[BUG] ocfs2: possible deadlock in ocfs2_write_begin_nolock on v7.2-rc3

[CJ <firefly0158@163.com>]

Hi,

I am reporting that possible deadlock in ocfs2_write_begin_nolock remains
reproducible with a migrated syzkaller reproducer on Linux 7.2-rc3. The
tested HEAD commit is a13c140cc289c0b7b3770bce5b3ad42ab35074aa.

The reproducer mounts a crafted OCFS2 image and performs file and xattr
operations that enter the buffered write path.

Lockdep reports a circular dependency when ocfs2_write_begin_nolock() starts
a transaction while the inode allocation semaphore is already held. The
reverse dependency is established through OCFS2 refcount-tree and xattr
handling.

This looks like an inconsistent lock ordering between the inode allocation
semaphore and the filesystem transaction lock in the OCFS2 write path.

This issue was previously reported at:

https://syzkaller.appspot.com/bug?extid=1782a3a587bc53dc10cf

Reproducer:

syz reproducer: https://pastebin.com/raw/Ch9E7gAi

console output: https://pastebin.com/raw/02Zj2gLb
kernel config: https://pastebin.com/raw/RLSmRk5j
Kernel:

HEAD commit: a13c140cc289c0b7b3770bce5b3ad42ab35074aa
git tree: torvalds/linux
kernel version: 7.2.0-rc3 #1 PREEMPT(full)
tested tag: v7.2-rc3

Let me know if you need more details or testing.

Best regards,
Xianying