segfault in http-push

Hi,

With 1.4.5.2, I see this whenever I try to push a new branch over http:

$ git-push hash long_ids
Fetching remote heads...
  refs/
  refs/heads/
  refs/tags/
  refs/remotes/
updating 'refs/heads/long_ids'
  from 0000000000000000000000000000000000000000
  to   b24938890f04f3ba42b88f56f3ff2bbe0fda6629
    sending 10200 objects
    done
Updating remote server info
error: http-push died with strange error
error: failed to push to 'https://user@machine/foo.git/'

Here's a backtrace:

#0  0xb7cbd583 in strlen () from /lib/tls/i686/cmov/libc.so.6
#1  0x08053b5a in start_put (request=0xdce9ad8) at http-push.c:520
#2  0x080550d7 in fill_active_slot (unused=0x0) at http-push.c:810
#3  0x0804da9d in fill_active_slots () at http.c:428
#4  0x0804dc8b in step_active_slots () at http.c:454
#5  0x0804dd0b in run_active_slot (slot=0x8e70148) at http.c:474
#6  0x080502c9 in remote_ls (path=0x808bdbd "refs/", flags=5, 
    userFunc=0x804f2b0 <add_remote_info_ref>, userData=0xbfa04434)
    at http-push.c:1467
#7  0x08050552 in update_remote_info_refs (lock=0x80e66c0) at http-push.c:1942
#8  0x080534ee in main (argc=3, argv=Cannot access memory at address 0x4
) at http-push.c:2421

Error is at: 520 request->url = xmalloc(strlen(remote->url) + 521 strlen(request->lock->token) + 51);

(gdb) p request->lock $1 = (struct remote_lock *) 0x80e7d40 (gdb) p request->lock->token $2 = 0x65707954 <Address 0x65707954 out of bounds>

If I go fix up the remote repository I can usually get it to work. Any thoughts?

Hi,

On Sun, 17 Feb 2008, Bob Copeland wrote:

I can only suspect that you mean 1.5.4.2, not 1.4.5.2 (which I could not find).

Alas, a fix I provided for exactly this case did not make it into master yet, let alone into any released version. Can you please test the current "next"?

FWIW, the fix is add8e8ce(http-push: avoid invalid memory accesses).

Thanks, Dscho

On Sun, Feb 17, 2008 at 11:23:38PM +0000, Johannes Schindelin wrote:

Just so...

Yep, next fixes it for me, thanks!

Hi,

On Sun, 17 Feb 2008, Bob Copeland wrote:

Thanks for testing!

Ciao, Dscho "who is pleased that we fix some bugs before they are reported"

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

Wonderful.

I take this as a go ahead to merge js/maint-http-push topic to 'maint' before we cut 1.5.4.3.

Thanks both of you, especially Dscho.