1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 1991, 1992 Linus Torvalds 4 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics 5 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE 6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> 7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> 8 * - July2000 9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 10 */ 11 12 /* 13 * This handles all read/write requests to block devices 14 */ 15 #include <linux/kernel.h> 16 #include <linux/module.h> 17 #include <linux/bio.h> 18 #include <linux/blkdev.h> 19 #include <linux/blk-pm.h> 20 #include <linux/blk-integrity.h> 21 #include <linux/highmem.h> 22 #include <linux/mm.h> 23 #include <linux/pagemap.h> 24 #include <linux/kernel_stat.h> 25 #include <linux/string.h> 26 #include <linux/init.h> 27 #include <linux/completion.h> 28 #include <linux/slab.h> 29 #include <linux/swap.h> 30 #include <linux/writeback.h> 31 #include <linux/task_io_accounting_ops.h> 32 #include <linux/fault-inject.h> 33 #include <linux/list_sort.h> 34 #include <linux/delay.h> 35 #include <linux/ratelimit.h> 36 #include <linux/pm_runtime.h> 37 #include <linux/t10-pi.h> 38 #include <linux/debugfs.h> 39 #include <linux/bpf.h> 40 #include <linux/part_stat.h> 41 #include <linux/sched/sysctl.h> 42 #include <linux/blk-crypto.h> 43 #include <linux/error-injection.h> 44 45 #define CREATE_TRACE_POINTS 46 #include <trace/events/block.h> 47 48 #include "blk.h" 49 #include "blk-mq-sched.h" 50 #include "blk-pm.h" 51 #include "blk-cgroup.h" 52 #include "blk-throttle.h" 53 #include "blk-ioprio.h" 54 #include "error-injection.h" 55 56 struct dentry *blk_debugfs_root; 57 58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); 59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); 60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); 61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); 62 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); 63 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert); 64 65 static DEFINE_IDA(blk_queue_ida); 66 67 /* 68 * For queue allocation 69 */ 70 static struct kmem_cache *blk_requestq_cachep; 71 72 /* 73 * Controlling structure to kblockd 74 */ 75 static struct workqueue_struct *kblockd_workqueue; 76 77 /** 78 * blk_queue_flag_set - atomically set a queue flag 79 * @flag: flag to be set 80 * @q: request queue 81 */ 82 void blk_queue_flag_set(unsigned int flag, struct request_queue *q) 83 { 84 set_bit(flag, &q->queue_flags); 85 } 86 EXPORT_SYMBOL(blk_queue_flag_set); 87 88 /** 89 * blk_queue_flag_clear - atomically clear a queue flag 90 * @flag: flag to be cleared 91 * @q: request queue 92 */ 93 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) 94 { 95 clear_bit(flag, &q->queue_flags); 96 } 97 EXPORT_SYMBOL(blk_queue_flag_clear); 98 99 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name 100 static const char *const blk_op_name[] = { 101 REQ_OP_NAME(READ), 102 REQ_OP_NAME(WRITE), 103 REQ_OP_NAME(FLUSH), 104 REQ_OP_NAME(DISCARD), 105 REQ_OP_NAME(SECURE_ERASE), 106 REQ_OP_NAME(ZONE_RESET), 107 REQ_OP_NAME(ZONE_RESET_ALL), 108 REQ_OP_NAME(ZONE_OPEN), 109 REQ_OP_NAME(ZONE_CLOSE), 110 REQ_OP_NAME(ZONE_FINISH), 111 REQ_OP_NAME(ZONE_APPEND), 112 REQ_OP_NAME(WRITE_ZEROES), 113 REQ_OP_NAME(DRV_IN), 114 REQ_OP_NAME(DRV_OUT), 115 }; 116 #undef REQ_OP_NAME 117 118 /** 119 * blk_op_str - Return the string "name" for an operation REQ_OP_name. 120 * @op: a request operation. 121 * 122 * Convert a request operation REQ_OP_name into the string "name". Useful for 123 * debugging and tracing BIOs and requests. For an invalid request operation 124 * code, the string "UNKNOWN" is returned. 125 */ 126 inline const char *blk_op_str(enum req_op op) 127 { 128 const char *op_str = "UNKNOWN"; 129 130 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) 131 op_str = blk_op_name[op]; 132 133 return op_str; 134 } 135 EXPORT_SYMBOL_GPL(blk_op_str); 136 137 enum req_op str_to_blk_op(const char *op) 138 { 139 int i; 140 141 for (i = 0; i < ARRAY_SIZE(blk_op_name); i++) 142 if (blk_op_name[i] && !strcmp(blk_op_name[i], op)) 143 return (enum req_op)i; 144 return REQ_OP_LAST; 145 } 146 147 #define ENT(_tag, _errno, _desc) \ 148 [BLK_STS_##_tag] = { \ 149 .errno = _errno, \ 150 .tag = __stringify(_tag), \ 151 .name = _desc, \ 152 } 153 static const struct { 154 int errno; 155 const char *tag; 156 const char *name; 157 } blk_errors[] = { 158 ENT(OK, 0, ""), 159 ENT(NOTSUPP, -EOPNOTSUPP, "operation not supported"), 160 ENT(TIMEOUT, -ETIMEDOUT, "timeout"), 161 ENT(NOSPC, -ENOSPC, "critical space allocation"), 162 ENT(TRANSPORT, -ENOLINK, "recoverable transport"), 163 ENT(TARGET, -EREMOTEIO, "critical target"), 164 ENT(RESV_CONFLICT, -EBADE, "reservation conflict"), 165 ENT(MEDIUM, -ENODATA, "critical medium"), 166 ENT(PROTECTION, -EILSEQ, "protection"), 167 ENT(RESOURCE, -ENOMEM, "kernel resource"), 168 ENT(DEV_RESOURCE, -EBUSY, "device resource"), 169 ENT(AGAIN, -EAGAIN, "nonblocking retry"), 170 ENT(OFFLINE, -ENODEV, "device offline"), 