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 /** 98 * blk_queue_flag_test_and_set - atomically test and set a queue flag 99 * @flag: flag to be set 100 * @q: request queue 101 * 102 * Returns the previous value of @flag - 0 if the flag was not set and 1 if 103 * the flag was already set. 104 */ 105 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) 106 { 107 return test_and_set_bit(flag, &q->queue_flags); 108 } 109 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); 110 111 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name 112 static const char *const blk_op_name[] = { 113 REQ_OP_NAME(READ), 114 REQ_OP_NAME(WRITE), 115 REQ_OP_NAME(FLUSH), 116 REQ_OP_NAME(DISCARD), 117 REQ_OP_NAME(SECURE_ERASE), 118 REQ_OP_NAME(ZONE_RESET), 119 REQ_OP_NAME(ZONE_RESET_ALL), 120 REQ_OP_NAME(ZONE_OPEN), 121 REQ_OP_NAME(ZONE_CLOSE), 122 REQ_OP_NAME(ZONE_FINISH), 123 REQ_OP_NAME(ZONE_APPEND), 124 REQ_OP_NAME(WRITE_ZEROES), 125 REQ_OP_NAME(DRV_IN), 126 REQ_OP_NAME(DRV_OUT), 127 }; 128 #undef REQ_OP_NAME 129 130 /** 131 * blk_op_str - Return string XXX in the REQ_OP_XXX. 132 * @op: REQ_OP_XXX. 133 * 134 * Description: Centralize block layer function to convert REQ_OP_XXX into 135 * string format. Useful in the debugging and tracing bio or request. For 136 * invalid REQ_OP_XXX it returns string "UNKNOWN". 137 */ 138 inline const char *blk_op_str(enum req_op op) 139 { 140 const char *op_str = "UNKNOWN"; 141 142 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) 143 op_str = blk_op_name[op]; 144 145 return op_str; 146 } 147 EXPORT_SYMBOL_GPL(blk_op_str); 148 149 static const struct { 150 int errno; 151 const char *name; 152 } blk_errors[] = { 153 [BLK_STS_OK] = { 0, "" }, 154 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, 155 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, 156 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, 157 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, 158 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, 159 [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" }, 160 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, 161 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, 162 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, 163 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, 164 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, 165 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" }, 166 167 /* device mapper special case, should not leak out: */ 168 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, 169 170 /* zone device specific errors */ 171 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, 172 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, 173 174 /* Command duration limit device-side timeout */ 175 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" }, 176 177 /* everything else not covered above: */ 178 [BLK_STS_IOERR] = { -EIO, "I/O" }, 179 }; 180 181 blk_status_t errno_to_blk_status(int errno) 182 { 183 int i; 184 185 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { 186 if (blk_errors[i].errno == errno) 187 return (__force blk_status_t)i; 188 } 189 190 return BLK_STS_IOERR; 191 } 192 EXPORT_SYMBOL_GPL(errno_to_blk_status); 193 194 int blk_status_to_errno(blk_status_t status) 195 { 196 int idx = (__force int)status; 197 198 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 199 return -EIO; 200 return blk_errors[idx].errno; 201 } 202 EXPORT_SYMBOL_GPL(blk_status_to_errno); 203 204 const char *blk_status_to_str(blk_status_t status) 205 { 206 int idx = (__force int)status; 207 208 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 209 return "<null>"; 210 return blk_errors[idx].name; 211 } 212 EXPORT_SYMBOL_GPL(blk_status_to_str); 213 214 /** 215 * blk_sync_queue - cancel any pending callbacks on a queue 216 * @q: the queue 217 * 218 * Description: 219 * The block layer may perform asynchronous callback activity 220 * on a queue, such as calling the unplug function after a timeout. 221 * A block device may call blk_sync_queue to ensure that any 222 * such activity is cancelled, thus allowing it to release resources 223 * that the callbacks might use. The caller must already have made sure 224 * that its ->submit_bio will not re-add plugging prior to calling 225 * this function. 