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> |
acc93d30 | 32 | #include <linux/badblocks.h> |
1da177e4 | 33 | #include <asm/uaccess.h> |
07f3f05c | 34 | #include "internal.h" |
1da177e4 LT |
35 | |
36 | struct bdev_inode { | |
37 | struct block_device bdev; | |
38 | struct inode vfs_inode; | |
39 | }; | |
40 | ||
4c54ac62 AB |
41 | static const struct address_space_operations def_blk_aops; |
42 | ||
1da177e4 LT |
43 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
44 | { | |
45 | return container_of(inode, struct bdev_inode, vfs_inode); | |
46 | } | |
47 | ||
ff5053f6 | 48 | struct block_device *I_BDEV(struct inode *inode) |
1da177e4 LT |
49 | { |
50 | return &BDEV_I(inode)->bdev; | |
51 | } | |
1da177e4 LT |
52 | EXPORT_SYMBOL(I_BDEV); |
53 | ||
2af3a815 TK |
54 | void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...) |
55 | { | |
56 | struct va_format vaf; | |
57 | va_list args; | |
58 | ||
59 | va_start(args, fmt); | |
60 | vaf.fmt = fmt; | |
61 | vaf.va = &args; | |
62 | printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf); | |
63 | va_end(args); | |
64 | } | |
65 | ||
dbd3ca50 | 66 | static void bdev_write_inode(struct block_device *bdev) |
564f00f6 | 67 | { |
dbd3ca50 VG |
68 | struct inode *inode = bdev->bd_inode; |
69 | int ret; | |
70 | ||
564f00f6 CH |
71 | spin_lock(&inode->i_lock); |
72 | while (inode->i_state & I_DIRTY) { | |
73 | spin_unlock(&inode->i_lock); | |
dbd3ca50 VG |
74 | ret = write_inode_now(inode, true); |
75 | if (ret) { | |
76 | char name[BDEVNAME_SIZE]; | |
77 | pr_warn_ratelimited("VFS: Dirty inode writeback failed " | |
78 | "for block device %s (err=%d).\n", | |
79 | bdevname(bdev, name), ret); | |
80 | } | |
564f00f6 CH |
81 | spin_lock(&inode->i_lock); |
82 | } | |
83 | spin_unlock(&inode->i_lock); | |
84 | } | |
85 | ||
f9a14399 | 86 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
ff01bb48 | 87 | void kill_bdev(struct block_device *bdev) |
1da177e4 | 88 | { |
ff01bb48 AV |
89 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
90 | ||
f9fe48be | 91 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
f9a14399 | 92 | return; |
ff01bb48 | 93 | |
f9a14399 | 94 | invalidate_bh_lrus(); |
ff01bb48 | 95 | truncate_inode_pages(mapping, 0); |
1da177e4 | 96 | } |
ff01bb48 AV |
97 | EXPORT_SYMBOL(kill_bdev); |
98 | ||
99 | /* Invalidate clean unused buffers and pagecache. */ | |
100 | void invalidate_bdev(struct block_device *bdev) | |
101 | { | |
102 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
103 | ||
104 | if (mapping->nrpages == 0) | |
105 | return; | |
106 | ||
107 | invalidate_bh_lrus(); | |
108 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
109 | invalidate_mapping_pages(mapping, 0, -1); | |
110 | /* 99% of the time, we don't need to flush the cleancache on the bdev. | |
111 | * But, for the strange corners, lets be cautious | |
112 | */ | |
3167760f | 113 | cleancache_invalidate_inode(mapping); |
ff01bb48 AV |
114 | } |
115 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 LT |
116 | |
117 | int set_blocksize(struct block_device *bdev, int size) | |
118 | { | |
119 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 120 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
121 | return -EINVAL; |
122 | ||
123 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 124 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
125 | return -EINVAL; |
126 | ||
127 | /* Don't change the size if it is same as current */ | |
128 | if (bdev->bd_block_size != size) { | |
129 | sync_blockdev(bdev); | |
130 | bdev->bd_block_size = size; | |
131 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
132 | kill_bdev(bdev); | |
133 | } | |
134 | return 0; | |
135 | } | |
136 | ||
137 | EXPORT_SYMBOL(set_blocksize); | |
138 | ||
139 | int sb_set_blocksize(struct super_block *sb, int size) | |
140 | { | |
1da177e4 LT |
141 | if (set_blocksize(sb->s_bdev, size)) |
142 | return 0; | |
143 | /* If we get here, we know size is power of two | |
144 | * and it's value is between 512 and PAGE_SIZE */ | |
145 | sb->s_blocksize = size; | |
38885bd4 | 146 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
147 | return sb->s_blocksize; |
148 | } | |
149 | ||
150 | EXPORT_SYMBOL(sb_set_blocksize); | |
151 | ||
152 | int sb_min_blocksize(struct super_block *sb, int size) | |
153 | { | |
e1defc4f | 154 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
155 | if (size < minsize) |
156 | size = minsize; | |
157 | return sb_set_blocksize(sb, size); | |
158 | } | |
159 | ||
160 | EXPORT_SYMBOL(sb_min_blocksize); | |
161 | ||
162 | static int | |
163 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
164 | struct buffer_head *bh, int create) | |
165 | { | |
1da177e4 LT |
166 | bh->b_bdev = I_BDEV(inode); |
167 | bh->b_blocknr = iblock; | |
168 | set_buffer_mapped(bh); | |
169 | return 0; | |
170 | } | |
171 | ||
4ebb16ca DW |
172 | static struct inode *bdev_file_inode(struct file *file) |
173 | { | |
174 | return file->f_mapping->host; | |
175 | } | |
176 | ||
b2e895db | 177 | static ssize_t |
c8b8e32d | 178 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
b2e895db AM |
179 | { |
180 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 181 | struct inode *inode = bdev_file_inode(file); |
b2e895db | 182 | |
bbab37dd | 183 | if (IS_DAX(inode)) |
c8b8e32d | 184 | return dax_do_io(iocb, inode, iter, blkdev_get_block, |
bbab37dd | 185 | NULL, DIO_SKIP_DIO_COUNT); |
c8b8e32d | 186 | return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, |
fe0f07d0 JA |
187 | blkdev_get_block, NULL, NULL, |
188 | DIO_SKIP_DIO_COUNT); | |
b2e895db AM |
189 | } |
190 | ||
5cee5815 JK |
191 | int __sync_blockdev(struct block_device *bdev, int wait) |
192 | { | |
193 | if (!bdev) | |
194 | return 0; | |
195 | if (!wait) | |
196 | return filemap_flush(bdev->bd_inode->i_mapping); | |
197 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
198 | } | |
199 | ||
585d3bc0 NP |
200 | /* |
201 | * Write out and wait upon all the dirty data associated with a block | |
202 | * device via its mapping. Does not take the superblock lock. | |
203 | */ | |
204 | int sync_blockdev(struct block_device *bdev) | |
205 | { | |
5cee5815 | 206 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
207 | } |
208 | EXPORT_SYMBOL(sync_blockdev); | |
209 | ||
210 | /* | |
211 | * Write out and wait upon all dirty data associated with this | |
212 | * device. Filesystem data as well as the underlying block | |
213 | * device. Takes the superblock lock. | |
214 | */ | |
215 | int fsync_bdev(struct block_device *bdev) | |
216 | { | |
217 | struct super_block *sb = get_super(bdev); | |
218 | if (sb) { | |
60b0680f | 219 | int res = sync_filesystem(sb); |
585d3bc0 NP |
220 | drop_super(sb); |
221 | return res; | |
222 | } | |
223 | return sync_blockdev(bdev); | |
224 | } | |
47e4491b | 225 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
226 | |
227 | /** | |
228 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
229 | * @bdev: blockdevice to lock | |
230 | * | |
585d3bc0 NP |
231 | * If a superblock is found on this device, we take the s_umount semaphore |
232 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
233 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
234 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
235 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
236 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
237 | * actually. | |
238 | */ | |
239 | struct super_block *freeze_bdev(struct block_device *bdev) | |
240 | { | |
241 | struct super_block *sb; | |
242 | int error = 0; | |
243 | ||
244 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
245 | if (++bdev->bd_fsfreeze_count > 1) { |
246 | /* | |
247 | * We don't even need to grab a reference - the first call | |
248 | * to freeze_bdev grab an active reference and only the last | |
249 | * thaw_bdev drops it. | |
250 | */ | |
585d3bc0 | 251 | sb = get_super(bdev); |
4504230a CH |
252 | drop_super(sb); |
253 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
254 | return sb; | |
255 | } | |
256 | ||
257 | sb = get_active_super(bdev); | |
258 | if (!sb) | |
259 | goto out; | |
48b6bca6 BM |
260 | if (sb->s_op->freeze_super) |
261 | error = sb->s_op->freeze_super(sb); | |
262 | else | |
263 | error = freeze_super(sb); | |
18e9e510 JB |
264 | if (error) { |
265 | deactivate_super(sb); | |
