Re: [PATCH 3/5] libceph: a couple tweaks for wait loops

[Alex Elder <elder@ieee.org>]

On 05/25/2015 05:38 AM, Ilya Dryomov wrote:
> On Thu, May 21, 2015 at 4:29 PM, Alex Elder <elder@ieee.org> wrote:
>> On 05/21/2015 07:35 AM, Ilya Dryomov wrote:
>>> - return -ETIMEDOUT instead of -EIO in case of timeout
>>> - wait_event_interruptible_timeout() returns time left until timeout
>>>   and since it can be almost LONG_MAX we had better assign it to long
>>
>> Any error returned by wait_event_interruptible_timeout()
>> can now be returned by __ceph_open_session().  It looks
>> like that may, in fact, be only -EINTR and -ERESTARTSYS.
>> But it's a change you could note in the log message.
> 
> I think it's just -ERESTARTSYS so I didn't bother.

My point was almost a little more philosophical.  It's conceivable
(though not likely) that wait_event_interruptible_timeout() could
be changed to return a value that your caller here does not expect.

>> It turns out the only caller ignores the return value of
>> ceph_monc_wait_osdmap() anyway.  That should maybe be fixed.
> 
> That's on purpose - rbd map tries to wait for a new enough osdmap only
> if the pool that the image is supposed to be in doesn't exist and we
> know we have a stale osdmap.  We ignore wait retval because if we
> timeout we should return "this pool doesn't exist", not -ETIMEDOUT.

Yes I realize that.  This second part of my response was
following on to my previous thought.  That is, the caller
might get a different return value that it didn't expect;
but since the only caller ignores what gets returned, it's
a moot point.

					-Alex

> Thanks,
> 
>                 Ilya
>