Btrfs: only adjust outstanding_extents when we do a short write
PATCH 1/3 Btrfs: only adjust outstanding_extents when we do a short write
We have this weird dance where we always inc outstanding_extents when we do a O_DIRECT write, even if we allocate the entire range. To get around this we also drop the metadata space if we successfully write. This is an unnecessary dance, we only need to jack up outstanding_extents if we don't satisfy the entire range request in get_blocks_direct, otherwise we are good using our original reservation. So drop the unconditional inc and the drop of the metadata space that we have for the unconditional inc. Thanks,
--- fs/btrfs/inode.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-)
fs/btrfs/inode.c +6 -5
@@ -7137,6 +7137,7 @@ static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,u64 start = iblock << inode->i_blkbits;u64 lockstart, lockend;u64 len = bh_result->b_size; +u64 orig_len = len;int unlock_bits =EXTENT_LOCKED ;int ret =0 ;
@@ -7272,9 +7273,11 @@ unlock:if (start + len >i_size_read (inode))i_size_write (inode, start + len); -spin_lock (&BTRFS_I (inode)->lock); -BTRFS_I (inode)->outstanding_extents++; -spin_unlock (&BTRFS_I (inode)->lock); +if (len < orig_len) { +spin_lock (&BTRFS_I (inode)->lock); +BTRFS_I (inode)->outstanding_extents++; +spin_unlock (&BTRFS_I (inode)->lock); + } ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockstart + len - 1, EXTENT_DELALLOC, NULL,
@@ -8056,8 +8059,6 @@ static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,else if (ret >=0 && (size_t )ret < count)btrfs_delalloc_release_space (inode, count - (size_t )ret); -else -btrfs_delalloc_release_metadata (inode,0 ); } out:if (wakeup)
Signature · 1 line
1.8.3.1
PATCH 2/3 Btrfs: don't set and clear delalloc for O_DIRECT writes
We do this to get the space accounting, but this is just needless churn on the io_tree, so just drop setting/clearing delalloc and just drop the reserved data space when we have a successfull allocation. Thanks,
--- fs/btrfs/inode.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)
fs/btrfs/inode.c +2 -6
@@ -7142,7 +7142,7 @@ static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,int ret =0 ;if (create) - unlock_bits |=EXTENT_DELALLOC |EXTENT_DIRTY ; + unlock_bits |=EXTENT_DIRTY ;else len =min_t (u64, len, root->sectorsize);
@@ -7278,11 +7278,7 @@ unlock:BTRFS_I (inode)->outstanding_extents++;spin_unlock (&BTRFS_I (inode)->lock); } - - ret =set_extent_bit (&BTRFS_I (inode)->io_tree, lockstart, - lockstart + len -1 ,EXTENT_DELALLOC ,NULL , - &cached_state,GFP_NOFS ); -BUG_ON (ret); +btrfs_free_reserved_data_space (inode, len); }/*
Signature · 1 line
1.8.3.1
Re: [PATCH 2/3] Btrfs: don't set and clear delalloc for O_DIRECT writes
Quoted 4 lines · Josef Bacik
We do this to get the space accounting, but this is just needless churn on the
io_tree, so just drop setting/clearing delalloc and just drop the reserved data space when we have a successfull allocation. Thanks,
Looks good.
Quoted 38 lines
Signed-off-by Josef Bacik <[email protected]>--- fs/btrfs/inode.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)fs/btrfs/inode.c +2
-6@@ -7142,7 +7142,7 @@ static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,int ret =0 ;if (create) - unlock_bits |=EXTENT_DELALLOC |EXTENT_DIRTY ; + unlock_bits |=EXTENT_DIRTY ;else len =min_t (u64, len, root->sectorsize);@@ -7278,11 +7278,7 @@ unlock:BTRFS_I (inode)->outstanding_extents++;spin_unlock (&BTRFS_I (inode)->lock); } - - ret =set_extent_bit (&BTRFS_I (inode)->io_tree, lockstart, - lockstart + len -1 ,EXTENT_DELALLOC ,NULL , - &cached_state,GFP_NOFS ); -BUG_ON (ret); +btrfs_free_reserved_data_space (inode, len); }/* Signature · 6 lines
1.8.3.1 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
PATCH 3/3 Btrfs: account for large extents with enospc
In reply to Josef BacikOn our gluster boxes we stream large tar balls of backups onto our fses. With 160gb of ram this means we get really large contiguous ranges of dirty data, but the way our ENOSPC stuff works is that as long as it's contiguous we only hold metadata reservation for one extent. The problem is we limit our extents to 128mb, so we'll end up with at least 800 extents so our enospc accounting is quite a bit lower than what we need. To keep track of this make sure we increase outstanding_extents for every multiple of the max extent size so we can be sure to have enough reserved metadata space. Thanks,
--- fs/btrfs/ctree.h | 2 ++ fs/btrfs/extent-tree.c | 16 +++++++++---- fs/btrfs/extent_io.c | 2 +- fs/btrfs/inode.c | 63 +++++++++++++++++++++++++++++++++++++++++++++++++- 4 files changed, 76 insertions(+), 7 deletions(-)
fs/btrfs/ctree.h +2 -0
@@ -198,6 +198,8 @@ static int btrfs_csum_sizes[] = { 4, 0 };#define BTRFS_DIRTY_METADATA_THRESH (32 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_SIZE (128 * 1024 * 1024) +/* * The key defines the order in the tree, and so it also defines (optimal) * block layout.
