1/* 2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. 3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. 4 * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the 10 * OpenIB.org BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or 13 * without modification, are permitted provided that the following 14 * conditions are met: 15 * 16 * - Redistributions of source code must retain the above 17 * copyright notice, this list of conditions and the following 18 * disclaimer. 19 * 20 * - Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials 23 * provided with the distribution. 24 * 25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 32 * SOFTWARE. 33 */ 34#include <linux/kernel.h> 35#include <linux/module.h> 36#include <linux/slab.h> 37#include <linux/delay.h> 38 39#include "iscsi_iser.h" 40 41#define ISCSI_ISER_MAX_CONN 8 42#define ISER_MAX_RX_LEN (ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN) 43#define ISER_MAX_TX_LEN (ISER_QP_MAX_REQ_DTOS * ISCSI_ISER_MAX_CONN) 44#define ISER_MAX_CQ_LEN (ISER_MAX_RX_LEN + ISER_MAX_TX_LEN + \ 45 ISCSI_ISER_MAX_CONN) 46 47static void iser_qp_event_callback(struct ib_event *cause, void *context) 48{ 49 iser_err("qp event %s (%d)\n", 50 ib_event_msg(cause->event), cause->event); 51} 52 53static void iser_event_handler(struct ib_event_handler *handler, 54 struct ib_event *event) 55{ 56 iser_err("async event %s (%d) on device %s port %d\n", 57 ib_event_msg(event->event), event->event, 58 dev_name(&event->device->dev), event->element.port_num); 59} 60 61/* 62 * iser_create_device_ib_res - creates Protection Domain (PD), Completion 63 * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with 64 * the adaptor. 65 * 66 * Return: 0 on success, -1 on failure 67 */ 68static int iser_create_device_ib_res(struct iser_device *device) 69{ 70 struct ib_device *ib_dev = device->ib_device; 71 72 if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS)) { 73 iser_err("IB device does not support memory registrations\n"); 74 return -1; 75 } 76 77 device->pd = ib_alloc_pd(ib_dev, 78 iser_always_reg ? 0 : IB_PD_UNSAFE_GLOBAL_RKEY); 79 if (IS_ERR(device->pd)) 80 goto pd_err; 81 82 INIT_IB_EVENT_HANDLER(&device->event_handler, ib_dev, 83 iser_event_handler); 84 ib_register_event_handler(&device->event_handler); 85 return 0; 86 87pd_err: 88 iser_err("failed to allocate an IB resource\n"); 89 return -1; 90} 91 92/* 93 * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR, 94 * CQ and PD created with the device associated with the adaptor. 95 */ 96static void iser_free_device_ib_res(struct iser_device *device) 97{ 98 ib_unregister_event_handler(&device->event_handler); 99 ib_dealloc_pd(device->pd); 100 101 device->pd = NULL; 102} 103 104static struct iser_fr_desc * 105iser_create_fastreg_desc(struct iser_device *device, 106 struct ib_pd *pd, 107 bool pi_enable, 108 unsigned int size) 109{ 110 struct iser_fr_desc *desc; 111 struct ib_device *ib_dev = device->ib_device; 112 enum ib_mr_type mr_type; 113 int ret; 114 115 desc = kzalloc(sizeof(*desc), GFP_KERNEL); 116 if (!desc) 117 return ERR_PTR(-ENOMEM); 118 119 if (ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG) 120 mr_type = IB_MR_TYPE_SG_GAPS; 121 else 122 mr_type = IB_MR_TYPE_MEM_REG; 123 124 desc->rsc.mr = ib_alloc_mr(pd, mr_type, size); 125 if (IS_ERR(desc->rsc.mr)) { 126 ret = PTR_ERR(desc->rsc.mr); 127 iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret); 128 goto err_alloc_mr; 129 } 130 131 if (pi_enable) { 132 