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> | |
66114cad | 17 | #include <linux/backing-dev.h> |
1da177e4 LT |
18 | #include <linux/module.h> |
19 | #include <linux/blkpg.h> | |
b502bd11 | 20 | #include <linux/magic.h> |
1da177e4 | 21 | #include <linux/buffer_head.h> |
ff01bb48 | 22 | #include <linux/swap.h> |
585d3bc0 | 23 | #include <linux/pagevec.h> |
811d736f | 24 | #include <linux/writeback.h> |
1da177e4 LT |
25 | #include <linux/mpage.h> |
26 | #include <linux/mount.h> | |
27 | #include <linux/uio.h> | |
28 | #include <linux/namei.h> | |
1368c4f2 | 29 | #include <linux/log2.h> |
ff01bb48 | 30 | #include <linux/cleancache.h> |
c94c2acf | 31 | #include <linux/dax.h> |
1da177e4 | 32 | #include <asm/uaccess.h> |
07f3f05c | 33 | #include "internal.h" |
1da177e4 LT |
34 | |
35 | struct bdev_inode { | |
36 | struct block_device bdev; | |
37 | struct inode vfs_inode; | |
38 | }; | |
39 | ||
4c54ac62 AB |
40 | static const struct address_space_operations def_blk_aops; |
41 | ||
1da177e4 LT |
42 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
43 | { | |
44 | return container_of(inode, struct bdev_inode, vfs_inode); | |
45 | } | |
46 | ||
ff5053f6 | 47 | struct block_device *I_BDEV(struct inode *inode) |
1da177e4 LT |
48 | { |
49 | return &BDEV_I(inode)->bdev; | |
50 | } | |
1da177e4 LT |
51 | EXPORT_SYMBOL(I_BDEV); |
52 | ||
2af3a815 TK |
53 | void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...) |
54 | { | |
55 | struct va_format vaf; | |
56 | va_list args; | |
57 | ||
58 | va_start(args, fmt); | |
59 | vaf.fmt = fmt; | |
60 | vaf.va = &args; | |
61 | printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf); | |
62 | va_end(args); | |
63 | } | |
64 | ||
dbd3ca50 | 65 | static void bdev_write_inode(struct block_device *bdev) |
564f00f6 | 66 | { |
dbd3ca50 VG |
67 | struct inode *inode = bdev->bd_inode; |
68 | int ret; | |
69 | ||
564f00f6 CH |
70 | spin_lock(&inode->i_lock); |
71 | while (inode->i_state & I_DIRTY) { | |
72 | spin_unlock(&inode->i_lock); | |
dbd3ca50 VG |
73 | ret = write_inode_now(inode, true); |
74 | if (ret) { | |
75 | char name[BDEVNAME_SIZE]; | |
76 | pr_warn_ratelimited("VFS: Dirty inode writeback failed " | |
77 | "for block device %s (err=%d).\n", | |
78 | bdevname(bdev, name), ret); | |
79 | } | |
564f00f6 CH |
80 | spin_lock(&inode->i_lock); |
81 | } | |
82 | spin_unlock(&inode->i_lock); | |
83 | } | |
84 | ||
f9a14399 | 85 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
ff01bb48 | 86 | void kill_bdev(struct block_device *bdev) |
1da177e4 | 87 | { |
ff01bb48 AV |
88 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
89 | ||
f9fe48be | 90 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
f9a14399 | 91 | return; |
ff01bb48 | 92 | |
f9a14399 | 93 | invalidate_bh_lrus(); |
ff01bb48 | 94 | truncate_inode_pages(mapping, 0); |
1da177e4 | 95 | } |
ff01bb48 AV |
96 | EXPORT_SYMBOL(kill_bdev); |
97 | ||
98 | /* Invalidate clean unused buffers and pagecache. */ | |
99 | void invalidate_bdev(struct block_device *bdev) | |
100 | { | |
101 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
102 | ||
103 | if (mapping->nrpages == 0) | |
104 | return; | |
105 | ||
106 | invalidate_bh_lrus(); | |
107 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
108 | invalidate_mapping_pages(mapping, 0, -1); | |
109 | /* 99% of the time, we don't need to flush the cleancache on the bdev. | |
110 | * But, for the strange corners, lets be cautious | |
111 | */ | |
3167760f | 112 | cleancache_invalidate_inode(mapping); |
ff01bb48 AV |
113 | } |
114 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 LT |
115 | |
116 | int set_blocksize(struct block_device *bdev, int size) | |
117 | { | |
118 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 119 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
120 | return -EINVAL; |
121 | ||
122 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 123 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
124 | return -EINVAL; |
125 | ||
126 | /* Don't change the size if it is same as current */ | |
127 | if (bdev->bd_block_size != size) { | |
128 | sync_blockdev(bdev); | |
129 | bdev->bd_block_size = size; | |
130 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
131 | kill_bdev(bdev); | |
132 | } | |
133 | return 0; | |
134 | } | |
135 | ||
136 | EXPORT_SYMBOL(set_blocksize); | |
137 | ||
138 | int sb_set_blocksize(struct super_block *sb, int size) | |
139 | { | |
1da177e4 LT |
140 | if (set_blocksize(sb->s_bdev, size)) |
141 | return 0; | |
142 | /* If we get here, we know size is power of two | |
143 | * and it's value is between 512 and PAGE_SIZE */ | |
144 | sb->s_blocksize = size; | |
38885bd4 | 145 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
146 | return sb->s_blocksize; |
147 | } | |
148 | ||
149 | EXPORT_SYMBOL(sb_set_blocksize); | |
150 | ||
151 | int sb_min_blocksize(struct super_block *sb, int size) | |
152 | { | |
e1defc4f | 153 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
154 | if (size < minsize) |
155 | size = minsize; | |
156 | return sb_set_blocksize(sb, size); | |
157 | } | |
158 | ||
159 | EXPORT_SYMBOL(sb_min_blocksize); | |
160 | ||
161 | static int | |
162 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
163 | struct buffer_head *bh, int create) | |
164 | { | |
1da177e4 LT |
165 | bh->b_bdev = I_BDEV(inode); |
166 | bh->b_blocknr = iblock; | |
167 | set_buffer_mapped(bh); | |
168 | return 0; | |
169 | } | |
170 | ||
4ebb16ca DW |
171 | static struct inode *bdev_file_inode(struct file *file) |
172 | { | |
173 | return file->f_mapping->host; | |
174 | } | |
175 | ||
b2e895db | 176 | static ssize_t |
22c6186e | 177 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset) |
b2e895db AM |
178 | { |
179 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 180 | struct inode *inode = bdev_file_inode(file); |
b2e895db | 181 | |
bbab37dd MW |
182 | if (IS_DAX(inode)) |
183 | return dax_do_io(iocb, inode, iter, offset, blkdev_get_block, | |
184 | NULL, DIO_SKIP_DIO_COUNT); | |
17f8c842 | 185 | return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset, |
fe0f07d0 JA |
186 | blkdev_get_block, NULL, NULL, |
187 | DIO_SKIP_DIO_COUNT); | |
b2e895db AM |
188 | } |
189 | ||
5cee5815 JK |
190 | int __sync_blockdev(struct block_device *bdev, int wait) |
191 | { | |
192 | if (!bdev) | |
193 | return 0; | |
194 | if (!wait) | |
195 | return filemap_flush(bdev->bd_inode->i_mapping); | |
196 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
197 | } | |
198 | ||
585d3bc0 NP |
199 | /* |
200 | * Write out and wait upon all the dirty data associated with a block | |
201 | * device via its mapping. Does not take the superblock lock. | |
202 | */ | |
203 | int sync_blockdev(struct block_device *bdev) | |
204 | { | |
5cee5815 | 205 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
206 | } |
207 | EXPORT_SYMBOL(sync_blockdev); | |
208 | ||
209 | /* | |
210 | * Write out and wait upon all dirty data associated with this | |
211 | * device. Filesystem data as well as the underlying block | |
212 | * device. Takes the superblock lock. | |
213 | */ | |
214 | int fsync_bdev(struct block_device *bdev) | |
215 | { | |
216 | struct super_block *sb = get_super(bdev); | |
217 | if (sb) { | |
60b0680f | 218 | int res = sync_filesystem(sb); |
585d3bc0 NP |
219 | drop_super(sb); |
220 | return res; | |
221 | } | |
222 | return sync_blockdev(bdev); | |
223 | } | |
47e4491b | 224 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
225 | |
226 | /** | |
227 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
228 | * @bdev: blockdevice to lock | |
229 | * | |
585d3bc0 NP |
230 | * If a superblock is found on this device, we take the s_umount semaphore |
231 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
232 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
233 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
234 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
235 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
236 | * actually. | |
237 | */ | |
238 | struct super_block *freeze_bdev(struct block_device *bdev) | |
239 | { | |
240 | struct super_block *sb; | |
241 | int error = 0; | |
242 | ||
243 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
244 | if (++bdev->bd_fsfreeze_count > 1) { |
245 | /* | |
246 | * We don't even need to grab a reference - the first call | |
247 | * to freeze_bdev grab an active reference and only the last | |
248 | * thaw_bdev drops it. | |
249 | */ | |
585d3bc0 | 250 | sb = get_super(bdev); |
4504230a CH |
251 | drop_super(sb); |
252 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
253 | return sb; | |
254 | } | |
255 | ||
256 | sb = get_active_super(bdev); | |
257 | if (!sb) | |
258 | goto out; | |
48b6bca6 BM |
259 | if (sb->s_op->freeze_super) |
260 | error = sb->s_op->freeze_super(sb); | |
261 | else | |
