Root/
1 | /* |
2 | * linux/fs/block_dev.c |
3 | * |
4 | * Copyright (C) 1991, 1992 Linus Torvalds |
5 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE |
6 | */ |
7 | |
8 | #include <linux/init.h> |
9 | #include <linux/mm.h> |
10 | #include <linux/fcntl.h> |
11 | #include <linux/slab.h> |
12 | #include <linux/kmod.h> |
13 | #include <linux/major.h> |
14 | #include <linux/smp_lock.h> |
15 | #include <linux/device_cgroup.h> |
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> |
21 | #include <linux/pagevec.h> |
22 | #include <linux/writeback.h> |
23 | #include <linux/mpage.h> |
24 | #include <linux/mount.h> |
25 | #include <linux/uio.h> |
26 | #include <linux/namei.h> |
27 | #include <linux/log2.h> |
28 | #include <linux/kmemleak.h> |
29 | #include <asm/uaccess.h> |
30 | #include "internal.h" |
31 | |
32 | struct bdev_inode { |
33 | struct block_device bdev; |
34 | struct inode vfs_inode; |
35 | }; |
36 | |
37 | static const struct address_space_operations def_blk_aops; |
38 | |
39 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
40 | { |
41 | return container_of(inode, struct bdev_inode, vfs_inode); |
42 | } |
43 | |
44 | inline struct block_device *I_BDEV(struct inode *inode) |
45 | { |
46 | return &BDEV_I(inode)->bdev; |
47 | } |
48 | |
49 | EXPORT_SYMBOL(I_BDEV); |
50 | |
51 | static sector_t max_block(struct block_device *bdev) |
52 | { |
53 | sector_t retval = ~((sector_t)0); |
54 | loff_t sz = i_size_read(bdev->bd_inode); |
55 | |
56 | if (sz) { |
57 | unsigned int size = block_size(bdev); |
58 | unsigned int sizebits = blksize_bits(size); |
59 | retval = (sz >> sizebits); |
60 | } |
61 | return retval; |
62 | } |
63 | |
64 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
65 | static void kill_bdev(struct block_device *bdev) |
66 | { |
67 | if (bdev->bd_inode->i_mapping->nrpages == 0) |
68 | return; |
69 | invalidate_bh_lrus(); |
70 | truncate_inode_pages(bdev->bd_inode->i_mapping, 0); |
71 | } |
72 | |
73 | int set_blocksize(struct block_device *bdev, int size) |
74 | { |
75 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ |
76 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
77 | return -EINVAL; |
78 | |
79 | /* Size cannot be smaller than the size supported by the device */ |
80 | if (size < bdev_logical_block_size(bdev)) |
81 | return -EINVAL; |
82 | |
83 | /* Don't change the size if it is same as current */ |
84 | if (bdev->bd_block_size != size) { |
85 | sync_blockdev(bdev); |
86 | bdev->bd_block_size = size; |
87 | bdev->bd_inode->i_blkbits = blksize_bits(size); |
88 | kill_bdev(bdev); |
89 | } |
90 | return 0; |
91 | } |
92 | |
93 | EXPORT_SYMBOL(set_blocksize); |
94 | |
95 | int sb_set_blocksize(struct super_block *sb, int size) |
96 | { |
97 | if (set_blocksize(sb->s_bdev, size)) |
98 | return 0; |
99 | /* If we get here, we know size is power of two |
100 | * and it's value is between 512 and PAGE_SIZE */ |
101 | sb->s_blocksize = size; |
102 | sb->s_blocksize_bits = blksize_bits(size); |
103 | return sb->s_blocksize; |
104 | } |
105 | |
106 | EXPORT_SYMBOL(sb_set_blocksize); |
107 | |
108 | int sb_min_blocksize(struct super_block *sb, int size) |
109 | { |
110 | int minsize = bdev_logical_block_size(sb->s_bdev); |
111 | if (size < minsize) |
112 | size = minsize; |
113 | return sb_set_blocksize(sb, size); |
114 | } |
115 | |
116 | EXPORT_SYMBOL(sb_min_blocksize); |
117 | |
118 | static int |
119 | blkdev_get_block(struct inode *inode, sector_t iblock, |
120 | struct buffer_head *bh, int create) |
121 | { |
122 | if (iblock >= max_block(I_BDEV(inode))) { |
123 | if (create) |
124 | return -EIO; |
125 | |
126 | /* |
127 | * for reads, we're just trying to fill a partial page. |
128 | * return a hole, they will have to call get_block again |
129 | * before they can fill it, and they will get -EIO at that |
130 | * time |
131 | */ |
132 | return 0; |
133 | } |
134 | bh->b_bdev = I_BDEV(inode); |
135 | bh->b_blocknr = iblock; |
136 | set_buffer_mapped(bh); |
137 | return 0; |
138 | } |
139 | |
140 | static int |
141 | blkdev_get_blocks(struct inode *inode, sector_t iblock, |
142 | struct buffer_head *bh, int create) |
143 | { |
144 | sector_t end_block = max_block(I_BDEV(inode)); |
145 | unsigned long max_blocks = bh->b_size >> inode->i_blkbits; |
146 | |
147 | if ((iblock + max_blocks) > end_block) { |
148 | max_blocks = end_block - iblock; |
149 | if ((long)max_blocks <= 0) { |
150 | if (create) |
151 | return -EIO; /* write fully beyond EOF */ |
152 | /* |
153 | * It is a read which is fully beyond EOF. We return |
154 | * a !buffer_mapped buffer |
155 | */ |
156 | max_blocks = 0; |
157 | } |
158 | } |
159 | |
160 | bh->b_bdev = I_BDEV(inode); |
161 | bh->b_blocknr = iblock; |
162 | bh->b_size = max_blocks << inode->i_blkbits; |
163 | if (max_blocks) |
164 | set_buffer_mapped(bh); |
165 | return 0; |
166 | } |
167 | |
168 | static ssize_t |
169 | blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, |
170 | loff_t offset, unsigned long nr_segs) |
171 | { |
172 | struct file *file = iocb->ki_filp; |
173 | struct inode *inode = file->f_mapping->host; |
174 | |
175 | return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode), |
176 | iov, offset, nr_segs, blkdev_get_blocks, NULL); |
177 | } |
178 | |
179 | int __sync_blockdev(struct block_device *bdev, int wait) |
180 | { |
181 | if (!bdev) |
182 | return 0; |
183 | if (!wait) |
184 | return filemap_flush(bdev->bd_inode->i_mapping); |
185 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); |
186 | } |
187 | |
188 | /* |
189 | * Write out and wait upon all the dirty data associated with a block |
190 | * device via its mapping. Does not take the superblock lock. |
191 | */ |
192 | int sync_blockdev(struct block_device *bdev) |
193 | { |
194 | return __sync_blockdev(bdev, 1); |
195 | } |
196 | EXPORT_SYMBOL(sync_blockdev); |
197 | |
198 | /* |
199 | * Write out and wait upon all dirty data associated with this |
200 | * device. Filesystem data as well as the underlying block |
201 | * device. Takes the superblock lock. |
202 | */ |
203 | int fsync_bdev(struct block_device *bdev) |
204 | { |
205 | struct super_block *sb = get_super(bdev); |
206 | if (sb) { |
207 | int res = sync_filesystem(sb); |
208 | drop_super(sb); |
