1 From foo@baz Wed Mar 28 18:38:30 CEST 2018
2 From: Soheil Hassas Yeganeh <soheil@google.com>
3 Date: Thu, 15 Mar 2018 12:09:13 -0400
4 Subject: tcp: reset sk_send_head in tcp_write_queue_purge
6 From: Soheil Hassas Yeganeh <soheil@google.com>
9 tcp_write_queue_purge clears all the SKBs in the write queue
10 but does not reset the sk_send_head. As a result, we can have
11 a NULL pointer dereference anywhere that we use tcp_send_head
12 instead of the tcp_write_queue_tail.
14 For example, after a27fd7a8ed38 (tcp: purge write queue upon RST),
15 we can purge the write queue on RST. Prior to
16 75c119afe14f (tcp: implement rb-tree based retransmit queue),
17 tcp_push will only check tcp_send_head and then accesses
18 tcp_write_queue_tail to send the actual SKB. As a result, it will
19 dereference a NULL pointer.
21 This has been reported twice for 4.14 where we don't have
26 [ 422.081094] BUG: unable to handle kernel NULL pointer dereference
28 [ 422.081254] IP: tcp_push+0x42/0x110
29 [ 422.081314] PGD 0 P4D 0
30 [ 422.081364] Oops: 0002 [#1] SMP PTI
34 BUG: unable to handle kernel NULL pointer dereference at 0000000000000038
35 IP: tcp_push+0x48/0x120
36 PGD 80000007ff77b067 P4D 80000007ff77b067 PUD 7fd989067 PMD 0
37 Oops: 0002 [#18] SMP PTI
38 Modules linked in: tcp_diag inet_diag tcp_bbr sch_fq iTCO_wdt
39 iTCO_vendor_support pcspkr ixgbe mdio i2c_i801 lpc_ich joydev input_leds shpchp
40 e1000e igb dca ptp pps_core hwmon mei_me mei ipmi_si ipmi_msghandler sg ses
41 scsi_transport_sas enclosure ext4 jbd2 mbcache sd_mod ahci libahci megaraid_sas
42 wmi ast ttm dm_mirror dm_region_hash dm_log dm_mod dax
43 CPU: 6 PID: 14156 Comm: [ET_NET 6] Tainted: G D 4.14.26-1.el6.x86_64 #1
44 Hardware name: LENOVO ThinkServer RD440 /ThinkServer RD440, BIOS A0TS80A
46 task: ffff8807d78d8140 task.stack: ffffc9000e944000
47 RIP: 0010:tcp_push+0x48/0x120
48 RSP: 0018:ffffc9000e947a88 EFLAGS: 00010246
49 RAX: 00000000000005b4 RBX: ffff880f7cce9c00 RCX: 0000000000000000
50 RDX: 0000000000000000 RSI: 0000000000000040 RDI: ffff8807d00f5000
51 RBP: ffffc9000e947aa8 R08: 0000000000001c84 R09: 0000000000000000
52 R10: ffff8807d00f5158 R11: 0000000000000000 R12: ffff8807d00f5000
53 R13: 0000000000000020 R14: 00000000000256d4 R15: 0000000000000000
54 FS: 00007f5916de9700(0000) GS:ffff88107fd00000(0000) knlGS:0000000000000000
55 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
56 CR2: 0000000000000038 CR3: 00000007f8226004 CR4: 00000000001606e0
58 tcp_sendmsg_locked+0x33d/0xe50
60 inet_sendmsg+0x39/0xc0
61 sock_sendmsg+0x49/0x60
62 sock_write_iter+0xb6/0x100
63 do_iter_readv_writev+0xec/0x130
64 ? rw_verify_area+0x49/0xb0
65 do_iter_write+0x97/0xd0
67 ? __wake_up_common_lock+0x80/0xa0
68 ? __fget_light+0x2c/0x70
69 ? __do_page_fault+0x1e7/0x530
71 ? inet_shutdown+0xac/0x110
73 do_syscall_64+0x6f/0x140
74 ? prepare_exit_to_usermode+0x8b/0xa0
75 entry_SYSCALL_64_after_hwframe+0x3d/0xa2
76 RIP: 0033:0x3135ce0c57
77 RSP: 002b:00007f5916de4b00 EFLAGS: 00000293 ORIG_RAX: 0000000000000014
78 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000003135ce0c57
79 RDX: 0000000000000002 RSI: 00007f5916de4b90 RDI: 000000000000606f
80 RBP: 0000000000000000 R08: 0000000000000000 R09: 00007f5916de8c38
81 R10: 0000000000000000 R11: 0000000000000293 R12: 00000000000464cc
82 R13: 00007f5916de8c30 R14: 00007f58d8bef080 R15: 0000000000000002
83 Code: 48 8b 97 60 01 00 00 4c 8d 97 58 01 00 00 41 b9 00 00 00 00 41 89 f3 4c 39
84 d2 49 0f 44 d1 41 81 e3 00 80 00 00 0f 85 b0 00 00 00 <80> 4a 38 08 44 8b 8f 74
85 06 00 00 44 89 8f 7c 06 00 00 83 e6 01
86 RIP: tcp_push+0x48/0x120 RSP: ffffc9000e947a88
88 ---[ end trace 8d545c2e93515549 ]---
90 Fixes: a27fd7a8ed38 (tcp: purge write queue upon RST)
91 Reported-by: Timofey Titovets <nefelim4ag@gmail.com>
92 Reported-by: Yongjian Xu <yongjianchn@gmail.com>
93 Signed-off-by: Eric Dumazet <edumazet@google.com>
94 Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
95 Tested-by: Yongjian Xu <yongjianchn@gmail.com>
97 Signed-off-by: David S. Miller <davem@davemloft.net>
98 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
100 include/net/tcp.h | 11 ++++++-----
101 1 file changed, 6 insertions(+), 5 deletions(-)
103 --- a/include/net/tcp.h
104 +++ b/include/net/tcp.h
105 @@ -1600,6 +1600,11 @@ enum tcp_chrono {
106 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
107 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
109 +static inline void tcp_init_send_head(struct sock *sk)
111 + sk->sk_send_head = NULL;
114 /* write queue abstraction */
115 static inline void tcp_write_queue_purge(struct sock *sk)
117 @@ -1610,6 +1615,7 @@ static inline void tcp_write_queue_purge
118 sk_wmem_free_skb(sk, skb);
120 tcp_clear_all_retrans_hints(tcp_sk(sk));
121 + tcp_init_send_head(sk);
124 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
125 @@ -1672,11 +1678,6 @@ static inline void tcp_check_send_head(s
126 tcp_sk(sk)->highest_sack = NULL;
129 -static inline void tcp_init_send_head(struct sock *sk)
131 - sk->sk_send_head = NULL;
134 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
136 __skb_queue_tail(&sk->sk_write_queue, skb);