262 | error = freeze_super(sb); | |
18e9e510 JB |
263 | if (error) { |
264 | deactivate_super(sb); | |
265 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 266 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 267 | return ERR_PTR(error); |
585d3bc0 | 268 | } |
18e9e510 | 269 | deactivate_super(sb); |
4504230a | 270 | out: |
585d3bc0 NP |
271 | sync_blockdev(bdev); |
272 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 273 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
274 | } |
275 | EXPORT_SYMBOL(freeze_bdev); | |
276 | ||
277 | /** | |
278 | * thaw_bdev -- unlock filesystem | |
279 | * @bdev: blockdevice to unlock | |
280 | * @sb: associated superblock | |
281 | * | |
282 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
283 | */ | |
284 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
285 | { | |
4504230a | 286 | int error = -EINVAL; |
585d3bc0 NP |
287 | |
288 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 289 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 290 | goto out; |
4504230a CH |
291 | |
292 | error = 0; | |
293 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 294 | goto out; |
4504230a CH |
295 | |
296 | if (!sb) | |
18e9e510 | 297 | goto out; |
4504230a | 298 | |
48b6bca6 BM |
299 | if (sb->s_op->thaw_super) |
300 | error = sb->s_op->thaw_super(sb); | |
301 | else | |
302 | error = thaw_super(sb); | |
18e9e510 JB |
303 | if (error) { |
304 | bdev->bd_fsfreeze_count++; | |
305 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
306 | return error; | |
307 | } | |
308 | out: | |
585d3bc0 NP |
309 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
310 | return 0; | |
311 | } | |
312 | EXPORT_SYMBOL(thaw_bdev); | |
313 | ||
1da177e4 LT |
314 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
315 | { | |
316 | return block_write_full_page(page, blkdev_get_block, wbc); | |
317 | } | |
318 | ||
319 | static int blkdev_readpage(struct file * file, struct page * page) | |
320 | { | |
321 | return block_read_full_page(page, blkdev_get_block); | |
322 | } | |
323 | ||
447f05bb AM |
324 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
325 | struct list_head *pages, unsigned nr_pages) | |
326 | { | |
327 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
328 | } | |
329 | ||
6272b5a5 NP |
330 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
331 | loff_t pos, unsigned len, unsigned flags, | |
332 | struct page **pagep, void **fsdata) | |
1da177e4 | 333 | { |
155130a4 CH |
334 | return block_write_begin(mapping, pos, len, flags, pagep, |
335 | blkdev_get_block); | |
1da177e4 LT |
336 | } |
337 | ||
6272b5a5 NP |
338 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
339 | loff_t pos, unsigned len, unsigned copied, | |
340 | struct page *page, void *fsdata) | |
1da177e4 | 341 | { |
6272b5a5 NP |
342 | int ret; |
343 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
344 | ||
345 | unlock_page(page); | |
09cbfeaf | 346 | put_page(page); |
6272b5a5 NP |
347 | |
348 | return ret; | |
1da177e4 LT |
349 | } |
350 | ||
351 | /* | |
352 | * private llseek: | |
496ad9aa | 353 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
354 | * so we compute the size by hand (just as in block_read/write above) |
355 | */ | |
965c8e59 | 356 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 357 | { |
4ebb16ca | 358 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
359 | loff_t retval; |
360 | ||
5955102c | 361 | inode_lock(bd_inode); |
5d48f3a2 | 362 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 363 | inode_unlock(bd_inode); |
1da177e4 LT |
364 | return retval; |
365 | } | |
366 | ||
02c24a82 | 367 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 368 | { |
4ebb16ca | 369 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 370 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 371 | int error; |
da5aa861 RW |
372 | |
373 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
374 | if (error) | |
375 | return error; | |
ab0a9735 | 376 | |
b8af67e2 AB |
377 | /* |
378 | * There is no need to serialise calls to blkdev_issue_flush with | |
379 | * i_mutex and doing so causes performance issues with concurrent | |
380 | * O_SYNC writers to a block device. | |
381 | */ | |
dd3932ed | 382 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
383 | if (error == -EOPNOTSUPP) |
384 | error = 0; | |
b8af67e2 | 385 | |
ab0a9735 | 386 | return error; |
1da177e4 | 387 | } |
b1dd3b28 | 388 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 389 | |
47a191fd MW |
390 | /** |
391 | * bdev_read_page() - Start reading a page from a block device | |
392 | * @bdev: The device to read the page from | |
393 | * @sector: The offset on the device to read the page to (need not be aligned) | |
394 | * @page: The page to read | |
395 | * | |
396 | * On entry, the page should be locked. It will be unlocked when the page | |
397 | * has been read. If the block driver implements rw_page synchronously, | |
398 | * that will be true on exit from this function, but it need not be. | |
399 | * | |
400 | * Errors returned by this function are usually "soft", eg out of memory, or | |
401 | * queue full; callers should try a different route to read this page rather | |
402 | * than propagate an error back up the stack. | |
403 | * | |
404 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
405 | */ | |
406 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
407 | struct page *page) | |
408 | { | |
409 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
410 | int result = -EOPNOTSUPP; |
411 | ||
f68eb1e7 | 412 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
413 | return result; |
414 | ||
6f3b0e8b | 415 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
416 | if (result) |
417 | return result; | |
418 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ); | |
419 | blk_queue_exit(bdev->bd_queue); | |
420 | return result; | |
47a191fd MW |
421 | } |
422 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
423 | ||
424 | /** | |
425 | * bdev_write_page() - Start writing a page to a block device | |
426 | * @bdev: The device to write the page to | |
427 | * @sector: The offset on the device to write the page to (need not be aligned) | |
428 | * @page: The page to write | |
429 | * @wbc: The writeback_control for the write | |
430 | * | |
431 | * On entry, the page should be locked and not currently under writeback. | |
432 | * On exit, if the write started successfully, the page will be unlocked and | |
433 | * under writeback. If the write failed already (eg the driver failed to | |
434 | * queue the page to the device), the page will still be locked. If the | |
435 | * caller is a ->writepage implementation, it will need to unlock the page. | |
436 | * | |
437 | * Errors returned by this function are usually "soft", eg out of memory, or | |
438 | * queue full; callers should try a different route to write this page rather | |
439 | * than propagate an error back up the stack. | |
440 | * | |
441 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
442 | */ | |
443 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
444 | struct page *page, struct writeback_control *wbc) | |
445 | { | |
446 | int result; | |
447 | int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE; | |
448 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 | 449 | |
f68eb1e7 | 450 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 451 | return -EOPNOTSUPP; |
6f3b0e8b | 452 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
453 | if (result) |
454 | return result; | |
455 | ||
47a191fd MW |
456 | set_page_writeback(page); |
457 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw); | |
458 | if (result) | |
459 | end_page_writeback(page); | |
460 | else | |
461 | unlock_page(page); | |
2e6edc95 | 462 | blk_queue_exit(bdev->bd_queue); |
47a191fd MW |
463 | return result; |
464 | } | |
465 | EXPORT_SYMBOL_GPL(bdev_write_page); | |
466 | ||
dd22f551 MW |
467 | /** |
468 | * bdev_direct_access() - Get the address for directly-accessibly memory | |
469 | * @bdev: The device containing the memory | |
b2e0d162 | 470 | * @dax: control and output parameters for ->direct_access |
dd22f551 MW |
471 | * |
472 | * If a block device is made up of directly addressable memory, this function | |
473 | * will tell the caller the PFN and the address of the memory. The address | |
474 | * may be directly dereferenced within the kernel without the need to call | |
475 | * ioremap(), kmap() or similar. The PFN is suitable for inserting into | |
476 | * page tables. | |
477 | * | |
478 | * Return: negative errno if an error occurs, otherwise the number of bytes | |
479 | * accessible at this address. | |
480 | */ | |
b2e0d162 | 481 | long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax) |
dd22f551 | 482 | { |
b2e0d162 DW |
483 | sector_t sector = dax->sector; |
