block/bdev.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  Copyright (C) 1991, 1992  Linus Torvalds
   4  *  Copyright (C) 2001  Andrea Arcangeli <andrea@suse.de> SuSE
   5  *  Copyright (C) 2016 - 2020 Christoph Hellwig
   6  */
   7
   8 #include <linux/init.h>
   9 #include <linux/mm.h>
  10 #include <linux/slab.h>
  11 #include <linux/kmod.h>
  12 #include <linux/major.h>
  13 #include <linux/device_cgroup.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/blk-integrity.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/module.h>
  18 #include <linux/blkpg.h>
  19 #include <linux/magic.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/swap.h>
  22 #include <linux/writeback.h>
  23 #include <linux/mount.h>
  24 #include <linux/pseudo_fs.h>
  25 #include <linux/uio.h>
  26 #include <linux/namei.h>
  27 #include <linux/part_stat.h>
  28 #include <linux/uaccess.h>
  29 #include <linux/stat.h>
  30 #include "../fs/internal.h"
  31 #include "blk.h"
  32
  33 struct bdev_inode {
  34         struct block_device bdev;
  35         struct inode vfs_inode;
  36 };
  37
  38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
  39 {
  40         return container_of(inode, struct bdev_inode, vfs_inode);
  41 }
  42
  43 struct block_device *I_BDEV(struct inode *inode)
  44 {
  45         return &BDEV_I(inode)->bdev;
  46 }
  47 EXPORT_SYMBOL(I_BDEV);
  48
  49 static void bdev_write_inode(struct block_device *bdev)
  50 {
  51         struct inode *inode = bdev->bd_inode;
  52         int ret;
  53
  54         spin_lock(&inode->i_lock);
  55         while (inode->i_state & I_DIRTY) {
  56                 spin_unlock(&inode->i_lock);
  57                 ret = write_inode_now(inode, true);
  58                 if (ret)
  59                         pr_warn_ratelimited(
  60         "VFS: Dirty inode writeback failed for block device %pg (err=%d).\n",
  61                                 bdev, ret);
  62                 spin_lock(&inode->i_lock);
  63         }
  64         spin_unlock(&inode->i_lock);
  65 }
  66
  67 /* Kill _all_ buffers and pagecache , dirty or not.. */
  68 static void kill_bdev(struct block_device *bdev)
  69 {
  70         struct address_space *mapping = bdev->bd_inode->i_mapping;
  71
  72         if (mapping_empty(mapping))
  73                 return;
  74
  75         invalidate_bh_lrus();
  76         truncate_inode_pages(mapping, 0);
  77 }
  78
  79 /* Invalidate clean unused buffers and pagecache. */
  80 void invalidate_bdev(struct block_device *bdev)
  81 {
  82         struct address_space *mapping = bdev->bd_inode->i_mapping;
  83
  84         if (mapping->nrpages) {
  85                 invalidate_bh_lrus();
  86                 lru_add_drain_all();    /* make sure all lru add caches are flushed */
  87                 invalidate_mapping_pages(mapping, 0, -1);
  88         }
  89 }
  90 EXPORT_SYMBOL(invalidate_bdev);
  91
  92 /*
  93  * Drop all buffers & page cache for given bdev range. This function bails
  94  * with error if bdev has other exclusive owner (such as filesystem).
  95  */
  96 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
  97                         loff_t lstart, loff_t lend)
  98 {
  99         /*
 100          * If we don't hold exclusive handle for the device, upgrade to it
 101          * while we discard the buffer cache to avoid discarding buffers
 102          * under live filesystem.
 103          */
 104         if (!(mode & BLK_OPEN_EXCL)) {
 105                 int err = bd_prepare_to_claim(bdev, truncate_bdev_range, NULL);
 106                 if (err)
 107                         goto invalidate;
 108         }
 109
 110         truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
 111         if (!(mode & BLK_OPEN_EXCL))
 112                 bd_abort_claiming(bdev, truncate_bdev_range);
 113         return 0;
 114
 115 invalidate:
 116         /*
 117          * Someone else has handle exclusively open. Try invalidating instead.
 118          * The 'end' argument is inclusive so the rounding is safe.
