btrfs-progs: Support for BG_TREE feature
Subsystems (2)
v2 cover btrfs-progs: Support for BG_TREE feature
This patchset can be fetched from github:
https://github.com/adam900710/btrfs-progs/tree/bg_tree
Which is based on v5.2.2 tag.
This patchset provides the needed user space infrastructure for BG_TREE feature.
Since it's an new incompatible feature, unlike SKINNY_METADATA, btrfs-progs is needed to convert existing fs (unmounted) to new format.
Now btrfstune can convert regular extent tree fs to bg tree fs to improve mount time.
For the performance improvement, please check the kernel patchset cover letter or the last patch.
(SPOILER ALERT: It's super fast)
Changelog:
v2:
- Rebase to v5.2.2 tag
- Add btrfstune ability to convert existing fs to BG_TREE feature
Qu Wenruo (7):
btrfs-progs: Refactor excluded extent functions to use fs_info
btrfs-progs: Refactor btrfs_read_block_groups()
btrfs-progs: Enable read-write ability for 'bg_tree' feature
btrfs-progs: mkfs: Introduce -O bg-tree
btrfs-progs: dump-tree/dump-super: Introduce support for bg tree
btrfs-progs: check: Introduce support for bg-tree feature
btrfs-progs: btrfstune: Allow to enable bg-tree feature offline
Documentation/btrfstune.asciidoc | 6 + btrfstune.c | 44 +++- check/main.c | 7 +- check/mode-lowmem.c | 9 +- cmds/inspect-dump-super.c | 3 +- cmds/inspect-dump-tree.c | 5 + common/fsfeatures.c | 6 + ctree.h | 18 +- disk-io.c | 29 ++- extent-tree.c | 364 +++++++++++++++++++++++-------- mkfs/common.c | 6 +- mkfs/main.c | 9 + print-tree.c | 3 + transaction.c | 1 + 14 files changed, 402 insertions(+), 108 deletions(-)
Signature · 1 line
2.23.0
PATCH v2 1/7 btrfs-progs: Refactor excluded extent functions to use fs_info
The following functions are just using @root to reach fs_info:
- exclude_super_stripes
- free_excluded_extents
- add_excluded_extent
Refactor them to use fs_info directly.
--- check/main.c | 4 ++-- ctree.h | 4 ++-- extent-tree.c | 20 ++++++++++---------- 3 files changed, 14 insertions(+), 14 deletions(-)
check/main.c +2 -2
@@ -5552,7 +5552,7 @@ static int check_space_cache(struct btrfs_root *root) }if (btrfs_fs_compat_ro (root->fs_info,FREE_SPACE_TREE )) { - ret =exclude_super_stripes (root, cache); + ret =exclude_super_stripes (root->fs_info, cache);if (ret) { errno = -ret; fprintf(stderr,
@@ -5561,7 +5561,7 @@ static int check_space_cache(struct btrfs_root *root)continue ; } ret =load_free_space_tree (root->fs_info, cache); -free_excluded_extents (root, cache); +free_excluded_extents (root->fs_info, cache);if (ret <0 ) { errno = -ret; fprintf(stderr,
ctree.h +2 -2
@@ -2568,9 +2568,9 @@ int btrfs_record_file_extent(struct btrfs_trans_handle *trans, u64 num_bytes);int btrfs_free_block_group (struct btrfs_trans_handle *trans,struct btrfs_fs_info *fs_info,u64 bytenr,u64 len); -void free_excluded_extents (struct btrfs_root *root, +void free_excluded_extents (struct btrfs_fs_info *fs_info,struct btrfs_block_group_cache *cache); -int exclude_super_stripes (struct btrfs_root *root, +int exclude_super_stripes (struct btrfs_fs_info *fs_info,struct btrfs_block_group_cache *cache);u64 add_new_free_space (struct btrfs_block_group_cache *block_group,struct btrfs_fs_info *info,u64 start,u64 end);
extent-tree.c +10 -10
@@ -2716,7 +2716,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)if (btrfs_chunk_readonly (info, cache->key.objectid)) cache->ro =1 ; -exclude_super_stripes (root, cache); +exclude_super_stripes (info, cache); ret = update_space_info(info, cache->flags, found_key.offset, btrfs_block_group_used(&cache->item),
@@ -2760,7 +2760,7 @@ btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type, cache->flags = type;btrfs_set_block_group_flags (&cache->item, type); -exclude_super_stripes (fs_info->extent_root, cache); +exclude_super_stripes (fs_info, cache); ret =update_space_info (fs_info, cache->flags, size, bytes_used, &cache->space_info);BUG_ON (ret);
@@ -3551,16 +3551,16 @@ int btrfs_record_file_extent(struct btrfs_trans_handle *trans, } -static int add_excluded_extent (struct btrfs_root *root, +static int add_excluded_extent (struct btrfs_fs_info *fs_info,u64 start,u64 num_bytes) {u64 end = start + num_bytes -1 ; -set_extent_bits (&root->fs_info->pinned_extents, +set_extent_bits (&fs_info->pinned_extents, start, end,EXTENT_UPTODATE );return 0 ; } -void free_excluded_extents (struct btrfs_root *root, +void free_excluded_extents (struct btrfs_fs_info *fs_info,struct btrfs_block_group_cache *cache) {u64 start, end;
@@ -3568,11 +3568,11 @@ void free_excluded_extents(struct btrfs_root *root, start = cache->key.objectid; end = start + cache->key.offset -1 ; -clear_extent_bits (&root->fs_info->pinned_extents, +clear_extent_bits (&fs_info->pinned_extents, start, end,EXTENT_UPTODATE ); } -int exclude_super_stripes (struct btrfs_root *root, +int exclude_super_stripes (struct btrfs_fs_info *fs_info,struct btrfs_block_group_cache *cache) {u64 bytenr;
@@ -3583,7 +3583,7 @@ int exclude_super_stripes(struct btrfs_root *root,if (cache->key.objectid <BTRFS_SUPER_INFO_OFFSET ) { stripe_len =BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; cache->bytes_super += stripe_len; - ret =add_excluded_extent (root, cache->key.objectid, + ret =add_excluded_extent (fs_info, cache->key.objectid, stripe_len);if (ret)return ret;
@@ -3591,7 +3591,7 @@ int exclude_super_stripes(struct btrfs_root *root,for (i =0 ; i <BTRFS_SUPER_MIRROR_MAX ; i++) { bytenr =btrfs_sb_offset (i); - ret =btrfs_rmap_block (root->fs_info, + ret =btrfs_rmap_block (fs_info, cache->key.objectid, bytenr, &logical, &nr, &stripe_len);if (ret)
@@ -3618,7 +3618,7 @@ int exclude_super_stripes(struct btrfs_root *root, } cache->bytes_super += len; - ret =add_excluded_extent (root, start, len); + ret =add_excluded_extent (fs_info, start, len);if (ret) {kfree (logical);return ret;
Signature · 1 line
2.23.0
Re: [PATCH v2 1/7] btrfs-progs: Refactor excluded extent functions to use fs_info
Looks good,
Signature · 8 lines
Johannes Thumshirn SUSE Labs Filesystems [email protected] +49 911 74053 689 SUSE Software Solutions Germany GmbH Maxfeldstr. 5 90409 Nürnberg Germany (HRB 247165, AG München) Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
Re: [PATCH v2 1/7] btrfs-progs: Refactor excluded extent functions to use fs_info
In reply to Qu WenruoQuoted 8 lines · Qu Wenruo
The following functions are just using @root to reach
fs_info:
-exclude_super_stripes
-free_excluded_extents
-add_excluded_extentRefactor them to use
fs_infodirectly.Signed-off-by Qu Wenruo <[email protected]>
PATCH v2 2/7 btrfs-progs: Refactor btrfs_read_block_groups()
In reply to Qu WenruoThis patch does the following refactor:
- Refactor parameter from @root to @fs_info
- Refactor the large loop body into another function Now we have a helper function, read_one_block_group(), to handle block group cache and space info related routine.
- Refactor the return value
Even we have the code handling ret > 0 from find_first_block_group(), it never works, as when there is no more block group, find_first_block_group() just return -ENOENT other than 1.
This is super confusing, it's almost a mircle it even works.
--- ctree.h | 2 +- disk-io.c | 9 ++- extent-tree.c | 160 ++++++++++++++++++++++++++++---------------------- 3 files changed, 97 insertions(+), 74 deletions(-)
ctree.h +1 -1
@@ -2550,7 +2550,7 @@ int update_space_info(struct btrfs_fs_info *info, u64 flags, u64 total_bytes, u64 bytes_used, struct btrfs_space_info **space_info);int btrfs_free_block_groups (struct btrfs_fs_info *info); -int btrfs_read_block_groups (struct btrfs_root *root); +int btrfs_read_block_groups (struct btrfs_fs_info *info);struct btrfs_block_group_cache *btrfs_add_block_group (struct btrfs_fs_info *fs_info,u64 bytes_used,u64 type,u64 chunk_offset,u64 size);
disk-io.c +6 -3
@@ -983,14 +983,17 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr, fs_info->last_trans_committed = generation;if (extent_buffer_uptodate (fs_info->extent_root->node) && !(flags &OPEN_CTREE_NO_BLOCK_GROUPS )) { - ret =btrfs_read_block_groups (fs_info->tree_root); + ret =btrfs_read_block_groups (fs_info);/* * If we don't find any blockgroups (ENOENT) we're either * restoring or creating the filesystem, where it's expected, * anything else is error */ -if (ret != -ENOENT ) -return -EIO ; +if (ret <0 && ret != -ENOENT ) { + errno = -ret; +error ("failed to read block groups: %m" ); +return ret; + } } key.objectid =BTRFS_FS_TREE_OBJECTID ;
extent-tree.c +90 -70
199 lines not on this page · raw message
Signature · 1 line
2.23.0
Re: [PATCH v2 2/7] btrfs-progs: Refactor btrfs_read_block_groups()
Quoted 11 lines · Qu Wenruo
This patch does the following refactor:
- Refactor parameter from @root to @fs_info- Refactor the large loop body into another function
Now we have a helper function,read_one_block_group(), to handle block group cache and space info related routine.- Refactor the return value
Even we have the code handling ret > 0 fromfind_first_block_group(), it never works, as when there is no more block group,find_first_block_group()just return-ENOENTother than 1.
Can it be separated into patches? My concern is as it alters the return value of the rescue command. So we shall have clarity of a discrete patch to blame. Otherwise I agree its a good change.
Quoted 35 lines
This is super confusing, it's almost a mircle it even works.
Signed-off-by Qu Wenruo <[email protected]>--- ctree.h | 2 +- disk-io.c | 9 ++- extent-tree.c | 160 ++++++++++++++++++++++++++++---------------------- 3 files changed, 97 insertions(+), 74 deletions(-)ctree.h +1
-1@@ -2550,7 +2550,7 @@ int update_space_info(struct btrfs_fs_info *info, u64 flags, u64 total_bytes, u64 bytes_used, struct btrfs_space_info **space_info);int btrfs_free_block_groups (struct btrfs_fs_info *info); -int btrfs_read_block_groups (struct btrfs_root *root); +int btrfs_read_block_groups (struct btrfs_fs_info *info);struct btrfs_block_group_cache *btrfs_add_block_group (struct btrfs_fs_info *fs_info,u64 bytes_used,u64 type,u64 chunk_offset,u64 size);disk-io.c +1
-1@@ -983,14 +983,17 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr, fs_info->last_trans_committed = generation;if (extent_buffer_uptodate (fs_info->extent_root->node) && !(flags &OPEN_CTREE_NO_BLOCK_GROUPS )) { - ret =btrfs_read_block_groups (fs_info->tree_root); + ret =btrfs_read_block_groups (fs_info);/* * If we don't find any blockgroups (ENOENT) we're either * restoring or creating the filesystem, where it's expected,
disk-io.c int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr, · line 997
Quoted 10 lines
disk-io.c +5
-2* anything else is error */ -if (ret != -ENOENT ) -return -EIO ; +if (ret <0 && ret != -ENOENT ) { + errno = -ret; +error ("failed to read block groups: %m" ); +return ret; + } }
As mentioned this alters the rescue command semantics as show below. Earlier we had only -EIO as error now its much more and accurate which is good. fstests is fine but anything else?