226 * 227 * This function does not cancel any asynchronous activity arising 228 * out of elevator or throttling code. That would require elevator_exit() 229 * and blkcg_exit_queue() to be called with queue lock initialized. 230 * 231 */ 232 void blk_sync_queue(struct request_queue *q) 233 { 234 del_timer_sync(&q->timeout); 235 cancel_work_sync(&q->timeout_work); 236 } 237 EXPORT_SYMBOL(blk_sync_queue); 238 239 /** 240 * blk_set_pm_only - increment pm_only counter 241 * @q: request queue pointer 242 */ 243 void blk_set_pm_only(struct request_queue *q) 244 { 245 atomic_inc(&q->pm_only); 246 } 247 EXPORT_SYMBOL_GPL(blk_set_pm_only); 248 249 void blk_clear_pm_only(struct request_queue *q) 250 { 251 int pm_only; 252 253 pm_only = atomic_dec_return(&q->pm_only); 254 WARN_ON_ONCE(pm_only < 0); 255 if (pm_only == 0) 256 wake_up_all(&q->mq_freeze_wq); 257 } 258 EXPORT_SYMBOL_GPL(blk_clear_pm_only); 259 260 static void blk_free_queue_rcu(struct rcu_head *rcu_head) 261 { 262 struct request_queue *q = container_of(rcu_head, 263 struct request_queue, rcu_head); 264 265 percpu_ref_exit(&q->q_usage_counter); 266 kmem_cache_free(blk_requestq_cachep, q); 267 } 268 269 static void blk_free_queue(struct request_queue *q) 270 { 271 blk_free_queue_stats(q->stats); 272 if (queue_is_mq(q)) 273 blk_mq_release(q); 274 275 ida_free(&blk_queue_ida, q->id); 276 call_rcu(&q->rcu_head, blk_free_queue_rcu); 277 } 278 279 /** 280 * blk_put_queue - decrement the request_queue refcount 281 * @q: the request_queue structure to decrement the refcount for 282 * 283 * Decrements the refcount of the request_queue and free it when the refcount 284 * reaches 0. 285 */ 286 void blk_put_queue(struct request_queue *q) 287 { 288 if (refcount_dec_and_test(&q->refs)) 289 blk_free_queue(q); 290 } 291 EXPORT_SYMBOL(blk_put_queue); 292 293 void blk_queue_start_drain(struct request_queue *q) 294 { 295 /* 296 * When queue DYING flag is set, we need to block new req 297 * entering queue, so we call blk_freeze_queue_start() to 298 * prevent I/O from crossing blk_queue_enter(). 299 */ 300 blk_freeze_queue_start(q); 301 if (queue_is_mq(q)) 302 blk_mq_wake_waiters(q); 303 /* Make blk_queue_enter() reexamine the DYING flag. */ 304 wake_up_all(&q->mq_freeze_wq); 305 } 306 307 /** 308 * blk_queue_enter() - try to increase q->q_usage_counter 309 * @q: request queue pointer 310 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM 311 */ 312 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) 313 { 314 const bool pm = flags & BLK_MQ_REQ_PM; 315 316 while (!blk_try_enter_queue(q, pm)) { 317 if (flags & BLK_MQ_REQ_NOWAIT) 318 return -EAGAIN; 319 320 /* 321 * read pair of barrier in blk_freeze_queue_start(), we need to 322 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 323 * reading .mq_freeze_depth or queue dying flag, otherwise the 324 * following wait may never return if the two reads are 325 * reordered. 326 */ 327 smp_rmb(); 328 wait_event(q->mq_freeze_wq, 329 (!q->mq_freeze_depth && 330 blk_pm_resume_queue(pm, q)) || 331 blk_queue_dying(q)); 332 if (blk_queue_dying(q)) 333 return -ENODEV; 334 } 335 336 return 0; 337 } 338 339 int __bio_queue_enter(struct request_queue *q, struct bio *bio) 340 { 341 while (!blk_try_enter_queue(q, false)) { 342 struct gendisk *disk = bio->bi_bdev->bd_disk; 343 344 if (bio->bi_opf & REQ_NOWAIT) { 345 if (test_bit(GD_DEAD, &disk->state)) 346 goto dead; 347 bio_wouldblock_error(bio); 348 return -EAGAIN; 349 } 350 351 /* 352 * read pair of barrier in blk_freeze_queue_start(), we need to 353 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 354 * reading .mq_freeze_depth or queue dying flag, otherwise the 355 * following wait may never return if the two reads are 356 * reordered. 357 */ 358 smp_rmb(); 359 wait_event(q->mq_freeze_wq, 360 (!q->mq_freeze_depth && 361 blk_pm_resume_queue(false, q)) || 362 test_bit(GD_DEAD, &disk->state)); 363 if (test_bit(GD_DEAD, &disk->state)) 364 goto dead; 365 } 366 367 return 0; 368 dead: 369 bio_io_error(bio); 370 return -ENODEV; 371 } 372 373 void blk_queue_exit(struct request_queue *q) 374 { 375 percpu_ref_put(&q->q_usage_counter); 376 } 377 378 static void blk_queue_usage_counter_release(struct percpu_ref *ref) 379 { 380 struct request_queue *q = 381 container_of(ref, struct request_queue, q_usage_counter); 382 383 wake_up_all(&q->mq_freeze_wq); 384 } 385 386 static void blk_rq_timed_out_timer(struct timer_list *t) 387 { 388 struct request_queue *q = from_timer(q, t, timeout); 