linux/fs/block_dev.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  Copyright (C) 1991, 1992  Linus Torvalds
   4 *  Copyright (C) 2001  Andrea Arcangeli <andrea@suse.de> SuSE
   5 *  Copyright (C) 2016 - 2020 Christoph Hellwig
   6 */
   7
   8#include <linux/init.h>
   9#include <linux/mm.h>
  10#include <linux/fcntl.h>
  11#include <linux/slab.h>
  12#include <linux/kmod.h>
  13#include <linux/major.h>
  14#include <linux/device_cgroup.h>
  15#include <linux/highmem.h>
  16#include <linux/blkdev.h>
  17#include <linux/backing-dev.h>
  18#include <linux/module.h>
  19#include <linux/blkpg.h>
  20#include <linux/magic.h>
  21#include <linux/buffer_head.h>
  22#include <linux/swap.h>
  23#include <linux/pagevec.h>
  24#include <linux/writeback.h>
  25#include <linux/mpage.h>
  26#include <linux/mount.h>
  27#include <linux/pseudo_fs.h>
  28#include <linux/uio.h>
  29#include <linux/namei.h>
  30#include <linux/log2.h>
  31#include <linux/cleancache.h>
  32#include <linux/task_io_accounting_ops.h>
  33#include <linux/falloc.h>
  34#include <linux/part_stat.h>
  35#include <linux/uaccess.h>
  36#include <linux/suspend.h>
  37#include "internal.h"
  38
  39struct bdev_inode {
  40        struct block_device bdev;
  41        struct inode vfs_inode;
  42};
  43
  44static const struct address_space_operations def_blk_aops;
  45
  46static inline struct bdev_inode *BDEV_I(struct inode *inode)
  47{
  48        return container_of(inode, struct bdev_inode, vfs_inode);
  49}
  50
  51struct block_device *I_BDEV(struct inode *inode)
  52{
  53        return &BDEV_I(inode)->bdev;
  54}
  55EXPORT_SYMBOL(I_BDEV);
  56
  57static void bdev_write_inode(struct block_device *bdev)
  58{
  59        struct inode *inode = bdev->bd_inode;
  60        int ret;
  61
  62        spin_lock(&inode->i_lock);
  63        while (inode->i_state & I_DIRTY) {
  64                spin_unlock(&inode->i_lock);
  65                ret = write_inode_now(inode, true);
  66                if (ret) {
  67                        char name[BDEVNAME_SIZE];
  68                        pr_warn_ratelimited("VFS: Dirty inode writeback failed "
  69                                            "for block device %s (err=%d).\n",
  70                                            bdevname(bdev, name), ret);
  71                }
  72                spin_lock(&inode->i_lock);
  73        }
  74        spin_unlock(&inode->i_lock);
  75}
  76
  77/* Kill _all_ buffers and pagecache , dirty or not.. */
  78static void kill_bdev(struct block_device *bdev)
  79{
  80        struct address_space *mapping = bdev->bd_inode->i_mapping;
  81
  82        if (mapping_empty(mapping))
  83                return;
  84
  85        invalidate_bh_lrus();
  86        truncate_inode_pages(mapping, 0);
  87}
  88
  89/* Invalidate clean unused buffers and pagecache. */
  90void invalidate_bdev(struct block_device *bdev)
  91{
  92        struct address_space *mapping = bdev->bd_inode->i_mapping;
  93
  94        if (mapping->nrpages) {
  95                invalidate_bh_lrus();
  96                lru_add_drain_all();    /* make sure all lru add caches are flushed */
  97                invalidate_mapping_pages(mapping, 0, -1);
  98        }
  99        /* 99% of the time, we don't need to flush the cleancache on the bdev.
 100         * But, for the strange corners, lets be cautious
 101         */
 102        cleancache_invalidate_inode(mapping);
 103}
 104EXPORT_SYMBOL(invalidate_bdev);
 105
 106/*
 107 * Drop all buffers & page cache for given bdev range. This function bails
 108 * with error if bdev has other exclusive owner (such as filesystem).
 109 */
 110int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
 111                        loff_t lstart, loff_t lend)
 112{
 113        /*
 114         * If we don't hold exclusive handle for the device, upgrade to it
 115         * while we discard the buffer cache to avoid discarding buffers
 116         * under live filesystem.
 117         */
 118        if (!(mode & FMODE_EXCL)) {
 119                int err = bd_prepare_to_claim(bdev, truncate_bdev_range);
 120                if (err)
 121                        goto invalidate;
 122        }
 123
 124        truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
 125        if (!(mode & FMODE_EXCL))
 126                bd_abort_claiming(bdev, truncate_bdev_range);
 127        return 0;
 128
 129invalidate:
 130        /*
 131         * Someone else has handle exclusively open. Try invalidating instead.
 132         * The 'end' argument is inclusive so the rounding is safe.
 133         */
 134        return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
 135                                             lstart >> PAGE_SHIFT,
 136                                             lend >> PAGE_SHIFT);
 137}
 138
 139static void set_init_blocksize(struct block_device *bdev)
 140{
 141        unsigned int bsize = bdev_logical_block_size(bdev);
 142        loff_t size = i_size_read(bdev->bd_inode);
 143
 144        while (bsize < PAGE_SIZE) {
 145                if (size & bsize)
 146                        break;
 147                bsize <<= 1;
 148        }
 149        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 150}
 151
 152int set_blocksize(struct block_device *bdev, int size)
 153{
 154        /* Size must be a power of two, and between 512 and PAGE_SIZE */
 155        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
 156                return -EINVAL;
 157
 158        /* Size cannot be smaller than the size supported by the device */
 159        if (size < bdev_logical_block_size(bdev))
 160                return -EINVAL;
 161
 162        /* Don't change the size if it is same as current */
 163        if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
 164                sync_blockdev(bdev);
 165                bdev->bd_inode->i_blkbits = blksize_bits(size);
 166                kill_bdev(bdev);
 167        }
 168        return 0;
 169}
 170
 171EXPORT_SYMBOL(set_blocksize);
 172
 173int sb_set_blocksize(struct super_block *sb, int size)
 174{
 175        if (set_blocksize(sb->s_bdev, size))
 176                return 0;
 177        /* If we get here, we know size is power of two
 178         * and it's value is between 512 and PAGE_SIZE */
 179        sb->s_blocksize = size;
 180        sb->s_blocksize_bits = blksize_bits(size);
 181        return sb->s_blocksize;
 182}
 183
 184EXPORT_SYMBOL(sb_set_blocksize);
 185
 186int sb_min_blocksize(struct super_block *sb, int size)
 187{
 188        int minsize = bdev_logical_block_size(sb->s_bdev);
 189        if (size < minsize)
 190                size = minsize;
 191        return sb_set_blocksize(sb, size);
 192}
 193
 194EXPORT_SYMBOL(sb_min_blocksize);
 195
 196static int
 197blkdev_get_block(struct inode *inode, sector_t iblock,
 198                struct buffer_head *bh, int create)
 199{
 200        bh->b_bdev = I_BDEV(inode);
 201        bh->b_blocknr = iblock;
 202        set_buffer_mapped(bh);
 203        return 0;
 204}
 205
 206static struct inode *bdev_file_inode(struct file *file)
 207{
 208        return file->f_mapping->host;
 209}
 210
 211static unsigned int dio_bio_write_op(struct kiocb *iocb)
 212{
 213        unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
 214
 215        /* avoid the need for a I/O completion work item */
 216        if (iocb->ki_flags & IOCB_DSYNC)
 217                op |= REQ_FUA;
 218        return op;
 219}
 220
 221#define DIO_INLINE_BIO_VECS 4
 222
 223static void blkdev_bio_end_io_simple(struct bio *bio)
 224{
 225        struct task_struct *waiter = bio->bi_private;
 226
 227        WRITE_ONCE(bio->bi_private, NULL);
 228        blk_wake_io_task(waiter);
 229}
 230
 231static ssize_t
 232__blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
 233                unsigned int nr_pages)
 234{
 235        struct file *file = iocb->ki_filp;
 236        struct block_device *bdev = I_BDEV(bdev_file_inode(file));
 237        struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
 238        loff_t pos = iocb->ki_pos;
