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_RESV_CONFLICT] = { -EBADE, "reservation conflict" }, 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 /* Command duration limit device-side timeout */ 174 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" }, 175 176 /* everything else not covered above: */ 177 [BLK_STS_IOERR] = { -EIO, "I/O" }, 178 }; 179 180 blk_status_t errno_to_blk_status(int errno) 181 { 182 int i; 183 184 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { 185 if (blk_errors[i].errno == errno) 186 return (__force blk_status_t)i; 187 } 188 189 return BLK_STS_IOERR; 190 } 191 EXPORT_SYMBOL_GPL(errno_to_blk_status); 192 193 int blk_status_to_errno(blk_status_t status) 194 { 195 int idx = (__force int)status; 196 197 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 198 return -EIO; 199 return blk_errors[idx].errno; 200 } 201 EXPORT_SYMBOL_GPL(blk_status_to_errno); 202 203 const char *blk_status_to_str(blk_status_t status) 204 { 205 int idx = (__force int)status; 206 207 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 208 return "<null>"; 209 return blk_errors[idx].name; 210 } 211 EXPORT_SYMBOL_GPL(blk_status_to_str); 212 213 /** 214 * blk_sync_queue - cancel any pending callbacks on a queue 215 * @q: the queue 216 * 217 * Description: 218 * The block layer may perform asynchronous callback activity 219 * on a queue, such as calling the unplug function after a timeout. 220 * A block device may call blk_sync_queue to ensure that any 221 * such activity is cancelled, thus allowing it to release resources 222 * that the callbacks might use. The caller must already have made sure 223 * that its ->submit_bio will not re-add plugging prior to calling 224 * this function. 225 * 226 * This function does not cancel any asynchronous activity arising 227 * out of elevator or throttling code. That would require elevator_exit() 228 * and blkcg_exit_queue() to be called with queue lock initialized. 229 * 230 */ 231 void blk_sync_queue(struct request_queue *q) 232 { 233 del_timer_sync(&q->timeout); 234 cancel_work_sync(&q->timeout_work); 235 } 236 EXPORT_SYMBOL(blk_sync_queue); 237 238 /** 239 * blk_set_pm_only - increment pm_only counter 240 * @q: request queue pointer 241 */ 242 void blk_set_pm_only(struct request_queue *q) 243 { 244 atomic_inc(&q->pm_only); 245 } 246 EXPORT_SYMBOL_GPL(blk_set_pm_only); 247 248 void blk_clear_pm_only(struct request_queue *q) 249 { 250 int pm_only; 251 252 pm_only = atomic_dec_return(&q->pm_only); 253 WARN_ON_ONCE(pm_only < 0); 254 if (pm_only == 0) 255 wake_up_all(&q->mq_freeze_wq); 256 } 257 EXPORT_SYMBOL_GPL(blk_clear_pm_only); 258 259 static void blk_free_queue_rcu(struct rcu_head *rcu_head) 260 { 261 struct request_queue *q = container_of(rcu_head, 262 struct request_queue, rcu_head); 263 264 percpu_ref_exit(&q->q_usage_counter); 265 kmem_cache_free(blk_requestq_cachep, q); 266 } 267 268 static void blk_free_queue(struct request_queue *q) 269 { 270 blk_free_queue_stats(q->stats); 271 if (queue_is_mq(q)) 272 blk_mq_release(q); 273 274 ida_free(&blk_queue_ida, q->id); 275 call_rcu(&q->rcu_head, blk_free_queue_rcu); 276 } 277 278 /** 279 * blk_put_queue - decrement the request_queue refcount 280 * @q: the request_queue structure to decrement the refcount for 281 * 282 * Decrements the refcount of the request_queue and free it when the refcount 283 * reaches 0. 284 */ 285 void blk_put_queue(struct request_queue *q) 286 { 287 if (refcount_dec_and_test(&q->refs)) 288 blk_free_queue(q); 289 } 290 EXPORT_SYMBOL(blk_put_queue); 291 292 void blk_queue_start_drain(struct request_queue *q) 293 { 294 /* 295 * When queue DYING flag is set, we need to block new req 296 * entering queue, so we call blk_freeze_queue_start() to 297 * prevent I/O from crossing blk_queue_enter(). 