Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
1da177e4 LT |
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> | |
1da177e4 | 14 | #include <linux/smp_lock.h> |
7db9cfd3 | 15 | #include <linux/device_cgroup.h> |
1da177e4 LT |
16 | #include <linux/highmem.h> |
17 | #include <linux/blkdev.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/blkpg.h> | |
20 | #include <linux/buffer_head.h> | |
585d3bc0 | 21 | #include <linux/pagevec.h> |
811d736f | 22 | #include <linux/writeback.h> |
1da177e4 LT |
23 | #include <linux/mpage.h> |
24 | #include <linux/mount.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/namei.h> | |
1368c4f2 | 27 | #include <linux/log2.h> |
1da177e4 | 28 | #include <asm/uaccess.h> |
07f3f05c | 29 | #include "internal.h" |
1da177e4 LT |
30 | |
31 | struct bdev_inode { | |
32 | struct block_device bdev; | |
33 | struct inode vfs_inode; | |
34 | }; | |
35 | ||
4c54ac62 AB |
36 | static const struct address_space_operations def_blk_aops; |
37 | ||
1da177e4 LT |
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 | inline struct block_device *I_BDEV(struct inode *inode) | |
44 | { | |
45 | return &BDEV_I(inode)->bdev; | |
46 | } | |
47 | ||
48 | EXPORT_SYMBOL(I_BDEV); | |
49 | ||
50 | static sector_t max_block(struct block_device *bdev) | |
51 | { | |
52 | sector_t retval = ~((sector_t)0); | |
53 | loff_t sz = i_size_read(bdev->bd_inode); | |
54 | ||
55 | if (sz) { | |
56 | unsigned int size = block_size(bdev); | |
57 | unsigned int sizebits = blksize_bits(size); | |
58 | retval = (sz >> sizebits); | |
59 | } | |
60 | return retval; | |
61 | } | |
62 | ||
f9a14399 | 63 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
1da177e4 LT |
64 | static void kill_bdev(struct block_device *bdev) |
65 | { | |
f9a14399 PZ |
66 | if (bdev->bd_inode->i_mapping->nrpages == 0) |
67 | return; | |
68 | invalidate_bh_lrus(); | |
1da177e4 LT |
69 | truncate_inode_pages(bdev->bd_inode->i_mapping, 0); |
70 | } | |
71 | ||
72 | int set_blocksize(struct block_device *bdev, int size) | |
73 | { | |
74 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 75 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
76 | return -EINVAL; |
77 | ||
78 | /* Size cannot be smaller than the size supported by the device */ | |
79 | if (size < bdev_hardsect_size(bdev)) | |
80 | return -EINVAL; | |
81 | ||
82 | /* Don't change the size if it is same as current */ | |
83 | if (bdev->bd_block_size != size) { | |
84 | sync_blockdev(bdev); | |
85 | bdev->bd_block_size = size; | |
86 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
87 | kill_bdev(bdev); | |
88 | } | |
89 | return 0; | |
90 | } | |
91 | ||
92 | EXPORT_SYMBOL(set_blocksize); | |
93 | ||
94 | int sb_set_blocksize(struct super_block *sb, int size) | |
95 | { | |
1da177e4 LT |
96 | if (set_blocksize(sb->s_bdev, size)) |
97 | return 0; | |
98 | /* If we get here, we know size is power of two | |
99 | * and it's value is between 512 and PAGE_SIZE */ | |
100 | sb->s_blocksize = size; | |
38885bd4 | 101 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
102 | return sb->s_blocksize; |
103 | } | |
104 | ||
105 | EXPORT_SYMBOL(sb_set_blocksize); | |
106 | ||
107 | int sb_min_blocksize(struct super_block *sb, int size) | |
108 | { | |
109 | int minsize = bdev_hardsect_size(sb->s_bdev); | |
110 | if (size < minsize) | |
111 | size = minsize; | |
112 | return sb_set_blocksize(sb, size); | |
113 | } | |
114 | ||
115 | EXPORT_SYMBOL(sb_min_blocksize); | |
116 | ||
117 | static int | |
118 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
119 | struct buffer_head *bh, int create) | |
120 | { | |
121 | if (iblock >= max_block(I_BDEV(inode))) { | |
122 | if (create) | |
123 | return -EIO; | |
124 | ||
125 | /* | |
126 | * for reads, we're just trying to fill a partial page. | |
127 | * return a hole, they will have to call get_block again | |
128 | * before they can fill it, and they will get -EIO at that | |
129 | * time | |
130 | */ | |
131 | return 0; | |
132 | } | |
133 | bh->b_bdev = I_BDEV(inode); | |
134 | bh->b_blocknr = iblock; | |
135 | set_buffer_mapped(bh); | |
136 | return 0; | |
137 | } | |
138 | ||
b2e895db AM |
139 | static int |
140 | blkdev_get_blocks(struct inode *inode, sector_t iblock, | |
141 | struct buffer_head *bh, int create) | |
142 | { | |
143 | sector_t end_block = max_block(I_BDEV(inode)); | |
144 | unsigned long max_blocks = bh->b_size >> inode->i_blkbits; | |
145 | ||
146 | if ((iblock + max_blocks) > end_block) { | |
147 | max_blocks = end_block - iblock; | |
148 | if ((long)max_blocks <= 0) { | |
149 | if (create) | |
150 | return -EIO; /* write fully beyond EOF */ | |
151 | /* | |
152 | * It is a read which is fully beyond EOF. We return | |
153 | * a !buffer_mapped buffer | |
154 | */ | |
155 | max_blocks = 0; | |
156 | } | |
157 | } | |
158 | ||
159 | bh->b_bdev = I_BDEV(inode); | |
160 | bh->b_blocknr = iblock; | |
161 | bh->b_size = max_blocks << inode->i_blkbits; | |
162 | if (max_blocks) | |
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_no_locking(rw, iocb, inode, I_BDEV(inode), | |
175 | iov, offset, nr_segs, blkdev_get_blocks, NULL); | |
176 | } | |
177 | ||
585d3bc0 NP |
178 | /* |
179 | * Write out and wait upon all the dirty data associated with a block | |
180 | * device via its mapping. Does not take the superblock lock. | |
181 | */ | |
182 | int sync_blockdev(struct block_device *bdev) | |
183 | { | |
184 | int ret = 0; | |
185 | ||
186 | if (bdev) | |
187 | ret = filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
188 | return ret; | |
189 | } | |
190 | EXPORT_SYMBOL(sync_blockdev); | |
191 | ||
192 | /* | |
193 | * Write out and wait upon all dirty data associated with this | |
194 | * device. Filesystem data as well as the underlying block | |
195 | * device. Takes the superblock lock. | |
196 | */ | |
197 | int fsync_bdev(struct block_device *bdev) | |
198 | { | |
199 | struct super_block *sb = get_super(bdev); | |
200 | if (sb) { | |
201 | int res = fsync_super(sb); | |
202 | drop_super(sb); | |
203 | return res; | |
204 | } | |
205 | return sync_blockdev(bdev); | |
206 | } | |
47e4491b | 207 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
208 | |
209 | /** | |
210 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
211 | * @bdev: blockdevice to lock | |
212 | * | |
213 | * This takes the block device bd_mount_sem to make sure no new mounts | |
214 | * happen on bdev until thaw_bdev() is called. | |
215 | * If a superblock is found on this device, we take the s_umount semaphore | |
216 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
217 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
