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/fault-inject.h> 30 31 #define CREATE_TRACE_POINTS 32 #include <trace/events/block.h> 33 34 #include "blk.h" 35 36 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap); 37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); 38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); 39 40 static int __make_request(struct request_queue *q, struct bio *bio); 41 42 /* 43 * For the allocated request tables 44 */ 45 static struct kmem_cache *request_cachep; 46 47 /* 48 * For queue allocation 49 */ 50 struct kmem_cache *blk_requestq_cachep; 51 52 /* 53 * Controlling structure to kblockd 54 */ 55 static struct workqueue_struct *kblockd_workqueue; 56 57 static void drive_stat_acct(struct request *rq, int new_io) 58 { 59 struct hd_struct *part; 60 int rw = rq_data_dir(rq); 61 int cpu; 62 63 if (!blk_do_io_stat(rq)) 64 return; 65 66 cpu = part_stat_lock(); 67 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); 68 69 if (!new_io) 70 part_stat_inc(cpu, part, merges[rw]); 71 else { 72 part_round_stats(cpu, part); 73 part_inc_in_flight(part, rw); 74 } 75 76 part_stat_unlock(); 77 } 78 79 void blk_queue_congestion_threshold(struct request_queue *q) 80 { 81 int nr; 82 83 nr = q->nr_requests - (q->nr_requests / 8) + 1; 84 if (nr > q->nr_requests) 85 nr = q->nr_requests; 86 q->nr_congestion_on = nr; 87 88 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; 89 if (nr < 1) 90 nr = 1; 91 q->nr_congestion_off = nr; 92 } 93 94 /** 95 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info 96 * @bdev: device 97 * 98 * Locates the passed device's request queue and returns the address of its 99 * backing_dev_info 100 * 101 * Will return NULL if the request queue cannot be located. 102 */ 103 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) 104 { 105 struct backing_dev_info *ret = NULL; 106 struct request_queue *q = bdev_get_queue(bdev); 107 108 if (q) 109 ret = &q->backing_dev_info; 110 return ret; 111 } 112 EXPORT_SYMBOL(blk_get_backing_dev_info); 113 114 void blk_rq_init(struct request_queue *q, struct request *rq) 115 { 116 memset(rq, 0, sizeof(*rq)); 117 118 INIT_LIST_HEAD(&rq->queuelist); 119 INIT_LIST_HEAD(&rq->timeout_list); 120 rq->cpu = -1; 121 rq->q = q; 122 rq->__sector = (sector_t) -1; 123 INIT_HLIST_NODE(&rq->hash); 124 RB_CLEAR_NODE(&rq->rb_node); 125 rq->cmd = rq->__cmd; 126 rq->cmd_len = BLK_MAX_CDB; 127 rq->tag = -1; 128 rq->ref_count = 1; 129 rq->start_time = jiffies; 130 set_start_time_ns(rq); 131 } 132 EXPORT_SYMBOL(blk_rq_init); 133 134 static void req_bio_endio(struct request *rq, struct bio *bio, 135 unsigned int nbytes, int error) 136 { 137 struct request_queue *q = rq->q; 138 139 if (&q->bar_rq != rq) { 140 if (error) 141 clear_bit(BIO_UPTODATE, &bio->bi_flags); 142 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) 143 error = -EIO; 144 145 if (unlikely(nbytes > bio->bi_size)) { 146 printk(KERN_ERR "%s: want %u bytes done, %u left\n", 147 __func__, nbytes, bio->bi_size); 148 nbytes = bio->bi_size; 149 } 150 151 if (unlikely(rq->cmd_flags & REQ_QUIET)) 152 set_bit(BIO_QUIET, &bio->bi_flags); 153 154 bio->bi_size -= nbytes; 155 bio->bi_sector += (nbytes >> 9); 156 157 if (bio_integrity(bio)) 158 bio_integrity_advance(bio, nbytes); 159 160 if (bio->bi_size == 0) 161 bio_endio(bio, error); 162 } else { 163 164 /* 165 * Okay, this is the barrier request in progress, just 166 * record the error; 167 */ 168 if (error && !q->orderr) 169 q->orderr = error; 170 } 171 } 172 173 void blk_dump_rq_flags(struct request *rq, char *msg) 174 { 175 int bit; 176 177 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, 178 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, 179 rq->cmd_flags); 180 181 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", 182 (unsigned long long)blk_rq_pos(rq), 183 blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); 184 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", 185 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); 186 187 if (blk_pc_request(rq)) { 188 printk(KERN_INFO " cdb: "); 189 for (bit = 0; bit < BLK_MAX_CDB; bit++) 190 printk("%02x ", rq->cmd[bit]); 191 printk("\n"); 192 } 193 } 194 EXPORT_SYMBOL(blk_dump_rq_flags); 195 196 /* 197 * "plug" the device if there are no outstanding requests: this will 198 * force the transfer to start only after we have put all the requests 199 * on the list. 200 * 201 * This is called with interrupts off and no requests on the queue and 202 * with the queue lock held. 203 */ 204 void blk_plug_device(struct request_queue *q) 205 { 206 WARN_ON(!irqs_disabled()); 207 208 /* 209 * don't plug a stopped queue, it must be paired with blk_start_queue() 210 * which will restart the queueing 211 */ 212 if (blk_queue_stopped(q)) 213 return; 214 215 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { 216 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); 217 trace_block_plug(q); 218 } 219 } 220 EXPORT_SYMBOL(blk_plug_device); 221 222 /** 223 * blk_plug_device_unlocked - plug a device without queue lock held 224 * @q: The &struct request_queue to plug 225 * 226 * Description: 227 * Like @blk_plug_device(), but grabs the queue lock and disables 228 * interrupts. 229 **/ 230 void blk_plug_device_unlocked(struct request_queue *q) 231 { 232 unsigned long flags; 233 234 spin_lock_irqsave(q->queue_lock, flags); 235 blk_plug_device(q); 236 spin_unlock_irqrestore(q->queue_lock, flags); 237 } 238 EXPORT_SYMBOL(blk_plug_device_unlocked); 239 240 /* 241 * remove the queue from the plugged list, if present. called with 242 * queue lock held and interrupts disabled. 243 */ 244 int blk_remove_plug(struct request_queue *q) 245 { 246 WARN_ON(!irqs_disabled()); 247 248 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) 249 return 0; 250 251 del_timer(&q->unplug_timer); 252 return 1; 253 } 254 EXPORT_SYMBOL(blk_remove_plug); 255 256 /* 257 * remove the plug and let it rip.. 258 */ 259 void __generic_unplug_device(struct request_queue *q) 260 { 261 if (unlikely(blk_queue_stopped(q))) 262 return; 263 if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) 264 return; 265 266 q->request_fn(q); 267 } 268 269 /** 270 * generic_unplug_device - fire a request queue 271 * @q: The &struct request_queue in question 272 * 273 * Description: 274 * Linux uses plugging to build bigger requests queues before letting 275 * the device have at them. If a queue is plugged, the I/O scheduler 276 * is still adding and merging requests on the queue. Once the queue 277 * gets unplugged, the request_fn defined for the queue is invoked and 278 * transfers started. 279 **/ 280 void generic_unplug_device(struct request_queue *q) 281 { 282 if (blk_queue_plugged(q)) { 283 spin_lock_irq(q->queue_lock); 284 __generic_unplug_device(q); 285 spin_unlock_irq(q->queue_lock); 286 } 287 } 288 EXPORT_SYMBOL(generic_unplug_device); 289 290 static void blk_backing_dev_unplug(struct backing_dev_info *bdi, 291 struct page *page) 292 { 293 struct request_queue *q = bdi->unplug_io_data; 294 295 blk_unplug(q); 296 } 297 298 void blk_unplug_work(struct work_struct *work) 299 { 300 struct request_queue *q = 301 container_of(work, struct request_queue, unplug_work); 302 303 trace_block_unplug_io(q); 304 q->unplug_fn(q); 305 } 306 307 void blk_unplug_timeout(unsigned long data) 308 { 309 struct request_queue *q = (struct request_queue *)data; 310 311 trace_block_unplug_timer(q); 312 kblockd_schedule_work(q, &q->unplug_work); 313 } 314 315 void blk_unplug(struct request_queue *q) 316 { 317 /* 318 * devices don't necessarily have an ->unplug_fn defined 319 */ 320 if (q->unplug_fn) { 321 trace_block_unplug_io(q); 322 q->unplug_fn(q); 323 } 324 } 325 EXPORT_SYMBOL(blk_unplug); 326 327 /** 328 * blk_start_queue - restart a previously stopped queue 329 * @q: The &struct request_queue in question 330 * 331 * Description: 332 * blk_start_queue() will clear the stop flag on the queue, and call 333 * the request_fn for the queue if it was in a stopped state when 334 * entered. Also see blk_stop_queue(). Queue lock must be held. 335 **/ 336 void blk_start_queue(struct request_queue *q) 337 { 338 WARN_ON(!irqs_disabled()); 339 340 queue_flag_clear(QUEUE_FLAG_STOPPED, q); 341 __blk_run_queue(q); 342 } 343 EXPORT_SYMBOL(blk_start_queue); 344 345 /** 346 * blk_stop_queue - stop a queue 347 * @q: The &struct request_queue in question 348 * 349 * Description: 350 * The Linux block layer assumes that a block driver will consume all 351 * entries on the request queue when the request_fn strategy is called. 352 * Often this will not happen, because of hardware limitations (queue 353 * depth settings). If a device driver gets a 'queue full' response, 354 * or if it simply chooses not to queue more I/O at one point, it can 355 * call this function to prevent the request_fn from being called until 356 * the driver has signalled it's ready to go again. This happens by calling 357 * blk_start_queue() to restart queue operations. Queue lock must be held. 358 **/ 359 void blk_stop_queue(struct request_queue *q) 360 { 361 blk_remove_plug(q); 362 queue_flag_set(QUEUE_FLAG_STOPPED, q); 363 } 364 EXPORT_SYMBOL(blk_stop_queue); 365 366 /** 367 * blk_sync_queue - cancel any pending callbacks on a queue 368 * @q: the queue 369 * 370 * Description: 371 * The block layer may perform asynchronous callback activity 372 * on a queue, such as calling the unplug function after a timeout. 