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/device_cgroup.h> |
15 | #include <linux/highmem.h> |
16 | #include <linux/blkdev.h> |
17 | #include <linux/module.h> |
18 | #include <linux/blkpg.h> |
19 | #include <linux/buffer_head.h> |
20 | #include <linux/pagevec.h> |
21 | #include <linux/writeback.h> |
22 | #include <linux/mpage.h> |
23 | #include <linux/mount.h> |
24 | #include <linux/uio.h> |
25 | #include <linux/namei.h> |
26 | #include <linux/log2.h> |
27 | #include <linux/kmemleak.h> |
28 | #include <asm/uaccess.h> |
29 | #include "internal.h" |
30 | |
31 | struct bdev_inode { |
32 | struct block_device bdev; |
33 | struct inode vfs_inode; |
34 | }; |
35 | |
36 | static const struct address_space_operations def_blk_aops; |
37 | |
38 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
39 | { |
40 | return container_of(inode, struct bdev_inode, vfs_inode); |
41 | } |
42 | |
43 | inline struct block_device *I_BDEV(struct inode *inode) |
44 | { |
45 | return &BDEV_I(inode)->bdev; |
46 | } |
47 | |
48 | EXPORT_SYMBOL(I_BDEV); |
49 | |
50 | /* |
51 | * move the inode from it's current bdi to the a new bdi. if the inode is dirty |
52 | * we need to move it onto the dirty list of @dst so that the inode is always |
53 | * on the right list. |
54 | */ |
55 | static void bdev_inode_switch_bdi(struct inode *inode, |
56 | struct backing_dev_info *dst) |
57 | { |
58 | spin_lock(&inode_lock); |
59 | inode->i_data.backing_dev_info = dst; |
60 | if (inode->i_state & I_DIRTY) |
61 | list_move(&inode->i_wb_list, &dst->wb.b_dirty); |
62 | spin_unlock(&inode_lock); |
63 | } |
64 | |
65 | static sector_t max_block(struct block_device *bdev) |
66 | { |
67 | sector_t retval = ~((sector_t)0); |
68 | loff_t sz = i_size_read(bdev->bd_inode); |
69 | |
70 | if (sz) { |
71 | unsigned int size = block_size(bdev); |
72 | unsigned int sizebits = blksize_bits(size); |
73 | retval = (sz >> sizebits); |
74 | } |
75 | return retval; |
76 | } |
77 | |
78 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
79 | static void kill_bdev(struct block_device *bdev) |
80 | { |
81 | if (bdev->bd_inode->i_mapping->nrpages == 0) |
82 | return; |
83 | invalidate_bh_lrus(); |
84 | truncate_inode_pages(bdev->bd_inode->i_mapping, 0); |
85 | } |
86 | |
87 | int set_blocksize(struct block_device *bdev, int size) |
88 | { |
89 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ |
90 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
91 | return -EINVAL; |
92 | |
93 | /* Size cannot be smaller than the size supported by the device */ |
94 | if (size < bdev_logical_block_size(bdev)) |
95 | return -EINVAL; |
96 | |
97 | /* Don't change the size if it is same as current */ |
98 | if (bdev->bd_block_size != size) { |
99 | sync_blockdev(bdev); |
100 | bdev->bd_block_size = size; |
101 | bdev->bd_inode->i_blkbits = blksize_bits(size); |
102 | kill_bdev(bdev); |
103 | } |
104 | return 0; |
105 | } |
106 | |
107 | EXPORT_SYMBOL(set_blocksize); |
108 | |
109 | int sb_set_blocksize(struct super_block *sb, int size) |
110 | { |
111 | if (set_blocksize(sb->s_bdev, size)) |
112 | return 0; |
113 | /* If we get here, we know size is power of two |
114 | * and it's value is between 512 and PAGE_SIZE */ |
115 | sb->s_blocksize = size; |
116 | sb->s_blocksize_bits = blksize_bits(size); |
117 | return sb->s_blocksize; |
118 | } |
119 | |
120 | EXPORT_SYMBOL(sb_set_blocksize); |
121 | |
122 | int sb_min_blocksize(struct super_block *sb, int size) |
123 | { |
124 | int minsize = bdev_logical_block_size(sb->s_bdev); |
125 | if (size < minsize) |
126 | size = minsize; |
127 | return sb_set_blocksize(sb, size); |
128 | } |
129 | |
130 | EXPORT_SYMBOL(sb_min_blocksize); |
131 | |
132 | static int |
133 | blkdev_get_block(struct inode *inode, sector_t iblock, |
134 | struct buffer_head *bh, int create) |
135 | { |
136 | if (iblock >= max_block(I_BDEV(inode))) { |
137 | if (create) |
138 | return -EIO; |
139 | |
140 | /* |
141 | * for reads, we're just trying to fill a partial page. |
142 | * return a hole, they will have to call get_block again |
143 | * before they can fill it, and they will get -EIO at that |
144 | * time |
145 | */ |
146 | return 0; |
147 | } |
148 | bh->b_bdev = I_BDEV(inode); |
149 | bh->b_blocknr = iblock; |
150 | set_buffer_mapped(bh); |
151 | return 0; |
152 | } |
153 | |
154 | static int |
155 | blkdev_get_blocks(struct inode *inode, sector_t iblock, |
156 | struct buffer_head *bh, int create) |
157 | { |
158 | sector_t end_block = max_block(I_BDEV(inode)); |
159 | unsigned long max_blocks = bh->b_size >> inode->i_blkbits; |
160 | |
161 | if ((iblock + max_blocks) > end_block) { |
162 | max_blocks = end_block - iblock; |
163 | if ((long)max_blocks <= 0) { |
164 | if (create) |
165 | return -EIO; /* write fully beyond EOF */ |
166 | /* |
167 | * It is a read which is fully beyond EOF. We return |
168 | * a !buffer_mapped buffer |
169 | */ |
170 | max_blocks = 0; |
171 | } |
172 | } |
173 | |
174 | bh->b_bdev = I_BDEV(inode); |
175 | bh->b_blocknr = iblock; |
176 | bh->b_size = max_blocks << inode->i_blkbits; |
177 | if (max_blocks) |
178 | set_buffer_mapped(bh); |
179 | return 0; |
180 | } |
181 | |
182 | static ssize_t |
183 | blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, |
184 | loff_t offset, unsigned long nr_segs) |
185 | { |
186 | struct file *file = iocb->ki_filp; |
187 | struct inode *inode = file->f_mapping->host; |
188 | |
189 | return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset, |
190 | nr_segs, blkdev_get_blocks, NULL, NULL, 0); |
191 | } |
192 | |
193 | int __sync_blockdev(struct block_device *bdev, int wait) |
194 | { |
195 | if (!bdev) |
196 | return 0; |
197 | if (!wait) |
198 | return filemap_flush(bdev->bd_inode->i_mapping); |
199 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); |
200 | } |
201 | |
202 | /* |
203 | * Write out and wait upon all the dirty data associated with a block |
204 | * device via its mapping. Does not take the superblock lock. |
205 | */ |
206 | int sync_blockdev(struct block_device *bdev) |
207 | { |
208 | return __sync_blockdev(bdev, 1); |
209 | } |
210 | EXPORT_SYMBOL(sync_blockdev); |
211 | |
212 | /* |
213 | * Write out and wait upon all dirty data associated with this |
214 | * device. Filesystem data as well as the underlying block |
215 | * device. Takes the superblock lock. |