218 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
219 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
220 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
221 | * actually. | |
222 | */ | |
223 | struct super_block *freeze_bdev(struct block_device *bdev) | |
224 | { | |
225 | struct super_block *sb; | |
226 | int error = 0; | |
227 | ||
228 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
229 | if (bdev->bd_fsfreeze_count > 0) { | |
230 | bdev->bd_fsfreeze_count++; | |
231 | sb = get_super(bdev); | |
232 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
233 | return sb; | |
234 | } | |
235 | bdev->bd_fsfreeze_count++; | |
236 | ||
237 | down(&bdev->bd_mount_sem); | |
238 | sb = get_super(bdev); | |
239 | if (sb && !(sb->s_flags & MS_RDONLY)) { | |
240 | sb->s_frozen = SB_FREEZE_WRITE; | |
241 | smp_wmb(); | |
242 | ||
243 | __fsync_super(sb); | |
244 | ||
245 | sb->s_frozen = SB_FREEZE_TRANS; | |
246 | smp_wmb(); | |
247 | ||
248 | sync_blockdev(sb->s_bdev); | |
249 | ||
250 | if (sb->s_op->freeze_fs) { | |
251 | error = sb->s_op->freeze_fs(sb); | |
252 | if (error) { | |
253 | printk(KERN_ERR | |
254 | "VFS:Filesystem freeze failed\n"); | |
255 | sb->s_frozen = SB_UNFROZEN; | |
256 | drop_super(sb); | |
257 | up(&bdev->bd_mount_sem); | |
258 | bdev->bd_fsfreeze_count--; | |
259 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
260 | return ERR_PTR(error); | |
261 | } | |
262 | } | |
263 | } | |
264 | ||
265 | sync_blockdev(bdev); | |
266 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
267 | ||
268 | return sb; /* thaw_bdev releases s->s_umount and bd_mount_sem */ | |
269 | } | |
270 | EXPORT_SYMBOL(freeze_bdev); | |
271 | ||
272 | /** | |
273 | * thaw_bdev -- unlock filesystem | |
274 | * @bdev: blockdevice to unlock | |
275 | * @sb: associated superblock | |
276 | * | |
277 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
278 | */ | |
279 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
280 | { | |
281 | int error = 0; | |
282 | ||
283 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
284 | if (!bdev->bd_fsfreeze_count) { | |
285 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
286 | return -EINVAL; | |
287 | } | |
288 | ||
289 | bdev->bd_fsfreeze_count--; | |
290 | if (bdev->bd_fsfreeze_count > 0) { | |
291 | if (sb) | |
292 | drop_super(sb); | |
293 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
294 | return 0; | |
295 | } | |
296 | ||
297 | if (sb) { | |
298 | BUG_ON(sb->s_bdev != bdev); | |
299 | if (!(sb->s_flags & MS_RDONLY)) { | |
300 | if (sb->s_op->unfreeze_fs) { | |
301 | error = sb->s_op->unfreeze_fs(sb); | |
302 | if (error) { | |
303 | printk(KERN_ERR | |
304 | "VFS:Filesystem thaw failed\n"); | |
305 | sb->s_frozen = SB_FREEZE_TRANS; | |
306 | bdev->bd_fsfreeze_count++; | |
307 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
308 | return error; | |
309 | } | |
310 | } | |
311 | sb->s_frozen = SB_UNFROZEN; | |
312 | smp_wmb(); | |
313 | wake_up(&sb->s_wait_unfrozen); | |
314 | } | |
315 | drop_super(sb); | |
316 | } | |
317 | ||
318 | up(&bdev->bd_mount_sem); | |
319 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
320 | return 0; | |
321 | } | |
322 | EXPORT_SYMBOL(thaw_bdev); | |
323 | ||
1da177e4 LT |
324 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
325 | { | |
326 | return block_write_full_page(page, blkdev_get_block, wbc); | |
327 | } | |
328 | ||
329 | static int blkdev_readpage(struct file * file, struct page * page) | |
330 | { | |
331 | return block_read_full_page(page, blkdev_get_block); | |
332 | } | |
333 | ||
db2dbb12 JM |
334 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
335 | struct list_head *pages, unsigned nr_pages) | |
336 | { | |
337 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
338 | } | |
339 | ||
6272b5a5 NP |
340 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
341 | loff_t pos, unsigned len, unsigned flags, | |
342 | struct page **pagep, void **fsdata) | |
1da177e4 | 343 | { |
6272b5a5 NP |
344 | *pagep = NULL; |
345 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
346 | blkdev_get_block); | |
1da177e4 LT |
347 | } |
348 | ||
6272b5a5 NP |
349 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
350 | loff_t pos, unsigned len, unsigned copied, | |
351 | struct page *page, void *fsdata) | |
1da177e4 | 352 | { |
6272b5a5 NP |
353 | int ret; |
354 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
355 | ||
356 | unlock_page(page); | |
357 | page_cache_release(page); | |
358 | ||
359 | return ret; | |
1da177e4 LT |
360 | } |
361 | ||
362 | /* | |
363 | * private llseek: | |
0f7fc9e4 | 364 | * for a block special file file->f_path.dentry->d_inode->i_size is zero |
1da177e4 LT |
365 | * so we compute the size by hand (just as in block_read/write above) |
366 | */ | |
367 | static loff_t block_llseek(struct file *file, loff_t offset, int origin) | |
368 | { | |
369 | struct inode *bd_inode = file->f_mapping->host; | |
370 | loff_t size; | |
371 | loff_t retval; | |
372 | ||
1b1dcc1b | 373 | mutex_lock(&bd_inode->i_mutex); |
1da177e4 LT |
374 | size = i_size_read(bd_inode); |
375 | ||
376 | switch (origin) { | |
377 | case 2: | |
378 | offset += size; | |
379 | break; | |
380 | case 1: | |
381 | offset += file->f_pos; | |
382 | } | |
383 | retval = -EINVAL; | |
384 | if (offset >= 0 && offset <= size) { | |
385 | if (offset != file->f_pos) { | |
386 | file->f_pos = offset; | |
387 | } | |
388 | retval = offset; | |
389 | } | |
1b1dcc1b | 390 | mutex_unlock(&bd_inode->i_mutex); |
1da177e4 LT |
391 | return retval; |
392 | } | |
393 | ||
394 | /* | |
395 | * Filp is never NULL; the only case when ->fsync() is called with | |
396 | * NULL first argument is nfsd_sync_dir() and that's not a directory. | |
397 | */ | |
398 | ||
399 | static int block_fsync(struct file *filp, struct dentry *dentry, int datasync) | |
400 | { | |
401 | return sync_blockdev(I_BDEV(filp->f_mapping->host)); | |
402 | } | |
403 | ||
404 | /* | |
405 | * pseudo-fs | |
406 | */ | |
407 | ||
408 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 409 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
410 | |
411 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
412 | { | |
e94b1766 | 413 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
414 | if (!ei) |
415 | return NULL; | |
416 | return &ei->vfs_inode; | |
417 | } | |
418 | ||
419 | static void bdev_destroy_inode(struct inode *inode) | |
420 | { | |
421 | struct bdev_inode *bdi = BDEV_I(inode); | |
422 | ||
423 | bdi->bdev.bd_inode_backing_dev_info = NULL; | |
424 | kmem_cache_free(bdev_cachep, bdi); | |
425 | } | |
426 | ||
51cc5068 | 427 | static void init_once(void *foo) |
