]>
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 | ||
2af3a815 TK |
57 | void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...) |
58 | { | |
59 | struct va_format vaf; | |
60 | va_list args; | |
61 | ||
62 | va_start(args, fmt); | |
63 | vaf.fmt = fmt; | |
64 | vaf.va = &args; | |
65 | printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf); | |
66 | va_end(args); | |
67 | } | |
68 | ||
dbd3ca50 | 69 | static void bdev_write_inode(struct block_device *bdev) |
564f00f6 | 70 | { |
dbd3ca50 VG |
71 | struct inode *inode = bdev->bd_inode; |
72 | int ret; | |
73 | ||
564f00f6 CH |
74 | spin_lock(&inode->i_lock); |
75 | while (inode->i_state & I_DIRTY) { | |
76 | spin_unlock(&inode->i_lock); | |
dbd3ca50 VG |
77 | ret = write_inode_now(inode, true); |
78 | if (ret) { | |
79 | char name[BDEVNAME_SIZE]; | |
80 | pr_warn_ratelimited("VFS: Dirty inode writeback failed " | |
81 | "for block device %s (err=%d).\n", | |
82 | bdevname(bdev, name), ret); | |
83 | } | |
564f00f6 CH |
84 | spin_lock(&inode->i_lock); |
85 | } | |
86 | spin_unlock(&inode->i_lock); | |
87 | } | |
88 | ||
f9a14399 | 89 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
ff01bb48 | 90 | void kill_bdev(struct block_device *bdev) |
1da177e4 | 91 | { |
ff01bb48 AV |
92 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
93 | ||
f9fe48be | 94 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
f9a14399 | 95 | return; |
ff01bb48 | 96 | |
f9a14399 | 97 | invalidate_bh_lrus(); |
ff01bb48 | 98 | truncate_inode_pages(mapping, 0); |
1da177e4 | 99 | } |
ff01bb48 AV |
100 | EXPORT_SYMBOL(kill_bdev); |
101 | ||
102 | /* Invalidate clean unused buffers and pagecache. */ | |
103 | void invalidate_bdev(struct block_device *bdev) | |
104 | { | |
105 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
106 | ||
a5f6a6a9 AR |
107 | if (mapping->nrpages) { |
108 | invalidate_bh_lrus(); | |
109 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
110 | invalidate_mapping_pages(mapping, 0, -1); | |
111 | } | |
ff01bb48 AV |
112 | /* 99% of the time, we don't need to flush the cleancache on the bdev. |
113 | * But, for the strange corners, lets be cautious | |
114 | */ | |
3167760f | 115 | cleancache_invalidate_inode(mapping); |
ff01bb48 AV |
116 | } |
117 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 LT |
118 | |
119 | int set_blocksize(struct block_device *bdev, int size) | |
120 | { | |
121 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 122 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
123 | return -EINVAL; |
124 | ||
125 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 126 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
127 | return -EINVAL; |
128 | ||
129 | /* Don't change the size if it is same as current */ | |
130 | if (bdev->bd_block_size != size) { | |
131 | sync_blockdev(bdev); | |
132 | bdev->bd_block_size = size; | |
133 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
134 | kill_bdev(bdev); | |
135 | } | |
136 | return 0; | |
137 | } | |
138 | ||
139 | EXPORT_SYMBOL(set_blocksize); | |
140 | ||
141 | int sb_set_blocksize(struct super_block *sb, int size) | |
142 | { | |
1da177e4 LT |
143 | if (set_blocksize(sb->s_bdev, size)) |
144 | return 0; | |
145 | /* If we get here, we know size is power of two | |
146 | * and it's value is between 512 and PAGE_SIZE */ | |
147 | sb->s_blocksize = size; | |
38885bd4 | 148 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
149 | return sb->s_blocksize; |
150 | } | |
151 | ||
152 | EXPORT_SYMBOL(sb_set_blocksize); | |
153 | ||
154 | int sb_min_blocksize(struct super_block *sb, int size) | |
155 | { | |
e1defc4f | 156 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
157 | if (size < minsize) |
158 | size = minsize; | |
159 | return sb_set_blocksize(sb, size); | |
160 | } | |
161 | ||
162 | EXPORT_SYMBOL(sb_min_blocksize); | |
163 | ||
164 | static int | |
165 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
166 | struct buffer_head *bh, int create) | |
167 | { | |
1da177e4 LT |
168 | bh->b_bdev = I_BDEV(inode); |
169 | bh->b_blocknr = iblock; | |
170 | set_buffer_mapped(bh); | |
171 | return 0; | |
172 | } | |
173 | ||
4ebb16ca DW |
174 | static struct inode *bdev_file_inode(struct file *file) |
175 | { | |
176 | return file->f_mapping->host; | |
177 | } | |
178 | ||
78250c02 JA |
179 | static unsigned int dio_bio_write_op(struct kiocb *iocb) |
180 | { | |
181 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; | |
182 | ||
183 | /* avoid the need for a I/O completion work item */ | |
184 | if (iocb->ki_flags & IOCB_DSYNC) | |
185 | op |= REQ_FUA; | |
186 | return op; | |
187 | } | |
188 | ||
189ce2b9 CH |
189 | #define DIO_INLINE_BIO_VECS 4 |
190 | ||
191 | static void blkdev_bio_end_io_simple(struct bio *bio) | |
192 | { | |
193 | struct task_struct *waiter = bio->bi_private; | |
194 | ||
195 | WRITE_ONCE(bio->bi_private, NULL); | |
196 | wake_up_process(waiter); | |
197 | } | |
198 | ||
199 | static ssize_t | |
200 | __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter, | |
201 | int nr_pages) | |
202 | { | |
203 | struct file *file = iocb->ki_filp; | |
204 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
72ecad22 | 205 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec; |
189ce2b9 CH |
206 | loff_t pos = iocb->ki_pos; |
207 | bool should_dirty = false; | |
208 | struct bio bio; | |
209 | ssize_t ret; | |
210 | blk_qc_t qc; | |
211 | int i; | |
212 | ||
9a794fb9 JA |
213 | if ((pos | iov_iter_alignment(iter)) & |
214 | (bdev_logical_block_size(bdev) - 1)) | |
189ce2b9 CH |
215 | return -EINVAL; |
216 | ||
72ecad22 JA |
217 | if (nr_pages <= DIO_INLINE_BIO_VECS) |
218 | vecs = inline_vecs; | |
219 | else { | |
220 | vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL); | |
221 | if (!vecs) | |
222 | return -ENOMEM; | |
223 | } | |
224 | ||
3a83f467 | 225 | bio_init(&bio, vecs, nr_pages); |
189ce2b9 | 226 | bio.bi_bdev = bdev; |
4d1a4765 | 227 | bio.bi_iter.bi_sector = pos >> 9; |
189ce2b9 CH |
228 | bio.bi_private = current; |
229 | bio.bi_end_io = blkdev_bio_end_io_simple; | |
230 | ||
231 | ret = bio_iov_iter_get_pages(&bio, iter); | |
232 | if (unlikely(ret)) | |
233 | return ret; | |
234 | ret = bio.bi_iter.bi_size; | |
235 | ||
236 | if (iov_iter_rw(iter) == READ) { | |
78250c02 | 237 | bio.bi_opf = REQ_OP_READ; |
189ce2b9 CH |
238 | if (iter_is_iovec(iter)) |
239 | should_dirty = true; | |
240 | } else { | |
78250c02 | 241 | bio.bi_opf = dio_bio_write_op(iocb); |
189ce2b9 CH |
242 | task_io_account_write(ret); |
243 | } | |
244 | ||
245 | qc = submit_bio(&bio); | |
246 | for (;;) { | |
247 | set_current_state(TASK_UNINTERRUPTIBLE); | |
248 | if (!READ_ONCE(bio.bi_private)) | |
249 | break; | |
250 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
251 | !blk_mq_poll(bdev_get_queue(bdev), qc)) | |
252 | io_schedule(); | |
253 | } | |
254 | __set_current_state(TASK_RUNNING); | |
255 | ||
256 | bio_for_each_segment_all(bvec, &bio, i) { | |
257 | if (should_dirty && !PageCompound(bvec->bv_page)) | |
258 | set_page_dirty_lock(bvec->bv_page); | |
259 | put_page(bvec->bv_page); | |
260 | } | |
261 | ||
72ecad22 JA |
262 | if (vecs != inline_vecs) |
263 | kfree(vecs); | |
264 | ||
189ce2b9 | 265 | if (unlikely(bio.bi_error)) |
9ae3b3f5 JA |
266 | ret = bio.bi_error; |
267 | ||
268 | bio_uninit(&bio); | |
269 | ||
189ce2b9 CH |
270 | return ret; |
271 | } | |
272 | ||
542ff7bf CH |
273 | struct blkdev_dio { |
274 | union { | |
275 | struct kiocb *iocb; | |
276 | struct task_struct *waiter; | |
277 | }; | |
278 | size_t size; | |
279 | atomic_t ref; | |
280 | bool multi_bio : 1; | |
281 | bool should_dirty : 1; | |
282 | bool is_sync : 1; | |
283 | struct bio bio; | |
284 | }; | |
285 | ||
286 | static struct bio_set *blkdev_dio_pool __read_mostly; | |
287 | ||
288 | static void blkdev_bio_end_io(struct bio *bio) | |
289 | { | |
290 | struct blkdev_dio *dio = bio->bi_private; | |
291 | bool should_dirty = dio->should_dirty; | |
292 | ||
293 | if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) { | |
294 | if (bio->bi_error && !dio->bio.bi_error) | |
295 | dio->bio.bi_error = bio->bi_error; | |
296 | } else { | |
297 | if (!dio->is_sync) { | |
298 | struct kiocb *iocb = dio->iocb; | |
299 | ssize_t ret = dio->bio.bi_error; | |
300 | ||
301 | if (likely(!ret)) { | |
302 | ret = dio->size; | |
303 | iocb->ki_pos += ret; | |
304 | } | |
305 | ||
306 | dio->iocb->ki_complete(iocb, ret, 0); | |
307 | bio_put(&dio->bio); | |
308 | } else { | |
309 | struct task_struct *waiter = dio->waiter; | |
310 | ||
311 | WRITE_ONCE(dio->waiter, NULL); | |
312 | wake_up_process(waiter); | |
313 | } | |
314 | } | |
315 | ||
316 | if (should_dirty) { | |
317 | bio_check_pages_dirty(bio); | |
318 | } else { | |
319 | struct bio_vec *bvec; | |
320 | int i; | |
321 | ||
322 | bio_for_each_segment_all(bvec, bio, i) | |
323 | put_page(bvec->bv_page); | |
324 | bio_put(bio); | |
325 | } | |
326 | } | |
327 | ||
b2e895db | 328 | static ssize_t |
542ff7bf | 329 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages) |
b2e895db AM |
330 | { |
331 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 332 | struct inode *inode = bdev_file_inode(file); |
542ff7bf | 333 | struct block_device *bdev = I_BDEV(inode); |
64d656a1 | 334 | struct blk_plug plug; |
542ff7bf CH |
335 | struct blkdev_dio *dio; |
336 | struct bio *bio; | |
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; | |
340 | int ret; | |
341 | ||
9a794fb9 JA |
342 | if ((pos | iov_iter_alignment(iter)) & |
