]>
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.. */ |
ff01bb48 | 78 | 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); |
1da177e4 | 87 | } |
ff01bb48 AV |
88 | EXPORT_SYMBOL(kill_bdev); |
89 | ||
90 | /* Invalidate clean unused buffers and pagecache. */ | |
91 | void invalidate_bdev(struct block_device *bdev) | |
92 | { | |
93 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
94 | ||
a5f6a6a9 AR |
95 | if (mapping->nrpages) { |
96 | invalidate_bh_lrus(); | |
97 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
98 | invalidate_mapping_pages(mapping, 0, -1); | |
99 | } | |
ff01bb48 AV |
100 | /* 99% of the time, we don't need to flush the cleancache on the bdev. |
101 | * But, for the strange corners, lets be cautious | |
102 | */ | |
3167760f | 103 | cleancache_invalidate_inode(mapping); |
ff01bb48 AV |
104 | } |
105 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 | 106 | |
04906b2f JK |
107 | static void set_init_blocksize(struct block_device *bdev) |
108 | { | |
109 | unsigned bsize = bdev_logical_block_size(bdev); | |
110 | loff_t size = i_size_read(bdev->bd_inode); | |
111 | ||
112 | while (bsize < PAGE_SIZE) { | |
113 | if (size & bsize) | |
114 | break; | |
115 | bsize <<= 1; | |
116 | } | |
117 | bdev->bd_block_size = bsize; | |
118 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
119 | } | |
120 | ||
1da177e4 LT |
121 | int set_blocksize(struct block_device *bdev, int size) |
122 | { | |
123 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 124 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
125 | return -EINVAL; |
126 | ||
127 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 128 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
129 | return -EINVAL; |
130 | ||
131 | /* Don't change the size if it is same as current */ | |
132 | if (bdev->bd_block_size != size) { | |
133 | sync_blockdev(bdev); | |
134 | bdev->bd_block_size = size; | |
135 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
136 | kill_bdev(bdev); | |
137 | } | |
138 | return 0; | |
139 | } | |
140 | ||
141 | EXPORT_SYMBOL(set_blocksize); | |
142 | ||
143 | int sb_set_blocksize(struct super_block *sb, int size) | |
144 | { | |
1da177e4 LT |
145 | if (set_blocksize(sb->s_bdev, size)) |
146 | return 0; | |
147 | /* If we get here, we know size is power of two | |
148 | * and it's value is between 512 and PAGE_SIZE */ | |
149 | sb->s_blocksize = size; | |
38885bd4 | 150 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
151 | return sb->s_blocksize; |
152 | } | |
153 | ||
154 | EXPORT_SYMBOL(sb_set_blocksize); | |
155 | ||
156 | int sb_min_blocksize(struct super_block *sb, int size) | |
157 | { | |
e1defc4f | 158 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
159 | if (size < minsize) |
160 | size = minsize; | |
161 | return sb_set_blocksize(sb, size); | |
162 | } | |
163 | ||
164 | EXPORT_SYMBOL(sb_min_blocksize); | |
165 | ||
166 | static int | |
167 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
168 | struct buffer_head *bh, int create) | |
169 | { | |
1da177e4 LT |
170 | bh->b_bdev = I_BDEV(inode); |
171 | bh->b_blocknr = iblock; | |
172 | set_buffer_mapped(bh); | |
173 | return 0; | |
174 | } | |
175 | ||
4ebb16ca DW |
176 | static struct inode *bdev_file_inode(struct file *file) |
177 | { | |
178 | return file->f_mapping->host; | |
179 | } | |
180 | ||
78250c02 JA |
181 | static unsigned int dio_bio_write_op(struct kiocb *iocb) |
182 | { | |
183 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; | |
184 | ||
185 | /* avoid the need for a I/O completion work item */ | |
186 | if (iocb->ki_flags & IOCB_DSYNC) | |
187 | op |= REQ_FUA; | |
188 | return op; | |
189 | } | |
190 | ||
189ce2b9 CH |
191 | #define DIO_INLINE_BIO_VECS 4 |
192 | ||
193 | static void blkdev_bio_end_io_simple(struct bio *bio) | |
194 | { | |
195 | struct task_struct *waiter = bio->bi_private; | |
196 | ||
197 | WRITE_ONCE(bio->bi_private, NULL); | |
0619317f | 198 | blk_wake_io_task(waiter); |
189ce2b9 CH |
199 | } |
200 | ||
201 | static ssize_t | |
202 | __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter, | |
203 | int nr_pages) | |
204 | { | |
205 | struct file *file = iocb->ki_filp; | |
206 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
9fec4a21 | 207 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs; |
189ce2b9 CH |
208 | loff_t pos = iocb->ki_pos; |
209 | bool should_dirty = false; | |
210 | struct bio bio; | |
211 | ssize_t ret; | |
212 | blk_qc_t qc; | |
189ce2b9 | 213 | |
9a794fb9 JA |
214 | if ((pos | iov_iter_alignment(iter)) & |
215 | (bdev_logical_block_size(bdev) - 1)) | |
189ce2b9 CH |
216 | return -EINVAL; |
217 | ||
72ecad22 JA |
218 | if (nr_pages <= DIO_INLINE_BIO_VECS) |
219 | vecs = inline_vecs; | |
220 | else { | |
6da2ec56 KC |
221 | vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec), |
222 | GFP_KERNEL); | |
72ecad22 JA |
223 | if (!vecs) |
224 | return -ENOMEM; | |
225 | } | |
226 | ||
3a83f467 | 227 | bio_init(&bio, vecs, nr_pages); |
74d46992 | 228 | bio_set_dev(&bio, bdev); |
4d1a4765 | 229 | bio.bi_iter.bi_sector = pos >> 9; |
45d06cf7 | 230 | bio.bi_write_hint = iocb->ki_hint; |
189ce2b9 CH |
231 | bio.bi_private = current; |
232 | bio.bi_end_io = blkdev_bio_end_io_simple; | |
074111ca | 233 | bio.bi_ioprio = iocb->ki_ioprio; |
189ce2b9 CH |
234 | |
235 | ret = bio_iov_iter_get_pages(&bio, iter); | |
236 | if (unlikely(ret)) | |
9362dd11 | 237 | goto out; |
189ce2b9 CH |
238 | ret = bio.bi_iter.bi_size; |
239 | ||
240 | if (iov_iter_rw(iter) == READ) { | |
78250c02 | 241 | bio.bi_opf = REQ_OP_READ; |
189ce2b9 CH |
242 | if (iter_is_iovec(iter)) |
243 | should_dirty = true; | |
244 | } else { | |
78250c02 | 245 | bio.bi_opf = dio_bio_write_op(iocb); |
189ce2b9 CH |
246 | task_io_account_write(ret); |
247 | } | |
d1e36282 | 248 | if (iocb->ki_flags & IOCB_HIPRI) |
0bbb280d | 249 | bio_set_polled(&bio, iocb); |
189ce2b9 CH |
250 | |
251 | qc = submit_bio(&bio); | |
252 | for (;;) { | |
1ac5cd49 | 253 | set_current_state(TASK_UNINTERRUPTIBLE); |
189ce2b9 CH |
254 | if (!READ_ONCE(bio.bi_private)) |
255 | break; | |
256 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
0a1b8b87 | 257 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
e6249cdd | 258 | blk_io_schedule(); |
189ce2b9 CH |
259 | } |
260 | __set_current_state(TASK_RUNNING); | |
261 | ||
9fec4a21 | 262 | bio_release_pages(&bio, should_dirty); |
4e4cbee9 | 263 | if (unlikely(bio.bi_status)) |
c6b1e36c | 264 | ret = blk_status_to_errno(bio.bi_status); |
9ae3b3f5 | 265 | |
9362dd11 MW |
266 | out: |
267 | if (vecs != inline_vecs) | |
268 | kfree(vecs); | |
269 | ||
9ae3b3f5 JA |
270 | bio_uninit(&bio); |
271 | ||
189ce2b9 CH |
272 | return ret; |
273 | } | |
274 | ||
542ff7bf CH |
275 | struct blkdev_dio { |
276 | union { | |
277 | struct kiocb *iocb; | |
278 | struct task_struct *waiter; | |
279 | }; | |
280 | size_t size; | |
281 | atomic_t ref; | |
282 | bool multi_bio : 1; | |
283 | bool should_dirty : 1; | |
284 | bool is_sync : 1; | |
285 | struct bio bio; | |
286 | }; | |
287 | ||
52190f8a | 288 | static struct bio_set blkdev_dio_pool; |
542ff7bf | 289 | |
eae83ce1 CH |
290 | static int blkdev_iopoll(struct kiocb *kiocb, bool wait) |
291 | { | |
292 | struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host); | |
293 | struct request_queue *q = bdev_get_queue(bdev); | |
294 | ||
295 | return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait); | |
296 | } | |
297 | ||
542ff7bf CH |
298 | static void blkdev_bio_end_io(struct bio *bio) |
299 | { | |
300 | struct blkdev_dio *dio = bio->bi_private; | |
301 | bool should_dirty = dio->should_dirty; | |
302 | ||
a89afe58 JY |
303 | if (bio->bi_status && !dio->bio.bi_status) |
304 | dio->bio.bi_status = bio->bi_status; | |
305 | ||
306 | if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) { | |
542ff7bf CH |
307 | if (!dio->is_sync) { |
308 | struct kiocb *iocb = dio->iocb; | |
4e4cbee9 | 309 | ssize_t ret; |
542ff7bf | 310 | |
4e4cbee9 | 311 | if (likely(!dio->bio.bi_status)) { |
542ff7bf CH |
312 | ret = dio->size; |
313 | iocb->ki_pos += ret; | |
4e4cbee9 CH |
314 | } else { |
315 | ret = blk_status_to_errno(dio->bio.bi_status); | |
542ff7bf CH |
316 | } |
317 | ||
318 | dio->iocb->ki_complete(iocb, ret, 0); | |
531724ab CH |
319 | if (dio->multi_bio) |
320 | bio_put(&dio->bio); | |
542ff7bf CH |
321 | } else { |
322 | struct task_struct *waiter = dio->waiter; | |
323 | ||
324 | WRITE_ONCE(dio->waiter, NULL); | |
0619317f | 325 | blk_wake_io_task(waiter); |
542ff7bf CH |
326 | } |
327 | } | |
328 | ||
329 | if (should_dirty) { | |
330 | bio_check_pages_dirty(bio); | |
331 | } else { | |
57dfe3ce | 332 | bio_release_pages(bio, false); |
542ff7bf CH |
333 | bio_put(bio); |
334 | } | |
335 | } | |
336 | ||
b2e895db | 337 | static ssize_t |
542ff7bf | 338 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages) |
b2e895db AM |
339 | { |
340 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 341 | struct inode *inode = bdev_file_inode(file); |
542ff7bf | 342 | struct block_device *bdev = I_BDEV(inode); |
64d656a1 | 343 | struct blk_plug plug; |
542ff7bf CH |
344 | struct blkdev_dio *dio; |
345 | struct bio *bio; | |
cb700eb3 | 346 | bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0; |
690e5325 | 347 | bool is_read = (iov_iter_rw(iter) == READ), is_sync; |
542ff7bf CH |
348 | loff_t pos = iocb->ki_pos; |
349 | blk_qc_t qc = BLK_QC_T_NONE; | |
7b6620d7 | 350 | int ret = 0; |
542ff7bf | 351 | |
9a794fb9 JA |
352 | if ((pos | iov_iter_alignment(iter)) & |
353 | (bdev_logical_block_size(bdev) - 1)) | |
542ff7bf CH |
354 | return -EINVAL; |
355 | ||
7b6620d7 | 356 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool); |
542ff7bf CH |
357 | |
358 | dio = container_of(bio, struct blkdev_dio, bio); | |
690e5325 | 359 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
531724ab | 360 | if (dio->is_sync) { |
542ff7bf | 361 | dio->waiter = current; |
531724ab CH |
362 | bio_get(bio); |
363 | } else { | |
542ff7bf | 364 | dio->iocb = iocb; |
531724ab | 365 | } |
542ff7bf CH |
366 | |
367 | dio->size = 0; | |
368 | dio->multi_bio = false; | |
00e23707 | 369 | dio->should_dirty = is_read && iter_is_iovec(iter); |
542ff7bf | 370 | |
cb700eb3 JA |
371 | /* |