 119          */
 120         return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
 121                                              lstart >> PAGE_SHIFT,
 122                                              lend >> PAGE_SHIFT);
 123 }
 124
 125 static void set_init_blocksize(struct block_device *bdev)
 126 {
 127         unsigned int bsize = bdev_logical_block_size(bdev);
 128         loff_t size = i_size_read(bdev->bd_inode);
 129
 130         while (bsize < PAGE_SIZE) {
 131                 if (size & bsize)
 132                         break;
 133                 bsize <<= 1;
 134         }
 135         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 136 }
 137
 138 int set_blocksize(struct block_device *bdev, int size)
 139 {
 140         /* Size must be a power of two, and between 512 and PAGE_SIZE */
 141         if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
 142                 return -EINVAL;
 143
 144         /* Size cannot be smaller than the size supported by the device */
 145         if (size < bdev_logical_block_size(bdev))
 146                 return -EINVAL;
 147
 148         /* Don't change the size if it is same as current */
 149         if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
 150                 sync_blockdev(bdev);
 151                 bdev->bd_inode->i_blkbits = blksize_bits(size);
 152                 kill_bdev(bdev);
 153         }
 154         return 0;
 155 }
 156
 157 EXPORT_SYMBOL(set_blocksize);
 158
 159 int sb_set_blocksize(struct super_block *sb, int size)
 160 {
 161         if (set_blocksize(sb->s_bdev, size))
 162                 return 0;
 163         /* If we get here, we know size is power of two
 164          * and it's value is between 512 and PAGE_SIZE */
 165         sb->s_blocksize = size;
 166         sb->s_blocksize_bits = blksize_bits(size);
 167         return sb->s_blocksize;
 168 }
 169
 170 EXPORT_SYMBOL(sb_set_blocksize);
 171
 172 int sb_min_blocksize(struct super_block *sb, int size)
 173 {
 174         int minsize = bdev_logical_block_size(sb->s_bdev);
 175         if (size < minsize)
 176                 size = minsize;
 177         return sb_set_blocksize(sb, size);
 178 }
 179
 180 EXPORT_SYMBOL(sb_min_blocksize);
 181
 182 int sync_blockdev_nowait(struct block_device *bdev)
 183 {
 184         if (!bdev)
 185                 return 0;
 186         return filemap_flush(bdev->bd_inode->i_mapping);
 187 }
 188 EXPORT_SYMBOL_GPL(sync_blockdev_nowait);
 189
 190 /*
 191  * Write out and wait upon all the dirty data associated with a block
 192  * device via its mapping.  Does not take the superblock lock.
 193  */
 194 int sync_blockdev(struct block_device *bdev)
 195 {
 196         if (!bdev)
 197                 return 0;
 198         return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 199 }
 200 EXPORT_SYMBOL(sync_blockdev);
 201
 202 int sync_blockdev_range(struct block_device *bdev, loff_t lstart, loff_t lend)
 203 {
 204         return filemap_write_and_wait_range(bdev->bd_inode->i_mapping,
 205                         lstart, lend);
 206 }
 207 EXPORT_SYMBOL(sync_blockdev_range);
 208
 209 /**
 210  * freeze_bdev - lock a filesystem and force it into a consistent state
 211  * @bdev:       blockdevice to lock
 212  *
 213  * If a superblock is found on this device, we take the s_umount semaphore
 214  * on it to make sure nobody unmounts until the snapshot creation is done.
 215  * The reference counter (bd_fsfreeze_count) guarantees that only the last
 216  * unfreeze process can unfreeze the frozen filesystem actually when multiple
 217  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 218  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 219  * actually.
 220  */
 221 int freeze_bdev(struct block_device *bdev)
 222 {
 223         struct super_block *sb;
 224         int error = 0;
 225
 226         mutex_lock(&bdev->bd_fsfreeze_mutex);
 227         if (++bdev->bd_fsfreeze_count > 1)
 228                 goto done;
 229
 230         sb = get_active_super(bdev);
 231         if (!sb)
 232                 goto sync;
 233         if (sb->s_op->freeze_super)
 234                 error = sb->s_op->freeze_super(sb, FREEZE_HOLDER_USERSPACE);
 235         else
 236                 error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
 237         deactivate_super(sb);
 238
 239         if (error) {
 240                 bdev->bd_fsfreeze_count--;
 241                 goto done;
 242         }
 243         bdev->bd_fsfreeze_sb = sb;
 244
 245 sync:
 246         sync_blockdev(bdev);
 247 done:
 248         mutex_unlock(&bdev->bd_fsfreeze_mutex);
 249         return error;
 250 }
 251 EXPORT_SYMBOL(freeze_bdev);
 252
 253 /**
 254  * thaw_bdev - unlock filesystem
 255  * @bdev:       blockdevice to unlock
 256  *
 257  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 258  */
 259 int thaw_bdev(struct block_device *bdev)
 260 {
 261         struct super_block *sb;
 262         int error = -EINVAL;
 263
 264         mutex_lock(&bdev->bd_fsfreeze_mutex);
 265         if (!bdev->bd_fsfreeze_count)
 266                 goto out;
 267
 268         error = 0;
 269         if (--bdev->bd_fsfreeze_count > 0)
 270                 goto out;
 271
 272         sb = bdev->bd_fsfreeze_sb;
 273         if (!sb)
 274                 goto out;
 275
 276         if (sb->s_op->thaw_super)
 277                 error = sb->s_op->thaw_super(sb, FREEZE_HOLDER_USERSPACE);
 278         else
 279                 error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
 280         if (error)
 281                 bdev->bd_fsfreeze_count++;
 282         else
