1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * RDMA Transport Layer 4 * 5 * Copyright (c) 2014 - 2018 ProfitBricks GmbH. All rights reserved. 6 * Copyright (c) 2018 - 2019 1&1 IONOS Cloud GmbH. All rights reserved. 7 * Copyright (c) 2019 - 2020 1&1 IONOS SE. All rights reserved. 8 */ 9 10 #undef pr_fmt 11 #define pr_fmt(fmt) KBUILD_MODNAME " L" __stringify(__LINE__) ": " fmt 12 13 #include <linux/module.h> 14 #include <linux/rculist.h> 15 #include <linux/random.h> 16 17 #include "rtrs-clt.h" 18 #include "rtrs-log.h" 19 #include "rtrs-clt-trace.h" 20 21 #define RTRS_CONNECT_TIMEOUT_MS 30000 22 /* 23 * Wait a bit before trying to reconnect after a failure 24 * in order to give server time to finish clean up which 25 * leads to "false positives" failed reconnect attempts 26 */ 27 #define RTRS_RECONNECT_BACKOFF 1000 28 /* 29 * Wait for additional random time between 0 and 8 seconds 30 * before starting to reconnect to avoid clients reconnecting 31 * all at once in case of a major network outage 32 */ 33 #define RTRS_RECONNECT_SEED 8 34 35 #define FIRST_CONN 0x01 36 /* limit to 128 * 4k = 512k max IO */ 37 #define RTRS_MAX_SEGMENTS 128 38 39 MODULE_DESCRIPTION("RDMA Transport Client"); 40 MODULE_LICENSE("GPL"); 41 42 static const struct rtrs_rdma_dev_pd_ops dev_pd_ops; 43 static struct rtrs_rdma_dev_pd dev_pd = { 44 .ops = &dev_pd_ops 45 }; 46 47 static struct workqueue_struct *rtrs_wq; 48 static const struct class rtrs_clt_dev_class = { 49 .name = "rtrs-client", 50 }; 51 52 static inline bool rtrs_clt_is_connected(const struct rtrs_clt_sess *clt) 53 { 54 struct rtrs_clt_path *clt_path; 55 bool connected = false; 56 57 rcu_read_lock(); 58 list_for_each_entry_rcu(clt_path, &clt->paths_list, s.entry) 59 if (READ_ONCE(clt_path->state) == RTRS_CLT_CONNECTED) { 60 connected = true; 61 break; 62 } 63 rcu_read_unlock(); 64 65 return connected; 66 } 67 68 static struct rtrs_permit * 69 __rtrs_get_permit(struct rtrs_clt_sess *clt, enum rtrs_clt_con_type con_type) 70 { 71 size_t max_depth = clt->queue_depth; 72 struct rtrs_permit *permit; 73 int bit; 74 75 /* 76 * Adapted from null_blk get_tag(). Callers from different cpus may 77 * grab the same bit, since find_first_zero_bit is not atomic. 78 * But then the test_and_set_bit_lock will fail for all the 79 * callers but one, so that they will loop again. 80 * This way an explicit spinlock is not required. 81 */ 82 do { 83 bit = find_first_zero_bit(clt->permits_map, max_depth); 84 if (bit >= max_depth) 85 return NULL; 86 } while (test_and_set_bit_lock(bit, clt->permits_map)); 87 88 permit = get_permit(clt, bit); 89 WARN_ON(permit->mem_id != bit); 90 permit->cpu_id = raw_smp_processor_id(); 91 permit->con_type = con_type; 92 93 return permit; 94 } 95 96 static inline void __rtrs_put_permit(struct rtrs_clt_sess *clt, 97 struct rtrs_permit *permit) 98 { 99 clear_bit_unlock(permit->mem_id, clt->permits_map); 100 } 101 102 /** 103 * rtrs_clt_get_permit() - allocates permit for future RDMA operation 104 * @clt: Current session 105 * @con_type: Type of connection to use with the permit 106 * @can_wait: Wait type 107 * 108 * Description: 109 * Allocates permit for the following RDMA operation. Permit is used 110 * to preallocate all resources and to propagate memory pressure 111 * up earlier. 112 * 113 * Context: 114 * Can sleep if @wait == RTRS_PERMIT_WAIT 115 */ 116 struct rtrs_permit *rtrs_clt_get_permit(struct rtrs_clt_sess *clt, 117 enum rtrs_clt_con_type con_type, 118 enum wait_type can_wait) 119 { 120 struct rtrs_permit *permit; 121 DEFINE_WAIT(wait); 122 123 permit = __rtrs_get_permit(clt, con_type); 124 if (permit || !can_wait) 125 return permit; 126 127 do { 128 prepare_to_wait(&clt->permits_wait, &wait, 129 TASK_UNINTERRUPTIBLE); 130 permit = __rtrs_get_permit(clt, con_type); 131 if (permit) 132 break; 133 134 io_schedule(); 135 } while (1); 136 137 finish_wait(&clt->permits_wait, &wait); 138 139 return permit; 140 } 141 EXPORT_SYMBOL(rtrs_clt_get_permit); 142 143 /** 144 * rtrs_clt_put_permit() - puts allocated permit 145 * @clt: Current session 146 * @permit: Permit to be freed 147 * 148 * Context: 149 * Does not matter 150 */ 151 void rtrs_clt_put_permit(struct rtrs_clt_sess *clt, 152 struct rtrs_permit *permit) 153 { 154 if (WARN_ON(!test_bit(permit->mem_id, clt->permits_map))) 155 return; 156 157 __rtrs_put_permit(clt, permit); 158 159 /* 160 * rtrs_clt_get_permit() adds itself to the &clt->permits_wait list 161 * before calling schedule(). So if rtrs_clt_get_permit() is sleeping 162 * it must have added itself to &clt->permits_wait before 163 * __rtrs_put_permit() finished. 164 * Hence it is safe to guard wake_up() with a waitqueue_active() test. 165 */ 166 if (waitqueue_active(&clt->permits_wait)) 167 wake_up(&clt->permits_wait); 168 } 169 EXPORT_SYMBOL(rtrs_clt_put_permit); 170 171 /** 172 * rtrs_permit_to_clt_con() - returns RDMA connection pointer by the permit 173 * @clt_path: client path pointer 174 * @permit: permit for the allocation of the RDMA buffer 175 * Note: 176 * IO connection starts from 1. 177 * 0 connection is for user messages. 178 */ 179 static 180 struct rtrs_clt_con *rtrs_permit_to_clt_con(struct rtrs_clt_path *clt_path, 181 struct rtrs_permit *permit) 182 { 183 int id = 0; 184 185 if (permit->con_type == RTRS_IO_CON) 186 id = (permit->cpu_id % (clt_path->s.irq_con_num - 1)) + 1; 187 188 return to_clt_con(clt_path->s.con[id]); 189 } 190 191 /** 192 * rtrs_clt_change_state() - change the session state through session state 193 * machine. 194 * 195 * @clt_path: client path to change the state of. 196 * @new_state: state to change to. 197 * 198 * returns true if sess's state is changed to new state, otherwise return false. 199 * 200 * Locks: 201 * state_wq lock must be hold. 202 */ 203 static bool rtrs_clt_change_state(struct rtrs_clt_path *clt_path, 204 enum rtrs_clt_state new_state) 205 { 206 enum rtrs_clt_state old_state; 207 bool changed = false; 208 209 lockdep_assert_held(&clt_path->state_wq.lock); 210 211 old_state = clt_path->state; 212 switch (new_state) { 213 case RTRS_CLT_CONNECTING: 214 switch (old_state) { 215 case RTRS_CLT_RECONNECTING: 216 changed = true; 217 fallthrough; 218 default: 219 break; 220 } 221 break; 222 case RTRS_CLT_RECONNECTING: 223 switch (old_state) { 224 case RTRS_CLT_CONNECTED: 225 case RTRS_CLT_CONNECTING_ERR: 226 case RTRS_CLT_CLOSED: 227 changed = true; 228 fallthrough; 229 default: 230 break; 231 } 232 break; 233 case RTRS_CLT_CONNECTED: 234 switch (old_state) { 235 case RTRS_CLT_CONNECTING: 236 changed = true; 237 fallthrough; 238 default: 239 break; 240 } 241 break; 242 case RTRS_CLT_CONNECTING_ERR: 243 switch (old_state) { 244 case RTRS_CLT_CONNECTING: 245 changed = true; 246 fallthrough; 247 default: 248 break; 249 } 250 break; 251 case RTRS_CLT_CLOSING: 252 switch (old_state) { 253 case RTRS_CLT_CONNECTING: 254 case RTRS_CLT_CONNECTING_ERR: 255 case RTRS_CLT_RECONNECTING: 256 case RTRS_CLT_CONNECTED: 257 changed = true; 258 fallthrough; 259 default: 260 break; 261 } 262 break; 263 case RTRS_CLT_CLOSED: 264 switch (old_state) { 265 case RTRS_CLT_CLOSING: 266 changed = true; 267 fallthrough; 268 default: 269 break; 270 } 271 break; 272 case RTRS_CLT_DEAD: 273 switch (old_state) { 274 case RTRS_CLT_CLOSED: 275 changed = true; 276 fallthrough; 277 default: 278 break; 279 } 280 break; 281 default: 282 break; 283 } 284 if (changed) { 285 clt_path->state = new_state; 286 wake_up_locked(&clt_path->state_wq); 287 } 288 289 return changed; 290 } 291 292 static bool rtrs_clt_change_state_from_to(struct rtrs_clt_path *clt_path, 293 enum rtrs_clt_state old_state, 294 enum rtrs_clt_state new_state) 295 { 296 bool changed = false; 297 298 spin_lock_irq(&clt_path->state_wq.lock); 299 if (clt_path->state == old_state) 300 changed = rtrs_clt_change_state(clt_path, new_state); 301 spin_unlock_irq(&clt_path->state_wq.lock); 302 303 return changed; 304 } 305 306 static void rtrs_clt_stop_and_destroy_conns(struct rtrs_clt_path *clt_path); 307 static void rtrs_rdma_error_recovery(struct rtrs_clt_con *con) 308 { 309 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 310 311 trace_rtrs_rdma_error_recovery(clt_path); 312 313 if (rtrs_clt_change_state_from_to(clt_path, 314 RTRS_CLT_CONNECTED, 315 RTRS_CLT_RECONNECTING)) { 316 queue_work(rtrs_wq, &clt_path->err_recovery_work); 317 } else { 318 /* 319 * Error can happen just on establishing new connection, 320 * so notify waiter with error state, waiter is responsible 321 * for cleaning the rest and reconnect if needed. 322 */ 323 rtrs_clt_change_state_from_to(clt_path, 324 RTRS_CLT_CONNECTING, 325 RTRS_CLT_CONNECTING_ERR); 326 } 327 } 328 329 static void rtrs_clt_fast_reg_done(struct ib_cq *cq, struct ib_wc *wc) 330 { 331 struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context); 332 333 if (wc->status != IB_WC_SUCCESS) { 334 rtrs_err(con->c.path, "Failed IB_WR_REG_MR: %s\n", 335 ib_wc_status_msg(wc->status)); 336 rtrs_rdma_error_recovery(con); 337 } 338 } 339 340 static struct ib_cqe fast_reg_cqe = { 341 .done = rtrs_clt_fast_reg_done 342 }; 343 344 static void complete_rdma_req(struct rtrs_clt_io_req *req, int errno, 345 bool notify, bool can_wait); 346 347 static void rtrs_clt_inv_rkey_done(struct ib_cq *cq, struct ib_wc *wc) 348 { 349 struct rtrs_clt_io_req *req = 350 container_of(wc->wr_cqe, typeof(*req), inv_cqe); 351 struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context); 352 353 if (wc->status != IB_WC_SUCCESS) { 354 rtrs_err(con->c.path, "Failed IB_WR_LOCAL_INV: %s\n", 355 ib_wc_status_msg(wc->status)); 356 rtrs_rdma_error_recovery(con); 357 } 358 req->need_inv = false; 359 if (req->need_inv_comp) 360 complete(&req->inv_comp); 361 else 362 /* Complete request from INV callback */ 363 complete_rdma_req(req, req->inv_errno, true, false); 364 } 365 366 static int rtrs_inv_rkey(struct rtrs_clt_io_req *req) 367 { 368 struct rtrs_clt_con *con = req->con; 369 struct ib_send_wr wr = { 370 .opcode = IB_WR_LOCAL_INV, 371 .wr_cqe = &req->inv_cqe, 372 .send_flags = IB_SEND_SIGNALED, 373 .ex.invalidate_rkey = req->mr->rkey, 374 }; 375 req->inv_cqe.done = rtrs_clt_inv_rkey_done; 376 377 return ib_post_send(con->c.qp, &wr, NULL); 378 } 379 380 static void complete_rdma_req(struct rtrs_clt_io_req *req, int errno, 381 bool notify, bool can_wait) 382 { 383 struct rtrs_clt_con *con = req->con; 384 struct rtrs_clt_path *clt_path; 385 int err; 386 387 if (!req->in_use) 388 return; 389 if (WARN_ON(!req->con)) 390 return; 391 clt_path = to_clt_path(con->c.path); 392 393 if (req->sg_cnt) { 394 if (req->dir == DMA_FROM_DEVICE && req->need_inv) { 395 /* 396 * We are here to invalidate read requests 397 * ourselves. In normal scenario server should 398 * send INV for all read requests, but 399 * we are here, thus two things could happen: 400 * 401 * 1. this is failover, when errno != 0 402 * and can_wait == 1, 403 * 404 * 2. something totally bad happened and 405 * server forgot to send INV, so we 406 * should do that ourselves. 407 */ 408 409 if (can_wait) { 410 req->need_inv_comp = true; 411 } else { 412 /* This should be IO path, so always notify */ 413 WARN_ON(!notify); 414 /* Save errno for INV callback */ 415 req->inv_errno = errno; 416 } 417 418 refcount_inc(&req->ref); 419 err = rtrs_inv_rkey(req); 420 if (err) { 421 rtrs_err(con->c.path, "Send INV WR key=%#x: %d\n", 422 req->mr->rkey, err); 423 } else if (can_wait) { 424 wait_for_completion(&req->inv_comp); 425 } else { 426 /* 427 * Something went wrong, so request will be 428 * completed from INV callback. 