linux/drivers/block/ll_rw_blk.c
<<
>>
Prefs
   1/*
   2 *  linux/drivers/block/ll_rw_blk.c
   3 *
   4 * Copyright (C) 1991, 1992 Linus Torvalds
   5 * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
   6 * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
   7 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
   8 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> -  July2000
   9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
  10 */
  11
  12/*
  13 * This handles all read/write requests to block devices
  14 */
  15#include <linux/config.h>
  16#include <linux/kernel.h>
  17#include <linux/module.h>
  18#include <linux/backing-dev.h>
  19#include <linux/bio.h>
  20#include <linux/blkdev.h>
  21#include <linux/highmem.h>
  22#include <linux/mm.h>
  23#include <linux/kernel_stat.h>
  24#include <linux/string.h>
  25#include <linux/init.h>
  26#include <linux/bootmem.h>      /* for max_pfn/max_low_pfn */
  27#include <linux/completion.h>
  28#include <linux/slab.h>
  29#include <linux/swap.h>
  30#include <linux/writeback.h>
  31
  32/*
  33 * for max sense size
  34 */
  35#include <scsi/scsi_cmnd.h>
  36
  37static void blk_unplug_work(void *data);
  38static void blk_unplug_timeout(unsigned long data);
  39
  40/*
  41 * For the allocated request tables
  42 */
  43static kmem_cache_t *request_cachep;
  44
  45/*
  46 * For queue allocation
  47 */
  48static kmem_cache_t *requestq_cachep;
  49
  50/*
  51 * For io context allocations
  52 */
  53static kmem_cache_t *iocontext_cachep;
  54
  55static wait_queue_head_t congestion_wqh[2] = {
  56                __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
  57                __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
  58        };
  59
  60/*
  61 * Controlling structure to kblockd
  62 */
  63static struct workqueue_struct *kblockd_workqueue; 
  64
  65unsigned long blk_max_low_pfn, blk_max_pfn;
  66
  67EXPORT_SYMBOL(blk_max_low_pfn);
  68EXPORT_SYMBOL(blk_max_pfn);
  69
  70/* Amount of time in which a process may batch requests */
  71#define BLK_BATCH_TIME  (HZ/50UL)
  72
  73/* Number of requests a "batching" process may submit */
  74#define BLK_BATCH_REQ   32
  75
  76/*
  77 * Return the threshold (number of used requests) at which the queue is
  78 * considered to be congested.  It include a little hysteresis to keep the
  79 * context switch rate down.
  80 */
  81static inline int queue_congestion_on_threshold(struct request_queue *q)
  82{
  83        return q->nr_congestion_on;
  84}
  85
  86/*
  87 * The threshold at which a queue is considered to be uncongested
  88 */
  89static inline int queue_congestion_off_threshold(struct request_queue *q)
  90{
  91        return q->nr_congestion_off;
  92}
  93
  94static void blk_queue_congestion_threshold(struct request_queue *q)
  95{
  96        int nr;
  97
  98        nr = q->nr_requests - (q->nr_requests / 8) + 1;
  99        if (nr > q->nr_requests)
 100                nr = q->nr_requests;
 101        q->nr_congestion_on = nr;
 102
 103        nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
 104        if (nr < 1)
 105                nr = 1;
 106        q->nr_congestion_off = nr;
 107}
 108
 109/*
 110 * A queue has just exitted congestion.  Note this in the global counter of
 111 * congested queues, and wake up anyone who was waiting for requests to be
 112 * put back.
 113 */
 114static void clear_queue_congested(request_queue_t *q, int rw)
 115{
 116        enum bdi_state bit;
 117        wait_queue_head_t *wqh = &congestion_wqh[rw];
 118
 119        bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
 120        clear_bit(bit, &q->backing_dev_info.state);
 121        smp_mb__after_clear_bit();
 122        if (waitqueue_active(wqh))
 123                wake_up(wqh);
 124}
 125
 126/*
 127 * A queue has just entered congestion.  Flag that in the queue's VM-visible
 128 * state flags and increment the global gounter of congested queues.
 129 */
 130static void set_queue_congested(request_queue_t *q, int rw)
 131{
 132        enum bdi_state bit;
 133
 134        bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
 135        set_bit(bit, &q->backing_dev_info.state);
 136}
 137
 138/**
 139 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
 140 * @bdev:       device
 141 *
 142 * Locates the passed device's request queue and returns the address of its
 143 * backing_dev_info
 144 *
 145 * Will return NULL if the request queue cannot be located.
 146 */
 147struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
 148{
 149        struct backing_dev_info *ret = NULL;
 150        request_queue_t *q = bdev_get_queue(bdev);
 151
 152        if (q)
 153                ret = &q->backing_dev_info;
 154        return ret;
 155}
 156
 157EXPORT_SYMBOL(blk_get_backing_dev_info);
 158
 159void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
 160{
 161        q->activity_fn = fn;
 162        q->activity_data = data;
 163}
 164
 165EXPORT_SYMBOL(blk_queue_activity_fn);
 166
 167/**
 168 * blk_queue_prep_rq - set a prepare_request function for queue
 169 * @q:          queue
 170 * @pfn:        prepare_request function
 171 *
 172 * It's possible for a queue to register a prepare_request callback which
 173 * is invoked before the request is handed to the request_fn. The goal of
 174 * the function is to prepare a request for I/O, it can be used to build a
 175 * cdb from the request data for instance.
 176 *
 177 */
 178void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
 179{
 180        q->prep_rq_fn = pfn;
 181}
 182
 183EXPORT_SYMBOL(blk_queue_prep_rq);
 184
 185/**
 186 * blk_queue_merge_bvec - set a merge_bvec function for queue
 187 * @q:          queue
 188 * @mbfn:       merge_bvec_fn
 189 *
 190 * Usually queues have static limitations on the max sectors or segments that
 191 * we can put in a request. Stacking drivers may have some settings that
 192 * are dynamic, and thus we have to query the queue whether it is ok to
 193 * add a new bio_vec to a bio at a given offset or not. If the block device
 194 * has such limitations, it needs to register a merge_bvec_fn to control
 195 * the size of bio's sent to it. Note that a block device *must* allow a
 196 * single page to be added to an empty bio. The block device driver may want
 197 * to use the bio_split() function to deal with these bio's. By default
 198 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
 199 * honored.
 200 */
 201void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
 202{
 203        q->merge_bvec_fn = mbfn;
 204}
 205
 206EXPORT_SYMBOL(blk_queue_merge_bvec);
 207
 208/**
 209 * blk_queue_make_request - define an alternate make_request function for a device
 210 * @q:  the request queue for the device to be affected
 211 * @mfn: the alternate make_request function
 212 *
 213 * Description:
 214 *    The normal way for &struct bios to be passed to a device
 215 *    driver is for them to be collected into requests on a request
 216 *    queue, and then to allow the device driver to select requests
 217 *    off that queue when it is ready.  This works well for many block
 218 *    devices. However some block devices (typically virtual devices
 219 *    such as md or lvm) do not benefit from the processing on the
 220 *    request queue, and are served best by having the requests passed
 221 *    directly to them.  This can be achieved by providing a function
 222 *    to blk_queue_make_request().
 223 *
 224 * Caveat:
 225 *    The driver that does this *must* be able to deal appropriately
 226 *    with buffers in "highmemory". This can be accomplished by either calling
 227 *    __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
 228 *    blk_queue_bounce() to create a buffer in normal memory.
 229 **/
 230void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
 231{
 232        /*
 233         * set defaults
 234         */
 235        q->nr_requests = BLKDEV_MAX_RQ;
 236        q->max_phys_segments = MAX_PHYS_SEGMENTS;
 237        q->max_hw_segments = MAX_HW_SEGMENTS;
 238        q->make_request_fn = mfn;
 239        q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 240        q->backing_dev_info.state = 0;
 241        q->backing_dev_info.memory_backed = 0;
 242        blk_queue_max_sectors(q, MAX_SECTORS);
 243        blk_queue_hardsect_size(q, 512);
 244        blk_queue_dma_alignment(q, 511);
 245        blk_queue_congestion_threshold(q);
 246        q->nr_batching = BLK_BATCH_REQ;
 247
 248        q->unplug_thresh = 4;           /* hmm */
 249        q->unplug_delay = (3 * HZ) / 1000;      /* 3 milliseconds */
 250        if (q->unplug_delay == 0)
 251                q->unplug_delay = 1;
 252
 253        INIT_WORK(&q->unplug_work, blk_unplug_work, q);
 254
 255        q->unplug_timer.function = blk_unplug_timeout;
 256        q->unplug_timer.data = (unsigned long)q;
 257
 258        /*
 259         * by default assume old behaviour and bounce for any highmem page
 260         */
 261        blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
 262
 263        blk_queue_activity_fn(q, NULL, NULL);
 264
 265        INIT_LIST_HEAD(&q->drain_list);
 266}
 267
 268EXPORT_SYMBOL(blk_queue_make_request);
 269
 270/**
 271 * blk_queue_ordered - does this queue support ordered writes
 272 * @q:     the request queue
 273 * @flag:  see below
 274 *
 275 * Description:
 276 *   For journalled file systems, doing ordered writes on a commit
 277 *   block instead of explicitly doing wait_on_buffer (which is bad
 278 *   for performance) can be a big win. Block drivers supporting this
 279 *   feature should call this function and indicate so.
 280 *
 281 **/
 282void blk_queue_ordered(request_queue_t *q, int flag)
 283{
 284        if (flag)
 285                set_bit(QUEUE_FLAG_ORDERED, &q->queue_flags);
 286        else
 287                clear_bit(QUEUE_FLAG_ORDERED, &q->queue_flags);
 288}
 289
 290EXPORT_SYMBOL(blk_queue_ordered);
 291
 292/**
 293 * blk_queue_issue_flush_fn - set function for issuing a flush
 294 * @q:     the request queue
 295 * @iff:   the function to be called issuing the flush
 296 *
 297 * Description:
 298 *   If a driver supports issuing a flush command, the support is notified
 299 *   to the block layer by defining it through this call.
 300 *
 301 **/
 302void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
 303{
 304        q->issue_flush_fn = iff;
 305}
 306
 307EXPORT_SYMBOL(blk_queue_issue_flush_fn);
 308
 309/**
 310 * blk_queue_bounce_limit - set bounce buffer limit for queue
 311 * @q:  the request queue for the device
 312 * @dma_addr:   bus address limit
 313 *
 314 * Description:
 315 *    Different hardware can have different requirements as to what pages
 316 *    it can do I/O directly to. A low level driver can call
 317 *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
 318 *    buffers for doing I/O to pages residing above @page. By default
 319 *    the block layer sets this to the highest numbered "low" memory page.
 320 **/
 321void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
 322{
 323        unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
 324
 325        /*
 326         * set appropriate bounce gfp mask -- unfortunately we don't have a
 327         * full 4GB zone, so we have to resort to low memory for any bounces.
 328         * ISA has its own < 16MB zone.
 329         */
 330        if (bounce_pfn < blk_max_low_pfn) {
 331                BUG_ON(dma_addr < BLK_BOUNCE_ISA);
 332                init_emergency_isa_pool();
 333                q->bounce_gfp = GFP_NOIO | GFP_DMA;
 334        } else
 335                q->bounce_gfp = GFP_NOIO;
 336
 337        q->bounce_pfn = bounce_pfn;
 338}
 339
 340EXPORT_SYMBOL(blk_queue_bounce_limit);
 341
 342/**
 343 * blk_queue_max_sectors - set max sectors for a request for this queue
 344 * @q:  the request queue for the device
 345 * @max_sectors:  max sectors in the usual 512b unit
 346 *
 347 * Description:
 348 *    Enables a low level driver to set an upper limit on the size of
 349 *    received requests.
 350 **/
 351void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors)
 352{
 353        if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
 354                max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
 355                printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
 356        }
 357
 358        q->max_sectors = q->max_hw_sectors = max_sectors;
 359}
 360
 361EXPORT_SYMBOL(blk_queue_max_sectors);
 362
 363/**
 364 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
 365 * @q:  the request queue for the device
 366 * @max_segments:  max number of segments
 367 *
 368 * Description:
 369 *    Enables a low level driver to set an upper limit on the number of
 370 *    physical data segments in a request.  This would be the largest sized
 371 *    scatter list the driver could handle.
 372 **/
 373void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
 374{
 375        if (!max_segments) {
 376                max_segments = 1;
 377                printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
 378        }
 379
 380        q->max_phys_segments = max_segments;
 381}
 382
 383EXPORT_SYMBOL(blk_queue_max_phys_segments);
 384
 385/**
 386 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
 387 * @q:  the request queue for the device
 388 * @max_segments:  max number of segments
 389 *
 390 * Description:
 391 *    Enables a low level driver to set an upper limit on the number of
 392 *    hw data segments in a request.  This would be the largest number of
 393 *    address/length pairs the host adapter can actually give as once
 394 *    to the device.
 395 **/
 396void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
 397{
 398        if (!max_segments) {
 399                max_segments = 1;
 400                printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
 401        }
 402
 403        q->max_hw_segments = max_segments;
 404}
 405
 406EXPORT_SYMBOL(blk_queue_max_hw_segments);
 407
 408/**
 409 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
 410 * @q:  the request queue for the device
 411 * @max_size:  max size of segment in bytes
 412 *
 413 * Description:
 414 *    Enables a low level driver to set an upper limit on the size of a
 415 *    coalesced segment
 416 **/
 417void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
 418{
 419        if (max_size < PAGE_CACHE_SIZE) {
 420                max_size = PAGE_CACHE_SIZE;
 421                printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
 422        }
 423
 424        q->max_segment_size = max_size;
 425}
 426
 427EXPORT_SYMBOL(blk_queue_max_segment_size);
 428
 429/**
 430 * blk_queue_hardsect_size - set hardware sector size for the queue
 431 * @q:  the request queue for the device
 432 * @size:  the hardware sector size, in bytes
 433 *
 434 * Description:
 435 *   This should typically be set to the lowest possible sector size
 436 *   that the hardware can operate on (possible without reverting to
 437 *   even internal read-modify-write operations). Usually the default
 438 *   of 512 covers most hardware.
