1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * RDMA Network Block Driver 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/blkdev.h> 15 #include <linux/hdreg.h> 16 #include <linux/scatterlist.h> 17 #include <linux/idr.h> 18 19 #include "rnbd-clt.h" 20 21 MODULE_DESCRIPTION("RDMA Network Block Device Client"); 22 MODULE_LICENSE("GPL"); 23 24 static int rnbd_client_major; 25 static DEFINE_IDA(index_ida); 26 static DEFINE_MUTEX(sess_lock); 27 static LIST_HEAD(sess_list); 28 static struct workqueue_struct *rnbd_clt_wq; 29 30 /* 31 * Maximum number of partitions an instance can have. 32 * 6 bits = 64 minors = 63 partitions (one minor is used for the device itself) 33 */ 34 #define RNBD_PART_BITS 6 35 36 static inline bool rnbd_clt_get_sess(struct rnbd_clt_session *sess) 37 { 38 return refcount_inc_not_zero(&sess->refcount); 39 } 40 41 static void free_sess(struct rnbd_clt_session *sess); 42 43 static void rnbd_clt_put_sess(struct rnbd_clt_session *sess) 44 { 45 might_sleep(); 46 47 if (refcount_dec_and_test(&sess->refcount)) 48 free_sess(sess); 49 } 50 51 static void rnbd_clt_put_dev(struct rnbd_clt_dev *dev) 52 { 53 might_sleep(); 54 55 if (!refcount_dec_and_test(&dev->refcount)) 56 return; 57 58 ida_free(&index_ida, dev->clt_device_id); 59 kfree(dev->hw_queues); 60 kfree(dev->pathname); 61 rnbd_clt_put_sess(dev->sess); 62 mutex_destroy(&dev->lock); 63 kfree(dev); 64 } 65 66 static inline bool rnbd_clt_get_dev(struct rnbd_clt_dev *dev) 67 { 68 return refcount_inc_not_zero(&dev->refcount); 69 } 70 71 static int rnbd_clt_set_dev_attr(struct rnbd_clt_dev *dev, 72 const struct rnbd_msg_open_rsp *rsp) 73 { 74 struct rnbd_clt_session *sess = dev->sess; 75 76 if (!rsp->logical_block_size) 77 return -EINVAL; 78 79 dev->device_id = le32_to_cpu(rsp->device_id); 80 dev->nsectors = le64_to_cpu(rsp->nsectors); 81 dev->logical_block_size = le16_to_cpu(rsp->logical_block_size); 82 dev->physical_block_size = le16_to_cpu(rsp->physical_block_size); 83 dev->max_discard_sectors = le32_to_cpu(rsp->max_discard_sectors); 84 dev->discard_granularity = le32_to_cpu(rsp->discard_granularity); 85 dev->discard_alignment = le32_to_cpu(rsp->discard_alignment); 86 dev->secure_discard = le16_to_cpu(rsp->secure_discard); 87 dev->wc = !!(rsp->cache_policy & RNBD_WRITEBACK); 88 dev->fua = !!(rsp->cache_policy & RNBD_FUA); 89 90 dev->max_hw_sectors = sess->max_io_size / SECTOR_SIZE; 91 dev->max_segments = sess->max_segments; 92 93 return 0; 94 } 95 96 static int rnbd_clt_change_capacity(struct rnbd_clt_dev *dev, 97 size_t new_nsectors) 98 { 99 rnbd_clt_info(dev, "Device size changed from %zu to %zu sectors\n", 100 dev->nsectors, new_nsectors); 101 dev->nsectors = new_nsectors; 102 set_capacity_and_notify(dev->gd, dev->nsectors); 103 return 0; 104 } 105 106 static int process_msg_open_rsp(struct rnbd_clt_dev *dev, 107 struct rnbd_msg_open_rsp *rsp) 108 { 109 struct kobject *gd_kobj; 110 int err = 0; 111 112 mutex_lock(&dev->lock); 113 if (dev->dev_state == DEV_STATE_UNMAPPED) { 114 rnbd_clt_info(dev, 115 "Ignoring Open-Response message from server for unmapped device\n"); 116 err = -ENOENT; 117 goto out; 118 } 119 if (dev->dev_state == DEV_STATE_MAPPED_DISCONNECTED) { 120 u64 nsectors = le64_to_cpu(rsp->nsectors); 121 122 /* 123 * If the device was remapped and the size changed in the 124 * meantime we need to revalidate it 125 */ 126 if (dev->nsectors != nsectors) 127 rnbd_clt_change_capacity(dev, nsectors); 128 gd_kobj = &disk_to_dev(dev->gd)->kobj; 129 kobject_uevent(gd_kobj, KOBJ_ONLINE); 130 rnbd_clt_info(dev, "Device online, device remapped successfully\n"); 131 } 132 err = rnbd_clt_set_dev_attr(dev, rsp); 133 if (err) 134 goto out; 135 dev->dev_state = DEV_STATE_MAPPED; 136 137 out: 138 mutex_unlock(&dev->lock); 139 140 return err; 141 } 142 143 int rnbd_clt_resize_disk(struct rnbd_clt_dev *dev, size_t newsize) 144 { 145 int ret = 0; 146 147 mutex_lock(&dev->lock); 148 if (dev->dev_state != DEV_STATE_MAPPED) { 149 pr_err("Failed to set new size of the device, device is not opened\n"); 150 ret = -ENOENT; 151 goto out; 152 } 153 ret = rnbd_clt_change_capacity(dev, newsize); 154 155 out: 156 mutex_unlock(&dev->lock); 157 158 return ret; 159 } 160 161 static inline void rnbd_clt_dev_requeue(struct rnbd_queue *q) 162 { 163 if (WARN_ON(!q->hctx)) 164 return; 165 166 /* We can come here from interrupt, thus async=true */ 167 blk_mq_run_hw_queue(q->hctx, true); 168 } 169 170 enum { 171 RNBD_DELAY_IFBUSY = -1, 172 }; 173 174 /** 175 * rnbd_get_cpu_qlist() - finds a list with HW queues to be rerun 176 * @sess: Session to find a queue for 177 * @cpu: Cpu to start the search from 178 * 179 * Description: 180 * Each CPU has a list of HW queues, which needs to be rerun. If a list 181 * is not empty - it is marked with a bit. This function finds first 182 * set bit in a bitmap and returns corresponding CPU list. 183 */ 184 static struct rnbd_cpu_qlist * 185 rnbd_get_cpu_qlist(struct rnbd_clt_session *sess, int cpu) 186 { 187 int bit; 188 189 /* Search from cpu to nr_cpu_ids */ 190 bit = find_next_bit(sess->cpu_queues_bm, nr_cpu_ids, cpu); 191 if (bit < nr_cpu_ids) { 192 return per_cpu_ptr(sess->cpu_queues, bit); 193 } else if (cpu != 0) { 194 /* Search from 0 to cpu */ 195 bit = find_first_bit(sess->cpu_queues_bm, cpu); 196 if (bit < cpu) 197 return per_cpu_ptr(sess->cpu_queues, bit); 198 } 199 200 return NULL; 201 } 202 203 static inline int nxt_cpu(int cpu) 204 { 205 return (cpu + 1) % nr_cpu_ids; 206 } 207 208 /** 209 * rnbd_rerun_if_needed() - rerun next queue marked as stopped 210 * @sess: Session to rerun a queue on 211 * 212 * Description: 213 * Each CPU has it's own list of HW queues, which should be rerun. 214 * Function finds such list with HW queues, takes a list lock, picks up 215 * the first HW queue out of the list and requeues it. 216 * 217 * Return: 218 * True if the queue was requeued, false otherwise. 219 * 220 * Context: 221 * Does not matter. 222 */ 223 static bool rnbd_rerun_if_needed(struct rnbd_clt_session *sess) 224 { 225 struct rnbd_queue *q = NULL; 226 struct rnbd_cpu_qlist *cpu_q; 227 unsigned long flags; 228 int *cpup; 229 230 /* 231 * To keep fairness and not to let other queues starve we always 232 * try to wake up someone else in round-robin manner. That of course 233 * increases latency but queues always have a chance to be executed. 