blob: 91762be58accaed7d8974e723b580fb3d7922fca [file] [log] [blame]
Herbert Xucf80e0e2016-01-24 21:20:23 +08001#include <linux/crypto.h>
Jerry Chu10467162012-08-31 12:29:11 +00002#include <linux/err.h>
Yuchung Cheng2100c8d2012-07-19 06:43:05 +00003#include <linux/init.h>
4#include <linux/kernel.h>
Jerry Chu10467162012-08-31 12:29:11 +00005#include <linux/list.h>
6#include <linux/tcp.h>
7#include <linux/rcupdate.h>
8#include <linux/rculist.h>
9#include <net/inetpeer.h>
10#include <net/tcp.h>
Yuchung Cheng2100c8d2012-07-19 06:43:05 +000011
Haishuang Yan43713842017-09-27 11:35:42 +080012void tcp_fastopen_init_key_once(struct net *net)
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +020013{
Haishuang Yan43713842017-09-27 11:35:42 +080014 u8 key[TCP_FASTOPEN_KEY_LENGTH];
15 struct tcp_fastopen_context *ctxt;
16
17 rcu_read_lock();
18 ctxt = rcu_dereference(net->ipv4.tcp_fastopen_ctx);
19 if (ctxt) {
20 rcu_read_unlock();
21 return;
22 }
23 rcu_read_unlock();
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +020024
25 /* tcp_fastopen_reset_cipher publishes the new context
26 * atomically, so we allow this race happening here.
27 *
28 * All call sites of tcp_fastopen_cookie_gen also check
29 * for a valid cookie, so this is an acceptable risk.
30 */
Haishuang Yan43713842017-09-27 11:35:42 +080031 get_random_bytes(key, sizeof(key));
Yuchung Cheng1fba70e2017-10-18 11:22:51 -070032 tcp_fastopen_reset_cipher(net, NULL, key, sizeof(key));
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +020033}
34
Jerry Chu10467162012-08-31 12:29:11 +000035static void tcp_fastopen_ctx_free(struct rcu_head *head)
36{
37 struct tcp_fastopen_context *ctx =
38 container_of(head, struct tcp_fastopen_context, rcu);
39 crypto_free_cipher(ctx->tfm);
40 kfree(ctx);
41}
42
Yuchung Cheng1fba70e2017-10-18 11:22:51 -070043void tcp_fastopen_destroy_cipher(struct sock *sk)
44{
45 struct tcp_fastopen_context *ctx;
46
47 ctx = rcu_dereference_protected(
48 inet_csk(sk)->icsk_accept_queue.fastopenq.ctx, 1);
49 if (ctx)
50 call_rcu(&ctx->rcu, tcp_fastopen_ctx_free);
51}
52
Haishuang Yan43713842017-09-27 11:35:42 +080053void tcp_fastopen_ctx_destroy(struct net *net)
54{
55 struct tcp_fastopen_context *ctxt;
56
57 spin_lock(&net->ipv4.tcp_fastopen_ctx_lock);
58
59 ctxt = rcu_dereference_protected(net->ipv4.tcp_fastopen_ctx,
60 lockdep_is_held(&net->ipv4.tcp_fastopen_ctx_lock));
61 rcu_assign_pointer(net->ipv4.tcp_fastopen_ctx, NULL);
62 spin_unlock(&net->ipv4.tcp_fastopen_ctx_lock);
63
64 if (ctxt)
65 call_rcu(&ctxt->rcu, tcp_fastopen_ctx_free);
66}
67
Yuchung Cheng1fba70e2017-10-18 11:22:51 -070068int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
69 void *key, unsigned int len)
Jerry Chu10467162012-08-31 12:29:11 +000070{
Jerry Chu10467162012-08-31 12:29:11 +000071 struct tcp_fastopen_context *ctx, *octx;
Yuchung Cheng1fba70e2017-10-18 11:22:51 -070072 struct fastopen_queue *q;
73 int err;
Jerry Chu10467162012-08-31 12:29:11 +000074
