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 the string "name" for an operation REQ_OP_name. 118 * @op: a request operation. 119 * 120 * Convert a request operation REQ_OP_name into the string "name". Useful for 121 * debugging and tracing BIOs and requests. For an invalid request operation 122 * code, the string "UNKNOWN" is returned. 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 } 200 201 /** 202 * blk_sync_queue - cancel any pending callbacks on a queue 203 * @q: the queue 204 * 205 * Description: 206 * The block layer may perform asynchronous callback activity 207 * on a queue, such as calling the unplug function after a timeout. 208 * A block device may call blk_sync_queue to ensure that any 209 * such activity is cancelled, thus allowing it to release resources 210 * that the callbacks might use. The caller must already have made sure 211 * that its ->submit_bio will not re-add plugging prior to calling 212 * this function. 213 * 214 * This function does not cancel any asynchronous activity arising 215 * out of elevator or throttling code. That would require elevator_exit() 216 * and blkcg_exit_queue() to be called with queue lock initialized. 217 * 218 */ 219 void blk_sync_queue(struct request_queue *q) 220 { 221 timer_delete_sync(&q->timeout); 222 cancel_work_sync(&q->timeout_work); 223 } 224 EXPORT_SYMBOL(blk_sync_queue); 225 226 /** 227 * blk_set_pm_only - increment pm_only counter 228 * @q: request queue pointer 229 */ 230 void blk_set_pm_only(struct request_queue *q) 231 { 232 atomic_inc(&q->pm_only); 233 } 234 EXPORT_SYMBOL_GPL(blk_set_pm_only); 235 236 void blk_clear_pm_only(struct request_queue *q) 237 { 238 int pm_only; 239 240 pm_only = atomic_dec_return(&q->pm_only); 241 WARN_ON_ONCE(pm_only < 0); 242 if (pm_only == 0) 243 wake_up_all(&q->mq_freeze_wq); 244 } 245 EXPORT_SYMBOL_GPL(blk_clear_pm_only); 246 247 static void blk_free_queue_rcu(struct rcu_head *rcu_head) 248 { 249 struct request_queue *q = container_of(rcu_head, 250 struct request_queue, rcu_head); 251 252 percpu_ref_exit(&q->q_usage_counter); 253 kmem_cache_free(blk_requestq_cachep, q); 254 } 255 256 static void blk_free_queue(struct request_queue *q) 257 { 258 blk_free_queue_stats(q->stats); 259 if (queue_is_mq(q)) 260 blk_mq_release(q); 261 262 ida_free(&blk_queue_ida, q->id); 263 lockdep_unregister_key(&q->io_lock_cls_key); 264 lockdep_unregister_key(&q->q_lock_cls_key); 265 call_rcu(&q->rcu_head, blk_free_queue_rcu); 266 } 267 268 /** 269 * blk_put_queue - decrement the request_queue refcount 270 * @q: the request_queue structure to decrement the refcount for 271 * 272 * Decrements the refcount of the request_queue and free it when the refcount 273 * reaches 0. 274 */ 275 void blk_put_queue(struct request_queue *q) 276 { 277 if (refcount_dec_and_test(&q->refs)) 278 blk_free_queue(q); 279 } 280 EXPORT_SYMBOL(blk_put_queue); 281 282 bool blk_queue_start_drain(struct request_queue *q) 283 { 284 /* 285 * When queue DYING flag is set, we need to block new req 286 * entering queue, so we call blk_freeze_queue_start() to 287 * prevent I/O from crossing blk_queue_enter(). 288 */ 289 bool freeze = __blk_freeze_queue_start(q, current); 290 if (queue_is_mq(q)) 291 blk_mq_wake_waiters(q); 292 /* Make blk_queue_enter() reexamine the DYING flag. */ 293 wake_up_all(&q->mq_freeze_wq); 294 295 return freeze; 296 } 297 298 /** 299 * blk_queue_enter() - try to increase q->q_usage_counter 300 * @q: request queue pointer 301 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM 302 */ 303 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) 304 { 305 const bool pm = flags & BLK_MQ_REQ_PM; 306 307 while (!blk_try_enter_queue(q, pm)) { 308 if (flags & BLK_MQ_REQ_NOWAIT) 309 return -EAGAIN; 310 311 /* 312 * read pair of barrier in blk_freeze_queue_start(), we need to 313 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 314 * reading .mq_freeze_depth or queue dying flag, otherwise the 315 * following wait may never return if the two reads are 316 * reordered. 