Bug 155771 – Kernel: get_fdb_entries() integer overflow (CVE-2006-5751)

• Status: RESOLVED FIXED
• Product: Gentoo Security

[Description Sune Kloppenborg Jeppesen (RETIRED) 2006-11-20 07:27:14 UTC]

This issue is EMBARGOED and the proposed unembargo date is Monday, 4th
of December 2006 at 14:00 UTC. It has been assigned CVE-2006-5751.

The vulnerability description is as follows:

Synopsis:  Linux kernel get_fdb_entries() integer overflow
Product:   Linux kernel
Version:   2.6.7 - 2.6.18.3
Vendor:    http://www.kernel.org/
Severity:  Important
Author:    Eugene Teo <[e-mail]>
Date:      November 19, 2006
CVE-ID:    CVE-2006-5751

Issue:
======

A potential security vulnerability has been found in the Linux kernel in the
get_fdb_entries() function code. Successful exploitation may potentially allow
execution of arbitrary code with escalated privileges.

Details:
========

The vulnerable code resides in net/bridge/br_ioctl.c:

  56 static _nt get_fdb_entries(struct net_bridge *br, void __user *userbuf,
  57                            unsigned long maxnum, unsigned long offset)
  58 {
  59         int num;
  60         void *buf;
  61         size_t size = maxnum * sizeof(struct __fdb_entry);
  62
  63         if (size > PAGE_SIZE) {
  64                 size = PAGE_SIZE;
  65                 maxnum = PAGE_SIZE/sizeof(struct __fdb_entry);
  66         }
  67
  68         buf = kmalloc(size, GFP_USER);
  69         if (!buf)
  70                 return -ENOMEM;
  71
  72         num = br_fdb_fillbuf(br, buf, maxnum, offset);

By supplying a sufficiently large value to maxnum, we can control the amount
of memory to allocate to buf (i.e. 32 bytes).

net/bridge/br_fdb.c:

219 int br_fdb_fillbuf(struct net_bridge *br, void *buf,
220                    unsigned long maxnum, unsigned long skip)
221 {
222         struct __fdb_entry *fe = buf;
/* ... */
227         memset(buf, 0, maxnum*sizeof(struct __fdb_entry));
228
229         rcu_read_lock();
230         for (i = 0; i < BR_HASH_SIZE; i++) {
231                 hlist_for_each_entry_rcu(f, h, &br->hash[i], hlist) {
232                         if (num >= maxnum)
233                                 goto out;
234
235                         if (has_expired(br, f))
236                                 continue;
237
238                         if (skip) {
239                                 --skip;
240                                 continue;
241                         }
242
243                         /* convert from internal format to API */
244                         memcpy(fe->mac_addr, f->addr.addr, ETH_ALEN);
245                         fe->port_no = f->dst->port_no;
246                         fe->is_local = f->is_local;
247                         if (!f->is_static)
248                                 fe->ageing_timer_value = 
jiffies_to_clock_t(jiffies - f->ageing_timer);
249                         ++fe;
250                         ++num;
251                 }
252         }

because fe = buf, and buf can be allocated with only 32 bytes, if one bridge has
more than two interfaces added, memcpy will be able to overwrite other slab
objects, which can be exploited to execute arbitrary code.

Credits:
========

Thanks a1rsupp1y for performing further research together.

POC:
====

        unsigned long args[4] = { BRCTL_GET_FDB_ENTRIES,
                      (unsigned long) fe,
                       LARGE_VALUE, offset};
        struct ifreq ifr;
        int retries = 0;

        strncpy(ifr.ifr_name, bridge, IFNAMSIZ);
        ifr.ifr_data = (char *) args;
        n = ioctl(br_socket_fd, SIOCDEVPRIVATE, &ifr);

[Comment 1 Harlan Lieberman-Berg (RETIRED) 2006-12-08 18:43:36 UTC]

Past the release date.

[Comment 2 Daniel Drake (RETIRED) 2007-01-01 20:21:54 UTC]

Fixed in Linux 2.6.18.4

[Comment 3 Harlan Lieberman-Berg (RETIRED) 2007-05-21 23:22:26 UTC]

Not affected in ~arch.