1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0 3 * 4 * Copyright (c) 2004-2007 Intel Corporation. All rights reserved. 5 * Copyright (c) 2004 Topspin Corporation. All rights reserved. 6 * Copyright (c) 2004, 2005 Voltaire Corporation. All rights reserved. 7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 8 * 9 * This software is available to you under a choice of one of two 10 * licenses. You may choose to be licensed under the terms of the GNU 11 * General Public License (GPL) Version 2, available from the file 12 * COPYING in the main directory of this source tree, or the 13 * OpenIB.org BSD license below: 14 * 15 * Redistribution and use in source and binary forms, with or 16 * without modification, are permitted provided that the following 17 * conditions are met: 18 * 19 * - Redistributions of source code must retain the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer. 22 * 23 * - Redistributions in binary form must reproduce the above 24 * copyright notice, this list of conditions and the following 25 * disclaimer in the documentation and/or other materials 26 * provided with the distribution. 27 * 28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 35 * SOFTWARE. 36 */ 37 38 #include <sys/cdefs.h> 39 #include <linux/completion.h> 40 #include <linux/dma-mapping.h> 41 #include <linux/device.h> 42 #include <linux/module.h> 43 #include <linux/err.h> 44 #include <linux/idr.h> 45 #include <linux/interrupt.h> 46 #include <linux/random.h> 47 #include <linux/rbtree.h> 48 #include <linux/spinlock.h> 49 #include <linux/slab.h> 50 #include <linux/sysfs.h> 51 #include <linux/workqueue.h> 52 #include <linux/kdev_t.h> 53 #include <linux/etherdevice.h> 54 55 #include <asm/atomic-long.h> 56 57 #include <rdma/ib_addr.h> 58 #include <rdma/ib_cache.h> 59 #include <rdma/ib_cm.h> 60 #include "cm_msgs.h" 61 #include "core_priv.h" 62 63 MODULE_AUTHOR("Sean Hefty"); 64 MODULE_DESCRIPTION("InfiniBand CM"); 65 MODULE_LICENSE("Dual BSD/GPL"); 66 67 static const char * const ibcm_rej_reason_strs[] = { 68 [IB_CM_REJ_NO_QP] = "no QP", 69 [IB_CM_REJ_NO_EEC] = "no EEC", 70 [IB_CM_REJ_NO_RESOURCES] = "no resources", 71 [IB_CM_REJ_TIMEOUT] = "timeout", 72 [IB_CM_REJ_UNSUPPORTED] = "unsupported", 73 [IB_CM_REJ_INVALID_COMM_ID] = "invalid comm ID", 74 [IB_CM_REJ_INVALID_COMM_INSTANCE] = "invalid comm instance", 75 [IB_CM_REJ_INVALID_SERVICE_ID] = "invalid service ID", 76 [IB_CM_REJ_INVALID_TRANSPORT_TYPE] = "invalid transport type", 77 [IB_CM_REJ_STALE_CONN] = "stale conn", 78 [IB_CM_REJ_RDC_NOT_EXIST] = "RDC not exist", 79 [IB_CM_REJ_INVALID_GID] = "invalid GID", 80 [IB_CM_REJ_INVALID_LID] = "invalid LID", 81 [IB_CM_REJ_INVALID_SL] = "invalid SL", 82 [IB_CM_REJ_INVALID_TRAFFIC_CLASS] = "invalid traffic class", 83 [IB_CM_REJ_INVALID_HOP_LIMIT] = "invalid hop limit", 84 [IB_CM_REJ_INVALID_PACKET_RATE] = "invalid packet rate", 85 [IB_CM_REJ_INVALID_ALT_GID] = "invalid alt GID", 86 [IB_CM_REJ_INVALID_ALT_LID] = "invalid alt LID", 87 [IB_CM_REJ_INVALID_ALT_SL] = "invalid alt SL", 88 [IB_CM_REJ_INVALID_ALT_TRAFFIC_CLASS] = "invalid alt traffic class", 89 [IB_CM_REJ_INVALID_ALT_HOP_LIMIT] = "invalid alt hop limit", 90 [IB_CM_REJ_INVALID_ALT_PACKET_RATE] = "invalid alt packet rate", 91 [IB_CM_REJ_PORT_CM_REDIRECT] = "port CM redirect", 92 [IB_CM_REJ_PORT_REDIRECT] = "port redirect", 93 [IB_CM_REJ_INVALID_MTU] = "invalid MTU", 94 [IB_CM_REJ_INSUFFICIENT_RESP_RESOURCES] = "insufficient resp resources", 95 [IB_CM_REJ_CONSUMER_DEFINED] = "consumer defined", 96 [IB_CM_REJ_INVALID_RNR_RETRY] = "invalid RNR retry", 97 [IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID] = "duplicate local comm ID", 98 [IB_CM_REJ_INVALID_CLASS_VERSION] = "invalid class version", 99 [IB_CM_REJ_INVALID_FLOW_LABEL] = "invalid flow label", 100 [IB_CM_REJ_INVALID_ALT_FLOW_LABEL] = "invalid alt flow label", 101 }; 102 103 const char *__attribute_const__ ibcm_reject_msg(int reason) 104 { 105 size_t index = reason; 106 107 if (index < ARRAY_SIZE(ibcm_rej_reason_strs) && 108 ibcm_rej_reason_strs[index]) 109 return ibcm_rej_reason_strs[index]; 110 else 111 return "unrecognized reason"; 112 } 113 EXPORT_SYMBOL(ibcm_reject_msg); 114 115 static void cm_add_one(struct ib_device *device); 116 static void cm_remove_one(struct ib_device *device, void *client_data); 117 118 static struct ib_client cm_client = { 119 .name = "cm", 120 .add = cm_add_one, 121 .remove = cm_remove_one 122 }; 123 124 static struct ib_cm { 125 spinlock_t lock; 126 struct list_head device_list; 127 rwlock_t device_lock; 128 struct rb_root listen_service_table; 129 u64 listen_service_id; 130 /* struct rb_root peer_service_table; todo: fix peer to peer */ 131 struct rb_root remote_qp_table; 132 struct rb_root remote_id_table; 133 struct rb_root remote_sidr_table; 134 struct idr local_id_table; 135 __be32 random_id_operand; 136 struct list_head timewait_list; 137 struct workqueue_struct *wq; 138 /* Sync on cm change port state */ 139 spinlock_t state_lock; 140 } cm; 141 142 /* Counter indexes ordered by attribute ID */ 143 enum { 144 CM_REQ_COUNTER, 145 CM_MRA_COUNTER, 146 CM_REJ_COUNTER, 147 CM_REP_COUNTER, 148 CM_RTU_COUNTER, 149 CM_DREQ_COUNTER, 150 CM_DREP_COUNTER, 151 CM_SIDR_REQ_COUNTER, 152 CM_SIDR_REP_COUNTER, 153 CM_LAP_COUNTER, 154 CM_APR_COUNTER, 155 CM_ATTR_COUNT, 156 CM_ATTR_ID_OFFSET = 0x0010, 157 }; 158 159 enum { 160 CM_XMIT, 161 CM_XMIT_RETRIES, 162 CM_RECV, 163 CM_RECV_DUPLICATES, 164 CM_COUNTER_GROUPS 165 }; 166 167 static char const counter_group_names[CM_COUNTER_GROUPS] 168 [sizeof("cm_rx_duplicates")] = { 169 "cm_tx_msgs", "cm_tx_retries", 170 "cm_rx_msgs", "cm_rx_duplicates" 171 }; 172 173 struct cm_counter_group { 174 struct kobject obj; 175 atomic_long_t counter[CM_ATTR_COUNT]; 176 }; 177 178 struct cm_counter_attribute { 179 struct attribute attr; 180 int index; 181 }; 182 183 #define CM_COUNTER_ATTR(_name, _index) \ 184 struct cm_counter_attribute cm_##_name##_counter_attr = { \ 185 .attr = { .name = __stringify(_name), .mode = 0444 }, \ 186 .index = _index \ 187 } 188 189 static CM_COUNTER_ATTR(req, CM_REQ_COUNTER); 190 static CM_COUNTER_ATTR(mra, CM_MRA_COUNTER); 191 static CM_COUNTER_ATTR(rej, CM_REJ_COUNTER); 192 static CM_COUNTER_ATTR(rep, CM_REP_COUNTER); 193 static CM_COUNTER_ATTR(rtu, CM_RTU_COUNTER); 194 static CM_COUNTER_ATTR(dreq, CM_DREQ_COUNTER); 195 static CM_COUNTER_ATTR(drep, CM_DREP_COUNTER); 196 static CM_COUNTER_ATTR(sidr_req, CM_SIDR_REQ_COUNTER); 197 static CM_COUNTER_ATTR(sidr_rep, CM_SIDR_REP_COUNTER); 198 static CM_COUNTER_ATTR(lap, CM_LAP_COUNTER); 199 static CM_COUNTER_ATTR(apr, CM_APR_COUNTER); 200 201 static struct attribute *cm_counter_default_attrs[] = { 202 &cm_req_counter_attr.attr, 203 &cm_mra_counter_attr.attr, 204 &cm_rej_counter_attr.attr, 205 &cm_rep_counter_attr.attr, 206 &cm_rtu_counter_attr.attr, 207 &cm_dreq_counter_attr.attr, 208 &cm_drep_counter_attr.attr, 209 &cm_sidr_req_counter_attr.attr, 210 &cm_sidr_rep_counter_attr.attr, 211 &cm_lap_counter_attr.attr, 212 &cm_apr_counter_attr.attr, 213 NULL 214 }; 215 216 struct cm_port { 217 struct cm_device *cm_dev; 218 struct ib_mad_agent *mad_agent; 219 struct kobject port_obj; 220 u8 port_num; 221 struct list_head cm_priv_prim_list; 222 struct list_head cm_priv_altr_list; 223 struct cm_counter_group counter_group[CM_COUNTER_GROUPS]; 224 }; 225 226 struct cm_device { 227 struct list_head list; 228 struct ib_device *ib_device; 229 u8 ack_delay; 230 int going_down; 231 struct cm_port *port[0]; 232 }; 233 234 struct cm_av { 235 struct cm_port *port; 236 union ib_gid dgid; 237 struct ib_ah_attr ah_attr; 238 u16 pkey_index; 239 u8 timeout; 240 }; 241 242 struct cm_work { 243 struct delayed_work work; 244 struct list_head list; 245 struct cm_port *port; 246 struct ib_mad_recv_wc *mad_recv_wc; /* Received MADs */ 247 __be32 local_id; /* Established / timewait */ 248 __be32 remote_id; 249 struct ib_cm_event cm_event; 250 struct ib_sa_path_rec path[0]; 251 }; 252 253 struct cm_timewait_info { 254 struct cm_work work; /* Must be first. */ 255 struct list_head list; 256 struct rb_node remote_qp_node; 257 struct rb_node remote_id_node; 258 __be64 remote_ca_guid; 259 __be32 remote_qpn; 260 u8 inserted_remote_qp; 261 u8 inserted_remote_id; 262 }; 263 264 struct cm_id_private { 265 struct ib_cm_id id; 266 267 struct rb_node service_node; 268 struct rb_node sidr_id_node; 269 spinlock_t lock; /* Do not acquire inside cm.lock */ 270 struct completion comp; 271 atomic_t refcount; 272 /* Number of clients sharing this ib_cm_id. Only valid for listeners. 273 * Protected by the cm.lock spinlock. */ 274 int listen_sharecount; 275 276 struct ib_mad_send_buf *msg; 277 struct cm_timewait_info *timewait_info; 278 /* todo: use alternate port on send failure */ 279 struct cm_av av; 280 struct cm_av alt_av; 281 282 void *private_data; 283 __be64 tid; 284 __be32 local_qpn; 285 __be32 remote_qpn; 286 enum ib_qp_type qp_type; 287 __be32 sq_psn; 288 __be32 rq_psn; 289 int timeout_ms; 290 enum ib_mtu path_mtu; 291 __be16 pkey; 292 u8 private_data_len; 293 u8 max_cm_retries; 294 u8 peer_to_peer; 295 u8 responder_resources; 296 u8 initiator_depth; 297 u8 retry_count; 298 u8 rnr_retry_count; 299 u8 service_timeout; 300 u8 target_ack_delay; 301 302 struct list_head prim_list; 303 struct list_head altr_list; 304 /* Indicates that the send port mad is registered and av is set */ 305 int prim_send_port_not_ready; 306 int altr_send_port_not_ready; 307 308 struct list_head work_list; 309 atomic_t work_count; 310 }; 311 312 static void cm_work_handler(struct work_struct *work); 313 314 static inline void cm_deref_id(struct cm_id_private *cm_id_priv) 315 { 316 if (atomic_dec_and_test(&cm_id_priv->refcount)) 317 complete(&cm_id_priv->comp); 318 } 319 320 static int cm_alloc_msg(struct cm_id_private *cm_id_priv, 321 struct ib_mad_send_buf **msg) 322 { 323 struct ib_mad_agent *mad_agent; 324 struct ib_mad_send_buf *m; 325 struct ib_ah *ah; 326 struct cm_av *av; 327 unsigned long flags, flags2; 328 int ret = 0; 329 330 /* don't let the port to be released till the agent is down */ 331 spin_lock_irqsave(&cm.state_lock, flags2); 332 spin_lock_irqsave(&cm.lock, flags); 333 if (!cm_id_priv->prim_send_port_not_ready) 334 av = &cm_id_priv->av; 335 else if (!cm_id_priv->altr_send_port_not_ready && 336 (cm_id_priv->alt_av.port)) 337 av = &cm_id_priv->alt_av; 338 else { 339 pr_info("%s: not valid CM id\n", __func__); 340 ret = -ENODEV; 341 spin_unlock_irqrestore(&cm.lock, flags); 342 goto out; 343 } 344 spin_unlock_irqrestore(&cm.lock, flags); 345 /* Make sure the port haven't released the mad yet */ 346 mad_agent = cm_id_priv->av.port->mad_agent; 347 if (!mad_agent) { 348 pr_info("%s: not a valid MAD agent\n", __func__); 349 ret = -ENODEV; 350 goto out; 351 } 352 ah = ib_create_ah(mad_agent->qp->pd, &av->ah_attr, 0); 353 if (IS_ERR(ah)) { 354 ret = PTR_ERR(ah); 355 goto out; 356 } 357 358 m = ib_create_send_mad(mad_agent, cm_id_priv->id.remote_cm_qpn, 359 av->pkey_index, 360 0, IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA, 361 GFP_ATOMIC, 362 IB_MGMT_BASE_VERSION); 363 if (IS_ERR(m)) { 364 ib_destroy_ah(ah, 0); 365 ret = PTR_ERR(m); 366 goto out; 367 } 368 369 /* Timeout set by caller if response is expected. */ 370 m->ah = ah; 371 m->retries = cm_id_priv->max_cm_retries; 372 373 atomic_inc(&cm_id_priv->refcount); 374 m->context[0] = cm_id_priv; 375 *msg = m; 376 377 out: 378 spin_unlock_irqrestore(&cm.state_lock, flags2); 379 return ret; 380 } 381 382 static struct ib_mad_send_buf *cm_alloc_response_msg_no_ah(struct cm_port *port, 383 struct ib_mad_recv_wc *mad_recv_wc) 384 { 385 return ib_create_send_mad(port->mad_agent, 1, mad_recv_wc->wc->pkey_index, 386 0, IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA, 387 GFP_ATOMIC, 388 IB_MGMT_BASE_VERSION); 389 } 390 391 static int cm_create_response_msg_ah(struct cm_port *port, 392 struct ib_mad_recv_wc *mad_recv_wc, 393 struct ib_mad_send_buf *msg) 394 { 395 struct ib_ah *ah; 396 397 ah = ib_create_ah_from_wc(port->mad_agent->qp->pd, mad_recv_wc->wc, 398 mad_recv_wc->recv_buf.grh, port->port_num); 399 if (IS_ERR(ah)) 400 return PTR_ERR(ah); 401 402 msg->ah = ah; 403 return 0; 404 } 405 406 static void cm_free_msg(struct ib_mad_send_buf *msg) 407 { 408 if (msg->ah) 409 ib_destroy_ah(msg->ah, 0); 410 if (msg->context[0]) 411 cm_deref_id(msg->context[0]); 412 ib_free_send_mad(msg); 413 } 414 415 static int cm_alloc_response_msg(struct cm_port *port, 416 struct ib_mad_recv_wc *mad_recv_wc, 417 struct ib_mad_send_buf **msg) 418 { 419 struct ib_mad_send_buf *m; 420 int ret; 421 422 m = cm_alloc_response_msg_no_ah(port, mad_recv_wc); 423 if (IS_ERR(m)) 424 return PTR_ERR(m); 425 426 ret = cm_create_response_msg_ah(port, mad_recv_wc, m); 427 if (ret) { 428 cm_free_msg(m); 429 return ret; 430 } 431 432 *msg = m; 433 return 0; 434 } 435 436 static void * cm_copy_private_data(const void *private_data, 437 u8 private_data_len) 438 { 439 void *data; 440 441 if (!private_data || !private_data_len) 442 return NULL; 443 444 data = kmemdup(private_data, private_data_len, GFP_KERNEL); 445 if (!data) 446 return ERR_PTR(-ENOMEM); 447 448 return data; 449 } 450 451 static void cm_set_private_data(struct cm_id_private *cm_id_priv, 452 void *private_data, u8 private_data_len) 453 { 454 if (cm_id_priv->private_data && cm_id_priv->private_data_len) 455 kfree(cm_id_priv->private_data); 456 457 cm_id_priv->private_data = private_data; 458 cm_id_priv->private_data_len = private_data_len; 459 } 460 461 static int cm_init_av_for_lap(struct cm_port *port, struct ib_wc *wc, 462 struct ib_grh *grh, struct cm_av *av) 463 { 464 struct ib_ah_attr new_ah_attr; 465 int ret; 466 467 av->port = port; 468 av->pkey_index = wc->pkey_index; 469 470 /* 471 * av->ah_attr might be initialized based on past wc during incoming 472 * connect request or while sending out connect request. So initialize 473 * a new ah_attr on stack. If initialization fails, old ah_attr is 474 * used for sending any responses. If initialization is successful, 475 * than new ah_attr is used by overwriting old one. 476 */ 477 ret = ib_init_ah_from_wc(port->cm_dev->ib_device, 478 port->port_num, wc, 479 grh, &new_ah_attr); 480 if (ret) 481 return ret; 482 483 memcpy(&av->ah_attr, &new_ah_attr, sizeof(new_ah_attr)); 484 return 0; 485 } 486 487 static int cm_init_av_for_response(struct cm_port *port, struct ib_wc *wc, 488 struct ib_grh *grh, struct cm_av *av) 489 { 490 av->port = port; 491 av->pkey_index = wc->pkey_index; 492 return ib_init_ah_from_wc(port->cm_dev->ib_device, port->port_num, wc, 493 grh, &av->ah_attr); 494 } 495 496 static int cm_init_av_by_path(struct ib_sa_path_rec *path, struct cm_av *av, 497 struct cm_id_private *cm_id_priv) 498 { 499 struct cm_device *cm_dev; 500 struct cm_port *port = NULL; 501 unsigned long flags; 502 int ret; 503 u8 p; 504 if_t ndev = ib_get_ndev_from_path(path); 505 506 read_lock_irqsave(&cm.device_lock, flags); 507 list_for_each_entry(cm_dev, &cm.device_list, list) { 508 if (!ib_find_cached_gid(cm_dev->ib_device, &path->sgid, 509 path->gid_type, ndev, &p, NULL)) { 510 port = cm_dev->port[p-1]; 511 break; 512 } 513 } 514 read_unlock_irqrestore(&cm.device_lock, flags); 515 516 if (ndev) 517 dev_put(ndev); 518 519 if (!port) 520 return -EINVAL; 521 522 ret = ib_find_cached_pkey(cm_dev->ib_device, port->port_num, 523 be16_to_cpu(path->pkey), &av->pkey_index); 524 if (ret) 525 return ret; 526 527 av->port = port; 528 ret = ib_init_ah_from_path(cm_dev->ib_device, port->port_num, 529 path, &av->ah_attr); 530 if (ret) 531 return ret; 532 533 av->timeout = path->packet_life_time + 1; 534 535 spin_lock_irqsave(&cm.lock, flags); 536 if (&cm_id_priv->av == av) 537 list_add_tail(&cm_id_priv->prim_list, &port->cm_priv_prim_list); 538 else if (&cm_id_priv->alt_av == av) 539 list_add_tail(&cm_id_priv->altr_list, &port->cm_priv_altr_list); 540 else 541 ret = -EINVAL; 542 543 spin_unlock_irqrestore(&cm.lock, flags); 544 545 return ret; 546 } 547 548 static int cm_alloc_id(struct cm_id_private *cm_id_priv) 549 { 550 unsigned long flags; 551 int id; 552 553 idr_preload(GFP_KERNEL); 554 spin_lock_irqsave(&cm.lock, flags); 555 556 id = idr_alloc_cyclic(&cm.local_id_table, cm_id_priv, 0, 0, GFP_NOWAIT); 557 558 spin_unlock_irqrestore(&cm.lock, flags); 559 idr_preload_end(); 560 561 cm_id_priv->id.local_id = (__force __be32)id ^ cm.random_id_operand; 562 return id < 0 ? id : 0; 563 } 564 565 static void cm_free_id(__be32 local_id) 566 { 567 spin_lock_irq(&cm.lock); 568 idr_remove(&cm.local_id_table, 569 (__force int) (local_id ^ cm.random_id_operand)); 570 spin_unlock_irq(&cm.lock); 571 } 572 573 static struct cm_id_private * cm_get_id(__be32 local_id, __be32 remote_id) 574 { 575 struct cm_id_private *cm_id_priv; 576 577 