Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/block_dev.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE | |
6 | */ | |
7 | ||
1da177e4 LT |
8 | #include <linux/init.h> |
9 | #include <linux/mm.h> | |
10 | #include <linux/fcntl.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/kmod.h> | |
13 | #include <linux/major.h> | |
7db9cfd3 | 14 | #include <linux/device_cgroup.h> |
1da177e4 LT |
15 | #include <linux/highmem.h> |
16 | #include <linux/blkdev.h> | |
66114cad | 17 | #include <linux/backing-dev.h> |
1da177e4 LT |
18 | #include <linux/module.h> |
19 | #include <linux/blkpg.h> | |
b502bd11 | 20 | #include <linux/magic.h> |
b0686260 | 21 | #include <linux/dax.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> | |
28 | #include <linux/uio.h> | |
29 | #include <linux/namei.h> | |
1368c4f2 | 30 | #include <linux/log2.h> |
ff01bb48 | 31 | #include <linux/cleancache.h> |
c94c2acf | 32 | #include <linux/dax.h> |
acc93d30 | 33 | #include <linux/badblocks.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 LT |
106 | |
107 | int set_blocksize(struct block_device *bdev, int size) | |
108 | { | |
109 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 110 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
111 | return -EINVAL; |
112 | ||
113 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 114 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
115 | return -EINVAL; |
116 | ||
117 | /* Don't change the size if it is same as current */ | |
118 | if (bdev->bd_block_size != size) { | |
119 | sync_blockdev(bdev); | |
120 | bdev->bd_block_size = size; | |
121 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
122 | kill_bdev(bdev); | |
123 | } | |
124 | return 0; | |
125 | } | |
126 | ||
127 | EXPORT_SYMBOL(set_blocksize); | |
128 | ||
129 | int sb_set_blocksize(struct super_block *sb, int size) | |
130 | { | |
1da177e4 LT |
131 | if (set_blocksize(sb->s_bdev, size)) |
132 | return 0; | |
133 | /* If we get here, we know size is power of two | |
134 | * and it's value is between 512 and PAGE_SIZE */ | |
135 | sb->s_blocksize = size; | |
38885bd4 | 136 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
137 | return sb->s_blocksize; |
138 | } | |
139 | ||
140 | EXPORT_SYMBOL(sb_set_blocksize); | |
141 | ||
142 | int sb_min_blocksize(struct super_block *sb, int size) | |
143 | { | |
e1defc4f | 144 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
145 | if (size < minsize) |
146 | size = minsize; | |
147 | return sb_set_blocksize(sb, size); | |
148 | } | |
149 | ||
150 | EXPORT_SYMBOL(sb_min_blocksize); | |
151 | ||
152 | static int | |
153 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
154 | struct buffer_head *bh, int create) | |
155 | { | |
1da177e4 LT |
156 | bh->b_bdev = I_BDEV(inode); |
157 | bh->b_blocknr = iblock; | |
158 | set_buffer_mapped(bh); | |
159 | return 0; | |
160 | } | |
161 | ||
4ebb16ca DW |
162 | static struct inode *bdev_file_inode(struct file *file) |
163 | { | |
164 | return file->f_mapping->host; | |
165 | } | |
166 | ||
78250c02 JA |
167 | static unsigned int dio_bio_write_op(struct kiocb *iocb) |
168 | { | |
169 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; | |
170 | ||
171 | /* avoid the need for a I/O completion work item */ | |
172 | if (iocb->ki_flags & IOCB_DSYNC) | |
173 | op |= REQ_FUA; | |
174 | return op; | |
175 | } | |
176 | ||
189ce2b9 CH |
177 | #define DIO_INLINE_BIO_VECS 4 |
178 | ||
179 | static void blkdev_bio_end_io_simple(struct bio *bio) | |
180 | { | |
181 | struct task_struct *waiter = bio->bi_private; | |
182 | ||
183 | WRITE_ONCE(bio->bi_private, NULL); | |
0619317f | 184 | blk_wake_io_task(waiter); |
189ce2b9 CH |
185 | } |
186 | ||
187 | static ssize_t | |
188 | __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter, | |
189 | int nr_pages) | |
190 | { | |
191 | struct file *file = iocb->ki_filp; | |
192 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
72ecad22 | 193 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec; |
189ce2b9 CH |
194 | loff_t pos = iocb->ki_pos; |
195 | bool should_dirty = false; | |
196 | struct bio bio; | |
197 | ssize_t ret; | |
198 | blk_qc_t qc; | |
199 | int i; | |
200 | ||
9a794fb9 JA |
201 | if ((pos | iov_iter_alignment(iter)) & |
202 | (bdev_logical_block_size(bdev) - 1)) | |
189ce2b9 CH |
203 | return -EINVAL; |
204 | ||
72ecad22 JA |
205 | if (nr_pages <= DIO_INLINE_BIO_VECS) |
206 | vecs = inline_vecs; | |
207 | else { | |
6da2ec56 KC |
208 | vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec), |
209 | GFP_KERNEL); | |
72ecad22 JA |
210 | if (!vecs) |
211 | return -ENOMEM; | |
212 | } | |
213 | ||
3a83f467 | 214 | bio_init(&bio, vecs, nr_pages); |
74d46992 | 215 | bio_set_dev(&bio, bdev); |
4d1a4765 | 216 | bio.bi_iter.bi_sector = pos >> 9; |
45d06cf7 | 217 | bio.bi_write_hint = iocb->ki_hint; |
189ce2b9 CH |
218 | bio.bi_private = current; |
219 | bio.bi_end_io = blkdev_bio_end_io_simple; | |
074111ca | 220 | bio.bi_ioprio = iocb->ki_ioprio; |
189ce2b9 CH |
221 | |
222 | ret = bio_iov_iter_get_pages(&bio, iter); | |
223 | if (unlikely(ret)) | |
9362dd11 | 224 | goto out; |
189ce2b9 CH |
225 | ret = bio.bi_iter.bi_size; |
226 | ||
227 | if (iov_iter_rw(iter) == READ) { | |
78250c02 | 228 | bio.bi_opf = REQ_OP_READ; |
189ce2b9 CH |
229 | if (iter_is_iovec(iter)) |
230 | should_dirty = true; | |
231 | } else { | |
78250c02 | 232 | bio.bi_opf = dio_bio_write_op(iocb); |
189ce2b9 CH |
233 | task_io_account_write(ret); |
234 | } | |
d1e36282 JA |
235 | if (iocb->ki_flags & IOCB_HIPRI) |
236 | bio.bi_opf |= REQ_HIPRI; | |
189ce2b9 CH |
237 | |
238 | qc = submit_bio(&bio); | |
239 | for (;;) { | |
849a3700 JA |
240 | __set_current_state(TASK_UNINTERRUPTIBLE); |
241 | ||
189ce2b9 CH |
242 | if (!READ_ONCE(bio.bi_private)) |
243 | break; | |
849a3700 | 244 | |
189ce2b9 | 245 | if (!(iocb->ki_flags & IOCB_HIPRI) || |
0a1b8b87 | 246 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
189ce2b9 CH |
247 | io_schedule(); |
248 | } | |
249 | __set_current_state(TASK_RUNNING); | |
250 | ||
251 | bio_for_each_segment_all(bvec, &bio, i) { | |
252 | if (should_dirty && !PageCompound(bvec->bv_page)) | |
253 | set_page_dirty_lock(bvec->bv_page); | |
254 | put_page(bvec->bv_page); | |
255 | } | |
256 | ||
4e4cbee9 | 257 | if (unlikely(bio.bi_status)) |
c6b1e36c | 258 | ret = blk_status_to_errno(bio.bi_status); |
9ae3b3f5 | 259 | |
9362dd11 MW |
260 | out: |
261 | if (vecs != inline_vecs) | |
262 | kfree(vecs); | |
263 | ||
9ae3b3f5 JA |
264 | bio_uninit(&bio); |
265 | ||
189ce2b9 CH |
266 | return ret; |
267 | } | |
268 | ||
542ff7bf CH |
269 | struct blkdev_dio { |
270 | union { | |
271 | struct kiocb *iocb; | |
272 | struct task_struct *waiter; | |
273 | }; | |
274 | size_t size; | |
275 | atomic_t ref; | |
276 | bool multi_bio : 1; | |
277 | bool should_dirty : 1; | |
278 | bool is_sync : 1; | |
279 | struct bio bio; | |
280 | }; | |
281 | ||
52190f8a | 282 | static struct bio_set blkdev_dio_pool; |
542ff7bf CH |
283 | |
284 | static void blkdev_bio_end_io(struct bio *bio) | |
285 | { | |
286 | struct blkdev_dio *dio = bio->bi_private; | |
287 | bool should_dirty = dio->should_dirty; | |
288 | ||
289 | if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) { | |
4e4cbee9 CH |
290 | if (bio->bi_status && !dio->bio.bi_status) |
291 | dio->bio.bi_status = bio->bi_status; | |
542ff7bf CH |
292 | } else { |
293 | if (!dio->is_sync) { | |
294 | struct kiocb *iocb = dio->iocb; | |
4e4cbee9 | 295 | ssize_t ret; |
542ff7bf | 296 | |
4e4cbee9 | 297 | if (likely(!dio->bio.bi_status)) { |
542ff7bf CH |
298 | ret = dio->size; |
299 | iocb->ki_pos += ret; | |
4e4cbee9 CH |
300 | } else { |
301 | ret = blk_status_to_errno(dio->bio.bi_status); | |
542ff7bf CH |
302 | } |
303 | ||
304 | dio->iocb->ki_complete(iocb, ret, 0); | |
531724ab CH |
305 | if (dio->multi_bio) |
306 | bio_put(&dio->bio); | |
542ff7bf CH |
307 | } else { |
308 | struct task_struct *waiter = dio->waiter; | |
309 | ||
310 | WRITE_ONCE(dio->waiter, NULL); | |
0619317f | 311 | blk_wake_io_task(waiter); |
542ff7bf CH |
312 | } |
313 | } | |
314 | ||
315 | if (should_dirty) { | |
316 | bio_check_pages_dirty(bio); | |
317 | } else { | |
318 | struct bio_vec *bvec; | |
319 | int i; | |
320 | ||
321 | bio_for_each_segment_all(bvec, bio, i) | |
322 | put_page(bvec->bv_page); | |
323 | bio_put(bio); | |
324 | } | |
325 | } | |
326 | ||
b2e895db | 327 | static ssize_t |
542ff7bf | 328 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages) |
b2e895db AM |
329 | { |
330 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 331 | struct inode *inode = bdev_file_inode(file); |
542ff7bf | 332 | struct block_device *bdev = I_BDEV(inode); |
64d656a1 | 333 | struct blk_plug plug; |
542ff7bf CH |
334 | struct blkdev_dio *dio; |
335 | struct bio *bio; | |
cb700eb3 | 336 | bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0; |
690e5325 | 337 | bool is_read = (iov_iter_rw(iter) == READ), is_sync; |
542ff7bf CH |
338 | loff_t pos = iocb->ki_pos; |
339 | blk_qc_t qc = BLK_QC_T_NONE; | |
36ffc6c1 | 340 | int ret = 0; |
542ff7bf | 341 | |
9a794fb9 JA |
342 | if ((pos | iov_iter_alignment(iter)) & |
343 | (bdev_logical_block_size(bdev) - 1)) | |
542ff7bf CH |
344 | return -EINVAL; |
345 | ||
52190f8a | 346 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool); |
542ff7bf CH |
347 | |
348 | dio = container_of(bio, struct blkdev_dio, bio); | |
690e5325 | 349 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
531724ab | 350 | if (dio->is_sync) { |
542ff7bf | 351 | dio->waiter = current; |
531724ab CH |
352 | bio_get(bio); |
353 | } else { | |
542ff7bf | 354 | dio->iocb = iocb; |
531724ab | 355 | } |
542ff7bf CH |
356 | |
357 | dio->size = 0; | |
358 | dio->multi_bio = false; | |
