1 /* 2 * Copyright (C) 1991, 1992 Linus Torvalds 3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics 4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE 5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> 6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> 7 * - July2000 8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 9 */ 10 11 /* 12 * This handles all read/write requests to block devices 13 */ 14 #include <linux/kernel.h> 15 #include <linux/module.h> 16 #include <linux/backing-dev.h> 17 #include <linux/bio.h> 18 #include <linux/blkdev.h> 19 #include <linux/blk-mq.h> 20 #include <linux/highmem.h> 21 #include <linux/mm.h> 22 #include <linux/kernel_stat.h> 23 #include <linux/string.h> 24 #include <linux/init.h> 25 #include <linux/completion.h> 26 #include <linux/slab.h> 27 #include <linux/swap.h> 28 #include <linux/writeback.h> 29 #include <linux/task_io_accounting_ops.h> 30 #include <linux/fault-inject.h> 31 #include <linux/list_sort.h> 32 #include <linux/delay.h> 33 #include <linux/ratelimit.h> 34 #include <linux/pm_runtime.h> 35 36 #define CREATE_TRACE_POINTS 37 #include <trace/events/block.h> 38 39 #include "blk.h" 40 #include "blk-cgroup.h" 41 #include "blk-mq.h" 42 43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); 44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); 45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); 46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); 47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); 48 49 DEFINE_IDA(blk_queue_ida); 50 51 /* 52 * For the allocated request tables 53 */ 54 struct kmem_cache *request_cachep = NULL; 55 56 /* 57 * For queue allocation 58 */ 59 struct kmem_cache *blk_requestq_cachep; 60 61 /* 62 * Controlling structure to kblockd 63 */ 64 static struct workqueue_struct *kblockd_workqueue; 65 66 void blk_queue_congestion_threshold(struct request_queue *q) 67 { 68 int nr; 69 70 nr = q->nr_requests - (q->nr_requests / 8) + 1; 71 if (nr > q->nr_requests) 72 nr = q->nr_requests; 73 q->nr_congestion_on = nr; 74 75 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; 76 if (nr < 1) 77 nr = 1; 78 q->nr_congestion_off = nr; 79 } 80 81 /** 82 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info 83 * @bdev: device 84 * 85 * Locates the passed device's request queue and returns the address of its 86 * backing_dev_info. This function can only be called if @bdev is opened 87 * and the return value is never NULL. 88 */ 89 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) 90 { 91 struct request_queue *q = bdev_get_queue(bdev); 92 93 return &q->backing_dev_info; 94 } 95 EXPORT_SYMBOL(blk_get_backing_dev_info); 96 97 void blk_rq_init(struct request_queue *q, struct request *rq) 98 { 99 memset(rq, 0, sizeof(*rq)); 100 101 INIT_LIST_HEAD(&rq->queuelist); 102 INIT_LIST_HEAD(&rq->timeout_list); 103 rq->cpu = -1; 104 rq->q = q; 105 rq->__sector = (sector_t) -1; 106 INIT_HLIST_NODE(&rq->hash); 107 RB_CLEAR_NODE(&rq->rb_node); 108 rq->cmd = rq->__cmd; 109 rq->cmd_len = BLK_MAX_CDB; 110 rq->tag = -1; 111 rq->start_time = jiffies; 112 set_start_time_ns(rq); 113 rq->part = NULL; 114 } 115 EXPORT_SYMBOL(blk_rq_init); 116 117 static void req_bio_endio(struct request *rq, struct bio *bio, 118 unsigned int nbytes, int error) 119 { 120 if (error) 121 clear_bit(BIO_UPTODATE, &bio->bi_flags); 122 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) 123 error = -EIO; 124 125 if (unlikely(rq->cmd_flags & REQ_QUIET)) 126 set_bit(BIO_QUIET, &bio->bi_flags); 127 128 bio_advance(bio, nbytes); 129 130 /* don't actually finish bio if it's part of flush sequence */ 131 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) 132 bio_endio(bio, error); 133 } 134 135 void blk_dump_rq_flags(struct request *rq, char *msg) 136 { 137 int bit; 138 139 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg, 140 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, 141 (unsigned long long) rq->cmd_flags); 142 143 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", 144 (unsigned long long)blk_rq_pos(rq), 145 blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); 146 printk(KERN_INFO " bio %p, biotail %p, len %u\n", 147 rq->bio, rq->biotail, blk_rq_bytes(rq)); 148 149 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { 150 printk(KERN_INFO " cdb: "); 151 for (bit = 0; bit < BLK_MAX_CDB; bit++) 152 printk("%02x ", rq->cmd[bit]); 153 printk("\n"); 154 } 155 } 156 EXPORT_SYMBOL(blk_dump_rq_flags); 157 158 static void blk_delay_work(struct work_struct *work) 159 { 160 struct request_queue *q; 161 162 q = container_of(work, struct request_queue, delay_work.work); 163 spin_lock_irq(q->queue_lock); 164 __blk_run_queue(q); 165 spin_unlock_irq(q->queue_lock); 166 } 167 168 /** 169 * blk_delay_queue - restart queueing after defined interval 170 * @q: The &struct request_queue in question 171 * @msecs: Delay in msecs 172 * 173 * Description: 174 * Sometimes queueing needs to be postponed for a little while, to allow 175 * resources to come back. This function will make sure that queueing is 176 * restarted around the specified time. Queue lock must be held. 177 */ 178 void blk_delay_queue(struct request_queue *q, unsigned long msecs) 179 { 180 if (likely(!blk_queue_dead(q))) 181 queue_delayed_work(kblockd_workqueue, &q->delay_work, 182 msecs_to_jiffies(msecs)); 183 } 184 EXPORT_SYMBOL(blk_delay_queue); 185 186 /** 187 * blk_start_queue - restart a previously stopped queue 188 * @q: The &struct request_queue in question 189 * 190 * Description: 191 * blk_start_queue() will clear the stop flag on the queue, and call 192 * the request_fn for the queue if it was in a stopped state when 193 * entered. Also see blk_stop_queue(). Queue lock must be held. 194 **/ 195 void blk_start_queue(struct request_queue *q) 196 { 197 WARN_ON(!irqs_disabled()); 198 199 queue_flag_clear(QUEUE_FLAG_STOPPED, q); 200 __blk_run_queue(q); 201 } 202 EXPORT_SYMBOL(blk_start_queue); 203 204 /** 205 * blk_stop_queue - stop a queue 206 * @q: The &struct request_queue in question 207 * 208 * Description: 209 * The Linux block layer assumes that a block driver will consume all 210 * entries on the request queue when the request_fn strategy is called. 211 * Often this will not happen, because of hardware limitations (queue 212 * depth settings). If a device driver gets a 'queue full' response, 213 * or if it simply chooses not to queue more I/O at one point, it can 214 * call this function to prevent the request_fn from being called until 215 * the driver has signalled it's ready to go again. This happens by calling 216 * blk_start_queue() to restart queue operations. Queue lock must be held. 217 **/ 218 void blk_stop_queue(struct request_queue *q) 219 { 220 cancel_delayed_work(&q->delay_work); 221 queue_flag_set(QUEUE_FLAG_STOPPED, q); 222 } 223 EXPORT_SYMBOL(blk_stop_queue); 224 225 /** 226 * blk_sync_queue - cancel any pending callbacks on a queue 227 * @q: the queue 228 * 229 * Description: 230 * The block layer may perform asynchronous callback activity 231 * on a queue, such as calling the unplug function after a timeout. 232 * A block device may call blk_sync_queue to ensure that any 233 * such activity is cancelled, thus allowing it to release resources 234 * that the callbacks might use. The caller must already have made sure 235 * that its ->make_request_fn will not re-add plugging prior to calling 236 * this function. 237 * 238 * This function does not cancel any asynchronous activity arising 239 * out of elevator or throttling code. That would require elevator_exit() 240 * and blkcg_exit_queue() to be called with queue lock initialized. 241 * 242 */ 243 void blk_sync_queue(struct request_queue *q) 244 { 245 del_timer_sync(&q->timeout); 246 247 if (q->mq_ops) { 248 struct blk_mq_hw_ctx *hctx; 249 int i; 250 251 queue_for_each_hw_ctx(q, hctx, i) { 252 cancel_delayed_work_sync(&hctx->run_work); 253 cancel_delayed_work_sync(&hctx->delay_work); 254 } 255 } else { 256 cancel_delayed_work_sync(&q->delay_work); 257 } 258 } 259 EXPORT_SYMBOL(blk_sync_queue); 260 261 /** 262 * __blk_run_queue_uncond - run a queue whether or not it has been stopped 263 * @q: The queue to run 264 * 265 * Description: 266 * Invoke request handling on a queue if there are any pending requests. 267 * May be used to restart request handling after a request has completed. 268 * This variant runs the queue whether or not the queue has been 269 * stopped. Must be called with the queue lock held and interrupts 270 * disabled. See also @blk_run_queue. 271 */ 272 inline void __blk_run_queue_uncond(struct request_queue *q) 273 { 274 if (unlikely(blk_queue_dead(q))) 275 return; 276 277 /* 278 * Some request_fn implementations, e.g. scsi_request_fn(), unlock 279 * the queue lock internally. As a result multiple threads may be 280 * running such a request function concurrently. Keep track of the 281 * number of active request_fn invocations such that blk_drain_queue() 282 * can wait until all these request_fn calls have finished. 283 */ 284 q->request_fn_active++; 285 q->request_fn(q); 286 q->request_fn_active--; 287 } 288 289 /** 290 * __blk_run_queue - run a single device queue 291 * @q: The queue to run 292 * 293 * Description: 294 * See @blk_run_queue. This variant must be called with the queue lock 295 * held and interrupts disabled. 296 */ 297 void __blk_run_queue(struct request_queue *q) 298 { 299 if (unlikely(blk_queue_stopped(q))) 300 return; 301 302 __blk_run_queue_uncond(q); 303 } 304 EXPORT_SYMBOL(__blk_run_queue); 305 306 /** 307 * blk_run_queue_async - run a single device queue in workqueue context 308 * @q: The queue to run 309 * 310 * Description: 311 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf 312 * of us. The caller must hold the queue lock. 313 */ 314 void blk_run_queue_async(struct request_queue *q) 315 { 316 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) 317 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); 318 } 319 EXPORT_SYMBOL(blk_run_queue_async); 320 321 /** 322 * blk_run_queue - run a single device queue 323 * @q: The queue to run 324 * 325 * Description: 326 * Invoke request handling on this queue, if it has pending work to do. 327 * May be used to restart queueing when a request has completed. 328 */ 329 void blk_run_queue(struct request_queue *q) 330 { 331 unsigned long flags; 332 333 spin_lock_irqsave(q->queue_lock, flags); 334 __blk_run_queue(q); 335 spin_unlock_irqrestore(q->queue_lock, flags); 336 } 337 EXPORT_SYMBOL(blk_run_queue); 338 339 void blk_put_queue(struct request_queue *q) 340 { 341 kobject_put(&q->kobj); 342 } 343 EXPORT_SYMBOL(blk_put_queue); 344 345 /** 346 * __blk_drain_queue - drain requests from request_queue 347 * @q: queue to drain 348 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV 349 * 350 * Drain requests from @q. If @drain_all is set, all requests are drained. 351 * If not, only ELVPRIV requests are drained. The caller is responsible 352 * for ensuring that no new requests which need to be drained are queued. 353 */ 354 static void __blk_drain_queue(struct request_queue *q, bool drain_all) 355 __releases(q->queue_lock) 356 __acquires(q->queue_lock) 357 { 358 int i; 359 360 lockdep_assert_held(q->queue_lock); 361 362 while (true) { 363 bool drain = false; 364 365 /* 366 * The caller might be trying to drain @q before its 367 * elevator is initialized. 368 */ 369 if (q->elevator) 370 elv_drain_elevator(q); 371 372 blkcg_drain_queue(q); 373 374 /* 375 * This function might be called on a queue which failed 376 * driver init after queue creation or is not yet fully 377 * active yet. Some drivers (e.g. fd and loop) get unhappy 378 * in such cases. Kick queue iff dispatch queue has 379 * something on it and @q has request_fn set. 380 */ 381 if (!list_empty(&q->queue_head) && q->request_fn) 382 __blk_run_queue(q); 383 384 drain |= q->nr_rqs_elvpriv; 385 drain |= q->request_fn_active; 386 387 /* 388 * Unfortunately, requests are queued at and tracked from 389 * multiple places and there's no single counter which can 390 * be drained. Check all the queues and counters. 391 */ 392 if (drain_all) { 393 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); 394 drain |= !list_empty(&q->queue_head); 395 for (i = 0; i < 2; i++) { 396 drain |= q->nr_rqs[i]; 397 drain |= q->in_flight[i]; 398 if (fq) 399 drain |= !list_empty(&fq->flush_queue[i]); 400 } 401 } 402 403 if (!drain) 404 break; 405 406 spin_unlock_irq(q->queue_lock); 407 408 msleep(10); 409 410 spin_lock_irq(q->queue_lock); 411 } 412 413 /* 414 * With queue marked dead, any woken up waiter will fail the 415 * allocation path, so the wakeup chaining is lost and we're 416 * left with hung waiters. We need to wake up those waiters. 