484 | long avail, size = dax->size; | |
dd22f551 MW |
485 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
486 | ||
43c3dd08 MW |
487 | /* |
488 | * The device driver is allowed to sleep, in order to make the | |
489 | * memory directly accessible. | |
490 | */ | |
491 | might_sleep(); | |
492 | ||
dd22f551 MW |
493 | if (size < 0) |
494 | return size; | |
495 | if (!ops->direct_access) | |
496 | return -EOPNOTSUPP; | |
497 | if ((sector + DIV_ROUND_UP(size, 512)) > | |
498 | part_nr_sects_read(bdev->bd_part)) | |
499 | return -ERANGE; | |
500 | sector += get_start_sect(bdev); | |
501 | if (sector % (PAGE_SIZE / 512)) | |
502 | return -EINVAL; | |
b2e0d162 | 503 | avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn); |
dd22f551 MW |
504 | if (!avail) |
505 | return -ERANGE; | |
fe683ada DW |
506 | if (avail > 0 && avail & ~PAGE_MASK) |
507 | return -ENXIO; | |
dd22f551 MW |
508 | return min(avail, size); |
509 | } | |
510 | EXPORT_SYMBOL_GPL(bdev_direct_access); | |
511 | ||
1da177e4 LT |
512 | /* |
513 | * pseudo-fs | |
514 | */ | |
515 | ||
516 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 517 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
518 | |
519 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
520 | { | |
e94b1766 | 521 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
522 | if (!ei) |
523 | return NULL; | |
524 | return &ei->vfs_inode; | |
525 | } | |
526 | ||
fa0d7e3d | 527 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 528 | { |
fa0d7e3d | 529 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
530 | struct bdev_inode *bdi = BDEV_I(inode); |
531 | ||
1da177e4 LT |
532 | kmem_cache_free(bdev_cachep, bdi); |
533 | } | |
534 | ||
fa0d7e3d NP |
535 | static void bdev_destroy_inode(struct inode *inode) |
536 | { | |
537 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
538 | } | |
539 | ||
51cc5068 | 540 | static void init_once(void *foo) |
1da177e4 LT |
541 | { |
542 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
543 | struct block_device *bdev = &ei->bdev; | |
544 | ||
a35afb83 CL |
545 | memset(bdev, 0, sizeof(*bdev)); |
546 | mutex_init(&bdev->bd_mutex); | |
a35afb83 CL |
547 | INIT_LIST_HEAD(&bdev->bd_inodes); |
548 | INIT_LIST_HEAD(&bdev->bd_list); | |
49731baa TH |
549 | #ifdef CONFIG_SYSFS |
550 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
551 | #endif | |
a35afb83 | 552 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
553 | /* Initialize mutex for freeze. */ |
554 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
555 | } |
556 | ||
557 | static inline void __bd_forget(struct inode *inode) | |
558 | { | |
559 | list_del_init(&inode->i_devices); | |
560 | inode->i_bdev = NULL; | |
561 | inode->i_mapping = &inode->i_data; | |
562 | } | |
563 | ||
b57922d9 | 564 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
565 | { |
566 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
567 | struct list_head *p; | |
91b0abe3 | 568 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 569 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 570 | clear_inode(inode); |
1da177e4 LT |
571 | spin_lock(&bdev_lock); |
572 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
573 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
574 | } | |
575 | list_del_init(&bdev->bd_list); | |
576 | spin_unlock(&bdev_lock); | |
577 | } | |
578 | ||
ee9b6d61 | 579 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
580 | .statfs = simple_statfs, |
581 | .alloc_inode = bdev_alloc_inode, | |
582 | .destroy_inode = bdev_destroy_inode, | |
583 | .drop_inode = generic_delete_inode, | |
b57922d9 | 584 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
585 | }; |
586 | ||
51139ada AV |
587 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
588 | int flags, const char *dev_name, void *data) | |
1da177e4 | 589 | { |
3684aa70 SL |
590 | struct dentry *dent; |
591 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); | |
592 | if (dent) | |
593 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; | |
594 | return dent; | |
1da177e4 LT |
595 | } |
596 | ||
597 | static struct file_system_type bd_type = { | |
598 | .name = "bdev", | |
51139ada | 599 | .mount = bd_mount, |
1da177e4 LT |
600 | .kill_sb = kill_anon_super, |
601 | }; | |
602 | ||
a212b105 TH |
603 | struct super_block *blockdev_superblock __read_mostly; |
604 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
605 | |
606 | void __init bdev_cache_init(void) | |
607 | { | |
608 | int err; | |
ace8577a | 609 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 610 | |
1da177e4 | 611 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 612 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 613 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 614 | init_once); |
1da177e4 LT |
615 | err = register_filesystem(&bd_type); |
616 | if (err) | |
617 | panic("Cannot register bdev pseudo-fs"); | |
618 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
619 | if (IS_ERR(bd_mnt)) |
620 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 621 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
622 | } |
623 | ||
624 | /* | |
625 | * Most likely _very_ bad one - but then it's hardly critical for small | |
626 | * /dev and can be fixed when somebody will need really large one. | |
627 | * Keep in mind that it will be fed through icache hash function too. | |
628 | */ | |
629 | static inline unsigned long hash(dev_t dev) | |
630 | { | |
631 | return MAJOR(dev)+MINOR(dev); | |
632 | } | |
633 | ||
634 | static int bdev_test(struct inode *inode, void *data) | |
635 | { | |
636 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
637 | } | |
638 | ||
639 | static int bdev_set(struct inode *inode, void *data) | |
640 | { | |
641 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
642 | return 0; | |
643 | } | |
644 | ||
645 | static LIST_HEAD(all_bdevs); | |
646 | ||
647 | struct block_device *bdget(dev_t dev) | |
648 | { | |
649 | struct block_device *bdev; | |
650 | struct inode *inode; | |
651 | ||
c2acf7b9 | 652 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
653 | bdev_test, bdev_set, &dev); |
654 | ||
655 | if (!inode) | |
656 | return NULL; | |
657 | ||
658 | bdev = &BDEV_I(inode)->bdev; | |
659 | ||
660 | if (inode->i_state & I_NEW) { | |
661 | bdev->bd_contains = NULL; | |
782b94cd | 662 | bdev->bd_super = NULL; |
1da177e4 LT |
663 | bdev->bd_inode = inode; |
664 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
665 | bdev->bd_part_count = 0; | |
666 | bdev->bd_invalidated = 0; | |
667 | inode->i_mode = S_IFBLK; | |
668 | inode->i_rdev = dev; | |
669 | inode->i_bdev = bdev; | |
670 | inode->i_data.a_ops = &def_blk_aops; | |
671 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
672 | spin_lock(&bdev_lock); |
673 | list_add(&bdev->bd_list, &all_bdevs); | |
674 | spin_unlock(&bdev_lock); | |
675 | unlock_new_inode(inode); | |
676 | } | |
677 | return bdev; | |
678 | } | |
679 | ||
680 | EXPORT_SYMBOL(bdget); | |
681 | ||
dddac6a7 AJ |
682 | /** |
683 | * bdgrab -- Grab a reference to an already referenced block device | |
684 | * @bdev: Block device to grab a reference to. | |
685 | */ | |
686 | struct block_device *bdgrab(struct block_device *bdev) | |
687 | { | |
7de9c6ee | 688 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
689 | return bdev; |
690 | } | |
c1681bf8 | 691 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 692 | |
1da177e4 LT |
693 | long nr_blockdev_pages(void) |
694 | { | |
203a2935 | 695 | struct block_device *bdev; |
1da177e4 LT |
696 | long ret = 0; |
697 | spin_lock(&bdev_lock); | |
203a2935 | 698 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
699 | ret += bdev->bd_inode->i_mapping->nrpages; |
700 | } | |
701 | spin_unlock(&bdev_lock); | |
702 | return ret; | |
703 | } | |
704 | ||
705 | void bdput(struct block_device *bdev) | |
706 | { | |
707 | iput(bdev->bd_inode); | |
708 | } | |
709 | ||
710 | EXPORT_SYMBOL(bdput); | |
711 | ||
712 | static struct block_device *bd_acquire(struct inode *inode) | |
713 | { | |
714 | struct block_device *bdev; | |
09d967c6 | 715 | |
1da177e4 LT |
716 | spin_lock(&bdev_lock); |
717 | bdev = inode->i_bdev; | |
09d967c6 | 718 | if (bdev) { |
ed8a9d2c | 719 | bdgrab(bdev); |
1da177e4 LT |
720 | spin_unlock(&bdev_lock); |
721 | return bdev; | |
722 | } | |
723 | spin_unlock(&bdev_lock); | |
09d967c6 | 724 | |
1da177e4 LT |
725 | bdev = bdget(inode->i_rdev); |
726 | if (bdev) { | |
727 | spin_lock(&bdev_lock); | |
09d967c6 OH |
728 | if (!inode->i_bdev) { |
729 | /* | |
7de9c6ee | 730 | * We take an additional reference to bd_inode, |
09d967c6 OH |
731 | * and it's released in clear_inode() of inode. |
732 | * So, we can access it via ->i_mapping always | |
733 | * without igrab(). | |
734 | */ | |
ed8a9d2c | 735 | bdgrab(bdev); |
09d967c6 OH |
736 | inode->i_bdev = bdev; |
