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 LT |
14 | #include <linux/smp_lock.h> |
15 | #include <linux/highmem.h> | |
16 | #include <linux/blkdev.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/blkpg.h> | |
19 | #include <linux/buffer_head.h> | |
20 | #include <linux/mpage.h> | |
21 | #include <linux/mount.h> | |
22 | #include <linux/uio.h> | |
23 | #include <linux/namei.h> | |
24 | #include <asm/uaccess.h> | |
07f3f05c | 25 | #include "internal.h" |
1da177e4 LT |
26 | |
27 | struct bdev_inode { | |
28 | struct block_device bdev; | |
29 | struct inode vfs_inode; | |
30 | }; | |
31 | ||
32 | static inline struct bdev_inode *BDEV_I(struct inode *inode) | |
33 | { | |
34 | return container_of(inode, struct bdev_inode, vfs_inode); | |
35 | } | |
36 | ||
37 | inline struct block_device *I_BDEV(struct inode *inode) | |
38 | { | |
39 | return &BDEV_I(inode)->bdev; | |
40 | } | |
41 | ||
42 | EXPORT_SYMBOL(I_BDEV); | |
43 | ||
44 | static sector_t max_block(struct block_device *bdev) | |
45 | { | |
46 | sector_t retval = ~((sector_t)0); | |
47 | loff_t sz = i_size_read(bdev->bd_inode); | |
48 | ||
49 | if (sz) { | |
50 | unsigned int size = block_size(bdev); | |
51 | unsigned int sizebits = blksize_bits(size); | |
52 | retval = (sz >> sizebits); | |
53 | } | |
54 | return retval; | |
55 | } | |
56 | ||
57 | /* Kill _all_ buffers, dirty or not.. */ | |
58 | static void kill_bdev(struct block_device *bdev) | |
59 | { | |
60 | invalidate_bdev(bdev, 1); | |
61 | truncate_inode_pages(bdev->bd_inode->i_mapping, 0); | |
62 | } | |
63 | ||
64 | int set_blocksize(struct block_device *bdev, int size) | |
65 | { | |
66 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
67 | if (size > PAGE_SIZE || size < 512 || (size & (size-1))) | |
68 | return -EINVAL; | |
69 | ||
70 | /* Size cannot be smaller than the size supported by the device */ | |
71 | if (size < bdev_hardsect_size(bdev)) | |
72 | return -EINVAL; | |
73 | ||
74 | /* Don't change the size if it is same as current */ | |
75 | if (bdev->bd_block_size != size) { | |
76 | sync_blockdev(bdev); | |
77 | bdev->bd_block_size = size; | |
78 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
79 | kill_bdev(bdev); | |
80 | } | |
81 | return 0; | |
82 | } | |
83 | ||
84 | EXPORT_SYMBOL(set_blocksize); | |
85 | ||
86 | int sb_set_blocksize(struct super_block *sb, int size) | |
87 | { | |
1da177e4 LT |
88 | if (set_blocksize(sb->s_bdev, size)) |
89 | return 0; | |
90 | /* If we get here, we know size is power of two | |
91 | * and it's value is between 512 and PAGE_SIZE */ | |
92 | sb->s_blocksize = size; | |
38885bd4 | 93 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
94 | return sb->s_blocksize; |
95 | } | |
96 | ||
97 | EXPORT_SYMBOL(sb_set_blocksize); | |
98 | ||
99 | int sb_min_blocksize(struct super_block *sb, int size) | |
100 | { | |
101 | int minsize = bdev_hardsect_size(sb->s_bdev); | |
102 | if (size < minsize) | |
103 | size = minsize; | |
104 | return sb_set_blocksize(sb, size); | |
105 | } | |
106 | ||
107 | EXPORT_SYMBOL(sb_min_blocksize); | |
108 | ||
109 | static int | |
110 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
111 | struct buffer_head *bh, int create) | |
112 | { | |
113 | if (iblock >= max_block(I_BDEV(inode))) { | |
114 | if (create) | |
115 | return -EIO; | |
116 | ||
117 | /* | |
118 | * for reads, we're just trying to fill a partial page. | |
119 | * return a hole, they will have to call get_block again | |
120 | * before they can fill it, and they will get -EIO at that | |
121 | * time | |
122 | */ | |
123 | return 0; | |
124 | } | |
125 | bh->b_bdev = I_BDEV(inode); | |
126 | bh->b_blocknr = iblock; | |
127 | set_buffer_mapped(bh); | |
128 | return 0; | |
129 | } | |
130 | ||
131 | static int | |
132 | blkdev_get_blocks(struct inode *inode, sector_t iblock, | |
1d8fa7a2 | 133 | struct buffer_head *bh, int create) |
1da177e4 LT |
134 | { |
135 | sector_t end_block = max_block(I_BDEV(inode)); | |
1d8fa7a2 | 136 | unsigned long max_blocks = bh->b_size >> inode->i_blkbits; |
1da177e4 LT |
137 | |
138 | if ((iblock + max_blocks) > end_block) { | |
139 | max_blocks = end_block - iblock; | |
140 | if ((long)max_blocks <= 0) { | |
141 | if (create) | |
142 | return -EIO; /* write fully beyond EOF */ | |
143 | /* | |
144 | * It is a read which is fully beyond EOF. We return | |
145 | * a !buffer_mapped buffer | |
146 | */ | |
147 | max_blocks = 0; | |
148 | } | |
149 | } | |
150 | ||
151 | bh->b_bdev = I_BDEV(inode); | |
152 | bh->b_blocknr = iblock; | |
153 | bh->b_size = max_blocks << inode->i_blkbits; | |
154 | if (max_blocks) | |
155 | set_buffer_mapped(bh); | |
156 | return 0; | |
157 | } | |
158 | ||
159 | static ssize_t | |
160 | blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, | |
161 | loff_t offset, unsigned long nr_segs) | |
162 | { | |
163 | struct file *file = iocb->ki_filp; | |
164 | struct inode *inode = file->f_mapping->host; | |
165 | ||
166 | return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode), | |
167 | iov, offset, nr_segs, blkdev_get_blocks, NULL); | |
168 | } | |
169 | ||
170 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) | |
171 | { | |
172 | return block_write_full_page(page, blkdev_get_block, wbc); | |
173 | } | |
174 | ||
175 | static int blkdev_readpage(struct file * file, struct page * page) | |
176 | { | |
177 | return block_read_full_page(page, blkdev_get_block); | |
178 | } | |
179 | ||
180 | static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to) | |
181 | { | |
