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/highmem.h> 20 #include <linux/mm.h> 21 #include <linux/kernel_stat.h> 22 #include <linux/string.h> 23 #include <linux/init.h> 24 #include <linux/completion.h> 25 #include <linux/slab.h> 26 #include <linux/swap.h> 27 #include <linux/writeback.h> 28 #include <linux/task_io_accounting_ops.h> 29 #include <linux/blktrace_api.h> 30 #include <linux/fault-inject.h> 31 #include <trace/block.h> 32 33 #include "blk.h" 34 35 DEFINE_TRACE(block_plug); 36 DEFINE_TRACE(block_unplug_io); 37 DEFINE_TRACE(block_unplug_timer); 38 DEFINE_TRACE(block_getrq); 39 DEFINE_TRACE(block_sleeprq); 40 DEFINE_TRACE(block_rq_requeue); 41 DEFINE_TRACE(block_bio_backmerge); 42 DEFINE_TRACE(block_bio_frontmerge); 43 DEFINE_TRACE(block_bio_queue); 44 DEFINE_TRACE(block_rq_complete); 45 DEFINE_TRACE(block_remap); /* Also used in drivers/md/dm.c */ 46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap); 47 48 static int __make_request(struct request_queue *q, struct bio *bio); 49 50 /* 51 * For the allocated request tables 52 */ 53 static struct kmem_cache *request_cachep; 54 55 /* 56 * For queue allocation 57 */ 58 struct kmem_cache *blk_requestq_cachep; 59 60 /* 61 * Controlling structure to kblockd 62 */ 63 static struct workqueue_struct *kblockd_workqueue; 64 65 static void drive_stat_acct(struct request *rq, int new_io) 66 { 67 struct hd_struct *part; 68 int rw = rq_data_dir(rq); 69 int cpu; 70 71 if (!blk_fs_request(rq) || !rq->rq_disk) 72 return; 73 74 cpu = part_stat_lock(); 75 part = disk_map_sector_rcu(rq->rq_disk, rq->sector); 76 77 if (!new_io) 78 part_stat_inc(cpu, part, merges[rw]); 79 else { 80 part_round_stats(cpu, part); 81 part_inc_in_flight(part); 82 } 83 84 part_stat_unlock(); 85 } 86 87 void blk_queue_congestion_threshold(struct request_queue *q) 88 { 89 int nr; 90 91 nr = q->nr_requests - (q->nr_requests / 8) + 1; 92 if (nr > q->nr_requests) 93 nr = q->nr_requests; 94 q->nr_congestion_on = nr; 95 96 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; 97 if (nr < 1) 98 nr = 1; 99 q->nr_congestion_off = nr; 100 } 101 102 /** 103 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info 104 * @bdev: device 105 * 106 * Locates the passed device's request queue and returns the address of its 107 * backing_dev_info 108 * 109 * Will return NULL if the request queue cannot be located. 110 */ 111 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) 112 { 113 struct backing_dev_info *ret = NULL; 114 struct request_queue *q = bdev_get_queue(bdev); 115 116 if (q) 117 ret = &q->backing_dev_info; 118 return ret; 119 } 120 EXPORT_SYMBOL(blk_get_backing_dev_info); 121 122 void blk_rq_init(struct request_queue *q, struct request *rq) 123 { 124 memset(rq, 0, sizeof(*rq)); 125 126 INIT_LIST_HEAD(&rq->queuelist); 127 INIT_LIST_HEAD(&rq->timeout_list); 128 rq->cpu = -1; 129 rq->q = q; 130 rq->sector = rq->hard_sector = (sector_t) -1; 131 INIT_HLIST_NODE(&rq->hash); 132 RB_CLEAR_NODE(&rq->rb_node); 133 rq->cmd = rq->__cmd; 134 rq->tag = -1; 135 rq->ref_count = 1; 136 } 137 EXPORT_SYMBOL(blk_rq_init); 138 139 static void req_bio_endio(struct request *rq, struct bio *bio, 140 unsigned int nbytes, int error) 141 { 142 struct request_queue *q = rq->q; 143 144 if (&q->bar_rq != rq) { 145 if (error) 146 clear_bit(BIO_UPTODATE, &bio->bi_flags); 147 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) 148 error = -EIO; 149 150 if (unlikely(nbytes > bio->bi_size)) { 151 printk(KERN_ERR "%s: want %u bytes done, %u left\n", 152 __func__, nbytes, bio->bi_size); 153 nbytes = bio->bi_size; 154 } 155 156 if (unlikely(rq->cmd_flags & REQ_QUIET)) 157 set_bit(BIO_QUIET, &bio->bi_flags); 158 159 bio->bi_size -= nbytes; 160 bio->bi_sector += (nbytes >> 9); 161 162 if (bio_integrity(bio)) 163 bio_integrity_advance(bio, nbytes); 164 165 if (bio->bi_size == 0) 166 bio_endio(bio, error); 167 } else { 168 169 /* 170 * Okay, this is the barrier request in progress, just 171 * record the error; 172 */ 173 if (error && !q->orderr) 174 q->orderr = error; 175 } 176 } 177 178 void blk_dump_rq_flags(struct request *rq, char *msg) 179 { 180 int bit; 181 182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, 183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, 184 rq->cmd_flags); 185 186 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", 187 (unsigned long long)rq->sector, 188 rq->nr_sectors, 189 rq->current_nr_sectors); 190 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", 191 rq->bio, rq->biotail, 192 rq->buffer, rq->data, 193 rq->data_len); 194 195 if (blk_pc_request(rq)) { 196 printk(KERN_INFO " cdb: "); 197 for (bit = 0; bit < BLK_MAX_CDB; bit++) 198 printk("%02x ", rq->cmd[bit]); 199 printk("\n"); 200 } 201 } 202 EXPORT_SYMBOL(blk_dump_rq_flags); 203 204 /* 205 * "plug" the device if there are no outstanding requests: this will 206 * force the transfer to start only after we have put all the requests 207 * on the list. 208 * 209 * This is called with interrupts off and no requests on the queue and 210 * with the queue lock held. 211 */ 212 void blk_plug_device(struct request_queue *q) 213 { 214 WARN_ON(!irqs_disabled()); 215 216 /* 217 * don't plug a stopped queue, it must be paired with blk_start_queue() 218 * which will restart the queueing 219 */ 220 if (blk_queue_stopped(q)) 221 return; 222 223 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { 224 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); 225 trace_block_plug(q); 226 } 227 } 228 EXPORT_SYMBOL(blk_plug_device); 229 230 /** 231 * blk_plug_device_unlocked - plug a device without queue lock held 232 * @q: The &struct request_queue to plug 233 * 234 * Description: 235 * Like @blk_plug_device(), but grabs the queue lock and disables 236 * interrupts. 237 **/ 238 void blk_plug_device_unlocked(struct request_queue *q) 239 { 240 unsigned long flags; 241 242 spin_lock_irqsave(q->queue_lock, flags); 243 blk_plug_device(q); 244 spin_unlock_irqrestore(q->queue_lock, flags); 245 } 246 EXPORT_SYMBOL(blk_plug_device_unlocked); 247 248 /* 249 * remove the queue from the plugged list, if present. called with 250 * queue lock held and interrupts disabled. 251 */ 252 int blk_remove_plug(struct request_queue *q) 253 { 254 WARN_ON(!irqs_disabled()); 255 256 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) 257 return 0; 258 259 del_timer(&q->unplug_timer); 260 return 1; 261 } 262 EXPORT_SYMBOL(blk_remove_plug); 263 264 /* 265 * remove the plug and let it rip.. 266 */ 267 void __generic_unplug_device(struct request_queue *q) 268 { 269 if (unlikely(blk_queue_stopped(q))) 270 return; 271 if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) 272 return; 273 274 q->request_fn(q); 275 } 276 277 /** 278 * generic_unplug_device - fire a request queue 279 * @q: The &struct request_queue in question 280 * 281 * Description: 282 * Linux uses plugging to build bigger requests queues before letting 283 * the device have at them. If a queue is plugged, the I/O scheduler 284 * is still adding and merging requests on the queue. Once the queue 285 * gets unplugged, the request_fn defined for the queue is invoked and 286 * transfers started. 