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/backing-dev.h> 18 #include <linux/bio.h> 19 #include <linux/blkdev.h> 20 #include <linux/blk-mq.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/blk-cgroup.h> 38 #include <linux/t10-pi.h> 39 #include <linux/debugfs.h> 40 #include <linux/bpf.h> 41 #include <linux/psi.h> 42 #include <linux/sched/sysctl.h> 43 #include <linux/blk-crypto.h> 44 45 #define CREATE_TRACE_POINTS 46 #include <trace/events/block.h> 47 48 #include "blk.h" 49 #include "blk-mq.h" 50 #include "blk-mq-sched.h" 51 #include "blk-pm.h" 52 #include "blk-rq-qos.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 62 DEFINE_IDA(blk_queue_ida); 63 64 /* 65 * For queue allocation 66 */ 67 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 void blk_rq_init(struct request_queue *q, struct request *rq) 111 { 112 memset(rq, 0, sizeof(*rq)); 113 114 INIT_LIST_HEAD(&rq->queuelist); 115 rq->q = q; 116 rq->__sector = (sector_t) -1; 117 INIT_HLIST_NODE(&rq->hash); 118 RB_CLEAR_NODE(&rq->rb_node); 119 rq->tag = -1; 120 rq->internal_tag = -1; 121 rq->start_time_ns = ktime_get_ns(); 122 rq->part = NULL; 123 refcount_set(&rq->ref, 1); 124 blk_crypto_rq_set_defaults(rq); 125 } 126 EXPORT_SYMBOL(blk_rq_init); 127 128 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name 129 static const char *const blk_op_name[] = { 130 REQ_OP_NAME(READ), 131 REQ_OP_NAME(WRITE), 132 REQ_OP_NAME(FLUSH), 133 REQ_OP_NAME(DISCARD), 134 REQ_OP_NAME(SECURE_ERASE), 135 REQ_OP_NAME(ZONE_RESET), 136 REQ_OP_NAME(ZONE_RESET_ALL), 137 REQ_OP_NAME(ZONE_OPEN), 138 REQ_OP_NAME(ZONE_CLOSE), 139 REQ_OP_NAME(ZONE_FINISH), 140 REQ_OP_NAME(ZONE_APPEND), 141 REQ_OP_NAME(WRITE_SAME), 142 REQ_OP_NAME(WRITE_ZEROES), 143 REQ_OP_NAME(SCSI_IN), 144 REQ_OP_NAME(SCSI_OUT), 145 REQ_OP_NAME(DRV_IN), 146 REQ_OP_NAME(DRV_OUT), 147 }; 148 #undef REQ_OP_NAME 149 150 /** 151 * blk_op_str - Return string XXX in the REQ_OP_XXX. 152 * @op: REQ_OP_XXX. 153 * 154 * Description: Centralize block layer function to convert REQ_OP_XXX into 155 * string format. Useful in the debugging and tracing bio or request. For 156 * invalid REQ_OP_XXX it returns string "UNKNOWN". 157 */ 158 inline const char *blk_op_str(unsigned int op) 159 { 160 const char *op_str = "UNKNOWN"; 161 162 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) 163 op_str = blk_op_name[op]; 164 165 return op_str; 166 } 167 EXPORT_SYMBOL_GPL(blk_op_str); 168 169 static const struct { 170 int errno; 171 const char *name; 172 } blk_errors[] = { 173 [BLK_STS_OK] = { 0, "" }, 174 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, 175 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, 176 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, 177 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, 178 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, 179 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, 180 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, 181 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, 182 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, 183 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, 184 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, 185 186 /* device mapper special case, should not leak out: */ 187 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, 188 189 /* everything else not covered above: */ 190 [BLK_STS_IOERR] = { -EIO, "I/O" }, 191 }; 192 193 blk_status_t errno_to_blk_status(int errno) 194 { 195 int i; 196 197 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { 198 if (blk_errors[i].errno == errno) 199 return (__force blk_status_t)i; 200 } 201 202 return BLK_STS_IOERR; 203 } 204 EXPORT_SYMBOL_GPL(errno_to_blk_status); 205 206 int blk_status_to_errno(blk_status_t status) 207 { 208 int idx = (__force int)status; 209 210 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 211 return -EIO; 212 return blk_errors[idx].errno; 213 } 214 EXPORT_SYMBOL_GPL(blk_status_to_errno); 215 216 static void print_req_error(struct request *req, blk_status_t status, 217 const char *caller) 218 { 219 int idx = (__force int)status; 220 221 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) 222 return; 223 224 printk_ratelimited(KERN_ERR 225 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " 226 "phys_seg %u prio class %u\n", 227 caller, blk_errors[idx].name, 228 req->rq_disk ? req->rq_disk->disk_name : "?", 229 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)), 230 req->cmd_flags & ~REQ_OP_MASK, 231 req->nr_phys_segments, 232 IOPRIO_PRIO_CLASS(req->ioprio)); 233 } 234 235 static void req_bio_endio(struct request *rq, struct bio *bio, 236 unsigned int nbytes, blk_status_t error) 237 { 238 if (error) 239 bio->bi_status = error; 240 241 if (unlikely(rq->rq_flags & RQF_QUIET)) 242 bio_set_flag(bio, BIO_QUIET); 243 244 bio_advance(bio, nbytes); 245 246 if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) { 247 /* 248 * Partial zone append completions cannot be supported as the 249 * BIO fragments may end up not being written sequentially. 250 */ 251 if (bio->bi_iter.bi_size) 252 bio->bi_status = BLK_STS_IOERR; 253 else 254 bio->bi_iter.bi_sector = rq->__sector; 255 } 256 257 /* don't actually finish bio if it's part of flush sequence */ 258 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) 259 bio_endio(bio); 260 } 261 262 void blk_dump_rq_flags(struct request *rq, char *msg) 263 { 264 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, 265 rq->rq_disk ? rq->rq_disk->disk_name : "?", 266 (unsigned long long) rq->cmd_flags); 267 268 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", 269 (unsigned long long)blk_rq_pos(rq), 270 blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); 271 printk(KERN_INFO " bio %p, biotail %p, len %u\n", 272 rq->bio, rq->biotail, blk_rq_bytes(rq)); 273 } 274 EXPORT_SYMBOL(blk_dump_rq_flags); 275 276 /** 277 * blk_sync_queue - cancel any pending callbacks on a queue 278 * @q: the queue 279 * 280 * Description: 281 * The block layer may perform asynchronous callback activity 282 * on a queue, such as calling the unplug function after a timeout. 283 * A block device may call blk_sync_queue to ensure that any 284 * such activity is cancelled, thus allowing it to release resources 285 * that the callbacks might use. The caller must already have made sure 286 * that its ->submit_bio will not re-add plugging prior to calling 287 * this function. 288 * 289 * This function does not cancel any asynchronous activity arising 290 * out of elevator or throttling code. That would require elevator_exit() 291 * and blkcg_exit_queue() to be called with queue lock initialized. 