linux/fs/block_dev.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/block_dev.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 *  Copyright (C) 2001  Andrea Arcangeli <andrea@suse.de> SuSE
   7 */
   8
   9#include <linux/init.h>
  10#include <linux/mm.h>
  11#include <linux/fcntl.h>
  12#include <linux/slab.h>
  13#include <linux/kmod.h>
  14#include <linux/major.h>
  15#include <linux/device_cgroup.h>
  16#include <linux/highmem.h>
  17#include <linux/blkdev.h>
  18#include <linux/backing-dev.h>
  19#include <linux/module.h>
  20#include <linux/blkpg.h>
  21#include <linux/magic.h>
  22#include <linux/dax.h>
  23#include <linux/buffer_head.h>
  24#include <linux/swap.h>
  25#include <linux/pagevec.h>
  26#include <linux/writeback.h>
  27#include <linux/mpage.h>
  28#include <linux/mount.h>
  29#include <linux/pseudo_fs.h>
  30#include <linux/uio.h>
  31#include <linux/namei.h>
  32#include <linux/log2.h>
  33#include <linux/cleancache.h>
  34#include <linux/task_io_accounting_ops.h>
  35#include <linux/falloc.h>
  36#include <linux/uaccess.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.. */
  78void kill_bdev(struct block_device *bdev)
  79{
  80        struct address_space *mapping = bdev->bd_inode->i_mapping;
  81
  82        if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
  83                return;
  84
  85        invalidate_bh_lrus();
  86        truncate_inode_pages(mapping, 0);
  87}       
  88EXPORT_SYMBOL(kill_bdev);
  89
  90/* Invalidate clean unused buffers and pagecache. */
  91void invalidate_bdev(struct block_device *bdev)
  92{
  93        struct address_space *mapping = bdev->bd_inode->i_mapping;
  94
  95        if (mapping->nrpages) {
  96                invalidate_bh_lrus();
  97                lru_add_drain_all();    /* make sure all lru add caches are flushed */
  98                invalidate_mapping_pages(mapping, 0, -1);
  99        }
 100        /* 99% of the time, we don't need to flush the cleancache on the bdev.
 101         * But, for the strange corners, lets be cautious
 102         */
 103        cleancache_invalidate_inode(mapping);
 104}
 105EXPORT_SYMBOL(invalidate_bdev);
 106
 107static void set_init_blocksize(struct block_device *bdev)
 108{
 109        unsigned bsize = bdev_logical_block_size(bdev);
 110        loff_t size = i_size_read(bdev->bd_inode);
 111
 112        while (bsize < PAGE_SIZE) {
 113                if (size & bsize)
 114                        break;
 115                bsize <<= 1;
 116        }
 117        bdev->bd_block_size = bsize;
 118        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 119}
 120
 121int set_blocksize(struct block_device *bdev, int size)
 122{
 123        /* Size must be a power of two, and between 512 and PAGE_SIZE */
 124        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
 125                return -EINVAL;
 126
 127        /* Size cannot be smaller than the size supported by the device */
 128        if (size < bdev_logical_block_size(bdev))
 129                return -EINVAL;
 130
 131        /* Don't change the size if it is same as current */
 132        if (bdev->bd_block_size != size) {
 133                sync_blockdev(bdev);
 134                bdev->bd_block_size = size;
 135                bdev->bd_inode->i_blkbits = blksize_bits(size);
 136                kill_bdev(bdev);
 137        }
 138        return 0;
 139}
 140
 141EXPORT_SYMBOL(set_blocksize);
 142
 143int sb_set_blocksize(struct super_block *sb, int size)
 144{
 145        if (set_blocksize(sb->s_bdev, size))
 146                return 0;
 147        /* If we get here, we know size is power of two
 148         * and it's value is between 512 and PAGE_SIZE */
 149        sb->s_blocksize = size;
 150        sb->s_blocksize_bits = blksize_bits(size);
 151        return sb->s_blocksize;
 152}
 153
 154EXPORT_SYMBOL(sb_set_blocksize);
 155
 156int sb_min_blocksize(struct super_block *sb, int size)
 157{
 158        int minsize = bdev_logical_block_size(sb->s_bdev);
 159        if (size < minsize)
 160                size = minsize;
 161        return sb_set_blocksize(sb, size);
 162}
 163
 164EXPORT_SYMBOL(sb_min_blocksize);
 165
 166static int
 167blkdev_get_block(struct inode *inode, sector_t iblock,
 168                struct buffer_head *bh, int create)
 169{
 170        bh->b_bdev = I_BDEV(inode);
 171        bh->b_blocknr = iblock;
 172        set_buffer_mapped(bh);
 173        return 0;
 174}
 175
 176static struct inode *bdev_file_inode(struct file *file)
 177{
 178        return file->f_mapping->host;
 179}
 180
 181static unsigned int dio_bio_write_op(struct kiocb *iocb)
 182{
 183        unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
 184
 185        /* avoid the need for a I/O completion work item */
 186        if (iocb->ki_flags & IOCB_DSYNC)
 187                op |= REQ_FUA;
 188        return op;
 189}
 190
 191#define DIO_INLINE_BIO_VECS 4
 192
 193static void blkdev_bio_end_io_simple(struct bio *bio)
 194{
 195        struct task_struct *waiter = bio->bi_private;
 196
 197        WRITE_ONCE(bio->bi_private, NULL);
 198        blk_wake_io_task(waiter);
 199}
 200
 201static ssize_t
 202__blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
 203                int nr_pages)
 204{
 205        struct file *file = iocb->ki_filp;
 206        struct block_device *bdev = I_BDEV(bdev_file_inode(file));
 207        struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
 208        loff_t pos = iocb->ki_pos;
 209        bool should_dirty = false;
 210        struct bio bio;
 211        ssize_t ret;
 212        blk_qc_t qc;
 213
 214        if ((pos | iov_iter_alignment(iter)) &
 215            (bdev_logical_block_size(bdev) - 1))
 216                return -EINVAL;
 217
 218        if (nr_pages <= DIO_INLINE_BIO_VECS)
 219                vecs = inline_vecs;
 220        else {
 221                vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
 222                                     GFP_KERNEL);
 223                if (!vecs)
 224                        return -ENOMEM;
 225        }
 226
 227        bio_init(&bio, vecs, nr_pages);
 228        bio_set_dev(&bio, bdev);
 229        bio.bi_iter.bi_sector = pos >> 9;
 230        bio.bi_write_hint = iocb->ki_hint;
 231        bio.bi_private = current;
 232        bio.bi_end_io = blkdev_bio_end_io_simple;
 233        bio.bi_ioprio = iocb->ki_ioprio;
 234
 235        ret = bio_iov_iter_get_pages(&bio, iter);
 236        if (unlikely(ret))
 237                goto out;
 238        ret = bio.bi_iter.bi_size;
 239
 240        if (iov_iter_rw(iter) == READ) {
 241                bio.bi_opf = REQ_OP_READ;
 242                if (iter_is_iovec(iter))
 243                        should_dirty = true;
 244        } else {
 245                bio.bi_opf = dio_bio_write_op(iocb);
 246                task_io_account_write(ret);
 247        }
 248        if (iocb->ki_flags & IOCB_HIPRI)
 249                bio_set_polled(&bio, iocb);
 250
 251        qc = submit_bio(&bio);
 252        for (;;) {
 253                set_current_state(TASK_UNINTERRUPTIBLE);
 254                if (!READ_ONCE(bio.bi_private))
 255                        break;
 256                if (!(iocb->ki_flags & IOCB_HIPRI) ||
 257                    !blk_poll(bdev_get_queue(bdev), qc, true))
 258                        io_schedule();
 259        }
 260        __set_current_state(TASK_RUNNING);
 261
 262        bio_release_pages(&bio, should_dirty);
 263        if (unlikely(bio.bi_status))
 264                ret = blk_status_to_errno(bio.bi_status);
 265
 266out:
 267        if (vecs != inline_vecs)
 268                kfree(vecs);
 269
 270        bio_uninit(&bio);
 271
 272        return ret;
 273}
 274
 275struct blkdev_dio {
 276        union {
 277                struct kiocb            *iocb;
 278                struct task_struct      *waiter;
 279        };
 280        size_t                  size;
 281        atomic_t                ref;
 282        bool                    multi_bio : 1;
 283        bool                    should_dirty : 1;
 284        bool                    is_sync : 1;
 285        struct bio              bio;
 286};
 287
 288static struct bio_set blkdev_dio_pool;
 289
 290static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
 291{
 292        struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
 293        struct request_queue *q = bdev_get_queue(bdev);
 294
 295        return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
 296}
 297
 298static void blkdev_bio_end_io(struct bio *bio)
 299{
 300        struct blkdev_dio *dio = bio->bi_private;
 301        bool should_dirty = dio->should_dirty;
 302
