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->flush_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 * Okay, this is the sequenced flush request in 165 * progress, just record the error; 166 */ 167 if (error && !q->flush_err) 168 q->flush_err = error; 169 } 170 } 171 172 void blk_dump_rq_flags(struct request *rq, char *msg) 173 { 174 int bit; 175 176 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, 177 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, 178 rq->cmd_flags); 179 180 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", 181 (unsigned long long)blk_rq_pos(rq), 182 blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); 183 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", 184 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); 185 186 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { 187 printk(KERN_INFO " cdb: "); 188 for (bit = 0; bit < BLK_MAX_CDB; bit++) 189 printk("%02x ", rq->cmd[bit]); 190 printk("\n"); 191 } 192 } 193 EXPORT_SYMBOL(blk_dump_rq_flags); 194 195 /* 196 * "plug" the device if there are no outstanding requests: this will 197 * force the transfer to start only after we have put all the requests 198 * on the list. 199 * 200 * This is called with interrupts off and no requests on the queue and 201 * with the queue lock held. 202 */ 203 void blk_plug_device(struct request_queue *q) 204 { 205 WARN_ON(!irqs_disabled()); 206 207 /* 208 * don't plug a stopped queue, it must be paired with blk_start_queue() 209 * which will restart the queueing 210 */ 211 if (blk_queue_stopped(q)) 212 return; 213 214 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { 215 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); 216 trace_block_plug(q); 217 } 218 } 219 EXPORT_SYMBOL(blk_plug_device); 220 221 /** 222 * blk_plug_device_unlocked - plug a device without queue lock held 223 * @q: The &struct request_queue to plug 224 * 225 * Description: 226 * Like @blk_plug_device(), but grabs the queue lock and disables 227 * interrupts. 228 **/ 229 void blk_plug_device_unlocked(struct request_queue *q) 230 { 231 unsigned long flags; 232 233 spin_lock_irqsave(q->queue_lock, flags); 234 blk_plug_device(q); 235 spin_unlock_irqrestore(q->queue_lock, flags); 236 } 237 EXPORT_SYMBOL(blk_plug_device_unlocked); 238 239 /* 240 * remove the queue from the plugged list, if present. called with 241 * queue lock held and interrupts disabled. 242 */ 243 int blk_remove_plug(struct request_queue *q) 244 { 245 WARN_ON(!irqs_disabled()); 246 247 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) 248 return 0; 249 250 del_timer(&q->unplug_timer); 251 return 1; 252 } 253 EXPORT_SYMBOL(blk_remove_plug); 254 255 /* 256 * remove the plug and let it rip.. 257 */ 258 void __generic_unplug_device(struct request_queue *q) 259 { 260 if (unlikely(blk_queue_stopped(q))) 261 return; 262 if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) 263 return; 264 265 q->request_fn(q); 266 } 267 268 /** 269 * generic_unplug_device - fire a request queue 270 * @q: The &struct request_queue in question 271 * 272 * Description: 273 * Linux uses plugging to build bigger requests queues before letting 274 * the device have at them. If a queue is plugged, the I/O scheduler 275 * is still adding and merging requests on the queue. Once the queue 276 * gets unplugged, the request_fn defined for the queue is invoked and 277 * transfers started. 278 **/ 279 void generic_unplug_device(struct request_queue *q) 280 { 281 if (blk_queue_plugged(q)) { 282 spin_lock_irq(q->queue_lock); 283 __generic_unplug_device(q); 284 spin_unlock_irq(q->queue_lock); 285 } 286 } 287 EXPORT_SYMBOL(generic_unplug_device); 288 289 static void blk_backing_dev_unplug(struct backing_dev_info *bdi, 290 struct page *page) 291 { 292 struct request_queue *q = bdi->unplug_io_data; 293 294 blk_unplug(q); 295 } 296 297 void blk_unplug_work(struct work_struct *work) 298 { 299 struct request_queue *q = 300 container_of(work, struct request_queue, unplug_work); 301 302 trace_block_unplug_io(q); 303 q->unplug_fn(q); 304 } 305 306 void blk_unplug_timeout(unsigned long data) 307 { 308 struct request_queue *q = (struct request_queue *)data; 309 310 trace_block_unplug_timer(q); 311 kblockd_schedule_work(q, &q->unplug_work); 312 } 313 314 void blk_unplug(struct request_queue *q) 315 { 316 /* 317 * devices don't necessarily have an ->unplug_fn defined 318 */ 319 if (q->unplug_fn) { 320 trace_block_unplug_io(q); 321 q->unplug_fn(q); 322 } 323 } 324 EXPORT_SYMBOL(blk_unplug); 325 326 /** 327 * blk_start_queue - restart a previously stopped queue 328 * @q: The &struct request_queue in question 329 * 330 * Description: 331 * blk_start_queue() will clear the stop flag on the queue, and call 332 * the request_fn for the queue if it was in a stopped state when 333 * entered. Also see blk_stop_queue(). Queue lock must be held. 334 **/ 335 void blk_start_queue(struct request_queue *q) 336 { 337 WARN_ON(!irqs_disabled()); 338 339 queue_flag_clear(QUEUE_FLAG_STOPPED, q); 340 __blk_run_queue(q); 341 } 342 EXPORT_SYMBOL(blk_start_queue); 343 344 /** 345 * blk_stop_queue - stop a queue 346 * @q: The &struct request_queue in question 347 * 348 * Description: 349 * The Linux block layer assumes that a block driver will consume all 350 * entries on the request queue when the request_fn strategy is called. 351 * Often this will not happen, because of hardware limitations (queue 352 * depth settings). If a device driver gets a 'queue full' response, 353 * or if it simply chooses not to queue more I/O at one point, it can 354 * call this function to prevent the request_fn from being called until 355 * the driver has signalled it's ready to go again. This happens by calling 356 * blk_start_queue() to restart queue operations. Queue lock must be held. 357 **/ 358 void blk_stop_queue(struct request_queue *q) 359 { 360 blk_remove_plug(q); 361 queue_flag_set(QUEUE_FLAG_STOPPED, q); 362 } 363 EXPORT_SYMBOL(blk_stop_queue); 364 365 /** 366 * blk_sync_queue - cancel any pending callbacks on a queue 367 * @q: the queue 368 * 369 * Description: 370 * The block layer may perform asynchronous callback activity 371 * on a queue, such as calling the unplug function after a timeout. 372 * A block device may call blk_sync_queue to ensure that any 373 * such activity is cancelled, thus allowing it to release resources 374 * that the callbacks might use. The caller must already have made sure 375 * that its ->make_request_fn will not re-add plugging prior to calling 376 * this function. 377 * 378 */ 379 void blk_sync_queue(struct request_queue *q) 380 { 381 del_timer_sync(&q->unplug_timer); 382 del_timer_sync(&q->timeout); 383 cancel_work_sync(&q->unplug_work); 384 throtl_shutdown_timer_wq(q); 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 if (blk_throtl_init(q)) { 519 kmem_cache_free(blk_requestq_cachep, q); 520 return NULL; 521 } 522 523 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, 524 laptop_mode_timer_fn, (unsigned long) q); 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_LIST_HEAD(&q->pending_flushes); 529 INIT_WORK(&q->unplug_work, blk_unplug_work); 530 531 kobject_init(&q->kobj, &blk_queue_ktype); 532 533 mutex_init(&q->sysfs_lock); 534 spin_lock_init(&q->__queue_lock); 535 536 return q; 537 } 538 EXPORT_SYMBOL(blk_alloc_queue_node); 539 540 /** 541 * blk_init_queue - prepare a request queue for use with a block device 542 * @rfn: The function to be called to process requests that have been 543 * placed on the queue. 544 * @lock: Request queue spin lock 545 * 546 * Description: 547 * If a block device wishes to use the standard request handling procedures, 548 * which sorts requests and coalesces adjacent requests, then it must 549 * call blk_init_queue(). The function @rfn will be called when there 550 * are requests on the queue that need to be processed. If the device 551 * supports plugging, then @rfn may not be called immediately when requests 552 * are available on the queue, but may be called at some time later instead. 553 * Plugged queues are generally unplugged when a buffer belonging to one 554 * of the requests on the queue is needed, or due to memory pressure. 555 * 556 * @rfn is not required, or even expected, to remove all requests off the 557 * queue, but only as many as it can handle at a time. If it does leave 558 * requests on the queue, it is responsible for arranging that the requests 559 * get dealt with eventually. 560 * 561 * The queue spin lock must be held while manipulating the requests on the 562 * request queue; this lock will be taken also from interrupt context, so irq 563 * disabling is needed for it. 564 * 565 * Function returns a pointer to the initialized request queue, or %NULL if 566 * it didn't succeed. 567 * 568 * Note: 569 * blk_init_queue() must be paired with a blk_cleanup_queue() call 570 * when the block device is deactivated (such as at module unload). 