373 * A block device may call blk_sync_queue to ensure that any 374 * such activity is cancelled, thus allowing it to release resources 375 * that the callbacks might use. The caller must already have made sure 376 * that its ->make_request_fn will not re-add plugging prior to calling 377 * this function. 378 * 379 */ 380 void blk_sync_queue(struct request_queue *q) 381 { 382 del_timer_sync(&q->unplug_timer); 383 del_timer_sync(&q->timeout); 384 cancel_work_sync(&q->unplug_work); 385 } 386 EXPORT_SYMBOL(blk_sync_queue); 387 388 /** 389 * __blk_run_queue - run a single device queue 390 * @q: The queue to run 391 * 392 * Description: 393 * See @blk_run_queue. This variant must be called with the queue lock 394 * held and interrupts disabled. 395 * 396 */ 397 void __blk_run_queue(struct request_queue *q) 398 { 399 blk_remove_plug(q); 400 401 if (unlikely(blk_queue_stopped(q))) 402 return; 403 404 if (elv_queue_empty(q)) 405 return; 406 407 /* 408 * Only recurse once to avoid overrunning the stack, let the unplug 409 * handling reinvoke the handler shortly if we already got there. 410 */ 411 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { 412 q->request_fn(q); 413 queue_flag_clear(QUEUE_FLAG_REENTER, q); 414 } else { 415 queue_flag_set(QUEUE_FLAG_PLUGGED, q); 416 kblockd_schedule_work(q, &q->unplug_work); 417 } 418 } 419 EXPORT_SYMBOL(__blk_run_queue); 420 421 /** 422 * blk_run_queue - run a single device queue 423 * @q: The queue to run 424 * 425 * Description: 426 * Invoke request handling on this queue, if it has pending work to do. 427 * May be used to restart queueing when a request has completed. 428 */ 429 void blk_run_queue(struct request_queue *q) 430 { 431 unsigned long flags; 432 433 spin_lock_irqsave(q->queue_lock, flags); 434 __blk_run_queue(q); 435 spin_unlock_irqrestore(q->queue_lock, flags); 436 } 437 EXPORT_SYMBOL(blk_run_queue); 438 439 void blk_put_queue(struct request_queue *q) 440 { 441 kobject_put(&q->kobj); 442 } 443 444 void blk_cleanup_queue(struct request_queue *q) 445 { 446 /* 447 * We know we have process context here, so we can be a little 448 * cautious and ensure that pending block actions on this device 449 * are done before moving on. Going into this function, we should 450 * not have processes doing IO to this device. 451 */ 452 blk_sync_queue(q); 453 454 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); 455 mutex_lock(&q->sysfs_lock); 456 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); 457 mutex_unlock(&q->sysfs_lock); 458 459 if (q->elevator) 460 elevator_exit(q->elevator); 461 462 blk_put_queue(q); 463 } 464 EXPORT_SYMBOL(blk_cleanup_queue); 465 466 static int blk_init_free_list(struct request_queue *q) 467 { 468 struct request_list *rl = &q->rq; 469 470 if (unlikely(rl->rq_pool)) 471 return 0; 472 473 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; 474 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; 475 rl->elvpriv = 0; 476 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); 477 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); 478 479 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, 480 mempool_free_slab, request_cachep, q->node); 481 482 if (!rl->rq_pool) 483 return -ENOMEM; 484 485 return 0; 486 } 487 488 struct request_queue *blk_alloc_queue(gfp_t gfp_mask) 489 { 490 return blk_alloc_queue_node(gfp_mask, -1); 491 } 492 EXPORT_SYMBOL(blk_alloc_queue); 493 494 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) 495 { 496 struct request_queue *q; 497 int err; 498 499 q = kmem_cache_alloc_node(blk_requestq_cachep, 500 gfp_mask | __GFP_ZERO, node_id); 501 if (!q) 502 return NULL; 503 504 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; 505 q->backing_dev_info.unplug_io_data = q; 506 q->backing_dev_info.ra_pages = 507 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; 508 q->backing_dev_info.state = 0; 509 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; 510 q->backing_dev_info.name = "block"; 511 512 err = bdi_init(&q->backing_dev_info); 513 if (err) { 514 kmem_cache_free(blk_requestq_cachep, q); 515 return NULL; 516 } 517 518 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, 519 laptop_mode_timer_fn, (unsigned long) q); 520 init_timer(&q->unplug_timer); 521 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); 522 INIT_LIST_HEAD(&q->timeout_list); 523 INIT_WORK(&q->unplug_work, blk_unplug_work); 524 525 kobject_init(&q->kobj, &blk_queue_ktype); 526 527 mutex_init(&q->sysfs_lock); 528 spin_lock_init(&q->__queue_lock); 529 530 return q; 531 } 532 EXPORT_SYMBOL(blk_alloc_queue_node); 533 534 /** 535 * blk_init_queue - prepare a request queue for use with a block device 536 * @rfn: The function to be called to process requests that have been 537 * placed on the queue. 538 * @lock: Request queue spin lock 539 * 540 * Description: 541 * If a block device wishes to use the standard request handling procedures, 542 * which sorts requests and coalesces adjacent requests, then it must 543 * call blk_init_queue(). The function @rfn will be called when there 544 * are requests on the queue that need to be processed. If the device 545 * supports plugging, then @rfn may not be called immediately when requests 546 * are available on the queue, but may be called at some time later instead. 547 * Plugged queues are generally unplugged when a buffer belonging to one 548 * of the requests on the queue is needed, or due to memory pressure. 549 * 550 * @rfn is not required, or even expected, to remove all requests off the 551 * queue, but only as many as it can handle at a time. If it does leave 552 * requests on the queue, it is responsible for arranging that the requests 553 * get dealt with eventually. 554 * 555 * The queue spin lock must be held while manipulating the requests on the 556 * request queue; this lock will be taken also from interrupt context, so irq 557 * disabling is needed for it. 558 * 559 * Function returns a pointer to the initialized request queue, or %NULL if 560 * it didn't succeed. 561 * 562 * Note: 563 * blk_init_queue() must be paired with a blk_cleanup_queue() call 564 * when the block device is deactivated (such as at module unload). 565 **/ 566 567 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) 568 { 569 return blk_init_queue_node(rfn, lock, -1); 570 } 571 EXPORT_SYMBOL(blk_init_queue); 572 573 struct request_queue * 574 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) 575 { 576 struct request_queue *uninit_q, *q; 577 578 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); 579 if (!uninit_q) 580 return NULL; 581 582 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id); 583 if (!q) 584 blk_cleanup_queue(uninit_q); 585 586 return q; 587 } 588 EXPORT_SYMBOL(blk_init_queue_node); 589 590 struct request_queue * 591 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, 592 spinlock_t *lock) 593 { 594 return blk_init_allocated_queue_node(q, rfn, lock, -1); 595 } 596 EXPORT_SYMBOL(blk_init_allocated_queue); 597 598 struct request_queue * 599 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn, 600 spinlock_t *lock, int node_id) 601 { 602 if (!q) 603 return NULL; 604 605 q->node = node_id; 606 if (blk_init_free_list(q)) 607 return NULL; 608 609 q->request_fn = rfn; 610 q->prep_rq_fn = NULL; 611 q->unplug_fn = generic_unplug_device; 612 q->queue_flags = QUEUE_FLAG_DEFAULT; 613 q->queue_lock = lock; 614 615 /* 616 * This also sets hw/phys segments, boundary and size 617 */ 618 blk_queue_make_request(q, __make_request); 619 620 q->sg_reserved_size = INT_MAX; 621 622 /* 623 * all done 624 */ 625 if (!elevator_init(q, NULL)) { 626 blk_queue_congestion_threshold(q); 627 return q; 628 } 629 630 return NULL; 631 } 632 EXPORT_SYMBOL(blk_init_allocated_queue_node); 633 634 int blk_get_queue(struct request_queue *q) 635 { 636 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { 637 kobject_get(&q->kobj); 638 return 0; 639 } 640 641 return 1; 642 } 643 644 static inline void blk_free_request(struct request_queue *q, struct request *rq) 645 { 646 if (rq->cmd_flags & REQ_ELVPRIV) 647 elv_put_request(q, rq); 648 mempool_free(rq, q->rq.rq_pool); 649 } 650 651 static struct request * 652 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) 653 { 654 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); 655 656 if (!rq) 657 return NULL; 658 659 blk_rq_init(q, rq); 660 661 rq->cmd_flags = flags | REQ_ALLOCED; 662 663 if (priv) { 664 if (unlikely(elv_set_request(q, rq, gfp_mask))) { 665 mempool_free(rq, q->rq.rq_pool); 666 return NULL; 667 } 668 rq->cmd_flags |= REQ_ELVPRIV; 669 } 670 671 return rq; 672 } 673 674 /* 675 * ioc_batching returns true if the ioc is a valid batching request and 676 * should be given priority access to a request. 677 */ 678 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) 679 { 680 if (!ioc) 681 return 0; 682 683 /* 684 * Make sure the process is able to allocate at least 1 request 685 * even if the batch times out, otherwise we could theoretically 686 * lose wakeups. 