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 /* 446 * Init percpu_ref in atomic mode so that it's faster to shutdown. 447 * See blk_register_queue() for details. 448 */ 449 error = percpu_ref_init(&q->q_usage_counter, 450 blk_queue_usage_counter_release, 451 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL); 452 if (error) 453 goto fail_stats; 454 455 q->nr_requests = BLKDEV_DEFAULT_RQ; 456 457 return q; 458 459 fail_stats: 460 blk_free_queue_stats(q->stats); 461 fail_id: 462 ida_free(&blk_queue_ida, q->id); 463 fail_q: 464 kmem_cache_free(blk_requestq_cachep, q); 465 return ERR_PTR(error); 466 } 467 468 /** 469 * blk_get_queue - increment the request_queue refcount 470 * @q: the request_queue structure to increment the refcount for 471 * 472 * Increment the refcount of the request_queue kobject. 473 * 474 * Context: Any context. 475 */ 476 bool blk_get_queue(struct request_queue *q) 477 { 478 if (unlikely(blk_queue_dying(q))) 479 return false; 480 refcount_inc(&q->refs); 481 return true; 482 } 483 EXPORT_SYMBOL(blk_get_queue); 484 485 #ifdef CONFIG_FAIL_MAKE_REQUEST 486 487 static DECLARE_FAULT_ATTR(fail_make_request); 488 489 static int __init setup_fail_make_request(char *str) 490 { 491 return setup_fault_attr(&fail_make_request, str); 492 } 493 __setup("fail_make_request=", setup_fail_make_request); 494 495 bool should_fail_request(struct block_device *part, unsigned int bytes) 496 { 497 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); 498 } 499 500 static int __init fail_make_request_debugfs(void) 501 { 502 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 503 NULL, &fail_make_request); 504 505 return PTR_ERR_OR_ZERO(dir); 506 } 507 508 late_initcall(fail_make_request_debugfs); 509 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 510 511 static inline void bio_check_ro(struct bio *bio) 512 { 513 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { 514 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) 515 return; 516 517 if (bio->bi_bdev->bd_ro_warned) 518 return; 519 520 bio->bi_bdev->bd_ro_warned = true; 521 /* 522 * Use ioctl to set underlying disk of raid/dm to read-only 523 * will trigger this. 524 */ 525 pr_warn("Trying to write to read-only block-device %pg\n", 526 bio->bi_bdev); 527 } 528 } 529 530 static noinline int should_fail_bio(struct bio *bio) 531 { 532 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) 533 return -EIO; 534 return 0; 535 } 536 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); 537 538 /* 539 * Check whether this bio extends beyond the end of the device or partition. 540 * This may well happen - the kernel calls bread() without checking the size of 541 * the device, e.g., when mounting a file system. 542 */ 543 static inline int bio_check_eod(struct bio *bio) 544 { 545 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 546 unsigned int nr_sectors = bio_sectors(bio); 547 548 if (nr_sectors && 549 (nr_sectors > maxsector || 550 bio->bi_iter.bi_sector > maxsector - nr_sectors)) { 551 pr_info_ratelimited("%s: attempt to access beyond end of device\n" 552 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", 553 current->comm, bio->bi_bdev, bio->bi_opf, 554 bio->bi_iter.bi_sector, nr_sectors, maxsector); 555 return -EIO; 556 } 557 return 0; 558 } 559 560 /* 561 * Remap block n of partition p to block n+start(p) of the disk. 562 */ 563 static int blk_partition_remap(struct bio *bio) 564 { 565 struct block_device *p = bio->bi_bdev; 566 567 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 568 return -EIO; 569 if (bio_sectors(bio)) { 570 bio->bi_iter.bi_sector += p->bd_start_sect; 571 trace_block_bio_remap(bio, p->bd_dev, 572 bio->bi_iter.bi_sector - 573 p->bd_start_sect); 574 } 575 bio_set_flag(bio, BIO_REMAPPED); 576 return 0; 577 } 578 579 /* 580 * Check write append to a zoned block device. 581 */ 582 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 583 struct bio *bio) 584 { 585 int nr_sectors = bio_sectors(bio); 586 587 /* Only applicable to zoned block devices */ 588 if (!bdev_is_zoned(bio->bi_bdev)) 589 return BLK_STS_NOTSUPP; 590 591 /* The bio sector must point to the start of a sequential zone */ 592 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector) || 593 !bio_zone_is_seq(bio)) 594 return BLK_STS_IOERR; 595 596 /* 597 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 598 * split and could result in non-contiguous sectors being written in 599 * different zones. 