blob: 4e777a3243f94457d9928e3967bb83947da563f6 [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
Yuchung Cheng0d41cca2013-10-31 09:19:32 -070012int sysctl_tcp_fastopen __read_mostly = TFO_CLIENT_ENABLE;
Jerry Chu10467162012-08-31 12:29:11 +000013
14struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
15
16static DEFINE_SPINLOCK(tcp_fastopen_ctx_lock);
17
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +020018void tcp_fastopen_init_key_once(bool publish)
19{
20 static u8 key[TCP_FASTOPEN_KEY_LENGTH];
21
22 /* tcp_fastopen_reset_cipher publishes the new context
23 * atomically, so we allow this race happening here.
24 *
25 * All call sites of tcp_fastopen_cookie_gen also check
26 * for a valid cookie, so this is an acceptable risk.
27 */
28 if (net_get_random_once(key, sizeof(key)) && publish)
29 tcp_fastopen_reset_cipher(key, sizeof(key));
30}
31
Jerry Chu10467162012-08-31 12:29:11 +000032static void tcp_fastopen_ctx_free(struct rcu_head *head)
33{
34 struct tcp_fastopen_context *ctx =
35 container_of(head, struct tcp_fastopen_context, rcu);
36 crypto_free_cipher(ctx->tfm);
37 kfree(ctx);
38}
39
40int tcp_fastopen_reset_cipher(void *key, unsigned int len)
41{
42 int err;
43 struct tcp_fastopen_context *ctx, *octx;
44
45 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
46 if (!ctx)
47 return -ENOMEM;
48 ctx->tfm = crypto_alloc_cipher("aes", 0, 0);
49
50 if (IS_ERR(ctx->tfm)) {
51 err = PTR_ERR(ctx->tfm);
52error: kfree(ctx);
53 pr_err("TCP: TFO aes cipher alloc error: %d\n", err);
54 return err;
55 }
56 err = crypto_cipher_setkey(ctx->tfm, key, len);
57 if (err) {
58 pr_err("TCP: TFO cipher key error: %d\n", err);
59 crypto_free_cipher(ctx->tfm);
60 goto error;
61 }
62 memcpy(ctx->key, key, len);
63
64 spin_lock(&tcp_fastopen_ctx_lock);
65
66 octx = rcu_dereference_protected(tcp_fastopen_ctx,
67 lockdep_is_held(&tcp_fastopen_ctx_lock));
68 rcu_assign_pointer(tcp_fastopen_ctx, ctx);
69 spin_unlock(&tcp_fastopen_ctx_lock);
70
71 if (octx)
72 call_rcu(&octx->rcu, tcp_fastopen_ctx_free);
73 return err;
74}
75
Daniel Lee3a19ce02014-05-11 20:22:13 -070076static bool __tcp_fastopen_cookie_gen(const void *path,
77 struct tcp_fastopen_cookie *foc)
Jerry Chu10467162012-08-31 12:29:11 +000078{
Jerry Chu10467162012-08-31 12:29:11 +000079 struct tcp_fastopen_context *ctx;
Daniel Lee3a19ce02014-05-11 20:22:13 -070080 bool ok = false;
Jerry Chu10467162012-08-31 12:29:11 +000081
82 rcu_read_lock();
83 ctx = rcu_dereference(tcp_fastopen_ctx);
84 if (ctx) {
Daniel Lee3a19ce02014-05-11 20:22:13 -070085 crypto_cipher_encrypt_one(ctx->tfm, foc->val, path);
Jerry Chu10467162012-08-31 12:29:11 +000086 foc->len = TCP_FASTOPEN_COOKIE_SIZE;
Daniel Lee3a19ce02014-05-11 20:22:13 -070087 ok = true;
Jerry Chu10467162012-08-31 12:29:11 +000088 }
89 rcu_read_unlock();
Daniel Lee3a19ce02014-05-11 20:22:13 -070090 return ok;
91}
92
93/* Generate the fastopen cookie by doing aes128 encryption on both
94 * the source and destination addresses. Pad 0s for IPv4 or IPv4-mapped-IPv6
95 * addresses. For the longer IPv6 addresses use CBC-MAC.