cm_id_priv = idr_find(&cm.local_id_table, 578 (__force int) (local_id ^ cm.random_id_operand)); 579 if (cm_id_priv) { 580 if (cm_id_priv->id.remote_id == remote_id) 581 atomic_inc(&cm_id_priv->refcount); 582 else 583 cm_id_priv = NULL; 584 } 585 586 return cm_id_priv; 587 } 588 589 static struct cm_id_private * cm_acquire_id(__be32 local_id, __be32 remote_id) 590 { 591 struct cm_id_private *cm_id_priv; 592 593 spin_lock_irq(&cm.lock); 594 cm_id_priv = cm_get_id(local_id, remote_id); 595 spin_unlock_irq(&cm.lock); 596 597 return cm_id_priv; 598 } 599 600 /* 601 * Trivial helpers to strip endian annotation and compare; the 602 * endianness doesn't actually matter since we just need a stable 603 * order for the RB tree. 604 */ 605 static int be32_lt(__be32 a, __be32 b) 606 { 607 return (__force u32) a < (__force u32) b; 608 } 609 610 static int be32_gt(__be32 a, __be32 b) 611 { 612 return (__force u32) a > (__force u32) b; 613 } 614 615 static int be64_lt(__be64 a, __be64 b) 616 { 617 return (__force u64) a < (__force u64) b; 618 } 619 620 static int be64_gt(__be64 a, __be64 b) 621 { 622 return (__force u64) a > (__force u64) b; 623 } 624 625 static struct cm_id_private * cm_insert_listen(struct cm_id_private *cm_id_priv) 626 { 627 struct rb_node **link = &cm.listen_service_table.rb_node; 628 struct rb_node *parent = NULL; 629 struct cm_id_private *cur_cm_id_priv; 630 __be64 service_id = cm_id_priv->id.service_id; 631 __be64 service_mask = cm_id_priv->id.service_mask; 632 633 while (*link) { 634 parent = *link; 635 cur_cm_id_priv = rb_entry(parent, struct cm_id_private, 636 service_node); 637 if ((cur_cm_id_priv->id.service_mask & service_id) == 638 (service_mask & cur_cm_id_priv->id.service_id) && 639 (cm_id_priv->id.device == cur_cm_id_priv->id.device)) 640 return cur_cm_id_priv; 641 642 if (cm_id_priv->id.device < cur_cm_id_priv->id.device) 643 link = &(*link)->rb_left; 644 else if (cm_id_priv->id.device > cur_cm_id_priv->id.device) 645 link = &(*link)->rb_right; 646 else if (be64_lt(service_id, cur_cm_id_priv->id.service_id)) 647 link = &(*link)->rb_left; 648 else if (be64_gt(service_id, cur_cm_id_priv->id.service_id)) 649 link = &(*link)->rb_right; 650 else 651 link = &(*link)->rb_right; 652 } 653 rb_link_node(&cm_id_priv->service_node, parent, link); 654 rb_insert_color(&cm_id_priv->service_node, &cm.listen_service_table); 655 return NULL; 656 } 657 658 static struct cm_id_private * cm_find_listen(struct ib_device *device, 659 __be64 service_id) 660 { 661 struct rb_node *node = cm.listen_service_table.rb_node; 662 struct cm_id_private *cm_id_priv; 663 664 while (node) { 665 cm_id_priv = rb_entry(node, struct cm_id_private, service_node); 666 if ((cm_id_priv->id.service_mask & service_id) == 667 cm_id_priv->id.service_id && 668 (cm_id_priv->id.device == device)) 669 return cm_id_priv; 670 671 if (device < cm_id_priv->id.device) 672 node = node->rb_left; 673 else if (device > cm_id_priv->id.device) 674 node = node->rb_right; 675 else if (be64_lt(service_id, cm_id_priv->id.service_id)) 676 node = node->rb_left; 677 else if (be64_gt(service_id, cm_id_priv->id.service_id)) 678 node = node->rb_right; 679 else 680 node = node->rb_right; 681 } 682 return NULL; 683 } 684 685 static struct cm_timewait_info * cm_insert_remote_id(struct cm_timewait_info 686 *timewait_info) 687 { 688 struct rb_node **link = &cm.remote_id_table.rb_node; 689 struct rb_node *parent = NULL; 690 struct cm_timewait_info *cur_timewait_info; 691 __be64 remote_ca_guid = timewait_info->remote_ca_guid; 692 __be32 remote_id = timewait_info->work.remote_id; 693 694 while (*link) { 695 parent = *link; 696 cur_timewait_info = rb_entry(parent, struct cm_timewait_info, 697 remote_id_node); 698 if (be32_lt(remote_id, cur_timewait_info->work.remote_id)) 699 link = &(*link)->rb_left; 700 else if (be32_gt(remote_id, cur_timewait_info->work.remote_id)) 701 link = &(*link)->rb_right; 702 else if (be64_lt(remote_ca_guid, cur_timewait_info->remote_ca_guid)) 703 link = &(*link)->rb_left; 704 else if (be64_gt(remote_ca_guid, cur_timewait_info->remote_ca_guid)) 705 link = &(*link)->rb_right; 706 else 707 return cur_timewait_info; 708 } 709 timewait_info->inserted_remote_id = 1; 710 rb_link_node(&timewait_info->remote_id_node, parent, link); 711 rb_insert_color(&timewait_info->remote_id_node, &cm.remote_id_table); 712 return NULL; 713 } 714 715 static struct cm_timewait_info * cm_find_remote_id(__be64 remote_ca_guid, 716 __be32 remote_id) 717 { 718 struct rb_node *node = cm.remote_id_table.rb_node; 719 struct cm_timewait_info *timewait_info; 720 721 while (node) { 722 timewait_info = rb_entry(node, struct cm_timewait_info, 723 remote_id_node); 724 if (be32_lt(remote_id, timewait_info->work.remote_id)) 725 node = node->rb_left; 726 else if (be32_gt(remote_id, timewait_info->work.remote_id)) 727 node = node->rb_right; 728 else if (be64_lt(remote_ca_guid, timewait_info->remote_ca_guid)) 729 node = node->rb_left; 730 else if (be64_gt(remote_ca_guid, timewait_info->remote_ca_guid)) 731 node = node->rb_right; 732 else 733 return timewait_info; 734 } 735 return NULL; 736 } 737 738 static struct cm_timewait_info * cm_insert_remote_qpn(struct cm_timewait_info 739 *timewait_info) 740 { 741 struct rb_node **link = &cm.remote_qp_table.rb_node; 742 struct rb_node *parent = NULL; 743 struct cm_timewait_info *cur_timewait_info; 744 __be64 remote_ca_guid = timewait_info->remote_ca_guid; 745 __be32 remote_qpn = timewait_info->remote_qpn; 746 747 while (*link) { 748 parent = *link; 749 cur_timewait_info = rb_entry(parent, struct cm_timewait_info, 750 remote_qp_node); 751 if (be32_lt(remote_qpn, cur_timewait_info->remote_qpn)) 752 link = &(*link)->rb_left; 753 else if (be32_gt(remote_qpn, cur_timewait_info->remote_qpn)) 754 link = &(*link)->rb_right; 755 else if (be64_lt(remote_ca_guid, cur_timewait_info->remote_ca_guid)) 756 link = &(*link)->rb_left; 757 else if (be64_gt(remote_ca_guid, cur_timewait_info->remote_ca_guid)) 758 link = &(*link)->rb_right; 759 else 760 return cur_timewait_info; 761 } 762 timewait_info->inserted_remote_qp = 1; 763 rb_link_node(&timewait_info->remote_qp_node, parent, link); 764 rb_insert_color(&timewait_info->remote_qp_node, &cm.remote_qp_table); 765 return NULL; 766 } 767 768 static struct cm_id_private * cm_insert_remote_sidr(struct cm_id_private 769 *cm_id_priv) 770 { 771 struct rb_node **link = &cm.remote_sidr_table.rb_node; 772 struct rb_node *parent = NULL; 773 struct cm_id_private *cur_cm_id_priv; 774 union ib_gid *port_gid = &cm_id_priv->av.dgid; 775 __be32 remote_id = cm_id_priv->id.remote_id; 776 777 while (*link) { 778 parent = *link; 779 cur_cm_id_priv = rb_entry(parent, struct cm_id_private, 780 sidr_id_node); 781 if (be32_lt(remote_id, cur_cm_id_priv->id.remote_id)) 782 link = &(*link)->rb_left; 783 else if (be32_gt(remote_id, cur_cm_id_priv->id.remote_id)) 784 link = &(*link)->rb_right; 785 else { 786 int cmp; 787 cmp = memcmp(port_gid, &cur_cm_id_priv->av.dgid, 788 sizeof *port_gid); 789 if (cmp < 0) 790 link = &(*link)->rb_left; 791 else if (cmp > 0) 792 link = &(*link)->rb_right; 793 else 794 return cur_cm_id_priv; 795 } 796 } 797 rb_link_node(&cm_id_priv->sidr_id_node, parent, link); 798 rb_insert_color(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table); 799 return NULL; 800 } 801 802 static void cm_reject_sidr_req(struct cm_id_private *cm_id_priv, 803 enum ib_cm_sidr_status status) 804 { 805 struct ib_cm_sidr_rep_param param; 806 807 memset(¶m, 0, sizeof param); 808 param.status = status; 809 ib_send_cm_sidr_rep(&cm_id_priv->id, ¶m); 810 } 811 812 struct ib_cm_id *ib_create_cm_id(struct ib_device *device, 813 ib_cm_handler cm_handler, 814 void *context) 815 { 816 struct cm_id_private *cm_id_priv; 817 int ret; 818 819 cm_id_priv = kzalloc(sizeof *cm_id_priv, GFP_KERNEL); 820 if (!cm_id_priv) 821 return ERR_PTR(-ENOMEM); 822 823 cm_id_priv->id.state = IB_CM_IDLE; 824 cm_id_priv->id.device = device; 825 cm_id_priv->id.cm_handler = cm_handler; 826 cm_id_priv->id.context = context; 827 cm_id_priv->id.remote_cm_qpn = 1; 828 ret = cm_alloc_id(cm_id_priv); 829 if (ret) 830 goto error; 831 832 spin_lock_init(&cm_id_priv->lock); 833 init_completion(&cm_id_priv->comp); 834 INIT_LIST_HEAD(&cm_id_priv->work_list); 835 INIT_LIST_HEAD(&cm_id_priv->prim_list); 836 INIT_LIST_HEAD(&cm_id_priv->altr_list); 837 atomic_set(&cm_id_priv->work_count, -1); 838 atomic_set(&cm_id_priv->refcount, 1); 839 return &cm_id_priv->id; 840 841 error: 842 kfree(cm_id_priv); 843 return ERR_PTR(-ENOMEM); 844 } 845 EXPORT_SYMBOL(ib_create_cm_id); 846 847 static struct cm_work * cm_dequeue_work(struct cm_id_private *cm_id_priv) 848 { 849 struct cm_work *work; 850 851 if (list_empty(&cm_id_priv->work_list)) 852 return NULL; 853 854 work = list_entry(cm_id_priv->work_list.next, struct cm_work, list); 855 list_del(&work->list); 856 return work; 857 } 858 859 static void cm_free_work(struct cm_work *work) 860 { 861 if (work->mad_recv_wc) 862 ib_free_recv_mad(work->mad_recv_wc); 863 kfree(work); 864 } 865 866 static inline int cm_convert_to_ms(int iba_time) 867 { 868 /* approximate conversion to ms from 4.096us x 2^iba_time */ 869 return 1 << max(iba_time - 8, 0); 870 } 871 872 /* 873 * calculate: 4.096x2^ack_timeout = 4.096x2^ack_delay + 2x4.096x2^life_time 874 * Because of how ack_timeout is stored, adding one doubles the timeout. 875 * To avoid large timeouts, select the max(ack_delay, life_time + 1), and 876 * increment it (round up) only if the other is within 50%. 877 */ 878 static u8 cm_ack_timeout(u8 ca_ack_delay, u8 packet_life_time) 879 { 880 int ack_timeout = packet_life_time + 1; 881 882 if (ack_timeout >= ca_ack_delay) 883 ack_timeout += (ca_ack_delay >= (ack_timeout - 1)); 884 else 885 ack_timeout = ca_ack_delay + 886 (ack_timeout >= (ca_ack_delay - 1)); 887 888 return min(31, ack_timeout); 889 } 890 891 static void cm_cleanup_timewait(struct cm_timewait_info *timewait_info) 892 { 893 if (timewait_info->inserted_remote_id) { 894 rb_erase(&timewait_info->remote_id_node, &cm.remote_id_table); 895 timewait_info->inserted_remote_id = 0; 896 } 897 898 if (timewait_info->inserted_remote_qp) { 899 rb_erase(&timewait_info->remote_qp_node, &cm.remote_qp_table); 900 timewait_info->inserted_remote_qp = 0; 901 } 902 } 903 904 static struct cm_timewait_info * cm_create_timewait_info(__be32 local_id) 905 { 906 struct cm_timewait_info *timewait_info; 907 908 timewait_info = kzalloc(sizeof *timewait_info, GFP_KERNEL); 909 if (!timewait_info) 910 return ERR_PTR(-ENOMEM); 911 912 timewait_info->work.local_id = local_id; 913 INIT_DELAYED_WORK(&timewait_info->work.work, cm_work_handler); 914 timewait_info->work.cm_event.event = IB_CM_TIMEWAIT_EXIT; 915 return timewait_info; 916 } 917 918 static void cm_enter_timewait(struct cm_id_private *cm_id_priv) 919 { 920 int wait_time; 921 unsigned long flags; 922 struct cm_device *cm_dev; 923 924 cm_dev = ib_get_client_data(cm_id_priv->id.device, &cm_client); 925 if (!cm_dev) 926 return; 927 928 spin_lock_irqsave(&cm.lock, flags); 929 cm_cleanup_timewait(cm_id_priv->timewait_info); 930 list_add_tail(&cm_id_priv->timewait_info->list, &cm.timewait_list); 931 spin_unlock_irqrestore(&cm.lock, flags); 932 933 /* 934 * The cm_id could be destroyed by the user before we exit timewait. 935 * To protect against this, we search for the cm_id after exiting 936 * timewait before notifying the user that we've exited timewait. 937 */ 938 cm_id_priv->id.state = IB_CM_TIMEWAIT; 939 wait_time = cm_convert_to_ms(cm_id_priv->av.timeout); 940 941 /* Check if the device started its remove_one */ 942 spin_lock_irqsave(&cm.lock, flags); 943 if (!cm_dev->going_down) 944 queue_delayed_work(cm.wq, &cm_id_priv->timewait_info->work.work, 945 msecs_to_jiffies(wait_time)); 946 spin_unlock_irqrestore(&cm.lock, flags); 947 948 cm_id_priv->timewait_info = NULL; 949 } 950 951 static void cm_reset_to_idle(struct cm_id_private *cm_id_priv) 952 { 953 unsigned long flags; 954 955 cm_id_priv->id.state = IB_CM_IDLE; 956 if (cm_id_priv->timewait_info) { 957 spin_lock_irqsave(&cm.lock, flags); 958 cm_cleanup_timewait(cm_id_priv->timewait_info); 959 spin_unlock_irqrestore(&cm.lock, flags); 960 kfree(cm_id_priv->timewait_info); 961 cm_id_priv->timewait_info = NULL; 962 } 963 } 964 965 static void cm_destroy_id(struct ib_cm_id *cm_id, int err) 966 { 967 struct cm_id_private *cm_id_priv; 968 struct cm_work *work; 969 970 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 971 retest: 972 spin_lock_irq(&cm_id_priv->lock); 973 switch (cm_id->state) { 974 case IB_CM_LISTEN: 975 spin_unlock_irq(&cm_id_priv->lock); 976 977 spin_lock_irq(&cm.lock); 978 if (--cm_id_priv->listen_sharecount > 0) { 979 /* The id is still shared. */ 980 cm_deref_id(cm_id_priv); 981 spin_unlock_irq(&cm.lock); 982 return; 983 } 984 rb_erase(&cm_id_priv->service_node, &cm.listen_service_table); 985 spin_unlock_irq(&cm.lock); 986 break; 987 case IB_CM_SIDR_REQ_SENT: 988 cm_id->state = IB_CM_IDLE; 989 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 990 spin_unlock_irq(&cm_id_priv->lock); 991 break; 992 case IB_CM_SIDR_REQ_RCVD: 993 spin_unlock_irq(&cm_id_priv->lock); 994 cm_reject_sidr_req(cm_id_priv, IB_SIDR_REJECT); 995 spin_lock_irq(&cm.lock); 996 if (!RB_EMPTY_NODE(&cm_id_priv->sidr_id_node)) 997 rb_erase(&cm_id_priv->sidr_id_node, 998 &cm.remote_sidr_table); 999 spin_unlock_irq(&cm.lock); 1000 break; 1001 case IB_CM_REQ_SENT: 1002 case IB_CM_MRA_REQ_RCVD: 1003 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 1004 spin_unlock_irq(&cm_id_priv->lock); 1005 ib_send_cm_rej(cm_id, IB_CM_REJ_TIMEOUT, 1006 &cm_id_priv->id.device->node_guid, 1007 sizeof cm_id_priv->id.device->node_guid, 1008 NULL, 0); 1009 break; 1010 case IB_CM_REQ_RCVD: 1011 if (err == -ENOMEM) { 1012 /* Do not reject to allow future retries. */ 1013 cm_reset_to_idle(cm_id_priv); 1014 spin_unlock_irq(&cm_id_priv->lock); 1015 } else { 1016 spin_unlock_irq(&cm_id_priv->lock); 1017 ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED, 1018 NULL, 0, NULL, 0); 1019 } 1020 break; 1021 case IB_CM_REP_SENT: 1022 case IB_CM_MRA_REP_RCVD: 1023 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 1024 /* Fall through */ 1025 case IB_CM_MRA_REQ_SENT: 1026 case IB_CM_REP_RCVD: 1027 case IB_CM_MRA_REP_SENT: 1028 spin_unlock_irq(&cm_id_priv->lock); 1029 ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED, 1030 NULL, 0, NULL, 0); 1031 break; 1032 case IB_CM_ESTABLISHED: 1033 spin_unlock_irq(&cm_id_priv->lock); 1034 if (cm_id_priv->qp_type == IB_QPT_XRC_TGT) 1035 break; 1036 ib_send_cm_dreq(cm_id, NULL, 0); 1037 goto retest; 1038 case IB_CM_DREQ_SENT: 1039 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 1040 cm_enter_timewait(cm_id_priv); 1041 spin_unlock_irq(&cm_id_priv->lock); 1042 break; 1043 case IB_CM_DREQ_RCVD: 1044 spin_unlock_irq(&cm_id_priv->lock); 1045 ib_send_cm_drep(cm_id, NULL, 0); 1046 break; 1047 default: 1048 spin_unlock_irq(&cm_id_priv->lock); 1049 break; 1050 } 1051 1052 spin_lock_irq(&cm.lock); 1053 if (!list_empty(&cm_id_priv->altr_list) && 1054 (!cm_id_priv->altr_send_port_not_ready)) 1055 list_del(&cm_id_priv->altr_list); 1056 if (!list_empty(&cm_id_priv->prim_list) && 1057 (!cm_id_priv->prim_send_port_not_ready)) 1058 list_del(&cm_id_priv->prim_list); 1059 spin_unlock_irq(&cm.lock); 1060 1061 cm_free_id(cm_id->local_id); 1062 cm_deref_id(cm_id_priv); 1063 wait_for_completion(&cm_id_priv->comp); 1064 while ((work = cm_dequeue_work(cm_id_priv)) != NULL) 1065 cm_free_work(work); 1066 kfree(cm_id_priv->private_data); 1067 kfree(cm_id_priv); 1068 } 1069 1070 void ib_destroy_cm_id(struct ib_cm_id *cm_id) 1071 { 1072 cm_destroy_id(cm_id, 0); 1073 } 1074 EXPORT_SYMBOL(ib_destroy_cm_id); 1075 1076 /** 1077 * __ib_cm_listen - Initiates listening on the specified service ID for 1078 * connection and service ID resolution requests. 1079 * @cm_id: Connection identifier associated with the listen request. 1080 * @service_id: Service identifier matched against incoming connection 1081 * and service ID resolution requests. The service ID should be specified 1082 * network-byte order. If set to IB_CM_ASSIGN_SERVICE_ID, the CM will 1083 * assign a service ID to the caller. 