266 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 267 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 268 | return ERR_PTR(error); |
585d3bc0 | 269 | } |
18e9e510 | 270 | deactivate_super(sb); |
4504230a | 271 | out: |
585d3bc0 NP |
272 | sync_blockdev(bdev); |
273 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 274 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
275 | } |
276 | EXPORT_SYMBOL(freeze_bdev); | |
277 | ||
278 | /** | |
279 | * thaw_bdev -- unlock filesystem | |
280 | * @bdev: blockdevice to unlock | |
281 | * @sb: associated superblock | |
282 | * | |
283 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
284 | */ | |
285 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
286 | { | |
4504230a | 287 | int error = -EINVAL; |
585d3bc0 NP |
288 | |
289 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 290 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 291 | goto out; |
4504230a CH |
292 | |
293 | error = 0; | |
294 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 295 | goto out; |
4504230a CH |
296 | |
297 | if (!sb) | |
18e9e510 | 298 | goto out; |
4504230a | 299 | |
48b6bca6 BM |
300 | if (sb->s_op->thaw_super) |
301 | error = sb->s_op->thaw_super(sb); | |
302 | else | |
303 | error = thaw_super(sb); | |
18e9e510 JB |
304 | if (error) { |
305 | bdev->bd_fsfreeze_count++; | |
306 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
307 | return error; | |
308 | } | |
309 | out: | |
585d3bc0 NP |
310 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
311 | return 0; | |
312 | } | |
313 | EXPORT_SYMBOL(thaw_bdev); | |
314 | ||
1da177e4 LT |
315 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
316 | { | |
317 | return block_write_full_page(page, blkdev_get_block, wbc); | |
318 | } | |
319 | ||
320 | static int blkdev_readpage(struct file * file, struct page * page) | |
321 | { | |
322 | return block_read_full_page(page, blkdev_get_block); | |
323 | } | |
324 | ||
447f05bb AM |
325 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
326 | struct list_head *pages, unsigned nr_pages) | |
327 | { | |
328 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
329 | } | |
330 | ||
6272b5a5 NP |
331 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
332 | loff_t pos, unsigned len, unsigned flags, | |
333 | struct page **pagep, void **fsdata) | |
1da177e4 | 334 | { |
155130a4 CH |
335 | return block_write_begin(mapping, pos, len, flags, pagep, |
336 | blkdev_get_block); | |
1da177e4 LT |
337 | } |
338 | ||
6272b5a5 NP |
339 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
340 | loff_t pos, unsigned len, unsigned copied, | |
341 | struct page *page, void *fsdata) | |
1da177e4 | 342 | { |
6272b5a5 NP |
343 | int ret; |
344 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
345 | ||
346 | unlock_page(page); | |
09cbfeaf | 347 | put_page(page); |
6272b5a5 NP |
348 | |
349 | return ret; | |
1da177e4 LT |
350 | } |
351 | ||
352 | /* | |
353 | * private llseek: | |
496ad9aa | 354 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
355 | * so we compute the size by hand (just as in block_read/write above) |
356 | */ | |
965c8e59 | 357 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 358 | { |
4ebb16ca | 359 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
360 | loff_t retval; |
361 | ||
5955102c | 362 | inode_lock(bd_inode); |
5d48f3a2 | 363 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 364 | inode_unlock(bd_inode); |
1da177e4 LT |
365 | return retval; |
366 | } | |
367 | ||
02c24a82 | 368 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 369 | { |
4ebb16ca | 370 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 371 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 372 | int error; |
da5aa861 RW |
373 | |
374 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
375 | if (error) | |
376 | return error; | |
ab0a9735 | 377 | |
b8af67e2 AB |
378 | /* |
379 | * There is no need to serialise calls to blkdev_issue_flush with | |
380 | * i_mutex and doing so causes performance issues with concurrent | |
381 | * O_SYNC writers to a block device. | |
382 | */ | |
dd3932ed | 383 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
384 | if (error == -EOPNOTSUPP) |
385 | error = 0; | |
b8af67e2 | 386 | |
ab0a9735 | 387 | return error; |
1da177e4 | 388 | } |
b1dd3b28 | 389 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 390 | |
47a191fd MW |
391 | /** |
392 | * bdev_read_page() - Start reading a page from a block device | |
393 | * @bdev: The device to read the page from | |
394 | * @sector: The offset on the device to read the page to (need not be aligned) | |
395 | * @page: The page to read | |
396 | * | |
397 | * On entry, the page should be locked. It will be unlocked when the page | |
398 | * has been read. If the block driver implements rw_page synchronously, | |
399 | * that will be true on exit from this function, but it need not be. | |
400 | * | |
401 | * Errors returned by this function are usually "soft", eg out of memory, or | |
402 | * queue full; callers should try a different route to read this page rather | |
403 | * than propagate an error back up the stack. | |
404 | * | |
405 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
406 | */ | |
407 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
408 | struct page *page) | |
409 | { | |
410 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
411 | int result = -EOPNOTSUPP; |
412 | ||
f68eb1e7 | 413 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
414 | return result; |
415 | ||
6f3b0e8b | 416 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
417 | if (result) |
418 | return result; | |
c11f0c0b | 419 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false); |
2e6edc95 DW |
420 | blk_queue_exit(bdev->bd_queue); |
421 | return result; | |
47a191fd MW |
422 | } |
423 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
424 | ||
425 | /** | |
426 | * bdev_write_page() - Start writing a page to a block device | |
427 | * @bdev: The device to write the page to | |
428 | * @sector: The offset on the device to write the page to (need not be aligned) | |
429 | * @page: The page to write | |
430 | * @wbc: The writeback_control for the write | |
431 | * | |
432 | * On entry, the page should be locked and not currently under writeback. | |
433 | * On exit, if the write started successfully, the page will be unlocked and | |
434 | * under writeback. If the write failed already (eg the driver failed to | |
435 | * queue the page to the device), the page will still be locked. If the | |
436 | * caller is a ->writepage implementation, it will need to unlock the page. | |
437 | * | |
438 | * Errors returned by this function are usually "soft", eg out of memory, or | |
439 | * queue full; callers should try a different route to write this page rather | |
440 | * than propagate an error back up the stack. | |
441 | * | |
442 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
443 | */ | |
444 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
445 | struct page *page, struct writeback_control *wbc) | |
446 | { | |
447 | int result; | |
47a191fd | 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 | 456 | set_page_writeback(page); |
c11f0c0b | 457 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true); |
47a191fd MW |
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; | |
163d4baa | 495 | if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access) |
dd22f551 MW |
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; | |
0a70bd43 | 503 | avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size); |
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 | ||
2d96afc8 TK |
512 | /** |
513 | * bdev_dax_supported() - Check if the device supports dax for filesystem | |
514 | * @sb: The superblock of the device | |
515 | * @blocksize: The block size of the device | |
516 | * | |
517 | * This is a library function for filesystems to check if the block device | |
518 | * can be mounted with dax option. | |
519 | * | |
520 | * Return: negative errno if unsupported, 0 if supported. | |
521 | */ | |
522 | int bdev_dax_supported(struct super_block *sb, int blocksize) | |
523 | { | |
524 | struct blk_dax_ctl dax = { | |
525 | .sector = 0, | |
526 | .size = PAGE_SIZE, | |
527 | }; | |
528 | int err; | |
529 | ||
530 | if (blocksize != PAGE_SIZE) { | |
