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 /* @q is and will stay empty, shutdown and put */ 326 blk_put_queue(q); 327 } 328 EXPORT_SYMBOL(blk_cleanup_queue); 329 330 /** 331 * blk_queue_enter() - try to increase q->q_usage_counter 332 * @q: request queue pointer 333 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM 334 */ 335 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) 336 { 337 const bool pm = flags & BLK_MQ_REQ_PM; 338 339 while (!blk_try_enter_queue(q, pm)) { 340 if (flags & BLK_MQ_REQ_NOWAIT) 341 return -EBUSY; 342 343 /* 344 * read pair of barrier in blk_freeze_queue_start(), we need to 345 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 346 * reading .mq_freeze_depth or queue dying flag, otherwise the 347 * following wait may never return if the two reads are 348 * reordered. 349 */ 350 smp_rmb(); 351 wait_event(q->mq_freeze_wq, 352 (!q->mq_freeze_depth && 353 blk_pm_resume_queue(pm, q)) || 354 blk_queue_dying(q)); 355 if (blk_queue_dying(q)) 356 return -ENODEV; 357 } 358 359 return 0; 360 } 361 362 int __bio_queue_enter(struct request_queue *q, struct bio *bio) 363 { 364 while (!blk_try_enter_queue(q, false)) { 365 struct gendisk *disk = bio->bi_bdev->bd_disk; 366 367 if (bio->bi_opf & REQ_NOWAIT) { 368 if (test_bit(GD_DEAD, &disk->state)) 369 goto dead; 370 bio_wouldblock_error(bio); 371 return -EBUSY; 372 } 373 374 /* 375 * read pair of barrier in blk_freeze_queue_start(), we need to 376 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 377 * reading .mq_freeze_depth or queue dying flag, otherwise the 378 * following wait may never return if the two reads are 379 * reordered. 380 */ 381 smp_rmb(); 382 wait_event(q->mq_freeze_wq, 383 (!q->mq_freeze_depth && 384 blk_pm_resume_queue(false, q)) || 385 test_bit(GD_DEAD, &disk->state)); 386 if (test_bit(GD_DEAD, &disk->state)) 387 goto dead; 388 } 389 390 return 0; 391 dead: 392 bio_io_error(bio); 393 return -ENODEV; 394 } 395 396 void blk_queue_exit(struct request_queue *q) 397 { 398 percpu_ref_put(&q->q_usage_counter); 399 } 400 401 static void blk_queue_usage_counter_release(struct percpu_ref *ref) 402 { 403 struct request_queue *q = 404 container_of(ref, struct request_queue, q_usage_counter); 405 406 wake_up_all(&q->mq_freeze_wq); 407 } 408 409 static void blk_rq_timed_out_timer(struct timer_list *t) 410 { 411 struct request_queue *q = from_timer(q, t, timeout); 412 413 kblockd_schedule_work(&q->timeout_work); 414 } 415 416 static void blk_timeout_work(struct work_struct *work) 417 { 418 } 419 420 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu) 421 { 422 struct request_queue *q; 423 int ret; 424 425 q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu), 426 GFP_KERNEL | __GFP_ZERO, node_id); 427 if (!q) 428 return NULL; 429 430 if (alloc_srcu) { 431 blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q); 432 if (init_srcu_struct(q->srcu) != 0) 433 goto fail_q; 434 } 435 436 q->last_merge = NULL; 437 438 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); 439 if (q->id < 0) 440 goto fail_srcu; 441 442 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0); 443 if (ret) 444 goto fail_id; 445 446 q->stats = blk_alloc_queue_stats(); 447 if (!q->stats) 448 goto fail_split; 449 450 q->node = node_id; 451 452 atomic_set(&q->nr_active_requests_shared_tags, 0); 453 454 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); 455 INIT_WORK(&q->timeout_work, blk_timeout_work); 456 INIT_LIST_HEAD(&q->icq_list); 457 458 kobject_init(&q->kobj, &blk_queue_ktype); 459 460 mutex_init(&q->debugfs_mutex); 461 mutex_init(&q->sysfs_lock); 462 mutex_init(&q->sysfs_dir_lock); 463 spin_lock_init(&q->queue_lock); 464 465 init_waitqueue_head(&q->mq_freeze_wq); 466 mutex_init(&q->mq_freeze_lock); 467 468 /* 469 * Init percpu_ref in atomic mode so that it's faster to shutdown. 470 * See blk_register_queue() for details. 