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