Audit alloc failure in theora xiph codec

RESOLVED FIXED in mozilla13

Status

()

--
critical
RESOLVED FIXED
10 years ago
7 years ago

People

(Reporter: timeless, Assigned: derf)

Tracking

({crash})

Trunk
mozilla13
crash
Points:
---

Firefox Tracking Flags

(Not tracked)

Details

(Whiteboard: [sg:audit], URL)

Attachments

(1 attachment, 1 obsolete attachment)

Whiteboard: [sg:investigate]

Updated

9 years ago
Whiteboard: [sg:investigate] → [sg:audit]

Comment 1

9 years ago
Tim, do you know if these have been addressed in libtheora?
(Assignee)

Comment 2

9 years ago
I can't go through and re-check all of the links above to double-check, because the files no longer exist (and wouldn't be the same version if they did). I did a quick re-audit, and it appears that while most of the malloc/calloc/realloc usage got NULL checks added, one hunk was dropped. The following is upstream in the 1.2 development branch, but was not in the 1.1 release:

Index: decinfo.c
===================================================================
--- decinfo.c   (revision 17275)
+++ decinfo.c   (working copy)
@@ -128,6 +128,10 @@
    _tc->comments*sizeof(_tc->comment_lengths[0]));
   _tc->user_comments=(char **)_ogg_malloc(
    _tc->comments*sizeof(_tc->user_comments[0]));
+  if(_tc->comment_lengths==NULL||_tc->user_comments==NULL){
+    _tc->comments=0;
+    return TH_EFAULT;
+  }
   for(i=0;i<_tc->comments;i++){
     len=oc_unpack_length(_opb);
     if(len<0||len>oc_pack_bytes_left(_opb)){
(Assignee)

Comment 3

7 years ago
Josh asked me if we could close this bug. It had been long enough that I had forgotten what exactly had been reviewed, so I reviewed things again, and found the following problem @ media/libtheora/lib/state.c:586:

  ref_frame_data=oc_aligned_malloc(ref_frame_data_sz,16);
  frag_buf_offs=_state->frag_buf_offs=
   _ogg_malloc(_state->nfrags*sizeof(*frag_buf_offs));
  if(ref_frame_data==NULL||frag_buf_offs==NULL){
    _ogg_free(frag_buf_offs);
    _ogg_free(ref_frame_data);
    return TH_EFAULT;
  }

I.e., ref_frame_data was being allocated with oc_aligned_malloc(), and freed with _ogg_free() instead of oc_aligned_free() on error. This doesn't work (the aligned block is guaranteed to start after a block of memory returned by _ogg_malloc(), because it stores the alignment offset there).

This would only cause a problem if there was just enough memory to satisfy the reference frame allocation (just over 4.5 or 9 bytes per pixel) but not enough for the fragment buffer offets (1/16 or 1/8th byte per pixel). The size of the allocations is attacker-controlled, but nothing is stored in the buffers, so I don't see how this can be exploited, but it could cause a crash with libc's that don't validate pointers before trying to free them (e.g., glibc).

This was introduced in upstream commit r17563 and local commit 57091:af89c96d (Nov. 8th 2010), so it should affect everything from Firefox 4 onwards.

I've fixed this upstream in r18219. Patch for m-c forthcoming.
(Assignee)

Comment 4

7 years ago
Assignee: nobody → tterribe
Status: NEW → ASSIGNED
Attachment #602536 - Flags: review?(kinetik)
Attachment #602536 - Flags: review?(kinetik) → review+
(Assignee)

Comment 5

7 years ago
This version drops a couple of extraneous hunks (which I included in the upstream commit, but in order to keep this fix to a minimum, meant to exclude here). I'd dropped them from the update.sh patch, but not dropped the actual changes to the source files.

Greenish on try: 
https://tbpl.mozilla.org/?tree=Try&rev=22bec58b158c

Carrying forward r=kinetik
Attachment #602536 - Attachment is obsolete: true
Attachment #602798 - Flags: review+
(Assignee)

Comment 7

7 years ago
https://hg.mozilla.org/mozilla-central/rev/129ab3b6dff8
Status: ASSIGNED → RESOLVED
Last Resolved: 7 years ago
Resolution: --- → FIXED
(Assignee)

Comment 8

7 years ago
I don't think there's any reason to keep this closed. This might be a potential DoS, but because we place limits on the maximum size of a video, the only way it can be triggered is by putting enough <video> tags on a page to actually run the system out of RAM, which would be a problem all by itself.

Updated

7 years ago
Group: core-security
You need to log in before you can comment on or make changes to this bug.