96 *
97 * XXX (TFO) - refactor when TCP_FASTOPEN_COOKIE_SIZE != AES_BLOCK_SIZE.
98 */
99static bool tcp_fastopen_cookie_gen(struct request_sock *req,
100 struct sk_buff *syn,
101 struct tcp_fastopen_cookie *foc)
102{
103 if (req->rsk_ops->family == AF_INET) {
104 const struct iphdr *iph = ip_hdr(syn);
105
106 __be32 path[4] = { iph->saddr, iph->daddr, 0, 0 };
107 return __tcp_fastopen_cookie_gen(path, foc);
108 }
109
110#if IS_ENABLED(CONFIG_IPV6)
111 if (req->rsk_ops->family == AF_INET6) {
112 const struct ipv6hdr *ip6h = ipv6_hdr(syn);
113 struct tcp_fastopen_cookie tmp;
114
115 if (__tcp_fastopen_cookie_gen(&ip6h->saddr, &tmp)) {
116 struct in6_addr *buf = (struct in6_addr *) tmp.val;
Li RongQing41c91992014-09-29 15:04:37 +0800117 int i;
Daniel Lee3a19ce02014-05-11 20:22:13 -0700118
119 for (i = 0; i < 4; i++)
120 buf->s6_addr32[i] ^= ip6h->daddr.s6_addr32[i];
121 return __tcp_fastopen_cookie_gen(buf, foc);
122 }
123 }
124#endif
125 return false;
Jerry Chu10467162012-08-31 12:29:11 +0000126}
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700127
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800128
129/* If an incoming SYN or SYNACK frame contains a payload and/or FIN,
130 * queue this additional data / FIN.
131 */
132void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb)
133{
134 struct tcp_sock *tp = tcp_sk(sk);
135
136 if (TCP_SKB_CB(skb)->end_seq == tp->rcv_nxt)
137 return;
138
139 skb = skb_clone(skb, GFP_ATOMIC);
140 if (!skb)
141 return;
142
143 skb_dst_drop(skb);
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -0700144 /* segs_in has been initialized to 1 in tcp_create_openreq_child().
145 * Hence, reset segs_in to 0 before calling tcp_segs_in()
146 * to avoid double counting. Also, tcp_segs_in() expects
147 * skb->len to include the tcp_hdrlen. Hence, it should
148 * be called before __skb_pull().
149 */
150 tp->segs_in = 0;
151 tcp_segs_in(tp, skb);
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800152 __skb_pull(skb, tcp_hdrlen(skb));
Eric Dumazet76061f62016-09-07 08:34:11 -0700153 sk_forced_mem_schedule(sk, skb->truesize);
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800154 skb_set_owner_r(skb, sk);
155
Eric Dumazet9d691532016-02-01 21:03:08 -0800156 TCP_SKB_CB(skb)->seq++;
157 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_SYN;
158
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800159 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
160 __skb_queue_tail(&sk->sk_receive_queue, skb);
161 tp->syn_data_acked = 1;
162
163 /* u64_stats_update_begin(&tp->syncp) not needed here,
164 * as we certainly are not changing upper 32bit value (0)
165 */
166 tp->bytes_received = skb->len;
Eric Dumazete3e17b72016-02-06 11:16:28 -0800167
168 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
169 tcp_fin(sk);
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800170}
171
Eric Dumazet7c85af82015-09-24 17:16:05 -0700172static struct sock *tcp_fastopen_create_child(struct sock *sk,
173 struct sk_buff *skb,
174 struct dst_entry *dst,
175 struct request_sock *req)
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700176{
Dave Jones17846372014-06-16 16:30:36 -0400177 struct tcp_sock *tp;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700178 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700179 struct sock *child;
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700180 bool own_req;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700181
182 req->num_retrans = 0;
183 req->num_timeout = 0;
184 req->sk = NULL;
185
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700186 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
187 NULL, &own_req);
Ian Morris51456b22015-04-03 09:17:26 +0100188 if (!child)
Eric Dumazet7c85af82015-09-24 17:16:05 -0700189 return NULL;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700190
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700191 spin_lock(&queue->fastopenq.lock);
192 queue->fastopenq.qlen++;
193 spin_unlock(&queue->fastopenq.lock);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700194
195 /* Initialize the child socket. Have to fix some values to take
196 * into account the child is a Fast Open socket and is created
197 * only out of the bits carried in the SYN packet.