75 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
76 if (!ctx)
77 return -ENOMEM;
78 ctx->tfm = crypto_alloc_cipher("aes", 0, 0);
79
80 if (IS_ERR(ctx->tfm)) {
81 err = PTR_ERR(ctx->tfm);
82error: kfree(ctx);
83 pr_err("TCP: TFO aes cipher alloc error: %d\n", err);
84 return err;
85 }
86 err = crypto_cipher_setkey(ctx->tfm, key, len);
87 if (err) {
88 pr_err("TCP: TFO cipher key error: %d\n", err);
89 crypto_free_cipher(ctx->tfm);
90 goto error;
91 }
92 memcpy(ctx->key, key, len);
93
Jerry Chu10467162012-08-31 12:29:11 +000094
Eric Dumazet9eba9352017-11-02 11:53:04 -070095 spin_lock(&net->ipv4.tcp_fastopen_ctx_lock);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -070096 if (sk) {
97 q = &inet_csk(sk)->icsk_accept_queue.fastopenq;
Yuchung Cheng1fba70e2017-10-18 11:22:51 -070098 octx = rcu_dereference_protected(q->ctx,
Eric Dumazet9eba9352017-11-02 11:53:04 -070099 lockdep_is_held(&net->ipv4.tcp_fastopen_ctx_lock));
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700100 rcu_assign_pointer(q->ctx, ctx);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700101 } else {
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700102 octx = rcu_dereference_protected(net->ipv4.tcp_fastopen_ctx,
103 lockdep_is_held(&net->ipv4.tcp_fastopen_ctx_lock));
104 rcu_assign_pointer(net->ipv4.tcp_fastopen_ctx, ctx);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700105 }
Eric Dumazet9eba9352017-11-02 11:53:04 -0700106 spin_unlock(&net->ipv4.tcp_fastopen_ctx_lock);
Jerry Chu10467162012-08-31 12:29:11 +0000107
108 if (octx)
109 call_rcu(&octx->rcu, tcp_fastopen_ctx_free);
110 return err;
111}
112
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700113static bool __tcp_fastopen_cookie_gen(struct sock *sk, const void *path,
Daniel Lee3a19ce02014-05-11 20:22:13 -0700114 struct tcp_fastopen_cookie *foc)
Jerry Chu10467162012-08-31 12:29:11 +0000115{
Jerry Chu10467162012-08-31 12:29:11 +0000116 struct tcp_fastopen_context *ctx;
Daniel Lee3a19ce02014-05-11 20:22:13 -0700117 bool ok = false;
Jerry Chu10467162012-08-31 12:29:11 +0000118
119 rcu_read_lock();
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700120
121 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
122 if (!ctx)
123 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
124
Jerry Chu10467162012-08-31 12:29:11 +0000125 if (ctx) {
Daniel Lee3a19ce02014-05-11 20:22:13 -0700126 crypto_cipher_encrypt_one(ctx->tfm, foc->val, path);
Jerry Chu10467162012-08-31 12:29:11 +0000127 foc->len = TCP_FASTOPEN_COOKIE_SIZE;
Daniel Lee3a19ce02014-05-11 20:22:13 -0700128 ok = true;
Jerry Chu10467162012-08-31 12:29:11 +0000129 }
130 rcu_read_unlock();
Daniel Lee3a19ce02014-05-11 20:22:13 -0700131 return ok;
132}
133
134/* Generate the fastopen cookie by doing aes128 encryption on both
135 * the source and destination addresses. Pad 0s for IPv4 or IPv4-mapped-IPv6
136 * addresses. For the longer IPv6 addresses use CBC-MAC.
137 *
138 * XXX (TFO) - refactor when TCP_FASTOPEN_COOKIE_SIZE != AES_BLOCK_SIZE.
