delalloc_start > hole_start) { -/* -* Our hole starts before our delalloc, so we have to -* return just the parts of the hole that go until the -* delalloc starts -*/ - em->len =min (hole_len, delalloc_start - hole_start); - em->start = hole_start; - em->orig_start = hole_start; -/* -* Don't adjust block start at all, it is fixed at -* EXTENT_MAP_HOLE -*/ - em->block_start = hole_em->block_start; - em->block_len = hole_len; -if (test_bit (EXTENT_FLAG_PREALLOC , &hole_em->flags)) -set_bit (EXTENT_FLAG_PREALLOC , &em->flags); - }else { -/* -* Hole is out of passed range or it starts after -* delalloc range -*/ - em->start = delalloc_start; - em->len = delalloc_len; - em->orig_start = delalloc_start; - em->block_start =EXTENT_MAP_DELALLOC ; - em->block_len = delalloc_len; - } - }else { -return hole_em; - } -out: - -free_extent_map (hole_em); -if (err) { -free_extent_map (em); -return ERR_PTR (err); - } -return em; -} - static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, const u64 start, const u64 len,@@ -8265,15 +8138,14 @@ static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,* in the compression of data (in an async thread) and will return * before the compression is done and writeback is started. A second * filemap_fdatawrite_range() is needed to wait for the compression to -* complete and writeback to start. Without this, our user is very -* likely to get stale results, because the extents and extent maps for -* delalloc regions are only allocated when writeback starts. +* complete and writeback to start. We also need to wait for ordered +* extents to complete, because our fiemap implementation uses mainly +* file extent items to list the extents, searching for extent maps +* only for file ranges with holes or prealloc extents to figure out +* if we have delalloc in those ranges.
fs/btrfs/inode.c static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, · line 8151
Quoted 10 lines
fs/btrfs/inode.c +1
-4*/ if (fieinfo->fi_flags &FIEMAP_FLAG_SYNC ) { - ret =btrfs_fdatawrite_range (inode,0 ,LLONG_MAX ); -if (ret) -return ret; - ret =filemap_fdatawait_range (inode->i_mapping,0 ,LLONG_MAX ); + ret =btrfs_wait_ordered_range (inode,0 ,LLONG_MAX );if (ret)return ret; }
Hmm this bit should be in "btrfs: properly flush delalloc when entering fiemap" instead. Thanks,
Josef
Re: [PATCH 10/10] btrfs: make fiemap more efficient and accurate reporting extent sharedness
Quoted 1,481 lines · Josef Bacik
Quoted 1,479 lines · On Thu, Sep 01, 2022 at 02:18:30PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current fiemap implementation does not scale very well with the number of extents a file has. This is both because the main algorithm to find out the extents has a high algorithmic complexity and because for each extent we have to check if it's shared. This second part, checking if an extent is shared, is significantly improved by the two previous patches in this patchset, while the first part is improved by this specific patch. Every now and then we get reports from users mentioning fiemap is too slow or even unusable for files with a very large number of extents, such as the two recent reports referred to by the Link tags at the bottom of this change log.
To understand why the part of finding which extents a file has is very inneficient, consider the example of doing a full ranged fiemap against a file that has over 100K extents (normal for example for a file with more than 10G of data and using compression, which limits the extent size to 128K). When we enter fiemap at
extent_fiemap(), the following happens:1) Before entering the main loop, we call
get_extent_skip_holes()to get the first extent map. This leads us tobtrfs_get_extent_fiemap(), which in turn callsbtrfs_get_extent(), to find the first extent map that covers the file range [0,LLONG_MAX).
btrfs_get_extent()will first search the inode's extent map tree, to see if we have an extent map there that covers the range. If it does not find one, then it will search the inode's subvolume b+tree for a fitting file extent item. After finding the file extent item, it will allocate an extent map, fill it in with information extracted from the file extent item, and add it to the inode's extent map tree (which requires a search for insertion in the tree).2) Then we enter the main loop at
extent_fiemap(), emit the details of the extent, and call againget_extent_skip_holes(), with a start offset matching the end of the extent map we previously processed.We end up at
btrfs_get_extent()again, will search the extent map tree and then search the subvolume b+tree for a file extent item if we could not find an extent map in the extent tree. We allocate an extent map, fill it in with the details in the file extent item, and then insert it into the extent map tree (yet another search in this tree).3) The second step is repeated over and over, until we have processed the whole file range. Each iteration ends at
btrfs_get_extent(), which does a red black tree search on the extent map tree, then searches the subvolume b+tree, allocates an extent map and then does another search in the extent map tree in order to insert the extent map.In the best scenario we have all the extent maps already in the extent tree, and so for each extent we do a single search on a red black tree, so we have a complexity of O(n log n).
In the worst scenario we don't have any extent map already loaded in the extent map tree, or have very few already there. In this case the complexity is much higher since we do:
- A red black tree search on the extent map tree, which has O(log n) complexity, initially very fast since the tree is empty or very small, but as we end up allocating extent maps and adding them to the tree when we don't find them there, each subsequent search on the tree gets slower, since it's getting bigger and bigger after each iteration.
- A search on the subvolume b+tree, also O(log n) complexity, but it has items for all inodes in the subvolume, not just items for our inode. Plus on a filesystem with concurrent operations on other inodes, we can block doing the search due to lock contention on b+tree nodes/leaves.
- Allocate an extent map - this can block, and can also fail if we are under serious memory pressure.
- Do another search on the extent maps red black tree, with the goal of inserting the extent map we just allocated. Again, after every iteration this tree is getting bigger by 1 element, so after many iterations the searches are slower and slower.
- We will not need the allocated extent map anymore, so it's pointless to add it to the extent map tree. It's just wasting time and memory.
In short we end up searching the extent map tree multiple times, on a tree that is growing bigger and bigger after each iteration. And besides that we visit the same leaf of the subvolume b+tree many times, since a leaf with the default size of 16K can easily have more than 200 file extent items.
This is very inneficient overall. This patch changes the algorithm to instead iterate over the subvolume b+tree, visiting each leaf only once, and only searching in the extent map tree for file ranges that have holes or prealloc extents, in order to figure out if we have delalloc there. It will never allocate an extent map and add it to the extent map tree. This is very similar to what was previously done for the lseek's hole and data seeking features.
Also, the current implementation relying on extent maps for figuring out which extents we have is not correct. This is because extent maps can be merged even if they represent different extents - we do this to minimize memory utilization and keep extent map trees smaller. For example if we have two extents that are contiguous on disk, once we load the two extent maps, they get merged into a single one - however if only one of the extents is shared, we end up reporting both as shared or both as not shared, which is incorrect.
This reproducer triggers that bug:
$ cat fiemap-bug.sh #!/bin/bashDEV=/dev/sdj
MNT=/mnt/sdjmkfs.btrfs -f $DEV
mount $DEV $MNT# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents.xfs_io-f -c "pwrite -S0xab0 256K" \
-c "fsync" \
-c "pwrite -S0xcd256K 256K" \
-c "fsync" \
$MNT/foo# Now clone only the second extent into another file.
xfs_io-f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/fooumount $MNT
Running the reproducer:
$ ./fiemap-bug.sh wrote 262144/262144 bytes at offset 0 256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec) wrote 262144/262144 bytes at offset 262144 256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec) linked 262144/262144 bytes at offset 0 256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec)Filesystem type is: 9123683e File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes) ext: logical_offset: physical_offset: length: expected: flags: 0: 0.. 127: 3328.. 3455: 128: last,eof /mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
This patch is part of a larger patchset that is comprised of the following patches:
btrfs: allow hole and data seeking to be interruptible
btrfs: make hole and data seeking a lot more efficient
btrfs: remove check for impossible block start for an extent map at fiemap
btrfs: remove zero length check when entering fiemap
btrfs: properly flush delalloc when entering fiemap
btrfs: allow fiemap to be interruptible
btrfs: renamebtrfs_check_shared()to a more descriptive name
btrfs: speedup checking for extent sharedness during fiemap
btrfs: skip unnecessary extent buffer sharedness checks during fiemap
btrfs: make fiemap more efficient and accurate reporting extent sharednessThe patchset was tested on a machine running a non-debug kernel (Debian's default config) and compared the tests below on a branch without the patchset versus the same branch with the whole patchset applied.
The following test for a large compressed file without holes:
$ cat fiemap-perf-test.sh #!/bin/bashDEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f $DEV mount -o compress=lzo $DEV $MNT # 40G gives 327680 128K file extents (due to compression). xfs_io -f -c "pwrite -S 0xab -b 1M 0 20G" $MNT/foobar umount $MNT mount -o compress=lzo $DEV $MNT start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata not cached)" start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata cached)" umount $MNTBefore patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 3597 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 2107 milliseconds (metadata cached)After patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 1214 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 684 milliseconds (metadata cached)That's a speedup of about 3x for both cases (no metadata cached and all metadata cached).
The test provided by Pavel (first Link tag at the bottom), which uses files with a large number of holes, was also used to measure the gains, and it consists on a small C program and a shell script to invoke it. The C program is the following:
$ cat pavels-test.c #include <stdio.h> #include <unistd.h> #include <stdlib.h> #include <fcntl.h>#include <sys/stat.h> #include <sys/time.h> #include <sys/ioctl.h> #include <linux/fs.h> #include <linux/fiemap.h> #define FILE_INTERVAL (1<<13) /* 8Kb */ long long interval(struct timeval t1, struct timeval t2) { long long val = 0; val += (t2.tv_usec - t1.tv_usec); val += (t2.tv_sec - t1.tv_sec) * 1000 * 1000; return val; } int main(int argc, char **argv) { struct fiemap fiemap = {}; struct timeval t1, t2; char data = 'a'; struct stat st; int fd, off, file_size = FILE_INTERVAL; if (argc != 3 && argc != 2) { printf("usage: %s <path> [size]\n", argv[0]); return 1; } if (argc == 3) file_size = atoi(argv[2]); if (file_size < FILE_INTERVAL) file_size = FILE_INTERVAL; file_size -= file_size % FILE_INTERVAL; fd = open(argv[1], O_RDWR | O_CREAT | O_TRUNC, 0644); if (fd < 0) { perror("open"); return 1; } for (off = 0; off < file_size; off += FILE_INTERVAL) { if (pwrite(fd, &data, 1, off) != 1) { perror("pwrite"); close(fd); return 1; } } if (ftruncate(fd, file_size)) { perror("ftruncate"); close(fd); return 1; } if (fstat(fd, &st) < 0) { perror("fstat"); close(fd); return 1; } printf("size: %ld\n", st.st_size); printf("actual size: %ld\n", st.st_blocks * 512); fiemap.fm_length = FIEMAP_MAX_OFFSET; gettimeofday(&t1, NULL); if (ioctl(fd, FS_IOC_FIEMAP, &fiemap) < 0) { perror("fiemap"); close(fd); return 1; } gettimeofday(&t2, NULL); printf("fiemap: fm_mapped_extents = %d\n", fiemap.fm_mapped_extents); printf("time = %lld us\n", interval(t1, t2)); close(fd); return 0; } $ gcc -o pavels_test pavels_test.cAnd the wrapper shell script:
$ cat fiemap-pavels-test.sh#!/bin/bash DEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f -O no-holes $DEV mount $DEV $MNT echo echo "*********** 256M ***********" echo ./pavels-test $MNT/testfile $((1 << 28)) echo ./pavels-test $MNT/testfile $((1 << 28)) echo echo "*********** 512M ***********" echo ./pavels-test $MNT/testfile $((1 << 29)) echo ./pavels-test $MNT/testfile $((1 << 29)) echo echo "*********** 1G ***********" echo ./pavels-test $MNT/testfile $((1 << 30)) echo ./pavels-test $MNT/testfile $((1 << 30)) umount $MNTRunning his reproducer before applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4003133 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4895330 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 30123675 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 33450934 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 224924074 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 217239242 usRunning it after applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29475 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29307 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 58996 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 59115 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 116251 time = 124141 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 119387 usThe speedup is massive, both on the first fiemap call and on the second one as well, as his test creates files with many holes and small extents (every extent follows a hole and precedes another hole).
For the 256M file we go from 4 seconds down to 29 milliseconds in the first run, and then from 4.9 seconds down to 29 milliseconds again in the second run, a speedup of 138x and 169x, respectively.
For the 512M file we go from 30.1 seconds down to 59 milliseconds in the first run, and then from 33.5 seconds down to 59 milliseconds again in the second run, a speedup of 510x and 568x, respectively.
For the 1G file, we go from 225 seconds down to 124 milliseconds in the first run, and then from 217 seconds down to 119 milliseconds in the second run, a speedup of 1815x and 1824x, respectively.
Reported-by Pavel Tikhomirov <[email protected]>Reported-by Dominique MARTINET <[email protected]>Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/ctree.h | 4 +- fs/btrfs/extent_io.c | 714 +++++++++++++++++++++++++++++-------------- fs/btrfs/file.c | 16 +- fs/btrfs/inode.c | 140 +-------- 4 files changed, 506 insertions(+), 368 deletions(-)fs/btrfs/ctree.h +2
-2@@ -3402,8 +3402,6 @@ unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio, u64 start, u64 end);int btrfs_check_data_csum (struct inode *inode,struct btrfs_bio *bbio,u32 bio_offset,struct page *page,u32 pgoff); -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len); noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,u64 *orig_start,u64 *orig_block_len,u64 *ram_bytes,bool strict);@@ -3583,6 +3581,8 @@ int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end);int btrfs_check_nocow_lock (struct btrfs_inode *inode,loff_t pos,size_t *write_bytes);void btrfs_check_nocow_unlock (struct btrfs_inode *inode); +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret);/* tree-defrag.c */ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,fs/btrfs/extent_io.c +490
-224@@ -5353,42 +5353,6 @@ int try_release_extent_mapping(struct page *page, gfp_t mask)return try_release_extent_state (tree, page, mask); } -/* -* helper function for fiemap, which doesn't want to see any holes. -* This maps until we find something past 'last' -*/ -static struct extent_map *get_extent_skip_holes (struct btrfs_inode *inode, -u64 offset,u64 last) -{ -u64 sectorsize =btrfs_inode_sectorsize (inode); -struct extent_map *em; -u64 len; - -if (offset >= last) -return NULL ; - -while (1 ) { - len = last - offset; -if (len ==0 ) -break ; - len =ALIGN (len, sectorsize); - em =btrfs_get_extent_fiemap (inode, offset, len); -if (IS_ERR (em)) -return em; - -/* if this isn't a hole return it */ -if (em->block_start !=EXTENT_MAP_HOLE ) -return em; - -/* this is a hole, advance to the next extent */ - offset =extent_map_end (em); -free_extent_map (em); -if (offset >= last) -break ; - } -return NULL ; -} -/* * To cache previous fiemap extent * @@ -5418,6 +5382,9 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, {int ret =0 ; +/* Set at the end of extent_fiemap(). */ +ASSERT ((flags &FIEMAP_EXTENT_LAST ) ==0 ); +if (!cache->cached) goto assign;@@ -5446,11 +5413,10 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,*/ if (cache->offset + cache->len == offset && cache->phys + cache->len == phys && - (cache->flags & ~FIEMAP_EXTENT_LAST ) == - (flags & ~FIEMAP_EXTENT_LAST )) { + cache->flags == flags) { cache->len += len; cache->flags |= flags; - goto try_submit_last; +return 0 ; }/* Not mergeable, need to submit cached one */ @@ -5465,13 +5431,8 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, cache->phys = phys; cache->len = len; cache->flags = flags; -try_submit_last: -if (cache->flags &FIEMAP_EXTENT_LAST ) { - ret =fiemap_fill_next_extent (fieinfo, cache->offset, - cache->phys, cache->len, cache->flags); - cache->cached = false; - } -return ret; + +return 0 ; }/* @@ -5501,229 +5462,534 @@ static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,return ret; } -int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, -u64 start,u64 len) +static int fiemap_next_leaf_item (struct btrfs_inode *inode, +struct btrfs_path *path) { -int ret =0 ; -u64 off; -u64 max = start + len; -u32 flags =0 ; -u32 found_type; -u64 last; -u64 last_for_get_extent =0 ; -u64 disko =0 ; -u64 isize =i_size_read (&inode->vfs_inode); -struct btrfs_key found_key; -struct extent_map *em =NULL ; -struct extent_state *cached_state =NULL ; -struct btrfs_path *path; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; + + path->slots[0 ]++; +if (path->slots[0 ] <btrfs_header_nritems (path->nodes[0 ])) +return 0 ; + + ret =btrfs_next_leaf (inode->root, path); +if (ret !=0 ) +return ret; + +/* +* Don't bother with cloning if there are no more file extent items for +* our inode. +*/ +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid !=btrfs_ino (inode) || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; + +/* See the comment at fiemap_search_slot() about why we clone. */ + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot; + +return 0 ; +} + +/* +* Search for the first file extent item that starts at a given file offset or +* the one that starts immediately before that offset. +* Returns: 0 on success, < 0 on error, 1 if not found. +*/ +static int fiemap_search_slot (struct btrfs_inode *inode,struct btrfs_path *path, +u64 file_offset) +{ +const u64 ino =btrfs_ino (inode);struct btrfs_root *root = inode->root; -struct fiemap_cache cache = {0 }; -struct btrfs_backref_shared_cache *backref_cache; -struct ulist *roots; -struct ulist *tmp_ulist; -int end =0 ; -u64 em_start =0 ; -u64 em_len =0 ; -u64 em_end =0 ; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; - backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); - path =btrfs_alloc_path (); - roots =ulist_alloc (GFP_KERNEL ); - tmp_ulist =ulist_alloc (GFP_KERNEL ); -if (!backref_cache || !path || !roots || !tmp_ulist) { - ret = -ENOMEM ; - goto out_free_ulist; + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = file_offset; + + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) +return ret; + +if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } + +if (path->slots[0 ] >=btrfs_header_nritems (path->nodes[0 ])) { + ret =btrfs_next_leaf (root, path); +if (ret !=0 ) +return ret; + +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; }/* -* We can't initialize that to 'start' as this could miss extents due -* to extent item merging +* We clone the leaf and use it during fiemap. This is because while +* using the leaf we do expensive things like checking if an extent is +* shared, which can take a long time. In order to prevent blocking +* other tasks for too long, we use a clone of the leaf. We have locked +* the file range in the inode's io tree, so we know none of our file +* extent items can change. This way we avoid blocking other tasks that +* want to insert items for other inodes in the same leaf or b+tree +* rebalance operations (triggered for example when someone is trying +* to push items into this leaf when trying to insert an item in a +* neighbour leaf). +* We also need the private clone because holding a read lock on an +* extent buffer of the subvolume's b+tree will make lockdep unhappy +* when we call fiemap_fill_next_extent(), because that may cause a page +* fault when filling the user space buffer with fiemap data. */ - off =0 ; - start =round_down (start,btrfs_inode_sectorsize (inode)); - len =round_up (max,btrfs_inode_sectorsize (inode)) - start; + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot; + +return 0 ; +} + +/* +* Process a range which is a hole or a prealloc extent in the inode's subvolume +* btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc +* extent. The end offset (@end) is inclusive. +*/ +static int fiemap_process_hole (struct btrfs_inode *inode, +struct fiemap_extent_info *fieinfo, +struct fiemap_cache *cache, +struct btrfs_backref_shared_cache *backref_cache, +u64 disk_bytenr,u64 extent_offset, +u64 extent_gen, +struct ulist *roots,struct ulist *tmp_ulist, +u64 start,u64 end) +{ +const u64 i_size =i_size_read (&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +u64 cur_offset = start; +u64 last_delalloc_end =0 ; +u32 prealloc_flags =FIEMAP_EXTENT_UNWRITTEN ; +bool checked_extent_shared = false; +int ret;/* -* lookup the last file extent. We're not using i_size here -* because there might be preallocation past i_size +* There can be no delalloc past i_size, so don't waste time looking for +* it beyond i_size. */ - ret =btrfs_lookup_file_extent (NULL , root, path,btrfs_ino (inode),-1 , -0 ); -if (ret <0 ) { - goto out_free_ulist; - }else { -WARN_ON (!ret); -if (ret ==1 ) - ret =0 ; - } +while (cur_offset < end && cur_offset < i_size) { +u64 delalloc_start; +u64 delalloc_end; +u64 prealloc_start; +u64 prealloc_len =0 ; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; - path->slots[0 ]--; -btrfs_item_key_to_cpu (path->nodes[0 ], &found_key, path->slots[0 ]); - found_type = found_key.type; - -/* No extents, but there might be delalloc bits */ -if (found_key.objectid !