[PATCH] Fix WARNING: suspicious RCU usage in bch2_bucket_ref_update
From: Pei Li
Date: Wed Jul 10 2024 - 11:33:37 EST
When dereference ca->bucket_gens, we are expecting one of the three
locks to be hold if ca->fs is not NULL.
This patch acquires mark_lock before entering __mark_pointer()
Note: testing is done by a robot and is best-effort only.
To: Kent Overstreet <kent.overstreet@xxxxxxxxx>
To: Brian Foster <bfoster@xxxxxxxxxx>
Cc: linux-bcachefs@xxxxxxxxxxxxxxx
Cc: linux-kernel@xxxxxxxxxxxxxxx
Cc: skhan@xxxxxxxxxxxxxxxxxxx
Cc: syzkaller-bugs@xxxxxxxxxxxxxxxx
Cc: linux-kernel-mentees@xxxxxxxxxxxxxxxxxxxxxxxxx
Reported-by: syzbot+e74fea078710bbca6f4b@xxxxxxxxxxxxxxxxxxxxxxxxx
Closes: https://syzkaller.appspot.com/bug?extid=e74fea078710bbca6f4b
Tested-by: syzbot+e74fea078710bbca6f4b@xxxxxxxxxxxxxxxxxxxxxxxxx
Signed-off-by: Pei Li <peili.dev@xxxxxxxxx>
---
Syzbot reported the following warning:
fs/bcachefs/buckets.h:107 suspicious rcu_dereference_check() usage!
When dereference ca->bucket_gens, we are expecting one of the three
locks to be hold if ca->fs is not NULL.
This patch acquires mark_lock before entering __mark_pointer()
Tested on:
commit: 34afb82a Merge tag '6.10-rc6-smb3-server-fixes' of git..
git tree: upstream
console output: https://syzkaller.appspot.com/x/log.txt?x=1492e87e980000
kernel config: https://syzkaller.appspot.com/x/.config?x=42a432cfd0e579e0
dashboard link: https://syzkaller.appspot.com/bug?extid=e74fea078710bbca6f4b
compiler: Debian clang version 15.0.6, GNU ld (GNU Binutils for Debian) 2.40
patch: https://syzkaller.appspot.com/x/patch.diff?x=15e2e87e980000
Note: testing is done by a robot and is best-effort only.
---
fs/bcachefs/buckets.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index 743d57eba760..03147e46dc47 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -1033,8 +1033,14 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
if (flags & BTREE_TRIGGER_transactional) {
struct bkey_i_alloc_v4 *a = bch2_trans_start_alloc_update(trans, bucket);
+
+ percpu_down_read(&c->mark_lock);
+
ret = PTR_ERR_OR_ZERO(a) ?:
__mark_pointer(trans, ca, k, &p.ptr, *sectors, bp.data_type, &a->v);
+
+ percpu_up_read(&c->mark_lock);
+
if (ret)
goto err;
---
base-commit: 256abd8e550ce977b728be79a74e1729438b4948
change-id: 20240709-bug10-6b354d193f20
Best regards,
--
Pei Li <peili.dev@xxxxxxxxx>