Btrfs: fix unwritten extent buffers and hangs on future writeback attempts
PATCH 1/1 Btrfs: fix unwritten extent buffers and hangs on future writeback attempts
From: Filipe Manana <[email protected]>
The lock_extent_buffer_io() returns 1 to the caller to tell it everything went fine and the callers needs to start writeback for the extent buffer (submit a bio, etc), 0 to tell the caller everything went fine but it does not need to start writeback for the extent buffer, and a negative value if some error happened.
When it's about to return 1 it tries to lock all pages, and if a try lock on a page fails, and we didn't flush any existing bio in our "epd", it calls flush_write_bio(epd) and overwrites the return value of 1 to 0 or an error. The page might have been locked elsewhere, not with the goal of starting writeback of the extent buffer, and even by some code other than btrfs, like page migration for example, so it does not mean the writeback of the extent buffer was already started by some other task, so returning a 0 tells the caller (btree_write_cache_pages()) to not start writeback for the extent buffer. Note that epd might currently have either no bio, so flush_write_bio() returns 0 (success) or it might have a bio for another extent buffer with a lower index (logical address).
Since we return 0 with the EXTENT_BUFFER_WRITEBACK bit set on the extent buffer and writeback is never started for the extent buffer, future attempts to writeback the extent buffer will hang forever waiting on that bit to be cleared, since it can only be cleared after writeback completes. Such hang is reported with a trace like the following:
[49887.347053] INFO: task btrfs-transacti:1752 blocked for more than 122 seconds. [49887.347059] Not tainted 5.2.13-gentoo #2 [49887.347060] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [49887.347062] btrfs-transacti D 0 1752 2 0x80004000 [49887.347064] Call Trace: [49887.347069] ? __schedule+0x265/0x830 [49887.347071] ? bit_wait+0x50/0x50 [49887.347072] ? bit_wait+0x50/0x50 [49887.347074] schedule+0x24/0x90 [49887.347075] io_schedule+0x3c/0x60 [49887.347077] bit_wait_io+0x8/0x50 [49887.347079] __wait_on_bit+0x6c/0x80 [49887.347081] ? __lock_release.isra.29+0x155/0x2d0 [49887.347083] out_of_line_wait_on_bit+0x7b/0x80 [49887.347084] ? var_wake_function+0x20/0x20 [49887.347087] lock_extent_buffer_for_io+0x28c/0x390 [49887.347089] btree_write_cache_pages+0x18e/0x340 [49887.347091] do_writepages+0x29/0xb0 [49887.347093] ? kmem_cache_free+0x132/0x160 [49887.347095] ? convert_extent_bit+0x544/0x680 [49887.347097] filemap_fdatawrite_range+0x70/0x90 [49887.347099] btrfs_write_marked_extents+0x53/0x120 [49887.347100] btrfs_write_and_wait_transaction.isra.4+0x38/0xa0 [49887.347102] btrfs_commit_transaction+0x6bb/0x990 [49887.347103] ? start_transaction+0x33e/0x500 [49887.347105] transaction_kthread+0x139/0x15c
So fix this by not overwriting the return value (ret) with the result from flush_write_bio(). We also need to clear the EXTENT_BUFFER_WRITEBACK bit in case flush_write_bio() returns an error, otherwise it will hang any future attempts to writeback the extent buffer.
This is a regression introduced in the 5.2 kernel.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=204377
--- fs/btrfs/extent_io.c | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-)
fs/btrfs/extent_io.c +14 -9
@@ -3628,6 +3628,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks *
@@ -3699,8 +3706,11 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *ebif (!trylock_page (p)) {if (!flush) { - ret =flush_write_bio (epd); -if (ret <0 ) { +int err; + + err =flush_write_bio (epd); +if (err <0 ) { + ret = err; failed_page_nr = i; goto err_unlock; }
@@ -3715,16 +3725,11 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb/* Unlock already locked pages */ for (i =0 ; i < failed_page_nr; i++)unlock_page (eb->pages[i]); +/* Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it. */ +end_extent_buffer_writeback (eb);return ret; } -static void end_extent_buffer_writeback (struct extent_buffer *eb) -{ -clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); -smp_mb__after_atomic (); -wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); -} -static void set_btree_ioerr (struct page *page) {struct extent_buffer *eb = (struct extent_buffer *)page->private;
Signature · 1 line
2.11.0
Quoted 53 lines · On 11 Sep 2019, at 15:55, [email protected] wrote:
From: Filipe Manana <[email protected]>
So fix this by not overwriting the return value (ret) with the result from
flush_write_bio(). We also need to clear theEXTENT_BUFFER_WRITEBACK
bit in caseflush_write_bio()returns an error, otherwise it will hang any future attempts to writeback the extent buffer.This is a regression introduced in the 5.2 kernel.
