| 1 | /* |
| 2 | * linux/fs/block_dev.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 5 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE |
| 6 | */ |
| 7 | |
| 8 | #include <linux/init.h> |
| 9 | #include <linux/mm.h> |
| 10 | #include <linux/fcntl.h> |
| 11 | #include <linux/slab.h> |
| 12 | #include <linux/kmod.h> |
| 13 | #include <linux/major.h> |
| 14 | #include <linux/device_cgroup.h> |
| 15 | #include <linux/highmem.h> |
| 16 | #include <linux/blkdev.h> |
| 17 | #include <linux/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/pagevec.h> |
| 23 | #include <linux/writeback.h> |
| 24 | #include <linux/mpage.h> |
| 25 | #include <linux/mount.h> |
| 26 | #include <linux/uio.h> |
| 27 | #include <linux/namei.h> |
| 28 | #include <linux/log2.h> |
| 29 | #include <linux/cleancache.h> |
| 30 | #include <linux/aio.h> |
| 31 | #include <asm/uaccess.h> |
| 32 | #include "internal.h" |
| 33 | |
| 34 | struct bdev_inode { |
| 35 | struct block_device bdev; |
| 36 | struct inode vfs_inode; |
| 37 | }; |
| 38 | |
| 39 | static const struct address_space_operations def_blk_aops; |
| 40 | |
| 41 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
| 42 | { |
| 43 | return container_of(inode, struct bdev_inode, vfs_inode); |
| 44 | } |
| 45 | |
| 46 | inline struct block_device *I_BDEV(struct inode *inode) |
| 47 | { |
| 48 | return &BDEV_I(inode)->bdev; |
| 49 | } |
| 50 | EXPORT_SYMBOL(I_BDEV); |
| 51 | |
| 52 | /* |
| 53 | * Move the inode from its current bdi to a new bdi. If the inode is dirty we |
| 54 | * need to move it onto the dirty list of @dst so that the inode is always on |
| 55 | * the right list. |
| 56 | */ |
| 57 | static void bdev_inode_switch_bdi(struct inode *inode, |
| 58 | struct backing_dev_info *dst) |
| 59 | { |
| 60 | struct backing_dev_info *old = inode->i_data.backing_dev_info; |
| 61 | bool wakeup_bdi = false; |
| 62 | |
| 63 | if (unlikely(dst == old)) /* deadlock avoidance */ |
| 64 | return; |
| 65 | bdi_lock_two(&old->wb, &dst->wb); |
| 66 | spin_lock(&inode->i_lock); |
| 67 | inode->i_data.backing_dev_info = dst; |
| 68 | if (inode->i_state & I_DIRTY) { |
| 69 | if (bdi_cap_writeback_dirty(dst) && !wb_has_dirty_io(&dst->wb)) |
| 70 | wakeup_bdi = true; |
| 71 | list_move(&inode->i_wb_list, &dst->wb.b_dirty); |
| 72 | } |
| 73 | spin_unlock(&inode->i_lock); |
| 74 | spin_unlock(&old->wb.list_lock); |
| 75 | spin_unlock(&dst->wb.list_lock); |
| 76 | |
| 77 | if (wakeup_bdi) |
| 78 | bdi_wakeup_thread_delayed(dst); |
| 79 | } |
| 80 | |
| 81 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
| 82 | void kill_bdev(struct block_device *bdev) |
| 83 | { |
| 84 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
| 85 | |
| 86 | if (mapping->nrpages == 0) |
| 87 | return; |
| 88 | |
| 89 | invalidate_bh_lrus(); |
| 90 | truncate_inode_pages(mapping, 0); |
| 91 | } |
| 92 | EXPORT_SYMBOL(kill_bdev); |
| 93 | |
| 94 | /* Invalidate clean unused buffers and pagecache. */ |
| 95 | void invalidate_bdev(struct block_device *bdev) |
| 96 | { |
| 97 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
| 98 | |
| 99 | if (mapping->nrpages == 0) |
| 100 | return; |
| 101 | |
| 102 | invalidate_bh_lrus(); |
| 103 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ |
| 104 | invalidate_mapping_pages(mapping, 0, -1); |
| 105 | /* 99% of the time, we don't need to flush the cleancache on the bdev. |
| 106 | * But, for the strange corners, lets be cautious |
| 107 | */ |
| 108 | cleancache_invalidate_inode(mapping); |
| 109 | } |
| 110 | EXPORT_SYMBOL(invalidate_bdev); |
| 111 | |
| 112 | int set_blocksize(struct block_device *bdev, int size) |
| 113 | { |
| 114 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ |
| 115 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
| 116 | return -EINVAL; |
| 117 | |
| 118 | /* Size cannot be smaller than the size supported by the device */ |
| 119 | if (size < bdev_logical_block_size(bdev)) |
| 120 | return -EINVAL; |
| 121 | |
| 122 | /* Don't change the size if it is same as current */ |
| 123 | if (bdev->bd_block_size != size) { |
| 124 | sync_blockdev(bdev); |
| 125 | bdev->bd_block_size = size; |
| 126 | bdev->bd_inode->i_blkbits = blksize_bits(size); |
| 127 | kill_bdev(bdev); |
| 128 | } |
| 129 | return 0; |
| 130 | } |
| 131 | |
| 132 | EXPORT_SYMBOL(set_blocksize); |
| 133 | |
| 134 | int sb_set_blocksize(struct super_block *sb, int size) |
| 135 | { |
| 136 | if (set_blocksize(sb->s_bdev, size)) |
| 137 | return 0; |
| 138 | /* If we get here, we know size is power of two |
| 139 | * and it's value is between 512 and PAGE_SIZE */ |
| 140 | sb->s_blocksize = size; |
| 141 | sb->s_blocksize_bits = blksize_bits(size); |
| 142 | return sb->s_blocksize; |
| 143 | } |
| 144 | |
| 145 | EXPORT_SYMBOL(sb_set_blocksize); |
| 146 | |
| 147 | int sb_min_blocksize(struct super_block *sb, int size) |
| 148 | { |
| 149 | int minsize = bdev_logical_block_size(sb->s_bdev); |
| 150 | if (size < minsize) |
| 151 | size = minsize; |
| 152 | return sb_set_blocksize(sb, size); |
| 153 | } |
| 154 | |
| 155 | EXPORT_SYMBOL(sb_min_blocksize); |
| 156 | |
| 157 | static int |
| 158 | blkdev_get_block(struct inode *inode, sector_t iblock, |
| 159 | struct buffer_head *bh, int create) |
| 160 | { |
| 161 | bh->b_bdev = I_BDEV(inode); |
| 162 | bh->b_blocknr = iblock; |
| 163 | set_buffer_mapped(bh); |
| 164 | return 0; |
| 165 | } |
| 166 | |
| 167 | static ssize_t |
| 168 | blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, |
| 169 | loff_t offset, unsigned long nr_segs) |
| 170 | { |
| 171 | struct file *file = iocb->ki_filp; |
| 172 | struct inode *inode = file->f_mapping->host; |
| 173 | |
| 174 | return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset, |
| 175 | nr_segs, blkdev_get_block, NULL, NULL, 0); |
| 176 | } |
| 177 | |
| 178 | int __sync_blockdev(struct block_device *bdev, int wait) |
| 179 | { |
| 180 | if (!bdev) |
| 181 | return 0; |
| 182 | if (!wait) |
| 183 | return filemap_flush(bdev->bd_inode->i_mapping); |
| 184 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); |
| 185 | } |
| 186 | |
| 187 | /* |
| 188 | * Write out and wait upon all the dirty data associated with a block |
| 189 | * device via its mapping. Does not take the superblock lock. |
| 190 | */ |
| 191 | int sync_blockdev(struct block_device *bdev) |
| 192 | { |
| 193 | return __sync_blockdev(bdev, 1); |
| 194 | } |
| 195 | EXPORT_SYMBOL(sync_blockdev); |
| 196 | |
| 197 | /* |
| 198 | * Write out and wait upon all dirty data associated with this |
| 199 | * device. Filesystem data as well as the underlying block |
| 200 | * device. Takes the superblock lock. |
| 201 | */ |
| 202 | int fsync_bdev(struct block_device *bdev) |
| 203 | { |
| 204 | struct super_block *sb = get_super(bdev); |
| 205 | if (sb) { |
| 206 | int res = sync_filesystem(sb); |
| 207 | drop_super(sb); |
| 208 | return res; |
| 209 | } |
| 210 | return sync_blockdev(bdev); |
| 211 | } |
| 212 | EXPORT_SYMBOL(fsync_bdev); |
| 213 | |
| 214 | /** |
| 215 | * freeze_bdev -- lock a filesystem and force it into a consistent state |
| 216 | * @bdev: blockdevice to lock |
| 217 | * |
| 218 | * If a superblock is found on this device, we take the s_umount semaphore |
