Re: [PATCH] ocfs2: fix circular locking dependency in ocfs2_dio_end_io_write

[Joseph Qi <joseph.qi@linux.alibaba.com>]

On 6/15/26 3:32 PM, Heming Zhao wrote:
> On Fri, Jun 12, 2026 at 11:50:20AM +0000, syzbot wrote:
>> From: Aleksandr Nogikh <nogikh@google.com>
>>
>> A circular locking dependency involves INODE_ALLOC_SYSTEM_INODE,
>> EXTENT_ALLOC_SYSTEM_INODE, and ORPHAN_DIR_SYSTEM_INODE.
>>
>> 1. ocfs2_mknod() acquires INODE_ALLOC then EXTENT_ALLOC.
>> 2. ocfs2_dio_end_io_write() acquires EXTENT_ALLOC for unwritten extents,
>> then ORPHAN_DIR via ocfs2_del_inode_from_orphan() while still holding
>> EXTENT_ALLOC.
>> 3. ocfs2_wipe_inode() acquires ORPHAN_DIR then INODE_ALLOC via
>> ocfs2_remove_inode.
>>
>> Break the cycle in ocfs2_dio_end_io_write() by freeing the allocation
>> contexts (releasing EXTENT_ALLOC) before acquiring ORPHAN_DIR.
>>
>> WARNING: possible circular locking dependency detected
>> ------------------------------------------------------
>> is trying to acquire lock:
>> ffff8881e78b33a0
>> (&ocfs2_sysfile_lock_key[INODE_ALLOC_SYSTEM_INODE]){+.+.}-{4:4}, at:
>> ocfs2_evict_inode+0x1539/0x43b0 fs/ocfs2/inode.c:1299
>>
>> but task is already holding lock:
>> ffff8881e78b4fa0
>> (&ocfs2_sysfile_lock_key[ORPHAN_DIR_SYSTEM_INODE]){+.+.}-{4:4}, at:
>> ocfs2_evict_inode+0xe97/0x43b0 fs/ocfs2/inode.c:1299
>>
>> the existing dependency chain (in reverse order) is:
>>
>> -> #2 (&ocfs2_sysfile_lock_key[ORPHAN_DIR_SYSTEM_INODE]){+.+.}-{4:4}:
>>        inode_lock include/linux/fs.h:1029 [inline]
>>        ocfs2_del_inode_from_orphan+0x12e/0x7a0 fs/ocfs2/namei.c:2728
>>        ocfs2_dio_end_io+0xf9c/0x1370 fs/ocfs2/aops.c:2418
>>        dio_complete+0x25b/0x790 fs/direct-io.c:281
>>
>> -> #1 (&ocfs2_sysfile_lock_key[EXTENT_ALLOC_SYSTEM_INODE]){+.+.}-{4:4}:
>>        inode_lock include/linux/fs.h:1029 [inline]
>>        ocfs2_reserve_suballoc_bits+0x16d/0x4840 fs/ocfs2/suballoc.c:882
>>        ocfs2_reserve_new_metadata_blocks+0x415/0x9a0
>>        fs/ocfs2/suballoc.c:1078
>>        ocfs2_mknod+0x10f3/0x2260 fs/ocfs2/namei.c:351
>>
>> -> #0 (&ocfs2_sysfile_lock_key[INODE_ALLOC_SYSTEM_INODE]){+.+.}-{4:4}:
>>        __lock_acquire+0x15a5/0x2cf0 kernel/locking/lockdep.c:5237
>>        lock_acquire+0x106/0x350 kernel/locking/lockdep.c:5868
>>        down_write+0x96/0x200 kernel/locking/rwsem.c:1625
>>        inode_lock include/linux/fs.h:1029 [inline]
>>        ocfs2_remove_inode fs/ocfs2/inode.c:733 [inline]
>>        ocfs2_wipe_inode fs/ocfs2/inode.c:896 [inline]
>>        ocfs2_delete_inode fs/ocfs2/inode.c:1157 [inline]
>>        ocfs2_evict_inode+0x1539/0x43b0 fs/ocfs2/inode.c:1299
>>
>> Chain exists of:
>>   &ocfs2_sysfile_lock_key[INODE_ALLOC_SYSTEM_INODE] -->
>>   &ocfs2_sysfile_lock_key[EXTENT_ALLOC_SYSTEM_INODE] -->
>>   &ocfs2_sysfile_lock_key[ORPHAN_DIR_SYSTEM_INODE]
>>
>>  Possible unsafe locking scenario:
>>
>>        CPU0                    CPU1
>>        ----                    ----
>>   lock(&ocfs2_sysfile_lock_key[ORPHAN_DIR_SYSTEM_INODE]);
>>                                lock(&ocfs2_sysfile_lock_key[EXTENT_ALLOC_SYSTEM_INODE]);
>>                                lock(&ocfs2_sysfile_lock_key[ORPHAN_DIR_SYSTEM_INODE]);
>>   lock(&ocfs2_sysfile_lock_key[INODE_ALLOC_SYSTEM_INODE]);
>>
>>  *** DEADLOCK ***
>>