317 */ 318 smp_rmb(); 319 wait_event(q->mq_freeze_wq, 320 (!q->mq_freeze_depth && 321 blk_pm_resume_queue(pm, q)) || 322 blk_queue_dying(q)); 323 if (blk_queue_dying(q)) 324 return -ENODEV; 325 } 326 327 rwsem_acquire_read(&q->q_lockdep_map, 0, 0, _RET_IP_); 328 rwsem_release(&q->q_lockdep_map, _RET_IP_); 329 return 0; 330 } 331 332 int __bio_queue_enter(struct request_queue *q, struct bio *bio) 333 { 334 while (!blk_try_enter_queue(q, false)) { 335 struct gendisk *disk = bio->bi_bdev->bd_disk; 336 337 if (bio->bi_opf & REQ_NOWAIT) { 338 if (test_bit(GD_DEAD, &disk->state)) 339 goto dead; 340 bio_wouldblock_error(bio); 341 return -EAGAIN; 342 } 343 344 /* 345 * read pair of barrier in blk_freeze_queue_start(), we need to 346 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 347 * reading .mq_freeze_depth or queue dying flag, otherwise the 348 * following wait may never return if the two reads are 349 * reordered. 350 */ 351 smp_rmb(); 352 wait_event(q->mq_freeze_wq, 353 (!q->mq_freeze_depth && 354 blk_pm_resume_queue(false, q)) || 355 test_bit(GD_DEAD, &disk->state)); 356 if (test_bit(GD_DEAD, &disk->state)) 357 goto dead; 358 } 359 360 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_); 361 rwsem_release(&q->io_lockdep_map, _RET_IP_); 362 return 0; 363 dead: 364 bio_io_error(bio); 365 return -ENODEV; 366 } 367 368 void blk_queue_exit(struct request_queue *q) 369 { 370 percpu_ref_put(&q->q_usage_counter); 371 } 372 373 static void blk_queue_usage_counter_release(struct percpu_ref *ref) 374 { 375 struct request_queue *q = 376 container_of(ref, struct request_queue, q_usage_counter); 377 378 wake_up_all(&q->mq_freeze_wq); 379 } 380 381 static void blk_rq_timed_out_timer(struct timer_list *t) 382 { 383 struct request_queue *q = timer_container_of(q, t, timeout); 384 385 kblockd_schedule_work(&q->timeout_work); 386 } 387 388 static void blk_timeout_work(struct work_struct *work) 389 { 390 } 391 392 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id) 393 { 394 struct request_queue *q; 395 int error; 396 397 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO, 398 node_id); 399 if (!q) 400 return ERR_PTR(-ENOMEM); 401 402 q->last_merge = NULL; 403 404 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL); 405 if (q->id < 0) { 406 error = q->id; 407 goto fail_q; 408 } 409 410 q->stats = blk_alloc_queue_stats(); 411 if (!q->stats) { 412 error = -ENOMEM; 413 goto fail_id; 414 } 415 416 error = blk_set_default_limits(lim); 417 if (error) 418 goto fail_stats; 419 q->limits = *lim; 420 421 q->node = node_id; 422 423 atomic_set(&q->nr_active_requests_shared_tags, 0); 424 425 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); 426 INIT_WORK(&q->timeout_work, blk_timeout_work); 427 INIT_LIST_HEAD(&q->icq_list); 428 429 refcount_set(&q->refs, 1); 430 mutex_init(&q->debugfs_mutex); 431 mutex_init(&q->elevator_lock); 432 mutex_init(&q->sysfs_lock); 433 mutex_init(&q->limits_lock); 434 mutex_init(&q->rq_qos_mutex); 435 spin_lock_init(&q->queue_lock); 436 437 init_waitqueue_head(&q->mq_freeze_wq); 438 mutex_init(&q->mq_freeze_lock); 439 440 blkg_init_queue(q); 441 442 /* 443 * Init percpu_ref in atomic mode so that it's faster to shutdown. 444 * See blk_register_queue() for details. 