]>
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 CH |
265 | if (unlikely(bio.bi_error)) |
266 | return bio.bi_error; | |
189ce2b9 CH |
267 | return ret; |
268 | } | |
269 | ||
542ff7bf CH |
270 | struct blkdev_dio { |
271 | union { | |
272 | struct kiocb *iocb; | |
273 | struct task_struct *waiter; | |
274 | }; | |
275 | size_t size; | |
276 | atomic_t ref; | |
277 | bool multi_bio : 1; | |
278 | bool should_dirty : 1; | |
279 | bool is_sync : 1; | |
280 | struct bio bio; | |
281 | }; | |
282 | ||
283 | static struct bio_set *blkdev_dio_pool __read_mostly; | |
284 | ||
285 | static void blkdev_bio_end_io(struct bio *bio) | |
286 | { | |
287 | struct blkdev_dio *dio = bio->bi_private; | |
288 | bool should_dirty = dio->should_dirty; | |
289 | ||
290 | if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) { | |
291 | if (bio->bi_error && !dio->bio.bi_error) | |
292 | dio->bio.bi_error = bio->bi_error; | |
293 | } else { | |
294 | if (!dio->is_sync) { | |
295 | struct kiocb *iocb = dio->iocb; | |
296 | ssize_t ret = dio->bio.bi_error; | |
297 | ||
298 | if (likely(!ret)) { | |
299 | ret = dio->size; | |
300 | iocb->ki_pos += ret; | |
301 | } | |
302 | ||
303 | dio->iocb->ki_complete(iocb, ret, 0); | |
304 | bio_put(&dio->bio); | |
305 | } else { | |
306 | struct task_struct *waiter = dio->waiter; | |
307 | ||
308 | WRITE_ONCE(dio->waiter, NULL); | |
309 | wake_up_process(waiter); | |
310 | } | |
311 | } | |
312 | ||
313 | if (should_dirty) { | |
314 | bio_check_pages_dirty(bio); | |
315 | } else { | |
316 | struct bio_vec *bvec; | |
317 | int i; | |
318 | ||
319 | bio_for_each_segment_all(bvec, bio, i) | |
320 | put_page(bvec->bv_page); | |
321 | bio_put(bio); | |
322 | } | |
323 | } | |
324 | ||
b2e895db | 325 | static ssize_t |
542ff7bf | 326 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages) |
b2e895db AM |
327 | { |
328 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 329 | struct inode *inode = bdev_file_inode(file); |
542ff7bf | 330 | struct block_device *bdev = I_BDEV(inode); |
64d656a1 | 331 | struct blk_plug plug; |
542ff7bf CH |
332 | struct blkdev_dio *dio; |
333 | struct bio *bio; | |
690e5325 | 334 | bool is_read = (iov_iter_rw(iter) == READ), is_sync; |
542ff7bf CH |
335 | loff_t pos = iocb->ki_pos; |
336 | blk_qc_t qc = BLK_QC_T_NONE; | |
36ffc6c1 | 337 | int ret = 0; |
542ff7bf | 338 | |
9a794fb9 JA |
339 | if ((pos | iov_iter_alignment(iter)) & |
340 | (bdev_logical_block_size(bdev) - 1)) | |
542ff7bf CH |
341 | return -EINVAL; |
342 | ||
343 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool); | |
344 | bio_get(bio); /* extra ref for the completion handler */ | |
345 | ||
346 | dio = container_of(bio, struct blkdev_dio, bio); | |
690e5325 | 347 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
542ff7bf CH |
348 | if (dio->is_sync) |
349 | dio->waiter = current; | |
350 | else | |
351 | dio->iocb = iocb; | |
352 | ||
353 | dio->size = 0; | |
354 | dio->multi_bio = false; | |
355 | dio->should_dirty = is_read && (iter->type == ITER_IOVEC); | |
356 | ||
64d656a1 | 357 | blk_start_plug(&plug); |
542ff7bf CH |
358 | for (;;) { |
359 | bio->bi_bdev = bdev; | |
4d1a4765 | 360 | bio->bi_iter.bi_sector = pos >> 9; |
542ff7bf CH |
361 | bio->bi_private = dio; |
362 | bio->bi_end_io = blkdev_bio_end_io; | |
363 | ||
364 | ret = bio_iov_iter_get_pages(bio, iter); | |
365 | if (unlikely(ret)) { | |
36ffc6c1 | 366 | bio->bi_error = -EIO; |
542ff7bf CH |
367 | bio_endio(bio); |
368 | break; | |
369 | } | |
370 | ||
371 | if (is_read) { | |
372 | bio->bi_opf = REQ_OP_READ; | |
373 | if (dio->should_dirty) | |
374 | bio_set_pages_dirty(bio); | |
375 | } else { | |
376 | bio->bi_opf = dio_bio_write_op(iocb); | |
377 | task_io_account_write(bio->bi_iter.bi_size); | |
378 | } | |
379 | ||
380 | dio->size += bio->bi_iter.bi_size; | |
381 | pos += bio->bi_iter.bi_size; | |
382 | ||
383 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); | |
384 | if (!nr_pages) { | |
385 | qc = submit_bio(bio); | |
386 | break; | |
387 | } | |
388 | ||
389 | if (!dio->multi_bio) { | |
390 | dio->multi_bio = true; | |
391 | atomic_set(&dio->ref, 2); | |
392 | } else { | |
393 | atomic_inc(&dio->ref); | |
394 | } | |
395 | ||
396 | submit_bio(bio); | |
397 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
398 | } | |
64d656a1 | 399 | blk_finish_plug(&plug); |
542ff7bf | 400 | |
690e5325 | 401 | if (!is_sync) |
542ff7bf CH |
402 | return -EIOCBQUEUED; |
