btrfs: relocation: Output current relocation stage at btrfs_relocate_block_group()
PATCH 1/1 btrfs: relocation: Output current relocation stage at btrfs_relocate_block_group()
There are several reports of hanging relocation, populating the dmesg with things like:
BTRFS info (device dm-5): found 1 extents
The investigation is still on going, but will never hurt to output a little more info.
This patch will also output the current relocation stage, making that output something like:
BTRFS info (device dm-5): balance: start -d -m -s
BTRFS info (device dm-5): relocating block group 30408704 flags metadata|dup BTRFS info (device dm-5): found 2 extents at MOVE_DATA_EXTENT stage BTRFS info (device dm-5): relocating block group 22020096 flags system|dup BTRFS info (device dm-5): found 1 extents at MOVE_DATA_EXTENT stage BTRFS info (device dm-5): relocating block group 13631488 flags data BTRFS info (device dm-5): found 1 extents at MOVE_DATA_EXTENT stage BTRFS info (device dm-5): found 1 extents at UPDATE_DATA_PTRS stage BTRFS info (device dm-5): balance: ended with status: 0
The string "MOVE_DATA_EXTENT" and "UPDATE_DATA_PTRS" is mostly from the macro MOVE_DATA_EXTENTS and UPDATE_DATA_PTRS, but the 'S' from MOVE_DATA_EXTENTS is removed in the output string to make the alignment better.
This patch will not increase the number of lines, but with extra info for us to debug the reported problem.
(Although it's very likely the bug is sticking at UPDATE_DATA_PTRS stage, even without the patch)
--- fs/btrfs/relocation.c | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-)
fs/btrfs/relocation.c +14 -2
@@ -4291,6 +4291,15 @@ static void describe_relocation(struct btrfs_fs_info *fs_info, block_group->start, buf); } +static const char *stage_to_string (int stage) +{ +if (stage ==MOVE_DATA_EXTENTS ) +return "MOVE_DATA_EXTENT" ; +if (stage ==UPDATE_DATA_PTRS ) +return "UPDATE_DATA_PTRS" ; +return "UNKNOWN" ; +} +/* * function to relocate all extents in a block group. */
@@ -4365,12 +4374,15 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) rc->block_group->length);while (1 ) { +int finishes_stage; +mutex_lock (&fs_info->cleaner_mutex); ret =relocate_block_group (rc);mutex_unlock (&fs_info->cleaner_mutex);if (ret <0 ) err = ret; + finishes_stage = rc->stage;/* * We may have gotten ENOSPC after we already dirtied some * extents. If writeout happens while we're relocating a
@@ -4396,8 +4408,8 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)if (rc->extents_found ==0 )break ; -btrfs_info (fs_info,"found %llu extents" , rc->extents_found); - +btrfs_info (fs_info,"found %llu extents at %s stage" , + rc->extents_found,stage_to_string (finishes_stage)); }WARN_ON (rc->block_group->pinned >0 );
Signature · 1 line
2.24.0
PATCH 1/1 btrfs: relocation: Output current relocation stage at btrfs_relocate_block_group()
There are several reports of hanging relocation, populating the dmesg with things like:
BTRFS info (device dm-5): found 1 extents
The investigation is still on going, but will never hurt to output a little more info.
This patch will also output the current relocation stage, making that output something like:
BTRFS info (device dm-5): balance: start -d -m -s
BTRFS info (device dm-5): relocating block group 30408704 flags metadata|dup BTRFS info (device dm-5): found 2 extents at MOVE_DATA_EXTENTS stage BTRFS info (device dm-5): relocating block group 22020096 flags system|dup BTRFS info (device dm-5): found 1 extents at MOVE_DATA_EXTENTS stage BTRFS info (device dm-5): relocating block group 13631488 flags data BTRFS info (device dm-5): found 1 extents at MOVE_DATA_EXTENTS stage BTRFS info (device dm-5): found 1 extents at UPDATE_DATA_PTRS stage BTRFS info (device dm-5): balance: ended with status: 0
--- fs/btrfs/relocation.c | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-)
fs/btrfs/relocation.c +14 -2
@@ -4291,6 +4291,15 @@ static void describe_relocation(struct btrfs_fs_info *fs_info, block_group->start, buf); } +static const char *stage_to_string (int stage) +{ +if (stage ==MOVE_DATA_EXTENTS ) +return "MOVE_DATA_EXTENTS" ; +if (stage ==UPDATE_DATA_PTRS ) +return "UPDATE_DATA_PTRS" ; +return "UNKNOWN" ; +} +/* * function to relocate all extents in a block group. */
@@ -4365,12 +4374,15 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) rc->block_group->length);while (1 ) { +int finishes_stage; +mutex_lock (&fs_info->cleaner_mutex); ret =relocate_block_group (rc);mutex_unlock (&fs_info->cleaner_mutex);if (ret <0 ) err = ret; + finishes_stage = rc->stage;/* * We may have gotten ENOSPC after we already dirtied some * extents. If writeout happens while we're relocating a
@@ -4396,8 +4408,8 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)if (rc->extents_found ==0 )break ; -btrfs_info (fs_info,"found %llu extents" , rc->extents_found); - +btrfs_info (fs_info,"found %llu extents at %s stage" , + rc->extents_found,stage_to_string (finishes_stage)); }WARN_ON (rc->block_group->pinned >0 );
Signature · 1 line
2.23.0
Quoted 56 lines · Qu Wenruo
There are several reports of hanging relocation, populating the dmesg with things like:
BTRFS info (device dm-5): found 1 extentsThe investigation is still on going, but will never hurt to output a little more info.