389 390 kblockd_schedule_work(&q->timeout_work); 391 } 392 393 static void blk_timeout_work(struct work_struct *work) 394 { 395 } 396 397 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id) 398 { 399 struct request_queue *q; 400 int error; 401 402 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO, 403 node_id); 404 if (!q) 405 return ERR_PTR(-ENOMEM); 406 407 q->last_merge = NULL; 408 409 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL); 410 if (q->id < 0) { 411 error = q->id; 412 goto fail_q; 413 } 414 415 q->stats = blk_alloc_queue_stats(); 416 if (!q->stats) { 417 error = -ENOMEM; 418 goto fail_id; 419 } 420 421 error = blk_set_default_limits(lim); 422 if (error) 423 goto fail_stats; 424 q->limits = *lim; 425 426 q->node = node_id; 427 428 atomic_set(&q->nr_active_requests_shared_tags, 0); 429 430 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); 431 INIT_WORK(&q->timeout_work, blk_timeout_work); 432 INIT_LIST_HEAD(&q->icq_list); 433 434 refcount_set(&q->refs, 1); 435 mutex_init(&q->debugfs_mutex); 436 mutex_init(&q->sysfs_lock); 437 mutex_init(&q->sysfs_dir_lock); 438 mutex_init(&q->limits_lock); 439 mutex_init(&q->rq_qos_mutex); 440 spin_lock_init(&q->queue_lock); 441 442 init_waitqueue_head(&q->mq_freeze_wq); 443 mutex_init(&q->mq_freeze_lock); 444 445 blkg_init_queue(q); 446 447 /* 448 * Init percpu_ref in atomic mode so that it's faster to shutdown. 449 * See blk_register_queue() for details. 450 */ 451 error = percpu_ref_init(&q->q_usage_counter, 452 blk_queue_usage_counter_release, 453 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL); 454 if (error) 455 goto fail_stats; 456 457 q->nr_requests = BLKDEV_DEFAULT_RQ; 458 459 return q; 460 461 fail_stats: 462 blk_free_queue_stats(q->stats); 463 fail_id: 464 ida_free(&blk_queue_ida, q->id); 465 fail_q: 466 kmem_cache_free(blk_requestq_cachep, q); 467 return ERR_PTR(error); 468 } 469 470 /** 471 * blk_get_queue - increment the request_queue refcount 472 * @q: the request_queue structure to increment the refcount for 473 * 474 * Increment the refcount of the request_queue kobject. 475 * 476 * Context: Any context. 477 */ 478 bool blk_get_queue(struct request_queue *q) 479 { 480 if (unlikely(blk_queue_dying(q))) 481 return false; 482 refcount_inc(&q->refs); 483 return true; 484 } 485 EXPORT_SYMBOL(blk_get_queue); 486 487 #ifdef CONFIG_FAIL_MAKE_REQUEST 488 489 static DECLARE_FAULT_ATTR(fail_make_request); 490 491 static int __init setup_fail_make_request(char *str) 492 { 493 return setup_fault_attr(&fail_make_request, str); 494 } 495 __setup("fail_make_request=", setup_fail_make_request); 496 497 bool should_fail_request(struct block_device *part, unsigned int bytes) 498 { 499 return bdev_test_flag(part, BD_MAKE_IT_FAIL) && 500 should_fail(&fail_make_request, bytes); 501 } 502 503 static int __init fail_make_request_debugfs(void) 504 { 505 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 506 NULL, &fail_make_request); 507 508 return PTR_ERR_OR_ZERO(dir); 509 } 510 511 late_initcall(fail_make_request_debugfs); 512 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 513 514 static inline void bio_check_ro(struct bio *bio) 515 { 516 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { 517 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) 518 return; 519 520 if (bdev_test_flag(bio->bi_bdev, BD_RO_WARNED)) 521 return; 522 523 bdev_set_flag(bio->bi_bdev, BD_RO_WARNED); 524 525 /* 526 * Use ioctl to set underlying disk of raid/dm to read-only 527 * will trigger this. 528 */ 529 pr_warn("Trying to write to read-only block-device %pg\n", 530 bio->bi_bdev); 531 } 532 } 533 534 static noinline int should_fail_bio(struct bio *bio) 535 { 536 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) 537 return -EIO; 538 return 0; 539 } 540 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); 541 542 /* 543 * Check whether this bio extends beyond the end of the device or partition. 544 * This may well happen - the kernel calls bread() without checking the size of 545 * the device, e.g., when mounting a file system. 