1084 * @service_mask: Mask applied to service ID used to listen across a 1085 * range of service IDs. If set to 0, the service ID is matched 1086 * exactly. This parameter is ignored if %service_id is set to 1087 * IB_CM_ASSIGN_SERVICE_ID. 1088 */ 1089 static int __ib_cm_listen(struct ib_cm_id *cm_id, __be64 service_id, 1090 __be64 service_mask) 1091 { 1092 struct cm_id_private *cm_id_priv, *cur_cm_id_priv; 1093 int ret = 0; 1094 1095 service_mask = service_mask ? service_mask : ~cpu_to_be64(0); 1096 service_id &= service_mask; 1097 if ((service_id & IB_SERVICE_ID_AGN_MASK) == IB_CM_ASSIGN_SERVICE_ID && 1098 (service_id != IB_CM_ASSIGN_SERVICE_ID)) 1099 return -EINVAL; 1100 1101 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 1102 if (cm_id->state != IB_CM_IDLE) 1103 return -EINVAL; 1104 1105 cm_id->state = IB_CM_LISTEN; 1106 ++cm_id_priv->listen_sharecount; 1107 1108 if (service_id == IB_CM_ASSIGN_SERVICE_ID) { 1109 cm_id->service_id = cpu_to_be64(cm.listen_service_id++); 1110 cm_id->service_mask = ~cpu_to_be64(0); 1111 } else { 1112 cm_id->service_id = service_id; 1113 cm_id->service_mask = service_mask; 1114 } 1115 cur_cm_id_priv = cm_insert_listen(cm_id_priv); 1116 1117 if (cur_cm_id_priv) { 1118 cm_id->state = IB_CM_IDLE; 1119 --cm_id_priv->listen_sharecount; 1120 ret = -EBUSY; 1121 } 1122 return ret; 1123 } 1124 1125 int ib_cm_listen(struct ib_cm_id *cm_id, __be64 service_id, __be64 service_mask) 1126 { 1127 unsigned long flags; 1128 int ret; 1129 1130 spin_lock_irqsave(&cm.lock, flags); 1131 ret = __ib_cm_listen(cm_id, service_id, service_mask); 1132 spin_unlock_irqrestore(&cm.lock, flags); 1133 1134 return ret; 1135 } 1136 EXPORT_SYMBOL(ib_cm_listen); 1137 1138 /** 1139 * Create a new listening ib_cm_id and listen on the given service ID. 1140 * 1141 * If there's an existing ID listening on that same device and service ID, 1142 * return it. 1143 * 1144 * @device: Device associated with the cm_id. All related communication will 1145 * be associated with the specified device. 1146 * @cm_handler: Callback invoked to notify the user of CM events. 1147 * @service_id: Service identifier matched against incoming connection 1148 * and service ID resolution requests. The service ID should be specified 1149 * network-byte order. If set to IB_CM_ASSIGN_SERVICE_ID, the CM will 1150 * assign a service ID to the caller. 1151 * 1152 * Callers should call ib_destroy_cm_id when done with the listener ID. 1153 */ 1154 struct ib_cm_id *ib_cm_insert_listen(struct ib_device *device, 1155 ib_cm_handler cm_handler, 1156 __be64 service_id) 1157 { 1158 struct cm_id_private *cm_id_priv; 1159 struct ib_cm_id *cm_id; 1160 unsigned long flags; 1161 int err = 0; 1162 1163 /* Create an ID in advance, since the creation may sleep */ 1164 cm_id = ib_create_cm_id(device, cm_handler, NULL); 1165 if (IS_ERR(cm_id)) 1166 return cm_id; 1167 1168 spin_lock_irqsave(&cm.lock, flags); 1169 1170 if (service_id == IB_CM_ASSIGN_SERVICE_ID) 1171 goto new_id; 1172 1173 /* Find an existing ID */ 1174 cm_id_priv = cm_find_listen(device, service_id); 1175 if (cm_id_priv) { 1176 if (cm_id->cm_handler != cm_handler || cm_id->context) { 1177 /* Sharing an ib_cm_id with different handlers is not 1178 * supported */ 1179 spin_unlock_irqrestore(&cm.lock, flags); 1180 ib_destroy_cm_id(cm_id); 1181 return ERR_PTR(-EINVAL); 1182 } 1183 atomic_inc(&cm_id_priv->refcount); 1184 ++cm_id_priv->listen_sharecount; 1185 spin_unlock_irqrestore(&cm.lock, flags); 1186 1187 ib_destroy_cm_id(cm_id); 1188 cm_id = &cm_id_priv->id; 1189 return cm_id; 1190 } 1191 1192 new_id: 1193 /* Use newly created ID */ 1194 err = __ib_cm_listen(cm_id, service_id, 0); 1195 1196 spin_unlock_irqrestore(&cm.lock, flags); 1197 1198 if (err) { 1199 ib_destroy_cm_id(cm_id); 1200 return ERR_PTR(err); 1201 } 1202 return cm_id; 1203 } 1204 EXPORT_SYMBOL(ib_cm_insert_listen); 1205 1206 static __be64 cm_form_tid(struct cm_id_private *cm_id_priv) 1207 { 1208 u64 hi_tid, low_tid; 1209 1210 hi_tid = ((u64) cm_id_priv->av.port->mad_agent->hi_tid) << 32; 1211 low_tid = (u64)cm_id_priv->id.local_id; 1212 return cpu_to_be64(hi_tid | low_tid); 1213 } 1214 1215 static void cm_format_mad_hdr(struct ib_mad_hdr *hdr, 1216 __be16 attr_id, __be64 tid) 1217 { 1218 hdr->base_version = IB_MGMT_BASE_VERSION; 1219 hdr->mgmt_class = IB_MGMT_CLASS_CM; 1220 hdr->class_version = IB_CM_CLASS_VERSION; 1221 hdr->method = IB_MGMT_METHOD_SEND; 1222 hdr->attr_id = attr_id; 1223 hdr->tid = tid; 1224 } 1225 1226 static void cm_format_req(struct cm_req_msg *req_msg, 1227 struct cm_id_private *cm_id_priv, 1228 struct ib_cm_req_param *param) 1229 { 1230 struct ib_sa_path_rec *pri_path = param->primary_path; 1231 struct ib_sa_path_rec *alt_path = param->alternate_path; 1232 1233 cm_format_mad_hdr(&req_msg->hdr, CM_REQ_ATTR_ID, 1234 cm_form_tid(cm_id_priv)); 1235 1236 req_msg->local_comm_id = cm_id_priv->id.local_id; 1237 req_msg->service_id = param->service_id; 1238 req_msg->local_ca_guid = cm_id_priv->id.device->node_guid; 1239 cm_req_set_local_qpn(req_msg, cpu_to_be32(param->qp_num)); 1240 cm_req_set_init_depth(req_msg, param->initiator_depth); 1241 cm_req_set_remote_resp_timeout(req_msg, 1242 param->remote_cm_response_timeout); 1243 cm_req_set_qp_type(req_msg, param->qp_type); 1244 cm_req_set_flow_ctrl(req_msg, param->flow_control); 1245 cm_req_set_starting_psn(req_msg, cpu_to_be32(param->starting_psn)); 1246 cm_req_set_local_resp_timeout(req_msg, 1247 param->local_cm_response_timeout); 1248 req_msg->pkey = param->primary_path->pkey; 1249 cm_req_set_path_mtu(req_msg, param->primary_path->mtu); 1250 cm_req_set_max_cm_retries(req_msg, param->max_cm_retries); 1251 1252 if (param->qp_type != IB_QPT_XRC_INI) { 1253 cm_req_set_resp_res(req_msg, param->responder_resources); 1254 cm_req_set_retry_count(req_msg, param->retry_count); 1255 cm_req_set_rnr_retry_count(req_msg, param->rnr_retry_count); 1256 cm_req_set_srq(req_msg, param->srq); 1257 } 1258 1259 if (pri_path->hop_limit <= 1) { 1260 req_msg->primary_local_lid = pri_path->slid; 1261 req_msg->primary_remote_lid = pri_path->dlid; 1262 } else { 1263 /* Work-around until there's a way to obtain remote LID info */ 1264 req_msg->primary_local_lid = IB_LID_PERMISSIVE; 1265 req_msg->primary_remote_lid = IB_LID_PERMISSIVE; 1266 } 1267 req_msg->primary_local_gid = pri_path->sgid; 1268 req_msg->primary_remote_gid = pri_path->dgid; 1269 cm_req_set_primary_flow_label(req_msg, pri_path->flow_label); 1270 cm_req_set_primary_packet_rate(req_msg, pri_path->rate); 1271 req_msg->primary_traffic_class = pri_path->traffic_class; 1272 req_msg->primary_hop_limit = pri_path->hop_limit; 1273 cm_req_set_primary_sl(req_msg, pri_path->sl); 1274 cm_req_set_primary_subnet_local(req_msg, (pri_path->hop_limit <= 1)); 1275 cm_req_set_primary_local_ack_timeout(req_msg, 1276 cm_ack_timeout(cm_id_priv->av.port->cm_dev->ack_delay, 1277 pri_path->packet_life_time)); 1278 1279 if (alt_path) { 1280 if (alt_path->hop_limit <= 1) { 1281 req_msg->alt_local_lid = alt_path->slid; 1282 req_msg->alt_remote_lid = alt_path->dlid; 1283 } else { 1284 req_msg->alt_local_lid = IB_LID_PERMISSIVE; 1285 req_msg->alt_remote_lid = IB_LID_PERMISSIVE; 1286 } 1287 req_msg->alt_local_gid = alt_path->sgid; 1288 req_msg->alt_remote_gid = alt_path->dgid; 1289 cm_req_set_alt_flow_label(req_msg, 1290 alt_path->flow_label); 1291 cm_req_set_alt_packet_rate(req_msg, alt_path->rate); 1292 req_msg->alt_traffic_class = alt_path->traffic_class; 1293 req_msg->alt_hop_limit = alt_path->hop_limit; 1294 cm_req_set_alt_sl(req_msg, alt_path->sl); 1295 cm_req_set_alt_subnet_local(req_msg, (alt_path->hop_limit <= 1)); 1296 cm_req_set_alt_local_ack_timeout(req_msg, 1297 cm_ack_timeout(cm_id_priv->av.port->cm_dev->ack_delay, 1298 alt_path->packet_life_time)); 1299 } 1300 1301 if (param->private_data && param->private_data_len) 1302 memcpy(req_msg->private_data, param->private_data, 1303 param->private_data_len); 1304 } 1305 1306 static int cm_validate_req_param(struct ib_cm_req_param *param) 1307 { 1308 /* peer-to-peer not supported */ 1309 if (param->peer_to_peer) 1310 return -EINVAL; 1311 1312 if (!param->primary_path) 1313 return -EINVAL; 1314 1315 if (param->qp_type != IB_QPT_RC && param->qp_type != IB_QPT_UC && 1316 param->qp_type != IB_QPT_XRC_INI) 1317 return -EINVAL; 1318 1319 if (param->private_data && 1320 param->private_data_len > IB_CM_REQ_PRIVATE_DATA_SIZE) 1321 return -EINVAL; 1322 1323 if (param->alternate_path && 1324 (param->alternate_path->pkey != param->primary_path->pkey || 1325 param->alternate_path->mtu != param->primary_path->mtu)) 1326 return -EINVAL; 1327 1328 return 0; 1329 } 1330 1331 int ib_send_cm_req(struct ib_cm_id *cm_id, 1332 struct ib_cm_req_param *param) 1333 { 1334 struct cm_id_private *cm_id_priv; 1335 struct cm_req_msg *req_msg; 1336 unsigned long flags; 1337 int ret; 1338 1339 ret = cm_validate_req_param(param); 1340 if (ret) 1341 return ret; 1342 1343 /* Verify that we're not in timewait. */ 1344 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 1345 spin_lock_irqsave(&cm_id_priv->lock, flags); 1346 if (cm_id->state != IB_CM_IDLE) { 1347 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1348 ret = -EINVAL; 1349 goto out; 1350 } 1351 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1352 1353 cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv-> 1354 id.local_id); 1355 if (IS_ERR(cm_id_priv->timewait_info)) { 1356 ret = PTR_ERR(cm_id_priv->timewait_info); 1357 goto out; 1358 } 1359 1360 ret = cm_init_av_by_path(param->primary_path, &cm_id_priv->av, 1361 cm_id_priv); 1362 if (ret) 1363 goto error1; 1364 if (param->alternate_path) { 1365 ret = cm_init_av_by_path(param->alternate_path, 1366 &cm_id_priv->alt_av, cm_id_priv); 1367 if (ret) 1368 goto error1; 1369 } 1370 cm_id->service_id = param->service_id; 1371 cm_id->service_mask = ~cpu_to_be64(0); 1372 cm_id_priv->timeout_ms = cm_convert_to_ms( 1373 param->primary_path->packet_life_time) * 2 + 1374 cm_convert_to_ms( 1375 param->remote_cm_response_timeout); 1376 cm_id_priv->max_cm_retries = param->max_cm_retries; 1377 cm_id_priv->initiator_depth = param->initiator_depth; 1378 cm_id_priv->responder_resources = param->responder_resources; 1379 cm_id_priv->retry_count = param->retry_count; 1380 cm_id_priv->path_mtu = param->primary_path->mtu; 1381 cm_id_priv->pkey = param->primary_path->pkey; 1382 cm_id_priv->qp_type = param->qp_type; 1383 1384 ret = cm_alloc_msg(cm_id_priv, &cm_id_priv->msg); 1385 if (ret) 1386 goto error1; 1387 1388 req_msg = (struct cm_req_msg *) cm_id_priv->msg->mad; 1389 cm_format_req(req_msg, cm_id_priv, param); 1390 cm_id_priv->tid = req_msg->hdr.tid; 1391 cm_id_priv->msg->timeout_ms = cm_id_priv->timeout_ms; 1392 cm_id_priv->msg->context[1] = (void *) (unsigned long) IB_CM_REQ_SENT; 1393 1394 cm_id_priv->local_qpn = cm_req_get_local_qpn(req_msg); 1395 cm_id_priv->rq_psn = cm_req_get_starting_psn(req_msg); 1396 1397 spin_lock_irqsave(&cm_id_priv->lock, flags); 1398 ret = ib_post_send_mad(cm_id_priv->msg, NULL); 1399 if (ret) { 1400 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1401 goto error2; 1402 } 1403 BUG_ON(cm_id->state != IB_CM_IDLE); 1404 cm_id->state = IB_CM_REQ_SENT; 1405 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1406 return 0; 1407 1408 error2: cm_free_msg(cm_id_priv->msg); 1409 error1: kfree(cm_id_priv->timewait_info); 1410 out: return ret; 1411 } 1412 EXPORT_SYMBOL(ib_send_cm_req); 1413 1414 static int cm_issue_rej(struct cm_port *port, 1415 struct ib_mad_recv_wc *mad_recv_wc, 1416 enum ib_cm_rej_reason reason, 1417 enum cm_msg_response msg_rejected, 1418 void *ari, u8 ari_length) 1419 { 1420 struct ib_mad_send_buf *msg = NULL; 1421 struct cm_rej_msg *rej_msg, *rcv_msg; 1422 int ret; 1423 1424 ret = cm_alloc_response_msg(port, mad_recv_wc, &msg); 1425 if (ret) 1426 return ret; 1427 1428 /* We just need common CM header information. Cast to any message. */ 1429 rcv_msg = (struct cm_rej_msg *) mad_recv_wc->recv_buf.mad; 1430 rej_msg = (struct cm_rej_msg *) msg->mad; 1431 1432 cm_format_mad_hdr(&rej_msg->hdr, CM_REJ_ATTR_ID, rcv_msg->hdr.tid); 1433 rej_msg->remote_comm_id = rcv_msg->local_comm_id; 1434 rej_msg->local_comm_id = rcv_msg->remote_comm_id; 1435 cm_rej_set_msg_rejected(rej_msg, msg_rejected); 1436 rej_msg->reason = cpu_to_be16(reason); 1437 1438 if (ari && ari_length) { 1439 cm_rej_set_reject_info_len(rej_msg, ari_length); 1440 memcpy(rej_msg->ari, ari, ari_length); 1441 } 1442 1443 ret = ib_post_send_mad(msg, NULL); 1444 if (ret) 1445 cm_free_msg(msg); 1446 1447 return ret; 1448 } 1449 1450 static void cm_format_paths_from_req(struct cm_req_msg *req_msg, 1451 struct ib_sa_path_rec *primary_path, 1452 struct ib_sa_path_rec *alt_path) 1453 { 1454 memset(primary_path, 0, sizeof *primary_path); 1455 primary_path->dgid = req_msg->primary_local_gid; 1456 primary_path->sgid = req_msg->primary_remote_gid; 1457 primary_path->dlid = req_msg->primary_local_lid; 1458 primary_path->slid = req_msg->primary_remote_lid; 1459 primary_path->flow_label = cm_req_get_primary_flow_label(req_msg); 1460 primary_path->hop_limit = req_msg->primary_hop_limit; 1461 primary_path->traffic_class = req_msg->primary_traffic_class; 1462 primary_path->reversible = 1; 1463 primary_path->pkey = req_msg->pkey; 1464 primary_path->sl = cm_req_get_primary_sl(req_msg); 1465 primary_path->mtu_selector = IB_SA_EQ; 1466 primary_path->mtu = cm_req_get_path_mtu(req_msg); 1467 primary_path->rate_selector = IB_SA_EQ; 1468 primary_path->rate = cm_req_get_primary_packet_rate(req_msg); 1469 primary_path->packet_life_time_selector = IB_SA_EQ; 1470 primary_path->packet_life_time = 1471 cm_req_get_primary_local_ack_timeout(req_msg); 1472 primary_path->packet_life_time -= (primary_path->packet_life_time > 0); 1473 primary_path->service_id = req_msg->service_id; 1474 1475 if (req_msg->alt_local_lid) { 1476 memset(alt_path, 0, sizeof *alt_path); 1477 alt_path->dgid = req_msg->alt_local_gid; 1478 alt_path->sgid = req_msg->alt_remote_gid; 1479 alt_path->dlid = req_msg->alt_local_lid; 1480 alt_path->slid = req_msg->alt_remote_lid; 1481 alt_path->flow_label = cm_req_get_alt_flow_label(req_msg); 1482 alt_path->hop_limit = req_msg->alt_hop_limit; 1483 alt_path->traffic_class = req_msg->alt_traffic_class; 1484 alt_path->reversible = 1; 1485 alt_path->pkey = req_msg->pkey; 1486 alt_path->sl = cm_req_get_alt_sl(req_msg); 1487 alt_path->mtu_selector = IB_SA_EQ; 1488 alt_path->mtu = cm_req_get_path_mtu(req_msg); 1489 alt_path->rate_selector = IB_SA_EQ; 1490 alt_path->rate = cm_req_get_alt_packet_rate(req_msg); 1491 alt_path->packet_life_time_selector = IB_SA_EQ; 1492 alt_path->packet_life_time = 1493 cm_req_get_alt_local_ack_timeout(req_msg); 1494 alt_path->packet_life_time -= (alt_path->packet_life_time > 0); 1495 alt_path->service_id = req_msg->service_id; 1496 } 1497 } 1498 1499 static u16 cm_get_bth_pkey(struct cm_work *work) 1500 { 1501 struct ib_device *ib_dev = work->port->cm_dev->ib_device; 1502 u8 port_num = work->port->port_num; 1503 u16 pkey_index = work->mad_recv_wc->wc->pkey_index; 1504 u16 pkey; 1505 int ret; 1506 1507 ret = ib_get_cached_pkey(ib_dev, port_num, pkey_index, &pkey); 1508 if (ret) { 1509 dev_warn_ratelimited(&ib_dev->dev, "ib_cm: Couldn't retrieve pkey for incoming request (port %d, pkey index %d). %d\n", 1510 port_num, pkey_index, ret); 1511 return 0; 1512 } 1513 1514 return pkey; 1515 } 1516 1517 static void cm_format_req_event(struct cm_work *work, 1518 struct cm_id_private *cm_id_priv, 1519 struct ib_cm_id *listen_id) 1520 { 1521 struct cm_req_msg *req_msg; 1522 struct ib_cm_req_event_param *param; 1523 1524 req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad; 1525 param = &work->cm_event.param.req_rcvd; 1526 param->listen_id = listen_id; 1527 param->bth_pkey = cm_get_bth_pkey(work); 1528 param->port = cm_id_priv->av.port->port_num; 1529 param->primary_path = &work->path[0]; 1530 if (req_msg->alt_local_lid) 1531 param->alternate_path = &work->path[1]; 1532 else 1533 param->alternate_path = NULL; 1534 param->remote_ca_guid = req_msg->local_ca_guid; 1535 param->remote_qkey = be32_to_cpu(req_msg->local_qkey); 1536 param->remote_qpn = be32_to_cpu(cm_req_get_local_qpn(req_msg)); 1537 param->qp_type = cm_req_get_qp_type(req_msg); 1538 param->starting_psn = be32_to_cpu(cm_req_get_starting_psn(req_msg)); 1539 param->responder_resources = cm_req_get_init_depth(req_msg); 1540 param->initiator_depth = cm_req_get_resp_res(req_msg); 1541 param->local_cm_response_timeout = 1542 cm_req_get_remote_resp_timeout(req_msg); 1543 param->flow_control = cm_req_get_flow_ctrl(req_msg); 1544 param->remote_cm_response_timeout = 1545 cm_req_get_local_resp_timeout(req_msg); 1546 param->retry_count = cm_req_get_retry_count(req_msg); 1547 param->rnr_retry_count = cm_req_get_rnr_retry_count(req_msg); 1548 param->srq = cm_req_get_srq(req_msg); 1549 param->ppath_sgid_index = cm_id_priv->av.ah_attr.grh.sgid_index; 1550 work->cm_event.private_data = &req_msg->private_data; 1551 } 1552 1553 static void cm_process_work(struct cm_id_private *cm_id_priv, 1554 struct cm_work *work) 1555 { 1556 int ret; 1557 1558 /* We will typically only have the current event to report. */ 1559 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, &work->cm_event); 1560 cm_free_work(work); 1561 1562 while (!ret && !atomic_add_negative(-1, &cm_id_priv->work_count)) { 