216 | */ |
217 | int fsync_bdev(struct block_device *bdev) |
218 | { |
219 | struct super_block *sb = get_super(bdev); |
220 | if (sb) { |
221 | int res = sync_filesystem(sb); |
222 | drop_super(sb); |
223 | return res; |
224 | } |
225 | return sync_blockdev(bdev); |
226 | } |
227 | EXPORT_SYMBOL(fsync_bdev); |
228 | |
229 | /** |
230 | * freeze_bdev -- lock a filesystem and force it into a consistent state |
231 | * @bdev: blockdevice to lock |
232 | * |
233 | * If a superblock is found on this device, we take the s_umount semaphore |
234 | * on it to make sure nobody unmounts until the snapshot creation is done. |
235 | * The reference counter (bd_fsfreeze_count) guarantees that only the last |
236 | * unfreeze process can unfreeze the frozen filesystem actually when multiple |
237 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and |
238 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze |
239 | * actually. |
240 | */ |
241 | struct super_block *freeze_bdev(struct block_device *bdev) |
242 | { |
243 | struct super_block *sb; |
244 | int error = 0; |
245 | |
246 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
247 | if (++bdev->bd_fsfreeze_count > 1) { |
248 | /* |
249 | * We don't even need to grab a reference - the first call |
250 | * to freeze_bdev grab an active reference and only the last |
251 | * thaw_bdev drops it. |
252 | */ |
253 | sb = get_super(bdev); |
254 | drop_super(sb); |
255 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
256 | return sb; |
257 | } |
258 | |
259 | sb = get_active_super(bdev); |
260 | if (!sb) |
261 | goto out; |
262 | error = freeze_super(sb); |
263 | if (error) { |
264 | deactivate_super(sb); |
265 | bdev->bd_fsfreeze_count--; |
266 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
267 | return ERR_PTR(error); |
268 | } |
269 | deactivate_super(sb); |
270 | out: |
271 | sync_blockdev(bdev); |
272 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
273 | return sb; /* thaw_bdev releases s->s_umount */ |
274 | } |
275 | EXPORT_SYMBOL(freeze_bdev); |
276 | |
277 | /** |
278 | * thaw_bdev -- unlock filesystem |
279 | * @bdev: blockdevice to unlock |
280 | * @sb: associated superblock |
281 | * |
282 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). |
283 | */ |
284 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) |
285 | { |
286 | int error = -EINVAL; |
287 | |
288 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
289 | if (!bdev->bd_fsfreeze_count) |
290 | goto out; |
291 | |
292 | error = 0; |
293 | if (--bdev->bd_fsfreeze_count > 0) |
294 | goto out; |
295 | |
296 | if (!sb) |
297 | goto out; |
298 | |
299 | error = thaw_super(sb); |
300 | if (error) { |
301 | bdev->bd_fsfreeze_count++; |
302 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
303 | return error; |
304 | } |
305 | out: |
306 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
307 | return 0; |
308 | } |
309 | EXPORT_SYMBOL(thaw_bdev); |
310 | |
311 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
312 | { |
313 | return block_write_full_page(page, blkdev_get_block, wbc); |
314 | } |
315 | |
316 | static int blkdev_readpage(struct file * file, struct page * page) |
317 | { |
318 | return block_read_full_page(page, blkdev_get_block); |
319 | } |
320 | |
321 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
322 | loff_t pos, unsigned len, unsigned flags, |
323 | struct page **pagep, void **fsdata) |
324 | { |
325 | return block_write_begin(mapping, pos, len, flags, pagep, |
326 | blkdev_get_block); |
327 | } |
328 | |
329 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
330 | loff_t pos, unsigned len, unsigned copied, |
331 | struct page *page, void *fsdata) |
332 | { |
333 | int ret; |
334 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
335 | |
336 | unlock_page(page); |
337 | page_cache_release(page); |
338 | |
339 | return ret; |
340 | } |
341 | |
342 | /* |
343 | * private llseek: |
344 | * for a block special file file->f_path.dentry->d_inode->i_size is zero |
345 | * so we compute the size by hand (just as in block_read/write above) |
346 | */ |
347 | static loff_t block_llseek(struct file *file, loff_t offset, int origin) |
348 | { |
349 | struct inode *bd_inode = file->f_mapping->host; |
350 | loff_t size; |
351 | loff_t retval; |
352 | |
353 | mutex_lock(&bd_inode->i_mutex); |
354 | size = i_size_read(bd_inode); |
355 | |
356 | switch (origin) { |
357 | case 2: |
358 | offset += size; |
359 | break; |
360 | case 1: |
361 | offset += file->f_pos; |
362 | } |
363 | retval = -EINVAL; |
364 | if (offset >= 0 && offset <= size) { |
365 | if (offset != file->f_pos) { |
366 | file->f_pos = offset; |
367 | } |
368 | retval = offset; |
369 | } |
370 | mutex_unlock(&bd_inode->i_mutex); |
371 | return retval; |
372 | } |
373 | |
374 | int blkdev_fsync(struct file *filp, int datasync) |
375 | { |
376 | struct inode *bd_inode = filp->f_mapping->host; |
377 | struct block_device *bdev = I_BDEV(bd_inode); |
378 | int error; |
379 | |
380 | /* |
381 | * There is no need to serialise calls to blkdev_issue_flush with |
382 | * i_mutex and doing so causes performance issues with concurrent |
383 | * O_SYNC writers to a block device. |
384 | */ |
385 | mutex_unlock(&bd_inode->i_mutex); |
386 | |
387 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
388 | if (error == -EOPNOTSUPP) |
389 | error = 0; |
390 | |
391 | mutex_lock(&bd_inode->i_mutex); |
392 | |
393 | return error; |
394 | } |
395 | EXPORT_SYMBOL(blkdev_fsync); |
396 | |
397 | /* |
398 | * pseudo-fs |
399 | */ |
400 | |
401 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); |
402 | static struct kmem_cache * bdev_cachep __read_mostly; |
403 | |
404 | static struct inode *bdev_alloc_inode(struct super_block *sb) |
405 | { |
406 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
407 | if (!ei) |
408 | return NULL; |
409 | return &ei->vfs_inode; |
410 | } |
411 | |
412 | static void bdev_i_callback(struct rcu_head *head) |
413 | { |
414 | struct inode *inode = container_of(head, struct inode, i_rcu); |
415 | struct bdev_inode *bdi = BDEV_I(inode); |
416 | |
417 | INIT_LIST_HEAD(&inode->i_dentry); |
418 | kmem_cache_free(bdev_cachep, bdi); |
419 | } |
420 | |
421 | static void bdev_destroy_inode(struct inode *inode) |
422 | { |
423 | call_rcu(&inode->i_rcu, bdev_i_callback); |
