net: icmp: fix data-race in cmp_global_allow()

This code reads two global variables without protection
of a lock. We need READ_ONCE()/WRITE_ONCE() pairs to
avoid load/store-tearing and better document the intent.

KCSAN reported :
BUG: KCSAN: data-race in icmp_global_allow / icmp_global_allow

read to 0xffffffff861a8014 of 4 bytes by task 11201 on cpu 0:
 icmp_global_allow+0x36/0x1b0 net/ipv4/icmp.c:254
 icmpv6_global_allow net/ipv6/icmp.c:184 [inline]
 icmpv6_global_allow net/ipv6/icmp.c:179 [inline]
 icmp6_send+0x493/0x1140 net/ipv6/icmp.c:514
 icmpv6_send+0x71/0xb0 net/ipv6/ip6_icmp.c:43
 ip6_link_failure+0x43/0x180 net/ipv6/route.c:2640
 dst_link_failure include/net/dst.h:419 [inline]
 vti_xmit net/ipv4/ip_vti.c:243 [inline]
 vti_tunnel_xmit+0x27f/0xa50 net/ipv4/ip_vti.c:279
 __netdev_start_xmit include/linux/netdevice.h:4420 [inline]
 netdev_start_xmit include/linux/netdevice.h:4434 [inline]
 xmit_one net/core/dev.c:3280 [inline]
 dev_hard_start_xmit+0xef/0x430 net/core/dev.c:3296
 __dev_queue_xmit+0x14c9/0x1b60 net/core/dev.c:3873
 dev_queue_xmit+0x21/0x30 net/core/dev.c:3906
 neigh_direct_output+0x1f/0x30 net/core/neighbour.c:1530
 neigh_output include/net/neighbour.h:511 [inline]
 ip6_finish_output2+0x7a6/0xec0 net/ipv6/ip6_output.c:116
 __ip6_finish_output net/ipv6/ip6_output.c:142 [inline]
 __ip6_finish_output+0x2d7/0x330 net/ipv6/ip6_output.c:127
 ip6_finish_output+0x41/0x160 net/ipv6/ip6_output.c:152
 NF_HOOK_COND include/linux/netfilter.h:294 [inline]
 ip6_output+0xf2/0x280 net/ipv6/ip6_output.c:175
 dst_output include/net/dst.h:436 [inline]
 ip6_local_out+0x74/0x90 net/ipv6/output_core.c:179

write to 0xffffffff861a8014 of 4 bytes by task 11183 on cpu 1:
 icmp_global_allow+0x174/0x1b0 net/ipv4/icmp.c:272
 icmpv6_global_allow net/ipv6/icmp.c:184 [inline]
 icmpv6_global_allow net/ipv6/icmp.c:179 [inline]
 icmp6_send+0x493/0x1140 net/ipv6/icmp.c:514
 icmpv6_send+0x71/0xb0 net/ipv6/ip6_icmp.c:43
 ip6_link_failure+0x43/0x180 net/ipv6/route.c:2640
 dst_link_failure include/net/dst.h:419 [inline]
 vti_xmit net/ipv4/ip_vti.c:243 [inline]
 vti_tunnel_xmit+0x27f/0xa50 net/ipv4/ip_vti.c:279
 __netdev_start_xmit include/linux/netdevice.h:4420 [inline]
 netdev_start_xmit include/linux/netdevice.h:4434 [inline]
 xmit_one net/core/dev.c:3280 [inline]
 dev_hard_start_xmit+0xef/0x430 net/core/dev.c:3296
 __dev_queue_xmit+0x14c9/0x1b60 net/core/dev.c:3873
 dev_queue_xmit+0x21/0x30 net/core/dev.c:3906
 neigh_direct_output+0x1f/0x30 net/core/neighbour.c:1530
 neigh_output include/net/neighbour.h:511 [inline]
 ip6_finish_output2+0x7a6/0xec0 net/ipv6/ip6_output.c:116
 __ip6_finish_output net/ipv6/ip6_output.c:142 [inline]
 __ip6_finish_output+0x2d7/0x330 net/ipv6/ip6_output.c:127
 ip6_finish_output+0x41/0x160 net/ipv6/ip6_output.c:152
 NF_HOOK_COND include/linux/netfilter.h:294 [inline]
 ip6_output+0xf2/0x280 net/ipv6/ip6_output.c:175

Reported by Kernel Concurrency Sanitizer on:
CPU: 1 PID: 11183 Comm: syz-executor.2 Not tainted 5.4.0-rc3+ #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011

Fixes: 4cdf507d ("icmp: add a global rate limitation")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/icmp.c

@@ -249,10 +249,11 @@ bool icmp_global_allow(void)
 	bool rc = false;
 
 	/* Check if token bucket is empty and cannot be refilled
-	 * without taking the spinlock.
+	 * without taking the spinlock. The READ_ONCE() are paired
+	 * with the following WRITE_ONCE() in this same function.
 	 */
-	if (!icmp_global.credit) {
-		delta = min_t(u32, now - icmp_global.stamp, HZ);
+	if (!READ_ONCE(icmp_global.credit)) {
+		delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
 		if (delta < HZ / 50)
 			return false;
 	}
@@ -262,14 +263,14 @@ bool icmp_global_allow(void)
 	if (delta >= HZ / 50) {
 		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
 		if (incr)
-			icmp_global.stamp = now;
+			WRITE_ONCE(icmp_global.stamp, now);
 	}
 	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
 	if (credit) {
 		credit--;
 		rc = true;
 	}
-	icmp_global.credit = credit;
+	WRITE_ONCE(icmp_global.credit, credit);
 	spin_unlock(&icmp_global.lock);
 	return rc;
 }