]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
cf80e0e4 | 2 | #include <linux/crypto.h> |
10467163 | 3 | #include <linux/err.h> |
2100c8d2 YC |
4 | #include <linux/init.h> |
5 | #include <linux/kernel.h> | |
10467163 JC |
6 | #include <linux/list.h> |
7 | #include <linux/tcp.h> | |
8 | #include <linux/rcupdate.h> | |
9 | #include <linux/rculist.h> | |
10 | #include <net/inetpeer.h> | |
11 | #include <net/tcp.h> | |
2100c8d2 | 12 | |
43713848 | 13 | void tcp_fastopen_init_key_once(struct net *net) |
222e83d2 | 14 | { |
43713848 HY |
15 | u8 key[TCP_FASTOPEN_KEY_LENGTH]; |
16 | struct tcp_fastopen_context *ctxt; | |
17 | ||
18 | rcu_read_lock(); | |
19 | ctxt = rcu_dereference(net->ipv4.tcp_fastopen_ctx); | |
20 | if (ctxt) { | |
21 | rcu_read_unlock(); | |
22 | return; | |
23 | } | |
24 | rcu_read_unlock(); | |
222e83d2 HFS |
25 | |
26 | /* tcp_fastopen_reset_cipher publishes the new context | |
27 | * atomically, so we allow this race happening here. | |
28 | * | |
29 | * All call sites of tcp_fastopen_cookie_gen also check | |
30 | * for a valid cookie, so this is an acceptable risk. | |
31 | */ | |
43713848 | 32 | get_random_bytes(key, sizeof(key)); |
1fba70e5 | 33 | tcp_fastopen_reset_cipher(net, NULL, key, sizeof(key)); |
222e83d2 HFS |
34 | } |
35 | ||
10467163 JC |
36 | static void tcp_fastopen_ctx_free(struct rcu_head *head) |
37 | { | |
38 | struct tcp_fastopen_context *ctx = | |
39 | container_of(head, struct tcp_fastopen_context, rcu); | |
40 | crypto_free_cipher(ctx->tfm); | |
41 | kfree(ctx); | |
42 | } | |
43 | ||
1fba70e5 YC |
44 | void tcp_fastopen_destroy_cipher(struct sock *sk) |
45 | { | |
46 | struct tcp_fastopen_context *ctx; | |
47 | ||
48 | ctx = rcu_dereference_protected( | |
49 | inet_csk(sk)->icsk_accept_queue.fastopenq.ctx, 1); | |
50 | if (ctx) | |
51 | call_rcu(&ctx->rcu, tcp_fastopen_ctx_free); | |
52 | } | |
53 | ||
43713848 HY |
54 | void tcp_fastopen_ctx_destroy(struct net *net) |
55 | { | |
56 | struct tcp_fastopen_context *ctxt; | |
57 | ||
58 | spin_lock(&net->ipv4.tcp_fastopen_ctx_lock); | |
59 | ||
60 | ctxt = rcu_dereference_protected(net->ipv4.tcp_fastopen_ctx, | |
61 | lockdep_is_held(&net->ipv4.tcp_fastopen_ctx_lock)); | |
62 | rcu_assign_pointer(net->ipv4.tcp_fastopen_ctx, NULL); | |
63 | spin_unlock(&net->ipv4.tcp_fastopen_ctx_lock); | |
64 | ||
65 | if (ctxt) | |
66 | call_rcu(&ctxt->rcu, tcp_fastopen_ctx_free); | |
67 | } | |
68 | ||
1fba70e5 YC |
69 | int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, |
70 | void *key, unsigned int len) | |
10467163 | 71 | { |
10467163 | 72 | struct tcp_fastopen_context *ctx, *octx; |
1fba70e5 YC |
73 | struct fastopen_queue *q; |
74 | int err; | |
10467163 JC |
75 | |
76 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); | |
77 | if (!ctx) | |
78 | return -ENOMEM; | |
79 | ctx->tfm = crypto_alloc_cipher("aes", 0, 0); | |
80 | ||
81 | if (IS_ERR(ctx->tfm)) { | |
82 | err = PTR_ERR(ctx->tfm); | |
83 | error: kfree(ctx); | |
84 | pr_err("TCP: TFO aes cipher alloc error: %d\n", err); | |
85 | return err; | |