1da177e4 LT |
428 | { |
429 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
430 | struct block_device *bdev = &ei->bdev; | |
431 | ||
a35afb83 CL |
432 | memset(bdev, 0, sizeof(*bdev)); |
433 | mutex_init(&bdev->bd_mutex); | |
434 | sema_init(&bdev->bd_mount_sem, 1); | |
435 | INIT_LIST_HEAD(&bdev->bd_inodes); | |
436 | INIT_LIST_HEAD(&bdev->bd_list); | |
641dc636 | 437 | #ifdef CONFIG_SYSFS |
a35afb83 | 438 | INIT_LIST_HEAD(&bdev->bd_holder_list); |
641dc636 | 439 | #endif |
a35afb83 | 440 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
441 | /* Initialize mutex for freeze. */ |
442 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
443 | } |
444 | ||
445 | static inline void __bd_forget(struct inode *inode) | |
446 | { | |
447 | list_del_init(&inode->i_devices); | |
448 | inode->i_bdev = NULL; | |
449 | inode->i_mapping = &inode->i_data; | |
450 | } | |
451 | ||
452 | static void bdev_clear_inode(struct inode *inode) | |
453 | { | |
454 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
455 | struct list_head *p; | |
456 | spin_lock(&bdev_lock); | |
457 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
458 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
459 | } | |
460 | list_del_init(&bdev->bd_list); | |
461 | spin_unlock(&bdev_lock); | |
462 | } | |
463 | ||
ee9b6d61 | 464 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
465 | .statfs = simple_statfs, |
466 | .alloc_inode = bdev_alloc_inode, | |
467 | .destroy_inode = bdev_destroy_inode, | |
468 | .drop_inode = generic_delete_inode, | |
469 | .clear_inode = bdev_clear_inode, | |
470 | }; | |
471 | ||
454e2398 DH |
472 | static int bd_get_sb(struct file_system_type *fs_type, |
473 | int flags, const char *dev_name, void *data, struct vfsmount *mnt) | |
1da177e4 | 474 | { |
454e2398 | 475 | return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt); |
1da177e4 LT |
476 | } |
477 | ||
478 | static struct file_system_type bd_type = { | |
479 | .name = "bdev", | |
480 | .get_sb = bd_get_sb, | |
481 | .kill_sb = kill_anon_super, | |
482 | }; | |
483 | ||
c2acf7b9 | 484 | struct super_block *blockdev_superblock __read_mostly; |
1da177e4 LT |
485 | |
486 | void __init bdev_cache_init(void) | |
487 | { | |
488 | int err; | |
c2acf7b9 DC |
489 | struct vfsmount *bd_mnt; |
490 | ||
1da177e4 | 491 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 PJ |
492 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
493 | SLAB_MEM_SPREAD|SLAB_PANIC), | |
20c2df83 | 494 | init_once); |
1da177e4 LT |
495 | err = register_filesystem(&bd_type); |
496 | if (err) | |
497 | panic("Cannot register bdev pseudo-fs"); | |
498 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
499 | if (IS_ERR(bd_mnt)) |
500 | panic("Cannot create bdev pseudo-fs"); | |
501 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ | |
502 | } | |
503 | ||
504 | /* | |
505 | * Most likely _very_ bad one - but then it's hardly critical for small | |
506 | * /dev and can be fixed when somebody will need really large one. | |
507 | * Keep in mind that it will be fed through icache hash function too. | |
508 | */ | |
509 | static inline unsigned long hash(dev_t dev) | |
510 | { | |
511 | return MAJOR(dev)+MINOR(dev); | |
512 | } | |
513 | ||
514 | static int bdev_test(struct inode *inode, void *data) | |
515 | { | |
516 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
517 | } | |
518 | ||
519 | static int bdev_set(struct inode *inode, void *data) | |
520 | { | |
521 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
522 | return 0; | |
523 | } | |
524 | ||
525 | static LIST_HEAD(all_bdevs); | |
526 | ||
527 | struct block_device *bdget(dev_t dev) | |
528 | { | |
529 | struct block_device *bdev; | |
530 | struct inode *inode; | |
531 | ||
c2acf7b9 | 532 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
533 | bdev_test, bdev_set, &dev); |
534 | ||
535 | if (!inode) | |
536 | return NULL; | |
537 | ||
538 | bdev = &BDEV_I(inode)->bdev; | |
539 | ||
540 | if (inode->i_state & I_NEW) { | |
541 | bdev->bd_contains = NULL; | |
542 | bdev->bd_inode = inode; | |
543 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
544 | bdev->bd_part_count = 0; | |
545 | bdev->bd_invalidated = 0; | |
546 | inode->i_mode = S_IFBLK; | |
547 | inode->i_rdev = dev; | |
548 | inode->i_bdev = bdev; | |
549 | inode->i_data.a_ops = &def_blk_aops; | |
550 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
551 | inode->i_data.backing_dev_info = &default_backing_dev_info; | |
552 | spin_lock(&bdev_lock); | |
553 | list_add(&bdev->bd_list, &all_bdevs); | |
554 | spin_unlock(&bdev_lock); | |
555 | unlock_new_inode(inode); | |
556 | } | |
557 | return bdev; | |
558 | } | |
559 | ||
560 | EXPORT_SYMBOL(bdget); | |
561 | ||
562 | long nr_blockdev_pages(void) | |
563 | { | |
203a2935 | 564 | struct block_device *bdev; |
1da177e4 LT |
565 | long ret = 0; |
566 | spin_lock(&bdev_lock); | |
203a2935 | 567 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
568 | ret += bdev->bd_inode->i_mapping->nrpages; |
569 | } | |
570 | spin_unlock(&bdev_lock); | |
571 | return ret; | |
572 | } | |
573 | ||
574 | void bdput(struct block_device *bdev) | |
575 | { | |
576 | iput(bdev->bd_inode); | |
577 | } | |
578 | ||
579 | EXPORT_SYMBOL(bdput); | |
580 | ||
581 | static struct block_device *bd_acquire(struct inode *inode) | |
582 | { | |
583 | struct block_device *bdev; | |
09d967c6 | 584 | |
1da177e4 LT |
585 | spin_lock(&bdev_lock); |
586 | bdev = inode->i_bdev; | |
09d967c6 OH |
587 | if (bdev) { |
588 | atomic_inc(&bdev->bd_inode->i_count); | |
1da177e4 LT |
589 | spin_unlock(&bdev_lock); |
590 | return bdev; | |
591 | } | |
592 | spin_unlock(&bdev_lock); | |
09d967c6 | 593 | |
1da177e4 LT |
594 | bdev = bdget(inode->i_rdev); |
595 | if (bdev) { | |
596 | spin_lock(&bdev_lock); | |
09d967c6 OH |
597 | if (!inode->i_bdev) { |
598 | /* | |
599 | * We take an additional bd_inode->i_count for inode, | |
600 | * and it's released in clear_inode() of inode. | |
601 | * So, we can access it via ->i_mapping always | |
602 | * without igrab(). | |
603 | */ | |
604 | atomic_inc(&bdev->bd_inode->i_count); | |
605 | inode->i_bdev = bdev; | |
606 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
607 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
608 | } | |
1da177e4 LT |
609 | spin_unlock(&bdev_lock); |
610 | } | |
611 | return bdev; | |
612 | } | |
613 | ||
614 | /* Call when you free inode */ | |
615 | ||
616 | void bd_forget(struct inode *inode) | |
617 | { | |
09d967c6 OH |
618 | struct block_device *bdev = NULL; |
619 | ||
1da177e4 | 620 | spin_lock(&bdev_lock); |
09d967c6 | 621 | if (inode->i_bdev) { |
c2acf7b9 | 622 | if (!sb_is_blkdev_sb(inode->i_sb)) |
