xref: /linux/net/ipv4/tcp_cong.c (revision 93df8a1ed6231727c5db94a80b1a6bd5ee67cec3)
1 /*
2  * Pluggable TCP congestion control support and newReno
3  * congestion control.
4  * Based on ideas from I/O scheduler support and Web100.
5  *
6  * Copyright (C) 2005 Stephen Hemminger <shemminger@osdl.org>
7  */
8 
9 #define pr_fmt(fmt) "TCP: " fmt
10 
11 #include <linux/module.h>
12 #include <linux/mm.h>
13 #include <linux/types.h>
14 #include <linux/list.h>
15 #include <linux/gfp.h>
16 #include <linux/jhash.h>
17 #include <net/tcp.h>
18 
19 static DEFINE_SPINLOCK(tcp_cong_list_lock);
20 static LIST_HEAD(tcp_cong_list);
21 
22 /* Simple linear search, don't expect many entries! */
23 static struct tcp_congestion_ops *tcp_ca_find(const char *name)
24 {
25 	struct tcp_congestion_ops *e;
26 
27 	list_for_each_entry_rcu(e, &tcp_cong_list, list) {
28 		if (strcmp(e->name, name) == 0)
29 			return e;
30 	}
31 
32 	return NULL;
33 }
34 
35 /* Must be called with rcu lock held */
36 static const struct tcp_congestion_ops *__tcp_ca_find_autoload(const char *name)
37 {
38 	const struct tcp_congestion_ops *ca = tcp_ca_find(name);
39 #ifdef CONFIG_MODULES
40 	if (!ca && capable(CAP_NET_ADMIN)) {
41 		rcu_read_unlock();
42 		request_module("tcp_%s", name);
43 		rcu_read_lock();
44 		ca = tcp_ca_find(name);
45 	}
46 #endif
47 	return ca;
48 }
49 
50 /* Simple linear search, not much in here. */
51 struct tcp_congestion_ops *tcp_ca_find_key(u32 key)
52 {
53 	struct tcp_congestion_ops *e;
54 
55 	list_for_each_entry_rcu(e, &tcp_cong_list, list) {
56 		if (e->key == key)
57 			return e;
58 	}
59 
60 	return NULL;
61 }
62 
63 /*
64  * Attach new congestion control algorithm to the list
65  * of available options.
66  */
67 int tcp_register_congestion_control(struct tcp_congestion_ops *ca)
68 {
69 	int ret = 0;
70 
71 	/* all algorithms must implement ssthresh and cong_avoid ops */
72 	if (!ca->ssthresh || !ca->cong_avoid) {
73 		pr_err("%s does not implement required ops\n", ca->name);
74 		return -EINVAL;
75 	}
76 
77 	ca->key = jhash(ca->name, sizeof(ca->name), strlen(ca->name));
78 
79 	spin_lock(&tcp_cong_list_lock);
80 	if (ca->key == TCP_CA_UNSPEC || tcp_ca_find_key(ca->key)) {
81 		pr_notice("%s already registered or non-unique key\n",
82 			  ca->name);
83 		ret = -EEXIST;
84 	} else {
85 		list_add_tail_rcu(&ca->list, &tcp_cong_list);
86 		pr_debug("%s registered\n", ca->name);
87 	}
88 	spin_unlock(&tcp_cong_list_lock);
89 
90 	return ret;
91 }
92 EXPORT_SYMBOL_GPL(tcp_register_congestion_control);
93 
94 /*
95  * Remove congestion control algorithm, called from
96  * the module's remove function.  Module ref counts are used
97  * to ensure that this can't be done till all sockets using
98  * that method are closed.
99  */
100 void tcp_unregister_congestion_control(struct tcp_congestion_ops *ca)
101 {
102 	spin_lock(&tcp_cong_list_lock);
103 	list_del_rcu(&ca->list);
104 	spin_unlock(&tcp_cong_list_lock);
105 
106 	/* Wait for outstanding readers to complete before the
107 	 * module gets removed entirely.
108 	 *
109 	 * A try_module_get() should fail by now as our module is
110 	 * in "going" state since no refs are held anymore and
111 	 * module_exit() handler being called.
