fs/block_dev.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
BDEV_I
I_BDEV
bdev_write_inode
kill_bdev
invalidate_bdev
set_init_blocksize
set_blocksize
sb_set_blocksize
sb_min_blocksize
blkdev_get_block
bdev_file_inode
dio_bio_write_op
blkdev_bio_end_io_simple
__blkdev_direct_IO_simple
blkdev_iopoll
blkdev_bio_end_io
__blkdev_direct_IO
blkdev_direct_IO
blkdev_init
__sync_blockdev
sync_blockdev
fsync_bdev
freeze_bdev
thaw_bdev
blkdev_writepage
blkdev_readpage
blkdev_readpages
blkdev_write_begin
blkdev_write_end
block_llseek
blkdev_fsync
bdev_read_page
bdev_write_page
bdev_alloc_inode
bdev_free_inode
init_once
bdev_evict_inode
bd_init_fs_context
bdev_cache_init
hash
bdev_test
bdev_set
bdev_unhash_inode
bdget
bdgrab
nr_blockdev_pages
bdput
bd_acquire
bd_forget
bd_may_claim
bd_prepare_to_claim
bdev_get_gendisk
bd_start_claiming
bd_clear_claiming
bd_finish_claiming
bd_abort_claiming
bd_find_holder_disk
add_symlink
del_symlink
bd_link_disk_holder
bd_unlink_disk_holder
flush_disk
check_disk_size_change
revalidate_disk
check_disk_change
bd_set_size
bdev_disk_changed
__blkdev_get
blkdev_get
blkdev_get_by_path
blkdev_get_by_dev
blkdev_open
__blkdev_put
blkdev_put
blkdev_close
block_ioctl
blkdev_write_iter
blkdev_read_iter
blkdev_releasepage
blkdev_writepages
blkdev_fallocate
ioctl_by_bdev
lookup_bdev
__invalidate_device
iterate_bdevs
   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 <linux/suspend.h>
  38 #include "internal.h"
  39 
  40 struct bdev_inode {
  41         struct block_device bdev;
  42         struct inode vfs_inode;
  43 };
  44 
  45 static const struct address_space_operations def_blk_aops;
  46 
  47 static inline struct bdev_inode *BDEV_I(struct inode *inode)
  48 {
  49         return container_of(inode, struct bdev_inode, vfs_inode);
  50 }
  51 
  52 struct block_device *I_BDEV(struct inode *inode)
  53 {
  54         return &BDEV_I(inode)->bdev;
  55 }
  56 EXPORT_SYMBOL(I_BDEV);
  57 
  58 static void bdev_write_inode(struct block_device *bdev)
  59 {
  60         struct inode *inode = bdev->bd_inode;
  61         int ret;
  62 
  63         spin_lock(&inode->i_lock);
  64         while (inode->i_state & I_DIRTY) {
  65                 spin_unlock(&inode->i_lock);
  66                 ret = write_inode_now(inode, true);
  67                 if (ret) {
  68                         char name[BDEVNAME_SIZE];
  69                         pr_warn_ratelimited("VFS: Dirty inode writeback failed "
  70                                             "for block device %s (err=%d).\n",
  71                                             bdevname(bdev, name), ret);
  72                 }
  73                 spin_lock(&inode->i_lock);
  74         }
  75         spin_unlock(&inode->i_lock);
  76 }
  77 
  78 /* Kill _all_ buffers and pagecache , dirty or not.. */
  79 void kill_bdev(struct block_device *bdev)
  80 {
  81         struct address_space *mapping = bdev->bd_inode->i_mapping;
  82 
  83         if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
  84                 return;
  85 
  86         invalidate_bh_lrus();
  87         truncate_inode_pages(mapping, 0);
  88 }       
  89 EXPORT_SYMBOL(kill_bdev);
  90 
  91 /* Invalidate clean unused buffers and pagecache. */
  92 void invalidate_bdev(struct block_device *bdev)
  93 {
  94         struct address_space *mapping = bdev->bd_inode->i_mapping;
  95 
  96         if (mapping->nrpages) {
  97                 invalidate_bh_lrus();
  98                 lru_add_drain_all();    /* make sure all lru add caches are flushed */
  99                 invalidate_mapping_pages(mapping, 0, -1);
 100         }
 101         /* 99% of the time, we don't need to flush the cleancache on the bdev.
 102          * But, for the strange corners, lets be cautious
 103          */
 104         cleancache_invalidate_inode(mapping);
 105 }
 106 EXPORT_SYMBOL(invalidate_bdev);
 107 
 108 static void set_init_blocksize(struct block_device *bdev)
 109 {
 110         unsigned bsize = bdev_logical_block_size(bdev);
 111         loff_t size = i_size_read(bdev->bd_inode);
 112 
 113         while (bsize < PAGE_SIZE) {
 114                 if (size & bsize)
 115                         break;
 116                 bsize <<= 1;
 117         }
 118         bdev->bd_block_size = bsize;
 119         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 120 }
 121 
 122 int set_blocksize(struct block_device *bdev, int size)
 123 {
 124         /* Size must be a power of two, and between 512 and PAGE_SIZE */
 125         if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
 126                 return -EINVAL;
 127 
 128         /* Size cannot be smaller than the size supported by the device */
 129         if (size < bdev_logical_block_size(bdev))
 130                 return -EINVAL;
 131 
 132         /* Don't change the size if it is same as current */
 133         if (bdev->bd_block_size != size) {
 134                 sync_blockdev(bdev);
 135                 bdev->bd_block_size = size;
 136                 bdev->bd_inode->i_blkbits = blksize_bits(size);
 137                 kill_bdev(bdev);
 138         }
 139         return 0;
 140 }
 141 
 142 EXPORT_SYMBOL(set_blocksize);
 143 
 144 int sb_set_blocksize(struct super_block *sb, int size)
 145 {
 146         if (set_blocksize(sb->s_bdev, size))
 147                 return 0;
 148         /* If we get here, we know size is power of two
 149          * and it's value is between 512 and PAGE_SIZE */
 150         sb->s_blocksize = size;
 151         sb->s_blocksize_bits = blksize_bits(size);
 152         return sb->s_blocksize;
 153 }
 154 
 155 EXPORT_SYMBOL(sb_set_blocksize);
 156 
 157 int sb_min_blocksize(struct super_block *sb, int size)
 158 {
 159         int minsize = bdev_logical_block_size(sb->s_bdev);
 160         if (size < minsize)
 161                 size = minsize;
 162         return sb_set_blocksize(sb, size);
 163 }
 164 
 165 EXPORT_SYMBOL(sb_min_blocksize);
 166 
 167 static int
 168 blkdev_get_block(struct inode *inode, sector_t iblock,
 169                 struct buffer_head *bh, int create)
 170 {
 171         bh->b_bdev = I_BDEV(inode);
 172         bh->b_blocknr = iblock;
 173         set_buffer_mapped(bh);
 174         return 0;
 175 }
 176 
 177 static struct inode *bdev_file_inode(struct file *file)
 178 {
 179         return file->f_mapping->host;
 180 }
 181 
 182 static unsigned int dio_bio_write_op(struct kiocb *iocb)
 183 {
 184         unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
 185 
 186         /* avoid the need for a I/O completion work item */
 187         if (iocb->ki_flags & IOCB_DSYNC)
 188                 op |= REQ_FUA;
 189         return op;
 190 }
 191 
 192 #define DIO_INLINE_BIO_VECS 4
 193 
 194 static void blkdev_bio_end_io_simple(struct bio *bio)
 195 {
 196         struct task_struct *waiter = bio->bi_private;
 197 
 198         WRITE_ONCE(bio->bi_private, NULL);
 199         blk_wake_io_task(waiter);
 200 }
 201 
 202 static ssize_t
 203 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
 204                 int nr_pages)
 205 {
 206         struct file *file = iocb->ki_filp;
 207         struct block_device *bdev = I_BDEV(bdev_file_inode(file));
 208         struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
 209         loff_t pos = iocb->ki_pos;
 210         bool should_dirty = false;
 211         struct bio bio;
 212         ssize_t ret;
 213         blk_qc_t qc;
 214 
 215         if ((pos | iov_iter_alignment(iter)) &
 216             (bdev_logical_block_size(bdev) - 1))
 217                 return -EINVAL;
 218 
 219         if (nr_pages <= DIO_INLINE_BIO_VECS)
 220                 vecs = inline_vecs;
 221         else {
 222                 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
 223                                      GFP_KERNEL);
 224                 if (!vecs)
 225                         return -ENOMEM;
 226         }
 227 
 228         bio_init(&bio, vecs, nr_pages);
 229         bio_set_dev(&bio, bdev);
 230         bio.bi_iter.bi_sector = pos >> 9;
 231         bio.bi_write_hint = iocb->ki_hint;
 232         bio.bi_private = current;
 233         bio.bi_end_io = blkdev_bio_end_io_simple;
 234         bio.bi_ioprio = iocb->ki_ioprio;
 235 
 236         ret = bio_iov_iter_get_pages(&bio, iter);
 237         if (unlikely(ret))
 238                 goto out;
 239         ret = bio.bi_iter.bi_size;
 240 
 241         if (iov_iter_rw(iter) == READ) {
 242                 bio.bi_opf = REQ_OP_READ;
 243                 if (iter_is_iovec(iter))
 244                         should_dirty = true;
 245         } else {
 246                 bio.bi_opf = dio_bio_write_op(iocb);
 247                 task_io_account_write(ret);
 248         }
 249         if (iocb->ki_flags & IOCB_HIPRI)
 250                 bio_set_polled(&bio, iocb);
 251 
 252         qc = submit_bio(&bio);
 253         for (;;) {
 254                 set_current_state(TASK_UNINTERRUPTIBLE);
 255                 if (!READ_ONCE(bio.bi_private))
 256                         break;
 257                 if (!(iocb->ki_flags & IOCB_HIPRI) ||
 258                     !blk_poll(bdev_get_queue(bdev), qc, true))
 259                         io_schedule();
 260         }
 261         __set_current_state(TASK_RUNNING);
 262 
 263         bio_release_pages(&bio, should_dirty);
 264         if (unlikely(bio.bi_status))
 265                 ret = blk_status_to_errno(bio.bi_status);
 266 
 267 out:
 268         if (vecs != inline_vecs)
 269                 kfree(vecs);
 270 
 271         bio_uninit(&bio);
 272 
 273         return ret;
 274 }
 275 
 276 struct blkdev_dio {
 277         union {
 278                 struct kiocb            *iocb;
 279                 struct task_struct      *waiter;
 280         };