 239        bool should_dirty = false;
 240        struct bio bio;
 241        ssize_t ret;
 242        blk_qc_t qc;
 243
 244        if ((pos | iov_iter_alignment(iter)) &
 245            (bdev_logical_block_size(bdev) - 1))
 246                return -EINVAL;
 247
 248        if (nr_pages <= DIO_INLINE_BIO_VECS)
 249                vecs = inline_vecs;
 250        else {
 251                vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
 252                                     GFP_KERNEL);
 253                if (!vecs)
 254                        return -ENOMEM;
 255        }
 256
 257        bio_init(&bio, vecs, nr_pages);
 258        bio_set_dev(&bio, bdev);
 259        bio.bi_iter.bi_sector = pos >> 9;
 260        bio.bi_write_hint = iocb->ki_hint;
 261        bio.bi_private = current;
 262        bio.bi_end_io = blkdev_bio_end_io_simple;
 263        bio.bi_ioprio = iocb->ki_ioprio;
 264
 265        ret = bio_iov_iter_get_pages(&bio, iter);
 266        if (unlikely(ret))
 267                goto out;
 268        ret = bio.bi_iter.bi_size;
 269
 270        if (iov_iter_rw(iter) == READ) {
 271                bio.bi_opf = REQ_OP_READ;
 272                if (iter_is_iovec(iter))
 273                        should_dirty = true;
 274        } else {
 275                bio.bi_opf = dio_bio_write_op(iocb);
 276                task_io_account_write(ret);
 277        }
 278        if (iocb->ki_flags & IOCB_NOWAIT)
 279                bio.bi_opf |= REQ_NOWAIT;
 280        if (iocb->ki_flags & IOCB_HIPRI)
 281                bio_set_polled(&bio, iocb);
 282
 283        qc = submit_bio(&bio);
 284        for (;;) {
 285                set_current_state(TASK_UNINTERRUPTIBLE);
 286                if (!READ_ONCE(bio.bi_private))
 287                        break;
 288                if (!(iocb->ki_flags & IOCB_HIPRI) ||
 289                    !blk_poll(bdev_get_queue(bdev), qc, true))
 290                        blk_io_schedule();
 291        }
 292        __set_current_state(TASK_RUNNING);
 293
 294        bio_release_pages(&bio, should_dirty);
 295        if (unlikely(bio.bi_status))
 296                ret = blk_status_to_errno(bio.bi_status);
 297
 298out:
 299        if (vecs != inline_vecs)
 300                kfree(vecs);
 301
 302        bio_uninit(&bio);
 303
 304        return ret;
 305}
 306
 307struct blkdev_dio {
 308        union {
 309                struct kiocb            *iocb;
 310                struct task_struct      *waiter;
 311        };
 312        size_t                  size;
 313        atomic_t                ref;
 314        bool                    multi_bio : 1;
 315        bool                    should_dirty : 1;
 316        bool                    is_sync : 1;
 317        struct bio              bio;
 318};
 319
 320static struct bio_set blkdev_dio_pool;
 321
 322static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
 323{
 324        struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
 325        struct request_queue *q = bdev_get_queue(bdev);
 326
 327        return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
 328}
 329
 330static void blkdev_bio_end_io(struct bio *bio)
 331{
 332        struct blkdev_dio *dio = bio->bi_private;
 333        bool should_dirty = dio->should_dirty;
 334
 335        if (bio->bi_status && !dio->bio.bi_status)
 336                dio->bio.bi_status = bio->bi_status;
 337
 338        if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
 339                if (!dio->is_sync) {
 340                        struct kiocb *iocb = dio->iocb;
 341                        ssize_t ret;
 342
 343                        if (likely(!dio->bio.bi_status)) {
 344                                ret = dio->size;
 345                                iocb->ki_pos += ret;
 346                        } else {
 347                                ret = blk_status_to_errno(dio->bio.bi_status);
 348                        }
 349
 350                        dio->iocb->ki_complete(iocb, ret, 0);
 351                        if (dio->multi_bio)
 352                                bio_put(&dio->bio);
 353                } else {
 354                        struct task_struct *waiter = dio->waiter;
 355
 356                        WRITE_ONCE(dio->waiter, NULL);
 357                        blk_wake_io_task(waiter);
 358                }
 359        }
 360
 361        if (should_dirty) {
 362                bio_check_pages_dirty(bio);
 363        } else {
 364                bio_release_pages(bio, false);
 365                bio_put(bio);
 366        }
 367}
 368
 369static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 370                unsigned int nr_pages)
 371{
 372        struct file *file = iocb->ki_filp;
 373        struct inode *inode = bdev_file_inode(file);
 374        struct block_device *bdev = I_BDEV(inode);
 375        struct blk_plug plug;
 376        struct blkdev_dio *dio;
 377        struct bio *bio;
 378        bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
 379        bool is_read = (iov_iter_rw(iter) == READ), is_sync;
 380        loff_t pos = iocb->ki_pos;
 381        blk_qc_t qc = BLK_QC_T_NONE;
 382        int ret = 0;
 383
 384        if ((pos | iov_iter_alignment(iter)) &
 385            (bdev_logical_block_size(bdev) - 1))
 386                return -EINVAL;
 387
 388        bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
 389
 390        dio = container_of(bio, struct blkdev_dio, bio);
 391        dio->is_sync = is_sync = is_sync_kiocb(iocb);
 392        if (dio->is_sync) {
 393                dio->waiter = current;
 394                bio_get(bio);
 395        } else {
 396                dio->iocb = iocb;
 397        }
 398
 399        dio->size = 0;
 400        dio->multi_bio = false;
 401        dio->should_dirty = is_read && iter_is_iovec(iter);
 402
 403        /*
 404         * Don't plug for HIPRI/polled IO, as those should go straight
 405         * to issue
 406         */
 407        if (!is_poll)
 408                blk_start_plug(&plug);
 409
 410        for (;;) {
 411                bio_set_dev(bio, bdev);
 412                bio->bi_iter.bi_sector = pos >> 9;
 413                bio->bi_write_hint = iocb->ki_hint;
 414                bio->bi_private = dio;
 415                bio->bi_end_io = blkdev_bio_end_io;
 416                bio->bi_ioprio = iocb->ki_ioprio;
 417
 418                ret = bio_iov_iter_get_pages(bio, iter);
 419                if (unlikely(ret)) {
 420                        bio->bi_status = BLK_STS_IOERR;
 421                        bio_endio(bio);
 422                        break;
 423                }
 424
 425                if (is_read) {
 426                        bio->bi_opf = REQ_OP_READ;
 427                        if (dio->should_dirty)
 428                                bio_set_pages_dirty(bio);
 429                } else {
 430                        bio->bi_opf = dio_bio_write_op(iocb);
 431                        task_io_account_write(bio->bi_iter.bi_size);
 432                }
 433                if (iocb->ki_flags & IOCB_NOWAIT)
 434                        bio->bi_opf |= REQ_NOWAIT;
 435
 436                dio->size += bio->bi_iter.bi_size;
 437                pos += bio->bi_iter.bi_size;
 438
 439                nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS);
 440                if (!nr_pages) {
 441                        bool polled = false;
 442
 443                        if (iocb->ki_flags & IOCB_HIPRI) {
 444                                bio_set_polled(bio, iocb);
 445                                polled = true;
 446                        }
 447
 448                        qc = submit_bio(bio);
 449
 450                        if (polled)
 451                                WRITE_ONCE(iocb->ki_cookie, qc);
 452                        break;
 453                }
 454
 455                if (!dio->multi_bio) {
 456                        /*
 457                         * AIO needs an extra reference to ensure the dio
 458                         * structure which is embedded into the first bio
 459                         * stays around.