 303        if (bio->bi_status && !dio->bio.bi_status)
 304                dio->bio.bi_status = bio->bi_status;
 305
 306        if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
 307                if (!dio->is_sync) {
 308                        struct kiocb *iocb = dio->iocb;
 309                        ssize_t ret;
 310
 311                        if (likely(!dio->bio.bi_status)) {
 312                                ret = dio->size;
 313                                iocb->ki_pos += ret;
 314                        } else {
 315                                ret = blk_status_to_errno(dio->bio.bi_status);
 316                        }
 317
 318                        dio->iocb->ki_complete(iocb, ret, 0);
 319                        if (dio->multi_bio)
 320                                bio_put(&dio->bio);
 321                } else {
 322                        struct task_struct *waiter = dio->waiter;
 323
 324                        WRITE_ONCE(dio->waiter, NULL);
 325                        blk_wake_io_task(waiter);
 326                }
 327        }
 328
 329        if (should_dirty) {
 330                bio_check_pages_dirty(bio);
 331        } else {
 332                bio_release_pages(bio, false);
 333                bio_put(bio);
 334        }
 335}
 336
 337static ssize_t
 338__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
 339{
 340        struct file *file = iocb->ki_filp;
 341        struct inode *inode = bdev_file_inode(file);
 342        struct block_device *bdev = I_BDEV(inode);
 343        struct blk_plug plug;
 344        struct blkdev_dio *dio;
 345        struct bio *bio;
 346        bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
 347        bool is_read = (iov_iter_rw(iter) == READ), is_sync;
 348        loff_t pos = iocb->ki_pos;
 349        blk_qc_t qc = BLK_QC_T_NONE;
 350        int ret = 0;
 351
 352        if ((pos | iov_iter_alignment(iter)) &
 353            (bdev_logical_block_size(bdev) - 1))
 354                return -EINVAL;
 355
 356        bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
 357
 358        dio = container_of(bio, struct blkdev_dio, bio);
 359        dio->is_sync = is_sync = is_sync_kiocb(iocb);
 360        if (dio->is_sync) {
 361                dio->waiter = current;
 362                bio_get(bio);
 363        } else {
 364                dio->iocb = iocb;
 365        }
 366
 367        dio->size = 0;
 368        dio->multi_bio = false;
 369        dio->should_dirty = is_read && iter_is_iovec(iter);
 370
 371        /*
 372         * Don't plug for HIPRI/polled IO, as those should go straight
 373         * to issue
 374         */
 375        if (!is_poll)
 376                blk_start_plug(&plug);
 377
 378        for (;;) {
 379                bio_set_dev(bio, bdev);
 380                bio->bi_iter.bi_sector = pos >> 9;
 381                bio->bi_write_hint = iocb->ki_hint;
 382                bio->bi_private = dio;
 383                bio->bi_end_io = blkdev_bio_end_io;
 384                bio->bi_ioprio = iocb->ki_ioprio;
 385
 386                ret = bio_iov_iter_get_pages(bio, iter);
 387                if (unlikely(ret)) {
 388                        bio->bi_status = BLK_STS_IOERR;
 389                        bio_endio(bio);
 390                        break;
 391                }
 392
 393                if (is_read) {
 394                        bio->bi_opf = REQ_OP_READ;
 395                        if (dio->should_dirty)
 396                                bio_set_pages_dirty(bio);
 397                } else {
 398                        bio->bi_opf = dio_bio_write_op(iocb);
 399                        task_io_account_write(bio->bi_iter.bi_size);
 400                }
 401
 402                dio->size += bio->bi_iter.bi_size;
 403                pos += bio->bi_iter.bi_size;
 404
 405                nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
 406                if (!nr_pages) {
 407                        bool polled = false;
 408
 409                        if (iocb->ki_flags & IOCB_HIPRI) {
 410                                bio_set_polled(bio, iocb);
 411                                polled = true;
 412                        }
 413
 414                        qc = submit_bio(bio);
 415
 416                        if (polled)
 417                                WRITE_ONCE(iocb->ki_cookie, qc);
 418                        break;
 419                }
 420
 421                if (!dio->multi_bio) {
 422                        /*
 423                         * AIO needs an extra reference to ensure the dio
 424                         * structure which is embedded into the first bio
 425                         * stays around.
 426                         */
 427                        if (!is_sync)
 428                                bio_get(bio);
 429                        dio->multi_bio = true;
 430                        atomic_set(&dio->ref, 2);
 431                } else {
 432                        atomic_inc(&dio->ref);
 433                }
 434
 435                submit_bio(bio);
 436                bio = bio_alloc(GFP_KERNEL, nr_pages);
 437        }
 438
 439        if (!is_poll)
 440                blk_finish_plug(&plug);
 441
 442        if (!is_sync)
 443                return -EIOCBQUEUED;
 444
 445        for (;;) {
 446                set_current_state(TASK_UNINTERRUPTIBLE);
 447                if (!READ_ONCE(dio->waiter))
 448                        break;
 449
 450                if (!(iocb->ki_flags & IOCB_HIPRI) ||
 451                    !blk_poll(bdev_get_queue(bdev), qc, true))
 452                        io_schedule();
 453        }
 454        __set_current_state(TASK_RUNNING);
 455
 456        if (!ret)
 457                ret = blk_status_to_errno(dio->bio.bi_status);
 458        if (likely(!ret))
 459                ret = dio->size;
 460
 461        bio_put(&dio->bio);
 462        return ret;
 463}
 464
 465static ssize_t
 466blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 467{
 468        int nr_pages;
 469
 470        nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
 471        if (!nr_pages)
 472                return 0;
 473        if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
 474                return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
 475
 476        return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
 477}
 478
 479static __init int blkdev_init(void)
 480{
 481        return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
 482}
 483module_init(blkdev_init);
 484
 485int __sync_blockdev(struct block_device *bdev, int wait)
 486{
 487        if (!bdev)
 488                return 0;
 489        if (!wait)
 490                return filemap_flush(bdev->bd_inode->i_mapping);
 491        return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 492}
 493
 494/*
 495 * Write out and wait upon all the dirty data associated with a block
 496 * device via its mapping.  Does not take the superblock lock.
 497 */
 498int sync_blockdev(struct block_device *bdev)
 499{
 500        return __sync_blockdev(bdev, 1);
 501}
 502EXPORT_SYMBOL(sync_blockdev);
 503
 504/*
 505 * Write out and wait upon all dirty data associated with this
 506 * device.   Filesystem data as well as the underlying block
 507 * device.  Takes the superblock lock.
 508 */
 509int fsync_bdev(struct block_device *bdev)
 510{
 511        struct super_block *sb = get_super(bdev);
 512        if (sb) {
 513                int res = sync_filesystem(sb);
 514                drop_super(sb);
 515                return res;
 516        }
 517        return sync_blockdev(bdev);
 518}
 519EXPORT_SYMBOL(fsync_bdev);
 520
 521/**
 522 * freeze_bdev  --  lock a filesystem and force it into a consistent state
 523 * @bdev:       blockdevice to lock
 524 *
 525 * If a superblock is found on this device, we take the s_umount semaphore
 526 * on it to make sure nobody unmounts until the snapshot creation is done.
 527 * The reference counter (bd_fsfreeze_count) guarantees that only the last
 528 * unfreeze process can unfreeze the frozen filesystem actually when multiple
 529 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 530 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 531 * actually.
 532 */
 533struct super_block *freeze_bdev(struct block_device *bdev)
 534{
 535        struct super_block *sb;
 536        int error = 0;
 537
 538        mutex_lock(&bdev->bd_fsfreeze_mutex);
 539        if (++bdev->bd_fsfreeze_count > 1) {
 540                /*
 541                 * We don't even need to grab a reference - the first call
 542                 * to freeze_bdev grab an active reference and only the last
 543                 * thaw_bdev drops it.
