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 53 struct dentry *blk_debugfs_root; 54 55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); 56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); 57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); 58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); 59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); 60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert); 61 62 static DEFINE_IDA(blk_queue_ida); 63 64 /* 65 * For queue allocation 66 */ 67 static struct kmem_cache *blk_requestq_cachep; 68 69 /* 70 * Controlling structure to kblockd 71 */ 72 static struct workqueue_struct *kblockd_workqueue; 73 74 /** 75 * blk_queue_flag_set - atomically set a queue flag 76 * @flag: flag to be set 77 * @q: request queue 78 */ 79 void blk_queue_flag_set(unsigned int flag, struct request_queue *q) 80 { 81 set_bit(flag, &q->queue_flags); 82 } 83 EXPORT_SYMBOL(blk_queue_flag_set); 84 85 /** 86 * blk_queue_flag_clear - atomically clear a queue flag 87 * @flag: flag to be cleared 88 * @q: request queue 89 */ 90 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) 91 { 92 clear_bit(flag, &q->queue_flags); 93 } 94 EXPORT_SYMBOL(blk_queue_flag_clear); 95 96 /** 97 * blk_queue_flag_test_and_set - atomically test and set a queue flag 98 * @flag: flag to be set 99 * @q: request queue 100 * 101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if 102 * the flag was already set. 103 */ 104 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) 105 { 106 return test_and_set_bit(flag, &q->queue_flags); 107 } 108 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); 109 110 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name 111 static const char *const blk_op_name[] = { 112 REQ_OP_NAME(READ), 113 REQ_OP_NAME(WRITE), 114 REQ_OP_NAME(FLUSH), 115 REQ_OP_NAME(DISCARD), 116 REQ_OP_NAME(SECURE_ERASE), 117 REQ_OP_NAME(ZONE_RESET), 118 REQ_OP_NAME(ZONE_RESET_ALL), 119 REQ_OP_NAME(ZONE_OPEN), 120 REQ_OP_NAME(ZONE_CLOSE), 121 REQ_OP_NAME(ZONE_FINISH), 122 REQ_OP_NAME(ZONE_APPEND), 123 REQ_OP_NAME(WRITE_ZEROES), 124 REQ_OP_NAME(DRV_IN), 125 REQ_OP_NAME(DRV_OUT), 126 }; 127 #undef REQ_OP_NAME 128 129 /** 130 * blk_op_str - Return string XXX in the REQ_OP_XXX. 131 * @op: REQ_OP_XXX. 132 * 133 * Description: Centralize block layer function to convert REQ_OP_XXX into 134 * string format. Useful in the debugging and tracing bio or request. For 135 * invalid REQ_OP_XXX it returns string "UNKNOWN". 136 */ 137 inline const char *blk_op_str(enum req_op op) 138 { 139 const char *op_str = "UNKNOWN"; 140 141 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) 142 op_str = blk_op_name[op]; 143 144 return op_str; 145 } 146 EXPORT_SYMBOL_GPL(blk_op_str); 147 148 static const struct { 149 int errno; 150 const char *name; 151 } blk_errors[] = { 152 [BLK_STS_OK] = { 0, "" }, 153 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, 154 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, 155 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, 156 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, 157 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, 158 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, 159 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, 160 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, 161 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, 162 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, 163 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, 164 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" }, 165 166 /* device mapper special case, should not leak out: */ 167 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, 168 169 /* zone device specific errors */ 170 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, 171 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, 172 173 /* everything else not covered above: */ 174 [BLK_STS_IOERR] = { -EIO, "I/O" }, 175 }; 176 177 blk_status_t errno_to_blk_status(int errno) 178 { 179 int i; 180 181 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { 182 if (blk_errors[i].errno == errno) 183 return (__force blk_status_t)i; 184 } 185 186 return BLK_STS_IOERR; 187 } 188 EXPORT_SYMBOL_GPL(errno_to_blk_status); 189 190 int blk_status_to_errno(blk_status_t status) 191 { 192 int idx = (__force int)status; 193 194 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 195 return -EIO; 196 return blk_errors[idx].errno; 197 } 198 EXPORT_SYMBOL_GPL(blk_status_to_errno); 199 200 const char *blk_status_to_str(blk_status_t status) 201 { 202 int idx = (__force int)status; 203 204 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 205 return "<null>"; 206 return blk_errors[idx].name; 207 } 208 209 /** 210 * blk_sync_queue - cancel any pending callbacks on a queue 211 * @q: the queue 212 * 213 * Description: 214 * The block layer may perform asynchronous callback activity 215 * on a queue, such as calling the unplug function after a timeout. 216 * A block device may call blk_sync_queue to ensure that any 217 * such activity is cancelled, thus allowing it to release resources 218 * that the callbacks might use. The caller must already have made sure 219 * that its ->submit_bio will not re-add plugging prior to calling 220 * this function. 221 * 222 * This function does not cancel any asynchronous activity arising 223 * out of elevator or throttling code. That would require elevator_exit() 224 * and blkcg_exit_queue() to be called with queue lock initialized. 225 * 226 */ 227 void blk_sync_queue(struct request_queue *q) 228 { 229 del_timer_sync(&q->timeout); 230 cancel_work_sync(&q->timeout_work); 231 } 232 EXPORT_SYMBOL(blk_sync_queue); 233 234 /** 235 * blk_set_pm_only - increment pm_only counter 236 * @q: request queue pointer 237 */ 238 void blk_set_pm_only(struct request_queue *q) 239 { 240 atomic_inc(&q->pm_only); 241 } 242 EXPORT_SYMBOL_GPL(blk_set_pm_only); 243 244 void blk_clear_pm_only(struct request_queue *q) 245 { 246 int pm_only; 247 248 pm_only = atomic_dec_return(&q->pm_only); 249 WARN_ON_ONCE(pm_only < 0); 250 if (pm_only == 0) 251 wake_up_all(&q->mq_freeze_wq); 252 } 253 EXPORT_SYMBOL_GPL(blk_clear_pm_only); 254 255 static void blk_free_queue_rcu(struct rcu_head *rcu_head) 256 { 257 struct request_queue *q = container_of(rcu_head, 258 struct request_queue, rcu_head); 259 260 percpu_ref_exit(&q->q_usage_counter); 261 kmem_cache_free(blk_requestq_cachep, q); 262 } 263 264 static void blk_free_queue(struct request_queue *q) 265 { 266 blk_free_queue_stats(q->stats); 267 if (queue_is_mq(q)) 268 blk_mq_release(q); 269 270 ida_free(&blk_queue_ida, q->id); 271 call_rcu(&q->rcu_head, blk_free_queue_rcu); 272 } 273 274 /** 275 * blk_put_queue - decrement the request_queue refcount 276 * @q: the request_queue structure to decrement the refcount for 277 * 278 * Decrements the refcount of the request_queue and free it when the refcount 279 * reaches 0. 280 */ 281 void blk_put_queue(struct request_queue *q) 282 { 283 if (refcount_dec_and_test(&q->refs)) 284 blk_free_queue(q); 285 } 286 EXPORT_SYMBOL(blk_put_queue); 287 288 void blk_queue_start_drain(struct request_queue *q) 289 { 290 /* 291 * When queue DYING flag is set, we need to block new req 292 * entering queue, so we call blk_freeze_queue_start() to 293 * prevent I/O from crossing blk_queue_enter(). 294 */ 295 blk_freeze_queue_start(q); 296 if (queue_is_mq(q)) 297 blk_mq_wake_waiters(q); 298 /* Make blk_queue_enter() reexamine the DYING flag. */ 299 wake_up_all(&q->mq_freeze_wq); 300 } 301 302 /** 303 * blk_queue_enter() - try to increase q->q_usage_counter 304 * @q: request queue pointer 305 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM 306 */ 307 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) 308 { 309 const bool pm = flags & BLK_MQ_REQ_PM; 310 311 while (!blk_try_enter_queue(q, pm)) { 312 if (flags & BLK_MQ_REQ_NOWAIT) 313 return -EAGAIN; 314 315 /* 316 * read pair of barrier in blk_freeze_queue_start(), we need to 317 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 318 * reading .mq_freeze_depth or queue dying flag, otherwise the 319 * following wait may never return if the two reads are 320 * reordered. 321 */ 322 smp_rmb(); 323 wait_event(q->mq_freeze_wq, 324 (!q->mq_freeze_depth && 325 blk_pm_resume_queue(pm, q)) || 326 blk_queue_dying(q)); 327 if (blk_queue_dying(q)) 328 return -ENODEV; 329 } 330 331 return 0; 332 } 333 334 int __bio_queue_enter(struct request_queue *q, struct bio *bio) 335 { 336 while (!blk_try_enter_queue(q, false)) { 337 struct gendisk *disk = bio->bi_bdev->bd_disk; 338 339 if (bio->bi_opf & REQ_NOWAIT) { 340 if (test_bit(GD_DEAD, &disk->state)) 341 goto dead; 342 bio_wouldblock_error(bio); 343 return -EAGAIN; 344 } 345 346 /* 347 * read pair of barrier in blk_freeze_queue_start(), we need to 348 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 349 * reading .mq_freeze_depth or queue dying flag, otherwise the 350 * following wait may never return if the two reads are 351 * reordered. 