112 	 */
113 	synchronize_rcu();
114 }
115 EXPORT_SYMBOL_GPL(tcp_unregister_congestion_control);
116 
117 u32 tcp_ca_get_key_by_name(const char *name)
118 {
119 	const struct tcp_congestion_ops *ca;
120 	u32 key;
121 
122 	might_sleep();
123 
124 	rcu_read_lock();
125 	ca = __tcp_ca_find_autoload(name);
126 	key = ca ? ca->key : TCP_CA_UNSPEC;
127 	rcu_read_unlock();
128 
129 	return key;
130 }
131 EXPORT_SYMBOL_GPL(tcp_ca_get_key_by_name);
132 
133 char *tcp_ca_get_name_by_key(u32 key, char *buffer)
134 {
135 	const struct tcp_congestion_ops *ca;
136 	char *ret = NULL;
137 
138 	rcu_read_lock();
139 	ca = tcp_ca_find_key(key);
140 	if (ca)
141 		ret = strncpy(buffer, ca->name,
142 			      TCP_CA_NAME_MAX);
143 	rcu_read_unlock();
144 
145 	return ret;
146 }
147 EXPORT_SYMBOL_GPL(tcp_ca_get_name_by_key);
148 
149 /* Assign choice of congestion control. */
150 void tcp_assign_congestion_control(struct sock *sk)
151 {
152 	struct inet_connection_sock *icsk = inet_csk(sk);
153 	struct tcp_congestion_ops *ca;
154 
155 	rcu_read_lock();
156 	list_for_each_entry_rcu(ca, &tcp_cong_list, list) {
157 		if (likely(try_module_get(ca->owner))) {
158 			icsk->icsk_ca_ops = ca;
159 			goto out;
160 		}
161 		/* Fallback to next available. The last really
162 		 * guaranteed fallback is Reno from this list.
163 		 */
164 	}
165 out:
166 	rcu_read_unlock();
167 
168 	/* Clear out private data before diag gets it and
169 	 * the ca has not been initialized.
170 	 */
171 	if (ca->get_info)
172 		memset(icsk->icsk_ca_priv, 0, sizeof(icsk->icsk_ca_priv));
173 }
174 
175 void tcp_init_congestion_control(struct sock *sk)
176 {
177 	const struct inet_connection_sock *icsk = inet_csk(sk);
178 
179 	if (icsk->icsk_ca_ops->init)
180 		icsk->icsk_ca_ops->init(sk);
181 }
182 
183 static void tcp_reinit_congestion_control(struct sock *sk,
184 					  const struct tcp_congestion_ops *ca)
185 {
186 	struct inet_connection_sock *icsk = inet_csk(sk);
187 
188 	tcp_cleanup_congestion_control(sk);
189 	icsk->icsk_ca_ops = ca;
190 	icsk->icsk_ca_setsockopt = 1;
191 
192 	if (sk->sk_state != TCP_CLOSE && icsk->icsk_ca_ops->init)
193 		icsk->icsk_ca_ops->init(sk);
194 }
195 
196 /* Manage refcounts on socket close. */
197 void tcp_cleanup_congestion_control(struct sock *sk)
198 {
199 	struct inet_connection_sock *icsk = inet_csk(sk);
200 
201 	if (icsk->icsk_ca_ops->release)
202 		icsk->icsk_ca_ops->release(sk);
203 	module_put(icsk->icsk_ca_ops->owner);
204 }
205 
206 /* Used by sysctl to change default congestion control */
207 int tcp_set_default_congestion_control(const char *name)
208 {
209 	struct tcp_congestion_ops *ca;
210 	int ret = -ENOENT;
211 
212 	spin_lock(&tcp_cong_list_lock);
213 	ca = tcp_ca_find(name);
214 #ifdef CONFIG_MODULES
215 	if (!ca && capable(CAP_NET_ADMIN)) {
216 		spin_unlock(&tcp_cong_list_lock);
217 
218 		request_module("tcp_%s", name);
219 		spin_lock(&tcp_cong_list_lock);
220 		ca = tcp_ca_find(name);
221 	}
222 #endif
223 
224 	if (ca) {
225 		ca->flags |= TCP_CONG_NON_RESTRICTED;	/* default is always allowed */
226 		list_move(&ca->list, &tcp_cong_list);
227 		ret = 0;
228 	}
229 	spin_unlock(&tcp_cong_list_lock);
230 
231 	return ret;
232 }
233 
234 /* Set default value from kernel configuration at bootup */
235 static int __init tcp_congestion_default(void)
236 {
237 	return tcp_set_default_congestion_control(CONFIG_DEFAULT_TCP_CONG);
238 }