Fixes 2e3c25136adfb ("btrfs: extent_io: add proper error handling to
lock_extent_buffer_for_io()")Fixes f4340622e0226 ("btrfs: extent_io: Move the BUG_ON() in
flush_write_bio()one level up")Reported-by Zdenek Sojka <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/GpO.2yos.3WGDOLpx6t%[email protected]/T/#uReported-by Stefan Priebe - Profihost AG <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/[email protected]/T/#tReported-by Drazen Kacar <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/DB8PR03MB562876ECE2319B3E579590F799C80@DB8PR03MB5628.eurprd03.prod.outlook.com/
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=204377Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/extent_io.c | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-)fs/btrfs/extent_io.c +7
-0@@ -3628,6 +3628,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks *
fs/btrfs/extent_io.c +0
-0@@ -3699,8 +3706,11 @@ static noinline_for_stack intlock_extent_buffer_for_io(struct
extent_buffer*ebfs/btrfs/extent_io.c +0
-0if (!trylock_page (p)) {if (!flush) {
fs/btrfs/extent_io.c static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb · line 3716
Quoted 10 lines
fs/btrfs/extent_io.c +5
-2- ret =flush_write_bio (epd); -if (ret <0 ) { +int err; + + err =flush_write_bio (epd); +if (err <0 ) { + ret = err; failed_page_nr = i; goto err_unlock; }
Dennis (cc'd) has been trying a similar fix against this in production, but sending it was interrupted by plumbing conferences. I think he found that it needs to undo this as well:
percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
-eb->len,
fs_info->dirty_metadata_batch);With the IO errors, we should end up aborting the FS. This function also flips the the extent buffer written and dirty flags, and his patch resets them as well. Given that we're aborting anyway, it's not critical, but it's probably a good idea to fix things up in the goto err_unlock just to make future bugs less likely.
-chris
Quoted 76 lines · Chris Mason
Quoted 53 lines · On 11 Sep 2019, at 15:55, [email protected] wrote:
From: Filipe Manana <[email protected]>
So fix this by not overwriting the return value (ret) with the result from
flush_write_bio(). We also need to clear theEXTENT_BUFFER_WRITEBACK
bit in caseflush_write_bio()returns an error, otherwise it will hang any future attempts to writeback the extent buffer.This is a regression introduced in the 5.2 kernel.
Fixes 2e3c25136adfb ("btrfs: extent_io: add proper error handling to
lock_extent_buffer_for_io()")Fixes f4340622e0226 ("btrfs: extent_io: Move the BUG_ON() in
flush_write_bio()one level up")Reported-by Zdenek Sojka <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/GpO.2yos.3WGDOLpx6t%[email protected]/T/#uReported-by Stefan Priebe - Profihost AG <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/[email protected]/T/#tReported-by Drazen Kacar <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/DB8PR03MB562876ECE2319B3E579590F799C80@DB8PR03MB5628.eurprd03.prod.outlook.com/
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=204377Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/extent_io.c | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-)fs/btrfs/extent_io.c +7
-0@@ -3628,6 +3628,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks *
fs/btrfs/extent_io.c +0
-0@@ -3699,8 +3706,11 @@ static noinline_for_stack intlock_extent_buffer_for_io(struct
extent_buffer*ebif (!trylock_page(p)) { if (!flush) {fs/btrfs/extent_io.c static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb · line 3716
Quoted 10 lines
fs/btrfs/extent_io.c +5
-2- ret =flush_write_bio (epd); -if (ret <0 ) { +int err; + + err =flush_write_bio (epd); +if (err <0 ) { + ret = err; failed_page_nr = i; goto err_unlock; }Dennis (cc'd) has been trying a similar fix against this in production, but sending it was interrupted by plumbing conferences. I think he found that it needs to undo this as well:
percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len, fs_info->dirty_metadata_batch);
Quoted 5 lines
With the IO errors, we should end up aborting the FS. This function also flips the the extent buffer written and dirty flags, and his patch resets them as well. Given that we're aborting anyway, it's not critical, but it's probably a good idea to fix things up in the goto
err_unlockjust to make future bugs less likely.
Yes, I considered that at some point (undo everything done so far, under the locks, etc) and thought it was pointless as well because we abort.
But we may not abort - if we never start the writeback for an eb because we returned error from flush_write_bio(), we can leave btree_write_cache_pages() without noticing the error. Since writeback never started, and btree_write_cache_pages() didn't return the error, the transaction commit path may never get an error from filemap_fdatawrite_range,
and we can commit the transaction despite failure to start writeback for some extent buffer.
A problem that existed before that regression in 5.2 anyway. Sending it as separate.
I'll include the undo of all operations in patch however, it doesn't hurt for sure.
Thanks.
Quoted 2 lines
-chris
Quoted 96 lines · Filipe Manana
Quoted 76 lines · Chris Mason
Quoted 53 lines · On 11 Sep 2019, at 15:55, [email protected] wrote:
From: Filipe Manana <[email protected]>
So fix this by not overwriting the return value (ret) with the result from
flush_write_bio(). We also need to clear theEXTENT_BUFFER_WRITEBACK
bit in caseflush_write_bio()returns an error, otherwise it will hang any future attempts to writeback the extent buffer.This is a regression introduced in the 5.2 kernel.