209 | return res; |
210 | } |
211 | return sync_blockdev(bdev); |
212 | } |
213 | EXPORT_SYMBOL(fsync_bdev); |
214 | |
215 | /** |
216 | * freeze_bdev -- lock a filesystem and force it into a consistent state |
217 | * @bdev: blockdevice to lock |
218 | * |
219 | * If a superblock is found on this device, we take the s_umount semaphore |
220 | * on it to make sure nobody unmounts until the snapshot creation is done. |
221 | * The reference counter (bd_fsfreeze_count) guarantees that only the last |
222 | * unfreeze process can unfreeze the frozen filesystem actually when multiple |
223 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and |
224 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze |
225 | * actually. |
226 | */ |
227 | struct super_block *freeze_bdev(struct block_device *bdev) |
228 | { |
229 | struct super_block *sb; |
230 | int error = 0; |
231 | |
232 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
233 | if (++bdev->bd_fsfreeze_count > 1) { |
234 | /* |
235 | * We don't even need to grab a reference - the first call |
236 | * to freeze_bdev grab an active reference and only the last |
237 | * thaw_bdev drops it. |
238 | */ |
239 | sb = get_super(bdev); |
240 | drop_super(sb); |
241 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
242 | return sb; |
243 | } |
244 | |
245 | sb = get_active_super(bdev); |
246 | if (!sb) |
247 | goto out; |
248 | if (sb->s_flags & MS_RDONLY) { |
249 | sb->s_frozen = SB_FREEZE_TRANS; |
250 | up_write(&sb->s_umount); |
251 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
252 | return sb; |
253 | } |
254 | |
255 | sb->s_frozen = SB_FREEZE_WRITE; |
256 | smp_wmb(); |
257 | |
258 | sync_filesystem(sb); |
259 | |
260 | sb->s_frozen = SB_FREEZE_TRANS; |
261 | smp_wmb(); |
262 | |
263 | sync_blockdev(sb->s_bdev); |
264 | |
265 | if (sb->s_op->freeze_fs) { |
266 | error = sb->s_op->freeze_fs(sb); |
267 | if (error) { |
268 | printk(KERN_ERR |
269 | "VFS:Filesystem freeze failed\n"); |
270 | sb->s_frozen = SB_UNFROZEN; |
271 | deactivate_locked_super(sb); |
272 | bdev->bd_fsfreeze_count--; |
273 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
274 | return ERR_PTR(error); |
275 | } |
276 | } |
277 | up_write(&sb->s_umount); |
278 | |
279 | out: |
280 | sync_blockdev(bdev); |
281 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
282 | return sb; /* thaw_bdev releases s->s_umount */ |
283 | } |
284 | EXPORT_SYMBOL(freeze_bdev); |
285 | |
286 | /** |
287 | * thaw_bdev -- unlock filesystem |
288 | * @bdev: blockdevice to unlock |
289 | * @sb: associated superblock |
290 | * |
291 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). |
292 | */ |
293 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) |
294 | { |
295 | int error = -EINVAL; |
296 | |
297 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
298 | if (!bdev->bd_fsfreeze_count) |
299 | goto out_unlock; |
300 | |
301 | error = 0; |
302 | if (--bdev->bd_fsfreeze_count > 0) |
303 | goto out_unlock; |
304 | |
305 | if (!sb) |
306 | goto out_unlock; |
307 | |
308 | BUG_ON(sb->s_bdev != bdev); |
309 | down_write(&sb->s_umount); |
310 | if (sb->s_flags & MS_RDONLY) |
311 | goto out_unfrozen; |
312 | |
313 | if (sb->s_op->unfreeze_fs) { |
314 | error = sb->s_op->unfreeze_fs(sb); |
315 | if (error) { |
316 | printk(KERN_ERR |
317 | "VFS:Filesystem thaw failed\n"); |
318 | sb->s_frozen = SB_FREEZE_TRANS; |
319 | bdev->bd_fsfreeze_count++; |
320 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
321 | return error; |
322 | } |
323 | } |
324 | |
325 | out_unfrozen: |
326 | sb->s_frozen = SB_UNFROZEN; |
327 | smp_wmb(); |
328 | wake_up(&sb->s_wait_unfrozen); |
329 | |
330 | if (sb) |
331 | deactivate_locked_super(sb); |
332 | out_unlock: |
333 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
334 | return 0; |
335 | } |
336 | EXPORT_SYMBOL(thaw_bdev); |
337 | |
338 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
339 | { |
340 | return block_write_full_page(page, blkdev_get_block, wbc); |
341 | } |
342 | |
343 | static int blkdev_readpage(struct file * file, struct page * page) |
344 | { |
345 | return block_read_full_page(page, blkdev_get_block); |
346 | } |
347 | |
348 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
349 | loff_t pos, unsigned len, unsigned flags, |
350 | struct page **pagep, void **fsdata) |
351 | { |
352 | *pagep = NULL; |
353 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
354 | blkdev_get_block); |
355 | } |
356 | |
357 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
358 | loff_t pos, unsigned len, unsigned copied, |
359 | struct page *page, void *fsdata) |
360 | { |
361 | int ret; |
362 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
363 | |
364 | unlock_page(page); |
365 | page_cache_release(page); |
366 | |
367 | return ret; |
368 | } |
369 | |
370 | /* |
371 | * private llseek: |
372 | * for a block special file file->f_path.dentry->d_inode->i_size is zero |
373 | * so we compute the size by hand (just as in block_read/write above) |
374 | */ |
375 | static loff_t block_llseek(struct file *file, loff_t offset, int origin) |
376 | { |
377 | struct inode *bd_inode = file->f_mapping->host; |
378 | loff_t size; |
379 | loff_t retval; |
380 | |
381 | mutex_lock(&bd_inode->i_mutex); |
382 | size = i_size_read(bd_inode); |
383 | |
384 | switch (origin) { |
385 | case 2: |
386 | offset += size; |
387 | break; |
388 | case 1: |
389 | offset += file->f_pos; |
390 | } |
391 | retval = -EINVAL; |
392 | if (offset >= 0 && offset <= size) { |
393 | if (offset != file->f_pos) { |
394 | file->f_pos = offset; |
395 | } |
396 | retval = offset; |
397 | } |
398 | mutex_unlock(&bd_inode->i_mutex); |
399 | return retval; |
400 | } |
401 | |
402 | /* |
403 | * Filp is never NULL; the only case when ->fsync() is called with |
404 | * NULL first argument is nfsd_sync_dir() and that's not a directory. |
405 | */ |
406 | |
407 | int blkdev_fsync(struct file *filp, struct dentry *dentry, int datasync) |
408 | { |
409 | struct inode *bd_inode = filp->f_mapping->host; |
410 | struct block_device *bdev = I_BDEV(bd_inode); |
411 | int error; |
412 | |
413 | /* |
414 | * There is no need to serialise calls to blkdev_issue_flush with |