00e23707 | 359 | dio->should_dirty = is_read && iter_is_iovec(iter); |
542ff7bf | 360 | |
cb700eb3 JA |
361 | /* |
362 | * Don't plug for HIPRI/polled IO, as those should go straight | |
363 | * to issue | |
364 | */ | |
365 | if (!is_poll) | |
366 | blk_start_plug(&plug); | |
367 | ||
542ff7bf | 368 | for (;;) { |
74d46992 | 369 | bio_set_dev(bio, bdev); |
4d1a4765 | 370 | bio->bi_iter.bi_sector = pos >> 9; |
45d06cf7 | 371 | bio->bi_write_hint = iocb->ki_hint; |
542ff7bf CH |
372 | bio->bi_private = dio; |
373 | bio->bi_end_io = blkdev_bio_end_io; | |
074111ca | 374 | bio->bi_ioprio = iocb->ki_ioprio; |
542ff7bf CH |
375 | |
376 | ret = bio_iov_iter_get_pages(bio, iter); | |
377 | if (unlikely(ret)) { | |
4e4cbee9 | 378 | bio->bi_status = BLK_STS_IOERR; |
542ff7bf CH |
379 | bio_endio(bio); |
380 | break; | |
381 | } | |
382 | ||
383 | if (is_read) { | |
384 | bio->bi_opf = REQ_OP_READ; | |
385 | if (dio->should_dirty) | |
386 | bio_set_pages_dirty(bio); | |
387 | } else { | |
388 | bio->bi_opf = dio_bio_write_op(iocb); | |
389 | task_io_account_write(bio->bi_iter.bi_size); | |
390 | } | |
391 | ||
392 | dio->size += bio->bi_iter.bi_size; | |
393 | pos += bio->bi_iter.bi_size; | |
394 | ||
395 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); | |
396 | if (!nr_pages) { | |
d34513d3 JA |
397 | if (iocb->ki_flags & IOCB_HIPRI) |
398 | bio->bi_opf |= REQ_HIPRI; | |
399 | ||
542ff7bf CH |
400 | qc = submit_bio(bio); |
401 | break; | |
402 | } | |
403 | ||
404 | if (!dio->multi_bio) { | |
531724ab CH |
405 | /* |
406 | * AIO needs an extra reference to ensure the dio | |
407 | * structure which is embedded into the first bio | |
408 | * stays around. | |
409 | */ | |
410 | if (!is_sync) | |
411 | bio_get(bio); | |
542ff7bf CH |
412 | dio->multi_bio = true; |
413 | atomic_set(&dio->ref, 2); | |
414 | } else { | |
415 | atomic_inc(&dio->ref); | |
416 | } | |
417 | ||
418 | submit_bio(bio); | |
419 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
420 | } | |
cb700eb3 JA |
421 | |
422 | if (!is_poll) | |
423 | blk_finish_plug(&plug); | |
542ff7bf | 424 | |
690e5325 | 425 | if (!is_sync) |
542ff7bf CH |
426 | return -EIOCBQUEUED; |
427 | ||
428 | for (;;) { | |
849a3700 JA |
429 | __set_current_state(TASK_UNINTERRUPTIBLE); |
430 | ||
542ff7bf CH |
431 | if (!READ_ONCE(dio->waiter)) |
432 | break; | |
433 | ||
434 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
0a1b8b87 | 435 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
542ff7bf CH |
436 | io_schedule(); |
437 | } | |
438 | __set_current_state(TASK_RUNNING); | |
439 | ||
36ffc6c1 | 440 | if (!ret) |
4e4cbee9 | 441 | ret = blk_status_to_errno(dio->bio.bi_status); |
7a62a523 | 442 | if (likely(!ret)) |
542ff7bf | 443 | ret = dio->size; |
542ff7bf CH |
444 | |
445 | bio_put(&dio->bio); | |
446 | return ret; | |
447 | } | |
448 | ||
449 | static ssize_t | |
450 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) | |
451 | { | |
189ce2b9 | 452 | int nr_pages; |
b2e895db | 453 | |
72ecad22 | 454 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
189ce2b9 CH |
455 | if (!nr_pages) |
456 | return 0; | |
72ecad22 | 457 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
189ce2b9 | 458 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
542ff7bf CH |
459 | |
460 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); | |
461 | } | |
462 | ||
463 | static __init int blkdev_init(void) | |
464 | { | |
52190f8a | 465 | return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS); |
b2e895db | 466 | } |
542ff7bf | 467 | module_init(blkdev_init); |
b2e895db | 468 | |
5cee5815 JK |
469 | int __sync_blockdev(struct block_device *bdev, int wait) |
470 | { | |
471 | if (!bdev) | |
472 | return 0; | |
473 | if (!wait) | |
474 | return filemap_flush(bdev->bd_inode->i_mapping); | |
475 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
476 | } | |
477 | ||
585d3bc0 NP |
478 | /* |
479 | * Write out and wait upon all the dirty data associated with a block | |
480 | * device via its mapping. Does not take the superblock lock. | |
481 | */ | |
482 | int sync_blockdev(struct block_device *bdev) | |
483 | { | |
5cee5815 | 484 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
485 | } |
486 | EXPORT_SYMBOL(sync_blockdev); | |
487 | ||
488 | /* | |
489 | * Write out and wait upon all dirty data associated with this | |
490 | * device. Filesystem data as well as the underlying block | |
491 | * device. Takes the superblock lock. | |
492 | */ | |
493 | int fsync_bdev(struct block_device *bdev) | |
494 | { | |
495 | struct super_block *sb = get_super(bdev); | |
496 | if (sb) { | |
60b0680f | 497 | int res = sync_filesystem(sb); |
585d3bc0 NP |
498 | drop_super(sb); |
499 | return res; | |
500 | } | |
501 | return sync_blockdev(bdev); | |
502 | } | |
47e4491b | 503 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
504 | |
505 | /** | |
506 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
507 | * @bdev: blockdevice to lock | |
508 | * | |
585d3bc0 NP |
509 | * If a superblock is found on this device, we take the s_umount semaphore |
510 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
511 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
512 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
513 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
514 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
515 | * actually. | |
516 | */ | |
517 | struct super_block *freeze_bdev(struct block_device *bdev) | |
518 | { | |
519 | struct super_block *sb; | |
520 | int error = 0; | |
521 | ||
522 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
523 | if (++bdev->bd_fsfreeze_count > 1) { |
524 | /* | |
525 | * We don't even need to grab a reference - the first call | |
526 | * to freeze_bdev grab an active reference and only the last | |
527 | * thaw_bdev drops it. | |
528 | */ | |
585d3bc0 | 529 | sb = get_super(bdev); |
5bb53c0f AR |
530 | if (sb) |
531 | drop_super(sb); | |
4504230a CH |
532 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
533 | return sb; | |
534 | } | |
535 | ||
536 | sb = get_active_super(bdev); | |
537 | if (!sb) | |
538 | goto out; | |
48b6bca6 BM |
539 | if (sb->s_op->freeze_super) |
540 | error = sb->s_op->freeze_super(sb); | |
541 | else | |
542 | error = freeze_super(sb); | |
18e9e510 JB |
543 | if (error) { |
544 | deactivate_super(sb); | |
545 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 546 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 547 | return ERR_PTR(error); |
585d3bc0 | 548 | } |
18e9e510 | 549 | deactivate_super(sb); |
4504230a | 550 | out: |
585d3bc0 NP |
551 | sync_blockdev(bdev); |
552 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 553 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
554 | } |
555 | EXPORT_SYMBOL(freeze_bdev); | |
556 | ||
557 | /** | |
558 | * thaw_bdev -- unlock filesystem | |
559 | * @bdev: blockdevice to unlock | |
560 | * @sb: associated superblock | |
561 | * | |
562 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
563 | */ | |
564 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
565 | { | |
4504230a | 566 | int error = -EINVAL; |
585d3bc0 NP |
567 | |
568 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 569 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 570 | goto out; |
4504230a CH |
571 | |
572 | error = 0; | |
573 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 574 | goto out; |
4504230a CH |
575 | |
576 | if (!sb) | |
18e9e510 | 577 | goto out; |
4504230a | 578 | |
48b6bca6 BM |
579 | if (sb->s_op->thaw_super) |
580 | error = sb->s_op->thaw_super(sb); | |
581 | else | |
582 | error = thaw_super(sb); | |
997198ba | 583 | if (error) |
18e9e510 | 584 | bdev->bd_fsfreeze_count++; |
18e9e510 | 585 | out: |
585d3bc0 | 586 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
997198ba | 587 | return error; |
585d3bc0 NP |
588 | } |
589 | EXPORT_SYMBOL(thaw_bdev); | |
590 | ||
1da177e4 LT |
591 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
592 | { | |
593 | return block_write_full_page(page, blkdev_get_block, wbc); | |
594 | } | |
595 | ||
596 | static int blkdev_readpage(struct file * file, struct page * page) | |
597 | { | |
598 | return block_read_full_page(page, blkdev_get_block); | |
599 | } | |
600 | ||
447f05bb AM |
601 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
602 | struct list_head *pages, unsigned nr_pages) | |
603 | { | |
604 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
605 | } | |
606 | ||
6272b5a5 NP |
607 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
608 | loff_t pos, unsigned len, unsigned flags, | |
609 | struct page **pagep, void **fsdata) | |
1da177e4 | 610 | { |
155130a4 CH |
611 | return block_write_begin(mapping, pos, len, flags, pagep, |
612 | blkdev_get_block); | |
1da177e4 LT |
613 | } |
614 | ||
6272b5a5 NP |
615 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
616 | loff_t pos, unsigned len, unsigned copied, | |
617 | struct page *page, void *fsdata) | |
1da177e4 | 618 | { |
6272b5a5 NP |
619 | int ret; |
620 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
621 | ||
622 | unlock_page(page); | |
09cbfeaf | 623 | put_page(page); |
6272b5a5 NP |
624 | |
625 | return ret; | |
1da177e4 LT |
626 | } |
627 | ||
628 | /* | |
629 | * private llseek: | |
496ad9aa | 630 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
631 | * so we compute the size by hand (just as in block_read/write above) |
632 | */ | |
965c8e59 | 633 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 634 | { |
4ebb16ca | 635 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
636 | loff_t retval; |
637 | ||
5955102c | 638 | inode_lock(bd_inode); |
5d48f3a2 | 639 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 640 | inode_unlock(bd_inode); |