287 **/ 288 void generic_unplug_device(struct request_queue *q) 289 { 290 if (blk_queue_plugged(q)) { 291 spin_lock_irq(q->queue_lock); 292 __generic_unplug_device(q); 293 spin_unlock_irq(q->queue_lock); 294 } 295 } 296 EXPORT_SYMBOL(generic_unplug_device); 297 298 static void blk_backing_dev_unplug(struct backing_dev_info *bdi, 299 struct page *page) 300 { 301 struct request_queue *q = bdi->unplug_io_data; 302 303 blk_unplug(q); 304 } 305 306 void blk_unplug_work(struct work_struct *work) 307 { 308 struct request_queue *q = 309 container_of(work, struct request_queue, unplug_work); 310 311 trace_block_unplug_io(q); 312 q->unplug_fn(q); 313 } 314 315 void blk_unplug_timeout(unsigned long data) 316 { 317 struct request_queue *q = (struct request_queue *)data; 318 319 trace_block_unplug_timer(q); 320 kblockd_schedule_work(q, &q->unplug_work); 321 } 322 323 void blk_unplug(struct request_queue *q) 324 { 325 /* 326 * devices don't necessarily have an ->unplug_fn defined 327 */ 328 if (q->unplug_fn) { 329 trace_block_unplug_io(q); 330 q->unplug_fn(q); 331 } 332 } 333 EXPORT_SYMBOL(blk_unplug); 334 335 static void blk_invoke_request_fn(struct request_queue *q) 336 { 337 if (unlikely(blk_queue_stopped(q))) 338 return; 339 340 /* 341 * one level of recursion is ok and is much faster than kicking 342 * the unplug handling 343 */ 344 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { 345 q->request_fn(q); 346 queue_flag_clear(QUEUE_FLAG_REENTER, q); 347 } else { 348 queue_flag_set(QUEUE_FLAG_PLUGGED, q); 349 kblockd_schedule_work(q, &q->unplug_work); 350 } 351 } 352 353 /** 354 * blk_start_queue - restart a previously stopped queue 355 * @q: The &struct request_queue in question 356 * 357 * Description: 358 * blk_start_queue() will clear the stop flag on the queue, and call 359 * the request_fn for the queue if it was in a stopped state when 360 * entered. Also see blk_stop_queue(). Queue lock must be held. 361 **/ 362 void blk_start_queue(struct request_queue *q) 363 { 364 WARN_ON(!irqs_disabled()); 365 366 queue_flag_clear(QUEUE_FLAG_STOPPED, q); 367 blk_invoke_request_fn(q); 368 } 369 EXPORT_SYMBOL(blk_start_queue); 370 371 /** 372 * blk_stop_queue - stop a queue 373 * @q: The &struct request_queue in question 374 * 375 * Description: 376 * The Linux block layer assumes that a block driver will consume all 377 * entries on the request queue when the request_fn strategy is called. 378 * Often this will not happen, because of hardware limitations (queue 379 * depth settings). If a device driver gets a 'queue full' response, 380 * or if it simply chooses not to queue more I/O at one point, it can 381 * call this function to prevent the request_fn from being called until 382 * the driver has signalled it's ready to go again. This happens by calling 383 * blk_start_queue() to restart queue operations. Queue lock must be held. 384 **/ 385 void blk_stop_queue(struct request_queue *q) 386 { 387 blk_remove_plug(q); 388 queue_flag_set(QUEUE_FLAG_STOPPED, q); 389 } 390 EXPORT_SYMBOL(blk_stop_queue); 391 392 /** 393 * blk_sync_queue - cancel any pending callbacks on a queue 394 * @q: the queue 395 * 396 * Description: 397 * The block layer may perform asynchronous callback activity 398 * on a queue, such as calling the unplug function after a timeout. 399 * A block device may call blk_sync_queue to ensure that any 400 * such activity is cancelled, thus allowing it to release resources 401 * that the callbacks might use. The caller must already have made sure 402 * that its ->make_request_fn will not re-add plugging prior to calling 403 * this function. 404 * 405 */ 406 void blk_sync_queue(struct request_queue *q) 407 { 408 del_timer_sync(&q->unplug_timer); 409 del_timer_sync(&q->timeout); 410 cancel_work_sync(&q->unplug_work); 411 } 412 EXPORT_SYMBOL(blk_sync_queue); 413 414 /** 415 * __blk_run_queue - run a single device queue 416 * @q: The queue to run 417 * 418 * Description: 419 * See @blk_run_queue. This variant must be called with the queue lock 420 * held and interrupts disabled. 421 * 422 */ 423 void __blk_run_queue(struct request_queue *q) 424 { 425 blk_remove_plug(q); 426 427 /* 428 * Only recurse once to avoid overrunning the stack, let the unplug 429 * handling reinvoke the handler shortly if we already got there. 430 */ 431 if (!elv_queue_empty(q)) 432 blk_invoke_request_fn(q); 433 } 434 EXPORT_SYMBOL(__blk_run_queue); 435 436 /** 437 * blk_run_queue - run a single device queue 438 * @q: The queue to run 439 * 440 * Description: 441 * Invoke request handling on this queue, if it has pending work to do. 442 * May be used to restart queueing when a request has completed. Also 443 * See @blk_start_queueing. 444 * 445 */ 446 void blk_run_queue(struct request_queue *q) 447 { 448 unsigned long flags; 449 450 spin_lock_irqsave(q->queue_lock, flags); 451 __blk_run_queue(q); 452 spin_unlock_irqrestore(q->queue_lock, flags); 453 } 454 EXPORT_SYMBOL(blk_run_queue); 455 456 void blk_put_queue(struct request_queue *q) 457 { 458 kobject_put(&q->kobj); 459 } 460 461 void blk_cleanup_queue(struct request_queue *q) 462 { 463 /* 464 * We know we have process context here, so we can be a little 465 * cautious and ensure that pending block actions on this device 466 * are done before moving on. Going into this function, we should 467 * not have processes doing IO to this device. 468 */ 469 blk_sync_queue(q); 470 471 mutex_lock(&q->sysfs_lock); 472 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); 473 mutex_unlock(&q->sysfs_lock); 474 475 if (q->elevator) 476 elevator_exit(q->elevator); 477 478 blk_put_queue(q); 479 } 480 EXPORT_SYMBOL(blk_cleanup_queue); 481 482 static int blk_init_free_list(struct request_queue *q) 483 { 484 struct request_list *rl = &q->rq; 485 486 rl->count[READ] = rl->count[WRITE] = 0; 487 rl->starved[READ] = rl->starved[WRITE] = 0; 488 rl->elvpriv = 0; 489 init_waitqueue_head(&rl->wait[READ]); 490 init_waitqueue_head(&rl->wait[WRITE]); 491 492 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, 493 mempool_free_slab, request_cachep, q->node); 494 495 if (!rl->rq_pool) 496 return -ENOMEM; 497 498 return 0; 499 } 500 501 struct request_queue *blk_alloc_queue(gfp_t gfp_mask) 502 { 503 return blk_alloc_queue_node(gfp_mask, -1); 504 } 505 EXPORT_SYMBOL(blk_alloc_queue); 506 507 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) 508 { 509 struct request_queue *q; 510 int err; 511 512 q = kmem_cache_alloc_node(blk_requestq_cachep, 513 gfp_mask | __GFP_ZERO, node_id); 514 if (!q) 515 return NULL; 516 517 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; 518 q->backing_dev_info.unplug_io_data = q; 519 err = bdi_init(&q->backing_dev_info); 520 if (err) { 521 kmem_cache_free(blk_requestq_cachep, q); 522 return NULL; 523 } 524 525 init_timer(&q->unplug_timer); 526 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); 527 INIT_LIST_HEAD(&q->timeout_list); 528 INIT_WORK(&q->unplug_work, blk_unplug_work); 529 530 kobject_init(&q->kobj, &blk_queue_ktype); 531 532 mutex_init(&q->sysfs_lock); 533 spin_lock_init(&q->__queue_lock); 534 535 return q; 536 } 537 EXPORT_SYMBOL(blk_alloc_queue_node); 538 539 /** 540 * blk_init_queue - prepare a request queue for use with a block device 541 * @rfn: The function to be called to process requests that have been 542 * placed on the queue. 543 * @lock: Request queue spin lock 544 * 545 * Description: 546 * If a block device wishes to use the standard request handling procedures, 547 * which sorts requests and coalesces adjacent requests, then it must 548 * call blk_init_queue(). The function @rfn will be called when there 549 * are requests on the queue that need to be processed. If the device 550 * supports plugging, then @rfn may not be called immediately when requests 551 * are available on the queue, but may be called at some time later instead. 552 * Plugged queues are generally unplugged when a buffer belonging to one 553 * of the requests on the queue is needed, or due to memory pressure. 554 * 555 * @rfn is not required, or even expected, to remove all requests off the 556 * queue, but only as many as it can handle at a time. If it does leave 557 * requests on the queue, it is responsible for arranging that the requests 558 * get dealt with eventually. 559 * 560 * The queue spin lock must be held while manipulating the requests on the 561 * request queue; this lock will be taken also from interrupt context, so irq 562 * disabling is needed for it. 563 * 564 * Function returns a pointer to the initialized request queue, or %NULL if 565 * it didn't succeed. 566 * 567 * Note: 568 * blk_init_queue() must be paired with a blk_cleanup_queue() call 569 * when the block device is deactivated (such as at module unload). 