403 | ||
404 | for (;;) { | |
405 | set_current_state(TASK_UNINTERRUPTIBLE); | |
406 | if (!READ_ONCE(dio->waiter)) | |
407 | break; | |
408 | ||
409 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
410 | !blk_mq_poll(bdev_get_queue(bdev), qc)) | |
411 | io_schedule(); | |
412 | } | |
413 | __set_current_state(TASK_RUNNING); | |
414 | ||
36ffc6c1 CH |
415 | if (!ret) |
416 | ret = dio->bio.bi_error; | |
7a62a523 | 417 | if (likely(!ret)) |
542ff7bf | 418 | ret = dio->size; |
542ff7bf CH |
419 | |
420 | bio_put(&dio->bio); | |
421 | return ret; | |
422 | } | |
423 | ||
424 | static ssize_t | |
425 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) | |
426 | { | |
189ce2b9 | 427 | int nr_pages; |
b2e895db | 428 | |
72ecad22 | 429 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
189ce2b9 CH |
430 | if (!nr_pages) |
431 | return 0; | |
72ecad22 | 432 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
189ce2b9 | 433 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
542ff7bf CH |
434 | |
435 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); | |
436 | } | |
437 | ||
438 | static __init int blkdev_init(void) | |
439 | { | |
440 | blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio)); | |
441 | if (!blkdev_dio_pool) | |
442 | return -ENOMEM; | |
443 | return 0; | |
b2e895db | 444 | } |
542ff7bf | 445 | module_init(blkdev_init); |
b2e895db | 446 | |
5cee5815 JK |
447 | int __sync_blockdev(struct block_device *bdev, int wait) |
448 | { | |
449 | if (!bdev) | |
450 | return 0; | |
451 | if (!wait) | |
452 | return filemap_flush(bdev->bd_inode->i_mapping); | |
453 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
454 | } | |
455 | ||
585d3bc0 NP |
456 | /* |
457 | * Write out and wait upon all the dirty data associated with a block | |
458 | * device via its mapping. Does not take the superblock lock. | |
459 | */ | |
460 | int sync_blockdev(struct block_device *bdev) | |
461 | { | |
5cee5815 | 462 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
463 | } |
464 | EXPORT_SYMBOL(sync_blockdev); | |
465 | ||
466 | /* | |
467 | * Write out and wait upon all dirty data associated with this | |
468 | * device. Filesystem data as well as the underlying block | |
469 | * device. Takes the superblock lock. | |
470 | */ | |
471 | int fsync_bdev(struct block_device *bdev) | |
472 | { | |
473 | struct super_block *sb = get_super(bdev); | |
474 | if (sb) { | |
60b0680f | 475 | int res = sync_filesystem(sb); |
585d3bc0 NP |
476 | drop_super(sb); |
477 | return res; | |
478 | } | |
479 | return sync_blockdev(bdev); | |
480 | } | |
47e4491b | 481 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
482 | |
483 | /** | |
484 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
485 | * @bdev: blockdevice to lock | |
486 | * | |
585d3bc0 NP |
487 | * If a superblock is found on this device, we take the s_umount semaphore |
488 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
489 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
490 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
491 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
492 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
493 | * actually. | |
494 | */ | |
495 | struct super_block *freeze_bdev(struct block_device *bdev) | |
496 | { | |
497 | struct super_block *sb; | |
498 | int error = 0; | |
499 | ||
500 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
501 | if (++bdev->bd_fsfreeze_count > 1) { |
502 | /* | |
503 | * We don't even need to grab a reference - the first call | |
504 | * to freeze_bdev grab an active reference and only the last | |
505 | * thaw_bdev drops it. | |
506 | */ | |
585d3bc0 | 507 | sb = get_super(bdev); |
5bb53c0f AR |
508 | if (sb) |
509 | drop_super(sb); | |
4504230a CH |
510 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
511 | return sb; | |
512 | } | |
513 | ||
514 | sb = get_active_super(bdev); | |
515 | if (!sb) | |
516 | goto out; | |
48b6bca6 BM |
517 | if (sb->s_op->freeze_super) |
518 | error = sb->s_op->freeze_super(sb); | |
519 | else | |
520 | error = freeze_super(sb); | |
18e9e510 JB |
521 | if (error) { |
522 | deactivate_super(sb); | |
523 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 524 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 525 | return ERR_PTR(error); |
585d3bc0 | 526 | } |
18e9e510 | 527 | deactivate_super(sb); |
4504230a | 528 | out: |
585d3bc0 NP |
529 | sync_blockdev(bdev); |