424 | } |
425 | |
426 | static void init_once(void *foo) |
427 | { |
428 | struct bdev_inode *ei = (struct bdev_inode *) foo; |
429 | struct block_device *bdev = &ei->bdev; |
430 | |
431 | memset(bdev, 0, sizeof(*bdev)); |
432 | mutex_init(&bdev->bd_mutex); |
433 | INIT_LIST_HEAD(&bdev->bd_inodes); |
434 | INIT_LIST_HEAD(&bdev->bd_list); |
435 | #ifdef CONFIG_SYSFS |
436 | INIT_LIST_HEAD(&bdev->bd_holder_disks); |
437 | #endif |
438 | inode_init_once(&ei->vfs_inode); |
439 | /* Initialize mutex for freeze. */ |
440 | mutex_init(&bdev->bd_fsfreeze_mutex); |
441 | } |
442 | |
443 | static inline void __bd_forget(struct inode *inode) |
444 | { |
445 | list_del_init(&inode->i_devices); |
446 | inode->i_bdev = NULL; |
447 | inode->i_mapping = &inode->i_data; |
448 | } |
449 | |
450 | static void bdev_evict_inode(struct inode *inode) |
451 | { |
452 | struct block_device *bdev = &BDEV_I(inode)->bdev; |
453 | struct list_head *p; |
454 | truncate_inode_pages(&inode->i_data, 0); |
455 | invalidate_inode_buffers(inode); /* is it needed here? */ |
456 | end_writeback(inode); |
457 | spin_lock(&bdev_lock); |
458 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { |
459 | __bd_forget(list_entry(p, struct inode, i_devices)); |
460 | } |
461 | list_del_init(&bdev->bd_list); |
462 | spin_unlock(&bdev_lock); |
463 | } |
464 | |
465 | static const struct super_operations bdev_sops = { |
466 | .statfs = simple_statfs, |
467 | .alloc_inode = bdev_alloc_inode, |
468 | .destroy_inode = bdev_destroy_inode, |
469 | .drop_inode = generic_delete_inode, |
470 | .evict_inode = bdev_evict_inode, |
471 | }; |
472 | |
473 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
474 | int flags, const char *dev_name, void *data) |
475 | { |
476 | return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576); |
477 | } |
478 | |
479 | static struct file_system_type bd_type = { |
480 | .name = "bdev", |
481 | .mount = bd_mount, |
482 | .kill_sb = kill_anon_super, |
483 | }; |
484 | |
485 | struct super_block *blockdev_superblock __read_mostly; |
486 | |
487 | void __init bdev_cache_init(void) |
488 | { |
489 | int err; |
490 | struct vfsmount *bd_mnt; |
491 | |
492 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
493 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
494 | SLAB_MEM_SPREAD|SLAB_PANIC), |
495 | init_once); |
496 | err = register_filesystem(&bd_type); |
497 | if (err) |
498 | panic("Cannot register bdev pseudo-fs"); |
499 | bd_mnt = kern_mount(&bd_type); |
500 | if (IS_ERR(bd_mnt)) |
501 | panic("Cannot create bdev pseudo-fs"); |
502 | /* |
503 | * This vfsmount structure is only used to obtain the |
504 | * blockdev_superblock, so tell kmemleak not to report it. |
505 | */ |
506 | kmemleak_not_leak(bd_mnt); |
507 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
508 | } |
509 | |
510 | /* |
511 | * Most likely _very_ bad one - but then it's hardly critical for small |
512 | * /dev and can be fixed when somebody will need really large one. |
513 | * Keep in mind that it will be fed through icache hash function too. |
514 | */ |
515 | static inline unsigned long hash(dev_t dev) |
516 | { |
517 | return MAJOR(dev)+MINOR(dev); |
518 | } |
519 | |
520 | static int bdev_test(struct inode *inode, void *data) |
521 | { |
522 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; |
523 | } |
524 | |
525 | static int bdev_set(struct inode *inode, void *data) |
526 | { |
527 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; |
528 | return 0; |
529 | } |
530 | |
531 | static LIST_HEAD(all_bdevs); |
532 | |
533 | struct block_device *bdget(dev_t dev) |
534 | { |
535 | struct block_device *bdev; |
536 | struct inode *inode; |
537 | |
538 | inode = iget5_locked(blockdev_superblock, hash(dev), |
539 | bdev_test, bdev_set, &dev); |
540 | |
541 | if (!inode) |
542 | return NULL; |
543 | |
544 | bdev = &BDEV_I(inode)->bdev; |
545 | |
546 | if (inode->i_state & I_NEW) { |
547 | bdev->bd_contains = NULL; |
548 | bdev->bd_inode = inode; |
549 | bdev->bd_block_size = (1 << inode->i_blkbits); |
550 | bdev->bd_part_count = 0; |
551 | bdev->bd_invalidated = 0; |
552 | inode->i_mode = S_IFBLK; |
553 | inode->i_rdev = dev; |
554 | inode->i_bdev = bdev; |
555 | inode->i_data.a_ops = &def_blk_aops; |
556 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); |
557 | inode->i_data.backing_dev_info = &default_backing_dev_info; |
558 | spin_lock(&bdev_lock); |
559 | list_add(&bdev->bd_list, &all_bdevs); |
560 | spin_unlock(&bdev_lock); |
561 | unlock_new_inode(inode); |
562 | } |
563 | return bdev; |
564 | } |
565 | |
566 | EXPORT_SYMBOL(bdget); |
567 | |
568 | /** |
569 | * bdgrab -- Grab a reference to an already referenced block device |
570 | * @bdev: Block device to grab a reference to. |
571 | */ |
572 | struct block_device *bdgrab(struct block_device *bdev) |
573 | { |
574 | ihold(bdev->bd_inode); |
575 | return bdev; |
576 | } |
577 | |
578 | long nr_blockdev_pages(void) |
579 | { |
580 | struct block_device *bdev; |
581 | long ret = 0; |
582 | spin_lock(&bdev_lock); |
583 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
584 | ret += bdev->bd_inode->i_mapping->nrpages; |
585 | } |
586 | spin_unlock(&bdev_lock); |
587 | return ret; |
588 | } |
589 | |
590 | void bdput(struct block_device *bdev) |
591 | { |
592 | iput(bdev->bd_inode); |
593 | } |
594 | |
595 | EXPORT_SYMBOL(bdput); |
596 | |
597 | static struct block_device *bd_acquire(struct inode *inode) |
598 | { |
599 | struct block_device *bdev; |
600 | |
601 | spin_lock(&bdev_lock); |
602 | bdev = inode->i_bdev; |
603 | if (bdev) { |
604 | ihold(bdev->bd_inode); |
605 | spin_unlock(&bdev_lock); |
606 | return bdev; |
607 | } |
608 | spin_unlock(&bdev_lock); |
609 | |
610 | bdev = bdget(inode->i_rdev); |
611 | if (bdev) { |
612 | spin_lock(&bdev_lock); |
613 | if (!inode->i_bdev) { |
614 | /* |
615 | * We take an additional reference to bd_inode, |
616 | * and it's released in clear_inode() of inode. |
617 | * So, we can access it via ->i_mapping always |
618 | * without igrab(). |
619 | */ |
620 | ihold(bdev->bd_inode); |
621 | inode->i_bdev = bdev; |
622 | inode->i_mapping = bdev->bd_inode->i_mapping; |
623 | list_add(&inode->i_devices, &bdev->bd_inodes); |
624 | } |
625 | spin_unlock(&bdev_lock); |
626 | } |
627 | return bdev; |
628 | } |
629 | |