This patch will also output the current relocation stage, making that output something like:
BTRFS info (device dm-5): balance: start -d -m -s
BTRFS info (device dm-5): relocating block group 30408704 flags metadata|dup BTRFS info (device dm-5): found 2 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): relocating block group 22020096 flags system|dup BTRFS info (device dm-5): found 1 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): relocating block group 13631488 flags data BTRFS info (device dm-5): found 1 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): found 1 extents atUPDATE_DATA_PTRSstage BTRFS info (device dm-5): balance: ended with status: 0The string "
MOVE_DATA_EXTENT" and "UPDATE_DATA_PTRS" is mostly from the macroMOVE_DATA_EXTENTSandUPDATE_DATA_PTRS, but the 'S' fromMOVE_DATA_EXTENTSis removed in the output string to make the alignment better.This patch will not increase the number of lines, but with extra info for us to debug the reported problem.
(Although it's very likely the bug is sticking atUPDATE_DATA_PTRSstage, even without the patch)Signed-off-by Qu Wenruo <[email protected]>--- fs/btrfs/relocation.c | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-)fs/btrfs/relocation.c +9
-0@@ -4291,6 +4291,15 @@ static void describe_relocation(struct btrfs_fs_info *fs_info, block_group->start, buf); } +static const char *stage_to_string (int stage) +{ +if (stage ==MOVE_DATA_EXTENTS ) +return "MOVE_DATA_EXTENT" ; +if (stage ==UPDATE_DATA_PTRS ) +return "UPDATE_DATA_PTRS" ; +return "UNKNOWN" ; +} +/* * function to relocate all extents in a block group. */
fs/btrfs/relocation.c int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) · line 4378
Quoted 5 lines
fs/btrfs/relocation.c +2
-0@@ -4365,12 +4374,15 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) rc->block_group->length);while (1 ) { +int finishes_stage; +
NIT: the rc::stage is an unsigned integer.
Quoted 22 lines
fs/btrfs/relocation.c +3
-2mutex_lock (&fs_info->cleaner_mutex); ret =relocate_block_group (rc);mutex_unlock (&fs_info->cleaner_mutex);if (ret <0 ) err = ret; + finishes_stage = rc->stage;/* * We may have gotten ENOSPC after we already dirtied some * extents. If writeout happens while we're relocating a @@ -4396,8 +4408,8 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)if (rc->extents_found ==0 )break ; -btrfs_info (fs_info,"found %llu extents" , rc->extents_found); - +btrfs_info (fs_info,"found %llu extents at %s stage" , + rc->extents_found,stage_to_string (finishes_stage)); }WARN_ON (rc->block_group->pinned >0 );
Quoted 65 lines · Su Yue
Quoted 62 lines · Qu Wenruo
There are several reports of hanging relocation, populating the dmesg with things like:
BTRFS info (device dm-5): found 1 extentsThe investigation is still on going, but will never hurt to output a little more info.
This patch will also output the current relocation stage, making that output something like:
BTRFS info (device dm-5): balance: start -d -m -s
BTRFS info (device dm-5): relocating block group 30408704 flags metadata|dup
BTRFS info (device dm-5): found 2 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): relocating block group 22020096 flags system|dup
BTRFS info (device dm-5): found 1 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): relocating block group 13631488 flags data BTRFS info (device dm-5): found 1 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): found 1 extents atUPDATE_DATA_PTRSstage BTRFS info (device dm-5): balance: ended with status: 0The string "
MOVE_DATA_EXTENT" and "UPDATE_DATA_PTRS" is mostly from the macroMOVE_DATA_EXTENTSandUPDATE_DATA_PTRS, but the 'S' fromMOVE_DATA_EXTENTSis removed in the output string to make the alignment better.This patch will not increase the number of lines, but with extra info for us to debug the reported problem.