239 late_initcall(tcp_congestion_default);
240 
241 /* Build string with list of available congestion control values */
242 void tcp_get_available_congestion_control(char *buf, size_t maxlen)
243 {
244 	struct tcp_congestion_ops *ca;
245 	size_t offs = 0;
246 
247 	rcu_read_lock();
248 	list_for_each_entry_rcu(ca, &tcp_cong_list, list) {
249 		offs += snprintf(buf + offs, maxlen - offs,
250 				 "%s%s",
251 				 offs == 0 ? "" : " ", ca->name);
252 	}
253 	rcu_read_unlock();
254 }
255 
256 /* Get current default congestion control */
257 void tcp_get_default_congestion_control(char *name)
258 {
259 	struct tcp_congestion_ops *ca;
260 	/* We will always have reno... */
261 	BUG_ON(list_empty(&tcp_cong_list));
262 
263 	rcu_read_lock();
264 	ca = list_entry(tcp_cong_list.next, struct tcp_congestion_ops, list);
265 	strncpy(name, ca->name, TCP_CA_NAME_MAX);
266 	rcu_read_unlock();
267 }
268 
269 /* Built list of non-restricted congestion control values */
270 void tcp_get_allowed_congestion_control(char *buf, size_t maxlen)
271 {
272 	struct tcp_congestion_ops *ca;
273 	size_t offs = 0;
274 
275 	*buf = '\0';
276 	rcu_read_lock();
277 	list_for_each_entry_rcu(ca, &tcp_cong_list, list) {
278 		if (!(ca->flags & TCP_CONG_NON_RESTRICTED))
279 			continue;
280 		offs += snprintf(buf + offs, maxlen - offs,
281 				 "%s%s",
282 				 offs == 0 ? "" : " ", ca->name);
283 	}
284 	rcu_read_unlock();
285 }
286 
287 /* Change list of non-restricted congestion control */
288 int tcp_set_allowed_congestion_control(char *val)
289 {
290 	struct tcp_congestion_ops *ca;
291 	char *saved_clone, *clone, *name;
292 	int ret = 0;
293 
294 	saved_clone = clone = kstrdup(val, GFP_USER);
295 	if (!clone)
296 		return -ENOMEM;
297 
298 	spin_lock(&tcp_cong_list_lock);
299 	/* pass 1 check for bad entries */
300 	while ((name = strsep(&clone, " ")) && *name) {
301 		ca = tcp_ca_find(name);
302 		if (!ca) {
303 			ret = -ENOENT;
304 			goto out;
305 		}
306 	}
307 
308 	/* pass 2 clear old values */
309 	list_for_each_entry_rcu(ca, &tcp_cong_list, list)
310 		ca->flags &= ~TCP_CONG_NON_RESTRICTED;
311 
312 	/* pass 3 mark as allowed */
313 	while ((name = strsep(&val, " ")) && *name) {
314 		ca = tcp_ca_find(name);
315 		WARN_ON(!ca);
316 		if (ca)
317 			ca->flags |= TCP_CONG_NON_RESTRICTED;
318 	}
319 out:
320 	spin_unlock(&tcp_cong_list_lock);
321 	kfree(saved_clone);
322 
323 	return ret;
324 }
325 
326 /* Change congestion control for socket */
327 int tcp_set_congestion_control(struct sock *sk, const char *name)
328 {
329 	struct inet_connection_sock *icsk = inet_csk(sk);
330 	const struct tcp_congestion_ops *ca;
331 	int err = 0;
332 
333 	if (icsk->icsk_ca_dst_locked)
334 		return -EPERM;
335 
336 	rcu_read_lock();
337 	ca = __tcp_ca_find_autoload(name);
338 	/* No change asking for existing value */
339 	if (ca == icsk->icsk_ca_ops) {
340 		icsk->icsk_ca_setsockopt = 1;
341 		goto out;
342 	}
343 	if (!ca)
344 		err = -ENOENT;
345 	else if (!((ca->flags & TCP_CONG_NON_RESTRICTED) ||
346 		   ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)))
347 		err = -EPERM;
348 	else if (!try_module_get(ca->owner))
349 		err = -EBUSY;
350 	else
351 		tcp_reinit_congestion_control(sk, ca);
352  out:
353 	rcu_read_unlock();
354 	return err;
355 }
356 
357 /* Slow start is used when congestion window is no greater than the slow start
358  * threshold. We base on RFC2581 and also handle stretch ACKs properly.