 281         size_t                  size;
 282         atomic_t                ref;
 283         bool                    multi_bio : 1;
 284         bool                    should_dirty : 1;
 285         bool                    is_sync : 1;
 286         struct bio              bio;
 287 };
 288 
 289 static struct bio_set blkdev_dio_pool;
 290 
 291 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
 292 {
 293         struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
 294         struct request_queue *q = bdev_get_queue(bdev);
 295 
 296         return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
 297 }
 298 
 299 static void blkdev_bio_end_io(struct bio *bio)
 300 {
 301         struct blkdev_dio *dio = bio->bi_private;
 302         bool should_dirty = dio->should_dirty;
 303 
 304         if (bio->bi_status && !dio->bio.bi_status)
 305                 dio->bio.bi_status = bio->bi_status;
 306 
 307         if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
 308                 if (!dio->is_sync) {
 309                         struct kiocb *iocb = dio->iocb;
 310                         ssize_t ret;
 311 
 312                         if (likely(!dio->bio.bi_status)) {
 313                                 ret = dio->size;
 314                                 iocb->ki_pos += ret;
 315                         } else {
 316                                 ret = blk_status_to_errno(dio->bio.bi_status);
 317                         }
 318 
 319                         dio->iocb->ki_complete(iocb, ret, 0);
 320                         if (dio->multi_bio)
 321                                 bio_put(&dio->bio);
 322                 } else {
 323                         struct task_struct *waiter = dio->waiter;
 324 
 325                         WRITE_ONCE(dio->waiter, NULL);
 326                         blk_wake_io_task(waiter);
 327                 }
 328         }
 329 
 330         if (should_dirty) {
 331                 bio_check_pages_dirty(bio);
 332         } else {
 333                 bio_release_pages(bio, false);
 334                 bio_put(bio);
 335         }
 336 }
 337 
 338 static ssize_t
 339 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
 340 {
 341         struct file *file = iocb->ki_filp;
 342         struct inode *inode = bdev_file_inode(file);
 343         struct block_device *bdev = I_BDEV(inode);
 344         struct blk_plug plug;
 345         struct blkdev_dio *dio;
 346         struct bio *bio;
 347         bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
 348         bool is_read = (iov_iter_rw(iter) == READ), is_sync;
 349         loff_t pos = iocb->ki_pos;
 350         blk_qc_t qc = BLK_QC_T_NONE;
 351         int ret = 0;
 352 
 353         if ((pos | iov_iter_alignment(iter)) &
 354             (bdev_logical_block_size(bdev) - 1))
 355                 return -EINVAL;
 356 
 357         bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
 358 
 359         dio = container_of(bio, struct blkdev_dio, bio);
 360         dio->is_sync = is_sync = is_sync_kiocb(iocb);
 361         if (dio->is_sync) {
 362                 dio->waiter = current;
 363                 bio_get(bio);
 364         } else {
 365                 dio->iocb = iocb;
 366         }
 367 
 368         dio->size = 0;
 369         dio->multi_bio = false;
 370         dio->should_dirty = is_read && iter_is_iovec(iter);
 371 
 372         /*
 373          * Don't plug for HIPRI/polled IO, as those should go straight
 374          * to issue
 375          */
 376         if (!is_poll)
 377                 blk_start_plug(&plug);
 378 
 379         for (;;) {
 380                 bio_set_dev(bio, bdev);
 381                 bio->bi_iter.bi_sector = pos >> 9;
 382                 bio->bi_write_hint = iocb->ki_hint;
 383                 bio->bi_private = dio;
 384                 bio->bi_end_io = blkdev_bio_end_io;
 385                 bio->bi_ioprio = iocb->ki_ioprio;
 386 
 387                 ret = bio_iov_iter_get_pages(bio, iter);
 388                 if (unlikely(ret)) {
 389                         bio->bi_status = BLK_STS_IOERR;
 390                         bio_endio(bio);
 391                         break;
 392                 }
 393 
 394                 if (is_read) {
 395                         bio->bi_opf = REQ_OP_READ;
 396                         if (dio->should_dirty)
 397                                 bio_set_pages_dirty(bio);
 398                 } else {
 399                         bio->bi_opf = dio_bio_write_op(iocb);
 400                         task_io_account_write(bio->bi_iter.bi_size);
 401                 }
 402 
 403                 dio->size += bio->bi_iter.bi_size;
 404                 pos += bio->bi_iter.bi_size;
 405 
 406                 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
 407                 if (!nr_pages) {
 408                         bool polled = false;
 409 
 410                         if (iocb->ki_flags & IOCB_HIPRI) {
 411                                 bio_set_polled(bio, iocb);
 412                                 polled = true;
 413                         }
 414 
 415                         qc = submit_bio(bio);
 416 
 417                         if (polled)
 418                                 WRITE_ONCE(iocb->ki_cookie, qc);
 419                         break;
 420                 }
 421 
 422                 if (!dio->multi_bio) {
 423                         /*
 424                          * AIO needs an extra reference to ensure the dio
 425                          * structure which is embedded into the first bio
 426                          * stays around.
 427                          */
 428                         if (!is_sync)
 429                                 bio_get(bio);
 430                         dio->multi_bio = true;
 431                         atomic_set(&dio->ref, 2);
 432                 } else {
 433                         atomic_inc(&dio->ref);
 434                 }
 435 
 436                 submit_bio(bio);
 437                 bio = bio_alloc(GFP_KERNEL, nr_pages);
 438         }
 439 
 440         if (!is_poll)
 441                 blk_finish_plug(&plug);
 442 
 443         if (!is_sync)
 444                 return -EIOCBQUEUED;
 445 
 446         for (;;) {
 447                 set_current_state(TASK_UNINTERRUPTIBLE);
 448                 if (!READ_ONCE(dio->waiter))
 449                         break;
 450 
 451                 if (!(iocb->ki_flags & IOCB_HIPRI) ||
 452                     !blk_poll(bdev_get_queue(bdev), qc, true))
 453                         io_schedule();
 454         }
 455         __set_current_state(TASK_RUNNING);
 456 
 457         if (!ret)
 458                 ret = blk_status_to_errno(dio->bio.bi_status);
 459         if (likely(!ret))
 460                 ret = dio->size;
 461 
 462         bio_put(&dio->bio);
 463         return ret;
 464 }
 465 
 466 static ssize_t
 467 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 468 {
 469         int nr_pages;
 470 
 471         nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
 472         if (!nr_pages)
 473                 return 0;
 474         if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
 475                 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
 476 
 477         return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
 478 }
 479 
 480 static __init int blkdev_init(void)
 481 {
 482         return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
 483 }
 484 module_init(blkdev_init);
 485 
 486 int __sync_blockdev(struct block_device *bdev, int wait)
 487 {
 488         if (!bdev)
 489                 return 0;
 490         if (!wait)
 491                 return filemap_flush(bdev->bd_inode->i_mapping);
 492         return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 493 }
 494 
 495 /*
 496  * Write out and wait upon all the dirty data associated with a block
 497  * device via its mapping.  Does not take the superblock lock.
 498  */
 499 int sync_blockdev(struct block_device *bdev)
 500 {
 501         return __sync_blockdev(bdev, 1);
 502 }
 503 EXPORT_SYMBOL(sync_blockdev);
 504 
 505 /*
 506  * Write out and wait upon all dirty data associated with this
 507  * device.   Filesystem data as well as the underlying block
 508  * device.  Takes the superblock lock.
 509  */
 510 int fsync_bdev(struct block_device *bdev)
 511 {
 512         struct super_block *sb = get_super(bdev);
 513         if (sb) {
 514                 int res = sync_filesystem(sb);
 515                 drop_super(sb);
 516                 return res;
 517         }
 518         return sync_blockdev(bdev);
 519 }
 520 EXPORT_SYMBOL(fsync_bdev);
 521 
 522 /**
 523  * freeze_bdev  --  lock a filesystem and force it into a consistent state
 524  * @bdev:       blockdevice to lock
 525  *
 526  * If a superblock is found on this device, we take the s_umount semaphore
 527  * on it to make sure nobody unmounts until the snapshot creation is done.
 528  * The reference counter (bd_fsfreeze_count) guarantees that only the last
 529  * unfreeze process can unfreeze the frozen filesystem actually when multiple
 530  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 531  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 532  * actually.
 533  */
 534 struct super_block *freeze_bdev(struct block_device *bdev)
 535 {
 536         struct super_block *sb;
 537         int error = 0;
 538 
 539         mutex_lock(&bdev->bd_fsfreeze_mutex);
 540         if (++bdev->bd_fsfreeze_count > 1) {
 541                 /*
 542                  * We don't even need to grab a reference - the first call
 543                  * to freeze_bdev grab an active reference and only the last
 544                  * thaw_bdev drops it.