417 */ 418 if (q->request_fn) { 419 struct request_list *rl; 420 421 blk_queue_for_each_rl(rl, q) 422 for (i = 0; i < ARRAY_SIZE(rl->wait); i++) 423 wake_up_all(&rl->wait[i]); 424 } 425 } 426 427 /** 428 * blk_queue_bypass_start - enter queue bypass mode 429 * @q: queue of interest 430 * 431 * In bypass mode, only the dispatch FIFO queue of @q is used. This 432 * function makes @q enter bypass mode and drains all requests which were 433 * throttled or issued before. On return, it's guaranteed that no request 434 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true 435 * inside queue or RCU read lock. 436 */ 437 void blk_queue_bypass_start(struct request_queue *q) 438 { 439 spin_lock_irq(q->queue_lock); 440 q->bypass_depth++; 441 queue_flag_set(QUEUE_FLAG_BYPASS, q); 442 spin_unlock_irq(q->queue_lock); 443 444 /* 445 * Queues start drained. Skip actual draining till init is 446 * complete. This avoids lenghty delays during queue init which 447 * can happen many times during boot. 448 */ 449 if (blk_queue_init_done(q)) { 450 spin_lock_irq(q->queue_lock); 451 __blk_drain_queue(q, false); 452 spin_unlock_irq(q->queue_lock); 453 454 /* ensure blk_queue_bypass() is %true inside RCU read lock */ 455 synchronize_rcu(); 456 } 457 } 458 EXPORT_SYMBOL_GPL(blk_queue_bypass_start); 459 460 /** 461 * blk_queue_bypass_end - leave queue bypass mode 462 * @q: queue of interest 463 * 464 * Leave bypass mode and restore the normal queueing behavior. 465 */ 466 void blk_queue_bypass_end(struct request_queue *q) 467 { 468 spin_lock_irq(q->queue_lock); 469 if (!--q->bypass_depth) 470 queue_flag_clear(QUEUE_FLAG_BYPASS, q); 471 WARN_ON_ONCE(q->bypass_depth < 0); 472 spin_unlock_irq(q->queue_lock); 473 } 474 EXPORT_SYMBOL_GPL(blk_queue_bypass_end); 475 476 void blk_set_queue_dying(struct request_queue *q) 477 { 478 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q); 479 480 if (q->mq_ops) 481 blk_mq_wake_waiters(q); 482 else { 483 struct request_list *rl; 484 485 blk_queue_for_each_rl(rl, q) { 486 if (rl->rq_pool) { 487 wake_up(&rl->wait[BLK_RW_SYNC]); 488 wake_up(&rl->wait[BLK_RW_ASYNC]); 489 } 490 } 491 } 492 } 493 EXPORT_SYMBOL_GPL(blk_set_queue_dying); 494 495 /** 496 * blk_cleanup_queue - shutdown a request queue 497 * @q: request queue to shutdown 498 * 499 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and 500 * put it. All future requests will be failed immediately with -ENODEV. 501 */ 502 void blk_cleanup_queue(struct request_queue *q) 503 { 504 spinlock_t *lock = q->queue_lock; 505 506 /* mark @q DYING, no new request or merges will be allowed afterwards */ 507 mutex_lock(&q->sysfs_lock); 508 blk_set_queue_dying(q); 509 spin_lock_irq(lock); 510 511 /* 512 * A dying queue is permanently in bypass mode till released. Note 513 * that, unlike blk_queue_bypass_start(), we aren't performing 514 * synchronize_rcu() after entering bypass mode to avoid the delay 515 * as some drivers create and destroy a lot of queues while 516 * probing. This is still safe because blk_release_queue() will be 517 * called only after the queue refcnt drops to zero and nothing, 518 * RCU or not, would be traversing the queue by then. 519 */ 520 q->bypass_depth++; 521 queue_flag_set(QUEUE_FLAG_BYPASS, q); 522 523 queue_flag_set(QUEUE_FLAG_NOMERGES, q); 524 queue_flag_set(QUEUE_FLAG_NOXMERGES, q); 525 queue_flag_set(QUEUE_FLAG_DYING, q); 526 spin_unlock_irq(lock); 527 mutex_unlock(&q->sysfs_lock); 528 529 /* 530 * Drain all requests queued before DYING marking. Set DEAD flag to 531 * prevent that q->request_fn() gets invoked after draining finished. 532 */ 533 if (q->mq_ops) { 534 blk_mq_freeze_queue(q); 535 spin_lock_irq(lock); 536 } else { 537 spin_lock_irq(lock); 538 __blk_drain_queue(q, true); 539 } 540 queue_flag_set(QUEUE_FLAG_DEAD, q); 541 spin_unlock_irq(lock); 542 543 /* @q won't process any more request, flush async actions */ 544 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); 545 blk_sync_queue(q); 546 547 if (q->mq_ops) 548 blk_mq_free_queue(q); 549 550 spin_lock_irq(lock); 551 if (q->queue_lock != &q->__queue_lock) 552 q->queue_lock = &q->__queue_lock; 553 spin_unlock_irq(lock); 554 555 bdi_destroy(&q->backing_dev_info); 556 557 /* @q is and will stay empty, shutdown and put */ 558 blk_put_queue(q); 559 } 560 EXPORT_SYMBOL(blk_cleanup_queue); 561 562 /* Allocate memory local to the request queue */ 563 static void *alloc_request_struct(gfp_t gfp_mask, void *data) 564 { 565 int nid = (int)(long)data; 566 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid); 567 } 568 569 static void free_request_struct(void *element, void *unused) 570 { 571 kmem_cache_free(request_cachep, element); 572 } 573 574 int blk_init_rl(struct request_list *rl, struct request_queue *q, 575 gfp_t gfp_mask) 576 { 577 if (unlikely(rl->rq_pool)) 578 return 0; 579 580 rl->q = q; 581 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; 582 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; 583 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); 584 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); 585 586 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct, 587 free_request_struct, 588 (void *)(long)q->node, gfp_mask, 589 q->node); 590 if (!rl->rq_pool) 591 return -ENOMEM; 592 593 return 0; 594 } 595 596 void blk_exit_rl(struct request_list *rl) 597 { 598 if (rl->rq_pool) 599 mempool_destroy(rl->rq_pool); 600 } 601 602 struct request_queue *blk_alloc_queue(gfp_t gfp_mask) 603 { 604 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); 605 } 606 EXPORT_SYMBOL(blk_alloc_queue); 607 608 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) 609 { 610 struct request_queue *q; 611 int err; 612 613 q = kmem_cache_alloc_node(blk_requestq_cachep, 614 gfp_mask | __GFP_ZERO, node_id); 615 if (!q) 616 return NULL; 617 618 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); 619 if (q->id < 0) 620 goto fail_q; 621 622 q->backing_dev_info.ra_pages = 623 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; 624 q->backing_dev_info.state = 0; 625 q->backing_dev_info.capabilities = 0; 626 q->backing_dev_info.name = "block"; 627 q->node = node_id; 628 629 err = bdi_init(&q->backing_dev_info); 630 if (err) 631 goto fail_id; 632 633 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, 634 laptop_mode_timer_fn, (unsigned long) q); 635 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); 636 INIT_LIST_HEAD(&q->queue_head); 637 INIT_LIST_HEAD(&q->timeout_list); 638 INIT_LIST_HEAD(&q->icq_list); 639 #ifdef CONFIG_BLK_CGROUP 640 INIT_LIST_HEAD(&q->blkg_list); 641 #endif 642 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); 643 644 kobject_init(&q->kobj, &blk_queue_ktype); 645 646 mutex_init(&q->sysfs_lock); 647 spin_lock_init(&q->__queue_lock); 648 649 /* 650 * By default initialize queue_lock to internal lock and driver can 651 * override it later if need be. 652 */ 653 q->queue_lock = &q->__queue_lock; 654 655 /* 656 * A queue starts its life with bypass turned on to avoid 657 * unnecessary bypass on/off overhead and nasty surprises during 658 * init. The initial bypass will be finished when the queue is 659 * registered by blk_register_queue(). 660 */ 661 q->bypass_depth = 1; 662 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); 663 664 init_waitqueue_head(&q->mq_freeze_wq); 665 666 if (blkcg_init_queue(q)) 667 goto fail_bdi; 668 669 return q; 670 671 fail_bdi: 672 bdi_destroy(&q->backing_dev_info); 673 fail_id: 674 ida_simple_remove(&blk_queue_ida, q->id); 675 fail_q: 676 kmem_cache_free(blk_requestq_cachep, q); 677 return NULL; 678 } 679 EXPORT_SYMBOL(blk_alloc_queue_node); 680 681 /** 682 * blk_init_queue - prepare a request queue for use with a block device 683 * @rfn: The function to be called to process requests that have been 684 * placed on the queue. 685 * @lock: Request queue spin lock 686 * 687 * Description: 688 * If a block device wishes to use the standard request handling procedures, 689 * which sorts requests and coalesces adjacent requests, then it must 690 * call blk_init_queue(). The function @rfn will be called when there 691 * are requests on the queue that need to be processed. If the device 692 * supports plugging, then @rfn may not be called immediately when requests 693 * are available on the queue, but may be called at some time later instead. 694 * Plugged queues are generally unplugged when a buffer belonging to one 695 * of the requests on the queue is needed, or due to memory pressure. 696 * 697 * @rfn is not required, or even expected, to remove all requests off the 698 * queue, but only as many as it can handle at a time. If it does leave 699 * requests on the queue, it is responsible for arranging that the requests 700 * get dealt with eventually. 701 * 702 * The queue spin lock must be held while manipulating the requests on the 703 * request queue; this lock will be taken also from interrupt context, so irq 704 * disabling is needed for it. 705 * 706 * Function returns a pointer to the initialized request queue, or %NULL if 707 * it didn't succeed. 708 * 709 * Note: 710 * blk_init_queue() must be paired with a blk_cleanup_queue() call 711 * when the block device is deactivated (such as at module unload). 712 **/ 713 714 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) 715 { 716 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); 717 } 718 EXPORT_SYMBOL(blk_init_queue); 719 720 struct request_queue * 721 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) 722 { 723 struct request_queue *uninit_q, *q; 724 725 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); 726 if (!uninit_q) 727 return NULL; 728 729 q = blk_init_allocated_queue(uninit_q, rfn, lock); 730 if (!q) 731 blk_cleanup_queue(uninit_q); 732 733 return q; 734 } 735 EXPORT_SYMBOL(blk_init_queue_node); 736 737 static void blk_queue_bio(struct request_queue *q, struct bio *bio); 738 739 struct request_queue * 740 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, 741 spinlock_t *lock) 742 { 743 if (!q) 744 return NULL; 745 746 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0); 747 if (!q->fq) 748 return NULL; 749 750 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) 751 goto fail; 752 753 q->request_fn = rfn; 754 q->prep_rq_fn = NULL; 755 q->unprep_rq_fn = NULL; 756 q->queue_flags |= QUEUE_FLAG_DEFAULT; 757 758 /* Override internal queue lock with supplied lock pointer */ 759 if (lock) 760 q->queue_lock = lock; 761 762 /* 763 * This also sets hw/phys segments, boundary and size 764 */ 765 blk_queue_make_request(q, blk_queue_bio); 766 767 q->sg_reserved_size = INT_MAX; 768 769 /* Protect q->elevator from elevator_change */ 770 mutex_lock(&q->sysfs_lock); 771 772 /* init elevator */ 773 if (elevator_init(q, NULL)) { 774 mutex_unlock(&q->sysfs_lock); 775 goto fail; 776 } 777 778 mutex_unlock(&q->sysfs_lock); 779 780 return q; 781 782 fail: 783 blk_free_flush_queue(q->fq); 784 return NULL; 785 } 786 EXPORT_SYMBOL(blk_init_allocated_queue); 787 788 bool blk_get_queue(struct request_queue *q) 789 { 790 if (likely(!blk_queue_dying(q))) { 791 __blk_get_queue(q); 792 return true; 793 } 794 795 return false; 796 } 797 EXPORT_SYMBOL(blk_get_queue); 798 799 static inline void blk_free_request(struct request_list *rl, struct request *rq) 800 { 801 if (rq->cmd_flags & REQ_ELVPRIV) { 802 elv_put_request(rl->q, rq); 803 if (rq->elv.icq) 804 put_io_context(rq->elv.icq->ioc); 805 } 806 807 mempool_free(rq, rl->rq_pool); 808 } 809 810 /* 811 * ioc_batching returns true if the ioc is a valid batching request and 812 * should be given priority access to a request. 813 */ 814 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) 815 { 816 if (!ioc) 817 return 0; 818 819 /* 820 * Make sure the process is able to allocate at least 1 request 821 * even if the batch times out, otherwise we could theoretically 822 * lose wakeups. 823 */ 824 return ioc->nr_batch_requests == q->nr_batching || 825 (ioc->nr_batch_requests > 0 826 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); 827 } 828 829 /* 830 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This 831 * will cause the process to be a "batcher" on all queues in the system. This 832 * is the behaviour we want though - once it gets a wakeup it should be given 833 * a nice run. 834 */ 835 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) 836 { 837 if (!ioc || ioc_batching(q, ioc)) 838 return; 839 840 ioc->nr_batch_requests = q->nr_batching; 841 ioc->last_waited = jiffies; 842 } 843 844 static void __freed_request(struct request_list *rl, int sync) 845 { 846 struct request_queue *q = rl->q; 847 848 /* 849 * bdi isn't aware of blkcg yet. As all async IOs end up root 850 * blkcg anyway, just use root blkcg state. 