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