On 04/27/2011 12:24 AM, Jim Meyering wrote: >> fdutimensat-tests still pulls in xalloc, via utimecmp (which >> calls xalloc_die if it can't manage a hash table). So this >> patch isn't quite perfect yet, but getting closer. Any ideas >> on whether it would be possible to make utimecmp work without >> xalloc? > > Looks feasible. > The hash table is solely to avoid incurring the cost of calculating > the time stamp resolution for every call. The alternative is simply > to incur the cost when hash table initialization or insertion fails.
Even worse, I just noticed that there is a NULL dereference on low memory conditions in utimecmp. The code tries to allocate ht, but then blindly uses ht even if allocation failed. I definitely agree that utimecmp's use of a hash-table is merely a caching optimization, and therefore memory failure should not be fatal. I'll have a patch shortly... -- Eric Blake ebl...@redhat.com +1-801-349-2682 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
