Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/block_dev.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE | |
6 | */ | |
7 | ||
1da177e4 LT |
8 | #include <linux/init.h> |
9 | #include <linux/mm.h> | |
10 | #include <linux/fcntl.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/kmod.h> | |
13 | #include <linux/major.h> | |
7db9cfd3 | 14 | #include <linux/device_cgroup.h> |
1da177e4 LT |
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> | |
ff01bb48 | 20 | #include <linux/swap.h> |
585d3bc0 | 21 | #include <linux/pagevec.h> |
811d736f | 22 | #include <linux/writeback.h> |
1da177e4 LT |
23 | #include <linux/mpage.h> |
24 | #include <linux/mount.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/namei.h> | |
1368c4f2 | 27 | #include <linux/log2.h> |
2e1483c9 | 28 | #include <linux/kmemleak.h> |
ff01bb48 | 29 | #include <linux/cleancache.h> |
1da177e4 | 30 | #include <asm/uaccess.h> |
07f3f05c | 31 | #include "internal.h" |
1da177e4 LT |
32 | |
33 | struct bdev_inode { | |
34 | struct block_device bdev; | |
35 | struct inode vfs_inode; | |
36 | }; | |
37 | ||
4c54ac62 AB |
38 | static const struct address_space_operations def_blk_aops; |
39 | ||
1da177e4 LT |
40 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
41 | { | |
42 | return container_of(inode, struct bdev_inode, vfs_inode); | |
43 | } | |
44 | ||
45 | inline struct block_device *I_BDEV(struct inode *inode) | |
46 | { | |
47 | return &BDEV_I(inode)->bdev; | |
48 | } | |
1da177e4 LT |
49 | EXPORT_SYMBOL(I_BDEV); |
50 | ||
a5491e0c | 51 | /* |
f758eeab CH |
52 | * Move the inode from its current bdi to a new bdi. If the inode is dirty we |
53 | * need to move it onto the dirty list of @dst so that the inode is always on | |
54 | * the right list. | |
a5491e0c DC |
55 | */ |
56 | static void bdev_inode_switch_bdi(struct inode *inode, | |
57 | struct backing_dev_info *dst) | |
58 | { | |
f758eeab CH |
59 | struct backing_dev_info *old = inode->i_data.backing_dev_info; |
60 | ||
61 | if (unlikely(dst == old)) /* deadlock avoidance */ | |
62 | return; | |
63 | bdi_lock_two(&old->wb, &dst->wb); | |
250df6ed | 64 | spin_lock(&inode->i_lock); |
a5491e0c DC |
65 | inode->i_data.backing_dev_info = dst; |
66 | if (inode->i_state & I_DIRTY) | |
7ccf19a8 | 67 | list_move(&inode->i_wb_list, &dst->wb.b_dirty); |
250df6ed | 68 | spin_unlock(&inode->i_lock); |
f758eeab CH |
69 | spin_unlock(&old->wb.list_lock); |
70 | spin_unlock(&dst->wb.list_lock); | |
a5491e0c DC |
71 | } |
72 | ||
1da177e4 LT |
73 | static sector_t max_block(struct block_device *bdev) |
74 | { | |
75 | sector_t retval = ~((sector_t)0); | |
76 | loff_t sz = i_size_read(bdev->bd_inode); | |
77 | ||
78 | if (sz) { | |
79 | unsigned int size = block_size(bdev); | |
80 | unsigned int sizebits = blksize_bits(size); | |
81 | retval = (sz >> sizebits); | |
82 | } | |
83 | return retval; | |
84 | } | |
85 | ||
f9a14399 | 86 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
ff01bb48 | 87 | void kill_bdev(struct block_device *bdev) |
1da177e4 | 88 | { |
ff01bb48 AV |
89 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
90 | ||
91 | if (mapping->nrpages == 0) | |
f9a14399 | 92 | return; |
ff01bb48 | 93 | |
f9a14399 | 94 | invalidate_bh_lrus(); |
ff01bb48 | 95 | truncate_inode_pages(mapping, 0); |
1da177e4 | 96 | } |
ff01bb48 AV |
97 | EXPORT_SYMBOL(kill_bdev); |
98 | ||
99 | /* Invalidate clean unused buffers and pagecache. */ | |
100 | void invalidate_bdev(struct block_device *bdev) | |
101 | { | |
102 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
103 | ||
104 | if (mapping->nrpages == 0) | |
105 | return; | |
106 | ||
107 | invalidate_bh_lrus(); | |
108 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
109 | invalidate_mapping_pages(mapping, 0, -1); | |
110 | /* 99% of the time, we don't need to flush the cleancache on the bdev. | |
111 | * But, for the strange corners, lets be cautious | |
112 | */ | |
113 | cleancache_flush_inode(mapping); | |
114 | } | |
115 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 LT |
116 | |
117 | int set_blocksize(struct block_device *bdev, int size) | |
118 | { | |
119 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 120 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
121 | return -EINVAL; |
122 | ||
123 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 124 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
125 | return -EINVAL; |
126 | ||
127 | /* Don't change the size if it is same as current */ | |
128 | if (bdev->bd_block_size != size) { | |
129 | sync_blockdev(bdev); | |
130 | bdev->bd_block_size = size; | |
131 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
132 | kill_bdev(bdev); | |
133 | } | |
134 | return 0; | |
135 | } | |
136 | ||
137 | EXPORT_SYMBOL(set_blocksize); | |
138 | ||
139 | int sb_set_blocksize(struct super_block *sb, int size) | |
140 | { | |
1da177e4 LT |
141 | if (set_blocksize(sb->s_bdev, size)) |
142 | return 0; | |
143 | /* If we get here, we know size is power of two | |
144 | * and it's value is between 512 and PAGE_SIZE */ | |
145 | sb->s_blocksize = size; | |
38885bd4 | 146 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
147 | return sb->s_blocksize; |
148 | } | |
149 | ||
150 | EXPORT_SYMBOL(sb_set_blocksize); | |
151 | ||
152 | int sb_min_blocksize(struct super_block *sb, int size) | |
153 | { | |
e1defc4f | 154 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
155 | if (size < minsize) |
156 | size = minsize; | |
157 | return sb_set_blocksize(sb, size); | |
158 | } | |
159 | ||
160 | EXPORT_SYMBOL(sb_min_blocksize); | |
161 | ||
162 | static int | |
163 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
164 | struct buffer_head *bh, int create) | |
165 | { | |
166 | if (iblock >= max_block(I_BDEV(inode))) { | |
167 | if (create) | |
168 | return -EIO; | |
169 | ||
170 | /* | |
171 | * for reads, we're just trying to fill a partial page. | |
172 | * return a hole, they will have to call get_block again | |
173 | * before they can fill it, and they will get -EIO at that | |
174 | * time | |
175 | */ | |
176 | return 0; | |
177 | } | |
178 | bh->b_bdev = I_BDEV(inode); | |
179 | bh->b_blocknr = iblock; | |
180 | set_buffer_mapped(bh); | |
181 | return 0; | |
182 | } | |
183 | ||
b2e895db AM |
184 | static int |
185 | blkdev_get_blocks(struct inode *inode, sector_t iblock, | |
186 | struct buffer_head *bh, int create) | |
187 | { | |
188 | sector_t end_block = max_block(I_BDEV(inode)); | |
189 | unsigned long max_blocks = bh->b_size >> inode->i_blkbits; | |
190 | ||
191 | if ((iblock + max_blocks) > end_block) { | |
192 | max_blocks = end_block - iblock; | |
193 | if ((long)max_blocks <= 0) { | |
194 | if (create) | |
195 | return -EIO; /* write fully beyond EOF */ | |
196 | /* | |
197 | * It is a read which is fully beyond EOF. We return | |
198 | * a !buffer_mapped buffer | |
199 | */ | |
200 | max_blocks = 0; | |
201 | } | |
202 | } | |
203 | ||
204 | bh->b_bdev = I_BDEV(inode); | |
205 | bh->b_blocknr = iblock; | |
206 | bh->b_size = max_blocks << inode->i_blkbits; | |
207 | if (max_blocks) | |
208 | set_buffer_mapped(bh); | |
209 | return 0; | |
210 | } | |
211 | ||
212 | static ssize_t | |
213 | blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, | |
214 | loff_t offset, unsigned long nr_segs) | |
215 | { | |
216 | struct file *file = iocb->ki_filp; | |
217 | struct inode *inode = file->f_mapping->host; | |
218 | ||
eafdc7d1 CH |
219 | return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset, |
220 | nr_segs, blkdev_get_blocks, NULL, NULL, 0); | |
b2e895db AM |
221 | } |
222 | ||
5cee5815 JK |