352 */ 353 smp_rmb(); 354 wait_event(q->mq_freeze_wq, 355 (!q->mq_freeze_depth && 356 blk_pm_resume_queue(false, q)) || 357 test_bit(GD_DEAD, &disk->state)); 358 if (test_bit(GD_DEAD, &disk->state)) 359 goto dead; 360 } 361 362 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 = from_timer(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(int node_id) 393 { 394 struct request_queue *q; 395 396 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO, 397 node_id); 398 if (!q) 399 return NULL; 400 401 q->last_merge = NULL; 402 403 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL); 404 if (q->id < 0) 405 goto fail_q; 406 407 q->stats = blk_alloc_queue_stats(); 408 if (!q->stats) 409 goto fail_id; 410 411 q->node = node_id; 412 413 atomic_set(&q->nr_active_requests_shared_tags, 0); 414 415 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); 416 INIT_WORK(&q->timeout_work, blk_timeout_work); 417 INIT_LIST_HEAD(&q->icq_list); 418 419 refcount_set(&q->refs, 1); 420 mutex_init(&q->debugfs_mutex); 421 mutex_init(&q->sysfs_lock); 422 mutex_init(&q->sysfs_dir_lock); 423 spin_lock_init(&q->queue_lock); 424 425 init_waitqueue_head(&q->mq_freeze_wq); 426 mutex_init(&q->mq_freeze_lock); 427 428 /* 429 * Init percpu_ref in atomic mode so that it's faster to shutdown. 430 * See blk_register_queue() for details. 431 */ 432 if (percpu_ref_init(&q->q_usage_counter, 433 blk_queue_usage_counter_release, 434 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) 435 goto fail_stats; 436 437 blk_set_default_limits(&q->limits); 438 q->nr_requests = BLKDEV_DEFAULT_RQ; 439 440 return q; 441 442 fail_stats: 443 blk_free_queue_stats(q->stats); 444 fail_id: 445 ida_free(&blk_queue_ida, q->id); 446 fail_q: 447 kmem_cache_free(blk_requestq_cachep, q); 448 return NULL; 449 } 450 451 /** 452 * blk_get_queue - increment the request_queue refcount 453 * @q: the request_queue structure to increment the refcount for 454 * 455 * Increment the refcount of the request_queue kobject. 456 * 457 * Context: Any context. 458 */ 459 bool blk_get_queue(struct request_queue *q) 460 { 461 if (unlikely(blk_queue_dying(q))) 462 return false; 463 refcount_inc(&q->refs); 464 return true; 465 } 466 EXPORT_SYMBOL(blk_get_queue); 467 468 #ifdef CONFIG_FAIL_MAKE_REQUEST 469 470 static DECLARE_FAULT_ATTR(fail_make_request); 471 472 static int __init setup_fail_make_request(char *str) 473 { 474 return setup_fault_attr(&fail_make_request, str); 475 } 476 __setup("fail_make_request=", setup_fail_make_request); 477 478 bool should_fail_request(struct block_device *part, unsigned int bytes) 479 { 480 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); 481 } 482 483 static int __init fail_make_request_debugfs(void) 484 { 485 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 486 NULL, &fail_make_request); 487 488 return PTR_ERR_OR_ZERO(dir); 489 } 490 491 late_initcall(fail_make_request_debugfs); 492 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 493 494 static inline void bio_check_ro(struct bio *bio) 495 { 496 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { 497 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) 498 return; 499 pr_warn("Trying to write to read-only block-device %pg\n", 500 bio->bi_bdev); 501 /* Older lvm-tools actually trigger this */ 502 } 503 } 504 505 static noinline int should_fail_bio(struct bio *bio) 506 { 507 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) 508 return -EIO; 509 return 0; 510 } 511 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); 512 513 /* 514 * Check whether this bio extends beyond the end of the device or partition. 515 * This may well happen - the kernel calls bread() without checking the size of 516 * the device, e.g., when mounting a file system. 