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/iocontext.h>
  23#include <linux/slab.h>
  24#include <linux/init.h>
  25#include <linux/kernel.h>
  26#include <linux/export.h>
  27#include <linux/mempool.h>
  28#include <linux/workqueue.h>
  29#include <linux/cgroup.h>
  30#include <scsi/sg.h>            /* for struct sg_iovec */
  31
  32#include <trace/events/block.h>
  33
  34/*
  35 * Test patch to inline a certain number of bi_io_vec's inside the bio
  36 * itself, to shrink a bio data allocation from two mempool calls to one
  37 */
  38#define BIO_INLINE_VECS         4
  39
  40static mempool_t *bio_split_pool __read_mostly;
  41
  42/*
  43 * if you change this list, also change bvec_alloc or things will
  44 * break badly! cannot be bigger than what you can fit into an
  45 * unsigned short
  46 */
  47#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
  48static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
  49        BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
  50};
  51#undef BV
  52
  53/*
  54 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
  55 * IO code that does not need private memory pools.
  56 */
  57struct bio_set *fs_bio_set;
  58
  59/*
  60 * Our slab pool management
  61 */
  62struct bio_slab {
  63        struct kmem_cache *slab;
  64        unsigned int slab_ref;
  65        unsigned int slab_size;
  66        char name[8];
  67};
  68static DEFINE_MUTEX(bio_slab_lock);
  69static struct bio_slab *bio_slabs;
  70static unsigned int bio_slab_nr, bio_slab_max;
  71
  72static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
  73{
  74        unsigned int sz = sizeof(struct bio) + extra_size;
  75        struct kmem_cache *slab = NULL;
  76        struct bio_slab *bslab, *new_bio_slabs;
  77        unsigned int i, entry = -1;
  78
  79        mutex_lock(&bio_slab_lock);
  80
  81        i = 0;
  82        while (i < bio_slab_nr) {
  83                bslab = &bio_slabs[i];
  84
  85                if (!bslab->slab && entry == -1)
  86                        entry = i;
  87                else if (bslab->slab_size == sz) {
  88                        slab = bslab->slab;
  89                        bslab->slab_ref++;
  90                        break;
  91                }
  92                i++;
  93        }
  94
  95        if (slab)
  96                goto out_unlock;
  97
  98        if (bio_slab_nr == bio_slab_max && entry == -1) {
  99                bio_slab_max <<= 1;
 100                new_bio_slabs = krealloc(bio_slabs,
 101                                         bio_slab_max * sizeof(struct bio_slab),
 102                                         GFP_KERNEL);
 103                if (!new_bio_slabs)
 104                        goto out_unlock;
 105                bio_slabs = new_bio_slabs;
 106        }
 107        if (entry == -1)
 108                entry = bio_slab_nr++;
 109
 110        bslab = &bio_slabs[entry];
 111
 112        snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
 113        slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
 114        if (!slab)
 115                goto out_unlock;
 116
 117        printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
 118        bslab->slab = slab;
 119        bslab->slab_ref = 1;
 120        bslab->slab_size = sz;
 121out_unlock:
 122        mutex_unlock(&bio_slab_lock);
 123        return slab;
 124}
 125
 126static void bio_put_slab(struct bio_set *bs)
 127{
 128        struct bio_slab *bslab = NULL;
 129        unsigned int i;
 130
 131        mutex_lock(&bio_slab_lock);
 132
 133        for (i = 0; i < bio_slab_nr; i++) {
 134                if (bs->bio_slab == bio_slabs[i].slab) {
 135                        bslab = &bio_slabs[i];
 136                        break;
 137                }
 138        }
 139
 140        if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
 141                goto out;
 142
 143        WARN_ON(!bslab->slab_ref);
 144
 145        if (--bslab->slab_ref)
 146                goto out;
 147
 148        kmem_cache_destroy(bslab->slab);
 149        bslab->slab = NULL;
 150
 151out:
 152        mutex_unlock(&bio_slab_lock);
 153}
 154
 155unsigned int bvec_nr_vecs(unsigned short idx)
 156{
 157        return bvec_slabs[idx].nr_vecs;
 158}
 159
 160void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
 161{
 162        BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
 163
 164        if (idx == BIOVEC_MAX_IDX)
 165                mempool_free(bv, bs->bvec_pool);
 166        else {
 167                struct biovec_slab *bvs = bvec_slabs + idx;
 168
 169                kmem_cache_free(bvs->slab, bv);
 170        }
 171}
 172
 173struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
 174                              struct bio_set *bs)
 175{
 176        struct bio_vec *bvl;
 177
 178        /*
 179         * see comment near bvec_array define!
 180         */
 181        switch (nr) {
 182        case 1:
 183                *idx = 0;
 184                break;
 185        case 2 ... 4:
 186                *idx = 1;
 187                break;
 188        case 5 ... 16:
 189                *idx = 2;
 190                break;
 191        case 17 ... 64:
 192                *idx = 3;
 193                break;
 194        case 65 ... 128:
 195                *idx = 4;
 196                break;
 197        case 129 ... BIO_MAX_PAGES:
 198                *idx = 5;
 199                break;
 200        default:
 201                return NULL;
 202        }
 203
 204        /*
 205         * idx now points to the pool we want to allocate from. only the
 206         * 1-vec entry pool is mempool backed.
 207         */
 208        if (*idx == BIOVEC_MAX_IDX) {
 209fallback:
 210                bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
 211        } else {
 212                struct biovec_slab *bvs = bvec_slabs + *idx;
 213                gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
 214
 215                /*
 216                 * Make this allocation restricted and don't dump info on
 217                 * allocation failures, since we'll fallback to the mempool
 218                 * in case of failure.
 219                 */
 220                __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
 221
 222                /*
 223                 * Try a slab allocation. If this fails and __GFP_WAIT
 224                 * is set, retry with the 1-entry mempool
 225                 */
 226                bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
 227                if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
 228                        *idx = BIOVEC_MAX_IDX;
 229                        goto fallback;
 230                }
 231        }
 232
 233        return bvl;
 234}
 235
 236void bio_free(struct bio *bio, struct bio_set *bs)
 237{
 238        void *p;
 239
 240        if (bio_has_allocated_vec(bio))
 241                bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
 242
 243        if (bio_integrity(bio))
 244                bio_integrity_free(bio, bs);
 245
 246        /*
 247         * If we have front padding, adjust the bio pointer before freeing
 248         */
 249        p = bio;
 250        if (bs->front_pad)
 251                p -= bs->front_pad;
 252
 253        mempool_free(p, bs->bio_pool);
 254}
 255EXPORT_SYMBOL(bio_free);
 256
 257void bio_init(struct bio *bio)
 258{
 259        memset(bio, 0, sizeof(*bio));
 260        bio->bi_flags = 1 << BIO_UPTODATE;
 261        atomic_set(&bio->bi_cnt, 1);
 262}
 263EXPORT_SYMBOL(bio_init);
 264
 265/**
 266 * bio_alloc_bioset - allocate a bio for I/O
 267 * @gfp_mask:   the GFP_ mask given to the slab allocator
 268 * @nr_iovecs:  number of iovecs to pre-allocate
 269 * @bs:         the bio_set to allocate from.