429 */ 430 WARN_ON_ONCE(1); 431 432 return; 433 } 434 if (!refcount_dec_and_test(&req->ref)) 435 return; 436 } 437 ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist, 438 req->sg_cnt, req->dir); 439 } 440 if (!refcount_dec_and_test(&req->ref)) 441 return; 442 if (req->mp_policy == MP_POLICY_MIN_INFLIGHT) 443 atomic_dec(&clt_path->stats->inflight); 444 445 req->in_use = false; 446 req->con = NULL; 447 448 if (errno) { 449 rtrs_err_rl(con->c.path, "IO request failed: error=%d path=%s [%s:%u] notify=%d\n", 450 errno, kobject_name(&clt_path->kobj), clt_path->hca_name, 451 clt_path->hca_port, notify); 452 } 453 454 if (notify) 455 req->conf(req->priv, errno); 456 } 457 458 static int rtrs_post_send_rdma(struct rtrs_clt_con *con, 459 struct rtrs_clt_io_req *req, 460 struct rtrs_rbuf *rbuf, u32 off, 461 u32 imm, struct ib_send_wr *wr) 462 { 463 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 464 enum ib_send_flags flags; 465 struct ib_sge sge; 466 467 if (!req->sg_size) { 468 rtrs_wrn(con->c.path, 469 "Doing RDMA Write failed, no data supplied\n"); 470 return -EINVAL; 471 } 472 473 /* user data and user message in the first list element */ 474 sge.addr = req->iu->dma_addr; 475 sge.length = req->sg_size; 476 sge.lkey = clt_path->s.dev->ib_pd->local_dma_lkey; 477 478 /* 479 * From time to time we have to post signalled sends, 480 * or send queue will fill up and only QP reset can help. 481 */ 482 flags = atomic_inc_return(&con->c.wr_cnt) % clt_path->s.signal_interval ? 483 0 : IB_SEND_SIGNALED; 484 485 ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev, 486 req->iu->dma_addr, 487 req->sg_size, DMA_TO_DEVICE); 488 489 return rtrs_iu_post_rdma_write_imm(&con->c, req->iu, &sge, 1, 490 rbuf->rkey, rbuf->addr + off, 491 imm, flags, wr, NULL); 492 } 493 494 static void process_io_rsp(struct rtrs_clt_path *clt_path, u32 msg_id, 495 s16 errno, bool w_inval) 496 { 497 struct rtrs_clt_io_req *req; 498 499 if (WARN_ON(msg_id >= clt_path->queue_depth)) 500 return; 501 502 req = &clt_path->reqs[msg_id]; 503 /* Drop need_inv if server responded with send with invalidation */ 504 req->need_inv &= !w_inval; 505 complete_rdma_req(req, errno, true, false); 506 } 507 508 static void rtrs_clt_recv_done(struct rtrs_clt_con *con, struct ib_wc *wc) 509 { 510 struct rtrs_iu *iu; 511 int err; 512 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 513 514 WARN_ON((clt_path->flags & RTRS_MSG_NEW_RKEY_F) == 0); 515 iu = container_of(wc->wr_cqe, struct rtrs_iu, 516 cqe); 517 err = rtrs_iu_post_recv(&con->c, iu); 518 if (err) { 519 rtrs_err(con->c.path, "post iu failed %d\n", err); 520 rtrs_rdma_error_recovery(con); 521 } 522 } 523 524 static void rtrs_clt_rkey_rsp_done(struct rtrs_clt_con *con, struct ib_wc *wc) 525 { 526 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 527 struct rtrs_msg_rkey_rsp *msg; 528 u32 imm_type, imm_payload; 529 bool w_inval = false; 530 struct rtrs_iu *iu; 531 u32 buf_id; 532 int err; 533 534 WARN_ON((clt_path->flags & RTRS_MSG_NEW_RKEY_F) == 0); 535 536 iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe); 537 538 if (wc->byte_len < sizeof(*msg)) { 539 rtrs_err(con->c.path, "rkey response is malformed: size %d\n", 540 wc->byte_len); 541 goto out; 542 } 543 ib_dma_sync_single_for_cpu(clt_path->s.dev->ib_dev, iu->dma_addr, 544 iu->size, DMA_FROM_DEVICE); 545 msg = iu->buf; 546 if (le16_to_cpu(msg->type) != RTRS_MSG_RKEY_RSP) { 547 rtrs_err(clt_path->clt, 548 "rkey response is malformed: type %d\n", 549 le16_to_cpu(msg->type)); 550 goto out; 551 } 552 buf_id = le16_to_cpu(msg->buf_id); 553 if (WARN_ON(buf_id >= clt_path->queue_depth)) 554 goto out; 555 556 rtrs_from_imm(be32_to_cpu(wc->ex.imm_data), &imm_type, &imm_payload); 557 if (imm_type == RTRS_IO_RSP_IMM || 558 imm_type == RTRS_IO_RSP_W_INV_IMM) { 559 u32 msg_id; 560 561 w_inval = (imm_type == RTRS_IO_RSP_W_INV_IMM); 562 rtrs_from_io_rsp_imm(imm_payload, &msg_id, &err); 563 564 if (WARN_ON(buf_id != msg_id)) 565 goto out; 566 clt_path->rbufs[buf_id].rkey = le32_to_cpu(msg->rkey); 567 process_io_rsp(clt_path, msg_id, err, w_inval); 568 } 569 ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev, iu->dma_addr, 570 iu->size, DMA_FROM_DEVICE); 571 return rtrs_clt_recv_done(con, wc); 572 out: 573 rtrs_rdma_error_recovery(con); 574 } 575 576 static void rtrs_clt_rdma_done(struct ib_cq *cq, struct ib_wc *wc); 577 578 static struct ib_cqe io_comp_cqe = { 579 .done = rtrs_clt_rdma_done 580 }; 581 582 /* 583 * Post x2 empty WRs: first is for this RDMA with IMM, 584 * second is for RECV with INV, which happened earlier. 585 */ 586 static int rtrs_post_recv_empty_x2(struct rtrs_con *con, struct ib_cqe *cqe) 587 { 588 struct ib_recv_wr wr_arr[2], *wr; 589 int i; 590 591 memset(wr_arr, 0, sizeof(wr_arr)); 592 for (i = 0; i < ARRAY_SIZE(wr_arr); i++) { 593 wr = &wr_arr[i]; 594 wr->wr_cqe = cqe; 595 if (i) 596 /* Chain backwards */ 597 wr->next = &wr_arr[i - 1]; 598 } 599 600 return ib_post_recv(con->qp, wr, NULL); 601 } 602 603 static void rtrs_clt_rdma_done(struct ib_cq *cq, struct ib_wc *wc) 604 { 605 struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context); 606 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 607 u32 imm_type, imm_payload; 608 bool w_inval = false; 609 int err; 610 611 if (wc->status != IB_WC_SUCCESS) { 612 if (wc->status != IB_WC_WR_FLUSH_ERR) { 613 rtrs_err(clt_path->clt, "RDMA failed: %s\n", 614 ib_wc_status_msg(wc->status)); 615 rtrs_rdma_error_recovery(con); 616 } 617 return; 618 } 619 rtrs_clt_update_wc_stats(con); 620 621 switch (wc->opcode) { 622 case IB_WC_RECV_RDMA_WITH_IMM: 623 /* 624 * post_recv() RDMA write completions of IO reqs (read/write) 625 * and hb 626 */ 627 if (WARN_ON(wc->wr_cqe->done != rtrs_clt_rdma_done)) 628 return; 629 clt_path->s.hb_missed_cnt = 0; 630 rtrs_from_imm(be32_to_cpu(wc->ex.imm_data), 631 &imm_type, &imm_payload); 632 if (imm_type == RTRS_IO_RSP_IMM || 633 imm_type == RTRS_IO_RSP_W_INV_IMM) { 634 u32 msg_id; 635 636 w_inval = (imm_type == RTRS_IO_RSP_W_INV_IMM); 637 rtrs_from_io_rsp_imm(imm_payload, &msg_id, &err); 638 639 process_io_rsp(clt_path, msg_id, err, w_inval); 640 } else if (imm_type == RTRS_HB_MSG_IMM) { 641 WARN_ON(con->c.cid); 642 rtrs_send_hb_ack(&clt_path->s); 643 if (clt_path->flags & RTRS_MSG_NEW_RKEY_F) 644 return rtrs_clt_recv_done(con, wc); 645 } else if (imm_type == RTRS_HB_ACK_IMM) { 646 WARN_ON(con->c.cid); 647 clt_path->s.hb_cur_latency = 648 ktime_sub(ktime_get(), clt_path->s.hb_last_sent); 649 if (clt_path->flags & RTRS_MSG_NEW_RKEY_F) 650 return rtrs_clt_recv_done(con, wc); 651 } else { 652 rtrs_wrn(con->c.path, "Unknown IMM type %u\n", 653 imm_type); 654 } 655 if (w_inval) 656 /* 657 * Post x2 empty WRs: first is for this RDMA with IMM, 658 * second is for RECV with INV, which happened earlier. 659 */ 660 err = rtrs_post_recv_empty_x2(&con->c, &io_comp_cqe); 661 else 662 err = rtrs_post_recv_empty(&con->c, &io_comp_cqe); 663 if (err) { 664 rtrs_err(con->c.path, "rtrs_post_recv_empty(): %d\n", 665 err); 666 rtrs_rdma_error_recovery(con); 667 } 668 break; 669 case IB_WC_RECV: 670 /* 671 * Key invalidations from server side 672 */ 673 clt_path->s.hb_missed_cnt = 0; 674 WARN_ON(!(wc->wc_flags & IB_WC_WITH_INVALIDATE || 675 wc->wc_flags & IB_WC_WITH_IMM)); 676 WARN_ON(wc->wr_cqe->done != rtrs_clt_rdma_done); 677 if (clt_path->flags & RTRS_MSG_NEW_RKEY_F) { 678 if (wc->wc_flags & IB_WC_WITH_INVALIDATE) 679 return rtrs_clt_recv_done(con, wc); 680 681 return rtrs_clt_rkey_rsp_done(con, wc); 682 } 683 break; 684 case IB_WC_RDMA_WRITE: 685 /* 686 * post_send() RDMA write completions of IO reqs (read/write) 687 * and hb. 688 */ 689 break; 690 691 default: 692 rtrs_wrn(clt_path->clt, "Unexpected WC type: %d\n", wc->opcode); 693 return; 694 } 695 } 696 697 static int post_recv_io(struct rtrs_clt_con *con, size_t q_size) 698 { 699 int err, i; 700 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 701 702 for (i = 0; i < q_size; i++) { 703 if (clt_path->flags & RTRS_MSG_NEW_RKEY_F) { 704 struct rtrs_iu *iu = &con->rsp_ius[i]; 705 706 err = rtrs_iu_post_recv(&con->c, iu); 707 } else { 708 err = rtrs_post_recv_empty(&con->c, &io_comp_cqe); 709 } 710 if (err) 711 return err; 712 } 713 714 return 0; 715 } 716 717 static int post_recv_path(struct rtrs_clt_path *clt_path) 718 { 719 size_t q_size = 0; 720 int err, cid; 721 722 for (cid = 0; cid < clt_path->s.con_num; cid++) { 723 if (cid == 0) 724 q_size = SERVICE_CON_QUEUE_DEPTH; 725 else 726 q_size = clt_path->queue_depth; 727 728 /* 729 * x2 for RDMA read responses + FR key invalidations, 730 * RDMA writes do not require any FR registrations. 731 */ 732 q_size *= 2; 733 734 err = post_recv_io(to_clt_con(clt_path->s.con[cid]), q_size); 735 if (err) { 736 rtrs_err(clt_path->clt, "post_recv_io(), err: %d\n", 737 err); 738 return err; 739 } 740 } 741 742 return 0; 743 } 744 745 struct path_it { 746 int i; 747 struct list_head skip_list; 748 struct rtrs_clt_sess *clt; 749 struct rtrs_clt_path *(*next_path)(struct path_it *it); 750 }; 751 752 /* 753 * rtrs_clt_get_next_path_or_null - get clt path from the list or return NULL 754 * @head: the head for the list. 755 * @clt_path: The element to take the next clt_path from. 756 * 757 * Next clt path returned in round-robin fashion, i.e. head will be skipped, 758 * but if list is observed as empty, NULL will be returned. 759 * 760 * This function may safely run concurrently with the _rcu list-mutation 761 * primitives such as list_add_rcu() as long as it's guarded by rcu_read_lock(). 762 */ 763 static inline struct rtrs_clt_path * 764 rtrs_clt_get_next_path_or_null(struct list_head *head, struct rtrs_clt_path *clt_path) 765 { 766 return list_next_or_null_rcu(head, &clt_path->s.entry, typeof(*clt_path), s.entry) ?: 767 list_next_or_null_rcu(head, 768 READ_ONCE((&clt_path->s.entry)->next), 769 typeof(*clt_path), s.entry); 770 } 771 772 /** 773 * get_next_path_rr() - Returns path in round-robin fashion. 774 * @it: the path pointer 775 * 776 * Related to @MP_POLICY_RR 777 * 778 * Locks: 779 * rcu_read_lock() must be hold. 780 */ 781 static struct rtrs_clt_path *get_next_path_rr(struct path_it *it) 782 { 783 struct rtrs_clt_path __rcu **ppcpu_path; 784 struct rtrs_clt_path *path; 785 struct rtrs_clt_sess *clt; 786 787 clt = it->clt; 788 789 /* 790 * Here we use two RCU objects: @paths_list and @pcpu_path 791 * pointer. See rtrs_clt_remove_path_from_arr() for details 792 * how that is handled. 793 */ 794 795 ppcpu_path = this_cpu_ptr(clt->pcpu_path); 796 path = rcu_dereference(*ppcpu_path); 797 if (!path) 798 path = list_first_or_null_rcu(&clt->paths_list, 799 typeof(*path), s.entry); 800 else 801 path = rtrs_clt_get_next_path_or_null(&clt->paths_list, path); 802 803 rcu_assign_pointer(*ppcpu_path, path); 804 805 return path; 806 } 807 808 /** 809 * get_next_path_min_inflight() - Returns path with minimal inflight count. 810 * @it: the path pointer 811 * 812 * Related to @MP_POLICY_MIN_INFLIGHT 813 * 814 * Locks: 815 * rcu_read_lock() must be hold. 816 */ 817 static struct rtrs_clt_path *get_next_path_min_inflight(struct path_it *it) 818 { 819 struct rtrs_clt_path *min_path = NULL; 820 struct rtrs_clt_sess *clt = it->clt; 821 struct rtrs_clt_path *clt_path; 822 int min_inflight = INT_MAX; 823 int inflight; 824 825 list_for_each_entry_rcu(clt_path, &clt->paths_list, s.entry) { 826 if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED) 827 continue; 828 829 if (!list_empty(raw_cpu_ptr(clt_path->mp_skip_entry))) 830 continue; 831 832 inflight = atomic_read(&clt_path->stats->inflight); 833 834 if (inflight < min_inflight) { 835 min_inflight = inflight; 836 min_path = clt_path; 837 } 838 } 839 840 /* 841 * add the path to the skip list, so that next time we can get 842 * a different one 843 */ 844 if (min_path) 845 list_add(raw_cpu_ptr(min_path->mp_skip_entry), &it->skip_list); 846 847 return min_path; 848 } 849 850 /** 851 * get_next_path_min_latency() - Returns path with minimal latency. 