1da177e4 LT |
641 | return retval; |
642 | } | |
643 | ||
02c24a82 | 644 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 645 | { |
4ebb16ca | 646 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 647 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 648 | int error; |
da5aa861 | 649 | |
372cf243 | 650 | error = file_write_and_wait_range(filp, start, end); |
da5aa861 RW |
651 | if (error) |
652 | return error; | |
ab0a9735 | 653 | |
b8af67e2 AB |
654 | /* |
655 | * There is no need to serialise calls to blkdev_issue_flush with | |
656 | * i_mutex and doing so causes performance issues with concurrent | |
657 | * O_SYNC writers to a block device. | |
658 | */ | |
dd3932ed | 659 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
660 | if (error == -EOPNOTSUPP) |
661 | error = 0; | |
b8af67e2 | 662 | |
ab0a9735 | 663 | return error; |
1da177e4 | 664 | } |
b1dd3b28 | 665 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 666 | |
47a191fd MW |
667 | /** |
668 | * bdev_read_page() - Start reading a page from a block device | |
669 | * @bdev: The device to read the page from | |
670 | * @sector: The offset on the device to read the page to (need not be aligned) | |
671 | * @page: The page to read | |
672 | * | |
673 | * On entry, the page should be locked. It will be unlocked when the page | |
674 | * has been read. If the block driver implements rw_page synchronously, | |
675 | * that will be true on exit from this function, but it need not be. | |
676 | * | |
677 | * Errors returned by this function are usually "soft", eg out of memory, or | |
678 | * queue full; callers should try a different route to read this page rather | |
679 | * than propagate an error back up the stack. | |
680 | * | |
681 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
682 | */ | |
683 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
684 | struct page *page) | |
685 | { | |
686 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
687 | int result = -EOPNOTSUPP; |
688 | ||
f68eb1e7 | 689 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
690 | return result; |
691 | ||
3a0a5299 | 692 | result = blk_queue_enter(bdev->bd_queue, 0); |
2e6edc95 DW |
693 | if (result) |
694 | return result; | |
3f289dcb TH |
695 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
696 | REQ_OP_READ); | |
2e6edc95 DW |
697 | blk_queue_exit(bdev->bd_queue); |
698 | return result; | |
47a191fd MW |
699 | } |
700 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
701 | ||
702 | /** | |
703 | * bdev_write_page() - Start writing a page to a block device | |
704 | * @bdev: The device to write the page to | |
705 | * @sector: The offset on the device to write the page to (need not be aligned) | |
706 | * @page: The page to write | |
707 | * @wbc: The writeback_control for the write | |
708 | * | |
709 | * On entry, the page should be locked and not currently under writeback. | |
710 | * On exit, if the write started successfully, the page will be unlocked and | |
711 | * under writeback. If the write failed already (eg the driver failed to | |
712 | * queue the page to the device), the page will still be locked. If the | |
713 | * caller is a ->writepage implementation, it will need to unlock the page. | |
714 | * | |
715 | * Errors returned by this function are usually "soft", eg out of memory, or | |
716 | * queue full; callers should try a different route to write this page rather | |
717 | * than propagate an error back up the stack. | |
718 | * | |
719 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
720 | */ | |
721 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
722 | struct page *page, struct writeback_control *wbc) | |
723 | { | |
724 | int result; | |
47a191fd | 725 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
2e6edc95 | 726 | |
f68eb1e7 | 727 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 728 | return -EOPNOTSUPP; |
3a0a5299 | 729 | result = blk_queue_enter(bdev->bd_queue, 0); |
2e6edc95 DW |
730 | if (result) |
731 | return result; | |
732 | ||
47a191fd | 733 | set_page_writeback(page); |
3f289dcb TH |
734 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
735 | REQ_OP_WRITE); | |
f892760a | 736 | if (result) { |
47a191fd | 737 | end_page_writeback(page); |
f892760a MW |
738 | } else { |
739 | clean_page_buffers(page); | |
47a191fd | 740 | unlock_page(page); |
f892760a | 741 | } |
2e6edc95 | 742 | blk_queue_exit(bdev->bd_queue); |
47a191fd MW |
743 | return result; |
744 | } | |
745 | EXPORT_SYMBOL_GPL(bdev_write_page); | |
746 | ||
1da177e4 LT |
747 | /* |
748 | * pseudo-fs | |
749 | */ | |
750 | ||
751 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 752 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
753 | |
754 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
755 | { | |
e94b1766 | 756 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
757 | if (!ei) |
758 | return NULL; | |
759 | return &ei->vfs_inode; | |
760 | } | |
761 | ||
fa0d7e3d | 762 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 763 | { |
fa0d7e3d | 764 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
765 | struct bdev_inode *bdi = BDEV_I(inode); |
766 | ||
1da177e4 LT |
767 | kmem_cache_free(bdev_cachep, bdi); |
768 | } | |
769 | ||
fa0d7e3d NP |
770 | static void bdev_destroy_inode(struct inode *inode) |
771 | { | |
772 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
773 | } | |
774 | ||
51cc5068 | 775 | static void init_once(void *foo) |
1da177e4 LT |
776 | { |
777 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
778 | struct block_device *bdev = &ei->bdev; | |
779 | ||
a35afb83 CL |
780 | memset(bdev, 0, sizeof(*bdev)); |
781 | mutex_init(&bdev->bd_mutex); | |
a35afb83 | 782 | INIT_LIST_HEAD(&bdev->bd_list); |
49731baa TH |
783 | #ifdef CONFIG_SYSFS |
784 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
785 | #endif | |
a5a79d00 | 786 | bdev->bd_bdi = &noop_backing_dev_info; |
a35afb83 | 787 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
788 | /* Initialize mutex for freeze. */ |
789 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
790 | } |
791 | ||
b57922d9 | 792 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
793 | { |
794 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
91b0abe3 | 795 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 796 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 797 | clear_inode(inode); |
1da177e4 | 798 | spin_lock(&bdev_lock); |
1da177e4 LT |
799 | list_del_init(&bdev->bd_list); |
800 | spin_unlock(&bdev_lock); | |
f759741d JK |
801 | /* Detach inode from wb early as bdi_put() may free bdi->wb */ |
802 | inode_detach_wb(inode); | |
a5a79d00 | 803 | if (bdev->bd_bdi != &noop_backing_dev_info) { |
b1d2dc56 | 804 | bdi_put(bdev->bd_bdi); |
a5a79d00 JK |
805 | bdev->bd_bdi = &noop_backing_dev_info; |
806 | } | |
1da177e4 LT |
807 | } |
808 | ||
ee9b6d61 | 809 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
810 | .statfs = simple_statfs, |
811 | .alloc_inode = bdev_alloc_inode, | |
812 | .destroy_inode = bdev_destroy_inode, | |
813 | .drop_inode = generic_delete_inode, | |
b57922d9 | 814 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
815 | }; |
816 | ||
51139ada AV |
817 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
818 | int flags, const char *dev_name, void *data) | |
1da177e4 | 819 | { |
3684aa70 SL |
820 | struct dentry *dent; |
821 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); | |
e9e5e3fa | 822 | if (!IS_ERR(dent)) |
3684aa70 SL |
823 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; |
824 | return dent; | |
1da177e4 LT |
825 | } |
826 | ||
827 | static struct file_system_type bd_type = { | |
828 | .name = "bdev", | |
51139ada | 829 | .mount = bd_mount, |
1da177e4 LT |
830 | .kill_sb = kill_anon_super, |
831 | }; | |
832 | ||
a212b105 TH |
833 | struct super_block *blockdev_superblock __read_mostly; |
834 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
835 | |
836 | void __init bdev_cache_init(void) | |
837 | { | |
838 | int err; | |
ace8577a | 839 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 840 | |
1da177e4 | 841 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 842 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 843 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 844 | init_once); |
1da177e4 LT |
845 | err = register_filesystem(&bd_type); |
846 | if (err) | |
847 | panic("Cannot register bdev pseudo-fs"); | |
848 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
849 | if (IS_ERR(bd_mnt)) |
850 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 851 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
852 | } |
853 | ||
854 | /* | |
855 | * Most likely _very_ bad one - but then it's hardly critical for small | |
856 | * /dev and can be fixed when somebody will need really large one. | |
857 | * Keep in mind that it will be fed through icache hash function too. | |
858 | */ | |
859 | static inline unsigned long hash(dev_t dev) | |
860 | { | |
861 | return MAJOR(dev)+MINOR(dev); | |
862 | } | |
863 | ||
864 | static int bdev_test(struct inode *inode, void *data) | |
865 | { | |
866 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
867 | } | |
868 | ||
869 | static int bdev_set(struct inode *inode, void *data) | |
870 | { | |
871 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
872 | return 0; | |
873 | } | |
874 | ||
875 | static LIST_HEAD(all_bdevs); | |
876 | ||
f44f1ab5 JK |
877 | /* |
878 | * If there is a bdev inode for this device, unhash it so that it gets evicted | |
879 | * as soon as last inode reference is dropped. | |
880 | */ | |
881 | void bdev_unhash_inode(dev_t dev) | |
882 | { | |
883 | struct inode *inode; | |
884 | ||