Fixes 2e3c25136adfb ("btrfs: extent_io: add proper error handling to
lock_extent_buffer_for_io()")Fixes f4340622e0226 ("btrfs: extent_io: Move the BUG_ON() in
flush_write_bio()one level up")Reported-by Zdenek Sojka <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/GpO.2yos.3WGDOLpx6t%[email protected]/T/#uReported-by Stefan Priebe - Profihost AG <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/[email protected]/T/#tReported-by Drazen Kacar <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/DB8PR03MB562876ECE2319B3E579590F799C80@DB8PR03MB5628.eurprd03.prod.outlook.com/
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=204377Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/extent_io.c | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-)fs/btrfs/extent_io.c +7
-0@@ -3628,6 +3628,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks *
fs/btrfs/extent_io.c +0
-0@@ -3699,8 +3706,11 @@ static noinline_for_stack intlock_extent_buffer_for_io(struct
extent_buffer*ebif (!trylock_page(p)) { if (!flush) {fs/btrfs/extent_io.c
Quoted 10 lines
fs/btrfs/extent_io.c +5
-2- ret =flush_write_bio (epd); -if (ret <0 ) { +int err; + + err =flush_write_bio (epd); +if (err <0 ) { + ret = err; failed_page_nr = i; goto err_unlock; }Dennis (cc'd) has been trying a similar fix against this in production, but sending it was interrupted by plumbing conferences. I think he found that it needs to undo this as well:
percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len, fs_info->dirty_metadata_batch);Quoted 5 lines
With the IO errors, we should end up aborting the FS. This function also flips the the extent buffer written and dirty flags, and his patch resets them as well. Given that we're aborting anyway, it's not critical, but it's probably a good idea to fix things up in the goto
err_unlockjust to make future bugs less likely.Yes, I considered that at some point (undo everything done so far, under the locks, etc) and thought it was pointless as well because we abort.
But we may not abort - if we never start the writeback for an eb because we returned error from
flush_write_bio(), we can leavebtree_write_cache_pages()without noticing the error. Since writeback never started, andbtree_write_cache_pages()didn't return the error, the transaction commit path may never get an error fromfilemap_fdatawrite_range,
and we can commit the transaction despite failure to start writeback for some extent buffer.
Quoted 10 lines
A problem that existed before that regression in 5.2 anyway. Sending it as separate.
I'll include the undo of all operations in patch however, it doesn't hurt for sure.
Thanks.
Quoted 2 lines
-chris
Hello,
I should have pushed this upstream sooner, I was hoping to have one of my test hosts hit my WARN_ON() in a separate patch, but it hasn't.
The following is what I have to unblock 5.2 testing.
I think your patch is missing resetting the header bits + the percpu metadata counter. I think on error we should break out of btree_write_cache_pages() and return it too.
Thanks,
Dennis
-----
From 1a57b5ee6e52c63bf7c8e3ae969c0df406e3cf69 Mon Sep 17 00:00:00 2001
From: Dennis Zhou <[email protected]>
Date: Wed, 4 Sep 2019 10:49:53 -0700
Subject: [PATCH] btrfs: fix stall on writeback bit extent buffer
In lock_extent_buffer_for_io(), if we encounter a blocking action, we try and flush the currently held onto bio. The failure mode here used to be a BUG_ON(). f4340622e022 changed this to move BUG_ON() up and incorrectly reset the current ret code. However, lock_extent_buffer_for_io() returns 1 on we should write out the pages. This caused the buffer to be skipped while keeping the writeback bit set.
Now that we can fail here, we also need to fix up dirty_metadata_bytes, clear BTRFS_HEADER_FLAG_WRITTEN and EXTENT_BUFFER_WRITEBACK, and set EXTENT_BUFFER_DIRTY again.
