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/psi.h> 41 #include <linux/part_stat.h> 42 #include <linux/sched/sysctl.h> 43 #include <linux/blk-crypto.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-rq-qos.h" 54 55 struct dentry *blk_debugfs_root; 56 57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); 58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); 59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); 60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); 61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); 62 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert); 63 64 DEFINE_IDA(blk_queue_ida); 65 66 /* 67 * For queue allocation 68 */ 69 struct kmem_cache *blk_requestq_cachep; 70 struct kmem_cache *blk_requestq_srcu_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 /** 100 * blk_queue_flag_test_and_set - atomically test and set a queue flag 101 * @flag: flag to be set 102 * @q: request queue 103 * 104 * Returns the previous value of @flag - 0 if the flag was not set and 1 if 105 * the flag was already set. 106 */ 107 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) 108 { 109 return test_and_set_bit(flag, &q->queue_flags); 110 } 111 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); 112 113 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name 114 static const char *const blk_op_name[] = { 115 REQ_OP_NAME(READ), 116 REQ_OP_NAME(WRITE), 117 REQ_OP_NAME(FLUSH), 118 REQ_OP_NAME(DISCARD), 119 REQ_OP_NAME(SECURE_ERASE), 120 REQ_OP_NAME(ZONE_RESET), 121 REQ_OP_NAME(ZONE_RESET_ALL), 122 REQ_OP_NAME(ZONE_OPEN), 123 REQ_OP_NAME(ZONE_CLOSE), 124 REQ_OP_NAME(ZONE_FINISH), 125 REQ_OP_NAME(ZONE_APPEND), 126 REQ_OP_NAME(WRITE_ZEROES), 127 REQ_OP_NAME(DRV_IN), 128 REQ_OP_NAME(DRV_OUT), 129 }; 130 #undef REQ_OP_NAME 131 132 /** 133 * blk_op_str - Return string XXX in the REQ_OP_XXX. 134 * @op: REQ_OP_XXX. 135 * 136 * Description: Centralize block layer function to convert REQ_OP_XXX into 137 * string format. Useful in the debugging and tracing bio or request. For 138 * invalid REQ_OP_XXX it returns string "UNKNOWN". 139 */ 140 inline const char *blk_op_str(unsigned int op) 141 { 142 const char *op_str = "UNKNOWN"; 143 144 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) 145 op_str = blk_op_name[op]; 146 147 return op_str; 148 } 149 EXPORT_SYMBOL_GPL(blk_op_str); 150 151 static const struct { 152 int errno; 153 const char *name; 154 } blk_errors[] = { 155 [BLK_STS_OK] = { 0, "" }, 156 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, 157 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, 158 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, 159 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, 160 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, 161 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, 162 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, 163 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, 164 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, 165 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, 166 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, 167 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" }, 168 169 /* device mapper special case, should not leak out: */ 170 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, 171 172 /* zone device specific errors */ 173 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, 174 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, 175 176 /* everything else not covered above: */ 177 [BLK_STS_IOERR] = { -EIO, "I/O" }, 178 }; 179 180 blk_status_t errno_to_blk_status(int errno) 181 { 182 int i; 183 184 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { 185 if (blk_errors[i].errno == errno) 186 return (__force blk_status_t)i; 187 } 188 189 return BLK_STS_IOERR; 190 } 191 EXPORT_SYMBOL_GPL(errno_to_blk_status); 192 193 int blk_status_to_errno(blk_status_t status) 194 { 195 int idx = (__force int)status; 196 197 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 198 return -EIO; 199 return blk_errors[idx].errno; 200 } 201 EXPORT_SYMBOL_GPL(blk_status_to_errno); 202 203 const char *blk_status_to_str(blk_status_t status) 204 { 205 int idx = (__force int)status; 206 207 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 208 return "<null>"; 209 return blk_errors[idx].name; 210 } 211 212 /** 213 * blk_sync_queue - cancel any pending callbacks on a queue 214 * @q: the queue 215 * 216 * Description: 217 * The block layer may perform asynchronous callback activity 218 * on a queue, such as calling the unplug function after a timeout. 