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> | |
b502bd11 | 19 | #include <linux/magic.h> |
1da177e4 | 20 | #include <linux/buffer_head.h> |
ff01bb48 | 21 | #include <linux/swap.h> |
585d3bc0 | 22 | #include <linux/pagevec.h> |
811d736f | 23 | #include <linux/writeback.h> |
1da177e4 LT |
24 | #include <linux/mpage.h> |
25 | #include <linux/mount.h> | |
26 | #include <linux/uio.h> | |
27 | #include <linux/namei.h> | |
1368c4f2 | 28 | #include <linux/log2.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 | ||
564f00f6 CH |
51 | static void bdev_write_inode(struct inode *inode) |
52 | { | |
53 | spin_lock(&inode->i_lock); | |
54 | while (inode->i_state & I_DIRTY) { | |
55 | spin_unlock(&inode->i_lock); | |
56 | WARN_ON_ONCE(write_inode_now(inode, true)); | |
57 | spin_lock(&inode->i_lock); | |
58 | } | |
59 | spin_unlock(&inode->i_lock); | |
60 | } | |
61 | ||
f9a14399 | 62 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
ff01bb48 | 63 | void kill_bdev(struct block_device *bdev) |
1da177e4 | 64 | { |
ff01bb48 AV |
65 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
66 | ||
91b0abe3 | 67 | if (mapping->nrpages == 0 && mapping->nrshadows == 0) |
f9a14399 | 68 | return; |
ff01bb48 | 69 | |
f9a14399 | 70 | invalidate_bh_lrus(); |
ff01bb48 | 71 | truncate_inode_pages(mapping, 0); |
1da177e4 | 72 | } |
ff01bb48 AV |
73 | EXPORT_SYMBOL(kill_bdev); |
74 | ||
75 | /* Invalidate clean unused buffers and pagecache. */ | |
76 | void invalidate_bdev(struct block_device *bdev) | |
77 | { | |
78 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
79 | ||
80 | if (mapping->nrpages == 0) | |
81 | return; | |
82 | ||
83 | invalidate_bh_lrus(); | |
84 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
85 | invalidate_mapping_pages(mapping, 0, -1); | |
86 | /* 99% of the time, we don't need to flush the cleancache on the bdev. | |
87 | * But, for the strange corners, lets be cautious | |
88 | */ | |
3167760f | 89 | cleancache_invalidate_inode(mapping); |
ff01bb48 AV |
90 | } |
91 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 LT |
92 | |
93 | int set_blocksize(struct block_device *bdev, int size) | |
94 | { | |
95 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 96 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
97 | return -EINVAL; |
98 | ||
99 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 100 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
101 | return -EINVAL; |
102 | ||
103 | /* Don't change the size if it is same as current */ | |
104 | if (bdev->bd_block_size != size) { | |
105 | sync_blockdev(bdev); | |
106 | bdev->bd_block_size = size; | |
107 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
108 | kill_bdev(bdev); | |
109 | } | |
110 | return 0; | |
111 | } | |
112 | ||
113 | EXPORT_SYMBOL(set_blocksize); | |
114 | ||
115 | int sb_set_blocksize(struct super_block *sb, int size) | |
116 | { | |
1da177e4 LT |
117 | if (set_blocksize(sb->s_bdev, size)) |
118 | return 0; | |
119 | /* If we get here, we know size is power of two | |
120 | * and it's value is between 512 and PAGE_SIZE */ | |
121 | sb->s_blocksize = size; | |
38885bd4 | 122 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
123 | return sb->s_blocksize; |
124 | } | |
125 | ||
126 | EXPORT_SYMBOL(sb_set_blocksize); | |
127 | ||
128 | int sb_min_blocksize(struct super_block *sb, int size) | |
129 | { | |
e1defc4f | 130 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
131 | if (size < minsize) |
132 | size = minsize; | |
133 | return sb_set_blocksize(sb, size); | |
134 | } | |
135 | ||
136 | EXPORT_SYMBOL(sb_min_blocksize); | |
137 | ||
138 | static int | |
139 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
140 | struct buffer_head *bh, int create) | |
141 | { | |
1da177e4 LT |
142 | bh->b_bdev = I_BDEV(inode); |
143 | bh->b_blocknr = iblock; | |
144 | set_buffer_mapped(bh); | |
145 | return 0; | |
146 | } | |
147 | ||
b2e895db | 148 | static ssize_t |
d8d3d94b AV |
149 | blkdev_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter, |
150 | loff_t offset) | |
b2e895db AM |
151 | { |
152 | struct file *file = iocb->ki_filp; | |
153 | struct inode *inode = file->f_mapping->host; | |
154 | ||
31b14039 AV |
155 | return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iter, |
156 | offset, blkdev_get_block, | |
d8d3d94b | 157 | NULL, NULL, 0); |
b2e895db AM |
158 | } |
159 | ||
5cee5815 JK |
160 | int __sync_blockdev(struct block_device *bdev, int wait) |
161 | { | |
162 | if (!bdev) | |
163 | return 0; | |
164 | if (!wait) | |
165 | return filemap_flush(bdev->bd_inode->i_mapping); | |
166 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
167 | } | |
168 | ||
585d3bc0 NP |
169 | /* |
170 | * Write out and wait upon all the dirty data associated with a block | |
171 | * device via its mapping. Does not take the superblock lock. | |
172 | */ | |
173 | int sync_blockdev(struct block_device *bdev) | |
174 | { | |
5cee5815 | 175 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
176 | } |
177 | EXPORT_SYMBOL(sync_blockdev); | |
178 | ||
179 | /* | |
180 | * Write out and wait upon all dirty data associated with this | |
181 | * device. Filesystem data as well as the underlying block | |
182 | * device. Takes the superblock lock. | |
183 | */ | |
184 | int fsync_bdev(struct block_device *bdev) | |
185 | { | |
186 | struct super_block *sb = get_super(bdev); | |
187 | if (sb) { | |
60b0680f | 188 | int res = sync_filesystem(sb); |
585d3bc0 NP |
189 | drop_super(sb); |
190 | return res; | |
191 | } | |
192 | return sync_blockdev(bdev); | |
193 | } | |
47e4491b | 194 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
195 | |
196 | /** | |
197 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
198 | * @bdev: blockdevice to lock | |
199 | * | |
585d3bc0 NP |
200 | * If a superblock is found on this device, we take the s_umount semaphore |
201 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
202 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
203 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
204 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
205 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
206 | * actually. | |
207 | */ | |
208 | struct super_block *freeze_bdev(struct block_device *bdev) | |
209 | { | |
210 | struct super_block *sb; | |
211 | int error = 0; | |
212 | ||
213 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
214 | if (++bdev->bd_fsfreeze_count > 1) { |
215 | /* | |
216 | * We don't even need to grab a reference - the first call | |
217 | * to freeze_bdev grab an active reference and only the last | |
218 | * thaw_bdev drops it. | |
219 | */ | |
585d3bc0 | 220 | sb = get_super(bdev); |
4504230a CH |
221 | drop_super(sb); |
222 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
223 | return sb; | |
224 | } | |
225 | ||
226 | sb = get_active_super(bdev); | |
227 | if (!sb) | |
228 | goto out; | |
48b6bca6 BM |
229 | if (sb->s_op->freeze_super) |
230 | error = sb->s_op->freeze_super(sb); | |
231 | else | |
232 | error = freeze_super(sb); | |
18e9e510 JB |
233 | if (error) { |
234 | deactivate_super(sb); | |
235 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 236 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 237 | return ERR_PTR(error); |
585d3bc0 | 238 | } |
18e9e510 | 239 | deactivate_super(sb); |
4504230a | 240 | out: |
585d3bc0 NP |
241 | sync_blockdev(bdev); |
242 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 243 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
244 | } |
245 | EXPORT_SYMBOL(freeze_bdev); | |
246 | ||
247 | /** | |
248 | * thaw_bdev -- unlock filesystem | |
249 | * @bdev: blockdevice to unlock | |
250 | * @sb: associated superblock | |
251 | * | |
252 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
253 | */ | |
254 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
255 | { | |
4504230a | 256 | int error = -EINVAL; |
585d3bc0 NP |
257 | |
258 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 259 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 260 | goto out; |
