| 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, fmode_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 & FMODE_EXCL)) { |
| 105 | int err = bd_prepare_to_claim(bdev, truncate_bdev_range); |
| 106 | if (err) |
| 107 | goto invalidate; |
| 108 | } |
| 109 | |
| 110 | truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend); |
| 111 | if (!(mode & FMODE_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 | * Write out and wait upon all dirty data associated with this |
| 211 | * device. Filesystem data as well as the underlying block |
| 212 | * device. Takes the superblock lock. |
| 213 | */ |
| 214 | int fsync_bdev(struct block_device *bdev) |
| 215 | { |
| 216 | struct super_block *sb = get_super(bdev); |
| 217 | if (sb) { |
| 218 | int res = sync_filesystem(sb); |
| 219 | drop_super(sb); |
| 220 | return res; |
| 221 | } |
| 222 | return sync_blockdev(bdev); |
| 223 | } |
| 224 | EXPORT_SYMBOL(fsync_bdev); |
| 225 | |
| 226 | /** |
| 227 | * freeze_bdev - lock a filesystem and force it into a consistent state |
| 228 | * @bdev: blockdevice to lock |
| 229 | * |
| 230 | * If a superblock is found on this device, we take the s_umount semaphore |
| 231 | * on it to make sure nobody unmounts until the snapshot creation is done. |
| 232 | * The reference counter (bd_fsfreeze_count) guarantees that only the last |
| 233 | * unfreeze process can unfreeze the frozen filesystem actually when multiple |
| 234 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and |
| 235 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze |
| 236 | * actually. |
| 237 | */ |
| 238 | int freeze_bdev(struct block_device *bdev) |
| 239 | { |
| 240 | struct super_block *sb; |
| 241 | int error = 0; |
| 242 | |
| 243 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
| 244 | if (++bdev->bd_fsfreeze_count > 1) |
| 245 | goto done; |
| 246 | |
| 247 | sb = get_active_super(bdev); |
| 248 | if (!sb) |
| 249 | goto sync; |
| 250 | if (sb->s_op->freeze_super) |
| 251 | error = sb->s_op->freeze_super(sb); |
| 252 | else |
| 253 | error = freeze_super(sb); |
| 254 | deactivate_super(sb); |
| 255 | |
| 256 | if (error) { |
| 257 | bdev->bd_fsfreeze_count--; |
| 258 | goto done; |
| 259 | } |
| 260 | bdev->bd_fsfreeze_sb = sb; |
| 261 | |
| 262 | sync: |
| 263 | sync_blockdev(bdev); |
| 264 | done: |
| 265 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 266 | return error; |
| 267 | } |
| 268 | EXPORT_SYMBOL(freeze_bdev); |
| 269 | |
| 270 | /** |
| 271 | * thaw_bdev - unlock filesystem |
| 272 | * @bdev: blockdevice to unlock |
| 273 | * |
| 274 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). |
| 275 | */ |
| 276 | int thaw_bdev(struct block_device *bdev) |
| 277 | { |
| 278 | struct super_block *sb; |
| 279 | int error = -EINVAL; |
| 280 | |
| 281 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
| 282 | if (!bdev->bd_fsfreeze_count) |
| 283 | goto out; |
| 284 | |
| 285 | error = 0; |
| 286 | if (--bdev->bd_fsfreeze_count > 0) |
| 287 | goto out; |
| 288 | |
| 289 | sb = bdev->bd_fsfreeze_sb; |
| 290 | if (!sb) |
| 291 | goto out; |
| 292 | |
| 293 | if (sb->s_op->thaw_super) |
| 294 | error = sb->s_op->thaw_super(sb); |
| 295 | else |
| 296 | error = thaw_super(sb); |
| 297 | if (error) |
| 298 | bdev->bd_fsfreeze_count++; |
| 299 | else |
| 300 | bdev->bd_fsfreeze_sb = NULL; |
| 301 | out: |
| 302 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 303 | return error; |
| 304 | } |
| 305 | EXPORT_SYMBOL(thaw_bdev); |
| 306 | |
| 307 | /** |
| 308 | * bdev_read_page() - Start reading a page from a block device |
| 309 | * @bdev: The device to read the page from |
| 310 | * @sector: The offset on the device to read the page to (need not be aligned) |
| 311 | * @page: The page to read |
| 312 | * |
| 313 | * On entry, the page should be locked. It will be unlocked when the page |
| 314 | * has been read. If the block driver implements rw_page synchronously, |
