Blame SOURCES/kvm-virtio-net-fix-buffer-overflow-on-invalid-state-load.patch

eb5a2f
From fbae1c1a80fc0c1495c86695f5f48ee1003c2a1a Mon Sep 17 00:00:00 2001
eb5a2f
From: Michael S. Tsirkin <mst@redhat.com>
eb5a2f
Date: Wed, 14 May 2014 08:07:39 +0200
eb5a2f
Subject: [PATCH 01/30] virtio-net: fix buffer overflow on invalid state load
eb5a2f
eb5a2f
RH-Author: Michael S. Tsirkin <mst@redhat.com>
eb5a2f
Message-id: <1400054498-4366-1-git-send-email-mst@redhat.com>
eb5a2f
Patchwork-id: 58839
eb5a2f
O-Subject: [PATCH qemu-kvm RHEL7.0] virtio-net: fix buffer overflow on invalid state load
eb5a2f
Bugzilla: 1095677
eb5a2f
RH-Acked-by: Dr. David Alan Gilbert (git) <dgilbert@redhat.com>
eb5a2f
RH-Acked-by: Xiao Wang <jasowang@redhat.com>
eb5a2f
RH-Acked-by: Amos Kong <akong@redhat.com>
eb5a2f
eb5a2f
CVE-2013-4148 QEMU 1.0 integer conversion in
eb5a2f
virtio_net_load()@hw/net/virtio-net.c
eb5a2f
eb5a2f
Deals with loading a corrupted savevm image.
eb5a2f
eb5a2f
>         n->mac_table.in_use = qemu_get_be32(f);
eb5a2f
eb5a2f
in_use is int so it can get negative when assigned 32bit unsigned value.
eb5a2f
eb5a2f
>         /* MAC_TABLE_ENTRIES may be different from the saved image */
eb5a2f
>         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
eb5a2f
eb5a2f
passing this check ^^^
eb5a2f
eb5a2f
>             qemu_get_buffer(f, n->mac_table.macs,
eb5a2f
>                             n->mac_table.in_use * ETH_ALEN);
eb5a2f
eb5a2f
with good in_use value, "n->mac_table.in_use * ETH_ALEN" can get
eb5a2f
positive and bigger than mac_table.macs. For example 0x81000000
eb5a2f
satisfies this condition when ETH_ALEN is 6.
eb5a2f
eb5a2f
Fix it by making the value unsigned.
eb5a2f
For consistency, change first_multi as well.
eb5a2f
eb5a2f
Note: all call sites were audited to confirm that
eb5a2f
making them unsigned didn't cause any issues:
eb5a2f
it turns out we actually never do math on them,
eb5a2f
so it's easy to validate because both values are
eb5a2f
always <= MAC_TABLE_ENTRIES.
eb5a2f
eb5a2f
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
eb5a2f
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
eb5a2f
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
eb5a2f
Signed-off-by: Juan Quintela <quintela@redhat.com>
eb5a2f
(cherry picked from commit 71f7fe48e10a8437c9d42d859389f37157f59980)
eb5a2f
eb5a2f
Bugzilla: 1095677
eb5a2f
Tested: on developer's box
eb5a2f
Brew build: http://brewweb.devel.redhat.com/brew/taskinfo?taskID=7450401
eb5a2f
---
eb5a2f
 include/hw/virtio/virtio-net.h | 4 ++--
eb5a2f
 1 file changed, 2 insertions(+), 2 deletions(-)
eb5a2f
eb5a2f
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
eb5a2f
---
eb5a2f
 include/hw/virtio/virtio-net.h |    4 ++--
eb5a2f
 1 files changed, 2 insertions(+), 2 deletions(-)
eb5a2f
eb5a2f
diff --git a/include/hw/virtio/virtio-net.h b/include/hw/virtio/virtio-net.h
eb5a2f
index 75723a8..a02fc50 100644
eb5a2f
--- a/include/hw/virtio/virtio-net.h
eb5a2f
+++ b/include/hw/virtio/virtio-net.h
eb5a2f
@@ -174,8 +174,8 @@ typedef struct VirtIONet {
eb5a2f
     uint8_t nobcast;
eb5a2f
     uint8_t vhost_started;
eb5a2f
     struct {
eb5a2f
-        int in_use;
eb5a2f
-        int first_multi;
eb5a2f
+        uint32_t in_use;
eb5a2f
+        uint32_t first_multi;
eb5a2f
         uint8_t multi_overflow;
eb5a2f
         uint8_t uni_overflow;
eb5a2f
         uint8_t *macs;
eb5a2f
-- 
eb5a2f
1.7.1
eb5a2f