cmd_rescue_chunk_recover()
btrfs_recover_chunk_tree()
open_ctree_with_broken_chunk()
btrfs_setup_all_roots()
Thanks, Anand
Quoted 208 lines
disk-io.c +0
-0key.objectid =BTRFS_FS_TREE_OBJECTID ;extent-tree.c +90
-70@@ -2607,6 +2607,13 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)return 0 ; } +/* +* Find a block group which starts >= @key->objectid in extent tree. +* +* Return 0 for found +* Retrun >0 for not found +* Return <0 for error +*/ static int find_first_block_group (struct btrfs_root *root,struct btrfs_path *path,struct btrfs_key *key) {@@ -2636,36 +2643,95 @@ static int find_first_block_group(struct btrfs_root *root,return 0 ; path->slots[0 ]++; } - ret = -ENOENT ; + ret =1 ; error:return ret; } -int btrfs_read_block_groups (struct btrfs_root *root) +/* +* Helper function to read out one BLOCK_GROUP_ITEM and insert it into +* block group cache. +* +* Return 0 if nothing wrong (either insert the bg cache or skip 0 sized bg) +* Return <0 for error. +*/ +static int read_one_block_group (struct btrfs_fs_info *fs_info, +struct btrfs_path *path) { -struct btrfs_path *path; -int ret; -int bit; -struct btrfs_block_group_cache *cache; -struct btrfs_fs_info *info = root->fs_info; +struct extent_io_tree *block_group_cache = &fs_info->block_group_cache; +struct extent_buffer *leaf = path->nodes[0 ];struct btrfs_space_info *space_info; -struct extent_io_tree *block_group_cache; +struct btrfs_block_group_cache *cache;struct btrfs_key key; -struct btrfs_key found_key; -struct extent_buffer *leaf; +int slot = path->slots[0 ]; +int bit =0 ; +int ret; - block_group_cache = &info->block_group_cache; +btrfs_item_key_to_cpu (leaf, &key, slot); +ASSERT (key.type ==BTRFS_BLOCK_GROUP_ITEM_KEY ); + +/* +* Skip 0 sized block group, don't insert them into block +* group cache tree, as its length is 0, it won't get +* freed at close_ctree() time. +*/ +if (key.offset ==0 ) +return 0 ; + + cache =kzalloc (sizeof (*cache),GFP_NOFS ); +if (!cache) +return -ENOMEM ; +read_extent_buffer (leaf, &cache->item, +btrfs_item_ptr_offset (leaf, slot), +sizeof (cache->item)); +memcpy (&cache->key, &key,sizeof (key)); + cache->cached =0 ; + cache->pinned =0 ; + cache->flags =btrfs_block_group_flags (&cache->item); +if (cache->flags &BTRFS_BLOCK_GROUP_DATA ) { + bit =BLOCK_GROUP_DATA ; + }else if (cache->flags &BTRFS_BLOCK_GROUP_SYSTEM ) { + bit =BLOCK_GROUP_SYSTEM ; + }else if (cache->flags &BTRFS_BLOCK_GROUP_METADATA ) { + bit =BLOCK_GROUP_METADATA ; + } +set_avail_alloc_bits (fs_info, cache->flags); +if (btrfs_chunk_readonly (fs_info, cache->key.objectid)) + cache->ro =1 ; +exclude_super_stripes (fs_info, cache); + + ret =update_space_info (fs_info, cache->flags, cache->key.offset, +btrfs_block_group_used (&cache->item), + &space_info); +if (ret <0 ) { +free (cache); +return ret; + } + cache->space_info = space_info; + +set_extent_bits (block_group_cache, cache->key.objectid, + cache->key.objectid + cache->key.offset -1 , + bit |EXTENT_LOCKED ); +set_state_private (block_group_cache, cache->key.objectid, + (unsigned long )cache); +return 0 ; +} - root = info->extent_root; +int btrfs_read_block_groups (struct btrfs_fs_info *fs_info) +{ +struct btrfs_path path; +struct btrfs_root *root; +int ret; +struct btrfs_key key; + + root = fs_info->extent_root; key.objectid =0 ; key.offset =0 ; key.type =BTRFS_BLOCK_GROUP_ITEM_KEY ; - path =btrfs_alloc_path (); -if (!path) -return -ENOMEM ; +btrfs_init_path (&path);while (1 ) { - ret =find_first_block_group (root, path, &key); + ret =find_first_block_group (root, &path, &key);if (ret >0 ) { ret =0 ; goto error;@@ -2673,67 +2739,21 @@ int btrfs_read_block_groups(struct btrfs_root *root)if (ret !=0 ) { goto error; } - leaf = path->nodes[0 ]; -btrfs_item_key_to_cpu (leaf, &found_key, path->slots[0 ]); +btrfs_item_key_to_cpu (path.nodes[0 ], &key, path.slots[0 ]); - cache =kzalloc (sizeof (*cache),GFP_NOFS ); -if (!cache) { - ret = -ENOMEM ; + ret =read_one_block_group (fs_info, &path); +if (ret <0 ) goto error; - } -read_extent_buffer (leaf, &cache->item, -btrfs_item_ptr_offset (leaf, path->slots[0 ]), -sizeof (cache->item)); -memcpy (&cache->key, &found_key,sizeof (found_key)); - cache->cached =0 ; - cache->pinned =0 ; - key.objectid = found_key.objectid + found_key.offset; -if (found_key.offset ==0 ) +if (key.offset ==0 ) key.objectid++; -btrfs_release_path (path); - -/* -* Skip 0 sized block group, don't insert them into block -* group cache tree, as its length is 0, it won't get -* freed at close_ctree() time. -*/ -if (found_key.offset ==0 ) { -free (cache); -continue ; - } - - cache->flags =btrfs_block_group_flags (&cache->item); - bit =0 ; -if (cache->flags &BTRFS_BLOCK_GROUP_DATA ) { - bit =BLOCK_GROUP_DATA ; - }else if (cache->flags &BTRFS_BLOCK_GROUP_SYSTEM ) { - bit =BLOCK_GROUP_SYSTEM ; - }else if (cache->flags &BTRFS_BLOCK_GROUP_METADATA ) { - bit =BLOCK_GROUP_METADATA ; - } -set_avail_alloc_bits (info, cache->flags); -if (btrfs_chunk_readonly (info, cache->key.objectid)) - cache->ro =1 ; - -exclude_super_stripes (info, cache); - - ret =update_space_info (info, cache->flags, found_key.offset, -btrfs_block_group_used (&cache->item), - &space_info); -BUG_ON (ret); - cache->space_info = space_info; - -/* use EXTENT_LOCKED to prevent merging */ -set_extent_bits (block_group_cache, found_key.objectid, - found_key.objectid + found_key.offset -1 , - bit |EXTENT_LOCKED ); -set_state_private (block_group_cache, found_key.objectid, - (unsigned long )cache); +else + key.objectid = key.objectid + key.offset; +btrfs_release_path (&path); } ret =0 ; error: -btrfs_free_path (path); +btrfs_release_path (&path);return ret; }
Re: [PATCH v2 2/7] btrfs-progs: Refactor btrfs_read_block_groups()
Quoted 14 lines · Anand Jain
Quoted 11 lines · Qu Wenruo
This patch does the following refactor:
- Refactor parameter from @root to @fs_info- Refactor the large loop body into another function
Now we have a helper function,read_one_block_group(), to handle block group cache and space info related routine.- Refactor the return value
Even we have the code handling ret > 0 fromfind_first_block_group(), it never works, as when there is no more block group,find_first_block_group()just return-ENOENTother than 1.
Quoted 3 lines
Can it be separated into patches? My concern is as it alters the return value of the rescue command. So we shall have clarity of a discrete patch to blame. Otherwise I agree its a good change.
No problem.
What about 3 patches split by the mentioned 3 refactors?
Quoted 56 lines
Quoted 42 lines
This is super confusing, it's almost a mircle it even works.
Signed-off-by Qu Wenruo <[email protected]>---
ctree.h | 2 +-
disk-io.c | 9 ++-
extent-tree.c| 160 ++++++++++++++++++++++++++++----------------------3 files changed, 97 insertions(+), 74 deletions(-)ctree.h +1
-1@@ -2550,7 +2550,7 @@ int update_space_info(struct btrfs_fs_info *info, u64 flags, u64 total_bytes, u64 bytes_used, struct btrfs_space_info **space_info); int btrfs_free_block_groups(struct btrfs_fs_info *info); -int btrfs_read_block_groups (struct btrfs_root *root); +int btrfs_read_block_groups (struct btrfs_fs_info *info);struct
btrfs_block_group_cache* btrfs_add_block_group(structbtrfs_fs_info*fs_info, u64bytes_used, u64 type,
u64chunk_offset, u64 size);disk-io.c +0
-0@@ -983,14 +983,17 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr, fs_info->last_trans_committed = generation;if (
extent_buffer_uptodate(fs_info->extent_root->node)&& !(flags &OPEN_CTREE_NO_BLOCK_GROUPS)) {
- ret =btrfs_read_block_groups(fs_info->tree_root); + ret =btrfs_read_block_groups(fs_info);
/*
* If we don't find any blockgroups (ENOENT) we're either
* restoring or creating the filesystem, where it's expected,
* anything else is error
*/Quoted 8 lines
- if (ret !=
-ENOENT)
- return-EIO;
+ if (ret < 0 && ret !=-ENOENT) {
+ errno = -ret;
+error("failed to read block groups: %m"); + return ret;
+ }
}As mentioned this alters the rescue command semantics as show below. Earlier we had only
-EIOas error now its much more and accurate which is good. fstests is fine but anything else?
Quoted 4 lines
cmd_rescue_chunk_recover()
btrfs_recover_chunk_tree()
open_ctree_with_broken_chunk()
btrfs_setup_all_roots()
I'm not sure if I got the point.
Although btrfs_setup_all_roots() get called in above call chain, it doesn't have any special handling of -EIO or others.
It just reads the extent tree root.
Would you mind to explain a little more?
Thanks,
Qu
Quoted 209 lines
Thanks, Anand
Quoted 205 lines
key.objectid =
BTRFS_FS_TREE_OBJECTID;extent-tree.c +0
-0@@ -2607,6 +2607,13 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info) return 0; }+/*
+ * Find a block group which starts >= @key->objectid in extent tree.
+ *
+ * Return 0 for found
+ * Retrun >0 for not found
+ * Return <0 for error
+ */
static intfind_first_block_group(struct btrfs_root *root, struct btrfs_path *path, struct btrfs_key *key){extent-tree.c +1
-1@@ -2636,36 +2643,95 @@ static int find_first_block_group(structbtrfs_root *root, return 0; path->slots[0]++; } - ret = -ENOENT; + ret = 1;error:
return ret;
}
-intbtrfs_read_block_groups(struct btrfs_root *root)
+/*
+ * Helper function to read out oneBLOCK_GROUP_ITEMand insert it into
+ * block group cache.
+ *
+ * Return 0 if nothing wrong (either insert the bg cache or skip 0 sized bg)
+ * Return <0 for error.
+ */
+static intread_one_block_group(struct btrfs_fs_info *fs_info, + struct btrfs_path *path)
{
- structbtrfs_path*path;
- int ret;
- int bit;
- structbtrfs_block_group_cache*cache;
- structbtrfs_fs_info*info =root->fs_info;
+ structextent_io_tree*block_group_cache=&fs_info->block_group_cache;
+ structextent_buffer*leaf = path->nodes[0];
structbtrfs_space_info*space_info;
- structextent_io_tree*block_group_cache;
+ structbtrfs_block_group_cache*cache;
structbtrfs_keykey;
- structbtrfs_keyfound_key;
- structextent_buffer*leaf;
+ int slot = path->slots[0];
+ int bit = 0;
+ int ret;
-block_group_cache=&info->block_group_cache;
+btrfs_item_key_to_cpu(leaf, &key, slot);
+ASSERT(key.type == BTRFS_BLOCK_GROUP_ITEM_KEY);extent-tree.c +47
-0+ + /* + * Skip 0 sized block group, don't insert them into block + * group cache tree, as its length is 0, it won't get + * freed at close_ctree() time. + */ + if (key.offset == 0) + return 0; + + cache = kzalloc(sizeof(*cache), GFP_NOFS); + if (!cache) + return -ENOMEM; + read_extent_buffer(leaf, &cache->item, + btrfs_item_ptr_offset(leaf, slot), + sizeof(cache->item)); + memcpy(&cache->key, &key, sizeof(key)); + cache->cached = 0; + cache->pinned = 0; + cache->flags = btrfs_block_group_flags(&cache->item); + if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { + bit = BLOCK_GROUP_DATA; + } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { + bit = BLOCK_GROUP_SYSTEM; + } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) { + bit = BLOCK_GROUP_METADATA; + } + set_avail_alloc_bits(fs_info, cache->flags); + if (btrfs_chunk_readonly(fs_info, cache->key.objectid)) + cache->ro = 1; + exclude_super_stripes(fs_info, cache); + + ret = update_space_info(fs_info, cache->flags, cache->key.offset, + btrfs_block_group_used(&cache->item), + &space_info); + if (ret < 0) { + free(cache); + return ret; + } + cache->space_info = space_info; + + set_extent_bits(block_group_cache, cache->key.objectid, + cache->key.objectid + cache->key.offset - 1, + bit | EXTENT_LOCKED); + set_state_private(block_group_cache, cache->key.objectid, + (unsigned long)cache); + return 0; +}- root =
info->extent_root;extent-tree.c +8
-0+int btrfs_read_block_groups (struct btrfs_fs_info *fs_info) +{ + struct btrfs_path path; + struct btrfs_root *root; + int ret; + struct btrfs_key key; + + root = fs_info->extent_root;key.objectid = 0;
key.offset = 0;
key.type =BTRFS_BLOCK_GROUP_ITEM_KEY;
- path =btrfs_alloc_path();
- if (!path)
- return-ENOMEM;
+btrfs_init_path(&path);
while(1) {
- ret =find_first_block_group(root, path, &key); + ret =find_first_block_group(root, &path, &key); if (ret > 0) {
ret = 0;
goto error;extent-tree.c +7
-51@@ -2673,67 +2739,21 @@ int btrfs_read_block_groups(struct btrfs_root *root) if (ret != 0) { goto error; } - leaf = path->nodes[0]; - btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); + btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]); - cache = kzalloc(sizeof(*cache), GFP_NOFS); - if (!cache) { - ret = -ENOMEM; + ret = read_one_block_group(fs_info, &path); + if (ret < 0) goto error; - } - read_extent_buffer(leaf, &cache->item, - btrfs_item_ptr_offset(leaf, path->slots[0]), - sizeof(cache->item)); - memcpy(&cache->key, &found_key, sizeof(found_key)); - cache->cached = 0; - cache->pinned = 0; - key.objectid = found_key.objectid + found_key.offset; - if (found_key.offset == 0) + if (key.offset == 0) key.objectid++; - btrfs_release_path(path); - - /* - * Skip 0 sized block group, don't insert them into block - * group cache tree, as its length is 0, it won't get - * freed at close_ctree() time. - */ - if (found_key.offset == 0) { - free(cache); - continue; - } - - cache->flags = btrfs_block_group_flags(&cache->item); - bit = 0; - if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { - bit = BLOCK_GROUP_DATA; - } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { - bit = BLOCK_GROUP_SYSTEM; - } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) { - bit = BLOCK_GROUP_METADATA; - } - set_avail_alloc_bits(info, cache->flags); - if (btrfs_chunk_readonly(info, cache->key.objectid)) - cache->ro = 1; - - exclude_super_stripes(info, cache); - - ret = update_space_info(info, cache->flags, found_key.offset, - btrfs_block_group_used(&cache->item), - &space_info); - BUG_ON(ret); - cache->space_info = space_info; - - /* use EXTENT_LOCKED to prevent merging */ - set_extent_bits(block_group_cache, found_key.objectid, - found_key.objectid + found_key.offset - 1, - bit | EXTENT_LOCKED); - set_state_private(block_group_cache, found_key.objectid, - (unsigned long)cache); + else + key.objectid = key.objectid + key.offset; + btrfs_release_path(&path); } ret = 0;error:
-btrfs_free_path(path);
+btrfs_release_path(&path);
return ret;
}
Re: [PATCH v2 2/7] btrfs-progs: Refactor btrfs_read_block_groups()
Quoted 85 lines · Qu Wenruo
Quoted 14 lines · Anand Jain
Quoted 11 lines · Qu Wenruo
This patch does the following refactor:
- Refactor parameter from @root to @fs_info- Refactor the large loop body into another function
Now we have a helper function,read_one_block_group(), to handle block group cache and space info related routine.- Refactor the return value
Even we have the code handling ret > 0 fromfind_first_block_group(), it never works, as when there is no more block group,find_first_block_group()just return-ENOENTother than 1.Quoted 3 lines
Can it be separated into patches? My concern is as it alters the return value of the rescue command. So we shall have clarity of a discrete patch to blame. Otherwise I agree its a good change.