198 */
199 tp = tcp_sk(child);
200
201 tp->fastopen_rsk = req;
Eric Dumazet9439ce02015-03-17 18:32:29 -0700202 tcp_rsk(req)->tfo_listener = true;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700203
204 /* RFC1323: The window in SYN & SYN/ACK segments is never
205 * scaled. So correct it appropriately.
206 */
207 tp->snd_wnd = ntohs(tcp_hdr(skb)->window);
208
209 /* Activate the retrans timer so that SYNACK can be retransmitted.
Eric Dumazetca6fb062015-10-02 11:43:35 -0700210 * The request socket is not added to the ehash
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700211 * because it's been added to the accept queue directly.
212 */
213 inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
214 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
215
Eric Dumazetca6fb062015-10-02 11:43:35 -0700216 atomic_set(&req->rsk_refcnt, 2);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700217
218 /* Now finish processing the fastopen child socket. */
219 inet_csk(child)->icsk_af_ops->rebuild_header(child);
220 tcp_init_congestion_control(child);
221 tcp_mtup_init(child);
222 tcp_init_metrics(child);
223 tcp_init_buffer_space(child);
224
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800225 tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
Eric Dumazetba34e6d2015-02-13 04:47:12 -0800226
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800227 tcp_fastopen_add_skb(child, skb);
Eric Dumazetd6549762015-05-21 21:51:19 -0700228
Eric Dumazet61d2bca2016-02-01 21:03:07 -0800229 tcp_rsk(req)->rcv_nxt = tp->rcv_nxt;
Neal Cardwell28b346c2016-08-30 11:55:23 -0400230 tp->rcv_wup = tp->rcv_nxt;
Eric Dumazet7656d842015-10-04 21:08:07 -0700231 /* tcp_conn_request() is sending the SYNACK,
232 * and queues the child into listener accept queue.
Eric Dumazet7c85af82015-09-24 17:16:05 -0700233 */
Eric Dumazet7c85af82015-09-24 17:16:05 -0700234 return child;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700235}
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700236
237static bool tcp_fastopen_queue_check(struct sock *sk)
238{
239 struct fastopen_queue *fastopenq;
240
241 /* Make sure the listener has enabled fastopen, and we don't
242 * exceed the max # of pending TFO requests allowed before trying
243 * to validating the cookie in order to avoid burning CPU cycles
244 * unnecessarily.
245 *
246 * XXX (TFO) - The implication of checking the max_qlen before
247 * processing a cookie request is that clients can't differentiate
248 * between qlen overflow causing Fast Open to be disabled
249 * temporarily vs a server not supporting Fast Open at all.
250 */
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700251 fastopenq = &inet_csk(sk)->icsk_accept_queue.fastopenq;
252 if (fastopenq->max_qlen == 0)
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700253 return false;
254
255 if (fastopenq->qlen >= fastopenq->max_qlen) {
256 struct request_sock *req1;
257 spin_lock(&fastopenq->lock);
258 req1 = fastopenq->rskq_rst_head;
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700259 if (!req1 || time_after(req1->rsk_timer.expires, jiffies)) {
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700260 __NET_INC_STATS(sock_net(sk),
261 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW);
Eric Dumazetc10d9312016-04-29 14:16:47 -0700262 spin_unlock(&fastopenq->lock);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700263 return false;
264 }
265 fastopenq->rskq_rst_head = req1->dl_next;
266 fastopenq->qlen--;
267 spin_unlock(&fastopenq->lock);
Eric Dumazet13854e52015-03-15 21:12:16 -0700268 reqsk_put(req1);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700269 }
270 return true;
271}
272
Yuchung Cheng89278c92014-05-11 20:22:10 -0700273/* Returns true if we should perform Fast Open on the SYN. The cookie (foc)
274 * may be updated and return the client in the SYN-ACK later. E.g., Fast Open
275 * cookie request (foc->len == 0).