223 | int __sync_blockdev(struct block_device *bdev, int wait) |
224 | { | |
225 | if (!bdev) | |
226 | return 0; | |
227 | if (!wait) | |
228 | return filemap_flush(bdev->bd_inode->i_mapping); | |
229 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
230 | } | |
231 | ||
585d3bc0 NP |
232 | /* |
233 | * Write out and wait upon all the dirty data associated with a block | |
234 | * device via its mapping. Does not take the superblock lock. | |
235 | */ | |
236 | int sync_blockdev(struct block_device *bdev) | |
237 | { | |
5cee5815 | 238 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
239 | } |
240 | EXPORT_SYMBOL(sync_blockdev); | |
241 | ||
242 | /* | |
243 | * Write out and wait upon all dirty data associated with this | |
244 | * device. Filesystem data as well as the underlying block | |
245 | * device. Takes the superblock lock. | |
246 | */ | |
247 | int fsync_bdev(struct block_device *bdev) | |
248 | { | |
249 | struct super_block *sb = get_super(bdev); | |
250 | if (sb) { | |
60b0680f | 251 | int res = sync_filesystem(sb); |
585d3bc0 NP |
252 | drop_super(sb); |
253 | return res; | |
254 | } | |
255 | return sync_blockdev(bdev); | |
256 | } | |
47e4491b | 257 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
258 | |
259 | /** | |
260 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
261 | * @bdev: blockdevice to lock | |
262 | * | |
585d3bc0 NP |
263 | * If a superblock is found on this device, we take the s_umount semaphore |
264 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
265 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
266 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
267 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
268 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
269 | * actually. | |
270 | */ | |
271 | struct super_block *freeze_bdev(struct block_device *bdev) | |
272 | { | |
273 | struct super_block *sb; | |
274 | int error = 0; | |
275 | ||
276 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
277 | if (++bdev->bd_fsfreeze_count > 1) { |
278 | /* | |
279 | * We don't even need to grab a reference - the first call | |
280 | * to freeze_bdev grab an active reference and only the last | |
281 | * thaw_bdev drops it. | |
282 | */ | |
585d3bc0 | 283 | sb = get_super(bdev); |
4504230a CH |
284 | drop_super(sb); |
285 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
286 | return sb; | |
287 | } | |
288 | ||
289 | sb = get_active_super(bdev); | |
290 | if (!sb) | |
291 | goto out; | |
18e9e510 JB |
292 | error = freeze_super(sb); |
293 | if (error) { | |
294 | deactivate_super(sb); | |
295 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 296 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 297 | return ERR_PTR(error); |
585d3bc0 | 298 | } |
18e9e510 | 299 | deactivate_super(sb); |
4504230a | 300 | out: |
585d3bc0 NP |
301 | sync_blockdev(bdev); |
302 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 303 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
304 | } |
305 | EXPORT_SYMBOL(freeze_bdev); | |
306 | ||
307 | /** | |
308 | * thaw_bdev -- unlock filesystem | |
309 | * @bdev: blockdevice to unlock | |
310 | * @sb: associated superblock | |
311 | * | |
312 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
313 | */ | |
314 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
315 | { | |
4504230a | 316 | int error = -EINVAL; |
585d3bc0 NP |
317 | |
318 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 319 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 320 | goto out; |
4504230a CH |
321 | |
322 | error = 0; | |
323 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 324 | goto out; |
4504230a CH |
325 | |
326 | if (!sb) | |
18e9e510 | 327 | goto out; |
4504230a | 328 | |
18e9e510 JB |
329 | error = thaw_super(sb); |
330 | if (error) { | |
331 | bdev->bd_fsfreeze_count++; | |
332 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
333 | return error; | |
334 | } | |
335 | out: | |
585d3bc0 NP |
336 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
337 | return 0; | |
338 | } | |
339 | EXPORT_SYMBOL(thaw_bdev); | |
340 | ||
1da177e4 LT |
341 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
342 | { | |
343 | return block_write_full_page(page, blkdev_get_block, wbc); | |
344 | } | |
345 | ||
346 | static int blkdev_readpage(struct file * file, struct page * page) | |
347 | { | |
348 | return block_read_full_page(page, blkdev_get_block); | |
349 | } | |
350 | ||
6272b5a5 NP |
351 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
352 | loff_t pos, unsigned len, unsigned flags, | |
353 | struct page **pagep, void **fsdata) | |
1da177e4 | 354 | { |
155130a4 CH |
355 | return block_write_begin(mapping, pos, len, flags, pagep, |
356 | blkdev_get_block); | |
1da177e4 LT |
357 | } |
358 | ||
6272b5a5 NP |
359 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
360 | loff_t pos, unsigned len, unsigned copied, | |
361 | struct page *page, void *fsdata) | |
1da177e4 | 362 | { |
6272b5a5 NP |
363 | int ret; |
364 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
365 | ||
366 | unlock_page(page); | |
367 | page_cache_release(page); | |
368 | ||
369 | return ret; | |
1da177e4 LT |
370 | } |
371 | ||
372 | /* | |
373 | * private llseek: | |
0f7fc9e4 | 374 | * for a block special file file->f_path.dentry->d_inode->i_size is zero |
1da177e4 LT |
375 | * so we compute the size by hand (just as in block_read/write above) |
376 | */ | |
377 | static loff_t block_llseek(struct file *file, loff_t offset, int origin) | |
378 | { | |
379 | struct inode *bd_inode = file->f_mapping->host; | |
380 | loff_t size; | |
381 | loff_t retval; | |
382 | ||
1b1dcc1b | 383 | mutex_lock(&bd_inode->i_mutex); |
1da177e4 LT |
384 | size = i_size_read(bd_inode); |
385 | ||
06222e49 | 386 | retval = -EINVAL; |
1da177e4 | 387 | switch (origin) { |
06222e49 | 388 | case SEEK_END: |
1da177e4 LT |
389 | offset += size; |
390 | break; | |
06222e49 | 391 | case SEEK_CUR: |
1da177e4 | 392 | offset += file->f_pos; |
06222e49 JB |
393 | case SEEK_SET: |
394 | break; | |
395 | default: | |
396 | goto out; | |
1da177e4 | 397 | } |
1da177e4 LT |
398 | if (offset >= 0 && offset <= size) { |
399 | if (offset != file->f_pos) { | |
400 | file->f_pos = offset; | |
401 | } | |
402 | retval = offset; | |
403 | } | |
06222e49 | 404 | out: |
1b1dcc1b | 405 | mutex_unlock(&bd_inode->i_mutex); |
1da177e4 LT |
406 | return retval; |
407 | } | |
408 | ||
02c24a82 | 409 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 410 | { |
b8af67e2 AB |
411 | struct inode *bd_inode = filp->f_mapping->host; |
412 | struct block_device *bdev = I_BDEV(bd_inode); | |
ab0a9735 | 413 | int error; |
da5aa861 RW |
414 | |
415 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
416 | if (error) | |
417 | return error; | |
ab0a9735 | 418 | |
b8af67e2 AB |
419 | /* |
420 | * There is no need to serialise calls to blkdev_issue_flush with | |
421 | * i_mutex and doing so causes performance issues with concurrent | |
422 | * O_SYNC writers to a block device. | |
423 | */ | |
dd3932ed | 424 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
425 | if (error == -EOPNOTSUPP) |
426 | error = 0; | |
b8af67e2 | 427 | |
ab0a9735 | 428 | return error; |
1da177e4 | 429 | } |
b1dd3b28 | 430 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 LT |
431 | |
432 | /* | |
433 | * pseudo-fs | |
434 | */ | |
435 | ||
436 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 437 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
438 | |
439 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
440 | { | |
e94b1766 | 441 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
442 | if (!ei) |