571 **/ 572 573 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) 574 { 575 return blk_init_queue_node(rfn, lock, -1); 576 } 577 EXPORT_SYMBOL(blk_init_queue); 578 579 struct request_queue * 580 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) 581 { 582 struct request_queue *uninit_q, *q; 583 584 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); 585 if (!uninit_q) 586 return NULL; 587 588 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id); 589 if (!q) 590 blk_cleanup_queue(uninit_q); 591 592 return q; 593 } 594 EXPORT_SYMBOL(blk_init_queue_node); 595 596 struct request_queue * 597 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, 598 spinlock_t *lock) 599 { 600 return blk_init_allocated_queue_node(q, rfn, lock, -1); 601 } 602 EXPORT_SYMBOL(blk_init_allocated_queue); 603 604 struct request_queue * 605 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn, 606 spinlock_t *lock, int node_id) 607 { 608 if (!q) 609 return NULL; 610 611 q->node = node_id; 612 if (blk_init_free_list(q)) 613 return NULL; 614 615 q->request_fn = rfn; 616 q->prep_rq_fn = NULL; 617 q->unprep_rq_fn = NULL; 618 q->unplug_fn = generic_unplug_device; 619 q->queue_flags = QUEUE_FLAG_DEFAULT; 620 q->queue_lock = lock; 621 622 /* 623 * This also sets hw/phys segments, boundary and size 624 */ 625 blk_queue_make_request(q, __make_request); 626 627 q->sg_reserved_size = INT_MAX; 628 629 /* 630 * all done 631 */ 632 if (!elevator_init(q, NULL)) { 633 blk_queue_congestion_threshold(q); 634 return q; 635 } 636 637 return NULL; 638 } 639 EXPORT_SYMBOL(blk_init_allocated_queue_node); 640 641 int blk_get_queue(struct request_queue *q) 642 { 643 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { 644 kobject_get(&q->kobj); 645 return 0; 646 } 647 648 return 1; 649 } 650 651 static inline void blk_free_request(struct request_queue *q, struct request *rq) 652 { 653 if (rq->cmd_flags & REQ_ELVPRIV) 654 elv_put_request(q, rq); 655 mempool_free(rq, q->rq.rq_pool); 656 } 657 658 static struct request * 659 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) 660 { 661 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); 662 663 if (!rq) 664 return NULL; 665 666 blk_rq_init(q, rq); 667 668 rq->cmd_flags = flags | REQ_ALLOCED; 669 670 if (priv) { 671 if (unlikely(elv_set_request(q, rq, gfp_mask))) { 672 mempool_free(rq, q->rq.rq_pool); 673 return NULL; 674 } 675 rq->cmd_flags |= REQ_ELVPRIV; 676 } 677 678 return rq; 679 } 680 681 /* 682 * ioc_batching returns true if the ioc is a valid batching request and 683 * should be given priority access to a request. 684 */ 685 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) 686 { 687 if (!ioc) 688 return 0; 689 690 /* 691 * Make sure the process is able to allocate at least 1 request 692 * even if the batch times out, otherwise we could theoretically 693 * lose wakeups. 694 */ 695 return ioc->nr_batch_requests == q->nr_batching || 696 (ioc->nr_batch_requests > 0 697 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); 698 } 699 700 /* 701 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This 702 * will cause the process to be a "batcher" on all queues in the system. This 703 * is the behaviour we want though - once it gets a wakeup it should be given 704 * a nice run. 705 */ 706 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) 707 { 708 if (!ioc || ioc_batching(q, ioc)) 709 return; 710 711 ioc->nr_batch_requests = q->nr_batching; 712 ioc->last_waited = jiffies; 713 } 714 715 static void __freed_request(struct request_queue *q, int sync) 716 { 717 struct request_list *rl = &q->rq; 718 719 if (rl->count[sync] < queue_congestion_off_threshold(q)) 720 blk_clear_queue_congested(q, sync); 721 722 if (rl->count[sync] + 1 <= q->nr_requests) { 723 if (waitqueue_active(&rl->wait[sync])) 724 wake_up(&rl->wait[sync]); 725 726 blk_clear_queue_full(q, sync); 727 } 728 } 729 730 /* 731 * A request has just been released. Account for it, update the full and 732 * congestion status, wake up any waiters. Called under q->queue_lock. 733 */ 734 static void freed_request(struct request_queue *q, int sync, int priv) 735 { 736 struct request_list *rl = &q->rq; 737 738 rl->count[sync]--; 739 if (priv) 740 rl->elvpriv--; 741 742 __freed_request(q, sync); 743 744 if (unlikely(rl->starved[sync ^ 1])) 745 __freed_request(q, sync ^ 1); 746 } 747 748 /* 749 * Get a free request, queue_lock must be held. 750 * Returns NULL on failure, with queue_lock held. 751 * Returns !NULL on success, with queue_lock *not held*. 752 */ 753 static struct request *get_request(struct request_queue *q, int rw_flags, 754 struct bio *bio, gfp_t gfp_mask) 755 { 756 struct request *rq = NULL; 757 struct request_list *rl = &q->rq; 758 struct io_context *ioc = NULL; 759 const bool is_sync = rw_is_sync(rw_flags) != 0; 760 int may_queue, priv; 761 762 may_queue = elv_may_queue(q, rw_flags); 763 if (may_queue == ELV_MQUEUE_NO) 764 goto rq_starved; 765 766 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { 767 if (rl->count[is_sync]+1 >= q->nr_requests) { 768 ioc = current_io_context(GFP_ATOMIC, q->node); 769 /* 770 * The queue will fill after this allocation, so set 771 * it as full, and mark this process as "batching". 772 * This process will be allowed to complete a batch of 773 * requests, others will be blocked. 774 */ 775 if (!blk_queue_full(q, is_sync)) { 776 ioc_set_batching(q, ioc); 777 blk_set_queue_full(q, is_sync); 778 } else { 779 if (may_queue != ELV_MQUEUE_MUST 780 && !ioc_batching(q, ioc)) { 781 /* 782 * The queue is full and the allocating 783 * process is not a "batcher", and not 784 * exempted by the IO scheduler 785 */ 786 goto out; 787 } 788 } 789 } 790 blk_set_queue_congested(q, is_sync); 791 } 792 793 /* 794 * Only allow batching queuers to allocate up to 50% over the defined 795 * limit of requests, otherwise we could have thousands of requests 796 * allocated with any setting of ->nr_requests 797 */ 798 if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) 799 goto out; 800 801 rl->count[is_sync]++; 802 rl->starved[is_sync] = 0; 803 804 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); 805 if (priv) 806 rl->elvpriv++; 807 808 if (blk_queue_io_stat(q)) 809 rw_flags |= REQ_IO_STAT; 810 spin_unlock_irq(q->queue_lock); 811 812 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); 813 if (unlikely(!rq)) { 814 /* 815 * Allocation failed presumably due to memory. Undo anything 816 * we might have messed up. 817 * 818 * Allocating task should really be put onto the front of the 819 * wait queue, but this is pretty rare. 820 */ 821 spin_lock_irq(q->queue_lock); 822 freed_request(q, is_sync, priv); 823 824 /* 825 * in the very unlikely event that allocation failed and no 826 * requests for this direction was pending, mark us starved 827 * so that freeing of a request in the other direction will 828 * notice us. another possible fix would be to split the 829 * rq mempool into READ and WRITE 830 */ 831 rq_starved: 832 if (unlikely(rl->count[is_sync] == 0)) 833 rl->starved[is_sync] = 1; 834 835 goto out; 836 } 837 838 /* 839 * ioc may be NULL here, and ioc_batching will be false. That's 840 * OK, if the queue is under the request limit then requests need 841 * not count toward the nr_batch_requests limit. There will always 842 * be some limit enforced by BLK_BATCH_TIME. 843 */ 844 if (ioc_batching(q, ioc)) 845 ioc->nr_batch_requests--; 846 847 trace_block_getrq(q, bio, rw_flags & 1); 848 out: 849 return rq; 850 } 851 852 /* 853 * No available requests for this queue, unplug the device and wait for some 854 * requests to become available. 855 * 856 * Called with q->queue_lock held, and returns with it unlocked. 857 */ 858 static struct request *get_request_wait(struct request_queue *q, int rw_flags, 859 struct bio *bio) 860 { 861 const bool is_sync = rw_is_sync(rw_flags) != 0; 862 struct request *rq; 863 864 rq = get_request(q, rw_flags, bio, GFP_NOIO); 865 while (!rq) { 866 DEFINE_WAIT(wait); 867 struct io_context *ioc; 868 struct request_list *rl = &q->rq; 869 870 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, 871 TASK_UNINTERRUPTIBLE); 872 873 trace_block_sleeprq(q, bio, rw_flags & 1); 874 875 __generic_unplug_device(q); 876 spin_unlock_irq(q->queue_lock); 877 io_schedule(); 878 879 /* 880 * After sleeping, we become a "batching" process and 881 * will be able to allocate at least one request, and 882 * up to a big batch of them for a small period time. 883 * See ioc_batching, ioc_set_batching 884 */ 885 ioc = current_io_context(GFP_NOIO, q->node); 886 ioc_set_batching(q, ioc); 887 888 spin_lock_irq(q->queue_lock); 889 finish_wait(&rl->wait[is_sync], &wait); 890 891 rq = get_request(q, rw_flags, bio, GFP_NOIO); 892 }; 893 894 return rq; 895 } 896 897 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) 898 { 899 struct request *rq; 900 901 BUG_ON(rw != READ && rw != WRITE); 902 903 spin_lock_irq(q->queue_lock); 904 if (gfp_mask & __GFP_WAIT) { 905 rq = get_request_wait(q, rw, NULL); 906 } else { 907 rq = get_request(q, rw, NULL, gfp_mask); 908 if (!rq) 909 spin_unlock_irq(q->queue_lock); 910 } 911 /* q->queue_lock is unlocked at this point */ 912 913 return rq; 914 } 915 EXPORT_SYMBOL(blk_get_request); 916 917 /** 918 * blk_make_request - given a bio, allocate a corresponding struct request. 