1563 spin_lock_irq(&cm_id_priv->lock); 1564 work = cm_dequeue_work(cm_id_priv); 1565 spin_unlock_irq(&cm_id_priv->lock); 1566 BUG_ON(!work); 1567 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, 1568 &work->cm_event); 1569 cm_free_work(work); 1570 } 1571 cm_deref_id(cm_id_priv); 1572 if (ret) 1573 cm_destroy_id(&cm_id_priv->id, ret); 1574 } 1575 1576 static void cm_format_mra(struct cm_mra_msg *mra_msg, 1577 struct cm_id_private *cm_id_priv, 1578 enum cm_msg_response msg_mraed, u8 service_timeout, 1579 const void *private_data, u8 private_data_len) 1580 { 1581 cm_format_mad_hdr(&mra_msg->hdr, CM_MRA_ATTR_ID, cm_id_priv->tid); 1582 cm_mra_set_msg_mraed(mra_msg, msg_mraed); 1583 mra_msg->local_comm_id = cm_id_priv->id.local_id; 1584 mra_msg->remote_comm_id = cm_id_priv->id.remote_id; 1585 cm_mra_set_service_timeout(mra_msg, service_timeout); 1586 1587 if (private_data && private_data_len) 1588 memcpy(mra_msg->private_data, private_data, private_data_len); 1589 } 1590 1591 static void cm_format_rej(struct cm_rej_msg *rej_msg, 1592 struct cm_id_private *cm_id_priv, 1593 enum ib_cm_rej_reason reason, 1594 void *ari, 1595 u8 ari_length, 1596 const void *private_data, 1597 u8 private_data_len) 1598 { 1599 cm_format_mad_hdr(&rej_msg->hdr, CM_REJ_ATTR_ID, cm_id_priv->tid); 1600 rej_msg->remote_comm_id = cm_id_priv->id.remote_id; 1601 1602 switch(cm_id_priv->id.state) { 1603 case IB_CM_REQ_RCVD: 1604 rej_msg->local_comm_id = 0; 1605 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REQ); 1606 break; 1607 case IB_CM_MRA_REQ_SENT: 1608 rej_msg->local_comm_id = cm_id_priv->id.local_id; 1609 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REQ); 1610 break; 1611 case IB_CM_REP_RCVD: 1612 case IB_CM_MRA_REP_SENT: 1613 rej_msg->local_comm_id = cm_id_priv->id.local_id; 1614 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REP); 1615 break; 1616 default: 1617 rej_msg->local_comm_id = cm_id_priv->id.local_id; 1618 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_OTHER); 1619 break; 1620 } 1621 1622 rej_msg->reason = cpu_to_be16(reason); 1623 if (ari && ari_length) { 1624 cm_rej_set_reject_info_len(rej_msg, ari_length); 1625 memcpy(rej_msg->ari, ari, ari_length); 1626 } 1627 1628 if (private_data && private_data_len) 1629 memcpy(rej_msg->private_data, private_data, private_data_len); 1630 } 1631 1632 static void cm_dup_req_handler(struct cm_work *work, 1633 struct cm_id_private *cm_id_priv) 1634 { 1635 struct ib_mad_send_buf *msg = NULL; 1636 int ret; 1637 1638 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 1639 counter[CM_REQ_COUNTER]); 1640 1641 /* Quick state check to discard duplicate REQs. */ 1642 if (cm_id_priv->id.state == IB_CM_REQ_RCVD) 1643 return; 1644 1645 ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg); 1646 if (ret) 1647 return; 1648 1649 spin_lock_irq(&cm_id_priv->lock); 1650 switch (cm_id_priv->id.state) { 1651 case IB_CM_MRA_REQ_SENT: 1652 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv, 1653 CM_MSG_RESPONSE_REQ, cm_id_priv->service_timeout, 1654 cm_id_priv->private_data, 1655 cm_id_priv->private_data_len); 1656 break; 1657 case IB_CM_TIMEWAIT: 1658 cm_format_rej((struct cm_rej_msg *) msg->mad, cm_id_priv, 1659 IB_CM_REJ_STALE_CONN, NULL, 0, NULL, 0); 1660 break; 1661 default: 1662 goto unlock; 1663 } 1664 spin_unlock_irq(&cm_id_priv->lock); 1665 1666 ret = ib_post_send_mad(msg, NULL); 1667 if (ret) 1668 goto free; 1669 return; 1670 1671 unlock: spin_unlock_irq(&cm_id_priv->lock); 1672 free: cm_free_msg(msg); 1673 } 1674 1675 static struct cm_id_private * cm_match_req(struct cm_work *work, 1676 struct cm_id_private *cm_id_priv) 1677 { 1678 struct cm_id_private *listen_cm_id_priv, *cur_cm_id_priv; 1679 struct cm_timewait_info *timewait_info; 1680 struct cm_req_msg *req_msg; 1681 struct ib_cm_id *cm_id; 1682 1683 req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad; 1684 1685 /* Check for possible duplicate REQ. */ 1686 spin_lock_irq(&cm.lock); 1687 timewait_info = cm_insert_remote_id(cm_id_priv->timewait_info); 1688 if (timewait_info) { 1689 cur_cm_id_priv = cm_get_id(timewait_info->work.local_id, 1690 timewait_info->work.remote_id); 1691 spin_unlock_irq(&cm.lock); 1692 if (cur_cm_id_priv) { 1693 cm_dup_req_handler(work, cur_cm_id_priv); 1694 cm_deref_id(cur_cm_id_priv); 1695 } 1696 return NULL; 1697 } 1698 1699 /* Check for stale connections. */ 1700 timewait_info = cm_insert_remote_qpn(cm_id_priv->timewait_info); 1701 if (timewait_info) { 1702 cm_cleanup_timewait(cm_id_priv->timewait_info); 1703 cur_cm_id_priv = cm_get_id(timewait_info->work.local_id, 1704 timewait_info->work.remote_id); 1705 1706 spin_unlock_irq(&cm.lock); 1707 cm_issue_rej(work->port, work->mad_recv_wc, 1708 IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REQ, 1709 NULL, 0); 1710 if (cur_cm_id_priv) { 1711 cm_id = &cur_cm_id_priv->id; 1712 ib_send_cm_dreq(cm_id, NULL, 0); 1713 cm_deref_id(cur_cm_id_priv); 1714 } 1715 return NULL; 1716 } 1717 1718 /* Find matching listen request. */ 1719 listen_cm_id_priv = cm_find_listen(cm_id_priv->id.device, 1720 req_msg->service_id); 1721 if (!listen_cm_id_priv) { 1722 cm_cleanup_timewait(cm_id_priv->timewait_info); 1723 spin_unlock_irq(&cm.lock); 1724 cm_issue_rej(work->port, work->mad_recv_wc, 1725 IB_CM_REJ_INVALID_SERVICE_ID, CM_MSG_RESPONSE_REQ, 1726 NULL, 0); 1727 goto out; 1728 } 1729 atomic_inc(&listen_cm_id_priv->refcount); 1730 atomic_inc(&cm_id_priv->refcount); 1731 cm_id_priv->id.state = IB_CM_REQ_RCVD; 1732 atomic_inc(&cm_id_priv->work_count); 1733 spin_unlock_irq(&cm.lock); 1734 out: 1735 return listen_cm_id_priv; 1736 } 1737 1738 /* 1739 * Work-around for inter-subnet connections. If the LIDs are permissive, 1740 * we need to override the LID/SL data in the REQ with the LID information 1741 * in the work completion. 1742 */ 1743 static void cm_process_routed_req(struct cm_req_msg *req_msg, struct ib_wc *wc) 1744 { 1745 if (!cm_req_get_primary_subnet_local(req_msg)) { 1746 if (req_msg->primary_local_lid == IB_LID_PERMISSIVE) { 1747 req_msg->primary_local_lid = cpu_to_be16(wc->slid); 1748 cm_req_set_primary_sl(req_msg, wc->sl); 1749 } 1750 1751 if (req_msg->primary_remote_lid == IB_LID_PERMISSIVE) 1752 req_msg->primary_remote_lid = cpu_to_be16(wc->dlid_path_bits); 1753 } 1754 1755 if (!cm_req_get_alt_subnet_local(req_msg)) { 1756 if (req_msg->alt_local_lid == IB_LID_PERMISSIVE) { 1757 req_msg->alt_local_lid = cpu_to_be16(wc->slid); 1758 cm_req_set_alt_sl(req_msg, wc->sl); 1759 } 1760 1761 if (req_msg->alt_remote_lid == IB_LID_PERMISSIVE) 1762 req_msg->alt_remote_lid = cpu_to_be16(wc->dlid_path_bits); 1763 } 1764 } 1765 1766 static int cm_req_handler(struct cm_work *work) 1767 { 1768 struct ib_cm_id *cm_id; 1769 struct cm_id_private *cm_id_priv, *listen_cm_id_priv; 1770 struct cm_req_msg *req_msg; 1771 union ib_gid gid; 1772 struct ib_gid_attr gid_attr; 1773 int ret; 1774 1775 req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad; 1776 1777 cm_id = ib_create_cm_id(work->port->cm_dev->ib_device, NULL, NULL); 1778 if (IS_ERR(cm_id)) 1779 return PTR_ERR(cm_id); 1780 1781 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 1782 cm_id_priv->id.remote_id = req_msg->local_comm_id; 1783 ret = cm_init_av_for_response(work->port, work->mad_recv_wc->wc, 1784 work->mad_recv_wc->recv_buf.grh, 1785 &cm_id_priv->av); 1786 if (ret) 1787 goto destroy; 1788 cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv-> 1789 id.local_id); 1790 if (IS_ERR(cm_id_priv->timewait_info)) { 1791 ret = PTR_ERR(cm_id_priv->timewait_info); 1792 goto destroy; 1793 } 1794 cm_id_priv->timewait_info->work.remote_id = req_msg->local_comm_id; 1795 cm_id_priv->timewait_info->remote_ca_guid = req_msg->local_ca_guid; 1796 cm_id_priv->timewait_info->remote_qpn = cm_req_get_local_qpn(req_msg); 1797 1798 listen_cm_id_priv = cm_match_req(work, cm_id_priv); 1799 if (!listen_cm_id_priv) { 1800 ret = -EINVAL; 1801 goto free_timeinfo; 1802 } 1803 1804 cm_id_priv->id.cm_handler = listen_cm_id_priv->id.cm_handler; 1805 cm_id_priv->id.context = listen_cm_id_priv->id.context; 1806 cm_id_priv->id.service_id = req_msg->service_id; 1807 cm_id_priv->id.service_mask = ~cpu_to_be64(0); 1808 1809 cm_process_routed_req(req_msg, work->mad_recv_wc->wc); 1810 cm_format_paths_from_req(req_msg, &work->path[0], &work->path[1]); 1811 1812 memcpy(work->path[0].dmac, cm_id_priv->av.ah_attr.dmac, ETH_ALEN); 1813 work->path[0].hop_limit = cm_id_priv->av.ah_attr.grh.hop_limit; 1814 ret = ib_get_cached_gid(work->port->cm_dev->ib_device, 1815 work->port->port_num, 1816 cm_id_priv->av.ah_attr.grh.sgid_index, 1817 &gid, &gid_attr); 1818 if (!ret) { 1819 if (gid_attr.ndev) { 1820 work->path[0].ifindex = if_getindex(gid_attr.ndev); 1821 work->path[0].net = dev_net(gid_attr.ndev); 1822 dev_put(gid_attr.ndev); 1823 } 1824 work->path[0].gid_type = gid_attr.gid_type; 1825 ret = cm_init_av_by_path(&work->path[0], &cm_id_priv->av, 1826 cm_id_priv); 1827 } 1828 if (ret) { 1829 int err = ib_get_cached_gid(work->port->cm_dev->ib_device, 1830 work->port->port_num, 0, 1831 &work->path[0].sgid, 1832 &gid_attr); 1833 if (!err && gid_attr.ndev) { 1834 work->path[0].ifindex = if_getindex(gid_attr.ndev); 1835 work->path[0].net = dev_net(gid_attr.ndev); 1836 dev_put(gid_attr.ndev); 1837 } 1838 work->path[0].gid_type = gid_attr.gid_type; 1839 ib_send_cm_rej(cm_id, IB_CM_REJ_INVALID_GID, 1840 &work->path[0].sgid, sizeof work->path[0].sgid, 1841 NULL, 0); 1842 goto rejected; 1843 } 1844 if (req_msg->alt_local_lid) { 1845 ret = cm_init_av_by_path(&work->path[1], &cm_id_priv->alt_av, 1846 cm_id_priv); 1847 if (ret) { 1848 ib_send_cm_rej(cm_id, IB_CM_REJ_INVALID_ALT_GID, 1849 &work->path[0].sgid, 1850 sizeof work->path[0].sgid, NULL, 0); 1851 goto rejected; 1852 } 1853 } 1854 cm_id_priv->tid = req_msg->hdr.tid; 1855 cm_id_priv->timeout_ms = cm_convert_to_ms( 1856 cm_req_get_local_resp_timeout(req_msg)); 1857 cm_id_priv->max_cm_retries = cm_req_get_max_cm_retries(req_msg); 1858 cm_id_priv->remote_qpn = cm_req_get_local_qpn(req_msg); 1859 cm_id_priv->initiator_depth = cm_req_get_resp_res(req_msg); 1860 cm_id_priv->responder_resources = cm_req_get_init_depth(req_msg); 1861 cm_id_priv->path_mtu = cm_req_get_path_mtu(req_msg); 1862 cm_id_priv->pkey = req_msg->pkey; 1863 cm_id_priv->sq_psn = cm_req_get_starting_psn(req_msg); 1864 cm_id_priv->retry_count = cm_req_get_retry_count(req_msg); 1865 cm_id_priv->rnr_retry_count = cm_req_get_rnr_retry_count(req_msg); 1866 cm_id_priv->qp_type = cm_req_get_qp_type(req_msg); 1867 1868 cm_format_req_event(work, cm_id_priv, &listen_cm_id_priv->id); 1869 cm_process_work(cm_id_priv, work); 1870 cm_deref_id(listen_cm_id_priv); 1871 return 0; 1872 1873 rejected: 1874 atomic_dec(&cm_id_priv->refcount); 1875 cm_deref_id(listen_cm_id_priv); 1876 free_timeinfo: 1877 kfree(cm_id_priv->timewait_info); 1878 destroy: 1879 ib_destroy_cm_id(cm_id); 1880 return ret; 1881 } 1882 1883 static void cm_format_rep(struct cm_rep_msg *rep_msg, 1884 struct cm_id_private *cm_id_priv, 1885 struct ib_cm_rep_param *param) 1886 { 1887 cm_format_mad_hdr(&rep_msg->hdr, CM_REP_ATTR_ID, cm_id_priv->tid); 1888 rep_msg->local_comm_id = cm_id_priv->id.local_id; 1889 rep_msg->remote_comm_id = cm_id_priv->id.remote_id; 1890 cm_rep_set_starting_psn(rep_msg, cpu_to_be32(param->starting_psn)); 1891 rep_msg->resp_resources = param->responder_resources; 1892 cm_rep_set_target_ack_delay(rep_msg, 1893 cm_id_priv->av.port->cm_dev->ack_delay); 1894 cm_rep_set_failover(rep_msg, param->failover_accepted); 1895 cm_rep_set_rnr_retry_count(rep_msg, param->rnr_retry_count); 1896 rep_msg->local_ca_guid = cm_id_priv->id.device->node_guid; 1897 1898 if (cm_id_priv->qp_type != IB_QPT_XRC_TGT) { 1899 rep_msg->initiator_depth = param->initiator_depth; 1900 cm_rep_set_flow_ctrl(rep_msg, param->flow_control); 1901 cm_rep_set_srq(rep_msg, param->srq); 1902 cm_rep_set_local_qpn(rep_msg, cpu_to_be32(param->qp_num)); 1903 } else { 1904 cm_rep_set_srq(rep_msg, 1); 1905 cm_rep_set_local_eecn(rep_msg, cpu_to_be32(param->qp_num)); 1906 } 1907 1908 if (param->private_data && param->private_data_len) 1909 memcpy(rep_msg->private_data, param->private_data, 1910 param->private_data_len); 1911 } 1912 1913 int ib_send_cm_rep(struct ib_cm_id *cm_id, 1914 struct ib_cm_rep_param *param) 1915 { 1916 struct cm_id_private *cm_id_priv; 1917 struct ib_mad_send_buf *msg; 1918 struct cm_rep_msg *rep_msg; 1919 unsigned long flags; 1920 int ret; 1921 1922 if (param->private_data && 1923 param->private_data_len > IB_CM_REP_PRIVATE_DATA_SIZE) 1924 return -EINVAL; 1925 1926 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 1927 spin_lock_irqsave(&cm_id_priv->lock, flags); 1928 if (cm_id->state != IB_CM_REQ_RCVD && 1929 cm_id->state != IB_CM_MRA_REQ_SENT) { 1930 ret = -EINVAL; 1931 goto out; 1932 } 1933 1934 ret = cm_alloc_msg(cm_id_priv, &msg); 1935 if (ret) 1936 goto out; 1937 1938 rep_msg = (struct cm_rep_msg *) msg->mad; 1939 cm_format_rep(rep_msg, cm_id_priv, param); 1940 msg->timeout_ms = cm_id_priv->timeout_ms; 1941 msg->context[1] = (void *) (unsigned long) IB_CM_REP_SENT; 1942 1943 ret = ib_post_send_mad(msg, NULL); 1944 if (ret) { 1945 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1946 cm_free_msg(msg); 1947 return ret; 1948 } 1949 1950 cm_id->state = IB_CM_REP_SENT; 1951 cm_id_priv->msg = msg; 1952 cm_id_priv->initiator_depth = param->initiator_depth; 1953 cm_id_priv->responder_resources = param->responder_resources; 1954 cm_id_priv->rq_psn = cm_rep_get_starting_psn(rep_msg); 1955 cm_id_priv->local_qpn = cpu_to_be32(param->qp_num & 0xFFFFFF); 1956 1957 out: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1958 return ret; 1959 } 1960 EXPORT_SYMBOL(ib_send_cm_rep); 1961 1962 static void cm_format_rtu(struct cm_rtu_msg *rtu_msg, 1963 struct cm_id_private *cm_id_priv, 1964 const void *private_data, 1965 u8 private_data_len) 1966 { 1967 cm_format_mad_hdr(&rtu_msg->hdr, CM_RTU_ATTR_ID, cm_id_priv->tid); 1968 rtu_msg->local_comm_id = cm_id_priv->id.local_id; 1969 rtu_msg->remote_comm_id = cm_id_priv->id.remote_id; 1970 1971 if (private_data && private_data_len) 1972 memcpy(rtu_msg->private_data, private_data, private_data_len); 1973 } 1974 1975 int ib_send_cm_rtu(struct ib_cm_id *cm_id, 1976 const void *private_data, 1977 u8 private_data_len) 1978 { 1979 struct cm_id_private *cm_id_priv; 1980 struct ib_mad_send_buf *msg; 1981 unsigned long flags; 1982 void *data; 1983 int ret; 1984 1985 if (private_data && private_data_len > IB_CM_RTU_PRIVATE_DATA_SIZE) 1986 return -EINVAL; 1987 1988 data = cm_copy_private_data(private_data, private_data_len); 1989 if (IS_ERR(data)) 1990 return PTR_ERR(data); 1991 1992 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 1993 spin_lock_irqsave(&cm_id_priv->lock, flags); 1994 if (cm_id->state != IB_CM_REP_RCVD && 1995 cm_id->state != IB_CM_MRA_REP_SENT) { 1996 ret = -EINVAL; 1997 goto error; 1998 } 1999 2000 ret = cm_alloc_msg(cm_id_priv, &msg); 2001 if (ret) 2002 goto error; 2003 2004 cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv, 2005 private_data, private_data_len); 2006 2007 ret = ib_post_send_mad(msg, NULL); 2008 if (ret) { 2009 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2010 cm_free_msg(msg); 2011 kfree(data); 2012 return ret; 2013 } 2014 2015 cm_id->state = IB_CM_ESTABLISHED; 2016 cm_set_private_data(cm_id_priv, data, private_data_len); 2017 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2018 return 0; 2019 2020 error: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2021 kfree(data); 2022 return ret; 2023 } 2024 EXPORT_SYMBOL(ib_send_cm_rtu); 2025 2026 static void cm_format_rep_event(struct cm_work *work, enum ib_qp_type qp_type) 2027 { 2028 struct cm_rep_msg *rep_msg; 2029 struct ib_cm_rep_event_param *param; 2030 2031 rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad; 2032 param = &work->cm_event.param.rep_rcvd; 2033 param->remote_ca_guid = rep_msg->local_ca_guid; 2034 param->remote_qkey = be32_to_cpu(rep_msg->local_qkey); 2035 param->remote_qpn = be32_to_cpu(cm_rep_get_qpn(rep_msg, qp_type)); 2036 param->starting_psn = be32_to_cpu(cm_rep_get_starting_psn(rep_msg)); 2037 param->responder_resources = rep_msg->initiator_depth; 2038 param->initiator_depth = rep_msg->resp_resources; 2039 param->target_ack_delay = cm_rep_get_target_ack_delay(rep_msg); 2040 param->failover_accepted = cm_rep_get_failover(rep_msg); 2041 param->flow_control = cm_rep_get_flow_ctrl(rep_msg); 2042 param->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg); 2043 param->srq = cm_rep_get_srq(rep_msg); 2044 work->cm_event.private_data = &rep_msg->private_data; 2045 } 2046 2047 static void cm_dup_rep_handler(struct cm_work *work) 2048 { 2049 struct cm_id_private *cm_id_priv; 2050 struct cm_rep_msg *rep_msg; 