 460                         */
 461                        if (!is_sync)
 462                                bio_get(bio);
 463                        dio->multi_bio = true;
 464                        atomic_set(&dio->ref, 2);
 465                } else {
 466                        atomic_inc(&dio->ref);
 467                }
 468
 469                submit_bio(bio);
 470                bio = bio_alloc(GFP_KERNEL, nr_pages);
 471        }
 472
 473        if (!is_poll)
 474                blk_finish_plug(&plug);
 475
 476        if (!is_sync)
 477                return -EIOCBQUEUED;
 478
 479        for (;;) {
 480                set_current_state(TASK_UNINTERRUPTIBLE);
 481                if (!READ_ONCE(dio->waiter))
 482                        break;
 483
 484                if (!(iocb->ki_flags & IOCB_HIPRI) ||
 485                    !blk_poll(bdev_get_queue(bdev), qc, true))
 486                        blk_io_schedule();
 487        }
 488        __set_current_state(TASK_RUNNING);
 489
 490        if (!ret)
 491                ret = blk_status_to_errno(dio->bio.bi_status);
 492        if (likely(!ret))
 493                ret = dio->size;
 494
 495        bio_put(&dio->bio);
 496        return ret;
 497}
 498
 499static ssize_t
 500blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 501{
 502        unsigned int nr_pages;
 503
 504        if (!iov_iter_count(iter))
 505                return 0;
 506
 507        nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
 508        if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_VECS)
 509                return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
 510
 511        return __blkdev_direct_IO(iocb, iter, bio_max_segs(nr_pages));
 512}
 513
 514static __init int blkdev_init(void)
 515{
 516        return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
 517}
 518module_init(blkdev_init);
 519
 520int __sync_blockdev(struct block_device *bdev, int wait)
 521{
 522        if (!bdev)
 523                return 0;
 524        if (!wait)
 525                return filemap_flush(bdev->bd_inode->i_mapping);
 526        return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 527}
 528
 529/*
 530 * Write out and wait upon all the dirty data associated with a block
 531 * device via its mapping.  Does not take the superblock lock.
 532 */
 533int sync_blockdev(struct block_device *bdev)
 534{
 535        return __sync_blockdev(bdev, 1);
 536}
 537EXPORT_SYMBOL(sync_blockdev);
 538
 539/*
 540 * Write out and wait upon all dirty data associated with this
 541 * device.   Filesystem data as well as the underlying block
 542 * device.  Takes the superblock lock.
 543 */
 544int fsync_bdev(struct block_device *bdev)
 545{
 546        struct super_block *sb = get_super(bdev);
 547        if (sb) {
 548                int res = sync_filesystem(sb);
 549                drop_super(sb);
 550                return res;
 551        }
 552        return sync_blockdev(bdev);
 553}
 554EXPORT_SYMBOL(fsync_bdev);
 555
 556/**
 557 * freeze_bdev  --  lock a filesystem and force it into a consistent state
 558 * @bdev:       blockdevice to lock
 559 *
 560 * If a superblock is found on this device, we take the s_umount semaphore
 561 * on it to make sure nobody unmounts until the snapshot creation is done.
 562 * The reference counter (bd_fsfreeze_count) guarantees that only the last
 563 * unfreeze process can unfreeze the frozen filesystem actually when multiple
 564 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 565 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 566 * actually.
 567 */
 568int freeze_bdev(struct block_device *bdev)
 569{
 570        struct super_block *sb;
 571        int error = 0;
 572
 573        mutex_lock(&bdev->bd_fsfreeze_mutex);
 574        if (++bdev->bd_fsfreeze_count > 1)
 575                goto done;
 576
 577        sb = get_active_super(bdev);
 578        if (!sb)
 579                goto sync;
 580        if (sb->s_op->freeze_super)
 581                error = sb->s_op->freeze_super(sb);
 582        else
 583                error = freeze_super(sb);
 584        deactivate_super(sb);
 585
 586        if (error) {
 587                bdev->bd_fsfreeze_count--;
 588                goto done;
 589        }
 590        bdev->bd_fsfreeze_sb = sb;
 591
 592sync:
 593        sync_blockdev(bdev);
 594done:
 595        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 596        return error;
 597}
 598EXPORT_SYMBOL(freeze_bdev);
 599
 600/**
 601 * thaw_bdev  -- unlock filesystem
 602 * @bdev:       blockdevice to unlock
 603 *
 604 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 605 */
 606int thaw_bdev(struct block_device *bdev)
 607{
 608        struct super_block *sb;
 609        int error = -EINVAL;
 610
 611        mutex_lock(&bdev->bd_fsfreeze_mutex);
 612        if (!bdev->bd_fsfreeze_count)
 613                goto out;
 614
 615        error = 0;
 616        if (--bdev->bd_fsfreeze_count > 0)
 617                goto out;
 618
 619        sb = bdev->bd_fsfreeze_sb;
 620        if (!sb)
 621                goto out;
 622
 623        if (sb->s_op->thaw_super)
 624                error = sb->s_op->thaw_super(sb);
 625        else
 626                error = thaw_super(sb);
 627        if (error)
 628                bdev->bd_fsfreeze_count++;
 629        else
 630                bdev->bd_fsfreeze_sb = NULL;
 631out:
 632        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 633        return error;
 634}
 635EXPORT_SYMBOL(thaw_bdev);
 636
 637static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 638{
 639        return block_write_full_page(page, blkdev_get_block, wbc);
 640}
 641
 642static int blkdev_readpage(struct file * file, struct page * page)
 643{
 644        return block_read_full_page(page, blkdev_get_block);
 645}
 646
 647static void blkdev_readahead(struct readahead_control *rac)
 648{
 649        mpage_readahead(rac, blkdev_get_block);
 650}
 651
 652static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 653                        loff_t pos, unsigned len, unsigned flags,
 654                        struct page **pagep, void **fsdata)
 655{
 656        return block_write_begin(mapping, pos, len, flags, pagep,
 657                                 blkdev_get_block);
 658}
 659
 660static int blkdev_write_end(struct file *file, struct address_space *mapping,
 661                        loff_t pos, unsigned len, unsigned copied,
 662                        struct page *page, void *fsdata)
 663{
 664        int ret;
 665        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 666
 667        unlock_page(page);
 668        put_page(page);
 669
 670        return ret;
 671}
 672
 673/*
 674 * private llseek:
 675 * for a block special file file_inode(file)->i_size is zero
 676 * so we compute the size by hand (just as in block_read/write above)
 677 */
 678static loff_t block_llseek(struct file *file, loff_t offset, int whence)
 679{
 680        struct inode *bd_inode = bdev_file_inode(file);
 681        loff_t retval;
 682
 683        inode_lock(bd_inode);
 684        retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
 685        inode_unlock(bd_inode);
 686        return retval;
 687}
 688        
 689int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 690{
 691        struct inode *bd_inode = bdev_file_inode(filp);
 692        struct block_device *bdev = I_BDEV(bd_inode);
 693        int error;
 694        
 695        error = file_write_and_wait_range(filp, start, end);
 696        if (error)
 697                return error;
 698
 699        /*
 700         * There is no need to serialise calls to blkdev_issue_flush with
 701         * i_mutex and doing so causes performance issues with concurrent
 702         * O_SYNC writers to a block device.