919 * @q: target request queue 920 * @bio: The bio describing the memory mappings that will be submitted for IO. 921 * It may be a chained-bio properly constructed by block/bio layer. 922 * @gfp_mask: gfp flags to be used for memory allocation 923 * 924 * blk_make_request is the parallel of generic_make_request for BLOCK_PC 925 * type commands. Where the struct request needs to be farther initialized by 926 * the caller. It is passed a &struct bio, which describes the memory info of 927 * the I/O transfer. 928 * 929 * The caller of blk_make_request must make sure that bi_io_vec 930 * are set to describe the memory buffers. That bio_data_dir() will return 931 * the needed direction of the request. (And all bio's in the passed bio-chain 932 * are properly set accordingly) 933 * 934 * If called under none-sleepable conditions, mapped bio buffers must not 935 * need bouncing, by calling the appropriate masked or flagged allocator, 936 * suitable for the target device. Otherwise the call to blk_queue_bounce will 937 * BUG. 938 * 939 * WARNING: When allocating/cloning a bio-chain, careful consideration should be 940 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for 941 * anything but the first bio in the chain. Otherwise you risk waiting for IO 942 * completion of a bio that hasn't been submitted yet, thus resulting in a 943 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead 944 * of bio_alloc(), as that avoids the mempool deadlock. 945 * If possible a big IO should be split into smaller parts when allocation 946 * fails. Partial allocation should not be an error, or you risk a live-lock. 947 */ 948 struct request *blk_make_request(struct request_queue *q, struct bio *bio, 949 gfp_t gfp_mask) 950 { 951 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); 952 953 if (unlikely(!rq)) 954 return ERR_PTR(-ENOMEM); 955 956 for_each_bio(bio) { 957 struct bio *bounce_bio = bio; 958 int ret; 959 960 blk_queue_bounce(q, &bounce_bio); 961 ret = blk_rq_append_bio(q, rq, bounce_bio); 962 if (unlikely(ret)) { 963 blk_put_request(rq); 964 return ERR_PTR(ret); 965 } 966 } 967 968 return rq; 969 } 970 EXPORT_SYMBOL(blk_make_request); 971 972 /** 973 * blk_requeue_request - put a request back on queue 974 * @q: request queue where request should be inserted 975 * @rq: request to be inserted 976 * 977 * Description: 978 * Drivers often keep queueing requests until the hardware cannot accept 979 * more, when that condition happens we need to put the request back 980 * on the queue. Must be called with queue lock held. 981 */ 982 void blk_requeue_request(struct request_queue *q, struct request *rq) 983 { 984 blk_delete_timer(rq); 985 blk_clear_rq_complete(rq); 986 trace_block_rq_requeue(q, rq); 987 988 if (blk_rq_tagged(rq)) 989 blk_queue_end_tag(q, rq); 990 991 BUG_ON(blk_queued_rq(rq)); 992 993 elv_requeue_request(q, rq); 994 } 995 EXPORT_SYMBOL(blk_requeue_request); 996 997 /** 998 * blk_insert_request - insert a special request into a request queue 999 * @q: request queue where request should be inserted 1000 * @rq: request to be inserted 1001 * @at_head: insert request at head or tail of queue 1002 * @data: private data 1003 * 1004 * Description: 1005 * Many block devices need to execute commands asynchronously, so they don't 1006 * block the whole kernel from preemption during request execution. This is 1007 * accomplished normally by inserting aritficial requests tagged as 1008 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them 1009 * be scheduled for actual execution by the request queue. 1010 * 1011 * We have the option of inserting the head or the tail of the queue. 1012 * Typically we use the tail for new ioctls and so forth. We use the head 1013 * of the queue for things like a QUEUE_FULL message from a device, or a 1014 * host that is unable to accept a particular command. 1015 */ 1016 void blk_insert_request(struct request_queue *q, struct request *rq, 1017 int at_head, void *data) 1018 { 1019 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; 1020 unsigned long flags; 1021 1022 /* 1023 * tell I/O scheduler that this isn't a regular read/write (ie it 1024 * must not attempt merges on this) and that it acts as a soft 1025 * barrier 1026 */ 1027 rq->cmd_type = REQ_TYPE_SPECIAL; 1028 1029 rq->special = data; 1030 1031 spin_lock_irqsave(q->queue_lock, flags); 1032 1033 /* 1034 * If command is tagged, release the tag 1035 */ 1036 if (blk_rq_tagged(rq)) 1037 blk_queue_end_tag(q, rq); 1038 1039 drive_stat_acct(rq, 1); 1040 __elv_add_request(q, rq, where, 0); 1041 __blk_run_queue(q); 1042 spin_unlock_irqrestore(q->queue_lock, flags); 1043 } 1044 EXPORT_SYMBOL(blk_insert_request); 1045 1046 static void part_round_stats_single(int cpu, struct hd_struct *part, 1047 unsigned long now) 1048 { 1049 if (now == part->stamp) 1050 return; 1051 1052 if (part_in_flight(part)) { 1053 __part_stat_add(cpu, part, time_in_queue, 1054 part_in_flight(part) * (now - part->stamp)); 1055 __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); 1056 } 1057 part->stamp = now; 1058 } 1059 1060 /** 1061 * part_round_stats() - Round off the performance stats on a struct disk_stats. 1062 * @cpu: cpu number for stats access 1063 * @part: target partition 1064 * 1065 * The average IO queue length and utilisation statistics are maintained 1066 * by observing the current state of the queue length and the amount of 1067 * time it has been in this state for. 1068 * 1069 * Normally, that accounting is done on IO completion, but that can result 1070 * in more than a second's worth of IO being accounted for within any one 1071 * second, leading to >100% utilisation. To deal with that, we call this 1072 * function to do a round-off before returning the results when reading 1073 * /proc/diskstats. This accounts immediately for all queue usage up to 1074 * the current jiffies and restarts the counters again. 1075 */ 1076 void part_round_stats(int cpu, struct hd_struct *part) 1077 { 1078 unsigned long now = jiffies; 1079 1080 if (part->partno) 1081 part_round_stats_single(cpu, &part_to_disk(part)->part0, now); 1082 part_round_stats_single(cpu, part, now); 1083 } 1084 EXPORT_SYMBOL_GPL(part_round_stats); 1085 1086 /* 1087 * queue lock must be held 1088 */ 1089 void __blk_put_request(struct request_queue *q, struct request *req) 1090 { 1091 if (unlikely(!q)) 1092 return; 1093 if (unlikely(--req->ref_count)) 1094 return; 1095 1096 elv_completed_request(q, req); 1097 1098 /* this is a bio leak */ 1099 WARN_ON(req->bio != NULL); 1100 1101 /* 1102 * Request may not have originated from ll_rw_blk. if not, 1103 * it didn't come out of our reserved rq pools 1104 */ 1105 if (req->cmd_flags & REQ_ALLOCED) { 1106 int is_sync = rq_is_sync(req) != 0; 1107 int priv = req->cmd_flags & REQ_ELVPRIV; 1108 1109 BUG_ON(!list_empty(&req->queuelist)); 1110 BUG_ON(!hlist_unhashed(&req->hash)); 1111 1112 blk_free_request(q, req); 1113 freed_request(q, is_sync, priv); 1114 } 1115 } 1116 EXPORT_SYMBOL_GPL(__blk_put_request); 1117 1118 void blk_put_request(struct request *req) 1119 { 1120 unsigned long flags; 1121 struct request_queue *q = req->q; 1122 1123 spin_lock_irqsave(q->queue_lock, flags); 1124 __blk_put_request(q, req); 1125 spin_unlock_irqrestore(q->queue_lock, flags); 1126 } 1127 EXPORT_SYMBOL(blk_put_request); 1128 1129 /** 1130 * blk_add_request_payload - add a payload to a request 1131 * @rq: request to update 1132 * @page: page backing the payload 1133 * @len: length of the payload. 1134 * 1135 * This allows to later add a payload to an already submitted request by 1136 * a block driver. The driver needs to take care of freeing the payload 1137 * itself. 1138 * 1139 * Note that this is a quite horrible hack and nothing but handling of 1140 * discard requests should ever use it. 1141 */ 1142 void blk_add_request_payload(struct request *rq, struct page *page, 1143 unsigned int len) 1144 { 1145 struct bio *bio = rq->bio; 1146 1147 bio->bi_io_vec->bv_page = page; 1148 bio->bi_io_vec->bv_offset = 0; 1149 bio->bi_io_vec->bv_len = len; 1150 1151 bio->bi_size = len; 1152 bio->bi_vcnt = 1; 1153 bio->bi_phys_segments = 1; 1154 1155 rq->__data_len = rq->resid_len = len; 1156 rq->nr_phys_segments = 1; 1157 rq->buffer = bio_data(bio); 1158 } 1159 EXPORT_SYMBOL_GPL(blk_add_request_payload); 1160 1161 void init_request_from_bio(struct request *req, struct bio *bio) 1162 { 1163 req->cpu = bio->bi_comp_cpu; 1164 req->cmd_type = REQ_TYPE_FS; 1165 1166 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; 1167 if (bio->bi_rw & REQ_RAHEAD) 1168 req->cmd_flags |= REQ_FAILFAST_MASK; 1169 1170 req->errors = 0; 1171 req->__sector = bio->bi_sector; 1172 req->ioprio = bio_prio(bio); 1173 blk_rq_bio_prep(req->q, req, bio); 1174 } 1175 1176 /* 1177 * Only disabling plugging for non-rotational devices if it does tagging 1178 * as well, otherwise we do need the proper merging 1179 */ 1180 static inline bool queue_should_plug(struct request_queue *q) 1181 { 1182 return !