Linux Audio

Check our new training course

Loading...
v3.15
 
 
 1#include <linux/err.h>
 2#include <linux/init.h>
 3#include <linux/kernel.h>
 4#include <linux/list.h>
 5#include <linux/tcp.h>
 6#include <linux/rcupdate.h>
 7#include <linux/rculist.h>
 8#include <net/inetpeer.h>
 9#include <net/tcp.h>
10
11int sysctl_tcp_fastopen __read_mostly = TFO_CLIENT_ENABLE;
12
13struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
14
15static DEFINE_SPINLOCK(tcp_fastopen_ctx_lock);
16
17void tcp_fastopen_init_key_once(bool publish)
18{
19	static u8 key[TCP_FASTOPEN_KEY_LENGTH];
 
 
 
 
 
 
 
 
 
20
21	/* tcp_fastopen_reset_cipher publishes the new context
22	 * atomically, so we allow this race happening here.
23	 *
24	 * All call sites of tcp_fastopen_cookie_gen also check
25	 * for a valid cookie, so this is an acceptable risk.
26	 */
27	if (net_get_random_once(key, sizeof(key)) && publish)
28		tcp_fastopen_reset_cipher(key, sizeof(key));
29}
30
31static void tcp_fastopen_ctx_free(struct rcu_head *head)
32{
33	struct tcp_fastopen_context *ctx =
34	    container_of(head, struct tcp_fastopen_context, rcu);
35	crypto_free_cipher(ctx->tfm);
36	kfree(ctx);
37}
38
39int tcp_fastopen_reset_cipher(void *key, unsigned int len)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
40{
41	int err;
42	struct tcp_fastopen_context *ctx, *octx;
 
 
43
44	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
45	if (!ctx)
46		return -ENOMEM;
47	ctx->tfm = crypto_alloc_cipher("aes", 0, 0);
 
 
 
 
 
 
 
 
 
 
 
48
49	if (IS_ERR(ctx->tfm)) {
50		err = PTR_ERR(ctx->tfm);
51error:		kfree(ctx);
52		pr_err("TCP: TFO aes cipher alloc error: %d\n", err);
53		return err;
54	}
55	err = crypto_cipher_setkey(ctx->tfm, key, len);
56	if (err) {
57		pr_err("TCP: TFO cipher key error: %d\n", err);
58		crypto_free_cipher(ctx->tfm);
59		goto error;
60	}
61	memcpy(ctx->key, key, len);
62
63	spin_lock(&tcp_fastopen_ctx_lock);
64
65	octx = rcu_dereference_protected(tcp_fastopen_ctx,
66				lockdep_is_held(&tcp_fastopen_ctx_lock));
67	rcu_assign_pointer(tcp_fastopen_ctx, ctx);
68	spin_unlock(&tcp_fastopen_ctx_lock);
69
70	if (octx)
71		call_rcu(&octx->rcu, tcp_fastopen_ctx_free);
 
72	return err;
73}
74
75/* Computes the fastopen cookie for the IP path.
76 * The path is a 128 bits long (pad with zeros for IPv4).
77 *
78 * The caller must check foc->len to determine if a valid cookie
79 * has been generated successfully.
80*/
81void tcp_fastopen_cookie_gen(__be32 src, __be32 dst,
82			     struct tcp_fastopen_cookie *foc)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
83{
84	__be32 path[4] = { src, dst, 0, 0 };
85	struct tcp_fastopen_context *ctx;
86
87	tcp_fastopen_init_key_once(true);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
88
89	rcu_read_lock();
90	ctx = rcu_dereference(tcp_fastopen_ctx);
91	if (ctx) {
92		crypto_cipher_encrypt_one(ctx->tfm, foc->val, (__u8 *)path);
93		foc->len = TCP_FASTOPEN_COOKIE_SIZE;
 
 
 
 
 
 
94	}
 
95	rcu_read_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
96}
v5.4
  1// SPDX-License-Identifier: GPL-2.0
  2#include <linux/crypto.h>
  3#include <linux/err.h>
  4#include <linux/init.h>
  5#include <linux/kernel.h>
  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>
 12
 13void tcp_fastopen_init_key_once(struct net *net)
 
 
 
 
 
 
 14{
 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();
 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	 */
 32	get_random_bytes(key, sizeof(key));
 33	tcp_fastopen_reset_cipher(net, NULL, key, NULL);
 34}
 35
 36static 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
 41	kzfree(ctx);
 42}
 43
 44void 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
 54void 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
 69int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
 70			      void *primary_key, void *backup_key)
 71{
 
 72	struct tcp_fastopen_context *ctx, *octx;
 73	struct fastopen_queue *q;
 74	int err = 0;
 75
 76	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 77	if (!ctx) {
 78		err = -ENOMEM;
 79		goto out;
 80	}
 81
 82	ctx->key[0].key[0] = get_unaligned_le64(primary_key);
 83	ctx->key[0].key[1] = get_unaligned_le64(primary_key + 8);
 84	if (backup_key) {
 85		ctx->key[1].key[0] = get_unaligned_le64(backup_key);
 86		ctx->key[1].key[1] = get_unaligned_le64(backup_key + 8);
 87		ctx->num = 2;
 88	} else {
 89		ctx->num = 1;
 90	}
 91
 92	spin_lock(&net->ipv4.tcp_fastopen_ctx_lock);
 93	if (sk) {
 94		q = &inet_csk(sk)->icsk_accept_queue.fastopenq;
 95		octx = rcu_dereference_protected(q->ctx,
 96			lockdep_is_held(&net->ipv4.tcp_fastopen_ctx_lock));
 97		rcu_assign_pointer(q->ctx, ctx);
 98	} else {
 99		octx = rcu_dereference_protected(net->ipv4.tcp_fastopen_ctx,
100			lockdep_is_held(&net->ipv4.tcp_fastopen_ctx_lock));
101		rcu_assign_pointer(net->ipv4.tcp_fastopen_ctx, ctx);
102	}
103	spin_unlock(&net->ipv4.tcp_fastopen_ctx_lock);
 
 
 
 
 
 
 
 
104
105	if (octx)
106		call_rcu(&octx->rcu, tcp_fastopen_ctx_free);
107out:
108	return err;
109}
110
111static bool __tcp_fastopen_cookie_gen_cipher(struct request_sock *req,
112					     struct sk_buff *syn,
113					     const siphash_key_t *key,
114					     struct tcp_fastopen_cookie *foc)
115{
116	BUILD_BUG_ON(TCP_FASTOPEN_COOKIE_SIZE != sizeof(u64));
117
118	if (req->rsk_ops->family == AF_INET) {
119		const struct iphdr *iph = ip_hdr(syn);
120
121		foc->val[0] = cpu_to_le64(siphash(&iph->saddr,
122					  sizeof(iph->saddr) +
123					  sizeof(iph->daddr),
124					  key));
125		foc->len = TCP_FASTOPEN_COOKIE_SIZE;
126		return true;
127	}
128#if IS_ENABLED(CONFIG_IPV6)
129	if (req->rsk_ops->family == AF_INET6) {
130		const struct ipv6hdr *ip6h = ipv6_hdr(syn);
131
132		foc->val[0] = cpu_to_le64(siphash(&ip6h->saddr,
133					  sizeof(ip6h->saddr) +
134					  sizeof(ip6h->daddr),
135					  key));
136		foc->len = TCP_FASTOPEN_COOKIE_SIZE;
137		return true;
138	}
139#endif
140	return false;
141}
142
143/* Generate the fastopen cookie by applying SipHash to both the source and
144 * destination addresses.
145 */
146static void tcp_fastopen_cookie_gen(struct sock *sk,
147				    struct request_sock *req,
148				    struct sk_buff *syn,
149				    struct tcp_fastopen_cookie *foc)
150{
 