298 */ 299 blk_freeze_queue_start(q); 300 if (queue_is_mq(q)) 301 blk_mq_wake_waiters(q); 302 /* Make blk_queue_enter() reexamine the DYING flag. */ 303 wake_up_all(&q->mq_freeze_wq); 304 } 305 306 /** 307 * blk_queue_enter() - try to increase q->q_usage_counter 308 * @q: request queue pointer 309 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM 310 */ 311 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) 312 { 313 const bool pm = flags & BLK_MQ_REQ_PM; 314 315 while (!blk_try_enter_queue(q, pm)) { 316 if (flags & BLK_MQ_REQ_NOWAIT) 317 return -EAGAIN; 318 319 /* 320 * read pair of barrier in blk_freeze_queue_start(), we need to 321 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 322 * reading .mq_freeze_depth or queue dying flag, otherwise the 323 * following wait may never return if the two reads are 324 * reordered. 325 */ 326 smp_rmb(); 327 wait_event(q->mq_freeze_wq, 328 (!q->mq_freeze_depth && 329 blk_pm_resume_queue(pm, q)) || 330 blk_queue_dying(q)); 331 if (blk_queue_dying(q)) 332 return -ENODEV; 333 } 334 335 return 0; 336 } 337 338 int __bio_queue_enter(struct request_queue *q, struct bio *bio) 339 { 340 while (!blk_try_enter_queue(q, false)) { 341 struct gendisk *disk = bio->bi_bdev->bd_disk; 342 343 if (bio->bi_opf & REQ_NOWAIT) { 344 if (test_bit(GD_DEAD, &disk->state)) 345 goto dead; 346 bio_wouldblock_error(bio); 347 return -EAGAIN; 348 } 349 350 /* 351 * read pair of barrier in blk_freeze_queue_start(), we need to 352 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and 353 * reading .mq_freeze_depth or queue dying flag, otherwise the 354 * following wait may never return if the two reads are 355 * reordered. 356 */ 357 smp_rmb(); 358 wait_event(q->mq_freeze_wq, 359 (!q->mq_freeze_depth && 360 blk_pm_resume_queue(false, q)) || 361 test_bit(GD_DEAD, &disk->state)); 362 if (test_bit(GD_DEAD, &disk->state)) 363 goto dead; 364 } 365 366 return 0; 367 dead: 368 bio_io_error(bio); 369 return -ENODEV; 370 } 371 372 void blk_queue_exit(struct request_queue *q) 373 { 374 percpu_ref_put(&q->q_usage_counter); 375 } 376 377 static void blk_queue_usage_counter_release(struct percpu_ref *ref) 378 { 379 struct request_queue *q = 380 container_of(ref, struct request_queue, q_usage_counter); 381 382 wake_up_all(&q->mq_freeze_wq); 383 } 384 385 static void blk_rq_timed_out_timer(struct timer_list *t) 386 { 387 struct request_queue *q = from_timer(q, t, timeout); 388 389 kblockd_schedule_work(&q->timeout_work); 390 } 391 392 static void blk_timeout_work(struct work_struct *work) 393 { 394 } 395 396 struct request_queue *blk_alloc_queue(int node_id) 397 { 398 struct request_queue *q; 399 400 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO, 401 node_id); 402 if (!q) 403 return NULL; 404 405 q->last_merge = NULL; 406 407 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL); 408 if (q->id < 0) 409 goto fail_q; 410 411 q->stats = blk_alloc_queue_stats(); 412 if (!q->stats) 413 goto fail_id; 414 415 q->node = node_id; 416 417 atomic_set(&q->nr_active_requests_shared_tags, 0); 418 419 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); 420 INIT_WORK(&q->timeout_work, blk_timeout_work); 421 INIT_LIST_HEAD(&q->icq_list); 422 423 refcount_set(&q->refs, 1); 424 mutex_init(&q->debugfs_mutex); 425 mutex_init(&q->sysfs_lock); 426 mutex_init(&q->sysfs_dir_lock); 427 mutex_init(&q->rq_qos_mutex); 428 spin_lock_init(&q->queue_lock); 429 430 init_waitqueue_head(&q->mq_freeze_wq); 431 mutex_init(&q->mq_freeze_lock); 432 433 /* 434 * Init percpu_ref in atomic mode so that it's faster to shutdown. 435 * See blk_register_queue() for details. 436 */ 437 if (percpu_ref_init(&q->q_usage_counter, 438 blk_queue_usage_counter_release, 439 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) 440 goto fail_stats; 441 442 blk_set_default_limits(&q->limits); 443 q->nr_requests = BLKDEV_DEFAULT_RQ; 444 445 return q; 446 447 fail_stats: 448 blk_free_queue_stats(q->stats); 449 fail_id: 450 ida_free(&blk_queue_ida, q->id); 451 fail_q: 452 kmem_cache_free(blk_requestq_cachep, q); 453 return NULL; 454 } 455 456 /** 457 * blk_get_queue - increment the request_queue refcount 458 * @q: the request_queue structure to increment the refcount for 459 * 460 * Increment the refcount of the request_queue kobject. 