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