630 | /* Call when you free inode */ |
631 | |
632 | void bd_forget(struct inode *inode) |
633 | { |
634 | struct block_device *bdev = NULL; |
635 | |
636 | spin_lock(&bdev_lock); |
637 | if (inode->i_bdev) { |
638 | if (!sb_is_blkdev_sb(inode->i_sb)) |
639 | bdev = inode->i_bdev; |
640 | __bd_forget(inode); |
641 | } |
642 | spin_unlock(&bdev_lock); |
643 | |
644 | if (bdev) |
645 | iput(bdev->bd_inode); |
646 | } |
647 | |
648 | /** |
649 | * bd_may_claim - test whether a block device can be claimed |
650 | * @bdev: block device of interest |
651 | * @whole: whole block device containing @bdev, may equal @bdev |
652 | * @holder: holder trying to claim @bdev |
653 | * |
654 | * Test whther @bdev can be claimed by @holder. |
655 | * |
656 | * CONTEXT: |
657 | * spin_lock(&bdev_lock). |
658 | * |
659 | * RETURNS: |
660 | * %true if @bdev can be claimed, %false otherwise. |
661 | */ |
662 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, |
663 | void *holder) |
664 | { |
665 | if (bdev->bd_holder == holder) |
666 | return true; /* already a holder */ |
667 | else if (bdev->bd_holder != NULL) |
668 | return false; /* held by someone else */ |
669 | else if (bdev->bd_contains == bdev) |
670 | return true; /* is a whole device which isn't held */ |
671 | |
672 | else if (whole->bd_holder == bd_may_claim) |
673 | return true; /* is a partition of a device that is being partitioned */ |
674 | else if (whole->bd_holder != NULL) |
675 | return false; /* is a partition of a held device */ |
676 | else |
677 | return true; /* is a partition of an un-held device */ |
678 | } |
679 | |
680 | /** |
681 | * bd_prepare_to_claim - prepare to claim a block device |
682 | * @bdev: block device of interest |
683 | * @whole: the whole device containing @bdev, may equal @bdev |
684 | * @holder: holder trying to claim @bdev |
685 | * |
686 | * Prepare to claim @bdev. This function fails if @bdev is already |
687 | * claimed by another holder and waits if another claiming is in |
688 | * progress. This function doesn't actually claim. On successful |
689 | * return, the caller has ownership of bd_claiming and bd_holder[s]. |
690 | * |
691 | * CONTEXT: |
692 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab |
693 | * it multiple times. |
694 | * |
695 | * RETURNS: |
696 | * 0 if @bdev can be claimed, -EBUSY otherwise. |
697 | */ |
698 | static int bd_prepare_to_claim(struct block_device *bdev, |
699 | struct block_device *whole, void *holder) |
700 | { |
701 | retry: |
702 | /* if someone else claimed, fail */ |
703 | if (!bd_may_claim(bdev, whole, holder)) |
704 | return -EBUSY; |
705 | |
706 | /* if claiming is already in progress, wait for it to finish */ |
707 | if (whole->bd_claiming) { |
708 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
709 | DEFINE_WAIT(wait); |
710 | |
711 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); |
712 | spin_unlock(&bdev_lock); |
713 | schedule(); |
714 | finish_wait(wq, &wait); |
715 | spin_lock(&bdev_lock); |
716 | goto retry; |
717 | } |
718 | |
719 | /* yay, all mine */ |
720 | return 0; |
721 | } |
722 | |
723 | /** |
724 | * bd_start_claiming - start claiming a block device |
725 | * @bdev: block device of interest |
726 | * @holder: holder trying to claim @bdev |
727 | * |
728 | * @bdev is about to be opened exclusively. Check @bdev can be opened |
729 | * exclusively and mark that an exclusive open is in progress. Each |
730 | * successful call to this function must be matched with a call to |
731 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
732 | * fail). |
733 | * |
734 | * This function is used to gain exclusive access to the block device |
735 | * without actually causing other exclusive open attempts to fail. It |
736 | * should be used when the open sequence itself requires exclusive |
737 | * access but may subsequently fail. |
738 | * |
739 | * CONTEXT: |
740 | * Might sleep. |
741 | * |
742 | * RETURNS: |
743 | * Pointer to the block device containing @bdev on success, ERR_PTR() |
744 | * value on failure. |
745 | */ |
746 | static struct block_device *bd_start_claiming(struct block_device *bdev, |
747 | void *holder) |
748 | { |
749 | struct gendisk *disk; |
750 | struct block_device *whole; |
751 | int partno, err; |
752 | |
753 | might_sleep(); |
754 | |
755 | /* |
756 | * @bdev might not have been initialized properly yet, look up |
757 | * and grab the outer block device the hard way. |
758 | */ |
759 | disk = get_gendisk(bdev->bd_dev, &partno); |
760 | if (!disk) |
761 | return ERR_PTR(-ENXIO); |
762 | |
763 | whole = bdget_disk(disk, 0); |
764 | module_put(disk->fops->owner); |
765 | put_disk(disk); |
766 | if (!whole) |
767 | return ERR_PTR(-ENOMEM); |
768 | |
769 | /* prepare to claim, if successful, mark claiming in progress */ |
770 | spin_lock(&bdev_lock); |
771 | |
772 | err = bd_prepare_to_claim(bdev, whole, holder); |
773 | if (err == 0) { |
774 | whole->bd_claiming = holder; |
775 | spin_unlock(&bdev_lock); |
776 | return whole; |
777 | } else { |
778 | spin_unlock(&bdev_lock); |
779 | bdput(whole); |
780 | return ERR_PTR(err); |
781 | } |
782 | } |
783 | |
784 | #ifdef CONFIG_SYSFS |
785 | struct bd_holder_disk { |
786 | struct list_head list; |
787 | struct gendisk *disk; |
788 | int refcnt; |
789 | }; |
790 | |
791 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, |
792 | struct gendisk *disk) |
793 | { |
794 | struct bd_holder_disk *holder; |
795 | |
796 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) |
797 | if (holder->disk == disk) |
798 | return holder; |
799 | return NULL; |
800 | } |
801 | |
802 | static int add_symlink(struct kobject *from, struct kobject *to) |
803 | { |
804 | return sysfs_create_link(from, to, kobject_name(to)); |
805 | } |
806 | |
807 | static void del_symlink(struct kobject *from, struct kobject *to) |
808 | { |
809 | sysfs_remove_link(from, kobject_name(to)); |
810 | } |
811 | |
812 | /** |
813 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
814 | * @bdev: the claimed slave bdev |
815 | * @disk: the holding disk |
816 | * |
817 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
818 | * |
819 | * This functions creates the following sysfs symlinks. |
820 | * |
821 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
822 | * - from "holders" directory of the @bdev to the holder @disk |
