linux/fs/bio.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
   3 *
   4 * This program is free software; you can redistribute it and/or modify
   5 * it under the terms of the GNU General Public License version 2 as
   6 * published by the Free Software Foundation.
   7 *
   8 * This program is distributed in the hope that it will be useful,
   9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11 * GNU General Public License for more details.
  12 *
  13 * You should have received a copy of the GNU General Public Licens
  14 * along with this program; if not, write to the Free Software
  15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
  16 *
  17 */
  18#include <linux/mm.h>
  19#include <linux/swap.h>
  20#include <linux/bio.h>
  21#include <linux/blkdev.h>
  22#include <linux/slab.h>
  23#include <linux/init.h>
  24#include <linux/kernel.h>
  25#include <linux/module.h>
  26#include <linux/mempool.h>
  27#include <linux/workqueue.h>
  28#include <scsi/sg.h>            /* for struct sg_iovec */
  29
  30#include <trace/events/block.h>
  31
  32/*
  33 * Test patch to inline a certain number of bi_io_vec's inside the bio
  34 * itself, to shrink a bio data allocation from two mempool calls to one
  35 */
  36#define BIO_INLINE_VECS         4
  37
  38static mempool_t *bio_split_pool __read_mostly;
  39
  40/*
  41 * if you change this list, also change bvec_alloc or things will
  42 * break badly! cannot be bigger than what you can fit into an
  43 * unsigned short
  44 */
  45#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
  46static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
  47        BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
  48};
  49#undef BV
  50
  51/*
  52 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
  53 * IO code that does not need private memory pools.
  54 */
  55struct bio_set *fs_bio_set;
  56
  57/*
  58 * Our slab pool management
  59 */
  60struct bio_slab {
  61        struct kmem_cache *slab;
  62        unsigned int slab_ref;
  63        unsigned int slab_size;
  64        char name[8];
  65};
  66static DEFINE_MUTEX(bio_slab_lock);
  67static struct bio_slab *bio_slabs;
  68static unsigned int bio_slab_nr, bio_slab_max;
  69
  70static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
  71{
  72        unsigned int sz = sizeof(struct bio) + extra_size;
  73        struct kmem_cache *slab = NULL;
  74        struct bio_slab *bslab;
  75        unsigned int i, entry = -1;
  76
  77        mutex_lock(&bio_slab_lock);
  78
  79        i = 0;
  80        while (i < bio_slab_nr) {
  81                bslab = &bio_slabs[i];
  82
  83                if (!bslab->slab && entry == -1)
  84                        entry = i;
  85                else if (bslab->slab_size == sz) {
  86                        slab = bslab->slab;
  87                        bslab->slab_ref++;
  88                        break;
  89                }
  90                i++;
  91        }
  92
  93        if (slab)
  94                goto out_unlock;
  95
  96        if (bio_slab_nr == bio_slab_max && entry == -1) {
  97                bio_slab_max <<= 1;
  98                bio_slabs = krealloc(bio_slabs,
  99                                     bio_slab_max * sizeof(struct bio_slab),
 100                                     GFP_KERNEL);
 101                if (!bio_slabs)
 102                        goto out_unlock;
 103        }
 104        if (entry == -1)
 105                entry = bio_slab_nr++;
 106
 107        bslab = &bio_slabs[entry];
 108
 109        snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
 110        slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
 111        if (!slab)
 112                goto out_unlock;
 113
 114        printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
 115        bslab->slab = slab;
 116        bslab->slab_ref = 1;
 117        bslab->slab_size = sz;
 118out_unlock:
 119        mutex_unlock(&bio_slab_lock);
 120        return slab;
 121}
 122
 123static void bio_put_slab(struct bio_set *bs)
 124{
 125        struct bio_slab *bslab = NULL;
 126        unsigned int i;
 127
 128        mutex_lock(&bio_slab_lock);
 129
 130        for (i = 0; i < bio_slab_nr; i++) {
 131                if (bs->bio_slab == bio_slabs[i].slab) {
 132                        bslab = &bio_slabs[i];
 133                        break;
 134                }
 135        }
 136
 137        if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
 138                goto out;
 139
 140        WARN_ON(!bslab->slab_ref);
 141
 142        if (--bslab->slab_ref)
 143                goto out;
 144
 145        kmem_cache_destroy(bslab->slab);
 146        bslab->slab = NULL;
 147
 148out:
 149        mutex_unlock(&bio_slab_lock);
 150}
 151
 152unsigned int bvec_nr_vecs(unsigned short idx)
 153{
 154        return bvec_slabs[idx].nr_vecs;
 155}
 156
 157void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
 158{
 159        BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
 160
 161        if (idx == BIOVEC_MAX_IDX)
 162                mempool_free(bv, bs->bvec_pool);
 163        else {
 164                struct biovec_slab *bvs = bvec_slabs + idx;
 165
 166                kmem_cache_free(bvs->slab, bv);
 167        }
 168}
 169
 170struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
 171                              struct bio_set *bs)
 172{
 173        struct bio_vec *bvl;
 174
 175        /*
 176         * see comment near bvec_array define!
 177         */
 178        switch (nr) {
 179        case 1:
 180                *idx = 0;
 181                break;
 182        case 2 ... 4:
 183                *idx = 1;
 184                break;
 185        case 5 ... 16:
 186                *idx = 2;
 187                break;
 188        case 17 ... 64:
 189                *idx = 3;
 190                break;
 191        case 65 ... 128:
 192                *idx = 4;
 193                break;
 194        case 129 ... BIO_MAX_PAGES:
 195                *idx = 5;
 196                break;
 197        default:
 198                return NULL;
 199        }
 200
 201        /*
 202         * idx now points to the pool we want to allocate from. only the
 203         * 1-vec entry pool is mempool backed.
 204         */
 205        if (*idx == BIOVEC_MAX_IDX) {
 206fallback:
 207                bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
 208        } else {
 209                struct biovec_slab *bvs = bvec_slabs + *idx;
 210                gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
 211
 212                /*
 213                 * Make this allocation restricted and don't dump info on
 214                 * allocation failures, since we'll fallback to the mempool
 215                 * in case of failure.
