tcp: fix race condition when creating child sockets from syncookies
Subsystems (3)
PATCH v6 1/1 tcp: fix race condition when creating child sockets from syncookies
When the TCP stack is in SYN flood mode, the server child socket is created from the SYN cookie received in a TCP packet with the ACK flag set.
The child socket is created when the server receives the first TCP packet with a valid SYN cookie from the client. Usually, this packet corresponds to the final step of the TCP 3-way handshake, the ACK packet. But is also possible to receive a valid SYN cookie from the first TCP data packet sent by the client, and thus create a child socket from that SYN cookie.
Since a client socket is ready to send data as soon as it receives the SYN+ACK packet from the server, the client can send the ACK packet (sent by the TCP stack code), and the first data packet (sent by the userspace program) almost at the same time, and thus the server will equally receive the two TCP packets with valid SYN cookies almost at the same instant.
When such event happens, the TCP stack code has a race condition that occurs between the momement a lookup is done to the established connections hashtable to check for the existence of a connection for the same client, and the moment that the child socket is added to the established connections hashtable. As a consequence, this race condition can lead to a situation where we add two child sockets to the established connections hashtable and deliver two sockets to the userspace program to the same client.
This patch fixes the race condition by checking if an existing child socket exists for the same client when we are adding the second child socket to the established connections socket. If an existing child socket exists, we return that socket and use it to process the TCP packet received, and discard the second child socket to the same client.
---
v6 (2020-11-17):
* Moved the ehash bucket list search for its own helper function.
v5 (2020-11-16):
- Not considered for review -
v4 (2020-11-12):
* Added struct sock **esk) parameter to inet_ehash_insert`.
* Fixed ref count increment in inet_ehash_insert.
* Fixed callers of inet_ehash_nolisten.
v3 (2020-11-11):
* Fixed IPv6 handling in inet_ehash_insert
* Removed unecessary comparison while traversing the ehash bucket list.
v2 (2020-11-09):
* Changed the author's email domain.
* Removed the helper function inet_ehash_insert_chk_dup and moved the logic to the existing inet_ehash_insert.
* Updated the callers of iner_ehash_nolisten to deal with the new logic.
include/net/inet_hashtables.h | 4 +- net/dccp/ipv4.c | 2 +- net/dccp/ipv6.c | 2 +- net/ipv4/inet_connection_sock.c | 2 +- net/ipv4/inet_hashtables.c | 72 +++++++++++++++++++++++++++++---- net/ipv4/syncookies.c | 5 ++- net/ipv4/tcp_ipv4.c | 9 ++++- net/ipv6/tcp_ipv6.c | 16 +++++++- 8 files changed, 96 insertions(+), 16 deletions(-)
include/net/inet_hashtables.h +2 -2
@@ -247,8 +247,8 @@ void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name, unsigned long high_limit);int inet_hashinfo2_init_mod (struct inet_hashinfo *h); -bool inet_ehash_insert (struct sock *sk,struct sock *osk); -bool inet_ehash_nolisten (struct sock *sk,struct sock *osk); +bool inet_ehash_insert (struct sock *sk,struct sock *osk,struct sock **esk); +bool inet_ehash_nolisten (struct sock *sk,struct sock *osk,struct sock **esk);int __inet_hash (struct sock *sk,struct sock *osk);int inet_hash (struct sock *sk);void inet_unhash (struct sock *sk);
net/dccp/ipv4.c +1 -1
@@ -427,7 +427,7 @@ struct sock *dccp_v4_request_recv_sock(const struct sock *sk,if (__inet_inherit_port (sk, newsk) <0 ) goto put_and_exit; - *own_req =inet_ehash_nolisten (newsk,req_to_sk (req_unhash)); + *own_req =inet_ehash_nolisten (newsk,req_to_sk (req_unhash),NULL );if (*own_req) ireq->ireq_opt =NULL ;else