823 | * |
824 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
825 | * passed to bd_link_disk_holder(), then: |
826 | * |
827 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
828 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 |
829 | * |
830 | * The caller must have claimed @bdev before calling this function and |
831 | * ensure that both @bdev and @disk are valid during the creation and |
832 | * lifetime of these symlinks. |
833 | * |
834 | * CONTEXT: |
835 | * Might sleep. |
836 | * |
837 | * RETURNS: |
838 | * 0 on success, -errno on failure. |
839 | */ |
840 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
841 | { |
842 | struct bd_holder_disk *holder; |
843 | int ret = 0; |
844 | |
845 | mutex_lock(&bdev->bd_mutex); |
846 | |
847 | WARN_ON_ONCE(!bdev->bd_holder); |
848 | |
849 | /* FIXME: remove the following once add_disk() handles errors */ |
850 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) |
851 | goto out_unlock; |
852 | |
853 | holder = bd_find_holder_disk(bdev, disk); |
854 | if (holder) { |
855 | holder->refcnt++; |
856 | goto out_unlock; |
857 | } |
858 | |
859 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
860 | if (!holder) { |
861 | ret = -ENOMEM; |
862 | goto out_unlock; |
863 | } |
864 | |
865 | INIT_LIST_HEAD(&holder->list); |
866 | holder->disk = disk; |
867 | holder->refcnt = 1; |
868 | |
869 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
870 | if (ret) |
871 | goto out_free; |
872 | |
873 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); |
874 | if (ret) |
875 | goto out_del; |
876 | /* |
877 | * bdev could be deleted beneath us which would implicitly destroy |
878 | * the holder directory. Hold on to it. |
879 | */ |
880 | kobject_get(bdev->bd_part->holder_dir); |
881 | |
882 | list_add(&holder->list, &bdev->bd_holder_disks); |
883 | goto out_unlock; |
884 | |
885 | out_del: |
886 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
887 | out_free: |
888 | kfree(holder); |
889 | out_unlock: |
890 | mutex_unlock(&bdev->bd_mutex); |
891 | return ret; |
892 | } |
893 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
894 | |
895 | /** |
896 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() |
897 | * @bdev: the calimed slave bdev |
898 | * @disk: the holding disk |
899 | * |
900 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
901 | * |
902 | * CONTEXT: |
903 | * Might sleep. |
904 | */ |
905 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) |
906 | { |
907 | struct bd_holder_disk *holder; |
908 | |
909 | mutex_lock(&bdev->bd_mutex); |
910 | |
911 | holder = bd_find_holder_disk(bdev, disk); |
912 | |
913 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { |
914 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
915 | del_symlink(bdev->bd_part->holder_dir, |
916 | &disk_to_dev(disk)->kobj); |
917 | kobject_put(bdev->bd_part->holder_dir); |
918 | list_del_init(&holder->list); |
919 | kfree(holder); |
920 | } |
921 | |
922 | mutex_unlock(&bdev->bd_mutex); |
923 | } |
924 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
925 | #endif |
926 | |
927 | /** |
928 | * flush_disk - invalidates all buffer-cache entries on a disk |
929 | * |
930 | * @bdev: struct block device to be flushed |
931 | * @kill_dirty: flag to guide handling of dirty inodes |
932 | * |
933 | * Invalidates all buffer-cache entries on a disk. It should be called |
934 | * when a disk has been changed -- either by a media change or online |
935 | * resize. |
936 | */ |
937 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
938 | { |
939 | if (__invalidate_device(bdev, kill_dirty)) { |
940 | char name[BDEVNAME_SIZE] = ""; |
941 | |
942 | if (bdev->bd_disk) |
943 | disk_name(bdev->bd_disk, 0, name); |
944 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
945 | "resized disk %s\n", name); |
946 | } |
947 | |
948 | if (!bdev->bd_disk) |
949 | return; |
950 | if (disk_partitionable(bdev->bd_disk)) |
951 | bdev->bd_invalidated = 1; |
952 | } |
953 | |
954 | /** |
955 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
956 | * @disk: struct gendisk to check |
957 | * @bdev: struct bdev to adjust. |
958 | * |
959 | * This routine checks to see if the bdev size does not match the disk size |
960 | * and adjusts it if it differs. |
961 | */ |
962 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) |
963 | { |
964 | loff_t disk_size, bdev_size; |
965 | |
966 | disk_size = (loff_t)get_capacity(disk) << 9; |
967 | bdev_size = i_size_read(bdev->bd_inode); |
968 | if (disk_size != bdev_size) { |
969 | char name[BDEVNAME_SIZE]; |
970 | |
971 | disk_name(disk, 0, name); |
972 | printk(KERN_INFO |
973 | "%s: detected capacity change from %lld to %lld\n", |
974 | name, bdev_size, disk_size); |
975 | i_size_write(bdev->bd_inode, disk_size); |
976 | flush_disk(bdev, false); |
977 | } |
978 | } |
979 | EXPORT_SYMBOL(check_disk_size_change); |
980 | |
981 | /** |
982 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
983 | * @disk: struct gendisk to be revalidated |
984 | * |
985 | * This routine is a wrapper for lower-level driver's revalidate_disk |
986 | * call-backs. It is used to do common pre and post operations needed |
987 | * for all revalidate_disk operations. |
988 | */ |
989 | int revalidate_disk(struct gendisk *disk) |
990 | { |
991 | struct block_device *bdev; |
992 | int ret = 0; |
993 | |
994 | if (disk->fops->revalidate_disk) |
995 | ret = disk->fops->revalidate_disk(disk); |
996 | |
997 | bdev = bdget_disk(disk, 0); |
998 | if (!bdev) |
999 | return ret; |
1000 | |
1001 | mutex_lock(&bdev->bd_mutex); |
1002 | check_disk_size_change(disk, bdev); |
1003 | mutex_unlock(&bdev->bd_mutex); |
1004 | bdput(bdev); |
1005 | return ret; |
1006 | } |
1007 | EXPORT_SYMBOL(revalidate_disk); |
1008 | |
1009 | /* |
1010 | * This routine checks whether a removable media has been changed, |
1011 | * and invalidates all buffer-cache-entries in that case. This |
1012 | * is a relatively slow routine, so we have to try to minimize using |
1013 | * it. Thus it is called only upon a 'mount' or 'open'. This |
1014 | * is the best way of combining speed and utility, I think. |
1015 | * People changing diskettes in the middle of an operation deserve |
1016 | * to lose :-) |
1017 | */ |
1018 | int check_disk_change(struct block_device *bdev) |
