https://sourceware.org/bugzilla/show_bug.cgi?id=29474
--- Comment #2 from Martin Liska <mliska at suse dot cz> ---
So it's the prefetching feature as documented here:
// 3) extract some number of prefetched entries (just into fdcache)
...
if (r != 0) // stage 3
{
// NB: now we know we have a complete reusable file; make fdcache
// responsible for unlinking it later.
obatched(clog) << "INTERN " << b_source1 << " filename=" << fn <<
endl;
fdcache.intern(b_source0, fn,
tmppath, archive_entry_size(e),
false); // prefetched ones go to the prefetch cache
prefetch_count --;
close (fd); // we're not saving this fd to make a mhd-response from!
continue;
}
So what's bad is that
handle_buildid_r_match begins with:
// check for a match in the fdcache first
int fd = fdcache.lookup(b_source0, b_source1);
while (fd >= 0) // got one!; NB: this is really an if() with a possible
branch out to the end
{
...
}
which was false, but happens to be true once another thread inters the file :/
--
You are receiving this mail because:
You are on the CC list for the bug.