415 | * i_mutex and doing so causes performance issues with concurrent |
416 | * O_SYNC writers to a block device. |
417 | */ |
418 | mutex_unlock(&bd_inode->i_mutex); |
419 | |
420 | error = blkdev_issue_flush(bdev, NULL); |
421 | if (error == -EOPNOTSUPP) |
422 | error = 0; |
423 | |
424 | mutex_lock(&bd_inode->i_mutex); |
425 | |
426 | return error; |
427 | } |
428 | EXPORT_SYMBOL(blkdev_fsync); |
429 | |
430 | /* |
431 | * pseudo-fs |
432 | */ |
433 | |
434 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); |
435 | static struct kmem_cache * bdev_cachep __read_mostly; |
436 | |
437 | static struct inode *bdev_alloc_inode(struct super_block *sb) |
438 | { |
439 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
440 | if (!ei) |
441 | return NULL; |
442 | return &ei->vfs_inode; |
443 | } |
444 | |
445 | static void bdev_destroy_inode(struct inode *inode) |
446 | { |
447 | struct bdev_inode *bdi = BDEV_I(inode); |
448 | |
449 | kmem_cache_free(bdev_cachep, bdi); |
450 | } |
451 | |
452 | static void init_once(void *foo) |
453 | { |
454 | struct bdev_inode *ei = (struct bdev_inode *) foo; |
455 | struct block_device *bdev = &ei->bdev; |
456 | |
457 | memset(bdev, 0, sizeof(*bdev)); |
458 | mutex_init(&bdev->bd_mutex); |
459 | INIT_LIST_HEAD(&bdev->bd_inodes); |
460 | INIT_LIST_HEAD(&bdev->bd_list); |
461 | #ifdef CONFIG_SYSFS |
462 | INIT_LIST_HEAD(&bdev->bd_holder_list); |
463 | #endif |
464 | inode_init_once(&ei->vfs_inode); |
465 | /* Initialize mutex for freeze. */ |
466 | mutex_init(&bdev->bd_fsfreeze_mutex); |
467 | } |
468 | |
469 | static inline void __bd_forget(struct inode *inode) |
470 | { |
471 | list_del_init(&inode->i_devices); |
472 | inode->i_bdev = NULL; |
473 | inode->i_mapping = &inode->i_data; |
474 | } |
475 | |
476 | static void bdev_clear_inode(struct inode *inode) |
477 | { |
478 | struct block_device *bdev = &BDEV_I(inode)->bdev; |
479 | struct list_head *p; |
480 | spin_lock(&bdev_lock); |
481 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { |
482 | __bd_forget(list_entry(p, struct inode, i_devices)); |
483 | } |
484 | list_del_init(&bdev->bd_list); |
485 | spin_unlock(&bdev_lock); |
486 | } |
487 | |
488 | static const struct super_operations bdev_sops = { |
489 | .statfs = simple_statfs, |
490 | .alloc_inode = bdev_alloc_inode, |
491 | .destroy_inode = bdev_destroy_inode, |
492 | .drop_inode = generic_delete_inode, |
493 | .clear_inode = bdev_clear_inode, |
494 | }; |
495 | |
496 | static int bd_get_sb(struct file_system_type *fs_type, |
497 | int flags, const char *dev_name, void *data, struct vfsmount *mnt) |
498 | { |
499 | return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt); |
500 | } |
501 | |
502 | static struct file_system_type bd_type = { |
503 | .name = "bdev", |
504 | .get_sb = bd_get_sb, |
505 | .kill_sb = kill_anon_super, |
506 | }; |
507 | |
508 | struct super_block *blockdev_superblock __read_mostly; |
509 | |
510 | void __init bdev_cache_init(void) |
511 | { |
512 | int err; |
513 | struct vfsmount *bd_mnt; |
514 | |
515 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
516 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
517 | SLAB_MEM_SPREAD|SLAB_PANIC), |
518 | init_once); |
519 | err = register_filesystem(&bd_type); |
520 | if (err) |
521 | panic("Cannot register bdev pseudo-fs"); |
522 | bd_mnt = kern_mount(&bd_type); |
523 | if (IS_ERR(bd_mnt)) |
524 | panic("Cannot create bdev pseudo-fs"); |
525 | /* |
526 | * This vfsmount structure is only used to obtain the |
527 | * blockdev_superblock, so tell kmemleak not to report it. |
528 | */ |
529 | kmemleak_not_leak(bd_mnt); |
530 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
531 | } |
532 | |
533 | /* |
534 | * Most likely _very_ bad one - but then it's hardly critical for small |
535 | * /dev and can be fixed when somebody will need really large one. |
536 | * Keep in mind that it will be fed through icache hash function too. |
537 | */ |
538 | static inline unsigned long hash(dev_t dev) |
539 | { |
540 | return MAJOR(dev)+MINOR(dev); |
541 | } |
542 | |
543 | static int bdev_test(struct inode *inode, void *data) |
544 | { |
545 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; |
546 | } |
547 | |
548 | static int bdev_set(struct inode *inode, void *data) |
549 | { |
550 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; |
551 | return 0; |
552 | } |
553 | |
554 | static LIST_HEAD(all_bdevs); |
555 | |
556 | struct block_device *bdget(dev_t dev) |
557 | { |
558 | struct block_device *bdev; |
559 | struct inode *inode; |
560 | |
561 | inode = iget5_locked(blockdev_superblock, hash(dev), |
562 | bdev_test, bdev_set, &dev); |
563 | |
564 | if (!inode) |
565 | return NULL; |
566 | |
567 | bdev = &BDEV_I(inode)->bdev; |
568 | |
569 | if (inode->i_state & I_NEW) { |
570 | bdev->bd_contains = NULL; |
571 | bdev->bd_inode = inode; |
572 | bdev->bd_block_size = (1 << inode->i_blkbits); |
573 | bdev->bd_part_count = 0; |
574 | bdev->bd_invalidated = 0; |
575 | inode->i_mode = S_IFBLK; |
576 | inode->i_rdev = dev; |
577 | inode->i_bdev = bdev; |
578 | inode->i_data.a_ops = &def_blk_aops; |
579 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); |
580 | inode->i_data.backing_dev_info = &default_backing_dev_info; |
581 | spin_lock(&bdev_lock); |
582 | list_add(&bdev->bd_list, &all_bdevs); |
583 | spin_unlock(&bdev_lock); |
584 | unlock_new_inode(inode); |
585 | } |
586 | return bdev; |
587 | } |
588 | |
589 | EXPORT_SYMBOL(bdget); |
590 | |
591 | /** |
592 | * bdgrab -- Grab a reference to an already referenced block device |
593 | * @bdev: Block device to grab a reference to. |
594 | */ |
595 | struct block_device *bdgrab(struct block_device *bdev) |
596 | { |
597 | atomic_inc(&bdev->bd_inode->i_count); |
598 | return bdev; |
599 | } |
600 | |
601 | long nr_blockdev_pages(void) |
602 | { |
603 | struct block_device *bdev; |
604 | long ret = 0; |
605 | spin_lock(&bdev_lock); |
606 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
607 | ret += bdev->bd_inode->i_mapping->nrpages; |
608 | } |
609 | spin_unlock(&bdev_lock); |
610 | return ret; |
611 | } |
612 | |
613 | void bdput(struct block_device *bdev) |
614 | { |
615 | iput(bdev->bd_inode); |
616 | } |
617 | |
618 | EXPORT_SYMBOL(bdput); |
619 | |