171 172 /* device mapper special case, should not leak out: */ 173 ENT(DM_REQUEUE, -EREMCHG, "dm internal retry"), 174 175 /* zone device specific errors */ 176 ENT(ZONE_OPEN_RESOURCE, -ETOOMANYREFS, "open zones exceeded"), 177 ENT(ZONE_ACTIVE_RESOURCE, -EOVERFLOW, "active zones exceeded"), 178 179 /* Command duration limit device-side timeout */ 180 ENT(DURATION_LIMIT, -ETIME, "duration limit exceeded"), 181 ENT(INVAL, -EINVAL, "invalid"), 182 183 /* everything else not covered above: */ 184 ENT(IOERR, -EIO, "I/O"), 185 }; 186 #undef ENT 187 188 blk_status_t errno_to_blk_status(int errno) 189 { 190 int i; 191 192 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { 193 if (blk_errors[i].errno == errno) 194 return (__force blk_status_t)i; 195 } 196 197 return BLK_STS_IOERR; 198 } 199 EXPORT_SYMBOL_GPL(errno_to_blk_status); 200 201 int blk_status_to_errno(blk_status_t status) 202 { 203 int idx = (__force int)status; 204 205 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 206 return -EIO; 207 return blk_errors[idx].errno; 208 } 209 EXPORT_SYMBOL_GPL(blk_status_to_errno); 210 211 const char *blk_status_to_str(blk_status_t status) 212 { 213 int idx = (__force int)status; 214 215 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 216 return "<null>"; 217 return blk_errors[idx].name; 218 } 219 220 const char *blk_status_to_tag(blk_status_t status) 221 { 222 int idx = (__force int)status; 223 224 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors) || !blk_errors[idx].tag)) 225 return "<null>"; 226 return blk_errors[idx].tag; 227 } 228 229 blk_status_t tag_to_blk_status(const char *tag) 230 { 231 int i; 232 233 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { 234 if (blk_errors[i].tag && 235 !strcmp(blk_errors[i].tag, tag)) 236 return (__force blk_status_t)i; 237 } 238 239 /* 240 * Return BLK_STS_OK for mismatches as this function is intended to 241 * parse error status values. 242 */ 243 return BLK_STS_OK; 244 } 245 246 /** 247 * blk_sync_queue - cancel any pending callbacks on a queue 248 * @q: the queue 249 * 250 * Description: 251 * The block layer may perform asynchronous callback activity 252 * on a queue, such as calling the unplug function after a timeout. 253 * A block device may call blk_sync_queue to ensure that any 254 * such activity is cancelled, thus allowing it to release resources 255 * that the callbacks might use. The caller must already have made sure 256 * that its ->submit_bio will not re-add plugging prior to calling 257 * this function. 258 * 259 * This function does not cancel any asynchronous activity arising 260 * out of elevator or throttling code. That would require elevator_exit() 261 * and blkcg_exit_queue() to be called with queue lock initialized. 262 * 263 */ 264 void blk_sync_queue(struct request_queue *q) 265 { 266 timer_delete_sync(&q->timeout); 267 cancel_work_sync(&q->timeout_work); 268 } 269 EXPORT_SYMBOL(blk_sync_queue); 270 271 /** 272 * blk_set_pm_only - increment pm_only counter 273 * @q: request queue pointer 274 */ 275 void blk_set_pm_only(struct request_queue *q) 276 { 277 atomic_inc(&q->pm_only); 278 } 279 EXPORT_SYMBOL_GPL(blk_set_pm_only); 280 281 void blk_clear_pm_only(struct request_queue *q) 282 { 283 int pm_only; 284 285 pm_only = atomic_dec_return(&q->pm_only); 286 WARN_ON_ONCE(pm_only < 0); 287 if (pm_only == 0) 288 wake_up_all(&q->mq_freeze_wq); 289 } 290 EXPORT_SYMBOL_GPL(blk_clear_pm_only); 291 292 static void blk_free_queue_rcu(struct rcu_head *rcu_head) 293 { 294 struct request_queue *q = container_of(rcu_head, 295 struct request_queue, rcu_head); 296 297 percpu_ref_exit(&q->q_usage_counter); 298 kmem_cache_free(blk_requestq_cachep, q); 299 } 300 301 static void blk_free_queue(struct request_queue *q) 302 { 303 blk_free_queue_stats(q->stats); 304 if (queue_is_mq(q)) 305 blk_mq_release(q); 306 307 ida_free(&blk_queue_ida, q->id); 308 lockdep_unregister_key(&q->io_lock_cls_key); 309 lockdep_unregister_key(&q->q_lock_cls_key); 310 call_rcu(&q->rcu_head, blk_free_queue_rcu); 311 } 312 313 /** 314 * blk_put_queue - decrement the request_queue refcount 315 * @q: the request_queue structure to decrement the refcount for 316 * 317 * Decrements the refcount of the request_queue and free it when the refcount 318 * reaches 0. 319 */ 320 void blk_put_queue(struct request_queue *q) 321 { 322 if (refcount_dec_and_test(&q->refs)) 323 blk_free_queue(q); 324 } 325 EXPORT_SYMBOL(blk_put_queue); 326 327 bool blk_queue_start_drain(struct request_queue *q) 328 { 329 /* 330 * When queue DYING flag is set, we need to block new req 331 * entering queue, so we call blk_freeze_queue_start() to 332 * prevent I/O from crossing blk_queue_enter(). 