852 * @it: the path pointer 853 * 854 * Return: a path with the lowest latency or NULL if all paths are tried 855 * 856 * Locks: 857 * rcu_read_lock() must be hold. 858 * 859 * Related to @MP_POLICY_MIN_LATENCY 860 * 861 * This DOES skip an already-tried path. 862 * There is a skip-list to skip a path if the path has tried but failed. 863 * It will try the minimum latency path and then the second minimum latency 864 * path and so on. Finally it will return NULL if all paths are tried. 865 * Therefore the caller MUST check the returned 866 * path is NULL and trigger the IO error. 867 */ 868 static struct rtrs_clt_path *get_next_path_min_latency(struct path_it *it) 869 { 870 struct rtrs_clt_path *min_path = NULL; 871 struct rtrs_clt_sess *clt = it->clt; 872 struct rtrs_clt_path *clt_path; 873 ktime_t min_latency = KTIME_MAX; 874 ktime_t latency; 875 876 list_for_each_entry_rcu(clt_path, &clt->paths_list, s.entry) { 877 if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED) 878 continue; 879 880 if (!list_empty(raw_cpu_ptr(clt_path->mp_skip_entry))) 881 continue; 882 883 latency = clt_path->s.hb_cur_latency; 884 885 if (latency < min_latency) { 886 min_latency = latency; 887 min_path = clt_path; 888 } 889 } 890 891 /* 892 * add the path to the skip list, so that next time we can get 893 * a different one 894 */ 895 if (min_path) 896 list_add(raw_cpu_ptr(min_path->mp_skip_entry), &it->skip_list); 897 898 return min_path; 899 } 900 901 static inline void path_it_init(struct path_it *it, struct rtrs_clt_sess *clt) 902 { 903 INIT_LIST_HEAD(&it->skip_list); 904 it->clt = clt; 905 it->i = 0; 906 907 if (clt->mp_policy == MP_POLICY_RR) 908 it->next_path = get_next_path_rr; 909 else if (clt->mp_policy == MP_POLICY_MIN_INFLIGHT) 910 it->next_path = get_next_path_min_inflight; 911 else 912 it->next_path = get_next_path_min_latency; 913 } 914 915 static inline void path_it_deinit(struct path_it *it) 916 { 917 struct list_head *skip, *tmp; 918 /* 919 * The skip_list is used only for the MIN_INFLIGHT and MIN_LATENCY policies. 920 * We need to remove paths from it, so that next IO can insert 921 * paths (->mp_skip_entry) into a skip_list again. 922 */ 923 list_for_each_safe(skip, tmp, &it->skip_list) 924 list_del_init(skip); 925 } 926 927 /** 928 * rtrs_clt_init_req() - Initialize an rtrs_clt_io_req holding information 929 * about an inflight IO. 930 * The user buffer holding user control message (not data) is copied into 931 * the corresponding buffer of rtrs_iu (req->iu->buf), which later on will 932 * also hold the control message of rtrs. 933 * @req: an io request holding information about IO. 934 * @clt_path: client path 935 * @conf: conformation callback function to notify upper layer. 936 * @permit: permit for allocation of RDMA remote buffer 937 * @priv: private pointer 938 * @vec: kernel vector containing control message 939 * @usr_len: length of the user message 940 * @sg: scater list for IO data 941 * @sg_cnt: number of scater list entries 942 * @data_len: length of the IO data 943 * @dir: direction of the IO. 944 */ 945 static void rtrs_clt_init_req(struct rtrs_clt_io_req *req, 946 struct rtrs_clt_path *clt_path, 947 void (*conf)(void *priv, int errno), 948 struct rtrs_permit *permit, void *priv, 949 const struct kvec *vec, size_t usr_len, 950 struct scatterlist *sg, size_t sg_cnt, 951 size_t data_len, int dir) 952 { 953 struct iov_iter iter; 954 size_t len; 955 956 req->permit = permit; 957 req->in_use = true; 958 req->usr_len = usr_len; 959 req->data_len = data_len; 960 req->sglist = sg; 961 req->sg_cnt = sg_cnt; 962 req->priv = priv; 963 req->dir = dir; 964 req->con = rtrs_permit_to_clt_con(clt_path, permit); 965 req->conf = conf; 966 req->need_inv = false; 967 req->need_inv_comp = false; 968 req->inv_errno = 0; 969 refcount_set(&req->ref, 1); 970 req->mp_policy = clt_path->clt->mp_policy; 971 972 iov_iter_kvec(&iter, ITER_SOURCE, vec, 1, usr_len); 973 len = _copy_from_iter(req->iu->buf, usr_len, &iter); 974 WARN_ON(len != usr_len); 975 976 reinit_completion(&req->inv_comp); 977 } 978 979 static struct rtrs_clt_io_req * 980 rtrs_clt_get_req(struct rtrs_clt_path *clt_path, 981 void (*conf)(void *priv, int errno), 982 struct rtrs_permit *permit, void *priv, 983 const struct kvec *vec, size_t usr_len, 984 struct scatterlist *sg, size_t sg_cnt, 985 size_t data_len, int dir) 986 { 987 struct rtrs_clt_io_req *req; 988 989 req = &clt_path->reqs[permit->mem_id]; 990 rtrs_clt_init_req(req, clt_path, conf, permit, priv, vec, usr_len, 991 sg, sg_cnt, data_len, dir); 992 return req; 993 } 994 995 static struct rtrs_clt_io_req * 996 rtrs_clt_get_copy_req(struct rtrs_clt_path *alive_path, 997 struct rtrs_clt_io_req *fail_req) 998 { 999 struct rtrs_clt_io_req *req; 1000 struct kvec vec = { 1001 .iov_base = fail_req->iu->buf, 1002 .iov_len = fail_req->usr_len 1003 }; 1004 1005 req = &alive_path->reqs[fail_req->permit->mem_id]; 1006 rtrs_clt_init_req(req, alive_path, fail_req->conf, fail_req->permit, 1007 fail_req->priv, &vec, fail_req->usr_len, 1008 fail_req->sglist, fail_req->sg_cnt, 1009 fail_req->data_len, fail_req->dir); 1010 return req; 1011 } 1012 1013 static int rtrs_post_rdma_write_sg(struct rtrs_clt_con *con, 1014 struct rtrs_clt_io_req *req, 1015 struct rtrs_rbuf *rbuf, bool fr_en, 1016 u32 count, u32 size, u32 imm, 1017 struct ib_send_wr *wr, 1018 struct ib_send_wr *tail) 1019 { 1020 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 1021 struct ib_sge *sge = req->sge; 1022 enum ib_send_flags flags; 1023 struct scatterlist *sg; 1024 size_t num_sge; 1025 int i; 1026 struct ib_send_wr *ptail = NULL; 1027 1028 if (fr_en) { 1029 i = 0; 1030 sge[i].addr = req->mr->iova; 1031 sge[i].length = req->mr->length; 1032 sge[i].lkey = req->mr->lkey; 1033 i++; 1034 num_sge = 2; 1035 ptail = tail; 1036 } else { 1037 for_each_sg(req->sglist, sg, count, i) { 1038 sge[i].addr = sg_dma_address(sg); 1039 sge[i].length = sg_dma_len(sg); 1040 sge[i].lkey = clt_path->s.dev->ib_pd->local_dma_lkey; 1041 } 1042 num_sge = 1 + count; 1043 } 1044 sge[i].addr = req->iu->dma_addr; 1045 sge[i].length = size; 1046 sge[i].lkey = clt_path->s.dev->ib_pd->local_dma_lkey; 1047 1048 /* 1049 * From time to time we have to post signalled sends, 1050 * or send queue will fill up and only QP reset can help. 1051 */ 1052 flags = atomic_inc_return(&con->c.wr_cnt) % clt_path->s.signal_interval ? 1053 0 : IB_SEND_SIGNALED; 1054 1055 ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev, 1056 req->iu->dma_addr, 1057 size, DMA_TO_DEVICE); 1058 1059 return rtrs_iu_post_rdma_write_imm(&con->c, req->iu, sge, num_sge, 1060 rbuf->rkey, rbuf->addr, imm, 1061 flags, wr, ptail); 1062 } 1063 1064 static int rtrs_map_sg_fr(struct rtrs_clt_io_req *req, size_t count) 1065 { 1066 int nr; 1067 1068 /* Align the MR to a 4K page size to match the block virt boundary */ 1069 nr = ib_map_mr_sg(req->mr, req->sglist, count, NULL, SZ_4K); 1070 if (nr != count) 1071 return nr < 0 ? nr : -EINVAL; 1072 ib_update_fast_reg_key(req->mr, ib_inc_rkey(req->mr->rkey)); 1073 1074 return nr; 1075 } 1076 1077 static int rtrs_clt_write_req(struct rtrs_clt_io_req *req) 1078 { 1079 struct rtrs_clt_con *con = req->con; 1080 struct rtrs_path *s = con->c.path; 1081 struct rtrs_clt_path *clt_path = to_clt_path(s); 1082 struct rtrs_msg_rdma_write *msg; 1083 1084 struct rtrs_rbuf *rbuf; 1085 int ret, count = 0; 1086 u32 imm, buf_id; 1087 struct ib_reg_wr rwr; 1088 struct ib_send_wr inv_wr; 1089 struct ib_send_wr *wr = NULL; 1090 bool fr_en = false; 1091 1092 const size_t tsize = sizeof(*msg) + req->data_len + req->usr_len; 1093 1094 if (tsize > clt_path->chunk_size) { 1095 rtrs_wrn(s, "Write request failed, size too big %zu > %d\n", 1096 tsize, clt_path->chunk_size); 1097 return -EMSGSIZE; 1098 } 1099 if (req->sg_cnt) { 1100 count = ib_dma_map_sg(clt_path->s.dev->ib_dev, req->sglist, 1101 req->sg_cnt, req->dir); 1102 if (!count) { 1103 rtrs_wrn(s, "Write request failed, map failed\n"); 1104 return -EINVAL; 1105 } 1106 } 1107 /* put rtrs msg after sg and user message */ 1108 msg = req->iu->buf + req->usr_len; 1109 msg->type = cpu_to_le16(RTRS_MSG_WRITE); 1110 msg->usr_len = cpu_to_le16(req->usr_len); 1111 1112 /* rtrs message on server side will be after user data and message */ 1113 imm = req->permit->mem_off + req->data_len + req->usr_len; 1114 imm = rtrs_to_io_req_imm(imm); 1115 buf_id = req->permit->mem_id; 1116 req->sg_size = tsize; 1117 rbuf = &clt_path->rbufs[buf_id]; 1118 1119 if (count) { 1120 ret = rtrs_map_sg_fr(req, count); 1121 if (ret < 0) { 1122 rtrs_err_rl(s, 1123 "Write request failed, failed to map fast reg. data, err: %d\n", 1124 ret); 1125 ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist, 1126 req->sg_cnt, req->dir); 1127 return ret; 1128 } 1129 inv_wr = (struct ib_send_wr) { 1130 .opcode = IB_WR_LOCAL_INV, 1131 .wr_cqe = &req->inv_cqe, 1132 .send_flags = IB_SEND_SIGNALED, 1133 .ex.invalidate_rkey = req->mr->rkey, 1134 }; 1135 req->inv_cqe.done = rtrs_clt_inv_rkey_done; 1136 rwr = (struct ib_reg_wr) { 1137 .wr.opcode = IB_WR_REG_MR, 1138 .wr.wr_cqe = &fast_reg_cqe, 1139 .mr = req->mr, 1140 .key = req->mr->rkey, 1141 .access = (IB_ACCESS_LOCAL_WRITE), 1142 }; 1143 wr = &rwr.wr; 1144 fr_en = true; 1145 refcount_inc(&req->ref); 1146 } 1147 /* 1148 * Update stats now, after request is successfully sent it is not 1149 * safe anymore to touch it. 1150 */ 1151 rtrs_clt_update_all_stats(req, WRITE); 1152 1153 ret = rtrs_post_rdma_write_sg(req->con, req, rbuf, fr_en, count, 1154 req->usr_len + sizeof(*msg), 1155 imm, wr, &inv_wr); 1156 if (ret) { 1157 rtrs_err_rl(s, 1158 "Write request failed: error=%d path=%s [%s:%u]\n", 1159 ret, kobject_name(&clt_path->kobj), clt_path->hca_name, 1160 clt_path->hca_port); 1161 if (req->mp_policy == MP_POLICY_MIN_INFLIGHT) 1162 atomic_dec(&clt_path->stats->inflight); 1163 if (req->sg_cnt) 1164 ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist, 1165 req->sg_cnt, req->dir); 1166 } 1167 1168 return ret; 1169 } 1170 1171 static int rtrs_clt_read_req(struct rtrs_clt_io_req *req) 1172 { 1173 struct rtrs_clt_con *con = req->con; 1174 struct rtrs_path *s = con->c.path; 1175 struct rtrs_clt_path *clt_path = to_clt_path(s); 1176 struct rtrs_msg_rdma_read *msg; 1177 struct rtrs_ib_dev *dev = clt_path->s.dev; 1178 1179 struct ib_reg_wr rwr; 1180 struct ib_send_wr *wr = NULL; 1181 1182 int ret, count = 0; 1183 u32 imm, buf_id; 1184 1185 const size_t tsize = sizeof(*msg) + req->data_len + req->usr_len; 1186 1187 if (tsize > clt_path->chunk_size) { 1188 rtrs_wrn(s, 1189 "Read request failed, message size is %zu, bigger than CHUNK_SIZE %d\n", 1190 tsize, clt_path->chunk_size); 1191 return -EMSGSIZE; 1192 } 1193 1194 if (req->sg_cnt) { 1195 count = ib_dma_map_sg(dev->ib_dev, req->sglist, req->sg_cnt, 1196 req->dir); 1197 if (!count) { 1198 rtrs_wrn(s, 1199 "Read request failed, dma map failed\n"); 1200 return -EINVAL; 1201 } 1202 } 1203 /* put our message into req->buf after user message*/ 1204 msg = req->iu->buf + req->usr_len; 1205 msg->type = cpu_to_le16(RTRS_MSG_READ); 1206 msg->usr_len = cpu_to_le16(req->usr_len); 1207 1208 if (count) { 1209 ret = rtrs_map_sg_fr(req, count); 1210 if (ret < 0) { 1211 rtrs_err_rl(s, 1212 "Read request failed, failed to map fast reg. data, err: %d\n", 1213 ret); 1214 ib_dma_unmap_sg(dev->ib_dev, req->sglist, req->sg_cnt, 1215 req->dir); 1216 return ret; 1217 } 1218 rwr = (struct ib_reg_wr) { 1219 .wr.opcode = IB_WR_REG_MR, 1220 .wr.wr_cqe = &fast_reg_cqe, 1221 .mr = req->mr, 1222 .key = req->mr->rkey, 1223 .access = (IB_ACCESS_LOCAL_WRITE | 1224 IB_ACCESS_REMOTE_WRITE), 1225 }; 1226 wr = &rwr.wr; 1227 1228 msg->sg_cnt = cpu_to_le16(1); 1229 msg->flags = cpu_to_le16(RTRS_MSG_NEED_INVAL_F); 1230 1231 msg->desc[0].addr = cpu_to_le64(req->mr->iova); 1232 msg->desc[0].key = cpu_to_le32(req->mr->rkey); 1233 msg->desc[0].len = cpu_to_le32(req->mr->length); 1234 1235 /* Further invalidation is required */ 1236 req->need_inv = !!RTRS_MSG_NEED_INVAL_F; 1237 1238 } else { 1239 msg->sg_cnt = 0; 1240 msg->flags = 0; 1241 } 1242 /* 1243 * rtrs message will be after the space reserved for disk data and 1244 * user message 1245 */ 1246 imm = req->permit->mem_off + req->data_len + req->usr_len; 1247 imm = rtrs_to_io_req_imm(imm); 1248 buf_id = req->permit->mem_id; 1249 1250 req->sg_size = sizeof(*msg); 1251 req->sg_size += le16_to_cpu(msg->sg_cnt) * sizeof(struct rtrs_sg_desc); 1252 req->sg_size += req->usr_len; 1253 1254 /* 1255 * Update stats now, after request is successfully sent it is not 1256 * safe anymore to touch it. 1257 */ 1258 rtrs_clt_update_all_stats(req, READ); 1259 1260 ret = rtrs_post_send_rdma(req->con, req, &clt_path->rbufs[buf_id], 1261 req->data_len, imm, wr); 1262 if (ret) { 1263 rtrs_err_rl(s, 1264 "Read request failed: error=%d path=%s [%s:%u]\n", 1265 ret, kobject_name(&clt_path->kobj), clt_path->hca_name, 1266 clt_path->hca_port); 1267 if (req->mp_policy == MP_POLICY_MIN_INFLIGHT) 1268 atomic_dec(&clt_path->stats->inflight); 1269 req->need_inv = false; 1270 if (req->sg_cnt) 1271 ib_dma_unmap_sg(dev->ib_dev, req->sglist, 1272 req->sg_cnt, req->dir); 1273 } 1274 1275 return ret; 1276 } 1277 1278 /** 1279 * rtrs_clt_failover_req() - Try to find an active path for a failed request 1280 * @clt: clt context 1281 * @fail_req: a failed io request. 