]>
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> |
b0686260 | 22 | #include <linux/dax.h> |
1da177e4 | 23 | #include <linux/buffer_head.h> |
ff01bb48 | 24 | #include <linux/swap.h> |
585d3bc0 | 25 | #include <linux/pagevec.h> |
811d736f | 26 | #include <linux/writeback.h> |
1da177e4 LT |
27 | #include <linux/mpage.h> |
28 | #include <linux/mount.h> | |
9030d16e | 29 | #include <linux/pseudo_fs.h> |
1da177e4 LT |
30 | #include <linux/uio.h> |
31 | #include <linux/namei.h> | |
1368c4f2 | 32 | #include <linux/log2.h> |
ff01bb48 | 33 | #include <linux/cleancache.h> |
189ce2b9 | 34 | #include <linux/task_io_accounting_ops.h> |
25f4c414 | 35 | #include <linux/falloc.h> |
7c0f6ba6 | 36 | #include <linux/uaccess.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)) |
189ce2b9 CH |
258 | io_schedule(); |
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; |
6a43074e | 348 | bool nowait = (iocb->ki_flags & IOCB_NOWAIT) != 0; |
542ff7bf CH |
349 | loff_t pos = iocb->ki_pos; |
350 | blk_qc_t qc = BLK_QC_T_NONE; | |
6a43074e JA |
351 | gfp_t gfp; |
352 | ssize_t ret; | |
542ff7bf | 353 | |
9a794fb9 JA |
354 | if ((pos | iov_iter_alignment(iter)) & |
355 | (bdev_logical_block_size(bdev) - 1)) | |
542ff7bf CH |
356 | return -EINVAL; |
357 | ||
6a43074e JA |
358 | if (nowait) |
359 | gfp = GFP_NOWAIT; | |
360 | else | |
361 | gfp = GFP_KERNEL; | |
362 | ||
363 | bio = bio_alloc_bioset(gfp, nr_pages, &blkdev_dio_pool); | |
364 | if (!bio) | |
365 | return -EAGAIN; | |
542ff7bf CH |
366 | |
367 | dio = container_of(bio, struct blkdev_dio, bio); | |
690e5325 | 368 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
531724ab | 369 | if (dio->is_sync) { |
542ff7bf | 370 | dio->waiter = current; |
531724ab CH |
371 | bio_get(bio); |
372 | } else { | |
542ff7bf | 373 | dio->iocb = iocb; |
531724ab | 374 | } |
542ff7bf CH |
375 | |
376 | dio->size = 0; | |
377 | dio->multi_bio = false; | |
00e23707 | 378 | dio->should_dirty = is_read && iter_is_iovec(iter); |
542ff7bf | 379 | |
cb700eb3 JA |
380 | /* |
381 | * Don't plug for HIPRI/polled IO, as those should go straight | |
382 | * to issue | |
383 | */ | |
384 | if (!is_poll) | |
385 | blk_start_plug(&plug); | |
386 | ||
6a43074e | 387 | ret = 0; |
542ff7bf | 388 | for (;;) { |
6a43074e JA |
389 | int err; |
390 | ||
74d46992 | 391 | bio_set_dev(bio, bdev); |
4d1a4765 | 392 | bio->bi_iter.bi_sector = pos >> 9; |
45d06cf7 | 393 | bio->bi_write_hint = iocb->ki_hint; |
542ff7bf CH |
394 | bio->bi_private = dio; |
395 | bio->bi_end_io = blkdev_bio_end_io; | |
074111ca | 396 | bio->bi_ioprio = iocb->ki_ioprio; |
542ff7bf | 397 | |
6a43074e JA |
398 | err = bio_iov_iter_get_pages(bio, iter); |
399 | if (unlikely(err)) { | |
400 | if (!ret) | |
401 | ret = err; | |
4e4cbee9 | 402 | bio->bi_status = BLK_STS_IOERR; |
542ff7bf CH |
403 | bio_endio(bio); |
404 | break; | |
405 | } | |
406 | ||
407 | if (is_read) { | |
408 | bio->bi_opf = REQ_OP_READ; | |
409 | if (dio->should_dirty) | |
410 | bio_set_pages_dirty(bio); | |
411 | } else { | |
412 | bio->bi_opf = dio_bio_write_op(iocb); | |
413 | task_io_account_write(bio->bi_iter.bi_size); | |
414 | } | |
415 | ||
6a43074e JA |
416 | /* |
417 | * Tell underlying layer to not block for resource shortage. | |
418 | * And if we would have blocked, return error inline instead | |
419 | * of through the bio->bi_end_io() callback. | |
420 | */ | |
421 | if (nowait) | |
422 | bio->bi_opf |= (REQ_NOWAIT | REQ_NOWAIT_INLINE); | |
423 | ||
542ff7bf CH |
424 | dio->size += bio->bi_iter.bi_size; |
425 | pos += bio->bi_iter.bi_size; | |
426 | ||
427 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); | |
428 | if (!nr_pages) { | |
eae83ce1 CH |
429 | bool polled = false; |
430 | ||
431 | if (iocb->ki_flags & IOCB_HIPRI) { | |
0bbb280d | 432 | bio_set_polled(bio, iocb); |
eae83ce1 CH |
433 | polled = true; |
434 | } | |
d34513d3 | 435 | |
542ff7bf | 436 | qc = submit_bio(bio); |
6a43074e JA |
437 | if (qc == BLK_QC_T_EAGAIN) { |
438 | if (!ret) | |
439 | ret = -EAGAIN; | |
440 | goto error; | |
441 | } | |
eae83ce1 CH |
442 | |
443 | if (polled) | |
444 | WRITE_ONCE(iocb->ki_cookie, qc); | |
542ff7bf CH |
445 | break; |
446 | } | |
447 | ||
448 | if (!dio->multi_bio) { | |
531724ab CH |
449 | /* |
450 | * AIO needs an extra reference to ensure the dio | |
451 | * structure which is embedded into the first bio | |
452 | * stays around. | |
453 | */ | |
454 | if (!is_sync) | |
455 | bio_get(bio); | |
542ff7bf CH |
456 | dio->multi_bio = true; |
457 | atomic_set(&dio->ref, 2); | |
458 | } else { | |
459 | atomic_inc(&dio->ref); | |
460 | } | |
461 | ||
6a43074e JA |
462 | qc = submit_bio(bio); |
463 | if (qc == BLK_QC_T_EAGAIN) { | |
464 | if (!ret) | |
465 | ret = -EAGAIN; | |
466 | goto error; | |
467 | } | |
468 | ret += bio->bi_iter.bi_size; | |
469 | ||
470 | bio = bio_alloc(gfp, nr_pages); | |
471 | if (!bio) { | |
472 | if (!ret) | |
473 | ret = -EAGAIN; | |
474 | goto error; | |
475 | } | |
542ff7bf | 476 | } |
cb700eb3 JA |
477 | |
478 | if (!is_poll) | |
479 | blk_finish_plug(&plug); | |
542ff7bf | 480 | |
690e5325 | 481 | if (!is_sync) |
542ff7bf CH |
482 | return -EIOCBQUEUED; |
483 | ||
484 | for (;;) { | |
1ac5cd49 | 485 | set_current_state(TASK_UNINTERRUPTIBLE); |
542ff7bf CH |
486 | if (!READ_ONCE(dio->waiter)) |
487 | break; | |
488 | ||
489 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
0a1b8b87 | 490 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
542ff7bf CH |
491 | io_schedule(); |
492 | } | |
493 | __set_current_state(TASK_RUNNING); | |
494 | ||
6a43074e | 495 | out: |
36ffc6c1 | 496 | if (!ret) |
4e4cbee9 | 497 | ret = blk_status_to_errno(dio->bio.bi_status); |
542ff7bf CH |
498 | |
499 | bio_put(&dio->bio); | |
500 | return ret; | |
6a43074e JA |
501 | error: |
502 | if (!is_poll) | |
503 | blk_finish_plug(&plug); | |
504 | goto out; | |
542ff7bf CH |
505 | } |
506 | ||
507 | static ssize_t | |
508 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) | |
509 | { | |
189ce2b9 | 510 | int nr_pages; |
b2e895db | 511 | |
72ecad22 | 512 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
189ce2b9 CH |
513 | if (!nr_pages) |
514 | return 0; | |
72ecad22 | 515 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
189ce2b9 | 516 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
542ff7bf CH |
517 | |
518 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); | |
519 | } | |
520 | ||
521 | static __init int blkdev_init(void) | |
522 | { | |
52190f8a | 523 | return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS); |
b2e895db | 524 | } |
542ff7bf | 525 | module_init(blkdev_init); |
b2e895db | 526 | |
5cee5815 JK |
527 | int __sync_blockdev(struct block_device *bdev, int wait) |
528 | { | |
529 | if (!bdev) | |
530 | return 0; | |
531 | if (!wait) | |
532 | return filemap_flush(bdev->bd_inode->i_mapping); | |
533 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
534 | } | |
535 | ||
585d3bc0 NP |
536 | /* |
537 | * Write out and wait upon all the dirty data associated with a block | |
538 | * device via its mapping. Does not take the superblock lock. | |
539 | */ | |
540 | int sync_blockdev(struct block_device *bdev) | |
541 | { | |
5cee5815 | 542 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
543 | } |
544 | EXPORT_SYMBOL(sync_blockdev); | |
545 | ||
546 | /* | |
547 | * Write out and wait upon all dirty data associated with this | |
548 | * device. Filesystem data as well as the underlying block | |
549 | * device. Takes the superblock lock. | |
550 | */ | |
551 | int fsync_bdev(struct block_device *bdev) | |
552 | { | |
553 | struct super_block *sb = get_super(bdev); | |
554 | if (sb) { | |
60b0680f | 555 | int res = sync_filesystem(sb); |
585d3bc0 NP |
556 | drop_super(sb); |
557 | return res; | |
558 | } | |
559 | return sync_blockdev(bdev); | |
560 | } | |
47e4491b | 561 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
562 | |
563 | /** | |
564 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
565 | * @bdev: blockdevice to lock | |
566 | * | |
585d3bc0 NP |
567 | * If a superblock is found on this device, we take the s_umount semaphore |
568 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
569 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
570 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
571 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
572 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
573 | * actually. | |
574 | */ | |
575 | struct super_block *freeze_bdev(struct block_device *bdev) | |
576 | { | |
577 | struct super_block *sb; | |
578 | int error = 0; | |
579 | ||
580 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
581 | if (++bdev->bd_fsfreeze_count > 1) { |
582 | /* | |
583 | * We don't even need to grab a reference - the first call | |
584 | * to freeze_bdev grab an active reference and only the last | |
585 | * thaw_bdev drops it. | |