 270 *
 271 * Description:
 272 *   bio_alloc_bioset will try its own mempool to satisfy the allocation.
 273 *   If %__GFP_WAIT is set then we will block on the internal pool waiting
 274 *   for a &struct bio to become free.
 275 *
 276 *   Note that the caller must set ->bi_destructor on successful return
 277 *   of a bio, to do the appropriate freeing of the bio once the reference
 278 *   count drops to zero.
 279 **/
 280struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
 281{
 282        unsigned long idx = BIO_POOL_NONE;
 283        struct bio_vec *bvl = NULL;
 284        struct bio *bio;
 285        void *p;
 286
 287        p = mempool_alloc(bs->bio_pool, gfp_mask);
 288        if (unlikely(!p))
 289                return NULL;
 290        bio = p + bs->front_pad;
 291
 292        bio_init(bio);
 293
 294        if (unlikely(!nr_iovecs))
 295                goto out_set;
 296
 297        if (nr_iovecs <= BIO_INLINE_VECS) {
 298                bvl = bio->bi_inline_vecs;
 299                nr_iovecs = BIO_INLINE_VECS;
 300        } else {
 301                bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
 302                if (unlikely(!bvl))
 303                        goto err_free;
 304
 305                nr_iovecs = bvec_nr_vecs(idx);
 306        }
 307out_set:
 308        bio->bi_flags |= idx << BIO_POOL_OFFSET;
 309        bio->bi_max_vecs = nr_iovecs;
 310        bio->bi_io_vec = bvl;
 311        return bio;
 312
 313err_free:
 314        mempool_free(p, bs->bio_pool);
 315        return NULL;
 316}
 317EXPORT_SYMBOL(bio_alloc_bioset);
 318
 319static void bio_fs_destructor(struct bio *bio)
 320{
 321        bio_free(bio, fs_bio_set);
 322}
 323
 324/**
 325 *      bio_alloc - allocate a new bio, memory pool backed
 326 *      @gfp_mask: allocation mask to use
 327 *      @nr_iovecs: number of iovecs
 328 *
 329 *      bio_alloc will allocate a bio and associated bio_vec array that can hold
 330 *      at least @nr_iovecs entries. Allocations will be done from the
 331 *      fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
 332 *
 333 *      If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
 334 *      a bio. This is due to the mempool guarantees. To make this work, callers
 335 *      must never allocate more than 1 bio at a time from this pool. Callers
 336 *      that need to allocate more than 1 bio must always submit the previously
 337 *      allocated bio for IO before attempting to allocate a new one. Failure to
 338 *      do so can cause livelocks under memory pressure.
 339 *
 340 *      RETURNS:
 341 *      Pointer to new bio on success, NULL on failure.
 342 */
 343struct bio *bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
 344{
 345        struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
 346
 347        if (bio)
 348                bio->bi_destructor = bio_fs_destructor;
 349
 350        return bio;
 351}
 352EXPORT_SYMBOL(bio_alloc);
 353
 354static void bio_kmalloc_destructor(struct bio *bio)
 355{
 356        if (bio_integrity(bio))
 357                bio_integrity_free(bio, fs_bio_set);
 358        kfree(bio);
 359}
 360
 361/**
 362 * bio_kmalloc - allocate a bio for I/O using kmalloc()
 363 * @gfp_mask:   the GFP_ mask given to the slab allocator
 364 * @nr_iovecs:  number of iovecs to pre-allocate
 365 *
 366 * Description:
 367 *   Allocate a new bio with @nr_iovecs bvecs.  If @gfp_mask contains
 368 *   %__GFP_WAIT, the allocation is guaranteed to succeed.
 369 *
 370 **/
 371struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned int nr_iovecs)
 372{
 373        struct bio *bio;
 374
 375        if (nr_iovecs > UIO_MAXIOV)
 376                return NULL;
 377
 378        bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
 379                      gfp_mask);
 380        if (unlikely(!bio))
 381                return NULL;
 382
 383        bio_init(bio);
 384        bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
 385        bio->bi_max_vecs = nr_iovecs;
 386        bio->bi_io_vec = bio->bi_inline_vecs;
 387        bio->bi_destructor = bio_kmalloc_destructor;
 388
 389        return bio;
 390}
 391EXPORT_SYMBOL(bio_kmalloc);
 392
 393void zero_fill_bio(struct bio *bio)
 394{
 395        unsigned long flags;
 396        struct bio_vec *bv;
 397        int i;
 398
 399        bio_for_each_segment(bv, bio, i) {
 400                char *data = bvec_kmap_irq(bv, &flags);
 401                memset(data, 0, bv->bv_len);
 402                flush_dcache_page(bv->bv_page);
 403                bvec_kunmap_irq(data, &flags);
 404        }
 405}
 406EXPORT_SYMBOL(zero_fill_bio);
 407
 408/**
 409 * bio_put - release a reference to a bio
 410 * @bio:   bio to release reference to
 411 *
 412 * Description:
 413 *   Put a reference to a &struct bio, either one you have gotten with
 414 *   bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
 415 **/
 416void bio_put(struct bio *bio)
 417{
 418        BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
 419
 420        /*
 421         * last put frees it
 422         */
 423        if (atomic_dec_and_test(&bio->bi_cnt)) {
 424                bio_disassociate_task(bio);
 425                bio->bi_next = NULL;
 426                bio->bi_destructor(bio);
 427        }
 428}
 429EXPORT_SYMBOL(bio_put);
 430
 431inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
 432{
 433        if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
 434                blk_recount_segments(q, bio);
 435
 436        return bio->bi_phys_segments;
 437}
 438EXPORT_SYMBOL(bio_phys_segments);
 439
 440/**
 441 *      __bio_clone     -       clone a bio
 442 *      @bio: destination bio
 443 *      @bio_src: bio to clone
 444 *
 445 *      Clone a &bio. Caller will own the returned bio, but not
 446 *      the actual data it points to. Reference count of returned
 447 *      bio will be one.