| 219 | * on it to make sure nobody unmounts until the snapshot creation is done. |
| 220 | * The reference counter (bd_fsfreeze_count) guarantees that only the last |
| 221 | * unfreeze process can unfreeze the frozen filesystem actually when multiple |
| 222 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and |
| 223 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze |
| 224 | * actually. |
| 225 | */ |
| 226 | struct super_block *freeze_bdev(struct block_device *bdev) |
| 227 | { |
| 228 | struct super_block *sb; |
| 229 | int error = 0; |
| 230 | |
| 231 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
| 232 | if (++bdev->bd_fsfreeze_count > 1) { |
| 233 | /* |
| 234 | * We don't even need to grab a reference - the first call |
| 235 | * to freeze_bdev grab an active reference and only the last |
| 236 | * thaw_bdev drops it. |
| 237 | */ |
| 238 | sb = get_super(bdev); |
| 239 | drop_super(sb); |
| 240 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 241 | return sb; |
| 242 | } |
| 243 | |
| 244 | sb = get_active_super(bdev); |
| 245 | if (!sb) |
| 246 | goto out; |
| 247 | error = freeze_super(sb); |
| 248 | if (error) { |
| 249 | deactivate_super(sb); |
| 250 | bdev->bd_fsfreeze_count--; |
| 251 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 252 | return ERR_PTR(error); |
| 253 | } |
| 254 | deactivate_super(sb); |
| 255 | out: |
| 256 | sync_blockdev(bdev); |
| 257 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 258 | return sb; /* thaw_bdev releases s->s_umount */ |
| 259 | } |
| 260 | EXPORT_SYMBOL(freeze_bdev); |
| 261 | |
| 262 | /** |
| 263 | * thaw_bdev -- unlock filesystem |
| 264 | * @bdev: blockdevice to unlock |
| 265 | * @sb: associated superblock |
| 266 | * |
| 267 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). |
| 268 | */ |
| 269 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) |
| 270 | { |
| 271 | int error = -EINVAL; |
| 272 | |
| 273 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
| 274 | if (!bdev->bd_fsfreeze_count) |
| 275 | goto out; |
| 276 | |
| 277 | error = 0; |
| 278 | if (--bdev->bd_fsfreeze_count > 0) |
| 279 | goto out; |
| 280 | |
| 281 | if (!sb) |
| 282 | goto out; |
| 283 | |
| 284 | error = thaw_super(sb); |
| 285 | if (error) { |
| 286 | bdev->bd_fsfreeze_count++; |
| 287 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 288 | return error; |
| 289 | } |
| 290 | out: |
| 291 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 292 | return 0; |
| 293 | } |
| 294 | EXPORT_SYMBOL(thaw_bdev); |
| 295 | |
| 296 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
| 297 | { |
| 298 | return block_write_full_page(page, blkdev_get_block, wbc); |
| 299 | } |
| 300 | |
| 301 | static int blkdev_readpage(struct file * file, struct page * page) |
| 302 | { |
| 303 | return block_read_full_page(page, blkdev_get_block); |
| 304 | } |
| 305 | |
| 306 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
| 307 | loff_t pos, unsigned len, unsigned flags, |
| 308 | struct page **pagep, void **fsdata) |
| 309 | { |
| 310 | return block_write_begin(mapping, pos, len, flags, pagep, |
| 311 | blkdev_get_block); |
| 312 | } |
| 313 | |
| 314 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
| 315 | loff_t pos, unsigned len, unsigned copied, |
| 316 | struct page *page, void *fsdata) |
| 317 | { |
| 318 | int ret; |
| 319 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
| 320 | |
| 321 | unlock_page(page); |
| 322 | page_cache_release(page); |
| 323 | |
| 324 | return ret; |
| 325 | } |
| 326 | |
| 327 | /* |
| 328 | * private llseek: |
| 329 | * for a block special file file_inode(file)->i_size is zero |
| 330 | * so we compute the size by hand (just as in block_read/write above) |
| 331 | */ |
| 332 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
| 333 | { |
| 334 | struct inode *bd_inode = file->f_mapping->host; |
| 335 | loff_t size; |
| 336 | loff_t retval; |
| 337 | |
| 338 | mutex_lock(&bd_inode->i_mutex); |
| 339 | size = i_size_read(bd_inode); |
| 340 | |
| 341 | retval = -EINVAL; |
| 342 | switch (whence) { |
| 343 | case SEEK_END: |
| 344 | offset += size; |
| 345 | break; |
| 346 | case SEEK_CUR: |
| 347 | offset += file->f_pos; |
| 348 | case SEEK_SET: |
| 349 | break; |
| 350 | default: |
| 351 | goto out; |
| 352 | } |
| 353 | if (offset >= 0 && offset <= size) { |
| 354 | if (offset != file->f_pos) { |
| 355 | file->f_pos = offset; |
| 356 | } |
| 357 | retval = offset; |
| 358 | } |
| 359 | out: |
| 360 | mutex_unlock(&bd_inode->i_mutex); |
| 361 | return retval; |
| 362 | } |
| 363 | |
| 364 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
| 365 | { |
| 366 | struct inode *bd_inode = filp->f_mapping->host; |
| 367 | struct block_device *bdev = I_BDEV(bd_inode); |
| 368 | int error; |
| 369 | |
| 370 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); |
| 371 | if (error) |
| 372 | return error; |
| 373 | |
| 374 | /* |
| 375 | * There is no need to serialise calls to blkdev_issue_flush with |
| 376 | * i_mutex and doing so causes performance issues with concurrent |
| 377 | * O_SYNC writers to a block device. |
| 378 | */ |
| 379 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
| 380 | if (error == -EOPNOTSUPP) |
| 381 | error = 0; |
| 382 | |
| 383 | return error; |
| 384 | } |
| 385 | EXPORT_SYMBOL(blkdev_fsync); |
| 386 | |
| 387 | /* |
| 388 | * pseudo-fs |
| 389 | */ |
| 390 | |
| 391 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); |
| 392 | static struct kmem_cache * bdev_cachep __read_mostly; |
| 393 | |
| 394 | static struct inode *bdev_alloc_inode(struct super_block *sb) |
| 395 | { |
| 396 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
| 397 | if (!ei) |
| 398 | return NULL; |
| 399 | return &ei->vfs_inode; |
| 400 | } |
| 401 | |
| 402 | static void bdev_i_callback(struct rcu_head *head) |
| 403 | { |
| 404 | struct inode *inode = container_of(head, struct inode, i_rcu); |
| 405 | struct bdev_inode *bdi = BDEV_I(inode); |
| 406 | |
| 407 | kmem_cache_free(bdev_cachep, bdi); |
| 408 | } |
| 409 | |
| 410 | static void bdev_destroy_inode(struct inode *inode) |
| 411 | { |
| 412 | call_rcu(&inode->i_rcu, bdev_i_callback); |
| 413 | } |
| 414 | |
| 415 | static void init_once(void *foo) |
| 416 | { |
| 417 | struct bdev_inode *ei = (struct bdev_inode *) foo; |
| 418 | struct block_device *bdev = &ei->bdev; |
| 419 | |
| 420 | memset(bdev, 0, sizeof(*bdev)); |
| 421 | mutex_init(&bdev->bd_mutex); |
| 422 | INIT_LIST_HEAD(&bdev->bd_inodes); |
| 423 | INIT_LIST_HEAD(&bdev->bd_list); |
| 424 | #ifdef CONFIG_SYSFS |
| 425 | INIT_LIST_HEAD(&bdev->bd_holder_disks); |
| 426 | #endif |
| 427 | inode_init_once(&ei->vfs_inode); |
| 428 | /* Initialize mutex for freeze. */ |
| 429 | mutex_init(&bdev->bd_fsfreeze_mutex); |
| 430 | } |
| 431 | |
| 432 | static inline void __bd_forget(struct inode *inode) |
| 433 | { |
| 434 | list_del_init(&inode->i_devices); |
| 435 | inode->i_bdev = NULL; |
| 436 | inode->i_mapping = &inode->i_data; |
| 437 | } |
| 438 | |
| 439 | static void bdev_evict_inode(struct inode *inode) |
| 440 | { |
| 441 | struct block_device *bdev = &BDEV_I(inode)->bdev; |
| 442 | struct list_head *p; |
| 443 | truncate_inode_pages(&inode->i_data, 0); |