343 | (bdev_logical_block_size(bdev) - 1)) | |
542ff7bf CH |
344 | return -EINVAL; |
345 | ||
346 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool); | |
347 | bio_get(bio); /* extra ref for the completion handler */ | |
348 | ||
349 | dio = container_of(bio, struct blkdev_dio, bio); | |
690e5325 | 350 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
542ff7bf CH |
351 | if (dio->is_sync) |
352 | dio->waiter = current; | |
353 | else | |
354 | dio->iocb = iocb; | |
355 | ||
356 | dio->size = 0; | |
357 | dio->multi_bio = false; | |
358 | dio->should_dirty = is_read && (iter->type == ITER_IOVEC); | |
359 | ||
64d656a1 | 360 | blk_start_plug(&plug); |
542ff7bf CH |
361 | for (;;) { |
362 | bio->bi_bdev = bdev; | |
4d1a4765 | 363 | bio->bi_iter.bi_sector = pos >> 9; |
542ff7bf CH |
364 | bio->bi_private = dio; |
365 | bio->bi_end_io = blkdev_bio_end_io; | |
366 | ||
367 | ret = bio_iov_iter_get_pages(bio, iter); | |
368 | if (unlikely(ret)) { | |
369 | bio->bi_error = ret; | |
370 | bio_endio(bio); | |
371 | break; | |
372 | } | |
373 | ||
374 | if (is_read) { | |
375 | bio->bi_opf = REQ_OP_READ; | |
376 | if (dio->should_dirty) | |
377 | bio_set_pages_dirty(bio); | |
378 | } else { | |
379 | bio->bi_opf = dio_bio_write_op(iocb); | |
380 | task_io_account_write(bio->bi_iter.bi_size); | |
381 | } | |
382 | ||
383 | dio->size += bio->bi_iter.bi_size; | |
384 | pos += bio->bi_iter.bi_size; | |
385 | ||
386 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); | |
387 | if (!nr_pages) { | |
388 | qc = submit_bio(bio); | |
389 | break; | |
390 | } | |
391 | ||
392 | if (!dio->multi_bio) { | |
393 | dio->multi_bio = true; | |
394 | atomic_set(&dio->ref, 2); | |
395 | } else { | |
396 | atomic_inc(&dio->ref); | |
397 | } | |
398 | ||
399 | submit_bio(bio); | |
400 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
401 | } | |
64d656a1 | 402 | blk_finish_plug(&plug); |
542ff7bf | 403 | |
690e5325 | 404 | if (!is_sync) |
542ff7bf CH |
405 | return -EIOCBQUEUED; |
406 | ||
407 | for (;;) { | |
408 | set_current_state(TASK_UNINTERRUPTIBLE); | |
409 | if (!READ_ONCE(dio->waiter)) | |
410 | break; | |
411 | ||
412 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
413 | !blk_mq_poll(bdev_get_queue(bdev), qc)) | |
414 | io_schedule(); | |
415 | } | |
416 | __set_current_state(TASK_RUNNING); | |
417 | ||
418 | ret = dio->bio.bi_error; | |
7a62a523 | 419 | if (likely(!ret)) |
542ff7bf | 420 | ret = dio->size; |
542ff7bf CH |
421 | |
422 | bio_put(&dio->bio); | |
423 | return ret; | |
424 | } | |
425 | ||
426 | static ssize_t | |
427 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) | |
428 | { | |
189ce2b9 | 429 | int nr_pages; |
b2e895db | 430 | |
72ecad22 | 431 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
189ce2b9 CH |
432 | if (!nr_pages) |
433 | return 0; | |
72ecad22 | 434 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
189ce2b9 | 435 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
542ff7bf CH |
436 | |
437 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); | |
438 | } | |
439 | ||
440 | static __init int blkdev_init(void) | |
441 | { | |
442 | blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio)); | |
443 | if (!blkdev_dio_pool) | |
444 | return -ENOMEM; | |
445 | return 0; | |
b2e895db | 446 | } |
542ff7bf | 447 | module_init(blkdev_init); |
b2e895db | 448 | |
5cee5815 JK |
449 | int __sync_blockdev(struct block_device *bdev, int wait) |
450 | { | |
451 | if (!bdev) | |
452 | return 0; | |
453 | if (!wait) | |
454 | return filemap_flush(bdev->bd_inode->i_mapping); | |
455 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
456 | } | |
457 | ||
585d3bc0 NP |
458 | /* |
459 | * Write out and wait upon all the dirty data associated with a block | |
460 | * device via its mapping. Does not take the superblock lock. | |
461 | */ | |
462 | int sync_blockdev(struct block_device *bdev) | |
463 | { | |
5cee5815 | 464 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
465 | } |
466 | EXPORT_SYMBOL(sync_blockdev); | |
467 | ||
468 | /* | |
469 | * Write out and wait upon all dirty data associated with this | |
470 | * device. Filesystem data as well as the underlying block | |
471 | * device. Takes the superblock lock. | |
472 | */ | |
473 | int fsync_bdev(struct block_device *bdev) | |
474 | { | |
475 | struct super_block *sb = get_super(bdev); | |
476 | if (sb) { | |
60b0680f | 477 | int res = sync_filesystem(sb); |
585d3bc0 NP |
478 | drop_super(sb); |
479 | return res; | |
480 | } | |
481 | return sync_blockdev(bdev); | |
482 | } | |
47e4491b | 483 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
484 | |
485 | /** | |
486 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
487 | * @bdev: blockdevice to lock | |
488 | * | |
585d3bc0 NP |
489 | * If a superblock is found on this device, we take the s_umount semaphore |
490 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
491 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
492 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
493 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
494 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
495 | * actually. | |
496 | */ | |
497 | struct super_block *freeze_bdev(struct block_device *bdev) | |
498 | { | |
499 | struct super_block *sb; | |
500 | int error = 0; | |
501 | ||
502 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
503 | if (++bdev->bd_fsfreeze_count > 1) { |
504 | /* | |
505 | * We don't even need to grab a reference - the first call | |
506 | * to freeze_bdev grab an active reference and only the last | |
507 | * thaw_bdev drops it. | |
508 | */ | |
585d3bc0 | 509 | sb = get_super(bdev); |
5bb53c0f AR |
510 | if (sb) |
511 | drop_super(sb); | |
4504230a CH |
512 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
513 | return sb; | |
514 | } | |
515 | ||
516 | sb = get_active_super(bdev); | |
517 | if (!sb) | |
518 | goto out; | |
48b6bca6 BM |
519 | if (sb->s_op->freeze_super) |
520 | error = sb->s_op->freeze_super(sb); | |
521 | else | |
522 | error = freeze_super(sb); | |
18e9e510 JB |
523 | if (error) { |
524 | deactivate_super(sb); | |
525 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 526 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 527 | return ERR_PTR(error); |
585d3bc0 | 528 | } |
18e9e510 | 529 | deactivate_super(sb); |
4504230a | 530 | out: |
585d3bc0 NP |
531 | sync_blockdev(bdev); |
532 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 533 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
534 | } |
535 | EXPORT_SYMBOL(freeze_bdev); | |
536 | ||
537 | /** | |
538 | * thaw_bdev -- unlock filesystem | |
539 | * @bdev: blockdevice to unlock | |
540 | * @sb: associated superblock | |
541 | * | |
542 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
543 | */ | |
544 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
545 | { | |
4504230a | 546 | int error = -EINVAL; |
585d3bc0 NP |
547 | |
548 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 549 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 550 | goto out; |
4504230a CH |
551 | |
552 | error = 0; | |
553 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 554 | goto out; |
4504230a CH |
555 | |
556 | if (!sb) | |
18e9e510 | 557 | goto out; |
4504230a | 558 | |
48b6bca6 BM |
559 | if (sb->s_op->thaw_super) |
560 | error = sb->s_op->thaw_super(sb); | |
561 | else | |
562 | error = thaw_super(sb); | |
997198ba | 563 | if (error) |
18e9e510 | 564 | bdev->bd_fsfreeze_count++; |
18e9e510 | 565 | out: |
585d3bc0 | 566 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
997198ba | 567 | return error; |
585d3bc0 NP |
568 | } |
569 | EXPORT_SYMBOL(thaw_bdev); | |
570 | ||
1da177e4 LT |
571 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
572 | { | |
573 | return block_write_full_page(page, blkdev_get_block, wbc); | |
574 | } | |
575 | ||
576 | static int blkdev_readpage(struct file * file, struct page * page) | |
577 | { | |
578 | return block_read_full_page(page, blkdev_get_block); | |
579 | } | |
580 | ||
447f05bb AM |
581 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
582 | struct list_head *pages, unsigned nr_pages) | |
583 | { | |
584 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
585 | } | |
586 | ||
6272b5a5 NP |
587 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
588 | loff_t pos, unsigned len, unsigned flags, | |
589 | struct page **pagep, void **fsdata) | |
1da177e4 | 590 | { |
155130a4 CH |
591 | return block_write_begin(mapping, pos, len, flags, pagep, |
592 | blkdev_get_block); | |
1da177e4 LT |
593 | } |
594 | ||
6272b5a5 NP |
595 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
596 | loff_t pos, unsigned len, unsigned copied, | |
597 | struct page *page, void *fsdata) | |
1da177e4 | 598 | { |
6272b5a5 NP |
599 | int ret; |
600 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
601 | ||
602 | unlock_page(page); | |
09cbfeaf | 603 | put_page(page); |
6272b5a5 NP |
604 | |
605 | return ret; | |
1da177e4 LT |
606 | } |
607 | ||
608 | /* | |
609 | * private llseek: | |
496ad9aa | 610 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
611 | * so we compute the size by hand (just as in block_read/write above) |
612 | */ | |
965c8e59 | 613 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 614 | { |
4ebb16ca | 615 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
616 | loff_t retval; |
617 | ||
5955102c | 618 | inode_lock(bd_inode); |
5d48f3a2 | 619 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 620 | inode_unlock(bd_inode); |