4504230a CH |
261 | |
262 | error = 0; | |
263 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 264 | goto out; |
4504230a CH |
265 | |
266 | if (!sb) | |
18e9e510 | 267 | goto out; |
4504230a | 268 | |
48b6bca6 BM |
269 | if (sb->s_op->thaw_super) |
270 | error = sb->s_op->thaw_super(sb); | |
271 | else | |
272 | error = thaw_super(sb); | |
18e9e510 JB |
273 | if (error) { |
274 | bdev->bd_fsfreeze_count++; | |
275 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
276 | return error; | |
277 | } | |
278 | out: | |
585d3bc0 NP |
279 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
280 | return 0; | |
281 | } | |
282 | EXPORT_SYMBOL(thaw_bdev); | |
283 | ||
1da177e4 LT |
284 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
285 | { | |
286 | return block_write_full_page(page, blkdev_get_block, wbc); | |
287 | } | |
288 | ||
289 | static int blkdev_readpage(struct file * file, struct page * page) | |
290 | { | |
291 | return block_read_full_page(page, blkdev_get_block); | |
292 | } | |
293 | ||
447f05bb AM |
294 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
295 | struct list_head *pages, unsigned nr_pages) | |
296 | { | |
297 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
298 | } | |
299 | ||
6272b5a5 NP |
300 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
301 | loff_t pos, unsigned len, unsigned flags, | |
302 | struct page **pagep, void **fsdata) | |
1da177e4 | 303 | { |
155130a4 CH |
304 | return block_write_begin(mapping, pos, len, flags, pagep, |
305 | blkdev_get_block); | |
1da177e4 LT |
306 | } |
307 | ||
6272b5a5 NP |
308 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
309 | loff_t pos, unsigned len, unsigned copied, | |
310 | struct page *page, void *fsdata) | |
1da177e4 | 311 | { |
6272b5a5 NP |
312 | int ret; |
313 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
314 | ||
315 | unlock_page(page); | |
316 | page_cache_release(page); | |
317 | ||
318 | return ret; | |
1da177e4 LT |
319 | } |
320 | ||
321 | /* | |
322 | * private llseek: | |
496ad9aa | 323 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
324 | * so we compute the size by hand (just as in block_read/write above) |
325 | */ | |
965c8e59 | 326 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 LT |
327 | { |
328 | struct inode *bd_inode = file->f_mapping->host; | |
1da177e4 LT |
329 | loff_t retval; |
330 | ||
1b1dcc1b | 331 | mutex_lock(&bd_inode->i_mutex); |
5d48f3a2 | 332 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
1b1dcc1b | 333 | mutex_unlock(&bd_inode->i_mutex); |
1da177e4 LT |
334 | return retval; |
335 | } | |
336 | ||
02c24a82 | 337 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 338 | { |
b8af67e2 AB |
339 | struct inode *bd_inode = filp->f_mapping->host; |
340 | struct block_device *bdev = I_BDEV(bd_inode); | |
ab0a9735 | 341 | int error; |
da5aa861 RW |
342 | |
343 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
344 | if (error) | |
345 | return error; | |
ab0a9735 | 346 | |
b8af67e2 AB |
347 | /* |
348 | * There is no need to serialise calls to blkdev_issue_flush with | |
349 | * i_mutex and doing so causes performance issues with concurrent | |
350 | * O_SYNC writers to a block device. | |
351 | */ | |
dd3932ed | 352 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
353 | if (error == -EOPNOTSUPP) |
354 | error = 0; | |
b8af67e2 | 355 | |
ab0a9735 | 356 | return error; |
1da177e4 | 357 | } |
b1dd3b28 | 358 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 359 | |
47a191fd MW |
360 | /** |
361 | * bdev_read_page() - Start reading a page from a block device | |
362 | * @bdev: The device to read the page from | |
363 | * @sector: The offset on the device to read the page to (need not be aligned) | |
364 | * @page: The page to read | |
365 | * | |
366 | * On entry, the page should be locked. It will be unlocked when the page | |
367 | * has been read. If the block driver implements rw_page synchronously, | |
368 | * that will be true on exit from this function, but it need not be. | |
369 | * | |
370 | * Errors returned by this function are usually "soft", eg out of memory, or | |
371 | * queue full; callers should try a different route to read this page rather | |
372 | * than propagate an error back up the stack. | |
373 | * | |
374 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
375 | */ | |
376 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
377 | struct page *page) | |
378 | { | |
379 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
380 | if (!ops->rw_page) | |
381 | return -EOPNOTSUPP; | |
382 | return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ); | |
383 | } | |
384 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
385 | ||
386 | /** | |
387 | * bdev_write_page() - Start writing a page to a block device | |
388 | * @bdev: The device to write the page to | |
389 | * @sector: The offset on the device to write the page to (need not be aligned) | |
390 | * @page: The page to write | |
391 | * @wbc: The writeback_control for the write | |
392 | * | |
393 | * On entry, the page should be locked and not currently under writeback. | |
394 | * On exit, if the write started successfully, the page will be unlocked and | |
395 | * under writeback. If the write failed already (eg the driver failed to | |
396 | * queue the page to the device), the page will still be locked. If the | |
397 | * caller is a ->writepage implementation, it will need to unlock the page. | |
398 | * | |
399 | * Errors returned by this function are usually "soft", eg out of memory, or | |
400 | * queue full; callers should try a different route to write this page rather | |
401 | * than propagate an error back up the stack. | |
402 | * | |
403 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
404 | */ | |
405 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
406 | struct page *page, struct writeback_control *wbc) | |
407 | { | |
408 | int result; | |
409 | int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE; | |
410 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
411 | if (!ops->rw_page) | |
412 | return -EOPNOTSUPP; | |
413 | set_page_writeback(page); | |
414 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw); | |
415 | if (result) | |
416 | end_page_writeback(page); | |
417 | else | |
418 | unlock_page(page); | |
419 | return result; | |
420 | } | |
421 | EXPORT_SYMBOL_GPL(bdev_write_page); | |
422 | ||
dd22f551 MW |
423 | /** |
424 | * bdev_direct_access() - Get the address for directly-accessibly memory | |
425 | * @bdev: The device containing the memory | |
426 | * @sector: The offset within the device | |
427 | * @addr: Where to put the address of the memory | |
428 | * @pfn: The Page Frame Number for the memory | |
429 | * @size: The number of bytes requested | |
430 | * | |
431 | * If a block device is made up of directly addressable memory, this function | |
432 | * will tell the caller the PFN and the address of the memory. The address | |
433 | * may be directly dereferenced within the kernel without the need to call | |
434 | * ioremap(), kmap() or similar. The PFN is suitable for inserting into | |
435 | * page tables. | |
436 | * | |
437 | * Return: negative errno if an error occurs, otherwise the number of bytes | |
438 | * accessible at this address. | |
439 | */ | |
440 | long bdev_direct_access(struct block_device *bdev, sector_t sector, | |
441 | void **addr, unsigned long *pfn, long size) | |
442 | { | |
443 | long avail; | |
444 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
445 | ||
446 | if (size < 0) | |
447 | return size; | |
448 | if (!ops->direct_access) | |
449 | return -EOPNOTSUPP; | |
450 | if ((sector + DIV_ROUND_UP(size, 512)) > | |
451 | part_nr_sects_read(bdev->bd_part)) | |
452 | return -ERANGE; | |
453 | sector += get_start_sect(bdev); | |
454 | if (sector % (PAGE_SIZE / 512)) | |
455 | return -EINVAL; | |
456 | avail = ops->direct_access(bdev, sector, addr, pfn, size); | |
457 | if (!avail) | |