471 */ 472 if (percpu_ref_init(&q->q_usage_counter, 473 blk_queue_usage_counter_release, 474 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) 475 goto fail_stats; 476 477 blk_queue_dma_alignment(q, 511); 478 blk_set_default_limits(&q->limits); 479 q->nr_requests = BLKDEV_DEFAULT_RQ; 480 481 return q; 482 483 fail_stats: 484 blk_free_queue_stats(q->stats); 485 fail_split: 486 bioset_exit(&q->bio_split); 487 fail_id: 488 ida_simple_remove(&blk_queue_ida, q->id); 489 fail_srcu: 490 if (alloc_srcu) 491 cleanup_srcu_struct(q->srcu); 492 fail_q: 493 kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q); 494 return NULL; 495 } 496 497 /** 498 * blk_get_queue - increment the request_queue refcount 499 * @q: the request_queue structure to increment the refcount for 500 * 501 * Increment the refcount of the request_queue kobject. 502 * 503 * Context: Any context. 504 */ 505 bool blk_get_queue(struct request_queue *q) 506 { 507 if (likely(!blk_queue_dying(q))) { 508 __blk_get_queue(q); 509 return true; 510 } 511 512 return false; 513 } 514 EXPORT_SYMBOL(blk_get_queue); 515 516 #ifdef CONFIG_FAIL_MAKE_REQUEST 517 518 static DECLARE_FAULT_ATTR(fail_make_request); 519 520 static int __init setup_fail_make_request(char *str) 521 { 522 return setup_fault_attr(&fail_make_request, str); 523 } 524 __setup("fail_make_request=", setup_fail_make_request); 525 526 bool should_fail_request(struct block_device *part, unsigned int bytes) 527 { 528 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); 529 } 530 531 static int __init fail_make_request_debugfs(void) 532 { 533 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 534 NULL, &fail_make_request); 535 536 return PTR_ERR_OR_ZERO(dir); 537 } 538 539 late_initcall(fail_make_request_debugfs); 540 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 541 542 static inline bool bio_check_ro(struct bio *bio) 543 { 544 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { 545 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) 546 return false; 547 pr_warn("Trying to write to read-only block-device %pg\n", 548 bio->bi_bdev); 549 /* Older lvm-tools actually trigger this */ 550 return false; 551 } 552 553 return false; 554 } 555 556 static noinline int should_fail_bio(struct bio *bio) 557 { 558 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) 559 return -EIO; 560 return 0; 561 } 562 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); 563 564 /* 565 * Check whether this bio extends beyond the end of the device or partition. 566 * This may well happen - the kernel calls bread() without checking the size of 567 * the device, e.g., when mounting a file system. 568 */ 569 static inline int bio_check_eod(struct bio *bio) 570 { 571 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 572 unsigned int nr_sectors = bio_sectors(bio); 573 574 if (nr_sectors && maxsector && 575 (nr_sectors > maxsector || 576 bio->bi_iter.bi_sector > maxsector - nr_sectors)) { 577 pr_info_ratelimited("%s: attempt to access beyond end of device\n" 578 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", 579 current->comm, bio->bi_bdev, bio->bi_opf, 580 bio->bi_iter.bi_sector, nr_sectors, maxsector); 581 return -EIO; 582 } 583 return 0; 584 } 585 586 /* 587 * Remap block n of partition p to block n+start(p) of the disk. 588 */ 589 static int blk_partition_remap(struct bio *bio) 590 { 591 struct block_device *p = bio->bi_bdev; 592 593 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 594 return -EIO; 595 if (bio_sectors(bio)) { 596 bio->bi_iter.bi_sector += p->bd_start_sect; 597 trace_block_bio_remap(bio, p->bd_dev, 598 bio->bi_iter.bi_sector - 599 p->bd_start_sect); 600 } 601 bio_set_flag(bio, BIO_REMAPPED); 602 return 0; 603 } 604 605 /* 606 * Check write append to a zoned block device. 607 */ 608 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 609 struct bio *bio) 610 { 611 sector_t pos = bio->bi_iter.bi_sector; 612 int nr_sectors = bio_sectors(bio); 613 614 /* Only applicable to zoned block devices */ 615 if (!blk_queue_is_zoned(q)) 616 return BLK_STS_NOTSUPP; 617 618 /* The bio sector must point to the start of a sequential zone */ 619 if (pos & (blk_queue_zone_sectors(q) - 1) || 620 !blk_queue_zone_is_seq(q, pos)) 621 return BLK_STS_IOERR; 622 623 /* 624 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 625 * split and could result in non-contiguous sectors being written in 626 * different zones. 