]>
Commit | Line | Data |
---|---|---|
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/swap.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/cleancache.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 | EXPORT_SYMBOL(I_BDEV); | |
49 | ||
50 | /* | |
51 | * Move the inode from its current bdi to a new bdi. If the inode is dirty we | |
52 | * need to move it onto the dirty list of @dst so that the inode is always on | |
53 | * the right list. | |
54 | */ | |
55 | static void bdev_inode_switch_bdi(struct inode *inode, | |
56 | struct backing_dev_info *dst) | |
57 | { | |
58 | struct backing_dev_info *old = inode->i_data.backing_dev_info; | |
59 | ||
60 | if (unlikely(dst == old)) /* deadlock avoidance */ | |
61 | return; | |
62 | bdi_lock_two(&old->wb, &dst->wb); | |
63 | spin_lock(&inode->i_lock); | |
64 | inode->i_data.backing_dev_info = dst; | |
65 | if (inode->i_state & I_DIRTY) | |
66 | list_move(&inode->i_wb_list, &dst->wb.b_dirty); | |
67 | spin_unlock(&inode->i_lock); | |
68 | spin_unlock(&old->wb.list_lock); | |
69 | spin_unlock(&dst->wb.list_lock); | |
70 | } | |
71 | ||
72 | static sector_t max_block(struct block_device *bdev) | |
73 | { | |
74 | sector_t retval = ~((sector_t)0); | |
75 | loff_t sz = i_size_read(bdev->bd_inode); | |
76 | ||
77 | if (sz) { | |
78 | unsigned int size = block_size(bdev); | |
79 | unsigned int sizebits = blksize_bits(size); | |
80 | retval = (sz >> sizebits); | |
81 | } | |
82 | return retval; | |
83 | } | |
84 | ||
85 | /* Kill _all_ buffers and pagecache , dirty or not.. */ | |
86 | void kill_bdev(struct block_device *bdev) | |
87 | { | |
88 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
89 | ||
90 | if (mapping->nrpages == 0) | |
91 | return; | |
92 | ||
93 | invalidate_bh_lrus(); | |
94 | truncate_inode_pages(mapping, 0); | |
95 | } | |
96 | EXPORT_SYMBOL(kill_bdev); | |
97 | ||
98 | /* Invalidate clean unused buffers and pagecache. */ | |
99 | void invalidate_bdev(struct block_device *bdev) | |
100 | { | |
101 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
102 | ||
103 | if (mapping->nrpages == 0) | |
104 | return; | |
105 | ||
106 | invalidate_bh_lrus(); | |
107 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
108 | invalidate_mapping_pages(mapping, 0, -1); | |
109 | /* 99% of the time, we don't need to flush the cleancache on the bdev. | |
110 | * But, for the strange corners, lets be cautious | |
111 | */ | |
112 | cleancache_flush_inode(mapping); | |
113 | } | |
114 | EXPORT_SYMBOL(invalidate_bdev); | |
115 | ||
116 | int set_blocksize(struct block_device *bdev, int size) | |
117 | { | |
118 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
119 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) | |
120 | return -EINVAL; | |
121 | ||
122 | /* Size cannot be smaller than the size supported by the device */ | |
123 | if (size < bdev_logical_block_size(bdev)) | |
124 | return -EINVAL; | |
125 | ||
126 | /* Don't change the size if it is same as current */ | |
127 | if (bdev->bd_block_size != size) { | |
128 | sync_blockdev(bdev); | |
129 | bdev->bd_block_size = size; | |
130 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
131 | kill_bdev(bdev); | |
132 | } | |
133 | return 0; | |
134 | } | |
135 | ||
136 | EXPORT_SYMBOL(set_blocksize); | |
137 | ||
138 | int sb_set_blocksize(struct super_block *sb, int size) | |
139 | { | |
140 | if (set_blocksize(sb->s_bdev, size)) | |
141 | return 0; | |
142 | /* If we get here, we know size is power of two | |
143 | * and it's value is between 512 and PAGE_SIZE */ | |
144 | sb->s_blocksize = size; | |
145 | sb->s_blocksize_bits = blksize_bits(size); | |
146 | return sb->s_blocksize; | |
147 | } | |
148 | ||
149 | EXPORT_SYMBOL(sb_set_blocksize); | |
150 | ||
151 | int sb_min_blocksize(struct super_block *sb, int size) | |
152 | { | |
153 | int minsize = bdev_logical_block_size(sb->s_bdev); | |
154 | if (size < minsize) | |
155 | size = minsize; | |
156 | return sb_set_blocksize(sb, size); | |
157 | } | |
158 | ||
159 | EXPORT_SYMBOL(sb_min_blocksize); | |
160 | ||
161 | static int | |
162 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
163 | struct buffer_head *bh, int create) | |
164 | { | |
165 | if (iblock >= max_block(I_BDEV(inode))) { | |
166 | if (create) | |
167 | return -EIO; | |
168 | ||
169 | /* | |
170 | * for reads, we're just trying to fill a partial page. | |
171 | * return a hole, they will have to call get_block again | |
172 | * before they can fill it, and they will get -EIO at that | |
173 | * time | |
174 | */ | |
175 | return 0; | |
176 | } | |
177 | bh->b_bdev = I_BDEV(inode); | |
178 | bh->b_blocknr = iblock; | |
179 | set_buffer_mapped(bh); | |
180 | return 0; | |
181 | } | |
182 | ||
183 | static int | |
184 | blkdev_get_blocks(struct inode *inode, sector_t iblock, | |
185 | struct buffer_head *bh, int create) | |
186 | { | |
187 | sector_t end_block = max_block(I_BDEV(inode)); | |
188 | unsigned long max_blocks = bh->b_size >> inode->i_blkbits; | |
189 | ||
190 | if ((iblock + max_blocks) > end_block) { | |
191 | max_blocks = end_block - iblock; | |
192 | if ((long)max_blocks <= 0) { | |
193 | if (create) | |
194 | return -EIO; /* write fully beyond EOF */ | |
195 | /* | |
196 | * It is a read which is fully beyond EOF. We return | |
197 | * a !buffer_mapped buffer | |
198 | */ | |
199 | max_blocks = 0; | |
200 | } | |
201 | } | |
202 | ||
203 | bh->b_bdev = I_BDEV(inode); | |
204 | bh->b_blocknr = iblock; | |
205 | bh->b_size = max_blocks << inode->i_blkbits; | |
206 | if (max_blocks) | |
207 | set_buffer_mapped(bh); | |
208 | return 0; | |
209 | } | |
210 | ||
211 | static ssize_t | |
212 | blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, | |
213 | loff_t offset, unsigned long nr_segs) | |
214 | { | |
215 | struct file *file = iocb->ki_filp; | |
216 | struct inode *inode = file->f_mapping->host; | |
217 | ||
218 | return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset, | |
219 | nr_segs, blkdev_get_blocks, NULL, NULL, 0); | |
220 | } | |
221 | ||
222 | int __sync_blockdev(struct block_device *bdev, int wait) | |
223 | { | |
224 | if (!bdev) | |
225 | return 0; | |
226 | if (!wait) | |
227 | return filemap_flush(bdev->bd_inode->i_mapping); | |
228 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
229 | } | |
230 | ||
231 | /* | |
232 | * Write out and wait upon all the dirty data associated with a block | |
233 | * device via its mapping. Does not take the superblock lock. | |
234 | */ | |
235 | int sync_blockdev(struct block_device *bdev) | |
236 | { | |
237 | return __sync_blockdev(bdev, 1); | |
238 | } | |
239 | EXPORT_SYMBOL(sync_blockdev); | |
240 | ||
241 | /* | |
242 | * Write out and wait upon all dirty data associated with this | |
243 | * device. Filesystem data as well as the underlying block | |
244 | * device. Takes the superblock lock. | |
245 | */ | |