1da177e4 LT |
621 | return retval; |
622 | } | |
623 | ||
02c24a82 | 624 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 625 | { |
4ebb16ca | 626 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 627 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 628 | int error; |
da5aa861 RW |
629 | |
630 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
631 | if (error) | |
632 | return error; | |
ab0a9735 | 633 | |
b8af67e2 AB |
634 | /* |
635 | * There is no need to serialise calls to blkdev_issue_flush with | |
636 | * i_mutex and doing so causes performance issues with concurrent | |
637 | * O_SYNC writers to a block device. | |
638 | */ | |
dd3932ed | 639 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
640 | if (error == -EOPNOTSUPP) |
641 | error = 0; | |
b8af67e2 | 642 | |
ab0a9735 | 643 | return error; |
1da177e4 | 644 | } |
b1dd3b28 | 645 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 646 | |
47a191fd MW |
647 | /** |
648 | * bdev_read_page() - Start reading a page from a block device | |
649 | * @bdev: The device to read the page from | |
650 | * @sector: The offset on the device to read the page to (need not be aligned) | |
651 | * @page: The page to read | |
652 | * | |
653 | * On entry, the page should be locked. It will be unlocked when the page | |
654 | * has been read. If the block driver implements rw_page synchronously, | |
655 | * that will be true on exit from this function, but it need not be. | |
656 | * | |
657 | * Errors returned by this function are usually "soft", eg out of memory, or | |
658 | * queue full; callers should try a different route to read this page rather | |
659 | * than propagate an error back up the stack. | |
660 | * | |
661 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
662 | */ | |
663 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
664 | struct page *page) | |
665 | { | |
666 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
667 | int result = -EOPNOTSUPP; |
668 | ||
f68eb1e7 | 669 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
670 | return result; |
671 | ||
6f3b0e8b | 672 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
673 | if (result) |
674 | return result; | |
c11f0c0b | 675 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false); |
2e6edc95 DW |
676 | blk_queue_exit(bdev->bd_queue); |
677 | return result; | |
47a191fd MW |
678 | } |
679 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
680 | ||
681 | /** | |
682 | * bdev_write_page() - Start writing a page to a block device | |
683 | * @bdev: The device to write the page to | |
684 | * @sector: The offset on the device to write the page to (need not be aligned) | |
685 | * @page: The page to write | |
686 | * @wbc: The writeback_control for the write | |
687 | * | |
688 | * On entry, the page should be locked and not currently under writeback. | |
689 | * On exit, if the write started successfully, the page will be unlocked and | |
690 | * under writeback. If the write failed already (eg the driver failed to | |
691 | * queue the page to the device), the page will still be locked. If the | |
692 | * caller is a ->writepage implementation, it will need to unlock the page. | |
693 | * | |
694 | * Errors returned by this function are usually "soft", eg out of memory, or | |
695 | * queue full; callers should try a different route to write this page rather | |
696 | * than propagate an error back up the stack. | |
697 | * | |
698 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
699 | */ | |
700 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
701 | struct page *page, struct writeback_control *wbc) | |
702 | { | |
703 | int result; | |
47a191fd | 704 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
2e6edc95 | 705 | |
f68eb1e7 | 706 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 707 | return -EOPNOTSUPP; |
6f3b0e8b | 708 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
709 | if (result) |
710 | return result; | |
711 | ||
47a191fd | 712 | set_page_writeback(page); |
c11f0c0b | 713 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true); |
47a191fd MW |
714 | if (result) |
715 | end_page_writeback(page); | |
716 | else | |
717 | unlock_page(page); | |
2e6edc95 | 718 | blk_queue_exit(bdev->bd_queue); |
47a191fd MW |
719 | return result; |
720 | } | |
721 | EXPORT_SYMBOL_GPL(bdev_write_page); | |
722 | ||
1da177e4 LT |
723 | /* |
724 | * pseudo-fs | |
725 | */ | |
726 | ||
727 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 728 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
729 | |
730 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
731 | { | |
e94b1766 | 732 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
733 | if (!ei) |
734 | return NULL; | |
735 | return &ei->vfs_inode; | |
736 | } | |
737 | ||
fa0d7e3d | 738 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 739 | { |
fa0d7e3d | 740 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
741 | struct bdev_inode *bdi = BDEV_I(inode); |
742 | ||
1da177e4 LT |
743 | kmem_cache_free(bdev_cachep, bdi); |
744 | } | |
745 | ||
fa0d7e3d NP |
746 | static void bdev_destroy_inode(struct inode *inode) |
747 | { | |
748 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
749 | } | |
750 | ||
51cc5068 | 751 | static void init_once(void *foo) |
1da177e4 LT |
752 | { |
753 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
754 | struct block_device *bdev = &ei->bdev; | |
755 | ||
a35afb83 CL |
756 | memset(bdev, 0, sizeof(*bdev)); |
757 | mutex_init(&bdev->bd_mutex); | |
a35afb83 | 758 | INIT_LIST_HEAD(&bdev->bd_list); |
49731baa TH |
759 | #ifdef CONFIG_SYSFS |
760 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
761 | #endif | |
a5a79d00 | 762 | bdev->bd_bdi = &noop_backing_dev_info; |
a35afb83 | 763 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
764 | /* Initialize mutex for freeze. */ |
765 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
766 | } |
767 | ||
b57922d9 | 768 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
769 | { |
770 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
91b0abe3 | 771 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 772 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 773 | clear_inode(inode); |
1da177e4 | 774 | spin_lock(&bdev_lock); |
1da177e4 LT |
775 | list_del_init(&bdev->bd_list); |
776 | spin_unlock(&bdev_lock); | |
f759741d JK |
777 | /* Detach inode from wb early as bdi_put() may free bdi->wb */ |
778 | inode_detach_wb(inode); | |
a5a79d00 | 779 | if (bdev->bd_bdi != &noop_backing_dev_info) { |
b1d2dc56 | 780 | bdi_put(bdev->bd_bdi); |
a5a79d00 JK |
781 | bdev->bd_bdi = &noop_backing_dev_info; |
782 | } | |
1da177e4 LT |
783 | } |
784 | ||
ee9b6d61 | 785 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
786 | .statfs = simple_statfs, |
787 | .alloc_inode = bdev_alloc_inode, | |
788 | .destroy_inode = bdev_destroy_inode, | |
789 | .drop_inode = generic_delete_inode, | |
b57922d9 | 790 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
791 | }; |
792 | ||
51139ada AV |
793 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
794 | int flags, const char *dev_name, void *data) | |
1da177e4 | 795 | { |
3684aa70 SL |
796 | struct dentry *dent; |
797 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); | |
e9e5e3fa | 798 | if (!IS_ERR(dent)) |
3684aa70 SL |
799 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; |
800 | return dent; | |
1da177e4 LT |
801 | } |
802 | ||
803 | static struct file_system_type bd_type = { | |
804 | .name = "bdev", | |
51139ada | 805 | .mount = bd_mount, |
1da177e4 LT |
806 | .kill_sb = kill_anon_super, |
807 | }; | |
808 | ||
a212b105 TH |
809 | struct super_block *blockdev_superblock __read_mostly; |
810 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
811 | |
812 | void __init bdev_cache_init(void) | |
813 | { | |
814 | int err; | |
ace8577a | 815 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 816 | |
1da177e4 | 817 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 818 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 819 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 820 | init_once); |
1da177e4 LT |
821 | err = register_filesystem(&bd_type); |
822 | if (err) | |
823 | panic("Cannot register bdev pseudo-fs"); | |
824 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
825 | if (IS_ERR(bd_mnt)) |
826 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 827 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
828 | } |
829 | ||
830 | /* | |
831 | * Most likely _very_ bad one - but then it's hardly critical for small | |
832 | * /dev and can be fixed when somebody will need really large one. | |
833 | * Keep in mind that it will be fed through icache hash function too. | |
834 | */ | |
835 | static inline unsigned long hash(dev_t dev) | |
836 | { | |
837 | return MAJOR(dev)+MINOR(dev); | |
838 | } | |
839 | ||
840 | static int bdev_test(struct inode *inode, void *data) | |
841 | { | |
842 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
843 | } | |
844 | ||
845 | static int bdev_set(struct inode *inode, void *data) | |
846 | { | |
847 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
848 | return 0; | |
849 | } | |
850 | ||
851 | static LIST_HEAD(all_bdevs); | |
852 | ||
f44f1ab5 JK |
853 | /* |
854 | * If there is a bdev inode for this device, unhash it so that it gets evicted | |
855 | * as soon as last inode reference is dropped. | |
856 | */ | |
857 | void bdev_unhash_inode(dev_t dev) | |