627 */ 628 if (nr_sectors > q->limits.chunk_sectors) 629 return BLK_STS_IOERR; 630 631 /* Make sure the BIO is small enough and will not get split */ 632 if (nr_sectors > q->limits.max_zone_append_sectors) 633 return BLK_STS_IOERR; 634 635 bio->bi_opf |= REQ_NOMERGE; 636 637 return BLK_STS_OK; 638 } 639 640 static void __submit_bio(struct bio *bio) 641 { 642 struct gendisk *disk = bio->bi_bdev->bd_disk; 643 644 if (unlikely(!blk_crypto_bio_prep(&bio))) 645 return; 646 647 if (!disk->fops->submit_bio) { 648 blk_mq_submit_bio(bio); 649 } else if (likely(bio_queue_enter(bio) == 0)) { 650 disk->fops->submit_bio(bio); 651 blk_queue_exit(disk->queue); 652 } 653 } 654 655 /* 656 * The loop in this function may be a bit non-obvious, and so deserves some 657 * explanation: 658 * 659 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 660 * that), so we have a list with a single bio. 661 * - We pretend that we have just taken it off a longer list, so we assign 662 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 663 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 664 * bios through a recursive call to submit_bio_noacct. If it did, we find a 665 * non-NULL value in bio_list and re-enter the loop from the top. 666 * - In this case we really did just take the bio of the top of the list (no 667 * pretending) and so remove it from bio_list, and call into ->submit_bio() 668 * again. 669 * 670 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 671 * bio_list_on_stack[1] contains bios that were submitted before the current 672 * ->submit_bio, but that haven't been processed yet. 673 */ 674 static void __submit_bio_noacct(struct bio *bio) 675 { 676 struct bio_list bio_list_on_stack[2]; 677 678 BUG_ON(bio->bi_next); 679 680 bio_list_init(&bio_list_on_stack[0]); 681 current->bio_list = bio_list_on_stack; 682 683 do { 684 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 685 struct bio_list lower, same; 686 687 /* 688 * Create a fresh bio_list for all subordinate requests. 689 */ 690 bio_list_on_stack[1] = bio_list_on_stack[0]; 691 bio_list_init(&bio_list_on_stack[0]); 692 693 __submit_bio(bio); 694 695 /* 696 * Sort new bios into those for a lower level and those for the 697 * same level. 698 */ 699 bio_list_init(&lower); 700 bio_list_init(&same); 701 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 702 if (q == bdev_get_queue(bio->bi_bdev)) 703 bio_list_add(&same, bio); 704 else 705 bio_list_add(&lower, bio); 706 707 /* 708 * Now assemble so we handle the lowest level first. 709 */ 710 bio_list_merge(&bio_list_on_stack[0], &lower); 711 bio_list_merge(&bio_list_on_stack[0], &same); 712 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 713 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 714 715 current->bio_list = NULL; 716 } 717 718 static void __submit_bio_noacct_mq(struct bio *bio) 719 { 720 struct bio_list bio_list[2] = { }; 721 722 current->bio_list = bio_list; 723 724 do { 725 __submit_bio(bio); 726 } while ((bio = bio_list_pop(&bio_list[0]))); 727 728 current->bio_list = NULL; 729 } 730 731 void submit_bio_noacct_nocheck(struct bio *bio) 732 { 733 /* 734 * We only want one ->submit_bio to be active at a time, else stack 735 * usage with stacked devices could be a problem. Use current->bio_list 736 * to collect a list of requests submited by a ->submit_bio method while 737 * it is active, and then process them after it returned. 738 */ 739 if (current->bio_list) 740 bio_list_add(¤t->bio_list[0], bio); 741 else if (!bio->bi_bdev->bd_disk->fops->submit_bio) 742 __submit_bio_noacct_mq(bio); 743 else 744 __submit_bio_noacct(bio); 745 } 746 747 /** 748 * submit_bio_noacct - re-submit a bio to the block device layer for I/O 749 * @bio: The bio describing the location in memory and on the device. 750 * 751 * This is a version of submit_bio() that shall only be used for I/O that is 752 * resubmitted to lower level drivers by stacking block drivers. All file 753 * systems and other upper level users of the block layer should use 754 * submit_bio() instead. 