No problem.
What about 3 patches split by the mentioned 3 refactors?
Quoted 56 lines
Quoted 40 lines
This is super confusing, it's almost a mircle it even works.
Signed-off-by Qu Wenruo <[email protected]>--- ctree.h | 2 +- disk-io.c | 9 ++- extent-tree.c | 160 ++++++++++++++++++++++++++++---------------------- 3 files changed, 97 insertions(+), 74 deletions(-)ctree.h +1
-1@@ -2550,7 +2550,7 @@ int update_space_info(struct btrfs_fs_info *info, u64 flags, u64 total_bytes, u64 bytes_used, struct btrfs_space_info **space_info); int btrfs_free_block_groups(struct btrfs_fs_info *info); -int btrfs_read_block_groups (struct btrfs_root *root); +int btrfs_read_block_groups (struct btrfs_fs_info *info); struct btrfs_block_group_cache * btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used,u64 type,
u64chunk_offset, u64 size);disk-io.c +1
-1@@ -983,14 +983,17 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr, fs_info->last_trans_committed = generation; if (extent_buffer_uptodate(fs_info->extent_root->node) && !(flags & OPEN_CTREE_NO_BLOCK_GROUPS)) { - ret = btrfs_read_block_groups(fs_info->tree_root); + ret = btrfs_read_block_groups(fs_info); /** If we don't find any blockgroups (ENOENT) we're either * restoring or creating the filesystem, where it's expected, disk-io.c int btrfs_setup_all_roots(struct btrfs_fs_info · line 998
Quoted 10 lines
disk-io.c +5
-2* anything else is error */ - if (ret != -ENOENT) - return -EIO; + if (ret < 0 && ret != -ENOENT) { + errno = -ret; + error("failed to read block groups: %m"); + return ret; + } }As mentioned this alters the rescue command semantics as show below. Earlier we had only
-EIOas error now its much more and accurate which is good. fstests is fine but anything else?Quoted 4 lines
cmd_rescue_chunk_recover()
btrfs_recover_chunk_tree()
open_ctree_with_broken_chunk()
btrfs_setup_all_roots()
Quoted 8 lines
I'm not sure if I got the point.
Although
btrfs_setup_all_roots()get called in above call chain, it doesn't have any special handling of-EIOor others.It just reads the extent tree root.
Would you mind to explain a little more?
sure.
The above thread is in the call chain of the command
btrfs rescue chunk-recover [options] <device>"
And as the its return error code is being changed for the same problem, so a separate patch not part of the bg-tree changes would make sense.
Thanks, Anand
PATCH v2 3/7 btrfs-progs: Enable read-write ability for 'bg_tree' feature
In reply to Qu WenruoAllow btrfs-progs to open, read and write 'bg_tree' enabled fs.
The modification itself is not large, as block groups items are only used at 4 timing:
1) open_ctree()
We only need to populate fs_info->bg_root and read block group items from fs_info->bg_root.
The obvious change is, we don't need to do btrfs_search_slot() for each block group item, but btrfs_next_item() is enough.
This should hugely reduce open_ctree() execution duration.
2) btrfs_commit_transaction()
We need to write back dirty block group items back to bg_root.
The modification here is to insert new block group item if we can't find one existing in bg_root, and delete the old one in extent tree if we're converting to bg_tree feature.
3) btrfs_make_block_group()
Just change @root for btrfs_insert_item() from @extent_root to @bg_root for fs with that feature.
This modification needs extra handling for converting case, where block group items can be either in extent tree or bg tree.
4) free_block_group_item()
Just delete the block group item in extent tree.
--- ctree.h | 11 +++- disk-io.c | 20 +++++++ extent-tree.c | 152 +++++++++++++++++++++++++++++++++++++++++++++----- 3 files changed, 168 insertions(+), 15 deletions(-)
ctree.h +10 -1
@@ -89,6 +89,9 @@ struct btrfs_free_space_ctl;/* tracks free space in block groups. */ #define BTRFS_FREE_SPACE_TREE_OBJECTID 10ULL +/* store BLOCK_GROUP_ITEMS in a seperate tree */ +#define BTRFS_BLOCK_GROUP_TREE_OBJECTID 11ULL +/* device stats in the device tree */ #define BTRFS_DEV_STATS_OBJECTID 0ULL
@@ -490,6 +493,7 @@ struct btrfs_super_block {#define BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA (1ULL << 8)#define BTRFS_FEATURE_INCOMPAT_NO_HOLES (1ULL << 9)#define BTRFS_FEATURE_INCOMPAT_METADATA_UUID (1ULL << 10) +#define BTRFS_FEATURE_INCOMPAT_BG_TREE (1ULL << 11)#define BTRFS_FEATURE_COMPAT_SUPP 0ULL
@@ -513,7 +517,8 @@ struct btrfs_super_block {BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS | \BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA | \BTRFS_FEATURE_INCOMPAT_NO_HOLES | \ - BTRFS_FEATURE_INCOMPAT_METADATA_UUID) +BTRFS_FEATURE_INCOMPAT_METADATA_UUID | \ + BTRFS_FEATURE_INCOMPAT_BG_TREE)/* * A leaf is full of items. offset and size tell us where to find
@@ -1123,6 +1128,7 @@ struct btrfs_fs_info {struct btrfs_root *quota_root;struct btrfs_root *free_space_root;struct btrfs_root *uuid_root; +struct btrfs_root *bg_root;struct rb_root fs_root_tree;
@@ -1175,6 +1181,9 @@ struct btrfs_fs_info { unsigned int avoid_sys_chunk_alloc:1; unsigned int finalize_on_close:1; +/* Converting from bg in extent tree to bg tree */ + unsigned int convert_to_bg_tree:1; +int transaction_aborted; int (*free_extent_hook)(struct btrfs_fs_info *fs_info,
disk-io.c +20 -0
@@ -716,6 +716,8 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,if (location->objectid ==BTRFS_FREE_SPACE_TREE_OBJECTID )return fs_info->free_space_root ? fs_info->free_space_root :ERR_PTR (-ENOENT ); +if (location->objectid ==BTRFS_BLOCK_GROUP_TREE_OBJECTID ) +return fs_info->bg_root ? fs_info->bg_root :ERR_PTR (-ENOENT );BUG_ON (location->objectid ==BTRFS_TREE_RELOC_OBJECTID || location->offset != (u64)-1 );
@@ -768,6 +770,7 @@ struct btrfs_fs_info *btrfs_new_fs_info(int writable, u64 sb_bytenr) fs_info->quota_root =calloc (1 ,sizeof (struct btrfs_root )); fs_info->free_space_root =calloc (1 ,sizeof (struct btrfs_root )); fs_info->uuid_root =calloc (1 ,sizeof (struct btrfs_root )); + fs_info->bg_root =calloc (1 ,sizeof (struct btrfs_root )); fs_info->super_copy =calloc (1 ,BTRFS_SUPER_INFO_SIZE ); if (!fs_info->tree_root || !fs_info->extent_root ||
@@ -930,6 +933,21 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr,return ret; fs_info->extent_root->track_dirty =1 ; +if (btrfs_fs_incompat (fs_info,BG_TREE )) { + ret =setup_root_or_create_block (fs_info, flags, + fs_info->bg_root, +BTRFS_BLOCK_GROUP_TREE_OBJECTID ,"bg" ); +if (ret <0 ) { +error ("Couldn't setup bg tree" ); +return ret; + } + fs_info->bg_root->track_dirty =1 ; + fs_info->bg_root->ref_cows =0 ; + }else { +free (fs_info->bg_root); + fs_info->bg_root =NULL ; + } + ret =find_and_setup_root (root, fs_info,BTRFS_DEV_TREE_OBJECTID , fs_info->dev_root);if (ret) {
@@ -1012,6 +1030,8 @@ void btrfs_release_all_roots(struct btrfs_fs_info *fs_info)free_extent_buffer (fs_info->free_space_root->node);if (fs_info->quota_root)free_extent_buffer (fs_info->quota_root->node); +if (fs_info->bg_root) +free_extent_buffer (fs_info->bg_root->node);if (fs_info->csum_root)free_extent_buffer (fs_info->csum_root->node);if (fs_info->dev_root)
extent-tree.c +138 -14
200 lines not on this page · raw message
Signature · 1 line
2.23.0
Re: [PATCH v2 3/7] btrfs-progs: Enable read-write ability for 'bg_tree' feature
Quoted 23 lines · Qu Wenruo
Allow btrfs-progs to open, read and write '
bg_tree' enabled fs.The modification itself is not large, as block groups items are only used at 4 timing:
1)open_ctree()
We only need to populatefs_info->bg_rootand read block group items fromfs_info->bg_root.
The obvious change is, we don't need to dobtrfs_search_slot()for each block group item, butbtrfs_next_item()is enough.This should hugely reduce
open_ctree()execution duration.2)
btrfs_commit_transaction()
We need to write back dirty block group items back tobg_root.The modification here is to insert new block group item if we can't find one existing in
bg_root, and delete the old one in extent tree if we're converting tobg_treefeature.3)
btrfs_make_block_group()
Just change @root forbtrfs_insert_item()from @extent_root to @bg_root for fs with that feature.
Quoted 7 lines
This modification needs extra handling for converting case, where block group items can be either in extent tree or bg tree.
4)
free_block_group_item()
Just delete the block group item in extent tree.Signed-off-by Qu Wenruo <[email protected]>
looks good.
two nit below.
Quoted 47 lines
--- ctree.h | 11 +++- disk-io.c | 20 +++++++ extent-tree.c | 152 +++++++++++++++++++++++++++++++++++++++++++++----- 3 files changed, 168 insertions(+), 15 deletions(-)ctree.h +7
-1@@ -89,6 +89,9 @@ struct btrfs_free_space_ctl;/* tracks free space in block groups. */ #define BTRFS_FREE_SPACE_TREE_OBJECTID 10ULL +/* store BLOCK_GROUP_ITEMS in a seperate tree */ +#define BTRFS_BLOCK_GROUP_TREE_OBJECTID 11ULL +/* device stats in the device tree */ #define BTRFS_DEV_STATS_OBJECTID 0ULL@@ -490,6 +493,7 @@ struct btrfs_super_block {#define BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA (1ULL << 8)#define BTRFS_FEATURE_INCOMPAT_NO_HOLES (1ULL << 9)#define BTRFS_FEATURE_INCOMPAT_METADATA_UUID (1ULL << 10) +#define BTRFS_FEATURE_INCOMPAT_BG_TREE (1ULL << 11)#define BTRFS_FEATURE_COMPAT_SUPP 0ULL@@ -513,7 +517,8 @@ struct btrfs_super_block {BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS | \BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA | \BTRFS_FEATURE_INCOMPAT_NO_HOLES | \ - BTRFS_FEATURE_INCOMPAT_METADATA_UUID) +BTRFS_FEATURE_INCOMPAT_METADATA_UUID | \ + BTRFS_FEATURE_INCOMPAT_BG_TREE)/* * A leaf is full of items. offset and size tell us where to find @@ -1123,6 +1128,7 @@ struct btrfs_fs_info {struct btrfs_root *quota_root;struct btrfs_root *free_space_root;struct btrfs_root *uuid_root; +struct btrfs_root *bg_root;struct rb_root fs_root_tree;
ctree.h struct btrfs_fs_info { · line 1186
Quoted 6 lines
ctree.h +3
-0@@ -1175,6 +1181,9 @@ struct btrfs_fs_info { unsigned int avoid_sys_chunk_alloc:1; unsigned int finalize_on_close:1; +/* Converting from bg in extent tree to bg tree */ + unsigned int convert_to_bg_tree:1; +
This should probably be part of the patch 7/7.
