1/* 2 * Copyright (c) 2004, 2005 Intel Corporation. All rights reserved. 3 * Copyright (c) 2004 Topspin Corporation. All rights reserved. 4 * Copyright (c) 2004, 2005 Voltaire Corporation. All rights reserved. 5 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 6 * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. 7 * Copyright (c) 2005 Network Appliance, 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 <linux/dma-mapping.h> 39#include <linux/err.h> 40#include <linux/idr.h> 41#include <linux/interrupt.h> 42#include <linux/rbtree.h> 43#include <linux/sched.h> 44#include <linux/spinlock.h> 45#include <linux/workqueue.h> 46#include <linux/completion.h> 47#include <linux/slab.h> 48#include <linux/module.h> 49#include <linux/sysctl.h> 50 51#include <rdma/iw_cm.h> 52#include <rdma/ib_addr.h> 53#include <rdma/iw_portmap.h> 54#include <rdma/rdma_netlink.h> 55 56#include "iwcm.h" 57 58MODULE_AUTHOR("Tom Tucker"); 59MODULE_DESCRIPTION("iWARP CM"); 60MODULE_LICENSE("Dual BSD/GPL"); 61 62static const char * const iwcm_rej_reason_strs[] = { 63 [ECONNRESET] = "reset by remote host", 64 [ECONNREFUSED] = "refused by remote application", 65 [ETIMEDOUT] = "setup timeout", 66}; 67 68const char *__attribute_const__ iwcm_reject_msg(int reason) 69{ 70 size_t index; 71 72 /* iWARP uses negative errnos */ 73 index = -reason; 74 75 if (index < ARRAY_SIZE(iwcm_rej_reason_strs) && 76 iwcm_rej_reason_strs[index]) 77 return iwcm_rej_reason_strs[index]; 78 else 79 return "unrecognized reason"; 80} 81EXPORT_SYMBOL(iwcm_reject_msg); 82 83static struct rdma_nl_cbs iwcm_nl_cb_table[RDMA_NL_IWPM_NUM_OPS] = { 84 [RDMA_NL_IWPM_REG_PID] = {.dump = iwpm_register_pid_cb}, 85 [RDMA_NL_IWPM_ADD_MAPPING] = {.dump = iwpm_add_mapping_cb}, 86 [RDMA_NL_IWPM_QUERY_MAPPING] = {.dump = iwpm_add_and_query_mapping_cb}, 87 [RDMA_NL_IWPM_REMOTE_INFO] = {.dump = iwpm_remote_info_cb}, 88 [RDMA_NL_IWPM_HANDLE_ERR] = {.dump = iwpm_mapping_error_cb}, 89 [RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb}, 90 [RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}, 91 [RDMA_NL_IWPM_HELLO] = {.dump = iwpm_hello_cb} 92}; 93 94static struct workqueue_struct *iwcm_wq; 95struct iwcm_work { 96 struct work_struct work; 97 struct iwcm_id_private *cm_id; 98 struct list_head list; 99 struct iw_cm_event event; 100 struct list_head free_list; 101}; 102 103static unsigned int default_backlog = 256; 104 105static struct ctl_table_header *iwcm_ctl_table_hdr; 106static struct ctl_table iwcm_ctl_table[] = { 107 { 108 .procname = "default_backlog", 109 .data = &default_backlog, 110 .maxlen = sizeof(default_backlog), 111 .mode = 0644, 112 .proc_handler = proc_dointvec, 113 }, 114 { } 115}; 116 117/* 118 * The following services provide a mechanism for pre-allocating iwcm_work 119 * elements. The design pre-allocates them based on the cm_id type: 120 * LISTENING IDS: Get enough elements preallocated to handle the 121 * listen backlog. 122 * ACTIVE IDS: 4: CONNECT_REPLY, ESTABLISHED, DISCONNECT, CLOSE 123 * PASSIVE IDS: 3: ESTABLISHED, DISCONNECT, CLOSE 124 * 125 * Allocating them in connect and listen avoids having to deal 126 * with allocation failures on the event upcall from the provider (which 127 * is called in the interrupt context). 128 * 129 * One exception is when creating the cm_id for incoming connection requests. 130 * There are two cases: 131 * 1) in the event upcall, cm_event_handler(), for a listening cm_id. If 132 * the backlog is exceeded, then no more connection request events will 133 * be processed. cm_event_handler() returns -ENOMEM in this case. Its up 134 * to the provider to reject the connection request. 135 * 2) in the connection request workqueue handler, cm_conn_req_handler(). 136 * If work elements cannot be allocated for the new connect request cm_id, 137 * then IWCM will call the provider reject method. This is ok since 138 * cm_conn_req_handler() runs in the workqueue thread context. 