 703         */
 704        error = blkdev_issue_flush(bdev);
 705        if (error == -EOPNOTSUPP)
 706                error = 0;
 707
 708        return error;
 709}
 710EXPORT_SYMBOL(blkdev_fsync);
 711
 712/**
 713 * bdev_read_page() - Start reading a page from a block device
 714 * @bdev: The device to read the page from
 715 * @sector: The offset on the device to read the page to (need not be aligned)
 716 * @page: The page to read
 717 *
 718 * On entry, the page should be locked.  It will be unlocked when the page
 719 * has been read.  If the block driver implements rw_page synchronously,
 720 * that will be true on exit from this function, but it need not be.
 721 *
 722 * Errors returned by this function are usually "soft", eg out of memory, or
 723 * queue full; callers should try a different route to read this page rather
 724 * than propagate an error back up the stack.
 725 *
 726 * Return: negative errno if an error occurs, 0 if submission was successful.
 727 */
 728int bdev_read_page(struct block_device *bdev, sector_t sector,
 729                        struct page *page)
 730{
 731        const struct block_device_operations *ops = bdev->bd_disk->fops;
 732        int result = -EOPNOTSUPP;
 733
 734        if (!ops->rw_page || bdev_get_integrity(bdev))
 735                return result;
 736
 737        result = blk_queue_enter(bdev->bd_disk->queue, 0);
 738        if (result)
 739                return result;
 740        result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
 741                              REQ_OP_READ);
 742        blk_queue_exit(bdev->bd_disk->queue);
 743        return result;
 744}
 745
 746/**
 747 * bdev_write_page() - Start writing a page to a block device
 748 * @bdev: The device to write the page to
 749 * @sector: The offset on the device to write the page to (need not be aligned)
 750 * @page: The page to write
 751 * @wbc: The writeback_control for the write
 752 *
 753 * On entry, the page should be locked and not currently under writeback.
 754 * On exit, if the write started successfully, the page will be unlocked and
 755 * under writeback.  If the write failed already (eg the driver failed to
 756 * queue the page to the device), the page will still be locked.  If the
 757 * caller is a ->writepage implementation, it will need to unlock the page.
 758 *
 759 * Errors returned by this function are usually "soft", eg out of memory, or
 760 * queue full; callers should try a different route to write this page rather
 761 * than propagate an error back up the stack.
 762 *
 763 * Return: negative errno if an error occurs, 0 if submission was successful.
 764 */
 765int bdev_write_page(struct block_device *bdev, sector_t sector,
 766                        struct page *page, struct writeback_control *wbc)
 767{
 768        int result;
 769        const struct block_device_operations *ops = bdev->bd_disk->fops;
 770
 771        if (!ops->rw_page || bdev_get_integrity(bdev))
 772                return -EOPNOTSUPP;
 773        result = blk_queue_enter(bdev->bd_disk->queue, 0);
 774        if (result)
 775                return result;
 776
 777        set_page_writeback(page);
 778        result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
 779                              REQ_OP_WRITE);
 780        if (result) {
 781                end_page_writeback(page);
 782        } else {
 783                clean_page_buffers(page);
 784                unlock_page(page);
 785        }
 786        blk_queue_exit(bdev->bd_disk->queue);
 787        return result;
 788}
 789
 790/*
 791 * pseudo-fs
 792 */
 793
 794static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 795static struct kmem_cache * bdev_cachep __read_mostly;
 796
 797static struct inode *bdev_alloc_inode(struct super_block *sb)
 798{
 799        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 800
 801        if (!ei)
 802                return NULL;
 803        memset(&ei->bdev, 0, sizeof(ei->bdev));
 804        ei->bdev.bd_bdi = &noop_backing_dev_info;
 805        return &ei->vfs_inode;
 806}
 807
 808static void bdev_free_inode(struct inode *inode)
 809{
 810        struct block_device *bdev = I_BDEV(inode);
 811
 812        free_percpu(bdev->bd_stats);
 813        kfree(bdev->bd_meta_info);
 814
 815        if (!bdev_is_partition(bdev))
 816                kfree(bdev->bd_disk);
 817        kmem_cache_free(bdev_cachep, BDEV_I(inode));
 818}
 819
 820static void init_once(void *data)
 821{
 822        struct bdev_inode *ei = data;
 823
 824        inode_init_once(&ei->vfs_inode);
 825}
 826
 827static void bdev_evict_inode(struct inode *inode)
 828{
 829        struct block_device *bdev = &BDEV_I(inode)->bdev;
 830        truncate_inode_pages_final(&inode->i_data);
 831        invalidate_inode_buffers(inode); /* is it needed here? */
 832        clear_inode(inode);
 833        /* Detach inode from wb early as bdi_put() may free bdi->wb */
 834        inode_detach_wb(inode);
 835        if (bdev->bd_bdi != &noop_backing_dev_info) {
 836                bdi_put(bdev->bd_bdi);
 837                bdev->bd_bdi = &noop_backing_dev_info;
 838        }
 839}
 840
 841static const struct super_operations bdev_sops = {
 842        .statfs = simple_statfs,
 843        .alloc_inode = bdev_alloc_inode,
 844        .free_inode = bdev_free_inode,
 845        .drop_inode = generic_delete_inode,
 846        .evict_inode = bdev_evict_inode,
 847};
 848
 849static int bd_init_fs_context(struct fs_context *fc)
 850{
 851        struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
 852        if (!ctx)
 853                return -ENOMEM;
 854        fc->s_iflags |= SB_I_CGROUPWB;
 855        ctx->ops = &bdev_sops;
 856        return 0;
 857}
 858
 859static struct file_system_type bd_type = {
 860        .name           = "bdev",
 861        .init_fs_context = bd_init_fs_context,
 862        .kill_sb        = kill_anon_super,
 863};
 864
 865struct super_block *blockdev_superblock __read_mostly;
 866EXPORT_SYMBOL_GPL(blockdev_superblock);
 867
 868void __init bdev_cache_init(void)
 869{
 870        int err;
 871        static struct vfsmount *bd_mnt;
 872
 873        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 874                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 875                                SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
 876                        init_once);
 877        err = register_filesystem(&bd_type);
 878        if (err)
 879                panic("Cannot register bdev pseudo-fs");
 880        bd_mnt = kern_mount(&bd_type);
 881        if (IS_ERR(bd_mnt))
 882                panic("Cannot create bdev pseudo-fs");
 883        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 884}
 885
 886struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
 887{
 888        struct block_device *bdev;
 889        struct inode *inode;
 890
 891        inode = new_inode(blockdev_superblock);
 892        if (!inode)
 893                return NULL;
 894        inode->i_mode = S_IFBLK;
 895        inode->i_rdev = 0;
 896        inode->i_data.a_ops = &def_blk_aops;
 897        mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 898
 899        bdev = I_BDEV(inode);
 900        mutex_init(&bdev->bd_fsfreeze_mutex);
 901        spin_lock_init(&bdev->bd_size_lock);
 902        bdev->bd_disk = disk;
 903        bdev->bd_partno = partno;
 904        bdev->bd_inode = inode;
 905#ifdef CONFIG_SYSFS
 906        INIT_LIST_HEAD(&bdev->bd_holder_disks);
 907#endif
 908        bdev->bd_stats = alloc_percpu(struct disk_stats);
 909        if (!bdev->bd_stats) {
 910                iput(inode);
 911                return NULL;
 912        }
 913        return bdev;
 914}
 915
 916void bdev_add(struct block_device *bdev, dev_t dev)
 917{
 918        bdev->bd_dev = dev;
 919        bdev->bd_inode->i_rdev = dev;
 920        bdev->bd_inode->i_ino = dev;
 921        insert_inode_hash(bdev->bd_inode);
 922}
 923
 924static struct block_device *bdget(dev_t dev)
 925{
 926        struct inode *inode;
 927
 928        inode = ilookup(blockdev_superblock, dev);
 929        if (!inode)
 930                return NULL;
 931        return &BDEV_I(inode)->bdev;
 932}
 933
 934/**
 935 * bdgrab -- Grab a reference to an already referenced block device
 936 * @bdev:       Block device to grab a reference to.