=btrfs_ino (inode) || - found_type !=BTRFS_EXTENT_DATA_KEY ) { -/* have to trust i_size as the end */ - last = (u64)-1 ; - last_for_get_extent = isize; - }else {/* -* remember the start of the last extent. There are a -* bunch of different factors that go into the length of the -* extent, so its much less complex to remember where it started +* If this is a prealloc extent we have to report every section +* of it that has no delalloc. */ - last = found_key.offset; - last_for_get_extent = last +1 ; +if (disk_bytenr !=0 ) { +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = delalloc_start - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = delalloc_start - prealloc_start; + } + } + +if (prealloc_len >0 ) { +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + + checked_extent_shared = true; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + extent_offset += prealloc_len; + } + + ret =emit_fiemap_extent (fieinfo, cache, delalloc_start,0 , + delalloc_end +1 - delalloc_start, +FIEMAP_EXTENT_DELALLOC | +FIEMAP_EXTENT_UNKNOWN ); +if (ret) +return ret; + + last_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; + extent_offset += cur_offset - delalloc_start; +cond_resched (); + } + +/* +* Either we found no delalloc for the whole prealloc extent or we have +* a prealloc extent that spans i_size or starts at or after i_size. +*/ +if (disk_bytenr !=0 && last_delalloc_end < end) { +u64 prealloc_start; +u64 prealloc_len; + +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = end +1 - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = end +1 - prealloc_start; + } + +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + } + +return 0 ; +} + +static int fiemap_find_last_extent_offset (struct btrfs_inode *inode, +struct btrfs_path *path, +u64 *last_extent_end_ret) +{ +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct extent_buffer *leaf; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 disk_bytenr; +int ret; + +/* +* Lookup the last file extent. We're not using i_size here because +* there might be preallocation past i_size. +*/ + ret =btrfs_lookup_file_extent (NULL , root, path, ino, (u64)-1 ,0 ); +/* There can't be a file extent item at offset (u64)-1 */ +ASSERT (ret !=0 ); +if (ret <0 ) +return ret; + +/* +* For a non-existing key, btrfs_search_slot() always leaves us at a +* slot > 0, except if the btree is empty, which is impossible because +* at least it has the inode item for this inode and all the items for +* the root inode 256. +*/ +ASSERT (path->slots[0 ] >0 ); + path->slots[0 ]--; + leaf = path->nodes[0 ]; +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) { +/* No file extent items in the subvolume tree. */ + *last_extent_end_ret =0 ; +return 0 ; } -btrfs_release_path (path);/* -* we might have some extents allocated but more delalloc past those -* extents. so, we trust isize unless the start of the last extent is -* beyond isize +* For an inline extent, the disk_bytenr is where inline data starts at, +* so first check if we have an inline extent item before checking if we +* have an implicit hole (disk_bytenr == 0). */ -if (last < isize) { - last = (u64)-1 ; - last_for_get_extent = isize; + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); +if (btrfs_file_extent_type (leaf, ei) ==BTRFS_FILE_EXTENT_INLINE ) { + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; } -lock_extent_bits (&inode->io_tree, start, start + len -1 , - &cached_state); +/* +* Find the last file extent item that is not a hole (when NO_HOLES is +* not enabled). This should take at most 2 iterations in the worst +* case: we have one hole file extent item at slot 0 of a leaf and +* another hole file extent item as the last item in the previous leaf. +* This is because we merge file extent items that represent holes. +*/ + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +while (disk_bytenr ==0 ) { + ret =btrfs_previous_item (root, path, ino,BTRFS_EXTENT_DATA_KEY ); +if (ret <0 ) { +return ret; + }else if (ret >0 ) { +/* No file extent items that are not holes. */ + *last_extent_end_ret =0 ; +return 0 ; + } + leaf = path->nodes[0 ]; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); + } - em =get_extent_skip_holes (inode, start, last_for_get_extent); -if (!em) - goto out; -if (IS_ERR (em)) { - ret =PTR_ERR (em); + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; +} + +int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, +u64 start,u64 len) +{ +const u64 ino =btrfs_ino (inode); +struct extent_state *cached_state =NULL ; +struct btrfs_path *path; +struct btrfs_root *root = inode->root; +struct fiemap_cache cache = {0 }; +struct btrfs_backref_shared_cache *backref_cache; +struct ulist *roots; +struct ulist *tmp_ulist; +u64 last_extent_end; +u64 prev_extent_end; +u64 lockstart; +u64 lockend; +bool stopped = false; +int ret; + + backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); + path =btrfs_alloc_path (); + roots =ulist_alloc (GFP_KERNEL ); + tmp_ulist =ulist_alloc (GFP_KERNEL ); +if (!backref_cache || !path || !roots || !tmp_ulist) { + ret = -ENOMEM ; goto out; } -while (!end) { -u64 offset_in_extent =0 ; + lockstart =round_down (start,btrfs_inode_sectorsize (inode)); + lockend =round_up (start + len,btrfs_inode_sectorsize (inode)); + prev_extent_end = lockstart; -/* break if the extent we found is outside the range */ -if (em->start >= max ||extent_map_end (em) < off) -break ; +lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); -/* -* get_extent may return an extent that starts before our -* requested range. We have to make sure the ranges -* we return to fiemap always move forward and don't -* overlap, so adjust the offsets here -*/ - em_start =max (em->start, off); + ret =fiemap_find_last_extent_offset (inode, path, &last_extent_end); +if (ret <0 ) + goto out_unlock; +btrfs_release_path (path); + path->reada =READA_FORWARD ; + ret =fiemap_search_slot (inode, path, lockstart); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) {/* -* record the offset from the start of the extent -* for adjusting the disk offset below. Only do this if the -* extent isn't compressed since our in ram offset may be past -* what we have actually allocated on disk. +* No file extent item found, but we may have delalloc between +* the current offset and i_size. So check for that. */ -if (!test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - offset_in_extent = em_start - em->start; - em_end =extent_map_end (em); - em_len = em_end - em_start; - flags =0 ; -if (em->block_start <EXTENT_MAP_LAST_BYTE ) - disko = em->block_start + offset_in_extent; -else - disko =0 ; + ret =0 ; + goto check_eof_delalloc; + } + +while (prev_extent_end < lockend) { +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 extent_end; +u64 extent_len; +u64 extent_offset =0 ; +u64 extent_gen; +u64 disk_bytenr =0 ; +u64 flags =0 ; +int extent_type; +u8 compression; + +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +break ; + + extent_end =btrfs_file_extent_end (path);/* -* bump off for our next call to get_extent +* The first iteration can leave us at an extent item that ends +* before our range's start. Move to the next item. */ - off =extent_map_end (em); -if (off >= max) - end =1 ; - -if (em->block_start ==EXTENT_MAP_INLINE ) { - flags |= (FIEMAP_EXTENT_DATA_INLINE | -FIEMAP_EXTENT_NOT_ALIGNED ); - }else if (em->block_start ==EXTENT_MAP_DELALLOC ) { - flags |= (FIEMAP_EXTENT_DELALLOC | -FIEMAP_EXTENT_UNKNOWN ); - }else if (fieinfo->fi_extents_max) { -u64 extent_gen; -u64 bytenr = em->block_start - - (em->start - em->orig_start); +if (extent_end <= lockstart) + goto next_item; -/* -* If two extent maps are merged, then their generation -* is set to the maximum between their generations. -* Otherwise its generation matches the one we have in -* corresponding file extent item. If we have a merged -* extent map, don't use its generation to speedup the -* sharedness check below. -*/ -if (test_bit (EXTENT_FLAG_MERGED , &em->flags)) - extent_gen =0 ; -else - extent_gen = em->generation; +/* We have in implicit hole (NO_HOLES feature enabled). */ +if (prev_extent_end < key.offset) { +const u64 range_end =min (key.offset, lockend) -1 ; -/* -* As btrfs supports shared space, this information -* can be exported to userspace tools via -* flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 -* then we're just getting a count and we can skip the -* lookup stuff. -*/ - ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), - bytenr, extent_gen, - roots, tmp_ulist, - backref_cache); -if (ret <0 ) - goto out_free; -if (ret) - flags |=FIEMAP_EXTENT_SHARED ; - ret =0 ; - } -if (test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - flags |=FIEMAP_EXTENT_ENCODED ; -if (test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) - flags |=FIEMAP_EXTENT_UNWRITTEN ; + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + prev_extent_end, range_end); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; + } -free_extent_map (em); - em =NULL ; -if ((em_start >= last) || em_len == (u64)-1 || - (last == (u64)-1 && isize <= em_end)) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; +/* We've reached the end of the fiemap range, stop. */ +if (key.offset >= lockend) { + stopped = true; +break ; + } } -/* now scan forward to see if this is really the last extent. */ - em =get_extent_skip_holes (inode, off, last_for_get_extent); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - goto out; + extent_len = extent_end - key.offset; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + compression =btrfs_file_extent_compression (leaf, ei); + extent_type =btrfs_file_extent_type (leaf, ei); + extent_gen =btrfs_file_extent_generation (leaf, ei); + +if (extent_type !=BTRFS_FILE_EXTENT_INLINE ) { + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +if (compression ==BTRFS_COMPRESS_NONE ) + extent_offset =btrfs_file_extent_offset (leaf, ei); } -if (!em) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; + +if (compression !=BTRFS_COMPRESS_NONE ) + flags |=FIEMAP_EXTENT_ENCODED ; + +if (extent_type ==BTRFS_FILE_EXTENT_INLINE ) { + flags |=FIEMAP_EXTENT_DATA_INLINE ; + flags |=FIEMAP_EXTENT_NOT_ALIGNED ; + ret =emit_fiemap_extent (fieinfo, &cache, key.offset,0 , + extent_len, flags); + }else if (extent_type ==BTRFS_FILE_EXTENT_PREALLOC ) { + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache, + disk_bytenr, extent_offset, + extent_gen, roots, tmp_ulist, + key.offset, extent_end -1 ); + }else if (disk_bytenr ==0 ) { +/* We have an explicit hole. */ + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + key.offset, extent_end -1 ); + }else { +/* We have a regular extent. */ +if (fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (root, ino, + disk_bytenr, + extent_gen, + roots, + tmp_ulist, + backref_cache); +if (ret <0 ) + goto out_unlock; +else if (ret >0 ) + flags |=FIEMAP_EXTENT_SHARED ; + } + + ret =emit_fiemap_extent (fieinfo, &cache, key.offset, + disk_bytenr + extent_offset, + extent_len, flags); } - ret =emit_fiemap_extent (fieinfo, &cache, em_start, disko, - em_len, flags); -if (ret) { -if (ret ==1 ) - ret =0 ; - goto out_free; + +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; } + prev_extent_end = extent_end; +next_item:if (fatal_signal_pending (current)) { ret = -EINTR ; - goto out_free; + goto out_unlock; } + + ret =fiemap_next_leaf_item (inode, path); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* No more file extent items for this inode. */ +break ; + } +cond_resched (); } -out_free: -if (!ret) - ret =emit_last_fiemap_cache (fieinfo, &cache); -free_extent_map (em); -out: -unlock_extent_cached (&inode->io_tree, start, start + len -1 , - &cached_state); -out_free_ulist: +check_eof_delalloc: +/* +* Release (and free) the path before emitting any final entries to +* fiemap_fill_next_extent() to keep lockdep happy. This is because +* once we find no more file extent items exist, we may have a +* non-cloned leaf, and fiemap_fill_next_extent() can trigger page +* faults when copying data to the user space buffer. +*/ +btrfs_free_path (path); + path =NULL ; + +if (!stopped && prev_extent_end < lockend) { + ret =fiemap_process_hole (inode, fieinfo, &cache, backref_cache, +0 ,0 ,0 , roots, tmp_ulist, + prev_extent_end, lockend -1 ); +if (ret <0 ) + goto out_unlock; + prev_extent_end = lockend; + } + +if (cache.cached && cache.offset + cache.len >= last_extent_end) { +const u64 i_size =i_size_read (&inode->vfs_inode); + +if (prev_extent_end < i_size) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, + prev_extent_end, + i_size -1 , + &delalloc_start, + &delalloc_end); +if (!delalloc) + cache.flags |=FIEMAP_EXTENT_LAST ; + }else { + cache.flags |=FIEMAP_EXTENT_LAST ; + } + } + + ret =emit_last_fiemap_cache (fieinfo, &cache); + +out_unlock: +unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); +out:kfree (backref_cache);btrfs_free_path (path);ulist_free (roots);fs/btrfs/file.c +8
-8@@ -3602,10 +3602,10 @@ static long btrfs_fallocate(struct file *file, int mode, }/* -* Helper for have_delalloc_in_range(). Find a subrange in a given range that -* has unflushed and/or flushing delalloc. There might be other adjacent -* subranges after the one it found, so have_delalloc_in_range() keeps looping -* while it gets adjacent subranges, and merging them together. +* Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range +* that has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so btrfs_find_delalloc_in_range() keeps +* looping while it gets adjacent subranges, and merging them together. */ static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end,u64 *delalloc_start_ret,u64 *delalloc_end_ret)@@ -3740,8 +3740,8 @@ static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and * end offsets of the subrange. */ -static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, -u64 *delalloc_start_ret,u64 *delalloc_end_ret) +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) {u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize);u64 prev_delalloc_end =0 ;@@ -3804,8 +3804,8 @@ static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,u64 delalloc_end;bool delalloc; - delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, - &delalloc_end); + delalloc =btrfs_find_delalloc_in_range (inode, start, end, + &delalloc_start, &delalloc_end);if (delalloc && whence ==SEEK_DATA ) { *start_ret = delalloc_start;return true;fs/btrfs/inode.c +5
-133@@ -7064,133 +7064,6 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,return em; } -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len) -{ -struct extent_map *em; -struct extent_map *hole_em =NULL ; -u64 delalloc_start = start; -u64 end; -u64 delalloc_len; -u64 delalloc_end; -int err =0 ; - - em =btrfs_get_extent (inode,NULL ,0 , start, len); -if (IS_ERR (em)) -return em; -/* -* If our em maps to: -* - a hole or -* - a pre-alloc extent, -* there might actually be delalloc bytes behind it. -*/ -if (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) -return em; -else - hole_em = em; - -/* check to see if we've wrapped (len == -1 or similar) */ - end = start + len; -if (end < start) - end = (u64)-1 ; -else - end -=1 ; - - em =NULL ; - -/* ok, we didn't find anything, lets look for delalloc */ - delalloc_len =count_range_bits (&inode->io_tree, &delalloc_start, - end, len,EXTENT_DELALLOC ,1 ); - delalloc_end = delalloc_start + delalloc_len; -if (delalloc_end < delalloc_start) - delalloc_end = (u64)-1 ; - -/* -* We didn't find anything useful, return the original results from -* get_extent() -*/ -if (delalloc_start > end || delalloc_end <= start) { - em = hole_em; - hole_em =NULL ; - goto out; - } - -/* -* Adjust the delalloc_start to make sure it doesn't go backwards from -* the start they passed in -*/ - delalloc_start =max (start, delalloc_start); - delalloc_len = delalloc_end - delalloc_start; - -if (delalloc_len >0 ) { -u64 hole_start; -u64 hole_len; -const u64 hole_end =extent_map_end (hole_em); - - em =alloc_extent_map (); -if (!em) { - err = -ENOMEM ; - goto out; - } - -ASSERT (hole_em); -/* -* When btrfs_get_extent can't find anything it returns one -* huge hole -* -* Make sure what it found really fits our range, and adjust to -* make sure it is based on the start from the caller -*/ -if (hole_end <= start || hole_em->start > end) { -free_extent_map (hole_em); - hole_em =NULL ; - }else { - hole_start =max (hole_em->start, start); - hole_len = hole_end - hole_start; - } - -if (hole_em && delalloc_start > hole_start) { -/* -* Our hole starts before our delalloc, so we have to -* return just the parts of the hole that go until the -* delalloc starts -*/ - em->len =min (hole_len, delalloc_start - hole_start); - em->start = hole_start; - em->orig_start = hole_start; -/* -* Don't adjust block start at all, it is fixed at -* EXTENT_MAP_HOLE -*/ - em->block_start = hole_em->block_start; - em->block_len = hole_len; -if (test_bit (EXTENT_FLAG_PREALLOC , &hole_em->flags)) -set_bit (EXTENT_FLAG_PREALLOC , &em->flags); - }else { -/* -* Hole is out of passed range or it starts after -* delalloc range -*/ - em->start = delalloc_start; - em->len = delalloc_len; - em->orig_start = delalloc_start; - em->block_start =EXTENT_MAP_DELALLOC ; - em->block_len = delalloc_len; - } - }else { -return hole_em; - } -out: - -free_extent_map (hole_em); -if (err) { -free_extent_map (em); -return ERR_PTR (err); - } -return em; -} - static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, const u64 start, const u64 len,@@ -8265,15 +8138,14 @@ static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,* in the compression of data (in an async thread) and will return * before the compression is done and writeback is started. A second * filemap_fdatawrite_range() is needed to wait for the compression to -* complete and writeback to start. Without this, our user is very -* likely to get stale results, because the extents and extent maps for -* delalloc regions are only allocated when writeback starts. +* complete and writeback to start. We also need to wait for ordered +* extents to complete, because our fiemap implementation uses mainly +* file extent items to list the extents, searching for extent maps +* only for file ranges with holes or prealloc extents to figure out +* if we have delalloc in those ranges. */ if (fieinfo->fi_flags &FIEMAP_FLAG_SYNC ) { - ret =btrfs_fdatawrite_range (inode,0 ,LLONG_MAX ); -if (ret) -return ret;
Quoted 8 lines
fs/btrfs/inode.c static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, · line 8151
Quoted 5 lines
fs/btrfs/inode.c +1
-1- ret =filemap_fdatawait_range (inode->i_mapping,0 ,LLONG_MAX ); + ret =btrfs_wait_ordered_range (inode,0 ,LLONG_MAX );if (ret)return ret; }Hmm this bit should be in "btrfs: properly flush delalloc when entering fiemap" instead. Thanks,
Nop, the change is done here for a good reason: before this change, we only needed
to wait for writeback to complete (actually just to start and create the new extent maps),
so that's why that other patch only waits for writeback to complete, just like the generic code.
After this change we need to wait for ordered extents to complete, since we use the file
extent items to get extent information for fiemap - that's why that change is in this patch.
Thanks.
Quoted 2 lines
Josef
Josef Bacik
Reviewed-byRe: [PATCH 10/10] btrfs: make fiemap more efficient and accurate reporting extent sharedness
Quoted 1,498 lines · Filipe Manana
Quoted 1,481 lines · Josef Bacik
Quoted 1,479 lines · On Thu, Sep 01, 2022 at 02:18:30PM +0100, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current fiemap implementation does not scale very well with the number of extents a file has. This is both because the main algorithm to find out the extents has a high algorithmic complexity and because for each extent we have to check if it's shared. This second part, checking if an extent is shared, is significantly improved by the two previous patches in this patchset, while the first part is improved by this specific patch. Every now and then we get reports from users mentioning fiemap is too slow or even unusable for files with a very large number of extents, such as the two recent reports referred to by the Link tags at the bottom of this change log.