885 | inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev); | |
886 | if (inode) { | |
887 | remove_inode_hash(inode); | |
888 | iput(inode); | |
889 | } | |
890 | } | |
891 | ||
1da177e4 LT |
892 | struct block_device *bdget(dev_t dev) |
893 | { | |
894 | struct block_device *bdev; | |
895 | struct inode *inode; | |
896 | ||
c2acf7b9 | 897 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
898 | bdev_test, bdev_set, &dev); |
899 | ||
900 | if (!inode) | |
901 | return NULL; | |
902 | ||
903 | bdev = &BDEV_I(inode)->bdev; | |
904 | ||
905 | if (inode->i_state & I_NEW) { | |
906 | bdev->bd_contains = NULL; | |
782b94cd | 907 | bdev->bd_super = NULL; |
1da177e4 | 908 | bdev->bd_inode = inode; |
93407472 | 909 | bdev->bd_block_size = i_blocksize(inode); |
1da177e4 LT |
910 | bdev->bd_part_count = 0; |
911 | bdev->bd_invalidated = 0; | |
912 | inode->i_mode = S_IFBLK; | |
913 | inode->i_rdev = dev; | |
914 | inode->i_bdev = bdev; | |
915 | inode->i_data.a_ops = &def_blk_aops; | |
916 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
917 | spin_lock(&bdev_lock); |
918 | list_add(&bdev->bd_list, &all_bdevs); | |
919 | spin_unlock(&bdev_lock); | |
920 | unlock_new_inode(inode); | |
921 | } | |
922 | return bdev; | |
923 | } | |
924 | ||
925 | EXPORT_SYMBOL(bdget); | |
926 | ||
dddac6a7 AJ |
927 | /** |
928 | * bdgrab -- Grab a reference to an already referenced block device | |
929 | * @bdev: Block device to grab a reference to. | |
930 | */ | |
931 | struct block_device *bdgrab(struct block_device *bdev) | |
932 | { | |
7de9c6ee | 933 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
934 | return bdev; |
935 | } | |
c1681bf8 | 936 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 937 | |
1da177e4 LT |
938 | long nr_blockdev_pages(void) |
939 | { | |
203a2935 | 940 | struct block_device *bdev; |
1da177e4 LT |
941 | long ret = 0; |
942 | spin_lock(&bdev_lock); | |
203a2935 | 943 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
944 | ret += bdev->bd_inode->i_mapping->nrpages; |
945 | } | |
946 | spin_unlock(&bdev_lock); | |
947 | return ret; | |
948 | } | |
949 | ||
950 | void bdput(struct block_device *bdev) | |
951 | { | |
952 | iput(bdev->bd_inode); | |
953 | } | |
954 | ||
955 | EXPORT_SYMBOL(bdput); | |
956 | ||
957 | static struct block_device *bd_acquire(struct inode *inode) | |
958 | { | |
959 | struct block_device *bdev; | |
09d967c6 | 960 | |
1da177e4 LT |
961 | spin_lock(&bdev_lock); |
962 | bdev = inode->i_bdev; | |
cccd9fb9 | 963 | if (bdev && !inode_unhashed(bdev->bd_inode)) { |
ed8a9d2c | 964 | bdgrab(bdev); |
1da177e4 LT |
965 | spin_unlock(&bdev_lock); |
966 | return bdev; | |
967 | } | |
968 | spin_unlock(&bdev_lock); | |
09d967c6 | 969 | |
cccd9fb9 JK |
970 | /* |
971 | * i_bdev references block device inode that was already shut down | |
972 | * (corresponding device got removed). Remove the reference and look | |
973 | * up block device inode again just in case new device got | |
974 | * reestablished under the same device number. | |
975 | */ | |
976 | if (bdev) | |
977 | bd_forget(inode); | |
978 | ||
1da177e4 LT |
979 | bdev = bdget(inode->i_rdev); |
980 | if (bdev) { | |
981 | spin_lock(&bdev_lock); | |
09d967c6 OH |
982 | if (!inode->i_bdev) { |
983 | /* | |
7de9c6ee | 984 | * We take an additional reference to bd_inode, |
09d967c6 OH |
985 | * and it's released in clear_inode() of inode. |
986 | * So, we can access it via ->i_mapping always | |
987 | * without igrab(). | |
988 | */ | |
ed8a9d2c | 989 | bdgrab(bdev); |
09d967c6 OH |
990 | inode->i_bdev = bdev; |
991 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
09d967c6 | 992 | } |
1da177e4 LT |
993 | spin_unlock(&bdev_lock); |
994 | } | |
995 | return bdev; | |
996 | } | |
997 | ||
998 | /* Call when you free inode */ | |
999 | ||
1000 | void bd_forget(struct inode *inode) | |
1001 | { | |
09d967c6 OH |
1002 | struct block_device *bdev = NULL; |
1003 | ||
1da177e4 | 1004 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
1005 | if (!sb_is_blkdev_sb(inode->i_sb)) |
1006 | bdev = inode->i_bdev; | |
a4a4f943 AV |
1007 | inode->i_bdev = NULL; |
1008 | inode->i_mapping = &inode->i_data; | |
1da177e4 | 1009 | spin_unlock(&bdev_lock); |
09d967c6 OH |
1010 | |
1011 | if (bdev) | |
ed8a9d2c | 1012 | bdput(bdev); |
1da177e4 LT |
1013 | } |
1014 | ||
1a3cbbc5 TH |
1015 | /** |
1016 | * bd_may_claim - test whether a block device can be claimed | |
1017 | * @bdev: block device of interest | |
1018 | * @whole: whole block device containing @bdev, may equal @bdev | |
1019 | * @holder: holder trying to claim @bdev | |
1020 | * | |
25985edc | 1021 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
1022 | * |
1023 | * CONTEXT: | |
1024 | * spin_lock(&bdev_lock). | |
1025 | * | |
1026 | * RETURNS: | |
1027 | * %true if @bdev can be claimed, %false otherwise. | |
1028 | */ | |
1029 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
1030 | void *holder) | |
1da177e4 | 1031 | { |
1da177e4 | 1032 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 1033 | return true; /* already a holder */ |
1da177e4 | 1034 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 1035 | return false; /* held by someone else */ |
bcc7f5b4 | 1036 | else if (whole == bdev) |
1a3cbbc5 | 1037 | return true; /* is a whole device which isn't held */ |
1da177e4 | 1038 | |
e525fd89 | 1039 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
1040 | return true; /* is a partition of a device that is being partitioned */ |
1041 | else if (whole->bd_holder != NULL) | |
1042 | return false; /* is a partition of a held device */ | |
1da177e4 | 1043 | else |
1a3cbbc5 TH |
1044 | return true; /* is a partition of an un-held device */ |
1045 | } | |
1046 | ||
6b4517a7 TH |
1047 | /** |
1048 | * bd_prepare_to_claim - prepare to claim a block device | |
1049 | * @bdev: block device of interest | |
1050 | * @whole: the whole device containing @bdev, may equal @bdev | |
1051 | * @holder: holder trying to claim @bdev | |
1052 | * | |
1053 | * Prepare to claim @bdev. This function fails if @bdev is already | |
1054 | * claimed by another holder and waits if another claiming is in | |
1055 | * progress. This function doesn't actually claim. On successful | |
1056 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
1057 | * | |
1058 | * CONTEXT: | |
1059 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
1060 | * it multiple times. | |
1061 | * | |
1062 | * RETURNS: | |
1063 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
1064 | */ | |
1065 | static int bd_prepare_to_claim(struct block_device *bdev, | |
1066 | struct block_device *whole, void *holder) | |
1067 | { | |
1068 | retry: | |
1069 | /* if someone else claimed, fail */ | |
1070 | if (!bd_may_claim(bdev, whole, holder)) | |
1071 | return -EBUSY; | |
1072 | ||
e75aa858 TH |
1073 | /* if claiming is already in progress, wait for it to finish */ |
1074 | if (whole->bd_claiming) { | |
6b4517a7 TH |
1075 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
1076 | DEFINE_WAIT(wait); | |
1077 | ||
1078 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
1079 | spin_unlock(&bdev_lock); | |
1080 | schedule(); | |
1081 | finish_wait(wq, &wait); | |
1082 | spin_lock(&bdev_lock); | |
1083 | goto retry; | |
1084 | } | |
1085 | ||
1086 | /* yay, all mine */ | |
1087 | return 0; | |
1088 | } | |
1089 | ||
560e7cb2 JK |
1090 | static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno) |
1091 | { | |
1092 | struct gendisk *disk = get_gendisk(bdev->bd_dev, partno); | |
1093 | ||
1094 | if (!disk) | |
1095 | return NULL; | |
1096 | /* | |
1097 | * Now that we hold gendisk reference we make sure bdev we looked up is | |
1098 | * not stale. If it is, it means device got removed and created before | |
1099 | * we looked up gendisk and we fail open in such case. Associating | |
1100 | * unhashed bdev with newly created gendisk could lead to two bdevs | |
1101 | * (and thus two independent caches) being associated with one device | |
1102 | * which is bad. | |
1103 | */ | |
1104 | if (inode_unhashed(bdev->bd_inode)) { | |
1105 | put_disk_and_module(disk); | |
1106 | return NULL; | |
1107 | } | |
1108 | return disk; | |
1109 | } | |
1110 | ||
6b4517a7 TH |
1111 | /** |
1112 | * bd_start_claiming - start claiming a block device | |
1113 | * @bdev: block device of interest | |
1114 | * @holder: holder trying to claim @bdev | |
1115 | * | |
1116 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
1117 | * exclusively and mark that an exclusive open is in progress. Each | |
1118 | * successful call to this function must be matched with a call to | |
b0018361 NP |
1119 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
1120 | * fail). | |
1121 | * | |
1122 | * This function is used to gain exclusive access to the block device | |
1123 | * without actually causing other exclusive open attempts to fail. It | |
1124 | * should be used when the open sequence itself requires exclusive | |
1125 | * access but may subsequently fail. | |
6b4517a7 TH |
1126 | * |
1127 | * CONTEXT: | |
1128 | * Might sleep. | |
1129 | * | |
1130 | * RETURNS: | |
1131 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
1132 | * value on failure. | |
1133 | */ | |
1134 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
1135 | void *holder) | |
1136 | { | |
1137 | struct gendisk *disk; | |
1138 | struct block_device *whole; | |
1139 | int partno, err; | |
1140 | ||
1141 | might_sleep(); | |
1142 | ||
1143 | /* | |
1144 | * @bdev might not have been initialized properly yet, look up | |
1145 | * and grab the outer block device the hard way. | |
1146 | */ | |
560e7cb2 | 1147 | disk = bdev_get_gendisk(bdev, &partno); |
6b4517a7 TH |
1148 | if (!disk) |
1149 | return ERR_PTR(-ENXIO); | |
1150 | ||
d4c208b8 TH |
1151 | /* |