139 */ 140 141static struct iwcm_work *get_work(struct iwcm_id_private *cm_id_priv) 142{ 143 struct iwcm_work *work; 144 145 if (list_empty(&cm_id_priv->work_free_list)) 146 return NULL; 147 work = list_entry(cm_id_priv->work_free_list.next, struct iwcm_work, 148 free_list); 149 list_del_init(&work->free_list); 150 return work; 151} 152 153static void put_work(struct iwcm_work *work) 154{ 155 list_add(&work->free_list, &work->cm_id->work_free_list); 156} 157 158static void dealloc_work_entries(struct iwcm_id_private *cm_id_priv) 159{ 160 struct list_head *e, *tmp; 161 162 list_for_each_safe(e, tmp, &cm_id_priv->work_free_list) { 163 list_del(e); 164 kfree(list_entry(e, struct iwcm_work, free_list)); 165 } 166} 167 168static int alloc_work_entries(struct iwcm_id_private *cm_id_priv, int count) 169{ 170 struct iwcm_work *work; 171 172 BUG_ON(!list_empty(&cm_id_priv->work_free_list)); 173 while (count--) { 174 work = kmalloc(sizeof(struct iwcm_work), GFP_KERNEL); 175 if (!work) { 176 dealloc_work_entries(cm_id_priv); 177 return -ENOMEM; 178 } 179 work->cm_id = cm_id_priv; 180 INIT_LIST_HEAD(&work->list); 181 put_work(work); 182 } 183 return 0; 184} 185 186/* 187 * Save private data from incoming connection requests to 188 * iw_cm_event, so the low level driver doesn't have to. Adjust 189 * the event ptr to point to the local copy. 190 */ 191static int copy_private_data(struct iw_cm_event *event) 192{ 193 void *p; 194 195 p = kmemdup(event->private_data, event->private_data_len, GFP_ATOMIC); 196 if (!p) 197 return -ENOMEM; 198 event->private_data = p; 199 return 0; 200} 201 202static void free_cm_id(struct iwcm_id_private *cm_id_priv) 203{ 204 dealloc_work_entries(cm_id_priv); 205 kfree(cm_id_priv); 206} 207 208/* 209 * Release a reference on cm_id. If the last reference is being 210 * released, free the cm_id and return 1. 211 */ 212static int iwcm_deref_id(struct iwcm_id_private *cm_id_priv) 213{ 214 if (refcount_dec_and_test(&cm_id_priv->refcount)) { 215 BUG_ON(!list_empty(&cm_id_priv->work_list)); 216 free_cm_id(cm_id_priv); 217 return 1; 218 } 219 220 return 0; 221} 222 223static void add_ref(struct iw_cm_id *cm_id) 224{ 225 struct iwcm_id_private *cm_id_priv; 226 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 227 refcount_inc(&cm_id_priv->refcount); 228} 229 230static void rem_ref(struct iw_cm_id *cm_id) 231{ 232 struct iwcm_id_private *cm_id_priv; 233 234 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 235 236 (void)iwcm_deref_id(cm_id_priv); 237} 238 239static int cm_event_handler(struct iw_cm_id *cm_id, struct iw_cm_event *event); 240 241struct iw_cm_id *iw_create_cm_id(struct ib_device *device, 242 iw_cm_handler cm_handler, 243 void *context) 244{ 245 struct iwcm_id_private *cm_id_priv; 246 247 cm_id_priv = kzalloc(sizeof(*cm_id_priv), GFP_KERNEL); 248 if (!cm_id_priv) 249 return ERR_PTR(-ENOMEM); 250 251 cm_id_priv->state = IW_CM_STATE_IDLE; 252 cm_id_priv->id.device = device; 253 cm_id_priv->id.cm_handler = cm_handler; 254 cm_id_priv->id.context = context; 255 cm_id_priv->id.event_handler = cm_event_handler; 256 cm_id_priv->id.add_ref = add_ref; 257 cm_id_priv->id.rem_ref = rem_ref; 258 spin_lock_init(&cm_id_priv->lock); 259 refcount_set(&cm_id_priv->refcount, 1); 260 init_waitqueue_head(&cm_id_priv->connect_wait); 261 init_completion(&cm_id_priv->destroy_comp); 262 INIT_LIST_HEAD(&cm_id_priv->work_list); 263 INIT_LIST_HEAD(&cm_id_priv->work_free_list); 264 265 return &cm_id_priv->id; 266} 267EXPORT_SYMBOL(iw_create_cm_id); 268 269 270static int iwcm_modify_qp_err(struct ib_qp *qp) 271{ 272 struct ib_qp_attr qp_attr; 273 274 if (!qp) 275 return -EINVAL; 276 277 qp_attr.qp_state = IB_QPS_ERR; 278 return ib_modify_qp(qp, &qp_attr, IB_QP_STATE); 279} 280 281/* 282 * This is really the RDMAC CLOSING state. It is most similar to the 283 * IB SQD QP state. 284 */ 285static int iwcm_modify_qp_sqd(struct ib_qp *qp) 286{ 287 struct ib_qp_attr qp_attr; 288 289 BUG_ON(qp == NULL); 290 qp_attr.qp_state = IB_QPS_SQD; 291 return ib_modify_qp(qp, &qp_attr, IB_QP_STATE); 292} 293 294/* 295 * CM_ID <-- CLOSING 296 * 297 * Block if a passive or active connection is currently being processed. Then 298 * process the event as follows: 299 * - If we are ESTABLISHED, move to CLOSING and modify the QP state 300 * based on the abrupt flag 301 * - If the connection is already in the CLOSING or IDLE state, the peer is 302 * disconnecting concurrently with us and we've already seen the 303 * DISCONNECT event -- ignore the request and return 0 304 * - Disconnect on a listening endpoint returns -EINVAL 305 */ 306int iw_cm_disconnect(struct iw_cm_id *cm_id, int abrupt) 307{ 308 struct iwcm_id_private *cm_id_priv; 309 unsigned long flags; 310 int ret = 0; 311 struct ib_qp *qp = NULL; 312 313 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 314 /* Wait if we're currently in a connect or accept downcall */ 315 wait_event(cm_id_priv->connect_wait, 316 !test_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags)); 317 318 spin_lock_irqsave(&cm_id_priv->lock, flags); 319 switch (cm_id_priv->state) { 320 case IW_CM_STATE_ESTABLISHED: 321 cm_id_priv->state = IW_CM_STATE_CLOSING; 322 323 /* QP could be <nul> for user-mode client */ 324 if (cm_id_priv->qp) 325 qp = cm_id_priv->qp; 326 else 327 ret = -EINVAL; 328 break; 329 case IW_CM_STATE_LISTEN: 330 ret = -EINVAL; 331 break; 332 case IW_CM_STATE_CLOSING: 333 /* remote peer closed first */ 334 case IW_CM_STATE_IDLE: 335 /* accept or connect returned !0 */ 336 break; 337 case IW_CM_STATE_CONN_RECV: 338 /* 339 * App called disconnect before/without calling accept after 340 * connect_request event delivered. 