 545                  */
 546                 sb = get_super(bdev);
 547                 if (sb)
 548                         drop_super(sb);
 549                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
 550                 return sb;
 551         }
 552 
 553         sb = get_active_super(bdev);
 554         if (!sb)
 555                 goto out;
 556         if (sb->s_op->freeze_super)
 557                 error = sb->s_op->freeze_super(sb);
 558         else
 559                 error = freeze_super(sb);
 560         if (error) {
 561                 deactivate_super(sb);
 562                 bdev->bd_fsfreeze_count--;
 563                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
 564                 return ERR_PTR(error);
 565         }
 566         deactivate_super(sb);
 567  out:
 568         sync_blockdev(bdev);
 569         mutex_unlock(&bdev->bd_fsfreeze_mutex);
 570         return sb;      /* thaw_bdev releases s->s_umount */
 571 }
 572 EXPORT_SYMBOL(freeze_bdev);
 573 
 574 /**
 575  * thaw_bdev  -- unlock filesystem
 576  * @bdev:       blockdevice to unlock
 577  * @sb:         associated superblock
 578  *
 579  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 580  */
 581 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 582 {
 583         int error = -EINVAL;
 584 
 585         mutex_lock(&bdev->bd_fsfreeze_mutex);
 586         if (!bdev->bd_fsfreeze_count)
 587                 goto out;
 588 
 589         error = 0;
 590         if (--bdev->bd_fsfreeze_count > 0)
 591                 goto out;
 592 
 593         if (!sb)
 594                 goto out;
 595 
 596         if (sb->s_op->thaw_super)
 597                 error = sb->s_op->thaw_super(sb);
 598         else
 599                 error = thaw_super(sb);
 600         if (error)
 601                 bdev->bd_fsfreeze_count++;
 602 out:
 603         mutex_unlock(&bdev->bd_fsfreeze_mutex);
 604         return error;
 605 }
 606 EXPORT_SYMBOL(thaw_bdev);
 607 
 608 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 609 {
 610         return block_write_full_page(page, blkdev_get_block, wbc);
 611 }
 612 
 613 static int blkdev_readpage(struct file * file, struct page * page)
 614 {
 615         return block_read_full_page(page, blkdev_get_block);
 616 }
 617 
 618 static int blkdev_readpages(struct file *file, struct address_space *mapping,
 619                         struct list_head *pages, unsigned nr_pages)
 620 {
 621         return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
 622 }
 623 
 624 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 625                         loff_t pos, unsigned len, unsigned flags,
 626                         struct page **pagep, void **fsdata)
 627 {
 628         return block_write_begin(mapping, pos, len, flags, pagep,
 629                                  blkdev_get_block);
 630 }
 631 
 632 static int blkdev_write_end(struct file *file, struct address_space *mapping,
 633                         loff_t pos, unsigned len, unsigned copied,
 634                         struct page *page, void *fsdata)
 635 {
 636         int ret;
 637         ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 638 
 639         unlock_page(page);
 640         put_page(page);
 641 
 642         return ret;
 643 }
 644 
 645 /*
 646  * private llseek:
 647  * for a block special file file_inode(file)->i_size is zero
 648  * so we compute the size by hand (just as in block_read/write above)
 649  */
 650 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
 651 {
 652         struct inode *bd_inode = bdev_file_inode(file);
 653         loff_t retval;
 654 
 655         inode_lock(bd_inode);
 656         retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
 657         inode_unlock(bd_inode);
 658         return retval;
 659 }
 660         
 661 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 662 {
 663         struct inode *bd_inode = bdev_file_inode(filp);
 664         struct block_device *bdev = I_BDEV(bd_inode);
 665         int error;
 666         
 667         error = file_write_and_wait_range(filp, start, end);
 668         if (error)
 669                 return error;
 670 
 671         /*
 672          * There is no need to serialise calls to blkdev_issue_flush with
 673          * i_mutex and doing so causes performance issues with concurrent
 674          * O_SYNC writers to a block device.
 675          */
 676         error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
 677         if (error == -EOPNOTSUPP)
 678                 error = 0;
 679 
 680         return error;
 681 }
 682 EXPORT_SYMBOL(blkdev_fsync);
 683 
 684 /**
 685  * bdev_read_page() - Start reading a page from a block device
 686  * @bdev: The device to read the page from
 687  * @sector: The offset on the device to read the page to (need not be aligned)
 688  * @page: The page to read
 689  *
 690  * On entry, the page should be locked.  It will be unlocked when the page
 691  * has been read.  If the block driver implements rw_page synchronously,
 692  * that will be true on exit from this function, but it need not be.
 693  *
 694  * Errors returned by this function are usually "soft", eg out of memory, or
 695  * queue full; callers should try a different route to read this page rather
 696  * than propagate an error back up the stack.
 697  *
 698  * Return: negative errno if an error occurs, 0 if submission was successful.
 699  */
 700 int bdev_read_page(struct block_device *bdev, sector_t sector,
 701                         struct page *page)
 702 {
 703         const struct block_device_operations *ops = bdev->bd_disk->fops;
 704         int result = -EOPNOTSUPP;
 705 
 706         if (!ops->rw_page || bdev_get_integrity(bdev))
 707                 return result;
 708 
 709         result = blk_queue_enter(bdev->bd_queue, 0);
 710         if (result)
 711                 return result;
 712         result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
 713                               REQ_OP_READ);
 714         blk_queue_exit(bdev->bd_queue);
 715         return result;
 716 }
 717 EXPORT_SYMBOL_GPL(bdev_read_page);
 718 
 719 /**
 720  * bdev_write_page() - Start writing a page to a block device
 721  * @bdev: The device to write the page to
 722  * @sector: The offset on the device to write the page to (need not be aligned)
 723  * @page: The page to write
 724  * @wbc: The writeback_control for the write
 725  *
 726  * On entry, the page should be locked and not currently under writeback.
 727  * On exit, if the write started successfully, the page will be unlocked and
 728  * under writeback.  If the write failed already (eg the driver failed to
 729  * queue the page to the device), the page will still be locked.  If the
 730  * caller is a ->writepage implementation, it will need to unlock the page.
 731  *
 732  * Errors returned by this function are usually "soft", eg out of memory, or
 733  * queue full; callers should try a different route to write this page rather
 734  * than propagate an error back up the stack.
 735  *
 736  * Return: negative errno if an error occurs, 0 if submission was successful.
 737  */
 738 int bdev_write_page(struct block_device *bdev, sector_t sector,
 739                         struct page *page, struct writeback_control *wbc)
 740 {
 741         int result;
 742         const struct block_device_operations *ops = bdev->bd_disk->fops;
 743 
 744         if (!ops->rw_page || bdev_get_integrity(bdev))
 745                 return -EOPNOTSUPP;
 746         result = blk_queue_enter(bdev->bd_queue, 0);
 747         if (result)
 748                 return result;
 749 
 750         set_page_writeback(page);
 751         result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
 752                               REQ_OP_WRITE);
 753         if (result) {
 754                 end_page_writeback(page);
 755         } else {
 756                 clean_page_buffers(page);
 757                 unlock_page(page);
 758         }
 759         blk_queue_exit(bdev->bd_queue);
 760         return result;
 761 }
 762 EXPORT_SYMBOL_GPL(bdev_write_page);
 763 
 764 /*
 765  * pseudo-fs
 766  */
 767 
 768 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 769 static struct kmem_cache * bdev_cachep __read_mostly;
 770 
 771 static struct inode *bdev_alloc_inode(struct super_block *sb)
 772 {
 773         struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 774         if (!ei)
 775                 return NULL;
 776         return &ei->vfs_inode;
 777 }
 778 
 779 static void bdev_free_inode(struct inode *inode)
 780 {
 781         kmem_cache_free(bdev_cachep, BDEV_I(inode));
 782 }
 783 
 784 static void init_once(void *foo)
 785 {
 786         struct bdev_inode *ei = (struct bdev_inode *) foo;
 787         struct block_device *bdev = &ei->bdev;
 788 
 789         memset(bdev, 0, sizeof(*bdev));
 790         mutex_init(&bdev->bd_mutex);
 791         INIT_LIST_HEAD(&bdev->bd_list);
 792 #ifdef CONFIG_SYSFS
 793         INIT_LIST_HEAD(&bdev->bd_holder_disks);
 794 #endif
 795         bdev->bd_bdi = &noop_backing_dev_info;
 796         inode_init_once(&ei->vfs_inode);
 797         /* Initialize mutex for freeze. */
 798         mutex_init(&bdev->bd_fsfreeze_mutex);
 799 }
 800 
 801 static void bdev_evict_inode(struct inode *inode)
 802 {
 803         struct block_device *bdev = &BDEV_I(inode)->bdev;
 804         truncate_inode_pages_final(&inode->i_data);
 805         invalidate_inode_buffers(inode); /* is it needed here? */
 806         clear_inode(inode);
 807         spin_lock(&bdev_lock);
 808         list_del_init(&bdev->bd_list);
 809         spin_unlock(&bdev_lock);
 810         /* Detach inode from wb early as bdi_put() may free bdi->wb */
 811         inode_detach_wb(inode);
 812         if (bdev->bd_bdi != &noop_backing_dev_info) {
 813                 bdi_put(bdev->bd_bdi);
 814                 bdev->bd_bdi = &noop_backing_dev_info;
 815         }
 816 }
 817 
 818 static const struct super_operations bdev_sops = {
 819         .statfs = simple_statfs,
 820         .alloc_inode = bdev_alloc_inode,
 821         .free_inode = bdev_free_inode,
 822         .drop_inode = generic_delete_inode,
 823         .evict_inode = bdev_evict_inode,
 824 };
 825 
 826 static int bd_init_fs_context(struct fs_context *fc)
 827 {
 828         struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
 829         if (!ctx)
 830                 return -ENOMEM;
 831         fc->s_iflags |= SB_I_CGROUPWB;
 832         ctx->ops = &bdev_sops;
 833         return 0;
 834 }
 835 
 836 static struct file_system_type bd_type = {
 837         .name           = "bdev",
 838         .init_fs_context = bd_init_fs_context,
 839         .kill_sb        = kill_anon_super,
 840 };
 841 
 842 struct super_block *blockdev_superblock __read_mostly;
 843 EXPORT_SYMBOL_GPL(blockdev_superblock);
 844 
 845 void __init bdev_cache_init(void)
 846 {
 847         int err;
 848         static struct vfsmount *bd_mnt;
 849 
 850         bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 851                         0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 852                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
 853                         init_once);
 854         err = register_filesystem(&bd_type);
 855         if (err)
 856                 panic("Cannot register bdev pseudo-fs");
 857         bd_mnt = kern_mount(&bd_type);
 858         if (IS_ERR(bd_mnt))
 859                 panic("Cannot create bdev pseudo-fs");
 860         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 861 }
 862 
 863 /*
 864  * Most likely _very_ bad one - but then it's hardly critical for small
 865  * /dev and can be fixed when somebody will need really large one.