443 | return NULL; | |
444 | return &ei->vfs_inode; | |
445 | } | |
446 | ||
fa0d7e3d | 447 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 448 | { |
fa0d7e3d | 449 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
450 | struct bdev_inode *bdi = BDEV_I(inode); |
451 | ||
1da177e4 LT |
452 | kmem_cache_free(bdev_cachep, bdi); |
453 | } | |
454 | ||
fa0d7e3d NP |
455 | static void bdev_destroy_inode(struct inode *inode) |
456 | { | |
457 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
458 | } | |
459 | ||
51cc5068 | 460 | static void init_once(void *foo) |
1da177e4 LT |
461 | { |
462 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
463 | struct block_device *bdev = &ei->bdev; | |
464 | ||
a35afb83 CL |
465 | memset(bdev, 0, sizeof(*bdev)); |
466 | mutex_init(&bdev->bd_mutex); | |
a35afb83 CL |
467 | INIT_LIST_HEAD(&bdev->bd_inodes); |
468 | INIT_LIST_HEAD(&bdev->bd_list); | |
49731baa TH |
469 | #ifdef CONFIG_SYSFS |
470 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
471 | #endif | |
a35afb83 | 472 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
473 | /* Initialize mutex for freeze. */ |
474 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
475 | } |
476 | ||
477 | static inline void __bd_forget(struct inode *inode) | |
478 | { | |
479 | list_del_init(&inode->i_devices); | |
480 | inode->i_bdev = NULL; | |
481 | inode->i_mapping = &inode->i_data; | |
482 | } | |
483 | ||
b57922d9 | 484 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
485 | { |
486 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
487 | struct list_head *p; | |
b57922d9 AV |
488 | truncate_inode_pages(&inode->i_data, 0); |
489 | invalidate_inode_buffers(inode); /* is it needed here? */ | |
490 | end_writeback(inode); | |
1da177e4 LT |
491 | spin_lock(&bdev_lock); |
492 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
493 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
494 | } | |
495 | list_del_init(&bdev->bd_list); | |
496 | spin_unlock(&bdev_lock); | |
497 | } | |
498 | ||
ee9b6d61 | 499 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
500 | .statfs = simple_statfs, |
501 | .alloc_inode = bdev_alloc_inode, | |
502 | .destroy_inode = bdev_destroy_inode, | |
503 | .drop_inode = generic_delete_inode, | |
b57922d9 | 504 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
505 | }; |
506 | ||
51139ada AV |
507 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
508 | int flags, const char *dev_name, void *data) | |
1da177e4 | 509 | { |
c74a1cbb | 510 | return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576); |
1da177e4 LT |
511 | } |
512 | ||
513 | static struct file_system_type bd_type = { | |
514 | .name = "bdev", | |
51139ada | 515 | .mount = bd_mount, |
1da177e4 LT |
516 | .kill_sb = kill_anon_super, |
517 | }; | |
518 | ||
f47ec3f2 | 519 | static struct super_block *blockdev_superblock __read_mostly; |
1da177e4 LT |
520 | |
521 | void __init bdev_cache_init(void) | |
522 | { | |
523 | int err; | |
c2acf7b9 DC |
524 | struct vfsmount *bd_mnt; |
525 | ||
1da177e4 | 526 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 PJ |
527 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
528 | SLAB_MEM_SPREAD|SLAB_PANIC), | |
20c2df83 | 529 | init_once); |
1da177e4 LT |
530 | err = register_filesystem(&bd_type); |
531 | if (err) | |
532 | panic("Cannot register bdev pseudo-fs"); | |
533 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
534 | if (IS_ERR(bd_mnt)) |
535 | panic("Cannot create bdev pseudo-fs"); | |
2e1483c9 CM |
536 | /* |
537 | * This vfsmount structure is only used to obtain the | |
538 | * blockdev_superblock, so tell kmemleak not to report it. | |
539 | */ | |
540 | kmemleak_not_leak(bd_mnt); | |
1da177e4 LT |
541 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
542 | } | |
543 | ||
544 | /* | |
545 | * Most likely _very_ bad one - but then it's hardly critical for small | |
546 | * /dev and can be fixed when somebody will need really large one. | |
547 | * Keep in mind that it will be fed through icache hash function too. | |
548 | */ | |
549 | static inline unsigned long hash(dev_t dev) | |
550 | { | |
551 | return MAJOR(dev)+MINOR(dev); | |
552 | } | |
553 | ||
554 | static int bdev_test(struct inode *inode, void *data) | |
555 | { | |
556 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
557 | } | |
558 | ||
559 | static int bdev_set(struct inode *inode, void *data) | |
560 | { | |
561 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
562 | return 0; | |
563 | } | |
564 | ||
565 | static LIST_HEAD(all_bdevs); | |
566 | ||
567 | struct block_device *bdget(dev_t dev) | |
568 | { | |
569 | struct block_device *bdev; | |
570 | struct inode *inode; | |
571 | ||
c2acf7b9 | 572 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
573 | bdev_test, bdev_set, &dev); |
574 | ||
575 | if (!inode) | |
576 | return NULL; | |
577 | ||
578 | bdev = &BDEV_I(inode)->bdev; | |
579 | ||
580 | if (inode->i_state & I_NEW) { | |
581 | bdev->bd_contains = NULL; | |
782b94cd | 582 | bdev->bd_super = NULL; |
1da177e4 LT |
583 | bdev->bd_inode = inode; |
584 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
585 | bdev->bd_part_count = 0; | |
586 | bdev->bd_invalidated = 0; | |
587 | inode->i_mode = S_IFBLK; | |
588 | inode->i_rdev = dev; | |
589 | inode->i_bdev = bdev; | |
590 | inode->i_data.a_ops = &def_blk_aops; | |
591 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
592 | inode->i_data.backing_dev_info = &default_backing_dev_info; | |
593 | spin_lock(&bdev_lock); | |
594 | list_add(&bdev->bd_list, &all_bdevs); | |
595 | spin_unlock(&bdev_lock); | |
596 | unlock_new_inode(inode); | |
597 | } | |
598 | return bdev; | |
599 | } | |
600 | ||
601 | EXPORT_SYMBOL(bdget); | |
602 | ||
dddac6a7 AJ |
603 | /** |
604 | * bdgrab -- Grab a reference to an already referenced block device | |
605 | * @bdev: Block device to grab a reference to. | |
606 | */ | |
607 | struct block_device *bdgrab(struct block_device *bdev) | |
608 | { | |
7de9c6ee | 609 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
610 | return bdev; |
611 | } | |
612 | ||
1da177e4 LT |
613 | long nr_blockdev_pages(void) |
614 | { | |
203a2935 | 615 | struct block_device *bdev; |
1da177e4 LT |
616 | long ret = 0; |
617 | spin_lock(&bdev_lock); | |
203a2935 | 618 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
619 | ret += bdev->bd_inode->i_mapping->nrpages; |
620 | } | |
621 | spin_unlock(&bdev_lock); | |
622 | return ret; | |
623 | } | |
624 | ||
625 | void bdput(struct block_device *bdev) | |
626 | { | |
627 | iput(bdev->bd_inode); | |
628 | } | |
629 | ||
630 | EXPORT_SYMBOL(bdput); | |
631 | ||
632 | static struct block_device *bd_acquire(struct inode *inode) | |
633 | { | |
634 | struct block_device *bdev; | |
09d967c6 | 635 | |
1da177e4 LT |
636 | spin_lock(&bdev_lock); |
637 | bdev = inode->i_bdev; | |
09d967c6 | 638 | if (bdev) { |
7de9c6ee | 639 | ihold(bdev->bd_inode); |
1da177e4 LT |
640 | spin_unlock(&bdev_lock); |
641 | return bdev; | |
642 | } | |
643 | spin_unlock(&bdev_lock); | |
09d967c6 | 644 | |
1da177e4 LT |
645 | bdev = bdget(inode->i_rdev); |
646 | if (bdev) { | |
647 | spin_lock(&bdev_lock); | |
09d967c6 OH |
648 | if (!inode->i_bdev) { |
649 | /* | |
7de9c6ee | 650 | * We take an additional reference to bd_inode, |
09d967c6 OH |
651 | * and it's released in clear_inode() of inode. |
652 | * So, we can access it via ->i_mapping always | |
653 | * without igrab(). | |
654 | */ | |
7de9c6ee | 655 | ihold(bdev->bd_inode); |
09d967c6 OH |
656 | inode->i_bdev = bdev; |
657 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
658 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