333 */ 334 bool freeze = __blk_freeze_queue_start(q, current); 335 if (queue_is_mq(q)) 336 blk_mq_wake_waiters(q); 337 /* Make blk_queue_enter() reexamine the DYING flag. */ 338 wake_up_all(&q->mq_freeze_wq); 339 340 return freeze; 341 } 342 343 /** 344 * blk_queue_enter() - try to increase q->q_usage_counter 345 * @q: request queue pointer 346 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM 347 */ 348 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) 349 { 350 const bool pm = flags & BLK_MQ_REQ_PM; 351 352 while (!blk_try_enter_queue(q, pm)) { 353 if (flags & BLK_MQ_REQ_NOWAIT) 354 return -EAGAIN; 355 356 /* 357 * read pair of barrier in blk_freeze_queue_start(), we need to 358 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 359 * reading .mq_freeze_depth or queue dying flag, otherwise the 360 * following wait may never return if the two reads are 361 * reordered. 362 */ 363 smp_rmb(); 364 wait_event(q->mq_freeze_wq, 365 (!q->mq_freeze_depth && 366 blk_pm_resume_queue(pm, q)) || 367 blk_queue_dying(q)); 368 if (blk_queue_dying(q)) 369 return -ENODEV; 370 } 371 372 rwsem_acquire_read(&q->q_lockdep_map, 0, 0, _RET_IP_); 373 rwsem_release(&q->q_lockdep_map, _RET_IP_); 374 return 0; 375 } 376 377 int __bio_queue_enter(struct request_queue *q, struct bio *bio) 378 { 379 while (!blk_try_enter_queue(q, false)) { 380 struct gendisk *disk = bio->bi_bdev->bd_disk; 381 382 if (bio->bi_opf & REQ_NOWAIT) { 383 if (test_bit(GD_DEAD, &disk->state)) 384 goto dead; 385 bio_wouldblock_error(bio); 386 return -EAGAIN; 387 } 388 389 /* 390 * read pair of barrier in blk_freeze_queue_start(), we need to 391 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 392 * reading .mq_freeze_depth or queue dying flag, otherwise the 393 * following wait may never return if the two reads are 394 * reordered. 395 */ 396 smp_rmb(); 397 wait_event(q->mq_freeze_wq, 398 (!q->mq_freeze_depth && 399 blk_pm_resume_queue(false, q)) || 400 test_bit(GD_DEAD, &disk->state)); 401 if (test_bit(GD_DEAD, &disk->state)) 402 goto dead; 403 } 404 405 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_); 406 rwsem_release(&q->io_lockdep_map, _RET_IP_); 407 return 0; 408 dead: 409 bio_io_error(bio); 410 return -ENODEV; 411 } 412 413 void blk_queue_exit(struct request_queue *q) 414 { 415 percpu_ref_put(&q->q_usage_counter); 416 } 417 418 static void blk_queue_usage_counter_release(struct percpu_ref *ref) 419 { 420 struct request_queue *q = 421 container_of(ref, struct request_queue, q_usage_counter); 422 423 wake_up_all(&q->mq_freeze_wq); 424 } 425 426 static void blk_rq_timed_out_timer(struct timer_list *t) 427 { 428 struct request_queue *q = timer_container_of(q, t, timeout); 429 430 kblockd_schedule_work(&q->timeout_work); 431 } 432 433 static void blk_timeout_work(struct work_struct *work) 434 { 435 } 436 437 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id) 438 { 439 struct request_queue *q; 440 int error; 441 442 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO, 443 node_id); 444 if (!q) 445 return ERR_PTR(-ENOMEM); 446 447 q->last_merge = NULL; 448 449 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL); 450 if (q->id < 0) { 451 error = q->id; 452 goto fail_q; 453 } 454 455 q->stats = blk_alloc_queue_stats(); 456 if (!q->stats) { 457 error = -ENOMEM; 458 goto fail_id; 459 } 460 461 error = blk_set_default_limits(lim); 462 if (error) 463 goto fail_stats; 464 q->limits = *lim; 465 466 q->node = node_id; 467 468 atomic_set(&q->nr_active_requests_shared_tags, 0); 469 470 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); 471 INIT_WORK(&q->timeout_work, blk_timeout_work); 472 INIT_LIST_HEAD(&q->icq_list); 473 474 refcount_set(&q->refs, 1); 475 mutex_init(&q->debugfs_mutex); 476 mutex_init(&q->elevator_lock); 477 mutex_init(&q->sysfs_lock); 478 mutex_init(&q->limits_lock); 479 mutex_init(&q->rq_qos_mutex); 480 spin_lock_init(&q->queue_lock); 481 482 init_waitqueue_head(&q->mq_freeze_wq); 483 mutex_init(&q->mq_freeze_lock); 484 485 blkg_init_queue(q); 486 487 /* 488 * Init percpu_ref in atomic mode so that it's faster to shutdown. 489 * See blk_register_queue() for details. 490 */ 491 error = percpu_ref_init(&q->q_usage_counter, 492 blk_queue_usage_counter_release, 493 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL); 494 if (error) 495 goto fail_stats; 496 lockdep_register_key(&q->io_lock_cls_key); 497 lockdep_register_key(&q->q_lock_cls_key); 498 lockdep_init_map(&q->io_lockdep_map, "&q->q_usage_counter(io)", 499 &q->io_lock_cls_key, 0); 500 lockdep_init_map(&q->q_lockdep_map, "&q->q_usage_counter(queue)", 501 &q->q_lock_cls_key, 0); 502 503 /* Teach lockdep about lock ordering (reclaim WRT queue freeze lock). */ 504 fs_reclaim_acquire(GFP_KERNEL); 505 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_); 506 rwsem_release(&q->io_lockdep_map, _RET_IP_); 507 fs_reclaim_release(GFP_KERNEL); 508 509 q->nr_requests = BLKDEV_DEFAULT_RQ; 510 q->async_depth = BLKDEV_DEFAULT_RQ; 511 512 return q; 513 514 fail_stats: 515 blk_free_queue_stats(q->stats); 516 fail_id: 517 ida_free(&blk_queue_ida, q->id); 518 fail_q: 519 kmem_cache_free(blk_requestq_cachep, q); 520 return ERR_PTR(error); 521 } 522 523 /** 524 * blk_get_queue - increment the request_queue refcount 525 * @q: the request_queue structure to increment the refcount for 526 * 527 * Increment the refcount of the request_queue kobject. 