To understand why the part of finding which extents a file has is very inneficient, consider the example of doing a full ranged fiemap against a file that has over 100K extents (normal for example for a file with more than 10G of data and using compression, which limits the extent size to 128K). When we enter fiemap at
extent_fiemap(), the following happens:1) Before entering the main loop, we call
get_extent_skip_holes()to get the first extent map. This leads us tobtrfs_get_extent_fiemap(), which in turn callsbtrfs_get_extent(), to find the first extent map that covers the file range [0,LLONG_MAX).
btrfs_get_extent()will first search the inode's extent map tree, to see if we have an extent map there that covers the range. If it does not find one, then it will search the inode's subvolume b+tree for a fitting file extent item. After finding the file extent item, it will allocate an extent map, fill it in with information extracted from the file extent item, and add it to the inode's extent map tree (which requires a search for insertion in the tree).2) Then we enter the main loop at
extent_fiemap(), emit the details of the extent, and call againget_extent_skip_holes(), with a start offset matching the end of the extent map we previously processed.We end up at
btrfs_get_extent()again, will search the extent map tree and then search the subvolume b+tree for a file extent item if we could not find an extent map in the extent tree. We allocate an extent map, fill it in with the details in the file extent item, and then insert it into the extent map tree (yet another search in this tree).3) The second step is repeated over and over, until we have processed the whole file range. Each iteration ends at
btrfs_get_extent(), which does a red black tree search on the extent map tree, then searches the subvolume b+tree, allocates an extent map and then does another search in the extent map tree in order to insert the extent map.In the best scenario we have all the extent maps already in the extent tree, and so for each extent we do a single search on a red black tree, so we have a complexity of O(n log n).
In the worst scenario we don't have any extent map already loaded in the extent map tree, or have very few already there. In this case the complexity is much higher since we do:
- A red black tree search on the extent map tree, which has O(log n) complexity, initially very fast since the tree is empty or very small, but as we end up allocating extent maps and adding them to the tree when we don't find them there, each subsequent search on the tree gets slower, since it's getting bigger and bigger after each iteration.
- A search on the subvolume b+tree, also O(log n) complexity, but it has items for all inodes in the subvolume, not just items for our inode. Plus on a filesystem with concurrent operations on other inodes, we can block doing the search due to lock contention on b+tree nodes/leaves.
- Allocate an extent map - this can block, and can also fail if we are under serious memory pressure.
- Do another search on the extent maps red black tree, with the goal of inserting the extent map we just allocated. Again, after every iteration this tree is getting bigger by 1 element, so after many iterations the searches are slower and slower.
- We will not need the allocated extent map anymore, so it's pointless to add it to the extent map tree. It's just wasting time and memory.
In short we end up searching the extent map tree multiple times, on a tree that is growing bigger and bigger after each iteration. And besides that we visit the same leaf of the subvolume b+tree many times, since a leaf with the default size of 16K can easily have more than 200 file extent items.
This is very inneficient overall. This patch changes the algorithm to instead iterate over the subvolume b+tree, visiting each leaf only once, and only searching in the extent map tree for file ranges that have holes or prealloc extents, in order to figure out if we have delalloc there. It will never allocate an extent map and add it to the extent map tree. This is very similar to what was previously done for the lseek's hole and data seeking features.
Also, the current implementation relying on extent maps for figuring out which extents we have is not correct. This is because extent maps can be merged even if they represent different extents - we do this to minimize memory utilization and keep extent map trees smaller. For example if we have two extents that are contiguous on disk, once we load the two extent maps, they get merged into a single one - however if only one of the extents is shared, we end up reporting both as shared or both as not shared, which is incorrect.
This reproducer triggers that bug:
$ cat fiemap-bug.sh #!/bin/bashDEV=/dev/sdj
MNT=/mnt/sdjmkfs.btrfs -f $DEV
mount $DEV $MNT# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents.xfs_io-f -c "pwrite -S0xab0 256K" \
-c "fsync" \
-c "pwrite -S0xcd256K 256K" \
-c "fsync" \
$MNT/foo# Now clone only the second extent into another file.
xfs_io-f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/fooumount $MNT
Running the reproducer:
$ ./fiemap-bug.sh wrote 262144/262144 bytes at offset 0 256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec) wrote 262144/262144 bytes at offset 262144 256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec) linked 262144/262144 bytes at offset 0 256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec)Filesystem type is: 9123683e File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes) ext: logical_offset: physical_offset: length: expected: flags: 0: 0.. 127: 3328.. 3455: 128: last,eof /mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
This patch is part of a larger patchset that is comprised of the following patches:
btrfs: allow hole and data seeking to be interruptible
btrfs: make hole and data seeking a lot more efficient
btrfs: remove check for impossible block start for an extent map at fiemap
btrfs: remove zero length check when entering fiemap
btrfs: properly flush delalloc when entering fiemap
btrfs: allow fiemap to be interruptible
btrfs: renamebtrfs_check_shared()to a more descriptive name
btrfs: speedup checking for extent sharedness during fiemap
btrfs: skip unnecessary extent buffer sharedness checks during fiemap
btrfs: make fiemap more efficient and accurate reporting extent sharednessThe patchset was tested on a machine running a non-debug kernel (Debian's default config) and compared the tests below on a branch without the patchset versus the same branch with the whole patchset applied.
The following test for a large compressed file without holes:
$ cat fiemap-perf-test.sh #!/bin/bashDEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f $DEV mount -o compress=lzo $DEV $MNT # 40G gives 327680 128K file extents (due to compression). xfs_io -f -c "pwrite -S 0xab -b 1M 0 20G" $MNT/foobar umount $MNT mount -o compress=lzo $DEV $MNT start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata not cached)" start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata cached)" umount $MNTBefore patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 3597 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 2107 milliseconds (metadata cached)After patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 1214 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 684 milliseconds (metadata cached)That's a speedup of about 3x for both cases (no metadata cached and all metadata cached).
The test provided by Pavel (first Link tag at the bottom), which uses files with a large number of holes, was also used to measure the gains, and it consists on a small C program and a shell script to invoke it. The C program is the following:
$ cat pavels-test.c #include <stdio.h> #include <unistd.h> #include <stdlib.h> #include <fcntl.h>#include <sys/stat.h> #include <sys/time.h> #include <sys/ioctl.h> #include <linux/fs.h> #include <linux/fiemap.h> #define FILE_INTERVAL (1<<13) /* 8Kb */ long long interval(struct timeval t1, struct timeval t2) { long long val = 0; val += (t2.tv_usec - t1.tv_usec); val += (t2.tv_sec - t1.tv_sec) * 1000 * 1000; return val; } int main(int argc, char **argv) { struct fiemap fiemap = {}; struct timeval t1, t2; char data = 'a'; struct stat st; int fd, off, file_size = FILE_INTERVAL; if (argc != 3 && argc != 2) { printf("usage: %s <path> [size]\n", argv[0]); return 1; } if (argc == 3) file_size = atoi(argv[2]); if (file_size < FILE_INTERVAL) file_size = FILE_INTERVAL; file_size -= file_size % FILE_INTERVAL; fd = open(argv[1], O_RDWR | O_CREAT | O_TRUNC, 0644); if (fd < 0) { perror("open"); return 1; } for (off = 0; off < file_size; off += FILE_INTERVAL) { if (pwrite(fd, &data, 1, off) != 1) { perror("pwrite"); close(fd); return 1; } } if (ftruncate(fd, file_size)) { perror("ftruncate"); close(fd); return 1; } if (fstat(fd, &st) < 0) { perror("fstat"); close(fd); return 1; } printf("size: %ld\n", st.st_size); printf("actual size: %ld\n", st.st_blocks * 512); fiemap.fm_length = FIEMAP_MAX_OFFSET; gettimeofday(&t1, NULL); if (ioctl(fd, FS_IOC_FIEMAP, &fiemap) < 0) { perror("fiemap"); close(fd); return 1; } gettimeofday(&t2, NULL); printf("fiemap: fm_mapped_extents = %d\n", fiemap.fm_mapped_extents); printf("time = %lld us\n", interval(t1, t2)); close(fd); return 0; } $ gcc -o pavels_test pavels_test.cAnd the wrapper shell script:
$ cat fiemap-pavels-test.sh#!/bin/bash DEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f -O no-holes $DEV mount $DEV $MNT echo echo "*********** 256M ***********" echo ./pavels-test $MNT/testfile $((1 << 28)) echo ./pavels-test $MNT/testfile $((1 << 28)) echo echo "*********** 512M ***********" echo ./pavels-test $MNT/testfile $((1 << 29)) echo ./pavels-test $MNT/testfile $((1 << 29)) echo echo "*********** 1G ***********" echo ./pavels-test $MNT/testfile $((1 << 30)) echo ./pavels-test $MNT/testfile $((1 << 30)) umount $MNTRunning his reproducer before applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4003133 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4895330 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 30123675 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 33450934 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 224924074 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 217239242 usRunning it after applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29475 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29307 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 58996 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 59115 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 116251 time = 124141 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 119387 usThe speedup is massive, both on the first fiemap call and on the second one as well, as his test creates files with many holes and small extents (every extent follows a hole and precedes another hole).
For the 256M file we go from 4 seconds down to 29 milliseconds in the first run, and then from 4.9 seconds down to 29 milliseconds again in the second run, a speedup of 138x and 169x, respectively.
For the 512M file we go from 30.1 seconds down to 59 milliseconds in the first run, and then from 33.5 seconds down to 59 milliseconds again in the second run, a speedup of 510x and 568x, respectively.
For the 1G file, we go from 225 seconds down to 124 milliseconds in the first run, and then from 217 seconds down to 119 milliseconds in the second run, a speedup of 1815x and 1824x, respectively.
Reported-by Pavel Tikhomirov <[email protected]>Reported-by Dominique MARTINET <[email protected]>Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/ctree.h | 4 +- fs/btrfs/extent_io.c | 714 +++++++++++++++++++++++++++++-------------- fs/btrfs/file.c | 16 +- fs/btrfs/inode.c | 140 +-------- 4 files changed, 506 insertions(+), 368 deletions(-)fs/btrfs/ctree.h +2
-2@@ -3402,8 +3402,6 @@ unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio, u64 start, u64 end);int btrfs_check_data_csum (struct inode *inode,struct btrfs_bio *bbio,u32 bio_offset,struct page *page,u32 pgoff); -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len); noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,u64 *orig_start,u64 *orig_block_len,u64 *ram_bytes,bool strict);@@ -3583,6 +3581,8 @@ int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end);int btrfs_check_nocow_lock (struct btrfs_inode *inode,loff_t pos,size_t *write_bytes);void btrfs_check_nocow_unlock (struct btrfs_inode *inode); +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret);/* tree-defrag.c */ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,fs/btrfs/extent_io.c +490
-224@@ -5353,42 +5353,6 @@ int try_release_extent_mapping(struct page *page, gfp_t mask)return try_release_extent_state (tree, page, mask); } -/* -* helper function for fiemap, which doesn't want to see any holes. -* This maps until we find something past 'last' -*/ -static struct extent_map *get_extent_skip_holes (struct btrfs_inode *inode, -u64 offset,u64 last) -{ -u64 sectorsize =btrfs_inode_sectorsize (inode); -struct extent_map *em; -u64 len; - -if (offset >= last) -return NULL ; - -while (1 ) { - len = last - offset; -if (len ==0 ) -break ; - len =ALIGN (len, sectorsize); - em =btrfs_get_extent_fiemap (inode, offset, len); -if (IS_ERR (em)) -return em; - -/* if this isn't a hole return it */ -if (em->block_start !=EXTENT_MAP_HOLE ) -return em; - -/* this is a hole, advance to the next extent */ - offset =extent_map_end (em); -free_extent_map (em); -if (offset >= last) -break ; - } -return NULL ; -} -/* * To cache previous fiemap extent * @@ -5418,6 +5382,9 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, {int ret =0 ; +/* Set at the end of extent_fiemap(). */ +ASSERT ((flags &FIEMAP_EXTENT_LAST ) ==0 ); +if (!cache->cached) goto assign;@@ -5446,11 +5413,10 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,*/ if (cache->offset + cache->len == offset && cache->phys + cache->len == phys && - (cache->flags & ~FIEMAP_EXTENT_LAST ) == - (flags & ~FIEMAP_EXTENT_LAST )) { + cache->flags == flags) { cache->len += len; cache->flags |= flags; - goto try_submit_last; +return 0 ; }/* Not mergeable, need to submit cached one */ @@ -5465,13 +5431,8 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, cache->phys = phys; cache->len = len; cache->flags = flags; -try_submit_last: -if (cache->flags &FIEMAP_EXTENT_LAST ) { - ret =fiemap_fill_next_extent (fieinfo, cache->offset, - cache->phys, cache->len, cache->flags); - cache->cached = false; - } -return ret; + +return 0 ; }/* @@ -5501,229 +5462,534 @@ static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,return ret; } -int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, -u64 start,u64 len) +static int fiemap_next_leaf_item (struct btrfs_inode *inode, +struct btrfs_path *path) { -int ret =0 ; -u64 off; -u64 max = start + len; -u32 flags =0 ; -u32 found_type; -u64 last; -u64 last_for_get_extent =0 ; -u64 disko =0 ; -u64 isize =i_size_read (&inode->vfs_inode); -struct btrfs_key found_key; -struct extent_map *em =NULL ; -struct extent_state *cached_state =NULL ; -struct btrfs_path *path; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; + + path->slots[0 ]++; +if (path->slots[0 ] <btrfs_header_nritems (path->nodes[0 ])) +return 0 ; + + ret =btrfs_next_leaf (inode->root, path); +if (ret !=0 ) +return ret; + +/* +* Don't bother with cloning if there are no more file extent items for +* our inode. +*/ +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid !=btrfs_ino (inode) || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; + +/* See the comment at fiemap_search_slot() about why we clone. */ + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot; + +return 0 ; +} + +/* +* Search for the first file extent item that starts at a given file offset or +* the one that starts immediately before that offset. +* Returns: 0 on success, < 0 on error, 1 if not found. +*/ +static int fiemap_search_slot (struct btrfs_inode *inode,struct btrfs_path *path, +u64 file_offset) +{ +const u64 ino =btrfs_ino (inode);struct btrfs_root *root = inode->root; -struct fiemap_cache cache = {0 }; -struct btrfs_backref_shared_cache *backref_cache; -struct ulist *roots; -struct ulist *tmp_ulist; -int end =0 ; -u64 em_start =0 ; -u64 em_len =0 ; -u64 em_end =0 ; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; - backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); - path =btrfs_alloc_path (); - roots =ulist_alloc (GFP_KERNEL ); - tmp_ulist =ulist_alloc (GFP_KERNEL ); -if (!backref_cache || !path || !roots || !tmp_ulist) { - ret = -ENOMEM ; - goto out_free_ulist; + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = file_offset; + + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) +return ret; + +if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } + +if (path->slots[0 ] >=btrfs_header_nritems (path->nodes[0 ])) { + ret =btrfs_next_leaf (root, path); +if (ret !=0 ) +return ret; + +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; }/* -* We can't initialize that to 'start' as this could miss extents due -* to extent item merging +* We clone the leaf and use it during fiemap. This is because while +* using the leaf we do expensive things like checking if an extent is +* shared, which can take a long time. In order to prevent blocking +* other tasks for too long, we use a clone of the leaf. We have locked +* the file range in the inode's io tree, so we know none of our file +* extent items can change. This way we avoid blocking other tasks that +* want to insert items for other inodes in the same leaf or b+tree +* rebalance operations (triggered for example when someone is trying +* to push items into this leaf when trying to insert an item in a +* neighbour leaf). +* We also need the private clone because holding a read lock on an +* extent buffer of the subvolume's b+tree will make lockdep unhappy +* when we call fiemap_fill_next_extent(), because that may cause a page +* fault when filling the user space buffer with fiemap data. */ - off =0 ; - start =round_down (start,btrfs_inode_sectorsize (inode)); - len =round_up (max,btrfs_inode_sectorsize (inode)) - start; + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot; + +return 0 ; +} + +/* +* Process a range which is a hole or a prealloc extent in the inode's subvolume +* btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc +* extent. The end offset (@end) is inclusive. +*/ +static int fiemap_process_hole (struct btrfs_inode *inode, +struct fiemap_extent_info *fieinfo, +struct fiemap_cache *cache, +struct btrfs_backref_shared_cache *backref_cache, +u64 disk_bytenr,u64 extent_offset, +u64 extent_gen, +struct ulist *roots,struct ulist *tmp_ulist, +u64 start,u64 end) +{ +const u64 i_size =i_size_read (&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +u64 cur_offset = start; +u64 last_delalloc_end =0 ; +u32 prealloc_flags =FIEMAP_EXTENT_UNWRITTEN ; +bool checked_extent_shared = false; +int ret;/* -* lookup the last file extent. We're not using i_size here -* because there might be preallocation past i_size +* There can be no delalloc past i_size, so don't waste time looking for +* it beyond i_size. */ - ret =btrfs_lookup_file_extent (NULL , root, path,btrfs_ino (inode),-1 , -0 ); -if (ret <0 ) { - goto out_free_ulist; - }else { -WARN_ON (!ret); -if (ret ==1 ) - ret =0 ; - } +while (cur_offset < end && cur_offset < i_size) { +u64 delalloc_start; +u64 delalloc_end; +u64 prealloc_start; +u64 prealloc_len =0 ; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; - path->slots[0 ]--; -btrfs_item_key_to_cpu (path->nodes[0 ], &found_key, path->slots[0 ]); - found_type = found_key.type; - -/* No extents, but there might be delalloc bits */ -if (found_key.objectid !=btrfs_ino (inode) || - found_type !=BTRFS_EXTENT_DATA_KEY ) { -/* have to trust i_size as the end */ - last = (u64)-1 ; - last_for_get_extent = isize; - }else {/* -* remember the start of the last extent. There are a -* bunch of different factors that go into the length of the -* extent, so its much less complex to remember where it started +* If this is a prealloc extent we have to report every section +* of it that has no delalloc. */ - last = found_key.offset; - last_for_get_extent = last +1 ; +if (disk_bytenr !=0 ) { +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = delalloc_start - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = delalloc_start - prealloc_start; + } + } + +if (prealloc_len >0 ) { +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + + checked_extent_shared = true; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + extent_offset += prealloc_len; + } + + ret =emit_fiemap_extent (fieinfo, cache, delalloc_start,0 , + delalloc_end +1 - delalloc_start, +FIEMAP_EXTENT_DELALLOC | +FIEMAP_EXTENT_UNKNOWN ); +if (ret) +return ret; + + last_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; + extent_offset += cur_offset - delalloc_start; +cond_resched (); + } + +/* +* Either we found no delalloc for the whole prealloc extent or we have +* a prealloc extent that spans i_size or starts at or after i_size. +*/ +if (disk_bytenr !=0 && last_delalloc_end < end) { +u64 prealloc_start; +u64 prealloc_len; + +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = end +1 - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = end +1 - prealloc_start; + } + +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + } + +return 0 ; +} + +static int fiemap_find_last_extent_offset (struct btrfs_inode *inode, +struct btrfs_path *path, +u64 *last_extent_end_ret) +{ +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct extent_buffer *leaf; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 disk_bytenr; +int ret; + +/* +* Lookup the last file extent. We're not using i_size here because +* there might be preallocation past i_size. +*/ + ret =btrfs_lookup_file_extent (NULL , root, path, ino, (u64)-1 ,0 ); +/* There can't be a file extent item at offset (u64)-1 */ +ASSERT (ret !