659 | } | |
1da177e4 LT |
660 | spin_unlock(&bdev_lock); |
661 | } | |
662 | return bdev; | |
663 | } | |
664 | ||
f47ec3f2 AV |
665 | static inline int sb_is_blkdev_sb(struct super_block *sb) |
666 | { | |
667 | return sb == blockdev_superblock; | |
668 | } | |
669 | ||
1da177e4 LT |
670 | /* Call when you free inode */ |
671 | ||
672 | void bd_forget(struct inode *inode) | |
673 | { | |
09d967c6 OH |
674 | struct block_device *bdev = NULL; |
675 | ||
1da177e4 | 676 | spin_lock(&bdev_lock); |
09d967c6 | 677 | if (inode->i_bdev) { |
c2acf7b9 | 678 | if (!sb_is_blkdev_sb(inode->i_sb)) |
09d967c6 | 679 | bdev = inode->i_bdev; |
1da177e4 | 680 | __bd_forget(inode); |
09d967c6 | 681 | } |
1da177e4 | 682 | spin_unlock(&bdev_lock); |
09d967c6 OH |
683 | |
684 | if (bdev) | |
685 | iput(bdev->bd_inode); | |
1da177e4 LT |
686 | } |
687 | ||
1a3cbbc5 TH |
688 | /** |
689 | * bd_may_claim - test whether a block device can be claimed | |
690 | * @bdev: block device of interest | |
691 | * @whole: whole block device containing @bdev, may equal @bdev | |
692 | * @holder: holder trying to claim @bdev | |
693 | * | |
25985edc | 694 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
695 | * |
696 | * CONTEXT: | |
697 | * spin_lock(&bdev_lock). | |
698 | * | |
699 | * RETURNS: | |
700 | * %true if @bdev can be claimed, %false otherwise. | |
701 | */ | |
702 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
703 | void *holder) | |
1da177e4 | 704 | { |
1da177e4 | 705 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 706 | return true; /* already a holder */ |
1da177e4 | 707 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 708 | return false; /* held by someone else */ |
1da177e4 | 709 | else if (bdev->bd_contains == bdev) |
1a3cbbc5 | 710 | return true; /* is a whole device which isn't held */ |
1da177e4 | 711 | |
e525fd89 | 712 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
713 | return true; /* is a partition of a device that is being partitioned */ |
714 | else if (whole->bd_holder != NULL) | |
715 | return false; /* is a partition of a held device */ | |
1da177e4 | 716 | else |
1a3cbbc5 TH |
717 | return true; /* is a partition of an un-held device */ |
718 | } | |
719 | ||
6b4517a7 TH |
720 | /** |
721 | * bd_prepare_to_claim - prepare to claim a block device | |
722 | * @bdev: block device of interest | |
723 | * @whole: the whole device containing @bdev, may equal @bdev | |
724 | * @holder: holder trying to claim @bdev | |
725 | * | |
726 | * Prepare to claim @bdev. This function fails if @bdev is already | |
727 | * claimed by another holder and waits if another claiming is in | |
728 | * progress. This function doesn't actually claim. On successful | |
729 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
730 | * | |
731 | * CONTEXT: | |
732 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
733 | * it multiple times. | |
734 | * | |
735 | * RETURNS: | |
736 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
737 | */ | |
738 | static int bd_prepare_to_claim(struct block_device *bdev, | |
739 | struct block_device *whole, void *holder) | |
740 | { | |
741 | retry: | |
742 | /* if someone else claimed, fail */ | |
743 | if (!bd_may_claim(bdev, whole, holder)) | |
744 | return -EBUSY; | |
745 | ||
e75aa858 TH |
746 | /* if claiming is already in progress, wait for it to finish */ |
747 | if (whole->bd_claiming) { | |
6b4517a7 TH |
748 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
749 | DEFINE_WAIT(wait); | |
750 | ||
751 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
752 | spin_unlock(&bdev_lock); | |
753 | schedule(); | |
754 | finish_wait(wq, &wait); | |
755 | spin_lock(&bdev_lock); | |
756 | goto retry; | |
757 | } | |
758 | ||
759 | /* yay, all mine */ | |
760 | return 0; | |
761 | } | |
762 | ||
763 | /** | |
764 | * bd_start_claiming - start claiming a block device | |
765 | * @bdev: block device of interest | |
766 | * @holder: holder trying to claim @bdev | |
767 | * | |
768 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
769 | * exclusively and mark that an exclusive open is in progress. Each | |
770 | * successful call to this function must be matched with a call to | |
b0018361 NP |
771 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
772 | * fail). | |
773 | * | |
774 | * This function is used to gain exclusive access to the block device | |
775 | * without actually causing other exclusive open attempts to fail. It | |
776 | * should be used when the open sequence itself requires exclusive | |
777 | * access but may subsequently fail. | |
6b4517a7 TH |
778 | * |
779 | * CONTEXT: | |
780 | * Might sleep. | |
781 | * | |
782 | * RETURNS: | |
783 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
784 | * value on failure. | |
785 | */ | |
786 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
787 | void *holder) | |
788 | { | |
789 | struct gendisk *disk; | |
790 | struct block_device *whole; | |
791 | int partno, err; | |
792 | ||
793 | might_sleep(); | |
794 | ||
795 | /* | |
796 | * @bdev might not have been initialized properly yet, look up | |
797 | * and grab the outer block device the hard way. | |
798 | */ | |
799 | disk = get_gendisk(bdev->bd_dev, &partno); | |
800 | if (!disk) | |
801 | return ERR_PTR(-ENXIO); | |
802 | ||
d4c208b8 TH |
803 | /* |
804 | * Normally, @bdev should equal what's returned from bdget_disk() | |
805 | * if partno is 0; however, some drivers (floppy) use multiple | |
806 | * bdev's for the same physical device and @bdev may be one of the | |
807 | * aliases. Keep @bdev if partno is 0. This means claimer | |
808 | * tracking is broken for those devices but it has always been that | |
809 | * way. | |
810 | */ | |
811 | if (partno) | |
812 | whole = bdget_disk(disk, 0); | |
813 | else | |
814 | whole = bdgrab(bdev); | |
815 | ||
cf342570 | 816 | module_put(disk->fops->owner); |
6b4517a7 TH |
817 | put_disk(disk); |
818 | if (!whole) | |
819 | return ERR_PTR(-ENOMEM); | |
820 | ||
821 | /* prepare to claim, if successful, mark claiming in progress */ | |
822 | spin_lock(&bdev_lock); | |
823 | ||
824 | err = bd_prepare_to_claim(bdev, whole, holder); | |
825 | if (err == 0) { | |
826 | whole->bd_claiming = holder; | |
827 | spin_unlock(&bdev_lock); | |
828 | return whole; | |
829 | } else { | |
830 | spin_unlock(&bdev_lock); | |
831 | bdput(whole); | |
832 | return ERR_PTR(err); | |
833 | } | |
834 | } | |
835 | ||
641dc636 | 836 | #ifdef CONFIG_SYSFS |
49731baa TH |
837 | struct bd_holder_disk { |
838 | struct list_head list; | |
839 | struct gendisk *disk; | |
840 | int refcnt; | |
841 | }; | |
842 | ||
843 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
844 | struct gendisk *disk) | |
845 | { | |
846 | struct bd_holder_disk *holder; | |
847 | ||
848 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
849 | if (holder->disk == disk) | |
850 | return holder; | |
851 | return NULL; | |
852 | } | |
853 | ||
4d7dd8fd | 854 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 855 | { |
4d7dd8fd | 856 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
857 | } |
858 | ||
859 | static void del_symlink(struct kobject *from, struct kobject *to) | |
860 | { | |
641dc636 JN |
861 | sysfs_remove_link(from, kobject_name(to)); |
862 | } | |
863 | ||
df6c0cd9 | 864 | /** |
e09b457b TH |
865 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
866 | * @bdev: the claimed slave bdev | |
867 | * @disk: the holding disk | |
df6c0cd9 | 868 | * |
49731baa TH |
869 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
870 | * | |
e09b457b | 871 | * This functions creates the following sysfs symlinks. |