586 | */ | |
585d3bc0 | 587 | sb = get_super(bdev); |
5bb53c0f AR |
588 | if (sb) |
589 | drop_super(sb); | |
4504230a CH |
590 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
591 | return sb; | |
592 | } | |
593 | ||
594 | sb = get_active_super(bdev); | |
595 | if (!sb) | |
596 | goto out; | |
48b6bca6 BM |
597 | if (sb->s_op->freeze_super) |
598 | error = sb->s_op->freeze_super(sb); | |
599 | else | |
600 | error = freeze_super(sb); | |
18e9e510 JB |
601 | if (error) { |
602 | deactivate_super(sb); | |
603 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 604 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 605 | return ERR_PTR(error); |
585d3bc0 | 606 | } |
18e9e510 | 607 | deactivate_super(sb); |
4504230a | 608 | out: |
585d3bc0 NP |
609 | sync_blockdev(bdev); |
610 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 611 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
612 | } |
613 | EXPORT_SYMBOL(freeze_bdev); | |
614 | ||
615 | /** | |
616 | * thaw_bdev -- unlock filesystem | |
617 | * @bdev: blockdevice to unlock | |
618 | * @sb: associated superblock | |
619 | * | |
620 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
621 | */ | |
622 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
623 | { | |
4504230a | 624 | int error = -EINVAL; |
585d3bc0 NP |
625 | |
626 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 627 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 628 | goto out; |
4504230a CH |
629 | |
630 | error = 0; | |
631 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 632 | goto out; |
4504230a CH |
633 | |
634 | if (!sb) | |
18e9e510 | 635 | goto out; |
4504230a | 636 | |
48b6bca6 BM |
637 | if (sb->s_op->thaw_super) |
638 | error = sb->s_op->thaw_super(sb); | |
639 | else | |
640 | error = thaw_super(sb); | |
997198ba | 641 | if (error) |
18e9e510 | 642 | bdev->bd_fsfreeze_count++; |
18e9e510 | 643 | out: |
585d3bc0 | 644 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
997198ba | 645 | return error; |
585d3bc0 NP |
646 | } |
647 | EXPORT_SYMBOL(thaw_bdev); | |
648 | ||
1da177e4 LT |
649 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
650 | { | |
651 | return block_write_full_page(page, blkdev_get_block, wbc); | |
652 | } | |
653 | ||
654 | static int blkdev_readpage(struct file * file, struct page * page) | |
655 | { | |
656 | return block_read_full_page(page, blkdev_get_block); | |
657 | } | |
658 | ||
447f05bb AM |
659 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
660 | struct list_head *pages, unsigned nr_pages) | |
661 | { | |
662 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
663 | } | |
664 | ||
6272b5a5 NP |
665 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
666 | loff_t pos, unsigned len, unsigned flags, | |
667 | struct page **pagep, void **fsdata) | |
1da177e4 | 668 | { |
155130a4 CH |
669 | return block_write_begin(mapping, pos, len, flags, pagep, |
670 | blkdev_get_block); | |
1da177e4 LT |
671 | } |
672 | ||
6272b5a5 NP |
673 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
674 | loff_t pos, unsigned len, unsigned copied, | |
675 | struct page *page, void *fsdata) | |
1da177e4 | 676 | { |
6272b5a5 NP |
677 | int ret; |
678 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
679 | ||
680 | unlock_page(page); | |
09cbfeaf | 681 | put_page(page); |
6272b5a5 NP |
682 | |
683 | return ret; | |
1da177e4 LT |
684 | } |
685 | ||
686 | /* | |
687 | * private llseek: | |
496ad9aa | 688 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
689 | * so we compute the size by hand (just as in block_read/write above) |
690 | */ | |
965c8e59 | 691 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 692 | { |
4ebb16ca | 693 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
694 | loff_t retval; |
695 | ||
5955102c | 696 | inode_lock(bd_inode); |
5d48f3a2 | 697 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 698 | inode_unlock(bd_inode); |
1da177e4 LT |
699 | return retval; |
700 | } | |
701 | ||
02c24a82 | 702 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 703 | { |
4ebb16ca | 704 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 705 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 706 | int error; |
da5aa861 | 707 | |
372cf243 | 708 | error = file_write_and_wait_range(filp, start, end); |
da5aa861 RW |
709 | if (error) |
710 | return error; | |
ab0a9735 | 711 | |
b8af67e2 AB |
712 | /* |
713 | * There is no need to serialise calls to blkdev_issue_flush with | |
714 | * i_mutex and doing so causes performance issues with concurrent | |
715 | * O_SYNC writers to a block device. | |
716 | */ | |
dd3932ed | 717 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
718 | if (error == -EOPNOTSUPP) |
719 | error = 0; | |
b8af67e2 | 720 | |
ab0a9735 | 721 | return error; |
1da177e4 | 722 | } |
b1dd3b28 | 723 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 724 | |
47a191fd MW |
725 | /** |
726 | * bdev_read_page() - Start reading a page from a block device | |
727 | * @bdev: The device to read the page from | |
728 | * @sector: The offset on the device to read the page to (need not be aligned) | |
729 | * @page: The page to read | |
730 | * | |
731 | * On entry, the page should be locked. It will be unlocked when the page | |
732 | * has been read. If the block driver implements rw_page synchronously, | |
733 | * that will be true on exit from this function, but it need not be. | |
734 | * | |
735 | * Errors returned by this function are usually "soft", eg out of memory, or | |
736 | * queue full; callers should try a different route to read this page rather | |
737 | * than propagate an error back up the stack. | |
738 | * | |
739 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
740 | */ | |
741 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
742 | struct page *page) | |
743 | { | |
744 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
745 | int result = -EOPNOTSUPP; |
746 | ||
f68eb1e7 | 747 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
748 | return result; |
749 | ||
3a0a5299 | 750 | result = blk_queue_enter(bdev->bd_queue, 0); |
2e6edc95 DW |
751 | if (result) |
752 | return result; | |
3f289dcb TH |
753 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
754 | REQ_OP_READ); | |
2e6edc95 DW |
755 | blk_queue_exit(bdev->bd_queue); |
756 | return result; | |
47a191fd MW |
757 | } |
758 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
759 | ||
760 | /** | |
761 | * bdev_write_page() - Start writing a page to a block device | |
762 | * @bdev: The device to write the page to | |
763 | * @sector: The offset on the device to write the page to (need not be aligned) | |
764 | * @page: The page to write | |
765 | * @wbc: The writeback_control for the write | |
766 | * | |
767 | * On entry, the page should be locked and not currently under writeback. | |
768 | * On exit, if the write started successfully, the page will be unlocked and | |
769 | * under writeback. If the write failed already (eg the driver failed to | |
770 | * queue the page to the device), the page will still be locked. If the | |
771 | * caller is a ->writepage implementation, it will need to unlock the page. | |
772 | * | |
773 | * Errors returned by this function are usually "soft", eg out of memory, or | |
774 | * queue full; callers should try a different route to write this page rather | |
775 | * than propagate an error back up the stack. | |
776 | * | |
777 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
778 | */ | |
779 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
780 | struct page *page, struct writeback_control *wbc) | |
781 | { | |
782 | int result; | |
47a191fd | 783 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
2e6edc95 | 784 | |
f68eb1e7 | 785 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 786 | return -EOPNOTSUPP; |
3a0a5299 | 787 | result = blk_queue_enter(bdev->bd_queue, 0); |
2e6edc95 DW |
788 | if (result) |
789 | return result; | |
790 | ||
47a191fd | 791 | set_page_writeback(page); |
3f289dcb TH |
792 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
793 | REQ_OP_WRITE); | |
f892760a | 794 | if (result) { |
47a191fd | 795 | end_page_writeback(page); |
f892760a MW |
796 | } else { |
797 | clean_page_buffers(page); | |
47a191fd | 798 | unlock_page(page); |
f892760a | 799 | } |
2e6edc95 | 800 | blk_queue_exit(bdev->bd_queue); |
47a191fd MW |
801 | return result; |
802 | } | |
803 | EXPORT_SYMBOL_GPL(bdev_write_page); | |
804 | ||
1da177e4 LT |
805 | /* |
806 | * pseudo-fs | |
807 | */ | |
808 | ||
809 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 810 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
811 | |
812 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
813 | { | |
e94b1766 | 814 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
815 | if (!ei) |
816 | return NULL; | |
817 | return &ei->vfs_inode; | |
818 | } | |