| 444 | invalidate_inode_buffers(inode); /* is it needed here? */ |
| 445 | clear_inode(inode); |
| 446 | spin_lock(&bdev_lock); |
| 447 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { |
| 448 | __bd_forget(list_entry(p, struct inode, i_devices)); |
| 449 | } |
| 450 | list_del_init(&bdev->bd_list); |
| 451 | spin_unlock(&bdev_lock); |
| 452 | } |
| 453 | |
| 454 | static const struct super_operations bdev_sops = { |
| 455 | .statfs = simple_statfs, |
| 456 | .alloc_inode = bdev_alloc_inode, |
| 457 | .destroy_inode = bdev_destroy_inode, |
| 458 | .drop_inode = generic_delete_inode, |
| 459 | .evict_inode = bdev_evict_inode, |
| 460 | }; |
| 461 | |
| 462 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
| 463 | int flags, const char *dev_name, void *data) |
| 464 | { |
| 465 | return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); |
| 466 | } |
| 467 | |
| 468 | static struct file_system_type bd_type = { |
| 469 | .name = "bdev", |
| 470 | .mount = bd_mount, |
| 471 | .kill_sb = kill_anon_super, |
| 472 | }; |
| 473 | |
| 474 | static struct super_block *blockdev_superblock __read_mostly; |
| 475 | |
| 476 | void __init bdev_cache_init(void) |
| 477 | { |
| 478 | int err; |
| 479 | static struct vfsmount *bd_mnt; |
| 480 | |
| 481 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
| 482 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
| 483 | SLAB_MEM_SPREAD|SLAB_PANIC), |
| 484 | init_once); |
| 485 | err = register_filesystem(&bd_type); |
| 486 | if (err) |
| 487 | panic("Cannot register bdev pseudo-fs"); |
| 488 | bd_mnt = kern_mount(&bd_type); |
| 489 | if (IS_ERR(bd_mnt)) |
| 490 | panic("Cannot create bdev pseudo-fs"); |
| 491 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
| 492 | } |
| 493 | |
| 494 | /* |
| 495 | * Most likely _very_ bad one - but then it's hardly critical for small |
| 496 | * /dev and can be fixed when somebody will need really large one. |
| 497 | * Keep in mind that it will be fed through icache hash function too. |
| 498 | */ |
| 499 | static inline unsigned long hash(dev_t dev) |
| 500 | { |
| 501 | return MAJOR(dev)+MINOR(dev); |
| 502 | } |
| 503 | |
| 504 | static int bdev_test(struct inode *inode, void *data) |
| 505 | { |
| 506 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; |
| 507 | } |
| 508 | |
| 509 | static int bdev_set(struct inode *inode, void *data) |
| 510 | { |
| 511 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; |
| 512 | return 0; |
| 513 | } |
| 514 | |
| 515 | static LIST_HEAD(all_bdevs); |
| 516 | |
| 517 | struct block_device *bdget(dev_t dev) |
| 518 | { |
| 519 | struct block_device *bdev; |
| 520 | struct inode *inode; |
| 521 | |
| 522 | inode = iget5_locked(blockdev_superblock, hash(dev), |
| 523 | bdev_test, bdev_set, &dev); |
| 524 | |
| 525 | if (!inode) |
| 526 | return NULL; |
| 527 | |
| 528 | bdev = &BDEV_I(inode)->bdev; |
| 529 | |
| 530 | if (inode->i_state & I_NEW) { |
| 531 | bdev->bd_contains = NULL; |
| 532 | bdev->bd_super = NULL; |
| 533 | bdev->bd_inode = inode; |
| 534 | bdev->bd_block_size = (1 << inode->i_blkbits); |
| 535 | bdev->bd_part_count = 0; |
| 536 | bdev->bd_invalidated = 0; |
| 537 | inode->i_mode = S_IFBLK; |
| 538 | inode->i_rdev = dev; |
| 539 | inode->i_bdev = bdev; |
| 540 | inode->i_data.a_ops = &def_blk_aops; |
| 541 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); |
| 542 | inode->i_data.backing_dev_info = &default_backing_dev_info; |
| 543 | spin_lock(&bdev_lock); |
| 544 | list_add(&bdev->bd_list, &all_bdevs); |
| 545 | spin_unlock(&bdev_lock); |
| 546 | unlock_new_inode(inode); |
| 547 | } |
| 548 | return bdev; |
| 549 | } |
| 550 | |
| 551 | EXPORT_SYMBOL(bdget); |
| 552 | |
| 553 | /** |
| 554 | * bdgrab -- Grab a reference to an already referenced block device |
| 555 | * @bdev: Block device to grab a reference to. |
| 556 | */ |
| 557 | struct block_device *bdgrab(struct block_device *bdev) |
| 558 | { |
| 559 | ihold(bdev->bd_inode); |
| 560 | return bdev; |
| 561 | } |
| 562 | EXPORT_SYMBOL(bdgrab); |
| 563 | |
| 564 | long nr_blockdev_pages(void) |
| 565 | { |
| 566 | struct block_device *bdev; |
| 567 | long ret = 0; |
| 568 | spin_lock(&bdev_lock); |
| 569 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
| 570 | ret += bdev->bd_inode->i_mapping->nrpages; |
| 571 | } |
| 572 | spin_unlock(&bdev_lock); |
| 573 | return ret; |
| 574 | } |
| 575 | |
| 576 | void bdput(struct block_device *bdev) |
| 577 | { |
| 578 | iput(bdev->bd_inode); |
| 579 | } |
| 580 | |
| 581 | EXPORT_SYMBOL(bdput); |
| 582 | |
| 583 | static struct block_device *bd_acquire(struct inode *inode) |
| 584 | { |
| 585 | struct block_device *bdev; |
| 586 | |
| 587 | spin_lock(&bdev_lock); |
| 588 | bdev = inode->i_bdev; |
| 589 | if (bdev) { |
| 590 | ihold(bdev->bd_inode); |
| 591 | spin_unlock(&bdev_lock); |
| 592 | return bdev; |
| 593 | } |
| 594 | spin_unlock(&bdev_lock); |
| 595 | |
| 596 | bdev = bdget(inode->i_rdev); |
| 597 | if (bdev) { |
| 598 | spin_lock(&bdev_lock); |
| 599 | if (!inode->i_bdev) { |
| 600 | /* |
| 601 | * We take an additional reference to bd_inode, |
| 602 | * and it's released in clear_inode() of inode. |
| 603 | * So, we can access it via ->i_mapping always |
| 604 | * without igrab(). |
| 605 | */ |
| 606 | ihold(bdev->bd_inode); |
| 607 | inode->i_bdev = bdev; |
| 608 | inode->i_mapping = bdev->bd_inode->i_mapping; |
| 609 | list_add(&inode->i_devices, &bdev->bd_inodes); |
| 610 | } |
| 611 | spin_unlock(&bdev_lock); |
| 612 | } |
| 613 | return bdev; |
| 614 | } |
| 615 | |
| 616 | static inline int sb_is_blkdev_sb(struct super_block *sb) |
| 617 | { |
| 618 | return sb == blockdev_superblock; |
| 619 | } |
| 620 | |
| 621 | /* Call when you free inode */ |
| 622 | |
| 623 | void bd_forget(struct inode *inode) |
| 624 | { |
| 625 | struct block_device *bdev = NULL; |
| 626 | |
| 627 | spin_lock(&bdev_lock); |
| 628 | if (!sb_is_blkdev_sb(inode->i_sb)) |
| 629 | bdev = inode->i_bdev; |
| 630 | __bd_forget(inode); |
| 631 | spin_unlock(&bdev_lock); |
| 632 | |
| 633 | if (bdev) |
| 634 | iput(bdev->bd_inode); |
| 635 | } |
| 636 | |
| 637 | /** |
| 638 | * bd_may_claim - test whether a block device can be claimed |
| 639 | * @bdev: block device of interest |
| 640 | * @whole: whole block device containing @bdev, may equal @bdev |
| 641 | * @holder: holder trying to claim @bdev |
| 642 | * |
| 643 | * Test whether @bdev can be claimed by @holder. |
| 644 | * |
| 645 | * CONTEXT: |
| 646 | * spin_lock(&bdev_lock). |
| 647 | * |
| 648 | * RETURNS: |
| 649 | * %true if @bdev can be claimed, %false otherwise. |
| 650 | */ |
| 651 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, |
| 652 | void *holder) |
| 653 | { |
| 654 | if (bdev->bd_holder == holder) |
| 655 | return true; /* already a holder */ |
| 656 | else if (bdev->bd_holder != NULL) |
| 657 | return false; /* held by someone else */ |
| 658 | else if (whole == bdev) |
| 659 | return true; /* is a whole device which isn't held */ |
| 660 | |
| 661 | else if (whole->bd_holder == bd_may_claim) |
| 662 | return true; /* is a partition of a device that is being partitioned */ |
| 663 | else if (whole->bd_holder != NULL) |
| 664 | return false; /* is a partition of a held device */ |