737 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
738 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
739 | } | |
1da177e4 LT |
740 | spin_unlock(&bdev_lock); |
741 | } | |
742 | return bdev; | |
743 | } | |
744 | ||
745 | /* Call when you free inode */ | |
746 | ||
747 | void bd_forget(struct inode *inode) | |
748 | { | |
09d967c6 OH |
749 | struct block_device *bdev = NULL; |
750 | ||
1da177e4 | 751 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
752 | if (!sb_is_blkdev_sb(inode->i_sb)) |
753 | bdev = inode->i_bdev; | |
754 | __bd_forget(inode); | |
1da177e4 | 755 | spin_unlock(&bdev_lock); |
09d967c6 OH |
756 | |
757 | if (bdev) | |
ed8a9d2c | 758 | bdput(bdev); |
1da177e4 LT |
759 | } |
760 | ||
1a3cbbc5 TH |
761 | /** |
762 | * bd_may_claim - test whether a block device can be claimed | |
763 | * @bdev: block device of interest | |
764 | * @whole: whole block device containing @bdev, may equal @bdev | |
765 | * @holder: holder trying to claim @bdev | |
766 | * | |
25985edc | 767 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
768 | * |
769 | * CONTEXT: | |
770 | * spin_lock(&bdev_lock). | |
771 | * | |
772 | * RETURNS: | |
773 | * %true if @bdev can be claimed, %false otherwise. | |
774 | */ | |
775 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
776 | void *holder) | |
1da177e4 | 777 | { |
1da177e4 | 778 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 779 | return true; /* already a holder */ |
1da177e4 | 780 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 781 | return false; /* held by someone else */ |
1da177e4 | 782 | else if (bdev->bd_contains == bdev) |
1a3cbbc5 | 783 | return true; /* is a whole device which isn't held */ |
1da177e4 | 784 | |
e525fd89 | 785 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
786 | return true; /* is a partition of a device that is being partitioned */ |
787 | else if (whole->bd_holder != NULL) | |
788 | return false; /* is a partition of a held device */ | |
1da177e4 | 789 | else |
1a3cbbc5 TH |
790 | return true; /* is a partition of an un-held device */ |
791 | } | |
792 | ||
6b4517a7 TH |
793 | /** |
794 | * bd_prepare_to_claim - prepare to claim a block device | |
795 | * @bdev: block device of interest | |
796 | * @whole: the whole device containing @bdev, may equal @bdev | |
797 | * @holder: holder trying to claim @bdev | |
798 | * | |
799 | * Prepare to claim @bdev. This function fails if @bdev is already | |
800 | * claimed by another holder and waits if another claiming is in | |
801 | * progress. This function doesn't actually claim. On successful | |
802 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
803 | * | |
804 | * CONTEXT: | |
805 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
806 | * it multiple times. | |
807 | * | |
808 | * RETURNS: | |
809 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
810 | */ | |
811 | static int bd_prepare_to_claim(struct block_device *bdev, | |
812 | struct block_device *whole, void *holder) | |
813 | { | |
814 | retry: | |
815 | /* if someone else claimed, fail */ | |
816 | if (!bd_may_claim(bdev, whole, holder)) | |
817 | return -EBUSY; | |
818 | ||
e75aa858 TH |
819 | /* if claiming is already in progress, wait for it to finish */ |
820 | if (whole->bd_claiming) { | |
6b4517a7 TH |
821 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
822 | DEFINE_WAIT(wait); | |
823 | ||
824 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
825 | spin_unlock(&bdev_lock); | |
826 | schedule(); | |
827 | finish_wait(wq, &wait); | |
828 | spin_lock(&bdev_lock); | |
829 | goto retry; | |
830 | } | |
831 | ||
832 | /* yay, all mine */ | |
833 | return 0; | |
834 | } | |
835 | ||
836 | /** | |
837 | * bd_start_claiming - start claiming a block device | |
838 | * @bdev: block device of interest | |
839 | * @holder: holder trying to claim @bdev | |
840 | * | |
841 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
842 | * exclusively and mark that an exclusive open is in progress. Each | |
843 | * successful call to this function must be matched with a call to | |
b0018361 NP |
844 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
845 | * fail). | |
846 | * | |
847 | * This function is used to gain exclusive access to the block device | |
848 | * without actually causing other exclusive open attempts to fail. It | |
849 | * should be used when the open sequence itself requires exclusive | |
850 | * access but may subsequently fail. | |
6b4517a7 TH |
851 | * |
852 | * CONTEXT: | |
853 | * Might sleep. | |
854 | * | |
855 | * RETURNS: | |
856 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
857 | * value on failure. | |
858 | */ | |
859 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
860 | void *holder) | |
861 | { | |
862 | struct gendisk *disk; | |
863 | struct block_device *whole; | |
864 | int partno, err; | |
865 | ||
866 | might_sleep(); | |
867 | ||
868 | /* | |
869 | * @bdev might not have been initialized properly yet, look up | |
870 | * and grab the outer block device the hard way. | |
871 | */ | |
872 | disk = get_gendisk(bdev->bd_dev, &partno); | |
873 | if (!disk) | |
874 | return ERR_PTR(-ENXIO); | |
875 | ||
d4c208b8 TH |
876 | /* |
877 | * Normally, @bdev should equal what's returned from bdget_disk() | |
878 | * if partno is 0; however, some drivers (floppy) use multiple | |
879 | * bdev's for the same physical device and @bdev may be one of the | |
880 | * aliases. Keep @bdev if partno is 0. This means claimer | |
881 | * tracking is broken for those devices but it has always been that | |
882 | * way. | |
883 | */ | |
884 | if (partno) | |
885 | whole = bdget_disk(disk, 0); | |
886 | else | |
887 | whole = bdgrab(bdev); | |
888 | ||
cf342570 | 889 | module_put(disk->fops->owner); |
6b4517a7 TH |
890 | put_disk(disk); |
891 | if (!whole) | |
892 | return ERR_PTR(-ENOMEM); | |
893 | ||
894 | /* prepare to claim, if successful, mark claiming in progress */ | |
895 | spin_lock(&bdev_lock); | |
896 | ||
897 | err = bd_prepare_to_claim(bdev, whole, holder); | |
898 | if (err == 0) { | |
899 | whole->bd_claiming = holder; | |
900 | spin_unlock(&bdev_lock); | |
901 | return whole; | |
902 | } else { | |
903 | spin_unlock(&bdev_lock); | |
904 | bdput(whole); | |
905 | return ERR_PTR(err); | |
906 | } | |
907 | } | |
908 | ||
641dc636 | 909 | #ifdef CONFIG_SYSFS |
49731baa TH |
910 | struct bd_holder_disk { |
911 | struct list_head list; | |
912 | struct gendisk *disk; | |
913 | int refcnt; | |
914 | }; | |
915 | ||
916 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
917 | struct gendisk *disk) | |
918 | { | |
919 | struct bd_holder_disk *holder; | |
920 | ||
921 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
922 | if (holder->disk == disk) | |
923 | return holder; | |
924 | return NULL; | |
925 | } | |
926 | ||
4d7dd8fd | 927 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 928 | { |
4d7dd8fd | 929 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
930 | } |
931 | ||
932 | static void del_symlink(struct kobject *from, struct kobject *to) | |
933 | { | |
641dc636 JN |
934 | sysfs_remove_link(from, kobject_name(to)); |
935 | } | |
936 | ||
df6c0cd9 | 937 | /** |
e09b457b TH |
938 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
939 | * @bdev: the claimed slave bdev | |
940 | * @disk: the holding disk | |
df6c0cd9 | 941 | * |
49731baa TH |
942 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
943 | * | |
e09b457b | 944 | * This functions creates the following sysfs symlinks. |
641dc636 | 945 | * |
e09b457b TH |
946 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
947 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 948 | * |
e09b457b TH |
949 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
950 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 951 | * |
e09b457b TH |
952 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
953 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 954 | * |
e09b457b TH |
955 | * The caller must have claimed @bdev before calling this function and |
956 | * ensure that both @bdev and @disk are valid during the creation and | |
957 | * lifetime of these symlinks. | |
641dc636 | 958 | * |
e09b457b TH |
959 | * CONTEXT: |
960 | * Might sleep. | |
641dc636 | 961 | * |
e09b457b TH |
962 | * RETURNS: |
963 | * 0 on success, -errno on failure. | |
641dc636 | 964 | */ |
e09b457b | 965 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 966 | { |
49731baa | 967 | struct bd_holder_disk *holder; |
e09b457b | 968 | int ret = 0; |
641dc636 | 969 | |
2e7b651d | 970 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 971 | |
49731baa | 972 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 973 | |
e09b457b TH |
974 | /* FIXME: remove the following once add_disk() handles errors */ |
975 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
976 | goto out_unlock; | |
4e91672c | 977 | |