1282 */ 1283 static int rtrs_clt_failover_req(struct rtrs_clt_sess *clt, 1284 struct rtrs_clt_io_req *fail_req) 1285 { 1286 struct rtrs_clt_path *alive_path; 1287 struct rtrs_clt_io_req *req; 1288 int err = -ECONNABORTED; 1289 struct path_it it; 1290 1291 rcu_read_lock(); 1292 for (path_it_init(&it, clt); 1293 (alive_path = it.next_path(&it)) && it.i < it.clt->paths_num; 1294 it.i++) { 1295 if (READ_ONCE(alive_path->state) != RTRS_CLT_CONNECTED) 1296 continue; 1297 req = rtrs_clt_get_copy_req(alive_path, fail_req); 1298 if (req->dir == DMA_TO_DEVICE) 1299 err = rtrs_clt_write_req(req); 1300 else 1301 err = rtrs_clt_read_req(req); 1302 if (err) { 1303 req->in_use = false; 1304 continue; 1305 } 1306 /* Success path */ 1307 rtrs_clt_inc_failover_cnt(alive_path->stats); 1308 break; 1309 } 1310 path_it_deinit(&it); 1311 rcu_read_unlock(); 1312 1313 return err; 1314 } 1315 1316 static void fail_all_outstanding_reqs(struct rtrs_clt_path *clt_path) 1317 { 1318 struct rtrs_clt_sess *clt = clt_path->clt; 1319 struct rtrs_clt_io_req *req; 1320 int i, err; 1321 1322 if (!clt_path->reqs) 1323 return; 1324 for (i = 0; i < clt_path->queue_depth; ++i) { 1325 req = &clt_path->reqs[i]; 1326 if (!req->in_use) 1327 continue; 1328 1329 /* 1330 * Safely (without notification) complete failed request. 1331 * After completion this request is still useble and can 1332 * be failovered to another path. 1333 */ 1334 complete_rdma_req(req, -ECONNABORTED, false, true); 1335 1336 err = rtrs_clt_failover_req(clt, req); 1337 if (err) 1338 /* Failover failed, notify anyway */ 1339 req->conf(req->priv, err); 1340 } 1341 } 1342 1343 static void free_path_reqs(struct rtrs_clt_path *clt_path) 1344 { 1345 struct rtrs_clt_io_req *req; 1346 int i; 1347 1348 if (!clt_path->reqs) 1349 return; 1350 for (i = 0; i < clt_path->queue_depth; ++i) { 1351 req = &clt_path->reqs[i]; 1352 if (req->mr) 1353 ib_dereg_mr(req->mr); 1354 kfree(req->sge); 1355 rtrs_iu_free(req->iu, clt_path->s.dev->ib_dev, 1); 1356 } 1357 kfree(clt_path->reqs); 1358 clt_path->reqs = NULL; 1359 } 1360 1361 static int alloc_path_reqs(struct rtrs_clt_path *clt_path) 1362 { 1363 struct rtrs_clt_io_req *req; 1364 int i, err = -ENOMEM; 1365 1366 clt_path->reqs = kcalloc(clt_path->queue_depth, 1367 sizeof(*clt_path->reqs), 1368 GFP_KERNEL); 1369 if (!clt_path->reqs) 1370 return -ENOMEM; 1371 1372 for (i = 0; i < clt_path->queue_depth; ++i) { 1373 req = &clt_path->reqs[i]; 1374 req->iu = rtrs_iu_alloc(1, clt_path->max_hdr_size, GFP_KERNEL, 1375 clt_path->s.dev->ib_dev, 1376 DMA_TO_DEVICE, 1377 rtrs_clt_rdma_done); 1378 if (!req->iu) 1379 goto out; 1380 1381 req->sge = kcalloc(2, sizeof(*req->sge), GFP_KERNEL); 1382 if (!req->sge) 1383 goto out; 1384 1385 req->mr = ib_alloc_mr(clt_path->s.dev->ib_pd, 1386 IB_MR_TYPE_MEM_REG, 1387 clt_path->max_pages_per_mr); 1388 if (IS_ERR(req->mr)) { 1389 err = PTR_ERR(req->mr); 1390 req->mr = NULL; 1391 pr_err("Failed to alloc clt_path->max_pages_per_mr %d\n", 1392 clt_path->max_pages_per_mr); 1393 goto out; 1394 } 1395 1396 init_completion(&req->inv_comp); 1397 } 1398 1399 return 0; 1400 1401 out: 1402 free_path_reqs(clt_path); 1403 1404 return err; 1405 } 1406 1407 static int alloc_permits(struct rtrs_clt_sess *clt) 1408 { 1409 unsigned int chunk_bits; 1410 int err, i; 1411 1412 clt->permits_map = bitmap_zalloc(clt->queue_depth, GFP_KERNEL); 1413 if (!clt->permits_map) { 1414 err = -ENOMEM; 1415 goto out_err; 1416 } 1417 clt->permits = kcalloc(clt->queue_depth, permit_size(clt), GFP_KERNEL); 1418 if (!clt->permits) { 1419 err = -ENOMEM; 1420 goto err_map; 1421 } 1422 chunk_bits = ilog2(clt->queue_depth - 1) + 1; 1423 for (i = 0; i < clt->queue_depth; i++) { 1424 struct rtrs_permit *permit; 1425 1426 permit = get_permit(clt, i); 1427 permit->mem_id = i; 1428 permit->mem_off = i << (MAX_IMM_PAYL_BITS - chunk_bits); 1429 } 1430 1431 return 0; 1432 1433 err_map: 1434 bitmap_free(clt->permits_map); 1435 clt->permits_map = NULL; 1436 out_err: 1437 return err; 1438 } 1439 1440 static void free_permits(struct rtrs_clt_sess *clt) 1441 { 1442 if (clt->permits_map) 1443 wait_event(clt->permits_wait, 1444 bitmap_empty(clt->permits_map, clt->queue_depth)); 1445 1446 bitmap_free(clt->permits_map); 1447 clt->permits_map = NULL; 1448 kfree(clt->permits); 1449 clt->permits = NULL; 1450 } 1451 1452 static void query_fast_reg_mode(struct rtrs_clt_path *clt_path) 1453 { 1454 struct ib_device *ib_dev; 1455 u64 max_pages_per_mr; 1456 int mr_page_shift; 1457 1458 ib_dev = clt_path->s.dev->ib_dev; 1459 1460 /* 1461 * Use the smallest page size supported by the HCA, down to a 1462 * minimum of 4096 bytes. We're unlikely to build large sglists 1463 * out of smaller entries. 1464 */ 1465 mr_page_shift = max(12, ffs(ib_dev->attrs.page_size_cap) - 1); 1466 max_pages_per_mr = ib_dev->attrs.max_mr_size; 1467 do_div(max_pages_per_mr, (1ull << mr_page_shift)); 1468 clt_path->max_pages_per_mr = 1469 min3(clt_path->max_pages_per_mr, (u32)max_pages_per_mr, 1470 ib_dev->attrs.max_fast_reg_page_list_len); 1471 clt_path->clt->max_segments = 1472 min(clt_path->max_pages_per_mr, clt_path->clt->max_segments); 1473 } 1474 1475 static bool rtrs_clt_change_state_get_old(struct rtrs_clt_path *clt_path, 1476 enum rtrs_clt_state new_state, 1477 enum rtrs_clt_state *old_state) 1478 { 1479 bool changed; 1480 1481 spin_lock_irq(&clt_path->state_wq.lock); 1482 if (old_state) 1483 *old_state = clt_path->state; 1484 changed = rtrs_clt_change_state(clt_path, new_state); 1485 spin_unlock_irq(&clt_path->state_wq.lock); 1486 1487 return changed; 1488 } 1489 1490 static void rtrs_clt_hb_err_handler(struct rtrs_con *c) 1491 { 1492 struct rtrs_clt_con *con = container_of(c, typeof(*con), c); 1493 1494 rtrs_rdma_error_recovery(con); 1495 } 1496 1497 static void rtrs_clt_init_hb(struct rtrs_clt_path *clt_path) 1498 { 1499 rtrs_init_hb(&clt_path->s, &io_comp_cqe, 1500 RTRS_HB_INTERVAL_MS, 1501 RTRS_HB_MISSED_MAX, 1502 rtrs_clt_hb_err_handler, 1503 rtrs_wq); 1504 } 1505 1506 static void rtrs_clt_reconnect_work(struct work_struct *work); 1507 static void rtrs_clt_close_work(struct work_struct *work); 1508 1509 static void rtrs_clt_err_recovery_work(struct work_struct *work) 1510 { 1511 struct rtrs_clt_path *clt_path; 1512 struct rtrs_clt_sess *clt; 1513 int delay_ms; 1514 1515 clt_path = container_of(work, struct rtrs_clt_path, err_recovery_work); 1516 clt = clt_path->clt; 1517 delay_ms = clt->reconnect_delay_sec * 1000; 1518 rtrs_clt_stop_and_destroy_conns(clt_path); 1519 queue_delayed_work(rtrs_wq, &clt_path->reconnect_dwork, 1520 msecs_to_jiffies(delay_ms + 1521 get_random_u32_below(RTRS_RECONNECT_SEED))); 1522 } 1523 1524 static struct rtrs_clt_path *alloc_path(struct rtrs_clt_sess *clt, 1525 const struct rtrs_addr *path, 1526 size_t con_num, u32 nr_poll_queues) 1527 { 1528 struct rtrs_clt_path *clt_path; 1529 int err = -ENOMEM; 1530 int cpu; 1531 size_t total_con; 1532 1533 clt_path = kzalloc(sizeof(*clt_path), GFP_KERNEL); 1534 if (!clt_path) 1535 goto err; 1536 1537 /* 1538 * irqmode and poll 1539 * +1: Extra connection for user messages 1540 */ 1541 total_con = con_num + nr_poll_queues + 1; 1542 clt_path->s.con = kcalloc(total_con, sizeof(*clt_path->s.con), 1543 GFP_KERNEL); 1544 if (!clt_path->s.con) 1545 goto err_free_path; 1546 1547 clt_path->s.con_num = total_con; 1548 clt_path->s.irq_con_num = con_num + 1; 1549 1550 clt_path->stats = kzalloc(sizeof(*clt_path->stats), GFP_KERNEL); 1551 if (!clt_path->stats) 1552 goto err_free_con; 1553 1554 mutex_init(&clt_path->init_mutex); 1555 uuid_gen(&clt_path->s.uuid); 1556 memcpy(&clt_path->s.dst_addr, path->dst, 1557 rdma_addr_size((struct sockaddr *)path->dst)); 1558 1559 /* 1560 * rdma_resolve_addr() passes src_addr to cma_bind_addr, which 1561 * checks the sa_family to be non-zero. If user passed src_addr=NULL 1562 * the sess->src_addr will contain only zeros, which is then fine. 1563 */ 1564 if (path->src) 1565 memcpy(&clt_path->s.src_addr, path->src, 1566 rdma_addr_size((struct sockaddr *)path->src)); 1567 strscpy(clt_path->s.sessname, clt->sessname, 1568 sizeof(clt_path->s.sessname)); 1569 clt_path->clt = clt; 1570 clt_path->max_pages_per_mr = RTRS_MAX_SEGMENTS; 1571 init_waitqueue_head(&clt_path->state_wq); 1572 clt_path->state = RTRS_CLT_CONNECTING; 1573 atomic_set(&clt_path->connected_cnt, 0); 1574 INIT_WORK(&clt_path->close_work, rtrs_clt_close_work); 1575 INIT_WORK(&clt_path->err_recovery_work, rtrs_clt_err_recovery_work); 1576 INIT_DELAYED_WORK(&clt_path->reconnect_dwork, rtrs_clt_reconnect_work); 1577 rtrs_clt_init_hb(clt_path); 1578 1579 clt_path->mp_skip_entry = alloc_percpu(typeof(*clt_path->mp_skip_entry)); 1580 if (!clt_path->mp_skip_entry) 1581 goto err_free_stats; 1582 1583 for_each_possible_cpu(cpu) 1584 INIT_LIST_HEAD(per_cpu_ptr(clt_path->mp_skip_entry, cpu)); 1585 1586 err = rtrs_clt_init_stats(clt_path->stats); 1587 if (err) 1588 goto err_free_percpu; 1589 1590 return clt_path; 1591 1592 err_free_percpu: 1593 free_percpu(clt_path->mp_skip_entry); 1594 err_free_stats: 1595 kfree(clt_path->stats); 1596 err_free_con: 1597 kfree(clt_path->s.con); 1598 err_free_path: 1599 kfree(clt_path); 1600 err: 1601 return ERR_PTR(err); 1602 } 1603 1604 void free_path(struct rtrs_clt_path *clt_path) 1605 { 1606 free_percpu(clt_path->mp_skip_entry); 1607 mutex_destroy(&clt_path->init_mutex); 1608 kfree(clt_path->s.con); 1609 kfree(clt_path->rbufs); 1610 kfree(clt_path); 1611 } 1612 1613 static int create_con(struct rtrs_clt_path *clt_path, unsigned int cid) 1614 { 1615 struct rtrs_clt_con *con; 1616 1617 con = kzalloc(sizeof(*con), GFP_KERNEL); 1618 if (!con) 1619 return -ENOMEM; 1620 1621 /* Map first two connections to the first CPU */ 1622 con->cpu = (cid ? cid - 1 : 0) % nr_cpu_ids; 1623 con->c.cid = cid; 1624 con->c.path = &clt_path->s; 1625 /* Align with srv, init as 1 */ 1626 atomic_set(&con->c.wr_cnt, 1); 1627 mutex_init(&con->con_mutex); 1628 1629 clt_path->s.con[cid] = &con->c; 1630 1631 return 0; 1632 } 1633 1634 static void destroy_con(struct rtrs_clt_con *con) 1635 { 1636 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 1637 1638 clt_path->s.con[con->c.cid] = NULL; 1639 mutex_destroy(&con->con_mutex); 1640 kfree(con); 1641 } 1642 1643 static int create_con_cq_qp(struct rtrs_clt_con *con) 1644 { 1645 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 1646 u32 max_send_wr, max_recv_wr, cq_num, max_send_sge, wr_limit; 1647 int err, cq_vector; 1648 struct rtrs_msg_rkey_rsp *rsp; 1649 1650 lockdep_assert_held(&con->con_mutex); 1651 if (con->c.cid == 0) { 1652 max_send_sge = 1; 1653 /* We must be the first here */ 1654 if (WARN_ON(clt_path->s.dev)) 1655 return -EINVAL; 1656 1657 /* 1658 * The whole session uses device from user connection. 