445 */ 446 error = percpu_ref_init(&q->q_usage_counter, 447 blk_queue_usage_counter_release, 448 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL); 449 if (error) 450 goto fail_stats; 451 lockdep_register_key(&q->io_lock_cls_key); 452 lockdep_register_key(&q->q_lock_cls_key); 453 lockdep_init_map(&q->io_lockdep_map, "&q->q_usage_counter(io)", 454 &q->io_lock_cls_key, 0); 455 lockdep_init_map(&q->q_lockdep_map, "&q->q_usage_counter(queue)", 456 &q->q_lock_cls_key, 0); 457 458 /* Teach lockdep about lock ordering (reclaim WRT queue freeze lock). */ 459 fs_reclaim_acquire(GFP_KERNEL); 460 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_); 461 rwsem_release(&q->io_lockdep_map, _RET_IP_); 462 fs_reclaim_release(GFP_KERNEL); 463 464 q->nr_requests = BLKDEV_DEFAULT_RQ; 465 q->async_depth = 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 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 if (!maxsector) 564 return -EIO; 565 pr_info_ratelimited("%s: attempt to access beyond end of device\n" 566 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", 567 current->comm, bio->bi_bdev, bio->bi_opf, 568 bio->bi_iter.bi_sector, nr_sectors, maxsector); 569 return -EIO; 570 } 571 return 0; 572 } 573 574 /* 575 * Remap block n of partition p to block n+start(p) of the disk. 576 */ 577 static int blk_partition_remap(struct bio *bio) 578 { 579 struct block_device *p = bio->bi_bdev; 580 581 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 582 return -EIO; 583 if (bio_sectors(bio)) { 584 bio->bi_iter.bi_sector += p->bd_start_sect; 585 trace_block_bio_remap(bio, p->bd_dev, 586 bio->bi_iter.bi_sector - 587 p->bd_start_sect); 588 } 589 bio_set_flag(bio, BIO_REMAPPED); 590 return 0; 591 } 592 593 /* 594 * Check write append to a zoned block device. 595 */ 596 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 597 struct bio *bio) 598 { 599 int nr_sectors = bio_sectors(bio); 600 601 /* Only applicable to zoned block devices */ 602 if (!bdev_is_zoned(bio->bi_bdev)) 603 return BLK_STS_NOTSUPP; 604 605 /* The bio sector must point to the start of a sequential zone */ 606 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector)) 607 return BLK_STS_IOERR; 608 609 /* 610 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 611 * split and could result in non-contiguous sectors being written in 612 * different zones. 613 */ 614 if (nr_sectors > q->limits.chunk_sectors) 615 return BLK_STS_IOERR; 616 617 /* Make sure the BIO is small enough and will not get split */ 618 if (nr_sectors > q->limits.max_zone_append_sectors) 619 return BLK_STS_IOERR; 620 621 bio->bi_opf |= REQ_NOMERGE; 622 623 return BLK_STS_OK; 624 } 625 626 static void __submit_bio(struct bio *bio) 627 { 628 /* If plug is not used, add new plug here to cache nsecs time. */ 629 struct blk_plug plug; 630 631 blk_start_plug(&plug); 632 633 if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) { 634 blk_mq_submit_bio(bio); 635 } else if (likely(bio_queue_enter(bio) == 0)) { 636 struct gendisk *disk = bio->bi_bdev->bd_disk; 637 638 if ((bio->bi_opf & REQ_POLLED) && 639 !(disk->queue->limits.features & BLK_FEAT_POLL)) 640 bio_endio_status(bio, BLK_STS_NOTSUPP); 641 else 642 disk->fops->submit_bio(bio); 643 blk_queue_exit(disk->queue); 644 } 645 646 blk_finish_plug(&plug); 647 } 648 649 /* 650 * The loop in this function may be a bit non-obvious, and so deserves some 651 * explanation: 652 * 653 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 654 * that), so we have a list with a single bio. 655 * - We pretend that we have just taken it off a longer list, so we assign 656 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 657 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 658 * bios through a recursive call to submit_bio_noacct. If it did, we find a 659 * non-NULL value in bio_list and re-enter the loop from the top. 660 * - In this case we really did just take the bio off the top of the list (no 661 * pretending) and so remove it from bio_list, and call into ->submit_bio() 662 * again. 