desc->rsc.sig_mr = ib_alloc_mr_integrity(pd, size, size); 133 if (IS_ERR(desc->rsc.sig_mr)) { 134 ret = PTR_ERR(desc->rsc.sig_mr); 135 iser_err("Failed to allocate sig_mr err=%d\n", ret); 136 goto err_alloc_mr_integrity; 137 } 138 } 139 140 return desc; 141 142err_alloc_mr_integrity: 143 ib_dereg_mr(desc->rsc.mr); 144err_alloc_mr: 145 kfree(desc); 146 147 return ERR_PTR(ret); 148} 149 150static void iser_destroy_fastreg_desc(struct iser_fr_desc *desc) 151{ 152 struct iser_reg_resources *res = &desc->rsc; 153 154 ib_dereg_mr(res->mr); 155 if (res->sig_mr) { 156 ib_dereg_mr(res->sig_mr); 157 res->sig_mr = NULL; 158 } 159 kfree(desc); 160} 161 162/** 163 * iser_alloc_fastreg_pool - Creates pool of fast_reg descriptors 164 * for fast registration work requests. 165 * @ib_conn: connection RDMA resources 166 * @cmds_max: max number of SCSI commands for this connection 167 * @size: max number of pages per map request 168 * 169 * Return: 0 on success, or errno code on failure 170 */ 171int iser_alloc_fastreg_pool(struct ib_conn *ib_conn, 172 unsigned cmds_max, 173 unsigned int size) 174{ 175 struct iser_device *device = ib_conn->device; 176 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; 177 struct iser_fr_desc *desc; 178 int i, ret; 179 180 INIT_LIST_HEAD(&fr_pool->list); 181 INIT_LIST_HEAD(&fr_pool->all_list); 182 spin_lock_init(&fr_pool->lock); 183 fr_pool->size = 0; 184 for (i = 0; i < cmds_max; i++) { 185 desc = iser_create_fastreg_desc(device, device->pd, 186 ib_conn->pi_support, size); 187 if (IS_ERR(desc)) { 188 ret = PTR_ERR(desc); 189 goto err; 190 } 191 192 list_add_tail(&desc->list, &fr_pool->list); 193 list_add_tail(&desc->all_list, &fr_pool->all_list); 194 fr_pool->size++; 195 } 196 197 return 0; 198 199err: 200 iser_free_fastreg_pool(ib_conn); 201 return ret; 202} 203 204/** 205 * iser_free_fastreg_pool - releases the pool of fast_reg descriptors 206 * @ib_conn: connection RDMA resources 207 */ 208void iser_free_fastreg_pool(struct ib_conn *ib_conn) 209{ 210 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; 211 struct iser_fr_desc *desc, *tmp; 212 int i = 0; 213 214 if (list_empty(&fr_pool->all_list)) 215 return; 216 217 iser_info("freeing conn %p fr pool\n", ib_conn); 218 219 list_for_each_entry_safe(desc, tmp, &fr_pool->all_list, all_list) { 220 list_del(&desc->all_list); 221 iser_destroy_fastreg_desc(desc); 222 ++i; 223 } 224 225 if (i < fr_pool->size) 226 iser_warn("pool still has %d regions registered\n", 227 fr_pool->size - i); 228} 229 230/* 231 * iser_create_ib_conn_res - Queue-Pair (QP) 232 * 233 * Return: 0 on success, -1 on failure 234 */ 235static int iser_create_ib_conn_res(struct ib_conn *ib_conn) 236{ 237 struct iser_conn *iser_conn = to_iser_conn(ib_conn); 238 struct iser_device *device; 239 struct ib_device *ib_dev; 240 struct ib_qp_init_attr init_attr; 241 int ret = -ENOMEM; 242 unsigned int max_send_wr, cq_size; 243 244 BUG_ON(ib_conn->device == NULL); 245 246 device = ib_conn->device; 247 ib_dev = device->ib_device; 248 249 if (ib_conn->pi_support) 250 max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1; 251 else 252 max_send_wr = ISER_QP_MAX_REQ_DTOS + 1; 253 max_send_wr = min_t(unsigned int, max_send_wr, 254 (unsigned int)ib_dev->attrs.max_qp_wr); 255 256 cq_size = max_send_wr + ISER_QP_MAX_RECV_DTOS; 257 ib_conn->cq = ib_cq_pool_get(ib_dev, cq_size, -1, IB_POLL_SOFTIRQ); 258 if (IS_ERR(ib_conn->cq)) { 259 ret = PTR_ERR(ib_conn->cq); 260 goto cq_err; 261 } 262 ib_conn->cq_size = cq_size; 