 544                 */
 545                sb = get_super(bdev);
 546                if (sb)
 547                        drop_super(sb);
 548                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 549                return sb;
 550        }
 551
 552        sb = get_active_super(bdev);
 553        if (!sb)
 554                goto out;
 555        if (sb->s_op->freeze_super)
 556                error = sb->s_op->freeze_super(sb);
 557        else
 558                error = freeze_super(sb);
 559        if (error) {
 560                deactivate_super(sb);
 561                bdev->bd_fsfreeze_count--;
 562                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 563                return ERR_PTR(error);
 564        }
 565        deactivate_super(sb);
 566 out:
 567        sync_blockdev(bdev);
 568        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 569        return sb;      /* thaw_bdev releases s->s_umount */
 570}
 571EXPORT_SYMBOL(freeze_bdev);
 572
 573/**
 574 * thaw_bdev  -- unlock filesystem
 575 * @bdev:       blockdevice to unlock
 576 * @sb:         associated superblock
 577 *
 578 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 579 */
 580int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 581{
 582        int error = -EINVAL;
 583
 584        mutex_lock(&bdev->bd_fsfreeze_mutex);
 585        if (!bdev->bd_fsfreeze_count)
 586                goto out;
 587
 588        error = 0;
 589        if (--bdev->bd_fsfreeze_count > 0)
 590                goto out;
 591
 592        if (!sb)
 593                goto out;
 594
 595        if (sb->s_op->thaw_super)
 596                error = sb->s_op->thaw_super(sb);
 597        else
 598                error = thaw_super(sb);
 599        if (error)
 600                bdev->bd_fsfreeze_count++;
 601out:
 602        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 603        return error;
 604}
 605EXPORT_SYMBOL(thaw_bdev);
 606
 607static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 608{
 609        return block_write_full_page(page, blkdev_get_block, wbc);
 610}
 611
 612static int blkdev_readpage(struct file * file, struct page * page)
 613{
 614        return block_read_full_page(page, blkdev_get_block);
 615}
 616
 617static int blkdev_readpages(struct file *file, struct address_space *mapping,
 618                        struct list_head *pages, unsigned nr_pages)
 619{
 620        return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
 621}
 622
 623static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 624                        loff_t pos, unsigned len, unsigned flags,
 625                        struct page **pagep, void **fsdata)
 626{
 627        return block_write_begin(mapping, pos, len, flags, pagep,
 628                                 blkdev_get_block);
 629}
 630
 631static int blkdev_write_end(struct file *file, struct address_space *mapping,
 632                        loff_t pos, unsigned len, unsigned copied,
 633                        struct page *page, void *fsdata)
 634{
 635        int ret;
 636        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 637
 638        unlock_page(page);
 639        put_page(page);
 640
 641        return ret;
 642}
 643
 644/*
 645 * private llseek:
 646 * for a block special file file_inode(file)->i_size is zero
 647 * so we compute the size by hand (just as in block_read/write above)
 648 */
 649static loff_t block_llseek(struct file *file, loff_t offset, int whence)
 650{
 651        struct inode *bd_inode = bdev_file_inode(file);
 652        loff_t retval;
 653
 654        inode_lock(bd_inode);
 655        retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
 656        inode_unlock(bd_inode);
 657        return retval;
 658}
 659        
 660int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 661{
 662        struct inode *bd_inode = bdev_file_inode(filp);
 663        struct block_device *bdev = I_BDEV(bd_inode);
 664        int error;
 665        
 666        error = file_write_and_wait_range(filp, start, end);
 667        if (error)
 668                return error;
 669
 670        /*
 671         * There is no need to serialise calls to blkdev_issue_flush with
 672         * i_mutex and doing so causes performance issues with concurrent
 673         * O_SYNC writers to a block device.
 674         */
 675        error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
 676        if (error == -EOPNOTSUPP)
 677                error = 0;
 678
 679        return error;
 680}
 681EXPORT_SYMBOL(blkdev_fsync);
 682
 683/**
 684 * bdev_read_page() - Start reading a page from a block device
 685 * @bdev: The device to read the page from
 686 * @sector: The offset on the device to read the page to (need not be aligned)
 687 * @page: The page to read
 688 *
 689 * On entry, the page should be locked.  It will be unlocked when the page
 690 * has been read.  If the block driver implements rw_page synchronously,
 691 * that will be true on exit from this function, but it need not be.
 692 *
 693 * Errors returned by this function are usually "soft", eg out of memory, or
 694 * queue full; callers should try a different route to read this page rather
 695 * than propagate an error back up the stack.
 696 *
 697 * Return: negative errno if an error occurs, 0 if submission was successful.
 698 */
 699int bdev_read_page(struct block_device *bdev, sector_t sector,
 700                        struct page *page)
 701{
 702        const struct block_device_operations *ops = bdev->bd_disk->fops;
 703        int result = -EOPNOTSUPP;
 704
 705        if (!ops->rw_page || bdev_get_integrity(bdev))
 706                return result;
 707
 708        result = blk_queue_enter(bdev->bd_queue, 0);
 709        if (result)
 710                return result;
 711        result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
 712                              REQ_OP_READ);
 713        blk_queue_exit(bdev->bd_queue);
 714        return result;
 715}
 716EXPORT_SYMBOL_GPL(bdev_read_page);
 717
 718/**
 719 * bdev_write_page() - Start writing a page to a block device
 720 * @bdev: The device to write the page to
 721 * @sector: The offset on the device to write the page to (need not be aligned)
 722 * @page: The page to write
 723 * @wbc: The writeback_control for the write
 724 *
 725 * On entry, the page should be locked and not currently under writeback.
 726 * On exit, if the write started successfully, the page will be unlocked and
 727 * under writeback.  If the write failed already (eg the driver failed to
 728 * queue the page to the device), the page will still be locked.  If the
 729 * caller is a ->writepage implementation, it will need to unlock the page.
 730 *
 731 * Errors returned by this function are usually "soft", eg out of memory, or
 732 * queue full; callers should try a different route to write this page rather
 733 * than propagate an error back up the stack.
 734 *
 735 * Return: negative errno if an error occurs, 0 if submission was successful.
 736 */
 737int bdev_write_page(struct block_device *bdev, sector_t sector,
 738                        struct page *page, struct writeback_control *wbc)
 739{
 740        int result;
 741        const struct block_device_operations *ops = bdev->bd_disk->fops;
 742
 743        if (!ops->rw_page || bdev_get_integrity(bdev))
 744                return -EOPNOTSUPP;
 745        result = blk_queue_enter(bdev->bd_queue, 0);
 746        if (result)
 747                return result;
 748
 749        set_page_writeback(page);
 750        result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
 751                              REQ_OP_WRITE);
 752        if (result) {
 753                end_page_writeback(page);
 754        } else {
 755                clean_page_buffers(page);
 756                unlock_page(page);
 757        }
 758        blk_queue_exit(bdev->bd_queue);
 759        return result;
 760}
 761EXPORT_SYMBOL_GPL(bdev_write_page);
 762
 763/*
 764 * pseudo-fs
 765 */
 766
 767static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 768static struct kmem_cache * bdev_cachep __read_mostly;
 769
 770static struct inode *bdev_alloc_inode(struct super_block *sb)
 771{
 772        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 773        if (!ei)
 774                return NULL;
 775        return &ei->vfs_inode;
 776}
 777
 778static void bdev_free_inode(struct inode *inode)
 779{
 780        kmem_cache_free(bdev_cachep, BDEV_I(inode));
 781}
 782
 783static void init_once(void *foo)
 784{
 785        struct bdev_inode *ei = (struct bdev_inode *) foo;
 786        struct block_device *bdev = &ei->bdev;
 787
 788        memset(bdev, 0, sizeof(*bdev));
 789        mutex_init(&bdev->bd_mutex);
 790        INIT_LIST_HEAD(&bdev->bd_list);
 791#ifdef CONFIG_SYSFS
 792        INIT_LIST_HEAD(&bdev->bd_holder_disks);
 793#endif
 794        bdev->bd_bdi = &noop_backing_dev_info;
 795        inode_init_once(&ei->vfs_inode);
 796        /* Initialize mutex for freeze. */
 797        mutex_init(&bdev->bd_fsfreeze_mutex);
 798}
 799
 800static void bdev_evict_inode(struct inode *inode)
 801{
 802        struct block_device *bdev = &BDEV_I(inode)->bdev;
 803        truncate_inode_pages_final(&inode->i_data);
 804        invalidate_inode_buffers(inode); /* is it needed here? */
 805        clear_inode(inode);
 806        spin_lock(&bdev_lock);
 807        list_del_init(&bdev->bd_list);
 808        spin_unlock(&bdev_lock);
 809        /* Detach inode from wb early as bdi_put() may free bdi->wb */
 810        inode_detach_wb(inode);
 811        if (bdev->bd_bdi != &noop_backing_dev_info) {
 812                bdi_put(bdev->bd_bdi);
 813                bdev->bd_bdi = &noop_backing_dev_info;
 814        }
 815}
 816
 817static const struct super_operations bdev_sops = {
 818        .statfs = simple_statfs,
 819        .alloc_inode = bdev_alloc_inode,
 820        .free_inode = bdev_free_inode,
 821        .drop_inode = generic_delete_inode,
 822        .evict_inode = bdev_evict_inode,
 823};
 824
 825static int bd_init_fs_context(struct fs_context *fc)
 826{
 827        struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
 828        if (!ctx)
 829                return -ENOMEM;
 830        fc->s_iflags |= SB_I_CGROUPWB;
 831        ctx->ops = &bdev_sops;
 832        return 0;
 833}
 834
 835static struct file_system_type bd_type = {
 836        .name           = "bdev",
 837        .init_fs_context = bd_init_fs_context,
 838        .kill_sb        = kill_anon_super,
 839};
 840
 841struct super_block *blockdev_superblock __read_mostly;
 842EXPORT_SYMBOL_GPL(blockdev_superblock);
 843
 844void __init bdev_cache_init(void)
 845{
 846        int err;
 847        static struct vfsmount *bd_mnt;
 848
 849        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 850                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 851                                SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
 852                        init_once);
 853        err = register_filesystem(&bd_type);
 854        if (err)
 855                panic("Cannot register bdev pseudo-fs");
 856        bd_mnt = kern_mount(&bd_type);
 857        if (IS_ERR(bd_mnt))
 858                panic("Cannot create bdev pseudo-fs");
 859        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 860}
 861
 862/*
 863 * Most likely _very_ bad one - but then it's hardly critical for small
 864 * /dev and can be fixed when somebody will need really large one.