139 */
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700140static bool tcp_fastopen_cookie_gen(struct sock *sk,
Haishuang Yan43713842017-09-27 11:35:42 +0800141 struct request_sock *req,
Daniel Lee3a19ce02014-05-11 20:22:13 -0700142 struct sk_buff *syn,
143 struct tcp_fastopen_cookie *foc)
144{
145 if (req->rsk_ops->family == AF_INET) {
146 const struct iphdr *iph = ip_hdr(syn);
147
148 __be32 path[4] = { iph->saddr, iph->daddr, 0, 0 };
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700149 return __tcp_fastopen_cookie_gen(sk, path, foc);
Daniel Lee3a19ce02014-05-11 20:22:13 -0700150 }
151
152#if IS_ENABLED(CONFIG_IPV6)
153 if (req->rsk_ops->family == AF_INET6) {
154 const struct ipv6hdr *ip6h = ipv6_hdr(syn);
155 struct tcp_fastopen_cookie tmp;
156
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700157 if (__tcp_fastopen_cookie_gen(sk, &ip6h->saddr, &tmp)) {
Shannon Nelson003c9412017-01-12 14:24:58 -0800158 struct in6_addr *buf = &tmp.addr;
Li RongQing41c91992014-09-29 15:04:37 +0800159 int i;
Daniel Lee3a19ce02014-05-11 20:22:13 -0700160
161 for (i = 0; i < 4; i++)
162 buf->s6_addr32[i] ^= ip6h->daddr.s6_addr32[i];
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700163 return __tcp_fastopen_cookie_gen(sk, buf, foc);
Daniel Lee3a19ce02014-05-11 20:22:13 -0700164 }
165 }
166#endif
167 return false;
Jerry Chu10467162012-08-31 12:29:11 +0000168}
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700169
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800170
171/* If an incoming SYN or SYNACK frame contains a payload and/or FIN,
172 * queue this additional data / FIN.
173 */
174void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb)
175{
176 struct tcp_sock *tp = tcp_sk(sk);
177
178 if (TCP_SKB_CB(skb)->end_seq == tp->rcv_nxt)
179 return;
180
181 skb = skb_clone(skb, GFP_ATOMIC);
182 if (!skb)
183 return;
184
185 skb_dst_drop(skb);
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -0700186 /* segs_in has been initialized to 1 in tcp_create_openreq_child().
187 * Hence, reset segs_in to 0 before calling tcp_segs_in()
188 * to avoid double counting. Also, tcp_segs_in() expects
189 * skb->len to include the tcp_hdrlen. Hence, it should
190 * be called before __skb_pull().
191 */
192 tp->segs_in = 0;
193 tcp_segs_in(tp, skb);
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800194 __skb_pull(skb, tcp_hdrlen(skb));
Eric Dumazet76061f62016-09-07 08:34:11 -0700195 sk_forced_mem_schedule(sk, skb->truesize);
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800196 skb_set_owner_r(skb, sk);
197
Eric Dumazet9d691532016-02-01 21:03:08 -0800198 TCP_SKB_CB(skb)->seq++;
199 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_SYN;
200
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800201 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
202 __skb_queue_tail(&sk->sk_receive_queue, skb);
203 tp->syn_data_acked = 1;
204
205 /* u64_stats_update_begin(&tp->syncp) not needed here,
206 * as we certainly are not changing upper 32bit value (0)
207 */
208 tp->bytes_received = skb->len;
Eric Dumazete3e17b72016-02-06 11:16:28 -0800209
210 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
211 tcp_fin(sk);
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800212}
213
Eric Dumazet7c85af82015-09-24 17:16:05 -0700214static struct sock *tcp_fastopen_create_child(struct sock *sk,
215 struct sk_buff *skb,
Eric Dumazet7c85af82015-09-24 17:16:05 -0700216 struct request_sock *req)
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700217{
Dave Jones17846372014-06-16 16:30:36 -0400218 struct tcp_sock *tp;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700219 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700220 struct sock *child;
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700221 bool own_req;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700222
223 req->num_retrans = 0;
224 req->num_timeout = 0;
225 req->sk = NULL;
226
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700227 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
228 NULL, &own_req);
Ian Morris51456b22015-04-03 09:17:26 +0100229 if (!child)
Eric Dumazet7c85af82015-09-24 17:16:05 -0700230 return NULL;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700231
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700232 spin_lock(&queue->fastopenq.lock);
233 queue->fastopenq.qlen++;
234 spin_unlock(&queue->fastopenq.lock);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700235
236 /* Initialize the child socket. Have to fix some values to take
237 * into account the child is a Fast Open socket and is created
238 * only out of the bits carried in the SYN packet.