819 | ||
41149cb0 | 820 | static void bdev_free_inode(struct inode *inode) |
1da177e4 | 821 | { |
41149cb0 | 822 | kmem_cache_free(bdev_cachep, BDEV_I(inode)); |
fa0d7e3d NP |
823 | } |
824 | ||
51cc5068 | 825 | static void init_once(void *foo) |
1da177e4 LT |
826 | { |
827 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
828 | struct block_device *bdev = &ei->bdev; | |
829 | ||
a35afb83 CL |
830 | memset(bdev, 0, sizeof(*bdev)); |
831 | mutex_init(&bdev->bd_mutex); | |
a35afb83 | 832 | INIT_LIST_HEAD(&bdev->bd_list); |
49731baa TH |
833 | #ifdef CONFIG_SYSFS |
834 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
835 | #endif | |
a5a79d00 | 836 | bdev->bd_bdi = &noop_backing_dev_info; |
a35afb83 | 837 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
838 | /* Initialize mutex for freeze. */ |
839 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
840 | } |
841 | ||
b57922d9 | 842 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
843 | { |
844 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
91b0abe3 | 845 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 846 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 847 | clear_inode(inode); |
1da177e4 | 848 | spin_lock(&bdev_lock); |
1da177e4 LT |
849 | list_del_init(&bdev->bd_list); |
850 | spin_unlock(&bdev_lock); | |
f759741d JK |
851 | /* Detach inode from wb early as bdi_put() may free bdi->wb */ |
852 | inode_detach_wb(inode); | |
a5a79d00 | 853 | if (bdev->bd_bdi != &noop_backing_dev_info) { |
b1d2dc56 | 854 | bdi_put(bdev->bd_bdi); |
a5a79d00 JK |
855 | bdev->bd_bdi = &noop_backing_dev_info; |
856 | } | |
1da177e4 LT |
857 | } |
858 | ||
ee9b6d61 | 859 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
860 | .statfs = simple_statfs, |
861 | .alloc_inode = bdev_alloc_inode, | |
41149cb0 | 862 | .free_inode = bdev_free_inode, |
1da177e4 | 863 | .drop_inode = generic_delete_inode, |
b57922d9 | 864 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
865 | }; |
866 | ||
9030d16e | 867 | static int bd_init_fs_context(struct fs_context *fc) |
1da177e4 | 868 | { |
9030d16e DH |
869 | struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC); |
870 | if (!ctx) | |
871 | return -ENOMEM; | |
872 | fc->s_iflags |= SB_I_CGROUPWB; | |
873 | ctx->ops = &bdev_sops; | |
874 | return 0; | |
1da177e4 LT |
875 | } |
876 | ||
877 | static struct file_system_type bd_type = { | |
878 | .name = "bdev", | |
9030d16e | 879 | .init_fs_context = bd_init_fs_context, |
1da177e4 LT |
880 | .kill_sb = kill_anon_super, |
881 | }; | |
882 | ||
a212b105 TH |
883 | struct super_block *blockdev_superblock __read_mostly; |
884 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
885 | |
886 | void __init bdev_cache_init(void) | |
887 | { | |
888 | int err; | |
ace8577a | 889 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 890 | |
1da177e4 | 891 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 892 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 893 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 894 | init_once); |
1da177e4 LT |
895 | err = register_filesystem(&bd_type); |
896 | if (err) | |
897 | panic("Cannot register bdev pseudo-fs"); | |
898 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
899 | if (IS_ERR(bd_mnt)) |
900 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 901 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
902 | } |
903 | ||
904 | /* | |
905 | * Most likely _very_ bad one - but then it's hardly critical for small | |
906 | * /dev and can be fixed when somebody will need really large one. | |
907 | * Keep in mind that it will be fed through icache hash function too. | |
908 | */ | |
909 | static inline unsigned long hash(dev_t dev) | |
910 | { | |
911 | return MAJOR(dev)+MINOR(dev); | |
912 | } | |
913 | ||
914 | static int bdev_test(struct inode *inode, void *data) | |
915 | { | |
916 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
917 | } | |
918 | ||
919 | static int bdev_set(struct inode *inode, void *data) | |
920 | { | |
921 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
922 | return 0; | |
923 | } | |
924 | ||
925 | static LIST_HEAD(all_bdevs); | |
926 | ||
f44f1ab5 JK |
927 | /* |
928 | * If there is a bdev inode for this device, unhash it so that it gets evicted | |
929 | * as soon as last inode reference is dropped. | |
930 | */ | |
931 | void bdev_unhash_inode(dev_t dev) | |
932 | { | |
933 | struct inode *inode; | |
934 | ||
935 | inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev); | |
936 | if (inode) { | |
937 | remove_inode_hash(inode); | |
938 | iput(inode); | |
939 | } | |
940 | } | |
941 | ||
1da177e4 LT |
942 | struct block_device *bdget(dev_t dev) |
943 | { | |
944 | struct block_device *bdev; | |
945 | struct inode *inode; | |
946 | ||
c2acf7b9 | 947 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
948 | bdev_test, bdev_set, &dev); |
949 | ||
950 | if (!inode) | |
951 | return NULL; | |
952 | ||
953 | bdev = &BDEV_I(inode)->bdev; | |
954 | ||
955 | if (inode->i_state & I_NEW) { | |
956 | bdev->bd_contains = NULL; | |
782b94cd | 957 | bdev->bd_super = NULL; |
1da177e4 | 958 | bdev->bd_inode = inode; |
93407472 | 959 | bdev->bd_block_size = i_blocksize(inode); |
1da177e4 LT |
960 | bdev->bd_part_count = 0; |
961 | bdev->bd_invalidated = 0; | |
962 | inode->i_mode = S_IFBLK; | |
963 | inode->i_rdev = dev; | |
964 | inode->i_bdev = bdev; | |
965 | inode->i_data.a_ops = &def_blk_aops; | |
966 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
967 | spin_lock(&bdev_lock); |
968 | list_add(&bdev->bd_list, &all_bdevs); | |
969 | spin_unlock(&bdev_lock); | |
970 | unlock_new_inode(inode); | |
971 | } | |
972 | return bdev; | |
973 | } | |
974 | ||
975 | EXPORT_SYMBOL(bdget); | |
976 | ||
dddac6a7 AJ |
977 | /** |
978 | * bdgrab -- Grab a reference to an already referenced block device | |
979 | * @bdev: Block device to grab a reference to. | |
980 | */ | |
981 | struct block_device *bdgrab(struct block_device *bdev) | |
982 | { | |
7de9c6ee | 983 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
984 | return bdev; |
985 | } | |
c1681bf8 | 986 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 987 | |
1da177e4 LT |
988 | long nr_blockdev_pages(void) |
989 | { | |
203a2935 | 990 | struct block_device *bdev; |
1da177e4 LT |
991 | long ret = 0; |
992 | spin_lock(&bdev_lock); | |
203a2935 | 993 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
994 | ret += bdev->bd_inode->i_mapping->nrpages; |
995 | } | |
996 | spin_unlock(&bdev_lock); | |
997 | return ret; | |
998 | } | |
999 | ||
1000 | void bdput(struct block_device *bdev) | |
1001 | { | |
1002 | iput(bdev->bd_inode); | |
1003 | } | |
1004 | ||
1005 | EXPORT_SYMBOL(bdput); | |
1006 | ||
1007 | static struct block_device *bd_acquire(struct inode *inode) | |
1008 | { | |
1009 | struct block_device *bdev; | |
09d967c6 | 1010 | |
1da177e4 LT |
1011 | spin_lock(&bdev_lock); |
1012 | bdev = inode->i_bdev; | |
cccd9fb9 | 1013 | if (bdev && !inode_unhashed(bdev->bd_inode)) { |
ed8a9d2c | 1014 | bdgrab(bdev); |
1da177e4 LT |
1015 | spin_unlock(&bdev_lock); |
1016 | return bdev; | |
1017 | } | |
1018 | spin_unlock(&bdev_lock); | |
09d967c6 | 1019 | |
cccd9fb9 JK |
1020 | /* |
1021 | * i_bdev references block device inode that was already shut down | |
1022 | * (corresponding device got removed). Remove the reference and look | |
1023 | * up block device inode again just in case new device got | |
1024 | * reestablished under the same device number. | |
1025 | */ | |
1026 | if (bdev) | |
1027 | bd_forget(inode); | |
1028 | ||
1da177e4 LT |
1029 | bdev = bdget(inode->i_rdev); |
1030 | if (bdev) { | |
1031 | spin_lock(&bdev_lock); | |
09d967c6 OH |
1032 | if (!inode->i_bdev) { |
1033 | /* | |
7de9c6ee | 1034 | * We take an additional reference to bd_inode, |
09d967c6 OH |
1035 | * and it's released in clear_inode() of inode. |
1036 | * So, we can access it via ->i_mapping always | |
1037 | * without igrab(). | |
1038 | */ | |
ed8a9d2c | 1039 | bdgrab(bdev); |
09d967c6 OH |
1040 | inode->i_bdev = bdev; |
1041 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
09d967c6 | 1042 | } |
1da177e4 LT |
1043 | spin_unlock(&bdev_lock); |
1044 | } | |
1045 | return bdev; | |
1046 | } | |
1047 | ||
1048 | /* Call when you free inode */ | |
1049 | ||
1050 | void bd_forget(struct inode *inode) | |
1051 | { | |
09d967c6 OH |
1052 | struct block_device *bdev = NULL; |
1053 | ||
1da177e4 | 1054 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
1055 | if (!sb_is_blkdev_sb(inode->i_sb)) |
1056 | bdev = inode->i_bdev; | |
a4a4f943 AV |
1057 | inode->i_bdev = NULL; |
1058 | inode->i_mapping = &inode->i_data; | |
1da177e4 | 1059 | spin_unlock(&bdev_lock); |
09d967c6 OH |
1060 | |
1061 | if (bdev) | |
ed8a9d2c | 1062 | bdput(bdev); |
1da177e4 LT |
1063 | } |
1064 | ||
1a3cbbc5 TH |
1065 | /** |