| 315 | * that will be true on exit from this function, but it need not be. |
| 316 | * |
| 317 | * Errors returned by this function are usually "soft", eg out of memory, or |
| 318 | * queue full; callers should try a different route to read this page rather |
| 319 | * than propagate an error back up the stack. |
| 320 | * |
| 321 | * Return: negative errno if an error occurs, 0 if submission was successful. |
| 322 | */ |
| 323 | int bdev_read_page(struct block_device *bdev, sector_t sector, |
| 324 | struct page *page) |
| 325 | { |
| 326 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
| 327 | int result = -EOPNOTSUPP; |
| 328 | |
| 329 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
| 330 | return result; |
| 331 | |
| 332 | result = blk_queue_enter(bdev_get_queue(bdev), 0); |
| 333 | if (result) |
| 334 | return result; |
| 335 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
| 336 | REQ_OP_READ); |
| 337 | blk_queue_exit(bdev_get_queue(bdev)); |
| 338 | return result; |
| 339 | } |
| 340 | |
| 341 | /** |
| 342 | * bdev_write_page() - Start writing a page to a block device |
| 343 | * @bdev: The device to write the page to |
| 344 | * @sector: The offset on the device to write the page to (need not be aligned) |
| 345 | * @page: The page to write |
| 346 | * @wbc: The writeback_control for the write |
| 347 | * |
| 348 | * On entry, the page should be locked and not currently under writeback. |
| 349 | * On exit, if the write started successfully, the page will be unlocked and |
| 350 | * under writeback. If the write failed already (eg the driver failed to |
| 351 | * queue the page to the device), the page will still be locked. If the |
| 352 | * caller is a ->writepage implementation, it will need to unlock the page. |
| 353 | * |
| 354 | * Errors returned by this function are usually "soft", eg out of memory, or |
| 355 | * queue full; callers should try a different route to write this page rather |
| 356 | * than propagate an error back up the stack. |
| 357 | * |
| 358 | * Return: negative errno if an error occurs, 0 if submission was successful. |
| 359 | */ |
| 360 | int bdev_write_page(struct block_device *bdev, sector_t sector, |
| 361 | struct page *page, struct writeback_control *wbc) |
| 362 | { |
| 363 | int result; |
| 364 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
| 365 | |
| 366 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
| 367 | return -EOPNOTSUPP; |
| 368 | result = blk_queue_enter(bdev_get_queue(bdev), 0); |
| 369 | if (result) |
| 370 | return result; |
| 371 | |
| 372 | set_page_writeback(page); |
| 373 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
| 374 | REQ_OP_WRITE); |
| 375 | if (result) { |
| 376 | end_page_writeback(page); |
| 377 | } else { |
| 378 | clean_page_buffers(page); |
| 379 | unlock_page(page); |
| 380 | } |
| 381 | blk_queue_exit(bdev_get_queue(bdev)); |
| 382 | return result; |
| 383 | } |
| 384 | |
| 385 | /* |
| 386 | * pseudo-fs |
| 387 | */ |
| 388 | |
| 389 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); |
| 390 | static struct kmem_cache * bdev_cachep __read_mostly; |
| 391 | |
| 392 | static struct inode *bdev_alloc_inode(struct super_block *sb) |
| 393 | { |
| 394 | struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL); |
| 395 | |
| 396 | if (!ei) |
| 397 | return NULL; |
| 398 | memset(&ei->bdev, 0, sizeof(ei->bdev)); |
| 399 | return &ei->vfs_inode; |
| 400 | } |
| 401 | |
| 402 | static void bdev_free_inode(struct inode *inode) |
| 403 | { |
| 404 | struct block_device *bdev = I_BDEV(inode); |
| 405 | |
| 406 | free_percpu(bdev->bd_stats); |
| 407 | kfree(bdev->bd_meta_info); |
| 408 | |
| 409 | if (!bdev_is_partition(bdev)) { |
| 410 | if (bdev->bd_disk && bdev->bd_disk->bdi) |
| 411 | bdi_put(bdev->bd_disk->bdi); |
| 412 | kfree(bdev->bd_disk); |
| 413 | } |
| 414 | |
| 415 | if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR) |