Quoted 73 lines
ctree.h +0
-0int transaction_aborted; int (*free_extent_hook)(struct btrfs_fs_info *fs_info,disk-io.c +20
-0@@ -716,6 +716,8 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,if (location->objectid ==BTRFS_FREE_SPACE_TREE_OBJECTID )return fs_info->free_space_root ? fs_info->free_space_root :ERR_PTR (-ENOENT ); +if (location->objectid ==BTRFS_BLOCK_GROUP_TREE_OBJECTID ) +return fs_info->bg_root ? fs_info->bg_root :ERR_PTR (-ENOENT );BUG_ON (location->objectid ==BTRFS_TREE_RELOC_OBJECTID || location->offset != (u64)-1 );@@ -768,6 +770,7 @@ struct btrfs_fs_info *btrfs_new_fs_info(int writable, u64 sb_bytenr) fs_info->quota_root =calloc (1 ,sizeof (struct btrfs_root )); fs_info->free_space_root =calloc (1 ,sizeof (struct btrfs_root )); fs_info->uuid_root =calloc (1 ,sizeof (struct btrfs_root )); + fs_info->bg_root =calloc (1 ,sizeof (struct btrfs_root )); fs_info->super_copy =calloc (1 ,BTRFS_SUPER_INFO_SIZE ); if (!fs_info->tree_root || !fs_info->extent_root ||@@ -930,6 +933,21 @@ int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr,return ret; fs_info->extent_root->track_dirty =1 ; +if (btrfs_fs_incompat (fs_info,BG_TREE )) { + ret =setup_root_or_create_block (fs_info, flags, + fs_info->bg_root, +BTRFS_BLOCK_GROUP_TREE_OBJECTID ,"bg" ); +if (ret <0 ) { +error ("Couldn't setup bg tree" ); +return ret; + } + fs_info->bg_root->track_dirty =1 ; + fs_info->bg_root->ref_cows =0 ; + }else { +free (fs_info->bg_root); + fs_info->bg_root =NULL ; + } + ret =find_and_setup_root (root, fs_info,BTRFS_DEV_TREE_OBJECTID , fs_info->dev_root);if (ret) {@@ -1012,6 +1030,8 @@ void btrfs_release_all_roots(struct btrfs_fs_info *fs_info)free_extent_buffer (fs_info->free_space_root->node);if (fs_info->quota_root)free_extent_buffer (fs_info->quota_root->node); +if (fs_info->bg_root) +free_extent_buffer (fs_info->bg_root->node);if (fs_info->csum_root)free_extent_buffer (fs_info->csum_root->node);if (fs_info->dev_root)extent-tree.c +4
-2@@ -1528,22 +1528,68 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,struct btrfs_path *path, struct btrfs_block_group_cache *cache) { + bool is_bg_tree = btrfs_fs_incompat(trans->fs_info, BG_TREE);int ret; -struct btrfs_root *extent_root = trans->fs_info->extent_root; +struct btrfs_fs_info *fs_info = trans->fs_info; +struct btrfs_root *root;unsigned long bi;struct extent_buffer *leaf; - ret =btrfs_search_slot (trans, extent_root, &cache->key, path,0 ,1 ); +if (is_bg_tree)
extent-tree.c static int write_one_cache_group(struct btrfs_trans_handle *trans, · line 1541
Quoted 10 lines
extent-tree.c +5
-2+ root = fs_info->bg_root; + else + root = fs_info->extent_root; + + ret =btrfs_search_slot (trans, root, &cache->key, path,0 ,1 );if (ret <0 ) - goto fail; -BUG_ON (ret);
Unrelated change ?
Thanks, Anand
Quoted 187 lines
extent-tree.c +129
-10+ goto out; - leaf = path->nodes[0 ]; - bi =btrfs_item_ptr_offset (leaf, path->slots[0 ]); -write_extent_buffer (leaf, &cache->item, bi,sizeof (cache->item)); -btrfs_mark_buffer_dirty (leaf); -btrfs_release_path (path); -fail: +if (ret ==0 ) { +/* Update existing bg */ + leaf = path->nodes[0 ]; + bi =btrfs_item_ptr_offset (leaf, path->slots[0 ]); +write_extent_buffer (leaf, &cache->item, bi,sizeof (cache->item)); +btrfs_mark_buffer_dirty (leaf); +btrfs_release_path (path); + }else { +btrfs_release_path (path); + +/* +* Insert new bg item +* +* This only happens for bg_tree feature +*/ +if (!is_bg_tree) { +error ("can't find block group item for bytenr %llu" , + cache->key.objectid); + ret = -ENOENT ; + goto out; + } + ret =btrfs_insert_item (trans, root, &cache->key, &cache->item, +sizeof (cache->item)); +if (ret <0 ) + goto out; + +/* Also delete the existing one in next tree if needed */ +if (fs_info->convert_to_bg_tree) { + ret =btrfs_search_slot (trans, fs_info->extent_root, + &cache->key, path,-1 ,1 ); +if (ret <0 ) { +btrfs_release_path (path); + goto out; + } +/* Good, already converted */ +if (ret >0 ) { + ret =0 ; +btrfs_release_path (path); + goto out; + } +/* Delete old block group item in extent tree */ + ret =btrfs_del_item (trans, fs_info->extent_root, path); +btrfs_release_path (path); + } + } +out:if (ret)return ret;return 0 ;@@ -2717,14 +2763,66 @@ static int read_one_block_group(struct btrfs_fs_info *fs_info,return 0 ; } +static int read_block_group_tree (struct btrfs_fs_info *fs_info) +{ +struct btrfs_root *root = fs_info->bg_root; +struct btrfs_key key = {0 }; +struct btrfs_path path; +int ret; + +btrfs_init_path (&path); + ret =btrfs_search_slot (NULL , root, &key, &path,0 ,0 ); +if (ret <0 ) { + errno = -ret; +error ("failed to search block group tree: %m" ); +return ret; + } +if (ret ==0 ) + goto invalid_key; + +while (1 ) { +btrfs_item_key_to_cpu (path.nodes[0 ], &key, path.slots[0 ]); +if (key.type !=BTRFS_BLOCK_GROUP_ITEM_KEY ) + goto invalid_key; + + ret =read_one_block_group (fs_info, &path); +if (ret <0 ) { + errno = -ret; +error ("failed to read one block group: %m" ); + goto out; + } + ret =btrfs_next_item (root, &path); +if (ret <0 ) { + errno = -ret; +error ("failed to search block group tree: %m" ); + goto out; + } +if (ret >0 ) { + ret =0 ; +break ; + } + } +out: +btrfs_release_path (&path); +return ret; + +invalid_key: +error ("invalid key (%llu, %u, %llu) found in block group tree" , + key.objectid, key.type, key.offset); +btrfs_release_path (&path); +return -EUCLEAN ; +} +int btrfs_read_block_groups (struct btrfs_fs_info *fs_info) {struct btrfs_path path; -struct btrfs_root *root; +struct btrfs_root *root = fs_info->extent_root;int ret;struct btrfs_key key; - root = fs_info->extent_root; +if (btrfs_fs_incompat (fs_info,BG_TREE )) +return read_block_group_tree (fs_info); + key.objectid =0 ; key.offset =0 ; key.type =BTRFS_BLOCK_GROUP_ITEM_KEY ;@@ -2804,12 +2902,17 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 type, u64 chunk_offset, u64 size) { int ret; -struct btrfs_root *extent_root = fs_info->extent_root; +struct btrfs_root *root;struct btrfs_block_group_cache *cache; +if (btrfs_fs_incompat (fs_info,BG_TREE )) + root = fs_info->bg_root; +else + root = fs_info->extent_root; + cache =btrfs_add_block_group (fs_info, bytes_used, type, chunk_offset, size); - ret =btrfs_insert_item (trans, extent_root, &cache->key, &cache->item, + ret =btrfs_insert_item (trans, root, &cache->key, &cache->item,sizeof (cache->item));BUG_ON (ret);@@ -2943,8 +3046,16 @@ static int free_block_group_item(struct btrfs_trans_handle *trans,if (!path)return -ENOMEM ; +/* Using bg tree only */ +if (btrfs_fs_incompat (fs_info,BG_TREE ) && !fs_info->convert_to_bg_tree) + goto bg_tree; + ret =btrfs_search_slot (trans, root, &key, path,-1 ,1 );if (ret >0 ) { +if (btrfs_fs_incompat (fs_info,BG_TREE )) { +btrfs_release_path (path); + goto bg_tree; + } ret = -ENOENT ; goto out; }@@ -2952,6 +3063,19 @@ static int free_block_group_item(struct btrfs_trans_handle *trans, goto out; ret =btrfs_del_item (trans, root, path); + goto out; + +bg_tree: + root = fs_info->bg_root; + ret =btrfs_search_slot (trans, fs_info->bg_root, &key, path,-1 ,1 ); +if (ret <0 ) + goto out; +if (ret >0 ) { + ret = -ENOENT ; + goto out; + } + ret =btrfs_del_item (trans, root, path); + out:btrfs_free_path (path);return ret;
PATCH v2 4/7 btrfs-progs: mkfs: Introduce -O bg-tree
In reply to Qu WenruoThis allow mkfs.btrfs to create a btrfs with bg-tree feature.
This patch introduce a global function, btrfs_convert_to_bg_tree() in extent-tree.c, to do the work.
The workflow is pretty simple:
- Create a new tree block for bg tree
- Set the BG_TREE feature for superblock
- Set the fs_info->convert_to_bg_tree flag
- Mark all block group items as dirty
- Commit transaction
* With fs_info->convert_to_bg_tree set, we will try to delete the BLOCK_GROUP_ITEM in extent tree first, then write the new BLOCK_GROUP_ITEM into bg tree.
This btrfs_convert_to_bg_tree() will be used in mkfs after the basic fs is created.
--- common/fsfeatures.c | 6 ++++++ ctree.h | 1 + extent-tree.c | 38 ++++++++++++++++++++++++++++++++++++++ mkfs/common.c | 6 ++++-- mkfs/main.c | 9 +++++++++ transaction.c | 1 + 6 files changed, 59 insertions(+), 2 deletions(-)
common/fsfeatures.c +6 -0
@@ -86,6 +86,12 @@ static const struct btrfs_fs_feature { VERSION_TO_STRING2(4,0), NULL, 0, "no explicit hole extents for files" }, + {"bg-tree" ,BTRFS_FEATURE_INCOMPAT_BG_TREE , +"bg_tree" , +VERSION_TO_STRING2 (5 ,0 ), +NULL ,0 , +NULL ,0 , + "store block group items in dedicated tree" },/* Keep this one last */ { "list-all", BTRFS_FEATURE_LIST_ALL, NULL } };
ctree.h +1 -0
@@ -2856,5 +2856,6 @@ int btrfs_read_file(struct btrfs_root *root, u64 ino, u64 start, int len,/* extent-tree.c */ int btrfs_run_delayed_refs (struct btrfs_trans_handle *trans,unsigned long nr); +int btrfs_convert_to_bg_tree (struct btrfs_trans_handle *trans);#endif
extent-tree.c +38 -0
@@ -1524,6 +1524,44 @@ int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,return __btrfs_mod_ref (trans, root, buf, record_parent,0 ); } +int btrfs_convert_to_bg_tree (struct btrfs_trans_handle *trans) +{ +struct btrfs_fs_info *fs_info = trans->fs_info; +struct btrfs_block_group_cache *bg; +struct btrfs_root *bg_root; +u64 features =btrfs_super_incompat_flags (fs_info->super_copy); +int ret; + +/* create bg tree first */ + bg_root =btrfs_create_tree (trans, fs_info,BTRFS_BLOCK_GROUP_TREE_OBJECTID ); +if (IS_ERR (bg_root)) { + ret =PTR_ERR (bg_root); + errno = -ret; +error ("failed to create bg tree: %m" ); +return ret; + } + fs_info->bg_root = bg_root; + fs_info->bg_root->track_dirty =1 ; + fs_info->bg_root->ref_cows =0 ; + +/* set BG_TREE feature and mark the fs into bg_tree convert status */ +btrfs_set_super_incompat_flags (fs_info->super_copy, + features |BTRFS_FEATURE_INCOMPAT_BG_TREE ); + fs_info->convert_to_bg_tree =1 ; + +/* +* Mark all block groups dirty so they will get converted to bg tree at +* commit transaction time +*/ +for (bg =btrfs_lookup_first_block_group (fs_info,0 ); bg; + bg =btrfs_lookup_first_block_group (fs_info, + bg->key.objectid + bg->key.offset)) +set_extent_bits (&fs_info->block_group_cache, bg->key.objectid, + bg->key.objectid + bg->key.offset -1 , +BLOCK_GROUP_DIRTY ); +return 0 ; +} +static int write_one_cache_group (struct btrfs_trans_handle *trans,struct btrfs_path *path,struct btrfs_block_group_cache *cache)
mkfs/common.c +4 -2
@@ -111,6 +111,9 @@ static int btrfs_create_tree_root(int fd, struct btrfs_mkfs_config *cfg,return ret; } +/* These features will not be set in the temporary fs */ +#define MASKED_FEATURES (~(BTRFS_FEATURE_INCOMPAT_BG_TREE)) +/* * @fs_uuid - if NULL, generates a UUID, returns back the new filesystem UUID *
@@ -204,7 +207,7 @@ int make_btrfs(int fd, struct btrfs_mkfs_config *cfg)btrfs_set_super_csum_type (&super,BTRFS_CSUM_TYPE_CRC32 );btrfs_set_super_chunk_root_generation (&super,1 );btrfs_set_super_cache_generation (&super,-1 ); -btrfs_set_super_incompat_flags (&super, cfg->features); +btrfs_set_super_incompat_flags (&super, cfg->features &MASKED_FEATURES );if (cfg->label)__strncpy_null (super.label, cfg->label,BTRFS_LABEL_SIZE -1 );
@@ -824,4 +827,3 @@ int test_minimum_size(const char *file, u64 min_dev_size)return 0 ; } -
mkfs/main.c +9 -0
@@ -1299,6 +1299,15 @@ raid_groups:warning ("unable to create uuid tree, will be created after mount: %d" , ret); +/* Bg tree are converted after the fs is created */ +if (mkfs_cfg.features &BTRFS_FEATURE_INCOMPAT_BG_TREE ) { + ret =btrfs_convert_to_bg_tree (trans); +if (ret <0 ) { + errno = -ret; +error ( +"bg-tree feature will not be enabled, due to error: %m" ); + } + } ret =btrfs_commit_transaction (trans, root);if (ret) {error ("unable to commit transaction: %d" , ret);
transaction.c +1 -0
@@ -225,6 +225,7 @@ commit_tree: root->commit_root =NULL ; fs_info->running_transaction =NULL ; fs_info->last_trans_committed = transid; + fs_info->convert_to_bg_tree =0 ;list_for_each_entry (sinfo, &fs_info->space_info, list) {if (sinfo->bytes_reserved) { warning(
Signature · 1 line
2.23.0
PATCH v2 4/7 btrfs-progs: mkfs: Introduce -O bg-tree
This allow mkfs.btrfs to create a btrfs with bg-tree feature.