 439 **/
 440void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
 441{
 442        q->hardsect_size = size;
 443}
 444
 445EXPORT_SYMBOL(blk_queue_hardsect_size);
 446
 447/*
 448 * Returns the minimum that is _not_ zero, unless both are zero.
 449 */
 450#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
 451
 452/**
 453 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
 454 * @t:  the stacking driver (top)
 455 * @b:  the underlying device (bottom)
 456 **/
 457void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
 458{
 459        /* zero is "infinity" */
 460        t->max_sectors = t->max_hw_sectors =
 461                min_not_zero(t->max_sectors,b->max_sectors);
 462
 463        t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
 464        t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
 465        t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
 466        t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
 467}
 468
 469EXPORT_SYMBOL(blk_queue_stack_limits);
 470
 471/**
 472 * blk_queue_segment_boundary - set boundary rules for segment merging
 473 * @q:  the request queue for the device
 474 * @mask:  the memory boundary mask
 475 **/
 476void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
 477{
 478        if (mask < PAGE_CACHE_SIZE - 1) {
 479                mask = PAGE_CACHE_SIZE - 1;
 480                printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
 481        }
 482
 483        q->seg_boundary_mask = mask;
 484}
 485
 486EXPORT_SYMBOL(blk_queue_segment_boundary);
 487
 488/**
 489 * blk_queue_dma_alignment - set dma length and memory alignment
 490 * @q:     the request queue for the device
 491 * @mask:  alignment mask
 492 *
 493 * description:
 494 *    set required memory and length aligment for direct dma transactions.
 495 *    this is used when buiding direct io requests for the queue.
 496 *
 497 **/
 498void blk_queue_dma_alignment(request_queue_t *q, int mask)
 499{
 500        q->dma_alignment = mask;
 501}
 502
 503EXPORT_SYMBOL(blk_queue_dma_alignment);
 504
 505/**
 506 * blk_queue_find_tag - find a request by its tag and queue
 507 *
 508 * @q:   The request queue for the device
 509 * @tag: The tag of the request
 510 *
 511 * Notes:
 512 *    Should be used when a device returns a tag and you want to match
 513 *    it with a request.
 514 *
 515 *    no locks need be held.
 516 **/
 517struct request *blk_queue_find_tag(request_queue_t *q, int tag)
 518{
 519        struct blk_queue_tag *bqt = q->queue_tags;
 520
 521        if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
 522                return NULL;
 523
 524        return bqt->tag_index[tag];
 525}
 526
 527EXPORT_SYMBOL(blk_queue_find_tag);
 528
 529/**
 530 * __blk_queue_free_tags - release tag maintenance info
 531 * @q:  the request queue for the device
 532 *
 533 *  Notes:
 534 *    blk_cleanup_queue() will take care of calling this function, if tagging
 535 *    has been used. So there's no need to call this directly.
 536 **/
 537static void __blk_queue_free_tags(request_queue_t *q)
 538{
 539        struct blk_queue_tag *bqt = q->queue_tags;
 540
 541        if (!bqt)
 542                return;
 543
 544        if (atomic_dec_and_test(&bqt->refcnt)) {
 545                BUG_ON(bqt->busy);
 546                BUG_ON(!list_empty(&bqt->busy_list));
 547
 548                kfree(bqt->tag_index);
 549                bqt->tag_index = NULL;
 550
 551                kfree(bqt->tag_map);
 552                bqt->tag_map = NULL;
 553
 554                kfree(bqt);
 555        }
 556
 557        q->queue_tags = NULL;
 558        q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
 559}
 560
 561/**
 562 * blk_queue_free_tags - release tag maintenance info
 563 * @q:  the request queue for the device
 564 *
 565 *  Notes:
 566 *      This is used to disabled tagged queuing to a device, yet leave
 567 *      queue in function.
 568 **/
 569void blk_queue_free_tags(request_queue_t *q)
 570{
 571        clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
 572}
 573
 574EXPORT_SYMBOL(blk_queue_free_tags);
 575
 576static int
 577init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
 578{
 579        int bits, i;
 580        struct request **tag_index;
 581        unsigned long *tag_map;
 582
 583        if (depth > q->nr_requests * 2) {
 584                depth = q->nr_requests * 2;
 585                printk(KERN_ERR "%s: adjusted depth to %d\n",
 586                                __FUNCTION__, depth);
 587        }
 588
 589        tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC);
 590        if (!tag_index)
 591                goto fail;
 592
 593        bits = (depth / BLK_TAGS_PER_LONG) + 1;
 594        tag_map = kmalloc(bits * sizeof(unsigned long), GFP_ATOMIC);
 595        if (!tag_map)
 596                goto fail;
 597
 598        memset(tag_index, 0, depth * sizeof(struct request *));
 599        memset(tag_map, 0, bits * sizeof(unsigned long));
 600        tags->max_depth = depth;
 601        tags->real_max_depth = bits * BITS_PER_LONG;
 602        tags->tag_index = tag_index;
 603        tags->tag_map = tag_map;
 604
 605        /*
 606         * set the upper bits if the depth isn't a multiple of the word size
 607         */
 608        for (i = depth; i < bits * BLK_TAGS_PER_LONG; i++)
 609                __set_bit(i, tag_map);
 610
 611        return 0;
 612fail:
 613        kfree(tag_index);
 614        return -ENOMEM;
 615}
 616
 617/**
 618 * blk_queue_init_tags - initialize the queue tag info
 619 * @q:  the request queue for the device
 620 * @depth:  the maximum queue depth supported
 621 **/
 622int blk_queue_init_tags(request_queue_t *q, int depth,
 623                        struct blk_queue_tag *tags)
 624{
 625        int rc;
 626
 627        BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
 628
 629        if (!tags && !q->queue_tags) {
 630                tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
 631                if (!tags)
 632                        goto fail;
 633
 634                if (init_tag_map(q, tags, depth))
 635                        goto fail;
 636
 637                INIT_LIST_HEAD(&tags->busy_list);
 638                tags->busy = 0;
 639                atomic_set(&tags->refcnt, 1);
 640        } else if (q->queue_tags) {
 641                if ((rc = blk_queue_resize_tags(q, depth)))
 642                        return rc;
 643                set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
 644                return 0;
 645        } else
 646                atomic_inc(&tags->refcnt);
 647
 648        /*
 649         * assign it, all done
 650         */
 651        q->queue_tags = tags;
 652        q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
 653        return 0;
 654fail:
 655        kfree(tags);
 656        return -ENOMEM;
 657}
 658
 659EXPORT_SYMBOL(blk_queue_init_tags);
 660
 661/**
 662 * blk_queue_resize_tags - change the queueing depth
 663 * @q:  the request queue for the device
 664 * @new_depth: the new max command queueing depth
 665 *
 666 *  Notes:
 667 *    Must be called with the queue lock held.
 668 **/
 669int blk_queue_resize_tags(request_queue_t *q, int new_depth)
 670{
 671        struct blk_queue_tag *bqt = q->queue_tags;
 672        struct request **tag_index;
 673        unsigned long *tag_map;
 674        int bits, max_depth;
 675
 676        if (!bqt)
 677                return -ENXIO;
 678
 679        /*
 680         * don't bother sizing down
 681         */
 682        if (new_depth <= bqt->real_max_depth) {
 683                bqt->max_depth = new_depth;
 684                return 0;
 685        }
 686
 687        /*
 688         * save the old state info, so we can copy it back
 689         */
 690        tag_index = bqt->tag_index;
 691        tag_map = bqt->tag_map;
 692        max_depth = bqt->real_max_depth;
 693
 694        if (init_tag_map(q, bqt, new_depth))
 695                return -ENOMEM;
 696
 697        memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
 698        bits = max_depth / BLK_TAGS_PER_LONG;
 699        memcpy(bqt->tag_map, tag_map, bits * sizeof(unsigned long));
 700
 701        kfree(tag_index);
 702        kfree(tag_map);
 703        return 0;
 704}
 705
 706EXPORT_SYMBOL(blk_queue_resize_tags);
 707
 708/**
 709 * blk_queue_end_tag - end tag operations for a request
 710 * @q:  the request queue for the device
 711 * @rq: the request that has completed
 712 *
 713 *  Description:
 714 *    Typically called when end_that_request_first() returns 0, meaning
 715 *    all transfers have been done for a request. It's important to call
 716 *    this function before end_that_request_last(), as that will put the
 717 *    request back on the free list thus corrupting the internal tag list.
 718 *
 719 *  Notes:
 720 *   queue lock must be held.
 721 **/
 722void blk_queue_end_tag(request_queue_t *q, struct request *rq)
 723{
 724        struct blk_queue_tag *bqt = q->queue_tags;
 725        int tag = rq->tag;
 726
 727        BUG_ON(tag == -1);
 728
 729        if (unlikely(tag >= bqt->real_max_depth))
 730                return;
 731
 732        if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
 733                printk("attempt to clear non-busy tag (%d)\n", tag);
 734                return;
 735        }
 736
 737        list_del_init(&rq->queuelist);
 738        rq->flags &= ~REQ_QUEUED;
 739        rq->tag = -1;
 740
 741        if (unlikely(bqt->tag_index[tag] == NULL))
 742                printk("tag %d is missing\n", tag);
 743
 744        bqt->tag_index[tag] = NULL;
 745        bqt->busy--;
 746}
 747
 748EXPORT_SYMBOL(blk_queue_end_tag);
 749
 750/**
 751 * blk_queue_start_tag - find a free tag and assign it
 752 * @q:  the request queue for the device
 753 * @rq:  the block request that needs tagging
 754 *
 755 *  Description:
 756 *    This can either be used as a stand-alone helper, or possibly be
 757 *    assigned as the queue &prep_rq_fn (in which case &struct request
 758 *    automagically gets a tag assigned). Note that this function
 759 *    assumes that any type of request can be queued! if this is not
 760 *    true for your device, you must check the request type before
 761 *    calling this function.  The request will also be removed from
 762 *    the request queue, so it's the drivers responsibility to readd
 763 *    it if it should need to be restarted for some reason.
 764 *
 765 *  Notes:
 766 *   queue lock must be held.
 767 **/
 768int blk_queue_start_tag(request_queue_t *q, struct request *rq)
 769{
 770        struct blk_queue_tag *bqt = q->queue_tags;
 771        unsigned long *map = bqt->tag_map;
 772        int tag = 0;
 773
 774        if (unlikely((rq->flags & REQ_QUEUED))) {
 775                printk(KERN_ERR 
 776                       "request %p for device [%s] already tagged %d",
 777                       rq, rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
 778                BUG();
 779        }
 780
 781        for (map = bqt->tag_map; *map == -1UL; map++) {
 782                tag += BLK_TAGS_PER_LONG;
 783
 784                if (tag >= bqt->max_depth)
 785                        return 1;
 786        }
 787
 788        tag += ffz(*map);
 789        __set_bit(tag, bqt->tag_map);
 790
 791        rq->flags |= REQ_QUEUED;
 792        rq->tag = tag;
 793        bqt->tag_index[tag] = rq;
 794        blkdev_dequeue_request(rq);
 795        list_add(&rq->queuelist, &bqt->busy_list);
 796        bqt->busy++;
 797        return 0;
 798}
 799
 800EXPORT_SYMBOL(blk_queue_start_tag);
 801
 802/**
 803 * blk_queue_invalidate_tags - invalidate all pending tags
 804 * @q:  the request queue for the device
 805 *
 806 *  Description:
 807 *   Hardware conditions may dictate a need to stop all pending requests.
 808 *   In this case, we will safely clear the block side of the tag queue and
 809 *   readd all requests to the request queue in the right order.
 810 *
 811 *  Notes:
 812 *   queue lock must be held.
 813 **/
 814void blk_queue_invalidate_tags(request_queue_t *q)
 815{
 816        struct blk_queue_tag *bqt = q->queue_tags;
 817        struct list_head *tmp, *n;
 818        struct request *rq;
 819
 820        list_for_each_safe(tmp, n, &bqt->busy_list) {
 821                rq = list_entry_rq(tmp);
 822
 823                if (rq->tag == -1) {
 824                        printk("bad tag found on list\n");
 825                        list_del_init(&rq->queuelist);
 826                        rq->flags &= ~REQ_QUEUED;
 827                } else
 828                        blk_queue_end_tag(q, rq);
 829
 830                rq->flags &= ~REQ_STARTED;
 831                __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
 832        }
 833}
 834
 835EXPORT_SYMBOL(blk_queue_invalidate_tags);
 836
 837static char *rq_flags[] = {
 838        "REQ_RW",
 839        "REQ_FAILFAST",
 840        "REQ_SOFTBARRIER",
 841        "REQ_HARDBARRIER",
 842        "REQ_CMD",
 843        "REQ_NOMERGE",
 844        "REQ_STARTED",
 845        "REQ_DONTPREP",
 846        "REQ_QUEUED",
 847        "REQ_PC",
 848        "REQ_BLOCK_PC",
 849        "REQ_SENSE",
 850        "REQ_FAILED",
 851        "REQ_QUIET",
 852        "REQ_SPECIAL",
 853        "REQ_DRIVE_CMD",
 854        "REQ_DRIVE_TASK",
 855        "REQ_DRIVE_TASKFILE",
 856        "REQ_PREEMPT",
 857        "REQ_PM_SUSPEND",
 858        "REQ_PM_RESUME",
 859        "REQ_PM_SHUTDOWN",
 860};
 861
 862void blk_dump_rq_flags(struct request *rq, char *msg)
 863{
 864        int bit;
 865
 866        printk("%s: dev %s: flags = ", msg,
 867                rq->rq_disk ? rq->rq_disk->disk_name : "?");
 868        bit = 0;
 869        do {
 870                if (rq->flags & (1 << bit))
 871                        printk("%s ", rq_flags[bit]);
 872                bit++;
 873        } while (bit < __REQ_NR_BITS);
 874
 875        printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
 876                                                       rq->nr_sectors,
 877                                                       rq->current_nr_sectors);
 878        printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
 879
 880        if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) {
 881                printk("cdb: ");
 882                for (bit = 0; bit < sizeof(rq->cmd); bit++)
 883                        printk("%02x ", rq->cmd[bit]);
 884                printk("\n");
 885        }
 886}
 887
 888EXPORT_SYMBOL(blk_dump_rq_flags);
 889
 890void blk_recount_segments(request_queue_t *q, struct bio *bio)
 891{
 892        struct bio_vec *bv, *bvprv = NULL;
 893        int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
 894        int high, highprv = 1;
 895
 896        if (unlikely(!bio->bi_io_vec))
 897                return;
 898
 899        cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
 900        hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
 901        bio_for_each_segment(bv, bio, i) {
 902                /*
 903                 * the trick here is making sure that a high page is never
 904                 * considered part of another segment, since that might
 905                 * change with the bounce page.
 906                 */
 907                high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
 908                if (high || highprv)
 909                        goto new_hw_segment;
 910                if (cluster) {
 911                        if (seg_size + bv->bv_len > q->max_segment_size)
 912                                goto new_segment;
 913                        if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
 914                                goto new_segment;
 915                        if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
 916                                goto new_segment;
 917                        if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
 918                                goto new_hw_segment;
 919
 920                        seg_size += bv->bv_len;
 921                        hw_seg_size += bv->bv_len;
 922                        bvprv = bv;
 923                        continue;
 924                }
 925new_segment:
 926                if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
 927                    !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
 928                        hw_seg_size += bv->bv_len;
 929                } else {
 930new_hw_segment:
 931                        if (hw_seg_size > bio->bi_hw_front_size)
 932                                bio->bi_hw_front_size = hw_seg_size;
 933                        hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
 934                        nr_hw_segs++;
 935                }
 936
 937                nr_phys_segs++;
 938                bvprv = bv;
 939                seg_size = bv->bv_len;
 940                highprv = high;
 941        }
 942        if (hw_seg_size > bio->bi_hw_back_size)
 943                bio->bi_hw_back_size = hw_seg_size;
 944        if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
 945                bio->bi_hw_front_size = hw_seg_size;
 946        bio->bi_phys_segments = nr_phys_segs;
 947        bio->bi_hw_segments = nr_hw_segs;
 948        bio->bi_flags |= (1 << BIO_SEG_VALID);
 949}
 950
 951
 952int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
 953                                   struct bio *nxt)
 954{
 955        if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
 956                return 0;
 957
 958        if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
 959                return 0;
 960        if (bio->bi_size + nxt->bi_size > q->max_segment_size)
 961                return 0;
 962
 963        /*
 964         * bio and nxt are contigous in memory, check if the queue allows
 965         * these two to be merged into one
 966         */
 967        if (BIO_SEG_BOUNDARY(q, bio, nxt))
 968                return 1;
 969
 970        return 0;
 971}
 972
 973EXPORT_SYMBOL(blk_phys_contig_segment);
 974
 975int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
 976                                 struct bio *nxt)
 977{
 978        if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
 979                blk_recount_segments(q, bio);
 980        if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
 981                blk_recount_segments(q, nxt);
 982        if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
 983            BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
 984                return 0;
 985        if (bio->bi_size + nxt->bi_size > q->max_segment_size)
 986                return 0;
 987
 988        return 1;
 989}
 990
 991EXPORT_SYMBOL(blk_hw_contig_segment);
 992
 993/*
 994 * map a request to scatterlist, return number of sg entries setup. Caller
 995 * must make sure sg can hold rq->nr_phys_segments entries
 996 */
 997int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
 998{
 999        struct bio_vec *bvec, *bvprv;
1000        struct bio *bio;
1001        int nsegs, i, cluster;
1002
1003        nsegs = 0;
1004        cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1005
1006        /*
1007         * for each bio in rq
1008         */
1009        bvprv = NULL;
1010        rq_for_each_bio(bio, rq) {
1011                /*
1012                 * for each segment in bio
1013                 */
1014                bio_for_each_segment(bvec, bio, i) {
1015                        int nbytes = bvec->bv_len;
1016
1017                        if (bvprv && cluster) {
1018                                if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1019                                        goto new_segment;
1020
1021                                if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1022                                        goto new_segment;
1023                                if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1024                                        goto new_segment;
1025
1026                                sg[nsegs - 1].length += nbytes;
1027                        } else {
1028new_segment:
1029                                memset(&sg[nsegs],0,sizeof(struct scatterlist));
1030                                sg[nsegs].page = bvec->bv_page;
1031                                sg[nsegs].length = nbytes;
1032                                sg[nsegs].offset = bvec->bv_offset;
1033
1034                                nsegs++;
1035                        }
1036                        bvprv = bvec;
1037                } /* segments in bio */
1038        } /* bios in rq */
1039
1040        return nsegs;
1041}
1042
1043EXPORT_SYMBOL(blk_rq_map_sg);
1044
1045/*
1046 * the standard queue merge functions, can be overridden with device
1047 * specific ones if so desired
1048 */
1049
1050static inline int ll_new_mergeable(request_queue_t *q,
1051                                   struct request *req,
1052                                   struct bio *bio)
1053{
1054        int nr_phys_segs = bio_phys_segments(q, bio);
1055
1056        if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1057                req->flags |= REQ_NOMERGE;
1058                if (req == q->last_merge)
1059                        q->last_merge = NULL;
1060                return 0;
1061        }
1062
1063        /*
1064         * A hw segment is just getting larger, bump just the phys
1065         * counter.
1066         */
1067        req->nr_phys_segments += nr_phys_segs;
1068        return 1;
1069}
1070
1071static inline int ll_new_hw_segment(request_queue_t *q,
1072                                    struct request *req,
1073                                    struct bio *bio)
1074{
1075        int nr_hw_segs = bio_hw_segments(q, bio);
1076        int nr_phys_segs = bio_phys_segments(q, bio);
1077
1078        if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1079            || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1080                req->flags |= REQ_NOMERGE;
1081                if (req == q->last_merge)
1082                        q->last_merge = NULL;
1083                return 0;
1084        }
1085
1086        /*
1087         * This will form the start of a new hw segment.  Bump both
1088         * counters.
1089         */
1090        req->nr_hw_segments += nr_hw_segs;
1091        req->nr_phys_segments += nr_phys_segs;
1092        return 1;
1093}
1094
1095static int ll_back_merge_fn(request_queue_t *q, struct request *req, 
1096                            struct bio *bio)
1097{
1098        int len;
1099
1100        if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1101                req->flags |= REQ_NOMERGE;
1102                if (req == q->last_merge)
1103                        q->last_merge = NULL;
1104                return 0;
1105        }
1106        if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1107                blk_recount_segments(q, req->biotail);
1108        if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1109                blk_recount_segments(q, bio);
1110        len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1111        if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1112            !BIOVEC_VIRT_OVERSIZE(len)) {
1113                int mergeable =  ll_new_mergeable(q, req, bio);
1114
1115                if (mergeable) {
1116                        if (req->nr_hw_segments == 1)
1117                                req->bio->bi_hw_front_size = len;
1118                        if (bio->bi_hw_segments == 1)
1119                                bio->bi_hw_back_size = len;
1120                }
1121                return mergeable;
1122        }
1123
1124        return ll_new_hw_segment(q, req, bio);
1125}
1126
1127static int ll_front_merge_fn(request_queue_t *q, struct request *req, 
1128                             struct bio *bio)
1129{
1130        int len;
1131
1132        if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1133                req->flags |= REQ_NOMERGE;
1134                if (req == q->last_merge)
1135                        q->last_merge = NULL;
1136                return 0;
1137        }
1138        len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1139        if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1140                blk_recount_segments(q, bio);
1141        if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1142                blk_recount_segments(q, req->bio);
1143        if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1144            !BIOVEC_VIRT_OVERSIZE(len)) {
1145                int mergeable =  ll_new_mergeable(q, req, bio);
1146
1147                if (mergeable) {
1148                        if (bio->bi_hw_segments == 1)
1149                                bio->bi_hw_front_size = len;
1150                        if (req->nr_hw_segments == 1)
1151                                req->biotail->bi_hw_back_size = len;
1152                }
1153                return mergeable;
1154        }
1155
1156        return ll_new_hw_segment(q, req, bio);
1157}
1158
1159static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1160                                struct request *next)
1161{
1162        int total_phys_segments = req->nr_phys_segments +next->nr_phys_segments;
1163        int total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1164
1165        /*
1166         * First check if the either of the requests are re-queued
1167         * requests.  Can't merge them if they are.
1168         */
1169        if (req->special || next->special)
1170                return 0;
1171
1172        /*
1173         * Will it become to large?
1174         */
1175        if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1176                return 0;
1177
1178        total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1179        if (blk_phys_contig_segment(q, req->biotail, next->bio))
1180                total_phys_segments--;
1181
1182        if (total_phys_segments > q->max_phys_segments)
1183                return 0;
1184
1185        total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1186        if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1187                int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1188                /*
1189                 * propagate the combined length to the end of the requests
1190                 */
1191                if (req->nr_hw_segments == 1)
1192                        req->bio->bi_hw_front_size = len;
1193                if (next->nr_hw_segments == 1)
1194                        next->biotail->bi_hw_back_size = len;
1195                total_hw_segments--;
1196        }
1197
1198        if (total_hw_segments > q->max_hw_segments)
1199                return 0;
1200
1201        /* Merge is OK... */
1202        req->nr_phys_segments = total_phys_segments;
1203        req->nr_hw_segments = total_hw_segments;
1204        return 1;
1205}
1206
1207/*
1208 * "plug" the device if there are no outstanding requests: this will
1209 * force the transfer to start only after we have put all the requests
1210 * on the list.
1211 *
1212 * This is called with interrupts off and no requests on the queue and
1213 * with the queue lock held.
1214 */
1215void blk_plug_device(request_queue_t *q)
1216{
1217        WARN_ON(!irqs_disabled());
1218
1219        /*
1220         * don't plug a stopped queue, it must be paired with blk_start_queue()
1221         * which will restart the queueing
1222         */
1223        if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1224                return;
1225
1226        if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1227                mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1228}
1229
1230EXPORT_SYMBOL(blk_plug_device);
1231
1232/*
1233 * remove the queue from the plugged list, if present. called with
1234 * queue lock held and interrupts disabled.
1235 */
1236int blk_remove_plug(request_queue_t *q)
1237{
1238        WARN_ON(!irqs_disabled());
1239
1240        if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1241                return 0;
1242
1243        del_timer(&q->unplug_timer);
1244        return 1;
1245}
1246
1247EXPORT_SYMBOL(blk_remove_plug);
1248
1249/*
1250 * remove the plug and let it rip..
1251 */
1252void __generic_unplug_device(request_queue_t *q)
1253{
1254        if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1255                return;
1256
1257        if (!blk_remove_plug(q))
1258                return;
1259
1260        /*
1261         * was plugged, fire request_fn if queue has stuff to do
1262         */
1263        if (elv_next_request(q))
1264                q->request_fn(q);
1265}
1266EXPORT_SYMBOL(__generic_unplug_device);
1267
1268/**
1269 * generic_unplug_device - fire a request queue
1270 * @q:    The &request_queue_t in question
1271 *
1272 * Description:
1273 *   Linux uses plugging to build bigger requests queues before letting
1274 *   the device have at them. If a queue is plugged, the I/O scheduler
1275 *   is still adding and merging requests on the queue. Once the queue
1276 *   gets unplugged, the request_fn defined for the queue is invoked and
1277 *   transfers started.
1278 **/
1279void generic_unplug_device(request_queue_t *q)
1280{
1281        spin_lock_irq(q->queue_lock);
1282        __generic_unplug_device(q);
1283        spin_unlock_irq(q->queue_lock);
1284}
1285EXPORT_SYMBOL(generic_unplug_device);
1286
1287static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1288                                   struct page *page)
1289{
1290        request_queue_t *q = bdi->unplug_io_data;
1291
1292        /*
1293         * devices don't necessarily have an ->unplug_fn defined
1294         */
1295        if (q->unplug_fn)
1296                q->unplug_fn(q);
1297}
1298
1299static void blk_unplug_work(void *data)
1300{
1301        request_queue_t *q = data;
1302
1303        q->unplug_fn(q);
1304}
1305
1306static void blk_unplug_timeout(unsigned long data)
1307{
1308        request_queue_t *q = (request_queue_t *)data;
1309
1310        kblockd_schedule_work(&q->unplug_work);
1311}
1312
1313/**
1314 * blk_start_queue - restart a previously stopped queue
1315 * @q:    The &request_queue_t in question
1316 *
1317 * Description:
1318 *   blk_start_queue() will clear the stop flag on the queue, and call
1319 *   the request_fn for the queue if it was in a stopped state when
1320 *   entered. Also see blk_stop_queue(). Queue lock must be held.