531 | vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax"); | |
532 | return -EINVAL; | |
533 | } | |
534 | ||
535 | err = bdev_direct_access(sb->s_bdev, &dax); | |
536 | if (err < 0) { | |
537 | switch (err) { | |
538 | case -EOPNOTSUPP: | |
539 | vfs_msg(sb, KERN_ERR, | |
540 | "error: device does not support dax"); | |
541 | break; | |
542 | case -EINVAL: | |
543 | vfs_msg(sb, KERN_ERR, | |
544 | "error: unaligned partition for dax"); | |
545 | break; | |
546 | default: | |
547 | vfs_msg(sb, KERN_ERR, | |
548 | "error: dax access failed (%d)", err); | |
549 | } | |
550 | return err; | |
551 | } | |
552 | ||
553 | return 0; | |
554 | } | |
555 | EXPORT_SYMBOL_GPL(bdev_dax_supported); | |
556 | ||
a8078b1f TK |
557 | /** |
558 | * bdev_dax_capable() - Return if the raw device is capable for dax | |
559 | * @bdev: The device for raw block device access | |
560 | */ | |
561 | bool bdev_dax_capable(struct block_device *bdev) | |
562 | { | |
a8078b1f TK |
563 | struct blk_dax_ctl dax = { |
564 | .size = PAGE_SIZE, | |
565 | }; | |
566 | ||
567 | if (!IS_ENABLED(CONFIG_FS_DAX)) | |
568 | return false; | |
569 | ||
570 | dax.sector = 0; | |
571 | if (bdev_direct_access(bdev, &dax) < 0) | |
572 | return false; | |
573 | ||
574 | dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512); | |
575 | if (bdev_direct_access(bdev, &dax) < 0) | |
576 | return false; | |
577 | ||
a8078b1f TK |
578 | return true; |
579 | } | |
580 | ||
1da177e4 LT |
581 | /* |
582 | * pseudo-fs | |
583 | */ | |
584 | ||
585 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 586 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
587 | |
588 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
589 | { | |
e94b1766 | 590 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
591 | if (!ei) |
592 | return NULL; | |
593 | return &ei->vfs_inode; | |
594 | } | |
595 | ||
fa0d7e3d | 596 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 597 | { |
fa0d7e3d | 598 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
599 | struct bdev_inode *bdi = BDEV_I(inode); |
600 | ||
1da177e4 LT |
601 | kmem_cache_free(bdev_cachep, bdi); |
602 | } | |
603 | ||
fa0d7e3d NP |
604 | static void bdev_destroy_inode(struct inode *inode) |
605 | { | |
606 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
607 | } | |
608 | ||
51cc5068 | 609 | static void init_once(void *foo) |
1da177e4 LT |
610 | { |
611 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
612 | struct block_device *bdev = &ei->bdev; | |
613 | ||
a35afb83 CL |
614 | memset(bdev, 0, sizeof(*bdev)); |
615 | mutex_init(&bdev->bd_mutex); | |
a35afb83 | 616 | INIT_LIST_HEAD(&bdev->bd_list); |
49731baa TH |
617 | #ifdef CONFIG_SYSFS |
618 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
619 | #endif | |
a35afb83 | 620 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
621 | /* Initialize mutex for freeze. */ |
622 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
623 | } |
624 | ||
b57922d9 | 625 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
626 | { |
627 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
91b0abe3 | 628 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 629 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 630 | clear_inode(inode); |
1da177e4 | 631 | spin_lock(&bdev_lock); |
1da177e4 LT |
632 | list_del_init(&bdev->bd_list); |
633 | spin_unlock(&bdev_lock); | |
634 | } | |
635 | ||
ee9b6d61 | 636 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
637 | .statfs = simple_statfs, |
638 | .alloc_inode = bdev_alloc_inode, | |
639 | .destroy_inode = bdev_destroy_inode, | |
640 | .drop_inode = generic_delete_inode, | |
b57922d9 | 641 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
642 | }; |
643 | ||
51139ada AV |
644 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
645 | int flags, const char *dev_name, void *data) | |
1da177e4 | 646 | { |
3684aa70 SL |
647 | struct dentry *dent; |
648 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); | |
649 | if (dent) | |
650 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; | |
651 | return dent; | |
1da177e4 LT |
652 | } |
653 | ||
654 | static struct file_system_type bd_type = { | |
655 | .name = "bdev", | |
51139ada | 656 | .mount = bd_mount, |
1da177e4 LT |
657 | .kill_sb = kill_anon_super, |
658 | }; | |
659 | ||
a212b105 TH |
660 | struct super_block *blockdev_superblock __read_mostly; |
661 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
662 | |
663 | void __init bdev_cache_init(void) | |
664 | { | |
665 | int err; | |
ace8577a | 666 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 667 | |
1da177e4 | 668 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 669 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 670 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 671 | init_once); |
1da177e4 LT |
672 | err = register_filesystem(&bd_type); |
673 | if (err) | |
674 | panic("Cannot register bdev pseudo-fs"); | |
675 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
676 | if (IS_ERR(bd_mnt)) |
677 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 678 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
679 | } |
680 | ||
681 | /* | |
682 | * Most likely _very_ bad one - but then it's hardly critical for small | |
683 | * /dev and can be fixed when somebody will need really large one. | |
684 | * Keep in mind that it will be fed through icache hash function too. | |
685 | */ | |
686 | static inline unsigned long hash(dev_t dev) | |
687 | { | |
688 | return MAJOR(dev)+MINOR(dev); | |
689 | } | |
690 | ||
691 | static int bdev_test(struct inode *inode, void *data) | |
692 | { | |
693 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
694 | } | |
695 | ||
696 | static int bdev_set(struct inode *inode, void *data) | |
697 | { | |
698 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
699 | return 0; | |
700 | } | |
701 | ||
702 | static LIST_HEAD(all_bdevs); | |
703 | ||
704 | struct block_device *bdget(dev_t dev) | |
705 | { | |
706 | struct block_device *bdev; | |
707 | struct inode *inode; | |
708 | ||
c2acf7b9 | 709 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
710 | bdev_test, bdev_set, &dev); |
711 | ||
712 | if (!inode) | |
713 | return NULL; | |
714 | ||
715 | bdev = &BDEV_I(inode)->bdev; | |
716 | ||
717 | if (inode->i_state & I_NEW) { | |
718 | bdev->bd_contains = NULL; | |
782b94cd | 719 | bdev->bd_super = NULL; |
1da177e4 LT |
720 | bdev->bd_inode = inode; |
721 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
722 | bdev->bd_part_count = 0; | |
723 | bdev->bd_invalidated = 0; | |
724 | inode->i_mode = S_IFBLK; | |
725 | inode->i_rdev = dev; | |
726 | inode->i_bdev = bdev; | |
727 | inode->i_data.a_ops = &def_blk_aops; | |
728 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
729 | spin_lock(&bdev_lock); |
730 | list_add(&bdev->bd_list, &all_bdevs); | |
731 | spin_unlock(&bdev_lock); | |
732 | unlock_new_inode(inode); | |
733 | } | |
734 | return bdev; | |
735 | } | |
736 | ||
737 | EXPORT_SYMBOL(bdget); | |
738 | ||
dddac6a7 AJ |
739 | /** |
740 | * bdgrab -- Grab a reference to an already referenced block device | |
741 | * @bdev: Block device to grab a reference to. | |
742 | */ | |
743 | struct block_device *bdgrab(struct block_device *bdev) | |
744 | { | |
7de9c6ee | 745 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
746 | return bdev; |
747 | } | |
c1681bf8 | 748 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 749 | |
1da177e4 LT |
750 | long nr_blockdev_pages(void) |
751 | { | |
203a2935 | 752 | struct block_device *bdev; |
1da177e4 LT |
753 | long ret = 0; |
754 | spin_lock(&bdev_lock); | |
203a2935 | 755 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
756 | ret += bdev->bd_inode->i_mapping->nrpages; |
757 | } | |
758 | spin_unlock(&bdev_lock); | |
759 | return ret; | |
760 | } | |
761 | ||
762 | void bdput(struct block_device *bdev) | |
763 | { | |
764 | iput(bdev->bd_inode); | |
765 | } | |
766 | ||
767 | EXPORT_SYMBOL(bdput); | |
768 | ||
769 | static struct block_device *bd_acquire(struct inode *inode) | |
770 | { | |
771 | struct block_device *bdev; | |
09d967c6 | 772 | |
1da177e4 LT |
773 | spin_lock(&bdev_lock); |
774 | bdev = inode->i_bdev; | |
09d967c6 | 775 | if (bdev) { |
ed8a9d2c | 776 | bdgrab(bdev); |
1da177e4 LT |
777 | spin_unlock(&bdev_lock); |
778 | return bdev; | |
779 | } | |