 448 */
 449void __bio_clone(struct bio *bio, struct bio *bio_src)
 450{
 451        memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
 452                bio_src->bi_max_vecs * sizeof(struct bio_vec));
 453
 454        /*
 455         * most users will be overriding ->bi_bdev with a new target,
 456         * so we don't set nor calculate new physical/hw segment counts here
 457         */
 458        bio->bi_sector = bio_src->bi_sector;
 459        bio->bi_bdev = bio_src->bi_bdev;
 460        bio->bi_flags |= 1 << BIO_CLONED;
 461        bio->bi_rw = bio_src->bi_rw;
 462        bio->bi_vcnt = bio_src->bi_vcnt;
 463        bio->bi_size = bio_src->bi_size;
 464        bio->bi_idx = bio_src->bi_idx;
 465}
 466EXPORT_SYMBOL(__bio_clone);
 467
 468/**
 469 *      bio_clone       -       clone a bio
 470 *      @bio: bio to clone
 471 *      @gfp_mask: allocation priority
 472 *
 473 *      Like __bio_clone, only also allocates the returned bio
 474 */
 475struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
 476{
 477        struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
 478
 479        if (!b)
 480                return NULL;
 481
 482        b->bi_destructor = bio_fs_destructor;
 483        __bio_clone(b, bio);
 484
 485        if (bio_integrity(bio)) {
 486                int ret;
 487
 488                ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
 489
 490                if (ret < 0) {
 491                        bio_put(b);
 492                        return NULL;
 493                }
 494        }
 495
 496        return b;
 497}
 498EXPORT_SYMBOL(bio_clone);
 499
 500/**
 501 *      bio_get_nr_vecs         - return approx number of vecs
 502 *      @bdev:  I/O target
 503 *
 504 *      Return the approximate number of pages we can send to this target.
 505 *      There's no guarantee that you will be able to fit this number of pages
 506 *      into a bio, it does not account for dynamic restrictions that vary
 507 *      on offset.
 508 */
 509int bio_get_nr_vecs(struct block_device *bdev)
 510{
 511        struct request_queue *q = bdev_get_queue(bdev);
 512        int nr_pages;
 513
 514        nr_pages = min_t(unsigned,
 515                     queue_max_segments(q),
 516                     queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
 517
 518        return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
 519
 520}
 521EXPORT_SYMBOL(bio_get_nr_vecs);
 522
 523static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
 524                          *page, unsigned int len, unsigned int offset,
 525                          unsigned short max_sectors)
 526{
 527        int retried_segments = 0;
 528        struct bio_vec *bvec;
 529
 530        /*
 531         * cloned bio must not modify vec list
 532         */
 533        if (unlikely(bio_flagged(bio, BIO_CLONED)))
 534                return 0;
 535
 536        if (((bio->bi_size + len) >> 9) > max_sectors)
 537                return 0;
 538
 539        /*
 540         * For filesystems with a blocksize smaller than the pagesize
 541         * we will often be called with the same page as last time and
 542         * a consecutive offset.  Optimize this special case.
 543         */
 544        if (bio->bi_vcnt > 0) {
 545                struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
 546
 547                if (page == prev->bv_page &&
 548                    offset == prev->bv_offset + prev->bv_len) {
 549                        unsigned int prev_bv_len = prev->bv_len;
 550                        prev->bv_len += len;
 551
 552                        if (q->merge_bvec_fn) {
 553                                struct bvec_merge_data bvm = {
 554                                        /* prev_bvec is already charged in
 555                                           bi_size, discharge it in order to
 556                                           simulate merging updated prev_bvec
 557                                           as new bvec. */
 558                                        .bi_bdev = bio->bi_bdev,
 559                                        .bi_sector = bio->bi_sector,
 560                                        .bi_size = bio->bi_size - prev_bv_len,
 561                                        .bi_rw = bio->bi_rw,
 562                                };
 563
 564                                if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
 565                                        prev->bv_len -= len;
 566                                        return 0;
 567                                }
 568                        }
 569
 570                        goto done;
 571                }
 572        }
 573
 574        if (bio->bi_vcnt >= bio->bi_max_vecs)
 575                return 0;
 576
 577        /*
 578         * we might lose a segment or two here, but rather that than
 579         * make this too complex.
 580         */
 581
 582        while (bio->bi_phys_segments >= queue_max_segments(q)) {
 583
 584                if (retried_segments)
 585                        return 0;
 586
 587                retried_segments = 1;
 588                blk_recount_segments(q, bio);
 589        }
 590
 591        /*
 592         * setup the new entry, we might clear it again later if we
 593         * cannot add the page
 594         */
 595        bvec = &bio->bi_io_vec[bio->bi_vcnt];
 596        bvec->bv_page = page;
 597        bvec->bv_len = len;
 598        bvec->bv_offset = offset;
 599
 600        /*
 601         * if queue has other restrictions (eg varying max sector size
 602         * depending on offset), it can specify a merge_bvec_fn in the
 603         * queue to get further control
 604         */
 605        if (q->merge_bvec_fn) {
 606                struct bvec_merge_data bvm = {
 607                        .bi_bdev = bio->bi_bdev,
 608                        .bi_sector = bio->bi_sector,
 609                        .bi_size = bio->bi_size,
 610                        .bi_rw = bio->bi_rw,
 611                };
 612
 613                /*
 614                 * merge_bvec_fn() returns number of bytes it can accept
 615                 * at this offset
 616                 */
 617                if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
 618                        bvec->bv_page = NULL;
 619                        bvec->bv_len = 0;
 620                        bvec->bv_offset = 0;
 621                        return 0;
 622                }
 623        }
 624
 625        /* If we may be able to merge these biovecs, force a recount */
 626        if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
 627                bio->bi_flags &= ~(1 << BIO_SEG_VALID);
 628
 629        bio->bi_vcnt++;
 630        bio->bi_phys_segments++;
 631 done:
 632        bio->bi_size += len;
 633        return len;
 634}
 635
 636/**
 637 *      bio_add_pc_page -       attempt to add page to bio
 638 *      @q: the target queue
 639 *      @bio: destination bio
 640 *      @page: page to add
 641 *      @len: vec entry length
 642 *      @offset: vec entry offset
 643 *
 644 *      Attempt to add a page to the bio_vec maplist. This can fail for a
 645 *      number of reasons, such as the bio being full or target block device
 646 *      limitations. The target block device must allow bio's up to PAGE_SIZE,
 647 *      so it is always possible to add a single page to an empty bio.
 648 *
 649 *      This should only be used by REQ_PC bios.
 650 */
 651int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
 652                    unsigned int len, unsigned int offset)
 653{
 654        return __bio_add_page(q, bio, page, len, offset,
 655                              queue_max_hw_sectors(q));
 656}
 657EXPORT_SYMBOL(bio_add_pc_page);
 658
 659/**
 660 *      bio_add_page    -       attempt to add page to bio
 661 *      @bio: destination bio
 662 *      @page: page to add
 663 *      @len: vec entry length
 664 *      @offset: vec entry offset
 665 *
 666 *      Attempt to add a page to the bio_vec maplist. This can fail for a
 667 *      number of reasons, such as the bio being full or target block device
 668 *      limitations. The target block device must allow bio's up to PAGE_SIZE,
 669 *      so it is always possible to add a single page to an empty bio.