 866  * Keep in mind that it will be fed through icache hash function too.
 867  */
 868 static inline unsigned long hash(dev_t dev)
 869 {
 870         return MAJOR(dev)+MINOR(dev);
 871 }
 872 
 873 static int bdev_test(struct inode *inode, void *data)
 874 {
 875         return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 876 }
 877 
 878 static int bdev_set(struct inode *inode, void *data)
 879 {
 880         BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 881         return 0;
 882 }
 883 
 884 static LIST_HEAD(all_bdevs);
 885 
 886 /*
 887  * If there is a bdev inode for this device, unhash it so that it gets evicted
 888  * as soon as last inode reference is dropped.
 889  */
 890 void bdev_unhash_inode(dev_t dev)
 891 {
 892         struct inode *inode;
 893 
 894         inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
 895         if (inode) {
 896                 remove_inode_hash(inode);
 897                 iput(inode);
 898         }
 899 }
 900 
 901 struct block_device *bdget(dev_t dev)
 902 {
 903         struct block_device *bdev;
 904         struct inode *inode;
 905 
 906         inode = iget5_locked(blockdev_superblock, hash(dev),
 907                         bdev_test, bdev_set, &dev);
 908 
 909         if (!inode)
 910                 return NULL;
 911 
 912         bdev = &BDEV_I(inode)->bdev;
 913 
 914         if (inode->i_state & I_NEW) {
 915                 bdev->bd_contains = NULL;
 916                 bdev->bd_super = NULL;
 917                 bdev->bd_inode = inode;
 918                 bdev->bd_block_size = i_blocksize(inode);
 919                 bdev->bd_part_count = 0;
 920                 bdev->bd_invalidated = 0;
 921                 inode->i_mode = S_IFBLK;
 922                 inode->i_rdev = dev;
 923                 inode->i_bdev = bdev;
 924                 inode->i_data.a_ops = &def_blk_aops;
 925                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 926                 spin_lock(&bdev_lock);
 927                 list_add(&bdev->bd_list, &all_bdevs);
 928                 spin_unlock(&bdev_lock);
 929                 unlock_new_inode(inode);
 930         }
 931         return bdev;
 932 }
 933 
 934 EXPORT_SYMBOL(bdget);
 935 
 936 /**
 937  * bdgrab -- Grab a reference to an already referenced block device
 938  * @bdev:       Block device to grab a reference to.
 939  */
 940 struct block_device *bdgrab(struct block_device *bdev)
 941 {
 942         ihold(bdev->bd_inode);
 943         return bdev;
 944 }
 945 EXPORT_SYMBOL(bdgrab);
 946 
 947 long nr_blockdev_pages(void)
 948 {
 949         struct block_device *bdev;
 950         long ret = 0;
 951         spin_lock(&bdev_lock);
 952         list_for_each_entry(bdev, &all_bdevs, bd_list) {
 953                 ret += bdev->bd_inode->i_mapping->nrpages;
 954         }
 955         spin_unlock(&bdev_lock);
 956         return ret;
 957 }
 958 
 959 void bdput(struct block_device *bdev)
 960 {
 961         iput(bdev->bd_inode);
 962 }
 963 
 964 EXPORT_SYMBOL(bdput);
 965  
 966 static struct block_device *bd_acquire(struct inode *inode)
 967 {
 968         struct block_device *bdev;
 969 
 970         spin_lock(&bdev_lock);
 971         bdev = inode->i_bdev;
 972         if (bdev && !inode_unhashed(bdev->bd_inode)) {
 973                 bdgrab(bdev);
 974                 spin_unlock(&bdev_lock);
 975                 return bdev;
 976         }
 977         spin_unlock(&bdev_lock);
 978 
 979         /*
 980          * i_bdev references block device inode that was already shut down
 981          * (corresponding device got removed).  Remove the reference and look
 982          * up block device inode again just in case new device got
 983          * reestablished under the same device number.
 984          */
 985         if (bdev)
 986                 bd_forget(inode);
 987 
 988         bdev = bdget(inode->i_rdev);
 989         if (bdev) {
 990                 spin_lock(&bdev_lock);
 991                 if (!inode->i_bdev) {
 992                         /*
 993                          * We take an additional reference to bd_inode,
 994                          * and it's released in clear_inode() of inode.
 995                          * So, we can access it via ->i_mapping always
 996                          * without igrab().
 997                          */
 998                         bdgrab(bdev);
 999                         inode->i_bdev = bdev;
1000                         inode->i_mapping = bdev->bd_inode->i_mapping;
1001                 }
1002                 spin_unlock(&bdev_lock);
1003         }
1004         return bdev;
1005 }
1006 
1007 /* Call when you free inode */
1008 
1009 void bd_forget(struct inode *inode)
1010 {
1011         struct block_device *bdev = NULL;
1012 
1013         spin_lock(&bdev_lock);
1014         if (!sb_is_blkdev_sb(inode->i_sb))
1015                 bdev = inode->i_bdev;
1016         inode->i_bdev = NULL;
1017         inode->i_mapping = &inode->i_data;
1018         spin_unlock(&bdev_lock);
1019 
1020         if (bdev)
1021                 bdput(bdev);
1022 }
1023 
1024 /**
1025  * bd_may_claim - test whether a block device can be claimed
1026  * @bdev: block device of interest
1027  * @whole: whole block device containing @bdev, may equal @bdev
1028  * @holder: holder trying to claim @bdev
1029  *
1030  * Test whether @bdev can be claimed by @holder.
1031  *
1032  * CONTEXT:
1033  * spin_lock(&bdev_lock).
1034  *
1035  * RETURNS:
1036  * %true if @bdev can be claimed, %false otherwise.
1037  */
1038 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1039                          void *holder)
1040 {
1041         if (bdev->bd_holder == holder)
1042                 return true;     /* already a holder */
1043         else if (bdev->bd_holder != NULL)
1044                 return false;    /* held by someone else */
1045         else if (whole == bdev)
1046                 return true;     /* is a whole device which isn't held */
1047 
1048         else if (whole->bd_holder == bd_may_claim)
1049                 return true;     /* is a partition of a device that is being partitioned */
1050         else if (whole->bd_holder != NULL)
1051                 return false;    /* is a partition of a held device */
1052         else
1053                 return true;     /* is a partition of an un-held device */
1054 }
1055 
1056 /**
1057  * bd_prepare_to_claim - prepare to claim a block device
1058  * @bdev: block device of interest
1059  * @whole: the whole device containing @bdev, may equal @bdev
1060  * @holder: holder trying to claim @bdev
1061  *
1062  * Prepare to claim @bdev.  This function fails if @bdev is already
1063  * claimed by another holder and waits if another claiming is in
1064  * progress.  This function doesn't actually claim.  On successful
1065  * return, the caller has ownership of bd_claiming and bd_holder[s].
1066  *
1067  * CONTEXT:
1068  * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
1069  * it multiple times.
1070  *
1071  * RETURNS:
1072  * 0 if @bdev can be claimed, -EBUSY otherwise.
1073  */
1074 static int bd_prepare_to_claim(struct block_device *bdev,
1075                                struct block_device *whole, void *holder)
1076 {
1077 retry:
1078         /* if someone else claimed, fail */
1079         if (!bd_may_claim(bdev, whole, holder))
1080                 return -EBUSY;
1081 
1082         /* if claiming is already in progress, wait for it to finish */
1083         if (whole->bd_claiming) {
1084                 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1085                 DEFINE_WAIT(wait);
1086 
1087                 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1088                 spin_unlock(&bdev_lock);
1089                 schedule();
1090                 finish_wait(wq, &wait);
1091                 spin_lock(&bdev_lock);
1092                 goto retry;
1093         }
1094 
1095         /* yay, all mine */
1096         return 0;
1097 }
1098 
1099 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1100 {
1101         struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1102 
1103         if (!disk)
1104                 return NULL;
1105         /*
1106          * Now that we hold gendisk reference we make sure bdev we looked up is
1107          * not stale. If it is, it means device got removed and created before
1108          * we looked up gendisk and we fail open in such case. Associating
1109          * unhashed bdev with newly created gendisk could lead to two bdevs
1110          * (and thus two independent caches) being associated with one device
1111          * which is bad.