 865 * Keep in mind that it will be fed through icache hash function too.
 866 */
 867static inline unsigned long hash(dev_t dev)
 868{
 869        return MAJOR(dev)+MINOR(dev);
 870}
 871
 872static int bdev_test(struct inode *inode, void *data)
 873{
 874        return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 875}
 876
 877static int bdev_set(struct inode *inode, void *data)
 878{
 879        BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 880        return 0;
 881}
 882
 883static LIST_HEAD(all_bdevs);
 884
 885/*
 886 * If there is a bdev inode for this device, unhash it so that it gets evicted
 887 * as soon as last inode reference is dropped.
 888 */
 889void bdev_unhash_inode(dev_t dev)
 890{
 891        struct inode *inode;
 892
 893        inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
 894        if (inode) {
 895                remove_inode_hash(inode);
 896                iput(inode);
 897        }
 898}
 899
 900struct block_device *bdget(dev_t dev)
 901{
 902        struct block_device *bdev;
 903        struct inode *inode;
 904
 905        inode = iget5_locked(blockdev_superblock, hash(dev),
 906                        bdev_test, bdev_set, &dev);
 907
 908        if (!inode)
 909                return NULL;
 910
 911        bdev = &BDEV_I(inode)->bdev;
 912
 913        if (inode->i_state & I_NEW) {
 914                bdev->bd_contains = NULL;
 915                bdev->bd_super = NULL;
 916                bdev->bd_inode = inode;
 917                bdev->bd_block_size = i_blocksize(inode);
 918                bdev->bd_part_count = 0;
 919                bdev->bd_invalidated = 0;
 920                inode->i_mode = S_IFBLK;
 921                inode->i_rdev = dev;
 922                inode->i_bdev = bdev;
 923                inode->i_data.a_ops = &def_blk_aops;
 924                mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 925                spin_lock(&bdev_lock);
 926                list_add(&bdev->bd_list, &all_bdevs);
 927                spin_unlock(&bdev_lock);
 928                unlock_new_inode(inode);
 929        }
 930        return bdev;
 931}
 932
 933EXPORT_SYMBOL(bdget);
 934
 935/**
 936 * bdgrab -- Grab a reference to an already referenced block device
 937 * @bdev:       Block device to grab a reference to.
 938 */
 939struct block_device *bdgrab(struct block_device *bdev)
 940{
 941        ihold(bdev->bd_inode);
 942        return bdev;
 943}
 944EXPORT_SYMBOL(bdgrab);
 945
 946long nr_blockdev_pages(void)
 947{
 948        struct block_device *bdev;
 949        long ret = 0;
 950        spin_lock(&bdev_lock);
 951        list_for_each_entry(bdev, &all_bdevs, bd_list) {
 952                ret += bdev->bd_inode->i_mapping->nrpages;
 953        }
 954        spin_unlock(&bdev_lock);
 955        return ret;
 956}
 957
 958void bdput(struct block_device *bdev)
 959{
 960        iput(bdev->bd_inode);
 961}
 962
 963EXPORT_SYMBOL(bdput);
 964 
 965static struct block_device *bd_acquire(struct inode *inode)
 966{
 967        struct block_device *bdev;
 968
 969        spin_lock(&bdev_lock);
 970        bdev = inode->i_bdev;
 971        if (bdev && !inode_unhashed(bdev->bd_inode)) {
 972                bdgrab(bdev);
 973                spin_unlock(&bdev_lock);
 974                return bdev;
 975        }
 976        spin_unlock(&bdev_lock);
 977
 978        /*
 979         * i_bdev references block device inode that was already shut down
 980         * (corresponding device got removed).  Remove the reference and look
 981         * up block device inode again just in case new device got
 982         * reestablished under the same device number.
 983         */
 984        if (bdev)
 985                bd_forget(inode);
 986
 987        bdev = bdget(inode->i_rdev);
 988        if (bdev) {
 989                spin_lock(&bdev_lock);
 990                if (!inode->i_bdev) {
 991                        /*
 992                         * We take an additional reference to bd_inode,
 993                         * and it's released in clear_inode() of inode.
 994                         * So, we can access it via ->i_mapping always
 995                         * without igrab().
 996                         */
 997                        bdgrab(bdev);
 998                        inode->i_bdev = bdev;
 999                        inode->i_mapping = bdev->bd_inode->i_mapping;
1000                }
1001                spin_unlock(&bdev_lock);
1002        }
1003        return bdev;
1004}
1005
1006/* Call when you free inode */
1007
1008void bd_forget(struct inode *inode)
1009{
1010        struct block_device *bdev = NULL;
1011
1012        spin_lock(&bdev_lock);
1013        if (!sb_is_blkdev_sb(inode->i_sb))
1014                bdev = inode->i_bdev;
1015        inode->i_bdev = NULL;
1016        inode->i_mapping = &inode->i_data;
1017        spin_unlock(&bdev_lock);
1018
1019        if (bdev)
1020                bdput(bdev);
1021}
1022
1023/**
1024 * bd_may_claim - test whether a block device can be claimed
1025 * @bdev: block device of interest
1026 * @whole: whole block device containing @bdev, may equal @bdev
1027 * @holder: holder trying to claim @bdev
1028 *
1029 * Test whether @bdev can be claimed by @holder.
1030 *
1031 * CONTEXT:
1032 * spin_lock(&bdev_lock).
1033 *
1034 * RETURNS:
1035 * %true if @bdev can be claimed, %false otherwise.
1036 */
1037static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1038                         void *holder)
1039{
1040        if (bdev->bd_holder == holder)
1041                return true;     /* already a holder */
1042        else if (bdev->bd_holder != NULL)
1043                return false;    /* held by someone else */
1044        else if (whole == bdev)
1045                return true;     /* is a whole device which isn't held */
1046
1047        else if (whole->bd_holder == bd_may_claim)
1048                return true;     /* is a partition of a device that is being partitioned */
1049        else if (whole->bd_holder != NULL)
1050                return false;    /* is a partition of a held device */
1051        else
1052                return true;     /* is a partition of an un-held device */
1053}
1054
1055/**
1056 * bd_prepare_to_claim - prepare to claim a block device
1057 * @bdev: block device of interest
1058 * @whole: the whole device containing @bdev, may equal @bdev
1059 * @holder: holder trying to claim @bdev
1060 *
1061 * Prepare to claim @bdev.  This function fails if @bdev is already
1062 * claimed by another holder and waits if another claiming is in
1063 * progress.  This function doesn't actually claim.  On successful
1064 * return, the caller has ownership of bd_claiming and bd_holder[s].
1065 *
1066 * CONTEXT:
1067 * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
1068 * it multiple times.
1069 *
1070 * RETURNS:
1071 * 0 if @bdev can be claimed, -EBUSY otherwise.
1072 */
1073static int bd_prepare_to_claim(struct block_device *bdev,
1074                               struct block_device *whole, void *holder)
1075{
1076retry:
1077        /* if someone else claimed, fail */
1078        if (!bd_may_claim(bdev, whole, holder))
1079                return -EBUSY;
1080
1081        /* if claiming is already in progress, wait for it to finish */
1082        if (whole->bd_claiming) {
1083                wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1084                DEFINE_WAIT(wait);
1085
1086                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1087                spin_unlock(&bdev_lock);
1088                schedule();
1089                finish_wait(wq, &wait);
1090                spin_lock(&bdev_lock);
1091                goto retry;
1092        }
1093
1094        /* yay, all mine */
1095        return 0;
1096}
1097
1098static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1099{
1100        struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1101
1102        if (!disk)
1103                return NULL;
1104        /*
1105         * Now that we hold gendisk reference we make sure bdev we looked up is
1106         * not stale. If it is, it means device got removed and created before
1107         * we looked up gendisk and we fail open in such case. Associating
1108         * unhashed bdev with newly created gendisk could lead to two bdevs
1109         * (and thus two independent caches) being associated with one device
1110         * which is bad.
1111         */
1112        if (inode_unhashed(bdev->bd_inode)) {
1113                put_disk_and_module(disk);
1114                return NULL;
1115        }
1116        return disk;
1117}
1118
1119/**
1120 * bd_start_claiming - start claiming a block device
1121 * @bdev: block device of interest
1122 * @holder: holder trying to claim @bdev
1123 *
1124 * @bdev is about to be opened exclusively.  Check @bdev can be opened
1125 * exclusively and mark that an exclusive open is in progress.  Each
1126 * successful call to this function must be matched with a call to
1127 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1128 * fail).
1129 *
1130 * This function is used to gain exclusive access to the block device
1131 * without actually causing other exclusive open attempts to fail. It
1132 * should be used when the open sequence itself requires exclusive
1133 * access but may subsequently fail.