780 | spin_unlock(&bdev_lock); | |
09d967c6 | 781 | |
1da177e4 LT |
782 | bdev = bdget(inode->i_rdev); |
783 | if (bdev) { | |
784 | spin_lock(&bdev_lock); | |
09d967c6 OH |
785 | if (!inode->i_bdev) { |
786 | /* | |
7de9c6ee | 787 | * We take an additional reference to bd_inode, |
09d967c6 OH |
788 | * and it's released in clear_inode() of inode. |
789 | * So, we can access it via ->i_mapping always | |
790 | * without igrab(). | |
791 | */ | |
ed8a9d2c | 792 | bdgrab(bdev); |
09d967c6 OH |
793 | inode->i_bdev = bdev; |
794 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
09d967c6 | 795 | } |
1da177e4 LT |
796 | spin_unlock(&bdev_lock); |
797 | } | |
798 | return bdev; | |
799 | } | |
800 | ||
801 | /* Call when you free inode */ | |
802 | ||
803 | void bd_forget(struct inode *inode) | |
804 | { | |
09d967c6 OH |
805 | struct block_device *bdev = NULL; |
806 | ||
1da177e4 | 807 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
808 | if (!sb_is_blkdev_sb(inode->i_sb)) |
809 | bdev = inode->i_bdev; | |
a4a4f943 AV |
810 | inode->i_bdev = NULL; |
811 | inode->i_mapping = &inode->i_data; | |
1da177e4 | 812 | spin_unlock(&bdev_lock); |
09d967c6 OH |
813 | |
814 | if (bdev) | |
ed8a9d2c | 815 | bdput(bdev); |
1da177e4 LT |
816 | } |
817 | ||
1a3cbbc5 TH |
818 | /** |
819 | * bd_may_claim - test whether a block device can be claimed | |
820 | * @bdev: block device of interest | |
821 | * @whole: whole block device containing @bdev, may equal @bdev | |
822 | * @holder: holder trying to claim @bdev | |
823 | * | |
25985edc | 824 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
825 | * |
826 | * CONTEXT: | |
827 | * spin_lock(&bdev_lock). | |
828 | * | |
829 | * RETURNS: | |
830 | * %true if @bdev can be claimed, %false otherwise. | |
831 | */ | |
832 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
833 | void *holder) | |
1da177e4 | 834 | { |
1da177e4 | 835 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 836 | return true; /* already a holder */ |
1da177e4 | 837 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 838 | return false; /* held by someone else */ |
1da177e4 | 839 | else if (bdev->bd_contains == bdev) |
1a3cbbc5 | 840 | return true; /* is a whole device which isn't held */ |
1da177e4 | 841 | |
e525fd89 | 842 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
843 | return true; /* is a partition of a device that is being partitioned */ |
844 | else if (whole->bd_holder != NULL) | |
845 | return false; /* is a partition of a held device */ | |
1da177e4 | 846 | else |
1a3cbbc5 TH |
847 | return true; /* is a partition of an un-held device */ |
848 | } | |
849 | ||
6b4517a7 TH |
850 | /** |
851 | * bd_prepare_to_claim - prepare to claim a block device | |
852 | * @bdev: block device of interest | |
853 | * @whole: the whole device containing @bdev, may equal @bdev | |
854 | * @holder: holder trying to claim @bdev | |
855 | * | |
856 | * Prepare to claim @bdev. This function fails if @bdev is already | |
857 | * claimed by another holder and waits if another claiming is in | |
858 | * progress. This function doesn't actually claim. On successful | |
859 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
860 | * | |
861 | * CONTEXT: | |
862 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
863 | * it multiple times. | |
864 | * | |
865 | * RETURNS: | |
866 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
867 | */ | |
868 | static int bd_prepare_to_claim(struct block_device *bdev, | |
869 | struct block_device *whole, void *holder) | |
870 | { | |
871 | retry: | |
872 | /* if someone else claimed, fail */ | |
873 | if (!bd_may_claim(bdev, whole, holder)) | |
874 | return -EBUSY; | |
875 | ||
e75aa858 TH |
876 | /* if claiming is already in progress, wait for it to finish */ |
877 | if (whole->bd_claiming) { | |
6b4517a7 TH |
878 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
879 | DEFINE_WAIT(wait); | |
880 | ||
881 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
882 | spin_unlock(&bdev_lock); | |
883 | schedule(); | |
884 | finish_wait(wq, &wait); | |
885 | spin_lock(&bdev_lock); | |
886 | goto retry; | |
887 | } | |
888 | ||
889 | /* yay, all mine */ | |
890 | return 0; | |
891 | } | |
892 | ||
893 | /** | |
894 | * bd_start_claiming - start claiming a block device | |
895 | * @bdev: block device of interest | |
896 | * @holder: holder trying to claim @bdev | |
897 | * | |
898 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
899 | * exclusively and mark that an exclusive open is in progress. Each | |
900 | * successful call to this function must be matched with a call to | |
b0018361 NP |
901 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
902 | * fail). | |
903 | * | |
904 | * This function is used to gain exclusive access to the block device | |
905 | * without actually causing other exclusive open attempts to fail. It | |
906 | * should be used when the open sequence itself requires exclusive | |
907 | * access but may subsequently fail. | |
6b4517a7 TH |
908 | * |
909 | * CONTEXT: | |
910 | * Might sleep. | |
911 | * | |
912 | * RETURNS: | |
913 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
914 | * value on failure. | |
915 | */ | |
916 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
917 | void *holder) | |
918 | { | |
919 | struct gendisk *disk; | |
920 | struct block_device *whole; | |
921 | int partno, err; | |
922 | ||
923 | might_sleep(); | |
924 | ||
925 | /* | |
926 | * @bdev might not have been initialized properly yet, look up | |
927 | * and grab the outer block device the hard way. | |
928 | */ | |
929 | disk = get_gendisk(bdev->bd_dev, &partno); | |
930 | if (!disk) | |
931 | return ERR_PTR(-ENXIO); | |
932 | ||
d4c208b8 TH |
933 | /* |
934 | * Normally, @bdev should equal what's returned from bdget_disk() | |
935 | * if partno is 0; however, some drivers (floppy) use multiple | |
936 | * bdev's for the same physical device and @bdev may be one of the | |
937 | * aliases. Keep @bdev if partno is 0. This means claimer | |
938 | * tracking is broken for those devices but it has always been that | |
939 | * way. | |
940 | */ | |
941 | if (partno) | |
942 | whole = bdget_disk(disk, 0); | |
943 | else | |
944 | whole = bdgrab(bdev); | |
945 | ||
cf342570 | 946 | module_put(disk->fops->owner); |
6b4517a7 TH |
947 | put_disk(disk); |
948 | if (!whole) | |
949 | return ERR_PTR(-ENOMEM); | |
950 | ||
951 | /* prepare to claim, if successful, mark claiming in progress */ | |
952 | spin_lock(&bdev_lock); | |
953 | ||
954 | err = bd_prepare_to_claim(bdev, whole, holder); | |
955 | if (err == 0) { | |
956 | whole->bd_claiming = holder; | |
957 | spin_unlock(&bdev_lock); | |
958 | return whole; | |
959 | } else { | |
960 | spin_unlock(&bdev_lock); | |
961 | bdput(whole); | |
962 | return ERR_PTR(err); | |
963 | } | |
964 | } | |
965 | ||
641dc636 | 966 | #ifdef CONFIG_SYSFS |
49731baa TH |
967 | struct bd_holder_disk { |
968 | struct list_head list; | |
969 | struct gendisk *disk; | |
970 | int refcnt; | |
971 | }; | |
972 | ||
973 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
974 | struct gendisk *disk) | |
975 | { | |
976 | struct bd_holder_disk *holder; | |
977 | ||
978 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
979 | if (holder->disk == disk) | |
980 | return holder; | |
981 | return NULL; | |
982 | } | |
983 | ||
4d7dd8fd | 984 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 985 | { |
4d7dd8fd | 986 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
987 | } |
988 | ||
989 | static void del_symlink(struct kobject *from, struct kobject *to) | |
990 | { | |
641dc636 JN |
991 | sysfs_remove_link(from, kobject_name(to)); |
992 | } | |
993 | ||
df6c0cd9 | 994 | /** |
e09b457b TH |
995 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
996 | * @bdev: the claimed slave bdev | |
997 | * @disk: the holding disk | |
df6c0cd9 | 998 | * |
49731baa TH |
999 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1000 | * | |
e09b457b | 1001 | * This functions creates the following sysfs symlinks. |
641dc636 | 1002 | * |
e09b457b TH |
1003 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1004 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 1005 | * |
e09b457b TH |
1006 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1007 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 1008 | * |
e09b457b TH |