 216                 */
 217                __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
 218
 219                /*
 220                 * Try a slab allocation. If this fails and __GFP_WAIT
 221                 * is set, retry with the 1-entry mempool
 222                 */
 223                bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
 224                if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
 225                        *idx = BIOVEC_MAX_IDX;
 226                        goto fallback;
 227                }
 228        }
 229
 230        return bvl;
 231}
 232
 233void bio_free(struct bio *bio, struct bio_set *bs)
 234{
 235        void *p;
 236
 237        if (bio_has_allocated_vec(bio))
 238                bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
 239
 240        if (bio_integrity(bio))
 241                bio_integrity_free(bio, bs);
 242
 243        /*
 244         * If we have front padding, adjust the bio pointer before freeing
 245         */
 246        p = bio;
 247        if (bs->front_pad)
 248                p -= bs->front_pad;
 249
 250        mempool_free(p, bs->bio_pool);
 251}
 252EXPORT_SYMBOL(bio_free);
 253
 254void bio_init(struct bio *bio)
 255{
 256        memset(bio, 0, sizeof(*bio));
 257        bio->bi_flags = 1 << BIO_UPTODATE;
 258        atomic_set(&bio->bi_cnt, 1);
 259}
 260EXPORT_SYMBOL(bio_init);
 261
 262/**
 263 * bio_alloc_bioset - allocate a bio for I/O
 264 * @gfp_mask:   the GFP_ mask given to the slab allocator
 265 * @nr_iovecs:  number of iovecs to pre-allocate
 266 * @bs:         the bio_set to allocate from.
 267 *
 268 * Description:
 269 *   bio_alloc_bioset will try its own mempool to satisfy the allocation.
 270 *   If %__GFP_WAIT is set then we will block on the internal pool waiting
 271 *   for a &struct bio to become free.
 272 *
 273 *   Note that the caller must set ->bi_destructor on successful return
 274 *   of a bio, to do the appropriate freeing of the bio once the reference
 275 *   count drops to zero.
 276 **/
 277struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
 278{
 279        unsigned long idx = BIO_POOL_NONE;
 280        struct bio_vec *bvl = NULL;
 281        struct bio *bio;
 282        void *p;
 283
 284        p = mempool_alloc(bs->bio_pool, gfp_mask);
 285        if (unlikely(!p))
 286                return NULL;
 287        bio = p + bs->front_pad;
 288
 289        bio_init(bio);
 290
 291        if (unlikely(!nr_iovecs))
 292                goto out_set;
 293
 294        if (nr_iovecs <= BIO_INLINE_VECS) {
 295                bvl = bio->bi_inline_vecs;
 296                nr_iovecs = BIO_INLINE_VECS;
 297        } else {
 298                bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
 299                if (unlikely(!bvl))
 300                        goto err_free;
 301
 302                nr_iovecs = bvec_nr_vecs(idx);
 303        }
 304out_set:
 305        bio->bi_flags |= idx << BIO_POOL_OFFSET;
 306        bio->bi_max_vecs = nr_iovecs;
 307        bio->bi_io_vec = bvl;
 308        return bio;
 309
 310err_free:
 311        mempool_free(p, bs->bio_pool);
 312        return NULL;
 313}
 314EXPORT_SYMBOL(bio_alloc_bioset);
 315
 316static void bio_fs_destructor(struct bio *bio)
 317{
 318        bio_free(bio, fs_bio_set);
 319}
 320
 321/**
 322 *      bio_alloc - allocate a new bio, memory pool backed
 323 *      @gfp_mask: allocation mask to use
 324 *      @nr_iovecs: number of iovecs
 325 *
 326 *      bio_alloc will allocate a bio and associated bio_vec array that can hold
 327 *      at least @nr_iovecs entries. Allocations will be done from the
 328 *      fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
 329 *
 330 *      If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
 331 *      a bio. This is due to the mempool guarantees. To make this work, callers
 332 *      must never allocate more than 1 bio at a time from this pool. Callers
 333 *      that need to allocate more than 1 bio must always submit the previously
 334 *      allocated bio for IO before attempting to allocate a new one. Failure to
 335 *      do so can cause livelocks under memory pressure.
 336 *
 337 *      RETURNS:
 338 *      Pointer to new bio on success, NULL on failure.
 339 */
 340struct bio *bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
 341{
 342        struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
 343
 344        if (bio)
 345                bio->bi_destructor = bio_fs_destructor;
 346
 347        return bio;
 348}
 349EXPORT_SYMBOL(bio_alloc);
 350
 351static void bio_kmalloc_destructor(struct bio *bio)
 352{
 353        if (bio_integrity(bio))
 354                bio_integrity_free(bio, fs_bio_set);
 355        kfree(bio);
 356}
 357
 358/**
 359 * bio_kmalloc - allocate a bio for I/O using kmalloc()
 360 * @gfp_mask:   the GFP_ mask given to the slab allocator
 361 * @nr_iovecs:  number of iovecs to pre-allocate
 362 *
 363 * Description:
 364 *   Allocate a new bio with @nr_iovecs bvecs.  If @gfp_mask contains
 365 *   %__GFP_WAIT, the allocation is guaranteed to succeed.
 366 *
 367 **/
 368struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned int nr_iovecs)
 369{
 370        struct bio *bio;
 371
 372        if (nr_iovecs > UIO_MAXIOV)
 373                return NULL;
 374
 375        bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
 376                      gfp_mask);
 377        if (unlikely(!bio))
 378                return NULL;
 379
 380        bio_init(bio);
 381        bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
 382        bio->bi_max_vecs = nr_iovecs;
 383        bio->bi_io_vec = bio->bi_inline_vecs;
 384        bio->bi_destructor = bio_kmalloc_destructor;
 385
 386        return bio;
 387}
 388EXPORT_SYMBOL(bio_kmalloc);
 389
 390void zero_fill_bio(struct bio *bio)
 391{
 392        unsigned long flags;
 393        struct bio_vec *bv;
 394        int i;
 395
 396        bio_for_each_segment(bv, bio, i) {
 397                char *data = bvec_kmap_irq(bv, &flags);
 398                memset(data, 0, bv->bv_len);
 399                flush_dcache_page(bv->bv_page);
 400                bvec_kunmap_irq(data, &flags);
 401        }
 402}
 403EXPORT_SYMBOL(zero_fill_bio);
 404
 405/**
 406 * bio_put - release a reference to a bio
 407 * @bio:   bio to release reference to
 408 *
 409 * Description:
 410 *   Put a reference to a &struct bio, either one you have gotten with
 411 *   bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
 412 **/
 413void bio_put(struct bio *bio)
 414{
 415        BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
 416
 417        /*
 418         * last put frees it
 419         */
 420        if (atomic_dec_and_test(&bio->bi_cnt)) {
 421                bio->bi_next = NULL;
 422                bio->bi_destructor(bio);
 423        }
 424}
 425EXPORT_SYMBOL(bio_put);
 426
 427inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
 428{
 429        if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
 430                blk_recount_segments(q, bio);
 431
 432        return bio->bi_phys_segments;
 433}
 434EXPORT_SYMBOL(bio_phys_segments);
 435
 436/**
 437 *      __bio_clone     -       clone a bio
 438 *      @bio: destination bio
 439 *      @bio_src: bio to clone
 440 *
 441 *      Clone a &bio. Caller will own the returned bio, but not
 442 *      the actual data it points to. Reference count of returned
 443 *      bio will be one.