663 * 664 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 665 * bio_list_on_stack[1] contains bios that were submitted before the current 666 * ->submit_bio(), but that haven't been processed yet. 667 */ 668 static void __submit_bio_noacct(struct bio *bio) 669 { 670 struct bio_list bio_list_on_stack[2]; 671 672 BUG_ON(bio->bi_next); 673 674 bio_list_init(&bio_list_on_stack[0]); 675 current->bio_list = bio_list_on_stack; 676 677 do { 678 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 679 struct bio_list lower, same; 680 681 /* 682 * Create a fresh bio_list for all subordinate requests. 683 */ 684 bio_list_on_stack[1] = bio_list_on_stack[0]; 685 bio_list_init(&bio_list_on_stack[0]); 686 687 __submit_bio(bio); 688 689 /* 690 * Sort new bios into those for a lower level and those for the 691 * same level. 692 */ 693 bio_list_init(&lower); 694 bio_list_init(&same); 695 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 696 if (q == bdev_get_queue(bio->bi_bdev)) 697 bio_list_add(&same, bio); 698 else 699 bio_list_add(&lower, bio); 700 701 /* 702 * Now assemble so we handle the lowest level first. 703 */ 704 bio_list_merge(&bio_list_on_stack[0], &lower); 705 bio_list_merge(&bio_list_on_stack[0], &same); 706 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 707 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 708 709 current->bio_list = NULL; 710 } 711 712 static void __submit_bio_noacct_mq(struct bio *bio) 713 { 714 struct bio_list bio_list[2] = { }; 715 716 current->bio_list = bio_list; 717 718 do { 719 __submit_bio(bio); 720 } while ((bio = bio_list_pop(&bio_list[0]))); 721 722 current->bio_list = NULL; 723 } 724 725 void submit_bio_noacct_nocheck(struct bio *bio, bool split) 726 { 727 blk_cgroup_bio_start(bio); 728 729 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 730 trace_block_bio_queue(bio); 731 /* 732 * Now that enqueuing has been traced, we need to trace 733 * completion as well. 734 */ 735 bio_set_flag(bio, BIO_TRACE_COMPLETION); 736 } 737 738 /* 739 * We only want one ->submit_bio to be active at a time, else stack 740 * usage with stacked devices could be a problem. Use current->bio_list 741 * to collect a list of requests submitted by a ->submit_bio method 742 * while it is active, and then process them after it returned. 743 */ 744 if (current->bio_list) { 745 if (split) 746 bio_list_add_head(¤t->bio_list[0], bio); 747 else 748 bio_list_add(¤t->bio_list[0], bio); 749 } else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) { 750 __submit_bio_noacct_mq(bio); 751 } else { 752 __submit_bio_noacct(bio); 753 } 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 (bio_has_crypt_ctx(bio)) { 793 if (WARN_ON_ONCE(!bio_has_data(bio))) 794 goto end_io; 795 if (!blk_crypto_supported(bio)) 796 goto not_supported; 797 } 798 799 if (should_fail_bio(bio)) 800 goto end_io; 801 bio_check_ro(bio); 802 if (!bio_flagged(bio, BIO_REMAPPED)) { 803 if (unlikely(bio_check_eod(bio))) 804 goto end_io; 805 if (bdev_is_partition(bdev) && 806 unlikely(blk_partition_remap(bio))) 807 goto end_io; 808 } 809 810 /* 811 * Filter flush bio's early so that bio based drivers without flush 812 * support don't have to worry about them. 813 */ 814 if (op_is_flush(bio->bi_opf)) { 815 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE && 816 bio_op(bio) != REQ_OP_ZONE_APPEND)) 817 goto end_io; 818 if (!bdev_write_cache(bdev)) { 819 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 820 if (!bio_sectors(bio)) { 821 status = BLK_STS_OK; 822 goto end_io; 823 } 824 } 825 } 826 827 switch (bio_op(bio)) { 828 case REQ_OP_READ: 829 break; 830 case REQ_OP_WRITE: 831 if (bio->bi_opf & REQ_ATOMIC) { 832 status = blk_validate_atomic_write_op_size(q, bio); 833 if (status != BLK_STS_OK) 834 goto end_io; 835 } 836 break; 837 case REQ_OP_FLUSH: 838 /* 839 * REQ_OP_FLUSH can't be submitted through bios, it is only 840 * synthetized in struct request by the flush state machine. 