755 */ 756 void submit_bio_noacct(struct bio *bio) 757 { 758 struct block_device *bdev = bio->bi_bdev; 759 struct request_queue *q = bdev_get_queue(bdev); 760 blk_status_t status = BLK_STS_IOERR; 761 struct blk_plug *plug; 762 763 might_sleep(); 764 765 plug = blk_mq_plug(q, bio); 766 if (plug && plug->nowait) 767 bio->bi_opf |= REQ_NOWAIT; 768 769 /* 770 * For a REQ_NOWAIT based request, return -EOPNOTSUPP 771 * if queue does not support NOWAIT. 772 */ 773 if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q)) 774 goto not_supported; 775 776 if (should_fail_bio(bio)) 777 goto end_io; 778 if (unlikely(bio_check_ro(bio))) 779 goto end_io; 780 if (!bio_flagged(bio, BIO_REMAPPED)) { 781 if (unlikely(bio_check_eod(bio))) 782 goto end_io; 783 if (bdev->bd_partno && unlikely(blk_partition_remap(bio))) 784 goto end_io; 785 } 786 787 /* 788 * Filter flush bio's early so that bio based drivers without flush 789 * support don't have to worry about them. 790 */ 791 if (op_is_flush(bio->bi_opf) && 792 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { 793 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 794 if (!bio_sectors(bio)) { 795 status = BLK_STS_OK; 796 goto end_io; 797 } 798 } 799 800 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 801 bio_clear_polled(bio); 802 803 switch (bio_op(bio)) { 804 case REQ_OP_DISCARD: 805 if (!bdev_max_discard_sectors(bdev)) 806 goto not_supported; 807 break; 808 case REQ_OP_SECURE_ERASE: 809 if (!bdev_max_secure_erase_sectors(bdev)) 810 goto not_supported; 811 break; 812 case REQ_OP_ZONE_APPEND: 813 status = blk_check_zone_append(q, bio); 814 if (status != BLK_STS_OK) 815 goto end_io; 816 break; 817 case REQ_OP_ZONE_RESET: 818 case REQ_OP_ZONE_OPEN: 819 case REQ_OP_ZONE_CLOSE: 820 case REQ_OP_ZONE_FINISH: 821 if (!blk_queue_is_zoned(q)) 822 goto not_supported; 823 break; 824 case REQ_OP_ZONE_RESET_ALL: 825 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q)) 826 goto not_supported; 827 break; 828 case REQ_OP_WRITE_ZEROES: 829 if (!q->limits.max_write_zeroes_sectors) 830 goto not_supported; 831 break; 832 default: 833 break; 834 } 835 836 if (blk_throtl_bio(bio)) 837 return; 838 839 blk_cgroup_bio_start(bio); 840 blkcg_bio_issue_init(bio); 841 842 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 843 trace_block_bio_queue(bio); 844 /* Now that enqueuing has been traced, we need to trace 845 * completion as well. 846 */ 847 bio_set_flag(bio, BIO_TRACE_COMPLETION); 848 } 849 submit_bio_noacct_nocheck(bio); 850 return; 851 852 not_supported: 853 status = BLK_STS_NOTSUPP; 854 end_io: 855 bio->bi_status = status; 856 bio_endio(bio); 857 } 858 EXPORT_SYMBOL(submit_bio_noacct); 859 860 /** 861 * submit_bio - submit a bio to the block device layer for I/O 862 * @bio: The &struct bio which describes the I/O 863 * 864 * submit_bio() is used to submit I/O requests to block devices. It is passed a 865 * fully set up &struct bio that describes the I/O that needs to be done. The 866 * bio will be send to the device described by the bi_bdev field. 867 * 868 * The success/failure status of the request, along with notification of 869 * completion, is delivered asynchronously through the ->bi_end_io() callback 870 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has 871 * been called. 872 */ 873 void submit_bio(struct bio *bio) 874 { 875 if (blkcg_punt_bio_submit(bio)) 876 return; 877 878 if (bio_op(bio) == REQ_OP_READ) { 879 task_io_account_read(bio->bi_iter.bi_size); 880 count_vm_events(PGPGIN, bio_sectors(bio)); 881 } else if (bio_op(bio) == REQ_OP_WRITE) { 882 count_vm_events(PGPGOUT, bio_sectors(bio)); 883 } 884 885 /* 886 * If we're reading data that is part of the userspace workingset, count 887 * submission time as memory stall. When the device is congested, or 888 * the submitting cgroup IO-throttled, submission can be a significant 889 * part of overall IO time. 890 */ 891 if (unlikely(bio_op(bio) == REQ_OP_READ && 892 bio_flagged(bio, BIO_WORKINGSET))) { 893 unsigned long pflags; 894 895 psi_memstall_enter(&pflags); 896 submit_bio_noacct(bio); 897 psi_memstall_leave(&pflags); 898 return; 899 } 900 901 submit_bio_noacct(bio); 902 } 903 EXPORT_SYMBOL(submit_bio); 904 905 /** 906 * bio_poll - poll for BIO completions 907 * @bio: bio to poll for 908 * @iob: batches of IO 909 * @flags: BLK_POLL_* flags that control the behavior 910 * 911 * Poll for completions on queue associated with the bio. Returns number of 912 * completed entries found. 