239 */
240 tp = tcp_sk(child);
241
242 tp->fastopen_rsk = req;
Eric Dumazet9439ce02015-03-17 18:32:29 -0700243 tcp_rsk(req)->tfo_listener = true;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700244
245 /* RFC1323: The window in SYN & SYN/ACK segments is never
246 * scaled. So correct it appropriately.
247 */
248 tp->snd_wnd = ntohs(tcp_hdr(skb)->window);
Alexey Kodanev0dbd7ff2017-01-19 16:36:39 +0300249 tp->max_window = tp->snd_wnd;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700250
251 /* Activate the retrans timer so that SYNACK can be retransmitted.
Eric Dumazetca6fb062015-10-02 11:43:35 -0700252 * The request socket is not added to the ehash
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700253 * because it's been added to the accept queue directly.
254 */
255 inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
256 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
257
Reshetova, Elena41c6d652017-06-30 13:08:01 +0300258 refcount_set(&req->rsk_refcnt, 2);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700259
260 /* Now finish processing the fastopen child socket. */
Wei Wang27204aa2017-10-04 10:03:44 -0700261 tcp_init_transfer(child, BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700262
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800263 tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
Eric Dumazetba34e6d2015-02-13 04:47:12 -0800264
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800265 tcp_fastopen_add_skb(child, skb);
Eric Dumazetd6549762015-05-21 21:51:19 -0700266
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800267 tcp_rsk(req)->rcv_nxt = tp->rcv_nxt;
Neal Cardwell28b346c2016-08-30 11:55:23 -0400268 tp->rcv_wup = tp->rcv_nxt;
Eric Dumazet7656d842015-10-04 21:08:07 -0700269 /* tcp_conn_request() is sending the SYNACK,
270 * and queues the child into listener accept queue.
Eric Dumazet7c85af82015-09-24 17:16:05 -0700271 */
Eric Dumazet7c85af82015-09-24 17:16:05 -0700272 return child;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700273}
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700274
275static bool tcp_fastopen_queue_check(struct sock *sk)
276{
277 struct fastopen_queue *fastopenq;
278
279 /* Make sure the listener has enabled fastopen, and we don't
280 * exceed the max # of pending TFO requests allowed before trying
281 * to validating the cookie in order to avoid burning CPU cycles
282 * unnecessarily.
283 *
284 * XXX (TFO) - The implication of checking the max_qlen before
285 * processing a cookie request is that clients can't differentiate
286 * between qlen overflow causing Fast Open to be disabled
287 * temporarily vs a server not supporting Fast Open at all.
288 */
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700289 fastopenq = &inet_csk(sk)->icsk_accept_queue.fastopenq;
290 if (fastopenq->max_qlen == 0)
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700291 return false;
292
293 if (fastopenq->qlen >= fastopenq->max_qlen) {
294 struct request_sock *req1;
295 spin_lock(&fastopenq->lock);
296 req1 = fastopenq->rskq_rst_head;
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700297 if (!req1 || time_after(req1->rsk_timer.expires, jiffies)) {
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700298 __NET_INC_STATS(sock_net(sk),
299 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW);
Eric Dumazetc10d9312016-04-29 14:16:47 -0700300 spin_unlock(&fastopenq->lock);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700301 return false;
302 }
303 fastopenq->rskq_rst_head = req1->dl_next;
304 fastopenq->qlen--;
305 spin_unlock(&fastopenq->lock);
Eric Dumazet13854e52015-03-15 21:12:16 -0700306 reqsk_put(req1);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700307 }
308 return true;
309}
310
Christoph Paasch71c02372017-10-23 13:22:23 -0700311static bool tcp_fastopen_no_cookie(const struct sock *sk,
312 const struct dst_entry *dst,
313 int flag)
314{
315 return (sock_net(sk)->ipv4.sysctl_tcp_fastopen & flag) ||
316 tcp_sk(sk)->fastopen_no_cookie ||
317 (dst && dst_metric(dst, RTAX_FASTOPEN_NO_COOKIE));
318}
319
Yuchung Cheng89278c92014-05-11 20:22:10 -0700320/* Returns true if we should perform Fast Open on the SYN. The cookie (foc)
321 * may be updated and return the client in the SYN-ACK later. E.g., Fast Open
322 * cookie request (foc->len == 0).