 444 */
 445void __bio_clone(struct bio *bio, struct bio *bio_src)
 446{
 447        memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
 448                bio_src->bi_max_vecs * sizeof(struct bio_vec));
 449
 450        /*
 451         * most users will be overriding ->bi_bdev with a new target,
 452         * so we don't set nor calculate new physical/hw segment counts here
 453         */
 454        bio->bi_sector = bio_src->bi_sector;
 455        bio->bi_bdev = bio_src->bi_bdev;
 456        bio->bi_flags |= 1 << BIO_CLONED;
 457        bio->bi_rw = bio_src->bi_rw;
 458        bio->bi_vcnt = bio_src->bi_vcnt;
 459        bio->bi_size = bio_src->bi_size;
 460        bio->bi_idx = bio_src->bi_idx;
 461}
 462EXPORT_SYMBOL(__bio_clone);
 463
 464/**
 465 *      bio_clone       -       clone a bio
 466 *      @bio: bio to clone
 467 *      @gfp_mask: allocation priority
 468 *
 469 *      Like __bio_clone, only also allocates the returned bio
 470 */
 471struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
 472{
 473        struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
 474
 475        if (!b)
 476                return NULL;
 477
 478        b->bi_destructor = bio_fs_destructor;
 479        __bio_clone(b, bio);
 480
 481        if (bio_integrity(bio)) {
 482                int ret;
 483
 484                ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
 485
 486                if (ret < 0) {
 487                        bio_put(b);
 488                        return NULL;
 489                }
 490        }
 491
 492        return b;
 493}
 494EXPORT_SYMBOL(bio_clone);
 495
 496/**
 497 *      bio_get_nr_vecs         - return approx number of vecs
 498 *      @bdev:  I/O target
 499 *
 500 *      Return the approximate number of pages we can send to this target.
 501 *      There's no guarantee that you will be able to fit this number of pages
 502 *      into a bio, it does not account for dynamic restrictions that vary
 503 *      on offset.
 504 */
 505int bio_get_nr_vecs(struct block_device *bdev)
 506{
 507        struct request_queue *q = bdev_get_queue(bdev);
 508        int nr_pages;
 509
 510        nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 511        if (nr_pages > queue_max_segments(q))
 512                nr_pages = queue_max_segments(q);
 513
 514        return nr_pages;
 515}
 516EXPORT_SYMBOL(bio_get_nr_vecs);
 517
 518static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
 519                          *page, unsigned int len, unsigned int offset,
 520                          unsigned short max_sectors)
 521{
 522        int retried_segments = 0;
 523        struct bio_vec *bvec;
 524
 525        /*
 526         * cloned bio must not modify vec list
 527         */
 528        if (unlikely(bio_flagged(bio, BIO_CLONED)))
 529                return 0;
 530
 531        if (((bio->bi_size + len) >> 9) > max_sectors)
 532                return 0;
 533
 534        /*
 535         * For filesystems with a blocksize smaller than the pagesize
 536         * we will often be called with the same page as last time and
 537         * a consecutive offset.  Optimize this special case.
 538         */
 539        if (bio->bi_vcnt > 0) {
 540                struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
 541
 542                if (page == prev->bv_page &&
 543                    offset == prev->bv_offset + prev->bv_len) {
 544                        unsigned int prev_bv_len = prev->bv_len;
 545                        prev->bv_len += len;
 546
 547                        if (q->merge_bvec_fn) {
 548                                struct bvec_merge_data bvm = {
 549                                        /* prev_bvec is already charged in
 550                                           bi_size, discharge it in order to
 551                                           simulate merging updated prev_bvec
 552                                           as new bvec. */
 553                                        .bi_bdev = bio->bi_bdev,
 554                                        .bi_sector = bio->bi_sector,
 555                                        .bi_size = bio->bi_size - prev_bv_len,
 556                                        .bi_rw = bio->bi_rw,
 557                                };
 558
 559                                if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
 560                                        prev->bv_len -= len;
 561                                        return 0;
 562                                }
 563                        }
 564
 565                        goto done;
 566                }
 567        }
 568
 569        if (bio->bi_vcnt >= bio->bi_max_vecs)
 570                return 0;
 571
 572        /*
 573         * we might lose a segment or two here, but rather that than
 574         * make this too complex.
 575         */
 576
 577        while (bio->bi_phys_segments >= queue_max_segments(q)) {
 578
 579                if (retried_segments)
 580                        return 0;
 581
 582                retried_segments = 1;
 583                blk_recount_segments(q, bio);
 584        }
 585
 586        /*
 587         * setup the new entry, we might clear it again later if we
 588         * cannot add the page
 589         */
 590        bvec = &bio->bi_io_vec[bio->bi_vcnt];
 591        bvec->bv_page = page;
 592        bvec->bv_len = len;
 593        bvec->bv_offset = offset;
 594
 595        /*
 596         * if queue has other restrictions (eg varying max sector size
 597         * depending on offset), it can specify a merge_bvec_fn in the
 598         * queue to get further control
 599         */
 600        if (q->merge_bvec_fn) {
 601                struct bvec_merge_data bvm = {
 602                        .bi_bdev = bio->bi_bdev,
 603                        .bi_sector = bio->bi_sector,
 604                        .bi_size = bio->bi_size,
 605                        .bi_rw = bio->bi_rw,
 606                };
 607
 608                /*
 609                 * merge_bvec_fn() returns number of bytes it can accept
 610                 * at this offset
 611                 */
 612                if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
 613                        bvec->bv_page = NULL;
 614                        bvec->bv_len = 0;
 615                        bvec->bv_offset = 0;
 616                        return 0;
 617                }
 618        }
 619
 620        /* If we may be able to merge these biovecs, force a recount */
 621        if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
 622                bio->bi_flags &= ~(1 << BIO_SEG_VALID);
 623
 624        bio->bi_vcnt++;
 625        bio->bi_phys_segments++;
 626 done:
 627        bio->bi_size += len;
 628        return len;
 629}
 630
 631/**
 632 *      bio_add_pc_page -       attempt to add page to bio
 633 *      @q: the target queue
 634 *      @bio: destination bio
 635 *      @page: page to add
 636 *      @len: vec entry length
 637 *      @offset: vec entry offset
 638 *
 639 *      Attempt to add a page to the bio_vec maplist. This can fail for a
 640 *      number of reasons, such as the bio being full or target block device
 641 *      limitations. The target block device must allow bio's up to PAGE_SIZE,
 642 *      so it is always possible to add a single page to an empty bio.