841 */ 842 goto not_supported; 843 case REQ_OP_DISCARD: 844 if (!bdev_max_discard_sectors(bdev)) 845 goto not_supported; 846 break; 847 case REQ_OP_SECURE_ERASE: 848 if (!bdev_max_secure_erase_sectors(bdev)) 849 goto not_supported; 850 break; 851 case REQ_OP_ZONE_APPEND: 852 status = blk_check_zone_append(q, bio); 853 if (status != BLK_STS_OK) 854 goto end_io; 855 break; 856 case REQ_OP_WRITE_ZEROES: 857 if (!q->limits.max_write_zeroes_sectors) 858 goto not_supported; 859 break; 860 case REQ_OP_ZONE_RESET: 861 case REQ_OP_ZONE_OPEN: 862 case REQ_OP_ZONE_CLOSE: 863 case REQ_OP_ZONE_FINISH: 864 case REQ_OP_ZONE_RESET_ALL: 865 if (!bdev_is_zoned(bio->bi_bdev)) 866 goto not_supported; 867 break; 868 case REQ_OP_DRV_IN: 869 case REQ_OP_DRV_OUT: 870 /* 871 * Driver private operations are only used with passthrough 872 * requests. 873 */ 874 fallthrough; 875 default: 876 goto not_supported; 877 } 878 879 if (blk_throtl_bio(bio)) 880 return; 881 submit_bio_noacct_nocheck(bio, false); 882 return; 883 884 not_supported: 885 status = BLK_STS_NOTSUPP; 886 end_io: 887 bio_endio_status(bio, status); 888 } 889 EXPORT_SYMBOL(submit_bio_noacct); 890 891 static void bio_set_ioprio(struct bio *bio) 892 { 893 /* Nobody set ioprio so far? Initialize it based on task's nice value */ 894 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE) 895 bio->bi_ioprio = get_current_ioprio(); 896 blkcg_set_ioprio(bio); 897 } 898 899 /** 900 * submit_bio - submit a bio to the block device layer for I/O 901 * @bio: The &struct bio which describes the I/O 902 * 903 * submit_bio() is used to submit I/O requests to block devices. It is passed a 904 * fully set up &struct bio that describes the I/O that needs to be done. The 905 * bio will be sent to the device described by the bi_bdev field. 906 * 907 * The success/failure status of the request, along with notification of 908 * completion, is delivered asynchronously through the ->bi_end_io() callback 909 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has 910 * been called. 911 */ 912 void submit_bio(struct bio *bio) 913 { 914 if (bio_op(bio) == REQ_OP_READ) { 915 task_io_account_read(bio->bi_iter.bi_size); 916 count_vm_events(PGPGIN, bio_sectors(bio)); 917 } else if (bio_op(bio) == REQ_OP_WRITE) { 918 count_vm_events(PGPGOUT, bio_sectors(bio)); 919 } 920 921 bio_set_ioprio(bio); 922 submit_bio_noacct(bio); 923 } 924 EXPORT_SYMBOL(submit_bio); 925 926 /** 927 * bio_poll - poll for BIO completions 928 * @bio: bio to poll for 929 * @iob: batches of IO 930 * @flags: BLK_POLL_* flags that control the behavior 931 * 932 * Poll for completions on queue associated with the bio. Returns number of 933 * completed entries found. 934 * 935 * Note: the caller must either be the context that submitted @bio, or 936 * be in a RCU critical section to prevent freeing of @bio. 937 */ 938 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) 939 { 940 blk_qc_t cookie = READ_ONCE(bio->bi_cookie); 941 struct block_device *bdev; 942 struct request_queue *q; 943 int ret = 0; 944 945 bdev = READ_ONCE(bio->bi_bdev); 946 if (!bdev) 947 return 0; 948 949 q = bdev_get_queue(bdev); 950 if (cookie == BLK_QC_T_NONE) 951 return 0; 952 953 blk_flush_plug(current->plug, false); 954 955 /* 956 * We need to be able to enter a frozen queue, similar to how 957 * timeouts also need to do that. If that is blocked, then we can 958 * have pending IO when a queue freeze is started, and then the 959 * wait for the freeze to finish will wait for polled requests to 960 * timeout as the poller is preventer from entering the queue and 961 * completing them. As long as we prevent new IO from being queued, 962 * that should be all that matters. 