[PATCH] fallocate returned values on failure

[Jeremy Allison]

On Wed, Dec 03, 2014 at 10:15:59PM -0800, Jeremy Allison wrote:
> On Thu, Dec 04, 2014 at 02:02:38PM +0800, Jones wrote:
> > Hello list,
> > 
> > There is samba-4.0.5 in the linux box,
> > and set strict allocate = yes to go through fallocate api.
> > 
> > Found if fallocate() returns -1 with errno = ENOSPC,
> > cannot catch by error handling as followings:
> >     ret = SMB_VFS_FALLOCATE(fsp, VFS_FALLOCATE_EXTEND_SIZE,
> >              offset, num_to_write);
> >   if (ret == ENOSPC) {
> >   errno = ENOSPC;
> > ret = -1;
> >         goto out;
> >   }
> > 
> > Per man pages said,
> > posix_fallocate() returns an error number on failure,
> > fallocate() returns -1 on failure.
> > 
> > Patch as attached could handle -1 returned from fallocate() on failure,
> > please help review and any suggestions are appreciated,
> > thanks.
> 
> Very good catch ! Thanks.
> 
> I think the best way to fix this is to
> standardize the return from SMB_VFS_FALLOCATE()
> to be -1,set errno on error, and then convert
> the lower level to do this if it uses posix_fallocate().
> 
> We should also fix vfs_slow_fallocate() to
> do the same - and then fix all the callers
> to expect -1,errno returns instead of errno.
> 
> Can you log a bug so we can track this and
> get it fixed in 4.0.x and above ?
> 
> I'll code a fix up for master, I'd appreciate
> it if you could test it !

Hmmm. This is a horrible mess... some of the
fallocate code paths (the gpfs or gluster versions for example)
return -1,errno, and many of the others return errno.

I'm going to go through and unify all these on
one return convention. It's the only way to stay
sane :-(.