| 665 | else |
| 666 | return true; /* is a partition of an un-held device */ |
| 667 | } |
| 668 | |
| 669 | /** |
| 670 | * bd_prepare_to_claim - prepare to claim a block device |
| 671 | * @bdev: block device of interest |
| 672 | * @whole: the whole device containing @bdev, may equal @bdev |
| 673 | * @holder: holder trying to claim @bdev |
| 674 | * |
| 675 | * Prepare to claim @bdev. This function fails if @bdev is already |
| 676 | * claimed by another holder and waits if another claiming is in |
| 677 | * progress. This function doesn't actually claim. On successful |
| 678 | * return, the caller has ownership of bd_claiming and bd_holder[s]. |
| 679 | * |
| 680 | * CONTEXT: |
| 681 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab |
| 682 | * it multiple times. |
| 683 | * |
| 684 | * RETURNS: |
| 685 | * 0 if @bdev can be claimed, -EBUSY otherwise. |
| 686 | */ |
| 687 | static int bd_prepare_to_claim(struct block_device *bdev, |
| 688 | struct block_device *whole, void *holder) |
| 689 | { |
| 690 | retry: |
| 691 | /* if someone else claimed, fail */ |
| 692 | if (!bd_may_claim(bdev, whole, holder)) |
| 693 | return -EBUSY; |
| 694 | |
| 695 | /* if claiming is already in progress, wait for it to finish */ |
| 696 | if (whole->bd_claiming) { |
| 697 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
| 698 | DEFINE_WAIT(wait); |
| 699 | |
| 700 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); |
| 701 | spin_unlock(&bdev_lock); |
| 702 | schedule(); |
| 703 | finish_wait(wq, &wait); |
| 704 | spin_lock(&bdev_lock); |
| 705 | goto retry; |
| 706 | } |
| 707 | |
| 708 | /* yay, all mine */ |
| 709 | return 0; |
| 710 | } |
| 711 | |
| 712 | /** |
| 713 | * bd_start_claiming - start claiming a block device |
| 714 | * @bdev: block device of interest |
| 715 | * @holder: holder trying to claim @bdev |
| 716 | * |
| 717 | * @bdev is about to be opened exclusively. Check @bdev can be opened |
| 718 | * exclusively and mark that an exclusive open is in progress. Each |
| 719 | * successful call to this function must be matched with a call to |
| 720 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
| 721 | * fail). |
| 722 | * |
| 723 | * This function is used to gain exclusive access to the block device |
| 724 | * without actually causing other exclusive open attempts to fail. It |
| 725 | * should be used when the open sequence itself requires exclusive |
| 726 | * access but may subsequently fail. |
| 727 | * |
| 728 | * CONTEXT: |
| 729 | * Might sleep. |
| 730 | * |
| 731 | * RETURNS: |
| 732 | * Pointer to the block device containing @bdev on success, ERR_PTR() |
| 733 | * value on failure. |
| 734 | */ |
| 735 | static struct block_device *bd_start_claiming(struct block_device *bdev, |
| 736 | void *holder) |
| 737 | { |
| 738 | struct gendisk *disk; |
| 739 | struct block_device *whole; |
| 740 | int partno, err; |
| 741 | |
| 742 | might_sleep(); |
| 743 | |
| 744 | /* |
| 745 | * @bdev might not have been initialized properly yet, look up |
| 746 | * and grab the outer block device the hard way. |
| 747 | */ |
| 748 | disk = get_gendisk(bdev->bd_dev, &partno); |
| 749 | if (!disk) |
| 750 | return ERR_PTR(-ENXIO); |
| 751 | |
| 752 | /* |
| 753 | * Normally, @bdev should equal what's returned from bdget_disk() |
| 754 | * if partno is 0; however, some drivers (floppy) use multiple |
| 755 | * bdev's for the same physical device and @bdev may be one of the |
| 756 | * aliases. Keep @bdev if partno is 0. This means claimer |
| 757 | * tracking is broken for those devices but it has always been that |
| 758 | * way. |
| 759 | */ |
| 760 | if (partno) |
| 761 | whole = bdget_disk(disk, 0); |
| 762 | else |
| 763 | whole = bdgrab(bdev); |
| 764 | |
| 765 | module_put(disk->fops->owner); |
| 766 | put_disk(disk); |
| 767 | if (!whole) |
| 768 | return ERR_PTR(-ENOMEM); |
| 769 | |
| 770 | /* prepare to claim, if successful, mark claiming in progress */ |
| 771 | spin_lock(&bdev_lock); |
| 772 | |
| 773 | err = bd_prepare_to_claim(bdev, whole, holder); |
| 774 | if (err == 0) { |
| 775 | whole->bd_claiming = holder; |
| 776 | spin_unlock(&bdev_lock); |
| 777 | return whole; |
| 778 | } else { |
| 779 | spin_unlock(&bdev_lock); |
| 780 | bdput(whole); |
| 781 | return ERR_PTR(err); |
| 782 | } |
| 783 | } |
| 784 | |
| 785 | #ifdef CONFIG_SYSFS |
| 786 | struct bd_holder_disk { |
| 787 | struct list_head list; |
| 788 | struct gendisk *disk; |
| 789 | int refcnt; |
| 790 | }; |
| 791 | |
| 792 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, |
| 793 | struct gendisk *disk) |
| 794 | { |
| 795 | struct bd_holder_disk *holder; |
| 796 | |
| 797 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) |
| 798 | if (holder->disk == disk) |
| 799 | return holder; |
| 800 | return NULL; |
| 801 | } |
| 802 | |
| 803 | static int add_symlink(struct kobject *from, struct kobject *to) |
| 804 | { |
| 805 | return sysfs_create_link(from, to, kobject_name(to)); |
| 806 | } |
| 807 | |
| 808 | static void del_symlink(struct kobject *from, struct kobject *to) |
| 809 | { |
| 810 | sysfs_remove_link(from, kobject_name(to)); |
| 811 | } |
| 812 | |
| 813 | /** |
| 814 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
| 815 | * @bdev: the claimed slave bdev |
| 816 | * @disk: the holding disk |
| 817 | * |
| 818 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
| 819 | * |
| 820 | * This functions creates the following sysfs symlinks. |
| 821 | * |
| 822 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
| 823 | * - from "holders" directory of the @bdev to the holder @disk |
| 824 | * |
| 825 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
| 826 | * passed to bd_link_disk_holder(), then: |
| 827 | * |
| 828 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
| 829 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 |
| 830 | * |
| 831 | * The caller must have claimed @bdev before calling this function and |
| 832 | * ensure that both @bdev and @disk are valid during the creation and |
| 833 | * lifetime of these symlinks. |
| 834 | * |
| 835 | * CONTEXT: |
| 836 | * Might sleep. |
| 837 | * |
| 838 | * RETURNS: |
| 839 | * 0 on success, -errno on failure. |
| 840 | */ |
| 841 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
| 842 | { |
| 843 | struct bd_holder_disk *holder; |
| 844 | int ret = 0; |
| 845 | |
| 846 | mutex_lock(&bdev->bd_mutex); |
| 847 | |
| 848 | WARN_ON_ONCE(!bdev->bd_holder); |
| 849 | |
| 850 | /* FIXME: remove the following once add_disk() handles errors */ |
| 851 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) |
| 852 | goto out_unlock; |
| 853 | |
| 854 | holder = bd_find_holder_disk(bdev, disk); |
| 855 | if (holder) { |
| 856 | holder->refcnt++; |
| 857 | goto out_unlock; |
| 858 | } |
| 859 | |
| 860 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
| 861 | if (!holder) { |
| 862 | ret = -ENOMEM; |
| 863 | goto out_unlock; |
| 864 | } |
| 865 | |
| 866 | INIT_LIST_HEAD(&holder->list); |
| 867 | holder->disk = disk; |
| 868 | holder->refcnt = 1; |
| 869 | |
| 870 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 871 | if (ret) |
| 872 | goto out_free; |
| 873 | |
| 874 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); |