458 | return -ERANGE; | |
459 | return min(avail, size); | |
460 | } | |
461 | EXPORT_SYMBOL_GPL(bdev_direct_access); | |
462 | ||
1da177e4 LT |
463 | /* |
464 | * pseudo-fs | |
465 | */ | |
466 | ||
467 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 468 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
469 | |
470 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
471 | { | |
e94b1766 | 472 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
473 | if (!ei) |
474 | return NULL; | |
475 | return &ei->vfs_inode; | |
476 | } | |
477 | ||
fa0d7e3d | 478 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 479 | { |
fa0d7e3d | 480 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
481 | struct bdev_inode *bdi = BDEV_I(inode); |
482 | ||
1da177e4 LT |
483 | kmem_cache_free(bdev_cachep, bdi); |
484 | } | |
485 | ||
fa0d7e3d NP |
486 | static void bdev_destroy_inode(struct inode *inode) |
487 | { | |
488 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
489 | } | |
490 | ||
51cc5068 | 491 | static void init_once(void *foo) |
1da177e4 LT |
492 | { |
493 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
494 | struct block_device *bdev = &ei->bdev; | |
495 | ||
a35afb83 CL |
496 | memset(bdev, 0, sizeof(*bdev)); |
497 | mutex_init(&bdev->bd_mutex); | |
a35afb83 CL |
498 | INIT_LIST_HEAD(&bdev->bd_inodes); |
499 | INIT_LIST_HEAD(&bdev->bd_list); | |
49731baa TH |
500 | #ifdef CONFIG_SYSFS |
501 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
502 | #endif | |
a35afb83 | 503 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
504 | /* Initialize mutex for freeze. */ |
505 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
506 | } |
507 | ||
508 | static inline void __bd_forget(struct inode *inode) | |
509 | { | |
510 | list_del_init(&inode->i_devices); | |
511 | inode->i_bdev = NULL; | |
512 | inode->i_mapping = &inode->i_data; | |
513 | } | |
514 | ||
b57922d9 | 515 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
516 | { |
517 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
518 | struct list_head *p; | |
91b0abe3 | 519 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 520 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 521 | clear_inode(inode); |
1da177e4 LT |
522 | spin_lock(&bdev_lock); |
523 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
524 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
525 | } | |
526 | list_del_init(&bdev->bd_list); | |
527 | spin_unlock(&bdev_lock); | |
528 | } | |
529 | ||
ee9b6d61 | 530 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
531 | .statfs = simple_statfs, |
532 | .alloc_inode = bdev_alloc_inode, | |
533 | .destroy_inode = bdev_destroy_inode, | |
534 | .drop_inode = generic_delete_inode, | |
b57922d9 | 535 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
536 | }; |
537 | ||
51139ada AV |
538 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
539 | int flags, const char *dev_name, void *data) | |
1da177e4 | 540 | { |
b502bd11 | 541 | return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); |
1da177e4 LT |
542 | } |
543 | ||
544 | static struct file_system_type bd_type = { | |
545 | .name = "bdev", | |
51139ada | 546 | .mount = bd_mount, |
1da177e4 LT |
547 | .kill_sb = kill_anon_super, |
548 | }; | |
549 | ||
f47ec3f2 | 550 | static struct super_block *blockdev_superblock __read_mostly; |
1da177e4 LT |
551 | |
552 | void __init bdev_cache_init(void) | |
553 | { | |
554 | int err; | |
ace8577a | 555 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 556 | |
1da177e4 | 557 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 PJ |
558 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
559 | SLAB_MEM_SPREAD|SLAB_PANIC), | |
20c2df83 | 560 | init_once); |
1da177e4 LT |
561 | err = register_filesystem(&bd_type); |
562 | if (err) | |
563 | panic("Cannot register bdev pseudo-fs"); | |
564 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
565 | if (IS_ERR(bd_mnt)) |
566 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 567 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
568 | } |
569 | ||
570 | /* | |
571 | * Most likely _very_ bad one - but then it's hardly critical for small | |
572 | * /dev and can be fixed when somebody will need really large one. | |
573 | * Keep in mind that it will be fed through icache hash function too. | |
574 | */ | |
575 | static inline unsigned long hash(dev_t dev) | |
576 | { | |
577 | return MAJOR(dev)+MINOR(dev); | |
578 | } | |
579 | ||
580 | static int bdev_test(struct inode *inode, void *data) | |
581 | { | |
582 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
583 | } | |
584 | ||
585 | static int bdev_set(struct inode *inode, void *data) | |
586 | { | |
587 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
588 | return 0; | |
589 | } | |
590 | ||
591 | static LIST_HEAD(all_bdevs); | |
592 | ||
593 | struct block_device *bdget(dev_t dev) | |
594 | { | |
595 | struct block_device *bdev; | |
596 | struct inode *inode; | |
597 | ||
c2acf7b9 | 598 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
599 | bdev_test, bdev_set, &dev); |
600 | ||
601 | if (!inode) | |
602 | return NULL; | |
603 | ||
604 | bdev = &BDEV_I(inode)->bdev; | |
605 | ||
606 | if (inode->i_state & I_NEW) { | |
607 | bdev->bd_contains = NULL; | |
782b94cd | 608 | bdev->bd_super = NULL; |
1da177e4 LT |
609 | bdev->bd_inode = inode; |
610 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
611 | bdev->bd_part_count = 0; | |
612 | bdev->bd_invalidated = 0; | |
613 | inode->i_mode = S_IFBLK; | |
614 | inode->i_rdev = dev; | |
615 | inode->i_bdev = bdev; | |
616 | inode->i_data.a_ops = &def_blk_aops; | |
617 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
618 | spin_lock(&bdev_lock); |
619 | list_add(&bdev->bd_list, &all_bdevs); | |
620 | spin_unlock(&bdev_lock); | |
621 | unlock_new_inode(inode); | |
622 | } | |
623 | return bdev; | |
624 | } | |
625 | ||
626 | EXPORT_SYMBOL(bdget); | |
627 | ||
dddac6a7 AJ |
628 | /** |
629 | * bdgrab -- Grab a reference to an already referenced block device | |
630 | * @bdev: Block device to grab a reference to. | |
631 | */ | |
632 | struct block_device *bdgrab(struct block_device *bdev) | |
633 | { | |
7de9c6ee | 634 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
635 | return bdev; |
636 | } | |
c1681bf8 | 637 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 638 | |
1da177e4 LT |
639 | long nr_blockdev_pages(void) |
640 | { | |
203a2935 | 641 | struct block_device *bdev; |
1da177e4 LT |
642 | long ret = 0; |
643 | spin_lock(&bdev_lock); | |
203a2935 | 644 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
645 | ret += bdev->bd_inode->i_mapping->nrpages; |
646 | } | |
647 | spin_unlock(&bdev_lock); | |
648 | return ret; | |
649 | } | |
650 | ||
651 | void bdput(struct block_device *bdev) | |
652 | { | |
653 | iput(bdev->bd_inode); | |
654 | } | |
655 | ||
656 | EXPORT_SYMBOL(bdput); | |
657 | ||
658 | static struct block_device *bd_acquire(struct inode *inode) | |
659 | { | |
660 | struct block_device *bdev; | |
09d967c6 | 661 | |
1da177e4 LT |
662 | spin_lock(&bdev_lock); |
663 | bdev = inode->i_bdev; | |
09d967c6 | 664 | if (bdev) { |
7de9c6ee | 665 | ihold(bdev->bd_inode); |
1da177e4 LT |
666 | spin_unlock(&bdev_lock); |
667 | return bdev; | |
668 | } | |
669 | spin_unlock(&bdev_lock); | |
09d967c6 | 670 | |
1da177e4 LT |
671 | bdev = bdget(inode->i_rdev); |
672 | if (bdev) { | |
673 | spin_lock(&bdev_lock); | |
09d967c6 OH |
674 | if (!inode->i_bdev) { |
675 | /* | |
7de9c6ee | 676 | * We take an additional reference to bd_inode, |
09d967c6 OH |
677 | * and it's released in clear_inode() of inode. |
678 | * So, we can access it via ->i_mapping always | |
679 | * without igrab(). | |
680 | */ | |
7de9c6ee | 681 | ihold(bdev->bd_inode); |
09d967c6 OH |
682 | inode->i_bdev = bdev; |
683 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
684 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
685 | } | |
1da177e4 LT |
686 | spin_unlock(&bdev_lock); |
687 | } | |