323 */
Eric Dumazet7c85af82015-09-24 17:16:05 -0700324struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
325 struct request_sock *req,
Christoph Paasch71c02372017-10-23 13:22:23 -0700326 struct tcp_fastopen_cookie *foc,
327 const struct dst_entry *dst)
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700328{
Yuchung Cheng89278c92014-05-11 20:22:10 -0700329 bool syn_data = TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1;
Haishuang Yane1cfcbe2017-09-27 11:35:40 +0800330 int tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
331 struct tcp_fastopen_cookie valid_foc = { .len = -1 };
Eric Dumazet7c85af82015-09-24 17:16:05 -0700332 struct sock *child;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700333
Yuchung Cheng531c94a2015-02-09 12:35:23 -0800334 if (foc->len == 0) /* Client requests a cookie */
Eric Dumazetc10d9312016-04-29 14:16:47 -0700335 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENCOOKIEREQD);
Yuchung Cheng531c94a2015-02-09 12:35:23 -0800336
Haishuang Yane1cfcbe2017-09-27 11:35:40 +0800337 if (!((tcp_fastopen & TFO_SERVER_ENABLE) &&
Yuchung Cheng89278c92014-05-11 20:22:10 -0700338 (syn_data || foc->len >= 0) &&
339 tcp_fastopen_queue_check(sk))) {
340 foc->len = -1;
Eric Dumazet7c85af82015-09-24 17:16:05 -0700341 return NULL;
Yuchung Cheng89278c92014-05-11 20:22:10 -0700342 }
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700343
Christoph Paasch71c02372017-10-23 13:22:23 -0700344 if (syn_data &&
345 tcp_fastopen_no_cookie(sk, dst, TFO_SERVER_COOKIE_NOT_REQD))
Yuchung Cheng89278c92014-05-11 20:22:10 -0700346 goto fastopen;
347
Yuchung Cheng531c94a2015-02-09 12:35:23 -0800348 if (foc->len >= 0 && /* Client presents or requests a cookie */
Yuchung Cheng1fba70e2017-10-18 11:22:51 -0700349 tcp_fastopen_cookie_gen(sk, req, skb, &valid_foc) &&
Daniel Lee3a19ce02014-05-11 20:22:13 -0700350 foc->len == TCP_FASTOPEN_COOKIE_SIZE &&
Yuchung Cheng89278c92014-05-11 20:22:10 -0700351 foc->len == valid_foc.len &&
352 !memcmp(foc->val, valid_foc.val, foc->len)) {
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700353 /* Cookie is valid. Create a (full) child socket to accept
354 * the data in SYN before returning a SYN-ACK to ack the
355 * data. If we fail to create the socket, fall back and
356 * ack the ISN only but includes the same cookie.
357 *
358 * Note: Data-less SYN with valid cookie is allowed to send
359 * data in SYN_RECV state.