 670 */
 671int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
 672                 unsigned int offset)
 673{
 674        struct request_queue *q = bdev_get_queue(bio->bi_bdev);
 675        return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
 676}
 677EXPORT_SYMBOL(bio_add_page);
 678
 679struct bio_map_data {
 680        struct bio_vec *iovecs;
 681        struct sg_iovec *sgvecs;
 682        int nr_sgvecs;
 683        int is_our_pages;
 684};
 685
 686static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
 687                             struct sg_iovec *iov, int iov_count,
 688                             int is_our_pages)
 689{
 690        memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
 691        memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
 692        bmd->nr_sgvecs = iov_count;
 693        bmd->is_our_pages = is_our_pages;
 694        bio->bi_private = bmd;
 695}
 696
 697static void bio_free_map_data(struct bio_map_data *bmd)
 698{
 699        kfree(bmd->iovecs);
 700        kfree(bmd->sgvecs);
 701        kfree(bmd);
 702}
 703
 704static struct bio_map_data *bio_alloc_map_data(int nr_segs,
 705                                               unsigned int iov_count,
 706                                               gfp_t gfp_mask)
 707{
 708        struct bio_map_data *bmd;
 709
 710        if (iov_count > UIO_MAXIOV)
 711                return NULL;
 712
 713        bmd = kmalloc(sizeof(*bmd), gfp_mask);
 714        if (!bmd)
 715                return NULL;
 716
 717        bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
 718        if (!bmd->iovecs) {
 719                kfree(bmd);
 720                return NULL;
 721        }
 722
 723        bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
 724        if (bmd->sgvecs)
 725                return bmd;
 726
 727        kfree(bmd->iovecs);
 728        kfree(bmd);
 729        return NULL;
 730}
 731
 732static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
 733                          struct sg_iovec *iov, int iov_count,
 734                          int to_user, int from_user, int do_free_page)
 735{
 736        int ret = 0, i;
 737        struct bio_vec *bvec;
 738        int iov_idx = 0;
 739        unsigned int iov_off = 0;
 740
 741        __bio_for_each_segment(bvec, bio, i, 0) {
 742                char *bv_addr = page_address(bvec->bv_page);
 743                unsigned int bv_len = iovecs[i].bv_len;
 744
 745                while (bv_len && iov_idx < iov_count) {
 746                        unsigned int bytes;
 747                        char __user *iov_addr;
 748
 749                        bytes = min_t(unsigned int,
 750                                      iov[iov_idx].iov_len - iov_off, bv_len);
 751                        iov_addr = iov[iov_idx].iov_base + iov_off;
 752
 753                        if (!ret) {
 754                                if (to_user)
 755                                        ret = copy_to_user(iov_addr, bv_addr,
 756                                                           bytes);
 757
 758                                if (from_user)
 759                                        ret = copy_from_user(bv_addr, iov_addr,
 760                                                             bytes);
 761
 762                                if (ret)
 763                                        ret = -EFAULT;
 764                        }
 765
 766                        bv_len -= bytes;
 767                        bv_addr += bytes;
 768                        iov_addr += bytes;
 769                        iov_off += bytes;
 770
 771                        if (iov[iov_idx].iov_len == iov_off) {
 772                                iov_idx++;
 773                                iov_off = 0;
 774                        }
 775                }
 776
 777                if (do_free_page)
 778                        __free_page(bvec->bv_page);
 779        }
 780
 781        return ret;
 782}
 783
 784/**
 785 *      bio_uncopy_user -       finish previously mapped bio
 786 *      @bio: bio being terminated
 787 *
 788 *      Free pages allocated from bio_copy_user() and write back data
 789 *      to user space in case of a read.
 790 */
 791int bio_uncopy_user(struct bio *bio)
 792{
 793        struct bio_map_data *bmd = bio->bi_private;
 794        int ret = 0;
 795
 796        if (!bio_flagged(bio, BIO_NULL_MAPPED))
 797                ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
 798                                     bmd->nr_sgvecs, bio_data_dir(bio) == READ,
 799                                     0, bmd->is_our_pages);
 800        bio_free_map_data(bmd);
 801        bio_put(bio);
 802        return ret;
 803}
 804EXPORT_SYMBOL(bio_uncopy_user);
 805
 806/**
 807 *      bio_copy_user_iov       -       copy user data to bio
 808 *      @q: destination block queue
 809 *      @map_data: pointer to the rq_map_data holding pages (if necessary)
 810 *      @iov:   the iovec.
 811 *      @iov_count: number of elements in the iovec
 812 *      @write_to_vm: bool indicating writing to pages or not
 813 *      @gfp_mask: memory allocation flags
 814 *
 815 *      Prepares and returns a bio for indirect user io, bouncing data
 816 *      to/from kernel pages as necessary. Must be paired with
 817 *      call bio_uncopy_user() on io completion.
 818 */
 819struct bio *bio_copy_user_iov(struct request_queue *q,
 820                              struct rq_map_data *map_data,
 821                              struct sg_iovec *iov, int iov_count,
 822                              int write_to_vm, gfp_t gfp_mask)
 823{
 824        struct bio_map_data *bmd;
 825        struct bio_vec *bvec;
 826        struct page *page;
 827        struct bio *bio;
 828        int i, ret;
 829        int nr_pages = 0;
 830        unsigned int len = 0;
 831        unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
 832
 833        for (i = 0; i < iov_count; i++) {
 834                unsigned long uaddr;
 835                unsigned long end;
 836                unsigned long start;
 837
 838                uaddr = (unsigned long)iov[i].iov_base;
 839                end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 840                start = uaddr >> PAGE_SHIFT;
 841
 842                /*
 843                 * Overflow, abort
 844                 */
 845                if (end < start)
 846                        return ERR_PTR(-EINVAL);
 847
 848                nr_pages += end - start;
 849                len += iov[i].iov_len;
 850        }
 851
 852        if (offset)
 853                nr_pages++;
 854
 855        bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
 856        if (!bmd)
 857                return ERR_PTR(-ENOMEM);
 858
 859        ret = -ENOMEM;
 860        bio = bio_kmalloc(gfp_mask, nr_pages);
 861        if (!bio)
 862                goto out_bmd;
 863
 864        if (!write_to_vm)
 865                bio->bi_rw |= REQ_WRITE;
 866
 867        ret = 0;
 868
 869        if (map_data) {
 870                nr_pages = 1 << map_data->page_order;
 871                i = map_data->offset / PAGE_SIZE;
 872        }
 873        while (len) {
 874                unsigned int bytes = PAGE_SIZE;
 875
 876                bytes -= offset;
 877
 878                if (bytes > len)
 879                        bytes = len;
 880
 881                if (map_data) {
 882                        if (i == map_data->nr_entries * nr_pages) {
 883                                ret = -ENOMEM;
 884                                break;
 885                        }
 886
 887                        page = map_data->pages[i / nr_pages];
 888                        page += (i % nr_pages);
 889
 890                        i++;
 891                } else {
 892                        page = alloc_page(q->bounce_gfp | gfp_mask);
 893                        if (!page) {
 894                                ret = -ENOMEM;
 895                                break;
 896                        }
 897                }
 898
 899                if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
 900                        break;
 901
 902                len -= bytes;
 903                offset = 0;
 904        }
 905
 906        if (ret)
 907                goto cleanup;
 908
 909        /*
 910         * success
 911         */
 912        if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
 913            (map_data && map_data->from_user)) {
 914                ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
 915                if (ret)
 916                        goto cleanup;
 917        }
 918
 919        bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
 920        return bio;
 921cleanup:
 922        if (!map_data)
 923                bio_for_each_segment(bvec, bio, i)
 924                        __free_page(bvec->bv_page);
 925
 926        bio_put(bio);
 927out_bmd:
 928        bio_free_map_data(bmd);
 929        return ERR_PTR(ret);
 930}
 931
 932/**
 933 *      bio_copy_user   -       copy user data to bio
 934 *      @q: destination block queue
 935 *      @map_data: pointer to the rq_map_data holding pages (if necessary)
 936 *      @uaddr: start of user address
 937 *      @len: length in bytes
 938 *      @write_to_vm: bool indicating writing to pages or not
 939 *      @gfp_mask: memory allocation flags
 940 *
 941 *      Prepares and returns a bio for indirect user io, bouncing data
 942 *      to/from kernel pages as necessary. Must be paired with
 943 *      call bio_uncopy_user() on io completion.