1019 | { |
1020 | struct gendisk *disk = bdev->bd_disk; |
1021 | const struct block_device_operations *bdops = disk->fops; |
1022 | unsigned int events; |
1023 | |
1024 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1025 | DISK_EVENT_EJECT_REQUEST); |
1026 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) |
1027 | return 0; |
1028 | |
1029 | flush_disk(bdev, true); |
1030 | if (bdops->revalidate_disk) |
1031 | bdops->revalidate_disk(bdev->bd_disk); |
1032 | return 1; |
1033 | } |
1034 | |
1035 | EXPORT_SYMBOL(check_disk_change); |
1036 | |
1037 | void bd_set_size(struct block_device *bdev, loff_t size) |
1038 | { |
1039 | unsigned bsize = bdev_logical_block_size(bdev); |
1040 | |
1041 | bdev->bd_inode->i_size = size; |
1042 | while (bsize < PAGE_CACHE_SIZE) { |
1043 | if (size & bsize) |
1044 | break; |
1045 | bsize <<= 1; |
1046 | } |
1047 | bdev->bd_block_size = bsize; |
1048 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); |
1049 | } |
1050 | EXPORT_SYMBOL(bd_set_size); |
1051 | |
1052 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
1053 | |
1054 | /* |
1055 | * bd_mutex locking: |
1056 | * |
1057 | * mutex_lock(part->bd_mutex) |
1058 | * mutex_lock_nested(whole->bd_mutex, 1) |
1059 | */ |
1060 | |
1061 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1062 | { |
1063 | struct gendisk *disk; |
1064 | int ret; |
1065 | int partno; |
1066 | int perm = 0; |
1067 | |
1068 | if (mode & FMODE_READ) |
1069 | perm |= MAY_READ; |
1070 | if (mode & FMODE_WRITE) |
1071 | perm |= MAY_WRITE; |
1072 | /* |
1073 | * hooks: /n/, see "layering violations". |
1074 | */ |
1075 | if (!for_part) { |
1076 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); |
1077 | if (ret != 0) { |
1078 | bdput(bdev); |
1079 | return ret; |
1080 | } |
1081 | } |
1082 | |
1083 | restart: |
1084 | |
1085 | ret = -ENXIO; |
1086 | disk = get_gendisk(bdev->bd_dev, &partno); |
1087 | if (!disk) |
1088 | goto out; |
1089 | |
1090 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1091 | if (!bdev->bd_openers) { |
1092 | bdev->bd_disk = disk; |
1093 | bdev->bd_contains = bdev; |
1094 | if (!partno) { |
1095 | struct backing_dev_info *bdi; |
1096 | |
1097 | ret = -ENXIO; |
1098 | bdev->bd_part = disk_get_part(disk, partno); |
1099 | if (!bdev->bd_part) |
1100 | goto out_clear; |
1101 | |
1102 | if (disk->fops->open) { |
1103 | ret = disk->fops->open(bdev, mode); |
1104 | if (ret == -ERESTARTSYS) { |
1105 | /* Lost a race with 'disk' being |
1106 | * deleted, try again. |
1107 | * See md.c |
1108 | */ |
1109 | disk_put_part(bdev->bd_part); |
1110 | bdev->bd_part = NULL; |
1111 | module_put(disk->fops->owner); |
1112 | put_disk(disk); |
1113 | bdev->bd_disk = NULL; |
1114 | mutex_unlock(&bdev->bd_mutex); |
1115 | goto restart; |
1116 | } |
1117 | if (ret) |
1118 | goto out_clear; |
1119 | } |
1120 | if (!bdev->bd_openers) { |
1121 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
1122 | bdi = blk_get_backing_dev_info(bdev); |
1123 | if (bdi == NULL) |
1124 | bdi = &default_backing_dev_info; |
1125 | bdev_inode_switch_bdi(bdev->bd_inode, bdi); |
1126 | } |
1127 | if (bdev->bd_invalidated) |
1128 | rescan_partitions(disk, bdev); |
1129 | } else { |
1130 | struct block_device *whole; |
1131 | whole = bdget_disk(disk, 0); |
1132 | ret = -ENOMEM; |
1133 | if (!whole) |
1134 | goto out_clear; |
1135 | BUG_ON(for_part); |
1136 | ret = __blkdev_get(whole, mode, 1); |
1137 | if (ret) |
1138 | goto out_clear; |
1139 | bdev->bd_contains = whole; |
1140 | bdev_inode_switch_bdi(bdev->bd_inode, |
1141 | whole->bd_inode->i_data.backing_dev_info); |
1142 | bdev->bd_part = disk_get_part(disk, partno); |
1143 | if (!(disk->flags & GENHD_FL_UP) || |
1144 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1145 | ret = -ENXIO; |
1146 | goto out_clear; |
1147 | } |
1148 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1149 | } |
1150 | } else { |
1151 | module_put(disk->fops->owner); |
1152 | put_disk(disk); |
1153 | disk = NULL; |
1154 | if (bdev->bd_contains == bdev) { |
1155 | if (bdev->bd_disk->fops->open) { |
1156 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1157 | if (ret) |
1158 | goto out_unlock_bdev; |
1159 | } |
1160 | if (bdev->bd_invalidated) |
1161 | rescan_partitions(bdev->bd_disk, bdev); |
1162 | } |
1163 | } |
1164 | bdev->bd_openers++; |
1165 | if (for_part) |
1166 | bdev->bd_part_count++; |
1167 | mutex_unlock(&bdev->bd_mutex); |
1168 | return 0; |
1169 | |
1170 | out_clear: |
1171 | disk_put_part(bdev->bd_part); |
1172 | bdev->bd_disk = NULL; |
1173 | bdev->bd_part = NULL; |
1174 | bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info); |
1175 | if (bdev != bdev->bd_contains) |
1176 | __blkdev_put(bdev->bd_contains, mode, 1); |
1177 | bdev->bd_contains = NULL; |
1178 | out_unlock_bdev: |
1179 | mutex_unlock(&bdev->bd_mutex); |
1180 | out: |
1181 | if (disk) |
1182 | module_put(disk->fops->owner); |
1183 | put_disk(disk); |
1184 | bdput(bdev); |
1185 | |
1186 | return ret; |
1187 | } |
1188 | |
1189 | /** |
1190 | * blkdev_get - open a block device |
1191 | * @bdev: block_device to open |
1192 | * @mode: FMODE_* mask |
1193 | * @holder: exclusive holder identifier |
1194 | * |
1195 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is |
1196 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL |
1197 | * @holder is invalid. Exclusive opens may nest for the same @holder. |
1198 | * |
1199 | * On success, the reference count of @bdev is unchanged. On failure, |
1200 | * @bdev is put. |
1201 | * |
1202 | * CONTEXT: |
1203 | * Might sleep. |
1204 | * |
1205 | * RETURNS: |
1206 | * 0 on success, -errno on failure. |
1207 | */ |
1208 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1209 | { |
1210 | struct block_device *whole = NULL; |
1211 | int res; |
1212 | |
1213 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); |
1214 | |
1215 | if ((mode & FMODE_EXCL) && holder) { |
1216 | whole = bd_start_claiming(bdev, holder); |
1217 | if (IS_ERR(whole)) { |
1218 | bdput(bdev); |
1219 | return PTR_ERR(whole); |
1220 | } |
1221 | } |
1222 | |
1223 | res = __blkdev_get(bdev, mode, 0); |
1224 | |
1225 | if (whole) { |
1226 | /* finish claiming */ |
1227 | mutex_lock(&bdev->bd_mutex); |
1228 | spin_lock(&bdev_lock); |
1229 | |
1230 | if (!res) { |
1231 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1232 | /* |