234 */ 235 cpup = get_cpu_ptr(sess->cpu_rr); 236 for (cpu_q = rnbd_get_cpu_qlist(sess, nxt_cpu(*cpup)); cpu_q; 237 cpu_q = rnbd_get_cpu_qlist(sess, nxt_cpu(cpu_q->cpu))) { 238 if (!spin_trylock_irqsave(&cpu_q->requeue_lock, flags)) 239 continue; 240 if (!test_bit(cpu_q->cpu, sess->cpu_queues_bm)) 241 goto unlock; 242 q = list_first_entry_or_null(&cpu_q->requeue_list, 243 typeof(*q), requeue_list); 244 if (WARN_ON(!q)) 245 goto clear_bit; 246 list_del_init(&q->requeue_list); 247 clear_bit_unlock(0, &q->in_list); 248 249 if (list_empty(&cpu_q->requeue_list)) { 250 /* Clear bit if nothing is left */ 251 clear_bit: 252 clear_bit(cpu_q->cpu, sess->cpu_queues_bm); 253 } 254 unlock: 255 spin_unlock_irqrestore(&cpu_q->requeue_lock, flags); 256 257 if (q) 258 break; 259 } 260 261 /** 262 * Saves the CPU that is going to be requeued on the per-cpu var. Just 263 * incrementing it doesn't work because rnbd_get_cpu_qlist() will 264 * always return the first CPU with something on the queue list when the 265 * value stored on the var is greater than the last CPU with something 266 * on the list. 267 */ 268 if (cpu_q) 269 *cpup = cpu_q->cpu; 270 put_cpu_ptr(sess->cpu_rr); 271 272 if (q) 273 rnbd_clt_dev_requeue(q); 274 275 return q; 276 } 277 278 /** 279 * rnbd_rerun_all_if_idle() - rerun all queues left in the list if 280 * session is idling (there are no requests 281 * in-flight). 282 * @sess: Session to rerun the queues on 283 * 284 * Description: 285 * This function tries to rerun all stopped queues if there are no 286 * requests in-flight anymore. This function tries to solve an obvious 287 * problem, when number of tags < than number of queues (hctx), which 288 * are stopped and put to sleep. If last permit, which has been just put, 289 * does not wake up all left queues (hctxs), IO requests hang forever. 290 * 291 * That can happen when all number of permits, say N, have been exhausted 292 * from one CPU, and we have many block devices per session, say M. 293 * Each block device has it's own queue (hctx) for each CPU, so eventually 294 * we can put that number of queues (hctxs) to sleep: M x nr_cpu_ids. 295 * If number of permits N < M x nr_cpu_ids finally we will get an IO hang. 296 * 297 * To avoid this hang last caller of rnbd_put_permit() (last caller is the 298 * one who observes sess->busy == 0) must wake up all remaining queues. 299 * 300 * Context: 301 * Does not matter. 302 */ 303 static void rnbd_rerun_all_if_idle(struct rnbd_clt_session *sess) 304 { 305 bool requeued; 306 307 do { 308 requeued = rnbd_rerun_if_needed(sess); 309 } while (atomic_read(&sess->busy) == 0 && requeued); 310 } 311 312 static struct rtrs_permit *rnbd_get_permit(struct rnbd_clt_session *sess, 313 enum rtrs_clt_con_type con_type, 314 enum wait_type wait) 315 { 316 struct rtrs_permit *permit; 317 318 permit = rtrs_clt_get_permit(sess->rtrs, con_type, wait); 319 if (permit) 320 /* We have a subtle rare case here, when all permits can be 321 * consumed before busy counter increased. This is safe, 322 * because loser will get NULL as a permit, observe 0 busy 323 * counter and immediately restart the queue himself. 324 */ 325 atomic_inc(&sess->busy); 326 327 return permit; 328 } 329 330 static void rnbd_put_permit(struct rnbd_clt_session *sess, 331 struct rtrs_permit *permit) 332 { 333 rtrs_clt_put_permit(sess->rtrs, permit); 334 atomic_dec(&sess->busy); 335 /* Paired with rnbd_clt_dev_add_to_requeue(). Decrement first 336 * and then check queue bits. 337 */ 338 smp_mb__after_atomic(); 339 rnbd_rerun_all_if_idle(sess); 340 } 341 342 static struct rnbd_iu *rnbd_get_iu(struct rnbd_clt_session *sess, 343 enum rtrs_clt_con_type con_type, 344 enum wait_type wait) 345 { 346 struct rnbd_iu *iu; 347 struct rtrs_permit *permit; 348 349 iu = kzalloc(sizeof(*iu), GFP_KERNEL); 350 if (!iu) 351 return NULL; 352 353 permit = rnbd_get_permit(sess, con_type, wait); 354 if (!permit) { 355 kfree(iu); 356 return NULL; 357 } 358 359 iu->permit = permit; 360 /* 361 * 1st reference is dropped after finishing sending a "user" message, 362 * 2nd reference is dropped after confirmation with the response is 363 * returned. 364 * 1st and 2nd can happen in any order, so the rnbd_iu should be 365 * released (rtrs_permit returned to rtrs) only after both 366 * are finished. 367 */ 368 atomic_set(&iu->refcount, 2); 369 init_waitqueue_head(&iu->comp.wait); 370 iu->comp.errno = INT_MAX; 371 372 if (sg_alloc_table(&iu->sgt, 1, GFP_KERNEL)) { 373 rnbd_put_permit(sess, permit); 374 kfree(iu); 375 return NULL; 376 } 377 378 return iu; 379 } 380 381 static void rnbd_put_iu(struct rnbd_clt_session *sess, struct rnbd_iu *iu) 382 { 383 if (atomic_dec_and_test(&iu->refcount)) { 384 sg_free_table(&iu->sgt); 385 rnbd_put_permit(sess, iu->permit); 386 kfree(iu); 387 } 388 } 389 390 static void rnbd_softirq_done_fn(struct request *rq) 391 { 392 struct rnbd_clt_dev *dev = rq->q->disk->private_data; 393 struct rnbd_clt_session *sess = dev->sess; 394 struct rnbd_iu *iu; 395 396 iu = blk_mq_rq_to_pdu(rq); 397 sg_free_table_chained(&iu->sgt, RNBD_INLINE_SG_CNT); 398 rnbd_put_permit(sess, iu->permit); 399 blk_mq_end_request(rq, errno_to_blk_status(iu->errno)); 400 } 401 402 static void msg_io_conf(void *priv, int errno) 403 { 404 struct rnbd_iu *iu = priv; 405 struct rnbd_clt_dev *dev = iu->dev; 406 struct request *rq = iu->rq; 407 int rw = rq_data_dir(rq); 408 409 iu->errno = errno; 410 411 blk_mq_complete_request(rq); 412 413 if (errno) 414 rnbd_clt_info_rl(dev, "%s I/O failed with err: %d\n", 415 rw == READ ? "read" : "write", errno); 416 } 417 418 static void wake_up_iu_comp(struct rnbd_iu *iu, int errno) 419 { 420 iu->comp.errno = errno; 421 wake_up(&iu->comp.wait); 422 } 423 424 static void msg_conf(void *priv, int errno) 425 { 426 struct rnbd_iu *iu = priv; 427 428 iu->errno = errno; 429 schedule_work(&iu->work); 430 } 431 432 static int send_usr_msg(struct rtrs_clt_sess *rtrs, int dir, 433 struct rnbd_iu *iu, struct kvec *vec, 434 size_t len, struct scatterlist *sg, unsigned int sg_len, 435 void (*conf)(struct work_struct *work), 436 int *errno, int wait) 437 { 438 int err; 439 struct rtrs_clt_req_ops req_ops; 440 441 INIT_WORK(&iu->work, conf); 442 req_ops = (struct rtrs_clt_req_ops) { 443 .priv = iu, 444 .conf_fn = msg_conf, 445 }; 446 err = rtrs_clt_request(dir, &req_ops, rtrs, iu->permit, 447 vec, 1, len, sg, sg_len); 448 if (!err && wait) { 449 wait_event(iu->comp.wait, iu->comp.errno != INT_MAX); 450 *errno = iu->comp.errno; 451 } else { 452 *errno = 0; 453 } 454 455 return err; 456 } 457 458 static void msg_close_conf(struct work_struct *work) 459 { 460 struct rnbd_iu *iu = container_of(work, struct rnbd_iu, work); 461 struct rnbd_clt_dev *dev = iu->dev; 462 463 wake_up_iu_comp(iu, iu->errno); 464 rnbd_put_iu(dev->sess, iu); 465 rnbd_clt_put_dev(dev); 466 } 467 468 static int send_msg_close(struct rnbd_clt_dev *dev, u32 device_id, 469 enum wait_type wait) 470 { 471 struct rnbd_clt_session *sess = dev->sess; 472 struct rnbd_msg_close msg; 473 struct rnbd_iu *iu; 474 struct kvec vec = { 475 .iov_base = &msg, 476 .iov_len = sizeof(msg) 477 }; 478 int err, errno; 479 480 iu = rnbd_get_iu(sess, RTRS_ADMIN_CON, RTRS_PERMIT_WAIT); 481 if (!iu) 482 return -ENOMEM; 483 484 iu->buf = NULL; 485 iu->dev = dev; 486 487 msg.hdr.type = cpu_to_le16(RNBD_MSG_CLOSE); 488 msg.device_id = cpu_to_le32(device_id); 489 490 WARN_ON(!rnbd_clt_get_dev(dev)); 491 err = send_usr_msg(sess->rtrs, WRITE, iu, &vec, 0, NULL, 0, 492 msg_close_conf, &errno, wait); 493 if (err) { 494 rnbd_clt_put_dev(dev); 495 rnbd_put_iu(sess, iu); 496 } else { 497 err = errno; 498 } 499 500 rnbd_put_iu(sess, iu); 501 return err; 502 } 503 504 static void msg_open_conf(struct work_struct *work) 505 { 506 struct rnbd_iu *iu = container_of(work, struct rnbd_iu, work); 507 struct rnbd_msg_open_rsp *rsp = iu->buf; 508 struct rnbd_clt_dev *dev = iu->dev; 509 int errno = iu->errno; 510 511 if (errno) { 512 rnbd_clt_err(dev, 513 "Opening failed, server responded: %d\n", 514 errno); 515 } else { 516 errno = process_msg_open_rsp(dev, rsp); 517 if (errno) { 518 u32 device_id = le32_to_cpu(rsp->device_id); 519 /* 520 * If server thinks its fine, but we fail to process 521 * then be nice and send a close to server. 522 */ 523 send_msg_close(dev, device_id, RTRS_PERMIT_NOWAIT); 524 } 525 } 526 kfree(rsp); 527 wake_up_iu_comp(iu, errno); 528 rnbd_put_iu(dev->sess, iu); 529 rnbd_clt_put_dev(dev); 530 } 531 532 static void msg_sess_info_conf(struct work_struct *work) 533 { 534 struct rnbd_iu *iu = container_of(work, struct rnbd_iu, work); 535 struct rnbd_msg_sess_info_rsp *rsp = iu->buf; 536 struct rnbd_clt_session *sess = iu->sess; 537 538 if (!iu->errno) 539 sess->ver = min_t(u8, rsp->ver, RNBD_PROTO_VER_MAJOR); 540 541 kfree(rsp); 542 wake_up_iu_comp(iu, iu->errno); 543 rnbd_put_iu(sess, iu); 544 rnbd_clt_put_sess(sess); 545 } 546 547 static int send_msg_open(struct rnbd_clt_dev *dev, enum wait_type wait) 548 { 549 struct rnbd_clt_session *sess = dev->sess; 550 struct rnbd_msg_open_rsp *rsp; 551 struct rnbd_msg_open msg; 552 struct rnbd_iu *iu; 553 struct kvec vec = { 554 .iov_base = &msg, 555 .iov_len = sizeof(msg) 556 }; 557 int err, errno; 558 559 rsp = kzalloc(sizeof(*rsp), GFP_KERNEL); 560 if (!rsp) 561 return -ENOMEM; 562 563 iu = rnbd_get_iu(sess, RTRS_ADMIN_CON, RTRS_PERMIT_WAIT); 564 if (!iu) { 565 kfree(rsp); 566 return -ENOMEM; 567 } 568 569 iu->buf = rsp; 570 iu->dev = dev; 571 572 sg_init_one(iu->sgt.sgl, rsp, sizeof(*rsp)); 573 574 msg.hdr.type = cpu_to_le16(RNBD_MSG_OPEN); 575 msg.access_mode = dev->access_mode; 576 strscpy(msg.dev_name, dev->pathname, sizeof(msg.dev_name)); 577 578 WARN_ON(!rnbd_clt_get_dev(dev)); 579 err = send_usr_msg(sess->rtrs, READ, iu, 580 &vec, sizeof(*rsp), iu->sgt.sgl, 1, 581 msg_open_conf, &errno, wait); 582 if (err) { 583 rnbd_clt_put_dev(dev); 584 rnbd_put_iu(sess, iu); 585 kfree(rsp); 586 } else { 587 err = errno; 588 } 589 590 rnbd_put_iu(sess, iu); 591 return err; 592 } 593 594 static int send_msg_sess_info(struct rnbd_clt_session *sess, enum wait_type wait) 595 { 596 struct rnbd_msg_sess_info_rsp *rsp; 597 struct rnbd_msg_sess_info msg; 598 struct rnbd_iu *iu; 599 struct kvec vec = { 600 .iov_base = &msg, 601 .iov_len = sizeof(msg) 602 }; 603 int err, errno; 604 605 rsp = kzalloc(sizeof(*rsp), GFP_KERNEL); 606 if (!rsp) 607 return -ENOMEM; 608 609 iu = rnbd_get_iu(sess, RTRS_ADMIN_CON, RTRS_PERMIT_WAIT); 610 if (!iu) { 611 kfree(rsp); 612 return -ENOMEM; 613 } 614 615 iu->buf = rsp; 616 iu->sess = sess; 617 sg_init_one(iu->sgt.sgl, rsp, sizeof(*rsp)); 618 619 msg.hdr.type = cpu_to_le16(RNBD_MSG_SESS_INFO); 620 msg.ver = RNBD_PROTO_VER_MAJOR; 621 622 if (!rnbd_clt_get_sess(sess)) { 623 /* 624 * That can happen only in one case, when RTRS has restablished 625 * the connection and link_ev() is called, but session is almost 626 * dead, last reference on session is put and caller is waiting 627 * for RTRS to close everything. 628 */ 629 err = -ENODEV; 630 goto put_iu; 631 } 632 err = send_usr_msg(sess->rtrs, READ, iu, 633 &vec, sizeof(*rsp), iu->sgt.sgl, 1, 634 msg_sess_info_conf, &errno, wait); 635 if (err) { 636 rnbd_clt_put_sess(sess); 637 put_iu: 638 rnbd_put_iu(sess, iu); 639 kfree(rsp); 640 } else { 641 err = errno; 642 } 643 rnbd_put_iu(sess, iu); 644 return err; 645 } 646 647 static void set_dev_states_to_disconnected(struct rnbd_clt_session *sess) 648 { 649 struct rnbd_clt_dev *dev; 650 struct kobject *gd_kobj; 651 652 mutex_lock(&sess->lock); 653 list_for_each_entry(dev, &sess->devs_list, list) { 654 rnbd_clt_err(dev, "Device disconnected.\n"); 655 656 mutex_lock(&dev->lock); 657 if (dev->dev_state == DEV_STATE_MAPPED) { 658 dev->dev_state = DEV_STATE_MAPPED_DISCONNECTED; 659 gd_kobj = &disk_to_dev(dev->gd)->kobj; 660 kobject_uevent(gd_kobj, KOBJ_OFFLINE); 661 } 662 mutex_unlock(&dev->lock); 663 } 664 mutex_unlock(&sess->lock); 665 } 666 667 static void remap_devs(struct rnbd_clt_session *sess) 668 { 669 struct rnbd_clt_dev *dev; 670 struct rtrs_attrs attrs; 671 int err; 672 673 /* 674 * Careful here: we are called from RTRS link event directly, 675 * thus we can't send any RTRS request and wait for response 676 * or RTRS will not be able to complete request with failure 677 * if something goes wrong (failing of outstanding requests 678 * happens exactly from the context where we are blocking now). 679 * 680 * So to avoid deadlocks each usr message sent from here must 681 * be asynchronous. 