341 */ 342 break; 343 case IW_CM_STATE_CONN_SENT: 344 /* Can only get here if wait above fails */ 345 default: 346 BUG(); 347 } 348 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 349 350 if (qp) { 351 if (abrupt) 352 ret = iwcm_modify_qp_err(qp); 353 else 354 ret = iwcm_modify_qp_sqd(qp); 355 356 /* 357 * If both sides are disconnecting the QP could 358 * already be in ERR or SQD states 359 */ 360 ret = 0; 361 } 362 363 return ret; 364} 365EXPORT_SYMBOL(iw_cm_disconnect); 366 367/* 368 * CM_ID <-- DESTROYING 369 * 370 * Clean up all resources associated with the connection and release 371 * the initial reference taken by iw_create_cm_id. 372 */ 373static void destroy_cm_id(struct iw_cm_id *cm_id) 374{ 375 struct iwcm_id_private *cm_id_priv; 376 struct ib_qp *qp; 377 unsigned long flags; 378 379 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 380 /* 381 * Wait if we're currently in a connect or accept downcall. A 382 * listening endpoint should never block here. 383 */ 384 wait_event(cm_id_priv->connect_wait, 385 !test_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags)); 386 387 /* 388 * Since we're deleting the cm_id, drop any events that 389 * might arrive before the last dereference. 390 */ 391 set_bit(IWCM_F_DROP_EVENTS, &cm_id_priv->flags); 392 393 spin_lock_irqsave(&cm_id_priv->lock, flags); 394 qp = cm_id_priv->qp; 395 cm_id_priv->qp = NULL; 396 397 switch (cm_id_priv->state) { 398 case IW_CM_STATE_LISTEN: 399 cm_id_priv->state = IW_CM_STATE_DESTROYING; 400 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 401 /* destroy the listening endpoint */ 402 cm_id->device->ops.iw_destroy_listen(cm_id); 403 spin_lock_irqsave(&cm_id_priv->lock, flags); 404 break; 405 case IW_CM_STATE_ESTABLISHED: 406 cm_id_priv->state = IW_CM_STATE_DESTROYING; 407 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 408 /* Abrupt close of the connection */ 409 (void)iwcm_modify_qp_err(qp); 410 spin_lock_irqsave(&cm_id_priv->lock, flags); 411 break; 412 case IW_CM_STATE_IDLE: 413 case IW_CM_STATE_CLOSING: 414 cm_id_priv->state = IW_CM_STATE_DESTROYING; 415 break; 416 case IW_CM_STATE_CONN_RECV: 417 /* 418 * App called destroy before/without calling accept after 419 * receiving connection request event notification or 420 * returned non zero from the event callback function. 421 * In either case, must tell the provider to reject. 422 */ 423 cm_id_priv->state = IW_CM_STATE_DESTROYING; 424 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 425 cm_id->device->ops.iw_reject(cm_id, NULL, 0); 426 spin_lock_irqsave(&cm_id_priv->lock, flags); 427 break; 428 case IW_CM_STATE_CONN_SENT: 429 case IW_CM_STATE_DESTROYING: 430 default: 431 BUG(); 432 break; 433 } 434 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 435 if (qp) 436 cm_id_priv->id.device->ops.iw_rem_ref(qp); 437 438 if (cm_id->mapped) { 439 iwpm_remove_mapinfo(&cm_id->local_addr, &cm_id->m_local_addr); 440 iwpm_remove_mapping(&cm_id->local_addr, RDMA_NL_IWCM); 441 } 442 443 (void)iwcm_deref_id(cm_id_priv); 444} 445 446/* 447 * This function is only called by the application thread and cannot 448 * be called by the event thread. The function will wait for all 449 * references to be released on the cm_id and then kfree the cm_id 450 * object. 451 */ 452void iw_destroy_cm_id(struct iw_cm_id *cm_id) 453{ 454 destroy_cm_id(cm_id); 455} 456EXPORT_SYMBOL(iw_destroy_cm_id); 457 458/** 459 * iw_cm_check_wildcard - If IP address is 0 then use original 460 * @pm_addr: sockaddr containing the ip to check for wildcard 461 * @cm_addr: sockaddr containing the actual IP address 462 * @cm_outaddr: sockaddr to set IP addr which leaving port 463 * 464 * Checks the pm_addr for wildcard and then sets cm_outaddr's 465 * IP to the actual (cm_addr). 