 944 */
 945struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
 946                          unsigned long uaddr, unsigned int len,
 947                          int write_to_vm, gfp_t gfp_mask)
 948{
 949        struct sg_iovec iov;
 950
 951        iov.iov_base = (void __user *)uaddr;
 952        iov.iov_len = len;
 953
 954        return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
 955}
 956EXPORT_SYMBOL(bio_copy_user);
 957
 958static struct bio *__bio_map_user_iov(struct request_queue *q,
 959                                      struct block_device *bdev,
 960                                      struct sg_iovec *iov, int iov_count,
 961                                      int write_to_vm, gfp_t gfp_mask)
 962{
 963        int i, j;
 964        int nr_pages = 0;
 965        struct page **pages;
 966        struct bio *bio;
 967        int cur_page = 0;
 968        int ret, offset;
 969
 970        for (i = 0; i < iov_count; i++) {
 971                unsigned long uaddr = (unsigned long)iov[i].iov_base;
 972                unsigned long len = iov[i].iov_len;
 973                unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 974                unsigned long start = uaddr >> PAGE_SHIFT;
 975
 976                /*
 977                 * Overflow, abort
 978                 */
 979                if (end < start)
 980                        return ERR_PTR(-EINVAL);
 981
 982                nr_pages += end - start;
 983                /*
 984                 * buffer must be aligned to at least hardsector size for now
 985                 */
 986                if (uaddr & queue_dma_alignment(q))
 987                        return ERR_PTR(-EINVAL);
 988        }
 989
 990        if (!nr_pages)
 991                return ERR_PTR(-EINVAL);
 992
 993        bio = bio_kmalloc(gfp_mask, nr_pages);
 994        if (!bio)
 995                return ERR_PTR(-ENOMEM);
 996
 997        ret = -ENOMEM;
 998        pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
 999        if (!pages)
1000                goto out;
1001
1002        for (i = 0; i < iov_count; i++) {
1003                unsigned long uaddr = (unsigned long)iov[i].iov_base;
1004                unsigned long len = iov[i].iov_len;
1005                unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1006                unsigned long start = uaddr >> PAGE_SHIFT;
1007                const int local_nr_pages = end - start;
1008                const int page_limit = cur_page + local_nr_pages;
1009
1010                ret = get_user_pages_fast(uaddr, local_nr_pages,
1011                                write_to_vm, &pages[cur_page]);
1012                if (ret < local_nr_pages) {
1013                        ret = -EFAULT;
1014                        goto out_unmap;
1015                }
1016
1017                offset = uaddr & ~PAGE_MASK;
1018                for (j = cur_page; j < page_limit; j++) {
1019                        unsigned int bytes = PAGE_SIZE - offset;
1020
1021                        if (len <= 0)
1022                                break;
1023                        
1024                        if (bytes > len)
1025                                bytes = len;
1026
1027                        /*
1028                         * sorry...
1029                         */
1030                        if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1031                                            bytes)
1032                                break;
1033
1034                        len -= bytes;
1035                        offset = 0;
1036                }
1037
1038                cur_page = j;
1039                /*
1040                 * release the pages we didn't map into the bio, if any
1041                 */
1042                while (j < page_limit)
1043                        page_cache_release(pages[j++]);
1044        }
1045
1046        kfree(pages);
1047
1048        /*
1049         * set data direction, and check if mapped pages need bouncing
1050         */
1051        if (!write_to_vm)
1052                bio->bi_rw |= REQ_WRITE;
1053
1054        bio->bi_bdev = bdev;
1055        bio->bi_flags |= (1 << BIO_USER_MAPPED);
1056        return bio;
1057
1058 out_unmap:
1059        for (i = 0; i < nr_pages; i++) {
1060                if(!pages[i])
1061                        break;
1062                page_cache_release(pages[i]);
1063        }
1064 out:
1065        kfree(pages);
1066        bio_put(bio);
1067        return ERR_PTR(ret);
1068}
1069
1070/**
1071 *      bio_map_user    -       map user address into bio
1072 *      @q: the struct request_queue for the bio
1073 *      @bdev: destination block device
1074 *      @uaddr: start of user address
1075 *      @len: length in bytes
1076 *      @write_to_vm: bool indicating writing to pages or not
1077 *      @gfp_mask: memory allocation flags
1078 *
1079 *      Map the user space address into a bio suitable for io to a block
1080 *      device. Returns an error pointer in case of error.
1081 */
1082struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1083                         unsigned long uaddr, unsigned int len, int write_to_vm,
1084                         gfp_t gfp_mask)
1085{
1086        struct sg_iovec iov;
1087
1088        iov.iov_base = (void __user *)uaddr;
1089        iov.iov_len = len;
1090
1091        return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1092}
1093EXPORT_SYMBOL(bio_map_user);
1094
1095/**
1096 *      bio_map_user_iov - map user sg_iovec table into bio
1097 *      @q: the struct request_queue for the bio
1098 *      @bdev: destination block device
1099 *      @iov:   the iovec.