1233 | * Note that for a whole device bd_holders |
1234 | * will be incremented twice, and bd_holder |
1235 | * will be set to bd_may_claim before being |
1236 | * set to holder |
1237 | */ |
1238 | whole->bd_holders++; |
1239 | whole->bd_holder = bd_may_claim; |
1240 | bdev->bd_holders++; |
1241 | bdev->bd_holder = holder; |
1242 | } |
1243 | |
1244 | /* tell others that we're done */ |
1245 | BUG_ON(whole->bd_claiming != holder); |
1246 | whole->bd_claiming = NULL; |
1247 | wake_up_bit(&whole->bd_claiming, 0); |
1248 | |
1249 | spin_unlock(&bdev_lock); |
1250 | |
1251 | /* |
1252 | * Block event polling for write claims. Any write |
1253 | * holder makes the write_holder state stick until all |
1254 | * are released. This is good enough and tracking |
1255 | * individual writeable reference is too fragile given |
1256 | * the way @mode is used in blkdev_get/put(). |
1257 | */ |
1258 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder) { |
1259 | bdev->bd_write_holder = true; |
1260 | disk_block_events(bdev->bd_disk); |
1261 | } |
1262 | |
1263 | mutex_unlock(&bdev->bd_mutex); |
1264 | bdput(whole); |
1265 | } |
1266 | |
1267 | return res; |
1268 | } |
1269 | EXPORT_SYMBOL(blkdev_get); |
1270 | |
1271 | /** |
1272 | * blkdev_get_by_path - open a block device by name |
1273 | * @path: path to the block device to open |
1274 | * @mode: FMODE_* mask |
1275 | * @holder: exclusive holder identifier |
1276 | * |
1277 | * Open the blockdevice described by the device file at @path. @mode |
1278 | * and @holder are identical to blkdev_get(). |
1279 | * |
1280 | * On success, the returned block_device has reference count of one. |
1281 | * |
1282 | * CONTEXT: |
1283 | * Might sleep. |
1284 | * |
1285 | * RETURNS: |
1286 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
1287 | */ |
1288 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, |
1289 | void *holder) |
1290 | { |
1291 | struct block_device *bdev; |
1292 | int err; |
1293 | |
1294 | bdev = lookup_bdev(path); |
1295 | if (IS_ERR(bdev)) |
1296 | return bdev; |
1297 | |
1298 | err = blkdev_get(bdev, mode, holder); |
1299 | if (err) |
1300 | return ERR_PTR(err); |
1301 | |
1302 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1303 | blkdev_put(bdev, mode); |
1304 | return ERR_PTR(-EACCES); |
1305 | } |
1306 | |
1307 | return bdev; |
1308 | } |
1309 | EXPORT_SYMBOL(blkdev_get_by_path); |
1310 | |
1311 | /** |
1312 | * blkdev_get_by_dev - open a block device by device number |
1313 | * @dev: device number of block device to open |
1314 | * @mode: FMODE_* mask |
1315 | * @holder: exclusive holder identifier |
1316 | * |
1317 | * Open the blockdevice described by device number @dev. @mode and |
1318 | * @holder are identical to blkdev_get(). |
1319 | * |
1320 | * Use it ONLY if you really do not have anything better - i.e. when |
1321 | * you are behind a truly sucky interface and all you are given is a |
1322 | * device number. _Never_ to be used for internal purposes. If you |
1323 | * ever need it - reconsider your API. |
1324 | * |
1325 | * On success, the returned block_device has reference count of one. |
1326 | * |
1327 | * CONTEXT: |
1328 | * Might sleep. |
1329 | * |
1330 | * RETURNS: |
1331 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
1332 | */ |
1333 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) |
1334 | { |
1335 | struct block_device *bdev; |
1336 | int err; |
1337 | |
1338 | bdev = bdget(dev); |
1339 | if (!bdev) |
1340 | return ERR_PTR(-ENOMEM); |
1341 | |
1342 | err = blkdev_get(bdev, mode, holder); |
1343 | if (err) |
1344 | return ERR_PTR(err); |
1345 | |
1346 | return bdev; |
1347 | } |
1348 | EXPORT_SYMBOL(blkdev_get_by_dev); |
1349 | |
1350 | static int blkdev_open(struct inode * inode, struct file * filp) |
1351 | { |
1352 | struct block_device *bdev; |
1353 | |
1354 | /* |
1355 | * Preserve backwards compatibility and allow large file access |
1356 | * even if userspace doesn't ask for it explicitly. Some mkfs |
1357 | * binary needs it. We might want to drop this workaround |
1358 | * during an unstable branch. |
1359 | */ |
1360 | filp->f_flags |= O_LARGEFILE; |
1361 | |
1362 | if (filp->f_flags & O_NDELAY) |
1363 | filp->f_mode |= FMODE_NDELAY; |
1364 | if (filp->f_flags & O_EXCL) |
1365 | filp->f_mode |= FMODE_EXCL; |
1366 | if ((filp->f_flags & O_ACCMODE) == 3) |
1367 | filp->f_mode |= FMODE_WRITE_IOCTL; |
1368 | |
1369 | bdev = bd_acquire(inode); |
1370 | if (bdev == NULL) |
1371 | return -ENOMEM; |
1372 | |
1373 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1374 | |
1375 | return blkdev_get(bdev, filp->f_mode, filp); |
1376 | } |
1377 | |
1378 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
1379 | { |
1380 | int ret = 0; |
1381 | struct gendisk *disk = bdev->bd_disk; |
1382 | struct block_device *victim = NULL; |
1383 | |
1384 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1385 | if (for_part) |
1386 | bdev->bd_part_count--; |
1387 | |
1388 | if (!--bdev->bd_openers) { |
1389 | WARN_ON_ONCE(bdev->bd_holders); |
1390 | sync_blockdev(bdev); |
1391 | kill_bdev(bdev); |
1392 | } |
1393 | if (bdev->bd_contains == bdev) { |
1394 | if (disk->fops->release) |
1395 | ret = disk->fops->release(disk, mode); |
1396 | } |
1397 | if (!bdev->bd_openers) { |
1398 | struct module *owner = disk->fops->owner; |
1399 | |
1400 | put_disk(disk); |
1401 | module_put(owner); |
1402 | disk_put_part(bdev->bd_part); |
1403 | bdev->bd_part = NULL; |
1404 | bdev->bd_disk = NULL; |
1405 | bdev_inode_switch_bdi(bdev->bd_inode, |
1406 | &default_backing_dev_info); |
1407 | if (bdev != bdev->bd_contains) |
1408 | victim = bdev->bd_contains; |
1409 | bdev->bd_contains = NULL; |
1410 | } |
1411 | mutex_unlock(&bdev->bd_mutex); |
1412 | bdput(bdev); |
1413 | if (victim) |
1414 | __blkdev_put(victim, mode, 1); |
1415 | return ret; |
1416 | } |
1417 | |
1418 | int blkdev_put(struct block_device *bdev, fmode_t mode) |
1419 | { |
1420 | if (mode & FMODE_EXCL) { |
1421 | bool bdev_free; |
1422 | |
1423 | /* |
1424 | * Release a claim on the device. The holder fields |
1425 | * are protected with bdev_lock. bd_mutex is to |
1426 | * synchronize disk_holder unlinking. |
1427 | */ |
1428 | mutex_lock(&bdev->bd_mutex); |
1429 | spin_lock(&bdev_lock); |
1430 | |
1431 | WARN_ON_ONCE(--bdev->bd_holders < 0); |
