[CRIU] Re: [PATCH] fdset: Make fdset open helpers to work in all-or-nothing manner

[Cyrill Gorcunov]

On Tue, Jan 31, 2012 at 01:52:01PM +0400, Pavel Emelyanov wrote:
> On 01/31/2012 01:38 PM, Cyrill Gorcunov wrote:
> > When cr_fdset_open or prep_cr_fdset_for_restore is called
> > the caller expects either all operations are successfull
> > or there is some error happened and NULL returned, but
> > instead at moment partially filled fdset might be returned.
> > 
> > Fix it -- we either should get all or nothing.
> > 
> > Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
> > ---
> >  crtools.c |   18 ++++++++++++------
> >  1 files changed, 12 insertions(+), 6 deletions(-)
> > 
> > diff --git a/crtools.c b/crtools.c
> > index 926d26b..d7f956d 100644
> > --- a/crtools.c
> > +++ b/crtools.c
> > @@ -157,15 +157,18 @@ struct cr_fdset *cr_fdset_open(int pid, unsigned long use_mask, struct cr_fdset
> >  			pr_perror("Unable to open %s\n", path);
> >  			goto err;
> >  		}
> > +		cr_fdset->fds[i] = ret;
> >  
> >  		pr_debug("Opened %s with %d\n", path, ret);
> >  		if (write_img(ret, &fdset_template[i].magic))
> >  			goto err;
> > -
> > -		cr_fdset->fds[i] = ret;
> >  	}
> > -err:
> > +
> >  	return cr_fdset;
> > +
> > +err:
> > +	close_cr_fdset(&cr_fdset);
> > +	return NULL;
> 
> NAK.
> 
> Code now works like this
> 
> 
> fdset = cr_fdset_open(NULL);
> 
>         fdset2 = cr_fdset_open(fdset);
> 
> cr_fdset_close(&fdset);
> 
> If you free the fdset in cr_fdset_open 2nd call you get the memory corruption.
> 

I see, it seems we need to pass **fdset so that close_cr_fdset(&cr_fdset) wont
screw on stack value.

	Cyrill