 283                 bdev->bd_fsfreeze_sb = NULL;
 284 out:
 285         mutex_unlock(&bdev->bd_fsfreeze_mutex);
 286         return error;
 287 }
 288 EXPORT_SYMBOL(thaw_bdev);
 289
 290 /*
 291  * pseudo-fs
 292  */
 293
 294 static  __cacheline_aligned_in_smp DEFINE_MUTEX(bdev_lock);
 295 static struct kmem_cache * bdev_cachep __read_mostly;
 296
 297 static struct inode *bdev_alloc_inode(struct super_block *sb)
 298 {
 299         struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL);
 300
 301         if (!ei)
 302                 return NULL;
 303         memset(&ei->bdev, 0, sizeof(ei->bdev));
 304         return &ei->vfs_inode;
 305 }
 306
 307 static void bdev_free_inode(struct inode *inode)
 308 {
 309         struct block_device *bdev = I_BDEV(inode);
 310
 311         free_percpu(bdev->bd_stats);
 312         kfree(bdev->bd_meta_info);
 313
 314         if (!bdev_is_partition(bdev)) {
 315                 if (bdev->bd_disk && bdev->bd_disk->bdi)
 316                         bdi_put(bdev->bd_disk->bdi);
 317                 kfree(bdev->bd_disk);
 318         }
 319
 320         if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
 321                 blk_free_ext_minor(MINOR(bdev->bd_dev));
 322
 323         kmem_cache_free(bdev_cachep, BDEV_I(inode));
 324 }
 325
 326 static void init_once(void *data)
 327 {
 328         struct bdev_inode *ei = data;
 329
 330         inode_init_once(&ei->vfs_inode);
 331 }
 332
 333 static void bdev_evict_inode(struct inode *inode)
 334 {
 335         truncate_inode_pages_final(&inode->i_data);
 336         invalidate_inode_buffers(inode); /* is it needed here? */
 337         clear_inode(inode);
 338 }
 339
 340 static const struct super_operations bdev_sops = {
 341         .statfs = simple_statfs,
 342         .alloc_inode = bdev_alloc_inode,
 343         .free_inode = bdev_free_inode,
 344         .drop_inode = generic_delete_inode,
 345         .evict_inode = bdev_evict_inode,
 346 };
 347
 348 static int bd_init_fs_context(struct fs_context *fc)
 349 {
 350         struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
 351         if (!ctx)
 352                 return -ENOMEM;
 353         fc->s_iflags |= SB_I_CGROUPWB;
 354         ctx->ops = &bdev_sops;
 355         return 0;
 356 }
 357
 358 static struct file_system_type bd_type = {
 359         .name           = "bdev",
 360         .init_fs_context = bd_init_fs_context,
 361         .kill_sb        = kill_anon_super,
 362 };
 363
 364 struct super_block *blockdev_superblock __read_mostly;
 365 EXPORT_SYMBOL_GPL(blockdev_superblock);
 366
 367 void __init bdev_cache_init(void)
 368 {
 369         int err;
 370         static struct vfsmount *bd_mnt;
 371
 372         bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 373                         0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 374                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
 375                         init_once);
 376         err = register_filesystem(&bd_type);
 377         if (err)
 378                 panic("Cannot register bdev pseudo-fs");
 379         bd_mnt = kern_mount(&bd_type);
 380         if (IS_ERR(bd_mnt))
 381                 panic("Cannot create bdev pseudo-fs");
 382         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 383 }
 384
 385 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
 386 {
 387         struct block_device *bdev;
 388         struct inode *inode;
 389
 390         inode = new_inode(blockdev_superblock);
 391         if (!inode)
 392                 return NULL;
 393         inode->i_mode = S_IFBLK;
 394         inode->i_rdev = 0;
 395         inode->i_data.a_ops = &def_blk_aops;
 396         mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 397
 398         bdev = I_BDEV(inode);
 399         mutex_init(&bdev->bd_fsfreeze_mutex);
 400         spin_lock_init(&bdev->bd_size_lock);
 401         mutex_init(&bdev->bd_holder_lock);
 402         bdev->bd_partno = partno;
 403         bdev->bd_inode = inode;
 404         bdev->bd_queue = disk->queue;
 405         if (partno)
 406                 bdev->bd_has_submit_bio = disk->part0->bd_has_submit_bio;
 407         else
 408                 bdev->bd_has_submit_bio = false;
 409         bdev->bd_stats = alloc_percpu(struct disk_stats);
 410         if (!bdev->bd_stats) {
 411                 iput(inode);
 412                 return NULL;
 413         }
 414         bdev->bd_disk = disk;
 415         return bdev;
 416 }
 417
 418 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
 419 {
 420         spin_lock(&bdev->bd_size_lock);
 421         i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
 422         bdev->bd_nr_sectors = sectors;
 423         spin_unlock(&bdev->bd_size_lock);
 424 }
 425
 426 void bdev_add(struct block_device *bdev, dev_t dev)
 427 {
 428         bdev->bd_dev = dev;
 429         bdev->bd_inode->i_rdev = dev;
 430         bdev->bd_inode->i_ino = dev;
 431         insert_inode_hash(bdev->bd_inode);
 432 }
 433
 434 long nr_blockdev_pages(void)
 435 {
 436         struct inode *inode;
 437         long ret = 0;
 438
 439         spin_lock(&blockdev_superblock->s_inode_list_lock);
 440         list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
 441                 ret += inode->i_mapping->nrpages;
 442         spin_unlock(&blockdev_superblock->s_inode_list_lock);
 443
 444         return ret;
 445 }
 446
 447 /**
 448  * bd_may_claim - test whether a block device can be claimed
 449  * @bdev: block device of interest
 450  * @holder: holder trying to claim @bdev
 451  * @hops: holder ops
 452  *
 453  * Test whether @bdev can be claimed by @holder.