86 | } | |
87 | err = crypto_cipher_setkey(ctx->tfm, key, len); | |
88 | if (err) { | |
89 | pr_err("TCP: TFO cipher key error: %d\n", err); | |
90 | crypto_free_cipher(ctx->tfm); | |
91 | goto error; | |
92 | } | |
93 | memcpy(ctx->key, key, len); | |
94 | ||
10467163 | 95 | |
9eba9353 | 96 | spin_lock(&net->ipv4.tcp_fastopen_ctx_lock); |
1fba70e5 YC |
97 | if (sk) { |
98 | q = &inet_csk(sk)->icsk_accept_queue.fastopenq; | |
1fba70e5 | 99 | octx = rcu_dereference_protected(q->ctx, |
9eba9353 | 100 | lockdep_is_held(&net->ipv4.tcp_fastopen_ctx_lock)); |
1fba70e5 | 101 | rcu_assign_pointer(q->ctx, ctx); |
1fba70e5 | 102 | } else { |
1fba70e5 YC |
103 | octx = rcu_dereference_protected(net->ipv4.tcp_fastopen_ctx, |
104 | lockdep_is_held(&net->ipv4.tcp_fastopen_ctx_lock)); | |
105 | rcu_assign_pointer(net->ipv4.tcp_fastopen_ctx, ctx); | |
1fba70e5 | 106 | } |
9eba9353 | 107 | spin_unlock(&net->ipv4.tcp_fastopen_ctx_lock); |
10467163 JC |
108 | |
109 | if (octx) | |
110 | call_rcu(&octx->rcu, tcp_fastopen_ctx_free); | |
111 | return err; | |
112 | } | |
113 | ||
1fba70e5 | 114 | static bool __tcp_fastopen_cookie_gen(struct sock *sk, const void *path, |
3a19ce0e | 115 | struct tcp_fastopen_cookie *foc) |
10467163 | 116 | { |
10467163 | 117 | struct tcp_fastopen_context *ctx; |
3a19ce0e | 118 | bool ok = false; |
10467163 JC |
119 | |
120 | rcu_read_lock(); | |
1fba70e5 YC |
121 | |
122 | ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); | |
123 | if (!ctx) | |
124 | ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); | |
125 | ||
10467163 | 126 | if (ctx) { |
3a19ce0e | 127 | crypto_cipher_encrypt_one(ctx->tfm, foc->val, path); |
10467163 | 128 | foc->len = TCP_FASTOPEN_COOKIE_SIZE; |
3a19ce0e | 129 | ok = true; |
10467163 JC |
130 | } |
131 | rcu_read_unlock(); | |
3a19ce0e DL |
132 | return ok; |
133 | } | |
134 | ||
135 | /* Generate the fastopen cookie by doing aes128 encryption on both | |
136 | * the source and destination addresses. Pad 0s for IPv4 or IPv4-mapped-IPv6 | |
137 | * addresses. For the longer IPv6 addresses use CBC-MAC. | |
138 | * | |
139 | * XXX (TFO) - refactor when TCP_FASTOPEN_COOKIE_SIZE != AES_BLOCK_SIZE. | |
140 | */ | |
1fba70e5 | 141 | static bool tcp_fastopen_cookie_gen(struct sock *sk, |
43713848 | 142 | struct request_sock *req, |
3a19ce0e DL |
143 | struct sk_buff *syn, |
144 | struct tcp_fastopen_cookie *foc) | |
145 | { | |
146 | if (req->rsk_ops->family == AF_INET) { | |
147 | const struct iphdr *iph = ip_hdr(syn); | |
148 | ||
149 | __be32 path[4] = { iph->saddr, iph->daddr, 0, 0 }; | |
1fba70e5 | 150 | return __tcp_fastopen_cookie_gen(sk, path, foc); |
3a19ce0e DL |
151 | } |
152 | ||
153 | #if IS_ENABLED(CONFIG_IPV6) | |
154 | if (req->rsk_ops->family == AF_INET6) { | |
155 | const struct ipv6hdr *ip6h = ipv6_hdr(syn); | |
156 | struct tcp_fastopen_cookie tmp; | |
157 | ||
1fba70e5 | 158 | if (__tcp_fastopen_cookie_gen(sk, &ip6h->saddr, &tmp)) { |
003c9410 | 159 | struct in6_addr *buf = &tmp.addr; |
41c91996 | 160 | int i; |
3a19ce0e DL |
161 | |
