]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/block_dev.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992 Linus Torvalds | |
6 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/init.h> |
10 | #include <linux/mm.h> | |
11 | #include <linux/fcntl.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/kmod.h> | |
14 | #include <linux/major.h> | |
7db9cfd3 | 15 | #include <linux/device_cgroup.h> |
1da177e4 LT |
16 | #include <linux/highmem.h> |
17 | #include <linux/blkdev.h> | |
66114cad | 18 | #include <linux/backing-dev.h> |
1da177e4 LT |
19 | #include <linux/module.h> |
20 | #include <linux/blkpg.h> | |
b502bd11 | 21 | #include <linux/magic.h> |
1da177e4 | 22 | #include <linux/buffer_head.h> |
ff01bb48 | 23 | #include <linux/swap.h> |
585d3bc0 | 24 | #include <linux/pagevec.h> |
811d736f | 25 | #include <linux/writeback.h> |
1da177e4 LT |
26 | #include <linux/mpage.h> |
27 | #include <linux/mount.h> | |
9030d16e | 28 | #include <linux/pseudo_fs.h> |
1da177e4 LT |
29 | #include <linux/uio.h> |
30 | #include <linux/namei.h> | |
1368c4f2 | 31 | #include <linux/log2.h> |
ff01bb48 | 32 | #include <linux/cleancache.h> |
189ce2b9 | 33 | #include <linux/task_io_accounting_ops.h> |
25f4c414 | 34 | #include <linux/falloc.h> |
7c0f6ba6 | 35 | #include <linux/uaccess.h> |
56939e01 | 36 | #include <linux/suspend.h> |
07f3f05c | 37 | #include "internal.h" |
1da177e4 LT |
38 | |
39 | struct bdev_inode { | |
40 | struct block_device bdev; | |
41 | struct inode vfs_inode; | |
42 | }; | |
43 | ||
4c54ac62 AB |
44 | static const struct address_space_operations def_blk_aops; |
45 | ||
1da177e4 LT |
46 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
47 | { | |
48 | return container_of(inode, struct bdev_inode, vfs_inode); | |
49 | } | |
50 | ||
ff5053f6 | 51 | struct block_device *I_BDEV(struct inode *inode) |
1da177e4 LT |
52 | { |
53 | return &BDEV_I(inode)->bdev; | |
54 | } | |
1da177e4 LT |
55 | EXPORT_SYMBOL(I_BDEV); |
56 | ||
dbd3ca50 | 57 | static void bdev_write_inode(struct block_device *bdev) |
564f00f6 | 58 | { |
dbd3ca50 VG |
59 | struct inode *inode = bdev->bd_inode; |
60 | int ret; | |
61 | ||
564f00f6 CH |
62 | spin_lock(&inode->i_lock); |
63 | while (inode->i_state & I_DIRTY) { | |
64 | spin_unlock(&inode->i_lock); | |
dbd3ca50 VG |
65 | ret = write_inode_now(inode, true); |
66 | if (ret) { | |
67 | char name[BDEVNAME_SIZE]; | |
68 | pr_warn_ratelimited("VFS: Dirty inode writeback failed " | |
69 | "for block device %s (err=%d).\n", | |
70 | bdevname(bdev, name), ret); | |
71 | } | |
564f00f6 CH |
72 | spin_lock(&inode->i_lock); |
73 | } | |
74 | spin_unlock(&inode->i_lock); | |
75 | } | |
76 | ||
f9a14399 | 77 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
3373a346 | 78 | static void kill_bdev(struct block_device *bdev) |
1da177e4 | 79 | { |
ff01bb48 AV |
80 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
81 | ||
f9fe48be | 82 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
f9a14399 | 83 | return; |
ff01bb48 | 84 | |
f9a14399 | 85 | invalidate_bh_lrus(); |
ff01bb48 | 86 | truncate_inode_pages(mapping, 0); |
3373a346 | 87 | } |
ff01bb48 AV |
88 | |
89 | /* Invalidate clean unused buffers and pagecache. */ | |
90 | void invalidate_bdev(struct block_device *bdev) | |
91 | { | |
92 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
93 | ||
a5f6a6a9 AR |
94 | if (mapping->nrpages) { |
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 | } | |
ff01bb48 AV |
99 | /* 99% of the time, we don't need to flush the cleancache on the bdev. |
100 | * But, for the strange corners, lets be cautious | |
101 | */ | |
3167760f | 102 | cleancache_invalidate_inode(mapping); |
ff01bb48 AV |
103 | } |
104 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 | 105 | |
384d87ef JK |
106 | /* |
107 | * Drop all buffers & page cache for given bdev range. This function bails | |
108 | * with error if bdev has other exclusive owner (such as filesystem). | |
109 | */ | |
110 | int truncate_bdev_range(struct block_device *bdev, fmode_t mode, | |
111 | loff_t lstart, loff_t lend) | |
112 | { | |
113 | struct block_device *claimed_bdev = NULL; | |
114 | int err; | |
115 | ||
116 | /* | |
117 | * If we don't hold exclusive handle for the device, upgrade to it | |
118 | * while we discard the buffer cache to avoid discarding buffers | |
119 | * under live filesystem. | |
120 | */ | |
121 | if (!(mode & FMODE_EXCL)) { | |
122 | claimed_bdev = bdev->bd_contains; | |
123 | err = bd_prepare_to_claim(bdev, claimed_bdev, | |
124 | truncate_bdev_range); | |
125 | if (err) | |
126 | return err; | |
127 | } | |
128 | truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend); | |
129 | if (claimed_bdev) | |
130 | bd_abort_claiming(bdev, claimed_bdev, truncate_bdev_range); | |
131 | return 0; | |
132 | } | |
133 | EXPORT_SYMBOL(truncate_bdev_range); | |
134 | ||
04906b2f JK |
135 | static void set_init_blocksize(struct block_device *bdev) |
136 | { | |
6b7b181b | 137 | bdev->bd_inode->i_blkbits = blksize_bits(bdev_logical_block_size(bdev)); |
04906b2f JK |
138 | } |
139 | ||
1da177e4 LT |
140 | int set_blocksize(struct block_device *bdev, int size) |
141 | { | |
142 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 143 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
144 | return -EINVAL; |
145 | ||
146 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 147 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
148 | return -EINVAL; |
149 | ||
150 | /* Don't change the size if it is same as current */ | |
6b7b181b | 151 | if (bdev->bd_inode->i_blkbits != blksize_bits(size)) { |
1da177e4 | 152 | sync_blockdev(bdev); |
1da177e4 LT |
153 | bdev->bd_inode->i_blkbits = blksize_bits(size); |
154 | kill_bdev(bdev); | |
155 | } | |
156 | return 0; | |
157 | } | |
158 | ||
159 | EXPORT_SYMBOL(set_blocksize); | |
160 | ||
161 | int sb_set_blocksize(struct super_block *sb, int size) | |
162 | { | |
1da177e4 LT |
163 | if (set_blocksize(sb->s_bdev, size)) |
164 | return 0; | |
165 | /* If we get here, we know size is power of two | |
166 | * and it's value is between 512 and PAGE_SIZE */ | |
167 | sb->s_blocksize = size; | |
38885bd4 | 168 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
169 | return sb->s_blocksize; |
170 | } | |
171 | ||
172 | EXPORT_SYMBOL(sb_set_blocksize); | |
173 | ||
174 | int sb_min_blocksize(struct super_block *sb, int size) | |
175 | { | |
e1defc4f | 176 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
177 | if (size < minsize) |
178 | size = minsize; | |
179 | return sb_set_blocksize(sb, size); | |
180 | } | |
181 | ||
182 | EXPORT_SYMBOL(sb_min_blocksize); | |
183 | ||
184 | static int | |
185 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
186 | struct buffer_head *bh, int create) | |
187 | { | |
1da177e4 LT |
188 | bh->b_bdev = I_BDEV(inode); |
189 | bh->b_blocknr = iblock; | |
190 | set_buffer_mapped(bh); | |
191 | return 0; | |
192 | } | |
193 | ||
4ebb16ca DW |
194 | static struct inode *bdev_file_inode(struct file *file) |
195 | { | |
196 | return file->f_mapping->host; | |
197 | } | |
198 | ||
78250c02 JA |
199 | static unsigned int dio_bio_write_op(struct kiocb *iocb) |
200 | { | |
201 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; | |
202 | ||
203 | /* avoid the need for a I/O completion work item */ | |
204 | if (iocb->ki_flags & IOCB_DSYNC) | |
205 | op |= REQ_FUA; | |
206 | return op; | |
207 | } | |
208 | ||
189ce2b9 CH |
209 | #define DIO_INLINE_BIO_VECS 4 |
210 | ||
211 | static void blkdev_bio_end_io_simple(struct bio *bio) | |
212 | { | |
213 | struct task_struct *waiter = bio->bi_private; | |
214 | ||
215 | WRITE_ONCE(bio->bi_private, NULL); | |
0619317f | 216 | blk_wake_io_task(waiter); |
189ce2b9 CH |
217 | } |
218 | ||
219 | static ssize_t | |
220 | __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter, | |
221 | int nr_pages) | |
222 | { | |
223 | struct file *file = iocb->ki_filp; | |
224 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
9fec4a21 | 225 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs; |
189ce2b9 CH |
226 | loff_t pos = iocb->ki_pos; |
227 | bool should_dirty = false; | |
228 | struct bio bio; | |
229 | ssize_t ret; | |
230 | blk_qc_t qc; | |
189ce2b9 | 231 | |
9a794fb9 JA |
232 | if ((pos | iov_iter_alignment(iter)) & |
233 | (bdev_logical_block_size(bdev) - 1)) | |
189ce2b9 CH |
234 | return -EINVAL; |
235 | ||
72ecad22 JA |
236 | if (nr_pages <= DIO_INLINE_BIO_VECS) |
237 | vecs = inline_vecs; | |
238 | else { | |
6da2ec56 KC |
239 | vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec), |
240 | GFP_KERNEL); | |
72ecad22 JA |
241 | if (!vecs) |
242 | return -ENOMEM; | |
243 | } | |
244 | ||
3a83f467 | 245 | bio_init(&bio, vecs, nr_pages); |
74d46992 | 246 | bio_set_dev(&bio, bdev); |
4d1a4765 | 247 | bio.bi_iter.bi_sector = pos >> 9; |
45d06cf7 | 248 | bio.bi_write_hint = iocb->ki_hint; |
189ce2b9 CH |
249 | bio.bi_private = current; |
250 | bio.bi_end_io = blkdev_bio_end_io_simple; | |
074111ca | 251 | bio.bi_ioprio = iocb->ki_ioprio; |
189ce2b9 CH |
252 | |
253 | ret = bio_iov_iter_get_pages(&bio, iter); | |
254 | if (unlikely(ret)) | |
9362dd11 | 255 | goto out; |
189ce2b9 CH |
256 | ret = bio.bi_iter.bi_size; |
257 | ||
258 | if (iov_iter_rw(iter) == READ) { | |
78250c02 | 259 | bio.bi_opf = REQ_OP_READ; |
189ce2b9 CH |
260 | if (iter_is_iovec(iter)) |
261 | should_dirty = true; | |
262 | } else { | |
78250c02 | 263 | bio.bi_opf = dio_bio_write_op(iocb); |
189ce2b9 CH |
264 | task_io_account_write(ret); |
265 | } | |
d1e36282 | 266 | if (iocb->ki_flags & IOCB_HIPRI) |
0bbb280d | 267 | bio_set_polled(&bio, iocb); |
189ce2b9 CH |
268 | |
269 | qc = submit_bio(&bio); | |
270 | for (;;) { | |
1ac5cd49 | 271 | set_current_state(TASK_UNINTERRUPTIBLE); |
189ce2b9 CH |
272 | if (!READ_ONCE(bio.bi_private)) |
273 | break; | |
274 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