688 | return bdev; | |
689 | } | |
690 | ||
a8855990 | 691 | int sb_is_blkdev_sb(struct super_block *sb) |
f47ec3f2 AV |
692 | { |
693 | return sb == blockdev_superblock; | |
694 | } | |
695 | ||
1da177e4 LT |
696 | /* Call when you free inode */ |
697 | ||
698 | void bd_forget(struct inode *inode) | |
699 | { | |
09d967c6 OH |
700 | struct block_device *bdev = NULL; |
701 | ||
1da177e4 | 702 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
703 | if (!sb_is_blkdev_sb(inode->i_sb)) |
704 | bdev = inode->i_bdev; | |
705 | __bd_forget(inode); | |
1da177e4 | 706 | spin_unlock(&bdev_lock); |
09d967c6 OH |
707 | |
708 | if (bdev) | |
709 | iput(bdev->bd_inode); | |
1da177e4 LT |
710 | } |
711 | ||
1a3cbbc5 TH |
712 | /** |
713 | * bd_may_claim - test whether a block device can be claimed | |
714 | * @bdev: block device of interest | |
715 | * @whole: whole block device containing @bdev, may equal @bdev | |
716 | * @holder: holder trying to claim @bdev | |
717 | * | |
25985edc | 718 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
719 | * |
720 | * CONTEXT: | |
721 | * spin_lock(&bdev_lock). | |
722 | * | |
723 | * RETURNS: | |
724 | * %true if @bdev can be claimed, %false otherwise. | |
725 | */ | |
726 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
727 | void *holder) | |
1da177e4 | 728 | { |
1da177e4 | 729 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 730 | return true; /* already a holder */ |
1da177e4 | 731 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 732 | return false; /* held by someone else */ |
1da177e4 | 733 | else if (bdev->bd_contains == bdev) |
1a3cbbc5 | 734 | return true; /* is a whole device which isn't held */ |
1da177e4 | 735 | |
e525fd89 | 736 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
737 | return true; /* is a partition of a device that is being partitioned */ |
738 | else if (whole->bd_holder != NULL) | |
739 | return false; /* is a partition of a held device */ | |
1da177e4 | 740 | else |
1a3cbbc5 TH |
741 | return true; /* is a partition of an un-held device */ |
742 | } | |
743 | ||
6b4517a7 TH |
744 | /** |
745 | * bd_prepare_to_claim - prepare to claim a block device | |
746 | * @bdev: block device of interest | |
747 | * @whole: the whole device containing @bdev, may equal @bdev | |
748 | * @holder: holder trying to claim @bdev | |
749 | * | |
750 | * Prepare to claim @bdev. This function fails if @bdev is already | |
751 | * claimed by another holder and waits if another claiming is in | |
752 | * progress. This function doesn't actually claim. On successful | |
753 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
754 | * | |
755 | * CONTEXT: | |
756 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
757 | * it multiple times. | |
758 | * | |
759 | * RETURNS: | |
760 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
761 | */ | |
762 | static int bd_prepare_to_claim(struct block_device *bdev, | |
763 | struct block_device *whole, void *holder) | |
764 | { | |
765 | retry: | |
766 | /* if someone else claimed, fail */ | |
767 | if (!bd_may_claim(bdev, whole, holder)) | |
768 | return -EBUSY; | |
769 | ||
e75aa858 TH |
770 | /* if claiming is already in progress, wait for it to finish */ |
771 | if (whole->bd_claiming) { | |
6b4517a7 TH |
772 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
773 | DEFINE_WAIT(wait); | |
774 | ||
775 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
776 | spin_unlock(&bdev_lock); | |
777 | schedule(); | |
778 | finish_wait(wq, &wait); | |
779 | spin_lock(&bdev_lock); | |
780 | goto retry; | |
781 | } | |
782 | ||
783 | /* yay, all mine */ | |
784 | return 0; | |
785 | } | |
786 | ||
787 | /** | |
788 | * bd_start_claiming - start claiming a block device | |
789 | * @bdev: block device of interest | |
790 | * @holder: holder trying to claim @bdev | |
791 | * | |
792 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
793 | * exclusively and mark that an exclusive open is in progress. Each | |
794 | * successful call to this function must be matched with a call to | |
b0018361 NP |
795 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
796 | * fail). | |
797 | * | |
798 | * This function is used to gain exclusive access to the block device | |
799 | * without actually causing other exclusive open attempts to fail. It | |
800 | * should be used when the open sequence itself requires exclusive | |
801 | * access but may subsequently fail. | |
6b4517a7 TH |
802 | * |
803 | * CONTEXT: | |
804 | * Might sleep. | |
805 | * | |
806 | * RETURNS: | |
807 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
808 | * value on failure. | |
809 | */ | |
810 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
811 | void *holder) | |
812 | { | |
813 | struct gendisk *disk; | |
814 | struct block_device *whole; | |
815 | int partno, err; | |
816 | ||
817 | might_sleep(); | |
818 | ||
819 | /* | |
820 | * @bdev might not have been initialized properly yet, look up | |
821 | * and grab the outer block device the hard way. | |
822 | */ | |
823 | disk = get_gendisk(bdev->bd_dev, &partno); | |
824 | if (!disk) | |
825 | return ERR_PTR(-ENXIO); | |
826 | ||
d4c208b8 TH |
827 | /* |
828 | * Normally, @bdev should equal what's returned from bdget_disk() | |
829 | * if partno is 0; however, some drivers (floppy) use multiple | |
830 | * bdev's for the same physical device and @bdev may be one of the | |
831 | * aliases. Keep @bdev if partno is 0. This means claimer | |
832 | * tracking is broken for those devices but it has always been that | |
833 | * way. | |
834 | */ | |
835 | if (partno) | |
836 | whole = bdget_disk(disk, 0); | |
837 | else | |
838 | whole = bdgrab(bdev); | |
839 | ||
cf342570 | 840 | module_put(disk->fops->owner); |
6b4517a7 TH |
841 | put_disk(disk); |
842 | if (!whole) | |
843 | return ERR_PTR(-ENOMEM); | |
844 | ||
845 | /* prepare to claim, if successful, mark claiming in progress */ | |
846 | spin_lock(&bdev_lock); | |
847 | ||
848 | err = bd_prepare_to_claim(bdev, whole, holder); | |
849 | if (err == 0) { | |
850 | whole->bd_claiming = holder; | |
851 | spin_unlock(&bdev_lock); | |
852 | return whole; | |
853 | } else { | |
854 | spin_unlock(&bdev_lock); | |
855 | bdput(whole); | |
856 | return ERR_PTR(err); | |
857 | } | |
858 | } | |
859 | ||
641dc636 | 860 | #ifdef CONFIG_SYSFS |
49731baa TH |
861 | struct bd_holder_disk { |
862 | struct list_head list; | |
863 | struct gendisk *disk; | |
864 | int refcnt; | |
865 | }; | |
866 | ||
867 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
868 | struct gendisk *disk) | |
869 | { | |
870 | struct bd_holder_disk *holder; | |
871 | ||
872 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
873 | if (holder->disk == disk) | |
874 | return holder; | |
875 | return NULL; | |
876 | } | |
877 | ||
4d7dd8fd | 878 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 879 | { |
4d7dd8fd | 880 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
881 | } |
882 | ||
883 | static void del_symlink(struct kobject *from, struct kobject *to) | |
884 | { | |
641dc636 JN |
885 | sysfs_remove_link(from, kobject_name(to)); |
886 | } | |
887 | ||
df6c0cd9 | 888 | /** |
e09b457b TH |
889 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
890 | * @bdev: the claimed slave bdev | |
891 | * @disk: the holding disk | |
df6c0cd9 | 892 | * |
49731baa TH |
893 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
894 | * | |
e09b457b | 895 | * This functions creates the following sysfs symlinks. |
641dc636 | 896 | * |
e09b457b TH |
897 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
898 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 899 | * |
e09b457b TH |
900 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
901 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 902 | * |
e09b457b TH |
903 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
904 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 905 | * |
e09b457b TH |
906 | * The caller must have claimed @bdev before calling this function and |