151	struct tcp_fastopen_context *ctx;
152
153	rcu_read_lock();
154	ctx = tcp_fastopen_get_ctx(sk);
155	if (ctx)
156		__tcp_fastopen_cookie_gen_cipher(req, syn, &ctx->key[0], foc);
157	rcu_read_unlock();
158}
159
160/* If an incoming SYN or SYNACK frame contains a payload and/or FIN,
161 * queue this additional data / FIN.
162 */
163void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb)
164{
165	struct tcp_sock *tp = tcp_sk(sk);
166
167	if (TCP_SKB_CB(skb)->end_seq == tp->rcv_nxt)
168		return;
169
170	skb = skb_clone(skb, GFP_ATOMIC);
171	if (!skb)
172		return;
173
174	skb_dst_drop(skb);
175	/* segs_in has been initialized to 1 in tcp_create_openreq_child().
176	 * Hence, reset segs_in to 0 before calling tcp_segs_in()
177	 * to avoid double counting.  Also, tcp_segs_in() expects
178	 * skb->len to include the tcp_hdrlen.  Hence, it should
179	 * be called before __skb_pull().
180	 */
181	tp->segs_in = 0;
182	tcp_segs_in(tp, skb);
183	__skb_pull(skb, tcp_hdrlen(skb));
184	sk_forced_mem_schedule(sk, skb->truesize);
185	skb_set_owner_r(skb, sk);
186
187	TCP_SKB_CB(skb)->seq++;
188	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_SYN;
189
190	tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
191	__skb_queue_tail(&sk->sk_receive_queue, skb);
192	tp->syn_data_acked = 1;
193
194	/* u64_stats_update_begin(&tp->syncp) not needed here,
195	 * as we certainly are not changing upper 32bit value (0)
196	 */
197	tp->bytes_received = skb->len;
198
199	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
200		tcp_fin(sk);
201}
202
203/* returns 0 - no key match, 1 for primary, 2 for backup */
204static int tcp_fastopen_cookie_gen_check(struct sock *sk,
205					 struct request_sock *req,
206					 struct sk_buff *syn,
207					 struct tcp_fastopen_cookie *orig,
208					 struct tcp_fastopen_cookie *valid_foc)
209{
210	struct tcp_fastopen_cookie search_foc = { .len = -1 };
211	struct tcp_fastopen_cookie *foc = valid_foc;
212	struct tcp_fastopen_context *ctx;
213	int i, ret = 0;
214
215	rcu_read_lock();
216	ctx = tcp_fastopen_get_ctx(sk);
217	if (!ctx)
218		goto out;
219	for (i = 0; i < tcp_fastopen_context_len(ctx); i++) {
220		__tcp_fastopen_cookie_gen_cipher(req, syn, &ctx->key[i], foc);
221		if (tcp_fastopen_cookie_match(foc, orig)) {
222			ret = i + 1;
223			goto out;
224		}
225		foc = &search_foc;
226	}
227out:
228	rcu_read_unlock();
229	return ret;
230}
231
232static struct sock *tcp_fastopen_create_child(struct sock *sk,
233					      struct sk_buff *skb,
234					      struct request_sock *req)
235{
236	struct tcp_sock *tp;
237	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
238	struct sock *child;
239	bool own_req;
240
241	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
242							 NULL, &own_req);
243	if (!child)
244		return NULL;
245
246	spin_lock(&queue->fastopenq.lock);
247	queue->fastopenq.qlen++;
248	spin_unlock(&queue->fastopenq.lock);
249
250	/* Initialize the child socket. Have to fix some values to take
251	 * into account the child is a Fast Open socket and is created
252	 * only out of the bits carried in the SYN packet.
253	 */
254	tp = tcp_sk(child);
255
256	rcu_assign_pointer(tp->fastopen_rsk, req);
257	tcp_rsk(req)->tfo_listener = true;
258
259	/* RFC1323: The window in SYN & SYN/ACK segments is never
260	 * scaled. So correct it appropriately.
261	 */
262	tp->snd_wnd = ntohs(tcp_hdr(skb)->window);
263	tp->max_window = tp->snd_wnd;
