btrfs: honor path->skip_locking in backref code
PATCH 1/1 btrfs: honor path->skip_locking in backref code
qgroups will do the old roots lookup at delayed ref time, which could be while walking down the extent root while running a delayed ref. This should be fine, except we specifically lock eb's in the backref walking code irrespective of path->skip_locking, which deadlocks the system. Fix up the backref code to honor path->skip_locking, nobody will be modifying the commit_root when we're searching so it's completely safe to do. Thanks,
--- fs/btrfs/backref.c | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-)
fs/btrfs/backref.c +10 -6
@@ -712,7 +712,7 @@ static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,* read tree blocks and add keys where required. */ static int add_missing_keys (struct btrfs_fs_info *fs_info, -struct preftrees *preftrees) +struct preftrees *preftrees,bool lock) {struct prelim_ref *ref;struct extent_buffer *eb;
@@ -737,12 +737,14 @@ static int add_missing_keys(struct btrfs_fs_info *fs_info,free_extent_buffer (eb);return -EIO ; } -btrfs_tree_read_lock (eb); +if (lock) +btrfs_tree_read_lock (eb);if (btrfs_header_level (eb) ==0 )btrfs_item_key_to_cpu (eb, &ref->key_for_search,0 );else btrfs_node_key_to_cpu (eb, &ref->key_for_search,0 ); -btrfs_tree_read_unlock (eb); +if (lock) +btrfs_tree_read_unlock (eb);free_extent_buffer (eb);prelim_ref_insert (fs_info, &preftrees->indirect, ref,NULL );cond_resched ();
@@ -1227,7 +1229,7 @@ static int find_parent_nodes(struct btrfs_trans_handle *trans,btrfs_release_path (path); - ret =add_missing_keys (fs_info, &preftrees); + ret =add_missing_keys (fs_info, &preftrees, path->skip_locking ==0 );if (ret) goto out;
@@ -1288,11 +1290,13 @@ static int find_parent_nodes(struct btrfs_trans_handle *trans, ret = -EIO ; goto out; } -btrfs_tree_read_lock (eb); +if (!path->skip_locking) +btrfs_tree_read_lock (eb);btrfs_set_lock_blocking_rw (eb,BTRFS_READ_LOCK ); ret =find_extent_in_eb (eb, bytenr, *extent_item_pos, &eie, ignore_offset); -btrfs_tree_read_unlock_blocking (eb); +if (!path->skip_locking) +btrfs_tree_read_unlock_blocking (eb);free_extent_buffer (eb);if (ret <0 ) goto out;
Signature · 1 line
2.14.3
Quoted 9 lines · Josef Bacik
qgroups will do the old roots lookup at delayed ref time, which could be while walking down the extent root while running a delayed ref. This should be fine, except we specifically lock eb's in the backref walking code irrespective of
path->skip_locking, which deadlocks the system. Fix up the backref code to honorpath->skip_locking, nobody will be modifying thecommit_rootwhen we're searching so it's completely safe to do. Thanks,Signed-off-by Josef Bacik <[email protected]>
Thanks,
Qu
Quoted 60 lines
--- fs/btrfs/backref.c | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-)fs/btrfs/backref.c +10
-6@@ -712,7 +712,7 @@ static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,* read tree blocks and add keys where required. */ static int add_missing_keys (struct btrfs_fs_info *fs_info, -struct preftrees *preftrees) +struct preftrees *preftrees,bool lock) {struct prelim_ref *ref;struct extent_buffer *eb;@@ -737,12 +737,14 @@ static int add_missing_keys(struct btrfs_fs_info *fs_info,free_extent_buffer (eb);return -EIO ; } -btrfs_tree_read_lock (eb); +if (lock) +btrfs_tree_read_lock (eb);if (btrfs_header_level (eb) ==0 )btrfs_item_key_to_cpu (eb, &ref->key_for_search,0 );else btrfs_node_key_to_cpu (eb, &ref->key_for_search,0 ); -btrfs_tree_read_unlock (eb); +if (lock) +btrfs_tree_read_unlock (eb);free_extent_buffer (eb);prelim_ref_insert (fs_info, &preftrees->indirect, ref,NULL );cond_resched ();@@ -1227,7 +1229,7 @@ static int find_parent_nodes(struct btrfs_trans_handle *trans,btrfs_release_path (path); - ret =add_missing_keys (fs_info, &preftrees); + ret =add_missing_keys (fs_info, &preftrees, path->skip_locking ==0 );if (ret) goto out;@@ -1288,11 +1290,13 @@ static int find_parent_nodes(struct btrfs_trans_handle *trans, ret = -EIO ; goto out; } -btrfs_tree_read_lock (eb); +if (!path->skip_locking) +btrfs_tree_read_lock (eb);btrfs_set_lock_blocking_rw (eb,BTRFS_READ_LOCK ); ret =find_extent_in_eb (eb, bytenr, *extent_item_pos, &eie, ignore_offset); -btrfs_tree_read_unlock_blocking (eb); +if (!path->skip_locking) +btrfs_tree_read_unlock_blocking (eb);free_extent_buffer (eb);if (ret <0 ) goto out;