1659 * Be careful not to close user connection before ib dev 1660 * is gracefully put. 1661 */ 1662 clt_path->s.dev = rtrs_ib_dev_find_or_add(con->c.cm_id->device, 1663 &dev_pd); 1664 if (!clt_path->s.dev) { 1665 rtrs_wrn(clt_path->clt, 1666 "rtrs_ib_dev_find_get_or_add(): no memory\n"); 1667 return -ENOMEM; 1668 } 1669 clt_path->s.dev_ref = 1; 1670 query_fast_reg_mode(clt_path); 1671 wr_limit = clt_path->s.dev->ib_dev->attrs.max_qp_wr; 1672 /* 1673 * Two (request + registration) completion for send 1674 * Two for recv if always_invalidate is set on server 1675 * or one for recv. 1676 * + 2 for drain and heartbeat 1677 * in case qp gets into error state. 1678 */ 1679 max_send_wr = 1680 min_t(int, wr_limit, SERVICE_CON_QUEUE_DEPTH * 2 + 2); 1681 max_recv_wr = max_send_wr; 1682 } else { 1683 /* 1684 * Here we assume that session members are correctly set. 1685 * This is always true if user connection (cid == 0) is 1686 * established first. 1687 */ 1688 if (WARN_ON(!clt_path->s.dev)) 1689 return -EINVAL; 1690 if (WARN_ON(!clt_path->queue_depth)) 1691 return -EINVAL; 1692 1693 wr_limit = clt_path->s.dev->ib_dev->attrs.max_qp_wr; 1694 /* Shared between connections */ 1695 clt_path->s.dev_ref++; 1696 max_send_wr = min_t(int, wr_limit, 1697 /* QD * (REQ + RSP + FR REGS or INVS) + drain */ 1698 clt_path->queue_depth * 4 + 1); 1699 max_recv_wr = min_t(int, wr_limit, 1700 clt_path->queue_depth * 3 + 1); 1701 max_send_sge = 2; 1702 } 1703 atomic_set(&con->c.sq_wr_avail, max_send_wr); 1704 cq_num = max_send_wr + max_recv_wr; 1705 /* alloc iu to recv new rkey reply when server reports flags set */ 1706 if (clt_path->flags & RTRS_MSG_NEW_RKEY_F || con->c.cid == 0) { 1707 con->rsp_ius = rtrs_iu_alloc(cq_num, sizeof(*rsp), 1708 GFP_KERNEL, 1709 clt_path->s.dev->ib_dev, 1710 DMA_FROM_DEVICE, 1711 rtrs_clt_rdma_done); 1712 if (!con->rsp_ius) 1713 return -ENOMEM; 1714 con->queue_num = cq_num; 1715 } 1716 cq_vector = con->cpu % clt_path->s.dev->ib_dev->num_comp_vectors; 1717 if (con->c.cid >= clt_path->s.irq_con_num) 1718 err = rtrs_cq_qp_create(&clt_path->s, &con->c, max_send_sge, 1719 cq_vector, cq_num, max_send_wr, 1720 max_recv_wr, IB_POLL_DIRECT); 1721 else 1722 err = rtrs_cq_qp_create(&clt_path->s, &con->c, max_send_sge, 1723 cq_vector, cq_num, max_send_wr, 1724 max_recv_wr, IB_POLL_SOFTIRQ); 1725 /* 1726 * In case of error we do not bother to clean previous allocations, 1727 * since destroy_con_cq_qp() must be called. 1728 */ 1729 return err; 1730 } 1731 1732 static void destroy_con_cq_qp(struct rtrs_clt_con *con) 1733 { 1734 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 1735 1736 /* 1737 * Be careful here: destroy_con_cq_qp() can be called even 1738 * create_con_cq_qp() failed, see comments there. 1739 */ 1740 lockdep_assert_held(&con->con_mutex); 1741 rtrs_cq_qp_destroy(&con->c); 1742 if (con->rsp_ius) { 1743 rtrs_iu_free(con->rsp_ius, clt_path->s.dev->ib_dev, 1744 con->queue_num); 1745 con->rsp_ius = NULL; 1746 con->queue_num = 0; 1747 } 1748 if (clt_path->s.dev_ref && !--clt_path->s.dev_ref) { 1749 rtrs_ib_dev_put(clt_path->s.dev); 1750 clt_path->s.dev = NULL; 1751 } 1752 } 1753 1754 static void stop_cm(struct rtrs_clt_con *con) 1755 { 1756 rdma_disconnect(con->c.cm_id); 1757 if (con->c.qp) 1758 ib_drain_qp(con->c.qp); 1759 } 1760 1761 static void destroy_cm(struct rtrs_clt_con *con) 1762 { 1763 rdma_destroy_id(con->c.cm_id); 1764 con->c.cm_id = NULL; 1765 } 1766 1767 static int rtrs_rdma_addr_resolved(struct rtrs_clt_con *con) 1768 { 1769 struct rtrs_path *s = con->c.path; 1770 int err; 1771 1772 mutex_lock(&con->con_mutex); 1773 err = create_con_cq_qp(con); 1774 mutex_unlock(&con->con_mutex); 1775 if (err) { 1776 rtrs_err(s, "create_con_cq_qp(), err: %d\n", err); 1777 return err; 1778 } 1779 err = rdma_resolve_route(con->c.cm_id, RTRS_CONNECT_TIMEOUT_MS); 1780 if (err) 1781 rtrs_err(s, "Resolving route failed, err: %d\n", err); 1782 1783 return err; 1784 } 1785 1786 static int rtrs_rdma_route_resolved(struct rtrs_clt_con *con) 1787 { 1788 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 1789 struct rtrs_clt_sess *clt = clt_path->clt; 1790 struct rtrs_msg_conn_req msg; 1791 struct rdma_conn_param param; 1792 1793 int err; 1794 1795 param = (struct rdma_conn_param) { 1796 .retry_count = 7, 1797 .rnr_retry_count = 7, 1798 .private_data = &msg, 1799 .private_data_len = sizeof(msg), 1800 }; 1801 1802 msg = (struct rtrs_msg_conn_req) { 1803 .magic = cpu_to_le16(RTRS_MAGIC), 1804 .version = cpu_to_le16(RTRS_PROTO_VER), 1805 .cid = cpu_to_le16(con->c.cid), 1806 .cid_num = cpu_to_le16(clt_path->s.con_num), 1807 .recon_cnt = cpu_to_le16(clt_path->s.recon_cnt), 1808 }; 1809 msg.first_conn = clt_path->for_new_clt ? FIRST_CONN : 0; 1810 uuid_copy(&msg.sess_uuid, &clt_path->s.uuid); 1811 uuid_copy(&msg.paths_uuid, &clt->paths_uuid); 1812 1813 err = rdma_connect_locked(con->c.cm_id, ¶m); 1814 if (err) 1815 rtrs_err(clt, "rdma_connect_locked(): %d\n", err); 1816 1817 return err; 1818 } 1819 1820 static int rtrs_rdma_conn_established(struct rtrs_clt_con *con, 1821 struct rdma_cm_event *ev) 1822 { 1823 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 1824 struct rtrs_clt_sess *clt = clt_path->clt; 1825 const struct rtrs_msg_conn_rsp *msg; 1826 u16 version, queue_depth; 1827 int errno; 1828 u8 len; 1829 1830 msg = ev->param.conn.private_data; 1831 len = ev->param.conn.private_data_len; 1832 if (len < sizeof(*msg)) { 1833 rtrs_err(clt, "Invalid RTRS connection response\n"); 1834 return -ECONNRESET; 1835 } 1836 if (le16_to_cpu(msg->magic) != RTRS_MAGIC) { 1837 rtrs_err(clt, "Invalid RTRS magic\n"); 1838 return -ECONNRESET; 1839 } 1840 version = le16_to_cpu(msg->version); 1841 if (version >> 8 != RTRS_PROTO_VER_MAJOR) { 1842 rtrs_err(clt, "Unsupported major RTRS version: %d, expected %d\n", 1843 version >> 8, RTRS_PROTO_VER_MAJOR); 1844 return -ECONNRESET; 1845 } 1846 errno = le16_to_cpu(msg->errno); 1847 if (errno) { 1848 rtrs_err(clt, "Invalid RTRS message: errno %d\n", 1849 errno); 1850 return -ECONNRESET; 1851 } 1852 if (con->c.cid == 0) { 1853 queue_depth = le16_to_cpu(msg->queue_depth); 1854 1855 if (clt_path->queue_depth > 0 && queue_depth != clt_path->queue_depth) { 1856 rtrs_err(clt, "Error: queue depth changed\n"); 1857 1858 /* 1859 * Stop any more reconnection attempts 1860 */ 1861 clt_path->reconnect_attempts = -1; 1862 rtrs_err(clt, 1863 "Disabling auto-reconnect. Trigger a manual reconnect after issue is resolved\n"); 1864 return -ECONNRESET; 1865 } 1866 1867 if (!clt_path->rbufs) { 1868 clt_path->rbufs = kcalloc(queue_depth, 1869 sizeof(*clt_path->rbufs), 1870 GFP_KERNEL); 1871 if (!clt_path->rbufs) 1872 return -ENOMEM; 1873 } 1874 clt_path->queue_depth = queue_depth; 1875 clt_path->s.signal_interval = min_not_zero(queue_depth, 1876 (unsigned short) SERVICE_CON_QUEUE_DEPTH); 1877 clt_path->max_hdr_size = le32_to_cpu(msg->max_hdr_size); 1878 clt_path->max_io_size = le32_to_cpu(msg->max_io_size); 1879 clt_path->flags = le32_to_cpu(msg->flags); 1880 clt_path->chunk_size = clt_path->max_io_size + clt_path->max_hdr_size; 1881 1882 /* 1883 * Global IO size is always a minimum. 1884 * If while a reconnection server sends us a value a bit 1885 * higher - client does not care and uses cached minimum. 1886 * 1887 * Since we can have several sessions (paths) restablishing 1888 * connections in parallel, use lock. 1889 */ 1890 mutex_lock(&clt->paths_mutex); 1891 clt->queue_depth = clt_path->queue_depth; 1892 clt->max_io_size = min_not_zero(clt_path->max_io_size, 1893 clt->max_io_size); 1894 mutex_unlock(&clt->paths_mutex); 1895 1896 /* 1897 * Cache the hca_port and hca_name for sysfs 1898 */ 1899 clt_path->hca_port = con->c.cm_id->port_num; 1900 scnprintf(clt_path->hca_name, sizeof(clt_path->hca_name), 1901 clt_path->s.dev->ib_dev->name); 1902 clt_path->s.src_addr = con->c.cm_id->route.addr.src_addr; 1903 /* set for_new_clt, to allow future reconnect on any path */ 1904 clt_path->for_new_clt = 1; 1905 } 1906 1907 return 0; 1908 } 1909 1910 static inline void flag_success_on_conn(struct rtrs_clt_con *con) 1911 { 1912 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 1913 1914 atomic_inc(&clt_path->connected_cnt); 1915 con->cm_err = 1; 1916 } 1917 1918 static int rtrs_rdma_conn_rejected(struct rtrs_clt_con *con, 1919 struct rdma_cm_event *ev) 1920 { 1921 struct rtrs_path *s = con->c.path; 1922 const struct rtrs_msg_conn_rsp *msg; 1923 const char *rej_msg; 1924 int status, errno; 1925 u8 data_len; 1926 1927 status = ev->status; 1928 rej_msg = rdma_reject_msg(con->c.cm_id, status); 1929 msg = rdma_consumer_reject_data(con->c.cm_id, ev, &data_len); 1930 1931 if (msg && data_len >= sizeof(*msg)) { 1932 errno = (int16_t)le16_to_cpu(msg->errno); 1933 if (errno == -EBUSY) 1934 rtrs_err(s, 1935 "Previous session is still exists on the server, please reconnect later\n"); 1936 else 1937 rtrs_err(s, 1938 "Connect rejected: status %d (%s), rtrs errno %d\n", 1939 status, rej_msg, errno); 1940 } else { 1941 rtrs_err(s, 1942 "Connect rejected but with malformed message: status %d (%s)\n", 1943 status, rej_msg); 1944 } 1945 1946 return -ECONNRESET; 1947 } 1948 1949 void rtrs_clt_close_conns(struct rtrs_clt_path *clt_path, bool wait) 1950 { 1951 trace_rtrs_clt_close_conns(clt_path); 1952 1953 if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CLOSING, NULL)) 1954 queue_work(rtrs_wq, &clt_path->close_work); 1955 if (wait) 1956 flush_work(&clt_path->close_work); 1957 } 1958 1959 static inline void flag_error_on_conn(struct rtrs_clt_con *con, int cm_err) 1960 { 1961 if (con->cm_err == 1) { 1962 struct rtrs_clt_path *clt_path; 1963 1964 clt_path = to_clt_path(con->c.path); 1965 if (atomic_dec_and_test(&clt_path->connected_cnt)) 1966 1967 wake_up(&clt_path->state_wq); 1968 } 1969 con->cm_err = cm_err; 1970 } 1971 1972 static int rtrs_clt_rdma_cm_handler(struct rdma_cm_id *cm_id, 1973 struct rdma_cm_event *ev) 1974 { 1975 struct rtrs_clt_con *con = cm_id->context; 1976 struct rtrs_path *s = con->c.path; 1977 struct rtrs_clt_path *clt_path = to_clt_path(s); 1978 int cm_err = 0; 1979 1980 switch (ev->event) { 1981 case RDMA_CM_EVENT_ADDR_RESOLVED: 1982 cm_err = rtrs_rdma_addr_resolved(con); 1983 break; 1984 case RDMA_CM_EVENT_ROUTE_RESOLVED: 1985 cm_err = rtrs_rdma_route_resolved(con); 1986 break; 1987 case RDMA_CM_EVENT_ESTABLISHED: 1988 cm_err = rtrs_rdma_conn_established(con, ev); 1989 if (!cm_err) { 1990 /* 1991 * Report success and wake up. Here we abuse state_wq, 1992 * i.e. wake up without state change, but we set cm_err. 1993 */ 1994 flag_success_on_conn(con); 1995 wake_up(&clt_path->state_wq); 1996 return 0; 1997 } 1998 break; 1999 case RDMA_CM_EVENT_REJECTED: 2000 cm_err = rtrs_rdma_conn_rejected(con, ev); 2001 break; 2002 case RDMA_CM_EVENT_DISCONNECTED: 2003 /* No message for disconnecting */ 2004 cm_err = -ECONNRESET; 2005 break; 2006 case RDMA_CM_EVENT_CONNECT_ERROR: 2007 case RDMA_CM_EVENT_UNREACHABLE: 2008 case RDMA_CM_EVENT_ADDR_CHANGE: 2009 case RDMA_CM_EVENT_TIMEWAIT_EXIT: 2010 rtrs_wrn(s, "CM error (CM event: %s, err: %d)\n", 2011 rdma_event_msg(ev->event), ev->status); 2012 cm_err = -ECONNRESET; 2013 break; 2014 case RDMA_CM_EVENT_ADDR_ERROR: 2015 case RDMA_CM_EVENT_ROUTE_ERROR: 2016 rtrs_wrn(s, "CM error (CM event: %s, err: %d)\n", 2017 rdma_event_msg(ev->event), ev->status); 2018 cm_err = -EHOSTUNREACH; 2019 break; 2020 case RDMA_CM_EVENT_DEVICE_REMOVAL: 2021 /* 2022 * Device removal is a special case. Queue close and return 0. 