1112          */
1113         if (inode_unhashed(bdev->bd_inode)) {
1114                 put_disk_and_module(disk);
1115                 return NULL;
1116         }
1117         return disk;
1118 }
1119 
1120 /**
1121  * bd_start_claiming - start claiming a block device
1122  * @bdev: block device of interest
1123  * @holder: holder trying to claim @bdev
1124  *
1125  * @bdev is about to be opened exclusively.  Check @bdev can be opened
1126  * exclusively and mark that an exclusive open is in progress.  Each
1127  * successful call to this function must be matched with a call to
1128  * either bd_finish_claiming() or bd_abort_claiming() (which do not
1129  * fail).
1130  *
1131  * This function is used to gain exclusive access to the block device
1132  * without actually causing other exclusive open attempts to fail. It
1133  * should be used when the open sequence itself requires exclusive
1134  * access but may subsequently fail.
1135  *
1136  * CONTEXT:
1137  * Might sleep.
1138  *
1139  * RETURNS:
1140  * Pointer to the block device containing @bdev on success, ERR_PTR()
1141  * value on failure.
1142  */
1143 struct block_device *bd_start_claiming(struct block_device *bdev, void *holder)
1144 {
1145         struct gendisk *disk;
1146         struct block_device *whole;
1147         int partno, err;
1148 
1149         might_sleep();
1150 
1151         /*
1152          * @bdev might not have been initialized properly yet, look up
1153          * and grab the outer block device the hard way.
1154          */
1155         disk = bdev_get_gendisk(bdev, &partno);
1156         if (!disk)
1157                 return ERR_PTR(-ENXIO);
1158 
1159         /*
1160          * Normally, @bdev should equal what's returned from bdget_disk()
1161          * if partno is 0; however, some drivers (floppy) use multiple
1162          * bdev's for the same physical device and @bdev may be one of the
1163          * aliases.  Keep @bdev if partno is 0.  This means claimer
1164          * tracking is broken for those devices but it has always been that
1165          * way.
1166          */
1167         if (partno)
1168                 whole = bdget_disk(disk, 0);
1169         else
1170                 whole = bdgrab(bdev);
1171 
1172         put_disk_and_module(disk);
1173         if (!whole)
1174                 return ERR_PTR(-ENOMEM);
1175 
1176         /* prepare to claim, if successful, mark claiming in progress */
1177         spin_lock(&bdev_lock);
1178 
1179         err = bd_prepare_to_claim(bdev, whole, holder);
1180         if (err == 0) {
1181                 whole->bd_claiming = holder;
1182                 spin_unlock(&bdev_lock);
1183                 return whole;
1184         } else {
1185                 spin_unlock(&bdev_lock);
1186                 bdput(whole);
1187                 return ERR_PTR(err);
1188         }
1189 }
1190 EXPORT_SYMBOL(bd_start_claiming);
1191 
1192 static void bd_clear_claiming(struct block_device *whole, void *holder)
1193 {
1194         lockdep_assert_held(&bdev_lock);
1195         /* tell others that we're done */
1196         BUG_ON(whole->bd_claiming != holder);
1197         whole->bd_claiming = NULL;
1198         wake_up_bit(&whole->bd_claiming, 0);
1199 }
1200 
1201 /**
1202  * bd_finish_claiming - finish claiming of a block device
1203  * @bdev: block device of interest
1204  * @whole: whole block device (returned from bd_start_claiming())
1205  * @holder: holder that has claimed @bdev
1206  *
1207  * Finish exclusive open of a block device. Mark the device as exlusively
1208  * open by the holder and wake up all waiters for exclusive open to finish.
1209  */
1210 void bd_finish_claiming(struct block_device *bdev, struct block_device *whole,
1211                         void *holder)
1212 {
1213         spin_lock(&bdev_lock);
1214         BUG_ON(!bd_may_claim(bdev, whole, holder));
1215         /*
1216          * Note that for a whole device bd_holders will be incremented twice,
1217          * and bd_holder will be set to bd_may_claim before being set to holder
1218          */
1219         whole->bd_holders++;
1220         whole->bd_holder = bd_may_claim;
1221         bdev->bd_holders++;
1222         bdev->bd_holder = holder;
1223         bd_clear_claiming(whole, holder);
1224         spin_unlock(&bdev_lock);
1225 }
1226 EXPORT_SYMBOL(bd_finish_claiming);
1227 
1228 /**
1229  * bd_abort_claiming - abort claiming of a block device
1230  * @bdev: block device of interest
1231  * @whole: whole block device (returned from bd_start_claiming())
1232  * @holder: holder that has claimed @bdev
1233  *
1234  * Abort claiming of a block device when the exclusive open failed. This can be
1235  * also used when exclusive open is not actually desired and we just needed
1236  * to block other exclusive openers for a while.
1237  */
1238 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1239                        void *holder)
1240 {
1241         spin_lock(&bdev_lock);
1242         bd_clear_claiming(whole, holder);
1243         spin_unlock(&bdev_lock);
1244 }
1245 EXPORT_SYMBOL(bd_abort_claiming);
1246 
1247 #ifdef CONFIG_SYSFS
1248 struct bd_holder_disk {
1249         struct list_head        list;
1250         struct gendisk          *disk;
1251         int                     refcnt;
1252 };
1253 
1254 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1255                                                   struct gendisk *disk)
1256 {
1257         struct bd_holder_disk *holder;
1258 
1259         list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1260                 if (holder->disk == disk)
1261                         return holder;
1262         return NULL;
1263 }
1264 
1265 static int add_symlink(struct kobject *from, struct kobject *to)
1266 {
1267         return sysfs_create_link(from, to, kobject_name(to));
1268 }
1269 
1270 static void del_symlink(struct kobject *from, struct kobject *to)
1271 {
1272         sysfs_remove_link(from, kobject_name(to));
1273 }
1274 
1275 /**
1276  * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1277  * @bdev: the claimed slave bdev
1278  * @disk: the holding disk
1279  *
1280  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1281  *
1282  * This functions creates the following sysfs symlinks.
1283  *
1284  * - from "slaves" directory of the holder @disk to the claimed @bdev
1285  * - from "holders" directory of the @bdev to the holder @disk
1286  *
1287  * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1288  * passed to bd_link_disk_holder(), then:
1289  *
1290  *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
1291  *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1292  *
1293  * The caller must have claimed @bdev before calling this function and
1294  * ensure that both @bdev and @disk are valid during the creation and
1295  * lifetime of these symlinks.
1296  *
1297  * CONTEXT:
1298  * Might sleep.
1299  *
1300  * RETURNS:
1301  * 0 on success, -errno on failure.
1302  */
1303 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1304 {
1305         struct bd_holder_disk *holder;
1306         int ret = 0;
1307 
1308         mutex_lock(&bdev->bd_mutex);
1309 
1310         WARN_ON_ONCE(!bdev->bd_holder);
1311 
1312         /* FIXME: remove the following once add_disk() handles errors */
1313         if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1314                 goto out_unlock;
1315 
1316         holder = bd_find_holder_disk(bdev, disk);
1317         if (holder) {
1318                 holder->refcnt++;
1319                 goto out_unlock;
1320         }
1321 
1322         holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1323         if (!holder) {
1324                 ret = -ENOMEM;
1325                 goto out_unlock;
1326         }
1327 
1328         INIT_LIST_HEAD(&holder->list);
1329         holder->disk = disk;
1330         holder->refcnt = 1;
1331 
1332         ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1333         if (ret)
1334                 goto out_free;
1335 
1336         ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1337         if (ret)
1338                 goto out_del;
1339         /*
1340          * bdev could be deleted beneath us which would implicitly destroy
1341          * the holder directory.  Hold on to it.
1342          */
1343         kobject_get(bdev->bd_part->holder_dir);
1344 
1345         list_add(&holder->list, &bdev->bd_holder_disks);
1346         goto out_unlock;
1347 
1348 out_del:
1349         del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1350 out_free:
1351         kfree(holder);
1352 out_unlock:
1353         mutex_unlock(&bdev->bd_mutex);
1354         return ret;
1355 }
1356 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1357 
1358 /**
1359  * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1360  * @bdev: the calimed slave bdev
1361  * @disk: the holding disk
1362  *
1363  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1364  *
1365  * CONTEXT:
1366  * Might sleep.
1367  */
1368 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1369 {
1370         struct bd_holder_disk *holder;
1371 
1372         mutex_lock(&bdev->bd_mutex);
1373 
1374         holder = bd_find_holder_disk(bdev, disk);
1375 
1376         if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1377                 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1378                 del_symlink(bdev->bd_part->holder_dir,
1379                             &disk_to_dev(disk)->kobj);
1380                 kobject_put(bdev->bd_part->holder_dir);
1381                 list_del_init(&holder->list);
1382                 kfree(holder);
1383         }
1384 
1385         mutex_unlock(&bdev->bd_mutex);
1386 }
1387 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1388 #endif
1389 
1390 /**
1391  * flush_disk - invalidates all buffer-cache entries on a disk
1392  *
1393  * @bdev:      struct block device to be flushed
1394  * @kill_dirty: flag to guide handling of dirty inodes
1395  *
1396  * Invalidates all buffer-cache entries on a disk. It should be called
1397  * when a disk has been changed -- either by a media change or online
1398  * resize.