687 */ 688 return ioc->nr_batch_requests == q->nr_batching || 689 (ioc->nr_batch_requests > 0 690 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); 691 } 692 693 /* 694 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This 695 * will cause the process to be a "batcher" on all queues in the system. This 696 * is the behaviour we want though - once it gets a wakeup it should be given 697 * a nice run. 698 */ 699 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) 700 { 701 if (!ioc || ioc_batching(q, ioc)) 702 return; 703 704 ioc->nr_batch_requests = q->nr_batching; 705 ioc->last_waited = jiffies; 706 } 707 708 static void __freed_request(struct request_queue *q, int sync) 709 { 710 struct request_list *rl = &q->rq; 711 712 if (rl->count[sync] < queue_congestion_off_threshold(q)) 713 blk_clear_queue_congested(q, sync); 714 715 if (rl->count[sync] + 1 <= q->nr_requests) { 716 if (waitqueue_active(&rl->wait[sync])) 717 wake_up(&rl->wait[sync]); 718 719 blk_clear_queue_full(q, sync); 720 } 721 } 722 723 /* 724 * A request has just been released. Account for it, update the full and 725 * congestion status, wake up any waiters. Called under q->queue_lock. 726 */ 727 static void freed_request(struct request_queue *q, int sync, int priv) 728 { 729 struct request_list *rl = &q->rq; 730 731 rl->count[sync]--; 732 if (priv) 733 rl->elvpriv--; 734 735 __freed_request(q, sync); 736 737 if (unlikely(rl->starved[sync ^ 1])) 738 __freed_request(q, sync ^ 1); 739 } 740 741 /* 742 * Get a free request, queue_lock must be held. 743 * Returns NULL on failure, with queue_lock held. 744 * Returns !NULL on success, with queue_lock *not held*. 745 */ 746 static struct request *get_request(struct request_queue *q, int rw_flags, 747 struct bio *bio, gfp_t gfp_mask) 748 { 749 struct request *rq = NULL; 750 struct request_list *rl = &q->rq; 751 struct io_context *ioc = NULL; 752 const bool is_sync = rw_is_sync(rw_flags) != 0; 753 int may_queue, priv; 754 755 may_queue = elv_may_queue(q, rw_flags); 756 if (may_queue == ELV_MQUEUE_NO) 757 goto rq_starved; 758 759 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { 760 if (rl->count[is_sync]+1 >= q->nr_requests) { 761 ioc = current_io_context(GFP_ATOMIC, q->node); 762 /* 763 * The queue will fill after this allocation, so set 764 * it as full, and mark this process as "batching". 765 * This process will be allowed to complete a batch of 766 * requests, others will be blocked. 767 */ 768 if (!blk_queue_full(q, is_sync)) { 769 ioc_set_batching(q, ioc); 770 blk_set_queue_full(q, is_sync); 771 } else { 772 if (may_queue != ELV_MQUEUE_MUST 773 && !ioc_batching(q, ioc)) { 774 /* 775 * The queue is full and the allocating 776 * process is not a "batcher", and not 777 * exempted by the IO scheduler 778 */ 779 goto out; 780 } 781 } 782 } 783 blk_set_queue_congested(q, is_sync); 784 } 785 786 /* 787 * Only allow batching queuers to allocate up to 50% over the defined 788 * limit of requests, otherwise we could have thousands of requests 789 * allocated with any setting of ->nr_requests 790 */ 791 if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) 792 goto out; 793 794 rl->count[is_sync]++; 795 rl->starved[is_sync] = 0; 796 797 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); 798 if (priv) 799 rl->elvpriv++; 800 801 if (blk_queue_io_stat(q)) 802 rw_flags |= REQ_IO_STAT; 803 spin_unlock_irq(q->queue_lock); 804 805 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); 806 if (unlikely(!rq)) { 807 /* 808 * Allocation failed presumably due to memory. Undo anything 809 * we might have messed up. 810 * 811 * Allocating task should really be put onto the front of the 812 * wait queue, but this is pretty rare. 813 */ 814 spin_lock_irq(q->queue_lock); 815 freed_request(q, is_sync, priv); 816 817 /* 818 * in the very unlikely event that allocation failed and no 819 * requests for this direction was pending, mark us starved 820 * so that freeing of a request in the other direction will 821 * notice us. another possible fix would be to split the 822 * rq mempool into READ and WRITE 823 */ 824 rq_starved: 825 if (unlikely(rl->count[is_sync] == 0)) 826 rl->starved[is_sync] = 1; 827 828 goto out; 829 } 830 831 /* 832 * ioc may be NULL here, and ioc_batching will be false. That's 833 * OK, if the queue is under the request limit then requests need 834 * not count toward the nr_batch_requests limit. There will always 835 * be some limit enforced by BLK_BATCH_TIME. 836 */ 837 if (ioc_batching(q, ioc)) 838 ioc->nr_batch_requests--; 839 840 trace_block_getrq(q, bio, rw_flags & 1); 841 out: 842 return rq; 843 } 844 845 /* 846 * No available requests for this queue, unplug the device and wait for some 847 * requests to become available. 848 * 849 * Called with q->queue_lock held, and returns with it unlocked. 850 */ 851 static struct request *get_request_wait(struct request_queue *q, int rw_flags, 852 struct bio *bio) 853 { 854 const bool is_sync = rw_is_sync(rw_flags) != 0; 855 struct request *rq; 856 857 rq = get_request(q, rw_flags, bio, GFP_NOIO); 858 while (!rq) { 859 DEFINE_WAIT(wait); 860 struct io_context *ioc; 861 struct request_list *rl = &q->rq; 862 863 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, 864 TASK_UNINTERRUPTIBLE); 865 866 trace_block_sleeprq(q, bio, rw_flags & 1); 867 868 __generic_unplug_device(q); 869 spin_unlock_irq(q->queue_lock); 870 io_schedule(); 871 872 /* 873 * After sleeping, we become a "batching" process and 874 * will be able to allocate at least one request, and 875 * up to a big batch of them for a small period time. 876 * See ioc_batching, ioc_set_batching 877 */ 878 ioc = current_io_context(GFP_NOIO, q->node); 879 ioc_set_batching(q, ioc); 880 881 spin_lock_irq(q->queue_lock); 882 finish_wait(&rl->wait[is_sync], &wait); 883 884 rq = get_request(q, rw_flags, bio, GFP_NOIO); 885 }; 886 887 return rq; 888 } 889 890 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) 891 { 892 struct request *rq; 893 894 BUG_ON(rw != READ && rw != WRITE); 895 896 spin_lock_irq(q->queue_lock); 897 if (gfp_mask & __GFP_WAIT) { 898 rq = get_request_wait(q, rw, NULL); 899 } else { 900 rq = get_request(q, rw, NULL, gfp_mask); 901 if (!rq) 902 spin_unlock_irq(q->queue_lock); 903 } 904 /* q->queue_lock is unlocked at this point */ 905 906 return rq; 907 } 908 EXPORT_SYMBOL(blk_get_request); 909 910 /** 911 * blk_make_request - given a bio, allocate a corresponding struct request. 912 * @q: target request queue 913 * @bio: The bio describing the memory mappings that will be submitted for IO. 914 * It may be a chained-bio properly constructed by block/bio layer. 915 * @gfp_mask: gfp flags to be used for memory allocation 916 * 917 * blk_make_request is the parallel of generic_make_request for BLOCK_PC 918 * type commands. Where the struct request needs to be farther initialized by 919 * the caller. It is passed a &struct bio, which describes the memory info of 920 * the I/O transfer. 921 * 922 * The caller of blk_make_request must make sure that bi_io_vec 923 * are set to describe the memory buffers. That bio_data_dir() will return 924 * the needed direction of the request. (And all bio's in the passed bio-chain 925 * are properly set accordingly) 926 * 927 * If called under none-sleepable conditions, mapped bio buffers must not 928 * need bouncing, by calling the appropriate masked or flagged allocator, 929 * suitable for the target device. Otherwise the call to blk_queue_bounce will 930 * BUG. 931 * 932 * WARNING: When allocating/cloning a bio-chain, careful consideration should be 933 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for 934 * anything but the first bio in the chain. Otherwise you risk waiting for IO 935 * completion of a bio that hasn't been submitted yet, thus resulting in a 936 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead 937 * of bio_alloc(), as that avoids the mempool deadlock. 938 * If possible a big IO should be split into smaller parts when allocation 939 * fails. Partial allocation should not be an error, or you risk a live-lock. 940 */ 941 struct request *blk_make_request(struct request_queue *q, struct bio *bio, 942 gfp_t gfp_mask) 943 { 944 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); 945 946 if (unlikely(!rq)) 947 return ERR_PTR(-ENOMEM); 948 949 for_each_bio(bio) { 950 struct bio *bounce_bio = bio; 951 int ret; 952 953 blk_queue_bounce(q, &bounce_bio); 954 ret = blk_rq_append_bio(q, rq, bounce_bio); 955 if (unlikely(ret)) { 956 blk_put_request(rq); 957 return ERR_PTR(ret); 958 } 959 } 960 961 return rq; 962 } 963 EXPORT_SYMBOL(blk_make_request); 964 965 /** 966 * blk_requeue_request - put a request back on queue 967 * @q: request queue where request should be inserted 968 * @rq: request to be inserted 969 * 970 * Description: 971 * Drivers often keep queueing requests until the hardware cannot accept 972 * more, when that condition happens we need to put the request back 973 * on the queue. Must be called with queue lock held. 974 */ 975 void blk_requeue_request(struct request_queue *q, struct request *rq) 976 { 977 blk_delete_timer(rq); 978 blk_clear_rq_complete(rq); 979 trace_block_rq_requeue(q, rq); 980 981 if (blk_rq_tagged(rq)) 982 blk_queue_end_tag(q, rq); 983 984 BUG_ON(blk_queued_rq(rq)); 985 986 elv_requeue_request(q, rq); 987 } 988 EXPORT_SYMBOL(blk_requeue_request); 989 990 /** 991 * blk_insert_request - insert a special request into a request queue 992 * @q: request queue where request should be inserted 993 * @rq: request to be inserted 994 * @at_head: insert request at head or tail of queue 995 * @data: private data 996 * 997 * Description: 998 * Many block devices need to execute commands asynchronously, so they don't 999 * block the whole kernel from preemption during request execution. This is 1000 * accomplished normally by inserting aritficial requests tagged as 1001 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them 1002 * be scheduled for actual execution by the request queue. 