| 875 | if (ret) |
| 876 | goto out_del; |
| 877 | /* |
| 878 | * bdev could be deleted beneath us which would implicitly destroy |
| 879 | * the holder directory. Hold on to it. |
| 880 | */ |
| 881 | kobject_get(bdev->bd_part->holder_dir); |
| 882 | |
| 883 | list_add(&holder->list, &bdev->bd_holder_disks); |
| 884 | goto out_unlock; |
| 885 | |
| 886 | out_del: |
| 887 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 888 | out_free: |
| 889 | kfree(holder); |
| 890 | out_unlock: |
| 891 | mutex_unlock(&bdev->bd_mutex); |
| 892 | return ret; |
| 893 | } |
| 894 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
| 895 | |
| 896 | /** |
| 897 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() |
| 898 | * @bdev: the calimed slave bdev |
| 899 | * @disk: the holding disk |
| 900 | * |
| 901 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
| 902 | * |
| 903 | * CONTEXT: |
| 904 | * Might sleep. |
| 905 | */ |
| 906 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) |
| 907 | { |
| 908 | struct bd_holder_disk *holder; |
| 909 | |
| 910 | mutex_lock(&bdev->bd_mutex); |
| 911 | |
| 912 | holder = bd_find_holder_disk(bdev, disk); |
| 913 | |
| 914 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { |
| 915 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 916 | del_symlink(bdev->bd_part->holder_dir, |
| 917 | &disk_to_dev(disk)->kobj); |
| 918 | kobject_put(bdev->bd_part->holder_dir); |
| 919 | list_del_init(&holder->list); |
| 920 | kfree(holder); |
| 921 | } |
| 922 | |
| 923 | mutex_unlock(&bdev->bd_mutex); |
| 924 | } |
| 925 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
| 926 | #endif |
| 927 | |
| 928 | /** |
| 929 | * flush_disk - invalidates all buffer-cache entries on a disk |
| 930 | * |
| 931 | * @bdev: struct block device to be flushed |
| 932 | * @kill_dirty: flag to guide handling of dirty inodes |
| 933 | * |
| 934 | * Invalidates all buffer-cache entries on a disk. It should be called |
| 935 | * when a disk has been changed -- either by a media change or online |
| 936 | * resize. |
| 937 | */ |
| 938 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
| 939 | { |
| 940 | if (__invalidate_device(bdev, kill_dirty)) { |
| 941 | char name[BDEVNAME_SIZE] = ""; |
| 942 | |
| 943 | if (bdev->bd_disk) |
| 944 | disk_name(bdev->bd_disk, 0, name); |
| 945 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
| 946 | "resized disk %s\n", name); |
| 947 | } |
| 948 | |
| 949 | if (!bdev->bd_disk) |
| 950 | return; |
| 951 | if (disk_part_scan_enabled(bdev->bd_disk)) |
| 952 | bdev->bd_invalidated = 1; |
| 953 | } |
| 954 | |
| 955 | /** |
| 956 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
| 957 | * @disk: struct gendisk to check |
| 958 | * @bdev: struct bdev to adjust. |
| 959 | * |
| 960 | * This routine checks to see if the bdev size does not match the disk size |
| 961 | * and adjusts it if it differs. |
| 962 | */ |
| 963 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) |
| 964 | { |
| 965 | loff_t disk_size, bdev_size; |
| 966 | |
| 967 | disk_size = (loff_t)get_capacity(disk) << 9; |
| 968 | bdev_size = i_size_read(bdev->bd_inode); |
| 969 | if (disk_size != bdev_size) { |
| 970 | char name[BDEVNAME_SIZE]; |
| 971 | |
| 972 | disk_name(disk, 0, name); |
| 973 | printk(KERN_INFO |
| 974 | "%s: detected capacity change from %lld to %lld\n", |
| 975 | name, bdev_size, disk_size); |
| 976 | i_size_write(bdev->bd_inode, disk_size); |
| 977 | flush_disk(bdev, false); |
| 978 | } |
| 979 | } |
| 980 | EXPORT_SYMBOL(check_disk_size_change); |
| 981 | |
| 982 | /** |
| 983 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
| 984 | * @disk: struct gendisk to be revalidated |
| 985 | * |
| 986 | * This routine is a wrapper for lower-level driver's revalidate_disk |
| 987 | * call-backs. It is used to do common pre and post operations needed |
| 988 | * for all revalidate_disk operations. |
| 989 | */ |
| 990 | int revalidate_disk(struct gendisk *disk) |
| 991 | { |
| 992 | struct block_device *bdev; |
| 993 | int ret = 0; |
| 994 | |
| 995 | if (disk->fops->revalidate_disk) |
| 996 | ret = disk->fops->revalidate_disk(disk); |
| 997 | |
| 998 | bdev = bdget_disk(disk, 0); |
| 999 | if (!bdev) |
| 1000 | return ret; |
| 1001 | |
| 1002 | mutex_lock(&bdev->bd_mutex); |
| 1003 | check_disk_size_change(disk, bdev); |
| 1004 | bdev->bd_invalidated = 0; |
| 1005 | mutex_unlock(&bdev->bd_mutex); |
| 1006 | bdput(bdev); |
| 1007 | return ret; |
| 1008 | } |
| 1009 | EXPORT_SYMBOL(revalidate_disk); |
| 1010 | |
| 1011 | /* |
| 1012 | * This routine checks whether a removable media has been changed, |
| 1013 | * and invalidates all buffer-cache-entries in that case. This |
| 1014 | * is a relatively slow routine, so we have to try to minimize using |
| 1015 | * it. Thus it is called only upon a 'mount' or 'open'. This |
| 1016 | * is the best way of combining speed and utility, I think. |
| 1017 | * People changing diskettes in the middle of an operation deserve |
| 1018 | * to lose :-) |
| 1019 | */ |
| 1020 | int check_disk_change(struct block_device *bdev) |
| 1021 | { |
| 1022 | struct gendisk *disk = bdev->bd_disk; |
| 1023 | const struct block_device_operations *bdops = disk->fops; |
| 1024 | unsigned int events; |
| 1025 | |
| 1026 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
| 1027 | #ifdef CONFIG_MTK_MULTI_PARTITION_MOUNT_ONLY_SUPPORT |
| 1028 | DISK_EVENT_EJECT_REQUEST | DISK_EVENT_MEDIA_DISAPPEAR); //add DISK_EVENT_MEDIA_DISAPPEAR for sd hotplug |
| 1029 | #else |
| 1030 | DISK_EVENT_EJECT_REQUEST); |
| 1031 | #endif |
| 1032 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) |
| 1033 | return 0; |
| 1034 | |
| 1035 | flush_disk(bdev, true); |
| 1036 | if (bdops->revalidate_disk) |
| 1037 | bdops->revalidate_disk(bdev->bd_disk); |
| 1038 | return 1; |
| 1039 | } |
| 1040 | |
| 1041 | EXPORT_SYMBOL(check_disk_change); |
| 1042 | |
| 1043 | void bd_set_size(struct block_device *bdev, loff_t size) |
| 1044 | { |
| 1045 | unsigned bsize = bdev_logical_block_size(bdev); |
| 1046 | |
| 1047 | mutex_lock(&bdev->bd_inode->i_mutex); |
| 1048 | i_size_write(bdev->bd_inode, size); |
| 1049 | mutex_unlock(&bdev->bd_inode->i_mutex); |
| 1050 | while (bsize < PAGE_CACHE_SIZE) { |
| 1051 | if (size & bsize) |
| 1052 | break; |
| 1053 | bsize <<= 1; |
| 1054 | } |
| 1055 | bdev->bd_block_size = bsize; |
| 1056 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); |
| 1057 | } |
| 1058 | EXPORT_SYMBOL(bd_set_size); |
| 1059 | |
| 1060 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
| 1061 | |
| 1062 | /* |
| 1063 | * bd_mutex locking: |
| 1064 | * |
| 1065 | * mutex_lock(part->bd_mutex) |
| 1066 | * mutex_lock_nested(whole->bd_mutex, 1) |
| 1067 | */ |
| 1068 | |
| 1069 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
| 1070 | { |
| 1071 | struct gendisk *disk; |
| 1072 | struct module *owner; |
| 1073 | int ret; |
| 1074 | int partno; |
| 1075 | int perm = 0; |
| 1076 | |
| 1077 | if (mode & FMODE_READ) |
| 1078 | perm |= MAY_READ; |
| 1079 | if (mode & FMODE_WRITE) |
| 1080 | perm |= MAY_WRITE; |
| 1081 | /* |
| 1082 | * hooks: /n/, see "layering violations". |
| 1083 | */ |
| 1084 | if (!for_part) { |