(blk_queue_nonrot(q) && blk_queue_tagged(q)); 1183 } 1184 1185 static int __make_request(struct request_queue *q, struct bio *bio) 1186 { 1187 struct request *req; 1188 int el_ret; 1189 unsigned int bytes = bio->bi_size; 1190 const unsigned short prio = bio_prio(bio); 1191 const bool sync = !!(bio->bi_rw & REQ_SYNC); 1192 const bool unplug = !!(bio->bi_rw & REQ_UNPLUG); 1193 const unsigned long ff = bio->bi_rw & REQ_FAILFAST_MASK; 1194 int where = ELEVATOR_INSERT_SORT; 1195 int rw_flags; 1196 1197 /* REQ_HARDBARRIER is no more */ 1198 if (WARN_ONCE(bio->bi_rw & REQ_HARDBARRIER, 1199 "block: HARDBARRIER is deprecated, use FLUSH/FUA instead\n")) { 1200 bio_endio(bio, -EOPNOTSUPP); 1201 return 0; 1202 } 1203 1204 /* 1205 * low level driver can indicate that it wants pages above a 1206 * certain limit bounced to low memory (ie for highmem, or even 1207 * ISA dma in theory) 1208 */ 1209 blk_queue_bounce(q, &bio); 1210 1211 spin_lock_irq(q->queue_lock); 1212 1213 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { 1214 where = ELEVATOR_INSERT_FRONT; 1215 goto get_rq; 1216 } 1217 1218 if (elv_queue_empty(q)) 1219 goto get_rq; 1220 1221 el_ret = elv_merge(q, &req, bio); 1222 switch (el_ret) { 1223 case ELEVATOR_BACK_MERGE: 1224 BUG_ON(!rq_mergeable(req)); 1225 1226 if (!ll_back_merge_fn(q, req, bio)) 1227 break; 1228 1229 trace_block_bio_backmerge(q, bio); 1230 1231 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 1232 blk_rq_set_mixed_merge(req); 1233 1234 req->biotail->bi_next = bio; 1235 req->biotail = bio; 1236 req->__data_len += bytes; 1237 req->ioprio = ioprio_best(req->ioprio, prio); 1238 if (!blk_rq_cpu_valid(req)) 1239 req->cpu = bio->bi_comp_cpu; 1240 drive_stat_acct(req, 0); 1241 elv_bio_merged(q, req, bio); 1242 if (!attempt_back_merge(q, req)) 1243 elv_merged_request(q, req, el_ret); 1244 goto out; 1245 1246 case ELEVATOR_FRONT_MERGE: 1247 BUG_ON(!rq_mergeable(req)); 1248 1249 if (!ll_front_merge_fn(q, req, bio)) 1250 break; 1251 1252 trace_block_bio_frontmerge(q, bio); 1253 1254 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) { 1255 blk_rq_set_mixed_merge(req); 1256 req->cmd_flags &= ~REQ_FAILFAST_MASK; 1257 req->cmd_flags |= ff; 1258 } 1259 1260 bio->bi_next = req->bio; 1261 req->bio = bio; 1262 1263 /* 1264 * may not be valid. if the low level driver said 1265 * it didn't need a bounce buffer then it better 1266 * not touch req->buffer either... 1267 */ 1268 req->buffer = bio_data(bio); 1269 req->__sector = bio->bi_sector; 1270 req->__data_len += bytes; 1271 req->ioprio = ioprio_best(req->ioprio, prio); 1272 if (!blk_rq_cpu_valid(req)) 1273 req->cpu = bio->bi_comp_cpu; 1274 drive_stat_acct(req, 0); 1275 elv_bio_merged(q, req, bio); 1276 if (!attempt_front_merge(q, req)) 1277 elv_merged_request(q, req, el_ret); 1278 goto out; 1279 1280 /* ELV_NO_MERGE: elevator says don't/can't merge. */ 1281 default: 1282 ; 1283 } 1284 1285 get_rq: 1286 /* 1287 * This sync check and mask will be re-done in init_request_from_bio(), 1288 * but we need to set it earlier to expose the sync flag to the 1289 * rq allocator and io schedulers. 1290 */ 1291 rw_flags = bio_data_dir(bio); 1292 if (sync) 1293 rw_flags |= REQ_SYNC; 1294 1295 /* 1296 * Grab a free request. This is might sleep but can not fail. 1297 * Returns with the queue unlocked. 1298 */ 1299 req = get_request_wait(q, rw_flags, bio); 1300 1301 /* 1302 * After dropping the lock and possibly sleeping here, our request 1303 * may now be mergeable after it had proven unmergeable (above). 1304 * We don't worry about that case for efficiency. It won't happen 1305 * often, and the elevators are able to handle it. 1306 */ 1307 init_request_from_bio(req, bio); 1308 1309 spin_lock_irq(q->queue_lock); 1310 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || 1311 bio_flagged(bio, BIO_CPU_AFFINE)) 1312 req->cpu = blk_cpu_to_group(smp_processor_id()); 1313 if (queue_should_plug(q) && elv_queue_empty(q)) 1314 blk_plug_device(q); 1315 1316 /* insert the request into the elevator */ 1317 drive_stat_acct(req, 1); 1318 __elv_add_request(q, req, where, 0); 1319 out: 1320 if (unplug || !queue_should_plug(q)) 1321 __generic_unplug_device(q); 1322 spin_unlock_irq(q->queue_lock); 1323 return 0; 1324 } 1325 1326 /* 1327 * If bio->bi_dev is a partition, remap the location 1328 */ 1329 static inline void blk_partition_remap(struct bio *bio) 1330 { 1331 struct block_device *bdev = bio->bi_bdev; 1332 1333 if (bio_sectors(bio) && bdev != bdev->bd_contains) { 1334 struct hd_struct *p = bdev->bd_part; 1335 1336 bio->bi_sector += p->start_sect; 1337 bio->bi_bdev = bdev->bd_contains; 1338 1339 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, 1340 bdev->bd_dev, 1341 bio->bi_sector - p->start_sect); 1342 } 1343 } 1344 1345 static void handle_bad_sector(struct bio *bio) 1346 { 1347 char b[BDEVNAME_SIZE]; 1348 1349 printk(KERN_INFO "attempt to access beyond end of device\n"); 1350 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", 1351 bdevname(bio->bi_bdev, b), 1352 bio->bi_rw, 1353 (unsigned long long)bio->bi_sector + bio_sectors(bio), 1354 (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); 1355 1356 set_bit(BIO_EOF, &bio->bi_flags); 1357 } 1358 1359 #ifdef CONFIG_FAIL_MAKE_REQUEST 1360 1361 static DECLARE_FAULT_ATTR(fail_make_request); 1362 1363 static int __init setup_fail_make_request(char *str) 1364 { 1365 return setup_fault_attr(&fail_make_request, str); 1366 } 1367 __setup("fail_make_request=", setup_fail_make_request); 1368 1369 static int should_fail_request(struct bio *bio) 1370 { 1371 struct hd_struct *part = bio->bi_bdev->bd_part; 1372 1373 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) 1374 return should_fail(&fail_make_request, bio->bi_size); 1375 1376 return 0; 1377 } 1378 1379 static int __init fail_make_request_debugfs(void) 1380 { 1381 return init_fault_attr_dentries(&fail_make_request, 1382 "fail_make_request"); 1383 } 1384 1385 late_initcall(fail_make_request_debugfs); 1386 1387 #else /* CONFIG_FAIL_MAKE_REQUEST */ 1388 1389 static inline int should_fail_request(struct bio *bio) 1390 { 1391 return 0; 1392 } 1393 1394 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 1395 1396 /* 1397 * Check whether this bio extends beyond the end of the device. 1398 */ 1399 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) 1400 { 1401 sector_t maxsector; 1402 1403 if (!nr_sectors) 1404 return 0; 1405 1406 /* Test device or partition size, when known. */ 1407 maxsector = bio->bi_bdev->bd_inode->i_size >> 9; 1408 if (maxsector) { 1409 sector_t sector = bio->bi_sector; 1410 1411 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { 1412 /* 1413 * This may well happen - the kernel calls bread() 1414 * without checking the size of the device, e.g., when 1415 * mounting a device. 1416 */ 1417 handle_bad_sector(bio); 1418 return 1; 1419 } 1420 } 1421 1422 return 0; 1423 } 1424 1425 /** 1426 * generic_make_request - hand a buffer to its device driver for I/O 1427 * @bio: The bio describing the location in memory and on the device. 1428 * 1429 * generic_make_request() is used to make I/O requests of block 1430 * devices. It is passed a &struct bio, which describes the I/O that needs 1431 * to be done. 1432 * 1433 * generic_make_request() does not return any status. The 1434 * success/failure status of the request, along with notification of 1435 * completion, is delivered asynchronously through the bio->bi_end_io 1436 * function described (one day) else where. 1437 * 1438 * The caller of generic_make_request must make sure that bi_io_vec 1439 * are set to describe the memory buffer, and that bi_dev and bi_sector are 1440 * set to describe the device address, and the 1441 * bi_end_io and optionally bi_private are set to describe how 1442 * completion notification should be signaled. 1443 * 1444 * generic_make_request and the drivers it calls may use bi_next if this 1445 * bio happens to be merged with someone else, and may change bi_dev and 1446 * bi_sector for remaps as it sees fit. So the values of these fields 1447 * should NOT be depended on after the call to generic_make_request. 