246 | int fsync_bdev(struct block_device *bdev) | |
247 | { | |
248 | struct super_block *sb = get_super(bdev); | |
249 | if (sb) { | |
250 | int res = sync_filesystem(sb); | |
251 | drop_super(sb); | |
252 | return res; | |
253 | } | |
254 | return sync_blockdev(bdev); | |
255 | } | |
256 | EXPORT_SYMBOL(fsync_bdev); | |
257 | ||
258 | /** | |
259 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
260 | * @bdev: blockdevice to lock | |
261 | * | |
262 | * If a superblock is found on this device, we take the s_umount semaphore | |
263 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
264 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
265 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
266 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
267 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
268 | * actually. | |
269 | */ | |
270 | struct super_block *freeze_bdev(struct block_device *bdev) | |
271 | { | |
272 | struct super_block *sb; | |
273 | int error = 0; | |
274 | ||
275 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
276 | if (++bdev->bd_fsfreeze_count > 1) { | |
277 | /* | |
278 | * We don't even need to grab a reference - the first call | |
279 | * to freeze_bdev grab an active reference and only the last | |
280 | * thaw_bdev drops it. | |
281 | */ | |
282 | sb = get_super(bdev); | |
283 | drop_super(sb); | |
284 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
285 | return sb; | |
286 | } | |
287 | ||
288 | sb = get_active_super(bdev); | |
289 | if (!sb) | |
290 | goto out; | |
291 | error = freeze_super(sb); | |
292 | if (error) { | |
293 | deactivate_super(sb); | |
294 | bdev->bd_fsfreeze_count--; | |
295 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
296 | return ERR_PTR(error); | |
297 | } | |
298 | deactivate_super(sb); | |
299 | out: | |
300 | sync_blockdev(bdev); | |
301 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
302 | return sb; /* thaw_bdev releases s->s_umount */ | |
303 | } | |
304 | EXPORT_SYMBOL(freeze_bdev); | |
305 | ||
306 | /** | |
307 | * thaw_bdev -- unlock filesystem | |
308 | * @bdev: blockdevice to unlock | |
309 | * @sb: associated superblock | |
310 | * | |
311 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
312 | */ | |
313 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
314 | { | |
315 | int error = -EINVAL; | |
316 | ||
317 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
318 | if (!bdev->bd_fsfreeze_count) | |
319 | goto out; | |
320 | ||
321 | error = 0; | |
322 | if (--bdev->bd_fsfreeze_count > 0) | |
323 | goto out; | |
324 | ||
325 | if (!sb) | |
326 | goto out; | |
327 | ||
328 | error = thaw_super(sb); | |
329 | if (error) { | |
330 | bdev->bd_fsfreeze_count++; | |
331 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
332 | return error; | |
333 | } | |
334 | out: | |
335 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
336 | return 0; | |
337 | } | |
338 | EXPORT_SYMBOL(thaw_bdev); | |
339 | ||
340 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) | |
341 | { | |
342 | return block_write_full_page(page, blkdev_get_block, wbc); | |
343 | } | |
344 | ||
345 | static int blkdev_readpage(struct file * file, struct page * page) | |
346 | { | |
347 | return block_read_full_page(page, blkdev_get_block); | |
348 | } | |
349 | ||
350 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, | |
351 | loff_t pos, unsigned len, unsigned flags, | |
352 | struct page **pagep, void **fsdata) | |
353 | { | |
354 | return block_write_begin(mapping, pos, len, flags, pagep, | |
355 | blkdev_get_block); | |
356 | } | |
357 | ||
358 | static int blkdev_write_end(struct file *file, struct address_space *mapping, | |
359 | loff_t pos, unsigned len, unsigned copied, | |
360 | struct page *page, void *fsdata) | |
361 | { | |
362 | int ret; | |
363 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
364 | ||
365 | unlock_page(page); | |
366 | page_cache_release(page); | |
367 | ||
368 | return ret; | |
369 | } | |
370 | ||
371 | /* | |
372 | * private llseek: | |
373 | * for a block special file file->f_path.dentry->d_inode->i_size is zero | |
374 | * so we compute the size by hand (just as in block_read/write above) | |
375 | */ | |
376 | static loff_t block_llseek(struct file *file, loff_t offset, int origin) | |
377 | { | |
378 | struct inode *bd_inode = file->f_mapping->host; | |
379 | loff_t size; | |
380 | loff_t retval; | |
381 | ||
382 | mutex_lock(&bd_inode->i_mutex); | |
383 | size = i_size_read(bd_inode); | |
384 | ||
385 | retval = -EINVAL; | |
386 | switch (origin) { | |
387 | case SEEK_END: | |
388 | offset += size; | |
389 | break; | |
390 | case SEEK_CUR: | |
391 | offset += file->f_pos; | |
392 | case SEEK_SET: | |
393 | break; | |
394 | default: | |
395 | goto out; | |
396 | } | |
397 | if (offset >= 0 && offset <= size) { | |
398 | if (offset != file->f_pos) { | |
399 | file->f_pos = offset; | |
400 | } | |
401 | retval = offset; | |
402 | } | |
403 | out: | |
404 | mutex_unlock(&bd_inode->i_mutex); | |
405 | return retval; | |
406 | } | |
407 | ||
408 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) | |
409 | { | |
410 | struct inode *bd_inode = filp->f_mapping->host; | |
411 | struct block_device *bdev = I_BDEV(bd_inode); | |
412 | int error; | |
413 | ||
414 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
415 | if (error) | |
416 | return error; | |
417 | ||
418 | /* | |
419 | * There is no need to serialise calls to blkdev_issue_flush with | |
420 | * i_mutex and doing so causes performance issues with concurrent | |
421 | * O_SYNC writers to a block device. | |
422 | */ | |
423 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); | |
424 | if (error == -EOPNOTSUPP) | |
425 | error = 0; | |
426 | ||
427 | return error; | |
428 | } | |
429 | EXPORT_SYMBOL(blkdev_fsync); | |
430 | ||
431 | /* | |
432 | * pseudo-fs | |
433 | */ | |
434 | ||
435 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
436 | static struct kmem_cache * bdev_cachep __read_mostly; | |
437 | ||
438 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
439 | { | |
440 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); | |
441 | if (!ei) | |
442 | return NULL; | |
443 | return &ei->vfs_inode; | |
444 | } | |
445 | ||
446 | static void bdev_i_callback(struct rcu_head *head) | |
447 | { | |
448 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
449 | struct bdev_inode *bdi = BDEV_I(inode); | |
450 | ||
451 | kmem_cache_free(bdev_cachep, bdi); | |
452 | } | |
453 | ||
454 | static void bdev_destroy_inode(struct inode *inode) | |
455 | { | |
456 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
457 | } | |
458 | ||
459 | static void init_once(void *foo) | |
460 | { | |
461 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
462 | struct block_device *bdev = &ei->bdev; | |
463 | ||
464 | memset(bdev, 0, sizeof(*bdev)); | |