net/dccp/ipv6.c +1 -1
@@ -533,7 +533,7 @@ static struct sock *dccp_v6_request_recv_sock(const struct sock *sk,dccp_done (newsk); goto out; } - *own_req =inet_ehash_nolisten (newsk,req_to_sk (req_unhash)); + *own_req =inet_ehash_nolisten (newsk,req_to_sk (req_unhash),NULL );/* Clone pktoptions received with SYN, if we own the req */ if (*own_req && ireq->pktopts) { newnp->pktoptions =skb_clone (ireq->pktopts,GFP_ATOMIC );
net/ipv4/inet_connection_sock.c +1 -1
@@ -787,7 +787,7 @@ static void reqsk_queue_hash_req(struct request_sock *req,timer_setup (&req->rsk_timer, reqsk_timer_handler,TIMER_PINNED );mod_timer (&req->rsk_timer, jiffies + timeout); -inet_ehash_insert (req_to_sk (req),NULL ); +inet_ehash_insert (req_to_sk (req),NULL ,NULL );/* before letting lookups find us, make sure all req fields * are committed to memory and refcnt initialized. */
net/ipv4/inet_hashtables.c +64 -8
117 lines not on this page · raw message
net/ipv4/syncookies.c +4 -1
@@ -208,7 +208,7 @@ struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, child = icsk->icsk_af_ops->syn_recv_sock (sk, skb, req, dst,NULL , &own_req); -if (child) { +if (child && own_req) {refcount_set (&req->rsk_refcnt,1 );tcp_sk (child)->tsoffset = tsoff;sock_rps_save_rxhash (child, skb);
@@ -223,6 +223,9 @@ struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,bh_unlock_sock (child);sock_put (child); + }else if (child && !own_req) { +__reqsk_free (req); +return child; }__reqsk_free (req);
net/ipv4/tcp_ipv4.c +8 -1
@@ -1501,6 +1501,7 @@ struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,int l3index;#endif struct ip_options_rcu *inet_opt; +struct sock *esk =NULL ;if (sk_acceptq_is_full (sk)) goto exit_overflow;
@@ -1565,11 +1566,17 @@ struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,if (__inet_inherit_port (sk, newsk) <0 ) goto put_and_exit; - *own_req =inet_ehash_nolisten (newsk,req_to_sk (req_unhash)); + *own_req =inet_ehash_nolisten (newsk,req_to_sk (req_unhash), &esk);if (likely (*own_req)) {tcp_move_syn (newtp, req); ireq->ireq_opt =NULL ; }else { +if (!req_unhash && esk) { +/* This code path should only be executed in the +* syncookie case only +*/ + newsk = esk; + } newinet->inet_opt =NULL ; }return newsk;
net/ipv6/tcp_ipv6.c +15 -1
@@ -1190,6 +1190,7 @@ static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *struct inet_sock *newinet;struct tcp_sock *newtp;struct sock *newsk; +struct sock *esk =NULL ;#ifdef CONFIG_TCP_MD5SIG struct tcp_md5sig_key *key;int l3index;
@@ -1206,6 +1207,12 @@ static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *if (!newsk)return NULL ; +else if (!req_unhash && !own_req) { +/* We're returning an existing child socket, probably +* created by a previous syncookie ACK. +*/ +return newsk; + }inet_sk (newsk)->pinet6 =tcp_inet6_sk (newsk);
@@ -1359,7 +1366,7 @@ static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *tcp_done (newsk); goto out; } - *own_req =inet_ehash_nolisten (newsk,req_to_sk (req_unhash)); + *own_req =inet_ehash_nolisten (newsk,req_to_sk (req_unhash), &esk);if (*own_req) {tcp_move_syn (newtp, req);
@@ -1374,6 +1381,13 @@ static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb_set_owner_r (newnp->pktoptions, newsk); } } + }else { +if (!req_unhash && esk) { +/* This code path should only be executed in the +* syncookie case only +*/ + newsk = esk; + } }return newsk;
Signature · 1 line
2.25.1
Quoted 4 lines · Ricardo Dias
When the TCP stack is in SYN flood mode, the server child socket is created from the SYN cookie received in a TCP packet with the ACK flag set.
...