292 * 293 */ 294 void blk_sync_queue(struct request_queue *q) 295 { 296 del_timer_sync(&q->timeout); 297 cancel_work_sync(&q->timeout_work); 298 } 299 EXPORT_SYMBOL(blk_sync_queue); 300 301 /** 302 * blk_set_pm_only - increment pm_only counter 303 * @q: request queue pointer 304 */ 305 void blk_set_pm_only(struct request_queue *q) 306 { 307 atomic_inc(&q->pm_only); 308 } 309 EXPORT_SYMBOL_GPL(blk_set_pm_only); 310 311 void blk_clear_pm_only(struct request_queue *q) 312 { 313 int pm_only; 314 315 pm_only = atomic_dec_return(&q->pm_only); 316 WARN_ON_ONCE(pm_only < 0); 317 if (pm_only == 0) 318 wake_up_all(&q->mq_freeze_wq); 319 } 320 EXPORT_SYMBOL_GPL(blk_clear_pm_only); 321 322 /** 323 * blk_put_queue - decrement the request_queue refcount 324 * @q: the request_queue structure to decrement the refcount for 325 * 326 * Decrements the refcount of the request_queue kobject. When this reaches 0 327 * we'll have blk_release_queue() called. 328 * 329 * Context: Any context, but the last reference must not be dropped from 330 * atomic context. 331 */ 332 void blk_put_queue(struct request_queue *q) 333 { 334 kobject_put(&q->kobj); 335 } 336 EXPORT_SYMBOL(blk_put_queue); 337 338 void blk_set_queue_dying(struct request_queue *q) 339 { 340 blk_queue_flag_set(QUEUE_FLAG_DYING, q); 341 342 /* 343 * When queue DYING flag is set, we need to block new req 344 * entering queue, so we call blk_freeze_queue_start() to 345 * prevent I/O from crossing blk_queue_enter(). 346 */ 347 blk_freeze_queue_start(q); 348 349 if (queue_is_mq(q)) 350 blk_mq_wake_waiters(q); 351 352 /* Make blk_queue_enter() reexamine the DYING flag. */ 353 wake_up_all(&q->mq_freeze_wq); 354 } 355 EXPORT_SYMBOL_GPL(blk_set_queue_dying); 356 357 /** 358 * blk_cleanup_queue - shutdown a request queue 359 * @q: request queue to shutdown 360 * 361 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and 362 * put it. All future requests will be failed immediately with -ENODEV. 363 * 364 * Context: can sleep 365 */ 366 void blk_cleanup_queue(struct request_queue *q) 367 { 368 /* cannot be called from atomic context */ 369 might_sleep(); 370 371 WARN_ON_ONCE(blk_queue_registered(q)); 372 373 /* mark @q DYING, no new request or merges will be allowed afterwards */ 374 blk_set_queue_dying(q); 375 376 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); 377 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); 378 379 /* 380 * Drain all requests queued before DYING marking. Set DEAD flag to 381 * prevent that blk_mq_run_hw_queues() accesses the hardware queues 382 * after draining finished. 383 */ 384 blk_freeze_queue(q); 385 386 rq_qos_exit(q); 387 388 blk_queue_flag_set(QUEUE_FLAG_DEAD, q); 389 390 /* for synchronous bio-based driver finish in-flight integrity i/o */ 391 blk_flush_integrity(); 392 393 /* @q won't process any more request, flush async actions */ 394 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); 395 blk_sync_queue(q); 396 397 if (queue_is_mq(q)) 398 blk_mq_exit_queue(q); 399 400 /* 401 * In theory, request pool of sched_tags belongs to request queue. 402 * However, the current implementation requires tag_set for freeing 403 * requests, so free the pool now. 404 * 405 * Queue has become frozen, there can't be any in-queue requests, so 406 * it is safe to free requests now. 407 */ 408 mutex_lock(&q->sysfs_lock); 409 if (q->elevator) 410 blk_mq_sched_free_requests(q); 411 mutex_unlock(&q->sysfs_lock); 412 413 percpu_ref_exit(&q->q_usage_counter); 414 415 /* @q is and will stay empty, shutdown and put */ 416 blk_put_queue(q); 417 } 418 EXPORT_SYMBOL(blk_cleanup_queue); 419 420 /** 421 * blk_queue_enter() - try to increase q->q_usage_counter 422 * @q: request queue pointer 423 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT 424 */ 425 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) 426 { 427 const bool pm = flags & BLK_MQ_REQ_PREEMPT; 428 429 while (true) { 430 bool success = false; 431 432 rcu_read_lock(); 433 if (percpu_ref_tryget_live(&q->q_usage_counter)) { 434 /* 435 * The code that increments the pm_only counter is 436 * responsible for ensuring that that counter is 437 * globally visible before the queue is unfrozen. 438 */ 439 if (pm || !blk_queue_pm_only(q)) { 440 success = true; 441 } else { 442 percpu_ref_put(&q->q_usage_counter); 443 } 444 } 445 rcu_read_unlock(); 446 447 if (success) 448 return 0; 449 450 if (flags & BLK_MQ_REQ_NOWAIT) 451 return -EBUSY; 452 453 /* 454 * read pair of barrier in blk_freeze_queue_start(), 455 * we need to order reading __PERCPU_REF_DEAD flag of 456 * .q_usage_counter and reading .mq_freeze_depth or 457 * queue dying flag, otherwise the following wait may 458 * never return if the two reads are reordered. 459 */ 460 smp_rmb(); 461 462 wait_event(q->mq_freeze_wq, 463 (!q->mq_freeze_depth && 464 (pm || (blk_pm_request_resume(q), 465 !blk_queue_pm_only(q)))) || 466 blk_queue_dying(q)); 467 if (blk_queue_dying(q)) 468 return -ENODEV; 469 } 470 } 471 472 static inline int bio_queue_enter(struct bio *bio) 473 { 474 struct request_queue *q = bio->bi_disk->queue; 475 bool nowait = bio->bi_opf & REQ_NOWAIT; 476 int ret; 477 478 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0); 479 if (unlikely(ret)) { 480 if (nowait && !blk_queue_dying(q)) 481 bio_wouldblock_error(bio); 482 else 483 bio_io_error(bio); 484 } 485 486 return ret; 487 } 488 489 void blk_queue_exit(struct request_queue *q) 490 { 491 percpu_ref_put(&q->q_usage_counter); 492 } 493 494 static void blk_queue_usage_counter_release(struct percpu_ref *ref) 495 { 496 struct request_queue *q = 497 container_of(ref, struct request_queue, q_usage_counter); 498 499 wake_up_all(&q->mq_freeze_wq); 500 } 501 502 static void blk_rq_timed_out_timer(struct timer_list *t) 503 { 504 struct request_queue *q = from_timer(q, t, timeout); 505 506 kblockd_schedule_work(&q->timeout_work); 507 } 508 509 static void blk_timeout_work(struct work_struct *work) 510 { 511 } 512 513 struct request_queue *blk_alloc_queue(int node_id) 514 { 515 struct request_queue *q; 516 int ret; 517 518 q = kmem_cache_alloc_node(blk_requestq_cachep, 519 GFP_KERNEL | __GFP_ZERO, node_id); 520 if (!q) 521 return NULL; 522 523 q->last_merge = NULL; 524 525 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); 526 if (q->id < 0) 527 goto fail_q; 528 529 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); 530 if (ret) 531 goto fail_id; 532 533 q->backing_dev_info = bdi_alloc(node_id); 534 if (!q->backing_dev_info) 535 goto fail_split; 536 537 q->stats = blk_alloc_queue_stats(); 538 if (!q->stats) 539 goto fail_stats; 540 541 q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES; 542 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; 543 q->node = node_id; 544 545 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, 546 laptop_mode_timer_fn, 0); 547 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); 548 INIT_WORK(&q->timeout_work, blk_timeout_work); 549 INIT_LIST_HEAD(&q->icq_list); 550 #ifdef CONFIG_BLK_CGROUP 551 INIT_LIST_HEAD(&q->blkg_list); 552 #endif 553 554 kobject_init(&q->kobj, &blk_queue_ktype); 555 556 mutex_init(&q->debugfs_mutex); 557 mutex_init(&q->sysfs_lock); 558 mutex_init(&q->sysfs_dir_lock); 559 spin_lock_init(&q->queue_lock); 560 561 init_waitqueue_head(&q->mq_freeze_wq); 562 mutex_init(&q->mq_freeze_lock); 563 564 /* 565 * Init percpu_ref in atomic mode so that it's faster to shutdown. 