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; | |
abf54548 MC |
419 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
420 | REQ_OP_READ); | |
2e6edc95 DW |
421 | blk_queue_exit(bdev->bd_queue); |
422 | return result; | |
47a191fd MW |
423 | } |
424 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
425 | ||
426 | /** | |
427 | * bdev_write_page() - Start writing a page to a block device | |
428 | * @bdev: The device to write the page to | |
429 | * @sector: The offset on the device to write the page to (need not be aligned) | |
430 | * @page: The page to write | |
431 | * @wbc: The writeback_control for the write | |
432 | * | |
433 | * On entry, the page should be locked and not currently under writeback. | |
434 | * On exit, if the write started successfully, the page will be unlocked and | |
435 | * under writeback. If the write failed already (eg the driver failed to | |
436 | * queue the page to the device), the page will still be locked. If the | |
437 | * caller is a ->writepage implementation, it will need to unlock the page. | |
438 | * | |
439 | * Errors returned by this function are usually "soft", eg out of memory, or | |
440 | * queue full; callers should try a different route to write this page rather | |
441 | * than propagate an error back up the stack. | |
442 | * | |
443 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
444 | */ | |
445 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
446 | struct page *page, struct writeback_control *wbc) | |
447 | { | |
448 | int result; | |
47a191fd | 449 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
2e6edc95 | 450 | |
f68eb1e7 | 451 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 452 | return -EOPNOTSUPP; |
6f3b0e8b | 453 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
454 | if (result) |
455 | return result; | |
456 | ||
47a191fd | 457 | set_page_writeback(page); |
abf54548 MC |
458 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
459 | REQ_OP_WRITE); | |
47a191fd MW |
460 | if (result) |
461 | end_page_writeback(page); | |
462 | else | |
463 | unlock_page(page); | |
2e6edc95 | 464 | blk_queue_exit(bdev->bd_queue); |
47a191fd MW |
465 | return result; |
466 | } | |
467 | EXPORT_SYMBOL_GPL(bdev_write_page); | |
468 | ||
dd22f551 MW |
469 | /** |
470 | * bdev_direct_access() - Get the address for directly-accessibly memory | |
471 | * @bdev: The device containing the memory | |
b2e0d162 | 472 | * @dax: control and output parameters for ->direct_access |
dd22f551 MW |
473 | * |
474 | * If a block device is made up of directly addressable memory, this function | |
475 | * will tell the caller the PFN and the address of the memory. The address | |
476 | * may be directly dereferenced within the kernel without the need to call | |
477 | * ioremap(), kmap() or similar. The PFN is suitable for inserting into | |
478 | * page tables. | |
479 | * | |
480 | * Return: negative errno if an error occurs, otherwise the number of bytes | |
481 | * accessible at this address. | |
482 | */ | |
b2e0d162 | 483 | long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax) |
dd22f551 | 484 | { |
b2e0d162 DW |
485 | sector_t sector = dax->sector; |
486 | long avail, size = dax->size; | |
dd22f551 MW |
487 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
488 | ||
43c3dd08 MW |
489 | /* |
490 | * The device driver is allowed to sleep, in order to make the | |
491 | * memory directly accessible. | |
492 | */ | |
493 | might_sleep(); | |
494 | ||
dd22f551 MW |
495 | if (size < 0) |
496 | return size; | |
163d4baa | 497 | if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access) |
dd22f551 MW |
498 | return -EOPNOTSUPP; |
499 | if ((sector + DIV_ROUND_UP(size, 512)) > | |
500 | part_nr_sects_read(bdev->bd_part)) | |
501 | return -ERANGE; | |
502 | sector += get_start_sect(bdev); | |
503 | if (sector % (PAGE_SIZE / 512)) | |
504 | return -EINVAL; | |
0a70bd43 | 505 | avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size); |
dd22f551 MW |
506 | if (!avail) |
507 | return -ERANGE; | |
fe683ada DW |
508 | if (avail > 0 && avail & ~PAGE_MASK) |
509 | return -ENXIO; | |
dd22f551 MW |
510 | return min(avail, size); |
511 | } | |
512 | EXPORT_SYMBOL_GPL(bdev_direct_access); | |
513 | ||
2d96afc8 TK |
514 | /** |
515 | * bdev_dax_supported() - Check if the device supports dax for filesystem | |
516 | * @sb: The superblock of the device | |
517 | * @blocksize: The block size of the device | |
518 | * | |
519 | * This is a library function for filesystems to check if the block device | |
520 | * can be mounted with dax option. | |
521 | * | |
522 | * Return: negative errno if unsupported, 0 if supported. | |
523 | */ | |
524 | int bdev_dax_supported(struct super_block *sb, int blocksize) | |
525 | { | |
526 | struct blk_dax_ctl dax = { | |
527 | .sector = 0, | |
528 | .size = PAGE_SIZE, | |
529 | }; | |
530 | int err; | |
531 | ||
532 | if (blocksize != PAGE_SIZE) { | |
533 | vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax"); | |
534 | return -EINVAL; | |
535 | } | |
536 | ||
537 | err = bdev_direct_access(sb->s_bdev, &dax); | |
538 | if (err < 0) { | |
539 | switch (err) { | |
540 | case -EOPNOTSUPP: | |
541 | vfs_msg(sb, KERN_ERR, | |
542 | "error: device does not support dax"); | |
543 | break; | |
544 | case -EINVAL: | |
545 | vfs_msg(sb, KERN_ERR, | |
546 | "error: unaligned partition for dax"); | |
547 | break; | |
548 | default: | |
549 | vfs_msg(sb, KERN_ERR, | |
550 | "error: dax access failed (%d)", err); | |
551 | } | |
552 | return err; | |
553 | } | |
554 | ||
555 | return 0; | |
556 | } | |
557 | EXPORT_SYMBOL_GPL(bdev_dax_supported); | |
558 | ||
a8078b1f TK |
559 | /** |
560 | * bdev_dax_capable() - Return if the raw device is capable for dax | |
561 | * @bdev: The device for raw block device access | |
562 | */ | |
563 | bool bdev_dax_capable(struct block_device *bdev) | |
564 | { | |
a8078b1f TK |
565 | struct blk_dax_ctl dax = { |
566 | .size = PAGE_SIZE, | |
567 | }; | |
568 | ||
569 | if (!IS_ENABLED(CONFIG_FS_DAX)) | |
570 | return false; | |
571 | ||
572 | dax.sector = 0; | |
573 | if (bdev_direct_access(bdev, &dax) < 0) | |
574 | return false; | |
575 | ||
576 | dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512); | |
577 | if (bdev_direct_access(bdev, &dax) < 0) | |
578 | return false; | |
579 | ||
a8078b1f TK |
580 | return true; |
581 | } | |
582 | ||
1da177e4 LT |
583 | /* |
584 | * pseudo-fs | |
585 | */ | |
586 | ||
587 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 588 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
589 | |
590 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
591 | { | |
e94b1766 | 592 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
593 | if (!ei) |
594 | return NULL; | |
595 | return &ei->vfs_inode; | |
596 | } | |
597 | ||
fa0d7e3d | 598 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 599 | { |