fs/btrfs/extent-tree.c +11 -5
@@ -4966,19 +4966,25 @@ void btrfs_subvolume_release_metadata(struct btrfs_root *root,/** * drop_outstanding_extent - drop an outstanding extent * @inode: the inode we're dropping the extent for +* @num_bytes: the number of bytes we're relaseing. * * This is called when we are freeing up an outstanding extent, either called * after an error or after an extent is written. This will return the number of * reserved extents that need to be freed. This must be called with * BTRFS_I(inode)->lock held. */ -static unsigned drop_outstanding_extent(struct inode *inode) +static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) { unsigned drop_inode_space = 0; unsigned dropped_extents = 0; + unsigned num_extents = 0; - BUG_ON(!BTRFS_I(inode)->outstanding_extents); - BTRFS_I(inode)->outstanding_extents--; + num_extents = (unsigned)div64_u64(num_bytes + + BTRFS_MAX_EXTENT_SIZE - 1, + BTRFS_MAX_EXTENT_SIZE); + ASSERT(num_extents); + ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); + BTRFS_I(inode)->outstanding_extents -= num_extents; if (BTRFS_I(inode)->outstanding_extents == 0 && test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
@@ -5149,7 +5155,7 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) out_fail:spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);/* * If the inodes csum_bytes is the same as the original * csum_bytes then we know we haven't raced with any free()ers
@@ -5228,7 +5234,7 @@ void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) num_bytes =ALIGN (num_bytes, root->sectorsize);spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);if (num_bytes) to_free =calc_csum_metadata_size (inode, num_bytes,0 );
fs/btrfs/extent_io.c +1 -1
@@ -3239,7 +3239,7 @@ static noinline_for_stack int writepage_delalloc(struct inode *inode, page, &delalloc_start, &delalloc_end, - 128 * 1024 * 1024); + BTRFS_MAX_EXTENT_SIZE);if (nr_delalloc ==0 ) { delalloc_start = delalloc_end +1 ;continue ;
fs/btrfs/inode.c +62 -1
95 lines not on this page · raw message
Signature · 1 line
1.8.3.1
Re: [PATCH 3/3] Btrfs: account for large extents with enospc
Quoted 47 lines · Josef Bacik
On our gluster boxes we stream large tar balls of backups onto our fses. With 160gb of ram this means we get really large contiguous ranges of dirty data, but the way our ENOSPC stuff works is that as long as it's contiguous we only hold metadata reservation for one extent. The problem is we limit our extents to 128mb, so we'll end up with at least 800 extents so our enospc accounting is quite a bit lower than what we need. To keep track of this make sure we increase
outstanding_extentsfor every multiple of the max extent size so we can be sure to have enough reserved metadata space. Thanks,Signed-off-by Josef Bacik <[email protected]>--- fs/btrfs/ctree.h | 2 ++ fs/btrfs/extent-tree.c | 16 +++++++++---- fs/btrfs/extent_io.c | 2 +- fs/btrfs/inode.c | 63 +++++++++++++++++++++++++++++++++++++++++++++++++- 4 files changed, 76 insertions(+), 7 deletions(-)fs/btrfs/ctree.h +2
-0@@ -198,6 +198,8 @@ static int btrfs_csum_sizes[] = { 4, 0 };#define BTRFS_DIRTY_METADATA_THRESH (32 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_SIZE (128 * 1024 * 1024) +/* * The key defines the order in the tree, and so it also defines (optimal) * block layout. fs/btrfs/extent-tree.c +2
-1@@ -4966,19 +4966,25 @@ void btrfs_subvolume_release_metadata(struct btrfs_root *root,/** * drop_outstanding_extent - drop an outstanding extent * @inode: the inode we're dropping the extent for +* @num_bytes: the number of bytes we're relaseing. * * This is called when we are freeing up an outstanding extent, either called * after an error or after an extent is written. This will return the number of * reserved extents that need to be freed. This must be called with * BTRFS_I(inode)->lock held. */ -static unsigned drop_outstanding_extent (struct inode *inode) +static unsigned drop_outstanding_extent (struct inode *inode,u64 num_bytes)
fs/btrfs/extent-tree.c void btrfs_subvolume_release_metadata(struct btrfs_root *root, · line 4984
Quoted 10 lines
fs/btrfs/extent-tree.c +4
-2{unsigned drop_inode_space =0 ;unsigned dropped_extents =0 ; +unsigned num_extents =0 ; -BUG_ON (!BTRFS_I (inode)->outstanding_extents); -BTRFS_I (inode)->outstanding_extents--; + num_extents = (unsigned )div64_u64 (num_bytes + +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE );
A fastpath is better, like btrfs_merge_extent_hook().
(num_extents < BTRFS_MAX_EXTENT_SIZE) ? num_extents = 1 : (div64_u64(...))