162 | for (i = 0; i < 4; i++) | |
163 | buf->s6_addr32[i] ^= ip6h->daddr.s6_addr32[i]; | |
1fba70e5 | 164 | return __tcp_fastopen_cookie_gen(sk, buf, foc); |
3a19ce0e DL |
165 | } |
166 | } | |
167 | #endif | |
168 | return false; | |
10467163 | 169 | } |
5b7ed089 | 170 | |
61d2bcae ED |
171 | |
172 | /* If an incoming SYN or SYNACK frame contains a payload and/or FIN, | |
173 | * queue this additional data / FIN. | |
174 | */ | |
175 | void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb) | |
176 | { | |
177 | struct tcp_sock *tp = tcp_sk(sk); | |
178 | ||
179 | if (TCP_SKB_CB(skb)->end_seq == tp->rcv_nxt) | |
180 | return; | |
181 | ||
182 | skb = skb_clone(skb, GFP_ATOMIC); | |
183 | if (!skb) | |
184 | return; | |
185 | ||
186 | skb_dst_drop(skb); | |
a44d6eac MKL |
187 | /* segs_in has been initialized to 1 in tcp_create_openreq_child(). |
188 | * Hence, reset segs_in to 0 before calling tcp_segs_in() | |
189 | * to avoid double counting. Also, tcp_segs_in() expects | |
190 | * skb->len to include the tcp_hdrlen. Hence, it should | |
191 | * be called before __skb_pull(). | |
192 | */ | |
193 | tp->segs_in = 0; | |
194 | tcp_segs_in(tp, skb); | |
61d2bcae | 195 | __skb_pull(skb, tcp_hdrlen(skb)); |
76061f63 | 196 | sk_forced_mem_schedule(sk, skb->truesize); |
61d2bcae ED |
197 | skb_set_owner_r(skb, sk); |
198 | ||
9d691539 ED |
199 | TCP_SKB_CB(skb)->seq++; |
200 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_SYN; | |
201 | ||
61d2bcae ED |
202 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
203 | __skb_queue_tail(&sk->sk_receive_queue, skb); | |
204 | tp->syn_data_acked = 1; | |
205 | ||
206 | /* u64_stats_update_begin(&tp->syncp) not needed here, | |
207 | * as we certainly are not changing upper 32bit value (0) | |
208 | */ | |
209 | tp->bytes_received = skb->len; | |
e3e17b77 ED |
210 | |
211 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) | |
212 | tcp_fin(sk); | |
61d2bcae ED |
213 | } |
214 | ||
7c85af88 ED |
215 | static struct sock *tcp_fastopen_create_child(struct sock *sk, |
216 | struct sk_buff *skb, | |
7c85af88 | 217 | struct request_sock *req) |
5b7ed089 | 218 | { |
17846376 | 219 | struct tcp_sock *tp; |
5b7ed089 | 220 | struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; |
5b7ed089 | 221 | struct sock *child; |
5e0724d0 | 222 | bool own_req; |
5b7ed089 YC |
223 | |
224 | req->num_retrans = 0; | |
225 | req->num_timeout = 0; | |
226 | req->sk = NULL; | |
227 | ||
5e0724d0 ED |
228 | child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL, |
229 | NULL, &own_req); | |
51456b29 | 230 | if (!child) |
7c85af88 | 231 | return NULL; |
5b7ed089 | 232 | |
0536fcc0 ED |
233 | spin_lock(&queue->fastopenq.lock); |
234 | queue->fastopenq.qlen++; | |
235 | spin_unlock(&queue->fastopenq.lock); | |
5b7ed089 YC |
236 | |
237 | /* Initialize the child socket. Have to fix some values to take | |
238 | * into account the child is a Fast Open socket and is created | |
239 | * only out of the bits carried in the SYN packet. | |
240 | */ | |
241 | tp = tcp_sk(child); | |