641dc636 | 872 | * |
e09b457b TH |
873 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
874 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 875 | * |
e09b457b TH |
876 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
877 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 878 | * |
e09b457b TH |
879 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
880 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 881 | * |
e09b457b TH |
882 | * The caller must have claimed @bdev before calling this function and |
883 | * ensure that both @bdev and @disk are valid during the creation and | |
884 | * lifetime of these symlinks. | |
641dc636 | 885 | * |
e09b457b TH |
886 | * CONTEXT: |
887 | * Might sleep. | |
641dc636 | 888 | * |
e09b457b TH |
889 | * RETURNS: |
890 | * 0 on success, -errno on failure. | |
641dc636 | 891 | */ |
e09b457b | 892 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 893 | { |
49731baa | 894 | struct bd_holder_disk *holder; |
e09b457b | 895 | int ret = 0; |
641dc636 | 896 | |
2e7b651d | 897 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 898 | |
49731baa | 899 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 900 | |
e09b457b TH |
901 | /* FIXME: remove the following once add_disk() handles errors */ |
902 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
903 | goto out_unlock; | |
4e91672c | 904 | |
49731baa TH |
905 | holder = bd_find_holder_disk(bdev, disk); |
906 | if (holder) { | |
907 | holder->refcnt++; | |
e09b457b | 908 | goto out_unlock; |
49731baa | 909 | } |
641dc636 | 910 | |
49731baa TH |
911 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
912 | if (!holder) { | |
913 | ret = -ENOMEM; | |
e09b457b TH |
914 | goto out_unlock; |
915 | } | |
641dc636 | 916 | |
49731baa TH |
917 | INIT_LIST_HEAD(&holder->list); |
918 | holder->disk = disk; | |
919 | holder->refcnt = 1; | |
920 | ||
921 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
922 | if (ret) | |
923 | goto out_free; | |
924 | ||
925 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
926 | if (ret) | |
927 | goto out_del; | |
e7407d16 TH |
928 | /* |
929 | * bdev could be deleted beneath us which would implicitly destroy | |
930 | * the holder directory. Hold on to it. | |
931 | */ | |
932 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
933 | |
934 | list_add(&holder->list, &bdev->bd_holder_disks); | |
935 | goto out_unlock; | |
936 | ||
937 | out_del: | |
938 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
939 | out_free: | |
940 | kfree(holder); | |
e09b457b | 941 | out_unlock: |
b4cf1b72 | 942 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 943 | return ret; |
641dc636 | 944 | } |
e09b457b | 945 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 946 | |
49731baa TH |
947 | /** |
948 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
949 | * @bdev: the calimed slave bdev | |
950 | * @disk: the holding disk | |
951 | * | |
952 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
953 | * | |
954 | * CONTEXT: | |
955 | * Might sleep. | |
956 | */ | |
957 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 958 | { |
49731baa | 959 | struct bd_holder_disk *holder; |
641dc636 | 960 | |
49731baa | 961 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 962 | |
49731baa TH |
963 | holder = bd_find_holder_disk(bdev, disk); |
964 | ||
965 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
966 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
967 | del_symlink(bdev->bd_part->holder_dir, | |
968 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 969 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
970 | list_del_init(&holder->list); |
971 | kfree(holder); | |
972 | } | |
973 | ||
974 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 975 | } |
49731baa | 976 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 977 | #endif |
1da177e4 | 978 | |
56ade44b AP |
979 | /** |
980 | * flush_disk - invalidates all buffer-cache entries on a disk | |
981 | * | |
982 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 983 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
984 | * |
985 | * Invalidates all buffer-cache entries on a disk. It should be called | |
986 | * when a disk has been changed -- either by a media change or online | |
987 | * resize. | |
988 | */ | |
93b270f7 | 989 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 990 | { |
93b270f7 | 991 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b AP |
992 | char name[BDEVNAME_SIZE] = ""; |
993 | ||
994 | if (bdev->bd_disk) | |
995 | disk_name(bdev->bd_disk, 0, name); | |
996 | printk(KERN_WARNING "VFS: busy inodes on changed media or " | |
997 | "resized disk %s\n", name); | |
998 | } | |
999 | ||
1000 | if (!bdev->bd_disk) | |
1001 | return; | |
d27769ec | 1002 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1003 | bdev->bd_invalidated = 1; |
1004 | } | |
1005 | ||
c3279d14 | 1006 | /** |
57d1b536 | 1007 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1008 | * @disk: struct gendisk to check |
1009 | * @bdev: struct bdev to adjust. | |
1010 | * | |
1011 | * This routine checks to see if the bdev size does not match the disk size | |
1012 | * and adjusts it if it differs. | |
1013 | */ | |
1014 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1015 | { | |
1016 | loff_t disk_size, bdev_size; | |
1017 | ||
1018 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1019 | bdev_size = i_size_read(bdev->bd_inode); | |
1020 | if (disk_size != bdev_size) { | |
1021 | char name[BDEVNAME_SIZE]; | |
1022 | ||
1023 | disk_name(disk, 0, name); | |
1024 | printk(KERN_INFO | |
1025 | "%s: detected capacity change from %lld to %lld\n", | |
1026 | name, bdev_size, disk_size); | |
1027 | i_size_write(bdev->bd_inode, disk_size); | |
93b270f7 | 1028 | flush_disk(bdev, false); |
c3279d14 AP |
1029 | } |
1030 | } | |
1031 | EXPORT_SYMBOL(check_disk_size_change); | |
1032 | ||
0c002c2f | 1033 | /** |
57d1b536 | 1034 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1035 | * @disk: struct gendisk to be revalidated |
1036 | * | |
1037 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1038 | * call-backs. It is used to do common pre and post operations needed | |
1039 | * for all revalidate_disk operations. | |
1040 | */ | |
1041 | int revalidate_disk(struct gendisk *disk) | |
1042 | { | |
c3279d14 | 1043 | struct block_device *bdev; |
0c002c2f AP |
1044 | int ret = 0; |
1045 | ||
1046 | if (disk->fops->revalidate_disk) | |
1047 | ret = disk->fops->revalidate_disk(disk); | |
1048 | ||
c3279d14 AP |
1049 | bdev = bdget_disk(disk, 0); |
1050 | if (!bdev) | |
1051 | return ret; | |
1052 | ||
1053 | mutex_lock(&bdev->bd_mutex); | |
1054 | check_disk_size_change(disk, bdev); | |
1055 | mutex_unlock(&bdev->bd_mutex); | |
1056 | bdput(bdev); | |
0c002c2f AP |
1057 | return ret; |
1058 | } | |
1059 | EXPORT_SYMBOL(revalidate_disk); | |
1060 | ||
1da177e4 LT |
1061 | /* |
1062 | * This routine checks whether a removable media has been changed, | |
1063 | * and invalidates all buffer-cache-entries in that case. This | |
1064 | * is a relatively slow routine, so we have to try to minimize using | |
1065 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1066 | * is the best way of combining speed and utility, I think. | |
1067 | * People changing diskettes in the middle of an operation deserve | |
1068 | * to lose :-) | |
1069 | */ | |
1070 | int check_disk_change(struct block_device *bdev) | |
1071 | { | |
1072 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1073 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1074 | unsigned int events; |
