btrfs: defrag: fix the wrong number of defragged sectors
PATCH v2 1/1 btrfs: defrag: fix the wrong number of defragged sectors
[BUG]
There are users using autodefrag mount option reporting obvious increase in IO:
Quoted 9 lines
If I compare the write average (in total, I don't have it per process) when taking idle periods on the same machine:
Linux 5.16:
without autodefrag: ~ 10KiB/s
with autodefrag: between 1 and 2MiB/s.Linux 5.15:
with autodefrag:~ 10KiB/s (around the same as without
autodefrag on 5.16)
[CAUSE]
When autodefrag mount option is enabled, btrfs_defrag_file() will be called with @max_sectors = BTRFS_DEFRAG_BATCH (1024) to limit how many sectors we can defrag in one try.
And then use the number of sectors defragged to determine if we need to re-defrag.
But commit b18c3ab2343d ("btrfs: defrag: introduce helper to defrag one cluster") uses wrong unit to increase @sectors_defragged, which should be in unit of sector, not byte.
This means, if we have defragged any sector, then @sectors_defragged will be >= sectorsize (normally 4096), which is larger than BTRFS_DEFRAG_BATCH.
This makes the @max_sectors check in defrag_one_cluster() to underflow, rendering the whole @max_sectors check useless.
Thus causing way more IO for autodefrag mount options, as now there is no limit on how many sectors can really be defragged.
[FIX]
Fix the problems by:
- Use sector as unit when increaseing @sectors_defragged
- Include @sectors_defragged > @max_sectors case to break the loop
- Add extra comment on the return value of btrfs_defrag_file()
---
Changelog:
v2:
- Update the commit message to include the root cause of extra IO
- Keep @sectors_defragged update where it is
Since the over-reported @sectors_defragged is not the real reason, keep the patch smaller.
--- fs/btrfs/ioctl.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
fs/btrfs/ioctl.c +7 -3
@@ -1442,8 +1442,8 @@ static int defrag_one_cluster(struct btrfs_inode *inode,list_for_each_entry (entry, &target_list, list) {u32 range_len = entry->len; -/* Reached the limit */ -if (max_sectors && max_sectors == *sectors_defragged) +/* Reached or beyond the limit */ +if (max_sectors && *sectors_defragged >= max_sectors)break ;if (max_sectors)
@@ -1465,7 +1465,8 @@ static int defrag_one_cluster(struct btrfs_inode *inode, extent_thresh, newer_than, do_compress);if (ret <0 )break ; - *sectors_defragged += range_len; + *sectors_defragged += range_len >> + inode->root->fs_info->sectorsize_bits; } out:list_for_each_entry_safe (entry, tmp, &target_list, list) {
@@ -1484,6 +1485,9 @@ static int defrag_one_cluster(struct btrfs_inode *inode,* @newer_than: minimum transid to defrag * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode * will be defragged. +* +* Return <0 for error. +* Return >=0 for the number of sectors defragged. */ int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra, struct btrfs_ioctl_defrag_range_args *range,
Signature · 1 line
2.34.1
Quoted 41 lines · Qu Wenruo
[BUG]
There are users using autodefrag mount option reporting obvious increase in IO:Quoted 9 lines
If I compare the write average (in total, I don't have it per process) when taking idle periods on the same machine:
Linux 5.16:
without autodefrag: ~ 10KiB/s
with autodefrag: between 1 and 2MiB/s.Linux 5.15:
with autodefrag:~ 10KiB/s (around the same as without
autodefrag on 5.16)[CAUSE]
When autodefrag mount option is enabled,btrfs_defrag_file()will be called with @max_sectors =BTRFS_DEFRAG_BATCH(1024) to limit how many sectors we can defrag in one try.And then use the number of sectors defragged to determine if we need to re-defrag.