--- fs/btrfs/extent_io.c | 52 ++++++++++++++++++++++++++++++++++++-------- 1 file changed, 43 insertions(+), 9 deletions(-)
fs/btrfs/extent_io.c +43 -9
@@ -3636,6 +3636,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks *
@@ -3707,9 +3714,11 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *ebif (!trylock_page (p)) {if (!flush) { - ret =flush_write_bio (epd); -if (ret <0 ) { +int flush_ret =flush_write_bio (epd); + +if (flush_ret <0 ) { failed_page_nr = i; + ret = flush_ret; goto err_unlock; } flush =1 ;
@@ -3723,24 +3732,45 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb/* Unlock already locked pages */ for (i =0 ; i < failed_page_nr; i++)unlock_page (eb->pages[i]); -return ret; -} -static void end_extent_buffer_writeback (struct extent_buffer *eb) -{ -clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); -smp_mb__after_atomic (); -wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +/* undo the above above because we failed */ +btrfs_tree_lock (eb); + +percpu_counter_add_batch (&fs_info->dirty_metadata_bytes, + eb->len, + fs_info->dirty_metadata_batch); + +btrfs_clear_header_flag (eb,BTRFS_HEADER_FLAG_WRITTEN ); + +spin_lock (&eb->refs_lock); +set_bit (EXTENT_BUFFER_DIRTY , &eb->bflags); +spin_unlock (&eb->refs_lock); + +btrfs_tree_unlock (eb); + +end_extent_buffer_writeback (eb); + +return ret; }static void set_btree_ioerr (struct page *page) {struct extent_buffer *eb = (struct extent_buffer *)page->private; +struct btrfs_fs_info *fs_info;SetPageError (page);if (test_and_set_bit (EXTENT_BUFFER_WRITE_ERR , &eb->bflags))return ; +/* +* We just marked the extent as bad, that means we need retry +* in the future, so fix up the dirty_metadata_bytes accounting. +*/ + fs_info = eb->fs_info; +percpu_counter_add_batch (&fs_info->dirty_metadata_bytes, + eb->len, + fs_info->dirty_metadata_batch); +/* * If writeback for a btree extent that doesn't belong to a log tree * failed, increment the counter transaction->eb_write_errors.
@@ -3977,6 +4007,10 @@ int btree_write_cache_pages(struct address_space *mapping,if (!ret) {free_extent_buffer (eb);continue ; + }else if (ret <0 ) { + done =1 ; +free_extent_buffer (eb); +break ; } ret =write_one_eb (eb, wbc, &epd);
Signature · 1 line
2.17.1
Quoted 111 lines · Dennis Zhou
Quoted 96 lines · Filipe Manana
Quoted 76 lines · Chris Mason
Quoted 53 lines · On 11 Sep 2019, at 15:55, [email protected] wrote:
From: Filipe Manana <[email protected]>
So fix this by not overwriting the return value (ret) with the result from
flush_write_bio(). We also need to clear theEXTENT_BUFFER_WRITEBACK
bit in caseflush_write_bio()returns an error, otherwise it will hang any future attempts to writeback the extent buffer.This is a regression introduced in the 5.2 kernel.
Fixes 2e3c25136adfb ("btrfs: extent_io: add proper error handling to
lock_extent_buffer_for_io()")Fixes f4340622e0226 ("btrfs: extent_io: Move the BUG_ON() in
flush_write_bio()one level up")Reported-by Zdenek Sojka <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/GpO.2yos.3WGDOLpx6t%[email protected]/T/#uReported-by Stefan Priebe - Profihost AG <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/[email protected]/T/#tReported-by Drazen Kacar <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/DB8PR03MB562876ECE2319B3E579590F799C80@DB8PR03MB5628.eurprd03.prod.outlook.com/
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=204377Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/extent_io.c | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-)fs/btrfs/extent_io.c +7
-0@@ -3628,6 +3628,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks *
fs/btrfs/extent_io.c +0
-0@@ -3699,8 +3706,11 @@ static noinline_for_stack intlock_extent_buffer_for_io(struct
extent_buffer*ebif (!trylock_page(p)) { if (!flush) {fs/btrfs/extent_io.c static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb · line 3723
Quoted 10 lines
fs/btrfs/extent_io.c +5
-2- ret =flush_write_bio (epd); -if (ret <0 ) { +int err; + + err =flush_write_bio (epd); +if (err <0 ) { + ret = err; failed_page_nr = i; goto err_unlock; }Dennis (cc'd) has been trying a similar fix against this in production, but sending it was interrupted by plumbing conferences. I think he found that it needs to undo this as well:
percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len, fs_info->dirty_metadata_batch);Quoted 5 lines
With the IO errors, we should end up aborting the FS. This function also flips the the extent buffer written and dirty flags, and his patch resets them as well. Given that we're aborting anyway, it's not critical, but it's probably a good idea to fix things up in the goto
err_unlockjust to make future bugs less likely.Yes, I considered that at some point (undo everything done so far, under the locks, etc) and thought it was pointless as well because we abort.
But we may not abort - if we never start the writeback for an eb because we returned error from
flush_write_bio(), we can leavebtree_write_cache_pages()without noticing the error. Since writeback never started, andbtree_write_cache_pages()didn't return the error, the transaction commit path may never get an error fromfilemap_fdatawrite_range,
and we can commit the transaction despite failure to start writeback for some extent buffer.Quoted 10 lines
A problem that existed before that regression in 5.2 anyway. Sending it as separate.
I'll include the undo of all operations in patch however, it doesn't hurt for sure.
Thanks.
Quoted 2 lines
-chris
Hello,
Quoted 7 lines
I should have pushed this upstream sooner, I was hoping to have one of my test hosts hit my
WARN_ON()in a separate patch, but it hasn't.The following is what I have to unblock 5.2 testing.
I think your patch is missing resetting the header bits + the percpu metadata counter.
