[RT WARNING] DEBUG_LOCKS_WARN_ON(rt_mutex_owner(lock) != current) with fsfreeze (4.19.25-rt16)

[Juri Lelli <juri.lelli@redhat.com>]

Hi,

Running this reproducer on a 4.19.25-rt16 kernel (with lock debugging
turned on) produces warning below.

 --->8---
 # dd if=/dev/zero of=fsfreezetest count=999999
 # mkfs -t xfs -q ./fsfreezetest
 # mkdir testmount
 # mount -t xfs -o loop ./fsfreezetest ./testmount
 # for I in `seq 10`; do fsfreeze -f ./testmount; sleep 1; fsfreeze -u ./testmount; done
 --->8---

 ------------[ cut here ]------------
 DEBUG_LOCKS_WARN_ON(rt_mutex_owner(lock) != current)
 WARNING: CPU: 10 PID: 1226 at kernel/locking/rtmutex-debug.c:145 debug_rt_mutex_unlock+0x9b/0xb0
 Modules linked in: xfs [...]
 CPU: 10 PID: 1226 Comm: fsfreeze Not tainted 4.19.25-rt16 #2
 Hardware name: LENOVO 30B6S2F900/1030, BIOS S01KT61A 09/28/2018
 RIP: 0010:debug_rt_mutex_unlock+0x9b/0xb0
 Code: e8 aa af 3c 00 4c 8b 04 24 85 c0 74 a9 8b 05 3c 9c a6 02 85 c0 75 9f 48 c7 c6 b8 b4 2d 98 48 c7 c7 9b d2 2b 98 e8 d9 e5 f8 ff <0f> 0b 4c 8b 04 24 eb 84 66 66 2e 0f 1f 84 00 00 00 00 00 66 90 c3
 RSP: 0018:ffffa7efa60cbdd0 EFLAGS: 00010086
 RAX: 0000000000000000 RBX: ffff991b72813920 RCX: 0000000000000000
 RDX: 0000000000000007 RSI: ffffffff98318de2 RDI: 00000000ffffffff
 RBP: 0000000000000246 R08: 0000000000000000 R09: 0000000000024200
 R10: 0000000000000000 R11: 0000000000000000 R12: ffffa7efa60cbe08
 R13: ffffa7efa60cbe18 R14: ffff991b72813478 R15: ffffffff9730718d
 FS:  00007f19baf6a540(0000) GS:ffff991b9fb00000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 00007f19bae87040 CR3: 000000103c6ee002 CR4: 00000000001606e0
 Call Trace:
  rt_mutex_slowunlock+0x24/0x70
  __rt_mutex_unlock+0x45/0x80
  percpu_up_write+0x4b/0x60
  thaw_super_locked+0xdb/0x110
  do_vfs_ioctl+0x647/0x6f0
  ksys_ioctl+0x60/0x90
  __x64_sys_ioctl+0x16/0x20
  do_syscall_64+0x60/0x1f0
  entry_SYSCALL_64_after_hwframe+0x49/0xbe
 RIP: 0033:0x7f19bae9704b
 Code: 0f 1e fa 48 8b 05 3d be 0c 00 64 c7 00 26 00 00 00 48 c7 c0 ff ff ff ff c3 66 0f 1f 44 00 00 f3 0f 1e fa b8 10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 0d be 0c 00 f7 d8 64 89 01 48
 RSP: 002b:00007ffc6d275358 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
 RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007f19bae9704b
 RDX: 0000000000000000 RSI: 00000000c0045878 RDI: 0000000000000003
 RBP: 0000000000000003 R08: 0000000000000000 R09: 0000000000000000
 R10: 0000000000000000 R11: 0000000000000246 R12: 00007ffc6d275755
 R13: 00007ffc6d275500 R14: 0000000000000000 R15: 0000000000000000
 irq event stamp: 8002
 hardirqs last  enabled at (8001): [<ffffffff97a25981>] _raw_spin_unlock_irqrestore+0x81/0x90
 hardirqs last disabled at (8002): [<ffffffff97a25aa0>] _raw_spin_lock_irqsave+0x20/0x60
 softirqs last  enabled at (0): [<ffffffff970c04ad>] copy_process.part.36+0x89d/0x2170
 softirqs last disabled at (0): [<0000000000000000>]           (null)
 ---[ end trace 0000000000000002 ]---

AFAIU, this is a legit warning, since

 fsfreeze -f ./testmount grabs rt_mutexes embedded into
 sb->s_writers.rw_sem[SB_FREEZE_LEVELS] (rt-rwsem) as part of executing
 sb_wait_write() (for each FREEZE_LEVEL) in freeze_super().

 We then return to userspace.

 fsfreeze -u ./testmount unlocks the rt_mutexes while doing
 sb_freeze_unlock() in thaw_super_locked(). This is a different process
 w.r.t. the one that did the freeze above.

I noticed that a very similar problem was fixed (for !rt rwsem) by
5a817641f68a ("locking/percpu-rwsem: Annotate rwsem ownership transfer
by setting RWSEM_OWNER_UNKNOWN"). However, RT has of course to deal with
PI, so I wonder if there is an easy fix for this problem.

Suggestions?

Thanks,

- Juri