But commit
b18c3ab2343d("btrfs: defrag: introduce helper to defrag one cluster") uses wrong unit to increase @sectors_defragged, which should be in unit of sector, not byte.This means, if we have defragged any sector, then @sectors_defragged will be >= sectorsize (normally 4096), which is larger than
BTRFS_DEFRAG_BATCH.This makes the @max_sectors check in
defrag_one_cluster()to underflow, rendering the whole @max_sectors check useless.Thus causing way more IO for autodefrag mount options, as now there is no limit on how many sectors can really be defragged.
[FIX]
Fix the problems by:- Use sector as unit when increaseing @sectors_defragged
Quoted 7 lines
- Include @sectors_defragged > @max_sectors case to break the loop
- Add extra comment on the return value of
btrfs_defrag_file()Reported-by Anthony Ruhier <[email protected]>Fixes b18c3ab2343d ("btrfs: defrag: introduce helper to defrag one cluster")Signed-off-by Qu Wenruo <[email protected]>
Looks good, thanks.
Please, in the future also add a link to the thread reporting the issue. This makes it much easier to find more details, instead of grepping in a mailbox or lore for the reporter's name...
Quoted 50 lines
---
Changelog:
v2:
- Update the commit message to include the root cause of extra IO- Keep @sectors_defragged update where it is
Since the over-reported @sectors_defragged is not the real reason, keep the patch smaller.--- fs/btrfs/ioctl.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)fs/btrfs/ioctl.c +7
-3@@ -1442,8 +1442,8 @@ static int defrag_one_cluster(struct btrfs_inode *inode,list_for_each_entry (entry, &target_list, list) {u32 range_len = entry->len; -/* Reached the limit */ -if (max_sectors && max_sectors == *sectors_defragged) +/* Reached or beyond the limit */ +if (max_sectors && *sectors_defragged >= max_sectors)break ;if (max_sectors)@@ -1465,7 +1465,8 @@ static int defrag_one_cluster(struct btrfs_inode *inode, extent_thresh, newer_than, do_compress);if (ret <0 )break ; - *sectors_defragged += range_len; + *sectors_defragged += range_len >> + inode->root->fs_info->sectorsize_bits; } out:list_for_each_entry_safe (entry, tmp, &target_list, list) {@@ -1484,6 +1485,9 @@ static int defrag_one_cluster(struct btrfs_inode *inode,* @newer_than: minimum transid to defrag * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode * will be defragged. +* +* Return <0 for error. +* Return >=0 for the number of sectors defragged. */ int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra, struct btrfs_ioctl_defrag_range_args *range,Signature · 1 line
2.34.1
Just to give you an update, I tested the 3 patches, and this time it seems to completely fix the issue.
Thanks a lot, Filipe and Qu!
Anthony
Le 18/01/2022 à 15:33, Filipe Manana a écrit :
Quoted 108 lines · Qu Wenruo
Quoted 40 lines
[BUG]
There are users using autodefrag mount option reporting obvious increase in IO:Quoted 9 lines
If I compare the write average (in total, I don't have it per process) when taking idle periods on the same machine:
Linux 5.16:
without autodefrag: ~ 10KiB/s
with autodefrag: between 1 and 2MiB/s.Linux 5.15:
with autodefrag:~ 10KiB/s (around the same as without
autodefrag on 5.16)[CAUSE]
When autodefrag mount option is enabled,btrfs_defrag_file()will be called with @max_sectors =BTRFS_DEFRAG_BATCH(1024) to limit how many sectors we can defrag in one try.And then use the number of sectors defragged to determine if we need to re-defrag.
But commit
b18c3ab2343d("btrfs: defrag: introduce helper to defrag one cluster") uses wrong unit to increase @sectors_defragged, which should be in unit of sector, not byte.This means, if we have defragged any sector, then @sectors_defragged will be >= sectorsize (normally 4096), which is larger than
BTRFS_DEFRAG_BATCH.This makes the @max_sectors check in
defrag_one_cluster()to underflow, rendering the whole @max_sectors check useless.Thus causing way more IO for autodefrag mount options, as now there is no limit on how many sectors can really be defragged.