219 * A block device may call blk_sync_queue to ensure that any 220 * such activity is cancelled, thus allowing it to release resources 221 * that the callbacks might use. The caller must already have made sure 222 * that its ->submit_bio will not re-add plugging prior to calling 223 * this function. 224 * 225 * This function does not cancel any asynchronous activity arising 226 * out of elevator or throttling code. That would require elevator_exit() 227 * and blkcg_exit_queue() to be called with queue lock initialized. 228 * 229 */ 230 void blk_sync_queue(struct request_queue *q) 231 { 232 del_timer_sync(&q->timeout); 233 cancel_work_sync(&q->timeout_work); 234 } 235 EXPORT_SYMBOL(blk_sync_queue); 236 237 /** 238 * blk_set_pm_only - increment pm_only counter 239 * @q: request queue pointer 240 */ 241 void blk_set_pm_only(struct request_queue *q) 242 { 243 atomic_inc(&q->pm_only); 244 } 245 EXPORT_SYMBOL_GPL(blk_set_pm_only); 246 247 void blk_clear_pm_only(struct request_queue *q) 248 { 249 int pm_only; 250 251 pm_only = atomic_dec_return(&q->pm_only); 252 WARN_ON_ONCE(pm_only < 0); 253 if (pm_only == 0) 254 wake_up_all(&q->mq_freeze_wq); 255 } 256 EXPORT_SYMBOL_GPL(blk_clear_pm_only); 257 258 /** 259 * blk_put_queue - decrement the request_queue refcount 260 * @q: the request_queue structure to decrement the refcount for 261 * 262 * Decrements the refcount of the request_queue kobject. When this reaches 0 263 * we'll have blk_release_queue() called. 264 * 265 * Context: Any context, but the last reference must not be dropped from 266 * atomic context. 267 */ 268 void blk_put_queue(struct request_queue *q) 269 { 270 kobject_put(&q->kobj); 271 } 272 EXPORT_SYMBOL(blk_put_queue); 273 274 void blk_queue_start_drain(struct request_queue *q) 275 { 276 /* 277 * When queue DYING flag is set, we need to block new req 278 * entering queue, so we call blk_freeze_queue_start() to 279 * prevent I/O from crossing blk_queue_enter(). 280 */ 281 blk_freeze_queue_start(q); 282 if (queue_is_mq(q)) 283 blk_mq_wake_waiters(q); 284 /* Make blk_queue_enter() reexamine the DYING flag. */ 285 wake_up_all(&q->mq_freeze_wq); 286 } 287 288 /** 289 * blk_cleanup_queue - shutdown a request queue 290 * @q: request queue to shutdown 291 * 292 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and 293 * put it. All future requests will be failed immediately with -ENODEV. 294 * 295 * Context: can sleep 296 */ 297 void blk_cleanup_queue(struct request_queue *q) 298 { 299 /* cannot be called from atomic context */ 300 might_sleep(); 301 302 WARN_ON_ONCE(blk_queue_registered(q)); 303 304 /* mark @q DYING, no new request or merges will be allowed afterwards */ 305 blk_queue_flag_set(QUEUE_FLAG_DYING, q); 306 blk_queue_start_drain(q); 307 308 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); 309 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); 310 311 /* 312 * Drain all requests queued before DYING marking. Set DEAD flag to 313 * prevent that blk_mq_run_hw_queues() accesses the hardware queues 314 * after draining finished. 315 */ 316 blk_freeze_queue(q); 317 318 /* cleanup rq qos structures for queue without disk */ 319 rq_qos_exit(q); 320 321 blk_queue_flag_set(QUEUE_FLAG_DEAD, q); 322 323 blk_sync_queue(q); 324 if (queue_is_mq(q)) { 325 blk_mq_cancel_work_sync(q); 326 blk_mq_exit_queue(q); 327 } 328 329 /* 330 * In theory, request pool of sched_tags belongs to request queue. 331 * However, the current implementation requires tag_set for freeing 332 * requests, so free the pool now. 333 * 334 * Queue has become frozen, there can't be any in-queue requests, so 335 * it is safe to free requests now. 336 */ 337 mutex_lock(&q->sysfs_lock); 338 if (q->elevator) 339 blk_mq_sched_free_rqs(q); 340 mutex_unlock(&q->sysfs_lock); 341 342 /* @q is and will stay empty, shutdown and put */ 343 blk_put_queue(q); 344 } 345 EXPORT_SYMBOL(blk_cleanup_queue); 346 347 /** 348 * blk_queue_enter() - try to increase q->q_usage_counter 349 * @q: request queue pointer 350 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM 351 */ 352 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) 353 { 354 const bool pm = flags & BLK_MQ_REQ_PM; 355 356 while (!blk_try_enter_queue(q, pm)) { 357 if (flags & BLK_MQ_REQ_NOWAIT) 358 return -EBUSY; 359 360 /* 361 * read pair of barrier in blk_freeze_queue_start(), we need to 362 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 363 * reading .mq_freeze_depth or queue dying flag, otherwise the 364 * following wait may never return if the two reads are 365 * reordered. 