263 264 memset(&init_attr, 0, sizeof(init_attr)); 265 266 init_attr.event_handler = iser_qp_event_callback; 267 init_attr.qp_context = (void *)ib_conn; 268 init_attr.send_cq = ib_conn->cq; 269 init_attr.recv_cq = ib_conn->cq; 270 init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS; 271 init_attr.cap.max_send_sge = 2; 272 init_attr.cap.max_recv_sge = 1; 273 init_attr.sq_sig_type = IB_SIGNAL_REQ_WR; 274 init_attr.qp_type = IB_QPT_RC; 275 init_attr.cap.max_send_wr = max_send_wr; 276 if (ib_conn->pi_support) 277 init_attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN; 278 iser_conn->max_cmds = ISER_GET_MAX_XMIT_CMDS(max_send_wr - 1); 279 280 ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); 281 if (ret) 282 goto out_err; 283 284 ib_conn->qp = ib_conn->cma_id->qp; 285 iser_info("setting conn %p cma_id %p qp %p max_send_wr %d\n", 286 ib_conn, ib_conn->cma_id, 287 ib_conn->cma_id->qp, max_send_wr); 288 return ret; 289 290out_err: 291 ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size); 292cq_err: 293 iser_err("unable to alloc mem or create resource, err %d\n", ret); 294 295 return ret; 296} 297 298/* 299 * based on the resolved device node GUID see if there already allocated 300 * device for this device. If there's no such, create one. 301 */ 302static 303struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id) 304{ 305 struct iser_device *device; 306 307 mutex_lock(&ig.device_list_mutex); 308 309 list_for_each_entry(device, &ig.device_list, ig_list) 310 /* find if there's a match using the node GUID */ 311 if (device->ib_device->node_guid == cma_id->device->node_guid) 312 goto inc_refcnt; 313 314 device = kzalloc(sizeof *device, GFP_KERNEL); 315 if (device == NULL) 316 goto out; 317 318 /* assign this device to the device */ 319 device->ib_device = cma_id->device; 320 /* init the device and link it into ig device list */ 321 if (iser_create_device_ib_res(device)) { 322 kfree(device); 323 device = NULL; 324 goto out; 325 } 326 list_add(&device->ig_list, &ig.device_list); 327 328inc_refcnt: 329 device->refcount++; 330out: 331 mutex_unlock(&ig.device_list_mutex); 332 return device; 333} 334 335/* if there's no demand for this device, release it */ 336static void iser_device_try_release(struct iser_device *device) 337{ 338 mutex_lock(&ig.device_list_mutex); 339 device->refcount--; 340 iser_info("device %p refcount %d\n", device, device->refcount); 341 if (!device->refcount) { 342 iser_free_device_ib_res(device); 343 list_del(&device->ig_list); 344 kfree(device); 345 } 346 mutex_unlock(&ig.device_list_mutex); 347} 348 349/* 350 * Called with state mutex held 351 */ 352static int iser_conn_state_comp_exch(struct iser_conn *iser_conn, 353 enum iser_conn_state comp, 354 enum iser_conn_state exch) 355{ 356 int ret; 357 358 ret = (iser_conn->state == comp); 359 if (ret) 360 iser_conn->state = exch; 361 362 return ret; 363} 364 365void iser_release_work(struct work_struct *work) 366{ 367 struct iser_conn *iser_conn; 368 369 iser_conn = container_of(work, struct iser_conn, release_work); 370 371 /* Wait for conn_stop to complete */ 372 wait_for_completion(&iser_conn->stop_completion); 373 /* Wait for IB resouces cleanup to complete */ 374 wait_for_completion(&iser_conn->ib_completion); 375 376 mutex_lock(&iser_conn->state_mutex); 377 iser_conn->state = ISER_CONN_DOWN; 378 mutex_unlock(&iser_conn->state_mutex); 379 380 iser_conn_release(iser_conn); 381} 382 383/** 384 * iser_free_ib_conn_res - release IB related resources 385 * @iser_conn: iser connection struct 386 * @destroy: indicator if we need to try to release the 387 * iser device and memory regoins pool (only iscsi 388 * shutdown and DEVICE_REMOVAL will use this). 