1066 | * bd_may_claim - test whether a block device can be claimed | |
1067 | * @bdev: block device of interest | |
1068 | * @whole: whole block device containing @bdev, may equal @bdev | |
1069 | * @holder: holder trying to claim @bdev | |
1070 | * | |
25985edc | 1071 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
1072 | * |
1073 | * CONTEXT: | |
1074 | * spin_lock(&bdev_lock). | |
1075 | * | |
1076 | * RETURNS: | |
1077 | * %true if @bdev can be claimed, %false otherwise. | |
1078 | */ | |
1079 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
1080 | void *holder) | |
1da177e4 | 1081 | { |
1da177e4 | 1082 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 1083 | return true; /* already a holder */ |
1da177e4 | 1084 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 1085 | return false; /* held by someone else */ |
bcc7f5b4 | 1086 | else if (whole == bdev) |
1a3cbbc5 | 1087 | return true; /* is a whole device which isn't held */ |
1da177e4 | 1088 | |
e525fd89 | 1089 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
1090 | return true; /* is a partition of a device that is being partitioned */ |
1091 | else if (whole->bd_holder != NULL) | |
1092 | return false; /* is a partition of a held device */ | |
1da177e4 | 1093 | else |
1a3cbbc5 TH |
1094 | return true; /* is a partition of an un-held device */ |
1095 | } | |
1096 | ||
6b4517a7 TH |
1097 | /** |
1098 | * bd_prepare_to_claim - prepare to claim a block device | |
1099 | * @bdev: block device of interest | |
1100 | * @whole: the whole device containing @bdev, may equal @bdev | |
1101 | * @holder: holder trying to claim @bdev | |
1102 | * | |
1103 | * Prepare to claim @bdev. This function fails if @bdev is already | |
1104 | * claimed by another holder and waits if another claiming is in | |
1105 | * progress. This function doesn't actually claim. On successful | |
1106 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
1107 | * | |
1108 | * CONTEXT: | |
1109 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
1110 | * it multiple times. | |
1111 | * | |
1112 | * RETURNS: | |
1113 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
1114 | */ | |
1115 | static int bd_prepare_to_claim(struct block_device *bdev, | |
1116 | struct block_device *whole, void *holder) | |
1117 | { | |
1118 | retry: | |
1119 | /* if someone else claimed, fail */ | |
1120 | if (!bd_may_claim(bdev, whole, holder)) | |
1121 | return -EBUSY; | |
1122 | ||
e75aa858 TH |
1123 | /* if claiming is already in progress, wait for it to finish */ |
1124 | if (whole->bd_claiming) { | |
6b4517a7 TH |
1125 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
1126 | DEFINE_WAIT(wait); | |
1127 | ||
1128 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
1129 | spin_unlock(&bdev_lock); | |
1130 | schedule(); | |
1131 | finish_wait(wq, &wait); | |
1132 | spin_lock(&bdev_lock); | |
1133 | goto retry; | |
1134 | } | |
1135 | ||
1136 | /* yay, all mine */ | |
1137 | return 0; | |
1138 | } | |
1139 | ||
560e7cb2 JK |
1140 | static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno) |
1141 | { | |
1142 | struct gendisk *disk = get_gendisk(bdev->bd_dev, partno); | |
1143 | ||
1144 | if (!disk) | |
1145 | return NULL; | |
1146 | /* | |
1147 | * Now that we hold gendisk reference we make sure bdev we looked up is | |
1148 | * not stale. If it is, it means device got removed and created before | |
1149 | * we looked up gendisk and we fail open in such case. Associating | |
1150 | * unhashed bdev with newly created gendisk could lead to two bdevs | |
1151 | * (and thus two independent caches) being associated with one device | |
1152 | * which is bad. | |
1153 | */ | |
1154 | if (inode_unhashed(bdev->bd_inode)) { | |
1155 | put_disk_and_module(disk); | |
1156 | return NULL; | |
1157 | } | |
1158 | return disk; | |
1159 | } | |
1160 | ||
6b4517a7 TH |
1161 | /** |
1162 | * bd_start_claiming - start claiming a block device | |
1163 | * @bdev: block device of interest | |
1164 | * @holder: holder trying to claim @bdev | |
1165 | * | |
1166 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
1167 | * exclusively and mark that an exclusive open is in progress. Each | |
1168 | * successful call to this function must be matched with a call to | |
b0018361 NP |
1169 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
1170 | * fail). | |
1171 | * | |
1172 | * This function is used to gain exclusive access to the block device | |
1173 | * without actually causing other exclusive open attempts to fail. It | |
1174 | * should be used when the open sequence itself requires exclusive | |
1175 | * access but may subsequently fail. | |
6b4517a7 TH |
1176 | * |
1177 | * CONTEXT: | |
1178 | * Might sleep. | |
1179 | * | |
1180 | * RETURNS: | |
1181 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
1182 | * value on failure. | |
1183 | */ | |
1184 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
1185 | void *holder) | |
1186 | { | |
1187 | struct gendisk *disk; | |
1188 | struct block_device *whole; | |
1189 | int partno, err; | |
1190 | ||
1191 | might_sleep(); | |
1192 | ||
1193 | /* | |
1194 | * @bdev might not have been initialized properly yet, look up | |
1195 | * and grab the outer block device the hard way. | |
1196 | */ | |
560e7cb2 | 1197 | disk = bdev_get_gendisk(bdev, &partno); |
6b4517a7 TH |
1198 | if (!disk) |
1199 | return ERR_PTR(-ENXIO); | |
1200 | ||
d4c208b8 TH |
1201 | /* |
1202 | * Normally, @bdev should equal what's returned from bdget_disk() | |
1203 | * if partno is 0; however, some drivers (floppy) use multiple | |
1204 | * bdev's for the same physical device and @bdev may be one of the | |
1205 | * aliases. Keep @bdev if partno is 0. This means claimer | |
1206 | * tracking is broken for those devices but it has always been that | |
1207 | * way. | |
1208 | */ | |
1209 | if (partno) | |
1210 | whole = bdget_disk(disk, 0); | |
1211 | else | |
1212 | whole = bdgrab(bdev); | |
1213 | ||
9df6c299 | 1214 | put_disk_and_module(disk); |
6b4517a7 TH |
1215 | if (!whole) |
1216 | return ERR_PTR(-ENOMEM); | |
1217 | ||
1218 | /* prepare to claim, if successful, mark claiming in progress */ | |
1219 | spin_lock(&bdev_lock); | |
1220 | ||
1221 | err = bd_prepare_to_claim(bdev, whole, holder); | |
1222 | if (err == 0) { | |
1223 | whole->bd_claiming = holder; | |
1224 | spin_unlock(&bdev_lock); | |
1225 | return whole; | |
1226 | } else { | |
1227 | spin_unlock(&bdev_lock); | |
1228 | bdput(whole); | |
1229 | return ERR_PTR(err); | |
1230 | } | |
1231 | } | |
1232 | ||
641dc636 | 1233 | #ifdef CONFIG_SYSFS |
49731baa TH |
1234 | struct bd_holder_disk { |
1235 | struct list_head list; | |
1236 | struct gendisk *disk; | |
1237 | int refcnt; | |
1238 | }; | |
1239 | ||
1240 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
1241 | struct gendisk *disk) | |
1242 | { | |
1243 | struct bd_holder_disk *holder; | |
1244 | ||
1245 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
1246 | if (holder->disk == disk) | |
1247 | return holder; | |
1248 | return NULL; | |
1249 | } | |
1250 | ||
4d7dd8fd | 1251 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 1252 | { |
4d7dd8fd | 1253 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
1254 | } |
1255 | ||
1256 | static void del_symlink(struct kobject *from, struct kobject *to) | |
1257 | { | |
641dc636 JN |
1258 | sysfs_remove_link(from, kobject_name(to)); |
1259 | } | |
1260 | ||
df6c0cd9 | 1261 | /** |
e09b457b TH |
1262 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
1263 | * @bdev: the claimed slave bdev | |
1264 | * @disk: the holding disk | |
df6c0cd9 | 1265 | * |
49731baa TH |
1266 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1267 | * | |
e09b457b | 1268 | * This functions creates the following sysfs symlinks. |
641dc636 | 1269 | * |
e09b457b TH |
1270 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1271 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 1272 | * |
e09b457b TH |
1273 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1274 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 1275 | * |
e09b457b TH |
1276 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1277 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 1278 | * |
e09b457b TH |
1279 | * The caller must have claimed @bdev before calling this function and |
1280 | * ensure that both @bdev and @disk are valid during the creation and | |
1281 | * lifetime of these symlinks. | |
641dc636 | 1282 | * |
e09b457b TH |
1283 | * CONTEXT: |
1284 | * Might sleep. | |
641dc636 | 1285 | * |
e09b457b TH |
1286 | * RETURNS: |
1287 | * 0 on success, -errno on failure. | |
641dc636 | 1288 | */ |
e09b457b | 1289 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 1290 | { |
49731baa | 1291 | struct bd_holder_disk *holder; |
e09b457b | 1292 | int ret = 0; |
641dc636 | 1293 | |
2e7b651d | 1294 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 1295 | |