| 416 | blk_free_ext_minor(MINOR(bdev->bd_dev)); |
| 417 | |
| 418 | kmem_cache_free(bdev_cachep, BDEV_I(inode)); |
| 419 | } |
| 420 | |
| 421 | static void init_once(void *data) |
| 422 | { |
| 423 | struct bdev_inode *ei = data; |
| 424 | |
| 425 | inode_init_once(&ei->vfs_inode); |
| 426 | } |
| 427 | |
| 428 | static void bdev_evict_inode(struct inode *inode) |
| 429 | { |
| 430 | truncate_inode_pages_final(&inode->i_data); |
| 431 | invalidate_inode_buffers(inode); /* is it needed here? */ |
| 432 | clear_inode(inode); |
| 433 | } |
| 434 | |
| 435 | static const struct super_operations bdev_sops = { |
| 436 | .statfs = simple_statfs, |
| 437 | .alloc_inode = bdev_alloc_inode, |
| 438 | .free_inode = bdev_free_inode, |
| 439 | .drop_inode = generic_delete_inode, |
| 440 | .evict_inode = bdev_evict_inode, |
| 441 | }; |
| 442 | |
| 443 | static int bd_init_fs_context(struct fs_context *fc) |
| 444 | { |
| 445 | struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC); |
| 446 | if (!ctx) |
| 447 | return -ENOMEM; |
| 448 | fc->s_iflags |= SB_I_CGROUPWB; |
| 449 | ctx->ops = &bdev_sops; |
| 450 | return 0; |
| 451 | } |
| 452 | |
| 453 | static struct file_system_type bd_type = { |
| 454 | .name = "bdev", |
| 455 | .init_fs_context = bd_init_fs_context, |
| 456 | .kill_sb = kill_anon_super, |
| 457 | }; |
| 458 | |
| 459 | struct super_block *blockdev_superblock __read_mostly; |
| 460 | EXPORT_SYMBOL_GPL(blockdev_superblock); |
| 461 | |
| 462 | void __init bdev_cache_init(void) |
| 463 | { |
| 464 | int err; |
| 465 | static struct vfsmount *bd_mnt; |
| 466 | |
| 467 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
| 468 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
| 469 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
| 470 | init_once); |
| 471 | err = register_filesystem(&bd_type); |
| 472 | if (err) |
| 473 | panic("Cannot register bdev pseudo-fs"); |
| 474 | bd_mnt = kern_mount(&bd_type); |
| 475 | if (IS_ERR(bd_mnt)) |
| 476 | panic("Cannot create bdev pseudo-fs"); |
| 477 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
| 478 | } |
| 479 | |
| 480 | struct block_device *bdev_alloc(struct gendisk *disk, u8 partno) |
| 481 | { |
| 482 | struct block_device *bdev; |
| 483 | struct inode *inode; |
| 484 | |
| 485 | inode = new_inode(blockdev_superblock); |
| 486 | if (!inode) |
| 487 | return NULL; |
| 488 | inode->i_mode = S_IFBLK; |
| 489 | inode->i_rdev = 0; |
| 490 | inode->i_data.a_ops = &def_blk_aops; |
| 491 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); |
| 492 | |
| 493 | bdev = I_BDEV(inode); |
| 494 | mutex_init(&bdev->bd_fsfreeze_mutex); |
| 495 | spin_lock_init(&bdev->bd_size_lock); |
| 496 | bdev->bd_partno = partno; |
| 497 | bdev->bd_inode = inode; |
| 498 | bdev->bd_queue = disk->queue; |
| 499 | bdev->bd_stats = alloc_percpu(struct disk_stats); |
| 500 | if (!bdev->bd_stats) { |
| 501 | iput(inode); |
| 502 | return NULL; |
| 503 | } |
| 504 | bdev->bd_disk = disk; |
| 505 | return bdev; |
| 506 | } |
| 507 | |
| 508 | void bdev_add(struct block_device *bdev, dev_t dev) |
| 509 | { |
| 510 | bdev->bd_dev = dev; |
| 511 | bdev->bd_inode->i_rdev = dev; |
| 512 | bdev->bd_inode->i_ino = dev; |
| 513 | insert_inode_hash(bdev->bd_inode); |
| 514 | } |
| 515 | |
| 516 | long nr_blockdev_pages(void) |
| 517 | { |
| 518 | struct inode *inode; |
| 519 | long ret = 0; |
| 520 | |
| 521 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
| 522 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) |
| 523 | ret += inode->i_mapping->nrpages; |
| 524 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
| 525 | |
| 526 | return ret; |
| 527 | } |
| 528 | |
| 529 | /** |
| 530 | * bd_may_claim - test whether a block device can be claimed |
| 531 | * @bdev: block device of interest |
| 532 | * @whole: whole block device containing @bdev, may equal @bdev |
| 533 | * @holder: holder trying to claim @bdev |
| 534 | * |