570 **/ 571 572 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) 573 { 574 return blk_init_queue_node(rfn, lock, -1); 575 } 576 EXPORT_SYMBOL(blk_init_queue); 577 578 struct request_queue * 579 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) 580 { 581 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); 582 583 if (!q) 584 return NULL; 585 586 q->node = node_id; 587 if (blk_init_free_list(q)) { 588 kmem_cache_free(blk_requestq_cachep, q); 589 return NULL; 590 } 591 592 /* 593 * if caller didn't supply a lock, they get per-queue locking with 594 * our embedded lock 595 */ 596 if (!lock) 597 lock = &q->__queue_lock; 598 599 q->request_fn = rfn; 600 q->prep_rq_fn = NULL; 601 q->unplug_fn = generic_unplug_device; 602 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER | 603 1 << QUEUE_FLAG_STACKABLE); 604 q->queue_lock = lock; 605 606 blk_queue_segment_boundary(q, BLK_SEG_BOUNDARY_MASK); 607 608 blk_queue_make_request(q, __make_request); 609 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); 610 611 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); 612 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); 613 614 q->sg_reserved_size = INT_MAX; 615 616 blk_set_cmd_filter_defaults(&q->cmd_filter); 617 618 /* 619 * all done 620 */ 621 if (!elevator_init(q, NULL)) { 622 blk_queue_congestion_threshold(q); 623 return q; 624 } 625 626 blk_put_queue(q); 627 return NULL; 628 } 629 EXPORT_SYMBOL(blk_init_queue_node); 630 631 int blk_get_queue(struct request_queue *q) 632 { 633 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { 634 kobject_get(&q->kobj); 635 return 0; 636 } 637 638 return 1; 639 } 640 641 static inline void blk_free_request(struct request_queue *q, struct request *rq) 642 { 643 if (rq->cmd_flags & REQ_ELVPRIV) 644 elv_put_request(q, rq); 645 mempool_free(rq, q->rq.rq_pool); 646 } 647 648 static struct request * 649 blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) 650 { 651 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); 652 653 if (!rq) 654 return NULL; 655 656 blk_rq_init(q, rq); 657 658 rq->cmd_flags = rw | REQ_ALLOCED; 659 660 if (priv) { 661 if (unlikely(elv_set_request(q, rq, gfp_mask))) { 662 mempool_free(rq, q->rq.rq_pool); 663 return NULL; 664 } 665 rq->cmd_flags |= REQ_ELVPRIV; 666 } 667 668 return rq; 669 } 670 671 /* 672 * ioc_batching returns true if the ioc is a valid batching request and 673 * should be given priority access to a request. 674 */ 675 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) 676 { 677 if (!ioc) 678 return 0; 679 680 /* 681 * Make sure the process is able to allocate at least 1 request 682 * even if the batch times out, otherwise we could theoretically 683 * lose wakeups. 684 */ 685 return ioc->nr_batch_requests == q->nr_batching || 686 (ioc->nr_batch_requests > 0 687 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); 688 } 689 690 /* 691 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This 692 * will cause the process to be a "batcher" on all queues in the system. This 693 * is the behaviour we want though - once it gets a wakeup it should be given 694 * a nice run. 695 */ 696 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) 697 { 698 if (!ioc || ioc_batching(q, ioc)) 699 return; 700 701 ioc->nr_batch_requests = q->nr_batching; 702 ioc->last_waited = jiffies; 703 } 704 705 static void __freed_request(struct request_queue *q, int rw) 706 { 707 struct request_list *rl = &q->rq; 708 709 if (rl->count[rw] < queue_congestion_off_threshold(q)) 710 blk_clear_queue_congested(q, rw); 711 712 if (rl->count[rw] + 1 <= q->nr_requests) { 713 if (waitqueue_active(&rl->wait[rw])) 714 wake_up(&rl->wait[rw]); 715 716 blk_clear_queue_full(q, rw); 717 } 718 } 719 720 /* 721 * A request has just been released. Account for it, update the full and 722 * congestion status, wake up any waiters. Called under q->queue_lock. 723 */ 724 static void freed_request(struct request_queue *q, int rw, int priv) 725 { 726 struct request_list *rl = &q->rq; 727 728 rl->count[rw]--; 729 if (priv) 730 rl->elvpriv--; 731 732 __freed_request(q, rw); 733 734 if (unlikely(rl->starved[rw ^ 1])) 735 __freed_request(q, rw ^ 1); 736 } 737 738 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) 739 /* 740 * Get a free request, queue_lock must be held. 741 * Returns NULL on failure, with queue_lock held. 742 * Returns !NULL on success, with queue_lock *not held*. 743 */ 744 static struct request *get_request(struct request_queue *q, int rw_flags, 745 struct bio *bio, gfp_t gfp_mask) 746 { 747 struct request *rq = NULL; 748 struct request_list *rl = &q->rq; 749 struct io_context *ioc = NULL; 750 const int rw = rw_flags & 0x01; 751 int may_queue, priv; 752 753 may_queue = elv_may_queue(q, rw_flags); 754 if (may_queue == ELV_MQUEUE_NO) 755 goto rq_starved; 756 757 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { 758 if (rl->count[rw]+1 >= q->nr_requests) { 759 ioc = current_io_context(GFP_ATOMIC, q->node); 760 /* 761 * The queue will fill after this allocation, so set 762 * it as full, and mark this process as "batching". 763 * This process will be allowed to complete a batch of 764 * requests, others will be blocked. 765 */ 766 if (!blk_queue_full(q, rw)) { 767 ioc_set_batching(q, ioc); 768 blk_set_queue_full(q, rw); 769 } else { 770 if (may_queue != ELV_MQUEUE_MUST 771 && !ioc_batching(q, ioc)) { 772 /* 773 * The queue is full and the allocating 774 * process is not a "batcher", and not 775 * exempted by the IO scheduler 776 */ 777 goto out; 778 } 779 } 780 } 781 blk_set_queue_congested(q, rw); 782 } 783 784 /* 785 * Only allow batching queuers to allocate up to 50% over the defined 786 * limit of requests, otherwise we could have thousands of requests 787 * allocated with any setting of ->nr_requests 788 */ 789 if (rl->count[rw] >= (3 * q->nr_requests / 2)) 790 goto out; 791 792 rl->count[rw]++; 793 rl->starved[rw] = 0; 794 795 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); 796 if (priv) 797 rl->elvpriv++; 798 799 spin_unlock_irq(q->queue_lock); 800 801 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); 802 if (unlikely(!rq)) { 803 /* 804 * Allocation failed presumably due to memory. Undo anything 805 * we might have messed up. 806 * 807 * Allocating task should really be put onto the front of the 808 * wait queue, but this is pretty rare. 809 */ 810 spin_lock_irq(q->queue_lock); 811 freed_request(q, rw, priv); 812 813 /* 814 * in the very unlikely event that allocation failed and no 815 * requests for this direction was pending, mark us starved 816 * so that freeing of a request in the other direction will 817 * notice us. another possible fix would be to split the 818 * rq mempool into READ and WRITE 819 */ 820 rq_starved: 821 if (unlikely(rl->count[rw] == 0)) 822 rl->starved[rw] = 1; 823 824 goto out; 825 } 826 827 /* 828 * ioc may be NULL here, and ioc_batching will be false. That's 829 * OK, if the queue is under the request limit then requests need 830 * not count toward the nr_batch_requests limit. There will always 831 * be some limit enforced by BLK_BATCH_TIME. 832 */ 833 if (ioc_batching(q, ioc)) 834 ioc->nr_batch_requests--; 835 836 trace_block_getrq(q, bio, rw); 837 out: 838 return rq; 839 } 840 841 /* 842 * No available requests for this queue, unplug the device and wait for some 843 * requests to become available. 844 * 845 * Called with q->queue_lock held, and returns with it unlocked. 