389 * 390 * This routine is called with the iser state mutex held 391 * so the cm_id removal is out of here. It is Safe to 392 * be invoked multiple times. 393 */ 394static void iser_free_ib_conn_res(struct iser_conn *iser_conn, 395 bool destroy) 396{ 397 struct ib_conn *ib_conn = &iser_conn->ib_conn; 398 struct iser_device *device = ib_conn->device; 399 400 iser_info("freeing conn %p cma_id %p qp %p\n", 401 iser_conn, ib_conn->cma_id, ib_conn->qp); 402 403 if (ib_conn->qp != NULL) { 404 rdma_destroy_qp(ib_conn->cma_id); 405 ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size); 406 ib_conn->qp = NULL; 407 } 408 409 if (destroy) { 410 if (iser_conn->rx_descs) 411 iser_free_rx_descriptors(iser_conn); 412 413 if (device != NULL) { 414 iser_device_try_release(device); 415 ib_conn->device = NULL; 416 } 417 } 418} 419 420/** 421 * iser_conn_release - Frees all conn objects and deallocs conn descriptor 422 * @iser_conn: iSER connection context 423 */ 424void iser_conn_release(struct iser_conn *iser_conn) 425{ 426 struct ib_conn *ib_conn = &iser_conn->ib_conn; 427 428 mutex_lock(&ig.connlist_mutex); 429 list_del(&iser_conn->conn_list); 430 mutex_unlock(&ig.connlist_mutex); 431 432 mutex_lock(&iser_conn->state_mutex); 433 /* In case we endup here without ep_disconnect being invoked. */ 434 if (iser_conn->state != ISER_CONN_DOWN) { 435 iser_warn("iser conn %p state %d, expected state down.\n", 436 iser_conn, iser_conn->state); 437 iscsi_destroy_endpoint(iser_conn->ep); 438 iser_conn->state = ISER_CONN_DOWN; 439 } 440 /* 441 * In case we never got to bind stage, we still need to 442 * release IB resources (which is safe to call more than once). 443 */ 444 iser_free_ib_conn_res(iser_conn, true); 445 mutex_unlock(&iser_conn->state_mutex); 446 447 if (ib_conn->cma_id != NULL) { 448 rdma_destroy_id(ib_conn->cma_id); 449 ib_conn->cma_id = NULL; 450 } 451 452 kfree(iser_conn); 453} 454 455/** 456 * iser_conn_terminate - triggers start of the disconnect procedures and 457 * waits for them to be done 458 * @iser_conn: iSER connection context 459 * 460 * Called with state mutex held 461 */ 462int iser_conn_terminate(struct iser_conn *iser_conn) 463{ 464 struct ib_conn *ib_conn = &iser_conn->ib_conn; 465 int err = 0; 466 467 /* terminate the iser conn only if the conn state is UP */ 468 if (!iser_conn_state_comp_exch(iser_conn, ISER_CONN_UP, 469 ISER_CONN_TERMINATING)) 470 return 0; 471 472 iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state); 473 474 /* suspend queuing of new iscsi commands */ 475 if (iser_conn->iscsi_conn) 476 iscsi_suspend_queue(iser_conn->iscsi_conn); 477 478 /* 479 * In case we didn't already clean up the cma_id (peer initiated 480 * a disconnection), we need to Cause the CMA to change the QP 481 * state to ERROR. 