49731baa | 1296 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 1297 | |
e09b457b TH |
1298 | /* FIXME: remove the following once add_disk() handles errors */ |
1299 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
1300 | goto out_unlock; | |
4e91672c | 1301 | |
49731baa TH |
1302 | holder = bd_find_holder_disk(bdev, disk); |
1303 | if (holder) { | |
1304 | holder->refcnt++; | |
e09b457b | 1305 | goto out_unlock; |
49731baa | 1306 | } |
641dc636 | 1307 | |
49731baa TH |
1308 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1309 | if (!holder) { | |
1310 | ret = -ENOMEM; | |
e09b457b TH |
1311 | goto out_unlock; |
1312 | } | |
641dc636 | 1313 | |
49731baa TH |
1314 | INIT_LIST_HEAD(&holder->list); |
1315 | holder->disk = disk; | |
1316 | holder->refcnt = 1; | |
1317 | ||
1318 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1319 | if (ret) | |
1320 | goto out_free; | |
1321 | ||
1322 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
1323 | if (ret) | |
1324 | goto out_del; | |
e7407d16 TH |
1325 | /* |
1326 | * bdev could be deleted beneath us which would implicitly destroy | |
1327 | * the holder directory. Hold on to it. | |
1328 | */ | |
1329 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1330 | |
1331 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1332 | goto out_unlock; | |
1333 | ||
1334 | out_del: | |
1335 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1336 | out_free: | |
1337 | kfree(holder); | |
e09b457b | 1338 | out_unlock: |
b4cf1b72 | 1339 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1340 | return ret; |
641dc636 | 1341 | } |
e09b457b | 1342 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1343 | |
49731baa TH |
1344 | /** |
1345 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1346 | * @bdev: the calimed slave bdev | |
1347 | * @disk: the holding disk | |
1348 | * | |
1349 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1350 | * | |
1351 | * CONTEXT: | |
1352 | * Might sleep. | |
1353 | */ | |
1354 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1355 | { |
49731baa | 1356 | struct bd_holder_disk *holder; |
641dc636 | 1357 | |
49731baa | 1358 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1359 | |
49731baa TH |
1360 | holder = bd_find_holder_disk(bdev, disk); |
1361 | ||
1362 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1363 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1364 | del_symlink(bdev->bd_part->holder_dir, | |
1365 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1366 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1367 | list_del_init(&holder->list); |
1368 | kfree(holder); | |
1369 | } | |
1370 | ||
1371 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1372 | } |
49731baa | 1373 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1374 | #endif |
1da177e4 | 1375 | |
56ade44b AP |
1376 | /** |
1377 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1378 | * | |
1379 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1380 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1381 | * |
1382 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1383 | * when a disk has been changed -- either by a media change or online | |
1384 | * resize. | |
1385 | */ | |
93b270f7 | 1386 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1387 | { |
93b270f7 | 1388 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1389 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1390 | "resized disk %s\n", |
1391 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b AP |
1392 | } |
1393 | ||
1394 | if (!bdev->bd_disk) | |
1395 | return; | |
d27769ec | 1396 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1397 | bdev->bd_invalidated = 1; |
1398 | } | |
1399 | ||
c3279d14 | 1400 | /** |
57d1b536 | 1401 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1402 | * @disk: struct gendisk to check |
1403 | * @bdev: struct bdev to adjust. | |
5afb7835 | 1404 | * @verbose: if %true log a message about a size change if there is any |
c3279d14 AP |
1405 | * |
1406 | * This routine checks to see if the bdev size does not match the disk size | |
849cf559 | 1407 | * and adjusts it if it differs. When shrinking the bdev size, its all caches |
1408 | * are freed. | |
c3279d14 | 1409 | */ |
5afb7835 CH |
1410 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev, |
1411 | bool verbose) | |
c3279d14 AP |
1412 | { |
1413 | loff_t disk_size, bdev_size; | |
1414 | ||
1415 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1416 | bdev_size = i_size_read(bdev->bd_inode); | |
1417 | if (disk_size != bdev_size) { | |
5afb7835 CH |
1418 | if (verbose) { |
1419 | printk(KERN_INFO | |
1420 | "%s: detected capacity change from %lld to %lld\n", | |
1421 | disk->disk_name, bdev_size, disk_size); | |
1422 | } | |
c3279d14 | 1423 | i_size_write(bdev->bd_inode, disk_size); |
849cf559 | 1424 | if (bdev_size > disk_size) |
1425 | flush_disk(bdev, false); | |
c3279d14 AP |
1426 | } |
1427 | } | |
c3279d14 | 1428 | |
0c002c2f | 1429 | /** |
57d1b536 | 1430 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1431 | * @disk: struct gendisk to be revalidated |
1432 | * | |
1433 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1434 | * call-backs. It is used to do common pre and post operations needed | |
1435 | * for all revalidate_disk operations. | |
1436 | */ | |
1437 | int revalidate_disk(struct gendisk *disk) | |
1438 | { | |
1439 | int ret = 0; | |
1440 | ||
1441 | if (disk->fops->revalidate_disk) | |
1442 | ret = disk->fops->revalidate_disk(disk); | |
c3279d14 | 1443 | |
31cb1d64 JK |
1444 | /* |
1445 | * Hidden disks don't have associated bdev so there's no point in | |
1446 | * revalidating it. | |
1447 | */ | |
1448 | if (!(disk->flags & GENHD_FL_HIDDEN)) { | |
1449 | struct block_device *bdev = bdget_disk(disk, 0); | |
1450 | ||
1451 | if (!bdev) | |
1452 | return ret; | |
1453 | ||
1454 | mutex_lock(&bdev->bd_mutex); | |
1455 | check_disk_size_change(disk, bdev, ret == 0); | |
1456 | bdev->bd_invalidated = 0; | |
1457 | mutex_unlock(&bdev->bd_mutex); | |
1458 | bdput(bdev); | |
1459 | } | |
0c002c2f AP |
1460 | return ret; |
1461 | } | |
1462 | EXPORT_SYMBOL(revalidate_disk); | |
1463 | ||
1da177e4 LT |
1464 | /* |
1465 | * This routine checks whether a removable media has been changed, | |
1466 | * and invalidates all buffer-cache-entries in that case. This | |
1467 | * is a relatively slow routine, so we have to try to minimize using | |
1468 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1469 | * is the best way of combining speed and utility, I think. | |
1470 | * People changing diskettes in the middle of an operation deserve | |
1471 | * to lose :-) | |
1472 | */ | |
1473 | int check_disk_change(struct block_device *bdev) | |
1474 | { | |
1475 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1476 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1477 | unsigned int events; |
1da177e4 | 1478 | |
77ea887e TH |
1479 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1480 | DISK_EVENT_EJECT_REQUEST); | |
1481 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1482 | return 0; |
1483 | ||
93b270f7 | 1484 | flush_disk(bdev, true); |
1da177e4 LT |
1485 | if (bdops->revalidate_disk) |
1486 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1487 | return 1; |
1488 | } | |
1489 | ||
1490 | EXPORT_SYMBOL(check_disk_change); | |
1491 | ||
1492 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1493 | { | |
5955102c | 1494 | inode_lock(bdev->bd_inode); |
d646a02a | 1495 | i_size_write(bdev->bd_inode, size); |
5955102c | 1496 | inode_unlock(bdev->bd_inode); |
1da177e4 LT |
1497 | } |
1498 | EXPORT_SYMBOL(bd_set_size); | |
1499 | ||
4385bab1 | 1500 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1501 | |
6d740cd5 PZ |
1502 | /* |
1503 | * bd_mutex locking: | |
1504 | * | |
1505 | * mutex_lock(part->bd_mutex) | |
1506 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1507 | */ | |
1508 | ||
572c4892 | 1509 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1510 | { |
1da177e4 | 1511 | struct gendisk *disk; |
7db9cfd3 | 1512 | int ret; |
cf771cb5 | 1513 | int partno; |
fe6e9c1f | 1514 | int perm = 0; |
89736653 | 1515 | bool first_open = false; |
fe6e9c1f | 1516 | |
572c4892 | 1517 | if (mode & FMODE_READ) |
fe6e9c1f | 1518 | perm |= MAY_READ; |
572c4892 | 1519 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1520 | perm |= MAY_WRITE; |
1521 | /* | |
1522 | * hooks: /n/, see "layering violations". | |
1523 | */ | |
b7300b78 CW |
1524 | if (!for_part) { |
1525 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1526 | if (ret != 0) { | |
1527 | bdput(bdev); | |
1528 | return ret; | |
1529 | } | |
82666020 | 1530 | } |
7db9cfd3 | 1531 | |
d3374825 | 1532 | restart: |
0762b8bd | 1533 | |
89f97496 | 1534 | ret = -ENXIO; |
560e7cb2 | 1535 | disk = bdev_get_gendisk(bdev, &partno); |
0762b8bd | 1536 | if (!disk) |