851 */ 852 if (rl == &q->root_rl && 853 rl->count[sync] < queue_congestion_off_threshold(q)) 854 blk_clear_queue_congested(q, sync); 855 856 if (rl->count[sync] + 1 <= q->nr_requests) { 857 if (waitqueue_active(&rl->wait[sync])) 858 wake_up(&rl->wait[sync]); 859 860 blk_clear_rl_full(rl, sync); 861 } 862 } 863 864 /* 865 * A request has just been released. Account for it, update the full and 866 * congestion status, wake up any waiters. Called under q->queue_lock. 867 */ 868 static void freed_request(struct request_list *rl, unsigned int flags) 869 { 870 struct request_queue *q = rl->q; 871 int sync = rw_is_sync(flags); 872 873 q->nr_rqs[sync]--; 874 rl->count[sync]--; 875 if (flags & REQ_ELVPRIV) 876 q->nr_rqs_elvpriv--; 877 878 __freed_request(rl, sync); 879 880 if (unlikely(rl->starved[sync ^ 1])) 881 __freed_request(rl, sync ^ 1); 882 } 883 884 int blk_update_nr_requests(struct request_queue *q, unsigned int nr) 885 { 886 struct request_list *rl; 887 888 spin_lock_irq(q->queue_lock); 889 q->nr_requests = nr; 890 blk_queue_congestion_threshold(q); 891 892 /* congestion isn't cgroup aware and follows root blkcg for now */ 893 rl = &q->root_rl; 894 895 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q)) 896 blk_set_queue_congested(q, BLK_RW_SYNC); 897 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q)) 898 blk_clear_queue_congested(q, BLK_RW_SYNC); 899 900 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q)) 901 blk_set_queue_congested(q, BLK_RW_ASYNC); 902 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q)) 903 blk_clear_queue_congested(q, BLK_RW_ASYNC); 904 905 blk_queue_for_each_rl(rl, q) { 906 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { 907 blk_set_rl_full(rl, BLK_RW_SYNC); 908 } else { 909 blk_clear_rl_full(rl, BLK_RW_SYNC); 910 wake_up(&rl->wait[BLK_RW_SYNC]); 911 } 912 913 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { 914 blk_set_rl_full(rl, BLK_RW_ASYNC); 915 } else { 916 blk_clear_rl_full(rl, BLK_RW_ASYNC); 917 wake_up(&rl->wait[BLK_RW_ASYNC]); 918 } 919 } 920 921 spin_unlock_irq(q->queue_lock); 922 return 0; 923 } 924 925 /* 926 * Determine if elevator data should be initialized when allocating the 927 * request associated with @bio. 928 */ 929 static bool blk_rq_should_init_elevator(struct bio *bio) 930 { 931 if (!bio) 932 return true; 933 934 /* 935 * Flush requests do not use the elevator so skip initialization. 936 * This allows a request to share the flush and elevator data. 937 */ 938 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) 939 return false; 940 941 return true; 942 } 943 944 /** 945 * rq_ioc - determine io_context for request allocation 946 * @bio: request being allocated is for this bio (can be %NULL) 947 * 948 * Determine io_context to use for request allocation for @bio. May return 949 * %NULL if %current->io_context doesn't exist. 950 */ 951 static struct io_context *rq_ioc(struct bio *bio) 952 { 953 #ifdef CONFIG_BLK_CGROUP 954 if (bio && bio->bi_ioc) 955 return bio->bi_ioc; 956 #endif 957 return current->io_context; 958 } 959 960 /** 961 * __get_request - get a free request 962 * @rl: request list to allocate from 963 * @rw_flags: RW and SYNC flags 964 * @bio: bio to allocate request for (can be %NULL) 965 * @gfp_mask: allocation mask 966 * 967 * Get a free request from @q. This function may fail under memory 968 * pressure or if @q is dead. 969 * 970 * Must be called with @q->queue_lock held and, 971 * Returns ERR_PTR on failure, with @q->queue_lock held. 972 * Returns request pointer on success, with @q->queue_lock *not held*. 973 */ 974 static struct request *__get_request(struct request_list *rl, int rw_flags, 975 struct bio *bio, gfp_t gfp_mask) 976 { 977 struct request_queue *q = rl->q; 978 struct request *rq; 979 struct elevator_type *et = q->elevator->type; 980 struct io_context *ioc = rq_ioc(bio); 981 struct io_cq *icq = NULL; 982 const bool is_sync = rw_is_sync(rw_flags) != 0; 983 int may_queue; 984 985 if (unlikely(blk_queue_dying(q))) 986 return ERR_PTR(-ENODEV); 987 988 may_queue = elv_may_queue(q, rw_flags); 989 if (may_queue == ELV_MQUEUE_NO) 990 goto rq_starved; 991 992 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { 993 if (rl->count[is_sync]+1 >= q->nr_requests) { 994 /* 995 * The queue will fill after this allocation, so set 996 * it as full, and mark this process as "batching". 997 * This process will be allowed to complete a batch of 998 * requests, others will be blocked. 999 */ 1000 if (!blk_rl_full(rl, is_sync)) { 1001 ioc_set_batching(q, ioc); 1002 blk_set_rl_full(rl, is_sync); 1003 } else { 1004 if (may_queue != ELV_MQUEUE_MUST 1005 && !ioc_batching(q, ioc)) { 1006 /* 1007 * The queue is full and the allocating 1008 * process is not a "batcher", and not 1009 * exempted by the IO scheduler 1010 */ 1011 return ERR_PTR(-ENOMEM); 1012 } 1013 } 1014 } 1015 /* 1016 * bdi isn't aware of blkcg yet. As all async IOs end up 1017 * root blkcg anyway, just use root blkcg state. 1018 */ 1019 if (rl == &q->root_rl) 1020 blk_set_queue_congested(q, is_sync); 1021 } 1022 1023 /* 1024 * Only allow batching queuers to allocate up to 50% over the defined 1025 * limit of requests, otherwise we could have thousands of requests 1026 * allocated with any setting of ->nr_requests 1027 */ 1028 if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) 1029 return ERR_PTR(-ENOMEM); 1030 1031 q->nr_rqs[is_sync]++; 1032 rl->count[is_sync]++; 1033 rl->starved[is_sync] = 0; 1034 1035 /* 1036 * Decide whether the new request will be managed by elevator. If 1037 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will 1038 * prevent the current elevator from being destroyed until the new 1039 * request is freed. This guarantees icq's won't be destroyed and 1040 * makes creating new ones safe. 1041 * 1042 * Also, lookup icq while holding queue_lock. If it doesn't exist, 1043 * it will be created after releasing queue_lock. 1044 */ 1045 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { 1046 rw_flags |= REQ_ELVPRIV; 1047 q->nr_rqs_elvpriv++; 1048 if (et->icq_cache && ioc) 1049 icq = ioc_lookup_icq(ioc, q); 1050 } 1051 1052 if (blk_queue_io_stat(q)) 1053 rw_flags |= REQ_IO_STAT; 1054 spin_unlock_irq(q->queue_lock); 1055 1056 /* allocate and init request */ 1057 rq = mempool_alloc(rl->rq_pool, gfp_mask); 1058 if (!rq) 1059 goto fail_alloc; 1060 1061 blk_rq_init(q, rq); 1062 blk_rq_set_rl(rq, rl); 1063 rq->cmd_flags = rw_flags | REQ_ALLOCED; 1064 1065 /* init elvpriv */ 1066 if (rw_flags & REQ_ELVPRIV) { 1067 if (unlikely(et->icq_cache && !icq)) { 1068 if (ioc) 1069 icq = ioc_create_icq(ioc, q, gfp_mask); 1070 if (!icq) 1071 goto fail_elvpriv; 1072 } 1073 1074 rq->elv.icq = icq; 1075 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) 1076 goto fail_elvpriv; 1077 1078 /* @rq->elv.icq holds io_context until @rq is freed */ 1079 if (icq) 1080 get_io_context(icq->ioc); 1081 } 1082 out: 1083 /* 1084 * ioc may be NULL here, and ioc_batching will be false. That's 1085 * OK, if the queue is under the request limit then requests need 1086 * not count toward the nr_batch_requests limit. There will always 1087 * be some limit enforced by BLK_BATCH_TIME. 1088 */ 1089 if (ioc_batching(q, ioc)) 1090 ioc->nr_batch_requests--; 1091 1092 trace_block_getrq(q, bio, rw_flags & 1); 1093 return rq; 1094 1095 fail_elvpriv: 1096 /* 1097 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed 1098 * and may fail indefinitely under memory pressure and thus 1099 * shouldn't stall IO. Treat this request as !elvpriv. This will 1100 * disturb iosched and blkcg but weird is bettern than dead. 1101 */ 1102 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", 1103 __func__, dev_name(q->backing_dev_info.dev)); 1104 1105 rq->cmd_flags &= ~REQ_ELVPRIV; 1106 rq->elv.icq = NULL; 1107 1108 spin_lock_irq(q->queue_lock); 1109 q->nr_rqs_elvpriv--; 1110 spin_unlock_irq(q->queue_lock); 1111 goto out; 1112 1113 fail_alloc: 1114 /* 1115 * Allocation failed presumably due to memory. Undo anything we 1116 * might have messed up. 1117 * 1118 * Allocating task should really be put onto the front of the wait 1119 * queue, but this is pretty rare. 1120 */ 1121 spin_lock_irq(q->queue_lock); 1122 freed_request(rl, rw_flags); 1123 1124 /* 1125 * in the very unlikely event that allocation failed and no 1126 * requests for this direction was pending, mark us starved so that 1127 * freeing of a request in the other direction will notice 1128 * us. another possible fix would be to split the rq mempool into 1129 * READ and WRITE 1130 */ 1131 rq_starved: 1132 if (unlikely(rl->count[is_sync] == 0)) 1133 rl->starved[is_sync] = 1; 1134 return ERR_PTR(-ENOMEM); 1135 } 1136 1137 /** 1138 * get_request - get a free request 1139 * @q: request_queue to allocate request from 1140 * @rw_flags: RW and SYNC flags 1141 * @bio: bio to allocate request for (can be %NULL) 1142 * @gfp_mask: allocation mask 1143 * 1144 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this 1145 * function keeps retrying under memory pressure and fails iff @q is dead. 1146 * 1147 * Must be called with @q->queue_lock held and, 1148 * Returns ERR_PTR on failure, with @q->queue_lock held. 1149 * Returns request pointer on success, with @q->queue_lock *not held*. 1150 */ 1151 static struct request *get_request(struct request_queue *q, int rw_flags, 1152 struct bio *bio, gfp_t gfp_mask) 1153 { 1154 const bool is_sync = rw_is_sync(rw_flags) != 0; 1155 DEFINE_WAIT(wait); 1156 struct request_list *rl; 1157 struct request *rq; 1158 1159 rl = blk_get_rl(q, bio); /* transferred to @rq on success */ 1160 retry: 1161 rq = __get_request(rl, rw_flags, bio, gfp_mask); 1162 if (!IS_ERR(rq)) 1163 return rq; 1164 1165 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) { 1166 blk_put_rl(rl); 1167 return rq; 1168 } 1169 1170 /* wait on @rl and retry */ 1171 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, 1172 TASK_UNINTERRUPTIBLE); 1173 1174 trace_block_sleeprq(q, bio, rw_flags & 1); 1175 1176 spin_unlock_irq(q->queue_lock); 1177 io_schedule(); 1178 1179 /* 1180 * After sleeping, we become a "batching" process and will be able 1181 * to allocate at least one request, and up to a big batch of them 1182 * for a small period time. See ioc_batching, ioc_set_batching 1183 */ 1184 ioc_set_batching(q, current->io_context); 1185 1186 spin_lock_irq(q->queue_lock); 1187 finish_wait(&rl->wait[is_sync], &wait); 1188 1189 goto retry; 1190 } 1191 1192 static struct request *blk_old_get_request(struct request_queue *q, int rw, 1193 gfp_t gfp_mask) 1194 { 1195 struct request *rq; 1196 1197 BUG_ON(rw != READ && rw != WRITE); 1198 1199 /* create ioc upfront */ 1200 create_io_context(gfp_mask, q->node); 1201 1202 spin_lock_irq(q->queue_lock); 1203 rq = get_request(q, rw, NULL, gfp_mask); 1204 if (IS_ERR(rq)) 1205 spin_unlock_irq(q->queue_lock); 1206 /* q->queue_lock is unlocked at this point */ 1207 1208 return rq; 1209 } 1210 1211 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) 1212 { 1213 if (q->mq_ops) 1214 return blk_mq_alloc_request(q, rw, gfp_mask, false); 1215 else 1216 return blk_old_get_request(q, rw, gfp_mask); 1217 } 1218 EXPORT_SYMBOL(blk_get_request); 1219 1220 /** 1221 * blk_make_request - given a bio, allocate a corresponding struct request. 1222 * @q: target request queue 1223 * @bio: The bio describing the memory mappings that will be submitted for IO. 1224 * It may be a chained-bio properly constructed by block/bio layer. 1225 * @gfp_mask: gfp flags to be used for memory allocation 1226 * 1227 * blk_make_request is the parallel of generic_make_request for BLOCK_PC 1228 * type commands. Where the struct request needs to be farther initialized by 1229 * the caller. It is passed a &struct bio, which describes the memory info of 1230 * the I/O transfer. 1231 * 1232 * The caller of blk_make_request must make sure that bi_io_vec 1233 * are set to describe the memory buffers. That bio_data_dir() will return 1234 * the needed direction of the request. (And all bio's in the passed bio-chain 1235 * are properly set accordingly) 1236 * 1237 * If called under none-sleepable conditions, mapped bio buffers must not 1238 * need bouncing, by calling the appropriate masked or flagged allocator, 1239 * suitable for the target device. Otherwise the call to blk_queue_bounce will 1240 * BUG. 1241 * 1242 * WARNING: When allocating/cloning a bio-chain, careful consideration should be 1243 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for 1244 * anything but the first bio in the chain. Otherwise you risk waiting for IO 1245 * completion of a bio that hasn't been submitted yet, thus resulting in a 1246 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead 1247 * of bio_alloc(), as that avoids the mempool deadlock. 