Re: [PATCH] fs: fix data races on inode->i_flctx

[Jeff Layton <jlayton@poochiereds.net>]

On Mon, 2015-10-19 at 17:18 +0200, William Dauchy wrote:
> Hello Dmitry,
> 
> On Mon, Sep 21, 2015 at 1:44 PM, Jeff Layton <jlayton@poochiereds.net
> > wrote:
> > Ok, thanks for the explanation. Patch looks fine to me. I'll go
> > ahead
> > and merge it for v4.4. Let me know though if you think this needs
> > to go
> > in sooner.
> 
> I am getting a null deref on a v4.1.x
> Do you think your patch could fix the following trace? It looks
> similar in my opinion.
> 
> BUG: unable to handle kernel NULL pointer dereference at
> 00000000000001c8
> IP: [<ffffffff811d0cf3>] locks_get_lock_context+0x3/0xc0
> PGD 0
> Oops: 0000 [#1] SMP
> CPU: 1 PID: 1773 Comm: kworker/1:1H Not tainted 4.1.11-rc1 #1
> Workqueue: rpciod ffffffff8164fff0
> task: ffff8810374deba0 ti: ffff8810374df150 task.ti: ffff8810374df150
> RIP: 0010:[<ffffffff811d0cf3>]  [<ffffffff811d0cf3>]
> locks_get_lock_context+0x3/0xc0
> RSP: 0000:ffff881036007bb0  EFLAGS: 00010246
> RAX: ffff881036007c30 RBX: ffff881001981880 RCX: 0000000000000002
> RDX: 00000000000006ed RSI: 0000000000000002 RDI: 0000000000000000
> RBP: ffff881036007c08 R08: 0000000000000000 R09: 0000000000000001
> R10: 0000000000000000 R11: ffff88101db69948 R12: ffff8810019818d8
> R13: ffff881036007bc8 R14: ffff880e225d81c0 R15: ffff881edfd2b400
> FS:  0000000000000000(0000) GS:ffff88103fc20000(0000)
> knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00000000000001c8 CR3: 000000000169b000 CR4: 00000000000606f0
> Stack:
>  ffffffff811d2710 ffff881036007bc8 ffffffff819f1af1 ffff881036007bc8
>  ffff881036007bc8 ffff881036007c08 ffff881001981880 ffff8810019818d8
>  ffff881036007c48 ffff880e225d81c0 ffff881edfd2b400 ffff881036007c88
> Call Trace:
>  [<ffffffff811d2710>] ? flock_lock_file+0x30/0x270
>  [<ffffffff811d3ad1>] flock_lock_file_wait+0x41/0xf0
>  [<ffffffff8168be66>] ? _raw_spin_unlock+0x26/0x40
>  [<ffffffff81268de9>] do_vfs_lock+0x19/0x40
>  [<ffffffff812695cc>] nfs4_locku_done+0x5c/0xf0
>  [<ffffffff8164f3f4>] rpc_exit_task+0x34/0xb0
>  [<ffffffff8164fcd9>] __rpc_execute+0x79/0x390
>  [<ffffffff81650000>] rpc_async_schedule+0x10/0x20
>  [<ffffffff81086095>] process_one_work+0x1a5/0x450
>  [<ffffffff81086024>] ? process_one_work+0x134/0x450
>  [<ffffffff8108638b>] worker_thread+0x4b/0x4a0
>  [<ffffffff81086340>] ? process_one_work+0x450/0x450
>  [<ffffffff81086340>] ? process_one_work+0x450/0x450
>  [<ffffffff8108d777>] kthread+0xf7/0x110
>  [<ffffffff8108d680>] ? __kthread_parkme+0xa0/0xa0
>  [<ffffffff8168ce3e>] ret_from_fork+0x3e/0x70
>  [<ffffffff8108d680>] ? __kthread_parkme+0xa0/0xa0
> Code: 48 b8 00 00 00 00 00 00 00 80 55 48 89 e5 48 09 c1 ff d1 5d 85
> c0 0f 95 c0 0f b6 c0 eb b9 66 2e 0f 1f 84 00 00 00 00 00 83 fe 02
> <48>
> 8b 87 c8 01 00 00 0f 84 a0 00 00 00 48 85 c0 0f 85 97 00 00
> RIP  [<ffffffff811d0cf3>] locks_get_lock_context+0x3/0xc0
>  RSP <ffff881036007bb0>
> CR2: 00000000000001c8
> ---[ end trace 2da9686dda1b5574 ]---
> 
> 
> Thanks,


This should be fixed by this series of four commits that are already in
mainline:

   
bcd7f78d078ff6197715c1ed070c92aca57ec12c..ee296d7c5709440f8abd36b5b65c6
b3e388538d9

The basic problem is that when the struct file is being torn down,
file_inode(file) may already return NULL. So we need to be able to pass
in the inode directly from the reference held by the nfs_open_context.

Are those patches reasonable to pull in?

-- 
Jeff Layton <jlayton@poochiereds.net>