1432 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); |
1433 | |
1434 | /* bd_contains might point to self, check in a separate step */ |
1435 | if ((bdev_free = !bdev->bd_holders)) |
1436 | bdev->bd_holder = NULL; |
1437 | if (!bdev->bd_contains->bd_holders) |
1438 | bdev->bd_contains->bd_holder = NULL; |
1439 | |
1440 | spin_unlock(&bdev_lock); |
1441 | |
1442 | /* |
1443 | * If this was the last claim, remove holder link and |
1444 | * unblock evpoll if it was a write holder. |
1445 | */ |
1446 | if (bdev_free) { |
1447 | if (bdev->bd_write_holder) { |
1448 | disk_unblock_events(bdev->bd_disk); |
1449 | bdev->bd_write_holder = false; |
1450 | } else |
1451 | disk_check_events(bdev->bd_disk); |
1452 | } |
1453 | |
1454 | mutex_unlock(&bdev->bd_mutex); |
1455 | } else |
1456 | disk_check_events(bdev->bd_disk); |
1457 | |
1458 | return __blkdev_put(bdev, mode, 0); |
1459 | } |
1460 | EXPORT_SYMBOL(blkdev_put); |
1461 | |
1462 | static int blkdev_close(struct inode * inode, struct file * filp) |
1463 | { |
1464 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); |
1465 | |
1466 | return blkdev_put(bdev, filp->f_mode); |
1467 | } |
1468 | |
1469 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1470 | { |
1471 | struct block_device *bdev = I_BDEV(file->f_mapping->host); |
1472 | fmode_t mode = file->f_mode; |
1473 | |
1474 | /* |
1475 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have |
1476 | * to updated it before every ioctl. |
1477 | */ |
1478 | if (file->f_flags & O_NDELAY) |
1479 | mode |= FMODE_NDELAY; |
1480 | else |
1481 | mode &= ~FMODE_NDELAY; |
1482 | |
1483 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1484 | } |
1485 | |
1486 | /* |
1487 | * Write data to the block device. Only intended for the block device itself |
1488 | * and the raw driver which basically is a fake block device. |
1489 | * |
1490 | * Does not take i_mutex for the write and thus is not for general purpose |
1491 | * use. |
1492 | */ |
1493 | ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov, |
1494 | unsigned long nr_segs, loff_t pos) |
1495 | { |
1496 | struct file *file = iocb->ki_filp; |
1497 | ssize_t ret; |
1498 | |
1499 | BUG_ON(iocb->ki_pos != pos); |
1500 | |
1501 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1502 | if (ret > 0 || ret == -EIOCBQUEUED) { |
1503 | ssize_t err; |
1504 | |
1505 | err = generic_write_sync(file, pos, ret); |
1506 | if (err < 0 && ret > 0) |
1507 | ret = err; |
1508 | } |
1509 | return ret; |
1510 | } |
1511 | EXPORT_SYMBOL_GPL(blkdev_aio_write); |
1512 | |
1513 | /* |
1514 | * Try to release a page associated with block device when the system |
1515 | * is under memory pressure. |
1516 | */ |
1517 | static int blkdev_releasepage(struct page *page, gfp_t wait) |
1518 | { |
1519 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; |
1520 | |
1521 | if (super && super->s_op->bdev_try_to_free_page) |
1522 | return super->s_op->bdev_try_to_free_page(super, page, wait); |
1523 | |
1524 | return try_to_free_buffers(page); |
1525 | } |
1526 | |
1527 | static const struct address_space_operations def_blk_aops = { |
1528 | .readpage = blkdev_readpage, |
1529 | .writepage = blkdev_writepage, |
1530 | .sync_page = block_sync_page, |
1531 | .write_begin = blkdev_write_begin, |
1532 | .write_end = blkdev_write_end, |
1533 | .writepages = generic_writepages, |
1534 | .releasepage = blkdev_releasepage, |
1535 | .direct_IO = blkdev_direct_IO, |
1536 | }; |
1537 | |
1538 | const struct file_operations def_blk_fops = { |
1539 | .open = blkdev_open, |
1540 | .release = blkdev_close, |
1541 | .llseek = block_llseek, |
1542 | .read = do_sync_read, |
1543 | .write = do_sync_write, |
1544 | .aio_read = generic_file_aio_read, |
1545 | .aio_write = blkdev_aio_write, |
1546 | .mmap = generic_file_mmap, |
1547 | .fsync = blkdev_fsync, |
1548 | .unlocked_ioctl = block_ioctl, |
1549 | #ifdef CONFIG_COMPAT |
1550 | .compat_ioctl = compat_blkdev_ioctl, |
1551 | #endif |
1552 | .splice_read = generic_file_splice_read, |
1553 | .splice_write = generic_file_splice_write, |
1554 | }; |
1555 | |
1556 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) |
1557 | { |
1558 | int res; |
1559 | mm_segment_t old_fs = get_fs(); |
1560 | set_fs(KERNEL_DS); |
1561 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1562 | set_fs(old_fs); |
1563 | return res; |
1564 | } |
1565 | |
1566 | EXPORT_SYMBOL(ioctl_by_bdev); |
1567 | |
1568 | /** |
1569 | * lookup_bdev - lookup a struct block_device by name |
1570 | * @pathname: special file representing the block device |
1571 | * |
1572 | * Get a reference to the blockdevice at @pathname in the current |
1573 | * namespace if possible and return it. Return ERR_PTR(error) |
1574 | * otherwise. |
1575 | */ |
1576 | struct block_device *lookup_bdev(const char *pathname) |
1577 | { |
1578 | struct block_device *bdev; |
1579 | struct inode *inode; |
1580 | struct path path; |
1581 | int error; |
1582 | |
1583 | if (!pathname || !*pathname) |
1584 | return ERR_PTR(-EINVAL); |
1585 | |
1586 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1587 | if (error) |
1588 | return ERR_PTR(error); |
1589 | |
1590 | inode = path.dentry->d_inode; |
1591 | error = -ENOTBLK; |
1592 | if (!S_ISBLK(inode->i_mode)) |
1593 | goto fail; |
1594 | error = -EACCES; |
1595 | if (path.mnt->mnt_flags & MNT_NODEV) |
1596 | goto fail; |
1597 | error = -ENOMEM; |
1598 | bdev = bd_acquire(inode); |
1599 | if (!bdev) |
1600 | goto fail; |
1601 | out: |
1602 | path_put(&path); |
1603 | return bdev; |
1604 | fail: |
1605 | bdev = ERR_PTR(error); |
1606 | goto out; |
1607 | } |
1608 | EXPORT_SYMBOL(lookup_bdev); |
1609 | |
1610 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
1611 | { |
1612 | struct super_block *sb = get_super(bdev); |
1613 | int res = 0; |
1614 | |
1615 | if (sb) { |
1616 | /* |
1617 | * no need to lock the super, get_super holds the |
1618 | * read mutex so the filesystem cannot go away |
1619 | * under us (->put_super runs with the write lock |
1620 | * hold). |
1621 | */ |
1622 | shrink_dcache_sb(sb); |
1623 | res = invalidate_inodes(sb, kill_dirty); |
1624 | drop_super(sb); |
1625 | } |
1626 | invalidate_bdev(bdev); |
1627 | return res; |
1628 | } |
1629 | EXPORT_SYMBOL(__invalidate_device); |
1630 |
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