1003 * 1004 * We have the option of inserting the head or the tail of the queue. 1005 * Typically we use the tail for new ioctls and so forth. We use the head 1006 * of the queue for things like a QUEUE_FULL message from a device, or a 1007 * host that is unable to accept a particular command. 1008 */ 1009 void blk_insert_request(struct request_queue *q, struct request *rq, 1010 int at_head, void *data) 1011 { 1012 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; 1013 unsigned long flags; 1014 1015 /* 1016 * tell I/O scheduler that this isn't a regular read/write (ie it 1017 * must not attempt merges on this) and that it acts as a soft 1018 * barrier 1019 */ 1020 rq->cmd_type = REQ_TYPE_SPECIAL; 1021 1022 rq->special = data; 1023 1024 spin_lock_irqsave(q->queue_lock, flags); 1025 1026 /* 1027 * If command is tagged, release the tag 1028 */ 1029 if (blk_rq_tagged(rq)) 1030 blk_queue_end_tag(q, rq); 1031 1032 drive_stat_acct(rq, 1); 1033 __elv_add_request(q, rq, where, 0); 1034 __blk_run_queue(q); 1035 spin_unlock_irqrestore(q->queue_lock, flags); 1036 } 1037 EXPORT_SYMBOL(blk_insert_request); 1038 1039 /* 1040 * add-request adds a request to the linked list. 1041 * queue lock is held and interrupts disabled, as we muck with the 1042 * request queue list. 1043 */ 1044 static inline void add_request(struct request_queue *q, struct request *req) 1045 { 1046 drive_stat_acct(req, 1); 1047 1048 /* 1049 * elevator indicated where it wants this request to be 1050 * inserted at elevator_merge time 1051 */ 1052 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); 1053 } 1054 1055 static void part_round_stats_single(int cpu, struct hd_struct *part, 1056 unsigned long now) 1057 { 1058 if (now == part->stamp) 1059 return; 1060 1061 if (part_in_flight(part)) { 1062 __part_stat_add(cpu, part, time_in_queue, 1063 part_in_flight(part) * (now - part->stamp)); 1064 __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); 1065 } 1066 part->stamp = now; 1067 } 1068 1069 /** 1070 * part_round_stats() - Round off the performance stats on a struct disk_stats. 1071 * @cpu: cpu number for stats access 1072 * @part: target partition 1073 * 1074 * The average IO queue length and utilisation statistics are maintained 1075 * by observing the current state of the queue length and the amount of 1076 * time it has been in this state for. 1077 * 1078 * Normally, that accounting is done on IO completion, but that can result 1079 * in more than a second's worth of IO being accounted for within any one 1080 * second, leading to >100% utilisation. To deal with that, we call this 1081 * function to do a round-off before returning the results when reading 1082 * /proc/diskstats. This accounts immediately for all queue usage up to 1083 * the current jiffies and restarts the counters again. 1084 */ 1085 void part_round_stats(int cpu, struct hd_struct *part) 1086 { 1087 unsigned long now = jiffies; 1088 1089 if (part->partno) 1090 part_round_stats_single(cpu, &part_to_disk(part)->part0, now); 1091 part_round_stats_single(cpu, part, now); 1092 } 1093 EXPORT_SYMBOL_GPL(part_round_stats); 1094 1095 /* 1096 * queue lock must be held 1097 */ 1098 void __blk_put_request(struct request_queue *q, struct request *req) 1099 { 1100 if (unlikely(!q)) 1101 return; 1102 if (unlikely(--req->ref_count)) 1103 return; 1104 1105 elv_completed_request(q, req); 1106 1107 /* this is a bio leak */ 1108 WARN_ON(req->bio != NULL); 1109 1110 /* 1111 * Request may not have originated from ll_rw_blk. if not, 1112 * it didn't come out of our reserved rq pools 1113 */ 1114 if (req->cmd_flags & REQ_ALLOCED) { 1115 int is_sync = rq_is_sync(req) != 0; 1116 int priv = req->cmd_flags & REQ_ELVPRIV; 1117 1118 BUG_ON(!list_empty(&req->queuelist)); 1119 BUG_ON(!hlist_unhashed(&req->hash)); 1120 1121 blk_free_request(q, req); 1122 freed_request(q, is_sync, priv); 1123 } 1124 } 1125 EXPORT_SYMBOL_GPL(__blk_put_request); 1126 1127 void blk_put_request(struct request *req) 1128 { 1129 unsigned long flags; 1130 struct request_queue *q = req->q; 1131 1132 spin_lock_irqsave(q->queue_lock, flags); 1133 __blk_put_request(q, req); 1134 spin_unlock_irqrestore(q->queue_lock, flags); 1135 } 1136 EXPORT_SYMBOL(blk_put_request); 1137 1138 void init_request_from_bio(struct request *req, struct bio *bio) 1139 { 1140 req->cpu = bio->bi_comp_cpu; 1141 req->cmd_type = REQ_TYPE_FS; 1142 1143 /* 1144 * Inherit FAILFAST from bio (for read-ahead, and explicit 1145 * FAILFAST). FAILFAST flags are identical for req and bio. 1146 */ 1147 if (bio_rw_flagged(bio, BIO_RW_AHEAD)) 1148 req->cmd_flags |= REQ_FAILFAST_MASK; 1149 else 1150 req->cmd_flags |= bio->bi_rw & REQ_FAILFAST_MASK; 1151 1152 if (bio_rw_flagged(bio, BIO_RW_DISCARD)) 1153 req->cmd_flags |= REQ_DISCARD; 1154 if (bio_rw_flagged(bio, BIO_RW_BARRIER)) 1155 req->cmd_flags |= REQ_HARDBARRIER; 1156 if (bio_rw_flagged(bio, BIO_RW_SYNCIO)) 1157 req->cmd_flags |= REQ_RW_SYNC; 1158 if (bio_rw_flagged(bio, BIO_RW_META)) 1159 req->cmd_flags |= REQ_RW_META; 1160 if (bio_rw_flagged(bio, BIO_RW_NOIDLE)) 1161 req->cmd_flags |= REQ_NOIDLE; 1162 1163 req->errors = 0; 1164 req->__sector = bio->bi_sector; 1165 req->ioprio = bio_prio(bio); 1166 blk_rq_bio_prep(req->q, req, bio); 1167 } 1168 1169 /* 1170 * Only disabling plugging for non-rotational devices if it does tagging 1171 * as well, otherwise we do need the proper merging 1172 */ 1173 static inline bool queue_should_plug(struct request_queue *q) 1174 { 1175 return !(blk_queue_nonrot(q) && blk_queue_tagged(q)); 1176 } 1177 1178 static int __make_request(struct request_queue *q, struct bio *bio) 1179 { 1180 struct request *req; 1181 int el_ret; 1182 unsigned int bytes = bio->bi_size; 1183 const unsigned short prio = bio_prio(bio); 1184 const bool sync = bio_rw_flagged(bio, BIO_RW_SYNCIO); 1185 const bool unplug = bio_rw_flagged(bio, BIO_RW_UNPLUG); 1186 const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK; 1187 int rw_flags; 1188 1189 if (bio_rw_flagged(bio, BIO_RW_BARRIER) && 1190 (q->next_ordered == QUEUE_ORDERED_NONE)) { 1191 bio_endio(bio, -EOPNOTSUPP); 1192 return 0; 1193 } 1194 /* 1195 * low level driver can indicate that it wants pages above a 1196 * certain limit bounced to low memory (ie for highmem, or even 1197 * ISA dma in theory) 1198 */ 1199 blk_queue_bounce(q, &bio); 1200 1201 spin_lock_irq(q->queue_lock); 1202 1203 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)) || elv_queue_empty(q)) 1204 goto get_rq; 1205 1206 el_ret = elv_merge(q, &req, bio); 1207 switch (el_ret) { 1208 case ELEVATOR_BACK_MERGE: 1209 BUG_ON(!rq_mergeable(req)); 1210 1211 if (!ll_back_merge_fn(q, req, bio)) 1212 break; 1213 1214 trace_block_bio_backmerge(q, bio); 1215 1216 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 1217 blk_rq_set_mixed_merge(req); 1218 1219 req->biotail->bi_next = bio; 1220 req->biotail = bio; 1221 req->__data_len += bytes; 1222 req->ioprio = ioprio_best(req->ioprio, prio); 1223 if (!blk_rq_cpu_valid(req)) 1224 req->cpu = bio->bi_comp_cpu; 1225 drive_stat_acct(req, 0); 1226 elv_bio_merged(q, req, bio); 1227 if (!attempt_back_merge(q, req)) 1228 elv_merged_request(q, req, el_ret); 1229 goto out; 1230 1231 case ELEVATOR_FRONT_MERGE: 1232 BUG_ON(!rq_mergeable(req)); 1233 1234 if (!ll_front_merge_fn(q, req, bio)) 1235 break; 1236 1237 trace_block_bio_frontmerge(q, bio); 1238 1239 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) { 1240 blk_rq_set_mixed_merge(req); 1241 req->cmd_flags &= ~REQ_FAILFAST_MASK; 1242 req->cmd_flags |= ff; 1243 } 1244 1245 bio->bi_next = req->bio; 1246 req->bio = bio; 1247 1248 /* 1249 * may not be valid. if the low level driver said 1250 * it didn't need a bounce buffer then it better 1251 * not touch req->buffer either... 1252 */ 1253 req->buffer = bio_data(bio); 1254 req->__sector = bio->bi_sector; 1255 req->__data_len += bytes; 1256 req->ioprio = ioprio_best(req->ioprio, prio); 1257 if (!blk_rq_cpu_valid(req)) 1258 req->cpu = bio->bi_comp_cpu; 1259 drive_stat_acct(req, 0); 1260 elv_bio_merged(q, req, bio); 1261 if (!attempt_front_merge(q, req)) 1262 elv_merged_request(q, req, el_ret); 1263 goto out; 1264 1265 /* ELV_NO_MERGE: elevator says don't/can't merge. */ 1266 default: 1267 ; 1268 } 1269 1270 get_rq: 1271 /* 1272 * This sync check and mask will be re-done in init_request_from_bio(), 1273 * but we need to set it earlier to expose the sync flag to the 1274 * rq allocator and io schedulers. 1275 */ 1276 rw_flags = bio_data_dir(bio); 1277 if (sync) 1278 rw_flags |= REQ_RW_SYNC; 1279 1280 /* 1281 * Grab a free request. This is might sleep but can not fail. 1282 * Returns with the queue unlocked. 1283 */ 1284 req = get_request_wait(q, rw_flags, bio); 1285 1286 /* 1287 * After dropping the lock and possibly sleeping here, our request 1288 * may now be mergeable after it had proven unmergeable (above). 