546 */ 547 static inline int bio_check_eod(struct bio *bio) 548 { 549 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 550 unsigned int nr_sectors = bio_sectors(bio); 551 552 if (nr_sectors && 553 (nr_sectors > maxsector || 554 bio->bi_iter.bi_sector > maxsector - nr_sectors)) { 555 pr_info_ratelimited("%s: attempt to access beyond end of device\n" 556 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", 557 current->comm, bio->bi_bdev, bio->bi_opf, 558 bio->bi_iter.bi_sector, nr_sectors, maxsector); 559 return -EIO; 560 } 561 return 0; 562 } 563 564 /* 565 * Remap block n of partition p to block n+start(p) of the disk. 566 */ 567 static int blk_partition_remap(struct bio *bio) 568 { 569 struct block_device *p = bio->bi_bdev; 570 571 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 572 return -EIO; 573 if (bio_sectors(bio)) { 574 bio->bi_iter.bi_sector += p->bd_start_sect; 575 trace_block_bio_remap(bio, p->bd_dev, 576 bio->bi_iter.bi_sector - 577 p->bd_start_sect); 578 } 579 bio_set_flag(bio, BIO_REMAPPED); 580 return 0; 581 } 582 583 /* 584 * Check write append to a zoned block device. 585 */ 586 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 587 struct bio *bio) 588 { 589 int nr_sectors = bio_sectors(bio); 590 591 /* Only applicable to zoned block devices */ 592 if (!bdev_is_zoned(bio->bi_bdev)) 593 return BLK_STS_NOTSUPP; 594 595 /* The bio sector must point to the start of a sequential zone */ 596 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector)) 597 return BLK_STS_IOERR; 598 599 /* 600 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 601 * split and could result in non-contiguous sectors being written in 602 * different zones. 603 */ 604 if (nr_sectors > q->limits.chunk_sectors) 605 return BLK_STS_IOERR; 606 607 /* Make sure the BIO is small enough and will not get split */ 608 if (nr_sectors > queue_max_zone_append_sectors(q)) 609 return BLK_STS_IOERR; 610 611 bio->bi_opf |= REQ_NOMERGE; 612 613 return BLK_STS_OK; 614 } 615 616 static void __submit_bio(struct bio *bio) 617 { 618 /* If plug is not used, add new plug here to cache nsecs time. */ 619 struct blk_plug plug; 620 621 if (unlikely(!blk_crypto_bio_prep(&bio))) 622 return; 623 624 blk_start_plug(&plug); 625 626 if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) { 627 blk_mq_submit_bio(bio); 628 } else if (likely(bio_queue_enter(bio) == 0)) { 629 struct gendisk *disk = bio->bi_bdev->bd_disk; 630 631 disk->fops->submit_bio(bio); 632 blk_queue_exit(disk->queue); 633 } 634 635 blk_finish_plug(&plug); 636 } 637 638 /* 639 * The loop in this function may be a bit non-obvious, and so deserves some 640 * explanation: 641 * 642 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 643 * that), so we have a list with a single bio. 644 * - We pretend that we have just taken it off a longer list, so we assign 645 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 646 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 647 * bios through a recursive call to submit_bio_noacct. If it did, we find a 648 * non-NULL value in bio_list and re-enter the loop from the top. 649 * - In this case we really did just take the bio of the top of the list (no 650 * pretending) and so remove it from bio_list, and call into ->submit_bio() 651 * again. 652 * 653 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 654 * bio_list_on_stack[1] contains bios that were submitted before the current 655 * ->submit_bio, but that haven't been processed yet. 656 */ 657 static void __submit_bio_noacct(struct bio *bio) 658 { 659 struct bio_list bio_list_on_stack[2]; 660 661 BUG_ON(bio->bi_next); 662 663 bio_list_init(&bio_list_on_stack[0]); 664 current->bio_list = bio_list_on_stack; 665 666 do { 667 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 668 struct bio_list lower, same; 669 670 /* 671 * Create a fresh bio_list for all subordinate requests. 672 */ 673 bio_list_on_stack[1] = bio_list_on_stack[0]; 674 bio_list_init(&bio_list_on_stack[0]); 675 676 __submit_bio(bio); 677 678 /* 679 * Sort new bios into those for a lower level and those for the 680 * same level. 681 */ 682 bio_list_init(&lower); 683 bio_list_init(&same); 684 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 685 if (q == bdev_get_queue(bio->bi_bdev)) 686 bio_list_add(&same, bio); 687 else 688 bio_list_add(&lower, bio); 689 690 /* 691 * Now assemble so we handle the lowest level first. 