264
265	/* Activate the retrans timer so that SYNACK can be retransmitted.
266	 * The request socket is not added to the ehash
267	 * because it's been added to the accept queue directly.
268	 */
269	inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
270				  TCP_TIMEOUT_INIT, TCP_RTO_MAX);
271
272	refcount_set(&req->rsk_refcnt, 2);
273
274	/* Now finish processing the fastopen child socket. */
275	tcp_init_transfer(child, BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB);
276
277	tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
278
279	tcp_fastopen_add_skb(child, skb);
280
281	tcp_rsk(req)->rcv_nxt = tp->rcv_nxt;
282	tp->rcv_wup = tp->rcv_nxt;
283	/* tcp_conn_request() is sending the SYNACK,
284	 * and queues the child into listener accept queue.
285	 */
286	return child;
287}
288
289static bool tcp_fastopen_queue_check(struct sock *sk)
290{
291	struct fastopen_queue *fastopenq;
292
293	/* Make sure the listener has enabled fastopen, and we don't
294	 * exceed the max # of pending TFO requests allowed before trying
295	 * to validating the cookie in order to avoid burning CPU cycles
296	 * unnecessarily.
297	 *
298	 * XXX (TFO) - The implication of checking the max_qlen before
299	 * processing a cookie request is that clients can't differentiate
300	 * between qlen overflow causing Fast Open to be disabled
301	 * temporarily vs a server not supporting Fast Open at all.
302	 */
303	fastopenq = &inet_csk(sk)->icsk_accept_queue.fastopenq;
304	if (fastopenq->max_qlen == 0)
305		return false;
306
307	if (fastopenq->qlen >= fastopenq->max_qlen) {
308		struct request_sock *req1;
309		spin_lock(&fastopenq->lock);
310		req1 = fastopenq->rskq_rst_head;
311		if (!req1 || time_after(req1->rsk_timer.expires, jiffies)) {
312			__NET_INC_STATS(sock_net(sk),
313					LINUX_MIB_TCPFASTOPENLISTENOVERFLOW);
314			spin_unlock(&fastopenq->lock);
315			return false;
316		}
317		fastopenq->rskq_rst_head = req1->dl_next;
318		fastopenq->qlen--;
319		spin_unlock(&fastopenq->lock);
320		reqsk_put(req1);
321	}
322	return true;
323}
324
325static bool tcp_fastopen_no_cookie(const struct sock *sk,
326				   const struct dst_entry *dst,
327				   int flag)
328{
329	return (sock_net(sk)->ipv4.sysctl_tcp_fastopen & flag) ||
330	       tcp_sk(sk)->fastopen_no_cookie ||
331	       (dst && dst_metric(dst, RTAX_FASTOPEN_NO_COOKIE));
332}
333
334/* Returns true if we should perform Fast Open on the SYN. The cookie (foc)
335 * may be updated and return the client in the SYN-ACK later. E.g., Fast Open
336 * cookie request (foc->len == 0).
337 */
338struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
339			      struct request_sock *req,
340			      struct tcp_fastopen_cookie *foc,
341			      const struct dst_entry *dst)
342{
343	bool syn_data = TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1;
344	int tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
345	struct tcp_fastopen_cookie valid_foc = { .len = -1 };
346	struct sock *child;
347	int ret = 0;
348
349	if (foc->len == 0) /* Client requests a cookie */
350		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENCOOKIEREQD);
351
352	if (!((tcp_fastopen & TFO_SERVER_ENABLE) &&
353	      (syn_data || foc->len >= 0) &&
354	      tcp_fastopen_queue_check(sk))) {
355		foc->len = -1;
356		return NULL;
357	}
358
359	if (syn_data &&
360	    tcp_fastopen_no_cookie(sk, dst, TFO_SERVER_COOKIE_NOT_REQD))