1289 * We don't worry about that case for efficiency. It won't happen 1290 * often, and the elevators are able to handle it. 1291 */ 1292 init_request_from_bio(req, bio); 1293 1294 spin_lock_irq(q->queue_lock); 1295 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || 1296 bio_flagged(bio, BIO_CPU_AFFINE)) 1297 req->cpu = blk_cpu_to_group(smp_processor_id()); 1298 if (queue_should_plug(q) && elv_queue_empty(q)) 1299 blk_plug_device(q); 1300 add_request(q, req); 1301 out: 1302 if (unplug || !queue_should_plug(q)) 1303 __generic_unplug_device(q); 1304 spin_unlock_irq(q->queue_lock); 1305 return 0; 1306 } 1307 1308 /* 1309 * If bio->bi_dev is a partition, remap the location 1310 */ 1311 static inline void blk_partition_remap(struct bio *bio) 1312 { 1313 struct block_device *bdev = bio->bi_bdev; 1314 1315 if (bio_sectors(bio) && bdev != bdev->bd_contains) { 1316 struct hd_struct *p = bdev->bd_part; 1317 1318 bio->bi_sector += p->start_sect; 1319 bio->bi_bdev = bdev->bd_contains; 1320 1321 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, 1322 bdev->bd_dev, 1323 bio->bi_sector - p->start_sect); 1324 } 1325 } 1326 1327 static void handle_bad_sector(struct bio *bio) 1328 { 1329 char b[BDEVNAME_SIZE]; 1330 1331 printk(KERN_INFO "attempt to access beyond end of device\n"); 1332 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", 1333 bdevname(bio->bi_bdev, b), 1334 bio->bi_rw, 1335 (unsigned long long)bio->bi_sector + bio_sectors(bio), 1336 (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); 1337 1338 set_bit(BIO_EOF, &bio->bi_flags); 1339 } 1340 1341 #ifdef CONFIG_FAIL_MAKE_REQUEST 1342 1343 static DECLARE_FAULT_ATTR(fail_make_request); 1344 1345 static int __init setup_fail_make_request(char *str) 1346 { 1347 return setup_fault_attr(&fail_make_request, str); 1348 } 1349 __setup("fail_make_request=", setup_fail_make_request); 1350 1351 static int should_fail_request(struct bio *bio) 1352 { 1353 struct hd_struct *part = bio->bi_bdev->bd_part; 1354 1355 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) 1356 return should_fail(&fail_make_request, bio->bi_size); 1357 1358 return 0; 1359 } 1360 1361 static int __init fail_make_request_debugfs(void) 1362 { 1363 return init_fault_attr_dentries(&fail_make_request, 1364 "fail_make_request"); 1365 } 1366 1367 late_initcall(fail_make_request_debugfs); 1368 1369 #else /* CONFIG_FAIL_MAKE_REQUEST */ 1370 1371 static inline int should_fail_request(struct bio *bio) 1372 { 1373 return 0; 1374 } 1375 1376 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 1377 1378 /* 1379 * Check whether this bio extends beyond the end of the device. 1380 */ 1381 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) 1382 { 1383 sector_t maxsector; 1384 1385 if (!nr_sectors) 1386 return 0; 1387 1388 /* Test device or partition size, when known. */ 1389 maxsector = bio->bi_bdev->bd_inode->i_size >> 9; 1390 if (maxsector) { 1391 sector_t sector = bio->bi_sector; 1392 1393 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { 1394 /* 1395 * This may well happen - the kernel calls bread() 1396 * without checking the size of the device, e.g., when 1397 * mounting a device. 1398 */ 1399 handle_bad_sector(bio); 1400 return 1; 1401 } 1402 } 1403 1404 return 0; 1405 } 1406 1407 /** 1408 * generic_make_request - hand a buffer to its device driver for I/O 1409 * @bio: The bio describing the location in memory and on the device. 1410 * 1411 * generic_make_request() is used to make I/O requests of block 1412 * devices. It is passed a &struct bio, which describes the I/O that needs 1413 * to be done. 1414 * 1415 * generic_make_request() does not return any status. The 1416 * success/failure status of the request, along with notification of 1417 * completion, is delivered asynchronously through the bio->bi_end_io 1418 * function described (one day) else where. 1419 * 1420 * The caller of generic_make_request must make sure that bi_io_vec 1421 * are set to describe the memory buffer, and that bi_dev and bi_sector are 1422 * set to describe the device address, and the 1423 * bi_end_io and optionally bi_private are set to describe how 1424 * completion notification should be signaled. 1425 * 1426 * generic_make_request and the drivers it calls may use bi_next if this 1427 * bio happens to be merged with someone else, and may change bi_dev and 1428 * bi_sector for remaps as it sees fit. So the values of these fields 1429 * should NOT be depended on after the call to generic_make_request. 1430 */ 1431 static inline void __generic_make_request(struct bio *bio) 1432 { 1433 struct request_queue *q; 1434 sector_t old_sector; 1435 int ret, nr_sectors = bio_sectors(bio); 1436 dev_t old_dev; 1437 int err = -EIO; 1438 1439 might_sleep(); 1440 1441 if (bio_check_eod(bio, nr_sectors)) 1442 goto end_io; 1443 1444 /* 1445 * Resolve the mapping until finished. (drivers are 1446 * still free to implement/resolve their own stacking 1447 * by explicitly returning 0) 1448 * 1449 * NOTE: we don't repeat the blk_size check for each new device. 1450 * Stacking drivers are expected to know what they are doing. 1451 */ 1452 old_sector = -1; 1453 old_dev = 0; 1454 do { 1455 char b[BDEVNAME_SIZE]; 1456 1457 q = bdev_get_queue(bio->bi_bdev); 1458 if (unlikely(!q)) { 1459 printk(KERN_ERR 1460 "generic_make_request: Trying to access " 1461 "nonexistent block-device %s (%Lu)\n", 1462 bdevname(bio->bi_bdev, b), 1463 (long long) bio->bi_sector); 1464 goto end_io; 1465 } 1466 1467 if (unlikely(!bio_rw_flagged(bio, BIO_RW_DISCARD) && 1468 nr_sectors > queue_max_hw_sectors(q))) { 1469 printk(KERN_ERR "bio too big device %s (%u > %u)\n", 1470 bdevname(bio->bi_bdev, b), 1471 bio_sectors(bio), 1472 queue_max_hw_sectors(q)); 1473 goto end_io; 1474 } 1475 1476 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) 1477 goto end_io; 1478 1479 if (should_fail_request(bio)) 1480 goto end_io; 1481 1482 /* 1483 * If this device has partitions, remap block n 1484 * of partition p to block n+start(p) of the disk. 1485 */ 1486 blk_partition_remap(bio); 1487 1488 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) 1489 goto end_io; 1490 1491 if (old_sector != -1) 1492 trace_block_remap(q, bio, old_dev, old_sector); 1493 1494 old_sector = bio->bi_sector; 1495 old_dev = bio->bi_bdev->bd_dev; 1496 1497 if (bio_check_eod(bio, nr_sectors)) 1498 goto end_io; 1499 1500 if (bio_rw_flagged(bio, BIO_RW_DISCARD) && 1501 !blk_queue_discard(q)) { 1502 err = -EOPNOTSUPP; 1503 goto end_io; 1504 } 1505 1506 trace_block_bio_queue(q, bio); 1507 1508 ret = q->make_request_fn(q, bio); 1509 } while (ret); 1510 1511 return; 1512 1513 end_io: 1514 bio_endio(bio, err); 1515 } 1516 1517 /* 1518 * We only want one ->make_request_fn to be active at a time, 1519 * else stack usage with stacked devices could be a problem. 1520 * So use current->bio_list to keep a list of requests 1521 * submited by a make_request_fn function. 1522 * current->bio_list is also used as a flag to say if 1523 * generic_make_request is currently active in this task or not. 1524 * If it is NULL, then no make_request is active. If it is non-NULL, 1525 * then a make_request is active, and new requests should be added 1526 * at the tail 1527 */ 1528 void generic_make_request(struct bio *bio) 1529 { 1530 struct bio_list bio_list_on_stack; 1531 1532 if (current->bio_list) { 1533 /* make_request is active */ 1534 bio_list_add(current->bio_list, bio); 1535 return; 1536 } 1537 /* following loop may be a bit non-obvious, and so deserves some 1538 * explanation. 1539 * Before entering the loop, bio->bi_next is NULL (as all callers 1540 * ensure that) so we have a list with a single bio. 1541 * We pretend that we have just taken it off a longer list, so 1542 * we assign bio_list to a pointer to the bio_list_on_stack, 1543 * thus initialising the bio_list of new bios to be 1544 * added. __generic_make_request may indeed add some more bios 1545 * through a recursive call to generic_make_request. If it 1546 * did, we find a non-NULL value in bio_list and re-enter the loop 1547 * from the top. In this case we really did just take the bio 1548 * of the top of the list (no pretending) and so remove it from 1549 * bio_list, and call into __generic_make_request again. 1550 * 1551 * The loop was structured like this to make only one call to 1552 * __generic_make_request (which is important as it is large and 1553 * inlined) and to keep the structure simple. 1554 */ 1555 BUG_ON(bio->bi_next); 1556 bio_list_init(&bio_list_on_stack); 1557 current->bio_list = &bio_list_on_stack; 1558 do { 1559 __generic_make_request(bio); 1560 bio = bio_list_pop(current->bio_list); 1561 } while (bio); 1562 current->bio_list = NULL; /* deactivate */ 1563 } 1564 EXPORT_SYMBOL(generic_make_request); 1565 1566 /** 1567 * submit_bio - submit a bio to the block device layer for I/O 1568 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) 1569 * @bio: The &struct bio which describes the I/O 1570 * 1571 * submit_bio() is very similar in purpose to generic_make_request(), and 1572 * uses that function to do most of the work. Both are fairly rough 1573 * interfaces; @bio must be presetup and ready for I/O. 1574 * 1575 */ 1576 void submit_bio(int rw, struct bio *bio) 1577 { 1578 int count = bio_sectors(bio); 1579 1580 bio->bi_rw |= rw; 1581 1582 /* 1583 * If it's a regular read/write or a barrier with data attached, 1584 * go through the normal accounting stuff before submission. 1585 */ 1586 if (bio_has_data(bio) && !