528 * 529 * Context: Any context. 530 */ 531 bool blk_get_queue(struct request_queue *q) 532 { 533 if (unlikely(blk_queue_dying(q))) 534 return false; 535 refcount_inc(&q->refs); 536 return true; 537 } 538 EXPORT_SYMBOL(blk_get_queue); 539 540 #ifdef CONFIG_FAIL_MAKE_REQUEST 541 542 static DECLARE_FAULT_ATTR(fail_make_request); 543 544 static int __init setup_fail_make_request(char *str) 545 { 546 return setup_fault_attr(&fail_make_request, str); 547 } 548 __setup("fail_make_request=", setup_fail_make_request); 549 550 bool should_fail_request(struct block_device *part, unsigned int bytes) 551 { 552 return bdev_test_flag(part, BD_MAKE_IT_FAIL) && 553 should_fail(&fail_make_request, bytes); 554 } 555 556 static int __init fail_make_request_debugfs(void) 557 { 558 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 559 NULL, &fail_make_request); 560 561 return PTR_ERR_OR_ZERO(dir); 562 } 563 564 late_initcall(fail_make_request_debugfs); 565 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 566 567 static inline void bio_check_ro(struct bio *bio) 568 { 569 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { 570 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) 571 return; 572 573 if (bdev_test_flag(bio->bi_bdev, BD_RO_WARNED)) 574 return; 575 576 bdev_set_flag(bio->bi_bdev, BD_RO_WARNED); 577 578 /* 579 * Use ioctl to set underlying disk of raid/dm to read-only 580 * will trigger this. 581 */ 582 pr_warn("Trying to write to read-only block-device %pg\n", 583 bio->bi_bdev); 584 } 585 } 586 587 int should_fail_bio(struct bio *bio) 588 { 589 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) 590 return -EIO; 591 return 0; 592 } 593 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); 594 595 /* 596 * Check whether this bio extends beyond the end of the device or partition. 597 * This may well happen - the kernel calls bread() without checking the size of 598 * the device, e.g., when mounting a file system. 599 */ 600 static inline int bio_check_eod(struct bio *bio) 601 { 602 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 603 unsigned int nr_sectors = bio_sectors(bio); 604 605 if (nr_sectors && 606 (nr_sectors > maxsector || 607 bio->bi_iter.bi_sector > maxsector - nr_sectors)) { 608 if (!maxsector) 609 return -EIO; 610 pr_info_ratelimited("%s: attempt to access beyond end of device\n" 611 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", 612 current->comm, bio->bi_bdev, bio->bi_opf, 613 bio->bi_iter.bi_sector, nr_sectors, maxsector); 614 return -EIO; 615 } 616 return 0; 617 } 618 619 /* 620 * Remap block n of partition p to block n+start(p) of the disk. 621 */ 622 static int blk_partition_remap(struct bio *bio) 623 { 624 struct block_device *p = bio->bi_bdev; 625 626 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 627 return -EIO; 628 if (bio_sectors(bio)) { 629 bio->bi_iter.bi_sector += p->bd_start_sect; 630 trace_block_bio_remap(bio, p->bd_dev, 631 bio->bi_iter.bi_sector - 632 p->bd_start_sect); 633 } 634 bio_set_flag(bio, BIO_REMAPPED); 635 return 0; 636 } 637 638 /* 639 * Check write append to a zoned block device. 640 */ 641 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 642 struct bio *bio) 643 { 644 int nr_sectors = bio_sectors(bio); 645 646 /* Only applicable to zoned block devices */ 647 if (!bdev_is_zoned(bio->bi_bdev)) 648 return BLK_STS_NOTSUPP; 649 650 /* The bio sector must point to the start of a sequential zone */ 651 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector)) 652 return BLK_STS_IOERR; 653 654 /* 655 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 656 * split and could result in non-contiguous sectors being written in 657 * different zones. 658 */ 659 if (nr_sectors > q->limits.chunk_sectors) 660 return BLK_STS_IOERR; 661 662 /* Make sure the BIO is small enough and will not get split */ 663 if (nr_sectors > q->limits.max_zone_append_sectors) 664 return BLK_STS_IOERR; 665 666 bio->bi_opf |= REQ_NOMERGE; 667 668 return BLK_STS_OK; 669 } 670 671 static void __submit_bio(struct bio *bio) 672 { 673 if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) { 674 blk_mq_submit_bio(bio); 675 } else if (likely(bio_queue_enter(bio) == 0)) { 676 struct gendisk *disk = bio->bi_bdev->bd_disk; 677 678 if ((bio->bi_opf & REQ_POLLED) && 679 !(disk->queue->limits.features & BLK_FEAT_POLL)) 680 bio_endio_status(bio, BLK_STS_NOTSUPP); 681 else 682 disk->fops->submit_bio(bio); 683 blk_queue_exit(disk->queue); 684 } 685 } 686 687 /* 688 * The loop in this function may be a bit non-obvious, and so deserves some 689 * explanation: 690 * 691 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 692 * that), so we have a list with a single bio. 693 * - We pretend that we have just taken it off a longer list, so we assign 694 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 695 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 696 * bios through a recursive call to submit_bio_noacct. If it did, we find a 697 * non-NULL value in bio_list and re-enter the loop from the top. 698 * - In this case we really did just take the bio off the top of the list (no 699 * pretending) and so remove it from bio_list, and call into ->submit_bio() 700 * again. 