482 */ 483 if (ib_conn->cma_id) { 484 err = rdma_disconnect(ib_conn->cma_id); 485 if (err) 486 iser_err("Failed to disconnect, conn: 0x%p err %d\n", 487 iser_conn, err); 488 489 /* block until all flush errors are consumed */ 490 ib_drain_sq(ib_conn->qp); 491 } 492 493 return 1; 494} 495 496/* 497 * Called with state mutex held 498 */ 499static void iser_connect_error(struct rdma_cm_id *cma_id) 500{ 501 struct iser_conn *iser_conn; 502 503 iser_conn = (struct iser_conn *)cma_id->context; 504 iser_conn->state = ISER_CONN_TERMINATING; 505} 506 507static void 508iser_calc_scsi_params(struct iser_conn *iser_conn, 509 unsigned int max_sectors) 510{ 511 struct iser_device *device = iser_conn->ib_conn.device; 512 struct ib_device_attr *attr = &device->ib_device->attrs; 513 unsigned short sg_tablesize, sup_sg_tablesize; 514 unsigned short reserved_mr_pages; 515 u32 max_num_sg; 516 517 /* 518 * FRs without SG_GAPS can only map up to a (device) page per entry, 519 * but if the first entry is misaligned we'll end up using two entries 520 * (head and tail) for a single page worth data, so one additional 521 * entry is required. 522 */ 523 if (attr->device_cap_flags & IB_DEVICE_SG_GAPS_REG) 524 reserved_mr_pages = 0; 525 else 526 reserved_mr_pages = 1; 527 528 if (iser_conn->ib_conn.pi_support) 529 max_num_sg = attr->max_pi_fast_reg_page_list_len; 530 else 531 max_num_sg = attr->max_fast_reg_page_list_len; 532 533 sg_tablesize = DIV_ROUND_UP(max_sectors * SECTOR_SIZE, SZ_4K); 534 sup_sg_tablesize = min_t(uint, ISCSI_ISER_MAX_SG_TABLESIZE, 535 max_num_sg - reserved_mr_pages); 536 iser_conn->scsi_sg_tablesize = min(sg_tablesize, sup_sg_tablesize); 537 iser_conn->pages_per_mr = 538 iser_conn->scsi_sg_tablesize + reserved_mr_pages; 539} 540 541/* 542 * Called with state mutex held 543 */ 544static void iser_addr_handler(struct rdma_cm_id *cma_id) 545{ 546 struct iser_device *device; 547 struct iser_conn *iser_conn; 548 struct ib_conn *ib_conn; 549 int ret; 550 551 iser_conn = (struct iser_conn *)cma_id->context; 552 if (iser_conn->state != ISER_CONN_PENDING) 553 /* bailout */ 554 return; 555 556 ib_conn = &iser_conn->ib_conn; 557 device = iser_device_find_by_ib_device(cma_id); 558 if (!device) { 559 iser_err("device lookup/creation failed\n"); 560 iser_connect_error(cma_id); 561 return; 562 } 563 564 ib_conn->device = device; 565 566 /* connection T10-PI support */ 567 if (iser_pi_enable) { 568 if (!(device->ib_device->attrs.device_cap_flags & 569 IB_DEVICE_INTEGRITY_HANDOVER)) { 570 iser_warn("T10-PI requested but not supported on %s, " 571 "continue without T10-PI\n", 572 dev_name(&ib_conn->device->ib_device->dev)); 573 ib_conn->pi_support = false; 574 } else { 575 ib_conn->pi_support = true; 576 } 577 } 578 579 iser_calc_scsi_params(iser_conn, iser_max_sectors); 580 581 ret = rdma_resolve_route(cma_id, 1000); 582 if (ret) { 583 iser_err("resolve route failed: %d\n", ret); 584 iser_connect_error(cma_id); 585 return; 586 } 587} 588 589/* 590 * Called with state mutex held 591 */ 592static void iser_route_handler(struct rdma_cm_id *cma_id) 593{ 594 struct rdma_conn_param conn_param; 595 int ret; 596 struct iser_cm_hdr req_hdr; 597 struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; 598 struct ib_conn *ib_conn = &iser_conn->ib_conn; 599 struct ib_device *ib_dev = ib_conn->device->ib_device; 600 601 if (iser_conn->state != ISER_CONN_PENDING) 602 /* bailout */ 603 return; 604 605 ret = iser_create_ib_conn_res(ib_conn); 606 if (ret) 607 goto failure; 608 609 memset(&conn_param, 0, sizeof conn_param); 610 conn_param.responder_resources = ib_dev->attrs.max_qp_rd_atom; 611 conn_param.initiator_depth = 1; 612 conn_param.retry_count = 7; 613 conn_param.rnr_retry_count = 6; 614 615 memset(&req_hdr, 0, sizeof(req_hdr)); 616 req_hdr.flags = ISER_ZBVA_NOT_SUP; 617 if (!iser_always_reg) 618 req_hdr.flags |= ISER_SEND_W_INV_NOT_SUP; 619 conn_param.private_data = (void *)&req_hdr; 620 conn_param.private_data_len = sizeof(struct iser_cm_hdr); 