2051 struct ib_mad_send_buf *msg = NULL; 2052 int ret; 2053 2054 rep_msg = (struct cm_rep_msg *) work->mad_recv_wc->recv_buf.mad; 2055 cm_id_priv = cm_acquire_id(rep_msg->remote_comm_id, 2056 rep_msg->local_comm_id); 2057 if (!cm_id_priv) 2058 return; 2059 2060 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 2061 counter[CM_REP_COUNTER]); 2062 ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg); 2063 if (ret) 2064 goto deref; 2065 2066 spin_lock_irq(&cm_id_priv->lock); 2067 if (cm_id_priv->id.state == IB_CM_ESTABLISHED) 2068 cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv, 2069 cm_id_priv->private_data, 2070 cm_id_priv->private_data_len); 2071 else if (cm_id_priv->id.state == IB_CM_MRA_REP_SENT) 2072 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv, 2073 CM_MSG_RESPONSE_REP, cm_id_priv->service_timeout, 2074 cm_id_priv->private_data, 2075 cm_id_priv->private_data_len); 2076 else 2077 goto unlock; 2078 spin_unlock_irq(&cm_id_priv->lock); 2079 2080 ret = ib_post_send_mad(msg, NULL); 2081 if (ret) 2082 goto free; 2083 goto deref; 2084 2085 unlock: spin_unlock_irq(&cm_id_priv->lock); 2086 free: cm_free_msg(msg); 2087 deref: cm_deref_id(cm_id_priv); 2088 } 2089 2090 static int cm_rep_handler(struct cm_work *work) 2091 { 2092 struct cm_id_private *cm_id_priv; 2093 struct cm_rep_msg *rep_msg; 2094 int ret; 2095 struct cm_id_private *cur_cm_id_priv; 2096 struct ib_cm_id *cm_id; 2097 struct cm_timewait_info *timewait_info; 2098 2099 rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad; 2100 cm_id_priv = cm_acquire_id(rep_msg->remote_comm_id, 0); 2101 if (!cm_id_priv) { 2102 cm_dup_rep_handler(work); 2103 return -EINVAL; 2104 } 2105 2106 cm_format_rep_event(work, cm_id_priv->qp_type); 2107 2108 spin_lock_irq(&cm_id_priv->lock); 2109 switch (cm_id_priv->id.state) { 2110 case IB_CM_REQ_SENT: 2111 case IB_CM_MRA_REQ_RCVD: 2112 break; 2113 default: 2114 spin_unlock_irq(&cm_id_priv->lock); 2115 ret = -EINVAL; 2116 goto error; 2117 } 2118 2119 cm_id_priv->timewait_info->work.remote_id = rep_msg->local_comm_id; 2120 cm_id_priv->timewait_info->remote_ca_guid = rep_msg->local_ca_guid; 2121 cm_id_priv->timewait_info->remote_qpn = cm_rep_get_qpn(rep_msg, cm_id_priv->qp_type); 2122 2123 spin_lock(&cm.lock); 2124 /* Check for duplicate REP. */ 2125 if (cm_insert_remote_id(cm_id_priv->timewait_info)) { 2126 spin_unlock(&cm.lock); 2127 spin_unlock_irq(&cm_id_priv->lock); 2128 ret = -EINVAL; 2129 goto error; 2130 } 2131 /* Check for a stale connection. */ 2132 timewait_info = cm_insert_remote_qpn(cm_id_priv->timewait_info); 2133 if (timewait_info) { 2134 rb_erase(&cm_id_priv->timewait_info->remote_id_node, 2135 &cm.remote_id_table); 2136 cm_id_priv->timewait_info->inserted_remote_id = 0; 2137 cur_cm_id_priv = cm_get_id(timewait_info->work.local_id, 2138 timewait_info->work.remote_id); 2139 2140 spin_unlock(&cm.lock); 2141 spin_unlock_irq(&cm_id_priv->lock); 2142 cm_issue_rej(work->port, work->mad_recv_wc, 2143 IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REP, 2144 NULL, 0); 2145 ret = -EINVAL; 2146 if (cur_cm_id_priv) { 2147 cm_id = &cur_cm_id_priv->id; 2148 ib_send_cm_dreq(cm_id, NULL, 0); 2149 cm_deref_id(cur_cm_id_priv); 2150 } 2151 2152 goto error; 2153 } 2154 spin_unlock(&cm.lock); 2155 2156 cm_id_priv->id.state = IB_CM_REP_RCVD; 2157 cm_id_priv->id.remote_id = rep_msg->local_comm_id; 2158 cm_id_priv->remote_qpn = cm_rep_get_qpn(rep_msg, cm_id_priv->qp_type); 2159 cm_id_priv->initiator_depth = rep_msg->resp_resources; 2160 cm_id_priv->responder_resources = rep_msg->initiator_depth; 2161 cm_id_priv->sq_psn = cm_rep_get_starting_psn(rep_msg); 2162 cm_id_priv->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg); 2163 cm_id_priv->target_ack_delay = cm_rep_get_target_ack_delay(rep_msg); 2164 cm_id_priv->av.timeout = 2165 cm_ack_timeout(cm_id_priv->target_ack_delay, 2166 cm_id_priv->av.timeout - 1); 2167 cm_id_priv->alt_av.timeout = 2168 cm_ack_timeout(cm_id_priv->target_ack_delay, 2169 cm_id_priv->alt_av.timeout - 1); 2170 2171 /* todo: handle peer_to_peer */ 2172 2173 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2174 ret = atomic_inc_and_test(&cm_id_priv->work_count); 2175 if (!ret) 2176 list_add_tail(&work->list, &cm_id_priv->work_list); 2177 spin_unlock_irq(&cm_id_priv->lock); 2178 2179 if (ret) 2180 cm_process_work(cm_id_priv, work); 2181 else 2182 cm_deref_id(cm_id_priv); 2183 return 0; 2184 2185 error: 2186 cm_deref_id(cm_id_priv); 2187 return ret; 2188 } 2189 2190 static int cm_establish_handler(struct cm_work *work) 2191 { 2192 struct cm_id_private *cm_id_priv; 2193 int ret; 2194 2195 /* See comment in cm_establish about lookup. */ 2196 cm_id_priv = cm_acquire_id(work->local_id, work->remote_id); 2197 if (!cm_id_priv) 2198 return -EINVAL; 2199 2200 spin_lock_irq(&cm_id_priv->lock); 2201 if (cm_id_priv->id.state != IB_CM_ESTABLISHED) { 2202 spin_unlock_irq(&cm_id_priv->lock); 2203 goto out; 2204 } 2205 2206 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2207 ret = atomic_inc_and_test(&cm_id_priv->work_count); 2208 if (!ret) 2209 list_add_tail(&work->list, &cm_id_priv->work_list); 2210 spin_unlock_irq(&cm_id_priv->lock); 2211 2212 if (ret) 2213 cm_process_work(cm_id_priv, work); 2214 else 2215 cm_deref_id(cm_id_priv); 2216 return 0; 2217 out: 2218 cm_deref_id(cm_id_priv); 2219 return -EINVAL; 2220 } 2221 2222 static int cm_rtu_handler(struct cm_work *work) 2223 { 2224 struct cm_id_private *cm_id_priv; 2225 struct cm_rtu_msg *rtu_msg; 2226 int ret; 2227 2228 rtu_msg = (struct cm_rtu_msg *)work->mad_recv_wc->recv_buf.mad; 2229 cm_id_priv = cm_acquire_id(rtu_msg->remote_comm_id, 2230 rtu_msg->local_comm_id); 2231 if (!cm_id_priv) 2232 return -EINVAL; 2233 2234 work->cm_event.private_data = &rtu_msg->private_data; 2235 2236 spin_lock_irq(&cm_id_priv->lock); 2237 if (cm_id_priv->id.state != IB_CM_REP_SENT && 2238 cm_id_priv->id.state != IB_CM_MRA_REP_RCVD) { 2239 spin_unlock_irq(&cm_id_priv->lock); 2240 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 2241 counter[CM_RTU_COUNTER]); 2242 goto out; 2243 } 2244 cm_id_priv->id.state = IB_CM_ESTABLISHED; 2245 2246 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2247 ret = atomic_inc_and_test(&cm_id_priv->work_count); 2248 if (!ret) 2249 list_add_tail(&work->list, &cm_id_priv->work_list); 2250 spin_unlock_irq(&cm_id_priv->lock); 2251 2252 if (ret) 2253 cm_process_work(cm_id_priv, work); 2254 else 2255 cm_deref_id(cm_id_priv); 2256 return 0; 2257 out: 2258 cm_deref_id(cm_id_priv); 2259 return -EINVAL; 2260 } 2261 2262 static void cm_format_dreq(struct cm_dreq_msg *dreq_msg, 2263 struct cm_id_private *cm_id_priv, 2264 const void *private_data, 2265 u8 private_data_len) 2266 { 2267 cm_format_mad_hdr(&dreq_msg->hdr, CM_DREQ_ATTR_ID, 2268 cm_form_tid(cm_id_priv)); 2269 dreq_msg->local_comm_id = cm_id_priv->id.local_id; 2270 dreq_msg->remote_comm_id = cm_id_priv->id.remote_id; 2271 cm_dreq_set_remote_qpn(dreq_msg, cm_id_priv->remote_qpn); 2272 2273 if (private_data && private_data_len) 2274 memcpy(dreq_msg->private_data, private_data, private_data_len); 2275 } 2276 2277 int ib_send_cm_dreq(struct ib_cm_id *cm_id, 2278 const void *private_data, 2279 u8 private_data_len) 2280 { 2281 struct cm_id_private *cm_id_priv; 2282 struct ib_mad_send_buf *msg; 2283 unsigned long flags; 2284 int ret; 2285 2286 if (private_data && private_data_len > IB_CM_DREQ_PRIVATE_DATA_SIZE) 2287 return -EINVAL; 2288 2289 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 2290 spin_lock_irqsave(&cm_id_priv->lock, flags); 2291 if (cm_id->state != IB_CM_ESTABLISHED) { 2292 ret = -EINVAL; 2293 goto out; 2294 } 2295 2296 if (cm_id->lap_state == IB_CM_LAP_SENT || 2297 cm_id->lap_state == IB_CM_MRA_LAP_RCVD) 2298 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2299 2300 ret = cm_alloc_msg(cm_id_priv, &msg); 2301 if (ret) { 2302 cm_enter_timewait(cm_id_priv); 2303 goto out; 2304 } 2305 2306 cm_format_dreq((struct cm_dreq_msg *) msg->mad, cm_id_priv, 2307 private_data, private_data_len); 2308 msg->timeout_ms = cm_id_priv->timeout_ms; 2309 msg->context[1] = (void *) (unsigned long) IB_CM_DREQ_SENT; 2310 2311 ret = ib_post_send_mad(msg, NULL); 2312 if (ret) { 2313 cm_enter_timewait(cm_id_priv); 2314 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2315 cm_free_msg(msg); 2316 return ret; 2317 } 2318 2319 cm_id->state = IB_CM_DREQ_SENT; 2320 cm_id_priv->msg = msg; 2321 out: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2322 return ret; 2323 } 2324 EXPORT_SYMBOL(ib_send_cm_dreq); 2325 2326 static void cm_format_drep(struct cm_drep_msg *drep_msg, 2327 struct cm_id_private *cm_id_priv, 2328 const void *private_data, 2329 u8 private_data_len) 2330 { 2331 cm_format_mad_hdr(&drep_msg->hdr, CM_DREP_ATTR_ID, cm_id_priv->tid); 2332 drep_msg->local_comm_id = cm_id_priv->id.local_id; 2333 drep_msg->remote_comm_id = cm_id_priv->id.remote_id; 2334 2335 if (private_data && private_data_len) 2336 memcpy(drep_msg->private_data, private_data, private_data_len); 2337 } 2338 2339 int ib_send_cm_drep(struct ib_cm_id *cm_id, 2340 const void *private_data, 2341 u8 private_data_len) 2342 { 2343 struct cm_id_private *cm_id_priv; 2344 struct ib_mad_send_buf *msg; 2345 unsigned long flags; 2346 void *data; 2347 int ret; 2348 2349 if (private_data && private_data_len > IB_CM_DREP_PRIVATE_DATA_SIZE) 2350 return -EINVAL; 2351 2352 data = cm_copy_private_data(private_data, private_data_len); 2353 if (IS_ERR(data)) 2354 return PTR_ERR(data); 2355 2356 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 2357 spin_lock_irqsave(&cm_id_priv->lock, flags); 2358 if (cm_id->state != IB_CM_DREQ_RCVD) { 2359 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2360 kfree(data); 2361 return -EINVAL; 2362 } 2363 2364 cm_set_private_data(cm_id_priv, data, private_data_len); 2365 cm_enter_timewait(cm_id_priv); 2366 2367 ret = cm_alloc_msg(cm_id_priv, &msg); 2368 if (ret) 2369 goto out; 2370 2371 cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv, 2372 private_data, private_data_len); 2373 2374 ret = ib_post_send_mad(msg, NULL); 2375 if (ret) { 2376 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2377 cm_free_msg(msg); 2378 return ret; 2379 } 2380 2381 out: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2382 return ret; 2383 } 2384 EXPORT_SYMBOL(ib_send_cm_drep); 2385 2386 static int cm_issue_drep(struct cm_port *port, 2387 struct ib_mad_recv_wc *mad_recv_wc) 2388 { 2389 struct ib_mad_send_buf *msg = NULL; 2390 struct cm_dreq_msg *dreq_msg; 2391 struct cm_drep_msg *drep_msg; 2392 int ret; 2393 2394 ret = cm_alloc_response_msg(port, mad_recv_wc, &msg); 2395 if (ret) 2396 return ret; 2397 2398 dreq_msg = (struct cm_dreq_msg *) mad_recv_wc->recv_buf.mad; 2399 drep_msg = (struct cm_drep_msg *) msg->mad; 2400 2401 cm_format_mad_hdr(&drep_msg->hdr, CM_DREP_ATTR_ID, dreq_msg->hdr.tid); 2402 drep_msg->remote_comm_id = dreq_msg->local_comm_id; 2403 drep_msg->local_comm_id = dreq_msg->remote_comm_id; 2404 2405 ret = ib_post_send_mad(msg, NULL); 2406 if (ret) 2407 cm_free_msg(msg); 2408 2409 return ret; 2410 } 2411 2412 static int cm_dreq_handler(struct cm_work *work) 2413 { 2414 struct cm_id_private *cm_id_priv; 2415 struct cm_dreq_msg *dreq_msg; 2416 struct ib_mad_send_buf *msg = NULL; 2417 int ret; 2418 2419 dreq_msg = (struct cm_dreq_msg *)work->mad_recv_wc->recv_buf.mad; 2420 cm_id_priv = cm_acquire_id(dreq_msg->remote_comm_id, 2421 dreq_msg->local_comm_id); 2422 if (!cm_id_priv) { 2423 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 2424 counter[CM_DREQ_COUNTER]); 2425 cm_issue_drep(work->port, work->mad_recv_wc); 2426 return -EINVAL; 2427 } 2428 2429 work->cm_event.private_data = &dreq_msg->private_data; 2430 2431 spin_lock_irq(&cm_id_priv->lock); 2432 if (cm_id_priv->local_qpn != cm_dreq_get_remote_qpn(dreq_msg)) 2433 goto unlock; 2434 2435 switch (cm_id_priv->id.state) { 2436 case IB_CM_REP_SENT: 2437 case IB_CM_DREQ_SENT: 2438 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2439 break; 2440 case IB_CM_ESTABLISHED: 2441 if (cm_id_priv->id.lap_state == IB_CM_LAP_SENT || 2442 cm_id_priv->id.lap_state == IB_CM_MRA_LAP_RCVD) 2443 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2444 break; 2445 case IB_CM_MRA_REP_RCVD: 2446 break; 2447 case IB_CM_TIMEWAIT: 2448 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 2449 counter[CM_DREQ_COUNTER]); 2450 msg = cm_alloc_response_msg_no_ah(work->port, work->mad_recv_wc); 2451 if (IS_ERR(msg)) 2452 goto unlock; 2453 2454 cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv, 2455 cm_id_priv->private_data, 2456 cm_id_priv->private_data_len); 2457 spin_unlock_irq(&cm_id_priv->lock); 2458 2459 if (cm_create_response_msg_ah(work->port, work->mad_recv_wc, msg) || 2460 ib_post_send_mad(msg, NULL)) 2461 cm_free_msg(msg); 2462 goto deref; 2463 case IB_CM_DREQ_RCVD: 2464 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 2465 counter[CM_DREQ_COUNTER]); 2466 goto unlock; 2467 default: 2468 goto unlock; 2469 } 2470 cm_id_priv->id.state = IB_CM_DREQ_RCVD; 2471 cm_id_priv->tid = dreq_msg->hdr.tid; 2472 ret = atomic_inc_and_test(&cm_id_priv->work_count); 2473 if (!ret) 2474 list_add_tail(&work->list, &cm_id_priv->work_list); 2475 spin_unlock_irq(&cm_id_priv->lock); 2476 2477 if (ret) 2478 cm_process_work(cm_id_priv, work); 2479 else 2480 cm_deref_id(cm_id_priv); 2481 return 0; 2482 2483 unlock: spin_unlock_irq(&cm_id_priv->lock); 2484 deref: cm_deref_id(cm_id_priv); 2485 return -EINVAL; 2486 } 2487 2488 static int cm_drep_handler(struct cm_work *work) 2489 { 2490 struct cm_id_private *cm_id_priv; 2491 struct cm_drep_msg *drep_msg; 2492 int ret; 2493 2494 drep_msg = (struct cm_drep_msg *)work->mad_recv_wc->recv_buf.mad; 2495 cm_id_priv = cm_acquire_id(drep_msg->remote_comm_id, 2496 drep_msg->local_comm_id); 2497 if (!cm_id_priv) 2498 return -EINVAL; 2499 2500 work->cm_event.private_data = &drep_msg->private_data; 2501 2502 spin_lock_irq(&cm_id_priv->lock); 2503 if (cm_id_priv->id.state != IB_CM_DREQ_SENT && 2504 cm_id_priv->id.state != IB_CM_DREQ_RCVD) { 2505 spin_unlock_irq(&cm_id_priv->lock); 2506 goto out; 2507 } 2508 cm_enter_timewait(cm_id_priv); 2509 2510 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2511 ret = atomic_inc_and_test(&cm_id_priv->work_count); 2512 if (!ret) 2513 list_add_tail(&work->list, &cm_id_priv->work_list); 2514 spin_unlock_irq(&cm_id_priv->lock); 2515 2516 if (ret) 2517 cm_process_work(cm_id_priv, work); 2518 else 2519 cm_deref_id(cm_id_priv); 2520 return 0; 2521 out: 2522 cm_deref_id(cm_id_priv); 2523 return -EINVAL; 2524 } 2525 2526 int ib_send_cm_rej(struct ib_cm_id *cm_id, 2527 enum ib_cm_rej_reason reason, 2528 void *ari, 2529 u8 ari_length, 2530 const void *private_data, 2531 u8 private_data_len) 2532 { 2533 struct cm_id_private *cm_id_priv; 2534 struct ib_mad_send_buf *msg; 2535 unsigned long flags; 2536 int ret; 2537 2538 if ((private_data && private_data_len > IB_CM_REJ_PRIVATE_DATA_SIZE) || 2539 (ari && ari_length > IB_CM_REJ_ARI_LENGTH)) 2540 return -EINVAL; 2541 2542 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 2543 2544 spin_lock_irqsave(&cm_id_priv->lock, flags); 2545 switch (cm_id->state) { 2546 case IB_CM_REQ_SENT: 2547 case IB_CM_MRA_REQ_RCVD: 2548 case IB_CM_REQ_RCVD: 2549 case IB_CM_MRA_REQ_SENT: 2550 case IB_CM_REP_RCVD: 2551 case IB_CM_MRA_REP_SENT: 2552 ret = cm_alloc_msg(cm_id_priv, &msg); 2553 if (!ret) 2554 cm_format_rej((struct cm_rej_msg *) msg->mad, 2555 cm_id_priv, reason, ari, ari_length, 2556 private_data, private_data_len); 2557 2558 cm_reset_to_idle(cm_id_priv); 2559 break; 2560 case IB_CM_REP_SENT: 2561 case IB_CM_MRA_REP_RCVD: 2562 ret = cm_alloc_msg(cm_id_priv, &msg); 2563 if (!ret) 2564 cm_format_rej((struct cm_rej_msg *) msg->mad, 2565 cm_id_priv, reason, ari, ari_length, 2566 private_data, private_data_len); 2567 2568 cm_enter_timewait(cm_id_priv); 2569 break; 2570 default: 2571 ret = -EINVAL; 2572 goto out; 2573 } 2574 2575 if (ret) 2576 goto out; 2577 2578 ret = ib_post_send_mad(msg, NULL); 2579 if (ret) 2580 cm_free_msg(msg); 2581 2582 out: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2583 return ret; 2584 } 2585 EXPORT_SYMBOL(ib_send_cm_rej); 2586 2587 static void cm_format_rej_event(struct cm_work *work) 2588 { 2589 struct cm_rej_msg *rej_msg; 2590 struct ib_cm_rej_event_param *param; 2591 2592 rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad; 2593 param = &work->cm_event.param.rej_rcvd; 2594 param->ari = rej_msg->ari; 2595 param->ari_length = cm_rej_get_reject_info_len(rej_msg); 2596 param->reason = __be16_to_cpu(rej_msg->reason); 2597 work->cm_event.private_data = &rej_msg->private_data; 2598 } 2599 2600 static struct cm_id_private * cm_acquire_rejected_id(struct cm_rej_msg *rej_msg) 2601 { 2602 struct cm_timewait_info *timewait_info; 