1448 */ 1449 static inline void __generic_make_request(struct bio *bio) 1450 { 1451 struct request_queue *q; 1452 sector_t old_sector; 1453 int ret, nr_sectors = bio_sectors(bio); 1454 dev_t old_dev; 1455 int err = -EIO; 1456 1457 might_sleep(); 1458 1459 if (bio_check_eod(bio, nr_sectors)) 1460 goto end_io; 1461 1462 /* 1463 * Resolve the mapping until finished. (drivers are 1464 * still free to implement/resolve their own stacking 1465 * by explicitly returning 0) 1466 * 1467 * NOTE: we don't repeat the blk_size check for each new device. 1468 * Stacking drivers are expected to know what they are doing. 1469 */ 1470 old_sector = -1; 1471 old_dev = 0; 1472 do { 1473 char b[BDEVNAME_SIZE]; 1474 1475 q = bdev_get_queue(bio->bi_bdev); 1476 if (unlikely(!q)) { 1477 printk(KERN_ERR 1478 "generic_make_request: Trying to access " 1479 "nonexistent block-device %s (%Lu)\n", 1480 bdevname(bio->bi_bdev, b), 1481 (long long) bio->bi_sector); 1482 goto end_io; 1483 } 1484 1485 if (unlikely(!(bio->bi_rw & REQ_DISCARD) && 1486 nr_sectors > queue_max_hw_sectors(q))) { 1487 printk(KERN_ERR "bio too big device %s (%u > %u)\n", 1488 bdevname(bio->bi_bdev, b), 1489 bio_sectors(bio), 1490 queue_max_hw_sectors(q)); 1491 goto end_io; 1492 } 1493 1494 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) 1495 goto end_io; 1496 1497 if (should_fail_request(bio)) 1498 goto end_io; 1499 1500 /* 1501 * If this device has partitions, remap block n 1502 * of partition p to block n+start(p) of the disk. 1503 */ 1504 blk_partition_remap(bio); 1505 1506 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) 1507 goto end_io; 1508 1509 if (old_sector != -1) 1510 trace_block_remap(q, bio, old_dev, old_sector); 1511 1512 old_sector = bio->bi_sector; 1513 old_dev = bio->bi_bdev->bd_dev; 1514 1515 if (bio_check_eod(bio, nr_sectors)) 1516 goto end_io; 1517 1518 /* 1519 * Filter flush bio's early so that make_request based 1520 * drivers without flush support don't have to worry 1521 * about them. 1522 */ 1523 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { 1524 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); 1525 if (!nr_sectors) { 1526 err = 0; 1527 goto end_io; 1528 } 1529 } 1530 1531 if ((bio->bi_rw & REQ_DISCARD) && 1532 (!blk_queue_discard(q) || 1533 ((bio->bi_rw & REQ_SECURE) && 1534 !blk_queue_secdiscard(q)))) { 1535 err = -EOPNOTSUPP; 1536 goto end_io; 1537 } 1538 1539 blk_throtl_bio(q, &bio); 1540 1541 /* 1542 * If bio = NULL, bio has been throttled and will be submitted 1543 * later. 1544 */ 1545 if (!bio) 1546 break; 1547 1548 trace_block_bio_queue(q, bio); 1549 1550 ret = q->make_request_fn(q, bio); 1551 } while (ret); 1552 1553 return; 1554 1555 end_io: 1556 bio_endio(bio, err); 1557 } 1558 1559 /* 1560 * We only want one ->make_request_fn to be active at a time, 1561 * else stack usage with stacked devices could be a problem. 1562 * So use current->bio_list to keep a list of requests 1563 * submited by a make_request_fn function. 1564 * current->bio_list is also used as a flag to say if 1565 * generic_make_request is currently active in this task or not. 1566 * If it is NULL, then no make_request is active. If it is non-NULL, 1567 * then a make_request is active, and new requests should be added 1568 * at the tail 1569 */ 1570 void generic_make_request(struct bio *bio) 1571 { 1572 struct bio_list bio_list_on_stack; 1573 1574 if (current->bio_list) { 1575 /* make_request is active */ 1576 bio_list_add(current->bio_list, bio); 1577 return; 1578 } 1579 /* following loop may be a bit non-obvious, and so deserves some 1580 * explanation. 1581 * Before entering the loop, bio->bi_next is NULL (as all callers 1582 * ensure that) so we have a list with a single bio. 1583 * We pretend that we have just taken it off a longer list, so 1584 * we assign bio_list to a pointer to the bio_list_on_stack, 1585 * thus initialising the bio_list of new bios to be 1586 * added. __generic_make_request may indeed add some more bios 1587 * through a recursive call to generic_make_request. If it 1588 * did, we find a non-NULL value in bio_list and re-enter the loop 1589 * from the top. In this case we really did just take the bio 1590 * of the top of the list (no pretending) and so remove it from 1591 * bio_list, and call into __generic_make_request again. 1592 * 1593 * The loop was structured like this to make only one call to 1594 * __generic_make_request (which is important as it is large and 1595 * inlined) and to keep the structure simple. 1596 */ 1597 BUG_ON(bio->bi_next); 1598 bio_list_init(&bio_list_on_stack); 1599 current->bio_list = &bio_list_on_stack; 1600 do { 1601 __generic_make_request(bio); 1602 bio = bio_list_pop(current->bio_list); 1603 } while (bio); 1604 current->bio_list = NULL; /* deactivate */ 1605 } 1606 EXPORT_SYMBOL(generic_make_request); 1607 1608 /** 1609 * submit_bio - submit a bio to the block device layer for I/O 1610 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) 1611 * @bio: The &struct bio which describes the I/O 1612 * 1613 * submit_bio() is very similar in purpose to generic_make_request(), and 1614 * uses that function to do most of the work. Both are fairly rough 1615 * interfaces; @bio must be presetup and ready for I/O. 1616 * 1617 */ 1618 void submit_bio(int rw, struct bio *bio) 1619 { 1620 int count = bio_sectors(bio); 1621 1622 bio->bi_rw |= rw; 1623 1624 /* 1625 * If it's a regular read/write or a barrier with data attached, 1626 * go through the normal accounting stuff before submission. 1627 */ 1628 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) { 1629 if (rw & WRITE) { 1630 count_vm_events(PGPGOUT, count); 1631 } else { 1632 task_io_account_read(bio->bi_size); 1633 count_vm_events(PGPGIN, count); 1634 } 1635 1636 if (unlikely(block_dump)) { 1637 char b[BDEVNAME_SIZE]; 1638 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", 1639 current->comm, task_pid_nr(current), 1640 (rw & WRITE) ? "WRITE" : "READ", 1641 (unsigned long long)bio->bi_sector, 1642 bdevname(bio->bi_bdev, b), 1643 count); 1644 } 1645 } 1646 1647 generic_make_request(bio); 1648 } 1649 EXPORT_SYMBOL(submit_bio); 1650 1651 /** 1652 * blk_rq_check_limits - Helper function to check a request for the queue limit 1653 * @q: the queue 1654 * @rq: the request being checked 1655 * 1656 * Description: 1657 * @rq may have been made based on weaker limitations of upper-level queues 1658 * in request stacking drivers, and it may violate the limitation of @q. 1659 * Since the block layer and the underlying device driver trust @rq 1660 * after it is inserted to @q, it should be checked against @q before 1661 * the insertion using this generic function. 1662 * 1663 * This function should also be useful for request stacking drivers 1664 * in some cases below, so export this function. 1665 * Request stacking drivers like request-based dm may change the queue 1666 * limits while requests are in the queue (e.g. dm's table swapping). 1667 * Such request stacking drivers should check those requests agaist 1668 * the new queue limits again when they dispatch those requests, 1669 * although such checkings are also done against the old queue limits 1670 * when submitting requests. 1671 */ 1672 int blk_rq_check_limits(struct request_queue *q, struct request *rq) 1673 { 1674 if (rq->cmd_flags & REQ_DISCARD) 1675 return 0; 1676 1677 if (blk_rq_sectors(rq) > queue_max_sectors(q) || 1678 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { 1679 printk(KERN_ERR "%s: over max size limit.\n", __func__); 1680 return -EIO; 1681 } 1682 1683 /* 1684 * queue's settings related to segment counting like q->bounce_pfn 1685 * may differ from that of other stacking queues. 1686 * Recalculate it to check the request correctly on this queue's 1687 * limitation. 1688 */ 1689 blk_recalc_rq_segments(rq); 1690 if (rq->nr_phys_segments > queue_max_segments(q)) { 1691 printk(KERN_ERR "%s: over max segments limit.\n", __func__); 1692 return -EIO; 1693 } 1694 1695 return 0; 1696 } 1697 EXPORT_SYMBOL_GPL(blk_rq_check_limits); 1698 1699 /** 1700 * blk_insert_cloned_request - Helper for stacking drivers to submit a request 1701 * @q: the queue to submit the request 1702 * @rq: the request being queued 1703 */ 1704 int blk_insert_cloned_request(struct request_queue *q, struct request *rq) 1705 { 1706 unsigned long flags; 1707 1708 if (blk_rq_check_limits(q, rq)) 1709 return -EIO; 1710 1711 #ifdef CONFIG_FAIL_MAKE_REQUEST 1712 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && 1713 should_fail(&fail_make_request, blk_rq_bytes(rq))) 1714 return -EIO; 1715 #endif 1716 1717 spin_lock_irqsave(q->queue_lock, flags); 1718 1719 /* 1720 * Submitting request must be dequeued before calling this function 1721 * because it will be linked to another request_queue 1722 */ 1723 BUG_ON(blk_queued_rq(rq)); 1724 1725 drive_stat_acct(rq, 1); 1726 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); 1727 1728 spin_unlock_irqrestore(q->queue_lock, flags); 1729 1730 return 0; 1731 } 1732 EXPORT_SYMBOL_GPL(blk_insert_cloned_request); 1733 1734 /** 1735 * blk_rq_err_bytes - determine number of bytes till the next failure boundary 1736 * @rq: request to examine 1737 * 1738 * Description: 1739 * A request could be merge of IOs which require different failure 1740 * handling. This function determines the number of bytes which 1741 * can be failed from the beginning of the request without 1742 * crossing into area which need to be retried further. 