466 */ 467static void iw_cm_check_wildcard(struct sockaddr_storage *pm_addr, 468 struct sockaddr_storage *cm_addr, 469 struct sockaddr_storage *cm_outaddr) 470{ 471 if (pm_addr->ss_family == AF_INET) { 472 struct sockaddr_in *pm4_addr = (struct sockaddr_in *)pm_addr; 473 474 if (pm4_addr->sin_addr.s_addr == htonl(INADDR_ANY)) { 475 struct sockaddr_in *cm4_addr = 476 (struct sockaddr_in *)cm_addr; 477 struct sockaddr_in *cm4_outaddr = 478 (struct sockaddr_in *)cm_outaddr; 479 480 cm4_outaddr->sin_addr = cm4_addr->sin_addr; 481 } 482 } else { 483 struct sockaddr_in6 *pm6_addr = (struct sockaddr_in6 *)pm_addr; 484 485 if (ipv6_addr_type(&pm6_addr->sin6_addr) == IPV6_ADDR_ANY) { 486 struct sockaddr_in6 *cm6_addr = 487 (struct sockaddr_in6 *)cm_addr; 488 struct sockaddr_in6 *cm6_outaddr = 489 (struct sockaddr_in6 *)cm_outaddr; 490 491 cm6_outaddr->sin6_addr = cm6_addr->sin6_addr; 492 } 493 } 494} 495 496/** 497 * iw_cm_map - Use portmapper to map the ports 498 * @cm_id: connection manager pointer 499 * @active: Indicates the active side when true 500 * returns nonzero for error only if iwpm_create_mapinfo() fails 501 * 502 * Tries to add a mapping for a port using the Portmapper. If 503 * successful in mapping the IP/Port it will check the remote 504 * mapped IP address for a wildcard IP address and replace the 505 * zero IP address with the remote_addr. 506 */ 507static int iw_cm_map(struct iw_cm_id *cm_id, bool active) 508{ 509 const char *devname = dev_name(&cm_id->device->dev); 510 const char *ifname = cm_id->device->iw_ifname; 511 struct iwpm_dev_data pm_reg_msg = {}; 512 struct iwpm_sa_data pm_msg; 513 int status; 514 515 if (strlen(devname) >= sizeof(pm_reg_msg.dev_name) || 516 strlen(ifname) >= sizeof(pm_reg_msg.if_name)) 517 return -EINVAL; 518 519 cm_id->m_local_addr = cm_id->local_addr; 520 cm_id->m_remote_addr = cm_id->remote_addr; 521 522 strcpy(pm_reg_msg.dev_name, devname); 523 strcpy(pm_reg_msg.if_name, ifname); 524 525 if (iwpm_register_pid(&pm_reg_msg, RDMA_NL_IWCM) || 526 !iwpm_valid_pid()) 527 return 0; 528 529 cm_id->mapped = true; 530 pm_msg.loc_addr = cm_id->local_addr; 531 pm_msg.rem_addr = cm_id->remote_addr; 532 pm_msg.flags = (cm_id->device->iw_driver_flags & IW_F_NO_PORT_MAP) ? 533 IWPM_FLAGS_NO_PORT_MAP : 0; 534 if (active) 535 status = iwpm_add_and_query_mapping(&pm_msg, 536 RDMA_NL_IWCM); 537 else 538 status = iwpm_add_mapping(&pm_msg, RDMA_NL_IWCM); 539 540 if (!status) { 541 cm_id->m_local_addr = pm_msg.mapped_loc_addr; 542 if (active) { 543 cm_id->m_remote_addr = pm_msg.mapped_rem_addr; 544 iw_cm_check_wildcard(&pm_msg.mapped_rem_addr, 545 &cm_id->remote_addr, 546 &cm_id->m_remote_addr); 547 } 548 } 549 550 return iwpm_create_mapinfo(&cm_id->local_addr, 551 &cm_id->m_local_addr, 552 RDMA_NL_IWCM, pm_msg.flags); 553} 554 555/* 556 * CM_ID <-- LISTEN 557 * 558 * Start listening for connect requests. Generates one CONNECT_REQUEST 559 * event for each inbound connect request. 560 */ 561int iw_cm_listen(struct iw_cm_id *cm_id, int backlog) 562{ 563 struct iwcm_id_private *cm_id_priv; 564 unsigned long flags; 565 int ret; 566 567 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 568 569 if (!backlog) 570 backlog = default_backlog; 571 572 ret = alloc_work_entries(cm_id_priv, backlog); 573 if (ret) 574 return ret; 575 576 spin_lock_irqsave(&cm_id_priv->lock, flags); 577 switch (cm_id_priv->state) { 578 case IW_CM_STATE_IDLE: 579 cm_id_priv->state = IW_CM_STATE_LISTEN; 580 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 581 ret = iw_cm_map(cm_id, false); 582 if (!ret) 583 ret = cm_id->device->ops.iw_create_listen(cm_id, 584 backlog); 585 if (ret) 586 cm_id_priv->state = IW_CM_STATE_IDLE; 587 spin_lock_irqsave(&cm_id_priv->lock, flags); 588 break; 589 default: 590 ret = -EINVAL; 591 } 592 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 593 594 return ret; 595} 596EXPORT_SYMBOL(iw_cm_listen); 597 598/* 599 * CM_ID <-- IDLE 600 * 601 * Rejects an inbound connection request. No events are generated. 602 */ 603int iw_cm_reject(struct iw_cm_id *cm_id, 604 const void *private_data, 605 u8 private_data_len) 606{ 607 struct iwcm_id_private *cm_id_priv; 608 unsigned long flags; 609 int ret; 610 611 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 612 set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 613 614 spin_lock_irqsave(&cm_id_priv->lock, flags); 615 if (cm_id_priv->state != IW_CM_STATE_CONN_RECV) { 616 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 617 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 618 wake_up_all(&cm_id_priv->connect_wait); 619 return -EINVAL; 620 } 621 cm_id_priv->state = IW_CM_STATE_IDLE; 622 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 623 624 ret = cm_id->device->ops.iw_reject(cm_id, private_data, 625 private_data_len); 626 627 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 628 wake_up_all(&cm_id_priv->connect_wait); 629 630 return ret; 631} 632EXPORT_SYMBOL(iw_cm_reject); 633 634/* 635 * CM_ID <-- ESTABLISHED 636 * 637 * Accepts an inbound connection request and generates an ESTABLISHED 638 * event. Callers of iw_cm_disconnect and iw_destroy_cm_id will block 639 * until the ESTABLISHED event is received from the provider. 