530 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 531 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
532 | } |
533 | EXPORT_SYMBOL(freeze_bdev); | |
534 | ||
535 | /** | |
536 | * thaw_bdev -- unlock filesystem | |
537 | * @bdev: blockdevice to unlock | |
538 | * @sb: associated superblock | |
539 | * | |
540 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
541 | */ | |
542 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
543 | { | |
4504230a | 544 | int error = -EINVAL; |
585d3bc0 NP |
545 | |
546 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 547 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 548 | goto out; |
4504230a CH |
549 | |
550 | error = 0; | |
551 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 552 | goto out; |
4504230a CH |
553 | |
554 | if (!sb) | |
18e9e510 | 555 | goto out; |
4504230a | 556 | |
48b6bca6 BM |
557 | if (sb->s_op->thaw_super) |
558 | error = sb->s_op->thaw_super(sb); | |
559 | else | |
560 | error = thaw_super(sb); | |
997198ba | 561 | if (error) |
18e9e510 | 562 | bdev->bd_fsfreeze_count++; |
18e9e510 | 563 | out: |
585d3bc0 | 564 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
997198ba | 565 | return error; |
585d3bc0 NP |
566 | } |
567 | EXPORT_SYMBOL(thaw_bdev); | |
568 | ||
1da177e4 LT |
569 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
570 | { | |
571 | return block_write_full_page(page, blkdev_get_block, wbc); | |
572 | } | |
573 | ||
574 | static int blkdev_readpage(struct file * file, struct page * page) | |
575 | { | |
576 | return block_read_full_page(page, blkdev_get_block); | |
577 | } | |
578 | ||
447f05bb AM |
579 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
580 | struct list_head *pages, unsigned nr_pages) | |
581 | { | |
582 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); | |
583 | } | |
584 | ||
6272b5a5 NP |
585 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
586 | loff_t pos, unsigned len, unsigned flags, | |
587 | struct page **pagep, void **fsdata) | |
1da177e4 | 588 | { |
155130a4 CH |
589 | return block_write_begin(mapping, pos, len, flags, pagep, |
590 | blkdev_get_block); | |
1da177e4 LT |
591 | } |
592 | ||
6272b5a5 NP |
593 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
594 | loff_t pos, unsigned len, unsigned copied, | |
595 | struct page *page, void *fsdata) | |
1da177e4 | 596 | { |
6272b5a5 NP |
597 | int ret; |
598 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
599 | ||
600 | unlock_page(page); | |
09cbfeaf | 601 | put_page(page); |
6272b5a5 NP |
602 | |
603 | return ret; | |
1da177e4 LT |
604 | } |
605 | ||
606 | /* | |
607 | * private llseek: | |
496ad9aa | 608 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
609 | * so we compute the size by hand (just as in block_read/write above) |
610 | */ | |
965c8e59 | 611 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 | 612 | { |
4ebb16ca | 613 | struct inode *bd_inode = bdev_file_inode(file); |
1da177e4 LT |
614 | loff_t retval; |
615 | ||
5955102c | 616 | inode_lock(bd_inode); |
5d48f3a2 | 617 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
5955102c | 618 | inode_unlock(bd_inode); |
1da177e4 LT |
619 | return retval; |
620 | } | |
621 | ||
02c24a82 | 622 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 623 | { |
4ebb16ca | 624 | struct inode *bd_inode = bdev_file_inode(filp); |
b8af67e2 | 625 | struct block_device *bdev = I_BDEV(bd_inode); |
ab0a9735 | 626 | int error; |
da5aa861 RW |
627 | |
628 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
629 | if (error) | |
630 | return error; | |
ab0a9735 | 631 | |
b8af67e2 AB |
632 | /* |
633 | * There is no need to serialise calls to blkdev_issue_flush with | |
634 | * i_mutex and doing so causes performance issues with concurrent | |
635 | * O_SYNC writers to a block device. | |
636 | */ | |
dd3932ed | 637 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
638 | if (error == -EOPNOTSUPP) |
639 | error = 0; | |
b8af67e2 | 640 | |
ab0a9735 | 641 | return error; |
1da177e4 | 642 | } |
b1dd3b28 | 643 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 | 644 | |
47a191fd MW |
645 | /** |
646 | * bdev_read_page() - Start reading a page from a block device | |
647 | * @bdev: The device to read the page from | |
648 | * @sector: The offset on the device to read the page to (need not be aligned) | |
649 | * @page: The page to read | |
650 | * | |
651 | * On entry, the page should be locked. It will be unlocked when the page | |