0a1b8b87 | 275 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
e6249cdd | 276 | blk_io_schedule(); |
189ce2b9 CH |
277 | } |
278 | __set_current_state(TASK_RUNNING); | |
279 | ||
9fec4a21 | 280 | bio_release_pages(&bio, should_dirty); |
4e4cbee9 | 281 | if (unlikely(bio.bi_status)) |
c6b1e36c | 282 | ret = blk_status_to_errno(bio.bi_status); |
9ae3b3f5 | 283 | |
9362dd11 MW |
284 | out: |
285 | if (vecs != inline_vecs) | |
286 | kfree(vecs); | |
287 | ||
9ae3b3f5 JA |
288 | bio_uninit(&bio); |
289 | ||
189ce2b9 CH |
290 | return ret; |
291 | } | |
292 | ||
542ff7bf CH |
293 | struct blkdev_dio { |
294 | union { | |
295 | struct kiocb *iocb; | |
296 | struct task_struct *waiter; | |
297 | }; | |
298 | size_t size; | |
299 | atomic_t ref; | |
300 | bool multi_bio : 1; | |
301 | bool should_dirty : 1; | |
302 | bool is_sync : 1; | |
303 | struct bio bio; | |
304 | }; | |
305 | ||
52190f8a | 306 | static struct bio_set blkdev_dio_pool; |
542ff7bf | 307 | |
eae83ce1 CH |
308 | static int blkdev_iopoll(struct kiocb *kiocb, bool wait) |
309 | { | |
310 | struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host); | |
311 | struct request_queue *q = bdev_get_queue(bdev); | |
312 | ||
313 | return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait); | |
314 | } | |
315 | ||
542ff7bf CH |
316 | static void blkdev_bio_end_io(struct bio *bio) |
317 | { | |
318 | struct blkdev_dio *dio = bio->bi_private; | |
319 | bool should_dirty = dio->should_dirty; | |
320 | ||
a89afe58 JY |
321 | if (bio->bi_status && !dio->bio.bi_status) |
322 | dio->bio.bi_status = bio->bi_status; | |
323 | ||
324 | if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) { | |
542ff7bf CH |
325 | if (!dio->is_sync) { |
326 | struct kiocb *iocb = dio->iocb; | |
4e4cbee9 | 327 | ssize_t ret; |
542ff7bf | 328 | |
4e4cbee9 | 329 | if (likely(!dio->bio.bi_status)) { |
542ff7bf CH |
330 | ret = dio->size; |
331 | iocb->ki_pos += ret; | |
4e4cbee9 CH |
332 | } else { |
333 | ret = blk_status_to_errno(dio->bio.bi_status); | |
542ff7bf CH |
334 | } |
335 | ||
336 | dio->iocb->ki_complete(iocb, ret, 0); | |
531724ab CH |
337 | if (dio->multi_bio) |
338 | bio_put(&dio->bio); | |
542ff7bf CH |
339 | } else { |
340 | struct task_struct *waiter = dio->waiter; | |
341 | ||
342 | WRITE_ONCE(dio->waiter, NULL); | |
0619317f | 343 | blk_wake_io_task(waiter); |
542ff7bf CH |
344 | } |
345 | } | |
346 | ||
347 | if (should_dirty) { | |
348 | bio_check_pages_dirty(bio); | |
349 | } else { | |
57dfe3ce | 350 | bio_release_pages(bio, false); |
542ff7bf CH |
351 | bio_put(bio); |
352 | } | |
353 | } | |
354 | ||
b2e895db | 355 | static ssize_t |
542ff7bf | 356 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages) |
b2e895db AM |
357 | { |
358 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 359 | struct inode *inode = bdev_file_inode(file); |
542ff7bf | 360 | struct block_device *bdev = I_BDEV(inode); |
64d656a1 | 361 | struct blk_plug plug; |
542ff7bf CH |
362 | struct blkdev_dio *dio; |
363 | struct bio *bio; | |
cb700eb3 | 364 | bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0; |
690e5325 | 365 | bool is_read = (iov_iter_rw(iter) == READ), is_sync; |
542ff7bf CH |
366 | loff_t pos = iocb->ki_pos; |
367 | blk_qc_t qc = BLK_QC_T_NONE; | |
7b6620d7 | 368 | int ret = 0; |
542ff7bf | 369 | |
9a794fb9 JA |
370 | if ((pos | iov_iter_alignment(iter)) & |
371 | (bdev_logical_block_size(bdev) - 1)) | |
542ff7bf CH |
372 | return -EINVAL; |
373 | ||
7b6620d7 | 374 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool); |
542ff7bf CH |
375 | |
376 | dio = container_of(bio, struct blkdev_dio, bio); | |
690e5325 | 377 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
531724ab | 378 | if (dio->is_sync) { |
542ff7bf | 379 | dio->waiter = current; |
531724ab CH |
380 | bio_get(bio); |
381 | } else { | |
542ff7bf | 382 | dio->iocb = iocb; |
531724ab | 383 | } |
542ff7bf CH |
384 | |
385 | dio->size = 0; | |
386 | dio->multi_bio = false; | |
00e23707 | 387 | dio->should_dirty = is_read && iter_is_iovec(iter); |
542ff7bf | 388 | |
cb700eb3 JA |
389 | /* |
390 | * Don't plug for HIPRI/polled IO, as those should go straight | |
391 | * to issue | |
392 | */ | |
393 | if (!is_poll) | |
394 | blk_start_plug(&plug); | |
395 | ||
542ff7bf | 396 | for (;;) { |
74d46992 | 397 | bio_set_dev(bio, bdev); |
4d1a4765 | 398 | bio->bi_iter.bi_sector = pos >> 9; |
45d06cf7 | 399 | bio->bi_write_hint = iocb->ki_hint; |
542ff7bf CH |
400 | bio->bi_private = dio; |
401 | bio->bi_end_io = blkdev_bio_end_io; | |
074111ca | 402 | bio->bi_ioprio = iocb->ki_ioprio; |
542ff7bf | 403 | |
e15c2ffa JA |
404 | ret = bio_iov_iter_get_pages(bio, iter); |
405 | if (unlikely(ret)) { | |
4e4cbee9 | 406 | bio->bi_status = BLK_STS_IOERR; |
542ff7bf CH |
407 | bio_endio(bio); |
408 | break; | |
409 | } | |
410 | ||
411 | if (is_read) { | |
412 | bio->bi_opf = REQ_OP_READ; | |
413 | if (dio->should_dirty) | |
414 | bio_set_pages_dirty(bio); | |
415 | } else { | |
416 | bio->bi_opf = dio_bio_write_op(iocb); | |
417 | task_io_account_write(bio->bi_iter.bi_size); | |
418 | } | |
419 | ||
7b6620d7 | 420 | dio->size += bio->bi_iter.bi_size; |
542ff7bf CH |
421 | pos += bio->bi_iter.bi_size; |
422 | ||
423 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); | |
424 | if (!nr_pages) { | |
eae83ce1 CH |
425 | bool polled = false; |
426 | ||
427 | if (iocb->ki_flags & IOCB_HIPRI) { | |
0bbb280d | 428 | bio_set_polled(bio, iocb); |
eae83ce1 CH |
429 | polled = true; |
430 | } | |
d34513d3 | 431 | |
542ff7bf | 432 | qc = submit_bio(bio); |
eae83ce1 CH |
433 | |
434 | if (polled) | |
435 | WRITE_ONCE(iocb->ki_cookie, qc); | |
542ff7bf CH |
436 | break; |
437 | } | |
438 | ||
439 | if (!dio->multi_bio) { | |
531724ab CH |
440 | /* |
441 | * AIO needs an extra reference to ensure the dio | |
442 | * structure which is embedded into the first bio | |
443 | * stays around. | |
444 | */ | |
445 | if (!is_sync) | |
446 | bio_get(bio); | |
542ff7bf CH |
447 | dio->multi_bio = true; |
448 | atomic_set(&dio->ref, 2); | |
449 | } else { | |
450 | atomic_inc(&dio->ref); | |
451 | } | |
452 | ||
7b6620d7 JA |
453 | submit_bio(bio); |
454 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
542ff7bf | 455 | } |
cb700eb3 JA |
456 | |
457 | if (!is_poll) | |
458 | blk_finish_plug(&plug); | |
542ff7bf | 459 | |
690e5325 | 460 | if (!is_sync) |
542ff7bf CH |
461 | return -EIOCBQUEUED; |
462 | ||
463 | for (;;) { | |
1ac5cd49 | 464 | set_current_state(TASK_UNINTERRUPTIBLE); |
542ff7bf CH |
465 | if (!READ_ONCE(dio->waiter)) |
466 | break; | |
467 | ||
468 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
0a1b8b87 | 469 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
e6249cdd | 470 | blk_io_schedule(); |
542ff7bf CH |
471 | } |
472 | __set_current_state(TASK_RUNNING); | |
473 | ||
36ffc6c1 | 474 | if (!ret) |
4e4cbee9 | 475 | ret = blk_status_to_errno(dio->bio.bi_status); |
e15c2ffa JA |
476 | if (likely(!ret)) |
477 | ret = dio->size; | |
542ff7bf CH |
478 | |
479 | bio_put(&dio->bio); | |
480 | return ret; | |
481 | } | |
482 | ||
483 | static ssize_t | |
484 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) | |
485 | { | |
189ce2b9 | 486 | int nr_pages; |
b2e895db | 487 | |
72ecad22 | 488 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
189ce2b9 CH |
489 | if (!nr_pages) |
490 | return 0; | |
72ecad22 | 491 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
189ce2b9 | 492 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
542ff7bf CH |
493 | |
494 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); | |
495 | } | |
496 | ||
497 | static __init int blkdev_init(void) | |
498 | { | |
52190f8a | 499 | return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS); |
b2e895db | 500 | } |
542ff7bf | 501 | module_init(blkdev_init); |
b2e895db | 502 | |
5cee5815 JK |
503 | int __sync_blockdev(struct block_device *bdev, int wait) |
504 | { | |
505 | if (!bdev) | |
506 | return 0; | |
507 | if (!wait) | |
508 | return filemap_flush(bdev->bd_inode->i_mapping); | |
509 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
510 | } | |
511 | ||
585d3bc0 NP |
512 | /* |
513 | * Write out and wait upon all the dirty data associated with a block | |
514 | * device via its mapping. Does not take the superblock lock. | |
515 | */ | |
516 | int sync_blockdev(struct block_device *bdev) | |
517 | { | |
5cee5815 | 518 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
519 | } |
520 | EXPORT_SYMBOL(sync_blockdev); | |
521 | ||
522 | /* | |
523 | * Write out and wait upon all dirty data associated with this | |
524 | * device. Filesystem data as well as the underlying block | |
525 | * device. Takes the superblock lock. | |
526 | */ | |
527 | int fsync_bdev(struct block_device *bdev) | |
528 | { | |
529 | struct super_block *sb = get_super(bdev); | |
530 | if (sb) { | |
60b0680f | 531 | int res = sync_filesystem(sb); |
585d3bc0 NP |
532 | drop_super(sb); |
533 | return res; | |
534 | } | |
535 | return sync_blockdev(bdev); | |
536 | } | |
47e4491b | 537 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
538 | |
539 | /** | |
540 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
541 | * @bdev: blockdevice to lock | |
542 | * | |
585d3bc0 NP |
543 | * If a superblock is found on this device, we take the s_umount semaphore |
544 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
545 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
546 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
547 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
548 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
549 | * actually. | |
550 | */ | |
040f04bd | 551 | int freeze_bdev(struct block_device *bdev) |
585d3bc0 NP |
552 | { |
553 | struct super_block *sb; | |
554 | int error = 0; | |
555 | ||
556 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
040f04bd CH |
