net: icmp: fix data-race in cmp_global_allow()
Subsystems (3)
PATCH 1/1 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 4cdf507d5452 ("icmp: add a global rate limitation")
Signed-off-by Eric Dumazet <[email protected]>
Reported-by syzbot <[email protected]>
--- net/ipv4/icmp.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-)
net/ipv4/icmp.c +6 -5
@@ -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; }
Signature · 1 line
2.24.0.432.g9d3f5f5b63-goog
David Miller
AppliedFrom: Eric Dumazet <[email protected]>
Date: Fri, 8 Nov 2019 10:34:47 -0800
Quoted 6 lines
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
...
Quoted 3 lines
Fixes 4cdf507d5452 ("icmp: add a global rate limitation")Signed-off-by Eric Dumazet <[email protected]>Reported-by syzbot <[email protected]>
Applied, thanks Eric.