372 | * Don't plug for HIPRI/polled IO, as those should go straight | |
373 | * to issue | |
374 | */ | |
375 | if (!is_poll) | |
376 | blk_start_plug(&plug); | |
377 | ||
542ff7bf | 378 | for (;;) { |
74d46992 | 379 | bio_set_dev(bio, bdev); |
4d1a4765 | 380 | bio->bi_iter.bi_sector = pos >> 9; |
45d06cf7 | 381 | bio->bi_write_hint = iocb->ki_hint; |
542ff7bf CH |
382 | bio->bi_private = dio; |
383 | bio->bi_end_io = blkdev_bio_end_io; | |
074111ca | 384 | bio->bi_ioprio = iocb->ki_ioprio; |
542ff7bf | 385 | |
e15c2ffa JA |
386 | ret = bio_iov_iter_get_pages(bio, iter); |
387 | if (unlikely(ret)) { | |
4e4cbee9 | 388 | bio->bi_status = BLK_STS_IOERR; |
542ff7bf CH |
389 | bio_endio(bio); |
390 | break; | |
391 | } | |
392 | ||
393 | if (is_read) { | |
394 | bio->bi_opf = REQ_OP_READ; | |
395 | if (dio->should_dirty) | |
396 | bio_set_pages_dirty(bio); | |
397 | } else { | |
398 | bio->bi_opf = dio_bio_write_op(iocb); | |
399 | task_io_account_write(bio->bi_iter.bi_size); | |
400 | } | |
401 | ||
7b6620d7 | 402 | dio->size += bio->bi_iter.bi_size; |
542ff7bf CH |
403 | pos += bio->bi_iter.bi_size; |
404 | ||
405 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); | |
406 | if (!nr_pages) { | |
eae83ce1 CH |
407 | bool polled = false; |
408 | ||
409 | if (iocb->ki_flags & IOCB_HIPRI) { | |
0bbb280d | 410 | bio_set_polled(bio, iocb); |
eae83ce1 CH |
411 | polled = true; |
412 | } | |
d34513d3 | 413 | |
542ff7bf | 414 | qc = submit_bio(bio); |
eae83ce1 CH |
415 | |
416 | if (polled) | |
417 | WRITE_ONCE(iocb->ki_cookie, qc); | |
542ff7bf CH |
418 | break; |
419 | } | |
420 | ||
421 | if (!dio->multi_bio) { | |
531724ab CH |
422 | /* |
423 | * AIO needs an extra reference to ensure the dio | |
424 | * structure which is embedded into the first bio | |
425 | * stays around. | |
426 | */ | |
427 | if (!is_sync) | |
428 | bio_get(bio); | |
542ff7bf CH |
429 | dio->multi_bio = true; |
430 | atomic_set(&dio->ref, 2); | |
431 | } else { | |
432 | atomic_inc(&dio->ref); | |
433 | } | |
434 | ||
7b6620d7 JA |
435 | submit_bio(bio); |
436 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
542ff7bf | 437 | } |
cb700eb3 JA |
438 | |
439 | if (!is_poll) | |
440 | blk_finish_plug(&plug); | |
542ff7bf | 441 | |
690e5325 | 442 | if (!is_sync) |
542ff7bf CH |
443 | return -EIOCBQUEUED; |
444 | ||
445 | for (;;) { | |
1ac5cd49 | 446 | set_current_state(TASK_UNINTERRUPTIBLE); |
542ff7bf CH |
447 | if (!READ_ONCE(dio->waiter)) |
448 | break; | |
449 | ||
450 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
0a1b8b87 | 451 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
e6249cdd | 452 | blk_io_schedule(); |
542ff7bf CH |
453 | } |
454 | __set_current_state(TASK_RUNNING); | |
455 | ||
36ffc6c1 | 456 | if (!ret) |
4e4cbee9 | 457 | ret = blk_status_to_errno(dio->bio.bi_status); |
e15c2ffa JA |
458 | if (likely(!ret)) |
459 | ret = dio->size; | |
542ff7bf CH |
460 | |
461 | bio_put(&dio->bio); | |
462 | return ret; | |
463 | } | |
464 | ||
465 | static ssize_t | |
466 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) | |
467 | { | |
189ce2b9 | 468 | int nr_pages; |
b2e895db | 469 | |
72ecad22 | 470 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
189ce2b9 CH |
471 | if (!nr_pages) |
472 | return 0; | |
72ecad22 | 473 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
189ce2b9 | 474 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
542ff7bf CH |
475 | |
476 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); | |
477 | } | |
478 | ||
479 | static __init int blkdev_init(void) | |
480 | { | |
52190f8a | 481 | return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS); |
b2e895db | 482 | } |
542ff7bf | 483 | module_init(blkdev_init); |
b2e895db | 484 | |
5cee5815 JK |
485 | int __sync_blockdev(struct block_device *bdev, int wait) |
486 | { | |
487 | if (!bdev) | |
488 | return 0; | |
489 | if (!wait) | |
490 | return filemap_flush(bdev->bd_inode->i_mapping); | |
491 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
492 | } | |
493 | ||
585d3bc0 NP |
494 | /* |
495 | * Write out and wait upon all the dirty data associated with a block | |
496 | * device via its mapping. Does not take the superblock lock. | |
497 | */ | |
498 | int sync_blockdev(struct block_device *bdev) | |
499 | { | |
5cee5815 | 500 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
501 | } |
502 | EXPORT_SYMBOL(sync_blockdev); | |
503 | ||
504 | /* | |
505 | * Write out and wait upon all dirty data associated with this | |
506 | * device. Filesystem data as well as the underlying block | |
507 | * device. Takes the superblock lock. | |
508 | */ | |
509 | int fsync_bdev(struct block_device *bdev) | |
510 | { | |
511 | struct super_block *sb = get_super(bdev); | |
512 | if (sb) { | |
60b0680f | 513 | int res = sync_filesystem(sb); |
585d3bc0 NP |
514 | drop_super(sb); |
515 | return res; | |
516 | } | |
517 | return sync_blockdev(bdev); | |
518 | } | |
47e4491b | 519 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
520 | |
521 | /** | |
522 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
523 | * @bdev: blockdevice to lock | |
524 | * | |
585d3bc0 NP |
525 | * If a superblock is found on this device, we take the s_umount semaphore |
526 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
527 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
528 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
529 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
530 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
531 | * actually. | |
532 | */ | |
533 | struct super_block *freeze_bdev(struct block_device *bdev) | |
534 | { | |
535 | struct super_block *sb; | |
536 | int error = 0; | |
537 | ||
538 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
539 | if (++bdev->bd_fsfreeze_count > 1) { |
540 | /* | |
541 | * We don't even need to grab a reference - the first call | |
542 | * to freeze_bdev grab an active reference and only the last | |
543 | * thaw_bdev drops it. | |
544 | */ | |
585d3bc0 | 545 | sb = get_super(bdev); |
5bb53c0f AR |
546 | if (sb) |
547 | drop_super(sb); | |
4504230a CH |
548 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
549 | return sb; | |
550 | } | |
551 | ||
552 | sb = get_active_super(bdev); | |
553 | if (!sb) | |
554 | goto out; | |
48b6bca6 BM |
555 | if (sb->s_op->freeze_super) |
556 | error = sb->s_op->freeze_super(sb); | |
557 | else | |
558 | error = freeze_super(sb); | |
18e9e510 JB |
559 | if (error) { |
560 | deactivate_super(sb); | |
561 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 562 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 563 | return ERR_PTR(error); |
585d3bc0 | 564 | } |
18e9e510 | 565 | deactivate_super(sb); |
4504230a | 566 | out: |
585d3bc0 NP |
567 | sync_blockdev(bdev); |
568 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 569 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
570 | } |
571 | EXPORT_SYMBOL(freeze_bdev); | |
572 | ||
573 | /** | |
574 | * thaw_bdev -- unlock filesystem | |
575 | * @bdev: blockdevice to unlock | |
576 | * @sb: associated superblock | |
577 | * | |
578 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
579 | */ | |
580 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
581 | { | |
4504230a | 582 | int error = -EINVAL; |
585d3bc0 NP |
583 | |
584 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 585 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 586 | goto out; |
4504230a CH |
587 | |
588 | error = 0; | |
589 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 590 | goto out; |
4504230a CH |
591 | |
592 | if (!sb) | |
18e9e510 | 593 | goto out; |
4504230a | 594 | |
48b6bca6 BM |
595 | if (sb->s_op->thaw_super) |
596 | error = sb->s_op->thaw_super(sb); | |
597 | else | |
598 | error = thaw_super(sb); | |
997198ba | 599 | if (error) |
18e9e510 | 600 | bdev->bd_fsfreeze_count++; |
18e9e510 | 601 | out: |
585d3bc0 | 602 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
997198ba | 603 | return error; |
585d3bc0 NP |
604 | } |
605 | EXPORT_SYMBOL(thaw_bdev); | |
606 | ||
1da177e4 LT |
607 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
608 | { | |
609 | return block_write_full_page(page, blkdev_get_block, wbc); | |
610 | } | |
611 | ||
612 | static int blkdev_readpage(struct file * file, struct page * page) | |
613 | { | |
614 | return block_read_full_page(page, blkdev_get_block); | |
615 | } | |
616 | ||
447f05bb AM |
617 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
618 | struct list_head *pages, unsigned nr_pages) | |
619 | { | |
620 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
621 | } | |
622 | ||
6272b5a5 NP |
623 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
624 | loff_t pos, unsigned len, unsigned flags, | |
625 | struct page **pagep, void **fsdata) | |
1da177e4 | 626 | { |
155130a4 CH |
627 | return block_write_begin(mapping, pos, len, flags, pagep, |
628 | blkdev_get_block); | |
1da177e4 LT |
629 | } |
630 | ||
6272b5a5 NP |
631 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
632 | loff_t pos, unsigned len, unsigned copied, | |
633 | struct page *page, void *fsdata) | |
1da177e4 | 634 | { |
6272b5a5 NP |
635 | int ret; |
636 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
637 | ||
638 | unlock_page(page); | |
09cbfeaf | 639 | put_page(page); |
6272b5a5 NP |
640 | |
641 | return ret; | |
1da177e4 LT |
642 | } |
643 | ||
644 | /* | |
645 | * private llseek: | |
496ad9aa | 646 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
647 | * so we compute the size by hand (just as in block_read/write above) |
648 | */ | |
965c8e59 | 649 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 650 | { |
4ebb16ca | 651 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
652 | loff_t retval; |
653 | ||
5955102c | 654 | inode_lock(bd_inode); |
5d48f3a2 | 655 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 656 | inode_unlock(bd_inode); |
1da177e4 LT |
657 | return retval; |
658 | } | |
659 | ||
02c24a82 | 660 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 661 | { |
4ebb16ca | 662 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 663 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 664 | int error; |
da5aa861 | 665 | |
372cf243 | 666 | error = file_write_and_wait_range(filp, start, end); |