557 | if (++bdev->bd_fsfreeze_count > 1) |
558 | goto done; | |
4504230a CH |
559 | |
560 | sb = get_active_super(bdev); | |
561 | if (!sb) | |
040f04bd | 562 | goto sync; |
48b6bca6 BM |
563 | if (sb->s_op->freeze_super) |
564 | error = sb->s_op->freeze_super(sb); | |
565 | else | |
566 | error = freeze_super(sb); | |
040f04bd CH |
567 | deactivate_super(sb); |
568 | ||
18e9e510 | 569 | if (error) { |
18e9e510 | 570 | bdev->bd_fsfreeze_count--; |
040f04bd | 571 | goto done; |
585d3bc0 | 572 | } |
040f04bd CH |
573 | bdev->bd_fsfreeze_sb = sb; |
574 | ||
575 | sync: | |
585d3bc0 | 576 | sync_blockdev(bdev); |
040f04bd | 577 | done: |
585d3bc0 | 578 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
040f04bd | 579 | return error; |
585d3bc0 NP |
580 | } |
581 | EXPORT_SYMBOL(freeze_bdev); | |
582 | ||
583 | /** | |
584 | * thaw_bdev -- unlock filesystem | |
585 | * @bdev: blockdevice to unlock | |
585d3bc0 NP |
586 | * |
587 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
588 | */ | |
040f04bd | 589 | int thaw_bdev(struct block_device *bdev) |
585d3bc0 | 590 | { |
040f04bd | 591 | struct super_block *sb; |
4504230a | 592 | int error = -EINVAL; |
585d3bc0 NP |
593 | |
594 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 595 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 596 | goto out; |
4504230a CH |
597 | |
598 | error = 0; | |
599 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 600 | goto out; |
4504230a | 601 | |
040f04bd | 602 | sb = bdev->bd_fsfreeze_sb; |
4504230a | 603 | if (!sb) |
18e9e510 | 604 | goto out; |
4504230a | 605 | |
48b6bca6 BM |
606 | if (sb->s_op->thaw_super) |
607 | error = sb->s_op->thaw_super(sb); | |
608 | else | |
609 | error = thaw_super(sb); | |
997198ba | 610 | if (error) |
18e9e510 | 611 | bdev->bd_fsfreeze_count++; |
18e9e510 | 612 | out: |
585d3bc0 | 613 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
997198ba | 614 | return error; |
585d3bc0 NP |
615 | } |
616 | EXPORT_SYMBOL(thaw_bdev); | |
617 | ||
1da177e4 LT |
618 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
619 | { | |
620 | return block_write_full_page(page, blkdev_get_block, wbc); | |
621 | } | |
622 | ||
623 | static int blkdev_readpage(struct file * file, struct page * page) | |
624 | { | |
625 | return block_read_full_page(page, blkdev_get_block); | |
626 | } | |
627 | ||
d4388340 | 628 | static void blkdev_readahead(struct readahead_control *rac) |
447f05bb | 629 | { |
d4388340 | 630 | mpage_readahead(rac, blkdev_get_block); |
447f05bb AM |
631 | } |
632 | ||
6272b5a5 NP |
633 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
634 | loff_t pos, unsigned len, unsigned flags, | |
635 | struct page **pagep, void **fsdata) | |
1da177e4 | 636 | { |
155130a4 CH |
637 | return block_write_begin(mapping, pos, len, flags, pagep, |
638 | blkdev_get_block); | |
1da177e4 LT |
639 | } |
640 | ||
6272b5a5 NP |
641 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
642 | loff_t pos, unsigned len, unsigned copied, | |
643 | struct page *page, void *fsdata) | |
1da177e4 | 644 | { |
6272b5a5 NP |
645 | int ret; |
646 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
647 | ||
648 | unlock_page(page); | |
09cbfeaf | 649 | put_page(page); |
6272b5a5 NP |
650 | |
651 | return ret; | |
1da177e4 LT |
652 | } |
653 | ||
654 | /* | |
655 | * private llseek: | |
496ad9aa | 656 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
657 | * so we compute the size by hand (just as in block_read/write above) |
658 | */ | |
965c8e59 | 659 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 660 | { |
4ebb16ca | 661 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
662 | loff_t retval; |
663 | ||
5955102c | 664 | inode_lock(bd_inode); |
5d48f3a2 | 665 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 666 | inode_unlock(bd_inode); |
1da177e4 LT |
667 | return retval; |
668 | } | |
669 | ||
02c24a82 | 670 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 671 | { |
4ebb16ca | 672 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 673 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 674 | int error; |
da5aa861 | 675 | |
372cf243 | 676 | error = file_write_and_wait_range(filp, start, end); |
da5aa861 RW |
677 | if (error) |
678 | return error; | |
ab0a9735 | 679 | |
b8af67e2 AB |
680 | /* |
681 | * There is no need to serialise calls to blkdev_issue_flush with | |
682 | * i_mutex and doing so causes performance issues with concurrent | |
683 | * O_SYNC writers to a block device. | |
684 | */ | |
9398554f | 685 | error = blkdev_issue_flush(bdev, GFP_KERNEL); |
ab0a9735 CH |
686 | if (error == -EOPNOTSUPP) |
687 | error = 0; | |
b8af67e2 | 688 | |
ab0a9735 | 689 | return error; |
1da177e4 | 690 | } |
b1dd3b28 | 691 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 692 | |
47a191fd MW |
693 | /** |
694 | * bdev_read_page() - Start reading a page from a block device | |
695 | * @bdev: The device to read the page from | |
696 | * @sector: The offset on the device to read the page to (need not be aligned) | |
697 | * @page: The page to read | |
698 | * | |
699 | * On entry, the page should be locked. It will be unlocked when the page | |
700 | * has been read. If the block driver implements rw_page synchronously, | |
701 | * that will be true on exit from this function, but it need not be. | |
702 | * | |
703 | * Errors returned by this function are usually "soft", eg out of memory, or | |
704 | * queue full; callers should try a different route to read this page rather | |
705 | * than propagate an error back up the stack. | |
706 | * | |
707 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
708 | */ | |
709 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
710 | struct page *page) | |
711 | { | |
712 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
713 | int result = -EOPNOTSUPP; |
714 | ||
f68eb1e7 | 715 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
716 | return result; |
717 | ||
e556f6ba | 718 | result = blk_queue_enter(bdev->bd_disk->queue, 0); |
2e6edc95 DW |
719 | if (result) |
720 | return result; | |
3f289dcb TH |
721 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
722 | REQ_OP_READ); | |
e556f6ba | 723 | blk_queue_exit(bdev->bd_disk->queue); |
2e6edc95 | 724 | return result; |
47a191fd | 725 | } |
47a191fd MW |
726 | |
727 | /** | |
728 | * bdev_write_page() - Start writing a page to a block device | |
729 | * @bdev: The device to write the page to | |
730 | * @sector: The offset on the device to write the page to (need not be aligned) | |
731 | * @page: The page to write | |
732 | * @wbc: The writeback_control for the write | |
733 | * | |
734 | * On entry, the page should be locked and not currently under writeback. | |
735 | * On exit, if the write started successfully, the page will be unlocked and | |
736 | * under writeback. If the write failed already (eg the driver failed to | |
737 | * queue the page to the device), the page will still be locked. If the | |
738 | * caller is a ->writepage implementation, it will need to unlock the page. | |
739 | * | |
740 | * Errors returned by this function are usually "soft", eg out of memory, or | |
741 | * queue full; callers should try a different route to write this page rather | |
742 | * than propagate an error back up the stack. | |
743 | * | |
744 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
745 | */ | |
746 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
747 | struct page *page, struct writeback_control *wbc) | |
748 | { | |
749 | int result; | |
47a191fd | 750 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
2e6edc95 | 751 | |
f68eb1e7 | 752 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 753 | return -EOPNOTSUPP; |
e556f6ba | 754 | result = blk_queue_enter(bdev->bd_disk->queue, 0); |
2e6edc95 DW |
755 | if (result) |
756 | return result; | |
757 | ||
47a191fd | 758 | set_page_writeback(page); |
3f289dcb TH |
759 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
760 | REQ_OP_WRITE); | |
f892760a | 761 | if (result) { |
47a191fd | 762 | end_page_writeback(page); |
f892760a MW |
763 | } else { |
764 | clean_page_buffers(page); | |
47a191fd | 765 | unlock_page(page); |
f892760a | 766 | } |
e556f6ba | 767 | blk_queue_exit(bdev->bd_disk->queue); |
47a191fd MW |
768 | return result; |
769 | } | |
47a191fd | 770 | |
1da177e4 LT |
771 | /* |
772 | * pseudo-fs | |
773 | */ | |
774 | ||
775 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 776 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
777 | |
778 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
779 | { | |
e94b1766 | 780 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
781 | if (!ei) |
782 | return NULL; | |
783 | return &ei->vfs_inode; | |
784 | } | |
785 | ||
41149cb0 | 786 | static void bdev_free_inode(struct inode *inode) |
1da177e4 | 787 | { |
41149cb0 | 788 | kmem_cache_free(bdev_cachep, BDEV_I(inode)); |
fa0d7e3d NP |
789 | } |
790 | ||
51cc5068 | 791 | static void init_once(void *foo) |
1da177e4 LT |
792 | { |
793 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
794 | struct block_device *bdev = &ei->bdev; | |
795 | ||
a35afb83 CL |
796 | memset(bdev, 0, sizeof(*bdev)); |
797 | mutex_init(&bdev->bd_mutex); | |
49731baa TH |
798 | #ifdef CONFIG_SYSFS |
799 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
800 | #endif | |
a5a79d00 | 801 | bdev->bd_bdi = &noop_backing_dev_info; |
a35afb83 | 802 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
803 | /* Initialize mutex for freeze. */ |
804 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
805 | } |
806 | ||
b57922d9 | 807 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
808 | { |
809 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
91b0abe3 | 810 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 811 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 812 | clear_inode(inode); |