366 */ 367 smp_rmb(); 368 wait_event(q->mq_freeze_wq, 369 (!q->mq_freeze_depth && 370 blk_pm_resume_queue(pm, q)) || 371 blk_queue_dying(q)); 372 if (blk_queue_dying(q)) 373 return -ENODEV; 374 } 375 376 return 0; 377 } 378 379 int __bio_queue_enter(struct request_queue *q, struct bio *bio) 380 { 381 while (!blk_try_enter_queue(q, false)) { 382 struct gendisk *disk = bio->bi_bdev->bd_disk; 383 384 if (bio->bi_opf & REQ_NOWAIT) { 385 if (test_bit(GD_DEAD, &disk->state)) 386 goto dead; 387 bio_wouldblock_error(bio); 388 return -EBUSY; 389 } 390 391 /* 392 * read pair of barrier in blk_freeze_queue_start(), we need to 393 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 394 * reading .mq_freeze_depth or queue dying flag, otherwise the 395 * following wait may never return if the two reads are 396 * reordered. 397 */ 398 smp_rmb(); 399 wait_event(q->mq_freeze_wq, 400 (!q->mq_freeze_depth && 401 blk_pm_resume_queue(false, q)) || 402 test_bit(GD_DEAD, &disk->state)); 403 if (test_bit(GD_DEAD, &disk->state)) 404 goto dead; 405 } 406 407 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 = from_timer(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(int node_id, bool alloc_srcu) 438 { 439 struct request_queue *q; 440 int ret; 441 442 q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu), 443 GFP_KERNEL | __GFP_ZERO, node_id); 444 if (!q) 445 return NULL; 446 447 if (alloc_srcu) { 448 blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q); 449 if (init_srcu_struct(q->srcu) != 0) 450 goto fail_q; 451 } 452 453 q->last_merge = NULL; 454 455 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); 456 if (q->id < 0) 457 goto fail_srcu; 458 459 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0); 460 if (ret) 461 goto fail_id; 462 463 q->stats = blk_alloc_queue_stats(); 464 if (!q->stats) 465 goto fail_split; 466 467 q->node = node_id; 468 469 atomic_set(&q->nr_active_requests_shared_tags, 0); 470 471 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); 472 INIT_WORK(&q->timeout_work, blk_timeout_work); 473 INIT_LIST_HEAD(&q->icq_list); 474 475 kobject_init(&q->kobj, &blk_queue_ktype); 476 477 mutex_init(&q->debugfs_mutex); 478 mutex_init(&q->sysfs_lock); 479 mutex_init(&q->sysfs_dir_lock); 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 /* 486 * Init percpu_ref in atomic mode so that it's faster to shutdown. 487 * See blk_register_queue() for details. 488 */ 489 if (percpu_ref_init(&q->q_usage_counter, 490 blk_queue_usage_counter_release, 491 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) 492 goto fail_stats; 493 494 blk_queue_dma_alignment(q, 511); 495 blk_set_default_limits(&q->limits); 496 q->nr_requests = BLKDEV_DEFAULT_RQ; 497 498 return q; 499 500 fail_stats: 501 blk_free_queue_stats(q->stats); 502 fail_split: 503 bioset_exit(&q->bio_split); 504 fail_id: 505 ida_simple_remove(&blk_queue_ida, q->id); 506 fail_srcu: 507 if (alloc_srcu) 508 cleanup_srcu_struct(q->srcu); 509 fail_q: 510 kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q); 511 return NULL; 512 } 513 514 /** 515 * blk_get_queue - increment the request_queue refcount 516 * @q: the request_queue structure to increment the refcount for 517 * 518 * Increment the refcount of the request_queue kobject. 519 * 520 * Context: Any context. 521 */ 522 bool blk_get_queue(struct request_queue *q) 523 { 524 if (likely(!blk_queue_dying(q))) { 525 __blk_get_queue(q); 526 return true; 527 } 528 529 return false; 530 } 531 EXPORT_SYMBOL(blk_get_queue); 532 533 #ifdef CONFIG_FAIL_MAKE_REQUEST 534 535 static DECLARE_FAULT_ATTR(fail_make_request); 536 537 static int __init setup_fail_make_request(char *str) 538 { 539 return setup_fault_attr(&fail_make_request, str); 540 } 541 __setup("fail_make_request=", setup_fail_make_request); 542 543 bool should_fail_request(struct block_device *part, unsigned int bytes) 544 { 545 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); 546 } 547 548 static int __init fail_make_request_debugfs(void) 549 { 550 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 551 NULL, &fail_make_request); 552 553 return PTR_ERR_OR_ZERO(dir); 554 } 555 556 late_initcall(fail_make_request_debugfs); 557 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 558 559 static inline bool bio_check_ro(struct bio *bio) 560 { 561 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { 562 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) 563 return false; 564 pr_warn("Trying to write to read-only block-device %pg\n", 565 bio->bi_bdev); 566 /* Older lvm-tools actually trigger this */ 567 return false; 568 } 569 570 return false; 571 } 572 573 static noinline int should_fail_bio(struct bio *bio) 574 { 575 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) 576 return -EIO; 577 return 0; 578 } 579 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); 580 581 /* 582 * Check whether this bio extends beyond the end of the device or partition. 