09d967c6 | 623 | bdev = inode->i_bdev; |
1da177e4 | 624 | __bd_forget(inode); |
09d967c6 | 625 | } |
1da177e4 | 626 | spin_unlock(&bdev_lock); |
09d967c6 OH |
627 | |
628 | if (bdev) | |
629 | iput(bdev->bd_inode); | |
1da177e4 LT |
630 | } |
631 | ||
632 | int bd_claim(struct block_device *bdev, void *holder) | |
633 | { | |
634 | int res; | |
635 | spin_lock(&bdev_lock); | |
636 | ||
637 | /* first decide result */ | |
638 | if (bdev->bd_holder == holder) | |
639 | res = 0; /* already a holder */ | |
640 | else if (bdev->bd_holder != NULL) | |
641 | res = -EBUSY; /* held by someone else */ | |
642 | else if (bdev->bd_contains == bdev) | |
643 | res = 0; /* is a whole device which isn't held */ | |
644 | ||
645 | else if (bdev->bd_contains->bd_holder == bd_claim) | |
646 | res = 0; /* is a partition of a device that is being partitioned */ | |
647 | else if (bdev->bd_contains->bd_holder != NULL) | |
648 | res = -EBUSY; /* is a partition of a held device */ | |
649 | else | |
650 | res = 0; /* is a partition of an un-held device */ | |
651 | ||
652 | /* now impose change */ | |
653 | if (res==0) { | |
654 | /* note that for a whole device bd_holders | |
655 | * will be incremented twice, and bd_holder will | |
656 | * be set to bd_claim before being set to holder | |
657 | */ | |
658 | bdev->bd_contains->bd_holders ++; | |
659 | bdev->bd_contains->bd_holder = bd_claim; | |
660 | bdev->bd_holders++; | |
661 | bdev->bd_holder = holder; | |
662 | } | |
663 | spin_unlock(&bdev_lock); | |
664 | return res; | |
665 | } | |
666 | ||
667 | EXPORT_SYMBOL(bd_claim); | |
668 | ||
669 | void bd_release(struct block_device *bdev) | |
670 | { | |
671 | spin_lock(&bdev_lock); | |
672 | if (!--bdev->bd_contains->bd_holders) | |
673 | bdev->bd_contains->bd_holder = NULL; | |
674 | if (!--bdev->bd_holders) | |
675 | bdev->bd_holder = NULL; | |
676 | spin_unlock(&bdev_lock); | |
677 | } | |
678 | ||
679 | EXPORT_SYMBOL(bd_release); | |
680 | ||
641dc636 JN |
681 | #ifdef CONFIG_SYSFS |
682 | /* | |
683 | * Functions for bd_claim_by_kobject / bd_release_from_kobject | |
684 | * | |
685 | * If a kobject is passed to bd_claim_by_kobject() | |
686 | * and the kobject has a parent directory, | |
687 | * following symlinks are created: | |
688 | * o from the kobject to the claimed bdev | |
689 | * o from "holders" directory of the bdev to the parent of the kobject | |
690 | * bd_release_from_kobject() removes these symlinks. | |
691 | * | |
692 | * Example: | |
693 | * If /dev/dm-0 maps to /dev/sda, kobject corresponding to | |
694 | * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then: | |
695 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda | |
696 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
697 | */ | |
698 | ||
4d7dd8fd | 699 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 JN |
700 | { |
701 | if (!from || !to) | |
4d7dd8fd AM |
702 | return 0; |
703 | return sysfs_create_link(from, to, kobject_name(to)); | |
641dc636 JN |
704 | } |
705 | ||
706 | static void del_symlink(struct kobject *from, struct kobject *to) | |
707 | { | |
708 | if (!from || !to) | |
709 | return; | |
710 | sysfs_remove_link(from, kobject_name(to)); | |
711 | } | |
712 | ||
713 | /* | |
714 | * 'struct bd_holder' contains pointers to kobjects symlinked by | |
715 | * bd_claim_by_kobject. | |
716 | * It's connected to bd_holder_list which is protected by bdev->bd_sem. | |
717 | */ | |
718 | struct bd_holder { | |
719 | struct list_head list; /* chain of holders of the bdev */ | |
720 | int count; /* references from the holder */ | |
721 | struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */ | |
722 | struct kobject *hdev; /* e.g. "/block/dm-0" */ | |
723 | struct kobject *hdir; /* e.g. "/block/sda/holders" */ | |
724 | struct kobject *sdev; /* e.g. "/block/sda" */ | |
725 | }; | |
726 | ||
727 | /* | |
728 | * Get references of related kobjects at once. | |
729 | * Returns 1 on success. 0 on failure. | |
730 | * | |
731 | * Should call bd_holder_release_dirs() after successful use. | |
732 | */ | |
733 | static int bd_holder_grab_dirs(struct block_device *bdev, | |
734 | struct bd_holder *bo) | |
735 | { | |
736 | if (!bdev || !bo) | |
737 | return 0; | |
738 | ||
739 | bo->sdir = kobject_get(bo->sdir); | |
740 | if (!bo->sdir) | |
741 | return 0; | |
742 | ||
743 | bo->hdev = kobject_get(bo->sdir->parent); | |
744 | if (!bo->hdev) | |
745 | goto fail_put_sdir; | |
746 | ||
0762b8bd | 747 | bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj); |
641dc636 JN |
748 | if (!bo->sdev) |
749 | goto fail_put_hdev; | |
750 | ||
4c46501d | 751 | bo->hdir = kobject_get(bdev->bd_part->holder_dir); |
641dc636 JN |
752 | if (!bo->hdir) |
753 | goto fail_put_sdev; | |
754 | ||
755 | return 1; | |
756 | ||
757 | fail_put_sdev: | |
758 | kobject_put(bo->sdev); | |
759 | fail_put_hdev: | |
760 | kobject_put(bo->hdev); | |
761 | fail_put_sdir: | |
762 | kobject_put(bo->sdir); | |
763 | ||
764 | return 0; | |
765 | } | |
766 | ||
767 | /* Put references of related kobjects at once. */ | |
768 | static void bd_holder_release_dirs(struct bd_holder *bo) | |
769 | { | |
770 | kobject_put(bo->hdir); | |
771 | kobject_put(bo->sdev); | |
772 | kobject_put(bo->hdev); | |
773 | kobject_put(bo->sdir); | |
774 | } | |
775 | ||
776 | static struct bd_holder *alloc_bd_holder(struct kobject *kobj) | |
777 | { | |
778 | struct bd_holder *bo; | |
779 | ||
780 | bo = kzalloc(sizeof(*bo), GFP_KERNEL); | |
781 | if (!bo) | |
782 | return NULL; | |
783 | ||
784 | bo->count = 1; | |
785 | bo->sdir = kobj; | |
786 | ||
787 | return bo; | |
788 | } | |
789 | ||
790 | static void free_bd_holder(struct bd_holder *bo) | |
791 | { | |
792 | kfree(bo); | |
793 | } | |
794 | ||
df6c0cd9 JN |
795 | /** |
796 | * find_bd_holder - find matching struct bd_holder from the block device | |
797 | * | |
798 | * @bdev: struct block device to be searched | |
799 | * @bo: target struct bd_holder | |
800 | * | |
801 | * Returns matching entry with @bo in @bdev->bd_holder_list. | |
802 | * If found, increment the reference count and return the pointer. | |
803 | * If not found, returns NULL. | |
804 | */ | |
36a561d6 AM |
805 | static struct bd_holder *find_bd_holder(struct block_device *bdev, |
806 | struct bd_holder *bo) | |
df6c0cd9 JN |
807 | { |
808 | struct bd_holder *tmp; | |
809 | ||
810 | list_for_each_entry(tmp, &bdev->bd_holder_list, list) | |
811 | if (tmp->sdir == bo->sdir) { | |
812 | tmp->count++; | |
813 | return tmp; | |
814 | } | |
815 | ||
816 | return NULL; | |
817 | } | |
818 | ||
641dc636 JN |
819 | /** |
820 | * add_bd_holder - create sysfs symlinks for bd_claim() relationship | |