620 | static struct block_device *bd_acquire(struct inode *inode) |
621 | { |
622 | struct block_device *bdev; |
623 | |
624 | spin_lock(&bdev_lock); |
625 | bdev = inode->i_bdev; |
626 | if (bdev) { |
627 | atomic_inc(&bdev->bd_inode->i_count); |
628 | spin_unlock(&bdev_lock); |
629 | return bdev; |
630 | } |
631 | spin_unlock(&bdev_lock); |
632 | |
633 | bdev = bdget(inode->i_rdev); |
634 | if (bdev) { |
635 | spin_lock(&bdev_lock); |
636 | if (!inode->i_bdev) { |
637 | /* |
638 | * We take an additional bd_inode->i_count for inode, |
639 | * and it's released in clear_inode() of inode. |
640 | * So, we can access it via ->i_mapping always |
641 | * without igrab(). |
642 | */ |
643 | atomic_inc(&bdev->bd_inode->i_count); |
644 | inode->i_bdev = bdev; |
645 | inode->i_mapping = bdev->bd_inode->i_mapping; |
646 | list_add(&inode->i_devices, &bdev->bd_inodes); |
647 | } |
648 | spin_unlock(&bdev_lock); |
649 | } |
650 | return bdev; |
651 | } |
652 | |
653 | /* Call when you free inode */ |
654 | |
655 | void bd_forget(struct inode *inode) |
656 | { |
657 | struct block_device *bdev = NULL; |
658 | |
659 | spin_lock(&bdev_lock); |
660 | if (inode->i_bdev) { |
661 | if (!sb_is_blkdev_sb(inode->i_sb)) |
662 | bdev = inode->i_bdev; |
663 | __bd_forget(inode); |
664 | } |
665 | spin_unlock(&bdev_lock); |
666 | |
667 | if (bdev) |
668 | iput(bdev->bd_inode); |
669 | } |
670 | |
671 | int bd_claim(struct block_device *bdev, void *holder) |
672 | { |
673 | int res; |
674 | spin_lock(&bdev_lock); |
675 | |
676 | /* first decide result */ |
677 | if (bdev->bd_holder == holder) |
678 | res = 0; /* already a holder */ |
679 | else if (bdev->bd_holder != NULL) |
680 | res = -EBUSY; /* held by someone else */ |
681 | else if (bdev->bd_contains == bdev) |
682 | res = 0; /* is a whole device which isn't held */ |
683 | |
684 | else if (bdev->bd_contains->bd_holder == bd_claim) |
685 | res = 0; /* is a partition of a device that is being partitioned */ |
686 | else if (bdev->bd_contains->bd_holder != NULL) |
687 | res = -EBUSY; /* is a partition of a held device */ |
688 | else |
689 | res = 0; /* is a partition of an un-held device */ |
690 | |
691 | /* now impose change */ |
692 | if (res==0) { |
693 | /* note that for a whole device bd_holders |
694 | * will be incremented twice, and bd_holder will |
695 | * be set to bd_claim before being set to holder |
696 | */ |
697 | bdev->bd_contains->bd_holders ++; |
698 | bdev->bd_contains->bd_holder = bd_claim; |
699 | bdev->bd_holders++; |
700 | bdev->bd_holder = holder; |
701 | } |
702 | spin_unlock(&bdev_lock); |
703 | return res; |
704 | } |
705 | |
706 | EXPORT_SYMBOL(bd_claim); |
707 | |
708 | void bd_release(struct block_device *bdev) |
709 | { |
710 | spin_lock(&bdev_lock); |
711 | if (!--bdev->bd_contains->bd_holders) |
712 | bdev->bd_contains->bd_holder = NULL; |
713 | if (!--bdev->bd_holders) |
714 | bdev->bd_holder = NULL; |
715 | spin_unlock(&bdev_lock); |
716 | } |
717 | |
718 | EXPORT_SYMBOL(bd_release); |
719 | |
720 | #ifdef CONFIG_SYSFS |
721 | /* |
722 | * Functions for bd_claim_by_kobject / bd_release_from_kobject |
723 | * |
724 | * If a kobject is passed to bd_claim_by_kobject() |
725 | * and the kobject has a parent directory, |
726 | * following symlinks are created: |
727 | * o from the kobject to the claimed bdev |
728 | * o from "holders" directory of the bdev to the parent of the kobject |
729 | * bd_release_from_kobject() removes these symlinks. |
730 | * |
731 | * Example: |
732 | * If /dev/dm-0 maps to /dev/sda, kobject corresponding to |
733 | * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then: |
734 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
735 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 |
736 | */ |
737 | |
738 | static int add_symlink(struct kobject *from, struct kobject *to) |
739 | { |
740 | if (!from || !to) |
741 | return 0; |
742 | return sysfs_create_link(from, to, kobject_name(to)); |
743 | } |
744 | |
745 | static void del_symlink(struct kobject *from, struct kobject *to) |
746 | { |
747 | if (!from || !to) |
748 | return; |
749 | sysfs_remove_link(from, kobject_name(to)); |
750 | } |
751 | |
752 | /* |
753 | * 'struct bd_holder' contains pointers to kobjects symlinked by |
754 | * bd_claim_by_kobject. |
755 | * It's connected to bd_holder_list which is protected by bdev->bd_sem. |
756 | */ |
757 | struct bd_holder { |
758 | struct list_head list; /* chain of holders of the bdev */ |
759 | int count; /* references from the holder */ |
760 | struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */ |
761 | struct kobject *hdev; /* e.g. "/block/dm-0" */ |
762 | struct kobject *hdir; /* e.g. "/block/sda/holders" */ |
763 | struct kobject *sdev; /* e.g. "/block/sda" */ |
764 | }; |
765 | |
766 | /* |
767 | * Get references of related kobjects at once. |
768 | * Returns 1 on success. 0 on failure. |
769 | * |
770 | * Should call bd_holder_release_dirs() after successful use. |
771 | */ |
772 | static int bd_holder_grab_dirs(struct block_device *bdev, |
773 | struct bd_holder *bo) |
774 | { |
775 | if (!bdev || !bo) |
776 | return 0; |
777 | |
778 | bo->sdir = kobject_get(bo->sdir); |
779 | if (!bo->sdir) |
780 | return 0; |
781 | |
782 | bo->hdev = kobject_get(bo->sdir->parent); |
783 | if (!bo->hdev) |
784 | goto fail_put_sdir; |
785 | |
786 | bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj); |
787 | if (!bo->sdev) |
788 | goto fail_put_hdev; |
789 | |
790 | bo->hdir = kobject_get(bdev->bd_part->holder_dir); |
791 | if (!bo->hdir) |
792 | goto fail_put_sdev; |
793 | |
794 | return 1; |
795 | |
796 | fail_put_sdev: |
797 | kobject_put(bo->sdev); |
798 | fail_put_hdev: |
799 | kobject_put(bo->hdev); |
800 | fail_put_sdir: |
801 | kobject_put(bo->sdir); |
802 | |
803 | return 0; |
804 | } |
805 | |
806 | /* Put references of related kobjects at once. */ |
807 | static void bd_holder_release_dirs(struct bd_holder *bo) |
808 | { |
809 | kobject_put(bo->hdir); |
810 | kobject_put(bo->sdev); |
811 | kobject_put(bo->hdev); |
812 | kobject_put(bo->sdir); |
813 | } |
814 | |
815 | static struct bd_holder *alloc_bd_holder(struct kobject *kobj) |
816 | { |
817 | struct bd_holder *bo; |
818 | |
819 | bo = kzalloc(sizeof(*bo), GFP_KERNEL); |