682 */ 683 684 err = send_msg_sess_info(sess, RTRS_PERMIT_NOWAIT); 685 if (err) { 686 pr_err("send_msg_sess_info(\"%s\"): %d\n", sess->sessname, err); 687 return; 688 } 689 690 err = rtrs_clt_query(sess->rtrs, &attrs); 691 if (err) { 692 pr_err("rtrs_clt_query(\"%s\"): %d\n", sess->sessname, err); 693 return; 694 } 695 mutex_lock(&sess->lock); 696 sess->max_io_size = attrs.max_io_size; 697 698 list_for_each_entry(dev, &sess->devs_list, list) { 699 bool skip; 700 701 mutex_lock(&dev->lock); 702 skip = (dev->dev_state == DEV_STATE_INIT); 703 mutex_unlock(&dev->lock); 704 if (skip) 705 /* 706 * When device is establishing connection for the first 707 * time - do not remap, it will be closed soon. 708 */ 709 continue; 710 711 rnbd_clt_info(dev, "session reconnected, remapping device\n"); 712 err = send_msg_open(dev, RTRS_PERMIT_NOWAIT); 713 if (err) { 714 rnbd_clt_err(dev, "send_msg_open(): %d\n", err); 715 break; 716 } 717 } 718 mutex_unlock(&sess->lock); 719 } 720 721 static void rnbd_clt_link_ev(void *priv, enum rtrs_clt_link_ev ev) 722 { 723 struct rnbd_clt_session *sess = priv; 724 725 switch (ev) { 726 case RTRS_CLT_LINK_EV_DISCONNECTED: 727 set_dev_states_to_disconnected(sess); 728 break; 729 case RTRS_CLT_LINK_EV_RECONNECTED: 730 remap_devs(sess); 731 break; 732 default: 733 pr_err("Unknown session event received (%d), session: %s\n", 734 ev, sess->sessname); 735 } 736 } 737 738 static void rnbd_init_cpu_qlists(struct rnbd_cpu_qlist __percpu *cpu_queues) 739 { 740 unsigned int cpu; 741 struct rnbd_cpu_qlist *cpu_q; 742 743 for_each_possible_cpu(cpu) { 744 cpu_q = per_cpu_ptr(cpu_queues, cpu); 745 746 cpu_q->cpu = cpu; 747 INIT_LIST_HEAD(&cpu_q->requeue_list); 748 spin_lock_init(&cpu_q->requeue_lock); 749 } 750 } 751 752 static void destroy_mq_tags(struct rnbd_clt_session *sess) 753 { 754 if (sess->tag_set.tags) 755 blk_mq_free_tag_set(&sess->tag_set); 756 } 757 758 static inline void wake_up_rtrs_waiters(struct rnbd_clt_session *sess) 759 { 760 sess->rtrs_ready = true; 761 wake_up_all(&sess->rtrs_waitq); 762 } 763 764 static void close_rtrs(struct rnbd_clt_session *sess) 765 { 766 might_sleep(); 767 768 if (!IS_ERR_OR_NULL(sess->rtrs)) { 769 rtrs_clt_close(sess->rtrs); 770 sess->rtrs = NULL; 771 wake_up_rtrs_waiters(sess); 772 } 773 } 774 775 static void free_sess(struct rnbd_clt_session *sess) 776 { 777 WARN_ON(!list_empty(&sess->devs_list)); 778 779 might_sleep(); 780 781 close_rtrs(sess); 782 destroy_mq_tags(sess); 783 if (!list_empty(&sess->list)) { 784 mutex_lock(&sess_lock); 785 list_del(&sess->list); 786 mutex_unlock(&sess_lock); 787 } 788 free_percpu(sess->cpu_queues); 789 free_percpu(sess->cpu_rr); 790 mutex_destroy(&sess->lock); 791 kfree(sess); 792 } 793 794 static struct rnbd_clt_session *alloc_sess(const char *sessname) 795 { 796 struct rnbd_clt_session *sess; 797 int err, cpu; 798 799 sess = kzalloc_node(sizeof(*sess), GFP_KERNEL, NUMA_NO_NODE); 800 if (!sess) 801 return ERR_PTR(-ENOMEM); 802 strscpy(sess->sessname, sessname, sizeof(sess->sessname)); 803 atomic_set(&sess->busy, 0); 804 mutex_init(&sess->lock); 805 INIT_LIST_HEAD(&sess->devs_list); 806 INIT_LIST_HEAD(&sess->list); 807 bitmap_zero(sess->cpu_queues_bm, num_possible_cpus()); 808 init_waitqueue_head(&sess->rtrs_waitq); 809 refcount_set(&sess->refcount, 1); 810 811 sess->cpu_queues = alloc_percpu(struct rnbd_cpu_qlist); 812 if (!sess->cpu_queues) { 813 err = -ENOMEM; 814 goto err; 815 } 816 rnbd_init_cpu_qlists(sess->cpu_queues); 817 818 /* 819 * That is simple percpu variable which stores cpu indices, which are 820 * incremented on each access. We need that for the sake of fairness 821 * to wake up queues in a round-robin manner. 822 */ 823 sess->cpu_rr = alloc_percpu(int); 824 if (!sess->cpu_rr) { 825 err = -ENOMEM; 826 goto err; 827 } 828 for_each_possible_cpu(cpu) 829 * per_cpu_ptr(sess->cpu_rr, cpu) = cpu; 830 831 return sess; 832 833 err: 834 free_sess(sess); 835 836 return ERR_PTR(err); 837 } 838 839 static int wait_for_rtrs_connection(struct rnbd_clt_session *sess) 840 { 841 wait_event(sess->rtrs_waitq, sess->rtrs_ready); 842 if (IS_ERR_OR_NULL(sess->rtrs)) 843 return -ECONNRESET; 844 845 return 0; 846 } 847 848 static void wait_for_rtrs_disconnection(struct rnbd_clt_session *sess) 849 __releases(&sess_lock) 850 __acquires(&sess_lock) 851 { 852 DEFINE_WAIT(wait); 853 854 prepare_to_wait(&sess->rtrs_waitq, &wait, TASK_UNINTERRUPTIBLE); 855 if (IS_ERR_OR_NULL(sess->rtrs)) { 856 finish_wait(&sess->rtrs_waitq, &wait); 857 return; 858 } 859 mutex_unlock(&sess_lock); 860 /* loop in caller, see __find_and_get_sess(). 861 * You can't leave mutex locked and call schedule(), you will catch a 862 * deadlock with a caller of free_sess(), which has just put the last 863 * reference and is about to take the sess_lock in order to delete 864 * the session from the list. 865 */ 866 schedule(); 867 mutex_lock(&sess_lock); 868 } 869 870 static struct rnbd_clt_session *__find_and_get_sess(const char *sessname) 871 __releases(&sess_lock) 872 __acquires(&sess_lock) 873 { 874 struct rnbd_clt_session *sess, *sn; 875 int err; 876 877 again: 878 list_for_each_entry_safe(sess, sn, &sess_list, list) { 879 if (strcmp(sessname, sess->sessname)) 880 continue; 881 882 if (sess->rtrs_ready && IS_ERR_OR_NULL(sess->rtrs)) 883 /* 884 * No RTRS connection, session is dying. 885 */ 886 continue; 887 888 if (rnbd_clt_get_sess(sess)) { 889 /* 890 * Alive session is found, wait for RTRS connection. 891 */ 892 mutex_unlock(&sess_lock); 893 err = wait_for_rtrs_connection(sess); 894 if (err) 895 rnbd_clt_put_sess(sess); 896 mutex_lock(&sess_lock); 897 898 if (err) 899 /* Session is dying, repeat the loop */ 900 goto again; 901 902 return sess; 903 } 904 /* 905 * Ref is 0, session is dying, wait for RTRS disconnect 906 * in order to avoid session names clashes. 907 */ 908 wait_for_rtrs_disconnection(sess); 909 /* 910 * RTRS is disconnected and soon session will be freed, 911 * so repeat a loop. 912 */ 913 goto again; 914 } 915 916 return NULL; 917 } 918 919 /* caller is responsible for initializing 'first' to false */ 920 static struct 921 rnbd_clt_session *find_or_create_sess(const char *sessname, bool *first) 922 { 923 struct rnbd_clt_session *sess = NULL; 924 925 mutex_lock(&sess_lock); 926 sess = __find_and_get_sess(sessname); 927 if (!sess) { 928 sess = alloc_sess(sessname); 929 if (IS_ERR(sess)) { 930 mutex_unlock(&sess_lock); 931 return sess; 932 } 933 list_add(&sess->list, &sess_list); 934 *first = true; 935 } 936 mutex_unlock(&sess_lock); 937 938 return sess; 939 } 940 941 static int rnbd_client_open(struct block_device *block_device, fmode_t mode) 942 { 943 struct rnbd_clt_dev *dev = block_device->bd_disk->private_data; 944 945 if (dev->read_only && (mode & FMODE_WRITE)) 946 return -EPERM; 947 948 if (dev->dev_state == DEV_STATE_UNMAPPED || 949 !rnbd_clt_get_dev(dev)) 950 return -EIO; 951 952 return 0; 953 } 954 955 static void rnbd_client_release(struct gendisk *gen, fmode_t mode) 956 { 957 struct rnbd_clt_dev *dev = gen->private_data; 958 959 rnbd_clt_put_dev(dev); 960 } 961 962 static int rnbd_client_getgeo(struct block_device *block_device, 963 struct hd_geometry *geo) 964 { 965 u64 size; 966 struct rnbd_clt_dev *dev; 967 968 dev = block_device->bd_disk->private_data; 969 size = dev->size * (dev->logical_block_size / SECTOR_SIZE); 970 geo->cylinders = size >> 6; /* size/64 */ 971 geo->heads = 4; 972 geo->sectors = 16; 973 geo->start = 0; 974 975 return 0; 976 } 977 978 static const struct block_device_operations rnbd_client_ops = { 979 .owner = THIS_MODULE, 980 .open = rnbd_client_open, 981 .release = rnbd_client_release, 982 .getgeo = rnbd_client_getgeo 983 }; 984 985 /* The amount of data that belongs to an I/O and the amount of data that 986 * should be read or written to the disk (bi_size) can differ. 