[FIX]
Fix the problems by:- Use sector as unit when increaseing @sectors_defragged
Quoted 7 lines
- Include @sectors_defragged > @max_sectors case to break the loop
- Add extra comment on the return value of
btrfs_defrag_file()Reported-by Anthony Ruhier <[email protected]>Fixes b18c3ab2343d ("btrfs: defrag: introduce helper to defrag one cluster")Signed-off-by Qu Wenruo <[email protected]>Looks good, thanks.
Reviewed-by Filipe Manana <[email protected]>Please, in the future also add a link to the thread reporting the issue. This makes it much easier to find more details, instead of grepping in a mailbox or lore for the reporter's name...
Quoted 50 lines
---
Changelog:
v2:
- Update the commit message to include the root cause of extra IO- Keep @sectors_defragged update where it is
Since the over-reported @sectors_defragged is not the real reason, keep the patch smaller.--- fs/btrfs/ioctl.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)fs/btrfs/ioctl.c +7
-3@@ -1442,8 +1442,8 @@ static int defrag_one_cluster(struct btrfs_inode *inode,list_for_each_entry (entry, &target_list, list) {u32 range_len = entry->len; -/* Reached the limit */ -if (max_sectors && max_sectors == *sectors_defragged) +/* Reached or beyond the limit */ +if (max_sectors && *sectors_defragged >= max_sectors)break ;if (max_sectors)@@ -1465,7 +1465,8 @@ static int defrag_one_cluster(struct btrfs_inode *inode, extent_thresh, newer_than, do_compress);if (ret <0 )break ; - *sectors_defragged += range_len; + *sectors_defragged += range_len >> + inode->root->fs_info->sectorsize_bits; } out:list_for_each_entry_safe (entry, tmp, &target_list, list) {@@ -1484,6 +1485,9 @@ static int defrag_one_cluster(struct btrfs_inode *inode,* @newer_than: minimum transid to defrag * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode * will be defragged. +* +* Return <0 for error. +* Return >=0 for the number of sectors defragged. */ int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra, struct btrfs_ioctl_defrag_range_args *range,Signature · 1 line
2.34.1
Quoted 41 lines · Qu Wenruo
[BUG]
There are users using autodefrag mount option reporting obvious increase in IO:Quoted 9 lines
If I compare the write average (in total, I don't have it per process) when taking idle periods on the same machine:
Linux 5.16:
without autodefrag: ~ 10KiB/s
with autodefrag: between 1 and 2MiB/s.Linux 5.15:
with autodefrag:~ 10KiB/s (around the same as without
autodefrag on 5.16)[CAUSE]
When autodefrag mount option is enabled,btrfs_defrag_file()will be called with @max_sectors =BTRFS_DEFRAG_BATCH(1024) to limit how many sectors we can defrag in one try.And then use the number of sectors defragged to determine if we need to re-defrag.
But commit
b18c3ab2343d("btrfs: defrag: introduce helper to defrag one cluster") uses wrong unit to increase @sectors_defragged, which should be in unit of sector, not byte.This means, if we have defragged any sector, then @sectors_defragged will be >= sectorsize (normally 4096), which is larger than
BTRFS_DEFRAG_BATCH.This makes the @max_sectors check in
defrag_one_cluster()to underflow, rendering the whole @max_sectors check useless.Thus causing way more IO for autodefrag mount options, as now there is no limit on how many sectors can really be defragged.
[FIX]
Fix the problems by:- Use sector as unit when increaseing @sectors_defragged
Quoted 7 lines
- Include @sectors_defragged > @max_sectors case to break the loop
- Add extra comment on the return value of
btrfs_defrag_file()Reported-by Anthony Ruhier <[email protected]>Fixes b18c3ab2343d ("btrfs: defrag: introduce helper to defrag one cluster")Signed-off-by Qu Wenruo <[email protected]>
Added to misc-next, thanks.