| 1085 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); |
| 1086 | if (ret != 0) { |
| 1087 | bdput(bdev); |
| 1088 | return ret; |
| 1089 | } |
| 1090 | } |
| 1091 | |
| 1092 | restart: |
| 1093 | |
| 1094 | ret = -ENXIO; |
| 1095 | disk = get_gendisk(bdev->bd_dev, &partno); |
| 1096 | if (!disk) |
| 1097 | goto out; |
| 1098 | owner = disk->fops->owner; |
| 1099 | |
| 1100 | disk_block_events(disk); |
| 1101 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
| 1102 | if (!bdev->bd_openers) { |
| 1103 | bdev->bd_disk = disk; |
| 1104 | bdev->bd_queue = disk->queue; |
| 1105 | bdev->bd_contains = bdev; |
| 1106 | if (!partno) { |
| 1107 | struct backing_dev_info *bdi; |
| 1108 | |
| 1109 | ret = -ENXIO; |
| 1110 | bdev->bd_part = disk_get_part(disk, partno); |
| 1111 | if (!bdev->bd_part) |
| 1112 | goto out_clear; |
| 1113 | |
| 1114 | ret = 0; |
| 1115 | if (disk->fops->open) { |
| 1116 | ret = disk->fops->open(bdev, mode); |
| 1117 | if (ret == -ERESTARTSYS) { |
| 1118 | /* Lost a race with 'disk' being |
| 1119 | * deleted, try again. |
| 1120 | * See md.c |
| 1121 | */ |
| 1122 | disk_put_part(bdev->bd_part); |
| 1123 | bdev->bd_part = NULL; |
| 1124 | bdev->bd_disk = NULL; |
| 1125 | bdev->bd_queue = NULL; |
| 1126 | mutex_unlock(&bdev->bd_mutex); |
| 1127 | disk_unblock_events(disk); |
| 1128 | put_disk(disk); |
| 1129 | module_put(owner); |
| 1130 | goto restart; |
| 1131 | } |
| 1132 | } |
| 1133 | |
| 1134 | if (!ret) { |
| 1135 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
| 1136 | bdi = blk_get_backing_dev_info(bdev); |
| 1137 | if (bdi == NULL) |
| 1138 | bdi = &default_backing_dev_info; |
| 1139 | bdev_inode_switch_bdi(bdev->bd_inode, bdi); |
| 1140 | } |
| 1141 | |
| 1142 | /* |
| 1143 | * If the device is invalidated, rescan partition |
| 1144 | * if open succeeded or failed with -ENOMEDIUM. |
| 1145 | * The latter is necessary to prevent ghost |
| 1146 | * partitions on a removed medium. |
| 1147 | */ |
| 1148 | if (bdev->bd_invalidated) { |
| 1149 | if (!ret) |
| 1150 | rescan_partitions(disk, bdev); |
| 1151 | else if (ret == -ENOMEDIUM) |
| 1152 | invalidate_partitions(disk, bdev); |
| 1153 | } |
| 1154 | if (ret) |
| 1155 | goto out_clear; |
| 1156 | } else { |
| 1157 | struct block_device *whole; |
| 1158 | whole = bdget_disk(disk, 0); |
| 1159 | ret = -ENOMEM; |
| 1160 | if (!whole) |
| 1161 | goto out_clear; |
| 1162 | BUG_ON(for_part); |
| 1163 | ret = __blkdev_get(whole, mode, 1); |
| 1164 | if (ret) |
| 1165 | goto out_clear; |
| 1166 | bdev->bd_contains = whole; |
| 1167 | bdev_inode_switch_bdi(bdev->bd_inode, |
| 1168 | whole->bd_inode->i_data.backing_dev_info); |
| 1169 | bdev->bd_part = disk_get_part(disk, partno); |
| 1170 | if (!(disk->flags & GENHD_FL_UP) || |
| 1171 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
| 1172 | ret = -ENXIO; |
| 1173 | goto out_clear; |
| 1174 | } |
| 1175 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
| 1176 | } |
| 1177 | } else { |
| 1178 | if (bdev->bd_contains == bdev) { |
| 1179 | ret = 0; |
| 1180 | if (bdev->bd_disk->fops->open) |
| 1181 | ret = bdev->bd_disk->fops->open(bdev, mode); |
| 1182 | /* the same as first opener case, read comment there */ |
| 1183 | if (bdev->bd_invalidated) { |
| 1184 | if (!ret) |
| 1185 | rescan_partitions(bdev->bd_disk, bdev); |
| 1186 | else if (ret == -ENOMEDIUM) |
| 1187 | invalidate_partitions(bdev->bd_disk, bdev); |
| 1188 | } |
| 1189 | if (ret) |
| 1190 | goto out_unlock_bdev; |
| 1191 | } |
| 1192 | /* only one opener holds refs to the module and disk */ |
| 1193 | put_disk(disk); |
| 1194 | module_put(owner); |
| 1195 | } |
| 1196 | bdev->bd_openers++; |
| 1197 | if (for_part) |
| 1198 | bdev->bd_part_count++; |
| 1199 | mutex_unlock(&bdev->bd_mutex); |
| 1200 | disk_unblock_events(disk); |
| 1201 | return 0; |
| 1202 | |
| 1203 | out_clear: |
| 1204 | disk_put_part(bdev->bd_part); |
| 1205 | bdev->bd_disk = NULL; |
| 1206 | bdev->bd_part = NULL; |
| 1207 | bdev->bd_queue = NULL; |
| 1208 | bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info); |
| 1209 | if (bdev != bdev->bd_contains) |
| 1210 | __blkdev_put(bdev->bd_contains, mode, 1); |
| 1211 | bdev->bd_contains = NULL; |
| 1212 | out_unlock_bdev: |
| 1213 | mutex_unlock(&bdev->bd_mutex); |
| 1214 | disk_unblock_events(disk); |
| 1215 | put_disk(disk); |
| 1216 | module_put(owner); |
| 1217 | out: |
| 1218 | bdput(bdev); |
| 1219 | |
| 1220 | return ret; |
| 1221 | } |
| 1222 | |
| 1223 | /** |
| 1224 | * blkdev_get - open a block device |
| 1225 | * @bdev: block_device to open |
| 1226 | * @mode: FMODE_* mask |
| 1227 | * @holder: exclusive holder identifier |
| 1228 | * |
| 1229 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is |
| 1230 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL |
| 1231 | * @holder is invalid. Exclusive opens may nest for the same @holder. |
| 1232 | * |
| 1233 | * On success, the reference count of @bdev is unchanged. On failure, |
| 1234 | * @bdev is put. |
| 1235 | * |
| 1236 | * CONTEXT: |
| 1237 | * Might sleep. |
| 1238 | * |
| 1239 | * RETURNS: |
| 1240 | * 0 on success, -errno on failure. |
| 1241 | */ |
| 1242 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
| 1243 | { |
| 1244 | struct block_device *whole = NULL; |
| 1245 | int res; |
| 1246 | |
| 1247 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); |
| 1248 | |
| 1249 | if ((mode & FMODE_EXCL) && holder) { |
| 1250 | whole = bd_start_claiming(bdev, holder); |
| 1251 | if (IS_ERR(whole)) { |
| 1252 | bdput(bdev); |
| 1253 | return PTR_ERR(whole); |
| 1254 | } |
| 1255 | } |
| 1256 | |
| 1257 | res = __blkdev_get(bdev, mode, 0); |
| 1258 | |
| 1259 | if (whole) { |
| 1260 | struct gendisk *disk = whole->bd_disk; |
| 1261 | |
| 1262 | /* finish claiming */ |
| 1263 | mutex_lock(&bdev->bd_mutex); |
| 1264 | spin_lock(&bdev_lock); |
| 1265 | |
| 1266 | if (!res) { |
| 1267 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
| 1268 | /* |
| 1269 | * Note that for a whole device bd_holders |
| 1270 | * will be incremented twice, and bd_holder |
| 1271 | * will be set to bd_may_claim before being |
| 1272 | * set to holder |
| 1273 | */ |
| 1274 | whole->bd_holders++; |
| 1275 | whole->bd_holder = bd_may_claim; |
| 1276 | bdev->bd_holders++; |
| 1277 | bdev->bd_holder = holder; |
| 1278 | } |
| 1279 | |
| 1280 | /* tell others that we're done */ |
| 1281 | BUG_ON(whole->bd_claiming != holder); |
| 1282 | whole->bd_claiming = NULL; |
| 1283 | wake_up_bit(&whole->bd_claiming, 0); |
| 1284 | |
| 1285 | spin_unlock(&bdev_lock); |
| 1286 | |
| 1287 | /* |
| 1288 | * Block event polling for write claims if requested. Any |
| 1289 | * write holder makes the write_holder state stick until |
| 1290 | * all are released. This is good enough and tracking |
| 1291 | * individual writeable reference is too fragile given the |
| 1292 | * way @mode is used in blkdev_get/put(). |
| 1293 | */ |
| 1294 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
| 1295 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { |
| 1296 | bdev->bd_write_holder = true; |
| 1297 | disk_block_events(disk); |
| 1298 | } |
| 1299 | |
| 1300 | mutex_unlock(&bdev->bd_mutex); |
| 1301 | bdput(whole); |
| 1302 | } |
| 1303 | |
| 1304 | return res; |