640 */ 641int iw_cm_accept(struct iw_cm_id *cm_id, 642 struct iw_cm_conn_param *iw_param) 643{ 644 struct iwcm_id_private *cm_id_priv; 645 struct ib_qp *qp; 646 unsigned long flags; 647 int ret; 648 649 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 650 set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 651 652 spin_lock_irqsave(&cm_id_priv->lock, flags); 653 if (cm_id_priv->state != IW_CM_STATE_CONN_RECV) { 654 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 655 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 656 wake_up_all(&cm_id_priv->connect_wait); 657 return -EINVAL; 658 } 659 /* Get the ib_qp given the QPN */ 660 qp = cm_id->device->ops.iw_get_qp(cm_id->device, iw_param->qpn); 661 if (!qp) { 662 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 663 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 664 wake_up_all(&cm_id_priv->connect_wait); 665 return -EINVAL; 666 } 667 cm_id->device->ops.iw_add_ref(qp); 668 cm_id_priv->qp = qp; 669 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 670 671 ret = cm_id->device->ops.iw_accept(cm_id, iw_param); 672 if (ret) { 673 /* An error on accept precludes provider events */ 674 BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_RECV); 675 cm_id_priv->state = IW_CM_STATE_IDLE; 676 spin_lock_irqsave(&cm_id_priv->lock, flags); 677 qp = cm_id_priv->qp; 678 cm_id_priv->qp = NULL; 679 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 680 if (qp) 681 cm_id->device->ops.iw_rem_ref(qp); 682 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 683 wake_up_all(&cm_id_priv->connect_wait); 684 } 685 686 return ret; 687} 688EXPORT_SYMBOL(iw_cm_accept); 689 690/* 691 * Active Side: CM_ID <-- CONN_SENT 692 * 693 * If successful, results in the generation of a CONNECT_REPLY 694 * event. iw_cm_disconnect and iw_cm_destroy will block until the 695 * CONNECT_REPLY event is received from the provider. 696 */ 697int iw_cm_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param) 698{ 699 struct iwcm_id_private *cm_id_priv; 700 int ret; 701 unsigned long flags; 702 struct ib_qp *qp = NULL; 703 704 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 705 706 ret = alloc_work_entries(cm_id_priv, 4); 707 if (ret) 708 return ret; 709 710 set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 711 spin_lock_irqsave(&cm_id_priv->lock, flags); 712 713 if (cm_id_priv->state != IW_CM_STATE_IDLE) { 714 ret = -EINVAL; 715 goto err; 716 } 717 718 /* Get the ib_qp given the QPN */ 719 qp = cm_id->device->ops.iw_get_qp(cm_id->device, iw_param->qpn); 720 if (!qp) { 721 ret = -EINVAL; 722 goto err; 723 } 724 cm_id->device->ops.iw_add_ref(qp); 725 cm_id_priv->qp = qp; 726 cm_id_priv->state = IW_CM_STATE_CONN_SENT; 727 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 728 729 ret = iw_cm_map(cm_id, true); 730 if (!ret) 731 ret = cm_id->device->ops.iw_connect(cm_id, iw_param); 732 if (!ret) 733 return 0; /* success */ 734 735 spin_lock_irqsave(&cm_id_priv->lock, flags); 736 qp = cm_id_priv->qp; 737 cm_id_priv->qp = NULL; 738 cm_id_priv->state = IW_CM_STATE_IDLE; 739err: 740 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 741 if (qp) 742 cm_id->device->ops.iw_rem_ref(qp); 743 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 744 wake_up_all(&cm_id_priv->connect_wait); 745 return ret; 746} 747EXPORT_SYMBOL(iw_cm_connect); 748 749/* 750 * Passive Side: new CM_ID <-- CONN_RECV 751 * 752 * Handles an inbound connect request. The function creates a new 753 * iw_cm_id to represent the new connection and inherits the client 754 * callback function and other attributes from the listening parent. 755 * 756 * The work item contains a pointer to the listen_cm_id and the event. The 757 * listen_cm_id contains the client cm_handler, context and 758 * device. These are copied when the device is cloned. The event 759 * contains the new four tuple. 760 * 761 * An error on the child should not affect the parent, so this 762 * function does not return a value. 