Yes, even though it's not a problem since we will end up aborting the transaction later.
The v2 I sent some minutes ago does it:
https://patchwork.kernel.org/patch/11141559/
Quoted 2 lines
I think on error we should break out of
btree_write_cache_pages()and return it too.
Yes, but that's a separate change.
It makes the code clear but it doesn't fix any problem, the errors will be marked in the end io callback and transaction and log commits will see them and abort.
Sent a patch for that as well some minutes ago:
https://patchwork.kernel.org/patch/11141561/
Quoted 74 lines
Thanks,
Dennis-----
From1a57b5ee6e52c63bf7c8e3ae969c0df406e3cf69Mon Sep 17 00:00:00 2001
From: Dennis Zhou <[email protected]>
Date: Wed, 4 Sep 2019 10:49:53 -0700
Subject: [PATCH] btrfs: fix stall on writeback bit extent bufferIn
lock_extent_buffer_for_io(), if we encounter a blocking action, we try and flush the currently held onto bio. The failure mode here used to be aBUG_ON().f4340622e022changed this to moveBUG_ON()up and incorrectly reset the current ret code. However,lock_extent_buffer_for_io()returns 1 on we should write out the pages. This caused the buffer to be skipped while keeping the writeback bit set.Now that we can fail here, we also need to fix up
dirty_metadata_bytes, clearBTRFS_HEADER_FLAG_WRITTENandEXTENT_BUFFER_WRITEBACK, and setEXTENT_BUFFER_DIRTYagain.Fixes f4340622e022 ("btrfs: extent_io: Move the BUG_ON() in flush_write_bio() one level up")Signed-off-by Dennis Zhou <[email protected]>--- fs/btrfs/extent_io.c | 52 ++++++++++++++++++++++++++++++++++++-------- 1 file changed, 43 insertions(+), 9 deletions(-)fs/btrfs/extent_io.c +13
-9@@ -3636,6 +3636,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks * @@ -3707,9 +3714,11 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *ebif (!trylock_page (p)) {if (!flush) { - ret =flush_write_bio (epd); -if (ret <0 ) { +int flush_ret =flush_write_bio (epd); + +if (flush_ret <0 ) { failed_page_nr = i; + ret = flush_ret; goto err_unlock; } flush =1 ;@@ -3723,24 +3732,45 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb/* Unlock already locked pages */ for (i =0 ; i < failed_page_nr; i++)unlock_page (eb->pages[i]); -return ret; -} -static void end_extent_buffer_writeback (struct extent_buffer *eb) -{ -clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); -smp_mb__after_atomic (); -wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +/* undo the above above because we failed */ +btrfs_tree_lock (eb);
fs/btrfs/extent_io.c static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb · line 3747
Quoted 10 lines
fs/btrfs/extent_io.c +10
-0+ +percpu_counter_add_batch (&fs_info->dirty_metadata_bytes, + eb->len, + fs_info->dirty_metadata_batch); + +btrfs_clear_header_flag (eb,BTRFS_HEADER_FLAG_WRITTEN ); + +spin_lock (&eb->refs_lock); +set_bit (EXTENT_BUFFER_DIRTY , &eb->bflags); +spin_unlock (&eb->refs_lock);
Clearing the writeback bit should also be done while holding eb->refs_lock. Everything else is equivalent to what I sent in v2.
Quoted 16 lines
fs/btrfs/extent_io.c +7
-0+ +btrfs_tree_unlock (eb); + +end_extent_buffer_writeback (eb); + +return ret; }static void set_btree_ioerr (struct page *page) {struct extent_buffer *eb = (struct extent_buffer *)page->private; +struct btrfs_fs_info *fs_info;SetPageError (page);if (test_and_set_bit (EXTENT_BUFFER_WRITE_ERR , &eb->bflags))return ;
fs/btrfs/extent_io.c static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb · line 3773
Quoted 10 lines
fs/btrfs/extent_io.c +9
-0+/* +* We just marked the extent as bad, that means we need retry +* in the future, so fix up the dirty_metadata_bytes accounting. +*/ + fs_info = eb->fs_info; +percpu_counter_add_batch (&fs_info->dirty_metadata_bytes, + eb->len, + fs_info->dirty_metadata_batch); +
This is a separate change from the 5.2 regression. Should be separate patch with a specific changelog IMHO.
Can you please submit that?
fs/btrfs/extent_io.c int btree_write_cache_pages(struct address_space *mapping, · line 4014
Quoted 12 lines
fs/btrfs/extent_io.c +4
-0/* * If writeback for a btree extent that doesn't belong to a log tree * failed, increment the counter transaction->eb_write_errors. @@ -3977,6 +4007,10 @@ int btree_write_cache_pages(struct address_space *mapping,if (!ret) {free_extent_buffer (eb);continue ; + }else if (ret <0 ) { + done =1 ; +free_extent_buffer (eb); +break ; }
Separate change as well, not cause by the 5.2 regression. So separate patch with a specific changelog as well IMHO.