465 | mutex_init(&bdev->bd_mutex); | |
466 | INIT_LIST_HEAD(&bdev->bd_inodes); | |
467 | INIT_LIST_HEAD(&bdev->bd_list); | |
468 | #ifdef CONFIG_SYSFS | |
469 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
470 | #endif | |
471 | inode_init_once(&ei->vfs_inode); | |
472 | /* Initialize mutex for freeze. */ | |
473 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
474 | } | |
475 | ||
476 | static inline void __bd_forget(struct inode *inode) | |
477 | { | |
478 | list_del_init(&inode->i_devices); | |
479 | inode->i_bdev = NULL; | |
480 | inode->i_mapping = &inode->i_data; | |
481 | } | |
482 | ||
483 | static void bdev_evict_inode(struct inode *inode) | |
484 | { | |
485 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
486 | struct list_head *p; | |
487 | truncate_inode_pages(&inode->i_data, 0); | |
488 | invalidate_inode_buffers(inode); /* is it needed here? */ | |
489 | end_writeback(inode); | |
490 | spin_lock(&bdev_lock); | |
491 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
492 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
493 | } | |
494 | list_del_init(&bdev->bd_list); | |
495 | spin_unlock(&bdev_lock); | |
496 | } | |
497 | ||
498 | static const struct super_operations bdev_sops = { | |
499 | .statfs = simple_statfs, | |
500 | .alloc_inode = bdev_alloc_inode, | |
501 | .destroy_inode = bdev_destroy_inode, | |
502 | .drop_inode = generic_delete_inode, | |
503 | .evict_inode = bdev_evict_inode, | |
504 | }; | |
505 | ||
506 | static struct dentry *bd_mount(struct file_system_type *fs_type, | |
507 | int flags, const char *dev_name, void *data) | |
508 | { | |
509 | return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576); | |
510 | } | |
511 | ||
512 | static struct file_system_type bd_type = { | |
513 | .name = "bdev", | |
514 | .mount = bd_mount, | |
515 | .kill_sb = kill_anon_super, | |
516 | }; | |
517 | ||
518 | static struct super_block *blockdev_superblock __read_mostly; | |
519 | ||
520 | void __init bdev_cache_init(void) | |
521 | { | |
522 | int err; | |
523 | static struct vfsmount *bd_mnt; | |
524 | ||
525 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), | |
526 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| | |
527 | SLAB_MEM_SPREAD|SLAB_PANIC), | |
528 | init_once); | |
529 | err = register_filesystem(&bd_type); | |
530 | if (err) | |
531 | panic("Cannot register bdev pseudo-fs"); | |
532 | bd_mnt = kern_mount(&bd_type); | |
533 | if (IS_ERR(bd_mnt)) | |
534 | panic("Cannot create bdev pseudo-fs"); | |
535 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ | |
536 | } | |
537 | ||
538 | /* | |
539 | * Most likely _very_ bad one - but then it's hardly critical for small | |
540 | * /dev and can be fixed when somebody will need really large one. | |
541 | * Keep in mind that it will be fed through icache hash function too. | |
542 | */ | |
543 | static inline unsigned long hash(dev_t dev) | |
544 | { | |
545 | return MAJOR(dev)+MINOR(dev); | |
546 | } | |
547 | ||
548 | static int bdev_test(struct inode *inode, void *data) | |
549 | { | |
550 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
551 | } | |
552 | ||
553 | static int bdev_set(struct inode *inode, void *data) | |
554 | { | |
555 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
556 | return 0; | |
557 | } | |
558 | ||
559 | static LIST_HEAD(all_bdevs); | |
560 | ||
561 | struct block_device *bdget(dev_t dev) | |
562 | { | |
563 | struct block_device *bdev; | |
564 | struct inode *inode; | |
565 | ||
566 | inode = iget5_locked(blockdev_superblock, hash(dev), | |
567 | bdev_test, bdev_set, &dev); | |
568 | ||
569 | if (!inode) | |
570 | return NULL; | |
571 | ||
572 | bdev = &BDEV_I(inode)->bdev; | |
573 | ||
574 | if (inode->i_state & I_NEW) { | |
575 | bdev->bd_contains = NULL; | |
576 | bdev->bd_super = NULL; | |
577 | bdev->bd_inode = inode; | |
578 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
579 | bdev->bd_part_count = 0; | |
580 | bdev->bd_invalidated = 0; | |
581 | inode->i_mode = S_IFBLK; | |
582 | inode->i_rdev = dev; | |
583 | inode->i_bdev = bdev; | |
584 | inode->i_data.a_ops = &def_blk_aops; | |
585 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
586 | inode->i_data.backing_dev_info = &default_backing_dev_info; | |
587 | spin_lock(&bdev_lock); | |
588 | list_add(&bdev->bd_list, &all_bdevs); | |
589 | spin_unlock(&bdev_lock); | |
590 | unlock_new_inode(inode); | |
591 | } | |
592 | return bdev; | |
593 | } | |
594 | ||
595 | EXPORT_SYMBOL(bdget); | |
596 | ||
597 | /** | |
598 | * bdgrab -- Grab a reference to an already referenced block device | |
599 | * @bdev: Block device to grab a reference to. | |
600 | */ | |
601 | struct block_device *bdgrab(struct block_device *bdev) | |
602 | { | |
603 | ihold(bdev->bd_inode); | |
604 | return bdev; | |
605 | } | |
606 | ||
607 | long nr_blockdev_pages(void) | |
608 | { | |
609 | struct block_device *bdev; | |
610 | long ret = 0; | |
611 | spin_lock(&bdev_lock); | |
612 | list_for_each_entry(bdev, &all_bdevs, bd_list) { | |
613 | ret += bdev->bd_inode->i_mapping->nrpages; | |
614 | } | |
615 | spin_unlock(&bdev_lock); | |
616 | return ret; | |
617 | } | |
618 | ||
619 | void bdput(struct block_device *bdev) | |
620 | { | |
621 | iput(bdev->bd_inode); | |
622 | } | |
623 | ||
624 | EXPORT_SYMBOL(bdput); | |
625 | ||
626 | static struct block_device *bd_acquire(struct inode *inode) | |
627 | { | |
628 | struct block_device *bdev; | |
629 | ||
630 | spin_lock(&bdev_lock); | |
631 | bdev = inode->i_bdev; | |
632 | if (bdev) { | |
633 | ihold(bdev->bd_inode); | |
634 | spin_unlock(&bdev_lock); | |
635 | return bdev; | |
636 | } | |
637 | spin_unlock(&bdev_lock); | |
638 | ||
639 | bdev = bdget(inode->i_rdev); | |
640 | if (bdev) { | |
641 | spin_lock(&bdev_lock); | |
642 | if (!inode->i_bdev) { | |
643 | /* | |
644 | * We take an additional reference to bd_inode, | |
645 | * and it's released in clear_inode() of inode. | |
646 | * So, we can access it via ->i_mapping always | |
647 | * without igrab(). | |
648 | */ | |
649 | ihold(bdev->bd_inode); | |
650 | inode->i_bdev = bdev; | |
651 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
652 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
653 | } | |
654 | spin_unlock(&bdev_lock); | |
655 | } | |
656 | return bdev; | |
657 | } | |
658 | ||
659 | static inline int sb_is_blkdev_sb(struct super_block *sb) | |
660 | { | |
661 | return sb == blockdev_superblock; | |
662 | } | |
663 | ||
664 | /* Call when you free inode */ | |
665 | ||
666 | void bd_forget(struct inode *inode) | |
667 | { | |
668 | struct block_device *bdev = NULL; | |
669 | ||
670 | spin_lock(&bdev_lock); | |
671 | if (inode->i_bdev) { | |
672 | if (!sb_is_blkdev_sb(inode->i_sb)) | |
673 | bdev = inode->i_bdev; | |
674 | __bd_forget(inode); | |
675 | } | |
676 | spin_unlock(&bdev_lock); | |
677 | ||
678 | if (bdev) | |
679 | iput(bdev->bd_inode); | |
680 | } | |
681 | ||
682 | /** | |
683 | * bd_may_claim - test whether a block device can be claimed | |
684 | * @bdev: block device of interest | |
685 | * @whole: whole block device containing @bdev, may equal @bdev | |
686 | * @holder: holder trying to claim @bdev | |
687 | * | |
688 | * Test whether @bdev can be claimed by @holder. | |
689 | * | |
690 | * CONTEXT: | |
691 | * spin_lock(&bdev_lock). | |
692 | * | |
693 | * RETURNS: | |
694 | * %true if @bdev can be claimed, %false otherwise. | |
695 | */ | |
696 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
697 | void *holder) | |
698 | { | |
699 | if (bdev->bd_holder == holder) | |
700 | return true; /* already a holder */ | |
701 | else if (bdev->bd_holder != NULL) | |
702 | return false; /* held by someone else */ | |
703 | else if (bdev->bd_contains == bdev) | |
704 | return true; /* is a whole device which isn't held */ | |
705 | ||
706 | else if (whole->bd_holder == bd_may_claim) | |
707 | return true; /* is a partition of a device that is being partitioned */ | |
708 | else if (whole->bd_holder != NULL) | |
709 | return false; /* is a partition of a held device */ | |
710 | else | |
711 | return true; /* is a partition of an un-held device */ | |
712 | } | |
713 | ||
714 | /** | |
715 | * bd_prepare_to_claim - prepare to claim a block device | |
716 | * @bdev: block device of interest | |
717 | * @whole: the whole device containing @bdev, may equal @bdev | |
718 | * @holder: holder trying to claim @bdev | |
719 | * | |
720 | * Prepare to claim @bdev. This function fails if @bdev is already | |
721 | * claimed by another holder and waits if another claiming is in | |
722 | * progress. This function doesn't actually claim. On successful | |
723 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
724 | * | |
725 | * CONTEXT: | |
726 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
727 | * it multiple times. | |
728 | * | |
729 | * RETURNS: | |
730 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
731 | */ | |
732 | static int bd_prepare_to_claim(struct block_device *bdev, | |
733 | struct block_device *whole, void *holder) | |
734 | { | |
735 | retry: | |
736 | /* if someone else claimed, fail */ | |
737 | if (!bd_may_claim(bdev, whole, holder)) | |
738 | return -EBUSY; | |
739 | ||
740 | /* if claiming is already in progress, wait for it to finish */ | |
741 | if (whole->bd_claiming) { | |
742 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); | |
743 | DEFINE_WAIT(wait); | |
744 | ||
745 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
746 | spin_unlock(&bdev_lock); | |
747 | schedule(); | |
748 | finish_wait(wq, &wait); | |
749 | spin_lock(&bdev_lock); | |
750 | goto retry; | |
751 | } | |
752 | ||
753 | /* yay, all mine */ | |
754 | return 0; | |
755 | } | |
756 | ||
757 | /** | |
758 | * bd_start_claiming - start claiming a block device | |
759 | * @bdev: block device of interest | |
760 | * @holder: holder trying to claim @bdev | |
761 | * | |
762 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
763 | * exclusively and mark that an exclusive open is in progress. Each | |
764 | * successful call to this function must be matched with a call to | |
765 | * either bd_finish_claiming() or bd_abort_claiming() (which do not | |
766 | * fail). | |
767 | * | |
768 | * This function is used to gain exclusive access to the block device | |
769 | * without actually causing other exclusive open attempts to fail. It | |
770 | * should be used when the open sequence itself requires exclusive | |
771 | * access but may subsequently fail. | |
772 | * | |
773 | * CONTEXT: | |
774 | * Might sleep. | |
775 | * | |
776 | * RETURNS: | |
777 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
778 | * value on failure. | |
779 | */ | |
780 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
781 | void *holder) | |
782 | { | |
783 | struct gendisk *disk; | |
784 | struct block_device *whole; | |
785 | int partno, err; | |
786 | ||
787 | might_sleep(); | |
788 | ||
789 | /* | |
790 | * @bdev might not have been initialized properly yet, look up | |
791 | * and grab the outer block device the hard way. | |
792 | */ | |
793 | disk = get_gendisk(bdev->bd_dev, &partno); | |
794 | if (!disk) | |
795 | return ERR_PTR(-ENXIO); | |
796 | ||
797 | /* | |
798 | * Normally, @bdev should equal what's returned from bdget_disk() | |
799 | * if partno is 0; however, some drivers (floppy) use multiple | |
800 | * bdev's for the same physical device and @bdev may be one of the | |
801 | * aliases. Keep @bdev if partno is 0. This means claimer | |
802 | * tracking is broken for those devices but it has always been that | |
803 | * way. | |
804 | */ | |
805 | if (partno) | |
806 | whole = bdget_disk(disk, 0); | |
807 | else | |
808 | whole = bdgrab(bdev); | |
809 | ||
810 | module_put(disk->fops->owner); | |
811 | put_disk(disk); | |
812 | if (!whole) | |
813 | return ERR_PTR(-ENOMEM); | |
814 | ||
815 | /* prepare to claim, if successful, mark claiming in progress */ | |
816 | spin_lock(&bdev_lock); | |
817 | ||
818 | err = bd_prepare_to_claim(bdev, whole, holder); | |
819 | if (err == 0) { | |
820 | whole->bd_claiming = holder; | |
821 | spin_unlock(&bdev_lock); | |
822 | return whole; | |
823 | } else { | |
824 | spin_unlock(&bdev_lock); | |
825 | bdput(whole); | |
826 | return ERR_PTR(err); | |
827 | } | |
828 | } | |
829 | ||
830 | #ifdef CONFIG_SYSFS | |
831 | struct bd_holder_disk { | |
832 | struct list_head list; | |
833 | struct gendisk *disk; | |
834 | int refcnt; | |
835 | }; | |
836 | ||
837 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
838 | struct gendisk *disk) | |
839 | { | |
840 | struct bd_holder_disk *holder; | |
841 | ||
842 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
843 | if (holder->disk == disk) | |
844 | return holder; | |
845 | return NULL; | |
846 | } | |
847 | ||
848 | static int add_symlink(struct kobject *from, struct kobject *to) | |
849 | { | |
850 | return sysfs_create_link(from, to, kobject_name(to)); | |
851 | } | |
852 | ||
853 | static void del_symlink(struct kobject *from, struct kobject *to) | |
854 | { | |
855 | sysfs_remove_link(from, kobject_name(to)); | |
856 | } | |
857 | ||
858 | /** | |
859 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev | |
860 | * @bdev: the claimed slave bdev | |
861 | * @disk: the holding disk | |
862 | * | |
863 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
864 | * | |
865 | * This functions creates the following sysfs symlinks. | |
866 | * | |
867 | * - from "slaves" directory of the holder @disk to the claimed @bdev | |
868 | * - from "holders" directory of the @bdev to the holder @disk | |
869 | * | |
870 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is | |
871 | * passed to bd_link_disk_holder(), then: | |
872 | * | |
873 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda | |
874 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
875 | * | |
876 | * The caller must have claimed @bdev before calling this function and | |
877 | * ensure that both @bdev and @disk are valid during the creation and | |
878 | * lifetime of these symlinks. | |
879 | * | |
880 | * CONTEXT: | |
881 | * Might sleep. | |
882 | * | |
883 | * RETURNS: | |
884 | * 0 on success, -errno on failure. | |
885 | */ | |
886 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
887 | { | |
888 | struct bd_holder_disk *holder; | |
889 | int ret = 0; | |
890 | ||
891 | mutex_lock(&bdev->bd_mutex); | |
892 | ||
893 | WARN_ON_ONCE(!bdev->bd_holder); | |
894 | ||
895 | /* FIXME: remove the following once add_disk() handles errors */ | |
896 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
897 | goto out_unlock; | |
898 | ||
899 | holder = bd_find_holder_disk(bdev, disk); | |
900 | if (holder) { | |
901 | holder->refcnt++; | |
902 | goto out_unlock; | |
903 | } | |
904 | ||
905 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); | |
906 | if (!holder) { | |
907 | ret = -ENOMEM; | |
908 | goto out_unlock; | |
909 | } | |
910 | ||
911 | INIT_LIST_HEAD(&holder->list); | |
912 | holder->disk = disk; | |
913 | holder->refcnt = 1; | |
914 | ||
915 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
916 | if (ret) | |
917 | goto out_free; | |
918 | ||
919 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
920 | if (ret) | |
921 | goto out_del; | |
922 | /* | |
923 | * bdev could be deleted beneath us which would implicitly destroy | |
924 | * the holder directory. Hold on to it. | |
925 | */ | |
926 | kobject_get(bdev->bd_part->holder_dir); | |
927 | ||
928 | list_add(&holder->list, &bdev->bd_holder_disks); | |
929 | goto out_unlock; | |
930 | ||
931 | out_del: | |
932 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
933 | out_free: | |
934 | kfree(holder); | |
935 | out_unlock: | |
936 | mutex_unlock(&bdev->bd_mutex); | |
937 | return ret; | |
938 | } | |
939 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); | |
940 | ||
941 | /** | |
942 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
943 | * @bdev: the calimed slave bdev | |
944 | * @disk: the holding disk | |
945 | * | |
946 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
947 | * | |
948 | * CONTEXT: | |
949 | * Might sleep. | |
950 | */ | |
951 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
952 | { | |
953 | struct bd_holder_disk *holder; | |
954 | ||
955 | mutex_lock(&bdev->bd_mutex); | |
956 | ||
957 | holder = bd_find_holder_disk(bdev, disk); | |
958 | ||
959 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
960 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
961 | del_symlink(bdev->bd_part->holder_dir, | |
962 | &disk_to_dev(disk)->kobj); | |
963 | kobject_put(bdev->bd_part->holder_dir); | |
964 | list_del_init(&holder->list); | |
965 | kfree(holder); | |
966 | } | |
967 | ||
968 | mutex_unlock(&bdev->bd_mutex); | |
969 | } | |
970 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); | |
971 | #endif | |
972 | ||
973 | /** | |
974 | * flush_disk - invalidates all buffer-cache entries on a disk | |
975 | * | |
976 | * @bdev: struct block device to be flushed | |
977 | * @kill_dirty: flag to guide handling of dirty inodes | |
978 | * | |
979 | * Invalidates all buffer-cache entries on a disk. It should be called | |
980 | * when a disk has been changed -- either by a media change or online | |
981 | * resize. | |
982 | */ | |
983 | static void flush_disk(struct block_device *bdev, bool kill_dirty) | |
984 | { | |
985 | if (__invalidate_device(bdev, kill_dirty)) { | |
986 | char name[BDEVNAME_SIZE] = ""; | |
987 | ||
988 | if (bdev->bd_disk) | |
989 | disk_name(bdev->bd_disk, 0, name); | |
990 | printk(KERN_WARNING "VFS: busy inodes on changed media or " | |
991 | "resized disk %s\n", name); | |
992 | } | |
993 | ||
994 | if (!bdev->bd_disk) | |
995 | return; | |
996 | if (disk_part_scan_enabled(bdev->bd_disk)) | |
997 | bdev->bd_invalidated = 1; | |
998 | } | |
999 | ||
1000 | /** | |
1001 | * check_disk_size_change - checks for disk size change and adjusts bdev size. | |
1002 | * @disk: struct gendisk to check | |
1003 | * @bdev: struct bdev to adjust. | |
1004 | * | |
1005 | * This routine checks to see if the bdev size does not match the disk size | |
1006 | * and adjusts it if it differs. | |
1007 | */ | |
1008 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1009 | { | |
1010 | loff_t disk_size, bdev_size; | |
1011 | ||
1012 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1013 | bdev_size = i_size_read(bdev->bd_inode); | |
1014 | if (disk_size != bdev_size) { | |
1015 | char name[BDEVNAME_SIZE]; | |
1016 | ||
1017 | disk_name(disk, 0, name); | |
1018 | printk(KERN_INFO | |
1019 | "%s: detected capacity change from %lld to %lld\n", | |
1020 | name, bdev_size, disk_size); | |
1021 | i_size_write(bdev->bd_inode, disk_size); | |
1022 | flush_disk(bdev, false); | |
1023 | } | |
1024 | } | |
1025 | EXPORT_SYMBOL(check_disk_size_change); | |
1026 | ||
1027 | /** | |
1028 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back | |
1029 | * @disk: struct gendisk to be revalidated | |
1030 | * | |
1031 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1032 | * call-backs. It is used to do common pre and post operations needed | |
1033 | * for all revalidate_disk operations. | |
1034 | */ | |
1035 | int revalidate_disk(struct gendisk *disk) | |
1036 | { | |
1037 | struct block_device *bdev; | |
1038 | int ret = 0; | |
1039 | ||
1040 | if (disk->fops->revalidate_disk) | |
1041 | ret = disk->fops->revalidate_disk(disk); | |
1042 | ||
1043 | bdev = bdget_disk(disk, 0); | |
1044 | if (!bdev) | |
1045 | return ret; | |
1046 | ||
1047 | mutex_lock(&bdev->bd_mutex); | |
1048 | check_disk_size_change(disk, bdev); | |
1049 | mutex_unlock(&bdev->bd_mutex); | |
1050 | bdput(bdev); | |
1051 | return ret; | |
1052 | } | |
1053 | EXPORT_SYMBOL(revalidate_disk); | |
1054 | ||
1055 | /* | |
1056 | * This routine checks whether a removable media has been changed, | |
1057 | * and invalidates all buffer-cache-entries in that case. This | |
1058 | * is a relatively slow routine, so we have to try to minimize using | |
1059 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1060 | * is the best way of combining speed and utility, I think. | |
1061 | * People changing diskettes in the middle of an operation deserve | |
1062 | * to lose :-) | |
1063 | */ | |
1064 | int check_disk_change(struct block_device *bdev) | |
1065 | { | |
1066 | struct gendisk *disk = bdev->bd_disk; | |
1067 | const struct block_device_operations *bdops = disk->fops; | |
1068 | unsigned int events; | |
1069 | ||
1070 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | | |
1071 | DISK_EVENT_EJECT_REQUEST); | |
1072 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1073 | return 0; | |