858 | { | |
859 | struct inode *inode; | |
860 | ||
861 | inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev); | |
862 | if (inode) { | |
863 | remove_inode_hash(inode); | |
864 | iput(inode); | |
865 | } | |
866 | } | |
867 | ||
1da177e4 LT |
868 | struct block_device *bdget(dev_t dev) |
869 | { | |
870 | struct block_device *bdev; | |
871 | struct inode *inode; | |
872 | ||
c2acf7b9 | 873 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
874 | bdev_test, bdev_set, &dev); |
875 | ||
876 | if (!inode) | |
877 | return NULL; | |
878 | ||
879 | bdev = &BDEV_I(inode)->bdev; | |
880 | ||
881 | if (inode->i_state & I_NEW) { | |
882 | bdev->bd_contains = NULL; | |
782b94cd | 883 | bdev->bd_super = NULL; |
1da177e4 | 884 | bdev->bd_inode = inode; |
93407472 | 885 | bdev->bd_block_size = i_blocksize(inode); |
1da177e4 LT |
886 | bdev->bd_part_count = 0; |
887 | bdev->bd_invalidated = 0; | |
888 | inode->i_mode = S_IFBLK; | |
889 | inode->i_rdev = dev; | |
890 | inode->i_bdev = bdev; | |
891 | inode->i_data.a_ops = &def_blk_aops; | |
892 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
893 | spin_lock(&bdev_lock); |
894 | list_add(&bdev->bd_list, &all_bdevs); | |
895 | spin_unlock(&bdev_lock); | |
896 | unlock_new_inode(inode); | |
897 | } | |
898 | return bdev; | |
899 | } | |
900 | ||
901 | EXPORT_SYMBOL(bdget); | |
902 | ||
dddac6a7 AJ |
903 | /** |
904 | * bdgrab -- Grab a reference to an already referenced block device | |
905 | * @bdev: Block device to grab a reference to. | |
906 | */ | |
907 | struct block_device *bdgrab(struct block_device *bdev) | |
908 | { | |
7de9c6ee | 909 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
910 | return bdev; |
911 | } | |
c1681bf8 | 912 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 913 | |
1da177e4 LT |
914 | long nr_blockdev_pages(void) |
915 | { | |
203a2935 | 916 | struct block_device *bdev; |
1da177e4 LT |
917 | long ret = 0; |
918 | spin_lock(&bdev_lock); | |
203a2935 | 919 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
920 | ret += bdev->bd_inode->i_mapping->nrpages; |
921 | } | |
922 | spin_unlock(&bdev_lock); | |
923 | return ret; | |
924 | } | |
925 | ||
926 | void bdput(struct block_device *bdev) | |
927 | { | |
928 | iput(bdev->bd_inode); | |
929 | } | |
930 | ||
931 | EXPORT_SYMBOL(bdput); | |
932 | ||
933 | static struct block_device *bd_acquire(struct inode *inode) | |
934 | { | |
935 | struct block_device *bdev; | |
09d967c6 | 936 | |
1da177e4 LT |
937 | spin_lock(&bdev_lock); |
938 | bdev = inode->i_bdev; | |
cccd9fb9 | 939 | if (bdev && !inode_unhashed(bdev->bd_inode)) { |
ed8a9d2c | 940 | bdgrab(bdev); |
1da177e4 LT |
941 | spin_unlock(&bdev_lock); |
942 | return bdev; | |
943 | } | |
944 | spin_unlock(&bdev_lock); | |
09d967c6 | 945 | |
cccd9fb9 JK |
946 | /* |
947 | * i_bdev references block device inode that was already shut down | |
948 | * (corresponding device got removed). Remove the reference and look | |
949 | * up block device inode again just in case new device got | |
950 | * reestablished under the same device number. | |
951 | */ | |
952 | if (bdev) | |
953 | bd_forget(inode); | |
954 | ||
1da177e4 LT |
955 | bdev = bdget(inode->i_rdev); |
956 | if (bdev) { | |
957 | spin_lock(&bdev_lock); | |
09d967c6 OH |
958 | if (!inode->i_bdev) { |
959 | /* | |
7de9c6ee | 960 | * We take an additional reference to bd_inode, |
09d967c6 OH |
961 | * and it's released in clear_inode() of inode. |
962 | * So, we can access it via ->i_mapping always | |
963 | * without igrab(). | |
964 | */ | |
ed8a9d2c | 965 | bdgrab(bdev); |
09d967c6 OH |
966 | inode->i_bdev = bdev; |
967 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
09d967c6 | 968 | } |
1da177e4 LT |
969 | spin_unlock(&bdev_lock); |
970 | } | |
971 | return bdev; | |
972 | } | |
973 | ||
974 | /* Call when you free inode */ | |
975 | ||
976 | void bd_forget(struct inode *inode) | |
977 | { | |
09d967c6 OH |
978 | struct block_device *bdev = NULL; |
979 | ||
1da177e4 | 980 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
981 | if (!sb_is_blkdev_sb(inode->i_sb)) |
982 | bdev = inode->i_bdev; | |
a4a4f943 AV |
983 | inode->i_bdev = NULL; |
984 | inode->i_mapping = &inode->i_data; | |
1da177e4 | 985 | spin_unlock(&bdev_lock); |
09d967c6 OH |
986 | |
987 | if (bdev) | |
ed8a9d2c | 988 | bdput(bdev); |
1da177e4 LT |
989 | } |
990 | ||
1a3cbbc5 TH |
991 | /** |
992 | * bd_may_claim - test whether a block device can be claimed | |
993 | * @bdev: block device of interest | |
994 | * @whole: whole block device containing @bdev, may equal @bdev | |
995 | * @holder: holder trying to claim @bdev | |
996 | * | |
25985edc | 997 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
998 | * |
999 | * CONTEXT: | |
1000 | * spin_lock(&bdev_lock). | |
1001 | * | |
1002 | * RETURNS: | |
1003 | * %true if @bdev can be claimed, %false otherwise. | |
1004 | */ | |
1005 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
1006 | void *holder) | |
1da177e4 | 1007 | { |
1da177e4 | 1008 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 1009 | return true; /* already a holder */ |
1da177e4 | 1010 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 1011 | return false; /* held by someone else */ |
bcc7f5b4 | 1012 | else if (whole == bdev) |
1a3cbbc5 | 1013 | return true; /* is a whole device which isn't held */ |
1da177e4 | 1014 | |
e525fd89 | 1015 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
1016 | return true; /* is a partition of a device that is being partitioned */ |
1017 | else if (whole->bd_holder != NULL) | |
1018 | return false; /* is a partition of a held device */ | |
1da177e4 | 1019 | else |
1a3cbbc5 TH |
1020 | return true; /* is a partition of an un-held device */ |
1021 | } | |
1022 | ||
6b4517a7 TH |
1023 | /** |
1024 | * bd_prepare_to_claim - prepare to claim a block device | |
1025 | * @bdev: block device of interest | |
1026 | * @whole: the whole device containing @bdev, may equal @bdev | |
1027 | * @holder: holder trying to claim @bdev | |
1028 | * | |
1029 | * Prepare to claim @bdev. This function fails if @bdev is already | |
1030 | * claimed by another holder and waits if another claiming is in | |
1031 | * progress. This function doesn't actually claim. On successful | |
1032 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
1033 | * | |
1034 | * CONTEXT: | |
1035 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
1036 | * it multiple times. | |
1037 | * | |
1038 | * RETURNS: | |
1039 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
1040 | */ | |
1041 | static int bd_prepare_to_claim(struct block_device *bdev, | |
1042 | struct block_device *whole, void *holder) | |
1043 | { | |
1044 | retry: | |
1045 | /* if someone else claimed, fail */ | |
1046 | if (!bd_may_claim(bdev, whole, holder)) | |
1047 | return -EBUSY; | |
1048 | ||
e75aa858 TH |
1049 | /* if claiming is already in progress, wait for it to finish */ |
1050 | if (whole->bd_claiming) { | |
6b4517a7 TH |
1051 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
1052 | DEFINE_WAIT(wait); | |
1053 | ||
1054 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
1055 | spin_unlock(&bdev_lock); | |
1056 | schedule(); | |
1057 | finish_wait(wq, &wait); | |
1058 | spin_lock(&bdev_lock); | |
1059 | goto retry; | |
1060 | } | |
1061 | ||
1062 | /* yay, all mine */ | |
1063 | return 0; | |
1064 | } | |
1065 | ||
1066 | /** | |
1067 | * bd_start_claiming - start claiming a block device | |
1068 | * @bdev: block device of interest | |
1069 | * @holder: holder trying to claim @bdev | |
1070 | * | |
1071 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
1072 | * exclusively and mark that an exclusive open is in progress. Each | |
1073 | * successful call to this function must be matched with a call to | |
b0018361 NP |
1074 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
1075 | * fail). | |
1076 | * | |
1077 | * This function is used to gain exclusive access to the block device | |
1078 | * without actually causing other exclusive open attempts to fail. It | |
1079 | * should be used when the open sequence itself requires exclusive | |
1080 | * access but may subsequently fail. | |
6b4517a7 TH |
1081 | * |
1082 | * CONTEXT: | |
1083 | * Might sleep. | |
1084 | * | |
1085 | * RETURNS: | |
1086 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
1087 | * value on failure. | |
1088 | */ | |
1089 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
1090 | void *holder) | |
1091 | { | |
1092 | struct gendisk *disk; | |
1093 | struct block_device *whole; | |
1094 | int partno, err; | |
1095 | ||
1096 | might_sleep(); | |
1097 | ||
1098 | /* | |
1099 | * @bdev might not have been initialized properly yet, look up | |
1100 | * and grab the outer block device the hard way. | |
1101 | */ | |
1102 | disk = get_gendisk(bdev->bd_dev, &partno); | |
1103 | if (!disk) | |
1104 | return ERR_PTR(-ENXIO); | |
1105 | ||
d4c208b8 TH |
1106 | /* |
1107 | * Normally, @bdev should equal what's returned from bdget_disk() | |
1108 | * if partno is 0; however, some drivers (floppy) use multiple | |
1109 | * bdev's for the same physical device and @bdev may be one of the | |
1110 | * aliases. Keep @bdev if partno is 0. This means claimer | |
1111 | * tracking is broken for those devices but it has always been that | |
1112 | * way. | |
1113 | */ | |
1114 | if (partno) | |
1115 | whole = bdget_disk(disk, 0); | |
1116 | else | |
1117 | whole = bdgrab(bdev); | |
1118 | ||
cf342570 | 1119 | module_put(disk->fops->owner); |