1321 **/
1322void blk_start_queue(request_queue_t *q)
1323{
1324        clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1325
1326        /*
1327         * one level of recursion is ok and is much faster than kicking
1328         * the unplug handling
1329         */
1330        if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1331                q->request_fn(q);
1332                clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1333        } else {
1334                blk_plug_device(q);
1335                kblockd_schedule_work(&q->unplug_work);
1336        }
1337}
1338
1339EXPORT_SYMBOL(blk_start_queue);
1340
1341/**
1342 * blk_stop_queue - stop a queue
1343 * @q:    The &request_queue_t in question
1344 *
1345 * Description:
1346 *   The Linux block layer assumes that a block driver will consume all
1347 *   entries on the request queue when the request_fn strategy is called.
1348 *   Often this will not happen, because of hardware limitations (queue
1349 *   depth settings). If a device driver gets a 'queue full' response,
1350 *   or if it simply chooses not to queue more I/O at one point, it can
1351 *   call this function to prevent the request_fn from being called until
1352 *   the driver has signalled it's ready to go again. This happens by calling
1353 *   blk_start_queue() to restart queue operations. Queue lock must be held.
1354 **/
1355void blk_stop_queue(request_queue_t *q)
1356{
1357        blk_remove_plug(q);
1358        set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1359}
1360EXPORT_SYMBOL(blk_stop_queue);
1361
1362/**
1363 * blk_sync_queue - cancel any pending callbacks on a queue
1364 * @q: the queue
1365 *
1366 * Description:
1367 *     The block layer may perform asynchronous callback activity
1368 *     on a queue, such as calling the unplug function after a timeout.
1369 *     A block device may call blk_sync_queue to ensure that any
1370 *     such activity is cancelled, thus allowing it to release resources
1371 *     the the callbacks might use. The caller must already have made sure
1372 *     that its ->make_request_fn will not re-add plugging prior to calling
1373 *     this function.
1374 *
1375 */
1376void blk_sync_queue(struct request_queue *q)
1377{
1378        del_timer_sync(&q->unplug_timer);
1379        kblockd_flush();
1380}
1381EXPORT_SYMBOL(blk_sync_queue);
1382
1383/**
1384 * blk_run_queue - run a single device queue
1385 * @q:  The queue to run
1386 */
1387void blk_run_queue(struct request_queue *q)
1388{
1389        unsigned long flags;
1390
1391        spin_lock_irqsave(q->queue_lock, flags);
1392        blk_remove_plug(q);
1393        q->request_fn(q);
1394        spin_unlock_irqrestore(q->queue_lock, flags);
1395}
1396EXPORT_SYMBOL(blk_run_queue);
1397
1398/**
1399 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1400 * @q:    the request queue to be released
1401 *
1402 * Description:
1403 *     blk_cleanup_queue is the pair to blk_init_queue() or
1404 *     blk_queue_make_request().  It should be called when a request queue is
1405 *     being released; typically when a block device is being de-registered.
1406 *     Currently, its primary task it to free all the &struct request
1407 *     structures that were allocated to the queue and the queue itself.
1408 *
1409 * Caveat:
1410 *     Hopefully the low level driver will have finished any
1411 *     outstanding requests first...
1412 **/
1413void blk_cleanup_queue(request_queue_t * q)
1414{
1415        struct request_list *rl = &q->rq;
1416
1417        if (!atomic_dec_and_test(&q->refcnt))
1418                return;
1419
1420        if (q->elevator)
1421                elevator_exit(q->elevator);
1422
1423        blk_sync_queue(q);
1424
1425        if (rl->rq_pool)
1426                mempool_destroy(rl->rq_pool);
1427
1428        if (q->queue_tags)
1429                __blk_queue_free_tags(q);
1430
1431        kmem_cache_free(requestq_cachep, q);
1432}
1433
1434EXPORT_SYMBOL(blk_cleanup_queue);
1435
1436static int blk_init_free_list(request_queue_t *q)
1437{
1438        struct request_list *rl = &q->rq;
1439
1440        rl->count[READ] = rl->count[WRITE] = 0;
1441        rl->starved[READ] = rl->starved[WRITE] = 0;
1442        init_waitqueue_head(&rl->wait[READ]);
1443        init_waitqueue_head(&rl->wait[WRITE]);
1444        init_waitqueue_head(&rl->drain);
1445
1446        rl->rq_pool = mempool_create(BLKDEV_MIN_RQ, mempool_alloc_slab, mempool_free_slab, request_cachep);
1447
1448        if (!rl->rq_pool)
1449                return -ENOMEM;
1450
1451        return 0;
1452}
1453
1454static int __make_request(request_queue_t *, struct bio *);
1455
1456request_queue_t *blk_alloc_queue(int gfp_mask)
1457{
1458        request_queue_t *q = kmem_cache_alloc(requestq_cachep, gfp_mask);
1459
1460        if (!q)
1461                return NULL;
1462
1463        memset(q, 0, sizeof(*q));
1464        init_timer(&q->unplug_timer);
1465        atomic_set(&q->refcnt, 1);
1466
1467        q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1468        q->backing_dev_info.unplug_io_data = q;
1469
1470        return q;
1471}
1472
1473EXPORT_SYMBOL(blk_alloc_queue);
1474
1475/**
1476 * blk_init_queue  - prepare a request queue for use with a block device
1477 * @rfn:  The function to be called to process requests that have been
1478 *        placed on the queue.
1479 * @lock: Request queue spin lock
1480 *
1481 * Description:
1482 *    If a block device wishes to use the standard request handling procedures,
1483 *    which sorts requests and coalesces adjacent requests, then it must
1484 *    call blk_init_queue().  The function @rfn will be called when there
1485 *    are requests on the queue that need to be processed.  If the device
1486 *    supports plugging, then @rfn may not be called immediately when requests
1487 *    are available on the queue, but may be called at some time later instead.
1488 *    Plugged queues are generally unplugged when a buffer belonging to one
1489 *    of the requests on the queue is needed, or due to memory pressure.
1490 *
1491 *    @rfn is not required, or even expected, to remove all requests off the
1492 *    queue, but only as many as it can handle at a time.  If it does leave
1493 *    requests on the queue, it is responsible for arranging that the requests
1494 *    get dealt with eventually.
1495 *
1496 *    The queue spin lock must be held while manipulating the requests on the
1497 *    request queue.
1498 *
1499 *    Function returns a pointer to the initialized request queue, or NULL if
1500 *    it didn't succeed.
1501 *
1502 * Note:
1503 *    blk_init_queue() must be paired with a blk_cleanup_queue() call
1504 *    when the block device is deactivated (such as at module unload).
1505 **/
1506request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1507{
1508        request_queue_t *q = blk_alloc_queue(GFP_KERNEL);
1509
1510        if (!q)
1511                return NULL;
1512
1513        if (blk_init_free_list(q))
1514                goto out_init;
1515
1516        q->request_fn           = rfn;
1517        q->back_merge_fn        = ll_back_merge_fn;
1518        q->front_merge_fn       = ll_front_merge_fn;
1519        q->merge_requests_fn    = ll_merge_requests_fn;
1520        q->prep_rq_fn           = NULL;
1521        q->unplug_fn            = generic_unplug_device;
1522        q->queue_flags          = (1 << QUEUE_FLAG_CLUSTER);
1523        q->queue_lock           = lock;
1524
1525        blk_queue_segment_boundary(q, 0xffffffff);
1526
1527        blk_queue_make_request(q, __make_request);
1528        blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1529
1530        blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1531        blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1532
1533        /*
1534         * all done
1535         */
1536        if (!elevator_init(q, NULL)) {
1537                blk_queue_congestion_threshold(q);
1538                return q;
1539        }
1540
1541        blk_cleanup_queue(q);
1542out_init:
1543        kmem_cache_free(requestq_cachep, q);
1544        return NULL;
1545}
1546
1547EXPORT_SYMBOL(blk_init_queue);
1548
1549int blk_get_queue(request_queue_t *q)
1550{
1551        if (!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
1552                atomic_inc(&q->refcnt);
1553                return 0;
1554        }
1555
1556        return 1;
1557}
1558
1559EXPORT_SYMBOL(blk_get_queue);
1560
1561static inline void blk_free_request(request_queue_t *q, struct request *rq)
1562{
1563        elv_put_request(q, rq);
1564        mempool_free(rq, q->rq.rq_pool);
1565}
1566
1567static inline struct request *blk_alloc_request(request_queue_t *q, int rw,
1568                                                int gfp_mask)
1569{
1570        struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1571
1572        if (!rq)
1573                return NULL;
1574
1575        /*
1576         * first three bits are identical in rq->flags and bio->bi_rw,
1577         * see bio.h and blkdev.h
1578         */
1579        rq->flags = rw;
1580
1581        if (!elv_set_request(q, rq, gfp_mask))
1582                return rq;
1583
1584        mempool_free(rq, q->rq.rq_pool);
1585        return NULL;
1586}
1587
1588/*
1589 * ioc_batching returns true if the ioc is a valid batching request and
1590 * should be given priority access to a request.
1591 */
1592static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
1593{
1594        if (!ioc)
1595                return 0;
1596
1597        /*
1598         * Make sure the process is able to allocate at least 1 request
1599         * even if the batch times out, otherwise we could theoretically
1600         * lose wakeups.
1601         */
1602        return ioc->nr_batch_requests == q->nr_batching ||
1603                (ioc->nr_batch_requests > 0
1604                && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1605}
1606
1607/*
1608 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1609 * will cause the process to be a "batcher" on all queues in the system. This
1610 * is the behaviour we want though - once it gets a wakeup it should be given
1611 * a nice run.
1612 */
1613void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
1614{
1615        if (!ioc || ioc_batching(q, ioc))
1616                return;
1617
1618        ioc->nr_batch_requests = q->nr_batching;
1619        ioc->last_waited = jiffies;
1620}
1621
1622static void __freed_request(request_queue_t *q, int rw)
1623{
1624        struct request_list *rl = &q->rq;
1625
1626        if (rl->count[rw] < queue_congestion_off_threshold(q))
1627                clear_queue_congested(q, rw);
1628
1629        if (rl->count[rw] + 1 <= q->nr_requests) {
1630                smp_mb();
1631                if (waitqueue_active(&rl->wait[rw]))
1632                        wake_up(&rl->wait[rw]);
1633
1634                blk_clear_queue_full(q, rw);
1635        }
1636}
1637
1638/*
1639 * A request has just been released.  Account for it, update the full and
1640 * congestion status, wake up any waiters.   Called under q->queue_lock.
1641 */
1642static void freed_request(request_queue_t *q, int rw)
1643{
1644        struct request_list *rl = &q->rq;
1645
1646        rl->count[rw]--;
1647
1648        __freed_request(q, rw);
1649
1650        if (unlikely(rl->starved[rw ^ 1]))
1651                __freed_request(q, rw ^ 1);
1652
1653        if (!rl->count[READ] && !rl->count[WRITE]) {
1654                smp_mb();
1655                if (unlikely(waitqueue_active(&rl->drain)))
1656                        wake_up(&rl->drain);
1657        }
1658}
1659
1660#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1661/*
1662 * Get a free request, queue_lock must not be held
1663 */
1664static struct request *get_request(request_queue_t *q, int rw, int gfp_mask)
1665{
1666        struct request *rq = NULL;
1667        struct request_list *rl = &q->rq;
1668        struct io_context *ioc = get_io_context(gfp_mask);
1669
1670        if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)))
1671                goto out;
1672
1673        spin_lock_irq(q->queue_lock);
1674        if (rl->count[rw]+1 >= q->nr_requests) {
1675                /*
1676                 * The queue will fill after this allocation, so set it as
1677                 * full, and mark this process as "batching". This process
1678                 * will be allowed to complete a batch of requests, others
1679                 * will be blocked.
1680                 */
1681                if (!blk_queue_full(q, rw)) {
1682                        ioc_set_batching(q, ioc);
1683                        blk_set_queue_full(q, rw);
1684                }
1685        }
1686
1687        switch (elv_may_queue(q, rw)) {
1688                case ELV_MQUEUE_NO:
1689                        goto rq_starved;
1690                case ELV_MQUEUE_MAY:
1691                        break;
1692                case ELV_MQUEUE_MUST:
1693                        goto get_rq;
1694        }
1695
1696        if (blk_queue_full(q, rw) && !ioc_batching(q, ioc)) {
1697                /*
1698                 * The queue is full and the allocating process is not a
1699                 * "batcher", and not exempted by the IO scheduler
1700                 */
1701                spin_unlock_irq(q->queue_lock);
1702                goto out;
1703        }
1704
1705get_rq:
1706        rl->count[rw]++;
1707        rl->starved[rw] = 0;
1708        if (rl->count[rw] >= queue_congestion_on_threshold(q))
1709                set_queue_congested(q, rw);
1710        spin_unlock_irq(q->queue_lock);
1711
1712        rq = blk_alloc_request(q, rw, gfp_mask);
1713        if (!rq) {
1714                /*
1715                 * Allocation failed presumably due to memory. Undo anything
1716                 * we might have messed up.
1717                 *
1718                 * Allocating task should really be put onto the front of the
1719                 * wait queue, but this is pretty rare.