621 622 ret = rdma_connect_locked(cma_id, &conn_param); 623 if (ret) { 624 iser_err("failure connecting: %d\n", ret); 625 goto failure; 626 } 627 628 return; 629failure: 630 iser_connect_error(cma_id); 631} 632 633static void iser_connected_handler(struct rdma_cm_id *cma_id, 634 const void *private_data) 635{ 636 struct iser_conn *iser_conn; 637 struct ib_qp_attr attr; 638 struct ib_qp_init_attr init_attr; 639 640 iser_conn = (struct iser_conn *)cma_id->context; 641 if (iser_conn->state != ISER_CONN_PENDING) 642 /* bailout */ 643 return; 644 645 (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr); 646 iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num); 647 648 if (private_data) { 649 u8 flags = *(u8 *)private_data; 650 651 iser_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP); 652 } 653 654 iser_info("conn %p: negotiated %s invalidation\n", 655 iser_conn, iser_conn->snd_w_inv ? "remote" : "local"); 656 657 iser_conn->state = ISER_CONN_UP; 658 complete(&iser_conn->up_completion); 659} 660 661static void iser_disconnected_handler(struct rdma_cm_id *cma_id) 662{ 663 struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; 664 665 if (iser_conn_terminate(iser_conn)) { 666 if (iser_conn->iscsi_conn) 667 iscsi_conn_failure(iser_conn->iscsi_conn, 668 ISCSI_ERR_CONN_FAILED); 669 else 670 iser_err("iscsi_iser connection isn't bound\n"); 671 } 672} 673 674static void iser_cleanup_handler(struct rdma_cm_id *cma_id, 675 bool destroy) 676{ 677 struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; 678 679 /* 680 * We are not guaranteed that we visited disconnected_handler 681 * by now, call it here to be safe that we handle CM drep 682 * and flush errors. 683 */ 684 iser_disconnected_handler(cma_id); 685 iser_free_ib_conn_res(iser_conn, destroy); 686 complete(&iser_conn->ib_completion); 687}; 688 689static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) 690{ 691 struct iser_conn *iser_conn; 692 int ret = 0; 693 694 iser_conn = (struct iser_conn *)cma_id->context; 695 iser_info("%s (%d): status %d conn %p id %p\n", 696 rdma_event_msg(event->event), event->event, 697 event->status, cma_id->context, cma_id); 698 699 mutex_lock(&iser_conn->state_mutex); 700 switch (event->event) { 701 case RDMA_CM_EVENT_ADDR_RESOLVED: 702 iser_addr_handler(cma_id); 703 break; 704 case RDMA_CM_EVENT_ROUTE_RESOLVED: 705 iser_route_handler(cma_id); 706 break; 707 case RDMA_CM_EVENT_ESTABLISHED: 708 iser_connected_handler(cma_id, event->param.conn.private_data); 709 break; 710 case RDMA_CM_EVENT_REJECTED: 711 iser_info("Connection rejected: %s\n", 712 rdma_reject_msg(cma_id, event->status)); 713 fallthrough; 714 case RDMA_CM_EVENT_ADDR_ERROR: 715 case RDMA_CM_EVENT_ROUTE_ERROR: 716 case RDMA_CM_EVENT_CONNECT_ERROR: 717 case RDMA_CM_EVENT_UNREACHABLE: 718 iser_connect_error(cma_id); 719 break; 720 case RDMA_CM_EVENT_DISCONNECTED: 721 case RDMA_CM_EVENT_ADDR_CHANGE: 722 case RDMA_CM_EVENT_TIMEWAIT_EXIT: 723 iser_cleanup_handler(cma_id, false); 724 break; 725 case RDMA_CM_EVENT_DEVICE_REMOVAL: 726 /* 727 * we *must* destroy the device as we cannot rely 728 * on iscsid to be around to initiate error handling. 