821 | * | |
822 | * @bdev: block device to be bd_claimed | |
823 | * @bo: preallocated and initialized by alloc_bd_holder() | |
824 | * | |
df6c0cd9 | 825 | * Add @bo to @bdev->bd_holder_list, create symlinks. |
641dc636 | 826 | * |
df6c0cd9 JN |
827 | * Returns 0 if symlinks are created. |
828 | * Returns -ve if something fails. | |
641dc636 JN |
829 | */ |
830 | static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo) | |
831 | { | |
4e91672c | 832 | int err; |
641dc636 JN |
833 | |
834 | if (!bo) | |
4d7dd8fd | 835 | return -EINVAL; |
641dc636 | 836 | |
641dc636 | 837 | if (!bd_holder_grab_dirs(bdev, bo)) |
4d7dd8fd | 838 | return -EBUSY; |
641dc636 | 839 | |
4e91672c JW |
840 | err = add_symlink(bo->sdir, bo->sdev); |
841 | if (err) | |
842 | return err; | |
843 | ||
844 | err = add_symlink(bo->hdir, bo->hdev); | |
845 | if (err) { | |
846 | del_symlink(bo->sdir, bo->sdev); | |
847 | return err; | |
4d7dd8fd | 848 | } |
4e91672c JW |
849 | |
850 | list_add_tail(&bo->list, &bdev->bd_holder_list); | |
851 | return 0; | |
641dc636 JN |
852 | } |
853 | ||
854 | /** | |
855 | * del_bd_holder - delete sysfs symlinks for bd_claim() relationship | |
856 | * | |
857 | * @bdev: block device to be bd_claimed | |
858 | * @kobj: holder's kobject | |
859 | * | |
860 | * If there is matching entry with @kobj in @bdev->bd_holder_list | |
861 | * and no other bd_claim() from the same kobject, | |
862 | * remove the struct bd_holder from the list, delete symlinks for it. | |
863 | * | |
864 | * Returns a pointer to the struct bd_holder when it's removed from the list | |
865 | * and ready to be freed. | |
866 | * Returns NULL if matching claim isn't found or there is other bd_claim() | |
867 | * by the same kobject. | |
868 | */ | |
869 | static struct bd_holder *del_bd_holder(struct block_device *bdev, | |
870 | struct kobject *kobj) | |
871 | { | |
872 | struct bd_holder *bo; | |
873 | ||
874 | list_for_each_entry(bo, &bdev->bd_holder_list, list) { | |
875 | if (bo->sdir == kobj) { | |
876 | bo->count--; | |
877 | BUG_ON(bo->count < 0); | |
878 | if (!bo->count) { | |
879 | list_del(&bo->list); | |
880 | del_symlink(bo->sdir, bo->sdev); | |
881 | del_symlink(bo->hdir, bo->hdev); | |
882 | bd_holder_release_dirs(bo); | |
883 | return bo; | |
884 | } | |
885 | break; | |
886 | } | |
887 | } | |
888 | ||
889 | return NULL; | |
890 | } | |
891 | ||
892 | /** | |
893 | * bd_claim_by_kobject - bd_claim() with additional kobject signature | |
894 | * | |
895 | * @bdev: block device to be claimed | |
896 | * @holder: holder's signature | |
897 | * @kobj: holder's kobject | |
898 | * | |
899 | * Do bd_claim() and if it succeeds, create sysfs symlinks between | |
900 | * the bdev and the holder's kobject. | |
901 | * Use bd_release_from_kobject() when relesing the claimed bdev. | |
902 | * | |
903 | * Returns 0 on success. (same as bd_claim()) | |
904 | * Returns errno on failure. | |
905 | */ | |
906 | static int bd_claim_by_kobject(struct block_device *bdev, void *holder, | |
907 | struct kobject *kobj) | |
908 | { | |
4e91672c | 909 | int err; |
df6c0cd9 | 910 | struct bd_holder *bo, *found; |
641dc636 JN |
911 | |
912 | if (!kobj) | |
913 | return -EINVAL; | |
914 | ||
915 | bo = alloc_bd_holder(kobj); | |
916 | if (!bo) | |
917 | return -ENOMEM; | |
918 | ||
2e7b651d | 919 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 920 | |
4e91672c JW |
921 | err = bd_claim(bdev, holder); |
922 | if (err) | |
4210df28 | 923 | goto fail; |
4e91672c JW |
924 | |
925 | found = find_bd_holder(bdev, bo); | |
926 | if (found) | |
4210df28 | 927 | goto fail; |
4e91672c JW |
928 | |
929 | err = add_bd_holder(bdev, bo); | |
930 | if (err) | |
931 | bd_release(bdev); | |
4210df28 AM |
932 | else |
933 | bo = NULL; | |
934 | fail: | |
b4cf1b72 | 935 | mutex_unlock(&bdev->bd_mutex); |
4210df28 | 936 | free_bd_holder(bo); |
4e91672c | 937 | return err; |
641dc636 JN |
938 | } |
939 | ||
940 | /** | |
941 | * bd_release_from_kobject - bd_release() with additional kobject signature | |
942 | * | |
943 | * @bdev: block device to be released | |
944 | * @kobj: holder's kobject | |
945 | * | |
946 | * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject(). | |
947 | */ | |
948 | static void bd_release_from_kobject(struct block_device *bdev, | |
949 | struct kobject *kobj) | |
950 | { | |
641dc636 JN |
951 | if (!kobj) |
952 | return; | |
953 | ||
2e7b651d | 954 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 955 | bd_release(bdev); |
4210df28 | 956 | free_bd_holder(del_bd_holder(bdev, kobj)); |
b4cf1b72 | 957 | mutex_unlock(&bdev->bd_mutex); |
641dc636 JN |
958 | } |
959 | ||
960 | /** | |
961 | * bd_claim_by_disk - wrapper function for bd_claim_by_kobject() | |
962 | * | |
963 | * @bdev: block device to be claimed | |
964 | * @holder: holder's signature | |
965 | * @disk: holder's gendisk | |
966 | * | |
967 | * Call bd_claim_by_kobject() with getting @disk->slave_dir. | |
968 | */ | |
969 | int bd_claim_by_disk(struct block_device *bdev, void *holder, | |
970 | struct gendisk *disk) | |
971 | { | |
972 | return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir)); | |
973 | } | |
974 | EXPORT_SYMBOL_GPL(bd_claim_by_disk); | |
975 | ||
976 | /** | |
977 | * bd_release_from_disk - wrapper function for bd_release_from_kobject() | |
978 | * | |
979 | * @bdev: block device to be claimed | |
980 | * @disk: holder's gendisk | |
981 | * | |
982 | * Call bd_release_from_kobject() and put @disk->slave_dir. | |
983 | */ | |
984 | void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk) | |
985 | { | |
986 | bd_release_from_kobject(bdev, disk->slave_dir); | |
987 | kobject_put(disk->slave_dir); | |
988 | } | |
989 | EXPORT_SYMBOL_GPL(bd_release_from_disk); | |
990 | #endif | |
991 | ||
1da177e4 LT |
992 | /* |
993 | * Tries to open block device by device number. Use it ONLY if you | |
994 | * really do not have anything better - i.e. when you are behind a | |
995 | * truly sucky interface and all you are given is a device number. _Never_ | |
996 | * to be used for internal purposes. If you ever need it - reconsider | |
997 | * your API. | |
998 | */ | |
aeb5d727 | 999 | struct block_device *open_by_devnum(dev_t dev, fmode_t mode) |
1da177e4 LT |
1000 | { |
1001 | struct block_device *bdev = bdget(dev); | |
1002 | int err = -ENOMEM; | |
1da177e4 | 1003 | if (bdev) |
572c4892 | 1004 | err = blkdev_get(bdev, mode); |
1da177e4 LT |
1005 | return err ? ERR_PTR(err) : bdev; |
1006 | } | |
1007 | ||
1008 | EXPORT_SYMBOL(open_by_devnum); | |
1009 | ||
56ade44b AP |
1010 | /** |
1011 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1012 | * | |