6e9624b8 | 1537 | goto out; |
1da177e4 | 1538 | |
69e02c59 | 1539 | disk_block_events(disk); |
6796bf54 | 1540 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 | 1541 | if (!bdev->bd_openers) { |
89736653 | 1542 | first_open = true; |
1da177e4 | 1543 | bdev->bd_disk = disk; |
87192a2a | 1544 | bdev->bd_queue = disk->queue; |
1da177e4 | 1545 | bdev->bd_contains = bdev; |
c2ee070f | 1546 | bdev->bd_partno = partno; |
03cdadb0 | 1547 | |
cf771cb5 | 1548 | if (!partno) { |
89f97496 TH |
1549 | ret = -ENXIO; |
1550 | bdev->bd_part = disk_get_part(disk, partno); | |
1551 | if (!bdev->bd_part) | |
1552 | goto out_clear; | |
1553 | ||
1196f8b8 | 1554 | ret = 0; |
1da177e4 | 1555 | if (disk->fops->open) { |
572c4892 | 1556 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1557 | if (ret == -ERESTARTSYS) { |
1558 | /* Lost a race with 'disk' being | |
1559 | * deleted, try again. | |
1560 | * See md.c | |
1561 | */ | |
1562 | disk_put_part(bdev->bd_part); | |
1563 | bdev->bd_part = NULL; | |
d3374825 | 1564 | bdev->bd_disk = NULL; |
87192a2a | 1565 | bdev->bd_queue = NULL; |
d3374825 | 1566 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1567 | disk_unblock_events(disk); |
9df6c299 | 1568 | put_disk_and_module(disk); |
d3374825 N |
1569 | goto restart; |
1570 | } | |
1da177e4 | 1571 | } |
7e69723f | 1572 | |
04906b2f | 1573 | if (!ret) { |
7e69723f | 1574 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
04906b2f JK |
1575 | set_init_blocksize(bdev); |
1576 | } | |
7e69723f | 1577 | |
1196f8b8 TH |
1578 | /* |
1579 | * If the device is invalidated, rescan partition | |
1580 | * if open succeeded or failed with -ENOMEDIUM. | |
1581 | * The latter is necessary to prevent ghost | |
1582 | * partitions on a removed medium. | |
1583 | */ | |
fe316bf2 JN |
1584 | if (bdev->bd_invalidated) { |
1585 | if (!ret) | |
1586 | rescan_partitions(disk, bdev); | |
1587 | else if (ret == -ENOMEDIUM) | |
1588 | invalidate_partitions(disk, bdev); | |
1589 | } | |
5a023cdb | 1590 | |
1196f8b8 TH |
1591 | if (ret) |
1592 | goto out_clear; | |
1da177e4 | 1593 | } else { |
1da177e4 LT |
1594 | struct block_device *whole; |
1595 | whole = bdget_disk(disk, 0); | |
1596 | ret = -ENOMEM; | |
1597 | if (!whole) | |
0762b8bd | 1598 | goto out_clear; |
37be4124 | 1599 | BUG_ON(for_part); |
572c4892 | 1600 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1601 | if (ret) |
0762b8bd | 1602 | goto out_clear; |
1da177e4 | 1603 | bdev->bd_contains = whole; |
89f97496 | 1604 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1605 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1606 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1607 | ret = -ENXIO; |
0762b8bd | 1608 | goto out_clear; |
1da177e4 | 1609 | } |
89f97496 | 1610 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
04906b2f | 1611 | set_init_blocksize(bdev); |
1da177e4 | 1612 | } |
03e26279 JK |
1613 | |
1614 | if (bdev->bd_bdi == &noop_backing_dev_info) | |
1615 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info); | |
1da177e4 | 1616 | } else { |
1da177e4 | 1617 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1618 | ret = 0; |
1619 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1620 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1621 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1622 | if (bdev->bd_invalidated) { |
1623 | if (!ret) | |
1624 | rescan_partitions(bdev->bd_disk, bdev); | |
1625 | else if (ret == -ENOMEDIUM) | |
1626 | invalidate_partitions(bdev->bd_disk, bdev); | |
1627 | } | |
1196f8b8 TH |
1628 | if (ret) |
1629 | goto out_unlock_bdev; | |
1da177e4 LT |
1630 | } |
1631 | } | |
1632 | bdev->bd_openers++; | |
37be4124 N |
1633 | if (for_part) |
1634 | bdev->bd_part_count++; | |
c039e313 | 1635 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1636 | disk_unblock_events(disk); |
89736653 JK |
1637 | /* only one opener holds refs to the module and disk */ |
1638 | if (!first_open) | |
1639 | put_disk_and_module(disk); | |
1da177e4 LT |
1640 | return 0; |
1641 | ||
0762b8bd | 1642 | out_clear: |
89f97496 | 1643 | disk_put_part(bdev->bd_part); |
1da177e4 | 1644 | bdev->bd_disk = NULL; |
0762b8bd | 1645 | bdev->bd_part = NULL; |
87192a2a | 1646 | bdev->bd_queue = NULL; |
1da177e4 | 1647 | if (bdev != bdev->bd_contains) |
572c4892 | 1648 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1649 | bdev->bd_contains = NULL; |
0762b8bd | 1650 | out_unlock_bdev: |
c039e313 | 1651 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1652 | disk_unblock_events(disk); |
9df6c299 | 1653 | put_disk_and_module(disk); |
4345caba | 1654 | out: |
0762b8bd TH |
1655 | bdput(bdev); |
1656 | ||
1da177e4 LT |
1657 | return ret; |
1658 | } | |
1659 | ||
d4d77629 TH |
1660 | /** |
1661 | * blkdev_get - open a block device | |
1662 | * @bdev: block_device to open | |
1663 | * @mode: FMODE_* mask | |
1664 | * @holder: exclusive holder identifier | |
1665 | * | |
1666 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1667 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1668 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1669 | * | |
1670 | * On success, the reference count of @bdev is unchanged. On failure, | |
1671 | * @bdev is put. | |
1672 | * | |
1673 | * CONTEXT: | |
1674 | * Might sleep. | |
1675 | * | |
1676 | * RETURNS: | |
1677 | * 0 on success, -errno on failure. | |
1678 | */ | |
e525fd89 | 1679 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1680 | { |
e525fd89 TH |
1681 | struct block_device *whole = NULL; |
1682 | int res; | |
1683 | ||
1684 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1685 | ||
1686 | if ((mode & FMODE_EXCL) && holder) { | |
1687 | whole = bd_start_claiming(bdev, holder); | |
1688 | if (IS_ERR(whole)) { | |
1689 | bdput(bdev); | |
1690 | return PTR_ERR(whole); | |
1691 | } | |
1692 | } | |
1693 | ||
1694 | res = __blkdev_get(bdev, mode, 0); | |
1695 | ||
1696 | if (whole) { | |
d4dc210f TH |
1697 | struct gendisk *disk = whole->bd_disk; |
1698 | ||
6a027eff | 1699 | /* finish claiming */ |
77ea887e | 1700 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1701 | spin_lock(&bdev_lock); |
1702 | ||
77ea887e | 1703 | if (!res) { |
6a027eff TH |
1704 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1705 | /* | |
1706 | * Note that for a whole device bd_holders | |
1707 | * will be incremented twice, and bd_holder | |
1708 | * will be set to bd_may_claim before being | |
1709 | * set to holder | |
1710 | */ | |
1711 | whole->bd_holders++; | |
1712 | whole->bd_holder = bd_may_claim; | |
1713 | bdev->bd_holders++; | |
1714 | bdev->bd_holder = holder; | |
1715 | } | |
1716 | ||
1717 | /* tell others that we're done */ | |
1718 | BUG_ON(whole->bd_claiming != holder); | |
1719 | whole->bd_claiming = NULL; | |
1720 | wake_up_bit(&whole->bd_claiming, 0); | |
1721 | ||
1722 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1723 | |
1724 | /* | |
d4dc210f TH |
1725 | * Block event polling for write claims if requested. Any |
1726 | * write holder makes the write_holder state stick until | |
1727 | * all are released. This is good enough and tracking | |
1728 | * individual writeable reference is too fragile given the | |
1729 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1730 | */ |
4c49ff3f TH |
1731 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1732 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1733 | bdev->bd_write_holder = true; |
d4dc210f | 1734 | disk_block_events(disk); |
77ea887e TH |
1735 | } |
1736 | ||
1737 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1738 | bdput(whole); |
e525fd89 TH |
1739 | } |
1740 | ||
1741 | return res; | |
37be4124 | 1742 | } |
1da177e4 LT |
1743 | EXPORT_SYMBOL(blkdev_get); |
1744 | ||
d4d77629 TH |
1745 | /** |
1746 | * blkdev_get_by_path - open a block device by name | |
1747 | * @path: path to the block device to open | |
1748 | * @mode: FMODE_* mask | |
1749 | * @holder: exclusive holder identifier | |
1750 | * | |
1751 | * Open the blockdevice described by the device file at @path. @mode | |
1752 | * and @holder are identical to blkdev_get(). | |
1753 | * | |
1754 | * On success, the returned block_device has reference count of one. | |
1755 | * | |
1756 | * CONTEXT: | |
1757 | * Might sleep. | |
1758 | * | |
1759 | * RETURNS: | |
1760 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1761 | */ | |
1762 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1763 | void *holder) | |
1764 | { | |
1765 | struct block_device *bdev; | |
1766 | int err; | |
1767 | ||
1768 | bdev = lookup_bdev(path); | |
1769 | if (IS_ERR(bdev)) | |
1770 | return bdev; | |
1771 | ||
1772 | err = blkdev_get(bdev, mode, holder); | |
1773 | if (err) | |
1774 | return ERR_PTR(err); | |
1775 | ||
e51900f7 CE |
1776 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1777 | blkdev_put(bdev, mode); | |
1778 | return ERR_PTR(-EACCES); | |