1399  */
1400 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1401 {
1402         if (__invalidate_device(bdev, kill_dirty)) {
1403                 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1404                        "resized disk %s\n",
1405                        bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1406         }
1407         bdev->bd_invalidated = 1;
1408 }
1409 
1410 /**
1411  * check_disk_size_change - checks for disk size change and adjusts bdev size.
1412  * @disk: struct gendisk to check
1413  * @bdev: struct bdev to adjust.
1414  * @verbose: if %true log a message about a size change if there is any
1415  *
1416  * This routine checks to see if the bdev size does not match the disk size
1417  * and adjusts it if it differs. When shrinking the bdev size, its all caches
1418  * are freed.
1419  */
1420 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1421                 bool verbose)
1422 {
1423         loff_t disk_size, bdev_size;
1424 
1425         disk_size = (loff_t)get_capacity(disk) << 9;
1426         bdev_size = i_size_read(bdev->bd_inode);
1427         if (disk_size != bdev_size) {
1428                 if (verbose) {
1429                         printk(KERN_INFO
1430                                "%s: detected capacity change from %lld to %lld\n",
1431                                disk->disk_name, bdev_size, disk_size);
1432                 }
1433                 i_size_write(bdev->bd_inode, disk_size);
1434                 if (bdev_size > disk_size)
1435                         flush_disk(bdev, false);
1436         }
1437 }
1438 
1439 /**
1440  * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1441  * @disk: struct gendisk to be revalidated
1442  *
1443  * This routine is a wrapper for lower-level driver's revalidate_disk
1444  * call-backs.  It is used to do common pre and post operations needed
1445  * for all revalidate_disk operations.
1446  */
1447 int revalidate_disk(struct gendisk *disk)
1448 {
1449         int ret = 0;
1450 
1451         if (disk->fops->revalidate_disk)
1452                 ret = disk->fops->revalidate_disk(disk);
1453 
1454         /*
1455          * Hidden disks don't have associated bdev so there's no point in
1456          * revalidating it.
1457          */
1458         if (!(disk->flags & GENHD_FL_HIDDEN)) {
1459                 struct block_device *bdev = bdget_disk(disk, 0);
1460 
1461                 if (!bdev)
1462                         return ret;
1463 
1464                 mutex_lock(&bdev->bd_mutex);
1465                 check_disk_size_change(disk, bdev, ret == 0);
1466                 bdev->bd_invalidated = 0;
1467                 mutex_unlock(&bdev->bd_mutex);
1468                 bdput(bdev);
1469         }
1470         return ret;
1471 }
1472 EXPORT_SYMBOL(revalidate_disk);
1473 
1474 /*
1475  * This routine checks whether a removable media has been changed,
1476  * and invalidates all buffer-cache-entries in that case. This
1477  * is a relatively slow routine, so we have to try to minimize using
1478  * it. Thus it is called only upon a 'mount' or 'open'. This
1479  * is the best way of combining speed and utility, I think.
1480  * People changing diskettes in the middle of an operation deserve
1481  * to lose :-)
1482  */
1483 int check_disk_change(struct block_device *bdev)
1484 {
1485         struct gendisk *disk = bdev->bd_disk;
1486         const struct block_device_operations *bdops = disk->fops;
1487         unsigned int events;
1488 
1489         events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1490                                    DISK_EVENT_EJECT_REQUEST);
1491         if (!(events & DISK_EVENT_MEDIA_CHANGE))
1492                 return 0;
1493 
1494         flush_disk(bdev, true);
1495         if (bdops->revalidate_disk)
1496                 bdops->revalidate_disk(bdev->bd_disk);
1497         return 1;
1498 }
1499 
1500 EXPORT_SYMBOL(check_disk_change);
1501 
1502 void bd_set_size(struct block_device *bdev, loff_t size)
1503 {
1504         inode_lock(bdev->bd_inode);
1505         i_size_write(bdev->bd_inode, size);
1506         inode_unlock(bdev->bd_inode);
1507 }
1508 EXPORT_SYMBOL(bd_set_size);
1509 
1510 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1511 
1512 static void bdev_disk_changed(struct block_device *bdev, bool invalidate)
1513 {
1514         if (disk_part_scan_enabled(bdev->bd_disk)) {
1515                 if (invalidate)
1516                         invalidate_partitions(bdev->bd_disk, bdev);
1517                 else
1518                         rescan_partitions(bdev->bd_disk, bdev);
1519         } else {
1520                 check_disk_size_change(bdev->bd_disk, bdev, !invalidate);
1521                 bdev->bd_invalidated = 0;
1522         }
1523 }
1524 
1525 /*
1526  * bd_mutex locking:
1527  *
1528  *  mutex_lock(part->bd_mutex)
1529  *    mutex_lock_nested(whole->bd_mutex, 1)
1530  */
1531 
1532 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1533 {
1534         struct gendisk *disk;
1535         int ret;
1536         int partno;
1537         int perm = 0;
1538         bool first_open = false;
1539 
1540         if (mode & FMODE_READ)
1541                 perm |= MAY_READ;
1542         if (mode & FMODE_WRITE)
1543                 perm |= MAY_WRITE;
1544         /*
1545          * hooks: /n/, see "layering violations".
1546          */
1547         if (!for_part) {
1548                 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1549                 if (ret != 0) {
1550                         bdput(bdev);
1551                         return ret;
1552                 }
1553         }
1554 
1555  restart:
1556 
1557         ret = -ENXIO;
1558         disk = bdev_get_gendisk(bdev, &partno);
1559         if (!disk)
1560                 goto out;
1561 
1562         disk_block_events(disk);
1563         mutex_lock_nested(&bdev->bd_mutex, for_part);
1564         if (!bdev->bd_openers) {
1565                 first_open = true;
1566                 bdev->bd_disk = disk;
1567                 bdev->bd_queue = disk->queue;
1568                 bdev->bd_contains = bdev;
1569                 bdev->bd_partno = partno;
1570 
1571                 if (!partno) {
1572                         ret = -ENXIO;
1573                         bdev->bd_part = disk_get_part(disk, partno);
1574                         if (!bdev->bd_part)
1575                                 goto out_clear;
1576 
1577                         ret = 0;
1578                         if (disk->fops->open) {
1579                                 ret = disk->fops->open(bdev, mode);
1580                                 if (ret == -ERESTARTSYS) {
1581                                         /* Lost a race with 'disk' being
1582                                          * deleted, try again.
1583                                          * See md.c
1584                                          */
1585                                         disk_put_part(bdev->bd_part);
1586                                         bdev->bd_part = NULL;
1587                                         bdev->bd_disk = NULL;
1588                                         bdev->bd_queue = NULL;
1589                                         mutex_unlock(&bdev->bd_mutex);
1590                                         disk_unblock_events(disk);
1591                                         put_disk_and_module(disk);
1592                                         goto restart;
1593                                 }
1594                         }
1595 
1596                         if (!ret) {
1597                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1598                                 set_init_blocksize(bdev);
1599                         }
1600 
1601                         /*
1602                          * If the device is invalidated, rescan partition
1603                          * if open succeeded or failed with -ENOMEDIUM.
1604                          * The latter is necessary to prevent ghost
1605                          * partitions on a removed medium.
1606                          */
1607                         if (bdev->bd_invalidated &&
1608                             (!ret || ret == -ENOMEDIUM))
1609                                 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1610 
1611                         if (ret)
1612                                 goto out_clear;
1613                 } else {
1614                         struct block_device *whole;
1615                         whole = bdget_disk(disk, 0);
1616                         ret = -ENOMEM;
1617                         if (!whole)
1618                                 goto out_clear;
1619                         BUG_ON(for_part);
1620                         ret = __blkdev_get(whole, mode, 1);
1621                         if (ret)
1622                                 goto out_clear;
1623                         bdev->bd_contains = whole;
1624                         bdev->bd_part = disk_get_part(disk, partno);
1625                         if (!(disk->flags & GENHD_FL_UP) ||
1626                             !bdev->bd_part || !bdev->bd_part->nr_sects) {
1627                                 ret = -ENXIO;
1628                                 goto out_clear;
1629                         }
1630                         bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1631                         set_init_blocksize(bdev);
1632                 }
1633 
1634                 if (bdev->bd_bdi == &noop_backing_dev_info)
1635                         bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1636         } else {
1637                 if (bdev->bd_contains == bdev) {
1638                         ret = 0;
1639                         if (bdev->bd_disk->fops->open)
1640                                 ret = bdev->bd_disk->fops->open(bdev, mode);
1641                         /* the same as first opener case, read comment there */
1642                         if (bdev->bd_invalidated &&
1643                             (!ret || ret == -ENOMEDIUM))
1644                                 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1645                         if (ret)
1646                                 goto out_unlock_bdev;
1647                 }
1648         }
1649         bdev->bd_openers++;
1650         if (for_part)
1651                 bdev->bd_part_count++;
1652         mutex_unlock(&bdev->bd_mutex);
1653         disk_unblock_events(disk);
1654         /* only one opener holds refs to the module and disk */
1655         if (!first_open)
1656                 put_disk_and_module(disk);
1657         return 0;
1658 
1659  out_clear:
1660         disk_put_part(bdev->bd_part);
1661         bdev->bd_disk = NULL;
1662         bdev->bd_part = NULL;
1663         bdev->bd_queue = NULL;
1664         if (bdev != bdev->bd_contains)
1665                 __blkdev_put(bdev->bd_contains, mode, 1);
1666         bdev->bd_contains = NULL;
1667  out_unlock_bdev:
1668         mutex_unlock(&bdev->bd_mutex);
1669         disk_unblock_events(disk);
1670         put_disk_and_module(disk);
1671  out:
1672         bdput(bdev);
1673 
1674         return ret;
1675 }
1676 
1677 /**
1678  * blkdev_get - open a block device
1679  * @bdev: block_device to open
1680  * @mode: FMODE_* mask
1681  * @holder: exclusive holder identifier
1682  *
1683  * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1684  * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1685  * @holder is invalid.  Exclusive opens may nest for the same @holder.