583 * This may well happen - the kernel calls bread() without checking the size of 584 * the device, e.g., when mounting a file system. 585 */ 586 static inline int bio_check_eod(struct bio *bio) 587 { 588 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 589 unsigned int nr_sectors = bio_sectors(bio); 590 591 if (nr_sectors && maxsector && 592 (nr_sectors > maxsector || 593 bio->bi_iter.bi_sector > maxsector - nr_sectors)) { 594 pr_info_ratelimited("%s: attempt to access beyond end of device\n" 595 "%pg: rw=%d, want=%llu, limit=%llu\n", 596 current->comm, 597 bio->bi_bdev, bio->bi_opf, 598 bio_end_sector(bio), maxsector); 599 return -EIO; 600 } 601 return 0; 602 } 603 604 /* 605 * Remap block n of partition p to block n+start(p) of the disk. 606 */ 607 static int blk_partition_remap(struct bio *bio) 608 { 609 struct block_device *p = bio->bi_bdev; 610 611 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 612 return -EIO; 613 if (bio_sectors(bio)) { 614 bio->bi_iter.bi_sector += p->bd_start_sect; 615 trace_block_bio_remap(bio, p->bd_dev, 616 bio->bi_iter.bi_sector - 617 p->bd_start_sect); 618 } 619 bio_set_flag(bio, BIO_REMAPPED); 620 return 0; 621 } 622 623 /* 624 * Check write append to a zoned block device. 625 */ 626 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 627 struct bio *bio) 628 { 629 sector_t pos = bio->bi_iter.bi_sector; 630 int nr_sectors = bio_sectors(bio); 631 632 /* Only applicable to zoned block devices */ 633 if (!blk_queue_is_zoned(q)) 634 return BLK_STS_NOTSUPP; 635 636 /* The bio sector must point to the start of a sequential zone */ 637 if (pos & (blk_queue_zone_sectors(q) - 1) || 638 !blk_queue_zone_is_seq(q, pos)) 639 return BLK_STS_IOERR; 640 641 /* 642 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 643 * split and could result in non-contiguous sectors being written in 644 * different zones. 645 */ 646 if (nr_sectors > q->limits.chunk_sectors) 647 return BLK_STS_IOERR; 648 649 /* Make sure the BIO is small enough and will not get split */ 650 if (nr_sectors > q->limits.max_zone_append_sectors) 651 return BLK_STS_IOERR; 652 653 bio->bi_opf |= REQ_NOMERGE; 654 655 return BLK_STS_OK; 656 } 657 658 static void __submit_bio(struct bio *bio) 659 { 660 struct gendisk *disk = bio->bi_bdev->bd_disk; 661 662 if (unlikely(!blk_crypto_bio_prep(&bio))) 663 return; 664 665 if (!disk->fops->submit_bio) { 666 blk_mq_submit_bio(bio); 667 } else if (likely(bio_queue_enter(bio) == 0)) { 668 disk->fops->submit_bio(bio); 669 blk_queue_exit(disk->queue); 670 } 671 } 672 673 /* 674 * The loop in this function may be a bit non-obvious, and so deserves some 675 * explanation: 676 * 677 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 678 * that), so we have a list with a single bio. 679 * - We pretend that we have just taken it off a longer list, so we assign 680 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 681 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 682 * bios through a recursive call to submit_bio_noacct. If it did, we find a 683 * non-NULL value in bio_list and re-enter the loop from the top. 684 * - In this case we really did just take the bio of the top of the list (no 685 * pretending) and so remove it from bio_list, and call into ->submit_bio() 686 * again. 687 * 688 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 689 * bio_list_on_stack[1] contains bios that were submitted before the current 690 * ->submit_bio, but that haven't been processed yet. 691 */ 692 static void __submit_bio_noacct(struct bio *bio) 693 { 694 struct bio_list bio_list_on_stack[2]; 695 696 BUG_ON(bio->bi_next); 697 698 bio_list_init(&bio_list_on_stack[0]); 699 current->bio_list = bio_list_on_stack; 700 701 do { 702 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 703 struct bio_list lower, same; 704 705 /* 706 * Create a fresh bio_list for all subordinate requests. 707 */ 708 bio_list_on_stack[1] = bio_list_on_stack[0]; 709 bio_list_init(&bio_list_on_stack[0]); 710 711 __submit_bio(bio); 712 713 /* 714 * Sort new bios into those for a lower level and those for the 715 * same level. 716 */ 717 bio_list_init(&lower); 718 bio_list_init(&same); 719 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 720 if (q == bdev_get_queue(bio->bi_bdev)) 721 bio_list_add(&same, bio); 722 else 723 bio_list_add(&lower, bio); 724 725 /* 726 * Now assemble so we handle the lowest level first. 