| 535 | * Test whether @bdev can be claimed by @holder. |
| 536 | * |
| 537 | * CONTEXT: |
| 538 | * spin_lock(&bdev_lock). |
| 539 | * |
| 540 | * RETURNS: |
| 541 | * %true if @bdev can be claimed, %false otherwise. |
| 542 | */ |
| 543 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, |
| 544 | void *holder) |
| 545 | { |
| 546 | if (bdev->bd_holder == holder) |
| 547 | return true; /* already a holder */ |
| 548 | else if (bdev->bd_holder != NULL) |
| 549 | return false; /* held by someone else */ |
| 550 | else if (whole == bdev) |
| 551 | return true; /* is a whole device which isn't held */ |
| 552 | |
| 553 | else if (whole->bd_holder == bd_may_claim) |
| 554 | return true; /* is a partition of a device that is being partitioned */ |
| 555 | else if (whole->bd_holder != NULL) |
| 556 | return false; /* is a partition of a held device */ |
| 557 | else |
| 558 | return true; /* is a partition of an un-held device */ |
| 559 | } |
| 560 | |
| 561 | /** |
| 562 | * bd_prepare_to_claim - claim a block device |
| 563 | * @bdev: block device of interest |
| 564 | * @holder: holder trying to claim @bdev |
| 565 | * |
| 566 | * Claim @bdev. This function fails if @bdev is already claimed by another |
| 567 | * holder and waits if another claiming is in progress. return, the caller |
| 568 | * has ownership of bd_claiming and bd_holder[s]. |
| 569 | * |
| 570 | * RETURNS: |
| 571 | * 0 if @bdev can be claimed, -EBUSY otherwise. |
| 572 | */ |
| 573 | int bd_prepare_to_claim(struct block_device *bdev, void *holder) |
| 574 | { |
| 575 | struct block_device *whole = bdev_whole(bdev); |
| 576 | |
| 577 | if (WARN_ON_ONCE(!holder)) |
| 578 | return -EINVAL; |
| 579 | retry: |
| 580 | spin_lock(&bdev_lock); |
| 581 | /* if someone else claimed, fail */ |
| 582 | if (!bd_may_claim(bdev, whole, holder)) { |
| 583 | spin_unlock(&bdev_lock); |
| 584 | return -EBUSY; |
| 585 | } |
| 586 | |
| 587 | /* if claiming is already in progress, wait for it to finish */ |
| 588 | if (whole->bd_claiming) { |
| 589 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
| 590 | DEFINE_WAIT(wait); |
| 591 | |
| 592 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); |
| 593 | spin_unlock(&bdev_lock); |
| 594 | schedule(); |
| 595 | finish_wait(wq, &wait); |
| 596 | goto retry; |
| 597 | } |
| 598 | |
| 599 | /* yay, all mine */ |
| 600 | whole->bd_claiming = holder; |
| 601 | spin_unlock(&bdev_lock); |
| 602 | return 0; |
| 603 | } |
| 604 | EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */ |
| 605 | |
| 606 | static void bd_clear_claiming(struct block_device *whole, void *holder) |
| 607 | { |
| 608 | lockdep_assert_held(&bdev_lock); |
| 609 | /* tell others that we're done */ |
| 610 | BUG_ON(whole->bd_claiming != holder); |
| 611 | whole->bd_claiming = NULL; |
| 612 | wake_up_bit(&whole->bd_claiming, 0); |
| 613 | } |
| 614 | |
| 615 | /** |
| 616 | * bd_finish_claiming - finish claiming of a block device |
| 617 | * @bdev: block device of interest |
| 618 | * @holder: holder that has claimed @bdev |
| 619 | * |
| 620 | * Finish exclusive open of a block device. Mark the device as exlusively |
| 621 | * open by the holder and wake up all waiters for exclusive open to finish. |
| 622 | */ |
| 623 | static void bd_finish_claiming(struct block_device *bdev, void *holder) |
| 624 | { |
| 625 | struct block_device *whole = bdev_whole(bdev); |
| 626 | |
| 627 | spin_lock(&bdev_lock); |
| 628 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
| 629 | /* |
| 630 | * Note that for a whole device bd_holders will be incremented twice, |
| 631 | * and bd_holder will be set to bd_may_claim before being set to holder |
| 632 | */ |
| 633 | whole->bd_holders++; |
| 634 | whole->bd_holder = bd_may_claim; |
| 635 | bdev->bd_holders++; |
| 636 | bdev->bd_holder = holder; |
| 637 | bd_clear_claiming(whole, holder); |
| 638 | spin_unlock(&bdev_lock); |
| 639 | } |
| 640 | |
| 641 | /** |