| 1305 | } |
| 1306 | EXPORT_SYMBOL(blkdev_get); |
| 1307 | |
| 1308 | /** |
| 1309 | * blkdev_get_by_path - open a block device by name |
| 1310 | * @path: path to the block device to open |
| 1311 | * @mode: FMODE_* mask |
| 1312 | * @holder: exclusive holder identifier |
| 1313 | * |
| 1314 | * Open the blockdevice described by the device file at @path. @mode |
| 1315 | * and @holder are identical to blkdev_get(). |
| 1316 | * |
| 1317 | * On success, the returned block_device has reference count of one. |
| 1318 | * |
| 1319 | * CONTEXT: |
| 1320 | * Might sleep. |
| 1321 | * |
| 1322 | * RETURNS: |
| 1323 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
| 1324 | */ |
| 1325 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, |
| 1326 | void *holder) |
| 1327 | { |
| 1328 | struct block_device *bdev; |
| 1329 | int err; |
| 1330 | |
| 1331 | bdev = lookup_bdev(path); |
| 1332 | if (IS_ERR(bdev)) |
| 1333 | return bdev; |
| 1334 | |
| 1335 | err = blkdev_get(bdev, mode, holder); |
| 1336 | if (err) |
| 1337 | return ERR_PTR(err); |
| 1338 | |
| 1339 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
| 1340 | blkdev_put(bdev, mode); |
| 1341 | return ERR_PTR(-EACCES); |
| 1342 | } |
| 1343 | |
| 1344 | return bdev; |
| 1345 | } |
| 1346 | EXPORT_SYMBOL(blkdev_get_by_path); |
| 1347 | |
| 1348 | /** |
| 1349 | * blkdev_get_by_dev - open a block device by device number |
| 1350 | * @dev: device number of block device to open |
| 1351 | * @mode: FMODE_* mask |
| 1352 | * @holder: exclusive holder identifier |
| 1353 | * |
| 1354 | * Open the blockdevice described by device number @dev. @mode and |
| 1355 | * @holder are identical to blkdev_get(). |
| 1356 | * |
| 1357 | * Use it ONLY if you really do not have anything better - i.e. when |
| 1358 | * you are behind a truly sucky interface and all you are given is a |
| 1359 | * device number. _Never_ to be used for internal purposes. If you |
| 1360 | * ever need it - reconsider your API. |
| 1361 | * |
| 1362 | * On success, the returned block_device has reference count of one. |
| 1363 | * |
| 1364 | * CONTEXT: |
| 1365 | * Might sleep. |
| 1366 | * |
| 1367 | * RETURNS: |
| 1368 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
| 1369 | */ |
| 1370 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) |
| 1371 | { |
| 1372 | struct block_device *bdev; |
| 1373 | int err; |
| 1374 | |
| 1375 | bdev = bdget(dev); |
| 1376 | if (!bdev) |
| 1377 | return ERR_PTR(-ENOMEM); |
| 1378 | |
| 1379 | err = blkdev_get(bdev, mode, holder); |
| 1380 | if (err) |
| 1381 | return ERR_PTR(err); |
| 1382 | |
| 1383 | return bdev; |
| 1384 | } |
| 1385 | EXPORT_SYMBOL(blkdev_get_by_dev); |
| 1386 | |
| 1387 | static int blkdev_open(struct inode * inode, struct file * filp) |
| 1388 | { |
| 1389 | struct block_device *bdev; |
| 1390 | |
| 1391 | /* |
| 1392 | * Preserve backwards compatibility and allow large file access |
| 1393 | * even if userspace doesn't ask for it explicitly. Some mkfs |
| 1394 | * binary needs it. We might want to drop this workaround |
| 1395 | * during an unstable branch. |
| 1396 | */ |
| 1397 | filp->f_flags |= O_LARGEFILE; |
| 1398 | |
| 1399 | if (filp->f_flags & O_NDELAY) |
| 1400 | filp->f_mode |= FMODE_NDELAY; |
| 1401 | if (filp->f_flags & O_EXCL) |
| 1402 | filp->f_mode |= FMODE_EXCL; |
| 1403 | if ((filp->f_flags & O_ACCMODE) == 3) |
| 1404 | filp->f_mode |= FMODE_WRITE_IOCTL; |
| 1405 | |
| 1406 | bdev = bd_acquire(inode); |
| 1407 | if (bdev == NULL) |
| 1408 | return -ENOMEM; |
| 1409 | |
| 1410 | filp->f_mapping = bdev->bd_inode->i_mapping; |
| 1411 | |
| 1412 | return blkdev_get(bdev, filp->f_mode, filp); |
| 1413 | } |
| 1414 | |
| 1415 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
| 1416 | { |
| 1417 | struct gendisk *disk = bdev->bd_disk; |
| 1418 | struct block_device *victim = NULL; |
| 1419 | |
| 1420 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
| 1421 | if (for_part) |
| 1422 | bdev->bd_part_count--; |
| 1423 | |
| 1424 | if (!--bdev->bd_openers) { |
| 1425 | WARN_ON_ONCE(bdev->bd_holders); |
| 1426 | sync_blockdev(bdev); |
| 1427 | kill_bdev(bdev); |
| 1428 | /* ->release can cause the old bdi to disappear, |
| 1429 | * so must switch it out first |
| 1430 | */ |
| 1431 | bdev_inode_switch_bdi(bdev->bd_inode, |
| 1432 | &default_backing_dev_info); |
| 1433 | } |
| 1434 | if (bdev->bd_contains == bdev) { |
| 1435 | if (disk->fops->release) |
| 1436 | disk->fops->release(disk, mode); |
| 1437 | } |
| 1438 | if (!bdev->bd_openers) { |
| 1439 | struct module *owner = disk->fops->owner; |
| 1440 | |
| 1441 | disk_put_part(bdev->bd_part); |
| 1442 | bdev->bd_part = NULL; |
| 1443 | bdev->bd_disk = NULL; |
| 1444 | if (bdev != bdev->bd_contains) |
| 1445 | victim = bdev->bd_contains; |
| 1446 | bdev->bd_contains = NULL; |
| 1447 | |
| 1448 | put_disk(disk); |
| 1449 | module_put(owner); |
| 1450 | } |
| 1451 | mutex_unlock(&bdev->bd_mutex); |
| 1452 | bdput(bdev); |
| 1453 | if (victim) |
| 1454 | __blkdev_put(victim, mode, 1); |
| 1455 | } |
| 1456 | |
| 1457 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
| 1458 | { |
| 1459 | mutex_lock(&bdev->bd_mutex); |
| 1460 | |
| 1461 | if (mode & FMODE_EXCL) { |
| 1462 | bool bdev_free; |
| 1463 | |
| 1464 | /* |
| 1465 | * Release a claim on the device. The holder fields |
| 1466 | * are protected with bdev_lock. bd_mutex is to |
| 1467 | * synchronize disk_holder unlinking. |
| 1468 | */ |
| 1469 | spin_lock(&bdev_lock); |
| 1470 | |
| 1471 | WARN_ON_ONCE(--bdev->bd_holders < 0); |
| 1472 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); |
| 1473 | |
| 1474 | /* bd_contains might point to self, check in a separate step */ |
| 1475 | if ((bdev_free = !bdev->bd_holders)) |
| 1476 | bdev->bd_holder = NULL; |
| 1477 | if (!bdev->bd_contains->bd_holders) |
| 1478 | bdev->bd_contains->bd_holder = NULL; |
| 1479 | |
| 1480 | spin_unlock(&bdev_lock); |
| 1481 | |
| 1482 | /* |
| 1483 | * If this was the last claim, remove holder link and |
| 1484 | * unblock evpoll if it was a write holder. |
| 1485 | */ |
| 1486 | if (bdev_free && bdev->bd_write_holder) { |
| 1487 | disk_unblock_events(bdev->bd_disk); |
| 1488 | bdev->bd_write_holder = false; |
| 1489 | } |
| 1490 | } |
| 1491 | |
| 1492 | /* |
| 1493 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE |
| 1494 | * event. This is to ensure detection of media removal commanded |
| 1495 | * from userland - e.g. eject(1). |
| 1496 | */ |
| 1497 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); |
| 1498 | |
| 1499 | mutex_unlock(&bdev->bd_mutex); |
| 1500 | |
| 1501 | __blkdev_put(bdev, mode, 0); |
| 1502 | } |
| 1503 | EXPORT_SYMBOL(blkdev_put); |
| 1504 | |
| 1505 | static int blkdev_close(struct inode * inode, struct file * filp) |
| 1506 | { |
| 1507 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); |
| 1508 | blkdev_put(bdev, filp->f_mode); |
| 1509 | return 0; |
| 1510 | } |
| 1511 | |
| 1512 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
| 1513 | { |
| 1514 | struct block_device *bdev = I_BDEV(file->f_mapping->host); |
| 1515 | fmode_t mode = file->f_mode; |
| 1516 | |
| 1517 | /* |
| 1518 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have |