 643 *
 644 *      This should only be used by REQ_PC bios.
 645 */
 646int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
 647                    unsigned int len, unsigned int offset)
 648{
 649        return __bio_add_page(q, bio, page, len, offset,
 650                              queue_max_hw_sectors(q));
 651}
 652EXPORT_SYMBOL(bio_add_pc_page);
 653
 654/**
 655 *      bio_add_page    -       attempt to add page to bio
 656 *      @bio: destination bio
 657 *      @page: page to add
 658 *      @len: vec entry length
 659 *      @offset: vec entry offset
 660 *
 661 *      Attempt to add a page to the bio_vec maplist. This can fail for a
 662 *      number of reasons, such as the bio being full or target block device
 663 *      limitations. The target block device must allow bio's up to PAGE_SIZE,
 664 *      so it is always possible to add a single page to an empty bio.
 665 */
 666int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
 667                 unsigned int offset)
 668{
 669        struct request_queue *q = bdev_get_queue(bio->bi_bdev);
 670        return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
 671}
 672EXPORT_SYMBOL(bio_add_page);
 673
 674struct bio_map_data {
 675        struct bio_vec *iovecs;
 676        struct sg_iovec *sgvecs;
 677        int nr_sgvecs;
 678        int is_our_pages;
 679};
 680
 681static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
 682                             struct sg_iovec *iov, int iov_count,
 683                             int is_our_pages)
 684{
 685        memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
 686        memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
 687        bmd->nr_sgvecs = iov_count;
 688        bmd->is_our_pages = is_our_pages;
 689        bio->bi_private = bmd;
 690}
 691
 692static void bio_free_map_data(struct bio_map_data *bmd)
 693{
 694        kfree(bmd->iovecs);
 695        kfree(bmd->sgvecs);
 696        kfree(bmd);
 697}
 698
 699static struct bio_map_data *bio_alloc_map_data(int nr_segs,
 700                                               unsigned int iov_count,
 701                                               gfp_t gfp_mask)
 702{
 703        struct bio_map_data *bmd;
 704
 705        if (iov_count > UIO_MAXIOV)
 706                return NULL;
 707
 708        bmd = kmalloc(sizeof(*bmd), gfp_mask);
 709        if (!bmd)
 710                return NULL;
 711
 712        bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
 713        if (!bmd->iovecs) {
 714                kfree(bmd);
 715                return NULL;
 716        }
 717
 718        bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
 719        if (bmd->sgvecs)
 720                return bmd;
 721
 722        kfree(bmd->iovecs);
 723        kfree(bmd);
 724        return NULL;
 725}
 726
 727static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
 728                          struct sg_iovec *iov, int iov_count,
 729                          int to_user, int from_user, int do_free_page)
 730{
 731        int ret = 0, i;
 732        struct bio_vec *bvec;
 733        int iov_idx = 0;
 734        unsigned int iov_off = 0;
 735
 736        __bio_for_each_segment(bvec, bio, i, 0) {
 737                char *bv_addr = page_address(bvec->bv_page);
 738                unsigned int bv_len = iovecs[i].bv_len;
 739
 740                while (bv_len && iov_idx < iov_count) {
 741                        unsigned int bytes;
 742                        char __user *iov_addr;
 743
 744                        bytes = min_t(unsigned int,
 745                                      iov[iov_idx].iov_len - iov_off, bv_len);
 746                        iov_addr = iov[iov_idx].iov_base + iov_off;
 747
 748                        if (!ret) {
 749                                if (to_user)
 750                                        ret = copy_to_user(iov_addr, bv_addr,
 751                                                           bytes);
 752
 753                                if (from_user)
 754                                        ret = copy_from_user(bv_addr, iov_addr,
 755                                                             bytes);
 756
 757                                if (ret)
 758                                        ret = -EFAULT;
 759                        }
 760
 761                        bv_len -= bytes;
 762                        bv_addr += bytes;
 763                        iov_addr += bytes;
 764                        iov_off += bytes;
 765
 766                        if (iov[iov_idx].iov_len == iov_off) {
 767                                iov_idx++;
 768                                iov_off = 0;
 769                        }
 770                }
 771
 772                if (do_free_page)
 773                        __free_page(bvec->bv_page);
 774        }
 775
 776        return ret;
 777}
 778
 779/**
 780 *      bio_uncopy_user -       finish previously mapped bio
 781 *      @bio: bio being terminated
 782 *
 783 *      Free pages allocated from bio_copy_user() and write back data
 784 *      to user space in case of a read.
 785 */
 786int bio_uncopy_user(struct bio *bio)
 787{
 788        struct bio_map_data *bmd = bio->bi_private;
 789        int ret = 0;
 790
 791        if (!bio_flagged(bio, BIO_NULL_MAPPED))
 792                ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
 793                                     bmd->nr_sgvecs, bio_data_dir(bio) == READ,
 794                                     0, bmd->is_our_pages);
 795        bio_free_map_data(bmd);
 796        bio_put(bio);
 797        return ret;
 798}
 799EXPORT_SYMBOL(bio_uncopy_user);
 800
 801/**
 802 *      bio_copy_user_iov       -       copy user data to bio
 803 *      @q: destination block queue
 804 *      @map_data: pointer to the rq_map_data holding pages (if necessary)
 805 *      @iov:   the iovec.
 806 *      @iov_count: number of elements in the iovec
 807 *      @write_to_vm: bool indicating writing to pages or not
 808 *      @gfp_mask: memory allocation flags
 809 *
 810 *      Prepares and returns a bio for indirect user io, bouncing data
 811 *      to/from kernel pages as necessary. Must be paired with
 812 *      call bio_uncopy_user() on io completion.