727 */ 728 bio_list_merge(&bio_list_on_stack[0], &lower); 729 bio_list_merge(&bio_list_on_stack[0], &same); 730 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 731 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 732 733 current->bio_list = NULL; 734 } 735 736 static void __submit_bio_noacct_mq(struct bio *bio) 737 { 738 struct bio_list bio_list[2] = { }; 739 740 current->bio_list = bio_list; 741 742 do { 743 __submit_bio(bio); 744 } while ((bio = bio_list_pop(&bio_list[0]))); 745 746 current->bio_list = NULL; 747 } 748 749 void submit_bio_noacct_nocheck(struct bio *bio) 750 { 751 /* 752 * We only want one ->submit_bio to be active at a time, else stack 753 * usage with stacked devices could be a problem. Use current->bio_list 754 * to collect a list of requests submited by a ->submit_bio method while 755 * it is active, and then process them after it returned. 756 */ 757 if (current->bio_list) 758 bio_list_add(¤t->bio_list[0], bio); 759 else if (!bio->bi_bdev->bd_disk->fops->submit_bio) 760 __submit_bio_noacct_mq(bio); 761 else 762 __submit_bio_noacct(bio); 763 } 764 765 /** 766 * submit_bio_noacct - re-submit a bio to the block device layer for I/O 767 * @bio: The bio describing the location in memory and on the device. 768 * 769 * This is a version of submit_bio() that shall only be used for I/O that is 770 * resubmitted to lower level drivers by stacking block drivers. All file 771 * systems and other upper level users of the block layer should use 772 * submit_bio() instead. 773 */ 774 void submit_bio_noacct(struct bio *bio) 775 { 776 struct block_device *bdev = bio->bi_bdev; 777 struct request_queue *q = bdev_get_queue(bdev); 778 blk_status_t status = BLK_STS_IOERR; 779 struct blk_plug *plug; 780 781 might_sleep(); 782 783 plug = blk_mq_plug(q, bio); 784 if (plug && plug->nowait) 785 bio->bi_opf |= REQ_NOWAIT; 786 787 /* 788 * For a REQ_NOWAIT based request, return -EOPNOTSUPP 789 * if queue does not support NOWAIT. 790 */ 791 if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q)) 792 goto not_supported; 793 794 if (should_fail_bio(bio)) 795 goto end_io; 796 if (unlikely(bio_check_ro(bio))) 797 goto end_io; 798 if (!bio_flagged(bio, BIO_REMAPPED)) { 799 if (unlikely(bio_check_eod(bio))) 800 goto end_io; 801 if (bdev->bd_partno && unlikely(blk_partition_remap(bio))) 802 goto end_io; 803 } 804 805 /* 806 * Filter flush bio's early so that bio based drivers without flush 807 * support don't have to worry about them. 808 */ 809 if (op_is_flush(bio->bi_opf) && 810 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { 811 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 812 if (!bio_sectors(bio)) { 813 status = BLK_STS_OK; 814 goto end_io; 815 } 816 } 817 818 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 819 bio_clear_polled(bio); 820 821 switch (bio_op(bio)) { 822 case REQ_OP_DISCARD: 823 if (!blk_queue_discard(q)) 824 goto not_supported; 825 break; 826 case REQ_OP_SECURE_ERASE: 827 if (!blk_queue_secure_erase(q)) 828 goto not_supported; 829 break; 830 case REQ_OP_ZONE_APPEND: 831 status = blk_check_zone_append(q, bio); 832 if (status != BLK_STS_OK) 833 goto end_io; 834 break; 835 case REQ_OP_ZONE_RESET: 836 case REQ_OP_ZONE_OPEN: 837 case REQ_OP_ZONE_CLOSE: 838 case REQ_OP_ZONE_FINISH: 839 if (!blk_queue_is_zoned(q)) 840 goto not_supported; 841 break; 842 case REQ_OP_ZONE_RESET_ALL: 843 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q)) 844 goto not_supported; 845 break; 846 case REQ_OP_WRITE_ZEROES: 847 if (!q->limits.max_write_zeroes_sectors) 848 goto not_supported; 849 break; 850 default: 851 break; 852 } 853 854 if (blk_throtl_bio(bio)) 855 return; 856 857 blk_cgroup_bio_start(bio); 858 blkcg_bio_issue_init(bio); 859 860 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 861 trace_block_bio_queue(bio); 862 /* Now that enqueuing has been traced, we need to trace 863 * completion as well. 864 */ 865 bio_set_flag(bio, BIO_TRACE_COMPLETION); 866 } 867 submit_bio_noacct_nocheck(bio); 868 return; 869 870 not_supported: 871 status = BLK_STS_NOTSUPP; 872 end_io: 873 bio->bi_status = status; 874 bio_endio(bio); 875 } 876 EXPORT_SYMBOL(submit_bio_noacct); 877 878 /** 879 * submit_bio - submit a bio to the block device layer for I/O 880 * @bio: The &struct bio which describes the I/O 881 * 882 * submit_bio() is used to submit I/O requests to block devices. It is passed a 883 * fully set up &struct bio that describes the I/O that needs to be done. The 884 * bio will be send to the device described by the bi_bdev field. 885 * 886 * The success/failure status of the request, along with notification of 887 * completion, is delivered asynchronously through the ->bi_end_io() callback 888 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has 889 * been called. 