1720                 */
1721                spin_lock_irq(q->queue_lock);
1722                freed_request(q, rw);
1723
1724                /*
1725                 * in the very unlikely event that allocation failed and no
1726                 * requests for this direction was pending, mark us starved
1727                 * so that freeing of a request in the other direction will
1728                 * notice us. another possible fix would be to split the
1729                 * rq mempool into READ and WRITE
1730                 */
1731rq_starved:
1732                if (unlikely(rl->count[rw] == 0))
1733                        rl->starved[rw] = 1;
1734
1735                spin_unlock_irq(q->queue_lock);
1736                goto out;
1737        }
1738
1739        if (ioc_batching(q, ioc))
1740                ioc->nr_batch_requests--;
1741        
1742        INIT_LIST_HEAD(&rq->queuelist);
1743
1744        rq->errors = 0;
1745        rq->rq_status = RQ_ACTIVE;
1746        rq->bio = rq->biotail = NULL;
1747        rq->buffer = NULL;
1748        rq->ref_count = 1;
1749        rq->q = q;
1750        rq->rl = rl;
1751        rq->waiting = NULL;
1752        rq->special = NULL;
1753        rq->data_len = 0;
1754        rq->data = NULL;
1755        rq->sense = NULL;
1756
1757out:
1758        put_io_context(ioc);
1759        return rq;
1760}
1761
1762/*
1763 * No available requests for this queue, unplug the device and wait for some
1764 * requests to become available.
1765 */
1766static struct request *get_request_wait(request_queue_t *q, int rw)
1767{
1768        DEFINE_WAIT(wait);
1769        struct request *rq;
1770
1771        generic_unplug_device(q);
1772        do {
1773                struct request_list *rl = &q->rq;
1774
1775                prepare_to_wait_exclusive(&rl->wait[rw], &wait,
1776                                TASK_UNINTERRUPTIBLE);
1777
1778                rq = get_request(q, rw, GFP_NOIO);
1779
1780                if (!rq) {
1781                        struct io_context *ioc;
1782
1783                        io_schedule();
1784
1785                        /*
1786                         * After sleeping, we become a "batching" process and
1787                         * will be able to allocate at least one request, and
1788                         * up to a big batch of them for a small period time.
1789                         * See ioc_batching, ioc_set_batching
1790                         */
1791                        ioc = get_io_context(GFP_NOIO);
1792                        ioc_set_batching(q, ioc);
1793                        put_io_context(ioc);
1794                }
1795                finish_wait(&rl->wait[rw], &wait);
1796        } while (!rq);
1797
1798        return rq;
1799}
1800
1801struct request *blk_get_request(request_queue_t *q, int rw, int gfp_mask)
1802{
1803        struct request *rq;
1804
1805        BUG_ON(rw != READ && rw != WRITE);
1806
1807        if (gfp_mask & __GFP_WAIT)
1808                rq = get_request_wait(q, rw);
1809        else
1810                rq = get_request(q, rw, gfp_mask);
1811
1812        return rq;
1813}
1814
1815EXPORT_SYMBOL(blk_get_request);
1816
1817/**
1818 * blk_requeue_request - put a request back on queue
1819 * @q:          request queue where request should be inserted
1820 * @rq:         request to be inserted
1821 *
1822 * Description:
1823 *    Drivers often keep queueing requests until the hardware cannot accept
1824 *    more, when that condition happens we need to put the request back
1825 *    on the queue. Must be called with queue lock held.
1826 */
1827void blk_requeue_request(request_queue_t *q, struct request *rq)
1828{
1829        if (blk_rq_tagged(rq))
1830                blk_queue_end_tag(q, rq);
1831
1832        elv_requeue_request(q, rq);
1833}
1834
1835EXPORT_SYMBOL(blk_requeue_request);
1836
1837/**
1838 * blk_insert_request - insert a special request in to a request queue
1839 * @q:          request queue where request should be inserted
1840 * @rq:         request to be inserted
1841 * @at_head:    insert request at head or tail of queue
1842 * @data:       private data
1843 * @reinsert:   true if request it a reinsertion of previously processed one
1844 *
1845 * Description:
1846 *    Many block devices need to execute commands asynchronously, so they don't
1847 *    block the whole kernel from preemption during request execution.  This is
1848 *    accomplished normally by inserting aritficial requests tagged as
1849 *    REQ_SPECIAL in to the corresponding request queue, and letting them be
1850 *    scheduled for actual execution by the request queue.
1851 *
1852 *    We have the option of inserting the head or the tail of the queue.
1853 *    Typically we use the tail for new ioctls and so forth.  We use the head
1854 *    of the queue for things like a QUEUE_FULL message from a device, or a
1855 *    host that is unable to accept a particular command.
1856 */
1857void blk_insert_request(request_queue_t *q, struct request *rq,
1858                        int at_head, void *data, int reinsert)
1859{
1860        unsigned long flags;
1861
1862        /*
1863         * tell I/O scheduler that this isn't a regular read/write (ie it
1864         * must not attempt merges on this) and that it acts as a soft
1865         * barrier
1866         */
1867        rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
1868
1869        rq->special = data;
1870
1871        spin_lock_irqsave(q->queue_lock, flags);
1872
1873        /*
1874         * If command is tagged, release the tag
1875         */
1876        if (reinsert)
1877                blk_requeue_request(q, rq);
1878        else {
1879                int where = ELEVATOR_INSERT_BACK;
1880
1881                if (at_head)
1882                        where = ELEVATOR_INSERT_FRONT;
1883
1884                if (blk_rq_tagged(rq))
1885                        blk_queue_end_tag(q, rq);
1886
1887                drive_stat_acct(rq, rq->nr_sectors, 1);
1888                __elv_add_request(q, rq, where, 0);
1889        }
1890        if (blk_queue_plugged(q))
1891                __generic_unplug_device(q);
1892        else
1893                q->request_fn(q);
1894        spin_unlock_irqrestore(q->queue_lock, flags);
1895}
1896
1897EXPORT_SYMBOL(blk_insert_request);
1898
1899/**
1900 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
1901 * @q:          request queue where request should be inserted
1902 * @rw:         READ or WRITE data
1903 * @ubuf:       the user buffer
1904 * @len:        length of user data
1905 *
1906 * Description:
1907 *    Data will be mapped directly for zero copy io, if possible. Otherwise
1908 *    a kernel bounce buffer is used.
1909 *
1910 *    A matching blk_rq_unmap_user() must be issued at the end of io, while
1911 *    still in process context.
1912 *
1913 *    Note: The mapped bio may need to be bounced through blk_queue_bounce()
1914 *    before being submitted to the device, as pages mapped may be out of
1915 *    reach. It's the callers responsibility to make sure this happens. The
1916 *    original bio must be passed back in to blk_rq_unmap_user() for proper
1917 *    unmapping.
1918 */
1919struct request *blk_rq_map_user(request_queue_t *q, int rw, void __user *ubuf,
1920                                unsigned int len)
1921{
1922        unsigned long uaddr;
1923        struct request *rq;
1924        struct bio *bio;
1925
1926        if (len > (q->max_sectors << 9))
1927                return ERR_PTR(-EINVAL);
1928        if ((!len && ubuf) || (len && !ubuf))
1929                return ERR_PTR(-EINVAL);
1930
1931        rq = blk_get_request(q, rw, __GFP_WAIT);
1932        if (!rq)
1933                return ERR_PTR(-ENOMEM);
1934
1935        /*
1936         * if alignment requirement is satisfied, map in user pages for
1937         * direct dma. else, set up kernel bounce buffers
1938         */
1939        uaddr = (unsigned long) ubuf;
1940        if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
1941                bio = bio_map_user(q, NULL, uaddr, len, rw == READ);
1942        else
1943                bio = bio_copy_user(q, uaddr, len, rw == READ);
1944
1945        if (!IS_ERR(bio)) {
1946                rq->bio = rq->biotail = bio;
1947                blk_rq_bio_prep(q, rq, bio);
1948
1949                rq->buffer = rq->data = NULL;
1950                rq->data_len = len;
1951                return rq;
1952        }
1953
1954        /*
1955         * bio is the err-ptr
1956         */
1957        blk_put_request(rq);
1958        return (struct request *) bio;
1959}
1960
1961EXPORT_SYMBOL(blk_rq_map_user);
1962
1963/**
1964 * blk_rq_unmap_user - unmap a request with user data
1965 * @rq:         request to be unmapped
1966 * @ubuf:       user buffer
1967 * @ulen:       length of user buffer
1968 *
1969 * Description:
1970 *    Unmap a request previously mapped by blk_rq_map_user().
1971 */
1972int blk_rq_unmap_user(struct request *rq, struct bio *bio, unsigned int ulen)
1973{
1974        int ret = 0;
1975
1976        if (bio) {
1977                if (bio_flagged(bio, BIO_USER_MAPPED))
1978                        bio_unmap_user(bio);
1979                else
1980                        ret = bio_uncopy_user(bio);
1981        }
1982
1983        blk_put_request(rq);
1984        return ret;
1985}
1986
1987EXPORT_SYMBOL(blk_rq_unmap_user);
1988
1989/**
1990 * blk_execute_rq - insert a request into queue for execution
1991 * @q:          queue to insert the request in
1992 * @bd_disk:    matching gendisk
1993 * @rq:         request to insert
1994 *
1995 * Description:
1996 *    Insert a fully prepared request at the back of the io scheduler queue
1997 *    for execution.
1998 */
1999int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
2000                   struct request *rq)
2001{
2002        DECLARE_COMPLETION(wait);
2003        char sense[SCSI_SENSE_BUFFERSIZE];
2004        int err = 0;
2005
2006        rq->rq_disk = bd_disk;
2007
2008        /*
2009         * we need an extra reference to the request, so we can look at
2010         * it after io completion
2011         */
2012        rq->ref_count++;
2013
2014        if (!rq->sense) {
2015                memset(sense, 0, sizeof(sense));
2016                rq->sense = sense;
2017                rq->sense_len = 0;
2018        }
2019
2020        rq->flags |= REQ_NOMERGE;
2021        if (!rq->waiting)
2022                rq->waiting = &wait;
2023        elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1);
2024        generic_unplug_device(q);
2025        wait_for_completion(rq->waiting);
2026        rq->waiting = NULL;
2027
2028        if (rq->errors)
2029                err = -EIO;
2030
2031        return err;
2032}
2033
2034EXPORT_SYMBOL(blk_execute_rq);
2035
2036/**
2037 * blkdev_issue_flush - queue a flush
2038 * @bdev:       blockdev to issue flush for
2039 * @error_sector:       error sector
2040 *
2041 * Description:
2042 *    Issue a flush for the block device in question. Caller can supply
2043 *    room for storing the error offset in case of a flush error, if they
2044 *    wish to.  Caller must run wait_for_completion() on its own.
2045 */
2046int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2047{
2048        request_queue_t *q;
2049
2050        if (bdev->bd_disk == NULL)
2051                return -ENXIO;
2052
2053        q = bdev_get_queue(bdev);
2054        if (!q)
2055                return -ENXIO;
2056        if (!q->issue_flush_fn)
2057                return -EOPNOTSUPP;
2058
2059        return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2060}
2061
2062EXPORT_SYMBOL(blkdev_issue_flush);
2063
2064/**
2065 * blkdev_scsi_issue_flush_fn - issue flush for SCSI devices
2066 * @q:          device queue
2067 * @disk:       gendisk
2068 * @error_sector:       error offset
2069 *
2070 * Description:
2071 *    Devices understanding the SCSI command set, can use this function as
2072 *    a helper for issuing a cache flush. Note: driver is required to store
2073 *    the error offset (in case of error flushing) in ->sector of struct
2074 *    request.
2075 */
2076int blkdev_scsi_issue_flush_fn(request_queue_t *q, struct gendisk *disk,
2077                               sector_t *error_sector)
2078{
2079        struct request *rq = blk_get_request(q, WRITE, __GFP_WAIT);
2080        int ret;
2081
2082        rq->flags |= REQ_BLOCK_PC | REQ_SOFTBARRIER;
2083        rq->sector = 0;
2084        memset(rq->cmd, 0, sizeof(rq->cmd));
2085        rq->cmd[0] = 0x35;
2086        rq->cmd_len = 12;
2087        rq->data = NULL;
2088        rq->data_len = 0;
2089        rq->timeout = 60 * HZ;
2090
2091        ret = blk_execute_rq(q, disk, rq);
2092
2093        if (ret && error_sector)
2094                *error_sector = rq->sector;
2095
2096        blk_put_request(rq);
2097        return ret;
2098}
2099
2100EXPORT_SYMBOL(blkdev_scsi_issue_flush_fn);
2101
2102void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
2103{
2104        int rw = rq_data_dir(rq);
2105
2106        if (!blk_fs_request(rq) || !rq->rq_disk)
2107                return;
2108
2109        if (rw == READ) {
2110                __disk_stat_add(rq->rq_disk, read_sectors, nr_sectors);
2111                if (!new_io)
2112                        __disk_stat_inc(rq->rq_disk, read_merges);
2113        } else if (rw == WRITE) {
2114                __disk_stat_add(rq->rq_disk, write_sectors, nr_sectors);
2115                if (!new_io)
2116                        __disk_stat_inc(rq->rq_disk, write_merges);
2117        }
2118        if (new_io) {
2119                disk_round_stats(rq->rq_disk);
2120                rq->rq_disk->in_flight++;
2121        }
2122}
2123
2124/*
2125 * add-request adds a request to the linked list.
2126 * queue lock is held and interrupts disabled, as we muck with the
2127 * request queue list.
2128 */
2129static inline void add_request(request_queue_t * q, struct request * req)
2130{
2131        drive_stat_acct(req, req->nr_sectors, 1);
2132
2133        if (q->activity_fn)
2134                q->activity_fn(q->activity_data, rq_data_dir(req));
2135
2136        /*
2137         * elevator indicated where it wants this request to be
2138         * inserted at elevator_merge time
2139         */
2140        __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2141}
2142 
2143/*
2144 * disk_round_stats()   - Round off the performance stats on a struct
2145 * disk_stats.