846 */ 847 static struct request *get_request_wait(struct request_queue *q, int rw_flags, 848 struct bio *bio) 849 { 850 const int rw = rw_flags & 0x01; 851 struct request *rq; 852 853 rq = get_request(q, rw_flags, bio, GFP_NOIO); 854 while (!rq) { 855 DEFINE_WAIT(wait); 856 struct io_context *ioc; 857 struct request_list *rl = &q->rq; 858 859 prepare_to_wait_exclusive(&rl->wait[rw], &wait, 860 TASK_UNINTERRUPTIBLE); 861 862 trace_block_sleeprq(q, bio, rw); 863 864 __generic_unplug_device(q); 865 spin_unlock_irq(q->queue_lock); 866 io_schedule(); 867 868 /* 869 * After sleeping, we become a "batching" process and 870 * will be able to allocate at least one request, and 871 * up to a big batch of them for a small period time. 872 * See ioc_batching, ioc_set_batching 873 */ 874 ioc = current_io_context(GFP_NOIO, q->node); 875 ioc_set_batching(q, ioc); 876 877 spin_lock_irq(q->queue_lock); 878 finish_wait(&rl->wait[rw], &wait); 879 880 rq = get_request(q, rw_flags, bio, GFP_NOIO); 881 }; 882 883 return rq; 884 } 885 886 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) 887 { 888 struct request *rq; 889 890 BUG_ON(rw != READ && rw != WRITE); 891 892 spin_lock_irq(q->queue_lock); 893 if (gfp_mask & __GFP_WAIT) { 894 rq = get_request_wait(q, rw, NULL); 895 } else { 896 rq = get_request(q, rw, NULL, gfp_mask); 897 if (!rq) 898 spin_unlock_irq(q->queue_lock); 899 } 900 /* q->queue_lock is unlocked at this point */ 901 902 return rq; 903 } 904 EXPORT_SYMBOL(blk_get_request); 905 906 /** 907 * blk_start_queueing - initiate dispatch of requests to device 908 * @q: request queue to kick into gear 909 * 910 * This is basically a helper to remove the need to know whether a queue 911 * is plugged or not if someone just wants to initiate dispatch of requests 912 * for this queue. Should be used to start queueing on a device outside 913 * of ->request_fn() context. Also see @blk_run_queue. 914 * 915 * The queue lock must be held with interrupts disabled. 916 */ 917 void blk_start_queueing(struct request_queue *q) 918 { 919 if (!blk_queue_plugged(q)) { 920 if (unlikely(blk_queue_stopped(q))) 921 return; 922 q->request_fn(q); 923 } else 924 __generic_unplug_device(q); 925 } 926 EXPORT_SYMBOL(blk_start_queueing); 927 928 /** 929 * blk_requeue_request - put a request back on queue 930 * @q: request queue where request should be inserted 931 * @rq: request to be inserted 932 * 933 * Description: 934 * Drivers often keep queueing requests until the hardware cannot accept 935 * more, when that condition happens we need to put the request back 936 * on the queue. Must be called with queue lock held. 937 */ 938 void blk_requeue_request(struct request_queue *q, struct request *rq) 939 { 940 blk_delete_timer(rq); 941 blk_clear_rq_complete(rq); 942 trace_block_rq_requeue(q, rq); 943 944 if (blk_rq_tagged(rq)) 945 blk_queue_end_tag(q, rq); 946 947 elv_requeue_request(q, rq); 948 } 949 EXPORT_SYMBOL(blk_requeue_request); 950 951 /** 952 * blk_insert_request - insert a special request into a request queue 953 * @q: request queue where request should be inserted 954 * @rq: request to be inserted 955 * @at_head: insert request at head or tail of queue 956 * @data: private data 957 * 958 * Description: 959 * Many block devices need to execute commands asynchronously, so they don't 960 * block the whole kernel from preemption during request execution. This is 961 * accomplished normally by inserting aritficial requests tagged as 962 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them 963 * be scheduled for actual execution by the request queue. 964 * 965 * We have the option of inserting the head or the tail of the queue. 966 * Typically we use the tail for new ioctls and so forth. We use the head 967 * of the queue for things like a QUEUE_FULL message from a device, or a 968 * host that is unable to accept a particular command. 969 */ 970 void blk_insert_request(struct request_queue *q, struct request *rq, 971 int at_head, void *data) 972 { 973 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; 974 unsigned long flags; 975 976 /* 977 * tell I/O scheduler that this isn't a regular read/write (ie it 978 * must not attempt merges on this) and that it acts as a soft 979 * barrier 980 */ 981 rq->cmd_type = REQ_TYPE_SPECIAL; 982 rq->cmd_flags |= REQ_SOFTBARRIER; 983 984 rq->special = data; 985 986 spin_lock_irqsave(q->queue_lock, flags); 987 988 /* 989 * If command is tagged, release the tag 990 */ 991 if (blk_rq_tagged(rq)) 992 blk_queue_end_tag(q, rq); 993 994 drive_stat_acct(rq, 1); 995 __elv_add_request(q, rq, where, 0); 996 blk_start_queueing(q); 997 spin_unlock_irqrestore(q->queue_lock, flags); 998 } 999 EXPORT_SYMBOL(blk_insert_request); 1000 1001 /* 1002 * add-request adds a request to the linked list. 1003 * queue lock is held and interrupts disabled, as we muck with the 1004 * request queue list. 1005 */ 1006 static inline void add_request(struct request_queue *q, struct request *req) 1007 { 1008 drive_stat_acct(req, 1); 1009 1010 /* 1011 * elevator indicated where it wants this request to be 1012 * inserted at elevator_merge time 1013 */ 1014 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); 1015 } 1016 1017 static void part_round_stats_single(int cpu, struct hd_struct *part, 1018 unsigned long now) 1019 { 1020 if (now == part->stamp) 1021 return; 1022 1023 if (part->in_flight) { 1024 __part_stat_add(cpu, part, time_in_queue, 1025 part->in_flight * (now - part->stamp)); 1026 __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); 1027 } 1028 part->stamp = now; 1029 } 1030 1031 /** 1032 * part_round_stats() - Round off the performance stats on a struct disk_stats. 1033 * @cpu: cpu number for stats access 1034 * @part: target partition 1035 * 1036 * The average IO queue length and utilisation statistics are maintained 1037 * by observing the current state of the queue length and the amount of 1038 * time it has been in this state for. 1039 * 1040 * Normally, that accounting is done on IO completion, but that can result 1041 * in more than a second's worth of IO being accounted for within any one 1042 * second, leading to >100% utilisation. To deal with that, we call this 1043 * function to do a round-off before returning the results when reading 1044 * /proc/diskstats. This accounts immediately for all queue usage up to 1045 * the current jiffies and restarts the counters again. 1046 */ 1047 void part_round_stats(int cpu, struct hd_struct *part) 1048 { 1049 unsigned long now = jiffies; 1050 1051 if (part->partno) 1052 part_round_stats_single(cpu, &part_to_disk(part)->part0, now); 1053 part_round_stats_single(cpu, part, now); 1054 } 1055 EXPORT_SYMBOL_GPL(part_round_stats); 1056 1057 /* 1058 * queue lock must be held 1059 */ 1060 void __blk_put_request(struct request_queue *q, struct request *req) 1061 { 1062 if (unlikely(!q)) 1063 return; 1064 if (unlikely(--req->ref_count)) 1065 return; 1066 1067 elv_completed_request(q, req); 1068 1069 /* 1070 * Request may not have originated from ll_rw_blk. if not, 1071 * it didn't come out of our reserved rq pools 1072 */ 1073 if (req->cmd_flags & REQ_ALLOCED) { 1074 int rw = rq_data_dir(req); 1075 int priv = req->cmd_flags & REQ_ELVPRIV; 1076 1077 BUG_ON(!list_empty(&req->queuelist)); 1078 BUG_ON(!hlist_unhashed(&req->hash)); 1079 1080 blk_free_request(q, req); 1081 freed_request(q, rw, priv); 1082 } 1083 } 1084 EXPORT_SYMBOL_GPL(__blk_put_request); 1085 1086 void blk_put_request(struct request *req) 1087 { 1088 unsigned long flags; 1089 struct request_queue *q = req->q; 1090 1091 spin_lock_irqsave(q->queue_lock, flags); 1092 __blk_put_request(q, req); 1093 spin_unlock_irqrestore(q->queue_lock, flags); 1094 } 1095 EXPORT_SYMBOL(blk_put_request); 1096 1097 void init_request_from_bio(struct request *req, struct bio *bio) 1098 { 1099 req->cpu = bio->bi_comp_cpu; 1100 req->cmd_type = REQ_TYPE_FS; 1101 1102 /* 1103 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) 1104 */ 1105 if (bio_rw_ahead(bio)) 1106 req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | 1107 REQ_FAILFAST_DRIVER); 1108 if (bio_failfast_dev(bio)) 1109 req->cmd_flags |= REQ_FAILFAST_DEV; 1110 if (bio_failfast_transport(bio)) 1111 req->cmd_flags |= REQ_FAILFAST_TRANSPORT; 1112 if (bio_failfast_driver(bio)) 1113 req->cmd_flags |= REQ_FAILFAST_DRIVER; 1114 1115 /* 1116 * REQ_BARRIER implies no merging, but lets make it explicit 1117 */ 1118 if (unlikely(bio_discard(bio))) { 1119 req->cmd_flags |= REQ_DISCARD; 1120 if (bio_barrier(bio)) 1121 req->cmd_flags |= REQ_SOFTBARRIER; 1122 req->q->prepare_discard_fn(req->q, req); 1123 } else if (unlikely(bio_barrier(bio))) 1124 