svn commit: r271753 - head/sys/fs/autofs

[Konstantin Belousov]

On Thu, Sep 18, 2014 at 03:12:16PM +0200, Edward Tomasz Napierala wrote:
> On 0918T1421, Konstantin Belousov wrote:
> > On Thu, Sep 18, 2014 at 10:33:23AM +0000, Edward Tomasz Napierala wrote:
> > > Author: trasz
> > > Date: Thu Sep 18 10:33:23 2014
> > > New Revision: 271753
> > > URL: http://svnweb.freebsd.org/changeset/base/271753
> > > 
> > > Log:
> > >   Fix typos.
> > >   
> > >   Sponsored by:	The FreeBSD Foundation
> > > 
> > > Modified:
> > >   head/sys/fs/autofs/autofs.c
> > > 
> > > Modified: head/sys/fs/autofs/autofs.c
> > > ==============================================================================
> > > --- head/sys/fs/autofs/autofs.c	Thu Sep 18 10:01:56 2014	(r271752)
> > > +++ head/sys/fs/autofs/autofs.c	Thu Sep 18 10:33:23 2014	(r271753)
> > > @@ -430,7 +430,7 @@ autofs_trigger_one(struct autofs_node *a
> > >  			autofs_restore_sigmask(&oldset);
> > >  			if (error != 0) {
> > >  				/*
> > > -				 * XXX: For some reson this returns -1
> > > +				 * XXX: For some reason this returns -1
> > >  				 *	instead of EINTR, wtf?!
> > >  				 */
> > >  				error = EINTR;
> > > @@ -542,7 +542,7 @@ autofs_ioctl_request(struct autofs_daemo
> > >  		    &autofs_softc->sc_lock);
> > >  		if (error != 0) {
> > >  			/*
> > > -			 * XXX: For some reson this returns -1 instead
> > > +			 * XXX: For some reason this returns -1 instead
> > >  			 * 	of EINTR, wtf?!
> > >  			 */
> > >  			error = EINTR;
> > The -1 is ERESTART, i.e. it is correct error value which indicates that
> > syscall return path must restart the syscall after the trip to userland.
> > This behaviour is controlled by SA_RESTART flag, see sigaction(2).
> 
> Ah, I didn't realize it's a valid error number.
> 
> > There are some syscalls, which explicitely deny restarting, mostly to
> > correctly handle timeout values, most prominent is select(2) family.
> > 
> > The comment above is definitely inappropriate, and translation of ERESTART
> > to EINTR may be wrong as well.
> 
> The whole point of this is to fail the syscall with EINTR.  However,
> I'm not sure what makes cv_wait_sig(9) to return with ERESTART instead
> of EINTR.  Is SA_RESTART set by default?
When SA_RESTART set, interruption of cv_wait_sig() by signal causes
it to return ERESTART instead of EINTR.

Default disposition is either ignoring of signals, or generating core
dump, so question about SA_RESTART does not make sense.  Despite this,
ps_sigintr signal mask of the restartable signals is clear by default,
so typical interrupt causes ERESTART before default termination action
happens.

Can you explain why this matters for autofs code ?  Normal action for
the kernel code is to pass errors from sig_wait class of functions to
upper levels verbatim, except when the timeout must not be restarted.
-------------- 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-head/attachments/20140918/0bcf6ff0/attachment.sig>