360 */
Yuchung Cheng89278c92014-05-11 20:22:10 -0700361fastopen:
Tonghao Zhang11199362017-08-21 23:33:49 -0700362 child = tcp_fastopen_create_child(sk, skb, req);
Eric Dumazet7c85af82015-09-24 17:16:05 -0700363 if (child) {
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700364 foc->len = -1;
Eric Dumazetc10d9312016-04-29 14:16:47 -0700365 NET_INC_STATS(sock_net(sk),
366 LINUX_MIB_TCPFASTOPENPASSIVE);
Eric Dumazet7c85af82015-09-24 17:16:05 -0700367 return child;
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700368 }
Eric Dumazetc10d9312016-04-29 14:16:47 -0700369 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
Yuchung Cheng531c94a2015-02-09 12:35:23 -0800370 } else if (foc->len > 0) /* Client presents an invalid cookie */
Eric Dumazetc10d9312016-04-29 14:16:47 -0700371 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700372
Daniel Lee7f9b8382015-04-06 14:37:26 -0700373 valid_foc.exp = foc->exp;
Yuchung Cheng89278c92014-05-11 20:22:10 -0700374 *foc = valid_foc;
Eric Dumazet7c85af82015-09-24 17:16:05 -0700375 return NULL;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700376}
Wei Wang065263f2017-01-23 10:59:20 -0800377
378bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
379 struct tcp_fastopen_cookie *cookie)
380{
381 unsigned long last_syn_loss = 0;
Christoph Paasch71c02372017-10-23 13:22:23 -0700382 const struct dst_entry *dst;
Wei Wang065263f2017-01-23 10:59:20 -0800383 int syn_loss = 0;
384
385 tcp_fastopen_cache_get(sk, mss, cookie, &syn_loss, &last_syn_loss);
386
387 /* Recurring FO SYN losses: no cookie or data in SYN */
388 if (syn_loss > 1 &&
389 time_before(jiffies, last_syn_loss + (60*HZ << syn_loss))) {
390 cookie->len = -1;
391 return false;
392 }
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700393
394 /* Firewall blackhole issue check */
395 if (tcp_fastopen_active_should_disable(sk)) {
396 cookie->len = -1;
397 return false;
398 }
399
Christoph Paasch71c02372017-10-23 13:22:23 -0700400 dst = __sk_dst_get(sk);
401
402 if (tcp_fastopen_no_cookie(sk, dst, TFO_CLIENT_NO_COOKIE)) {
Wei Wang065263f2017-01-23 10:59:20 -0800403 cookie->len = -1;
404 return true;
405 }
406 return cookie->len > 0;
407}
Wei Wang19f6d3f2017-01-23 10:59:22 -0800408
409/* This function checks if we want to defer sending SYN until the first
410 * write(). We defer under the following conditions:
411 * 1. fastopen_connect sockopt is set
412 * 2. we have a valid cookie
413 * Return value: return true if we want to defer until application writes data
414 * return false if we want to send out SYN immediately
415 */
416bool tcp_fastopen_defer_connect(struct sock *sk, int *err)
417{
418 struct tcp_fastopen_cookie cookie = { .len = 0 };
419 struct tcp_sock *tp = tcp_sk(sk);
420 u16 mss;
421
422 if (tp->fastopen_connect && !tp->fastopen_req) {
423 if (tcp_fastopen_cookie_check(sk, &mss, &cookie)) {
424 inet_sk(sk)->defer_connect = 1;
425 return true;
426 }
427
428 /* Alloc fastopen_req in order for FO option to be included
429 * in SYN
430 */
431 tp->fastopen_req = kzalloc(sizeof(*tp->fastopen_req),
432 sk->sk_allocation);
433 if (tp->fastopen_req)
434 tp->fastopen_req->cookie = cookie;
435 else
436 *err = -ENOBUFS;
437 }
438 return false;
439}
440EXPORT_SYMBOL(tcp_fastopen_defer_connect);
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700441
442/*
443 * The following code block is to deal with middle box issues with TFO:
444 * Middlebox firewall issues can potentially cause server's data being
445 * blackholed after a successful 3WHS using TFO.
446 * The proposed solution is to disable active TFO globally under the
447 * following circumstances:
448 * 1. client side TFO socket receives out of order FIN
449 * 2. client side TFO socket receives out of order RST
450 * We disable active side TFO globally for 1hr at first. Then if it
451 * happens again, we disable it for 2h, then 4h, 8h, ...