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); 1125 1126 if (bio_sync(bio)) 1127 req->cmd_flags |= REQ_RW_SYNC; 1128 if (bio_rw_meta(bio)) 1129 req->cmd_flags |= REQ_RW_META; 1130 1131 req->errors = 0; 1132 req->hard_sector = req->sector = bio->bi_sector; 1133 req->ioprio = bio_prio(bio); 1134 req->start_time = jiffies; 1135 blk_rq_bio_prep(req->q, req, bio); 1136 } 1137 1138 static int __make_request(struct request_queue *q, struct bio *bio) 1139 { 1140 struct request *req; 1141 int el_ret, nr_sectors; 1142 const unsigned short prio = bio_prio(bio); 1143 const int sync = bio_sync(bio); 1144 int rw_flags; 1145 1146 nr_sectors = bio_sectors(bio); 1147 1148 /* 1149 * low level driver can indicate that it wants pages above a 1150 * certain limit bounced to low memory (ie for highmem, or even 1151 * ISA dma in theory) 1152 */ 1153 blk_queue_bounce(q, &bio); 1154 1155 spin_lock_irq(q->queue_lock); 1156 1157 if (unlikely(bio_barrier(bio)) || elv_queue_empty(q)) 1158 goto get_rq; 1159 1160 el_ret = elv_merge(q, &req, bio); 1161 switch (el_ret) { 1162 case ELEVATOR_BACK_MERGE: 1163 BUG_ON(!rq_mergeable(req)); 1164 1165 if (!ll_back_merge_fn(q, req, bio)) 1166 break; 1167 1168 trace_block_bio_backmerge(q, bio); 1169 1170 req->biotail->bi_next = bio; 1171 req->biotail = bio; 1172 req->nr_sectors = req->hard_nr_sectors += nr_sectors; 1173 req->ioprio = ioprio_best(req->ioprio, prio); 1174 if (!blk_rq_cpu_valid(req)) 1175 req->cpu = bio->bi_comp_cpu; 1176 drive_stat_acct(req, 0); 1177 if (!attempt_back_merge(q, req)) 1178 elv_merged_request(q, req, el_ret); 1179 goto out; 1180 1181 case ELEVATOR_FRONT_MERGE: 1182 BUG_ON(!rq_mergeable(req)); 1183 1184 if (!ll_front_merge_fn(q, req, bio)) 1185 break; 1186 1187 trace_block_bio_frontmerge(q, bio); 1188 1189 bio->bi_next = req->bio; 1190 req->bio = bio; 1191 1192 /* 1193 * may not be valid. if the low level driver said 1194 * it didn't need a bounce buffer then it better 1195 * not touch req->buffer either... 1196 */ 1197 req->buffer = bio_data(bio); 1198 req->current_nr_sectors = bio_cur_sectors(bio); 1199 req->hard_cur_sectors = req->current_nr_sectors; 1200 req->sector = req->hard_sector = bio->bi_sector; 1201 req->nr_sectors = req->hard_nr_sectors += nr_sectors; 1202 req->ioprio = ioprio_best(req->ioprio, prio); 1203 if (!blk_rq_cpu_valid(req)) 1204 req->cpu = bio->bi_comp_cpu; 1205 drive_stat_acct(req, 0); 1206 if (!attempt_front_merge(q, req)) 1207 elv_merged_request(q, req, el_ret); 1208 goto out; 1209 1210 /* ELV_NO_MERGE: elevator says don't/can't merge. */ 1211 default: 1212 ; 1213 } 1214 1215 get_rq: 1216 /* 1217 * This sync check and mask will be re-done in init_request_from_bio(), 1218 * but we need to set it earlier to expose the sync flag to the 1219 * rq allocator and io schedulers. 1220 */ 1221 rw_flags = bio_data_dir(bio); 1222 if (sync) 1223 rw_flags |= REQ_RW_SYNC; 1224 1225 /* 1226 * Grab a free request. This is might sleep but can not fail. 1227 * Returns with the queue unlocked. 1228 */ 1229 req = get_request_wait(q, rw_flags, bio); 1230 1231 /* 1232 * After dropping the lock and possibly sleeping here, our request 1233 * may now be mergeable after it had proven unmergeable (above). 1234 * We don't worry about that case for efficiency. It won't happen 1235 * often, and the elevators are able to handle it. 1236 */ 1237 init_request_from_bio(req, bio); 1238 1239 spin_lock_irq(q->queue_lock); 1240 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || 1241 bio_flagged(bio, BIO_CPU_AFFINE)) 1242 req->cpu = blk_cpu_to_group(smp_processor_id()); 1243 if (!blk_queue_nonrot(q) && elv_queue_empty(q)) 1244 blk_plug_device(q); 1245 add_request(q, req); 1246 out: 1247 if (sync || blk_queue_nonrot(q)) 1248 __generic_unplug_device(q); 1249 spin_unlock_irq(q->queue_lock); 1250 return 0; 1251 } 1252 1253 /* 1254 * If bio->bi_dev is a partition, remap the location 1255 */ 1256 static inline void blk_partition_remap(struct bio *bio) 1257 { 1258 struct block_device *bdev = bio->bi_bdev; 1259 1260 if (bio_sectors(bio) && bdev != bdev->bd_contains) { 1261 struct hd_struct *p = bdev->bd_part; 1262 1263 bio->bi_sector += p->start_sect; 1264 bio->bi_bdev = bdev->bd_contains; 1265 1266 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, 1267 bdev->bd_dev, bio->bi_sector, 1268 bio->bi_sector - p->start_sect); 1269 } 1270 } 1271 1272 static void handle_bad_sector(struct bio *bio) 1273 { 1274 char b[BDEVNAME_SIZE]; 1275 1276 printk(KERN_INFO "attempt to access beyond end of device\n"); 1277 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", 1278 bdevname(bio->bi_bdev, b), 1279 bio->bi_rw, 1280 (unsigned long long)bio->bi_sector + bio_sectors(bio), 1281 (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); 1282 1283 set_bit(BIO_EOF, &bio->bi_flags); 1284 } 1285 1286 #ifdef CONFIG_FAIL_MAKE_REQUEST 1287 1288 static DECLARE_FAULT_ATTR(fail_make_request); 1289 1290 static int __init setup_fail_make_request(char *str) 1291 { 1292 return setup_fault_attr(&fail_make_request, str); 1293 } 1294 __setup("fail_make_request=", setup_fail_make_request); 1295 1296 static int should_fail_request(struct bio *bio) 1297 { 1298 struct hd_struct *part = bio->bi_bdev->bd_part; 1299 1300 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) 1301 return should_fail(&fail_make_request, bio->bi_size); 1302 1303 return 0; 1304 } 1305 1306 static int __init fail_make_request_debugfs(void) 1307 { 1308 return init_fault_attr_dentries(&fail_make_request, 1309 "fail_make_request"); 1310 } 1311 1312 late_initcall(fail_make_request_debugfs); 1313 1314 #else /* CONFIG_FAIL_MAKE_REQUEST */ 1315 1316 static inline int should_fail_request(struct bio *bio) 1317 { 1318 return 0; 1319 } 1320 1321 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 1322 1323 /* 1324 * Check whether this bio extends beyond the end of the device. 1325 */ 1326 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) 1327 { 1328 sector_t maxsector; 1329 1330 if (!nr_sectors) 1331 return 0; 1332 1333 /* Test device or partition size, when known. */ 1334 maxsector = bio->bi_bdev->bd_inode->i_size >> 9; 1335 if (maxsector) { 1336 sector_t sector = bio->bi_sector; 1337 1338 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { 1339 /* 1340 * This may well happen - the kernel calls bread() 1341 * without checking the size of the device, e.g., when 1342 * mounting a device. 1343 */ 1344 handle_bad_sector(bio); 1345 return 1; 1346 } 1347 } 1348 1349 return 0; 1350 } 1351 1352 /** 1353 * generic_make_request - hand a buffer to its device driver for I/O 1354 * @bio: The bio describing the location in memory and on the device. 1355 * 1356 * generic_make_request() is used to make I/O requests of block 1357 * devices. It is passed a &struct bio, which describes the I/O that needs 1358 * to be done. 1359 * 1360 * generic_make_request() does not return any status. The 1361 * success/failure status of the request, along with notification of 1362 * completion, is delivered asynchronously through the bio->bi_end_io 1363 * function described (one day) else where. 1364 * 1365 * The caller of generic_make_request must make sure that bi_io_vec 1366 * are set to describe the memory buffer, and that bi_dev and bi_sector are 1367 * set to describe the device address, and the 1368 * bi_end_io and optionally bi_private are set to describe how 1369 * completion notification should be signaled. 1370 * 1371 * generic_make_request and the drivers it calls may use bi_next if this 1372 * bio happens to be merged with someone else, and may change bi_dev and 1373 * bi_sector for remaps as it sees fit. So the values of these fields 1374 * should NOT be depended on after the call to generic_make_request. 1375 */ 1376 static inline void __generic_make_request(struct bio *bio) 1377 { 1378 struct request_queue *q; 1379 sector_t old_sector; 1380 int ret, nr_sectors = bio_sectors(bio); 1381 dev_t old_dev; 1382 int err = -EIO; 1383 1384 might_sleep(); 1385 1386 if (bio_check_eod(bio, nr_sectors)) 1387 goto end_io; 1388 1389 /* 1390 * Resolve the mapping until finished. (drivers are 1391 * still free to implement/resolve their own stacking 1392 * by explicitly returning 0) 1393 * 1394 * NOTE: we don't repeat the blk_size check for each new device. 