566 * See blk_register_queue() for details. 567 */ 568 if (percpu_ref_init(&q->q_usage_counter, 569 blk_queue_usage_counter_release, 570 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) 571 goto fail_bdi; 572 573 if (blkcg_init_queue(q)) 574 goto fail_ref; 575 576 blk_queue_dma_alignment(q, 511); 577 blk_set_default_limits(&q->limits); 578 q->nr_requests = BLKDEV_MAX_RQ; 579 580 return q; 581 582 fail_ref: 583 percpu_ref_exit(&q->q_usage_counter); 584 fail_bdi: 585 blk_free_queue_stats(q->stats); 586 fail_stats: 587 bdi_put(q->backing_dev_info); 588 fail_split: 589 bioset_exit(&q->bio_split); 590 fail_id: 591 ida_simple_remove(&blk_queue_ida, q->id); 592 fail_q: 593 kmem_cache_free(blk_requestq_cachep, q); 594 return NULL; 595 } 596 EXPORT_SYMBOL(blk_alloc_queue); 597 598 /** 599 * blk_get_queue - increment the request_queue refcount 600 * @q: the request_queue structure to increment the refcount for 601 * 602 * Increment the refcount of the request_queue kobject. 603 * 604 * Context: Any context. 605 */ 606 bool blk_get_queue(struct request_queue *q) 607 { 608 if (likely(!blk_queue_dying(q))) { 609 __blk_get_queue(q); 610 return true; 611 } 612 613 return false; 614 } 615 EXPORT_SYMBOL(blk_get_queue); 616 617 /** 618 * blk_get_request - allocate a request 619 * @q: request queue to allocate a request for 620 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. 621 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. 622 */ 623 struct request *blk_get_request(struct request_queue *q, unsigned int op, 624 blk_mq_req_flags_t flags) 625 { 626 struct request *req; 627 628 WARN_ON_ONCE(op & REQ_NOWAIT); 629 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT)); 630 631 req = blk_mq_alloc_request(q, op, flags); 632 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) 633 q->mq_ops->initialize_rq_fn(req); 634 635 return req; 636 } 637 EXPORT_SYMBOL(blk_get_request); 638 639 void blk_put_request(struct request *req) 640 { 641 blk_mq_free_request(req); 642 } 643 EXPORT_SYMBOL(blk_put_request); 644 645 static void blk_account_io_merge_bio(struct request *req) 646 { 647 if (!blk_do_io_stat(req)) 648 return; 649 650 part_stat_lock(); 651 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); 652 part_stat_unlock(); 653 } 654 655 bool bio_attempt_back_merge(struct request *req, struct bio *bio, 656 unsigned int nr_segs) 657 { 658 const int ff = bio->bi_opf & REQ_FAILFAST_MASK; 659 660 if (!ll_back_merge_fn(req, bio, nr_segs)) 661 return false; 662 663 trace_block_bio_backmerge(req->q, req, bio); 664 rq_qos_merge(req->q, req, bio); 665 666 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 667 blk_rq_set_mixed_merge(req); 668 669 req->biotail->bi_next = bio; 670 req->biotail = bio; 671 req->__data_len += bio->bi_iter.bi_size; 672 673 bio_crypt_free_ctx(bio); 674 675 blk_account_io_merge_bio(req); 676 return true; 677 } 678 679 bool bio_attempt_front_merge(struct request *req, struct bio *bio, 680 unsigned int nr_segs) 681 { 682 const int ff = bio->bi_opf & REQ_FAILFAST_MASK; 683 684 if (!ll_front_merge_fn(req, bio, nr_segs)) 685 return false; 686 687 trace_block_bio_frontmerge(req->q, req, bio); 688 rq_qos_merge(req->q, req, bio); 689 690 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 691 blk_rq_set_mixed_merge(req); 692 693 bio->bi_next = req->bio; 694 req->bio = bio; 695 696 req->__sector = bio->bi_iter.bi_sector; 697 req->__data_len += bio->bi_iter.bi_size; 698 699 bio_crypt_do_front_merge(req, bio); 700 701 blk_account_io_merge_bio(req); 702 return true; 703 } 704 705 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, 706 struct bio *bio) 707 { 708 unsigned short segments = blk_rq_nr_discard_segments(req); 709 710 if (segments >= queue_max_discard_segments(q)) 711 goto no_merge; 712 if (blk_rq_sectors(req) + bio_sectors(bio) > 713 blk_rq_get_max_sectors(req, blk_rq_pos(req))) 714 goto no_merge; 715 716 rq_qos_merge(q, req, bio); 717 718 req->biotail->bi_next = bio; 719 req->biotail = bio; 720 req->__data_len += bio->bi_iter.bi_size; 721 req->nr_phys_segments = segments + 1; 722 723 blk_account_io_merge_bio(req); 724 return true; 725 no_merge: 726 req_set_nomerge(q, req); 727 return false; 728 } 729 730 /** 731 * blk_attempt_plug_merge - try to merge with %current's plugged list 732 * @q: request_queue new bio is being queued at 733 * @bio: new bio being queued 734 * @nr_segs: number of segments in @bio 735 * @same_queue_rq: pointer to &struct request that gets filled in when 736 * another request associated with @q is found on the plug list 737 * (optional, may be %NULL) 738 * 739 * Determine whether @bio being queued on @q can be merged with a request 740 * on %current's plugged list. Returns %true if merge was successful, 741 * otherwise %false. 742 * 743 * Plugging coalesces IOs from the same issuer for the same purpose without 744 * going through @q->queue_lock. As such it's more of an issuing mechanism 745 * than scheduling, and the request, while may have elvpriv data, is not 746 * added on the elevator at this point. In addition, we don't have 747 * reliable access to the elevator outside queue lock. Only check basic 748 * merging parameters without querying the elevator. 749 * 750 * Caller must ensure !blk_queue_nomerges(q) beforehand. 751 */ 752 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, 753 unsigned int nr_segs, struct request **same_queue_rq) 754 { 755 struct blk_plug *plug; 756 struct request *rq; 757 struct list_head *plug_list; 758 759 plug = blk_mq_plug(q, bio); 760 if (!plug) 761 return false; 762 763 plug_list = &plug->mq_list; 764 765 list_for_each_entry_reverse(rq, plug_list, queuelist) { 766 bool merged = false; 767 768 if (rq->q == q && same_queue_rq) { 769 /* 770 * Only blk-mq multiple hardware queues case checks the 771 * rq in the same queue, there should be only one such 772 * rq in a queue 773 **/ 774 *same_queue_rq = rq; 775 } 776 777 if (rq->q != q || !blk_rq_merge_ok(rq, bio)) 778 continue; 779 780 switch (blk_try_merge(rq, bio)) { 781 case ELEVATOR_BACK_MERGE: 782 merged = bio_attempt_back_merge(rq, bio, nr_segs); 783 break; 784 case ELEVATOR_FRONT_MERGE: 785 merged = bio_attempt_front_merge(rq, bio, nr_segs); 786 break; 787 case ELEVATOR_DISCARD_MERGE: 788 merged = bio_attempt_discard_merge(q, rq, bio); 789 break; 790 default: 791 break; 792 } 793 794 if (merged) 795 return true; 796 } 797 798 return false; 799 } 800 801 static void