907 | * ensure that both @bdev and @disk are valid during the creation and | |
908 | * lifetime of these symlinks. | |
641dc636 | 909 | * |
e09b457b TH |
910 | * CONTEXT: |
911 | * Might sleep. | |
641dc636 | 912 | * |
e09b457b TH |
913 | * RETURNS: |
914 | * 0 on success, -errno on failure. | |
641dc636 | 915 | */ |
e09b457b | 916 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 917 | { |
49731baa | 918 | struct bd_holder_disk *holder; |
e09b457b | 919 | int ret = 0; |
641dc636 | 920 | |
2e7b651d | 921 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 922 | |
49731baa | 923 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 924 | |
e09b457b TH |
925 | /* FIXME: remove the following once add_disk() handles errors */ |
926 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
927 | goto out_unlock; | |
4e91672c | 928 | |
49731baa TH |
929 | holder = bd_find_holder_disk(bdev, disk); |
930 | if (holder) { | |
931 | holder->refcnt++; | |
e09b457b | 932 | goto out_unlock; |
49731baa | 933 | } |
641dc636 | 934 | |
49731baa TH |
935 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
936 | if (!holder) { | |
937 | ret = -ENOMEM; | |
e09b457b TH |
938 | goto out_unlock; |
939 | } | |
641dc636 | 940 | |
49731baa TH |
941 | INIT_LIST_HEAD(&holder->list); |
942 | holder->disk = disk; | |
943 | holder->refcnt = 1; | |
944 | ||
945 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
946 | if (ret) | |
947 | goto out_free; | |
948 | ||
949 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
950 | if (ret) | |
951 | goto out_del; | |
e7407d16 TH |
952 | /* |
953 | * bdev could be deleted beneath us which would implicitly destroy | |
954 | * the holder directory. Hold on to it. | |
955 | */ | |
956 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
957 | |
958 | list_add(&holder->list, &bdev->bd_holder_disks); | |
959 | goto out_unlock; | |
960 | ||
961 | out_del: | |
962 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
963 | out_free: | |
964 | kfree(holder); | |
e09b457b | 965 | out_unlock: |
b4cf1b72 | 966 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 967 | return ret; |
641dc636 | 968 | } |
e09b457b | 969 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 970 | |
49731baa TH |
971 | /** |
972 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
973 | * @bdev: the calimed slave bdev | |
974 | * @disk: the holding disk | |
975 | * | |
976 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
977 | * | |
978 | * CONTEXT: | |
979 | * Might sleep. | |
980 | */ | |
981 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 982 | { |
49731baa | 983 | struct bd_holder_disk *holder; |
641dc636 | 984 | |
49731baa | 985 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 986 | |
49731baa TH |
987 | holder = bd_find_holder_disk(bdev, disk); |
988 | ||
989 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
990 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
991 | del_symlink(bdev->bd_part->holder_dir, | |
992 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 993 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
994 | list_del_init(&holder->list); |
995 | kfree(holder); | |
996 | } | |
997 | ||
998 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 999 | } |
49731baa | 1000 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1001 | #endif |
1da177e4 | 1002 | |
56ade44b AP |
1003 | /** |
1004 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1005 | * | |
1006 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1007 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1008 | * |
1009 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1010 | * when a disk has been changed -- either by a media change or online | |
1011 | * resize. | |
1012 | */ | |
93b270f7 | 1013 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1014 | { |
93b270f7 | 1015 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b AP |
1016 | char name[BDEVNAME_SIZE] = ""; |
1017 | ||
1018 | if (bdev->bd_disk) | |
1019 | disk_name(bdev->bd_disk, 0, name); | |
1020 | printk(KERN_WARNING "VFS: busy inodes on changed media or " | |
1021 | "resized disk %s\n", name); | |
1022 | } | |
1023 | ||
1024 | if (!bdev->bd_disk) | |
1025 | return; | |
d27769ec | 1026 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1027 | bdev->bd_invalidated = 1; |
1028 | } | |
1029 | ||
c3279d14 | 1030 | /** |
57d1b536 | 1031 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1032 | * @disk: struct gendisk to check |
1033 | * @bdev: struct bdev to adjust. | |
1034 | * | |
1035 | * This routine checks to see if the bdev size does not match the disk size | |
1036 | * and adjusts it if it differs. | |
1037 | */ | |
1038 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1039 | { | |
1040 | loff_t disk_size, bdev_size; | |
1041 | ||
1042 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1043 | bdev_size = i_size_read(bdev->bd_inode); | |
1044 | if (disk_size != bdev_size) { | |
1045 | char name[BDEVNAME_SIZE]; | |
1046 | ||
1047 | disk_name(disk, 0, name); | |
1048 | printk(KERN_INFO | |
1049 | "%s: detected capacity change from %lld to %lld\n", | |
1050 | name, bdev_size, disk_size); | |
1051 | i_size_write(bdev->bd_inode, disk_size); | |
93b270f7 | 1052 | flush_disk(bdev, false); |
c3279d14 AP |
1053 | } |
1054 | } | |
1055 | EXPORT_SYMBOL(check_disk_size_change); | |
1056 | ||
0c002c2f | 1057 | /** |
57d1b536 | 1058 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1059 | * @disk: struct gendisk to be revalidated |
1060 | * | |
1061 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1062 | * call-backs. It is used to do common pre and post operations needed | |
1063 | * for all revalidate_disk operations. | |
1064 | */ | |
1065 | int revalidate_disk(struct gendisk *disk) | |
1066 | { | |
c3279d14 | 1067 | struct block_device *bdev; |
0c002c2f AP |
1068 | int ret = 0; |
1069 | ||
1070 | if (disk->fops->revalidate_disk) | |
1071 | ret = disk->fops->revalidate_disk(disk); | |
1072 | ||
c3279d14 AP |
1073 | bdev = bdget_disk(disk, 0); |
1074 | if (!bdev) | |
1075 | return ret; | |
1076 | ||
1077 | mutex_lock(&bdev->bd_mutex); | |
1078 | check_disk_size_change(disk, bdev); | |
7630b661 | 1079 | bdev->bd_invalidated = 0; |
c3279d14 AP |
1080 | mutex_unlock(&bdev->bd_mutex); |
1081 | bdput(bdev); | |
0c002c2f AP |
1082 | return ret; |
1083 | } | |
1084 | EXPORT_SYMBOL(revalidate_disk); | |
1085 | ||
1da177e4 LT |
1086 | /* |
1087 | * This routine checks whether a removable media has been changed, | |
1088 | * and invalidates all buffer-cache-entries in that case. This | |
1089 | * is a relatively slow routine, so we have to try to minimize using | |
1090 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1091 | * is the best way of combining speed and utility, I think. | |
1092 | * People changing diskettes in the middle of an operation deserve | |
1093 | * to lose :-) | |
1094 | */ | |
1095 | int check_disk_change(struct block_device *bdev) | |
1096 | { | |
1097 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1098 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1099 | unsigned int events; |
1da177e4 | 1100 | |
77ea887e TH |
1101 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1102 | DISK_EVENT_EJECT_REQUEST); | |
1103 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1104 | return 0; |
1105 | ||
93b270f7 | 1106 | flush_disk(bdev, true); |
1da177e4 LT |
1107 | if (bdops->revalidate_disk) |
1108 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1109 | return 1; |
1110 | } | |
1111 | ||
1112 | EXPORT_SYMBOL(check_disk_change); | |
1113 | ||
1114 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1115 | { | |
e1defc4f | 1116 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1117 | |
d646a02a GC |
1118 | mutex_lock(&bdev->bd_inode->i_mutex); |
1119 | i_size_write(bdev->bd_inode, size); | |
1120 | mutex_unlock(&bdev->bd_inode->i_mutex); | |
1da177e4 LT |
1121 | while (bsize < PAGE_CACHE_SIZE) { |
1122 | if (size & bsize) | |
1123 | break; | |
1124 | bsize <<= 1; | |