461 * 462 * Context: Any context. 463 */ 464 bool blk_get_queue(struct request_queue *q) 465 { 466 if (unlikely(blk_queue_dying(q))) 467 return false; 468 refcount_inc(&q->refs); 469 return true; 470 } 471 EXPORT_SYMBOL(blk_get_queue); 472 473 #ifdef CONFIG_FAIL_MAKE_REQUEST 474 475 static DECLARE_FAULT_ATTR(fail_make_request); 476 477 static int __init setup_fail_make_request(char *str) 478 { 479 return setup_fault_attr(&fail_make_request, str); 480 } 481 __setup("fail_make_request=", setup_fail_make_request); 482 483 bool should_fail_request(struct block_device *part, unsigned int bytes) 484 { 485 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); 486 } 487 488 static int __init fail_make_request_debugfs(void) 489 { 490 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 491 NULL, &fail_make_request); 492 493 return PTR_ERR_OR_ZERO(dir); 494 } 495 496 late_initcall(fail_make_request_debugfs); 497 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 498 499 static inline void bio_check_ro(struct bio *bio) 500 { 501 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { 502 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) 503 return; 504 505 if (bio->bi_bdev->bd_ro_warned) 506 return; 507 508 bio->bi_bdev->bd_ro_warned = true; 509 /* 510 * Use ioctl to set underlying disk of raid/dm to read-only 511 * will trigger this. 512 */ 513 pr_warn("Trying to write to read-only block-device %pg\n", 514 bio->bi_bdev); 515 } 516 } 517 518 static noinline int should_fail_bio(struct bio *bio) 519 { 520 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) 521 return -EIO; 522 return 0; 523 } 524 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); 525 526 /* 527 * Check whether this bio extends beyond the end of the device or partition. 528 * This may well happen - the kernel calls bread() without checking the size of 529 * the device, e.g., when mounting a file system. 530 */ 531 static inline int bio_check_eod(struct bio *bio) 532 { 533 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 534 unsigned int nr_sectors = bio_sectors(bio); 535 536 if (nr_sectors && 537 (nr_sectors > maxsector || 538 bio->bi_iter.bi_sector > maxsector - nr_sectors)) { 539 pr_info_ratelimited("%s: attempt to access beyond end of device\n" 540 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", 541 current->comm, bio->bi_bdev, bio->bi_opf, 542 bio->bi_iter.bi_sector, nr_sectors, maxsector); 543 return -EIO; 544 } 545 return 0; 546 } 547 548 /* 549 * Remap block n of partition p to block n+start(p) of the disk. 550 */ 551 static int blk_partition_remap(struct bio *bio) 552 { 553 struct block_device *p = bio->bi_bdev; 554 555 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 556 return -EIO; 557 if (bio_sectors(bio)) { 558 bio->bi_iter.bi_sector += p->bd_start_sect; 559 trace_block_bio_remap(bio, p->bd_dev, 560 bio->bi_iter.bi_sector - 561 p->bd_start_sect); 562 } 563 bio_set_flag(bio, BIO_REMAPPED); 564 return 0; 565 } 566 567 /* 568 * Check write append to a zoned block device. 569 */ 570 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 571 struct bio *bio) 572 { 573 int nr_sectors = bio_sectors(bio); 574 575 /* Only applicable to zoned block devices */ 576 if (!bdev_is_zoned(bio->bi_bdev)) 577 return BLK_STS_NOTSUPP; 578 579 /* The bio sector must point to the start of a sequential zone */ 580 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector) || 581 !bio_zone_is_seq(bio)) 582 return BLK_STS_IOERR; 583 584 /* 585 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 586 * split and could result in non-contiguous sectors being written in 587 * different zones. 