Re: [syzbot] [reiserfs?] INFO: task hung in flush_old_commits

On 5/30/2023 1:21 PM, Jan Kara wrote:
> On Fri 26-05-23 11:45:57, Roberto Sassu wrote:
>> On Wed, 2023-05-24 at 17:57 -0400, Paul Moore wrote:
>>> On Wed, May 24, 2023 at 11:50 AM Roberto Sassu
>>> <[email protected]> wrote:
>>>> On Wed, 2023-05-24 at 11:11 -0400, Paul Moore wrote:
>>>>> On Wed, May 24, 2023 at 5:59 AM syzbot
>>>>> <[email protected]> wrote:
>>>>>> syzbot has bisected this issue to:
>>>>>>
>>>>>> commit d82dcd9e21b77d338dc4875f3d4111f0db314a7c
>>>>>> Author: Roberto Sassu <[email protected]>
>>>>>> Date:   Fri Mar 31 12:32:18 2023 +0000
>>>>>>
>>>>>>      reiserfs: Add security prefix to xattr name in reiserfs_security_write()
>>>>>>
>>>>>> bisection log:  https://syzkaller.appspot.com/x/bisect.txt?x=11c39639280000
>>>>>> start commit:   421ca22e3138 Merge tag 'nfs-for-6.4-2' of git://git.linux-..
>>>>>> git tree:       upstream
>>>>>> final oops:     https://syzkaller.appspot.com/x/report.txt?x=13c39639280000
>>>>>> console output: https://syzkaller.appspot.com/x/log.txt?x=15c39639280000
>>>>>> kernel config:  https://syzkaller.appspot.com/x/.config?x=7d8067683055e3f5
>>>>>> dashboard link: https://syzkaller.appspot.com/bug?extid=0a684c061589dcc30e51
>>>>>> syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=14312791280000
>>>>>> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=12da8605280000
>>>>>>
>>>>>> Reported-by: [email protected]
>>>>>> Fixes: d82dcd9e21b7 ("reiserfs: Add security prefix to xattr name in reiserfs_security_write()")
>>>>>>
>>>>>> For information about bisection process see: https://goo.gl/tpsmEJ#bisection
>>>>>
>>>>> Roberto, I think we need to resolve this somehow.  As I mentioned
>>>>> earlier, I don't believe this to be a fault in your patch, rather that
>>>>> patch simply triggered a situation that had not been present before,
>>>>> likely because the reiserfs code always failed when writing LSM
>>>>> xattrs.  Regardless, we still need to fix the deadlocks that sysbot
>>>>> has been reporting.
>>>>
>>>> Hi Paul
>>>>
>>>> ok, I will try.
>>>
>>> Thanks Roberto.  If it gets to be too challenging, let us know and we
>>> can look into safely disabling the LSM xattrs for reiserfs, I'll be
>>> shocked if anyone is successfully using LSM xattrs on reiserfs.
>>
>> Ok, at least I know what happens...
>>
>> + Jan, Jeff
>>
>> I'm focusing on this reproducer, which works 100% of the times:
>>
>> https://syzkaller.appspot.com/text?tag=ReproSyz&x=163079f9280000
> 
> Well, the commit d82dcd9e21b ("reiserfs: Add security prefix to xattr name
> in reiserfs_security_write()") looks obviously broken to me. It does:
> 
> char xattr_name[XATTR_NAME_MAX + 1] = XATTR_SECURITY_PREFIX;
> 
> Which is not how we can initialize strings in C... ;)

Thanks for having a look!

Sorry for the silly question, do I need to patch it? It is already in 
stable kernels... (next time I document myself better)