f759741d JK |
813 | /* Detach inode from wb early as bdi_put() may free bdi->wb */ |
814 | inode_detach_wb(inode); | |
a5a79d00 | 815 | if (bdev->bd_bdi != &noop_backing_dev_info) { |
b1d2dc56 | 816 | bdi_put(bdev->bd_bdi); |
a5a79d00 JK |
817 | bdev->bd_bdi = &noop_backing_dev_info; |
818 | } | |
1da177e4 LT |
819 | } |
820 | ||
ee9b6d61 | 821 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
822 | .statfs = simple_statfs, |
823 | .alloc_inode = bdev_alloc_inode, | |
41149cb0 | 824 | .free_inode = bdev_free_inode, |
1da177e4 | 825 | .drop_inode = generic_delete_inode, |
b57922d9 | 826 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
827 | }; |
828 | ||
9030d16e | 829 | static int bd_init_fs_context(struct fs_context *fc) |
1da177e4 | 830 | { |
9030d16e DH |
831 | struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC); |
832 | if (!ctx) | |
833 | return -ENOMEM; | |
834 | fc->s_iflags |= SB_I_CGROUPWB; | |
835 | ctx->ops = &bdev_sops; | |
836 | return 0; | |
1da177e4 LT |
837 | } |
838 | ||
839 | static struct file_system_type bd_type = { | |
840 | .name = "bdev", | |
9030d16e | 841 | .init_fs_context = bd_init_fs_context, |
1da177e4 LT |
842 | .kill_sb = kill_anon_super, |
843 | }; | |
844 | ||
a212b105 TH |
845 | struct super_block *blockdev_superblock __read_mostly; |
846 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
847 | |
848 | void __init bdev_cache_init(void) | |
849 | { | |
850 | int err; | |
ace8577a | 851 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 852 | |
1da177e4 | 853 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 854 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 855 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 856 | init_once); |
1da177e4 LT |
857 | err = register_filesystem(&bd_type); |
858 | if (err) | |
859 | panic("Cannot register bdev pseudo-fs"); | |
860 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
861 | if (IS_ERR(bd_mnt)) |
862 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 863 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
864 | } |
865 | ||
10ed1666 | 866 | static struct block_device *bdget(dev_t dev) |
1da177e4 LT |
867 | { |
868 | struct block_device *bdev; | |
869 | struct inode *inode; | |
870 | ||
612c6aa7 | 871 | inode = iget_locked(blockdev_superblock, dev); |
1da177e4 LT |
872 | if (!inode) |
873 | return NULL; | |
874 | ||
875 | bdev = &BDEV_I(inode)->bdev; | |
876 | ||
877 | if (inode->i_state & I_NEW) { | |
c2b4bb8c | 878 | spin_lock_init(&bdev->bd_size_lock); |
1da177e4 | 879 | bdev->bd_contains = NULL; |
782b94cd | 880 | bdev->bd_super = NULL; |
1da177e4 | 881 | bdev->bd_inode = inode; |
1da177e4 | 882 | bdev->bd_part_count = 0; |
612c6aa7 | 883 | bdev->bd_dev = dev; |
1da177e4 LT |
884 | inode->i_mode = S_IFBLK; |
885 | inode->i_rdev = dev; | |
886 | inode->i_bdev = bdev; | |
887 | inode->i_data.a_ops = &def_blk_aops; | |
888 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
889 | unlock_new_inode(inode); |
890 | } | |
891 | return bdev; | |
892 | } | |
893 | ||
dddac6a7 AJ |
894 | /** |
895 | * bdgrab -- Grab a reference to an already referenced block device | |
896 | * @bdev: Block device to grab a reference to. | |
3a4174e6 CH |
897 | * |
898 | * Returns the block_device with an additional reference when successful, | |
899 | * or NULL if the inode is already beeing freed. | |
dddac6a7 AJ |
900 | */ |
901 | struct block_device *bdgrab(struct block_device *bdev) | |
902 | { | |
3a4174e6 CH |
903 | if (!igrab(bdev->bd_inode)) |
904 | return NULL; | |
dddac6a7 AJ |
905 | return bdev; |
906 | } | |
c1681bf8 | 907 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 908 | |
10ed1666 CH |
909 | struct block_device *bdget_part(struct hd_struct *part) |
910 | { | |
911 | return bdget(part_devt(part)); | |
912 | } | |
913 | ||
1da177e4 LT |
914 | long nr_blockdev_pages(void) |
915 | { | |
1008fe6d | 916 | struct inode *inode; |
1da177e4 | 917 | long ret = 0; |
1008fe6d CH |
918 | |
919 | spin_lock(&blockdev_superblock->s_inode_list_lock); | |
920 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) | |
921 | ret += inode->i_mapping->nrpages; | |
922 | spin_unlock(&blockdev_superblock->s_inode_list_lock); | |
923 | ||
1da177e4 LT |
924 | return ret; |
925 | } | |
926 | ||
927 | void bdput(struct block_device *bdev) | |
928 | { | |
929 | iput(bdev->bd_inode); | |
930 | } | |
931 | ||
932 | EXPORT_SYMBOL(bdput); | |
933 | ||
934 | static struct block_device *bd_acquire(struct inode *inode) | |
935 | { | |
936 | struct block_device *bdev; | |
09d967c6 | 937 | |
1da177e4 LT |
938 | spin_lock(&bdev_lock); |
939 | bdev = inode->i_bdev; | |
cccd9fb9 | 940 | if (bdev && !inode_unhashed(bdev->bd_inode)) { |
ed8a9d2c | 941 | bdgrab(bdev); |
1da177e4 LT |
942 | spin_unlock(&bdev_lock); |
943 | return bdev; | |
944 | } | |
945 | spin_unlock(&bdev_lock); | |
09d967c6 | 946 | |
cccd9fb9 JK |
947 | /* |
948 | * i_bdev references block device inode that was already shut down | |
949 | * (corresponding device got removed). Remove the reference and look | |
950 | * up block device inode again just in case new device got | |
951 | * reestablished under the same device number. | |
952 | */ | |
953 | if (bdev) | |
954 | bd_forget(inode); | |
955 | ||
1da177e4 LT |
956 | bdev = bdget(inode->i_rdev); |
957 | if (bdev) { | |
958 | spin_lock(&bdev_lock); | |
09d967c6 OH |
959 | if (!inode->i_bdev) { |
960 | /* | |
7de9c6ee | 961 | * We take an additional reference to bd_inode, |
09d967c6 OH |
962 | * and it's released in clear_inode() of inode. |
963 | * So, we can access it via ->i_mapping always | |
964 | * without igrab(). | |
965 | */ | |
ed8a9d2c | 966 | bdgrab(bdev); |
09d967c6 OH |
967 | inode->i_bdev = bdev; |
968 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
09d967c6 | 969 | } |
1da177e4 LT |
970 | spin_unlock(&bdev_lock); |
971 | } | |
972 | return bdev; | |
973 | } | |
974 | ||
975 | /* Call when you free inode */ | |
976 | ||
977 | void bd_forget(struct inode *inode) | |
978 | { | |
09d967c6 OH |
979 | struct block_device *bdev = NULL; |
980 | ||
1da177e4 | 981 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
982 | if (!sb_is_blkdev_sb(inode->i_sb)) |
983 | bdev = inode->i_bdev; | |
a4a4f943 AV |
984 | inode->i_bdev = NULL; |
985 | inode->i_mapping = &inode->i_data; | |
1da177e4 | 986 | spin_unlock(&bdev_lock); |
09d967c6 OH |
987 | |
988 | if (bdev) | |
ed8a9d2c | 989 | bdput(bdev); |
1da177e4 LT |
990 | } |
991 | ||
1a3cbbc5 TH |
992 | /** |
993 | * bd_may_claim - test whether a block device can be claimed | |
994 | * @bdev: block device of interest | |
995 | * @whole: whole block device containing @bdev, may equal @bdev | |
996 | * @holder: holder trying to claim @bdev | |
997 | * | |
25985edc | 998 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
999 | * |
1000 | * CONTEXT: | |
1001 | * spin_lock(&bdev_lock). | |
1002 | * | |
1003 | * RETURNS: | |
1004 | * %true if @bdev can be claimed, %false otherwise. | |
1005 | */ | |
1006 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
1007 | void *holder) | |
1da177e4 | 1008 | { |
1da177e4 | 1009 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 1010 | return true; /* already a holder */ |
1da177e4 | 1011 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 1012 | return false; /* held by someone else */ |
bcc7f5b4 | 1013 | else if (whole == bdev) |
1a3cbbc5 | 1014 | return true; /* is a whole device which isn't held */ |
1da177e4 | 1015 | |
e525fd89 | 1016 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
1017 | return true; /* is a partition of a device that is being partitioned */ |
1018 | else if (whole->bd_holder != NULL) | |
1019 | return false; /* is a partition of a held device */ | |
1da177e4 | 1020 | else |
1a3cbbc5 TH |
1021 | return true; /* is a partition of an un-held device */ |
1022 | } | |
1023 | ||
6b4517a7 | 1024 | /** |
58e46ed9 | 1025 | * bd_prepare_to_claim - claim a block device |
6b4517a7 TH |
1026 | * @bdev: block device of interest |
1027 | * @whole: the whole device containing @bdev, may equal @bdev | |
1028 | * @holder: holder trying to claim @bdev | |
1029 | * | |
58e46ed9 CH |
1030 | * Claim @bdev. This function fails if @bdev is already claimed by another |
1031 | * holder and waits if another claiming is in progress. return, the caller | |
1032 | * has ownership of bd_claiming and bd_holder[s]. | |
6b4517a7 TH |
1033 | * |
1034 | * RETURNS: | |
1035 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
1036 | */ | |
ecbe6bc0 CH |
1037 | int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole, |
1038 | void *holder) | |
6b4517a7 TH |
1039 | { |
1040 | retry: | |
58e46ed9 | 1041 | spin_lock(&bdev_lock); |
6b4517a7 | 1042 | /* if someone else claimed, fail */ |
58e46ed9 CH |
1043 | if (!bd_may_claim(bdev, whole, holder)) { |
1044 | spin_unlock(&bdev_lock); | |
6b4517a7 | 1045 | return -EBUSY; |
58e46ed9 | 1046 | } |
6b4517a7 | 1047 | |
e75aa858 TH |
1048 | /* if claiming is already in progress, wait for it to finish */ |
1049 | if (whole->bd_claiming) { | |
6b4517a7 TH |
1050 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
1051 | DEFINE_WAIT(wait); | |
1052 | ||
1053 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
1054 | spin_unlock(&bdev_lock); | |
1055 | schedule(); | |
1056 | finish_wait(wq, &wait); | |
6b4517a7 TH |
1057 | goto retry; |
1058 | } | |
1059 | ||
1060 | /* yay, all mine */ | |
58e46ed9 CH |
1061 | whole->bd_claiming = holder; |
1062 | spin_unlock(&bdev_lock); | |
6b4517a7 TH |
1063 | return 0; |
1064 | } | |
ecbe6bc0 | 1065 | EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */ |
6b4517a7 | 1066 | |
560e7cb2 JK |
1067 | static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno) |
1068 | { | |
1069 | struct gendisk *disk = get_gendisk(bdev->bd_dev, partno); | |
1070 | ||
1071 | if (!disk) | |
1072 | return NULL; | |
1073 | /* | |