182 | return block_prepare_write(page, from, to, blkdev_get_block); | |
183 | } | |
184 | ||
185 | static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to) | |
186 | { | |
187 | return block_commit_write(page, from, to); | |
188 | } | |
189 | ||
190 | /* | |
191 | * private llseek: | |
192 | * for a block special file file->f_dentry->d_inode->i_size is zero | |
193 | * so we compute the size by hand (just as in block_read/write above) | |
194 | */ | |
195 | static loff_t block_llseek(struct file *file, loff_t offset, int origin) | |
196 | { | |
197 | struct inode *bd_inode = file->f_mapping->host; | |
198 | loff_t size; | |
199 | loff_t retval; | |
200 | ||
1b1dcc1b | 201 | mutex_lock(&bd_inode->i_mutex); |
1da177e4 LT |
202 | size = i_size_read(bd_inode); |
203 | ||
204 | switch (origin) { | |
205 | case 2: | |
206 | offset += size; | |
207 | break; | |
208 | case 1: | |
209 | offset += file->f_pos; | |
210 | } | |
211 | retval = -EINVAL; | |
212 | if (offset >= 0 && offset <= size) { | |
213 | if (offset != file->f_pos) { | |
214 | file->f_pos = offset; | |
215 | } | |
216 | retval = offset; | |
217 | } | |
1b1dcc1b | 218 | mutex_unlock(&bd_inode->i_mutex); |
1da177e4 LT |
219 | return retval; |
220 | } | |
221 | ||
222 | /* | |
223 | * Filp is never NULL; the only case when ->fsync() is called with | |
224 | * NULL first argument is nfsd_sync_dir() and that's not a directory. | |
225 | */ | |
226 | ||
227 | static int block_fsync(struct file *filp, struct dentry *dentry, int datasync) | |
228 | { | |
229 | return sync_blockdev(I_BDEV(filp->f_mapping->host)); | |
230 | } | |
231 | ||
232 | /* | |
233 | * pseudo-fs | |
234 | */ | |
235 | ||
236 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
fa3536cc | 237 | static kmem_cache_t * bdev_cachep __read_mostly; |
1da177e4 LT |
238 | |
239 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
240 | { | |
241 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL); | |
242 | if (!ei) | |
243 | return NULL; | |
244 | return &ei->vfs_inode; | |
245 | } | |
246 | ||
247 | static void bdev_destroy_inode(struct inode *inode) | |
248 | { | |
249 | struct bdev_inode *bdi = BDEV_I(inode); | |
250 | ||
251 | bdi->bdev.bd_inode_backing_dev_info = NULL; | |
252 | kmem_cache_free(bdev_cachep, bdi); | |
253 | } | |
254 | ||
255 | static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags) | |
256 | { | |
257 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
258 | struct block_device *bdev = &ei->bdev; | |
259 | ||
260 | if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == | |
261 | SLAB_CTOR_CONSTRUCTOR) | |
262 | { | |
263 | memset(bdev, 0, sizeof(*bdev)); | |
c039e313 AV |
264 | mutex_init(&bdev->bd_mutex); |
265 | mutex_init(&bdev->bd_mount_mutex); | |
1da177e4 LT |
266 | INIT_LIST_HEAD(&bdev->bd_inodes); |
267 | INIT_LIST_HEAD(&bdev->bd_list); | |
641dc636 JN |
268 | #ifdef CONFIG_SYSFS |
269 | INIT_LIST_HEAD(&bdev->bd_holder_list); | |
270 | #endif | |
1da177e4 LT |
271 | inode_init_once(&ei->vfs_inode); |
272 | } | |
273 | } | |
274 | ||
275 | static inline void __bd_forget(struct inode *inode) | |
276 | { | |
277 | list_del_init(&inode->i_devices); | |
278 | inode->i_bdev = NULL; | |
279 | inode->i_mapping = &inode->i_data; | |
280 | } | |
281 | ||
282 | static void bdev_clear_inode(struct inode *inode) | |
283 | { | |
284 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
285 | struct list_head *p; | |
286 | spin_lock(&bdev_lock); | |
287 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
288 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
289 | } | |
290 | list_del_init(&bdev->bd_list); | |
291 | spin_unlock(&bdev_lock); | |
292 | } | |
293 | ||
294 | static struct super_operations bdev_sops = { | |
295 | .statfs = simple_statfs, | |
296 | .alloc_inode = bdev_alloc_inode, | |
297 | .destroy_inode = bdev_destroy_inode, | |
298 | .drop_inode = generic_delete_inode, | |
299 | .clear_inode = bdev_clear_inode, | |
300 | }; | |
301 | ||
454e2398 DH |
302 | static int bd_get_sb(struct file_system_type *fs_type, |
303 | int flags, const char *dev_name, void *data, struct vfsmount *mnt) | |
1da177e4 | 304 | { |
454e2398 | 305 | return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt); |
1da177e4 LT |
306 | } |
307 | ||
308 | static struct file_system_type bd_type = { | |
309 | .name = "bdev", | |
310 | .get_sb = bd_get_sb, | |
311 | .kill_sb = kill_anon_super, | |
312 | }; | |
313 | ||
fa3536cc | 314 | static struct vfsmount *bd_mnt __read_mostly; |
1da177e4 LT |
315 | struct super_block *blockdev_superblock; |
316 | ||
317 | void __init bdev_cache_init(void) | |
318 | { | |
319 | int err; | |
320 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), | |
fffb60f9 PJ |
321 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
322 | SLAB_MEM_SPREAD|SLAB_PANIC), | |
1da177e4 LT |
323 | init_once, NULL); |
324 | err = register_filesystem(&bd_type); | |
325 | if (err) | |
326 | panic("Cannot register bdev pseudo-fs"); | |
327 | bd_mnt = kern_mount(&bd_type); | |
328 | err = PTR_ERR(bd_mnt); | |
329 | if (IS_ERR(bd_mnt)) | |
330 | panic("Cannot create bdev pseudo-fs"); | |
331 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ | |
332 | } | |
333 | ||
334 | /* | |
335 | * Most likely _very_ bad one - but then it's hardly critical for small | |
336 | * /dev and can be fixed when somebody will need really large one. | |
337 | * Keep in mind that it will be fed through icache hash function too. | |
338 | */ | |
339 | static inline unsigned long hash(dev_t dev) | |
340 | { | |
341 | return MAJOR(dev)+MINOR(dev); | |