 813 */
 814struct bio *bio_copy_user_iov(struct request_queue *q,
 815                              struct rq_map_data *map_data,
 816                              struct sg_iovec *iov, int iov_count,
 817                              int write_to_vm, gfp_t gfp_mask)
 818{
 819        struct bio_map_data *bmd;
 820        struct bio_vec *bvec;
 821        struct page *page;
 822        struct bio *bio;
 823        int i, ret;
 824        int nr_pages = 0;
 825        unsigned int len = 0;
 826        unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
 827
 828        for (i = 0; i < iov_count; i++) {
 829                unsigned long uaddr;
 830                unsigned long end;
 831                unsigned long start;
 832
 833                uaddr = (unsigned long)iov[i].iov_base;
 834                end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 835                start = uaddr >> PAGE_SHIFT;
 836
 837                /*
 838                 * Overflow, abort
 839                 */
 840                if (end < start)
 841                        return ERR_PTR(-EINVAL);
 842
 843                nr_pages += end - start;
 844                len += iov[i].iov_len;
 845        }
 846
 847        if (offset)
 848                nr_pages++;
 849
 850        bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
 851        if (!bmd)
 852                return ERR_PTR(-ENOMEM);
 853
 854        ret = -ENOMEM;
 855        bio = bio_kmalloc(gfp_mask, nr_pages);
 856        if (!bio)
 857                goto out_bmd;
 858
 859        if (!write_to_vm)
 860                bio->bi_rw |= REQ_WRITE;
 861
 862        ret = 0;
 863
 864        if (map_data) {
 865                nr_pages = 1 << map_data->page_order;
 866                i = map_data->offset / PAGE_SIZE;
 867        }
 868        while (len) {
 869                unsigned int bytes = PAGE_SIZE;
 870
 871                bytes -= offset;
 872
 873                if (bytes > len)
 874                        bytes = len;
 875
 876                if (map_data) {
 877                        if (i == map_data->nr_entries * nr_pages) {
 878                                ret = -ENOMEM;
 879                                break;
 880                        }
 881
 882                        page = map_data->pages[i / nr_pages];
 883                        page += (i % nr_pages);
 884
 885                        i++;
 886                } else {
 887                        page = alloc_page(q->bounce_gfp | gfp_mask);
 888                        if (!page) {
 889                                ret = -ENOMEM;
 890                                break;
 891                        }
 892                }
 893
 894                if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
 895                        break;
 896
 897                len -= bytes;
 898                offset = 0;
 899        }
 900
 901        if (ret)
 902                goto cleanup;
 903
 904        /*
 905         * success
 906         */
 907        if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
 908            (map_data && map_data->from_user)) {
 909                ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
 910                if (ret)
 911                        goto cleanup;
 912        }
 913
 914        bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
 915        return bio;
 916cleanup:
 917        if (!map_data)
 918                bio_for_each_segment(bvec, bio, i)
 919                        __free_page(bvec->bv_page);
 920
 921        bio_put(bio);
 922out_bmd:
 923        bio_free_map_data(bmd);
 924        return ERR_PTR(ret);
 925}
 926
 927/**
 928 *      bio_copy_user   -       copy user data to bio
 929 *      @q: destination block queue
 930 *      @map_data: pointer to the rq_map_data holding pages (if necessary)
 931 *      @uaddr: start of user address
 932 *      @len: length in bytes
 933 *      @write_to_vm: bool indicating writing to pages or not
 934 *      @gfp_mask: memory allocation flags
 935 *
 936 *      Prepares and returns a bio for indirect user io, bouncing data
 937 *      to/from kernel pages as necessary. Must be paired with
 938 *      call bio_uncopy_user() on io completion.
 939 */
 940struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
 941                          unsigned long uaddr, unsigned int len,
 942                          int write_to_vm, gfp_t gfp_mask)
 943{
 944        struct sg_iovec iov;
 945
 946        iov.iov_base = (void __user *)uaddr;
 947        iov.iov_len = len;
 948
 949        return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
 950}
 951EXPORT_SYMBOL(bio_copy_user);
 952
 953static struct bio *__bio_map_user_iov(struct request_queue *q,
 954                                      struct block_device *bdev,
 955                                      struct sg_iovec *iov, int iov_count,
 956                                      int write_to_vm, gfp_t gfp_mask)
 957{
 958        int i, j;
 959        int nr_pages = 0;
 960        struct page **pages;
 961        struct bio *bio;
 962        int cur_page = 0;
 963        int ret, offset;
 964
 965        for (i = 0; i < iov_count; i++) {
 966                unsigned long uaddr = (unsigned long)iov[i].iov_base;
 967                unsigned long len = iov[i].iov_len;
 968                unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 969                unsigned long start = uaddr >> PAGE_SHIFT;
 970
 971                /*
 972                 * Overflow, abort
 973                 */
 974                if (end < start)
 975                        return ERR_PTR(-EINVAL);
 976
 977                nr_pages += end - start;
 978                /*
 979                 * buffer must be aligned to at least hardsector size for now
 980                 */
 981                if (uaddr & queue_dma_alignment(q))
 982                        return ERR_PTR(-EINVAL);
 983        }
 984
 985        if (!nr_pages)
 986                return ERR_PTR(-EINVAL);
 987
 988        bio = bio_kmalloc(gfp_mask, nr_pages);
 989        if (!bio)
 990                return ERR_PTR(-ENOMEM);
 991
 992        ret = -ENOMEM;
 993        pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
 994        if (!pages)
 995                goto out;
 996
 997        for (i = 0; i < iov_count; i++) {
 998                unsigned long uaddr = (unsigned long)iov[i].iov_base;
 999                unsigned long len = iov[i].iov_len;
1000                unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1001                unsigned long start = uaddr >> PAGE_SHIFT;
1002                const int local_nr_pages = end - start;
1003                const int page_limit = cur_page + local_nr_pages;
1004
1005                ret = get_user_pages_fast(uaddr, local_nr_pages,
1006                                write_to_vm, &pages[cur_page]);
1007                if (ret < local_nr_pages) {
1008                        ret = -EFAULT;
1009                        goto out_unmap;
1010                }
1011
1012                offset = uaddr & ~PAGE_MASK;
1013                for (j = cur_page; j < page_limit; j++) {
1014                        unsigned int bytes = PAGE_SIZE - offset;
1015
1016                        if (len <= 0)
1017                                break;
1018                        
1019                        if (bytes > len)
1020                                bytes = len;
1021
1022                        /*
1023                         * sorry...
1024                         */
1025                        if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1026                                            bytes)
1027                                break;
1028
1029                        len -= bytes;
1030                        offset = 0;
1031                }
1032
1033                cur_page = j;
1034                /*
1035                 * release the pages we didn't map into the bio, if any
1036                 */
1037                while (j < page_limit)
1038                        page_cache_release(pages[j++]);
1039        }
1040
1041        kfree(pages);
1042
1043        /*
1044         * set data direction, and check if mapped pages need bouncing
1045         */
1046        if (!write_to_vm)
1047                bio->bi_rw |= REQ_WRITE;
1048
1049        bio->bi_bdev = bdev;
1050        bio->bi_flags |= (1 << BIO_USER_MAPPED);
1051        return bio;
1052
1053 out_unmap:
1054        for (i = 0; i < nr_pages; i++) {
1055                if(!pages[i])
1056                        break;
1057                page_cache_release(pages[i]);
1058        }
1059 out:
1060        kfree(pages);
1061        bio_put(bio);
1062        return ERR_PTR(ret);
1063}
1064
1065/**
1066 *      bio_map_user    -       map user address into bio
1067 *      @q: the struct request_queue for the bio
1068 *      @bdev: destination block device
1069 *      @uaddr: start of user address
1070 *      @len: length in bytes
1071 *      @write_to_vm: bool indicating writing to pages or not
1072 *      @gfp_mask: memory allocation flags
1073 *
1074 *      Map the user space address into a bio suitable for io to a block
1075 *      device. Returns an error pointer in case of error.