1da177e4 | 1075 | |
77ea887e TH |
1076 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1077 | DISK_EVENT_EJECT_REQUEST); | |
1078 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1079 | return 0; |
1080 | ||
93b270f7 | 1081 | flush_disk(bdev, true); |
1da177e4 LT |
1082 | if (bdops->revalidate_disk) |
1083 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1084 | return 1; |
1085 | } | |
1086 | ||
1087 | EXPORT_SYMBOL(check_disk_change); | |
1088 | ||
1089 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1090 | { | |
e1defc4f | 1091 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 LT |
1092 | |
1093 | bdev->bd_inode->i_size = size; | |
1094 | while (bsize < PAGE_CACHE_SIZE) { | |
1095 | if (size & bsize) | |
1096 | break; | |
1097 | bsize <<= 1; | |
1098 | } | |
1099 | bdev->bd_block_size = bsize; | |
1100 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1101 | } | |
1102 | EXPORT_SYMBOL(bd_set_size); | |
1103 | ||
9a1c3542 | 1104 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1105 | |
6d740cd5 PZ |
1106 | /* |
1107 | * bd_mutex locking: | |
1108 | * | |
1109 | * mutex_lock(part->bd_mutex) | |
1110 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1111 | */ | |
1112 | ||
572c4892 | 1113 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1114 | { |
1da177e4 | 1115 | struct gendisk *disk; |
523e1d39 | 1116 | struct module *owner; |
7db9cfd3 | 1117 | int ret; |
cf771cb5 | 1118 | int partno; |
fe6e9c1f AV |
1119 | int perm = 0; |
1120 | ||
572c4892 | 1121 | if (mode & FMODE_READ) |
fe6e9c1f | 1122 | perm |= MAY_READ; |
572c4892 | 1123 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1124 | perm |= MAY_WRITE; |
1125 | /* | |
1126 | * hooks: /n/, see "layering violations". | |
1127 | */ | |
b7300b78 CW |
1128 | if (!for_part) { |
1129 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1130 | if (ret != 0) { | |
1131 | bdput(bdev); | |
1132 | return ret; | |
1133 | } | |
82666020 | 1134 | } |
7db9cfd3 | 1135 | |
d3374825 | 1136 | restart: |
0762b8bd | 1137 | |
89f97496 | 1138 | ret = -ENXIO; |
cf771cb5 | 1139 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1140 | if (!disk) |
6e9624b8 | 1141 | goto out; |
523e1d39 | 1142 | owner = disk->fops->owner; |
1da177e4 | 1143 | |
69e02c59 | 1144 | disk_block_events(disk); |
6796bf54 | 1145 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1146 | if (!bdev->bd_openers) { |
1147 | bdev->bd_disk = disk; | |
1148 | bdev->bd_contains = bdev; | |
cf771cb5 | 1149 | if (!partno) { |
1da177e4 | 1150 | struct backing_dev_info *bdi; |
89f97496 TH |
1151 | |
1152 | ret = -ENXIO; | |
1153 | bdev->bd_part = disk_get_part(disk, partno); | |
1154 | if (!bdev->bd_part) | |
1155 | goto out_clear; | |
1156 | ||
1196f8b8 | 1157 | ret = 0; |
1da177e4 | 1158 | if (disk->fops->open) { |
572c4892 | 1159 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1160 | if (ret == -ERESTARTSYS) { |
1161 | /* Lost a race with 'disk' being | |
1162 | * deleted, try again. | |
1163 | * See md.c | |
1164 | */ | |
1165 | disk_put_part(bdev->bd_part); | |
1166 | bdev->bd_part = NULL; | |
d3374825 N |
1167 | bdev->bd_disk = NULL; |
1168 | mutex_unlock(&bdev->bd_mutex); | |
69e02c59 | 1169 | disk_unblock_events(disk); |
69e02c59 | 1170 | put_disk(disk); |
523e1d39 | 1171 | module_put(owner); |
d3374825 N |
1172 | goto restart; |
1173 | } | |
1da177e4 | 1174 | } |
7e69723f TH |
1175 | |
1176 | if (!ret && !bdev->bd_openers) { | |
1177 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); | |
1178 | bdi = blk_get_backing_dev_info(bdev); | |
1179 | if (bdi == NULL) | |
1180 | bdi = &default_backing_dev_info; | |
1181 | bdev_inode_switch_bdi(bdev->bd_inode, bdi); | |
1182 | } | |
1183 | ||
1196f8b8 TH |
1184 | /* |
1185 | * If the device is invalidated, rescan partition | |
1186 | * if open succeeded or failed with -ENOMEDIUM. | |
1187 | * The latter is necessary to prevent ghost | |
1188 | * partitions on a removed medium. | |
1189 | */ | |
1190 | if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM)) | |
1191 | rescan_partitions(disk, bdev); | |
1192 | if (ret) | |
1193 | goto out_clear; | |
1da177e4 | 1194 | } else { |
1da177e4 LT |
1195 | struct block_device *whole; |
1196 | whole = bdget_disk(disk, 0); | |
1197 | ret = -ENOMEM; | |
1198 | if (!whole) | |
0762b8bd | 1199 | goto out_clear; |
37be4124 | 1200 | BUG_ON(for_part); |
572c4892 | 1201 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1202 | if (ret) |
0762b8bd | 1203 | goto out_clear; |
1da177e4 | 1204 | bdev->bd_contains = whole; |
a5491e0c DC |
1205 | bdev_inode_switch_bdi(bdev->bd_inode, |
1206 | whole->bd_inode->i_data.backing_dev_info); | |
89f97496 | 1207 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1208 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1209 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1210 | ret = -ENXIO; |
0762b8bd | 1211 | goto out_clear; |
1da177e4 | 1212 | } |
89f97496 | 1213 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1da177e4 LT |
1214 | } |
1215 | } else { | |
1da177e4 | 1216 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1217 | ret = 0; |
1218 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1219 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 TH |
1220 | /* the same as first opener case, read comment there */ |
1221 | if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM)) | |
1da177e4 | 1222 | rescan_partitions(bdev->bd_disk, bdev); |
1196f8b8 TH |
1223 | if (ret) |
1224 | goto out_unlock_bdev; | |
1da177e4 | 1225 | } |
69e02c59 | 1226 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1227 | put_disk(disk); |
523e1d39 | 1228 | module_put(owner); |
1da177e4 LT |
1229 | } |
1230 | bdev->bd_openers++; | |
37be4124 N |
1231 | if (for_part) |
1232 | bdev->bd_part_count++; | |
c039e313 | 1233 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1234 | disk_unblock_events(disk); |
1da177e4 LT |
1235 | return 0; |
1236 | ||
0762b8bd | 1237 | out_clear: |
89f97496 | 1238 | disk_put_part(bdev->bd_part); |
1da177e4 | 1239 | bdev->bd_disk = NULL; |
0762b8bd | 1240 | bdev->bd_part = NULL; |
a5491e0c | 1241 | bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info); |
1da177e4 | 1242 | if (bdev != bdev->bd_contains) |
572c4892 | 1243 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1244 | bdev->bd_contains = NULL; |
0762b8bd | 1245 | out_unlock_bdev: |
c039e313 | 1246 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1247 | disk_unblock_events(disk); |
0762b8bd | 1248 | put_disk(disk); |
523e1d39 | 1249 | module_put(owner); |
4345caba | 1250 | out: |
0762b8bd TH |
1251 | bdput(bdev); |
1252 | ||
1da177e4 LT |
1253 | return ret; |
1254 | } | |
1255 | ||
d4d77629 TH |
1256 | /** |
1257 | * blkdev_get - open a block device | |
1258 | * @bdev: block_device to open | |
1259 | * @mode: FMODE_* mask | |
1260 | * @holder: exclusive holder identifier | |
1261 | * | |
1262 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1263 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1264 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1265 | * | |
1266 | * On success, the reference count of @bdev is unchanged. On failure, | |
1267 | * @bdev is put. | |
1268 | * | |
1269 | * CONTEXT: | |
1270 | * Might sleep. | |
1271 | * | |
1272 | * RETURNS: | |
1273 | * 0 on success, -errno on failure. | |
1274 | */ | |
e525fd89 | 1275 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1276 | { |
e525fd89 TH |
1277 | struct block_device *whole = NULL; |
1278 | int res; | |
1279 | ||
1280 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1281 | ||
1282 | if ((mode & FMODE_EXCL) && holder) { | |
1283 | whole = bd_start_claiming(bdev, holder); | |