6b4517a7 TH |
1120 | put_disk(disk); |
1121 | if (!whole) | |
1122 | return ERR_PTR(-ENOMEM); | |
1123 | ||
1124 | /* prepare to claim, if successful, mark claiming in progress */ | |
1125 | spin_lock(&bdev_lock); | |
1126 | ||
1127 | err = bd_prepare_to_claim(bdev, whole, holder); | |
1128 | if (err == 0) { | |
1129 | whole->bd_claiming = holder; | |
1130 | spin_unlock(&bdev_lock); | |
1131 | return whole; | |
1132 | } else { | |
1133 | spin_unlock(&bdev_lock); | |
1134 | bdput(whole); | |
1135 | return ERR_PTR(err); | |
1136 | } | |
1137 | } | |
1138 | ||
641dc636 | 1139 | #ifdef CONFIG_SYSFS |
49731baa TH |
1140 | struct bd_holder_disk { |
1141 | struct list_head list; | |
1142 | struct gendisk *disk; | |
1143 | int refcnt; | |
1144 | }; | |
1145 | ||
1146 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
1147 | struct gendisk *disk) | |
1148 | { | |
1149 | struct bd_holder_disk *holder; | |
1150 | ||
1151 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
1152 | if (holder->disk == disk) | |
1153 | return holder; | |
1154 | return NULL; | |
1155 | } | |
1156 | ||
4d7dd8fd | 1157 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 1158 | { |
4d7dd8fd | 1159 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
1160 | } |
1161 | ||
1162 | static void del_symlink(struct kobject *from, struct kobject *to) | |
1163 | { | |
641dc636 JN |
1164 | sysfs_remove_link(from, kobject_name(to)); |
1165 | } | |
1166 | ||
df6c0cd9 | 1167 | /** |
e09b457b TH |
1168 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
1169 | * @bdev: the claimed slave bdev | |
1170 | * @disk: the holding disk | |
df6c0cd9 | 1171 | * |
49731baa TH |
1172 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1173 | * | |
e09b457b | 1174 | * This functions creates the following sysfs symlinks. |
641dc636 | 1175 | * |
e09b457b TH |
1176 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1177 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 1178 | * |
e09b457b TH |
1179 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1180 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 1181 | * |
e09b457b TH |
1182 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1183 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 1184 | * |
e09b457b TH |
1185 | * The caller must have claimed @bdev before calling this function and |
1186 | * ensure that both @bdev and @disk are valid during the creation and | |
1187 | * lifetime of these symlinks. | |
641dc636 | 1188 | * |
e09b457b TH |
1189 | * CONTEXT: |
1190 | * Might sleep. | |
641dc636 | 1191 | * |
e09b457b TH |
1192 | * RETURNS: |
1193 | * 0 on success, -errno on failure. | |
641dc636 | 1194 | */ |
e09b457b | 1195 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 1196 | { |
49731baa | 1197 | struct bd_holder_disk *holder; |
e09b457b | 1198 | int ret = 0; |
641dc636 | 1199 | |
2e7b651d | 1200 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 1201 | |
49731baa | 1202 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 1203 | |
e09b457b TH |
1204 | /* FIXME: remove the following once add_disk() handles errors */ |
1205 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
1206 | goto out_unlock; | |
4e91672c | 1207 | |
49731baa TH |
1208 | holder = bd_find_holder_disk(bdev, disk); |
1209 | if (holder) { | |
1210 | holder->refcnt++; | |
e09b457b | 1211 | goto out_unlock; |
49731baa | 1212 | } |
641dc636 | 1213 | |
49731baa TH |
1214 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1215 | if (!holder) { | |
1216 | ret = -ENOMEM; | |
e09b457b TH |
1217 | goto out_unlock; |
1218 | } | |
641dc636 | 1219 | |
49731baa TH |
1220 | INIT_LIST_HEAD(&holder->list); |
1221 | holder->disk = disk; | |
1222 | holder->refcnt = 1; | |
1223 | ||
1224 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1225 | if (ret) | |
1226 | goto out_free; | |
1227 | ||
1228 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
1229 | if (ret) | |
1230 | goto out_del; | |
e7407d16 TH |
1231 | /* |
1232 | * bdev could be deleted beneath us which would implicitly destroy | |
1233 | * the holder directory. Hold on to it. | |
1234 | */ | |
1235 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1236 | |
1237 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1238 | goto out_unlock; | |
1239 | ||
1240 | out_del: | |
1241 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1242 | out_free: | |
1243 | kfree(holder); | |
e09b457b | 1244 | out_unlock: |
b4cf1b72 | 1245 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1246 | return ret; |
641dc636 | 1247 | } |
e09b457b | 1248 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1249 | |
49731baa TH |
1250 | /** |
1251 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1252 | * @bdev: the calimed slave bdev | |
1253 | * @disk: the holding disk | |
1254 | * | |
1255 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1256 | * | |
1257 | * CONTEXT: | |
1258 | * Might sleep. | |
1259 | */ | |
1260 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1261 | { |
49731baa | 1262 | struct bd_holder_disk *holder; |
641dc636 | 1263 | |
49731baa | 1264 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1265 | |
49731baa TH |
1266 | holder = bd_find_holder_disk(bdev, disk); |
1267 | ||
1268 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1269 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1270 | del_symlink(bdev->bd_part->holder_dir, | |
1271 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1272 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1273 | list_del_init(&holder->list); |
1274 | kfree(holder); | |
1275 | } | |
1276 | ||
1277 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1278 | } |
49731baa | 1279 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1280 | #endif |
1da177e4 | 1281 | |
56ade44b AP |
1282 | /** |
1283 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1284 | * | |
1285 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1286 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1287 | * |
1288 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1289 | * when a disk has been changed -- either by a media change or online | |
1290 | * resize. | |
1291 | */ | |
93b270f7 | 1292 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1293 | { |
93b270f7 | 1294 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1295 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1296 | "resized disk %s\n", |
1297 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b AP |
1298 | } |
1299 | ||
1300 | if (!bdev->bd_disk) | |
1301 | return; | |
d27769ec | 1302 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1303 | bdev->bd_invalidated = 1; |
1304 | } | |
1305 | ||
c3279d14 | 1306 | /** |
57d1b536 | 1307 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1308 | * @disk: struct gendisk to check |
1309 | * @bdev: struct bdev to adjust. | |
1310 | * | |
1311 | * This routine checks to see if the bdev size does not match the disk size | |
1312 | * and adjusts it if it differs. | |
1313 | */ | |
1314 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1315 | { | |
1316 | loff_t disk_size, bdev_size; | |
1317 | ||
1318 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1319 | bdev_size = i_size_read(bdev->bd_inode); | |
1320 | if (disk_size != bdev_size) { | |
c3279d14 AP |
1321 | printk(KERN_INFO |
1322 | "%s: detected capacity change from %lld to %lld\n", | |
424081f3 | 1323 | disk->disk_name, bdev_size, disk_size); |
c3279d14 | 1324 | i_size_write(bdev->bd_inode, disk_size); |
93b270f7 | 1325 | flush_disk(bdev, false); |
c3279d14 AP |
1326 | } |
1327 | } | |
1328 | EXPORT_SYMBOL(check_disk_size_change); | |
1329 | ||
0c002c2f | 1330 | /** |
57d1b536 | 1331 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1332 | * @disk: struct gendisk to be revalidated |
1333 | * | |
1334 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1335 | * call-backs. It is used to do common pre and post operations needed | |
1336 | * for all revalidate_disk operations. | |
1337 | */ | |
1338 | int revalidate_disk(struct gendisk *disk) | |
1339 | { | |
c3279d14 | 1340 | struct block_device *bdev; |
0c002c2f AP |
1341 | int ret = 0; |
1342 | ||
1343 | if (disk->fops->revalidate_disk) | |
1344 | ret = disk->fops->revalidate_disk(disk); | |
c3279d14 AP |
1345 | bdev = bdget_disk(disk, 0); |
1346 | if (!bdev) | |
1347 | return ret; | |
1348 | ||
1349 | mutex_lock(&bdev->bd_mutex); | |
1350 | check_disk_size_change(disk, bdev); | |
7630b661 | 1351 | bdev->bd_invalidated = 0; |
c3279d14 AP |
1352 | mutex_unlock(&bdev->bd_mutex); |
1353 | bdput(bdev); | |
0c002c2f AP |
1354 | return ret; |
1355 | } | |
1356 | EXPORT_SYMBOL(revalidate_disk); | |
1357 | ||
1da177e4 LT |
1358 | /* |
1359 | * This routine checks whether a removable media has been changed, | |
1360 | * and invalidates all buffer-cache-entries in that case. This | |
1361 | * is a relatively slow routine, so we have to try to minimize using | |
1362 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1363 | * is the best way of combining speed and utility, I think. | |
1364 | * People changing diskettes in the middle of an operation deserve | |
1365 | * to lose :-) | |
1366 | */ | |
1367 | int check_disk_change(struct block_device *bdev) | |
1368 | { | |
1369 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1370 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1371 | unsigned int events; |
1da177e4 | 1372 | |
77ea887e TH |
1373 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1374 | DISK_EVENT_EJECT_REQUEST); | |