1248 * If possible a big IO should be split into smaller parts when allocation 1249 * fails. Partial allocation should not be an error, or you risk a live-lock. 1250 */ 1251 struct request *blk_make_request(struct request_queue *q, struct bio *bio, 1252 gfp_t gfp_mask) 1253 { 1254 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); 1255 1256 if (IS_ERR(rq)) 1257 return rq; 1258 1259 blk_rq_set_block_pc(rq); 1260 1261 for_each_bio(bio) { 1262 struct bio *bounce_bio = bio; 1263 int ret; 1264 1265 blk_queue_bounce(q, &bounce_bio); 1266 ret = blk_rq_append_bio(q, rq, bounce_bio); 1267 if (unlikely(ret)) { 1268 blk_put_request(rq); 1269 return ERR_PTR(ret); 1270 } 1271 } 1272 1273 return rq; 1274 } 1275 EXPORT_SYMBOL(blk_make_request); 1276 1277 /** 1278 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC 1279 * @rq: request to be initialized 1280 * 1281 */ 1282 void blk_rq_set_block_pc(struct request *rq) 1283 { 1284 rq->cmd_type = REQ_TYPE_BLOCK_PC; 1285 rq->__data_len = 0; 1286 rq->__sector = (sector_t) -1; 1287 rq->bio = rq->biotail = NULL; 1288 memset(rq->__cmd, 0, sizeof(rq->__cmd)); 1289 } 1290 EXPORT_SYMBOL(blk_rq_set_block_pc); 1291 1292 /** 1293 * blk_requeue_request - put a request back on queue 1294 * @q: request queue where request should be inserted 1295 * @rq: request to be inserted 1296 * 1297 * Description: 1298 * Drivers often keep queueing requests until the hardware cannot accept 1299 * more, when that condition happens we need to put the request back 1300 * on the queue. Must be called with queue lock held. 1301 */ 1302 void blk_requeue_request(struct request_queue *q, struct request *rq) 1303 { 1304 blk_delete_timer(rq); 1305 blk_clear_rq_complete(rq); 1306 trace_block_rq_requeue(q, rq); 1307 1308 if (rq->cmd_flags & REQ_QUEUED) 1309 blk_queue_end_tag(q, rq); 1310 1311 BUG_ON(blk_queued_rq(rq)); 1312 1313 elv_requeue_request(q, rq); 1314 } 1315 EXPORT_SYMBOL(blk_requeue_request); 1316 1317 static void add_acct_request(struct request_queue *q, struct request *rq, 1318 int where) 1319 { 1320 blk_account_io_start(rq, true); 1321 __elv_add_request(q, rq, where); 1322 } 1323 1324 static void part_round_stats_single(int cpu, struct hd_struct *part, 1325 unsigned long now) 1326 { 1327 int inflight; 1328 1329 if (now == part->stamp) 1330 return; 1331 1332 inflight = part_in_flight(part); 1333 if (inflight) { 1334 __part_stat_add(cpu, part, time_in_queue, 1335 inflight * (now - part->stamp)); 1336 __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); 1337 } 1338 part->stamp = now; 1339 } 1340 1341 /** 1342 * part_round_stats() - Round off the performance stats on a struct disk_stats. 1343 * @cpu: cpu number for stats access 1344 * @part: target partition 1345 * 1346 * The average IO queue length and utilisation statistics are maintained 1347 * by observing the current state of the queue length and the amount of 1348 * time it has been in this state for. 1349 * 1350 * Normally, that accounting is done on IO completion, but that can result 1351 * in more than a second's worth of IO being accounted for within any one 1352 * second, leading to >100% utilisation. To deal with that, we call this 1353 * function to do a round-off before returning the results when reading 1354 * /proc/diskstats. This accounts immediately for all queue usage up to 1355 * the current jiffies and restarts the counters again. 1356 */ 1357 void part_round_stats(int cpu, struct hd_struct *part) 1358 { 1359 unsigned long now = jiffies; 1360 1361 if (part->partno) 1362 part_round_stats_single(cpu, &part_to_disk(part)->part0, now); 1363 part_round_stats_single(cpu, part, now); 1364 } 1365 EXPORT_SYMBOL_GPL(part_round_stats); 1366 1367 #ifdef CONFIG_PM 1368 static void blk_pm_put_request(struct request *rq) 1369 { 1370 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending) 1371 pm_runtime_mark_last_busy(rq->q->dev); 1372 } 1373 #else 1374 static inline void blk_pm_put_request(struct request *rq) {} 1375 #endif 1376 1377 /* 1378 * queue lock must be held 1379 */ 1380 void __blk_put_request(struct request_queue *q, struct request *req) 1381 { 1382 if (unlikely(!q)) 1383 return; 1384 1385 if (q->mq_ops) { 1386 blk_mq_free_request(req); 1387 return; 1388 } 1389 1390 blk_pm_put_request(req); 1391 1392 elv_completed_request(q, req); 1393 1394 /* this is a bio leak */ 1395 WARN_ON(req->bio != NULL); 1396 1397 /* 1398 * Request may not have originated from ll_rw_blk. if not, 1399 * it didn't come out of our reserved rq pools 1400 */ 1401 if (req->cmd_flags & REQ_ALLOCED) { 1402 unsigned int flags = req->cmd_flags; 1403 struct request_list *rl = blk_rq_rl(req); 1404 1405 BUG_ON(!list_empty(&req->queuelist)); 1406 BUG_ON(ELV_ON_HASH(req)); 1407 1408 blk_free_request(rl, req); 1409 freed_request(rl, flags); 1410 blk_put_rl(rl); 1411 } 1412 } 1413 EXPORT_SYMBOL_GPL(__blk_put_request); 1414 1415 void blk_put_request(struct request *req) 1416 { 1417 struct request_queue *q = req->q; 1418 1419 if (q->mq_ops) 1420 blk_mq_free_request(req); 1421 else { 1422 unsigned long flags; 1423 1424 spin_lock_irqsave(q->queue_lock, flags); 1425 __blk_put_request(q, req); 1426 spin_unlock_irqrestore(q->queue_lock, flags); 1427 } 1428 } 1429 EXPORT_SYMBOL(blk_put_request); 1430 1431 /** 1432 * blk_add_request_payload - add a payload to a request 1433 * @rq: request to update 1434 * @page: page backing the payload 1435 * @len: length of the payload. 1436 * 1437 * This allows to later add a payload to an already submitted request by 1438 * a block driver. The driver needs to take care of freeing the payload 1439 * itself. 1440 * 1441 * Note that this is a quite horrible hack and nothing but handling of 1442 * discard requests should ever use it. 1443 */ 1444 void blk_add_request_payload(struct request *rq, struct page *page, 1445 unsigned int len) 1446 { 1447 struct bio *bio = rq->bio; 1448 1449 bio->bi_io_vec->bv_page = page; 1450 bio->bi_io_vec->bv_offset = 0; 1451 bio->bi_io_vec->bv_len = len; 1452 1453 bio->bi_iter.bi_size = len; 1454 bio->bi_vcnt = 1; 1455 bio->bi_phys_segments = 1; 1456 1457 rq->__data_len = rq->resid_len = len; 1458 rq->nr_phys_segments = 1; 1459 } 1460 EXPORT_SYMBOL_GPL(blk_add_request_payload); 1461 1462 bool bio_attempt_back_merge(struct request_queue *q, struct request *req, 1463 struct bio *bio) 1464 { 1465 const int ff = bio->bi_rw & REQ_FAILFAST_MASK; 1466 1467 if (!ll_back_merge_fn(q, req, bio)) 1468 return false; 1469 1470 trace_block_bio_backmerge(q, req, bio); 1471 1472 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 1473 blk_rq_set_mixed_merge(req); 1474 1475 req->biotail->bi_next = bio; 1476 req->biotail = bio; 1477 req->__data_len += bio->bi_iter.bi_size; 1478 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); 1479 1480 blk_account_io_start(req, false); 1481 return true; 1482 } 1483 1484 bool bio_attempt_front_merge(struct request_queue *q, struct request *req, 1485 struct bio *bio) 1486 { 1487 const int ff = bio->bi_rw & REQ_FAILFAST_MASK; 1488 1489 if (!ll_front_merge_fn(q, req, bio)) 1490 return false; 1491 1492 trace_block_bio_frontmerge(q, req, bio); 1493 1494 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 1495 blk_rq_set_mixed_merge(req); 1496 1497 bio->bi_next = req->bio; 1498 req->bio = bio; 1499 1500 req->__sector = bio->bi_iter.bi_sector; 1501 req->__data_len += bio->bi_iter.bi_size; 1502 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); 1503 1504 blk_account_io_start(req, false); 1505 return true; 1506 } 1507 1508 /** 1509 * blk_attempt_plug_merge - try to merge with %current's plugged list 1510 * @q: request_queue new bio is being queued at 1511 * @bio: new bio being queued 1512 * @request_count: out parameter for number of traversed plugged requests 1513 * 1514 * Determine whether @bio being queued on @q can be merged with a request 1515 * on %current's plugged list. Returns %true if merge was successful, 1516 * otherwise %false. 1517 * 1518 * Plugging coalesces IOs from the same issuer for the same purpose without 1519 * going through @q->queue_lock. As such it's more of an issuing mechanism 1520 * than scheduling, and the request, while may have elvpriv data, is not 1521 * added on the elevator at this point. In addition, we don't have 1522 * reliable access to the elevator outside queue lock. Only check basic 1523 * merging parameters without querying the elevator. 1524 * 1525 * Caller must ensure !blk_queue_nomerges(q) beforehand. 1526 */ 1527 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, 1528 unsigned int *request_count) 1529 { 1530 struct blk_plug *plug; 1531 struct request *rq; 1532 bool ret = false; 1533 struct list_head *plug_list; 1534 1535 plug = current->plug; 1536 if (!plug) 1537 goto out; 1538 *request_count = 0; 1539 1540 if (q->mq_ops) 1541 plug_list = &plug->mq_list; 1542 else 1543 plug_list = &plug->list; 1544 1545 list_for_each_entry_reverse(rq, plug_list, queuelist) { 1546 int el_ret; 1547 1548 if (rq->q == q) 1549 (*request_count)++; 1550 1551 if (rq->q != q || !blk_rq_merge_ok(rq, bio)) 1552 continue; 1553 1554 el_ret = blk_try_merge(rq, bio); 1555 if (el_ret == ELEVATOR_BACK_MERGE) { 1556 ret = bio_attempt_back_merge(q, rq, bio); 1557 if (ret) 1558 break; 1559 } else if (el_ret == ELEVATOR_FRONT_MERGE) { 1560 ret = bio_attempt_front_merge(q, rq, bio); 1561 if (ret) 1562 break; 1563 } 1564 } 1565 out: 1566 return ret; 1567 } 1568 1569 void init_request_from_bio(struct request *req, struct bio *bio) 1570 { 1571 req->cmd_type = REQ_TYPE_FS; 1572 1573 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; 1574 if (bio->bi_rw & REQ_RAHEAD) 1575 req->cmd_flags |= REQ_FAILFAST_MASK; 1576 1577 req->errors = 0; 1578 req->__sector = bio->bi_iter.bi_sector; 1579 req->ioprio = bio_prio(bio); 1580 blk_rq_bio_prep(req->q, req, bio); 1581 } 1582 1583 static void blk_queue_bio(struct request_queue *q, struct bio *bio) 1584 { 1585 const bool sync = !!(bio->bi_rw & REQ_SYNC); 1586 struct blk_plug *plug; 1587 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; 1588 struct request *req; 1589 unsigned int request_count = 0; 1590 1591 /* 1592 * low level driver can indicate that it wants pages above a 1593 * certain limit bounced to low memory (ie for highmem, or even 1594 * ISA dma in theory) 1595 */ 1596 blk_queue_bounce(q, &bio); 1597 1598 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) { 1599 bio_endio(bio, -EIO); 1600 return; 1601 } 1602 1603 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { 1604 spin_lock_irq(q->queue_lock); 1605 where = ELEVATOR_INSERT_FLUSH; 1606 goto get_rq; 1607 } 1608 1609 /* 1610 * Check if we can merge with the plugged list before grabbing 1611 * any locks. 1612 */ 1613 if (!blk_queue_nomerges(q) && 1614 blk_attempt_plug_merge(q, bio, &request_count)) 1615 return; 1616 1617 spin_lock_irq(q->queue_lock); 1618 1619 el_ret = elv_merge(q, &req, bio); 1620 if (el_ret == ELEVATOR_BACK_MERGE) { 1621 if (bio_attempt_back_merge(q, req, bio)) { 1622 elv_bio_merged(q, req, bio); 1623 if (!attempt_back_merge(q, req)) 1624 elv_merged_request(q, req, el_ret); 1625 goto out_unlock; 1626 } 1627 } else if (el_ret == ELEVATOR_FRONT_MERGE) { 1628 if (bio_attempt_front_merge(q, req, bio)) { 1629 elv_bio_merged(q, req, bio); 1630 if (!attempt_front_merge(q, req)) 1631 elv_merged_request(q, req, el_ret); 1632 goto out_unlock; 1633 } 1634 } 1635 1636 get_rq: 1637 /* 1638 * This sync check and mask will be re-done in init_request_from_bio(), 1639 * but we need to set it earlier to expose the sync flag to the 1640 * rq allocator and io schedulers. 1641 */ 1642 rw_flags = bio_data_dir(bio); 1643 if (sync) 1644 rw_flags |= REQ_SYNC; 1645 1646 /* 1647 * Grab a free request. This is might sleep but can not fail. 1648 * Returns with the queue unlocked. 1649 */ 1650 req = get_request(q, rw_flags, bio, GFP_NOIO); 1651 if (IS_ERR(req)) { 1652 bio_endio(bio, PTR_ERR(req)); /* @q is dead */ 1653 goto out_unlock; 1654 } 1655 1656 /* 1657 * After dropping the lock and possibly sleeping here, our request 1658 * may now be mergeable after it had proven unmergeable (above). 1659 * We don't worry about that case for efficiency. It won't happen 1660 * often, and the elevators are able to handle it. 1661 */ 1662 init_request_from_bio(req, bio); 1663 1664 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) 1665 req->cpu = raw_smp_processor_id(); 1666 1667 plug = current->plug; 1668 if (plug) { 1669 /* 1670 * If this is the first request added after a plug, fire 1671 * of a plug trace. 