(rw & (1 << BIO_RW_DISCARD))) { 1587 if (rw & WRITE) { 1588 count_vm_events(PGPGOUT, count); 1589 } else { 1590 task_io_account_read(bio->bi_size); 1591 count_vm_events(PGPGIN, count); 1592 } 1593 1594 if (unlikely(block_dump)) { 1595 char b[BDEVNAME_SIZE]; 1596 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", 1597 current->comm, task_pid_nr(current), 1598 (rw & WRITE) ? "WRITE" : "READ", 1599 (unsigned long long)bio->bi_sector, 1600 bdevname(bio->bi_bdev, b)); 1601 } 1602 } 1603 1604 generic_make_request(bio); 1605 } 1606 EXPORT_SYMBOL(submit_bio); 1607 1608 /** 1609 * blk_rq_check_limits - Helper function to check a request for the queue limit 1610 * @q: the queue 1611 * @rq: the request being checked 1612 * 1613 * Description: 1614 * @rq may have been made based on weaker limitations of upper-level queues 1615 * in request stacking drivers, and it may violate the limitation of @q. 1616 * Since the block layer and the underlying device driver trust @rq 1617 * after it is inserted to @q, it should be checked against @q before 1618 * the insertion using this generic function. 1619 * 1620 * This function should also be useful for request stacking drivers 1621 * in some cases below, so export this fuction. 1622 * Request stacking drivers like request-based dm may change the queue 1623 * limits while requests are in the queue (e.g. dm's table swapping). 1624 * Such request stacking drivers should check those requests agaist 1625 * the new queue limits again when they dispatch those requests, 1626 * although such checkings are also done against the old queue limits 1627 * when submitting requests. 1628 */ 1629 int blk_rq_check_limits(struct request_queue *q, struct request *rq) 1630 { 1631 if (blk_rq_sectors(rq) > queue_max_sectors(q) || 1632 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { 1633 printk(KERN_ERR "%s: over max size limit.\n", __func__); 1634 return -EIO; 1635 } 1636 1637 /* 1638 * queue's settings related to segment counting like q->bounce_pfn 1639 * may differ from that of other stacking queues. 1640 * Recalculate it to check the request correctly on this queue's 1641 * limitation. 1642 */ 1643 blk_recalc_rq_segments(rq); 1644 if (rq->nr_phys_segments > queue_max_segments(q)) { 1645 printk(KERN_ERR "%s: over max segments limit.\n", __func__); 1646 return -EIO; 1647 } 1648 1649 return 0; 1650 } 1651 EXPORT_SYMBOL_GPL(blk_rq_check_limits); 1652 1653 /** 1654 * blk_insert_cloned_request - Helper for stacking drivers to submit a request 1655 * @q: the queue to submit the request 1656 * @rq: the request being queued 1657 */ 1658 int blk_insert_cloned_request(struct request_queue *q, struct request *rq) 1659 { 1660 unsigned long flags; 1661 1662 if (blk_rq_check_limits(q, rq)) 1663 return -EIO; 1664 1665 #ifdef CONFIG_FAIL_MAKE_REQUEST 1666 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && 1667 should_fail(&fail_make_request, blk_rq_bytes(rq))) 1668 return -EIO; 1669 #endif 1670 1671 spin_lock_irqsave(q->queue_lock, flags); 1672 1673 /* 1674 * Submitting request must be dequeued before calling this function 1675 * because it will be linked to another request_queue 1676 */ 1677 BUG_ON(blk_queued_rq(rq)); 1678 1679 drive_stat_acct(rq, 1); 1680 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); 1681 1682 spin_unlock_irqrestore(q->queue_lock, flags); 1683 1684 return 0; 1685 } 1686 EXPORT_SYMBOL_GPL(blk_insert_cloned_request); 1687 1688 /** 1689 * blk_rq_err_bytes - determine number of bytes till the next failure boundary 1690 * @rq: request to examine 1691 * 1692 * Description: 1693 * A request could be merge of IOs which require different failure 1694 * handling. This function determines the number of bytes which 1695 * can be failed from the beginning of the request without 1696 * crossing into area which need to be retried further. 1697 * 1698 * Return: 1699 * The number of bytes to fail. 1700 * 1701 * Context: 1702 * queue_lock must be held. 1703 */ 1704 unsigned int blk_rq_err_bytes(const struct request *rq) 1705 { 1706 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; 1707 unsigned int bytes = 0; 1708 struct bio *bio; 1709 1710 if (!(rq->cmd_flags & REQ_MIXED_MERGE)) 1711 return blk_rq_bytes(rq); 1712 1713 /* 1714 * Currently the only 'mixing' which can happen is between 1715 * different fastfail types. We can safely fail portions 1716 * which have all the failfast bits that the first one has - 1717 * the ones which are at least as eager to fail as the first 1718 * one. 1719 */ 1720 for (bio = rq->bio; bio; bio = bio->bi_next) { 1721 if ((bio->bi_rw & ff) != ff) 1722 break; 1723 bytes += bio->bi_size; 1724 } 1725 1726 /* this could lead to infinite loop */ 1727 BUG_ON(blk_rq_bytes(rq) && !bytes); 1728 return bytes; 1729 } 1730 EXPORT_SYMBOL_GPL(blk_rq_err_bytes); 1731 1732 static void blk_account_io_completion(struct request *req, unsigned int bytes) 1733 { 1734 if (blk_do_io_stat(req)) { 1735 const int rw = rq_data_dir(req); 1736 struct hd_struct *part; 1737 int cpu; 1738 1739 cpu = part_stat_lock(); 1740 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req)); 1741 part_stat_add(cpu, part, sectors[rw], bytes >> 9); 1742 part_stat_unlock(); 1743 } 1744 } 1745 1746 static void blk_account_io_done(struct request *req) 1747 { 1748 /* 1749 * Account IO completion. bar_rq isn't accounted as a normal 1750 * IO on queueing nor completion. Accounting the containing 1751 * request is enough. 1752 */ 1753 if (blk_do_io_stat(req) && req != &req->q->bar_rq) { 1754 unsigned long duration = jiffies - req->start_time; 1755 const int rw = rq_data_dir(req); 1756 struct hd_struct *part; 1757 int cpu; 1758 1759 cpu = part_stat_lock(); 1760 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req)); 1761 1762 part_stat_inc(cpu, part, ios[rw]); 1763 part_stat_add(cpu, part, ticks[rw], duration); 1764 part_round_stats(cpu, part); 1765 part_dec_in_flight(part, rw); 1766 1767 part_stat_unlock(); 1768 } 1769 } 1770 1771 /** 1772 * blk_peek_request - peek at the top of a request queue 1773 * @q: request queue to peek at 1774 * 1775 * Description: 1776 * Return the request at the top of @q. The returned request 1777 * should be started using blk_start_request() before LLD starts 1778 * processing it. 1779 * 1780 * Return: 1781 * Pointer to the request at the top of @q if available. Null 1782 * otherwise. 1783 * 1784 * Context: 1785 * queue_lock must be held. 1786 */ 1787 struct request *blk_peek_request(struct request_queue *q) 1788 { 1789 struct request *rq; 1790 int ret; 1791 1792 while ((rq = __elv_next_request(q)) != NULL) { 1793 if (!(rq->cmd_flags & REQ_STARTED)) { 1794 /* 1795 * This is the first time the device driver 1796 * sees this request (possibly after 1797 * requeueing). Notify IO scheduler. 1798 */ 1799 if (blk_sorted_rq(rq)) 1800 elv_activate_rq(q, rq); 1801 1802 /* 1803 * just mark as started even if we don't start 1804 * it, a request that has been delayed should 1805 * not be passed by new incoming requests 1806 */ 1807 rq->cmd_flags |= REQ_STARTED; 1808 trace_block_rq_issue(q, rq); 1809 } 1810 1811 if (!q->boundary_rq || q->boundary_rq == rq) { 1812 q->end_sector = rq_end_sector(rq); 1813 q->boundary_rq = NULL; 1814 } 1815 1816 if (rq->cmd_flags & REQ_DONTPREP) 1817 break; 1818 1819 if (q->dma_drain_size && blk_rq_bytes(rq)) { 1820 /* 1821 * make sure space for the drain appears we 1822 * know we can do this because max_hw_segments 1823 * has been adjusted to be one fewer than the 1824 * device can handle 1825 */ 1826 rq->nr_phys_segments++; 1827 } 1828 1829 if (!q->prep_rq_fn) 1830 break; 1831 1832 ret = q->prep_rq_fn(q, rq); 1833 if (ret == BLKPREP_OK) { 1834 break; 1835 } else if (ret == BLKPREP_DEFER) { 1836 /* 1837 * the request may have been (partially) prepped. 1838 * we need to keep this request in the front to 1839 * avoid resource deadlock. REQ_STARTED will 1840 * prevent other fs requests from passing this one. 1841 */ 1842 if (q->dma_drain_size && blk_rq_bytes(rq) && 1843 !(rq->cmd_flags & REQ_DONTPREP)) { 1844 /* 1845 * remove the space for the drain we added 1846 * so that we don't add it again 1847 */ 1848 --rq->nr_phys_segments; 1849 } 1850 1851 rq = NULL; 1852 break; 1853 } else if (ret == BLKPREP_KILL) { 1854 rq->cmd_flags |= REQ_QUIET; 1855 /* 1856 * Mark this request as started so we don't trigger 1857 * any debug logic in the end I/O path. 1858 */ 1859 blk_start_request(rq); 1860 __blk_end_request_all(rq, -EIO); 1861 } else { 1862 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); 1863 break; 1864 } 1865 } 1866 1867 return rq; 1868 } 1869 EXPORT_SYMBOL(blk_peek_request); 1870 1871 void blk_dequeue_request(struct request *rq) 1872 { 1873 struct request_queue *q = rq->q; 1874 1875 BUG_ON(list_empty(&rq->queuelist)); 1876 BUG_ON(ELV_ON_HASH(rq)); 1877 1878 list_del_init(&rq->queuelist); 1879 1880 /* 1881 * the time frame between a request being removed from the lists 1882 * and to it is freed is accounted as io that is in progress at 1883 * the driver side. 1884 */ 1885 if (blk_account_rq(rq)) { 1886 q->in_flight[rq_is_sync(rq)]++; 1887 set_io_start_time_ns(rq); 1888 } 1889 } 1890 1891 /** 1892 * blk_start_request - start request processing on the driver 1893 * @req: request to dequeue 1894 * 1895 * Description: 1896 * Dequeue @req and start timeout timer on it. This hands off the 1897 * request to the driver. 1898 * 1899 * Block internal functions which don't want to start timer should 1900 * call blk_dequeue_request(). 1901 * 1902 * Context: 1903 * queue_lock must be held. 1904 */ 1905 void blk_start_request(struct request *req) 1906 { 1907 blk_dequeue_request(req); 1908 1909 /* 1910 * We are now handing the request to the hardware, initialize 1911 * resid_len to full count and add the timeout handler. 