Re: LOR on nfs: vfs_vnops.c:301 kern_descrip.c:1580

On Tue, Aug 17, 2010 at 4:04 PM, Kostik Belousov <kostikbel_at_gmail.com> wrote:
> On Tue, Aug 17, 2010 at 07:42:41PM +0400, pluknet wrote:
>> 2010/8/16 Kostik Belousov <kostikbel_at_gmail.com>:
>> > On Mon, Aug 16, 2010 at 09:07:24PM +0400, pluknet wrote:
>> >> On 16 August 2010 21:05, pluknet <pluknet_at_gmail.com> wrote:
>> >> > Hi.
>> >> >
>> >> > Seeing on mostly idle, recently updated current, while closing a file.
>> >> > Presumably never reported on ML.
>> [...]
>> >>
>> > Both LORs are valid. The fork performed deep inside the VFS call stack
>> > is obviously problematic. As a workaround, you may fix the number of
>> > nfsiods.
>> >
>> > Proper fix might consist of creating a shepherd thread which only task
>> > is to act on the requests on creating new nfsiods.
>> >
>> > Would you try to implement this ? I will provide the assistance, if needed.
>>
>> Hmm.. I tried to move kproc_create() under shepherd thread and now stuck
>> with cp process lockup in [bo_wwait] when cp'ing something on nfs: cp a b.
>> Did I screw up something?
>> See weird draft patch attached (weird, as I have no idea how to nicely
>> exchange data between nfs_nfsiodnew() and shep_thread() thread).
> Most likely, you loose the requests to create nfsiods since the
> existing request in the global variable shep_chan can be overwritten
> by new request. You should either sleep till existing request is serviced,
> or form a queue.

If you sleep for the request to be serviced, this presumably has the
same LOR/deadlock possibility (unless locks are released before
sleep), except now WITNESS can't see the LOR.

Thanks,
matthew