1743 * 1744 * Return: 1745 * The number of bytes to fail. 1746 * 1747 * Context: 1748 * queue_lock must be held. 1749 */ 1750 unsigned int blk_rq_err_bytes(const struct request *rq) 1751 { 1752 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; 1753 unsigned int bytes = 0; 1754 struct bio *bio; 1755 1756 if (!(rq->cmd_flags & REQ_MIXED_MERGE)) 1757 return blk_rq_bytes(rq); 1758 1759 /* 1760 * Currently the only 'mixing' which can happen is between 1761 * different fastfail types. We can safely fail portions 1762 * which have all the failfast bits that the first one has - 1763 * the ones which are at least as eager to fail as the first 1764 * one. 1765 */ 1766 for (bio = rq->bio; bio; bio = bio->bi_next) { 1767 if ((bio->bi_rw & ff) != ff) 1768 break; 1769 bytes += bio->bi_size; 1770 } 1771 1772 /* this could lead to infinite loop */ 1773 BUG_ON(blk_rq_bytes(rq) && !bytes); 1774 return bytes; 1775 } 1776 EXPORT_SYMBOL_GPL(blk_rq_err_bytes); 1777 1778 static void blk_account_io_completion(struct request *req, unsigned int bytes) 1779 { 1780 if (blk_do_io_stat(req)) { 1781 const int rw = rq_data_dir(req); 1782 struct hd_struct *part; 1783 int cpu; 1784 1785 cpu = part_stat_lock(); 1786 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req)); 1787 part_stat_add(cpu, part, sectors[rw], bytes >> 9); 1788 part_stat_unlock(); 1789 } 1790 } 1791 1792 static void blk_account_io_done(struct request *req) 1793 { 1794 /* 1795 * Account IO completion. flush_rq isn't accounted as a 1796 * normal IO on queueing nor completion. Accounting the 1797 * containing request is enough. 1798 */ 1799 if (blk_do_io_stat(req) && req != &req->q->flush_rq) { 1800 unsigned long duration = jiffies - req->start_time; 1801 const int rw = rq_data_dir(req); 1802 struct hd_struct *part; 1803 int cpu; 1804 1805 cpu = part_stat_lock(); 1806 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req)); 1807 1808 part_stat_inc(cpu, part, ios[rw]); 1809 part_stat_add(cpu, part, ticks[rw], duration); 1810 part_round_stats(cpu, part); 1811 part_dec_in_flight(part, rw); 1812 1813 part_stat_unlock(); 1814 } 1815 } 1816 1817 /** 1818 * blk_peek_request - peek at the top of a request queue 1819 * @q: request queue to peek at 1820 * 1821 * Description: 1822 * Return the request at the top of @q. The returned request 1823 * should be started using blk_start_request() before LLD starts 1824 * processing it. 1825 * 1826 * Return: 1827 * Pointer to the request at the top of @q if available. Null 1828 * otherwise. 1829 * 1830 * Context: 1831 * queue_lock must be held. 1832 */ 1833 struct request *blk_peek_request(struct request_queue *q) 1834 { 1835 struct request *rq; 1836 int ret; 1837 1838 while ((rq = __elv_next_request(q)) != NULL) { 1839 if (!(rq->cmd_flags & REQ_STARTED)) { 1840 /* 1841 * This is the first time the device driver 1842 * sees this request (possibly after 1843 * requeueing). Notify IO scheduler. 1844 */ 1845 if (rq->cmd_flags & REQ_SORTED) 1846 elv_activate_rq(q, rq); 1847 1848 /* 1849 * just mark as started even if we don't start 1850 * it, a request that has been delayed should 1851 * not be passed by new incoming requests 1852 */ 1853 rq->cmd_flags |= REQ_STARTED; 1854 trace_block_rq_issue(q, rq); 1855 } 1856 1857 if (!q->boundary_rq || q->boundary_rq == rq) { 1858 q->end_sector = rq_end_sector(rq); 1859 q->boundary_rq = NULL; 1860 } 1861 1862 if (rq->cmd_flags & REQ_DONTPREP) 1863 break; 1864 1865 if (q->dma_drain_size && blk_rq_bytes(rq)) { 1866 /* 1867 * make sure space for the drain appears we 1868 * know we can do this because max_hw_segments 1869 * has been adjusted to be one fewer than the 1870 * device can handle 1871 */ 1872 rq->nr_phys_segments++; 1873 } 1874 1875 if (!q->prep_rq_fn) 1876 break; 1877 1878 ret = q->prep_rq_fn(q, rq); 1879 if (ret == BLKPREP_OK) { 1880 break; 1881 } else if (ret == BLKPREP_DEFER) { 1882 /* 1883 * the request may have been (partially) prepped. 1884 * we need to keep this request in the front to 1885 * avoid resource deadlock. REQ_STARTED will 1886 * prevent other fs requests from passing this one. 1887 */ 1888 if (q->dma_drain_size && blk_rq_bytes(rq) && 1889 !(rq->cmd_flags & REQ_DONTPREP)) { 1890 /* 1891 * remove the space for the drain we added 1892 * so that we don't add it again 1893 */ 1894 --rq->nr_phys_segments; 1895 } 1896 1897 rq = NULL; 1898 break; 1899 } else if (ret == BLKPREP_KILL) { 1900 rq->cmd_flags |= REQ_QUIET; 1901 /* 1902 * Mark this request as started so we don't trigger 1903 * any debug logic in the end I/O path. 1904 */ 1905 blk_start_request(rq); 1906 __blk_end_request_all(rq, -EIO); 1907 } else { 1908 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); 1909 break; 1910 } 1911 } 1912 1913 return rq; 1914 } 1915 EXPORT_SYMBOL(blk_peek_request); 1916 1917 void blk_dequeue_request(struct request *rq) 1918 { 1919 struct request_queue *q = rq->q; 1920 1921 BUG_ON(list_empty(&rq->queuelist)); 1922 BUG_ON(ELV_ON_HASH(rq)); 1923 1924 list_del_init(&rq->queuelist); 1925 1926 /* 1927 * the time frame between a request being removed from the lists 1928 * and to it is freed is accounted as io that is in progress at 1929 * the driver side. 1930 */ 1931 if (blk_account_rq(rq)) { 1932 q->in_flight[rq_is_sync(rq)]++; 1933 set_io_start_time_ns(rq); 1934 } 1935 } 1936 1937 /** 1938 * blk_start_request - start request processing on the driver 1939 * @req: request to dequeue 1940 * 1941 * Description: 1942 * Dequeue @req and start timeout timer on it. This hands off the 1943 * request to the driver. 1944 * 1945 * Block internal functions which don't want to start timer should 1946 * call blk_dequeue_request(). 1947 * 1948 * Context: 1949 * queue_lock must be held. 1950 */ 1951 void blk_start_request(struct request *req) 1952 { 1953 blk_dequeue_request(req); 1954 1955 /* 1956 * We are now handing the request to the hardware, initialize 1957 * resid_len to full count and add the timeout handler. 1958 */ 1959 req->resid_len = blk_rq_bytes(req); 1960 if (unlikely(blk_bidi_rq(req))) 1961 req->next_rq->resid_len = blk_rq_bytes(req->next_rq); 1962 1963 blk_add_timer(req); 1964 } 1965 EXPORT_SYMBOL(blk_start_request); 1966 1967 /** 1968 * blk_fetch_request - fetch a request from a request queue 1969 * @q: request queue to fetch a request from 1970 * 1971 * Description: 1972 * Return the request at the top of @q. The request is started on 1973 * return and LLD can start processing it immediately. 1974 * 1975 * Return: 1976 * Pointer to the request at the top of @q if available. Null 1977 * otherwise. 1978 * 1979 * Context: 1980 * queue_lock must be held. 1981 */ 1982 struct request *blk_fetch_request(struct request_queue *q) 1983 { 1984 struct request *rq; 1985 1986 rq = blk_peek_request(q); 1987 if (rq) 1988 blk_start_request(rq); 1989 return rq; 1990 } 1991 EXPORT_SYMBOL(blk_fetch_request); 1992 1993 /** 1994 * blk_update_request - Special helper function for request stacking drivers 1995 * @req: the request being processed 1996 * @error: %0 for success, < %0 for error 1997 * @nr_bytes: number of bytes to complete @req 1998 * 1999 * Description: 2000 * Ends I/O on a number of bytes attached to @req, but doesn't complete 2001 * the request structure even if @req doesn't have leftover. 2002 * If @req has leftover, sets it up for the next range of segments. 2003 * 2004 * This special helper function is only for request stacking drivers 2005 * (e.g. request-based dm) so that they can handle partial completion. 2006 * Actual device drivers should use blk_end_request instead. 2007 * 2008 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees 2009 * %false return from this function. 2010 * 2011 * Return: 2012 * %false - this request doesn't have any more data 2013 * %true - this request has more data 2014 **/ 2015 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) 2016 { 2017 int total_bytes, bio_nbytes, next_idx = 0; 2018 struct bio *bio; 2019 2020 if (!req->bio) 2021 return false; 2022 2023 trace_block_rq_complete(req->q, req); 2024 2025 /* 2026 * For fs requests, rq is just carrier of independent bio's 2027 * and each partial completion should be handled separately. 2028 * Reset per-request error on each partial completion. 2029 * 2030 * TODO: tj: This is too subtle. It would be better to let 2031 * low level drivers do what they see fit. 2032 */ 2033 if (req->cmd_type == REQ_TYPE_FS) 2034 req->errors = 0; 2035 2036 if (error && req->cmd_type == REQ_TYPE_FS && 2037 !