342 | } | |
343 | ||
344 | static int bdev_test(struct inode *inode, void *data) | |
345 | { | |
346 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
347 | } | |
348 | ||
349 | static int bdev_set(struct inode *inode, void *data) | |
350 | { | |
351 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
352 | return 0; | |
353 | } | |
354 | ||
355 | static LIST_HEAD(all_bdevs); | |
356 | ||
357 | struct block_device *bdget(dev_t dev) | |
358 | { | |
359 | struct block_device *bdev; | |
360 | struct inode *inode; | |
361 | ||
362 | inode = iget5_locked(bd_mnt->mnt_sb, hash(dev), | |
363 | bdev_test, bdev_set, &dev); | |
364 | ||
365 | if (!inode) | |
366 | return NULL; | |
367 | ||
368 | bdev = &BDEV_I(inode)->bdev; | |
369 | ||
370 | if (inode->i_state & I_NEW) { | |
371 | bdev->bd_contains = NULL; | |
372 | bdev->bd_inode = inode; | |
373 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
374 | bdev->bd_part_count = 0; | |
375 | bdev->bd_invalidated = 0; | |
376 | inode->i_mode = S_IFBLK; | |
377 | inode->i_rdev = dev; | |
378 | inode->i_bdev = bdev; | |
379 | inode->i_data.a_ops = &def_blk_aops; | |
380 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
381 | inode->i_data.backing_dev_info = &default_backing_dev_info; | |
382 | spin_lock(&bdev_lock); | |
383 | list_add(&bdev->bd_list, &all_bdevs); | |
384 | spin_unlock(&bdev_lock); | |
385 | unlock_new_inode(inode); | |
386 | } | |
387 | return bdev; | |
388 | } | |
389 | ||
390 | EXPORT_SYMBOL(bdget); | |
391 | ||
392 | long nr_blockdev_pages(void) | |
393 | { | |
394 | struct list_head *p; | |
395 | long ret = 0; | |
396 | spin_lock(&bdev_lock); | |
397 | list_for_each(p, &all_bdevs) { | |
398 | struct block_device *bdev; | |
399 | bdev = list_entry(p, struct block_device, bd_list); | |
400 | ret += bdev->bd_inode->i_mapping->nrpages; | |
401 | } | |
402 | spin_unlock(&bdev_lock); | |
403 | return ret; | |
404 | } | |
405 | ||
406 | void bdput(struct block_device *bdev) | |
407 | { | |
408 | iput(bdev->bd_inode); | |
409 | } | |
410 | ||
411 | EXPORT_SYMBOL(bdput); | |
412 | ||
413 | static struct block_device *bd_acquire(struct inode *inode) | |
414 | { | |
415 | struct block_device *bdev; | |
09d967c6 | 416 | |
1da177e4 LT |
417 | spin_lock(&bdev_lock); |
418 | bdev = inode->i_bdev; | |
09d967c6 OH |
419 | if (bdev) { |
420 | atomic_inc(&bdev->bd_inode->i_count); | |
1da177e4 LT |
421 | spin_unlock(&bdev_lock); |
422 | return bdev; | |
423 | } | |
424 | spin_unlock(&bdev_lock); | |
09d967c6 | 425 | |
1da177e4 LT |
426 | bdev = bdget(inode->i_rdev); |
427 | if (bdev) { | |
428 | spin_lock(&bdev_lock); | |
09d967c6 OH |
429 | if (!inode->i_bdev) { |
430 | /* | |
431 | * We take an additional bd_inode->i_count for inode, | |
432 | * and it's released in clear_inode() of inode. | |
433 | * So, we can access it via ->i_mapping always | |
434 | * without igrab(). | |
435 | */ | |
436 | atomic_inc(&bdev->bd_inode->i_count); | |
437 | inode->i_bdev = bdev; | |
438 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
439 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
440 | } | |
1da177e4 LT |
441 | spin_unlock(&bdev_lock); |
442 | } | |
443 | return bdev; | |
444 | } | |
445 | ||
446 | /* Call when you free inode */ | |
447 | ||
448 | void bd_forget(struct inode *inode) | |
449 | { | |
09d967c6 OH |
450 | struct block_device *bdev = NULL; |
451 | ||
1da177e4 | 452 | spin_lock(&bdev_lock); |
09d967c6 OH |
453 | if (inode->i_bdev) { |
454 | if (inode->i_sb != blockdev_superblock) | |
455 | bdev = inode->i_bdev; | |
1da177e4 | 456 | __bd_forget(inode); |
09d967c6 | 457 | } |
1da177e4 | 458 | spin_unlock(&bdev_lock); |
09d967c6 OH |
459 | |
460 | if (bdev) | |
461 | iput(bdev->bd_inode); | |
1da177e4 LT |
462 | } |
463 | ||
464 | int bd_claim(struct block_device *bdev, void *holder) | |
465 | { | |
466 | int res; | |
467 | spin_lock(&bdev_lock); | |
468 | ||
469 | /* first decide result */ | |
470 | if (bdev->bd_holder == holder) | |
471 | res = 0; /* already a holder */ | |
472 | else if (bdev->bd_holder != NULL) | |
473 | res = -EBUSY; /* held by someone else */ | |
474 | else if (bdev->bd_contains == bdev) | |
475 | res = 0; /* is a whole device which isn't held */ | |
476 | ||
477 | else if (bdev->bd_contains->bd_holder == bd_claim) | |
478 | res = 0; /* is a partition of a device that is being partitioned */ | |
479 | else if (bdev->bd_contains->bd_holder != NULL) | |
480 | res = -EBUSY; /* is a partition of a held device */ | |
481 | else | |
482 | res = 0; /* is a partition of an un-held device */ | |
483 | ||
484 | /* now impose change */ | |
485 | if (res==0) { | |
486 | /* note that for a whole device bd_holders | |
487 | * will be incremented twice, and bd_holder will | |
488 | * be set to bd_claim before being set to holder | |
489 | */ | |
490 | bdev->bd_contains->bd_holders ++; | |
491 | bdev->bd_contains->bd_holder = bd_claim; | |
492 | bdev->bd_holders++; | |
493 | bdev->bd_holder = holder; | |
494 | } | |
495 | spin_unlock(&bdev_lock); | |
496 | return res; | |
497 | } | |
498 | ||
499 | EXPORT_SYMBOL(bd_claim); | |
500 | ||
501 | void bd_release(struct block_device *bdev) | |
502 | { | |
503 | spin_lock(&bdev_lock); | |
504 | if (!--bdev->bd_contains->bd_holders) | |
505 | bdev->bd_contains->bd_holder = NULL; | |
506 | if (!--bdev->bd_holders) | |
507 | bdev->bd_holder = NULL; | |
508 | spin_unlock(&bdev_lock); | |
509 | } | |
510 | ||
511 | EXPORT_SYMBOL(bd_release); | |
512 | ||
641dc636 JN |
513 | #ifdef CONFIG_SYSFS |
514 | /* | |
