Re: Linux 4.10-rc7

[Dave Jones]

On Sun, Feb 05, 2017 at 03:41:54PM -0800, Linus Torvalds wrote:
 > Hey, look at that - it's all been very quiet, and unless anything bad
 > happens, we're all back to the regular schedule with this being the
 > last rc.
 

..

 > Christoph Hellwig (1):
 >       nfsd: special case truncates some more


This one smells funny to me.

[  145.983182] =============================================
[  145.983201] [ INFO: possible recursive locking detected ]
[  145.983220] 4.10.0-rc7+ #1 Not tainted
[  145.983236] ---------------------------------------------
[  145.984868] nfsd/865 is trying to acquire lock:
[  145.986497]  (sb_writers#7){.+.+.+}, at: [<ffffffff9b1e0a25>] mnt_want_write+0x25/0x50
[  145.988153] 
but task is already holding lock:
[  145.991350]  (sb_writers#7){.+.+.+}, at: [<ffffffff9b1e0a25>] mnt_want_write+0x25/0x50
[  145.992931] 
other info that might help us debug this:
[  145.995991]  Possible unsafe locking scenario:

[  145.998936]        CPU0
[  146.000358]        ----
[  146.001749]   lock(sb_writers#7);
[  146.003109]   lock(sb_writers#7);
[  146.004431] 
 *** DEADLOCK ***

[  146.008179]  May be due to missing lock nesting notation

[  146.010541] 1 lock held by nfsd/865:
[  146.011681]  #0:  (sb_writers#7){.+.+.+}, at: [<ffffffff9b1e0a25>] mnt_want_write+0x25/0x50
[  146.012820] 
stack backtrace:
[  146.015033] CPU: 0 PID: 865 Comm: nfsd Not tainted 4.10.0-rc7+ #1
[  146.016147] Call Trace:
[  146.017225]  dump_stack+0x68/0xa0
[  146.018273]  __lock_acquire+0x6ea/0x11d0
[  146.019251]  ? trace_hardirqs_on_caller+0xef/0x1b0
[  146.020119]  lock_acquire+0x109/0x270
[  146.020957]  ? mnt_want_write+0x25/0x50
[  146.021770]  __sb_start_write+0xd8/0x230
[  146.022555]  ? mnt_want_write+0x25/0x50
[  146.023313]  ? nfsd_permission+0x7f/0x110
[  146.024042]  mnt_want_write+0x25/0x50
[  146.024745]  vfs_truncate+0x4e/0x200
[  146.025421]  nfsd_setattr+0x13a/0x300
[  146.026070]  nfsd4_setattr+0x109/0x130
[  146.026690]  nfsd4_proc_compound+0x443/0x6e0
[  146.027286]  nfsd_dispatch+0x91/0x170
[  146.027853]  svc_process+0x6e6/0xa00
[  146.028393]  nfsd+0x193/0x2a0
[  146.028903]  ? nfsd+0x5/0x2a0
[  146.029448]  kthread+0x111/0x150
[  146.029964]  ? nfsd_destroy+0x190/0x190
[  146.030474]  ? __kthread_create_on_node+0x250/0x250
[  146.030988]  ret_from_fork+0x2e/0x40


I'll double check that it's that commit in the morning, but nfs+truncate seems to point
in this direction..

	Dave