890 */ 891 void submit_bio(struct bio *bio) 892 { 893 if (blkcg_punt_bio_submit(bio)) 894 return; 895 896 /* 897 * If it's a regular read/write or a barrier with data attached, 898 * go through the normal accounting stuff before submission. 899 */ 900 if (bio_has_data(bio)) { 901 unsigned int count = bio_sectors(bio); 902 903 if (op_is_write(bio_op(bio))) { 904 count_vm_events(PGPGOUT, count); 905 } else { 906 task_io_account_read(bio->bi_iter.bi_size); 907 count_vm_events(PGPGIN, count); 908 } 909 } 910 911 /* 912 * If we're reading data that is part of the userspace workingset, count 913 * submission time as memory stall. When the device is congested, or 914 * the submitting cgroup IO-throttled, submission can be a significant 915 * part of overall IO time. 916 */ 917 if (unlikely(bio_op(bio) == REQ_OP_READ && 918 bio_flagged(bio, BIO_WORKINGSET))) { 919 unsigned long pflags; 920 921 psi_memstall_enter(&pflags); 922 submit_bio_noacct(bio); 923 psi_memstall_leave(&pflags); 924 return; 925 } 926 927 submit_bio_noacct(bio); 928 } 929 EXPORT_SYMBOL(submit_bio); 930 931 /** 932 * bio_poll - poll for BIO completions 933 * @bio: bio to poll for 934 * @iob: batches of IO 935 * @flags: BLK_POLL_* flags that control the behavior 936 * 937 * Poll for completions on queue associated with the bio. Returns number of 938 * completed entries found. 939 * 940 * Note: the caller must either be the context that submitted @bio, or 941 * be in a RCU critical section to prevent freeing of @bio. 942 */ 943 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) 944 { 945 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 946 blk_qc_t cookie = READ_ONCE(bio->bi_cookie); 947 int ret = 0; 948 949 if (cookie == BLK_QC_T_NONE || 950 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 951 return 0; 952 953 blk_flush_plug(current->plug, false); 954 955 if (blk_queue_enter(q, BLK_MQ_REQ_NOWAIT)) 956 return 0; 957 if (queue_is_mq(q)) { 958 ret = blk_mq_poll(q, cookie, iob, flags); 959 } else { 960 struct gendisk *disk = q->disk; 961 962 if (disk && disk->fops->poll_bio) 963 ret = disk->fops->poll_bio(bio, iob, flags); 964 } 965 blk_queue_exit(q); 966 return ret; 967 } 968 EXPORT_SYMBOL_GPL(bio_poll); 969 970 /* 971 * Helper to implement file_operations.iopoll. Requires the bio to be stored 972 * in iocb->private, and cleared before freeing the bio. 973 */ 974 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, 975 unsigned int flags) 976 { 977 struct bio *bio; 978 int ret = 0; 979 980 /* 981 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can 982 * point to a freshly allocated bio at this point. If that happens 983 * we have a few cases to consider: 984 * 985 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just 986 * simply nothing in this case 987 * 2) the bio points to a not poll enabled device. bio_poll will catch 988 * this and return 0 989 * 3) the bio points to a poll capable device, including but not 990 * limited to the one that the original bio pointed to. In this 991 * case we will call into the actual poll method and poll for I/O, 992 * even if we don't need to, but it won't cause harm either. 993 * 994 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev 995 * is still allocated. Because partitions hold a reference to the whole 996 * device bdev and thus disk, the disk is also still valid. Grabbing 997 * a reference to the queue in bio_poll() ensures the hctxs and requests 998 * are still valid as well. 999 */ 1000 rcu_read_lock(); 1001 bio = READ_ONCE(kiocb->private); 1002 if (bio && bio->bi_bdev) 1003 ret = bio_poll(bio, iob, flags); 1004 rcu_read_unlock(); 1005 1006 return ret; 1007 } 1008 EXPORT_SYMBOL_GPL(iocb_bio_iopoll); 1009 1010 void update_io_ticks(struct block_device *part, unsigned long now, bool end) 1011 { 1012 unsigned long stamp; 1013 again: 1014 stamp = READ_ONCE(part->bd_stamp); 1015 if (unlikely(time_after(now, stamp))) { 1016 if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp)) 1017 __part_stat_add(part, io_ticks, end ? now - stamp : 1); 1018 } 1019 if (part->bd_partno) { 1020 part = bdev_whole(part); 1021 goto again; 1022 } 1023 } 1024 1025 static unsigned long __part_start_io_acct(struct block_device *part, 1026 unsigned int sectors, unsigned int op, 1027 unsigned long start_time) 1028 { 1029 const int sgrp = op_stat_group(op); 1030 1031 part_stat_lock(); 1032 update_io_ticks(part, start_time, false); 1033 part_stat_inc(part, ios[sgrp]); 1034 part_stat_add(part, sectors[sgrp], sectors); 1035 part_stat_local_inc(part, in_flight[op_is_write(op)]); 1036 part_stat_unlock(); 1037 1038 return start_time; 1039 } 1040 1041 /** 1042 * bio_start_io_acct_time - start I/O accounting for bio based drivers 1043 * @bio: bio to start account for 1044 * @start_time: start time that should be passed back to bio_end_io_acct(). 