This patch introduce a global function, btrfs_convert_to_bg_tree() in extent-tree.c, to do the work.
The workflow is pretty simple:
- Create a new tree block for bg tree
- Set the BG_TREE feature for superblock
- Set the fs_info->convert_to_bg_tree flag
- Mark all block group items as dirty
- Commit transaction
* With fs_info->convert_to_bg_tree set, we will try to delete the BLOCK_GROUP_ITEM in extent tree first, then write the new BLOCK_GROUP_ITEM into bg tree.
This btrfs_convert_to_bg_tree() will be used in mkfs after the basic fs is created.
---
Changelog:
v2.1->v2:
- Convert to bg tree after cleaning up temp chunks
Exposed by btrfs/157 where a test case grep doesn't work as intented, due to SINGLE chunks.
--- common/fsfeatures.c | 6 ++++++ ctree.h | 1 + extent-tree.c | 38 ++++++++++++++++++++++++++++++++++++++ mkfs/common.c | 6 ++++-- mkfs/main.c | 25 +++++++++++++++++++++++++ transaction.c | 1 + 6 files changed, 75 insertions(+), 2 deletions(-)
common/fsfeatures.c +6 -0
@@ -86,6 +86,12 @@ static const struct btrfs_fs_feature { VERSION_TO_STRING2(4,0), NULL, 0, "no explicit hole extents for files" }, + {"bg-tree" ,BTRFS_FEATURE_INCOMPAT_BG_TREE , +"bg_tree" , +VERSION_TO_STRING2 (5 ,0 ), +NULL ,0 , +NULL ,0 , + "store block group items in dedicated tree" },/* Keep this one last */ { "list-all", BTRFS_FEATURE_LIST_ALL, NULL } };
ctree.h +1 -0
@@ -2856,5 +2856,6 @@ int btrfs_read_file(struct btrfs_root *root, u64 ino, u64 start, int len,/* extent-tree.c */ int btrfs_run_delayed_refs (struct btrfs_trans_handle *trans,unsigned long nr); +int btrfs_convert_to_bg_tree (struct btrfs_trans_handle *trans);#endif
extent-tree.c +38 -0
@@ -1524,6 +1524,44 @@ int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,return __btrfs_mod_ref (trans, root, buf, record_parent,0 ); } +int btrfs_convert_to_bg_tree (struct btrfs_trans_handle *trans) +{ +struct btrfs_fs_info *fs_info = trans->fs_info; +struct btrfs_block_group_cache *bg; +struct btrfs_root *bg_root; +u64 features =btrfs_super_incompat_flags (fs_info->super_copy); +int ret; + +/* create bg tree first */ + bg_root =btrfs_create_tree (trans, fs_info,BTRFS_BLOCK_GROUP_TREE_OBJECTID ); +if (IS_ERR (bg_root)) { + ret =PTR_ERR (bg_root); + errno = -ret; +error ("failed to create bg tree: %m" ); +return ret; + } + fs_info->bg_root = bg_root; + fs_info->bg_root->track_dirty =1 ; + fs_info->bg_root->ref_cows =0 ; + +/* set BG_TREE feature and mark the fs into bg_tree convert status */ +btrfs_set_super_incompat_flags (fs_info->super_copy, + features |BTRFS_FEATURE_INCOMPAT_BG_TREE ); + fs_info->convert_to_bg_tree =1 ; + +/* +* Mark all block groups dirty so they will get converted to bg tree at +* commit transaction time +*/ +for (bg =btrfs_lookup_first_block_group (fs_info,0 ); bg; + bg =btrfs_lookup_first_block_group (fs_info, + bg->key.objectid + bg->key.offset)) +set_extent_bits (&fs_info->block_group_cache, bg->key.objectid, + bg->key.objectid + bg->key.offset -1 , +BLOCK_GROUP_DIRTY ); +return 0 ; +} +static int write_one_cache_group (struct btrfs_trans_handle *trans,struct btrfs_path *path,struct btrfs_block_group_cache *cache)
mkfs/common.c +4 -2
@@ -111,6 +111,9 @@ static int btrfs_create_tree_root(int fd, struct btrfs_mkfs_config *cfg,return ret; } +/* These features will not be set in the temporary fs */ +#define MASKED_FEATURES (~(BTRFS_FEATURE_INCOMPAT_BG_TREE)) +/* * @fs_uuid - if NULL, generates a UUID, returns back the new filesystem UUID *
@@ -204,7 +207,7 @@ int make_btrfs(int fd, struct btrfs_mkfs_config *cfg)btrfs_set_super_csum_type (&super,BTRFS_CSUM_TYPE_CRC32 );btrfs_set_super_chunk_root_generation (&super,1 );btrfs_set_super_cache_generation (&super,-1 ); -btrfs_set_super_incompat_flags (&super, cfg->features); +btrfs_set_super_incompat_flags (&super, cfg->features &MASKED_FEATURES );if (cfg->label)__strncpy_null (super.label, cfg->label,BTRFS_LABEL_SIZE -1 );
@@ -824,4 +827,3 @@ int test_minimum_size(const char *file, u64 min_dev_size)return 0 ; } -
mkfs/main.c +25 -0
@@ -1312,6 +1312,31 @@ raid_groups: goto out; } +/* +* Bg tree are converted after temp chunks cleaned up, or we can +* populate temp chunks. +*/ +if (mkfs_cfg.features &BTRFS_FEATURE_INCOMPAT_BG_TREE ) { + trans =btrfs_start_transaction (fs_info->tree_root,1 ); +if (IS_ERR (trans)) { +error ("failed to start transaction: %d" , ret); + goto out; + } + ret =btrfs_convert_to_bg_tree (trans); +if (ret <0 ) { + errno = -ret; +error ( +"bg-tree feature will not be enabled, due to error: %m" ); +btrfs_abort_transaction (trans, ret); + goto out; + } + ret =btrfs_commit_transaction (trans, fs_info->tree_root); +if (ret <0 ) { +error ("failed to commit transaction: %d" , ret); + goto out; + } + } +if (source_dir_set) { ret =btrfs_mkfs_fill_dir (source_dir, root, verbose);if (ret) {
transaction.c +1 -0
@@ -225,6 +225,7 @@ commit_tree: root->commit_root =NULL ; fs_info->running_transaction =NULL ; fs_info->last_trans_committed = transid; + fs_info->convert_to_bg_tree =0 ;list_for_each_entry (sinfo, &fs_info->space_info, list) {if (sinfo->bytes_reserved) { warning(
Signature · 1 line
2.23.0
PATCH v2 5/7 btrfs-progs: dump-tree/dump-super: Introduce support for bg tree
In reply to Qu WenruoJust a new tree called BLOCK_GROUP_TREE.
--- cmds/inspect-dump-super.c | 3 ++- cmds/inspect-dump-tree.c | 5 +++++ print-tree.c | 3 +++ 3 files changed, 10 insertions(+), 1 deletion(-)
cmds/inspect-dump-super.c +2 -1
@@ -229,7 +229,8 @@ static struct readable_flag_entry incompat_flags_array[] = {DEF_INCOMPAT_FLAG_ENTRY (RAID56 ),DEF_INCOMPAT_FLAG_ENTRY (SKINNY_METADATA ),DEF_INCOMPAT_FLAG_ENTRY (NO_HOLES ), - DEF_INCOMPAT_FLAG_ENTRY(METADATA_UUID) +DEF_INCOMPAT_FLAG_ENTRY (METADATA_UUID ), + DEF_INCOMPAT_FLAG_ENTRY(BG_TREE) };static const int incompat_flags_num =sizeof (incompat_flags_array) /sizeof (struct readable_flag_entry );
cmds/inspect-dump-tree.c +5 -0
@@ -150,6 +150,7 @@ static u64 treeid_from_string(const char *str, const char **end) { "CSUM", BTRFS_CSUM_TREE_OBJECTID }, { "CHECKSUM", BTRFS_CSUM_TREE_OBJECTID }, { "QUOTA", BTRFS_QUOTA_TREE_OBJECTID }, + { "BG", BTRFS_BLOCK_GROUP_TREE_OBJECTID }, { "UUID", BTRFS_UUID_TREE_OBJECTID }, { "FREE_SPACE", BTRFS_FREE_SPACE_TREE_OBJECTID }, { "TREE_LOG_FIXUP", BTRFS_TREE_LOG_FIXUP_OBJECTID },
@@ -661,6 +662,10 @@ again:if (!skip)printf ("free space" );break ; +case BTRFS_BLOCK_GROUP_TREE_OBJECTID : +if (!skip) +printf ("block group" ); +break ;case BTRFS_MULTIPLE_OBJECTIDS :if (!skip) {printf ("multiple" );
print-tree.c +3 -0
@@ -770,6 +770,9 @@ void print_objectid(FILE *stream, u64 objectid, u8 type)case BTRFS_FREE_SPACE_TREE_OBJECTID :fprintf (stream,"FREE_SPACE_TREE" );break ; +case BTRFS_BLOCK_GROUP_TREE_OBJECTID : +fprintf (stream,"BLOCK_GROUP_TREE" ); +break ;case BTRFS_MULTIPLE_OBJECTIDS :fprintf (stream,"MULTIPLE" );break ;
Signature · 1 line
2.23.0
PATCH v2 6/7 btrfs-progs: check: Introduce support for bg-tree feature
In reply to Qu WenruoJust some minor modification.
- original mode:
* Block group item can occur in extent tree and bg tree.
- lowmem mode:
* search block group items in bg tree if BG_TREE feature is set.
--- check/main.c | 3 ++- check/mode-lowmem.c | 9 +++++++-- 2 files changed, 9 insertions(+), 3 deletions(-)
check/main.c +2 -1
@@ -6034,7 +6034,8 @@ static int check_type_with_root(u64 rootid, u8 key_type)case BTRFS_EXTENT_ITEM_KEY :case BTRFS_METADATA_ITEM_KEY :case BTRFS_BLOCK_GROUP_ITEM_KEY : -if (rootid !=BTRFS_EXTENT_TREE_OBJECTID ) +if (rootid !=BTRFS_EXTENT_TREE_OBJECTID && + rootid !=BTRFS_BLOCK_GROUP_TREE_OBJECTID ) goto err;break ;case BTRFS_ROOT_ITEM_KEY :
check/mode-lowmem.c +7 -2
@@ -4365,7 +4365,7 @@ next:static int check_chunk_item (struct btrfs_fs_info *fs_info,struct extent_buffer *eb,int slot) { -struct btrfs_root *extent_root = fs_info->extent_root; +struct btrfs_root *root;struct btrfs_root *dev_root = fs_info->dev_root;struct btrfs_path path;struct btrfs_key chunk_key;
@@ -4387,6 +4387,11 @@ static int check_chunk_item(struct btrfs_fs_info *fs_info,int ret;int err =0 ; +if (btrfs_fs_incompat (fs_info,BG_TREE )) + root = fs_info->bg_root; +else + root = fs_info->extent_root; +btrfs_item_key_to_cpu (eb, &chunk_key, slot); chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk); length =btrfs_chunk_length (eb, chunk);
@@ -4406,7 +4411,7 @@ static int check_chunk_item(struct btrfs_fs_info *fs_info, bg_key.offset = length;btrfs_init_path (&path); - ret =btrfs_search_slot (NULL , extent_root, &bg_key, &path,0 ,0 ); + ret =btrfs_search_slot (NULL , root, &bg_key, &path,0 ,0 );if (ret) { error( "chunk[%llu %llu) did not find the related block group item",
Signature · 1 line
2.23.0
PATCH v2 7/7 btrfs-progs: btrfstune: Allow to enable bg-tree feature offline
In reply to Qu WenruoAdd a new option '-b' for btrfstune, to enable bg-tree feature for a unmounted fs.