588 */ 589 if (nr_sectors > q->limits.chunk_sectors) 590 return BLK_STS_IOERR; 591 592 /* Make sure the BIO is small enough and will not get split */ 593 if (nr_sectors > q->limits.max_zone_append_sectors) 594 return BLK_STS_IOERR; 595 596 bio->bi_opf |= REQ_NOMERGE; 597 598 return BLK_STS_OK; 599 } 600 601 static void __submit_bio(struct bio *bio) 602 { 603 if (unlikely(!blk_crypto_bio_prep(&bio))) 604 return; 605 606 if (!bio->bi_bdev->bd_has_submit_bio) { 607 blk_mq_submit_bio(bio); 608 } else if (likely(bio_queue_enter(bio) == 0)) { 609 struct gendisk *disk = bio->bi_bdev->bd_disk; 610 611 disk->fops->submit_bio(bio); 612 blk_queue_exit(disk->queue); 613 } 614 } 615 616 /* 617 * The loop in this function may be a bit non-obvious, and so deserves some 618 * explanation: 619 * 620 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 621 * that), so we have a list with a single bio. 622 * - We pretend that we have just taken it off a longer list, so we assign 623 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 624 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 625 * bios through a recursive call to submit_bio_noacct. If it did, we find a 626 * non-NULL value in bio_list and re-enter the loop from the top. 627 * - In this case we really did just take the bio of the top of the list (no 628 * pretending) and so remove it from bio_list, and call into ->submit_bio() 629 * again. 630 * 631 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 632 * bio_list_on_stack[1] contains bios that were submitted before the current 633 * ->submit_bio, but that haven't been processed yet. 634 */ 635 static void __submit_bio_noacct(struct bio *bio) 636 { 637 struct bio_list bio_list_on_stack[2]; 638 639 BUG_ON(bio->bi_next); 640 641 bio_list_init(&bio_list_on_stack[0]); 642 current->bio_list = bio_list_on_stack; 643 644 do { 645 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 646 struct bio_list lower, same; 647 648 /* 649 * Create a fresh bio_list for all subordinate requests. 650 */ 651 bio_list_on_stack[1] = bio_list_on_stack[0]; 652 bio_list_init(&bio_list_on_stack[0]); 653 654 __submit_bio(bio); 655 656 /* 657 * Sort new bios into those for a lower level and those for the 658 * same level. 659 */ 660 bio_list_init(&lower); 661 bio_list_init(&same); 662 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 663 if (q == bdev_get_queue(bio->bi_bdev)) 664 bio_list_add(&same, bio); 665 else 666 bio_list_add(&lower, bio); 667 668 /* 669 * Now assemble so we handle the lowest level first. 670 */ 671 bio_list_merge(&bio_list_on_stack[0], &lower); 672 bio_list_merge(&bio_list_on_stack[0], &same); 673 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 674 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 675 676 current->bio_list = NULL; 677 } 678 679 static void __submit_bio_noacct_mq(struct bio *bio) 680 { 681 struct bio_list bio_list[2] = { }; 682 683 current->bio_list = bio_list; 684 685 do { 686 __submit_bio(bio); 687 } while ((bio = bio_list_pop(&bio_list[0]))); 688 689 current->bio_list = NULL; 690 } 691 692 void submit_bio_noacct_nocheck(struct bio *bio) 693 { 694 blk_cgroup_bio_start(bio); 695 blkcg_bio_issue_init(bio); 696 697 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 698 trace_block_bio_queue(bio); 699 /* 700 * Now that enqueuing has been traced, we need to trace 701 * completion as well. 702 */ 703 bio_set_flag(bio, BIO_TRACE_COMPLETION); 704 } 705 706 /* 707 * We only want one ->submit_bio to be active at a time, else stack 708 * usage with stacked devices could be a problem. Use current->bio_list 709 * to collect a list of requests submited by a ->submit_bio method while 710 * it is active, and then process them after it returned. 711 */ 712 if (current->bio_list) 713 bio_list_add(¤t->bio_list[0], bio); 714 else if (!bio->bi_bdev->bd_has_submit_bio) 715 __submit_bio_noacct_mq(bio); 716 else 717 __submit_bio_noacct(bio); 718 } 719 720 /** 721 * submit_bio_noacct - re-submit a bio to the block device layer for I/O 722 * @bio: The bio describing the location in memory and on the device. 723 * 724 * This is a version of submit_bio() that shall only be used for I/O that is 725 * resubmitted to lower level drivers by stacking block drivers. All file 726 * systems and other upper level users of the block layer should use 727 * submit_bio() instead. 