1009 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1010 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 1011 | * |
e09b457b TH |
1012 | * The caller must have claimed @bdev before calling this function and |
1013 | * ensure that both @bdev and @disk are valid during the creation and | |
1014 | * lifetime of these symlinks. | |
641dc636 | 1015 | * |
e09b457b TH |
1016 | * CONTEXT: |
1017 | * Might sleep. | |
641dc636 | 1018 | * |
e09b457b TH |
1019 | * RETURNS: |
1020 | * 0 on success, -errno on failure. | |
641dc636 | 1021 | */ |
e09b457b | 1022 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 1023 | { |
49731baa | 1024 | struct bd_holder_disk *holder; |
e09b457b | 1025 | int ret = 0; |
641dc636 | 1026 | |
2e7b651d | 1027 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 1028 | |
49731baa | 1029 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 1030 | |
e09b457b TH |
1031 | /* FIXME: remove the following once add_disk() handles errors */ |
1032 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
1033 | goto out_unlock; | |
4e91672c | 1034 | |
49731baa TH |
1035 | holder = bd_find_holder_disk(bdev, disk); |
1036 | if (holder) { | |
1037 | holder->refcnt++; | |
e09b457b | 1038 | goto out_unlock; |
49731baa | 1039 | } |
641dc636 | 1040 | |
49731baa TH |
1041 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1042 | if (!holder) { | |
1043 | ret = -ENOMEM; | |
e09b457b TH |
1044 | goto out_unlock; |
1045 | } | |
641dc636 | 1046 | |
49731baa TH |
1047 | INIT_LIST_HEAD(&holder->list); |
1048 | holder->disk = disk; | |
1049 | holder->refcnt = 1; | |
1050 | ||
1051 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1052 | if (ret) | |
1053 | goto out_free; | |
1054 | ||
1055 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
1056 | if (ret) | |
1057 | goto out_del; | |
e7407d16 TH |
1058 | /* |
1059 | * bdev could be deleted beneath us which would implicitly destroy | |
1060 | * the holder directory. Hold on to it. | |
1061 | */ | |
1062 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1063 | |
1064 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1065 | goto out_unlock; | |
1066 | ||
1067 | out_del: | |
1068 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1069 | out_free: | |
1070 | kfree(holder); | |
e09b457b | 1071 | out_unlock: |
b4cf1b72 | 1072 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1073 | return ret; |
641dc636 | 1074 | } |
e09b457b | 1075 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1076 | |
49731baa TH |
1077 | /** |
1078 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1079 | * @bdev: the calimed slave bdev | |
1080 | * @disk: the holding disk | |
1081 | * | |
1082 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1083 | * | |
1084 | * CONTEXT: | |
1085 | * Might sleep. | |
1086 | */ | |
1087 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1088 | { |
49731baa | 1089 | struct bd_holder_disk *holder; |
641dc636 | 1090 | |
49731baa | 1091 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1092 | |
49731baa TH |
1093 | holder = bd_find_holder_disk(bdev, disk); |
1094 | ||
1095 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1096 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1097 | del_symlink(bdev->bd_part->holder_dir, | |
1098 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1099 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1100 | list_del_init(&holder->list); |
1101 | kfree(holder); | |
1102 | } | |
1103 | ||
1104 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1105 | } |
49731baa | 1106 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1107 | #endif |
1da177e4 | 1108 | |
56ade44b AP |
1109 | /** |
1110 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1111 | * | |
1112 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1113 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1114 | * |
1115 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1116 | * when a disk has been changed -- either by a media change or online | |
1117 | * resize. | |
1118 | */ | |
93b270f7 | 1119 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1120 | { |
93b270f7 | 1121 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1122 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1123 | "resized disk %s\n", |
1124 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b AP |
1125 | } |
1126 | ||
1127 | if (!bdev->bd_disk) | |
1128 | return; | |
d27769ec | 1129 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1130 | bdev->bd_invalidated = 1; |
1131 | } | |
1132 | ||
c3279d14 | 1133 | /** |
57d1b536 | 1134 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1135 | * @disk: struct gendisk to check |
1136 | * @bdev: struct bdev to adjust. | |
1137 | * | |
1138 | * This routine checks to see if the bdev size does not match the disk size | |
1139 | * and adjusts it if it differs. | |
1140 | */ | |
1141 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1142 | { | |
1143 | loff_t disk_size, bdev_size; | |
1144 | ||
1145 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1146 | bdev_size = i_size_read(bdev->bd_inode); | |
1147 | if (disk_size != bdev_size) { | |
c3279d14 AP |
1148 | printk(KERN_INFO |
1149 | "%s: detected capacity change from %lld to %lld\n", | |
424081f3 | 1150 | disk->disk_name, bdev_size, disk_size); |
c3279d14 | 1151 | i_size_write(bdev->bd_inode, disk_size); |
93b270f7 | 1152 | flush_disk(bdev, false); |
c3279d14 AP |
1153 | } |
1154 | } | |
1155 | EXPORT_SYMBOL(check_disk_size_change); | |
1156 | ||
0c002c2f | 1157 | /** |
57d1b536 | 1158 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1159 | * @disk: struct gendisk to be revalidated |
1160 | * | |
1161 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1162 | * call-backs. It is used to do common pre and post operations needed | |
1163 | * for all revalidate_disk operations. | |
1164 | */ | |
1165 | int revalidate_disk(struct gendisk *disk) | |
1166 | { | |
c3279d14 | 1167 | struct block_device *bdev; |
0c002c2f AP |
1168 | int ret = 0; |
1169 | ||
1170 | if (disk->fops->revalidate_disk) | |
1171 | ret = disk->fops->revalidate_disk(disk); | |
25520d55 | 1172 | blk_integrity_revalidate(disk); |
c3279d14 AP |
1173 | bdev = bdget_disk(disk, 0); |
1174 | if (!bdev) | |
1175 | return ret; | |
1176 | ||
1177 | mutex_lock(&bdev->bd_mutex); | |
1178 | check_disk_size_change(disk, bdev); | |
7630b661 | 1179 | bdev->bd_invalidated = 0; |
c3279d14 AP |
1180 | mutex_unlock(&bdev->bd_mutex); |
1181 | bdput(bdev); | |
0c002c2f AP |
1182 | return ret; |
1183 | } | |
1184 | EXPORT_SYMBOL(revalidate_disk); | |
1185 | ||
1da177e4 LT |
1186 | /* |
1187 | * This routine checks whether a removable media has been changed, | |
1188 | * and invalidates all buffer-cache-entries in that case. This | |
1189 | * is a relatively slow routine, so we have to try to minimize using | |
1190 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1191 | * is the best way of combining speed and utility, I think. | |
1192 | * People changing diskettes in the middle of an operation deserve | |
1193 | * to lose :-) | |
1194 | */ | |
1195 | int check_disk_change(struct block_device *bdev) | |
1196 | { | |
1197 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1198 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1199 | unsigned int events; |
1da177e4 | 1200 | |
77ea887e TH |
1201 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1202 | DISK_EVENT_EJECT_REQUEST); | |
1203 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1204 | return 0; |
1205 | ||
93b270f7 | 1206 | flush_disk(bdev, true); |
1da177e4 LT |
1207 | if (bdops->revalidate_disk) |
1208 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1209 | return 1; |
1210 | } | |
1211 | ||
1212 | EXPORT_SYMBOL(check_disk_change); | |
1213 | ||
1214 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1215 | { | |
e1defc4f | 1216 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1217 | |
5955102c | 1218 | inode_lock(bdev->bd_inode); |
d646a02a | 1219 | i_size_write(bdev->bd_inode, size); |
5955102c | 1220 | inode_unlock(bdev->bd_inode); |
09cbfeaf | 1221 | while (bsize < PAGE_SIZE) { |
1da177e4 LT |
1222 | if (size & bsize) |
1223 | break; | |
1224 | bsize <<= 1; | |
1225 | } | |
1226 | bdev->bd_block_size = bsize; | |
1227 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1228 | } | |
1229 | EXPORT_SYMBOL(bd_set_size); | |