1100 *      @iov_count: number of elements in the iovec
1101 *      @write_to_vm: bool indicating writing to pages or not
1102 *      @gfp_mask: memory allocation flags
1103 *
1104 *      Map the user space address into a bio suitable for io to a block
1105 *      device. Returns an error pointer in case of error.
1106 */
1107struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1108                             struct sg_iovec *iov, int iov_count,
1109                             int write_to_vm, gfp_t gfp_mask)
1110{
1111        struct bio *bio;
1112
1113        bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1114                                 gfp_mask);
1115        if (IS_ERR(bio))
1116                return bio;
1117
1118        /*
1119         * subtle -- if __bio_map_user() ended up bouncing a bio,
1120         * it would normally disappear when its bi_end_io is run.
1121         * however, we need it for the unmap, so grab an extra
1122         * reference to it
1123         */
1124        bio_get(bio);
1125
1126        return bio;
1127}
1128
1129static void __bio_unmap_user(struct bio *bio)
1130{
1131        struct bio_vec *bvec;
1132        int i;
1133
1134        /*
1135         * make sure we dirty pages we wrote to
1136         */
1137        __bio_for_each_segment(bvec, bio, i, 0) {
1138                if (bio_data_dir(bio) == READ)
1139                        set_page_dirty_lock(bvec->bv_page);
1140
1141                page_cache_release(bvec->bv_page);
1142        }
1143
1144        bio_put(bio);
1145}
1146
1147/**
1148 *      bio_unmap_user  -       unmap a bio
1149 *      @bio:           the bio being unmapped
1150 *
1151 *      Unmap a bio previously mapped by bio_map_user(). Must be called with
1152 *      a process context.
1153 *
1154 *      bio_unmap_user() may sleep.
1155 */
1156void bio_unmap_user(struct bio *bio)
1157{
1158        __bio_unmap_user(bio);
1159        bio_put(bio);
1160}
1161EXPORT_SYMBOL(bio_unmap_user);
1162
1163static void bio_map_kern_endio(struct bio *bio, int err)
1164{
1165        bio_put(bio);
1166}
1167
1168static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1169                                  unsigned int len, gfp_t gfp_mask)
1170{
1171        unsigned long kaddr = (unsigned long)data;
1172        unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1173        unsigned long start = kaddr >> PAGE_SHIFT;
1174        const int nr_pages = end - start;
1175        int offset, i;
1176        struct bio *bio;
1177
1178        bio = bio_kmalloc(gfp_mask, nr_pages);
1179        if (!bio)
1180                return ERR_PTR(-ENOMEM);
1181
1182        offset = offset_in_page(kaddr);
1183        for (i = 0; i < nr_pages; i++) {
1184                unsigned int bytes = PAGE_SIZE - offset;
1185
1186                if (len <= 0)
1187                        break;
1188
1189                if (bytes > len)
1190                        bytes = len;
1191
1192                if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1193                                    offset) < bytes)
1194                        break;
1195
1196                data += bytes;
1197                len -= bytes;
1198                offset = 0;
1199        }
1200
1201        bio->bi_end_io = bio_map_kern_endio;
1202        return bio;
1203}
1204
1205/**
1206 *      bio_map_kern    -       map kernel address into bio
1207 *      @q: the struct request_queue for the bio
1208 *      @data: pointer to buffer to map
1209 *      @len: length in bytes
1210 *      @gfp_mask: allocation flags for bio allocation
1211 *
1212 *      Map the kernel address into a bio suitable for io to a block
1213 *      device. Returns an error pointer in case of error.
1214 */
1215struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1216                         gfp_t gfp_mask)
1217{
1218        struct bio *bio;
1219
1220        bio = __bio_map_kern(q, data, len, gfp_mask);
1221        if (IS_ERR(bio))
1222                return bio;
1223
1224        if (bio->bi_size == len)
1225                return bio;
1226
1227        /*
1228         * Don't support partial mappings.
1229         */
1230        bio_put(bio);
1231        return ERR_PTR(-EINVAL);
1232}
1233EXPORT_SYMBOL(bio_map_kern);
1234
1235static void bio_copy_kern_endio(struct bio *bio, int err)
1236{
1237        struct bio_vec *bvec;
1238        const int read = bio_data_dir(bio) == READ;
1239        struct bio_map_data *bmd = bio->bi_private;
1240        int i;
1241        char *p = bmd->sgvecs[0].iov_base;
1242
1243        __bio_for_each_segment(bvec, bio, i, 0) {
1244                char *addr = page_address(bvec->bv_page);
1245                int len = bmd->iovecs[i].bv_len;
1246
1247                if (read)
1248                        memcpy(p, addr, len);
1249
1250                __free_page(bvec->bv_page);
1251                p += len;
1252        }
1253
1254        bio_free_map_data(bmd);
1255        bio_put(bio);
1256}
1257
1258/**
1259 *      bio_copy_kern   -       copy kernel address into bio
1260 *      @q: the struct request_queue for the bio
1261 *      @data: pointer to buffer to copy
1262 *      @len: length in bytes
1263 *      @gfp_mask: allocation flags for bio and page allocation
1264 *      @reading: data direction is READ
1265 *
1266 *      copy the kernel address into a bio suitable for io to a block
1267 *      device. Returns an error pointer in case of error.
1268 */
1269struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1270                          gfp_t gfp_mask, int reading)
1271{
1272        struct bio *bio;
1273        struct bio_vec *bvec;
1274        int i;
1275
1276        bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1277        if (IS_ERR(bio))
1278                return bio;
1279
1280        if (!reading) {
1281                void *p = data;
1282
1283                bio_for_each_segment(bvec, bio, i) {
1284                        char *addr = page_address(bvec->bv_page);
1285
1286                        memcpy(addr, p, bvec->bv_len);
1287                        p += bvec->bv_len;
1288                }
1289        }
1290
1291        bio->bi_end_io = bio_copy_kern_endio;
1292
1293        return bio;
1294}
1295EXPORT_SYMBOL(bio_copy_kern);
1296
1297/*
1298 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1299 * for performing direct-IO in BIOs.
1300 *
1301 * The problem is that we cannot run set_page_dirty() from interrupt context
1302 * because the required locks are not interrupt-safe.  So what we can do is to
1303 * mark the pages dirty _before_ performing IO.  And in interrupt context,
1304 * check that the pages are still dirty.   If so, fine.  If not, redirty them
1305 * in process context.
1306 *
1307 * We special-case compound pages here: normally this means reads into hugetlb
1308 * pages.  The logic in here doesn't really work right for compound pages
1309 * because the VM does not uniformly chase down the head page in all cases.
1310 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1311 * handle them at all.  So we skip compound pages here at an early stage.