| 642 | * bd_abort_claiming - abort claiming of a block device |
| 643 | * @bdev: block device of interest |
| 644 | * @holder: holder that has claimed @bdev |
| 645 | * |
| 646 | * Abort claiming of a block device when the exclusive open failed. This can be |
| 647 | * also used when exclusive open is not actually desired and we just needed |
| 648 | * to block other exclusive openers for a while. |
| 649 | */ |
| 650 | void bd_abort_claiming(struct block_device *bdev, void *holder) |
| 651 | { |
| 652 | spin_lock(&bdev_lock); |
| 653 | bd_clear_claiming(bdev_whole(bdev), holder); |
| 654 | spin_unlock(&bdev_lock); |
| 655 | } |
| 656 | EXPORT_SYMBOL(bd_abort_claiming); |
| 657 | |
| 658 | static void blkdev_flush_mapping(struct block_device *bdev) |
| 659 | { |
| 660 | WARN_ON_ONCE(bdev->bd_holders); |
| 661 | sync_blockdev(bdev); |
| 662 | kill_bdev(bdev); |
| 663 | bdev_write_inode(bdev); |
| 664 | } |
| 665 | |
| 666 | static int blkdev_get_whole(struct block_device *bdev, fmode_t mode) |
| 667 | { |
| 668 | struct gendisk *disk = bdev->bd_disk; |
| 669 | int ret; |
| 670 | |
| 671 | if (disk->fops->open) { |
| 672 | ret = disk->fops->open(bdev, mode); |
| 673 | if (ret) { |
| 674 | /* avoid ghost partitions on a removed medium */ |
| 675 | if (ret == -ENOMEDIUM && |
| 676 | test_bit(GD_NEED_PART_SCAN, &disk->state)) |
| 677 | bdev_disk_changed(disk, true); |
| 678 | return ret; |
| 679 | } |
| 680 | } |
| 681 | |
| 682 | if (!atomic_read(&bdev->bd_openers)) |
| 683 | set_init_blocksize(bdev); |
| 684 | if (test_bit(GD_NEED_PART_SCAN, &disk->state)) |
| 685 | bdev_disk_changed(disk, false); |
| 686 | atomic_inc(&bdev->bd_openers); |
| 687 | return 0; |
| 688 | } |
| 689 | |
| 690 | static void blkdev_put_whole(struct block_device *bdev, fmode_t mode) |
| 691 | { |
| 692 | if (atomic_dec_and_test(&bdev->bd_openers)) |
| 693 | blkdev_flush_mapping(bdev); |
| 694 | if (bdev->bd_disk->fops->release) |
| 695 | bdev->bd_disk->fops->release(bdev->bd_disk, mode); |
| 696 | } |
| 697 | |
| 698 | static int blkdev_get_part(struct block_device *part, fmode_t mode) |
| 699 | { |
| 700 | struct gendisk *disk = part->bd_disk; |
| 701 | int ret; |
| 702 | |
| 703 | if (atomic_read(&part->bd_openers)) |
| 704 | goto done; |
| 705 | |
| 706 | ret = blkdev_get_whole(bdev_whole(part), mode); |
| 707 | if (ret) |
| 708 | return ret; |
| 709 | |
| 710 | ret = -ENXIO; |
| 711 | if (!bdev_nr_sectors(part)) |
| 712 | goto out_blkdev_put; |
| 713 | |
| 714 | disk->open_partitions++; |
| 715 | set_init_blocksize(part); |
| 716 | done: |
| 717 | atomic_inc(&part->bd_openers); |
| 718 | return 0; |
| 719 | |
| 720 | out_blkdev_put: |
| 721 | blkdev_put_whole(bdev_whole(part), mode); |
| 722 | return ret; |
| 723 | } |
| 724 | |
| 725 | static void blkdev_put_part(struct block_device *part, fmode_t mode) |
| 726 | { |
| 727 | struct block_device *whole = bdev_whole(part); |
| 728 | |
| 729 | if (!atomic_dec_and_test(&part->bd_openers)) |
| 730 | return; |
| 731 | blkdev_flush_mapping(part); |
| 732 | whole->bd_disk->open_partitions--; |
| 733 | blkdev_put_whole(whole, mode); |
| 734 | } |
| 735 | |
| 736 | struct block_device *blkdev_get_no_open(dev_t dev) |
| 737 | { |
| 738 | struct block_device *bdev; |
| 739 | struct inode *inode; |
| 740 | |
| 741 | inode = ilookup(blockdev_superblock, dev); |
| 742 | if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) { |
| 743 | blk_request_module(dev); |
| 744 | inode = ilookup(blockdev_superblock, dev); |
| 745 | if (inode) |
| 746 | pr_warn_ratelimited( |
| 747 | "block device autoloading is deprecated and will be removed.\n"); |
| 748 | } |
| 749 | if (!inode) |
| 750 | return NULL; |
| 751 | |
| 752 | /* switch from the inode reference to a device mode one: */ |
| 753 | bdev = &BDEV_I(inode)->bdev; |
| 754 | if (!kobject_get_unless_zero(&bdev->bd_device.kobj)) |
| 755 | bdev = NULL; |
| 756 | iput(inode); |
| 757 | return bdev; |
| 758 | } |