2603 struct cm_id_private *cm_id_priv; 2604 __be32 remote_id; 2605 2606 remote_id = rej_msg->local_comm_id; 2607 2608 if (__be16_to_cpu(rej_msg->reason) == IB_CM_REJ_TIMEOUT) { 2609 spin_lock_irq(&cm.lock); 2610 timewait_info = cm_find_remote_id( *((__be64 *) rej_msg->ari), 2611 remote_id); 2612 if (!timewait_info) { 2613 spin_unlock_irq(&cm.lock); 2614 return NULL; 2615 } 2616 cm_id_priv = idr_find(&cm.local_id_table, (__force int) 2617 (timewait_info->work.local_id ^ 2618 cm.random_id_operand)); 2619 if (cm_id_priv) { 2620 if (cm_id_priv->id.remote_id == remote_id) 2621 atomic_inc(&cm_id_priv->refcount); 2622 else 2623 cm_id_priv = NULL; 2624 } 2625 spin_unlock_irq(&cm.lock); 2626 } else if (cm_rej_get_msg_rejected(rej_msg) == CM_MSG_RESPONSE_REQ) 2627 cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, 0); 2628 else 2629 cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, remote_id); 2630 2631 return cm_id_priv; 2632 } 2633 2634 static int cm_rej_handler(struct cm_work *work) 2635 { 2636 struct cm_id_private *cm_id_priv; 2637 struct cm_rej_msg *rej_msg; 2638 int ret; 2639 2640 rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad; 2641 cm_id_priv = cm_acquire_rejected_id(rej_msg); 2642 if (!cm_id_priv) 2643 return -EINVAL; 2644 2645 cm_format_rej_event(work); 2646 2647 spin_lock_irq(&cm_id_priv->lock); 2648 switch (cm_id_priv->id.state) { 2649 case IB_CM_REQ_SENT: 2650 case IB_CM_MRA_REQ_RCVD: 2651 case IB_CM_REP_SENT: 2652 case IB_CM_MRA_REP_RCVD: 2653 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2654 /* fall through */ 2655 case IB_CM_REQ_RCVD: 2656 case IB_CM_MRA_REQ_SENT: 2657 if (__be16_to_cpu(rej_msg->reason) == IB_CM_REJ_STALE_CONN) 2658 cm_enter_timewait(cm_id_priv); 2659 else 2660 cm_reset_to_idle(cm_id_priv); 2661 break; 2662 case IB_CM_DREQ_SENT: 2663 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 2664 /* fall through */ 2665 case IB_CM_REP_RCVD: 2666 case IB_CM_MRA_REP_SENT: 2667 cm_enter_timewait(cm_id_priv); 2668 break; 2669 case IB_CM_ESTABLISHED: 2670 if (cm_id_priv->id.lap_state == IB_CM_LAP_UNINIT || 2671 cm_id_priv->id.lap_state == IB_CM_LAP_SENT) { 2672 if (cm_id_priv->id.lap_state == IB_CM_LAP_SENT) 2673 ib_cancel_mad(cm_id_priv->av.port->mad_agent, 2674 cm_id_priv->msg); 2675 cm_enter_timewait(cm_id_priv); 2676 break; 2677 } 2678 /* fall through */ 2679 default: 2680 spin_unlock_irq(&cm_id_priv->lock); 2681 ret = -EINVAL; 2682 goto out; 2683 } 2684 2685 ret = atomic_inc_and_test(&cm_id_priv->work_count); 2686 if (!ret) 2687 list_add_tail(&work->list, &cm_id_priv->work_list); 2688 spin_unlock_irq(&cm_id_priv->lock); 2689 2690 if (ret) 2691 cm_process_work(cm_id_priv, work); 2692 else 2693 cm_deref_id(cm_id_priv); 2694 return 0; 2695 out: 2696 cm_deref_id(cm_id_priv); 2697 return -EINVAL; 2698 } 2699 2700 int ib_send_cm_mra(struct ib_cm_id *cm_id, 2701 u8 service_timeout, 2702 const void *private_data, 2703 u8 private_data_len) 2704 { 2705 struct cm_id_private *cm_id_priv; 2706 struct ib_mad_send_buf *msg; 2707 enum ib_cm_state cm_state; 2708 enum ib_cm_lap_state lap_state; 2709 enum cm_msg_response msg_response; 2710 void *data; 2711 unsigned long flags; 2712 int ret; 2713 2714 if (private_data && private_data_len > IB_CM_MRA_PRIVATE_DATA_SIZE) 2715 return -EINVAL; 2716 2717 data = cm_copy_private_data(private_data, private_data_len); 2718 if (IS_ERR(data)) 2719 return PTR_ERR(data); 2720 2721 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 2722 2723 spin_lock_irqsave(&cm_id_priv->lock, flags); 2724 switch(cm_id_priv->id.state) { 2725 case IB_CM_REQ_RCVD: 2726 cm_state = IB_CM_MRA_REQ_SENT; 2727 lap_state = cm_id->lap_state; 2728 msg_response = CM_MSG_RESPONSE_REQ; 2729 break; 2730 case IB_CM_REP_RCVD: 2731 cm_state = IB_CM_MRA_REP_SENT; 2732 lap_state = cm_id->lap_state; 2733 msg_response = CM_MSG_RESPONSE_REP; 2734 break; 2735 case IB_CM_ESTABLISHED: 2736 if (cm_id->lap_state == IB_CM_LAP_RCVD) { 2737 cm_state = cm_id->state; 2738 lap_state = IB_CM_MRA_LAP_SENT; 2739 msg_response = CM_MSG_RESPONSE_OTHER; 2740 break; 2741 } 2742 default: 2743 ret = -EINVAL; 2744 goto error1; 2745 } 2746 2747 if (!(service_timeout & IB_CM_MRA_FLAG_DELAY)) { 2748 ret = cm_alloc_msg(cm_id_priv, &msg); 2749 if (ret) 2750 goto error1; 2751 2752 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv, 2753 msg_response, service_timeout, 2754 private_data, private_data_len); 2755 ret = ib_post_send_mad(msg, NULL); 2756 if (ret) 2757 goto error2; 2758 } 2759 2760 cm_id->state = cm_state; 2761 cm_id->lap_state = lap_state; 2762 cm_id_priv->service_timeout = service_timeout; 2763 cm_set_private_data(cm_id_priv, data, private_data_len); 2764 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2765 return 0; 2766 2767 error1: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2768 kfree(data); 2769 return ret; 2770 2771 error2: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2772 kfree(data); 2773 cm_free_msg(msg); 2774 return ret; 2775 } 2776 EXPORT_SYMBOL(ib_send_cm_mra); 2777 2778 static struct cm_id_private * cm_acquire_mraed_id(struct cm_mra_msg *mra_msg) 2779 { 2780 switch (cm_mra_get_msg_mraed(mra_msg)) { 2781 case CM_MSG_RESPONSE_REQ: 2782 return cm_acquire_id(mra_msg->remote_comm_id, 0); 2783 case CM_MSG_RESPONSE_REP: 2784 case CM_MSG_RESPONSE_OTHER: 2785 return cm_acquire_id(mra_msg->remote_comm_id, 2786 mra_msg->local_comm_id); 2787 default: 2788 return NULL; 2789 } 2790 } 2791 2792 static int cm_mra_handler(struct cm_work *work) 2793 { 2794 struct cm_id_private *cm_id_priv; 2795 struct cm_mra_msg *mra_msg; 2796 int timeout, ret; 2797 2798 mra_msg = (struct cm_mra_msg *)work->mad_recv_wc->recv_buf.mad; 2799 cm_id_priv = cm_acquire_mraed_id(mra_msg); 2800 if (!cm_id_priv) 2801 return -EINVAL; 2802 2803 work->cm_event.private_data = &mra_msg->private_data; 2804 work->cm_event.param.mra_rcvd.service_timeout = 2805 cm_mra_get_service_timeout(mra_msg); 2806 timeout = cm_convert_to_ms(cm_mra_get_service_timeout(mra_msg)) + 2807 cm_convert_to_ms(cm_id_priv->av.timeout); 2808 2809 spin_lock_irq(&cm_id_priv->lock); 2810 switch (cm_id_priv->id.state) { 2811 case IB_CM_REQ_SENT: 2812 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REQ || 2813 ib_modify_mad(cm_id_priv->av.port->mad_agent, 2814 cm_id_priv->msg, timeout)) 2815 goto out; 2816 cm_id_priv->id.state = IB_CM_MRA_REQ_RCVD; 2817 break; 2818 case IB_CM_REP_SENT: 2819 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REP || 2820 ib_modify_mad(cm_id_priv->av.port->mad_agent, 2821 cm_id_priv->msg, timeout)) 2822 goto out; 2823 cm_id_priv->id.state = IB_CM_MRA_REP_RCVD; 2824 break; 2825 case IB_CM_ESTABLISHED: 2826 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_OTHER || 2827 cm_id_priv->id.lap_state != IB_CM_LAP_SENT || 2828 ib_modify_mad(cm_id_priv->av.port->mad_agent, 2829 cm_id_priv->msg, timeout)) { 2830 if (cm_id_priv->id.lap_state == IB_CM_MRA_LAP_RCVD) 2831 atomic_long_inc(&work->port-> 2832 counter_group[CM_RECV_DUPLICATES]. 2833 counter[CM_MRA_COUNTER]); 2834 goto out; 2835 } 2836 cm_id_priv->id.lap_state = IB_CM_MRA_LAP_RCVD; 2837 break; 2838 case IB_CM_MRA_REQ_RCVD: 2839 case IB_CM_MRA_REP_RCVD: 2840 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 2841 counter[CM_MRA_COUNTER]); 2842 /* fall through */ 2843 default: 2844 goto out; 2845 } 2846 2847 cm_id_priv->msg->context[1] = (void *) (unsigned long) 2848 cm_id_priv->id.state; 2849 ret = atomic_inc_and_test(&cm_id_priv->work_count); 2850 if (!ret) 2851 list_add_tail(&work->list, &cm_id_priv->work_list); 2852 spin_unlock_irq(&cm_id_priv->lock); 2853 2854 if (ret) 2855 cm_process_work(cm_id_priv, work); 2856 else 2857 cm_deref_id(cm_id_priv); 2858 return 0; 2859 out: 2860 spin_unlock_irq(&cm_id_priv->lock); 2861 cm_deref_id(cm_id_priv); 2862 return -EINVAL; 2863 } 2864 2865 static void cm_format_lap(struct cm_lap_msg *lap_msg, 2866 struct cm_id_private *cm_id_priv, 2867 struct ib_sa_path_rec *alternate_path, 2868 const void *private_data, 2869 u8 private_data_len) 2870 { 2871 cm_format_mad_hdr(&lap_msg->hdr, CM_LAP_ATTR_ID, 2872 cm_form_tid(cm_id_priv)); 2873 lap_msg->local_comm_id = cm_id_priv->id.local_id; 2874 lap_msg->remote_comm_id = cm_id_priv->id.remote_id; 2875 cm_lap_set_remote_qpn(lap_msg, cm_id_priv->remote_qpn); 2876 /* todo: need remote CM response timeout */ 2877 cm_lap_set_remote_resp_timeout(lap_msg, 0x1F); 2878 lap_msg->alt_local_lid = alternate_path->slid; 2879 lap_msg->alt_remote_lid = alternate_path->dlid; 2880 lap_msg->alt_local_gid = alternate_path->sgid; 2881 lap_msg->alt_remote_gid = alternate_path->dgid; 2882 cm_lap_set_flow_label(lap_msg, alternate_path->flow_label); 2883 cm_lap_set_traffic_class(lap_msg, alternate_path->traffic_class); 2884 lap_msg->alt_hop_limit = alternate_path->hop_limit; 2885 cm_lap_set_packet_rate(lap_msg, alternate_path->rate); 2886 cm_lap_set_sl(lap_msg, alternate_path->sl); 2887 cm_lap_set_subnet_local(lap_msg, 1); /* local only... */ 2888 cm_lap_set_local_ack_timeout(lap_msg, 2889 cm_ack_timeout(cm_id_priv->av.port->cm_dev->ack_delay, 2890 alternate_path->packet_life_time)); 2891 2892 if (private_data && private_data_len) 2893 memcpy(lap_msg->private_data, private_data, private_data_len); 2894 } 2895 2896 int ib_send_cm_lap(struct ib_cm_id *cm_id, 2897 struct ib_sa_path_rec *alternate_path, 2898 const void *private_data, 2899 u8 private_data_len) 2900 { 2901 struct cm_id_private *cm_id_priv; 2902 struct ib_mad_send_buf *msg; 2903 unsigned long flags; 2904 int ret; 2905 2906 if (private_data && private_data_len > IB_CM_LAP_PRIVATE_DATA_SIZE) 2907 return -EINVAL; 2908 2909 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 2910 spin_lock_irqsave(&cm_id_priv->lock, flags); 2911 if (cm_id->state != IB_CM_ESTABLISHED || 2912 (cm_id->lap_state != IB_CM_LAP_UNINIT && 2913 cm_id->lap_state != IB_CM_LAP_IDLE)) { 2914 ret = -EINVAL; 2915 goto out; 2916 } 2917 2918 ret = cm_init_av_by_path(alternate_path, &cm_id_priv->alt_av, 2919 cm_id_priv); 2920 if (ret) 2921 goto out; 2922 cm_id_priv->alt_av.timeout = 2923 cm_ack_timeout(cm_id_priv->target_ack_delay, 2924 cm_id_priv->alt_av.timeout - 1); 2925 2926 ret = cm_alloc_msg(cm_id_priv, &msg); 2927 if (ret) 2928 goto out; 2929 2930 cm_format_lap((struct cm_lap_msg *) msg->mad, cm_id_priv, 2931 alternate_path, private_data, private_data_len); 2932 msg->timeout_ms = cm_id_priv->timeout_ms; 2933 msg->context[1] = (void *) (unsigned long) IB_CM_ESTABLISHED; 2934 2935 ret = ib_post_send_mad(msg, NULL); 2936 if (ret) { 2937 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2938 cm_free_msg(msg); 2939 return ret; 2940 } 2941 2942 cm_id->lap_state = IB_CM_LAP_SENT; 2943 cm_id_priv->msg = msg; 2944 2945 out: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 2946 return ret; 2947 } 2948 EXPORT_SYMBOL(ib_send_cm_lap); 2949 2950 static void cm_format_path_from_lap(struct cm_id_private *cm_id_priv, 2951 struct ib_sa_path_rec *path, 2952 struct cm_lap_msg *lap_msg) 2953 { 2954 memset(path, 0, sizeof *path); 2955 path->dgid = lap_msg->alt_local_gid; 2956 path->sgid = lap_msg->alt_remote_gid; 2957 path->dlid = lap_msg->alt_local_lid; 2958 path->slid = lap_msg->alt_remote_lid; 2959 path->flow_label = cm_lap_get_flow_label(lap_msg); 2960 path->hop_limit = lap_msg->alt_hop_limit; 2961 path->traffic_class = cm_lap_get_traffic_class(lap_msg); 2962 path->reversible = 1; 2963 path->pkey = cm_id_priv->pkey; 2964 path->sl = cm_lap_get_sl(lap_msg); 2965 path->mtu_selector = IB_SA_EQ; 2966 path->mtu = cm_id_priv->path_mtu; 2967 path->rate_selector = IB_SA_EQ; 2968 path->rate = cm_lap_get_packet_rate(lap_msg); 2969 path->packet_life_time_selector = IB_SA_EQ; 2970 path->packet_life_time = cm_lap_get_local_ack_timeout(lap_msg); 2971 path->packet_life_time -= (path->packet_life_time > 0); 2972 } 2973 2974 static int cm_lap_handler(struct cm_work *work) 2975 { 2976 struct cm_id_private *cm_id_priv; 2977 struct cm_lap_msg *lap_msg; 2978 struct ib_cm_lap_event_param *param; 2979 struct ib_mad_send_buf *msg = NULL; 2980 int ret; 2981 2982 /* Currently Alternate path messages are not supported for 2983 * RoCE link layer. 2984 */ 2985 if (rdma_protocol_roce(work->port->cm_dev->ib_device, 2986 work->port->port_num)) 2987 return -EINVAL; 2988 2989 /* todo: verify LAP request and send reject APR if invalid. */ 2990 lap_msg = (struct cm_lap_msg *)work->mad_recv_wc->recv_buf.mad; 2991 cm_id_priv = cm_acquire_id(lap_msg->remote_comm_id, 2992 lap_msg->local_comm_id); 2993 if (!cm_id_priv) 2994 return -EINVAL; 2995 2996 param = &work->cm_event.param.lap_rcvd; 2997 param->alternate_path = &work->path[0]; 2998 cm_format_path_from_lap(cm_id_priv, param->alternate_path, lap_msg); 2999 work->cm_event.private_data = &lap_msg->private_data; 3000 3001 spin_lock_irq(&cm_id_priv->lock); 3002 if (cm_id_priv->id.state != IB_CM_ESTABLISHED) 3003 goto unlock; 3004 3005 switch (cm_id_priv->id.lap_state) { 3006 case IB_CM_LAP_UNINIT: 3007 case IB_CM_LAP_IDLE: 3008 break; 3009 case IB_CM_MRA_LAP_SENT: 3010 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 3011 counter[CM_LAP_COUNTER]); 3012 msg = cm_alloc_response_msg_no_ah(work->port, work->mad_recv_wc); 3013 if (IS_ERR(msg)) 3014 goto unlock; 3015 3016 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv, 3017 CM_MSG_RESPONSE_OTHER, 3018 cm_id_priv->service_timeout, 3019 cm_id_priv->private_data, 3020 cm_id_priv->private_data_len); 3021 spin_unlock_irq(&cm_id_priv->lock); 3022 3023 if (cm_create_response_msg_ah(work->port, work->mad_recv_wc, msg) || 3024 ib_post_send_mad(msg, NULL)) 3025 cm_free_msg(msg); 3026 goto deref; 3027 case IB_CM_LAP_RCVD: 3028 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 3029 counter[CM_LAP_COUNTER]); 3030 goto unlock; 3031 default: 3032 goto unlock; 3033 } 3034 3035 cm_id_priv->id.lap_state = IB_CM_LAP_RCVD; 3036 cm_id_priv->tid = lap_msg->hdr.tid; 3037 ret = cm_init_av_for_response(work->port, work->mad_recv_wc->wc, 3038 work->mad_recv_wc->recv_buf.grh, 3039 &cm_id_priv->av); 3040 if (ret) 3041 goto unlock; 3042 ret = cm_init_av_by_path(param->alternate_path, &cm_id_priv->alt_av, 3043 cm_id_priv); 3044 if (ret) 3045 goto unlock; 3046 ret = atomic_inc_and_test(&cm_id_priv->work_count); 3047 if (!ret) 3048 list_add_tail(&work->list, &cm_id_priv->work_list); 3049 spin_unlock_irq(&cm_id_priv->lock); 3050 3051 if (ret) 3052 cm_process_work(cm_id_priv, work); 3053 else 3054 cm_deref_id(cm_id_priv); 3055 return 0; 3056 3057 unlock: spin_unlock_irq(&cm_id_priv->lock); 3058 deref: cm_deref_id(cm_id_priv); 3059 return -EINVAL; 3060 } 3061 3062 static void cm_format_apr(struct cm_apr_msg *apr_msg, 3063 struct cm_id_private *cm_id_priv, 3064 enum ib_cm_apr_status status, 3065 void *info, 3066 u8 info_length, 3067 const void *private_data, 3068 u8 private_data_len) 3069 { 3070 cm_format_mad_hdr(&apr_msg->hdr, CM_APR_ATTR_ID, cm_id_priv->tid); 3071 apr_msg->local_comm_id = cm_id_priv->id.local_id; 3072 apr_msg->remote_comm_id = cm_id_priv->id.remote_id; 3073 apr_msg->ap_status = (u8) status; 3074 3075 if (info && info_length) { 3076 apr_msg->info_length = info_length; 3077 memcpy(apr_msg->info, info, info_length); 3078 } 3079 3080 if (private_data && private_data_len) 3081 memcpy(apr_msg->private_data, private_data, private_data_len); 3082 } 3083 3084 int ib_send_cm_apr(struct ib_cm_id *cm_id, 3085 enum ib_cm_apr_status status, 3086 void *info, 3087 u8 info_length, 3088 const void *private_data, 3089 u8 private_data_len) 3090 { 3091 struct cm_id_private *cm_id_priv; 3092 struct ib_mad_send_buf *msg; 3093 unsigned long flags; 3094 int ret; 3095 3096 if ((private_data && private_data_len > IB_CM_APR_PRIVATE_DATA_SIZE) || 3097 (info && info_length > IB_CM_APR_INFO_LENGTH)) 3098 return -EINVAL; 3099 3100 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 3101 spin_lock_irqsave(&cm_id_priv->lock, flags); 3102 if (cm_id->state != IB_CM_ESTABLISHED || 3103 (cm_id->lap_state != IB_CM_LAP_RCVD && 3104 cm_id->lap_state != IB_CM_MRA_LAP_SENT)) { 3105 ret = -EINVAL; 3106 goto out; 3107 } 3108 3109 ret = cm_alloc_msg(cm_id_priv, &msg); 3110 if (ret) 3111 goto out; 3112 3113 cm_format_apr((struct cm_apr_msg *) msg->mad, cm_id_priv, status, 3114 info, info_length, private_data, private_data_len); 3115 ret = ib_post_send_mad(msg, NULL); 3116 if (ret) { 3117 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3118 cm_free_msg(msg); 3119 return ret; 3120 } 3121 3122 cm_id->lap_state = IB_CM_LAP_IDLE; 3123 out: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3124 return ret; 3125 } 3126 EXPORT_SYMBOL(ib_send_cm_apr); 3127 3128 static int cm_apr_handler(struct cm_work *work) 3129 { 3130 struct cm_id_private *cm_id_priv; 3131 struct cm_apr_msg *apr_msg; 3132 int ret; 3133 3134 /* Currently Alternate path messages are not supported for 3135 * RoCE link layer. 