2023 */ 2024 rtrs_clt_close_conns(clt_path, false); 2025 return 0; 2026 default: 2027 rtrs_err(s, "Unexpected RDMA CM error (CM event: %s, err: %d)\n", 2028 rdma_event_msg(ev->event), ev->status); 2029 cm_err = -ECONNRESET; 2030 break; 2031 } 2032 2033 if (cm_err) { 2034 /* 2035 * cm error makes sense only on connection establishing, 2036 * in other cases we rely on normal procedure of reconnecting. 2037 */ 2038 flag_error_on_conn(con, cm_err); 2039 rtrs_rdma_error_recovery(con); 2040 } 2041 2042 return 0; 2043 } 2044 2045 /* The caller should do the cleanup in case of error */ 2046 static int create_cm(struct rtrs_clt_con *con) 2047 { 2048 struct rtrs_path *s = con->c.path; 2049 struct rtrs_clt_path *clt_path = to_clt_path(s); 2050 struct rdma_cm_id *cm_id; 2051 int err; 2052 2053 cm_id = rdma_create_id(&init_net, rtrs_clt_rdma_cm_handler, con, 2054 clt_path->s.dst_addr.ss_family == AF_IB ? 2055 RDMA_PS_IB : RDMA_PS_TCP, IB_QPT_RC); 2056 if (IS_ERR(cm_id)) { 2057 err = PTR_ERR(cm_id); 2058 rtrs_err(s, "Failed to create CM ID, err: %d\n", err); 2059 2060 return err; 2061 } 2062 con->c.cm_id = cm_id; 2063 con->cm_err = 0; 2064 /* allow the port to be reused */ 2065 err = rdma_set_reuseaddr(cm_id, 1); 2066 if (err != 0) { 2067 rtrs_err(s, "Set address reuse failed, err: %d\n", err); 2068 return err; 2069 } 2070 err = rdma_resolve_addr(cm_id, (struct sockaddr *)&clt_path->s.src_addr, 2071 (struct sockaddr *)&clt_path->s.dst_addr, 2072 RTRS_CONNECT_TIMEOUT_MS); 2073 if (err) { 2074 rtrs_err(s, "Failed to resolve address, err: %d\n", err); 2075 return err; 2076 } 2077 /* 2078 * Combine connection status and session events. This is needed 2079 * for waiting two possible cases: cm_err has something meaningful 2080 * or session state was really changed to error by device removal. 2081 */ 2082 err = wait_event_interruptible_timeout( 2083 clt_path->state_wq, 2084 con->cm_err || clt_path->state != RTRS_CLT_CONNECTING, 2085 msecs_to_jiffies(RTRS_CONNECT_TIMEOUT_MS)); 2086 if (err == 0 || err == -ERESTARTSYS) { 2087 if (err == 0) 2088 err = -ETIMEDOUT; 2089 /* Timedout or interrupted */ 2090 return err; 2091 } 2092 if (con->cm_err < 0) 2093 return con->cm_err; 2094 if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTING) 2095 /* Device removal */ 2096 return -ECONNABORTED; 2097 2098 return 0; 2099 } 2100 2101 static void rtrs_clt_path_up(struct rtrs_clt_path *clt_path) 2102 { 2103 struct rtrs_clt_sess *clt = clt_path->clt; 2104 int up; 2105 2106 /* 2107 * We can fire RECONNECTED event only when all paths were 2108 * connected on rtrs_clt_open(), then each was disconnected 2109 * and the first one connected again. That's why this nasty 2110 * game with counter value. 2111 */ 2112 2113 mutex_lock(&clt->paths_ev_mutex); 2114 up = ++clt->paths_up; 2115 /* 2116 * Here it is safe to access paths num directly since up counter 2117 * is greater than MAX_PATHS_NUM only while rtrs_clt_open() is 2118 * in progress, thus paths removals are impossible. 2119 */ 2120 if (up > MAX_PATHS_NUM && up == MAX_PATHS_NUM + clt->paths_num) 2121 clt->paths_up = clt->paths_num; 2122 else if (up == 1) 2123 clt->link_ev(clt->priv, RTRS_CLT_LINK_EV_RECONNECTED); 2124 mutex_unlock(&clt->paths_ev_mutex); 2125 2126 /* Mark session as established */ 2127 clt_path->established = true; 2128 clt_path->reconnect_attempts = 0; 2129 clt_path->stats->reconnects.successful_cnt++; 2130 } 2131 2132 static void rtrs_clt_path_down(struct rtrs_clt_path *clt_path) 2133 { 2134 struct rtrs_clt_sess *clt = clt_path->clt; 2135 2136 if (!clt_path->established) 2137 return; 2138 2139 clt_path->established = false; 2140 mutex_lock(&clt->paths_ev_mutex); 2141 WARN_ON(!clt->paths_up); 2142 if (--clt->paths_up == 0) 2143 clt->link_ev(clt->priv, RTRS_CLT_LINK_EV_DISCONNECTED); 2144 mutex_unlock(&clt->paths_ev_mutex); 2145 } 2146 2147 static void rtrs_clt_stop_and_destroy_conns(struct rtrs_clt_path *clt_path) 2148 { 2149 struct rtrs_clt_con *con; 2150 unsigned int cid; 2151 2152 WARN_ON(READ_ONCE(clt_path->state) == RTRS_CLT_CONNECTED); 2153 2154 /* 2155 * Possible race with rtrs_clt_open(), when DEVICE_REMOVAL comes 2156 * exactly in between. Start destroying after it finishes. 2157 */ 2158 mutex_lock(&clt_path->init_mutex); 2159 mutex_unlock(&clt_path->init_mutex); 2160 2161 /* 2162 * All IO paths must observe !CONNECTED state before we 2163 * free everything. 2164 */ 2165 synchronize_rcu(); 2166 2167 rtrs_stop_hb(&clt_path->s); 2168 2169 /* 2170 * The order it utterly crucial: firstly disconnect and complete all 2171 * rdma requests with error (thus set in_use=false for requests), 2172 * then fail outstanding requests checking in_use for each, and 2173 * eventually notify upper layer about session disconnection. 2174 */ 2175 2176 for (cid = 0; cid < clt_path->s.con_num; cid++) { 2177 if (!clt_path->s.con[cid]) 2178 break; 2179 con = to_clt_con(clt_path->s.con[cid]); 2180 stop_cm(con); 2181 } 2182 fail_all_outstanding_reqs(clt_path); 2183 free_path_reqs(clt_path); 2184 rtrs_clt_path_down(clt_path); 2185 2186 /* 2187 * Wait for graceful shutdown, namely when peer side invokes 2188 * rdma_disconnect(). 'connected_cnt' is decremented only on 2189 * CM events, thus if other side had crashed and hb has detected 2190 * something is wrong, here we will stuck for exactly timeout ms, 2191 * since CM does not fire anything. That is fine, we are not in 2192 * hurry. 2193 */ 2194 wait_event_timeout(clt_path->state_wq, 2195 !atomic_read(&clt_path->connected_cnt), 2196 msecs_to_jiffies(RTRS_CONNECT_TIMEOUT_MS)); 2197 2198 for (cid = 0; cid < clt_path->s.con_num; cid++) { 2199 if (!clt_path->s.con[cid]) 2200 break; 2201 con = to_clt_con(clt_path->s.con[cid]); 2202 mutex_lock(&con->con_mutex); 2203 destroy_con_cq_qp(con); 2204 mutex_unlock(&con->con_mutex); 2205 destroy_cm(con); 2206 destroy_con(con); 2207 } 2208 } 2209 2210 static void rtrs_clt_remove_path_from_arr(struct rtrs_clt_path *clt_path) 2211 { 2212 struct rtrs_clt_sess *clt = clt_path->clt; 2213 struct rtrs_clt_path *next; 2214 bool wait_for_grace = false; 2215 int cpu; 2216 2217 mutex_lock(&clt->paths_mutex); 2218 list_del_rcu(&clt_path->s.entry); 2219 2220 /* Make sure everybody observes path removal. */ 2221 synchronize_rcu(); 2222 2223 /* 2224 * At this point nobody sees @sess in the list, but still we have 2225 * dangling pointer @pcpu_path which _can_ point to @sess. Since 2226 * nobody can observe @sess in the list, we guarantee that IO path 2227 * will not assign @sess to @pcpu_path, i.e. @pcpu_path can be equal 2228 * to @sess, but can never again become @sess. 2229 */ 2230 2231 /* 2232 * Decrement paths number only after grace period, because 2233 * caller of do_each_path() must firstly observe list without 2234 * path and only then decremented paths number. 2235 * 2236 * Otherwise there can be the following situation: 2237 * o Two paths exist and IO is coming. 2238 * o One path is removed: 2239 * CPU#0 CPU#1 2240 * do_each_path(): rtrs_clt_remove_path_from_arr(): 2241 * path = get_next_path() 2242 * ^^^ list_del_rcu(path) 2243 * [!CONNECTED path] clt->paths_num-- 2244 * ^^^^^^^^^ 2245 * load clt->paths_num from 2 to 1 2246 * ^^^^^^^^^ 2247 * sees 1 2248 * 2249 * path is observed as !CONNECTED, but do_each_path() loop 2250 * ends, because expression i < clt->paths_num is false. 2251 */ 2252 clt->paths_num--; 2253 2254 /* 2255 * Get @next connection from current @sess which is going to be 2256 * removed. If @sess is the last element, then @next is NULL. 2257 */ 2258 rcu_read_lock(); 2259 next = rtrs_clt_get_next_path_or_null(&clt->paths_list, clt_path); 2260 rcu_read_unlock(); 2261 2262 /* 2263 * @pcpu paths can still point to the path which is going to be 2264 * removed, so change the pointer manually. 2265 */ 2266 for_each_possible_cpu(cpu) { 2267 struct rtrs_clt_path __rcu **ppcpu_path; 2268 2269 ppcpu_path = per_cpu_ptr(clt->pcpu_path, cpu); 2270 if (rcu_dereference_protected(*ppcpu_path, 2271 lockdep_is_held(&clt->paths_mutex)) != clt_path) 2272 /* 2273 * synchronize_rcu() was called just after deleting 2274 * entry from the list, thus IO code path cannot 2275 * change pointer back to the pointer which is going 2276 * to be removed, we are safe here. 2277 */ 2278 continue; 2279 2280 /* 2281 * We race with IO code path, which also changes pointer, 2282 * thus we have to be careful not to overwrite it. 2283 */ 2284 if (try_cmpxchg((struct rtrs_clt_path **)ppcpu_path, &clt_path, 2285 next)) 2286 /* 2287 * @ppcpu_path was successfully replaced with @next, 2288 * that means that someone could also pick up the 2289 * @sess and dereferencing it right now, so wait for 2290 * a grace period is required. 2291 */ 2292 wait_for_grace = true; 2293 } 2294 if (wait_for_grace) 2295 synchronize_rcu(); 2296 2297 mutex_unlock(&clt->paths_mutex); 2298 } 2299 2300 static void rtrs_clt_add_path_to_arr(struct rtrs_clt_path *clt_path) 2301 { 2302 struct rtrs_clt_sess *clt = clt_path->clt; 2303 2304 mutex_lock(&clt->paths_mutex); 2305 clt->paths_num++; 2306 2307 list_add_tail_rcu(&clt_path->s.entry, &clt->paths_list); 2308 mutex_unlock(&clt->paths_mutex); 2309 } 2310 2311 static void rtrs_clt_close_work(struct work_struct *work) 2312 { 2313 struct rtrs_clt_path *clt_path; 2314 2315 clt_path = container_of(work, struct rtrs_clt_path, close_work); 2316 2317 cancel_work_sync(&clt_path->err_recovery_work); 2318 cancel_delayed_work_sync(&clt_path->reconnect_dwork); 2319 rtrs_clt_stop_and_destroy_conns(clt_path); 2320 rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CLOSED, NULL); 2321 } 2322 2323 static int init_conns(struct rtrs_clt_path *clt_path) 2324 { 2325 unsigned int cid; 2326 int err, i; 2327 2328 /* 2329 * On every new session connections increase reconnect counter 2330 * to avoid clashes with previous sessions not yet closed 2331 * sessions on a server side. 2332 */ 2333 clt_path->s.recon_cnt++; 2334 2335 /* Establish all RDMA connections */ 2336 for (cid = 0; cid < clt_path->s.con_num; cid++) { 2337 err = create_con(clt_path, cid); 2338 if (err) 2339 goto destroy; 2340 2341 err = create_cm(to_clt_con(clt_path->s.con[cid])); 2342 if (err) 2343 goto destroy; 2344 } 2345 2346 /* 2347 * Set the cid to con_num - 1, since if we fail later, we want to stay in bounds. 2348 */ 2349 cid = clt_path->s.con_num - 1; 2350 2351 err = alloc_path_reqs(clt_path); 2352 if (err) 2353 goto destroy; 2354 2355 return 0; 2356 2357 destroy: 2358 /* Make sure we do the cleanup in the order they are created */ 2359 for (i = 0; i <= cid; i++) { 2360 struct rtrs_clt_con *con; 2361 2362 if (!clt_path->s.con[i]) 2363 break; 2364 2365 con = to_clt_con(clt_path->s.con[i]); 2366 if (con->c.cm_id) { 2367 stop_cm(con); 2368 mutex_lock(&con->con_mutex); 2369 destroy_con_cq_qp(con); 2370 mutex_unlock(&con->con_mutex); 2371 destroy_cm(con); 2372 } 2373 destroy_con(con); 2374 } 2375 /* 2376 * If we've never taken async path and got an error, say, 2377 * doing rdma_resolve_addr(), switch to CONNECTION_ERR state 2378 * manually to keep reconnecting. 2379 */ 2380 rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING_ERR, NULL); 2381 2382 return err; 2383 } 2384 2385 static void rtrs_clt_info_req_done(struct ib_cq *cq, struct ib_wc *wc) 2386 { 2387 struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context); 2388 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 2389 struct rtrs_iu *iu; 2390 2391 iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe); 2392 rtrs_iu_free(iu, clt_path->s.dev->ib_dev, 1); 2393 2394 if (wc->status != IB_WC_SUCCESS) { 2395 rtrs_err(clt_path->clt, "Path info request send failed: %s\n", 2396 ib_wc_status_msg(wc->status)); 2397 rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING_ERR, NULL); 2398 return; 2399 } 2400 2401 rtrs_clt_update_wc_stats(con); 2402 } 2403 2404 static int process_info_rsp(struct rtrs_clt_path *clt_path, 2405 const struct rtrs_msg_info_rsp *msg) 2406 { 2407 unsigned int sg_cnt, total_len; 2408 int i, sgi; 2409 2410 sg_cnt = le16_to_cpu(msg->sg_cnt); 2411 if (!sg_cnt || (clt_path->queue_depth % sg_cnt)) { 2412 rtrs_err(clt_path->clt, 2413 "Incorrect sg_cnt %d, is not multiple\n", 2414 sg_cnt); 2415 return -EINVAL; 2416 } 2417 2418 /* 2419 * Check if IB immediate data size is enough to hold the mem_id and 2420 * the offset inside the memory chunk. 