1779 | } | |
1780 | ||
d4d77629 TH |
1781 | return bdev; |
1782 | } | |
1783 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1784 | ||
1785 | /** | |
1786 | * blkdev_get_by_dev - open a block device by device number | |
1787 | * @dev: device number of block device to open | |
1788 | * @mode: FMODE_* mask | |
1789 | * @holder: exclusive holder identifier | |
1790 | * | |
1791 | * Open the blockdevice described by device number @dev. @mode and | |
1792 | * @holder are identical to blkdev_get(). | |
1793 | * | |
1794 | * Use it ONLY if you really do not have anything better - i.e. when | |
1795 | * you are behind a truly sucky interface and all you are given is a | |
1796 | * device number. _Never_ to be used for internal purposes. If you | |
1797 | * ever need it - reconsider your API. | |
1798 | * | |
1799 | * On success, the returned block_device has reference count of one. | |
1800 | * | |
1801 | * CONTEXT: | |
1802 | * Might sleep. | |
1803 | * | |
1804 | * RETURNS: | |
1805 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1806 | */ | |
1807 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1808 | { | |
1809 | struct block_device *bdev; | |
1810 | int err; | |
1811 | ||
1812 | bdev = bdget(dev); | |
1813 | if (!bdev) | |
1814 | return ERR_PTR(-ENOMEM); | |
1815 | ||
1816 | err = blkdev_get(bdev, mode, holder); | |
1817 | if (err) | |
1818 | return ERR_PTR(err); | |
1819 | ||
1820 | return bdev; | |
1821 | } | |
1822 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1823 | ||
1da177e4 LT |
1824 | static int blkdev_open(struct inode * inode, struct file * filp) |
1825 | { | |
1826 | struct block_device *bdev; | |
1da177e4 LT |
1827 | |
1828 | /* | |
1829 | * Preserve backwards compatibility and allow large file access | |
1830 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1831 | * binary needs it. We might want to drop this workaround | |
1832 | * during an unstable branch. | |
1833 | */ | |
1834 | filp->f_flags |= O_LARGEFILE; | |
1835 | ||
c35fc7a5 CH |
1836 | filp->f_mode |= FMODE_NOWAIT; |
1837 | ||
572c4892 AV |
1838 | if (filp->f_flags & O_NDELAY) |
1839 | filp->f_mode |= FMODE_NDELAY; | |
1840 | if (filp->f_flags & O_EXCL) | |
1841 | filp->f_mode |= FMODE_EXCL; | |
1842 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1843 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1844 | ||
1da177e4 | 1845 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1846 | if (bdev == NULL) |
1847 | return -ENOMEM; | |
1da177e4 | 1848 | |
572c4892 | 1849 | filp->f_mapping = bdev->bd_inode->i_mapping; |
5660e13d | 1850 | filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping); |
572c4892 | 1851 | |
e525fd89 | 1852 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1853 | } |
1854 | ||
4385bab1 | 1855 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1856 | { |
2e7b651d | 1857 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1858 | struct block_device *victim = NULL; |
2e7b651d | 1859 | |
6796bf54 | 1860 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1861 | if (for_part) |
1862 | bdev->bd_part_count--; | |
1863 | ||
2e7b651d | 1864 | if (!--bdev->bd_openers) { |
6a027eff | 1865 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1866 | sync_blockdev(bdev); |
1867 | kill_bdev(bdev); | |
43d1c0eb ID |
1868 | |
1869 | bdev_write_inode(bdev); | |
2e7b651d PZ |
1870 | } |
1871 | if (bdev->bd_contains == bdev) { | |
1872 | if (disk->fops->release) | |
db2a144b | 1873 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1874 | } |
1875 | if (!bdev->bd_openers) { | |
0762b8bd TH |
1876 | disk_put_part(bdev->bd_part); |
1877 | bdev->bd_part = NULL; | |
2e7b651d | 1878 | bdev->bd_disk = NULL; |
37be4124 N |
1879 | if (bdev != bdev->bd_contains) |
1880 | victim = bdev->bd_contains; | |
2e7b651d | 1881 | bdev->bd_contains = NULL; |
523e1d39 | 1882 | |
9df6c299 | 1883 | put_disk_and_module(disk); |
2e7b651d | 1884 | } |
2e7b651d PZ |
1885 | mutex_unlock(&bdev->bd_mutex); |
1886 | bdput(bdev); | |
37be4124 | 1887 | if (victim) |
9a1c3542 | 1888 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1889 | } |
1890 | ||
4385bab1 | 1891 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1892 | { |
85ef06d1 TH |
1893 | mutex_lock(&bdev->bd_mutex); |
1894 | ||
e525fd89 | 1895 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1896 | bool bdev_free; |
1897 | ||
1898 | /* | |
1899 | * Release a claim on the device. The holder fields | |
1900 | * are protected with bdev_lock. bd_mutex is to | |
1901 | * synchronize disk_holder unlinking. | |
1902 | */ | |
6a027eff TH |
1903 | spin_lock(&bdev_lock); |
1904 | ||
1905 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1906 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1907 | ||
1908 | /* bd_contains might point to self, check in a separate step */ | |
1909 | if ((bdev_free = !bdev->bd_holders)) | |
1910 | bdev->bd_holder = NULL; | |
1911 | if (!bdev->bd_contains->bd_holders) | |
1912 | bdev->bd_contains->bd_holder = NULL; | |
1913 | ||
1914 | spin_unlock(&bdev_lock); | |
1915 | ||
77ea887e TH |
1916 | /* |
1917 | * If this was the last claim, remove holder link and | |
1918 | * unblock evpoll if it was a write holder. | |
1919 | */ | |
85ef06d1 TH |
1920 | if (bdev_free && bdev->bd_write_holder) { |
1921 | disk_unblock_events(bdev->bd_disk); | |
1922 | bdev->bd_write_holder = false; | |
77ea887e | 1923 | } |
6936217c | 1924 | } |
77ea887e | 1925 | |
85ef06d1 TH |
1926 | /* |
1927 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1928 | * event. This is to ensure detection of media removal commanded | |
1929 | * from userland - e.g. eject(1). | |
1930 | */ | |
1931 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1932 | ||
1933 | mutex_unlock(&bdev->bd_mutex); | |
1934 | ||
4385bab1 | 1935 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1936 | } |
2e7b651d PZ |
1937 | EXPORT_SYMBOL(blkdev_put); |
1938 | ||
1da177e4 LT |
1939 | static int blkdev_close(struct inode * inode, struct file * filp) |
1940 | { | |
4ebb16ca | 1941 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1942 | blkdev_put(bdev, filp->f_mode); |
1943 | return 0; | |
1da177e4 LT |
1944 | } |
1945 | ||
bb93e3a5 | 1946 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1947 | { |
4ebb16ca | 1948 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1949 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1950 | |
1951 | /* | |
1952 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1953 | * to updated it before every ioctl. | |
1954 | */ | |
56b26add | 1955 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1956 | mode |= FMODE_NDELAY; |
1957 | else | |
1958 | mode &= ~FMODE_NDELAY; | |
1959 | ||
56b26add | 1960 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1961 | } |
1962 | ||
eef99380 CH |
1963 | /* |
1964 | * Write data to the block device. Only intended for the block device itself | |
1965 | * and the raw driver which basically is a fake block device. | |
1966 | * | |
1967 | * Does not take i_mutex for the write and thus is not for general purpose | |
1968 | * use. | |
1969 | */ | |
1456c0a8 | 1970 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1971 | { |
1972 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1973 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1974 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1975 | struct blk_plug plug; |
eef99380 | 1976 | ssize_t ret; |
5f380c7f | 1977 | |
7ec7b94a AV |
1978 | if (bdev_read_only(I_BDEV(bd_inode))) |
1979 | return -EPERM; | |
5f380c7f | 1980 | |
7ec7b94a | 1981 | if (!iov_iter_count(from)) |
5f380c7f AV |
1982 | return 0; |
1983 | ||
7ec7b94a AV |
1984 | if (iocb->ki_pos >= size) |
1985 | return -ENOSPC; | |
1986 | ||
c35fc7a5 CH |
1987 | if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) |
1988 | return -EOPNOTSUPP; | |
1989 | ||
7ec7b94a | 1990 | iov_iter_truncate(from, size - iocb->ki_pos); |
eef99380 | 1991 | |
53362a05 | 1992 | blk_start_plug(&plug); |
1456c0a8 | 1993 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
1994 | if (ret > 0) |
1995 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 1996 | blk_finish_plug(&plug); |
eef99380 CH |
1997 | return ret; |
1998 | } | |
1456c0a8 | 1999 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 2000 | |
b2de525f | 2001 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
2002 | { |
2003 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 2004 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 2005 | loff_t size = i_size_read(bd_inode); |
a886038b | 2006 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
2007 | |
2008 | if (pos >= size) | |
2009 | return 0; | |
2010 | ||
2011 | size -= pos; | |
a886038b AV |
2012 | iov_iter_truncate(to, size); |
2013 | return generic_file_read_iter(iocb, to); | |
684c9aae | 2014 | } |
b2de525f | 2015 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 2016 | |