701 * 702 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 703 * bio_list_on_stack[1] contains bios that were submitted before the current 704 * ->submit_bio(), but that haven't been processed yet. 705 */ 706 static void __submit_bio_noacct(struct bio *bio) 707 { 708 struct bio_list bio_list_on_stack[2]; 709 710 BUG_ON(bio->bi_next); 711 712 bio_list_init(&bio_list_on_stack[0]); 713 current->bio_list = bio_list_on_stack; 714 715 do { 716 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 717 struct bio_list lower, same; 718 719 /* 720 * Create a fresh bio_list for all subordinate requests. 721 */ 722 bio_list_on_stack[1] = bio_list_on_stack[0]; 723 bio_list_init(&bio_list_on_stack[0]); 724 725 __submit_bio(bio); 726 727 /* 728 * Sort new bios into those for a lower level and those for the 729 * same level. 730 */ 731 bio_list_init(&lower); 732 bio_list_init(&same); 733 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 734 if (q == bdev_get_queue(bio->bi_bdev)) 735 bio_list_add(&same, bio); 736 else 737 bio_list_add(&lower, bio); 738 739 /* 740 * Now assemble so we handle the lowest level first. 741 */ 742 bio_list_merge(&bio_list_on_stack[0], &lower); 743 bio_list_merge(&bio_list_on_stack[0], &same); 744 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 745 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 746 747 current->bio_list = NULL; 748 } 749 750 static void __submit_bio_noacct_mq(struct bio *bio) 751 { 752 struct bio_list bio_list[2] = { }; 753 754 current->bio_list = bio_list; 755 756 do { 757 __submit_bio(bio); 758 } while ((bio = bio_list_pop(&bio_list[0]))); 759 760 current->bio_list = NULL; 761 } 762 763 void submit_bio_noacct_nocheck(struct bio *bio, bool split) 764 { 765 if (unlikely(blk_error_inject(bio))) 766 return; 767 768 blk_cgroup_bio_start(bio); 769 770 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 771 trace_block_bio_queue(bio); 772 /* 773 * Now that enqueuing has been traced, we need to trace 774 * completion as well. 775 */ 776 bio_set_flag(bio, BIO_TRACE_COMPLETION); 777 } 778 779 /* 780 * We only want one ->submit_bio to be active at a time, else stack 781 * usage with stacked devices could be a problem. Use current->bio_list 782 * to collect a list of requests submitted by a ->submit_bio method 783 * while it is active, and then process them after it returned. 784 */ 785 if (current->bio_list) { 786 if (split) 787 bio_list_add_head(¤t->bio_list[0], bio); 788 else 789 bio_list_add(¤t->bio_list[0], bio); 790 } else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) { 791 __submit_bio_noacct_mq(bio); 792 } else { 793 __submit_bio_noacct(bio); 794 } 795 } 796 797 static blk_status_t blk_validate_atomic_write_op_size(struct request_queue *q, 798 struct bio *bio) 799 { 800 if (bio->bi_iter.bi_size > queue_atomic_write_unit_max_bytes(q)) 801 return BLK_STS_INVAL; 802 803 if (bio->bi_iter.bi_size % queue_atomic_write_unit_min_bytes(q)) 804 return BLK_STS_INVAL; 805 806 return BLK_STS_OK; 807 } 808 809 /** 810 * submit_bio_noacct - re-submit a bio to the block device layer for I/O 811 * @bio: The bio describing the location in memory and on the device. 812 * 813 * This is a version of submit_bio() that shall only be used for I/O that is 814 * resubmitted to lower level drivers by stacking block drivers. All file 815 * systems and other upper level users of the block layer should use 816 * submit_bio() instead. 817 */ 818 void submit_bio_noacct(struct bio *bio) 819 { 820 struct block_device *bdev = bio->bi_bdev; 821 struct request_queue *q = bdev_get_queue(bdev); 822 blk_status_t status = BLK_STS_IOERR; 823 824 might_sleep(); 825 826 /* 827 * For a REQ_NOWAIT based request, return -EOPNOTSUPP 828 * if queue does not support NOWAIT. 829 */ 830 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev)) 831 goto not_supported; 832 833 if (bio_has_crypt_ctx(bio)) { 834 if (WARN_ON_ONCE(!bio_has_data(bio))) 835 goto end_io; 836 if (!blk_crypto_supported(bio)) 837 goto not_supported; 838 } 839 840 if (should_fail_bio(bio)) 841 goto end_io; 842 bio_check_ro(bio); 843 if (!bio_flagged(bio, BIO_REMAPPED)) { 844 if (unlikely(bio_check_eod(bio))) 845 goto end_io; 846 if (bdev_is_partition(bdev) && 847 unlikely(blk_partition_remap(bio))) 848 goto end_io; 849 } 850 851 /* 852 * Filter flush bio's early so that bio based drivers without flush 853 * support don't have to worry about them. 854 */ 855 if (op_is_flush(bio->bi_opf)) { 856 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE && 857 bio_op(bio) != REQ_OP_ZONE_APPEND)) 858 goto end_io; 859 if (!bdev_write_cache(bdev)) { 860 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 861 if (!bio_sectors(bio)) { 862 status = BLK_STS_OK; 863 goto end_io; 864 } 865 } 866 } 867 868 switch (bio_op(bio)) { 869 case REQ_OP_READ: 870 break; 871 case REQ_OP_WRITE: 872 if (bio->bi_opf & REQ_ATOMIC) { 873 status = blk_validate_atomic_write_op_size(q, bio); 874 if (status != BLK_STS_OK) 875 goto end_io; 876 } 877 break; 878 case REQ_OP_FLUSH: 879 /* 880 * REQ_OP_FLUSH can't be submitted through bios, it is only 881 * synthetized in struct request by the flush state machine. 