2146 *
2147 * The average IO queue length and utilisation statistics are maintained
2148 * by observing the current state of the queue length and the amount of
2149 * time it has been in this state for.
2150 *
2151 * Normally, that accounting is done on IO completion, but that can result
2152 * in more than a second's worth of IO being accounted for within any one
2153 * second, leading to >100% utilisation.  To deal with that, we call this
2154 * function to do a round-off before returning the results when reading
2155 * /proc/diskstats.  This accounts immediately for all queue usage up to
2156 * the current jiffies and restarts the counters again.
2157 */
2158void disk_round_stats(struct gendisk *disk)
2159{
2160        unsigned long now = jiffies;
2161
2162        __disk_stat_add(disk, time_in_queue,
2163                        disk->in_flight * (now - disk->stamp));
2164        disk->stamp = now;
2165
2166        if (disk->in_flight)
2167                __disk_stat_add(disk, io_ticks, (now - disk->stamp_idle));
2168        disk->stamp_idle = now;
2169}
2170
2171/*
2172 * queue lock must be held
2173 */
2174void __blk_put_request(request_queue_t *q, struct request *req)
2175{
2176        struct request_list *rl = req->rl;
2177
2178        if (unlikely(!q))
2179                return;
2180        if (unlikely(--req->ref_count))
2181                return;
2182
2183        req->rq_status = RQ_INACTIVE;
2184        req->q = NULL;
2185        req->rl = NULL;
2186
2187        /*
2188         * Request may not have originated from ll_rw_blk. if not,
2189         * it didn't come out of our reserved rq pools
2190         */
2191        if (rl) {
2192                int rw = rq_data_dir(req);
2193
2194                elv_completed_request(q, req);
2195
2196                BUG_ON(!list_empty(&req->queuelist));
2197
2198                blk_free_request(q, req);
2199                freed_request(q, rw);
2200        }
2201}
2202
2203void blk_put_request(struct request *req)
2204{
2205        /*
2206         * if req->rl isn't set, this request didnt originate from the
2207         * block layer, so it's safe to just disregard it
2208         */
2209        if (req->rl) {
2210                unsigned long flags;
2211                request_queue_t *q = req->q;
2212
2213                spin_lock_irqsave(q->queue_lock, flags);
2214                __blk_put_request(q, req);
2215                spin_unlock_irqrestore(q->queue_lock, flags);
2216        }
2217}
2218
2219EXPORT_SYMBOL(blk_put_request);
2220
2221/**
2222 * blk_congestion_wait - wait for a queue to become uncongested
2223 * @rw: READ or WRITE
2224 * @timeout: timeout in jiffies
2225 *
2226 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2227 * If no queues are congested then just wait for the next request to be
2228 * returned.
2229 */
2230long blk_congestion_wait(int rw, long timeout)
2231{
2232        long ret;
2233        DEFINE_WAIT(wait);
2234        wait_queue_head_t *wqh = &congestion_wqh[rw];
2235
2236        prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2237        ret = io_schedule_timeout(timeout);
2238        finish_wait(wqh, &wait);
2239        return ret;
2240}
2241
2242EXPORT_SYMBOL(blk_congestion_wait);
2243
2244/*
2245 * Has to be called with the request spinlock acquired
2246 */
2247static int attempt_merge(request_queue_t *q, struct request *req,
2248                          struct request *next)
2249{
2250        if (!rq_mergeable(req) || !rq_mergeable(next))
2251                return 0;
2252
2253        /*
2254         * not contigious
2255         */
2256        if (req->sector + req->nr_sectors != next->sector)
2257                return 0;
2258
2259        if (rq_data_dir(req) != rq_data_dir(next)
2260            || req->rq_disk != next->rq_disk
2261            || next->waiting || next->special)
2262                return 0;
2263
2264        /*
2265         * If we are allowed to merge, then append bio list
2266         * from next to rq and release next. merge_requests_fn
2267         * will have updated segment counts, update sector
2268         * counts here.
2269         */
2270        if (!q->merge_requests_fn(q, req, next))
2271                return 0;
2272
2273        /*
2274         * At this point we have either done a back merge
2275         * or front merge. We need the smaller start_time of
2276         * the merged requests to be the current request
2277         * for accounting purposes.
2278         */
2279        if (time_after(req->start_time, next->start_time))
2280                req->start_time = next->start_time;
2281
2282        req->biotail->bi_next = next->bio;
2283        req->biotail = next->biotail;
2284
2285        req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2286
2287        elv_merge_requests(q, req, next);
2288
2289        if (req->rq_disk) {
2290                disk_round_stats(req->rq_disk);
2291                req->rq_disk->in_flight--;
2292        }
2293
2294        __blk_put_request(q, next);
2295        return 1;
2296}
2297
2298static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2299{
2300        struct request *next = elv_latter_request(q, rq);
2301
2302        if (next)
2303                return attempt_merge(q, rq, next);
2304
2305        return 0;
2306}
2307
2308static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2309{
2310        struct request *prev = elv_former_request(q, rq);
2311
2312        if (prev)
2313                return attempt_merge(q, prev, rq);
2314
2315        return 0;
2316}
2317
2318/**
2319 * blk_attempt_remerge  - attempt to remerge active head with next request
2320 * @q:    The &request_queue_t belonging to the device
2321 * @rq:   The head request (usually)
2322 *
2323 * Description:
2324 *    For head-active devices, the queue can easily be unplugged so quickly
2325 *    that proper merging is not done on the front request. This may hurt
2326 *    performance greatly for some devices. The block layer cannot safely
2327 *    do merging on that first request for these queues, but the driver can
2328 *    call this function and make it happen any way. Only the driver knows
2329 *    when it is safe to do so.
2330 **/
2331void blk_attempt_remerge(request_queue_t *q, struct request *rq)
2332{
2333        unsigned long flags;
2334
2335        spin_lock_irqsave(q->queue_lock, flags);
2336        attempt_back_merge(q, rq);
2337        spin_unlock_irqrestore(q->queue_lock, flags);
2338}
2339
2340EXPORT_SYMBOL(blk_attempt_remerge);
2341
2342/*
2343 * Non-locking blk_attempt_remerge variant.
2344 */
2345void __blk_attempt_remerge(request_queue_t *q, struct request *rq)
2346{
2347        attempt_back_merge(q, rq);
2348}
2349
2350EXPORT_SYMBOL(__blk_attempt_remerge);
2351
2352static int __make_request(request_queue_t *q, struct bio *bio)
2353{
2354        struct request *req, *freereq = NULL;
2355        int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err;
2356        sector_t sector;
2357
2358        sector = bio->bi_sector;
2359        nr_sectors = bio_sectors(bio);
2360        cur_nr_sectors = bio_cur_sectors(bio);
2361
2362        rw = bio_data_dir(bio);
2363
2364        /*
2365         * low level driver can indicate that it wants pages above a
2366         * certain limit bounced to low memory (ie for highmem, or even
2367         * ISA dma in theory)
2368         */
2369        blk_queue_bounce(q, &bio);
2370
2371        spin_lock_prefetch(q->queue_lock);
2372
2373        barrier = bio_barrier(bio);
2374        if (barrier && !(q->queue_flags & (1 << QUEUE_FLAG_ORDERED))) {
2375                err = -EOPNOTSUPP;
2376                goto end_io;
2377        }
2378
2379again:
2380        spin_lock_irq(q->queue_lock);
2381
2382        if (elv_queue_empty(q)) {
2383                blk_plug_device(q);
2384                goto get_rq;
2385        }
2386        if (barrier)
2387                goto get_rq;
2388
2389        el_ret = elv_merge(q, &req, bio);
2390        switch (el_ret) {
2391                case ELEVATOR_BACK_MERGE:
2392                        BUG_ON(!rq_mergeable(req));
2393
2394                        if (!q->back_merge_fn(q, req, bio))
2395                                break;
2396
2397                        req->biotail->bi_next = bio;
2398                        req->biotail = bio;
2399                        req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2400                        drive_stat_acct(req, nr_sectors, 0);
2401                        if (!attempt_back_merge(q, req))
2402                                elv_merged_request(q, req);
2403                        goto out;
2404
2405                case ELEVATOR_FRONT_MERGE:
2406                        BUG_ON(!rq_mergeable(req));
2407
2408                        if (!q->front_merge_fn(q, req, bio))
2409                                break;
2410
2411                        bio->bi_next = req->bio;
2412                        req->bio = bio;
2413
2414                        /*
2415                         * may not be valid. if the low level driver said
2416                         * it didn't need a bounce buffer then it better
2417                         * not touch req->buffer either...
2418                         */
2419                        req->buffer = bio_data(bio);
2420                        req->current_nr_sectors = cur_nr_sectors;
2421                        req->hard_cur_sectors = cur_nr_sectors;
2422                        req->sector = req->hard_sector = sector;
2423                        req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2424                        drive_stat_acct(req, nr_sectors, 0);
2425                        if (!attempt_front_merge(q, req))
2426                                elv_merged_request(q, req);
2427                        goto out;
2428
2429                /*
2430                 * elevator says don't/can't merge. get new request
2431                 */
2432                case ELEVATOR_NO_MERGE:
2433                        break;
2434
2435                default:
2436                        printk("elevator returned crap (%d)\n", el_ret);
2437                        BUG();
2438        }
2439
2440        /*
2441         * Grab a free request from the freelist - if that is empty, check
2442         * if we are doing read ahead and abort instead of blocking for
2443         * a free slot.
2444         */
2445get_rq:
2446        if (freereq) {
2447                req = freereq;
2448                freereq = NULL;
2449        } else {
2450                spin_unlock_irq(q->queue_lock);
2451                if ((freereq = get_request(q, rw, GFP_ATOMIC)) == NULL) {
2452                        /*
2453                         * READA bit set
2454                         */
2455                        err = -EWOULDBLOCK;
2456                        if (bio_rw_ahead(bio))
2457                                goto end_io;
2458        
2459                        freereq = get_request_wait(q, rw);
2460                }
2461                goto again;
2462        }
2463
2464        req->flags |= REQ_CMD;
2465
2466        /*
2467         * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2468         */
2469        if (bio_rw_ahead(bio) || bio_failfast(bio))
2470                req->flags |= REQ_FAILFAST;
2471
2472        /*
2473         * REQ_BARRIER implies no merging, but lets make it explicit
2474         */
2475        if (barrier)
2476                req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2477
2478        req->errors = 0;
2479        req->hard_sector = req->sector = sector;
2480        req->hard_nr_sectors = req->nr_sectors = nr_sectors;
2481        req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors;
2482        req->nr_phys_segments = bio_phys_segments(q, bio);
2483        req->nr_hw_segments = bio_hw_segments(q, bio);
2484        req->buffer = bio_data(bio);    /* see ->buffer comment above */
2485        req->waiting = NULL;
2486        req->bio = req->biotail = bio;
2487        req->rq_disk = bio->bi_bdev->bd_disk;
2488        req->start_time = jiffies;
2489
2490        add_request(q, req);
2491out:
2492        if (freereq)
2493                __blk_put_request(q, freereq);
2494        if (bio_sync(bio))
2495                __generic_unplug_device(q);
2496
2497        spin_unlock_irq(q->queue_lock);
2498        return 0;
2499
2500end_io:
2501        bio_endio(bio, nr_sectors << 9, err);
2502        return 0;
2503}
2504
2505/*
2506 * If bio->bi_dev is a partition, remap the location
2507 */
2508static inline void blk_partition_remap(struct bio *bio)
2509{
2510        struct block_device *bdev = bio->bi_bdev;
2511
2512        if (bdev != bdev->bd_contains) {
2513                struct hd_struct *p = bdev->bd_part;
2514
2515                switch (bio->bi_rw) {
2516                case READ:
2517                        p->read_sectors += bio_sectors(bio);
2518                        p->reads++;
2519                        break;
2520                case WRITE:
2521                        p->write_sectors += bio_sectors(bio);
2522                        p->writes++;
2523                        break;
2524                }
2525                bio->bi_sector += p->start_sect;
2526                bio->bi_bdev = bdev->bd_contains;
2527        }
2528}
2529
2530void blk_finish_queue_drain(request_queue_t *q)
2531{
2532        struct request_list *rl = &q->rq;
2533        struct request *rq;
2534
2535        spin_lock_irq(q->queue_lock);
2536        clear_bit(QUEUE_FLAG_DRAIN, &q->queue_flags);
2537
2538        while (!list_empty(&q->drain_list)) {
2539                rq = list_entry_rq(q->drain_list.next);
2540
2541                list_del_init(&rq->queuelist);
2542                __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1);
2543        }
2544
2545        spin_unlock_irq(q->queue_lock);
2546
2547        wake_up(&rl->wait[0]);
2548        wake_up(&rl->wait[1]);
2549        wake_up(&rl->drain);
2550}
2551
2552static int wait_drain(request_queue_t *q, struct request_list *rl, int dispatch)
2553{
2554        int wait = rl->count[READ] + rl->count[WRITE];
2555
2556        if (dispatch)
2557                wait += !list_empty(&q->queue_head);
2558
2559        return wait;
2560}
2561
2562/*
2563 * We rely on the fact that only requests allocated through blk_alloc_request()
2564 * have io scheduler private data structures associated with them. Any other
2565 * type of request (allocated on stack or through kmalloc()) should not go
2566 * to the io scheduler core, but be attached to the queue head instead.