1912 */ 1913 req->resid_len = blk_rq_bytes(req); 1914 if (unlikely(blk_bidi_rq(req))) 1915 req->next_rq->resid_len = blk_rq_bytes(req->next_rq); 1916 1917 blk_add_timer(req); 1918 } 1919 EXPORT_SYMBOL(blk_start_request); 1920 1921 /** 1922 * blk_fetch_request - fetch a request from a request queue 1923 * @q: request queue to fetch a request from 1924 * 1925 * Description: 1926 * Return the request at the top of @q. The request is started on 1927 * return and LLD can start processing it immediately. 1928 * 1929 * Return: 1930 * Pointer to the request at the top of @q if available. Null 1931 * otherwise. 1932 * 1933 * Context: 1934 * queue_lock must be held. 1935 */ 1936 struct request *blk_fetch_request(struct request_queue *q) 1937 { 1938 struct request *rq; 1939 1940 rq = blk_peek_request(q); 1941 if (rq) 1942 blk_start_request(rq); 1943 return rq; 1944 } 1945 EXPORT_SYMBOL(blk_fetch_request); 1946 1947 /** 1948 * blk_update_request - Special helper function for request stacking drivers 1949 * @req: the request being processed 1950 * @error: %0 for success, < %0 for error 1951 * @nr_bytes: number of bytes to complete @req 1952 * 1953 * Description: 1954 * Ends I/O on a number of bytes attached to @req, but doesn't complete 1955 * the request structure even if @req doesn't have leftover. 1956 * If @req has leftover, sets it up for the next range of segments. 1957 * 1958 * This special helper function is only for request stacking drivers 1959 * (e.g. request-based dm) so that they can handle partial completion. 1960 * Actual device drivers should use blk_end_request instead. 1961 * 1962 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees 1963 * %false return from this function. 1964 * 1965 * Return: 1966 * %false - this request doesn't have any more data 1967 * %true - this request has more data 1968 **/ 1969 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) 1970 { 1971 int total_bytes, bio_nbytes, next_idx = 0; 1972 struct bio *bio; 1973 1974 if (!req->bio) 1975 return false; 1976 1977 trace_block_rq_complete(req->q, req); 1978 1979 /* 1980 * For fs requests, rq is just carrier of independent bio's 1981 * and each partial completion should be handled separately. 1982 * Reset per-request error on each partial completion. 1983 * 1984 * TODO: tj: This is too subtle. It would be better to let 1985 * low level drivers do what they see fit. 1986 */ 1987 if (blk_fs_request(req)) 1988 req->errors = 0; 1989 1990 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { 1991 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", 1992 req->rq_disk ? req->rq_disk->disk_name : "?", 1993 (unsigned long long)blk_rq_pos(req)); 1994 } 1995 1996 blk_account_io_completion(req, nr_bytes); 1997 1998 total_bytes = bio_nbytes = 0; 1999 while ((bio = req->bio) != NULL) { 2000 int nbytes; 2001 2002 if (nr_bytes >= bio->bi_size) { 2003 req->bio = bio->bi_next; 2004 nbytes = bio->bi_size; 2005 req_bio_endio(req, bio, nbytes, error); 2006 next_idx = 0; 2007 bio_nbytes = 0; 2008 } else { 2009 int idx = bio->bi_idx + next_idx; 2010 2011 if (unlikely(idx >= bio->bi_vcnt)) { 2012 blk_dump_rq_flags(req, "__end_that"); 2013 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", 2014 __func__, idx, bio->bi_vcnt); 2015 break; 2016 } 2017 2018 nbytes = bio_iovec_idx(bio, idx)->bv_len; 2019 BIO_BUG_ON(nbytes > bio->bi_size); 2020 2021 /* 2022 * not a complete bvec done 2023 */ 2024 if (unlikely(nbytes > nr_bytes)) { 2025 bio_nbytes += nr_bytes; 2026 total_bytes += nr_bytes; 2027 break; 2028 } 2029 2030 /* 2031 * advance to the next vector 2032 */ 2033 next_idx++; 2034 bio_nbytes += nbytes; 2035 } 2036 2037 total_bytes += nbytes; 2038 nr_bytes -= nbytes; 2039 2040 bio = req->bio; 2041 if (bio) { 2042 /* 2043 * end more in this run, or just return 'not-done' 2044 */ 2045 if (unlikely(nr_bytes <= 0)) 2046 break; 2047 } 2048 } 2049 2050 /* 2051 * completely done 2052 */ 2053 if (!req->bio) { 2054 /* 2055 * Reset counters so that the request stacking driver 2056 * can find how many bytes remain in the request 2057 * later. 2058 */ 2059 req->__data_len = 0; 2060 return false; 2061 } 2062 2063 /* 2064 * if the request wasn't completed, update state 2065 */ 2066 if (bio_nbytes) { 2067 req_bio_endio(req, bio, bio_nbytes, error); 2068 bio->bi_idx += next_idx; 2069 bio_iovec(bio)->bv_offset += nr_bytes; 2070 bio_iovec(bio)->bv_len -= nr_bytes; 2071 } 2072 2073 req->__data_len -= total_bytes; 2074 req->buffer = bio_data(req->bio); 2075 2076 /* update sector only for requests with clear definition of sector */ 2077 if (blk_fs_request(req) || blk_discard_rq(req)) 2078 req->__sector += total_bytes >> 9; 2079 2080 /* mixed attributes always follow the first bio */ 2081 if (req->cmd_flags & REQ_MIXED_MERGE) { 2082 req->cmd_flags &= ~REQ_FAILFAST_MASK; 2083 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; 2084 } 2085 2086 /* 2087 * If total number of sectors is less than the first segment 2088 * size, something has gone terribly wrong. 2089 */ 2090 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { 2091 printk(KERN_ERR "blk: request botched\n"); 2092 req->__data_len = blk_rq_cur_bytes(req); 2093 } 2094 2095 /* recalculate the number of segments */ 2096 blk_recalc_rq_segments(req); 2097 2098 return true; 2099 } 2100 EXPORT_SYMBOL_GPL(blk_update_request); 2101 2102 static bool blk_update_bidi_request(struct request *rq, int error, 2103 unsigned int nr_bytes, 2104 unsigned int bidi_bytes) 2105 { 2106 if (blk_update_request(rq, error, nr_bytes)) 2107 return true; 2108 2109 /* Bidi request must be completed as a whole */ 2110 if (unlikely(blk_bidi_rq(rq)) && 2111 blk_update_request(rq->next_rq, error, bidi_bytes)) 2112 return true; 2113 2114 add_disk_randomness(rq->rq_disk); 2115 2116 return false; 2117 } 2118 2119 /* 2120 * queue lock must be held 2121 */ 2122 static void blk_finish_request(struct request *req, int error) 2123 { 2124 if (blk_rq_tagged(req)) 2125 blk_queue_end_tag(req->q, req); 2126 2127 BUG_ON(blk_queued_rq(req)); 2128 2129 if (unlikely(laptop_mode) && blk_fs_request(req)) 2130 laptop_io_completion(&req->q->backing_dev_info); 2131 2132 blk_delete_timer(req); 2133 2134 blk_account_io_done(req); 2135 2136 if (req->end_io) 2137 req->end_io(req, error); 2138 else { 2139 if (blk_bidi_rq(req)) 2140 __blk_put_request(req->next_rq->q, req->next_rq); 2141 2142 __blk_put_request(req->q, req); 2143 } 2144 } 2145 2146 /** 2147 * blk_end_bidi_request - Complete a bidi request 2148 * @rq: the request to complete 2149 * @error: %0 for success, < %0 for error 2150 * @nr_bytes: number of bytes to complete @rq 2151 * @bidi_bytes: number of bytes to complete @rq->next_rq 2152 * 2153 * Description: 2154 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. 2155 * Drivers that supports bidi can safely call this member for any 2156 * type of request, bidi or uni. In the later case @bidi_bytes is 2157 * just ignored. 2158 * 2159 * Return: 2160 * %false - we are done with this request 2161 * %true - still buffers pending for this request 2162 **/ 2163 static bool blk_end_bidi_request(struct request *rq, int error, 2164 unsigned int nr_bytes, unsigned int bidi_bytes) 2165 { 2166 struct request_queue *q = rq->q; 2167 unsigned long flags; 2168 2169 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) 2170 return true; 2171 2172 spin_lock_irqsave(q->queue_lock, flags); 2173 blk_finish_request(rq, error); 2174 spin_unlock_irqrestore(q->queue_lock, flags); 2175 2176 return false; 2177 } 2178 2179 /** 2180 * __blk_end_bidi_request - Complete a bidi request with queue lock held 2181 * @rq: the request to complete 2182 * @error: %0 for success, < %0 for error 2183 * @nr_bytes: number of bytes to complete @rq 2184 * @bidi_bytes: number of bytes to complete @rq->next_rq 2185 * 2186 * Description: 2187 * Identical to blk_end_bidi_request() except that queue lock is 2188 * assumed to be locked on entry and remains so on return. 2189 * 2190 * Return: 2191 * %false - we are done with this request 2192 * %true - still buffers pending for this request 2193 **/ 2194 static bool __blk_end_bidi_request(struct request *rq, int error, 2195 unsigned int nr_bytes, unsigned int bidi_bytes) 2196 { 2197 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) 2198 return true; 2199 2200 blk_finish_request(rq, error); 2201 2202 return false; 2203 } 2204 2205 /** 2206 * blk_end_request - Helper function for drivers to complete the request. 2207 * @rq: the request being processed 2208 * @error: %0 for success, < %0 for error 2209 * @nr_bytes: number of bytes to complete 2210 * 2211 * Description: 2212 * Ends I/O on a number of bytes attached to @rq. 2213 * If @rq has leftover, sets it up for the next range of segments. 2214 * 2215 * Return: 2216 * %false - we are done with this request 2217 * %true - still buffers pending for this request 2218 **/ 2219 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 2220 { 2221 return blk_end_bidi_request(rq, error, nr_bytes, 0); 2222 } 2223 EXPORT_SYMBOL(blk_end_request); 2224 2225 /** 2226 * blk_end_request_all - Helper function for drives to finish the request. 2227 * @rq: the request to finish 2228 * @error: %0 for success, < %0 for error 2229 * 2230 * Description: 2231 * Completely finish @rq. 2232 */ 2233 void blk_end_request_all(struct request *rq, int error) 2234 { 2235 bool pending; 2236 unsigned int bidi_bytes = 0; 2237 2238 if (unlikely(blk_bidi_rq(rq))) 2239 bidi_bytes = blk_rq_bytes(rq->next_rq); 2240 2241 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); 2242 BUG_ON(pending); 2243 } 2244 EXPORT_SYMBOL(blk_end_request_all); 2245 2246 /** 2247 * blk_end_request_cur - Helper function to finish the current request chunk. 