1395 * Stacking drivers are expected to know what they are doing. 1396 */ 1397 old_sector = -1; 1398 old_dev = 0; 1399 do { 1400 char b[BDEVNAME_SIZE]; 1401 1402 q = bdev_get_queue(bio->bi_bdev); 1403 if (unlikely(!q)) { 1404 printk(KERN_ERR 1405 "generic_make_request: Trying to access " 1406 "nonexistent block-device %s (%Lu)\n", 1407 bdevname(bio->bi_bdev, b), 1408 (long long) bio->bi_sector); 1409 goto end_io; 1410 } 1411 1412 if (unlikely(nr_sectors > q->max_hw_sectors)) { 1413 printk(KERN_ERR "bio too big device %s (%u > %u)\n", 1414 bdevname(bio->bi_bdev, b), 1415 bio_sectors(bio), 1416 q->max_hw_sectors); 1417 goto end_io; 1418 } 1419 1420 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) 1421 goto end_io; 1422 1423 if (should_fail_request(bio)) 1424 goto end_io; 1425 1426 /* 1427 * If this device has partitions, remap block n 1428 * of partition p to block n+start(p) of the disk. 1429 */ 1430 blk_partition_remap(bio); 1431 1432 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) 1433 goto end_io; 1434 1435 if (old_sector != -1) 1436 trace_block_remap(q, bio, old_dev, bio->bi_sector, 1437 old_sector); 1438 1439 trace_block_bio_queue(q, bio); 1440 1441 old_sector = bio->bi_sector; 1442 old_dev = bio->bi_bdev->bd_dev; 1443 1444 if (bio_check_eod(bio, nr_sectors)) 1445 goto end_io; 1446 1447 if (bio_discard(bio) && !q->prepare_discard_fn) { 1448 err = -EOPNOTSUPP; 1449 goto end_io; 1450 } 1451 1452 ret = q->make_request_fn(q, bio); 1453 } while (ret); 1454 1455 return; 1456 1457 end_io: 1458 bio_endio(bio, err); 1459 } 1460 1461 /* 1462 * We only want one ->make_request_fn to be active at a time, 1463 * else stack usage with stacked devices could be a problem. 1464 * So use current->bio_{list,tail} to keep a list of requests 1465 * submited by a make_request_fn function. 1466 * current->bio_tail is also used as a flag to say if 1467 * generic_make_request is currently active in this task or not. 1468 * If it is NULL, then no make_request is active. If it is non-NULL, 1469 * then a make_request is active, and new requests should be added 1470 * at the tail 1471 */ 1472 void generic_make_request(struct bio *bio) 1473 { 1474 if (current->bio_tail) { 1475 /* make_request is active */ 1476 *(current->bio_tail) = bio; 1477 bio->bi_next = NULL; 1478 current->bio_tail = &bio->bi_next; 1479 return; 1480 } 1481 /* following loop may be a bit non-obvious, and so deserves some 1482 * explanation. 1483 * Before entering the loop, bio->bi_next is NULL (as all callers 1484 * ensure that) so we have a list with a single bio. 1485 * We pretend that we have just taken it off a longer list, so 1486 * we assign bio_list to the next (which is NULL) and bio_tail 1487 * to &bio_list, thus initialising the bio_list of new bios to be 1488 * added. __generic_make_request may indeed add some more bios 1489 * through a recursive call to generic_make_request. If it 1490 * did, we find a non-NULL value in bio_list and re-enter the loop 1491 * from the top. In this case we really did just take the bio 1492 * of the top of the list (no pretending) and so fixup bio_list and 1493 * bio_tail or bi_next, and call into __generic_make_request again. 1494 * 1495 * The loop was structured like this to make only one call to 1496 * __generic_make_request (which is important as it is large and 1497 * inlined) and to keep the structure simple. 1498 */ 1499 BUG_ON(bio->bi_next); 1500 do { 1501 current->bio_list = bio->bi_next; 1502 if (bio->bi_next == NULL) 1503 current->bio_tail = ¤t->bio_list; 1504 else 1505 bio->bi_next = NULL; 1506 __generic_make_request(bio); 1507 bio = current->bio_list; 1508 } while (bio); 1509 current->bio_tail = NULL; /* deactivate */ 1510 } 1511 EXPORT_SYMBOL(generic_make_request); 1512 1513 /** 1514 * submit_bio - submit a bio to the block device layer for I/O 1515 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) 1516 * @bio: The &struct bio which describes the I/O 1517 * 1518 * submit_bio() is very similar in purpose to generic_make_request(), and 1519 * uses that function to do most of the work. Both are fairly rough 1520 * interfaces; @bio must be presetup and ready for I/O. 1521 * 1522 */ 1523 void submit_bio(int rw, struct bio *bio) 1524 { 1525 int count = bio_sectors(bio); 1526 1527 bio->bi_rw |= rw; 1528 1529 /* 1530 * If it's a regular read/write or a barrier with data attached, 1531 * go through the normal accounting stuff before submission. 1532 */ 1533 if (bio_has_data(bio)) { 1534 if (rw & WRITE) { 1535 count_vm_events(PGPGOUT, count); 1536 } else { 1537 task_io_account_read(bio->bi_size); 1538 count_vm_events(PGPGIN, count); 1539 } 1540 1541 if (unlikely(block_dump)) { 1542 char b[BDEVNAME_SIZE]; 1543 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", 1544 current->comm, task_pid_nr(current), 1545 (rw & WRITE) ? "WRITE" : "READ", 1546 (unsigned long long)bio->bi_sector, 1547 bdevname(bio->bi_bdev, b)); 1548 } 1549 } 1550 1551 generic_make_request(bio); 1552 } 1553 EXPORT_SYMBOL(submit_bio); 1554 1555 /** 1556 * blk_rq_check_limits - Helper function to check a request for the queue limit 1557 * @q: the queue 1558 * @rq: the request being checked 1559 * 1560 * Description: 1561 * @rq may have been made based on weaker limitations of upper-level queues 1562 * in request stacking drivers, and it may violate the limitation of @q. 1563 * Since the block layer and the underlying device driver trust @rq 1564 * after it is inserted to @q, it should be checked against @q before 1565 * the insertion using this generic function. 1566 * 1567 * This function should also be useful for request stacking drivers 1568 * in some cases below, so export this fuction. 1569 * Request stacking drivers like request-based dm may change the queue 1570 * limits while requests are in the queue (e.g. dm's table swapping). 1571 * Such request stacking drivers should check those requests agaist 1572 * the new queue limits again when they dispatch those requests, 1573 * although such checkings are also done against the old queue limits 1574 * when submitting requests. 1575 */ 1576 int blk_rq_check_limits(struct request_queue *q, struct request *rq) 1577 { 1578 if (rq->nr_sectors > q->max_sectors || 1579 rq->data_len > q->max_hw_sectors << 9) { 1580 printk(KERN_ERR "%s: over max size limit.\n", __func__); 1581 return -EIO; 1582 } 1583 1584 /* 1585 * queue's settings related to segment counting like q->bounce_pfn 1586 * may differ from that of other stacking queues. 1587 * Recalculate it to check the request correctly on this queue's 1588 * limitation. 1589 */ 1590 blk_recalc_rq_segments(rq); 1591 if (rq->nr_phys_segments > q->max_phys_segments || 1592 rq->nr_phys_segments > q->max_hw_segments) { 1593 printk(KERN_ERR "%s: over max segments limit.\n", __func__); 1594 return -EIO; 1595 } 1596 1597 return 0; 1598 } 1599 EXPORT_SYMBOL_GPL(blk_rq_check_limits); 1600 1601 /** 1602 * blk_insert_cloned_request - Helper for stacking drivers to submit a request 1603 * @q: the queue to submit the request 1604 * @rq: the request being queued 1605 */ 1606 int blk_insert_cloned_request(struct request_queue *q, struct request *rq) 1607 { 1608 unsigned long flags; 1609 1610 if (blk_rq_check_limits(q, rq)) 1611 return -EIO; 1612 1613 #ifdef CONFIG_FAIL_MAKE_REQUEST 1614 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && 1615 should_fail(&fail_make_request, blk_rq_bytes(rq))) 1616 return -EIO; 1617 #endif 1618 1619 spin_lock_irqsave(q->queue_lock, flags); 1620 1621 /* 1622 * Submitting request must be dequeued before calling this function 1623 * because it will be linked to another request_queue 1624 */ 1625 BUG_ON(blk_queued_rq(rq)); 1626 1627 drive_stat_acct(rq, 1); 1628 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); 1629 1630 spin_unlock_irqrestore(q->queue_lock, flags); 1631 1632 return 0; 1633 } 1634 EXPORT_SYMBOL_GPL(blk_insert_cloned_request); 1635 1636 /** 1637 * blkdev_dequeue_request - dequeue request and start timeout timer 1638 * @req: request to dequeue 1639 * 1640 * Dequeue @req and start timeout timer on it. This hands off the 1641 * request to the driver. 