[PATCH 3.2 041/140] net: ipv4: fix for a race condition in raw_sendmsg

From: Ben Hutchings
Date: Wed Feb 28 2018 - 10:26:32 EST


3.2.100-rc1 review patch. If anyone has any objections, please let me know.

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

From: Mohamed Ghannam <simo.ghannam@xxxxxxxxx>

commit 8f659a03a0ba9289b9aeb9b4470e6fb263d6f483 upstream.

inet->hdrincl is racy, and could lead to uninitialized stack pointer
usage, so its value should be read only once.

Fixes: c008ba5bdc9f ("ipv4: Avoid reading user iov twice after raw_probe_proto_opt")
Signed-off-by: Mohamed Ghannam <simo.ghannam@xxxxxxxxx>
Reviewed-by: Eric Dumazet <edumazet@xxxxxxxxxx>
Signed-off-by: David S. Miller <davem@xxxxxxxxxxxxx>
[bwh: Backported to 3.2:
- flowi4 flags don't depend on hdrincl
- Adjust context]
Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx>
---
--- a/net/ipv4/raw.c
+++ b/net/ipv4/raw.c
@@ -484,11 +484,16 @@ static int raw_sendmsg(struct kiocb *ioc
int err;
struct ip_options_data opt_copy;
struct raw_frag_vec rfv;
+ int hdrincl;

err = -EMSGSIZE;
if (len > 0xFFFF)
goto out;

+ /* hdrincl should be READ_ONCE(inet->hdrincl)
+ * but READ_ONCE() doesn't work with bit fields
+ */
+ hdrincl = inet->hdrincl;
/*
* Check the flags.
*/
@@ -564,7 +569,7 @@ static int raw_sendmsg(struct kiocb *ioc
/* Linux does not mangle headers on raw sockets,
* so that IP options + IP_HDRINCL is non-sense.
*/
- if (inet->hdrincl)
+ if (hdrincl)
goto done;
if (ipc.opt->opt.srr) {
if (!daddr)
@@ -585,11 +590,11 @@ static int raw_sendmsg(struct kiocb *ioc

flowi4_init_output(&fl4, ipc.oif, sk->sk_mark, tos,
RT_SCOPE_UNIVERSE,
- inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
+ hdrincl ? IPPROTO_RAW : sk->sk_protocol,
inet_sk_flowi_flags(sk) | FLOWI_FLAG_CAN_SLEEP,
daddr, saddr, 0, 0);

- if (!inet->hdrincl) {
+ if (!hdrincl) {
rfv.iov = msg->msg_iov;
rfv.hlen = 0;

@@ -614,7 +619,7 @@ static int raw_sendmsg(struct kiocb *ioc
goto do_confirm;
back_from_confirm:

- if (inet->hdrincl)
+ if (hdrincl)
err = raw_send_hdrinc(sk, &fl4, msg->msg_iov, len,
&rt, msg->msg_flags);