=0 ); +if (ret <0 ) +return ret; + +/* +* For a non-existing key, btrfs_search_slot() always leaves us at a +* slot > 0, except if the btree is empty, which is impossible because +* at least it has the inode item for this inode and all the items for +* the root inode 256. +*/ +ASSERT (path->slots[0 ] >0 ); + path->slots[0 ]--; + leaf = path->nodes[0 ]; +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) { +/* No file extent items in the subvolume tree. */ + *last_extent_end_ret =0 ; +return 0 ; } -btrfs_release_path (path);/* -* we might have some extents allocated but more delalloc past those -* extents. so, we trust isize unless the start of the last extent is -* beyond isize +* For an inline extent, the disk_bytenr is where inline data starts at, +* so first check if we have an inline extent item before checking if we +* have an implicit hole (disk_bytenr == 0). */ -if (last < isize) { - last = (u64)-1 ; - last_for_get_extent = isize; + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); +if (btrfs_file_extent_type (leaf, ei) ==BTRFS_FILE_EXTENT_INLINE ) { + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; } -lock_extent_bits (&inode->io_tree, start, start + len -1 , - &cached_state); +/* +* Find the last file extent item that is not a hole (when NO_HOLES is +* not enabled). This should take at most 2 iterations in the worst +* case: we have one hole file extent item at slot 0 of a leaf and +* another hole file extent item as the last item in the previous leaf. +* This is because we merge file extent items that represent holes. +*/ + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +while (disk_bytenr ==0 ) { + ret =btrfs_previous_item (root, path, ino,BTRFS_EXTENT_DATA_KEY ); +if (ret <0 ) { +return ret; + }else if (ret >0 ) { +/* No file extent items that are not holes. */ + *last_extent_end_ret =0 ; +return 0 ; + } + leaf = path->nodes[0 ]; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); + } - em =get_extent_skip_holes (inode, start, last_for_get_extent); -if (!em) - goto out; -if (IS_ERR (em)) { - ret =PTR_ERR (em); + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; +} + +int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, +u64 start,u64 len) +{ +const u64 ino =btrfs_ino (inode); +struct extent_state *cached_state =NULL ; +struct btrfs_path *path; +struct btrfs_root *root = inode->root; +struct fiemap_cache cache = {0 }; +struct btrfs_backref_shared_cache *backref_cache; +struct ulist *roots; +struct ulist *tmp_ulist; +u64 last_extent_end; +u64 prev_extent_end; +u64 lockstart; +u64 lockend; +bool stopped = false; +int ret; + + backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); + path =btrfs_alloc_path (); + roots =ulist_alloc (GFP_KERNEL ); + tmp_ulist =ulist_alloc (GFP_KERNEL ); +if (!backref_cache || !path || !roots || !tmp_ulist) { + ret = -ENOMEM ; goto out; } -while (!end) { -u64 offset_in_extent =0 ; + lockstart =round_down (start,btrfs_inode_sectorsize (inode)); + lockend =round_up (start + len,btrfs_inode_sectorsize (inode)); + prev_extent_end = lockstart; -/* break if the extent we found is outside the range */ -if (em->start >= max ||extent_map_end (em) < off) -break ; +lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); -/* -* get_extent may return an extent that starts before our -* requested range. We have to make sure the ranges -* we return to fiemap always move forward and don't -* overlap, so adjust the offsets here -*/ - em_start =max (em->start, off); + ret =fiemap_find_last_extent_offset (inode, path, &last_extent_end); +if (ret <0 ) + goto out_unlock; +btrfs_release_path (path); + path->reada =READA_FORWARD ; + ret =fiemap_search_slot (inode, path, lockstart); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) {/* -* record the offset from the start of the extent -* for adjusting the disk offset below. Only do this if the -* extent isn't compressed since our in ram offset may be past -* what we have actually allocated on disk. +* No file extent item found, but we may have delalloc between +* the current offset and i_size. So check for that. */ -if (!test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - offset_in_extent = em_start - em->start; - em_end =extent_map_end (em); - em_len = em_end - em_start; - flags =0 ; -if (em->block_start <EXTENT_MAP_LAST_BYTE ) - disko = em->block_start + offset_in_extent; -else - disko =0 ; + ret =0 ; + goto check_eof_delalloc; + } + +while (prev_extent_end < lockend) { +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 extent_end; +u64 extent_len; +u64 extent_offset =0 ; +u64 extent_gen; +u64 disk_bytenr =0 ; +u64 flags =0 ; +int extent_type; +u8 compression; + +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +break ; + + extent_end =btrfs_file_extent_end (path);/* -* bump off for our next call to get_extent +* The first iteration can leave us at an extent item that ends +* before our range's start. Move to the next item. */ - off =extent_map_end (em); -if (off >= max) - end =1 ; - -if (em->block_start ==EXTENT_MAP_INLINE ) { - flags |= (FIEMAP_EXTENT_DATA_INLINE | -FIEMAP_EXTENT_NOT_ALIGNED ); - }else if (em->block_start ==EXTENT_MAP_DELALLOC ) { - flags |= (FIEMAP_EXTENT_DELALLOC | -FIEMAP_EXTENT_UNKNOWN ); - }else if (fieinfo->fi_extents_max) { -u64 extent_gen; -u64 bytenr = em->block_start - - (em->start - em->orig_start); +if (extent_end <= lockstart) + goto next_item; -/* -* If two extent maps are merged, then their generation -* is set to the maximum between their generations. -* Otherwise its generation matches the one we have in -* corresponding file extent item. If we have a merged -* extent map, don't use its generation to speedup the -* sharedness check below. -*/ -if (test_bit (EXTENT_FLAG_MERGED , &em->flags)) - extent_gen =0 ; -else - extent_gen = em->generation; +/* We have in implicit hole (NO_HOLES feature enabled). */ +if (prev_extent_end < key.offset) { +const u64 range_end =min (key.offset, lockend) -1 ; -/* -* As btrfs supports shared space, this information -* can be exported to userspace tools via -* flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 -* then we're just getting a count and we can skip the -* lookup stuff. -*/ - ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), - bytenr, extent_gen, - roots, tmp_ulist, - backref_cache); -if (ret <0 ) - goto out_free; -if (ret) - flags |=FIEMAP_EXTENT_SHARED ; - ret =0 ; - } -if (test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - flags |=FIEMAP_EXTENT_ENCODED ; -if (test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) - flags |=FIEMAP_EXTENT_UNWRITTEN ; + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + prev_extent_end, range_end); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; + } -free_extent_map (em); - em =NULL ; -if ((em_start >= last) || em_len == (u64)-1 || - (last == (u64)-1 && isize <= em_end)) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; +/* We've reached the end of the fiemap range, stop. */ +if (key.offset >= lockend) { + stopped = true; +break ; + } } -/* now scan forward to see if this is really the last extent. */ - em =get_extent_skip_holes (inode, off, last_for_get_extent); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - goto out; + extent_len = extent_end - key.offset; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + compression =btrfs_file_extent_compression (leaf, ei); + extent_type =btrfs_file_extent_type (leaf, ei); + extent_gen =btrfs_file_extent_generation (leaf, ei); + +if (extent_type !=BTRFS_FILE_EXTENT_INLINE ) { + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +if (compression ==BTRFS_COMPRESS_NONE ) + extent_offset =btrfs_file_extent_offset (leaf, ei); } -if (!em) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; + +if (compression !=BTRFS_COMPRESS_NONE ) + flags |=FIEMAP_EXTENT_ENCODED ; + +if (extent_type ==BTRFS_FILE_EXTENT_INLINE ) { + flags |=FIEMAP_EXTENT_DATA_INLINE ; + flags |=FIEMAP_EXTENT_NOT_ALIGNED ; + ret =emit_fiemap_extent (fieinfo, &cache, key.offset,0 , + extent_len, flags); + }else if (extent_type ==BTRFS_FILE_EXTENT_PREALLOC ) { + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache, + disk_bytenr, extent_offset, + extent_gen, roots, tmp_ulist, + key.offset, extent_end -1 ); + }else if (disk_bytenr ==0 ) { +/* We have an explicit hole. */ + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + key.offset, extent_end -1 ); + }else { +/* We have a regular extent. */ +if (fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (root, ino, + disk_bytenr, + extent_gen, + roots, + tmp_ulist, + backref_cache); +if (ret <0 ) + goto out_unlock; +else if (ret >0 ) + flags |=FIEMAP_EXTENT_SHARED ; + } + + ret =emit_fiemap_extent (fieinfo, &cache, key.offset, + disk_bytenr + extent_offset, + extent_len, flags); } - ret =emit_fiemap_extent (fieinfo, &cache, em_start, disko, - em_len, flags); -if (ret) { -if (ret ==1 ) - ret =0 ; - goto out_free; + +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; } + prev_extent_end = extent_end; +next_item:if (fatal_signal_pending (current)) { ret = -EINTR ; - goto out_free; + goto out_unlock; } + + ret =fiemap_next_leaf_item (inode, path); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* No more file extent items for this inode. */ +break ; + } +cond_resched (); } -out_free: -if (!ret) - ret =emit_last_fiemap_cache (fieinfo, &cache); -free_extent_map (em); -out: -unlock_extent_cached (&inode->io_tree, start, start + len -1 , - &cached_state); -out_free_ulist: +check_eof_delalloc: +/* +* Release (and free) the path before emitting any final entries to +* fiemap_fill_next_extent() to keep lockdep happy. This is because +* once we find no more file extent items exist, we may have a +* non-cloned leaf, and fiemap_fill_next_extent() can trigger page +* faults when copying data to the user space buffer. +*/ +btrfs_free_path (path); + path =NULL ; + +if (!stopped && prev_extent_end < lockend) { + ret =fiemap_process_hole (inode, fieinfo, &cache, backref_cache, +0 ,0 ,0 , roots, tmp_ulist, + prev_extent_end, lockend -1 ); +if (ret <0 ) + goto out_unlock; + prev_extent_end = lockend; + } + +if (cache.cached && cache.offset + cache.len >= last_extent_end) { +const u64 i_size =i_size_read (&inode->vfs_inode); + +if (prev_extent_end < i_size) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, + prev_extent_end, + i_size -1 , + &delalloc_start, + &delalloc_end); +if (!delalloc) + cache.flags |=FIEMAP_EXTENT_LAST ; + }else { + cache.flags |=FIEMAP_EXTENT_LAST ; + } + } + + ret =emit_last_fiemap_cache (fieinfo, &cache); + +out_unlock: +unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); +out:kfree (backref_cache);btrfs_free_path (path);ulist_free (roots);fs/btrfs/file.c +8
-8@@ -3602,10 +3602,10 @@ static long btrfs_fallocate(struct file *file, int mode, }/* -* Helper for have_delalloc_in_range(). Find a subrange in a given range that -* has unflushed and/or flushing delalloc. There might be other adjacent -* subranges after the one it found, so have_delalloc_in_range() keeps looping -* while it gets adjacent subranges, and merging them together. +* Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range +* that has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so btrfs_find_delalloc_in_range() keeps +* looping while it gets adjacent subranges, and merging them together. */ static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end,u64 *delalloc_start_ret,u64 *delalloc_end_ret)@@ -3740,8 +3740,8 @@ static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and * end offsets of the subrange. */ -static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, -u64 *delalloc_start_ret,u64 *delalloc_end_ret) +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) {u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize);u64 prev_delalloc_end =0 ;@@ -3804,8 +3804,8 @@ static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,u64 delalloc_end;bool delalloc; - delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, - &delalloc_end); + delalloc =btrfs_find_delalloc_in_range (inode, start, end, + &delalloc_start, &delalloc_end);if (delalloc && whence ==SEEK_DATA ) { *start_ret = delalloc_start;return true;fs/btrfs/inode.c +5
-133@@ -7064,133 +7064,6 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,return em; } -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len) -{ -struct extent_map *em; -struct extent_map *hole_em =NULL ; -u64 delalloc_start = start; -u64 end; -u64 delalloc_len; -u64 delalloc_end; -int err =0 ; - - em =btrfs_get_extent (inode,NULL ,0 , start, len); -if (IS_ERR (em)) -return em; -/* -* If our em maps to: -* - a hole or -* - a pre-alloc extent, -* there might actually be delalloc bytes behind it. -*/ -if (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) -return em; -else - hole_em = em; - -/* check to see if we've wrapped (len == -1 or similar) */ - end = start + len; -if (end < start) - end = (u64)-1 ; -else - end -=1 ; - - em =NULL ; - -/* ok, we didn't find anything, lets look for delalloc */ - delalloc_len =count_range_bits (&inode->io_tree, &delalloc_start, - end, len,EXTENT_DELALLOC ,1 ); - delalloc_end = delalloc_start + delalloc_len; -if (delalloc_end < delalloc_start) - delalloc_end = (u64)-1 ; - -/* -* We didn't find anything useful, return the original results from -* get_extent() -*/ -if (delalloc_start > end || delalloc_end <= start) { - em = hole_em; - hole_em =NULL ; - goto out; - } - -/* -* Adjust the delalloc_start to make sure it doesn't go backwards from -* the start they passed in -*/ - delalloc_start =max (start, delalloc_start); - delalloc_len = delalloc_end - delalloc_start; - -if (delalloc_len >0 ) { -u64 hole_start; -u64 hole_len; -const u64 hole_end =extent_map_end (hole_em); - - em =alloc_extent_map (); -if (!em) { - err = -ENOMEM ; - goto out; - } - -ASSERT (hole_em); -/* -* When btrfs_get_extent can't find anything it returns one -* huge hole -* -* Make sure what it found really fits our range, and adjust to -* make sure it is based on the start from the caller -*/ -if (hole_end <= start || hole_em->start > end) { -free_extent_map (hole_em); - hole_em =NULL ; - }else { - hole_start =max (hole_em->start, start); - hole_len = hole_end - hole_start; - } - -if (hole_em && delalloc_start > hole_start) { -/* -* Our hole starts before our delalloc, so we have to -* return just the parts of the hole that go until the -* delalloc starts -*/ - em->len =min (hole_len, delalloc_start - hole_start); - em->start = hole_start; - em->orig_start = hole_start; -/* -* Don't adjust block start at all, it is fixed at -* EXTENT_MAP_HOLE -*/ - em->block_start = hole_em->block_start; - em->block_len = hole_len; -if (test_bit (EXTENT_FLAG_PREALLOC , &hole_em->flags)) -set_bit (EXTENT_FLAG_PREALLOC , &em->flags); - }else { -/* -* Hole is out of passed range or it starts after -* delalloc range -*/ - em->start = delalloc_start; - em->len = delalloc_len; - em->orig_start = delalloc_start; - em->block_start =EXTENT_MAP_DELALLOC ; - em->block_len = delalloc_len; - } - }else { -return hole_em; - } -out: - -free_extent_map (hole_em); -if (err) { -free_extent_map (em); -return ERR_PTR (err); - } -return em; -} - static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, const u64 start, const u64 len,@@ -8265,15 +8138,14 @@ static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,* in the compression of data (in an async thread) and will return * before the compression is done and writeback is started. A second * filemap_fdatawrite_range() is needed to wait for the compression to -* complete and writeback to start. Without this, our user is very -* likely to get stale results, because the extents and extent maps for -* delalloc regions are only allocated when writeback starts. +* complete and writeback to start. We also need to wait for ordered +* extents to complete, because our fiemap implementation uses mainly +* file extent items to list the extents, searching for extent maps +* only for file ranges with holes or prealloc extents to figure out +* if we have delalloc in those ranges. */ if (fieinfo->fi_flags &FIEMAP_FLAG_SYNC ) { - ret =btrfs_fdatawrite_range (inode,0 ,LLONG_MAX ); -if (ret) -return ret;Quoted 8 lines
fs/btrfs/inode.c static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, · line 8151
Quoted 5 lines
fs/btrfs/inode.c +1
-1- ret =filemap_fdatawait_range (inode->i_mapping,0 ,LLONG_MAX ); + ret =btrfs_wait_ordered_range (inode,0 ,LLONG_MAX );if (ret)return ret; }Hmm this bit should be in "btrfs: properly flush delalloc when entering fiemap" instead. Thanks,
Nop, the change is done here for a good reason: before this change, we only needed
to wait for writeback to complete (actually just to start and create the new extent maps),
so that's why that other patch only waits for writeback to complete, just like the generic code.
Quoted 5 lines
After this change we need to wait for ordered extents to complete, since we use the file
extent items to get extent information for fiemap - that's why that change is in this patch.
Got it, you can add
Thanks,
Josef
Re: [PATCH 10/10] btrfs: make fiemap more efficient and accurate reporting extent sharedness
In reply to [email protected]Quoted 94 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current fiemap implementation does not scale very well with the number of extents a file has. This is both because the main algorithm to find out the extents has a high algorithmic complexity and because for each extent we have to check if it's shared. This second part, checking if an extent is shared, is significantly improved by the two previous patches in this patchset, while the first part is improved by this specific patch. Every now and then we get reports from users mentioning fiemap is too slow or even unusable for files with a very large number of extents, such as the two recent reports referred to by the Link tags at the bottom of this change log.
To understand why the part of finding which extents a file has is very inneficient, consider the example of doing a full ranged fiemap against a file that has over 100K extents (normal for example for a file with more than 10G of data and using compression, which limits the extent size to 128K). When we enter fiemap at
extent_fiemap(), the following happens:1) Before entering the main loop, we call
get_extent_skip_holes()to get the first extent map. This leads us tobtrfs_get_extent_fiemap(), which in turn callsbtrfs_get_extent(), to find the first extent map that covers the file range [0,LLONG_MAX).
btrfs_get_extent()will first search the inode's extent map tree, to see if we have an extent map there that covers the range. If it does not find one, then it will search the inode's subvolume b+tree for a fitting file extent item. After finding the file extent item, it will allocate an extent map, fill it in with information extracted from the file extent item, and add it to the inode's extent map tree (which requires a search for insertion in the tree).2) Then we enter the main loop at
extent_fiemap(), emit the details of the extent, and call againget_extent_skip_holes(), with a start offset matching the end of the extent map we previously processed.We end up at
btrfs_get_extent()again, will search the extent map tree and then search the subvolume b+tree for a file extent item if we could not find an extent map in the extent tree. We allocate an extent map, fill it in with the details in the file extent item, and then insert it into the extent map tree (yet another search in this tree).3) The second step is repeated over and over, until we have processed the whole file range. Each iteration ends at
btrfs_get_extent(), which does a red black tree search on the extent map tree, then searches the subvolume b+tree, allocates an extent map and then does another search in the extent map tree in order to insert the extent map.In the best scenario we have all the extent maps already in the extent tree, and so for each extent we do a single search on a red black tree, so we have a complexity of O(n log n).
In the worst scenario we don't have any extent map already loaded in the extent map tree, or have very few already there. In this case the complexity is much higher since we do:
- A red black tree search on the extent map tree, which has O(log n) complexity, initially very fast since the tree is empty or very small, but as we end up allocating extent maps and adding them to the tree when we don't find them there, each subsequent search on the tree gets slower, since it's getting bigger and bigger after each iteration.
- A search on the subvolume b+tree, also O(log n) complexity, but it has items for all inodes in the subvolume, not just items for our inode. Plus on a filesystem with concurrent operations on other inodes, we can block doing the search due to lock contention on b+tree nodes/leaves.
- Allocate an extent map - this can block, and can also fail if we are under serious memory pressure.
- Do another search on the extent maps red black tree, with the goal of inserting the extent map we just allocated. Again, after every iteration this tree is getting bigger by 1 element, so after many iterations the searches are slower and slower.
- We will not need the allocated extent map anymore, so it's pointless to add it to the extent map tree. It's just wasting time and memory.
In short we end up searching the extent map tree multiple times, on a tree that is growing bigger and bigger after each iteration. And besides that we visit the same leaf of the subvolume b+tree many times, since a leaf with the default size of 16K can easily have more than 200 file extent items.