600 */ 601 if (nr_sectors > q->limits.chunk_sectors) 602 return BLK_STS_IOERR; 603 604 /* Make sure the BIO is small enough and will not get split */ 605 if (nr_sectors > q->limits.max_zone_append_sectors) 606 return BLK_STS_IOERR; 607 608 bio->bi_opf |= REQ_NOMERGE; 609 610 return BLK_STS_OK; 611 } 612 613 static void __submit_bio(struct bio *bio) 614 { 615 if (unlikely(!blk_crypto_bio_prep(&bio))) 616 return; 617 618 if (!bio->bi_bdev->bd_has_submit_bio) { 619 blk_mq_submit_bio(bio); 620 } else if (likely(bio_queue_enter(bio) == 0)) { 621 struct gendisk *disk = bio->bi_bdev->bd_disk; 622 623 disk->fops->submit_bio(bio); 624 blk_queue_exit(disk->queue); 625 } 626 } 627 628 /* 629 * The loop in this function may be a bit non-obvious, and so deserves some 630 * explanation: 631 * 632 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 633 * that), so we have a list with a single bio. 634 * - We pretend that we have just taken it off a longer list, so we assign 635 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 636 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 637 * bios through a recursive call to submit_bio_noacct. If it did, we find a 638 * non-NULL value in bio_list and re-enter the loop from the top. 639 * - In this case we really did just take the bio of the top of the list (no 640 * pretending) and so remove it from bio_list, and call into ->submit_bio() 641 * again. 642 * 643 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 644 * bio_list_on_stack[1] contains bios that were submitted before the current 645 * ->submit_bio, but that haven't been processed yet. 646 */ 647 static void __submit_bio_noacct(struct bio *bio) 648 { 649 struct bio_list bio_list_on_stack[2]; 650 651 BUG_ON(bio->bi_next); 652 653 bio_list_init(&bio_list_on_stack[0]); 654 current->bio_list = bio_list_on_stack; 655 656 do { 657 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 658 struct bio_list lower, same; 659 660 /* 661 * Create a fresh bio_list for all subordinate requests. 662 */ 663 bio_list_on_stack[1] = bio_list_on_stack[0]; 664 bio_list_init(&bio_list_on_stack[0]); 665 666 __submit_bio(bio); 667 668 /* 669 * Sort new bios into those for a lower level and those for the 670 * same level. 671 */ 672 bio_list_init(&lower); 673 bio_list_init(&same); 674 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 675 if (q == bdev_get_queue(bio->bi_bdev)) 676 bio_list_add(&same, bio); 677 else 678 bio_list_add(&lower, bio); 679 680 /* 681 * Now assemble so we handle the lowest level first. 682 */ 683 bio_list_merge(&bio_list_on_stack[0], &lower); 684 bio_list_merge(&bio_list_on_stack[0], &same); 685 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 686 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 687 688 current->bio_list = NULL; 689 } 690 691 static void __submit_bio_noacct_mq(struct bio *bio) 692 { 693 struct bio_list bio_list[2] = { }; 694 695 current->bio_list = bio_list; 696 697 do { 698 __submit_bio(bio); 699 } while ((bio = bio_list_pop(&bio_list[0]))); 700 701 current->bio_list = NULL; 702 } 703 704 void submit_bio_noacct_nocheck(struct bio *bio) 705 { 706 blk_cgroup_bio_start(bio); 707 blkcg_bio_issue_init(bio); 708 709 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 710 trace_block_bio_queue(bio); 711 /* 712 * Now that enqueuing has been traced, we need to trace 713 * completion as well. 714 */ 715 bio_set_flag(bio, BIO_TRACE_COMPLETION); 716 } 717 718 /* 719 * We only want one ->submit_bio to be active at a time, else stack 720 * usage with stacked devices could be a problem. Use current->bio_list 721 * to collect a list of requests submited by a ->submit_bio method while 722 * it is active, and then process them after it returned. 