1134 *
1135 * CONTEXT:
1136 * Might sleep.
1137 *
1138 * RETURNS:
1139 * Pointer to the block device containing @bdev on success, ERR_PTR()
1140 * value on failure.
1141 */
1142struct block_device *bd_start_claiming(struct block_device *bdev, void *holder)
1143{
1144        struct gendisk *disk;
1145        struct block_device *whole;
1146        int partno, err;
1147
1148        might_sleep();
1149
1150        /*
1151         * @bdev might not have been initialized properly yet, look up
1152         * and grab the outer block device the hard way.
1153         */
1154        disk = bdev_get_gendisk(bdev, &partno);
1155        if (!disk)
1156                return ERR_PTR(-ENXIO);
1157
1158        /*
1159         * Normally, @bdev should equal what's returned from bdget_disk()
1160         * if partno is 0; however, some drivers (floppy) use multiple
1161         * bdev's for the same physical device and @bdev may be one of the
1162         * aliases.  Keep @bdev if partno is 0.  This means claimer
1163         * tracking is broken for those devices but it has always been that
1164         * way.
1165         */
1166        if (partno)
1167                whole = bdget_disk(disk, 0);
1168        else
1169                whole = bdgrab(bdev);
1170
1171        put_disk_and_module(disk);
1172        if (!whole)
1173                return ERR_PTR(-ENOMEM);
1174
1175        /* prepare to claim, if successful, mark claiming in progress */
1176        spin_lock(&bdev_lock);
1177
1178        err = bd_prepare_to_claim(bdev, whole, holder);
1179        if (err == 0) {
1180                whole->bd_claiming = holder;
1181                spin_unlock(&bdev_lock);
1182                return whole;
1183        } else {
1184                spin_unlock(&bdev_lock);
1185                bdput(whole);
1186                return ERR_PTR(err);
1187        }
1188}
1189EXPORT_SYMBOL(bd_start_claiming);
1190
1191static void bd_clear_claiming(struct block_device *whole, void *holder)
1192{
1193        lockdep_assert_held(&bdev_lock);
1194        /* tell others that we're done */
1195        BUG_ON(whole->bd_claiming != holder);
1196        whole->bd_claiming = NULL;
1197        wake_up_bit(&whole->bd_claiming, 0);
1198}
1199
1200/**
1201 * bd_finish_claiming - finish claiming of a block device
1202 * @bdev: block device of interest
1203 * @whole: whole block device (returned from bd_start_claiming())
1204 * @holder: holder that has claimed @bdev
1205 *
1206 * Finish exclusive open of a block device. Mark the device as exlusively
1207 * open by the holder and wake up all waiters for exclusive open to finish.
1208 */
1209void bd_finish_claiming(struct block_device *bdev, struct block_device *whole,
1210                        void *holder)
1211{
1212        spin_lock(&bdev_lock);
1213        BUG_ON(!bd_may_claim(bdev, whole, holder));
1214        /*
1215         * Note that for a whole device bd_holders will be incremented twice,
1216         * and bd_holder will be set to bd_may_claim before being set to holder
1217         */
1218        whole->bd_holders++;
1219        whole->bd_holder = bd_may_claim;
1220        bdev->bd_holders++;
1221        bdev->bd_holder = holder;
1222        bd_clear_claiming(whole, holder);
1223        spin_unlock(&bdev_lock);
1224}
1225EXPORT_SYMBOL(bd_finish_claiming);
1226
1227/**
1228 * bd_abort_claiming - abort claiming of a block device
1229 * @bdev: block device of interest
1230 * @whole: whole block device (returned from bd_start_claiming())
1231 * @holder: holder that has claimed @bdev
1232 *
1233 * Abort claiming of a block device when the exclusive open failed. This can be
1234 * also used when exclusive open is not actually desired and we just needed
1235 * to block other exclusive openers for a while.
1236 */
1237void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1238                       void *holder)
1239{
1240        spin_lock(&bdev_lock);
1241        bd_clear_claiming(whole, holder);
1242        spin_unlock(&bdev_lock);
1243}
1244EXPORT_SYMBOL(bd_abort_claiming);
1245
1246#ifdef CONFIG_SYSFS
1247struct bd_holder_disk {
1248        struct list_head        list;
1249        struct gendisk          *disk;
1250        int                     refcnt;
1251};
1252
1253static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1254                                                  struct gendisk *disk)
1255{
1256        struct bd_holder_disk *holder;
1257
1258        list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1259                if (holder->disk == disk)
1260                        return holder;
1261        return NULL;
1262}
1263
1264static int add_symlink(struct kobject *from, struct kobject *to)
1265{
1266        return sysfs_create_link(from, to, kobject_name(to));
1267}
1268
1269static void del_symlink(struct kobject *from, struct kobject *to)
1270{
1271        sysfs_remove_link(from, kobject_name(to));
1272}
1273
1274/**
1275 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1276 * @bdev: the claimed slave bdev
1277 * @disk: the holding disk
1278 *
1279 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1280 *
1281 * This functions creates the following sysfs symlinks.
1282 *
1283 * - from "slaves" directory of the holder @disk to the claimed @bdev
1284 * - from "holders" directory of the @bdev to the holder @disk
1285 *
1286 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1287 * passed to bd_link_disk_holder(), then:
1288 *
1289 *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
1290 *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1291 *
1292 * The caller must have claimed @bdev before calling this function and
1293 * ensure that both @bdev and @disk are valid during the creation and
1294 * lifetime of these symlinks.
1295 *
1296 * CONTEXT:
1297 * Might sleep.
1298 *
1299 * RETURNS:
1300 * 0 on success, -errno on failure.
1301 */
1302int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1303{
1304        struct bd_holder_disk *holder;
1305        int ret = 0;
1306
1307        mutex_lock(&bdev->bd_mutex);
1308
1309        WARN_ON_ONCE(!bdev->bd_holder);
1310
1311        /* FIXME: remove the following once add_disk() handles errors */
1312        if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1313                goto out_unlock;
1314
1315        holder = bd_find_holder_disk(bdev, disk);
1316        if (holder) {
1317                holder->refcnt++;
1318                goto out_unlock;
1319        }
1320
1321        holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1322        if (!holder) {
1323                ret = -ENOMEM;
1324                goto out_unlock;
1325        }
1326
1327        INIT_LIST_HEAD(&holder->list);
1328        holder->disk = disk;
1329        holder->refcnt = 1;
1330
1331        ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1332        if (ret)
1333                goto out_free;
1334
1335        ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1336        if (ret)
1337                goto out_del;
1338        /*
1339         * bdev could be deleted beneath us which would implicitly destroy
1340         * the holder directory.  Hold on to it.
1341         */
1342        kobject_get(bdev->bd_part->holder_dir);
1343
1344        list_add(&holder->list, &bdev->bd_holder_disks);
1345        goto out_unlock;
1346
1347out_del:
1348        del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1349out_free:
1350        kfree(holder);
1351out_unlock:
1352        mutex_unlock(&bdev->bd_mutex);
1353        return ret;
1354}
1355EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1356
1357/**
1358 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1359 * @bdev: the calimed slave bdev
1360 * @disk: the holding disk
1361 *
1362 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1363 *
1364 * CONTEXT:
1365 * Might sleep.
1366 */
1367void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1368{
1369        struct bd_holder_disk *holder;
1370
1371        mutex_lock(&bdev->bd_mutex);
1372
1373        holder = bd_find_holder_disk(bdev, disk);
1374
1375        if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1376                del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1377                del_symlink(bdev->bd_part->holder_dir,
1378                            &disk_to_dev(disk)->kobj);
1379                kobject_put(bdev->bd_part->holder_dir);
1380                list_del_init(&holder->list);
1381                kfree(holder);
1382        }
1383
1384        mutex_unlock(&bdev->bd_mutex);
1385}
1386EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1387#endif
1388
1389/**
1390 * flush_disk - invalidates all buffer-cache entries on a disk
1391 *
1392 * @bdev:      struct block device to be flushed
1393 * @kill_dirty: flag to guide handling of dirty inodes
1394 *
1395 * Invalidates all buffer-cache entries on a disk. It should be called
1396 * when a disk has been changed -- either by a media change or online
1397 * resize.
1398 */
1399static void flush_disk(struct block_device *bdev, bool kill_dirty)
1400{
1401        if (__invalidate_device(bdev, kill_dirty)) {
1402                printk(KERN_WARNING "VFS: busy inodes on changed media or "
1403                       "resized disk %s\n",
1404                       bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1405        }
1406
1407        if (!bdev->bd_disk)
1408                return;
1409        if (disk_part_scan_enabled(bdev->bd_disk))
1410                bdev->bd_invalidated = 1;
1411}
1412
1413/**
1414 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1415 * @disk: struct gendisk to check
1416 * @bdev: struct bdev to adjust.
1417 * @verbose: if %true log a message about a size change if there is any
1418 *
1419 * This routine checks to see if the bdev size does not match the disk size
1420 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1421 * are freed.