da5aa861 RW |
667 | if (error) |
668 | return error; | |
ab0a9735 | 669 | |
b8af67e2 AB |
670 | /* |
671 | * There is no need to serialise calls to blkdev_issue_flush with | |
672 | * i_mutex and doing so causes performance issues with concurrent | |
673 | * O_SYNC writers to a block device. | |
674 | */ | |
dd3932ed | 675 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
676 | if (error == -EOPNOTSUPP) |
677 | error = 0; | |
b8af67e2 | 678 | |
ab0a9735 | 679 | return error; |
1da177e4 | 680 | } |
b1dd3b28 | 681 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 682 | |
47a191fd MW |
683 | /** |
684 | * bdev_read_page() - Start reading a page from a block device | |
685 | * @bdev: The device to read the page from | |
686 | * @sector: The offset on the device to read the page to (need not be aligned) | |
687 | * @page: The page to read | |
688 | * | |
689 | * On entry, the page should be locked. It will be unlocked when the page | |
690 | * has been read. If the block driver implements rw_page synchronously, | |
691 | * that will be true on exit from this function, but it need not be. | |
692 | * | |
693 | * Errors returned by this function are usually "soft", eg out of memory, or | |
694 | * queue full; callers should try a different route to read this page rather | |
695 | * than propagate an error back up the stack. | |
696 | * | |
697 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
698 | */ | |
699 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
700 | struct page *page) | |
701 | { | |
702 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
703 | int result = -EOPNOTSUPP; |
704 | ||
f68eb1e7 | 705 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
706 | return result; |
707 | ||
3a0a5299 | 708 | result = blk_queue_enter(bdev->bd_queue, 0); |
2e6edc95 DW |
709 | if (result) |
710 | return result; | |
3f289dcb TH |
711 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
712 | REQ_OP_READ); | |
2e6edc95 DW |
713 | blk_queue_exit(bdev->bd_queue); |
714 | return result; | |
47a191fd | 715 | } |
47a191fd MW |
716 | |
717 | /** | |
718 | * bdev_write_page() - Start writing a page to a block device | |
719 | * @bdev: The device to write the page to | |
720 | * @sector: The offset on the device to write the page to (need not be aligned) | |
721 | * @page: The page to write | |
722 | * @wbc: The writeback_control for the write | |
723 | * | |
724 | * On entry, the page should be locked and not currently under writeback. | |
725 | * On exit, if the write started successfully, the page will be unlocked and | |
726 | * under writeback. If the write failed already (eg the driver failed to | |
727 | * queue the page to the device), the page will still be locked. If the | |
728 | * caller is a ->writepage implementation, it will need to unlock the page. | |
729 | * | |
730 | * Errors returned by this function are usually "soft", eg out of memory, or | |
731 | * queue full; callers should try a different route to write this page rather | |
732 | * than propagate an error back up the stack. | |
733 | * | |
734 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
735 | */ | |
736 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
737 | struct page *page, struct writeback_control *wbc) | |
738 | { | |
739 | int result; | |
47a191fd | 740 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
2e6edc95 | 741 | |
f68eb1e7 | 742 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 743 | return -EOPNOTSUPP; |
3a0a5299 | 744 | result = blk_queue_enter(bdev->bd_queue, 0); |
2e6edc95 DW |
745 | if (result) |
746 | return result; | |
747 | ||
47a191fd | 748 | set_page_writeback(page); |
3f289dcb TH |
749 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
750 | REQ_OP_WRITE); | |
f892760a | 751 | if (result) { |
47a191fd | 752 | end_page_writeback(page); |
f892760a MW |
753 | } else { |
754 | clean_page_buffers(page); | |
47a191fd | 755 | unlock_page(page); |
f892760a | 756 | } |
2e6edc95 | 757 | blk_queue_exit(bdev->bd_queue); |
47a191fd MW |
758 | return result; |
759 | } | |
47a191fd | 760 | |
1da177e4 LT |
761 | /* |
762 | * pseudo-fs | |
763 | */ | |
764 | ||
765 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 766 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
767 | |
768 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
769 | { | |
e94b1766 | 770 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
771 | if (!ei) |
772 | return NULL; | |
773 | return &ei->vfs_inode; | |
774 | } | |
775 | ||
41149cb0 | 776 | static void bdev_free_inode(struct inode *inode) |
1da177e4 | 777 | { |
41149cb0 | 778 | kmem_cache_free(bdev_cachep, BDEV_I(inode)); |
fa0d7e3d NP |
779 | } |
780 | ||
51cc5068 | 781 | static void init_once(void *foo) |
1da177e4 LT |
782 | { |
783 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
784 | struct block_device *bdev = &ei->bdev; | |
785 | ||
a35afb83 CL |
786 | memset(bdev, 0, sizeof(*bdev)); |
787 | mutex_init(&bdev->bd_mutex); | |
a35afb83 | 788 | INIT_LIST_HEAD(&bdev->bd_list); |
49731baa TH |
789 | #ifdef CONFIG_SYSFS |
790 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
791 | #endif | |
a5a79d00 | 792 | bdev->bd_bdi = &noop_backing_dev_info; |
a35afb83 | 793 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
794 | /* Initialize mutex for freeze. */ |
795 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
796 | } |
797 | ||
b57922d9 | 798 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
799 | { |
800 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
91b0abe3 | 801 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 802 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 803 | clear_inode(inode); |
1da177e4 | 804 | spin_lock(&bdev_lock); |
1da177e4 LT |
805 | list_del_init(&bdev->bd_list); |
806 | spin_unlock(&bdev_lock); | |
f759741d JK |
807 | /* Detach inode from wb early as bdi_put() may free bdi->wb */ |
808 | inode_detach_wb(inode); | |
a5a79d00 | 809 | if (bdev->bd_bdi != &noop_backing_dev_info) { |
b1d2dc56 | 810 | bdi_put(bdev->bd_bdi); |
a5a79d00 JK |
811 | bdev->bd_bdi = &noop_backing_dev_info; |
812 | } | |
1da177e4 LT |
813 | } |
814 | ||
ee9b6d61 | 815 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
816 | .statfs = simple_statfs, |
817 | .alloc_inode = bdev_alloc_inode, | |
41149cb0 | 818 | .free_inode = bdev_free_inode, |
1da177e4 | 819 | .drop_inode = generic_delete_inode, |
b57922d9 | 820 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
821 | }; |
822 | ||
9030d16e | 823 | static int bd_init_fs_context(struct fs_context *fc) |
1da177e4 | 824 | { |
9030d16e DH |
825 | struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC); |
826 | if (!ctx) | |
827 | return -ENOMEM; | |
828 | fc->s_iflags |= SB_I_CGROUPWB; | |
829 | ctx->ops = &bdev_sops; | |
830 | return 0; | |
1da177e4 LT |
831 | } |
832 | ||
833 | static struct file_system_type bd_type = { | |
834 | .name = "bdev", | |
9030d16e | 835 | .init_fs_context = bd_init_fs_context, |
1da177e4 LT |
836 | .kill_sb = kill_anon_super, |
837 | }; | |
838 | ||
a212b105 TH |
839 | struct super_block *blockdev_superblock __read_mostly; |
840 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
841 | |
842 | void __init bdev_cache_init(void) | |
843 | { | |
844 | int err; | |
ace8577a | 845 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 846 | |
1da177e4 | 847 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 848 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 849 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 850 | init_once); |
1da177e4 LT |
851 | err = register_filesystem(&bd_type); |
852 | if (err) | |
853 | panic("Cannot register bdev pseudo-fs"); | |
854 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
855 | if (IS_ERR(bd_mnt)) |
856 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 857 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
858 | } |
859 | ||
860 | /* | |
861 | * Most likely _very_ bad one - but then it's hardly critical for small | |
862 | * /dev and can be fixed when somebody will need really large one. | |
863 | * Keep in mind that it will be fed through icache hash function too. | |
864 | */ | |
865 | static inline unsigned long hash(dev_t dev) | |
866 | { | |
867 | return MAJOR(dev)+MINOR(dev); | |
868 | } | |
869 | ||
870 | static int bdev_test(struct inode *inode, void *data) | |
871 | { | |
872 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
873 | } | |
874 | ||
875 | static int bdev_set(struct inode *inode, void *data) | |
876 | { | |
877 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
878 | return 0; | |
879 | } | |
880 | ||
881 | static LIST_HEAD(all_bdevs); | |
882 | ||
883 | struct block_device *bdget(dev_t dev) | |
884 | { | |
885 | struct block_device *bdev; | |
886 | struct inode *inode; | |
887 | ||
c2acf7b9 | 888 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
889 | bdev_test, bdev_set, &dev); |
890 | ||
891 | if (!inode) | |
892 | return NULL; | |
893 | ||
894 | bdev = &BDEV_I(inode)->bdev; | |
895 | ||
896 | if (inode->i_state & I_NEW) { | |
897 | bdev->bd_contains = NULL; | |
782b94cd | 898 | bdev->bd_super = NULL; |
1da177e4 | 899 | bdev->bd_inode = inode; |
93407472 | 900 | bdev->bd_block_size = i_blocksize(inode); |
1da177e4 LT |
901 | bdev->bd_part_count = 0; |
902 | bdev->bd_invalidated = 0; | |
903 | inode->i_mode = S_IFBLK; | |
904 | inode->i_rdev = dev; | |
905 | inode->i_bdev = bdev; | |
906 | inode->i_data.a_ops = &def_blk_aops; | |
907 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
908 | spin_lock(&bdev_lock); |
909 | list_add(&bdev->bd_list, &all_bdevs); | |
910 | spin_unlock(&bdev_lock); | |
911 | unlock_new_inode(inode); | |
912 | } | |
913 | return bdev; | |
914 | } | |
915 | ||
916 | EXPORT_SYMBOL(bdget); | |
917 | ||
dddac6a7 AJ |
918 | /** |
919 | * bdgrab -- Grab a reference to an already referenced block device | |
920 | * @bdev: Block device to grab a reference to. | |
921 | */ | |
922 | struct block_device *bdgrab(struct block_device *bdev) | |
923 | { | |
7de9c6ee | 924 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
925 | return bdev; |
926 | } | |
c1681bf8 | 927 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 928 | |
1da177e4 LT |