728 */ 729 void submit_bio_noacct(struct bio *bio) 730 { 731 struct block_device *bdev = bio->bi_bdev; 732 struct request_queue *q = bdev_get_queue(bdev); 733 blk_status_t status = BLK_STS_IOERR; 734 735 might_sleep(); 736 737 /* 738 * For a REQ_NOWAIT based request, return -EOPNOTSUPP 739 * if queue does not support NOWAIT. 740 */ 741 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev)) 742 goto not_supported; 743 744 if (should_fail_bio(bio)) 745 goto end_io; 746 bio_check_ro(bio); 747 if (!bio_flagged(bio, BIO_REMAPPED)) { 748 if (unlikely(bio_check_eod(bio))) 749 goto end_io; 750 if (bdev->bd_partno && unlikely(blk_partition_remap(bio))) 751 goto end_io; 752 } 753 754 /* 755 * Filter flush bio's early so that bio based drivers without flush 756 * support don't have to worry about them. 757 */ 758 if (op_is_flush(bio->bi_opf)) { 759 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE && 760 bio_op(bio) != REQ_OP_ZONE_APPEND)) 761 goto end_io; 762 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { 763 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 764 if (!bio_sectors(bio)) { 765 status = BLK_STS_OK; 766 goto end_io; 767 } 768 } 769 } 770 771 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 772 bio_clear_polled(bio); 773 774 switch (bio_op(bio)) { 775 case REQ_OP_READ: 776 case REQ_OP_WRITE: 777 break; 778 case REQ_OP_FLUSH: 779 /* 780 * REQ_OP_FLUSH can't be submitted through bios, it is only 781 * synthetized in struct request by the flush state machine. 782 */ 783 goto not_supported; 784 case REQ_OP_DISCARD: 785 if (!bdev_max_discard_sectors(bdev)) 786 goto not_supported; 787 break; 788 case REQ_OP_SECURE_ERASE: 789 if (!bdev_max_secure_erase_sectors(bdev)) 790 goto not_supported; 791 break; 792 case REQ_OP_ZONE_APPEND: 793 status = blk_check_zone_append(q, bio); 794 if (status != BLK_STS_OK) 795 goto end_io; 796 break; 797 case REQ_OP_WRITE_ZEROES: 798 if (!q->limits.max_write_zeroes_sectors) 799 goto not_supported; 800 break; 801 case REQ_OP_ZONE_RESET: 802 case REQ_OP_ZONE_OPEN: 803 case REQ_OP_ZONE_CLOSE: 804 case REQ_OP_ZONE_FINISH: 805 if (!bdev_is_zoned(bio->bi_bdev)) 806 goto not_supported; 807 break; 808 case REQ_OP_ZONE_RESET_ALL: 809 if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q)) 810 goto not_supported; 811 break; 812 case REQ_OP_DRV_IN: 813 case REQ_OP_DRV_OUT: 814 /* 815 * Driver private operations are only used with passthrough 816 * requests. 817 */ 818 fallthrough; 819 default: 820 goto not_supported; 821 } 822 823 if (blk_throtl_bio(bio)) 824 return; 825 submit_bio_noacct_nocheck(bio); 826 return; 827 828 not_supported: 829 status = BLK_STS_NOTSUPP; 830 end_io: 831 bio->bi_status = status; 832 bio_endio(bio); 833 } 834 EXPORT_SYMBOL(submit_bio_noacct); 835 836 /** 837 * submit_bio - submit a bio to the block device layer for I/O 838 * @bio: The &struct bio which describes the I/O 839 * 840 * submit_bio() is used to submit I/O requests to block devices. It is passed a 841 * fully set up &struct bio that describes the I/O that needs to be done. The 842 * bio will be send to the device described by the bi_bdev field. 843 * 844 * The success/failure status of the request, along with notification of 845 * completion, is delivered asynchronously through the ->bi_end_io() callback 846 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has 847 * been called. 848 */ 849 void submit_bio(struct bio *bio) 850 { 851 if (bio_op(bio) == REQ_OP_READ) { 852 task_io_account_read(bio->bi_iter.bi_size); 853 count_vm_events(PGPGIN, bio_sectors(bio)); 854 } else if (bio_op(bio) == REQ_OP_WRITE) { 855 count_vm_events(PGPGOUT, bio_sectors(bio)); 856 } 857 858 submit_bio_noacct(bio); 859 } 860 EXPORT_SYMBOL(submit_bio); 861 862 /** 863 * bio_poll - poll for BIO completions 864 * @bio: bio to poll for 865 * @iob: batches of IO 866 * @flags: BLK_POLL_* flags that control the behavior 867 * 868 * Poll for completions on queue associated with the bio. Returns number of 869 * completed entries found. 