Quoted 52 lines
fs/btrfs/extent-tree.c +5
-2+ASSERT (num_extents); +ASSERT (BTRFS_I (inode)->outstanding_extents >= num_extents); +BTRFS_I (inode)->outstanding_extents -= num_extents;if (BTRFS_I (inode)->outstanding_extents ==0 && test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,@@ -5149,7 +5155,7 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) out_fail:spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);/* * If the inodes csum_bytes is the same as the original * csum_bytes then we know we haven't raced with any free()ers @@ -5228,7 +5234,7 @@ void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) num_bytes =ALIGN (num_bytes, root->sectorsize);spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);if (num_bytes) to_free =calc_csum_metadata_size (inode, num_bytes,0 );fs/btrfs/extent_io.c +1
-1@@ -3239,7 +3239,7 @@ static noinline_for_stack int writepage_delalloc(struct inode *inode, page, &delalloc_start, &delalloc_end, - 128 * 1024 * 1024); + BTRFS_MAX_EXTENT_SIZE);if (nr_delalloc ==0 ) { delalloc_start = delalloc_end +1 ;continue ;fs/btrfs/inode.c +3
-0@@ -1530,10 +1530,45 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,static void btrfs_split_extent_hook (struct inode *inode,struct extent_state *orig,u64 split) { +u64 size; +/* not delalloc, ignore it */ if (!(orig->state &EXTENT_DELALLOC ))return ; + size = orig->end - orig->start +1 ;
fs/btrfs/inode.c static int run_delalloc_range(struct inode *inode, struct page *locked_page, · line 1549
Quoted 10 lines
fs/btrfs/inode.c +10
-0+if (size >BTRFS_MAX_EXTENT_SIZE ) { +u64 num_extents; +u64 new_size; + +/* +* We need the largest size of the remaining extent to see if we +* need to add a new outstanding extent. Think of the following +* case +* +* [MEAX_EXTENT_SIZEx2 - 4k][4k]
s/MEAX/MAX
Quoted 9 lines
fs/btrfs/inode.c +9
-0+* +* The new_size would just be 4k and we'd think we had enough +* outstanding extents for this if we only took one side of the +* split, same goes for the other direction. We need to see if +* the larger size still is the same amount of extents as the +* original size, because if it is we need to add a new +* outstanding extent. But if we split up and the larger size +* is less than the original then we are good to go since we've +* already accounted for the extra extent in our original
fs/btrfs/inode.c static int run_delalloc_range(struct inode *inode, struct page *locked_page, · line 1568
Quoted 10 lines
fs/btrfs/inode.c +10
-0+* accounting. +*/ + new_size = orig->end - split +1 ; +if ((split - orig->start) > new_size) + new_size = split - orig->start; + + num_extents =div64_u64 (size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ); +if (div64_u64 (new_size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ) < num_extents)
BTRFS_MAX_EXTENT_SIZE (128mb) means 1 << BTRFS_MAX_EXTENT_BIT (27), can we use '>>' instead of div64_u64?
Thanks,
-liubo
Quoted 66 lines
fs/btrfs/inode.c +30
-1+return ; + } +spin_lock (&BTRFS_I (inode)->lock);BTRFS_I (inode)->outstanding_extents++;spin_unlock (&BTRFS_I (inode)->lock);@@ -1549,10 +1584,34 @@ static void btrfs_merge_extent_hook(struct inode *inode,struct extent_state *new, struct extent_state *other) { + u64 new_size, old_size; +u64 num_extents; +/* not delalloc, ignore it */ if (!(other->state &EXTENT_DELALLOC ))return ; + old_size = other->end - other->start +1 ; + new_size = old_size + (new->end - new->start +1 ); + +/* we're not bigger than the max, unreserve the space and go */ +if (new_size <=BTRFS_MAX_EXTENT_SIZE ) { +spin_lock (&BTRFS_I (inode)->lock); +BTRFS_I (inode)->outstanding_extents--; +spin_unlock (&BTRFS_I (inode)->lock); +return ; + } + +/* +* If we grew by another max_extent, just return, we want to keep that +* reserved amount. +*/ + num_extents =div64_u64 (old_size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ); +if (div64_u64 (new_size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ) > num_extents) +return ; +spin_lock (&BTRFS_I (inode)->lock);BTRFS_I (inode)->outstanding_extents--;spin_unlock (&BTRFS_I (inode)->lock);@@ -1648,6 +1707,8 @@ static void btrfs_clear_bit_hook(struct inode *inode, unsigned long *bits) {u64 len = state->end +1 - state->start; +u64 num_extents =div64_u64 (len +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE );spin_lock (&BTRFS_I (inode)->lock);if ((state->state &EXTENT_DEFRAG ) && (*bits &EXTENT_DEFRAG ))@@ -1667,7 +1728,7 @@ static void btrfs_clear_bit_hook(struct inode *inode, *bits &= ~EXTENT_FIRST_DELALLOC ; }else if (!(*bits &EXTENT_DO_ACCOUNTING )) {spin_lock (&BTRFS_I (inode)->lock); -BTRFS_I (inode)->outstanding_extents--; +BTRFS_I (inode)->outstanding_extents -= num_extents;spin_unlock (&BTRFS_I (inode)->lock); }Signature · 6 lines
1.8.3.1 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/3] Btrfs: account for large extents with enospc
Quoted 61 lines · Liu Bo
Quoted 47 lines · Josef Bacik