723 */ 724 if (current->bio_list) 725 bio_list_add(¤t->bio_list[0], bio); 726 else if (!bio->bi_bdev->bd_has_submit_bio) 727 __submit_bio_noacct_mq(bio); 728 else 729 __submit_bio_noacct(bio); 730 } 731 732 /** 733 * submit_bio_noacct - re-submit a bio to the block device layer for I/O 734 * @bio: The bio describing the location in memory and on the device. 735 * 736 * This is a version of submit_bio() that shall only be used for I/O that is 737 * resubmitted to lower level drivers by stacking block drivers. All file 738 * systems and other upper level users of the block layer should use 739 * submit_bio() instead. 740 */ 741 void submit_bio_noacct(struct bio *bio) 742 { 743 struct block_device *bdev = bio->bi_bdev; 744 struct request_queue *q = bdev_get_queue(bdev); 745 blk_status_t status = BLK_STS_IOERR; 746 747 might_sleep(); 748 749 /* 750 * For a REQ_NOWAIT based request, return -EOPNOTSUPP 751 * if queue does not support NOWAIT. 752 */ 753 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev)) 754 goto not_supported; 755 756 if (should_fail_bio(bio)) 757 goto end_io; 758 bio_check_ro(bio); 759 if (!bio_flagged(bio, BIO_REMAPPED)) { 760 if (unlikely(bio_check_eod(bio))) 761 goto end_io; 762 if (bdev->bd_partno && unlikely(blk_partition_remap(bio))) 763 goto end_io; 764 } 765 766 /* 767 * Filter flush bio's early so that bio based drivers without flush 768 * support don't have to worry about them. 769 */ 770 if (op_is_flush(bio->bi_opf)) { 771 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE && 772 bio_op(bio) != REQ_OP_ZONE_APPEND)) 773 goto end_io; 774 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { 775 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 776 if (!bio_sectors(bio)) { 777 status = BLK_STS_OK; 778 goto end_io; 779 } 780 } 781 } 782 783 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 784 bio_clear_polled(bio); 785 786 switch (bio_op(bio)) { 787 case REQ_OP_READ: 788 case REQ_OP_WRITE: 789 break; 790 case REQ_OP_FLUSH: 791 /* 792 * REQ_OP_FLUSH can't be submitted through bios, it is only 793 * synthetized in struct request by the flush state machine. 794 */ 795 goto not_supported; 796 case REQ_OP_DISCARD: 797 if (!bdev_max_discard_sectors(bdev)) 798 goto not_supported; 799 break; 800 case REQ_OP_SECURE_ERASE: 801 if (!bdev_max_secure_erase_sectors(bdev)) 802 goto not_supported; 803 break; 804 case REQ_OP_ZONE_APPEND: 805 status = blk_check_zone_append(q, bio); 806 if (status != BLK_STS_OK) 807 goto end_io; 808 break; 809 case REQ_OP_WRITE_ZEROES: 810 if (!q->limits.max_write_zeroes_sectors) 811 goto not_supported; 812 break; 813 case REQ_OP_ZONE_RESET: 814 case REQ_OP_ZONE_OPEN: 815 case REQ_OP_ZONE_CLOSE: 816 case REQ_OP_ZONE_FINISH: 817 if (!bdev_is_zoned(bio->bi_bdev)) 818 goto not_supported; 819 break; 820 case REQ_OP_ZONE_RESET_ALL: 821 if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q)) 822 goto not_supported; 823 break; 824 case REQ_OP_DRV_IN: 825 case REQ_OP_DRV_OUT: 826 /* 827 * Driver private operations are only used with passthrough 828 * requests. 829 */ 830 fallthrough; 831 default: 832 goto not_supported; 833 } 834 835 if (blk_throtl_bio(bio)) 836 return; 837 submit_bio_noacct_nocheck(bio); 838 return; 839 840 not_supported: 841 status = BLK_STS_NOTSUPP; 842 end_io: 843 bio->bi_status = status; 844 bio_endio(bio); 845 } 846 EXPORT_SYMBOL(submit_bio_noacct); 847 848 static void bio_set_ioprio(struct bio *bio) 849 { 850 /* Nobody set ioprio so far? Initialize it based on task's nice value */ 851 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE) 852 bio->bi_ioprio = get_current_ioprio(); 853 blkcg_set_ioprio(bio); 854 } 855 856 /** 857 * submit_bio - submit a bio to the block device layer for I/O 858 * @bio: The &struct bio which describes the I/O 859 * 860 * submit_bio() is used to submit I/O requests to block devices. It is passed a 861 * fully set up &struct bio that describes the I/O that needs to be done. The 862 * bio will be send to the device described by the bi_bdev field. 863 * 864 * The success/failure status of the request, along with notification of 865 * completion, is delivered asynchronously through the ->bi_end_io() callback 866 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has 867 * been called. 