1642 * 1643 * Block internal functions which don't want to start timer should 1644 * call elv_dequeue_request(). 1645 */ 1646 void blkdev_dequeue_request(struct request *req) 1647 { 1648 elv_dequeue_request(req->q, req); 1649 1650 /* 1651 * We are now handing the request to the hardware, add the 1652 * timeout handler. 1653 */ 1654 blk_add_timer(req); 1655 } 1656 EXPORT_SYMBOL(blkdev_dequeue_request); 1657 1658 /** 1659 * __end_that_request_first - end I/O on a request 1660 * @req: the request being processed 1661 * @error: %0 for success, < %0 for error 1662 * @nr_bytes: number of bytes to complete 1663 * 1664 * Description: 1665 * Ends I/O on a number of bytes attached to @req, and sets it up 1666 * for the next range of segments (if any) in the cluster. 1667 * 1668 * Return: 1669 * %0 - we are done with this request, call end_that_request_last() 1670 * %1 - still buffers pending for this request 1671 **/ 1672 static int __end_that_request_first(struct request *req, int error, 1673 int nr_bytes) 1674 { 1675 int total_bytes, bio_nbytes, next_idx = 0; 1676 struct bio *bio; 1677 1678 trace_block_rq_complete(req->q, req); 1679 1680 /* 1681 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual 1682 * sense key with us all the way through 1683 */ 1684 if (!blk_pc_request(req)) 1685 req->errors = 0; 1686 1687 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { 1688 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", 1689 req->rq_disk ? req->rq_disk->disk_name : "?", 1690 (unsigned long long)req->sector); 1691 } 1692 1693 if (blk_fs_request(req) && req->rq_disk) { 1694 const int rw = rq_data_dir(req); 1695 struct hd_struct *part; 1696 int cpu; 1697 1698 cpu = part_stat_lock(); 1699 part = disk_map_sector_rcu(req->rq_disk, req->sector); 1700 part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9); 1701 part_stat_unlock(); 1702 } 1703 1704 total_bytes = bio_nbytes = 0; 1705 while ((bio = req->bio) != NULL) { 1706 int nbytes; 1707 1708 if (nr_bytes >= bio->bi_size) { 1709 req->bio = bio->bi_next; 1710 nbytes = bio->bi_size; 1711 req_bio_endio(req, bio, nbytes, error); 1712 next_idx = 0; 1713 bio_nbytes = 0; 1714 } else { 1715 int idx = bio->bi_idx + next_idx; 1716 1717 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { 1718 blk_dump_rq_flags(req, "__end_that"); 1719 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", 1720 __func__, bio->bi_idx, bio->bi_vcnt); 1721 break; 1722 } 1723 1724 nbytes = bio_iovec_idx(bio, idx)->bv_len; 1725 BIO_BUG_ON(nbytes > bio->bi_size); 1726 1727 /* 1728 * not a complete bvec done 1729 */ 1730 if (unlikely(nbytes > nr_bytes)) { 1731 bio_nbytes += nr_bytes; 1732 total_bytes += nr_bytes; 1733 break; 1734 } 1735 1736 /* 1737 * advance to the next vector 1738 */ 1739 next_idx++; 1740 bio_nbytes += nbytes; 1741 } 1742 1743 total_bytes += nbytes; 1744 nr_bytes -= nbytes; 1745 1746 bio = req->bio; 1747 if (bio) { 1748 /* 1749 * end more in this run, or just return 'not-done' 1750 */ 1751 if (unlikely(nr_bytes <= 0)) 1752 break; 1753 } 1754 } 1755 1756 /* 1757 * completely done 1758 */ 1759 if (!req->bio) 1760 return 0; 1761 1762 /* 1763 * if the request wasn't completed, update state 1764 */ 1765 if (bio_nbytes) { 1766 req_bio_endio(req, bio, bio_nbytes, error); 1767 bio->bi_idx += next_idx; 1768 bio_iovec(bio)->bv_offset += nr_bytes; 1769 bio_iovec(bio)->bv_len -= nr_bytes; 1770 } 1771 1772 blk_recalc_rq_sectors(req, total_bytes >> 9); 1773 blk_recalc_rq_segments(req); 1774 return 1; 1775 } 1776 1777 /* 1778 * queue lock must be held 1779 */ 1780 static void end_that_request_last(struct request *req, int error) 1781 { 1782 struct gendisk *disk = req->rq_disk; 1783 1784 if (blk_rq_tagged(req)) 1785 blk_queue_end_tag(req->q, req); 1786 1787 if (blk_queued_rq(req)) 1788 elv_dequeue_request(req->q, req); 1789 1790 if (unlikely(laptop_mode) && blk_fs_request(req)) 1791 laptop_io_completion(); 1792 1793 blk_delete_timer(req); 1794 1795 /* 1796 * Account IO completion. bar_rq isn't accounted as a normal 1797 * IO on queueing nor completion. Accounting the containing 1798 * request is enough. 1799 */ 1800 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { 1801 unsigned long duration = jiffies - req->start_time; 1802 const int rw = rq_data_dir(req); 1803 struct hd_struct *part; 1804 int cpu; 1805 1806 cpu = part_stat_lock(); 1807 part = disk_map_sector_rcu(disk, req->sector); 1808 1809 part_stat_inc(cpu, part, ios[rw]); 1810 part_stat_add(cpu, part, ticks[rw], duration); 1811 part_round_stats(cpu, part); 1812 part_dec_in_flight(part); 1813 1814 part_stat_unlock(); 1815 } 1816 1817 if (req->end_io) 1818 req->end_io(req, error); 1819 else { 1820 if (blk_bidi_rq(req)) 1821 __blk_put_request(req->next_rq->q, req->next_rq); 1822 1823 __blk_put_request(req->q, req); 1824 } 1825 } 1826 1827 /** 1828 * blk_rq_bytes - Returns bytes left to complete in the entire request 1829 * @rq: the request being processed 1830 **/ 1831 unsigned int blk_rq_bytes(struct request *rq) 1832 { 1833 if (blk_fs_request(rq)) 1834 return rq->hard_nr_sectors << 9; 1835 1836 return rq->data_len; 1837 } 1838 EXPORT_SYMBOL_GPL(blk_rq_bytes); 1839 1840 /** 1841 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment 1842 * @rq: the request being processed 1843 **/ 1844 unsigned int blk_rq_cur_bytes(struct request *rq) 1845 { 1846 if (blk_fs_request(rq)) 1847 return rq->current_nr_sectors << 9; 1848 1849 if (rq->bio) 1850 return rq->bio->bi_size; 1851 1852 return rq->data_len; 1853 } 1854 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); 1855 1856 /** 1857 * end_request - end I/O on the current segment of the request 1858 * @req: the request being processed 1859 * @uptodate: error value or %0/%1 uptodate flag 1860 * 1861 * Description: 1862 * Ends I/O on the current segment of a request. If that is the only 1863 * remaining segment, the request is also completed and freed. 1864 * 1865 * This is a remnant of how older block drivers handled I/O completions. 1866 * Modern drivers typically end I/O on the full request in one go, unless 1867 * they have a residual value to account for. For that case this function 1868 * isn't really useful, unless the residual just happens to be the 1869 * full current segment. In other words, don't use this function in new 1870 * code. Use blk_end_request() or __blk_end_request() to end a request. 1871 **/ 1872 void end_request(struct request *req, int uptodate) 1873 { 1874 int error = 0; 1875 1876 if (uptodate <= 0) 1877 error = uptodate ? uptodate : -EIO; 1878 1879 __blk_end_request(req, error, req->hard_cur_sectors << 9); 1880 } 1881 EXPORT_SYMBOL(end_request); 1882 1883 static int end_that_request_data(struct request *rq, int error, 1884 unsigned int nr_bytes, unsigned int bidi_bytes) 1885 { 1886 if (rq->bio) { 1887 if (__end_that_request_first(rq, error, nr_bytes)) 1888 return 1; 1889 1890 /* Bidi request must be completed as a whole */ 1891 if (blk_bidi_rq(rq) && 1892 __end_that_request_first(rq->next_rq, error, bidi_bytes)) 1893 return 1; 1894 } 1895 1896 return 0; 1897 } 1898 1899 /** 1900 * blk_end_io - Generic end_io function to complete a request. 1901 * @rq: the request being processed 1902 * @error: %0 for success, < %0 for error 1903 * @nr_bytes: number of bytes to complete @rq 1904 * @bidi_bytes: number of bytes to complete @rq->next_rq 1905 * @drv_callback: function called between completion of bios in the request 1906 * and completion of the request. 1907 * If the callback returns non %0, this helper returns without 1908 * completion of the request. 