763 */ 764static void cm_conn_req_handler(struct iwcm_id_private *listen_id_priv, 765 struct iw_cm_event *iw_event) 766{ 767 unsigned long flags; 768 struct iw_cm_id *cm_id; 769 struct iwcm_id_private *cm_id_priv; 770 int ret; 771 772 /* 773 * The provider should never generate a connection request 774 * event with a bad status. 775 */ 776 BUG_ON(iw_event->status); 777 778 cm_id = iw_create_cm_id(listen_id_priv->id.device, 779 listen_id_priv->id.cm_handler, 780 listen_id_priv->id.context); 781 /* If the cm_id could not be created, ignore the request */ 782 if (IS_ERR(cm_id)) 783 goto out; 784 785 cm_id->provider_data = iw_event->provider_data; 786 cm_id->m_local_addr = iw_event->local_addr; 787 cm_id->m_remote_addr = iw_event->remote_addr; 788 cm_id->local_addr = listen_id_priv->id.local_addr; 789 790 ret = iwpm_get_remote_info(&listen_id_priv->id.m_local_addr, 791 &iw_event->remote_addr, 792 &cm_id->remote_addr, 793 RDMA_NL_IWCM); 794 if (ret) { 795 cm_id->remote_addr = iw_event->remote_addr; 796 } else { 797 iw_cm_check_wildcard(&listen_id_priv->id.m_local_addr, 798 &iw_event->local_addr, 799 &cm_id->local_addr); 800 iw_event->local_addr = cm_id->local_addr; 801 iw_event->remote_addr = cm_id->remote_addr; 802 } 803 804 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 805 cm_id_priv->state = IW_CM_STATE_CONN_RECV; 806 807 /* 808 * We could be destroying the listening id. If so, ignore this 809 * upcall. 810 */ 811 spin_lock_irqsave(&listen_id_priv->lock, flags); 812 if (listen_id_priv->state != IW_CM_STATE_LISTEN) { 813 spin_unlock_irqrestore(&listen_id_priv->lock, flags); 814 iw_cm_reject(cm_id, NULL, 0); 815 iw_destroy_cm_id(cm_id); 816 goto out; 817 } 818 spin_unlock_irqrestore(&listen_id_priv->lock, flags); 819 820 ret = alloc_work_entries(cm_id_priv, 3); 821 if (ret) { 822 iw_cm_reject(cm_id, NULL, 0); 823 iw_destroy_cm_id(cm_id); 824 goto out; 825 } 826 827 /* Call the client CM handler */ 828 ret = cm_id->cm_handler(cm_id, iw_event); 829 if (ret) { 830 iw_cm_reject(cm_id, NULL, 0); 831 iw_destroy_cm_id(cm_id); 832 } 833 834out: 835 if (iw_event->private_data_len) 836 kfree(iw_event->private_data); 837} 838 839/* 840 * Passive Side: CM_ID <-- ESTABLISHED 841 * 842 * The provider generated an ESTABLISHED event which means that 843 * the MPA negotion has completed successfully and we are now in MPA 844 * FPDU mode. 845 * 846 * This event can only be received in the CONN_RECV state. If the 847 * remote peer closed, the ESTABLISHED event would be received followed 848 * by the CLOSE event. If the app closes, it will block until we wake 849 * it up after processing this event. 850 */ 851static int cm_conn_est_handler(struct iwcm_id_private *cm_id_priv, 852 struct iw_cm_event *iw_event) 853{ 854 unsigned long flags; 855 int ret; 856 857 spin_lock_irqsave(&cm_id_priv->lock, flags); 858 859 /* 860 * We clear the CONNECT_WAIT bit here to allow the callback 861 * function to call iw_cm_disconnect. Calling iw_destroy_cm_id 862 * from a callback handler is not allowed. 863 */ 864 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 865 BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_RECV); 866 cm_id_priv->state = IW_CM_STATE_ESTABLISHED; 867 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 868 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event); 869 wake_up_all(&cm_id_priv->connect_wait); 870 871 return ret; 872} 873 874/* 875 * Active Side: CM_ID <-- ESTABLISHED 876 * 877 * The app has called connect and is waiting for the established event to 878 * post it's requests to the server. This event will wake up anyone 879 * blocked in iw_cm_disconnect or iw_destroy_id. 880 */ 881static int cm_conn_rep_handler(struct iwcm_id_private *cm_id_priv, 882 struct iw_cm_event *iw_event) 883{ 884 struct ib_qp *qp = NULL; 885 unsigned long flags; 886 int ret; 887 888 spin_lock_irqsave(&cm_id_priv->lock, flags); 889 /* 890 * Clear the connect wait bit so a callback function calling 891 * iw_cm_disconnect will not wait and deadlock this thread 892 */ 893 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); 894 BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_SENT); 895 if (iw_event->status == 0) { 896 cm_id_priv->id.m_local_addr = iw_event->local_addr; 897 cm_id_priv->id.m_remote_addr = iw_event->remote_addr; 898 iw_event->local_addr = cm_id_priv->id.local_addr; 899 iw_event->remote_addr = cm_id_priv->id.remote_addr; 900 cm_id_priv->state = IW_CM_STATE_ESTABLISHED; 901 } else { 902 /* REJECTED or RESET */ 903 qp = cm_id_priv->qp; 904 cm_id_priv->qp = NULL; 905 cm_id_priv->state = IW_CM_STATE_IDLE; 906 } 907 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 908 if (qp) 909 cm_id_priv->id.device->ops.iw_rem_ref(qp); 910 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event); 911 912 if (iw_event->private_data_len) 913 kfree(iw_event->private_data); 914 915 /* Wake up waiters on connect complete */ 916 wake_up_all(&cm_id_priv->connect_wait); 917 918 return ret; 919} 920 921/* 922 * CM_ID <-- CLOSING 923 * 924 * If in the ESTABLISHED state, move to CLOSING. 