1076 */
1077struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1078                         unsigned long uaddr, unsigned int len, int write_to_vm,
1079                         gfp_t gfp_mask)
1080{
1081        struct sg_iovec iov;
1082
1083        iov.iov_base = (void __user *)uaddr;
1084        iov.iov_len = len;
1085
1086        return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1087}
1088EXPORT_SYMBOL(bio_map_user);
1089
1090/**
1091 *      bio_map_user_iov - map user sg_iovec table into bio
1092 *      @q: the struct request_queue for the bio
1093 *      @bdev: destination block device
1094 *      @iov:   the iovec.
1095 *      @iov_count: number of elements in the iovec
1096 *      @write_to_vm: bool indicating writing to pages or not
1097 *      @gfp_mask: memory allocation flags
1098 *
1099 *      Map the user space address into a bio suitable for io to a block
1100 *      device. Returns an error pointer in case of error.
1101 */
1102struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1103                             struct sg_iovec *iov, int iov_count,
1104                             int write_to_vm, gfp_t gfp_mask)
1105{
1106        struct bio *bio;
1107
1108        bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1109                                 gfp_mask);
1110        if (IS_ERR(bio))
1111                return bio;
1112
1113        /*
1114         * subtle -- if __bio_map_user() ended up bouncing a bio,
1115         * it would normally disappear when its bi_end_io is run.
1116         * however, we need it for the unmap, so grab an extra
1117         * reference to it
1118         */
1119        bio_get(bio);
1120
1121        return bio;
1122}
1123
1124static void __bio_unmap_user(struct bio *bio)
1125{
1126        struct bio_vec *bvec;
1127        int i;
1128
1129        /*
1130         * make sure we dirty pages we wrote to
1131         */
1132        __bio_for_each_segment(bvec, bio, i, 0) {
1133                if (bio_data_dir(bio) == READ)
1134                        set_page_dirty_lock(bvec->bv_page);
1135
1136                page_cache_release(bvec->bv_page);
1137        }
1138
1139        bio_put(bio);
1140}
1141
1142/**
1143 *      bio_unmap_user  -       unmap a bio
1144 *      @bio:           the bio being unmapped
1145 *
1146 *      Unmap a bio previously mapped by bio_map_user(). Must be called with
1147 *      a process context.
1148 *
1149 *      bio_unmap_user() may sleep.
1150 */
1151void bio_unmap_user(struct bio *bio)
1152{
1153        __bio_unmap_user(bio);
1154        bio_put(bio);
1155}
1156EXPORT_SYMBOL(bio_unmap_user);
1157
1158static void bio_map_kern_endio(struct bio *bio, int err)
1159{
1160        bio_put(bio);
1161}
1162
1163static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1164                                  unsigned int len, gfp_t gfp_mask)
1165{
1166        unsigned long kaddr = (unsigned long)data;
1167        unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1168        unsigned long start = kaddr >> PAGE_SHIFT;
1169        const int nr_pages = end - start;
1170        int offset, i;
1171        struct bio *bio;
1172
1173        bio = bio_kmalloc(gfp_mask, nr_pages);
1174        if (!bio)
1175                return ERR_PTR(-ENOMEM);
1176
1177        offset = offset_in_page(kaddr);
1178        for (i = 0; i < nr_pages; i++) {
1179                unsigned int bytes = PAGE_SIZE - offset;
1180
1181                if (len <= 0)
1182                        break;
1183
1184                if (bytes > len)
1185                        bytes = len;
1186
1187                if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1188                                    offset) < bytes)
1189                        break;
1190
1191                data += bytes;
1192                len -= bytes;
1193                offset = 0;
1194        }
1195
1196        bio->bi_end_io = bio_map_kern_endio;
1197        return bio;
1198}
1199
1200/**
1201 *      bio_map_kern    -       map kernel address into bio
1202 *      @q: the struct request_queue for the bio
1203 *      @data: pointer to buffer to map
1204 *      @len: length in bytes
1205 *      @gfp_mask: allocation flags for bio allocation
1206 *
1207 *      Map the kernel address into a bio suitable for io to a block
1208 *      device. Returns an error pointer in case of error.
1209 */
1210struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1211                         gfp_t gfp_mask)
1212{
1213        struct bio *bio;
1214
1215        bio = __bio_map_kern(q, data, len, gfp_mask);
1216        if (IS_ERR(bio))
1217                return bio;
1218
1219        if (bio->bi_size == len)
1220                return bio;
1221
1222        /*
1223         * Don't support partial mappings.
1224         */
1225        bio_put(bio);
1226        return ERR_PTR(-EINVAL);
1227}
1228EXPORT_SYMBOL(bio_map_kern);
1229
1230static void bio_copy_kern_endio(struct bio *bio, int err)
1231{
1232        struct bio_vec *bvec;
1233        const int read = bio_data_dir(bio) == READ;
1234        struct bio_map_data *bmd = bio->bi_private;
1235        int i;
1236        char *p = bmd->sgvecs[0].iov_base;
1237
1238        __bio_for_each_segment(bvec, bio, i, 0) {
1239                char *addr = page_address(bvec->bv_page);
1240                int len = bmd->iovecs[i].bv_len;
1241
1242                if (read)
1243                        memcpy(p, addr, len);
1244
1245                __free_page(bvec->bv_page);
1246                p += len;
1247        }
1248
1249        bio_free_map_data(bmd);
1250        bio_put(bio);
1251}
1252
1253/**
1254 *      bio_copy_kern   -       copy kernel address into bio
1255 *      @q: the struct request_queue for the bio
1256 *      @data: pointer to buffer to copy
1257 *      @len: length in bytes
1258 *      @gfp_mask: allocation flags for bio and page allocation
1259 *      @reading: data direction is READ
1260 *
1261 *      copy the kernel address into a bio suitable for io to a block
1262 *      device. Returns an error pointer in case of error.