1672 */ 1673 if (!request_count) 1674 trace_block_plug(q); 1675 else { 1676 if (request_count >= BLK_MAX_REQUEST_COUNT) { 1677 blk_flush_plug_list(plug, false); 1678 trace_block_plug(q); 1679 } 1680 } 1681 list_add_tail(&req->queuelist, &plug->list); 1682 blk_account_io_start(req, true); 1683 } else { 1684 spin_lock_irq(q->queue_lock); 1685 add_acct_request(q, req, where); 1686 __blk_run_queue(q); 1687 out_unlock: 1688 spin_unlock_irq(q->queue_lock); 1689 } 1690 } 1691 1692 /* 1693 * If bio->bi_dev is a partition, remap the location 1694 */ 1695 static inline void blk_partition_remap(struct bio *bio) 1696 { 1697 struct block_device *bdev = bio->bi_bdev; 1698 1699 if (bio_sectors(bio) && bdev != bdev->bd_contains) { 1700 struct hd_struct *p = bdev->bd_part; 1701 1702 bio->bi_iter.bi_sector += p->start_sect; 1703 bio->bi_bdev = bdev->bd_contains; 1704 1705 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, 1706 bdev->bd_dev, 1707 bio->bi_iter.bi_sector - p->start_sect); 1708 } 1709 } 1710 1711 static void handle_bad_sector(struct bio *bio) 1712 { 1713 char b[BDEVNAME_SIZE]; 1714 1715 printk(KERN_INFO "attempt to access beyond end of device\n"); 1716 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", 1717 bdevname(bio->bi_bdev, b), 1718 bio->bi_rw, 1719 (unsigned long long)bio_end_sector(bio), 1720 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); 1721 1722 set_bit(BIO_EOF, &bio->bi_flags); 1723 } 1724 1725 #ifdef CONFIG_FAIL_MAKE_REQUEST 1726 1727 static DECLARE_FAULT_ATTR(fail_make_request); 1728 1729 static int __init setup_fail_make_request(char *str) 1730 { 1731 return setup_fault_attr(&fail_make_request, str); 1732 } 1733 __setup("fail_make_request=", setup_fail_make_request); 1734 1735 static bool should_fail_request(struct hd_struct *part, unsigned int bytes) 1736 { 1737 return part->make_it_fail && should_fail(&fail_make_request, bytes); 1738 } 1739 1740 static int __init fail_make_request_debugfs(void) 1741 { 1742 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 1743 NULL, &fail_make_request); 1744 1745 return PTR_ERR_OR_ZERO(dir); 1746 } 1747 1748 late_initcall(fail_make_request_debugfs); 1749 1750 #else /* CONFIG_FAIL_MAKE_REQUEST */ 1751 1752 static inline bool should_fail_request(struct hd_struct *part, 1753 unsigned int bytes) 1754 { 1755 return false; 1756 } 1757 1758 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 1759 1760 /* 1761 * Check whether this bio extends beyond the end of the device. 1762 */ 1763 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) 1764 { 1765 sector_t maxsector; 1766 1767 if (!nr_sectors) 1768 return 0; 1769 1770 /* Test device or partition size, when known. */ 1771 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; 1772 if (maxsector) { 1773 sector_t sector = bio->bi_iter.bi_sector; 1774 1775 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { 1776 /* 1777 * This may well happen - the kernel calls bread() 1778 * without checking the size of the device, e.g., when 1779 * mounting a device. 1780 */ 1781 handle_bad_sector(bio); 1782 return 1; 1783 } 1784 } 1785 1786 return 0; 1787 } 1788 1789 static noinline_for_stack bool 1790 generic_make_request_checks(struct bio *bio) 1791 { 1792 struct request_queue *q; 1793 int nr_sectors = bio_sectors(bio); 1794 int err = -EIO; 1795 char b[BDEVNAME_SIZE]; 1796 struct hd_struct *part; 1797 1798 might_sleep(); 1799 1800 if (bio_check_eod(bio, nr_sectors)) 1801 goto end_io; 1802 1803 q = bdev_get_queue(bio->bi_bdev); 1804 if (unlikely(!q)) { 1805 printk(KERN_ERR 1806 "generic_make_request: Trying to access " 1807 "nonexistent block-device %s (%Lu)\n", 1808 bdevname(bio->bi_bdev, b), 1809 (long long) bio->bi_iter.bi_sector); 1810 goto end_io; 1811 } 1812 1813 if (likely(bio_is_rw(bio) && 1814 nr_sectors > queue_max_hw_sectors(q))) { 1815 printk(KERN_ERR "bio too big device %s (%u > %u)\n", 1816 bdevname(bio->bi_bdev, b), 1817 bio_sectors(bio), 1818 queue_max_hw_sectors(q)); 1819 goto end_io; 1820 } 1821 1822 part = bio->bi_bdev->bd_part; 1823 if (should_fail_request(part, bio->bi_iter.bi_size) || 1824 should_fail_request(&part_to_disk(part)->part0, 1825 bio->bi_iter.bi_size)) 1826 goto end_io; 1827 1828 /* 1829 * If this device has partitions, remap block n 1830 * of partition p to block n+start(p) of the disk. 1831 */ 1832 blk_partition_remap(bio); 1833 1834 if (bio_check_eod(bio, nr_sectors)) 1835 goto end_io; 1836 1837 /* 1838 * Filter flush bio's early so that make_request based 1839 * drivers without flush support don't have to worry 1840 * about them. 1841 */ 1842 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { 1843 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); 1844 if (!nr_sectors) { 1845 err = 0; 1846 goto end_io; 1847 } 1848 } 1849 1850 if ((bio->bi_rw & REQ_DISCARD) && 1851 (!blk_queue_discard(q) || 1852 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) { 1853 err = -EOPNOTSUPP; 1854 goto end_io; 1855 } 1856 1857 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) { 1858 err = -EOPNOTSUPP; 1859 goto end_io; 1860 } 1861 1862 /* 1863 * Various block parts want %current->io_context and lazy ioc 1864 * allocation ends up trading a lot of pain for a small amount of 1865 * memory. Just allocate it upfront. This may fail and block 1866 * layer knows how to live with it. 1867 */ 1868 create_io_context(GFP_ATOMIC, q->node); 1869 1870 if (blk_throtl_bio(q, bio)) 1871 return false; /* throttled, will be resubmitted later */ 1872 1873 trace_block_bio_queue(q, bio); 1874 return true; 1875 1876 end_io: 1877 bio_endio(bio, err); 1878 return false; 1879 } 1880 1881 /** 1882 * generic_make_request - hand a buffer to its device driver for I/O 1883 * @bio: The bio describing the location in memory and on the device. 1884 * 1885 * generic_make_request() is used to make I/O requests of block 1886 * devices. It is passed a &struct bio, which describes the I/O that needs 1887 * to be done. 1888 * 1889 * generic_make_request() does not return any status. The 1890 * success/failure status of the request, along with notification of 1891 * completion, is delivered asynchronously through the bio->bi_end_io 1892 * function described (one day) else where. 1893 * 1894 * The caller of generic_make_request must make sure that bi_io_vec 1895 * are set to describe the memory buffer, and that bi_dev and bi_sector are 1896 * set to describe the device address, and the 1897 * bi_end_io and optionally bi_private are set to describe how 1898 * completion notification should be signaled. 1899 * 1900 * generic_make_request and the drivers it calls may use bi_next if this 1901 * bio happens to be merged with someone else, and may resubmit the bio to 1902 * a lower device by calling into generic_make_request recursively, which 1903 * means the bio should NOT be touched after the call to ->make_request_fn. 1904 */ 1905 void generic_make_request(struct bio *bio) 1906 { 1907 struct bio_list bio_list_on_stack; 1908 1909 if (!generic_make_request_checks(bio)) 1910 return; 1911 1912 /* 1913 * We only want one ->make_request_fn to be active at a time, else 1914 * stack usage with stacked devices could be a problem. So use 1915 * current->bio_list to keep a list of requests submited by a 1916 * make_request_fn function. current->bio_list is also used as a 1917 * flag to say if generic_make_request is currently active in this 1918 * task or not. If it is NULL, then no make_request is active. If 1919 * it is non-NULL, then a make_request is active, and new requests 1920 * should be added at the tail 1921 */ 1922 if (current->bio_list) { 1923 bio_list_add(current->bio_list, bio); 1924 return; 1925 } 1926 1927 /* following loop may be a bit non-obvious, and so deserves some 1928 * explanation. 1929 * Before entering the loop, bio->bi_next is NULL (as all callers 1930 * ensure that) so we have a list with a single bio. 1931 * We pretend that we have just taken it off a longer list, so 1932 * we assign bio_list to a pointer to the bio_list_on_stack, 1933 * thus initialising the bio_list of new bios to be 1934 * added. ->make_request() may indeed add some more bios 1935 * through a recursive call to generic_make_request. If it 1936 * did, we find a non-NULL value in bio_list and re-enter the loop 1937 * from the top. In this case we really did just take the bio 1938 * of the top of the list (no pretending) and so remove it from 1939 * bio_list, and call into ->make_request() again. 1940 */ 1941 BUG_ON(bio->bi_next); 1942 bio_list_init(&bio_list_on_stack); 1943 current->bio_list = &bio_list_on_stack; 1944 do { 1945 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1946 1947 q->make_request_fn(q, bio); 1948 1949 bio = bio_list_pop(current->bio_list); 1950 } while (bio); 1951 current->bio_list = NULL; /* deactivate */ 1952 } 1953 EXPORT_SYMBOL(generic_make_request); 1954 1955 /** 1956 * submit_bio - submit a bio to the block device layer for I/O 1957 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) 1958 * @bio: The &struct bio which describes the I/O 1959 * 1960 * submit_bio() is very similar in purpose to generic_make_request(), and 1961 * uses that function to do most of the work. Both are fairly rough 1962 * interfaces; @bio must be presetup and ready for I/O. 1963 * 1964 */ 1965 void submit_bio(int rw, struct bio *bio) 1966 { 1967 bio->bi_rw |= rw; 1968 1969 /* 1970 * If it's a regular read/write or a barrier with data attached, 1971 * go through the normal accounting stuff before submission. 1972 */ 1973 if (bio_has_data(bio)) { 1974 unsigned int count; 1975 1976 if (unlikely(rw & REQ_WRITE_SAME)) 1977 count = bdev_logical_block_size(bio->bi_bdev) >> 9; 1978 else 1979 count = bio_sectors(bio); 1980 1981 if (rw & WRITE) { 1982 count_vm_events(PGPGOUT, count); 1983 } else { 1984 task_io_account_read(bio->bi_iter.bi_size); 1985 count_vm_events(PGPGIN, count); 1986 } 1987 1988 if (unlikely(block_dump)) { 1989 char b[BDEVNAME_SIZE]; 1990 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", 1991 current->comm, task_pid_nr(current), 1992 (rw & WRITE) ? "WRITE" : "READ", 1993 (unsigned long long)bio->bi_iter.bi_sector, 1994 bdevname(bio->bi_bdev, b), 1995 count); 1996 } 1997 } 1998 1999 generic_make_request(bio); 2000 } 2001 EXPORT_SYMBOL(submit_bio); 2002 2003 /** 2004 * blk_rq_check_limits - Helper function to check a request for the queue limit 2005 * @q: the queue 2006 * @rq: the request being checked 2007 * 2008 * Description: 2009 * @rq may have been made based on weaker limitations of upper-level queues 2010 * in request stacking drivers, and it may violate the limitation of @q. 2011 * Since the block layer and the underlying device driver trust @rq 2012 * after it is inserted to @q, it should be checked against @q before 2013 * the insertion using this generic function. 2014 * 2015 * This function should also be useful for request stacking drivers 2016 * in some cases below, so export this function. 2017 * Request stacking drivers like request-based dm may change the queue 2018 * limits while requests are in the queue (e.g. dm's table swapping). 2019 * Such request stacking drivers should check those requests against 2020 * the new queue limits again when they dispatch those requests, 2021 * although such checkings are also done against the old queue limits 2022 * when submitting requests. 2023 */ 2024 int blk_rq_check_limits(struct request_queue *q, struct request *rq) 2025 { 2026 if (!rq_mergeable(rq)) 2027 return 0; 2028 2029 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) { 2030 printk(KERN_ERR "%s: over max size limit.\n", __func__); 2031 return -EIO; 2032 } 2033 2034 /* 2035 * queue's settings related to segment counting like q->bounce_pfn 2036 * may differ from that of other stacking queues. 2037 * Recalculate it to check the request correctly on this queue's 2038 * limitation. 2039 */ 2040 blk_recalc_rq_segments(rq); 2041 if (rq->nr_phys_segments > queue_max_segments(q)) { 2042 printk(KERN_ERR "%s: over max segments limit.\n", __func__); 2043 return -EIO; 2044 } 2045 2046 return 0; 2047 } 2048 EXPORT_SYMBOL_GPL(blk_rq_check_limits); 2049 2050 /** 2051 * blk_insert_cloned_request - Helper for stacking drivers to submit a request 2052 * @q: the queue to submit the request 2053 * @rq: the request being queued 2054 */ 2055 int blk_insert_cloned_request(struct request_queue *q, struct request *rq) 2056 { 2057 unsigned long flags; 2058 int where = ELEVATOR_INSERT_BACK; 2059 2060 if (blk_rq_check_limits(q, rq)) 2061 return -EIO; 2062 2063 if (rq->rq_disk && 2064 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) 2065 return -EIO; 2066 2067 if (q->mq_ops) { 2068 if (blk_queue_io_stat(q)) 2069 blk_account_io_start(rq, true); 2070 blk_mq_insert_request(rq, false, true, true); 2071 return 0; 2072 } 2073 2074 spin_lock_irqsave(q->queue_lock, flags); 2075 if (unlikely(blk_queue_dying(q))) { 2076 spin_unlock_irqrestore(q->queue_lock, flags); 2077 return -ENODEV; 2078 } 2079 2080 /* 2081 * Submitting request must be dequeued before calling this function 2082 * because it will be linked to another request_queue 2083 */ 2084 BUG_ON(blk_queued_rq(rq)); 2085 2086 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) 2087 where = ELEVATOR_INSERT_FLUSH; 2088 2089 add_acct_request(q, rq, where); 2090 if (where == ELEVATOR_INSERT_FLUSH) 2091 __blk_run_queue(q); 2092 spin_unlock_irqrestore(q->queue_lock, flags); 2093 2094 return 0; 2095 } 2096 EXPORT_SYMBOL_GPL(blk_insert_cloned_request); 2097 2098 /** 2099 * blk_rq_err_bytes - determine number of bytes till the next failure boundary 2100 * @rq: request to examine 2101 * 2102 * Description: 2103 * A request could be merge of IOs which require different failure 2104 * handling. This function determines the number of bytes which 2105 * can be failed from the beginning of the request without 2106 * crossing into area which need to be retried further. 