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