Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux

net/sched: annotate lockless accesses to qdisc->empty

KCSAN reported the following race [1]

BUG: KCSAN: data-race in __dev_queue_xmit / net_tx_action

read to 0xffff8880ba403508 of 1 bytes by task 21814 on cpu 1:
__dev_xmit_skb net/core/dev.c:3389 [inline]
__dev_queue_xmit+0x9db/0x1b40 net/core/dev.c:3761
dev_queue_xmit+0x21/0x30 net/core/dev.c:3825
neigh_hh_output include/net/neighbour.h:500 [inline]
neigh_output include/net/neighbour.h:509 [inline]
ip6_finish_output2+0x873/0xec0 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
ip6_send_skb+0x53/0x110 net/ipv6/ip6_output.c:1795
udp_v6_send_skb.isra.0+0x3ec/0xa70 net/ipv6/udp.c:1173
udpv6_sendmsg+0x1906/0x1c20 net/ipv6/udp.c:1471
inet6_sendmsg+0x6d/0x90 net/ipv6/af_inet6.c:576
sock_sendmsg_nosec net/socket.c:637 [inline]
sock_sendmsg+0x9f/0xc0 net/socket.c:657
___sys_sendmsg+0x2b7/0x5d0 net/socket.c:2311
__sys_sendmmsg+0x123/0x350 net/socket.c:2413
__do_sys_sendmmsg net/socket.c:2442 [inline]
__se_sys_sendmmsg net/socket.c:2439 [inline]
__x64_sys_sendmmsg+0x64/0x80 net/socket.c:2439
do_syscall_64+0xcc/0x370 arch/x86/entry/common.c:290
entry_SYSCALL_64_after_hwframe+0x44/0xa9

write to 0xffff8880ba403508 of 1 bytes by interrupt on cpu 0:
qdisc_run_begin include/net/sch_generic.h:160 [inline]
qdisc_run include/net/pkt_sched.h:120 [inline]
net_tx_action+0x2b1/0x6c0 net/core/dev.c:4551
__do_softirq+0x115/0x33f kernel/softirq.c:292
do_softirq_own_stack+0x2a/0x40 arch/x86/entry/entry_64.S:1082
do_softirq.part.0+0x6b/0x80 kernel/softirq.c:337
do_softirq kernel/softirq.c:329 [inline]
__local_bh_enable_ip+0x76/0x80 kernel/softirq.c:189
local_bh_enable include/linux/bottom_half.h:32 [inline]
rcu_read_unlock_bh include/linux/rcupdate.h:688 [inline]
ip6_finish_output2+0x7bb/0xec0 net/ipv6/ip6_output.c:117
__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
ip6_send_skb+0x53/0x110 net/ipv6/ip6_output.c:1795
udp_v6_send_skb.isra.0+0x3ec/0xa70 net/ipv6/udp.c:1173
udpv6_sendmsg+0x1906/0x1c20 net/ipv6/udp.c:1471
inet6_sendmsg+0x6d/0x90 net/ipv6/af_inet6.c:576
sock_sendmsg_nosec net/socket.c:637 [inline]
sock_sendmsg+0x9f/0xc0 net/socket.c:657
___sys_sendmsg+0x2b7/0x5d0 net/socket.c:2311
__sys_sendmmsg+0x123/0x350 net/socket.c:2413
__do_sys_sendmmsg net/socket.c:2442 [inline]
__se_sys_sendmmsg net/socket.c:2439 [inline]
__x64_sys_sendmmsg+0x64/0x80 net/socket.c:2439
do_syscall_64+0xcc/0x370 arch/x86/entry/common.c:290
entry_SYSCALL_64_after_hwframe+0x44/0xa9

Reported by Kernel Concurrency Sanitizer on:
CPU: 0 PID: 21817 Comm: syz-executor.2 Not tainted 5.4.0-rc6+ #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011

Fixes: d518d2ed8640 ("net/sched: fix race between deactivation and dequeue for NOLOCK qdisc")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

authored by

Eric Dumazet and committed by
David S. Miller
90b2be27 97c20ea8

+5 -5
+3 -3
include/net/sch_generic.h
··· 148 148 static inline bool qdisc_is_empty(const struct Qdisc *qdisc) 149 149 { 150 150 if (qdisc_is_percpu_stats(qdisc)) 151 - return qdisc->empty; 152 - return !qdisc->q.qlen; 151 + return READ_ONCE(qdisc->empty); 152 + return !READ_ONCE(qdisc->q.qlen); 153 153 } 154 154 155 155 static inline bool qdisc_run_begin(struct Qdisc *qdisc) ··· 157 157 if (qdisc->flags & TCQ_F_NOLOCK) { 158 158 if (!spin_trylock(&qdisc->seqlock)) 159 159 return false; 160 - qdisc->empty = false; 160 + WRITE_ONCE(qdisc->empty, false); 161 161 } else if (qdisc_is_running(qdisc)) { 162 162 return false; 163 163 }
+1 -1
net/core/dev.c
··· 3607 3607 qdisc_calculate_pkt_len(skb, q); 3608 3608 3609 3609 if (q->flags & TCQ_F_NOLOCK) { 3610 - if ((q->flags & TCQ_F_CAN_BYPASS) && q->empty && 3610 + if ((q->flags & TCQ_F_CAN_BYPASS) && READ_ONCE(q->empty) && 3611 3611 qdisc_run_begin(q)) { 3612 3612 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 3613 3613 &q->state))) {
+1 -1
net/sched/sch_generic.c
··· 652 652 if (likely(skb)) { 653 653 qdisc_update_stats_at_dequeue(qdisc, skb); 654 654 } else { 655 - qdisc->empty = true; 655 + WRITE_ONCE(qdisc->empty, true); 656 656 } 657 657 658 658 return skb;