1074 | * Now that we hold gendisk reference we make sure bdev we looked up is | |
1075 | * not stale. If it is, it means device got removed and created before | |
1076 | * we looked up gendisk and we fail open in such case. Associating | |
1077 | * unhashed bdev with newly created gendisk could lead to two bdevs | |
1078 | * (and thus two independent caches) being associated with one device | |
1079 | * which is bad. | |
1080 | */ | |
1081 | if (inode_unhashed(bdev->bd_inode)) { | |
1082 | put_disk_and_module(disk); | |
1083 | return NULL; | |
1084 | } | |
1085 | return disk; | |
1086 | } | |
1087 | ||
89e524c0 JK |
1088 | static void bd_clear_claiming(struct block_device *whole, void *holder) |
1089 | { | |
1090 | lockdep_assert_held(&bdev_lock); | |
1091 | /* tell others that we're done */ | |
1092 | BUG_ON(whole->bd_claiming != holder); | |
1093 | whole->bd_claiming = NULL; | |
1094 | wake_up_bit(&whole->bd_claiming, 0); | |
1095 | } | |
1096 | ||
1097 | /** | |
1098 | * bd_finish_claiming - finish claiming of a block device | |
1099 | * @bdev: block device of interest | |
5b642d8b | 1100 | * @whole: whole block device |
89e524c0 JK |
1101 | * @holder: holder that has claimed @bdev |
1102 | * | |
1103 | * Finish exclusive open of a block device. Mark the device as exlusively | |
1104 | * open by the holder and wake up all waiters for exclusive open to finish. | |
1105 | */ | |
764b23bd CH |
1106 | static void bd_finish_claiming(struct block_device *bdev, |
1107 | struct block_device *whole, void *holder) | |
89e524c0 JK |
1108 | { |
1109 | spin_lock(&bdev_lock); | |
1110 | BUG_ON(!bd_may_claim(bdev, whole, holder)); | |
1111 | /* | |
1112 | * Note that for a whole device bd_holders will be incremented twice, | |
1113 | * and bd_holder will be set to bd_may_claim before being set to holder | |
1114 | */ | |
1115 | whole->bd_holders++; | |
1116 | whole->bd_holder = bd_may_claim; | |
1117 | bdev->bd_holders++; | |
1118 | bdev->bd_holder = holder; | |
1119 | bd_clear_claiming(whole, holder); | |
1120 | spin_unlock(&bdev_lock); | |
1121 | } | |
89e524c0 JK |
1122 | |
1123 | /** | |
1124 | * bd_abort_claiming - abort claiming of a block device | |
1125 | * @bdev: block device of interest | |
5b642d8b | 1126 | * @whole: whole block device |
89e524c0 JK |
1127 | * @holder: holder that has claimed @bdev |
1128 | * | |
1129 | * Abort claiming of a block device when the exclusive open failed. This can be | |
1130 | * also used when exclusive open is not actually desired and we just needed | |
1131 | * to block other exclusive openers for a while. | |
1132 | */ | |
1133 | void bd_abort_claiming(struct block_device *bdev, struct block_device *whole, | |
1134 | void *holder) | |
1135 | { | |
1136 | spin_lock(&bdev_lock); | |
1137 | bd_clear_claiming(whole, holder); | |
1138 | spin_unlock(&bdev_lock); | |
1139 | } | |
1140 | EXPORT_SYMBOL(bd_abort_claiming); | |
6b4517a7 | 1141 | |
641dc636 | 1142 | #ifdef CONFIG_SYSFS |
49731baa TH |
1143 | struct bd_holder_disk { |
1144 | struct list_head list; | |
1145 | struct gendisk *disk; | |
1146 | int refcnt; | |
1147 | }; | |
1148 | ||
1149 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
1150 | struct gendisk *disk) | |
1151 | { | |
1152 | struct bd_holder_disk *holder; | |
1153 | ||
1154 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
1155 | if (holder->disk == disk) | |
1156 | return holder; | |
1157 | return NULL; | |
1158 | } | |
1159 | ||
4d7dd8fd | 1160 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 1161 | { |
4d7dd8fd | 1162 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
1163 | } |
1164 | ||
1165 | static void del_symlink(struct kobject *from, struct kobject *to) | |
1166 | { | |
641dc636 JN |
1167 | sysfs_remove_link(from, kobject_name(to)); |
1168 | } | |
1169 | ||
df6c0cd9 | 1170 | /** |
e09b457b TH |
1171 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
1172 | * @bdev: the claimed slave bdev | |
1173 | * @disk: the holding disk | |
df6c0cd9 | 1174 | * |
49731baa TH |
1175 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1176 | * | |
e09b457b | 1177 | * This functions creates the following sysfs symlinks. |
641dc636 | 1178 | * |
e09b457b TH |
1179 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1180 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 1181 | * |
e09b457b TH |
1182 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1183 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 1184 | * |
e09b457b TH |
1185 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1186 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 1187 | * |
e09b457b TH |
1188 | * The caller must have claimed @bdev before calling this function and |
1189 | * ensure that both @bdev and @disk are valid during the creation and | |
1190 | * lifetime of these symlinks. | |
641dc636 | 1191 | * |
e09b457b TH |
1192 | * CONTEXT: |
1193 | * Might sleep. | |
641dc636 | 1194 | * |
e09b457b TH |
1195 | * RETURNS: |
1196 | * 0 on success, -errno on failure. | |
641dc636 | 1197 | */ |
e09b457b | 1198 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 1199 | { |
49731baa | 1200 | struct bd_holder_disk *holder; |
e09b457b | 1201 | int ret = 0; |
641dc636 | 1202 | |
2e7b651d | 1203 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 1204 | |
49731baa | 1205 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 1206 | |
e09b457b TH |
1207 | /* FIXME: remove the following once add_disk() handles errors */ |
1208 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
1209 | goto out_unlock; | |
4e91672c | 1210 | |
49731baa TH |
1211 | holder = bd_find_holder_disk(bdev, disk); |
1212 | if (holder) { | |
1213 | holder->refcnt++; | |
e09b457b | 1214 | goto out_unlock; |
49731baa | 1215 | } |
641dc636 | 1216 | |
49731baa TH |
1217 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1218 | if (!holder) { | |
1219 | ret = -ENOMEM; | |
e09b457b TH |
1220 | goto out_unlock; |
1221 | } | |
641dc636 | 1222 | |
49731baa TH |
1223 | INIT_LIST_HEAD(&holder->list); |
1224 | holder->disk = disk; | |
1225 | holder->refcnt = 1; | |
1226 | ||
8d65269f | 1227 | ret = add_symlink(disk->slave_dir, bdev_kobj(bdev)); |
49731baa TH |
1228 | if (ret) |
1229 | goto out_free; | |
1230 | ||
1231 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
1232 | if (ret) | |
1233 | goto out_del; | |
e7407d16 TH |
1234 | /* |
1235 | * bdev could be deleted beneath us which would implicitly destroy | |
1236 | * the holder directory. Hold on to it. | |
1237 | */ | |
1238 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1239 | |
1240 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1241 | goto out_unlock; | |
1242 | ||
1243 | out_del: | |
8d65269f | 1244 | del_symlink(disk->slave_dir, bdev_kobj(bdev)); |
49731baa TH |
1245 | out_free: |
1246 | kfree(holder); | |
e09b457b | 1247 | out_unlock: |
b4cf1b72 | 1248 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1249 | return ret; |
641dc636 | 1250 | } |
e09b457b | 1251 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1252 | |
49731baa TH |
1253 | /** |
1254 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1255 | * @bdev: the calimed slave bdev | |
1256 | * @disk: the holding disk | |
1257 | * | |
1258 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1259 | * | |
1260 | * CONTEXT: | |
1261 | * Might sleep. | |
1262 | */ | |
1263 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1264 | { |
49731baa | 1265 | struct bd_holder_disk *holder; |
641dc636 | 1266 | |
49731baa | 1267 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1268 | |
49731baa TH |
1269 | holder = bd_find_holder_disk(bdev, disk); |
1270 | ||
1271 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
8d65269f | 1272 | del_symlink(disk->slave_dir, bdev_kobj(bdev)); |
49731baa TH |
1273 | del_symlink(bdev->bd_part->holder_dir, |
1274 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1275 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1276 | list_del_init(&holder->list); |
1277 | kfree(holder); | |
1278 | } | |
1279 | ||
1280 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1281 | } |
49731baa | 1282 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1283 | #endif |
1da177e4 | 1284 | |
c3279d14 | 1285 | /** |
57d1b536 | 1286 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1287 | * @disk: struct gendisk to check |
1288 | * @bdev: struct bdev to adjust. | |
5afb7835 | 1289 | * @verbose: if %true log a message about a size change if there is any |
c3279d14 AP |
1290 | * |
1291 | * This routine checks to see if the bdev size does not match the disk size | |
849cf559 | 1292 | * and adjusts it if it differs. When shrinking the bdev size, its all caches |
1293 | * are freed. | |
c3279d14 | 1294 | */ |
a1548b67 CH |
1295 | static void check_disk_size_change(struct gendisk *disk, |
1296 | struct block_device *bdev, bool verbose) | |
c3279d14 AP |
1297 | { |
1298 | loff_t disk_size, bdev_size; | |
1299 | ||
c2b4bb8c | 1300 | spin_lock(&bdev->bd_size_lock); |
c3279d14 AP |
1301 | disk_size = (loff_t)get_capacity(disk) << 9; |
1302 | bdev_size = i_size_read(bdev->bd_inode); | |
1303 | if (disk_size != bdev_size) { | |
5afb7835 CH |
1304 | if (verbose) { |
1305 | printk(KERN_INFO | |
1306 | "%s: detected capacity change from %lld to %lld\n", | |
1307 | disk->disk_name, bdev_size, disk_size); | |
1308 | } | |
c3279d14 | 1309 | i_size_write(bdev->bd_inode, disk_size); |
c2b4bb8c | 1310 | } |
c2b4bb8c | 1311 | spin_unlock(&bdev->bd_size_lock); |
c3279d14 | 1312 | } |
c3279d14 | 1313 | |
659e56ba CH |
1314 | /** |
1315 | * revalidate_disk_size - checks for disk size change and adjusts bdev size. | |
1316 | * @disk: struct gendisk to check | |
1317 | * @verbose: if %true log a message about a size change if there is any | |
1318 | * | |
1319 | * This routine checks to see if the bdev size does not match the disk size | |