(req->cmd_flags & REQ_QUIET)) { 2038 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", 2039 req->rq_disk ? req->rq_disk->disk_name : "?", 2040 (unsigned long long)blk_rq_pos(req)); 2041 } 2042 2043 blk_account_io_completion(req, nr_bytes); 2044 2045 total_bytes = bio_nbytes = 0; 2046 while ((bio = req->bio) != NULL) { 2047 int nbytes; 2048 2049 if (nr_bytes >= bio->bi_size) { 2050 req->bio = bio->bi_next; 2051 nbytes = bio->bi_size; 2052 req_bio_endio(req, bio, nbytes, error); 2053 next_idx = 0; 2054 bio_nbytes = 0; 2055 } else { 2056 int idx = bio->bi_idx + next_idx; 2057 2058 if (unlikely(idx >= bio->bi_vcnt)) { 2059 blk_dump_rq_flags(req, "__end_that"); 2060 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", 2061 __func__, idx, bio->bi_vcnt); 2062 break; 2063 } 2064 2065 nbytes = bio_iovec_idx(bio, idx)->bv_len; 2066 BIO_BUG_ON(nbytes > bio->bi_size); 2067 2068 /* 2069 * not a complete bvec done 2070 */ 2071 if (unlikely(nbytes > nr_bytes)) { 2072 bio_nbytes += nr_bytes; 2073 total_bytes += nr_bytes; 2074 break; 2075 } 2076 2077 /* 2078 * advance to the next vector 2079 */ 2080 next_idx++; 2081 bio_nbytes += nbytes; 2082 } 2083 2084 total_bytes += nbytes; 2085 nr_bytes -= nbytes; 2086 2087 bio = req->bio; 2088 if (bio) { 2089 /* 2090 * end more in this run, or just return 'not-done' 2091 */ 2092 if (unlikely(nr_bytes <= 0)) 2093 break; 2094 } 2095 } 2096 2097 /* 2098 * completely done 2099 */ 2100 if (!req->bio) { 2101 /* 2102 * Reset counters so that the request stacking driver 2103 * can find how many bytes remain in the request 2104 * later. 2105 */ 2106 req->__data_len = 0; 2107 return false; 2108 } 2109 2110 /* 2111 * if the request wasn't completed, update state 2112 */ 2113 if (bio_nbytes) { 2114 req_bio_endio(req, bio, bio_nbytes, error); 2115 bio->bi_idx += next_idx; 2116 bio_iovec(bio)->bv_offset += nr_bytes; 2117 bio_iovec(bio)->bv_len -= nr_bytes; 2118 } 2119 2120 req->__data_len -= total_bytes; 2121 req->buffer = bio_data(req->bio); 2122 2123 /* update sector only for requests with clear definition of sector */ 2124 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD)) 2125 req->__sector += total_bytes >> 9; 2126 2127 /* mixed attributes always follow the first bio */ 2128 if (req->cmd_flags & REQ_MIXED_MERGE) { 2129 req->cmd_flags &= ~REQ_FAILFAST_MASK; 2130 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; 2131 } 2132 2133 /* 2134 * If total number of sectors is less than the first segment 2135 * size, something has gone terribly wrong. 2136 */ 2137 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { 2138 printk(KERN_ERR "blk: request botched\n"); 2139 req->__data_len = blk_rq_cur_bytes(req); 2140 } 2141 2142 /* recalculate the number of segments */ 2143 blk_recalc_rq_segments(req); 2144 2145 return true; 2146 } 2147 EXPORT_SYMBOL_GPL(blk_update_request); 2148 2149 static bool blk_update_bidi_request(struct request *rq, int error, 2150 unsigned int nr_bytes, 2151 unsigned int bidi_bytes) 2152 { 2153 if (blk_update_request(rq, error, nr_bytes)) 2154 return true; 2155 2156 /* Bidi request must be completed as a whole */ 2157 if (unlikely(blk_bidi_rq(rq)) && 2158 blk_update_request(rq->next_rq, error, bidi_bytes)) 2159 return true; 2160 2161 if (blk_queue_add_random(rq->q)) 2162 add_disk_randomness(rq->rq_disk); 2163 2164 return false; 2165 } 2166 2167 /** 2168 * blk_unprep_request - unprepare a request 2169 * @req: the request 2170 * 2171 * This function makes a request ready for complete resubmission (or 2172 * completion). It happens only after all error handling is complete, 2173 * so represents the appropriate moment to deallocate any resources 2174 * that were allocated to the request in the prep_rq_fn. The queue 2175 * lock is held when calling this. 2176 */ 2177 void blk_unprep_request(struct request *req) 2178 { 2179 struct request_queue *q = req->q; 2180 2181 req->cmd_flags &= ~REQ_DONTPREP; 2182 if (q->unprep_rq_fn) 2183 q->unprep_rq_fn(q, req); 2184 } 2185 EXPORT_SYMBOL_GPL(blk_unprep_request); 2186 2187 /* 2188 * queue lock must be held 2189 */ 2190 static void blk_finish_request(struct request *req, int error) 2191 { 2192 if (blk_rq_tagged(req)) 2193 blk_queue_end_tag(req->q, req); 2194 2195 BUG_ON(blk_queued_rq(req)); 2196 2197 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) 2198 laptop_io_completion(&req->q->backing_dev_info); 2199 2200 blk_delete_timer(req); 2201 2202 if (req->cmd_flags & REQ_DONTPREP) 2203 blk_unprep_request(req); 2204 2205 2206 blk_account_io_done(req); 2207 2208 if (req->end_io) 2209 req->end_io(req, error); 2210 else { 2211 if (blk_bidi_rq(req)) 2212 __blk_put_request(req->next_rq->q, req->next_rq); 2213 2214 __blk_put_request(req->q, req); 2215 } 2216 } 2217 2218 /** 2219 * blk_end_bidi_request - Complete a bidi request 2220 * @rq: the request to complete 2221 * @error: %0 for success, < %0 for error 2222 * @nr_bytes: number of bytes to complete @rq 2223 * @bidi_bytes: number of bytes to complete @rq->next_rq 2224 * 2225 * Description: 2226 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. 2227 * Drivers that supports bidi can safely call this member for any 2228 * type of request, bidi or uni. In the later case @bidi_bytes is 2229 * just ignored. 2230 * 2231 * Return: 2232 * %false - we are done with this request 2233 * %true - still buffers pending for this request 2234 **/ 2235 static bool blk_end_bidi_request(struct request *rq, int error, 2236 unsigned int nr_bytes, unsigned int bidi_bytes) 2237 { 2238 struct request_queue *q = rq->q; 2239 unsigned long flags; 2240 2241 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) 2242 return true; 2243 2244 spin_lock_irqsave(q->queue_lock, flags); 2245 blk_finish_request(rq, error); 2246 spin_unlock_irqrestore(q->queue_lock, flags); 2247 2248 return false; 2249 } 2250 2251 /** 2252 * __blk_end_bidi_request - Complete a bidi request with queue lock held 2253 * @rq: the request to complete 2254 * @error: %0 for success, < %0 for error 2255 * @nr_bytes: number of bytes to complete @rq 2256 * @bidi_bytes: number of bytes to complete @rq->next_rq 2257 * 2258 * Description: 2259 * Identical to blk_end_bidi_request() except that queue lock is 2260 * assumed to be locked on entry and remains so on return. 2261 * 2262 * Return: 2263 * %false - we are done with this request 2264 * %true - still buffers pending for this request 2265 **/ 2266 static bool __blk_end_bidi_request(struct request *rq, int error, 2267 unsigned int nr_bytes, unsigned int bidi_bytes) 2268 { 2269 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) 2270 return true; 2271 2272 blk_finish_request(rq, error); 2273 2274 return false; 2275 } 2276 2277 /** 2278 * blk_end_request - Helper function for drivers to complete the request. 2279 * @rq: the request being processed 2280 * @error: %0 for success, < %0 for error 2281 * @nr_bytes: number of bytes to complete 2282 * 2283 * Description: 2284 * Ends I/O on a number of bytes attached to @rq. 2285 * If @rq has leftover, sets it up for the next range of segments. 2286 * 2287 * Return: 2288 * %false - we are done with this request 2289 * %true - still buffers pending for this request 2290 **/ 2291 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 2292 { 2293 return blk_end_bidi_request(rq, error, nr_bytes, 0); 2294 } 2295 EXPORT_SYMBOL(blk_end_request); 2296 2297 /** 2298 * blk_end_request_all - Helper function for drives to finish the request. 2299 * @rq: the request to finish 2300 * @error: %0 for success, < %0 for error 2301 * 2302 * Description: 2303 * Completely finish @rq. 2304 */ 2305 void blk_end_request_all(struct request *rq, int error) 2306 { 2307 bool pending; 2308 unsigned int bidi_bytes = 0; 2309 2310 if (unlikely(blk_bidi_rq(rq))) 2311 bidi_bytes = blk_rq_bytes(rq->next_rq); 2312 2313 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); 2314 BUG_ON(pending); 2315 } 2316 EXPORT_SYMBOL(blk_end_request_all); 2317 2318 /** 2319 * blk_end_request_cur - Helper function to finish the current request chunk. 2320 * @rq: the request to finish the current chunk for 2321 * @error: %0 for success, < %0 for error 2322 * 2323 * Description: 2324 * Complete the current consecutively mapped chunk from @rq. 2325 * 2326 * Return: 2327 * %false - we are done with this request 2328 * %true - still buffers pending for this request 2329 */ 2330 bool blk_end_request_cur(struct request *rq, int error) 2331 { 2332 return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); 2333 } 2334 EXPORT_SYMBOL(blk_end_request_cur); 2335 2336 /** 2337 * blk_end_request_err - Finish a request till the next failure boundary. 2338 * @rq: the request to finish till the next failure boundary for 2339 * @error: must be negative errno 2340 * 2341 * Description: 2342 * Complete @rq till the next failure boundary. 