515 | * Functions for bd_claim_by_kobject / bd_release_from_kobject | |
516 | * | |
517 | * If a kobject is passed to bd_claim_by_kobject() | |
518 | * and the kobject has a parent directory, | |
519 | * following symlinks are created: | |
520 | * o from the kobject to the claimed bdev | |
521 | * o from "holders" directory of the bdev to the parent of the kobject | |
522 | * bd_release_from_kobject() removes these symlinks. | |
523 | * | |
524 | * Example: | |
525 | * If /dev/dm-0 maps to /dev/sda, kobject corresponding to | |
526 | * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then: | |
527 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda | |
528 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
529 | */ | |
530 | ||
531 | static struct kobject *bdev_get_kobj(struct block_device *bdev) | |
532 | { | |
533 | if (bdev->bd_contains != bdev) | |
534 | return kobject_get(&bdev->bd_part->kobj); | |
535 | else | |
536 | return kobject_get(&bdev->bd_disk->kobj); | |
537 | } | |
538 | ||
539 | static struct kobject *bdev_get_holder(struct block_device *bdev) | |
540 | { | |
541 | if (bdev->bd_contains != bdev) | |
542 | return kobject_get(bdev->bd_part->holder_dir); | |
543 | else | |
544 | return kobject_get(bdev->bd_disk->holder_dir); | |
545 | } | |
546 | ||
4d7dd8fd | 547 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 JN |
548 | { |
549 | if (!from || !to) | |
4d7dd8fd AM |
550 | return 0; |
551 | return sysfs_create_link(from, to, kobject_name(to)); | |
641dc636 JN |
552 | } |
553 | ||
554 | static void del_symlink(struct kobject *from, struct kobject *to) | |
555 | { | |
556 | if (!from || !to) | |
557 | return; | |
558 | sysfs_remove_link(from, kobject_name(to)); | |
559 | } | |
560 | ||
561 | /* | |
562 | * 'struct bd_holder' contains pointers to kobjects symlinked by | |
563 | * bd_claim_by_kobject. | |
564 | * It's connected to bd_holder_list which is protected by bdev->bd_sem. | |
565 | */ | |
566 | struct bd_holder { | |
567 | struct list_head list; /* chain of holders of the bdev */ | |
568 | int count; /* references from the holder */ | |
569 | struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */ | |
570 | struct kobject *hdev; /* e.g. "/block/dm-0" */ | |
571 | struct kobject *hdir; /* e.g. "/block/sda/holders" */ | |
572 | struct kobject *sdev; /* e.g. "/block/sda" */ | |
573 | }; | |
574 | ||
575 | /* | |
576 | * Get references of related kobjects at once. | |
577 | * Returns 1 on success. 0 on failure. | |
578 | * | |
579 | * Should call bd_holder_release_dirs() after successful use. | |
580 | */ | |
581 | static int bd_holder_grab_dirs(struct block_device *bdev, | |
582 | struct bd_holder *bo) | |
583 | { | |
584 | if (!bdev || !bo) | |
585 | return 0; | |
586 | ||
587 | bo->sdir = kobject_get(bo->sdir); | |
588 | if (!bo->sdir) | |
589 | return 0; | |
590 | ||
591 | bo->hdev = kobject_get(bo->sdir->parent); | |
592 | if (!bo->hdev) | |
593 | goto fail_put_sdir; | |
594 | ||
595 | bo->sdev = bdev_get_kobj(bdev); | |
596 | if (!bo->sdev) | |
597 | goto fail_put_hdev; | |
598 | ||
599 | bo->hdir = bdev_get_holder(bdev); | |
600 | if (!bo->hdir) | |
601 | goto fail_put_sdev; | |
602 | ||
603 | return 1; | |
604 | ||
605 | fail_put_sdev: | |
606 | kobject_put(bo->sdev); | |
607 | fail_put_hdev: | |
608 | kobject_put(bo->hdev); | |
609 | fail_put_sdir: | |
610 | kobject_put(bo->sdir); | |
611 | ||
612 | return 0; | |
613 | } | |
614 | ||
615 | /* Put references of related kobjects at once. */ | |
616 | static void bd_holder_release_dirs(struct bd_holder *bo) | |
617 | { | |
618 | kobject_put(bo->hdir); | |
619 | kobject_put(bo->sdev); | |
620 | kobject_put(bo->hdev); | |
621 | kobject_put(bo->sdir); | |
622 | } | |
623 | ||
624 | static struct bd_holder *alloc_bd_holder(struct kobject *kobj) | |
625 | { | |
626 | struct bd_holder *bo; | |
627 | ||
628 | bo = kzalloc(sizeof(*bo), GFP_KERNEL); | |
629 | if (!bo) | |
630 | return NULL; | |
631 | ||
632 | bo->count = 1; | |
633 | bo->sdir = kobj; | |
634 | ||
635 | return bo; | |
636 | } | |
637 | ||
638 | static void free_bd_holder(struct bd_holder *bo) | |
639 | { | |
640 | kfree(bo); | |
641 | } | |
642 | ||
643 | /** | |
644 | * add_bd_holder - create sysfs symlinks for bd_claim() relationship | |
645 | * | |
646 | * @bdev: block device to be bd_claimed | |
647 | * @bo: preallocated and initialized by alloc_bd_holder() | |
648 | * | |
649 | * If there is no matching entry with @bo in @bdev->bd_holder_list, | |
650 | * add @bo to the list, create symlinks. | |
651 | * | |
4d7dd8fd AM |
652 | * Returns 0 if symlinks are created or already there. |
653 | * Returns -ve if something fails and @bo can be freed. | |
641dc636 JN |
654 | */ |
655 | static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo) | |
656 | { | |
657 | struct bd_holder *tmp; | |
4d7dd8fd | 658 | int ret; |
641dc636 JN |
659 | |
660 | if (!bo) | |
4d7dd8fd | 661 | return -EINVAL; |
641dc636 JN |
662 | |
663 | list_for_each_entry(tmp, &bdev->bd_holder_list, list) { | |
664 | if (tmp->sdir == bo->sdir) { | |
665 | tmp->count++; | |
4d7dd8fd AM |
666 | /* We've already done what we need to do here. */ |
667 | free_bd_holder(bo); | |
641dc636 JN |
668 | return 0; |
669 | } | |
670 | } | |
671 | ||
672 | if (!bd_holder_grab_dirs(bdev, bo)) | |
4d7dd8fd | 673 | return -EBUSY; |
641dc636 | 674 | |
4d7dd8fd AM |
675 | ret = add_symlink(bo->sdir, bo->sdev); |
676 | if (ret == 0) { | |
677 | ret = add_symlink(bo->hdir, bo->hdev); | |
678 | if (ret) | |
679 | del_symlink(bo->sdir, bo->sdev); | |
680 | } | |