| 759 | |
| 760 | void blkdev_put_no_open(struct block_device *bdev) |
| 761 | { |
| 762 | put_device(&bdev->bd_device); |
| 763 | } |
| 764 | |
| 765 | /** |
| 766 | * blkdev_get_by_dev - open a block device by device number |
| 767 | * @dev: device number of block device to open |
| 768 | * @mode: FMODE_* mask |
| 769 | * @holder: exclusive holder identifier |
| 770 | * |
| 771 | * Open the block device described by device number @dev. If @mode includes |
| 772 | * %FMODE_EXCL, the block device is opened with exclusive access. Specifying |
| 773 | * %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may nest for |
| 774 | * the same @holder. |
| 775 | * |
| 776 | * Use this interface ONLY if you really do not have anything better - i.e. when |
| 777 | * you are behind a truly sucky interface and all you are given is a device |
| 778 | * number. Everything else should use blkdev_get_by_path(). |
| 779 | * |
| 780 | * CONTEXT: |
| 781 | * Might sleep. |
| 782 | * |
| 783 | * RETURNS: |
| 784 | * Reference to the block_device on success, ERR_PTR(-errno) on failure. |
| 785 | */ |
| 786 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) |
| 787 | { |
| 788 | bool unblock_events = true; |
| 789 | struct block_device *bdev; |
| 790 | struct gendisk *disk; |
| 791 | int ret; |
| 792 | |
| 793 | ret = devcgroup_check_permission(DEVCG_DEV_BLOCK, |
| 794 | MAJOR(dev), MINOR(dev), |
| 795 | ((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) | |
| 796 | ((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0)); |
| 797 | if (ret) |
| 798 | return ERR_PTR(ret); |
| 799 | |
| 800 | bdev = blkdev_get_no_open(dev); |
| 801 | if (!bdev) |
| 802 | return ERR_PTR(-ENXIO); |
| 803 | disk = bdev->bd_disk; |
| 804 | |
| 805 | if (mode & FMODE_EXCL) { |
| 806 | ret = bd_prepare_to_claim(bdev, holder); |
| 807 | if (ret) |
| 808 | goto put_blkdev; |
| 809 | } |
| 810 | |
| 811 | disk_block_events(disk); |
| 812 | |
| 813 | mutex_lock(&disk->open_mutex); |
| 814 | ret = -ENXIO; |
| 815 | if (!disk_live(disk)) |
| 816 | goto abort_claiming; |
| 817 | if (!try_module_get(disk->fops->owner)) |
| 818 | goto abort_claiming; |
| 819 | if (bdev_is_partition(bdev)) |
| 820 | ret = blkdev_get_part(bdev, mode); |
| 821 | else |
| 822 | ret = blkdev_get_whole(bdev, mode); |
| 823 | if (ret) |
| 824 | goto put_module; |
| 825 | if (mode & FMODE_EXCL) { |
| 826 | bd_finish_claiming(bdev, holder); |
| 827 | |
| 828 | /* |
| 829 | * Block event polling for write claims if requested. Any write |
| 830 | * holder makes the write_holder state stick until all are |
| 831 | * released. This is good enough and tracking individual |
| 832 | * writeable reference is too fragile given the way @mode is |
| 833 | * used in blkdev_get/put(). |
| 834 | */ |
| 835 | if ((mode & FMODE_WRITE) && !bdev->bd_write_holder && |
| 836 | (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) { |
| 837 | bdev->bd_write_holder = true; |
| 838 | unblock_events = false; |
| 839 | } |
| 840 | } |
| 841 | mutex_unlock(&disk->open_mutex); |
| 842 | |
| 843 | if (unblock_events) |
| 844 | disk_unblock_events(disk); |
| 845 | return bdev; |
| 846 | put_module: |
| 847 | module_put(disk->fops->owner); |
| 848 | abort_claiming: |
| 849 | if (mode & FMODE_EXCL) |
| 850 | bd_abort_claiming(bdev, holder); |
| 851 | mutex_unlock(&disk->open_mutex); |
| 852 | disk_unblock_events(disk); |
| 853 | put_blkdev: |
| 854 | blkdev_put_no_open(bdev); |
| 855 | return ERR_PTR(ret); |
| 856 | } |
| 857 | EXPORT_SYMBOL(blkdev_get_by_dev); |
| 858 | |
| 859 | /** |
| 860 | * blkdev_get_by_path - open a block device by name |
| 861 | * @path: path to the block device to open |
| 862 | * @mode: FMODE_* mask |
| 863 | * @holder: exclusive holder identifier |
| 864 | * |
| 865 | * Open the block device described by the device file at @path. If @mode |
| 866 | * includes %FMODE_EXCL, the block device is opened with exclusive access. |