2567 */
2568void blk_wait_queue_drained(request_queue_t *q, int wait_dispatch)
2569{
2570        struct request_list *rl = &q->rq;
2571        DEFINE_WAIT(wait);
2572
2573        spin_lock_irq(q->queue_lock);
2574        set_bit(QUEUE_FLAG_DRAIN, &q->queue_flags);
2575
2576        while (wait_drain(q, rl, wait_dispatch)) {
2577                prepare_to_wait(&rl->drain, &wait, TASK_UNINTERRUPTIBLE);
2578
2579                if (wait_drain(q, rl, wait_dispatch)) {
2580                        __generic_unplug_device(q);
2581                        spin_unlock_irq(q->queue_lock);
2582                        io_schedule();
2583                        spin_lock_irq(q->queue_lock);
2584                }
2585
2586                finish_wait(&rl->drain, &wait);
2587        }
2588
2589        spin_unlock_irq(q->queue_lock);
2590}
2591
2592/*
2593 * block waiting for the io scheduler being started again.
2594 */
2595static inline void block_wait_queue_running(request_queue_t *q)
2596{
2597        DEFINE_WAIT(wait);
2598
2599        while (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)) {
2600                struct request_list *rl = &q->rq;
2601
2602                prepare_to_wait_exclusive(&rl->drain, &wait,
2603                                TASK_UNINTERRUPTIBLE);
2604
2605                /*
2606                 * re-check the condition. avoids using prepare_to_wait()
2607                 * in the fast path (queue is running)
2608                 */
2609                if (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))
2610                        io_schedule();
2611
2612                finish_wait(&rl->drain, &wait);
2613        }
2614}
2615
2616static void handle_bad_sector(struct bio *bio)
2617{
2618        char b[BDEVNAME_SIZE];
2619
2620        printk(KERN_INFO "attempt to access beyond end of device\n");
2621        printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2622                        bdevname(bio->bi_bdev, b),
2623                        bio->bi_rw,
2624                        (unsigned long long)bio->bi_sector + bio_sectors(bio),
2625                        (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
2626
2627        set_bit(BIO_EOF, &bio->bi_flags);
2628}
2629
2630/**
2631 * generic_make_request: hand a buffer to its device driver for I/O
2632 * @bio:  The bio describing the location in memory and on the device.
2633 *
2634 * generic_make_request() is used to make I/O requests of block
2635 * devices. It is passed a &struct bio, which describes the I/O that needs
2636 * to be done.
2637 *
2638 * generic_make_request() does not return any status.  The
2639 * success/failure status of the request, along with notification of
2640 * completion, is delivered asynchronously through the bio->bi_end_io
2641 * function described (one day) else where.
2642 *
2643 * The caller of generic_make_request must make sure that bi_io_vec
2644 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2645 * set to describe the device address, and the
2646 * bi_end_io and optionally bi_private are set to describe how
2647 * completion notification should be signaled.
2648 *
2649 * generic_make_request and the drivers it calls may use bi_next if this
2650 * bio happens to be merged with someone else, and may change bi_dev and
2651 * bi_sector for remaps as it sees fit.  So the values of these fields
2652 * should NOT be depended on after the call to generic_make_request.
2653 */
2654void generic_make_request(struct bio *bio)
2655{
2656        request_queue_t *q;
2657        sector_t maxsector;
2658        int ret, nr_sectors = bio_sectors(bio);
2659
2660        might_sleep();
2661        /* Test device or partition size, when known. */
2662        maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
2663        if (maxsector) {
2664                sector_t sector = bio->bi_sector;
2665
2666                if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2667                        /*
2668                         * This may well happen - the kernel calls bread()
2669                         * without checking the size of the device, e.g., when
2670                         * mounting a device.
2671                         */
2672                        handle_bad_sector(bio);
2673                        goto end_io;
2674                }
2675        }
2676
2677        /*
2678         * Resolve the mapping until finished. (drivers are
2679         * still free to implement/resolve their own stacking
2680         * by explicitly returning 0)
2681         *
2682         * NOTE: we don't repeat the blk_size check for each new device.
2683         * Stacking drivers are expected to know what they are doing.
2684         */
2685        do {
2686                char b[BDEVNAME_SIZE];
2687
2688                q = bdev_get_queue(bio->bi_bdev);
2689                if (!q) {
2690                        printk(KERN_ERR
2691                               "generic_make_request: Trying to access "
2692                                "nonexistent block-device %s (%Lu)\n",
2693                                bdevname(bio->bi_bdev, b),
2694                                (long long) bio->bi_sector);
2695end_io:
2696                        bio_endio(bio, bio->bi_size, -EIO);
2697                        break;
2698                }
2699
2700                if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
2701                        printk("bio too big device %s (%u > %u)\n", 
2702                                bdevname(bio->bi_bdev, b),
2703                                bio_sectors(bio),
2704                                q->max_hw_sectors);
2705                        goto end_io;
2706                }
2707
2708                if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))
2709                        goto end_io;
2710
2711                block_wait_queue_running(q);
2712
2713                /*
2714                 * If this device has partitions, remap block n
2715                 * of partition p to block n+start(p) of the disk.
2716                 */
2717                blk_partition_remap(bio);
2718
2719                ret = q->make_request_fn(q, bio);
2720        } while (ret);
2721}
2722
2723EXPORT_SYMBOL(generic_make_request);
2724
2725/**
2726 * submit_bio: submit a bio to the block device layer for I/O
2727 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2728 * @bio: The &struct bio which describes the I/O
2729 *
2730 * submit_bio() is very similar in purpose to generic_make_request(), and
2731 * uses that function to do most of the work. Both are fairly rough
2732 * interfaces, @bio must be presetup and ready for I/O.
2733 *
2734 */
2735void submit_bio(int rw, struct bio *bio)
2736{
2737        int count = bio_sectors(bio);
2738
2739        BIO_BUG_ON(!bio->bi_size);
2740        BIO_BUG_ON(!bio->bi_io_vec);
2741        bio->bi_rw = rw;
2742        if (rw & WRITE)
2743                mod_page_state(pgpgout, count);
2744        else
2745                mod_page_state(pgpgin, count);
2746
2747        if (unlikely(block_dump)) {
2748                char b[BDEVNAME_SIZE];
2749                printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
2750                        current->comm, current->pid,
2751                        (rw & WRITE) ? "WRITE" : "READ",
2752                        (unsigned long long)bio->bi_sector,
2753                        bdevname(bio->bi_bdev,b));
2754        }
2755
2756        generic_make_request(bio);
2757}
2758
2759EXPORT_SYMBOL(submit_bio);
2760
2761void blk_recalc_rq_segments(struct request *rq)
2762{
2763        struct bio *bio, *prevbio = NULL;
2764        int nr_phys_segs, nr_hw_segs;
2765        unsigned int phys_size, hw_size;
2766        request_queue_t *q = rq->q;
2767
2768        if (!rq->bio)
2769                return;
2770
2771        phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
2772        rq_for_each_bio(bio, rq) {
2773                /* Force bio hw/phys segs to be recalculated. */
2774                bio->bi_flags &= ~(1 << BIO_SEG_VALID);
2775
2776                nr_phys_segs += bio_phys_segments(q, bio);
2777                nr_hw_segs += bio_hw_segments(q, bio);
2778                if (prevbio) {
2779                        int pseg = phys_size + prevbio->bi_size + bio->bi_size;
2780                        int hseg = hw_size + prevbio->bi_size + bio->bi_size;
2781
2782                        if (blk_phys_contig_segment(q, prevbio, bio) &&
2783                            pseg <= q->max_segment_size) {
2784                                nr_phys_segs--;
2785                                phys_size += prevbio->bi_size + bio->bi_size;
2786                        } else
2787                                phys_size = 0;
2788
2789                        if (blk_hw_contig_segment(q, prevbio, bio) &&
2790                            hseg <= q->max_segment_size) {
2791                                nr_hw_segs--;
2792                                hw_size += prevbio->bi_size + bio->bi_size;
2793                        } else
2794                                hw_size = 0;
2795                }
2796                prevbio = bio;
2797        }
2798
2799        rq->nr_phys_segments = nr_phys_segs;
2800        rq->nr_hw_segments = nr_hw_segs;
2801}
2802
2803void blk_recalc_rq_sectors(struct request *rq, int nsect)
2804{
2805        if (blk_fs_request(rq)) {
2806                rq->hard_sector += nsect;
2807                rq->hard_nr_sectors -= nsect;
2808
2809                /*
2810                 * Move the I/O submission pointers ahead if required.
2811                 */
2812                if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
2813                    (rq->sector <= rq->hard_sector)) {
2814                        rq->sector = rq->hard_sector;
2815                        rq->nr_sectors = rq->hard_nr_sectors;
2816                        rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
2817                        rq->current_nr_sectors = rq->hard_cur_sectors;
2818                        rq->buffer = bio_data(rq->bio);
2819                }
2820
2821                /*
2822                 * if total number of sectors is less than the first segment
2823                 * size, something has gone terribly wrong
2824                 */
2825                if (rq->nr_sectors < rq->current_nr_sectors) {
2826                        printk("blk: request botched\n");
2827                        rq->nr_sectors = rq->current_nr_sectors;
2828                }
2829        }
2830}
2831
2832static int __end_that_request_first(struct request *req, int uptodate,
2833                                    int nr_bytes)
2834{
2835        int total_bytes, bio_nbytes, error, next_idx = 0;
2836        struct bio *bio;
2837
2838        /*
2839         * extend uptodate bool to allow < 0 value to be direct io error
2840         */
2841        error = 0;
2842        if (end_io_error(uptodate))
2843                error = !uptodate ? -EIO : uptodate;
2844
2845        /*
2846         * for a REQ_BLOCK_PC request, we want to carry any eventual
2847         * sense key with us all the way through
2848         */
2849        if (!blk_pc_request(req))
2850                req->errors = 0;
2851
2852        if (!uptodate) {
2853                if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
2854                        printk("end_request: I/O error, dev %s, sector %llu\n",
2855                                req->rq_disk ? req->rq_disk->disk_name : "?",
2856                                (unsigned long long)req->sector);
2857        }
2858
2859        total_bytes = bio_nbytes = 0;
2860        while ((bio = req->bio) != NULL) {
2861                int nbytes;
2862
2863                if (nr_bytes >= bio->bi_size) {
2864                        req->bio = bio->bi_next;
2865                        nbytes = bio->bi_size;
2866                        bio_endio(bio, nbytes, error);
2867                        next_idx = 0;
2868                        bio_nbytes = 0;
2869                } else {
2870                        int idx = bio->bi_idx + next_idx;
2871
2872                        if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
2873                                blk_dump_rq_flags(req, "__end_that");
2874                                printk("%s: bio idx %d >= vcnt %d\n",
2875                                                __FUNCTION__,
2876                                                bio->bi_idx, bio->bi_vcnt);
2877                                break;
2878                        }
2879
2880                        nbytes = bio_iovec_idx(bio, idx)->bv_len;
2881                        BIO_BUG_ON(nbytes > bio->bi_size);
2882
2883                        /*
2884                         * not a complete bvec done
2885                         */
2886                        if (unlikely(nbytes > nr_bytes)) {
2887                                bio_nbytes += nr_bytes;
2888                                total_bytes += nr_bytes;
2889                                break;
2890                        }
2891
2892                        /*
2893                         * advance to the next vector
2894                         */
2895                        next_idx++;
2896                        bio_nbytes += nbytes;
2897                }
2898
2899                total_bytes += nbytes;
2900                nr_bytes -= nbytes;
2901
2902                if ((bio = req->bio)) {
2903                        /*
2904                         * end more in this run, or just return 'not-done'
2905                         */
2906                        if (unlikely(nr_bytes <= 0))
2907                                break;
2908                }
2909        }
2910
2911        /*
2912         * completely done
2913         */
2914        if (!req->bio)
2915                return 0;
2916
2917        /*
2918         * if the request wasn't completed, update state
2919         */
2920        if (bio_nbytes) {
2921                bio_endio(bio, bio_nbytes, error);
2922                bio->bi_idx += next_idx;
2923                bio_iovec(bio)->bv_offset += nr_bytes;
2924                bio_iovec(bio)->bv_len -= nr_bytes;
2925        }
2926
2927        blk_recalc_rq_sectors(req, total_bytes >> 9);
2928        blk_recalc_rq_segments(req);
2929        return 1;
2930}
2931
2932/**
2933 * end_that_request_first - end I/O on a request
2934 * @req:      the request being processed
2935 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
2936 * @nr_sectors: number of sectors to end I/O on
2937 *
2938 * Description:
2939 *     Ends I/O on a number of sectors attached to @req, and sets it up
2940 *     for the next range of segments (if any) in the cluster.
2941 *
2942 * Return:
2943 *     0 - we are done with this request, call end_that_request_last()
2944 *     1 - still buffers pending for this request
2945 **/
2946int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
2947{
2948        return __end_that_request_first(req, uptodate, nr_sectors << 9);
2949}
2950
2951EXPORT_SYMBOL(end_that_request_first);
2952
2953/**
2954 * end_that_request_chunk - end I/O on a request
2955 * @req:      the request being processed
2956 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
2957 * @nr_bytes: number of bytes to complete
2958 *
2959 * Description:
2960 *     Ends I/O on a number of bytes attached to @req, and sets it up
2961 *     for the next range of segments (if any). Like end_that_request_first(),
2962 *     but deals with bytes instead of sectors.