870 * 871 * Note: the caller must either be the context that submitted @bio, or 872 * be in a RCU critical section to prevent freeing of @bio. 873 */ 874 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) 875 { 876 blk_qc_t cookie = READ_ONCE(bio->bi_cookie); 877 struct block_device *bdev; 878 struct request_queue *q; 879 int ret = 0; 880 881 bdev = READ_ONCE(bio->bi_bdev); 882 if (!bdev) 883 return 0; 884 885 q = bdev_get_queue(bdev); 886 if (cookie == BLK_QC_T_NONE || 887 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 888 return 0; 889 890 /* 891 * As the requests that require a zone lock are not plugged in the 892 * first place, directly accessing the plug instead of using 893 * blk_mq_plug() should not have any consequences during flushing for 894 * zoned devices. 895 */ 896 blk_flush_plug(current->plug, false); 897 898 /* 899 * We need to be able to enter a frozen queue, similar to how 900 * timeouts also need to do that. If that is blocked, then we can 901 * have pending IO when a queue freeze is started, and then the 902 * wait for the freeze to finish will wait for polled requests to 903 * timeout as the poller is preventer from entering the queue and 904 * completing them. As long as we prevent new IO from being queued, 905 * that should be all that matters. 906 */ 907 if (!percpu_ref_tryget(&q->q_usage_counter)) 908 return 0; 909 if (queue_is_mq(q)) { 910 ret = blk_mq_poll(q, cookie, iob, flags); 911 } else { 912 struct gendisk *disk = q->disk; 913 914 if (disk && disk->fops->poll_bio) 915 ret = disk->fops->poll_bio(bio, iob, flags); 916 } 917 blk_queue_exit(q); 918 return ret; 919 } 920 EXPORT_SYMBOL_GPL(bio_poll); 921 922 /* 923 * Helper to implement file_operations.iopoll. Requires the bio to be stored 924 * in iocb->private, and cleared before freeing the bio. 925 */ 926 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, 927 unsigned int flags) 928 { 929 struct bio *bio; 930 int ret = 0; 931 932 /* 933 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can 934 * point to a freshly allocated bio at this point. If that happens 935 * we have a few cases to consider: 936 * 937 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just 938 * simply nothing in this case 939 * 2) the bio points to a not poll enabled device. bio_poll will catch 940 * this and return 0 941 * 3) the bio points to a poll capable device, including but not 942 * limited to the one that the original bio pointed to. In this 943 * case we will call into the actual poll method and poll for I/O, 944 * even if we don't need to, but it won't cause harm either. 945 * 946 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev 947 * is still allocated. Because partitions hold a reference to the whole 948 * device bdev and thus disk, the disk is also still valid. Grabbing 949 * a reference to the queue in bio_poll() ensures the hctxs and requests 950 * are still valid as well. 951 */ 952 rcu_read_lock(); 953 bio = READ_ONCE(kiocb->private); 954 if (bio) 955 ret = bio_poll(bio, iob, flags); 956 rcu_read_unlock(); 957 958 return ret; 959 } 960 EXPORT_SYMBOL_GPL(iocb_bio_iopoll); 961 962 void update_io_ticks(struct block_device *part, unsigned long now, bool end) 963 { 964 unsigned long stamp; 965 again: 966 stamp = READ_ONCE(part->bd_stamp); 967 if (unlikely(time_after(now, stamp))) { 968 if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now))) 969 __part_stat_add(part, io_ticks, end ? now - stamp : 1); 970 } 971 if (part->bd_partno) { 972 part = bdev_whole(part); 973 goto again; 974 } 975 } 976 977 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op, 978 unsigned long start_time) 979 { 980 part_stat_lock(); 981 update_io_ticks(bdev, start_time, false); 982 part_stat_local_inc(bdev, in_flight[op_is_write(op)]); 983 part_stat_unlock(); 984 985 return start_time; 986 } 987 EXPORT_SYMBOL(bdev_start_io_acct); 988 989 /** 990 * bio_start_io_acct - start I/O accounting for bio based drivers 991 * @bio: bio to start account for 992 * 993 * Returns the start time that should be passed back to bio_end_io_acct(). 