Re: [PATCH v4 3/3] upload-archive: use start_command instead of fork

On Tue, Nov 15, 2011 at 6:37 PM, Jeff King [off-list ref] wrote:

On Tue, Nov 15, 2011 at 01:11:46PM +0100, Thomas Rast wrote:

quoted

But after a closer look I think this patch just prodded it enough to
unearth long-existing undefined behaviour: prepare_argv() summarizes
to something like

static void prepare_argv(const char **sent_argv, const char **argv)
{
      char *p, buf[4096];

      for (p = buf;;) {
              len = packet_read_line(0, p, (buf + sizeof buf) - p);
              /* ... p always points into buf ... */
              sent_argv[sent_argc++] = p;
              p += len;
              *p++ = 0;
      }
      sent_argv[sent_argc] = NULL;
}

The code appears to have looked like this ever since the addition of
that file back in 39345a2 (Add git-upload-archive, 2006-09-07).  So
the elements of sent_argv have apparently always pointed into the
stack-allocated 'buf'.

Oh, yikes. That is definitely the problem, but it does come from
c09cd77e. The prepare_argv function used to be "run_upload_archive", and
it would prepare argv on the stack, call into write_archive with it, and
then return; nobody else cares about the value afterwards.

Erik's patch converts it into a function that writes the new argv into a
parameter and returns, and the now-invalid stack-allocated memory is
used by the calling function.

Outch. Thanks for spotting.

quoted

A quick band-aid would be to heap-allocate it instead:

That works.  An even shorter band-aid is to mark it as "static".

Hmm, I seem to remember spotting it myself at some point and fixing it
by marking it as static. I guess I must have forgot to push it...

I think the code would be more readable if it just used the new
argv_array.

Oooh, nice. The whole argv_array slipped past me, I like it!