1045 */ 1046 void bio_start_io_acct_time(struct bio *bio, unsigned long start_time) 1047 { 1048 __part_start_io_acct(bio->bi_bdev, bio_sectors(bio), 1049 bio_op(bio), start_time); 1050 } 1051 EXPORT_SYMBOL_GPL(bio_start_io_acct_time); 1052 1053 /** 1054 * bio_start_io_acct - start I/O accounting for bio based drivers 1055 * @bio: bio to start account for 1056 * 1057 * Returns the start time that should be passed back to bio_end_io_acct(). 1058 */ 1059 unsigned long bio_start_io_acct(struct bio *bio) 1060 { 1061 return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio), 1062 bio_op(bio), jiffies); 1063 } 1064 EXPORT_SYMBOL_GPL(bio_start_io_acct); 1065 1066 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors, 1067 unsigned int op) 1068 { 1069 return __part_start_io_acct(disk->part0, sectors, op, jiffies); 1070 } 1071 EXPORT_SYMBOL(disk_start_io_acct); 1072 1073 static void __part_end_io_acct(struct block_device *part, unsigned int op, 1074 unsigned long start_time) 1075 { 1076 const int sgrp = op_stat_group(op); 1077 unsigned long now = READ_ONCE(jiffies); 1078 unsigned long duration = now - start_time; 1079 1080 part_stat_lock(); 1081 update_io_ticks(part, now, true); 1082 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 1083 part_stat_local_dec(part, in_flight[op_is_write(op)]); 1084 part_stat_unlock(); 1085 } 1086 1087 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, 1088 struct block_device *orig_bdev) 1089 { 1090 __part_end_io_acct(orig_bdev, bio_op(bio), start_time); 1091 } 1092 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); 1093 1094 void disk_end_io_acct(struct gendisk *disk, unsigned int op, 1095 unsigned long start_time) 1096 { 1097 __part_end_io_acct(disk->part0, op, start_time); 1098 } 1099 EXPORT_SYMBOL(disk_end_io_acct); 1100 1101 /** 1102 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1103 * @q : the queue of the device being checked 1104 * 1105 * Description: 1106 * Check if underlying low-level drivers of a device are busy. 1107 * If the drivers want to export their busy state, they must set own 1108 * exporting function using blk_queue_lld_busy() first. 1109 * 1110 * Basically, this function is used only by request stacking drivers 1111 * to stop dispatching requests to underlying devices when underlying 1112 * devices are busy. This behavior helps more I/O merging on the queue 1113 * of the request stacking driver and prevents I/O throughput regression 1114 * on burst I/O load. 1115 * 1116 * Return: 1117 * 0 - Not busy (The request stacking driver should dispatch request) 1118 * 1 - Busy (The request stacking driver should stop dispatching request) 1119 */ 1120 int blk_lld_busy(struct request_queue *q) 1121 { 1122 if (queue_is_mq(q) && q->mq_ops->busy) 1123 return q->mq_ops->busy(q); 1124 1125 return 0; 1126 } 1127 EXPORT_SYMBOL_GPL(blk_lld_busy); 1128 1129 int kblockd_schedule_work(struct work_struct *work) 1130 { 1131 return queue_work(kblockd_workqueue, work); 1132 } 1133 EXPORT_SYMBOL(kblockd_schedule_work); 1134 1135 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1136 unsigned long delay) 1137 { 1138 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1139 } 1140 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1141 1142 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) 1143 { 1144 struct task_struct *tsk = current; 1145 1146 /* 1147 * If this is a nested plug, don't actually assign it. 1148 */ 1149 if (tsk->plug) 1150 return; 1151 1152 plug->mq_list = NULL; 1153 plug->cached_rq = NULL; 1154 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); 1155 plug->rq_count = 0; 1156 plug->multiple_queues = false; 1157 plug->has_elevator = false; 1158 plug->nowait = false; 1159 INIT_LIST_HEAD(&plug->cb_list); 1160 1161 /* 1162 * Store ordering should not be needed here, since a potential 1163 * preempt will imply a full memory barrier 1164 */ 1165 tsk->plug = plug; 1166 } 1167 1168 /** 1169 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1170 * @plug: The &struct blk_plug that needs to be initialized 1171 * 1172 * Description: 1173 * blk_start_plug() indicates to the block layer an intent by the caller 1174 * to submit multiple I/O requests in a batch. The block layer may use 1175 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1176 * is called. However, the block layer may choose to submit requests 1177 * before a call to blk_finish_plug() if the number of queued I/Os 1178 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1179 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1180 * the task schedules (see below). 