1013 | * @bdev: struct block device to be flushed | |
1014 | * | |
1015 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1016 | * when a disk has been changed -- either by a media change or online | |
1017 | * resize. | |
1018 | */ | |
1019 | static void flush_disk(struct block_device *bdev) | |
1020 | { | |
1021 | if (__invalidate_device(bdev)) { | |
1022 | char name[BDEVNAME_SIZE] = ""; | |
1023 | ||
1024 | if (bdev->bd_disk) | |
1025 | disk_name(bdev->bd_disk, 0, name); | |
1026 | printk(KERN_WARNING "VFS: busy inodes on changed media or " | |
1027 | "resized disk %s\n", name); | |
1028 | } | |
1029 | ||
1030 | if (!bdev->bd_disk) | |
1031 | return; | |
1032 | if (disk_partitionable(bdev->bd_disk)) | |
1033 | bdev->bd_invalidated = 1; | |
1034 | } | |
1035 | ||
c3279d14 | 1036 | /** |
57d1b536 | 1037 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1038 | * @disk: struct gendisk to check |
1039 | * @bdev: struct bdev to adjust. | |
1040 | * | |
1041 | * This routine checks to see if the bdev size does not match the disk size | |
1042 | * and adjusts it if it differs. | |
1043 | */ | |
1044 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1045 | { | |
1046 | loff_t disk_size, bdev_size; | |
1047 | ||
1048 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1049 | bdev_size = i_size_read(bdev->bd_inode); | |
1050 | if (disk_size != bdev_size) { | |
1051 | char name[BDEVNAME_SIZE]; | |
1052 | ||
1053 | disk_name(disk, 0, name); | |
1054 | printk(KERN_INFO | |
1055 | "%s: detected capacity change from %lld to %lld\n", | |
1056 | name, bdev_size, disk_size); | |
1057 | i_size_write(bdev->bd_inode, disk_size); | |
608aeef1 | 1058 | flush_disk(bdev); |
c3279d14 AP |
1059 | } |
1060 | } | |
1061 | EXPORT_SYMBOL(check_disk_size_change); | |
1062 | ||
0c002c2f | 1063 | /** |
57d1b536 | 1064 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1065 | * @disk: struct gendisk to be revalidated |
1066 | * | |
1067 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1068 | * call-backs. It is used to do common pre and post operations needed | |
1069 | * for all revalidate_disk operations. | |
1070 | */ | |
1071 | int revalidate_disk(struct gendisk *disk) | |
1072 | { | |
c3279d14 | 1073 | struct block_device *bdev; |
0c002c2f AP |
1074 | int ret = 0; |
1075 | ||
1076 | if (disk->fops->revalidate_disk) | |
1077 | ret = disk->fops->revalidate_disk(disk); | |
1078 | ||
c3279d14 AP |
1079 | bdev = bdget_disk(disk, 0); |
1080 | if (!bdev) | |
1081 | return ret; | |
1082 | ||
1083 | mutex_lock(&bdev->bd_mutex); | |
1084 | check_disk_size_change(disk, bdev); | |
1085 | mutex_unlock(&bdev->bd_mutex); | |
1086 | bdput(bdev); | |
0c002c2f AP |
1087 | return ret; |
1088 | } | |
1089 | EXPORT_SYMBOL(revalidate_disk); | |
1090 | ||
1da177e4 LT |
1091 | /* |
1092 | * This routine checks whether a removable media has been changed, | |
1093 | * and invalidates all buffer-cache-entries in that case. This | |
1094 | * is a relatively slow routine, so we have to try to minimize using | |
1095 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1096 | * is the best way of combining speed and utility, I think. | |
1097 | * People changing diskettes in the middle of an operation deserve | |
1098 | * to lose :-) | |
1099 | */ | |
1100 | int check_disk_change(struct block_device *bdev) | |
1101 | { | |
1102 | struct gendisk *disk = bdev->bd_disk; | |
1103 | struct block_device_operations * bdops = disk->fops; | |
1104 | ||
1105 | if (!bdops->media_changed) | |
1106 | return 0; | |
1107 | if (!bdops->media_changed(bdev->bd_disk)) | |
1108 | return 0; | |
1109 | ||
56ade44b | 1110 | flush_disk(bdev); |
1da177e4 LT |
1111 | if (bdops->revalidate_disk) |
1112 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1113 | return 1; |
1114 | } | |
1115 | ||
1116 | EXPORT_SYMBOL(check_disk_change); | |
1117 | ||
1118 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1119 | { | |
1120 | unsigned bsize = bdev_hardsect_size(bdev); | |
1121 | ||
1122 | bdev->bd_inode->i_size = size; | |
1123 | while (bsize < PAGE_CACHE_SIZE) { | |
1124 | if (size & bsize) | |
1125 | break; | |
1126 | bsize <<= 1; | |
1127 | } | |
1128 | bdev->bd_block_size = bsize; | |
1129 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1130 | } | |
1131 | EXPORT_SYMBOL(bd_set_size); | |
1132 | ||
9a1c3542 | 1133 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1134 | |
6d740cd5 PZ |
1135 | /* |
1136 | * bd_mutex locking: | |
1137 | * | |
1138 | * mutex_lock(part->bd_mutex) | |
1139 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1140 | */ | |
1141 | ||
572c4892 | 1142 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1143 | { |
1da177e4 | 1144 | struct gendisk *disk; |
7db9cfd3 | 1145 | int ret; |
cf771cb5 | 1146 | int partno; |
fe6e9c1f AV |
1147 | int perm = 0; |
1148 | ||
572c4892 | 1149 | if (mode & FMODE_READ) |
fe6e9c1f | 1150 | perm |= MAY_READ; |
572c4892 | 1151 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1152 | perm |= MAY_WRITE; |
1153 | /* | |
1154 | * hooks: /n/, see "layering violations". | |
1155 | */ | |
1156 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
82666020 AV |
1157 | if (ret != 0) { |
1158 | bdput(bdev); | |
7db9cfd3 | 1159 | return ret; |
82666020 | 1160 | } |
7db9cfd3 | 1161 | |
1da177e4 | 1162 | lock_kernel(); |
d3374825 | 1163 | restart: |
0762b8bd | 1164 | |
89f97496 | 1165 | ret = -ENXIO; |
cf771cb5 | 1166 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd TH |
1167 | if (!disk) |
1168 | goto out_unlock_kernel; | |
1da177e4 | 1169 | |
6796bf54 | 1170 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1171 | if (!bdev->bd_openers) { |
1172 | bdev->bd_disk = disk; | |
1173 | bdev->bd_contains = bdev; | |
cf771cb5 | 1174 | if (!partno) { |
1da177e4 | 1175 | struct backing_dev_info *bdi; |
89f97496 TH |
1176 | |
1177 | ret = -ENXIO; | |
1178 | bdev->bd_part = disk_get_part(disk, partno); | |
1179 | if (!bdev->bd_part) | |
1180 | goto out_clear; | |
1181 | ||
1da177e4 | 1182 | if (disk->fops->open) { |
572c4892 | 1183 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1184 | if (ret == -ERESTARTSYS) { |
1185 | /* Lost a race with 'disk' being | |
1186 | * deleted, try again. | |
1187 | * See md.c | |
1188 | */ | |
1189 | disk_put_part(bdev->bd_part); | |
1190 | bdev->bd_part = NULL; | |
1191 | module_put(disk->fops->owner); | |
1192 | put_disk(disk); | |
1193 | bdev->bd_disk = NULL; | |
1194 | mutex_unlock(&bdev->bd_mutex); | |
1195 | goto restart; | |
1196 | } | |
1da177e4 | 1197 | if (ret) |
0762b8bd | 1198 | goto out_clear; |
1da177e4 LT |
1199 | } |
1200 | if (!bdev->bd_openers) { | |
1201 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); | |