handle_bad_sector(struct bio *bio, sector_t maxsector) 802 { 803 char b[BDEVNAME_SIZE]; 804 805 printk(KERN_INFO "attempt to access beyond end of device\n"); 806 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", 807 bio_devname(bio, b), bio->bi_opf, 808 (unsigned long long)bio_end_sector(bio), 809 (long long)maxsector); 810 } 811 812 #ifdef CONFIG_FAIL_MAKE_REQUEST 813 814 static DECLARE_FAULT_ATTR(fail_make_request); 815 816 static int __init setup_fail_make_request(char *str) 817 { 818 return setup_fault_attr(&fail_make_request, str); 819 } 820 __setup("fail_make_request=", setup_fail_make_request); 821 822 static bool should_fail_request(struct hd_struct *part, unsigned int bytes) 823 { 824 return part->make_it_fail && should_fail(&fail_make_request, bytes); 825 } 826 827 static int __init fail_make_request_debugfs(void) 828 { 829 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 830 NULL, &fail_make_request); 831 832 return PTR_ERR_OR_ZERO(dir); 833 } 834 835 late_initcall(fail_make_request_debugfs); 836 837 #else /* CONFIG_FAIL_MAKE_REQUEST */ 838 839 static inline bool should_fail_request(struct hd_struct *part, 840 unsigned int bytes) 841 { 842 return false; 843 } 844 845 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 846 847 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) 848 { 849 const int op = bio_op(bio); 850 851 if (part->policy && op_is_write(op)) { 852 char b[BDEVNAME_SIZE]; 853 854 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) 855 return false; 856 857 WARN_ONCE(1, 858 "Trying to write to read-only block-device %s (partno %d)\n", 859 bio_devname(bio, b), part->partno); 860 /* Older lvm-tools actually trigger this */ 861 return false; 862 } 863 864 return false; 865 } 866 867 static noinline int should_fail_bio(struct bio *bio) 868 { 869 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) 870 return -EIO; 871 return 0; 872 } 873 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); 874 875 /* 876 * Check whether this bio extends beyond the end of the device or partition. 877 * This may well happen - the kernel calls bread() without checking the size of 878 * the device, e.g., when mounting a file system. 879 */ 880 static inline int bio_check_eod(struct bio *bio, sector_t maxsector) 881 { 882 unsigned int nr_sectors = bio_sectors(bio); 883 884 if (nr_sectors && maxsector && 885 (nr_sectors > maxsector || 886 bio->bi_iter.bi_sector > maxsector - nr_sectors)) { 887 handle_bad_sector(bio, maxsector); 888 return -EIO; 889 } 890 return 0; 891 } 892 893 /* 894 * Remap block n of partition p to block n+start(p) of the disk. 895 */ 896 static inline int blk_partition_remap(struct bio *bio) 897 { 898 struct hd_struct *p; 899 int ret = -EIO; 900 901 rcu_read_lock(); 902 p = __disk_get_part(bio->bi_disk, bio->bi_partno); 903 if (unlikely(!p)) 904 goto out; 905 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) 906 goto out; 907 if (unlikely(bio_check_ro(bio, p))) 908 goto out; 909 910 if (bio_sectors(bio)) { 911 if (bio_check_eod(bio, part_nr_sects_read(p))) 912 goto out; 913 bio->bi_iter.bi_sector += p->start_sect; 914 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), 915 bio->bi_iter.bi_sector - p->start_sect); 916 } 917 bio->bi_partno = 0; 918 ret = 0; 919 out: 920 rcu_read_unlock(); 921 return ret; 922 } 923 924 /* 925 * Check write append to a zoned block device. 926 */ 927 static inline blk_status_t blk_check_zone_append(struct request_queue *q, 928 struct bio *bio) 929 { 930 sector_t pos = bio->bi_iter.bi_sector; 931 int nr_sectors = bio_sectors(bio); 932 933 /* Only applicable to zoned block devices */ 934 if (!blk_queue_is_zoned(q)) 935 return BLK_STS_NOTSUPP; 936 937 /* The bio sector must point to the start of a sequential zone */ 938 if (pos & (blk_queue_zone_sectors(q) - 1) || 939 !blk_queue_zone_is_seq(q, pos)) 940 return BLK_STS_IOERR; 941 942 /* 943 * Not allowed to cross zone boundaries. Otherwise, the BIO will be 944 * split and could result in non-contiguous sectors being written in 945 * different zones. 946 */ 947 if (nr_sectors > q->limits.chunk_sectors) 948 return BLK_STS_IOERR; 949 950 /* Make sure the BIO is small enough and will not get split */ 951 if (nr_sectors > q->limits.max_zone_append_sectors) 952 return BLK_STS_IOERR; 953 954 bio->bi_opf |= REQ_NOMERGE; 955 956 return BLK_STS_OK; 957 } 958 959 static noinline_for_stack bool submit_bio_checks(struct bio *bio) 960 { 961 struct request_queue *q = bio->bi_disk->queue; 962 blk_status_t status = BLK_STS_IOERR; 963 964 might_sleep(); 965 966 /* 967 * For a REQ_NOWAIT based request, return -EOPNOTSUPP 968 * if queue is not a request based queue. 969 */ 970 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q)) 971 goto not_supported; 972 973 if (should_fail_bio(bio)) 974 goto end_io; 975 976 if (bio->bi_partno) { 977 if (unlikely(blk_partition_remap(bio))) 978 goto end_io; 979 } else { 980 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0))) 981 goto end_io; 982 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk)))) 983 goto end_io; 984 } 985 986 /* 987 * Filter flush bio's early so that bio based drivers without flush 988 * support don't have to worry about them. 989 */ 990 if (op_is_flush(bio->bi_opf) && 991 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { 992 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); 993 if (!bio_sectors(bio)) { 994 status = BLK_STS_OK; 995 goto end_io; 996 } 997 } 998 999 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) 1000 bio->bi_opf &= ~REQ_HIPRI; 1001 1002 switch (bio_op(bio)) { 1003 case REQ_OP_DISCARD: 1004 if (!blk_queue_discard(q)) 1005 goto not_supported; 1006 break; 1007 case REQ_OP_SECURE_ERASE: 1008 if (!blk_queue_secure_erase(q)) 1009 goto not_supported; 1010 break; 1011 case REQ_OP_WRITE_SAME: 1012 if (!q->limits.max_write_same_sectors) 1013 goto not_supported; 1014 break; 1015 case REQ_OP_ZONE_APPEND: 1016 status = blk_check_zone_append(q, bio); 1017 if (status != BLK_STS_OK) 1018 goto end_io; 1019 break; 1020 case REQ_OP_ZONE_RESET: 1021 case REQ_OP_ZONE_OPEN: 1022 case REQ_OP_ZONE_CLOSE: 1023 case REQ_OP_ZONE_FINISH: 1024 if (!blk_queue_is_zoned(q)) 1025 goto not_supported; 1026 break; 1027 case REQ_OP_ZONE_RESET_ALL: 1028 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q)) 1029 goto not_supported; 1030 break; 1031 case REQ_OP_WRITE_ZEROES: 1032 if (!q->limits.max_write_zeroes_sectors) 1033 goto not_supported; 1034 break; 1035 default: 1036 break; 1037 } 1038 1039 /* 1040 * Various block parts want %current->io_context, so allocate it up 1041 * front rather than dealing with lots of pain to allocate it only 1042 * where needed. This may fail and the block layer knows how to live 1043 * with it. 1044 */ 1045 if (unlikely(!current->io_context)) 1046 create_task_io_context(current, GFP_ATOMIC, q->node); 