 454  *
 455  * RETURNS:
 456  * %true if @bdev can be claimed, %false otherwise.
 457  */
 458 static bool bd_may_claim(struct block_device *bdev, void *holder,
 459                 const struct blk_holder_ops *hops)
 460 {
 461         struct block_device *whole = bdev_whole(bdev);
 462
 463         lockdep_assert_held(&bdev_lock);
 464
 465         if (bdev->bd_holder) {
 466                 /*
 467                  * The same holder can always re-claim.
 468                  */
 469                 if (bdev->bd_holder == holder) {
 470                         if (WARN_ON_ONCE(bdev->bd_holder_ops != hops))
 471                                 return false;
 472                         return true;
 473                 }
 474                 return false;
 475         }
 476
 477         /*
 478          * If the whole devices holder is set to bd_may_claim, a partition on
 479          * the device is claimed, but not the whole device.
 480          */
 481         if (whole != bdev &&
 482             whole->bd_holder && whole->bd_holder != bd_may_claim)
 483                 return false;
 484         return true;
 485 }
 486
 487 /**
 488  * bd_prepare_to_claim - claim a block device
 489  * @bdev: block device of interest
 490  * @holder: holder trying to claim @bdev
 491  * @hops: holder ops.
 492  *
 493  * Claim @bdev.  This function fails if @bdev is already claimed by another
 494  * holder and waits if another claiming is in progress. return, the caller
 495  * has ownership of bd_claiming and bd_holder[s].
 496  *
 497  * RETURNS:
 498  * 0 if @bdev can be claimed, -EBUSY otherwise.
 499  */
 500 int bd_prepare_to_claim(struct block_device *bdev, void *holder,
 501                 const struct blk_holder_ops *hops)
 502 {
 503         struct block_device *whole = bdev_whole(bdev);
 504
 505         if (WARN_ON_ONCE(!holder))
 506                 return -EINVAL;
 507 retry:
 508         mutex_lock(&bdev_lock);
 509         /* if someone else claimed, fail */
 510         if (!bd_may_claim(bdev, holder, hops)) {
 511                 mutex_unlock(&bdev_lock);
 512                 return -EBUSY;
 513         }
 514
 515         /* if claiming is already in progress, wait for it to finish */
 516         if (whole->bd_claiming) {
 517                 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
 518                 DEFINE_WAIT(wait);
 519
 520                 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 521                 mutex_unlock(&bdev_lock);
 522                 schedule();
 523                 finish_wait(wq, &wait);
 524                 goto retry;
 525         }
 526
 527         /* yay, all mine */
 528         whole->bd_claiming = holder;
 529         mutex_unlock(&bdev_lock);
 530         return 0;
 531 }
 532 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
 533
 534 static void bd_clear_claiming(struct block_device *whole, void *holder)
 535 {
 536         lockdep_assert_held(&bdev_lock);
 537         /* tell others that we're done */
 538         BUG_ON(whole->bd_claiming != holder);
 539         whole->bd_claiming = NULL;
 540         wake_up_bit(&whole->bd_claiming, 0);
 541 }
 542
 543 /**
 544  * bd_finish_claiming - finish claiming of a block device
 545  * @bdev: block device of interest
 546  * @holder: holder that has claimed @bdev
 547  * @hops: block device holder operations
 548  *
 549  * Finish exclusive open of a block device. Mark the device as exlusively
 550  * open by the holder and wake up all waiters for exclusive open to finish.