1909 * 1910 * Description: 1911 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. 1912 * If @rq has leftover, sets it up for the next range of segments. 1913 * 1914 * Return: 1915 * %0 - we are done with this request 1916 * %1 - this request is not freed yet, it still has pending buffers. 1917 **/ 1918 static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, 1919 unsigned int bidi_bytes, 1920 int (drv_callback)(struct request *)) 1921 { 1922 struct request_queue *q = rq->q; 1923 unsigned long flags = 0UL; 1924 1925 if (end_that_request_data(rq, error, nr_bytes, bidi_bytes)) 1926 return 1; 1927 1928 /* Special feature for tricky drivers */ 1929 if (drv_callback && drv_callback(rq)) 1930 return 1; 1931 1932 add_disk_randomness(rq->rq_disk); 1933 1934 spin_lock_irqsave(q->queue_lock, flags); 1935 end_that_request_last(rq, error); 1936 spin_unlock_irqrestore(q->queue_lock, flags); 1937 1938 return 0; 1939 } 1940 1941 /** 1942 * blk_end_request - Helper function for drivers to complete the request. 1943 * @rq: the request being processed 1944 * @error: %0 for success, < %0 for error 1945 * @nr_bytes: number of bytes to complete 1946 * 1947 * Description: 1948 * Ends I/O on a number of bytes attached to @rq. 1949 * If @rq has leftover, sets it up for the next range of segments. 1950 * 1951 * Return: 1952 * %0 - we are done with this request 1953 * %1 - still buffers pending for this request 1954 **/ 1955 int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 1956 { 1957 return blk_end_io(rq, error, nr_bytes, 0, NULL); 1958 } 1959 EXPORT_SYMBOL_GPL(blk_end_request); 1960 1961 /** 1962 * __blk_end_request - Helper function for drivers to complete the request. 1963 * @rq: the request being processed 1964 * @error: %0 for success, < %0 for error 1965 * @nr_bytes: number of bytes to complete 1966 * 1967 * Description: 1968 * Must be called with queue lock held unlike blk_end_request(). 1969 * 1970 * Return: 1971 * %0 - we are done with this request 1972 * %1 - still buffers pending for this request 1973 **/ 1974 int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 1975 { 1976 if (rq->bio && __end_that_request_first(rq, error, nr_bytes)) 1977 return 1; 1978 1979 add_disk_randomness(rq->rq_disk); 1980 1981 end_that_request_last(rq, error); 1982 1983 return 0; 1984 } 1985 EXPORT_SYMBOL_GPL(__blk_end_request); 1986 1987 /** 1988 * blk_end_bidi_request - Helper function for drivers to complete bidi request. 1989 * @rq: the bidi request being processed 1990 * @error: %0 for success, < %0 for error 1991 * @nr_bytes: number of bytes to complete @rq 1992 * @bidi_bytes: number of bytes to complete @rq->next_rq 1993 * 1994 * Description: 1995 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. 1996 * 1997 * Return: 1998 * %0 - we are done with this request 1999 * %1 - still buffers pending for this request 2000 **/ 2001 int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, 2002 unsigned int bidi_bytes) 2003 { 2004 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); 2005 } 2006 EXPORT_SYMBOL_GPL(blk_end_bidi_request); 2007 2008 /** 2009 * blk_update_request - Special helper function for request stacking drivers 2010 * @rq: the request being processed 2011 * @error: %0 for success, < %0 for error 2012 * @nr_bytes: number of bytes to complete @rq 2013 * 2014 * Description: 2015 * Ends I/O on a number of bytes attached to @rq, but doesn't complete 2016 * the request structure even if @rq doesn't have leftover. 2017 * If @rq has leftover, sets it up for the next range of segments. 2018 * 2019 * This special helper function is only for request stacking drivers 2020 * (e.g. request-based dm) so that they can handle partial completion. 2021 * Actual device drivers should use blk_end_request instead. 2022 */ 2023 void blk_update_request(struct request *rq, int error, unsigned int nr_bytes) 2024 { 2025 if (!end_that_request_data(rq, error, nr_bytes, 0)) { 2026 /* 2027 * These members are not updated in end_that_request_data() 2028 * when all bios are completed. 2029 * Update them so that the request stacking driver can find 2030 * how many bytes remain in the request later. 2031 */ 2032 rq->nr_sectors = rq->hard_nr_sectors = 0; 2033 rq->current_nr_sectors = rq->hard_cur_sectors = 0; 2034 } 2035 } 2036 EXPORT_SYMBOL_GPL(blk_update_request); 2037 2038 /** 2039 * blk_end_request_callback - Special helper function for tricky drivers 2040 * @rq: the request being processed 2041 * @error: %0 for success, < %0 for error 2042 * @nr_bytes: number of bytes to complete 2043 * @drv_callback: function called between completion of bios in the request 2044 * and completion of the request. 2045 * If the callback returns non %0, this helper returns without 2046 * completion of the request. 2047 * 2048 * Description: 2049 * Ends I/O on a number of bytes attached to @rq. 2050 * If @rq has leftover, sets it up for the next range of segments. 2051 * 2052 * This special helper function is used only for existing tricky drivers. 2053 * (e.g. cdrom_newpc_intr() of ide-cd) 2054 * This interface will be removed when such drivers are rewritten. 2055 * Don't use this interface in other places anymore. 2056 * 2057 * Return: 2058 * %0 - we are done with this request 2059 * %1 - this request is not freed yet. 2060 * this request still has pending buffers or 2061 * the driver doesn't want to finish this request yet. 2062 **/ 2063 int blk_end_request_callback(struct request *rq, int error, 2064 unsigned int nr_bytes, 2065 int (drv_callback)(struct request *)) 2066 { 2067 return blk_end_io(rq, error, nr_bytes, 0, drv_callback); 2068 } 2069 EXPORT_SYMBOL_GPL(blk_end_request_callback); 2070 2071 void blk_rq_bio_prep(struct request_queue *q, struct request *rq, 2072 struct bio *bio) 2073 { 2074 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and 2075 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ 2076 rq->cmd_flags |= (bio->bi_rw & 3); 2077 2078 if (bio_has_data(bio)) { 2079 rq->nr_phys_segments = bio_phys_segments(q, bio); 2080 rq->buffer = bio_data(bio); 2081 } 2082 rq->current_nr_sectors = bio_cur_sectors(bio); 2083 rq->hard_cur_sectors = rq->current_nr_sectors; 2084 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); 2085 rq->data_len = bio->bi_size; 2086 2087 rq->bio = rq->biotail = bio; 2088 2089 if (bio->bi_bdev) 2090 rq->rq_disk = bio->bi_bdev->bd_disk; 2091 } 2092 2093 /** 2094 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 2095 * @q : the queue of the device being checked 2096 * 2097 * Description: 2098 * Check if underlying low-level drivers of a device are busy. 2099 * If the drivers want to export their busy state, they must set own 2100 * exporting function using blk_queue_lld_busy() first. 2101 * 2102 * Basically, this function is used only by request stacking drivers 2103 * to stop dispatching requests to underlying devices when underlying 2104 * devices are busy. This behavior helps more I/O merging on the queue 2105 * of the request stacking driver and prevents I/O throughput regression 2106 * on burst I/O load. 2107 * 2108 * Return: 2109 * 0 - Not busy (The request stacking driver should dispatch request) 2110 * 1 - Busy (The request stacking driver should stop dispatching request) 2111 */ 2112 int blk_lld_busy(struct request_queue *q) 2113 { 2114 if (q->lld_busy_fn) 2115 return q->lld_busy_fn(q); 2116 2117 return 0; 2118 } 2119 EXPORT_SYMBOL_GPL(blk_lld_busy); 2120 2121 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) 2122 { 2123 return queue_work(kblockd_workqueue, work); 2124 } 2125 EXPORT_SYMBOL(kblockd_schedule_work); 2126 2127 int __init blk_dev_init(void) 2128 { 2129 kblockd_workqueue = create_workqueue("kblockd"); 2130 if (!kblockd_workqueue) 2131 panic("Failed to create kblockd\n"); 2132 2133 request_cachep = kmem_cache_create("blkdev_requests", 2134 sizeof(struct request), 0, SLAB_PANIC, NULL); 2135 2136 blk_requestq_cachep = kmem_cache_create("blkdev_queue", 2137 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 2138 2139 return 0; 2140 } 2141 2142