1375 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1376 | return 0; |
1377 | ||
93b270f7 | 1378 | flush_disk(bdev, true); |
1da177e4 LT |
1379 | if (bdops->revalidate_disk) |
1380 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1381 | return 1; |
1382 | } | |
1383 | ||
1384 | EXPORT_SYMBOL(check_disk_change); | |
1385 | ||
1386 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1387 | { | |
e1defc4f | 1388 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1389 | |
5955102c | 1390 | inode_lock(bdev->bd_inode); |
d646a02a | 1391 | i_size_write(bdev->bd_inode, size); |
5955102c | 1392 | inode_unlock(bdev->bd_inode); |
09cbfeaf | 1393 | while (bsize < PAGE_SIZE) { |
1da177e4 LT |
1394 | if (size & bsize) |
1395 | break; | |
1396 | bsize <<= 1; | |
1397 | } | |
1398 | bdev->bd_block_size = bsize; | |
1399 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1400 | } | |
1401 | EXPORT_SYMBOL(bd_set_size); | |
1402 | ||
4385bab1 | 1403 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1404 | |
6d740cd5 PZ |
1405 | /* |
1406 | * bd_mutex locking: | |
1407 | * | |
1408 | * mutex_lock(part->bd_mutex) | |
1409 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1410 | */ | |
1411 | ||
572c4892 | 1412 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1413 | { |
1da177e4 | 1414 | struct gendisk *disk; |
523e1d39 | 1415 | struct module *owner; |
7db9cfd3 | 1416 | int ret; |
cf771cb5 | 1417 | int partno; |
fe6e9c1f AV |
1418 | int perm = 0; |
1419 | ||
572c4892 | 1420 | if (mode & FMODE_READ) |
fe6e9c1f | 1421 | perm |= MAY_READ; |
572c4892 | 1422 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1423 | perm |= MAY_WRITE; |
1424 | /* | |
1425 | * hooks: /n/, see "layering violations". | |
1426 | */ | |
b7300b78 CW |
1427 | if (!for_part) { |
1428 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1429 | if (ret != 0) { | |
1430 | bdput(bdev); | |
1431 | return ret; | |
1432 | } | |
82666020 | 1433 | } |
7db9cfd3 | 1434 | |
d3374825 | 1435 | restart: |
0762b8bd | 1436 | |
89f97496 | 1437 | ret = -ENXIO; |
cf771cb5 | 1438 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1439 | if (!disk) |
6e9624b8 | 1440 | goto out; |
523e1d39 | 1441 | owner = disk->fops->owner; |
1da177e4 | 1442 | |
69e02c59 | 1443 | disk_block_events(disk); |
6796bf54 | 1444 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1445 | if (!bdev->bd_openers) { |
1446 | bdev->bd_disk = disk; | |
87192a2a | 1447 | bdev->bd_queue = disk->queue; |
1da177e4 | 1448 | bdev->bd_contains = bdev; |
03cdadb0 | 1449 | |
cf771cb5 | 1450 | if (!partno) { |
89f97496 TH |
1451 | ret = -ENXIO; |
1452 | bdev->bd_part = disk_get_part(disk, partno); | |
1453 | if (!bdev->bd_part) | |
1454 | goto out_clear; | |
1455 | ||
1196f8b8 | 1456 | ret = 0; |
1da177e4 | 1457 | if (disk->fops->open) { |
572c4892 | 1458 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1459 | if (ret == -ERESTARTSYS) { |
1460 | /* Lost a race with 'disk' being | |
1461 | * deleted, try again. | |
1462 | * See md.c | |
1463 | */ | |
1464 | disk_put_part(bdev->bd_part); | |
1465 | bdev->bd_part = NULL; | |
d3374825 | 1466 | bdev->bd_disk = NULL; |
87192a2a | 1467 | bdev->bd_queue = NULL; |
d3374825 | 1468 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1469 | disk_unblock_events(disk); |
69e02c59 | 1470 | put_disk(disk); |
523e1d39 | 1471 | module_put(owner); |
d3374825 N |
1472 | goto restart; |
1473 | } | |
1da177e4 | 1474 | } |
7e69723f | 1475 | |
22375701 | 1476 | if (!ret) |
7e69723f | 1477 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
7e69723f | 1478 | |
1196f8b8 TH |
1479 | /* |
1480 | * If the device is invalidated, rescan partition | |
1481 | * if open succeeded or failed with -ENOMEDIUM. | |
1482 | * The latter is necessary to prevent ghost | |
1483 | * partitions on a removed medium. | |
1484 | */ | |
fe316bf2 JN |
1485 | if (bdev->bd_invalidated) { |
1486 | if (!ret) | |
1487 | rescan_partitions(disk, bdev); | |
1488 | else if (ret == -ENOMEDIUM) | |
1489 | invalidate_partitions(disk, bdev); | |
1490 | } | |
5a023cdb | 1491 | |
1196f8b8 TH |
1492 | if (ret) |
1493 | goto out_clear; | |
1da177e4 | 1494 | } else { |
1da177e4 LT |
1495 | struct block_device *whole; |
1496 | whole = bdget_disk(disk, 0); | |
1497 | ret = -ENOMEM; | |
1498 | if (!whole) | |
0762b8bd | 1499 | goto out_clear; |
37be4124 | 1500 | BUG_ON(for_part); |
572c4892 | 1501 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1502 | if (ret) |
0762b8bd | 1503 | goto out_clear; |
1da177e4 | 1504 | bdev->bd_contains = whole; |
89f97496 | 1505 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1506 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1507 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1508 | ret = -ENXIO; |
0762b8bd | 1509 | goto out_clear; |
1da177e4 | 1510 | } |
89f97496 | 1511 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1da177e4 | 1512 | } |
03e26279 JK |
1513 | |
1514 | if (bdev->bd_bdi == &noop_backing_dev_info) | |
1515 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info); | |
1da177e4 | 1516 | } else { |
1da177e4 | 1517 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1518 | ret = 0; |
1519 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1520 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1521 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1522 | if (bdev->bd_invalidated) { |
1523 | if (!ret) | |
1524 | rescan_partitions(bdev->bd_disk, bdev); | |
1525 | else if (ret == -ENOMEDIUM) | |
1526 | invalidate_partitions(bdev->bd_disk, bdev); | |
1527 | } | |
1196f8b8 TH |
1528 | if (ret) |
1529 | goto out_unlock_bdev; | |
1da177e4 | 1530 | } |
69e02c59 | 1531 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1532 | put_disk(disk); |
523e1d39 | 1533 | module_put(owner); |
1da177e4 LT |
1534 | } |
1535 | bdev->bd_openers++; | |
37be4124 N |
1536 | if (for_part) |
1537 | bdev->bd_part_count++; | |
c039e313 | 1538 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1539 | disk_unblock_events(disk); |
1da177e4 LT |
1540 | return 0; |
1541 | ||
0762b8bd | 1542 | out_clear: |
89f97496 | 1543 | disk_put_part(bdev->bd_part); |
1da177e4 | 1544 | bdev->bd_disk = NULL; |
0762b8bd | 1545 | bdev->bd_part = NULL; |
87192a2a | 1546 | bdev->bd_queue = NULL; |
1da177e4 | 1547 | if (bdev != bdev->bd_contains) |
572c4892 | 1548 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1549 | bdev->bd_contains = NULL; |
0762b8bd | 1550 | out_unlock_bdev: |
c039e313 | 1551 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1552 | disk_unblock_events(disk); |
0762b8bd | 1553 | put_disk(disk); |
523e1d39 | 1554 | module_put(owner); |
4345caba | 1555 | out: |
0762b8bd TH |
1556 | bdput(bdev); |
1557 | ||
1da177e4 LT |
1558 | return ret; |
1559 | } | |
1560 | ||
d4d77629 TH |
1561 | /** |
1562 | * blkdev_get - open a block device | |
1563 | * @bdev: block_device to open | |
1564 | * @mode: FMODE_* mask | |
1565 | * @holder: exclusive holder identifier | |
1566 | * | |
1567 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1568 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1569 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1570 | * | |
1571 | * On success, the reference count of @bdev is unchanged. On failure, | |
1572 | * @bdev is put. | |
1573 | * | |
1574 | * CONTEXT: | |
1575 | * Might sleep. | |
1576 | * | |
1577 | * RETURNS: | |
1578 | * 0 on success, -errno on failure. | |
1579 | */ | |
e525fd89 | 1580 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1581 | { |
e525fd89 TH |
1582 | struct block_device *whole = NULL; |
1583 | int res; | |
1584 | ||
1585 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1586 | ||
1587 | if ((mode & FMODE_EXCL) && holder) { | |
1588 | whole = bd_start_claiming(bdev, holder); | |
1589 | if (IS_ERR(whole)) { | |
1590 | bdput(bdev); | |
1591 | return PTR_ERR(whole); | |
1592 | } | |
1593 | } | |
1594 | ||
1595 | res = __blkdev_get(bdev, mode, 0); | |
1596 | ||
1597 | if (whole) { | |
d4dc210f TH |
1598 | struct gendisk *disk = whole->bd_disk; |
1599 | ||
6a027eff | 1600 | /* finish claiming */ |
77ea887e | 1601 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1602 | spin_lock(&bdev_lock); |
1603 | ||
77ea887e | 1604 | if (!res) { |
6a027eff TH |
1605 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1606 | /* | |
1607 | * Note that for a whole device bd_holders | |
1608 | * will be incremented twice, and bd_holder | |
1609 | * will be set to bd_may_claim before being | |
1610 | * set to holder | |
1611 | */ | |
1612 | whole->bd_holders++; | |
1613 | whole->bd_holder = bd_may_claim; | |
1614 | bdev->bd_holders++; | |
1615 | bdev->bd_holder = holder; | |
1616 | } | |
1617 | ||
1618 | /* tell others that we're done */ | |
1619 | BUG_ON(whole->bd_claiming != holder); | |
1620 | whole->bd_claiming = NULL; | |
1621 | wake_up_bit(&whole->bd_claiming, 0); | |
1622 | ||
1623 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1624 | |
1625 | /* | |
d4dc210f TH |
1626 | * Block event polling for write claims if requested. Any |
1627 | * write holder makes the write_holder state stick until | |
1628 | * all are released. This is good enough and tracking | |
1629 | * individual writeable reference is too fragile given the | |
1630 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1631 | */ |
4c49ff3f TH |
1632 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1633 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1634 | bdev->bd_write_holder = true; |
d4dc210f | 1635 | disk_block_events(disk); |