 551  */
 552 static void bd_finish_claiming(struct block_device *bdev, void *holder,
 553                 const struct blk_holder_ops *hops)
 554 {
 555         struct block_device *whole = bdev_whole(bdev);
 556
 557         mutex_lock(&bdev_lock);
 558         BUG_ON(!bd_may_claim(bdev, holder, hops));
 559         /*
 560          * Note that for a whole device bd_holders will be incremented twice,
 561          * and bd_holder will be set to bd_may_claim before being set to holder
 562          */
 563         whole->bd_holders++;
 564         whole->bd_holder = bd_may_claim;
 565         bdev->bd_holders++;
 566         mutex_lock(&bdev->bd_holder_lock);
 567         bdev->bd_holder = holder;
 568         bdev->bd_holder_ops = hops;
 569         mutex_unlock(&bdev->bd_holder_lock);
 570         bd_clear_claiming(whole, holder);
 571         mutex_unlock(&bdev_lock);
 572 }
 573
 574 /**
 575  * bd_abort_claiming - abort claiming of a block device
 576  * @bdev: block device of interest
 577  * @holder: holder that has claimed @bdev
 578  *
 579  * Abort claiming of a block device when the exclusive open failed. This can be
 580  * also used when exclusive open is not actually desired and we just needed
 581  * to block other exclusive openers for a while.
 582  */
 583 void bd_abort_claiming(struct block_device *bdev, void *holder)
 584 {
 585         mutex_lock(&bdev_lock);
 586         bd_clear_claiming(bdev_whole(bdev), holder);
 587         mutex_unlock(&bdev_lock);
 588 }
 589 EXPORT_SYMBOL(bd_abort_claiming);
 590
 591 static void bd_end_claim(struct block_device *bdev, void *holder)
 592 {
 593         struct block_device *whole = bdev_whole(bdev);
 594         bool unblock = false;
 595
 596         /*
 597          * Release a claim on the device.  The holder fields are protected with
 598          * bdev_lock.  open_mutex is used to synchronize disk_holder unlinking.
 599          */
 600         mutex_lock(&bdev_lock);
 601         WARN_ON_ONCE(bdev->bd_holder != holder);
 602         WARN_ON_ONCE(--bdev->bd_holders < 0);
 603         WARN_ON_ONCE(--whole->bd_holders < 0);
 604         if (!bdev->bd_holders) {
 605                 mutex_lock(&bdev->bd_holder_lock);
 606                 bdev->bd_holder = NULL;
 607                 bdev->bd_holder_ops = NULL;
 608                 mutex_unlock(&bdev->bd_holder_lock);
 609                 if (bdev->bd_write_holder)
 610                         unblock = true;
 611         }
 612         if (!whole->bd_holders)
 613                 whole->bd_holder = NULL;
 614         mutex_unlock(&bdev_lock);
 615
 616         /*
 617          * If this was the last claim, remove holder link and unblock evpoll if
 618          * it was a write holder.
 619          */
 620         if (unblock) {
 621                 disk_unblock_events(bdev->bd_disk);
 622                 bdev->bd_write_holder = false;
 623         }
 624 }
 625
 626 static void blkdev_flush_mapping(struct block_device *bdev)
 627 {
 628         WARN_ON_ONCE(bdev->bd_holders);
 629         sync_blockdev(bdev);
 630         kill_bdev(bdev);
 631         bdev_write_inode(bdev);
 632 }
 633
 634 static int blkdev_get_whole(struct block_device *bdev, blk_mode_t mode)
 635 {
 636         struct gendisk *disk = bdev->bd_disk;
 637         int ret;
 638
 639         if (disk->fops->open) {
 640                 ret = disk->fops->open(disk, mode);
 641                 if (ret) {
 642                         /* avoid ghost partitions on a removed medium */
 643                         if (ret == -ENOMEDIUM &&
 644                              test_bit(GD_NEED_PART_SCAN, &disk->state))
 645                                 bdev_disk_changed(disk, true);
 646                         return ret;
 647                 }
 648         }
 649
 650         if (!atomic_read(&bdev->bd_openers))
 651                 set_init_blocksize(bdev);
 652         if (test_bit(GD_NEED_PART_SCAN, &disk->state))
 653                 bdev_disk_changed(disk, false);
 654         atomic_inc(&bdev->bd_openers);
 655         return 0;
 656 }
 657
 658 static void blkdev_put_whole(struct block_device *bdev)
 659 {
 660         if (atomic_dec_and_test(&bdev->bd_openers))
 661                 blkdev_flush_mapping(bdev);
 662         if (bdev->bd_disk->fops->release)
 663                 bdev->bd_disk->fops->release(bdev->bd_disk);
 664 }
 665
 666 static int blkdev_get_part(struct block_device *part, blk_mode_t mode)
 667 {
 668         struct gendisk *disk = part->bd_disk;
 669         int ret;
 670
 671         ret = blkdev_get_whole(bdev_whole(part), mode);
 672         if (ret)
 673                 return ret;
 674
 675         ret = -ENXIO;
 676         if (!bdev_nr_sectors(part))
 677                 goto out_blkdev_put;