868 */ 869 void submit_bio(struct bio *bio) 870 { 871 if (bio_op(bio) == REQ_OP_READ) { 872 task_io_account_read(bio->bi_iter.bi_size); 873 count_vm_events(PGPGIN, bio_sectors(bio)); 874 } else if (bio_op(bio) == REQ_OP_WRITE) { 875 count_vm_events(PGPGOUT, bio_sectors(bio)); 876 } 877 878 bio_set_ioprio(bio); 879 submit_bio_noacct(bio); 880 } 881 EXPORT_SYMBOL(submit_bio); 882 883 /** 884 * bio_poll - poll for BIO completions 885 * @bio: bio to poll for 886 * @iob: batches of IO 887 * @flags: BLK_POLL_* flags that control the behavior 888 * 889 * Poll for completions on queue associated with the bio. Returns number of 890 * completed entries found. 891 * 892 * Note: the caller must either be the context that submitted @bio, or 893 * be in a RCU critical section to prevent freeing of @bio. 894 */ 895 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) 896 { 897 blk_qc_t cookie = READ_ONCE(bio->bi_cookie); 898 struct block_device *bdev; 899 struct request_queue *q; 900 int ret = 0; 901 902 bdev = READ_ONCE(bio->bi_bdev); 903 if (!bdev) 904 return 0; 905 906 q = bdev_get_queue(bdev); 907 if (cookie == BLK_QC_T_NONE || 908 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 909 return 0; 910 911 /* 912 * As the requests that require a zone lock are not plugged in the 913 * first place, directly accessing the plug instead of using 914 * blk_mq_plug() should not have any consequences during flushing for 915 * zoned devices. 916 */ 917 blk_flush_plug(current->plug, false); 918 919 /* 920 * We need to be able to enter a frozen queue, similar to how 921 * timeouts also need to do that. If that is blocked, then we can 922 * have pending IO when a queue freeze is started, and then the 923 * wait for the freeze to finish will wait for polled requests to 924 * timeout as the poller is preventer from entering the queue and 925 * completing them. As long as we prevent new IO from being queued, 926 * that should be all that matters. 927 */ 928 if (!percpu_ref_tryget(&q->q_usage_counter)) 929 return 0; 930 if (queue_is_mq(q)) { 931 ret = blk_mq_poll(q, cookie, iob, flags); 932 } else { 933 struct gendisk *disk = q->disk; 934 935 if (disk && disk->fops->poll_bio) 936 ret = disk->fops->poll_bio(bio, iob, flags); 937 } 938 blk_queue_exit(q); 939 return ret; 940 } 941 EXPORT_SYMBOL_GPL(bio_poll); 942 943 /* 944 * Helper to implement file_operations.iopoll. Requires the bio to be stored 945 * in iocb->private, and cleared before freeing the bio. 946 */ 947 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, 948 unsigned int flags) 949 { 950 struct bio *bio; 951 int ret = 0; 952 953 /* 954 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can 955 * point to a freshly allocated bio at this point. If that happens 956 * we have a few cases to consider: 957 * 958 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just 959 * simply nothing in this case 960 * 2) the bio points to a not poll enabled device. bio_poll will catch 961 * this and return 0 962 * 3) the bio points to a poll capable device, including but not 963 * limited to the one that the original bio pointed to. In this 964 * case we will call into the actual poll method and poll for I/O, 965 * even if we don't need to, but it won't cause harm either. 966 * 967 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev 968 * is still allocated. Because partitions hold a reference to the whole 969 * device bdev and thus disk, the disk is also still valid. Grabbing 970 * a reference to the queue in bio_poll() ensures the hctxs and requests 971 * are still valid as well. 972 */ 973 rcu_read_lock(); 974 bio = READ_ONCE(kiocb->private); 975 if (bio) 976 ret = bio_poll(bio, iob, flags); 977 rcu_read_unlock(); 978 979 return ret; 980 } 981 EXPORT_SYMBOL_GPL(iocb_bio_iopoll); 982 983 void update_io_ticks(struct block_device *part, unsigned long now, bool end) 984 { 985 unsigned long stamp; 986 again: 987 stamp = READ_ONCE(part->bd_stamp); 988 if (unlikely(time_after(now, stamp))) { 989 if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now))) 990 __part_stat_add(part, io_ticks, end ? now - stamp : 1); 991 } 992 if (part->bd_partno) { 993 part = bdev_whole(part); 994 goto again; 995 } 996 } 997 998 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op, 999 unsigned long start_time) 1000 { 1001 part_stat_lock(); 1002 update_io_ticks(bdev, start_time, false); 1003 part_stat_local_inc(bdev, in_flight[op_is_write(op)]); 1004 part_stat_unlock(); 1005 1006 return start_time; 1007 } 1008 EXPORT_SYMBOL(bdev_start_io_acct); 1009 1010 /** 1011 * bio_start_io_acct - start I/O accounting for bio based drivers 1012 * @bio: bio to start account for 1013 * 1014 * Returns the start time that should be passed back to bio_end_io_acct(). 