963 */ 964 if (!percpu_ref_tryget(&q->q_usage_counter)) 965 return 0; 966 if (queue_is_mq(q)) { 967 ret = blk_mq_poll(q, cookie, iob, flags); 968 } else { 969 struct gendisk *disk = q->disk; 970 971 if ((q->limits.features & BLK_FEAT_POLL) && disk && 972 disk->fops->poll_bio) 973 ret = disk->fops->poll_bio(bio, iob, flags); 974 } 975 blk_queue_exit(q); 976 return ret; 977 } 978 EXPORT_SYMBOL_GPL(bio_poll); 979 980 /* 981 * Helper to implement file_operations.iopoll. Requires the bio to be stored 982 * in iocb->private, and cleared before freeing the bio. 983 */ 984 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, 985 unsigned int flags) 986 { 987 struct bio *bio; 988 int ret = 0; 989 990 /* 991 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can 992 * point to a freshly allocated bio at this point. If that happens 993 * we have a few cases to consider: 994 * 995 * 1) the bio is being initialized and bi_bdev is NULL. We can just 996 * simply nothing in this case 997 * 2) the bio points to a not poll enabled device. bio_poll will catch 998 * this and return 0 999 * 3) the bio points to a poll capable device, including but not 1000 * limited to the one that the original bio pointed to. In this 1001 * case we will call into the actual poll method and poll for I/O, 1002 * even if we don't need to, but it won't cause harm either. 1003 * 1004 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev 1005 * is still allocated. Because partitions hold a reference to the whole 1006 * device bdev and thus disk, the disk is also still valid. Grabbing 1007 * a reference to the queue in bio_poll() ensures the hctxs and requests 1008 * are still valid as well. 1009 */ 1010 rcu_read_lock(); 1011 bio = READ_ONCE(kiocb->private); 1012 if (bio) 1013 ret = bio_poll(bio, iob, flags); 1014 rcu_read_unlock(); 1015 1016 return ret; 1017 } 1018 EXPORT_SYMBOL_GPL(iocb_bio_iopoll); 1019 1020 void update_io_ticks(struct block_device *part, unsigned long now, bool end) 1021 { 1022 unsigned long stamp; 1023 again: 1024 stamp = READ_ONCE(part->bd_stamp); 1025 if (unlikely(time_after(now, stamp)) && 1026 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) && 1027 (end || bdev_count_inflight(part))) 1028 __part_stat_add(part, io_ticks, now - stamp); 1029 1030 if (bdev_is_partition(part)) { 1031 part = bdev_whole(part); 1032 goto again; 1033 } 1034 } 1035 1036 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op, 1037 unsigned long start_time) 1038 { 1039 part_stat_lock(); 1040 update_io_ticks(bdev, start_time, false); 1041 part_stat_local_inc(bdev, in_flight[op_is_write(op)]); 1042 part_stat_unlock(); 1043 1044 return start_time; 1045 } 1046 EXPORT_SYMBOL(bdev_start_io_acct); 1047 1048 /** 1049 * bio_start_io_acct - start I/O accounting for bio based drivers 1050 * @bio: bio to start account for 1051 * 1052 * Returns the start time that should be passed back to bio_end_io_acct(). 