| 867 | * Specifying %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may |
| 868 | * nest for the same @holder. |
| 869 | * |
| 870 | * CONTEXT: |
| 871 | * Might sleep. |
| 872 | * |
| 873 | * RETURNS: |
| 874 | * Reference to the block_device on success, ERR_PTR(-errno) on failure. |
| 875 | */ |
| 876 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, |
| 877 | void *holder) |
| 878 | { |
| 879 | struct block_device *bdev; |
| 880 | dev_t dev; |
| 881 | int error; |
| 882 | |
| 883 | error = lookup_bdev(path, &dev); |
| 884 | if (error) |
| 885 | return ERR_PTR(error); |
| 886 | |
| 887 | bdev = blkdev_get_by_dev(dev, mode, holder); |
| 888 | if (!IS_ERR(bdev) && (mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
| 889 | blkdev_put(bdev, mode); |
| 890 | return ERR_PTR(-EACCES); |
| 891 | } |
| 892 | |
| 893 | return bdev; |
| 894 | } |
| 895 | EXPORT_SYMBOL(blkdev_get_by_path); |
| 896 | |
| 897 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
| 898 | { |
| 899 | struct gendisk *disk = bdev->bd_disk; |
| 900 | |
| 901 | /* |
| 902 | * Sync early if it looks like we're the last one. If someone else |
| 903 | * opens the block device between now and the decrement of bd_openers |
| 904 | * then we did a sync that we didn't need to, but that's not the end |
| 905 | * of the world and we want to avoid long (could be several minute) |
| 906 | * syncs while holding the mutex. |
| 907 | */ |
| 908 | if (atomic_read(&bdev->bd_openers) == 1) |
| 909 | sync_blockdev(bdev); |
| 910 | |
| 911 | mutex_lock(&disk->open_mutex); |
| 912 | if (mode & FMODE_EXCL) { |
| 913 | struct block_device *whole = bdev_whole(bdev); |
| 914 | bool bdev_free; |
| 915 | |
| 916 | /* |
| 917 | * Release a claim on the device. The holder fields |
| 918 | * are protected with bdev_lock. open_mutex is to |
| 919 | * synchronize disk_holder unlinking. |
| 920 | */ |
| 921 | spin_lock(&bdev_lock); |
| 922 | |
| 923 | WARN_ON_ONCE(--bdev->bd_holders < 0); |
| 924 | WARN_ON_ONCE(--whole->bd_holders < 0); |
| 925 | |
| 926 | if ((bdev_free = !bdev->bd_holders)) |
| 927 | bdev->bd_holder = NULL; |
| 928 | if (!whole->bd_holders) |
| 929 | whole->bd_holder = NULL; |
| 930 | |
| 931 | spin_unlock(&bdev_lock); |
| 932 | |
| 933 | /* |
| 934 | * If this was the last claim, remove holder link and |
| 935 | * unblock evpoll if it was a write holder. |
| 936 | */ |
| 937 | if (bdev_free && bdev->bd_write_holder) { |
| 938 | disk_unblock_events(disk); |
| 939 | bdev->bd_write_holder = false; |
| 940 | } |
| 941 | } |
| 942 | |
| 943 | /* |
| 944 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE |
| 945 | * event. This is to ensure detection of media removal commanded |
| 946 | * from userland - e.g. eject(1). |
| 947 | */ |
| 948 | disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE); |
| 949 | |
| 950 | if (bdev_is_partition(bdev)) |
| 951 | blkdev_put_part(bdev, mode); |
| 952 | else |
| 953 | blkdev_put_whole(bdev, mode); |
| 954 | mutex_unlock(&disk->open_mutex); |
| 955 | |
| 956 | module_put(disk->fops->owner); |
| 957 | blkdev_put_no_open(bdev); |
| 958 | } |
| 959 | EXPORT_SYMBOL(blkdev_put); |
| 960 | |
| 961 | /** |
| 962 | * lookup_bdev() - Look up a struct block_device by name. |
| 963 | * @pathname: Name of the block device in the filesystem. |
| 964 | * @dev: Pointer to the block device's dev_t, if found. |
| 965 | * |
| 966 | * Lookup the block device's dev_t at @pathname in the current |
| 967 | * namespace if possible and return it in @dev. |
| 968 | * |
| 969 | * Context: May sleep. |
| 970 | * Return: 0 if succeeded, negative errno otherwise. |
| 971 | */ |
| 972 | int lookup_bdev(const char *pathname, dev_t *dev) |
| 973 | { |
| 974 | struct inode *inode; |
| 975 | struct path path; |
| 976 | int error; |
| 977 | |
| 978 | if (!pathname || !*pathname) |
| 979 | return -EINVAL; |
| 980 | |
| 981 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
| 982 | if (error) |
