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