This feature will convert all BLOCK_GROUP_ITEMs in extent tree to bg tree, by reusing the existing btrfs_convert_to_bg_tree() function.
--- Documentation/btrfstune.asciidoc | 6 +++++ btrfstune.c | 44 ++++++++++++++++++++++++++++++-- 2 files changed, 48 insertions(+), 2 deletions(-)
Documentation/btrfstune.asciidoc +6 -0
@@ -26,6 +26,12 @@ means. Please refer to the 'FILESYSTEM FEATURES' in `btrfs`(5). OPTIONS ------- +-b:: +(since kernel: 5.x) ++ +enable bg-tree feature (faster mount time for large fs), enabled by mkfs +feature 'bg-tree'. + -f:: Allow dangerous changes, e.g. clear the seeding flag or change fsid. Make sure that you are aware of the dangers.
btrfstune.c +42 -2
85 lines not on this page · raw message
Signature · 1 line
2.23.0
Re: [PATCH v2 7/7] btrfs-progs: btrfstune: Allow to enable bg-tree feature offline
Depending on the size of the FS the convert may take longer, further fatal error (power loss; pid kill) may leave the FS in a state where the bg items are in both extent-tree and bg-tree.
The lessons which lead to the implementation of metadata_uuid fsid suggests, for conversions its better to use the btrfstune to only flag the bg convert requirement and let the kernel handle of migration of the bg items from the extent-tree to the bg-tree as and when the bg-items are written.
Thanks, Anand
Quoted 125 lines · Qu Wenruo
Add a new option '-b' for btrfstune, to enable bg-tree feature for a unmounted fs.
This feature will convert all
BLOCK_GROUP_ITEMsin extent tree to bg tree, by reusing the existingbtrfs_convert_to_bg_tree()function.Signed-off-by Qu Wenruo <[email protected]>--- Documentation/btrfstune.asciidoc | 6 +++++ btrfstune.c | 44 ++++++++++++++++++++++++++++++-- 2 files changed, 48 insertions(+), 2 deletions(-)Documentation/btrfstune.asciidoc +6
-0@@ -26,6 +26,12 @@ means. Please refer to the 'FILESYSTEM FEATURES' in `btrfs`(5). OPTIONS ------- +-b:: +(since kernel: 5.x) ++ +enable bg-tree feature (faster mount time for large fs), enabled by mkfs +feature 'bg-tree'. + -f:: Allow dangerous changes, e.g. clear the seeding flag or change fsid. Make sure that you are aware of the dangers.btrfstune.c +42
-2@@ -476,11 +476,39 @@ static void print_usage(void)printf ("\t-m change fsid in metadata_uuid to a random UUID\n" );printf ("\t (incompat change, more lightweight than -u|-U)\n" );printf ("\t-M UUID change fsid in metadata_uuid to UUID\n" ); +printf ("\t-b enable bg-tree feature (mkfs: bg-tree, for faster mount time)\n" );printf (" general:\n" );printf ("\t-f allow dangerous operations, make sure that you are aware of the dangers\n" );printf ("\t--help print this help\n" ); } +static int convert_to_bg_tree (struct btrfs_fs_info *fs_info) +{ +struct btrfs_trans_handle *trans; +int ret; + + trans =btrfs_start_transaction (fs_info->tree_root,1 ); +if (IS_ERR (trans)) { + ret =PTR_ERR (trans); + errno = -ret; +error ("failed to start transaction: %m" ); +return ret; + } + ret =btrfs_convert_to_bg_tree (trans); +if (ret <0 ) { + errno = -ret; +error ("failed to convert: %m" ); +btrfs_abort_transaction (trans, ret); +return ret; + } + ret =btrfs_commit_transaction (trans, fs_info->tree_root); +if (ret <0 ) { + errno = -ret; +error ("failed to commit transaction: %m" ); + } +return ret; +} +int BOX_MAIN (btrfstune )(int argc,char *argv[]) {struct btrfs_root *root;@@ -491,6 +519,7 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[])u64 seeding_value =0 ;int random_fsid =0 ;int change_metadata_uuid =0 ; +bool to_bg_tree = false;char *new_fsid_str =NULL ;int ret;u64 super_flags =0 ;@@ -501,7 +530,7 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[]) { "help", no_argument, NULL, GETOPT_VAL_HELP}, { NULL, 0, NULL, 0 } }; -int c =getopt_long (argc, argv,"S:rxfuU:nmM:" , long_options,NULL ); +int c =getopt_long (argc, argv,"S:rxfuU:nmM:b" , long_options,NULL );if (c <0 )break ;@@ -539,6 +568,9 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[]) ctree_flags |=OPEN_CTREE_IGNORE_FSID_MISMATCH ; change_metadata_uuid =1 ;break ; +case 'b' : + to_bg_tree = true; +break ;case GETOPT_VAL_HELP :default :print_usage ();@@ -556,7 +588,7 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[])return 1 ; }if (!super_flags && !seeding_flag && !(random_fsid || new_fsid_str) && - !change_metadata_uuid) { + !change_metadata_uuid && !to_bg_tree) {error ("at least one option should be specified" );print_usage ();return 1 ;@@ -602,6 +634,14 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[])return 1 ; } +if (to_bg_tree) { + ret =convert_to_bg_tree (root->fs_info); +if (ret <0 ) { + errno = -ret; +error ("failed to convert to bg-tree feature: %m" ); + goto out; + } + }if (seeding_flag) {if (btrfs_fs_incompat (root->fs_info,METADATA_UUID )) {fprintf (stderr,"SEED flag cannot be changed on a metadata-uuid changed fs\n" );
Re: [PATCH v2 7/7] btrfs-progs: btrfstune: Allow to enable bg-tree feature offline
Quoted 4 lines · Anand Jain
Depending on the size of the FS the convert may take longer, further fatal error (power loss; pid kill) may leave the FS in a state where the bg items are in both extent-tree and bg-tree.
That's why I'm using one transaction to convert them all.
So if the convert get interrutped, we're still safe.
Thanks,
Qu
Quoted 144 lines
The lessons which lead to the implementation of
metadata_uuidfsid suggests, for conversions its better to use the btrfstune to only flag the bg convert requirement and let the kernel handle of migration of the bg items from the extent-tree to the bg-tree as and when the bg-items are written.Thanks, Anand
Quoted 132 lines · Qu Wenruo
Add a new option '-b' for btrfstune, to enable bg-tree feature for a unmounted fs.
This feature will convert all
BLOCK_GROUP_ITEMsin extent tree to bg tree, by reusing the existingbtrfs_convert_to_bg_tree()function.Signed-off-by Qu Wenruo <[email protected]>---
Documentation/btrfstune.asciidoc| 6 +++++
btrfstune.c | 44 ++++++++++++++++++++++++++++++--2 files changed, 48 insertions(+), 2 deletions(-)Documentation/btrfstune.asciidoc +0
-0@@ -26,6 +26,12 @@ means. Please refer to the 'FILESYSTEM FEATURES'in
btrfs(5).
OPTIONS
-------
+-b::Documentation/btrfstune.asciidoc +5
-0+(since kernel: 5.x) ++ +enable bg-tree feature (faster mount time for large fs), enabled by mkfs +feature 'bg-tree'. +-f::
Allow dangerous changes, e.g. clear the seeding flag or change fsid. Make sure
that you are aware of the dangers.btrfstune.c +0
-0@@ -476,11 +476,39 @@ static void print_usage(void)
printf("\t-m change fsid in metadata_uuid to a random UUID\n");
printf("\t (incompat change, more lightweight than -u|-U)\n");
printf("\t-M UUID change fsid in metadata_uuid to UUID\n"); +printf("\t-b enable bg-tree feature (mkfs: bg-tree, for faster mount time)\n");
printf(" general:\n");
printf("\t-f allow dangerous operations, make sure that you are aware of the dangers\n");
printf("\t--help print this help\n");
}
+static intconvert_to_bg_tree(struct btrfs_fs_info *fs_info)btrfstune.c +26
-0+{ + struct btrfs_trans_handle *trans; + int ret; + + trans = btrfs_start_transaction(fs_info->tree_root, 1); + if (IS_ERR(trans)) { + ret = PTR_ERR(trans); + errno = -ret; + error("failed to start transaction: %m"); + return ret; + } + ret = btrfs_convert_to_bg_tree(trans); + if (ret < 0) { + errno = -ret; + error("failed to convert: %m"); + btrfs_abort_transaction(trans, ret); + return ret; + } + ret = btrfs_commit_transaction(trans, fs_info->tree_root); + if (ret < 0) { + errno = -ret; + error("failed to commit transaction: %m"); + } + return ret; +} +int BOX_MAIN(btrfstune)(int argc, char *argv[])
{
structbtrfs_root*root;btrfstune.c +2
-1@@ -491,6 +519,7 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[]) u64 seeding_value = 0; int random_fsid = 0; int change_metadata_uuid = 0; + bool to_bg_tree = false; char *new_fsid_str = NULL; int ret; u64 super_flags = 0;@@ -501,7 +530,7 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[]) { "help", no_argument, NULL, GETOPT_VAL_HELP}, { NULL, 0, NULL, 0 } }; - int c = getopt_long(argc, argv, "S:rxfuU:nmM:", long_options, NULL); + int c = getopt_long(argc, argv, "S:rxfuU:nmM:b", long_options,NULL ); if (c < 0)break;
btrfstune.c +3
-0@@ -539,6 +568,9 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[]) ctree_flags |= OPEN_CTREE_IGNORE_FSID_MISMATCH; change_metadata_uuid = 1; break; + case 'b': + to_bg_tree = true; + break;case
GETOPT_VAL_HELP:
default:
print_usage();btrfstune.c +0
-0@@ -556,7 +588,7 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[]) return 1; }if (!super_flags && !seeding_flag && !(random_fsid ||
new_fsid_str) &&
- !change_metadata_uuid) {
+ !change_metadata_uuid && !to_bg_tree) {error("at least one option should be specified");print_usage();
return 1;btrfstune.c +7
-0@@ -602,6 +634,14 @@ int BOX_MAIN(btrfstune)(int argc, char *argv[]) return 1; } + if (to_bg_tree) { + ret = convert_to_bg_tree(root->fs_info); + if (ret < 0) { + errno = -ret; + error("failed to convert to bg-tree feature: %m"); + goto out; + } + } if (seeding_flag) { if (btrfs_fs_incompat(root->fs_info, METADATA_UUID)) {
fprintf(stderr, "SEED flag cannot be changed on a metadata-uuid changed fs\n");
Quoted 4 lines · Qu Wenruo
This patchset can be fetched from github:
https://github.com/adam900710/btrfs-progs/tree/bg_tree
Which is based on v5.2.2 tag.
Quoted 8 lines
This patchset provides the needed user space infrastructure for
BG_TREEfeature.Since it's an new incompatible feature, unlike
SKINNY_METADATA, btrfs-progs is needed to convert existing fs (unmounted) to new format.Now btrfstune can convert regular extent tree fs to bg tree fs to improve mount time.
Have we settled the argument whether to use a new tree or key tricks for the blocgroup data? I think we have not and will read the previous discussions. For a feature like this I want to be sure we understand all the pros and cons.
Quoted 11 lines · David Sterba
Quoted 10 lines · Qu Wenruo
This patchset can be fetched from github:
https://github.com/adam900710/btrfs-progs/tree/bg_tree
Which is based on v5.2.2 tag.This patchset provides the needed user space infrastructure for
BG_TREEfeature.Since it's an new incompatible feature, unlike
SKINNY_METADATA, btrfs-progs is needed to convert existing fs (unmounted) to new format.
Quoted 8 lines
Quoted 2 lines
Now btrfstune can convert regular extent tree fs to bg tree fs to improve mount time.
Have we settled the argument whether to use a new tree or key tricks for the blocgroup data? I think we have not and will read the previous discussions. For a feature like this I want to be sure we understand all the pros and cons.
Yep, we haven't settled on the whether creating a new tree, or re-organize the keys.
But as my last discussion said, I see no obvious pro using the existing extent tree to hold the new block group item keys, even we can pack them all together.
And for backup roots, indeed I forgot to add this feature. But to me that's a minor point, not a show stopper.
The most important aspect to me is, to allow real world user of super large fs to try this feature, to prove the usefulness of this design, other than my on-paper analyse.
That's why I'm pushing the patchset, even it may not pass any review. I just want to hold a up-to-date branch so that when some one needs, it can grab and try them themselves.
Thanks,
Qu
Quoted 11 lines · Qu Wenruo
Quoted 5 lines
Have we settled the argument whether to use a new tree or key tricks for the blocgroup data? I think we have not and will read the previous discussions. For a feature like this I want to be sure we understand all the pros and cons.
Yep, we haven't settled on the whether creating a new tree, or re-organize the keys.
But as my last discussion said, I see no obvious pro using the existing extent tree to hold the new block group item keys, even we can pack them all together.
For me the obvious pro is minimum change to existing set of trees.