1047 1048 if (blk_throtl_bio(bio)) { 1049 blkcg_bio_issue_init(bio); 1050 return false; 1051 } 1052 1053 blk_cgroup_bio_start(bio); 1054 blkcg_bio_issue_init(bio); 1055 1056 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1057 trace_block_bio_queue(q, bio); 1058 /* Now that enqueuing has been traced, we need to trace 1059 * completion as well. 1060 */ 1061 bio_set_flag(bio, BIO_TRACE_COMPLETION); 1062 } 1063 return true; 1064 1065 not_supported: 1066 status = BLK_STS_NOTSUPP; 1067 end_io: 1068 bio->bi_status = status; 1069 bio_endio(bio); 1070 return false; 1071 } 1072 1073 static blk_qc_t __submit_bio(struct bio *bio) 1074 { 1075 struct gendisk *disk = bio->bi_disk; 1076 blk_qc_t ret = BLK_QC_T_NONE; 1077 1078 if (blk_crypto_bio_prep(&bio)) { 1079 if (!disk->fops->submit_bio) 1080 return blk_mq_submit_bio(bio); 1081 ret = disk->fops->submit_bio(bio); 1082 } 1083 blk_queue_exit(disk->queue); 1084 return ret; 1085 } 1086 1087 /* 1088 * The loop in this function may be a bit non-obvious, and so deserves some 1089 * explanation: 1090 * 1091 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure 1092 * that), so we have a list with a single bio. 1093 * - We pretend that we have just taken it off a longer list, so we assign 1094 * bio_list to a pointer to the bio_list_on_stack, thus initialising the 1095 * bio_list of new bios to be added. ->submit_bio() may indeed add some more 1096 * bios through a recursive call to submit_bio_noacct. If it did, we find a 1097 * non-NULL value in bio_list and re-enter the loop from the top. 1098 * - In this case we really did just take the bio of the top of the list (no 1099 * pretending) and so remove it from bio_list, and call into ->submit_bio() 1100 * again. 1101 * 1102 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. 1103 * bio_list_on_stack[1] contains bios that were submitted before the current 1104 * ->submit_bio_bio, but that haven't been processed yet. 1105 */ 1106 static blk_qc_t __submit_bio_noacct(struct bio *bio) 1107 { 1108 struct bio_list bio_list_on_stack[2]; 1109 blk_qc_t ret = BLK_QC_T_NONE; 1110 1111 BUG_ON(bio->bi_next); 1112 1113 bio_list_init(&bio_list_on_stack[0]); 1114 current->bio_list = bio_list_on_stack; 1115 1116 do { 1117 struct request_queue *q = bio->bi_disk->queue; 1118 struct bio_list lower, same; 1119 1120 if (unlikely(bio_queue_enter(bio) != 0)) 1121 continue; 1122 1123 /* 1124 * Create a fresh bio_list for all subordinate requests. 1125 */ 1126 bio_list_on_stack[1] = bio_list_on_stack[0]; 1127 bio_list_init(&bio_list_on_stack[0]); 1128 1129 ret = __submit_bio(bio); 1130 1131 /* 1132 * Sort new bios into those for a lower level and those for the 1133 * same level. 1134 */ 1135 bio_list_init(&lower); 1136 bio_list_init(&same); 1137 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) 1138 if (q == bio->bi_disk->queue) 1139 bio_list_add(&same, bio); 1140 else 1141 bio_list_add(&lower, bio); 1142 1143 /* 1144 * Now assemble so we handle the lowest level first. 1145 */ 1146 bio_list_merge(&bio_list_on_stack[0], &lower); 1147 bio_list_merge(&bio_list_on_stack[0], &same); 1148 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); 1149 } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); 1150 1151 current->bio_list = NULL; 1152 return ret; 1153 } 1154 1155 static blk_qc_t __submit_bio_noacct_mq(struct bio *bio) 1156 { 1157 struct bio_list bio_list[2] = { }; 1158 blk_qc_t ret = BLK_QC_T_NONE; 1159 1160 current->bio_list = bio_list; 1161 1162 do { 1163 struct gendisk *disk = bio->bi_disk; 1164 1165 if (unlikely(bio_queue_enter(bio) != 0)) 1166 continue; 1167 1168 if (!blk_crypto_bio_prep(&bio)) { 1169 blk_queue_exit(disk->queue); 1170 ret = BLK_QC_T_NONE; 1171 continue; 1172 } 1173 1174 ret = blk_mq_submit_bio(bio); 1175 } while ((bio = bio_list_pop(&bio_list[0]))); 1176 1177 current->bio_list = NULL; 1178 return ret; 1179 } 1180 1181 /** 1182 * submit_bio_noacct - re-submit a bio to the block device layer for I/O 1183 * @bio: The bio describing the location in memory and on the device. 1184 * 1185 * This is a version of submit_bio() that shall only be used for I/O that is 1186 * resubmitted to lower level drivers by stacking block drivers. All file 1187 * systems and other upper level users of the block layer should use 1188 * submit_bio() instead. 1189 */ 1190 blk_qc_t submit_bio_noacct(struct bio *bio) 1191 { 1192 if (!submit_bio_checks(bio)) 1193 return BLK_QC_T_NONE; 1194 1195 /* 1196 * We only want one ->submit_bio to be active at a time, else stack 1197 * usage with stacked devices could be a problem. Use current->bio_list 1198 * to collect a list of requests submited by a ->submit_bio method while 1199 * it is active, and then process them after it returned. 1200 */ 1201 if (current->bio_list) { 1202 bio_list_add(¤t->bio_list[0], bio); 1203 return BLK_QC_T_NONE; 1204 } 1205 1206 if (!bio->bi_disk->fops->submit_bio) 1207 return __submit_bio_noacct_mq(bio); 1208 return __submit_bio_noacct(bio); 1209 } 1210 EXPORT_SYMBOL(submit_bio_noacct); 1211 1212 /** 1213 * submit_bio - submit a bio to the block device layer for I/O 1214 * @bio: The &struct bio which describes the I/O 1215 * 1216 * submit_bio() is used to submit I/O requests to block devices. It is passed a 1217 * fully set up &struct bio that describes the I/O that needs to be done. The 1218 * bio will be send to the device described by the bi_disk and bi_partno fields. 1219 * 1220 * The success/failure status of the request, along with notification of 1221 * completion, is delivered asynchronously through the ->bi_end_io() callback 1222 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has 1223 * been called. 1224 */ 1225 blk_qc_t submit_bio(struct bio *bio) 1226 { 1227 if (blkcg_punt_bio_submit(bio)) 1228 return BLK_QC_T_NONE; 1229 1230 /* 1231 * If it's a regular read/write or a barrier with data attached, 1232 * go through the normal accounting stuff before submission. 