1202 | bdi = blk_get_backing_dev_info(bdev); | |
1203 | if (bdi == NULL) | |
1204 | bdi = &default_backing_dev_info; | |
1205 | bdev->bd_inode->i_data.backing_dev_info = bdi; | |
1206 | } | |
1207 | if (bdev->bd_invalidated) | |
1208 | rescan_partitions(disk, bdev); | |
1209 | } else { | |
1da177e4 LT |
1210 | struct block_device *whole; |
1211 | whole = bdget_disk(disk, 0); | |
1212 | ret = -ENOMEM; | |
1213 | if (!whole) | |
0762b8bd | 1214 | goto out_clear; |
37be4124 | 1215 | BUG_ON(for_part); |
572c4892 | 1216 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1217 | if (ret) |
0762b8bd | 1218 | goto out_clear; |
1da177e4 | 1219 | bdev->bd_contains = whole; |
1da177e4 LT |
1220 | bdev->bd_inode->i_data.backing_dev_info = |
1221 | whole->bd_inode->i_data.backing_dev_info; | |
89f97496 | 1222 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1223 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1224 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1225 | ret = -ENXIO; |
0762b8bd | 1226 | goto out_clear; |
1da177e4 | 1227 | } |
89f97496 | 1228 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1da177e4 LT |
1229 | } |
1230 | } else { | |
1231 | put_disk(disk); | |
0762b8bd | 1232 | module_put(disk->fops->owner); |
0762b8bd | 1233 | disk = NULL; |
1da177e4 LT |
1234 | if (bdev->bd_contains == bdev) { |
1235 | if (bdev->bd_disk->fops->open) { | |
572c4892 | 1236 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1da177e4 | 1237 | if (ret) |
0762b8bd | 1238 | goto out_unlock_bdev; |
1da177e4 LT |
1239 | } |
1240 | if (bdev->bd_invalidated) | |
1241 | rescan_partitions(bdev->bd_disk, bdev); | |
1da177e4 LT |
1242 | } |
1243 | } | |
1244 | bdev->bd_openers++; | |
37be4124 N |
1245 | if (for_part) |
1246 | bdev->bd_part_count++; | |
c039e313 | 1247 | mutex_unlock(&bdev->bd_mutex); |
1da177e4 LT |
1248 | unlock_kernel(); |
1249 | return 0; | |
1250 | ||
0762b8bd | 1251 | out_clear: |
89f97496 | 1252 | disk_put_part(bdev->bd_part); |
1da177e4 | 1253 | bdev->bd_disk = NULL; |
0762b8bd | 1254 | bdev->bd_part = NULL; |
1da177e4 LT |
1255 | bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info; |
1256 | if (bdev != bdev->bd_contains) | |
572c4892 | 1257 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1258 | bdev->bd_contains = NULL; |
0762b8bd | 1259 | out_unlock_bdev: |
c039e313 | 1260 | mutex_unlock(&bdev->bd_mutex); |
0762b8bd | 1261 | out_unlock_kernel: |
1da177e4 | 1262 | unlock_kernel(); |
0762b8bd | 1263 | |
0762b8bd TH |
1264 | if (disk) |
1265 | module_put(disk->fops->owner); | |
1266 | put_disk(disk); | |
1267 | bdput(bdev); | |
1268 | ||
1da177e4 LT |
1269 | return ret; |
1270 | } | |
1271 | ||
572c4892 | 1272 | int blkdev_get(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1273 | { |
572c4892 | 1274 | return __blkdev_get(bdev, mode, 0); |
37be4124 | 1275 | } |
1da177e4 LT |
1276 | EXPORT_SYMBOL(blkdev_get); |
1277 | ||
1278 | static int blkdev_open(struct inode * inode, struct file * filp) | |
1279 | { | |
1280 | struct block_device *bdev; | |
1281 | int res; | |
1282 | ||
1283 | /* | |
1284 | * Preserve backwards compatibility and allow large file access | |
1285 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1286 | * binary needs it. We might want to drop this workaround | |
1287 | * during an unstable branch. | |
1288 | */ | |
1289 | filp->f_flags |= O_LARGEFILE; | |
1290 | ||
572c4892 AV |
1291 | if (filp->f_flags & O_NDELAY) |
1292 | filp->f_mode |= FMODE_NDELAY; | |
1293 | if (filp->f_flags & O_EXCL) | |
1294 | filp->f_mode |= FMODE_EXCL; | |
1295 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1296 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1297 | ||
1da177e4 | 1298 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1299 | if (bdev == NULL) |
1300 | return -ENOMEM; | |
1da177e4 | 1301 | |
572c4892 AV |
1302 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1303 | ||
1304 | res = blkdev_get(bdev, filp->f_mode); | |
1da177e4 LT |
1305 | if (res) |
1306 | return res; | |
1307 | ||
ebbefc01 CH |
1308 | if (filp->f_mode & FMODE_EXCL) { |
1309 | res = bd_claim(bdev, filp); | |
1310 | if (res) | |
1311 | goto out_blkdev_put; | |
1312 | } | |
1da177e4 | 1313 | |
ebbefc01 | 1314 | return 0; |
1da177e4 | 1315 | |
ebbefc01 | 1316 | out_blkdev_put: |
9a1c3542 | 1317 | blkdev_put(bdev, filp->f_mode); |
1da177e4 LT |
1318 | return res; |
1319 | } | |
1320 | ||
9a1c3542 | 1321 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d PZ |
1322 | { |
1323 | int ret = 0; | |
2e7b651d | 1324 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1325 | struct block_device *victim = NULL; |
2e7b651d | 1326 | |
6796bf54 | 1327 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
2e7b651d | 1328 | lock_kernel(); |
37be4124 N |
1329 | if (for_part) |
1330 | bdev->bd_part_count--; | |
1331 | ||
2e7b651d PZ |
1332 | if (!--bdev->bd_openers) { |
1333 | sync_blockdev(bdev); | |
1334 | kill_bdev(bdev); | |
1335 | } | |
1336 | if (bdev->bd_contains == bdev) { | |
1337 | if (disk->fops->release) | |
9a1c3542 | 1338 | ret = disk->fops->release(disk, mode); |
2e7b651d PZ |
1339 | } |
1340 | if (!bdev->bd_openers) { | |
1341 | struct module *owner = disk->fops->owner; | |
1342 | ||
1343 | put_disk(disk); | |
1344 | module_put(owner); | |
0762b8bd TH |
1345 | disk_put_part(bdev->bd_part); |
1346 | bdev->bd_part = NULL; | |
2e7b651d PZ |
1347 | bdev->bd_disk = NULL; |
1348 | bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info; | |
37be4124 N |
1349 | if (bdev != bdev->bd_contains) |
1350 | victim = bdev->bd_contains; | |
2e7b651d PZ |
1351 | bdev->bd_contains = NULL; |
1352 | } | |
1353 | unlock_kernel(); | |
1354 | mutex_unlock(&bdev->bd_mutex); | |
1355 | bdput(bdev); | |
37be4124 | 1356 | if (victim) |
9a1c3542 | 1357 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1358 | return ret; |
1359 | } | |
1360 | ||
9a1c3542 | 1361 | int blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1362 | { |
9a1c3542 | 1363 | return __blkdev_put(bdev, mode, 0); |
37be4124 | 1364 | } |
2e7b651d PZ |
1365 | EXPORT_SYMBOL(blkdev_put); |
1366 | ||
1da177e4 LT |
1367 | static int blkdev_close(struct inode * inode, struct file * filp) |
1368 | { | |
1369 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); | |
1370 | if (bdev->bd_holder == filp) | |
1371 | bd_release(bdev); | |
9a1c3542 | 1372 | return blkdev_put(bdev, filp->f_mode); |
1da177e4 LT |
1373 | } |
1374 | ||
bb93e3a5 | 1375 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1376 | { |
56b26add AV |
1377 | struct block_device *bdev = I_BDEV(file->f_mapping->host); |
1378 | fmode_t mode = file->f_mode; | |
fd4ce1ac CH |
1379 | |