1074 | ||
1075 | flush_disk(bdev, true); | |
1076 | if (bdops->revalidate_disk) | |
1077 | bdops->revalidate_disk(bdev->bd_disk); | |
1078 | return 1; | |
1079 | } | |
1080 | ||
1081 | EXPORT_SYMBOL(check_disk_change); | |
1082 | ||
1083 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1084 | { | |
1085 | unsigned bsize = bdev_logical_block_size(bdev); | |
1086 | ||
1087 | bdev->bd_inode->i_size = size; | |
1088 | while (bsize < PAGE_CACHE_SIZE) { | |
1089 | if (size & bsize) | |
1090 | break; | |
1091 | bsize <<= 1; | |
1092 | } | |
1093 | bdev->bd_block_size = bsize; | |
1094 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1095 | } | |
1096 | EXPORT_SYMBOL(bd_set_size); | |
1097 | ||
1098 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); | |
1099 | ||
1100 | /* | |
1101 | * bd_mutex locking: | |
1102 | * | |
1103 | * mutex_lock(part->bd_mutex) | |
1104 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1105 | */ | |
1106 | ||
1107 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) | |
1108 | { | |
1109 | struct gendisk *disk; | |
1110 | struct module *owner; | |
1111 | int ret; | |
1112 | int partno; | |
1113 | int perm = 0; | |
1114 | ||
1115 | if (mode & FMODE_READ) | |
1116 | perm |= MAY_READ; | |
1117 | if (mode & FMODE_WRITE) | |
1118 | perm |= MAY_WRITE; | |
1119 | /* | |
1120 | * hooks: /n/, see "layering violations". | |
1121 | */ | |
1122 | if (!for_part) { | |
1123 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1124 | if (ret != 0) { | |
1125 | bdput(bdev); | |
1126 | return ret; | |
1127 | } | |
1128 | } | |
1129 | ||
1130 | restart: | |
1131 | ||
1132 | ret = -ENXIO; | |
1133 | disk = get_gendisk(bdev->bd_dev, &partno); | |
1134 | if (!disk) | |
1135 | goto out; | |
1136 | owner = disk->fops->owner; | |
1137 | ||
1138 | disk_block_events(disk); | |
1139 | mutex_lock_nested(&bdev->bd_mutex, for_part); | |
1140 | if (!bdev->bd_openers) { | |
1141 | bdev->bd_disk = disk; | |
1142 | bdev->bd_contains = bdev; | |
1143 | if (!partno) { | |
1144 | struct backing_dev_info *bdi; | |
1145 | ||
1146 | ret = -ENXIO; | |
1147 | bdev->bd_part = disk_get_part(disk, partno); | |
1148 | if (!bdev->bd_part) | |
1149 | goto out_clear; | |
1150 | ||
1151 | ret = 0; | |
1152 | if (disk->fops->open) { | |
1153 | ret = disk->fops->open(bdev, mode); | |
1154 | if (ret == -ERESTARTSYS) { | |
1155 | /* Lost a race with 'disk' being | |
1156 | * deleted, try again. | |
1157 | * See md.c | |
1158 | */ | |
1159 | disk_put_part(bdev->bd_part); | |
1160 | bdev->bd_part = NULL; | |
1161 | bdev->bd_disk = NULL; | |
1162 | mutex_unlock(&bdev->bd_mutex); | |
1163 | disk_unblock_events(disk); | |
1164 | put_disk(disk); | |
1165 | module_put(owner); | |
1166 | goto restart; | |
1167 | } | |
1168 | } | |
1169 | ||
1170 | if (!ret && !bdev->bd_openers) { | |
1171 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); | |
1172 | bdi = blk_get_backing_dev_info(bdev); | |
1173 | if (bdi == NULL) | |
1174 | bdi = &default_backing_dev_info; | |
1175 | bdev_inode_switch_bdi(bdev->bd_inode, bdi); | |
1176 | } | |
1177 | ||
1178 | /* | |
1179 | * If the device is invalidated, rescan partition | |
1180 | * if open succeeded or failed with -ENOMEDIUM. | |
1181 | * The latter is necessary to prevent ghost | |
1182 | * partitions on a removed medium. | |
1183 | */ | |
1184 | if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM)) | |
1185 | rescan_partitions(disk, bdev); | |
1186 | if (ret) | |
1187 | goto out_clear; | |
1188 | } else { | |
1189 | struct block_device *whole; | |
1190 | whole = bdget_disk(disk, 0); | |
1191 | ret = -ENOMEM; | |
1192 | if (!whole) | |
1193 | goto out_clear; | |
1194 | BUG_ON(for_part); | |
1195 | ret = __blkdev_get(whole, mode, 1); | |
1196 | if (ret) | |
1197 | goto out_clear; | |
1198 | bdev->bd_contains = whole; | |
1199 | bdev_inode_switch_bdi(bdev->bd_inode, | |
1200 | whole->bd_inode->i_data.backing_dev_info); | |
1201 | bdev->bd_part = disk_get_part(disk, partno); | |
1202 | if (!(disk->flags & GENHD_FL_UP) || | |
1203 | !bdev->bd_part || !bdev->bd_part->nr_sects) { | |
1204 | ret = -ENXIO; | |
1205 | goto out_clear; | |
1206 | } | |
1207 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); | |
1208 | } | |
1209 | } else { | |
1210 | if (bdev->bd_contains == bdev) { | |
1211 | ret = 0; | |
1212 | if (bdev->bd_disk->fops->open) | |
1213 | ret = bdev->bd_disk->fops->open(bdev, mode); | |
1214 | /* the same as first opener case, read comment there */ | |
1215 | if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM)) | |
1216 | rescan_partitions(bdev->bd_disk, bdev); | |
1217 | if (ret) | |
1218 | goto out_unlock_bdev; | |
1219 | } | |
1220 | /* only one opener holds refs to the module and disk */ | |
1221 | put_disk(disk); | |
1222 | module_put(owner); | |
1223 | } | |
1224 | bdev->bd_openers++; | |
1225 | if (for_part) | |
1226 | bdev->bd_part_count++; | |
1227 | mutex_unlock(&bdev->bd_mutex); | |
1228 | disk_unblock_events(disk); | |
1229 | return 0; | |
1230 | ||
1231 | out_clear: | |
1232 | disk_put_part(bdev->bd_part); | |
1233 | bdev->bd_disk = NULL; | |
1234 | bdev->bd_part = NULL; | |
1235 | bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info); | |
1236 | if (bdev != bdev->bd_contains) | |
1237 | __blkdev_put(bdev->bd_contains, mode, 1); | |
1238 | bdev->bd_contains = NULL; | |
1239 | out_unlock_bdev: | |
1240 | mutex_unlock(&bdev->bd_mutex); | |
1241 | disk_unblock_events(disk); | |
1242 | put_disk(disk); | |
1243 | module_put(owner); | |
1244 | out: | |
1245 | bdput(bdev); | |
1246 | ||
1247 | return ret; | |
1248 | } | |
1249 | ||
1250 | /** | |
1251 | * blkdev_get - open a block device | |
1252 | * @bdev: block_device to open | |
1253 | * @mode: FMODE_* mask | |
1254 | * @holder: exclusive holder identifier | |
1255 | * | |
1256 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1257 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1258 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1259 | * | |
1260 | * On success, the reference count of @bdev is unchanged. On failure, | |
1261 | * @bdev is put. | |
1262 | * | |
1263 | * CONTEXT: | |
1264 | * Might sleep. | |
1265 | * | |
1266 | * RETURNS: | |
1267 | * 0 on success, -errno on failure. | |
1268 | */ | |
1269 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) | |
1270 | { | |
1271 | struct block_device *whole = NULL; | |
1272 | int res; | |
1273 | ||
1274 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1275 | ||
1276 | if ((mode & FMODE_EXCL) && holder) { | |
1277 | whole = bd_start_claiming(bdev, holder); | |
1278 | if (IS_ERR(whole)) { | |
1279 | bdput(bdev); | |
1280 | return PTR_ERR(whole); | |
1281 | } | |
1282 | } | |
1283 | ||
1284 | res = __blkdev_get(bdev, mode, 0); | |
1285 | ||
1286 | if (whole) { | |
1287 | struct gendisk *disk = whole->bd_disk; | |
1288 | ||
1289 | /* finish claiming */ | |