929 | long nr_blockdev_pages(void) |
930 | { | |
203a2935 | 931 | struct block_device *bdev; |
1da177e4 LT |
932 | long ret = 0; |
933 | spin_lock(&bdev_lock); | |
203a2935 | 934 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
935 | ret += bdev->bd_inode->i_mapping->nrpages; |
936 | } | |
937 | spin_unlock(&bdev_lock); | |
938 | return ret; | |
939 | } | |
940 | ||
941 | void bdput(struct block_device *bdev) | |
942 | { | |
943 | iput(bdev->bd_inode); | |
944 | } | |
945 | ||
946 | EXPORT_SYMBOL(bdput); | |
947 | ||
948 | static struct block_device *bd_acquire(struct inode *inode) | |
949 | { | |
950 | struct block_device *bdev; | |
09d967c6 | 951 | |
1da177e4 LT |
952 | spin_lock(&bdev_lock); |
953 | bdev = inode->i_bdev; | |
cccd9fb9 | 954 | if (bdev && !inode_unhashed(bdev->bd_inode)) { |
ed8a9d2c | 955 | bdgrab(bdev); |
1da177e4 LT |
956 | spin_unlock(&bdev_lock); |
957 | return bdev; | |
958 | } | |
959 | spin_unlock(&bdev_lock); | |
09d967c6 | 960 | |
cccd9fb9 JK |
961 | /* |
962 | * i_bdev references block device inode that was already shut down | |
963 | * (corresponding device got removed). Remove the reference and look | |
964 | * up block device inode again just in case new device got | |
965 | * reestablished under the same device number. | |
966 | */ | |
967 | if (bdev) | |
968 | bd_forget(inode); | |
969 | ||
1da177e4 LT |
970 | bdev = bdget(inode->i_rdev); |
971 | if (bdev) { | |
972 | spin_lock(&bdev_lock); | |
09d967c6 OH |
973 | if (!inode->i_bdev) { |
974 | /* | |
7de9c6ee | 975 | * We take an additional reference to bd_inode, |
09d967c6 OH |
976 | * and it's released in clear_inode() of inode. |
977 | * So, we can access it via ->i_mapping always | |
978 | * without igrab(). | |
979 | */ | |
ed8a9d2c | 980 | bdgrab(bdev); |
09d967c6 OH |
981 | inode->i_bdev = bdev; |
982 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
09d967c6 | 983 | } |
1da177e4 LT |
984 | spin_unlock(&bdev_lock); |
985 | } | |
986 | return bdev; | |
987 | } | |
988 | ||
989 | /* Call when you free inode */ | |
990 | ||
991 | void bd_forget(struct inode *inode) | |
992 | { | |
09d967c6 OH |
993 | struct block_device *bdev = NULL; |
994 | ||
1da177e4 | 995 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
996 | if (!sb_is_blkdev_sb(inode->i_sb)) |
997 | bdev = inode->i_bdev; | |
a4a4f943 AV |
998 | inode->i_bdev = NULL; |
999 | inode->i_mapping = &inode->i_data; | |
1da177e4 | 1000 | spin_unlock(&bdev_lock); |
09d967c6 OH |
1001 | |
1002 | if (bdev) | |
ed8a9d2c | 1003 | bdput(bdev); |
1da177e4 LT |
1004 | } |
1005 | ||
1a3cbbc5 TH |
1006 | /** |
1007 | * bd_may_claim - test whether a block device can be claimed | |
1008 | * @bdev: block device of interest | |
1009 | * @whole: whole block device containing @bdev, may equal @bdev | |
1010 | * @holder: holder trying to claim @bdev | |
1011 | * | |
25985edc | 1012 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
1013 | * |
1014 | * CONTEXT: | |
1015 | * spin_lock(&bdev_lock). | |
1016 | * | |
1017 | * RETURNS: | |
1018 | * %true if @bdev can be claimed, %false otherwise. | |
1019 | */ | |
1020 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
1021 | void *holder) | |
1da177e4 | 1022 | { |
1da177e4 | 1023 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 1024 | return true; /* already a holder */ |
1da177e4 | 1025 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 1026 | return false; /* held by someone else */ |
bcc7f5b4 | 1027 | else if (whole == bdev) |
1a3cbbc5 | 1028 | return true; /* is a whole device which isn't held */ |
1da177e4 | 1029 | |
e525fd89 | 1030 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
1031 | return true; /* is a partition of a device that is being partitioned */ |
1032 | else if (whole->bd_holder != NULL) | |
1033 | return false; /* is a partition of a held device */ | |
1da177e4 | 1034 | else |
1a3cbbc5 TH |
1035 | return true; /* is a partition of an un-held device */ |
1036 | } | |
1037 | ||
6b4517a7 TH |
1038 | /** |
1039 | * bd_prepare_to_claim - prepare to claim a block device | |
1040 | * @bdev: block device of interest | |
1041 | * @whole: the whole device containing @bdev, may equal @bdev | |
1042 | * @holder: holder trying to claim @bdev | |
1043 | * | |
1044 | * Prepare to claim @bdev. This function fails if @bdev is already | |
1045 | * claimed by another holder and waits if another claiming is in | |
1046 | * progress. This function doesn't actually claim. On successful | |
1047 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
1048 | * | |
1049 | * CONTEXT: | |
1050 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
1051 | * it multiple times. | |
1052 | * | |
1053 | * RETURNS: | |
1054 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
1055 | */ | |
1056 | static int bd_prepare_to_claim(struct block_device *bdev, | |
1057 | struct block_device *whole, void *holder) | |
1058 | { | |
1059 | retry: | |
1060 | /* if someone else claimed, fail */ | |
1061 | if (!bd_may_claim(bdev, whole, holder)) | |
1062 | return -EBUSY; | |
1063 | ||
e75aa858 TH |
1064 | /* if claiming is already in progress, wait for it to finish */ |
1065 | if (whole->bd_claiming) { | |
6b4517a7 TH |
1066 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
1067 | DEFINE_WAIT(wait); | |
1068 | ||
1069 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
1070 | spin_unlock(&bdev_lock); | |
1071 | schedule(); | |
1072 | finish_wait(wq, &wait); | |
1073 | spin_lock(&bdev_lock); | |
1074 | goto retry; | |
1075 | } | |
1076 | ||
1077 | /* yay, all mine */ | |
1078 | return 0; | |
1079 | } | |
1080 | ||
560e7cb2 JK |
1081 | static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno) |
1082 | { | |
1083 | struct gendisk *disk = get_gendisk(bdev->bd_dev, partno); | |
1084 | ||
1085 | if (!disk) | |
1086 | return NULL; | |
1087 | /* | |
1088 | * Now that we hold gendisk reference we make sure bdev we looked up is | |
1089 | * not stale. If it is, it means device got removed and created before | |
1090 | * we looked up gendisk and we fail open in such case. Associating | |
1091 | * unhashed bdev with newly created gendisk could lead to two bdevs | |
1092 | * (and thus two independent caches) being associated with one device | |
1093 | * which is bad. | |
1094 | */ | |
1095 | if (inode_unhashed(bdev->bd_inode)) { | |
1096 | put_disk_and_module(disk); | |
1097 | return NULL; | |
1098 | } | |
1099 | return disk; | |
1100 | } | |
1101 | ||
6b4517a7 TH |
1102 | /** |
1103 | * bd_start_claiming - start claiming a block device | |
1104 | * @bdev: block device of interest | |
1105 | * @holder: holder trying to claim @bdev | |
1106 | * | |
1107 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
1108 | * exclusively and mark that an exclusive open is in progress. Each | |
1109 | * successful call to this function must be matched with a call to | |
b0018361 NP |
1110 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
1111 | * fail). | |
1112 | * | |
1113 | * This function is used to gain exclusive access to the block device | |
1114 | * without actually causing other exclusive open attempts to fail. It | |
1115 | * should be used when the open sequence itself requires exclusive | |
1116 | * access but may subsequently fail. | |
6b4517a7 TH |
1117 | * |
1118 | * CONTEXT: | |
1119 | * Might sleep. | |
1120 | * | |
1121 | * RETURNS: | |
1122 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
1123 | * value on failure. | |
1124 | */ | |
89e524c0 | 1125 | struct block_device *bd_start_claiming(struct block_device *bdev, void *holder) |
6b4517a7 TH |
1126 | { |
1127 | struct gendisk *disk; | |
1128 | struct block_device *whole; | |
1129 | int partno, err; | |
1130 | ||
1131 | might_sleep(); | |
1132 | ||
1133 | /* | |
1134 | * @bdev might not have been initialized properly yet, look up | |
1135 | * and grab the outer block device the hard way. | |
1136 | */ | |
560e7cb2 | 1137 | disk = bdev_get_gendisk(bdev, &partno); |
6b4517a7 TH |
1138 | if (!disk) |
1139 | return ERR_PTR(-ENXIO); | |
1140 | ||
d4c208b8 TH |
1141 | /* |
1142 | * Normally, @bdev should equal what's returned from bdget_disk() | |
1143 | * if partno is 0; however, some drivers (floppy) use multiple | |
1144 | * bdev's for the same physical device and @bdev may be one of the | |
1145 | * aliases. Keep @bdev if partno is 0. This means claimer | |
1146 | * tracking is broken for those devices but it has always been that | |
1147 | * way. | |
1148 | */ | |
1149 | if (partno) | |
1150 | whole = bdget_disk(disk, 0); | |
1151 | else | |
1152 | whole = bdgrab(bdev); | |
1153 | ||
9df6c299 | 1154 | put_disk_and_module(disk); |
6b4517a7 TH |
1155 | if (!whole) |
1156 | return ERR_PTR(-ENOMEM); | |
1157 | ||
1158 | /* prepare to claim, if successful, mark claiming in progress */ | |
1159 | spin_lock(&bdev_lock); | |
1160 | ||
1161 | err = bd_prepare_to_claim(bdev, whole, holder); | |
1162 | if (err == 0) { | |
1163 | whole->bd_claiming = holder; | |
1164 | spin_unlock(&bdev_lock); | |
1165 | return whole; | |
1166 | } else { | |
1167 | spin_unlock(&bdev_lock); | |
1168 | bdput(whole); | |
1169 | return ERR_PTR(err); | |
1170 | } | |
1171 | } | |
89e524c0 JK |
1172 | EXPORT_SYMBOL(bd_start_claiming); |
1173 | ||
1174 | static void bd_clear_claiming(struct block_device *whole, void *holder) | |
1175 | { | |
1176 | lockdep_assert_held(&bdev_lock); | |
1177 | /* tell others that we're done */ | |
1178 | BUG_ON(whole->bd_claiming != holder); | |
1179 | whole->bd_claiming = NULL; | |
1180 | wake_up_bit(&whole->bd_claiming, 0); | |
1181 | } | |
1182 | ||
1183 | /** | |
1184 | * bd_finish_claiming - finish claiming of a block device | |
1185 | * @bdev: block device of interest | |
1186 | * @whole: whole block device (returned from bd_start_claiming()) | |
1187 | * @holder: holder that has claimed @bdev | |
1188 | * | |
1189 | * Finish exclusive open of a block device. Mark the device as exlusively | |
1190 | * open by the holder and wake up all waiters for exclusive open to finish. | |
1191 | */ | |
1192 | void bd_finish_claiming(struct block_device *bdev, struct block_device *whole, | |
1193 | void *holder) | |
1194 | { | |
1195 | spin_lock(&bdev_lock); | |
1196 | BUG_ON(!bd_may_claim(bdev, whole, holder)); | |
1197 | /* | |