1284 | if (IS_ERR(whole)) { | |
1285 | bdput(bdev); | |
1286 | return PTR_ERR(whole); | |
1287 | } | |
1288 | } | |
1289 | ||
1290 | res = __blkdev_get(bdev, mode, 0); | |
1291 | ||
1292 | if (whole) { | |
d4dc210f TH |
1293 | struct gendisk *disk = whole->bd_disk; |
1294 | ||
6a027eff | 1295 | /* finish claiming */ |
77ea887e | 1296 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1297 | spin_lock(&bdev_lock); |
1298 | ||
77ea887e | 1299 | if (!res) { |
6a027eff TH |
1300 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1301 | /* | |
1302 | * Note that for a whole device bd_holders | |
1303 | * will be incremented twice, and bd_holder | |
1304 | * will be set to bd_may_claim before being | |
1305 | * set to holder | |
1306 | */ | |
1307 | whole->bd_holders++; | |
1308 | whole->bd_holder = bd_may_claim; | |
1309 | bdev->bd_holders++; | |
1310 | bdev->bd_holder = holder; | |
1311 | } | |
1312 | ||
1313 | /* tell others that we're done */ | |
1314 | BUG_ON(whole->bd_claiming != holder); | |
1315 | whole->bd_claiming = NULL; | |
1316 | wake_up_bit(&whole->bd_claiming, 0); | |
1317 | ||
1318 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1319 | |
1320 | /* | |
d4dc210f TH |
1321 | * Block event polling for write claims if requested. Any |
1322 | * write holder makes the write_holder state stick until | |
1323 | * all are released. This is good enough and tracking | |
1324 | * individual writeable reference is too fragile given the | |
1325 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1326 | */ |
4c49ff3f TH |
1327 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1328 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1329 | bdev->bd_write_holder = true; |
d4dc210f | 1330 | disk_block_events(disk); |
77ea887e TH |
1331 | } |
1332 | ||
1333 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1334 | bdput(whole); |
e525fd89 TH |
1335 | } |
1336 | ||
1337 | return res; | |
37be4124 | 1338 | } |
1da177e4 LT |
1339 | EXPORT_SYMBOL(blkdev_get); |
1340 | ||
d4d77629 TH |
1341 | /** |
1342 | * blkdev_get_by_path - open a block device by name | |
1343 | * @path: path to the block device to open | |
1344 | * @mode: FMODE_* mask | |
1345 | * @holder: exclusive holder identifier | |
1346 | * | |
1347 | * Open the blockdevice described by the device file at @path. @mode | |
1348 | * and @holder are identical to blkdev_get(). | |
1349 | * | |
1350 | * On success, the returned block_device has reference count of one. | |
1351 | * | |
1352 | * CONTEXT: | |
1353 | * Might sleep. | |
1354 | * | |
1355 | * RETURNS: | |
1356 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1357 | */ | |
1358 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1359 | void *holder) | |
1360 | { | |
1361 | struct block_device *bdev; | |
1362 | int err; | |
1363 | ||
1364 | bdev = lookup_bdev(path); | |
1365 | if (IS_ERR(bdev)) | |
1366 | return bdev; | |
1367 | ||
1368 | err = blkdev_get(bdev, mode, holder); | |
1369 | if (err) | |
1370 | return ERR_PTR(err); | |
1371 | ||
e51900f7 CE |
1372 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1373 | blkdev_put(bdev, mode); | |
1374 | return ERR_PTR(-EACCES); | |
1375 | } | |
1376 | ||
d4d77629 TH |
1377 | return bdev; |
1378 | } | |
1379 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1380 | ||
1381 | /** | |
1382 | * blkdev_get_by_dev - open a block device by device number | |
1383 | * @dev: device number of block device to open | |
1384 | * @mode: FMODE_* mask | |
1385 | * @holder: exclusive holder identifier | |
1386 | * | |
1387 | * Open the blockdevice described by device number @dev. @mode and | |
1388 | * @holder are identical to blkdev_get(). | |
1389 | * | |
1390 | * Use it ONLY if you really do not have anything better - i.e. when | |
1391 | * you are behind a truly sucky interface and all you are given is a | |
1392 | * device number. _Never_ to be used for internal purposes. If you | |
1393 | * ever need it - reconsider your API. | |
1394 | * | |
1395 | * On success, the returned block_device has reference count of one. | |
1396 | * | |
1397 | * CONTEXT: | |
1398 | * Might sleep. | |
1399 | * | |
1400 | * RETURNS: | |
1401 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1402 | */ | |
1403 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1404 | { | |
1405 | struct block_device *bdev; | |
1406 | int err; | |
1407 | ||
1408 | bdev = bdget(dev); | |
1409 | if (!bdev) | |
1410 | return ERR_PTR(-ENOMEM); | |
1411 | ||
1412 | err = blkdev_get(bdev, mode, holder); | |
1413 | if (err) | |
1414 | return ERR_PTR(err); | |
1415 | ||
1416 | return bdev; | |
1417 | } | |
1418 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1419 | ||
1da177e4 LT |
1420 | static int blkdev_open(struct inode * inode, struct file * filp) |
1421 | { | |
1422 | struct block_device *bdev; | |
1da177e4 LT |
1423 | |
1424 | /* | |
1425 | * Preserve backwards compatibility and allow large file access | |
1426 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1427 | * binary needs it. We might want to drop this workaround | |
1428 | * during an unstable branch. | |
1429 | */ | |
1430 | filp->f_flags |= O_LARGEFILE; | |
1431 | ||
572c4892 AV |
1432 | if (filp->f_flags & O_NDELAY) |
1433 | filp->f_mode |= FMODE_NDELAY; | |
1434 | if (filp->f_flags & O_EXCL) | |
1435 | filp->f_mode |= FMODE_EXCL; | |
1436 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1437 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1438 | ||
1da177e4 | 1439 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1440 | if (bdev == NULL) |
1441 | return -ENOMEM; | |
1da177e4 | 1442 | |
572c4892 AV |
1443 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1444 | ||
e525fd89 | 1445 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1446 | } |
1447 | ||
9a1c3542 | 1448 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d PZ |
1449 | { |
1450 | int ret = 0; | |
2e7b651d | 1451 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1452 | struct block_device *victim = NULL; |
2e7b651d | 1453 | |
6796bf54 | 1454 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1455 | if (for_part) |
1456 | bdev->bd_part_count--; | |
1457 | ||
2e7b651d | 1458 | if (!--bdev->bd_openers) { |
6a027eff | 1459 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1460 | sync_blockdev(bdev); |
1461 | kill_bdev(bdev); | |
94007751 N |
1462 | /* ->release can cause the old bdi to disappear, |
1463 | * so must switch it out first | |
1464 | */ | |
1465 | bdev_inode_switch_bdi(bdev->bd_inode, | |
1466 | &default_backing_dev_info); | |
2e7b651d PZ |
1467 | } |
1468 | if (bdev->bd_contains == bdev) { | |
1469 | if (disk->fops->release) | |
9a1c3542 | 1470 | ret = disk->fops->release(disk, mode); |
2e7b651d PZ |
1471 | } |
1472 | if (!bdev->bd_openers) { | |
1473 | struct module *owner = disk->fops->owner; | |
1474 | ||
0762b8bd TH |
1475 | disk_put_part(bdev->bd_part); |
1476 | bdev->bd_part = NULL; | |
2e7b651d | 1477 | bdev->bd_disk = NULL; |
37be4124 N |
1478 | if (bdev != bdev->bd_contains) |
1479 | victim = bdev->bd_contains; | |
2e7b651d | 1480 | bdev->bd_contains = NULL; |
523e1d39 TH |
1481 | |
1482 | put_disk(disk); | |
1483 | module_put(owner); | |
2e7b651d | 1484 | } |
2e7b651d PZ |
1485 | mutex_unlock(&bdev->bd_mutex); |
1486 | bdput(bdev); | |
37be4124 | 1487 | if (victim) |
9a1c3542 | 1488 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1489 | return ret; |
1490 | } | |
1491 | ||
9a1c3542 | 1492 | int blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1493 | { |
85ef06d1 TH |
1494 | mutex_lock(&bdev->bd_mutex); |
1495 | ||
e525fd89 | 1496 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1497 | bool bdev_free; |