fa0d7e3d | 600 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
601 | struct bdev_inode *bdi = BDEV_I(inode); |
602 | ||
1da177e4 LT |
603 | kmem_cache_free(bdev_cachep, bdi); |
604 | } | |
605 | ||
fa0d7e3d NP |
606 | static void bdev_destroy_inode(struct inode *inode) |
607 | { | |
608 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
609 | } | |
610 | ||
51cc5068 | 611 | static void init_once(void *foo) |
1da177e4 LT |
612 | { |
613 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
614 | struct block_device *bdev = &ei->bdev; | |
615 | ||
a35afb83 CL |
616 | memset(bdev, 0, sizeof(*bdev)); |
617 | mutex_init(&bdev->bd_mutex); | |
a35afb83 | 618 | INIT_LIST_HEAD(&bdev->bd_list); |
49731baa TH |
619 | #ifdef CONFIG_SYSFS |
620 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
621 | #endif | |
a35afb83 | 622 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
623 | /* Initialize mutex for freeze. */ |
624 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
625 | } |
626 | ||
b57922d9 | 627 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
628 | { |
629 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
91b0abe3 | 630 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 631 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 632 | clear_inode(inode); |
1da177e4 | 633 | spin_lock(&bdev_lock); |
1da177e4 LT |
634 | list_del_init(&bdev->bd_list); |
635 | spin_unlock(&bdev_lock); | |
636 | } | |
637 | ||
ee9b6d61 | 638 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
639 | .statfs = simple_statfs, |
640 | .alloc_inode = bdev_alloc_inode, | |
641 | .destroy_inode = bdev_destroy_inode, | |
642 | .drop_inode = generic_delete_inode, | |
b57922d9 | 643 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
644 | }; |
645 | ||
51139ada AV |
646 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
647 | int flags, const char *dev_name, void *data) | |
1da177e4 | 648 | { |
3684aa70 SL |
649 | struct dentry *dent; |
650 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); | |
651 | if (dent) | |
652 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; | |
653 | return dent; | |
1da177e4 LT |
654 | } |
655 | ||
656 | static struct file_system_type bd_type = { | |
657 | .name = "bdev", | |
51139ada | 658 | .mount = bd_mount, |
1da177e4 LT |
659 | .kill_sb = kill_anon_super, |
660 | }; | |
661 | ||
a212b105 TH |
662 | struct super_block *blockdev_superblock __read_mostly; |
663 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
664 | |
665 | void __init bdev_cache_init(void) | |
666 | { | |
667 | int err; | |
ace8577a | 668 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 669 | |
1da177e4 | 670 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 671 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 672 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 673 | init_once); |
1da177e4 LT |
674 | err = register_filesystem(&bd_type); |
675 | if (err) | |
676 | panic("Cannot register bdev pseudo-fs"); | |
677 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
678 | if (IS_ERR(bd_mnt)) |
679 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 680 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
681 | } |
682 | ||
683 | /* | |
684 | * Most likely _very_ bad one - but then it's hardly critical for small | |
685 | * /dev and can be fixed when somebody will need really large one. | |
686 | * Keep in mind that it will be fed through icache hash function too. | |
687 | */ | |
688 | static inline unsigned long hash(dev_t dev) | |
689 | { | |
690 | return MAJOR(dev)+MINOR(dev); | |
691 | } | |
692 | ||
693 | static int bdev_test(struct inode *inode, void *data) | |
694 | { | |
695 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
696 | } | |
697 | ||
698 | static int bdev_set(struct inode *inode, void *data) | |
699 | { | |
700 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
701 | return 0; | |
702 | } | |
703 | ||
704 | static LIST_HEAD(all_bdevs); | |
705 | ||
706 | struct block_device *bdget(dev_t dev) | |
707 | { | |
708 | struct block_device *bdev; | |
709 | struct inode *inode; | |
710 | ||
c2acf7b9 | 711 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
712 | bdev_test, bdev_set, &dev); |
713 | ||
714 | if (!inode) | |
715 | return NULL; | |
716 | ||
717 | bdev = &BDEV_I(inode)->bdev; | |
718 | ||
719 | if (inode->i_state & I_NEW) { | |
720 | bdev->bd_contains = NULL; | |
782b94cd | 721 | bdev->bd_super = NULL; |
1da177e4 LT |
722 | bdev->bd_inode = inode; |
723 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
724 | bdev->bd_part_count = 0; | |
725 | bdev->bd_invalidated = 0; | |
726 | inode->i_mode = S_IFBLK; | |
727 | inode->i_rdev = dev; | |
728 | inode->i_bdev = bdev; | |
729 | inode->i_data.a_ops = &def_blk_aops; | |
730 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
731 | spin_lock(&bdev_lock); |
732 | list_add(&bdev->bd_list, &all_bdevs); | |
733 | spin_unlock(&bdev_lock); | |
734 | unlock_new_inode(inode); | |
735 | } | |
736 | return bdev; | |
737 | } | |
738 | ||
739 | EXPORT_SYMBOL(bdget); | |
740 | ||
dddac6a7 AJ |
741 | /** |
742 | * bdgrab -- Grab a reference to an already referenced block device | |
743 | * @bdev: Block device to grab a reference to. | |
744 | */ | |
745 | struct block_device *bdgrab(struct block_device *bdev) | |
746 | { | |
7de9c6ee | 747 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
748 | return bdev; |
749 | } | |
c1681bf8 | 750 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 751 | |
1da177e4 LT |
752 | long nr_blockdev_pages(void) |
753 | { | |
203a2935 | 754 | struct block_device *bdev; |
1da177e4 LT |
755 | long ret = 0; |
756 | spin_lock(&bdev_lock); | |
203a2935 | 757 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
758 | ret += bdev->bd_inode->i_mapping->nrpages; |
759 | } | |
760 | spin_unlock(&bdev_lock); | |
761 | return ret; | |
762 | } | |
763 | ||
764 | void bdput(struct block_device *bdev) | |
765 | { | |
766 | iput(bdev->bd_inode); | |
767 | } | |
768 | ||
769 | EXPORT_SYMBOL(bdput); | |
770 | ||
771 | static struct block_device *bd_acquire(struct inode *inode) | |
772 | { | |
773 | struct block_device *bdev; | |
09d967c6 | 774 | |
1da177e4 LT |
775 | spin_lock(&bdev_lock); |
776 | bdev = inode->i_bdev; | |
09d967c6 | 777 | if (bdev) { |
ed8a9d2c | 778 | bdgrab(bdev); |
1da177e4 LT |
779 | spin_unlock(&bdev_lock); |
780 | return bdev; | |
781 | } | |
782 | spin_unlock(&bdev_lock); | |
09d967c6 | 783 | |
1da177e4 LT |
784 | bdev = bdget(inode->i_rdev); |
785 | if (bdev) { | |
786 | spin_lock(&bdev_lock); | |
09d967c6 OH |
787 | if (!inode->i_bdev) { |
788 | /* | |
7de9c6ee | 789 | * We take an additional reference to bd_inode, |
09d967c6 OH |
790 | * and it's released in clear_inode() of inode. |
791 | * So, we can access it via ->i_mapping always | |
792 | * without igrab(). | |
793 | */ | |
ed8a9d2c | 794 | bdgrab(bdev); |
09d967c6 OH |
795 | inode->i_bdev = bdev; |
796 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
09d967c6 | 797 | } |
1da177e4 LT |
798 | spin_unlock(&bdev_lock); |
799 | } | |
800 | return bdev; | |
801 | } | |
802 | ||
803 | /* Call when you free inode */ | |
804 | ||
805 | void bd_forget(struct inode *inode) | |
806 | { | |
09d967c6 OH |
807 | struct block_device *bdev = NULL; |
808 | ||
1da177e4 | 809 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
810 | if (!sb_is_blkdev_sb(inode->i_sb)) |
811 | bdev = inode->i_bdev; | |
a4a4f943 AV |
812 | inode->i_bdev = NULL; |
813 | inode->i_mapping = &inode->i_data; | |
1da177e4 | 814 | spin_unlock(&bdev_lock); |
09d967c6 OH |
815 | |
816 | if (bdev) | |
ed8a9d2c | 817 | bdput(bdev); |
1da177e4 LT |
818 | } |
819 | ||
1a3cbbc5 TH |
820 | /** |
821 | * bd_may_claim - test whether a block device can be claimed | |
822 | * @bdev: block device of interest | |
823 | * @whole: whole block device containing @bdev, may equal @bdev | |
824 | * @holder: holder trying to claim @bdev | |
825 | * | |
25985edc | 826 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
827 | * |
828 | * CONTEXT: | |
829 | * spin_lock(&bdev_lock). | |
830 | * | |
831 | * RETURNS: | |
832 | * %true if @bdev can be claimed, %false otherwise. | |
833 | */ | |
834 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
835 | void *holder) | |
1da177e4 | 836 | { |
1da177e4 | 837 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 838 | return true; /* already a holder */ |
1da177e4 | 839 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 840 | return false; /* held by someone else */ |
1da177e4 | 841 | else if (bdev->bd_contains == bdev) |
1a3cbbc5 | 842 | return true; /* is a whole device which isn't held */ |
1da177e4 | 843 | |
e525fd89 | 844 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
845 | return true; /* is a partition of a device that is being partitioned */ |
846 | else if (whole->bd_holder != NULL) | |
847 | return false; /* is a partition of a held device */ | |
1da177e4 | 848 | else |
1a3cbbc5 TH |
849 | return true; /* is a partition of an un-held device */ |
850 | } | |
851 | ||
6b4517a7 TH |
852 | /** |
853 | * bd_prepare_to_claim - prepare to claim a block device | |
854 | * @bdev: block device of interest | |
855 | * @whole: the whole device containing @bdev, may equal @bdev | |
856 | * @holder: holder trying to claim @bdev | |
857 | * | |
858 | * Prepare to claim @bdev. This function fails if @bdev is already | |
859 | * claimed by another holder and waits if another claiming is in | |
860 | * progress. This function doesn't actually claim. On successful | |
861 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
862 | * | |
863 | * CONTEXT: | |
864 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
865 | * it multiple times. | |
866 | * | |
867 | * RETURNS: | |
868 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
869 | */ | |
870 | static int bd_prepare_to_claim(struct block_device *bdev, | |
871 | struct block_device *whole, void *holder) | |
872 | { | |
873 | retry: | |
874 | /* if someone else claimed, fail */ | |
875 | if (!bd_may_claim(bdev, whole, holder)) | |
876 | return -EBUSY; | |
877 | ||
e75aa858 TH |
878 | /* if claiming is already in progress, wait for it to finish */ |
879 | if (whole->bd_claiming) { | |
6b4517a7 TH |
880 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
881 | DEFINE_WAIT(wait); | |
882 | ||
883 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
884 | spin_unlock(&bdev_lock); | |
885 | schedule(); | |
886 | finish_wait(wq, &wait); | |
887 | spin_lock(&bdev_lock); | |
888 | goto retry; | |
889 | } | |
890 | ||
891 | /* yay, all mine */ | |
892 | return 0; | |
893 | } | |
894 | ||
895 | /** | |
896 | * bd_start_claiming - start claiming a block device | |
897 | * @bdev: block device of interest | |
898 | * @holder: holder trying to claim @bdev | |
899 | * | |
900 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
901 | * exclusively and mark that an exclusive open is in progress. Each | |
902 | * successful call to this function must be matched with a call to | |
b0018361 NP |
903 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
904 | * fail). | |
905 | * | |
906 | * This function is used to gain exclusive access to the block device | |
907 | * without actually causing other exclusive open attempts to fail. It | |
908 | * should be used when the open sequence itself requires exclusive | |
909 | * access but may subsequently fail. | |
6b4517a7 TH |
910 | * |
911 | * CONTEXT: | |
912 | * Might sleep. | |
913 | * | |
914 | * RETURNS: | |
915 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
916 | * value on failure. | |
917 | */ | |
918 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
919 | void *holder) | |
920 | { | |
921 | struct gendisk *disk; | |
922 | struct block_device *whole; | |
923 | int partno, err; | |
924 | ||
925 | might_sleep(); | |
926 | ||
927 | /* | |
928 | * @bdev might not have been initialized properly yet, look up | |
929 | * and grab the outer block device the hard way. | |
930 | */ | |
931 | disk = get_gendisk(bdev->bd_dev, &partno); | |
932 | if (!disk) | |
933 | return ERR_PTR(-ENXIO); | |
934 | ||
d4c208b8 TH |
935 | /* |
936 | * Normally, @bdev should equal what's returned from bdget_disk() | |
937 | * if partno is 0; however, some drivers (floppy) use multiple | |
938 | * bdev's for the same physical device and @bdev may be one of the | |
939 | * aliases. Keep @bdev if partno is 0. This means claimer | |
940 | * tracking is broken for those devices but it has always been that | |
941 | * way. | |
942 | */ | |
943 | if (partno) | |
944 | whole = bdget_disk(disk, 0); | |
945 | else | |
946 | whole = bdgrab(bdev); | |
947 | ||
cf342570 | 948 | module_put(disk->fops->owner); |
6b4517a7 TH |
949 | put_disk(disk); |
950 | if (!whole) | |
951 | return ERR_PTR(-ENOMEM); | |
952 | ||
953 | /* prepare to claim, if successful, mark claiming in progress */ | |
954 | spin_lock(&bdev_lock); | |
955 | ||
956 | err = bd_prepare_to_claim(bdev, whole, holder); | |
957 | if (err == 0) { | |
958 | whole->bd_claiming = holder; | |
959 | spin_unlock(&bdev_lock); | |
960 | return whole; | |
961 | } else { | |
962 | spin_unlock(&bdev_lock); | |
963 | bdput(whole); | |
964 | return ERR_PTR(err); | |
965 | } | |
966 | } | |
967 | ||
641dc636 | 968 | #ifdef CONFIG_SYSFS |
49731baa TH |
969 | struct bd_holder_disk { |
970 | struct list_head list; | |
971 | struct gendisk *disk; | |
972 | int refcnt; | |
973 | }; | |
974 | ||
975 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
976 | struct gendisk *disk) | |
977 | { | |
978 | struct bd_holder_disk *holder; | |
979 | ||
980 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
981 | if (holder->disk == disk) | |
982 | return holder; | |
983 | return NULL; | |
984 | } | |
985 | ||
4d7dd8fd | 986 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 987 | { |
4d7dd8fd | 988 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
989 | } |
990 | ||
991 | static void del_symlink(struct kobject *from, struct kobject *to) | |
992 | { | |
641dc636 JN |
993 | sysfs_remove_link(from, kobject_name(to)); |
994 | } | |
995 | ||
df6c0cd9 | 996 | /** |
e09b457b TH |
997 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
998 | * @bdev: the claimed slave bdev | |
999 | * @disk: the holding disk | |
df6c0cd9 | 1000 | * |
49731baa TH |
1001 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1002 | * | |
e09b457b | 1003 | * This functions creates the following sysfs symlinks. |
641dc636 | 1004 | * |
e09b457b TH |
1005 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1006 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 1007 | * |
e09b457b TH |
1008 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1009 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 1010 | * |
e09b457b TH |
1011 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1012 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 1013 | * |
e09b457b TH |
1014 | * The caller must have claimed @bdev before calling this function and |
1015 | * ensure that both @bdev and @disk are valid during the creation and | |
1016 | * lifetime of these symlinks. | |
641dc636 | 1017 | * |
e09b457b TH |
1018 | * CONTEXT: |
1019 | * Might sleep. | |
641dc636 | 1020 | * |
e09b457b TH |
1021 | * RETURNS: |
1022 | * 0 on success, -errno on failure. | |
641dc636 | 1023 | */ |
e09b457b | 1024 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 1025 | { |
49731baa | 1026 | struct bd_holder_disk *holder; |
e09b457b | 1027 | int ret = 0; |
641dc636 | 1028 | |
2e7b651d | 1029 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 1030 | |
49731baa | 1031 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 1032 | |
e09b457b TH |
1033 | /* FIXME: remove the following once add_disk() handles errors */ |
1034 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
1035 | goto out_unlock; | |
4e91672c | 1036 | |
49731baa TH |
1037 | holder = bd_find_holder_disk(bdev, disk); |
1038 | if (holder) { | |
1039 | holder->refcnt++; | |
e09b457b | 1040 | goto out_unlock; |
49731baa | 1041 | } |
641dc636 | 1042 | |
49731baa TH |
1043 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1044 | if (!holder) { | |
1045 | ret = -ENOMEM; | |
e09b457b TH |
1046 | goto out_unlock; |
1047 | } | |
641dc636 | 1048 | |
49731baa TH |
1049 | INIT_LIST_HEAD(&holder->list); |
1050 | holder->disk = disk; | |
1051 | holder->refcnt = 1; | |
1052 | ||
1053 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1054 | if (ret) | |
1055 | goto out_free; | |
1056 | ||
1057 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
1058 | if (ret) | |
1059 | goto out_del; | |
e7407d16 TH |
1060 | /* |
1061 | * bdev could be deleted beneath us which would implicitly destroy | |
1062 | * the holder directory. Hold on to it. | |
1063 | */ | |
1064 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1065 | |
1066 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1067 | goto out_unlock; | |
1068 | ||
1069 | out_del: | |
1070 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1071 | out_free: | |
1072 | kfree(holder); | |
e09b457b | 1073 | out_unlock: |
b4cf1b72 | 1074 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1075 | return ret; |
641dc636 | 1076 | } |
e09b457b | 1077 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1078 | |
49731baa TH |
1079 | /** |
1080 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1081 | * @bdev: the calimed slave bdev | |
1082 | * @disk: the holding disk | |
1083 | * | |
1084 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1085 | * | |
1086 | * CONTEXT: | |
1087 | * Might sleep. | |
1088 | */ | |
1089 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1090 | { |
49731baa | 1091 | struct bd_holder_disk *holder; |
641dc636 | 1092 | |
49731baa | 1093 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1094 | |
49731baa TH |
1095 | holder = bd_find_holder_disk(bdev, disk); |
1096 | ||
1097 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1098 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1099 | del_symlink(bdev->bd_part->holder_dir, | |
1100 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1101 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1102 | list_del_init(&holder->list); |
1103 | kfree(holder); | |
1104 | } | |
1105 | ||
1106 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1107 | } |
49731baa | 1108 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1109 | #endif |
1da177e4 | 1110 | |
56ade44b AP |
1111 | /** |
1112 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1113 | * | |
1114 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1115 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1116 | * |
1117 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1118 | * when a disk has been changed -- either by a media change or online | |
1119 | * resize. | |
1120 | */ | |
93b270f7 | 1121 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1122 | { |
93b270f7 | 1123 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1124 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1125 | "resized disk %s\n", |
1126 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b AP |
1127 | } |
1128 | ||
1129 | if (!bdev->bd_disk) | |
1130 | return; | |
d27769ec | 1131 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1132 | bdev->bd_invalidated = 1; |
1133 | } | |
1134 | ||
c3279d14 | 1135 | /** |
57d1b536 | 1136 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1137 | * @disk: struct gendisk to check |
1138 | * @bdev: struct bdev to adjust. | |
1139 | * | |
1140 | * This routine checks to see if the bdev size does not match the disk size | |
1141 | * and adjusts it if it differs. | |
1142 | */ | |
1143 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1144 | { | |
1145 | loff_t disk_size, bdev_size; | |
1146 | ||
1147 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1148 | bdev_size = i_size_read(bdev->bd_inode); | |
1149 | if (disk_size != bdev_size) { | |
c3279d14 AP |
1150 | printk(KERN_INFO |
1151 | "%s: detected capacity change from %lld to %lld\n", | |
424081f3 | 1152 | disk->disk_name, bdev_size, disk_size); |
c3279d14 | 1153 | i_size_write(bdev->bd_inode, disk_size); |
93b270f7 | 1154 | flush_disk(bdev, false); |
c3279d14 AP |
1155 | } |
1156 | } | |
1157 | EXPORT_SYMBOL(check_disk_size_change); | |
1158 | ||
0c002c2f | 1159 | /** |
57d1b536 | 1160 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1161 | * @disk: struct gendisk to be revalidated |
1162 | * | |
1163 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1164 | * call-backs. It is used to do common pre and post operations needed | |
1165 | * for all revalidate_disk operations. | |
1166 | */ | |
1167 | int revalidate_disk(struct gendisk *disk) | |
1168 | { | |
c3279d14 | 1169 | struct block_device *bdev; |
0c002c2f AP |
1170 | int ret = 0; |
1171 | ||
1172 | if (disk->fops->revalidate_disk) | |
1173 | ret = disk->fops->revalidate_disk(disk); | |
25520d55 | 1174 | blk_integrity_revalidate(disk); |
c3279d14 AP |
1175 | bdev = bdget_disk(disk, 0); |
1176 | if (!bdev) | |
1177 | return ret; | |
1178 | ||
1179 | mutex_lock(&bdev->bd_mutex); | |
1180 | check_disk_size_change(disk, bdev); | |
7630b661 | 1181 | bdev->bd_invalidated = 0; |
c3279d14 AP |
1182 | mutex_unlock(&bdev->bd_mutex); |
1183 | bdput(bdev); | |
0c002c2f AP |
1184 | return ret; |
1185 | } | |
1186 | EXPORT_SYMBOL(revalidate_disk); | |
1187 | ||
1da177e4 LT |
1188 | /* |
1189 | * This routine checks whether a removable media has been changed, | |
1190 | * and invalidates all buffer-cache-entries in that case. This | |
1191 | * is a relatively slow routine, so we have to try to minimize using | |
1192 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1193 | * is the best way of combining speed and utility, I think. | |
1194 | * People changing diskettes in the middle of an operation deserve | |
1195 | * to lose :-) | |
1196 | */ | |
1197 | int check_disk_change(struct block_device *bdev) | |
1198 | { | |
1199 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1200 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1201 | unsigned int events; |
1da177e4 | 1202 | |
77ea887e TH |
1203 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1204 | DISK_EVENT_EJECT_REQUEST); | |
1205 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1206 | return 0; |
1207 | ||
93b270f7 | 1208 | flush_disk(bdev, true); |
1da177e4 LT |
1209 | if (bdops->revalidate_disk) |
1210 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1211 | return 1; |
1212 | } | |
1213 | ||
1214 | EXPORT_SYMBOL(check_disk_change); | |
1215 | ||
1216 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1217 | { | |
e1defc4f | 1218 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1219 | |
5955102c | 1220 | inode_lock(bdev->bd_inode); |
d646a02a | 1221 | i_size_write(bdev->bd_inode, size); |
5955102c | 1222 | inode_unlock(bdev->bd_inode); |
09cbfeaf | 1223 | while (bsize < PAGE_SIZE) { |
1da177e4 LT |
1224 | if (size & bsize) |
1225 | break; | |
1226 | bsize <<= 1; | |
1227 | } | |
1228 | bdev->bd_block_size = bsize; | |
1229 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1230 | } | |
1231 | EXPORT_SYMBOL(bd_set_size); | |
1232 | ||
4385bab1 | 1233 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1234 | |
6d740cd5 PZ |
1235 | /* |
1236 | * bd_mutex locking: | |
1237 | * | |
1238 | * mutex_lock(part->bd_mutex) | |
1239 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1240 | */ | |
1241 | ||
572c4892 | 1242 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1243 | { |
1da177e4 | 1244 | struct gendisk *disk; |
523e1d39 | 1245 | struct module *owner; |
7db9cfd3 | 1246 | int ret; |
cf771cb5 | 1247 | int partno; |
fe6e9c1f AV |
1248 | int perm = 0; |
1249 | ||
572c4892 | 1250 | if (mode & FMODE_READ) |
fe6e9c1f | 1251 | perm |= MAY_READ; |
572c4892 | 1252 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1253 | perm |= MAY_WRITE; |
1254 | /* | |
1255 | * hooks: /n/, see "layering violations". | |
1256 | */ | |
b7300b78 CW |
1257 | if (!for_part) { |
1258 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1259 | if (ret != 0) { | |
1260 | bdput(bdev); | |
1261 | return ret; | |
1262 | } | |
82666020 | 1263 | } |
7db9cfd3 | 1264 | |
d3374825 | 1265 | restart: |
0762b8bd | 1266 | |
89f97496 | 1267 | ret = -ENXIO; |
cf771cb5 | 1268 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1269 | if (!disk) |
6e9624b8 | 1270 | goto out; |
523e1d39 | 1271 | owner = disk->fops->owner; |
1da177e4 | 1272 | |
69e02c59 | 1273 | disk_block_events(disk); |
6796bf54 | 1274 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1275 | if (!bdev->bd_openers) { |
1276 | bdev->bd_disk = disk; | |
87192a2a | 1277 | bdev->bd_queue = disk->queue; |
1da177e4 | 1278 | bdev->bd_contains = bdev; |
99a01cdf | 1279 | bdev->bd_inode->i_flags = 0; |
03cdadb0 | 1280 | |
cf771cb5 | 1281 | if (!partno) { |
89f97496 TH |
1282 | ret = -ENXIO; |
1283 | bdev->bd_part = disk_get_part(disk, partno); | |
1284 | if (!bdev->bd_part) | |
1285 | goto out_clear; | |
1286 | ||
1196f8b8 | 1287 | ret = 0; |
1da177e4 | 1288 | if (disk->fops->open) { |
572c4892 | 1289 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1290 | if (ret == -ERESTARTSYS) { |
1291 | /* Lost a race with 'disk' being | |
1292 | * deleted, try again. | |
1293 | * See md.c | |
1294 | */ | |
1295 | disk_put_part(bdev->bd_part); | |
1296 | bdev->bd_part = NULL; | |
d3374825 | 1297 | bdev->bd_disk = NULL; |
87192a2a | 1298 | bdev->bd_queue = NULL; |
d3374825 | 1299 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1300 | disk_unblock_events(disk); |
69e02c59 | 1301 | put_disk(disk); |
523e1d39 | 1302 | module_put(owner); |
d3374825 N |
1303 | goto restart; |
1304 | } | |
1da177e4 | 1305 | } |
7e69723f | 1306 | |
5a023cdb | 1307 | if (!ret) { |
7e69723f | 1308 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
a8078b1f | 1309 | if (!bdev_dax_capable(bdev)) |
5a023cdb DW |
1310 | bdev->bd_inode->i_flags &= ~S_DAX; |
1311 | } | |
7e69723f | 1312 | |
1196f8b8 TH |
1313 | /* |
1314 | * If the device is invalidated, rescan partition | |
1315 | * if open succeeded or failed with -ENOMEDIUM. | |
1316 | * The latter is necessary to prevent ghost | |
1317 | * partitions on a removed medium. | |
1318 | */ | |
fe316bf2 JN |
1319 | if (bdev->bd_invalidated) { |
1320 | if (!ret) | |
1321 | rescan_partitions(disk, bdev); | |
1322 | else if (ret == -ENOMEDIUM) | |
1323 | invalidate_partitions(disk, bdev); | |
1324 | } | |
5a023cdb | 1325 | |
1196f8b8 TH |
1326 | if (ret) |
1327 | goto out_clear; | |
1da177e4 | 1328 | } else { |
1da177e4 LT |
1329 | struct block_device *whole; |
1330 | whole = bdget_disk(disk, 0); | |
1331 | ret = -ENOMEM; | |
1332 | if (!whole) | |
0762b8bd | 1333 | goto out_clear; |
37be4124 | 1334 | BUG_ON(for_part); |
572c4892 | 1335 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1336 | if (ret) |
0762b8bd | 1337 | goto out_clear; |
1da177e4 | 1338 | bdev->bd_contains = whole; |
89f97496 | 1339 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1340 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1341 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1342 | ret = -ENXIO; |
0762b8bd | 1343 | goto out_clear; |
1da177e4 | 1344 | } |
89f97496 | 1345 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
a8078b1f | 1346 | if (!bdev_dax_capable(bdev)) |
f0b2e563 | 1347 | bdev->bd_inode->i_flags &= ~S_DAX; |
1da177e4 LT |
1348 | } |
1349 | } else { | |
1da177e4 | 1350 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1351 | ret = 0; |
1352 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1353 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1354 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1355 | if (bdev->bd_invalidated) { |
1356 | if (!ret) | |
1357 | rescan_partitions(bdev->bd_disk, bdev); | |
1358 | else if (ret == -ENOMEDIUM) | |
1359 | invalidate_partitions(bdev->bd_disk, bdev); | |
1360 | } | |
1196f8b8 TH |
1361 | if (ret) |
1362 | goto out_unlock_bdev; | |
1da177e4 | 1363 | } |
69e02c59 | 1364 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1365 | put_disk(disk); |
523e1d39 | 1366 | module_put(owner); |
1da177e4 LT |
1367 | } |
1368 | bdev->bd_openers++; | |
37be4124 N |
1369 | if (for_part) |
1370 | bdev->bd_part_count++; | |
c039e313 | 1371 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1372 | disk_unblock_events(disk); |
1da177e4 LT |
1373 | return 0; |
1374 | ||
0762b8bd | 1375 | out_clear: |
89f97496 | 1376 | disk_put_part(bdev->bd_part); |
1da177e4 | 1377 | bdev->bd_disk = NULL; |
0762b8bd | 1378 | bdev->bd_part = NULL; |
87192a2a | 1379 | bdev->bd_queue = NULL; |
1da177e4 | 1380 | if (bdev != bdev->bd_contains) |
572c4892 | 1381 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1382 | bdev->bd_contains = NULL; |
0762b8bd | 1383 | out_unlock_bdev: |
c039e313 | 1384 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1385 | disk_unblock_events(disk); |
0762b8bd | 1386 | put_disk(disk); |
523e1d39 | 1387 | module_put(owner); |
4345caba | 1388 | out: |
0762b8bd TH |
1389 | bdput(bdev); |
1390 | ||
1da177e4 LT |
1391 | return ret; |
1392 | } | |
1393 | ||
d4d77629 TH |
1394 | /** |
1395 | * blkdev_get - open a block device | |
1396 | * @bdev: block_device to open | |
1397 | * @mode: FMODE_* mask | |
1398 | * @holder: exclusive holder identifier | |
1399 | * | |
1400 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1401 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1402 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1403 | * | |
1404 | * On success, the reference count of @bdev is unchanged. On failure, | |
1405 | * @bdev is put. | |
1406 | * | |
1407 | * CONTEXT: | |
1408 | * Might sleep. | |
1409 | * | |
1410 | * RETURNS: | |
1411 | * 0 on success, -errno on failure. | |
1412 | */ | |
e525fd89 | 1413 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1414 | { |
e525fd89 TH |
1415 | struct block_device *whole = NULL; |
1416 | int res; | |
1417 | ||
1418 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1419 | ||
1420 | if ((mode & FMODE_EXCL) && holder) { | |
1421 | whole = bd_start_claiming(bdev, holder); | |
1422 | if (IS_ERR(whole)) { | |
1423 | bdput(bdev); | |
1424 | return PTR_ERR(whole); | |
1425 | } | |
1426 | } | |
1427 | ||
1428 | res = __blkdev_get(bdev, mode, 0); | |
1429 | ||
1430 | if (whole) { | |
d4dc210f TH |
1431 | struct gendisk *disk = whole->bd_disk; |
1432 | ||
6a027eff | 1433 | /* finish claiming */ |
77ea887e | 1434 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1435 | spin_lock(&bdev_lock); |
1436 | ||
77ea887e | 1437 | if (!res) { |
6a027eff TH |
1438 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1439 | /* | |
1440 | * Note that for a whole device bd_holders | |
1441 | * will be incremented twice, and bd_holder | |
1442 | * will be set to bd_may_claim before being | |
1443 | * set to holder | |
1444 | */ | |
1445 | whole->bd_holders++; | |
1446 | whole->bd_holder = bd_may_claim; | |
1447 | bdev->bd_holders++; | |
1448 | bdev->bd_holder = holder; | |
1449 | } | |
1450 | ||
1451 | /* tell others that we're done */ | |
1452 | BUG_ON(whole->bd_claiming != holder); | |
1453 | whole->bd_claiming = NULL; | |
1454 | wake_up_bit(&whole->bd_claiming, 0); | |
1455 | ||
1456 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1457 | |
1458 | /* | |
d4dc210f TH |
1459 | * Block event polling for write claims if requested. Any |
1460 | * write holder makes the write_holder state stick until | |
1461 | * all are released. This is good enough and tracking | |
1462 | * individual writeable reference is too fragile given the | |
1463 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1464 | */ |
4c49ff3f TH |
1465 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1466 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1467 | bdev->bd_write_holder = true; |
d4dc210f | 1468 | disk_block_events(disk); |
77ea887e TH |
1469 | } |
1470 | ||
1471 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1472 | bdput(whole); |
e525fd89 TH |
1473 | } |
1474 | ||
1475 | return res; | |
37be4124 | 1476 | } |
1da177e4 LT |
1477 | EXPORT_SYMBOL(blkdev_get); |
1478 | ||
d4d77629 TH |
1479 | /** |
1480 | * blkdev_get_by_path - open a block device by name | |
1481 | * @path: path to the block device to open | |
1482 | * @mode: FMODE_* mask | |
1483 | * @holder: exclusive holder identifier | |
1484 | * | |
1485 | * Open the blockdevice described by the device file at @path. @mode | |
1486 | * and @holder are identical to blkdev_get(). | |
1487 | * | |
1488 | * On success, the returned block_device has reference count of one. | |
1489 | * | |
1490 | * CONTEXT: | |
1491 | * Might sleep. | |
1492 | * | |
1493 | * RETURNS: | |
1494 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1495 | */ | |
1496 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1497 | void *holder) | |
1498 | { | |
1499 | struct block_device *bdev; | |
1500 | int err; | |
1501 | ||
1502 | bdev = lookup_bdev(path); | |
1503 | if (IS_ERR(bdev)) | |
1504 | return bdev; | |
1505 | ||
1506 | err = blkdev_get(bdev, mode, holder); | |
1507 | if (err) | |
1508 | return ERR_PTR(err); | |
1509 | ||
e51900f7 CE |
1510 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1511 | blkdev_put(bdev, mode); | |
1512 | return ERR_PTR(-EACCES); | |
1513 | } | |
1514 | ||
d4d77629 TH |
1515 | return bdev; |
1516 | } | |
1517 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1518 | ||
1519 | /** | |
1520 | * blkdev_get_by_dev - open a block device by device number | |
1521 | * @dev: device number of block device to open | |
1522 | * @mode: FMODE_* mask | |
1523 | * @holder: exclusive holder identifier | |
1524 | * | |
1525 | * Open the blockdevice described by device number @dev. @mode and | |
1526 | * @holder are identical to blkdev_get(). | |
1527 | * | |
1528 | * Use it ONLY if you really do not have anything better - i.e. when | |
1529 | * you are behind a truly sucky interface and all you are given is a | |
1530 | * device number. _Never_ to be used for internal purposes. If you | |
1531 | * ever need it - reconsider your API. | |
1532 | * | |
1533 | * On success, the returned block_device has reference count of one. | |
1534 | * | |
1535 | * CONTEXT: | |
1536 | * Might sleep. | |
1537 | * | |
1538 | * RETURNS: | |
1539 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1540 | */ | |
1541 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1542 | { | |
1543 | struct block_device *bdev; | |
1544 | int err; | |
1545 | ||
1546 | bdev = bdget(dev); | |
1547 | if (!bdev) | |
1548 | return ERR_PTR(-ENOMEM); | |
1549 | ||
1550 | err = blkdev_get(bdev, mode, holder); | |
1551 | if (err) | |
1552 | return ERR_PTR(err); | |
1553 | ||
1554 | return bdev; | |
1555 | } | |
1556 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1557 | ||
1da177e4 LT |
1558 | static int blkdev_open(struct inode * inode, struct file * filp) |
1559 | { | |
1560 | struct block_device *bdev; | |
1da177e4 LT |
1561 | |
1562 | /* | |
1563 | * Preserve backwards compatibility and allow large file access | |
1564 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1565 | * binary needs it. We might want to drop this workaround | |
1566 | * during an unstable branch. | |
1567 | */ | |
1568 | filp->f_flags |= O_LARGEFILE; | |
1569 | ||
572c4892 AV |
1570 | if (filp->f_flags & O_NDELAY) |
1571 | filp->f_mode |= FMODE_NDELAY; | |
1572 | if (filp->f_flags & O_EXCL) | |
1573 | filp->f_mode |= FMODE_EXCL; | |
1574 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1575 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1576 | ||
1da177e4 | 1577 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1578 | if (bdev == NULL) |
1579 | return -ENOMEM; | |
1da177e4 | 1580 | |
572c4892 AV |
1581 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1582 | ||
e525fd89 | 1583 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1584 | } |
1585 | ||
4385bab1 | 1586 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1587 | { |
2e7b651d | 1588 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1589 | struct block_device *victim = NULL; |
2e7b651d | 1590 | |
6796bf54 | 1591 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1592 | if (for_part) |
1593 | bdev->bd_part_count--; | |
1594 | ||
2e7b651d | 1595 | if (!--bdev->bd_openers) { |
6a027eff | 1596 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1597 | sync_blockdev(bdev); |
1598 | kill_bdev(bdev); | |
43d1c0eb ID |
1599 | |
1600 | bdev_write_inode(bdev); | |
564f00f6 | 1601 | /* |
43d1c0eb ID |
1602 | * Detaching bdev inode from its wb in __destroy_inode() |
1603 | * is too late: the queue which embeds its bdi (along with | |
1604 | * root wb) can be gone as soon as we put_disk() below. | |
94007751 | 1605 | */ |
43d1c0eb | 1606 | inode_detach_wb(bdev->bd_inode); |
2e7b651d PZ |
1607 | } |
1608 | if (bdev->bd_contains == bdev) { | |
1609 | if (disk->fops->release) | |
db2a144b | 1610 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1611 | } |
1612 | if (!bdev->bd_openers) { | |
1613 | struct module *owner = disk->fops->owner; | |
1614 | ||
0762b8bd TH |
1615 | disk_put_part(bdev->bd_part); |
1616 | bdev->bd_part = NULL; | |
2e7b651d | 1617 | bdev->bd_disk = NULL; |
37be4124 N |
1618 | if (bdev != bdev->bd_contains) |
1619 | victim = bdev->bd_contains; | |
2e7b651d | 1620 | bdev->bd_contains = NULL; |
523e1d39 TH |
1621 | |
1622 | put_disk(disk); | |
1623 | module_put(owner); | |
2e7b651d | 1624 | } |
2e7b651d PZ |
1625 | mutex_unlock(&bdev->bd_mutex); |
1626 | bdput(bdev); | |
37be4124 | 1627 | if (victim) |
9a1c3542 | 1628 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1629 | } |
1630 | ||
4385bab1 | 1631 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1632 | { |
85ef06d1 TH |
1633 | mutex_lock(&bdev->bd_mutex); |
1634 | ||
e525fd89 | 1635 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1636 | bool bdev_free; |
1637 | ||
1638 | /* | |
1639 | * Release a claim on the device. The holder fields | |
1640 | * are protected with bdev_lock. bd_mutex is to | |
1641 | * synchronize disk_holder unlinking. | |
1642 | */ | |
6a027eff TH |
1643 | spin_lock(&bdev_lock); |
1644 | ||
1645 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1646 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1647 | ||
1648 | /* bd_contains might point to self, check in a separate step */ | |
1649 | if ((bdev_free = !bdev->bd_holders)) | |
1650 | bdev->bd_holder = NULL; | |
1651 | if (!bdev->bd_contains->bd_holders) | |
1652 | bdev->bd_contains->bd_holder = NULL; | |
1653 | ||
1654 | spin_unlock(&bdev_lock); | |
1655 | ||
77ea887e TH |
1656 | /* |
1657 | * If this was the last claim, remove holder link and | |
1658 | * unblock evpoll if it was a write holder. | |
1659 | */ | |
85ef06d1 TH |
1660 | if (bdev_free && bdev->bd_write_holder) { |
1661 | disk_unblock_events(bdev->bd_disk); | |
1662 | bdev->bd_write_holder = false; | |
77ea887e | 1663 | } |
6936217c | 1664 | } |
77ea887e | 1665 | |
85ef06d1 TH |
1666 | /* |
1667 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1668 | * event. This is to ensure detection of media removal commanded | |
1669 | * from userland - e.g. eject(1). | |
1670 | */ | |
1671 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1672 | ||
1673 | mutex_unlock(&bdev->bd_mutex); | |
1674 | ||
4385bab1 | 1675 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1676 | } |
2e7b651d PZ |
1677 | EXPORT_SYMBOL(blkdev_put); |
1678 | ||
1da177e4 LT |
1679 | static int blkdev_close(struct inode * inode, struct file * filp) |
1680 | { | |
4ebb16ca | 1681 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1682 | blkdev_put(bdev, filp->f_mode); |
1683 | return 0; | |
1da177e4 LT |
1684 | } |
1685 | ||
bb93e3a5 | 1686 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1687 | { |
4ebb16ca | 1688 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1689 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1690 | |
1691 | /* | |
1692 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1693 | * to updated it before every ioctl. | |
1694 | */ | |
56b26add | 1695 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1696 | mode |= FMODE_NDELAY; |
1697 | else | |
1698 | mode &= ~FMODE_NDELAY; | |
1699 | ||
56b26add | 1700 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1701 | } |
1702 | ||
eef99380 CH |
1703 | /* |
1704 | * Write data to the block device. Only intended for the block device itself | |
1705 | * and the raw driver which basically is a fake block device. | |
1706 | * | |
1707 | * Does not take i_mutex for the write and thus is not for general purpose | |
1708 | * use. | |
1709 | */ | |
1456c0a8 | 1710 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1711 | { |
1712 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1713 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1714 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1715 | struct blk_plug plug; |
eef99380 | 1716 | ssize_t ret; |
5f380c7f | 1717 | |
7ec7b94a AV |
1718 | if (bdev_read_only(I_BDEV(bd_inode))) |
1719 | return -EPERM; | |
5f380c7f | 1720 | |
7ec7b94a | 1721 | if (!iov_iter_count(from)) |
5f380c7f AV |
1722 | return 0; |
1723 | ||
7ec7b94a AV |
1724 | if (iocb->ki_pos >= size) |
1725 | return -ENOSPC; | |
1726 | ||
1727 | iov_iter_truncate(from, size - iocb->ki_pos); | |
eef99380 | 1728 | |
53362a05 | 1729 | blk_start_plug(&plug); |
1456c0a8 | 1730 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
1731 | if (ret > 0) |
1732 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 1733 | blk_finish_plug(&plug); |
eef99380 CH |
1734 | return ret; |
1735 | } | |
1456c0a8 | 1736 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1737 | |
b2de525f | 1738 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1739 | { |
1740 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1741 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 1742 | loff_t size = i_size_read(bd_inode); |
a886038b | 1743 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1744 | |
1745 | if (pos >= size) | |
1746 | return 0; | |
1747 | ||
1748 | size -= pos; | |
a886038b AV |
1749 | iov_iter_truncate(to, size); |
1750 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1751 | } |
b2de525f | 1752 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1753 | |
87d8fe1e TT |
1754 | /* |
1755 | * Try to release a page associated with block device when the system | |
1756 | * is under memory pressure. | |
1757 | */ | |
1758 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1759 | { | |
1760 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1761 | ||
1762 | if (super && super->s_op->bdev_try_to_free_page) | |
1763 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1764 | ||
1765 | return try_to_free_buffers(page); | |
1766 | } | |
1767 | ||
7f6d5b52 RZ |
1768 | static int blkdev_writepages(struct address_space *mapping, |
1769 | struct writeback_control *wbc) | |
1770 | { | |
1771 | if (dax_mapping(mapping)) { | |
1772 | struct block_device *bdev = I_BDEV(mapping->host); | |
1773 | ||
1774 | return dax_writeback_mapping_range(mapping, bdev, wbc); | |
1775 | } | |
1776 | return generic_writepages(mapping, wbc); | |
1777 | } | |
1778 | ||
4c54ac62 | 1779 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1780 | .readpage = blkdev_readpage, |
447f05bb | 1781 | .readpages = blkdev_readpages, |
1da177e4 | 1782 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1783 | .write_begin = blkdev_write_begin, |
1784 | .write_end = blkdev_write_end, | |
7f6d5b52 | 1785 | .writepages = blkdev_writepages, |
87d8fe1e | 1786 | .releasepage = blkdev_releasepage, |
1da177e4 | 1787 | .direct_IO = blkdev_direct_IO, |
b4597226 | 1788 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1789 | }; |
1790 | ||
4b6f5d20 | 1791 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1792 | .open = blkdev_open, |
1793 | .release = blkdev_close, | |
1794 | .llseek = block_llseek, | |
a886038b | 1795 | .read_iter = blkdev_read_iter, |
1456c0a8 | 1796 | .write_iter = blkdev_write_iter, |
acc93d30 | 1797 | .mmap = generic_file_mmap, |
b1dd3b28 | 1798 | .fsync = blkdev_fsync, |
bb93e3a5 | 1799 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1800 | #ifdef CONFIG_COMPAT |
1801 | .compat_ioctl = compat_blkdev_ioctl, | |
1802 | #endif | |
1e8b3332 | 1803 | .splice_read = generic_file_splice_read, |
8d020765 | 1804 | .splice_write = iter_file_splice_write, |
1da177e4 LT |
1805 | }; |
1806 | ||
1807 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1808 | { | |
1809 | int res; | |
1810 | mm_segment_t old_fs = get_fs(); | |
1811 | set_fs(KERNEL_DS); | |
56b26add | 1812 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1813 | set_fs(old_fs); |
1814 | return res; | |
1815 | } | |
1816 | ||
1817 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1818 | ||
1819 | /** | |
1820 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1821 | * @pathname: special file representing the block device |
1da177e4 | 1822 | * |
57d1b536 | 1823 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1824 | * namespace if possible and return it. Return ERR_PTR(error) |
1825 | * otherwise. | |
1826 | */ | |
421748ec | 1827 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1828 | { |
1829 | struct block_device *bdev; | |
1830 | struct inode *inode; | |
421748ec | 1831 | struct path path; |
1da177e4 LT |
1832 | int error; |
1833 | ||
421748ec | 1834 | if (!pathname || !*pathname) |
1da177e4 LT |
1835 | return ERR_PTR(-EINVAL); |
1836 | ||
421748ec | 1837 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1838 | if (error) |
1839 | return ERR_PTR(error); | |
1840 | ||
bb668734 | 1841 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
1842 | error = -ENOTBLK; |
1843 | if (!S_ISBLK(inode->i_mode)) | |
1844 | goto fail; | |
1845 | error = -EACCES; | |
a2982cc9 | 1846 | if (!may_open_dev(&path)) |
1da177e4 LT |
1847 | goto fail; |
1848 | error = -ENOMEM; | |
1849 | bdev = bd_acquire(inode); | |
1850 | if (!bdev) | |
1851 | goto fail; | |
1852 | out: | |
421748ec | 1853 | path_put(&path); |
1da177e4 LT |
1854 | return bdev; |
1855 | fail: | |
1856 | bdev = ERR_PTR(error); | |
1857 | goto out; | |
1858 | } | |
d5686b44 | 1859 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 1860 | |
93b270f7 | 1861 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
1862 | { |
1863 | struct super_block *sb = get_super(bdev); | |
1864 | int res = 0; | |
1865 | ||
1866 | if (sb) { | |
1867 | /* | |
1868 | * no need to lock the super, get_super holds the | |
1869 | * read mutex so the filesystem cannot go away | |
1870 | * under us (->put_super runs with the write lock | |
1871 | * hold). | |
1872 | */ | |
1873 | shrink_dcache_sb(sb); | |
93b270f7 | 1874 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
1875 | drop_super(sb); |
1876 | } | |
f98393a6 | 1877 | invalidate_bdev(bdev); |
b71e8a4c DH |
1878 | return res; |
1879 | } | |
1880 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
1881 | |
1882 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
1883 | { | |
1884 | struct inode *inode, *old_inode = NULL; | |
1885 | ||
74278da9 | 1886 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1887 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
1888 | struct address_space *mapping = inode->i_mapping; | |
1889 | ||
1890 | spin_lock(&inode->i_lock); | |
1891 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
1892 | mapping->nrpages == 0) { | |
1893 | spin_unlock(&inode->i_lock); | |
1894 | continue; | |
1895 | } | |
1896 | __iget(inode); | |
1897 | spin_unlock(&inode->i_lock); | |
74278da9 | 1898 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1899 | /* |
1900 | * We hold a reference to 'inode' so it couldn't have been | |
1901 | * removed from s_inodes list while we dropped the | |
74278da9 | 1902 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 1903 | * be holding the last reference and we cannot iput it under |
74278da9 | 1904 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
1905 | * later. |
1906 | */ | |
1907 | iput(old_inode); | |
1908 | old_inode = inode; | |
1909 | ||
1910 | func(I_BDEV(inode), arg); | |
1911 | ||
74278da9 | 1912 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 1913 | } |
74278da9 | 1914 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1915 | iput(old_inode); |
1916 | } |