Quoted 10 lines
And for backup roots, indeed I forgot to add this feature. But to me that's a minor point, not a show stopper.
The most important aspect to me is, to allow real world user of super large fs to try this feature, to prove the usefulness of this design, other than my on-paper analyse.
That's why I'm pushing the patchset, even it may not pass any review. I just want to hold a up-to-date branch so that when some one needs, it can grab and try them themselves.
Ok that's fine and I can add the branch to for-next for ease of testing. I'm working on a prototype that does it the bg item key way, it compiles and creates almost correct filesystem, so I have to fix it before posting. The patches are on top of your bg-tree feature so we could have both in the same kernel for testing.
Quoted 6 lines · David Sterba
Quoted 5 lines · Qu Wenruo
Quoted 5 lines
Have we settled the argument whether to use a new tree or key tricks for the blocgroup data? I think we have not and will read the previous discussions. For a feature like this I want to be sure we understand all the pros and cons.
Quoted 8 lines
Quoted 6 lines
Yep, we haven't settled on the whether creating a new tree, or re-organize the keys.
But as my last discussion said, I see no obvious pro using the existing extent tree to hold the new block group item keys, even we can pack them all together.
For me the obvious pro is minimum change to existing set of trees.
That's interesting.
And indeed, since we're dealing one less tree, there is no chance to cause the bug mentioned by Josef.
Quoted 12 lines
Quoted 11 lines
And for backup roots, indeed I forgot to add this feature. But to me that's a minor point, not a show stopper.
The most important aspect to me is, to allow real world user of super large fs to try this feature, to prove the usefulness of this design, other than my on-paper analyse.
That's why I'm pushing the patchset, even it may not pass any review. I just want to hold a up-to-date branch so that when some one needs, it can grab and try them themselves.
Quoted 5 lines
Ok that's fine and I can add the branch to for-next for ease of testing. I'm working on a prototype that does it the bg item key way, it compiles and creates almost correct filesystem, so I have to fix it before posting. The patches are on top of your bg-tree feature so we could have both in the same kernel for testing.
That's great!
As long as we're pushing a solution to the mount time problem, I can't be more happier!
Then I guess no matter which version get merged to upstream, the patchset is already meaningful.
Thanks,
Qu
Quoted 14 lines · Qu WenRuo
Quoted 8 lines
Quoted 7 lines
The most important aspect to me is, to allow real world user of super large fs to try this feature, to prove the usefulness of this design, other than my on-paper analyse.
That's why I'm pushing the patchset, even it may not pass any review. I just want to hold a up-to-date branch so that when some one needs, it can grab and try them themselves.
Quoted 5 lines
Ok that's fine and I can add the branch to for-next for ease of testing. I'm working on a prototype that does it the bg item key way, it compiles and creates almost correct filesystem, so I have to fix it before posting. The patches are on top of your bg-tree feature so we could have both in the same kernel for testing.
Quoted 7 lines
That's great!
As long as we're pushing a solution to the mount time problem, I can't be more happier!
Then I guess no matter which version get merged to upstream, the patchset is already meaningful.
We'll see what works in the end, I'm getting to the point where the prototype almost works and am debugging weird problems or making sure it's correct. So I'll dump the ideas here and link to the code so you can have a look.
We agree on the point that the block group items must be packed. The key approach should move the new BGI to the beginning, ie. key type is smaller than anything that appears in the extent tree. I chose 100 for the prototype, it could change.
To keep changes to minimum, the new BGI uses the same block group item, so the only difference then becomes how we search for the items.
Packing of the items is done by swapping the key objectid and offset.
Normal BGI has bg.start == key.objectid and bg.length == key.offset. As the objectid is the thing that scatters the items all over the tree.
So the new BGI has bg.length == key.objectid and bg.start == key.offset. As most of block groups are of same size, or from a small set, they're packed.
The nice thing is that a lot of code can be shared between BGI and new BGI, just needs some care with searches, inserts and search key advances.
I'm now stuck a bit at mkfs, where the 8M block groups are separated by some METADATA_ITEMs
item 0 key (8388608 BLOCK_GROUP_ITEM_NEW 13631488) itemoff 16259 itemsize 24
block group NEW used 0 chunk_objectid 256 flags DATA
item 1 key (8388608 BLOCK_GROUP_ITEM_NEW 22020096) itemoff 16235 itemsize 24
block group NEW used 16384 chunk_objectid 256 flags SYSTEM|DUP
item 2 key (22036480 METADATA_ITEM 0) itemoff 16202 itemsize 33
refs 1 gen 5 flags TREE_BLOCK
tree block skinny level 0
tree block backref root CHUNK_TREE
item 3 key (30408704 METADATA_ITEM 0) itemoff 16169 itemsize 33
refs 1 gen 4 flags TREE_BLOCK
tree block skinny level 0
tree block backref root FS_TREE
item 4 key (30474240 METADATA_ITEM 0) itemoff 16136 itemsize 33
refs 1 gen 4 flags TREE_BLOCK
tree block skinny level 0
tree block backref root CSUM_TREE
item 5 key (30490624 METADATA_ITEM 0) itemoff 16103 itemsize 33
refs 1 gen 4 flags TREE_BLOCK
tree block skinny level 0
tree block backref root DATA_RELOC_TREE
item 6 key (30507008 METADATA_ITEM 0) itemoff 16070 itemsize 33
refs 1 gen 4 flags TREE_BLOCK
tree block skinny level 0
tree block backref root UUID_TREE
item 7 key (30523392 METADATA_ITEM 0) itemoff 16037 itemsize 33
refs 1 gen 5 flags TREE_BLOCK
tree block skinny level 0
tree block backref root EXTENT_TREE
item 8 key (30539776 METADATA_ITEM 0) itemoff 16004 itemsize 33
refs 1 gen 5 flags TREE_BLOCK
tree block skinny level 0
tree block backref root DEV_TREE
item 9 key (30556160 METADATA_ITEM 0) itemoff 15971 itemsize 33
refs 1 gen 5 flags TREE_BLOCK
tree block skinny level 0
tree block backref root ROOT_TREE
item 10 key (107347968 BLOCK_GROUP_ITEM_NEW 30408704) itemoff 15947 itemsize 24
block group NEW used 114688 chunk_objectid 256 flags METADATA|DUP
After item 10, the rest of the block group would appear, and basically the rest of the extent tree, many other items.
I don't want
to make hardcoded assumptins, that maximum objecit is 1G, but so far was not able to come up with a generic and reliable formula how to set up key for next search to reach item (107347968 BLOCK_GROUP_ITEM_NEW 30408704) once (8388608 BLOCK_GROUP_ITEM_NEW 22020096) has been processed.
The swapped objectid and offset is the hard part for search because we lose the linearity of block group start. Advance objectid by one and search again ie. (8388608+1 BGI_NEW 22020096) will land on the next metadata item. Iterating objectid by one would eventually reach the 1G block group item, but what to do after the last 1G item is found and we want do decide wheather to continue or not?
This would be easy with the bg_tree, because we'd know that all items in the tree are just the block group items. Some sort of enumeration could work for bg_key too, but I don't have something solid.
The WIP is in my github repository branch dev/bg-key. It's not on top of the bg_tree branch for now. The kernel part will be very similar once the progs side is done.
Feedback welcome.
Quoted 20 lines · David Sterba
Quoted 19 lines · Qu WenRuo
Quoted 8 lines
Quoted 7 lines
The most important aspect to me is, to allow real world user of super large fs to try this feature, to prove the usefulness of this design, other than my on-paper analyse.
That's why I'm pushing the patchset, even it may not pass any review. I just want to hold a up-to-date branch so that when some one needs, it can grab and try them themselves.
Quoted 5 lines
Ok that's fine and I can add the branch to for-next for ease of testing. I'm working on a prototype that does it the bg item key way, it compiles and creates almost correct filesystem, so I have to fix it before posting. The patches are on top of your bg-tree feature so we could have both in the same kernel for testing.
That's great!
As long as we're pushing a solution to the mount time problem, I can't be more happier!
Quoted 7 lines
Quoted 2 lines
Then I guess no matter which version get merged to upstream, the patchset is already meaningful.
We'll see what works in the end, I'm getting to the point where the prototype almost works and am debugging weird problems or making sure it's correct. So I'll dump the ideas here and link to the code so you can have a look.
That's wonderful.
Although I guess my patchset should provide the hint of where to modify the code, since there are only a limited number of places we modify block group item.
Quoted 8 lines
We agree on the point that the block group items must be packed. The key approach should move the new BGI to the beginning, ie. key type is smaller than anything that appears in the extent tree. I chose 100 for the prototype, it could change.
To keep changes to minimum, the new BGI uses the same block group item, so the only difference then becomes how we search for the items.
If we're introducing new block group item, I hope to do a minor change.
Remove the chunk_objectid member, or even flags. to make it more compact. So that you can make the BGI subtree even smaller.
But I guess since you don't want to modify the BGI structure, and keep the code modification minimal, it may not be a good idea right now.
Quoted 9 lines
Packing of the items is done by swapping the key objectid and offset.
Normal BGI has bg.start == key.objectid and bg.length == key.offset. As the objectid is the thing that scatters the items all over the tree.
So the new BGI has bg.length == key.objectid and bg.start == key.offset. As most of block groups are of same size, or from a small set, they're packed.
That doesn't look optimized enough.
bg.length can be at 1G, that means if extents starts below 1G can still be before BGIs.
I believe we should have a fixed objectid for this new BGIs, so that they are ensured to be at the beginning of extent tree.
Quoted 4 lines
The nice thing is that a lot of code can be shared between BGI and new BGI, just needs some care with searches, inserts and search key advances.
Exactly, but since we're introducing a new key type, I prefer to perfect it. Not only change the key, but also the block group item structure to make it more compact.
Although from the design aspect, it looks like BG tree along with new BGI would be the best design.
New BG key goes (bg start, NEW BGI TYPE, used) no data. It would provide the most compact on-disk format.
Quoted 4 lines
I'm now stuck a bit at mkfs, where the 8M block groups are separated by some
METADATA_ITEMs
Quoted 12 lines
item 0 key (8388608
BLOCK_GROUP_ITEM_NEW13631488) itemoff 16259 itemsize 24
block group NEW used 0chunk_objectid256 flags DATA
item 1 key (8388608BLOCK_GROUP_ITEM_NEW22020096) itemoff 16235 itemsize 24
block group NEW used 16384chunk_objectid256 flags SYSTEM|DUP
item 2 key (22036480METADATA_ITEM0) itemoff 16202 itemsize 33
refs 1 gen 5 flagsTREE_BLOCK
tree block skinny level 0
tree block backref rootCHUNK_TREE
item 3 key (30408704METADATA_ITEM0) itemoff 16169 itemsize 33
refs 1 gen 4 flagsTREE_BLOCK
tree block skinny level 0
tree block backref rootFS_TREE
Exactly the problem I described.
Quoted 44 lines
item 4 key (30474240
METADATA_ITEM0) itemoff 16136 itemsize 33
refs 1 gen 4 flagsTREE_BLOCK
tree block skinny level 0
tree block backref rootCSUM_TREE
item 5 key (30490624METADATA_ITEM0) itemoff 16103 itemsize 33
refs 1 gen 4 flagsTREE_BLOCK
tree block skinny level 0
tree block backref rootDATA_RELOC_TREE
item 6 key (30507008METADATA_ITEM0) itemoff 16070 itemsize 33
refs 1 gen 4 flagsTREE_BLOCK
tree block skinny level 0
tree block backref rootUUID_TREE
item 7 key (30523392METADATA_ITEM0) itemoff 16037 itemsize 33
refs 1 gen 5 flagsTREE_BLOCK
tree block skinny level 0
tree block backref rootEXTENT_TREE
item 8 key (30539776METADATA_ITEM0) itemoff 16004 itemsize 33
refs 1 gen 5 flagsTREE_BLOCK
tree block skinny level 0
tree block backref rootDEV_TREE
item 9 key (30556160METADATA_ITEM0) itemoff 15971 itemsize 33
refs 1 gen 5 flagsTREE_BLOCK
tree block skinny level 0
tree block backref rootROOT_TREE
item 10 key (107347968BLOCK_GROUP_ITEM_NEW30408704) itemoff 15947 itemsize 24
block group NEW used 114688chunk_objectid256 flags METADATA|DUPAfter item 10, the rest of the block group would appear, and basically the rest of the extent tree, many other items.
I don't want
to make hardcoded assumptins, that maximum objecit is 1G, but so far was not able to come up with a generic and reliable formula how to set up key for next search to reach item (107347968BLOCK_GROUP_ITEM_NEW30408704) once (8388608BLOCK_GROUP_ITEM_NEW22020096) has been processed.The swapped objectid and offset is the hard part for search because we lose the linearity of block group start. Advance objectid by one and search again ie. (8388608+1
BGI_NEW22020096) will land on the next metadata item. Iterating objectid by one would eventually reach the 1G block group item, but what to do after the last 1G item is found and we want do decide wheather to continue or not?
Quoted 3 lines
This would be easy with the
bg_tree, because we'd know that all items in the tree are just the block group items. Some sort of enumeration could work forbg_keytoo, but I don't have something solid.
Why not fixed objectid for BGI and just ignore the bg.len part?
We have chunk<->BGI verification code, no bg.len is not a problem at all, we can still make sure chunk<->bg is 1:1 mapped and still verify the bg.used.
Thanks,
Qu
Quoted 7 lines
The WIP is in my github repository branch dev/bg-key. It's not on top of the
bg_treebranch for now. The kernel part will be very similar once the progs side is done.Feedback welcome.