49731baa TH |
978 | holder = bd_find_holder_disk(bdev, disk); |
979 | if (holder) { | |
980 | holder->refcnt++; | |
e09b457b | 981 | goto out_unlock; |
49731baa | 982 | } |
641dc636 | 983 | |
49731baa TH |
984 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
985 | if (!holder) { | |
986 | ret = -ENOMEM; | |
e09b457b TH |
987 | goto out_unlock; |
988 | } | |
641dc636 | 989 | |
49731baa TH |
990 | INIT_LIST_HEAD(&holder->list); |
991 | holder->disk = disk; | |
992 | holder->refcnt = 1; | |
993 | ||
994 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
995 | if (ret) | |
996 | goto out_free; | |
997 | ||
998 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
999 | if (ret) | |
1000 | goto out_del; | |
e7407d16 TH |
1001 | /* |
1002 | * bdev could be deleted beneath us which would implicitly destroy | |
1003 | * the holder directory. Hold on to it. | |
1004 | */ | |
1005 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1006 | |
1007 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1008 | goto out_unlock; | |
1009 | ||
1010 | out_del: | |
1011 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1012 | out_free: | |
1013 | kfree(holder); | |
e09b457b | 1014 | out_unlock: |
b4cf1b72 | 1015 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1016 | return ret; |
641dc636 | 1017 | } |
e09b457b | 1018 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1019 | |
49731baa TH |
1020 | /** |
1021 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1022 | * @bdev: the calimed slave bdev | |
1023 | * @disk: the holding disk | |
1024 | * | |
1025 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1026 | * | |
1027 | * CONTEXT: | |
1028 | * Might sleep. | |
1029 | */ | |
1030 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1031 | { |
49731baa | 1032 | struct bd_holder_disk *holder; |
641dc636 | 1033 | |
49731baa | 1034 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1035 | |
49731baa TH |
1036 | holder = bd_find_holder_disk(bdev, disk); |
1037 | ||
1038 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1039 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1040 | del_symlink(bdev->bd_part->holder_dir, | |
1041 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1042 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1043 | list_del_init(&holder->list); |
1044 | kfree(holder); | |
1045 | } | |
1046 | ||
1047 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1048 | } |
49731baa | 1049 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1050 | #endif |
1da177e4 | 1051 | |
56ade44b AP |
1052 | /** |
1053 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1054 | * | |
1055 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1056 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1057 | * |
1058 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1059 | * when a disk has been changed -- either by a media change or online | |
1060 | * resize. | |
1061 | */ | |
93b270f7 | 1062 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1063 | { |
93b270f7 | 1064 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1065 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1066 | "resized disk %s\n", |
1067 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b AP |
1068 | } |
1069 | ||
1070 | if (!bdev->bd_disk) | |
1071 | return; | |
d27769ec | 1072 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1073 | bdev->bd_invalidated = 1; |
1074 | } | |
1075 | ||
c3279d14 | 1076 | /** |
57d1b536 | 1077 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1078 | * @disk: struct gendisk to check |
1079 | * @bdev: struct bdev to adjust. | |
1080 | * | |
1081 | * This routine checks to see if the bdev size does not match the disk size | |
1082 | * and adjusts it if it differs. | |
1083 | */ | |
1084 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1085 | { | |
1086 | loff_t disk_size, bdev_size; | |
1087 | ||
1088 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1089 | bdev_size = i_size_read(bdev->bd_inode); | |
1090 | if (disk_size != bdev_size) { | |
c3279d14 AP |
1091 | printk(KERN_INFO |
1092 | "%s: detected capacity change from %lld to %lld\n", | |
424081f3 | 1093 | disk->disk_name, bdev_size, disk_size); |
c3279d14 | 1094 | i_size_write(bdev->bd_inode, disk_size); |
93b270f7 | 1095 | flush_disk(bdev, false); |
c3279d14 AP |
1096 | } |
1097 | } | |
1098 | EXPORT_SYMBOL(check_disk_size_change); | |
1099 | ||
0c002c2f | 1100 | /** |
57d1b536 | 1101 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1102 | * @disk: struct gendisk to be revalidated |
1103 | * | |
1104 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1105 | * call-backs. It is used to do common pre and post operations needed | |
1106 | * for all revalidate_disk operations. | |
1107 | */ | |
1108 | int revalidate_disk(struct gendisk *disk) | |
1109 | { | |
c3279d14 | 1110 | struct block_device *bdev; |
0c002c2f AP |
1111 | int ret = 0; |
1112 | ||
1113 | if (disk->fops->revalidate_disk) | |
1114 | ret = disk->fops->revalidate_disk(disk); | |
25520d55 | 1115 | blk_integrity_revalidate(disk); |
c3279d14 AP |
1116 | bdev = bdget_disk(disk, 0); |
1117 | if (!bdev) | |
1118 | return ret; | |
1119 | ||
1120 | mutex_lock(&bdev->bd_mutex); | |
1121 | check_disk_size_change(disk, bdev); | |
7630b661 | 1122 | bdev->bd_invalidated = 0; |
c3279d14 AP |
1123 | mutex_unlock(&bdev->bd_mutex); |
1124 | bdput(bdev); | |
0c002c2f AP |
1125 | return ret; |
1126 | } | |
1127 | EXPORT_SYMBOL(revalidate_disk); | |
1128 | ||
1da177e4 LT |
1129 | /* |
1130 | * This routine checks whether a removable media has been changed, | |
1131 | * and invalidates all buffer-cache-entries in that case. This | |
1132 | * is a relatively slow routine, so we have to try to minimize using | |
1133 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1134 | * is the best way of combining speed and utility, I think. | |
1135 | * People changing diskettes in the middle of an operation deserve | |
1136 | * to lose :-) | |
1137 | */ | |
1138 | int check_disk_change(struct block_device *bdev) | |
1139 | { | |
1140 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1141 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1142 | unsigned int events; |
1da177e4 | 1143 | |
77ea887e TH |
1144 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1145 | DISK_EVENT_EJECT_REQUEST); | |
1146 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1147 | return 0; |
1148 | ||
93b270f7 | 1149 | flush_disk(bdev, true); |
1da177e4 LT |
1150 | if (bdops->revalidate_disk) |
1151 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1152 | return 1; |
1153 | } | |
1154 | ||
1155 | EXPORT_SYMBOL(check_disk_change); | |
1156 | ||
1157 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1158 | { | |
e1defc4f | 1159 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1160 | |
5955102c | 1161 | inode_lock(bdev->bd_inode); |
d646a02a | 1162 | i_size_write(bdev->bd_inode, size); |
5955102c | 1163 | inode_unlock(bdev->bd_inode); |
09cbfeaf | 1164 | while (bsize < PAGE_SIZE) { |
1da177e4 LT |
1165 | if (size & bsize) |
1166 | break; | |
1167 | bsize <<= 1; | |
1168 | } | |
1169 | bdev->bd_block_size = bsize; | |
1170 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1171 | } | |
1172 | EXPORT_SYMBOL(bd_set_size); | |
1173 | ||
4385bab1 | 1174 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1175 | |
6d740cd5 PZ |
1176 | /* |
1177 | * bd_mutex locking: | |
1178 | * | |
1179 | * mutex_lock(part->bd_mutex) | |
1180 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1181 | */ | |
1182 | ||
572c4892 | 1183 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1184 | { |
1da177e4 | 1185 | struct gendisk *disk; |
523e1d39 | 1186 | struct module *owner; |
7db9cfd3 | 1187 | int ret; |
cf771cb5 | 1188 | int partno; |
fe6e9c1f AV |
1189 | int perm = 0; |
1190 | ||
572c4892 | 1191 | if (mode & FMODE_READ) |
fe6e9c1f | 1192 | perm |= MAY_READ; |
572c4892 | 1193 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1194 | perm |= MAY_WRITE; |
1195 | /* | |
1196 | * hooks: /n/, see "layering violations". | |
1197 | */ | |
b7300b78 CW |
1198 | if (!for_part) { |
1199 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1200 | if (ret != 0) { | |
1201 | bdput(bdev); | |
1202 | return ret; | |
1203 | } | |
82666020 | 1204 | } |
7db9cfd3 | 1205 | |
d3374825 | 1206 | restart: |
0762b8bd | 1207 | |
89f97496 | 1208 | ret = -ENXIO; |
cf771cb5 | 1209 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1210 | if (!disk) |
6e9624b8 | 1211 | goto out; |
523e1d39 | 1212 | owner = disk->fops->owner; |
1da177e4 | 1213 | |
69e02c59 | 1214 | disk_block_events(disk); |
6796bf54 | 1215 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1216 | if (!bdev->bd_openers) { |
1217 | bdev->bd_disk = disk; | |
87192a2a | 1218 | bdev->bd_queue = disk->queue; |
1da177e4 | 1219 | bdev->bd_contains = bdev; |