1233 */ 1234 if (bio_has_data(bio)) { 1235 unsigned int count; 1236 1237 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) 1238 count = queue_logical_block_size(bio->bi_disk->queue) >> 9; 1239 else 1240 count = bio_sectors(bio); 1241 1242 if (op_is_write(bio_op(bio))) { 1243 count_vm_events(PGPGOUT, count); 1244 } else { 1245 task_io_account_read(bio->bi_iter.bi_size); 1246 count_vm_events(PGPGIN, count); 1247 } 1248 1249 if (unlikely(block_dump)) { 1250 char b[BDEVNAME_SIZE]; 1251 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", 1252 current->comm, task_pid_nr(current), 1253 op_is_write(bio_op(bio)) ? "WRITE" : "READ", 1254 (unsigned long long)bio->bi_iter.bi_sector, 1255 bio_devname(bio, b), count); 1256 } 1257 } 1258 1259 /* 1260 * If we're reading data that is part of the userspace workingset, count 1261 * submission time as memory stall. When the device is congested, or 1262 * the submitting cgroup IO-throttled, submission can be a significant 1263 * part of overall IO time. 1264 */ 1265 if (unlikely(bio_op(bio) == REQ_OP_READ && 1266 bio_flagged(bio, BIO_WORKINGSET))) { 1267 unsigned long pflags; 1268 blk_qc_t ret; 1269 1270 psi_memstall_enter(&pflags); 1271 ret = submit_bio_noacct(bio); 1272 psi_memstall_leave(&pflags); 1273 1274 return ret; 1275 } 1276 1277 return submit_bio_noacct(bio); 1278 } 1279 EXPORT_SYMBOL(submit_bio); 1280 1281 /** 1282 * blk_cloned_rq_check_limits - Helper function to check a cloned request 1283 * for the new queue limits 1284 * @q: the queue 1285 * @rq: the request being checked 1286 * 1287 * Description: 1288 * @rq may have been made based on weaker limitations of upper-level queues 1289 * in request stacking drivers, and it may violate the limitation of @q. 1290 * Since the block layer and the underlying device driver trust @rq 1291 * after it is inserted to @q, it should be checked against @q before 1292 * the insertion using this generic function. 1293 * 1294 * Request stacking drivers like request-based dm may change the queue 1295 * limits when retrying requests on other queues. Those requests need 1296 * to be checked against the new queue limits again during dispatch. 1297 */ 1298 static int blk_cloned_rq_check_limits(struct request_queue *q, 1299 struct request *rq) 1300 { 1301 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { 1302 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", 1303 __func__, blk_rq_sectors(rq), 1304 blk_queue_get_max_sectors(q, req_op(rq))); 1305 return -EIO; 1306 } 1307 1308 /* 1309 * queue's settings related to segment counting like q->bounce_pfn 1310 * may differ from that of other stacking queues. 1311 * Recalculate it to check the request correctly on this queue's 1312 * limitation. 1313 */ 1314 rq->nr_phys_segments = blk_recalc_rq_segments(rq); 1315 if (rq->nr_phys_segments > queue_max_segments(q)) { 1316 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", 1317 __func__, rq->nr_phys_segments, queue_max_segments(q)); 1318 return -EIO; 1319 } 1320 1321 return 0; 1322 } 1323 1324 /** 1325 * blk_insert_cloned_request - Helper for stacking drivers to submit a request 1326 * @q: the queue to submit the request 1327 * @rq: the request being queued 1328 */ 1329 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) 1330 { 1331 if (blk_cloned_rq_check_limits(q, rq)) 1332 return BLK_STS_IOERR; 1333 1334 if (rq->rq_disk && 1335 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) 1336 return BLK_STS_IOERR; 1337 1338 if (blk_crypto_insert_cloned_request(rq)) 1339 return BLK_STS_IOERR; 1340 1341 if (blk_queue_io_stat(q)) 1342 blk_account_io_start(rq); 1343 1344 /* 1345 * Since we have a scheduler attached on the top device, 1346 * bypass a potential scheduler on the bottom device for 1347 * insert. 1348 */ 1349 return blk_mq_request_issue_directly(rq, true); 1350 } 1351 EXPORT_SYMBOL_GPL(blk_insert_cloned_request); 1352 1353 /** 1354 * blk_rq_err_bytes - determine number of bytes till the next failure boundary 1355 * @rq: request to examine 1356 * 1357 * Description: 1358 * A request could be merge of IOs which require different failure 1359 * handling. This function determines the number of bytes which 1360 * can be failed from the beginning of the request without 1361 * crossing into area which need to be retried further. 1362 * 1363 * Return: 1364 * The number of bytes to fail. 1365 */ 1366 unsigned int blk_rq_err_bytes(const struct request *rq) 1367 { 1368 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; 1369 unsigned int bytes = 0; 1370 struct bio *bio; 1371 1372 if (!(rq->rq_flags & RQF_MIXED_MERGE)) 1373 return blk_rq_bytes(rq); 1374 1375 /* 1376 * Currently the only 'mixing' which can happen is between 1377 * different fastfail types. We can safely fail portions 1378 * which have all the failfast bits that the first one has - 1379 * the ones which are at least as eager to fail as the first 1380 * one. 1381 */ 1382 for (bio = rq->bio; bio; bio = bio->bi_next) { 1383 if ((bio->bi_opf & ff) != ff) 1384 break; 1385 bytes += bio->bi_iter.bi_size; 1386 } 1387 1388 /* this could lead to infinite loop */ 1389 BUG_ON(blk_rq_bytes(rq) && !bytes); 1390 return bytes; 1391 } 1392 EXPORT_SYMBOL_GPL(blk_rq_err_bytes); 1393 1394 static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end) 1395 { 1396 unsigned long stamp; 1397 again: 1398 stamp = READ_ONCE(part->stamp); 1399 if (unlikely(stamp != now)) { 1400 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) 1401 __part_stat_add(part, io_ticks, end ? now - stamp : 1); 1402 } 1403 if (part->partno) { 1404 part = &part_to_disk(part)->part0; 1405 goto again; 1406 } 1407 } 1408 1409 static void blk_account_io_completion(struct request *req, unsigned int bytes) 1410 { 1411 if (req->part && blk_do_io_stat(req)) { 1412 const int sgrp = op_stat_group(req_op(req)); 1413 struct hd_struct *part; 1414 1415 part_stat_lock(); 1416 part = req->part; 1417 part_stat_add(part, sectors[sgrp], bytes >> 9); 1418 part_stat_unlock(); 1419 } 1420 } 1421 1422 void blk_account_io_done(struct request *req, u64 now) 1423 { 1424 /* 1425 * Account IO completion. flush_rq isn't accounted as a 1426 * normal IO on queueing nor completion. Accounting the 1427 * containing request is enough. 1428 */ 1429 if (req->part && blk_do_io_stat(req) && 1430 !(req->rq_flags & RQF_FLUSH_SEQ)) { 1431 const int sgrp = op_stat_group(req_op(req)); 1432 struct hd_struct *part; 1433 1434 part_stat_lock(); 1435 part = req->part; 1436 1437 update_io_ticks(part, jiffies, true); 1438 part_stat_inc(part, ios[sgrp]); 1439 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns); 1440 part_stat_unlock(); 1441 1442 hd_struct_put(part); 1443 } 1444 } 1445 1446 void blk_account_io_start(struct request *rq) 1447 { 1448 if (!blk_do_io_stat(rq)) 1449 return; 1450 1451 rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); 1452 1453 part_stat_lock(); 1454 update_io_ticks(rq->part, jiffies, false); 1455 part_stat_unlock(); 1456 } 1457 1458 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors, 1459 unsigned int op) 1460 { 1461 struct hd_struct *part = &disk->part0; 1462 const int sgrp = op_stat_group(op); 1463 unsigned long now = READ_ONCE(jiffies); 1464 1465 part_stat_lock(); 1466 update_io_ticks(part, now, false); 1467 part_stat_inc(part, ios[sgrp]); 1468 part_stat_add(part, sectors[sgrp], sectors); 1469 part_stat_local_inc(part, in_flight[op_is_write(op)]); 1470 part_stat_unlock(); 1471 1472 return now; 1473 } 1474 EXPORT_SYMBOL(disk_start_io_acct); 1475 1476 void disk_end_io_acct(struct gendisk *disk, unsigned int op, 1477 unsigned long start_time) 1478 { 1479 struct hd_struct *part = &disk->part0; 1480 const int sgrp = op_stat_group(op); 1481 unsigned long now = READ_ONCE(jiffies); 1482 unsigned long duration = now - start_time; 1483 1484 part_stat_lock(); 1485 update_io_ticks(part, now, true); 1486 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 1487 part_stat_local_dec(part, in_flight[op_is_write(op)]); 1488 part_stat_unlock(); 1489 } 1490 EXPORT_SYMBOL(disk_end_io_acct); 1491 1492 /* 1493 * Steal bios from a request and add them to a bio list. 1494 * The request must not have been partially completed before. 