 937 *
 938 * Returns the block_device with an additional reference when successful,
 939 * or NULL if the inode is already beeing freed.
 940 */
 941struct block_device *bdgrab(struct block_device *bdev)
 942{
 943        if (!igrab(bdev->bd_inode))
 944                return NULL;
 945        return bdev;
 946}
 947EXPORT_SYMBOL(bdgrab);
 948
 949long nr_blockdev_pages(void)
 950{
 951        struct inode *inode;
 952        long ret = 0;
 953
 954        spin_lock(&blockdev_superblock->s_inode_list_lock);
 955        list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
 956                ret += inode->i_mapping->nrpages;
 957        spin_unlock(&blockdev_superblock->s_inode_list_lock);
 958
 959        return ret;
 960}
 961
 962void bdput(struct block_device *bdev)
 963{
 964        iput(bdev->bd_inode);
 965}
 966EXPORT_SYMBOL(bdput);
 967 
 968/**
 969 * bd_may_claim - test whether a block device can be claimed
 970 * @bdev: block device of interest
 971 * @whole: whole block device containing @bdev, may equal @bdev
 972 * @holder: holder trying to claim @bdev
 973 *
 974 * Test whether @bdev can be claimed by @holder.
 975 *
 976 * CONTEXT:
 977 * spin_lock(&bdev_lock).
 978 *
 979 * RETURNS:
 980 * %true if @bdev can be claimed, %false otherwise.
 981 */
 982static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
 983                         void *holder)
 984{
 985        if (bdev->bd_holder == holder)
 986                return true;     /* already a holder */
 987        else if (bdev->bd_holder != NULL)
 988                return false;    /* held by someone else */
 989        else if (whole == bdev)
 990                return true;     /* is a whole device which isn't held */
 991
 992        else if (whole->bd_holder == bd_may_claim)
 993                return true;     /* is a partition of a device that is being partitioned */
 994        else if (whole->bd_holder != NULL)
 995                return false;    /* is a partition of a held device */
 996        else
 997                return true;     /* is a partition of an un-held device */
 998}
 999
1000/**
1001 * bd_prepare_to_claim - claim a block device
1002 * @bdev: block device of interest
1003 * @holder: holder trying to claim @bdev
1004 *
1005 * Claim @bdev.  This function fails if @bdev is already claimed by another
1006 * holder and waits if another claiming is in progress. return, the caller
1007 * has ownership of bd_claiming and bd_holder[s].
1008 *
1009 * RETURNS:
1010 * 0 if @bdev can be claimed, -EBUSY otherwise.
1011 */
1012int bd_prepare_to_claim(struct block_device *bdev, void *holder)
1013{
1014        struct block_device *whole = bdev_whole(bdev);
1015
1016        if (WARN_ON_ONCE(!holder))
1017                return -EINVAL;
1018retry:
1019        spin_lock(&bdev_lock);
1020        /* if someone else claimed, fail */
1021        if (!bd_may_claim(bdev, whole, holder)) {
1022                spin_unlock(&bdev_lock);
1023                return -EBUSY;
1024        }
1025
1026        /* if claiming is already in progress, wait for it to finish */
1027        if (whole->bd_claiming) {
1028                wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1029                DEFINE_WAIT(wait);
1030
1031                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1032                spin_unlock(&bdev_lock);
1033                schedule();
1034                finish_wait(wq, &wait);
1035                goto retry;
1036        }
1037
1038        /* yay, all mine */
1039        whole->bd_claiming = holder;
1040        spin_unlock(&bdev_lock);
1041        return 0;
1042}
1043EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1044
1045static void bd_clear_claiming(struct block_device *whole, void *holder)
1046{
1047        lockdep_assert_held(&bdev_lock);
1048        /* tell others that we're done */
1049        BUG_ON(whole->bd_claiming != holder);
1050        whole->bd_claiming = NULL;
1051        wake_up_bit(&whole->bd_claiming, 0);
1052}
1053
1054/**
1055 * bd_finish_claiming - finish claiming of a block device
1056 * @bdev: block device of interest
1057 * @holder: holder that has claimed @bdev
1058 *
1059 * Finish exclusive open of a block device. Mark the device as exlusively
1060 * open by the holder and wake up all waiters for exclusive open to finish.
1061 */
1062static void bd_finish_claiming(struct block_device *bdev, void *holder)
1063{
1064        struct block_device *whole = bdev_whole(bdev);
1065
1066        spin_lock(&bdev_lock);
1067        BUG_ON(!bd_may_claim(bdev, whole, holder));
1068        /*
1069         * Note that for a whole device bd_holders will be incremented twice,
1070         * and bd_holder will be set to bd_may_claim before being set to holder
1071         */
1072        whole->bd_holders++;
1073        whole->bd_holder = bd_may_claim;
1074        bdev->bd_holders++;
1075        bdev->bd_holder = holder;
1076        bd_clear_claiming(whole, holder);
1077        spin_unlock(&bdev_lock);
1078}
1079
1080/**
1081 * bd_abort_claiming - abort claiming of a block device
1082 * @bdev: block device of interest
1083 * @holder: holder that has claimed @bdev
1084 *
1085 * Abort claiming of a block device when the exclusive open failed. This can be
1086 * also used when exclusive open is not actually desired and we just needed
1087 * to block other exclusive openers for a while.
1088 */
1089void bd_abort_claiming(struct block_device *bdev, void *holder)
1090{
1091        spin_lock(&bdev_lock);
1092        bd_clear_claiming(bdev_whole(bdev), holder);
1093        spin_unlock(&bdev_lock);
1094}
1095EXPORT_SYMBOL(bd_abort_claiming);
1096
1097#ifdef CONFIG_SYSFS
1098struct bd_holder_disk {
1099        struct list_head        list;
1100        struct gendisk          *disk;
1101        int                     refcnt;
1102};
1103
1104static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1105                                                  struct gendisk *disk)
1106{
1107        struct bd_holder_disk *holder;
1108
1109        list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1110                if (holder->disk == disk)
1111                        return holder;
1112        return NULL;
1113}
1114
1115static int add_symlink(struct kobject *from, struct kobject *to)
1116{
1117        return sysfs_create_link(from, to, kobject_name(to));
1118}
1119
1120static void del_symlink(struct kobject *from, struct kobject *to)
1121{
1122        sysfs_remove_link(from, kobject_name(to));
1123}
1124
1125/**
1126 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1127 * @bdev: the claimed slave bdev
1128 * @disk: the holding disk
1129 *
1130 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1131 *
1132 * This functions creates the following sysfs symlinks.
1133 *
1134 * - from "slaves" directory of the holder @disk to the claimed @bdev
1135 * - from "holders" directory of the @bdev to the holder @disk
1136 *
1137 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1138 * passed to bd_link_disk_holder(), then:
1139 *
1140 *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
1141 *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1142 *
1143 * The caller must have claimed @bdev before calling this function and
1144 * ensure that both @bdev and @disk are valid during the creation and
1145 * lifetime of these symlinks.
1146 *
1147 * CONTEXT:
1148 * Might sleep.
1149 *
1150 * RETURNS:
1151 * 0 on success, -errno on failure.