1498 | ||
1499 | /* | |
1500 | * Release a claim on the device. The holder fields | |
1501 | * are protected with bdev_lock. bd_mutex is to | |
1502 | * synchronize disk_holder unlinking. | |
1503 | */ | |
6a027eff TH |
1504 | spin_lock(&bdev_lock); |
1505 | ||
1506 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1507 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1508 | ||
1509 | /* bd_contains might point to self, check in a separate step */ | |
1510 | if ((bdev_free = !bdev->bd_holders)) | |
1511 | bdev->bd_holder = NULL; | |
1512 | if (!bdev->bd_contains->bd_holders) | |
1513 | bdev->bd_contains->bd_holder = NULL; | |
1514 | ||
1515 | spin_unlock(&bdev_lock); | |
1516 | ||
77ea887e TH |
1517 | /* |
1518 | * If this was the last claim, remove holder link and | |
1519 | * unblock evpoll if it was a write holder. | |
1520 | */ | |
85ef06d1 TH |
1521 | if (bdev_free && bdev->bd_write_holder) { |
1522 | disk_unblock_events(bdev->bd_disk); | |
1523 | bdev->bd_write_holder = false; | |
77ea887e | 1524 | } |
6936217c | 1525 | } |
77ea887e | 1526 | |
85ef06d1 TH |
1527 | /* |
1528 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1529 | * event. This is to ensure detection of media removal commanded | |
1530 | * from userland - e.g. eject(1). | |
1531 | */ | |
1532 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1533 | ||
1534 | mutex_unlock(&bdev->bd_mutex); | |
1535 | ||
9a1c3542 | 1536 | return __blkdev_put(bdev, mode, 0); |
37be4124 | 1537 | } |
2e7b651d PZ |
1538 | EXPORT_SYMBOL(blkdev_put); |
1539 | ||
1da177e4 LT |
1540 | static int blkdev_close(struct inode * inode, struct file * filp) |
1541 | { | |
1542 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); | |
e525fd89 | 1543 | |
9a1c3542 | 1544 | return blkdev_put(bdev, filp->f_mode); |
1da177e4 LT |
1545 | } |
1546 | ||
bb93e3a5 | 1547 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1548 | { |
56b26add AV |
1549 | struct block_device *bdev = I_BDEV(file->f_mapping->host); |
1550 | fmode_t mode = file->f_mode; | |
fd4ce1ac CH |
1551 | |
1552 | /* | |
1553 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1554 | * to updated it before every ioctl. | |
1555 | */ | |
56b26add | 1556 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1557 | mode |= FMODE_NDELAY; |
1558 | else | |
1559 | mode &= ~FMODE_NDELAY; | |
1560 | ||
56b26add | 1561 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1562 | } |
1563 | ||
eef99380 CH |
1564 | /* |
1565 | * Write data to the block device. Only intended for the block device itself | |
1566 | * and the raw driver which basically is a fake block device. | |
1567 | * | |
1568 | * Does not take i_mutex for the write and thus is not for general purpose | |
1569 | * use. | |
1570 | */ | |
1571 | ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
1572 | unsigned long nr_segs, loff_t pos) | |
1573 | { | |
1574 | struct file *file = iocb->ki_filp; | |
1575 | ssize_t ret; | |
1576 | ||
1577 | BUG_ON(iocb->ki_pos != pos); | |
1578 | ||
1579 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); | |
1580 | if (ret > 0 || ret == -EIOCBQUEUED) { | |
1581 | ssize_t err; | |
1582 | ||
1583 | err = generic_write_sync(file, pos, ret); | |
1584 | if (err < 0 && ret > 0) | |
1585 | ret = err; | |
1586 | } | |
1587 | return ret; | |
1588 | } | |
1589 | EXPORT_SYMBOL_GPL(blkdev_aio_write); | |
1590 | ||
87d8fe1e TT |
1591 | /* |
1592 | * Try to release a page associated with block device when the system | |
1593 | * is under memory pressure. | |
1594 | */ | |
1595 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1596 | { | |
1597 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1598 | ||
1599 | if (super && super->s_op->bdev_try_to_free_page) | |
1600 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1601 | ||
1602 | return try_to_free_buffers(page); | |
1603 | } | |
1604 | ||
4c54ac62 | 1605 | static const struct address_space_operations def_blk_aops = { |
1da177e4 LT |
1606 | .readpage = blkdev_readpage, |
1607 | .writepage = blkdev_writepage, | |
6272b5a5 NP |
1608 | .write_begin = blkdev_write_begin, |
1609 | .write_end = blkdev_write_end, | |
1da177e4 | 1610 | .writepages = generic_writepages, |
87d8fe1e | 1611 | .releasepage = blkdev_releasepage, |
1da177e4 LT |
1612 | .direct_IO = blkdev_direct_IO, |
1613 | }; | |
1614 | ||
4b6f5d20 | 1615 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1616 | .open = blkdev_open, |
1617 | .release = blkdev_close, | |
1618 | .llseek = block_llseek, | |
543ade1f BP |
1619 | .read = do_sync_read, |
1620 | .write = do_sync_write, | |
1da177e4 | 1621 | .aio_read = generic_file_aio_read, |
eef99380 | 1622 | .aio_write = blkdev_aio_write, |
1da177e4 | 1623 | .mmap = generic_file_mmap, |
b1dd3b28 | 1624 | .fsync = blkdev_fsync, |
bb93e3a5 | 1625 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1626 | #ifdef CONFIG_COMPAT |
1627 | .compat_ioctl = compat_blkdev_ioctl, | |
1628 | #endif | |
7f9c51f0 JA |
1629 | .splice_read = generic_file_splice_read, |
1630 | .splice_write = generic_file_splice_write, | |
1da177e4 LT |
1631 | }; |
1632 | ||
1633 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1634 | { | |
1635 | int res; | |
1636 | mm_segment_t old_fs = get_fs(); | |
1637 | set_fs(KERNEL_DS); | |
56b26add | 1638 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1639 | set_fs(old_fs); |
1640 | return res; | |
1641 | } | |
1642 | ||
1643 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1644 | ||
1645 | /** | |
1646 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1647 | * @pathname: special file representing the block device |
1da177e4 | 1648 | * |
57d1b536 | 1649 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1650 | * namespace if possible and return it. Return ERR_PTR(error) |
1651 | * otherwise. | |
1652 | */ | |
421748ec | 1653 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1654 | { |
1655 | struct block_device *bdev; | |
1656 | struct inode *inode; | |
421748ec | 1657 | struct path path; |
1da177e4 LT |
1658 | int error; |
1659 | ||
421748ec | 1660 | if (!pathname || !*pathname) |
1da177e4 LT |
1661 | return ERR_PTR(-EINVAL); |
1662 | ||
421748ec | 1663 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1664 | if (error) |
1665 | return ERR_PTR(error); | |
1666 | ||
421748ec | 1667 | inode = path.dentry->d_inode; |
1da177e4 LT |
1668 | error = -ENOTBLK; |
1669 | if (!S_ISBLK(inode->i_mode)) | |
1670 | goto fail; | |
1671 | error = -EACCES; | |
421748ec | 1672 | if (path.mnt->mnt_flags & MNT_NODEV) |
1da177e4 LT |
1673 | goto fail; |
1674 | error = -ENOMEM; | |
1675 | bdev = bd_acquire(inode); | |
1676 | if (!bdev) | |
1677 | goto fail; | |
1678 | out: | |
421748ec | 1679 | path_put(&path); |
1da177e4 LT |
1680 | return bdev; |
1681 | fail: | |
1682 | bdev = ERR_PTR(error); | |
1683 | goto out; | |
1684 | } | |
d5686b44 | 1685 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 1686 | |
93b270f7 | 1687 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
1688 | { |
1689 | struct super_block *sb = get_super(bdev); | |
1690 | int res = 0; | |
1691 | ||
1692 | if (sb) { | |
1693 | /* | |
1694 | * no need to lock the super, get_super holds the | |
1695 | * read mutex so the filesystem cannot go away | |
1696 | * under us (->put_super runs with the write lock | |
1697 | * hold). | |
1698 | */ | |
1699 | shrink_dcache_sb(sb); | |
93b270f7 | 1700 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
1701 | drop_super(sb); |
1702 | } | |
f98393a6 | 1703 | invalidate_bdev(bdev); |
b71e8a4c DH |
1704 | return res; |
1705 | } | |
1706 | EXPORT_SYMBOL(__invalidate_device); |