1312 *
1313 * Note that this code is very hard to test under normal circumstances because
1314 * direct-io pins the pages with get_user_pages().  This makes
1315 * is_page_cache_freeable return false, and the VM will not clean the pages.
1316 * But other code (eg, flusher threads) could clean the pages if they are mapped
1317 * pagecache.
1318 *
1319 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1320 * deferred bio dirtying paths.
1321 */
1322
1323/*
1324 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1325 */
1326void bio_set_pages_dirty(struct bio *bio)
1327{
1328        struct bio_vec *bvec = bio->bi_io_vec;
1329        int i;
1330
1331        for (i = 0; i < bio->bi_vcnt; i++) {
1332                struct page *page = bvec[i].bv_page;
1333
1334                if (page && !PageCompound(page))
1335                        set_page_dirty_lock(page);
1336        }
1337}
1338
1339static void bio_release_pages(struct bio *bio)
1340{
1341        struct bio_vec *bvec = bio->bi_io_vec;
1342        int i;
1343
1344        for (i = 0; i < bio->bi_vcnt; i++) {
1345                struct page *page = bvec[i].bv_page;
1346
1347                if (page)
1348                        put_page(page);
1349        }
1350}
1351
1352/*
1353 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1354 * If they are, then fine.  If, however, some pages are clean then they must
1355 * have been written out during the direct-IO read.  So we take another ref on
1356 * the BIO and the offending pages and re-dirty the pages in process context.
1357 *
1358 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1359 * here on.  It will run one page_cache_release() against each page and will
1360 * run one bio_put() against the BIO.
1361 */
1362
1363static void bio_dirty_fn(struct work_struct *work);
1364
1365static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1366static DEFINE_SPINLOCK(bio_dirty_lock);
1367static struct bio *bio_dirty_list;
1368
1369/*
1370 * This runs in process context
1371 */
1372static void bio_dirty_fn(struct work_struct *work)
1373{
1374        unsigned long flags;
1375        struct bio *bio;
1376
1377        spin_lock_irqsave(&bio_dirty_lock, flags);
1378        bio = bio_dirty_list;
1379        bio_dirty_list = NULL;
1380        spin_unlock_irqrestore(&bio_dirty_lock, flags);
1381
1382        while (bio) {
1383                struct bio *next = bio->bi_private;
1384
1385                bio_set_pages_dirty(bio);
1386                bio_release_pages(bio);
1387                bio_put(bio);
1388                bio = next;
1389        }
1390}
1391
1392void bio_check_pages_dirty(struct bio *bio)
1393{
1394        struct bio_vec *bvec = bio->bi_io_vec;
1395        int nr_clean_pages = 0;
1396        int i;
1397
1398        for (i = 0; i < bio->bi_vcnt; i++) {
1399                struct page *page = bvec[i].bv_page;
1400
1401                if (PageDirty(page) || PageCompound(page)) {
1402                        page_cache_release(page);
1403                        bvec[i].bv_page = NULL;
1404                } else {
1405                        nr_clean_pages++;
1406                }
1407        }
1408
1409        if (nr_clean_pages) {
1410                unsigned long flags;
1411
1412                spin_lock_irqsave(&bio_dirty_lock, flags);
1413                bio->bi_private = bio_dirty_list;
1414                bio_dirty_list = bio;
1415                spin_unlock_irqrestore(&bio_dirty_lock, flags);
1416                schedule_work(&bio_dirty_work);
1417        } else {
1418                bio_put(bio);
1419        }
1420}
1421
1422#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1423void bio_flush_dcache_pages(struct bio *bi)
1424{
1425        int i;
1426        struct bio_vec *bvec;
1427
1428        bio_for_each_segment(bvec, bi, i)
1429                flush_dcache_page(bvec->bv_page);
1430}
1431EXPORT_SYMBOL(bio_flush_dcache_pages);
1432#endif
1433
1434/**
1435 * bio_endio - end I/O on a bio
1436 * @bio:        bio
1437 * @error:      error, if any
1438 *
1439 * Description:
1440 *   bio_endio() will end I/O on the whole bio. bio_endio() is the
1441 *   preferred way to end I/O on a bio, it takes care of clearing
1442 *   BIO_UPTODATE on error. @error is 0 on success, and and one of the
1443 *   established -Exxxx (-EIO, for instance) error values in case
1444 *   something went wrong. No one should call bi_end_io() directly on a
1445 *   bio unless they own it and thus know that it has an end_io
1446 *   function.
1447 **/
1448void bio_endio(struct bio *bio, int error)
1449{
1450        if (error)
1451                clear_bit(BIO_UPTODATE, &bio->bi_flags);
1452        else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1453                error = -EIO;
1454
1455        if (bio->bi_end_io)
1456                bio->bi_end_io(bio, error);
1457}
1458EXPORT_SYMBOL(bio_endio);
1459
1460void bio_pair_release(struct bio_pair *bp)
1461{
1462        if (atomic_dec_and_test(&bp->cnt)) {
1463                struct bio *master = bp->bio1.bi_private;
1464
1465                bio_endio(master, bp->error);
1466                mempool_free(bp, bp->bio2.bi_private);
1467        }
1468}
1469EXPORT_SYMBOL(bio_pair_release);
1470
1471static void bio_pair_end_1(struct bio *bi, int err)
1472{
1473        struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1474
1475        if (err)
1476                bp->error = err;
1477
1478        bio_pair_release(bp);
1479}
1480
1481static void bio_pair_end_2(struct bio *bi, int err)
1482{
1483        struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1484
1485        if (err)
1486                bp->error = err;
1487
1488        bio_pair_release(bp);
1489}
1490
1491/*
1492 * split a bio - only worry about a bio with a single page in its iovec
1493 */
1494struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1495{
1496        struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1497
1498        if (!bp)
1499                return bp;
1500
1501        trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1502                                bi->bi_sector + first_sectors);
1503
1504        BUG_ON(bi->bi_vcnt != 1);
1505        BUG_ON(bi->bi_idx != 0);
1506        atomic_set(&bp->cnt, 3);
1507        bp->error = 0;
1508        bp->bio1 = *bi;
1509        bp->bio2 = *bi;
1510        bp->bio2.bi_sector += first_sectors;
1511        bp->bio2.bi_size -= first_sectors << 9;
1512        bp->bio1.bi_size = first_sectors << 9;
1513
1514        bp->bv1 = bi->bi_io_vec[0];
1515        bp->bv2 = bi->bi_io_vec[0];
1516        bp->bv2.bv_offset += first_sectors << 9;
1517        bp->bv2.bv_len -= first_sectors << 9;
1518        bp->bv1.bv_len = first_sectors << 9;
1519
1520        bp->bio1.bi_io_vec = &bp->bv1;
1521        bp->bio2.bi_io_vec = &bp->bv2;
1522
1523        bp->bio1.bi_max_vecs = 1;
1524        bp->bio2.bi_max_vecs = 1;
1525
1526        bp->bio1.bi_end_io = bio_pair_end_1;
1527        bp->bio2.bi_end_io = bio_pair_end_2;
1528
1529        bp->bio1.bi_private = bi;
1530        bp->bio2.bi_private = bio_split_pool;
1531
1532        if (bio_integrity(bi))
1533                bio_integrity_split(bi, bp, first_sectors);
1534
1535        return bp;
1536}
1537EXPORT_SYMBOL(bio_split);
1538
1539/**
1540 *      bio_sector_offset - Find hardware sector offset in bio
1541 *      @bio:           bio to inspect
1542 *      @index:         bio_vec index
1543 *      @offset:        offset in bv_page
1544 *
1545 *      Return the number of hardware sectors between beginning of bio
1546 *      and an end point indicated by a bio_vec index and an offset
1547 *      within that vector's page.