03cdadb0 DW |
1220 | if (IS_ENABLED(CONFIG_BLK_DEV_DAX) && disk->fops->direct_access) |
1221 | bdev->bd_inode->i_flags = S_DAX; | |
1222 | else | |
1223 | bdev->bd_inode->i_flags = 0; | |
1224 | ||
cf771cb5 | 1225 | if (!partno) { |
89f97496 TH |
1226 | ret = -ENXIO; |
1227 | bdev->bd_part = disk_get_part(disk, partno); | |
1228 | if (!bdev->bd_part) | |
1229 | goto out_clear; | |
1230 | ||
1196f8b8 | 1231 | ret = 0; |
1da177e4 | 1232 | if (disk->fops->open) { |
572c4892 | 1233 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1234 | if (ret == -ERESTARTSYS) { |
1235 | /* Lost a race with 'disk' being | |
1236 | * deleted, try again. | |
1237 | * See md.c | |
1238 | */ | |
1239 | disk_put_part(bdev->bd_part); | |
1240 | bdev->bd_part = NULL; | |
d3374825 | 1241 | bdev->bd_disk = NULL; |
87192a2a | 1242 | bdev->bd_queue = NULL; |
d3374825 | 1243 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1244 | disk_unblock_events(disk); |
69e02c59 | 1245 | put_disk(disk); |
523e1d39 | 1246 | module_put(owner); |
d3374825 N |
1247 | goto restart; |
1248 | } | |
1da177e4 | 1249 | } |
7e69723f | 1250 | |
5a023cdb | 1251 | if (!ret) { |
7e69723f | 1252 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
5a023cdb DW |
1253 | if (!blkdev_dax_capable(bdev)) |
1254 | bdev->bd_inode->i_flags &= ~S_DAX; | |
1255 | } | |
7e69723f | 1256 | |
1196f8b8 TH |
1257 | /* |
1258 | * If the device is invalidated, rescan partition | |
1259 | * if open succeeded or failed with -ENOMEDIUM. | |
1260 | * The latter is necessary to prevent ghost | |
1261 | * partitions on a removed medium. | |
1262 | */ | |
fe316bf2 JN |
1263 | if (bdev->bd_invalidated) { |
1264 | if (!ret) | |
1265 | rescan_partitions(disk, bdev); | |
1266 | else if (ret == -ENOMEDIUM) | |
1267 | invalidate_partitions(disk, bdev); | |
1268 | } | |
5a023cdb | 1269 | |
1196f8b8 TH |
1270 | if (ret) |
1271 | goto out_clear; | |
1da177e4 | 1272 | } else { |
1da177e4 LT |
1273 | struct block_device *whole; |
1274 | whole = bdget_disk(disk, 0); | |
1275 | ret = -ENOMEM; | |
1276 | if (!whole) | |
0762b8bd | 1277 | goto out_clear; |
37be4124 | 1278 | BUG_ON(for_part); |
572c4892 | 1279 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1280 | if (ret) |
0762b8bd | 1281 | goto out_clear; |
1da177e4 | 1282 | bdev->bd_contains = whole; |
89f97496 | 1283 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1284 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1285 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1286 | ret = -ENXIO; |
0762b8bd | 1287 | goto out_clear; |
1da177e4 | 1288 | } |
89f97496 | 1289 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
5a023cdb | 1290 | if (!blkdev_dax_capable(bdev)) |
f0b2e563 | 1291 | bdev->bd_inode->i_flags &= ~S_DAX; |
1da177e4 LT |
1292 | } |
1293 | } else { | |
1da177e4 | 1294 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1295 | ret = 0; |
1296 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1297 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1298 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1299 | if (bdev->bd_invalidated) { |
1300 | if (!ret) | |
1301 | rescan_partitions(bdev->bd_disk, bdev); | |
1302 | else if (ret == -ENOMEDIUM) | |
1303 | invalidate_partitions(bdev->bd_disk, bdev); | |
1304 | } | |
1196f8b8 TH |
1305 | if (ret) |
1306 | goto out_unlock_bdev; | |
1da177e4 | 1307 | } |
69e02c59 | 1308 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1309 | put_disk(disk); |
523e1d39 | 1310 | module_put(owner); |
1da177e4 LT |
1311 | } |
1312 | bdev->bd_openers++; | |
37be4124 N |
1313 | if (for_part) |
1314 | bdev->bd_part_count++; | |
c039e313 | 1315 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1316 | disk_unblock_events(disk); |
1da177e4 LT |
1317 | return 0; |
1318 | ||
0762b8bd | 1319 | out_clear: |
89f97496 | 1320 | disk_put_part(bdev->bd_part); |
1da177e4 | 1321 | bdev->bd_disk = NULL; |
0762b8bd | 1322 | bdev->bd_part = NULL; |
87192a2a | 1323 | bdev->bd_queue = NULL; |
1da177e4 | 1324 | if (bdev != bdev->bd_contains) |
572c4892 | 1325 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1326 | bdev->bd_contains = NULL; |
0762b8bd | 1327 | out_unlock_bdev: |
c039e313 | 1328 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1329 | disk_unblock_events(disk); |
0762b8bd | 1330 | put_disk(disk); |
523e1d39 | 1331 | module_put(owner); |
4345caba | 1332 | out: |
0762b8bd TH |
1333 | bdput(bdev); |
1334 | ||
1da177e4 LT |
1335 | return ret; |
1336 | } | |
1337 | ||
d4d77629 TH |
1338 | /** |
1339 | * blkdev_get - open a block device | |
1340 | * @bdev: block_device to open | |
1341 | * @mode: FMODE_* mask | |
1342 | * @holder: exclusive holder identifier | |
1343 | * | |
1344 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1345 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1346 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1347 | * | |
1348 | * On success, the reference count of @bdev is unchanged. On failure, | |
1349 | * @bdev is put. | |
1350 | * | |
1351 | * CONTEXT: | |
1352 | * Might sleep. | |
1353 | * | |
1354 | * RETURNS: | |
1355 | * 0 on success, -errno on failure. | |
1356 | */ | |
e525fd89 | 1357 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1358 | { |
e525fd89 TH |
1359 | struct block_device *whole = NULL; |
1360 | int res; | |
1361 | ||
1362 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1363 | ||
1364 | if ((mode & FMODE_EXCL) && holder) { | |
1365 | whole = bd_start_claiming(bdev, holder); | |
1366 | if (IS_ERR(whole)) { | |
1367 | bdput(bdev); | |
1368 | return PTR_ERR(whole); | |
1369 | } | |
1370 | } | |
1371 | ||
1372 | res = __blkdev_get(bdev, mode, 0); | |
1373 | ||
1374 | if (whole) { | |
d4dc210f TH |
1375 | struct gendisk *disk = whole->bd_disk; |
1376 | ||
6a027eff | 1377 | /* finish claiming */ |
77ea887e | 1378 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1379 | spin_lock(&bdev_lock); |
1380 | ||
77ea887e | 1381 | if (!res) { |
6a027eff TH |
1382 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1383 | /* | |
1384 | * Note that for a whole device bd_holders | |
1385 | * will be incremented twice, and bd_holder | |
1386 | * will be set to bd_may_claim before being | |
1387 | * set to holder | |
1388 | */ | |
1389 | whole->bd_holders++; | |
1390 | whole->bd_holder = bd_may_claim; | |
1391 | bdev->bd_holders++; | |
1392 | bdev->bd_holder = holder; | |
1393 | } | |
1394 | ||
1395 | /* tell others that we're done */ | |
1396 | BUG_ON(whole->bd_claiming != holder); | |
1397 | whole->bd_claiming = NULL; | |
1398 | wake_up_bit(&whole->bd_claiming, 0); | |
1399 | ||
1400 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1401 | |
1402 | /* | |
d4dc210f TH |
1403 | * Block event polling for write claims if requested. Any |
1404 | * write holder makes the write_holder state stick until | |
1405 | * all are released. This is good enough and tracking | |
1406 | * individual writeable reference is too fragile given the | |
1407 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1408 | */ |
4c49ff3f TH |
1409 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1410 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1411 | bdev->bd_write_holder = true; |
d4dc210f | 1412 | disk_block_events(disk); |
77ea887e TH |
1413 | } |
1414 | ||
1415 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1416 | bdput(whole); |
e525fd89 TH |
1417 | } |
1418 | ||
1419 | return res; | |
37be4124 | 1420 | } |
1da177e4 LT |
1421 | EXPORT_SYMBOL(blkdev_get); |
1422 | ||
d4d77629 TH |
1423 | /** |
1424 | * blkdev_get_by_path - open a block device by name | |
1425 | * @path: path to the block device to open | |
1426 | * @mode: FMODE_* mask | |
1427 | * @holder: exclusive holder identifier | |
1428 | * | |
1429 | * Open the blockdevice described by the device file at @path. @mode | |
1430 | * and @holder are identical to blkdev_get(). | |
1431 | * | |
1432 | * On success, the returned block_device has reference count of one. | |
1433 | * | |
1434 | * CONTEXT: | |
1435 | * Might sleep. | |
1436 | * | |
1437 | * RETURNS: | |
1438 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1439 | */ | |
1440 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1441 | void *holder) | |
1442 | { | |
1443 | struct block_device *bdev; | |
1444 | int err; | |
1445 | ||
1446 | bdev = lookup_bdev(path); | |
1447 | if (IS_ERR(bdev)) | |
1448 | return bdev; | |
1449 | ||
1450 | err = blkdev_get(bdev, mode, holder); | |
1451 | if (err) | |
1452 | return ERR_PTR(err); | |
1453 | ||
e51900f7 CE |
1454 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1455 | blkdev_put(bdev, mode); | |
1456 | return ERR_PTR(-EACCES); | |
1457 | } | |
1458 | ||
d4d77629 TH |
1459 | return bdev; |
1460 | } | |
1461 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1462 | ||
1463 | /** | |
1464 | * blkdev_get_by_dev - open a block device by device number | |