 678
 679         if (!atomic_read(&part->bd_openers)) {
 680                 disk->open_partitions++;
 681                 set_init_blocksize(part);
 682         }
 683         atomic_inc(&part->bd_openers);
 684         return 0;
 685
 686 out_blkdev_put:
 687         blkdev_put_whole(bdev_whole(part));
 688         return ret;
 689 }
 690
 691 static void blkdev_put_part(struct block_device *part)
 692 {
 693         struct block_device *whole = bdev_whole(part);
 694
 695         if (atomic_dec_and_test(&part->bd_openers)) {
 696                 blkdev_flush_mapping(part);
 697                 whole->bd_disk->open_partitions--;
 698         }
 699         blkdev_put_whole(whole);
 700 }
 701
 702 struct block_device *blkdev_get_no_open(dev_t dev)
 703 {
 704         struct block_device *bdev;
 705         struct inode *inode;
 706
 707         inode = ilookup(blockdev_superblock, dev);
 708         if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) {
 709                 blk_request_module(dev);
 710                 inode = ilookup(blockdev_superblock, dev);
 711                 if (inode)
 712                         pr_warn_ratelimited(
 713 "block device autoloading is deprecated and will be removed.\n");
 714         }
 715         if (!inode)
 716                 return NULL;
 717
 718         /* switch from the inode reference to a device mode one: */
 719         bdev = &BDEV_I(inode)->bdev;
 720         if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
 721                 bdev = NULL;
 722         iput(inode);
 723         return bdev;
 724 }
 725
 726 void blkdev_put_no_open(struct block_device *bdev)
 727 {
 728         put_device(&bdev->bd_device);
 729 }
 730
 731 /**
 732  * blkdev_get_by_dev - open a block device by device number
 733  * @dev: device number of block device to open
 734  * @mode: open mode (BLK_OPEN_*)
 735  * @holder: exclusive holder identifier
 736  * @hops: holder operations
 737  *
 738  * Open the block device described by device number @dev. If @holder is not
 739  * %NULL, the block device is opened with exclusive access.  Exclusive opens may
 740  * nest for the same @holder.
 741  *
 742  * Use this interface ONLY if you really do not have anything better - i.e. when
 743  * you are behind a truly sucky interface and all you are given is a device
 744  * number.  Everything else should use blkdev_get_by_path().
 745  *
 746  * CONTEXT:
 747  * Might sleep.
 748  *
 749  * RETURNS:
 750  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
 751  */
 752 struct block_device *blkdev_get_by_dev(dev_t dev, blk_mode_t mode, void *holder,
 753                 const struct blk_holder_ops *hops)
 754 {
 755         bool unblock_events = true;
 756         struct block_device *bdev;
 757         struct gendisk *disk;
 758         int ret;
 759
 760         ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
 761                         MAJOR(dev), MINOR(dev),
 762                         ((mode & BLK_OPEN_READ) ? DEVCG_ACC_READ : 0) |
 763                         ((mode & BLK_OPEN_WRITE) ? DEVCG_ACC_WRITE : 0));
 764         if (ret)
 765                 return ERR_PTR(ret);
 766
 767         bdev = blkdev_get_no_open(dev);
 768         if (!bdev)
 769                 return ERR_PTR(-ENXIO);
 770         disk = bdev->bd_disk;
 771
 772         if (holder) {
 773                 mode |= BLK_OPEN_EXCL;
 774                 ret = bd_prepare_to_claim(bdev, holder, hops);
 775                 if (ret)
 776                         goto put_blkdev;
 777         } else {
 778                 if (WARN_ON_ONCE(mode & BLK_OPEN_EXCL)) {
 779                         ret = -EIO;
 780                         goto put_blkdev;
 781                 }
 782         }
 783
 784         disk_block_events(disk);
 785
 786         mutex_lock(&disk->open_mutex);
 787         ret = -ENXIO;
 788         if (!disk_live(disk))
 789                 goto abort_claiming;
 790         if (!try_module_get(disk->fops->owner))
 791                 goto abort_claiming;
 792         if (bdev_is_partition(bdev))
 793                 ret = blkdev_get_part(bdev, mode);
 794         else
 795                 ret = blkdev_get_whole(bdev, mode);
 796         if (ret)
 797                 goto put_module;
 798         if (holder) {
 799                 bd_finish_claiming(bdev, holder, hops);
 800
 801                 /*
 802                  * Block event polling for write claims if requested.  Any write
 803                  * holder makes the write_holder state stick until all are
 804                  * released.  This is good enough and tracking individual
 805                  * writeable reference is too fragile given the way @mode is
 806                  * used in blkdev_get/put().