882 */ 883 goto not_supported; 884 case REQ_OP_DISCARD: 885 if (!bdev_max_discard_sectors(bdev)) 886 goto not_supported; 887 break; 888 case REQ_OP_SECURE_ERASE: 889 if (!bdev_max_secure_erase_sectors(bdev)) 890 goto not_supported; 891 break; 892 case REQ_OP_ZONE_APPEND: 893 status = blk_check_zone_append(q, bio); 894 if (status != BLK_STS_OK) 895 goto end_io; 896 break; 897 case REQ_OP_WRITE_ZEROES: 898 if (!q->limits.max_write_zeroes_sectors) 899 goto not_supported; 900 break; 901 case REQ_OP_ZONE_OPEN: 902 case REQ_OP_ZONE_CLOSE: 903 case REQ_OP_ZONE_RESET: 904 case REQ_OP_ZONE_FINISH: 905 /* 906 * Zone management operations require sequential zones that are 907 * not offline nor read-only. 908 */ 909 if (!bdev_zone_mgmt_allowed(bdev, bio->bi_iter.bi_sector)) 910 goto end_io; 911 break; 912 case REQ_OP_ZONE_RESET_ALL: 913 if (!bdev_is_zoned(bio->bi_bdev)) 914 goto not_supported; 915 break; 916 case REQ_OP_DRV_IN: 917 case REQ_OP_DRV_OUT: 918 /* 919 * Driver private operations are only used with passthrough 920 * requests. 921 */ 922 fallthrough; 923 default: 924 goto not_supported; 925 } 926 927 if (blk_throtl_bio(bio)) 928 return; 929 submit_bio_noacct_nocheck(bio, false); 930 return; 931 932 not_supported: 933 status = BLK_STS_NOTSUPP; 934 end_io: 935 bio_endio_status(bio, status); 936 } 937 EXPORT_SYMBOL(submit_bio_noacct); 938 939 static void bio_set_ioprio(struct bio *bio) 940 { 941 /* Nobody set ioprio so far? Initialize it based on task's nice value */ 942 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE) 943 bio->bi_ioprio = get_current_ioprio(); 944 blkcg_set_ioprio(bio); 945 } 946 947 /** 948 * submit_bio - submit a bio to the block device layer for I/O 949 * @bio: The &struct bio which describes the I/O 950 * 951 * submit_bio() is used to submit I/O requests to block devices. It is passed a 952 * fully set up &struct bio that describes the I/O that needs to be done. The 953 * bio will be sent to the device described by the bi_bdev field. 954 * 955 * The success/failure status of the request, along with notification of 956 * completion, is delivered asynchronously through the ->bi_end_io() callback 957 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has 958 * been called. 959 */ 960 void submit_bio(struct bio *bio) 961 { 962 if (bio_op(bio) == REQ_OP_READ) { 963 task_io_account_read(bio->bi_iter.bi_size); 964 count_vm_events(PGPGIN, bio_sectors(bio)); 965 } else if (bio_op(bio) == REQ_OP_WRITE) { 966 count_vm_events(PGPGOUT, bio_sectors(bio)); 967 } 968 969 bio_set_ioprio(bio); 970 submit_bio_noacct(bio); 971 } 972 EXPORT_SYMBOL(submit_bio); 973 974 /** 975 * bio_poll - poll for BIO completions 976 * @bio: bio to poll for 977 * @iob: batches of IO 978 * @flags: BLK_POLL_* flags that control the behavior 979 * 980 * Poll for completions on queue associated with the bio. Returns number of 981 * completed entries found. 982 * 983 * Note: the caller must either be the context that submitted @bio, or 984 * be in a RCU critical section to prevent freeing of @bio. 985 */ 986 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) 987 { 988 blk_qc_t cookie = READ_ONCE(bio->bi_cookie); 989 struct block_device *bdev; 990 struct request_queue *q; 991 int ret = 0; 992 993 bdev = READ_ONCE(bio->bi_bdev); 994 if (!bdev) 995 return 0; 996 997 q = bdev_get_queue(bdev); 998 if (cookie == BLK_QC_T_NONE) 999 return 0; 1000 1001 blk_flush_plug(current->plug, false); 1002 1003 /* 1004 * We need to be able to enter a frozen queue, similar to how 1005 * timeouts also need to do that. If that is blocked, then we can 1006 * have pending IO when a queue freeze is started, and then the 1007 * wait for the freeze to finish will wait for polled requests to 1008 * timeout as the poller is preventer from entering the queue and 1009 * completing them. As long as we prevent new IO from being queued, 1010 * that should be all that matters. 1011 */ 1012 if (!percpu_ref_tryget(&q->q_usage_counter)) 1013 return 0; 1014 if (queue_is_mq(q)) { 1015 ret = blk_mq_poll(q, cookie, iob, flags); 1016 } else { 1017 struct gendisk *disk = q->disk; 1018 1019 if ((q->limits.features & BLK_FEAT_POLL) && disk && 1020 disk->fops->poll_bio) 1021 ret = disk->fops->poll_bio(bio, iob, flags); 1022 } 1023 blk_queue_exit(q); 1024 return ret; 1025 } 1026 EXPORT_SYMBOL_GPL(bio_poll); 1027 1028 /* 1029 * Helper to implement file_operations.iopoll. Requires the bio to be stored 1030 * in iocb->private, and cleared before freeing the bio. 1031 */ 1032 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, 1033 unsigned int flags) 1034 { 1035 struct bio *bio; 1036 int ret = 0; 1037 1038 /* 1039 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can 1040 * point to a freshly allocated bio at this point. If that happens 1041 * we have a few cases to consider: 1042 * 1043 * 1) the bio is being initialized and bi_bdev is NULL. We can just 1044 * simply nothing in this case 1045 * 2) the bio points to a not poll enabled device. bio_poll will catch 1046 * this and return 0 1047 * 3) the bio points to a poll capable device, including but not 1048 * limited to the one that the original bio pointed to. In this 1049 * case we will call into the actual poll method and poll for I/O, 1050 * even if we don't need to, but it won't cause harm either. 