517 */ 518 static inline int bio_check_eod(struct bio *bio) 519 { 520 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 521 unsigned int nr_sectors = bio_sectors(bio); 522 523 if (nr_sectors && maxsector && 524 (nr_sectors > maxsector || 525 bio->bi_iter.bi_sector > maxsector - nr_sectors)) { 526 pr_info_ratelimited("%s: attempt to access beyond end of device\n" 527 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", 528 current->comm, bio->bi_bdev, bio->bi_opf, 529 bio->bi_iter.bi_sector, nr_sectors, maxsector); 530 return -EIO; 531 } 532 return 0; 533 } 534 535 /* 536 * Remap block n of partition p to block n+start(p) of the disk. 537 */ 538 static int blk_partition_remap(struct bio *bio) 539 { 540 struct block_device *p = bio->bi_bdev; 541 542 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 543 return -EIO; 544 if (bio_sectors(bio)) { 545 bio->bi_iter.bi_sector += p->bd_start_sect; 546 trace_block_bio_remap(bio, p->bd_dev, 547 bio->bi_iter.bi_sector - 548 p->bd_start_sect); 549 } 550 bio_set_flag(bio, BIO_REMAPPED); 551 return 0; 552 } 553 554 /* 555 * Check write append to a zoned block device. 556 */ 557 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 558 struct bio *bio) 559 { 560 int nr_sectors = bio_sectors(bio); 561 562 /* Only applicable to zoned block devices */ 563 if (!bdev_is_zoned(bio->bi_bdev)) 564 return BLK_STS_NOTSUPP; 565 566 /* The bio sector must point to the start of a sequential zone */ 567 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector) || 568 !bio_zone_is_seq(bio)) 569 return BLK_STS_IOERR; 570 571 /* 572 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 573 * split and could result in non-contiguous sectors being written in 574 * different zones. 575 */ 576 if (nr_sectors > q->limits.chunk_sectors) 577 return BLK_STS_IOERR; 578 579 /* Make sure the BIO is small enough and will not get split */ 580 if (nr_sectors > q->limits.max_zone_append_sectors) 581 return BLK_STS_IOERR; 582 583 bio->bi_opf |= REQ_NOMERGE; 584 585 return BLK_STS_OK; 586 } 587 588 static void __submit_bio(struct bio *bio) 589 { 590 if (unlikely(!blk_crypto_bio_prep(&bio))) 591 return; 592 593 if (!bio->bi_bdev->bd_has_submit_bio) { 594 blk_mq_submit_bio(bio); 595 } else if (likely(bio_queue_enter(bio) == 0)) { 596 struct gendisk *disk = bio->bi_bdev->bd_disk; 597 598 disk->fops->submit_bio(bio); 599 blk_queue_exit(disk->queue); 600 } 601 } 602 603 /* 604 * The loop in this function may be a bit non-obvious, and so deserves some 605 * explanation: 606 * 607 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 608 * that), so we have a list with a single bio. 609 * - We pretend that we have just taken it off a longer list, so we assign 610 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 611 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 612 * bios through a recursive call to submit_bio_noacct. If it did, we find a 613 * non-NULL value in bio_list and re-enter the loop from the top. 614 * - In this case we really did just take the bio of the top of the list (no 615 * pretending) and so remove it from bio_list, and call into ->submit_bio() 616 * again. 617 * 618 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 619 * bio_list_on_stack[1] contains bios that were submitted before the current 620 * ->submit_bio, but that haven't been processed yet. 621 */ 622 static void __submit_bio_noacct(struct bio *bio) 623 { 624 struct bio_list bio_list_on_stack[2]; 625 626 BUG_ON(bio->bi_next); 627 628 bio_list_init(&bio_list_on_stack[0]); 629 current->bio_list = bio_list_on_stack; 630 631 do { 632 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 633 struct bio_list lower, same; 634 635 /* 636 * Create a fresh bio_list for all subordinate requests. 637 */ 638 bio_list_on_stack[1] = bio_list_on_stack[0]; 639 bio_list_init(&bio_list_on_stack[0]); 640 641 __submit_bio(bio); 642 643 /* 644 * Sort new bios into those for a lower level and those for the 645 * same level. 646 */ 647 bio_list_init(&lower); 648 bio_list_init(&same); 649 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 650 if (q == bdev_get_queue(bio->bi_bdev)) 651 bio_list_add(&same, bio); 652 else 653 bio_list_add(&lower, bio); 654 655 /* 656 * Now assemble so we handle the lowest level first. 