77ea887e TH |
1636 | } |
1637 | ||
1638 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1639 | bdput(whole); |
e525fd89 TH |
1640 | } |
1641 | ||
1642 | return res; | |
37be4124 | 1643 | } |
1da177e4 LT |
1644 | EXPORT_SYMBOL(blkdev_get); |
1645 | ||
d4d77629 TH |
1646 | /** |
1647 | * blkdev_get_by_path - open a block device by name | |
1648 | * @path: path to the block device to open | |
1649 | * @mode: FMODE_* mask | |
1650 | * @holder: exclusive holder identifier | |
1651 | * | |
1652 | * Open the blockdevice described by the device file at @path. @mode | |
1653 | * and @holder are identical to blkdev_get(). | |
1654 | * | |
1655 | * On success, the returned block_device has reference count of one. | |
1656 | * | |
1657 | * CONTEXT: | |
1658 | * Might sleep. | |
1659 | * | |
1660 | * RETURNS: | |
1661 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1662 | */ | |
1663 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1664 | void *holder) | |
1665 | { | |
1666 | struct block_device *bdev; | |
1667 | int err; | |
1668 | ||
1669 | bdev = lookup_bdev(path); | |
1670 | if (IS_ERR(bdev)) | |
1671 | return bdev; | |
1672 | ||
1673 | err = blkdev_get(bdev, mode, holder); | |
1674 | if (err) | |
1675 | return ERR_PTR(err); | |
1676 | ||
e51900f7 CE |
1677 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1678 | blkdev_put(bdev, mode); | |
1679 | return ERR_PTR(-EACCES); | |
1680 | } | |
1681 | ||
d4d77629 TH |
1682 | return bdev; |
1683 | } | |
1684 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1685 | ||
1686 | /** | |
1687 | * blkdev_get_by_dev - open a block device by device number | |
1688 | * @dev: device number of block device to open | |
1689 | * @mode: FMODE_* mask | |
1690 | * @holder: exclusive holder identifier | |
1691 | * | |
1692 | * Open the blockdevice described by device number @dev. @mode and | |
1693 | * @holder are identical to blkdev_get(). | |
1694 | * | |
1695 | * Use it ONLY if you really do not have anything better - i.e. when | |
1696 | * you are behind a truly sucky interface and all you are given is a | |
1697 | * device number. _Never_ to be used for internal purposes. If you | |
1698 | * ever need it - reconsider your API. | |
1699 | * | |
1700 | * On success, the returned block_device has reference count of one. | |
1701 | * | |
1702 | * CONTEXT: | |
1703 | * Might sleep. | |
1704 | * | |
1705 | * RETURNS: | |
1706 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1707 | */ | |
1708 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1709 | { | |
1710 | struct block_device *bdev; | |
1711 | int err; | |
1712 | ||
1713 | bdev = bdget(dev); | |
1714 | if (!bdev) | |
1715 | return ERR_PTR(-ENOMEM); | |
1716 | ||
1717 | err = blkdev_get(bdev, mode, holder); | |
1718 | if (err) | |
1719 | return ERR_PTR(err); | |
1720 | ||
1721 | return bdev; | |
1722 | } | |
1723 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1724 | ||
1da177e4 LT |
1725 | static int blkdev_open(struct inode * inode, struct file * filp) |
1726 | { | |
1727 | struct block_device *bdev; | |
1da177e4 LT |
1728 | |
1729 | /* | |
1730 | * Preserve backwards compatibility and allow large file access | |
1731 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1732 | * binary needs it. We might want to drop this workaround | |
1733 | * during an unstable branch. | |
1734 | */ | |
1735 | filp->f_flags |= O_LARGEFILE; | |
1736 | ||
572c4892 AV |
1737 | if (filp->f_flags & O_NDELAY) |
1738 | filp->f_mode |= FMODE_NDELAY; | |
1739 | if (filp->f_flags & O_EXCL) | |
1740 | filp->f_mode |= FMODE_EXCL; | |
1741 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1742 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1743 | ||
1da177e4 | 1744 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1745 | if (bdev == NULL) |
1746 | return -ENOMEM; | |
1da177e4 | 1747 | |
572c4892 AV |
1748 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1749 | ||
e525fd89 | 1750 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1751 | } |
1752 | ||
4385bab1 | 1753 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1754 | { |
2e7b651d | 1755 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1756 | struct block_device *victim = NULL; |
2e7b651d | 1757 | |
6796bf54 | 1758 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1759 | if (for_part) |
1760 | bdev->bd_part_count--; | |
1761 | ||
2e7b651d | 1762 | if (!--bdev->bd_openers) { |
6a027eff | 1763 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1764 | sync_blockdev(bdev); |
1765 | kill_bdev(bdev); | |
43d1c0eb ID |
1766 | |
1767 | bdev_write_inode(bdev); | |
2e7b651d PZ |
1768 | } |
1769 | if (bdev->bd_contains == bdev) { | |
1770 | if (disk->fops->release) | |
db2a144b | 1771 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1772 | } |
1773 | if (!bdev->bd_openers) { | |
1774 | struct module *owner = disk->fops->owner; | |
1775 | ||
0762b8bd TH |
1776 | disk_put_part(bdev->bd_part); |
1777 | bdev->bd_part = NULL; | |
2e7b651d | 1778 | bdev->bd_disk = NULL; |
37be4124 N |
1779 | if (bdev != bdev->bd_contains) |
1780 | victim = bdev->bd_contains; | |
2e7b651d | 1781 | bdev->bd_contains = NULL; |
523e1d39 TH |
1782 | |
1783 | put_disk(disk); | |
1784 | module_put(owner); | |
2e7b651d | 1785 | } |
2e7b651d PZ |
1786 | mutex_unlock(&bdev->bd_mutex); |
1787 | bdput(bdev); | |
37be4124 | 1788 | if (victim) |
9a1c3542 | 1789 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1790 | } |
1791 | ||
4385bab1 | 1792 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1793 | { |
85ef06d1 TH |
1794 | mutex_lock(&bdev->bd_mutex); |
1795 | ||
e525fd89 | 1796 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1797 | bool bdev_free; |
1798 | ||
1799 | /* | |
1800 | * Release a claim on the device. The holder fields | |
1801 | * are protected with bdev_lock. bd_mutex is to | |
1802 | * synchronize disk_holder unlinking. | |
1803 | */ | |
6a027eff TH |
1804 | spin_lock(&bdev_lock); |
1805 | ||
1806 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1807 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1808 | ||
1809 | /* bd_contains might point to self, check in a separate step */ | |
1810 | if ((bdev_free = !bdev->bd_holders)) | |
1811 | bdev->bd_holder = NULL; | |
1812 | if (!bdev->bd_contains->bd_holders) | |
1813 | bdev->bd_contains->bd_holder = NULL; | |
1814 | ||
1815 | spin_unlock(&bdev_lock); | |
1816 | ||
77ea887e TH |
1817 | /* |
1818 | * If this was the last claim, remove holder link and | |
1819 | * unblock evpoll if it was a write holder. | |
1820 | */ | |
85ef06d1 TH |
1821 | if (bdev_free && bdev->bd_write_holder) { |
1822 | disk_unblock_events(bdev->bd_disk); | |
1823 | bdev->bd_write_holder = false; | |
77ea887e | 1824 | } |
6936217c | 1825 | } |
77ea887e | 1826 | |
85ef06d1 TH |
1827 | /* |
1828 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1829 | * event. This is to ensure detection of media removal commanded | |
1830 | * from userland - e.g. eject(1). | |
1831 | */ | |
1832 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1833 | ||
1834 | mutex_unlock(&bdev->bd_mutex); | |
1835 | ||
4385bab1 | 1836 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1837 | } |
2e7b651d PZ |
1838 | EXPORT_SYMBOL(blkdev_put); |
1839 | ||
1da177e4 LT |
1840 | static int blkdev_close(struct inode * inode, struct file * filp) |
1841 | { | |
4ebb16ca | 1842 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1843 | blkdev_put(bdev, filp->f_mode); |
1844 | return 0; | |
1da177e4 LT |
1845 | } |
1846 | ||
bb93e3a5 | 1847 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1848 | { |
4ebb16ca | 1849 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1850 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1851 | |
1852 | /* | |
1853 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1854 | * to updated it before every ioctl. | |
1855 | */ | |
56b26add | 1856 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1857 | mode |= FMODE_NDELAY; |
1858 | else | |
1859 | mode &= ~FMODE_NDELAY; | |
1860 | ||
56b26add | 1861 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1862 | } |
1863 | ||
eef99380 CH |
1864 | /* |
1865 | * Write data to the block device. Only intended for the block device itself | |
1866 | * and the raw driver which basically is a fake block device. | |
1867 | * | |
1868 | * Does not take i_mutex for the write and thus is not for general purpose | |
1869 | * use. | |
1870 | */ | |
1456c0a8 | 1871 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1872 | { |
1873 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1874 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1875 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1876 | struct blk_plug plug; |
eef99380 | 1877 | ssize_t ret; |
5f380c7f | 1878 | |
7ec7b94a AV |
1879 | if (bdev_read_only(I_BDEV(bd_inode))) |
1880 | return -EPERM; | |
5f380c7f | 1881 | |
7ec7b94a | 1882 | if (!iov_iter_count(from)) |
5f380c7f AV |
1883 | return 0; |
1884 | ||
7ec7b94a AV |
1885 | if (iocb->ki_pos >= size) |
1886 | return -ENOSPC; | |
1887 | ||
1888 | iov_iter_truncate(from, size - iocb->ki_pos); | |
eef99380 | 1889 | |
53362a05 | 1890 | blk_start_plug(&plug); |
1456c0a8 | 1891 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
1892 | if (ret > 0) |
1893 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 1894 | blk_finish_plug(&plug); |
eef99380 CH |
1895 | return ret; |
1896 | } | |
1456c0a8 | 1897 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1898 | |
b2de525f | 1899 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1900 | { |