242 | ||
243 | tp->fastopen_rsk = req; | |
9439ce00 | 244 | tcp_rsk(req)->tfo_listener = true; |
5b7ed089 YC |
245 | |
246 | /* RFC1323: The window in SYN & SYN/ACK segments is never | |
247 | * scaled. So correct it appropriately. | |
248 | */ | |
249 | tp->snd_wnd = ntohs(tcp_hdr(skb)->window); | |
0dbd7ff3 | 250 | tp->max_window = tp->snd_wnd; |
5b7ed089 YC |
251 | |
252 | /* Activate the retrans timer so that SYNACK can be retransmitted. | |
ca6fb065 | 253 | * The request socket is not added to the ehash |
5b7ed089 YC |
254 | * because it's been added to the accept queue directly. |
255 | */ | |
256 | inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS, | |
257 | TCP_TIMEOUT_INIT, TCP_RTO_MAX); | |
258 | ||
41c6d650 | 259 | refcount_set(&req->rsk_refcnt, 2); |
5b7ed089 YC |
260 | |
261 | /* Now finish processing the fastopen child socket. */ | |
27204aaa | 262 | tcp_init_transfer(child, BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB); |
5b7ed089 | 263 | |
61d2bcae ED |
264 | tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; |
265 | ||
266 | tcp_fastopen_add_skb(child, skb); | |
267 | ||
268 | tcp_rsk(req)->rcv_nxt = tp->rcv_nxt; | |
28b346cb | 269 | tp->rcv_wup = tp->rcv_nxt; |
7656d842 ED |
270 | /* tcp_conn_request() is sending the SYNACK, |
271 | * and queues the child into listener accept queue. | |
7c85af88 | 272 | */ |
7c85af88 | 273 | return child; |
5b7ed089 | 274 | } |
5b7ed089 YC |
275 | |
276 | static bool tcp_fastopen_queue_check(struct sock *sk) | |
277 | { | |
278 | struct fastopen_queue *fastopenq; | |
279 | ||
280 | /* Make sure the listener has enabled fastopen, and we don't | |
281 | * exceed the max # of pending TFO requests allowed before trying | |
282 | * to validating the cookie in order to avoid burning CPU cycles | |
283 | * unnecessarily. | |
284 | * | |
285 | * XXX (TFO) - The implication of checking the max_qlen before | |
286 | * processing a cookie request is that clients can't differentiate | |
287 | * between qlen overflow causing Fast Open to be disabled | |
288 | * temporarily vs a server not supporting Fast Open at all. | |
289 | */ | |
0536fcc0 ED |
290 | fastopenq = &inet_csk(sk)->icsk_accept_queue.fastopenq; |
291 | if (fastopenq->max_qlen == 0) | |
5b7ed089 YC |
292 | return false; |
293 | ||
294 | if (fastopenq->qlen >= fastopenq->max_qlen) { | |
295 | struct request_sock *req1; | |
296 | spin_lock(&fastopenq->lock); | |
297 | req1 = fastopenq->rskq_rst_head; | |
fa76ce73 | 298 | if (!req1 || time_after(req1->rsk_timer.expires, jiffies)) { |
02a1d6e7 ED |
299 | __NET_INC_STATS(sock_net(sk), |
300 | LINUX_MIB_TCPFASTOPENLISTENOVERFLOW); | |
c10d9310 | 301 | spin_unlock(&fastopenq->lock); |
5b7ed089 YC |
302 | return false; |
303 | } | |
304 | fastopenq->rskq_rst_head = req1->dl_next; | |
305 | fastopenq->qlen--; | |
306 | spin_unlock(&fastopenq->lock); | |
13854e5a | 307 | reqsk_put(req1); |
5b7ed089 YC |
308 | } |
309 | return true; | |
310 | } | |
311 | ||
71c02379 CP |
312 | static bool tcp_fastopen_no_cookie(const struct sock *sk, |