681 | if (ret == 0) | |
682 | list_add_tail(&bo->list, &bdev->bd_holder_list); | |
683 | return ret; | |
641dc636 JN |
684 | } |
685 | ||
686 | /** | |
687 | * del_bd_holder - delete sysfs symlinks for bd_claim() relationship | |
688 | * | |
689 | * @bdev: block device to be bd_claimed | |
690 | * @kobj: holder's kobject | |
691 | * | |
692 | * If there is matching entry with @kobj in @bdev->bd_holder_list | |
693 | * and no other bd_claim() from the same kobject, | |
694 | * remove the struct bd_holder from the list, delete symlinks for it. | |
695 | * | |
696 | * Returns a pointer to the struct bd_holder when it's removed from the list | |
697 | * and ready to be freed. | |
698 | * Returns NULL if matching claim isn't found or there is other bd_claim() | |
699 | * by the same kobject. | |
700 | */ | |
701 | static struct bd_holder *del_bd_holder(struct block_device *bdev, | |
702 | struct kobject *kobj) | |
703 | { | |
704 | struct bd_holder *bo; | |
705 | ||
706 | list_for_each_entry(bo, &bdev->bd_holder_list, list) { | |
707 | if (bo->sdir == kobj) { | |
708 | bo->count--; | |
709 | BUG_ON(bo->count < 0); | |
710 | if (!bo->count) { | |
711 | list_del(&bo->list); | |
712 | del_symlink(bo->sdir, bo->sdev); | |
713 | del_symlink(bo->hdir, bo->hdev); | |
714 | bd_holder_release_dirs(bo); | |
715 | return bo; | |
716 | } | |
717 | break; | |
718 | } | |
719 | } | |
720 | ||
721 | return NULL; | |
722 | } | |
723 | ||
724 | /** | |
725 | * bd_claim_by_kobject - bd_claim() with additional kobject signature | |
726 | * | |
727 | * @bdev: block device to be claimed | |
728 | * @holder: holder's signature | |
729 | * @kobj: holder's kobject | |
730 | * | |
731 | * Do bd_claim() and if it succeeds, create sysfs symlinks between | |
732 | * the bdev and the holder's kobject. | |
733 | * Use bd_release_from_kobject() when relesing the claimed bdev. | |
734 | * | |
735 | * Returns 0 on success. (same as bd_claim()) | |
736 | * Returns errno on failure. | |
737 | */ | |
738 | static int bd_claim_by_kobject(struct block_device *bdev, void *holder, | |
739 | struct kobject *kobj) | |
740 | { | |
741 | int res; | |
742 | struct bd_holder *bo; | |
743 | ||
744 | if (!kobj) | |
745 | return -EINVAL; | |
746 | ||
747 | bo = alloc_bd_holder(kobj); | |
748 | if (!bo) | |
749 | return -ENOMEM; | |
750 | ||
663d440e | 751 | mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION); |
641dc636 | 752 | res = bd_claim(bdev, holder); |
4d7dd8fd AM |
753 | if (res == 0) |
754 | res = add_bd_holder(bdev, bo); | |
755 | if (res) | |
641dc636 | 756 | free_bd_holder(bo); |
b4cf1b72 | 757 | mutex_unlock(&bdev->bd_mutex); |
641dc636 JN |
758 | |
759 | return res; | |
760 | } | |
761 | ||
762 | /** | |
763 | * bd_release_from_kobject - bd_release() with additional kobject signature | |
764 | * | |
765 | * @bdev: block device to be released | |
766 | * @kobj: holder's kobject | |
767 | * | |
768 | * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject(). | |
769 | */ | |
770 | static void bd_release_from_kobject(struct block_device *bdev, | |
771 | struct kobject *kobj) | |
772 | { | |
773 | struct bd_holder *bo; | |
774 | ||
775 | if (!kobj) | |
776 | return; | |
777 | ||
663d440e | 778 | mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION); |
641dc636 JN |
779 | bd_release(bdev); |
780 | if ((bo = del_bd_holder(bdev, kobj))) | |
781 | free_bd_holder(bo); | |
b4cf1b72 | 782 | mutex_unlock(&bdev->bd_mutex); |
641dc636 JN |
783 | } |
784 | ||
785 | /** | |
786 | * bd_claim_by_disk - wrapper function for bd_claim_by_kobject() | |
787 | * | |
788 | * @bdev: block device to be claimed | |
789 | * @holder: holder's signature | |
790 | * @disk: holder's gendisk | |
791 | * | |
792 | * Call bd_claim_by_kobject() with getting @disk->slave_dir. | |
793 | */ | |
794 | int bd_claim_by_disk(struct block_device *bdev, void *holder, | |
795 | struct gendisk *disk) | |
796 | { | |
797 | return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir)); | |
798 | } | |
799 | EXPORT_SYMBOL_GPL(bd_claim_by_disk); | |
800 | ||
801 | /** | |
802 | * bd_release_from_disk - wrapper function for bd_release_from_kobject() | |
803 | * | |
804 | * @bdev: block device to be claimed | |
805 | * @disk: holder's gendisk | |
806 | * | |
807 | * Call bd_release_from_kobject() and put @disk->slave_dir. | |
808 | */ | |
809 | void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk) | |
810 | { | |
811 | bd_release_from_kobject(bdev, disk->slave_dir); | |
812 | kobject_put(disk->slave_dir); | |
813 | } | |
814 | EXPORT_SYMBOL_GPL(bd_release_from_disk); | |
815 | #endif | |
816 | ||
1da177e4 LT |
817 | /* |
818 | * Tries to open block device by device number. Use it ONLY if you | |
819 | * really do not have anything better - i.e. when you are behind a | |
820 | * truly sucky interface and all you are given is a device number. _Never_ | |
821 | * to be used for internal purposes. If you ever need it - reconsider | |
822 | * your API. | |
823 | */ | |
824 | struct block_device *open_by_devnum(dev_t dev, unsigned mode) | |
825 | { | |
826 | struct block_device *bdev = bdget(dev); | |
827 | int err = -ENOMEM; | |
828 | int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY; | |
829 | if (bdev) | |
830 | err = blkdev_get(bdev, mode, flags); | |
831 | return err ? ERR_PTR(err) : bdev; | |
832 | } | |
833 | ||
834 | EXPORT_SYMBOL(open_by_devnum); | |
835 | ||
663d440e IM |
836 | static int |
837 | blkdev_get_partition(struct block_device *bdev, mode_t mode, unsigned flags); | |
838 | ||