1901 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1902 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 1903 | loff_t size = i_size_read(bd_inode); |
a886038b | 1904 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1905 | |
1906 | if (pos >= size) | |
1907 | return 0; | |
1908 | ||
1909 | size -= pos; | |
a886038b AV |
1910 | iov_iter_truncate(to, size); |
1911 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1912 | } |
b2de525f | 1913 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1914 | |
87d8fe1e TT |
1915 | /* |
1916 | * Try to release a page associated with block device when the system | |
1917 | * is under memory pressure. | |
1918 | */ | |
1919 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1920 | { | |
1921 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1922 | ||
1923 | if (super && super->s_op->bdev_try_to_free_page) | |
1924 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1925 | ||
1926 | return try_to_free_buffers(page); | |
1927 | } | |
1928 | ||
7f6d5b52 RZ |
1929 | static int blkdev_writepages(struct address_space *mapping, |
1930 | struct writeback_control *wbc) | |
1931 | { | |
1932 | if (dax_mapping(mapping)) { | |
1933 | struct block_device *bdev = I_BDEV(mapping->host); | |
1934 | ||
1935 | return dax_writeback_mapping_range(mapping, bdev, wbc); | |
1936 | } | |
1937 | return generic_writepages(mapping, wbc); | |
1938 | } | |
1939 | ||
4c54ac62 | 1940 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1941 | .readpage = blkdev_readpage, |
447f05bb | 1942 | .readpages = blkdev_readpages, |
1da177e4 | 1943 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1944 | .write_begin = blkdev_write_begin, |
1945 | .write_end = blkdev_write_end, | |
7f6d5b52 | 1946 | .writepages = blkdev_writepages, |
87d8fe1e | 1947 | .releasepage = blkdev_releasepage, |
1da177e4 | 1948 | .direct_IO = blkdev_direct_IO, |
b4597226 | 1949 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1950 | }; |
1951 | ||
25f4c414 DW |
1952 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
1953 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
1954 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) | |
1955 | ||
1956 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, | |
1957 | loff_t len) | |
1958 | { | |
1959 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
25f4c414 DW |
1960 | struct address_space *mapping; |
1961 | loff_t end = start + len - 1; | |
1962 | loff_t isize; | |
1963 | int error; | |
1964 | ||
1965 | /* Fail if we don't recognize the flags. */ | |
1966 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) | |
1967 | return -EOPNOTSUPP; | |
1968 | ||
1969 | /* Don't go off the end of the device. */ | |
1970 | isize = i_size_read(bdev->bd_inode); | |
1971 | if (start >= isize) | |
1972 | return -EINVAL; | |
1973 | if (end >= isize) { | |
1974 | if (mode & FALLOC_FL_KEEP_SIZE) { | |
1975 | len = isize - start; | |
1976 | end = start + len - 1; | |
1977 | } else | |
1978 | return -EINVAL; | |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * Don't allow IO that isn't aligned to logical block size. | |
1983 | */ | |
1984 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) | |
1985 | return -EINVAL; | |
1986 | ||
1987 | /* Invalidate the page cache, including dirty pages. */ | |
1988 | mapping = bdev->bd_inode->i_mapping; | |
1989 | truncate_inode_pages_range(mapping, start, end); | |
1990 | ||
1991 | switch (mode) { | |
1992 | case FALLOC_FL_ZERO_RANGE: | |
1993 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: | |
1994 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, | |
ee472d83 | 1995 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP); |
25f4c414 DW |
1996 | break; |
1997 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: | |
34045129 CH |
1998 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
1999 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK); | |
25f4c414 DW |
2000 | break; |
2001 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: | |
25f4c414 DW |
2002 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
2003 | GFP_KERNEL, 0); | |
2004 | break; | |
2005 | default: | |
2006 | return -EOPNOTSUPP; | |
2007 | } | |
2008 | if (error) | |
2009 | return error; | |
2010 | ||
2011 | /* | |
2012 | * Invalidate again; if someone wandered in and dirtied a page, | |
2013 | * the caller will be given -EBUSY. The third argument is | |
2014 | * inclusive, so the rounding here is safe. | |
2015 | */ | |
2016 | return invalidate_inode_pages2_range(mapping, | |
2017 | start >> PAGE_SHIFT, | |
2018 | end >> PAGE_SHIFT); | |
2019 | } | |
2020 | ||
4b6f5d20 | 2021 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
2022 | .open = blkdev_open, |
2023 | .release = blkdev_close, | |
2024 | .llseek = block_llseek, | |
a886038b | 2025 | .read_iter = blkdev_read_iter, |
1456c0a8 | 2026 | .write_iter = blkdev_write_iter, |
acc93d30 | 2027 | .mmap = generic_file_mmap, |
b1dd3b28 | 2028 | .fsync = blkdev_fsync, |
bb93e3a5 | 2029 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
2030 | #ifdef CONFIG_COMPAT |
2031 | .compat_ioctl = compat_blkdev_ioctl, | |
2032 | #endif | |
1e8b3332 | 2033 | .splice_read = generic_file_splice_read, |
8d020765 | 2034 | .splice_write = iter_file_splice_write, |
25f4c414 | 2035 | .fallocate = blkdev_fallocate, |
1da177e4 LT |
2036 | }; |
2037 | ||
2038 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
2039 | { | |
2040 | int res; | |
2041 | mm_segment_t old_fs = get_fs(); | |
2042 | set_fs(KERNEL_DS); | |
56b26add | 2043 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
2044 | set_fs(old_fs); |
2045 | return res; | |
2046 | } | |
2047 | ||
2048 | EXPORT_SYMBOL(ioctl_by_bdev); | |
2049 | ||
2050 | /** | |
2051 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 2052 | * @pathname: special file representing the block device |
1da177e4 | 2053 | * |
57d1b536 | 2054 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
2055 | * namespace if possible and return it. Return ERR_PTR(error) |
2056 | * otherwise. | |
2057 | */ | |
421748ec | 2058 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
2059 | { |
2060 | struct block_device *bdev; | |
2061 | struct inode *inode; | |
421748ec | 2062 | struct path path; |
1da177e4 LT |
2063 | int error; |
2064 | ||
421748ec | 2065 | if (!pathname || !*pathname) |
1da177e4 LT |
2066 | return ERR_PTR(-EINVAL); |
2067 | ||
421748ec | 2068 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
2069 | if (error) |
2070 | return ERR_PTR(error); | |
2071 | ||
bb668734 | 2072 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
2073 | error = -ENOTBLK; |
2074 | if (!S_ISBLK(inode->i_mode)) | |
2075 | goto fail; | |
2076 | error = -EACCES; | |
a2982cc9 | 2077 | if (!may_open_dev(&path)) |
1da177e4 LT |
2078 | goto fail; |
2079 | error = -ENOMEM; | |
2080 | bdev = bd_acquire(inode); | |
2081 | if (!bdev) | |
2082 | goto fail; | |
2083 | out: | |
421748ec | 2084 | path_put(&path); |
1da177e4 LT |
2085 | return bdev; |
2086 | fail: | |
2087 | bdev = ERR_PTR(error); | |
2088 | goto out; | |
2089 | } | |
d5686b44 | 2090 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 2091 | |
93b270f7 | 2092 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
2093 | { |
2094 | struct super_block *sb = get_super(bdev); | |
2095 | int res = 0; | |
2096 | ||
2097 | if (sb) { | |
2098 | /* | |
2099 | * no need to lock the super, get_super holds the | |
2100 | * read mutex so the filesystem cannot go away | |
2101 | * under us (->put_super runs with the write lock | |
2102 | * hold). | |
2103 | */ | |
2104 | shrink_dcache_sb(sb); | |
93b270f7 | 2105 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
2106 | drop_super(sb); |
2107 | } | |
f98393a6 | 2108 | invalidate_bdev(bdev); |
b71e8a4c DH |
2109 | return res; |
2110 | } | |
2111 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
2112 | |
2113 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
2114 | { | |
2115 | struct inode *inode, *old_inode = NULL; | |
2116 | ||
74278da9 | 2117 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2118 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
2119 | struct address_space *mapping = inode->i_mapping; | |
af309226 | 2120 | struct block_device *bdev; |
5c0d6b60 JK |
2121 | |
2122 | spin_lock(&inode->i_lock); | |
2123 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
2124 | mapping->nrpages == 0) { | |
2125 | spin_unlock(&inode->i_lock); | |
2126 | continue; | |
2127 | } | |
2128 | __iget(inode); | |
2129 | spin_unlock(&inode->i_lock); | |
74278da9 | 2130 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2131 | /* |
2132 | * We hold a reference to 'inode' so it couldn't have been | |
2133 | * removed from s_inodes list while we dropped the | |
74278da9 | 2134 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 2135 | * be holding the last reference and we cannot iput it under |
74278da9 | 2136 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
2137 | * later. |
2138 | */ | |
2139 | iput(old_inode); | |
2140 | old_inode = inode; | |
af309226 | 2141 | bdev = I_BDEV(inode); |
5c0d6b60 | 2142 | |
af309226 RV |
2143 | mutex_lock(&bdev->bd_mutex); |
2144 | if (bdev->bd_openers) | |
2145 | func(bdev, arg); | |
2146 | mutex_unlock(&bdev->bd_mutex); | |
5c0d6b60 | 2147 | |
74278da9 | 2148 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 2149 | } |
74278da9 | 2150 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2151 | iput(old_inode); |
2152 | } |