925 */ 926static void cm_disconnect_handler(struct iwcm_id_private *cm_id_priv, 927 struct iw_cm_event *iw_event) 928{ 929 unsigned long flags; 930 931 spin_lock_irqsave(&cm_id_priv->lock, flags); 932 if (cm_id_priv->state == IW_CM_STATE_ESTABLISHED) 933 cm_id_priv->state = IW_CM_STATE_CLOSING; 934 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 935} 936 937/* 938 * CM_ID <-- IDLE 939 * 940 * If in the ESTBLISHED or CLOSING states, the QP will have have been 941 * moved by the provider to the ERR state. Disassociate the CM_ID from 942 * the QP, move to IDLE, and remove the 'connected' reference. 943 * 944 * If in some other state, the cm_id was destroyed asynchronously. 945 * This is the last reference that will result in waking up 946 * the app thread blocked in iw_destroy_cm_id. 947 */ 948static int cm_close_handler(struct iwcm_id_private *cm_id_priv, 949 struct iw_cm_event *iw_event) 950{ 951 struct ib_qp *qp; 952 unsigned long flags; 953 int ret = 0, notify_event = 0; 954 spin_lock_irqsave(&cm_id_priv->lock, flags); 955 qp = cm_id_priv->qp; 956 cm_id_priv->qp = NULL; 957 958 switch (cm_id_priv->state) { 959 case IW_CM_STATE_ESTABLISHED: 960 case IW_CM_STATE_CLOSING: 961 cm_id_priv->state = IW_CM_STATE_IDLE; 962 notify_event = 1; 963 break; 964 case IW_CM_STATE_DESTROYING: 965 break; 966 default: 967 BUG(); 968 } 969 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 970 971 if (qp) 972 cm_id_priv->id.device->ops.iw_rem_ref(qp); 973 if (notify_event) 974 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event); 975 return ret; 976} 977 978static int process_event(struct iwcm_id_private *cm_id_priv, 979 struct iw_cm_event *iw_event) 980{ 981 int ret = 0; 982 983 switch (iw_event->event) { 984 case IW_CM_EVENT_CONNECT_REQUEST: 985 cm_conn_req_handler(cm_id_priv, iw_event); 986 break; 987 case IW_CM_EVENT_CONNECT_REPLY: 988 ret = cm_conn_rep_handler(cm_id_priv, iw_event); 989 break; 990 case IW_CM_EVENT_ESTABLISHED: 991 ret = cm_conn_est_handler(cm_id_priv, iw_event); 992 break; 993 case IW_CM_EVENT_DISCONNECT: 994 cm_disconnect_handler(cm_id_priv, iw_event); 995 break; 996 case IW_CM_EVENT_CLOSE: 997 ret = cm_close_handler(cm_id_priv, iw_event); 998 break; 999 default: 1000 BUG(); 1001 } 1002 1003 return ret; 1004} 1005 1006/* 1007 * Process events on the work_list for the cm_id. If the callback 1008 * function requests that the cm_id be deleted, a flag is set in the 1009 * cm_id flags to indicate that when the last reference is 1010 * removed, the cm_id is to be destroyed. This is necessary to 1011 * distinguish between an object that will be destroyed by the app 1012 * thread asleep on the destroy_comp list vs. an object destroyed 1013 * here synchronously when the last reference is removed. 1014 */ 1015static void cm_work_handler(struct work_struct *_work) 1016{ 1017 struct iwcm_work *work = container_of(_work, struct iwcm_work, work); 1018 struct iw_cm_event levent; 1019 struct iwcm_id_private *cm_id_priv = work->cm_id; 1020 unsigned long flags; 1021 int empty; 1022 int ret = 0; 1023 1024 spin_lock_irqsave(&cm_id_priv->lock, flags); 1025 empty = list_empty(&cm_id_priv->work_list); 1026 while (!empty) { 1027 work = list_entry(cm_id_priv->work_list.next, 1028 struct iwcm_work, list); 1029 list_del_init(&work->list); 1030 empty = list_empty(&cm_id_priv->work_list); 1031 levent = work->event; 1032 put_work(work); 1033 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1034 1035 if (!test_bit(IWCM_F_DROP_EVENTS, &cm_id_priv->flags)) { 1036 ret = process_event(cm_id_priv, &levent); 1037 if (ret) 1038 destroy_cm_id(&cm_id_priv->id); 1039 } else 1040 pr_debug("dropping event %d\n", levent.event); 1041 if (iwcm_deref_id(cm_id_priv)) 1042 return; 1043 if (empty) 1044 return; 1045 spin_lock_irqsave(&cm_id_priv->lock, flags); 1046 } 1047 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1048} 1049 1050/* 1051 * This function is called on interrupt context. Schedule events on 1052 * the iwcm_wq thread to allow callback functions to downcall into 1053 * the CM and/or block. Events are queued to a per-CM_ID 1054 * work_list. If this is the first event on the work_list, the work 1055 * element is also queued on the iwcm_wq thread. 