987 * 988 * E.g. When WRITE_SAME is used, only a small amount of data is 989 * transferred that is then written repeatedly over a lot of sectors. 990 * 991 * Get the size of data to be transferred via RTRS by summing up the size 992 * of the scather-gather list entries. 993 */ 994 static size_t rnbd_clt_get_sg_size(struct scatterlist *sglist, u32 len) 995 { 996 struct scatterlist *sg; 997 size_t tsize = 0; 998 int i; 999 1000 for_each_sg(sglist, sg, len, i) 1001 tsize += sg->length; 1002 return tsize; 1003 } 1004 1005 static int rnbd_client_xfer_request(struct rnbd_clt_dev *dev, 1006 struct request *rq, 1007 struct rnbd_iu *iu) 1008 { 1009 struct rtrs_clt_sess *rtrs = dev->sess->rtrs; 1010 struct rtrs_permit *permit = iu->permit; 1011 struct rnbd_msg_io msg; 1012 struct rtrs_clt_req_ops req_ops; 1013 unsigned int sg_cnt = 0; 1014 struct kvec vec; 1015 size_t size; 1016 int err; 1017 1018 iu->rq = rq; 1019 iu->dev = dev; 1020 msg.sector = cpu_to_le64(blk_rq_pos(rq)); 1021 msg.bi_size = cpu_to_le32(blk_rq_bytes(rq)); 1022 msg.rw = cpu_to_le32(rq_to_rnbd_flags(rq)); 1023 msg.prio = cpu_to_le16(req_get_ioprio(rq)); 1024 1025 /* 1026 * We only support discards with single segment for now. 1027 * See queue limits. 1028 */ 1029 if (req_op(rq) != REQ_OP_DISCARD) 1030 sg_cnt = blk_rq_map_sg(dev->queue, rq, iu->sgt.sgl); 1031 1032 if (sg_cnt == 0) 1033 sg_mark_end(&iu->sgt.sgl[0]); 1034 1035 msg.hdr.type = cpu_to_le16(RNBD_MSG_IO); 1036 msg.device_id = cpu_to_le32(dev->device_id); 1037 1038 vec = (struct kvec) { 1039 .iov_base = &msg, 1040 .iov_len = sizeof(msg) 1041 }; 1042 size = rnbd_clt_get_sg_size(iu->sgt.sgl, sg_cnt); 1043 req_ops = (struct rtrs_clt_req_ops) { 1044 .priv = iu, 1045 .conf_fn = msg_io_conf, 1046 }; 1047 err = rtrs_clt_request(rq_data_dir(rq), &req_ops, rtrs, permit, 1048 &vec, 1, size, iu->sgt.sgl, sg_cnt); 1049 if (err) { 1050 rnbd_clt_err_rl(dev, "RTRS failed to transfer IO, err: %d\n", 1051 err); 1052 return err; 1053 } 1054 1055 return 0; 1056 } 1057 1058 /** 1059 * rnbd_clt_dev_add_to_requeue() - add device to requeue if session is busy 1060 * @dev: Device to be checked 1061 * @q: Queue to be added to the requeue list if required 1062 * 1063 * Description: 1064 * If session is busy, that means someone will requeue us when resources 1065 * are freed. If session is not doing anything - device is not added to 1066 * the list and @false is returned. 1067 */ 1068 static bool rnbd_clt_dev_add_to_requeue(struct rnbd_clt_dev *dev, 1069 struct rnbd_queue *q) 1070 { 1071 struct rnbd_clt_session *sess = dev->sess; 1072 struct rnbd_cpu_qlist *cpu_q; 1073 unsigned long flags; 1074 bool added = true; 1075 bool need_set; 1076 1077 cpu_q = get_cpu_ptr(sess->cpu_queues); 1078 spin_lock_irqsave(&cpu_q->requeue_lock, flags); 1079 1080 if (!test_and_set_bit_lock(0, &q->in_list)) { 1081 if (WARN_ON(!list_empty(&q->requeue_list))) 1082 goto unlock; 1083 1084 need_set = !test_bit(cpu_q->cpu, sess->cpu_queues_bm); 1085 if (need_set) { 1086 set_bit(cpu_q->cpu, sess->cpu_queues_bm); 1087 /* Paired with rnbd_put_permit(). Set a bit first 1088 * and then observe the busy counter. 1089 */ 1090 smp_mb__before_atomic(); 1091 } 1092 if (atomic_read(&sess->busy)) { 1093 list_add_tail(&q->requeue_list, &cpu_q->requeue_list); 1094 } else { 1095 /* Very unlikely, but possible: busy counter was 1096 * observed as zero. Drop all bits and return 1097 * false to restart the queue by ourselves. 1098 */ 1099 if (need_set) 1100 clear_bit(cpu_q->cpu, sess->cpu_queues_bm); 1101 clear_bit_unlock(0, &q->in_list); 1102 added = false; 1103 } 1104 } 1105 unlock: 1106 spin_unlock_irqrestore(&cpu_q->requeue_lock, flags); 1107 put_cpu_ptr(sess->cpu_queues); 1108 1109 return added; 1110 } 1111 1112 static void rnbd_clt_dev_kick_mq_queue(struct rnbd_clt_dev *dev, 1113 struct blk_mq_hw_ctx *hctx, 1114 int delay) 1115 { 1116 struct rnbd_queue *q = hctx->driver_data; 1117 1118 if (delay != RNBD_DELAY_IFBUSY) 1119 blk_mq_delay_run_hw_queue(hctx, delay); 1120 else if (!rnbd_clt_dev_add_to_requeue(dev, q)) 1121 /* 1122 * If session is not busy we have to restart 1123 * the queue ourselves. 1124 */ 1125 blk_mq_delay_run_hw_queue(hctx, 10/*ms*/); 1126 } 1127 1128 static blk_status_t rnbd_queue_rq(struct blk_mq_hw_ctx *hctx, 1129 const struct blk_mq_queue_data *bd) 1130 { 1131 struct request *rq = bd->rq; 1132 struct rnbd_clt_dev *dev = rq->q->disk->private_data; 1133 struct rnbd_iu *iu = blk_mq_rq_to_pdu(rq); 1134 int err; 1135 blk_status_t ret = BLK_STS_IOERR; 1136 1137 if (dev->dev_state != DEV_STATE_MAPPED) 1138 return BLK_STS_IOERR; 1139 1140 iu->permit = rnbd_get_permit(dev->sess, RTRS_IO_CON, 1141 RTRS_PERMIT_NOWAIT); 1142 if (!iu->permit) { 1143 rnbd_clt_dev_kick_mq_queue(dev, hctx, RNBD_DELAY_IFBUSY); 1144 return BLK_STS_RESOURCE; 1145 } 1146 1147 iu->sgt.sgl = iu->first_sgl; 1148 err = sg_alloc_table_chained(&iu->sgt, 1149 /* Even-if the request has no segment, 1150 * sglist must have one entry at least. 