This is very inneficient overall. This patch changes the algorithm to instead iterate over the subvolume b+tree, visiting each leaf only once, and only searching in the extent map tree for file ranges that have holes or prealloc extents, in order to figure out if we have delalloc there. It will never allocate an extent map and add it to the extent map tree. This is very similar to what was previously done for the lseek's hole and data seeking features.
Quoted 8 lines
Also, the current implementation relying on extent maps for figuring out which extents we have is not correct. This is because extent maps can be merged even if they represent different extents - we do this to minimize memory utilization and keep extent map trees smaller. For example if we have two extents that are contiguous on disk, once we load the two extent maps, they get merged into a single one - however if only one of the extents is shared, we end up reporting both as shared or both as not shared, which is incorrect.
Is there any other major usage for extent map now?
I can only think of read, which uses extent map to grab the logical bytenr of the real extent.
In that case, the SHARED flag doesn't make much sense anyway, can we do a cleanup for those flags? Since fiemap/lseek no longer relies on extent map anymore.
Quoted 597 lines
This reproducer triggers that bug:
$ cat fiemap-bug.sh #!/bin/bashDEV=/dev/sdj
MNT=/mnt/sdjmkfs.btrfs -f $DEV
mount $DEV $MNT# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents.xfs_io-f -c "pwrite -S0xab0 256K" \
-c "fsync" \
-c "pwrite -S0xcd256K 256K" \
-c "fsync" \
$MNT/foo# Now clone only the second extent into another file.
xfs_io-f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/fooumount $MNT
Running the reproducer:
$ ./fiemap-bug.sh wrote 262144/262144 bytes at offset 0 256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec) wrote 262144/262144 bytes at offset 262144 256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec) linked 262144/262144 bytes at offset 0 256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec)Filesystem type is: 9123683e File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes) ext: logical_offset: physical_offset: length: expected: flags: 0: 0.. 127: 3328.. 3455: 128: last,eof /mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
This patch is part of a larger patchset that is comprised of the following patches:
btrfs: allow hole and data seeking to be interruptible
btrfs: make hole and data seeking a lot more efficient
btrfs: remove check for impossible block start for an extent map at fiemap
btrfs: remove zero length check when entering fiemap
btrfs: properly flush delalloc when entering fiemap
btrfs: allow fiemap to be interruptible
btrfs: renamebtrfs_check_shared()to a more descriptive name
btrfs: speedup checking for extent sharedness during fiemap
btrfs: skip unnecessary extent buffer sharedness checks during fiemap
btrfs: make fiemap more efficient and accurate reporting extent sharednessThe patchset was tested on a machine running a non-debug kernel (Debian's default config) and compared the tests below on a branch without the patchset versus the same branch with the whole patchset applied.
The following test for a large compressed file without holes:
$ cat fiemap-perf-test.sh #!/bin/bashDEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f $DEV mount -o compress=lzo $DEV $MNT # 40G gives 327680 128K file extents (due to compression). xfs_io -f -c "pwrite -S 0xab -b 1M 0 20G" $MNT/foobar umount $MNT mount -o compress=lzo $DEV $MNT start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata not cached)" start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata cached)" umount $MNTBefore patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 3597 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 2107 milliseconds (metadata cached)After patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 1214 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 684 milliseconds (metadata cached)That's a speedup of about 3x for both cases (no metadata cached and all metadata cached).
The test provided by Pavel (first Link tag at the bottom), which uses files with a large number of holes, was also used to measure the gains, and it consists on a small C program and a shell script to invoke it. The C program is the following:
$ cat pavels-test.c #include <stdio.h> #include <unistd.h> #include <stdlib.h> #include <fcntl.h>#include <sys/stat.h> #include <sys/time.h> #include <sys/ioctl.h> #include <linux/fs.h> #include <linux/fiemap.h> #define FILE_INTERVAL (1<<13) /* 8Kb */ long long interval(struct timeval t1, struct timeval t2) { long long val = 0; val += (t2.tv_usec - t1.tv_usec); val += (t2.tv_sec - t1.tv_sec) * 1000 * 1000; return val; } int main(int argc, char **argv) { struct fiemap fiemap = {}; struct timeval t1, t2; char data = 'a'; struct stat st; int fd, off, file_size = FILE_INTERVAL; if (argc != 3 && argc != 2) { printf("usage: %s <path> [size]\n", argv[0]); return 1; } if (argc == 3) file_size = atoi(argv[2]); if (file_size < FILE_INTERVAL) file_size = FILE_INTERVAL; file_size -= file_size % FILE_INTERVAL; fd = open(argv[1], O_RDWR | O_CREAT | O_TRUNC, 0644); if (fd < 0) { perror("open"); return 1; } for (off = 0; off < file_size; off += FILE_INTERVAL) { if (pwrite(fd, &data, 1, off) != 1) { perror("pwrite"); close(fd); return 1; } } if (ftruncate(fd, file_size)) { perror("ftruncate"); close(fd); return 1; } if (fstat(fd, &st) < 0) { perror("fstat"); close(fd); return 1; } printf("size: %ld\n", st.st_size); printf("actual size: %ld\n", st.st_blocks * 512); fiemap.fm_length = FIEMAP_MAX_OFFSET; gettimeofday(&t1, NULL); if (ioctl(fd, FS_IOC_FIEMAP, &fiemap) < 0) { perror("fiemap"); close(fd); return 1; } gettimeofday(&t2, NULL); printf("fiemap: fm_mapped_extents = %d\n", fiemap.fm_mapped_extents); printf("time = %lld us\n", interval(t1, t2)); close(fd); return 0; } $ gcc -o pavels_test pavels_test.cAnd the wrapper shell script:
$ cat fiemap-pavels-test.sh#!/bin/bash DEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f -O no-holes $DEV mount $DEV $MNT echo echo "*********** 256M ***********" echo ./pavels-test $MNT/testfile $((1 << 28)) echo ./pavels-test $MNT/testfile $((1 << 28)) echo echo "*********** 512M ***********" echo ./pavels-test $MNT/testfile $((1 << 29)) echo ./pavels-test $MNT/testfile $((1 << 29)) echo echo "*********** 1G ***********" echo ./pavels-test $MNT/testfile $((1 << 30)) echo ./pavels-test $MNT/testfile $((1 << 30)) umount $MNTRunning his reproducer before applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4003133 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4895330 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 30123675 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 33450934 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 224924074 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 217239242 usRunning it after applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29475 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29307 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 58996 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 59115 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 116251 time = 124141 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 119387 usThe speedup is massive, both on the first fiemap call and on the second one as well, as his test creates files with many holes and small extents (every extent follows a hole and precedes another hole).
For the 256M file we go from 4 seconds down to 29 milliseconds in the first run, and then from 4.9 seconds down to 29 milliseconds again in the second run, a speedup of 138x and 169x, respectively.
For the 512M file we go from 30.1 seconds down to 59 milliseconds in the first run, and then from 33.5 seconds down to 59 milliseconds again in the second run, a speedup of 510x and 568x, respectively.
For the 1G file, we go from 225 seconds down to 124 milliseconds in the first run, and then from 217 seconds down to 119 milliseconds in the second run, a speedup of 1815x and 1824x, respectively.
Reported-by Pavel Tikhomirov <[email protected]>Reported-by Dominique MARTINET <[email protected]>Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/ctree.h | 4 +- fs/btrfs/extent_io.c | 714 +++++++++++++++++++++++++++++-------------- fs/btrfs/file.c | 16 +- fs/btrfs/inode.c | 140 +-------- 4 files changed, 506 insertions(+), 368 deletions(-)fs/btrfs/ctree.h +2
-2@@ -3402,8 +3402,6 @@ unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio, u64 start, u64 end);int btrfs_check_data_csum (struct inode *inode,struct btrfs_bio *bbio,u32 bio_offset,struct page *page,u32 pgoff); -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len); noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,u64 *orig_start,u64 *orig_block_len,u64 *ram_bytes,bool strict);@@ -3583,6 +3581,8 @@ int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end);int btrfs_check_nocow_lock (struct btrfs_inode *inode,loff_t pos,size_t *write_bytes);void btrfs_check_nocow_unlock (struct btrfs_inode *inode); +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret);/* tree-defrag.c */ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,fs/btrfs/extent_io.c +92
-79@@ -5353,42 +5353,6 @@ int try_release_extent_mapping(struct page *page, gfp_t mask)return try_release_extent_state (tree, page, mask); } -/* -* helper function for fiemap, which doesn't want to see any holes. -* This maps until we find something past 'last' -*/ -static struct extent_map *get_extent_skip_holes (struct btrfs_inode *inode, -u64 offset,u64 last) -{ -u64 sectorsize =btrfs_inode_sectorsize (inode); -struct extent_map *em; -u64 len; - -if (offset >= last) -return NULL ; - -while (1 ) { - len = last - offset; -if (len ==0 ) -break ; - len =ALIGN (len, sectorsize); - em =btrfs_get_extent_fiemap (inode, offset, len); -if (IS_ERR (em)) -return em; - -/* if this isn't a hole return it */ -if (em->block_start !=EXTENT_MAP_HOLE ) -return em; - -/* this is a hole, advance to the next extent */ - offset =extent_map_end (em); -free_extent_map (em); -if (offset >= last) -break ; - } -return NULL ; -} -/* * To cache previous fiemap extent * @@ -5418,6 +5382,9 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, {int ret =0 ; +/* Set at the end of extent_fiemap(). */ +ASSERT ((flags &FIEMAP_EXTENT_LAST ) ==0 ); +if (!cache->cached) goto assign;@@ -5446,11 +5413,10 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,*/ if (cache->offset + cache->len == offset && cache->phys + cache->len == phys && - (cache->flags & ~FIEMAP_EXTENT_LAST ) == - (flags & ~FIEMAP_EXTENT_LAST )) { + cache->flags == flags) { cache->len += len; cache->flags |= flags; - goto try_submit_last; +return 0 ; }/* Not mergeable, need to submit cached one */ @@ -5465,13 +5431,8 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, cache->phys = phys; cache->len = len; cache->flags = flags; -try_submit_last: -if (cache->flags &FIEMAP_EXTENT_LAST ) { - ret =fiemap_fill_next_extent (fieinfo, cache->offset, - cache->phys, cache->len, cache->flags); - cache->cached = false; - } -return ret; + +return 0 ; }/* @@ -5501,229 +5462,534 @@ static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,return ret; } -int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, -u64 start,u64 len) +static int fiemap_next_leaf_item (struct btrfs_inode *inode, +struct btrfs_path *path) { -int ret =0 ; -u64 off; -u64 max = start + len; -u32 flags =0 ; -u32 found_type; -u64 last; -u64 last_for_get_extent =0 ; -u64 disko =0 ; -u64 isize =i_size_read (&inode->vfs_inode); -struct btrfs_key found_key; -struct extent_map *em =NULL ; -struct extent_state *cached_state =NULL ; -struct btrfs_path *path; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; + + path->slots[0 ]++; +if (path->slots[0 ] <btrfs_header_nritems (path->nodes[0 ])) +return 0 ; + + ret =btrfs_next_leaf (inode->root, path); +if (ret !=0 ) +return ret; + +/* +* Don't bother with cloning if there are no more file extent items for +* our inode. +*/ +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid !=btrfs_ino (inode) || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; + +/* See the comment at fiemap_search_slot() about why we clone. */ + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot; + +return 0 ; +} + +/* +* Search for the first file extent item that starts at a given file offset or +* the one that starts immediately before that offset. +* Returns: 0 on success, < 0 on error, 1 if not found. +*/ +static int fiemap_search_slot (struct btrfs_inode *inode,struct btrfs_path *path, +u64 file_offset) +{ +const u64 ino =btrfs_ino (inode);struct btrfs_root *root = inode->root; -struct fiemap_cache cache = {0 }; -struct btrfs_backref_shared_cache *backref_cache; -struct ulist *roots; -struct ulist *tmp_ulist; -int end =0 ; -u64 em_start =0 ; -u64 em_len =0 ; -u64 em_end =0 ; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; - backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); - path =btrfs_alloc_path (); - roots =ulist_alloc (GFP_KERNEL ); - tmp_ulist =ulist_alloc (GFP_KERNEL ); -if (!backref_cache || !path || !roots || !tmp_ulist) { - ret = -ENOMEM ; - goto out_free_ulist; + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = file_offset; + + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) +return ret; + +if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } + +if (path->slots[0 ] >=btrfs_header_nritems (path->nodes[0 ])) { + ret =btrfs_next_leaf (root, path); +if (ret !=0 ) +return ret; + +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; }/* -* We can't initialize that to 'start' as this could miss extents due -* to extent item merging +* We clone the leaf and use it during fiemap. This is because while +* using the leaf we do expensive things like checking if an extent is +* shared, which can take a long time. In order to prevent blocking +* other tasks for too long, we use a clone of the leaf. We have locked +* the file range in the inode's io tree, so we know none of our file +* extent items can change. This way we avoid blocking other tasks that +* want to insert items for other inodes in the same leaf or b+tree +* rebalance operations (triggered for example when someone is trying +* to push items into this leaf when trying to insert an item in a +* neighbour leaf). +* We also need the private clone because holding a read lock on an +* extent buffer of the subvolume's b+tree will make lockdep unhappy +* when we call fiemap_fill_next_extent(), because that may cause a page +* fault when filling the user space buffer with fiemap data. */ - off =0 ;
fs/btrfs/extent_io.c static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, · line 5564
Quoted 10 lines
fs/btrfs/extent_io.c +8
-2- start =round_down (start,btrfs_inode_sectorsize (inode)); - len =round_up (max,btrfs_inode_sectorsize (inode)) - start; + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot;
Although this is correct, it still looks a little tricky.
We rely on btrfs_release_path() to release all tree blocks in the subvolume tree, including unlocking the tree blocks, thus path->locks[0] is also 0, meaning next time we call btrfs_release_path() we won't try to unlock the cloned eb.
But I'd say it's still pretty tricky, and unfortunately I don't have any better alternative.
fs/btrfs/extent_io.c static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, · line 5574
Quoted 10 lines
fs/btrfs/extent_io.c +10
-0+ +return 0 ; +} + +/* +* Process a range which is a hole or a prealloc extent in the inode's subvolume +* btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc +* extent. The end offset (@end) is inclusive. +*/ +static int fiemap_process_hole(struct btrfs_inode *inode,
Does the name still make sense as we're handling both hole and prealloc range?
And I always find the delalloc search a big pain during lseek/fiemap.
I guess except using certain flags, there is some hard requirement for delalloc range reporting?
Quoted 152 lines
fs/btrfs/extent_io.c +119
-27+struct fiemap_extent_info *fieinfo, + struct fiemap_cache *cache, + struct btrfs_backref_shared_cache *backref_cache, + u64 disk_bytenr, u64 extent_offset, + u64 extent_gen, + struct ulist *roots, struct ulist *tmp_ulist, + u64 start, u64 end) +{ + const u64 i_size = i_size_read(&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +u64 cur_offset = start; +u64 last_delalloc_end =0 ; +u32 prealloc_flags =FIEMAP_EXTENT_UNWRITTEN ; +bool checked_extent_shared = false; +int ret;/* -* lookup the last file extent. We're not using i_size here -* because there might be preallocation past i_size +* There can be no delalloc past i_size, so don't waste time looking for +* it beyond i_size. */ - ret =btrfs_lookup_file_extent (NULL , root, path,btrfs_ino (inode),-1 , -0 ); -if (ret <0 ) { - goto out_free_ulist; - }else { -WARN_ON (!ret); -if (ret ==1 ) - ret =0 ; - } +while (cur_offset < end && cur_offset < i_size) { +u64 delalloc_start; +u64 delalloc_end; +u64 prealloc_start; +u64 prealloc_len =0 ; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; - path->slots[0 ]--; -btrfs_item_key_to_cpu (path->nodes[0 ], &found_key, path->slots[0 ]); - found_type = found_key.type; - -/* No extents, but there might be delalloc bits */ -if (found_key.objectid !=btrfs_ino (inode) || - found_type !=BTRFS_EXTENT_DATA_KEY ) { -/* have to trust i_size as the end */ - last = (u64)-1 ; - last_for_get_extent = isize; - }else {/* -* remember the start of the last extent. There are a -* bunch of different factors that go into the length of the -* extent, so its much less complex to remember where it started +* If this is a prealloc extent we have to report every section +* of it that has no delalloc. */ - last = found_key.offset; - last_for_get_extent = last +1 ; +if (disk_bytenr !=0 ) { +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = delalloc_start - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = delalloc_start - prealloc_start; + } + } + +if (prealloc_len >0 ) { +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + + checked_extent_shared = true; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + extent_offset += prealloc_len; + } + + ret =emit_fiemap_extent (fieinfo, cache, delalloc_start,0 , + delalloc_end +1 - delalloc_start, +FIEMAP_EXTENT_DELALLOC | +FIEMAP_EXTENT_UNKNOWN ); +if (ret) +return ret; + + last_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; + extent_offset += cur_offset - delalloc_start; +cond_resched (); + } + +/* +* Either we found no delalloc for the whole prealloc extent or we have +* a prealloc extent that spans i_size or starts at or after i_size. +*/ +if (disk_bytenr !=0 && last_delalloc_end < end) { +u64 prealloc_start; +u64 prealloc_len; + +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = end +1 - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = end +1 - prealloc_start; + } + +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + } + +return 0 ; +} + +static int fiemap_find_last_extent_offset (struct btrfs_inode *inode, +struct btrfs_path *path, +u64 *last_extent_end_ret) +{ +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root;
fs/btrfs/extent_io.c static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, · line 5709
Quoted 10 lines
fs/btrfs/extent_io.c +10
-0+struct extent_buffer *leaf; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 disk_bytenr; +int ret; + +/* +* Lookup the last file extent. We're not using i_size here because +* there might be preallocation past i_size. +*/
I'm wondering how could this happen?
Normally if we're truncating an inode, the extents starting after round_up(i_size, sectorsize) should be dropped.
Or if we later enlarge the inode, we may hit old extents and read out some stale data other than expected zeros.
Thus searching using round_up(i_size, sectorsize) should still let us to reach the slot after the last file extent.
Or did I miss something?