1320 | * and adjusts it if it differs. When shrinking the bdev size, its all caches | |
1321 | * are freed. | |
1322 | */ | |
1323 | void revalidate_disk_size(struct gendisk *disk, bool verbose) | |
1324 | { | |
1325 | struct block_device *bdev; | |
1326 | ||
1327 | /* | |
1328 | * Hidden disks don't have associated bdev so there's no point in | |
1329 | * revalidating them. | |
1330 | */ | |
1331 | if (disk->flags & GENHD_FL_HIDDEN) | |
1332 | return; | |
1333 | ||
1334 | bdev = bdget_disk(disk, 0); | |
1335 | if (bdev) { | |
1336 | check_disk_size_change(disk, bdev, verbose); | |
1337 | bdput(bdev); | |
1338 | } | |
1339 | } | |
659e56ba | 1340 | |
611bee52 | 1341 | void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors) |
1da177e4 | 1342 | { |
c2b4bb8c | 1343 | spin_lock(&bdev->bd_size_lock); |
611bee52 | 1344 | i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); |
c2b4bb8c | 1345 | spin_unlock(&bdev->bd_size_lock); |
1da177e4 | 1346 | } |
611bee52 | 1347 | EXPORT_SYMBOL(bd_set_nr_sectors); |
1da177e4 | 1348 | |
4385bab1 | 1349 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1350 | |
142fe8f4 | 1351 | int bdev_disk_changed(struct block_device *bdev, bool invalidate) |
a1548b67 | 1352 | { |
142fe8f4 | 1353 | struct gendisk *disk = bdev->bd_disk; |
a1548b67 CH |
1354 | int ret; |
1355 | ||
f0b870df CH |
1356 | lockdep_assert_held(&bdev->bd_mutex); |
1357 | ||
38430f08 | 1358 | clear_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state); |
6540fbf6 | 1359 | |
a1548b67 | 1360 | rescan: |
d46430bf | 1361 | ret = blk_drop_partitions(bdev); |
a1548b67 CH |
1362 | if (ret) |
1363 | return ret; | |
1364 | ||
d981cb5b CH |
1365 | /* |
1366 | * Historically we only set the capacity to zero for devices that | |
1367 | * support partitions (independ of actually having partitions created). | |
1368 | * Doing that is rather inconsistent, but changing it broke legacy | |
1369 | * udisks polling for legacy ide-cdrom devices. Use the crude check | |
1370 | * below to get the sane behavior for most device while not breaking | |
1371 | * userspace for this particular setup. | |
1372 | */ | |
1373 | if (invalidate) { | |
1374 | if (disk_part_scan_enabled(disk) || | |
1375 | !(disk->flags & GENHD_FL_REMOVABLE)) | |
1376 | set_capacity(disk, 0); | |
1377 | } else { | |
1378 | if (disk->fops->revalidate_disk) | |
1379 | disk->fops->revalidate_disk(disk); | |
1380 | } | |
a1548b67 CH |
1381 | |
1382 | check_disk_size_change(disk, bdev, !invalidate); | |
a1548b67 | 1383 | |
142fe8f4 CH |
1384 | if (get_capacity(disk)) { |
1385 | ret = blk_add_partitions(disk, bdev); | |
1386 | if (ret == -EAGAIN) | |
1387 | goto rescan; | |
490547ca | 1388 | } else if (invalidate) { |
a1548b67 CH |
1389 | /* |
1390 | * Tell userspace that the media / partition table may have | |
1391 | * changed. | |
1392 | */ | |
1393 | kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); | |
a1548b67 CH |
1394 | } |
1395 | ||
a1548b67 CH |
1396 | return ret; |
1397 | } | |
f0b870df CH |
1398 | /* |
1399 | * Only exported for for loop and dasd for historic reasons. Don't use in new | |
1400 | * code! | |
1401 | */ | |
1402 | EXPORT_SYMBOL_GPL(bdev_disk_changed); | |
a1548b67 | 1403 | |
6d740cd5 PZ |
1404 | /* |
1405 | * bd_mutex locking: | |
1406 | * | |
1407 | * mutex_lock(part->bd_mutex) | |
1408 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1409 | */ | |
5b56b6ed CH |
1410 | static int __blkdev_get(struct block_device *bdev, struct gendisk *disk, |
1411 | int partno, fmode_t mode) | |
1da177e4 | 1412 | { |
7db9cfd3 | 1413 | int ret; |
5b642d8b | 1414 | |
1da177e4 LT |
1415 | if (!bdev->bd_openers) { |
1416 | bdev->bd_disk = disk; | |
1417 | bdev->bd_contains = bdev; | |
c2ee070f | 1418 | bdev->bd_partno = partno; |
03cdadb0 | 1419 | |
cf771cb5 | 1420 | if (!partno) { |
89f97496 TH |
1421 | ret = -ENXIO; |
1422 | bdev->bd_part = disk_get_part(disk, partno); | |
1423 | if (!bdev->bd_part) | |
1424 | goto out_clear; | |
1425 | ||
1196f8b8 | 1426 | ret = 0; |
5b56b6ed | 1427 | if (disk->fops->open) |
572c4892 | 1428 | ret = disk->fops->open(bdev, mode); |
7e69723f | 1429 | |
04906b2f | 1430 | if (!ret) { |
611bee52 | 1431 | bd_set_nr_sectors(bdev, get_capacity(disk)); |
04906b2f JK |
1432 | set_init_blocksize(bdev); |
1433 | } | |
7e69723f | 1434 | |
1196f8b8 TH |
1435 | /* |
1436 | * If the device is invalidated, rescan partition | |
1437 | * if open succeeded or failed with -ENOMEDIUM. | |
1438 | * The latter is necessary to prevent ghost | |
1439 | * partitions on a removed medium. | |
1440 | */ | |
38430f08 | 1441 | if (test_bit(GD_NEED_PART_SCAN, &disk->state) && |
731dc486 JK |
1442 | (!ret || ret == -ENOMEDIUM)) |
1443 | bdev_disk_changed(bdev, ret == -ENOMEDIUM); | |
5a023cdb | 1444 | |
1196f8b8 TH |
1445 | if (ret) |
1446 | goto out_clear; | |
1da177e4 | 1447 | } else { |
5b56b6ed CH |
1448 | struct block_device *whole = bdget_disk(disk, 0); |
1449 | ||
1450 | mutex_lock_nested(&whole->bd_mutex, 1); | |
1451 | ret = __blkdev_get(whole, disk, 0, mode); | |
1452 | if (ret) { | |
1453 | mutex_unlock(&whole->bd_mutex); | |
1454 | bdput(whole); | |
0762b8bd | 1455 | goto out_clear; |
5b56b6ed CH |
1456 | } |
1457 | whole->bd_part_count++; | |
1458 | mutex_unlock(&whole->bd_mutex); | |
1459 | ||
1460 | bdev->bd_contains = whole; | |
89f97496 | 1461 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1462 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1463 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
5b56b6ed | 1464 | __blkdev_put(whole, mode, 1); |
63d9932c | 1465 | bdput(whole); |
1da177e4 | 1466 | ret = -ENXIO; |
0762b8bd | 1467 | goto out_clear; |
1da177e4 | 1468 | } |
611bee52 | 1469 | bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects); |
04906b2f | 1470 | set_init_blocksize(bdev); |
1da177e4 | 1471 | } |
03e26279 JK |
1472 | |
1473 | if (bdev->bd_bdi == &noop_backing_dev_info) | |
1474 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info); | |
1da177e4 | 1475 | } else { |
1da177e4 | 1476 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1477 | ret = 0; |
1478 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1479 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1480 | /* the same as first opener case, read comment there */ |
38430f08 | 1481 | if (test_bit(GD_NEED_PART_SCAN, &disk->state) && |
731dc486 JK |
1482 | (!ret || ret == -ENOMEDIUM)) |
1483 | bdev_disk_changed(bdev, ret == -ENOMEDIUM); | |
1196f8b8 | 1484 | if (ret) |
5b56b6ed | 1485 | return ret; |
1da177e4 LT |
1486 | } |
1487 | } | |
1488 | bdev->bd_openers++; | |
1da177e4 LT |
1489 | return 0; |
1490 | ||
0762b8bd | 1491 | out_clear: |
89f97496 | 1492 | disk_put_part(bdev->bd_part); |
1da177e4 | 1493 | bdev->bd_disk = NULL; |
0762b8bd | 1494 | bdev->bd_part = NULL; |
1da177e4 | 1495 | bdev->bd_contains = NULL; |
1da177e4 LT |
1496 | return ret; |
1497 | } | |
1498 | ||
d4d77629 TH |
1499 | /** |
1500 | * blkdev_get - open a block device | |
1501 | * @bdev: block_device to open | |
1502 | * @mode: FMODE_* mask | |
1503 | * @holder: exclusive holder identifier | |
1504 | * | |
1505 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1506 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1507 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1508 | * | |
1509 | * On success, the reference count of @bdev is unchanged. On failure, | |
1510 | * @bdev is put. | |
1511 | * | |
1512 | * CONTEXT: | |
1513 | * Might sleep. | |
1514 | * | |
1515 | * RETURNS: | |
1516 | * 0 on success, -errno on failure. | |
1517 | */ | |
1fb1a2ad | 1518 | static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1519 | { |
5b56b6ed CH |
1520 | struct block_device *claiming; |
1521 | bool unblock_events = true; | |
1522 | struct gendisk *disk; | |
1523 | int perm = 0; | |
1524 | int partno; | |
1525 | int ret; | |
e525fd89 | 1526 | |
e5c7fb40 CH |
1527 | if (mode & FMODE_READ) |
1528 | perm |= MAY_READ; | |
1529 | if (mode & FMODE_WRITE) | |
1530 | perm |= MAY_WRITE; | |
1531 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1532 | if (ret) | |
1533 | goto bdput; | |
1534 | ||
5b56b6ed CH |
1535 | /* |
1536 | * If we lost a race with 'disk' being deleted, try again. See md.c. | |
1537 | */ | |
1538 | retry: | |
1539 | ret = -ENXIO; | |
1540 | disk = bdev_get_gendisk(bdev, &partno); | |
1541 | if (!disk) | |
e5c7fb40 | 1542 | goto bdput; |
e5c7fb40 | 1543 | |
5b56b6ed CH |
1544 | if (mode & FMODE_EXCL) { |
1545 | WARN_ON_ONCE(!holder); | |
1546 | ||
1547 | ret = -ENOMEM; | |
1548 | claiming = bdget_disk(disk, 0); | |
1549 | if (!claiming) | |
1550 | goto put_disk; | |
1551 | ret = bd_prepare_to_claim(bdev, claiming, holder); | |
1552 | if (ret) | |
1553 | goto put_claiming; | |
1554 | } | |
1555 | ||
1556 | disk_block_events(disk); | |
1557 | ||
1558 | mutex_lock(&bdev->bd_mutex); | |
1559 | ret =__blkdev_get(bdev, disk, partno, mode); | |
1560 | if (!(mode & FMODE_EXCL)) { | |
1561 | ; /* nothing to do here */ | |
1562 | } else if (ret) { | |
1563 | bd_abort_claiming(bdev, claiming, holder); | |
1564 | } else { | |
1565 | bd_finish_claiming(bdev, claiming, holder); | |
1566 | ||
1567 | /* | |
1568 | * Block event polling for write claims if requested. Any write | |
1569 | * holder makes the write_holder state stick until all are | |
1570 | * released. This is good enough and tracking individual | |
1571 | * writeable reference is too fragile given the way @mode is | |
1572 | * used in blkdev_get/put(). | |
1573 | */ | |
1574 | if ((mode & FMODE_WRITE) && !bdev->bd_write_holder && | |
1575 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
1576 | bdev->bd_write_holder = true; | |
1577 | unblock_events = false; | |
1578 | } | |
1579 | } | |
1580 | mutex_unlock(&bdev->bd_mutex); | |
1581 | ||
1582 | if (unblock_events) | |
1583 | disk_unblock_events(disk); | |
1584 | ||
1585 | put_claiming: | |
1586 | if (mode & FMODE_EXCL) | |