1230 | ||
4385bab1 | 1231 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1232 | |
6d740cd5 PZ |
1233 | /* |
1234 | * bd_mutex locking: | |
1235 | * | |
1236 | * mutex_lock(part->bd_mutex) | |
1237 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1238 | */ | |
1239 | ||
572c4892 | 1240 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1241 | { |
1da177e4 | 1242 | struct gendisk *disk; |
523e1d39 | 1243 | struct module *owner; |
7db9cfd3 | 1244 | int ret; |
cf771cb5 | 1245 | int partno; |
fe6e9c1f AV |
1246 | int perm = 0; |
1247 | ||
572c4892 | 1248 | if (mode & FMODE_READ) |
fe6e9c1f | 1249 | perm |= MAY_READ; |
572c4892 | 1250 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1251 | perm |= MAY_WRITE; |
1252 | /* | |
1253 | * hooks: /n/, see "layering violations". | |
1254 | */ | |
b7300b78 CW |
1255 | if (!for_part) { |
1256 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1257 | if (ret != 0) { | |
1258 | bdput(bdev); | |
1259 | return ret; | |
1260 | } | |
82666020 | 1261 | } |
7db9cfd3 | 1262 | |
d3374825 | 1263 | restart: |
0762b8bd | 1264 | |
89f97496 | 1265 | ret = -ENXIO; |
cf771cb5 | 1266 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1267 | if (!disk) |
6e9624b8 | 1268 | goto out; |
523e1d39 | 1269 | owner = disk->fops->owner; |
1da177e4 | 1270 | |
69e02c59 | 1271 | disk_block_events(disk); |
6796bf54 | 1272 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1273 | if (!bdev->bd_openers) { |
1274 | bdev->bd_disk = disk; | |
87192a2a | 1275 | bdev->bd_queue = disk->queue; |
1da177e4 | 1276 | bdev->bd_contains = bdev; |
99a01cdf | 1277 | bdev->bd_inode->i_flags = 0; |
03cdadb0 | 1278 | |
cf771cb5 | 1279 | if (!partno) { |
89f97496 TH |
1280 | ret = -ENXIO; |
1281 | bdev->bd_part = disk_get_part(disk, partno); | |
1282 | if (!bdev->bd_part) | |
1283 | goto out_clear; | |
1284 | ||
1196f8b8 | 1285 | ret = 0; |
1da177e4 | 1286 | if (disk->fops->open) { |
572c4892 | 1287 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1288 | if (ret == -ERESTARTSYS) { |
1289 | /* Lost a race with 'disk' being | |
1290 | * deleted, try again. | |
1291 | * See md.c | |
1292 | */ | |
1293 | disk_put_part(bdev->bd_part); | |
1294 | bdev->bd_part = NULL; | |
d3374825 | 1295 | bdev->bd_disk = NULL; |
87192a2a | 1296 | bdev->bd_queue = NULL; |
d3374825 | 1297 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1298 | disk_unblock_events(disk); |
69e02c59 | 1299 | put_disk(disk); |
523e1d39 | 1300 | module_put(owner); |
d3374825 N |
1301 | goto restart; |
1302 | } | |
1da177e4 | 1303 | } |
7e69723f | 1304 | |
5a023cdb | 1305 | if (!ret) { |
7e69723f | 1306 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
a8078b1f | 1307 | if (!bdev_dax_capable(bdev)) |
5a023cdb DW |
1308 | bdev->bd_inode->i_flags &= ~S_DAX; |
1309 | } | |
7e69723f | 1310 | |
1196f8b8 TH |
1311 | /* |
1312 | * If the device is invalidated, rescan partition | |
1313 | * if open succeeded or failed with -ENOMEDIUM. | |
1314 | * The latter is necessary to prevent ghost | |
1315 | * partitions on a removed medium. | |
1316 | */ | |
fe316bf2 JN |
1317 | if (bdev->bd_invalidated) { |
1318 | if (!ret) | |
1319 | rescan_partitions(disk, bdev); | |
1320 | else if (ret == -ENOMEDIUM) | |
1321 | invalidate_partitions(disk, bdev); | |
1322 | } | |
5a023cdb | 1323 | |
1196f8b8 TH |
1324 | if (ret) |
1325 | goto out_clear; | |
1da177e4 | 1326 | } else { |
1da177e4 LT |
1327 | struct block_device *whole; |
1328 | whole = bdget_disk(disk, 0); | |
1329 | ret = -ENOMEM; | |
1330 | if (!whole) | |
0762b8bd | 1331 | goto out_clear; |
37be4124 | 1332 | BUG_ON(for_part); |
572c4892 | 1333 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1334 | if (ret) |
0762b8bd | 1335 | goto out_clear; |
1da177e4 | 1336 | bdev->bd_contains = whole; |
89f97496 | 1337 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1338 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1339 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1340 | ret = -ENXIO; |
0762b8bd | 1341 | goto out_clear; |
1da177e4 | 1342 | } |
89f97496 | 1343 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
a8078b1f | 1344 | if (!bdev_dax_capable(bdev)) |
f0b2e563 | 1345 | bdev->bd_inode->i_flags &= ~S_DAX; |
1da177e4 LT |
1346 | } |
1347 | } else { | |
1da177e4 | 1348 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1349 | ret = 0; |
1350 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1351 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1352 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1353 | if (bdev->bd_invalidated) { |
1354 | if (!ret) | |
1355 | rescan_partitions(bdev->bd_disk, bdev); | |
1356 | else if (ret == -ENOMEDIUM) | |
1357 | invalidate_partitions(bdev->bd_disk, bdev); | |
1358 | } | |
1196f8b8 TH |
1359 | if (ret) |
1360 | goto out_unlock_bdev; | |
1da177e4 | 1361 | } |
69e02c59 | 1362 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1363 | put_disk(disk); |
523e1d39 | 1364 | module_put(owner); |
1da177e4 LT |
1365 | } |
1366 | bdev->bd_openers++; | |
37be4124 N |
1367 | if (for_part) |
1368 | bdev->bd_part_count++; | |
c039e313 | 1369 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1370 | disk_unblock_events(disk); |
1da177e4 LT |
1371 | return 0; |
1372 | ||
0762b8bd | 1373 | out_clear: |
89f97496 | 1374 | disk_put_part(bdev->bd_part); |
1da177e4 | 1375 | bdev->bd_disk = NULL; |
0762b8bd | 1376 | bdev->bd_part = NULL; |
87192a2a | 1377 | bdev->bd_queue = NULL; |
1da177e4 | 1378 | if (bdev != bdev->bd_contains) |
572c4892 | 1379 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1380 | bdev->bd_contains = NULL; |
0762b8bd | 1381 | out_unlock_bdev: |
c039e313 | 1382 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1383 | disk_unblock_events(disk); |
0762b8bd | 1384 | put_disk(disk); |
523e1d39 | 1385 | module_put(owner); |
4345caba | 1386 | out: |
0762b8bd TH |
1387 | bdput(bdev); |
1388 | ||
1da177e4 LT |
1389 | return ret; |
1390 | } | |
1391 | ||
d4d77629 TH |
1392 | /** |
1393 | * blkdev_get - open a block device | |
1394 | * @bdev: block_device to open | |
1395 | * @mode: FMODE_* mask | |
1396 | * @holder: exclusive holder identifier | |
1397 | * | |
1398 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1399 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1400 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1401 | * | |
1402 | * On success, the reference count of @bdev is unchanged. On failure, | |
1403 | * @bdev is put. | |
1404 | * | |
1405 | * CONTEXT: | |
1406 | * Might sleep. | |
1407 | * | |
1408 | * RETURNS: | |
1409 | * 0 on success, -errno on failure. | |
1410 | */ | |
e525fd89 | 1411 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1412 | { |
e525fd89 TH |
1413 | struct block_device *whole = NULL; |
1414 | int res; | |
1415 | ||
1416 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1417 | ||
1418 | if ((mode & FMODE_EXCL) && holder) { | |
1419 | whole = bd_start_claiming(bdev, holder); | |
1420 | if (IS_ERR(whole)) { | |
1421 | bdput(bdev); | |
1422 | return PTR_ERR(whole); | |
1423 | } | |
1424 | } | |
1425 | ||
1426 | res = __blkdev_get(bdev, mode, 0); | |
1427 | ||
1428 | if (whole) { | |
d4dc210f TH |
1429 | struct gendisk *disk = whole->bd_disk; |
1430 | ||
6a027eff | 1431 | /* finish claiming */ |
77ea887e | 1432 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1433 | spin_lock(&bdev_lock); |
1434 | ||
77ea887e | 1435 | if (!res) { |
6a027eff TH |
1436 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1437 | /* | |
1438 | * Note that for a whole device bd_holders | |
1439 | * will be incremented twice, and bd_holder | |
1440 | * will be set to bd_may_claim before being | |
1441 | * set to holder | |
1442 | */ | |
1443 | whole->bd_holders++; | |
1444 | whole->bd_holder = bd_may_claim; | |
1445 | bdev->bd_holders++; | |
1446 | bdev->bd_holder = holder; | |
1447 | } | |
1448 | ||
1449 | /* tell others that we're done */ | |
1450 | BUG_ON(whole->bd_claiming != holder); | |
1451 | whole->bd_claiming = NULL; | |
1452 | wake_up_bit(&whole->bd_claiming, 0); | |
1453 | ||
1454 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1455 | |
1456 | /* | |
d4dc210f TH |
1457 | * Block event polling for write claims if requested. Any |
1458 | * write holder makes the write_holder state stick until | |
1459 | * all are released. This is good enough and tracking | |