Anyway, this hunk is exactly like the separate patch I sent some minutes ago.
Thanks!
Quoted 5 lines
fs/btrfs/extent_io.c +0
-1ret =write_one_eb (eb, wbc, &epd); --2.17.1
Quoted 117 lines · Filipe Manana
Quoted 116 lines · Dennis Zhou
Quoted 96 lines · Filipe Manana
Quoted 76 lines · Chris Mason
Quoted 53 lines · On 11 Sep 2019, at 15:55, [email protected] wrote:
From: Filipe Manana <[email protected]>
So fix this by not overwriting the return value (ret) with the result from
flush_write_bio(). We also need to clear theEXTENT_BUFFER_WRITEBACK
bit in caseflush_write_bio()returns an error, otherwise it will hang any future attempts to writeback the extent buffer.This is a regression introduced in the 5.2 kernel.
Fixes 2e3c25136adfb ("btrfs: extent_io: add proper error handling to
lock_extent_buffer_for_io()")Fixes f4340622e0226 ("btrfs: extent_io: Move the BUG_ON() in
flush_write_bio()one level up")Reported-by Zdenek Sojka <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/GpO.2yos.3WGDOLpx6t%[email protected]/T/#uReported-by Stefan Priebe - Profihost AG <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/[email protected]/T/#tReported-by Drazen Kacar <[email protected]>Link:
https://lore.kernel.org/linux-btrfs/DB8PR03MB562876ECE2319B3E579590F799C80@DB8PR03MB5628.eurprd03.prod.outlook.com/
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=204377Signed-off-by Filipe Manana <[email protected]>--- fs/btrfs/extent_io.c | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-)fs/btrfs/extent_io.c +7
-0@@ -3628,6 +3628,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks *
fs/btrfs/extent_io.c +0
-0@@ -3699,8 +3706,11 @@ static noinline_for_stack intlock_extent_buffer_for_io(struct
extent_buffer*ebif (!trylock_page(p)) { if (!flush) {fs/btrfs/extent_io.c static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb · line 3723
Quoted 10 lines
fs/btrfs/extent_io.c +5
-2- ret =flush_write_bio (epd); -if (ret <0 ) { +int err; + + err =flush_write_bio (epd); +if (err <0 ) { + ret = err; failed_page_nr = i; goto err_unlock; }Dennis (cc'd) has been trying a similar fix against this in production, but sending it was interrupted by plumbing conferences. I think he found that it needs to undo this as well:
percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len, fs_info->dirty_metadata_batch);Quoted 5 lines
With the IO errors, we should end up aborting the FS. This function also flips the the extent buffer written and dirty flags, and his patch resets them as well. Given that we're aborting anyway, it's not critical, but it's probably a good idea to fix things up in the goto
err_unlockjust to make future bugs less likely.Yes, I considered that at some point (undo everything done so far, under the locks, etc) and thought it was pointless as well because we abort.
But we may not abort - if we never start the writeback for an eb because we returned error from
flush_write_bio(), we can leavebtree_write_cache_pages()without noticing the error. Since writeback never started, andbtree_write_cache_pages()didn't return the error, the transaction commit path may never get an error fromfilemap_fdatawrite_range,
and we can commit the transaction despite failure to start writeback for some extent buffer.Quoted 10 lines
A problem that existed before that regression in 5.2 anyway. Sending it as separate.
I'll include the undo of all operations in patch however, it doesn't hurt for sure.
Thanks.
Quoted 2 lines
-chris
Hello,
I should have pushed this upstream sooner, I was hoping to have one of my test hosts hit my
WARN_ON()in a separate patch, but it hasn't.The following is what I have to unblock 5.2 testing.
Quoted 8 lines
Quoted 2 lines
I think your patch is missing resetting the header bits + the percpu metadata counter.
Yes, even though it's not a problem since we will end up aborting the transaction later.
The v2 I sent some minutes ago does it:
https://patchwork.kernel.org/patch/11141559/
I saw it as I refreshed the page after I sent my email :).
Quoted 115 lines
Quoted 2 lines
I think on error we should break out of
btree_write_cache_pages()and return it too.Yes, but that's a separate change.
It makes the code clear but it doesn't fix any problem, the errors will be marked in the end io callback and transaction and log commits will see them and abort.