87d8fe1e TT |
2017 | /* |
2018 | * Try to release a page associated with block device when the system | |
2019 | * is under memory pressure. | |
2020 | */ | |
2021 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
2022 | { | |
2023 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
2024 | ||
2025 | if (super && super->s_op->bdev_try_to_free_page) | |
2026 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
2027 | ||
2028 | return try_to_free_buffers(page); | |
2029 | } | |
2030 | ||
7f6d5b52 RZ |
2031 | static int blkdev_writepages(struct address_space *mapping, |
2032 | struct writeback_control *wbc) | |
2033 | { | |
7f6d5b52 RZ |
2034 | return generic_writepages(mapping, wbc); |
2035 | } | |
2036 | ||
4c54ac62 | 2037 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 2038 | .readpage = blkdev_readpage, |
447f05bb | 2039 | .readpages = blkdev_readpages, |
1da177e4 | 2040 | .writepage = blkdev_writepage, |
6272b5a5 NP |
2041 | .write_begin = blkdev_write_begin, |
2042 | .write_end = blkdev_write_end, | |
7f6d5b52 | 2043 | .writepages = blkdev_writepages, |
87d8fe1e | 2044 | .releasepage = blkdev_releasepage, |
1da177e4 | 2045 | .direct_IO = blkdev_direct_IO, |
88dbcbb3 | 2046 | .migratepage = buffer_migrate_page_norefs, |
b4597226 | 2047 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
2048 | }; |
2049 | ||
25f4c414 DW |
2050 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
2051 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
2052 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) | |
2053 | ||
2054 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, | |
2055 | loff_t len) | |
2056 | { | |
2057 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
25f4c414 DW |
2058 | struct address_space *mapping; |
2059 | loff_t end = start + len - 1; | |
2060 | loff_t isize; | |
2061 | int error; | |
2062 | ||
2063 | /* Fail if we don't recognize the flags. */ | |
2064 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) | |
2065 | return -EOPNOTSUPP; | |
2066 | ||
2067 | /* Don't go off the end of the device. */ | |
2068 | isize = i_size_read(bdev->bd_inode); | |
2069 | if (start >= isize) | |
2070 | return -EINVAL; | |
2071 | if (end >= isize) { | |
2072 | if (mode & FALLOC_FL_KEEP_SIZE) { | |
2073 | len = isize - start; | |
2074 | end = start + len - 1; | |
2075 | } else | |
2076 | return -EINVAL; | |
2077 | } | |
2078 | ||
2079 | /* | |
2080 | * Don't allow IO that isn't aligned to logical block size. | |
2081 | */ | |
2082 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) | |
2083 | return -EINVAL; | |
2084 | ||
2085 | /* Invalidate the page cache, including dirty pages. */ | |
2086 | mapping = bdev->bd_inode->i_mapping; | |
2087 | truncate_inode_pages_range(mapping, start, end); | |
2088 | ||
2089 | switch (mode) { | |
2090 | case FALLOC_FL_ZERO_RANGE: | |
2091 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: | |
2092 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, | |
ee472d83 | 2093 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP); |
25f4c414 DW |
2094 | break; |
2095 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: | |
34045129 CH |
2096 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
2097 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK); | |
25f4c414 DW |
2098 | break; |
2099 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: | |
25f4c414 DW |
2100 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
2101 | GFP_KERNEL, 0); | |
2102 | break; | |
2103 | default: | |
2104 | return -EOPNOTSUPP; | |
2105 | } | |
2106 | if (error) | |
2107 | return error; | |
2108 | ||
2109 | /* | |
2110 | * Invalidate again; if someone wandered in and dirtied a page, | |
2111 | * the caller will be given -EBUSY. The third argument is | |
2112 | * inclusive, so the rounding here is safe. | |
2113 | */ | |
2114 | return invalidate_inode_pages2_range(mapping, | |
2115 | start >> PAGE_SHIFT, | |
2116 | end >> PAGE_SHIFT); | |
2117 | } | |
2118 | ||
4b6f5d20 | 2119 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
2120 | .open = blkdev_open, |
2121 | .release = blkdev_close, | |
2122 | .llseek = block_llseek, | |
a886038b | 2123 | .read_iter = blkdev_read_iter, |
1456c0a8 | 2124 | .write_iter = blkdev_write_iter, |
eae83ce1 | 2125 | .iopoll = blkdev_iopoll, |
acc93d30 | 2126 | .mmap = generic_file_mmap, |
b1dd3b28 | 2127 | .fsync = blkdev_fsync, |
bb93e3a5 | 2128 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
2129 | #ifdef CONFIG_COMPAT |
2130 | .compat_ioctl = compat_blkdev_ioctl, | |
2131 | #endif | |
1e8b3332 | 2132 | .splice_read = generic_file_splice_read, |
8d020765 | 2133 | .splice_write = iter_file_splice_write, |
25f4c414 | 2134 | .fallocate = blkdev_fallocate, |
1da177e4 LT |
2135 | }; |
2136 | ||
2137 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
2138 | { | |
2139 | int res; | |
2140 | mm_segment_t old_fs = get_fs(); | |
2141 | set_fs(KERNEL_DS); | |
56b26add | 2142 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
2143 | set_fs(old_fs); |
2144 | return res; | |
2145 | } | |
2146 | ||
2147 | EXPORT_SYMBOL(ioctl_by_bdev); | |
2148 | ||
2149 | /** | |
2150 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 2151 | * @pathname: special file representing the block device |
1da177e4 | 2152 | * |
57d1b536 | 2153 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
2154 | * namespace if possible and return it. Return ERR_PTR(error) |
2155 | * otherwise. | |
2156 | */ | |
421748ec | 2157 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
2158 | { |
2159 | struct block_device *bdev; | |
2160 | struct inode *inode; | |
421748ec | 2161 | struct path path; |
1da177e4 LT |
2162 | int error; |
2163 | ||
421748ec | 2164 | if (!pathname || !*pathname) |
1da177e4 LT |
2165 | return ERR_PTR(-EINVAL); |
2166 | ||
421748ec | 2167 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
2168 | if (error) |
2169 | return ERR_PTR(error); | |
2170 | ||
bb668734 | 2171 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
2172 | error = -ENOTBLK; |
2173 | if (!S_ISBLK(inode->i_mode)) | |
2174 | goto fail; | |
2175 | error = -EACCES; | |
a2982cc9 | 2176 | if (!may_open_dev(&path)) |
1da177e4 LT |
2177 | goto fail; |
2178 | error = -ENOMEM; | |
2179 | bdev = bd_acquire(inode); | |
2180 | if (!bdev) | |
2181 | goto fail; | |
2182 | out: | |
421748ec | 2183 | path_put(&path); |
1da177e4 LT |
2184 | return bdev; |
2185 | fail: | |
2186 | bdev = ERR_PTR(error); | |
2187 | goto out; | |
2188 | } | |
d5686b44 | 2189 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 2190 | |
93b270f7 | 2191 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
2192 | { |
2193 | struct super_block *sb = get_super(bdev); | |
2194 | int res = 0; | |
2195 | ||
2196 | if (sb) { | |
2197 | /* | |
2198 | * no need to lock the super, get_super holds the | |
2199 | * read mutex so the filesystem cannot go away | |
2200 | * under us (->put_super runs with the write lock | |
2201 | * hold). | |
2202 | */ | |
2203 | shrink_dcache_sb(sb); | |
93b270f7 | 2204 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
2205 | drop_super(sb); |
2206 | } | |
f98393a6 | 2207 | invalidate_bdev(bdev); |
b71e8a4c DH |
2208 | return res; |
2209 | } | |
2210 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
2211 | |
2212 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
2213 | { | |
2214 | struct inode *inode, *old_inode = NULL; | |
2215 | ||
74278da9 | 2216 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2217 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
2218 | struct address_space *mapping = inode->i_mapping; | |
af309226 | 2219 | struct block_device *bdev; |
5c0d6b60 JK |
2220 | |
2221 | spin_lock(&inode->i_lock); | |
2222 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
2223 | mapping->nrpages == 0) { | |
2224 | spin_unlock(&inode->i_lock); | |
2225 | continue; | |
2226 | } | |
2227 | __iget(inode); | |
2228 | spin_unlock(&inode->i_lock); | |
74278da9 | 2229 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2230 | /* |
2231 | * We hold a reference to 'inode' so it couldn't have been | |
2232 | * removed from s_inodes list while we dropped the | |
74278da9 | 2233 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 2234 | * be holding the last reference and we cannot iput it under |
74278da9 | 2235 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
2236 | * later. |
2237 | */ | |
2238 | iput(old_inode); | |
2239 | old_inode = inode; | |
af309226 | 2240 | bdev = I_BDEV(inode); |
5c0d6b60 | 2241 | |
af309226 RV |
2242 | mutex_lock(&bdev->bd_mutex); |
2243 | if (bdev->bd_openers) | |
2244 | func(bdev, arg); | |
2245 | mutex_unlock(&bdev->bd_mutex); | |
5c0d6b60 | 2246 | |
74278da9 | 2247 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 2248 | } |
74278da9 | 2249 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2250 | iput(old_inode); |
2251 | } |