2248 * @rq: the request to finish the current chunk for 2249 * @error: %0 for success, < %0 for error 2250 * 2251 * Description: 2252 * Complete the current consecutively mapped chunk from @rq. 2253 * 2254 * Return: 2255 * %false - we are done with this request 2256 * %true - still buffers pending for this request 2257 */ 2258 bool blk_end_request_cur(struct request *rq, int error) 2259 { 2260 return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); 2261 } 2262 EXPORT_SYMBOL(blk_end_request_cur); 2263 2264 /** 2265 * blk_end_request_err - Finish a request till the next failure boundary. 2266 * @rq: the request to finish till the next failure boundary for 2267 * @error: must be negative errno 2268 * 2269 * Description: 2270 * Complete @rq till the next failure boundary. 2271 * 2272 * Return: 2273 * %false - we are done with this request 2274 * %true - still buffers pending for this request 2275 */ 2276 bool blk_end_request_err(struct request *rq, int error) 2277 { 2278 WARN_ON(error >= 0); 2279 return blk_end_request(rq, error, blk_rq_err_bytes(rq)); 2280 } 2281 EXPORT_SYMBOL_GPL(blk_end_request_err); 2282 2283 /** 2284 * __blk_end_request - Helper function for drivers to complete the request. 2285 * @rq: the request being processed 2286 * @error: %0 for success, < %0 for error 2287 * @nr_bytes: number of bytes to complete 2288 * 2289 * Description: 2290 * Must be called with queue lock held unlike blk_end_request(). 2291 * 2292 * Return: 2293 * %false - we are done with this request 2294 * %true - still buffers pending for this request 2295 **/ 2296 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 2297 { 2298 return __blk_end_bidi_request(rq, error, nr_bytes, 0); 2299 } 2300 EXPORT_SYMBOL(__blk_end_request); 2301 2302 /** 2303 * __blk_end_request_all - Helper function for drives to finish the request. 2304 * @rq: the request to finish 2305 * @error: %0 for success, < %0 for error 2306 * 2307 * Description: 2308 * Completely finish @rq. Must be called with queue lock held. 2309 */ 2310 void __blk_end_request_all(struct request *rq, int error) 2311 { 2312 bool pending; 2313 unsigned int bidi_bytes = 0; 2314 2315 if (unlikely(blk_bidi_rq(rq))) 2316 bidi_bytes = blk_rq_bytes(rq->next_rq); 2317 2318 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); 2319 BUG_ON(pending); 2320 } 2321 EXPORT_SYMBOL(__blk_end_request_all); 2322 2323 /** 2324 * __blk_end_request_cur - Helper function to finish the current request chunk. 2325 * @rq: the request to finish the current chunk for 2326 * @error: %0 for success, < %0 for error 2327 * 2328 * Description: 2329 * Complete the current consecutively mapped chunk from @rq. Must 2330 * be called with queue lock held. 2331 * 2332 * Return: 2333 * %false - we are done with this request 2334 * %true - still buffers pending for this request 2335 */ 2336 bool __blk_end_request_cur(struct request *rq, int error) 2337 { 2338 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); 2339 } 2340 EXPORT_SYMBOL(__blk_end_request_cur); 2341 2342 /** 2343 * __blk_end_request_err - Finish a request till the next failure boundary. 2344 * @rq: the request to finish till the next failure boundary for 2345 * @error: must be negative errno 2346 * 2347 * Description: 2348 * Complete @rq till the next failure boundary. Must be called 2349 * with queue lock held. 2350 * 2351 * Return: 2352 * %false - we are done with this request 2353 * %true - still buffers pending for this request 2354 */ 2355 bool __blk_end_request_err(struct request *rq, int error) 2356 { 2357 WARN_ON(error >= 0); 2358 return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); 2359 } 2360 EXPORT_SYMBOL_GPL(__blk_end_request_err); 2361 2362 void blk_rq_bio_prep(struct request_queue *q, struct request *rq, 2363 struct bio *bio) 2364 { 2365 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ 2366 rq->cmd_flags |= bio->bi_rw & REQ_RW; 2367 2368 if (bio_has_data(bio)) { 2369 rq->nr_phys_segments = bio_phys_segments(q, bio); 2370 rq->buffer = bio_data(bio); 2371 } 2372 rq->__data_len = bio->bi_size; 2373 rq->bio = rq->biotail = bio; 2374 2375 if (bio->bi_bdev) 2376 rq->rq_disk = bio->bi_bdev->bd_disk; 2377 } 2378 2379 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 2380 /** 2381 * rq_flush_dcache_pages - Helper function to flush all pages in a request 2382 * @rq: the request to be flushed 2383 * 2384 * Description: 2385 * Flush all pages in @rq. 2386 */ 2387 void rq_flush_dcache_pages(struct request *rq) 2388 { 2389 struct req_iterator iter; 2390 struct bio_vec *bvec; 2391 2392 rq_for_each_segment(bvec, rq, iter) 2393 flush_dcache_page(bvec->bv_page); 2394 } 2395 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); 2396 #endif 2397 2398 /** 2399 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 2400 * @q : the queue of the device being checked 2401 * 2402 * Description: 2403 * Check if underlying low-level drivers of a device are busy. 2404 * If the drivers want to export their busy state, they must set own 2405 * exporting function using blk_queue_lld_busy() first. 2406 * 2407 * Basically, this function is used only by request stacking drivers 2408 * to stop dispatching requests to underlying devices when underlying 2409 * devices are busy. This behavior helps more I/O merging on the queue 2410 * of the request stacking driver and prevents I/O throughput regression 2411 * on burst I/O load. 2412 * 2413 * Return: 2414 * 0 - Not busy (The request stacking driver should dispatch request) 2415 * 1 - Busy (The request stacking driver should stop dispatching request) 2416 */ 2417 int blk_lld_busy(struct request_queue *q) 2418 { 2419 if (q->lld_busy_fn) 2420 return q->lld_busy_fn(q); 2421 2422 return 0; 2423 } 2424 EXPORT_SYMBOL_GPL(blk_lld_busy); 2425 2426 /** 2427 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request 2428 * @rq: the clone request to be cleaned up 2429 * 2430 * Description: 2431 * Free all bios in @rq for a cloned request. 2432 */ 2433 void blk_rq_unprep_clone(struct request *rq) 2434 { 2435 struct bio *bio; 2436 2437 while ((bio = rq->bio) != NULL) { 2438 rq->bio = bio->bi_next; 2439 2440 bio_put(bio); 2441 } 2442 } 2443 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); 2444 2445 /* 2446 * Copy attributes of the original request to the clone request. 2447 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. 2448 */ 2449 static void __blk_rq_prep_clone(struct request *dst, struct request *src) 2450 { 2451 dst->cpu = src->cpu; 2452 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE); 2453 dst->cmd_type = src->cmd_type; 2454 dst->__sector = blk_rq_pos(src); 2455 dst->__data_len = blk_rq_bytes(src); 2456 dst->nr_phys_segments = src->nr_phys_segments; 2457 dst->ioprio = src->ioprio; 2458 dst->extra_len = src->extra_len; 2459 } 2460 2461 /** 2462 * blk_rq_prep_clone - Helper function to setup clone request 2463 * @rq: the request to be setup 2464 * @rq_src: original request to be cloned 2465 * @bs: bio_set that bios for clone are allocated from 2466 * @gfp_mask: memory allocation mask for bio 2467 * @bio_ctr: setup function to be called for each clone bio. 2468 * Returns %0 for success, non %0 for failure. 2469 * @data: private data to be passed to @bio_ctr 2470 * 2471 * Description: 2472 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. 2473 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) 2474 * are not copied, and copying such parts is the caller's responsibility. 2475 * Also, pages which the original bios are pointing to are not copied 2476 * and the cloned bios just point same pages. 2477 * So cloned bios must be completed before original bios, which means 2478 * the caller must complete @rq before @rq_src. 2479 */ 2480 int blk_rq_prep_clone(struct request *rq, struct request *rq_src, 2481 struct bio_set *bs, gfp_t gfp_mask, 2482 int (*bio_ctr)(struct bio *, struct bio *, void *), 2483 void *data) 2484 { 2485 struct bio *bio, *bio_src; 2486 2487 if (!bs) 2488 bs = fs_bio_set; 2489 2490 blk_rq_init(NULL, rq); 2491 2492 __rq_for_each_bio(bio_src, rq_src) { 2493 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); 2494 if (!bio) 2495 goto free_and_out; 2496 2497 __bio_clone(bio, bio_src); 2498 2499 if (bio_integrity(bio_src) && 2500 bio_integrity_clone(bio, bio_src, gfp_mask, bs)) 2501 goto free_and_out; 2502 2503 if (bio_ctr && bio_ctr(bio, bio_src, data)) 2504 goto free_and_out; 2505 2506 if (rq->bio) { 2507 rq->biotail->bi_next = bio; 2508 rq->biotail = bio; 2509 } else 2510 rq->bio = rq->biotail = bio; 2511 } 2512 2513 __blk_rq_prep_clone(rq, rq_src); 2514 2515 return 0; 2516 2517 free_and_out: 2518 if (bio) 2519 bio_free(bio, bs); 2520 blk_rq_unprep_clone(rq); 2521 2522 return -ENOMEM; 2523 } 2524 EXPORT_SYMBOL_GPL(blk_rq_prep_clone); 2525 2526 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) 2527 { 2528 return queue_work(kblockd_workqueue, work); 2529 } 2530 EXPORT_SYMBOL(kblockd_schedule_work); 2531 2532 int __init blk_dev_init(void) 2533 { 2534 BUILD_BUG_ON(__REQ_NR_BITS > 8 * 2535 sizeof(((struct request *)0)->cmd_flags)); 2536 2537 kblockd_workqueue = create_workqueue("kblockd"); 2538 if (!kblockd_workqueue) 2539 panic("Failed to create kblockd\n"); 2540 2541 request_cachep = kmem_cache_create("blkdev_requests", 2542 sizeof(struct request), 0, SLAB_PANIC, NULL); 2543 2544 blk_requestq_cachep = kmem_cache_create("blkdev_queue", 2545 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 2546 2547 return 0; 2548 } 2549