1152 | * Normally, @bdev should equal what's returned from bdget_disk() | |
1153 | * if partno is 0; however, some drivers (floppy) use multiple | |
1154 | * bdev's for the same physical device and @bdev may be one of the | |
1155 | * aliases. Keep @bdev if partno is 0. This means claimer | |
1156 | * tracking is broken for those devices but it has always been that | |
1157 | * way. | |
1158 | */ | |
1159 | if (partno) | |
1160 | whole = bdget_disk(disk, 0); | |
1161 | else | |
1162 | whole = bdgrab(bdev); | |
1163 | ||
9df6c299 | 1164 | put_disk_and_module(disk); |
6b4517a7 TH |
1165 | if (!whole) |
1166 | return ERR_PTR(-ENOMEM); | |
1167 | ||
1168 | /* prepare to claim, if successful, mark claiming in progress */ | |
1169 | spin_lock(&bdev_lock); | |
1170 | ||
1171 | err = bd_prepare_to_claim(bdev, whole, holder); | |
1172 | if (err == 0) { | |
1173 | whole->bd_claiming = holder; | |
1174 | spin_unlock(&bdev_lock); | |
1175 | return whole; | |
1176 | } else { | |
1177 | spin_unlock(&bdev_lock); | |
1178 | bdput(whole); | |
1179 | return ERR_PTR(err); | |
1180 | } | |
1181 | } | |
1182 | ||
641dc636 | 1183 | #ifdef CONFIG_SYSFS |
49731baa TH |
1184 | struct bd_holder_disk { |
1185 | struct list_head list; | |
1186 | struct gendisk *disk; | |
1187 | int refcnt; | |
1188 | }; | |
1189 | ||
1190 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
1191 | struct gendisk *disk) | |
1192 | { | |
1193 | struct bd_holder_disk *holder; | |
1194 | ||
1195 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
1196 | if (holder->disk == disk) | |
1197 | return holder; | |
1198 | return NULL; | |
1199 | } | |
1200 | ||
4d7dd8fd | 1201 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 1202 | { |
4d7dd8fd | 1203 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
1204 | } |
1205 | ||
1206 | static void del_symlink(struct kobject *from, struct kobject *to) | |
1207 | { | |
641dc636 JN |
1208 | sysfs_remove_link(from, kobject_name(to)); |
1209 | } | |
1210 | ||
df6c0cd9 | 1211 | /** |
e09b457b TH |
1212 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
1213 | * @bdev: the claimed slave bdev | |
1214 | * @disk: the holding disk | |
df6c0cd9 | 1215 | * |
49731baa TH |
1216 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1217 | * | |
e09b457b | 1218 | * This functions creates the following sysfs symlinks. |
641dc636 | 1219 | * |
e09b457b TH |
1220 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1221 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 1222 | * |
e09b457b TH |
1223 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1224 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 1225 | * |
e09b457b TH |
1226 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1227 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 1228 | * |
e09b457b TH |
1229 | * The caller must have claimed @bdev before calling this function and |
1230 | * ensure that both @bdev and @disk are valid during the creation and | |
1231 | * lifetime of these symlinks. | |
641dc636 | 1232 | * |
e09b457b TH |
1233 | * CONTEXT: |
1234 | * Might sleep. | |
641dc636 | 1235 | * |
e09b457b TH |
1236 | * RETURNS: |
1237 | * 0 on success, -errno on failure. | |
641dc636 | 1238 | */ |
e09b457b | 1239 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 1240 | { |
49731baa | 1241 | struct bd_holder_disk *holder; |
e09b457b | 1242 | int ret = 0; |
641dc636 | 1243 | |
2e7b651d | 1244 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 1245 | |
49731baa | 1246 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 1247 | |
e09b457b TH |
1248 | /* FIXME: remove the following once add_disk() handles errors */ |
1249 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
1250 | goto out_unlock; | |
4e91672c | 1251 | |
49731baa TH |
1252 | holder = bd_find_holder_disk(bdev, disk); |
1253 | if (holder) { | |
1254 | holder->refcnt++; | |
e09b457b | 1255 | goto out_unlock; |
49731baa | 1256 | } |
641dc636 | 1257 | |
49731baa TH |
1258 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1259 | if (!holder) { | |
1260 | ret = -ENOMEM; | |
e09b457b TH |
1261 | goto out_unlock; |
1262 | } | |
641dc636 | 1263 | |
49731baa TH |
1264 | INIT_LIST_HEAD(&holder->list); |
1265 | holder->disk = disk; | |
1266 | holder->refcnt = 1; | |
1267 | ||
1268 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1269 | if (ret) | |
1270 | goto out_free; | |
1271 | ||
1272 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
1273 | if (ret) | |
1274 | goto out_del; | |
e7407d16 TH |
1275 | /* |
1276 | * bdev could be deleted beneath us which would implicitly destroy | |
1277 | * the holder directory. Hold on to it. | |
1278 | */ | |
1279 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1280 | |
1281 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1282 | goto out_unlock; | |
1283 | ||
1284 | out_del: | |
1285 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1286 | out_free: | |
1287 | kfree(holder); | |
e09b457b | 1288 | out_unlock: |
b4cf1b72 | 1289 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1290 | return ret; |
641dc636 | 1291 | } |
e09b457b | 1292 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1293 | |
49731baa TH |
1294 | /** |
1295 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1296 | * @bdev: the calimed slave bdev | |
1297 | * @disk: the holding disk | |
1298 | * | |
1299 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1300 | * | |
1301 | * CONTEXT: | |
1302 | * Might sleep. | |
1303 | */ | |
1304 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1305 | { |
49731baa | 1306 | struct bd_holder_disk *holder; |
641dc636 | 1307 | |
49731baa | 1308 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1309 | |
49731baa TH |
1310 | holder = bd_find_holder_disk(bdev, disk); |
1311 | ||
1312 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1313 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1314 | del_symlink(bdev->bd_part->holder_dir, | |
1315 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1316 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1317 | list_del_init(&holder->list); |
1318 | kfree(holder); | |
1319 | } | |
1320 | ||
1321 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1322 | } |
49731baa | 1323 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1324 | #endif |
1da177e4 | 1325 | |
56ade44b AP |
1326 | /** |
1327 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1328 | * | |
1329 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1330 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1331 | * |
1332 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1333 | * when a disk has been changed -- either by a media change or online | |
1334 | * resize. | |
1335 | */ | |
93b270f7 | 1336 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1337 | { |
93b270f7 | 1338 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1339 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1340 | "resized disk %s\n", |
1341 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b AP |
1342 | } |
1343 | ||
1344 | if (!bdev->bd_disk) | |
1345 | return; | |
d27769ec | 1346 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1347 | bdev->bd_invalidated = 1; |
1348 | } | |
1349 | ||
c3279d14 | 1350 | /** |
57d1b536 | 1351 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1352 | * @disk: struct gendisk to check |
1353 | * @bdev: struct bdev to adjust. | |
5afb7835 | 1354 | * @verbose: if %true log a message about a size change if there is any |
c3279d14 AP |
1355 | * |
1356 | * This routine checks to see if the bdev size does not match the disk size | |
849cf559 | 1357 | * and adjusts it if it differs. When shrinking the bdev size, its all caches |
1358 | * are freed. | |
c3279d14 | 1359 | */ |
5afb7835 CH |
1360 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev, |
1361 | bool verbose) | |
c3279d14 AP |
1362 | { |
1363 | loff_t disk_size, bdev_size; | |
1364 | ||
1365 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1366 | bdev_size = i_size_read(bdev->bd_inode); | |
1367 | if (disk_size != bdev_size) { | |
5afb7835 CH |
1368 | if (verbose) { |
1369 | printk(KERN_INFO | |
1370 | "%s: detected capacity change from %lld to %lld\n", | |
1371 | disk->disk_name, bdev_size, disk_size); | |
1372 | } | |
c3279d14 | 1373 | i_size_write(bdev->bd_inode, disk_size); |
849cf559 | 1374 | if (bdev_size > disk_size) |
1375 | flush_disk(bdev, false); | |
c3279d14 AP |
1376 | } |
1377 | } | |
c3279d14 | 1378 | |
0c002c2f | 1379 | /** |
57d1b536 | 1380 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1381 | * @disk: struct gendisk to be revalidated |
1382 | * | |
1383 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1384 | * call-backs. It is used to do common pre and post operations needed | |
1385 | * for all revalidate_disk operations. | |
1386 | */ | |
1387 | int revalidate_disk(struct gendisk *disk) | |
1388 | { | |
c3279d14 | 1389 | struct block_device *bdev; |
0c002c2f AP |
1390 | int ret = 0; |
1391 | ||
1392 | if (disk->fops->revalidate_disk) | |
1393 | ret = disk->fops->revalidate_disk(disk); | |
c3279d14 AP |
1394 | bdev = bdget_disk(disk, 0); |
1395 | if (!bdev) | |
1396 | return ret; | |
1397 | ||
1398 | mutex_lock(&bdev->bd_mutex); | |
5afb7835 | 1399 | check_disk_size_change(disk, bdev, ret == 0); |
7630b661 | 1400 | bdev->bd_invalidated = 0; |
c3279d14 AP |
1401 | mutex_unlock(&bdev->bd_mutex); |
1402 | bdput(bdev); | |
0c002c2f AP |
1403 | return ret; |
1404 | } | |
1405 | EXPORT_SYMBOL(revalidate_disk); | |
1406 | ||
1da177e4 LT |
1407 | /* |
1408 | * This routine checks whether a removable media has been changed, | |
1409 | * and invalidates all buffer-cache-entries in that case. This | |
1410 | * is a relatively slow routine, so we have to try to minimize using | |