313 | const struct dst_entry *dst, | |
314 | int flag) | |
315 | { | |
316 | return (sock_net(sk)->ipv4.sysctl_tcp_fastopen & flag) || | |
317 | tcp_sk(sk)->fastopen_no_cookie || | |
318 | (dst && dst_metric(dst, RTAX_FASTOPEN_NO_COOKIE)); | |
319 | } | |
320 | ||
89278c9d YC |
321 | /* Returns true if we should perform Fast Open on the SYN. The cookie (foc) |
322 | * may be updated and return the client in the SYN-ACK later. E.g., Fast Open | |
323 | * cookie request (foc->len == 0). | |
324 | */ | |
7c85af88 ED |
325 | struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, |
326 | struct request_sock *req, | |
71c02379 CP |
327 | struct tcp_fastopen_cookie *foc, |
328 | const struct dst_entry *dst) | |
5b7ed089 | 329 | { |
89278c9d | 330 | bool syn_data = TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1; |
e1cfcbe8 HY |
331 | int tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen; |
332 | struct tcp_fastopen_cookie valid_foc = { .len = -1 }; | |
7c85af88 | 333 | struct sock *child; |
5b7ed089 | 334 | |
531c94a9 | 335 | if (foc->len == 0) /* Client requests a cookie */ |
c10d9310 | 336 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENCOOKIEREQD); |
531c94a9 | 337 | |
e1cfcbe8 | 338 | if (!((tcp_fastopen & TFO_SERVER_ENABLE) && |
89278c9d YC |
339 | (syn_data || foc->len >= 0) && |
340 | tcp_fastopen_queue_check(sk))) { | |
341 | foc->len = -1; | |
7c85af88 | 342 | return NULL; |
5b7ed089 YC |
343 | } |
344 | ||
71c02379 CP |
345 | if (syn_data && |
346 | tcp_fastopen_no_cookie(sk, dst, TFO_SERVER_COOKIE_NOT_REQD)) | |
89278c9d YC |
347 | goto fastopen; |
348 | ||
531c94a9 | 349 | if (foc->len >= 0 && /* Client presents or requests a cookie */ |
1fba70e5 | 350 | tcp_fastopen_cookie_gen(sk, req, skb, &valid_foc) && |
3a19ce0e | 351 | foc->len == TCP_FASTOPEN_COOKIE_SIZE && |
89278c9d YC |
352 | foc->len == valid_foc.len && |
353 | !memcmp(foc->val, valid_foc.val, foc->len)) { | |
843f4a55 YC |
354 | /* Cookie is valid. Create a (full) child socket to accept |
355 | * the data in SYN before returning a SYN-ACK to ack the | |
356 | * data. If we fail to create the socket, fall back and | |
357 | * ack the ISN only but includes the same cookie. | |
358 | * | |
359 | * Note: Data-less SYN with valid cookie is allowed to send | |
360 | * data in SYN_RECV state. | |
361 | */ | |
89278c9d | 362 | fastopen: |
11199369 | 363 | child = tcp_fastopen_create_child(sk, skb, req); |
7c85af88 | 364 | if (child) { |
843f4a55 | 365 | foc->len = -1; |
c10d9310 ED |
366 | NET_INC_STATS(sock_net(sk), |
367 | LINUX_MIB_TCPFASTOPENPASSIVE); | |
7c85af88 | 368 | return child; |
843f4a55 | 369 | } |
c10d9310 | 370 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVEFAIL); |
531c94a9 | 371 | } else if (foc->len > 0) /* Client presents an invalid cookie */ |
c10d9310 | 372 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVEFAIL); |
89278c9d | 373 | |
7f9b838b | 374 | valid_foc.exp = foc->exp; |
89278c9d | 375 | *foc = valid_foc; |
7c85af88 | 376 | return NULL; |
5b7ed089 | 377 | } |