1263 */
1264struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1265                          gfp_t gfp_mask, int reading)
1266{
1267        struct bio *bio;
1268        struct bio_vec *bvec;
1269        int i;
1270
1271        bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1272        if (IS_ERR(bio))
1273                return bio;
1274
1275        if (!reading) {
1276                void *p = data;
1277
1278                bio_for_each_segment(bvec, bio, i) {
1279                        char *addr = page_address(bvec->bv_page);
1280
1281                        memcpy(addr, p, bvec->bv_len);
1282                        p += bvec->bv_len;
1283                }
1284        }
1285
1286        bio->bi_end_io = bio_copy_kern_endio;
1287
1288        return bio;
1289}
1290EXPORT_SYMBOL(bio_copy_kern);
1291
1292/*
1293 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1294 * for performing direct-IO in BIOs.
1295 *
1296 * The problem is that we cannot run set_page_dirty() from interrupt context
1297 * because the required locks are not interrupt-safe.  So what we can do is to
1298 * mark the pages dirty _before_ performing IO.  And in interrupt context,
1299 * check that the pages are still dirty.   If so, fine.  If not, redirty them
1300 * in process context.
1301 *
1302 * We special-case compound pages here: normally this means reads into hugetlb
1303 * pages.  The logic in here doesn't really work right for compound pages
1304 * because the VM does not uniformly chase down the head page in all cases.
1305 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1306 * handle them at all.  So we skip compound pages here at an early stage.
1307 *
1308 * Note that this code is very hard to test under normal circumstances because
1309 * direct-io pins the pages with get_user_pages().  This makes
1310 * is_page_cache_freeable return false, and the VM will not clean the pages.
1311 * But other code (eg, pdflush) could clean the pages if they are mapped
1312 * pagecache.
1313 *
1314 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1315 * deferred bio dirtying paths.
1316 */
1317
1318/*
1319 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1320 */
1321void bio_set_pages_dirty(struct bio *bio)
1322{
1323        struct bio_vec *bvec = bio->bi_io_vec;
1324        int i;
1325
1326        for (i = 0; i < bio->bi_vcnt; i++) {
1327                struct page *page = bvec[i].bv_page;
1328
1329                if (page && !PageCompound(page))
1330                        set_page_dirty_lock(page);
1331        }
1332}
1333
1334static void bio_release_pages(struct bio *bio)
1335{
1336        struct bio_vec *bvec = bio->bi_io_vec;
1337        int i;
1338
1339        for (i = 0; i < bio->bi_vcnt; i++) {
1340                struct page *page = bvec[i].bv_page;
1341
1342                if (page)
1343                        put_page(page);
1344        }
1345}
1346
1347/*
1348 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1349 * If they are, then fine.  If, however, some pages are clean then they must
1350 * have been written out during the direct-IO read.  So we take another ref on
1351 * the BIO and the offending pages and re-dirty the pages in process context.
1352 *
1353 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1354 * here on.  It will run one page_cache_release() against each page and will
1355 * run one bio_put() against the BIO.
1356 */
1357
1358static void bio_dirty_fn(struct work_struct *work);
1359
1360static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1361static DEFINE_SPINLOCK(bio_dirty_lock);
1362static struct bio *bio_dirty_list;
1363
1364/*
1365 * This runs in process context
1366 */
1367static void bio_dirty_fn(struct work_struct *work)
1368{
1369        unsigned long flags;
1370        struct bio *bio;
1371
1372        spin_lock_irqsave(&bio_dirty_lock, flags);
1373        bio = bio_dirty_list;
1374        bio_dirty_list = NULL;
1375        spin_unlock_irqrestore(&bio_dirty_lock, flags);
1376
1377        while (bio) {
1378                struct bio *next = bio->bi_private;
1379
1380                bio_set_pages_dirty(bio);
1381                bio_release_pages(bio);
1382                bio_put(bio);
1383                bio = next;
1384        }
1385}
1386
1387void bio_check_pages_dirty(struct bio *bio)
1388{
1389        struct bio_vec *bvec = bio->bi_io_vec;
1390        int nr_clean_pages = 0;
1391        int i;
1392
1393        for (i = 0; i < bio->bi_vcnt; i++) {
1394                struct page *page = bvec[i].bv_page;
1395
1396                if (PageDirty(page) || PageCompound(page)) {
1397                        page_cache_release(page);
1398                        bvec[i].bv_page = NULL;
1399                } else {
1400                        nr_clean_pages++;
1401                }
1402        }
1403
1404        if (nr_clean_pages) {
1405                unsigned long flags;
1406
1407                spin_lock_irqsave(&bio_dirty_lock, flags);
1408                bio->bi_private = bio_dirty_list;
1409                bio_dirty_list = bio;
1410                spin_unlock_irqrestore(&bio_dirty_lock, flags);
1411                schedule_work(&bio_dirty_work);
1412        } else {
1413                bio_put(bio);
1414        }
1415}
1416
1417#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1418void bio_flush_dcache_pages(struct bio *bi)
1419{
1420        int i;
1421        struct bio_vec *bvec;
1422
1423        bio_for_each_segment(bvec, bi, i)
1424                flush_dcache_page(bvec->bv_page);
1425}
1426EXPORT_SYMBOL(bio_flush_dcache_pages);
1427#endif
1428
1429/**
1430 * bio_endio - end I/O on a bio
1431 * @bio:        bio
1432 * @error:      error, if any
1433 *
1434 * Description:
1435 *   bio_endio() will end I/O on the whole bio. bio_endio() is the
1436 *   preferred way to end I/O on a bio, it takes care of clearing
1437 *   BIO_UPTODATE on error. @error is 0 on success, and and one of the
1438 *   established -Exxxx (-EIO, for instance) error values in case
1439 *   something went wrong. No one should call bi_end_io() directly on a
1440 *   bio unless they own it and thus know that it has an end_io
1441 *   function.