1422 */
1423void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1424                bool verbose)
1425{
1426        loff_t disk_size, bdev_size;
1427
1428        disk_size = (loff_t)get_capacity(disk) << 9;
1429        bdev_size = i_size_read(bdev->bd_inode);
1430        if (disk_size != bdev_size) {
1431                if (verbose) {
1432                        printk(KERN_INFO
1433                               "%s: detected capacity change from %lld to %lld\n",
1434                               disk->disk_name, bdev_size, disk_size);
1435                }
1436                i_size_write(bdev->bd_inode, disk_size);
1437                if (bdev_size > disk_size)
1438                        flush_disk(bdev, false);
1439        }
1440}
1441
1442/**
1443 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1444 * @disk: struct gendisk to be revalidated
1445 *
1446 * This routine is a wrapper for lower-level driver's revalidate_disk
1447 * call-backs.  It is used to do common pre and post operations needed
1448 * for all revalidate_disk operations.
1449 */
1450int revalidate_disk(struct gendisk *disk)
1451{
1452        int ret = 0;
1453
1454        if (disk->fops->revalidate_disk)
1455                ret = disk->fops->revalidate_disk(disk);
1456
1457        /*
1458         * Hidden disks don't have associated bdev so there's no point in
1459         * revalidating it.
1460         */
1461        if (!(disk->flags & GENHD_FL_HIDDEN)) {
1462                struct block_device *bdev = bdget_disk(disk, 0);
1463
1464                if (!bdev)
1465                        return ret;
1466
1467                mutex_lock(&bdev->bd_mutex);
1468                check_disk_size_change(disk, bdev, ret == 0);
1469                bdev->bd_invalidated = 0;
1470                mutex_unlock(&bdev->bd_mutex);
1471                bdput(bdev);
1472        }
1473        return ret;
1474}
1475EXPORT_SYMBOL(revalidate_disk);
1476
1477/*
1478 * This routine checks whether a removable media has been changed,
1479 * and invalidates all buffer-cache-entries in that case. This
1480 * is a relatively slow routine, so we have to try to minimize using
1481 * it. Thus it is called only upon a 'mount' or 'open'. This
1482 * is the best way of combining speed and utility, I think.
1483 * People changing diskettes in the middle of an operation deserve
1484 * to lose :-)
1485 */
1486int check_disk_change(struct block_device *bdev)
1487{
1488        struct gendisk *disk = bdev->bd_disk;
1489        const struct block_device_operations *bdops = disk->fops;
1490        unsigned int events;
1491
1492        events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1493                                   DISK_EVENT_EJECT_REQUEST);
1494        if (!(events & DISK_EVENT_MEDIA_CHANGE))
1495                return 0;
1496
1497        flush_disk(bdev, true);
1498        if (bdops->revalidate_disk)
1499                bdops->revalidate_disk(bdev->bd_disk);
1500        return 1;
1501}
1502
1503EXPORT_SYMBOL(check_disk_change);
1504
1505void bd_set_size(struct block_device *bdev, loff_t size)
1506{
1507        inode_lock(bdev->bd_inode);
1508        i_size_write(bdev->bd_inode, size);
1509        inode_unlock(bdev->bd_inode);
1510}
1511EXPORT_SYMBOL(bd_set_size);
1512
1513static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1514
1515/*
1516 * bd_mutex locking:
1517 *
1518 *  mutex_lock(part->bd_mutex)
1519 *    mutex_lock_nested(whole->bd_mutex, 1)
1520 */
1521
1522static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1523{
1524        struct gendisk *disk;
1525        int ret;
1526        int partno;
1527        int perm = 0;
1528        bool first_open = false;
1529
1530        if (mode & FMODE_READ)
1531                perm |= MAY_READ;
1532        if (mode & FMODE_WRITE)
1533                perm |= MAY_WRITE;
1534        /*
1535         * hooks: /n/, see "layering violations".
1536         */
1537        if (!for_part) {
1538                ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1539                if (ret != 0) {
1540                        bdput(bdev);
1541                        return ret;
1542                }
1543        }
1544
1545 restart:
1546
1547        ret = -ENXIO;
1548        disk = bdev_get_gendisk(bdev, &partno);
1549        if (!disk)
1550                goto out;
1551
1552        disk_block_events(disk);
1553        mutex_lock_nested(&bdev->bd_mutex, for_part);
1554        if (!bdev->bd_openers) {
1555                first_open = true;
1556                bdev->bd_disk = disk;
1557                bdev->bd_queue = disk->queue;
1558                bdev->bd_contains = bdev;
1559                bdev->bd_partno = partno;
1560
1561                if (!partno) {
1562                        ret = -ENXIO;
1563                        bdev->bd_part = disk_get_part(disk, partno);
1564                        if (!bdev->bd_part)
1565                                goto out_clear;
1566
1567                        ret = 0;
1568                        if (disk->fops->open) {
1569                                ret = disk->fops->open(bdev, mode);
1570                                if (ret == -ERESTARTSYS) {
1571                                        /* Lost a race with 'disk' being
1572                                         * deleted, try again.
1573                                         * See md.c
1574                                         */
1575                                        disk_put_part(bdev->bd_part);
1576                                        bdev->bd_part = NULL;
1577                                        bdev->bd_disk = NULL;
1578                                        bdev->bd_queue = NULL;
1579                                        mutex_unlock(&bdev->bd_mutex);
1580                                        disk_unblock_events(disk);
1581                                        put_disk_and_module(disk);
1582                                        goto restart;
1583                                }
1584                        }
1585
1586                        if (!ret) {
1587                                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1588                                set_init_blocksize(bdev);
1589                        }
1590
1591                        /*
1592                         * If the device is invalidated, rescan partition
1593                         * if open succeeded or failed with -ENOMEDIUM.
1594                         * The latter is necessary to prevent ghost
1595                         * partitions on a removed medium.
1596                         */
1597                        if (bdev->bd_invalidated) {
1598                                if (!ret)
1599                                        rescan_partitions(disk, bdev);
1600                                else if (ret == -ENOMEDIUM)
1601                                        invalidate_partitions(disk, bdev);
1602                        }
1603
1604                        if (ret)
1605                                goto out_clear;
1606                } else {
1607                        struct block_device *whole;
1608                        whole = bdget_disk(disk, 0);
1609                        ret = -ENOMEM;
1610                        if (!whole)
1611                                goto out_clear;
1612                        BUG_ON(for_part);
1613                        ret = __blkdev_get(whole, mode, 1);
1614                        if (ret)
1615                                goto out_clear;
1616                        bdev->bd_contains = whole;
1617                        bdev->bd_part = disk_get_part(disk, partno);
1618                        if (!(disk->flags & GENHD_FL_UP) ||
1619                            !bdev->bd_part || !bdev->bd_part->nr_sects) {
1620                                ret = -ENXIO;
1621                                goto out_clear;
1622                        }
1623                        bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1624                        set_init_blocksize(bdev);
1625                }
1626
1627                if (bdev->bd_bdi == &noop_backing_dev_info)
1628                        bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1629        } else {
1630                if (bdev->bd_contains == bdev) {
1631                        ret = 0;
1632                        if (bdev->bd_disk->fops->open)
1633                                ret = bdev->bd_disk->fops->open(bdev, mode);
1634                        /* the same as first opener case, read comment there */
1635                        if (bdev->bd_invalidated) {
1636                                if (!ret)
1637                                        rescan_partitions(bdev->bd_disk, bdev);
1638                                else if (ret == -ENOMEDIUM)
1639                                        invalidate_partitions(bdev->bd_disk, bdev);
1640                        }
1641                        if (ret)
1642                                goto out_unlock_bdev;
1643                }
1644        }
1645        bdev->bd_openers++;
1646        if (for_part)
1647                bdev->bd_part_count++;
1648        mutex_unlock(&bdev->bd_mutex);
1649        disk_unblock_events(disk);
1650        /* only one opener holds refs to the module and disk */
1651        if (!first_open)
1652                put_disk_and_module(disk);
1653        return 0;
1654
1655 out_clear:
1656        disk_put_part(bdev->bd_part);
1657        bdev->bd_disk = NULL;
1658        bdev->bd_part = NULL;
1659        bdev->bd_queue = NULL;
1660        if (bdev != bdev->bd_contains)
1661                __blkdev_put(bdev->bd_contains, mode, 1);
1662        bdev->bd_contains = NULL;
1663 out_unlock_bdev:
1664        mutex_unlock(&bdev->bd_mutex);
1665        disk_unblock_events(disk);
1666        put_disk_and_module(disk);
1667 out:
1668        bdput(bdev);
1669
1670        return ret;
1671}
1672
1673/**
1674 * blkdev_get - open a block device
1675 * @bdev: block_device to open
1676 * @mode: FMODE_* mask
1677 * @holder: exclusive holder identifier
1678 *
1679 * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1680 * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1681 * @holder is invalid.  Exclusive opens may nest for the same @holder.
1682 *
1683 * On success, the reference count of @bdev is unchanged.  On failure,
1684 * @bdev is put.
