1 /* 2 * Copyright (c) 2006 Oracle. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 * 32 */ 33 #include <linux/kernel.h> 34 #include <linux/list.h> 35 #include <linux/slab.h> 36 #include <net/inet_hashtables.h> 37 38 #include "rds.h" 39 #include "loop.h" 40 #include "rdma.h" 41 42 #define RDS_CONNECTION_HASH_BITS 12 43 #define RDS_CONNECTION_HASH_ENTRIES (1 << RDS_CONNECTION_HASH_BITS) 44 #define RDS_CONNECTION_HASH_MASK (RDS_CONNECTION_HASH_ENTRIES - 1) 45 46 /* converting this to RCU is a chore for another day.. */ 47 static DEFINE_SPINLOCK(rds_conn_lock); 48 static unsigned long rds_conn_count; 49 static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES]; 50 static struct kmem_cache *rds_conn_slab; 51 52 static struct hlist_head *rds_conn_bucket(__be32 laddr, __be32 faddr) 53 { 54 /* Pass NULL, don't need struct net for hash */ 55 unsigned long hash = inet_ehashfn(NULL, 56 be32_to_cpu(laddr), 0, 57 be32_to_cpu(faddr), 0); 58 return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK]; 59 } 60 61 #define rds_conn_info_set(var, test, suffix) do { \ 62 if (test) \ 63 var |= RDS_INFO_CONNECTION_FLAG_##suffix; \ 64 } while (0) 65 66 static inline int rds_conn_is_sending(struct rds_connection *conn) 67 { 68 int ret = 0; 69 70 if (!mutex_trylock(&conn->c_send_lock)) 71 ret = 1; 72 else 73 mutex_unlock(&conn->c_send_lock); 74 75 return ret; 76 } 77 78 static struct rds_connection *rds_conn_lookup(struct hlist_head *head, 79 __be32 laddr, __be32 faddr, 80 struct rds_transport *trans) 81 { 82 struct rds_connection *conn, *ret = NULL; 83 struct hlist_node *pos; 84 85 hlist_for_each_entry(conn, pos, head, c_hash_node) { 86 if (conn->c_faddr == faddr && conn->c_laddr == laddr && 87 conn->c_trans == trans) { 88 ret = conn; 89 break; 90 } 91 } 92 rdsdebug("returning conn %p for %pI4 -> %pI4\n", ret, 93 &laddr, &faddr); 94 return ret; 95 } 96 97 /* 98 * This is called by transports as they're bringing down a connection. 99 * It clears partial message state so that the transport can start sending 100 * and receiving over this connection again in the future. It is up to 101 * the transport to have serialized this call with its send and recv. 102 */ 103 void rds_conn_reset(struct rds_connection *conn) 104 { 105 rdsdebug("connection %pI4 to %pI4 reset\n", 106 &conn->c_laddr, &conn->c_faddr); 107 108 rds_stats_inc(s_conn_reset); 109 rds_send_reset(conn); 110 conn->c_flags = 0; 111 112 /* Do not clear next_rx_seq here, else we cannot distinguish 113 * retransmitted packets from new packets, and will hand all 114 * of them to the application. That is not consistent with the 115 * reliability guarantees of RDS. */ 116 } 117 118 /* 119 * There is only every one 'conn' for a given pair of addresses in the 120 * system at a time. They contain messages to be retransmitted and so 121 * span the lifetime of the actual underlying transport connections. 122 * 123 * For now they are not garbage collected once they're created. They 124 * are torn down as the module is removed, if ever. 125 */ 126 static struct rds_connection *__rds_conn_create(__be32 laddr, __be32 faddr, 127 struct rds_transport *trans, gfp_t gfp, 128 int is_outgoing) 129 { 130 struct rds_connection *conn, *parent = NULL; 131 struct hlist_head *head = rds_conn_bucket(laddr, faddr); 132 unsigned long flags; 133 int ret; 134 135 spin_lock_irqsave(&rds_conn_lock, flags); 136 conn = rds_conn_lookup(head, laddr, faddr, trans); 137 if (conn && conn->c_loopback && conn->c_trans != &rds_loop_transport && 138 !is_outgoing) { 139 /* This is a looped back IB connection, and we're 140 * called by the code handling the incoming connect. 