2107 * 2108 * Return: 2109 * The number of bytes to fail. 2110 * 2111 * Context: 2112 * queue_lock must be held. 2113 */ 2114 unsigned int blk_rq_err_bytes(const struct request *rq) 2115 { 2116 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; 2117 unsigned int bytes = 0; 2118 struct bio *bio; 2119 2120 if (!(rq->cmd_flags & REQ_MIXED_MERGE)) 2121 return blk_rq_bytes(rq); 2122 2123 /* 2124 * Currently the only 'mixing' which can happen is between 2125 * different fastfail types. We can safely fail portions 2126 * which have all the failfast bits that the first one has - 2127 * the ones which are at least as eager to fail as the first 2128 * one. 2129 */ 2130 for (bio = rq->bio; bio; bio = bio->bi_next) { 2131 if ((bio->bi_rw & ff) != ff) 2132 break; 2133 bytes += bio->bi_iter.bi_size; 2134 } 2135 2136 /* this could lead to infinite loop */ 2137 BUG_ON(blk_rq_bytes(rq) && !bytes); 2138 return bytes; 2139 } 2140 EXPORT_SYMBOL_GPL(blk_rq_err_bytes); 2141 2142 void blk_account_io_completion(struct request *req, unsigned int bytes) 2143 { 2144 if (blk_do_io_stat(req)) { 2145 const int rw = rq_data_dir(req); 2146 struct hd_struct *part; 2147 int cpu; 2148 2149 cpu = part_stat_lock(); 2150 part = req->part; 2151 part_stat_add(cpu, part, sectors[rw], bytes >> 9); 2152 part_stat_unlock(); 2153 } 2154 } 2155 2156 void blk_account_io_done(struct request *req) 2157 { 2158 /* 2159 * Account IO completion. flush_rq isn't accounted as a 2160 * normal IO on queueing nor completion. Accounting the 2161 * containing request is enough. 2162 */ 2163 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { 2164 unsigned long duration = jiffies - req->start_time; 2165 const int rw = rq_data_dir(req); 2166 struct hd_struct *part; 2167 int cpu; 2168 2169 cpu = part_stat_lock(); 2170 part = req->part; 2171 2172 part_stat_inc(cpu, part, ios[rw]); 2173 part_stat_add(cpu, part, ticks[rw], duration); 2174 part_round_stats(cpu, part); 2175 part_dec_in_flight(part, rw); 2176 2177 hd_struct_put(part); 2178 part_stat_unlock(); 2179 } 2180 } 2181 2182 #ifdef CONFIG_PM 2183 /* 2184 * Don't process normal requests when queue is suspended 2185 * or in the process of suspending/resuming 2186 */ 2187 static struct request *blk_pm_peek_request(struct request_queue *q, 2188 struct request *rq) 2189 { 2190 if (q->dev && (q->rpm_status == RPM_SUSPENDED || 2191 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM)))) 2192 return NULL; 2193 else 2194 return rq; 2195 } 2196 #else 2197 static inline struct request *blk_pm_peek_request(struct request_queue *q, 2198 struct request *rq) 2199 { 2200 return rq; 2201 } 2202 #endif 2203 2204 void blk_account_io_start(struct request *rq, bool new_io) 2205 { 2206 struct hd_struct *part; 2207 int rw = rq_data_dir(rq); 2208 int cpu; 2209 2210 if (!blk_do_io_stat(rq)) 2211 return; 2212 2213 cpu = part_stat_lock(); 2214 2215 if (!new_io) { 2216 part = rq->part; 2217 part_stat_inc(cpu, part, merges[rw]); 2218 } else { 2219 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); 2220 if (!hd_struct_try_get(part)) { 2221 /* 2222 * The partition is already being removed, 2223 * the request will be accounted on the disk only 2224 * 2225 * We take a reference on disk->part0 although that 2226 * partition will never be deleted, so we can treat 2227 * it as any other partition. 2228 */ 2229 part = &rq->rq_disk->part0; 2230 hd_struct_get(part); 2231 } 2232 part_round_stats(cpu, part); 2233 part_inc_in_flight(part, rw); 2234 rq->part = part; 2235 } 2236 2237 part_stat_unlock(); 2238 } 2239 2240 /** 2241 * blk_peek_request - peek at the top of a request queue 2242 * @q: request queue to peek at 2243 * 2244 * Description: 2245 * Return the request at the top of @q. The returned request 2246 * should be started using blk_start_request() before LLD starts 2247 * processing it. 2248 * 2249 * Return: 2250 * Pointer to the request at the top of @q if available. Null 2251 * otherwise. 2252 * 2253 * Context: 2254 * queue_lock must be held. 2255 */ 2256 struct request *blk_peek_request(struct request_queue *q) 2257 { 2258 struct request *rq; 2259 int ret; 2260 2261 while ((rq = __elv_next_request(q)) != NULL) { 2262 2263 rq = blk_pm_peek_request(q, rq); 2264 if (!rq) 2265 break; 2266 2267 if (!(rq->cmd_flags & REQ_STARTED)) { 2268 /* 2269 * This is the first time the device driver 2270 * sees this request (possibly after 2271 * requeueing). Notify IO scheduler. 2272 */ 2273 if (rq->cmd_flags & REQ_SORTED) 2274 elv_activate_rq(q, rq); 2275 2276 /* 2277 * just mark as started even if we don't start 2278 * it, a request that has been delayed should 2279 * not be passed by new incoming requests 2280 */ 2281 rq->cmd_flags |= REQ_STARTED; 2282 trace_block_rq_issue(q, rq); 2283 } 2284 2285 if (!q->boundary_rq || q->boundary_rq == rq) { 2286 q->end_sector = rq_end_sector(rq); 2287 q->boundary_rq = NULL; 2288 } 2289 2290 if (rq->cmd_flags & REQ_DONTPREP) 2291 break; 2292 2293 if (q->dma_drain_size && blk_rq_bytes(rq)) { 2294 /* 2295 * make sure space for the drain appears we 2296 * know we can do this because max_hw_segments 2297 * has been adjusted to be one fewer than the 2298 * device can handle 2299 */ 2300 rq->nr_phys_segments++; 2301 } 2302 2303 if (!q->prep_rq_fn) 2304 break; 2305 2306 ret = q->prep_rq_fn(q, rq); 2307 if (ret == BLKPREP_OK) { 2308 break; 2309 } else if (ret == BLKPREP_DEFER) { 2310 /* 2311 * the request may have been (partially) prepped. 2312 * we need to keep this request in the front to 2313 * avoid resource deadlock. REQ_STARTED will 2314 * prevent other fs requests from passing this one. 2315 */ 2316 if (q->dma_drain_size && blk_rq_bytes(rq) && 2317 !(rq->cmd_flags & REQ_DONTPREP)) { 2318 /* 2319 * remove the space for the drain we added 2320 * so that we don't add it again 2321 */ 2322 --rq->nr_phys_segments; 2323 } 2324 2325 rq = NULL; 2326 break; 2327 } else if (ret == BLKPREP_KILL) { 2328 rq->cmd_flags |= REQ_QUIET; 2329 /* 2330 * Mark this request as started so we don't trigger 2331 * any debug logic in the end I/O path. 2332 */ 2333 blk_start_request(rq); 2334 __blk_end_request_all(rq, -EIO); 2335 } else { 2336 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); 2337 break; 2338 } 2339 } 2340 2341 return rq; 2342 } 2343 EXPORT_SYMBOL(blk_peek_request); 2344 2345 void blk_dequeue_request(struct request *rq) 2346 { 2347 struct request_queue *q = rq->q; 2348 2349 BUG_ON(list_empty(&rq->queuelist)); 2350 BUG_ON(ELV_ON_HASH(rq)); 2351 2352 list_del_init(&rq->queuelist); 2353 2354 /* 2355 * the time frame between a request being removed from the lists 2356 * and to it is freed is accounted as io that is in progress at 2357 * the driver side. 2358 */ 2359 if (blk_account_rq(rq)) { 2360 q->in_flight[rq_is_sync(rq)]++; 2361 set_io_start_time_ns(rq); 2362 } 2363 } 2364 2365 /** 2366 * blk_start_request - start request processing on the driver 2367 * @req: request to dequeue 2368 * 2369 * Description: 2370 * Dequeue @req and start timeout timer on it. This hands off the 2371 * request to the driver. 2372 * 2373 * Block internal functions which don't want to start timer should 2374 * call blk_dequeue_request(). 2375 * 2376 * Context: 2377 * queue_lock must be held. 2378 */ 2379 void blk_start_request(struct request *req) 2380 { 2381 blk_dequeue_request(req); 2382 2383 /* 2384 * We are now handing the request to the hardware, initialize 2385 * resid_len to full count and add the timeout handler. 2386 */ 2387 req->resid_len = blk_rq_bytes(req); 2388 if (unlikely(blk_bidi_rq(req))) 2389 req->next_rq->resid_len = blk_rq_bytes(req->next_rq); 2390 2391 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); 2392 blk_add_timer(req); 2393 } 2394 EXPORT_SYMBOL(blk_start_request); 2395 2396 /** 2397 * blk_fetch_request - fetch a request from a request queue 2398 * @q: request queue to fetch a request from 2399 * 2400 * Description: 2401 * Return the request at the top of @q. The request is started on 2402 * return and LLD can start processing it immediately. 2403 * 2404 * Return: 2405 * Pointer to the request at the top of @q if available. Null 2406 * otherwise. 2407 * 2408 * Context: 2409 * queue_lock must be held. 2410 */ 2411 struct request *blk_fetch_request(struct request_queue *q) 2412 { 2413 struct request *rq; 2414 2415 rq = blk_peek_request(q); 2416 if (rq) 2417 blk_start_request(rq); 2418 return rq; 2419 } 2420 EXPORT_SYMBOL(blk_fetch_request); 2421 2422 /** 2423 * blk_update_request - Special helper function for request stacking drivers 2424 * @req: the request being processed 2425 * @error: %0 for success, < %0 for error 2426 * @nr_bytes: number of bytes to complete @req 2427 * 2428 * Description: 2429 * Ends I/O on a number of bytes attached to @req, but doesn't complete 2430 * the request structure even if @req doesn't have leftover. 2431 * If @req has leftover, sets it up for the next range of segments. 2432 * 2433 * This special helper function is only for request stacking drivers 2434 * (e.g. request-based dm) so that they can handle partial completion. 2435 * Actual device drivers should use blk_end_request instead. 2436 * 2437 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees 2438 * %false return from this function. 2439 * 2440 * Return: 2441 * %false - this request doesn't have any more data 2442 * %true - this request has more data 2443 **/ 2444 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) 2445 { 2446 int total_bytes; 2447 2448 trace_block_rq_complete(req->q, req, nr_bytes); 2449 2450 if (!req->bio) 2451 return false; 2452 2453 /* 2454 * For fs requests, rq is just carrier of independent bio's 2455 * and each partial completion should be handled separately. 2456 * Reset per-request error on each partial completion. 2457 * 2458 * TODO: tj: This is too subtle. It would be better to let 2459 * low level drivers do what they see fit. 2460 */ 2461 if (req->cmd_type == REQ_TYPE_FS) 2462 req->errors = 0; 2463 2464 if (error && req->cmd_type == REQ_TYPE_FS && 2465 !(req->cmd_flags & REQ_QUIET)) { 2466 char *error_type; 2467 2468 switch (error) { 2469 case -ENOLINK: 2470 error_type = "recoverable transport"; 2471 break; 2472 case -EREMOTEIO: 2473 error_type = "critical target"; 2474 break; 2475 case -EBADE: 2476 error_type = "critical nexus"; 2477 break; 2478 case -ETIMEDOUT: 2479 error_type = "timeout"; 2480 break; 2481 case -ENOSPC: 2482 error_type = "critical space allocation"; 2483 break; 2484 case -ENODATA: 2485 error_type = "critical medium"; 2486 break; 2487 case -EIO: 2488 default: 2489 error_type = "I/O"; 2490 break; 2491 } 2492 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n", 2493 __func__, error_type, req->rq_disk ? 2494 req->rq_disk->disk_name : "?", 2495 (unsigned long long)blk_rq_pos(req)); 2496 2497 } 2498 2499 blk_account_io_completion(req, nr_bytes); 2500 2501 total_bytes = 0; 2502 while (req->bio) { 2503 struct bio *bio = req->bio; 2504 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); 2505 2506 if (bio_bytes == bio->bi_iter.bi_size) 2507 req->bio = bio->bi_next; 2508 2509 req_bio_endio(req, bio, bio_bytes, error); 2510 2511 total_bytes += bio_bytes; 2512 nr_bytes -= bio_bytes; 2513 2514 if (!nr_bytes) 2515 break; 2516 } 2517 2518 /* 2519 * completely done 2520 */ 2521 if (!req->bio) { 2522 /* 2523 * Reset counters so that the request stacking driver 2524 * can find how many bytes remain in the request 2525 * later. 2526 */ 2527 req->__data_len = 0; 2528 return false; 2529 } 2530 2531 req->__data_len -= total_bytes; 2532 2533 /* update sector only for requests with clear definition of sector */ 2534 if (req->cmd_type == REQ_TYPE_FS) 2535 req->__sector += total_bytes >> 9; 2536 2537 /* mixed attributes always follow the first bio */ 2538 if (req->cmd_flags & REQ_MIXED_MERGE) { 2539 req->cmd_flags &= ~REQ_FAILFAST_MASK; 2540 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; 2541 } 2542 2543 /* 2544 * If total number of sectors is less than the first segment 2545 * size, something has gone terribly wrong. 