Quoted 11 lines · Qu Wenruo
Quoted 8 lines · Josef Bacik
qgroups will do the old roots lookup at delayed ref time, which could be while walking down the extent root while running a delayed ref. This should be fine, except we specifically lock eb's in the backref walking code irrespective of
path->skip_locking, which deadlocks the system. Fix up the backref code to honorpath->skip_locking, nobody will be modifying thecommit_rootwhen we're searching so it's completely safe to do. Thanks,
Quoted 3 lines
Quoted 1 line
Signed-off-by Josef Bacik <[email protected]>Reviewed-by Qu Wenruo <[email protected]>
What can we use from https://patchwork.kernel.org/patch/10725371/ to add to the changelog? Is the deadlock caused by fb235dc06fac? The fix applies to stable 4.19+ directly and to 4.14 with context fixups. Thanks.
Quoted 12 lines · David Sterba
Quoted 11 lines · Qu Wenruo
Quoted 8 lines · Josef Bacik
qgroups will do the old roots lookup at delayed ref time, which could be while walking down the extent root while running a delayed ref. This should be fine, except we specifically lock eb's in the backref walking code irrespective of
path->skip_locking, which deadlocks the system. Fix up the backref code to honorpath->skip_locking, nobody will be modifying thecommit_rootwhen we're searching so it's completely safe to do. Thanks,
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
Signed-off-by Josef Bacik <[email protected]>Reviewed-by Qu Wenruo <[email protected]>What can we use from https://patchwork.kernel.org/patch/10725371/ to add to the changelog? Is the deadlock caused by
fb235dc06fac?
The error and cause part should be the same.
Just Josef's is much better than my patch.
Only the [FIX] part needs some update.
Thanks
Qu
Quoted 4 lines
The fix
applies to stable 4.19+ directly and to 4.14 with context fixups. Thanks.
Quoted 15 lines · Qu Wenruo
Quoted 12 lines · David Sterba
Quoted 11 lines · Qu Wenruo
Quoted 8 lines · Josef Bacik
qgroups will do the old roots lookup at delayed ref time, which could be while walking down the extent root while running a delayed ref. This should be fine, except we specifically lock eb's in the backref walking code irrespective of
path->skip_locking, which deadlocks the system. Fix up the backref code to honorpath->skip_locking, nobody will be modifying thecommit_rootwhen we're searching so it's completely safe to do. Thanks,
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
Signed-off-by Josef Bacik <[email protected]>Reviewed-by Qu Wenruo <[email protected]>What can we use from https://patchwork.kernel.org/patch/10725371/ to add to the changelog? Is the deadlock caused by
fb235dc06fac?The error and cause part should be the same.
Ok, thanks I've copied the stacktrace and deadlock analysis and tags.
Quoted 9 lines · Josef Bacik
qgroups will do the old roots lookup at delayed ref time, which could be while walking down the extent root while running a delayed ref. This should be fine, except we specifically lock eb's in the backref walking code irrespective of
path->skip_locking, which deadlocks the system. Fix up the backref code to honorpath->skip_locking, nobody will be modifying thecommit_rootwhen we're searching so it's completely safe to do. Thanks,Signed-off-by Josef Bacik <[email protected]>
Overnight test of the untar/snap/snapdel/rm stress test is fine now, there's a lockdep warning in the logs that I think have seen already.
Quoted 9 lines · Josef Bacik
qgroups will do the old roots lookup at delayed ref time, which could be while walking down the extent root while running a delayed ref. This should be fine, except we specifically lock eb's in the backref walking code irrespective of
path->skip_locking, which deadlocks the system. Fix up the backref code to honorpath->skip_locking, nobody will be modifying thecommit_rootwhen we're searching so it's completely safe to do. Thanks,Signed-off-by Josef Bacik <[email protected]>
So this explodes at btrfs/007, I had the patch in a separate branch out of misc-next and for-next and did not notice that, the functional tests passed but fstests do not.
