Skip to contentSkip to conversation

btrfs: defrag: fix the wrong number of defragged sectors

PATCH In review Qu Wenruo 4 participants 4 messages Jan 18, 2022 – Jan 19, 2022 Btrfs file system

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

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()

Fixes b18c3ab2343d ("btrfs: defrag: introduce helper to defrag one cluster")

---
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
index 6ad2bc2e5af3..2a77273d91fe 100644
@@ -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
Reviewed-by on v2 1/1
#1 lore raw
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()

Fixes b18c3ab2343d ("btrfs: defrag: introduce helper to defrag one cluster")

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
index 6ad2bc2e5af3..2a77273d91fe 100644
@@ -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
#2 lore raw

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()

Fixes b18c3ab2343d ("btrfs: defrag: introduce helper to defrag one cluster")

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
index 6ad2bc2e5af3..2a77273d91fe 100644
@@ -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
#3 lore raw
In reply to Qu Wenruo
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()

Fixes b18c3ab2343d ("btrfs: defrag: introduce helper to defrag one cluster")

Added to misc-next, thanks.