1198 | * Note that for a whole device bd_holders will be incremented twice, | |
1199 | * and bd_holder will be set to bd_may_claim before being set to holder | |
1200 | */ | |
1201 | whole->bd_holders++; | |
1202 | whole->bd_holder = bd_may_claim; | |
1203 | bdev->bd_holders++; | |
1204 | bdev->bd_holder = holder; | |
1205 | bd_clear_claiming(whole, holder); | |
1206 | spin_unlock(&bdev_lock); | |
1207 | } | |
1208 | EXPORT_SYMBOL(bd_finish_claiming); | |
1209 | ||
1210 | /** | |
1211 | * bd_abort_claiming - abort claiming of a block device | |
1212 | * @bdev: block device of interest | |
1213 | * @whole: whole block device (returned from bd_start_claiming()) | |
1214 | * @holder: holder that has claimed @bdev | |
1215 | * | |
1216 | * Abort claiming of a block device when the exclusive open failed. This can be | |
1217 | * also used when exclusive open is not actually desired and we just needed | |
1218 | * to block other exclusive openers for a while. | |
1219 | */ | |
1220 | void bd_abort_claiming(struct block_device *bdev, struct block_device *whole, | |
1221 | void *holder) | |
1222 | { | |
1223 | spin_lock(&bdev_lock); | |
1224 | bd_clear_claiming(whole, holder); | |
1225 | spin_unlock(&bdev_lock); | |
1226 | } | |
1227 | EXPORT_SYMBOL(bd_abort_claiming); | |
6b4517a7 | 1228 | |
641dc636 | 1229 | #ifdef CONFIG_SYSFS |
49731baa TH |
1230 | struct bd_holder_disk { |
1231 | struct list_head list; | |
1232 | struct gendisk *disk; | |
1233 | int refcnt; | |
1234 | }; | |
1235 | ||
1236 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
1237 | struct gendisk *disk) | |
1238 | { | |
1239 | struct bd_holder_disk *holder; | |
1240 | ||
1241 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
1242 | if (holder->disk == disk) | |
1243 | return holder; | |
1244 | return NULL; | |
1245 | } | |
1246 | ||
4d7dd8fd | 1247 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 1248 | { |
4d7dd8fd | 1249 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
1250 | } |
1251 | ||
1252 | static void del_symlink(struct kobject *from, struct kobject *to) | |
1253 | { | |
641dc636 JN |
1254 | sysfs_remove_link(from, kobject_name(to)); |
1255 | } | |
1256 | ||
df6c0cd9 | 1257 | /** |
e09b457b TH |
1258 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
1259 | * @bdev: the claimed slave bdev | |
1260 | * @disk: the holding disk | |
df6c0cd9 | 1261 | * |
49731baa TH |
1262 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1263 | * | |
e09b457b | 1264 | * This functions creates the following sysfs symlinks. |
641dc636 | 1265 | * |
e09b457b TH |
1266 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1267 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 1268 | * |
e09b457b TH |
1269 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1270 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 1271 | * |
e09b457b TH |
1272 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1273 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 1274 | * |
e09b457b TH |
1275 | * The caller must have claimed @bdev before calling this function and |
1276 | * ensure that both @bdev and @disk are valid during the creation and | |
1277 | * lifetime of these symlinks. | |
641dc636 | 1278 | * |
e09b457b TH |
1279 | * CONTEXT: |
1280 | * Might sleep. | |
641dc636 | 1281 | * |
e09b457b TH |
1282 | * RETURNS: |
1283 | * 0 on success, -errno on failure. | |
641dc636 | 1284 | */ |
e09b457b | 1285 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 1286 | { |
49731baa | 1287 | struct bd_holder_disk *holder; |
e09b457b | 1288 | int ret = 0; |
641dc636 | 1289 | |
2e7b651d | 1290 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 1291 | |
49731baa | 1292 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 1293 | |
e09b457b TH |
1294 | /* FIXME: remove the following once add_disk() handles errors */ |
1295 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
1296 | goto out_unlock; | |
4e91672c | 1297 | |
49731baa TH |
1298 | holder = bd_find_holder_disk(bdev, disk); |
1299 | if (holder) { | |
1300 | holder->refcnt++; | |
e09b457b | 1301 | goto out_unlock; |
49731baa | 1302 | } |
641dc636 | 1303 | |
49731baa TH |
1304 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1305 | if (!holder) { | |
1306 | ret = -ENOMEM; | |
e09b457b TH |
1307 | goto out_unlock; |
1308 | } | |
641dc636 | 1309 | |
49731baa TH |
1310 | INIT_LIST_HEAD(&holder->list); |
1311 | holder->disk = disk; | |
1312 | holder->refcnt = 1; | |
1313 | ||
1314 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1315 | if (ret) | |
1316 | goto out_free; | |
1317 | ||
1318 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
1319 | if (ret) | |
1320 | goto out_del; | |
e7407d16 TH |
1321 | /* |
1322 | * bdev could be deleted beneath us which would implicitly destroy | |
1323 | * the holder directory. Hold on to it. | |
1324 | */ | |
1325 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1326 | |
1327 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1328 | goto out_unlock; | |
1329 | ||
1330 | out_del: | |
1331 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1332 | out_free: | |
1333 | kfree(holder); | |
e09b457b | 1334 | out_unlock: |
b4cf1b72 | 1335 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1336 | return ret; |
641dc636 | 1337 | } |
e09b457b | 1338 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1339 | |
49731baa TH |
1340 | /** |
1341 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1342 | * @bdev: the calimed slave bdev | |
1343 | * @disk: the holding disk | |
1344 | * | |
1345 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1346 | * | |
1347 | * CONTEXT: | |
1348 | * Might sleep. | |
1349 | */ | |
1350 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1351 | { |
49731baa | 1352 | struct bd_holder_disk *holder; |
641dc636 | 1353 | |
49731baa | 1354 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1355 | |
49731baa TH |
1356 | holder = bd_find_holder_disk(bdev, disk); |
1357 | ||
1358 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1359 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1360 | del_symlink(bdev->bd_part->holder_dir, | |
1361 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1362 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1363 | list_del_init(&holder->list); |
1364 | kfree(holder); | |
1365 | } | |
1366 | ||
1367 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1368 | } |
49731baa | 1369 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1370 | #endif |
1da177e4 | 1371 | |
56ade44b AP |
1372 | /** |
1373 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1374 | * | |
1375 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1376 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1377 | * |
1378 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1379 | * when a disk has been changed -- either by a media change or online | |
1380 | * resize. | |
1381 | */ | |
93b270f7 | 1382 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1383 | { |
93b270f7 | 1384 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1385 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1386 | "resized disk %s\n", |
1387 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b | 1388 | } |
cba22d86 | 1389 | bdev->bd_invalidated = 1; |
56ade44b AP |
1390 | } |
1391 | ||
c3279d14 | 1392 | /** |
57d1b536 | 1393 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1394 | * @disk: struct gendisk to check |
1395 | * @bdev: struct bdev to adjust. | |
5afb7835 | 1396 | * @verbose: if %true log a message about a size change if there is any |
c3279d14 AP |
1397 | * |
1398 | * This routine checks to see if the bdev size does not match the disk size | |
849cf559 | 1399 | * and adjusts it if it differs. When shrinking the bdev size, its all caches |
1400 | * are freed. | |
c3279d14 | 1401 | */ |
a1548b67 CH |
1402 | static void check_disk_size_change(struct gendisk *disk, |
1403 | struct block_device *bdev, bool verbose) | |
c3279d14 AP |
1404 | { |
1405 | loff_t disk_size, bdev_size; | |
1406 | ||
1407 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1408 | bdev_size = i_size_read(bdev->bd_inode); | |
1409 | if (disk_size != bdev_size) { | |
5afb7835 CH |
1410 | if (verbose) { |
1411 | printk(KERN_INFO | |
1412 | "%s: detected capacity change from %lld to %lld\n", | |
1413 | disk->disk_name, bdev_size, disk_size); | |
1414 | } | |
c3279d14 | 1415 | i_size_write(bdev->bd_inode, disk_size); |
849cf559 | 1416 | if (bdev_size > disk_size) |
1417 | flush_disk(bdev, false); | |
c3279d14 | 1418 | } |
979c690d | 1419 | bdev->bd_invalidated = 0; |
c3279d14 | 1420 | } |
c3279d14 | 1421 | |
0c002c2f | 1422 | /** |
57d1b536 | 1423 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1424 | * @disk: struct gendisk to be revalidated |
1425 | * | |
1426 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1427 | * call-backs. It is used to do common pre and post operations needed | |
1428 | * for all revalidate_disk operations. | |
1429 | */ | |
1430 | int revalidate_disk(struct gendisk *disk) | |
1431 | { | |
1432 | int ret = 0; | |
1433 | ||
1434 | if (disk->fops->revalidate_disk) | |
1435 | ret = disk->fops->revalidate_disk(disk); | |
c3279d14 | 1436 | |
31cb1d64 JK |
1437 | /* |
1438 | * Hidden disks don't have associated bdev so there's no point in | |
1439 | * revalidating it. | |
1440 | */ | |
1441 | if (!(disk->flags & GENHD_FL_HIDDEN)) { | |
1442 | struct block_device *bdev = bdget_disk(disk, 0); | |
1443 | ||
1444 | if (!bdev) | |
1445 | return ret; | |
1446 | ||
1447 | mutex_lock(&bdev->bd_mutex); | |
1448 | check_disk_size_change(disk, bdev, ret == 0); | |
31cb1d64 JK |
1449 | mutex_unlock(&bdev->bd_mutex); |
1450 | bdput(bdev); | |
1451 | } | |
0c002c2f AP |
1452 | return ret; |
1453 | } | |
1454 | EXPORT_SYMBOL(revalidate_disk); | |
1455 | ||
1da177e4 LT |
1456 | /* |
1457 | * This routine checks whether a removable media has been changed, | |
1458 | * and invalidates all buffer-cache-entries in that case. This | |
1459 | * is a relatively slow routine, so we have to try to minimize using | |
1460 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1461 | * is the best way of combining speed and utility, I think. | |
1462 | * People changing diskettes in the middle of an operation deserve | |