Thanks,
Qu
Quoted 603 lines
fs/btrfs/extent_io.c +251
-116+ ret =btrfs_lookup_file_extent (NULL , root, path, ino, (u64)-1 ,0 ); +/* There can't be a file extent item at offset (u64)-1 */ +ASSERT (ret !=0 ); +if (ret <0 ) +return ret; + +/* +* For a non-existing key, btrfs_search_slot() always leaves us at a +* slot > 0, except if the btree is empty, which is impossible because +* at least it has the inode item for this inode and all the items for +* the root inode 256. +*/ +ASSERT (path->slots[0 ] >0 ); + path->slots[0 ]--; + leaf = path->nodes[0 ]; +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) { +/* No file extent items in the subvolume tree. */ + *last_extent_end_ret =0 ; +return 0 ; } -btrfs_release_path (path);/* -* we might have some extents allocated but more delalloc past those -* extents. so, we trust isize unless the start of the last extent is -* beyond isize +* For an inline extent, the disk_bytenr is where inline data starts at, +* so first check if we have an inline extent item before checking if we +* have an implicit hole (disk_bytenr == 0). */ -if (last < isize) { - last = (u64)-1 ; - last_for_get_extent = isize; + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); +if (btrfs_file_extent_type (leaf, ei) ==BTRFS_FILE_EXTENT_INLINE ) { + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; } -lock_extent_bits (&inode->io_tree, start, start + len -1 , - &cached_state); +/* +* Find the last file extent item that is not a hole (when NO_HOLES is +* not enabled). This should take at most 2 iterations in the worst +* case: we have one hole file extent item at slot 0 of a leaf and +* another hole file extent item as the last item in the previous leaf. +* This is because we merge file extent items that represent holes. +*/ + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +while (disk_bytenr ==0 ) { + ret =btrfs_previous_item (root, path, ino,BTRFS_EXTENT_DATA_KEY ); +if (ret <0 ) { +return ret; + }else if (ret >0 ) { +/* No file extent items that are not holes. */ + *last_extent_end_ret =0 ; +return 0 ; + } + leaf = path->nodes[0 ]; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); + } - em =get_extent_skip_holes (inode, start, last_for_get_extent); -if (!em) - goto out; -if (IS_ERR (em)) { - ret =PTR_ERR (em); + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; +} + +int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, +u64 start,u64 len) +{ +const u64 ino =btrfs_ino (inode); +struct extent_state *cached_state =NULL ; +struct btrfs_path *path; +struct btrfs_root *root = inode->root; +struct fiemap_cache cache = {0 }; +struct btrfs_backref_shared_cache *backref_cache; +struct ulist *roots; +struct ulist *tmp_ulist; +u64 last_extent_end; +u64 prev_extent_end; +u64 lockstart; +u64 lockend; +bool stopped = false; +int ret; + + backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); + path =btrfs_alloc_path (); + roots =ulist_alloc (GFP_KERNEL ); + tmp_ulist =ulist_alloc (GFP_KERNEL ); +if (!backref_cache || !path || !roots || !tmp_ulist) { + ret = -ENOMEM ; goto out; } -while (!end) { -u64 offset_in_extent =0 ; + lockstart =round_down (start,btrfs_inode_sectorsize (inode)); + lockend =round_up (start + len,btrfs_inode_sectorsize (inode)); + prev_extent_end = lockstart; -/* break if the extent we found is outside the range */ -if (em->start >= max ||extent_map_end (em) < off) -break ; +lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); -/* -* get_extent may return an extent that starts before our -* requested range. We have to make sure the ranges -* we return to fiemap always move forward and don't -* overlap, so adjust the offsets here -*/ - em_start =max (em->start, off); + ret =fiemap_find_last_extent_offset (inode, path, &last_extent_end); +if (ret <0 ) + goto out_unlock; +btrfs_release_path (path); + path->reada =READA_FORWARD ; + ret =fiemap_search_slot (inode, path, lockstart); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) {/* -* record the offset from the start of the extent -* for adjusting the disk offset below. Only do this if the -* extent isn't compressed since our in ram offset may be past -* what we have actually allocated on disk. +* No file extent item found, but we may have delalloc between +* the current offset and i_size. So check for that. */ -if (!test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - offset_in_extent = em_start - em->start; - em_end =extent_map_end (em); - em_len = em_end - em_start; - flags =0 ; -if (em->block_start <EXTENT_MAP_LAST_BYTE ) - disko = em->block_start + offset_in_extent; -else - disko =0 ; + ret =0 ; + goto check_eof_delalloc; + } + +while (prev_extent_end < lockend) { +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 extent_end; +u64 extent_len; +u64 extent_offset =0 ; +u64 extent_gen; +u64 disk_bytenr =0 ; +u64 flags =0 ; +int extent_type; +u8 compression; + +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +break ; + + extent_end =btrfs_file_extent_end (path);/* -* bump off for our next call to get_extent +* The first iteration can leave us at an extent item that ends +* before our range's start. Move to the next item. */ - off =extent_map_end (em); -if (off >= max) - end =1 ; - -if (em->block_start ==EXTENT_MAP_INLINE ) { - flags |= (FIEMAP_EXTENT_DATA_INLINE | -FIEMAP_EXTENT_NOT_ALIGNED ); - }else if (em->block_start ==EXTENT_MAP_DELALLOC ) { - flags |= (FIEMAP_EXTENT_DELALLOC | -FIEMAP_EXTENT_UNKNOWN ); - }else if (fieinfo->fi_extents_max) { -u64 extent_gen; -u64 bytenr = em->block_start - - (em->start - em->orig_start); +if (extent_end <= lockstart) + goto next_item; -/* -* If two extent maps are merged, then their generation -* is set to the maximum between their generations. -* Otherwise its generation matches the one we have in -* corresponding file extent item. If we have a merged -* extent map, don't use its generation to speedup the -* sharedness check below. -*/ -if (test_bit (EXTENT_FLAG_MERGED , &em->flags)) - extent_gen =0 ; -else - extent_gen = em->generation; +/* We have in implicit hole (NO_HOLES feature enabled). */ +if (prev_extent_end < key.offset) { +const u64 range_end =min (key.offset, lockend) -1 ; -/* -* As btrfs supports shared space, this information -* can be exported to userspace tools via -* flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 -* then we're just getting a count and we can skip the -* lookup stuff. -*/ - ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), - bytenr, extent_gen, - roots, tmp_ulist, - backref_cache); -if (ret <0 ) - goto out_free; -if (ret) - flags |=FIEMAP_EXTENT_SHARED ; - ret =0 ; - } -if (test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - flags |=FIEMAP_EXTENT_ENCODED ; -if (test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) - flags |=FIEMAP_EXTENT_UNWRITTEN ; + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + prev_extent_end, range_end); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; + } -free_extent_map (em); - em =NULL ; -if ((em_start >= last) || em_len == (u64)-1 || - (last == (u64)-1 && isize <= em_end)) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; +/* We've reached the end of the fiemap range, stop. */ +if (key.offset >= lockend) { + stopped = true; +break ; + } } -/* now scan forward to see if this is really the last extent. */ - em =get_extent_skip_holes (inode, off, last_for_get_extent); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - goto out; + extent_len = extent_end - key.offset; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + compression =btrfs_file_extent_compression (leaf, ei); + extent_type =btrfs_file_extent_type (leaf, ei); + extent_gen =btrfs_file_extent_generation (leaf, ei); + +if (extent_type !=BTRFS_FILE_EXTENT_INLINE ) { + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +if (compression ==BTRFS_COMPRESS_NONE ) + extent_offset =btrfs_file_extent_offset (leaf, ei); } -if (!em) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; + +if (compression !=BTRFS_COMPRESS_NONE ) + flags |=FIEMAP_EXTENT_ENCODED ; + +if (extent_type ==BTRFS_FILE_EXTENT_INLINE ) { + flags |=FIEMAP_EXTENT_DATA_INLINE ; + flags |=FIEMAP_EXTENT_NOT_ALIGNED ; + ret =emit_fiemap_extent (fieinfo, &cache, key.offset,0 , + extent_len, flags); + }else if (extent_type ==BTRFS_FILE_EXTENT_PREALLOC ) { + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache, + disk_bytenr, extent_offset, + extent_gen, roots, tmp_ulist, + key.offset, extent_end -1 ); + }else if (disk_bytenr ==0 ) { +/* We have an explicit hole. */ + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + key.offset, extent_end -1 ); + }else { +/* We have a regular extent. */ +if (fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (root, ino, + disk_bytenr, + extent_gen, + roots, + tmp_ulist, + backref_cache); +if (ret <0 ) + goto out_unlock; +else if (ret >0 ) + flags |=FIEMAP_EXTENT_SHARED ; + } + + ret =emit_fiemap_extent (fieinfo, &cache, key.offset, + disk_bytenr + extent_offset, + extent_len, flags); } - ret =emit_fiemap_extent (fieinfo, &cache, em_start, disko, - em_len, flags); -if (ret) { -if (ret ==1 ) - ret =0 ; - goto out_free; + +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; } + prev_extent_end = extent_end; +next_item:if (fatal_signal_pending (current)) { ret = -EINTR ; - goto out_free; + goto out_unlock; } + + ret =fiemap_next_leaf_item (inode, path); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* No more file extent items for this inode. */ +break ; + } +cond_resched (); } -out_free: -if (!ret) - ret =emit_last_fiemap_cache (fieinfo, &cache); -free_extent_map (em); -out: -unlock_extent_cached (&inode->io_tree, start, start + len -1 , - &cached_state); -out_free_ulist: +check_eof_delalloc: +/* +* Release (and free) the path before emitting any final entries to +* fiemap_fill_next_extent() to keep lockdep happy. This is because +* once we find no more file extent items exist, we may have a +* non-cloned leaf, and fiemap_fill_next_extent() can trigger page +* faults when copying data to the user space buffer. +*/ +btrfs_free_path (path); + path =NULL ; + +if (!stopped && prev_extent_end < lockend) { + ret =fiemap_process_hole (inode, fieinfo, &cache, backref_cache, +0 ,0 ,0 , roots, tmp_ulist, + prev_extent_end, lockend -1 ); +if (ret <0 ) + goto out_unlock; + prev_extent_end = lockend; + } + +if (cache.cached && cache.offset + cache.len >= last_extent_end) { +const u64 i_size =i_size_read (&inode->vfs_inode); + +if (prev_extent_end < i_size) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, + prev_extent_end, + i_size -1 , + &delalloc_start, + &delalloc_end); +if (!delalloc) + cache.flags |=FIEMAP_EXTENT_LAST ; + }else { + cache.flags |=FIEMAP_EXTENT_LAST ; + } + } + + ret =emit_last_fiemap_cache (fieinfo, &cache); + +out_unlock: +unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); +out:kfree (backref_cache);btrfs_free_path (path);ulist_free (roots);fs/btrfs/file.c +8
-8@@ -3602,10 +3602,10 @@ static long btrfs_fallocate(struct file *file, int mode, }/* -* Helper for have_delalloc_in_range(). Find a subrange in a given range that -* has unflushed and/or flushing delalloc. There might be other adjacent -* subranges after the one it found, so have_delalloc_in_range() keeps looping -* while it gets adjacent subranges, and merging them together. +* Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range +* that has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so btrfs_find_delalloc_in_range() keeps +* looping while it gets adjacent subranges, and merging them together. */ static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end,u64 *delalloc_start_ret,u64 *delalloc_end_ret)@@ -3740,8 +3740,8 @@ static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and * end offsets of the subrange. */ -static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, -u64 *delalloc_start_ret,u64 *delalloc_end_ret) +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) {u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize);u64 prev_delalloc_end =0 ;@@ -3804,8 +3804,8 @@ static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,u64 delalloc_end;bool delalloc; - delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, - &delalloc_end); + delalloc =btrfs_find_delalloc_in_range (inode, start, end, + &delalloc_start, &delalloc_end);if (delalloc && whence ==SEEK_DATA ) { *start_ret = delalloc_start;return true;fs/btrfs/inode.c +6
-134@@ -7064,133 +7064,6 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,return em; } -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len) -{ -struct extent_map *em; -struct extent_map *hole_em =NULL ; -u64 delalloc_start = start; -u64 end; -u64 delalloc_len; -u64 delalloc_end; -int err =0 ; - - em =btrfs_get_extent (inode,NULL ,0 , start, len); -if (IS_ERR (em)) -return em; -/* -* If our em maps to: -* - a hole or -* - a pre-alloc extent, -* there might actually be delalloc bytes behind it. -*/ -if (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) -return em; -else - hole_em = em; - -/* check to see if we've wrapped (len == -1 or similar) */ - end = start + len; -if (end < start) - end = (u64)-1 ; -else - end -=1 ; - - em =NULL ; - -/* ok, we didn't find anything, lets look for delalloc */ - delalloc_len =count_range_bits (&inode->io_tree, &delalloc_start, - end, len,EXTENT_DELALLOC ,1 ); - delalloc_end = delalloc_start + delalloc_len; -if (delalloc_end < delalloc_start) - delalloc_end = (u64)-1 ; - -/* -* We didn't find anything useful, return the original results from -* get_extent() -*/ -if (delalloc_start > end || delalloc_end <= start) { - em = hole_em; - hole_em =NULL ; - goto out; - } - -/* -* Adjust the delalloc_start to make sure it doesn't go backwards from -* the start they passed in -*/ - delalloc_start =max (start, delalloc_start); - delalloc_len = delalloc_end - delalloc_start; - -if (delalloc_len >0 ) { -u64 hole_start; -u64 hole_len; -const u64 hole_end =extent_map_end (hole_em); - - em =alloc_extent_map (); -if (!em) { - err = -ENOMEM ; - goto out; - } - -ASSERT (hole_em); -/* -* When btrfs_get_extent can't find anything it returns one -* huge hole -* -* Make sure what it found really fits our range, and adjust to -* make sure it is based on the start from the caller -*/ -if (hole_end <= start || hole_em->start > end) { -free_extent_map (hole_em); - hole_em =NULL ; - }else { - hole_start =max (hole_em->start, start); - hole_len = hole_end - hole_start; - } - -if (hole_em && delalloc_start > hole_start) { -/* -* Our hole starts before our delalloc, so we have to -* return just the parts of the hole that go until the -* delalloc starts -*/ - em->len =min (hole_len, delalloc_start - hole_start); - em->start = hole_start; - em->orig_start = hole_start; -/* -* Don't adjust block start at all, it is fixed at -* EXTENT_MAP_HOLE -*/ - em->block_start = hole_em->block_start; - em->block_len = hole_len; -if (test_bit (EXTENT_FLAG_PREALLOC , &hole_em->flags)) -set_bit (EXTENT_FLAG_PREALLOC , &em->flags); - }else { -/* -* Hole is out of passed range or it starts after -* delalloc range -*/ - em->start = delalloc_start; - em->len = delalloc_len; - em->orig_start = delalloc_start; - em->block_start =EXTENT_MAP_DELALLOC ; - em->block_len = delalloc_len; - } - }else { -return hole_em; - } -out: - -free_extent_map (hole_em); -if (err) { -free_extent_map (em); -return ERR_PTR (err); - } -return em; -} - static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, const u64 start, const u64 len,@@ -8265,15 +8138,14 @@ static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,* in the compression of data (in an async thread) and will return * before the compression is done and writeback is started. A second * filemap_fdatawrite_range() is needed to wait for the compression to -* complete and writeback to start. Without this, our user is very -* likely to get stale results, because the extents and extent maps for -* delalloc regions are only allocated when writeback starts. +* complete and writeback to start. We also need to wait for ordered +* extents to complete, because our fiemap implementation uses mainly +* file extent items to list the extents, searching for extent maps +* only for file ranges with holes or prealloc extents to figure out +* if we have delalloc in those ranges. */ if (fieinfo->fi_flags &FIEMAP_FLAG_SYNC ) { - ret =btrfs_fdatawrite_range (inode,0 ,LLONG_MAX ); -if (ret) -return ret; - ret =filemap_fdatawait_range (inode->i_mapping,0 ,LLONG_MAX ); + ret =btrfs_wait_ordered_range (inode,0 ,LLONG_MAX );if (ret)return ret; }
Re: [PATCH 10/10] btrfs: make fiemap more efficient and accurate reporting extent sharedness
Quoted 107 lines · Qu Wenruo
Quoted 94 lines · On 2022/9/1 21:18, [email protected] wrote:
From: Filipe Manana <[email protected]>
The current fiemap implementation does not scale very well with the number of extents a file has. This is both because the main algorithm to find out the extents has a high algorithmic complexity and because for each extent we have to check if it's shared. This second part, checking if an extent is shared, is significantly improved by the two previous patches in this patchset, while the first part is improved by this specific patch. Every now and then we get reports from users mentioning fiemap is too slow or even unusable for files with a very large number of extents, such as the two recent reports referred to by the Link tags at the bottom of this change log.
To understand why the part of finding which extents a file has is very inneficient, consider the example of doing a full ranged fiemap against a file that has over 100K extents (normal for example for a file with more than 10G of data and using compression, which limits the extent size to 128K). When we enter fiemap at
extent_fiemap(), the following happens:1) Before entering the main loop, we call
get_extent_skip_holes()to get the first extent map. This leads us tobtrfs_get_extent_fiemap(), which in turn callsbtrfs_get_extent(), to find the first extent map that covers the file range [0,LLONG_MAX).
btrfs_get_extent()will first search the inode's extent map tree, to see if we have an extent map there that covers the range. If it does not find one, then it will search the inode's subvolume b+tree for a fitting file extent item. After finding the file extent item, it will allocate an extent map, fill it in with information extracted from the file extent item, and add it to the inode's extent map tree (which requires a search for insertion in the tree).2) Then we enter the main loop at
extent_fiemap(), emit the details of the extent, and call againget_extent_skip_holes(), with a start offset matching the end of the extent map we previously processed.We end up at
btrfs_get_extent()again, will search the extent map tree and then search the subvolume b+tree for a file extent item if we could not find an extent map in the extent tree. We allocate an extent map, fill it in with the details in the file extent item, and then insert it into the extent map tree (yet another search in this tree).3) The second step is repeated over and over, until we have processed the whole file range. Each iteration ends at
btrfs_get_extent(), which does a red black tree search on the extent map tree, then searches the subvolume b+tree, allocates an extent map and then does another search in the extent map tree in order to insert the extent map.In the best scenario we have all the extent maps already in the extent tree, and so for each extent we do a single search on a red black tree, so we have a complexity of O(n log n).
In the worst scenario we don't have any extent map already loaded in the extent map tree, or have very few already there. In this case the complexity is much higher since we do:
- A red black tree search on the extent map tree, which has O(log n) complexity, initially very fast since the tree is empty or very small, but as we end up allocating extent maps and adding them to the tree when we don't find them there, each subsequent search on the tree gets slower, since it's getting bigger and bigger after each iteration.
- A search on the subvolume b+tree, also O(log n) complexity, but it has items for all inodes in the subvolume, not just items for our inode. Plus on a filesystem with concurrent operations on other inodes, we can block doing the search due to lock contention on b+tree nodes/leaves.
- Allocate an extent map - this can block, and can also fail if we are under serious memory pressure.
- Do another search on the extent maps red black tree, with the goal of inserting the extent map we just allocated. Again, after every iteration this tree is getting bigger by 1 element, so after many iterations the searches are slower and slower.
- We will not need the allocated extent map anymore, so it's pointless to add it to the extent map tree. It's just wasting time and memory.
In short we end up searching the extent map tree multiple times, on a tree that is growing bigger and bigger after each iteration. And besides that we visit the same leaf of the subvolume b+tree many times, since a leaf with the default size of 16K can easily have more than 200 file extent items.
This is very inneficient overall. This patch changes the algorithm to instead iterate over the subvolume b+tree, visiting each leaf only once, and only searching in the extent map tree for file ranges that have holes or prealloc extents, in order to figure out if we have delalloc there. It will never allocate an extent map and add it to the extent map tree. This is very similar to what was previously done for the lseek's hole and data seeking features.
Quoted 8 lines
Also, the current implementation relying on extent maps for figuring out which extents we have is not correct. This is because extent maps can be merged even if they represent different extents - we do this to minimize memory utilization and keep extent map trees smaller. For example if we have two extents that are contiguous on disk, once we load the two extent maps, they get merged into a single one - however if only one of the extents is shared, we end up reporting both as shared or both as not shared, which is incorrect.
Quoted 1 line
Is there any other major usage for extent map now?
For fsync at least is very important.
Also for reads it's nice to not have to go to the b+tree.
If someone reads several pages of an extent, being able to get it directly from the extent map tree is faster than having to go to the btree for every read.
Extent map trees are per inode, but subvolume b+trees can have a lot of concurrent write and read access.
Quoted 7 lines
I can only think of read, which uses extent map to grab the logical bytenr of the real extent.
In that case, the SHARED flag doesn't make much sense anyway, can we do a cleanup for those flags? Since fiemap/lseek no longer relies on extent map anymore.
I don't get it. What SHARED flag are talking about? And which "flags", where? We have nothing specific for lseek/fiemap in the extent maps, so I don't understand.
Quoted 611 lines
Quoted 598 lines
This reproducer triggers that bug:
$ cat fiemap-bug.sh #!/bin/bashDEV=/dev/sdj
MNT=/mnt/sdjmkfs.btrfs -f $DEV
mount $DEV $MNT# Create a file with two 256K extents.