Sent a patch for that as well some minutes ago:
https://patchwork.kernel.org/patch/11141561/Quoted 74 lines
Thanks,
Dennis-----
From1a57b5ee6e52c63bf7c8e3ae969c0df406e3cf69Mon Sep 17 00:00:00 2001
From: Dennis Zhou <[email protected]>
Date: Wed, 4 Sep 2019 10:49:53 -0700
Subject: [PATCH] btrfs: fix stall on writeback bit extent bufferIn
lock_extent_buffer_for_io(), if we encounter a blocking action, we try and flush the currently held onto bio. The failure mode here used to be aBUG_ON().f4340622e022changed this to moveBUG_ON()up and incorrectly reset the current ret code. However,lock_extent_buffer_for_io()returns 1 on we should write out the pages. This caused the buffer to be skipped while keeping the writeback bit set.Now that we can fail here, we also need to fix up
dirty_metadata_bytes, clearBTRFS_HEADER_FLAG_WRITTENandEXTENT_BUFFER_WRITEBACK, and setEXTENT_BUFFER_DIRTYagain.Fixes f4340622e022 ("btrfs: extent_io: Move the BUG_ON() in flush_write_bio() one level up")Signed-off-by Dennis Zhou <[email protected]>--- fs/btrfs/extent_io.c | 52 ++++++++++++++++++++++++++++++++++++-------- 1 file changed, 43 insertions(+), 9 deletions(-)fs/btrfs/extent_io.c +13
-9@@ -3636,6 +3636,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks * @@ -3707,9 +3714,11 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *ebif (!trylock_page (p)) {if (!flush) { - ret =flush_write_bio (epd); -if (ret <0 ) { +int flush_ret =flush_write_bio (epd); + +if (flush_ret <0 ) { failed_page_nr = i; + ret = flush_ret; goto err_unlock; } flush =1 ;@@ -3723,24 +3732,45 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb/* Unlock already locked pages */ for (i =0 ; i < failed_page_nr; i++)unlock_page (eb->pages[i]); -return ret; -} -static void end_extent_buffer_writeback (struct extent_buffer *eb) -{ -clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); -smp_mb__after_atomic (); -wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +/* undo the above above because we failed */ +btrfs_tree_lock (eb);fs/btrfs/extent_io.c static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb · line 3747
Quoted 10 lines
fs/btrfs/extent_io.c +10
-0+ +percpu_counter_add_batch (&fs_info->dirty_metadata_bytes, + eb->len, + fs_info->dirty_metadata_batch); + +btrfs_clear_header_flag (eb,BTRFS_HEADER_FLAG_WRITTEN ); + +spin_lock (&eb->refs_lock); +set_bit (EXTENT_BUFFER_DIRTY , &eb->bflags); +spin_unlock (&eb->refs_lock);Clearing the writeback bit should also be done while holding
eb->refs_lock. Everything else is equivalent to what I sent in v2.Quoted 17 lines
fs/btrfs/extent_io.c +7
-0+ +btrfs_tree_unlock (eb); + +end_extent_buffer_writeback (eb); + +return ret; }static void set_btree_ioerr (struct page *page) {struct extent_buffer *eb = (struct extent_buffer *)page->private; +struct btrfs_fs_info *fs_info;SetPageError (page);if (test_and_set_bit (EXTENT_BUFFER_WRITE_ERR , &eb->bflags))return ;
Quoted 14 lines
fs/btrfs/extent_io.c static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb · line 3773
Quoted 9 lines
fs/btrfs/extent_io.c +9
-0+/* +* We just marked the extent as bad, that means we need retry +* in the future, so fix up the dirty_metadata_bytes accounting. +*/ + fs_info = eb->fs_info; +percpu_counter_add_batch (&fs_info->dirty_metadata_bytes, + eb->len, + fs_info->dirty_metadata_batch); +This is a separate change from the 5.2 regression. Should be separate patch with a specific changelog IMHO.
Can you please submit that?
Will do.
Quoted 17 lines
fs/btrfs/extent_io.c int btree_write_cache_pages(struct address_space *mapping, · line 4014
Quoted 12 lines
fs/btrfs/extent_io.c +4
-0/* * If writeback for a btree extent that doesn't belong to a log tree * failed, increment the counter transaction->eb_write_errors. @@ -3977,6 +4007,10 @@ int btree_write_cache_pages(struct address_space *mapping,if (!ret) {free_extent_buffer (eb);continue ; + }else if (ret <0 ) { + done =1 ; +free_extent_buffer (eb); +break ; }Separate change as well, not cause by the 5.2 regression. So separate patch with a specific changelog as well IMHO.
Anyway, this hunk is exactly like the separate patch I sent some minutes ago.
Quoted 7 lines
Thanks!
Quoted 5 lines
fs/btrfs/extent_io.c +0
-1ret =write_one_eb (eb, wbc, &epd); --2.17.1
Thanks,
Dennis
PATCH v2 1/1 Btrfs: fix unwritten extent buffers and hangs on future writeback attempts
In reply to [email protected]From: Filipe Manana <[email protected]>
The lock_extent_buffer_io() returns 1 to the caller to tell it everything went fine and the callers needs to start writeback for the extent buffer (submit a bio, etc), 0 to tell the caller everything went fine but it does not need to start writeback for the extent buffer, and a negative value if some error happened.