1152 */
1153int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1154{
1155        struct bd_holder_disk *holder;
1156        int ret = 0;
1157
1158        mutex_lock(&bdev->bd_disk->open_mutex);
1159
1160        WARN_ON_ONCE(!bdev->bd_holder);
1161
1162        /* FIXME: remove the following once add_disk() handles errors */
1163        if (WARN_ON(!disk->slave_dir || !bdev->bd_holder_dir))
1164                goto out_unlock;
1165
1166        holder = bd_find_holder_disk(bdev, disk);
1167        if (holder) {
1168                holder->refcnt++;
1169                goto out_unlock;
1170        }
1171
1172        holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1173        if (!holder) {
1174                ret = -ENOMEM;
1175                goto out_unlock;
1176        }
1177
1178        INIT_LIST_HEAD(&holder->list);
1179        holder->disk = disk;
1180        holder->refcnt = 1;
1181
1182        ret = add_symlink(disk->slave_dir, bdev_kobj(bdev));
1183        if (ret)
1184                goto out_free;
1185
1186        ret = add_symlink(bdev->bd_holder_dir, &disk_to_dev(disk)->kobj);
1187        if (ret)
1188                goto out_del;
1189        /*
1190         * bdev could be deleted beneath us which would implicitly destroy
1191         * the holder directory.  Hold on to it.
1192         */
1193        kobject_get(bdev->bd_holder_dir);
1194
1195        list_add(&holder->list, &bdev->bd_holder_disks);
1196        goto out_unlock;
1197
1198out_del:
1199        del_symlink(disk->slave_dir, bdev_kobj(bdev));
1200out_free:
1201        kfree(holder);
1202out_unlock:
1203        mutex_unlock(&bdev->bd_disk->open_mutex);
1204        return ret;
1205}
1206EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1207
1208/**
1209 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1210 * @bdev: the calimed slave bdev
1211 * @disk: the holding disk
1212 *
1213 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1214 *
1215 * CONTEXT:
1216 * Might sleep.
1217 */
1218void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1219{
1220        struct bd_holder_disk *holder;
1221
1222        mutex_lock(&bdev->bd_disk->open_mutex);
1223
1224        holder = bd_find_holder_disk(bdev, disk);
1225
1226        if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1227                del_symlink(disk->slave_dir, bdev_kobj(bdev));
1228                del_symlink(bdev->bd_holder_dir, &disk_to_dev(disk)->kobj);
1229                kobject_put(bdev->bd_holder_dir);
1230                list_del_init(&holder->list);
1231                kfree(holder);
1232        }
1233
1234        mutex_unlock(&bdev->bd_disk->open_mutex);
1235}
1236EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1237#endif
1238
1239static void blkdev_flush_mapping(struct block_device *bdev)
1240{
1241        WARN_ON_ONCE(bdev->bd_holders);
1242        sync_blockdev(bdev);
1243        kill_bdev(bdev);
1244        bdev_write_inode(bdev);
1245}
1246
1247static int blkdev_get_whole(struct block_device *bdev, fmode_t mode)
1248{
1249        struct gendisk *disk = bdev->bd_disk;
1250        int ret = 0;
1251
1252        if (disk->fops->open) {
1253                ret = disk->fops->open(bdev, mode);
1254                if (ret) {
1255                        /* avoid ghost partitions on a removed medium */
1256                        if (ret == -ENOMEDIUM &&
1257                             test_bit(GD_NEED_PART_SCAN, &disk->state))
1258                                bdev_disk_changed(disk, true);
1259                        return ret;
1260                }
1261        }
1262
1263        if (!bdev->bd_openers) {
1264                set_init_blocksize(bdev);
1265                if (bdev->bd_bdi == &noop_backing_dev_info)
1266                        bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1267        }
1268        if (test_bit(GD_NEED_PART_SCAN, &disk->state))
1269                bdev_disk_changed(disk, false);
1270        bdev->bd_openers++;
1271        return 0;;
1272}
1273
1274static void blkdev_put_whole(struct block_device *bdev, fmode_t mode)
1275{
1276        if (!--bdev->bd_openers)
1277                blkdev_flush_mapping(bdev);
1278        if (bdev->bd_disk->fops->release)
1279                bdev->bd_disk->fops->release(bdev->bd_disk, mode);
1280}
1281
1282static int blkdev_get_part(struct block_device *part, fmode_t mode)
1283{
1284        struct gendisk *disk = part->bd_disk;
1285        struct block_device *whole;
1286        int ret;
1287
1288        if (part->bd_openers)
1289                goto done;
1290
1291        whole = bdgrab(disk->part0);
1292        ret = blkdev_get_whole(whole, mode);
1293        if (ret)
1294                goto out_put_whole;
1295
1296        ret = -ENXIO;
1297        if (!bdev_nr_sectors(part))
1298                goto out_blkdev_put;
1299
1300        disk->open_partitions++;
1301        set_init_blocksize(part);
1302        if (part->bd_bdi == &noop_backing_dev_info)
1303                part->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1304done:
1305        part->bd_openers++;
1306        return 0;
1307
1308out_blkdev_put:
1309        blkdev_put_whole(whole, mode);
1310out_put_whole:
1311        bdput(whole);
1312        return ret;
1313}
1314
1315static void blkdev_put_part(struct block_device *part, fmode_t mode)
1316{
1317        struct block_device *whole = bdev_whole(part);
1318
1319        if (--part->bd_openers)
1320                return;
1321        blkdev_flush_mapping(part);
1322        whole->bd_disk->open_partitions--;
1323        blkdev_put_whole(whole, mode);
1324        bdput(whole);
1325}
1326
1327struct block_device *blkdev_get_no_open(dev_t dev)
1328{
1329        struct block_device *bdev;
1330        struct gendisk *disk;
1331
1332        bdev = bdget(dev);
1333        if (!bdev) {
1334                blk_request_module(dev);
1335                bdev = bdget(dev);
1336                if (!bdev)
1337                        return NULL;
1338        }
1339
1340        disk = bdev->bd_disk;
1341        if (!kobject_get_unless_zero(&disk_to_dev(disk)->kobj))
1342                goto bdput;
1343        if ((disk->flags & (GENHD_FL_UP | GENHD_FL_HIDDEN)) != GENHD_FL_UP)
1344                goto put_disk;
1345        if (!try_module_get(bdev->bd_disk->fops->owner))
1346                goto put_disk;
1347        return bdev;
1348put_disk:
1349        put_disk(disk);
1350bdput:
1351        bdput(bdev);
1352        return NULL;
1353}
1354
1355void blkdev_put_no_open(struct block_device *bdev)
1356{
1357        module_put(bdev->bd_disk->fops->owner);
1358        put_disk(bdev->bd_disk);
1359        bdput(bdev);
1360}
1361
1362/**
1363 * blkdev_get_by_dev - open a block device by device number
1364 * @dev: device number of block device to open
1365 * @mode: FMODE_* mask
1366 * @holder: exclusive holder identifier
1367 *
1368 * Open the block device described by device number @dev. If @mode includes
1369 * %FMODE_EXCL, the block device is opened with exclusive access.  Specifying
1370 * %FMODE_EXCL with a %NULL @holder is invalid.  Exclusive opens may nest for
1371 * the same @holder.
1372 *
1373 * Use this interface ONLY if you really do not have anything better - i.e. when
1374 * you are behind a truly sucky interface and all you are given is a device
1375 * number.  Everything else should use blkdev_get_by_path().
1376 *
1377 * CONTEXT:
1378 * Might sleep.