692 */ 693 bio_list_merge(&bio_list_on_stack[0], &lower); 694 bio_list_merge(&bio_list_on_stack[0], &same); 695 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 696 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 697 698 current->bio_list = NULL; 699 } 700 701 static void __submit_bio_noacct_mq(struct bio *bio) 702 { 703 struct bio_list bio_list[2] = { }; 704 705 current->bio_list = bio_list; 706 707 do { 708 __submit_bio(bio); 709 } while ((bio = bio_list_pop(&bio_list[0]))); 710 711 current->bio_list = NULL; 712 } 713 714 void submit_bio_noacct_nocheck(struct bio *bio) 715 { 716 blk_cgroup_bio_start(bio); 717 blkcg_bio_issue_init(bio); 718 719 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 720 trace_block_bio_queue(bio); 721 /* 722 * Now that enqueuing has been traced, we need to trace 723 * completion as well. 724 */ 725 bio_set_flag(bio, BIO_TRACE_COMPLETION); 726 } 727 728 /* 729 * We only want one ->submit_bio to be active at a time, else stack 730 * usage with stacked devices could be a problem. Use current->bio_list 731 * to collect a list of requests submited by a ->submit_bio method while 732 * it is active, and then process them after it returned. 733 */ 734 if (current->bio_list) 735 bio_list_add(¤t->bio_list[0], bio); 736 else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) 737 __submit_bio_noacct_mq(bio); 738 else 739 __submit_bio_noacct(bio); 740 } 741 742 /** 743 * submit_bio_noacct - re-submit a bio to the block device layer for I/O 744 * @bio: The bio describing the location in memory and on the device. 745 * 746 * This is a version of submit_bio() that shall only be used for I/O that is 747 * resubmitted to lower level drivers by stacking block drivers. All file 748 * systems and other upper level users of the block layer should use 749 * submit_bio() instead. 750 */ 751 void submit_bio_noacct(struct bio *bio) 752 { 753 struct block_device *bdev = bio->bi_bdev; 754 struct request_queue *q = bdev_get_queue(bdev); 755 blk_status_t status = BLK_STS_IOERR; 756 757 might_sleep(); 758 759 /* 760 * For a REQ_NOWAIT based request, return -EOPNOTSUPP 761 * if queue does not support NOWAIT. 762 */ 763 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev)) 764 goto not_supported; 765 766 if (should_fail_bio(bio)) 767 goto end_io; 768 bio_check_ro(bio); 769 if (!bio_flagged(bio, BIO_REMAPPED)) { 770 if (unlikely(bio_check_eod(bio))) 771 goto end_io; 772 if (bdev_is_partition(bdev) && 773 unlikely(blk_partition_remap(bio))) 774 goto end_io; 775 } 776 777 /* 778 * Filter flush bio's early so that bio based drivers without flush 779 * support don't have to worry about them. 780 */ 781 if (op_is_flush(bio->bi_opf)) { 782 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE && 783 bio_op(bio) != REQ_OP_ZONE_APPEND)) 784 goto end_io; 785 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { 786 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 787 if (!bio_sectors(bio)) { 788 status = BLK_STS_OK; 789 goto end_io; 790 } 791 } 792 } 793 794 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 795 bio_clear_polled(bio); 796 797 switch (bio_op(bio)) { 798 case REQ_OP_READ: 799 case REQ_OP_WRITE: 800 break; 801 case REQ_OP_FLUSH: 802 /* 803 * REQ_OP_FLUSH can't be submitted through bios, it is only 804 * synthetized in struct request by the flush state machine. 805 */ 806 goto not_supported; 807 case REQ_OP_DISCARD: 808 if (!bdev_max_discard_sectors(bdev)) 809 goto not_supported; 810 break; 811 case REQ_OP_SECURE_ERASE: 812 if (!bdev_max_secure_erase_sectors(bdev)) 813 goto not_supported; 814 break; 815 case REQ_OP_ZONE_APPEND: 816 status = blk_check_zone_append(q, bio); 817 if (status != BLK_STS_OK) 818 goto end_io; 819 break; 820 case REQ_OP_WRITE_ZEROES: 821 if (!q->limits.max_write_zeroes_sectors) 822 goto not_supported; 823 break; 824 case REQ_OP_ZONE_RESET: 825 case REQ_OP_ZONE_OPEN: 826 case REQ_OP_ZONE_CLOSE: 827 case REQ_OP_ZONE_FINISH: 828 if (!bdev_is_zoned(bio->bi_bdev)) 829 goto not_supported; 830 break; 831 case REQ_OP_ZONE_RESET_ALL: 832 if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q)) 833 goto not_supported; 834 break; 835 case REQ_OP_DRV_IN: 836 case REQ_OP_DRV_OUT: 837 /* 838 * Driver private operations are only used with passthrough 839 * requests. 