[ 90.693679] kernel BUG at fs/btrfs/locking.c:286! [ 90.694914] invalid opcode: 0000 [#1] PREEMPT SMP [ 90.696152] CPU: 2 PID: 22976 Comm: btrfs Not tainted 4.20.0-rc7-default+ #419 [ 90.698100] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.11.2-0-gf9626cc-prebuilt.qemu-project.org 04/01/2014 [ 90.700893] RIP: 0010:btrfs_assert_tree_read_locked.part.2+0x5/0x10 [btrfs] [ 90.706111] RSP: 0018:ffffa1b046447840 EFLAGS: 00010246 [ 90.707035] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000 [ 90.707987] RDX: 0000000069fc7000 RSI: 0000000000000002 RDI: ffff91c73784f480 [ 90.709962] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000001 [ 90.711894] R10: ffff91c73ffc6000 R11: 0000000001a7f1c0 R12: ffff91c723d8dc60 [ 90.713543] R13: ffff91c73784f480 R14: 0000000000000001 R15: ffff91c723d8daa8 [ 90.715023] FS: 00007fdedfe778c0(0000) GS:ffff91c73d800000(0000) knlGS:0000000000000000 [ 90.717170] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 90.718176] CR2: 00007fdedf58ae38 CR3: 0000000073d09000 CR4: 00000000000006e0 [ 90.719112] Call Trace: [ 90.720810] btrfs_set_lock_blocking_rw+0x9d/0xb0 [btrfs] [ 90.722187] find_parent_nodes+0xaf5/0xf30 [btrfs] [ 90.723102] ? __find_iref+0xc0/0xc0 [btrfs] [ 90.723798] iterate_extent_inodes+0xf6/0x320 [btrfs] [ 90.724574] ? btrfs_get_token_64+0x103/0x120 [btrfs] [ 90.725700] ? kvm_sched_clock_read+0x5/0x10 [ 90.726688] ? sched_clock+0x5/0x10 [ 90.727695] ? process_extent+0x503/0x1210 [btrfs] [ 90.728899] process_extent+0x503/0x1210 [btrfs] [ 90.729910] ? __record_deleted_ref+0x20/0x20 [btrfs] [ 90.731130] changed_cb+0xb9e/0x1060 [btrfs] [ 90.732105] send_subvol+0x4a0/0x580 [btrfs] [ 90.733142] btrfs_ioctl_send+0x7c4/0xb20 [btrfs] [ 90.734188] ? sched_clock_cpu+0xc/0xc0 [ 90.735076] ? _copy_from_user+0x63/0x90 [ 90.736004] _btrfs_ioctl_send+0xdd/0x110 [btrfs] [ 90.737077] ? __lock_is_held+0x5a/0xa0 [ 90.737962] btrfs_ioctl+0xcd3/0x2f10 [btrfs] [ 90.738931] ? __lock_acquire+0x263/0xf10 [ 90.740025] ? lockdep_hardirqs_on+0xe8/0x180 [ 90.741082] ? do_vfs_ioctl+0xa2/0x6d0 [ 90.742077] do_vfs_ioctl+0xa2/0x6d0 [ 90.743151] ? __fget+0x109/0x1e0 [ 90.744117] ksys_ioctl+0x3a/0x70 [ 90.744988] __x64_sys_ioctl+0x16/0x20 [ 90.746035] do_syscall_64+0x54/0x180 [ 90.746997] entry_SYSCALL_64_after_hwframe+0x49/0xbe [ 90.748090] RIP: 0033:0x7fdedf67daa7
Quoted 53 lines · Josef Bacik
qgroups will do the old roots lookup at delayed ref time, which could be while walking down the extent root while running a delayed ref. This should be fine, except we specifically lock eb's in the backref walking code irrespective of
path->skip_locking, which deadlocks the system. Fix up the backref code to honorpath->skip_locking, nobody will be modifying thecommit_rootwhen we're searching so it's completely safe to do. Thanks,Signed-off-by Josef Bacik <[email protected]>--- fs/btrfs/backref.c | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-)fs/btrfs/backref.c +6
