Re: ipop3d (x2) / pine (x2) / Linux kernel (x2) / Midnight

[okir () MONAD SWB DE (Olaf Kirch)]

--qDbXVdCdHGoSgWSk
Content-Type: text/plain; charset=us-ascii

--------
On Thu, 08 Apr 1999 18:16:48 PDT, Mark Crispin wrote:
> The maximum exposure is that one byte of stack frame may be set to zero,
> because of a "buf[x] = 0" for a stack buffer of length x.
[snip]
> It is *NOT* a security bug in the normal sense.

I haven't done a thorough analysis, but let me make a general remark.
Even a single byte overflow can have deadly consequences.

When compiling imap on an intel box (gcc -O2), the stack frame of the
procedure we're talking about looks like this:

        word            return address
        word            frame pointer of calling process
        1024 bytes      buffer

Writing a NUL byte past the end of the buffer will clobber the first byte
of the frame pointer (which is its LSB, since we're little-endian). This
shifts down the stack frame of the calling procedure by an amount between
0 and 252 bytes. When the calling procedure returns, it will pick up a
bogus return address, which _may_ be just inside a data buffer filled
with client-supplied data.

This is not just a theoretic possibility. I've seen this happen once with
the linux user space nfsd. It gave me enough of a headache debugging it
that I feel no urge to find out whether this applies to imapd as well...

> Now, we'll talk about the 20% that is fact.  Yes, it is possible to write
> a negative process ID in the lock file.  This requires that the attacker
> have shell access; it can't be mounted remotely.  It also requires that
> the attacker have a program running at the time that the victim opens his
> mail file.

There's a feature in imap that's rarely if ever mentioned which lets
you configure the lock protection for mailbox locks. In /etc/client.cf,
you can do `set lock-protection 0600'. The source says it's totally
unsupported and may go away anytime, though.

Combined with the appended patch this may make both Mark and the people
who dislike his lock files happy. (AFAIK a very similar patch is in
4.5 already).

Olaf Kirch
--
Olaf Kirch         |  --- o --- Nous sommes du soleil we love when we play
okir () monad swb de  |    / | \   sol.dhoop.naytheet.ah kin.ir.samse.qurax

--qDbXVdCdHGoSgWSk
Content-Type: application/x-patch
Content-Disposition: attachment; filename="imap-4.4.patch"

diff -ur imap-4.4.orig/src/osdep/unix/env_unix.c imap-4.4/src/osdep/unix/env_unix.c
--- imap-4.4.orig/src/osdep/unix/env_unix.c     Wed Aug 26 20:04:50 1998
+++ imap-4.4/src/osdep/unix/env_unix.c  Fri Apr  9 10:21:39 1999
@@ -709,19 +709,52 @@

 int lock_work (char *lock)
 {
-  int fd;
-  long nlinks = chk_notsymlink (lock);
-  if (!nlinks) return -1;      /* fail if symbolic link */
-  if (nlinks > 1) {            /* extra hard link to the file? */
-    mm_log ("SECURITY ALERT: hard link to lock name!",ERROR);
-    syslog (LOG_CRIT,"SECURITY PROBLEM: lock file %s has a hard link",lock);
-    return -1;
+  struct stat  sbuf, fsbuf;
+  int          prot, fd;
+
+  prot = (int) mail_parameters (NIL,GET_LOCKPROTECTION,NIL);
+  if (lstat (lock, &sbuf) < 0) {
+    fd = open(lock, O_RDWR|O_CREAT|O_EXCL, prot);
+  } else if ((fd = open(lock, O_RDWR)) >= 0) {
+    if (fstat(fd, &fsbuf) < 0) {
+       syslog(LOG_NOTICE, "unable to fstat %s?!", lock);
+       goto fail;
+    }
+    if (sbuf.st_dev != fsbuf.st_dev || sbuf.st_ino != fsbuf.st_ino) {
+       mm_log ("SECURITY ALERT: Someone is playing symlink "
+               "tricks with lock file", ERROR);
+       syslog (LOG_CRIT,
+               "SECURITY PROBLEM: Someone is playing symlink "
+               "tricks with lock file %s", lock);
+       goto fail;
+    }
+    if (fsbuf.st_nlink > 1) {
+       mm_log ("SECURITY ALERT: hard link to lock file", ERROR);
+       syslog (LOG_CRIT,
+               "SECURITY PROBLEM: hard link to lock file %s",
+               lock);
+       goto fail;
+    }
+    if (fsbuf.st_uid != getuid() || fsbuf.st_gid != getgid()) {
+       mm_log ("SECURITY ALERT: bad owner for lock file", ERROR);
+       syslog (LOG_CRIT,
+               "SECURITY PROBLEM: bad owner for lock file %s (uid %d gid %d)",
+               lock, fsbuf.st_uid, fsbuf.st_gid);
+       goto fail;
+    }
+    if (fsbuf.st_mode & ~prot) {
+       mm_log ("SECURITY ALERT: bad permissions on lock file", ERROR);
+       syslog (LOG_CRIT,
+               "SECURITY PROBLEM: bad permissions on file %s "
+               "(mode 0%o, should be 0%o)",
+               lock, fsbuf.st_mode & O_ACCMODE, prot);
+       goto fail;
+    }
   }
-  if ((fd = open (lock,O_RDWR | O_CREAT | ((nlinks < 0) ? O_EXCL : NIL),
-                 (int) mail_parameters (NIL,GET_LOCKPROTECTION,NIL))) >= 0)
-                               /* make sure mode OK (don't use fchmod()) */
-    chmod (lock,(int) mail_parameters (NIL,GET_LOCKPROTECTION,NIL));
   return fd;
+fail:
+  close(fd);
+  return -1;
 }



--qDbXVdCdHGoSgWSk--