3136 */ 3137 if (rdma_protocol_roce(work->port->cm_dev->ib_device, 3138 work->port->port_num)) 3139 return -EINVAL; 3140 3141 apr_msg = (struct cm_apr_msg *)work->mad_recv_wc->recv_buf.mad; 3142 cm_id_priv = cm_acquire_id(apr_msg->remote_comm_id, 3143 apr_msg->local_comm_id); 3144 if (!cm_id_priv) 3145 return -EINVAL; /* Unmatched reply. */ 3146 3147 work->cm_event.param.apr_rcvd.ap_status = apr_msg->ap_status; 3148 work->cm_event.param.apr_rcvd.apr_info = &apr_msg->info; 3149 work->cm_event.param.apr_rcvd.info_len = apr_msg->info_length; 3150 work->cm_event.private_data = &apr_msg->private_data; 3151 3152 spin_lock_irq(&cm_id_priv->lock); 3153 if (cm_id_priv->id.state != IB_CM_ESTABLISHED || 3154 (cm_id_priv->id.lap_state != IB_CM_LAP_SENT && 3155 cm_id_priv->id.lap_state != IB_CM_MRA_LAP_RCVD)) { 3156 spin_unlock_irq(&cm_id_priv->lock); 3157 goto out; 3158 } 3159 cm_id_priv->id.lap_state = IB_CM_LAP_IDLE; 3160 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 3161 cm_id_priv->msg = NULL; 3162 3163 ret = atomic_inc_and_test(&cm_id_priv->work_count); 3164 if (!ret) 3165 list_add_tail(&work->list, &cm_id_priv->work_list); 3166 spin_unlock_irq(&cm_id_priv->lock); 3167 3168 if (ret) 3169 cm_process_work(cm_id_priv, work); 3170 else 3171 cm_deref_id(cm_id_priv); 3172 return 0; 3173 out: 3174 cm_deref_id(cm_id_priv); 3175 return -EINVAL; 3176 } 3177 3178 static int cm_timewait_handler(struct cm_work *work) 3179 { 3180 struct cm_timewait_info *timewait_info; 3181 struct cm_id_private *cm_id_priv; 3182 int ret; 3183 3184 timewait_info = (struct cm_timewait_info *)work; 3185 spin_lock_irq(&cm.lock); 3186 list_del(&timewait_info->list); 3187 spin_unlock_irq(&cm.lock); 3188 3189 cm_id_priv = cm_acquire_id(timewait_info->work.local_id, 3190 timewait_info->work.remote_id); 3191 if (!cm_id_priv) 3192 return -EINVAL; 3193 3194 spin_lock_irq(&cm_id_priv->lock); 3195 if (cm_id_priv->id.state != IB_CM_TIMEWAIT || 3196 cm_id_priv->remote_qpn != timewait_info->remote_qpn) { 3197 spin_unlock_irq(&cm_id_priv->lock); 3198 goto out; 3199 } 3200 cm_id_priv->id.state = IB_CM_IDLE; 3201 ret = atomic_inc_and_test(&cm_id_priv->work_count); 3202 if (!ret) 3203 list_add_tail(&work->list, &cm_id_priv->work_list); 3204 spin_unlock_irq(&cm_id_priv->lock); 3205 3206 if (ret) 3207 cm_process_work(cm_id_priv, work); 3208 else 3209 cm_deref_id(cm_id_priv); 3210 return 0; 3211 out: 3212 cm_deref_id(cm_id_priv); 3213 return -EINVAL; 3214 } 3215 3216 static void cm_format_sidr_req(struct cm_sidr_req_msg *sidr_req_msg, 3217 struct cm_id_private *cm_id_priv, 3218 struct ib_cm_sidr_req_param *param) 3219 { 3220 cm_format_mad_hdr(&sidr_req_msg->hdr, CM_SIDR_REQ_ATTR_ID, 3221 cm_form_tid(cm_id_priv)); 3222 sidr_req_msg->request_id = cm_id_priv->id.local_id; 3223 sidr_req_msg->pkey = param->path->pkey; 3224 sidr_req_msg->service_id = param->service_id; 3225 3226 if (param->private_data && param->private_data_len) 3227 memcpy(sidr_req_msg->private_data, param->private_data, 3228 param->private_data_len); 3229 } 3230 3231 int ib_send_cm_sidr_req(struct ib_cm_id *cm_id, 3232 struct ib_cm_sidr_req_param *param) 3233 { 3234 struct cm_id_private *cm_id_priv; 3235 struct ib_mad_send_buf *msg; 3236 unsigned long flags; 3237 int ret; 3238 3239 if (!param->path || (param->private_data && 3240 param->private_data_len > IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE)) 3241 return -EINVAL; 3242 3243 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 3244 ret = cm_init_av_by_path(param->path, &cm_id_priv->av, cm_id_priv); 3245 if (ret) 3246 goto out; 3247 3248 cm_id->service_id = param->service_id; 3249 cm_id->service_mask = ~cpu_to_be64(0); 3250 cm_id_priv->timeout_ms = param->timeout_ms; 3251 cm_id_priv->max_cm_retries = param->max_cm_retries; 3252 ret = cm_alloc_msg(cm_id_priv, &msg); 3253 if (ret) 3254 goto out; 3255 3256 cm_format_sidr_req((struct cm_sidr_req_msg *) msg->mad, cm_id_priv, 3257 param); 3258 msg->timeout_ms = cm_id_priv->timeout_ms; 3259 msg->context[1] = (void *) (unsigned long) IB_CM_SIDR_REQ_SENT; 3260 3261 spin_lock_irqsave(&cm_id_priv->lock, flags); 3262 if (cm_id->state == IB_CM_IDLE) 3263 ret = ib_post_send_mad(msg, NULL); 3264 else 3265 ret = -EINVAL; 3266 3267 if (ret) { 3268 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3269 cm_free_msg(msg); 3270 goto out; 3271 } 3272 cm_id->state = IB_CM_SIDR_REQ_SENT; 3273 cm_id_priv->msg = msg; 3274 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3275 out: 3276 return ret; 3277 } 3278 EXPORT_SYMBOL(ib_send_cm_sidr_req); 3279 3280 static void cm_format_sidr_req_event(struct cm_work *work, 3281 const struct cm_id_private *rx_cm_id, 3282 struct ib_cm_id *listen_id) 3283 { 3284 struct cm_sidr_req_msg *sidr_req_msg; 3285 struct ib_cm_sidr_req_event_param *param; 3286 3287 sidr_req_msg = (struct cm_sidr_req_msg *) 3288 work->mad_recv_wc->recv_buf.mad; 3289 param = &work->cm_event.param.sidr_req_rcvd; 3290 param->pkey = __be16_to_cpu(sidr_req_msg->pkey); 3291 param->listen_id = listen_id; 3292 param->service_id = sidr_req_msg->service_id; 3293 param->bth_pkey = cm_get_bth_pkey(work); 3294 param->port = work->port->port_num; 3295 param->sgid_index = rx_cm_id->av.ah_attr.grh.sgid_index; 3296 work->cm_event.private_data = &sidr_req_msg->private_data; 3297 } 3298 3299 static int cm_sidr_req_handler(struct cm_work *work) 3300 { 3301 struct ib_cm_id *cm_id; 3302 struct cm_id_private *cm_id_priv, *cur_cm_id_priv; 3303 struct cm_sidr_req_msg *sidr_req_msg; 3304 struct ib_wc *wc; 3305 int ret; 3306 3307 cm_id = ib_create_cm_id(work->port->cm_dev->ib_device, NULL, NULL); 3308 if (IS_ERR(cm_id)) 3309 return PTR_ERR(cm_id); 3310 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 3311 3312 /* Record SGID/SLID and request ID for lookup. */ 3313 sidr_req_msg = (struct cm_sidr_req_msg *) 3314 work->mad_recv_wc->recv_buf.mad; 3315 wc = work->mad_recv_wc->wc; 3316 cm_id_priv->av.dgid.global.subnet_prefix = cpu_to_be64(wc->slid); 3317 cm_id_priv->av.dgid.global.interface_id = 0; 3318 ret = cm_init_av_for_lap(work->port, work->mad_recv_wc->wc, 3319 work->mad_recv_wc->recv_buf.grh, 3320 &cm_id_priv->av); 3321 if (ret) 3322 goto out; 3323 cm_id_priv->id.remote_id = sidr_req_msg->request_id; 3324 cm_id_priv->tid = sidr_req_msg->hdr.tid; 3325 atomic_inc(&cm_id_priv->work_count); 3326 3327 spin_lock_irq(&cm.lock); 3328 cur_cm_id_priv = cm_insert_remote_sidr(cm_id_priv); 3329 if (cur_cm_id_priv) { 3330 spin_unlock_irq(&cm.lock); 3331 atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES]. 3332 counter[CM_SIDR_REQ_COUNTER]); 3333 goto out; /* Duplicate message. */ 3334 } 3335 cm_id_priv->id.state = IB_CM_SIDR_REQ_RCVD; 3336 cur_cm_id_priv = cm_find_listen(cm_id->device, 3337 sidr_req_msg->service_id); 3338 if (!cur_cm_id_priv) { 3339 spin_unlock_irq(&cm.lock); 3340 cm_reject_sidr_req(cm_id_priv, IB_SIDR_UNSUPPORTED); 3341 goto out; /* No match. */ 3342 } 3343 atomic_inc(&cur_cm_id_priv->refcount); 3344 atomic_inc(&cm_id_priv->refcount); 3345 spin_unlock_irq(&cm.lock); 3346 3347 cm_id_priv->id.cm_handler = cur_cm_id_priv->id.cm_handler; 3348 cm_id_priv->id.context = cur_cm_id_priv->id.context; 3349 cm_id_priv->id.service_id = sidr_req_msg->service_id; 3350 cm_id_priv->id.service_mask = ~cpu_to_be64(0); 3351 3352 cm_format_sidr_req_event(work, cm_id_priv, &cur_cm_id_priv->id); 3353 cm_process_work(cm_id_priv, work); 3354 cm_deref_id(cur_cm_id_priv); 3355 return 0; 3356 out: 3357 ib_destroy_cm_id(&cm_id_priv->id); 3358 return -EINVAL; 3359 } 3360 3361 static void cm_format_sidr_rep(struct cm_sidr_rep_msg *sidr_rep_msg, 3362 struct cm_id_private *cm_id_priv, 3363 struct ib_cm_sidr_rep_param *param) 3364 { 3365 cm_format_mad_hdr(&sidr_rep_msg->hdr, CM_SIDR_REP_ATTR_ID, 3366 cm_id_priv->tid); 3367 sidr_rep_msg->request_id = cm_id_priv->id.remote_id; 3368 sidr_rep_msg->status = param->status; 3369 cm_sidr_rep_set_qpn(sidr_rep_msg, cpu_to_be32(param->qp_num)); 3370 sidr_rep_msg->service_id = cm_id_priv->id.service_id; 3371 sidr_rep_msg->qkey = cpu_to_be32(param->qkey); 3372 3373 if (param->info && param->info_length) 3374 memcpy(sidr_rep_msg->info, param->info, param->info_length); 3375 3376 if (param->private_data && param->private_data_len) 3377 memcpy(sidr_rep_msg->private_data, param->private_data, 3378 param->private_data_len); 3379 } 3380 3381 int ib_send_cm_sidr_rep(struct ib_cm_id *cm_id, 3382 struct ib_cm_sidr_rep_param *param) 3383 { 3384 struct cm_id_private *cm_id_priv; 3385 struct ib_mad_send_buf *msg; 3386 unsigned long flags; 3387 int ret; 3388 3389 if ((param->info && param->info_length > IB_CM_SIDR_REP_INFO_LENGTH) || 3390 (param->private_data && 3391 param->private_data_len > IB_CM_SIDR_REP_PRIVATE_DATA_SIZE)) 3392 return -EINVAL; 3393 3394 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 3395 spin_lock_irqsave(&cm_id_priv->lock, flags); 3396 if (cm_id->state != IB_CM_SIDR_REQ_RCVD) { 3397 ret = -EINVAL; 3398 goto error; 3399 } 3400 3401 ret = cm_alloc_msg(cm_id_priv, &msg); 3402 if (ret) 3403 goto error; 3404 3405 cm_format_sidr_rep((struct cm_sidr_rep_msg *) msg->mad, cm_id_priv, 3406 param); 3407 ret = ib_post_send_mad(msg, NULL); 3408 if (ret) { 3409 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3410 cm_free_msg(msg); 3411 return ret; 3412 } 3413 cm_id->state = IB_CM_IDLE; 3414 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3415 3416 spin_lock_irqsave(&cm.lock, flags); 3417 if (!RB_EMPTY_NODE(&cm_id_priv->sidr_id_node)) { 3418 rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table); 3419 RB_CLEAR_NODE(&cm_id_priv->sidr_id_node); 3420 } 3421 spin_unlock_irqrestore(&cm.lock, flags); 3422 return 0; 3423 3424 error: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3425 return ret; 3426 } 3427 EXPORT_SYMBOL(ib_send_cm_sidr_rep); 3428 3429 static void cm_format_sidr_rep_event(struct cm_work *work) 3430 { 3431 struct cm_sidr_rep_msg *sidr_rep_msg; 3432 struct ib_cm_sidr_rep_event_param *param; 3433 3434 sidr_rep_msg = (struct cm_sidr_rep_msg *) 3435 work->mad_recv_wc->recv_buf.mad; 3436 param = &work->cm_event.param.sidr_rep_rcvd; 3437 param->status = sidr_rep_msg->status; 3438 param->qkey = be32_to_cpu(sidr_rep_msg->qkey); 3439 param->qpn = be32_to_cpu(cm_sidr_rep_get_qpn(sidr_rep_msg)); 3440 param->info = &sidr_rep_msg->info; 3441 param->info_len = sidr_rep_msg->info_length; 3442 work->cm_event.private_data = &sidr_rep_msg->private_data; 3443 } 3444 3445 static int cm_sidr_rep_handler(struct cm_work *work) 3446 { 3447 struct cm_sidr_rep_msg *sidr_rep_msg; 3448 struct cm_id_private *cm_id_priv; 3449 3450 sidr_rep_msg = (struct cm_sidr_rep_msg *) 3451 work->mad_recv_wc->recv_buf.mad; 3452 cm_id_priv = cm_acquire_id(sidr_rep_msg->request_id, 0); 3453 if (!cm_id_priv) 3454 return -EINVAL; /* Unmatched reply. */ 3455 3456 spin_lock_irq(&cm_id_priv->lock); 3457 if (cm_id_priv->id.state != IB_CM_SIDR_REQ_SENT) { 3458 spin_unlock_irq(&cm_id_priv->lock); 3459 goto out; 3460 } 3461 cm_id_priv->id.state = IB_CM_IDLE; 3462 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 3463 spin_unlock_irq(&cm_id_priv->lock); 3464 3465 cm_format_sidr_rep_event(work); 3466 cm_process_work(cm_id_priv, work); 3467 return 0; 3468 out: 3469 cm_deref_id(cm_id_priv); 3470 return -EINVAL; 3471 } 3472 3473 static void cm_process_send_error(struct ib_mad_send_buf *msg, 3474 enum ib_wc_status wc_status) 3475 { 3476 struct cm_id_private *cm_id_priv; 3477 struct ib_cm_event cm_event; 3478 enum ib_cm_state state; 3479 int ret; 3480 3481 memset(&cm_event, 0, sizeof cm_event); 3482 cm_id_priv = msg->context[0]; 3483 3484 /* Discard old sends or ones without a response. */ 3485 spin_lock_irq(&cm_id_priv->lock); 3486 state = (enum ib_cm_state) (unsigned long) msg->context[1]; 3487 if (msg != cm_id_priv->msg || state != cm_id_priv->id.state) 3488 goto discard; 3489 3490 switch (state) { 3491 case IB_CM_REQ_SENT: 3492 case IB_CM_MRA_REQ_RCVD: 3493 cm_reset_to_idle(cm_id_priv); 3494 cm_event.event = IB_CM_REQ_ERROR; 3495 break; 3496 case IB_CM_REP_SENT: 3497 case IB_CM_MRA_REP_RCVD: 3498 cm_reset_to_idle(cm_id_priv); 3499 cm_event.event = IB_CM_REP_ERROR; 3500 break; 3501 case IB_CM_DREQ_SENT: 3502 cm_enter_timewait(cm_id_priv); 3503 cm_event.event = IB_CM_DREQ_ERROR; 3504 break; 3505 case IB_CM_SIDR_REQ_SENT: 3506 cm_id_priv->id.state = IB_CM_IDLE; 3507 cm_event.event = IB_CM_SIDR_REQ_ERROR; 3508 break; 3509 default: 3510 goto discard; 3511 } 3512 spin_unlock_irq(&cm_id_priv->lock); 3513 cm_event.param.send_status = wc_status; 3514 3515 /* No other events can occur on the cm_id at this point. */ 3516 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, &cm_event); 3517 cm_free_msg(msg); 3518 if (ret) 3519 ib_destroy_cm_id(&cm_id_priv->id); 3520 return; 3521 discard: 3522 spin_unlock_irq(&cm_id_priv->lock); 3523 cm_free_msg(msg); 3524 } 3525 3526 static void cm_send_handler(struct ib_mad_agent *mad_agent, 3527 struct ib_mad_send_wc *mad_send_wc) 3528 { 3529 struct ib_mad_send_buf *msg = mad_send_wc->send_buf; 3530 struct cm_port *port; 3531 u16 attr_index; 3532 3533 port = mad_agent->context; 3534 attr_index = be16_to_cpu(((struct ib_mad_hdr *) 3535 msg->mad)->attr_id) - CM_ATTR_ID_OFFSET; 3536 3537 /* 3538 * If the send was in response to a received message (context[0] is not 3539 * set to a cm_id), and is not a REJ, then it is a send that was 3540 * manually retried. 3541 */ 3542 if (!msg->context[0] && (attr_index != CM_REJ_COUNTER)) 3543 msg->retries = 1; 3544 3545 atomic_long_add(1 + msg->retries, 3546 &port->counter_group[CM_XMIT].counter[attr_index]); 3547 if (msg->retries) 3548 atomic_long_add(msg->retries, 3549 &port->counter_group[CM_XMIT_RETRIES]. 3550 counter[attr_index]); 3551 3552 switch (mad_send_wc->status) { 3553 case IB_WC_SUCCESS: 3554 case IB_WC_WR_FLUSH_ERR: 3555 cm_free_msg(msg); 3556 break; 3557 default: 3558 if (msg->context[0] && msg->context[1]) 3559 cm_process_send_error(msg, mad_send_wc->status); 3560 else 3561 cm_free_msg(msg); 3562 break; 3563 } 3564 } 3565 3566 static void cm_work_handler(struct work_struct *_work) 3567 { 3568 struct cm_work *work = container_of(_work, struct cm_work, work.work); 3569 int ret; 3570 3571 switch (work->cm_event.event) { 3572 case IB_CM_REQ_RECEIVED: 3573 ret = cm_req_handler(work); 3574 break; 3575 case IB_CM_MRA_RECEIVED: 3576 ret = cm_mra_handler(work); 3577 break; 3578 case IB_CM_REJ_RECEIVED: 3579 ret = cm_rej_handler(work); 3580 break; 3581 case IB_CM_REP_RECEIVED: 3582 ret = cm_rep_handler(work); 3583 break; 3584 case IB_CM_RTU_RECEIVED: 3585 ret = cm_rtu_handler(work); 3586 break; 3587 case IB_CM_USER_ESTABLISHED: 3588 ret = cm_establish_handler(work); 3589 break; 3590 case IB_CM_DREQ_RECEIVED: 3591 ret = cm_dreq_handler(work); 3592 break; 3593 case IB_CM_DREP_RECEIVED: 3594 ret = cm_drep_handler(work); 3595 break; 3596 case IB_CM_SIDR_REQ_RECEIVED: 3597 ret = cm_sidr_req_handler(work); 3598 break; 3599 case IB_CM_SIDR_REP_RECEIVED: 3600 ret = cm_sidr_rep_handler(work); 3601 break; 3602 case IB_CM_LAP_RECEIVED: 3603 ret = cm_lap_handler(work); 3604 break; 3605 case IB_CM_APR_RECEIVED: 3606 ret = cm_apr_handler(work); 3607 break; 3608 case IB_CM_TIMEWAIT_EXIT: 3609 ret = cm_timewait_handler(work); 3610 break; 3611 default: 3612 ret = -EINVAL; 3613 break; 3614 } 3615 if (ret) 3616 cm_free_work(work); 3617 } 3618 3619 static int cm_establish(struct ib_cm_id *cm_id) 3620 { 3621 struct cm_id_private *cm_id_priv; 3622 struct cm_work *work; 3623 unsigned long flags; 3624 int ret = 0; 3625 struct cm_device *cm_dev; 3626 3627 cm_dev = ib_get_client_data(cm_id->device, &cm_client); 3628 if (!cm_dev) 3629 return -ENODEV; 3630 3631 work = kmalloc(sizeof *work, GFP_ATOMIC); 3632 if (!work) 3633 return -ENOMEM; 3634 3635 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 3636 spin_lock_irqsave(&cm_id_priv->lock, flags); 3637 switch (cm_id->state) 3638 { 3639 case IB_CM_REP_SENT: 3640 case IB_CM_MRA_REP_RCVD: 3641 cm_id->state = IB_CM_ESTABLISHED; 3642 break; 3643 case IB_CM_ESTABLISHED: 3644 ret = -EISCONN; 3645 break; 3646 default: 3647 ret = -EINVAL; 3648 break; 3649 } 3650 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3651 3652 if (ret) { 3653 kfree(work); 3654 goto out; 3655 } 3656 3657 /* 3658 * The CM worker thread may try to destroy the cm_id before it 3659 * can execute this work item. To prevent potential deadlock, 3660 * we need to find the cm_id once we're in the context of the 3661 * worker thread, rather than holding a reference on it. 3662 */ 3663 INIT_DELAYED_WORK(&work->work, cm_work_handler); 3664 work->local_id = cm_id->local_id; 3665 work->remote_id = cm_id->remote_id; 3666 work->mad_recv_wc = NULL; 3667 work->cm_event.event = IB_CM_USER_ESTABLISHED; 3668 3669 /* Check if the device started its remove_one */ 