1460 | * individual writeable reference is too fragile given the | |
1461 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1462 | */ |
4c49ff3f TH |
1463 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1464 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1465 | bdev->bd_write_holder = true; |
d4dc210f | 1466 | disk_block_events(disk); |
77ea887e TH |
1467 | } |
1468 | ||
1469 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1470 | bdput(whole); |
e525fd89 TH |
1471 | } |
1472 | ||
1473 | return res; | |
37be4124 | 1474 | } |
1da177e4 LT |
1475 | EXPORT_SYMBOL(blkdev_get); |
1476 | ||
d4d77629 TH |
1477 | /** |
1478 | * blkdev_get_by_path - open a block device by name | |
1479 | * @path: path to the block device to open | |
1480 | * @mode: FMODE_* mask | |
1481 | * @holder: exclusive holder identifier | |
1482 | * | |
1483 | * Open the blockdevice described by the device file at @path. @mode | |
1484 | * and @holder are identical to blkdev_get(). | |
1485 | * | |
1486 | * On success, the returned block_device has reference count of one. | |
1487 | * | |
1488 | * CONTEXT: | |
1489 | * Might sleep. | |
1490 | * | |
1491 | * RETURNS: | |
1492 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1493 | */ | |
1494 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1495 | void *holder) | |
1496 | { | |
1497 | struct block_device *bdev; | |
1498 | int err; | |
1499 | ||
1500 | bdev = lookup_bdev(path); | |
1501 | if (IS_ERR(bdev)) | |
1502 | return bdev; | |
1503 | ||
1504 | err = blkdev_get(bdev, mode, holder); | |
1505 | if (err) | |
1506 | return ERR_PTR(err); | |
1507 | ||
e51900f7 CE |
1508 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1509 | blkdev_put(bdev, mode); | |
1510 | return ERR_PTR(-EACCES); | |
1511 | } | |
1512 | ||
d4d77629 TH |
1513 | return bdev; |
1514 | } | |
1515 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1516 | ||
1517 | /** | |
1518 | * blkdev_get_by_dev - open a block device by device number | |
1519 | * @dev: device number of block device to open | |
1520 | * @mode: FMODE_* mask | |
1521 | * @holder: exclusive holder identifier | |
1522 | * | |
1523 | * Open the blockdevice described by device number @dev. @mode and | |
1524 | * @holder are identical to blkdev_get(). | |
1525 | * | |
1526 | * Use it ONLY if you really do not have anything better - i.e. when | |
1527 | * you are behind a truly sucky interface and all you are given is a | |
1528 | * device number. _Never_ to be used for internal purposes. If you | |
1529 | * ever need it - reconsider your API. | |
1530 | * | |
1531 | * On success, the returned block_device has reference count of one. | |
1532 | * | |
1533 | * CONTEXT: | |
1534 | * Might sleep. | |
1535 | * | |
1536 | * RETURNS: | |
1537 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1538 | */ | |
1539 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1540 | { | |
1541 | struct block_device *bdev; | |
1542 | int err; | |
1543 | ||
1544 | bdev = bdget(dev); | |
1545 | if (!bdev) | |
1546 | return ERR_PTR(-ENOMEM); | |
1547 | ||
1548 | err = blkdev_get(bdev, mode, holder); | |
1549 | if (err) | |
1550 | return ERR_PTR(err); | |
1551 | ||
1552 | return bdev; | |
1553 | } | |
1554 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1555 | ||
1da177e4 LT |
1556 | static int blkdev_open(struct inode * inode, struct file * filp) |
1557 | { | |
1558 | struct block_device *bdev; | |
1da177e4 LT |
1559 | |
1560 | /* | |
1561 | * Preserve backwards compatibility and allow large file access | |
1562 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1563 | * binary needs it. We might want to drop this workaround | |
1564 | * during an unstable branch. | |
1565 | */ | |
1566 | filp->f_flags |= O_LARGEFILE; | |
1567 | ||
572c4892 AV |
1568 | if (filp->f_flags & O_NDELAY) |
1569 | filp->f_mode |= FMODE_NDELAY; | |
1570 | if (filp->f_flags & O_EXCL) | |
1571 | filp->f_mode |= FMODE_EXCL; | |
1572 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1573 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1574 | ||
1da177e4 | 1575 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1576 | if (bdev == NULL) |
1577 | return -ENOMEM; | |
1da177e4 | 1578 | |
572c4892 AV |
1579 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1580 | ||
e525fd89 | 1581 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1582 | } |
1583 | ||
4385bab1 | 1584 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1585 | { |
2e7b651d | 1586 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1587 | struct block_device *victim = NULL; |
2e7b651d | 1588 | |
6796bf54 | 1589 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1590 | if (for_part) |
1591 | bdev->bd_part_count--; | |
1592 | ||
2e7b651d | 1593 | if (!--bdev->bd_openers) { |
6a027eff | 1594 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1595 | sync_blockdev(bdev); |
1596 | kill_bdev(bdev); | |
43d1c0eb ID |
1597 | |
1598 | bdev_write_inode(bdev); | |
564f00f6 | 1599 | /* |
43d1c0eb ID |
1600 | * Detaching bdev inode from its wb in __destroy_inode() |
1601 | * is too late: the queue which embeds its bdi (along with | |
1602 | * root wb) can be gone as soon as we put_disk() below. | |
94007751 | 1603 | */ |
43d1c0eb | 1604 | inode_detach_wb(bdev->bd_inode); |
2e7b651d PZ |
1605 | } |
1606 | if (bdev->bd_contains == bdev) { | |
1607 | if (disk->fops->release) | |
db2a144b | 1608 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1609 | } |
1610 | if (!bdev->bd_openers) { | |
1611 | struct module *owner = disk->fops->owner; | |
1612 | ||
0762b8bd TH |
1613 | disk_put_part(bdev->bd_part); |
1614 | bdev->bd_part = NULL; | |
2e7b651d | 1615 | bdev->bd_disk = NULL; |
37be4124 N |
1616 | if (bdev != bdev->bd_contains) |
1617 | victim = bdev->bd_contains; | |
2e7b651d | 1618 | bdev->bd_contains = NULL; |
523e1d39 TH |
1619 | |
1620 | put_disk(disk); | |
1621 | module_put(owner); | |
2e7b651d | 1622 | } |
2e7b651d PZ |
1623 | mutex_unlock(&bdev->bd_mutex); |
1624 | bdput(bdev); | |
37be4124 | 1625 | if (victim) |
9a1c3542 | 1626 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1627 | } |
1628 | ||
4385bab1 | 1629 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1630 | { |
85ef06d1 TH |
1631 | mutex_lock(&bdev->bd_mutex); |
1632 | ||
e525fd89 | 1633 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1634 | bool bdev_free; |
1635 | ||
1636 | /* | |
1637 | * Release a claim on the device. The holder fields | |
1638 | * are protected with bdev_lock. bd_mutex is to | |
1639 | * synchronize disk_holder unlinking. | |
1640 | */ | |
6a027eff TH |
1641 | spin_lock(&bdev_lock); |
1642 | ||
1643 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1644 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1645 | ||
1646 | /* bd_contains might point to self, check in a separate step */ | |
1647 | if ((bdev_free = !bdev->bd_holders)) | |
1648 | bdev->bd_holder = NULL; | |
1649 | if (!bdev->bd_contains->bd_holders) | |
1650 | bdev->bd_contains->bd_holder = NULL; | |
1651 | ||
1652 | spin_unlock(&bdev_lock); | |
1653 | ||
77ea887e TH |
1654 | /* |
1655 | * If this was the last claim, remove holder link and | |
1656 | * unblock evpoll if it was a write holder. | |
1657 | */ | |
85ef06d1 TH |
1658 | if (bdev_free && bdev->bd_write_holder) { |
1659 | disk_unblock_events(bdev->bd_disk); | |
1660 | bdev->bd_write_holder = false; | |
77ea887e | 1661 | } |
6936217c | 1662 | } |
77ea887e | 1663 | |
85ef06d1 TH |
1664 | /* |
1665 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1666 | * event. This is to ensure detection of media removal commanded | |
1667 | * from userland - e.g. eject(1). | |
1668 | */ | |
1669 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1670 | ||
1671 | mutex_unlock(&bdev->bd_mutex); | |
1672 | ||
4385bab1 | 1673 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1674 | } |
2e7b651d PZ |
1675 | EXPORT_SYMBOL(blkdev_put); |
1676 | ||
1da177e4 LT |
1677 | static int blkdev_close(struct inode * inode, struct file * filp) |
1678 | { | |
4ebb16ca | 1679 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1680 | blkdev_put(bdev, filp->f_mode); |
1681 | return 0; | |
1da177e4 LT |
1682 | } |
1683 | ||
bb93e3a5 | 1684 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1685 | { |
4ebb16ca | 1686 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1687 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1688 | |
1689 | /* | |