820 | if (!bo) |
821 | return NULL; |
822 | |
823 | bo->count = 1; |
824 | bo->sdir = kobj; |
825 | |
826 | return bo; |
827 | } |
828 | |
829 | static void free_bd_holder(struct bd_holder *bo) |
830 | { |
831 | kfree(bo); |
832 | } |
833 | |
834 | /** |
835 | * find_bd_holder - find matching struct bd_holder from the block device |
836 | * |
837 | * @bdev: struct block device to be searched |
838 | * @bo: target struct bd_holder |
839 | * |
840 | * Returns matching entry with @bo in @bdev->bd_holder_list. |
841 | * If found, increment the reference count and return the pointer. |
842 | * If not found, returns NULL. |
843 | */ |
844 | static struct bd_holder *find_bd_holder(struct block_device *bdev, |
845 | struct bd_holder *bo) |
846 | { |
847 | struct bd_holder *tmp; |
848 | |
849 | list_for_each_entry(tmp, &bdev->bd_holder_list, list) |
850 | if (tmp->sdir == bo->sdir) { |
851 | tmp->count++; |
852 | return tmp; |
853 | } |
854 | |
855 | return NULL; |
856 | } |
857 | |
858 | /** |
859 | * add_bd_holder - create sysfs symlinks for bd_claim() relationship |
860 | * |
861 | * @bdev: block device to be bd_claimed |
862 | * @bo: preallocated and initialized by alloc_bd_holder() |
863 | * |
864 | * Add @bo to @bdev->bd_holder_list, create symlinks. |
865 | * |
866 | * Returns 0 if symlinks are created. |
867 | * Returns -ve if something fails. |
868 | */ |
869 | static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo) |
870 | { |
871 | int err; |
872 | |
873 | if (!bo) |
874 | return -EINVAL; |
875 | |
876 | if (!bd_holder_grab_dirs(bdev, bo)) |
877 | return -EBUSY; |
878 | |
879 | err = add_symlink(bo->sdir, bo->sdev); |
880 | if (err) |
881 | return err; |
882 | |
883 | err = add_symlink(bo->hdir, bo->hdev); |
884 | if (err) { |
885 | del_symlink(bo->sdir, bo->sdev); |
886 | return err; |
887 | } |
888 | |
889 | list_add_tail(&bo->list, &bdev->bd_holder_list); |
890 | return 0; |
891 | } |
892 | |
893 | /** |
894 | * del_bd_holder - delete sysfs symlinks for bd_claim() relationship |
895 | * |
896 | * @bdev: block device to be bd_claimed |
897 | * @kobj: holder's kobject |
898 | * |
899 | * If there is matching entry with @kobj in @bdev->bd_holder_list |
900 | * and no other bd_claim() from the same kobject, |
901 | * remove the struct bd_holder from the list, delete symlinks for it. |
902 | * |
903 | * Returns a pointer to the struct bd_holder when it's removed from the list |
904 | * and ready to be freed. |
905 | * Returns NULL if matching claim isn't found or there is other bd_claim() |
906 | * by the same kobject. |
907 | */ |
908 | static struct bd_holder *del_bd_holder(struct block_device *bdev, |
909 | struct kobject *kobj) |
910 | { |
911 | struct bd_holder *bo; |
912 | |
913 | list_for_each_entry(bo, &bdev->bd_holder_list, list) { |
914 | if (bo->sdir == kobj) { |
915 | bo->count--; |
916 | BUG_ON(bo->count < 0); |
917 | if (!bo->count) { |
918 | list_del(&bo->list); |
919 | del_symlink(bo->sdir, bo->sdev); |
920 | del_symlink(bo->hdir, bo->hdev); |
921 | bd_holder_release_dirs(bo); |
922 | return bo; |
923 | } |
924 | break; |
925 | } |
926 | } |
927 | |
928 | return NULL; |
929 | } |
930 | |
931 | /** |
932 | * bd_claim_by_kobject - bd_claim() with additional kobject signature |
933 | * |
934 | * @bdev: block device to be claimed |
935 | * @holder: holder's signature |
936 | * @kobj: holder's kobject |
937 | * |
938 | * Do bd_claim() and if it succeeds, create sysfs symlinks between |
939 | * the bdev and the holder's kobject. |
940 | * Use bd_release_from_kobject() when relesing the claimed bdev. |
941 | * |
942 | * Returns 0 on success. (same as bd_claim()) |
943 | * Returns errno on failure. |
944 | */ |
945 | static int bd_claim_by_kobject(struct block_device *bdev, void *holder, |
946 | struct kobject *kobj) |
947 | { |
948 | int err; |
949 | struct bd_holder *bo, *found; |
950 | |
951 | if (!kobj) |
952 | return -EINVAL; |
953 | |
954 | bo = alloc_bd_holder(kobj); |
955 | if (!bo) |
956 | return -ENOMEM; |
957 | |
958 | mutex_lock(&bdev->bd_mutex); |
959 | |
960 | err = bd_claim(bdev, holder); |
961 | if (err) |
962 | goto fail; |
963 | |
964 | found = find_bd_holder(bdev, bo); |
965 | if (found) |
966 | goto fail; |
967 | |
968 | err = add_bd_holder(bdev, bo); |
969 | if (err) |
970 | bd_release(bdev); |
971 | else |
972 | bo = NULL; |
973 | fail: |
974 | mutex_unlock(&bdev->bd_mutex); |
975 | free_bd_holder(bo); |
976 | return err; |
977 | } |
978 | |
979 | /** |
980 | * bd_release_from_kobject - bd_release() with additional kobject signature |
981 | * |
982 | * @bdev: block device to be released |
983 | * @kobj: holder's kobject |
984 | * |
985 | * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject(). |
986 | */ |
987 | static void bd_release_from_kobject(struct block_device *bdev, |
988 | struct kobject *kobj) |
989 | { |
990 | if (!kobj) |
991 | return; |
992 | |
993 | mutex_lock(&bdev->bd_mutex); |
994 | bd_release(bdev); |
995 | free_bd_holder(del_bd_holder(bdev, kobj)); |
996 | mutex_unlock(&bdev->bd_mutex); |
997 | } |
998 | |
999 | /** |
1000 | * bd_claim_by_disk - wrapper function for bd_claim_by_kobject() |
1001 | * |
1002 | * @bdev: block device to be claimed |
1003 | * @holder: holder's signature |
1004 | * @disk: holder's gendisk |
1005 | * |
1006 | * Call bd_claim_by_kobject() with getting @disk->slave_dir. |
1007 | */ |
1008 | int bd_claim_by_disk(struct block_device *bdev, void *holder, |
1009 | struct gendisk *disk) |
1010 | { |
1011 | return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir)); |
1012 | } |
1013 | EXPORT_SYMBOL_GPL(bd_claim_by_disk); |
1014 | |
1015 | /** |
1016 | * bd_release_from_disk - wrapper function for bd_release_from_kobject() |
1017 | * |
1018 | * @bdev: block device to be claimed |
1019 | * @disk: holder's gendisk |
1020 | * |
1021 | * Call bd_release_from_kobject() and put @disk->slave_dir. |
1022 | */ |
1023 | void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk) |
1024 | { |
1025 | bd_release_from_kobject(bdev, disk->slave_dir); |
1026 | kobject_put(disk->slave_dir); |
1027 | } |
1028 | EXPORT_SYMBOL_GPL(bd_release_from_disk); |
1029 | #endif |
1030 | |
1031 | /* |
1032 | * Tries to open block device by device number. Use it ONLY if you |