 807                  */
 808                 if ((mode & BLK_OPEN_WRITE) && !bdev->bd_write_holder &&
 809                     (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) {
 810                         bdev->bd_write_holder = true;
 811                         unblock_events = false;
 812                 }
 813         }
 814         mutex_unlock(&disk->open_mutex);
 815
 816         if (unblock_events)
 817                 disk_unblock_events(disk);
 818         return bdev;
 819 put_module:
 820         module_put(disk->fops->owner);
 821 abort_claiming:
 822         if (holder)
 823                 bd_abort_claiming(bdev, holder);
 824         mutex_unlock(&disk->open_mutex);
 825         disk_unblock_events(disk);
 826 put_blkdev:
 827         blkdev_put_no_open(bdev);
 828         return ERR_PTR(ret);
 829 }
 830 EXPORT_SYMBOL(blkdev_get_by_dev);
 831
 832 /**
 833  * blkdev_get_by_path - open a block device by name
 834  * @path: path to the block device to open
 835  * @mode: open mode (BLK_OPEN_*)
 836  * @holder: exclusive holder identifier
 837  * @hops: holder operations
 838  *
 839  * Open the block device described by the device file at @path.  If @holder is
 840  * not %NULL, the block device is opened with exclusive access.  Exclusive opens
 841  * may nest for the same @holder.
 842  *
 843  * CONTEXT:
 844  * Might sleep.
 845  *
 846  * RETURNS:
 847  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
 848  */
 849 struct block_device *blkdev_get_by_path(const char *path, blk_mode_t mode,
 850                 void *holder, const struct blk_holder_ops *hops)
 851 {
 852         struct block_device *bdev;
 853         dev_t dev;
 854         int error;
 855
 856         error = lookup_bdev(path, &dev);
 857         if (error)
 858                 return ERR_PTR(error);
 859
 860         bdev = blkdev_get_by_dev(dev, mode, holder, hops);
 861         if (!IS_ERR(bdev) && (mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
 862                 blkdev_put(bdev, holder);
 863                 return ERR_PTR(-EACCES);
 864         }
 865
 866         return bdev;
 867 }
 868 EXPORT_SYMBOL(blkdev_get_by_path);
 869
 870 void blkdev_put(struct block_device *bdev, void *holder)
 871 {
 872         struct gendisk *disk = bdev->bd_disk;
 873
 874         /*
 875          * Sync early if it looks like we're the last one.  If someone else
 876          * opens the block device between now and the decrement of bd_openers
 877          * then we did a sync that we didn't need to, but that's not the end
 878          * of the world and we want to avoid long (could be several minute)
 879          * syncs while holding the mutex.
 880          */
 881         if (atomic_read(&bdev->bd_openers) == 1)
 882                 sync_blockdev(bdev);
 883
 884         mutex_lock(&disk->open_mutex);
 885         if (holder)
 886                 bd_end_claim(bdev, holder);
 887
 888         /*
 889          * Trigger event checking and tell drivers to flush MEDIA_CHANGE
 890          * event.  This is to ensure detection of media removal commanded
 891          * from userland - e.g. eject(1).
 892          */
 893         disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
 894
 895         if (bdev_is_partition(bdev))
 896                 blkdev_put_part(bdev);
 897         else
 898                 blkdev_put_whole(bdev);
 899         mutex_unlock(&disk->open_mutex);
 900
 901         module_put(disk->fops->owner);
 902         blkdev_put_no_open(bdev);
 903 }
 904 EXPORT_SYMBOL(blkdev_put);
 905
 906 /**
 907  * lookup_bdev() - Look up a struct block_device by name.
 908  * @pathname: Name of the block device in the filesystem.
 909  * @dev: Pointer to the block device's dev_t, if found.
 910  *
 911  * Lookup the block device's dev_t at @pathname in the current
 912  * namespace if possible and return it in @dev.
 913  *
 914  * Context: May sleep.
 915  * Return: 0 if succeeded, negative errno otherwise.