1463 | * to lose :-) | |
1464 | */ | |
1465 | int check_disk_change(struct block_device *bdev) | |
1466 | { | |
1467 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1468 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1469 | unsigned int events; |
1da177e4 | 1470 | |
77ea887e TH |
1471 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1472 | DISK_EVENT_EJECT_REQUEST); | |
1473 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1474 | return 0; |
1475 | ||
93b270f7 | 1476 | flush_disk(bdev, true); |
1da177e4 LT |
1477 | if (bdops->revalidate_disk) |
1478 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1479 | return 1; |
1480 | } | |
1481 | ||
1482 | EXPORT_SYMBOL(check_disk_change); | |
1483 | ||
1484 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1485 | { | |
5955102c | 1486 | inode_lock(bdev->bd_inode); |
d646a02a | 1487 | i_size_write(bdev->bd_inode, size); |
5955102c | 1488 | inode_unlock(bdev->bd_inode); |
1da177e4 LT |
1489 | } |
1490 | EXPORT_SYMBOL(bd_set_size); | |
1491 | ||
4385bab1 | 1492 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1493 | |
142fe8f4 | 1494 | int bdev_disk_changed(struct block_device *bdev, bool invalidate) |
a1548b67 | 1495 | { |
142fe8f4 | 1496 | struct gendisk *disk = bdev->bd_disk; |
a1548b67 CH |
1497 | int ret; |
1498 | ||
f0b870df CH |
1499 | lockdep_assert_held(&bdev->bd_mutex); |
1500 | ||
a1548b67 | 1501 | rescan: |
d46430bf | 1502 | ret = blk_drop_partitions(bdev); |
a1548b67 CH |
1503 | if (ret) |
1504 | return ret; | |
1505 | ||
d981cb5b CH |
1506 | /* |
1507 | * Historically we only set the capacity to zero for devices that | |
1508 | * support partitions (independ of actually having partitions created). | |
1509 | * Doing that is rather inconsistent, but changing it broke legacy | |
1510 | * udisks polling for legacy ide-cdrom devices. Use the crude check | |
1511 | * below to get the sane behavior for most device while not breaking | |
1512 | * userspace for this particular setup. | |
1513 | */ | |
1514 | if (invalidate) { | |
1515 | if (disk_part_scan_enabled(disk) || | |
1516 | !(disk->flags & GENHD_FL_REMOVABLE)) | |
1517 | set_capacity(disk, 0); | |
1518 | } else { | |
1519 | if (disk->fops->revalidate_disk) | |
1520 | disk->fops->revalidate_disk(disk); | |
1521 | } | |
a1548b67 CH |
1522 | |
1523 | check_disk_size_change(disk, bdev, !invalidate); | |
a1548b67 | 1524 | |
142fe8f4 CH |
1525 | if (get_capacity(disk)) { |
1526 | ret = blk_add_partitions(disk, bdev); | |
1527 | if (ret == -EAGAIN) | |
1528 | goto rescan; | |
490547ca | 1529 | } else if (invalidate) { |
a1548b67 CH |
1530 | /* |
1531 | * Tell userspace that the media / partition table may have | |
1532 | * changed. | |
1533 | */ | |
1534 | kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); | |
a1548b67 CH |
1535 | } |
1536 | ||
a1548b67 CH |
1537 | return ret; |
1538 | } | |
f0b870df CH |
1539 | /* |
1540 | * Only exported for for loop and dasd for historic reasons. Don't use in new | |
1541 | * code! | |
1542 | */ | |
1543 | EXPORT_SYMBOL_GPL(bdev_disk_changed); | |
a1548b67 | 1544 | |
6d740cd5 PZ |
1545 | /* |
1546 | * bd_mutex locking: | |
1547 | * | |
1548 | * mutex_lock(part->bd_mutex) | |
1549 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1550 | */ | |
1551 | ||
572c4892 | 1552 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1553 | { |
1da177e4 | 1554 | struct gendisk *disk; |
7db9cfd3 | 1555 | int ret; |
cf771cb5 | 1556 | int partno; |
fe6e9c1f | 1557 | int perm = 0; |
89736653 | 1558 | bool first_open = false; |
fe6e9c1f | 1559 | |
572c4892 | 1560 | if (mode & FMODE_READ) |
fe6e9c1f | 1561 | perm |= MAY_READ; |
572c4892 | 1562 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1563 | perm |= MAY_WRITE; |
1564 | /* | |
1565 | * hooks: /n/, see "layering violations". | |
1566 | */ | |
b7300b78 CW |
1567 | if (!for_part) { |
1568 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1569 | if (ret != 0) { | |
1570 | bdput(bdev); | |
1571 | return ret; | |
1572 | } | |
82666020 | 1573 | } |
7db9cfd3 | 1574 | |
d3374825 | 1575 | restart: |
0762b8bd | 1576 | |
89f97496 | 1577 | ret = -ENXIO; |
560e7cb2 | 1578 | disk = bdev_get_gendisk(bdev, &partno); |
0762b8bd | 1579 | if (!disk) |
6e9624b8 | 1580 | goto out; |
1da177e4 | 1581 | |
69e02c59 | 1582 | disk_block_events(disk); |
6796bf54 | 1583 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 | 1584 | if (!bdev->bd_openers) { |
89736653 | 1585 | first_open = true; |
1da177e4 | 1586 | bdev->bd_disk = disk; |
87192a2a | 1587 | bdev->bd_queue = disk->queue; |
1da177e4 | 1588 | bdev->bd_contains = bdev; |
c2ee070f | 1589 | bdev->bd_partno = partno; |
03cdadb0 | 1590 | |
cf771cb5 | 1591 | if (!partno) { |
89f97496 TH |
1592 | ret = -ENXIO; |
1593 | bdev->bd_part = disk_get_part(disk, partno); | |
1594 | if (!bdev->bd_part) | |
1595 | goto out_clear; | |
1596 | ||
1196f8b8 | 1597 | ret = 0; |
1da177e4 | 1598 | if (disk->fops->open) { |
572c4892 | 1599 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1600 | if (ret == -ERESTARTSYS) { |
1601 | /* Lost a race with 'disk' being | |
1602 | * deleted, try again. | |
1603 | * See md.c | |
1604 | */ | |
1605 | disk_put_part(bdev->bd_part); | |
1606 | bdev->bd_part = NULL; | |
d3374825 | 1607 | bdev->bd_disk = NULL; |
87192a2a | 1608 | bdev->bd_queue = NULL; |
d3374825 | 1609 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1610 | disk_unblock_events(disk); |
9df6c299 | 1611 | put_disk_and_module(disk); |
d3374825 N |
1612 | goto restart; |
1613 | } | |
1da177e4 | 1614 | } |
7e69723f | 1615 | |
04906b2f | 1616 | if (!ret) { |
7e69723f | 1617 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
04906b2f JK |
1618 | set_init_blocksize(bdev); |
1619 | } | |
7e69723f | 1620 | |
1196f8b8 TH |
1621 | /* |
1622 | * If the device is invalidated, rescan partition | |
1623 | * if open succeeded or failed with -ENOMEDIUM. | |
1624 | * The latter is necessary to prevent ghost | |
1625 | * partitions on a removed medium. | |
1626 | */ | |
731dc486 JK |
1627 | if (bdev->bd_invalidated && |
1628 | (!ret || ret == -ENOMEDIUM)) | |
1629 | bdev_disk_changed(bdev, ret == -ENOMEDIUM); | |
5a023cdb | 1630 | |
1196f8b8 TH |
1631 | if (ret) |
1632 | goto out_clear; | |
1da177e4 | 1633 | } else { |
1da177e4 LT |
1634 | struct block_device *whole; |
1635 | whole = bdget_disk(disk, 0); | |
1636 | ret = -ENOMEM; | |
1637 | if (!whole) | |
0762b8bd | 1638 | goto out_clear; |
37be4124 | 1639 | BUG_ON(for_part); |
572c4892 | 1640 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1641 | if (ret) |
0762b8bd | 1642 | goto out_clear; |
1da177e4 | 1643 | bdev->bd_contains = whole; |
89f97496 | 1644 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1645 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1646 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1647 | ret = -ENXIO; |
0762b8bd | 1648 | goto out_clear; |
1da177e4 | 1649 | } |
89f97496 | 1650 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
04906b2f | 1651 | set_init_blocksize(bdev); |
1da177e4 | 1652 | } |
03e26279 JK |
1653 | |
1654 | if (bdev->bd_bdi == &noop_backing_dev_info) | |
1655 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info); | |
1da177e4 | 1656 | } else { |
1da177e4 | 1657 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1658 | ret = 0; |
1659 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1660 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1661 | /* the same as first opener case, read comment there */ |
731dc486 JK |
1662 | if (bdev->bd_invalidated && |
1663 | (!ret || ret == -ENOMEDIUM)) | |
1664 | bdev_disk_changed(bdev, ret == -ENOMEDIUM); | |
1196f8b8 TH |
1665 | if (ret) |
1666 | goto out_unlock_bdev; | |
1da177e4 LT |
1667 | } |
1668 | } | |
1669 | bdev->bd_openers++; | |
37be4124 N |
1670 | if (for_part) |
1671 | bdev->bd_part_count++; | |
c039e313 | 1672 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1673 | disk_unblock_events(disk); |
89736653 JK |
1674 | /* only one opener holds refs to the module and disk */ |
1675 | if (!first_open) | |
1676 | put_disk_and_module(disk); | |
1da177e4 LT |
1677 | return 0; |
1678 | ||
0762b8bd | 1679 | out_clear: |
89f97496 | 1680 | disk_put_part(bdev->bd_part); |
1da177e4 | 1681 | bdev->bd_disk = NULL; |
0762b8bd | 1682 | bdev->bd_part = NULL; |
87192a2a | 1683 | bdev->bd_queue = NULL; |
1da177e4 | 1684 | if (bdev != bdev->bd_contains) |
572c4892 | 1685 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1686 | bdev->bd_contains = NULL; |
0762b8bd | 1687 | out_unlock_bdev: |
c039e313 | 1688 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1689 | disk_unblock_events(disk); |
9df6c299 | 1690 | put_disk_and_module(disk); |
4345caba | 1691 | out: |
0762b8bd TH |
1692 | bdput(bdev); |
1693 | ||
1da177e4 LT |
1694 | return ret; |
1695 | } | |
1696 | ||
d4d77629 TH |
1697 | /** |
1698 | * blkdev_get - open a block device | |
1699 | * @bdev: block_device to open | |
1700 | * @mode: FMODE_* mask | |
1701 | * @holder: exclusive holder identifier | |
1702 | * | |
1703 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1704 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1705 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1706 | * | |
1707 | * On success, the reference count of @bdev is unchanged. On failure, | |
1708 | * @bdev is put. | |
1709 | * | |
1710 | * CONTEXT: | |
1711 | * Might sleep. | |
1712 | * | |
1713 | * RETURNS: | |
1714 | * 0 on success, -errno on failure. | |
1715 | */ | |
e525fd89 | 1716 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1717 | { |
e525fd89 TH |
1718 | struct block_device *whole = NULL; |
1719 | int res; | |
1720 | ||
1721 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1722 | ||
1723 | if ((mode & FMODE_EXCL) && holder) { | |
1724 | whole = bd_start_claiming(bdev, holder); | |
1725 | if (IS_ERR(whole)) { | |
1726 | bdput(bdev); | |
1727 | return PTR_ERR(whole); | |
1728 | } | |
1729 | } | |
1730 | ||
1731 | res = __blkdev_get(bdev, mode, 0); | |
1732 | ||
1733 | if (whole) { | |
d4dc210f TH |
1734 | struct gendisk *disk = whole->bd_disk; |
1735 | ||
6a027eff | 1736 | /* finish claiming */ |
77ea887e | 1737 | mutex_lock(&bdev->bd_mutex); |
e91455ba JK |
1738 | if (!res) |
1739 | bd_finish_claiming(bdev, whole, holder); | |