065263f4 WW |
378 | |
379 | bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, | |
380 | struct tcp_fastopen_cookie *cookie) | |
381 | { | |
382 | unsigned long last_syn_loss = 0; | |
71c02379 | 383 | const struct dst_entry *dst; |
065263f4 WW |
384 | int syn_loss = 0; |
385 | ||
386 | tcp_fastopen_cache_get(sk, mss, cookie, &syn_loss, &last_syn_loss); | |
387 | ||
388 | /* Recurring FO SYN losses: no cookie or data in SYN */ | |
389 | if (syn_loss > 1 && | |
390 | time_before(jiffies, last_syn_loss + (60*HZ << syn_loss))) { | |
391 | cookie->len = -1; | |
392 | return false; | |
393 | } | |
cf1ef3f0 WW |
394 | |
395 | /* Firewall blackhole issue check */ | |
396 | if (tcp_fastopen_active_should_disable(sk)) { | |
397 | cookie->len = -1; | |
398 | return false; | |
399 | } | |
400 | ||
71c02379 CP |
401 | dst = __sk_dst_get(sk); |
402 | ||
403 | if (tcp_fastopen_no_cookie(sk, dst, TFO_CLIENT_NO_COOKIE)) { | |
065263f4 WW |
404 | cookie->len = -1; |
405 | return true; | |
406 | } | |
407 | return cookie->len > 0; | |
408 | } | |
19f6d3f3 WW |
409 | |
410 | /* This function checks if we want to defer sending SYN until the first | |
411 | * write(). We defer under the following conditions: | |
412 | * 1. fastopen_connect sockopt is set | |
413 | * 2. we have a valid cookie | |
414 | * Return value: return true if we want to defer until application writes data | |
415 | * return false if we want to send out SYN immediately | |
416 | */ | |
417 | bool tcp_fastopen_defer_connect(struct sock *sk, int *err) | |
418 | { | |
419 | struct tcp_fastopen_cookie cookie = { .len = 0 }; | |
420 | struct tcp_sock *tp = tcp_sk(sk); | |
421 | u16 mss; | |
422 | ||
423 | if (tp->fastopen_connect && !tp->fastopen_req) { | |
424 | if (tcp_fastopen_cookie_check(sk, &mss, &cookie)) { | |
425 | inet_sk(sk)->defer_connect = 1; | |
426 | return true; | |
427 | } | |
428 | ||
429 | /* Alloc fastopen_req in order for FO option to be included | |
430 | * in SYN | |
431 | */ | |
432 | tp->fastopen_req = kzalloc(sizeof(*tp->fastopen_req), | |
433 | sk->sk_allocation); | |
434 | if (tp->fastopen_req) | |
435 | tp->fastopen_req->cookie = cookie; | |
436 | else | |
437 | *err = -ENOBUFS; | |
438 | } | |
439 | return false; | |
440 | } | |
441 | EXPORT_SYMBOL(tcp_fastopen_defer_connect); | |
cf1ef3f0 WW |
442 | |
443 | /* | |
444 | * The following code block is to deal with middle box issues with TFO: | |
445 | * Middlebox firewall issues can potentially cause server's data being | |
446 | * blackholed after a successful 3WHS using TFO. | |
447 | * The proposed solution is to disable active TFO globally under the | |
448 | * following circumstances: | |
449 | * 1. client side TFO socket receives out of order FIN | |
450 | * 2. client side TFO socket receives out of order RST | |
451 | * We disable active side TFO globally for 1hr at first. Then if it | |
452 | * happens again, we disable it for 2h, then 4h, 8h, ... | |
453 | * And we reset the timeout back to 1hr when we see a successful active | |
454 | * TFO connection with data exchanges. | |