2343 * 2344 * Return: 2345 * %false - we are done with this request 2346 * %true - still buffers pending for this request 2347 */ 2348 bool blk_end_request_err(struct request *rq, int error) 2349 { 2350 WARN_ON(error >= 0); 2351 return blk_end_request(rq, error, blk_rq_err_bytes(rq)); 2352 } 2353 EXPORT_SYMBOL_GPL(blk_end_request_err); 2354 2355 /** 2356 * __blk_end_request - Helper function for drivers to complete the request. 2357 * @rq: the request being processed 2358 * @error: %0 for success, < %0 for error 2359 * @nr_bytes: number of bytes to complete 2360 * 2361 * Description: 2362 * Must be called with queue lock held unlike blk_end_request(). 2363 * 2364 * Return: 2365 * %false - we are done with this request 2366 * %true - still buffers pending for this request 2367 **/ 2368 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 2369 { 2370 return __blk_end_bidi_request(rq, error, nr_bytes, 0); 2371 } 2372 EXPORT_SYMBOL(__blk_end_request); 2373 2374 /** 2375 * __blk_end_request_all - Helper function for drives to finish the request. 2376 * @rq: the request to finish 2377 * @error: %0 for success, < %0 for error 2378 * 2379 * Description: 2380 * Completely finish @rq. Must be called with queue lock held. 2381 */ 2382 void __blk_end_request_all(struct request *rq, int error) 2383 { 2384 bool pending; 2385 unsigned int bidi_bytes = 0; 2386 2387 if (unlikely(blk_bidi_rq(rq))) 2388 bidi_bytes = blk_rq_bytes(rq->next_rq); 2389 2390 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); 2391 BUG_ON(pending); 2392 } 2393 EXPORT_SYMBOL(__blk_end_request_all); 2394 2395 /** 2396 * __blk_end_request_cur - Helper function to finish the current request chunk. 2397 * @rq: the request to finish the current chunk for 2398 * @error: %0 for success, < %0 for error 2399 * 2400 * Description: 2401 * Complete the current consecutively mapped chunk from @rq. Must 2402 * be called with queue lock held. 2403 * 2404 * Return: 2405 * %false - we are done with this request 2406 * %true - still buffers pending for this request 2407 */ 2408 bool __blk_end_request_cur(struct request *rq, int error) 2409 { 2410 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); 2411 } 2412 EXPORT_SYMBOL(__blk_end_request_cur); 2413 2414 /** 2415 * __blk_end_request_err - Finish a request till the next failure boundary. 2416 * @rq: the request to finish till the next failure boundary for 2417 * @error: must be negative errno 2418 * 2419 * Description: 2420 * Complete @rq till the next failure boundary. Must be called 2421 * with queue lock held. 2422 * 2423 * Return: 2424 * %false - we are done with this request 2425 * %true - still buffers pending for this request 2426 */ 2427 bool __blk_end_request_err(struct request *rq, int error) 2428 { 2429 WARN_ON(error >= 0); 2430 return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); 2431 } 2432 EXPORT_SYMBOL_GPL(__blk_end_request_err); 2433 2434 void blk_rq_bio_prep(struct request_queue *q, struct request *rq, 2435 struct bio *bio) 2436 { 2437 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ 2438 rq->cmd_flags |= bio->bi_rw & REQ_WRITE; 2439 2440 if (bio_has_data(bio)) { 2441 rq->nr_phys_segments = bio_phys_segments(q, bio); 2442 rq->buffer = bio_data(bio); 2443 } 2444 rq->__data_len = bio->bi_size; 2445 rq->bio = rq->biotail = bio; 2446 2447 if (bio->bi_bdev) 2448 rq->rq_disk = bio->bi_bdev->bd_disk; 2449 } 2450 2451 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 2452 /** 2453 * rq_flush_dcache_pages - Helper function to flush all pages in a request 2454 * @rq: the request to be flushed 2455 * 2456 * Description: 2457 * Flush all pages in @rq. 2458 */ 2459 void rq_flush_dcache_pages(struct request *rq) 2460 { 2461 struct req_iterator iter; 2462 struct bio_vec *bvec; 2463 2464 rq_for_each_segment(bvec, rq, iter) 2465 flush_dcache_page(bvec->bv_page); 2466 } 2467 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); 2468 #endif 2469 2470 /** 2471 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 2472 * @q : the queue of the device being checked 2473 * 2474 * Description: 2475 * Check if underlying low-level drivers of a device are busy. 2476 * If the drivers want to export their busy state, they must set own 2477 * exporting function using blk_queue_lld_busy() first. 2478 * 2479 * Basically, this function is used only by request stacking drivers 2480 * to stop dispatching requests to underlying devices when underlying 2481 * devices are busy. This behavior helps more I/O merging on the queue 2482 * of the request stacking driver and prevents I/O throughput regression 2483 * on burst I/O load. 2484 * 2485 * Return: 2486 * 0 - Not busy (The request stacking driver should dispatch request) 2487 * 1 - Busy (The request stacking driver should stop dispatching request) 2488 */ 2489 int blk_lld_busy(struct request_queue *q) 2490 { 2491 if (q->lld_busy_fn) 2492 return q->lld_busy_fn(q); 2493 2494 return 0; 2495 } 2496 EXPORT_SYMBOL_GPL(blk_lld_busy); 2497 2498 /** 2499 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request 2500 * @rq: the clone request to be cleaned up 2501 * 2502 * Description: 2503 * Free all bios in @rq for a cloned request. 2504 */ 2505 void blk_rq_unprep_clone(struct request *rq) 2506 { 2507 struct bio *bio; 2508 2509 while ((bio = rq->bio) != NULL) { 2510 rq->bio = bio->bi_next; 2511 2512 bio_put(bio); 2513 } 2514 } 2515 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); 2516 2517 /* 2518 * Copy attributes of the original request to the clone request. 2519 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. 2520 */ 2521 static void __blk_rq_prep_clone(struct request *dst, struct request *src) 2522 { 2523 dst->cpu = src->cpu; 2524 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; 2525 dst->cmd_type = src->cmd_type; 2526 dst->__sector = blk_rq_pos(src); 2527 dst->__data_len = blk_rq_bytes(src); 2528 dst->nr_phys_segments = src->nr_phys_segments; 2529 dst->ioprio = src->ioprio; 2530 dst->extra_len = src->extra_len; 2531 } 2532 2533 /** 2534 * blk_rq_prep_clone - Helper function to setup clone request 2535 * @rq: the request to be setup 2536 * @rq_src: original request to be cloned 2537 * @bs: bio_set that bios for clone are allocated from 2538 * @gfp_mask: memory allocation mask for bio 2539 * @bio_ctr: setup function to be called for each clone bio. 2540 * Returns %0 for success, non %0 for failure. 2541 * @data: private data to be passed to @bio_ctr 2542 * 2543 * Description: 2544 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. 2545 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) 2546 * are not copied, and copying such parts is the caller's responsibility. 2547 * Also, pages which the original bios are pointing to are not copied 2548 * and the cloned bios just point same pages. 2549 * So cloned bios must be completed before original bios, which means 2550 * the caller must complete @rq before @rq_src. 2551 */ 2552 int blk_rq_prep_clone(struct request *rq, struct request *rq_src, 2553 struct bio_set *bs, gfp_t gfp_mask, 2554 int (*bio_ctr)(struct bio *, struct bio *, void *), 2555 void *data) 2556 { 2557 struct bio *bio, *bio_src; 2558 2559 if (!bs) 2560 bs = fs_bio_set; 2561 2562 blk_rq_init(NULL, rq); 2563 2564 __rq_for_each_bio(bio_src, rq_src) { 2565 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); 2566 if (!bio) 2567 goto free_and_out; 2568 2569 __bio_clone(bio, bio_src); 2570 2571 if (bio_integrity(bio_src) && 2572 bio_integrity_clone(bio, bio_src, gfp_mask, bs)) 2573 goto free_and_out; 2574 2575 if (bio_ctr && bio_ctr(bio, bio_src, data)) 2576 goto free_and_out; 2577 2578 if (rq->bio) { 2579 rq->biotail->bi_next = bio; 2580 rq->biotail = bio; 2581 } else 2582 rq->bio = rq->biotail = bio; 2583 } 2584 2585 __blk_rq_prep_clone(rq, rq_src); 2586 2587 return 0; 2588 2589 free_and_out: 2590 if (bio) 2591 bio_free(bio, bs); 2592 blk_rq_unprep_clone(rq); 2593 2594 return -ENOMEM; 2595 } 2596 EXPORT_SYMBOL_GPL(blk_rq_prep_clone); 2597 2598 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) 2599 { 2600 return queue_work(kblockd_workqueue, work); 2601 } 2602 EXPORT_SYMBOL(kblockd_schedule_work); 2603 2604 int kblockd_schedule_delayed_work(struct request_queue *q, 2605 struct delayed_work *dwork, unsigned long delay) 2606 { 2607 return queue_delayed_work(kblockd_workqueue, dwork, delay); 2608 } 2609 EXPORT_SYMBOL(kblockd_schedule_delayed_work); 2610 2611 int __init blk_dev_init(void) 2612 { 2613 BUILD_BUG_ON(__REQ_NR_BITS > 8 * 2614 sizeof(((struct request *)0)->cmd_flags)); 2615 2616 kblockd_workqueue = create_workqueue("kblockd"); 2617 if (!kblockd_workqueue) 2618 panic("Failed to create kblockd\n"); 2619 2620 request_cachep = kmem_cache_create("blkdev_requests", 2621 sizeof(struct request), 0, SLAB_PANIC, NULL); 2622 2623 blk_requestq_cachep = kmem_cache_create("blkdev_queue", 2624 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 2625 2626 return 0; 2627 } 2628