1495 */ 1496 void blk_steal_bios(struct bio_list *list, struct request *rq) 1497 { 1498 if (rq->bio) { 1499 if (list->tail) 1500 list->tail->bi_next = rq->bio; 1501 else 1502 list->head = rq->bio; 1503 list->tail = rq->biotail; 1504 1505 rq->bio = NULL; 1506 rq->biotail = NULL; 1507 } 1508 1509 rq->__data_len = 0; 1510 } 1511 EXPORT_SYMBOL_GPL(blk_steal_bios); 1512 1513 /** 1514 * blk_update_request - Special helper function for request stacking drivers 1515 * @req: the request being processed 1516 * @error: block status code 1517 * @nr_bytes: number of bytes to complete @req 1518 * 1519 * Description: 1520 * Ends I/O on a number of bytes attached to @req, but doesn't complete 1521 * the request structure even if @req doesn't have leftover. 1522 * If @req has leftover, sets it up for the next range of segments. 1523 * 1524 * This special helper function is only for request stacking drivers 1525 * (e.g. request-based dm) so that they can handle partial completion. 1526 * Actual device drivers should use blk_mq_end_request instead. 1527 * 1528 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees 1529 * %false return from this function. 1530 * 1531 * Note: 1532 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both 1533 * blk_rq_bytes() and in blk_update_request(). 1534 * 1535 * Return: 1536 * %false - this request doesn't have any more data 1537 * %true - this request has more data 1538 **/ 1539 bool blk_update_request(struct request *req, blk_status_t error, 1540 unsigned int nr_bytes) 1541 { 1542 int total_bytes; 1543 1544 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); 1545 1546 if (!req->bio) 1547 return false; 1548 1549 #ifdef CONFIG_BLK_DEV_INTEGRITY 1550 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ && 1551 error == BLK_STS_OK) 1552 req->q->integrity.profile->complete_fn(req, nr_bytes); 1553 #endif 1554 1555 if (unlikely(error && !blk_rq_is_passthrough(req) && 1556 !(req->rq_flags & RQF_QUIET))) 1557 print_req_error(req, error, __func__); 1558 1559 blk_account_io_completion(req, nr_bytes); 1560 1561 total_bytes = 0; 1562 while (req->bio) { 1563 struct bio *bio = req->bio; 1564 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); 1565 1566 if (bio_bytes == bio->bi_iter.bi_size) 1567 req->bio = bio->bi_next; 1568 1569 /* Completion has already been traced */ 1570 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1571 req_bio_endio(req, bio, bio_bytes, error); 1572 1573 total_bytes += bio_bytes; 1574 nr_bytes -= bio_bytes; 1575 1576 if (!nr_bytes) 1577 break; 1578 } 1579 1580 /* 1581 * completely done 1582 */ 1583 if (!req->bio) { 1584 /* 1585 * Reset counters so that the request stacking driver 1586 * can find how many bytes remain in the request 1587 * later. 1588 */ 1589 req->__data_len = 0; 1590 return false; 1591 } 1592 1593 req->__data_len -= total_bytes; 1594 1595 /* update sector only for requests with clear definition of sector */ 1596 if (!blk_rq_is_passthrough(req)) 1597 req->__sector += total_bytes >> 9; 1598 1599 /* mixed attributes always follow the first bio */ 1600 if (req->rq_flags & RQF_MIXED_MERGE) { 1601 req->cmd_flags &= ~REQ_FAILFAST_MASK; 1602 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; 1603 } 1604 1605 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { 1606 /* 1607 * If total number of sectors is less than the first segment 1608 * size, something has gone terribly wrong. 1609 */ 1610 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { 1611 blk_dump_rq_flags(req, "request botched"); 1612 req->__data_len = blk_rq_cur_bytes(req); 1613 } 1614 1615 /* recalculate the number of segments */ 1616 req->nr_phys_segments = blk_recalc_rq_segments(req); 1617 } 1618 1619 return true; 1620 } 1621 EXPORT_SYMBOL_GPL(blk_update_request); 1622 1623 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1624 /** 1625 * rq_flush_dcache_pages - Helper function to flush all pages in a request 1626 * @rq: the request to be flushed 1627 * 1628 * Description: 1629 * Flush all pages in @rq. 1630 */ 1631 void rq_flush_dcache_pages(struct request *rq) 1632 { 1633 struct req_iterator iter; 1634 struct bio_vec bvec; 1635 1636 rq_for_each_segment(bvec, rq, iter) 1637 flush_dcache_page(bvec.bv_page); 1638 } 1639 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); 1640 #endif 1641 1642 /** 1643 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 1644 * @q : the queue of the device being checked 1645 * 1646 * Description: 1647 * Check if underlying low-level drivers of a device are busy. 1648 * If the drivers want to export their busy state, they must set own 1649 * exporting function using blk_queue_lld_busy() first. 1650 * 1651 * Basically, this function is used only by request stacking drivers 1652 * to stop dispatching requests to underlying devices when underlying 1653 * devices are busy. This behavior helps more I/O merging on the queue 1654 * of the request stacking driver and prevents I/O throughput regression 1655 * on burst I/O load. 1656 * 1657 * Return: 1658 * 0 - Not busy (The request stacking driver should dispatch request) 1659 * 1 - Busy (The request stacking driver should stop dispatching request) 1660 */ 1661 int blk_lld_busy(struct request_queue *q) 1662 { 1663 if (queue_is_mq(q) && q->mq_ops->busy) 1664 return q->mq_ops->busy(q); 1665 1666 return 0; 1667 } 1668 EXPORT_SYMBOL_GPL(blk_lld_busy); 1669 1670 /** 1671 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request 1672 * @rq: the clone request to be cleaned up 1673 * 1674 * Description: 1675 * Free all bios in @rq for a cloned request. 1676 */ 1677 void blk_rq_unprep_clone(struct request *rq) 1678 { 1679 struct bio *bio; 1680 1681 while ((bio = rq->bio) != NULL) { 1682 rq->bio = bio->bi_next; 1683 1684 bio_put(bio); 1685 } 1686 } 1687 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); 1688 1689 /** 1690 * blk_rq_prep_clone - Helper function to setup clone request 1691 * @rq: the request to be setup 1692 * @rq_src: original request to be cloned 1693 * @bs: bio_set that bios for clone are allocated from 1694 * @gfp_mask: memory allocation mask for bio 1695 * @bio_ctr: setup function to be called for each clone bio. 1696 * Returns %0 for success, non %0 for failure. 1697 * @data: private data to be passed to @bio_ctr 1698 * 1699 * Description: 1700 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. 