1290 | mutex_lock(&bdev->bd_mutex); | |
1291 | spin_lock(&bdev_lock); | |
1292 | ||
1293 | if (!res) { | |
1294 | BUG_ON(!bd_may_claim(bdev, whole, holder)); | |
1295 | /* | |
1296 | * Note that for a whole device bd_holders | |
1297 | * will be incremented twice, and bd_holder | |
1298 | * will be set to bd_may_claim before being | |
1299 | * set to holder | |
1300 | */ | |
1301 | whole->bd_holders++; | |
1302 | whole->bd_holder = bd_may_claim; | |
1303 | bdev->bd_holders++; | |
1304 | bdev->bd_holder = holder; | |
1305 | } | |
1306 | ||
1307 | /* tell others that we're done */ | |
1308 | BUG_ON(whole->bd_claiming != holder); | |
1309 | whole->bd_claiming = NULL; | |
1310 | wake_up_bit(&whole->bd_claiming, 0); | |
1311 | ||
1312 | spin_unlock(&bdev_lock); | |
1313 | ||
1314 | /* | |
1315 | * Block event polling for write claims if requested. Any | |
1316 | * write holder makes the write_holder state stick until | |
1317 | * all are released. This is good enough and tracking | |
1318 | * individual writeable reference is too fragile given the | |
1319 | * way @mode is used in blkdev_get/put(). | |
1320 | */ | |
1321 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && | |
1322 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
1323 | bdev->bd_write_holder = true; | |
1324 | disk_block_events(disk); | |
1325 | } | |
1326 | ||
1327 | mutex_unlock(&bdev->bd_mutex); | |
1328 | bdput(whole); | |
1329 | } | |
1330 | ||
1331 | return res; | |
1332 | } | |
1333 | EXPORT_SYMBOL(blkdev_get); | |
1334 | ||
1335 | /** | |
1336 | * blkdev_get_by_path - open a block device by name | |
1337 | * @path: path to the block device to open | |
1338 | * @mode: FMODE_* mask | |
1339 | * @holder: exclusive holder identifier | |
1340 | * | |
1341 | * Open the blockdevice described by the device file at @path. @mode | |
1342 | * and @holder are identical to blkdev_get(). | |
1343 | * | |
1344 | * On success, the returned block_device has reference count of one. | |
1345 | * | |
1346 | * CONTEXT: | |
1347 | * Might sleep. | |
1348 | * | |
1349 | * RETURNS: | |
1350 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1351 | */ | |
1352 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1353 | void *holder) | |
1354 | { | |
1355 | struct block_device *bdev; | |
1356 | int err; | |
1357 | ||
1358 | bdev = lookup_bdev(path); | |
1359 | if (IS_ERR(bdev)) | |
1360 | return bdev; | |
1361 | ||
1362 | err = blkdev_get(bdev, mode, holder); | |
1363 | if (err) | |
1364 | return ERR_PTR(err); | |
1365 | ||
1366 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { | |
1367 | blkdev_put(bdev, mode); | |
1368 | return ERR_PTR(-EACCES); | |
1369 | } | |
1370 | ||
1371 | return bdev; | |
1372 | } | |
1373 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1374 | ||
1375 | /** | |
1376 | * blkdev_get_by_dev - open a block device by device number | |
1377 | * @dev: device number of block device to open | |
1378 | * @mode: FMODE_* mask | |
1379 | * @holder: exclusive holder identifier | |
1380 | * | |
1381 | * Open the blockdevice described by device number @dev. @mode and | |
1382 | * @holder are identical to blkdev_get(). | |
1383 | * | |
1384 | * Use it ONLY if you really do not have anything better - i.e. when | |
1385 | * you are behind a truly sucky interface and all you are given is a | |
1386 | * device number. _Never_ to be used for internal purposes. If you | |
1387 | * ever need it - reconsider your API. | |
1388 | * | |
1389 | * On success, the returned block_device has reference count of one. | |
1390 | * | |
1391 | * CONTEXT: | |
1392 | * Might sleep. | |
1393 | * | |
1394 | * RETURNS: | |
1395 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1396 | */ | |
1397 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1398 | { | |
1399 | struct block_device *bdev; | |
1400 | int err; | |
1401 | ||
1402 | bdev = bdget(dev); | |
1403 | if (!bdev) | |
1404 | return ERR_PTR(-ENOMEM); | |
1405 | ||
1406 | err = blkdev_get(bdev, mode, holder); | |
1407 | if (err) | |
1408 | return ERR_PTR(err); | |
1409 | ||
1410 | return bdev; | |
1411 | } | |
1412 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1413 | ||
1414 | static int blkdev_open(struct inode * inode, struct file * filp) | |
1415 | { | |
1416 | struct block_device *bdev; | |
1417 | ||
1418 | /* | |
1419 | * Preserve backwards compatibility and allow large file access | |
1420 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1421 | * binary needs it. We might want to drop this workaround | |
1422 | * during an unstable branch. | |
1423 | */ | |
1424 | filp->f_flags |= O_LARGEFILE; | |
1425 | ||
1426 | if (filp->f_flags & O_NDELAY) | |
1427 | filp->f_mode |= FMODE_NDELAY; | |
1428 | if (filp->f_flags & O_EXCL) | |
1429 | filp->f_mode |= FMODE_EXCL; | |
1430 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1431 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1432 | ||
1433 | bdev = bd_acquire(inode); | |
1434 | if (bdev == NULL) | |
1435 | return -ENOMEM; | |
1436 | ||
1437 | filp->f_mapping = bdev->bd_inode->i_mapping; | |
1438 | ||
1439 | return blkdev_get(bdev, filp->f_mode, filp); | |
1440 | } | |
1441 | ||
1442 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) | |
1443 | { | |
1444 | int ret = 0; | |
1445 | struct gendisk *disk = bdev->bd_disk; | |
1446 | struct block_device *victim = NULL; | |
1447 | ||
1448 | mutex_lock_nested(&bdev->bd_mutex, for_part); | |
1449 | if (for_part) | |
1450 | bdev->bd_part_count--; | |
1451 | ||
1452 | if (!--bdev->bd_openers) { | |
1453 | WARN_ON_ONCE(bdev->bd_holders); | |
1454 | sync_blockdev(bdev); | |
1455 | kill_bdev(bdev); | |
1456 | /* ->release can cause the old bdi to disappear, | |
1457 | * so must switch it out first | |
1458 | */ | |
1459 | bdev_inode_switch_bdi(bdev->bd_inode, | |
1460 | &default_backing_dev_info); | |
1461 | } | |
1462 | if (bdev->bd_contains == bdev) { | |
1463 | if (disk->fops->release) | |
1464 | ret = disk->fops->release(disk, mode); | |
1465 | } | |
1466 | if (!bdev->bd_openers) { | |
1467 | struct module *owner = disk->fops->owner; | |
1468 | ||
1469 | disk_put_part(bdev->bd_part); | |
1470 | bdev->bd_part = NULL; | |
1471 | bdev->bd_disk = NULL; | |
1472 | if (bdev != bdev->bd_contains) | |
1473 | victim = bdev->bd_contains; | |
1474 | bdev->bd_contains = NULL; | |
1475 | ||
1476 | put_disk(disk); | |
1477 | module_put(owner); | |
1478 | } | |
1479 | mutex_unlock(&bdev->bd_mutex); | |
1480 | bdput(bdev); | |
1481 | if (victim) | |
1482 | __blkdev_put(victim, mode, 1); | |
1483 | return ret; | |
1484 | } | |
1485 | ||
1486 | int blkdev_put(struct block_device *bdev, fmode_t mode) | |
1487 | { | |
1488 | mutex_lock(&bdev->bd_mutex); | |
1489 | ||
1490 | if (mode & FMODE_EXCL) { | |
1491 | bool bdev_free; | |
1492 | ||
1493 | /* | |
1494 | * Release a claim on the device. The holder fields | |
1495 | * are protected with bdev_lock. bd_mutex is to | |