Review static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff * · line 1389
Quoted 11 lines
(quoted diff) +6
-0@@ -1374,6 +1381,13 @@ static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff * skb_set_owner_r(newnp->pktoptions, newsk); } } + } else { + if (!req_unhash && esk) { + /* This code path should only be executed in the + * syncookie case only + */ + newsk = esk;
What happens to the socket we just created and was stored in @newsk ?
We are going to leak kernel memory quite fast, if we just forget about it.
Review static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff * · line 1393
Quoted 5 lines
(quoted diff) +1
-0+ } } return newsk;
Anyway, I have tested your patch (see below) and had many errors in
[ 669.578597] cleanup rbuf bug: copied 1A5AD52D seq 1A5AD52D rcvnxt 1A5AD52D [ 669.578604] WARNING: CPU: 30 PID: 13267 at net/ipv4/tcp.c:1553 tcp_cleanup_rbuf+0x48/0x100 [ 669.578604] Modules linked in: vfat fat w1_therm wire i2c_mux_pca954x i2c_mux cdc_acm ehci_pci ehci_hcd mlx4_en mlx4_ib ib_uverbs ib_core mlx4_core [ 669.578610] CPU: 30 PID: 13267 Comm: ld-2.15.so.orig Tainted: G S W 5.10.0-smp-DEV #394 [ 669.578611] Hardware name: Intel RML,PCH/Iota_QC_19, BIOS 2.60.0 10/02/2019 [ 669.578612] RIP: 0010:tcp_cleanup_rbuf+0x48/0x100 [ 669.578612] Code: f3 48 39 d0 74 26 48 85 c0 74 21 8b 50 2c 8b b7 7c 05 00 00 39 d6 78 14 8b 8f 78 05 00 00 48 c7 c7 08 8a 57 a5 e8 1d b6 10 00 <0f> 0b 41 0f b6 84 24 c8 04 00 00 a8 01 74 2a 41 0f b7 8c 24 de 04 [ 669.578613] RSP: 0018:ffff893bd898fca8 EFLAGS: 00010286 [ 669.578614] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 00000000ffff7fff [ 669.578615] RDX: 00000000ffff7fff RSI: 00000000ffffffea RDI: 0000000000000000 [ 669.578615] RBP: ffff893bd898fcb8 R08: ffffffffa5bed108 R09: 0000000000027ffb [ 669.578616] R10: 00000000ffff8000 R11: 3fffffffffffffff R12: ffff891cd3530140 [ 669.578617] R13: ffff891cd3530140 R14: 0000000000000000 R15: 0000000000000000 [ 669.578618] FS: 00007fb7908ea700(0000) GS:ffff893affc80000(0000) knlGS:0000000000000000 [ 669.578619] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 669.578620] CR2: 000034fedd58981c CR3: 00000020d87b2001 CR4: 00000000001706e0 [ 669.578620] Call Trace: [ 669.578621] tcp_recvmsg+0x21f/0xac0 [ 669.578624] inet6_recvmsg+0x5f/0x120 [ 669.578625] ? add_wait_queue_exclusive+0x80/0x80 [ 669.578627] sock_recvmsg+0x54/0x80 [ 669.578628] __sys_recvfrom+0x19e/0x1d0 [ 669.578630] ? __fput+0x11c/0x240 [ 669.578632] __x64_sys_recvfrom+0x29/0x30 [ 669.578634] do_syscall_64+0x38/0x50 [ 669.578635] entry_SYSCALL_64_after_hwframe+0x44/0xa9
I have used following method on DUT host, and patch [1] in get_rps_cpus()
DUT has 48 hyper threads, I force RPS to spread packets on all of them.
for q in {0..15}; do echo ffff,ffffffff >/sys/class/net/eth0/queues/rx-$q/rps_cpus; done echo 2 >/proc/sys/net/ipv4/tcp_syncookies
netserver
Then from another host :
for i in {1..20}
do
super_netperf 200 -H DUT -t TCP_CRR -l2 -- -p0
done
After the test, I had more than 100,000 TCPv6 sockets that were mem leaked.
[1]
net/core/dev.c +1 -1
@@ -4344,7 +4344,7 @@ static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, goto done;skb_reset_network_header (skb); - hash =skb_get_hash (skb); + hash =prandom_u32 ();if (!hash) goto done;