657 */ 658 bio_list_merge(&bio_list_on_stack[0], &lower); 659 bio_list_merge(&bio_list_on_stack[0], &same); 660 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 661 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 662 663 current->bio_list = NULL; 664 } 665 666 static void __submit_bio_noacct_mq(struct bio *bio) 667 { 668 struct bio_list bio_list[2] = { }; 669 670 current->bio_list = bio_list; 671 672 do { 673 __submit_bio(bio); 674 } while ((bio = bio_list_pop(&bio_list[0]))); 675 676 current->bio_list = NULL; 677 } 678 679 void submit_bio_noacct_nocheck(struct bio *bio) 680 { 681 blk_cgroup_bio_start(bio); 682 blkcg_bio_issue_init(bio); 683 684 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 685 trace_block_bio_queue(bio); 686 /* 687 * Now that enqueuing has been traced, we need to trace 688 * completion as well. 689 */ 690 bio_set_flag(bio, BIO_TRACE_COMPLETION); 691 } 692 693 /* 694 * We only want one ->submit_bio to be active at a time, else stack 695 * usage with stacked devices could be a problem. Use current->bio_list 696 * to collect a list of requests submited by a ->submit_bio method while 697 * it is active, and then process them after it returned. 698 */ 699 if (current->bio_list) 700 bio_list_add(¤t->bio_list[0], bio); 701 else if (!bio->bi_bdev->bd_has_submit_bio) 702 __submit_bio_noacct_mq(bio); 703 else 704 __submit_bio_noacct(bio); 705 } 706 707 /** 708 * submit_bio_noacct - re-submit a bio to the block device layer for I/O 709 * @bio: The bio describing the location in memory and on the device. 710 * 711 * This is a version of submit_bio() that shall only be used for I/O that is 712 * resubmitted to lower level drivers by stacking block drivers. All file 713 * systems and other upper level users of the block layer should use 714 * submit_bio() instead. 715 */ 716 void submit_bio_noacct(struct bio *bio) 717 { 718 struct block_device *bdev = bio->bi_bdev; 719 struct request_queue *q = bdev_get_queue(bdev); 720 blk_status_t status = BLK_STS_IOERR; 721 struct blk_plug *plug; 722 723 might_sleep(); 724 725 plug = blk_mq_plug(bio); 726 if (plug && plug->nowait) 727 bio->bi_opf |= REQ_NOWAIT; 728 729 /* 730 * For a REQ_NOWAIT based request, return -EOPNOTSUPP 731 * if queue does not support NOWAIT. 732 */ 733 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev)) 734 goto not_supported; 735 736 if (should_fail_bio(bio)) 737 goto end_io; 738 bio_check_ro(bio); 739 if (!bio_flagged(bio, BIO_REMAPPED)) { 740 if (unlikely(bio_check_eod(bio))) 741 goto end_io; 742 if (bdev->bd_partno && unlikely(blk_partition_remap(bio))) 743 goto end_io; 744 } 745 746 /* 747 * Filter flush bio's early so that bio based drivers without flush 748 * support don't have to worry about them. 749 */ 750 if (op_is_flush(bio->bi_opf)) { 751 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE && 752 bio_op(bio) != REQ_OP_ZONE_APPEND)) 753 goto end_io; 754 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { 755 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 756 if (!bio_sectors(bio)) { 757 status = BLK_STS_OK; 758 goto end_io; 759 } 760 } 761 } 762 763 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 764 bio_clear_polled(bio); 765 766 switch (bio_op(bio)) { 767 case REQ_OP_DISCARD: 768 if (!bdev_max_discard_sectors(bdev)) 769 goto not_supported; 770 break; 771 case REQ_OP_SECURE_ERASE: 772 if (!bdev_max_secure_erase_sectors(bdev)) 773 goto not_supported; 774 break; 775 case REQ_OP_ZONE_APPEND: 776 status = blk_check_zone_append(q, bio); 777 if (status != BLK_STS_OK) 778 goto end_io; 779 break; 780 case REQ_OP_ZONE_RESET: 781 case REQ_OP_ZONE_OPEN: 782 case REQ_OP_ZONE_CLOSE: 783 case REQ_OP_ZONE_FINISH: 784 if (!bdev_is_zoned(bio->bi_bdev)) 785 goto not_supported; 786 break; 787 case REQ_OP_ZONE_RESET_ALL: 788 if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q)) 789 goto not_supported; 790 break; 791 case REQ_OP_WRITE_ZEROES: 792 if (!q->limits.max_write_zeroes_sectors) 793 goto not_supported; 794 break; 795 default: 796 break; 797 } 798 799 if (blk_throtl_bio(bio)) 800 return; 801 submit_bio_noacct_nocheck(bio); 802 return; 803 804 not_supported: 805 status = BLK_STS_NOTSUPP; 806 end_io: 807 bio->bi_status = status; 808 bio_endio(bio); 809 } 810 EXPORT_SYMBOL(submit_bio_noacct); 811 812 /** 813 * submit_bio - submit a bio to the block device layer for I/O 814 * @bio: The &struct bio which describes the I/O 815 * 816 * submit_bio() is used to submit I/O requests to block devices. It is passed a 817 * fully set up &struct bio that describes the I/O that needs to be done. The 818 * bio will be send to the device described by the bi_bdev field. 