1496 | * synchronize disk_holder unlinking. | |
1497 | */ | |
1498 | spin_lock(&bdev_lock); | |
1499 | ||
1500 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1501 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1502 | ||
1503 | /* bd_contains might point to self, check in a separate step */ | |
1504 | if ((bdev_free = !bdev->bd_holders)) | |
1505 | bdev->bd_holder = NULL; | |
1506 | if (!bdev->bd_contains->bd_holders) | |
1507 | bdev->bd_contains->bd_holder = NULL; | |
1508 | ||
1509 | spin_unlock(&bdev_lock); | |
1510 | ||
1511 | /* | |
1512 | * If this was the last claim, remove holder link and | |
1513 | * unblock evpoll if it was a write holder. | |
1514 | */ | |
1515 | if (bdev_free && bdev->bd_write_holder) { | |
1516 | disk_unblock_events(bdev->bd_disk); | |
1517 | bdev->bd_write_holder = false; | |
1518 | } | |
1519 | } | |
1520 | ||
1521 | /* | |
1522 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1523 | * event. This is to ensure detection of media removal commanded | |
1524 | * from userland - e.g. eject(1). | |
1525 | */ | |
1526 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1527 | ||
1528 | mutex_unlock(&bdev->bd_mutex); | |
1529 | ||
1530 | return __blkdev_put(bdev, mode, 0); | |
1531 | } | |
1532 | EXPORT_SYMBOL(blkdev_put); | |
1533 | ||
1534 | static int blkdev_close(struct inode * inode, struct file * filp) | |
1535 | { | |
1536 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); | |
1537 | ||
1538 | return blkdev_put(bdev, filp->f_mode); | |
1539 | } | |
1540 | ||
1541 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) | |
1542 | { | |
1543 | struct block_device *bdev = I_BDEV(file->f_mapping->host); | |
1544 | fmode_t mode = file->f_mode; | |
1545 | ||
1546 | /* | |
1547 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1548 | * to updated it before every ioctl. | |
1549 | */ | |
1550 | if (file->f_flags & O_NDELAY) | |
1551 | mode |= FMODE_NDELAY; | |
1552 | else | |
1553 | mode &= ~FMODE_NDELAY; | |
1554 | ||
1555 | return blkdev_ioctl(bdev, mode, cmd, arg); | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * Write data to the block device. Only intended for the block device itself | |
1560 | * and the raw driver which basically is a fake block device. | |
1561 | * | |
1562 | * Does not take i_mutex for the write and thus is not for general purpose | |
1563 | * use. | |
1564 | */ | |
1565 | ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
1566 | unsigned long nr_segs, loff_t pos) | |
1567 | { | |
1568 | struct file *file = iocb->ki_filp; | |
1569 | ssize_t ret; | |
1570 | ||
1571 | BUG_ON(iocb->ki_pos != pos); | |
1572 | ||
1573 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); | |
1574 | if (ret > 0 || ret == -EIOCBQUEUED) { | |
1575 | ssize_t err; | |
1576 | ||
1577 | err = generic_write_sync(file, pos, ret); | |
1578 | if (err < 0 && ret > 0) | |
1579 | ret = err; | |
1580 | } | |
1581 | return ret; | |
1582 | } | |
1583 | EXPORT_SYMBOL_GPL(blkdev_aio_write); | |
1584 | ||
1585 | /* | |
1586 | * Try to release a page associated with block device when the system | |
1587 | * is under memory pressure. | |
1588 | */ | |
1589 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1590 | { | |
1591 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1592 | ||
1593 | if (super && super->s_op->bdev_try_to_free_page) | |
1594 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1595 | ||
1596 | return try_to_free_buffers(page); | |
1597 | } | |
1598 | ||
1599 | static const struct address_space_operations def_blk_aops = { | |
1600 | .readpage = blkdev_readpage, | |
1601 | .writepage = blkdev_writepage, | |
1602 | .write_begin = blkdev_write_begin, | |
1603 | .write_end = blkdev_write_end, | |
1604 | .writepages = generic_writepages, | |
1605 | .releasepage = blkdev_releasepage, | |
1606 | .direct_IO = blkdev_direct_IO, | |
1607 | }; | |
1608 | ||
1609 | const struct file_operations def_blk_fops = { | |
1610 | .open = blkdev_open, | |
1611 | .release = blkdev_close, | |
1612 | .llseek = block_llseek, | |
1613 | .read = do_sync_read, | |
1614 | .write = do_sync_write, | |
1615 | .aio_read = generic_file_aio_read, | |
1616 | .aio_write = blkdev_aio_write, | |
1617 | .mmap = generic_file_mmap, | |
1618 | .fsync = blkdev_fsync, | |
1619 | .unlocked_ioctl = block_ioctl, | |
1620 | #ifdef CONFIG_COMPAT | |
1621 | .compat_ioctl = compat_blkdev_ioctl, | |
1622 | #endif | |
1623 | .splice_read = generic_file_splice_read, | |
1624 | .splice_write = generic_file_splice_write, | |
1625 | }; | |
1626 | ||
1627 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1628 | { | |
1629 | int res; | |
1630 | mm_segment_t old_fs = get_fs(); | |
1631 | set_fs(KERNEL_DS); | |
1632 | res = blkdev_ioctl(bdev, 0, cmd, arg); | |
1633 | set_fs(old_fs); | |
1634 | return res; | |
1635 | } | |
1636 | ||
1637 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1638 | ||
1639 | /** | |
1640 | * lookup_bdev - lookup a struct block_device by name | |
1641 | * @pathname: special file representing the block device | |
1642 | * | |
1643 | * Get a reference to the blockdevice at @pathname in the current | |
1644 | * namespace if possible and return it. Return ERR_PTR(error) | |
1645 | * otherwise. | |
1646 | */ | |
1647 | struct block_device *lookup_bdev(const char *pathname) | |
1648 | { | |
1649 | struct block_device *bdev; | |
1650 | struct inode *inode; | |
1651 | struct path path; | |
1652 | int error; | |
1653 | ||
1654 | if (!pathname || !*pathname) | |
1655 | return ERR_PTR(-EINVAL); | |
1656 | ||
1657 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); | |
1658 | if (error) | |
1659 | return ERR_PTR(error); | |
1660 | ||
1661 | inode = path.dentry->d_inode; | |
1662 | error = -ENOTBLK; | |
1663 | if (!S_ISBLK(inode->i_mode)) | |
1664 | goto fail; | |
1665 | error = -EACCES; | |
1666 | if (path.mnt->mnt_flags & MNT_NODEV) | |
1667 | goto fail; | |
1668 | error = -ENOMEM; | |
1669 | bdev = bd_acquire(inode); | |
1670 | if (!bdev) | |
1671 | goto fail; | |
1672 | out: | |
1673 | path_put(&path); | |
1674 | return bdev; | |
1675 | fail: | |
1676 | bdev = ERR_PTR(error); | |
1677 | goto out; | |
1678 | } | |
1679 | EXPORT_SYMBOL(lookup_bdev); | |
1680 | ||
1681 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) | |
1682 | { | |
1683 | struct super_block *sb = get_super(bdev); | |
1684 | int res = 0; | |
1685 | ||
1686 | if (sb) { | |
1687 | /* | |
1688 | * no need to lock the super, get_super holds the | |
1689 | * read mutex so the filesystem cannot go away | |
1690 | * under us (->put_super runs with the write lock | |
1691 | * hold). | |
1692 | */ | |
1693 | shrink_dcache_sb(sb); | |
1694 | res = invalidate_inodes(sb, kill_dirty); | |
1695 | drop_super(sb); | |
1696 | } | |
1697 | invalidate_bdev(bdev); | |
1698 | return res; | |
1699 | } | |
1700 | EXPORT_SYMBOL(__invalidate_device); |