1379 *
1380 * RETURNS:
1381 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
1382 */
1383struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1384{
1385        bool unblock_events = true;
1386        struct block_device *bdev;
1387        struct gendisk *disk;
1388        int ret;
1389
1390        ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
1391                        MAJOR(dev), MINOR(dev),
1392                        ((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) |
1393                        ((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0));
1394        if (ret)
1395                return ERR_PTR(ret);
1396
1397        bdev = blkdev_get_no_open(dev);
1398        if (!bdev)
1399                return ERR_PTR(-ENXIO);
1400        disk = bdev->bd_disk;
1401
1402        if (mode & FMODE_EXCL) {
1403                ret = bd_prepare_to_claim(bdev, holder);
1404                if (ret)
1405                        goto put_blkdev;
1406        }
1407
1408        disk_block_events(disk);
1409
1410        mutex_lock(&disk->open_mutex);
1411        ret = -ENXIO;
1412        if (!(disk->flags & GENHD_FL_UP))
1413                goto abort_claiming;
1414        if (bdev_is_partition(bdev))
1415                ret = blkdev_get_part(bdev, mode);
1416        else
1417                ret = blkdev_get_whole(bdev, mode);
1418        if (ret)
1419                goto abort_claiming;
1420        if (mode & FMODE_EXCL) {
1421                bd_finish_claiming(bdev, holder);
1422
1423                /*
1424                 * Block event polling for write claims if requested.  Any write
1425                 * holder makes the write_holder state stick until all are
1426                 * released.  This is good enough and tracking individual
1427                 * writeable reference is too fragile given the way @mode is
1428                 * used in blkdev_get/put().
1429                 */
1430                if ((mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1431                    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1432                        bdev->bd_write_holder = true;
1433                        unblock_events = false;
1434                }
1435        }
1436        mutex_unlock(&disk->open_mutex);
1437
1438        if (unblock_events)
1439                disk_unblock_events(disk);
1440        return bdev;
1441
1442abort_claiming:
1443        if (mode & FMODE_EXCL)
1444                bd_abort_claiming(bdev, holder);
1445        mutex_unlock(&disk->open_mutex);
1446        disk_unblock_events(disk);
1447put_blkdev:
1448        blkdev_put_no_open(bdev);
1449        return ERR_PTR(ret);
1450}
1451EXPORT_SYMBOL(blkdev_get_by_dev);
1452
1453/**
1454 * blkdev_get_by_path - open a block device by name
1455 * @path: path to the block device to open
1456 * @mode: FMODE_* mask
1457 * @holder: exclusive holder identifier
1458 *
1459 * Open the block device described by the device file at @path.  If @mode
1460 * includes %FMODE_EXCL, the block device is opened with exclusive access.
1461 * Specifying %FMODE_EXCL with a %NULL @holder is invalid.  Exclusive opens may
1462 * nest for the same @holder.
1463 *
1464 * CONTEXT:
1465 * Might sleep.
1466 *
1467 * RETURNS:
1468 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
1469 */
1470struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1471                                        void *holder)
1472{
1473        struct block_device *bdev;
1474        dev_t dev;
1475        int error;
1476
1477        error = lookup_bdev(path, &dev);
1478        if (error)
1479                return ERR_PTR(error);
1480
1481        bdev = blkdev_get_by_dev(dev, mode, holder);
1482        if (!IS_ERR(bdev) && (mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1483                blkdev_put(bdev, mode);
1484                return ERR_PTR(-EACCES);
1485        }
1486
1487        return bdev;
1488}
1489EXPORT_SYMBOL(blkdev_get_by_path);
1490
1491static int blkdev_open(struct inode * inode, struct file * filp)
1492{
1493        struct block_device *bdev;
1494
1495        /*
1496         * Preserve backwards compatibility and allow large file access
1497         * even if userspace doesn't ask for it explicitly. Some mkfs
1498         * binary needs it. We might want to drop this workaround
1499         * during an unstable branch.
1500         */
1501        filp->f_flags |= O_LARGEFILE;
1502
1503        filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1504
1505        if (filp->f_flags & O_NDELAY)
1506                filp->f_mode |= FMODE_NDELAY;
1507        if (filp->f_flags & O_EXCL)
1508                filp->f_mode |= FMODE_EXCL;
1509        if ((filp->f_flags & O_ACCMODE) == 3)
1510                filp->f_mode |= FMODE_WRITE_IOCTL;
1511
1512        bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
1513        if (IS_ERR(bdev))
1514                return PTR_ERR(bdev);
1515        filp->f_mapping = bdev->bd_inode->i_mapping;
1516        filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1517        return 0;
1518}
1519
1520void blkdev_put(struct block_device *bdev, fmode_t mode)
1521{
1522        struct gendisk *disk = bdev->bd_disk;
1523
1524        /*
1525         * Sync early if it looks like we're the last one.  If someone else
1526         * opens the block device between now and the decrement of bd_openers
1527         * then we did a sync that we didn't need to, but that's not the end
1528         * of the world and we want to avoid long (could be several minute)
1529         * syncs while holding the mutex.
1530         */
1531        if (bdev->bd_openers == 1)
1532                sync_blockdev(bdev);
1533
1534        mutex_lock(&disk->open_mutex);
1535        if (mode & FMODE_EXCL) {
1536                struct block_device *whole = bdev_whole(bdev);
1537                bool bdev_free;
1538
1539                /*
1540                 * Release a claim on the device.  The holder fields
1541                 * are protected with bdev_lock.  open_mutex is to
1542                 * synchronize disk_holder unlinking.
1543                 */
1544                spin_lock(&bdev_lock);
1545
1546                WARN_ON_ONCE(--bdev->bd_holders < 0);
1547                WARN_ON_ONCE(--whole->bd_holders < 0);
1548
1549                if ((bdev_free = !bdev->bd_holders))
1550                        bdev->bd_holder = NULL;
1551                if (!whole->bd_holders)
1552                        whole->bd_holder = NULL;
1553
1554                spin_unlock(&bdev_lock);
1555
1556                /*
1557                 * If this was the last claim, remove holder link and
1558                 * unblock evpoll if it was a write holder.
1559                 */
1560                if (bdev_free && bdev->bd_write_holder) {
1561                        disk_unblock_events(disk);
1562                        bdev->bd_write_holder = false;
1563                }
1564        }
1565
1566        /*
1567         * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1568         * event.  This is to ensure detection of media removal commanded
1569         * from userland - e.g. eject(1).
1570         */
1571        disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
1572
1573        if (bdev_is_partition(bdev))
1574                blkdev_put_part(bdev, mode);
1575        else
1576                blkdev_put_whole(bdev, mode);
1577        mutex_unlock(&disk->open_mutex);
1578
1579        blkdev_put_no_open(bdev);
1580}
1581EXPORT_SYMBOL(blkdev_put);
1582
1583static int blkdev_close(struct inode * inode, struct file * filp)
1584{
1585        struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1586        blkdev_put(bdev, filp->f_mode);
1587        return 0;
1588}
1589
1590static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1591{
1592        struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1593        fmode_t mode = file->f_mode;
1594
1595        /*
1596         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1597         * to updated it before every ioctl.
1598         */
1599        if (file->f_flags & O_NDELAY)
1600                mode |= FMODE_NDELAY;
1601        else
1602                mode &= ~FMODE_NDELAY;
1603
1604        return blkdev_ioctl(bdev, mode, cmd, arg);
1605}
1606
1607/*
1608 * Write data to the block device.  Only intended for the block device itself
1609 * and the raw driver which basically is a fake block device.
1610 *
1611 * Does not take i_mutex for the write and thus is not for general purpose
1612 * use.