1056 * 1057 * Each event holds a reference on the cm_id. Until the last posted 1058 * event has been delivered and processed, the cm_id cannot be 1059 * deleted. 1060 * 1061 * Returns: 1062 * 0 - the event was handled. 1063 * -ENOMEM - the event was not handled due to lack of resources. 1064 */ 1065static int cm_event_handler(struct iw_cm_id *cm_id, 1066 struct iw_cm_event *iw_event) 1067{ 1068 struct iwcm_work *work; 1069 struct iwcm_id_private *cm_id_priv; 1070 unsigned long flags; 1071 int ret = 0; 1072 1073 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 1074 1075 spin_lock_irqsave(&cm_id_priv->lock, flags); 1076 work = get_work(cm_id_priv); 1077 if (!work) { 1078 ret = -ENOMEM; 1079 goto out; 1080 } 1081 1082 INIT_WORK(&work->work, cm_work_handler); 1083 work->cm_id = cm_id_priv; 1084 work->event = *iw_event; 1085 1086 if ((work->event.event == IW_CM_EVENT_CONNECT_REQUEST || 1087 work->event.event == IW_CM_EVENT_CONNECT_REPLY) && 1088 work->event.private_data_len) { 1089 ret = copy_private_data(&work->event); 1090 if (ret) { 1091 put_work(work); 1092 goto out; 1093 } 1094 } 1095 1096 refcount_inc(&cm_id_priv->refcount); 1097 if (list_empty(&cm_id_priv->work_list)) { 1098 list_add_tail(&work->list, &cm_id_priv->work_list); 1099 queue_work(iwcm_wq, &work->work); 1100 } else 1101 list_add_tail(&work->list, &cm_id_priv->work_list); 1102out: 1103 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1104 return ret; 1105} 1106 1107static int iwcm_init_qp_init_attr(struct iwcm_id_private *cm_id_priv, 1108 struct ib_qp_attr *qp_attr, 1109 int *qp_attr_mask) 1110{ 1111 unsigned long flags; 1112 int ret; 1113 1114 spin_lock_irqsave(&cm_id_priv->lock, flags); 1115 switch (cm_id_priv->state) { 1116 case IW_CM_STATE_IDLE: 1117 case IW_CM_STATE_CONN_SENT: 1118 case IW_CM_STATE_CONN_RECV: 1119 case IW_CM_STATE_ESTABLISHED: 1120 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS; 1121 qp_attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE| 1122 IB_ACCESS_REMOTE_READ; 1123 ret = 0; 1124 break; 1125 default: 1126 ret = -EINVAL; 1127 break; 1128 } 1129 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1130 return ret; 1131} 1132 1133static int iwcm_init_qp_rts_attr(struct iwcm_id_private *cm_id_priv, 1134 struct ib_qp_attr *qp_attr, 1135 int *qp_attr_mask) 1136{ 1137 unsigned long flags; 1138 int ret; 1139 1140 spin_lock_irqsave(&cm_id_priv->lock, flags); 1141 switch (cm_id_priv->state) { 1142 case IW_CM_STATE_IDLE: 1143 case IW_CM_STATE_CONN_SENT: 1144 case IW_CM_STATE_CONN_RECV: 1145 case IW_CM_STATE_ESTABLISHED: 1146 *qp_attr_mask = 0; 1147 ret = 0; 1148 break; 1149 default: 1150 ret = -EINVAL; 1151 break; 1152 } 1153 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1154 return ret; 1155} 1156 1157int iw_cm_init_qp_attr(struct iw_cm_id *cm_id, 1158 struct ib_qp_attr *qp_attr, 1159 int *qp_attr_mask) 1160{ 1161 struct iwcm_id_private *cm_id_priv; 1162 int ret; 1163 1164 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 1165 switch (qp_attr->qp_state) { 1166 case IB_QPS_INIT: 1167 case IB_QPS_RTR: 1168 ret = iwcm_init_qp_init_attr(cm_id_priv, 1169 qp_attr, qp_attr_mask); 1170 break; 1171 case IB_QPS_RTS: 1172 ret = iwcm_init_qp_rts_attr(cm_id_priv, 1173 qp_attr, qp_attr_mask); 1174 break; 1175 default: 1176 ret = -EINVAL; 1177 break; 1178 } 1179 return ret; 1180} 1181EXPORT_SYMBOL(iw_cm_init_qp_attr); 1182 1183static int __init iw_cm_init(void) 1184{ 1185 int ret; 1186 1187 ret = iwpm_init(RDMA_NL_IWCM); 1188 if (ret) 1189 return ret; 1190 1191 iwcm_wq = alloc_ordered_workqueue("iw_cm_wq", 0); 1192 if (!iwcm_wq) 1193 goto err_alloc; 1194 1195 iwcm_ctl_table_hdr = register_net_sysctl(&init_net, "net/iw_cm", 1196 iwcm_ctl_table); 1197 if (!iwcm_ctl_table_hdr) { 1198 pr_err("iw_cm: couldn't register sysctl paths\n"); 1199 goto err_sysctl; 1200 } 1201 1202 rdma_nl_register(RDMA_NL_IWCM, iwcm_nl_cb_table); 1203 return 0; 1204 1205err_sysctl: 1206 destroy_workqueue(iwcm_wq); 1207err_alloc: 1208 iwpm_exit(RDMA_NL_IWCM); 1209 return -ENOMEM; 1210} 1211 1212static void __exit iw_cm_cleanup(void) 1213{ 1214 rdma_nl_unregister(RDMA_NL_IWCM); 1215 unregister_net_sysctl_table(iwcm_ctl_table_hdr); 1216 destroy_workqueue(iwcm_wq); 1217 iwpm_exit(RDMA_NL_IWCM); 1218} 1219 1220MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_IWCM, 2); 1221 1222module_init(iw_cm_init); 1223module_exit(iw_cm_cleanup); 1224