1033 | * really do not have anything better - i.e. when you are behind a |
1034 | * truly sucky interface and all you are given is a device number. _Never_ |
1035 | * to be used for internal purposes. If you ever need it - reconsider |
1036 | * your API. |
1037 | */ |
1038 | struct block_device *open_by_devnum(dev_t dev, fmode_t mode) |
1039 | { |
1040 | struct block_device *bdev = bdget(dev); |
1041 | int err = -ENOMEM; |
1042 | if (bdev) |
1043 | err = blkdev_get(bdev, mode); |
1044 | return err ? ERR_PTR(err) : bdev; |
1045 | } |
1046 | |
1047 | EXPORT_SYMBOL(open_by_devnum); |
1048 | |
1049 | /** |
1050 | * flush_disk - invalidates all buffer-cache entries on a disk |
1051 | * |
1052 | * @bdev: struct block device to be flushed |
1053 | * |
1054 | * Invalidates all buffer-cache entries on a disk. It should be called |
1055 | * when a disk has been changed -- either by a media change or online |
1056 | * resize. |
1057 | */ |
1058 | static void flush_disk(struct block_device *bdev) |
1059 | { |
1060 | if (__invalidate_device(bdev)) { |
1061 | char name[BDEVNAME_SIZE] = ""; |
1062 | |
1063 | if (bdev->bd_disk) |
1064 | disk_name(bdev->bd_disk, 0, name); |
1065 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
1066 | "resized disk %s\n", name); |
1067 | } |
1068 | |
1069 | if (!bdev->bd_disk) |
1070 | return; |
1071 | if (disk_partitionable(bdev->bd_disk)) |
1072 | bdev->bd_invalidated = 1; |
1073 | } |
1074 | |
1075 | /** |
1076 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
1077 | * @disk: struct gendisk to check |
1078 | * @bdev: struct bdev to adjust. |
1079 | * |
1080 | * This routine checks to see if the bdev size does not match the disk size |
1081 | * and adjusts it if it differs. |
1082 | */ |
1083 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) |
1084 | { |
1085 | loff_t disk_size, bdev_size; |
1086 | |
1087 | disk_size = (loff_t)get_capacity(disk) << 9; |
1088 | bdev_size = i_size_read(bdev->bd_inode); |
1089 | if (disk_size != bdev_size) { |
1090 | char name[BDEVNAME_SIZE]; |
1091 | |
1092 | disk_name(disk, 0, name); |
1093 | printk(KERN_INFO |
1094 | "%s: detected capacity change from %lld to %lld\n", |
1095 | name, bdev_size, disk_size); |
1096 | i_size_write(bdev->bd_inode, disk_size); |
1097 | flush_disk(bdev); |
1098 | } |
1099 | } |
1100 | EXPORT_SYMBOL(check_disk_size_change); |
1101 | |
1102 | /** |
1103 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
1104 | * @disk: struct gendisk to be revalidated |
1105 | * |
1106 | * This routine is a wrapper for lower-level driver's revalidate_disk |
1107 | * call-backs. It is used to do common pre and post operations needed |
1108 | * for all revalidate_disk operations. |
1109 | */ |
1110 | int revalidate_disk(struct gendisk *disk) |
1111 | { |
1112 | struct block_device *bdev; |
1113 | int ret = 0; |
1114 | |
1115 | if (disk->fops->revalidate_disk) |
1116 | ret = disk->fops->revalidate_disk(disk); |
1117 | |
1118 | bdev = bdget_disk(disk, 0); |
1119 | if (!bdev) |
1120 | return ret; |
1121 | |
1122 | mutex_lock(&bdev->bd_mutex); |
1123 | check_disk_size_change(disk, bdev); |
1124 | mutex_unlock(&bdev->bd_mutex); |
1125 | bdput(bdev); |
1126 | return ret; |
1127 | } |
1128 | EXPORT_SYMBOL(revalidate_disk); |
1129 | |
1130 | /* |
1131 | * This routine checks whether a removable media has been changed, |
1132 | * and invalidates all buffer-cache-entries in that case. This |
1133 | * is a relatively slow routine, so we have to try to minimize using |
1134 | * it. Thus it is called only upon a 'mount' or 'open'. This |
1135 | * is the best way of combining speed and utility, I think. |
1136 | * People changing diskettes in the middle of an operation deserve |
1137 | * to lose :-) |
1138 | */ |
1139 | int check_disk_change(struct block_device *bdev) |
1140 | { |
1141 | struct gendisk *disk = bdev->bd_disk; |
1142 | const struct block_device_operations *bdops = disk->fops; |
1143 | |
1144 | if (!bdops->media_changed) |
1145 | return 0; |
1146 | if (!bdops->media_changed(bdev->bd_disk)) |
1147 | return 0; |
1148 | |
1149 | flush_disk(bdev); |
1150 | if (bdops->revalidate_disk) |
1151 | bdops->revalidate_disk(bdev->bd_disk); |
1152 | return 1; |
1153 | } |
1154 | |
1155 | EXPORT_SYMBOL(check_disk_change); |
1156 | |
1157 | void bd_set_size(struct block_device *bdev, loff_t size) |
1158 | { |
1159 | unsigned bsize = bdev_logical_block_size(bdev); |
1160 | |
1161 | bdev->bd_inode->i_size = size; |
1162 | while (bsize < PAGE_CACHE_SIZE) { |
1163 | if (size & bsize) |
1164 | break; |
1165 | bsize <<= 1; |
1166 | } |
1167 | bdev->bd_block_size = bsize; |
1168 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); |
1169 | } |
1170 | EXPORT_SYMBOL(bd_set_size); |
1171 | |
1172 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
1173 | |
1174 | /* |
1175 | * bd_mutex locking: |
1176 | * |
1177 | * mutex_lock(part->bd_mutex) |
1178 | * mutex_lock_nested(whole->bd_mutex, 1) |
1179 | */ |
1180 | |
1181 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1182 | { |
1183 | struct gendisk *disk; |
1184 | int ret; |
1185 | int partno; |
1186 | int perm = 0; |
1187 | |
1188 | if (mode & FMODE_READ) |
1189 | perm |= MAY_READ; |
1190 | if (mode & FMODE_WRITE) |
1191 | perm |= MAY_WRITE; |
1192 | /* |
1193 | * hooks: /n/, see "layering violations". |
1194 | */ |
1195 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); |
1196 | if (ret != 0) { |
1197 | bdput(bdev); |
1198 | return ret; |
1199 | } |
1200 | |
1201 | lock_kernel(); |
1202 | restart: |
1203 | |
1204 | ret = -ENXIO; |
1205 | disk = get_gendisk(bdev->bd_dev, &partno); |
1206 | if (!disk) |
1207 | goto out_unlock_kernel; |
1208 | |
1209 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1210 | if (!bdev->bd_openers) { |
1211 | bdev->bd_disk = disk; |
1212 | bdev->bd_contains = bdev; |
1213 | if (!partno) { |
1214 | struct backing_dev_info *bdi; |
1215 | |
1216 | ret = -ENXIO; |
1217 | bdev->bd_part = disk_get_part(disk, partno); |
1218 | if (!bdev->bd_part) |
1219 | goto out_clear; |
1220 | |
1221 | if (disk->fops->open) { |
1222 | ret = disk->fops->open(bdev, mode); |
1223 | if (ret == -ERESTARTSYS) { |
1224 | /* Lost a race with 'disk' being |