1125 | } | |
1126 | bdev->bd_block_size = bsize; | |
1127 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1128 | } | |
1129 | EXPORT_SYMBOL(bd_set_size); | |
1130 | ||
4385bab1 | 1131 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1132 | |
6d740cd5 PZ |
1133 | /* |
1134 | * bd_mutex locking: | |
1135 | * | |
1136 | * mutex_lock(part->bd_mutex) | |
1137 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1138 | */ | |
1139 | ||
572c4892 | 1140 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1141 | { |
1da177e4 | 1142 | struct gendisk *disk; |
523e1d39 | 1143 | struct module *owner; |
7db9cfd3 | 1144 | int ret; |
cf771cb5 | 1145 | int partno; |
fe6e9c1f AV |
1146 | int perm = 0; |
1147 | ||
572c4892 | 1148 | if (mode & FMODE_READ) |
fe6e9c1f | 1149 | perm |= MAY_READ; |
572c4892 | 1150 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1151 | perm |= MAY_WRITE; |
1152 | /* | |
1153 | * hooks: /n/, see "layering violations". | |
1154 | */ | |
b7300b78 CW |
1155 | if (!for_part) { |
1156 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1157 | if (ret != 0) { | |
1158 | bdput(bdev); | |
1159 | return ret; | |
1160 | } | |
82666020 | 1161 | } |
7db9cfd3 | 1162 | |
d3374825 | 1163 | restart: |
0762b8bd | 1164 | |
89f97496 | 1165 | ret = -ENXIO; |
cf771cb5 | 1166 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1167 | if (!disk) |
6e9624b8 | 1168 | goto out; |
523e1d39 | 1169 | owner = disk->fops->owner; |
1da177e4 | 1170 | |
69e02c59 | 1171 | disk_block_events(disk); |
6796bf54 | 1172 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1173 | if (!bdev->bd_openers) { |
1174 | bdev->bd_disk = disk; | |
87192a2a | 1175 | bdev->bd_queue = disk->queue; |
1da177e4 | 1176 | bdev->bd_contains = bdev; |
cf771cb5 | 1177 | if (!partno) { |
89f97496 TH |
1178 | ret = -ENXIO; |
1179 | bdev->bd_part = disk_get_part(disk, partno); | |
1180 | if (!bdev->bd_part) | |
1181 | goto out_clear; | |
1182 | ||
1196f8b8 | 1183 | ret = 0; |
1da177e4 | 1184 | if (disk->fops->open) { |
572c4892 | 1185 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1186 | if (ret == -ERESTARTSYS) { |
1187 | /* Lost a race with 'disk' being | |
1188 | * deleted, try again. | |
1189 | * See md.c | |
1190 | */ | |
1191 | disk_put_part(bdev->bd_part); | |
1192 | bdev->bd_part = NULL; | |
d3374825 | 1193 | bdev->bd_disk = NULL; |
87192a2a | 1194 | bdev->bd_queue = NULL; |
d3374825 | 1195 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1196 | disk_unblock_events(disk); |
69e02c59 | 1197 | put_disk(disk); |
523e1d39 | 1198 | module_put(owner); |
d3374825 N |
1199 | goto restart; |
1200 | } | |
1da177e4 | 1201 | } |
7e69723f | 1202 | |
b83ae6d4 | 1203 | if (!ret) |
7e69723f | 1204 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
7e69723f | 1205 | |
1196f8b8 TH |
1206 | /* |
1207 | * If the device is invalidated, rescan partition | |
1208 | * if open succeeded or failed with -ENOMEDIUM. | |
1209 | * The latter is necessary to prevent ghost | |
1210 | * partitions on a removed medium. | |
1211 | */ | |
fe316bf2 JN |
1212 | if (bdev->bd_invalidated) { |
1213 | if (!ret) | |
1214 | rescan_partitions(disk, bdev); | |
1215 | else if (ret == -ENOMEDIUM) | |
1216 | invalidate_partitions(disk, bdev); | |
1217 | } | |
1196f8b8 TH |
1218 | if (ret) |
1219 | goto out_clear; | |
1da177e4 | 1220 | } else { |
1da177e4 LT |
1221 | struct block_device *whole; |
1222 | whole = bdget_disk(disk, 0); | |
1223 | ret = -ENOMEM; | |
1224 | if (!whole) | |
0762b8bd | 1225 | goto out_clear; |
37be4124 | 1226 | BUG_ON(for_part); |
572c4892 | 1227 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1228 | if (ret) |
0762b8bd | 1229 | goto out_clear; |
1da177e4 | 1230 | bdev->bd_contains = whole; |
89f97496 | 1231 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1232 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1233 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1234 | ret = -ENXIO; |
0762b8bd | 1235 | goto out_clear; |
1da177e4 | 1236 | } |
89f97496 | 1237 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1da177e4 LT |
1238 | } |
1239 | } else { | |
1da177e4 | 1240 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1241 | ret = 0; |
1242 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1243 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1244 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1245 | if (bdev->bd_invalidated) { |
1246 | if (!ret) | |
1247 | rescan_partitions(bdev->bd_disk, bdev); | |
1248 | else if (ret == -ENOMEDIUM) | |
1249 | invalidate_partitions(bdev->bd_disk, bdev); | |
1250 | } | |
1196f8b8 TH |
1251 | if (ret) |
1252 | goto out_unlock_bdev; | |
1da177e4 | 1253 | } |
69e02c59 | 1254 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1255 | put_disk(disk); |
523e1d39 | 1256 | module_put(owner); |
1da177e4 LT |
1257 | } |
1258 | bdev->bd_openers++; | |
37be4124 N |
1259 | if (for_part) |
1260 | bdev->bd_part_count++; | |
c039e313 | 1261 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1262 | disk_unblock_events(disk); |
1da177e4 LT |
1263 | return 0; |
1264 | ||
0762b8bd | 1265 | out_clear: |
89f97496 | 1266 | disk_put_part(bdev->bd_part); |
1da177e4 | 1267 | bdev->bd_disk = NULL; |
0762b8bd | 1268 | bdev->bd_part = NULL; |
87192a2a | 1269 | bdev->bd_queue = NULL; |
1da177e4 | 1270 | if (bdev != bdev->bd_contains) |
572c4892 | 1271 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1272 | bdev->bd_contains = NULL; |
0762b8bd | 1273 | out_unlock_bdev: |
c039e313 | 1274 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1275 | disk_unblock_events(disk); |
0762b8bd | 1276 | put_disk(disk); |
523e1d39 | 1277 | module_put(owner); |
4345caba | 1278 | out: |
0762b8bd TH |
1279 | bdput(bdev); |
1280 | ||
1da177e4 LT |
1281 | return ret; |
1282 | } | |
1283 | ||
d4d77629 TH |
1284 | /** |
1285 | * blkdev_get - open a block device | |
1286 | * @bdev: block_device to open | |
1287 | * @mode: FMODE_* mask | |
1288 | * @holder: exclusive holder identifier | |
1289 | * | |
1290 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1291 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1292 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1293 | * | |
1294 | * On success, the reference count of @bdev is unchanged. On failure, | |
1295 | * @bdev is put. | |
1296 | * | |
1297 | * CONTEXT: | |
1298 | * Might sleep. | |
1299 | * | |
1300 | * RETURNS: | |
1301 | * 0 on success, -errno on failure. | |
1302 | */ | |
e525fd89 | 1303 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1304 | { |
e525fd89 TH |
1305 | struct block_device *whole = NULL; |
1306 | int res; | |
1307 | ||
1308 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1309 | ||
1310 | if ((mode & FMODE_EXCL) && holder) { | |
1311 | whole = bd_start_claiming(bdev, holder); | |
1312 | if (IS_ERR(whole)) { | |
1313 | bdput(bdev); | |
1314 | return PTR_ERR(whole); | |
1315 | } | |
1316 | } | |
1317 | ||
1318 | res = __blkdev_get(bdev, mode, 0); | |
1319 | ||
1320 | if (whole) { | |
d4dc210f TH |
1321 | struct gendisk *disk = whole->bd_disk; |
1322 | ||
6a027eff | 1323 | /* finish claiming */ |
77ea887e | 1324 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1325 | spin_lock(&bdev_lock); |
1326 | ||
77ea887e | 1327 | if (!res) { |
6a027eff TH |
1328 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1329 | /* | |
1330 | * Note that for a whole device bd_holders | |
1331 | * will be incremented twice, and bd_holder | |
1332 | * will be set to bd_may_claim before being | |
1333 | * set to holder | |
1334 | */ | |
1335 | whole->bd_holders++; | |
1336 | whole->bd_holder = bd_may_claim; | |
1337 | bdev->bd_holders++; | |
1338 | bdev->bd_holder = holder; | |
1339 | } | |
1340 | ||
1341 | /* tell others that we're done */ | |
1342 | BUG_ON(whole->bd_claiming != holder); | |
1343 | whole->bd_claiming = NULL; | |
1344 | wake_up_bit(&whole->bd_claiming, 0); | |
1345 | ||
1346 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1347 | |
1348 | /* | |
d4dc210f TH |
