svn commit: r268493 - head/lib/libfetch

[Konstantin Belousov]

On Thu, Jul 10, 2014 at 03:52:53PM +0200, Baptiste Daroussin wrote:
> On Thu, Jul 10, 2014 at 04:28:13PM +0300, Konstantin Belousov wrote:
> > On Thu, Jul 10, 2014 at 01:04:53PM +0000, Baptiste Daroussin wrote:
> > > Author: bapt
> > > Date: Thu Jul 10 13:04:52 2014
> > > New Revision: 268493
> > > URL: http://svnweb.freebsd.org/changeset/base/268493
> > > 
> > > Log:
> > >   Support EAGAIN in fetch_writev
> > >   
> > >   Reviewed by:	des
> > >   Approved by:	des
> > > 
> > > Modified:
> > >   head/lib/libfetch/common.c
> > > 
> > > Modified: head/lib/libfetch/common.c
> > > ==============================================================================
> > > --- head/lib/libfetch/common.c	Thu Jul 10 12:41:58 2014	(r268492)
> > > +++ head/lib/libfetch/common.c	Thu Jul 10 13:04:52 2014	(r268493)
> > > @@ -1110,6 +1110,8 @@ fetch_writev(conn_t *conn, struct iovec 
> > >  			errno = 0;
> > >  			pfd.revents = 0;
> > >  			if (poll(&pfd, 1, deltams) < 0) {
> > > +				if (errno == EAGAIN)
> > > +					continue;
> > >  				if (errno == EINTR && fetchRestartCalls)
> > >  					continue;
> > >  				return (-1);
> > 
> > How is this error possible ? poll(2) converts EAGAIN from seltdwait(9)
> > to no error.
> 
> For posix compliance, libfetch is not only used on FreeBSD, reading at:
> http://pubs.opengroup.org/onlinepubs/009695399/functions/poll.html I can read:
> 
> [EAGAIN]
>     The allocation of internal data structures failed but a subsequent request
>     may succeed.
> 
> Am I missing something?

At least for me, it was very surprising.  I accustomed to the facts that
poll(2) is never restarted after signal, and that you only should test
for EINTR as the error condition.

Apparently, the behaviour of converting EAGAIN to no error can be tracked
back to at least 4.4 lite.  On the other hand, just looking into the man
page for poll(2) on STREAMS-based systems indeed mention EAGAIN.

My suggestion is to add a comment, noting that FreeBSD does not return
EAGAIN from poll(2), but SUSv4 allows it.  This way, the addition would
be not removed in some code cleanup.  Does it sound reasonable ?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/svn-src-all/attachments/20140710/32978a52/attachment.sig>