729 * also if we are not in state DOWN implicitly destroy 730 * the cma_id. 731 */ 732 iser_cleanup_handler(cma_id, true); 733 if (iser_conn->state != ISER_CONN_DOWN) { 734 iser_conn->ib_conn.cma_id = NULL; 735 ret = 1; 736 } 737 break; 738 default: 739 iser_err("Unexpected RDMA CM event: %s (%d)\n", 740 rdma_event_msg(event->event), event->event); 741 break; 742 } 743 mutex_unlock(&iser_conn->state_mutex); 744 745 return ret; 746} 747 748void iser_conn_init(struct iser_conn *iser_conn) 749{ 750 struct ib_conn *ib_conn = &iser_conn->ib_conn; 751 752 iser_conn->state = ISER_CONN_INIT; 753 init_completion(&iser_conn->stop_completion); 754 init_completion(&iser_conn->ib_completion); 755 init_completion(&iser_conn->up_completion); 756 INIT_LIST_HEAD(&iser_conn->conn_list); 757 mutex_init(&iser_conn->state_mutex); 758 759 ib_conn->post_recv_buf_count = 0; 760 ib_conn->reg_cqe.done = iser_reg_comp; 761} 762 763 /** 764 * starts the process of connecting to the target 765 * sleeps until the connection is established or rejected 766 */ 767int iser_connect(struct iser_conn *iser_conn, 768 struct sockaddr *src_addr, 769 struct sockaddr *dst_addr, 770 int non_blocking) 771{ 772 struct ib_conn *ib_conn = &iser_conn->ib_conn; 773 int err = 0; 774 775 mutex_lock(&iser_conn->state_mutex); 776 777 sprintf(iser_conn->name, "%pISp", dst_addr); 778 779 iser_info("connecting to: %s\n", iser_conn->name); 780 781 /* the device is known only --after-- address resolution */ 782 ib_conn->device = NULL; 783 784 iser_conn->state = ISER_CONN_PENDING; 785 786 ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler, 787 (void *)iser_conn, 788 RDMA_PS_TCP, IB_QPT_RC); 789 if (IS_ERR(ib_conn->cma_id)) { 790 err = PTR_ERR(ib_conn->cma_id); 791 iser_err("rdma_create_id failed: %d\n", err); 792 goto id_failure; 793 } 794 795 err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000); 796 if (err) { 797 iser_err("rdma_resolve_addr failed: %d\n", err); 798 goto addr_failure; 799 } 800 801 if (!non_blocking) { 802 wait_for_completion_interruptible(&iser_conn->up_completion); 803 804 if (iser_conn->state != ISER_CONN_UP) { 805 err = -EIO; 806 goto connect_failure; 807 } 808 } 809 mutex_unlock(&iser_conn->state_mutex); 810 811 mutex_lock(&ig.connlist_mutex); 812 list_add(&iser_conn->conn_list, &ig.connlist); 813 mutex_unlock(&ig.connlist_mutex); 814 return 0; 815 816id_failure: 817 ib_conn->cma_id = NULL; 818addr_failure: 819 iser_conn->state = ISER_CONN_DOWN; 820connect_failure: 821 mutex_unlock(&iser_conn->state_mutex); 822 iser_conn_release(iser_conn); 823 return err; 824} 825 826int iser_post_recvl(struct iser_conn *iser_conn) 827{ 828 struct ib_conn *ib_conn = &iser_conn->ib_conn; 829 struct iser_login_desc *desc = &iser_conn->login_desc; 830 struct ib_recv_wr wr; 831 int ib_ret; 832 833 desc->sge.addr = desc->rsp_dma; 834 desc->sge.length = ISER_RX_LOGIN_SIZE; 835 desc->sge.lkey = ib_conn->device->pd->local_dma_lkey; 836 837 desc->cqe.done = iser_login_rsp; 838 wr.wr_cqe = &desc->cqe; 839 wr.sg_list = &desc->sge; 840 wr.num_sge = 1; 841 wr.next = NULL; 842 843 ib_conn->post_recv_buf_count++; 844 ib_ret = ib_post_recv(ib_conn->qp, &wr, NULL); 845 if (ib_ret) { 846 iser_err("ib_post_recv failed ret=%d\n", ib_ret); 847 ib_conn->post_recv_buf_count--; 848 } 849 850 return ib_ret; 851} 852 853int iser_post_recvm(struct iser_conn *iser_conn, int count) 854{ 855 struct ib_conn *ib_conn = &iser_conn->ib_conn; 856 unsigned int my_rx_head = iser_conn->rx_desc_head; 857 struct iser_rx_desc *rx_desc; 858 struct ib_recv_wr *wr; 859 int i, ib_ret; 860 861 for (wr = ib_conn->rx_wr, i = 0; i < count; i++, wr++) { 862 rx_desc = &iser_conn->rx_descs[my_rx_head]; 863 rx_desc->cqe.done = iser_task_rsp; 864 wr->wr_cqe = &rx_desc->cqe; 865 wr->sg_list = &rx_desc->rx_sg; 866 wr->num_sge = 1; 867 wr->next = wr + 1; 868 my_rx_head = (my_rx_head + 1) & iser_conn->qp_max_recv_dtos_mask; 869 } 870 871 wr--; 872 wr->next = NULL; /* mark end of work requests list */ 873 874 ib_conn->post_recv_buf_count += count; 875 ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, NULL); 876 if (unlikely(ib_ret)) { 877 iser_err("ib_post_recv failed ret=%d\n", ib_ret); 878 ib_conn->post_recv_buf_count -= count; 879 } else 880 iser_conn->rx_desc_head = my_rx_head; 881 882 return ib_ret; 883} 884 885 886/** 887 * iser_post_send - Initiate a Send DTO operation 888 * @ib_conn: connection RDMA resources 889 * @tx_desc: iSER TX descriptor 890 * @signal: true to send work request as SIGNALED 891 * 892 * Return: 0 on success, -1 on failure 893 */ 894int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc, 895 bool signal) 896{ 897 struct ib_send_wr *wr = &tx_desc->send_wr; 898 struct ib_send_wr *first_wr; 899 int ib_ret; 900 901 ib_dma_sync_single_for_device(ib_conn->device->ib_device, 902 tx_desc->dma_addr, ISER_HEADERS_LEN, 903 DMA_TO_DEVICE); 904 905 wr->next = NULL; 906 wr->wr_cqe = &tx_desc->cqe; 907 wr->sg_list = tx_desc->tx_sg; 908 wr->num_sge = tx_desc->num_sge; 909 wr->opcode = IB_WR_SEND; 910 wr->send_flags = signal ? IB_SEND_SIGNALED : 0; 911 912 if (tx_desc->inv_wr.next) 913 first_wr = &tx_desc->inv_wr; 914 else if (tx_desc->reg_wr.wr.next) 915 first_wr = &tx_desc->reg_wr.wr; 916 else 917 first_wr = wr; 918 919 ib_ret = ib_post_send(ib_conn->qp, first_wr, NULL); 920 if (unlikely(ib_ret)) 921 iser_err("ib_post_send failed, ret:%d opcode:%d\n", 922 ib_ret, wr->opcode); 923 924 return ib_ret; 925} 926 927u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task, 928 enum iser_data_dir cmd_dir, sector_t *sector) 929{ 930 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir]; 931 struct iser_fr_desc *desc = reg->mem_h; 932 unsigned long sector_size = iser_task->sc->device->sector_size; 933 struct ib_mr_status mr_status; 934 int ret; 935 936 if (desc && desc->sig_protected) { 937 desc->sig_protected = false; 938 ret = ib_check_mr_status(desc->rsc.sig_mr, 939 IB_MR_CHECK_SIG_STATUS, &mr_status); 940 if (ret) { 941 iser_err("ib_check_mr_status failed, ret %d\n", ret); 942 /* Not a lot we can do, return ambiguous guard error */ 943 *sector = 0; 944 return 0x1; 945 } 946 947 if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) { 948 sector_t sector_off = mr_status.sig_err.sig_err_offset; 949 950 sector_div(sector_off, sector_size + 8); 951 *sector = scsi_get_lba(iser_task->sc) + sector_off; 952 953 iser_err("PI error found type %d at sector %llx " 954 "expected %x vs actual %x\n", 955 mr_status.sig_err.err_type, 956 (unsigned long long)*sector, 957 mr_status.sig_err.expected, 958 mr_status.sig_err.actual); 959 960 switch (mr_status.sig_err.err_type) { 961 case IB_SIG_BAD_GUARD: 962 return 0x1; 963 case IB_SIG_BAD_REFTAG: 964 return 0x3; 965 case IB_SIG_BAD_APPTAG: 966 return 0x2; 967 } 968 } 969 } 970 971 return 0; 972} 973 974void iser_err_comp(struct ib_wc *wc, const char *type) 975{ 976 if (wc->status != IB_WC_WR_FLUSH_ERR) { 977 struct iser_conn *iser_conn = to_iser_conn(wc->qp->qp_context); 978 979 iser_err("%s failure: %s (%d) vend_err %#x\n", type, 980 ib_wc_status_msg(wc->status), wc->status, 981 wc->vendor_err); 982 983 if (iser_conn->iscsi_conn) 984 iscsi_conn_failure(iser_conn->iscsi_conn, 985 ISCSI_ERR_CONN_FAILED); 986 } else { 987 iser_dbg("%s failure: %s (%d)\n", type, 988 ib_wc_status_msg(wc->status), wc->status); 989 } 990} 991