1740 | else | |
1741 | bd_abort_claiming(bdev, whole, holder); | |
77ea887e | 1742 | /* |
d4dc210f TH |
1743 | * Block event polling for write claims if requested. Any |
1744 | * write holder makes the write_holder state stick until | |
1745 | * all are released. This is good enough and tracking | |
1746 | * individual writeable reference is too fragile given the | |
1747 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1748 | */ |
4c49ff3f TH |
1749 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1750 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1751 | bdev->bd_write_holder = true; |
d4dc210f | 1752 | disk_block_events(disk); |
77ea887e TH |
1753 | } |
1754 | ||
1755 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1756 | bdput(whole); |
e525fd89 TH |
1757 | } |
1758 | ||
1759 | return res; | |
37be4124 | 1760 | } |
1da177e4 LT |
1761 | EXPORT_SYMBOL(blkdev_get); |
1762 | ||
d4d77629 TH |
1763 | /** |
1764 | * blkdev_get_by_path - open a block device by name | |
1765 | * @path: path to the block device to open | |
1766 | * @mode: FMODE_* mask | |
1767 | * @holder: exclusive holder identifier | |
1768 | * | |
1769 | * Open the blockdevice described by the device file at @path. @mode | |
1770 | * and @holder are identical to blkdev_get(). | |
1771 | * | |
1772 | * On success, the returned block_device has reference count of one. | |
1773 | * | |
1774 | * CONTEXT: | |
1775 | * Might sleep. | |
1776 | * | |
1777 | * RETURNS: | |
1778 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1779 | */ | |
1780 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1781 | void *holder) | |
1782 | { | |
1783 | struct block_device *bdev; | |
1784 | int err; | |
1785 | ||
1786 | bdev = lookup_bdev(path); | |
1787 | if (IS_ERR(bdev)) | |
1788 | return bdev; | |
1789 | ||
1790 | err = blkdev_get(bdev, mode, holder); | |
1791 | if (err) | |
1792 | return ERR_PTR(err); | |
1793 | ||
e51900f7 CE |
1794 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1795 | blkdev_put(bdev, mode); | |
1796 | return ERR_PTR(-EACCES); | |
1797 | } | |
1798 | ||
d4d77629 TH |
1799 | return bdev; |
1800 | } | |
1801 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1802 | ||
1803 | /** | |
1804 | * blkdev_get_by_dev - open a block device by device number | |
1805 | * @dev: device number of block device to open | |
1806 | * @mode: FMODE_* mask | |
1807 | * @holder: exclusive holder identifier | |
1808 | * | |
1809 | * Open the blockdevice described by device number @dev. @mode and | |
1810 | * @holder are identical to blkdev_get(). | |
1811 | * | |
1812 | * Use it ONLY if you really do not have anything better - i.e. when | |
1813 | * you are behind a truly sucky interface and all you are given is a | |
1814 | * device number. _Never_ to be used for internal purposes. If you | |
1815 | * ever need it - reconsider your API. | |
1816 | * | |
1817 | * On success, the returned block_device has reference count of one. | |
1818 | * | |
1819 | * CONTEXT: | |
1820 | * Might sleep. | |
1821 | * | |
1822 | * RETURNS: | |
1823 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1824 | */ | |
1825 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1826 | { | |
1827 | struct block_device *bdev; | |
1828 | int err; | |
1829 | ||
1830 | bdev = bdget(dev); | |
1831 | if (!bdev) | |
1832 | return ERR_PTR(-ENOMEM); | |
1833 | ||
1834 | err = blkdev_get(bdev, mode, holder); | |
1835 | if (err) | |
1836 | return ERR_PTR(err); | |
1837 | ||
1838 | return bdev; | |
1839 | } | |
1840 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1841 | ||
1da177e4 LT |
1842 | static int blkdev_open(struct inode * inode, struct file * filp) |
1843 | { | |
1844 | struct block_device *bdev; | |
1da177e4 LT |
1845 | |
1846 | /* | |
1847 | * Preserve backwards compatibility and allow large file access | |
1848 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1849 | * binary needs it. We might want to drop this workaround | |
1850 | * during an unstable branch. | |
1851 | */ | |
1852 | filp->f_flags |= O_LARGEFILE; | |
1853 | ||
c35fc7a5 CH |
1854 | filp->f_mode |= FMODE_NOWAIT; |
1855 | ||
572c4892 AV |
1856 | if (filp->f_flags & O_NDELAY) |
1857 | filp->f_mode |= FMODE_NDELAY; | |
1858 | if (filp->f_flags & O_EXCL) | |
1859 | filp->f_mode |= FMODE_EXCL; | |
1860 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1861 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1862 | ||
1da177e4 | 1863 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1864 | if (bdev == NULL) |
1865 | return -ENOMEM; | |
1da177e4 | 1866 | |
572c4892 | 1867 | filp->f_mapping = bdev->bd_inode->i_mapping; |
5660e13d | 1868 | filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping); |
572c4892 | 1869 | |
e525fd89 | 1870 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1871 | } |
1872 | ||
4385bab1 | 1873 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1874 | { |
2e7b651d | 1875 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1876 | struct block_device *victim = NULL; |
2e7b651d | 1877 | |
b849dd84 DA |
1878 | /* |
1879 | * Sync early if it looks like we're the last one. If someone else | |
1880 | * opens the block device between now and the decrement of bd_openers | |
1881 | * then we did a sync that we didn't need to, but that's not the end | |
1882 | * of the world and we want to avoid long (could be several minute) | |
1883 | * syncs while holding the mutex. | |
1884 | */ | |
1885 | if (bdev->bd_openers == 1) | |
1886 | sync_blockdev(bdev); | |
1887 | ||
6796bf54 | 1888 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1889 | if (for_part) |
1890 | bdev->bd_part_count--; | |
1891 | ||
2e7b651d | 1892 | if (!--bdev->bd_openers) { |
6a027eff | 1893 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1894 | sync_blockdev(bdev); |
1895 | kill_bdev(bdev); | |
43d1c0eb ID |
1896 | |
1897 | bdev_write_inode(bdev); | |
2e7b651d PZ |
1898 | } |
1899 | if (bdev->bd_contains == bdev) { | |
1900 | if (disk->fops->release) | |
db2a144b | 1901 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1902 | } |
1903 | if (!bdev->bd_openers) { | |
0762b8bd TH |
1904 | disk_put_part(bdev->bd_part); |
1905 | bdev->bd_part = NULL; | |
2e7b651d | 1906 | bdev->bd_disk = NULL; |
37be4124 N |
1907 | if (bdev != bdev->bd_contains) |
1908 | victim = bdev->bd_contains; | |
2e7b651d | 1909 | bdev->bd_contains = NULL; |
523e1d39 | 1910 | |
9df6c299 | 1911 | put_disk_and_module(disk); |
2e7b651d | 1912 | } |
2e7b651d PZ |
1913 | mutex_unlock(&bdev->bd_mutex); |
1914 | bdput(bdev); | |
37be4124 | 1915 | if (victim) |
9a1c3542 | 1916 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1917 | } |
1918 | ||
4385bab1 | 1919 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1920 | { |
85ef06d1 TH |
1921 | mutex_lock(&bdev->bd_mutex); |
1922 | ||
e525fd89 | 1923 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1924 | bool bdev_free; |
1925 | ||
1926 | /* | |
1927 | * Release a claim on the device. The holder fields | |
1928 | * are protected with bdev_lock. bd_mutex is to | |
1929 | * synchronize disk_holder unlinking. | |
1930 | */ | |
6a027eff TH |
1931 | spin_lock(&bdev_lock); |
1932 | ||
1933 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1934 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1935 | ||
1936 | /* bd_contains might point to self, check in a separate step */ | |
1937 | if ((bdev_free = !bdev->bd_holders)) | |
1938 | bdev->bd_holder = NULL; | |
1939 | if (!bdev->bd_contains->bd_holders) | |
1940 | bdev->bd_contains->bd_holder = NULL; | |
1941 | ||
1942 | spin_unlock(&bdev_lock); | |
1943 | ||
77ea887e TH |
1944 | /* |
1945 | * If this was the last claim, remove holder link and | |
1946 | * unblock evpoll if it was a write holder. | |
1947 | */ | |
85ef06d1 TH |
1948 | if (bdev_free && bdev->bd_write_holder) { |
1949 | disk_unblock_events(bdev->bd_disk); | |
1950 | bdev->bd_write_holder = false; | |
77ea887e | 1951 | } |
6936217c | 1952 | } |
77ea887e | 1953 | |
85ef06d1 TH |
1954 | /* |
1955 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1956 | * event. This is to ensure detection of media removal commanded | |
1957 | * from userland - e.g. eject(1). | |
1958 | */ | |
1959 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1960 | ||
1961 | mutex_unlock(&bdev->bd_mutex); | |
1962 | ||
4385bab1 | 1963 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1964 | } |
2e7b651d PZ |
1965 | EXPORT_SYMBOL(blkdev_put); |
1966 | ||
1da177e4 LT |
1967 | static int blkdev_close(struct inode * inode, struct file * filp) |
1968 | { | |
4ebb16ca | 1969 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1970 | blkdev_put(bdev, filp->f_mode); |
1971 | return 0; | |
1da177e4 LT |
1972 | } |
1973 | ||
bb93e3a5 | 1974 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1975 | { |
4ebb16ca | 1976 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1977 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1978 | |
1979 | /* | |
1980 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1981 | * to updated it before every ioctl. | |
1982 | */ | |
56b26add | 1983 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1984 | mode |= FMODE_NDELAY; |
1985 | else | |
1986 | mode &= ~FMODE_NDELAY; | |
1987 | ||
56b26add | 1988 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1989 | } |
1990 | ||
eef99380 CH |
1991 | /* |
1992 | * Write data to the block device. Only intended for the block device itself | |
1993 | * and the raw driver which basically is a fake block device. | |
1994 | * | |
1995 | * Does not take i_mutex for the write and thus is not for general purpose | |
1996 | * use. | |
1997 | */ | |
1456c0a8 | 1998 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1999 | { |
2000 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 2001 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 2002 | loff_t size = i_size_read(bd_inode); |
53362a05 | 2003 | struct blk_plug plug; |
eef99380 | 2004 | ssize_t ret; |
5f380c7f | 2005 | |
7ec7b94a AV |
2006 | if (bdev_read_only(I_BDEV(bd_inode))) |
2007 | return -EPERM; | |
5f380c7f | 2008 | |
56939e01 DA |
2009 | /* uswsusp needs write permission to the swap */ |
2010 | if (IS_SWAPFILE(bd_inode) && !hibernation_available()) | |
dc617f29 DW |