3670 spin_lock_irqsave(&cm.lock, flags); 3671 if (!cm_dev->going_down) { 3672 queue_delayed_work(cm.wq, &work->work, 0); 3673 } else { 3674 kfree(work); 3675 ret = -ENODEV; 3676 } 3677 spin_unlock_irqrestore(&cm.lock, flags); 3678 3679 out: 3680 return ret; 3681 } 3682 3683 static int cm_migrate(struct ib_cm_id *cm_id) 3684 { 3685 struct cm_id_private *cm_id_priv; 3686 struct cm_av tmp_av; 3687 unsigned long flags; 3688 int tmp_send_port_not_ready; 3689 int ret = 0; 3690 3691 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 3692 spin_lock_irqsave(&cm_id_priv->lock, flags); 3693 if (cm_id->state == IB_CM_ESTABLISHED && 3694 (cm_id->lap_state == IB_CM_LAP_UNINIT || 3695 cm_id->lap_state == IB_CM_LAP_IDLE)) { 3696 cm_id->lap_state = IB_CM_LAP_IDLE; 3697 /* Swap address vector */ 3698 tmp_av = cm_id_priv->av; 3699 cm_id_priv->av = cm_id_priv->alt_av; 3700 cm_id_priv->alt_av = tmp_av; 3701 /* Swap port send ready state */ 3702 tmp_send_port_not_ready = cm_id_priv->prim_send_port_not_ready; 3703 cm_id_priv->prim_send_port_not_ready = cm_id_priv->altr_send_port_not_ready; 3704 cm_id_priv->altr_send_port_not_ready = tmp_send_port_not_ready; 3705 } else 3706 ret = -EINVAL; 3707 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3708 3709 return ret; 3710 } 3711 3712 int ib_cm_notify(struct ib_cm_id *cm_id, enum ib_event_type event) 3713 { 3714 int ret; 3715 3716 switch (event) { 3717 case IB_EVENT_COMM_EST: 3718 ret = cm_establish(cm_id); 3719 break; 3720 case IB_EVENT_PATH_MIG: 3721 ret = cm_migrate(cm_id); 3722 break; 3723 default: 3724 ret = -EINVAL; 3725 } 3726 return ret; 3727 } 3728 EXPORT_SYMBOL(ib_cm_notify); 3729 3730 static void cm_recv_handler(struct ib_mad_agent *mad_agent, 3731 struct ib_mad_send_buf *send_buf, 3732 struct ib_mad_recv_wc *mad_recv_wc) 3733 { 3734 struct cm_port *port = mad_agent->context; 3735 struct cm_work *work; 3736 enum ib_cm_event_type event; 3737 u16 attr_id; 3738 int paths = 0; 3739 int going_down = 0; 3740 3741 switch (mad_recv_wc->recv_buf.mad->mad_hdr.attr_id) { 3742 case CM_REQ_ATTR_ID: 3743 paths = 1 + (((struct cm_req_msg *) mad_recv_wc->recv_buf.mad)-> 3744 alt_local_lid != 0); 3745 event = IB_CM_REQ_RECEIVED; 3746 break; 3747 case CM_MRA_ATTR_ID: 3748 event = IB_CM_MRA_RECEIVED; 3749 break; 3750 case CM_REJ_ATTR_ID: 3751 event = IB_CM_REJ_RECEIVED; 3752 break; 3753 case CM_REP_ATTR_ID: 3754 event = IB_CM_REP_RECEIVED; 3755 break; 3756 case CM_RTU_ATTR_ID: 3757 event = IB_CM_RTU_RECEIVED; 3758 break; 3759 case CM_DREQ_ATTR_ID: 3760 event = IB_CM_DREQ_RECEIVED; 3761 break; 3762 case CM_DREP_ATTR_ID: 3763 event = IB_CM_DREP_RECEIVED; 3764 break; 3765 case CM_SIDR_REQ_ATTR_ID: 3766 event = IB_CM_SIDR_REQ_RECEIVED; 3767 break; 3768 case CM_SIDR_REP_ATTR_ID: 3769 event = IB_CM_SIDR_REP_RECEIVED; 3770 break; 3771 case CM_LAP_ATTR_ID: 3772 paths = 1; 3773 event = IB_CM_LAP_RECEIVED; 3774 break; 3775 case CM_APR_ATTR_ID: 3776 event = IB_CM_APR_RECEIVED; 3777 break; 3778 default: 3779 ib_free_recv_mad(mad_recv_wc); 3780 return; 3781 } 3782 3783 attr_id = be16_to_cpu(mad_recv_wc->recv_buf.mad->mad_hdr.attr_id); 3784 atomic_long_inc(&port->counter_group[CM_RECV]. 3785 counter[attr_id - CM_ATTR_ID_OFFSET]); 3786 3787 work = kmalloc(sizeof *work + sizeof(struct ib_sa_path_rec) * paths, 3788 GFP_KERNEL); 3789 if (!work) { 3790 ib_free_recv_mad(mad_recv_wc); 3791 return; 3792 } 3793 3794 INIT_DELAYED_WORK(&work->work, cm_work_handler); 3795 work->cm_event.event = event; 3796 work->mad_recv_wc = mad_recv_wc; 3797 work->port = port; 3798 3799 /* Check if the device started its remove_one */ 3800 spin_lock_irq(&cm.lock); 3801 if (!port->cm_dev->going_down) 3802 queue_delayed_work(cm.wq, &work->work, 0); 3803 else 3804 going_down = 1; 3805 spin_unlock_irq(&cm.lock); 3806 3807 if (going_down) { 3808 kfree(work); 3809 ib_free_recv_mad(mad_recv_wc); 3810 } 3811 } 3812 3813 static int cm_init_qp_init_attr(struct cm_id_private *cm_id_priv, 3814 struct ib_qp_attr *qp_attr, 3815 int *qp_attr_mask) 3816 { 3817 unsigned long flags; 3818 int ret; 3819 3820 spin_lock_irqsave(&cm_id_priv->lock, flags); 3821 switch (cm_id_priv->id.state) { 3822 case IB_CM_REQ_SENT: 3823 case IB_CM_MRA_REQ_RCVD: 3824 case IB_CM_REQ_RCVD: 3825 case IB_CM_MRA_REQ_SENT: 3826 case IB_CM_REP_RCVD: 3827 case IB_CM_MRA_REP_SENT: 3828 case IB_CM_REP_SENT: 3829 case IB_CM_MRA_REP_RCVD: 3830 case IB_CM_ESTABLISHED: 3831 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS | 3832 IB_QP_PKEY_INDEX | IB_QP_PORT; 3833 qp_attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE; 3834 if (cm_id_priv->responder_resources) 3835 qp_attr->qp_access_flags |= IB_ACCESS_REMOTE_READ | 3836 IB_ACCESS_REMOTE_ATOMIC; 3837 qp_attr->pkey_index = cm_id_priv->av.pkey_index; 3838 qp_attr->port_num = cm_id_priv->av.port->port_num; 3839 ret = 0; 3840 break; 3841 default: 3842 ret = -EINVAL; 3843 break; 3844 } 3845 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3846 return ret; 3847 } 3848 3849 static int cm_init_qp_rtr_attr(struct cm_id_private *cm_id_priv, 3850 struct ib_qp_attr *qp_attr, 3851 int *qp_attr_mask) 3852 { 3853 unsigned long flags; 3854 int ret; 3855 3856 spin_lock_irqsave(&cm_id_priv->lock, flags); 3857 switch (cm_id_priv->id.state) { 3858 case IB_CM_REQ_RCVD: 3859 case IB_CM_MRA_REQ_SENT: 3860 case IB_CM_REP_RCVD: 3861 case IB_CM_MRA_REP_SENT: 3862 case IB_CM_REP_SENT: 3863 case IB_CM_MRA_REP_RCVD: 3864 case IB_CM_ESTABLISHED: 3865 *qp_attr_mask = IB_QP_STATE | IB_QP_AV | IB_QP_PATH_MTU | 3866 IB_QP_DEST_QPN | IB_QP_RQ_PSN; 3867 qp_attr->ah_attr = cm_id_priv->av.ah_attr; 3868 qp_attr->path_mtu = cm_id_priv->path_mtu; 3869 qp_attr->dest_qp_num = be32_to_cpu(cm_id_priv->remote_qpn); 3870 qp_attr->rq_psn = be32_to_cpu(cm_id_priv->rq_psn); 3871 if (cm_id_priv->qp_type == IB_QPT_RC || 3872 cm_id_priv->qp_type == IB_QPT_XRC_TGT) { 3873 *qp_attr_mask |= IB_QP_MAX_DEST_RD_ATOMIC | 3874 IB_QP_MIN_RNR_TIMER; 3875 qp_attr->max_dest_rd_atomic = 3876 cm_id_priv->responder_resources; 3877 qp_attr->min_rnr_timer = 0; 3878 } 3879 if (cm_id_priv->alt_av.ah_attr.dlid) { 3880 *qp_attr_mask |= IB_QP_ALT_PATH; 3881 qp_attr->alt_port_num = cm_id_priv->alt_av.port->port_num; 3882 qp_attr->alt_pkey_index = cm_id_priv->alt_av.pkey_index; 3883 qp_attr->alt_timeout = cm_id_priv->alt_av.timeout; 3884 qp_attr->alt_ah_attr = cm_id_priv->alt_av.ah_attr; 3885 } 3886 ret = 0; 3887 break; 3888 default: 3889 ret = -EINVAL; 3890 break; 3891 } 3892 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3893 return ret; 3894 } 3895 3896 static int cm_init_qp_rts_attr(struct cm_id_private *cm_id_priv, 3897 struct ib_qp_attr *qp_attr, 3898 int *qp_attr_mask) 3899 { 3900 unsigned long flags; 3901 int ret; 3902 3903 spin_lock_irqsave(&cm_id_priv->lock, flags); 3904 switch (cm_id_priv->id.state) { 3905 /* Allow transition to RTS before sending REP */ 3906 case IB_CM_REQ_RCVD: 3907 case IB_CM_MRA_REQ_SENT: 3908 3909 case IB_CM_REP_RCVD: 3910 case IB_CM_MRA_REP_SENT: 3911 case IB_CM_REP_SENT: 3912 case IB_CM_MRA_REP_RCVD: 3913 case IB_CM_ESTABLISHED: 3914 if (cm_id_priv->id.lap_state == IB_CM_LAP_UNINIT) { 3915 *qp_attr_mask = IB_QP_STATE | IB_QP_SQ_PSN; 3916 qp_attr->sq_psn = be32_to_cpu(cm_id_priv->sq_psn); 3917 switch (cm_id_priv->qp_type) { 3918 case IB_QPT_RC: 3919 case IB_QPT_XRC_INI: 3920 *qp_attr_mask |= IB_QP_RETRY_CNT | IB_QP_RNR_RETRY | 3921 IB_QP_MAX_QP_RD_ATOMIC; 3922 qp_attr->retry_cnt = cm_id_priv->retry_count; 3923 qp_attr->rnr_retry = cm_id_priv->rnr_retry_count; 3924 qp_attr->max_rd_atomic = cm_id_priv->initiator_depth; 3925 /* fall through */ 3926 case IB_QPT_XRC_TGT: 3927 *qp_attr_mask |= IB_QP_TIMEOUT; 3928 qp_attr->timeout = cm_id_priv->av.timeout; 3929 break; 3930 default: 3931 break; 3932 } 3933 if (cm_id_priv->alt_av.ah_attr.dlid) { 3934 *qp_attr_mask |= IB_QP_PATH_MIG_STATE; 3935 qp_attr->path_mig_state = IB_MIG_REARM; 3936 } 3937 } else { 3938 *qp_attr_mask = IB_QP_ALT_PATH | IB_QP_PATH_MIG_STATE; 3939 qp_attr->alt_port_num = cm_id_priv->alt_av.port->port_num; 3940 qp_attr->alt_pkey_index = cm_id_priv->alt_av.pkey_index; 3941 qp_attr->alt_timeout = cm_id_priv->alt_av.timeout; 3942 qp_attr->alt_ah_attr = cm_id_priv->alt_av.ah_attr; 3943 qp_attr->path_mig_state = IB_MIG_REARM; 3944 } 3945 ret = 0; 3946 break; 3947 default: 3948 ret = -EINVAL; 3949 break; 3950 } 3951 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 3952 return ret; 3953 } 3954 3955 int ib_cm_init_qp_attr(struct ib_cm_id *cm_id, 3956 struct ib_qp_attr *qp_attr, 3957 int *qp_attr_mask) 3958 { 3959 struct cm_id_private *cm_id_priv; 3960 int ret; 3961 3962 cm_id_priv = container_of(cm_id, struct cm_id_private, id); 3963 switch (qp_attr->qp_state) { 3964 case IB_QPS_INIT: 3965 ret = cm_init_qp_init_attr(cm_id_priv, qp_attr, qp_attr_mask); 3966 break; 3967 case IB_QPS_RTR: 3968 ret = cm_init_qp_rtr_attr(cm_id_priv, qp_attr, qp_attr_mask); 3969 break; 3970 case IB_QPS_RTS: 3971 ret = cm_init_qp_rts_attr(cm_id_priv, qp_attr, qp_attr_mask); 3972 break; 3973 default: 3974 ret = -EINVAL; 3975 break; 3976 } 3977 return ret; 3978 } 3979 EXPORT_SYMBOL(ib_cm_init_qp_attr); 3980 3981 static ssize_t cm_show_counter(struct kobject *obj, struct attribute *attr, 3982 char *buf) 3983 { 3984 struct cm_counter_group *group; 3985 struct cm_counter_attribute *cm_attr; 3986 3987 group = container_of(obj, struct cm_counter_group, obj); 3988 cm_attr = container_of(attr, struct cm_counter_attribute, attr); 3989 3990 return sprintf(buf, "%ld\n", 3991 atomic_long_read(&group->counter[cm_attr->index])); 3992 } 3993 3994 static const struct sysfs_ops cm_counter_ops = { 3995 .show = cm_show_counter 3996 }; 3997 3998 static struct kobj_type cm_counter_obj_type = { 3999 .sysfs_ops = &cm_counter_ops, 4000 .default_attrs = cm_counter_default_attrs 4001 }; 4002 4003 static char *cm_devnode(struct device *dev, umode_t *mode) 4004 { 4005 if (mode) 4006 *mode = 0666; 4007 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev)); 4008 } 4009 4010 struct class cm_class = { 4011 .owner = THIS_MODULE, 4012 .name = "infiniband_cm", 4013 .devnode = cm_devnode, 4014 }; 4015 EXPORT_SYMBOL(cm_class); 4016 4017 static int cm_create_port_fs(struct cm_port *port) 4018 { 4019 int i, ret; 4020 4021 for (i = 0; i < CM_COUNTER_GROUPS; i++) { 4022 ret = ib_port_register_module_stat(port->cm_dev->ib_device, 4023 port->port_num, 4024 &port->counter_group[i].obj, 4025 &cm_counter_obj_type, 4026 counter_group_names[i]); 4027 if (ret) 4028 goto error; 4029 } 4030 4031 return 0; 4032 4033 error: 4034 while (i--) 4035 ib_port_unregister_module_stat(&port->counter_group[i].obj); 4036 return ret; 4037 4038 } 4039 4040 static void cm_remove_port_fs(struct cm_port *port) 4041 { 4042 int i; 4043 4044 for (i = 0; i < CM_COUNTER_GROUPS; i++) 4045 ib_port_unregister_module_stat(&port->counter_group[i].obj); 4046 4047 } 4048 4049 static void cm_add_one(struct ib_device *ib_device) 4050 { 4051 struct cm_device *cm_dev; 4052 struct cm_port *port; 4053 struct ib_mad_reg_req reg_req = { 4054 .mgmt_class = IB_MGMT_CLASS_CM, 4055 .mgmt_class_version = IB_CM_CLASS_VERSION, 4056 }; 4057 struct ib_port_modify port_modify = { 4058 .set_port_cap_mask = IB_PORT_CM_SUP 4059 }; 4060 unsigned long flags; 4061 int ret; 4062 int count = 0; 4063 u8 i; 4064 4065 cm_dev = kzalloc(sizeof(*cm_dev) + sizeof(*port) * 4066 ib_device->phys_port_cnt, GFP_KERNEL); 4067 if (!cm_dev) 4068 return; 4069 4070 cm_dev->ib_device = ib_device; 4071 cm_dev->ack_delay = ib_device->attrs.local_ca_ack_delay; 4072 cm_dev->going_down = 0; 4073 4074 set_bit(IB_MGMT_METHOD_SEND, reg_req.method_mask); 4075 for (i = 1; i <= ib_device->phys_port_cnt; i++) { 4076 if (!rdma_cap_ib_cm(ib_device, i)) 4077 continue; 4078 4079 port = kzalloc(sizeof *port, GFP_KERNEL); 4080 if (!port) 4081 goto error1; 4082 4083 cm_dev->port[i-1] = port; 4084 port->cm_dev = cm_dev; 4085 port->port_num = i; 4086 4087 INIT_LIST_HEAD(&port->cm_priv_prim_list); 4088 INIT_LIST_HEAD(&port->cm_priv_altr_list); 4089 4090 ret = cm_create_port_fs(port); 4091 if (ret) 4092 goto error1; 4093 4094 port->mad_agent = ib_register_mad_agent(ib_device, i, 4095 IB_QPT_GSI, 4096 ®_req, 4097 0, 4098 cm_send_handler, 4099 cm_recv_handler, 4100 port, 4101 0); 4102 if (IS_ERR(port->mad_agent)) 4103 goto error2; 4104 4105 ret = ib_modify_port(ib_device, i, 0, &port_modify); 4106 if (ret) 4107 goto error3; 4108 4109 count++; 4110 } 4111 4112 if (!count) 4113 goto free; 4114 4115 ib_set_client_data(ib_device, &cm_client, cm_dev); 4116 4117 write_lock_irqsave(&cm.device_lock, flags); 4118 list_add_tail(&cm_dev->list, &cm.device_list); 4119 write_unlock_irqrestore(&cm.device_lock, flags); 4120 return; 4121 4122 error3: 4123 ib_unregister_mad_agent(port->mad_agent); 4124 error2: 4125 cm_remove_port_fs(port); 4126 error1: 4127 port_modify.set_port_cap_mask = 0; 4128 port_modify.clr_port_cap_mask = IB_PORT_CM_SUP; 4129 kfree(port); 4130 while (--i) { 4131 if (!rdma_cap_ib_cm(ib_device, i)) 4132 continue; 4133 4134 port = cm_dev->port[i-1]; 4135 ib_modify_port(ib_device, port->port_num, 0, &port_modify); 4136 ib_unregister_mad_agent(port->mad_agent); 4137 cm_remove_port_fs(port); 4138 kfree(port); 4139 } 4140 free: 4141 kfree(cm_dev); 4142 } 4143 4144 static void cm_remove_one(struct ib_device *ib_device, void *client_data) 4145 { 4146 struct cm_device *cm_dev = client_data; 4147 struct cm_port *port; 4148 struct cm_id_private *cm_id_priv; 4149 struct ib_mad_agent *cur_mad_agent; 4150 struct ib_port_modify port_modify = { 4151 .clr_port_cap_mask = IB_PORT_CM_SUP 4152 }; 4153 unsigned long flags; 4154 int i; 4155 4156 if (!cm_dev) 4157 return; 4158 4159 write_lock_irqsave(&cm.device_lock, flags); 4160 list_del(&cm_dev->list); 4161 write_unlock_irqrestore(&cm.device_lock, flags); 4162 4163 spin_lock_irq(&cm.lock); 4164 cm_dev->going_down = 1; 4165 spin_unlock_irq(&cm.lock); 4166 4167 for (i = 1; i <= ib_device->phys_port_cnt; i++) { 4168 if (!rdma_cap_ib_cm(ib_device, i)) 4169 continue; 4170 4171 port = cm_dev->port[i-1]; 4172 ib_modify_port(ib_device, port->port_num, 0, &port_modify); 4173 /* Mark all the cm_id's as not valid */ 4174 spin_lock_irq(&cm.lock); 4175 list_for_each_entry(cm_id_priv, &port->cm_priv_altr_list, altr_list) 4176 cm_id_priv->altr_send_port_not_ready = 1; 4177 list_for_each_entry(cm_id_priv, &port->cm_priv_prim_list, prim_list) 4178 cm_id_priv->prim_send_port_not_ready = 1; 4179 spin_unlock_irq(&cm.lock); 4180 /* 4181 * We flush the queue here after the going_down set, this 4182 * verify that no new works will be queued in the recv handler, 4183 * after that we can call the unregister_mad_agent 4184 */ 4185 flush_workqueue(cm.wq); 4186 spin_lock_irq(&cm.state_lock); 4187 cur_mad_agent = port->mad_agent; 4188 port->mad_agent = NULL; 4189 spin_unlock_irq(&cm.state_lock); 4190 ib_unregister_mad_agent(cur_mad_agent); 4191 cm_remove_port_fs(port); 4192 kfree(port); 4193 } 4194 4195 kfree(cm_dev); 4196 } 4197 4198 static int __init ib_cm_init(void) 4199 { 4200 int ret; 4201 4202 memset(&cm, 0, sizeof cm); 4203 INIT_LIST_HEAD(&cm.device_list); 4204 rwlock_init(&cm.device_lock); 4205 spin_lock_init(&cm.lock); 4206 spin_lock_init(&cm.state_lock); 4207 cm.listen_service_table = RB_ROOT; 4208 cm.listen_service_id = be64_to_cpu(IB_CM_ASSIGN_SERVICE_ID); 4209 cm.remote_id_table = RB_ROOT; 4210 cm.remote_qp_table = RB_ROOT; 4211 cm.remote_sidr_table = RB_ROOT; 4212 idr_init(&cm.local_id_table); 4213 get_random_bytes(&cm.random_id_operand, sizeof cm.random_id_operand); 4214 INIT_LIST_HEAD(&cm.timewait_list); 4215 4216 ret = class_register(&cm_class); 4217 if (ret) { 4218 ret = -ENOMEM; 4219 goto error1; 4220 } 4221 4222 cm.wq = create_workqueue("ib_cm"); 4223 if (!cm.wq) { 4224 ret = -ENOMEM; 4225 goto error2; 4226 } 4227 4228 ret = ib_register_client(&cm_client); 4229 if (ret) 4230 goto error3; 4231 4232 return 0; 4233 error3: 4234 destroy_workqueue(cm.wq); 4235 error2: 4236 class_unregister(&cm_class); 4237 error1: 4238 idr_destroy(&cm.local_id_table); 4239 return ret; 4240 } 4241 4242 static void __exit ib_cm_cleanup(void) 4243 { 4244 struct cm_timewait_info *timewait_info, *tmp; 4245 4246 spin_lock_irq(&cm.lock); 4247 list_for_each_entry(timewait_info, &cm.timewait_list, list) 4248 cancel_delayed_work(&timewait_info->work.work); 4249 spin_unlock_irq(&cm.lock); 4250 4251 ib_unregister_client(&cm_client); 4252 destroy_workqueue(cm.wq); 4253 4254 list_for_each_entry_safe(timewait_info, tmp, &cm.timewait_list, list) { 4255 cancel_delayed_work_sync(&timewait_info->work.work); 4256 list_del(&timewait_info->list); 4257 kfree(timewait_info); 4258 } 4259 4260 class_unregister(&cm_class); 4261 idr_destroy(&cm.local_id_table); 4262 } 4263 4264 module_init_order(ib_cm_init, SI_ORDER_SECOND); 4265 module_exit_order(ib_cm_cleanup, SI_ORDER_SECOND); 4266 4267