2546 */ 2547 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { 2548 blk_dump_rq_flags(req, "request botched"); 2549 req->__data_len = blk_rq_cur_bytes(req); 2550 } 2551 2552 /* recalculate the number of segments */ 2553 blk_recalc_rq_segments(req); 2554 2555 return true; 2556 } 2557 EXPORT_SYMBOL_GPL(blk_update_request); 2558 2559 static bool blk_update_bidi_request(struct request *rq, int error, 2560 unsigned int nr_bytes, 2561 unsigned int bidi_bytes) 2562 { 2563 if (blk_update_request(rq, error, nr_bytes)) 2564 return true; 2565 2566 /* Bidi request must be completed as a whole */ 2567 if (unlikely(blk_bidi_rq(rq)) && 2568 blk_update_request(rq->next_rq, error, bidi_bytes)) 2569 return true; 2570 2571 if (blk_queue_add_random(rq->q)) 2572 add_disk_randomness(rq->rq_disk); 2573 2574 return false; 2575 } 2576 2577 /** 2578 * blk_unprep_request - unprepare a request 2579 * @req: the request 2580 * 2581 * This function makes a request ready for complete resubmission (or 2582 * completion). It happens only after all error handling is complete, 2583 * so represents the appropriate moment to deallocate any resources 2584 * that were allocated to the request in the prep_rq_fn. The queue 2585 * lock is held when calling this. 2586 */ 2587 void blk_unprep_request(struct request *req) 2588 { 2589 struct request_queue *q = req->q; 2590 2591 req->cmd_flags &= ~REQ_DONTPREP; 2592 if (q->unprep_rq_fn) 2593 q->unprep_rq_fn(q, req); 2594 } 2595 EXPORT_SYMBOL_GPL(blk_unprep_request); 2596 2597 /* 2598 * queue lock must be held 2599 */ 2600 void blk_finish_request(struct request *req, int error) 2601 { 2602 if (req->cmd_flags & REQ_QUEUED) 2603 blk_queue_end_tag(req->q, req); 2604 2605 BUG_ON(blk_queued_rq(req)); 2606 2607 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) 2608 laptop_io_completion(&req->q->backing_dev_info); 2609 2610 blk_delete_timer(req); 2611 2612 if (req->cmd_flags & REQ_DONTPREP) 2613 blk_unprep_request(req); 2614 2615 blk_account_io_done(req); 2616 2617 if (req->end_io) 2618 req->end_io(req, error); 2619 else { 2620 if (blk_bidi_rq(req)) 2621 __blk_put_request(req->next_rq->q, req->next_rq); 2622 2623 __blk_put_request(req->q, req); 2624 } 2625 } 2626 EXPORT_SYMBOL(blk_finish_request); 2627 2628 /** 2629 * blk_end_bidi_request - Complete a bidi request 2630 * @rq: the request to complete 2631 * @error: %0 for success, < %0 for error 2632 * @nr_bytes: number of bytes to complete @rq 2633 * @bidi_bytes: number of bytes to complete @rq->next_rq 2634 * 2635 * Description: 2636 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. 2637 * Drivers that supports bidi can safely call this member for any 2638 * type of request, bidi or uni. In the later case @bidi_bytes is 2639 * just ignored. 2640 * 2641 * Return: 2642 * %false - we are done with this request 2643 * %true - still buffers pending for this request 2644 **/ 2645 static bool blk_end_bidi_request(struct request *rq, int error, 2646 unsigned int nr_bytes, unsigned int bidi_bytes) 2647 { 2648 struct request_queue *q = rq->q; 2649 unsigned long flags; 2650 2651 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) 2652 return true; 2653 2654 spin_lock_irqsave(q->queue_lock, flags); 2655 blk_finish_request(rq, error); 2656 spin_unlock_irqrestore(q->queue_lock, flags); 2657 2658 return false; 2659 } 2660 2661 /** 2662 * __blk_end_bidi_request - Complete a bidi request with queue lock held 2663 * @rq: the request to complete 2664 * @error: %0 for success, < %0 for error 2665 * @nr_bytes: number of bytes to complete @rq 2666 * @bidi_bytes: number of bytes to complete @rq->next_rq 2667 * 2668 * Description: 2669 * Identical to blk_end_bidi_request() except that queue lock is 2670 * assumed to be locked on entry and remains so on return. 2671 * 2672 * Return: 2673 * %false - we are done with this request 2674 * %true - still buffers pending for this request 2675 **/ 2676 bool __blk_end_bidi_request(struct request *rq, int error, 2677 unsigned int nr_bytes, unsigned int bidi_bytes) 2678 { 2679 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) 2680 return true; 2681 2682 blk_finish_request(rq, error); 2683 2684 return false; 2685 } 2686 2687 /** 2688 * blk_end_request - Helper function for drivers to complete the request. 2689 * @rq: the request being processed 2690 * @error: %0 for success, < %0 for error 2691 * @nr_bytes: number of bytes to complete 2692 * 2693 * Description: 2694 * Ends I/O on a number of bytes attached to @rq. 2695 * If @rq has leftover, sets it up for the next range of segments. 2696 * 2697 * Return: 2698 * %false - we are done with this request 2699 * %true - still buffers pending for this request 2700 **/ 2701 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 2702 { 2703 return blk_end_bidi_request(rq, error, nr_bytes, 0); 2704 } 2705 EXPORT_SYMBOL(blk_end_request); 2706 2707 /** 2708 * blk_end_request_all - Helper function for drives to finish the request. 2709 * @rq: the request to finish 2710 * @error: %0 for success, < %0 for error 2711 * 2712 * Description: 2713 * Completely finish @rq. 2714 */ 2715 void blk_end_request_all(struct request *rq, int error) 2716 { 2717 bool pending; 2718 unsigned int bidi_bytes = 0; 2719 2720 if (unlikely(blk_bidi_rq(rq))) 2721 bidi_bytes = blk_rq_bytes(rq->next_rq); 2722 2723 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); 2724 BUG_ON(pending); 2725 } 2726 EXPORT_SYMBOL(blk_end_request_all); 2727 2728 /** 2729 * blk_end_request_cur - Helper function to finish the current request chunk. 2730 * @rq: the request to finish the current chunk for 2731 * @error: %0 for success, < %0 for error 2732 * 2733 * Description: 2734 * Complete the current consecutively mapped chunk from @rq. 2735 * 2736 * Return: 2737 * %false - we are done with this request 2738 * %true - still buffers pending for this request 2739 */ 2740 bool blk_end_request_cur(struct request *rq, int error) 2741 { 2742 return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); 2743 } 2744 EXPORT_SYMBOL(blk_end_request_cur); 2745 2746 /** 2747 * blk_end_request_err - Finish a request till the next failure boundary. 2748 * @rq: the request to finish till the next failure boundary for 2749 * @error: must be negative errno 2750 * 2751 * Description: 2752 * Complete @rq till the next failure boundary. 2753 * 2754 * Return: 2755 * %false - we are done with this request 2756 * %true - still buffers pending for this request 2757 */ 2758 bool blk_end_request_err(struct request *rq, int error) 2759 { 2760 WARN_ON(error >= 0); 2761 return blk_end_request(rq, error, blk_rq_err_bytes(rq)); 2762 } 2763 EXPORT_SYMBOL_GPL(blk_end_request_err); 2764 2765 /** 2766 * __blk_end_request - Helper function for drivers to complete the request. 2767 * @rq: the request being processed 2768 * @error: %0 for success, < %0 for error 2769 * @nr_bytes: number of bytes to complete 2770 * 2771 * Description: 2772 * Must be called with queue lock held unlike blk_end_request(). 2773 * 2774 * Return: 2775 * %false - we are done with this request 2776 * %true - still buffers pending for this request 2777 **/ 2778 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 2779 { 2780 return __blk_end_bidi_request(rq, error, nr_bytes, 0); 2781 } 2782 EXPORT_SYMBOL(__blk_end_request); 2783 2784 /** 2785 * __blk_end_request_all - Helper function for drives to finish the request. 2786 * @rq: the request to finish 2787 * @error: %0 for success, < %0 for error 2788 * 2789 * Description: 2790 * Completely finish @rq. Must be called with queue lock held. 2791 */ 2792 void __blk_end_request_all(struct request *rq, int error) 2793 { 2794 bool pending; 2795 unsigned int bidi_bytes = 0; 2796 2797 if (unlikely(blk_bidi_rq(rq))) 2798 bidi_bytes = blk_rq_bytes(rq->next_rq); 2799 2800 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); 2801 BUG_ON(pending); 2802 } 2803 EXPORT_SYMBOL(__blk_end_request_all); 2804 2805 /** 2806 * __blk_end_request_cur - Helper function to finish the current request chunk. 2807 * @rq: the request to finish the current chunk for 2808 * @error: %0 for success, < %0 for error 2809 * 2810 * Description: 2811 * Complete the current consecutively mapped chunk from @rq. Must 2812 * be called with queue lock held. 2813 * 2814 * Return: 2815 * %false - we are done with this request 2816 * %true - still buffers pending for this request 2817 */ 2818 bool __blk_end_request_cur(struct request *rq, int error) 2819 { 2820 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); 2821 } 2822 EXPORT_SYMBOL(__blk_end_request_cur); 2823 2824 /** 2825 * __blk_end_request_err - Finish a request till the next failure boundary. 2826 * @rq: the request to finish till the next failure boundary for 2827 * @error: must be negative errno 2828 * 2829 * Description: 2830 * Complete @rq till the next failure boundary. Must be called 2831 * with queue lock held. 2832 * 2833 * Return: 2834 * %false - we are done with this request 2835 * %true - still buffers pending for this request 2836 */ 2837 bool __blk_end_request_err(struct request *rq, int error) 2838 { 2839 WARN_ON(error >= 0); 2840 return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); 2841 } 2842 EXPORT_SYMBOL_GPL(__blk_end_request_err); 2843 2844 void blk_rq_bio_prep(struct request_queue *q, struct request *rq, 2845 struct bio *bio) 2846 { 2847 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ 2848 rq->cmd_flags |= bio->bi_rw & REQ_WRITE; 2849 2850 if (bio_has_data(bio)) 2851 rq->nr_phys_segments = bio_phys_segments(q, bio); 2852 2853 rq->__data_len = bio->bi_iter.bi_size; 2854 rq->bio = rq->biotail = bio; 2855 2856 if (bio->bi_bdev) 2857 rq->rq_disk = bio->bi_bdev->bd_disk; 2858 } 2859 2860 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 2861 /** 2862 * rq_flush_dcache_pages - Helper function to flush all pages in a request 2863 * @rq: the request to be flushed 2864 * 2865 * Description: 2866 * Flush all pages in @rq. 2867 */ 2868 void rq_flush_dcache_pages(struct request *rq) 2869 { 2870 struct req_iterator iter; 2871 struct bio_vec bvec; 2872 2873 rq_for_each_segment(bvec, rq, iter) 2874 flush_dcache_page(bvec.bv_page); 2875 } 2876 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); 2877 #endif 2878 2879 /** 2880 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 2881 * @q : the queue of the device being checked 2882 * 2883 * Description: 2884 * Check if underlying low-level drivers of a device are busy. 2885 * If the drivers want to export their busy state, they must set own 2886 * exporting function using blk_queue_lld_busy() first. 2887 * 2888 * Basically, this function is used only by request stacking drivers 2889 * to stop dispatching requests to underlying devices when underlying 2890 * devices are busy. This behavior helps more I/O merging on the queue 2891 * of the request stacking driver and prevents I/O throughput regression 2892 * on burst I/O load. 2893 * 2894 * Return: 2895 * 0 - Not busy (The request stacking driver should dispatch request) 2896 * 1 - Busy (The request stacking driver should stop dispatching request) 2897 */ 2898 int blk_lld_busy(struct request_queue *q) 2899 { 2900 if (q->lld_busy_fn) 2901 return q->lld_busy_fn(q); 2902 2903 return 0; 2904 } 2905 EXPORT_SYMBOL_GPL(blk_lld_busy); 2906 2907 /** 2908 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request 2909 * @rq: the clone request to be cleaned up 2910 * 2911 * Description: 2912 * Free all bios in @rq for a cloned request. 2913 */ 2914 void blk_rq_unprep_clone(struct request *rq) 2915 { 2916 struct bio *bio; 2917 2918 while ((bio = rq->bio) != NULL) { 2919 rq->bio = bio->bi_next; 2920 2921 bio_put(bio); 2922 } 2923 } 2924 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); 2925 2926 /* 2927 * Copy attributes of the original request to the clone request. 2928 * The actual data parts (e.g. ->cmd, ->sense) are not copied. 2929 */ 2930 static void __blk_rq_prep_clone(struct request *dst, struct request *src) 2931 { 2932 dst->cpu = src->cpu; 2933 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; 2934 dst->cmd_type = src->cmd_type; 2935 dst->__sector = blk_rq_pos(src); 2936 dst->__data_len = blk_rq_bytes(src); 2937 dst->nr_phys_segments = src->nr_phys_segments; 2938 dst->ioprio = src->ioprio; 2939 dst->extra_len = src->extra_len; 2940 } 2941 2942 /** 2943 * blk_rq_prep_clone - Helper function to setup clone request 2944 * @rq: the request to be setup 2945 * @rq_src: original request to be cloned 2946 * @bs: bio_set that bios for clone are allocated from 2947 * @gfp_mask: memory allocation mask for bio 2948 * @bio_ctr: setup function to be called for each clone bio. 2949 * Returns %0 for success, non %0 for failure. 2950 * @data: private data to be passed to @bio_ctr 2951 * 2952 * Description: 2953 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. 