994 */ 995 unsigned long bio_start_io_acct(struct bio *bio) 996 { 997 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies); 998 } 999 EXPORT_SYMBOL_GPL(bio_start_io_acct); 1000 1001 void bdev_end_io_acct(struct block_device *bdev, enum req_op op, 1002 unsigned int sectors, unsigned long start_time) 1003 { 1004 const int sgrp = op_stat_group(op); 1005 unsigned long now = READ_ONCE(jiffies); 1006 unsigned long duration = now - start_time; 1007 1008 part_stat_lock(); 1009 update_io_ticks(bdev, now, true); 1010 part_stat_inc(bdev, ios[sgrp]); 1011 part_stat_add(bdev, sectors[sgrp], sectors); 1012 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); 1013 part_stat_local_dec(bdev, in_flight[op_is_write(op)]); 1014 part_stat_unlock(); 1015 } 1016 EXPORT_SYMBOL(bdev_end_io_acct); 1017 1018 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, 1019 struct block_device *orig_bdev) 1020 { 1021 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time); 1022 } 1023 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); 1024 1025 /** 1026 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1027 * @q : the queue of the device being checked 1028 * 1029 * Description: 1030 * Check if underlying low-level drivers of a device are busy. 1031 * If the drivers want to export their busy state, they must set own 1032 * exporting function using blk_queue_lld_busy() first. 1033 * 1034 * Basically, this function is used only by request stacking drivers 1035 * to stop dispatching requests to underlying devices when underlying 1036 * devices are busy. This behavior helps more I/O merging on the queue 1037 * of the request stacking driver and prevents I/O throughput regression 1038 * on burst I/O load. 1039 * 1040 * Return: 1041 * 0 - Not busy (The request stacking driver should dispatch request) 1042 * 1 - Busy (The request stacking driver should stop dispatching request) 1043 */ 1044 int blk_lld_busy(struct request_queue *q) 1045 { 1046 if (queue_is_mq(q) && q->mq_ops->busy) 1047 return q->mq_ops->busy(q); 1048 1049 return 0; 1050 } 1051 EXPORT_SYMBOL_GPL(blk_lld_busy); 1052 1053 int kblockd_schedule_work(struct work_struct *work) 1054 { 1055 return queue_work(kblockd_workqueue, work); 1056 } 1057 EXPORT_SYMBOL(kblockd_schedule_work); 1058 1059 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1060 unsigned long delay) 1061 { 1062 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1063 } 1064 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1065 1066 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) 1067 { 1068 struct task_struct *tsk = current; 1069 1070 /* 1071 * If this is a nested plug, don't actually assign it. 1072 */ 1073 if (tsk->plug) 1074 return; 1075 1076 plug->mq_list = NULL; 1077 plug->cached_rq = NULL; 1078 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); 1079 plug->rq_count = 0; 1080 plug->multiple_queues = false; 1081 plug->has_elevator = false; 1082 INIT_LIST_HEAD(&plug->cb_list); 1083 1084 /* 1085 * Store ordering should not be needed here, since a potential 1086 * preempt will imply a full memory barrier 1087 */ 1088 tsk->plug = plug; 1089 } 1090 1091 /** 1092 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1093 * @plug: The &struct blk_plug that needs to be initialized 1094 * 1095 * Description: 1096 * blk_start_plug() indicates to the block layer an intent by the caller 1097 * to submit multiple I/O requests in a batch. The block layer may use 1098 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1099 * is called. However, the block layer may choose to submit requests 1100 * before a call to blk_finish_plug() if the number of queued I/Os 1101 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1102 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1103 * the task schedules (see below). 