913 * 914 * Note: the caller must either be the context that submitted @bio, or 915 * be in a RCU critical section to prevent freeing of @bio. 916 */ 917 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) 918 { 919 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 920 blk_qc_t cookie = READ_ONCE(bio->bi_cookie); 921 int ret = 0; 922 923 if (cookie == BLK_QC_T_NONE || 924 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 925 return 0; 926 927 blk_flush_plug(current->plug, false); 928 929 if (bio_queue_enter(bio)) 930 return 0; 931 if (queue_is_mq(q)) { 932 ret = blk_mq_poll(q, cookie, iob, flags); 933 } else { 934 struct gendisk *disk = q->disk; 935 936 if (disk && disk->fops->poll_bio) 937 ret = disk->fops->poll_bio(bio, iob, flags); 938 } 939 blk_queue_exit(q); 940 return ret; 941 } 942 EXPORT_SYMBOL_GPL(bio_poll); 943 944 /* 945 * Helper to implement file_operations.iopoll. Requires the bio to be stored 946 * in iocb->private, and cleared before freeing the bio. 947 */ 948 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, 949 unsigned int flags) 950 { 951 struct bio *bio; 952 int ret = 0; 953 954 /* 955 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can 956 * point to a freshly allocated bio at this point. If that happens 957 * we have a few cases to consider: 958 * 959 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just 960 * simply nothing in this case 961 * 2) the bio points to a not poll enabled device. bio_poll will catch 962 * this and return 0 963 * 3) the bio points to a poll capable device, including but not 964 * limited to the one that the original bio pointed to. In this 965 * case we will call into the actual poll method and poll for I/O, 966 * even if we don't need to, but it won't cause harm either. 967 * 968 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev 969 * is still allocated. Because partitions hold a reference to the whole 970 * device bdev and thus disk, the disk is also still valid. Grabbing 971 * a reference to the queue in bio_poll() ensures the hctxs and requests 972 * are still valid as well. 973 */ 974 rcu_read_lock(); 975 bio = READ_ONCE(kiocb->private); 976 if (bio && bio->bi_bdev) 977 ret = bio_poll(bio, iob, flags); 978 rcu_read_unlock(); 979 980 return ret; 981 } 982 EXPORT_SYMBOL_GPL(iocb_bio_iopoll); 983 984 void update_io_ticks(struct block_device *part, unsigned long now, bool end) 985 { 986 unsigned long stamp; 987 again: 988 stamp = READ_ONCE(part->bd_stamp); 989 if (unlikely(time_after(now, stamp))) { 990 if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp)) 991 __part_stat_add(part, io_ticks, end ? now - stamp : 1); 992 } 993 if (part->bd_partno) { 994 part = bdev_whole(part); 995 goto again; 996 } 997 } 998 999 unsigned long bdev_start_io_acct(struct block_device *bdev, 1000 unsigned int sectors, unsigned int op, 1001 unsigned long start_time) 1002 { 1003 const int sgrp = op_stat_group(op); 1004 1005 part_stat_lock(); 1006 update_io_ticks(bdev, start_time, false); 1007 part_stat_inc(bdev, ios[sgrp]); 1008 part_stat_add(bdev, sectors[sgrp], sectors); 1009 part_stat_local_inc(bdev, in_flight[op_is_write(op)]); 1010 part_stat_unlock(); 1011 1012 return start_time; 1013 } 1014 EXPORT_SYMBOL(bdev_start_io_acct); 1015 1016 /** 1017 * bio_start_io_acct_time - start I/O accounting for bio based drivers 1018 * @bio: bio to start account for 1019 * @start_time: start time that should be passed back to bio_end_io_acct(). 1020 */ 1021 void bio_start_io_acct_time(struct bio *bio, unsigned long start_time) 1022 { 1023 bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio), 1024 bio_op(bio), start_time); 1025 } 1026 EXPORT_SYMBOL_GPL(bio_start_io_acct_time); 1027 1028 /** 1029 * bio_start_io_acct - start I/O accounting for bio based drivers 1030 * @bio: bio to start account for 1031 * 1032 * Returns the start time that should be passed back to bio_end_io_acct(). 