839 | struct block_device *open_partition_by_devnum(dev_t dev, unsigned mode) | |
840 | { | |
841 | struct block_device *bdev = bdget(dev); | |
842 | int err = -ENOMEM; | |
843 | int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY; | |
844 | if (bdev) | |
845 | err = blkdev_get_partition(bdev, mode, flags); | |
846 | return err ? ERR_PTR(err) : bdev; | |
847 | } | |
848 | ||
849 | EXPORT_SYMBOL(open_partition_by_devnum); | |
850 | ||
851 | ||
1da177e4 LT |
852 | /* |
853 | * This routine checks whether a removable media has been changed, | |
854 | * and invalidates all buffer-cache-entries in that case. This | |
855 | * is a relatively slow routine, so we have to try to minimize using | |
856 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
857 | * is the best way of combining speed and utility, I think. | |
858 | * People changing diskettes in the middle of an operation deserve | |
859 | * to lose :-) | |
860 | */ | |
861 | int check_disk_change(struct block_device *bdev) | |
862 | { | |
863 | struct gendisk *disk = bdev->bd_disk; | |
864 | struct block_device_operations * bdops = disk->fops; | |
865 | ||
866 | if (!bdops->media_changed) | |
867 | return 0; | |
868 | if (!bdops->media_changed(bdev->bd_disk)) | |
869 | return 0; | |
870 | ||
2ef41634 | 871 | if (__invalidate_device(bdev)) |
1da177e4 LT |
872 | printk("VFS: busy inodes on changed media.\n"); |
873 | ||
874 | if (bdops->revalidate_disk) | |
875 | bdops->revalidate_disk(bdev->bd_disk); | |
876 | if (bdev->bd_disk->minors > 1) | |
877 | bdev->bd_invalidated = 1; | |
878 | return 1; | |
879 | } | |
880 | ||
881 | EXPORT_SYMBOL(check_disk_change); | |
882 | ||
883 | void bd_set_size(struct block_device *bdev, loff_t size) | |
884 | { | |
885 | unsigned bsize = bdev_hardsect_size(bdev); | |
886 | ||
887 | bdev->bd_inode->i_size = size; | |
888 | while (bsize < PAGE_CACHE_SIZE) { | |
889 | if (size & bsize) | |
890 | break; | |
891 | bsize <<= 1; | |
892 | } | |
893 | bdev->bd_block_size = bsize; | |
894 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
895 | } | |
896 | EXPORT_SYMBOL(bd_set_size); | |
897 | ||
6946bd63 PZ |
898 | static int __blkdev_put(struct block_device *bdev, unsigned int subclass) |
899 | { | |
900 | int ret = 0; | |
901 | struct inode *bd_inode = bdev->bd_inode; | |
902 | struct gendisk *disk = bdev->bd_disk; | |
903 | ||
904 | mutex_lock_nested(&bdev->bd_mutex, subclass); | |
905 | lock_kernel(); | |
906 | if (!--bdev->bd_openers) { | |
907 | sync_blockdev(bdev); | |
908 | kill_bdev(bdev); | |
909 | } | |
910 | if (bdev->bd_contains == bdev) { | |
911 | if (disk->fops->release) | |
912 | ret = disk->fops->release(bd_inode, NULL); | |
913 | } else { | |
914 | mutex_lock_nested(&bdev->bd_contains->bd_mutex, | |
915 | subclass + 1); | |
916 | bdev->bd_contains->bd_part_count--; | |
917 | mutex_unlock(&bdev->bd_contains->bd_mutex); | |
918 | } | |
919 | if (!bdev->bd_openers) { | |
920 | struct module *owner = disk->fops->owner; | |
921 | ||
922 | put_disk(disk); | |
923 | module_put(owner); | |
924 | ||
925 | if (bdev->bd_contains != bdev) { | |
926 | kobject_put(&bdev->bd_part->kobj); | |
927 | bdev->bd_part = NULL; | |
928 | } | |
929 | bdev->bd_disk = NULL; | |
930 | bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info; | |
931 | if (bdev != bdev->bd_contains) | |
932 | __blkdev_put(bdev->bd_contains, subclass + 1); | |
933 | bdev->bd_contains = NULL; | |
934 | } | |
935 | unlock_kernel(); | |
936 | mutex_unlock(&bdev->bd_mutex); | |
937 | bdput(bdev); | |
938 | return ret; | |
939 | } | |
940 | ||
941 | int blkdev_put(struct block_device *bdev) | |
942 | { | |
943 | return __blkdev_put(bdev, BD_MUTEX_NORMAL); | |
944 | } | |
945 | EXPORT_SYMBOL(blkdev_put); | |
946 | ||
947 | int blkdev_put_partition(struct block_device *bdev) | |
948 | { | |
949 | return __blkdev_put(bdev, BD_MUTEX_PARTITION); | |
950 | } | |
951 | EXPORT_SYMBOL(blkdev_put_partition); | |
952 | ||
663d440e IM |
953 | static int |
954 | blkdev_get_whole(struct block_device *bdev, mode_t mode, unsigned flags); | |
955 | ||
956 | static int | |
957 | do_open(struct block_device *bdev, struct file *file, unsigned int subclass) | |
1da177e4 LT |
958 | { |
959 | struct module *owner = NULL; | |
960 | struct gendisk *disk; | |
961 | int ret = -ENXIO; | |
962 | int part; | |
963 | ||
964 | file->f_mapping = bdev->bd_inode->i_mapping; | |
965 | lock_kernel(); | |
966 | disk = get_gendisk(bdev->bd_dev, &part); | |
967 | if (!disk) { | |
968 | unlock_kernel(); | |
969 | bdput(bdev); | |
970 | return ret; | |
971 | } | |
972 | owner = disk->fops->owner; | |
973 | ||
663d440e IM |
974 | mutex_lock_nested(&bdev->bd_mutex, subclass); |
975 | ||
1da177e4 LT |
976 | if (!bdev->bd_openers) { |
977 | bdev->bd_disk = disk; | |
978 | bdev->bd_contains = bdev; | |
979 | if (!part) { | |
980 | struct backing_dev_info *bdi; | |
981 | if (disk->fops->open) { | |
982 | ret = disk->fops->open(bdev->bd_inode, file); | |
983 | if (ret) | |
984 | goto out_first; | |
985 | } | |
986 | if (!bdev->bd_openers) { | |
987 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); | |
988 | bdi = blk_get_backing_dev_info(bdev); | |
989 | if (bdi == NULL) | |
990 | bdi = &default_backing_dev_info; | |
991 | bdev->bd_inode->i_data.backing_dev_info = bdi; | |
992 | } | |
993 | if (bdev->bd_invalidated) | |
994 | rescan_partitions(disk, bdev); | |
995 | } else { | |
996 | struct hd_struct *p; | |
997 | struct block_device *whole; | |
998 | whole = bdget_disk(disk, 0); | |
999 | ret = -ENOMEM; | |
1000 | if (!whole) | |
1001 | goto out_first; | |