1411 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1412 | * is the best way of combining speed and utility, I think. | |
1413 | * People changing diskettes in the middle of an operation deserve | |
1414 | * to lose :-) | |
1415 | */ | |
1416 | int check_disk_change(struct block_device *bdev) | |
1417 | { | |
1418 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1419 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1420 | unsigned int events; |
1da177e4 | 1421 | |
77ea887e TH |
1422 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1423 | DISK_EVENT_EJECT_REQUEST); | |
1424 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1425 | return 0; |
1426 | ||
93b270f7 | 1427 | flush_disk(bdev, true); |
1da177e4 LT |
1428 | if (bdops->revalidate_disk) |
1429 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1430 | return 1; |
1431 | } | |
1432 | ||
1433 | EXPORT_SYMBOL(check_disk_change); | |
1434 | ||
1435 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1436 | { | |
e1defc4f | 1437 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1438 | |
5955102c | 1439 | inode_lock(bdev->bd_inode); |
d646a02a | 1440 | i_size_write(bdev->bd_inode, size); |
5955102c | 1441 | inode_unlock(bdev->bd_inode); |
09cbfeaf | 1442 | while (bsize < PAGE_SIZE) { |
1da177e4 LT |
1443 | if (size & bsize) |
1444 | break; | |
1445 | bsize <<= 1; | |
1446 | } | |
1447 | bdev->bd_block_size = bsize; | |
1448 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1449 | } | |
1450 | EXPORT_SYMBOL(bd_set_size); | |
1451 | ||
4385bab1 | 1452 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1453 | |
6d740cd5 PZ |
1454 | /* |
1455 | * bd_mutex locking: | |
1456 | * | |
1457 | * mutex_lock(part->bd_mutex) | |
1458 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1459 | */ | |
1460 | ||
572c4892 | 1461 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1462 | { |
1da177e4 | 1463 | struct gendisk *disk; |
7db9cfd3 | 1464 | int ret; |
cf771cb5 | 1465 | int partno; |
fe6e9c1f | 1466 | int perm = 0; |
89736653 | 1467 | bool first_open = false; |
fe6e9c1f | 1468 | |
572c4892 | 1469 | if (mode & FMODE_READ) |
fe6e9c1f | 1470 | perm |= MAY_READ; |
572c4892 | 1471 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1472 | perm |= MAY_WRITE; |
1473 | /* | |
1474 | * hooks: /n/, see "layering violations". | |
1475 | */ | |
b7300b78 CW |
1476 | if (!for_part) { |
1477 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1478 | if (ret != 0) { | |
1479 | bdput(bdev); | |
1480 | return ret; | |
1481 | } | |
82666020 | 1482 | } |
7db9cfd3 | 1483 | |
d3374825 | 1484 | restart: |
0762b8bd | 1485 | |
89f97496 | 1486 | ret = -ENXIO; |
560e7cb2 | 1487 | disk = bdev_get_gendisk(bdev, &partno); |
0762b8bd | 1488 | if (!disk) |
6e9624b8 | 1489 | goto out; |
1da177e4 | 1490 | |
69e02c59 | 1491 | disk_block_events(disk); |
6796bf54 | 1492 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 | 1493 | if (!bdev->bd_openers) { |
89736653 | 1494 | first_open = true; |
1da177e4 | 1495 | bdev->bd_disk = disk; |
87192a2a | 1496 | bdev->bd_queue = disk->queue; |
1da177e4 | 1497 | bdev->bd_contains = bdev; |
c2ee070f | 1498 | bdev->bd_partno = partno; |
03cdadb0 | 1499 | |
cf771cb5 | 1500 | if (!partno) { |
89f97496 TH |
1501 | ret = -ENXIO; |
1502 | bdev->bd_part = disk_get_part(disk, partno); | |
1503 | if (!bdev->bd_part) | |
1504 | goto out_clear; | |
1505 | ||
1196f8b8 | 1506 | ret = 0; |
1da177e4 | 1507 | if (disk->fops->open) { |
572c4892 | 1508 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1509 | if (ret == -ERESTARTSYS) { |
1510 | /* Lost a race with 'disk' being | |
1511 | * deleted, try again. | |
1512 | * See md.c | |
1513 | */ | |
1514 | disk_put_part(bdev->bd_part); | |
1515 | bdev->bd_part = NULL; | |
d3374825 | 1516 | bdev->bd_disk = NULL; |
87192a2a | 1517 | bdev->bd_queue = NULL; |
d3374825 | 1518 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1519 | disk_unblock_events(disk); |
9df6c299 | 1520 | put_disk_and_module(disk); |
d3374825 N |
1521 | goto restart; |
1522 | } | |
1da177e4 | 1523 | } |
7e69723f | 1524 | |
22375701 | 1525 | if (!ret) |
7e69723f | 1526 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
7e69723f | 1527 | |
1196f8b8 TH |
1528 | /* |
1529 | * If the device is invalidated, rescan partition | |
1530 | * if open succeeded or failed with -ENOMEDIUM. | |
1531 | * The latter is necessary to prevent ghost | |
1532 | * partitions on a removed medium. | |
1533 | */ | |
fe316bf2 JN |
1534 | if (bdev->bd_invalidated) { |
1535 | if (!ret) | |
1536 | rescan_partitions(disk, bdev); | |
1537 | else if (ret == -ENOMEDIUM) | |
1538 | invalidate_partitions(disk, bdev); | |
1539 | } | |
5a023cdb | 1540 | |
1196f8b8 TH |
1541 | if (ret) |
1542 | goto out_clear; | |
1da177e4 | 1543 | } else { |
1da177e4 LT |
1544 | struct block_device *whole; |
1545 | whole = bdget_disk(disk, 0); | |
1546 | ret = -ENOMEM; | |
1547 | if (!whole) | |
0762b8bd | 1548 | goto out_clear; |
37be4124 | 1549 | BUG_ON(for_part); |
572c4892 | 1550 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1551 | if (ret) |
0762b8bd | 1552 | goto out_clear; |
1da177e4 | 1553 | bdev->bd_contains = whole; |
89f97496 | 1554 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1555 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1556 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1557 | ret = -ENXIO; |
0762b8bd | 1558 | goto out_clear; |
1da177e4 | 1559 | } |
89f97496 | 1560 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1da177e4 | 1561 | } |
03e26279 JK |
1562 | |
1563 | if (bdev->bd_bdi == &noop_backing_dev_info) | |
1564 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info); | |
1da177e4 | 1565 | } else { |
1da177e4 | 1566 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1567 | ret = 0; |
1568 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1569 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1570 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1571 | if (bdev->bd_invalidated) { |
1572 | if (!ret) | |
1573 | rescan_partitions(bdev->bd_disk, bdev); | |
1574 | else if (ret == -ENOMEDIUM) | |
1575 | invalidate_partitions(bdev->bd_disk, bdev); | |
1576 | } | |
1196f8b8 TH |
1577 | if (ret) |
1578 | goto out_unlock_bdev; | |
1da177e4 LT |
1579 | } |
1580 | } | |
1581 | bdev->bd_openers++; | |
37be4124 N |
1582 | if (for_part) |
1583 | bdev->bd_part_count++; | |
c039e313 | 1584 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1585 | disk_unblock_events(disk); |
89736653 JK |
1586 | /* only one opener holds refs to the module and disk */ |
1587 | if (!first_open) | |
1588 | put_disk_and_module(disk); | |
1da177e4 LT |
1589 | return 0; |
1590 | ||
0762b8bd | 1591 | out_clear: |
89f97496 | 1592 | disk_put_part(bdev->bd_part); |
1da177e4 | 1593 | bdev->bd_disk = NULL; |
0762b8bd | 1594 | bdev->bd_part = NULL; |
87192a2a | 1595 | bdev->bd_queue = NULL; |
1da177e4 | 1596 | if (bdev != bdev->bd_contains) |
572c4892 | 1597 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1598 | bdev->bd_contains = NULL; |
0762b8bd | 1599 | out_unlock_bdev: |
c039e313 | 1600 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1601 | disk_unblock_events(disk); |
9df6c299 | 1602 | put_disk_and_module(disk); |
4345caba | 1603 | out: |
0762b8bd TH |
1604 | bdput(bdev); |
1605 | ||
1da177e4 LT |
1606 | return ret; |
1607 | } | |
1608 | ||
d4d77629 TH |
1609 | /** |
1610 | * blkdev_get - open a block device | |
1611 | * @bdev: block_device to open | |
1612 | * @mode: FMODE_* mask | |
1613 | * @holder: exclusive holder identifier | |
1614 | * | |
1615 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1616 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1617 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1618 | * | |
1619 | * On success, the reference count of @bdev is unchanged. On failure, | |
1620 | * @bdev is put. | |
1621 | * | |
1622 | * CONTEXT: | |
1623 | * Might sleep. | |
1624 | * | |
1625 | * RETURNS: | |
1626 | * 0 on success, -errno on failure. | |
1627 | */ | |
e525fd89 | 1628 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1629 | { |
e525fd89 TH |
1630 | struct block_device *whole = NULL; |
1631 | int res; | |
1632 | ||
1633 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1634 | ||
1635 | if ((mode & FMODE_EXCL) && holder) { | |
1636 | whole = bd_start_claiming(bdev, holder); | |
1637 | if (IS_ERR(whole)) { | |
1638 | bdput(bdev); | |
1639 | return PTR_ERR(whole); | |
1640 | } | |
1641 | } | |
1642 | ||
1643 | res = __blkdev_get(bdev, mode, 0); | |
1644 | ||
1645 | if (whole) { | |
d4dc210f TH |
1646 | struct gendisk *disk = whole->bd_disk; |
1647 | ||
6a027eff | 1648 | /* finish claiming */ |
77ea887e | 1649 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1650 | spin_lock(&bdev_lock); |
1651 | ||
77ea887e | 1652 | if (!res) { |
6a027eff TH |
1653 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1654 | /* | |
1655 | * Note that for a whole device bd_holders | |
1656 | * will be incremented twice, and bd_holder | |
1657 | * will be set to bd_may_claim before being | |
1658 | * set to holder | |
1659 | */ | |
1660 | whole->bd_holders++; | |
1661 | whole->bd_holder = bd_may_claim; | |
1662 | bdev->bd_holders++; | |
1663 | bdev->bd_holder = holder; | |
1664 | } | |
1665 | ||
1666 | /* tell others that we're done */ | |
1667 | BUG_ON(whole->bd_claiming != holder); | |
1668 | whole->bd_claiming = NULL; | |
1669 | wake_up_bit(&whole->bd_claiming, 0); | |
1670 | ||
1671 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1672 | |
1673 | /* | |
d4dc210f TH |
1674 | * Block event polling for write claims if requested. Any |
1675 | * write holder makes the write_holder state stick until | |
1676 | * all are released. This is good enough and tracking | |