1349 | * Block event polling for write claims if requested. Any |
1350 | * write holder makes the write_holder state stick until | |
1351 | * all are released. This is good enough and tracking | |
1352 | * individual writeable reference is too fragile given the | |
1353 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1354 | */ |
4c49ff3f TH |
1355 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1356 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1357 | bdev->bd_write_holder = true; |
d4dc210f | 1358 | disk_block_events(disk); |
77ea887e TH |
1359 | } |
1360 | ||
1361 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1362 | bdput(whole); |
e525fd89 TH |
1363 | } |
1364 | ||
1365 | return res; | |
37be4124 | 1366 | } |
1da177e4 LT |
1367 | EXPORT_SYMBOL(blkdev_get); |
1368 | ||
d4d77629 TH |
1369 | /** |
1370 | * blkdev_get_by_path - open a block device by name | |
1371 | * @path: path to the block device to open | |
1372 | * @mode: FMODE_* mask | |
1373 | * @holder: exclusive holder identifier | |
1374 | * | |
1375 | * Open the blockdevice described by the device file at @path. @mode | |
1376 | * and @holder are identical to blkdev_get(). | |
1377 | * | |
1378 | * On success, the returned block_device has reference count of one. | |
1379 | * | |
1380 | * CONTEXT: | |
1381 | * Might sleep. | |
1382 | * | |
1383 | * RETURNS: | |
1384 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1385 | */ | |
1386 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1387 | void *holder) | |
1388 | { | |
1389 | struct block_device *bdev; | |
1390 | int err; | |
1391 | ||
1392 | bdev = lookup_bdev(path); | |
1393 | if (IS_ERR(bdev)) | |
1394 | return bdev; | |
1395 | ||
1396 | err = blkdev_get(bdev, mode, holder); | |
1397 | if (err) | |
1398 | return ERR_PTR(err); | |
1399 | ||
e51900f7 CE |
1400 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1401 | blkdev_put(bdev, mode); | |
1402 | return ERR_PTR(-EACCES); | |
1403 | } | |
1404 | ||
d4d77629 TH |
1405 | return bdev; |
1406 | } | |
1407 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1408 | ||
1409 | /** | |
1410 | * blkdev_get_by_dev - open a block device by device number | |
1411 | * @dev: device number of block device to open | |
1412 | * @mode: FMODE_* mask | |
1413 | * @holder: exclusive holder identifier | |
1414 | * | |
1415 | * Open the blockdevice described by device number @dev. @mode and | |
1416 | * @holder are identical to blkdev_get(). | |
1417 | * | |
1418 | * Use it ONLY if you really do not have anything better - i.e. when | |
1419 | * you are behind a truly sucky interface and all you are given is a | |
1420 | * device number. _Never_ to be used for internal purposes. If you | |
1421 | * ever need it - reconsider your API. | |
1422 | * | |
1423 | * On success, the returned block_device has reference count of one. | |
1424 | * | |
1425 | * CONTEXT: | |
1426 | * Might sleep. | |
1427 | * | |
1428 | * RETURNS: | |
1429 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1430 | */ | |
1431 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1432 | { | |
1433 | struct block_device *bdev; | |
1434 | int err; | |
1435 | ||
1436 | bdev = bdget(dev); | |
1437 | if (!bdev) | |
1438 | return ERR_PTR(-ENOMEM); | |
1439 | ||
1440 | err = blkdev_get(bdev, mode, holder); | |
1441 | if (err) | |
1442 | return ERR_PTR(err); | |
1443 | ||
1444 | return bdev; | |
1445 | } | |
1446 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1447 | ||
1da177e4 LT |
1448 | static int blkdev_open(struct inode * inode, struct file * filp) |
1449 | { | |
1450 | struct block_device *bdev; | |
1da177e4 LT |
1451 | |
1452 | /* | |
1453 | * Preserve backwards compatibility and allow large file access | |
1454 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1455 | * binary needs it. We might want to drop this workaround | |
1456 | * during an unstable branch. | |
1457 | */ | |
1458 | filp->f_flags |= O_LARGEFILE; | |
1459 | ||
572c4892 AV |
1460 | if (filp->f_flags & O_NDELAY) |
1461 | filp->f_mode |= FMODE_NDELAY; | |
1462 | if (filp->f_flags & O_EXCL) | |
1463 | filp->f_mode |= FMODE_EXCL; | |
1464 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1465 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1466 | ||
1da177e4 | 1467 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1468 | if (bdev == NULL) |
1469 | return -ENOMEM; | |
1da177e4 | 1470 | |
572c4892 AV |
1471 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1472 | ||
e525fd89 | 1473 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1474 | } |
1475 | ||
4385bab1 | 1476 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1477 | { |
2e7b651d | 1478 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1479 | struct block_device *victim = NULL; |
2e7b651d | 1480 | |
6796bf54 | 1481 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1482 | if (for_part) |
1483 | bdev->bd_part_count--; | |
1484 | ||
2e7b651d | 1485 | if (!--bdev->bd_openers) { |
6a027eff | 1486 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1487 | sync_blockdev(bdev); |
1488 | kill_bdev(bdev); | |
564f00f6 CH |
1489 | /* |
1490 | * ->release can cause the queue to disappear, so flush all | |
1491 | * dirty data before. | |
94007751 | 1492 | */ |
564f00f6 | 1493 | bdev_write_inode(bdev->bd_inode); |
2e7b651d PZ |
1494 | } |
1495 | if (bdev->bd_contains == bdev) { | |
1496 | if (disk->fops->release) | |
db2a144b | 1497 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1498 | } |
1499 | if (!bdev->bd_openers) { | |
1500 | struct module *owner = disk->fops->owner; | |
1501 | ||
0762b8bd TH |
1502 | disk_put_part(bdev->bd_part); |
1503 | bdev->bd_part = NULL; | |
2e7b651d | 1504 | bdev->bd_disk = NULL; |
37be4124 N |
1505 | if (bdev != bdev->bd_contains) |
1506 | victim = bdev->bd_contains; | |
2e7b651d | 1507 | bdev->bd_contains = NULL; |
523e1d39 TH |
1508 | |
1509 | put_disk(disk); | |
1510 | module_put(owner); | |
2e7b651d | 1511 | } |
2e7b651d PZ |
1512 | mutex_unlock(&bdev->bd_mutex); |
1513 | bdput(bdev); | |
37be4124 | 1514 | if (victim) |
9a1c3542 | 1515 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1516 | } |
1517 | ||
4385bab1 | 1518 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1519 | { |
85ef06d1 TH |
1520 | mutex_lock(&bdev->bd_mutex); |
1521 | ||
e525fd89 | 1522 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1523 | bool bdev_free; |
1524 | ||
1525 | /* | |
1526 | * Release a claim on the device. The holder fields | |
1527 | * are protected with bdev_lock. bd_mutex is to | |
1528 | * synchronize disk_holder unlinking. | |
1529 | */ | |
6a027eff TH |
1530 | spin_lock(&bdev_lock); |
1531 | ||
1532 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1533 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1534 | ||
1535 | /* bd_contains might point to self, check in a separate step */ | |
1536 | if ((bdev_free = !bdev->bd_holders)) | |
1537 | bdev->bd_holder = NULL; | |
1538 | if (!bdev->bd_contains->bd_holders) | |
1539 | bdev->bd_contains->bd_holder = NULL; | |
1540 | ||
1541 | spin_unlock(&bdev_lock); | |
1542 | ||
77ea887e TH |
1543 | /* |
1544 | * If this was the last claim, remove holder link and | |
1545 | * unblock evpoll if it was a write holder. | |
1546 | */ | |
85ef06d1 TH |
1547 | if (bdev_free && bdev->bd_write_holder) { |
1548 | disk_unblock_events(bdev->bd_disk); | |
1549 | bdev->bd_write_holder = false; | |
77ea887e | 1550 | } |
6936217c | 1551 | } |
77ea887e | 1552 | |
85ef06d1 TH |
1553 | /* |
1554 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1555 | * event. This is to ensure detection of media removal commanded | |
1556 | * from userland - e.g. eject(1). | |
1557 | */ | |
1558 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1559 | ||
1560 | mutex_unlock(&bdev->bd_mutex); | |
1561 | ||
4385bab1 | 1562 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1563 | } |
2e7b651d PZ |
1564 | EXPORT_SYMBOL(blkdev_put); |
1565 | ||
1da177e4 LT |
1566 | static int blkdev_close(struct inode * inode, struct file * filp) |
1567 | { | |