663d440e | 1002 | ret = blkdev_get_whole(whole, file->f_mode, file->f_flags); |
1da177e4 LT |
1003 | if (ret) |
1004 | goto out_first; | |
1005 | bdev->bd_contains = whole; | |
663d440e | 1006 | mutex_lock_nested(&whole->bd_mutex, BD_MUTEX_WHOLE); |
1da177e4 LT |
1007 | whole->bd_part_count++; |
1008 | p = disk->part[part - 1]; | |
1009 | bdev->bd_inode->i_data.backing_dev_info = | |
1010 | whole->bd_inode->i_data.backing_dev_info; | |
1011 | if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) { | |
1012 | whole->bd_part_count--; | |
c039e313 | 1013 | mutex_unlock(&whole->bd_mutex); |
1da177e4 LT |
1014 | ret = -ENXIO; |
1015 | goto out_first; | |
1016 | } | |
1017 | kobject_get(&p->kobj); | |
1018 | bdev->bd_part = p; | |
1019 | bd_set_size(bdev, (loff_t) p->nr_sects << 9); | |
c039e313 | 1020 | mutex_unlock(&whole->bd_mutex); |
1da177e4 LT |
1021 | } |
1022 | } else { | |
1023 | put_disk(disk); | |
1024 | module_put(owner); | |
1025 | if (bdev->bd_contains == bdev) { | |
1026 | if (bdev->bd_disk->fops->open) { | |
1027 | ret = bdev->bd_disk->fops->open(bdev->bd_inode, file); | |
1028 | if (ret) | |
1029 | goto out; | |
1030 | } | |
1031 | if (bdev->bd_invalidated) | |
1032 | rescan_partitions(bdev->bd_disk, bdev); | |
1033 | } else { | |
663d440e | 1034 | mutex_lock_nested(&bdev->bd_contains->bd_mutex, |
87d7c8ac | 1035 | BD_MUTEX_WHOLE); |
1da177e4 | 1036 | bdev->bd_contains->bd_part_count++; |
c039e313 | 1037 | mutex_unlock(&bdev->bd_contains->bd_mutex); |
1da177e4 LT |
1038 | } |
1039 | } | |
1040 | bdev->bd_openers++; | |
c039e313 | 1041 | mutex_unlock(&bdev->bd_mutex); |
1da177e4 LT |
1042 | unlock_kernel(); |
1043 | return 0; | |
1044 | ||
1045 | out_first: | |
1046 | bdev->bd_disk = NULL; | |
1047 | bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info; | |
1048 | if (bdev != bdev->bd_contains) | |
6946bd63 | 1049 | __blkdev_put(bdev->bd_contains, BD_MUTEX_WHOLE); |
1da177e4 LT |
1050 | bdev->bd_contains = NULL; |
1051 | put_disk(disk); | |
1052 | module_put(owner); | |
1053 | out: | |
c039e313 | 1054 | mutex_unlock(&bdev->bd_mutex); |
1da177e4 LT |
1055 | unlock_kernel(); |
1056 | if (ret) | |
1057 | bdput(bdev); | |
1058 | return ret; | |
1059 | } | |
1060 | ||
1061 | int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags) | |
1062 | { | |
1063 | /* | |
1064 | * This crockload is due to bad choice of ->open() type. | |
1065 | * It will go away. | |
1066 | * For now, block device ->open() routine must _not_ | |
1067 | * examine anything in 'inode' argument except ->i_rdev. | |
1068 | */ | |
1069 | struct file fake_file = {}; | |
1070 | struct dentry fake_dentry = {}; | |
1071 | fake_file.f_mode = mode; | |
1072 | fake_file.f_flags = flags; | |
1073 | fake_file.f_dentry = &fake_dentry; | |
1074 | fake_dentry.d_inode = bdev->bd_inode; | |
1075 | ||
663d440e | 1076 | return do_open(bdev, &fake_file, BD_MUTEX_NORMAL); |
1da177e4 LT |
1077 | } |
1078 | ||
1079 | EXPORT_SYMBOL(blkdev_get); | |
1080 | ||
663d440e IM |
1081 | static int |
1082 | blkdev_get_whole(struct block_device *bdev, mode_t mode, unsigned flags) | |
1083 | { | |
1084 | /* | |
1085 | * This crockload is due to bad choice of ->open() type. | |
1086 | * It will go away. | |
1087 | * For now, block device ->open() routine must _not_ | |
1088 | * examine anything in 'inode' argument except ->i_rdev. | |
1089 | */ | |
1090 | struct file fake_file = {}; | |
1091 | struct dentry fake_dentry = {}; | |
1092 | fake_file.f_mode = mode; | |
1093 | fake_file.f_flags = flags; | |
1094 | fake_file.f_dentry = &fake_dentry; | |
1095 | fake_dentry.d_inode = bdev->bd_inode; | |
1096 | ||
1097 | return do_open(bdev, &fake_file, BD_MUTEX_WHOLE); | |
1098 | } | |
1099 | ||
1100 | static int | |
1101 | blkdev_get_partition(struct block_device *bdev, mode_t mode, unsigned flags) | |
1102 | { | |
1103 | /* | |
1104 | * This crockload is due to bad choice of ->open() type. | |
1105 | * It will go away. | |
1106 | * For now, block device ->open() routine must _not_ | |
1107 | * examine anything in 'inode' argument except ->i_rdev. | |
1108 | */ | |
1109 | struct file fake_file = {}; | |
1110 | struct dentry fake_dentry = {}; | |
1111 | fake_file.f_mode = mode; | |
1112 | fake_file.f_flags = flags; | |
1113 | fake_file.f_dentry = &fake_dentry; | |
1114 | fake_dentry.d_inode = bdev->bd_inode; | |
1115 | ||
1116 | return do_open(bdev, &fake_file, BD_MUTEX_PARTITION); | |
1117 | } | |
1118 | ||
1da177e4 LT |
1119 | static int blkdev_open(struct inode * inode, struct file * filp) |
1120 | { | |
1121 | struct block_device *bdev; | |
1122 | int res; | |
1123 | ||
1124 | /* | |
1125 | * Preserve backwards compatibility and allow large file access | |
1126 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1127 | * binary needs it. We might want to drop this workaround | |
1128 | * during an unstable branch. | |
1129 | */ | |
1130 | filp->f_flags |= O_LARGEFILE; | |
1131 | ||
1132 | bdev = bd_acquire(inode); | |
1133 | ||
663d440e | 1134 | res = do_open(bdev, filp, BD_MUTEX_NORMAL); |
1da177e4 LT |
1135 | if (res) |
1136 | return res; | |
1137 | ||
1138 | if (!(filp->f_flags & O_EXCL) ) | |
1139 | return 0; | |
1140 | ||
1141 | if (!(res = bd_claim(bdev, filp))) | |
1142 | return 0; | |
1143 | ||
1144 | blkdev_put(bdev); | |
1145 | return res; | |
1146 | } | |
1147 | ||
1da177e4 LT |
1148 | static int blkdev_close(struct inode * inode, struct file * filp) |
1149 | { | |
1150 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); | |
1151 | if (bdev->bd_holder == filp) | |