1033 */ 1034 unsigned long bio_start_io_acct(struct bio *bio) 1035 { 1036 return bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio), 1037 bio_op(bio), jiffies); 1038 } 1039 EXPORT_SYMBOL_GPL(bio_start_io_acct); 1040 1041 void bdev_end_io_acct(struct block_device *bdev, unsigned int op, 1042 unsigned long start_time) 1043 { 1044 const int sgrp = op_stat_group(op); 1045 unsigned long now = READ_ONCE(jiffies); 1046 unsigned long duration = now - start_time; 1047 1048 part_stat_lock(); 1049 update_io_ticks(bdev, now, true); 1050 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); 1051 part_stat_local_dec(bdev, in_flight[op_is_write(op)]); 1052 part_stat_unlock(); 1053 } 1054 EXPORT_SYMBOL(bdev_end_io_acct); 1055 1056 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, 1057 struct block_device *orig_bdev) 1058 { 1059 bdev_end_io_acct(orig_bdev, bio_op(bio), start_time); 1060 } 1061 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); 1062 1063 /** 1064 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1065 * @q : the queue of the device being checked 1066 * 1067 * Description: 1068 * Check if underlying low-level drivers of a device are busy. 1069 * If the drivers want to export their busy state, they must set own 1070 * exporting function using blk_queue_lld_busy() first. 1071 * 1072 * Basically, this function is used only by request stacking drivers 1073 * to stop dispatching requests to underlying devices when underlying 1074 * devices are busy. This behavior helps more I/O merging on the queue 1075 * of the request stacking driver and prevents I/O throughput regression 1076 * on burst I/O load. 1077 * 1078 * Return: 1079 * 0 - Not busy (The request stacking driver should dispatch request) 1080 * 1 - Busy (The request stacking driver should stop dispatching request) 1081 */ 1082 int blk_lld_busy(struct request_queue *q) 1083 { 1084 if (queue_is_mq(q) && q->mq_ops->busy) 1085 return q->mq_ops->busy(q); 1086 1087 return 0; 1088 } 1089 EXPORT_SYMBOL_GPL(blk_lld_busy); 1090 1091 int kblockd_schedule_work(struct work_struct *work) 1092 { 1093 return queue_work(kblockd_workqueue, work); 1094 } 1095 EXPORT_SYMBOL(kblockd_schedule_work); 1096 1097 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1098 unsigned long delay) 1099 { 1100 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1101 } 1102 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1103 1104 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) 1105 { 1106 struct task_struct *tsk = current; 1107 1108 /* 1109 * If this is a nested plug, don't actually assign it. 1110 */ 1111 if (tsk->plug) 1112 return; 1113 1114 plug->mq_list = NULL; 1115 plug->cached_rq = NULL; 1116 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); 1117 plug->rq_count = 0; 1118 plug->multiple_queues = false; 1119 plug->has_elevator = false; 1120 plug->nowait = false; 1121 INIT_LIST_HEAD(&plug->cb_list); 1122 1123 /* 1124 * Store ordering should not be needed here, since a potential 1125 * preempt will imply a full memory barrier 1126 */ 1127 tsk->plug = plug; 1128 } 1129 1130 /** 1131 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1132 * @plug: The &struct blk_plug that needs to be initialized 1133 * 1134 * Description: 1135 * blk_start_plug() indicates to the block layer an intent by the caller 1136 * to submit multiple I/O requests in a batch. The block layer may use 1137 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1138 * is called. However, the block layer may choose to submit requests 1139 * before a call to blk_finish_plug() if the number of queued I/Os 1140 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1141 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1142 * the task schedules (see below). 1143 * 1144 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1145 * pending I/O should the task end up blocking between blk_start_plug() and 1146 * blk_finish_plug(). This is important from a performance perspective, but 1147 * also ensures that we don't deadlock. For instance, if the task is blocking 1148 * for a memory allocation, memory reclaim could end up wanting to free a 1149 * page belonging to that request that is currently residing in our private 1150 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1151 * this kind of deadlock. 