ipv4: fix suspicious RCU usage in fib_dump_info_fnhe()
authorEric Dumazet <edumazet@google.com>
Wed, 26 Jun 2019 10:04:50 +0000 (03:04 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 26 Jun 2019 16:07:39 +0000 (09:07 -0700)
sysbot reported that we lack appropriate rcu_read_lock()
protection in fib_dump_info_fnhe()

net/ipv4/route.c:2875 suspicious rcu_dereference_check() usage!

other info that might help us debug this:

rcu_scheduler_active = 2, debug_locks = 1
1 lock held by syz-executor609/8966:
 #0: 00000000b7dbe288 (rtnl_mutex){+.+.}, at: netlink_dump+0xe7/0xfb0 net/netlink/af_netlink.c:2199

stack backtrace:
CPU: 0 PID: 8966 Comm: syz-executor609 Not tainted 5.2.0-rc5+ #43
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x172/0x1f0 lib/dump_stack.c:113
 lockdep_rcu_suspicious+0x153/0x15d kernel/locking/lockdep.c:5250
 fib_dump_info_fnhe+0x9d9/0x1080 net/ipv4/route.c:2875
 fn_trie_dump_leaf net/ipv4/fib_trie.c:2141 [inline]
 fib_table_dump+0x64a/0xd00 net/ipv4/fib_trie.c:2175
 inet_dump_fib+0x83c/0xa90 net/ipv4/fib_frontend.c:1004
 rtnl_dump_all+0x295/0x490 net/core/rtnetlink.c:3445
 netlink_dump+0x558/0xfb0 net/netlink/af_netlink.c:2244
 __netlink_dump_start+0x5b1/0x7d0 net/netlink/af_netlink.c:2352
 netlink_dump_start include/linux/netlink.h:226 [inline]
 rtnetlink_rcv_msg+0x73d/0xb00 net/core/rtnetlink.c:5182
 netlink_rcv_skb+0x177/0x450 net/netlink/af_netlink.c:2477
 rtnetlink_rcv+0x1d/0x30 net/core/rtnetlink.c:5237
 netlink_unicast_kernel net/netlink/af_netlink.c:1302 [inline]
 netlink_unicast+0x531/0x710 net/netlink/af_netlink.c:1328
 netlink_sendmsg+0x8ae/0xd70 net/netlink/af_netlink.c:1917
 sock_sendmsg_nosec net/socket.c:646 [inline]
 sock_sendmsg+0xd7/0x130 net/socket.c:665
 sock_write_iter+0x27c/0x3e0 net/socket.c:994
 call_write_iter include/linux/fs.h:1872 [inline]
 new_sync_write+0x4d3/0x770 fs/read_write.c:483
 __vfs_write+0xe1/0x110 fs/read_write.c:496
 vfs_write+0x20c/0x580 fs/read_write.c:558
 ksys_write+0x14f/0x290 fs/read_write.c:611
 __do_sys_write fs/read_write.c:623 [inline]
 __se_sys_write fs/read_write.c:620 [inline]
 __x64_sys_write+0x73/0xb0 fs/read_write.c:620
 do_syscall_64+0xfd/0x680 arch/x86/entry/common.c:301
 entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x4401b9
Code: 18 89 d0 c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 fb 13 fc ff c3 66 2e 0f 1f 84 00 00 00 00
RSP: 002b:00007ffc8e134978 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 00000000004002c8 RCX: 00000000004401b9
RDX: 000000000000001c RSI: 0000000020000000 RDI: 0000000000000003
RBP: 00000000006ca018 R08: 00000000004002c8 R09: 00000000004002c8
R10: 0000000000000010 R11: 0000000000000246 R12: 0000000000401a40
R13: 0000000000401ad0 R14: 0000000000000000 R15: 0000000000000000

Fixes: ee28906fd7a1 ("ipv4: Dump route exceptions if requested")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Stefano Brivio <sbrivio@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/route.c

index 6aee412a68bdd3c24a6a0eb9883e04b7a83998e0..59670fafcd2612b94c237cbe30109adb196cf3f0 100644 (file)
@@ -2872,12 +2872,13 @@ int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
                if (nhc->nhc_flags & RTNH_F_DEAD)
                        continue;
 
+               rcu_read_lock();
                bucket = rcu_dereference(nhc->nhc_exceptions);
-               if (!bucket)
-                       continue;
-
-               err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, genid,
-                                      fa_index, fa_start);
+               err = 0;
+               if (bucket)
+                       err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
+                                              genid, fa_index, fa_start);
+               rcu_read_unlock();
                if (err)
                        return err;
        }