1701 * Also, pages which the original bios are pointing to are not copied 1702 * and the cloned bios just point same pages. 1703 * So cloned bios must be completed before original bios, which means 1704 * the caller must complete @rq before @rq_src. 1705 */ 1706 int blk_rq_prep_clone(struct request *rq, struct request *rq_src, 1707 struct bio_set *bs, gfp_t gfp_mask, 1708 int (*bio_ctr)(struct bio *, struct bio *, void *), 1709 void *data) 1710 { 1711 struct bio *bio, *bio_src; 1712 1713 if (!bs) 1714 bs = &fs_bio_set; 1715 1716 __rq_for_each_bio(bio_src, rq_src) { 1717 bio = bio_clone_fast(bio_src, gfp_mask, bs); 1718 if (!bio) 1719 goto free_and_out; 1720 1721 if (bio_ctr && bio_ctr(bio, bio_src, data)) 1722 goto free_and_out; 1723 1724 if (rq->bio) { 1725 rq->biotail->bi_next = bio; 1726 rq->biotail = bio; 1727 } else 1728 rq->bio = rq->biotail = bio; 1729 } 1730 1731 /* Copy attributes of the original request to the clone request. */ 1732 rq->__sector = blk_rq_pos(rq_src); 1733 rq->__data_len = blk_rq_bytes(rq_src); 1734 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) { 1735 rq->rq_flags |= RQF_SPECIAL_PAYLOAD; 1736 rq->special_vec = rq_src->special_vec; 1737 } 1738 rq->nr_phys_segments = rq_src->nr_phys_segments; 1739 rq->ioprio = rq_src->ioprio; 1740 1741 if (rq->bio) 1742 blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask); 1743 1744 return 0; 1745 1746 free_and_out: 1747 if (bio) 1748 bio_put(bio); 1749 blk_rq_unprep_clone(rq); 1750 1751 return -ENOMEM; 1752 } 1753 EXPORT_SYMBOL_GPL(blk_rq_prep_clone); 1754 1755 int kblockd_schedule_work(struct work_struct *work) 1756 { 1757 return queue_work(kblockd_workqueue, work); 1758 } 1759 EXPORT_SYMBOL(kblockd_schedule_work); 1760 1761 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, 1762 unsigned long delay) 1763 { 1764 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 1765 } 1766 EXPORT_SYMBOL(kblockd_mod_delayed_work_on); 1767 1768 /** 1769 * blk_start_plug - initialize blk_plug and track it inside the task_struct 1770 * @plug: The &struct blk_plug that needs to be initialized 1771 * 1772 * Description: 1773 * blk_start_plug() indicates to the block layer an intent by the caller 1774 * to submit multiple I/O requests in a batch. The block layer may use 1775 * this hint to defer submitting I/Os from the caller until blk_finish_plug() 1776 * is called. However, the block layer may choose to submit requests 1777 * before a call to blk_finish_plug() if the number of queued I/Os 1778 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than 1779 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if 1780 * the task schedules (see below). 1781 * 1782 * Tracking blk_plug inside the task_struct will help with auto-flushing the 1783 * pending I/O should the task end up blocking between blk_start_plug() and 1784 * blk_finish_plug(). This is important from a performance perspective, but 1785 * also ensures that we don't deadlock. For instance, if the task is blocking 1786 * for a memory allocation, memory reclaim could end up wanting to free a 1787 * page belonging to that request that is currently residing in our private 1788 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 1789 * this kind of deadlock. 1790 */ 1791 void blk_start_plug(struct blk_plug *plug) 1792 { 1793 struct task_struct *tsk = current; 1794 1795 /* 1796 * If this is a nested plug, don't actually assign it. 1797 */ 1798 if (tsk->plug) 1799 return; 1800 1801 INIT_LIST_HEAD(&plug->mq_list); 1802 INIT_LIST_HEAD(&plug->cb_list); 1803 plug->rq_count = 0; 1804 plug->multiple_queues = false; 1805 1806 /* 1807 * Store ordering should not be needed here, since a potential 1808 * preempt will imply a full memory barrier 1809 */ 1810 tsk->plug = plug; 1811 } 1812 EXPORT_SYMBOL(blk_start_plug); 1813 1814 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 1815 { 1816 LIST_HEAD(callbacks); 1817 1818 while (!list_empty(&plug->cb_list)) { 1819 list_splice_init(&plug->cb_list, &callbacks); 1820 1821 while (!list_empty(&callbacks)) { 1822 struct blk_plug_cb *cb = list_first_entry(&callbacks, 1823 struct blk_plug_cb, 1824 list); 1825 list_del(&cb->list); 1826 cb->callback(cb, from_schedule); 1827 } 1828 } 1829 } 1830 1831 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 1832 int size) 1833 { 1834 struct blk_plug *plug = current->plug; 1835 struct blk_plug_cb *cb; 1836 1837 if (!plug) 1838 return NULL; 1839 1840 list_for_each_entry(cb, &plug->cb_list, list) 1841 if (cb->callback == unplug && cb->data == data) 1842 return cb; 1843 1844 /* Not currently on the callback list */ 1845 BUG_ON(size < sizeof(*cb)); 1846 cb = kzalloc(size, GFP_ATOMIC); 1847 if (cb) { 1848 cb->data = data; 1849 cb->callback = unplug; 1850 list_add(&cb->list, &plug->cb_list); 1851 } 1852 return cb; 1853 } 1854 EXPORT_SYMBOL(blk_check_plugged); 1855 1856 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) 1857 { 1858 flush_plug_callbacks(plug, from_schedule); 1859 1860 if (!list_empty(&plug->mq_list)) 1861 blk_mq_flush_plug_list(plug, from_schedule); 1862 } 1863 1864 /** 1865 * blk_finish_plug - mark the end of a batch of submitted I/O 1866 * @plug: The &struct blk_plug passed to blk_start_plug() 1867 * 1868 * Description: 1869 * Indicate that a batch of I/O submissions is complete. This function 1870 * must be paired with an initial call to blk_start_plug(). The intent 1871 * is to allow the block layer to optimize I/O submission. See the 1872 * documentation for blk_start_plug() for more information. 1873 */ 1874 void blk_finish_plug(struct blk_plug *plug) 1875 { 1876 if (plug != current->plug) 1877 return; 1878 blk_flush_plug_list(plug, false); 1879 1880 current->plug = NULL; 1881 } 1882 EXPORT_SYMBOL(blk_finish_plug); 1883 1884 void blk_io_schedule(void) 1885 { 1886 /* Prevent hang_check timer from firing at us during very long I/O */ 1887 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 1888 1889 if (timeout) 1890 io_schedule_timeout(timeout); 1891 else 1892 io_schedule(); 1893 } 1894 EXPORT_SYMBOL_GPL(blk_io_schedule); 1895 1896 int __init blk_dev_init(void) 1897 { 1898 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); 1899 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1900 sizeof_field(struct request, cmd_flags)); 1901 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * 1902 sizeof_field(struct bio, bi_opf)); 1903 1904 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 1905 kblockd_workqueue = alloc_workqueue("kblockd", 1906 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 1907 if (!kblockd_workqueue) 1908 panic("Failed to create kblockd\n"); 1909 1910 blk_requestq_cachep = kmem_cache_create("request_queue", 1911 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 1912 1913 blk_debugfs_root = debugfs_create_dir("block", NULL); 1914 1915 return 0; 1916 } 1917