1548 */
1549sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1550                           unsigned int offset)
1551{
1552        unsigned int sector_sz;
1553        struct bio_vec *bv;
1554        sector_t sectors;
1555        int i;
1556
1557        sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1558        sectors = 0;
1559
1560        if (index >= bio->bi_idx)
1561                index = bio->bi_vcnt - 1;
1562
1563        __bio_for_each_segment(bv, bio, i, 0) {
1564                if (i == index) {
1565                        if (offset > bv->bv_offset)
1566                                sectors += (offset - bv->bv_offset) / sector_sz;
1567                        break;
1568                }
1569
1570                sectors += bv->bv_len / sector_sz;
1571        }
1572
1573        return sectors;
1574}
1575EXPORT_SYMBOL(bio_sector_offset);
1576
1577/*
1578 * create memory pools for biovec's in a bio_set.
1579 * use the global biovec slabs created for general use.
1580 */
1581static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1582{
1583        struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1584
1585        bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1586        if (!bs->bvec_pool)
1587                return -ENOMEM;
1588
1589        return 0;
1590}
1591
1592static void biovec_free_pools(struct bio_set *bs)
1593{
1594        mempool_destroy(bs->bvec_pool);
1595}
1596
1597void bioset_free(struct bio_set *bs)
1598{
1599        if (bs->bio_pool)
1600                mempool_destroy(bs->bio_pool);
1601
1602        bioset_integrity_free(bs);
1603        biovec_free_pools(bs);
1604        bio_put_slab(bs);
1605
1606        kfree(bs);
1607}
1608EXPORT_SYMBOL(bioset_free);
1609
1610/**
1611 * bioset_create  - Create a bio_set
1612 * @pool_size:  Number of bio and bio_vecs to cache in the mempool
1613 * @front_pad:  Number of bytes to allocate in front of the returned bio
1614 *
1615 * Description:
1616 *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1617 *    to ask for a number of bytes to be allocated in front of the bio.
1618 *    Front pad allocation is useful for embedding the bio inside
1619 *    another structure, to avoid allocating extra data to go with the bio.
1620 *    Note that the bio must be embedded at the END of that structure always,
1621 *    or things will break badly.
1622 */
1623struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1624{
1625        unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1626        struct bio_set *bs;
1627
1628        bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1629        if (!bs)
1630                return NULL;
1631
1632        bs->front_pad = front_pad;
1633
1634        bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1635        if (!bs->bio_slab) {
1636                kfree(bs);
1637                return NULL;
1638        }
1639
1640        bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1641        if (!bs->bio_pool)
1642                goto bad;
1643
1644        if (!biovec_create_pools(bs, pool_size))
1645                return bs;
1646
1647bad:
1648        bioset_free(bs);
1649        return NULL;
1650}
1651EXPORT_SYMBOL(bioset_create);
1652
1653#ifdef CONFIG_BLK_CGROUP
1654/**
1655 * bio_associate_current - associate a bio with %current
1656 * @bio: target bio
1657 *
1658 * Associate @bio with %current if it hasn't been associated yet.  Block
1659 * layer will treat @bio as if it were issued by %current no matter which
1660 * task actually issues it.
1661 *
1662 * This function takes an extra reference of @task's io_context and blkcg
1663 * which will be put when @bio is released.  The caller must own @bio,
1664 * ensure %current->io_context exists, and is responsible for synchronizing
1665 * calls to this function.
1666 */
1667int bio_associate_current(struct bio *bio)
1668{
1669        struct io_context *ioc;
1670        struct cgroup_subsys_state *css;
1671
1672        if (bio->bi_ioc)
1673                return -EBUSY;
1674
1675        ioc = current->io_context;
1676        if (!ioc)
1677                return -ENOENT;
1678
1679        /* acquire active ref on @ioc and associate */
1680        get_io_context_active(ioc);
1681        bio->bi_ioc = ioc;
1682
1683        /* associate blkcg if exists */
1684        rcu_read_lock();
1685        css = task_subsys_state(current, blkio_subsys_id);
1686        if (css && css_tryget(css))
1687                bio->bi_css = css;
1688        rcu_read_unlock();
1689
1690        return 0;
1691}
1692
1693/**
1694 * bio_disassociate_task - undo bio_associate_current()
1695 * @bio: target bio
1696 */
1697void bio_disassociate_task(struct bio *bio)
1698{
1699        if (bio->bi_ioc) {
1700                put_io_context(bio->bi_ioc);
1701                bio->bi_ioc = NULL;
1702        }
1703        if (bio->bi_css) {
1704                css_put(bio->bi_css);
1705                bio->bi_css = NULL;
1706        }
1707}
1708
1709#endif /* CONFIG_BLK_CGROUP */
1710
1711static void __init biovec_init_slabs(void)
1712{
1713        int i;
1714
1715        for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1716                int size;
1717                struct biovec_slab *bvs = bvec_slabs + i;
1718
1719                if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1720                        bvs->slab = NULL;
1721                        continue;
1722                }
1723
1724                size = bvs->nr_vecs * sizeof(struct bio_vec);
1725                bvs->slab = kmem_cache_create(bvs->name, size, 0,
1726                                SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1727        }
1728}
1729
1730static int __init init_bio(void)
1731{
1732        bio_slab_max = 2;
1733        bio_slab_nr = 0;
1734        bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1735        if (!bio_slabs)
1736                panic("bio: can't allocate bios\n");
1737
1738        bio_integrity_init();
1739        biovec_init_slabs();
1740
1741        fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1742        if (!fs_bio_set)
1743                panic("bio: can't allocate bios\n");
1744
1745        if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1746                panic("bio: can't create integrity pool\n");
1747
1748        bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1749                                                     sizeof(struct bio_pair));
1750        if (!bio_split_pool)
1751                panic("bio: can't create split pool\n");
1752
1753        return 0;
1754}
1755subsys_initcall(init_bio);
1756
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.