819 * 820 * The success/failure status of the request, along with notification of 821 * completion, is delivered asynchronously through the ->bi_end_io() callback 822 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has 823 * been called. 824 */ 825 void submit_bio(struct bio *bio) 826 { 827 if (bio_op(bio) == REQ_OP_READ) { 828 task_io_account_read(bio->bi_iter.bi_size); 829 count_vm_events(PGPGIN, bio_sectors(bio)); 830 } else if (bio_op(bio) == REQ_OP_WRITE) { 831 count_vm_events(PGPGOUT, bio_sectors(bio)); 832 } 833 834 submit_bio_noacct(bio); 835 } 836 EXPORT_SYMBOL(submit_bio); 837 838 /** 839 * bio_poll - poll for BIO completions 840 * @bio: bio to poll for 841 * @iob: batches of IO 842 * @flags: BLK_POLL_* flags that control the behavior 843 * 844 * Poll for completions on queue associated with the bio. Returns number of 845 * completed entries found. 846 * 847 * Note: the caller must either be the context that submitted @bio, or 848 * be in a RCU critical section to prevent freeing of @bio. 849 */ 850 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) 851 { 852 blk_qc_t cookie = READ_ONCE(bio->bi_cookie); 853 struct block_device *bdev; 854 struct request_queue *q; 855 int ret = 0; 856 857 bdev = READ_ONCE(bio->bi_bdev); 858 if (!bdev) 859 return 0; 860 861 q = bdev_get_queue(bdev); 862 if (cookie == BLK_QC_T_NONE || 863 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 864 return 0; 865 866 /* 867 * As the requests that require a zone lock are not plugged in the 868 * first place, directly accessing the plug instead of using 869 * blk_mq_plug() should not have any consequences during flushing for 870 * zoned devices. 871 */ 872 blk_flush_plug(current->plug, false); 873 874 /* 875 * We need to be able to enter a frozen queue, similar to how 876 * timeouts also need to do that. If that is blocked, then we can 877 * have pending IO when a queue freeze is started, and then the 878 * wait for the freeze to finish will wait for polled requests to 879 * timeout as the poller is preventer from entering the queue and 880 * completing them. As long as we prevent new IO from being queued, 881 * that should be all that matters. 882 */ 883 if (!percpu_ref_tryget(&q->q_usage_counter)) 884 return 0; 885 if (queue_is_mq(q)) { 886 ret = blk_mq_poll(q, cookie, iob, flags); 887 } else { 888 struct gendisk *disk = q->disk; 889 890 if (disk && disk->fops->poll_bio) 891 ret = disk->fops->poll_bio(bio, iob, flags); 892 } 893 blk_queue_exit(q); 894 return ret; 895 } 896 EXPORT_SYMBOL_GPL(bio_poll); 897 898 /* 899 * Helper to implement file_operations.iopoll. Requires the bio to be stored 900 * in iocb->private, and cleared before freeing the bio. 901 */ 902 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, 903 unsigned int flags) 904 { 905 struct bio *bio; 906 int ret = 0; 907 908 /* 909 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can 910 * point to a freshly allocated bio at this point. If that happens 911 * we have a few cases to consider: 912 * 913 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just 914 * simply nothing in this case 915 * 2) the bio points to a not poll enabled device. bio_poll will catch 916 * this and return 0 917 * 3) the bio points to a poll capable device, including but not 918 * limited to the one that the original bio pointed to. In this 919 * case we will call into the actual poll method and poll for I/O, 920 * even if we don't need to, but it won't cause harm either. 921 * 922 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev 923 * is still allocated. Because partitions hold a reference to the whole 924 * device bdev and thus disk, the disk is also still valid. Grabbing 925 * a reference to the queue in bio_poll() ensures the hctxs and requests 926 * are still valid as well. 927 */ 928 rcu_read_lock(); 929 bio = READ_ONCE(kiocb->private); 930 if (bio) 931 ret = bio_poll(bio, iob, flags); 932 rcu_read_unlock(); 933 934 return ret; 935 } 936 EXPORT_SYMBOL_GPL(iocb_bio_iopoll); 937 938 void update_io_ticks(struct block_device *part, unsigned long now, bool end) 939 { 940 unsigned long stamp; 941 again: 942 stamp = READ_ONCE(part->bd_stamp); 943 if (unlikely(time_after(now, stamp))) { 