1685 *
1686 * CONTEXT:
1687 * Might sleep.
1688 *
1689 * RETURNS:
1690 * 0 on success, -errno on failure.
1691 */
1692int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1693{
1694        struct block_device *whole = NULL;
1695        int res;
1696
1697        WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1698
1699        if ((mode & FMODE_EXCL) && holder) {
1700                whole = bd_start_claiming(bdev, holder);
1701                if (IS_ERR(whole)) {
1702                        bdput(bdev);
1703                        return PTR_ERR(whole);
1704                }
1705        }
1706
1707        res = __blkdev_get(bdev, mode, 0);
1708
1709        if (whole) {
1710                struct gendisk *disk = whole->bd_disk;
1711
1712                /* finish claiming */
1713                mutex_lock(&bdev->bd_mutex);
1714                if (!res)
1715                        bd_finish_claiming(bdev, whole, holder);
1716                else
1717                        bd_abort_claiming(bdev, whole, holder);
1718                /*
1719                 * Block event polling for write claims if requested.  Any
1720                 * write holder makes the write_holder state stick until
1721                 * all are released.  This is good enough and tracking
1722                 * individual writeable reference is too fragile given the
1723                 * way @mode is used in blkdev_get/put().
1724                 */
1725                if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1726                    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1727                        bdev->bd_write_holder = true;
1728                        disk_block_events(disk);
1729                }
1730
1731                mutex_unlock(&bdev->bd_mutex);
1732                bdput(whole);
1733        }
1734
1735        return res;
1736}
1737EXPORT_SYMBOL(blkdev_get);
1738
1739/**
1740 * blkdev_get_by_path - open a block device by name
1741 * @path: path to the block device to open
1742 * @mode: FMODE_* mask
1743 * @holder: exclusive holder identifier
1744 *
1745 * Open the blockdevice described by the device file at @path.  @mode
1746 * and @holder are identical to blkdev_get().
1747 *
1748 * On success, the returned block_device has reference count of one.
1749 *
1750 * CONTEXT:
1751 * Might sleep.
1752 *
1753 * RETURNS:
1754 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1755 */
1756struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1757                                        void *holder)
1758{
1759        struct block_device *bdev;
1760        int err;
1761
1762        bdev = lookup_bdev(path);
1763        if (IS_ERR(bdev))
1764                return bdev;
1765
1766        err = blkdev_get(bdev, mode, holder);
1767        if (err)
1768                return ERR_PTR(err);
1769
1770        if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1771                blkdev_put(bdev, mode);
1772                return ERR_PTR(-EACCES);
1773        }
1774
1775        return bdev;
1776}
1777EXPORT_SYMBOL(blkdev_get_by_path);
1778
1779/**
1780 * blkdev_get_by_dev - open a block device by device number
1781 * @dev: device number of block device to open
1782 * @mode: FMODE_* mask
1783 * @holder: exclusive holder identifier
1784 *
1785 * Open the blockdevice described by device number @dev.  @mode and
1786 * @holder are identical to blkdev_get().
1787 *
1788 * Use it ONLY if you really do not have anything better - i.e. when
1789 * you are behind a truly sucky interface and all you are given is a
1790 * device number.  _Never_ to be used for internal purposes.  If you
1791 * ever need it - reconsider your API.
1792 *
1793 * On success, the returned block_device has reference count of one.
1794 *
1795 * CONTEXT:
1796 * Might sleep.
1797 *
1798 * RETURNS:
1799 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1800 */
1801struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1802{
1803        struct block_device *bdev;
1804        int err;
1805
1806        bdev = bdget(dev);
1807        if (!bdev)
1808                return ERR_PTR(-ENOMEM);
1809
1810        err = blkdev_get(bdev, mode, holder);
1811        if (err)
1812                return ERR_PTR(err);
1813
1814        return bdev;
1815}
1816EXPORT_SYMBOL(blkdev_get_by_dev);
1817
1818static int blkdev_open(struct inode * inode, struct file * filp)
1819{
1820        struct block_device *bdev;
1821
1822        /*
1823         * Preserve backwards compatibility and allow large file access
1824         * even if userspace doesn't ask for it explicitly. Some mkfs
1825         * binary needs it. We might want to drop this workaround
1826         * during an unstable branch.
1827         */
1828        filp->f_flags |= O_LARGEFILE;
1829
1830        filp->f_mode |= FMODE_NOWAIT;
1831
1832        if (filp->f_flags & O_NDELAY)
1833                filp->f_mode |= FMODE_NDELAY;
1834        if (filp->f_flags & O_EXCL)
1835                filp->f_mode |= FMODE_EXCL;
1836        if ((filp->f_flags & O_ACCMODE) == 3)
1837                filp->f_mode |= FMODE_WRITE_IOCTL;
1838
1839        bdev = bd_acquire(inode);
1840        if (bdev == NULL)
1841                return -ENOMEM;
1842
1843        filp->f_mapping = bdev->bd_inode->i_mapping;
1844        filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1845
1846        return blkdev_get(bdev, filp->f_mode, filp);
1847}
1848
1849static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1850{
1851        struct gendisk *disk = bdev->bd_disk;
1852        struct block_device *victim = NULL;
1853
1854        mutex_lock_nested(&bdev->bd_mutex, for_part);
1855        if (for_part)
1856                bdev->bd_part_count--;
1857
1858        if (!--bdev->bd_openers) {
1859                WARN_ON_ONCE(bdev->bd_holders);
1860                sync_blockdev(bdev);
1861                kill_bdev(bdev);
1862
1863                bdev_write_inode(bdev);
1864        }
1865        if (bdev->bd_contains == bdev) {
1866                if (disk->fops->release)
1867                        disk->fops->release(disk, mode);
1868        }
1869        if (!bdev->bd_openers) {
1870                disk_put_part(bdev->bd_part);
1871                bdev->bd_part = NULL;
1872                bdev->bd_disk = NULL;
1873                if (bdev != bdev->bd_contains)
1874                        victim = bdev->bd_contains;
1875                bdev->bd_contains = NULL;
1876
1877                put_disk_and_module(disk);
1878        }
1879        mutex_unlock(&bdev->bd_mutex);
1880        bdput(bdev);
1881        if (victim)
1882                __blkdev_put(victim, mode, 1);
1883}
1884
1885void blkdev_put(struct block_device *bdev, fmode_t mode)
1886{
1887        mutex_lock(&bdev->bd_mutex);
1888
1889        if (mode & FMODE_EXCL) {
1890                bool bdev_free;
1891
1892                /*
1893                 * Release a claim on the device.  The holder fields
1894                 * are protected with bdev_lock.  bd_mutex is to
1895                 * synchronize disk_holder unlinking.
1896                 */
1897                spin_lock(&bdev_lock);
1898
1899                WARN_ON_ONCE(--bdev->bd_holders < 0);
1900                WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1901
1902                /* bd_contains might point to self, check in a separate step */
1903                if ((bdev_free = !bdev->bd_holders))
1904                        bdev->bd_holder = NULL;
1905                if (!bdev->bd_contains->bd_holders)
1906                        bdev->bd_contains->bd_holder = NULL;
1907
1908                spin_unlock(&bdev_lock);
1909
1910                /*
1911                 * If this was the last claim, remove holder link and
1912                 * unblock evpoll if it was a write holder.
1913                 */
1914                if (bdev_free && bdev->bd_write_holder) {
1915                        disk_unblock_events(bdev->bd_disk);
1916                        bdev->bd_write_holder = false;
1917                }
1918        }
1919
1920        /*
1921         * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1922         * event.  This is to ensure detection of media removal commanded
1923         * from userland - e.g. eject(1).
1924         */
1925        disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1926
1927        mutex_unlock(&bdev->bd_mutex);
1928
1929        __blkdev_put(bdev, mode, 0);
1930}
1931EXPORT_SYMBOL(blkdev_put);
1932
1933static int blkdev_close(struct inode * inode, struct file * filp)
1934{
1935        struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1936        blkdev_put(bdev, filp->f_mode);
1937        return 0;
1938}
1939
1940static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1941{
1942        struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1943        fmode_t mode = file->f_mode;
1944
1945        /*
1946         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1947         * to updated it before every ioctl.
1948         */
1949        if (file->f_flags & O_NDELAY)
1950                mode |= FMODE_NDELAY;
1951        else
1952                mode &= ~FMODE_NDELAY;
1953
1954        return blkdev_ioctl(bdev, mode, cmd, arg);
1955}
1956
1957/*
1958 * Write data to the block device.  Only intended for the block device itself
1959 * and the raw driver which basically is a fake block device.
1960 *
1961 * Does not take i_mutex for the write and thus is not for general purpose
1962 * use.