Thanks

Roberto

>> This is the last lock, before things go wrong:
>>
>> Thread 5 hit Breakpoint 2, reiserfs_write_lock (s=s@entry=0xffff888066e28000) at fs/reiserfs/lock.c:24
>> 24	{
>> (gdb) bt
>> #0  reiserfs_write_lock (s=s@entry=0xffff888066e28000) at fs/reiserfs/lock.c:24
>> #1  0xffffffff821a559a in reiserfs_get_block (inode=inode@entry=0xffff888069fd0190, block=block@entry=15, bh_result=bh_result@entry=0xffff888075940000, create=create@entry=1) at fs/reiserfs/inode.c:680
>> #2  0xffffffff81f50254 in __block_write_begin_int (folio=0xffffea00019a9180, pos=pos@entry=61440, len=len@entry=1, get_block=get_block@entry=0xffffffff821a5390 <reiserfs_get_block>, iomap=iomap@entry=0x0 <fixed_percpu_data>) at fs/buffer.c:2064
>> #3  0xffffffff81f5165a in __block_write_begin (page=page@entry=0xffffea00019a9180, pos=pos@entry=61440, len=len@entry=1, get_block=get_block@entry=0xffffffff821a5390 <reiserfs_get_block>) at ./arch/x86/include/asm/jump_label.h:27
>> #4  0xffffffff821a3e3d in reiserfs_write_begin (file=<optimized out>, mapping=<optimized out>, pos=61440, len=1, pagep=<optimized out>, fsdata=<optimized out>) at fs/reiserfs/inode.c:2779
>> #5  0xffffffff81aec252 in generic_perform_write (iocb=iocb@entry=0xffffc9002130fb60, i=i@entry=0xffffc9002130fd00) at mm/filemap.c:3923
>> #6  0xffffffff81b0604e in __generic_file_write_iter (iocb=iocb@entry=0xffffc9002130fb60, from=from@entry=0xffffc9002130fd00) at mm/filemap.c:4051
>> #7  0xffffffff81b06383 in generic_file_write_iter (iocb=0xffffc9002130fb60, from=0xffffc9002130fd00) at mm/filemap.c:4083
>> #8  0xffffffff81e3240b in call_write_iter (file=0xffff888012692d00, iter=0xffffc9002130fd00, kio=0xffffc9002130fb60) at ./include/linux/fs.h:1868
>> #9  do_iter_readv_writev (filp=filp@entry=0xffff888012692d00, iter=iter@entry=0xffffc9002130fd00, ppos=ppos@entry=0xffffc9002130fe90, type=type@entry=1, flags=flags@entry=0) at fs/read_write.c:735
>> #10 0xffffffff81e33da4 in do_iter_write (flags=0, pos=0xffffc9002130fe90, iter=0xffffc9002130fd00, file=0xffff888012692d00) at fs/read_write.c:860
>> #11 do_iter_write (file=0xffff888012692d00, iter=0xffffc9002130fd00, pos=0xffffc9002130fe90, flags=0) at fs/read_write.c:841
>> #12 0xffffffff81e34611 in vfs_writev (file=file@entry=0xffff888012692d00, vec=vec@entry=0x20000480, vlen=vlen@entry=1, pos=pos@entry=0xffffc9002130fe90, flags=flags@entry=0) at fs/read_write.c:933
>> #13 0xffffffff81e34fd6 in do_pwritev (fd=fd@entry=5, vec=vec@entry=0x20000480, vlen=vlen@entry=1, pos=pos@entry=61440, flags=flags@entry=0) at fs/read_write.c:1030
>> #14 0xffffffff81e3b61f in __do_sys_pwritev2 (pos_h=<optimized out>, flags=0, pos_l=61440, vlen=1, vec=0x20000480, fd=5) at fs/read_write.c:1089
>> #15 __se_sys_pwritev2 (pos_h=<optimized out>, flags=0, pos_l=61440, vlen=1, vec=536872064, fd=5) at fs/read_write.c:1080
>> #16 __x64_sys_pwritev2 (regs=0xffffc9002130ff58) at fs/read_write.c:1080
>> #17 0xffffffff880dd279 in do_syscall_x64 (nr=<optimized out>, regs=0xffffc9002130ff58) at arch/x86/entry/common.c:50
>> #18 do_syscall_64 (regs=0xffffc9002130ff58, nr=<optimized out>) at arch/x86/entry/common.c:80
>> #19 0xffffffff8820008b in entry_SYSCALL_64 () at arch/x86/entry/entry_64.S:120
>> #20 0x0000000000406e00 in ?? ()
>> #21 0x00007f99e21b5000 in ?? ()
>> #22 0x0000000000000000 in ?? ()
>>
>> After that, there is a very long loop doing:
>>
>> Thread 5 hit Breakpoint 3, reiserfs_read_bitmap_block (sb=sb@entry=0xffff888066e28000, bitmap=bitmap@entry=1) at fs/reiserfs/bitmap.c:1417
>> 1417	{
>> (gdb) c
>> Continuing.
>>
>> Thread 5 hit Breakpoint 3, reiserfs_read_bitmap_block (sb=sb@entry=0xffff888066e28000, bitmap=bitmap@entry=2) at fs/reiserfs/bitmap.c:1417
>> 1417	{
>> (gdb)
>> Continuing.
>>
>> and so on...
>>
>> [  628.589974][ T6003] REISERFS warning (device loop0): sh-2029: %s: bitmap block (#%u) reading failed reiserfs_read_bitmap_block: reiserfs_read_bitmap_block
>>
>> This message appears because we are here:
>>
>> struct buffer_head *reiserfs_read_bitmap_block(struct super_block *sb,
>>                                                 unsigned int bitmap)
>> {
>>
>> [...]
>>
>> 	bh = sb_bread(sb, block);
>> 	if (bh == NULL)
>> 		reiserfs_warning(sb, "sh-2029: %s: bitmap block (#%u) "
>> 		                 "reading failed", __func__, block);
>>
>> The hanging task (kthread) is trying to hold the same lock, which
>> unfortunately is not going to be released soon:
>>
>> static int reiserfs_sync_fs(struct super_block *s, int wait)
>> {
>>
>> [...]
>>
>> 	reiserfs_write_lock(s);
>>
>> I didn't get yet if the reason of this long loop is because we cannot
>> flush at this point, or just because of the test. I tried to
>> synchronously flush, but didn't make any difference.
>>
>> I did a very simple change, which can be totally wrong:
>>
>> @@ -94,7 +96,7 @@ static void flush_old_commits(struct work_struct *work)
>>           * trylock as reiserfs_cancel_old_flush() may be waiting for this work
>>           * to complete with s_umount held.
>>           */
>> -       if (!down_read_trylock(&s->s_umount)) {
>> +       if (sbi->lock_owner || !down_read_trylock(&s->s_umount)) {
>>                  /* Requeue work if we are not cancelling it */
>>                  spin_lock(&sbi->old_work_lock);
>>                  if (sbi->work_queued == 1)
>>
>>
>> If the lock is held, instead of waiting, reschedule the flush.
>>
>> Anyway, at least this report does not seem to be related to fixing
>> security xattrs.
>>
>> Roberto
>>