1015 */ 1016 unsigned long bio_start_io_acct(struct bio *bio) 1017 { 1018 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies); 1019 } 1020 EXPORT_SYMBOL_GPL(bio_start_io_acct); 1021 1022 void bdev_end_io_acct(struct block_device *bdev, enum req_op op, 1023 unsigned int sectors, unsigned long start_time) 1024 { 1025 const int sgrp = op_stat_group(op); 1026 unsigned long now = READ_ONCE(jiffies); 1027 unsigned long duration = now - start_time; 1028 1029 part_stat_lock(); 1030 update_io_ticks(bdev, now, true); 1031 part_stat_inc(bdev, ios[sgrp]); 1032 part_stat_add(bdev, sectors[sgrp], sectors); 1033 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); 1034 part_stat_local_dec(bdev, in_flight[op_is_write(op)]); 1035 part_stat_unlock(); 1036 } 1037 EXPORT_SYMBOL(bdev_end_io_acct); 1038 1039 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, 1040 struct block_device *orig_bdev) 1041 { 1042 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time); 1043 } 1044 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); 1045 1046 /** 1047 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1048 * @q : the queue of the device being checked 1049 * 1050 * Description: 1051 * Check if underlying low-level drivers of a device are busy. 1052 * If the drivers want to export their busy state, they must set own 1053 * exporting function using blk_queue_lld_busy() first. 1054 * 1055 * Basically, this function is used only by request stacking drivers 1056 * to stop dispatching requests to underlying devices when underlying 1057 * devices are busy. This behavior helps more I/O merging on the queue 1058 * of the request stacking driver and prevents I/O throughput regression 1059 * on burst I/O load. 1060 * 1061 * Return: 1062 * 0 - Not busy (The request stacking driver should dispatch request) 1063 * 1 - Busy (The request stacking driver should stop dispatching request) 1064 */ 1065 int blk_lld_busy(struct request_queue *q) 1066 { 1067 if (queue_is_mq(q) && q->mq_ops->busy) 1068 return q->mq_ops->busy(q); 1069 1070 return 0; 1071 } 1072 EXPORT_SYMBOL_GPL(blk_lld_busy); 1073 1074 int kblockd_schedule_work(struct work_struct *work) 1075 { 1076 return queue_work(kblockd_workqueue, work); 1077 } 1078 EXPORT_SYMBOL(kblockd_schedule_work); 1079 1080 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1081 unsigned long delay) 1082 { 1083 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1084 } 1085 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1086 1087 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) 1088 { 1089 struct task_struct *tsk = current; 1090 1091 /* 1092 * If this is a nested plug, don't actually assign it. 1093 */ 1094 if (tsk->plug) 1095 return; 1096 1097 plug->cur_ktime = 0; 1098 plug->mq_list = NULL; 1099 plug->cached_rq = NULL; 1100 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); 1101 plug->rq_count = 0; 1102 plug->multiple_queues = false; 1103 plug->has_elevator = false; 1104 INIT_LIST_HEAD(&plug->cb_list); 1105 1106 /* 1107 * Store ordering should not be needed here, since a potential 1108 * preempt will imply a full memory barrier 1109 */ 1110 tsk->plug = plug; 1111 } 1112 1113 /** 1114 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1115 * @plug: The &struct blk_plug that needs to be initialized 1116 * 1117 * Description: 1118 * blk_start_plug() indicates to the block layer an intent by the caller 1119 * to submit multiple I/O requests in a batch. The block layer may use 1120 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1121 * is called. However, the block layer may choose to submit requests 1122 * before a call to blk_finish_plug() if the number of queued I/Os 1123 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1124 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1125 * the task schedules (see below). 