1442 **/
1443void bio_endio(struct bio *bio, int error)
1444{
1445        if (error)
1446                clear_bit(BIO_UPTODATE, &bio->bi_flags);
1447        else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1448                error = -EIO;
1449
1450        if (bio->bi_end_io)
1451                bio->bi_end_io(bio, error);
1452}
1453EXPORT_SYMBOL(bio_endio);
1454
1455void bio_pair_release(struct bio_pair *bp)
1456{
1457        if (atomic_dec_and_test(&bp->cnt)) {
1458                struct bio *master = bp->bio1.bi_private;
1459
1460                bio_endio(master, bp->error);
1461                mempool_free(bp, bp->bio2.bi_private);
1462        }
1463}
1464EXPORT_SYMBOL(bio_pair_release);
1465
1466static void bio_pair_end_1(struct bio *bi, int err)
1467{
1468        struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1469
1470        if (err)
1471                bp->error = err;
1472
1473        bio_pair_release(bp);
1474}
1475
1476static void bio_pair_end_2(struct bio *bi, int err)
1477{
1478        struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1479
1480        if (err)
1481                bp->error = err;
1482
1483        bio_pair_release(bp);
1484}
1485
1486/*
1487 * split a bio - only worry about a bio with a single page in its iovec
1488 */
1489struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1490{
1491        struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1492
1493        if (!bp)
1494                return bp;
1495
1496        trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1497                                bi->bi_sector + first_sectors);
1498
1499        BUG_ON(bi->bi_vcnt != 1);
1500        BUG_ON(bi->bi_idx != 0);
1501        atomic_set(&bp->cnt, 3);
1502        bp->error = 0;
1503        bp->bio1 = *bi;
1504        bp->bio2 = *bi;
1505        bp->bio2.bi_sector += first_sectors;
1506        bp->bio2.bi_size -= first_sectors << 9;
1507        bp->bio1.bi_size = first_sectors << 9;
1508
1509        bp->bv1 = bi->bi_io_vec[0];
1510        bp->bv2 = bi->bi_io_vec[0];
1511        bp->bv2.bv_offset += first_sectors << 9;
1512        bp->bv2.bv_len -= first_sectors << 9;
1513        bp->bv1.bv_len = first_sectors << 9;
1514
1515        bp->bio1.bi_io_vec = &bp->bv1;
1516        bp->bio2.bi_io_vec = &bp->bv2;
1517
1518        bp->bio1.bi_max_vecs = 1;
1519        bp->bio2.bi_max_vecs = 1;
1520
1521        bp->bio1.bi_end_io = bio_pair_end_1;
1522        bp->bio2.bi_end_io = bio_pair_end_2;
1523
1524        bp->bio1.bi_private = bi;
1525        bp->bio2.bi_private = bio_split_pool;
1526
1527        if (bio_integrity(bi))
1528                bio_integrity_split(bi, bp, first_sectors);
1529
1530        return bp;
1531}
1532EXPORT_SYMBOL(bio_split);
1533
1534/**
1535 *      bio_sector_offset - Find hardware sector offset in bio
1536 *      @bio:           bio to inspect
1537 *      @index:         bio_vec index
1538 *      @offset:        offset in bv_page
1539 *
1540 *      Return the number of hardware sectors between beginning of bio
1541 *      and an end point indicated by a bio_vec index and an offset
1542 *      within that vector's page.
1543 */
1544sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1545                           unsigned int offset)
1546{
1547        unsigned int sector_sz;
1548        struct bio_vec *bv;
1549        sector_t sectors;
1550        int i;
1551
1552        sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1553        sectors = 0;
1554
1555        if (index >= bio->bi_idx)
1556                index = bio->bi_vcnt - 1;
1557
1558        __bio_for_each_segment(bv, bio, i, 0) {
1559                if (i == index) {
1560                        if (offset > bv->bv_offset)
1561                                sectors += (offset - bv->bv_offset) / sector_sz;
1562                        break;
1563                }
1564
1565                sectors += bv->bv_len / sector_sz;
1566        }
1567
1568        return sectors;
1569}
1570EXPORT_SYMBOL(bio_sector_offset);
1571
1572/*
1573 * create memory pools for biovec's in a bio_set.
1574 * use the global biovec slabs created for general use.
1575 */
1576static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1577{
1578        struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1579
1580        bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1581        if (!bs->bvec_pool)
1582                return -ENOMEM;
1583
1584        return 0;
1585}
1586
1587static void biovec_free_pools(struct bio_set *bs)
1588{
1589        mempool_destroy(bs->bvec_pool);
1590}
1591
1592void bioset_free(struct bio_set *bs)
1593{
1594        if (bs->bio_pool)
1595                mempool_destroy(bs->bio_pool);
1596
1597        bioset_integrity_free(bs);
1598        biovec_free_pools(bs);
1599        bio_put_slab(bs);
1600
1601        kfree(bs);
1602}
1603EXPORT_SYMBOL(bioset_free);
1604
1605/**
1606 * bioset_create  - Create a bio_set
1607 * @pool_size:  Number of bio and bio_vecs to cache in the mempool
1608 * @front_pad:  Number of bytes to allocate in front of the returned bio
1609 *
1610 * Description:
1611 *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1612 *    to ask for a number of bytes to be allocated in front of the bio.
1613 *    Front pad allocation is useful for embedding the bio inside
1614 *    another structure, to avoid allocating extra data to go with the bio.
1615 *    Note that the bio must be embedded at the END of that structure always,
1616 *    or things will break badly.
1617 */
1618struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1619{
1620        unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1621        struct bio_set *bs;
1622
1623        bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1624        if (!bs)
1625                return NULL;
1626
1627        bs->front_pad = front_pad;
1628
1629        bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1630        if (!bs->bio_slab) {
1631                kfree(bs);
1632                return NULL;
1633        }
1634
1635        bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1636        if (!bs->bio_pool)
1637                goto bad;
1638
1639        if (!biovec_create_pools(bs, pool_size))
1640                return bs;
1641
1642bad:
1643        bioset_free(bs);
1644        return NULL;
1645}
1646EXPORT_SYMBOL(bioset_create);
1647
1648static void __init biovec_init_slabs(void)
1649{
1650        int i;
1651
1652        for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1653                int size;
1654                struct biovec_slab *bvs = bvec_slabs + i;
1655
1656                if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1657                        bvs->slab = NULL;
1658                        continue;
1659                }
1660
1661                size = bvs->nr_vecs * sizeof(struct bio_vec);
1662                bvs->slab = kmem_cache_create(bvs->name, size, 0,
1663                                SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1664        }
1665}
1666
1667static int __init init_bio(void)
1668{
1669        bio_slab_max = 2;
1670        bio_slab_nr = 0;
1671        bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1672        if (!bio_slabs)
1673                panic("bio: can't allocate bios\n");
1674
1675        bio_integrity_init();
1676        biovec_init_slabs();
1677
1678        fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1679        if (!fs_bio_set)
1680                panic("bio: can't allocate bios\n");
1681
1682        if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1683                panic("bio: can't create integrity pool\n");
1684
1685        bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1686                                                     sizeof(struct bio_pair));
1687        if (!bio_split_pool)
1688                panic("bio: can't create split pool\n");
1689
1690        return 0;
1691}
1692subsys_initcall(init_bio);
1693