1677 | * individual writeable reference is too fragile given the | |
1678 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1679 | */ |
4c49ff3f TH |
1680 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1681 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1682 | bdev->bd_write_holder = true; |
d4dc210f | 1683 | disk_block_events(disk); |
77ea887e TH |
1684 | } |
1685 | ||
1686 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1687 | bdput(whole); |
e525fd89 TH |
1688 | } |
1689 | ||
1690 | return res; | |
37be4124 | 1691 | } |
1da177e4 LT |
1692 | EXPORT_SYMBOL(blkdev_get); |
1693 | ||
d4d77629 TH |
1694 | /** |
1695 | * blkdev_get_by_path - open a block device by name | |
1696 | * @path: path to the block device to open | |
1697 | * @mode: FMODE_* mask | |
1698 | * @holder: exclusive holder identifier | |
1699 | * | |
1700 | * Open the blockdevice described by the device file at @path. @mode | |
1701 | * and @holder are identical to blkdev_get(). | |
1702 | * | |
1703 | * On success, the returned block_device has reference count of one. | |
1704 | * | |
1705 | * CONTEXT: | |
1706 | * Might sleep. | |
1707 | * | |
1708 | * RETURNS: | |
1709 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1710 | */ | |
1711 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1712 | void *holder) | |
1713 | { | |
1714 | struct block_device *bdev; | |
1715 | int err; | |
1716 | ||
1717 | bdev = lookup_bdev(path); | |
1718 | if (IS_ERR(bdev)) | |
1719 | return bdev; | |
1720 | ||
1721 | err = blkdev_get(bdev, mode, holder); | |
1722 | if (err) | |
1723 | return ERR_PTR(err); | |
1724 | ||
e51900f7 CE |
1725 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1726 | blkdev_put(bdev, mode); | |
1727 | return ERR_PTR(-EACCES); | |
1728 | } | |
1729 | ||
d4d77629 TH |
1730 | return bdev; |
1731 | } | |
1732 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1733 | ||
1734 | /** | |
1735 | * blkdev_get_by_dev - open a block device by device number | |
1736 | * @dev: device number of block device to open | |
1737 | * @mode: FMODE_* mask | |
1738 | * @holder: exclusive holder identifier | |
1739 | * | |
1740 | * Open the blockdevice described by device number @dev. @mode and | |
1741 | * @holder are identical to blkdev_get(). | |
1742 | * | |
1743 | * Use it ONLY if you really do not have anything better - i.e. when | |
1744 | * you are behind a truly sucky interface and all you are given is a | |
1745 | * device number. _Never_ to be used for internal purposes. If you | |
1746 | * ever need it - reconsider your API. | |
1747 | * | |
1748 | * On success, the returned block_device has reference count of one. | |
1749 | * | |
1750 | * CONTEXT: | |
1751 | * Might sleep. | |
1752 | * | |
1753 | * RETURNS: | |
1754 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1755 | */ | |
1756 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1757 | { | |
1758 | struct block_device *bdev; | |
1759 | int err; | |
1760 | ||
1761 | bdev = bdget(dev); | |
1762 | if (!bdev) | |
1763 | return ERR_PTR(-ENOMEM); | |
1764 | ||
1765 | err = blkdev_get(bdev, mode, holder); | |
1766 | if (err) | |
1767 | return ERR_PTR(err); | |
1768 | ||
1769 | return bdev; | |
1770 | } | |
1771 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1772 | ||
1da177e4 LT |
1773 | static int blkdev_open(struct inode * inode, struct file * filp) |
1774 | { | |
1775 | struct block_device *bdev; | |
1da177e4 LT |
1776 | |
1777 | /* | |
1778 | * Preserve backwards compatibility and allow large file access | |
1779 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1780 | * binary needs it. We might want to drop this workaround | |
1781 | * during an unstable branch. | |
1782 | */ | |
1783 | filp->f_flags |= O_LARGEFILE; | |
1784 | ||
c35fc7a5 CH |
1785 | filp->f_mode |= FMODE_NOWAIT; |
1786 | ||
572c4892 AV |
1787 | if (filp->f_flags & O_NDELAY) |
1788 | filp->f_mode |= FMODE_NDELAY; | |
1789 | if (filp->f_flags & O_EXCL) | |
1790 | filp->f_mode |= FMODE_EXCL; | |
1791 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1792 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1793 | ||
1da177e4 | 1794 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1795 | if (bdev == NULL) |
1796 | return -ENOMEM; | |
1da177e4 | 1797 | |
572c4892 | 1798 | filp->f_mapping = bdev->bd_inode->i_mapping; |
5660e13d | 1799 | filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping); |
572c4892 | 1800 | |
e525fd89 | 1801 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1802 | } |
1803 | ||
4385bab1 | 1804 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1805 | { |
2e7b651d | 1806 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1807 | struct block_device *victim = NULL; |
2e7b651d | 1808 | |
6796bf54 | 1809 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1810 | if (for_part) |
1811 | bdev->bd_part_count--; | |
1812 | ||
2e7b651d | 1813 | if (!--bdev->bd_openers) { |
6a027eff | 1814 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1815 | sync_blockdev(bdev); |
1816 | kill_bdev(bdev); | |
43d1c0eb ID |
1817 | |
1818 | bdev_write_inode(bdev); | |
2e7b651d PZ |
1819 | } |
1820 | if (bdev->bd_contains == bdev) { | |
1821 | if (disk->fops->release) | |
db2a144b | 1822 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1823 | } |
1824 | if (!bdev->bd_openers) { | |
0762b8bd TH |
1825 | disk_put_part(bdev->bd_part); |
1826 | bdev->bd_part = NULL; | |
2e7b651d | 1827 | bdev->bd_disk = NULL; |
37be4124 N |
1828 | if (bdev != bdev->bd_contains) |
1829 | victim = bdev->bd_contains; | |
2e7b651d | 1830 | bdev->bd_contains = NULL; |
523e1d39 | 1831 | |
9df6c299 | 1832 | put_disk_and_module(disk); |
2e7b651d | 1833 | } |
2e7b651d PZ |
1834 | mutex_unlock(&bdev->bd_mutex); |
1835 | bdput(bdev); | |
37be4124 | 1836 | if (victim) |
9a1c3542 | 1837 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1838 | } |
1839 | ||
4385bab1 | 1840 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1841 | { |
85ef06d1 TH |
1842 | mutex_lock(&bdev->bd_mutex); |
1843 | ||
e525fd89 | 1844 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1845 | bool bdev_free; |
1846 | ||
1847 | /* | |
1848 | * Release a claim on the device. The holder fields | |
1849 | * are protected with bdev_lock. bd_mutex is to | |
1850 | * synchronize disk_holder unlinking. | |
1851 | */ | |
6a027eff TH |
1852 | spin_lock(&bdev_lock); |
1853 | ||
1854 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1855 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1856 | ||
1857 | /* bd_contains might point to self, check in a separate step */ | |
1858 | if ((bdev_free = !bdev->bd_holders)) | |
1859 | bdev->bd_holder = NULL; | |
1860 | if (!bdev->bd_contains->bd_holders) | |
1861 | bdev->bd_contains->bd_holder = NULL; | |
1862 | ||
1863 | spin_unlock(&bdev_lock); | |
1864 | ||
77ea887e TH |
1865 | /* |
1866 | * If this was the last claim, remove holder link and | |
1867 | * unblock evpoll if it was a write holder. | |
1868 | */ | |
85ef06d1 TH |
1869 | if (bdev_free && bdev->bd_write_holder) { |
1870 | disk_unblock_events(bdev->bd_disk); | |
1871 | bdev->bd_write_holder = false; | |
77ea887e | 1872 | } |
6936217c | 1873 | } |
77ea887e | 1874 | |
85ef06d1 TH |
1875 | /* |
1876 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1877 | * event. This is to ensure detection of media removal commanded | |
1878 | * from userland - e.g. eject(1). | |
1879 | */ | |
1880 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1881 | ||
1882 | mutex_unlock(&bdev->bd_mutex); | |
1883 | ||
4385bab1 | 1884 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1885 | } |
2e7b651d PZ |
1886 | EXPORT_SYMBOL(blkdev_put); |
1887 | ||
1da177e4 LT |
1888 | static int blkdev_close(struct inode * inode, struct file * filp) |
1889 | { | |
4ebb16ca | 1890 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1891 | blkdev_put(bdev, filp->f_mode); |
1892 | return 0; | |
1da177e4 LT |
1893 | } |
1894 | ||
bb93e3a5 | 1895 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1896 | { |
4ebb16ca | 1897 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1898 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1899 | |
1900 | /* | |
1901 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1902 | * to updated it before every ioctl. | |
1903 | */ | |
56b26add | 1904 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1905 | mode |= FMODE_NDELAY; |
1906 | else | |
1907 | mode &= ~FMODE_NDELAY; | |
1908 | ||
56b26add | 1909 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1910 | } |
1911 | ||
eef99380 CH |
1912 | /* |
1913 | * Write data to the block device. Only intended for the block device itself | |
1914 | * and the raw driver which basically is a fake block device. | |
1915 | * | |
1916 | * Does not take i_mutex for the write and thus is not for general purpose | |
1917 | * use. | |
1918 | */ | |
1456c0a8 | 1919 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1920 | { |
1921 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1922 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1923 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1924 | struct blk_plug plug; |
eef99380 | 1925 | ssize_t ret; |
5f380c7f | 1926 | |
7ec7b94a AV |
1927 | if (bdev_read_only(I_BDEV(bd_inode))) |
1928 | return -EPERM; | |
5f380c7f | 1929 | |
7ec7b94a | 1930 | if (!iov_iter_count(from)) |
5f380c7f AV |
1931 | return 0; |
1932 | ||
7ec7b94a AV |
1933 | if (iocb->ki_pos >= size) |
1934 | return -ENOSPC; | |
1935 | ||
c35fc7a5 CH |
1936 | if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) |
1937 | return -EOPNOTSUPP; | |
1938 | ||
7ec7b94a | 1939 | iov_iter_truncate(from, size - iocb->ki_pos); |
eef99380 | 1940 | |
53362a05 | 1941 | blk_start_plug(&plug); |