2421 */ 2422 if ((ilog2(sg_cnt - 1) + 1) + (ilog2(clt_path->chunk_size - 1) + 1) > 2423 MAX_IMM_PAYL_BITS) { 2424 rtrs_err(clt_path->clt, 2425 "RDMA immediate size (%db) not enough to encode %d buffers of size %dB\n", 2426 MAX_IMM_PAYL_BITS, sg_cnt, clt_path->chunk_size); 2427 return -EINVAL; 2428 } 2429 total_len = 0; 2430 for (sgi = 0, i = 0; sgi < sg_cnt && i < clt_path->queue_depth; sgi++) { 2431 const struct rtrs_sg_desc *desc = &msg->desc[sgi]; 2432 u32 len, rkey; 2433 u64 addr; 2434 2435 addr = le64_to_cpu(desc->addr); 2436 rkey = le32_to_cpu(desc->key); 2437 len = le32_to_cpu(desc->len); 2438 2439 total_len += len; 2440 2441 if (!len || (len % clt_path->chunk_size)) { 2442 rtrs_err(clt_path->clt, "Incorrect [%d].len %d\n", 2443 sgi, 2444 len); 2445 return -EINVAL; 2446 } 2447 for ( ; len && i < clt_path->queue_depth; i++) { 2448 clt_path->rbufs[i].addr = addr; 2449 clt_path->rbufs[i].rkey = rkey; 2450 2451 len -= clt_path->chunk_size; 2452 addr += clt_path->chunk_size; 2453 } 2454 } 2455 /* Sanity check */ 2456 if (sgi != sg_cnt || i != clt_path->queue_depth) { 2457 rtrs_err(clt_path->clt, 2458 "Incorrect sg vector, not fully mapped\n"); 2459 return -EINVAL; 2460 } 2461 if (total_len != clt_path->chunk_size * clt_path->queue_depth) { 2462 rtrs_err(clt_path->clt, "Incorrect total_len %d\n", total_len); 2463 return -EINVAL; 2464 } 2465 2466 return 0; 2467 } 2468 2469 static void rtrs_clt_info_rsp_done(struct ib_cq *cq, struct ib_wc *wc) 2470 { 2471 struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context); 2472 struct rtrs_clt_path *clt_path = to_clt_path(con->c.path); 2473 struct rtrs_msg_info_rsp *msg; 2474 enum rtrs_clt_state state; 2475 struct rtrs_iu *iu; 2476 size_t rx_sz; 2477 int err; 2478 2479 state = RTRS_CLT_CONNECTING_ERR; 2480 2481 WARN_ON(con->c.cid); 2482 iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe); 2483 if (wc->status != IB_WC_SUCCESS) { 2484 rtrs_err(clt_path->clt, "Path info response recv failed: %s\n", 2485 ib_wc_status_msg(wc->status)); 2486 goto out; 2487 } 2488 WARN_ON(wc->opcode != IB_WC_RECV); 2489 2490 if (wc->byte_len < sizeof(*msg)) { 2491 rtrs_err(clt_path->clt, "Path info response is malformed: size %d\n", 2492 wc->byte_len); 2493 goto out; 2494 } 2495 ib_dma_sync_single_for_cpu(clt_path->s.dev->ib_dev, iu->dma_addr, 2496 iu->size, DMA_FROM_DEVICE); 2497 msg = iu->buf; 2498 if (le16_to_cpu(msg->type) != RTRS_MSG_INFO_RSP) { 2499 rtrs_err(clt_path->clt, "Path info response is malformed: type %d\n", 2500 le16_to_cpu(msg->type)); 2501 goto out; 2502 } 2503 rx_sz = sizeof(*msg); 2504 rx_sz += sizeof(msg->desc[0]) * le16_to_cpu(msg->sg_cnt); 2505 if (wc->byte_len < rx_sz) { 2506 rtrs_err(clt_path->clt, "Path info response is malformed: size %d\n", 2507 wc->byte_len); 2508 goto out; 2509 } 2510 err = process_info_rsp(clt_path, msg); 2511 if (err) 2512 goto out; 2513 2514 err = post_recv_path(clt_path); 2515 if (err) 2516 goto out; 2517 2518 state = RTRS_CLT_CONNECTED; 2519 2520 out: 2521 rtrs_clt_update_wc_stats(con); 2522 rtrs_iu_free(iu, clt_path->s.dev->ib_dev, 1); 2523 rtrs_clt_change_state_get_old(clt_path, state, NULL); 2524 } 2525 2526 static int rtrs_send_path_info(struct rtrs_clt_path *clt_path) 2527 { 2528 struct rtrs_clt_con *usr_con = to_clt_con(clt_path->s.con[0]); 2529 struct rtrs_msg_info_req *msg; 2530 struct rtrs_iu *tx_iu, *rx_iu; 2531 size_t rx_sz; 2532 int err; 2533 2534 rx_sz = sizeof(struct rtrs_msg_info_rsp); 2535 rx_sz += sizeof(struct rtrs_sg_desc) * clt_path->queue_depth; 2536 2537 tx_iu = rtrs_iu_alloc(1, sizeof(struct rtrs_msg_info_req), GFP_KERNEL, 2538 clt_path->s.dev->ib_dev, DMA_TO_DEVICE, 2539 rtrs_clt_info_req_done); 2540 rx_iu = rtrs_iu_alloc(1, rx_sz, GFP_KERNEL, clt_path->s.dev->ib_dev, 2541 DMA_FROM_DEVICE, rtrs_clt_info_rsp_done); 2542 if (!tx_iu || !rx_iu) { 2543 err = -ENOMEM; 2544 goto out; 2545 } 2546 /* Prepare for getting info response */ 2547 err = rtrs_iu_post_recv(&usr_con->c, rx_iu); 2548 if (err) { 2549 rtrs_err(clt_path->clt, "rtrs_iu_post_recv(), err: %d\n", err); 2550 goto out; 2551 } 2552 rx_iu = NULL; 2553 2554 msg = tx_iu->buf; 2555 msg->type = cpu_to_le16(RTRS_MSG_INFO_REQ); 2556 memcpy(msg->pathname, clt_path->s.sessname, sizeof(msg->pathname)); 2557 2558 ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev, 2559 tx_iu->dma_addr, 2560 tx_iu->size, DMA_TO_DEVICE); 2561 2562 /* Send info request */ 2563 err = rtrs_iu_post_send(&usr_con->c, tx_iu, sizeof(*msg), NULL); 2564 if (err) { 2565 rtrs_err(clt_path->clt, "rtrs_iu_post_send(), err: %d\n", err); 2566 goto out; 2567 } 2568 tx_iu = NULL; 2569 2570 /* Wait for state change */ 2571 wait_event_interruptible_timeout(clt_path->state_wq, 2572 clt_path->state != RTRS_CLT_CONNECTING, 2573 msecs_to_jiffies( 2574 RTRS_CONNECT_TIMEOUT_MS)); 2575 if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED) { 2576 if (READ_ONCE(clt_path->state) == RTRS_CLT_CONNECTING_ERR) 2577 err = -ECONNRESET; 2578 else 2579 err = -ETIMEDOUT; 2580 } 2581 2582 out: 2583 if (tx_iu) 2584 rtrs_iu_free(tx_iu, clt_path->s.dev->ib_dev, 1); 2585 if (rx_iu) 2586 rtrs_iu_free(rx_iu, clt_path->s.dev->ib_dev, 1); 2587 if (err) 2588 /* If we've never taken async path because of malloc problems */ 2589 rtrs_clt_change_state_get_old(clt_path, 2590 RTRS_CLT_CONNECTING_ERR, NULL); 2591 2592 return err; 2593 } 2594 2595 /** 2596 * init_path() - establishes all path connections and does handshake 2597 * @clt_path: client path. 2598 * In case of error full close or reconnect procedure should be taken, 2599 * because reconnect or close async works can be started. 2600 */ 2601 static int init_path(struct rtrs_clt_path *clt_path) 2602 { 2603 int err; 2604 char str[NAME_MAX]; 2605 struct rtrs_addr path = { 2606 .src = &clt_path->s.src_addr, 2607 .dst = &clt_path->s.dst_addr, 2608 }; 2609 2610 rtrs_addr_to_str(&path, str, sizeof(str)); 2611 2612 mutex_lock(&clt_path->init_mutex); 2613 err = init_conns(clt_path); 2614 if (err) { 2615 rtrs_err(clt_path->clt, 2616 "init_conns() failed: err=%d path=%s [%s:%u]\n", err, 2617 str, clt_path->hca_name, clt_path->hca_port); 2618 goto out; 2619 } 2620 err = rtrs_send_path_info(clt_path); 2621 if (err) { 2622 rtrs_err(clt_path->clt, 2623 "rtrs_send_path_info() failed: err=%d path=%s [%s:%u]\n", 2624 err, str, clt_path->hca_name, clt_path->hca_port); 2625 goto out; 2626 } 2627 rtrs_clt_path_up(clt_path); 2628 rtrs_start_hb(&clt_path->s); 2629 out: 2630 mutex_unlock(&clt_path->init_mutex); 2631 2632 return err; 2633 } 2634 2635 static void rtrs_clt_reconnect_work(struct work_struct *work) 2636 { 2637 struct rtrs_clt_path *clt_path; 2638 struct rtrs_clt_sess *clt; 2639 int err; 2640 2641 clt_path = container_of(to_delayed_work(work), struct rtrs_clt_path, 2642 reconnect_dwork); 2643 clt = clt_path->clt; 2644 2645 trace_rtrs_clt_reconnect_work(clt_path); 2646 2647 if (READ_ONCE(clt_path->state) != RTRS_CLT_RECONNECTING) 2648 return; 2649 2650 if (clt_path->reconnect_attempts >= clt->max_reconnect_attempts) { 2651 /* Close a path completely if max attempts is reached */ 2652 rtrs_clt_close_conns(clt_path, false); 2653 return; 2654 } 2655 clt_path->reconnect_attempts++; 2656 2657 msleep(RTRS_RECONNECT_BACKOFF); 2658 if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING, NULL)) { 2659 err = init_path(clt_path); 2660 if (err) 2661 goto reconnect_again; 2662 } 2663 2664 return; 2665 2666 reconnect_again: 2667 if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_RECONNECTING, NULL)) { 2668 clt_path->stats->reconnects.fail_cnt++; 2669 queue_work(rtrs_wq, &clt_path->err_recovery_work); 2670 } 2671 } 2672 2673 static void rtrs_clt_dev_release(struct device *dev) 2674 { 2675 struct rtrs_clt_sess *clt = container_of(dev, struct rtrs_clt_sess, 2676 dev); 2677 2678 mutex_destroy(&clt->paths_ev_mutex); 2679 mutex_destroy(&clt->paths_mutex); 2680 kfree(clt); 2681 } 2682 2683 static struct rtrs_clt_sess *alloc_clt(const char *sessname, size_t paths_num, 2684 u16 port, size_t pdu_sz, void *priv, 2685 void (*link_ev)(void *priv, 2686 enum rtrs_clt_link_ev ev), 2687 unsigned int reconnect_delay_sec, 2688 unsigned int max_reconnect_attempts) 2689 { 2690 struct rtrs_clt_sess *clt; 2691 int err; 2692 2693 if (!paths_num || paths_num > MAX_PATHS_NUM) 2694 return ERR_PTR(-EINVAL); 2695 2696 if (strlen(sessname) >= sizeof(clt->sessname)) 2697 return ERR_PTR(-EINVAL); 2698 2699 clt = kzalloc(sizeof(*clt), GFP_KERNEL); 2700 if (!clt) 2701 return ERR_PTR(-ENOMEM); 2702 2703 clt->pcpu_path = alloc_percpu(typeof(*clt->pcpu_path)); 2704 if (!clt->pcpu_path) { 2705 kfree(clt); 2706 return ERR_PTR(-ENOMEM); 2707 } 2708 2709 clt->dev.class = &rtrs_clt_dev_class; 2710 clt->dev.release = rtrs_clt_dev_release; 2711 uuid_gen(&clt->paths_uuid); 2712 INIT_LIST_HEAD_RCU(&clt->paths_list); 2713 clt->paths_num = paths_num; 2714 clt->paths_up = MAX_PATHS_NUM; 2715 clt->port = port; 2716 clt->pdu_sz = pdu_sz; 2717 clt->max_segments = RTRS_MAX_SEGMENTS; 2718 clt->reconnect_delay_sec = reconnect_delay_sec; 2719 clt->max_reconnect_attempts = max_reconnect_attempts; 2720 clt->priv = priv; 2721 clt->link_ev = link_ev; 2722 clt->mp_policy = MP_POLICY_MIN_INFLIGHT; 2723 strscpy(clt->sessname, sessname, sizeof(clt->sessname)); 2724 init_waitqueue_head(&clt->permits_wait); 2725 mutex_init(&clt->paths_ev_mutex); 2726 mutex_init(&clt->paths_mutex); 2727 device_initialize(&clt->dev); 2728 2729 err = dev_set_name(&clt->dev, "%s", sessname); 2730 if (err) 2731 goto err_put; 2732 2733 /* 2734 * Suppress user space notification until 2735 * sysfs files are created 2736 */ 2737 dev_set_uevent_suppress(&clt->dev, true); 2738 err = device_add(&clt->dev); 2739 if (err) 2740 goto err_put; 2741 2742 clt->kobj_paths = kobject_create_and_add("paths", &clt->dev.kobj); 2743 if (!clt->kobj_paths) { 2744 err = -ENOMEM; 2745 goto err_del; 2746 } 2747 err = rtrs_clt_create_sysfs_root_files(clt); 2748 if (err) { 2749 kobject_del(clt->kobj_paths); 2750 kobject_put(clt->kobj_paths); 2751 goto err_del; 2752 } 2753 dev_set_uevent_suppress(&clt->dev, false); 2754 kobject_uevent(&clt->dev.kobj, KOBJ_ADD); 2755 2756 return clt; 2757 err_del: 2758 device_del(&clt->dev); 2759 err_put: 2760 free_percpu(clt->pcpu_path); 2761 put_device(&clt->dev); 2762 return ERR_PTR(err); 2763 } 2764 2765 static void free_clt(struct rtrs_clt_sess *clt) 2766 { 2767 free_percpu(clt->pcpu_path); 2768 2769 /* 2770 * release callback will free clt and destroy mutexes in last put 2771 */ 2772 device_unregister(&clt->dev); 2773 } 2774 2775 /** 2776 * rtrs_clt_open() - Open a path to an RTRS server 2777 * @ops: holds the link event callback and the private pointer. 2778 * @pathname: name of the path to an RTRS server 2779 * @paths: Paths to be established defined by their src and dst addresses 2780 * @paths_num: Number of elements in the @paths array 2781 * @port: port to be used by the RTRS session 2782 * @pdu_sz: Size of extra payload which can be accessed after permit allocation. 2783 * @reconnect_delay_sec: time between reconnect tries 2784 * @max_reconnect_attempts: Number of times to reconnect on error before giving 2785 * up, 0 for * disabled, -1 for forever 2786 * @nr_poll_queues: number of polling mode connection using IB_POLL_DIRECT flag 2787 * 2788 * Starts session establishment with the rtrs_server. The function can block 2789 * up to ~2000ms before it returns. 