1053 */ 1054 unsigned long bio_start_io_acct(struct bio *bio) 1055 { 1056 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies); 1057 } 1058 EXPORT_SYMBOL_GPL(bio_start_io_acct); 1059 1060 void bdev_end_io_acct(struct block_device *bdev, enum req_op op, 1061 unsigned int sectors, unsigned long start_time) 1062 { 1063 const int sgrp = op_stat_group(op); 1064 unsigned long now = READ_ONCE(jiffies); 1065 unsigned long duration = now - start_time; 1066 1067 part_stat_lock(); 1068 update_io_ticks(bdev, now, true); 1069 part_stat_inc(bdev, ios[sgrp]); 1070 part_stat_add(bdev, sectors[sgrp], sectors); 1071 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); 1072 part_stat_local_dec(bdev, in_flight[op_is_write(op)]); 1073 part_stat_unlock(); 1074 } 1075 EXPORT_SYMBOL(bdev_end_io_acct); 1076 1077 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, 1078 struct block_device *orig_bdev) 1079 { 1080 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time); 1081 } 1082 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); 1083 1084 /** 1085 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1086 * @q : the queue of the device being checked 1087 * 1088 * Description: 1089 * Check if underlying low-level drivers of a device are busy. 1090 * If the drivers want to export their busy state, they must set own 1091 * exporting function using blk_queue_lld_busy() first. 1092 * 1093 * Basically, this function is used only by request stacking drivers 1094 * to stop dispatching requests to underlying devices when underlying 1095 * devices are busy. This behavior helps more I/O merging on the queue 1096 * of the request stacking driver and prevents I/O throughput regression 1097 * on burst I/O load. 1098 * 1099 * Return: 1100 * 0 - Not busy (The request stacking driver should dispatch request) 1101 * 1 - Busy (The request stacking driver should stop dispatching request) 1102 */ 1103 int blk_lld_busy(struct request_queue *q) 1104 { 1105 if (queue_is_mq(q) && q->mq_ops->busy) 1106 return q->mq_ops->busy(q); 1107 1108 return 0; 1109 } 1110 EXPORT_SYMBOL_GPL(blk_lld_busy); 1111 1112 int kblockd_schedule_work(struct work_struct *work) 1113 { 1114 return queue_work(kblockd_workqueue, work); 1115 } 1116 EXPORT_SYMBOL(kblockd_schedule_work); 1117 1118 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1119 unsigned long delay) 1120 { 1121 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1122 } 1123 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1124 1125 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) 1126 { 1127 struct task_struct *tsk = current; 1128 1129 /* 1130 * If this is a nested plug, don't actually assign it. 1131 */ 1132 if (tsk->plug) 1133 return; 1134 1135 plug->cur_ktime = 0; 1136 rq_list_init(&plug->mq_list); 1137 rq_list_init(&plug->cached_rqs); 1138 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); 1139 plug->rq_count = 0; 1140 plug->multiple_queues = false; 1141 plug->has_elevator = false; 1142 INIT_LIST_HEAD(&plug->cb_list); 1143 1144 /* 1145 * Store ordering should not be needed here, since a potential 1146 * preempt will imply a full memory barrier 1147 */ 1148 tsk->plug = plug; 1149 } 1150 1151 /** 1152 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1153 * @plug: The &struct blk_plug that needs to be initialized 1154 * 1155 * Description: 1156 * blk_start_plug() indicates to the block layer an intent by the caller 1157 * to submit multiple I/O requests in a batch. The block layer may use 1158 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1159 * is called. However, the block layer may choose to submit requests 1160 * before a call to blk_finish_plug() if the number of queued I/Os 1161 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1162 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1163 * the task schedules (see below). 1164 * 1165 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1166 * pending I/O should the task end up blocking between blk_start_plug() and 1167 * blk_finish_plug(). This is important from a performance perspective, but 1168 * also ensures that we don't deadlock. For instance, if the task is blocking 1169 * for a memory allocation, memory reclaim could end up wanting to free a 1170 * page belonging to that request that is currently residing in our private 1171 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1172 * this kind of deadlock. 