1568 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); | |
4385bab1 AV |
1569 | blkdev_put(bdev, filp->f_mode); |
1570 | return 0; | |
1da177e4 LT |
1571 | } |
1572 | ||
bb93e3a5 | 1573 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1574 | { |
56b26add AV |
1575 | struct block_device *bdev = I_BDEV(file->f_mapping->host); |
1576 | fmode_t mode = file->f_mode; | |
fd4ce1ac CH |
1577 | |
1578 | /* | |
1579 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1580 | * to updated it before every ioctl. | |
1581 | */ | |
56b26add | 1582 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1583 | mode |= FMODE_NDELAY; |
1584 | else | |
1585 | mode &= ~FMODE_NDELAY; | |
1586 | ||
56b26add | 1587 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1588 | } |
1589 | ||
eef99380 CH |
1590 | /* |
1591 | * Write data to the block device. Only intended for the block device itself | |
1592 | * and the raw driver which basically is a fake block device. | |
1593 | * | |
1594 | * Does not take i_mutex for the write and thus is not for general purpose | |
1595 | * use. | |
1596 | */ | |
1456c0a8 | 1597 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1598 | { |
1599 | struct file *file = iocb->ki_filp; | |
53362a05 | 1600 | struct blk_plug plug; |
eef99380 CH |
1601 | ssize_t ret; |
1602 | ||
53362a05 | 1603 | blk_start_plug(&plug); |
1456c0a8 | 1604 | ret = __generic_file_write_iter(iocb, from); |
02afc27f | 1605 | if (ret > 0) { |
eef99380 | 1606 | ssize_t err; |
1456c0a8 | 1607 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
45d4f855 | 1608 | if (err < 0) |
eef99380 CH |
1609 | ret = err; |
1610 | } | |
53362a05 | 1611 | blk_finish_plug(&plug); |
eef99380 CH |
1612 | return ret; |
1613 | } | |
1456c0a8 | 1614 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1615 | |
b2de525f | 1616 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1617 | { |
1618 | struct file *file = iocb->ki_filp; | |
1619 | struct inode *bd_inode = file->f_mapping->host; | |
1620 | loff_t size = i_size_read(bd_inode); | |
a886038b | 1621 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1622 | |
1623 | if (pos >= size) | |
1624 | return 0; | |
1625 | ||
1626 | size -= pos; | |
a886038b AV |
1627 | iov_iter_truncate(to, size); |
1628 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1629 | } |
b2de525f | 1630 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1631 | |
87d8fe1e TT |
1632 | /* |
1633 | * Try to release a page associated with block device when the system | |
1634 | * is under memory pressure. | |
1635 | */ | |
1636 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1637 | { | |
1638 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1639 | ||
1640 | if (super && super->s_op->bdev_try_to_free_page) | |
1641 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1642 | ||
1643 | return try_to_free_buffers(page); | |
1644 | } | |
1645 | ||
4c54ac62 | 1646 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1647 | .readpage = blkdev_readpage, |
447f05bb | 1648 | .readpages = blkdev_readpages, |
1da177e4 | 1649 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1650 | .write_begin = blkdev_write_begin, |
1651 | .write_end = blkdev_write_end, | |
1da177e4 | 1652 | .writepages = generic_writepages, |
87d8fe1e | 1653 | .releasepage = blkdev_releasepage, |
1da177e4 | 1654 | .direct_IO = blkdev_direct_IO, |
b4597226 | 1655 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1656 | }; |
1657 | ||
4b6f5d20 | 1658 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1659 | .open = blkdev_open, |
1660 | .release = blkdev_close, | |
1661 | .llseek = block_llseek, | |
a886038b | 1662 | .read = new_sync_read, |
1456c0a8 | 1663 | .write = new_sync_write, |
a886038b | 1664 | .read_iter = blkdev_read_iter, |
1456c0a8 | 1665 | .write_iter = blkdev_write_iter, |
1e8b3332 | 1666 | .mmap = generic_file_mmap, |
b1dd3b28 | 1667 | .fsync = blkdev_fsync, |
bb93e3a5 | 1668 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1669 | #ifdef CONFIG_COMPAT |
1670 | .compat_ioctl = compat_blkdev_ioctl, | |
1671 | #endif | |
1e8b3332 | 1672 | .splice_read = generic_file_splice_read, |
8d020765 | 1673 | .splice_write = iter_file_splice_write, |
1da177e4 LT |
1674 | }; |
1675 | ||
1676 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1677 | { | |
1678 | int res; | |
1679 | mm_segment_t old_fs = get_fs(); | |
1680 | set_fs(KERNEL_DS); | |
56b26add | 1681 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1682 | set_fs(old_fs); |
1683 | return res; | |
1684 | } | |
1685 | ||
1686 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1687 | ||
1688 | /** | |
1689 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1690 | * @pathname: special file representing the block device |
1da177e4 | 1691 | * |
57d1b536 | 1692 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1693 | * namespace if possible and return it. Return ERR_PTR(error) |
1694 | * otherwise. | |
1695 | */ | |
421748ec | 1696 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1697 | { |
1698 | struct block_device *bdev; | |
1699 | struct inode *inode; | |
421748ec | 1700 | struct path path; |
1da177e4 LT |
1701 | int error; |
1702 | ||
421748ec | 1703 | if (!pathname || !*pathname) |
1da177e4 LT |
1704 | return ERR_PTR(-EINVAL); |
1705 | ||
421748ec | 1706 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1707 | if (error) |
1708 | return ERR_PTR(error); | |
1709 | ||
421748ec | 1710 | inode = path.dentry->d_inode; |
1da177e4 LT |
1711 | error = -ENOTBLK; |
1712 | if (!S_ISBLK(inode->i_mode)) | |
1713 | goto fail; | |
1714 | error = -EACCES; | |
421748ec | 1715 | if (path.mnt->mnt_flags & MNT_NODEV) |
1da177e4 LT |
1716 | goto fail; |
1717 | error = -ENOMEM; | |
1718 | bdev = bd_acquire(inode); | |
1719 | if (!bdev) | |
1720 | goto fail; | |
1721 | out: | |
421748ec | 1722 | path_put(&path); |
1da177e4 LT |
1723 | return bdev; |
1724 | fail: | |
1725 | bdev = ERR_PTR(error); | |
1726 | goto out; | |
1727 | } | |
d5686b44 | 1728 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 1729 | |
93b270f7 | 1730 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
1731 | { |
1732 | struct super_block *sb = get_super(bdev); | |
1733 | int res = 0; | |
1734 | ||
1735 | if (sb) { | |
1736 | /* | |
1737 | * no need to lock the super, get_super holds the | |
1738 | * read mutex so the filesystem cannot go away | |
1739 | * under us (->put_super runs with the write lock | |
1740 | * hold). | |
1741 | */ | |
1742 | shrink_dcache_sb(sb); | |
93b270f7 | 1743 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
1744 | drop_super(sb); |
1745 | } | |
f98393a6 | 1746 | invalidate_bdev(bdev); |
b71e8a4c DH |
1747 | return res; |
1748 | } | |
1749 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
1750 | |
1751 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
1752 | { | |
1753 | struct inode *inode, *old_inode = NULL; | |
1754 | ||
1755 | spin_lock(&inode_sb_list_lock); | |
1756 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { | |
1757 | struct address_space *mapping = inode->i_mapping; | |
1758 | ||
1759 | spin_lock(&inode->i_lock); | |
1760 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
1761 | mapping->nrpages == 0) { | |
1762 | spin_unlock(&inode->i_lock); | |
1763 | continue; | |
1764 | } | |
1765 | __iget(inode); | |
1766 | spin_unlock(&inode->i_lock); | |
1767 | spin_unlock(&inode_sb_list_lock); | |
1768 | /* | |
1769 | * We hold a reference to 'inode' so it couldn't have been | |
1770 | * removed from s_inodes list while we dropped the | |
1771 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1772 | * be holding the last reference and we cannot iput it under | |
1773 | * inode_sb_list_lock. So we keep the reference and iput it | |
1774 | * later. | |
1775 | */ | |
1776 | iput(old_inode); | |
1777 | old_inode = inode; | |
1778 | ||
1779 | func(I_BDEV(inode), arg); | |
1780 | ||
1781 | spin_lock(&inode_sb_list_lock); | |
1782 | } | |
1783 | spin_unlock(&inode_sb_list_lock); | |
1784 | iput(old_inode); | |
1785 | } |