452 * And we reset the timeout back to 1hr when we see a successful active
453 * TFO connection with data exchanges.
454 */
455
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700456/* Disable active TFO and record current jiffies and
457 * tfo_active_disable_times
458 */
Wei Wang46c2fa32017-04-20 14:45:47 -0700459void tcp_fastopen_active_disable(struct sock *sk)
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700460{
Haishuang Yan3733be12017-09-27 11:35:43 +0800461 struct net *net = sock_net(sk);
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700462
Haishuang Yan3733be12017-09-27 11:35:43 +0800463 atomic_inc(&net->ipv4.tfo_active_disable_times);
464 net->ipv4.tfo_active_disable_stamp = jiffies;
465 NET_INC_STATS(net, LINUX_MIB_TCPFASTOPENBLACKHOLE);
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700466}
467
468/* Calculate timeout for tfo active disable
469 * Return true if we are still in the active TFO disable period
470 * Return false if timeout already expired and we should use active TFO
471 */
472bool tcp_fastopen_active_should_disable(struct sock *sk)
473{
Haishuang Yan3733be12017-09-27 11:35:43 +0800474 unsigned int tfo_bh_timeout = sock_net(sk)->ipv4.sysctl_tcp_fastopen_blackhole_timeout;
475 int tfo_da_times = atomic_read(&sock_net(sk)->ipv4.tfo_active_disable_times);
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700476 unsigned long timeout;
Haishuang Yan3733be12017-09-27 11:35:43 +0800477 int multiplier;
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700478
479 if (!tfo_da_times)
480 return false;
481
482 /* Limit timout to max: 2^6 * initial timeout */
483 multiplier = 1 << min(tfo_da_times - 1, 6);
Haishuang Yan3733be12017-09-27 11:35:43 +0800484 timeout = multiplier * tfo_bh_timeout * HZ;
485 if (time_before(jiffies, sock_net(sk)->ipv4.tfo_active_disable_stamp + timeout))
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700486 return true;
487
488 /* Mark check bit so we can check for successful active TFO
489 * condition and reset tfo_active_disable_times
490 */
491 tcp_sk(sk)->syn_fastopen_ch = 1;
492 return false;
493}
494
495/* Disable active TFO if FIN is the only packet in the ofo queue
496 * and no data is received.
497 * Also check if we can reset tfo_active_disable_times if data is
498 * received successfully on a marked active TFO sockets opened on
499 * a non-loopback interface
500 */
501void tcp_fastopen_active_disable_ofo_check(struct sock *sk)
502{
503 struct tcp_sock *tp = tcp_sk(sk);
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700504 struct dst_entry *dst;
Eric Dumazet18a4c0e2017-10-05 22:21:21 -0700505 struct sk_buff *skb;
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700506
507 if (!tp->syn_fastopen)
508 return;
509
510 if (!tp->data_segs_in) {
Eric Dumazet18a4c0e2017-10-05 22:21:21 -0700511 skb = skb_rb_first(&tp->out_of_order_queue);
512 if (skb && !skb_rb_next(skb)) {
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700513 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
Wei Wang46c2fa32017-04-20 14:45:47 -0700514 tcp_fastopen_active_disable(sk);
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700515 return;
516 }
517 }
518 } else if (tp->syn_fastopen_ch &&
Haishuang Yan3733be12017-09-27 11:35:43 +0800519 atomic_read(&sock_net(sk)->ipv4.tfo_active_disable_times)) {
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700520 dst = sk_dst_get(sk);
521 if (!(dst && dst->dev && (dst->dev->flags & IFF_LOOPBACK)))
Haishuang Yan3733be12017-09-27 11:35:43 +0800522 atomic_set(&sock_net(sk)->ipv4.tfo_active_disable_times, 0);
Wei Wangcf1ef3f2017-04-20 14:45:46 -0700523 dst_release(dst);
524 }
525}