944 if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now))) 945 __part_stat_add(part, io_ticks, end ? now - stamp : 1); 946 } 947 if (part->bd_partno) { 948 part = bdev_whole(part); 949 goto again; 950 } 951 } 952 953 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op, 954 unsigned long start_time) 955 { 956 part_stat_lock(); 957 update_io_ticks(bdev, start_time, false); 958 part_stat_local_inc(bdev, in_flight[op_is_write(op)]); 959 part_stat_unlock(); 960 961 return start_time; 962 } 963 EXPORT_SYMBOL(bdev_start_io_acct); 964 965 /** 966 * bio_start_io_acct - start I/O accounting for bio based drivers 967 * @bio: bio to start account for 968 * 969 * Returns the start time that should be passed back to bio_end_io_acct(). 970 */ 971 unsigned long bio_start_io_acct(struct bio *bio) 972 { 973 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies); 974 } 975 EXPORT_SYMBOL_GPL(bio_start_io_acct); 976 977 void bdev_end_io_acct(struct block_device *bdev, enum req_op op, 978 unsigned int sectors, unsigned long start_time) 979 { 980 const int sgrp = op_stat_group(op); 981 unsigned long now = READ_ONCE(jiffies); 982 unsigned long duration = now - start_time; 983 984 part_stat_lock(); 985 update_io_ticks(bdev, now, true); 986 part_stat_inc(bdev, ios[sgrp]); 987 part_stat_add(bdev, sectors[sgrp], sectors); 988 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); 989 part_stat_local_dec(bdev, in_flight[op_is_write(op)]); 990 part_stat_unlock(); 991 } 992 EXPORT_SYMBOL(bdev_end_io_acct); 993 994 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, 995 struct block_device *orig_bdev) 996 { 997 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time); 998 } 999 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); 1000 1001 /** 1002 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1003 * @q : the queue of the device being checked 1004 * 1005 * Description: 1006 * Check if underlying low-level drivers of a device are busy. 1007 * If the drivers want to export their busy state, they must set own 1008 * exporting function using blk_queue_lld_busy() first. 1009 * 1010 * Basically, this function is used only by request stacking drivers 1011 * to stop dispatching requests to underlying devices when underlying 1012 * devices are busy. This behavior helps more I/O merging on the queue 1013 * of the request stacking driver and prevents I/O throughput regression 1014 * on burst I/O load. 1015 * 1016 * Return: 1017 * 0 - Not busy (The request stacking driver should dispatch request) 1018 * 1 - Busy (The request stacking driver should stop dispatching request) 1019 */ 1020 int blk_lld_busy(struct request_queue *q) 1021 { 1022 if (queue_is_mq(q) && q->mq_ops->busy) 1023 return q->mq_ops->busy(q); 1024 1025 return 0; 1026 } 1027 EXPORT_SYMBOL_GPL(blk_lld_busy); 1028 1029 int kblockd_schedule_work(struct work_struct *work) 1030 { 1031 return queue_work(kblockd_workqueue, work); 1032 } 1033 EXPORT_SYMBOL(kblockd_schedule_work); 1034 1035 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1036 unsigned long delay) 1037 { 1038 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1039 } 1040 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1041 1042 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) 1043 { 1044 struct task_struct *tsk = current; 1045 1046 /* 1047 * If this is a nested plug, don't actually assign it. 1048 */ 1049 if (tsk->plug) 1050 return; 1051 1052 plug->mq_list = NULL; 1053 plug->cached_rq = NULL; 1054 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); 1055 plug->rq_count = 0; 1056 plug->multiple_queues = false; 1057 plug->has_elevator = false; 1058 plug->nowait = false; 1059 INIT_LIST_HEAD(&plug->cb_list); 1060 1061 /* 1062 * Store ordering should not be needed here, since a potential 1063 * preempt will imply a full memory barrier 1064 */ 1065 tsk->plug = plug; 1066 } 1067 1068 /** 1069 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1070 * @plug: The &struct blk_plug that needs to be initialized 1071 * 1072 * Description: 1073 * blk_start_plug() indicates to the block layer an intent by the caller 1074 * to submit multiple I/O requests in a batch. The block layer may use 1075 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1076 * is called. However, the block layer may choose to submit requests 1077 * before a call to blk_finish_plug() if the number of queued I/Os 1078 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1079 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1080 * the task schedules (see below). 