141 * We need a second connection object into which we 142 * can stick the other QP. */ 143 parent = conn; 144 conn = parent->c_passive; 145 } 146 spin_unlock_irqrestore(&rds_conn_lock, flags); 147 if (conn) 148 goto out; 149 150 conn = kmem_cache_zalloc(rds_conn_slab, gfp); 151 if (conn == NULL) { 152 conn = ERR_PTR(-ENOMEM); 153 goto out; 154 } 155 156 INIT_HLIST_NODE(&conn->c_hash_node); 157 conn->c_laddr = laddr; 158 conn->c_faddr = faddr; 159 spin_lock_init(&conn->c_lock); 160 conn->c_next_tx_seq = 1; 161 162 mutex_init(&conn->c_send_lock); 163 INIT_LIST_HEAD(&conn->c_send_queue); 164 INIT_LIST_HEAD(&conn->c_retrans); 165 166 ret = rds_cong_get_maps(conn); 167 if (ret) { 168 kmem_cache_free(rds_conn_slab, conn); 169 conn = ERR_PTR(ret); 170 goto out; 171 } 172 173 /* 174 * This is where a connection becomes loopback. If *any* RDS sockets 175 * can bind to the destination address then we'd rather the messages 176 * flow through loopback rather than either transport. 177 */ 178 if (rds_trans_get_preferred(faddr)) { 179 conn->c_loopback = 1; 180 if (is_outgoing && trans->t_prefer_loopback) { 181 /* "outgoing" connection - and the transport 182 * says it wants the connection handled by the 183 * loopback transport. This is what TCP does. 184 */ 185 trans = &rds_loop_transport; 186 } 187 } 188 189 conn->c_trans = trans; 190 191 ret = trans->conn_alloc(conn, gfp); 192 if (ret) { 193 kmem_cache_free(rds_conn_slab, conn); 194 conn = ERR_PTR(ret); 195 goto out; 196 } 197 198 atomic_set(&conn->c_state, RDS_CONN_DOWN); 199 conn->c_reconnect_jiffies = 0; 200 INIT_DELAYED_WORK(&conn->c_send_w, rds_send_worker); 201 INIT_DELAYED_WORK(&conn->c_recv_w, rds_recv_worker); 202 INIT_DELAYED_WORK(&conn->c_conn_w, rds_connect_worker); 203 INIT_WORK(&conn->c_down_w, rds_shutdown_worker); 204 mutex_init(&conn->c_cm_lock); 205 conn->c_flags = 0; 206 207 rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n", 208 conn, &laddr, &faddr, 209 trans->t_name ? trans->t_name : "[unknown]", 210 is_outgoing ? "(outgoing)" : ""); 211 212 /* 213 * Since we ran without holding the conn lock, someone could 214 * have created the same conn (either normal or passive) in the 215 * interim. We check while holding the lock. If we won, we complete 216 * init and return our conn. If we lost, we rollback and return the 217 * other one. 218 */ 219 spin_lock_irqsave(&rds_conn_lock, flags); 220 if (parent) { 221 /* Creating passive conn */ 222 if (parent->c_passive) { 223 trans->conn_free(conn->c_transport_data); 224 kmem_cache_free(rds_conn_slab, conn); 225 conn = parent->c_passive; 226 } else { 227 parent->c_passive = conn; 228 rds_cong_add_conn(conn); 229 rds_conn_count++; 230 } 231 } else { 232 /* Creating normal conn */ 233 struct rds_connection *found; 234 235 found = rds_conn_lookup(head, laddr, faddr, trans); 236 if (found) { 237 trans->conn_free(conn->c_transport_data); 238 kmem_cache_free(rds_conn_slab, conn); 239 conn = found; 240 } else { 241 hlist_add_head(&conn->c_hash_node, head); 242 rds_cong_add_conn(conn); 243 rds_conn_count++; 244 } 245 } 246 spin_unlock_irqrestore(&rds_conn_lock, flags); 247 248 out: 249 return conn; 250 } 251 252 struct