359  * We do not implement RFC3465 Appropriate Byte Counting (ABC) per se but
360  * something better;) a packet is only considered (s)acked in its entirety to
361  * defend the ACK attacks described in the RFC. Slow start processes a stretch
362  * ACK of degree N as if N acks of degree 1 are received back to back except
363  * ABC caps N to 2. Slow start exits when cwnd grows over ssthresh and
364  * returns the leftover acks to adjust cwnd in congestion avoidance mode.
365  */
366 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked)
367 {
368 	u32 cwnd = tp->snd_cwnd + acked;
369 
370 	if (cwnd > tp->snd_ssthresh)
371 		cwnd = tp->snd_ssthresh + 1;
372 	acked -= cwnd - tp->snd_cwnd;
373 	tp->snd_cwnd = min(cwnd, tp->snd_cwnd_clamp);
374 
375 	return acked;
376 }
377 EXPORT_SYMBOL_GPL(tcp_slow_start);
378 
379 /* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd (or alternative w),
380  * for every packet that was ACKed.
381  */
382 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked)
383 {
384 	/* If credits accumulated at a higher w, apply them gently now. */
385 	if (tp->snd_cwnd_cnt >= w) {
386 		tp->snd_cwnd_cnt = 0;
387 		tp->snd_cwnd++;
388 	}
389 
390 	tp->snd_cwnd_cnt += acked;
391 	if (tp->snd_cwnd_cnt >= w) {
392 		u32 delta = tp->snd_cwnd_cnt / w;
393 
394 		tp->snd_cwnd_cnt -= delta * w;
395 		tp->snd_cwnd += delta;
396 	}
397 	tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_cwnd_clamp);
398 }
399 EXPORT_SYMBOL_GPL(tcp_cong_avoid_ai);
400 
401 /*
402  * TCP Reno congestion control
403  * This is special case used for fallback as well.
404  */
405 /* This is Jacobson's slow start and congestion avoidance.
406  * SIGCOMM '88, p. 328.
407  */
408 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked)
409 {
410 	struct tcp_sock *tp = tcp_sk(sk);
411 
412 	if (!tcp_is_cwnd_limited(sk))
413 		return;
414 
415 	/* In "safe" area, increase. */
416 	if (tp->snd_cwnd <= tp->snd_ssthresh) {
417 		acked = tcp_slow_start(tp, acked);
418 		if (!acked)
419 			return;
420 	}
421 	/* In dangerous area, increase slowly. */
422 	tcp_cong_avoid_ai(tp, tp->snd_cwnd, acked);
423 }
424 EXPORT_SYMBOL_GPL(tcp_reno_cong_avoid);
425 
426 /* Slow start threshold is half the congestion window (min 2) */
427 u32 tcp_reno_ssthresh(struct sock *sk)
428 {
429 	const struct tcp_sock *tp = tcp_sk(sk);
430 
431 	return max(tp->snd_cwnd >> 1U, 2U);
432 }
433 EXPORT_SYMBOL_GPL(tcp_reno_ssthresh);
434 
435 struct tcp_congestion_ops tcp_reno = {
436 	.flags		= TCP_CONG_NON_RESTRICTED,
437 	.name		= "reno",
438 	.owner		= THIS_MODULE,
439 	.ssthresh	= tcp_reno_ssthresh,
440 	.cong_avoid	= tcp_reno_cong_avoid,
441 };
442