361		goto fastopen;
362
363	if (foc->len == 0) {
364		/* Client requests a cookie. */
365		tcp_fastopen_cookie_gen(sk, req, skb, &valid_foc);
366	} else if (foc->len > 0) {
367		ret = tcp_fastopen_cookie_gen_check(sk, req, skb, foc,
368						    &valid_foc);
369		if (!ret) {
370			NET_INC_STATS(sock_net(sk),
371				      LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
372		} else {
373			/* Cookie is valid. Create a (full) child socket to
374			 * accept the data in SYN before returning a SYN-ACK to
375			 * ack the data. If we fail to create the socket, fall
376			 * back and ack the ISN only but includes the same
377			 * cookie.
378			 *
379			 * Note: Data-less SYN with valid cookie is allowed to
380			 * send data in SYN_RECV state.
381			 */
382fastopen:
383			child = tcp_fastopen_create_child(sk, skb, req);
384			if (child) {
385				if (ret == 2) {
386					valid_foc.exp = foc->exp;
387					*foc = valid_foc;
388					NET_INC_STATS(sock_net(sk),
389						      LINUX_MIB_TCPFASTOPENPASSIVEALTKEY);
390				} else {
391					foc->len = -1;
392				}
393				NET_INC_STATS(sock_net(sk),
394					      LINUX_MIB_TCPFASTOPENPASSIVE);
395				return child;
396			}
397			NET_INC_STATS(sock_net(sk),
398				      LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
399		}
400	}
401	valid_foc.exp = foc->exp;
402	*foc = valid_foc;
403	return NULL;
404}
405
406bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
407			       struct tcp_fastopen_cookie *cookie)
408{
409	const struct dst_entry *dst;
410
411	tcp_fastopen_cache_get(sk, mss, cookie);
412
413	/* Firewall blackhole issue check */
414	if (tcp_fastopen_active_should_disable(sk)) {
415		cookie->len = -1;
416		return false;
417	}
418
419	dst = __sk_dst_get(sk);
420
421	if (tcp_fastopen_no_cookie(sk, dst, TFO_CLIENT_NO_COOKIE)) {
422		cookie->len = -1;
423		return true;
424	}
425	return cookie->len > 0;
426}
427
428/* This function checks if we want to defer sending SYN until the first
429 * write().  We defer under the following conditions:
430 * 1. fastopen_connect sockopt is set
431 * 2. we have a valid cookie
432 * Return value: return true if we want to defer until application writes data
433 *               return false if we want to send out SYN immediately
434 */
435bool tcp_fastopen_defer_connect(struct sock *sk, int *err)
436{
437	struct tcp_fastopen_cookie cookie = { .len = 0 };
438	struct tcp_sock *tp = tcp_sk(sk);
439	u16 mss;
440
441	if (tp->fastopen_connect && !tp->fastopen_req) {
442		if (tcp_fastopen_cookie_check(sk, &mss, &cookie)) {
443			inet_sk(sk)->defer_connect = 1;
444			return true;
445		}
446
447		/* Alloc fastopen_req in order for FO option to be included
448		 * in SYN
449		 */
450		tp->fastopen_req = kzalloc(sizeof(*tp->fastopen_req),
451					   sk->sk_allocation);
452		if (tp->fastopen_req)
453			tp->fastopen_req->cookie = cookie;
454		else
455			*err = -ENOBUFS;
456	}
457	return false;
458}
459EXPORT_SYMBOL(tcp_fastopen_defer_connect);
460
461/*
462 * The following code block is to deal with middle box issues with TFO:
463 * Middlebox firewall issues can potentially cause server's data being
464 * blackholed after a successful 3WHS using TFO.
465 * The proposed solution is to disable active TFO globally under the
466 * following circumstances:
467 *   1. client side TFO socket receives out of order FIN
468 *   2. client side TFO socket receives out of order RST