840 */ 841 fallthrough; 842 default: 843 goto not_supported; 844 } 845 846 if (blk_throtl_bio(bio)) 847 return; 848 submit_bio_noacct_nocheck(bio); 849 return; 850 851 not_supported: 852 status = BLK_STS_NOTSUPP; 853 end_io: 854 bio->bi_status = status; 855 bio_endio(bio); 856 } 857 EXPORT_SYMBOL(submit_bio_noacct); 858 859 static void bio_set_ioprio(struct bio *bio) 860 { 861 /* Nobody set ioprio so far? Initialize it based on task's nice value */ 862 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE) 863 bio->bi_ioprio = get_current_ioprio(); 864 blkcg_set_ioprio(bio); 865 } 866 867 /** 868 * submit_bio - submit a bio to the block device layer for I/O 869 * @bio: The &struct bio which describes the I/O 870 * 871 * submit_bio() is used to submit I/O requests to block devices. It is passed a 872 * fully set up &struct bio that describes the I/O that needs to be done. The 873 * bio will be send to the device described by the bi_bdev field. 874 * 875 * The success/failure status of the request, along with notification of 876 * completion, is delivered asynchronously through the ->bi_end_io() callback 877 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has 878 * been called. 879 */ 880 void submit_bio(struct bio *bio) 881 { 882 if (bio_op(bio) == REQ_OP_READ) { 883 task_io_account_read(bio->bi_iter.bi_size); 884 count_vm_events(PGPGIN, bio_sectors(bio)); 885 } else if (bio_op(bio) == REQ_OP_WRITE) { 886 count_vm_events(PGPGOUT, bio_sectors(bio)); 887 } 888 889 bio_set_ioprio(bio); 890 submit_bio_noacct(bio); 891 } 892 EXPORT_SYMBOL(submit_bio); 893 894 /** 895 * bio_poll - poll for BIO completions 896 * @bio: bio to poll for 897 * @iob: batches of IO 898 * @flags: BLK_POLL_* flags that control the behavior 899 * 900 * Poll for completions on queue associated with the bio. Returns number of 901 * completed entries found. 902 * 903 * Note: the caller must either be the context that submitted @bio, or 904 * be in a RCU critical section to prevent freeing of @bio. 905 */ 906 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) 907 { 908 blk_qc_t cookie = READ_ONCE(bio->bi_cookie); 909 struct block_device *bdev; 910 struct request_queue *q; 911 int ret = 0; 912 913 bdev = READ_ONCE(bio->bi_bdev); 914 if (!bdev) 915 return 0; 916 917 q = bdev_get_queue(bdev); 918 if (cookie == BLK_QC_T_NONE || 919 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 920 return 0; 921 922 blk_flush_plug(current->plug, false); 923 924 /* 925 * We need to be able to enter a frozen queue, similar to how 926 * timeouts also need to do that. If that is blocked, then we can 927 * have pending IO when a queue freeze is started, and then the 928 * wait for the freeze to finish will wait for polled requests to 929 * timeout as the poller is preventer from entering the queue and 930 * completing them. As long as we prevent new IO from being queued, 931 * that should be all that matters. 932 */ 933 if (!percpu_ref_tryget(&q->q_usage_counter)) 934 return 0; 935 if (queue_is_mq(q)) { 936 ret = blk_mq_poll(q, cookie, iob, flags); 937 } else { 938 struct gendisk *disk = q->disk; 939 940 if (disk && disk->fops->poll_bio) 941 ret = disk->fops->poll_bio(bio, iob, flags); 942 } 943 blk_queue_exit(q); 944 return ret; 945 } 946 EXPORT_SYMBOL_GPL(bio_poll); 947 948 /* 949 * Helper to implement file_operations.iopoll. Requires the bio to be stored 950 * in iocb->private, and cleared before freeing the bio. 951 */ 952 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, 953 unsigned int flags) 954 { 955 struct bio *bio; 956 int ret = 0; 957 958 /* 959 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can 960 * point to a freshly allocated bio at this point. If that happens 961 * we have a few cases to consider: 962 * 963 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just 964 * simply nothing in this case 965 * 2) the bio points to a not poll enabled device. bio_poll will catch 966 * this and return 0 967 * 3) the bio points to a poll capable device, including but not 968 * limited to the one that the original bio pointed to. In this 969 * case we will call into the actual poll method and poll for I/O, 970 * even if we don't need to, but it won't cause harm either. 