2963 *
2964 * Return:
2965 *     0 - we are done with this request, call end_that_request_last()
2966 *     1 - still buffers pending for this request
2967 **/
2968int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
2969{
2970        return __end_that_request_first(req, uptodate, nr_bytes);
2971}
2972
2973EXPORT_SYMBOL(end_that_request_chunk);
2974
2975/*
2976 * queue lock must be held
2977 */
2978void end_that_request_last(struct request *req)
2979{
2980        struct gendisk *disk = req->rq_disk;
2981        struct completion *waiting = req->waiting;
2982
2983        if (unlikely(laptop_mode) && blk_fs_request(req))
2984                laptop_io_completion();
2985
2986        if (disk && blk_fs_request(req)) {
2987                unsigned long duration = jiffies - req->start_time;
2988                switch (rq_data_dir(req)) {
2989                    case WRITE:
2990                        __disk_stat_inc(disk, writes);
2991                        __disk_stat_add(disk, write_ticks, duration);
2992                        break;
2993                    case READ:
2994                        __disk_stat_inc(disk, reads);
2995                        __disk_stat_add(disk, read_ticks, duration);
2996                        break;
2997                }
2998                disk_round_stats(disk);
2999                disk->in_flight--;
3000        }
3001        __blk_put_request(req->q, req);
3002        /* Do this LAST! The structure may be freed immediately afterwards */
3003        if (waiting)
3004                complete(waiting);
3005}
3006
3007EXPORT_SYMBOL(end_that_request_last);
3008
3009void end_request(struct request *req, int uptodate)
3010{
3011        if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3012                add_disk_randomness(req->rq_disk);
3013                blkdev_dequeue_request(req);
3014                end_that_request_last(req);
3015        }
3016}
3017
3018EXPORT_SYMBOL(end_request);
3019
3020void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3021{
3022        /* first three bits are identical in rq->flags and bio->bi_rw */
3023        rq->flags |= (bio->bi_rw & 7);
3024
3025        rq->nr_phys_segments = bio_phys_segments(q, bio);
3026        rq->nr_hw_segments = bio_hw_segments(q, bio);
3027        rq->current_nr_sectors = bio_cur_sectors(bio);
3028        rq->hard_cur_sectors = rq->current_nr_sectors;
3029        rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3030        rq->buffer = bio_data(bio);
3031
3032        rq->bio = rq->biotail = bio;
3033}
3034
3035EXPORT_SYMBOL(blk_rq_bio_prep);
3036
3037int kblockd_schedule_work(struct work_struct *work)
3038{
3039        return queue_work(kblockd_workqueue, work);
3040}
3041
3042EXPORT_SYMBOL(kblockd_schedule_work);
3043
3044void kblockd_flush(void)
3045{
3046        flush_workqueue(kblockd_workqueue);
3047}
3048EXPORT_SYMBOL(kblockd_flush);
3049
3050int __init blk_dev_init(void)
3051{
3052        kblockd_workqueue = create_workqueue("kblockd");
3053        if (!kblockd_workqueue)
3054                panic("Failed to create kblockd\n");
3055
3056        request_cachep = kmem_cache_create("blkdev_requests",
3057                        sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3058
3059        requestq_cachep = kmem_cache_create("blkdev_queue",
3060                        sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3061
3062        iocontext_cachep = kmem_cache_create("blkdev_ioc",
3063                        sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3064
3065        blk_max_low_pfn = max_low_pfn;
3066        blk_max_pfn = max_pfn;
3067
3068        return 0;
3069}
3070
3071/*
3072 * IO Context helper functions
3073 */
3074void put_io_context(struct io_context *ioc)
3075{
3076        if (ioc == NULL)
3077                return;
3078
3079        BUG_ON(atomic_read(&ioc->refcount) == 0);
3080
3081        if (atomic_dec_and_test(&ioc->refcount)) {
3082                if (ioc->aic && ioc->aic->dtor)
3083                        ioc->aic->dtor(ioc->aic);
3084                if (ioc->cic && ioc->cic->dtor)
3085                        ioc->cic->dtor(ioc->cic);
3086
3087                kmem_cache_free(iocontext_cachep, ioc);
3088        }
3089}
3090EXPORT_SYMBOL(put_io_context);
3091
3092/* Called by the exitting task */
3093void exit_io_context(void)
3094{
3095        unsigned long flags;
3096        struct io_context *ioc;
3097
3098        local_irq_save(flags);
3099        ioc = current->io_context;
3100        current->io_context = NULL;
3101        local_irq_restore(flags);
3102
3103        if (ioc->aic && ioc->aic->exit)
3104                ioc->aic->exit(ioc->aic);
3105        if (ioc->cic && ioc->cic->exit)
3106                ioc->cic->exit(ioc->cic);
3107
3108        put_io_context(ioc);
3109}
3110
3111/*
3112 * If the current task has no IO context then create one and initialise it.
3113 * If it does have a context, take a ref on it.
3114 *
3115 * This is always called in the context of the task which submitted the I/O.
3116 * But weird things happen, so we disable local interrupts to ensure exclusive
3117 * access to *current.
3118 */
3119struct io_context *get_io_context(int gfp_flags)
3120{
3121        struct task_struct *tsk = current;
3122        unsigned long flags;
3123        struct io_context *ret;
3124
3125        local_irq_save(flags);
3126        ret = tsk->io_context;
3127        if (ret)
3128                goto out;
3129
3130        local_irq_restore(flags);
3131
3132        ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3133        if (ret) {
3134                atomic_set(&ret->refcount, 1);
3135                ret->pid = tsk->pid;
3136                ret->last_waited = jiffies; /* doesn't matter... */
3137                ret->nr_batch_requests = 0; /* because this is 0 */
3138                ret->aic = NULL;
3139                ret->cic = NULL;
3140                spin_lock_init(&ret->lock);
3141
3142                local_irq_save(flags);
3143
3144                /*
3145                 * very unlikely, someone raced with us in setting up the task
3146                 * io context. free new context and just grab a reference.
3147                 */
3148                if (!tsk->io_context)
3149                        tsk->io_context = ret;
3150                else {
3151                        kmem_cache_free(iocontext_cachep, ret);
3152                        ret = tsk->io_context;
3153                }
3154
3155out:
3156                atomic_inc(&ret->refcount);
3157                local_irq_restore(flags);
3158        }
3159
3160        return ret;
3161}
3162EXPORT_SYMBOL(get_io_context);
3163
3164void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3165{
3166        struct io_context *src = *psrc;
3167        struct io_context *dst = *pdst;
3168
3169        if (src) {
3170                BUG_ON(atomic_read(&src->refcount) == 0);
3171                atomic_inc(&src->refcount);
3172                put_io_context(dst);
3173                *pdst = src;
3174        }
3175}
3176EXPORT_SYMBOL(copy_io_context);
3177
3178void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3179{
3180        struct io_context *temp;
3181        temp = *ioc1;
3182        *ioc1 = *ioc2;
3183        *ioc2 = temp;
3184}
3185EXPORT_SYMBOL(swap_io_context);
3186
3187/*
3188 * sysfs parts below
3189 */
3190struct queue_sysfs_entry {
3191        struct attribute attr;
3192        ssize_t (*show)(struct request_queue *, char *);
3193        ssize_t (*store)(struct request_queue *, const char *, size_t);
3194};
3195
3196static ssize_t
3197queue_var_show(unsigned int var, char *page)
3198{
3199        return sprintf(page, "%d\n", var);
3200}
3201
3202static ssize_t
3203queue_var_store(unsigned long *var, const char *page, size_t count)
3204{
3205        char *p = (char *) page;
3206
3207        *var = simple_strtoul(p, &p, 10);
3208        return count;
3209}
3210
3211static ssize_t queue_requests_show(struct request_queue *q, char *page)
3212{
3213        return queue_var_show(q->nr_requests, (page));
3214}
3215
3216static ssize_t
3217queue_requests_store(struct request_queue *q, const char *page, size_t count)
3218{
3219        struct request_list *rl = &q->rq;
3220
3221        int ret = queue_var_store(&q->nr_requests, page, count);
3222        if (q->nr_requests < BLKDEV_MIN_RQ)
3223                q->nr_requests = BLKDEV_MIN_RQ;
3224        blk_queue_congestion_threshold(q);
3225
3226        if (rl->count[READ] >= queue_congestion_on_threshold(q))
3227                set_queue_congested(q, READ);
3228        else if (rl->count[READ] < queue_congestion_off_threshold(q))
3229                clear_queue_congested(q, READ);
3230
3231        if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3232                set_queue_congested(q, WRITE);
3233        else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3234                clear_queue_congested(q, WRITE);
3235
3236        if (rl->count[READ] >= q->nr_requests) {
3237                blk_set_queue_full(q, READ);
3238        } else if (rl->count[READ]+1 <= q->nr_requests) {
3239                blk_clear_queue_full(q, READ);
3240                wake_up(&rl->wait[READ]);
3241        }
3242
3243        if (rl->count[WRITE] >= q->nr_requests) {
3244                blk_set_queue_full(q, WRITE);
3245        } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3246                blk_clear_queue_full(q, WRITE);
3247                wake_up(&rl->wait[WRITE]);
3248        }
3249        return ret;
3250}
3251
3252static ssize_t queue_ra_show(struct request_queue *q, char *page)
3253{
3254        int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3255
3256        return queue_var_show(ra_kb, (page));
3257}
3258
3259static ssize_t
3260queue_ra_store(struct request_queue *q, const char *page, size_t count)
3261{
3262        unsigned long ra_kb;
3263        ssize_t ret = queue_var_store(&ra_kb, page, count);
3264
3265        spin_lock_irq(q->queue_lock);
3266        if (ra_kb > (q->max_sectors >> 1))
3267                ra_kb = (q->max_sectors >> 1);
3268
3269        q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3270        spin_unlock_irq(q->queue_lock);
3271
3272        return ret;
3273}
3274
3275static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3276{
3277        int max_sectors_kb = q->max_sectors >> 1;
3278
3279        return queue_var_show(max_sectors_kb, (page));
3280}
3281
3282static ssize_t
3283queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3284{
3285        unsigned long max_sectors_kb,
3286                        max_hw_sectors_kb = q->max_hw_sectors >> 1,
3287                        page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3288        ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3289        int ra_kb;
3290
3291        if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3292                return -EINVAL;
3293        /*
3294         * Take the queue lock to update the readahead and max_sectors
3295         * values synchronously:
3296         */
3297        spin_lock_irq(q->queue_lock);
3298        /*
3299         * Trim readahead window as well, if necessary:
3300         */
3301        ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3302        if (ra_kb > max_sectors_kb)
3303                q->backing_dev_info.ra_pages =
3304                                max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3305
3306        q->max_sectors = max_sectors_kb << 1;
3307        spin_unlock_irq(q->queue_lock);
3308
3309        return ret;
3310}
3311
3312static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3313{
3314        int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3315
3316        return queue_var_show(max_hw_sectors_kb, (page));
3317}
3318
3319
3320static struct queue_sysfs_entry queue_requests_entry = {
3321        .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3322        .show = queue_requests_show,
3323        .store = queue_requests_store,
3324};
3325
3326static struct queue_sysfs_entry queue_ra_entry = {
3327        .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3328        .show = queue_ra_show,
3329        .store = queue_ra_store,
3330};
3331
3332static struct queue_sysfs_entry queue_max_sectors_entry = {
3333        .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3334        .show = queue_max_sectors_show,
3335        .store = queue_max_sectors_store,
3336};
3337
3338static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3339        .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3340        .show = queue_max_hw_sectors_show,
3341};
3342
3343static struct queue_sysfs_entry queue_iosched_entry = {
3344        .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3345        .show = elv_iosched_show,
3346        .store = elv_iosched_store,
3347};
3348
3349static struct attribute *default_attrs[] = {
3350        &queue_requests_entry.attr,
3351        &queue_ra_entry.attr,
3352        &queue_max_hw_sectors_entry.attr,
3353        &queue_max_sectors_entry.attr,
3354        &queue_iosched_entry.attr,
3355        NULL,
3356};
3357
3358#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3359
3360static ssize_t
3361queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3362{
3363        struct queue_sysfs_entry *entry = to_queue(attr);
3364        struct request_queue *q;
3365
3366        q = container_of(kobj, struct request_queue, kobj);
3367        if (!entry->show)
3368                return 0;
3369
3370        return entry->show(q, page);
3371}
3372
3373static ssize_t
3374queue_attr_store(struct kobject *kobj, struct attribute *attr,
3375                    const char *page, size_t length)
3376{
3377        struct queue_sysfs_entry *entry = to_queue(attr);
3378        struct request_queue *q;
3379
3380        q = container_of(kobj, struct request_queue, kobj);
3381        if (!entry->store)
3382                return -EINVAL;
3383
3384        return entry->store(q, page, length);
3385}
3386
3387static struct sysfs_ops queue_sysfs_ops = {
3388        .show   = queue_attr_show,
3389        .store  = queue_attr_store,
3390};
3391
3392struct kobj_type queue_ktype = {
3393        .sysfs_ops      = &queue_sysfs_ops,
3394        .default_attrs  = default_attrs,
3395};
3396
3397int blk_register_queue(struct gendisk *disk)
3398{
3399        int ret;
3400
3401        request_queue_t *q = disk->queue;
3402
3403        if (!q || !q->request_fn)
3404                return -ENXIO;
3405
3406        q->kobj.parent = kobject_get(&disk->kobj);
3407        if (!q->kobj.parent)
3408                return -EBUSY;
3409
3410        snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
3411        q->kobj.ktype = &queue_ktype;
3412
3413        ret = kobject_register(&q->kobj);
3414        if (ret < 0)
3415                return ret;
3416
3417        ret = elv_register_queue(q);
3418        if (ret) {
3419                kobject_unregister(&q->kobj);
3420                return ret;
3421        }
3422
3423        return 0;
3424}
3425
3426void blk_unregister_queue(struct gendisk *disk)
3427{
3428        request_queue_t *q = disk->queue;
3429
3430        if (q && q->request_fn) {
3431                elv_unregister_queue(q);
3432
3433                kobject_unregister(&q->kobj);
3434                kobject_put(&disk->kobj);
3435        }
3436}
3437
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.