 916  */
 917 int lookup_bdev(const char *pathname, dev_t *dev)
 918 {
 919         struct inode *inode;
 920         struct path path;
 921         int error;
 922
 923         if (!pathname || !*pathname)
 924                 return -EINVAL;
 925
 926         error = kern_path(pathname, LOOKUP_FOLLOW, &path);
 927         if (error)
 928                 return error;
 929
 930         inode = d_backing_inode(path.dentry);
 931         error = -ENOTBLK;
 932         if (!S_ISBLK(inode->i_mode))
 933                 goto out_path_put;
 934         error = -EACCES;
 935         if (!may_open_dev(&path))
 936                 goto out_path_put;
 937
 938         *dev = inode->i_rdev;
 939         error = 0;
 940 out_path_put:
 941         path_put(&path);
 942         return error;
 943 }
 944 EXPORT_SYMBOL(lookup_bdev);
 945
 946 /**
 947  * bdev_mark_dead - mark a block device as dead
 948  * @bdev: block device to operate on
 949  * @surprise: indicate a surprise removal
 950  *
 951  * Tell the file system that this devices or media is dead.  If @surprise is set
 952  * to %true the device or media is already gone, if not we are preparing for an
 953  * orderly removal.
 954  *
 955  * This calls into the file system, which then typicall syncs out all dirty data
 956  * and writes back inodes and then invalidates any cached data in the inodes on
 957  * the file system.  In addition we also invalidate the block device mapping.
 958  */
 959 void bdev_mark_dead(struct block_device *bdev, bool surprise)
 960 {
 961         mutex_lock(&bdev->bd_holder_lock);
 962         if (bdev->bd_holder_ops && bdev->bd_holder_ops->mark_dead)
 963                 bdev->bd_holder_ops->mark_dead(bdev, surprise);
 964         else
 965                 sync_blockdev(bdev);
 966         mutex_unlock(&bdev->bd_holder_lock);
 967
 968         invalidate_bdev(bdev);
 969 }
 970 #ifdef CONFIG_DASD_MODULE
 971 /*
 972  * Drivers should not use this directly, but the DASD driver has historically
 973  * had a shutdown to offline mode that doesn't actually remove the gendisk
 974  * that otherwise looks a lot like a safe device removal.
 975  */
 976 EXPORT_SYMBOL_GPL(bdev_mark_dead);
 977 #endif
 978
 979 void sync_bdevs(bool wait)
 980 {
 981         struct inode *inode, *old_inode = NULL;
 982
 983         spin_lock(&blockdev_superblock->s_inode_list_lock);
 984         list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
 985                 struct address_space *mapping = inode->i_mapping;
 986                 struct block_device *bdev;
 987
 988                 spin_lock(&inode->i_lock);
 989                 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
 990                     mapping->nrpages == 0) {
 991                         spin_unlock(&inode->i_lock);
 992                         continue;
 993                 }
 994                 __iget(inode);
 995                 spin_unlock(&inode->i_lock);
 996                 spin_unlock(&blockdev_superblock->s_inode_list_lock);
 997                 /*
 998                  * We hold a reference to 'inode' so it couldn't have been
 999                  * removed from s_inodes list while we dropped the
1000                  * s_inode_list_lock  We cannot iput the inode now as we can
1001                  * be holding the last reference and we cannot iput it under
1002                  * s_inode_list_lock. So we keep the reference and iput it
1003                  * later.
1004                  */
1005                 iput(old_inode);
1006                 old_inode = inode;
1007                 bdev = I_BDEV(inode);
1008
1009                 mutex_lock(&bdev->bd_disk->open_mutex);
1010                 if (!atomic_read(&bdev->bd_openers)) {
1011                         ; /* skip */
1012                 } else if (wait) {
1013                         /*
1014                          * We keep the error status of individual mapping so
1015                          * that applications can catch the writeback error using
1016                          * fsync(2). See filemap_fdatawait_keep_errors() for
1017                          * details.
1018                          */
1019                         filemap_fdatawait_keep_errors(inode->i_mapping);
1020                 } else {
1021                         filemap_fdatawrite(inode->i_mapping);
1022                 }
1023                 mutex_unlock(&bdev->bd_disk->open_mutex);
1024
1025                 spin_lock(&blockdev_superblock->s_inode_list_lock);
1026         }
1027         spin_unlock(&blockdev_superblock->s_inode_list_lock);
1028         iput(old_inode);
1029 }
1030
1031 /*
1032  * Handle STATX_DIOALIGN for block devices.
1033  *
1034  * Note that the inode passed to this is the inode of a block device node file,
1035  * not the block device's internal inode.  Therefore it is *not* valid to use
1036  * I_BDEV() here; the block device has to be looked up by i_rdev instead.
1037  */
1038 void bdev_statx_dioalign(struct inode *inode, struct kstat *stat)
1039 {
1040         struct block_device *bdev;
1041
1042         bdev = blkdev_get_no_open(inode->i_rdev);
1043         if (!bdev)
1044                 return;
1045
1046         stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
1047         stat->dio_offset_align = bdev_logical_block_size(bdev);
1048         stat->result_mask |= STATX_DIOALIGN;
1049
1050         blkdev_put_no_open(bdev);
1051 }