1225 | * deleted, try again. |
1226 | * See md.c |
1227 | */ |
1228 | disk_put_part(bdev->bd_part); |
1229 | bdev->bd_part = NULL; |
1230 | module_put(disk->fops->owner); |
1231 | put_disk(disk); |
1232 | bdev->bd_disk = NULL; |
1233 | mutex_unlock(&bdev->bd_mutex); |
1234 | goto restart; |
1235 | } |
1236 | if (ret) |
1237 | goto out_clear; |
1238 | } |
1239 | if (!bdev->bd_openers) { |
1240 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
1241 | bdi = blk_get_backing_dev_info(bdev); |
1242 | if (bdi == NULL) |
1243 | bdi = &default_backing_dev_info; |
1244 | bdev->bd_inode->i_data.backing_dev_info = bdi; |
1245 | } |
1246 | if (bdev->bd_invalidated) |
1247 | rescan_partitions(disk, bdev); |
1248 | } else { |
1249 | struct block_device *whole; |
1250 | whole = bdget_disk(disk, 0); |
1251 | ret = -ENOMEM; |
1252 | if (!whole) |
1253 | goto out_clear; |
1254 | BUG_ON(for_part); |
1255 | ret = __blkdev_get(whole, mode, 1); |
1256 | if (ret) |
1257 | goto out_clear; |
1258 | bdev->bd_contains = whole; |
1259 | bdev->bd_inode->i_data.backing_dev_info = |
1260 | whole->bd_inode->i_data.backing_dev_info; |
1261 | bdev->bd_part = disk_get_part(disk, partno); |
1262 | if (!(disk->flags & GENHD_FL_UP) || |
1263 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1264 | ret = -ENXIO; |
1265 | goto out_clear; |
1266 | } |
1267 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1268 | } |
1269 | } else { |
1270 | module_put(disk->fops->owner); |
1271 | put_disk(disk); |
1272 | disk = NULL; |
1273 | if (bdev->bd_contains == bdev) { |
1274 | if (bdev->bd_disk->fops->open) { |
1275 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1276 | if (ret) |
1277 | goto out_unlock_bdev; |
1278 | } |
1279 | if (bdev->bd_invalidated) |
1280 | rescan_partitions(bdev->bd_disk, bdev); |
1281 | } |
1282 | } |
1283 | bdev->bd_openers++; |
1284 | if (for_part) |
1285 | bdev->bd_part_count++; |
1286 | mutex_unlock(&bdev->bd_mutex); |
1287 | unlock_kernel(); |
1288 | return 0; |
1289 | |
1290 | out_clear: |
1291 | disk_put_part(bdev->bd_part); |
1292 | bdev->bd_disk = NULL; |
1293 | bdev->bd_part = NULL; |
1294 | bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info; |
1295 | if (bdev != bdev->bd_contains) |
1296 | __blkdev_put(bdev->bd_contains, mode, 1); |
1297 | bdev->bd_contains = NULL; |
1298 | out_unlock_bdev: |
1299 | mutex_unlock(&bdev->bd_mutex); |
1300 | out_unlock_kernel: |
1301 | unlock_kernel(); |
1302 | |
1303 | if (disk) |
1304 | module_put(disk->fops->owner); |
1305 | put_disk(disk); |
1306 | bdput(bdev); |
1307 | |
1308 | return ret; |
1309 | } |
1310 | |
1311 | int blkdev_get(struct block_device *bdev, fmode_t mode) |
1312 | { |
1313 | return __blkdev_get(bdev, mode, 0); |
1314 | } |
1315 | EXPORT_SYMBOL(blkdev_get); |
1316 | |
1317 | static int blkdev_open(struct inode * inode, struct file * filp) |
1318 | { |
1319 | struct block_device *bdev; |
1320 | int res; |
1321 | |
1322 | /* |
1323 | * Preserve backwards compatibility and allow large file access |
1324 | * even if userspace doesn't ask for it explicitly. Some mkfs |
1325 | * binary needs it. We might want to drop this workaround |
1326 | * during an unstable branch. |
1327 | */ |
1328 | filp->f_flags |= O_LARGEFILE; |
1329 | |
1330 | if (filp->f_flags & O_NDELAY) |
1331 | filp->f_mode |= FMODE_NDELAY; |
1332 | if (filp->f_flags & O_EXCL) |
1333 | filp->f_mode |= FMODE_EXCL; |
1334 | if ((filp->f_flags & O_ACCMODE) == 3) |
1335 | filp->f_mode |= FMODE_WRITE_IOCTL; |
1336 | |
1337 | bdev = bd_acquire(inode); |
1338 | if (bdev == NULL) |
1339 | return -ENOMEM; |
1340 | |
1341 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1342 | |
1343 | res = blkdev_get(bdev, filp->f_mode); |
1344 | if (res) |
1345 | return res; |
1346 | |
1347 | if (filp->f_mode & FMODE_EXCL) { |
1348 | res = bd_claim(bdev, filp); |
1349 | if (res) |
1350 | goto out_blkdev_put; |
1351 | } |
1352 | |
1353 | return 0; |
1354 | |
1355 | out_blkdev_put: |
1356 | blkdev_put(bdev, filp->f_mode); |
1357 | return res; |
1358 | } |
1359 | |
1360 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
1361 | { |
1362 | int ret = 0; |
1363 | struct gendisk *disk = bdev->bd_disk; |
1364 | struct block_device *victim = NULL; |
1365 | |
1366 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1367 | lock_kernel(); |
1368 | if (for_part) |
1369 | bdev->bd_part_count--; |
1370 | |
1371 | if (!--bdev->bd_openers) { |
1372 | sync_blockdev(bdev); |
1373 | kill_bdev(bdev); |
1374 | } |
1375 | if (bdev->bd_contains == bdev) { |
1376 | if (disk->fops->release) |
1377 | ret = disk->fops->release(disk, mode); |
1378 | } |
1379 | if (!bdev->bd_openers) { |
1380 | struct module *owner = disk->fops->owner; |
1381 | |
1382 | put_disk(disk); |
1383 | module_put(owner); |
1384 | disk_put_part(bdev->bd_part); |
1385 | bdev->bd_part = NULL; |
1386 | bdev->bd_disk = NULL; |
1387 | bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info; |
1388 | if (bdev != bdev->bd_contains) |
1389 | victim = bdev->bd_contains; |
1390 | bdev->bd_contains = NULL; |
1391 | } |
1392 | unlock_kernel(); |
1393 | mutex_unlock(&bdev->bd_mutex); |
1394 | bdput(bdev); |
1395 | if (victim) |
1396 | __blkdev_put(victim, mode, 1); |
1397 | return ret; |
1398 | } |
1399 | |
1400 | int blkdev_put(struct block_device *bdev, fmode_t mode) |
1401 | { |
1402 | return __blkdev_put(bdev, mode, 0); |
1403 | } |
1404 | EXPORT_SYMBOL(blkdev_put); |
1405 | |
1406 | static int blkdev_close(struct inode * inode, struct file * filp) |
1407 | { |
1408 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); |
1409 | if (bdev->bd_holder == filp) |
1410 | bd_release(bdev); |
1411 | return blkdev_put(bdev, filp->f_mode); |
1412 | } |
1413 | |
1414 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1415 | { |
1416 | struct block_device *bdev = I_BDEV(file->f_mapping->host); |
1417 | fmode_t mode = file->f_mode; |
1418 | |
1419 | /* |
1420 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have |
1421 | * to updated it before every ioctl. |
1422 | */ |
1423 | if (file->f_flags & O_NDELAY) |
1424 | mode |= FMODE_NDELAY; |
1425 | else |
1426 | mode &= ~FMODE_NDELAY; |
1427 | |
1428 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1429 | } |
1430 | |
1431 | /* |