2954 * The actual data parts of @rq_src (e.g. ->cmd, ->sense) 2955 * are not copied, and copying such parts is the caller's responsibility. 2956 * Also, pages which the original bios are pointing to are not copied 2957 * and the cloned bios just point same pages. 2958 * So cloned bios must be completed before original bios, which means 2959 * the caller must complete @rq before @rq_src. 2960 */ 2961 int blk_rq_prep_clone(struct request *rq, struct request *rq_src, 2962 struct bio_set *bs, gfp_t gfp_mask, 2963 int (*bio_ctr)(struct bio *, struct bio *, void *), 2964 void *data) 2965 { 2966 struct bio *bio, *bio_src; 2967 2968 if (!bs) 2969 bs = fs_bio_set; 2970 2971 __rq_for_each_bio(bio_src, rq_src) { 2972 bio = bio_clone_fast(bio_src, gfp_mask, bs); 2973 if (!bio) 2974 goto free_and_out; 2975 2976 if (bio_ctr && bio_ctr(bio, bio_src, data)) 2977 goto free_and_out; 2978 2979 if (rq->bio) { 2980 rq->biotail->bi_next = bio; 2981 rq->biotail = bio; 2982 } else 2983 rq->bio = rq->biotail = bio; 2984 } 2985 2986 __blk_rq_prep_clone(rq, rq_src); 2987 2988 return 0; 2989 2990 free_and_out: 2991 if (bio) 2992 bio_put(bio); 2993 blk_rq_unprep_clone(rq); 2994 2995 return -ENOMEM; 2996 } 2997 EXPORT_SYMBOL_GPL(blk_rq_prep_clone); 2998 2999 int kblockd_schedule_work(struct work_struct *work) 3000 { 3001 return queue_work(kblockd_workqueue, work); 3002 } 3003 EXPORT_SYMBOL(kblockd_schedule_work); 3004 3005 int kblockd_schedule_delayed_work(struct delayed_work *dwork, 3006 unsigned long delay) 3007 { 3008 return queue_delayed_work(kblockd_workqueue, dwork, delay); 3009 } 3010 EXPORT_SYMBOL(kblockd_schedule_delayed_work); 3011 3012 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, 3013 unsigned long delay) 3014 { 3015 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); 3016 } 3017 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on); 3018 3019 /** 3020 * blk_start_plug - initialize blk_plug and track it inside the task_struct 3021 * @plug: The &struct blk_plug that needs to be initialized 3022 * 3023 * Description: 3024 * Tracking blk_plug inside the task_struct will help with auto-flushing the 3025 * pending I/O should the task end up blocking between blk_start_plug() and 3026 * blk_finish_plug(). This is important from a performance perspective, but 3027 * also ensures that we don't deadlock. For instance, if the task is blocking 3028 * for a memory allocation, memory reclaim could end up wanting to free a 3029 * page belonging to that request that is currently residing in our private 3030 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 3031 * this kind of deadlock. 3032 */ 3033 void blk_start_plug(struct blk_plug *plug) 3034 { 3035 struct task_struct *tsk = current; 3036 3037 INIT_LIST_HEAD(&plug->list); 3038 INIT_LIST_HEAD(&plug->mq_list); 3039 INIT_LIST_HEAD(&plug->cb_list); 3040 3041 /* 3042 * If this is a nested plug, don't actually assign it. It will be 3043 * flushed on its own. 3044 */ 3045 if (!tsk->plug) { 3046 /* 3047 * Store ordering should not be needed here, since a potential 3048 * preempt will imply a full memory barrier 3049 */ 3050 tsk->plug = plug; 3051 } 3052 } 3053 EXPORT_SYMBOL(blk_start_plug); 3054 3055 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) 3056 { 3057 struct request *rqa = container_of(a, struct request, queuelist); 3058 struct request *rqb = container_of(b, struct request, queuelist); 3059 3060 return !(rqa->q < rqb->q || 3061 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); 3062 } 3063 3064 /* 3065 * If 'from_schedule' is true, then postpone the dispatch of requests 3066 * until a safe kblockd context. We due this to avoid accidental big 3067 * additional stack usage in driver dispatch, in places where the originally 3068 * plugger did not intend it. 3069 */ 3070 static void queue_unplugged(struct request_queue *q, unsigned int depth, 3071 bool from_schedule) 3072 __releases(q->queue_lock) 3073 { 3074 trace_block_unplug(q, depth, !from_schedule); 3075 3076 if (from_schedule) 3077 blk_run_queue_async(q); 3078 else 3079 __blk_run_queue(q); 3080 spin_unlock(q->queue_lock); 3081 } 3082 3083 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 3084 { 3085 LIST_HEAD(callbacks); 3086 3087 while (!list_empty(&plug->cb_list)) { 3088 list_splice_init(&plug->cb_list, &callbacks); 3089 3090 while (!list_empty(&callbacks)) { 3091 struct blk_plug_cb *cb = list_first_entry(&callbacks, 3092 struct blk_plug_cb, 3093 list); 3094 list_del(&cb->list); 3095 cb->callback(cb, from_schedule); 3096 } 3097 } 3098 } 3099 3100 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 3101 int size) 3102 { 3103 struct blk_plug *plug = current->plug; 3104 struct blk_plug_cb *cb; 3105 3106 if (!plug) 3107 return NULL; 3108 3109 list_for_each_entry(cb, &plug->cb_list, list) 3110 if (cb->callback == unplug && cb->data == data) 3111 return cb; 3112 3113 /* Not currently on the callback list */ 3114 BUG_ON(size < sizeof(*cb)); 3115 cb = kzalloc(size, GFP_ATOMIC); 3116 if (cb) { 3117 cb->data = data; 3118 cb->callback = unplug; 3119 list_add(&cb->list, &plug->cb_list); 3120 } 3121 return cb; 3122 } 3123 EXPORT_SYMBOL(blk_check_plugged); 3124 3125 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) 3126 { 3127 struct request_queue *q; 3128 unsigned long flags; 3129 struct request *rq; 3130 LIST_HEAD(list); 3131 unsigned int depth; 3132 3133 flush_plug_callbacks(plug, from_schedule); 3134 3135 if (!list_empty(&plug->mq_list)) 3136 blk_mq_flush_plug_list(plug, from_schedule); 3137 3138 if (list_empty(&plug->list)) 3139 return; 3140 3141 list_splice_init(&plug->list, &list); 3142 3143 list_sort(NULL, &list, plug_rq_cmp); 3144 3145 q = NULL; 3146 depth = 0; 3147 3148 /* 3149 * Save and disable interrupts here, to avoid doing it for every 3150 * queue lock we have to take. 3151 */ 3152 local_irq_save(flags); 3153 while (!list_empty(&list)) { 3154 rq = list_entry_rq(list.next); 3155 list_del_init(&rq->queuelist); 3156 BUG_ON(!rq->q); 3157 if (rq->q != q) { 3158 /* 3159 * This drops the queue lock 3160 */ 3161 if (q) 3162 queue_unplugged(q, depth, from_schedule); 3163 q = rq->q; 3164 depth = 0; 3165 spin_lock(q->queue_lock); 3166 } 3167 3168 /* 3169 * Short-circuit if @q is dead 3170 */ 3171 if (unlikely(blk_queue_dying(q))) { 3172 __blk_end_request_all(rq, -ENODEV); 3173 continue; 3174 } 3175 3176 /* 3177 * rq is already accounted, so use raw insert 3178 */ 3179 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) 3180 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); 3181 else 3182 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); 3183 3184 depth++; 3185 } 3186 3187 /* 3188 * This drops the queue lock 3189 */ 3190 if (q) 3191 queue_unplugged(q, depth, from_schedule); 3192 3193 local_irq_restore(flags); 3194 } 3195 3196 void blk_finish_plug(struct blk_plug *plug) 3197 { 3198 blk_flush_plug_list(plug, false); 3199 3200 if (plug == current->plug) 3201 current->plug = NULL; 3202 } 3203 EXPORT_SYMBOL(blk_finish_plug); 3204 3205 #ifdef CONFIG_PM 3206 /** 3207 * blk_pm_runtime_init - Block layer runtime PM initialization routine 3208 * @q: the queue of the device 3209 * @dev: the device the queue belongs to 3210 * 3211 * Description: 3212 * Initialize runtime-PM-related fields for @q and start auto suspend for 3213 * @dev. Drivers that want to take advantage of request-based runtime PM 3214 * should call this function after @dev has been initialized, and its 3215 * request queue @q has been allocated, and runtime PM for it can not happen 3216 * yet(either due to disabled/forbidden or its usage_count > 0). In most 3217 * cases, driver should call this function before any I/O has taken place. 3218 * 3219 * This function takes care of setting up using auto suspend for the device, 3220 * the autosuspend delay is set to -1 to make runtime suspend impossible 3221 * until an updated value is either set by user or by driver. Drivers do 3222 * not need to touch other autosuspend settings. 3223 * 3224 * The block layer runtime PM is request based, so only works for drivers 3225 * that use request as their IO unit instead of those directly use bio's. 3226 */ 3227 void blk_pm_runtime_init(struct request_queue *q, struct device *dev) 3228 { 3229 q->dev = dev; 3230 q->rpm_status = RPM_ACTIVE; 3231 pm_runtime_set_autosuspend_delay(q->dev, -1); 3232 pm_runtime_use_autosuspend(q->dev); 3233 } 3234 EXPORT_SYMBOL(blk_pm_runtime_init); 3235 3236 /** 3237 * blk_pre_runtime_suspend - Pre runtime suspend check 3238 * @q: the queue of the device 3239 * 3240 * Description: 3241 * This function will check if runtime suspend is allowed for the device 3242 * by examining if there are any requests pending in the queue. If there 3243 * are requests pending, the device can not be runtime suspended; otherwise, 3244 * the queue's status will be updated to SUSPENDING and the driver can 3245 * proceed to suspend the device. 3246 * 3247 * For the not allowed case, we mark last busy for the device so that 3248 * runtime PM core will try to autosuspend it some time later. 3249 * 3250 * This function should be called near the start of the device's 3251 * runtime_suspend callback. 3252 * 3253 * Return: 3254 * 0 - OK to runtime suspend the device 3255 * -EBUSY - Device should not be runtime suspended 3256 */ 3257 int blk_pre_runtime_suspend(struct request_queue *q) 3258 { 3259 int ret = 0; 3260 3261 spin_lock_irq(q->queue_lock); 3262 if (q->nr_pending) { 3263 ret = -EBUSY; 3264 pm_runtime_mark_last_busy(q->dev); 3265 } else { 3266 q->rpm_status = RPM_SUSPENDING; 3267 } 3268 spin_unlock_irq(q->queue_lock); 3269 return ret; 3270 } 3271 EXPORT_SYMBOL(blk_pre_runtime_suspend); 3272 3273 /** 3274 * blk_post_runtime_suspend - Post runtime suspend processing 3275 * @q: the queue of the device 3276 * @err: return value of the device's runtime_suspend function 3277 * 3278 * Description: 3279 * Update the queue's runtime status according to the return value of the 3280 * device's runtime suspend function and mark last busy for the device so 3281 * that PM core will try to auto suspend the device at a later time. 3282 * 3283 * This function should be called near the end of the device's 3284 * runtime_suspend callback. 3285 */ 3286 void blk_post_runtime_suspend(struct request_queue *q, int err) 3287 { 3288 spin_lock_irq(q->queue_lock); 3289 if (!err) { 3290 q->rpm_status = RPM_SUSPENDED; 3291 } else { 3292 q->rpm_status = RPM_ACTIVE; 3293 pm_runtime_mark_last_busy(q->dev); 3294 } 3295 spin_unlock_irq(q->queue_lock); 3296 } 3297 EXPORT_SYMBOL(blk_post_runtime_suspend); 3298 3299 /** 3300 * blk_pre_runtime_resume - Pre runtime resume processing 3301 * @q: the queue of the device 3302 * 3303 * Description: 3304 * Update the queue's runtime status to RESUMING in preparation for the 3305 * runtime resume of the device. 3306 * 3307 * This function should be called near the start of the device's 3308 * runtime_resume callback. 3309 */ 3310 void blk_pre_runtime_resume(struct request_queue *q) 3311 { 3312 spin_lock_irq(q->queue_lock); 3313 q->rpm_status = RPM_RESUMING; 3314 spin_unlock_irq(q->queue_lock); 3315 } 3316 EXPORT_SYMBOL(blk_pre_runtime_resume); 3317 3318 /** 3319 * blk_post_runtime_resume - Post runtime resume processing 3320 * @q: the queue of the device 3321 * @err: return value of the device's runtime_resume function 3322 * 3323 * Description: 3324 * Update the queue's runtime status according to the return value of the 3325 * device's runtime_resume function. If it is successfully resumed, process 3326 * the requests that are queued into the device's queue when it is resuming 3327 * and then mark last busy and initiate autosuspend for it. 3328 * 3329 * This function should be called near the end of the device's 3330 * runtime_resume callback. 3331 */ 3332 void blk_post_runtime_resume(struct request_queue *q, int err) 3333 { 3334 spin_lock_irq(q->queue_lock); 3335 if (!err) { 3336 q->rpm_status = RPM_ACTIVE; 3337 __blk_run_queue(q); 3338 pm_runtime_mark_last_busy(q->dev); 3339 pm_request_autosuspend(q->dev); 3340 } else { 3341 q->rpm_status = RPM_SUSPENDED; 3342 } 3343 spin_unlock_irq(q->queue_lock); 3344 } 3345 EXPORT_SYMBOL(blk_post_runtime_resume); 3346 #endif 3347 3348 int __init blk_dev_init(void) 3349 { 3350 BUILD_BUG_ON(__REQ_NR_BITS > 8 * 3351 sizeof(((struct request *)0)->cmd_flags)); 3352 3353 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 3354 kblockd_workqueue = alloc_workqueue("kblockd", 3355 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 3356 if (!kblockd_workqueue) 3357 panic("Failed to create kblockd\n"); 3358 3359 request_cachep = kmem_cache_create("blkdev_requests", 3360 sizeof(struct request), 0, SLAB_PANIC, NULL); 3361 3362 blk_requestq_cachep = kmem_cache_create("blkdev_queue", 3363 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 3364 3365 return 0; 3366 } 3367