971 * 972 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev 973 * is still allocated. Because partitions hold a reference to the whole 974 * device bdev and thus disk, the disk is also still valid. Grabbing 975 * a reference to the queue in bio_poll() ensures the hctxs and requests 976 * are still valid as well. 977 */ 978 rcu_read_lock(); 979 bio = READ_ONCE(kiocb->private); 980 if (bio) 981 ret = bio_poll(bio, iob, flags); 982 rcu_read_unlock(); 983 984 return ret; 985 } 986 EXPORT_SYMBOL_GPL(iocb_bio_iopoll); 987 988 void update_io_ticks(struct block_device *part, unsigned long now, bool end) 989 { 990 unsigned long stamp; 991 again: 992 stamp = READ_ONCE(part->bd_stamp); 993 if (unlikely(time_after(now, stamp)) && 994 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) && 995 (end || part_in_flight(part))) 996 __part_stat_add(part, io_ticks, now - stamp); 997 998 if (bdev_is_partition(part)) { 999 part = bdev_whole(part); 1000 goto again; 1001 } 1002 } 1003 1004 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op, 1005 unsigned long start_time) 1006 { 1007 part_stat_lock(); 1008 update_io_ticks(bdev, start_time, false); 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 - start I/O accounting for bio based drivers 1018 * @bio: bio to start account for 1019 * 1020 * Returns the start time that should be passed back to bio_end_io_acct(). 1021 */ 1022 unsigned long bio_start_io_acct(struct bio *bio) 1023 { 1024 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies); 1025 } 1026 EXPORT_SYMBOL_GPL(bio_start_io_acct); 1027 1028 void bdev_end_io_acct(struct block_device *bdev, enum req_op op, 1029 unsigned int sectors, unsigned long start_time) 1030 { 1031 const int sgrp = op_stat_group(op); 1032 unsigned long now = READ_ONCE(jiffies); 1033 unsigned long duration = now - start_time; 1034 1035 part_stat_lock(); 1036 update_io_ticks(bdev, now, true); 1037 part_stat_inc(bdev, ios[sgrp]); 1038 part_stat_add(bdev, sectors[sgrp], sectors); 1039 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); 1040 part_stat_local_dec(bdev, in_flight[op_is_write(op)]); 1041 part_stat_unlock(); 1042 } 1043 EXPORT_SYMBOL(bdev_end_io_acct); 1044 1045 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, 1046 struct block_device *orig_bdev) 1047 { 1048 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time); 1049 } 1050 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); 1051 1052 /** 1053 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1054 * @q : the queue of the device being checked 1055 * 1056 * Description: 1057 * Check if underlying low-level drivers of a device are busy. 1058 * If the drivers want to export their busy state, they must set own 1059 * exporting function using blk_queue_lld_busy() first. 1060 * 1061 * Basically, this function is used only by request stacking drivers 1062 * to stop dispatching requests to underlying devices when underlying 1063 * devices are busy. This behavior helps more I/O merging on the queue 1064 * of the request stacking driver and prevents I/O throughput regression 1065 * on burst I/O load. 1066 * 1067 * Return: 1068 * 0 - Not busy (The request stacking driver should dispatch request) 1069 * 1 - Busy (The request stacking driver should stop dispatching request) 1070 */ 1071 int blk_lld_busy(struct request_queue *q) 1072 { 1073 if (queue_is_mq(q) && q->mq_ops->busy) 1074 return q->mq_ops->busy(q); 1075 1076 return 0; 1077 } 1078 EXPORT_SYMBOL_GPL(blk_lld_busy); 1079 1080 int kblockd_schedule_work(struct work_struct *work) 1081 { 1082 return queue_work(kblockd_workqueue, work); 1083 } 1084 EXPORT_SYMBOL(kblockd_schedule_work); 1085 1086 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1087 unsigned long delay) 1088 { 1089 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1090 } 1091 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1092 1093 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) 1094 { 1095 struct task_struct *tsk = current; 1096 1097 /* 1098 * If this is a nested plug, don't actually assign it. 1099 */ 1100 if (tsk->plug) 1101 return; 1102 1103 plug->cur_ktime = 0; 1104 plug->mq_list = NULL; 1105 plug->cached_rq = NULL; 1106 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); 1107 plug->rq_count = 0; 1108 plug->multiple_queues = false; 1109 plug->has_elevator = false; 1110 INIT_LIST_HEAD(&plug->cb_list); 1111 1112 /* 1113 * Store ordering should not be needed here, since a potential 1114 * preempt will imply a full memory barrier 1115 */ 1116 tsk->plug = plug; 1117 } 1118 1119 /** 1120 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1121 * @plug: The &struct blk_plug that needs to be initialized 1122 * 1123 * Description: 1124 * blk_start_plug() indicates to the block layer an intent by the caller 1125 * to submit multiple I/O requests in a batch. The block layer may use 1126 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1127 * is called. However, the block layer may choose to submit requests 1128 * before a call to blk_finish_plug() if the number of queued I/Os 1129 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1130 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1131 * the task schedules (see below). 