[PATCH] Fix bug 12268

[Simo]

On Thu, 2016-09-15 at 16:07 +0200, Volker Lendecke wrote:
> On Thu, Sep 15, 2016 at 09:25:14AM -0400, Simo wrote:
> > 
> > On Thu, 2016-09-15 at 13:57 +0200, Volker Lendecke wrote:
> > > 
> > > Review appreciated!
> > 
> > Is this file descriptor being accessed in a separate thread then ?
> > Or does a read() on it risk blocking the whole smbd ?
> 
> That's our standard aio way of doing things. With aio, it's in a
> thread, without aio, it's blocking smbd.

If it is in aiothread feel free to add my ack/reviewd-by

>  If you look at the pread
> susv4 definition, ETIMEDOUT is not defined as an error condition, so
> you might argue that it's GPFS's fault.

I noticed that, and I was almost going to comment that GPFS should
return EAGAIN and not ETIMEDOUT, but I've seen this happen in at least
another syscall that does not have ETIMEDOUT documented as a valid
error code (sorry do not recall what was the situation, might have been
a read() from a AF_INET socket)

>  But I thought as this might also happen in other situations, I put
> it into main smbd. If you think it's better done in vfs_gpfs, I can
> also put it there.

No I think smbd is fine, we do not know if gpfs is the only, and if no
other file system does this then we get no harm anyway, right ?

Simo.