On our gluster boxes we stream large tar balls of backups onto our fses. With 160gb of ram this means we get really large contiguous ranges of dirty data, but the way our ENOSPC stuff works is that as long as it's contiguous we only hold metadata reservation for one extent. The problem is we limit our extents to 128mb, so we'll end up with at least 800 extents so our enospc accounting is quite a bit lower than what we need. To keep track of this make sure we increase
outstanding_extentsfor every multiple of the max extent size so we can be sure to have enough reserved metadata space. Thanks,Signed-off-by Josef Bacik <[email protected]>--- fs/btrfs/ctree.h | 2 ++ fs/btrfs/extent-tree.c | 16 +++++++++---- fs/btrfs/extent_io.c | 2 +- fs/btrfs/inode.c | 63 +++++++++++++++++++++++++++++++++++++++++++++++++- 4 files changed, 76 insertions(+), 7 deletions(-)fs/btrfs/ctree.h +2
-0@@ -198,6 +198,8 @@ static int btrfs_csum_sizes[] = { 4, 0 };#define BTRFS_DIRTY_METADATA_THRESH (32 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_SIZE (128 * 1024 * 1024) +/* * The key defines the order in the tree, and so it also defines (optimal) * block layout. fs/btrfs/extent-tree.c +2
-1@@ -4966,19 +4966,25 @@ void btrfs_subvolume_release_metadata(struct btrfs_root *root,/** * drop_outstanding_extent - drop an outstanding extent * @inode: the inode we're dropping the extent for +* @num_bytes: the number of bytes we're relaseing. * * This is called when we are freeing up an outstanding extent, either called * after an error or after an extent is written. This will return the number of * reserved extents that need to be freed. This must be called with * BTRFS_I(inode)->lock held. */ -static unsigned drop_outstanding_extent (struct inode *inode) +static unsigned drop_outstanding_extent (struct inode *inode,u64 num_bytes)fs/btrfs/extent-tree.c void btrfs_subvolume_release_metadata(struct btrfs_root *root, · line 4984
Quoted 10 lines
fs/btrfs/extent-tree.c +4
-2{unsigned drop_inode_space =0 ;unsigned dropped_extents =0 ; +unsigned num_extents =0 ; -BUG_ON (!BTRFS_I (inode)->outstanding_extents); -BTRFS_I (inode)->outstanding_extents--; + num_extents = (unsigned )div64_u64 (num_bytes + +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE );A fastpath is better, like
btrfs_merge_extent_hook().
Quoted 1 line
(
num_extents<BTRFS_MAX_EXTENT_SIZE) ?num_extents= 1 : (div64_u64(...))
Agreed.
Quoted 79 lines
Quoted 53 lines
fs/btrfs/extent-tree.c +5
-2+ASSERT (num_extents); +ASSERT (BTRFS_I (inode)->outstanding_extents >= num_extents); +BTRFS_I (inode)->outstanding_extents -= num_extents;if (BTRFS_I (inode)->outstanding_extents ==0 && test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,@@ -5149,7 +5155,7 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) out_fail:spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);/* * If the inodes csum_bytes is the same as the original * csum_bytes then we know we haven't raced with any free()ers @@ -5228,7 +5234,7 @@ void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) num_bytes =ALIGN (num_bytes, root->sectorsize);spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);if (num_bytes) to_free =calc_csum_metadata_size (inode, num_bytes,0 );fs/btrfs/extent_io.c +1
-1@@ -3239,7 +3239,7 @@ static noinline_for_stack int writepage_delalloc(struct inode *inode, page, &delalloc_start, &delalloc_end, - 128 * 1024 * 1024); + BTRFS_MAX_EXTENT_SIZE);if (nr_delalloc ==0 ) { delalloc_start = delalloc_end +1 ;continue ;fs/btrfs/inode.c +3
-0@@ -1530,10 +1530,45 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,static void btrfs_split_extent_hook (struct inode *inode,struct extent_state *orig,u64 split) { +u64 size; +/* not delalloc, ignore it */ if (!(orig->state &EXTENT_DELALLOC ))return ; + size = orig->end - orig->start +1 ;fs/btrfs/inode.c static int run_delalloc_range(struct inode *inode, struct page *locked_page, · line 1549
Quoted 10 lines
fs/btrfs/inode.c +10
-0+if (size >BTRFS_MAX_EXTENT_SIZE ) { +u64 num_extents; +u64 new_size; + +/* +* We need the largest size of the remaining extent to see if we +* need to add a new outstanding extent. Think of the following +* case +* +* [MEAX_EXTENT_SIZEx2 - 4k][4k] s/MEAX/MAX
Quoted 13 lines
fs/btrfs/inode.c +13
-0+* +* The new_size would just be 4k and we'd think we had enough +* outstanding extents for this if we only took one side of the +* split, same goes for the other direction. We need to see if +* the larger size still is the same amount of extents as the +* original size, because if it is we need to add a new +* outstanding extent. But if we split up and the larger size +* is less than the original then we are good to go since we've +* already accounted for the extra extent in our original +* accounting. +*/ + new_size = orig->end - split +1 ; +if ((split - orig->start) > new_size)
Quoted 9 lines
fs/btrfs/inode.c static int run_delalloc_range(struct inode *inode, struct page *locked_page, · line 1568
Quoted 6 lines
fs/btrfs/inode.c +6
-0+ new_size = split - orig->start; + + num_extents =div64_u64 (size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ); +if (div64_u64 (new_size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ) < num_extents)
BTRFS_MAX_EXTENT_SIZE(128mb) means 1 <<BTRFS_MAX_EXTENT_BIT(27), can we use '>>' instead ofdiv64_u64?