# Since there is no other write activity, they will be contiguous, # and their extent maps merged, despite having two distinct extents.xfs_io-f -c "pwrite -S0xab0 256K" \
-c "fsync" \
-c "pwrite -S0xcd256K 256K" \
-c "fsync" \
$MNT/foo# Now clone only the second extent into another file.
xfs_io-f -c "reflink $MNT/foo 256K 0 256K" $MNT/bar# Filefrag will report a single 512K extent, and say it's not shared. echo
filefrag -v $MNT/fooumount $MNT
Running the reproducer:
$ ./fiemap-bug.sh wrote 262144/262144 bytes at offset 0 256 KiB, 64 ops; 0.0038 sec (65.479 MiB/sec and 16762.7030 ops/sec) wrote 262144/262144 bytes at offset 262144 256 KiB, 64 ops; 0.0040 sec (61.125 MiB/sec and 15647.9218 ops/sec) linked 262144/262144 bytes at offset 0 256 KiB, 1 ops; 0.0002 sec (1.034 GiB/sec and 4237.2881 ops/sec)Filesystem type is: 9123683e File size of /mnt/sdj/foo is 524288 (128 blocks of 4096 bytes) ext: logical_offset: physical_offset: length: expected: flags: 0: 0.. 127: 3328.. 3455: 128: last,eof /mnt/sdj/foo: 1 extent foundWe end up reporting that we have a single 512K that is not shared, however we have two 256K extents, and the second one is shared. Changing the reproducer to clone instead the first extent into file 'bar', makes us report a single 512K extent that is shared, which is algo incorrect since we have two 256K extents and only the first one is shared.
This patch is part of a larger patchset that is comprised of the following patches:
btrfs: allow hole and data seeking to be interruptible
btrfs: make hole and data seeking a lot more efficient
btrfs: remove check for impossible block start for an extent map at fiemap
btrfs: remove zero length check when entering fiemap
btrfs: properly flush delalloc when entering fiemap
btrfs: allow fiemap to be interruptible
btrfs: renamebtrfs_check_shared()to a more descriptive name
btrfs: speedup checking for extent sharedness during fiemap
btrfs: skip unnecessary extent buffer sharedness checks during fiemap
btrfs: make fiemap more efficient and accurate reporting extent sharednessThe patchset was tested on a machine running a non-debug kernel (Debian's default config) and compared the tests below on a branch without the patchset versus the same branch with the whole patchset applied.
The following test for a large compressed file without holes:
$ cat fiemap-perf-test.sh #!/bin/bashDEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f $DEV mount -o compress=lzo $DEV $MNT # 40G gives 327680 128K file extents (due to compression). xfs_io -f -c "pwrite -S 0xab -b 1M 0 20G" $MNT/foobar umount $MNT mount -o compress=lzo $DEV $MNT start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata not cached)" start=$(date +%s%N) filefrag $MNT/foobar end=$(date +%s%N) dur=$(( (end - start) / 1000000 )) echo "fiemap took $dur milliseconds (metadata cached)" umount $MNTBefore patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 3597 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 2107 milliseconds (metadata cached)After patchset:
$ ./fiemap-perf-test.sh (...) /mnt/sdi/foobar: 327680 extents found fiemap took 1214 milliseconds (metadata not cached) /mnt/sdi/foobar: 327680 extents found fiemap took 684 milliseconds (metadata cached)That's a speedup of about 3x for both cases (no metadata cached and all metadata cached).
The test provided by Pavel (first Link tag at the bottom), which uses files with a large number of holes, was also used to measure the gains, and it consists on a small C program and a shell script to invoke it. The C program is the following:
$ cat pavels-test.c #include <stdio.h> #include <unistd.h> #include <stdlib.h> #include <fcntl.h>#include <sys/stat.h> #include <sys/time.h> #include <sys/ioctl.h> #include <linux/fs.h> #include <linux/fiemap.h> #define FILE_INTERVAL (1<<13) /* 8Kb */ long long interval(struct timeval t1, struct timeval t2) { long long val = 0; val += (t2.tv_usec - t1.tv_usec); val += (t2.tv_sec - t1.tv_sec) * 1000 * 1000; return val; } int main(int argc, char **argv) { struct fiemap fiemap = {}; struct timeval t1, t2; char data = 'a'; struct stat st; int fd, off, file_size = FILE_INTERVAL; if (argc != 3 && argc != 2) { printf("usage: %s <path> [size]\n", argv[0]); return 1; } if (argc == 3) file_size = atoi(argv[2]); if (file_size < FILE_INTERVAL) file_size = FILE_INTERVAL; file_size -= file_size % FILE_INTERVAL; fd = open(argv[1], O_RDWR | O_CREAT | O_TRUNC, 0644); if (fd < 0) { perror("open"); return 1; } for (off = 0; off < file_size; off += FILE_INTERVAL) { if (pwrite(fd, &data, 1, off) != 1) { perror("pwrite"); close(fd); return 1; } } if (ftruncate(fd, file_size)) { perror("ftruncate"); close(fd); return 1; } if (fstat(fd, &st) < 0) { perror("fstat"); close(fd); return 1; } printf("size: %ld\n", st.st_size); printf("actual size: %ld\n", st.st_blocks * 512); fiemap.fm_length = FIEMAP_MAX_OFFSET; gettimeofday(&t1, NULL); if (ioctl(fd, FS_IOC_FIEMAP, &fiemap) < 0) { perror("fiemap"); close(fd); return 1; } gettimeofday(&t2, NULL); printf("fiemap: fm_mapped_extents = %d\n", fiemap.fm_mapped_extents); printf("time = %lld us\n", interval(t1, t2)); close(fd); return 0; } $ gcc -o pavels_test pavels_test.cAnd the wrapper shell script:
$ cat fiemap-pavels-test.sh#!/bin/bash DEV=/dev/sdi MNT=/mnt/sdi mkfs.btrfs -f -O no-holes $DEV mount $DEV $MNT echo echo "*********** 256M ***********" echo ./pavels-test $MNT/testfile $((1 << 28)) echo ./pavels-test $MNT/testfile $((1 << 28)) echo echo "*********** 512M ***********" echo ./pavels-test $MNT/testfile $((1 << 29)) echo ./pavels-test $MNT/testfile $((1 << 29)) echo echo "*********** 1G ***********" echo ./pavels-test $MNT/testfile $((1 << 30)) echo ./pavels-test $MNT/testfile $((1 << 30)) umount $MNTRunning his reproducer before applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4003133 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 4895330 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 30123675 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 33450934 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 224924074 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 217239242 usRunning it after applying the patchset:
*********** 256M *********** size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29475 us size: 268435456 actual size: 134217728 fiemap: fm_mapped_extents = 32768 time = 29307 us *********** 512M *********** size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 58996 us size: 536870912 actual size: 268435456 fiemap: fm_mapped_extents = 65536 time = 59115 us *********** 1G *********** size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 116251 time = 124141 us size: 1073741824 actual size: 536870912 fiemap: fm_mapped_extents = 131072 time = 119387 usThe speedup is massive, both on the first fiemap call and on the second one as well, as his test creates files with many holes and small extents (every extent follows a hole and precedes another hole).
For the 256M file we go from 4 seconds down to 29 milliseconds in the first run, and then from 4.9 seconds down to 29 milliseconds again in the second run, a speedup of 138x and 169x, respectively.
For the 512M file we go from 30.1 seconds down to 59 milliseconds in the first run, and then from 33.5 seconds down to 59 milliseconds again in the second run, a speedup of 510x and 568x, respectively.
For the 1G file, we go from 225 seconds down to 124 milliseconds in the first run, and then from 217 seconds down to 119 milliseconds in the second run, a speedup of 1815x and 1824x, respectively.
Reported-by Pavel Tikhomirov <[email protected]>Reported-by Dominique MARTINET <[email protected]>Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/ctree.h | 4 +- fs/btrfs/extent_io.c | 714 +++++++++++++++++++++++++++++-------------- fs/btrfs/file.c | 16 +- fs/btrfs/inode.c | 140 +-------- 4 files changed, 506 insertions(+), 368 deletions(-)fs/btrfs/ctree.h +2
-2@@ -3402,8 +3402,6 @@ unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio, u64 start, u64 end);int btrfs_check_data_csum (struct inode *inode,struct btrfs_bio *bbio,u32 bio_offset,struct page *page,u32 pgoff); -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len); noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,u64 *orig_start,u64 *orig_block_len,u64 *ram_bytes,bool strict);@@ -3583,6 +3581,8 @@ int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end);int btrfs_check_nocow_lock (struct btrfs_inode *inode,loff_t pos,size_t *write_bytes);void btrfs_check_nocow_unlock (struct btrfs_inode *inode); +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret);/* tree-defrag.c */ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,fs/btrfs/extent_io.c +92
-79@@ -5353,42 +5353,6 @@ int try_release_extent_mapping(struct page *page, gfp_t mask)return try_release_extent_state (tree, page, mask); } -/* -* helper function for fiemap, which doesn't want to see any holes. -* This maps until we find something past 'last' -*/ -static struct extent_map *get_extent_skip_holes (struct btrfs_inode *inode, -u64 offset,u64 last) -{ -u64 sectorsize =btrfs_inode_sectorsize (inode); -struct extent_map *em; -u64 len; - -if (offset >= last) -return NULL ; - -while (1 ) { - len = last - offset; -if (len ==0 ) -break ; - len =ALIGN (len, sectorsize); - em =btrfs_get_extent_fiemap (inode, offset, len); -if (IS_ERR (em)) -return em; - -/* if this isn't a hole return it */ -if (em->block_start !=EXTENT_MAP_HOLE ) -return em; - -/* this is a hole, advance to the next extent */ - offset =extent_map_end (em); -free_extent_map (em); -if (offset >= last) -break ; - } -return NULL ; -} -/* * To cache previous fiemap extent * @@ -5418,6 +5382,9 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, {int ret =0 ; +/* Set at the end of extent_fiemap(). */ +ASSERT ((flags &FIEMAP_EXTENT_LAST ) ==0 ); +if (!cache->cached) goto assign;@@ -5446,11 +5413,10 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,*/ if (cache->offset + cache->len == offset && cache->phys + cache->len == phys && - (cache->flags & ~FIEMAP_EXTENT_LAST ) == - (flags & ~FIEMAP_EXTENT_LAST )) { + cache->flags == flags) { cache->len += len; cache->flags |= flags; - goto try_submit_last; +return 0 ; }/* Not mergeable, need to submit cached one */ @@ -5465,13 +5431,8 @@ static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, cache->phys = phys; cache->len = len; cache->flags = flags; -try_submit_last: -if (cache->flags &FIEMAP_EXTENT_LAST ) { - ret =fiemap_fill_next_extent (fieinfo, cache->offset, - cache->phys, cache->len, cache->flags); - cache->cached = false; - } -return ret; + +return 0 ; }/* @@ -5501,229 +5462,534 @@ static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,return ret; } -int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, -u64 start,u64 len) +static int fiemap_next_leaf_item (struct btrfs_inode *inode, +struct btrfs_path *path) { -int ret =0 ; -u64 off; -u64 max = start + len; -u32 flags =0 ; -u32 found_type; -u64 last; -u64 last_for_get_extent =0 ; -u64 disko =0 ; -u64 isize =i_size_read (&inode->vfs_inode); -struct btrfs_key found_key; -struct extent_map *em =NULL ; -struct extent_state *cached_state =NULL ; -struct btrfs_path *path; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; + + path->slots[0 ]++; +if (path->slots[0 ] <btrfs_header_nritems (path->nodes[0 ])) +return 0 ; + + ret =btrfs_next_leaf (inode->root, path); +if (ret !=0 ) +return ret; + +/* +* Don't bother with cloning if there are no more file extent items for +* our inode. +*/ +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid !=btrfs_ino (inode) || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; + +/* See the comment at fiemap_search_slot() about why we clone. */ + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot; + +return 0 ; +} + +/* +* Search for the first file extent item that starts at a given file offset or +* the one that starts immediately before that offset. +* Returns: 0 on success, < 0 on error, 1 if not found. +*/ +static int fiemap_search_slot (struct btrfs_inode *inode,struct btrfs_path *path, +u64 file_offset) +{ +const u64 ino =btrfs_ino (inode);struct btrfs_root *root = inode->root; -struct fiemap_cache cache = {0 }; -struct btrfs_backref_shared_cache *backref_cache; -struct ulist *roots; -struct ulist *tmp_ulist; -int end =0 ; -u64 em_start =0 ; -u64 em_len =0 ; -u64 em_end =0 ; +struct extent_buffer *clone; +struct btrfs_key key; +int slot; +int ret; - backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); - path =btrfs_alloc_path (); - roots =ulist_alloc (GFP_KERNEL ); - tmp_ulist =ulist_alloc (GFP_KERNEL ); -if (!backref_cache || !path || !roots || !tmp_ulist) { - ret = -ENOMEM ; - goto out_free_ulist; + key.objectid = ino; + key.type =BTRFS_EXTENT_DATA_KEY ; + key.offset = file_offset; + + ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); +if (ret <0 ) +return ret; + +if (ret >0 && path->slots[0 ] >0 ) { +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ] -1 ); +if (key.objectid == ino && key.type ==BTRFS_EXTENT_DATA_KEY ) + path->slots[0 ]--; + } + +if (path->slots[0 ] >=btrfs_header_nritems (path->nodes[0 ])) { + ret =btrfs_next_leaf (root, path); +if (ret !=0 ) +return ret; + +btrfs_item_key_to_cpu (path->nodes[0 ], &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +return 1 ; }/* -* We can't initialize that to 'start' as this could miss extents due -* to extent item merging +* We clone the leaf and use it during fiemap. This is because while +* using the leaf we do expensive things like checking if an extent is +* shared, which can take a long time. In order to prevent blocking +* other tasks for too long, we use a clone of the leaf. We have locked +* the file range in the inode's io tree, so we know none of our file +* extent items can change. This way we avoid blocking other tasks that +* want to insert items for other inodes in the same leaf or b+tree +* rebalance operations (triggered for example when someone is trying +* to push items into this leaf when trying to insert an item in a +* neighbour leaf). +* We also need the private clone because holding a read lock on an +* extent buffer of the subvolume's b+tree will make lockdep unhappy +* when we call fiemap_fill_next_extent(), because that may cause a page +* fault when filling the user space buffer with fiemap data. */ - off =0 ;fs/btrfs/extent_io.c static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, · line 5564
Quoted 10 lines
fs/btrfs/extent_io.c +8
-2- start =round_down (start,btrfs_inode_sectorsize (inode)); - len =round_up (max,btrfs_inode_sectorsize (inode)) - start; + clone =btrfs_clone_extent_buffer (path->nodes[0 ]); +if (!clone) +return -ENOMEM ; + + slot = path->slots[0 ]; +btrfs_release_path (path); + path->nodes[0 ] = clone; + path->slots[0 ] = slot;Although this is correct, it still looks a little tricky.
Quoted 4 lines
We rely on
btrfs_release_path()to release all tree blocks in the subvolume tree, including unlocking the tree blocks, thus path->locks[0] is also 0, meaning next time we callbtrfs_release_path()we won't try to unlock the cloned eb.
We're not taking any lock on the cloned extent buffer. It's not needed, it's private
to the task.
Quoted 4 lines
But I'd say it's still pretty tricky, and unfortunately I don't have any better alternative.
Quoted 13 lines
fs/btrfs/extent_io.c static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, · line 5574
Quoted 10 lines
fs/btrfs/extent_io.c +10
-0+ +return 0 ; +} + +/* +* Process a range which is a hole or a prealloc extent in the inode's subvolume +* btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc +* extent. The end offset (@end) is inclusive. +*/ +static int fiemap_process_hole(struct btrfs_inode *inode,Does the name still make sense as we're handling both hole and prealloc range?
I chose that name because hole and prealloc are treated the same way. Sure, I could name it fiemap_process_hole_or_prealloc() or something like that, but
I decided to keep the name shorter, make them explicit in the comments and code.
The old code did the same, get_extent_skip_holes() skipped holes and prealloc extents without delalloc.
Quoted 6 lines
And I always find the delalloc search a big pain during lseek/fiemap.
I guess except using certain flags, there is some hard requirement for delalloc range reporting?
Yes. Delalloc is not meant to be flushed for fiemap unless FIEMAP_FLAG_SYNC is given by the user.
For lseek it's just not needed, but that was already mentioned / discussed in patch 2/10.
Quoted 157 lines
Quoted 157 lines
fs/btrfs/extent_io.c +123
-27+struct fiemap_extent_info *fieinfo, + struct fiemap_cache *cache, + struct btrfs_backref_shared_cache *backref_cache, + u64 disk_bytenr, u64 extent_offset, + u64 extent_gen, + struct ulist *roots, struct ulist *tmp_ulist, + u64 start, u64 end) +{ + const u64 i_size = i_size_read(&inode->vfs_inode); +const u64 ino =btrfs_ino (inode); +u64 cur_offset = start; +u64 last_delalloc_end =0 ; +u32 prealloc_flags =FIEMAP_EXTENT_UNWRITTEN ; +bool checked_extent_shared = false; +int ret;/* -* lookup the last file extent. We're not using i_size here -* because there might be preallocation past i_size +* There can be no delalloc past i_size, so don't waste time looking for +* it beyond i_size. */ - ret =btrfs_lookup_file_extent (NULL , root, path,btrfs_ino (inode),-1 , -0 ); -if (ret <0 ) { - goto out_free_ulist; - }else { -WARN_ON (!ret); -if (ret ==1 ) - ret =0 ; - } +while (cur_offset < end && cur_offset < i_size) { +u64 delalloc_start; +u64 delalloc_end; +u64 prealloc_start; +u64 prealloc_len =0 ; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, cur_offset, end, + &delalloc_start, + &delalloc_end); +if (!delalloc) +break ; - path->slots[0 ]--; -btrfs_item_key_to_cpu (path->nodes[0 ], &found_key, path->slots[0 ]); - found_type = found_key.type; - -/* No extents, but there might be delalloc bits */ -if (found_key.objectid !=btrfs_ino (inode) || - found_type !=BTRFS_EXTENT_DATA_KEY ) { -/* have to trust i_size as the end */ - last = (u64)-1 ; - last_for_get_extent = isize; - }else {/* -* remember the start of the last extent. There are a -* bunch of different factors that go into the length of the -* extent, so its much less complex to remember where it started +* If this is a prealloc extent we have to report every section +* of it that has no delalloc. */ - last = found_key.offset; - last_for_get_extent = last +1 ; +if (disk_bytenr !=0 ) { +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = delalloc_start - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = delalloc_start - prealloc_start; + } + } + +if (prealloc_len >0 ) { +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + + checked_extent_shared = true; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + extent_offset += prealloc_len; + } + + ret =emit_fiemap_extent (fieinfo, cache, delalloc_start,0 , + delalloc_end +1 - delalloc_start, +FIEMAP_EXTENT_DELALLOC | +FIEMAP_EXTENT_UNKNOWN ); +if (ret) +return ret; + + last_delalloc_end = delalloc_end; + cur_offset = delalloc_end +1 ; + extent_offset += cur_offset - delalloc_start; +cond_resched (); + } + +/* +* Either we found no delalloc for the whole prealloc extent or we have +* a prealloc extent that spans i_size or starts at or after i_size. +*/ +if (disk_bytenr !=0 && last_delalloc_end < end) { +u64 prealloc_start; +u64 prealloc_len; + +if (last_delalloc_end ==0 ) { + prealloc_start = start; + prealloc_len = end +1 - start; + }else { + prealloc_start = last_delalloc_end +1 ; + prealloc_len = end +1 - prealloc_start; + } + +if (!checked_extent_shared && fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (inode->root, + ino, disk_bytenr, + extent_gen, roots, + tmp_ulist, + backref_cache); +if (ret <0 ) +return ret; +else if (ret >0 ) + prealloc_flags |=FIEMAP_EXTENT_SHARED ; + } + ret =emit_fiemap_extent (fieinfo, cache, prealloc_start, + disk_bytenr + extent_offset, + prealloc_len, prealloc_flags); +if (ret) +return ret; + } + +return 0 ; +} + +static int fiemap_find_last_extent_offset (struct btrfs_inode *inode, +struct btrfs_path *path, +u64 *last_extent_end_ret) +{ +const u64 ino =btrfs_ino (inode); +struct btrfs_root *root = inode->root; +struct extent_buffer *leaf; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 disk_bytenr;
Quoted 8 lines
fs/btrfs/extent_io.c static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, · line 5709
Quoted 6 lines
fs/btrfs/extent_io.c +6
-0+int ret; + +/* +* Lookup the last file extent. We're not using i_size here because +* there might be preallocation past i_size. +*/ I'm wondering how could this happen?