1051 * 1052 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev 1053 * is still allocated. Because partitions hold a reference to the whole 1054 * device bdev and thus disk, the disk is also still valid. Grabbing 1055 * a reference to the queue in bio_poll() ensures the hctxs and requests 1056 * are still valid as well. 1057 */ 1058 rcu_read_lock(); 1059 bio = READ_ONCE(kiocb->private); 1060 if (bio) 1061 ret = bio_poll(bio, iob, flags); 1062 rcu_read_unlock(); 1063 1064 return ret; 1065 } 1066 EXPORT_SYMBOL_GPL(iocb_bio_iopoll); 1067 1068 void update_io_ticks(struct block_device *part, unsigned long now, bool end) 1069 { 1070 unsigned long stamp; 1071 again: 1072 stamp = READ_ONCE(part->bd_stamp); 1073 if (unlikely(time_after(now, stamp)) && 1074 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) && 1075 (end || bdev_count_inflight(part))) 1076 __part_stat_add(part, io_ticks, now - stamp); 1077 1078 if (bdev_is_partition(part)) { 1079 part = bdev_whole(part); 1080 goto again; 1081 } 1082 } 1083 1084 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op, 1085 unsigned long start_time) 1086 { 1087 part_stat_lock(); 1088 update_io_ticks(bdev, start_time, false); 1089 bdev_inc_in_flight(bdev, op); 1090 part_stat_unlock(); 1091 1092 return start_time; 1093 } 1094 EXPORT_SYMBOL(bdev_start_io_acct); 1095 1096 /** 1097 * bio_start_io_acct - start I/O accounting for bio based drivers 1098 * @bio: bio to start account for 1099 * 1100 * Returns the start time that should be passed back to bio_end_io_acct(). 1101 */ 1102 unsigned long bio_start_io_acct(struct bio *bio) 1103 { 1104 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies); 1105 } 1106 EXPORT_SYMBOL_GPL(bio_start_io_acct); 1107 1108 void bdev_end_io_acct(struct block_device *bdev, enum req_op op, 1109 unsigned int sectors, unsigned long start_time) 1110 { 1111 const int sgrp = op_stat_group(op); 1112 unsigned long now = READ_ONCE(jiffies); 1113 unsigned long duration = now - start_time; 1114 1115 part_stat_lock(); 1116 update_io_ticks(bdev, now, true); 1117 part_stat_inc(bdev, ios[sgrp]); 1118 part_stat_add(bdev, sectors[sgrp], sectors); 1119 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); 1120 bdev_dec_in_flight(bdev, op); 1121 part_stat_unlock(); 1122 } 1123 EXPORT_SYMBOL(bdev_end_io_acct); 1124 1125 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, 1126 struct block_device *orig_bdev) 1127 { 1128 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time); 1129 } 1130 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); 1131 1132 /** 1133 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1134 * @q : the queue of the device being checked 1135 * 1136 * Description: 1137 * Check if underlying low-level drivers of a device are busy. 1138 * If the drivers want to export their busy state, they must set own 1139 * exporting function using blk_queue_lld_busy() first. 1140 * 1141 * Basically, this function is used only by request stacking drivers 1142 * to stop dispatching requests to underlying devices when underlying 1143 * devices are busy. This behavior helps more I/O merging on the queue 1144 * of the request stacking driver and prevents I/O throughput regression 1145 * on burst I/O load. 1146 * 1147 * Return: 1148 * 0 - Not busy (The request stacking driver should dispatch request) 1149 * 1 - Busy (The request stacking driver should stop dispatching request) 1150 */ 1151 int blk_lld_busy(struct request_queue *q) 1152 { 1153 if (queue_is_mq(q) && q->mq_ops->busy) 1154 return q->mq_ops->busy(q); 1155 1156 return 0; 1157 } 1158 EXPORT_SYMBOL_GPL(blk_lld_busy); 1159 1160 int kblockd_schedule_work(struct work_struct *work) 1161 { 1162 return queue_work(kblockd_workqueue, work); 1163 } 1164 EXPORT_SYMBOL(kblockd_schedule_work); 1165 1166 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1167 unsigned long delay) 1168 { 1169 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1170 } 1171 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1172 1173 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) 1174 { 1175 struct task_struct *tsk = current; 1176 1177 /* 1178 * If this is a nested plug, don't actually assign it. 1179 */ 1180 if (tsk->plug) 1181 return; 1182 1183 plug->cur_ktime = 0; 1184 rq_list_init(&plug->mq_list); 1185 rq_list_init(&plug->cached_rqs); 1186 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); 