Re: ipc-msg broken again on 3.11-rc7?

[Manfred Spraul]

On 09/03/2013 11:16 AM, Vineet Gupta wrote:
> On 09/03/2013 02:27 PM, Manfred Spraul wrote:
> > On 09/03/2013 10:44 AM, Vineet Gupta wrote:
> > > > b) Could you check that it is not just a performance regression?
> > > >        Does ./msgctl08 1000 16 hang, too?
> > > Nope that doesn't hang. The minimal configuration that hangs reliably is msgctl
> > > 50000 2
> > >
> > > With this config there are 3 processes.
> > > ...
> > >     555   554 root     S     1208  0.4   0  0.0 ./msgctl08 50000 2
> > >     554   551 root     S     1208  0.4   0  0.0 ./msgctl08 50000 2
> > >     551   496 root     S     1208  0.4   0  0.0 ./msgctl08 50000 2
> > > ...
> > >
> > > [ARCLinux]$ cat /proc/551/stack
> > > [<80aec3c6>] do_wait+0xa02/0xc94
> > > [<80aecad2>] SyS_wait4+0x52/0xa4
> > > [<80ae24fc>] ret_from_system_call+0x0/0x4
> > >
> > > [ARCLinux]$ cat /proc/555/stack
> > > [<80c2950e>] SyS_msgrcv+0x252/0x420
> > > [<80ae24fc>] ret_from_system_call+0x0/0x4
> > >
> > > [ARCLinux]$ cat /proc/554/stack
> > > [<80c28c82>] do_msgsnd+0x116/0x35c
> > > [<80ae24fc>] ret_from_system_call+0x0/0x4
> > >
> > > Is this a case of lost wakeup or some such. I'm running with some more diagnostics
> > > and will report soon ...
> > What is the output of ipcs -q? Is the queue full or empty when it hangs?
> > I.e. do we forget to wake up a receiver or forget to wake up a sender?
> / # ipcs -q
>
> ------ Message Queues --------
> key        msqid      owner      perms      used-bytes   messages
> 0x72d83160 163841     root       600        0            0
Ok, a sender is sleeping - even though there are no messages in the queue.
Perhaps it is the race that I mentioned in a previous mail:
>       for (;;) {
>                 struct msg_sender s;
>
>                 err = -EACCES;
>                 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
>                         goto out_unlock1;
>
>                 err = security_msg_queue_msgsnd(msq, msg, msgflg);
>                 if (err)
>                         goto out_unlock1;
>
>                 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
>                                 1 + msq->q_qnum <= msq->q_qbytes) {
>                         break;
>                 }
[snip]
>         if (!pipelined_send(msq, msg)) {
>                 /* no one is waiting for this message, enqueue it */
>                 list_add_tail(&msg->m_list, &msq->q_messages);
>                 msq->q_cbytes += msgsz;
>                 msq->q_qnum++;
>                 atomic_add(msgsz, &ns->msg_bytes);

The access to msq->q_cbytes is not protected.

Vineet, could you try to move the test for free space after ipc_lock?
I.e. the lock must not get dropped between testing for free space and 
enqueueing the messages.

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