Yeah that's not a bad idea, I'll do that. Thanks,
Josef
PATCH v2 3/3 Btrfs: account for large extents with enospc
In reply to Liu BoOn our gluster boxes we stream large tar balls of backups onto our fses. With 160gb of ram this means we get really large contiguous ranges of dirty data, but the way our ENOSPC stuff works is that as long as it's contiguous we only hold metadata reservation for one extent. The problem is we limit our extents to 128mb, so we'll end up with at least 800 extents so our enospc accounting is quite a bit lower than what we need. To keep track of this make sure we increase outstanding_extents for every multiple of the max extent size so we can be sure to have enough reserved metadata space. Thanks,
---V1->V2:
-use bit shift instead of div64_u64.
-skip math if num_bytes < max_size in drop_outstanding_extents
fs/btrfs/ctree.h | 8 ++++++++ fs/btrfs/extent-tree.c | 16 ++++++++++----- fs/btrfs/extent_io.c | 2 +- fs/btrfs/inode.c | 54 +++++++++++++++++++++++++++++++++++++++++++++++++- 4 files changed, 73 insertions(+), 7 deletions(-)
fs/btrfs/ctree.h +8 -0
@@ -198,6 +198,14 @@ static int btrfs_csum_sizes[] = { 4, 0 };#define BTRFS_DIRTY_METADATA_THRESH (32 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_SIZE (128 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_BIT 27 + +static inline u64 btrfs_num_extents (u64 num_bytes) +{ +return (num_bytes +BTRFS_MAX_EXTENT_SIZE -1 ) >>BTRFS_MAX_EXTENT_BIT ; +} +/* * The key defines the order in the tree, and so it also defines (optimal) * block layout.
fs/btrfs/extent-tree.c +11 -5
@@ -4966,19 +4966,25 @@ void btrfs_subvolume_release_metadata(struct btrfs_root *root,/** * drop_outstanding_extent - drop an outstanding extent * @inode: the inode we're dropping the extent for +* @num_bytes: the number of bytes we're relaseing. * * This is called when we are freeing up an outstanding extent, either called * after an error or after an extent is written. This will return the number of * reserved extents that need to be freed. This must be called with * BTRFS_I(inode)->lock held. */ -static unsigned drop_outstanding_extent(struct inode *inode) +static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) { unsigned drop_inode_space = 0; unsigned dropped_extents = 0; + unsigned num_extents = 0; - BUG_ON(!BTRFS_I(inode)->outstanding_extents); - BTRFS_I(inode)->outstanding_extents--; + num_extents = (num_bytes > BTRFS_MAX_EXTENT_SIZE) ? + (unsigned)btrfs_num_extents(num_bytes) : 1; + + ASSERT(num_extents); + ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); + BTRFS_I(inode)->outstanding_extents -= num_extents; if (BTRFS_I(inode)->outstanding_extents == 0 && test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
@@ -5149,7 +5155,7 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) out_fail:spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);/* * If the inodes csum_bytes is the same as the original * csum_bytes then we know we haven't raced with any free()ers
@@ -5228,7 +5234,7 @@ void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) num_bytes =ALIGN (num_bytes, root->sectorsize);spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);if (num_bytes) to_free =calc_csum_metadata_size (inode, num_bytes,0 );
fs/btrfs/extent_io.c +1 -1
@@ -3239,7 +3239,7 @@ static noinline_for_stack int writepage_delalloc(struct inode *inode, page, &delalloc_start, &delalloc_end, - 128 * 1024 * 1024); + BTRFS_MAX_EXTENT_SIZE);if (nr_delalloc ==0 ) { delalloc_start = delalloc_end +1 ;continue ;
fs/btrfs/inode.c +53 -1
86 lines not on this page · raw message
Signature · 1 line
1.8.3.1
Re: [PATCH 3/3 v2] Btrfs: account for large extents with enospc
Quoted 8 lines · Josef Bacik
On our gluster boxes we stream large tar balls of backups onto our fses. With 160gb of ram this means we get really large contiguous ranges of dirty data, but the way our ENOSPC stuff works is that as long as it's contiguous we only hold metadata reservation for one extent. The problem is we limit our extents to 128mb, so we'll end up with at least 800 extents so our enospc accounting is quite a bit lower than what we need. To keep track of this make sure we increase
outstanding_extentsfor every multiple of the max extent size so we can be sure to have enough reserved metadata space. Thanks,
Looks good.
Thanks,
-liubo
Quoted 197 lines
Signed-off-by Josef Bacik <[email protected]>---
V1->V2:(quoted diff) +0
-2-use bit shift instead of div64_u64. -skip math if num_bytes < max_size in drop_outstanding_extents fs/btrfs/ctree.h | 8 ++++++++ fs/btrfs/extent-tree.c | 16 ++++++++++----- fs/btrfs/extent_io.c | 2 +- fs/btrfs/inode.c | 54 +++++++++++++++++++++++++++++++++++++++++++++++++- 4 files changed, 73 insertions(+), 7 deletions(-)fs/btrfs/ctree.h +8
-0@@ -198,6 +198,14 @@ static int btrfs_csum_sizes[] = { 4, 0 };#define BTRFS_DIRTY_METADATA_THRESH (32 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_SIZE (128 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_BIT 27 + +static inline u64 btrfs_num_extents (u64 num_bytes) +{ +return (num_bytes +BTRFS_MAX_EXTENT_SIZE -1 ) >>BTRFS_MAX_EXTENT_BIT ; +} +/* * The key defines the order in the tree, and so it also defines (optimal) * block layout. fs/btrfs/extent-tree.c +11
-5@@ -4966,19 +4966,25 @@ void btrfs_subvolume_release_metadata(struct btrfs_root *root,/** * drop_outstanding_extent - drop an outstanding extent * @inode: the inode we're dropping the extent for +* @num_bytes: the number of bytes we're relaseing. * * This is called when we are freeing up an outstanding extent, either called * after an error or after an extent is written. This will return the number of * reserved extents that need to be freed. This must be called with * BTRFS_I(inode)->lock held. */ -static unsigned drop_outstanding_extent(struct inode *inode) +static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) { unsigned drop_inode_space = 0; unsigned dropped_extents = 0; + unsigned num_extents = 0; - BUG_ON(!BTRFS_I(inode)->outstanding_extents); - BTRFS_I(inode)->outstanding_extents--; + num_extents = (num_bytes > BTRFS_MAX_EXTENT_SIZE) ? + (unsigned)btrfs_num_extents(num_bytes) : 1; + + ASSERT(num_extents); + ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); + BTRFS_I(inode)->outstanding_extents -= num_extents; if (BTRFS_I(inode)->outstanding_extents == 0 && test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,@@ -5149,7 +5155,7 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) out_fail:spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);/* * If the inodes csum_bytes is the same as the original * csum_bytes then we know we haven't raced with any free()ers @@ -5228,7 +5234,7 @@ void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) num_bytes =ALIGN (num_bytes, root->sectorsize);spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);if (num_bytes) to_free =calc_csum_metadata_size (inode, num_bytes,0 );fs/btrfs/extent_io.c +1
-1@@ -3239,7 +3239,7 @@ static noinline_for_stack int writepage_delalloc(struct inode *inode, page, &delalloc_start, &delalloc_end, - 128 * 1024 * 1024); + BTRFS_MAX_EXTENT_SIZE);if (nr_delalloc ==0 ) { delalloc_start = delalloc_end +1 ;continue ;fs/btrfs/inode.c +53
-1@@ -1530,10 +1530,41 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,static void btrfs_split_extent_hook (struct inode *inode,struct extent_state *orig,u64 split) { +u64 size; +/* not delalloc, ignore it */ if (!(orig->state &EXTENT_DELALLOC ))return ; + size = orig->end - orig->start +1 ; +if (size >BTRFS_MAX_EXTENT_SIZE ) { +u64 new_size; + +/* +* We need the largest size of the remaining extent to see if we +* need to add a new outstanding extent. Think of the following +* case +* +* [MEAX_EXTENT_SIZEx2 - 4k][4k] +* +* The new_size would just be 4k and we'd think we had enough +* outstanding extents for this if we only took one side of the +* split, same goes for the other direction. We need to see if +* the larger size still is the same amount of extents as the +* original size, because if it is we need to add a new +* outstanding extent. But if we split up and the larger size +* is less than the original then we are good to go since we've +* already accounted for the extra extent in our original +* accounting. +*/ + new_size = orig->end - split +1 ; +if ((split - orig->start) > new_size) + new_size = split - orig->start; + +if (btrfs_num_extents (size) >btrfs_num_extents (new_size)) +return ; + } +spin_lock (&BTRFS_I (inode)->lock);BTRFS_I (inode)->outstanding_extents++;spin_unlock (&BTRFS_I (inode)->lock);@@ -1549,10 +1580,30 @@ static void btrfs_merge_extent_hook(struct inode *inode,struct extent_state *new, struct extent_state *other) { + u64 new_size, old_size; +/* not delalloc, ignore it */ if (!(other->state &EXTENT_DELALLOC ))return ; + old_size = other->end - other->start +1 ; + new_size = old_size + (new->end - new->start +1 ); + +/* we're not bigger than the max, unreserve the space and go */ +if (new_size <=BTRFS_MAX_EXTENT_SIZE ) { +spin_lock (&BTRFS_I (inode)->lock); +BTRFS_I (inode)->outstanding_extents--; +spin_unlock (&BTRFS_I (inode)->lock); +return ; + } + +/* +* If we grew by another max_extent, just return, we want to keep that +* reserved amount. +*/ +if (btrfs_num_extents (new_size) >btrfs_num_extents (old_size)) +return ; +spin_lock (&BTRFS_I (inode)->lock);BTRFS_I (inode)->outstanding_extents--;spin_unlock (&BTRFS_I (inode)->lock);@@ -1648,6 +1699,7 @@ static void btrfs_clear_bit_hook(struct inode *inode, unsigned long *bits) {u64 len = state->end +1 - state->start; +u64 num_extents =btrfs_num_extents (len);spin_lock (&BTRFS_I (inode)->lock);if ((state->state &EXTENT_DEFRAG ) && (*bits &EXTENT_DEFRAG ))@@ -1667,7 +1719,7 @@ static void btrfs_clear_bit_hook(struct inode *inode, *bits &= ~EXTENT_FIRST_DELALLOC ; }else if (!(*bits &EXTENT_DO_ACCOUNTING )) {spin_lock (&BTRFS_I (inode)->lock); -BTRFS_I (inode)->outstanding_extents--; +BTRFS_I (inode)->outstanding_extents -= num_extents;spin_unlock (&BTRFS_I (inode)->lock); }Signature · 6 lines
1.8.3.1 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/3 v2] Btrfs: account for large extents with enospc
In reply to Josef Bacikfs/btrfs/ctree.h static int btrfs_csum_sizes[] = { 4, 0 }; · line 202
Quoted 8 lines · Josef Bacik
fs/btrfs/ctree.h +2
-0@@ -198,6 +198,14 @@ static int btrfs_csum_sizes[] = { 4, 0 };#define BTRFS_DIRTY_METADATA_THRESH (32 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_SIZE (128 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_BIT 27
#include <linux/ilog2.h>
#define BTRFS_MAX_EXTENT_SHIFT ilog2(BTRFS_MAX_EXTENT_SIZE)
Re: [PATCH 3/3] Btrfs: account for large extents with enospc
In reply to Josef BacikQuoted 10 lines · Josef Bacik
On our gluster boxes we stream large tar balls of backups onto our fses. With 160gb of ram this means we get really large contiguous ranges of dirty data, but the way our ENOSPC stuff works is that as long as it's contiguous we only hold metadata reservation for one extent. The problem is we limit our extents to 128mb, so we'll end up with at least 800 extents so our enospc accounting is quite a bit lower than what we need. To keep track of this make sure we increase
outstanding_extentsfor every multiple of the max extent size so we can be sure to have enough reserved metadata space. Thanks,Signed-off-by Josef Bacik <[email protected]>
Hi Josef,
So this change causes the outstanding_extents to never decrease to zero or stay with very large values due to underflow, causing warnings such as the ones reported at
https://bugzilla.kernel.org/show_bug.cgi?id=94401 or the ASSERT() you added to drop_outstanding_extent() when assertions are enabled.