-4@@ -712,7 +712,7 @@ static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,* read tree blocks and add keys where required. */ static int add_missing_keys (struct btrfs_fs_info *fs_info, -struct preftrees *preftrees) +struct preftrees *preftrees,bool lock) {struct prelim_ref *ref;struct extent_buffer *eb;@@ -737,12 +737,14 @@ static int add_missing_keys(struct btrfs_fs_info *fs_info,free_extent_buffer (eb);return -EIO ; } -btrfs_tree_read_lock (eb); +if (lock) +btrfs_tree_read_lock (eb);if (btrfs_header_level (eb) ==0 )btrfs_item_key_to_cpu (eb, &ref->key_for_search,0 );else btrfs_node_key_to_cpu (eb, &ref->key_for_search,0 ); -btrfs_tree_read_unlock (eb); +if (lock) +btrfs_tree_read_unlock (eb);free_extent_buffer (eb);prelim_ref_insert (fs_info, &preftrees->indirect, ref,NULL );cond_resched ();@@ -1227,7 +1229,7 @@ static int find_parent_nodes(struct btrfs_trans_handle *trans,btrfs_release_path (path); - ret =add_missing_keys (fs_info, &preftrees); + ret =add_missing_keys (fs_info, &preftrees, path->skip_locking ==0 );if (ret) goto out;
fs/btrfs/backref.c static int find_parent_nodes(struct btrfs_trans_handle *trans, · line 1295
Quoted 7 lines
fs/btrfs/backref.c +2
-1@@ -1288,11 +1290,13 @@ static int find_parent_nodes(struct btrfs_trans_handle *trans, ret = -EIO ; goto out; } -btrfs_tree_read_lock (eb); +if (!path->skip_locking) +btrfs_tree_read_lock (eb);btrfs_set_lock_blocking_rw (eb,BTRFS_READ_LOCK );
This btrfs_set_lock_blocking_rw() or the btrfs_set_lock_blocking_read() in latest misc-next call need @eb to be read locked first.
So this line should also be in the (!path->skip_locking) branch, and such modification solves the BUG_ON() caused by btrfs/007.
Thanks,
Qu
Quoted 9 lines
fs/btrfs/backref.c +2
-1ret =find_extent_in_eb (eb, bytenr, *extent_item_pos, &eie, ignore_offset); -btrfs_tree_read_unlock_blocking (eb); +if (!path->skip_locking) +btrfs_tree_read_unlock_blocking (eb);free_extent_buffer (eb);if (ret <0 ) goto out;
Quoted 12 lines · Qu Wenruo
fs/btrfs/backref.c static int find_parent_nodes(struct btrfs_trans_handle *trans, · line 1295
Quoted 8 lines
fs/btrfs/backref.c +2
-1@@ -1288,11 +1290,13 @@ static int find_parent_nodes(struct btrfs_trans_handle *trans, ret = -EIO ; goto out; } -btrfs_tree_read_lock (eb); +if (!path->skip_locking) +btrfs_tree_read_lock (eb);btrfs_set_lock_blocking_rw (eb,BTRFS_READ_LOCK );This
btrfs_set_lock_blocking_rw()or thebtrfs_set_lock_blocking_read()in latest misc-next call need @eb to be read locked first.
Quoted 2 lines
So this line should also be in the (!path->skip_locking) branch, and such modification solves the
BUG_ON()caused by btrfs/007.
Thanks. btrfs_set_lock_blocking_rw is gone from misc-next so the patch needs a refresh and resend, besides passing fstests of course.
Quoted 13 lines · David Sterba
Quoted 12 lines · Qu Wenruo
fs/btrfs/backref.c static int find_parent_nodes(struct btrfs_trans_handle *trans, · line 1295
Quoted 8 lines
fs/btrfs/backref.c +2
-1@@ -1288,11 +1290,13 @@ static int find_parent_nodes(struct btrfs_trans_handle *trans, ret = -EIO ; goto out; } -btrfs_tree_read_lock (eb); +if (!path->skip_locking) +btrfs_tree_read_lock (eb);btrfs_set_lock_blocking_rw (eb,BTRFS_READ_LOCK );This
btrfs_set_lock_blocking_rw()or thebtrfs_set_lock_blocking_read()in latest misc-next call need @eb to be read locked first.
Quoted 6 lines
Quoted 2 lines
So this line should also be in the (!path->skip_locking) branch, and such modification solves the
BUG_ON()caused by btrfs/007.Thanks.
btrfs_set_lock_blocking_rwis gone from misc-next so the patch needs a refresh and resend, besides passing fstests of course.
Yup, although the conflict is pretty small, just change the btrfs_set_lock_blocking_rw() to btrfs_set_lock_blocking_read() will do it.
Does Josef or me need to resend the patch or would you mind to fold the change?
Thanks,
Qu