1613 */
1614static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1615{
1616        struct file *file = iocb->ki_filp;
1617        struct inode *bd_inode = bdev_file_inode(file);
1618        loff_t size = i_size_read(bd_inode);
1619        struct blk_plug plug;
1620        size_t shorted = 0;
1621        ssize_t ret;
1622
1623        if (bdev_read_only(I_BDEV(bd_inode)))
1624                return -EPERM;
1625
1626        if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
1627                return -ETXTBSY;
1628
1629        if (!iov_iter_count(from))
1630                return 0;
1631
1632        if (iocb->ki_pos >= size)
1633                return -ENOSPC;
1634
1635        if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1636                return -EOPNOTSUPP;
1637
1638        size -= iocb->ki_pos;
1639        if (iov_iter_count(from) > size) {
1640                shorted = iov_iter_count(from) - size;
1641                iov_iter_truncate(from, size);
1642        }
1643
1644        blk_start_plug(&plug);
1645        ret = __generic_file_write_iter(iocb, from);
1646        if (ret > 0)
1647                ret = generic_write_sync(iocb, ret);
1648        iov_iter_reexpand(from, iov_iter_count(from) + shorted);
1649        blk_finish_plug(&plug);
1650        return ret;
1651}
1652
1653static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1654{
1655        struct file *file = iocb->ki_filp;
1656        struct inode *bd_inode = bdev_file_inode(file);
1657        loff_t size = i_size_read(bd_inode);
1658        loff_t pos = iocb->ki_pos;
1659        size_t shorted = 0;
1660        ssize_t ret;
1661
1662        if (pos >= size)
1663                return 0;
1664
1665        size -= pos;
1666        if (iov_iter_count(to) > size) {
1667                shorted = iov_iter_count(to) - size;
1668                iov_iter_truncate(to, size);
1669        }
1670
1671        ret = generic_file_read_iter(iocb, to);
1672        iov_iter_reexpand(to, iov_iter_count(to) + shorted);
1673        return ret;
1674}
1675
1676static int blkdev_writepages(struct address_space *mapping,
1677                             struct writeback_control *wbc)
1678{
1679        return generic_writepages(mapping, wbc);
1680}
1681
1682static const struct address_space_operations def_blk_aops = {
1683        .set_page_dirty = __set_page_dirty_buffers,
1684        .readpage       = blkdev_readpage,
1685        .readahead      = blkdev_readahead,
1686        .writepage      = blkdev_writepage,
1687        .write_begin    = blkdev_write_begin,
1688        .write_end      = blkdev_write_end,
1689        .writepages     = blkdev_writepages,
1690        .direct_IO      = blkdev_direct_IO,
1691        .migratepage    = buffer_migrate_page_norefs,
1692        .is_dirty_writeback = buffer_check_dirty_writeback,
1693};
1694
1695#define BLKDEV_FALLOC_FL_SUPPORTED                                      \
1696                (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |           \
1697                 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1698
1699static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1700                             loff_t len)
1701{
1702        struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1703        loff_t end = start + len - 1;
1704        loff_t isize;
1705        int error;
1706
1707        /* Fail if we don't recognize the flags. */
1708        if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1709                return -EOPNOTSUPP;
1710
1711        /* Don't go off the end of the device. */
1712        isize = i_size_read(bdev->bd_inode);
1713        if (start >= isize)
1714                return -EINVAL;
1715        if (end >= isize) {
1716                if (mode & FALLOC_FL_KEEP_SIZE) {
1717                        len = isize - start;
1718                        end = start + len - 1;
1719                } else
1720                        return -EINVAL;
1721        }
1722
1723        /*
1724         * Don't allow IO that isn't aligned to logical block size.
1725         */
1726        if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1727                return -EINVAL;
1728
1729        /* Invalidate the page cache, including dirty pages. */
1730        error = truncate_bdev_range(bdev, file->f_mode, start, end);
1731        if (error)
1732                return error;
1733
1734        switch (mode) {
1735        case FALLOC_FL_ZERO_RANGE:
1736        case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1737                error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1738                                            GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
1739                break;
1740        case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1741                error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1742                                             GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
1743                break;
1744        case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1745                error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1746                                             GFP_KERNEL, 0);
1747                break;
1748        default:
1749                return -EOPNOTSUPP;
1750        }
1751        if (error)
1752                return error;
1753
1754        /*
1755         * Invalidate the page cache again; if someone wandered in and dirtied
1756         * a page, we just discard it - userspace has no way of knowing whether
1757         * the write happened before or after discard completing...
1758         */
1759        return truncate_bdev_range(bdev, file->f_mode, start, end);
1760}
1761
1762const struct file_operations def_blk_fops = {
1763        .open           = blkdev_open,
1764        .release        = blkdev_close,
1765        .llseek         = block_llseek,
1766        .read_iter      = blkdev_read_iter,
1767        .write_iter     = blkdev_write_iter,
1768        .iopoll         = blkdev_iopoll,
1769        .mmap           = generic_file_mmap,
1770        .fsync          = blkdev_fsync,
1771        .unlocked_ioctl = block_ioctl,
1772#ifdef CONFIG_COMPAT
1773        .compat_ioctl   = compat_blkdev_ioctl,
1774#endif
1775        .splice_read    = generic_file_splice_read,
1776        .splice_write   = iter_file_splice_write,
1777        .fallocate      = blkdev_fallocate,
1778};
1779
1780/**
1781 * lookup_bdev  - lookup a struct block_device by name
1782 * @pathname:   special file representing the block device
1783 * @dev:        return value of the block device's dev_t
1784 *
1785 * Get a reference to the blockdevice at @pathname in the current
1786 * namespace if possible and return it.  Return ERR_PTR(error)
1787 * otherwise.
1788 */
1789int lookup_bdev(const char *pathname, dev_t *dev)
1790{
1791        struct inode *inode;
1792        struct path path;
1793        int error;
1794
1795        if (!pathname || !*pathname)
1796                return -EINVAL;
1797
1798        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1799        if (error)
1800                return error;
1801
1802        inode = d_backing_inode(path.dentry);
1803        error = -ENOTBLK;
1804        if (!S_ISBLK(inode->i_mode))
1805                goto out_path_put;
1806        error = -EACCES;
1807        if (!may_open_dev(&path))
1808                goto out_path_put;
1809
1810        *dev = inode->i_rdev;
1811        error = 0;
1812out_path_put:
1813        path_put(&path);
1814        return error;
1815}
1816EXPORT_SYMBOL(lookup_bdev);
1817
1818int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1819{
1820        struct super_block *sb = get_super(bdev);
1821        int res = 0;
1822
1823        if (sb) {
1824                /*
1825                 * no need to lock the super, get_super holds the
1826                 * read mutex so the filesystem cannot go away
1827                 * under us (->put_super runs with the write lock
1828                 * hold).
1829                 */
1830                shrink_dcache_sb(sb);
1831                res = invalidate_inodes(sb, kill_dirty);
1832                drop_super(sb);
1833        }
1834        invalidate_bdev(bdev);
1835        return res;
1836}
1837EXPORT_SYMBOL(__invalidate_device);
1838
1839void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1840{
1841        struct inode *inode, *old_inode = NULL;
1842
1843        spin_lock(&blockdev_superblock->s_inode_list_lock);
1844        list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1845                struct address_space *mapping = inode->i_mapping;
1846                struct block_device *bdev;
1847
1848                spin_lock(&inode->i_lock);
1849                if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1850                    mapping->nrpages == 0) {
1851                        spin_unlock(&inode->i_lock);
1852                        continue;
1853                }
1854                __iget(inode);
1855                spin_unlock(&inode->i_lock);
1856                spin_unlock(&blockdev_superblock->s_inode_list_lock);
1857                /*
1858                 * We hold a reference to 'inode' so it couldn't have been
1859                 * removed from s_inodes list while we dropped the
1860                 * s_inode_list_lock  We cannot iput the inode now as we can
1861                 * be holding the last reference and we cannot iput it under
1862                 * s_inode_list_lock. So we keep the reference and iput it
1863                 * later.
1864                 */
1865                iput(old_inode);
1866                old_inode = inode;
1867                bdev = I_BDEV(inode);
1868
1869                mutex_lock(&bdev->bd_disk->open_mutex);
1870                if (bdev->bd_openers)
1871                        func(bdev, arg);
1872                mutex_unlock(&bdev->bd_disk->open_mutex);
1873
1874                spin_lock(&blockdev_superblock->s_inode_list_lock);
1875        }
1876        spin_unlock(&blockdev_superblock->s_inode_list_lock);
1877        iput(old_inode);
1878}
1879
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.