1587 | bdput(claiming); | |
1588 | put_disk: | |
1589 | if (ret) | |
1590 | put_disk_and_module(disk); | |
1591 | if (ret == -ERESTARTSYS) | |
1592 | goto retry; | |
e5c7fb40 | 1593 | bdput: |
5b56b6ed CH |
1594 | if (ret) |
1595 | bdput(bdev); | |
e5c7fb40 | 1596 | return ret; |
37be4124 | 1597 | } |
1da177e4 | 1598 | |
d4d77629 TH |
1599 | /** |
1600 | * blkdev_get_by_path - open a block device by name | |
1601 | * @path: path to the block device to open | |
1602 | * @mode: FMODE_* mask | |
1603 | * @holder: exclusive holder identifier | |
1604 | * | |
1605 | * Open the blockdevice described by the device file at @path. @mode | |
1606 | * and @holder are identical to blkdev_get(). | |
1607 | * | |
1608 | * On success, the returned block_device has reference count of one. | |
1609 | * | |
1610 | * CONTEXT: | |
1611 | * Might sleep. | |
1612 | * | |
1613 | * RETURNS: | |
1614 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1615 | */ | |
1616 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1617 | void *holder) | |
1618 | { | |
1619 | struct block_device *bdev; | |
1620 | int err; | |
1621 | ||
1622 | bdev = lookup_bdev(path); | |
1623 | if (IS_ERR(bdev)) | |
1624 | return bdev; | |
1625 | ||
1626 | err = blkdev_get(bdev, mode, holder); | |
1627 | if (err) | |
1628 | return ERR_PTR(err); | |
1629 | ||
e51900f7 CE |
1630 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1631 | blkdev_put(bdev, mode); | |
1632 | return ERR_PTR(-EACCES); | |
1633 | } | |
1634 | ||
d4d77629 TH |
1635 | return bdev; |
1636 | } | |
1637 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1638 | ||
1639 | /** | |
1640 | * blkdev_get_by_dev - open a block device by device number | |
1641 | * @dev: device number of block device to open | |
1642 | * @mode: FMODE_* mask | |
1643 | * @holder: exclusive holder identifier | |
1644 | * | |
1645 | * Open the blockdevice described by device number @dev. @mode and | |
1646 | * @holder are identical to blkdev_get(). | |
1647 | * | |
1648 | * Use it ONLY if you really do not have anything better - i.e. when | |
1649 | * you are behind a truly sucky interface and all you are given is a | |
1650 | * device number. _Never_ to be used for internal purposes. If you | |
1651 | * ever need it - reconsider your API. | |
1652 | * | |
1653 | * On success, the returned block_device has reference count of one. | |
1654 | * | |
1655 | * CONTEXT: | |
1656 | * Might sleep. | |
1657 | * | |
1658 | * RETURNS: | |
1659 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1660 | */ | |
1661 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1662 | { | |
1663 | struct block_device *bdev; | |
1664 | int err; | |
1665 | ||
1666 | bdev = bdget(dev); | |
1667 | if (!bdev) | |
1668 | return ERR_PTR(-ENOMEM); | |
1669 | ||
1670 | err = blkdev_get(bdev, mode, holder); | |
1671 | if (err) | |
1672 | return ERR_PTR(err); | |
1673 | ||
1674 | return bdev; | |
1675 | } | |
1676 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1677 | ||
1da177e4 LT |
1678 | static int blkdev_open(struct inode * inode, struct file * filp) |
1679 | { | |
1680 | struct block_device *bdev; | |
1da177e4 LT |
1681 | |
1682 | /* | |
1683 | * Preserve backwards compatibility and allow large file access | |
1684 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1685 | * binary needs it. We might want to drop this workaround | |
1686 | * during an unstable branch. | |
1687 | */ | |
1688 | filp->f_flags |= O_LARGEFILE; | |
1689 | ||
a304f074 | 1690 | filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC; |
c35fc7a5 | 1691 | |
572c4892 AV |
1692 | if (filp->f_flags & O_NDELAY) |
1693 | filp->f_mode |= FMODE_NDELAY; | |
1694 | if (filp->f_flags & O_EXCL) | |
1695 | filp->f_mode |= FMODE_EXCL; | |
1696 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1697 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1698 | ||
1da177e4 | 1699 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1700 | if (bdev == NULL) |
1701 | return -ENOMEM; | |
1da177e4 | 1702 | |
572c4892 | 1703 | filp->f_mapping = bdev->bd_inode->i_mapping; |
5660e13d | 1704 | filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping); |
572c4892 | 1705 | |
e525fd89 | 1706 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1707 | } |
1708 | ||
4385bab1 | 1709 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1710 | { |
2e7b651d | 1711 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1712 | struct block_device *victim = NULL; |
2e7b651d | 1713 | |
b849dd84 DA |
1714 | /* |
1715 | * Sync early if it looks like we're the last one. If someone else | |
1716 | * opens the block device between now and the decrement of bd_openers | |
1717 | * then we did a sync that we didn't need to, but that's not the end | |
1718 | * of the world and we want to avoid long (could be several minute) | |
1719 | * syncs while holding the mutex. | |
1720 | */ | |
1721 | if (bdev->bd_openers == 1) | |
1722 | sync_blockdev(bdev); | |
1723 | ||
6796bf54 | 1724 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1725 | if (for_part) |
1726 | bdev->bd_part_count--; | |
1727 | ||
2e7b651d | 1728 | if (!--bdev->bd_openers) { |
6a027eff | 1729 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1730 | sync_blockdev(bdev); |
1731 | kill_bdev(bdev); | |
43d1c0eb | 1732 | bdev_write_inode(bdev); |
ec5d4514 CH |
1733 | |
1734 | if (!bdev_is_partition(bdev) && disk->fops->release) | |
db2a144b | 1735 | disk->fops->release(disk, mode); |
ec5d4514 | 1736 | |
0762b8bd TH |
1737 | disk_put_part(bdev->bd_part); |
1738 | bdev->bd_part = NULL; | |
2e7b651d | 1739 | bdev->bd_disk = NULL; |
ec5d4514 | 1740 | if (bdev_is_partition(bdev)) |
37be4124 | 1741 | victim = bdev->bd_contains; |
2e7b651d | 1742 | bdev->bd_contains = NULL; |
ec5d4514 CH |
1743 | } else { |
1744 | if (!bdev_is_partition(bdev) && disk->fops->release) | |
1745 | disk->fops->release(disk, mode); | |
2e7b651d | 1746 | } |
2e7b651d | 1747 | mutex_unlock(&bdev->bd_mutex); |
63d9932c | 1748 | if (victim) { |
9a1c3542 | 1749 | __blkdev_put(victim, mode, 1); |
63d9932c CH |
1750 | bdput(victim); |
1751 | } | |
2e7b651d PZ |
1752 | } |
1753 | ||
4385bab1 | 1754 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1755 | { |
5b56b6ed CH |
1756 | struct gendisk *disk = bdev->bd_disk; |
1757 | ||
85ef06d1 TH |
1758 | mutex_lock(&bdev->bd_mutex); |
1759 | ||
e525fd89 | 1760 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1761 | bool bdev_free; |
1762 | ||
1763 | /* | |
1764 | * Release a claim on the device. The holder fields | |
1765 | * are protected with bdev_lock. bd_mutex is to | |
1766 | * synchronize disk_holder unlinking. | |
1767 | */ | |
6a027eff TH |
1768 | spin_lock(&bdev_lock); |
1769 | ||
1770 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1771 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1772 | ||
1773 | /* bd_contains might point to self, check in a separate step */ | |
1774 | if ((bdev_free = !bdev->bd_holders)) | |
1775 | bdev->bd_holder = NULL; | |
1776 | if (!bdev->bd_contains->bd_holders) | |
1777 | bdev->bd_contains->bd_holder = NULL; | |
1778 | ||
1779 | spin_unlock(&bdev_lock); | |
1780 | ||
77ea887e TH |
1781 | /* |
1782 | * If this was the last claim, remove holder link and | |
1783 | * unblock evpoll if it was a write holder. | |
1784 | */ | |
85ef06d1 | 1785 | if (bdev_free && bdev->bd_write_holder) { |
5b56b6ed | 1786 | disk_unblock_events(disk); |
85ef06d1 | 1787 | bdev->bd_write_holder = false; |
77ea887e | 1788 | } |
6936217c | 1789 | } |
77ea887e | 1790 | |
85ef06d1 TH |
1791 | /* |
1792 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1793 | * event. This is to ensure detection of media removal commanded | |
1794 | * from userland - e.g. eject(1). | |
1795 | */ | |
5b56b6ed | 1796 | disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE); |
85ef06d1 TH |
1797 | |
1798 | mutex_unlock(&bdev->bd_mutex); | |
1799 | ||
4385bab1 | 1800 | __blkdev_put(bdev, mode, 0); |
63d9932c | 1801 | bdput(bdev); |
5b56b6ed | 1802 | put_disk_and_module(disk); |
37be4124 | 1803 | } |
2e7b651d PZ |
1804 | EXPORT_SYMBOL(blkdev_put); |
1805 | ||
1da177e4 LT |
1806 | static int blkdev_close(struct inode * inode, struct file * filp) |
1807 | { | |
4ebb16ca | 1808 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1809 | blkdev_put(bdev, filp->f_mode); |
1810 | return 0; | |
1da177e4 LT |
1811 | } |
1812 | ||
bb93e3a5 | 1813 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1814 | { |
4ebb16ca | 1815 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1816 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1817 | |
1818 | /* | |
1819 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1820 | * to updated it before every ioctl. | |
1821 | */ | |
56b26add | 1822 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1823 | mode |= FMODE_NDELAY; |
1824 | else | |
1825 | mode &= ~FMODE_NDELAY; | |
1826 | ||
56b26add | 1827 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1828 | } |
1829 | ||
eef99380 CH |
1830 | /* |
1831 | * Write data to the block device. Only intended for the block device itself | |
1832 | * and the raw driver which basically is a fake block device. | |
1833 | * | |
1834 | * Does not take i_mutex for the write and thus is not for general purpose | |
1835 | * use. | |
1836 | */ | |
1456c0a8 | 1837 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1838 | { |
1839 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1840 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1841 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1842 | struct blk_plug plug; |
eef99380 | 1843 | ssize_t ret; |
5f380c7f | 1844 | |
7ec7b94a AV |
1845 | if (bdev_read_only(I_BDEV(bd_inode))) |
1846 | return -EPERM; | |
5f380c7f | 1847 | |
bb3247a3 | 1848 | if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev)) |
dc617f29 DW |
1849 | return -ETXTBSY; |
1850 | ||
7ec7b94a | 1851 | if (!iov_iter_count(from)) |