rds_connection *rds_conn_create(__be32 laddr, __be32 faddr, 253 struct rds_transport *trans, gfp_t gfp) 254 { 255 return __rds_conn_create(laddr, faddr, trans, gfp, 0); 256 } 257 EXPORT_SYMBOL_GPL(rds_conn_create); 258 259 struct rds_connection *rds_conn_create_outgoing(__be32 laddr, __be32 faddr, 260 struct rds_transport *trans, gfp_t gfp) 261 { 262 return __rds_conn_create(laddr, faddr, trans, gfp, 1); 263 } 264 EXPORT_SYMBOL_GPL(rds_conn_create_outgoing); 265 266 void rds_conn_destroy(struct rds_connection *conn) 267 { 268 struct rds_message *rm, *rtmp; 269 270 rdsdebug("freeing conn %p for %pI4 -> " 271 "%pI4\n", conn, &conn->c_laddr, 272 &conn->c_faddr); 273 274 hlist_del_init(&conn->c_hash_node); 275 276 /* wait for the rds thread to shut it down */ 277 atomic_set(&conn->c_state, RDS_CONN_ERROR); 278 cancel_delayed_work(&conn->c_conn_w); 279 queue_work(rds_wq, &conn->c_down_w); 280 flush_workqueue(rds_wq); 281 282 /* tear down queued messages */ 283 list_for_each_entry_safe(rm, rtmp, 284 &conn->c_send_queue, 285 m_conn_item) { 286 list_del_init(&rm->m_conn_item); 287 BUG_ON(!list_empty(&rm->m_sock_item)); 288 rds_message_put(rm); 289 } 290 if (conn->c_xmit_rm) 291 rds_message_put(conn->c_xmit_rm); 292 293 conn->c_trans->conn_free(conn->c_transport_data); 294 295 /* 296 * The congestion maps aren't freed up here. They're 297 * freed by rds_cong_exit() after all the connections 298 * have been freed. 299 */ 300 rds_cong_remove_conn(conn); 301 302 BUG_ON(!list_empty(&conn->c_retrans)); 303 kmem_cache_free(rds_conn_slab, conn); 304 305 rds_conn_count--; 306 } 307 EXPORT_SYMBOL_GPL(rds_conn_destroy); 308 309 static void rds_conn_message_info(struct socket *sock, unsigned int len, 310 struct rds_info_iterator *iter, 311 struct rds_info_lengths *lens, 312 int want_send) 313 { 314 struct hlist_head *head; 315 struct hlist_node *pos; 316 struct list_head *list; 317 struct rds_connection *conn; 318 struct rds_message *rm; 319 unsigned long flags; 320 unsigned int total = 0; 321 size_t i; 322 323 len /= sizeof(struct rds_info_message); 324 325 spin_lock_irqsave(&rds_conn_lock, flags); 326 327 for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); 328 i++, head++) { 329 hlist_for_each_entry(conn, pos, head, c_hash_node) { 330 if (want_send) 331 list = &conn->c_send_queue; 332 else 333 list = &conn->c_retrans; 334 335 spin_lock(&conn->c_lock); 336 337 /* XXX too lazy to maintain counts.. */ 338 list_for_each_entry(rm, list, m_conn_item) { 339 total++; 340 if (total <= len) 341 rds_inc_info_copy(&rm->m_inc, iter, 342 conn->c_laddr, 343 conn->c_faddr, 0); 344 } 345 346 spin_unlock(&conn->c_lock); 347 } 348 } 349 350 spin_unlock_irqrestore(&rds_conn_lock, flags); 351 352 lens->nr = total; 353 lens->each = sizeof(struct rds_info_message); 354 } 355 356 static void rds_conn_message_info_send(struct socket *sock, unsigned int len, 357 struct rds_info_iterator *iter, 358 struct rds_info_lengths *lens) 359 { 360 rds_conn_message_info(sock, len, iter, lens, 1); 361 } 362 363 static void rds_conn_message_info_retrans(struct socket *sock, 364 unsigned int len, 365 struct rds_info_iterator *iter, 366 struct rds_info_lengths *lens) 367 { 368 rds_conn_message_info(sock, len, iter, lens, 0); 369 } 370 371 void rds_for_each_conn_info(struct socket *sock, unsigned int len, 372 struct rds_info_iterator *iter, 373 struct rds_info_lengths *lens, 374 int (*visitor)(struct