Re: PANIC: "Oops: 0000 [#1] PREEMPT SMP PTI" starting from 5.17 on dual socket Intel Xeon Gold servers

[Minchan Kim <minchan@kernel.org>]

On Thu, Apr 21, 2022 at 06:47:41AM -1000, Tejun Heo wrote:
> Sorry about late reply.
> 
> On Wed, Apr 20, 2022 at 10:02:20AM +0200, Jirka Hladky wrote:
> > > Based on your report, kernel was crashed due to kn_mondata was NULL
> > >
> > >   rdt_kill_sb
> > >     rmdir_all_sub
> > >       ..
> > >       kernfs_remove(kn_mondata);
> > >         struct kernfs_root *root = kernfs_root(kn); <-- crashed
> > >
> > >
> > > Before the my patch[1], it worked like this.
> > >
> > >   rdt_kill_sb
> > >     rmdir_all_sub
> > >       ..
> > >       kernfs_remove(kn_mondata);
> > >         down_write(&kernfs_rwsem);
> > >           if (!kn)
> > >             return;
> > >         up_write(&kernfs_rwsem);
> > >
> > > IOW, before, kernfs_remove worked with NULL argument via just bailing
> > > but with the my patch[1], it doesn't work any longer.
> > >
> > > It makes me have questions for kernfs maintainers:
> > >
> > > Should kernfs_remove API support NULL parameter? If so, can we support
> > > it atomically without old global kernfs_rwsem?
> > >
> > > [1] 393c3714081a, kernfs: switch global kernfs_rwsem lock to per-fs lock
> 
> Yes, I mean, kernfs_remove() used to support NULL arg, so it should do the
> same after the locking change too. Can you send a patch?

Thanks for checking, Tejun.

Jirka, Could you test the patch? Once it's confirmed, I need to resend
it with Ccing stable.

Thanks.

From c7441bc659d2869f2d751b43f27356156e028513 Mon Sep 17 00:00:00 2001
From: Minchan Kim <minchan@kernel.org>
Date: Fri, 22 Apr 2022 11:16:45 -0700
Subject: [PATCH] kernfs: fix NULL dereferencing in kernfs_remove

kernfs_remove supported NULL kernfs_node param to bail out but revent
per-fs lock change introduced regression that dereferencing the
param without NULL check so kernel goes crash.

This patch checks the NULL kernfs_node in kernfs_remove and if so,
just return.

Quote from bug report by Jirka

```
The bug is triggered by running NAS Parallel benchmark suite on
SuperMicro servers with 2x Xeon(R) Gold 6126 CPU. Here is the error
log:

[  247.035564] BUG: kernel NULL pointer dereference, address: 0000000000000008
[  247.036009] #PF: supervisor read access in kernel mode
[  247.036009] #PF: error_code(0x0000) - not-present page
[  247.036009] PGD 0 P4D 0
[  247.036009] Oops: 0000 [#1] PREEMPT SMP PTI
[  247.058060] CPU: 1 PID: 6546 Comm: umount Not tainted
5.16.0393c3714081a53795bbff0e985d24146def6f57f+ #16
[  247.058060] Hardware name: Supermicro Super Server/X11DDW-L, BIOS
2.0b 03/07/2018
[  247.058060] RIP: 0010:kernfs_remove+0x8/0x50
[  247.058060] Code: 4c 89 e0 5b 5d 41 5c 41 5d 41 5e c3 49 c7 c4 f4
ff ff ff eb b2 66 66 2e 0f 1f 84 00 00 00 00 00 66 90 0f 1f 44 00 00
41 54 55 <48> 8b 47 08 48 89 fd 48 85 c0 48 0f 44 c7 4c 8b 60 50 49 83
c4 60
[  247.058060] RSP: 0018:ffffbbfa48a27e48 EFLAGS: 00010246
[  247.058060] RAX: 0000000000000001 RBX: ffffffff89e31f98 RCX: 0000000080200018
[  247.058060] RDX: 0000000080200019 RSI: fffff6760786c900 RDI: 0000000000000000
[  247.058060] RBP: ffffffff89e31f98 R08: ffff926b61b24d00 R09: 0000000080200018
[  247.122048] R10: ffff926b61b24d00 R11: ffff926a8040c000 R12: ffff927bd09a2000
[  247.122048] R13: ffffffff89e31fa0 R14: dead000000000122 R15: dead000000000100
[  247.122048] FS:  00007f01be0a8c40(0000) GS:ffff926fa8e40000(0000)
knlGS:0000000000000000
[  247.122048] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  247.122048] CR2: 0000000000000008 CR3: 00000001145c6003 CR4: 00000000007706e0
[  247.122048] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  247.122048] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  247.122048] PKRU: 55555554
[  247.122048] Call Trace:
[  247.122048]  <TASK>
[  247.122048]  rdt_kill_sb+0x29d/0x350
[  247.122048]  deactivate_locked_super+0x36/0xa0
[  247.122048]  cleanup_mnt+0x131/0x190
[  247.122048]  task_work_run+0x5c/0x90
[  247.122048]  exit_to_user_mode_prepare+0x229/0x230
[  247.122048]  syscall_exit_to_user_mode+0x18/0x40
[  247.122048]  do_syscall_64+0x48/0x90
[  247.122048]  entry_SYSCALL_64_after_hwframe+0x44/0xae
[  247.122048] RIP: 0033:0x7f01be2d735b
```

Fixes: 393c3714081a (kernfs: switch global kernfs_rwsem lock to per-fs lock)
Reported-by: Jirka Hladky <jhladky@redhat.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
---
 fs/kernfs/dir.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/fs/kernfs/dir.c b/fs/kernfs/dir.c
index 61a8edc4ba8b..e205fde7163a 100644
--- a/fs/kernfs/dir.c
+++ b/fs/kernfs/dir.c
@@ -1406,7 +1406,12 @@ static void __kernfs_remove(struct kernfs_node *kn)
  */
 void kernfs_remove(struct kernfs_node *kn)
 {
-	struct kernfs_root *root = kernfs_root(kn);
+	struct kernfs_root *root;
+
+	if (!kn)
+		return;
+
+	root = kernfs_root(kn);
 
 	down_write(&root->kernfs_rwsem);
 	__kernfs_remove(kn);
-- 
2.36.0.rc2.479.g8af0fa9b8e-goog