1173 */ 1174 void blk_start_plug(struct blk_plug *plug) 1175 { 1176 blk_start_plug_nr_ios(plug, 1); 1177 } 1178 EXPORT_SYMBOL(blk_start_plug); 1179 1180 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1181 { 1182 LIST_HEAD(callbacks); 1183 1184 while (!list_empty(&plug->cb_list)) { 1185 list_splice_init(&plug->cb_list, &callbacks); 1186 1187 while (!list_empty(&callbacks)) { 1188 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1189 struct blk_plug_cb, 1190 list); 1191 list_del(&cb->list); 1192 cb->callback(cb, from_schedule); 1193 } 1194 } 1195 } 1196 1197 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1198 int size) 1199 { 1200 struct blk_plug *plug = current->plug; 1201 struct blk_plug_cb *cb; 1202 1203 if (!plug) 1204 return NULL; 1205 1206 list_for_each_entry(cb, &plug->cb_list, list) 1207 if (cb->callback == unplug && cb->data == data) 1208 return cb; 1209 1210 /* Not currently on the callback list */ 1211 BUG_ON(size < sizeof(*cb)); 1212 cb = kzalloc(size, GFP_ATOMIC); 1213 if (cb) { 1214 cb->data = data; 1215 cb->callback = unplug; 1216 list_add(&cb->list, &plug->cb_list); 1217 } 1218 return cb; 1219 } 1220 EXPORT_SYMBOL(blk_check_plugged); 1221 1222 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) 1223 { 1224 if (!list_empty(&plug->cb_list)) 1225 flush_plug_callbacks(plug, from_schedule); 1226 blk_mq_flush_plug_list(plug, from_schedule); 1227 /* 1228 * Unconditionally flush out cached requests, even if the unplug 1229 * event came from schedule. Since we know hold references to the 1230 * queue for cached requests, we don't want a blocked task holding 1231 * up a queue freeze/quiesce event. 1232 */ 1233 if (unlikely(!rq_list_empty(&plug->cached_rqs))) 1234 blk_mq_free_plug_rqs(plug); 1235 1236 plug->cur_ktime = 0; 1237 current->flags &= ~PF_BLOCK_TS; 1238 } 1239 1240 /** 1241 * blk_finish_plug - mark the end of a batch of submitted I/O 1242 * @plug: The &struct blk_plug passed to blk_start_plug() 1243 * 1244 * Description: 1245 * Indicate that a batch of I/O submissions is complete. This function 1246 * must be paired with an initial call to blk_start_plug(). The intent 1247 * is to allow the block layer to optimize I/O submission. See the 1248 * documentation for blk_start_plug() for more information. 1249 */ 1250 void blk_finish_plug(struct blk_plug *plug) 1251 { 1252 if (plug == current->plug) { 1253 __blk_flush_plug(plug, false); 1254 current->plug = NULL; 1255 } 1256 } 1257 EXPORT_SYMBOL(blk_finish_plug); 1258 1259 void blk_io_schedule(void) 1260 { 1261 /* Prevent hang_check timer from firing at us during very long I/O */ 1262 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1263 1264 if (timeout) 1265 io_schedule_timeout(timeout); 1266 else 1267 io_schedule(); 1268 } 1269 1270 int __init blk_dev_init(void) 1271 { 1272 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1273 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1274 sizeof_field(struct request, cmd_flags)); 1275 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1276 sizeof_field(struct bio, bi_opf)); 1277 1278 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1279 kblockd_workqueue = alloc_workqueue("kblockd", 1280 WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_PERCPU, 0); 1281 if (!kblockd_workqueue) 1282 panic("Failed to create kblockd\n"); 1283 1284 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC); 1285 1286 blk_debugfs_root = debugfs_create_dir("block", NULL); 1287 1288 return 0; 1289 } 1290