1963 */
1964ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1965{
1966        struct file *file = iocb->ki_filp;
1967        struct inode *bd_inode = bdev_file_inode(file);
1968        loff_t size = i_size_read(bd_inode);
1969        struct blk_plug plug;
1970        ssize_t ret;
1971
1972        if (bdev_read_only(I_BDEV(bd_inode)))
1973                return -EPERM;
1974
1975        if (IS_SWAPFILE(bd_inode))
1976                return -ETXTBSY;
1977
1978        if (!iov_iter_count(from))
1979                return 0;
1980
1981        if (iocb->ki_pos >= size)
1982                return -ENOSPC;
1983
1984        if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1985                return -EOPNOTSUPP;
1986
1987        iov_iter_truncate(from, size - iocb->ki_pos);
1988
1989        blk_start_plug(&plug);
1990        ret = __generic_file_write_iter(iocb, from);
1991        if (ret > 0)
1992                ret = generic_write_sync(iocb, ret);
1993        blk_finish_plug(&plug);
1994        return ret;
1995}
1996EXPORT_SYMBOL_GPL(blkdev_write_iter);
1997
1998ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1999{
2000        struct file *file = iocb->ki_filp;
2001        struct inode *bd_inode = bdev_file_inode(file);
2002        loff_t size = i_size_read(bd_inode);
2003        loff_t pos = iocb->ki_pos;
2004
2005        if (pos >= size)
2006                return 0;
2007
2008        size -= pos;
2009        iov_iter_truncate(to, size);
2010        return generic_file_read_iter(iocb, to);
2011}
2012EXPORT_SYMBOL_GPL(blkdev_read_iter);
2013
2014/*
2015 * Try to release a page associated with block device when the system
2016 * is under memory pressure.
2017 */
2018static int blkdev_releasepage(struct page *page, gfp_t wait)
2019{
2020        struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2021
2022        if (super && super->s_op->bdev_try_to_free_page)
2023                return super->s_op->bdev_try_to_free_page(super, page, wait);
2024
2025        return try_to_free_buffers(page);
2026}
2027
2028static int blkdev_writepages(struct address_space *mapping,
2029                             struct writeback_control *wbc)
2030{
2031        return generic_writepages(mapping, wbc);
2032}
2033
2034static const struct address_space_operations def_blk_aops = {
2035        .readpage       = blkdev_readpage,
2036        .readpages      = blkdev_readpages,
2037        .writepage      = blkdev_writepage,
2038        .write_begin    = blkdev_write_begin,
2039        .write_end      = blkdev_write_end,
2040        .writepages     = blkdev_writepages,
2041        .releasepage    = blkdev_releasepage,
2042        .direct_IO      = blkdev_direct_IO,
2043        .migratepage    = buffer_migrate_page_norefs,
2044        .is_dirty_writeback = buffer_check_dirty_writeback,
2045};
2046
2047#define BLKDEV_FALLOC_FL_SUPPORTED                                      \
2048                (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |           \
2049                 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2050
2051static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2052                             loff_t len)
2053{
2054        struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2055        struct address_space *mapping;
2056        loff_t end = start + len - 1;
2057        loff_t isize;
2058        int error;
2059
2060        /* Fail if we don't recognize the flags. */
2061        if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2062                return -EOPNOTSUPP;
2063
2064        /* Don't go off the end of the device. */
2065        isize = i_size_read(bdev->bd_inode);
2066        if (start >= isize)
2067                return -EINVAL;
2068        if (end >= isize) {
2069                if (mode & FALLOC_FL_KEEP_SIZE) {
2070                        len = isize - start;
2071                        end = start + len - 1;
2072                } else
2073                        return -EINVAL;
2074        }
2075
2076        /*
2077         * Don't allow IO that isn't aligned to logical block size.
2078         */
2079        if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2080                return -EINVAL;
2081
2082        /* Invalidate the page cache, including dirty pages. */
2083        mapping = bdev->bd_inode->i_mapping;
2084        truncate_inode_pages_range(mapping, start, end);
2085
2086        switch (mode) {
2087        case FALLOC_FL_ZERO_RANGE:
2088        case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2089                error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2090                                            GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2091                break;
2092        case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2093                error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2094                                             GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2095                break;
2096        case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2097                error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2098                                             GFP_KERNEL, 0);
2099                break;
2100        default:
2101                return -EOPNOTSUPP;
2102        }
2103        if (error)
2104                return error;
2105
2106        /*
2107         * Invalidate again; if someone wandered in and dirtied a page,
2108         * the caller will be given -EBUSY.  The third argument is
2109         * inclusive, so the rounding here is safe.
2110         */
2111        return invalidate_inode_pages2_range(mapping,
2112                                             start >> PAGE_SHIFT,
2113                                             end >> PAGE_SHIFT);
2114}
2115
2116const struct file_operations def_blk_fops = {
2117        .open           = blkdev_open,
2118        .release        = blkdev_close,
2119        .llseek         = block_llseek,
2120        .read_iter      = blkdev_read_iter,
2121        .write_iter     = blkdev_write_iter,
2122        .iopoll         = blkdev_iopoll,
2123        .mmap           = generic_file_mmap,
2124        .fsync          = blkdev_fsync,
2125        .unlocked_ioctl = block_ioctl,
2126#ifdef CONFIG_COMPAT
2127        .compat_ioctl   = compat_blkdev_ioctl,
2128#endif
2129        .splice_read    = generic_file_splice_read,
2130        .splice_write   = iter_file_splice_write,
2131        .fallocate      = blkdev_fallocate,
2132};
2133
2134int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2135{
2136        int res;
2137        mm_segment_t old_fs = get_fs();
2138        set_fs(KERNEL_DS);
2139        res = blkdev_ioctl(bdev, 0, cmd, arg);
2140        set_fs(old_fs);
2141        return res;
2142}
2143
2144EXPORT_SYMBOL(ioctl_by_bdev);
2145
2146/**
2147 * lookup_bdev  - lookup a struct block_device by name
2148 * @pathname:   special file representing the block device
2149 *
2150 * Get a reference to the blockdevice at @pathname in the current
2151 * namespace if possible and return it.  Return ERR_PTR(error)
2152 * otherwise.
2153 */
2154struct block_device *lookup_bdev(const char *pathname)
2155{
2156        struct block_device *bdev;
2157        struct inode *inode;
2158        struct path path;
2159        int error;
2160
2161        if (!pathname || !*pathname)
2162                return ERR_PTR(-EINVAL);
2163
2164        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2165        if (error)
2166                return ERR_PTR(error);
2167
2168        inode = d_backing_inode(path.dentry);
2169        error = -ENOTBLK;
2170        if (!S_ISBLK(inode->i_mode))
2171                goto fail;
2172        error = -EACCES;
2173        if (!may_open_dev(&path))
2174                goto fail;
2175        error = -ENOMEM;
2176        bdev = bd_acquire(inode);
2177        if (!bdev)
2178                goto fail;
2179out:
2180        path_put(&path);
2181        return bdev;
2182fail:
2183        bdev = ERR_PTR(error);
2184        goto out;
2185}
2186EXPORT_SYMBOL(lookup_bdev);
2187
2188int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2189{
2190        struct super_block *sb = get_super(bdev);
2191        int res = 0;
2192
2193        if (sb) {
2194                /*
2195                 * no need to lock the super, get_super holds the
2196                 * read mutex so the filesystem cannot go away
2197                 * under us (->put_super runs with the write lock
2198                 * hold).
2199                 */
2200                shrink_dcache_sb(sb);
2201                res = invalidate_inodes(sb, kill_dirty);
2202                drop_super(sb);
2203        }
2204        invalidate_bdev(bdev);
2205        return res;
2206}
2207EXPORT_SYMBOL(__invalidate_device);
2208
2209void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2210{
2211        struct inode *inode, *old_inode = NULL;
2212
2213        spin_lock(&blockdev_superblock->s_inode_list_lock);
2214        list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2215                struct address_space *mapping = inode->i_mapping;
2216                struct block_device *bdev;
2217
2218                spin_lock(&inode->i_lock);
2219                if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2220                    mapping->nrpages == 0) {
2221                        spin_unlock(&inode->i_lock);
2222                        continue;
2223                }
2224                __iget(inode);
2225                spin_unlock(&inode->i_lock);
2226                spin_unlock(&blockdev_superblock->s_inode_list_lock);
2227                /*
2228                 * We hold a reference to 'inode' so it couldn't have been
2229                 * removed from s_inodes list while we dropped the
2230                 * s_inode_list_lock  We cannot iput the inode now as we can
2231                 * be holding the last reference and we cannot iput it under
2232                 * s_inode_list_lock. So we keep the reference and iput it
2233                 * later.
2234                 */
2235                iput(old_inode);
2236                old_inode = inode;
2237                bdev = I_BDEV(inode);
2238
2239                mutex_lock(&bdev->bd_mutex);
2240                if (bdev->bd_openers)
2241                        func(bdev, arg);
2242                mutex_unlock(&bdev->bd_mutex);
2243
2244                spin_lock(&blockdev_superblock->s_inode_list_lock);
2245        }
2246        spin_unlock(&blockdev_superblock->s_inode_list_lock);
2247        iput(old_inode);
2248}
2249