1104 * 1105 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1106 * pending I/O should the task end up blocking between blk_start_plug() and 1107 * blk_finish_plug(). This is important from a performance perspective, but 1108 * also ensures that we don't deadlock. For instance, if the task is blocking 1109 * for a memory allocation, memory reclaim could end up wanting to free a 1110 * page belonging to that request that is currently residing in our private 1111 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1112 * this kind of deadlock. 1113 */ 1114 void blk_start_plug(struct blk_plug *plug) 1115 { 1116 blk_start_plug_nr_ios(plug, 1); 1117 } 1118 EXPORT_SYMBOL(blk_start_plug); 1119 1120 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1121 { 1122 LIST_HEAD(callbacks); 1123 1124 while (!list_empty(&plug->cb_list)) { 1125 list_splice_init(&plug->cb_list, &callbacks); 1126 1127 while (!list_empty(&callbacks)) { 1128 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1129 struct blk_plug_cb, 1130 list); 1131 list_del(&cb->list); 1132 cb->callback(cb, from_schedule); 1133 } 1134 } 1135 } 1136 1137 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1138 int size) 1139 { 1140 struct blk_plug *plug = current->plug; 1141 struct blk_plug_cb *cb; 1142 1143 if (!plug) 1144 return NULL; 1145 1146 list_for_each_entry(cb, &plug->cb_list, list) 1147 if (cb->callback == unplug && cb->data == data) 1148 return cb; 1149 1150 /* Not currently on the callback list */ 1151 BUG_ON(size < sizeof(*cb)); 1152 cb = kzalloc(size, GFP_ATOMIC); 1153 if (cb) { 1154 cb->data = data; 1155 cb->callback = unplug; 1156 list_add(&cb->list, &plug->cb_list); 1157 } 1158 return cb; 1159 } 1160 EXPORT_SYMBOL(blk_check_plugged); 1161 1162 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) 1163 { 1164 if (!list_empty(&plug->cb_list)) 1165 flush_plug_callbacks(plug, from_schedule); 1166 blk_mq_flush_plug_list(plug, from_schedule); 1167 /* 1168 * Unconditionally flush out cached requests, even if the unplug 1169 * event came from schedule. Since we know hold references to the 1170 * queue for cached requests, we don't want a blocked task holding 1171 * up a queue freeze/quiesce event. 1172 */ 1173 if (unlikely(!rq_list_empty(plug->cached_rq))) 1174 blk_mq_free_plug_rqs(plug); 1175 } 1176 1177 /** 1178 * blk_finish_plug - mark the end of a batch of submitted I/O 1179 * @plug: The &struct blk_plug passed to blk_start_plug() 1180 * 1181 * Description: 1182 * Indicate that a batch of I/O submissions is complete. This function 1183 * must be paired with an initial call to blk_start_plug(). The intent 1184 * is to allow the block layer to optimize I/O submission. See the 1185 * documentation for blk_start_plug() for more information. 1186 */ 1187 void blk_finish_plug(struct blk_plug *plug) 1188 { 1189 if (plug == current->plug) { 1190 __blk_flush_plug(plug, false); 1191 current->plug = NULL; 1192 } 1193 } 1194 EXPORT_SYMBOL(blk_finish_plug); 1195 1196 void blk_io_schedule(void) 1197 { 1198 /* Prevent hang_check timer from firing at us during very long I/O */ 1199 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1200 1201 if (timeout) 1202 io_schedule_timeout(timeout); 1203 else 1204 io_schedule(); 1205 } 1206 EXPORT_SYMBOL_GPL(blk_io_schedule); 1207 1208 int __init blk_dev_init(void) 1209 { 1210 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1211 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1212 sizeof_field(struct request, cmd_flags)); 1213 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1214 sizeof_field(struct bio, bi_opf)); 1215 1216 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1217 kblockd_workqueue = alloc_workqueue("kblockd", 1218 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 1219 if (!kblockd_workqueue) 1220 panic("Failed to create kblockd\n"); 1221 1222 blk_requestq_cachep = kmem_cache_create("request_queue", 1223 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 1224 1225 blk_debugfs_root = debugfs_create_dir("block", NULL); 1226 1227 return 0; 1228 } 1229