Better to describe the solution here.

>> Fixes: d647c5b2fbf8 ("ocfs2: split transactions in dio completion to avoid credit exhaustion")
>> Assisted-by: Gemini:gemini-3.1-pro-preview Gemini:gemini-3-flash-preview syzbot
>> Reported-by: syzbot+b225d4dfce6219600c42@syzkaller.appspotmail.com
>> Closes: https://syzkaller.appspot.com/bug?extid=b225d4dfce6219600c42
>> Link: https://syzkaller.appspot.com/ai_job?id=0b53ce1e-2972-4192-aa85-8097a702762c
>> Signed-off-by: Aleksandr Nogikh <nogikh@google.com>
> 
> LGTM.
> After d647c5b2fbf8, ocfs2_dio_end_io_write() spends a significant amount of time
> looping through the unwritten extent list before it finally acquires the
> ORPHAN_DIR lock. This gives other functions (such as ocfs2_wipe_inode() mentioned
> in this patch) a chance to preemptively grab the ORPHAN_DIR lock, thereby
> triggering the circular locking.
> What d647c5b2fbf8 actually did was widen the race window, turning a pre-existing,
> low-probability issue into an easily reproducible one.
> 

Ummm... before d647c5b2fbf8, ocfs2_del_inode_from_orphan() is called
before down_write(&ip_alloc_sem) and ocfs2_lock_allocators().
So I think the 'Fixes' tag looks reasonable.

Thanks,
Joseph

> Reviewed-by: Heming Zhao <heming.zhao@suse.com>
>>
>> ---
>> diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
>> index 6ec198bda..4acdbb708 100644
>> --- a/fs/ocfs2/aops.c
>> +++ b/fs/ocfs2/aops.c
>> @@ -2372,6 +2372,15 @@ static int ocfs2_dio_end_io_write(struct inode *inode,
>>  unlock:
>>  	up_write(&oi->ip_alloc_sem);
>>  
>> +	if (data_ac) {
>> +		ocfs2_free_alloc_context(data_ac);
>> +		data_ac = NULL;
>> +	}
>> +	if (meta_ac) {
>> +		ocfs2_free_alloc_context(meta_ac);
>> +		meta_ac = NULL;
>> +	}
>> +
>>  	/* everything looks good, let's start the cleanup */
>>  	if (!ret && dwc->dw_orphaned) {
>>  		BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
>> @@ -2383,10 +2392,6 @@ static int ocfs2_dio_end_io_write(struct inode *inode,
>>  	ocfs2_inode_unlock(inode, 1);
>>  	brelse(di_bh);
>>  out:
>> -	if (data_ac)
>> -		ocfs2_free_alloc_context(data_ac);
>> -	if (meta_ac)
>> -		ocfs2_free_alloc_context(meta_ac);
>>  	ocfs2_run_deallocs(osb, &dealloc);
>>  	ocfs2_dio_free_write_ctx(inode, dwc);
>>  
>>
>>
>> base-commit: e7ae89a0c97ce2b68b0983cd01eda67cf373517d
>> -- 
>> See https://goo.gle/syzbot-ai-patches for information about AI-generated patches.
>> You can comment on the patch as usual, syzbot will try to address
>> the comments and send a new version of the patch if necessary.
>> syzbot engineers can be reached at syzkaller@googlegroups.com.
>>