1151 */ 1152 blk_rq_nr_phys_segments(rq) ? : 1, 1153 iu->sgt.sgl, 1154 RNBD_INLINE_SG_CNT); 1155 if (err) { 1156 rnbd_clt_err_rl(dev, "sg_alloc_table_chained ret=%d\n", err); 1157 rnbd_clt_dev_kick_mq_queue(dev, hctx, 10/*ms*/); 1158 rnbd_put_permit(dev->sess, iu->permit); 1159 return BLK_STS_RESOURCE; 1160 } 1161 1162 blk_mq_start_request(rq); 1163 err = rnbd_client_xfer_request(dev, rq, iu); 1164 if (err == 0) 1165 return BLK_STS_OK; 1166 if (err == -EAGAIN || err == -ENOMEM) { 1167 rnbd_clt_dev_kick_mq_queue(dev, hctx, 10/*ms*/); 1168 ret = BLK_STS_RESOURCE; 1169 } 1170 sg_free_table_chained(&iu->sgt, RNBD_INLINE_SG_CNT); 1171 rnbd_put_permit(dev->sess, iu->permit); 1172 return ret; 1173 } 1174 1175 static int rnbd_rdma_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob) 1176 { 1177 struct rnbd_queue *q = hctx->driver_data; 1178 struct rnbd_clt_dev *dev = q->dev; 1179 int cnt; 1180 1181 cnt = rtrs_clt_rdma_cq_direct(dev->sess->rtrs, hctx->queue_num); 1182 return cnt; 1183 } 1184 1185 static int rnbd_rdma_map_queues(struct blk_mq_tag_set *set) 1186 { 1187 struct rnbd_clt_session *sess = set->driver_data; 1188 1189 /* shared read/write queues */ 1190 set->map[HCTX_TYPE_DEFAULT].nr_queues = num_online_cpus(); 1191 set->map[HCTX_TYPE_DEFAULT].queue_offset = 0; 1192 set->map[HCTX_TYPE_READ].nr_queues = num_online_cpus(); 1193 set->map[HCTX_TYPE_READ].queue_offset = 0; 1194 blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]); 1195 blk_mq_map_queues(&set->map[HCTX_TYPE_READ]); 1196 1197 if (sess->nr_poll_queues) { 1198 /* dedicated queue for poll */ 1199 set->map[HCTX_TYPE_POLL].nr_queues = sess->nr_poll_queues; 1200 set->map[HCTX_TYPE_POLL].queue_offset = set->map[HCTX_TYPE_READ].queue_offset + 1201 set->map[HCTX_TYPE_READ].nr_queues; 1202 blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]); 1203 pr_info("[session=%s] mapped %d/%d/%d default/read/poll queues.\n", 1204 sess->sessname, 1205 set->map[HCTX_TYPE_DEFAULT].nr_queues, 1206 set->map[HCTX_TYPE_READ].nr_queues, 1207 set->map[HCTX_TYPE_POLL].nr_queues); 1208 } else { 1209 pr_info("[session=%s] mapped %d/%d default/read queues.\n", 1210 sess->sessname, 1211 set->map[HCTX_TYPE_DEFAULT].nr_queues, 1212 set->map[HCTX_TYPE_READ].nr_queues); 1213 } 1214 1215 return 0; 1216 } 1217 1218 static struct blk_mq_ops rnbd_mq_ops = { 1219 .queue_rq = rnbd_queue_rq, 1220 .complete = rnbd_softirq_done_fn, 1221 .map_queues = rnbd_rdma_map_queues, 1222 .poll = rnbd_rdma_poll, 1223 }; 1224 1225 static int setup_mq_tags(struct rnbd_clt_session *sess) 1226 { 1227 struct blk_mq_tag_set *tag_set = &sess->tag_set; 1228 1229 memset(tag_set, 0, sizeof(*tag_set)); 1230 tag_set->ops = &rnbd_mq_ops; 1231 tag_set->queue_depth = sess->queue_depth; 1232 tag_set->numa_node = NUMA_NO_NODE; 1233 tag_set->flags = BLK_MQ_F_SHOULD_MERGE | 1234 BLK_MQ_F_TAG_QUEUE_SHARED; 1235 tag_set->cmd_size = sizeof(struct rnbd_iu) + RNBD_RDMA_SGL_SIZE; 1236 1237 /* for HCTX_TYPE_DEFAULT, HCTX_TYPE_READ, HCTX_TYPE_POLL */ 1238 tag_set->nr_maps = sess->nr_poll_queues ? HCTX_MAX_TYPES : 2; 1239 /* 1240 * HCTX_TYPE_DEFAULT and HCTX_TYPE_READ share one set of queues 1241 * others are for HCTX_TYPE_POLL 1242 */ 1243 tag_set->nr_hw_queues = num_online_cpus() + sess->nr_poll_queues; 1244 tag_set->driver_data = sess; 1245 1246 return blk_mq_alloc_tag_set(tag_set); 1247 } 1248 1249 static struct rnbd_clt_session * 1250 find_and_get_or_create_sess(const char *sessname, 1251 const struct rtrs_addr *paths, 1252 size_t path_cnt, u16 port_nr, u32 nr_poll_queues) 1253 { 1254 struct rnbd_clt_session *sess; 1255 struct rtrs_attrs attrs; 1256 int err; 1257 bool first = false; 1258 struct rtrs_clt_ops rtrs_ops; 1259 1260 sess = find_or_create_sess(sessname, &first); 1261 if (sess == ERR_PTR(-ENOMEM)) { 1262 return ERR_PTR(-ENOMEM); 1263 } else if ((nr_poll_queues && !first) || (!nr_poll_queues && sess->nr_poll_queues)) { 1264 /* 1265 * A device MUST have its own session to use the polling-mode. 1266 * It must fail to map new device with the same session. 1267 */ 1268 err = -EINVAL; 1269 goto put_sess; 1270 } 1271 1272 if (!first) 1273 return sess; 1274 1275 if (!path_cnt) { 1276 pr_err("Session %s not found, and path parameter not given", sessname); 1277 err = -ENXIO; 1278 goto put_sess; 1279 } 1280 1281 rtrs_ops = (struct rtrs_clt_ops) { 1282 .priv = sess, 1283 .link_ev = rnbd_clt_link_ev, 1284 }; 1285 /* 1286 * Nothing was found, establish rtrs connection and proceed further. 1287 */ 1288 sess->rtrs = rtrs_clt_open(&rtrs_ops, sessname, 1289 paths, path_cnt, port_nr, 1290 0, /* Do not use pdu of rtrs */ 1291 RECONNECT_DELAY, 1292 MAX_RECONNECTS, nr_poll_queues); 1293 if (IS_ERR(sess->rtrs)) { 1294 err = PTR_ERR(sess->rtrs); 1295 goto wake_up_and_put; 1296 } 1297 1298 err = rtrs_clt_query(sess->rtrs, &attrs); 1299 if (err) 1300 goto close_rtrs; 1301 1302 sess->max_io_size = attrs.max_io_size; 1303 sess->queue_depth = attrs.queue_depth; 1304 sess->nr_poll_queues = nr_poll_queues; 1305 sess->max_segments = attrs.max_segments; 1306 1307 err = setup_mq_tags(sess); 1308 if (err) 1309 goto close_rtrs; 1310 1311 err = send_msg_sess_info(sess, RTRS_PERMIT_WAIT); 1312 if (err) 1313 goto close_rtrs; 1314 1315 wake_up_rtrs_waiters(sess); 1316 1317 return sess; 1318 1319 close_rtrs: 1320 close_rtrs(sess); 1321 put_sess: 1322 rnbd_clt_put_sess(sess); 1323 1324 return ERR_PTR(err); 1325 1326 wake_up_and_put: 1327 wake_up_rtrs_waiters(sess); 1328 goto put_sess; 1329 } 1330 1331 static inline void rnbd_init_hw_queue(struct rnbd_clt_dev *dev, 1332 struct rnbd_queue *q, 1333 struct blk_mq_hw_ctx *hctx) 1334 { 1335 INIT_LIST_HEAD(&q->requeue_list); 1336 q->dev = dev; 1337 q->hctx = hctx; 1338 } 1339 1340 static void rnbd_init_mq_hw_queues(struct rnbd_clt_dev *dev) 1341 { 1342 unsigned long i; 1343 struct blk_mq_hw_ctx *hctx; 1344 struct rnbd_queue *q; 1345 1346 queue_for_each_hw_ctx(dev->queue, hctx, i) { 1347 q = &dev->hw_queues[i]; 1348 rnbd_init_hw_queue(dev, q, hctx); 1349 hctx->driver_data = q; 1350 } 1351 } 1352 1353 static void setup_request_queue(struct rnbd_clt_dev *dev) 1354 { 1355 blk_queue_logical_block_size(dev->queue, dev->logical_block_size); 1356 blk_queue_physical_block_size(dev->queue, dev->physical_block_size); 1357 blk_queue_max_hw_sectors(dev->queue, dev->max_hw_sectors); 1358 1359 /* 1360 * we don't support discards to "discontiguous" segments 1361 * in on request 1362 */ 1363 blk_queue_max_discard_segments(dev->queue, 1); 1364 1365 blk_queue_max_discard_sectors(dev->queue, dev->max_discard_sectors); 1366 dev->queue->limits.discard_granularity = dev->discard_granularity; 1367 dev->queue->limits.discard_alignment = dev->discard_alignment; 1368 if (dev->secure_discard) 1369 blk_queue_max_secure_erase_sectors(dev->queue, 1370 dev->max_discard_sectors); 1371 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, dev->queue); 1372 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, dev->queue); 1373 blk_queue_max_segments(dev->queue, dev->max_segments); 