(Although it's very likely the bug is sticking atUPDATE_DATA_PTRSstage, even without the patch)Signed-off-by Qu Wenruo <[email protected]>---
fs/btrfs/relocation.c| 16 ++++++++++++++--1 file changed, 14 insertions(+), 2 deletions(-)fs/btrfs/relocation.c +9
-0@@ -4291,6 +4291,15 @@ static void describe_relocation(structbtrfs_fs_info *fs_info, block_group->start, buf); } +static const char *stage_to_string (int stage) +{ + if (stage == MOVE_DATA_EXTENTS) + return "MOVE_DATA_EXTENT"; + if (stage == UPDATE_DATA_PTRS) + return "UPDATE_DATA_PTRS"; + return "UNKNOWN"; +} + /** function to relocate all extents in a block group. */
fs/btrfs/relocation.c +0
-0@@ -4365,12 +4374,15 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) rc->block_group->length);
Quoted 5 lines
fs/btrfs/relocation.c int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) · line 4378
Quoted 3 lines
fs/btrfs/relocation.c +2
-0while (1) { + int finishes_stage; +NIT: the
rc::stageis an unsigned integer.
It doesn't matter. rc::stage is only a 8bit value.
Both unsigned int/int can handle it without any problem.
And the strage_to_string() function can handle any value, so no problem at all.
Thanks,
Qu
Quoted 25 lines
Quoted 25 lines
fs/btrfs/relocation.c +1
-0mutex_lock(&fs_info->cleaner_mutex); ret = relocate_block_group(rc); mutex_unlock(&fs_info->cleaner_mutex); if (ret < 0) err = ret; + finishes_stage = rc->stage; /** We may have gotten ENOSPC after we already dirtied some * extents. If writeout happens while we're relocating a
fs/btrfs/relocation.c +2
-2@@ -4396,8 +4408,8 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) if (rc->extents_found == 0) break; - btrfs_info(fs_info, "found %llu extents", rc->extents_found); - + btrfs_info(fs_info, "found %llu extents at %s stage", + rc->extents_found, stage_to_string(finishes_stage)); }
WARN_ON(rc->block_group->pinned > 0);
Quoted 20 lines · Qu Wenruo
There are several reports of hanging relocation, populating the dmesg with things like:
BTRFS info (device dm-5): found 1 extentsThe investigation is still on going, but will never hurt to output a little more info.
This patch will also output the current relocation stage, making that output something like:
BTRFS info (device dm-5): balance: start -d -m -s
BTRFS info (device dm-5): relocating block group 30408704 flags metadata|dup BTRFS info (device dm-5): found 2 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): relocating block group 22020096 flags system|dup BTRFS info (device dm-5): found 1 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): relocating block group 13631488 flags data BTRFS info (device dm-5): found 1 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): found 1 extents atUPDATE_DATA_PTRSstage BTRFS info (device dm-5): balance: ended with status: 0
Quoted 7 lines
The string "
MOVE_DATA_EXTENT" and "UPDATE_DATA_PTRS" is mostly from the macroMOVE_DATA_EXTENTSandUPDATE_DATA_PTRS, but the 'S' fromMOVE_DATA_EXTENTSis removed in the output string to make the alignment better.This patch will not increase the number of lines, but with extra info for us to debug the reported problem.
Nice. I'd suggest to make it more user friendly
relocation: found 111 extents, stage: move data blocks
relocation: found 111 extents, stage: update data pointers
The identifier can be understood what it means but it's IMHO not important to copy it to the message verbatim.
Quoted 26 lines · David Sterba
Quoted 25 lines · Qu Wenruo
There are several reports of hanging relocation, populating the dmesg with things like:
BTRFS info (device dm-5): found 1 extentsThe investigation is still on going, but will never hurt to output a little more info.
This patch will also output the current relocation stage, making that output something like:
BTRFS info (device dm-5): balance: start -d -m -s
BTRFS info (device dm-5): relocating block group 30408704 flags metadata|dup BTRFS info (device dm-5): found 2 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): relocating block group 22020096 flags system|dup BTRFS info (device dm-5): found 1 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): relocating block group 13631488 flags data BTRFS info (device dm-5): found 1 extents atMOVE_DATA_EXTENTstage BTRFS info (device dm-5): found 1 extents atUPDATE_DATA_PTRSstage BTRFS info (device dm-5): balance: ended with status: 0The string "
MOVE_DATA_EXTENT" and "UPDATE_DATA_PTRS" is mostly from the macroMOVE_DATA_EXTENTSandUPDATE_DATA_PTRS, but the 'S' fromMOVE_DATA_EXTENTSis removed in the output string to make the alignment better.
Quoted 7 lines
Quoted 2 lines
This patch will not increase the number of lines, but with extra info for us to debug the reported problem.
Nice. I'd suggest to make it more user friendly
relocation: found 111 extents, stage: move data blocks
relocation: found 111 extents, stage: update data pointers
This is much better, keeps the indent while provide better readability.
I'll go this way in the next version.
Thanks,
Qu
Quoted 4 lines
The identifier can be understood what it means but it's IMHO not important to copy it to the message verbatim.