1126 * 1127 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1128 * pending I/O should the task end up blocking between blk_start_plug() and 1129 * blk_finish_plug(). This is important from a performance perspective, but 1130 * also ensures that we don't deadlock. For instance, if the task is blocking 1131 * for a memory allocation, memory reclaim could end up wanting to free a 1132 * page belonging to that request that is currently residing in our private 1133 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1134 * this kind of deadlock. 1135 */ 1136 void blk_start_plug(struct blk_plug *plug) 1137 { 1138 blk_start_plug_nr_ios(plug, 1); 1139 } 1140 EXPORT_SYMBOL(blk_start_plug); 1141 1142 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1143 { 1144 LIST_HEAD(callbacks); 1145 1146 while (!list_empty(&plug->cb_list)) { 1147 list_splice_init(&plug->cb_list, &callbacks); 1148 1149 while (!list_empty(&callbacks)) { 1150 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1151 struct blk_plug_cb, 1152 list); 1153 list_del(&cb->list); 1154 cb->callback(cb, from_schedule); 1155 } 1156 } 1157 } 1158 1159 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1160 int size) 1161 { 1162 struct blk_plug *plug = current->plug; 1163 struct blk_plug_cb *cb; 1164 1165 if (!plug) 1166 return NULL; 1167 1168 list_for_each_entry(cb, &plug->cb_list, list) 1169 if (cb->callback == unplug && cb->data == data) 1170 return cb; 1171 1172 /* Not currently on the callback list */ 1173 BUG_ON(size < sizeof(*cb)); 1174 cb = kzalloc(size, GFP_ATOMIC); 1175 if (cb) { 1176 cb->data = data; 1177 cb->callback = unplug; 1178 list_add(&cb->list, &plug->cb_list); 1179 } 1180 return cb; 1181 } 1182 EXPORT_SYMBOL(blk_check_plugged); 1183 1184 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) 1185 { 1186 if (!list_empty(&plug->cb_list)) 1187 flush_plug_callbacks(plug, from_schedule); 1188 blk_mq_flush_plug_list(plug, from_schedule); 1189 /* 1190 * Unconditionally flush out cached requests, even if the unplug 1191 * event came from schedule. Since we know hold references to the 1192 * queue for cached requests, we don't want a blocked task holding 1193 * up a queue freeze/quiesce event. 1194 */ 1195 if (unlikely(!rq_list_empty(plug->cached_rq))) 1196 blk_mq_free_plug_rqs(plug); 1197 1198 current->flags &= ~PF_BLOCK_TS; 1199 } 1200 1201 /** 1202 * blk_finish_plug - mark the end of a batch of submitted I/O 1203 * @plug: The &struct blk_plug passed to blk_start_plug() 1204 * 1205 * Description: 1206 * Indicate that a batch of I/O submissions is complete. This function 1207 * must be paired with an initial call to blk_start_plug(). The intent 1208 * is to allow the block layer to optimize I/O submission. See the 1209 * documentation for blk_start_plug() for more information. 1210 */ 1211 void blk_finish_plug(struct blk_plug *plug) 1212 { 1213 if (plug == current->plug) { 1214 __blk_flush_plug(plug, false); 1215 current->plug = NULL; 1216 } 1217 } 1218 EXPORT_SYMBOL(blk_finish_plug); 1219 1220 void blk_io_schedule(void) 1221 { 1222 /* Prevent hang_check timer from firing at us during very long I/O */ 1223 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1224 1225 if (timeout) 1226 io_schedule_timeout(timeout); 1227 else 1228 io_schedule(); 1229 } 1230 EXPORT_SYMBOL_GPL(blk_io_schedule); 1231 1232 int __init blk_dev_init(void) 1233 { 1234 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1235 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1236 sizeof_field(struct request, cmd_flags)); 1237 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1238 sizeof_field(struct bio, bi_opf)); 1239 1240 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1241 kblockd_workqueue = alloc_workqueue("kblockd", 1242 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 1243 if (!kblockd_workqueue) 1244 panic("Failed to create kblockd\n"); 1245 1246 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC); 1247 1248 blk_debugfs_root = debugfs_create_dir("block", NULL); 1249 1250 return 0; 1251 } 1252