Re: [Xen-devel] Re: [PATCH] libxl: initialize domid to 0 in libxl__create_stubdom

[Ian Jackson <Ian.Jackson@xxxxxxxxxxxxx>]

Stefano Stabellini writes ("Re: [Xen-devel] Re: [PATCH] libxl: initialize domid 
to 0 in libxl__create_stubdom"):
> On Fri, 17 Jun 2011, Ian Jackson wrote:
> >    char *something = NULL;
> >    uint32_t domid = -1;
> > 
> >    ...
> >    something = allocate();
> >    if (!something) goto error_exit;
> >    ...
> >    ret = libxl__domain_make(&domid);
> >    if (ret) goto error_exit;
> >    ...
> > 
> >    return successfully somehow;
> > 
> >   error_exit:
> >    free(something);
> >    if (libxl_domid_valid_guest(domid))
> >        libxl_domain_destroy(domid);
...
> If we decide to make domid an output parameter only, then
> 
> uint32_t domid;
> 
> isn't a bug anymore.

Look at the code above.  Without the initialisation, if allocate()
returns NULL, it calls:
  libxl_domid_valid_guest(UNDEFINED)
If it's an output parameter only then neither valgrind nor the
compiler will ever spot this bug unless allocate() actually fails.

The arrangement with the caller initialising and the check in
libxl__domain_make is there to support this error-handling paradim
(which is a good paradigm because it means you don't have to carefully
match up allocations with frees on every error exit path), and which
tries to give us a chance of spotting missing initialisation bugs.

> Have you read http://marc.info/?l=xen-devel&m=130763064528133?

Yes.  I don't agree.

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel