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 | ||
dbd3ca50 | 54 | static void bdev_write_inode(struct block_device *bdev) |
564f00f6 | 55 | { |
dbd3ca50 VG |
56 | struct inode *inode = bdev->bd_inode; |
57 | int ret; | |
58 | ||
564f00f6 CH |
59 | spin_lock(&inode->i_lock); |
60 | while (inode->i_state & I_DIRTY) { | |
61 | spin_unlock(&inode->i_lock); | |
dbd3ca50 VG |
62 | ret = write_inode_now(inode, true); |
63 | if (ret) { | |
64 | char name[BDEVNAME_SIZE]; | |
65 | pr_warn_ratelimited("VFS: Dirty inode writeback failed " | |
66 | "for block device %s (err=%d).\n", | |
67 | bdevname(bdev, name), ret); | |
68 | } | |
564f00f6 CH |
69 | spin_lock(&inode->i_lock); |
70 | } | |
71 | spin_unlock(&inode->i_lock); | |
72 | } | |
73 | ||
f9a14399 | 74 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
ff01bb48 | 75 | void kill_bdev(struct block_device *bdev) |
1da177e4 | 76 | { |
ff01bb48 AV |
77 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
78 | ||
f9fe48be | 79 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
f9a14399 | 80 | return; |
ff01bb48 | 81 | |
f9a14399 | 82 | invalidate_bh_lrus(); |
ff01bb48 | 83 | truncate_inode_pages(mapping, 0); |
1da177e4 | 84 | } |
ff01bb48 AV |
85 | EXPORT_SYMBOL(kill_bdev); |
86 | ||
87 | /* Invalidate clean unused buffers and pagecache. */ | |
88 | void invalidate_bdev(struct block_device *bdev) | |
89 | { | |
90 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
91 | ||
92 | if (mapping->nrpages == 0) | |
93 | return; | |
94 | ||
95 | invalidate_bh_lrus(); | |
96 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
97 | invalidate_mapping_pages(mapping, 0, -1); | |
98 | /* 99% of the time, we don't need to flush the cleancache on the bdev. | |
99 | * But, for the strange corners, lets be cautious | |
100 | */ | |
3167760f | 101 | cleancache_invalidate_inode(mapping); |
ff01bb48 AV |
102 | } |
103 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 LT |
104 | |
105 | int set_blocksize(struct block_device *bdev, int size) | |
106 | { | |
107 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 108 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
109 | return -EINVAL; |
110 | ||
111 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 112 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
113 | return -EINVAL; |
114 | ||
115 | /* Don't change the size if it is same as current */ | |
116 | if (bdev->bd_block_size != size) { | |
117 | sync_blockdev(bdev); | |
118 | bdev->bd_block_size = size; | |
119 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
120 | kill_bdev(bdev); | |
121 | } | |
122 | return 0; | |
123 | } | |
124 | ||
125 | EXPORT_SYMBOL(set_blocksize); | |
126 | ||
127 | int sb_set_blocksize(struct super_block *sb, int size) | |
128 | { | |
1da177e4 LT |
129 | if (set_blocksize(sb->s_bdev, size)) |
130 | return 0; | |
131 | /* If we get here, we know size is power of two | |
132 | * and it's value is between 512 and PAGE_SIZE */ | |
133 | sb->s_blocksize = size; | |
38885bd4 | 134 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
135 | return sb->s_blocksize; |
136 | } | |
137 | ||
138 | EXPORT_SYMBOL(sb_set_blocksize); | |
139 | ||
140 | int sb_min_blocksize(struct super_block *sb, int size) | |
141 | { | |
e1defc4f | 142 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
143 | if (size < minsize) |
144 | size = minsize; | |
145 | return sb_set_blocksize(sb, size); | |
146 | } | |
147 | ||
148 | EXPORT_SYMBOL(sb_min_blocksize); | |
149 | ||
150 | static int | |
151 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
152 | struct buffer_head *bh, int create) | |
153 | { | |
1da177e4 LT |
154 | bh->b_bdev = I_BDEV(inode); |
155 | bh->b_blocknr = iblock; | |
156 | set_buffer_mapped(bh); | |
157 | return 0; | |
158 | } | |
159 | ||
4ebb16ca DW |
160 | static struct inode *bdev_file_inode(struct file *file) |
161 | { | |
162 | return file->f_mapping->host; | |
163 | } | |
164 | ||
b2e895db | 165 | static ssize_t |
c8b8e32d | 166 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
b2e895db AM |
167 | { |
168 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 169 | struct inode *inode = bdev_file_inode(file); |
b2e895db | 170 | |
bbab37dd | 171 | if (IS_DAX(inode)) |
c8b8e32d | 172 | return dax_do_io(iocb, inode, iter, blkdev_get_block, |
bbab37dd | 173 | NULL, DIO_SKIP_DIO_COUNT); |
c8b8e32d | 174 | return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, |
fe0f07d0 JA |
175 | blkdev_get_block, NULL, NULL, |
176 | DIO_SKIP_DIO_COUNT); | |
b2e895db AM |
177 | } |
178 | ||
5cee5815 JK |
179 | int __sync_blockdev(struct block_device *bdev, int wait) |
180 | { | |
181 | if (!bdev) | |
182 | return 0; | |
183 | if (!wait) | |
184 | return filemap_flush(bdev->bd_inode->i_mapping); | |
185 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
186 | } | |
187 | ||
585d3bc0 NP |
188 | /* |
189 | * Write out and wait upon all the dirty data associated with a block | |
190 | * device via its mapping. Does not take the superblock lock. | |
191 | */ | |
192 | int sync_blockdev(struct block_device *bdev) | |
193 | { | |
5cee5815 | 194 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
195 | } |
196 | EXPORT_SYMBOL(sync_blockdev); | |
197 | ||
198 | /* | |
199 | * Write out and wait upon all dirty data associated with this | |
200 | * device. Filesystem data as well as the underlying block | |
201 | * device. Takes the superblock lock. | |
202 | */ | |
203 | int fsync_bdev(struct block_device *bdev) | |
204 | { | |
205 | struct super_block *sb = get_super(bdev); | |
206 | if (sb) { | |
60b0680f | 207 | int res = sync_filesystem(sb); |
585d3bc0 NP |
208 | drop_super(sb); |
209 | return res; | |
210 | } | |
211 | return sync_blockdev(bdev); | |
212 | } | |
47e4491b | 213 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
214 | |
215 | /** | |
216 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
217 | * @bdev: blockdevice to lock | |
218 | * | |
585d3bc0 NP |
219 | * If a superblock is found on this device, we take the s_umount semaphore |
220 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
221 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
222 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
223 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
224 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
225 | * actually. | |
226 | */ | |
227 | struct super_block *freeze_bdev(struct block_device *bdev) | |
228 | { | |
229 | struct super_block *sb; | |
230 | int error = 0; | |
231 | ||
232 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
233 | if (++bdev->bd_fsfreeze_count > 1) { |
234 | /* | |
235 | * We don't even need to grab a reference - the first call | |
236 | * to freeze_bdev grab an active reference and only the last | |
237 | * thaw_bdev drops it. | |
238 | */ | |
585d3bc0 | 239 | sb = get_super(bdev); |
4504230a CH |
240 | drop_super(sb); |
241 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
242 | return sb; | |
243 | } | |
244 | ||
245 | sb = get_active_super(bdev); | |
246 | if (!sb) | |
247 | goto out; | |
48b6bca6 BM |
248 | if (sb->s_op->freeze_super) |
249 | error = sb->s_op->freeze_super(sb); | |
250 | else | |
251 | error = freeze_super(sb); | |
18e9e510 JB |
252 | if (error) { |
253 | deactivate_super(sb); | |
254 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 255 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 256 | return ERR_PTR(error); |
585d3bc0 | 257 | } |
18e9e510 | 258 | deactivate_super(sb); |
4504230a | 259 | out: |
585d3bc0 NP |
260 | sync_blockdev(bdev); |
261 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 262 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
263 | } |
264 | EXPORT_SYMBOL(freeze_bdev); | |
265 | ||
266 | /** | |
267 | * thaw_bdev -- unlock filesystem | |
268 | * @bdev: blockdevice to unlock | |
269 | * @sb: associated superblock | |
270 | * | |
271 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
272 | */ | |
273 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
274 | { | |
4504230a | 275 | int error = -EINVAL; |
585d3bc0 NP |
276 | |
277 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 278 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 279 | goto out; |
4504230a CH |
280 | |
281 | error = 0; | |
282 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 283 | goto out; |
4504230a CH |
284 | |
285 | if (!sb) | |
18e9e510 | 286 | goto out; |
4504230a | 287 | |
48b6bca6 BM |
288 | if (sb->s_op->thaw_super) |
289 | error = sb->s_op->thaw_super(sb); | |
290 | else | |
291 | error = thaw_super(sb); | |
18e9e510 JB |
292 | if (error) { |
293 | bdev->bd_fsfreeze_count++; | |
294 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
295 | return error; | |
296 | } | |
297 | out: | |
585d3bc0 NP |
298 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
299 | return 0; | |
300 | } | |
301 | EXPORT_SYMBOL(thaw_bdev); | |
302 | ||
1da177e4 LT |
303 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
304 | { | |
305 | return block_write_full_page(page, blkdev_get_block, wbc); | |
306 | } | |
307 | ||
308 | static int blkdev_readpage(struct file * file, struct page * page) | |
309 | { | |
310 | return block_read_full_page(page, blkdev_get_block); | |
311 | } | |
312 | ||
447f05bb AM |
313 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
314 | struct list_head *pages, unsigned nr_pages) | |
315 | { | |
316 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
317 | } | |
318 | ||
6272b5a5 NP |
319 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
320 | loff_t pos, unsigned len, unsigned flags, | |
321 | struct page **pagep, void **fsdata) | |
1da177e4 | 322 | { |
155130a4 CH |
323 | return block_write_begin(mapping, pos, len, flags, pagep, |
324 | blkdev_get_block); | |
1da177e4 LT |
325 | } |
326 | ||
6272b5a5 NP |
327 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
328 | loff_t pos, unsigned len, unsigned copied, | |
329 | struct page *page, void *fsdata) | |
1da177e4 | 330 | { |
6272b5a5 NP |
331 | int ret; |
332 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
333 | ||
334 | unlock_page(page); | |
09cbfeaf | 335 | put_page(page); |
6272b5a5 NP |
336 | |
337 | return ret; | |
1da177e4 LT |
338 | } |
339 | ||
340 | /* | |
341 | * private llseek: | |
496ad9aa | 342 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
343 | * so we compute the size by hand (just as in block_read/write above) |
344 | */ | |
965c8e59 | 345 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 346 | { |
4ebb16ca | 347 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
348 | loff_t retval; |
349 | ||
5955102c | 350 | inode_lock(bd_inode); |
5d48f3a2 | 351 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 352 | inode_unlock(bd_inode); |
1da177e4 LT |
353 | return retval; |
354 | } | |
355 | ||
02c24a82 | 356 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 357 | { |
4ebb16ca | 358 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 359 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 360 | int error; |
da5aa861 RW |
361 | |
362 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
363 | if (error) | |
364 | return error; | |
ab0a9735 | 365 | |
b8af67e2 AB |
366 | /* |
367 | * There is no need to serialise calls to blkdev_issue_flush with | |
368 | * i_mutex and doing so causes performance issues with concurrent | |
369 | * O_SYNC writers to a block device. | |
370 | */ | |
dd3932ed | 371 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
372 | if (error == -EOPNOTSUPP) |
373 | error = 0; | |
b8af67e2 | 374 | |
ab0a9735 | 375 | return error; |
1da177e4 | 376 | } |
b1dd3b28 | 377 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 378 | |
47a191fd MW |
379 | /** |
380 | * bdev_read_page() - Start reading a page from a block device | |
381 | * @bdev: The device to read the page from | |
382 | * @sector: The offset on the device to read the page to (need not be aligned) | |
383 | * @page: The page to read | |
384 | * | |
385 | * On entry, the page should be locked. It will be unlocked when the page | |
386 | * has been read. If the block driver implements rw_page synchronously, | |
387 | * that will be true on exit from this function, but it need not be. | |
388 | * | |
389 | * Errors returned by this function are usually "soft", eg out of memory, or | |
390 | * queue full; callers should try a different route to read this page rather | |
391 | * than propagate an error back up the stack. | |
392 | * | |
393 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
394 | */ | |
395 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
396 | struct page *page) | |
397 | { | |
398 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
399 | int result = -EOPNOTSUPP; |
400 | ||
f68eb1e7 | 401 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
402 | return result; |
403 | ||
6f3b0e8b | 404 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
405 | if (result) |
406 | return result; | |
407 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ); | |
408 | blk_queue_exit(bdev->bd_queue); | |
409 | return result; | |
47a191fd MW |
410 | } |
411 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
412 | ||
413 | /** | |
414 | * bdev_write_page() - Start writing a page to a block device | |
415 | * @bdev: The device to write the page to | |
416 | * @sector: The offset on the device to write the page to (need not be aligned) | |
417 | * @page: The page to write | |
418 | * @wbc: The writeback_control for the write | |
419 | * | |
420 | * On entry, the page should be locked and not currently under writeback. | |
421 | * On exit, if the write started successfully, the page will be unlocked and | |
422 | * under writeback. If the write failed already (eg the driver failed to | |
423 | * queue the page to the device), the page will still be locked. If the | |
424 | * caller is a ->writepage implementation, it will need to unlock the page. | |
425 | * | |
426 | * Errors returned by this function are usually "soft", eg out of memory, or | |
427 | * queue full; callers should try a different route to write this page rather | |
428 | * than propagate an error back up the stack. | |
429 | * | |
430 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
431 | */ | |
432 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
433 | struct page *page, struct writeback_control *wbc) | |
434 | { | |
435 | int result; | |
436 | int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE; | |
437 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 | 438 | |
f68eb1e7 | 439 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 440 | return -EOPNOTSUPP; |
6f3b0e8b | 441 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
442 | if (result) |
443 | return result; | |
444 | ||
47a191fd MW |
445 | set_page_writeback(page); |
446 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw); | |
447 | if (result) | |
448 | end_page_writeback(page); | |
449 | else | |
450 | unlock_page(page); | |
2e6edc95 | 451 | blk_queue_exit(bdev->bd_queue); |
47a191fd MW |
452 | return result; |
453 | } | |
454 | EXPORT_SYMBOL_GPL(bdev_write_page); | |
455 | ||
dd22f551 MW |
456 | /** |
457 | * bdev_direct_access() - Get the address for directly-accessibly memory | |
458 | * @bdev: The device containing the memory | |
b2e0d162 | 459 | * @dax: control and output parameters for ->direct_access |
dd22f551 MW |
460 | * |
461 | * If a block device is made up of directly addressable memory, this function | |
462 | * will tell the caller the PFN and the address of the memory. The address | |
463 | * may be directly dereferenced within the kernel without the need to call | |
464 | * ioremap(), kmap() or similar. The PFN is suitable for inserting into | |
465 | * page tables. | |
466 | * | |
467 | * Return: negative errno if an error occurs, otherwise the number of bytes | |
468 | * accessible at this address. | |
469 | */ | |
b2e0d162 | 470 | long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax) |
dd22f551 | 471 | { |
b2e0d162 DW |
472 | sector_t sector = dax->sector; |
473 | long avail, size = dax->size; | |
dd22f551 MW |
474 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
475 | ||
43c3dd08 MW |
476 | /* |
477 | * The device driver is allowed to sleep, in order to make the | |
478 | * memory directly accessible. | |
479 | */ | |
480 | might_sleep(); | |
481 | ||
dd22f551 MW |
482 | if (size < 0) |
483 | return size; | |
484 | if (!ops->direct_access) | |
485 | return -EOPNOTSUPP; | |
486 | if ((sector + DIV_ROUND_UP(size, 512)) > | |
487 | part_nr_sects_read(bdev->bd_part)) | |
488 | return -ERANGE; | |
489 | sector += get_start_sect(bdev); | |
490 | if (sector % (PAGE_SIZE / 512)) | |
491 | return -EINVAL; | |
b2e0d162 | 492 | avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn); |
dd22f551 MW |
493 | if (!avail) |
494 | return -ERANGE; | |
fe683ada DW |
495 | if (avail > 0 && avail & ~PAGE_MASK) |
496 | return -ENXIO; | |
dd22f551 MW |
497 | return min(avail, size); |
498 | } | |
499 | EXPORT_SYMBOL_GPL(bdev_direct_access); | |
500 | ||
1da177e4 LT |
501 | /* |
502 | * pseudo-fs | |
503 | */ | |
504 | ||
505 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 506 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
507 | |
508 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
509 | { | |
e94b1766 | 510 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
511 | if (!ei) |
512 | return NULL; | |
513 | return &ei->vfs_inode; | |
514 | } | |
515 | ||
fa0d7e3d | 516 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 517 | { |
fa0d7e3d | 518 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
519 | struct bdev_inode *bdi = BDEV_I(inode); |
520 | ||
1da177e4 LT |
521 | kmem_cache_free(bdev_cachep, bdi); |
522 | } | |
523 | ||
fa0d7e3d NP |
524 | static void bdev_destroy_inode(struct inode *inode) |
525 | { | |
526 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
527 | } | |
528 | ||
51cc5068 | 529 | static void init_once(void *foo) |
1da177e4 LT |
530 | { |
531 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
532 | struct block_device *bdev = &ei->bdev; | |
533 | ||
a35afb83 CL |
534 | memset(bdev, 0, sizeof(*bdev)); |
535 | mutex_init(&bdev->bd_mutex); | |
a35afb83 CL |
536 | INIT_LIST_HEAD(&bdev->bd_inodes); |
537 | INIT_LIST_HEAD(&bdev->bd_list); | |
49731baa TH |
538 | #ifdef CONFIG_SYSFS |
539 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
540 | #endif | |
a35afb83 | 541 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
542 | /* Initialize mutex for freeze. */ |
543 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
544 | } |
545 | ||
546 | static inline void __bd_forget(struct inode *inode) | |
547 | { | |
548 | list_del_init(&inode->i_devices); | |
549 | inode->i_bdev = NULL; | |
550 | inode->i_mapping = &inode->i_data; | |
551 | } | |
552 | ||
b57922d9 | 553 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
554 | { |
555 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
556 | struct list_head *p; | |
91b0abe3 | 557 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 558 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 559 | clear_inode(inode); |
1da177e4 LT |
560 | spin_lock(&bdev_lock); |
561 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
562 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
563 | } | |
564 | list_del_init(&bdev->bd_list); | |
565 | spin_unlock(&bdev_lock); | |
566 | } | |
567 | ||
ee9b6d61 | 568 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
569 | .statfs = simple_statfs, |
570 | .alloc_inode = bdev_alloc_inode, | |
571 | .destroy_inode = bdev_destroy_inode, | |
572 | .drop_inode = generic_delete_inode, | |
b57922d9 | 573 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
574 | }; |
575 | ||
51139ada AV |
576 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
577 | int flags, const char *dev_name, void *data) | |
1da177e4 | 578 | { |
3684aa70 SL |
579 | struct dentry *dent; |
580 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); | |
581 | if (dent) | |
582 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; | |
583 | return dent; | |
1da177e4 LT |
584 | } |
585 | ||
586 | static struct file_system_type bd_type = { | |
587 | .name = "bdev", | |
51139ada | 588 | .mount = bd_mount, |
1da177e4 LT |
589 | .kill_sb = kill_anon_super, |
590 | }; | |
591 | ||
a212b105 TH |
592 | struct super_block *blockdev_superblock __read_mostly; |
593 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
594 | |
595 | void __init bdev_cache_init(void) | |
596 | { | |
597 | int err; | |
ace8577a | 598 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 599 | |
1da177e4 | 600 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 601 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 602 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 603 | init_once); |
1da177e4 LT |
604 | err = register_filesystem(&bd_type); |
605 | if (err) | |
606 | panic("Cannot register bdev pseudo-fs"); | |
607 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
608 | if (IS_ERR(bd_mnt)) |
609 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 610 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
611 | } |
612 | ||
613 | /* | |
614 | * Most likely _very_ bad one - but then it's hardly critical for small | |
615 | * /dev and can be fixed when somebody will need really large one. | |
616 | * Keep in mind that it will be fed through icache hash function too. | |
617 | */ | |
618 | static inline unsigned long hash(dev_t dev) | |
619 | { | |
620 | return MAJOR(dev)+MINOR(dev); | |
621 | } | |
622 | ||
623 | static int bdev_test(struct inode *inode, void *data) | |
624 | { | |
625 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
626 | } | |
627 | ||
628 | static int bdev_set(struct inode *inode, void *data) | |
629 | { | |
630 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
631 | return 0; | |
632 | } | |
633 | ||
634 | static LIST_HEAD(all_bdevs); | |
635 | ||
636 | struct block_device *bdget(dev_t dev) | |
637 | { | |
638 | struct block_device *bdev; | |
639 | struct inode *inode; | |
640 | ||
c2acf7b9 | 641 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
642 | bdev_test, bdev_set, &dev); |
643 | ||
644 | if (!inode) | |
645 | return NULL; | |
646 | ||
647 | bdev = &BDEV_I(inode)->bdev; | |
648 | ||
649 | if (inode->i_state & I_NEW) { | |
650 | bdev->bd_contains = NULL; | |
782b94cd | 651 | bdev->bd_super = NULL; |
1da177e4 LT |
652 | bdev->bd_inode = inode; |
653 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
654 | bdev->bd_part_count = 0; | |
655 | bdev->bd_invalidated = 0; | |
656 | inode->i_mode = S_IFBLK; | |
657 | inode->i_rdev = dev; | |
658 | inode->i_bdev = bdev; | |
659 | inode->i_data.a_ops = &def_blk_aops; | |
660 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
661 | spin_lock(&bdev_lock); |
662 | list_add(&bdev->bd_list, &all_bdevs); | |
663 | spin_unlock(&bdev_lock); | |
664 | unlock_new_inode(inode); | |
665 | } | |
666 | return bdev; | |
667 | } | |
668 | ||
669 | EXPORT_SYMBOL(bdget); | |
670 | ||
dddac6a7 AJ |
671 | /** |
672 | * bdgrab -- Grab a reference to an already referenced block device | |
673 | * @bdev: Block device to grab a reference to. | |
674 | */ | |
675 | struct block_device *bdgrab(struct block_device *bdev) | |
676 | { | |
7de9c6ee | 677 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
678 | return bdev; |
679 | } | |
c1681bf8 | 680 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 681 | |
1da177e4 LT |
682 | long nr_blockdev_pages(void) |
683 | { | |
203a2935 | 684 | struct block_device *bdev; |
1da177e4 LT |
685 | long ret = 0; |
686 | spin_lock(&bdev_lock); | |
203a2935 | 687 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
688 | ret += bdev->bd_inode->i_mapping->nrpages; |
689 | } | |
690 | spin_unlock(&bdev_lock); | |
691 | return ret; | |
692 | } | |
693 | ||
694 | void bdput(struct block_device *bdev) | |
695 | { | |
696 | iput(bdev->bd_inode); | |
697 | } | |
698 | ||
699 | EXPORT_SYMBOL(bdput); | |
700 | ||
701 | static struct block_device *bd_acquire(struct inode *inode) | |
702 | { | |
703 | struct block_device *bdev; | |
09d967c6 | 704 | |
1da177e4 LT |
705 | spin_lock(&bdev_lock); |
706 | bdev = inode->i_bdev; | |
09d967c6 | 707 | if (bdev) { |
ed8a9d2c | 708 | bdgrab(bdev); |
1da177e4 LT |
709 | spin_unlock(&bdev_lock); |
710 | return bdev; | |
711 | } | |
712 | spin_unlock(&bdev_lock); | |
09d967c6 | 713 | |
1da177e4 LT |
714 | bdev = bdget(inode->i_rdev); |
715 | if (bdev) { | |
716 | spin_lock(&bdev_lock); | |
09d967c6 OH |
717 | if (!inode->i_bdev) { |
718 | /* | |
7de9c6ee | 719 | * We take an additional reference to bd_inode, |
09d967c6 OH |
720 | * and it's released in clear_inode() of inode. |
721 | * So, we can access it via ->i_mapping always | |
722 | * without igrab(). | |
723 | */ | |
ed8a9d2c | 724 | bdgrab(bdev); |
09d967c6 OH |
725 | inode->i_bdev = bdev; |
726 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
727 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
728 | } | |
1da177e4 LT |
729 | spin_unlock(&bdev_lock); |
730 | } | |
731 | return bdev; | |
732 | } | |
733 | ||
734 | /* Call when you free inode */ | |
735 | ||
736 | void bd_forget(struct inode *inode) | |
737 | { | |
09d967c6 OH |
738 | struct block_device *bdev = NULL; |
739 | ||
1da177e4 | 740 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
741 | if (!sb_is_blkdev_sb(inode->i_sb)) |
742 | bdev = inode->i_bdev; | |
743 | __bd_forget(inode); | |
1da177e4 | 744 | spin_unlock(&bdev_lock); |
09d967c6 OH |
745 | |
746 | if (bdev) | |
ed8a9d2c | 747 | bdput(bdev); |
1da177e4 LT |
748 | } |
749 | ||
1a3cbbc5 TH |
750 | /** |
751 | * bd_may_claim - test whether a block device can be claimed | |
752 | * @bdev: block device of interest | |
753 | * @whole: whole block device containing @bdev, may equal @bdev | |
754 | * @holder: holder trying to claim @bdev | |
755 | * | |
25985edc | 756 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
757 | * |
758 | * CONTEXT: | |
759 | * spin_lock(&bdev_lock). | |
760 | * | |
761 | * RETURNS: | |
762 | * %true if @bdev can be claimed, %false otherwise. | |
763 | */ | |
764 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
765 | void *holder) | |
1da177e4 | 766 | { |
1da177e4 | 767 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 768 | return true; /* already a holder */ |
1da177e4 | 769 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 770 | return false; /* held by someone else */ |
1da177e4 | 771 | else if (bdev->bd_contains == bdev) |
1a3cbbc5 | 772 | return true; /* is a whole device which isn't held */ |
1da177e4 | 773 | |
e525fd89 | 774 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
775 | return true; /* is a partition of a device that is being partitioned */ |
776 | else if (whole->bd_holder != NULL) | |
777 | return false; /* is a partition of a held device */ | |
1da177e4 | 778 | else |
1a3cbbc5 TH |
779 | return true; /* is a partition of an un-held device */ |
780 | } | |
781 | ||
6b4517a7 TH |
782 | /** |
783 | * bd_prepare_to_claim - prepare to claim a block device | |
784 | * @bdev: block device of interest | |
785 | * @whole: the whole device containing @bdev, may equal @bdev | |
786 | * @holder: holder trying to claim @bdev | |
787 | * | |
788 | * Prepare to claim @bdev. This function fails if @bdev is already | |
789 | * claimed by another holder and waits if another claiming is in | |
790 | * progress. This function doesn't actually claim. On successful | |
791 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
792 | * | |
793 | * CONTEXT: | |
794 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
795 | * it multiple times. | |
796 | * | |
797 | * RETURNS: | |
798 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
799 | */ | |
800 | static int bd_prepare_to_claim(struct block_device *bdev, | |
801 | struct block_device *whole, void *holder) | |
802 | { | |
803 | retry: | |
804 | /* if someone else claimed, fail */ | |
805 | if (!bd_may_claim(bdev, whole, holder)) | |
806 | return -EBUSY; | |
807 | ||
e75aa858 TH |
808 | /* if claiming is already in progress, wait for it to finish */ |
809 | if (whole->bd_claiming) { | |
6b4517a7 TH |
810 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
811 | DEFINE_WAIT(wait); | |
812 | ||
813 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
814 | spin_unlock(&bdev_lock); | |
815 | schedule(); | |
816 | finish_wait(wq, &wait); | |
817 | spin_lock(&bdev_lock); | |
818 | goto retry; | |
819 | } | |
820 | ||
821 | /* yay, all mine */ | |
822 | return 0; | |
823 | } | |
824 | ||
825 | /** | |
826 | * bd_start_claiming - start claiming a block device | |
827 | * @bdev: block device of interest | |
828 | * @holder: holder trying to claim @bdev | |
829 | * | |
830 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
831 | * exclusively and mark that an exclusive open is in progress. Each | |
832 | * successful call to this function must be matched with a call to | |
b0018361 NP |
833 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
834 | * fail). | |
835 | * | |
836 | * This function is used to gain exclusive access to the block device | |
837 | * without actually causing other exclusive open attempts to fail. It | |
838 | * should be used when the open sequence itself requires exclusive | |
839 | * access but may subsequently fail. | |
6b4517a7 TH |
840 | * |
841 | * CONTEXT: | |
842 | * Might sleep. | |
843 | * | |
844 | * RETURNS: | |
845 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
846 | * value on failure. | |
847 | */ | |
848 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
849 | void *holder) | |
850 | { | |
851 | struct gendisk *disk; | |
852 | struct block_device *whole; | |
853 | int partno, err; | |
854 | ||
855 | might_sleep(); | |
856 | ||
857 | /* | |
858 | * @bdev might not have been initialized properly yet, look up | |
859 | * and grab the outer block device the hard way. | |
860 | */ | |
861 | disk = get_gendisk(bdev->bd_dev, &partno); | |
862 | if (!disk) | |
863 | return ERR_PTR(-ENXIO); | |
864 | ||
d4c208b8 TH |
865 | /* |
866 | * Normally, @bdev should equal what's returned from bdget_disk() | |
867 | * if partno is 0; however, some drivers (floppy) use multiple | |
868 | * bdev's for the same physical device and @bdev may be one of the | |
869 | * aliases. Keep @bdev if partno is 0. This means claimer | |
870 | * tracking is broken for those devices but it has always been that | |
871 | * way. | |
872 | */ | |
873 | if (partno) | |
874 | whole = bdget_disk(disk, 0); | |
875 | else | |
876 | whole = bdgrab(bdev); | |
877 | ||
cf342570 | 878 | module_put(disk->fops->owner); |
6b4517a7 TH |
879 | put_disk(disk); |
880 | if (!whole) | |
881 | return ERR_PTR(-ENOMEM); | |
882 | ||
883 | /* prepare to claim, if successful, mark claiming in progress */ | |
884 | spin_lock(&bdev_lock); | |
885 | ||
886 | err = bd_prepare_to_claim(bdev, whole, holder); | |
887 | if (err == 0) { | |
888 | whole->bd_claiming = holder; | |
889 | spin_unlock(&bdev_lock); | |
890 | return whole; | |
891 | } else { | |
892 | spin_unlock(&bdev_lock); | |
893 | bdput(whole); | |
894 | return ERR_PTR(err); | |
895 | } | |
896 | } | |
897 | ||
641dc636 | 898 | #ifdef CONFIG_SYSFS |
49731baa TH |
899 | struct bd_holder_disk { |
900 | struct list_head list; | |
901 | struct gendisk *disk; | |
902 | int refcnt; | |
903 | }; | |
904 | ||
905 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
906 | struct gendisk *disk) | |
907 | { | |
908 | struct bd_holder_disk *holder; | |
909 | ||
910 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
911 | if (holder->disk == disk) | |
912 | return holder; | |
913 | return NULL; | |
914 | } | |
915 | ||
4d7dd8fd | 916 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 917 | { |
4d7dd8fd | 918 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
919 | } |
920 | ||
921 | static void del_symlink(struct kobject *from, struct kobject *to) | |
922 | { | |
641dc636 JN |
923 | sysfs_remove_link(from, kobject_name(to)); |
924 | } | |
925 | ||
df6c0cd9 | 926 | /** |
e09b457b TH |
927 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
928 | * @bdev: the claimed slave bdev | |
929 | * @disk: the holding disk | |
df6c0cd9 | 930 | * |
49731baa TH |
931 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
932 | * | |
e09b457b | 933 | * This functions creates the following sysfs symlinks. |
641dc636 | 934 | * |
e09b457b TH |
935 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
936 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 937 | * |
e09b457b TH |
938 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
939 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 940 | * |
e09b457b TH |
941 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
942 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 943 | * |
e09b457b TH |
944 | * The caller must have claimed @bdev before calling this function and |
945 | * ensure that both @bdev and @disk are valid during the creation and | |
946 | * lifetime of these symlinks. | |
641dc636 | 947 | * |
e09b457b TH |
948 | * CONTEXT: |
949 | * Might sleep. | |
641dc636 | 950 | * |
e09b457b TH |
951 | * RETURNS: |
952 | * 0 on success, -errno on failure. | |
641dc636 | 953 | */ |
e09b457b | 954 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 955 | { |
49731baa | 956 | struct bd_holder_disk *holder; |
e09b457b | 957 | int ret = 0; |
641dc636 | 958 | |
2e7b651d | 959 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 960 | |
49731baa | 961 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 962 | |
e09b457b TH |
963 | /* FIXME: remove the following once add_disk() handles errors */ |
964 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
965 | goto out_unlock; | |
4e91672c | 966 | |
49731baa TH |
967 | holder = bd_find_holder_disk(bdev, disk); |
968 | if (holder) { | |
969 | holder->refcnt++; | |
e09b457b | 970 | goto out_unlock; |
49731baa | 971 | } |
641dc636 | 972 | |
49731baa TH |
973 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
974 | if (!holder) { | |
975 | ret = -ENOMEM; | |
e09b457b TH |
976 | goto out_unlock; |
977 | } | |
641dc636 | 978 | |
49731baa TH |
979 | INIT_LIST_HEAD(&holder->list); |
980 | holder->disk = disk; | |
981 | holder->refcnt = 1; | |
982 | ||
983 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
984 | if (ret) | |
985 | goto out_free; | |
986 | ||
987 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
988 | if (ret) | |
989 | goto out_del; | |
e7407d16 TH |
990 | /* |
991 | * bdev could be deleted beneath us which would implicitly destroy | |
992 | * the holder directory. Hold on to it. | |
993 | */ | |
994 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
995 | |
996 | list_add(&holder->list, &bdev->bd_holder_disks); | |
997 | goto out_unlock; | |
998 | ||
999 | out_del: | |
1000 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1001 | out_free: | |
1002 | kfree(holder); | |
e09b457b | 1003 | out_unlock: |
b4cf1b72 | 1004 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1005 | return ret; |
641dc636 | 1006 | } |
e09b457b | 1007 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1008 | |
49731baa TH |
1009 | /** |
1010 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1011 | * @bdev: the calimed slave bdev | |
1012 | * @disk: the holding disk | |
1013 | * | |
1014 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1015 | * | |
1016 | * CONTEXT: | |
1017 | * Might sleep. | |
1018 | */ | |
1019 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1020 | { |
49731baa | 1021 | struct bd_holder_disk *holder; |
641dc636 | 1022 | |
49731baa | 1023 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1024 | |
49731baa TH |
1025 | holder = bd_find_holder_disk(bdev, disk); |
1026 | ||
1027 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1028 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1029 | del_symlink(bdev->bd_part->holder_dir, | |
1030 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1031 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1032 | list_del_init(&holder->list); |
1033 | kfree(holder); | |
1034 | } | |
1035 | ||
1036 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1037 | } |
49731baa | 1038 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1039 | #endif |
1da177e4 | 1040 | |
56ade44b AP |
1041 | /** |
1042 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1043 | * | |
1044 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1045 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1046 | * |
1047 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1048 | * when a disk has been changed -- either by a media change or online | |
1049 | * resize. | |
1050 | */ | |
93b270f7 | 1051 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1052 | { |
93b270f7 | 1053 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1054 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1055 | "resized disk %s\n", |
1056 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b AP |
1057 | } |
1058 | ||
1059 | if (!bdev->bd_disk) | |
1060 | return; | |
d27769ec | 1061 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1062 | bdev->bd_invalidated = 1; |
1063 | } | |
1064 | ||
c3279d14 | 1065 | /** |
57d1b536 | 1066 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1067 | * @disk: struct gendisk to check |
1068 | * @bdev: struct bdev to adjust. | |
1069 | * | |
1070 | * This routine checks to see if the bdev size does not match the disk size | |
1071 | * and adjusts it if it differs. | |
1072 | */ | |
1073 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1074 | { | |
1075 | loff_t disk_size, bdev_size; | |
1076 | ||
1077 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1078 | bdev_size = i_size_read(bdev->bd_inode); | |
1079 | if (disk_size != bdev_size) { | |
c3279d14 AP |
1080 | printk(KERN_INFO |
1081 | "%s: detected capacity change from %lld to %lld\n", | |
424081f3 | 1082 | disk->disk_name, bdev_size, disk_size); |
c3279d14 | 1083 | i_size_write(bdev->bd_inode, disk_size); |
93b270f7 | 1084 | flush_disk(bdev, false); |
c3279d14 AP |
1085 | } |
1086 | } | |
1087 | EXPORT_SYMBOL(check_disk_size_change); | |
1088 | ||
0c002c2f | 1089 | /** |
57d1b536 | 1090 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1091 | * @disk: struct gendisk to be revalidated |
1092 | * | |
1093 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1094 | * call-backs. It is used to do common pre and post operations needed | |
1095 | * for all revalidate_disk operations. | |
1096 | */ | |
1097 | int revalidate_disk(struct gendisk *disk) | |
1098 | { | |
c3279d14 | 1099 | struct block_device *bdev; |
0c002c2f AP |
1100 | int ret = 0; |
1101 | ||
1102 | if (disk->fops->revalidate_disk) | |
1103 | ret = disk->fops->revalidate_disk(disk); | |
25520d55 | 1104 | blk_integrity_revalidate(disk); |
c3279d14 AP |
1105 | bdev = bdget_disk(disk, 0); |
1106 | if (!bdev) | |
1107 | return ret; | |
1108 | ||
1109 | mutex_lock(&bdev->bd_mutex); | |
1110 | check_disk_size_change(disk, bdev); | |
7630b661 | 1111 | bdev->bd_invalidated = 0; |
c3279d14 AP |
1112 | mutex_unlock(&bdev->bd_mutex); |
1113 | bdput(bdev); | |
0c002c2f AP |
1114 | return ret; |
1115 | } | |
1116 | EXPORT_SYMBOL(revalidate_disk); | |
1117 | ||
1da177e4 LT |
1118 | /* |
1119 | * This routine checks whether a removable media has been changed, | |
1120 | * and invalidates all buffer-cache-entries in that case. This | |
1121 | * is a relatively slow routine, so we have to try to minimize using | |
1122 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1123 | * is the best way of combining speed and utility, I think. | |
1124 | * People changing diskettes in the middle of an operation deserve | |
1125 | * to lose :-) | |
1126 | */ | |
1127 | int check_disk_change(struct block_device *bdev) | |
1128 | { | |
1129 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1130 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1131 | unsigned int events; |
1da177e4 | 1132 | |
77ea887e TH |
1133 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1134 | DISK_EVENT_EJECT_REQUEST); | |
1135 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1136 | return 0; |
1137 | ||
93b270f7 | 1138 | flush_disk(bdev, true); |
1da177e4 LT |
1139 | if (bdops->revalidate_disk) |
1140 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1141 | return 1; |
1142 | } | |
1143 | ||
1144 | EXPORT_SYMBOL(check_disk_change); | |
1145 | ||
1146 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1147 | { | |
e1defc4f | 1148 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1149 | |
5955102c | 1150 | inode_lock(bdev->bd_inode); |
d646a02a | 1151 | i_size_write(bdev->bd_inode, size); |
5955102c | 1152 | inode_unlock(bdev->bd_inode); |
09cbfeaf | 1153 | while (bsize < PAGE_SIZE) { |
1da177e4 LT |
1154 | if (size & bsize) |
1155 | break; | |
1156 | bsize <<= 1; | |
1157 | } | |
1158 | bdev->bd_block_size = bsize; | |
1159 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1160 | } | |
1161 | EXPORT_SYMBOL(bd_set_size); | |
1162 | ||
acc93d30 DW |
1163 | static bool blkdev_dax_capable(struct block_device *bdev) |
1164 | { | |
1165 | struct gendisk *disk = bdev->bd_disk; | |
1166 | ||
1167 | if (!disk->fops->direct_access || !IS_ENABLED(CONFIG_FS_DAX)) | |
1168 | return false; | |
1169 | ||
1170 | /* | |
1171 | * If the partition is not aligned on a page boundary, we can't | |
1172 | * do dax I/O to it. | |
1173 | */ | |
1174 | if ((bdev->bd_part->start_sect % (PAGE_SIZE / 512)) | |
1175 | || (bdev->bd_part->nr_sects % (PAGE_SIZE / 512))) | |
1176 | return false; | |
1177 | ||
1178 | /* | |
1179 | * If the device has known bad blocks, force all I/O through the | |
1180 | * driver / page cache. | |
1181 | * | |
1182 | * TODO: support finer grained dax error handling | |
1183 | */ | |
1184 | if (disk->bb && disk->bb->count) | |
1185 | return false; | |
1186 | ||
1187 | return true; | |
1188 | } | |
1189 | ||
4385bab1 | 1190 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1191 | |
6d740cd5 PZ |
1192 | /* |
1193 | * bd_mutex locking: | |
1194 | * | |
1195 | * mutex_lock(part->bd_mutex) | |
1196 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1197 | */ | |
1198 | ||
572c4892 | 1199 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1200 | { |
1da177e4 | 1201 | struct gendisk *disk; |
523e1d39 | 1202 | struct module *owner; |
7db9cfd3 | 1203 | int ret; |
cf771cb5 | 1204 | int partno; |
fe6e9c1f AV |
1205 | int perm = 0; |
1206 | ||
572c4892 | 1207 | if (mode & FMODE_READ) |
fe6e9c1f | 1208 | perm |= MAY_READ; |
572c4892 | 1209 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1210 | perm |= MAY_WRITE; |
1211 | /* | |
1212 | * hooks: /n/, see "layering violations". | |
1213 | */ | |
b7300b78 CW |
1214 | if (!for_part) { |
1215 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1216 | if (ret != 0) { | |
1217 | bdput(bdev); | |
1218 | return ret; | |
1219 | } | |
82666020 | 1220 | } |
7db9cfd3 | 1221 | |
d3374825 | 1222 | restart: |
0762b8bd | 1223 | |
89f97496 | 1224 | ret = -ENXIO; |
cf771cb5 | 1225 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1226 | if (!disk) |
6e9624b8 | 1227 | goto out; |
523e1d39 | 1228 | owner = disk->fops->owner; |
1da177e4 | 1229 | |
69e02c59 | 1230 | disk_block_events(disk); |
6796bf54 | 1231 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1232 | if (!bdev->bd_openers) { |
1233 | bdev->bd_disk = disk; | |
87192a2a | 1234 | bdev->bd_queue = disk->queue; |
1da177e4 | 1235 | bdev->bd_contains = bdev; |
03cdadb0 DW |
1236 | if (IS_ENABLED(CONFIG_BLK_DEV_DAX) && disk->fops->direct_access) |
1237 | bdev->bd_inode->i_flags = S_DAX; | |
1238 | else | |
1239 | bdev->bd_inode->i_flags = 0; | |
1240 | ||
cf771cb5 | 1241 | if (!partno) { |
89f97496 TH |
1242 | ret = -ENXIO; |
1243 | bdev->bd_part = disk_get_part(disk, partno); | |
1244 | if (!bdev->bd_part) | |
1245 | goto out_clear; | |
1246 | ||
1196f8b8 | 1247 | ret = 0; |
1da177e4 | 1248 | if (disk->fops->open) { |
572c4892 | 1249 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1250 | if (ret == -ERESTARTSYS) { |
1251 | /* Lost a race with 'disk' being | |
1252 | * deleted, try again. | |
1253 | * See md.c | |
1254 | */ | |
1255 | disk_put_part(bdev->bd_part); | |
1256 | bdev->bd_part = NULL; | |
d3374825 | 1257 | bdev->bd_disk = NULL; |
87192a2a | 1258 | bdev->bd_queue = NULL; |
d3374825 | 1259 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1260 | disk_unblock_events(disk); |
69e02c59 | 1261 | put_disk(disk); |
523e1d39 | 1262 | module_put(owner); |
d3374825 N |
1263 | goto restart; |
1264 | } | |
1da177e4 | 1265 | } |
7e69723f | 1266 | |
5a023cdb | 1267 | if (!ret) { |
7e69723f | 1268 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
5a023cdb DW |
1269 | if (!blkdev_dax_capable(bdev)) |
1270 | bdev->bd_inode->i_flags &= ~S_DAX; | |
1271 | } | |
7e69723f | 1272 | |
1196f8b8 TH |
1273 | /* |
1274 | * If the device is invalidated, rescan partition | |
1275 | * if open succeeded or failed with -ENOMEDIUM. | |
1276 | * The latter is necessary to prevent ghost | |
1277 | * partitions on a removed medium. | |
1278 | */ | |
fe316bf2 JN |
1279 | if (bdev->bd_invalidated) { |
1280 | if (!ret) | |
1281 | rescan_partitions(disk, bdev); | |
1282 | else if (ret == -ENOMEDIUM) | |
1283 | invalidate_partitions(disk, bdev); | |
1284 | } | |
5a023cdb | 1285 | |
1196f8b8 TH |
1286 | if (ret) |
1287 | goto out_clear; | |
1da177e4 | 1288 | } else { |
1da177e4 LT |
1289 | struct block_device *whole; |
1290 | whole = bdget_disk(disk, 0); | |
1291 | ret = -ENOMEM; | |
1292 | if (!whole) | |
0762b8bd | 1293 | goto out_clear; |
37be4124 | 1294 | BUG_ON(for_part); |
572c4892 | 1295 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1296 | if (ret) |
0762b8bd | 1297 | goto out_clear; |
1da177e4 | 1298 | bdev->bd_contains = whole; |
89f97496 | 1299 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1300 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1301 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1302 | ret = -ENXIO; |
0762b8bd | 1303 | goto out_clear; |
1da177e4 | 1304 | } |
89f97496 | 1305 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
5a023cdb | 1306 | if (!blkdev_dax_capable(bdev)) |
f0b2e563 | 1307 | bdev->bd_inode->i_flags &= ~S_DAX; |
1da177e4 LT |
1308 | } |
1309 | } else { | |
1da177e4 | 1310 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1311 | ret = 0; |
1312 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1313 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1314 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1315 | if (bdev->bd_invalidated) { |
1316 | if (!ret) | |
1317 | rescan_partitions(bdev->bd_disk, bdev); | |
1318 | else if (ret == -ENOMEDIUM) | |
1319 | invalidate_partitions(bdev->bd_disk, bdev); | |
1320 | } | |
1196f8b8 TH |
1321 | if (ret) |
1322 | goto out_unlock_bdev; | |
1da177e4 | 1323 | } |
69e02c59 | 1324 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1325 | put_disk(disk); |
523e1d39 | 1326 | module_put(owner); |
1da177e4 LT |
1327 | } |
1328 | bdev->bd_openers++; | |
37be4124 N |
1329 | if (for_part) |
1330 | bdev->bd_part_count++; | |
c039e313 | 1331 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1332 | disk_unblock_events(disk); |
1da177e4 LT |
1333 | return 0; |
1334 | ||
0762b8bd | 1335 | out_clear: |
89f97496 | 1336 | disk_put_part(bdev->bd_part); |
1da177e4 | 1337 | bdev->bd_disk = NULL; |
0762b8bd | 1338 | bdev->bd_part = NULL; |
87192a2a | 1339 | bdev->bd_queue = NULL; |
1da177e4 | 1340 | if (bdev != bdev->bd_contains) |
572c4892 | 1341 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1342 | bdev->bd_contains = NULL; |
0762b8bd | 1343 | out_unlock_bdev: |
c039e313 | 1344 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1345 | disk_unblock_events(disk); |
0762b8bd | 1346 | put_disk(disk); |
523e1d39 | 1347 | module_put(owner); |
4345caba | 1348 | out: |
0762b8bd TH |
1349 | bdput(bdev); |
1350 | ||
1da177e4 LT |
1351 | return ret; |
1352 | } | |
1353 | ||
d4d77629 TH |
1354 | /** |
1355 | * blkdev_get - open a block device | |
1356 | * @bdev: block_device to open | |
1357 | * @mode: FMODE_* mask | |
1358 | * @holder: exclusive holder identifier | |
1359 | * | |
1360 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1361 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1362 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1363 | * | |
1364 | * On success, the reference count of @bdev is unchanged. On failure, | |
1365 | * @bdev is put. | |
1366 | * | |
1367 | * CONTEXT: | |
1368 | * Might sleep. | |
1369 | * | |
1370 | * RETURNS: | |
1371 | * 0 on success, -errno on failure. | |
1372 | */ | |
e525fd89 | 1373 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1374 | { |
e525fd89 TH |
1375 | struct block_device *whole = NULL; |
1376 | int res; | |
1377 | ||
1378 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1379 | ||
1380 | if ((mode & FMODE_EXCL) && holder) { | |
1381 | whole = bd_start_claiming(bdev, holder); | |
1382 | if (IS_ERR(whole)) { | |
1383 | bdput(bdev); | |
1384 | return PTR_ERR(whole); | |
1385 | } | |
1386 | } | |
1387 | ||
1388 | res = __blkdev_get(bdev, mode, 0); | |
1389 | ||
1390 | if (whole) { | |
d4dc210f TH |
1391 | struct gendisk *disk = whole->bd_disk; |
1392 | ||
6a027eff | 1393 | /* finish claiming */ |
77ea887e | 1394 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1395 | spin_lock(&bdev_lock); |
1396 | ||
77ea887e | 1397 | if (!res) { |
6a027eff TH |
1398 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1399 | /* | |
1400 | * Note that for a whole device bd_holders | |
1401 | * will be incremented twice, and bd_holder | |
1402 | * will be set to bd_may_claim before being | |
1403 | * set to holder | |
1404 | */ | |
1405 | whole->bd_holders++; | |
1406 | whole->bd_holder = bd_may_claim; | |
1407 | bdev->bd_holders++; | |
1408 | bdev->bd_holder = holder; | |
1409 | } | |
1410 | ||
1411 | /* tell others that we're done */ | |
1412 | BUG_ON(whole->bd_claiming != holder); | |
1413 | whole->bd_claiming = NULL; | |
1414 | wake_up_bit(&whole->bd_claiming, 0); | |
1415 | ||
1416 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1417 | |
1418 | /* | |
d4dc210f TH |
1419 | * Block event polling for write claims if requested. Any |
1420 | * write holder makes the write_holder state stick until | |
1421 | * all are released. This is good enough and tracking | |
1422 | * individual writeable reference is too fragile given the | |
1423 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1424 | */ |
4c49ff3f TH |
1425 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1426 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1427 | bdev->bd_write_holder = true; |
d4dc210f | 1428 | disk_block_events(disk); |
77ea887e TH |
1429 | } |
1430 | ||
1431 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1432 | bdput(whole); |
e525fd89 TH |
1433 | } |
1434 | ||
1435 | return res; | |
37be4124 | 1436 | } |
1da177e4 LT |
1437 | EXPORT_SYMBOL(blkdev_get); |
1438 | ||
d4d77629 TH |
1439 | /** |
1440 | * blkdev_get_by_path - open a block device by name | |
1441 | * @path: path to the block device to open | |
1442 | * @mode: FMODE_* mask | |
1443 | * @holder: exclusive holder identifier | |
1444 | * | |
1445 | * Open the blockdevice described by the device file at @path. @mode | |
1446 | * and @holder are identical to blkdev_get(). | |
1447 | * | |
1448 | * On success, the returned block_device has reference count of one. | |
1449 | * | |
1450 | * CONTEXT: | |
1451 | * Might sleep. | |
1452 | * | |
1453 | * RETURNS: | |
1454 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1455 | */ | |
1456 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1457 | void *holder) | |
1458 | { | |
1459 | struct block_device *bdev; | |
1460 | int err; | |
1461 | ||
1462 | bdev = lookup_bdev(path); | |
1463 | if (IS_ERR(bdev)) | |
1464 | return bdev; | |
1465 | ||
1466 | err = blkdev_get(bdev, mode, holder); | |
1467 | if (err) | |
1468 | return ERR_PTR(err); | |
1469 | ||
e51900f7 CE |
1470 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1471 | blkdev_put(bdev, mode); | |
1472 | return ERR_PTR(-EACCES); | |
1473 | } | |
1474 | ||
d4d77629 TH |
1475 | return bdev; |
1476 | } | |
1477 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1478 | ||
1479 | /** | |
1480 | * blkdev_get_by_dev - open a block device by device number | |
1481 | * @dev: device number of block device to open | |
1482 | * @mode: FMODE_* mask | |
1483 | * @holder: exclusive holder identifier | |
1484 | * | |
1485 | * Open the blockdevice described by device number @dev. @mode and | |
1486 | * @holder are identical to blkdev_get(). | |
1487 | * | |
1488 | * Use it ONLY if you really do not have anything better - i.e. when | |
1489 | * you are behind a truly sucky interface and all you are given is a | |
1490 | * device number. _Never_ to be used for internal purposes. If you | |
1491 | * ever need it - reconsider your API. | |
1492 | * | |
1493 | * On success, the returned block_device has reference count of one. | |
1494 | * | |
1495 | * CONTEXT: | |
1496 | * Might sleep. | |
1497 | * | |
1498 | * RETURNS: | |
1499 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1500 | */ | |
1501 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1502 | { | |
1503 | struct block_device *bdev; | |
1504 | int err; | |
1505 | ||
1506 | bdev = bdget(dev); | |
1507 | if (!bdev) | |
1508 | return ERR_PTR(-ENOMEM); | |
1509 | ||
1510 | err = blkdev_get(bdev, mode, holder); | |
1511 | if (err) | |
1512 | return ERR_PTR(err); | |
1513 | ||
1514 | return bdev; | |
1515 | } | |
1516 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1517 | ||
1da177e4 LT |
1518 | static int blkdev_open(struct inode * inode, struct file * filp) |
1519 | { | |
1520 | struct block_device *bdev; | |
1da177e4 LT |
1521 | |
1522 | /* | |
1523 | * Preserve backwards compatibility and allow large file access | |
1524 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1525 | * binary needs it. We might want to drop this workaround | |
1526 | * during an unstable branch. | |
1527 | */ | |
1528 | filp->f_flags |= O_LARGEFILE; | |
1529 | ||
572c4892 AV |
1530 | if (filp->f_flags & O_NDELAY) |
1531 | filp->f_mode |= FMODE_NDELAY; | |
1532 | if (filp->f_flags & O_EXCL) | |
1533 | filp->f_mode |= FMODE_EXCL; | |
1534 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1535 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1536 | ||
1da177e4 | 1537 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1538 | if (bdev == NULL) |
1539 | return -ENOMEM; | |
1da177e4 | 1540 | |
572c4892 AV |
1541 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1542 | ||
e525fd89 | 1543 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1544 | } |
1545 | ||
4385bab1 | 1546 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1547 | { |
2e7b651d | 1548 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1549 | struct block_device *victim = NULL; |
2e7b651d | 1550 | |
6796bf54 | 1551 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1552 | if (for_part) |
1553 | bdev->bd_part_count--; | |
1554 | ||
2e7b651d | 1555 | if (!--bdev->bd_openers) { |
6a027eff | 1556 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1557 | sync_blockdev(bdev); |
1558 | kill_bdev(bdev); | |
43d1c0eb ID |
1559 | |
1560 | bdev_write_inode(bdev); | |
564f00f6 | 1561 | /* |
43d1c0eb ID |
1562 | * Detaching bdev inode from its wb in __destroy_inode() |
1563 | * is too late: the queue which embeds its bdi (along with | |
1564 | * root wb) can be gone as soon as we put_disk() below. | |
94007751 | 1565 | */ |
43d1c0eb | 1566 | inode_detach_wb(bdev->bd_inode); |
2e7b651d PZ |
1567 | } |
1568 | if (bdev->bd_contains == bdev) { | |
1569 | if (disk->fops->release) | |
db2a144b | 1570 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1571 | } |
1572 | if (!bdev->bd_openers) { | |
1573 | struct module *owner = disk->fops->owner; | |
1574 | ||
0762b8bd TH |
1575 | disk_put_part(bdev->bd_part); |
1576 | bdev->bd_part = NULL; | |
2e7b651d | 1577 | bdev->bd_disk = NULL; |
37be4124 N |
1578 | if (bdev != bdev->bd_contains) |
1579 | victim = bdev->bd_contains; | |
2e7b651d | 1580 | bdev->bd_contains = NULL; |
523e1d39 TH |
1581 | |
1582 | put_disk(disk); | |
1583 | module_put(owner); | |
2e7b651d | 1584 | } |
2e7b651d PZ |
1585 | mutex_unlock(&bdev->bd_mutex); |
1586 | bdput(bdev); | |
37be4124 | 1587 | if (victim) |
9a1c3542 | 1588 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1589 | } |
1590 | ||
4385bab1 | 1591 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1592 | { |
85ef06d1 TH |
1593 | mutex_lock(&bdev->bd_mutex); |
1594 | ||
e525fd89 | 1595 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1596 | bool bdev_free; |
1597 | ||
1598 | /* | |
1599 | * Release a claim on the device. The holder fields | |
1600 | * are protected with bdev_lock. bd_mutex is to | |
1601 | * synchronize disk_holder unlinking. | |
1602 | */ | |
6a027eff TH |
1603 | spin_lock(&bdev_lock); |
1604 | ||
1605 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1606 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1607 | ||
1608 | /* bd_contains might point to self, check in a separate step */ | |
1609 | if ((bdev_free = !bdev->bd_holders)) | |
1610 | bdev->bd_holder = NULL; | |
1611 | if (!bdev->bd_contains->bd_holders) | |
1612 | bdev->bd_contains->bd_holder = NULL; | |
1613 | ||
1614 | spin_unlock(&bdev_lock); | |
1615 | ||
77ea887e TH |
1616 | /* |
1617 | * If this was the last claim, remove holder link and | |
1618 | * unblock evpoll if it was a write holder. | |
1619 | */ | |
85ef06d1 TH |
1620 | if (bdev_free && bdev->bd_write_holder) { |
1621 | disk_unblock_events(bdev->bd_disk); | |
1622 | bdev->bd_write_holder = false; | |
77ea887e | 1623 | } |
6936217c | 1624 | } |
77ea887e | 1625 | |
85ef06d1 TH |
1626 | /* |
1627 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1628 | * event. This is to ensure detection of media removal commanded | |
1629 | * from userland - e.g. eject(1). | |
1630 | */ | |
1631 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1632 | ||
1633 | mutex_unlock(&bdev->bd_mutex); | |
1634 | ||
4385bab1 | 1635 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1636 | } |
2e7b651d PZ |
1637 | EXPORT_SYMBOL(blkdev_put); |
1638 | ||
1da177e4 LT |
1639 | static int blkdev_close(struct inode * inode, struct file * filp) |
1640 | { | |
4ebb16ca | 1641 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1642 | blkdev_put(bdev, filp->f_mode); |
1643 | return 0; | |
1da177e4 LT |
1644 | } |
1645 | ||
bb93e3a5 | 1646 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1647 | { |
4ebb16ca | 1648 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1649 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1650 | |