2011 | return -ETXTBSY; |
2012 | ||
7ec7b94a | 2013 | if (!iov_iter_count(from)) |
5f380c7f AV |
2014 | return 0; |
2015 | ||
7ec7b94a AV |
2016 | if (iocb->ki_pos >= size) |
2017 | return -ENOSPC; | |
2018 | ||
c35fc7a5 CH |
2019 | if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) |
2020 | return -EOPNOTSUPP; | |
2021 | ||
7ec7b94a | 2022 | iov_iter_truncate(from, size - iocb->ki_pos); |
eef99380 | 2023 | |
53362a05 | 2024 | blk_start_plug(&plug); |
1456c0a8 | 2025 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
2026 | if (ret > 0) |
2027 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 2028 | blk_finish_plug(&plug); |
eef99380 CH |
2029 | return ret; |
2030 | } | |
1456c0a8 | 2031 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 2032 | |
b2de525f | 2033 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
2034 | { |
2035 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 2036 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 2037 | loff_t size = i_size_read(bd_inode); |
a886038b | 2038 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
2039 | |
2040 | if (pos >= size) | |
2041 | return 0; | |
2042 | ||
2043 | size -= pos; | |
a886038b AV |
2044 | iov_iter_truncate(to, size); |
2045 | return generic_file_read_iter(iocb, to); | |
684c9aae | 2046 | } |
b2de525f | 2047 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 2048 | |
87d8fe1e TT |
2049 | /* |
2050 | * Try to release a page associated with block device when the system | |
2051 | * is under memory pressure. | |
2052 | */ | |
2053 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
2054 | { | |
2055 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
2056 | ||
2057 | if (super && super->s_op->bdev_try_to_free_page) | |
2058 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
2059 | ||
2060 | return try_to_free_buffers(page); | |
2061 | } | |
2062 | ||
7f6d5b52 RZ |
2063 | static int blkdev_writepages(struct address_space *mapping, |
2064 | struct writeback_control *wbc) | |
2065 | { | |
7f6d5b52 RZ |
2066 | return generic_writepages(mapping, wbc); |
2067 | } | |
2068 | ||
4c54ac62 | 2069 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 2070 | .readpage = blkdev_readpage, |
447f05bb | 2071 | .readpages = blkdev_readpages, |
1da177e4 | 2072 | .writepage = blkdev_writepage, |
6272b5a5 NP |
2073 | .write_begin = blkdev_write_begin, |
2074 | .write_end = blkdev_write_end, | |
7f6d5b52 | 2075 | .writepages = blkdev_writepages, |
87d8fe1e | 2076 | .releasepage = blkdev_releasepage, |
1da177e4 | 2077 | .direct_IO = blkdev_direct_IO, |
88dbcbb3 | 2078 | .migratepage = buffer_migrate_page_norefs, |
b4597226 | 2079 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
2080 | }; |
2081 | ||
25f4c414 DW |
2082 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
2083 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
2084 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) | |
2085 | ||
2086 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, | |
2087 | loff_t len) | |
2088 | { | |
2089 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
25f4c414 DW |
2090 | struct address_space *mapping; |
2091 | loff_t end = start + len - 1; | |
2092 | loff_t isize; | |
2093 | int error; | |
2094 | ||
2095 | /* Fail if we don't recognize the flags. */ | |
2096 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) | |
2097 | return -EOPNOTSUPP; | |
2098 | ||
2099 | /* Don't go off the end of the device. */ | |
2100 | isize = i_size_read(bdev->bd_inode); | |
2101 | if (start >= isize) | |
2102 | return -EINVAL; | |
2103 | if (end >= isize) { | |
2104 | if (mode & FALLOC_FL_KEEP_SIZE) { | |
2105 | len = isize - start; | |
2106 | end = start + len - 1; | |
2107 | } else | |
2108 | return -EINVAL; | |
2109 | } | |
2110 | ||
2111 | /* | |
2112 | * Don't allow IO that isn't aligned to logical block size. | |
2113 | */ | |
2114 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) | |
2115 | return -EINVAL; | |
2116 | ||
2117 | /* Invalidate the page cache, including dirty pages. */ | |
2118 | mapping = bdev->bd_inode->i_mapping; | |
2119 | truncate_inode_pages_range(mapping, start, end); | |
2120 | ||
2121 | switch (mode) { | |
2122 | case FALLOC_FL_ZERO_RANGE: | |
2123 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: | |
2124 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, | |
ee472d83 | 2125 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP); |
25f4c414 DW |
2126 | break; |
2127 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: | |
34045129 CH |
2128 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
2129 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK); | |
25f4c414 DW |
2130 | break; |
2131 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: | |
25f4c414 DW |
2132 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
2133 | GFP_KERNEL, 0); | |
2134 | break; | |
2135 | default: | |
2136 | return -EOPNOTSUPP; | |
2137 | } | |
2138 | if (error) | |
2139 | return error; | |
2140 | ||
2141 | /* | |
2142 | * Invalidate again; if someone wandered in and dirtied a page, | |
2143 | * the caller will be given -EBUSY. The third argument is | |
2144 | * inclusive, so the rounding here is safe. | |
2145 | */ | |
2146 | return invalidate_inode_pages2_range(mapping, | |
2147 | start >> PAGE_SHIFT, | |
2148 | end >> PAGE_SHIFT); | |
2149 | } | |
2150 | ||
4b6f5d20 | 2151 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
2152 | .open = blkdev_open, |
2153 | .release = blkdev_close, | |
2154 | .llseek = block_llseek, | |
a886038b | 2155 | .read_iter = blkdev_read_iter, |
1456c0a8 | 2156 | .write_iter = blkdev_write_iter, |
eae83ce1 | 2157 | .iopoll = blkdev_iopoll, |
acc93d30 | 2158 | .mmap = generic_file_mmap, |
b1dd3b28 | 2159 | .fsync = blkdev_fsync, |
bb93e3a5 | 2160 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
2161 | #ifdef CONFIG_COMPAT |
2162 | .compat_ioctl = compat_blkdev_ioctl, | |
2163 | #endif | |
1e8b3332 | 2164 | .splice_read = generic_file_splice_read, |
8d020765 | 2165 | .splice_write = iter_file_splice_write, |
25f4c414 | 2166 | .fallocate = blkdev_fallocate, |
1da177e4 LT |
2167 | }; |
2168 | ||
2169 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
2170 | { | |
2171 | int res; | |
2172 | mm_segment_t old_fs = get_fs(); | |
2173 | set_fs(KERNEL_DS); | |
56b26add | 2174 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
2175 | set_fs(old_fs); |
2176 | return res; | |
2177 | } | |
2178 | ||
2179 | EXPORT_SYMBOL(ioctl_by_bdev); | |
2180 | ||
2181 | /** | |
2182 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 2183 | * @pathname: special file representing the block device |
1da177e4 | 2184 | * |
57d1b536 | 2185 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
2186 | * namespace if possible and return it. Return ERR_PTR(error) |
2187 | * otherwise. | |
2188 | */ | |
421748ec | 2189 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
2190 | { |
2191 | struct block_device *bdev; | |
2192 | struct inode *inode; | |
421748ec | 2193 | struct path path; |
1da177e4 LT |
2194 | int error; |
2195 | ||
421748ec | 2196 | if (!pathname || !*pathname) |
1da177e4 LT |
2197 | return ERR_PTR(-EINVAL); |
2198 | ||
421748ec | 2199 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
2200 | if (error) |
2201 | return ERR_PTR(error); | |
2202 | ||
bb668734 | 2203 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
2204 | error = -ENOTBLK; |
2205 | if (!S_ISBLK(inode->i_mode)) | |
2206 | goto fail; | |
2207 | error = -EACCES; | |
a2982cc9 | 2208 | if (!may_open_dev(&path)) |
1da177e4 LT |
2209 | goto fail; |
2210 | error = -ENOMEM; | |
2211 | bdev = bd_acquire(inode); | |
2212 | if (!bdev) | |
2213 | goto fail; | |
2214 | out: | |
421748ec | 2215 | path_put(&path); |
1da177e4 LT |
2216 | return bdev; |
2217 | fail: | |
2218 | bdev = ERR_PTR(error); | |
2219 | goto out; | |
2220 | } | |
d5686b44 | 2221 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 2222 | |
93b270f7 | 2223 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
2224 | { |
2225 | struct super_block *sb = get_super(bdev); | |
2226 | int res = 0; | |
2227 | ||
2228 | if (sb) { | |
2229 | /* | |
2230 | * no need to lock the super, get_super holds the | |
2231 | * read mutex so the filesystem cannot go away | |
2232 | * under us (->put_super runs with the write lock | |
2233 | * hold). | |
2234 | */ | |
2235 | shrink_dcache_sb(sb); | |
93b270f7 | 2236 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
2237 | drop_super(sb); |
2238 | } | |
f98393a6 | 2239 | invalidate_bdev(bdev); |
b71e8a4c DH |
2240 | return res; |
2241 | } | |
2242 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
2243 | |
2244 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
2245 | { | |
2246 | struct inode *inode, *old_inode = NULL; | |
2247 | ||
74278da9 | 2248 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2249 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
2250 | struct address_space *mapping = inode->i_mapping; | |
af309226 | 2251 | struct block_device *bdev; |
5c0d6b60 JK |
2252 | |
2253 | spin_lock(&inode->i_lock); | |
2254 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
2255 | mapping->nrpages == 0) { | |
2256 | spin_unlock(&inode->i_lock); | |
2257 | continue; | |
2258 | } | |
2259 | __iget(inode); | |
2260 | spin_unlock(&inode->i_lock); | |
74278da9 | 2261 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2262 | /* |
2263 | * We hold a reference to 'inode' so it couldn't have been | |
2264 | * removed from s_inodes list while we dropped the | |
74278da9 | 2265 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 2266 | * be holding the last reference and we cannot iput it under |
74278da9 | 2267 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
2268 | * later. |
2269 | */ | |
2270 | iput(old_inode); | |
2271 | old_inode = inode; | |
af309226 | 2272 | bdev = I_BDEV(inode); |
5c0d6b60 | 2273 | |
af309226 RV |
2274 | mutex_lock(&bdev->bd_mutex); |
2275 | if (bdev->bd_openers) | |
2276 | func(bdev, arg); | |
2277 | mutex_unlock(&bdev->bd_mutex); | |
5c0d6b60 | 2278 | |
74278da9 | 2279 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 2280 | } |
74278da9 | 2281 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2282 | iput(old_inode); |
2283 | } |