When it's about to return 1 it tries to lock all pages, and if a try lock on a page fails, and we didn't flush any existing bio in our "epd", it calls flush_write_bio(epd) and overwrites the return value of 1 to 0 or an error. The page might have been locked elsewhere, not with the goal of starting writeback of the extent buffer, and even by some code other than btrfs, like page migration for example, so it does not mean the writeback of the extent buffer was already started by some other task, so returning a 0 tells the caller (btree_write_cache_pages()) to not start writeback for the extent buffer. Note that epd might currently have either no bio, so flush_write_bio() returns 0 (success) or it might have a bio for another extent buffer with a lower index (logical address).
Since we return 0 with the EXTENT_BUFFER_WRITEBACK bit set on the extent buffer and writeback is never started for the extent buffer, future attempts to writeback the extent buffer will hang forever waiting on that bit to be cleared, since it can only be cleared after writeback completes. Such hang is reported with a trace like the following:
[49887.347053] INFO: task btrfs-transacti:1752 blocked for more than 122 seconds. [49887.347059] Not tainted 5.2.13-gentoo #2 [49887.347060] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [49887.347062] btrfs-transacti D 0 1752 2 0x80004000 [49887.347064] Call Trace: [49887.347069] ? __schedule+0x265/0x830 [49887.347071] ? bit_wait+0x50/0x50 [49887.347072] ? bit_wait+0x50/0x50 [49887.347074] schedule+0x24/0x90 [49887.347075] io_schedule+0x3c/0x60 [49887.347077] bit_wait_io+0x8/0x50 [49887.347079] __wait_on_bit+0x6c/0x80 [49887.347081] ? __lock_release.isra.29+0x155/0x2d0 [49887.347083] out_of_line_wait_on_bit+0x7b/0x80 [49887.347084] ? var_wake_function+0x20/0x20 [49887.347087] lock_extent_buffer_for_io+0x28c/0x390 [49887.347089] btree_write_cache_pages+0x18e/0x340 [49887.347091] do_writepages+0x29/0xb0 [49887.347093] ? kmem_cache_free+0x132/0x160 [49887.347095] ? convert_extent_bit+0x544/0x680 [49887.347097] filemap_fdatawrite_range+0x70/0x90 [49887.347099] btrfs_write_marked_extents+0x53/0x120 [49887.347100] btrfs_write_and_wait_transaction.isra.4+0x38/0xa0 [49887.347102] btrfs_commit_transaction+0x6bb/0x990 [49887.347103] ? start_transaction+0x33e/0x500 [49887.347105] transaction_kthread+0x139/0x15c
So fix this by not overwriting the return value (ret) with the result from flush_write_bio(). We also need to clear the EXTENT_BUFFER_WRITEBACK bit in case flush_write_bio() returns an error, otherwise it will hang any future attempts to writeback the extent buffer, and undo all work done before (set back EXTENT_BUFFER_DIRTY, etc).
This is a regression introduced in the 5.2 kernel.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=204377
---
V2: Also set back EXTENT_BUFFER_DIRTY on error and undo all work done previously.
fs/btrfs/extent_io.c | 35 ++++++++++++++++++++++++++--------- 1 file changed, 26 insertions(+), 9 deletions(-)
fs/btrfs/extent_io.c +26 -9
@@ -3628,6 +3628,13 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb) TASK_UNINTERRUPTIBLE); } +static void end_extent_buffer_writeback (struct extent_buffer *eb) +{ +clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); +smp_mb__after_atomic (); +wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); +} +/* * Lock eb pages and flush the bio if we can't the locks *
@@ -3699,8 +3706,11 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *ebif (!trylock_page (p)) {if (!flush) { - ret =flush_write_bio (epd); -if (ret <0 ) { +int err; + + err =flush_write_bio (epd); +if (err <0 ) { + ret = err; failed_page_nr = i; goto err_unlock; }
@@ -3715,16 +3725,23 @@ static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb/* Unlock already locked pages */ for (i =0 ; i < failed_page_nr; i++)unlock_page (eb->pages[i]); +/* +* Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it. +* Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can +* be made and undo everything done before. +*/ +btrfs_tree_lock (eb); +spin_lock (&eb->refs_lock); +set_bit (EXTENT_BUFFER_DIRTY , &eb->bflags); +end_extent_buffer_writeback (eb); +spin_unlock (&eb->refs_lock); +percpu_counter_add_batch (&fs_info->dirty_metadata_bytes, eb->len, + fs_info->dirty_metadata_batch); +btrfs_clear_header_flag (eb,BTRFS_HEADER_FLAG_WRITTEN ); +btrfs_tree_unlock (eb);return ret; } -static void end_extent_buffer_writeback (struct extent_buffer *eb) -{ -clear_bit (EXTENT_BUFFER_WRITEBACK , &eb->bflags); -smp_mb__after_atomic (); -wake_up_bit (&eb->bflags,EXTENT_BUFFER_WRITEBACK ); -} -static void set_btree_ioerr (struct page *page) {struct extent_buffer *eb = (struct extent_buffer *)page->private;
Signature · 1 line
2.11.0