| 1519 | * to updated it before every ioctl. |
| 1520 | */ |
| 1521 | if (file->f_flags & O_NDELAY) |
| 1522 | mode |= FMODE_NDELAY; |
| 1523 | else |
| 1524 | mode &= ~FMODE_NDELAY; |
| 1525 | |
| 1526 | return blkdev_ioctl(bdev, mode, cmd, arg); |
| 1527 | } |
| 1528 | |
| 1529 | /* |
| 1530 | * Write data to the block device. Only intended for the block device itself |
| 1531 | * and the raw driver which basically is a fake block device. |
| 1532 | * |
| 1533 | * Does not take i_mutex for the write and thus is not for general purpose |
| 1534 | * use. |
| 1535 | */ |
| 1536 | ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov, |
| 1537 | unsigned long nr_segs, loff_t pos) |
| 1538 | { |
| 1539 | struct file *file = iocb->ki_filp; |
| 1540 | struct blk_plug plug; |
| 1541 | ssize_t ret; |
| 1542 | |
| 1543 | BUG_ON(iocb->ki_pos != pos); |
| 1544 | |
| 1545 | blk_start_plug(&plug); |
| 1546 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
| 1547 | if (ret > 0 || ret == -EIOCBQUEUED) { |
| 1548 | ssize_t err; |
| 1549 | |
| 1550 | err = generic_write_sync(file, pos, ret); |
| 1551 | if (err < 0 && ret > 0) |
| 1552 | ret = err; |
| 1553 | } |
| 1554 | blk_finish_plug(&plug); |
| 1555 | return ret; |
| 1556 | } |
| 1557 | EXPORT_SYMBOL_GPL(blkdev_aio_write); |
| 1558 | |
| 1559 | static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov, |
| 1560 | unsigned long nr_segs, loff_t pos) |
| 1561 | { |
| 1562 | struct file *file = iocb->ki_filp; |
| 1563 | struct inode *bd_inode = file->f_mapping->host; |
| 1564 | loff_t size = i_size_read(bd_inode); |
| 1565 | |
| 1566 | if (pos >= size) |
| 1567 | return 0; |
| 1568 | |
| 1569 | size -= pos; |
| 1570 | if (size < iocb->ki_left) |
| 1571 | nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size); |
| 1572 | return generic_file_aio_read(iocb, iov, nr_segs, pos); |
| 1573 | } |
| 1574 | |
| 1575 | /* |
| 1576 | * Try to release a page associated with block device when the system |
| 1577 | * is under memory pressure. |
| 1578 | */ |
| 1579 | static int blkdev_releasepage(struct page *page, gfp_t wait) |
| 1580 | { |
| 1581 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; |
| 1582 | |
| 1583 | if (super && super->s_op->bdev_try_to_free_page) |
| 1584 | return super->s_op->bdev_try_to_free_page(super, page, wait); |
| 1585 | |
| 1586 | return try_to_free_buffers(page); |
| 1587 | } |
| 1588 | |
| 1589 | static const struct address_space_operations def_blk_aops = { |
| 1590 | .readpage = blkdev_readpage, |
| 1591 | .writepage = blkdev_writepage, |
| 1592 | .write_begin = blkdev_write_begin, |
| 1593 | .write_end = blkdev_write_end, |
| 1594 | .writepages = generic_writepages, |
| 1595 | .releasepage = blkdev_releasepage, |
| 1596 | .direct_IO = blkdev_direct_IO, |
| 1597 | }; |
| 1598 | |
| 1599 | const struct file_operations def_blk_fops = { |
| 1600 | .open = blkdev_open, |
| 1601 | .release = blkdev_close, |
| 1602 | .llseek = block_llseek, |
| 1603 | .read = do_sync_read, |
| 1604 | .write = do_sync_write, |
| 1605 | .aio_read = blkdev_aio_read, |
| 1606 | .aio_write = blkdev_aio_write, |
| 1607 | .mmap = generic_file_mmap, |
| 1608 | .fsync = blkdev_fsync, |
| 1609 | .unlocked_ioctl = block_ioctl, |
| 1610 | #ifdef CONFIG_COMPAT |
| 1611 | .compat_ioctl = compat_blkdev_ioctl, |
| 1612 | #endif |
| 1613 | .splice_read = generic_file_splice_read, |
| 1614 | .splice_write = generic_file_splice_write, |
| 1615 | }; |
| 1616 | |
| 1617 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) |
| 1618 | { |
| 1619 | int res; |
| 1620 | mm_segment_t old_fs = get_fs(); |
| 1621 | set_fs(KERNEL_DS); |
| 1622 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
| 1623 | set_fs(old_fs); |
| 1624 | return res; |
| 1625 | } |
| 1626 | |
| 1627 | EXPORT_SYMBOL(ioctl_by_bdev); |
| 1628 | |
| 1629 | /** |
| 1630 | * lookup_bdev - lookup a struct block_device by name |
| 1631 | * @pathname: special file representing the block device |
| 1632 | * |
| 1633 | * Get a reference to the blockdevice at @pathname in the current |
| 1634 | * namespace if possible and return it. Return ERR_PTR(error) |
| 1635 | * otherwise. |
| 1636 | */ |
| 1637 | struct block_device *lookup_bdev(const char *pathname) |
| 1638 | { |
| 1639 | struct block_device *bdev; |
| 1640 | struct inode *inode; |
| 1641 | struct path path; |
| 1642 | int error; |
| 1643 | |
| 1644 | if (!pathname || !*pathname) |
| 1645 | return ERR_PTR(-EINVAL); |
| 1646 | |
| 1647 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
| 1648 | if (error) |
| 1649 | return ERR_PTR(error); |
| 1650 | |
| 1651 | inode = path.dentry->d_inode; |
| 1652 | error = -ENOTBLK; |
| 1653 | if (!S_ISBLK(inode->i_mode)) |
| 1654 | goto fail; |
| 1655 | error = -EACCES; |
| 1656 | if (path.mnt->mnt_flags & MNT_NODEV) |
| 1657 | goto fail; |
| 1658 | error = -ENOMEM; |
| 1659 | bdev = bd_acquire(inode); |
| 1660 | if (!bdev) |
| 1661 | goto fail; |
| 1662 | out: |
| 1663 | path_put(&path); |
| 1664 | return bdev; |
| 1665 | fail: |
| 1666 | bdev = ERR_PTR(error); |
| 1667 | goto out; |
| 1668 | } |
| 1669 | EXPORT_SYMBOL(lookup_bdev); |
| 1670 | |
| 1671 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
| 1672 | { |
| 1673 | struct super_block *sb = get_super(bdev); |
| 1674 | int res = 0; |
| 1675 | |
| 1676 | if (sb) { |
| 1677 | /* |
| 1678 | * no need to lock the super, get_super holds the |
| 1679 | * read mutex so the filesystem cannot go away |
| 1680 | * under us (->put_super runs with the write lock |
| 1681 | * hold). |
| 1682 | */ |
| 1683 | shrink_dcache_sb(sb); |
| 1684 | res = invalidate_inodes(sb, kill_dirty); |
| 1685 | drop_super(sb); |
| 1686 | } |
| 1687 | invalidate_bdev(bdev); |
| 1688 | return res; |
| 1689 | } |
| 1690 | EXPORT_SYMBOL(__invalidate_device); |
| 1691 | |
| 1692 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) |
| 1693 | { |
| 1694 | struct inode *inode, *old_inode = NULL; |
| 1695 | |
| 1696 | spin_lock(&inode_sb_list_lock); |
| 1697 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
| 1698 | struct address_space *mapping = inode->i_mapping; |
| 1699 | struct block_device *bdev; |
| 1700 | |
| 1701 | spin_lock(&inode->i_lock); |
| 1702 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || |
| 1703 | mapping->nrpages == 0) { |
| 1704 | spin_unlock(&inode->i_lock); |
| 1705 | continue; |
| 1706 | } |
| 1707 | __iget(inode); |
| 1708 | spin_unlock(&inode->i_lock); |
| 1709 | spin_unlock(&inode_sb_list_lock); |
| 1710 | /* |
| 1711 | * We hold a reference to 'inode' so it couldn't have been |
| 1712 | * removed from s_inodes list while we dropped the |
| 1713 | * inode_sb_list_lock. We cannot iput the inode now as we can |
| 1714 | * be holding the last reference and we cannot iput it under |
| 1715 | * inode_sb_list_lock. So we keep the reference and iput it |
| 1716 | * later. |
| 1717 | */ |
| 1718 | iput(old_inode); |
| 1719 | old_inode = inode; |
| 1720 | bdev = I_BDEV(inode); |
| 1721 | |
| 1722 | mutex_lock(&bdev->bd_mutex); |
| 1723 | if (bdev->bd_openers) |
| 1724 | func(bdev, arg); |
| 1725 | mutex_unlock(&bdev->bd_mutex); |
| 1726 | |
| 1727 | spin_lock(&inode_sb_list_lock); |
| 1728 | } |
| 1729 | spin_unlock(&inode_sb_list_lock); |
| 1730 | iput(old_inode); |
| 1731 | } |