1374 blk_queue_io_opt(dev->queue, dev->sess->max_io_size); 1375 blk_queue_virt_boundary(dev->queue, SZ_4K - 1); 1376 blk_queue_write_cache(dev->queue, dev->wc, dev->fua); 1377 } 1378 1379 static int rnbd_clt_setup_gen_disk(struct rnbd_clt_dev *dev, int idx) 1380 { 1381 int err; 1382 1383 dev->gd->major = rnbd_client_major; 1384 dev->gd->first_minor = idx << RNBD_PART_BITS; 1385 dev->gd->minors = 1 << RNBD_PART_BITS; 1386 dev->gd->fops = &rnbd_client_ops; 1387 dev->gd->queue = dev->queue; 1388 dev->gd->private_data = dev; 1389 snprintf(dev->gd->disk_name, sizeof(dev->gd->disk_name), "rnbd%d", 1390 idx); 1391 pr_debug("disk_name=%s, capacity=%zu\n", 1392 dev->gd->disk_name, 1393 dev->nsectors * (dev->logical_block_size / SECTOR_SIZE) 1394 ); 1395 1396 set_capacity(dev->gd, dev->nsectors); 1397 1398 if (dev->access_mode == RNBD_ACCESS_RO) { 1399 dev->read_only = true; 1400 set_disk_ro(dev->gd, true); 1401 } else { 1402 dev->read_only = false; 1403 } 1404 1405 /* 1406 * Network device does not need rotational 1407 */ 1408 blk_queue_flag_set(QUEUE_FLAG_NONROT, dev->queue); 1409 err = add_disk(dev->gd); 1410 if (err) 1411 blk_cleanup_disk(dev->gd); 1412 1413 return err; 1414 } 1415 1416 static int rnbd_client_setup_device(struct rnbd_clt_dev *dev) 1417 { 1418 int idx = dev->clt_device_id; 1419 1420 dev->size = dev->nsectors * dev->logical_block_size; 1421 1422 dev->gd = blk_mq_alloc_disk(&dev->sess->tag_set, dev); 1423 if (IS_ERR(dev->gd)) 1424 return PTR_ERR(dev->gd); 1425 dev->queue = dev->gd->queue; 1426 rnbd_init_mq_hw_queues(dev); 1427 1428 setup_request_queue(dev); 1429 return rnbd_clt_setup_gen_disk(dev, idx); 1430 } 1431 1432 static struct rnbd_clt_dev *init_dev(struct rnbd_clt_session *sess, 1433 enum rnbd_access_mode access_mode, 1434 const char *pathname, 1435 u32 nr_poll_queues) 1436 { 1437 struct rnbd_clt_dev *dev; 1438 int ret; 1439 1440 dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, NUMA_NO_NODE); 1441 if (!dev) 1442 return ERR_PTR(-ENOMEM); 1443 1444 /* 1445 * nr_cpu_ids: the number of softirq queues 1446 * nr_poll_queues: the number of polling queues 1447 */ 1448 dev->hw_queues = kcalloc(nr_cpu_ids + nr_poll_queues, 1449 sizeof(*dev->hw_queues), 1450 GFP_KERNEL); 1451 if (!dev->hw_queues) { 1452 ret = -ENOMEM; 1453 goto out_alloc; 1454 } 1455 1456 ret = ida_alloc_max(&index_ida, 1 << (MINORBITS - RNBD_PART_BITS), 1457 GFP_KERNEL); 1458 if (ret < 0) { 1459 pr_err("Failed to initialize device '%s' from session %s, allocating idr failed, err: %d\n", 1460 pathname, sess->sessname, ret); 1461 goto out_queues; 1462 } 1463 1464 dev->pathname = kstrdup(pathname, GFP_KERNEL); 1465 if (!dev->pathname) { 1466 ret = -ENOMEM; 1467 goto out_queues; 1468 } 1469 1470 dev->clt_device_id = ret; 1471 dev->sess = sess; 1472 dev->access_mode = access_mode; 1473 dev->nr_poll_queues = nr_poll_queues; 1474 mutex_init(&dev->lock); 1475 refcount_set(&dev->refcount, 1); 1476 dev->dev_state = DEV_STATE_INIT; 1477 1478 /* 1479 * Here we called from sysfs entry, thus clt-sysfs is 1480 * responsible that session will not disappear. 1481 */ 1482 WARN_ON(!rnbd_clt_get_sess(sess)); 1483 1484 return dev; 1485 1486 out_queues: 1487 kfree(dev->hw_queues); 1488 out_alloc: 1489 kfree(dev); 1490 return ERR_PTR(ret); 1491 } 1492 1493 static bool __exists_dev(const char *pathname, const char *sessname) 1494 { 1495 struct rnbd_clt_session *sess; 1496 struct rnbd_clt_dev *dev; 1497 bool found = false; 1498 1499 list_for_each_entry(sess, &sess_list, list) { 1500 if (sessname && strncmp(sess->sessname, sessname, 1501 sizeof(sess->sessname))) 1502 continue; 1503 mutex_lock(&sess->lock); 1504 list_for_each_entry(dev, &sess->devs_list, list) { 1505 if (strlen(dev->pathname) == strlen(pathname) && 1506 !strcmp(dev->pathname, pathname)) { 1507 found = true; 1508 break; 1509 } 1510 } 1511 mutex_unlock(&sess->lock); 1512 if (found) 1513 break; 1514 } 1515 1516 return found; 1517 } 1518 1519 static bool exists_devpath(const char *pathname, const char *sessname) 1520 { 1521 bool found; 1522 1523 mutex_lock(&sess_lock); 1524 found = __exists_dev(pathname, sessname); 1525 mutex_unlock(&sess_lock); 1526 1527 return found; 1528 } 1529 1530 static bool insert_dev_if_not_exists_devpath(struct rnbd_clt_dev *dev) 1531 { 1532 bool found; 1533 struct rnbd_clt_session *sess = dev->sess; 1534 1535 mutex_lock(&sess_lock); 1536 found = __exists_dev(dev->pathname, sess->sessname); 1537 if (!found) { 1538 mutex_lock(&sess->lock); 1539 list_add_tail(&dev->list, &sess->devs_list); 1540 mutex_unlock(&sess->lock); 1541 } 1542 mutex_unlock(&sess_lock); 1543 1544 return found; 1545 } 1546 1547 static void delete_dev(struct rnbd_clt_dev *dev) 1548 { 1549 struct rnbd_clt_session *sess = dev->sess; 1550 1551 mutex_lock(&sess->lock); 1552 list_del(&dev->list); 1553 mutex_unlock(&sess->lock); 1554 } 1555 1556 struct rnbd_clt_dev *rnbd_clt_map_device(const char *sessname, 1557 struct rtrs_addr *paths, 1558 size_t path_cnt, u16 port_nr, 1559 const char *pathname, 1560 enum rnbd_access_mode access_mode, 1561 u32 nr_poll_queues) 1562 { 1563 struct rnbd_clt_session *sess; 1564 struct rnbd_clt_dev *dev; 1565 int ret; 1566 1567 if (exists_devpath(pathname, sessname)) 1568 return ERR_PTR(-EEXIST); 1569 1570 sess = find_and_get_or_create_sess(sessname, paths, path_cnt, port_nr, nr_poll_queues); 1571 if (IS_ERR(sess)) 1572 return ERR_CAST(sess); 1573 1574 dev = init_dev(sess, access_mode, pathname, nr_poll_queues); 1575 if (IS_ERR(dev)) { 1576 pr_err("map_device: failed to map device '%s' from session %s, can't initialize device, err: %ld\n", 1577 pathname, sess->sessname, PTR_ERR(dev)); 1578 ret = PTR_ERR(dev); 1579 goto put_sess; 1580 } 1581 if (insert_dev_if_not_exists_devpath(dev)) { 1582 ret = -EEXIST; 1583 goto put_dev; 1584 } 1585 ret = send_msg_open(dev, RTRS_PERMIT_WAIT); 1586 if (ret) { 1587 rnbd_clt_err(dev, 1588 "map_device: failed, can't open remote device, err: %d\n", 1589 ret); 1590 goto del_dev; 1591 } 1592 mutex_lock(&dev->lock); 1593 pr_debug("Opened remote device: session=%s, path='%s'\n", 1594 sess->sessname, pathname); 1595 ret = rnbd_client_setup_device(dev); 1596 if (ret) { 1597 rnbd_clt_err(dev, 1598 "map_device: Failed to configure device, err: %d\n", 1599 ret); 1600 mutex_unlock(&dev->lock); 1601 goto send_close; 1602 } 1603 1604 rnbd_clt_info(dev, 1605 "map_device: Device mapped as %s (nsectors: %zu, logical_block_size: %d, physical_block_size: %d, max_discard_sectors: %d, discard_granularity: %d, discard_alignment: %d, secure_discard: %d, max_segments: %d, max_hw_sectors: %d, wc: %d, fua: %d)\n", 1606 dev->gd->disk_name, dev->nsectors, 1607 dev->logical_block_size, dev->physical_block_size, 1608 dev->max_discard_sectors, 1609 dev->discard_granularity, dev->discard_alignment, 1610 dev->secure_discard, dev->max_segments, 1611 dev->max_hw_sectors, dev->wc, dev->fua); 1612 1613 mutex_unlock(&dev->lock); 1614 rnbd_clt_put_sess(sess); 1615 1616 return dev; 1617 1618 send_close: 1619 send_msg_close(dev, dev->device_id, RTRS_PERMIT_WAIT); 1620 del_dev: 1621 delete_dev(dev); 1622 put_dev: 1623 rnbd_clt_put_dev(dev); 1624 put_sess: 1625 rnbd_clt_put_sess(sess); 1626 1627 return ERR_PTR(ret); 1628 } 1629 1630 static void destroy_gen_disk(struct rnbd_clt_dev *dev) 1631 { 1632 del_gendisk(dev->gd); 1633 blk_cleanup_disk(dev->gd); 1634 } 1635 1636 static void destroy_sysfs(struct rnbd_clt_dev *dev, 1637 const struct attribute *sysfs_self) 1638 { 1639 rnbd_clt_remove_dev_symlink(dev); 1640 if (dev->kobj.state_initialized) { 1641 if (sysfs_self) 1642 /* To avoid deadlock firstly remove itself */ 1643 sysfs_remove_file_self(&dev->kobj, sysfs_self); 1644 kobject_del(&dev->kobj); 1645 kobject_put(&dev->kobj); 1646 } 1647 } 1648 1649 int rnbd_clt_unmap_device(struct rnbd_clt_dev *dev, bool force, 1650 const struct attribute *sysfs_self) 1651 { 1652 struct rnbd_clt_session *sess = dev->sess; 1653 int refcount, ret = 0; 1654 bool was_mapped; 1655 1656 mutex_lock(&dev->lock); 1657 if (dev->dev_state == DEV_STATE_UNMAPPED) { 1658 rnbd_clt_info(dev, "Device is already being unmapped\n"); 1659 ret = -EALREADY; 1660 goto err; 1661 } 1662 refcount = refcount_read(&dev->refcount); 1663 if (!force && refcount > 1) { 1664 rnbd_clt_err(dev, 1665 "Closing device failed, device is in use, (%d device users)\n", 1666 refcount - 1); 1667 ret = -EBUSY; 1668 goto err; 1669 } 1670 was_mapped = (dev->dev_state == DEV_STATE_MAPPED); 1671 dev->dev_state = DEV_STATE_UNMAPPED; 1672 mutex_unlock(&dev->lock); 1673 1674 delete_dev(dev); 1675 destroy_sysfs(dev, sysfs_self); 1676 destroy_gen_disk(dev); 1677 if (was_mapped && sess->rtrs) 1678 send_msg_close(dev, dev->device_id, RTRS_PERMIT_WAIT); 1679 1680 rnbd_clt_info(dev, "Device is unmapped\n"); 1681 1682 /* Likely last reference put */ 1683 rnbd_clt_put_dev(dev); 1684 1685 /* 1686 * Here device and session can be vanished! 1687 */ 1688 1689 return 0; 1690 err: 1691 mutex_unlock(&dev->lock); 1692 1693 return ret; 1694 } 1695 1696 int rnbd_clt_remap_device(struct rnbd_clt_dev *dev) 1697 { 1698 int err; 1699 1700 mutex_lock(&dev->lock); 1701 if (dev->dev_state == DEV_STATE_MAPPED_DISCONNECTED) 1702 err = 0; 1703 else if (dev->dev_state == DEV_STATE_UNMAPPED) 1704 err = -ENODEV; 1705 else if (dev->dev_state == DEV_STATE_MAPPED) 1706 err = -EALREADY; 1707 else 1708 err = -EBUSY; 1709 mutex_unlock(&dev->lock); 1710 if (!err) { 1711 rnbd_clt_info(dev, "Remapping device.\n"); 1712 err = send_msg_open(dev, RTRS_PERMIT_WAIT); 1713 if (err) 1714 rnbd_clt_err(dev, "remap_device: %d\n", err); 1715 } 1716 1717 return err; 1718 } 1719 1720 static void unmap_device_work(struct work_struct *work) 1721 { 1722 struct rnbd_clt_dev *dev; 1723 1724 dev = container_of(work, typeof(*dev), unmap_on_rmmod_work); 1725 rnbd_clt_unmap_device(dev, true, NULL); 1726 } 1727 1728 static void rnbd_destroy_sessions(void) 1729 { 1730 struct rnbd_clt_session *sess, *sn; 1731 struct rnbd_clt_dev *dev, *tn; 1732 1733 /* Firstly forbid access through sysfs interface */ 1734 rnbd_clt_destroy_sysfs_files(); 1735 1736 /* 1737 * Here at this point there is no any concurrent access to sessions 1738 * list and devices list: 1739 * 1. New session or device can't be created - session sysfs files 1740 * are removed. 1741 * 2. Device or session can't be removed - module reference is taken 1742 * into account in unmap device sysfs callback. 1743 * 3. No IO requests inflight - each file open of block_dev increases 1744 * module reference in get_disk(). 1745 * 1746 * But still there can be user requests inflights, which are sent by 1747 * asynchronous send_msg_*() functions, thus before unmapping devices 1748 * RTRS session must be explicitly closed. 1749 */ 1750 1751 list_for_each_entry_safe(sess, sn, &sess_list, list) { 1752 if (!rnbd_clt_get_sess(sess)) 1753 continue; 1754 close_rtrs(sess); 1755 list_for_each_entry_safe(dev, tn, &sess->devs_list, list) { 1756 /* 1757 * Here unmap happens in parallel for only one reason: 1758 * blk_cleanup_queue() takes around half a second, so 1759 * on huge amount of devices the whole module unload 1760 * procedure takes minutes. 1761 */ 1762 INIT_WORK(&dev->unmap_on_rmmod_work, unmap_device_work); 1763 queue_work(rnbd_clt_wq, &dev->unmap_on_rmmod_work); 1764 } 1765 rnbd_clt_put_sess(sess); 1766 } 1767 /* Wait for all scheduled unmap works */ 1768 flush_workqueue(rnbd_clt_wq); 1769 WARN_ON(!list_empty(&sess_list)); 1770 } 1771 1772 static int __init rnbd_client_init(void) 1773 { 1774 int err = 0; 1775 1776 BUILD_BUG_ON(sizeof(struct rnbd_msg_hdr) != 4); 1777 BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info) != 36); 1778 BUILD_BUG_ON(sizeof(struct rnbd_msg_sess_info_rsp) != 36); 1779 BUILD_BUG_ON(sizeof(struct rnbd_msg_open) != 264); 1780 BUILD_BUG_ON(sizeof(struct rnbd_msg_close) != 8); 1781 BUILD_BUG_ON(sizeof(struct rnbd_msg_open_rsp) != 56); 1782 rnbd_client_major = register_blkdev(rnbd_client_major, "rnbd"); 1783 if (rnbd_client_major <= 0) { 1784 pr_err("Failed to load module, block device registration failed\n"); 1785 return -EBUSY; 1786 } 1787 1788 err = rnbd_clt_create_sysfs_files(); 1789 if (err) { 1790 pr_err("Failed to load module, creating sysfs device files failed, err: %d\n", 1791 err); 1792 unregister_blkdev(rnbd_client_major, "rnbd"); 1793 return err; 1794 } 1795 rnbd_clt_wq = alloc_workqueue("rnbd_clt_wq", 0, 0); 1796 if (!rnbd_clt_wq) { 1797 pr_err("Failed to load module, alloc_workqueue failed.\n"); 1798 rnbd_clt_destroy_sysfs_files(); 1799 unregister_blkdev(rnbd_client_major, "rnbd"); 1800 err = -ENOMEM; 1801 } 1802 1803 return err; 1804 } 1805 1806 static void __exit rnbd_client_exit(void) 1807 { 1808 rnbd_destroy_sessions(); 1809 unregister_blkdev(rnbd_client_major, "rnbd"); 1810 ida_destroy(&index_ida); 1811 destroy_workqueue(rnbd_clt_wq); 1812 } 1813 1814 module_init(rnbd_client_init); 1815 module_exit(rnbd_client_exit); 1816