It's very easy to reproduce this, like with the following fio job config:
[global]
ioengine=sync
direct=1
bssplit=130M/100
fallocate=none
fadvise_hint=0
filename=foobar
[job1]
numjobs=1
size=600M
rw=randwrite
So the following happens:
1) btrfs_delalloc_reserve_metadata, num_bytes 136314880 (num_extents 2), outstanding extents == 0, incremented outstanding extents by 1 (and not num_extents)
2) btrfs_drop_outstanding_extent num_bytes 136314880 (num_extents 2), outstanding extents == 1, decrement outstanding extents by 2 (num_extents) giving 4294967295
For buffered writes (by setting direct=0 in that fio job config), outstanding_extents rarely gets decremented to 0, but I haven't checked why exactly (extent state merges need to happen at least for all the special cases in btrfs_merge_extent_hook).
thanks
Quoted 194 lines
--- fs/btrfs/ctree.h | 2 ++ fs/btrfs/extent-tree.c | 16 +++++++++---- fs/btrfs/extent_io.c | 2 +- fs/btrfs/inode.c | 63 +++++++++++++++++++++++++++++++++++++++++++++++++- 4 files changed, 76 insertions(+), 7 deletions(-)fs/btrfs/ctree.h +2
-0@@ -198,6 +198,8 @@ static int btrfs_csum_sizes[] = { 4, 0 };#define BTRFS_DIRTY_METADATA_THRESH (32 * 1024 * 1024) +#define BTRFS_MAX_EXTENT_SIZE (128 * 1024 * 1024) +/* * The key defines the order in the tree, and so it also defines (optimal) * block layout. fs/btrfs/extent-tree.c +11
-5@@ -4966,19 +4966,25 @@ void btrfs_subvolume_release_metadata(struct btrfs_root *root,/** * drop_outstanding_extent - drop an outstanding extent * @inode: the inode we're dropping the extent for +* @num_bytes: the number of bytes we're relaseing. * * This is called when we are freeing up an outstanding extent, either called * after an error or after an extent is written. This will return the number of * reserved extents that need to be freed. This must be called with * BTRFS_I(inode)->lock held. */ -static unsigned drop_outstanding_extent(struct inode *inode) +static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) { unsigned drop_inode_space = 0; unsigned dropped_extents = 0; + unsigned num_extents = 0; - BUG_ON(!BTRFS_I(inode)->outstanding_extents); - BTRFS_I(inode)->outstanding_extents--; + num_extents = (unsigned)div64_u64(num_bytes + + BTRFS_MAX_EXTENT_SIZE - 1, + BTRFS_MAX_EXTENT_SIZE); + ASSERT(num_extents); + ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); + BTRFS_I(inode)->outstanding_extents -= num_extents; if (BTRFS_I(inode)->outstanding_extents == 0 && test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,@@ -5149,7 +5155,7 @@ int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) out_fail:spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);/* * If the inodes csum_bytes is the same as the original * csum_bytes then we know we haven't raced with any free()ers @@ -5228,7 +5234,7 @@ void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) num_bytes =ALIGN (num_bytes, root->sectorsize);spin_lock (&BTRFS_I (inode)->lock); - dropped =drop_outstanding_extent (inode); + dropped =drop_outstanding_extent (inode, num_bytes);if (num_bytes) to_free =calc_csum_metadata_size (inode, num_bytes,0 );fs/btrfs/extent_io.c +1
-1@@ -3239,7 +3239,7 @@ static noinline_for_stack int writepage_delalloc(struct inode *inode, page, &delalloc_start, &delalloc_end, - 128 * 1024 * 1024); + BTRFS_MAX_EXTENT_SIZE);if (nr_delalloc ==0 ) { delalloc_start = delalloc_end +1 ;continue ;fs/btrfs/inode.c +62
-2@@ -1530,10 +1530,45 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,static void btrfs_split_extent_hook (struct inode *inode,struct extent_state *orig,u64 split) { +u64 size; +/* not delalloc, ignore it */ if (!(orig->state &EXTENT_DELALLOC ))return ; + size = orig->end - orig->start +1 ; +if (size >BTRFS_MAX_EXTENT_SIZE ) { +u64 num_extents; +u64 new_size; + +/* +* We need the largest size of the remaining extent to see if we +* need to add a new outstanding extent. Think of the following +* case +* +* [MEAX_EXTENT_SIZEx2 - 4k][4k] +* +* The new_size would just be 4k and we'd think we had enough +* outstanding extents for this if we only took one side of the +* split, same goes for the other direction. We need to see if +* the larger size still is the same amount of extents as the +* original size, because if it is we need to add a new +* outstanding extent. But if we split up and the larger size +* is less than the original then we are good to go since we've +* already accounted for the extra extent in our original +* accounting. +*/ + new_size = orig->end - split +1 ; +if ((split - orig->start) > new_size) + new_size = split - orig->start; + + num_extents =div64_u64 (size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ); +if (div64_u64 (new_size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ) < num_extents) +return ; + } +spin_lock (&BTRFS_I (inode)->lock);BTRFS_I (inode)->outstanding_extents++;spin_unlock (&BTRFS_I (inode)->lock);@@ -1549,10 +1584,34 @@ static void btrfs_merge_extent_hook(struct inode *inode,struct extent_state *new, struct extent_state *other) { + u64 new_size, old_size; +u64 num_extents; +/* not delalloc, ignore it */ if (!(other->state &EXTENT_DELALLOC ))return ; + old_size = other->end - other->start +1 ; + new_size = old_size + (new->end - new->start +1 ); + +/* we're not bigger than the max, unreserve the space and go */ +if (new_size <=BTRFS_MAX_EXTENT_SIZE ) { +spin_lock (&BTRFS_I (inode)->lock); +BTRFS_I (inode)->outstanding_extents--; +spin_unlock (&BTRFS_I (inode)->lock); +return ; + } + +/* +* If we grew by another max_extent, just return, we want to keep that +* reserved amount. +*/ + num_extents =div64_u64 (old_size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ); +if (div64_u64 (new_size +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE ) > num_extents) +return ; +spin_lock (&BTRFS_I (inode)->lock);BTRFS_I (inode)->outstanding_extents--;spin_unlock (&BTRFS_I (inode)->lock);@@ -1648,6 +1707,8 @@ static void btrfs_clear_bit_hook(struct inode *inode, unsigned long *bits) {u64 len = state->end +1 - state->start; +u64 num_extents =div64_u64 (len +BTRFS_MAX_EXTENT_SIZE -1 , +BTRFS_MAX_EXTENT_SIZE );spin_lock (&BTRFS_I (inode)->lock);if ((state->state &EXTENT_DEFRAG ) && (*bits &EXTENT_DEFRAG ))@@ -1667,7 +1728,7 @@ static void btrfs_clear_bit_hook(struct inode *inode, *bits &= ~EXTENT_FIRST_DELALLOC ; }else if (!(*bits &EXTENT_DO_ACCOUNTING )) {spin_lock (&BTRFS_I (inode)->lock); -BTRFS_I (inode)->outstanding_extents--; +BTRFS_I (inode)->outstanding_extents -= num_extents;spin_unlock (&BTRFS_I (inode)->lock); } --1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Signature · 5 lines
Filipe David Manana, "Reasonable men adapt themselves to the world. Unreasonable men adapt the world to themselves. That's why all progress depends on unreasonable men."
Quoted 7 lines · Josef Bacik
We have this weird dance where we always inc
outstanding_extentswhen we do aO_DIRECTwrite, even if we allocate the entire range. To get around this we also drop the metadata space if we successfully write. This is an unnecessary dance, we only need to jack upoutstanding_extentsif we don't satisfy the entire range request inget_blocks_direct, otherwise we are good using our original reservation. So drop the unconditional inc and the drop of the metadata space that we have for the unconditional inc. Thanks,
Looks good.
Thanks,
-liubo
Quoted 49 lines
Signed-off-by Josef Bacik <[email protected]>--- fs/btrfs/inode.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-)fs/btrfs/inode.c +6
-5@@ -7137,6 +7137,7 @@ static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,u64 start = iblock << inode->i_blkbits;u64 lockstart, lockend;u64 len = bh_result->b_size; +u64 orig_len = len;int unlock_bits =EXTENT_LOCKED ;int ret =0 ;@@ -7272,9 +7273,11 @@ unlock:if (start + len >i_size_read (inode))i_size_write (inode, start + len); -spin_lock (&BTRFS_I (inode)->lock); -BTRFS_I (inode)->outstanding_extents++; -spin_unlock (&BTRFS_I (inode)->lock); +if (len < orig_len) { +spin_lock (&BTRFS_I (inode)->lock); +BTRFS_I (inode)->outstanding_extents++; +spin_unlock (&BTRFS_I (inode)->lock); + } ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockstart + len - 1, EXTENT_DELALLOC, NULL,@@ -8056,8 +8059,6 @@ static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,else if (ret >=0 && (size_t )ret < count)btrfs_delalloc_release_space (inode, count - (size_t )ret); -else -btrfs_delalloc_release_metadata (inode,0 ); } out:if (wakeup)Signature · 6 lines
1.8.3.1 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html