1132 * 1133 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1134 * pending I/O should the task end up blocking between blk_start_plug() and 1135 * blk_finish_plug(). This is important from a performance perspective, but 1136 * also ensures that we don't deadlock. For instance, if the task is blocking 1137 * for a memory allocation, memory reclaim could end up wanting to free a 1138 * page belonging to that request that is currently residing in our private 1139 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1140 * this kind of deadlock. 1141 */ 1142 void blk_start_plug(struct blk_plug *plug) 1143 { 1144 blk_start_plug_nr_ios(plug, 1); 1145 } 1146 EXPORT_SYMBOL(blk_start_plug); 1147 1148 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1149 { 1150 LIST_HEAD(callbacks); 1151 1152 while (!list_empty(&plug->cb_list)) { 1153 list_splice_init(&plug->cb_list, &callbacks); 1154 1155 while (!list_empty(&callbacks)) { 1156 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1157 struct blk_plug_cb, 1158 list); 1159 list_del(&cb->list); 1160 cb->callback(cb, from_schedule); 1161 } 1162 } 1163 } 1164 1165 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1166 int size) 1167 { 1168 struct blk_plug *plug = current->plug; 1169 struct blk_plug_cb *cb; 1170 1171 if (!plug) 1172 return NULL; 1173 1174 list_for_each_entry(cb, &plug->cb_list, list) 1175 if (cb->callback == unplug && cb->data == data) 1176 return cb; 1177 1178 /* Not currently on the callback list */ 1179 BUG_ON(size < sizeof(*cb)); 1180 cb = kzalloc(size, GFP_ATOMIC); 1181 if (cb) { 1182 cb->data = data; 1183 cb->callback = unplug; 1184 list_add(&cb->list, &plug->cb_list); 1185 } 1186 return cb; 1187 } 1188 EXPORT_SYMBOL(blk_check_plugged); 1189 1190 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) 1191 { 1192 if (!list_empty(&plug->cb_list)) 1193 flush_plug_callbacks(plug, from_schedule); 1194 blk_mq_flush_plug_list(plug, from_schedule); 1195 /* 1196 * Unconditionally flush out cached requests, even if the unplug 1197 * event came from schedule. Since we know hold references to the 1198 * queue for cached requests, we don't want a blocked task holding 1199 * up a queue freeze/quiesce event. 1200 */ 1201 if (unlikely(!rq_list_empty(plug->cached_rq))) 1202 blk_mq_free_plug_rqs(plug); 1203 1204 plug->cur_ktime = 0; 1205 current->flags &= ~PF_BLOCK_TS; 1206 } 1207 1208 /** 1209 * blk_finish_plug - mark the end of a batch of submitted I/O 1210 * @plug: The &struct blk_plug passed to blk_start_plug() 1211 * 1212 * Description: 1213 * Indicate that a batch of I/O submissions is complete. This function 1214 * must be paired with an initial call to blk_start_plug(). The intent 1215 * is to allow the block layer to optimize I/O submission. See the 1216 * documentation for blk_start_plug() for more information. 1217 */ 1218 void blk_finish_plug(struct blk_plug *plug) 1219 { 1220 if (plug == current->plug) { 1221 __blk_flush_plug(plug, false); 1222 current->plug = NULL; 1223 } 1224 } 1225 EXPORT_SYMBOL(blk_finish_plug); 1226 1227 void blk_io_schedule(void) 1228 { 1229 /* Prevent hang_check timer from firing at us during very long I/O */ 1230 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1231 1232 if (timeout) 1233 io_schedule_timeout(timeout); 1234 else 1235 io_schedule(); 1236 } 1237 EXPORT_SYMBOL_GPL(blk_io_schedule); 1238 1239 int __init blk_dev_init(void) 1240 { 1241 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1242 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1243 sizeof_field(struct request, cmd_flags)); 1244 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1245 sizeof_field(struct bio, bi_opf)); 1246 1247 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1248 kblockd_workqueue = alloc_workqueue("kblockd", 1249 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 1250 if (!kblockd_workqueue) 1251 panic("Failed to create kblockd\n"); 1252 1253 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC); 1254 1255 blk_debugfs_root = debugfs_create_dir("block", NULL); 1256 1257 return 0; 1258 } 1259