1181 * 1182 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1183 * pending I/O should the task end up blocking between blk_start_plug() and 1184 * blk_finish_plug(). This is important from a performance perspective, but 1185 * also ensures that we don't deadlock. For instance, if the task is blocking 1186 * for a memory allocation, memory reclaim could end up wanting to free a 1187 * page belonging to that request that is currently residing in our private 1188 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1189 * this kind of deadlock. 1190 */ 1191 void blk_start_plug(struct blk_plug *plug) 1192 { 1193 blk_start_plug_nr_ios(plug, 1); 1194 } 1195 EXPORT_SYMBOL(blk_start_plug); 1196 1197 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1198 { 1199 LIST_HEAD(callbacks); 1200 1201 while (!list_empty(&plug->cb_list)) { 1202 list_splice_init(&plug->cb_list, &callbacks); 1203 1204 while (!list_empty(&callbacks)) { 1205 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1206 struct blk_plug_cb, 1207 list); 1208 list_del(&cb->list); 1209 cb->callback(cb, from_schedule); 1210 } 1211 } 1212 } 1213 1214 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1215 int size) 1216 { 1217 struct blk_plug *plug = current->plug; 1218 struct blk_plug_cb *cb; 1219 1220 if (!plug) 1221 return NULL; 1222 1223 list_for_each_entry(cb, &plug->cb_list, list) 1224 if (cb->callback == unplug && cb->data == data) 1225 return cb; 1226 1227 /* Not currently on the callback list */ 1228 BUG_ON(size < sizeof(*cb)); 1229 cb = kzalloc(size, GFP_ATOMIC); 1230 if (cb) { 1231 cb->data = data; 1232 cb->callback = unplug; 1233 list_add(&cb->list, &plug->cb_list); 1234 } 1235 return cb; 1236 } 1237 EXPORT_SYMBOL(blk_check_plugged); 1238 1239 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) 1240 { 1241 if (!list_empty(&plug->cb_list)) 1242 flush_plug_callbacks(plug, from_schedule); 1243 if (!rq_list_empty(plug->mq_list)) 1244 blk_mq_flush_plug_list(plug, from_schedule); 1245 /* 1246 * Unconditionally flush out cached requests, even if the unplug 1247 * event came from schedule. Since we know hold references to the 1248 * queue for cached requests, we don't want a blocked task holding 1249 * up a queue freeze/quiesce event. 1250 */ 1251 if (unlikely(!rq_list_empty(plug->cached_rq))) 1252 blk_mq_free_plug_rqs(plug); 1253 } 1254 1255 /** 1256 * blk_finish_plug - mark the end of a batch of submitted I/O 1257 * @plug: The &struct blk_plug passed to blk_start_plug() 1258 * 1259 * Description: 1260 * Indicate that a batch of I/O submissions is complete. This function 1261 * must be paired with an initial call to blk_start_plug(). The intent 1262 * is to allow the block layer to optimize I/O submission. See the 1263 * documentation for blk_start_plug() for more information. 1264 */ 1265 void blk_finish_plug(struct blk_plug *plug) 1266 { 1267 if (plug == current->plug) { 1268 __blk_flush_plug(plug, false); 1269 current->plug = NULL; 1270 } 1271 } 1272 EXPORT_SYMBOL(blk_finish_plug); 1273 1274 void blk_io_schedule(void) 1275 { 1276 /* Prevent hang_check timer from firing at us during very long I/O */ 1277 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1278 1279 if (timeout) 1280 io_schedule_timeout(timeout); 1281 else 1282 io_schedule(); 1283 } 1284 EXPORT_SYMBOL_GPL(blk_io_schedule); 1285 1286 int __init blk_dev_init(void) 1287 { 1288 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1289 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1290 sizeof_field(struct request, cmd_flags)); 1291 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1292 sizeof_field(struct bio, bi_opf)); 1293 BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu), 1294 __alignof__(struct request_queue)) != 1295 sizeof(struct request_queue)); 1296 1297 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1298 kblockd_workqueue = alloc_workqueue("kblockd", 1299 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 1300 if (!kblockd_workqueue) 1301 panic("Failed to create kblockd\n"); 1302 1303 blk_requestq_cachep = kmem_cache_create("request_queue", 1304 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 1305 1306 blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu", 1307 sizeof(struct request_queue) + 1308 sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL); 1309 1310 blk_debugfs_root = debugfs_create_dir("block", NULL); 1311 1312 return 0; 1313 } 1314