| 983 | return error; |
| 984 | |
| 985 | inode = d_backing_inode(path.dentry); |
| 986 | error = -ENOTBLK; |
| 987 | if (!S_ISBLK(inode->i_mode)) |
| 988 | goto out_path_put; |
| 989 | error = -EACCES; |
| 990 | if (!may_open_dev(&path)) |
| 991 | goto out_path_put; |
| 992 | |
| 993 | *dev = inode->i_rdev; |
| 994 | error = 0; |
| 995 | out_path_put: |
| 996 | path_put(&path); |
| 997 | return error; |
| 998 | } |
| 999 | EXPORT_SYMBOL(lookup_bdev); |
| 1000 | |
| 1001 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
| 1002 | { |
| 1003 | struct super_block *sb = get_super(bdev); |
| 1004 | int res = 0; |
| 1005 | |
| 1006 | if (sb) { |
| 1007 | /* |
| 1008 | * no need to lock the super, get_super holds the |
| 1009 | * read mutex so the filesystem cannot go away |
| 1010 | * under us (->put_super runs with the write lock |
| 1011 | * hold). |
| 1012 | */ |
| 1013 | shrink_dcache_sb(sb); |
| 1014 | res = invalidate_inodes(sb, kill_dirty); |
| 1015 | drop_super(sb); |
| 1016 | } |
| 1017 | invalidate_bdev(bdev); |
| 1018 | return res; |
| 1019 | } |
| 1020 | EXPORT_SYMBOL(__invalidate_device); |
| 1021 | |
| 1022 | void sync_bdevs(bool wait) |
| 1023 | { |
| 1024 | struct inode *inode, *old_inode = NULL; |
| 1025 | |
| 1026 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
| 1027 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
| 1028 | struct address_space *mapping = inode->i_mapping; |
| 1029 | struct block_device *bdev; |
| 1030 | |
| 1031 | spin_lock(&inode->i_lock); |
| 1032 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || |
| 1033 | mapping->nrpages == 0) { |
| 1034 | spin_unlock(&inode->i_lock); |
| 1035 | continue; |
| 1036 | } |
| 1037 | __iget(inode); |
| 1038 | spin_unlock(&inode->i_lock); |
| 1039 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
| 1040 | /* |
| 1041 | * We hold a reference to 'inode' so it couldn't have been |
| 1042 | * removed from s_inodes list while we dropped the |
| 1043 | * s_inode_list_lock We cannot iput the inode now as we can |
| 1044 | * be holding the last reference and we cannot iput it under |
| 1045 | * s_inode_list_lock. So we keep the reference and iput it |
| 1046 | * later. |
| 1047 | */ |
| 1048 | iput(old_inode); |
| 1049 | old_inode = inode; |
| 1050 | bdev = I_BDEV(inode); |
| 1051 | |
| 1052 | mutex_lock(&bdev->bd_disk->open_mutex); |
| 1053 | if (!atomic_read(&bdev->bd_openers)) { |
| 1054 | ; /* skip */ |
| 1055 | } else if (wait) { |
| 1056 | /* |
| 1057 | * We keep the error status of individual mapping so |
| 1058 | * that applications can catch the writeback error using |
| 1059 | * fsync(2). See filemap_fdatawait_keep_errors() for |
| 1060 | * details. |
| 1061 | */ |
| 1062 | filemap_fdatawait_keep_errors(inode->i_mapping); |
| 1063 | } else { |
| 1064 | filemap_fdatawrite(inode->i_mapping); |
| 1065 | } |
| 1066 | mutex_unlock(&bdev->bd_disk->open_mutex); |
| 1067 | |
| 1068 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
| 1069 | } |
| 1070 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
| 1071 | iput(old_inode); |
| 1072 | } |
| 1073 | |
| 1074 | /* |
| 1075 | * Handle STATX_DIOALIGN for block devices. |
| 1076 | * |
| 1077 | * Note that the inode passed to this is the inode of a block device node file, |
| 1078 | * not the block device's internal inode. Therefore it is *not* valid to use |
| 1079 | * I_BDEV() here; the block device has to be looked up by i_rdev instead. |
| 1080 | */ |
| 1081 | void bdev_statx_dioalign(struct inode *inode, struct kstat *stat) |
| 1082 | { |
| 1083 | struct block_device *bdev; |
| 1084 | |
| 1085 | bdev = blkdev_get_no_open(inode->i_rdev); |
| 1086 | if (!bdev) |
| 1087 | return; |
| 1088 | |
| 1089 | stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; |
| 1090 | stat->dio_offset_align = bdev_logical_block_size(bdev); |
| 1091 | stat->result_mask |= STATX_DIOALIGN; |
| 1092 | |
| 1093 | blkdev_put_no_open(bdev); |
| 1094 | } |