1081 * 1082 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1083 * pending I/O should the task end up blocking between blk_start_plug() and 1084 * blk_finish_plug(). This is important from a performance perspective, but 1085 * also ensures that we don't deadlock. For instance, if the task is blocking 1086 * for a memory allocation, memory reclaim could end up wanting to free a 1087 * page belonging to that request that is currently residing in our private 1088 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1089 * this kind of deadlock. 1090 */ 1091 void blk_start_plug(struct blk_plug *plug) 1092 { 1093 blk_start_plug_nr_ios(plug, 1); 1094 } 1095 EXPORT_SYMBOL(blk_start_plug); 1096 1097 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1098 { 1099 LIST_HEAD(callbacks); 1100 1101 while (!list_empty(&plug->cb_list)) { 1102 list_splice_init(&plug->cb_list, &callbacks); 1103 1104 while (!list_empty(&callbacks)) { 1105 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1106 struct blk_plug_cb, 1107 list); 1108 list_del(&cb->list); 1109 cb->callback(cb, from_schedule); 1110 } 1111 } 1112 } 1113 1114 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1115 int size) 1116 { 1117 struct blk_plug *plug = current->plug; 1118 struct blk_plug_cb *cb; 1119 1120 if (!plug) 1121 return NULL; 1122 1123 list_for_each_entry(cb, &plug->cb_list, list) 1124 if (cb->callback == unplug && cb->data == data) 1125 return cb; 1126 1127 /* Not currently on the callback list */ 1128 BUG_ON(size < sizeof(*cb)); 1129 cb = kzalloc(size, GFP_ATOMIC); 1130 if (cb) { 1131 cb->data = data; 1132 cb->callback = unplug; 1133 list_add(&cb->list, &plug->cb_list); 1134 } 1135 return cb; 1136 } 1137 EXPORT_SYMBOL(blk_check_plugged); 1138 1139 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) 1140 { 1141 if (!list_empty(&plug->cb_list)) 1142 flush_plug_callbacks(plug, from_schedule); 1143 if (!rq_list_empty(plug->mq_list)) 1144 blk_mq_flush_plug_list(plug, from_schedule); 1145 /* 1146 * Unconditionally flush out cached requests, even if the unplug 1147 * event came from schedule. Since we know hold references to the 1148 * queue for cached requests, we don't want a blocked task holding 1149 * up a queue freeze/quiesce event. 1150 */ 1151 if (unlikely(!rq_list_empty(plug->cached_rq))) 1152 blk_mq_free_plug_rqs(plug); 1153 } 1154 1155 /** 1156 * blk_finish_plug - mark the end of a batch of submitted I/O 1157 * @plug: The &struct blk_plug passed to blk_start_plug() 1158 * 1159 * Description: 1160 * Indicate that a batch of I/O submissions is complete. This function 1161 * must be paired with an initial call to blk_start_plug(). The intent 1162 * is to allow the block layer to optimize I/O submission. See the 1163 * documentation for blk_start_plug() for more information. 1164 */ 1165 void blk_finish_plug(struct blk_plug *plug) 1166 { 1167 if (plug == current->plug) { 1168 __blk_flush_plug(plug, false); 1169 current->plug = NULL; 1170 } 1171 } 1172 EXPORT_SYMBOL(blk_finish_plug); 1173 1174 void blk_io_schedule(void) 1175 { 1176 /* Prevent hang_check timer from firing at us during very long I/O */ 1177 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1178 1179 if (timeout) 1180 io_schedule_timeout(timeout); 1181 else 1182 io_schedule(); 1183 } 1184 EXPORT_SYMBOL_GPL(blk_io_schedule); 1185 1186 int __init blk_dev_init(void) 1187 { 1188 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1189 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1190 sizeof_field(struct request, cmd_flags)); 1191 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1192 sizeof_field(struct bio, bi_opf)); 1193 1194 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1195 kblockd_workqueue = alloc_workqueue("kblockd", 1196 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 1197 if (!kblockd_workqueue) 1198 panic("Failed to create kblockd\n"); 1199 1200 blk_requestq_cachep = kmem_cache_create("request_queue", 1201 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 1202 1203 blk_debugfs_root = debugfs_create_dir("block", NULL); 1204 1205 return 0; 1206 } 1207