1152 | bd_release(bdev); | |
1153 | return blkdev_put(bdev); | |
1154 | } | |
1155 | ||
1156 | static ssize_t blkdev_file_write(struct file *file, const char __user *buf, | |
1157 | size_t count, loff_t *ppos) | |
1158 | { | |
1159 | struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count }; | |
1160 | ||
1161 | return generic_file_write_nolock(file, &local_iov, 1, ppos); | |
1162 | } | |
1163 | ||
1164 | static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf, | |
1165 | size_t count, loff_t pos) | |
1166 | { | |
1167 | struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count }; | |
1168 | ||
1169 | return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos); | |
1170 | } | |
1171 | ||
bb93e3a5 | 1172 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 LT |
1173 | { |
1174 | return blkdev_ioctl(file->f_mapping->host, file, cmd, arg); | |
1175 | } | |
1176 | ||
f5e54d6e | 1177 | const struct address_space_operations def_blk_aops = { |
1da177e4 LT |
1178 | .readpage = blkdev_readpage, |
1179 | .writepage = blkdev_writepage, | |
1180 | .sync_page = block_sync_page, | |
1181 | .prepare_write = blkdev_prepare_write, | |
1182 | .commit_write = blkdev_commit_write, | |
1183 | .writepages = generic_writepages, | |
1184 | .direct_IO = blkdev_direct_IO, | |
1185 | }; | |
1186 | ||
4b6f5d20 | 1187 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1188 | .open = blkdev_open, |
1189 | .release = blkdev_close, | |
1190 | .llseek = block_llseek, | |
1191 | .read = generic_file_read, | |
1192 | .write = blkdev_file_write, | |
1193 | .aio_read = generic_file_aio_read, | |
1194 | .aio_write = blkdev_file_aio_write, | |
1195 | .mmap = generic_file_mmap, | |
1196 | .fsync = block_fsync, | |
bb93e3a5 | 1197 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1198 | #ifdef CONFIG_COMPAT |
1199 | .compat_ioctl = compat_blkdev_ioctl, | |
1200 | #endif | |
1201 | .readv = generic_file_readv, | |
1202 | .writev = generic_file_write_nolock, | |
1203 | .sendfile = generic_file_sendfile, | |
7f9c51f0 JA |
1204 | .splice_read = generic_file_splice_read, |
1205 | .splice_write = generic_file_splice_write, | |
1da177e4 LT |
1206 | }; |
1207 | ||
1208 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1209 | { | |
1210 | int res; | |
1211 | mm_segment_t old_fs = get_fs(); | |
1212 | set_fs(KERNEL_DS); | |
1213 | res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg); | |
1214 | set_fs(old_fs); | |
1215 | return res; | |
1216 | } | |
1217 | ||
1218 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1219 | ||
1220 | /** | |
1221 | * lookup_bdev - lookup a struct block_device by name | |
1222 | * | |
1223 | * @path: special file representing the block device | |
1224 | * | |
1225 | * Get a reference to the blockdevice at @path in the current | |
1226 | * namespace if possible and return it. Return ERR_PTR(error) | |
1227 | * otherwise. | |
1228 | */ | |
1229 | struct block_device *lookup_bdev(const char *path) | |
1230 | { | |
1231 | struct block_device *bdev; | |
1232 | struct inode *inode; | |
1233 | struct nameidata nd; | |
1234 | int error; | |
1235 | ||
1236 | if (!path || !*path) | |
1237 | return ERR_PTR(-EINVAL); | |
1238 | ||
1239 | error = path_lookup(path, LOOKUP_FOLLOW, &nd); | |
1240 | if (error) | |
1241 | return ERR_PTR(error); | |
1242 | ||
1243 | inode = nd.dentry->d_inode; | |
1244 | error = -ENOTBLK; | |
1245 | if (!S_ISBLK(inode->i_mode)) | |
1246 | goto fail; | |
1247 | error = -EACCES; | |
1248 | if (nd.mnt->mnt_flags & MNT_NODEV) | |
1249 | goto fail; | |
1250 | error = -ENOMEM; | |
1251 | bdev = bd_acquire(inode); | |
1252 | if (!bdev) | |
1253 | goto fail; | |
1254 | out: | |
1255 | path_release(&nd); | |
1256 | return bdev; | |
1257 | fail: | |
1258 | bdev = ERR_PTR(error); | |
1259 | goto out; | |
1260 | } | |
1261 | ||
1262 | /** | |
1263 | * open_bdev_excl - open a block device by name and set it up for use | |
1264 | * | |
1265 | * @path: special file representing the block device | |
1266 | * @flags: %MS_RDONLY for opening read-only | |
1267 | * @holder: owner for exclusion | |
1268 | * | |
1269 | * Open the blockdevice described by the special file at @path, claim it | |
1270 | * for the @holder. | |
1271 | */ | |
1272 | struct block_device *open_bdev_excl(const char *path, int flags, void *holder) | |
1273 | { | |
1274 | struct block_device *bdev; | |
1275 | mode_t mode = FMODE_READ; | |
1276 | int error = 0; | |
1277 | ||
1278 | bdev = lookup_bdev(path); | |
1279 | if (IS_ERR(bdev)) | |
1280 | return bdev; | |
1281 | ||
1282 | if (!(flags & MS_RDONLY)) | |
1283 | mode |= FMODE_WRITE; | |
1284 | error = blkdev_get(bdev, mode, 0); | |
1285 | if (error) | |
1286 | return ERR_PTR(error); | |
1287 | error = -EACCES; | |
1288 | if (!(flags & MS_RDONLY) && bdev_read_only(bdev)) | |
1289 | goto blkdev_put; | |
1290 | error = bd_claim(bdev, holder); | |
1291 | if (error) | |
1292 | goto blkdev_put; | |
1293 | ||
1294 | return bdev; | |
1295 | ||
1296 | blkdev_put: | |
1297 | blkdev_put(bdev); | |
1298 | return ERR_PTR(error); | |
1299 | } | |
1300 | ||
1301 | EXPORT_SYMBOL(open_bdev_excl); | |
1302 | ||
1303 | /** | |
1304 | * close_bdev_excl - release a blockdevice openen by open_bdev_excl() | |
1305 | * | |
1306 | * @bdev: blockdevice to close | |
1307 | * | |
1308 | * This is the counterpart to open_bdev_excl(). | |
1309 | */ | |
1310 | void close_bdev_excl(struct block_device *bdev) | |
1311 | { | |
1312 | bd_release(bdev); | |
1313 | blkdev_put(bdev); | |
1314 | } | |
1315 | ||
1316 | EXPORT_SYMBOL(close_bdev_excl); |