1187 plug->rq_count = 0; 1188 plug->multiple_queues = false; 1189 plug->has_elevator = false; 1190 INIT_LIST_HEAD(&plug->cb_list); 1191 1192 /* 1193 * Store ordering should not be needed here, since a potential 1194 * preempt will imply a full memory barrier 1195 */ 1196 tsk->plug = plug; 1197 } 1198 1199 /** 1200 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1201 * @plug: The &struct blk_plug that needs to be initialized 1202 * 1203 * Description: 1204 * blk_start_plug() indicates to the block layer an intent by the caller 1205 * to submit multiple I/O requests in a batch. The block layer may use 1206 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1207 * is called. However, the block layer may choose to submit requests 1208 * before a call to blk_finish_plug() if the number of queued I/Os 1209 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1210 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1211 * the task schedules (see below). 1212 * 1213 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1214 * pending I/O should the task end up blocking between blk_start_plug() and 1215 * blk_finish_plug(). This is important from a performance perspective, but 1216 * also ensures that we don't deadlock. For instance, if the task is blocking 1217 * for a memory allocation, memory reclaim could end up wanting to free a 1218 * page belonging to that request that is currently residing in our private 1219 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1220 * this kind of deadlock. 1221 */ 1222 void blk_start_plug(struct blk_plug *plug) 1223 { 1224 blk_start_plug_nr_ios(plug, 1); 1225 } 1226 EXPORT_SYMBOL(blk_start_plug); 1227 1228 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1229 { 1230 LIST_HEAD(callbacks); 1231 1232 while (!list_empty(&plug->cb_list)) { 1233 list_splice_init(&plug->cb_list, &callbacks); 1234 1235 while (!list_empty(&callbacks)) { 1236 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1237 struct blk_plug_cb, 1238 list); 1239 list_del(&cb->list); 1240 cb->callback(cb, from_schedule); 1241 } 1242 } 1243 } 1244 1245 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1246 int size) 1247 { 1248 struct blk_plug *plug = current->plug; 1249 struct blk_plug_cb *cb; 1250 1251 if (!plug) 1252 return NULL; 1253 1254 list_for_each_entry(cb, &plug->cb_list, list) 1255 if (cb->callback == unplug && cb->data == data) 1256 return cb; 1257 1258 /* Not currently on the callback list */ 1259 BUG_ON(size < sizeof(*cb)); 1260 cb = kzalloc(size, GFP_ATOMIC); 1261 if (cb) { 1262 cb->data = data; 1263 cb->callback = unplug; 1264 list_add(&cb->list, &plug->cb_list); 1265 } 1266 return cb; 1267 } 1268 EXPORT_SYMBOL(blk_check_plugged); 1269 1270 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) 1271 { 1272 if (!list_empty(&plug->cb_list)) 1273 flush_plug_callbacks(plug, from_schedule); 1274 blk_mq_flush_plug_list(plug, from_schedule); 1275 /* 1276 * Unconditionally flush out cached requests, even if the unplug 1277 * event came from schedule. Since we know hold references to the 1278 * queue for cached requests, we don't want a blocked task holding 1279 * up a queue freeze/quiesce event. 1280 */ 1281 if (unlikely(!rq_list_empty(&plug->cached_rqs))) 1282 blk_mq_free_plug_rqs(plug); 1283 1284 plug->cur_ktime = 0; 1285 current->flags &= ~PF_BLOCK_TS; 1286 } 1287 1288 /** 1289 * blk_finish_plug - mark the end of a batch of submitted I/O 1290 * @plug: The &struct blk_plug passed to blk_start_plug() 1291 * 1292 * Description: 1293 * Indicate that a batch of I/O submissions is complete. This function 1294 * must be paired with an initial call to blk_start_plug(). The intent 1295 * is to allow the block layer to optimize I/O submission. See the 1296 * documentation for blk_start_plug() for more information. 1297 */ 1298 void blk_finish_plug(struct blk_plug *plug) 1299 { 1300 if (plug == current->plug) { 1301 __blk_flush_plug(plug, false); 1302 current->plug = NULL; 1303 } 1304 } 1305 EXPORT_SYMBOL(blk_finish_plug); 1306 1307 void blk_io_schedule(void) 1308 { 1309 /* Prevent hang_check timer from firing at us during very long I/O */ 1310 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1311 1312 if (timeout) 1313 io_schedule_timeout(timeout); 1314 else 1315 io_schedule(); 1316 } 1317 1318 int __init blk_dev_init(void) 1319 { 1320 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1321 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1322 sizeof_field(struct request, cmd_flags)); 1323 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1324 sizeof_field(struct bio, bi_opf)); 1325 1326 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1327 kblockd_workqueue = alloc_workqueue("kblockd", 1328 WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_PERCPU, 0); 1329 if (!kblockd_workqueue) 1330 panic("Failed to create kblockd\n"); 1331 1332 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC); 1333 1334 blk_debugfs_root = debugfs_create_dir("block", NULL); 1335 1336 return 0; 1337 } 1338