469 *   3. client side TFO socket has timed out three times consecutively during
470 *      or after handshake
471 * We disable active side TFO globally for 1hr at first. Then if it
472 * happens again, we disable it for 2h, then 4h, 8h, ...
473 * And we reset the timeout back to 1hr when we see a successful active
474 * TFO connection with data exchanges.
475 */
476
477/* Disable active TFO and record current jiffies and
478 * tfo_active_disable_times
479 */
480void tcp_fastopen_active_disable(struct sock *sk)
481{
482	struct net *net = sock_net(sk);
483
484	atomic_inc(&net->ipv4.tfo_active_disable_times);
485	net->ipv4.tfo_active_disable_stamp = jiffies;
486	NET_INC_STATS(net, LINUX_MIB_TCPFASTOPENBLACKHOLE);
487}
488
489/* Calculate timeout for tfo active disable
490 * Return true if we are still in the active TFO disable period
491 * Return false if timeout already expired and we should use active TFO
492 */
493bool tcp_fastopen_active_should_disable(struct sock *sk)
494{
495	unsigned int tfo_bh_timeout = sock_net(sk)->ipv4.sysctl_tcp_fastopen_blackhole_timeout;
496	int tfo_da_times = atomic_read(&sock_net(sk)->ipv4.tfo_active_disable_times);
497	unsigned long timeout;
498	int multiplier;
499
500	if (!tfo_da_times)
501		return false;
502
503	/* Limit timout to max: 2^6 * initial timeout */
504	multiplier = 1 << min(tfo_da_times - 1, 6);
505	timeout = multiplier * tfo_bh_timeout * HZ;
506	if (time_before(jiffies, sock_net(sk)->ipv4.tfo_active_disable_stamp + timeout))
507		return true;
508
509	/* Mark check bit so we can check for successful active TFO
510	 * condition and reset tfo_active_disable_times
511	 */
512	tcp_sk(sk)->syn_fastopen_ch = 1;
513	return false;
514}
515
516/* Disable active TFO if FIN is the only packet in the ofo queue
517 * and no data is received.
518 * Also check if we can reset tfo_active_disable_times if data is
519 * received successfully on a marked active TFO sockets opened on
520 * a non-loopback interface
521 */
522void tcp_fastopen_active_disable_ofo_check(struct sock *sk)
523{
524	struct tcp_sock *tp = tcp_sk(sk);
525	struct dst_entry *dst;
526	struct sk_buff *skb;
527
528	if (!tp->syn_fastopen)
529		return;
530
531	if (!tp->data_segs_in) {
532		skb = skb_rb_first(&tp->out_of_order_queue);
533		if (skb && !skb_rb_next(skb)) {
534			if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
535				tcp_fastopen_active_disable(sk);
536				return;
537			}
538		}
539	} else if (tp->syn_fastopen_ch &&
540		   atomic_read(&sock_net(sk)->ipv4.tfo_active_disable_times)) {
541		dst = sk_dst_get(sk);
542		if (!(dst && dst->dev && (dst->dev->flags & IFF_LOOPBACK)))
543			atomic_set(&sock_net(sk)->ipv4.tfo_active_disable_times, 0);
544		dst_release(dst);
545	}
546}
547
548void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired)
549{
550	u32 timeouts = inet_csk(sk)->icsk_retransmits;
551	struct tcp_sock *tp = tcp_sk(sk);
552
553	/* Broken middle-boxes may black-hole Fast Open connection during or
554	 * even after the handshake. Be extremely conservative and pause
555	 * Fast Open globally after hitting the third consecutive timeout or
556	 * exceeding the configured timeout limit.
557	 */
558	if ((tp->syn_fastopen || tp->syn_data || tp->syn_data_acked) &&
559	    (timeouts == 2 || (timeouts < 2 && expired))) {
560		tcp_fastopen_active_disable(sk);
561		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENACTIVEFAIL);
562	}
563}