1152 */ 1153 void blk_start_plug(struct blk_plug *plug) 1154 { 1155 blk_start_plug_nr_ios(plug, 1); 1156 } 1157 EXPORT_SYMBOL(blk_start_plug); 1158 1159 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1160 { 1161 LIST_HEAD(callbacks); 1162 1163 while (!list_empty(&plug->cb_list)) { 1164 list_splice_init(&plug->cb_list, &callbacks); 1165 1166 while (!list_empty(&callbacks)) { 1167 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1168 struct blk_plug_cb, 1169 list); 1170 list_del(&cb->list); 1171 cb->callback(cb, from_schedule); 1172 } 1173 } 1174 } 1175 1176 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1177 int size) 1178 { 1179 struct blk_plug *plug = current->plug; 1180 struct blk_plug_cb *cb; 1181 1182 if (!plug) 1183 return NULL; 1184 1185 list_for_each_entry(cb, &plug->cb_list, list) 1186 if (cb->callback == unplug && cb->data == data) 1187 return cb; 1188 1189 /* Not currently on the callback list */ 1190 BUG_ON(size < sizeof(*cb)); 1191 cb = kzalloc(size, GFP_ATOMIC); 1192 if (cb) { 1193 cb->data = data; 1194 cb->callback = unplug; 1195 list_add(&cb->list, &plug->cb_list); 1196 } 1197 return cb; 1198 } 1199 EXPORT_SYMBOL(blk_check_plugged); 1200 1201 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) 1202 { 1203 if (!list_empty(&plug->cb_list)) 1204 flush_plug_callbacks(plug, from_schedule); 1205 if (!rq_list_empty(plug->mq_list)) 1206 blk_mq_flush_plug_list(plug, from_schedule); 1207 /* 1208 * Unconditionally flush out cached requests, even if the unplug 1209 * event came from schedule. Since we know hold references to the 1210 * queue for cached requests, we don't want a blocked task holding 1211 * up a queue freeze/quiesce event. 1212 */ 1213 if (unlikely(!rq_list_empty(plug->cached_rq))) 1214 blk_mq_free_plug_rqs(plug); 1215 } 1216 1217 /** 1218 * blk_finish_plug - mark the end of a batch of submitted I/O 1219 * @plug: The &struct blk_plug passed to blk_start_plug() 1220 * 1221 * Description: 1222 * Indicate that a batch of I/O submissions is complete. This function 1223 * must be paired with an initial call to blk_start_plug(). The intent 1224 * is to allow the block layer to optimize I/O submission. See the 1225 * documentation for blk_start_plug() for more information. 1226 */ 1227 void blk_finish_plug(struct blk_plug *plug) 1228 { 1229 if (plug == current->plug) { 1230 __blk_flush_plug(plug, false); 1231 current->plug = NULL; 1232 } 1233 } 1234 EXPORT_SYMBOL(blk_finish_plug); 1235 1236 void blk_io_schedule(void) 1237 { 1238 /* Prevent hang_check timer from firing at us during very long I/O */ 1239 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1240 1241 if (timeout) 1242 io_schedule_timeout(timeout); 1243 else 1244 io_schedule(); 1245 } 1246 EXPORT_SYMBOL_GPL(blk_io_schedule); 1247 1248 int __init blk_dev_init(void) 1249 { 1250 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1251 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1252 sizeof_field(struct request, cmd_flags)); 1253 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1254 sizeof_field(struct bio, bi_opf)); 1255 BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu), 1256 __alignof__(struct request_queue)) != 1257 sizeof(struct request_queue)); 1258 1259 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1260 kblockd_workqueue = alloc_workqueue("kblockd", 1261 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 1262 if (!kblockd_workqueue) 1263 panic("Failed to create kblockd\n"); 1264 1265 blk_requestq_cachep = kmem_cache_create("request_queue", 1266 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 1267 1268 blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu", 1269 sizeof(struct request_queue) + 1270 sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL); 1271 1272 blk_debugfs_root = debugfs_create_dir("block", NULL); 1273 1274 return 0; 1275 } 1276