Quoted 4 lines · Qu Wenruo
That's wonderful.
Although I guess my patchset should provide the hint of where to modify the code, since there are only a limited number of places we modify block group item.
I indeed started at your patchset and grepped fro BG_TREE, adding another branch.
Quoted 5 lines
Quoted 5 lines
We agree on the point that the block group items must be packed. The key approach should move the new BGI to the beginning, ie. key type is smaller than anything that appears in the extent tree. I chose 100 for the prototype, it could change.
Quoted 7 lines
Quoted 2 lines
To keep changes to minimum, the new BGI uses the same block group item, so the only difference then becomes how we search for the items.
If we're introducing new block group item, I hope to do a minor change.
Remove the
chunk_objectidmember, or even flags. to make it more compact. So that you can make the BGI subtree even smaller.
Yeah that can be done.
Quoted 2 lines
But I guess since you don't want to modify the BGI structure, and keep the code modification minimal, it may not be a good idea right now.
As long as the changes are bearable, eg. a minor refactoring of block group access (the cache.key serving a as offset and length) is fine. And if we can make the b-tree item more then let's do it.
Quoted 5 lines
Quoted 5 lines
Packing of the items is done by swapping the key objectid and offset.
Normal BGI has bg.start == key.objectid and bg.length == key.offset. As the objectid is the thing that scatters the items all over the tree.
Quoted 8 lines
Quoted 3 lines
So the new BGI has bg.length == key.objectid and bg.start == key.offset. As most of block groups are of same size, or from a small set, they're packed.
That doesn't look optimized enough.
bg.length can be at 1G, that means if extents starts below 1G can still be before BGIs.
This shold not be a big problem, the items are still grouped togethers. Mkfs does 8M, we can have 256M or 1G. On average there could be several packed groups, which I think is fine and the estimated overhead would be a few more seeks.
Quoted 2 lines
I believe we should have a fixed objectid for this new BGIs, so that they are ensured to be at the beginning of extent tree.
That was my idea too, but that also requires to add one more member to the item to track the length. Currently the key is saves the bytes. With the proposed changes to drop chunk_objectid, the overall size of BGI would not increase so this still sounds ok. And all the problems with searching would go away.
Quoted 8 lines
Quoted 3 lines
The nice thing is that a lot of code can be shared between BGI and new BGI, just needs some care with searches, inserts and search key advances.
Exactly, but since we're introducing a new key type, I prefer to perfect it. Not only change the key, but also the block group item structure to make it more compact.
Quoted 5 lines
Although from the design aspect, it looks like BG tree along with new BGI would be the best design.
New BG key goes (bg start, NEW BGI TYPE, used) no data. It would provide the most compact on-disk format.
That's very compact. Given the 'bg start' we can't use the same for the extent tree item.
Quoted 5 lines
Quoted 3 lines
This would be easy with the
bg_tree, because we'd know that all items in the tree are just the block group items. Some sort of enumeration could work forbg_keytoo, but I don't have something solid.Why not fixed objectid for BGI and just ignore the bg.len part?
I wanted to avoid storing a useless number, it costs 8 bytes per item, and the simple swap of objectid/offset was first idea how to avoid it.
Quoted 3 lines
We have chunk<->BGI verification code, no bg.len is not a problem at all, we can still make sure chunk<->bg is 1:1 mapped and still verify the bg.used.
This is all great, sure, and makes the extensions easy. Let's try to work out best for each approach we have so far. Having a separate tree for block groups may open some future options.
Quoted 85 lines · David Sterba
Quoted 4 lines · Qu Wenruo
That's wonderful.
Although I guess my patchset should provide the hint of where to modify the code, since there are only a limited number of places we modify block group item.I indeed started at your patchset and grepped fro
BG_TREE, adding another branch.Quoted 5 lines
Quoted 5 lines
We agree on the point that the block group items must be packed. The key approach should move the new BGI to the beginning, ie. key type is smaller than anything that appears in the extent tree. I chose 100 for the prototype, it could change.
Quoted 7 lines
Quoted 2 lines
To keep changes to minimum, the new BGI uses the same block group item, so the only difference then becomes how we search for the items.
If we're introducing new block group item, I hope to do a minor change.
Remove the
chunk_objectidmember, or even flags. to make it more compact. So that you can make the BGI subtree even smaller.Yeah that can be done.
Quoted 2 lines
But I guess since you don't want to modify the BGI structure, and keep the code modification minimal, it may not be a good idea right now.
As long as the changes are bearable, eg. a minor refactoring of block group access (the cache.key serving a as offset and length) is fine. And if we can make the b-tree item more then let's do it.
Quoted 5 lines
Quoted 5 lines
Packing of the items is done by swapping the key objectid and offset.
Normal BGI has bg.start == key.objectid and bg.length == key.offset. As the objectid is the thing that scatters the items all over the tree.
Quoted 8 lines
Quoted 3 lines
So the new BGI has bg.length == key.objectid and bg.start == key.offset. As most of block groups are of same size, or from a small set, they're packed.
That doesn't look optimized enough.
bg.length can be at 1G, that means if extents starts below 1G can still be before BGIs.
This shold not be a big problem, the items are still grouped togethers. Mkfs does 8M, we can have 256M or 1G. On average there could be several packed groups, which I think is fine and the estimated overhead would be a few more seeks.
Quoted 2 lines
I believe we should have a fixed objectid for this new BGIs, so that they are ensured to be at the beginning of extent tree.
That was my idea too, but that also requires to add one more member to the item to track the length. Currently the key is saves the bytes. With the proposed changes to drop
chunk_objectid, the overall size of BGI would not increase so this still sounds ok. And all the problems with searching would go away.Quoted 8 lines
Quoted 3 lines
The nice thing is that a lot of code can be shared between BGI and new BGI, just needs some care with searches, inserts and search key advances.
Exactly, but since we're introducing a new key type, I prefer to perfect it. Not only change the key, but also the block group item structure to make it more compact.
Quoted 5 lines
Although from the design aspect, it looks like BG tree along with new BGI would be the best design.
New BG key goes (bg start, NEW BGI TYPE, used) no data. It would provide the most compact on-disk format.
That's very compact. Given the 'bg start' we can't use the same for the extent tree item.
Quoted 5 lines
Quoted 3 lines
This would be easy with the
bg_tree, because we'd know that all items in the tree are just the block group items. Some sort of enumeration could work forbg_keytoo, but I don't have something solid.Why not fixed objectid for BGI and just ignore the bg.len part?
I wanted to avoid storing a useless number, it costs 8 bytes per item, and the simple swap of objectid/offset was first idea how to avoid it.
Quoted 7 lines
Quoted 3 lines
We have chunk<->BGI verification code, no bg.len is not a problem at all, we can still make sure chunk<->bg is 1:1 mapped and still verify the bg.used.
This is all great, sure, and makes the extensions easy. Let's try to work out best for each approach we have so far. Having a separate tree for block groups may open some future options.
Great, I'll explore the most compact key only method with BG_TREE.
And maybe also try the fixed key objectid solution, just dropping the bg.length, while keep the current BGI structure.
Thanks,
Qu
BTW, there is one important compatibility problem related to all the BGI related features.
Although I'm putting the BG_TREE feature as incompatible feature, but in theory, it should be RO compatible.
As except extent/bg tree, we *should* read the fs without any problem.
But the problem is, current btrfs mount (including btrfs-check) still need to go through the block group item search, even for permanent RO mount.
This get my rescue mount option patchset to be involved. If we have such skip_bg feature earlier, we can completely afford to make all these potential features as RO compatible.
Now my question is, should we put this feature still as incompatible feature?
Thanks,
Qu
Quoted 104 lines · Qu Wenruo
Quoted 85 lines · David Sterba
Quoted 4 lines · Qu Wenruo
That's wonderful.
Although I guess my patchset should provide the hint of where to modify the code, since there are only a limited number of places we modify block group item.I indeed started at your patchset and grepped fro
BG_TREE, adding another branch.Quoted 5 lines
Quoted 5 lines
We agree on the point that the block group items must be packed. The key approach should move the new BGI to the beginning, ie. key type is smaller than anything that appears in the extent tree. I chose 100 for the prototype, it could change.
Quoted 7 lines
Quoted 2 lines
To keep changes to minimum, the new BGI uses the same block group item, so the only difference then becomes how we search for the items.
If we're introducing new block group item, I hope to do a minor change.
Remove the
chunk_objectidmember, or even flags. to make it more compact. So that you can make the BGI subtree even smaller.Yeah that can be done.
Quoted 2 lines
But I guess since you don't want to modify the BGI structure, and keep the code modification minimal, it may not be a good idea right now.
As long as the changes are bearable, eg. a minor refactoring of block group access (the cache.key serving a as offset and length) is fine. And if we can make the b-tree item more then let's do it.
Quoted 5 lines
Quoted 5 lines
Packing of the items is done by swapping the key objectid and offset.
Normal BGI has bg.start == key.objectid and bg.length == key.offset. As the objectid is the thing that scatters the items all over the tree.
Quoted 8 lines
Quoted 3 lines
So the new BGI has bg.length == key.objectid and bg.start == key.offset. As most of block groups are of same size, or from a small set, they're packed.
That doesn't look optimized enough.
bg.length can be at 1G, that means if extents starts below 1G can still be before BGIs.
This shold not be a big problem, the items are still grouped togethers. Mkfs does 8M, we can have 256M or 1G. On average there could be several packed groups, which I think is fine and the estimated overhead would be a few more seeks.
Quoted 2 lines
I believe we should have a fixed objectid for this new BGIs, so that they are ensured to be at the beginning of extent tree.
That was my idea too, but that also requires to add one more member to the item to track the length. Currently the key is saves the bytes. With the proposed changes to drop
chunk_objectid, the overall size of BGI would not increase so this still sounds ok. And all the problems with searching would go away.Quoted 8 lines
Quoted 3 lines
The nice thing is that a lot of code can be shared between BGI and new BGI, just needs some care with searches, inserts and search key advances.
Exactly, but since we're introducing a new key type, I prefer to perfect it. Not only change the key, but also the block group item structure to make it more compact.
Quoted 5 lines
Although from the design aspect, it looks like BG tree along with new BGI would be the best design.
New BG key goes (bg start, NEW BGI TYPE, used) no data. It would provide the most compact on-disk format.
That's very compact. Given the 'bg start' we can't use the same for the extent tree item.
Quoted 5 lines
Quoted 3 lines
This would be easy with the
bg_tree, because we'd know that all items in the tree are just the block group items. Some sort of enumeration could work forbg_keytoo, but I don't have something solid.Why not fixed objectid for BGI and just ignore the bg.len part?
I wanted to avoid storing a useless number, it costs 8 bytes per item, and the simple swap of objectid/offset was first idea how to avoid it.
Quoted 7 lines
Quoted 3 lines
We have chunk<->BGI verification code, no bg.len is not a problem at all, we can still make sure chunk<->bg is 1:1 mapped and still verify the bg.used.
This is all great, sure, and makes the extensions easy. Let's try to work out best for each approach we have so far. Having a separate tree for block groups may open some future options.
Great, I'll explore the most compact key only method with
BG_TREE.And maybe also try the fixed key objectid solution, just dropping the bg.length, while keep the current BGI structure.
Thanks,
Qu
Quoted 5 lines · Qu Wenruo
BTW, there is one important compatibility problem related to all the BGI related features.
Although I'm putting the
BG_TREEfeature as incompatible feature, but in theory, it should be RO compatible.
It could be RO compatible yes.
Quoted 5 lines
As except extent/bg tree, we *should* read the fs without any problem.
But the problem is, current btrfs mount (including btrfs-check) still need to go through the block group item search, even for permanent RO mount.
Quoted 7 lines
This get my rescue mount option patchset to be involved. If we have such
skip_bgfeature earlier, we can completely afford to make all these potential features as RO compatible.Now my question is, should we put this feature still as incompatible feature?
In some way it would probably have to be incompat, either full or RO. As unknown tree items are ignored, if the rest of the filesystem provides enough information to access the data, then incompat RO sounds like the best option. And that's probably independent of how exactly the new BGI is done.
Quoted 19 lines · David Sterba
Quoted 5 lines · Qu Wenruo
BTW, there is one important compatibility problem related to all the BGI related features.
Although I'm putting the
BG_TREEfeature as incompatible feature, but in theory, it should be RO compatible.It could be RO compatible yes.
Quoted 10 lines
As except extent/bg tree, we *should* read the fs without any problem.
But the problem is, current btrfs mount (including btrfs-check) still need to go through the block group item search, even for permanent RO mount.
This get my rescue mount option patchset to be involved. If we have such
skip_bgfeature earlier, we can completely afford to make all these potential features as RO compatible.
Quoted 9 lines
Quoted 2 lines
Now my question is, should we put this feature still as incompatible feature?
In some way it would probably have to be incompat, either full or RO. As unknown tree items are ignored, if the rest of the filesystem provides enough information to access the data, then incompat RO sounds like the best option. And that's probably independent of how exactly the new BGI is done.
But the problem is, older fs can't mount it at all, no matter RO or RW.
It's now completely dependent on how older kernel handles missing block group items.
If current kernel has something like skip_bg or fall back to RO + skip_bg by default, then it's really RO compatible.
But we all know how picky current btrfs is about missing block group items...
So, I guess we will have a RO compatible feature that can't really be mounted RO by older kernel at last?
Thanks,
Qu