1432 | * Write data to the block device. Only intended for the block device itself |
1433 | * and the raw driver which basically is a fake block device. |
1434 | * |
1435 | * Does not take i_mutex for the write and thus is not for general purpose |
1436 | * use. |
1437 | */ |
1438 | ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov, |
1439 | unsigned long nr_segs, loff_t pos) |
1440 | { |
1441 | struct file *file = iocb->ki_filp; |
1442 | ssize_t ret; |
1443 | |
1444 | BUG_ON(iocb->ki_pos != pos); |
1445 | |
1446 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1447 | if (ret > 0 || ret == -EIOCBQUEUED) { |
1448 | ssize_t err; |
1449 | |
1450 | err = generic_write_sync(file, pos, ret); |
1451 | if (err < 0 && ret > 0) |
1452 | ret = err; |
1453 | } |
1454 | return ret; |
1455 | } |
1456 | EXPORT_SYMBOL_GPL(blkdev_aio_write); |
1457 | |
1458 | /* |
1459 | * Try to release a page associated with block device when the system |
1460 | * is under memory pressure. |
1461 | */ |
1462 | static int blkdev_releasepage(struct page *page, gfp_t wait) |
1463 | { |
1464 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; |
1465 | |
1466 | if (super && super->s_op->bdev_try_to_free_page) |
1467 | return super->s_op->bdev_try_to_free_page(super, page, wait); |
1468 | |
1469 | return try_to_free_buffers(page); |
1470 | } |
1471 | |
1472 | static const struct address_space_operations def_blk_aops = { |
1473 | .readpage = blkdev_readpage, |
1474 | .writepage = blkdev_writepage, |
1475 | .sync_page = block_sync_page, |
1476 | .write_begin = blkdev_write_begin, |
1477 | .write_end = blkdev_write_end, |
1478 | .writepages = generic_writepages, |
1479 | .releasepage = blkdev_releasepage, |
1480 | .direct_IO = blkdev_direct_IO, |
1481 | }; |
1482 | |
1483 | const struct file_operations def_blk_fops = { |
1484 | .open = blkdev_open, |
1485 | .release = blkdev_close, |
1486 | .llseek = block_llseek, |
1487 | .read = do_sync_read, |
1488 | .write = do_sync_write, |
1489 | .aio_read = generic_file_aio_read, |
1490 | .aio_write = blkdev_aio_write, |
1491 | .mmap = generic_file_mmap, |
1492 | .fsync = blkdev_fsync, |
1493 | .unlocked_ioctl = block_ioctl, |
1494 | #ifdef CONFIG_COMPAT |
1495 | .compat_ioctl = compat_blkdev_ioctl, |
1496 | #endif |
1497 | .splice_read = generic_file_splice_read, |
1498 | .splice_write = generic_file_splice_write, |
1499 | }; |
1500 | |
1501 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) |
1502 | { |
1503 | int res; |
1504 | mm_segment_t old_fs = get_fs(); |
1505 | set_fs(KERNEL_DS); |
1506 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1507 | set_fs(old_fs); |
1508 | return res; |
1509 | } |
1510 | |
1511 | EXPORT_SYMBOL(ioctl_by_bdev); |
1512 | |
1513 | /** |
1514 | * lookup_bdev - lookup a struct block_device by name |
1515 | * @pathname: special file representing the block device |
1516 | * |
1517 | * Get a reference to the blockdevice at @pathname in the current |
1518 | * namespace if possible and return it. Return ERR_PTR(error) |
1519 | * otherwise. |
1520 | */ |
1521 | struct block_device *lookup_bdev(const char *pathname) |
1522 | { |
1523 | struct block_device *bdev; |
1524 | struct inode *inode; |
1525 | struct path path; |
1526 | int error; |
1527 | |
1528 | if (!pathname || !*pathname) |
1529 | return ERR_PTR(-EINVAL); |
1530 | |
1531 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1532 | if (error) |
1533 | return ERR_PTR(error); |
1534 | |
1535 | inode = path.dentry->d_inode; |
1536 | error = -ENOTBLK; |
1537 | if (!S_ISBLK(inode->i_mode)) |
1538 | goto fail; |
1539 | error = -EACCES; |
1540 | if (path.mnt->mnt_flags & MNT_NODEV) |
1541 | goto fail; |
1542 | error = -ENOMEM; |
1543 | bdev = bd_acquire(inode); |
1544 | if (!bdev) |
1545 | goto fail; |
1546 | out: |
1547 | path_put(&path); |
1548 | return bdev; |
1549 | fail: |
1550 | bdev = ERR_PTR(error); |
1551 | goto out; |
1552 | } |
1553 | EXPORT_SYMBOL(lookup_bdev); |
1554 | |
1555 | /** |
1556 | * open_bdev_exclusive - open a block device by name and set it up for use |
1557 | * |
1558 | * @path: special file representing the block device |
1559 | * @mode: FMODE_... combination to pass be used |
1560 | * @holder: owner for exclusion |
1561 | * |
1562 | * Open the blockdevice described by the special file at @path, claim it |
1563 | * for the @holder. |
1564 | */ |
1565 | struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder) |
1566 | { |
1567 | struct block_device *bdev; |
1568 | int error = 0; |
1569 | |
1570 | bdev = lookup_bdev(path); |
1571 | if (IS_ERR(bdev)) |
1572 | return bdev; |
1573 | |
1574 | error = blkdev_get(bdev, mode); |
1575 | if (error) |
1576 | return ERR_PTR(error); |
1577 | error = -EACCES; |
1578 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) |
1579 | goto blkdev_put; |
1580 | error = bd_claim(bdev, holder); |
1581 | if (error) |
1582 | goto blkdev_put; |
1583 | |
1584 | return bdev; |
1585 | |
1586 | blkdev_put: |
1587 | blkdev_put(bdev, mode); |
1588 | return ERR_PTR(error); |
1589 | } |
1590 | |
1591 | EXPORT_SYMBOL(open_bdev_exclusive); |
1592 | |
1593 | /** |
1594 | * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive() |
1595 | * |
1596 | * @bdev: blockdevice to close |
1597 | * @mode: mode, must match that used to open. |
1598 | * |
1599 | * This is the counterpart to open_bdev_exclusive(). |
1600 | */ |
1601 | void close_bdev_exclusive(struct block_device *bdev, fmode_t mode) |
1602 | { |
1603 | bd_release(bdev); |
1604 | blkdev_put(bdev, mode); |
1605 | } |
1606 | |
1607 | EXPORT_SYMBOL(close_bdev_exclusive); |
1608 | |
1609 | int __invalidate_device(struct block_device *bdev) |
1610 | { |
1611 | struct super_block *sb = get_super(bdev); |
1612 | int res = 0; |
1613 | |
1614 | if (sb) { |
1615 | /* |
1616 | * no need to lock the super, get_super holds the |
1617 | * read mutex so the filesystem cannot go away |
1618 | * under us (->put_super runs with the write lock |
1619 | * hold). |
1620 | */ |
1621 | shrink_dcache_sb(sb); |
1622 | res = invalidate_inodes(sb); |
1623 | drop_super(sb); |
1624 | } |
1625 | invalidate_bdev(bdev); |
1626 | return res; |
1627 | } |
1628 | EXPORT_SYMBOL(__invalidate_device); |
1629 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9