1465 | * @dev: device number of block device to open | |
1466 | * @mode: FMODE_* mask | |
1467 | * @holder: exclusive holder identifier | |
1468 | * | |
1469 | * Open the blockdevice described by device number @dev. @mode and | |
1470 | * @holder are identical to blkdev_get(). | |
1471 | * | |
1472 | * Use it ONLY if you really do not have anything better - i.e. when | |
1473 | * you are behind a truly sucky interface and all you are given is a | |
1474 | * device number. _Never_ to be used for internal purposes. If you | |
1475 | * ever need it - reconsider your API. | |
1476 | * | |
1477 | * On success, the returned block_device has reference count of one. | |
1478 | * | |
1479 | * CONTEXT: | |
1480 | * Might sleep. | |
1481 | * | |
1482 | * RETURNS: | |
1483 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1484 | */ | |
1485 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1486 | { | |
1487 | struct block_device *bdev; | |
1488 | int err; | |
1489 | ||
1490 | bdev = bdget(dev); | |
1491 | if (!bdev) | |
1492 | return ERR_PTR(-ENOMEM); | |
1493 | ||
1494 | err = blkdev_get(bdev, mode, holder); | |
1495 | if (err) | |
1496 | return ERR_PTR(err); | |
1497 | ||
1498 | return bdev; | |
1499 | } | |
1500 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1501 | ||
1da177e4 LT |
1502 | static int blkdev_open(struct inode * inode, struct file * filp) |
1503 | { | |
1504 | struct block_device *bdev; | |
1da177e4 LT |
1505 | |
1506 | /* | |
1507 | * Preserve backwards compatibility and allow large file access | |
1508 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1509 | * binary needs it. We might want to drop this workaround | |
1510 | * during an unstable branch. | |
1511 | */ | |
1512 | filp->f_flags |= O_LARGEFILE; | |
1513 | ||
572c4892 AV |
1514 | if (filp->f_flags & O_NDELAY) |
1515 | filp->f_mode |= FMODE_NDELAY; | |
1516 | if (filp->f_flags & O_EXCL) | |
1517 | filp->f_mode |= FMODE_EXCL; | |
1518 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1519 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1520 | ||
1da177e4 | 1521 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1522 | if (bdev == NULL) |
1523 | return -ENOMEM; | |
1da177e4 | 1524 | |
572c4892 AV |
1525 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1526 | ||
e525fd89 | 1527 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1528 | } |
1529 | ||
4385bab1 | 1530 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1531 | { |
2e7b651d | 1532 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1533 | struct block_device *victim = NULL; |
2e7b651d | 1534 | |
6796bf54 | 1535 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1536 | if (for_part) |
1537 | bdev->bd_part_count--; | |
1538 | ||
2e7b651d | 1539 | if (!--bdev->bd_openers) { |
6a027eff | 1540 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1541 | sync_blockdev(bdev); |
1542 | kill_bdev(bdev); | |
43d1c0eb ID |
1543 | |
1544 | bdev_write_inode(bdev); | |
564f00f6 | 1545 | /* |
43d1c0eb ID |
1546 | * Detaching bdev inode from its wb in __destroy_inode() |
1547 | * is too late: the queue which embeds its bdi (along with | |
1548 | * root wb) can be gone as soon as we put_disk() below. | |
94007751 | 1549 | */ |
43d1c0eb | 1550 | inode_detach_wb(bdev->bd_inode); |
2e7b651d PZ |
1551 | } |
1552 | if (bdev->bd_contains == bdev) { | |
1553 | if (disk->fops->release) | |
db2a144b | 1554 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1555 | } |
1556 | if (!bdev->bd_openers) { | |
1557 | struct module *owner = disk->fops->owner; | |
1558 | ||
0762b8bd TH |
1559 | disk_put_part(bdev->bd_part); |
1560 | bdev->bd_part = NULL; | |
2e7b651d | 1561 | bdev->bd_disk = NULL; |
37be4124 N |
1562 | if (bdev != bdev->bd_contains) |
1563 | victim = bdev->bd_contains; | |
2e7b651d | 1564 | bdev->bd_contains = NULL; |
523e1d39 TH |
1565 | |
1566 | put_disk(disk); | |
1567 | module_put(owner); | |
2e7b651d | 1568 | } |
2e7b651d PZ |
1569 | mutex_unlock(&bdev->bd_mutex); |
1570 | bdput(bdev); | |
37be4124 | 1571 | if (victim) |
9a1c3542 | 1572 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1573 | } |
1574 | ||
4385bab1 | 1575 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1576 | { |
85ef06d1 TH |
1577 | mutex_lock(&bdev->bd_mutex); |
1578 | ||
e525fd89 | 1579 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1580 | bool bdev_free; |
1581 | ||
1582 | /* | |
1583 | * Release a claim on the device. The holder fields | |
1584 | * are protected with bdev_lock. bd_mutex is to | |
1585 | * synchronize disk_holder unlinking. | |
1586 | */ | |
6a027eff TH |
1587 | spin_lock(&bdev_lock); |
1588 | ||
1589 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1590 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1591 | ||
1592 | /* bd_contains might point to self, check in a separate step */ | |
1593 | if ((bdev_free = !bdev->bd_holders)) | |
1594 | bdev->bd_holder = NULL; | |
1595 | if (!bdev->bd_contains->bd_holders) | |
1596 | bdev->bd_contains->bd_holder = NULL; | |
1597 | ||
1598 | spin_unlock(&bdev_lock); | |
1599 | ||
77ea887e TH |
1600 | /* |
1601 | * If this was the last claim, remove holder link and | |
1602 | * unblock evpoll if it was a write holder. | |
1603 | */ | |
85ef06d1 TH |
1604 | if (bdev_free && bdev->bd_write_holder) { |
1605 | disk_unblock_events(bdev->bd_disk); | |
1606 | bdev->bd_write_holder = false; | |
77ea887e | 1607 | } |
6936217c | 1608 | } |
77ea887e | 1609 | |
85ef06d1 TH |
1610 | /* |
1611 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1612 | * event. This is to ensure detection of media removal commanded | |
1613 | * from userland - e.g. eject(1). | |
1614 | */ | |
1615 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1616 | ||
1617 | mutex_unlock(&bdev->bd_mutex); | |
1618 | ||
4385bab1 | 1619 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1620 | } |
2e7b651d PZ |
1621 | EXPORT_SYMBOL(blkdev_put); |
1622 | ||
1da177e4 LT |
1623 | static int blkdev_close(struct inode * inode, struct file * filp) |
1624 | { | |
4ebb16ca | 1625 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1626 | blkdev_put(bdev, filp->f_mode); |
1627 | return 0; | |
1da177e4 LT |
1628 | } |
1629 | ||
bb93e3a5 | 1630 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1631 | { |
4ebb16ca | 1632 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1633 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1634 | |
1635 | /* | |
1636 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1637 | * to updated it before every ioctl. | |
1638 | */ | |
56b26add | 1639 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1640 | mode |= FMODE_NDELAY; |
1641 | else | |
1642 | mode &= ~FMODE_NDELAY; | |
1643 | ||
56b26add | 1644 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1645 | } |
1646 | ||
eef99380 CH |
1647 | /* |
1648 | * Write data to the block device. Only intended for the block device itself | |
1649 | * and the raw driver which basically is a fake block device. | |
1650 | * | |
1651 | * Does not take i_mutex for the write and thus is not for general purpose | |
1652 | * use. | |
1653 | */ | |
1456c0a8 | 1654 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1655 | { |
1656 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1657 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1658 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1659 | struct blk_plug plug; |
eef99380 | 1660 | ssize_t ret; |
5f380c7f | 1661 | |
7ec7b94a AV |
1662 | if (bdev_read_only(I_BDEV(bd_inode))) |
1663 | return -EPERM; | |
5f380c7f | 1664 | |
7ec7b94a | 1665 | if (!iov_iter_count(from)) |
5f380c7f AV |
1666 | return 0; |
1667 | ||
7ec7b94a AV |
1668 | if (iocb->ki_pos >= size) |
1669 | return -ENOSPC; | |
1670 | ||
1671 | iov_iter_truncate(from, size - iocb->ki_pos); | |
eef99380 | 1672 | |
53362a05 | 1673 | blk_start_plug(&plug); |
1456c0a8 | 1674 | ret = __generic_file_write_iter(iocb, from); |
02afc27f | 1675 | if (ret > 0) { |
eef99380 | 1676 | ssize_t err; |
1456c0a8 | 1677 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
45d4f855 | 1678 | if (err < 0) |
eef99380 CH |
1679 | ret = err; |
1680 | } | |
53362a05 | 1681 | blk_finish_plug(&plug); |
eef99380 CH |
1682 | return ret; |
1683 | } | |
1456c0a8 | 1684 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1685 | |
b2de525f | 1686 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1687 | { |
1688 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1689 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 1690 | loff_t size = i_size_read(bd_inode); |
a886038b | 1691 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1692 | |
1693 | if (pos >= size) | |
1694 | return 0; | |
1695 | ||
1696 | size -= pos; | |
a886038b AV |
1697 | iov_iter_truncate(to, size); |
1698 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1699 | } |