455 | */ | |
456 | ||
cf1ef3f0 WW |
457 | /* Disable active TFO and record current jiffies and |
458 | * tfo_active_disable_times | |
459 | */ | |
46c2fa39 | 460 | void tcp_fastopen_active_disable(struct sock *sk) |
cf1ef3f0 | 461 | { |
3733be14 | 462 | struct net *net = sock_net(sk); |
cf1ef3f0 | 463 | |
3733be14 HY |
464 | atomic_inc(&net->ipv4.tfo_active_disable_times); |
465 | net->ipv4.tfo_active_disable_stamp = jiffies; | |
466 | NET_INC_STATS(net, LINUX_MIB_TCPFASTOPENBLACKHOLE); | |
cf1ef3f0 WW |
467 | } |
468 | ||
469 | /* Calculate timeout for tfo active disable | |
470 | * Return true if we are still in the active TFO disable period | |
471 | * Return false if timeout already expired and we should use active TFO | |
472 | */ | |
473 | bool tcp_fastopen_active_should_disable(struct sock *sk) | |
474 | { | |
3733be14 HY |
475 | unsigned int tfo_bh_timeout = sock_net(sk)->ipv4.sysctl_tcp_fastopen_blackhole_timeout; |
476 | int tfo_da_times = atomic_read(&sock_net(sk)->ipv4.tfo_active_disable_times); | |
cf1ef3f0 | 477 | unsigned long timeout; |
3733be14 | 478 | int multiplier; |
cf1ef3f0 WW |
479 | |
480 | if (!tfo_da_times) | |
481 | return false; | |
482 | ||
483 | /* Limit timout to max: 2^6 * initial timeout */ | |
484 | multiplier = 1 << min(tfo_da_times - 1, 6); | |
3733be14 HY |
485 | timeout = multiplier * tfo_bh_timeout * HZ; |
486 | if (time_before(jiffies, sock_net(sk)->ipv4.tfo_active_disable_stamp + timeout)) | |
cf1ef3f0 WW |
487 | return true; |
488 | ||
489 | /* Mark check bit so we can check for successful active TFO | |
490 | * condition and reset tfo_active_disable_times | |
491 | */ | |
492 | tcp_sk(sk)->syn_fastopen_ch = 1; | |
493 | return false; | |
494 | } | |
495 | ||
496 | /* Disable active TFO if FIN is the only packet in the ofo queue | |
497 | * and no data is received. | |
498 | * Also check if we can reset tfo_active_disable_times if data is | |
499 | * received successfully on a marked active TFO sockets opened on | |
500 | * a non-loopback interface | |
501 | */ | |
502 | void tcp_fastopen_active_disable_ofo_check(struct sock *sk) | |
503 | { | |
504 | struct tcp_sock *tp = tcp_sk(sk); | |
cf1ef3f0 | 505 | struct dst_entry *dst; |
18a4c0ea | 506 | struct sk_buff *skb; |
cf1ef3f0 WW |
507 | |
508 | if (!tp->syn_fastopen) | |
509 | return; | |
510 | ||
511 | if (!tp->data_segs_in) { | |
18a4c0ea ED |
512 | skb = skb_rb_first(&tp->out_of_order_queue); |
513 | if (skb && !skb_rb_next(skb)) { | |
cf1ef3f0 | 514 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { |
46c2fa39 | 515 | tcp_fastopen_active_disable(sk); |
cf1ef3f0 WW |
516 | return; |
517 | } | |
518 | } | |
519 | } else if (tp->syn_fastopen_ch && | |
3733be14 | 520 | atomic_read(&sock_net(sk)->ipv4.tfo_active_disable_times)) { |
cf1ef3f0 WW |
521 | dst = sk_dst_get(sk); |
522 | if (!(dst && dst->dev && (dst->dev->flags & IFF_LOOPBACK))) | |
3733be14 | 523 | atomic_set(&sock_net(sk)->ipv4.tfo_active_disable_times, 0); |
cf1ef3f0 WW |
524 | dst_release(dst); |
525 | } | |
526 | } |