1380 | /* | |
1381 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1382 | * to updated it before every ioctl. | |
1383 | */ | |
56b26add | 1384 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1385 | mode |= FMODE_NDELAY; |
1386 | else | |
1387 | mode &= ~FMODE_NDELAY; | |
1388 | ||
56b26add | 1389 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1390 | } |
1391 | ||
87d8fe1e TT |
1392 | /* |
1393 | * Try to release a page associated with block device when the system | |
1394 | * is under memory pressure. | |
1395 | */ | |
1396 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1397 | { | |
1398 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1399 | ||
1400 | if (super && super->s_op->bdev_try_to_free_page) | |
1401 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1402 | ||
1403 | return try_to_free_buffers(page); | |
1404 | } | |
1405 | ||
4c54ac62 | 1406 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1407 | .readpage = blkdev_readpage, |
db2dbb12 | 1408 | .readpages = blkdev_readpages, |
1da177e4 LT |
1409 | .writepage = blkdev_writepage, |
1410 | .sync_page = block_sync_page, | |
6272b5a5 NP |
1411 | .write_begin = blkdev_write_begin, |
1412 | .write_end = blkdev_write_end, | |
1da177e4 | 1413 | .writepages = generic_writepages, |
87d8fe1e | 1414 | .releasepage = blkdev_releasepage, |
1da177e4 LT |
1415 | .direct_IO = blkdev_direct_IO, |
1416 | }; | |
1417 | ||
4b6f5d20 | 1418 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1419 | .open = blkdev_open, |
1420 | .release = blkdev_close, | |
1421 | .llseek = block_llseek, | |
543ade1f BP |
1422 | .read = do_sync_read, |
1423 | .write = do_sync_write, | |
1da177e4 | 1424 | .aio_read = generic_file_aio_read, |
027445c3 | 1425 | .aio_write = generic_file_aio_write_nolock, |
1da177e4 LT |
1426 | .mmap = generic_file_mmap, |
1427 | .fsync = block_fsync, | |
bb93e3a5 | 1428 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1429 | #ifdef CONFIG_COMPAT |
1430 | .compat_ioctl = compat_blkdev_ioctl, | |
1431 | #endif | |
7f9c51f0 JA |
1432 | .splice_read = generic_file_splice_read, |
1433 | .splice_write = generic_file_splice_write, | |
1da177e4 LT |
1434 | }; |
1435 | ||
1436 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1437 | { | |
1438 | int res; | |
1439 | mm_segment_t old_fs = get_fs(); | |
1440 | set_fs(KERNEL_DS); | |
56b26add | 1441 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1442 | set_fs(old_fs); |
1443 | return res; | |
1444 | } | |
1445 | ||
1446 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1447 | ||
1448 | /** | |
1449 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1450 | * @pathname: special file representing the block device |
1da177e4 | 1451 | * |
57d1b536 | 1452 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1453 | * namespace if possible and return it. Return ERR_PTR(error) |
1454 | * otherwise. | |
1455 | */ | |
421748ec | 1456 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1457 | { |
1458 | struct block_device *bdev; | |
1459 | struct inode *inode; | |
421748ec | 1460 | struct path path; |
1da177e4 LT |
1461 | int error; |
1462 | ||
421748ec | 1463 | if (!pathname || !*pathname) |
1da177e4 LT |
1464 | return ERR_PTR(-EINVAL); |
1465 | ||
421748ec | 1466 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1467 | if (error) |
1468 | return ERR_PTR(error); | |
1469 | ||
421748ec | 1470 | inode = path.dentry->d_inode; |
1da177e4 LT |
1471 | error = -ENOTBLK; |
1472 | if (!S_ISBLK(inode->i_mode)) | |
1473 | goto fail; | |
1474 | error = -EACCES; | |
421748ec | 1475 | if (path.mnt->mnt_flags & MNT_NODEV) |
1da177e4 LT |
1476 | goto fail; |
1477 | error = -ENOMEM; | |
1478 | bdev = bd_acquire(inode); | |
1479 | if (!bdev) | |
1480 | goto fail; | |
1481 | out: | |
421748ec | 1482 | path_put(&path); |
1da177e4 LT |
1483 | return bdev; |
1484 | fail: | |
1485 | bdev = ERR_PTR(error); | |
1486 | goto out; | |
1487 | } | |
d5686b44 | 1488 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 LT |
1489 | |
1490 | /** | |
30c40d2c | 1491 | * open_bdev_exclusive - open a block device by name and set it up for use |
1da177e4 LT |
1492 | * |
1493 | * @path: special file representing the block device | |
30c40d2c | 1494 | * @mode: FMODE_... combination to pass be used |
1da177e4 LT |
1495 | * @holder: owner for exclusion |
1496 | * | |
1497 | * Open the blockdevice described by the special file at @path, claim it | |
1498 | * for the @holder. | |
1499 | */ | |
30c40d2c | 1500 | struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder) |
1da177e4 LT |
1501 | { |
1502 | struct block_device *bdev; | |
1da177e4 LT |
1503 | int error = 0; |
1504 | ||
1505 | bdev = lookup_bdev(path); | |
1506 | if (IS_ERR(bdev)) | |
1507 | return bdev; | |
1508 | ||
572c4892 | 1509 | error = blkdev_get(bdev, mode); |
1da177e4 LT |
1510 | if (error) |
1511 | return ERR_PTR(error); | |
1512 | error = -EACCES; | |
30c40d2c | 1513 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) |
1da177e4 LT |
1514 | goto blkdev_put; |
1515 | error = bd_claim(bdev, holder); | |
1516 | if (error) | |
1517 | goto blkdev_put; | |
1518 | ||
1519 | return bdev; | |
1520 | ||
1521 | blkdev_put: | |
9a1c3542 | 1522 | blkdev_put(bdev, mode); |
1da177e4 LT |
1523 | return ERR_PTR(error); |
1524 | } | |
1525 | ||
30c40d2c | 1526 | EXPORT_SYMBOL(open_bdev_exclusive); |
1da177e4 LT |
1527 | |
1528 | /** | |
30c40d2c | 1529 | * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive() |
1da177e4 LT |
1530 | * |
1531 | * @bdev: blockdevice to close | |
30c40d2c | 1532 | * @mode: mode, must match that used to open. |
1da177e4 | 1533 | * |
30c40d2c | 1534 | * This is the counterpart to open_bdev_exclusive(). |
1da177e4 | 1535 | */ |
30c40d2c | 1536 | void close_bdev_exclusive(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
1537 | { |
1538 | bd_release(bdev); | |
30c40d2c | 1539 | blkdev_put(bdev, mode); |
1da177e4 LT |
1540 | } |
1541 | ||
30c40d2c | 1542 | EXPORT_SYMBOL(close_bdev_exclusive); |
b71e8a4c DH |
1543 | |
1544 | int __invalidate_device(struct block_device *bdev) | |
1545 | { | |
1546 | struct super_block *sb = get_super(bdev); | |
1547 | int res = 0; | |
1548 | ||
1549 | if (sb) { | |
1550 | /* | |
1551 | * no need to lock the super, get_super holds the | |
1552 | * read mutex so the filesystem cannot go away | |
1553 | * under us (->put_super runs with the write lock | |
1554 | * hold). | |
1555 | */ | |
1556 | shrink_dcache_sb(sb); | |
1557 | res = invalidate_inodes(sb); | |
1558 | drop_super(sb); | |
1559 | } | |
f98393a6 | 1560 | invalidate_bdev(bdev); |
b71e8a4c DH |
1561 | return res; |
1562 | } | |
1563 | EXPORT_SYMBOL(__invalidate_device); |