b2de525f | 1700 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1701 | |
87d8fe1e TT |
1702 | /* |
1703 | * Try to release a page associated with block device when the system | |
1704 | * is under memory pressure. | |
1705 | */ | |
1706 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1707 | { | |
1708 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1709 | ||
1710 | if (super && super->s_op->bdev_try_to_free_page) | |
1711 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1712 | ||
1713 | return try_to_free_buffers(page); | |
1714 | } | |
1715 | ||
7f6d5b52 RZ |
1716 | static int blkdev_writepages(struct address_space *mapping, |
1717 | struct writeback_control *wbc) | |
1718 | { | |
1719 | if (dax_mapping(mapping)) { | |
1720 | struct block_device *bdev = I_BDEV(mapping->host); | |
1721 | ||
1722 | return dax_writeback_mapping_range(mapping, bdev, wbc); | |
1723 | } | |
1724 | return generic_writepages(mapping, wbc); | |
1725 | } | |
1726 | ||
4c54ac62 | 1727 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1728 | .readpage = blkdev_readpage, |
447f05bb | 1729 | .readpages = blkdev_readpages, |
1da177e4 | 1730 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1731 | .write_begin = blkdev_write_begin, |
1732 | .write_end = blkdev_write_end, | |
7f6d5b52 | 1733 | .writepages = blkdev_writepages, |
87d8fe1e | 1734 | .releasepage = blkdev_releasepage, |
1da177e4 | 1735 | .direct_IO = blkdev_direct_IO, |
b4597226 | 1736 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1737 | }; |
1738 | ||
5a023cdb DW |
1739 | #ifdef CONFIG_FS_DAX |
1740 | /* | |
1741 | * In the raw block case we do not need to contend with truncation nor | |
1742 | * unwritten file extents. Without those concerns there is no need for | |
1743 | * additional locking beyond the mmap_sem context that these routines | |
1744 | * are already executing under. | |
1745 | * | |
1746 | * Note, there is no protection if the block device is dynamically | |
1747 | * resized (partition grow/shrink) during a fault. A stable block device | |
1748 | * size is already not enforced in the blkdev_direct_IO path. | |
1749 | * | |
1750 | * For DAX, it is the responsibility of the block device driver to | |
1751 | * ensure the whole-disk device size is stable while requests are in | |
1752 | * flight. | |
1753 | * | |
1754 | * Finally, unlike the filemap_page_mkwrite() case there is no | |
1755 | * filesystem superblock to sync against freezing. We still include a | |
1756 | * pfn_mkwrite callback for dax drivers to receive write fault | |
1757 | * notifications. | |
1758 | */ | |
1759 | static int blkdev_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
1760 | { | |
02fbd139 | 1761 | return __dax_fault(vma, vmf, blkdev_get_block); |
5a023cdb DW |
1762 | } |
1763 | ||
9c5a05bc RZ |
1764 | static int blkdev_dax_pfn_mkwrite(struct vm_area_struct *vma, |
1765 | struct vm_fault *vmf) | |
1766 | { | |
1767 | return dax_pfn_mkwrite(vma, vmf); | |
1768 | } | |
1769 | ||
5a023cdb DW |
1770 | static int blkdev_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr, |
1771 | pmd_t *pmd, unsigned int flags) | |
1772 | { | |
02fbd139 | 1773 | return __dax_pmd_fault(vma, addr, pmd, flags, blkdev_get_block); |
5a023cdb DW |
1774 | } |
1775 | ||
5a023cdb | 1776 | static const struct vm_operations_struct blkdev_dax_vm_ops = { |
5a023cdb DW |
1777 | .fault = blkdev_dax_fault, |
1778 | .pmd_fault = blkdev_dax_pmd_fault, | |
9c5a05bc | 1779 | .pfn_mkwrite = blkdev_dax_pfn_mkwrite, |
5a023cdb DW |
1780 | }; |
1781 | ||
1782 | static const struct vm_operations_struct blkdev_default_vm_ops = { | |
5a023cdb DW |
1783 | .fault = filemap_fault, |
1784 | .map_pages = filemap_map_pages, | |
1785 | }; | |
1786 | ||
1787 | static int blkdev_mmap(struct file *file, struct vm_area_struct *vma) | |
1788 | { | |
1789 | struct inode *bd_inode = bdev_file_inode(file); | |
5a023cdb DW |
1790 | |
1791 | file_accessed(file); | |
5a023cdb DW |
1792 | if (IS_DAX(bd_inode)) { |
1793 | vma->vm_ops = &blkdev_dax_vm_ops; | |
1794 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; | |
1795 | } else { | |
1796 | vma->vm_ops = &blkdev_default_vm_ops; | |
1797 | } | |
5a023cdb DW |
1798 | |
1799 | return 0; | |
1800 | } | |
1801 | #else | |
1802 | #define blkdev_mmap generic_file_mmap | |
1803 | #endif | |
1804 | ||
4b6f5d20 | 1805 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1806 | .open = blkdev_open, |
1807 | .release = blkdev_close, | |
1808 | .llseek = block_llseek, | |
a886038b | 1809 | .read_iter = blkdev_read_iter, |
1456c0a8 | 1810 | .write_iter = blkdev_write_iter, |
5a023cdb | 1811 | .mmap = blkdev_mmap, |
b1dd3b28 | 1812 | .fsync = blkdev_fsync, |
bb93e3a5 | 1813 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1814 | #ifdef CONFIG_COMPAT |
1815 | .compat_ioctl = compat_blkdev_ioctl, | |
1816 | #endif | |
1e8b3332 | 1817 | .splice_read = generic_file_splice_read, |
8d020765 | 1818 | .splice_write = iter_file_splice_write, |
1da177e4 LT |
1819 | }; |
1820 | ||
1821 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1822 | { | |
1823 | int res; | |
1824 | mm_segment_t old_fs = get_fs(); | |
1825 | set_fs(KERNEL_DS); | |
56b26add | 1826 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1827 | set_fs(old_fs); |
1828 | return res; | |
1829 | } | |
1830 | ||
1831 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1832 | ||
1833 | /** | |
1834 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1835 | * @pathname: special file representing the block device |
1da177e4 | 1836 | * |
57d1b536 | 1837 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1838 | * namespace if possible and return it. Return ERR_PTR(error) |
1839 | * otherwise. | |
1840 | */ | |
421748ec | 1841 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1842 | { |
1843 | struct block_device *bdev; | |
1844 | struct inode *inode; | |
421748ec | 1845 | struct path path; |
1da177e4 LT |
1846 | int error; |
1847 | ||
421748ec | 1848 | if (!pathname || !*pathname) |
1da177e4 LT |
1849 | return ERR_PTR(-EINVAL); |
1850 | ||
421748ec | 1851 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1852 | if (error) |
1853 | return ERR_PTR(error); | |
1854 | ||
bb668734 | 1855 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
1856 | error = -ENOTBLK; |
1857 | if (!S_ISBLK(inode->i_mode)) | |
1858 | goto fail; | |
1859 | error = -EACCES; | |
421748ec | 1860 | if (path.mnt->mnt_flags & MNT_NODEV) |
1da177e4 LT |
1861 | goto fail; |
1862 | error = -ENOMEM; | |
1863 | bdev = bd_acquire(inode); | |
1864 | if (!bdev) | |
1865 | goto fail; | |
1866 | out: | |
421748ec | 1867 | path_put(&path); |
1da177e4 LT |
1868 | return bdev; |
1869 | fail: | |
1870 | bdev = ERR_PTR(error); | |
1871 | goto out; | |
1872 | } | |
d5686b44 | 1873 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 1874 | |
93b270f7 | 1875 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
1876 | { |
1877 | struct super_block *sb = get_super(bdev); | |
1878 | int res = 0; | |
1879 | ||
1880 | if (sb) { | |
1881 | /* | |
1882 | * no need to lock the super, get_super holds the | |
1883 | * read mutex so the filesystem cannot go away | |
1884 | * under us (->put_super runs with the write lock | |
1885 | * hold). | |
1886 | */ | |
1887 | shrink_dcache_sb(sb); | |
93b270f7 | 1888 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
1889 | drop_super(sb); |
1890 | } | |
f98393a6 | 1891 | invalidate_bdev(bdev); |
b71e8a4c DH |
1892 | return res; |
1893 | } | |
1894 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
1895 | |
1896 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
1897 | { | |
1898 | struct inode *inode, *old_inode = NULL; | |
1899 | ||
74278da9 | 1900 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1901 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
1902 | struct address_space *mapping = inode->i_mapping; | |
1903 | ||
1904 | spin_lock(&inode->i_lock); | |
1905 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
1906 | mapping->nrpages == 0) { | |
1907 | spin_unlock(&inode->i_lock); | |
1908 | continue; | |
1909 | } | |
1910 | __iget(inode); | |
1911 | spin_unlock(&inode->i_lock); | |
74278da9 | 1912 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1913 | /* |
1914 | * We hold a reference to 'inode' so it couldn't have been | |
1915 | * removed from s_inodes list while we dropped the | |
74278da9 | 1916 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 1917 | * be holding the last reference and we cannot iput it under |
74278da9 | 1918 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
1919 | * later. |
1920 | */ | |
1921 | iput(old_inode); | |
1922 | old_inode = inode; | |
1923 | ||
1924 | func(I_BDEV(inode), arg); | |
1925 | ||
74278da9 | 1926 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 1927 | } |
74278da9 | 1928 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1929 | iput(old_inode); |
1930 | } |