5f380c7f AV |
1852 | return 0; |
1853 | ||
7ec7b94a AV |
1854 | if (iocb->ki_pos >= size) |
1855 | return -ENOSPC; | |
1856 | ||
c35fc7a5 CH |
1857 | if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) |
1858 | return -EOPNOTSUPP; | |
1859 | ||
7ec7b94a | 1860 | iov_iter_truncate(from, size - iocb->ki_pos); |
eef99380 | 1861 | |
53362a05 | 1862 | blk_start_plug(&plug); |
1456c0a8 | 1863 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
1864 | if (ret > 0) |
1865 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 1866 | blk_finish_plug(&plug); |
eef99380 CH |
1867 | return ret; |
1868 | } | |
1456c0a8 | 1869 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1870 | |
b2de525f | 1871 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1872 | { |
1873 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1874 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 1875 | loff_t size = i_size_read(bd_inode); |
a886038b | 1876 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1877 | |
1878 | if (pos >= size) | |
1879 | return 0; | |
1880 | ||
1881 | size -= pos; | |
a886038b AV |
1882 | iov_iter_truncate(to, size); |
1883 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1884 | } |
b2de525f | 1885 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1886 | |
87d8fe1e TT |
1887 | /* |
1888 | * Try to release a page associated with block device when the system | |
1889 | * is under memory pressure. | |
1890 | */ | |
1891 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1892 | { | |
1893 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1894 | ||
1895 | if (super && super->s_op->bdev_try_to_free_page) | |
1896 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1897 | ||
1898 | return try_to_free_buffers(page); | |
1899 | } | |
1900 | ||
7f6d5b52 RZ |
1901 | static int blkdev_writepages(struct address_space *mapping, |
1902 | struct writeback_control *wbc) | |
1903 | { | |
7f6d5b52 RZ |
1904 | return generic_writepages(mapping, wbc); |
1905 | } | |
1906 | ||
4c54ac62 | 1907 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1908 | .readpage = blkdev_readpage, |
d4388340 | 1909 | .readahead = blkdev_readahead, |
1da177e4 | 1910 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1911 | .write_begin = blkdev_write_begin, |
1912 | .write_end = blkdev_write_end, | |
7f6d5b52 | 1913 | .writepages = blkdev_writepages, |
87d8fe1e | 1914 | .releasepage = blkdev_releasepage, |
1da177e4 | 1915 | .direct_IO = blkdev_direct_IO, |
88dbcbb3 | 1916 | .migratepage = buffer_migrate_page_norefs, |
b4597226 | 1917 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1918 | }; |
1919 | ||
25f4c414 DW |
1920 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
1921 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
1922 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) | |
1923 | ||
1924 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, | |
1925 | loff_t len) | |
1926 | { | |
1927 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
25f4c414 DW |
1928 | loff_t end = start + len - 1; |
1929 | loff_t isize; | |
1930 | int error; | |
1931 | ||
1932 | /* Fail if we don't recognize the flags. */ | |
1933 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) | |
1934 | return -EOPNOTSUPP; | |
1935 | ||
1936 | /* Don't go off the end of the device. */ | |
1937 | isize = i_size_read(bdev->bd_inode); | |
1938 | if (start >= isize) | |
1939 | return -EINVAL; | |
1940 | if (end >= isize) { | |
1941 | if (mode & FALLOC_FL_KEEP_SIZE) { | |
1942 | len = isize - start; | |
1943 | end = start + len - 1; | |
1944 | } else | |
1945 | return -EINVAL; | |
1946 | } | |
1947 | ||
1948 | /* | |
1949 | * Don't allow IO that isn't aligned to logical block size. | |
1950 | */ | |
1951 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) | |
1952 | return -EINVAL; | |
1953 | ||
1954 | /* Invalidate the page cache, including dirty pages. */ | |
384d87ef JK |
1955 | error = truncate_bdev_range(bdev, file->f_mode, start, end); |
1956 | if (error) | |
1957 | return error; | |
25f4c414 DW |
1958 | |
1959 | switch (mode) { | |
1960 | case FALLOC_FL_ZERO_RANGE: | |
1961 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: | |
1962 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, | |
ee472d83 | 1963 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP); |
25f4c414 DW |
1964 | break; |
1965 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: | |
34045129 CH |
1966 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
1967 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK); | |
25f4c414 DW |
1968 | break; |
1969 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: | |
25f4c414 DW |
1970 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
1971 | GFP_KERNEL, 0); | |
1972 | break; | |
1973 | default: | |
1974 | return -EOPNOTSUPP; | |
1975 | } | |
1976 | if (error) | |
1977 | return error; | |
1978 | ||
1979 | /* | |
1980 | * Invalidate again; if someone wandered in and dirtied a page, | |
1981 | * the caller will be given -EBUSY. The third argument is | |
1982 | * inclusive, so the rounding here is safe. | |
1983 | */ | |
384d87ef | 1984 | return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping, |
25f4c414 DW |
1985 | start >> PAGE_SHIFT, |
1986 | end >> PAGE_SHIFT); | |
1987 | } | |
1988 | ||
4b6f5d20 | 1989 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1990 | .open = blkdev_open, |
1991 | .release = blkdev_close, | |
1992 | .llseek = block_llseek, | |
a886038b | 1993 | .read_iter = blkdev_read_iter, |
1456c0a8 | 1994 | .write_iter = blkdev_write_iter, |
eae83ce1 | 1995 | .iopoll = blkdev_iopoll, |
acc93d30 | 1996 | .mmap = generic_file_mmap, |
b1dd3b28 | 1997 | .fsync = blkdev_fsync, |
bb93e3a5 | 1998 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1999 | #ifdef CONFIG_COMPAT |
2000 | .compat_ioctl = compat_blkdev_ioctl, | |
2001 | #endif | |
1e8b3332 | 2002 | .splice_read = generic_file_splice_read, |
8d020765 | 2003 | .splice_write = iter_file_splice_write, |
25f4c414 | 2004 | .fallocate = blkdev_fallocate, |
1da177e4 LT |
2005 | }; |
2006 | ||
1da177e4 LT |
2007 | /** |
2008 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 2009 | * @pathname: special file representing the block device |
1da177e4 | 2010 | * |
57d1b536 | 2011 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
2012 | * namespace if possible and return it. Return ERR_PTR(error) |
2013 | * otherwise. | |
2014 | */ | |
421748ec | 2015 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
2016 | { |
2017 | struct block_device *bdev; | |
2018 | struct inode *inode; | |
421748ec | 2019 | struct path path; |
1da177e4 LT |
2020 | int error; |
2021 | ||
421748ec | 2022 | if (!pathname || !*pathname) |
1da177e4 LT |
2023 | return ERR_PTR(-EINVAL); |
2024 | ||
421748ec | 2025 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
2026 | if (error) |
2027 | return ERR_PTR(error); | |
2028 | ||
bb668734 | 2029 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
2030 | error = -ENOTBLK; |
2031 | if (!S_ISBLK(inode->i_mode)) | |
2032 | goto fail; | |
2033 | error = -EACCES; | |
a2982cc9 | 2034 | if (!may_open_dev(&path)) |
1da177e4 LT |
2035 | goto fail; |
2036 | error = -ENOMEM; | |
2037 | bdev = bd_acquire(inode); | |
2038 | if (!bdev) | |
2039 | goto fail; | |
2040 | out: | |
421748ec | 2041 | path_put(&path); |
1da177e4 LT |
2042 | return bdev; |
2043 | fail: | |
2044 | bdev = ERR_PTR(error); | |
2045 | goto out; | |
2046 | } | |
d5686b44 | 2047 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 2048 | |
93b270f7 | 2049 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
2050 | { |
2051 | struct super_block *sb = get_super(bdev); | |
2052 | int res = 0; | |
2053 | ||
2054 | if (sb) { | |
2055 | /* | |
2056 | * no need to lock the super, get_super holds the | |
2057 | * read mutex so the filesystem cannot go away | |
2058 | * under us (->put_super runs with the write lock | |
2059 | * hold). | |
2060 | */ | |
2061 | shrink_dcache_sb(sb); | |
93b270f7 | 2062 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
2063 | drop_super(sb); |
2064 | } | |
f98393a6 | 2065 | invalidate_bdev(bdev); |
b71e8a4c DH |
2066 | return res; |
2067 | } | |
2068 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
2069 | |
2070 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
2071 | { | |
2072 | struct inode *inode, *old_inode = NULL; | |
2073 | ||
74278da9 | 2074 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2075 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
2076 | struct address_space *mapping = inode->i_mapping; | |
af309226 | 2077 | struct block_device *bdev; |
5c0d6b60 JK |
2078 | |
2079 | spin_lock(&inode->i_lock); | |
2080 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
2081 | mapping->nrpages == 0) { | |
2082 | spin_unlock(&inode->i_lock); | |
2083 | continue; | |
2084 | } | |
2085 | __iget(inode); | |
2086 | spin_unlock(&inode->i_lock); | |
74278da9 | 2087 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2088 | /* |
2089 | * We hold a reference to 'inode' so it couldn't have been | |
2090 | * removed from s_inodes list while we dropped the | |
74278da9 | 2091 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 2092 | * be holding the last reference and we cannot iput it under |
74278da9 | 2093 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
2094 | * later. |
2095 | */ | |
2096 | iput(old_inode); | |
2097 | old_inode = inode; | |
af309226 | 2098 | bdev = I_BDEV(inode); |
5c0d6b60 | 2099 | |
af309226 RV |
2100 | mutex_lock(&bdev->bd_mutex); |
2101 | if (bdev->bd_openers) | |
2102 | func(bdev, arg); | |
2103 | mutex_unlock(&bdev->bd_mutex); | |
5c0d6b60 | 2104 | |
74278da9 | 2105 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 2106 | } |
74278da9 | 2107 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2108 | iput(old_inode); |
2109 | } |