1686  *
1687  * On success, the reference count of @bdev is unchanged.  On failure,
1688  * @bdev is put.
1689  *
1690  * CONTEXT:
1691  * Might sleep.
1692  *
1693  * RETURNS:
1694  * 0 on success, -errno on failure.
1695  */
1696 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1697 {
1698         struct block_device *whole = NULL;
1699         int res;
1700 
1701         WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1702 
1703         if ((mode & FMODE_EXCL) && holder) {
1704                 whole = bd_start_claiming(bdev, holder);
1705                 if (IS_ERR(whole)) {
1706                         bdput(bdev);
1707                         return PTR_ERR(whole);
1708                 }
1709         }
1710 
1711         res = __blkdev_get(bdev, mode, 0);
1712 
1713         if (whole) {
1714                 struct gendisk *disk = whole->bd_disk;
1715 
1716                 /* finish claiming */
1717                 mutex_lock(&bdev->bd_mutex);
1718                 if (!res)
1719                         bd_finish_claiming(bdev, whole, holder);
1720                 else
1721                         bd_abort_claiming(bdev, whole, holder);
1722                 /*
1723                  * Block event polling for write claims if requested.  Any
1724                  * write holder makes the write_holder state stick until
1725                  * all are released.  This is good enough and tracking
1726                  * individual writeable reference is too fragile given the
1727                  * way @mode is used in blkdev_get/put().
1728                  */
1729                 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1730                     (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1731                         bdev->bd_write_holder = true;
1732                         disk_block_events(disk);
1733                 }
1734 
1735                 mutex_unlock(&bdev->bd_mutex);
1736                 bdput(whole);
1737         }
1738 
1739         return res;
1740 }
1741 EXPORT_SYMBOL(blkdev_get);
1742 
1743 /**
1744  * blkdev_get_by_path - open a block device by name
1745  * @path: path to the block device to open
1746  * @mode: FMODE_* mask
1747  * @holder: exclusive holder identifier
1748  *
1749  * Open the blockdevice described by the device file at @path.  @mode
1750  * and @holder are identical to blkdev_get().
1751  *
1752  * On success, the returned block_device has reference count of one.
1753  *
1754  * CONTEXT:
1755  * Might sleep.
1756  *
1757  * RETURNS:
1758  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1759  */
1760 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1761                                         void *holder)
1762 {
1763         struct block_device *bdev;
1764         int err;
1765 
1766         bdev = lookup_bdev(path);
1767         if (IS_ERR(bdev))
1768                 return bdev;
1769 
1770         err = blkdev_get(bdev, mode, holder);
1771         if (err)
1772                 return ERR_PTR(err);
1773 
1774         if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1775                 blkdev_put(bdev, mode);
1776                 return ERR_PTR(-EACCES);
1777         }
1778 
1779         return bdev;
1780 }
1781 EXPORT_SYMBOL(blkdev_get_by_path);
1782 
1783 /**
1784  * blkdev_get_by_dev - open a block device by device number
1785  * @dev: device number of block device to open
1786  * @mode: FMODE_* mask
1787  * @holder: exclusive holder identifier
1788  *
1789  * Open the blockdevice described by device number @dev.  @mode and
1790  * @holder are identical to blkdev_get().
1791  *
1792  * Use it ONLY if you really do not have anything better - i.e. when
1793  * you are behind a truly sucky interface and all you are given is a
1794  * device number.  _Never_ to be used for internal purposes.  If you
1795  * ever need it - reconsider your API.
1796  *
1797  * On success, the returned block_device has reference count of one.
1798  *
1799  * CONTEXT:
1800  * Might sleep.
1801  *
1802  * RETURNS:
1803  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1804  */
1805 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1806 {
1807         struct block_device *bdev;
1808         int err;
1809 
1810         bdev = bdget(dev);
1811         if (!bdev)
1812                 return ERR_PTR(-ENOMEM);
1813 
1814         err = blkdev_get(bdev, mode, holder);
1815         if (err)
1816                 return ERR_PTR(err);
1817 
1818         return bdev;
1819 }
1820 EXPORT_SYMBOL(blkdev_get_by_dev);
1821 
1822 static int blkdev_open(struct inode * inode, struct file * filp)
1823 {
1824         struct block_device *bdev;
1825 
1826         /*
1827          * Preserve backwards compatibility and allow large file access
1828          * even if userspace doesn't ask for it explicitly. Some mkfs
1829          * binary needs it. We might want to drop this workaround
1830          * during an unstable branch.
1831          */
1832         filp->f_flags |= O_LARGEFILE;
1833 
1834         filp->f_mode |= FMODE_NOWAIT;
1835 
1836         if (filp->f_flags & O_NDELAY)
1837                 filp->f_mode |= FMODE_NDELAY;
1838         if (filp->f_flags & O_EXCL)
1839                 filp->f_mode |= FMODE_EXCL;
1840         if ((filp->f_flags & O_ACCMODE) == 3)
1841                 filp->f_mode |= FMODE_WRITE_IOCTL;
1842 
1843         bdev = bd_acquire(inode);
1844         if (bdev == NULL)
1845                 return -ENOMEM;
1846 
1847         filp->f_mapping = bdev->bd_inode->i_mapping;
1848         filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1849 
1850         return blkdev_get(bdev, filp->f_mode, filp);
1851 }
1852 
1853 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1854 {
1855         struct gendisk *disk = bdev->bd_disk;
1856         struct block_device *victim = NULL;
1857 
1858         mutex_lock_nested(&bdev->bd_mutex, for_part);
1859         if (for_part)
1860                 bdev->bd_part_count--;
1861 
1862         if (!--bdev->bd_openers) {
1863                 WARN_ON_ONCE(bdev->bd_holders);
1864                 sync_blockdev(bdev);
1865                 kill_bdev(bdev);
1866 
1867                 bdev_write_inode(bdev);
1868         }
1869         if (bdev->bd_contains == bdev) {
1870                 if (disk->fops->release)
1871                         disk->fops->release(disk, mode);
1872         }
1873         if (!bdev->bd_openers) {
1874                 disk_put_part(bdev->bd_part);
1875                 bdev->bd_part = NULL;
1876                 bdev->bd_disk = NULL;
1877                 if (bdev != bdev->bd_contains)
1878                         victim = bdev->bd_contains;
1879                 bdev->bd_contains = NULL;
1880 
1881                 put_disk_and_module(disk);
1882         }
1883         mutex_unlock(&bdev->bd_mutex);
1884         bdput(bdev);
1885         if (victim)
1886                 __blkdev_put(victim, mode, 1);
1887 }
1888 
1889 void blkdev_put(struct block_device *bdev, fmode_t mode)
1890 {
1891         mutex_lock(&bdev->bd_mutex);
1892 
1893         if (mode & FMODE_EXCL) {
1894                 bool bdev_free;
1895 
1896                 /*
1897                  * Release a claim on the device.  The holder fields
1898                  * are protected with bdev_lock.  bd_mutex is to
1899                  * synchronize disk_holder unlinking.
1900                  */
1901                 spin_lock(&bdev_lock);
1902 
1903                 WARN_ON_ONCE(--bdev->bd_holders < 0);
1904                 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1905 
1906                 /* bd_contains might point to self, check in a separate step */
1907                 if ((bdev_free = !bdev->bd_holders))
1908                         bdev->bd_holder = NULL;
1909                 if (!bdev->bd_contains->bd_holders)
1910                         bdev->bd_contains->bd_holder = NULL;
1911 
1912                 spin_unlock(&bdev_lock);
1913 
1914                 /*
1915                  * If this was the last claim, remove holder link and
1916                  * unblock evpoll if it was a write holder.
1917                  */
1918                 if (bdev_free && bdev->bd_write_holder) {
1919                         disk_unblock_events(bdev->bd_disk);
1920                         bdev->bd_write_holder = false;
1921                 }
1922         }
1923 
1924         /*
1925          * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1926          * event.  This is to ensure detection of media removal commanded
1927          * from userland - e.g. eject(1).
1928          */
1929         disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1930 
1931         mutex_unlock(&bdev->bd_mutex);
1932 
1933         __blkdev_put(bdev, mode, 0);
1934 }
1935 EXPORT_SYMBOL(blkdev_put);
1936 
1937 static int blkdev_close(struct inode * inode, struct file * filp)
1938 {
1939         struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1940         blkdev_put(bdev, filp->f_mode);
1941         return 0;
1942 }
1943 
1944 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1945 {
1946         struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1947         fmode_t mode = file->f_mode;
1948 
1949         /*
1950          * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1951          * to updated it before every ioctl.
1952          */
1953         if (file->f_flags & O_NDELAY)
1954                 mode |= FMODE_NDELAY;
1955         else
1956                 mode &= ~FMODE_NDELAY;
1957 
1958         return blkdev_ioctl(bdev, mode, cmd, arg);
1959 }
1960 
1961 /*
1962  * Write data to the block device.  Only intended for the block device itself
1963  * and the raw driver which basically is a fake block device.
1964  *
1965  * Does not take i_mutex for the write and thus is not for general purpose
1966  * use.