1456c0a8 | 1942 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
1943 | if (ret > 0) |
1944 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 1945 | blk_finish_plug(&plug); |
eef99380 CH |
1946 | return ret; |
1947 | } | |
1456c0a8 | 1948 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1949 | |
b2de525f | 1950 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1951 | { |
1952 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1953 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 1954 | loff_t size = i_size_read(bd_inode); |
a886038b | 1955 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1956 | |
1957 | if (pos >= size) | |
1958 | return 0; | |
1959 | ||
1960 | size -= pos; | |
a886038b AV |
1961 | iov_iter_truncate(to, size); |
1962 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1963 | } |
b2de525f | 1964 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1965 | |
87d8fe1e TT |
1966 | /* |
1967 | * Try to release a page associated with block device when the system | |
1968 | * is under memory pressure. | |
1969 | */ | |
1970 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1971 | { | |
1972 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1973 | ||
1974 | if (super && super->s_op->bdev_try_to_free_page) | |
1975 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1976 | ||
1977 | return try_to_free_buffers(page); | |
1978 | } | |
1979 | ||
7f6d5b52 RZ |
1980 | static int blkdev_writepages(struct address_space *mapping, |
1981 | struct writeback_control *wbc) | |
1982 | { | |
7f6d5b52 RZ |
1983 | return generic_writepages(mapping, wbc); |
1984 | } | |
1985 | ||
4c54ac62 | 1986 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1987 | .readpage = blkdev_readpage, |
447f05bb | 1988 | .readpages = blkdev_readpages, |
1da177e4 | 1989 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1990 | .write_begin = blkdev_write_begin, |
1991 | .write_end = blkdev_write_end, | |
7f6d5b52 | 1992 | .writepages = blkdev_writepages, |
87d8fe1e | 1993 | .releasepage = blkdev_releasepage, |
1da177e4 | 1994 | .direct_IO = blkdev_direct_IO, |
88dbcbb3 | 1995 | .migratepage = buffer_migrate_page_norefs, |
b4597226 | 1996 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1997 | }; |
1998 | ||
25f4c414 DW |
1999 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
2000 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
2001 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) | |
2002 | ||
2003 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, | |
2004 | loff_t len) | |
2005 | { | |
2006 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
25f4c414 DW |
2007 | struct address_space *mapping; |
2008 | loff_t end = start + len - 1; | |
2009 | loff_t isize; | |
2010 | int error; | |
2011 | ||
2012 | /* Fail if we don't recognize the flags. */ | |
2013 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) | |
2014 | return -EOPNOTSUPP; | |
2015 | ||
2016 | /* Don't go off the end of the device. */ | |
2017 | isize = i_size_read(bdev->bd_inode); | |
2018 | if (start >= isize) | |
2019 | return -EINVAL; | |
2020 | if (end >= isize) { | |
2021 | if (mode & FALLOC_FL_KEEP_SIZE) { | |
2022 | len = isize - start; | |
2023 | end = start + len - 1; | |
2024 | } else | |
2025 | return -EINVAL; | |
2026 | } | |
2027 | ||
2028 | /* | |
2029 | * Don't allow IO that isn't aligned to logical block size. | |
2030 | */ | |
2031 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) | |
2032 | return -EINVAL; | |
2033 | ||
2034 | /* Invalidate the page cache, including dirty pages. */ | |
2035 | mapping = bdev->bd_inode->i_mapping; | |
2036 | truncate_inode_pages_range(mapping, start, end); | |
2037 | ||
2038 | switch (mode) { | |
2039 | case FALLOC_FL_ZERO_RANGE: | |
2040 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: | |
2041 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, | |
ee472d83 | 2042 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP); |
25f4c414 DW |
2043 | break; |
2044 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: | |
34045129 CH |
2045 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
2046 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK); | |
25f4c414 DW |
2047 | break; |
2048 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: | |
25f4c414 DW |
2049 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
2050 | GFP_KERNEL, 0); | |
2051 | break; | |
2052 | default: | |
2053 | return -EOPNOTSUPP; | |
2054 | } | |
2055 | if (error) | |
2056 | return error; | |
2057 | ||
2058 | /* | |
2059 | * Invalidate again; if someone wandered in and dirtied a page, | |
2060 | * the caller will be given -EBUSY. The third argument is | |
2061 | * inclusive, so the rounding here is safe. | |
2062 | */ | |
2063 | return invalidate_inode_pages2_range(mapping, | |
2064 | start >> PAGE_SHIFT, | |
2065 | end >> PAGE_SHIFT); | |
2066 | } | |
2067 | ||
4b6f5d20 | 2068 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
2069 | .open = blkdev_open, |
2070 | .release = blkdev_close, | |
2071 | .llseek = block_llseek, | |
a886038b | 2072 | .read_iter = blkdev_read_iter, |
1456c0a8 | 2073 | .write_iter = blkdev_write_iter, |
acc93d30 | 2074 | .mmap = generic_file_mmap, |
b1dd3b28 | 2075 | .fsync = blkdev_fsync, |
bb93e3a5 | 2076 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
2077 | #ifdef CONFIG_COMPAT |
2078 | .compat_ioctl = compat_blkdev_ioctl, | |
2079 | #endif | |
1e8b3332 | 2080 | .splice_read = generic_file_splice_read, |
8d020765 | 2081 | .splice_write = iter_file_splice_write, |
25f4c414 | 2082 | .fallocate = blkdev_fallocate, |
1da177e4 LT |
2083 | }; |
2084 | ||
2085 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
2086 | { | |
2087 | int res; | |
2088 | mm_segment_t old_fs = get_fs(); | |
2089 | set_fs(KERNEL_DS); | |
56b26add | 2090 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
2091 | set_fs(old_fs); |
2092 | return res; | |
2093 | } | |
2094 | ||
2095 | EXPORT_SYMBOL(ioctl_by_bdev); | |
2096 | ||
2097 | /** | |
2098 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 2099 | * @pathname: special file representing the block device |
1da177e4 | 2100 | * |
57d1b536 | 2101 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
2102 | * namespace if possible and return it. Return ERR_PTR(error) |
2103 | * otherwise. | |
2104 | */ | |
421748ec | 2105 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
2106 | { |
2107 | struct block_device *bdev; | |
2108 | struct inode *inode; | |
421748ec | 2109 | struct path path; |
1da177e4 LT |
2110 | int error; |
2111 | ||
421748ec | 2112 | if (!pathname || !*pathname) |
1da177e4 LT |
2113 | return ERR_PTR(-EINVAL); |
2114 | ||
421748ec | 2115 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
2116 | if (error) |
2117 | return ERR_PTR(error); | |
2118 | ||
bb668734 | 2119 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
2120 | error = -ENOTBLK; |
2121 | if (!S_ISBLK(inode->i_mode)) | |
2122 | goto fail; | |
2123 | error = -EACCES; | |
a2982cc9 | 2124 | if (!may_open_dev(&path)) |
1da177e4 LT |
2125 | goto fail; |
2126 | error = -ENOMEM; | |
2127 | bdev = bd_acquire(inode); | |
2128 | if (!bdev) | |
2129 | goto fail; | |
2130 | out: | |
421748ec | 2131 | path_put(&path); |
1da177e4 LT |
2132 | return bdev; |
2133 | fail: | |
2134 | bdev = ERR_PTR(error); | |
2135 | goto out; | |
2136 | } | |
d5686b44 | 2137 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 2138 | |
93b270f7 | 2139 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
2140 | { |
2141 | struct super_block *sb = get_super(bdev); | |
2142 | int res = 0; | |
2143 | ||
2144 | if (sb) { | |
2145 | /* | |
2146 | * no need to lock the super, get_super holds the | |
2147 | * read mutex so the filesystem cannot go away | |
2148 | * under us (->put_super runs with the write lock | |
2149 | * hold). | |
2150 | */ | |
2151 | shrink_dcache_sb(sb); | |
93b270f7 | 2152 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
2153 | drop_super(sb); |
2154 | } | |
f98393a6 | 2155 | invalidate_bdev(bdev); |
b71e8a4c DH |
2156 | return res; |
2157 | } | |
2158 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
2159 | |
2160 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
2161 | { | |
2162 | struct inode *inode, *old_inode = NULL; | |
2163 | ||
74278da9 | 2164 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2165 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
2166 | struct address_space *mapping = inode->i_mapping; | |
af309226 | 2167 | struct block_device *bdev; |
5c0d6b60 JK |
2168 | |
2169 | spin_lock(&inode->i_lock); | |
2170 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
2171 | mapping->nrpages == 0) { | |
2172 | spin_unlock(&inode->i_lock); | |
2173 | continue; | |
2174 | } | |
2175 | __iget(inode); | |
2176 | spin_unlock(&inode->i_lock); | |
74278da9 | 2177 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2178 | /* |
2179 | * We hold a reference to 'inode' so it couldn't have been | |
2180 | * removed from s_inodes list while we dropped the | |
74278da9 | 2181 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 2182 | * be holding the last reference and we cannot iput it under |
74278da9 | 2183 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
2184 | * later. |
2185 | */ | |
2186 | iput(old_inode); | |
2187 | old_inode = inode; | |
af309226 | 2188 | bdev = I_BDEV(inode); |
5c0d6b60 | 2189 | |
af309226 RV |
2190 | mutex_lock(&bdev->bd_mutex); |
2191 | if (bdev->bd_openers) | |
2192 | func(bdev, arg); | |
2193 | mutex_unlock(&bdev->bd_mutex); | |
5c0d6b60 | 2194 | |
74278da9 | 2195 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 2196 | } |
74278da9 | 2197 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2198 | iput(old_inode); |
2199 | } |