1690 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1691 | * to updated it before every ioctl. | |
1692 | */ | |
56b26add | 1693 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1694 | mode |= FMODE_NDELAY; |
1695 | else | |
1696 | mode &= ~FMODE_NDELAY; | |
1697 | ||
56b26add | 1698 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1699 | } |
1700 | ||
eef99380 CH |
1701 | /* |
1702 | * Write data to the block device. Only intended for the block device itself | |
1703 | * and the raw driver which basically is a fake block device. | |
1704 | * | |
1705 | * Does not take i_mutex for the write and thus is not for general purpose | |
1706 | * use. | |
1707 | */ | |
1456c0a8 | 1708 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1709 | { |
1710 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1711 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1712 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1713 | struct blk_plug plug; |
eef99380 | 1714 | ssize_t ret; |
5f380c7f | 1715 | |
7ec7b94a AV |
1716 | if (bdev_read_only(I_BDEV(bd_inode))) |
1717 | return -EPERM; | |
5f380c7f | 1718 | |
7ec7b94a | 1719 | if (!iov_iter_count(from)) |
5f380c7f AV |
1720 | return 0; |
1721 | ||
7ec7b94a AV |
1722 | if (iocb->ki_pos >= size) |
1723 | return -ENOSPC; | |
1724 | ||
1725 | iov_iter_truncate(from, size - iocb->ki_pos); | |
eef99380 | 1726 | |
53362a05 | 1727 | blk_start_plug(&plug); |
1456c0a8 | 1728 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
1729 | if (ret > 0) |
1730 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 1731 | blk_finish_plug(&plug); |
eef99380 CH |
1732 | return ret; |
1733 | } | |
1456c0a8 | 1734 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1735 | |
b2de525f | 1736 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1737 | { |
1738 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1739 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 1740 | loff_t size = i_size_read(bd_inode); |
a886038b | 1741 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1742 | |
1743 | if (pos >= size) | |
1744 | return 0; | |
1745 | ||
1746 | size -= pos; | |
a886038b AV |
1747 | iov_iter_truncate(to, size); |
1748 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1749 | } |
b2de525f | 1750 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1751 | |
87d8fe1e TT |
1752 | /* |
1753 | * Try to release a page associated with block device when the system | |
1754 | * is under memory pressure. | |
1755 | */ | |
1756 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1757 | { | |
1758 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1759 | ||
1760 | if (super && super->s_op->bdev_try_to_free_page) | |
1761 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1762 | ||
1763 | return try_to_free_buffers(page); | |
1764 | } | |
1765 | ||
7f6d5b52 RZ |
1766 | static int blkdev_writepages(struct address_space *mapping, |
1767 | struct writeback_control *wbc) | |
1768 | { | |
1769 | if (dax_mapping(mapping)) { | |
1770 | struct block_device *bdev = I_BDEV(mapping->host); | |
1771 | ||
1772 | return dax_writeback_mapping_range(mapping, bdev, wbc); | |
1773 | } | |
1774 | return generic_writepages(mapping, wbc); | |
1775 | } | |
1776 | ||
4c54ac62 | 1777 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1778 | .readpage = blkdev_readpage, |
447f05bb | 1779 | .readpages = blkdev_readpages, |
1da177e4 | 1780 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1781 | .write_begin = blkdev_write_begin, |
1782 | .write_end = blkdev_write_end, | |
7f6d5b52 | 1783 | .writepages = blkdev_writepages, |
87d8fe1e | 1784 | .releasepage = blkdev_releasepage, |
1da177e4 | 1785 | .direct_IO = blkdev_direct_IO, |
b4597226 | 1786 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1787 | }; |
1788 | ||
4b6f5d20 | 1789 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1790 | .open = blkdev_open, |
1791 | .release = blkdev_close, | |
1792 | .llseek = block_llseek, | |
a886038b | 1793 | .read_iter = blkdev_read_iter, |
1456c0a8 | 1794 | .write_iter = blkdev_write_iter, |
acc93d30 | 1795 | .mmap = generic_file_mmap, |
b1dd3b28 | 1796 | .fsync = blkdev_fsync, |
bb93e3a5 | 1797 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1798 | #ifdef CONFIG_COMPAT |
1799 | .compat_ioctl = compat_blkdev_ioctl, | |
1800 | #endif | |
1e8b3332 | 1801 | .splice_read = generic_file_splice_read, |
8d020765 | 1802 | .splice_write = iter_file_splice_write, |
1da177e4 LT |
1803 | }; |
1804 | ||
1805 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1806 | { | |
1807 | int res; | |
1808 | mm_segment_t old_fs = get_fs(); | |
1809 | set_fs(KERNEL_DS); | |
56b26add | 1810 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1811 | set_fs(old_fs); |
1812 | return res; | |
1813 | } | |
1814 | ||
1815 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1816 | ||
1817 | /** | |
1818 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1819 | * @pathname: special file representing the block device |
1da177e4 | 1820 | * |
57d1b536 | 1821 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1822 | * namespace if possible and return it. Return ERR_PTR(error) |
1823 | * otherwise. | |
1824 | */ | |
421748ec | 1825 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1826 | { |
1827 | struct block_device *bdev; | |
1828 | struct inode *inode; | |
421748ec | 1829 | struct path path; |
1da177e4 LT |
1830 | int error; |
1831 | ||
421748ec | 1832 | if (!pathname || !*pathname) |
1da177e4 LT |
1833 | return ERR_PTR(-EINVAL); |
1834 | ||
421748ec | 1835 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1836 | if (error) |
1837 | return ERR_PTR(error); | |
1838 | ||
bb668734 | 1839 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
1840 | error = -ENOTBLK; |
1841 | if (!S_ISBLK(inode->i_mode)) | |
1842 | goto fail; | |
1843 | error = -EACCES; | |
a2982cc9 | 1844 | if (!may_open_dev(&path)) |
1da177e4 LT |
1845 | goto fail; |
1846 | error = -ENOMEM; | |
1847 | bdev = bd_acquire(inode); | |
1848 | if (!bdev) | |
1849 | goto fail; | |
1850 | out: | |
421748ec | 1851 | path_put(&path); |
1da177e4 LT |
1852 | return bdev; |
1853 | fail: | |
1854 | bdev = ERR_PTR(error); | |
1855 | goto out; | |
1856 | } | |
d5686b44 | 1857 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 1858 | |
93b270f7 | 1859 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
1860 | { |
1861 | struct super_block *sb = get_super(bdev); | |
1862 | int res = 0; | |
1863 | ||
1864 | if (sb) { | |
1865 | /* | |
1866 | * no need to lock the super, get_super holds the | |
1867 | * read mutex so the filesystem cannot go away | |
1868 | * under us (->put_super runs with the write lock | |
1869 | * hold). | |
1870 | */ | |
1871 | shrink_dcache_sb(sb); | |
93b270f7 | 1872 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
1873 | drop_super(sb); |
1874 | } | |
f98393a6 | 1875 | invalidate_bdev(bdev); |
b71e8a4c DH |
1876 | return res; |
1877 | } | |
1878 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
1879 | |
1880 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
1881 | { | |
1882 | struct inode *inode, *old_inode = NULL; | |
1883 | ||
74278da9 | 1884 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1885 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
1886 | struct address_space *mapping = inode->i_mapping; | |
1887 | ||
1888 | spin_lock(&inode->i_lock); | |
1889 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
1890 | mapping->nrpages == 0) { | |
1891 | spin_unlock(&inode->i_lock); | |
1892 | continue; | |
1893 | } | |
1894 | __iget(inode); | |
1895 | spin_unlock(&inode->i_lock); | |
74278da9 | 1896 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1897 | /* |
1898 | * We hold a reference to 'inode' so it couldn't have been | |
1899 | * removed from s_inodes list while we dropped the | |
74278da9 | 1900 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 1901 | * be holding the last reference and we cannot iput it under |
74278da9 | 1902 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
1903 | * later. |
1904 | */ | |
1905 | iput(old_inode); | |
1906 | old_inode = inode; | |
1907 | ||
1908 | func(I_BDEV(inode), arg); | |
1909 | ||
74278da9 | 1910 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 1911 | } |
74278da9 | 1912 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1913 | iput(old_inode); |
1914 | } |