1651 | /* | |
1652 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1653 | * to updated it before every ioctl. | |
1654 | */ | |
56b26add | 1655 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1656 | mode |= FMODE_NDELAY; |
1657 | else | |
1658 | mode &= ~FMODE_NDELAY; | |
1659 | ||
56b26add | 1660 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1661 | } |
1662 | ||
eef99380 CH |
1663 | /* |
1664 | * Write data to the block device. Only intended for the block device itself | |
1665 | * and the raw driver which basically is a fake block device. | |
1666 | * | |
1667 | * Does not take i_mutex for the write and thus is not for general purpose | |
1668 | * use. | |
1669 | */ | |
1456c0a8 | 1670 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1671 | { |
1672 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1673 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1674 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1675 | struct blk_plug plug; |
eef99380 | 1676 | ssize_t ret; |
5f380c7f | 1677 | |
7ec7b94a AV |
1678 | if (bdev_read_only(I_BDEV(bd_inode))) |
1679 | return -EPERM; | |
5f380c7f | 1680 | |
7ec7b94a | 1681 | if (!iov_iter_count(from)) |
5f380c7f AV |
1682 | return 0; |
1683 | ||
7ec7b94a AV |
1684 | if (iocb->ki_pos >= size) |
1685 | return -ENOSPC; | |
1686 | ||
1687 | iov_iter_truncate(from, size - iocb->ki_pos); | |
eef99380 | 1688 | |
53362a05 | 1689 | blk_start_plug(&plug); |
1456c0a8 | 1690 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
1691 | if (ret > 0) |
1692 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 1693 | blk_finish_plug(&plug); |
eef99380 CH |
1694 | return ret; |
1695 | } | |
1456c0a8 | 1696 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1697 | |
b2de525f | 1698 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1699 | { |
1700 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1701 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 1702 | loff_t size = i_size_read(bd_inode); |
a886038b | 1703 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1704 | |
1705 | if (pos >= size) | |
1706 | return 0; | |
1707 | ||
1708 | size -= pos; | |
a886038b AV |
1709 | iov_iter_truncate(to, size); |
1710 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1711 | } |
b2de525f | 1712 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1713 | |
87d8fe1e TT |
1714 | /* |
1715 | * Try to release a page associated with block device when the system | |
1716 | * is under memory pressure. | |
1717 | */ | |
1718 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1719 | { | |
1720 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1721 | ||
1722 | if (super && super->s_op->bdev_try_to_free_page) | |
1723 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1724 | ||
1725 | return try_to_free_buffers(page); | |
1726 | } | |
1727 | ||
7f6d5b52 RZ |
1728 | static int blkdev_writepages(struct address_space *mapping, |
1729 | struct writeback_control *wbc) | |
1730 | { | |
1731 | if (dax_mapping(mapping)) { | |
1732 | struct block_device *bdev = I_BDEV(mapping->host); | |
1733 | ||
1734 | return dax_writeback_mapping_range(mapping, bdev, wbc); | |
1735 | } | |
1736 | return generic_writepages(mapping, wbc); | |
1737 | } | |
1738 | ||
4c54ac62 | 1739 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1740 | .readpage = blkdev_readpage, |
447f05bb | 1741 | .readpages = blkdev_readpages, |
1da177e4 | 1742 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1743 | .write_begin = blkdev_write_begin, |
1744 | .write_end = blkdev_write_end, | |
7f6d5b52 | 1745 | .writepages = blkdev_writepages, |
87d8fe1e | 1746 | .releasepage = blkdev_releasepage, |
1da177e4 | 1747 | .direct_IO = blkdev_direct_IO, |
b4597226 | 1748 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1749 | }; |
1750 | ||
4b6f5d20 | 1751 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1752 | .open = blkdev_open, |
1753 | .release = blkdev_close, | |
1754 | .llseek = block_llseek, | |
a886038b | 1755 | .read_iter = blkdev_read_iter, |
1456c0a8 | 1756 | .write_iter = blkdev_write_iter, |
acc93d30 | 1757 | .mmap = generic_file_mmap, |
b1dd3b28 | 1758 | .fsync = blkdev_fsync, |
bb93e3a5 | 1759 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1760 | #ifdef CONFIG_COMPAT |
1761 | .compat_ioctl = compat_blkdev_ioctl, | |
1762 | #endif | |
1e8b3332 | 1763 | .splice_read = generic_file_splice_read, |
8d020765 | 1764 | .splice_write = iter_file_splice_write, |
1da177e4 LT |
1765 | }; |
1766 | ||
1767 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1768 | { | |
1769 | int res; | |
1770 | mm_segment_t old_fs = get_fs(); | |
1771 | set_fs(KERNEL_DS); | |
56b26add | 1772 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1773 | set_fs(old_fs); |
1774 | return res; | |
1775 | } | |
1776 | ||
1777 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1778 | ||
1779 | /** | |
1780 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1781 | * @pathname: special file representing the block device |
1da177e4 | 1782 | * |
57d1b536 | 1783 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1784 | * namespace if possible and return it. Return ERR_PTR(error) |
1785 | * otherwise. | |
1786 | */ | |
421748ec | 1787 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1788 | { |
1789 | struct block_device *bdev; | |
1790 | struct inode *inode; | |
421748ec | 1791 | struct path path; |
1da177e4 LT |
1792 | int error; |
1793 | ||
421748ec | 1794 | if (!pathname || !*pathname) |
1da177e4 LT |
1795 | return ERR_PTR(-EINVAL); |
1796 | ||
421748ec | 1797 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1798 | if (error) |
1799 | return ERR_PTR(error); | |
1800 | ||
bb668734 | 1801 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
1802 | error = -ENOTBLK; |
1803 | if (!S_ISBLK(inode->i_mode)) | |
1804 | goto fail; | |
1805 | error = -EACCES; | |
421748ec | 1806 | if (path.mnt->mnt_flags & MNT_NODEV) |
1da177e4 LT |
1807 | goto fail; |
1808 | error = -ENOMEM; | |
1809 | bdev = bd_acquire(inode); | |
1810 | if (!bdev) | |
1811 | goto fail; | |
1812 | out: | |
421748ec | 1813 | path_put(&path); |
1da177e4 LT |
1814 | return bdev; |
1815 | fail: | |
1816 | bdev = ERR_PTR(error); | |
1817 | goto out; | |
1818 | } | |
d5686b44 | 1819 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 1820 | |
93b270f7 | 1821 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
1822 | { |
1823 | struct super_block *sb = get_super(bdev); | |
1824 | int res = 0; | |
1825 | ||
1826 | if (sb) { | |
1827 | /* | |
1828 | * no need to lock the super, get_super holds the | |
1829 | * read mutex so the filesystem cannot go away | |
1830 | * under us (->put_super runs with the write lock | |
1831 | * hold). | |
1832 | */ | |
1833 | shrink_dcache_sb(sb); | |
93b270f7 | 1834 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
1835 | drop_super(sb); |
1836 | } | |
f98393a6 | 1837 | invalidate_bdev(bdev); |
b71e8a4c DH |
1838 | return res; |
1839 | } | |
1840 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
1841 | |
1842 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
1843 | { | |
1844 | struct inode *inode, *old_inode = NULL; | |
1845 | ||
74278da9 | 1846 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1847 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
1848 | struct address_space *mapping = inode->i_mapping; | |
1849 | ||
1850 | spin_lock(&inode->i_lock); | |
1851 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
1852 | mapping->nrpages == 0) { | |
1853 | spin_unlock(&inode->i_lock); | |
1854 | continue; | |
1855 | } | |
1856 | __iget(inode); | |
1857 | spin_unlock(&inode->i_lock); | |
74278da9 | 1858 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1859 | /* |
1860 | * We hold a reference to 'inode' so it couldn't have been | |
1861 | * removed from s_inodes list while we dropped the | |
74278da9 | 1862 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 1863 | * be holding the last reference and we cannot iput it under |
74278da9 | 1864 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
1865 | * later. |
1866 | */ | |
1867 | iput(old_inode); | |
1868 | old_inode = inode; | |
1869 | ||
1870 | func(I_BDEV(inode), arg); | |
1871 | ||
74278da9 | 1872 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 1873 | } |
74278da9 | 1874 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1875 | iput(old_inode); |
1876 | } |