1967  */
1968 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1969 {
1970         struct file *file = iocb->ki_filp;
1971         struct inode *bd_inode = bdev_file_inode(file);
1972         loff_t size = i_size_read(bd_inode);
1973         struct blk_plug plug;
1974         ssize_t ret;
1975 
1976         if (bdev_read_only(I_BDEV(bd_inode)))
1977                 return -EPERM;
1978 
1979         /* uswsusp needs write permission to the swap */
1980         if (IS_SWAPFILE(bd_inode) && !hibernation_available())
1981                 return -ETXTBSY;
1982 
1983         if (!iov_iter_count(from))
1984                 return 0;
1985 
1986         if (iocb->ki_pos >= size)
1987                 return -ENOSPC;
1988 
1989         if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1990                 return -EOPNOTSUPP;
1991 
1992         iov_iter_truncate(from, size - iocb->ki_pos);
1993 
1994         blk_start_plug(&plug);
1995         ret = __generic_file_write_iter(iocb, from);
1996         if (ret > 0)
1997                 ret = generic_write_sync(iocb, ret);
1998         blk_finish_plug(&plug);
1999         return ret;
2000 }
2001 EXPORT_SYMBOL_GPL(blkdev_write_iter);
2002 
2003 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
2004 {
2005         struct file *file = iocb->ki_filp;
2006         struct inode *bd_inode = bdev_file_inode(file);
2007         loff_t size = i_size_read(bd_inode);
2008         loff_t pos = iocb->ki_pos;
2009 
2010         if (pos >= size)
2011                 return 0;
2012 
2013         size -= pos;
2014         iov_iter_truncate(to, size);
2015         return generic_file_read_iter(iocb, to);
2016 }
2017 EXPORT_SYMBOL_GPL(blkdev_read_iter);
2018 
2019 /*
2020  * Try to release a page associated with block device when the system
2021  * is under memory pressure.
2022  */
2023 static int blkdev_releasepage(struct page *page, gfp_t wait)
2024 {
2025         struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2026 
2027         if (super && super->s_op->bdev_try_to_free_page)
2028                 return super->s_op->bdev_try_to_free_page(super, page, wait);
2029 
2030         return try_to_free_buffers(page);
2031 }
2032 
2033 static int blkdev_writepages(struct address_space *mapping,
2034                              struct writeback_control *wbc)
2035 {
2036         return generic_writepages(mapping, wbc);
2037 }
2038 
2039 static const struct address_space_operations def_blk_aops = {
2040         .readpage       = blkdev_readpage,
2041         .readpages      = blkdev_readpages,
2042         .writepage      = blkdev_writepage,
2043         .write_begin    = blkdev_write_begin,
2044         .write_end      = blkdev_write_end,
2045         .writepages     = blkdev_writepages,
2046         .releasepage    = blkdev_releasepage,
2047         .direct_IO      = blkdev_direct_IO,
2048         .migratepage    = buffer_migrate_page_norefs,
2049         .is_dirty_writeback = buffer_check_dirty_writeback,
2050 };
2051 
2052 #define BLKDEV_FALLOC_FL_SUPPORTED                                      \
2053                 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |           \
2054                  FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2055 
2056 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2057                              loff_t len)
2058 {
2059         struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2060         struct address_space *mapping;
2061         loff_t end = start + len - 1;
2062         loff_t isize;
2063         int error;
2064 
2065         /* Fail if we don't recognize the flags. */
2066         if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2067                 return -EOPNOTSUPP;
2068 
2069         /* Don't go off the end of the device. */
2070         isize = i_size_read(bdev->bd_inode);
2071         if (start >= isize)
2072                 return -EINVAL;
2073         if (end >= isize) {
2074                 if (mode & FALLOC_FL_KEEP_SIZE) {
2075                         len = isize - start;
2076                         end = start + len - 1;
2077                 } else
2078                         return -EINVAL;
2079         }
2080 
2081         /*
2082          * Don't allow IO that isn't aligned to logical block size.
2083          */
2084         if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2085                 return -EINVAL;
2086 
2087         /* Invalidate the page cache, including dirty pages. */
2088         mapping = bdev->bd_inode->i_mapping;
2089         truncate_inode_pages_range(mapping, start, end);
2090 
2091         switch (mode) {
2092         case FALLOC_FL_ZERO_RANGE:
2093         case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2094                 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2095                                             GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2096                 break;
2097         case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2098                 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2099                                              GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2100                 break;
2101         case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2102                 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2103                                              GFP_KERNEL, 0);
2104                 break;
2105         default:
2106                 return -EOPNOTSUPP;
2107         }
2108         if (error)
2109                 return error;
2110 
2111         /*
2112          * Invalidate again; if someone wandered in and dirtied a page,
2113          * the caller will be given -EBUSY.  The third argument is
2114          * inclusive, so the rounding here is safe.
2115          */
2116         return invalidate_inode_pages2_range(mapping,
2117                                              start >> PAGE_SHIFT,
2118                                              end >> PAGE_SHIFT);
2119 }
2120 
2121 const struct file_operations def_blk_fops = {
2122         .open           = blkdev_open,
2123         .release        = blkdev_close,
2124         .llseek         = block_llseek,
2125         .read_iter      = blkdev_read_iter,
2126         .write_iter     = blkdev_write_iter,
2127         .iopoll         = blkdev_iopoll,
2128         .mmap           = generic_file_mmap,
2129         .fsync          = blkdev_fsync,
2130         .unlocked_ioctl = block_ioctl,
2131 #ifdef CONFIG_COMPAT
2132         .compat_ioctl   = compat_blkdev_ioctl,
2133 #endif
2134         .splice_read    = generic_file_splice_read,
2135         .splice_write   = iter_file_splice_write,
2136         .fallocate      = blkdev_fallocate,
2137 };
2138 
2139 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2140 {
2141         int res;
2142         mm_segment_t old_fs = get_fs();
2143         set_fs(KERNEL_DS);
2144         res = blkdev_ioctl(bdev, 0, cmd, arg);
2145         set_fs(old_fs);
2146         return res;
2147 }
2148 
2149 EXPORT_SYMBOL(ioctl_by_bdev);
2150 
2151 /**
2152  * lookup_bdev  - lookup a struct block_device by name
2153  * @pathname:   special file representing the block device
2154  *
2155  * Get a reference to the blockdevice at @pathname in the current
2156  * namespace if possible and return it.  Return ERR_PTR(error)
2157  * otherwise.
2158  */
2159 struct block_device *lookup_bdev(const char *pathname)
2160 {
2161         struct block_device *bdev;
2162         struct inode *inode;
2163         struct path path;
2164         int error;
2165 
2166         if (!pathname || !*pathname)
2167                 return ERR_PTR(-EINVAL);
2168 
2169         error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2170         if (error)
2171                 return ERR_PTR(error);
2172 
2173         inode = d_backing_inode(path.dentry);
2174         error = -ENOTBLK;
2175         if (!S_ISBLK(inode->i_mode))
2176                 goto fail;
2177         error = -EACCES;
2178         if (!may_open_dev(&path))
2179                 goto fail;
2180         error = -ENOMEM;
2181         bdev = bd_acquire(inode);
2182         if (!bdev)
2183                 goto fail;
2184 out:
2185         path_put(&path);
2186         return bdev;
2187 fail:
2188         bdev = ERR_PTR(error);
2189         goto out;
2190 }
2191 EXPORT_SYMBOL(lookup_bdev);
2192 
2193 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2194 {
2195         struct super_block *sb = get_super(bdev);
2196         int res = 0;
2197 
2198         if (sb) {
2199                 /*
2200                  * no need to lock the super, get_super holds the
2201                  * read mutex so the filesystem cannot go away
2202                  * under us (->put_super runs with the write lock
2203                  * hold).
2204                  */
2205                 shrink_dcache_sb(sb);
2206                 res = invalidate_inodes(sb, kill_dirty);
2207                 drop_super(sb);
2208         }
2209         invalidate_bdev(bdev);
2210         return res;
2211 }
2212 EXPORT_SYMBOL(__invalidate_device);
2213 
2214 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2215 {
2216         struct inode *inode, *old_inode = NULL;
2217 
2218         spin_lock(&blockdev_superblock->s_inode_list_lock);
2219         list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2220                 struct address_space *mapping = inode->i_mapping;
2221                 struct block_device *bdev;
2222 
2223                 spin_lock(&inode->i_lock);
2224                 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2225                     mapping->nrpages == 0) {
2226                         spin_unlock(&inode->i_lock);
2227                         continue;
2228                 }
2229                 __iget(inode);
2230                 spin_unlock(&inode->i_lock);
2231                 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2232                 /*
2233                  * We hold a reference to 'inode' so it couldn't have been
2234                  * removed from s_inodes list while we dropped the
2235                  * s_inode_list_lock  We cannot iput the inode now as we can
2236                  * be holding the last reference and we cannot iput it under
2237                  * s_inode_list_lock. So we keep the reference and iput it
2238                  * later.
2239                  */
2240                 iput(old_inode);
2241                 old_inode = inode;
2242                 bdev = I_BDEV(inode);
2243 
2244                 mutex_lock(&bdev->bd_mutex);
2245                 if (bdev->bd_openers)
2246                         func(bdev, arg);
2247                 mutex_unlock(&bdev->bd_mutex);
2248 
2249                 spin_lock(&blockdev_superblock->s_inode_list_lock);
2250         }
2251         spin_unlock(&blockdev_superblock->s_inode_list_lock);
2252         iput(old_inode);
2253 }
/* [<][>][^][v][top][bottom][index][help] */
root/fs/block_dev.c

DEFINITIONS