276 */
Eric Dumazet7c85af82015-09-24 17:16:05 -0700277struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
278 struct request_sock *req,
279 struct tcp_fastopen_cookie *foc,
280 struct dst_entry *dst)
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700281{
Yuchung Cheng89278c92014-05-11 20:22:10 -0700282 struct tcp_fastopen_cookie valid_foc = { .len = -1 };
283 bool syn_data = TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1;
Eric Dumazet7c85af82015-09-24 17:16:05 -0700284 struct sock *child;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700285
Yuchung Cheng531c94a2015-02-09 12:35:23 -0800286 if (foc->len == 0) /* Client requests a cookie */
Eric Dumazetc10d9312016-04-29 14:16:47 -0700287 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENCOOKIEREQD);
Yuchung Cheng531c94a2015-02-09 12:35:23 -0800288
Yuchung Cheng89278c92014-05-11 20:22:10 -0700289 if (!((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) &&
290 (syn_data || foc->len >= 0) &&
291 tcp_fastopen_queue_check(sk))) {
292 foc->len = -1;
Eric Dumazet7c85af82015-09-24 17:16:05 -0700293 return NULL;
Yuchung Cheng89278c92014-05-11 20:22:10 -0700294 }
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700295
Yuchung Cheng89278c92014-05-11 20:22:10 -0700296 if (syn_data && (sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_REQD))
297 goto fastopen;
298
Yuchung Cheng531c94a2015-02-09 12:35:23 -0800299 if (foc->len >= 0 && /* Client presents or requests a cookie */
300 tcp_fastopen_cookie_gen(req, skb, &valid_foc) &&
Daniel Lee3a19ce02014-05-11 20:22:13 -0700301 foc->len == TCP_FASTOPEN_COOKIE_SIZE &&
Yuchung Cheng89278c92014-05-11 20:22:10 -0700302 foc->len == valid_foc.len &&
303 !memcmp(foc->val, valid_foc.val, foc->len)) {
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700304 /* Cookie is valid. Create a (full) child socket to accept
305 * the data in SYN before returning a SYN-ACK to ack the
306 * data. If we fail to create the socket, fall back and
307 * ack the ISN only but includes the same cookie.
308 *
309 * Note: Data-less SYN with valid cookie is allowed to send
310 * data in SYN_RECV state.
311 */
Yuchung Cheng89278c92014-05-11 20:22:10 -0700312fastopen:
Eric Dumazet7c85af82015-09-24 17:16:05 -0700313 child = tcp_fastopen_create_child(sk, skb, dst, req);
314 if (child) {
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700315 foc->len = -1;
Eric Dumazetc10d9312016-04-29 14:16:47 -0700316 NET_INC_STATS(sock_net(sk),
317 LINUX_MIB_TCPFASTOPENPASSIVE);
Eric Dumazet7c85af82015-09-24 17:16:05 -0700318 return child;
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700319 }
Eric Dumazetc10d9312016-04-29 14:16:47 -0700320 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
Yuchung Cheng531c94a2015-02-09 12:35:23 -0800321 } else if (foc->len > 0) /* Client presents an invalid cookie */
Eric Dumazetc10d9312016-04-29 14:16:47 -0700322 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700323
Daniel Lee7f9b8382015-04-06 14:37:26 -0700324 valid_foc.exp = foc->exp;
Yuchung Cheng89278c92014-05-11 20:22:10 -0700325 *foc = valid_foc;
Eric Dumazet7c85af82015-09-24 17:16:05 -0700326 return NULL;
Yuchung Cheng5b7ed082014-05-11 20:22:09 -0700327}