rds_connection *, void *), 375 size_t item_len) 376 { 377 uint64_t buffer[(item_len + 7) / 8]; 378 struct hlist_head *head; 379 struct hlist_node *pos; 380 struct hlist_node *tmp; 381 struct rds_connection *conn; 382 unsigned long flags; 383 size_t i; 384 385 spin_lock_irqsave(&rds_conn_lock, flags); 386 387 lens->nr = 0; 388 lens->each = item_len; 389 390 for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); 391 i++, head++) { 392 hlist_for_each_entry_safe(conn, pos, tmp, head, c_hash_node) { 393 394 /* XXX no c_lock usage.. */ 395 if (!visitor(conn, buffer)) 396 continue; 397 398 /* We copy as much as we can fit in the buffer, 399 * but we count all items so that the caller 400 * can resize the buffer. */ 401 if (len >= item_len) { 402 rds_info_copy(iter, buffer, item_len); 403 len -= item_len; 404 } 405 lens->nr++; 406 } 407 } 408 409 spin_unlock_irqrestore(&rds_conn_lock, flags); 410 } 411 EXPORT_SYMBOL_GPL(rds_for_each_conn_info); 412 413 static int rds_conn_info_visitor(struct rds_connection *conn, 414 void *buffer) 415 { 416 struct rds_info_connection *cinfo = buffer; 417 418 cinfo->next_tx_seq = conn->c_next_tx_seq; 419 cinfo->next_rx_seq = conn->c_next_rx_seq; 420 cinfo->laddr = conn->c_laddr; 421 cinfo->faddr = conn->c_faddr; 422 strncpy(cinfo->transport, conn->c_trans->t_name, 423 sizeof(cinfo->transport)); 424 cinfo->flags = 0; 425 426 rds_conn_info_set(cinfo->flags, 427 rds_conn_is_sending(conn), SENDING); 428 /* XXX Future: return the state rather than these funky bits */ 429 rds_conn_info_set(cinfo->flags, 430 atomic_read(&conn->c_state) == RDS_CONN_CONNECTING, 431 CONNECTING); 432 rds_conn_info_set(cinfo->flags, 433 atomic_read(&conn->c_state) == RDS_CONN_UP, 434 CONNECTED); 435 return 1; 436 } 437 438 static void rds_conn_info(struct socket *sock, unsigned int len, 439 struct rds_info_iterator *iter, 440 struct rds_info_lengths *lens) 441 { 442 rds_for_each_conn_info(sock, len, iter, lens, 443 rds_conn_info_visitor, 444 sizeof(struct rds_info_connection)); 445 } 446 447 int __init rds_conn_init(void) 448 { 449 rds_conn_slab = kmem_cache_create("rds_connection", 450 sizeof(struct rds_connection), 451 0, 0, NULL); 452 if (rds_conn_slab == NULL) 453 return -ENOMEM; 454 455 rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info); 456 rds_info_register_func(RDS_INFO_SEND_MESSAGES, 457 rds_conn_message_info_send); 458 rds_info_register_func(RDS_INFO_RETRANS_MESSAGES, 459 rds_conn_message_info_retrans); 460 461 return 0; 462 } 463 464 void rds_conn_exit(void) 465 { 466 rds_loop_exit(); 467 468 WARN_ON(!hlist_empty(rds_conn_hash)); 469 470 kmem_cache_destroy(rds_conn_slab); 471 472 rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info); 473 rds_info_deregister_func(RDS_INFO_SEND_MESSAGES, 474 rds_conn_message_info_send); 475 rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES, 476 rds_conn_message_info_retrans); 477 } 478 479 /* 480 * Force a disconnect 481 */ 482 void rds_conn_drop(struct rds_connection *conn) 483 { 484 atomic_set(&conn->c_state, RDS_CONN_ERROR); 485 queue_work(rds_wq, &conn->c_down_w); 486 } 487 EXPORT_SYMBOL_GPL(rds_conn_drop); 488 489 /* 490 * An error occurred on the connection 491 */ 492 void 493 __rds_conn_error(struct rds_connection *conn, const char *fmt, ...) 494 { 495 va_list ap; 496 497 va_start(ap, fmt); 498 vprintk(fmt, ap); 499 va_end(ap); 500 501 rds_conn_drop(conn); 502 } 503