2790 * 2791 * Return a valid pointer on success otherwise PTR_ERR. 2792 */ 2793 struct rtrs_clt_sess *rtrs_clt_open(struct rtrs_clt_ops *ops, 2794 const char *pathname, 2795 const struct rtrs_addr *paths, 2796 size_t paths_num, u16 port, 2797 size_t pdu_sz, u8 reconnect_delay_sec, 2798 s16 max_reconnect_attempts, u32 nr_poll_queues) 2799 { 2800 struct rtrs_clt_path *clt_path, *tmp; 2801 struct rtrs_clt_sess *clt; 2802 int err, i; 2803 2804 if (strchr(pathname, '/') || strchr(pathname, '.')) { 2805 pr_err("pathname cannot contain / and .\n"); 2806 err = -EINVAL; 2807 goto out; 2808 } 2809 2810 clt = alloc_clt(pathname, paths_num, port, pdu_sz, ops->priv, 2811 ops->link_ev, 2812 reconnect_delay_sec, 2813 max_reconnect_attempts); 2814 if (IS_ERR(clt)) { 2815 err = PTR_ERR(clt); 2816 goto out; 2817 } 2818 for (i = 0; i < paths_num; i++) { 2819 struct rtrs_clt_path *clt_path; 2820 2821 clt_path = alloc_path(clt, &paths[i], nr_cpu_ids, 2822 nr_poll_queues); 2823 if (IS_ERR(clt_path)) { 2824 err = PTR_ERR(clt_path); 2825 goto close_all_path; 2826 } 2827 if (!i) 2828 clt_path->for_new_clt = 1; 2829 list_add_tail_rcu(&clt_path->s.entry, &clt->paths_list); 2830 2831 err = init_path(clt_path); 2832 if (err) { 2833 list_del_rcu(&clt_path->s.entry); 2834 rtrs_clt_close_conns(clt_path, true); 2835 free_percpu(clt_path->stats->pcpu_stats); 2836 kfree(clt_path->stats); 2837 free_path(clt_path); 2838 goto close_all_path; 2839 } 2840 2841 err = rtrs_clt_create_path_files(clt_path); 2842 if (err) { 2843 list_del_rcu(&clt_path->s.entry); 2844 rtrs_clt_close_conns(clt_path, true); 2845 free_percpu(clt_path->stats->pcpu_stats); 2846 kfree(clt_path->stats); 2847 free_path(clt_path); 2848 goto close_all_path; 2849 } 2850 } 2851 err = alloc_permits(clt); 2852 if (err) 2853 goto close_all_path; 2854 2855 return clt; 2856 2857 close_all_path: 2858 list_for_each_entry_safe(clt_path, tmp, &clt->paths_list, s.entry) { 2859 rtrs_clt_destroy_path_files(clt_path, NULL); 2860 rtrs_clt_close_conns(clt_path, true); 2861 kobject_put(&clt_path->kobj); 2862 } 2863 rtrs_clt_destroy_sysfs_root(clt); 2864 free_clt(clt); 2865 2866 out: 2867 return ERR_PTR(err); 2868 } 2869 EXPORT_SYMBOL(rtrs_clt_open); 2870 2871 /** 2872 * rtrs_clt_close() - Close a path 2873 * @clt: Session handle. Session is freed upon return. 2874 */ 2875 void rtrs_clt_close(struct rtrs_clt_sess *clt) 2876 { 2877 struct rtrs_clt_path *clt_path, *tmp; 2878 2879 /* Firstly forbid sysfs access */ 2880 rtrs_clt_destroy_sysfs_root(clt); 2881 2882 /* Now it is safe to iterate over all paths without locks */ 2883 list_for_each_entry_safe(clt_path, tmp, &clt->paths_list, s.entry) { 2884 rtrs_clt_close_conns(clt_path, true); 2885 rtrs_clt_destroy_path_files(clt_path, NULL); 2886 kobject_put(&clt_path->kobj); 2887 } 2888 free_permits(clt); 2889 free_clt(clt); 2890 } 2891 EXPORT_SYMBOL(rtrs_clt_close); 2892 2893 int rtrs_clt_reconnect_from_sysfs(struct rtrs_clt_path *clt_path) 2894 { 2895 enum rtrs_clt_state old_state; 2896 int err = -EBUSY; 2897 bool changed; 2898 2899 changed = rtrs_clt_change_state_get_old(clt_path, 2900 RTRS_CLT_RECONNECTING, 2901 &old_state); 2902 if (changed) { 2903 clt_path->reconnect_attempts = 0; 2904 rtrs_clt_stop_and_destroy_conns(clt_path); 2905 queue_delayed_work(rtrs_wq, &clt_path->reconnect_dwork, 0); 2906 } 2907 if (changed || old_state == RTRS_CLT_RECONNECTING) { 2908 /* 2909 * flush_delayed_work() queues pending work for immediate 2910 * execution, so do the flush if we have queued something 2911 * right now or work is pending. 2912 */ 2913 flush_delayed_work(&clt_path->reconnect_dwork); 2914 err = (READ_ONCE(clt_path->state) == 2915 RTRS_CLT_CONNECTED ? 0 : -ENOTCONN); 2916 } 2917 2918 return err; 2919 } 2920 2921 int rtrs_clt_remove_path_from_sysfs(struct rtrs_clt_path *clt_path, 2922 const struct attribute *sysfs_self) 2923 { 2924 enum rtrs_clt_state old_state; 2925 bool changed; 2926 2927 /* 2928 * Continue stopping path till state was changed to DEAD or 2929 * state was observed as DEAD: 2930 * 1. State was changed to DEAD - we were fast and nobody 2931 * invoked rtrs_clt_reconnect(), which can again start 2932 * reconnecting. 2933 * 2. State was observed as DEAD - we have someone in parallel 2934 * removing the path. 2935 */ 2936 do { 2937 rtrs_clt_close_conns(clt_path, true); 2938 changed = rtrs_clt_change_state_get_old(clt_path, 2939 RTRS_CLT_DEAD, 2940 &old_state); 2941 } while (!changed && old_state != RTRS_CLT_DEAD); 2942 2943 if (changed) { 2944 rtrs_clt_remove_path_from_arr(clt_path); 2945 rtrs_clt_destroy_path_files(clt_path, sysfs_self); 2946 kobject_put(&clt_path->kobj); 2947 } 2948 2949 return 0; 2950 } 2951 2952 void rtrs_clt_set_max_reconnect_attempts(struct rtrs_clt_sess *clt, int value) 2953 { 2954 clt->max_reconnect_attempts = (unsigned int)value; 2955 } 2956 2957 int rtrs_clt_get_max_reconnect_attempts(const struct rtrs_clt_sess *clt) 2958 { 2959 return (int)clt->max_reconnect_attempts; 2960 } 2961 2962 /** 2963 * rtrs_clt_request() - Request data transfer to/from server via RDMA. 2964 * 2965 * @dir: READ/WRITE 2966 * @ops: callback function to be called as confirmation, and the pointer. 2967 * @clt: Session 2968 * @permit: Preallocated permit 2969 * @vec: Message that is sent to server together with the request. 2970 * Sum of len of all @vec elements limited to <= IO_MSG_SIZE. 2971 * Since the msg is copied internally it can be allocated on stack. 2972 * @nr: Number of elements in @vec. 2973 * @data_len: length of data sent to/from server 2974 * @sg: Pages to be sent/received to/from server. 2975 * @sg_cnt: Number of elements in the @sg 2976 * 2977 * Return: 2978 * 0: Success 2979 * <0: Error 2980 * 2981 * On dir=READ rtrs client will request a data transfer from Server to client. 2982 * The data that the server will respond with will be stored in @sg when 2983 * the user receives an %RTRS_CLT_RDMA_EV_RDMA_REQUEST_WRITE_COMPL event. 2984 * On dir=WRITE rtrs client will rdma write data in sg to server side. 2985 */ 2986 int rtrs_clt_request(int dir, struct rtrs_clt_req_ops *ops, 2987 struct rtrs_clt_sess *clt, struct rtrs_permit *permit, 2988 const struct kvec *vec, size_t nr, size_t data_len, 2989 struct scatterlist *sg, unsigned int sg_cnt) 2990 { 2991 struct rtrs_clt_io_req *req; 2992 struct rtrs_clt_path *clt_path; 2993 2994 enum dma_data_direction dma_dir; 2995 int err = -ECONNABORTED, i; 2996 size_t usr_len, hdr_len; 2997 struct path_it it; 2998 2999 /* Get kvec length */ 3000 for (i = 0, usr_len = 0; i < nr; i++) 3001 usr_len += vec[i].iov_len; 3002 3003 if (dir == READ) { 3004 hdr_len = sizeof(struct rtrs_msg_rdma_read) + 3005 sg_cnt * sizeof(struct rtrs_sg_desc); 3006 dma_dir = DMA_FROM_DEVICE; 3007 } else { 3008 hdr_len = sizeof(struct rtrs_msg_rdma_write); 3009 dma_dir = DMA_TO_DEVICE; 3010 } 3011 3012 rcu_read_lock(); 3013 for (path_it_init(&it, clt); 3014 (clt_path = it.next_path(&it)) && it.i < it.clt->paths_num; it.i++) { 3015 if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED) 3016 continue; 3017 3018 if (usr_len + hdr_len > clt_path->max_hdr_size) { 3019 rtrs_wrn_rl(clt_path->clt, 3020 "%s request failed, user message size is %zu and header length %zu, but max size is %u\n", 3021 dir == READ ? "Read" : "Write", 3022 usr_len, hdr_len, clt_path->max_hdr_size); 3023 err = -EMSGSIZE; 3024 break; 3025 } 3026 req = rtrs_clt_get_req(clt_path, ops->conf_fn, permit, ops->priv, 3027 vec, usr_len, sg, sg_cnt, data_len, 3028 dma_dir); 3029 if (dir == READ) 3030 err = rtrs_clt_read_req(req); 3031 else 3032 err = rtrs_clt_write_req(req); 3033 if (err) { 3034 req->in_use = false; 3035 continue; 3036 } 3037 /* Success path */ 3038 break; 3039 } 3040 path_it_deinit(&it); 3041 rcu_read_unlock(); 3042 3043 return err; 3044 } 3045 EXPORT_SYMBOL(rtrs_clt_request); 3046 3047 int rtrs_clt_rdma_cq_direct(struct rtrs_clt_sess *clt, unsigned int index) 3048 { 3049 /* If no path, return -1 for block layer not to try again */ 3050 int cnt = -1; 3051 struct rtrs_con *con; 3052 struct rtrs_clt_path *clt_path; 3053 struct path_it it; 3054 3055 rcu_read_lock(); 3056 for (path_it_init(&it, clt); 3057 (clt_path = it.next_path(&it)) && it.i < it.clt->paths_num; it.i++) { 3058 if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED) 3059 continue; 3060 3061 con = clt_path->s.con[index + 1]; 3062 cnt = ib_process_cq_direct(con->cq, -1); 3063 if (cnt) 3064 break; 3065 } 3066 path_it_deinit(&it); 3067 rcu_read_unlock(); 3068 3069 return cnt; 3070 } 3071 EXPORT_SYMBOL(rtrs_clt_rdma_cq_direct); 3072 3073 /** 3074 * rtrs_clt_query() - queries RTRS session attributes 3075 *@clt: session pointer 3076 *@attr: query results for session attributes. 3077 * Returns: 3078 * 0 on success 3079 * -ECOMM no connection to the server 3080 */ 3081 int rtrs_clt_query(struct rtrs_clt_sess *clt, struct rtrs_attrs *attr) 3082 { 3083 if (!rtrs_clt_is_connected(clt)) 3084 return -ECOMM; 3085 3086 attr->queue_depth = clt->queue_depth; 3087 attr->max_segments = clt->max_segments; 3088 /* Cap max_io_size to min of remote buffer size and the fr pages */ 3089 attr->max_io_size = min_t(int, clt->max_io_size, 3090 clt->max_segments * SZ_4K); 3091 3092 return 0; 3093 } 3094 EXPORT_SYMBOL(rtrs_clt_query); 3095 3096 int rtrs_clt_create_path_from_sysfs(struct rtrs_clt_sess *clt, 3097 struct rtrs_addr *addr) 3098 { 3099 struct rtrs_clt_path *clt_path; 3100 int err; 3101 3102 clt_path = alloc_path(clt, addr, nr_cpu_ids, 0); 3103 if (IS_ERR(clt_path)) 3104 return PTR_ERR(clt_path); 3105 3106 mutex_lock(&clt->paths_mutex); 3107 if (clt->paths_num == 0) { 3108 /* 3109 * When all the paths are removed for a session, 3110 * the addition of the first path is like a new session for 3111 * the storage server 3112 */ 3113 clt_path->for_new_clt = 1; 3114 } 3115 3116 mutex_unlock(&clt->paths_mutex); 3117 3118 /* 3119 * It is totally safe to add path in CONNECTING state: coming 3120 * IO will never grab it. Also it is very important to add 3121 * path before init, since init fires LINK_CONNECTED event. 3122 */ 3123 rtrs_clt_add_path_to_arr(clt_path); 3124 3125 err = init_path(clt_path); 3126 if (err) 3127 goto close_path; 3128 3129 err = rtrs_clt_create_path_files(clt_path); 3130 if (err) 3131 goto close_path; 3132 3133 return 0; 3134 3135 close_path: 3136 rtrs_clt_remove_path_from_arr(clt_path); 3137 rtrs_clt_close_conns(clt_path, true); 3138 free_percpu(clt_path->stats->pcpu_stats); 3139 kfree(clt_path->stats); 3140 free_path(clt_path); 3141 3142 return err; 3143 } 3144 3145 static int rtrs_clt_ib_dev_init(struct rtrs_ib_dev *dev) 3146 { 3147 if (!(dev->ib_dev->attrs.device_cap_flags & 3148 IB_DEVICE_MEM_MGT_EXTENSIONS)) { 3149 pr_err("Memory registrations not supported.\n"); 3150 return -ENOTSUPP; 3151 } 3152 3153 return 0; 3154 } 3155 3156 static const struct rtrs_rdma_dev_pd_ops dev_pd_ops = { 3157 .init = rtrs_clt_ib_dev_init 3158 }; 3159 3160 static int __init rtrs_client_init(void) 3161 { 3162 int ret = 0; 3163 3164 rtrs_rdma_dev_pd_init(0, &dev_pd); 3165 ret = class_register(&rtrs_clt_dev_class); 3166 if (ret) { 3167 pr_err("Failed to create rtrs-client dev class\n"); 3168 return ret; 3169 } 3170 rtrs_wq = alloc_workqueue("rtrs_client_wq", 0, 0); 3171 if (!rtrs_wq) { 3172 class_unregister(&rtrs_clt_dev_class); 3173 return -ENOMEM; 3174 } 3175 3176 return 0; 3177 } 3178 3179 static void __exit rtrs_client_exit(void) 3180 { 3181 destroy_workqueue(rtrs_wq); 3182 class_unregister(&rtrs_clt_dev_class); 3183 rtrs_rdma_dev_pd_deinit(&dev_pd); 3184 } 3185 3186 module_init(rtrs_client_init); 3187 module_exit(rtrs_client_exit); 3188