You can fallocate an extent at or after i_size.
Quoted 3 lines
Normally if we're truncating an inode, the extents starting after
round_up(i_size, sectorsize)should be dropped.
It has nothing to do with truncate, just fallocate.
Quoted 8 lines
Or if we later enlarge the inode, we may hit old extents and read out some stale data other than expected zeros.
Thus searching using
round_up(i_size, sectorsize)should still let us to reach the slot after the last file extent.Or did I miss something?
Yes, it's about prealloc extents at or after i_size.
Thanks.
Quoted 607 lines
Thanks,
QuQuoted 603 lines
fs/btrfs/extent_io.c +251
-116+ ret =btrfs_lookup_file_extent (NULL , root, path, ino, (u64)-1 ,0 ); +/* There can't be a file extent item at offset (u64)-1 */ +ASSERT (ret !=0 ); +if (ret <0 ) +return ret; + +/* +* For a non-existing key, btrfs_search_slot() always leaves us at a +* slot > 0, except if the btree is empty, which is impossible because +* at least it has the inode item for this inode and all the items for +* the root inode 256. +*/ +ASSERT (path->slots[0 ] >0 ); + path->slots[0 ]--; + leaf = path->nodes[0 ]; +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) { +/* No file extent items in the subvolume tree. */ + *last_extent_end_ret =0 ; +return 0 ; } -btrfs_release_path (path);/* -* we might have some extents allocated but more delalloc past those -* extents. so, we trust isize unless the start of the last extent is -* beyond isize +* For an inline extent, the disk_bytenr is where inline data starts at, +* so first check if we have an inline extent item before checking if we +* have an implicit hole (disk_bytenr == 0). */ -if (last < isize) { - last = (u64)-1 ; - last_for_get_extent = isize; + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); +if (btrfs_file_extent_type (leaf, ei) ==BTRFS_FILE_EXTENT_INLINE ) { + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; } -lock_extent_bits (&inode->io_tree, start, start + len -1 , - &cached_state); +/* +* Find the last file extent item that is not a hole (when NO_HOLES is +* not enabled). This should take at most 2 iterations in the worst +* case: we have one hole file extent item at slot 0 of a leaf and +* another hole file extent item as the last item in the previous leaf. +* This is because we merge file extent items that represent holes. +*/ + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +while (disk_bytenr ==0 ) { + ret =btrfs_previous_item (root, path, ino,BTRFS_EXTENT_DATA_KEY ); +if (ret <0 ) { +return ret; + }else if (ret >0 ) { +/* No file extent items that are not holes. */ + *last_extent_end_ret =0 ; +return 0 ; + } + leaf = path->nodes[0 ]; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); + } - em =get_extent_skip_holes (inode, start, last_for_get_extent); -if (!em) - goto out; -if (IS_ERR (em)) { - ret =PTR_ERR (em); + *last_extent_end_ret =btrfs_file_extent_end (path); +return 0 ; +} + +int extent_fiemap (struct btrfs_inode *inode,struct fiemap_extent_info *fieinfo, +u64 start,u64 len) +{ +const u64 ino =btrfs_ino (inode); +struct extent_state *cached_state =NULL ; +struct btrfs_path *path; +struct btrfs_root *root = inode->root; +struct fiemap_cache cache = {0 }; +struct btrfs_backref_shared_cache *backref_cache; +struct ulist *roots; +struct ulist *tmp_ulist; +u64 last_extent_end; +u64 prev_extent_end; +u64 lockstart; +u64 lockend; +bool stopped = false; +int ret; + + backref_cache =kzalloc (sizeof (*backref_cache),GFP_KERNEL ); + path =btrfs_alloc_path (); + roots =ulist_alloc (GFP_KERNEL ); + tmp_ulist =ulist_alloc (GFP_KERNEL ); +if (!backref_cache || !path || !roots || !tmp_ulist) { + ret = -ENOMEM ; goto out; } -while (!end) { -u64 offset_in_extent =0 ; + lockstart =round_down (start,btrfs_inode_sectorsize (inode)); + lockend =round_up (start + len,btrfs_inode_sectorsize (inode)); + prev_extent_end = lockstart; -/* break if the extent we found is outside the range */ -if (em->start >= max ||extent_map_end (em) < off) -break ; +lock_extent_bits (&inode->io_tree, lockstart, lockend, &cached_state); -/* -* get_extent may return an extent that starts before our -* requested range. We have to make sure the ranges -* we return to fiemap always move forward and don't -* overlap, so adjust the offsets here -*/ - em_start =max (em->start, off); + ret =fiemap_find_last_extent_offset (inode, path, &last_extent_end); +if (ret <0 ) + goto out_unlock; +btrfs_release_path (path); + path->reada =READA_FORWARD ; + ret =fiemap_search_slot (inode, path, lockstart); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) {/* -* record the offset from the start of the extent -* for adjusting the disk offset below. Only do this if the -* extent isn't compressed since our in ram offset may be past -* what we have actually allocated on disk. +* No file extent item found, but we may have delalloc between +* the current offset and i_size. So check for that. */ -if (!test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - offset_in_extent = em_start - em->start; - em_end =extent_map_end (em); - em_len = em_end - em_start; - flags =0 ; -if (em->block_start <EXTENT_MAP_LAST_BYTE ) - disko = em->block_start + offset_in_extent; -else - disko =0 ; + ret =0 ; + goto check_eof_delalloc; + } + +while (prev_extent_end < lockend) { +struct extent_buffer *leaf = path->nodes[0 ]; +struct btrfs_file_extent_item *ei; +struct btrfs_key key; +u64 extent_end; +u64 extent_len; +u64 extent_offset =0 ; +u64 extent_gen; +u64 disk_bytenr =0 ; +u64 flags =0 ; +int extent_type; +u8 compression; + +btrfs_item_key_to_cpu (leaf, &key, path->slots[0 ]); +if (key.objectid != ino || key.type !=BTRFS_EXTENT_DATA_KEY ) +break ; + + extent_end =btrfs_file_extent_end (path);/* -* bump off for our next call to get_extent +* The first iteration can leave us at an extent item that ends +* before our range's start. Move to the next item. */ - off =extent_map_end (em); -if (off >= max) - end =1 ; - -if (em->block_start ==EXTENT_MAP_INLINE ) { - flags |= (FIEMAP_EXTENT_DATA_INLINE | -FIEMAP_EXTENT_NOT_ALIGNED ); - }else if (em->block_start ==EXTENT_MAP_DELALLOC ) { - flags |= (FIEMAP_EXTENT_DELALLOC | -FIEMAP_EXTENT_UNKNOWN ); - }else if (fieinfo->fi_extents_max) { -u64 extent_gen; -u64 bytenr = em->block_start - - (em->start - em->orig_start); +if (extent_end <= lockstart) + goto next_item; -/* -* If two extent maps are merged, then their generation -* is set to the maximum between their generations. -* Otherwise its generation matches the one we have in -* corresponding file extent item. If we have a merged -* extent map, don't use its generation to speedup the -* sharedness check below. -*/ -if (test_bit (EXTENT_FLAG_MERGED , &em->flags)) - extent_gen =0 ; -else - extent_gen = em->generation; +/* We have in implicit hole (NO_HOLES feature enabled). */ +if (prev_extent_end < key.offset) { +const u64 range_end =min (key.offset, lockend) -1 ; -/* -* As btrfs supports shared space, this information -* can be exported to userspace tools via -* flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 -* then we're just getting a count and we can skip the -* lookup stuff. -*/ - ret =btrfs_is_data_extent_shared (root,btrfs_ino (inode), - bytenr, extent_gen, - roots, tmp_ulist, - backref_cache); -if (ret <0 ) - goto out_free; -if (ret) - flags |=FIEMAP_EXTENT_SHARED ; - ret =0 ; - } -if (test_bit (EXTENT_FLAG_COMPRESSED , &em->flags)) - flags |=FIEMAP_EXTENT_ENCODED ; -if (test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) - flags |=FIEMAP_EXTENT_UNWRITTEN ; + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + prev_extent_end, range_end); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; + } -free_extent_map (em); - em =NULL ; -if ((em_start >= last) || em_len == (u64)-1 || - (last == (u64)-1 && isize <= em_end)) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; +/* We've reached the end of the fiemap range, stop. */ +if (key.offset >= lockend) { + stopped = true; +break ; + } } -/* now scan forward to see if this is really the last extent. */ - em =get_extent_skip_holes (inode, off, last_for_get_extent); -if (IS_ERR (em)) { - ret =PTR_ERR (em); - goto out; + extent_len = extent_end - key.offset; + ei =btrfs_item_ptr (leaf, path->slots[0 ], + struct btrfs_file_extent_item); + compression =btrfs_file_extent_compression (leaf, ei); + extent_type =btrfs_file_extent_type (leaf, ei); + extent_gen =btrfs_file_extent_generation (leaf, ei); + +if (extent_type !=BTRFS_FILE_EXTENT_INLINE ) { + disk_bytenr =btrfs_file_extent_disk_bytenr (leaf, ei); +if (compression ==BTRFS_COMPRESS_NONE ) + extent_offset =btrfs_file_extent_offset (leaf, ei); } -if (!em) { - flags |=FIEMAP_EXTENT_LAST ; - end =1 ; + +if (compression !=BTRFS_COMPRESS_NONE ) + flags |=FIEMAP_EXTENT_ENCODED ; + +if (extent_type ==BTRFS_FILE_EXTENT_INLINE ) { + flags |=FIEMAP_EXTENT_DATA_INLINE ; + flags |=FIEMAP_EXTENT_NOT_ALIGNED ; + ret =emit_fiemap_extent (fieinfo, &cache, key.offset,0 , + extent_len, flags); + }else if (extent_type ==BTRFS_FILE_EXTENT_PREALLOC ) { + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache, + disk_bytenr, extent_offset, + extent_gen, roots, tmp_ulist, + key.offset, extent_end -1 ); + }else if (disk_bytenr ==0 ) { +/* We have an explicit hole. */ + ret =fiemap_process_hole (inode, fieinfo, &cache, + backref_cache,0 ,0 ,0 , + roots, tmp_ulist, + key.offset, extent_end -1 ); + }else { +/* We have a regular extent. */ +if (fieinfo->fi_extents_max) { + ret =btrfs_is_data_extent_shared (root, ino, + disk_bytenr, + extent_gen, + roots, + tmp_ulist, + backref_cache); +if (ret <0 ) + goto out_unlock; +else if (ret >0 ) + flags |=FIEMAP_EXTENT_SHARED ; + } + + ret =emit_fiemap_extent (fieinfo, &cache, key.offset, + disk_bytenr + extent_offset, + extent_len, flags); } - ret =emit_fiemap_extent (fieinfo, &cache, em_start, disko, - em_len, flags); -if (ret) { -if (ret ==1 ) - ret =0 ; - goto out_free; + +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* fiemap_fill_next_extent() told us to stop. */ + stopped = true; +break ; } + prev_extent_end = extent_end; +next_item:if (fatal_signal_pending (current)) { ret = -EINTR ; - goto out_free; + goto out_unlock; } + + ret =fiemap_next_leaf_item (inode, path); +if (ret <0 ) { + goto out_unlock; + }else if (ret >0 ) { +/* No more file extent items for this inode. */ +break ; + } +cond_resched (); } -out_free: -if (!ret) - ret =emit_last_fiemap_cache (fieinfo, &cache); -free_extent_map (em); -out: -unlock_extent_cached (&inode->io_tree, start, start + len -1 , - &cached_state); -out_free_ulist: +check_eof_delalloc: +/* +* Release (and free) the path before emitting any final entries to +* fiemap_fill_next_extent() to keep lockdep happy. This is because +* once we find no more file extent items exist, we may have a +* non-cloned leaf, and fiemap_fill_next_extent() can trigger page +* faults when copying data to the user space buffer. +*/ +btrfs_free_path (path); + path =NULL ; + +if (!stopped && prev_extent_end < lockend) { + ret =fiemap_process_hole (inode, fieinfo, &cache, backref_cache, +0 ,0 ,0 , roots, tmp_ulist, + prev_extent_end, lockend -1 ); +if (ret <0 ) + goto out_unlock; + prev_extent_end = lockend; + } + +if (cache.cached && cache.offset + cache.len >= last_extent_end) { +const u64 i_size =i_size_read (&inode->vfs_inode); + +if (prev_extent_end < i_size) { +u64 delalloc_start; +u64 delalloc_end; +bool delalloc; + + delalloc =btrfs_find_delalloc_in_range (inode, + prev_extent_end, + i_size -1 , + &delalloc_start, + &delalloc_end); +if (!delalloc) + cache.flags |=FIEMAP_EXTENT_LAST ; + }else { + cache.flags |=FIEMAP_EXTENT_LAST ; + } + } + + ret =emit_last_fiemap_cache (fieinfo, &cache); + +out_unlock: +unlock_extent_cached (&inode->io_tree, lockstart, lockend, &cached_state); +out:kfree (backref_cache);btrfs_free_path (path);ulist_free (roots);fs/btrfs/file.c +8
-8@@ -3602,10 +3602,10 @@ static long btrfs_fallocate(struct file *file, int mode, }/* -* Helper for have_delalloc_in_range(). Find a subrange in a given range that -* has unflushed and/or flushing delalloc. There might be other adjacent -* subranges after the one it found, so have_delalloc_in_range() keeps looping -* while it gets adjacent subranges, and merging them together. +* Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range +* that has unflushed and/or flushing delalloc. There might be other adjacent +* subranges after the one it found, so btrfs_find_delalloc_in_range() keeps +* looping while it gets adjacent subranges, and merging them together. */ static bool find_delalloc_subrange (struct btrfs_inode *inode,u64 start,u64 end,u64 *delalloc_start_ret,u64 *delalloc_end_ret)@@ -3740,8 +3740,8 @@ static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end* if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and * end offsets of the subrange. */ -static bool have_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, -u64 *delalloc_start_ret,u64 *delalloc_end_ret) +bool btrfs_find_delalloc_in_range (struct btrfs_inode *inode,u64 start,u64 end, +u64 *delalloc_start_ret,u64 *delalloc_end_ret) {u64 cur_offset =round_down (start, inode->root->fs_info->sectorsize);u64 prev_delalloc_end =0 ;@@ -3804,8 +3804,8 @@ static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,u64 delalloc_end;bool delalloc; - delalloc =have_delalloc_in_range (inode, start, end, &delalloc_start, - &delalloc_end); + delalloc =btrfs_find_delalloc_in_range (inode, start, end, + &delalloc_start, &delalloc_end);if (delalloc && whence ==SEEK_DATA ) { *start_ret = delalloc_start;return true;fs/btrfs/inode.c +6
-134@@ -7064,133 +7064,6 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,return em; } -struct extent_map *btrfs_get_extent_fiemap (struct btrfs_inode *inode, -u64 start,u64 len) -{ -struct extent_map *em; -struct extent_map *hole_em =NULL ; -u64 delalloc_start = start; -u64 end; -u64 delalloc_len; -u64 delalloc_end; -int err =0 ; - - em =btrfs_get_extent (inode,NULL ,0 , start, len); -if (IS_ERR (em)) -return em; -/* -* If our em maps to: -* - a hole or -* - a pre-alloc extent, -* there might actually be delalloc bytes behind it. -*/ -if (em->block_start !=EXTENT_MAP_HOLE && - !test_bit (EXTENT_FLAG_PREALLOC , &em->flags)) -return em; -else - hole_em = em; - -/* check to see if we've wrapped (len == -1 or similar) */ - end = start + len; -if (end < start) - end = (u64)-1 ; -else - end -=1 ; - - em =NULL ; - -/* ok, we didn't find anything, lets look for delalloc */ - delalloc_len =count_range_bits (&inode->io_tree, &delalloc_start, - end, len,EXTENT_DELALLOC ,1 ); - delalloc_end = delalloc_start + delalloc_len; -if (delalloc_end < delalloc_start) - delalloc_end = (u64)-1 ; - -/* -* We didn't find anything useful, return the original results from -* get_extent() -*/ -if (delalloc_start > end || delalloc_end <= start) { - em = hole_em; - hole_em =NULL ; - goto out; - } - -/* -* Adjust the delalloc_start to make sure it doesn't go backwards from -* the start they passed in -*/ - delalloc_start =max (start, delalloc_start); - delalloc_len = delalloc_end - delalloc_start; - -if (delalloc_len >0 ) { -u64 hole_start; -u64 hole_len; -const u64 hole_end =extent_map_end (hole_em); - - em =alloc_extent_map (); -if (!em) { - err = -ENOMEM ; - goto out; - } - -ASSERT (hole_em); -/* -* When btrfs_get_extent can't find anything it returns one -* huge hole -* -* Make sure what it found really fits our range, and adjust to -* make sure it is based on the start from the caller -*/ -if (hole_end <= start || hole_em->start > end) { -free_extent_map (hole_em); - hole_em =NULL ; - }else { - hole_start =max (hole_em->start, start); - hole_len = hole_end - hole_start; - } - -if (hole_em && delalloc_start > hole_start) { -/* -* Our hole starts before our delalloc, so we have to -* return just the parts of the hole that go until the -* delalloc starts -*/ - em->len =min (hole_len, delalloc_start - hole_start); - em->start = hole_start; - em->orig_start = hole_start; -/* -* Don't adjust block start at all, it is fixed at -* EXTENT_MAP_HOLE -*/ - em->block_start = hole_em->block_start; - em->block_len = hole_len; -if (test_bit (EXTENT_FLAG_PREALLOC , &hole_em->flags)) -set_bit (EXTENT_FLAG_PREALLOC , &em->flags); - }else { -/* -* Hole is out of passed range or it starts after -* delalloc range -*/ - em->start = delalloc_start; - em->len = delalloc_len; - em->orig_start = delalloc_start; - em->block_start =EXTENT_MAP_DELALLOC ; - em->block_len = delalloc_len; - } - }else { -return hole_em; - } -out: - -free_extent_map (hole_em); -if (err) { -free_extent_map (em); -return ERR_PTR (err); - } -return em; -} - static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, const u64 start, const u64 len,@@ -8265,15 +8138,14 @@ static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,* in the compression of data (in an async thread) and will return * before the compression is done and writeback is started. A second * filemap_fdatawrite_range() is needed to wait for the compression to -* complete and writeback to start. Without this, our user is very -* likely to get stale results, because the extents and extent maps for -* delalloc regions are only allocated when writeback starts. +* complete and writeback to start. We also need to wait for ordered +* extents to complete, because our fiemap implementation uses mainly +* file extent items to list the extents, searching for extent maps +* only for file ranges with holes or prealloc extents to figure out +* if we have delalloc in those ranges. */ if (fieinfo->fi_flags &FIEMAP_FLAG_SYNC ) { - ret =btrfs_fdatawrite_range (inode,0 ,LLONG_MAX ); -if (ret) -return ret; - ret =filemap_fdatawait_range (inode->i_mapping,0 ,LLONG_MAX ); + ret =btrfs_wait_ordered_range (inode,0 ,LLONG_MAX );if (ret)return ret; }