lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Tue, 20 May 2014 15:16:47 +0200 (CEST)
From:	Sebastian Ott <sebott@...ux.vnet.ibm.com>
To:	Benjamin LaHaise <bcrl@...ck.org>
cc:	Anatol Pomozov <anatol.pomozov@...il.com>, linux-aio@...ck.org,
	linux-kernel@...r.kernel.org
Subject: Re: hanging aio process

On Tue, 20 May 2014, Sebastian Ott wrote:
> On Mon, 19 May 2014, Benjamin LaHaise wrote:
> > It is entirely possible the bug isn't 
> > caused by the referenced commit, as the commit you're pointing to merely 
> > makes io_destroy() syscall wait for all aio outstanding to complete 
> > before returning.
> 
> I cannot reproduce this when I revert said commit (on top of 14186fe). If
> that matters - the arch is s390.

Hm, ok - maybe that commit is really just highlighting a refcounting bug.
I just compared traces for a good and a few bad cases. The good case:
# tracer: function
#
# entries-in-buffer/entries-written: 16/16   #P:4
#
#                              _-----=> irqs-off
#                             / _----=> need-resched
#                            | / _---=> hardirq/softirq
#                            || / _--=> preempt-depth
#                            ||| /     delay
#           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
#              | |       |   ||||       |         |
             fio-732   [003] ....    17.989315: kill_ioctx <-SyS_io_destroy
             fio-739   [003] ....    18.000563: kill_ioctx <-SyS_io_destroy
     ksoftirqd/3-19    [003] ..s.    18.031673: free_ioctx_users <-percpu_ref_kill_rcu
     ksoftirqd/3-19    [003] ..s.    18.031679: free_ioctx_users <-percpu_ref_kill_rcu
             fio-737   [003] ....    18.038765: kill_ioctx <-SyS_io_destroy
     ksoftirqd/3-19    [003] ..s.    18.062488: free_ioctx_reqs <-percpu_ref_kill_rcu
     ksoftirqd/3-19    [003] ..s.    18.062494: free_ioctx_reqs <-percpu_ref_kill_rcu
     kworker/3:1-57    [003] ....    18.062499: free_ioctx <-process_one_work
     kworker/3:1-57    [003] ....    18.062506: free_ioctx <-process_one_work
     ksoftirqd/3-19    [003] ..s.    18.072275: free_ioctx_users <-percpu_ref_kill_rcu
             fio-738   [003] ....    18.102419: kill_ioctx <-SyS_io_destroy
          <idle>-0     [003] .ns.    18.111668: free_ioctx_reqs <-percpu_ref_kill_rcu
     kworker/3:1-57    [003] ....    18.111675: free_ioctx <-process_one_work
     ksoftirqd/3-19    [003] ..s.    18.138035: free_ioctx_users <-percpu_ref_kill_rcu
          <idle>-0     [003] .ns.    18.191665: free_ioctx_reqs <-percpu_ref_kill_rcu
     kworker/3:1-57    [003] ....    18.191671: free_ioctx <-process_one_work

(4 fio workers, free_ioctx_reqs is called 4 times)

One of the bad cases:
# tracer: function
#
# entries-in-buffer/entries-written: 14/14   #P:4
#
#                              _-----=> irqs-off
#                             / _----=> need-resched
#                            | / _---=> hardirq/softirq
#                            || / _--=> preempt-depth
#                            ||| /     delay
#           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
#              | |       |   ||||       |         |
             fio-834   [000] ....    51.127359: kill_ioctx <-SyS_io_destroy
          <idle>-0     [000] ..s.    51.170237: free_ioctx_users <-percpu_ref_kill_rcu
             fio-828   [001] ....    51.189717: kill_ioctx <-SyS_io_destroy
             fio-833   [001] ..s.    51.220178: free_ioctx_users <-percpu_ref_kill_rcu
          <idle>-0     [000] .ns.    51.220230: free_ioctx_reqs <-percpu_ref_kill_rcu
     kworker/0:3-661   [000] ....    51.220238: free_ioctx <-process_one_work
          <idle>-0     [001] .ns.    51.260188: free_ioctx_reqs <-percpu_ref_kill_rcu
     kworker/1:2-103   [001] ....    51.260198: free_ioctx <-process_one_work
             fio-833   [002] ....    51.287602: kill_ioctx <-SyS_io_destroy
           udevd-868   [002] ..s1    51.332519: free_ioctx_users <-percpu_ref_kill_rcu
          <idle>-0     [002] .ns.    51.450180: free_ioctx_reqs <-percpu_ref_kill_rcu
     kworker/2:2-191   [002] ....    51.450191: free_ioctx <-process_one_work
             fio-835   [003] ....    51.907530: kill_ioctx <-SyS_io_destroy
     ksoftirqd/3-19    [003] ..s.    52.000232: free_ioctx_users <-percpu_ref_kill_rcu

(1 fio worker in D state, free_ioctx_reqs is called 3 times)

Regards,
Sebastian

> > 
> > > git bisect points to:
> > > 	commit e02ba72aabfade4c9cd6e3263e9b57bf890ad25c
> > > 	Author: Anatol Pomozov <anatol.pomozov@...il.com>
> > > 	Date:   Tue Apr 15 11:31:33 2014 -0700
> > > 
> > > 	    aio: block io_destroy() until all context requests are completed
> > > 
> > > 
> > > The fio workers are on the wait_for_completion in sys_io_destroy.
> > > 
> > > Regards,
> > > Sebastian
> > > [global]
> > > blocksize=4K
> > > size=256M
> > > rw=randrw
> > > verify=md5
> > > iodepth=32
> > > ioengine=libaio
> > > direct=1
> > > end_fsync=1
> > > 
> > > [file1]
> > > filename=/dev/scma
> > > 
> > > [file2]
> > > filename=/dev/scmbw
> > > 
> > > [file3]
> > > filename=/dev/scmc
> > > 
> > > [file4]
> > > filename=/dev/scmx
> > 
> > 
> > -- 
> > "Thought is the essence of where you are now."
> > 
> > 
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists