When allocating the skb within netlink_sendmsg, with certain supplied
len arguments, extra bytes are allocated at the end of the data buffer,
due to SKB_DATA_ALIGN giving a larger size within __alloc_skb for
alignment reasons. This means that after using skb_put with the same
len value and then copying data into the skb, the skb tail area is
non-zero in size and contains uninitialised bytes. Wiping this area
(if it exists) fixes a KMSAN-found uninit-value bug reported by syzbot at:
https://syzkaller.appspot.com/bug?id=3e63bcec536b7136b54c72e06adeb87dc6519f69
Reported-by: syzbot+a73d24a22eeeebe5f244@syzkaller.appspotmail.com
Signed-off-by: Phillip Potter <phil@philpotter.co.uk>
---
net/netlink/af_netlink.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 3a62f97acf39..e54321b63f98 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -1914,6 +1914,9 @@ static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
goto out;
}
+ if (skb->end - skb->tail)
+ memset(skb_tail_pointer(skb), 0, skb->end - skb->tail);
+
err = security_netlink_send(sk, skb);
if (err) {
kfree_skb(skb);--
2.30.2
Phillip Potter [off-list ref] wrote:
When allocating the skb within netlink_sendmsg, with certain supplied
len arguments, extra bytes are allocated at the end of the data buffer,
due to SKB_DATA_ALIGN giving a larger size within __alloc_skb for
alignment reasons. This means that after using skb_put with the same
len value and then copying data into the skb, the skb tail area is
non-zero in size and contains uninitialised bytes. Wiping this area
(if it exists) fixes a KMSAN-found uninit-value bug reported by syzbot at:
https://syzkaller.appspot.com/bug?id=3e63bcec536b7136b54c72e06adeb87dc6519f69
This patch papers over the real bug.
Please fix TIPC instead.
Incomplete patch as a starting point:
diff --git a/net/tipc/node.c b/net/tipc/node.c
--- a/net/tipc/node.c
+++ b/net/tipc/node.c
@@ -2481,7 +2481,6 @@ int tipc_nl_node_get_link(struct sk_buff *skb, struct genl_info *info)
struct net *net = genl_info_net(info);
struct nlattr *attrs[TIPC_NLA_LINK_MAX + 1];
struct tipc_nl_msg msg;
- char *name;
int err;
msg.portid = info->snd_portid;
@@ -2499,13 +2498,11 @@ int tipc_nl_node_get_link(struct sk_buff *skb, struct genl_info *info)
if (!attrs[TIPC_NLA_LINK_NAME])
return -EINVAL;
- name = nla_data(attrs[TIPC_NLA_LINK_NAME]);
-
msg.skb = nlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
if (!msg.skb)
return -ENOMEM;
- if (strcmp(name, tipc_bclink_name) == 0) {
+ if (nla_strcmp(attrs[TIPC_NLA_LINK_NAME], tipc_bclink_name) == 0) {
err = tipc_nl_add_bc_link(net, &msg, tipc_net(net)->bcl);
if (err)
goto err_free;
You will also need to change tipc_node_find_by_name() to pass the nla
attr.
Alternatively TIPC_NLA_LINK_NAME policy can be changed:
diff --git a/net/tipc/netlink.c b/net/tipc/netlink.c
--- a/net/tipc/netlink.c
+++ b/net/tipc/netlink.c
@@ -88,7 +88,7 @@ const struct nla_policy tipc_nl_net_policy[TIPC_NLA_NET_MAX + 1] = {
const struct nla_policy tipc_nl_link_policy[TIPC_NLA_LINK_MAX + 1] = {
[TIPC_NLA_LINK_UNSPEC] = { .type = NLA_UNSPEC },
- [TIPC_NLA_LINK_NAME] = { .type = NLA_STRING,
+ [TIPC_NLA_LINK_NAME] = { .type = NLA_NUL_STRING,
... which makes it safe to treat the raw attribute payload as a c-string,
but this might break existing userspace applications.
Its probably a good idea to audit all NLA_STRING attributes in tipc for
similar problems.
On Sun, May 09, 2021 at 03:10:51PM +0200, Florian Westphal wrote:
quoted hunk
Phillip Potter [off-list ref] wrote:
quoted
When allocating the skb within netlink_sendmsg, with certain supplied
len arguments, extra bytes are allocated at the end of the data buffer,
due to SKB_DATA_ALIGN giving a larger size within __alloc_skb for
alignment reasons. This means that after using skb_put with the same
len value and then copying data into the skb, the skb tail area is
non-zero in size and contains uninitialised bytes. Wiping this area
(if it exists) fixes a KMSAN-found uninit-value bug reported by syzbot at:
https://syzkaller.appspot.com/bug?id=3e63bcec536b7136b54c72e06adeb87dc6519f69
This patch papers over the real bug.
Please fix TIPC instead.
Incomplete patch as a starting point:
diff --git a/net/tipc/node.c b/net/tipc/node.c
--- a/net/tipc/node.c
+++ b/net/tipc/node.c
@@ -2481,7 +2481,6 @@ int tipc_nl_node_get_link(struct sk_buff *skb, struct genl_info *info)
struct net *net = genl_info_net(info);
struct nlattr *attrs[TIPC_NLA_LINK_MAX + 1];
struct tipc_nl_msg msg;
- char *name;
int err;
msg.portid = info->snd_portid;
@@ -2499,13 +2498,11 @@ int tipc_nl_node_get_link(struct sk_buff *skb, struct genl_info *info)
if (!attrs[TIPC_NLA_LINK_NAME])
return -EINVAL;
- name = nla_data(attrs[TIPC_NLA_LINK_NAME]);
-
msg.skb = nlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
if (!msg.skb)
return -ENOMEM;
- if (strcmp(name, tipc_bclink_name) == 0) {
+ if (nla_strcmp(attrs[TIPC_NLA_LINK_NAME], tipc_bclink_name) == 0) {
err = tipc_nl_add_bc_link(net, &msg, tipc_net(net)->bcl);
if (err)
goto err_free;
You will also need to change tipc_node_find_by_name() to pass the nla
attr.
Alternatively TIPC_NLA_LINK_NAME policy can be changed:
diff --git a/net/tipc/netlink.c b/net/tipc/netlink.c
--- a/net/tipc/netlink.c
+++ b/net/tipc/netlink.c
@@ -88,7 +88,7 @@ const struct nla_policy tipc_nl_net_policy[TIPC_NLA_NET_MAX + 1] = {
const struct nla_policy tipc_nl_link_policy[TIPC_NLA_LINK_MAX + 1] = {
[TIPC_NLA_LINK_UNSPEC] = { .type = NLA_UNSPEC },
- [TIPC_NLA_LINK_NAME] = { .type = NLA_STRING,
+ [TIPC_NLA_LINK_NAME] = { .type = NLA_NUL_STRING,
... which makes it safe to treat the raw attribute payload as a c-string,
but this might break existing userspace applications.
Its probably a good idea to audit all NLA_STRING attributes in tipc for
similar problems.
Dear Florian,
Thank you for your feedback and code + suggestions, I will take a look at this over
the next few days and then resubmit.
Regards,
Phil