Hi Neil-

I'm testing v6.8-rc3 + nfsd-next. This series includes:

   nfsd: fix RELEASE_LOCKOWNER

and

   nfsd: don't call locks_release_private() twice concurrently


======================================================
WARNING: possible circular locking dependency detected
6.8.0-rc3-00052-gc20ad5c2d60c #1 Not tainted
------------------------------------------------------
nfsd/1218 is trying to acquire lock:
ffff88814d754198 (&ctx->flc_lock){+.+.}-{2:2}, at: check_for_locks+0xf6/0x1d0 [nfsd]

but task is already holding lock:
ffff8881210e61f0 (&fp->fi_lock){+.+.}-{2:2}, at: check_for_locks+0x2d/0x1d0 [nfsd]

which lock already depends on the new lock.


the existing dependency chain (in reverse order) is:

-> #1 (&fp->fi_lock){+.+.}-{2:2}:
       _raw_spin_lock+0x2f/0x42
       nfsd_break_deleg_cb+0x295/0x2f6 [nfsd]
       __break_lease+0x38b/0x864
       break_lease+0x87/0xc2
       do_dentry_open+0x5df/0xc8d
       vfs_open+0xbb/0xc4
       dentry_open+0x5a/0x7a
       __nfsd_open.isra.0+0x1ed/0x2a3 [nfsd]
       nfsd_open_verified+0x16/0x1c [nfsd]
       nfsd_file_do_acquire+0x149c/0x1650 [nfsd]
       nfsd_file_acquire+0x16/0x1c [nfsd]
       nfsd4_commit+0x72/0x10c [nfsd]
       nfsd4_proc_compound+0x12c5/0x17a8 [nfsd]
       nfsd_dispatch+0x32f/0x590 [nfsd]
       svc_process_common+0xb64/0x13b8 [sunrpc]
       svc_process+0x3b8/0x40c [sunrpc]
       svc_recv+0x1478/0x1803 [sunrpc]
       nfsd+0x2a1/0x2e3 [nfsd]
       kthread+0x2cb/0x2da
       ret_from_fork+0x2a/0x62
       ret_from_fork_asm+0x1b/0x30

-> #0 (&ctx->flc_lock){+.+.}-{2:2}:
       __lock_acquire+0x1e49/0x27f7
       lock_acquire+0x25c/0x3df
       _raw_spin_lock+0x2f/0x42
       check_for_locks+0xf6/0x1d0 [nfsd]
       nfsd4_release_lockowner+0x2b9/0x3a4 [nfsd]
       nfsd4_proc_compound+0x12c5/0x17a8 [nfsd]
       nfsd_dispatch+0x32f/0x590 [nfsd]
       svc_process_common+0xb64/0x13b8 [sunrpc]
       svc_process+0x3b8/0x40c [sunrpc]
       svc_recv+0x1478/0x1803 [sunrpc]
       nfsd+0x2a1/0x2e3 [nfsd]
       kthread+0x2cb/0x2da
       ret_from_fork+0x2a/0x62
       ret_from_fork_asm+0x1b/0x30

other info that might help us debug this:

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&fp->fi_lock);
                               lock(&ctx->flc_lock);
                               lock(&fp->fi_lock);
  lock(&ctx->flc_lock);

 *** DEADLOCK ***

2 locks held by nfsd/1218:
 #0: ffff88823a3ccf90 (&clp->cl_lock#2){+.+.}-{2:2}, at: nfsd4_release_lockowner+0x22d/0x3a4 [nfsd]
 #1: ffff8881210e61f0 (&fp->fi_lock){+.+.}-{2:2}, at: check_for_locks+0x2d/0x1d0 [nfsd]

stack backtrace:
CPU: 2 PID: 1218 Comm: nfsd Not tainted 6.8.0-rc3-00052-gc20ad5c2d60c #1
Hardware name: Supermicro Super Server/X10SRL-F, BIOS 3.3 10/28/2020
Call Trace:
 <TASK>
 dump_stack_lvl+0x70/0xa4
 dump_stack+0x10/0x16
 print_circular_bug+0x37c/0x38f
 check_noncircular+0x16d/0x19a
 ? __pfx_mark_lock+0x10/0x10
 ? __pfx_check_noncircular+0x10/0x10
 ? add_chain_block+0x142/0x19c
 __lock_acquire+0x1e49/0x27f7
 ? __pfx___lock_acquire+0x10/0x10
 ? check_for_locks+0xf6/0x1d0 [nfsd]
 lock_acquire+0x25c/0x3df
 ? check_for_locks+0xf6/0x1d0 [nfsd]
 ? __pfx_lock_acquire+0x10/0x10
 ? __kasan_check_write+0x14/0x1a
 ? do_raw_spin_lock+0x146/0x1ea
 _raw_spin_lock+0x2f/0x42
 ? check_for_locks+0xf6/0x1d0 [nfsd]
 check_for_locks+0xf6/0x1d0 [nfsd]
 nfsd4_release_lockowner+0x2b9/0x3a4 [nfsd]
 ? __pfx_nfsd4_release_lockowner+0x10/0x10 [nfsd]
 nfsd4_proc_compound+0x12c5/0x17a8 [nfsd]
 ? nfsd_read_splice_ok+0xe/0x1f [nfsd]
 nfsd_dispatch+0x32f/0x590 [nfsd]
 ? __pfx_nfsd_dispatch+0x10/0x10 [nfsd]
 ? svc_reserve+0xed/0xfc [sunrpc]
 svc_process_common+0xb64/0x13b8 [sunrpc]
 ? __pfx_svc_process_common+0x10/0x10 [sunrpc]
 ? __pfx_nfsd_dispatch+0x10/0x10 [nfsd]
 ? xdr_inline_decode+0x92/0x1cb [sunrpc]
 svc_process+0x3b8/0x40c [sunrpc]
 svc_recv+0x1478/0x1803 [sunrpc]
 nfsd+0x2a1/0x2e3 [nfsd]
 ? __kthread_parkme+0xcc/0x19c
 ? __pfx_nfsd+0x10/0x10 [nfsd]
 kthread+0x2cb/0x2da
 ? __pfx_kthread+0x10/0x10
 ret_from_fork+0x2a/0x62
 ? __pfx_kthread+0x10/0x10
 ret_from_fork_asm+0x1b/0x30
 </TASK>

I get a very similar splat while testing v5.15.y and v5.10.y
with "nfsd: fix RELEASE_LOCKOWNER" applied. I'm circling back
to v6.1.y as well, but I expect I will see the same there.


--
Chuck Lever