652 | * has been read. If the block driver implements rw_page synchronously, | |
653 | * that will be true on exit from this function, but it need not be. | |
654 | * | |
655 | * Errors returned by this function are usually "soft", eg out of memory, or | |
656 | * queue full; callers should try a different route to read this page rather | |
657 | * than propagate an error back up the stack. | |
658 | * | |
659 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
660 | */ | |
661 | int bdev_read_page(struct block_device *bdev, sector_t sector, | |
662 | struct page *page) | |
663 | { | |
664 | const struct block_device_operations *ops = bdev->bd_disk->fops; | |
2e6edc95 DW |
665 | int result = -EOPNOTSUPP; |
666 | ||
f68eb1e7 | 667 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
2e6edc95 DW |
668 | return result; |
669 | ||
6f3b0e8b | 670 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
671 | if (result) |
672 | return result; | |
c11f0c0b | 673 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false); |
2e6edc95 DW |
674 | blk_queue_exit(bdev->bd_queue); |
675 | return result; | |
47a191fd MW |
676 | } |
677 | EXPORT_SYMBOL_GPL(bdev_read_page); | |
678 | ||
679 | /** | |
680 | * bdev_write_page() - Start writing a page to a block device | |
681 | * @bdev: The device to write the page to | |
682 | * @sector: The offset on the device to write the page to (need not be aligned) | |
683 | * @page: The page to write | |
684 | * @wbc: The writeback_control for the write | |
685 | * | |
686 | * On entry, the page should be locked and not currently under writeback. | |
687 | * On exit, if the write started successfully, the page will be unlocked and | |
688 | * under writeback. If the write failed already (eg the driver failed to | |
689 | * queue the page to the device), the page will still be locked. If the | |
690 | * caller is a ->writepage implementation, it will need to unlock the page. | |
691 | * | |
692 | * Errors returned by this function are usually "soft", eg out of memory, or | |
693 | * queue full; callers should try a different route to write this page rather | |
694 | * than propagate an error back up the stack. | |
695 | * | |
696 | * Return: negative errno if an error occurs, 0 if submission was successful. | |
697 | */ | |
698 | int bdev_write_page(struct block_device *bdev, sector_t sector, | |
699 | struct page *page, struct writeback_control *wbc) | |
700 | { | |
701 | int result; | |
47a191fd | 702 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
2e6edc95 | 703 | |
f68eb1e7 | 704 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
47a191fd | 705 | return -EOPNOTSUPP; |
6f3b0e8b | 706 | result = blk_queue_enter(bdev->bd_queue, false); |
2e6edc95 DW |
707 | if (result) |
708 | return result; | |
709 | ||
47a191fd | 710 | set_page_writeback(page); |
c11f0c0b | 711 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true); |
47a191fd MW |
712 | if (result) |
713 | end_page_writeback(page); | |
714 | else | |
715 | unlock_page(page); | |
2e6edc95 | 716 | blk_queue_exit(bdev->bd_queue); |
47a191fd MW |
717 | return result; |
718 | } | |
719 | EXPORT_SYMBOL_GPL(bdev_write_page); | |
720 | ||
1da177e4 LT |
721 | /* |
722 | * pseudo-fs | |
723 | */ | |
724 | ||
725 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 726 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
727 | |
728 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
729 | { | |
e94b1766 | 730 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
731 | if (!ei) |
732 | return NULL; | |
733 | return &ei->vfs_inode; | |
734 | } | |
735 | ||
fa0d7e3d | 736 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 737 | { |
fa0d7e3d | 738 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
739 | struct bdev_inode *bdi = BDEV_I(inode); |
740 | ||
1da177e4 LT |
741 | kmem_cache_free(bdev_cachep, bdi); |
742 | } | |
743 | ||
fa0d7e3d NP |
744 | static void bdev_destroy_inode(struct inode *inode) |
745 | { | |
746 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
747 | } | |
748 | ||
51cc5068 | 749 | static void init_once(void *foo) |
1da177e4 LT |
750 | { |
751 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
752 | struct block_device *bdev = &ei->bdev; | |
753 | ||
a35afb83 CL |
754 | memset(bdev, 0, sizeof(*bdev)); |
755 | mutex_init(&bdev->bd_mutex); | |
a35afb83 | 756 | INIT_LIST_HEAD(&bdev->bd_list); |
49731baa TH |
757 | #ifdef CONFIG_SYSFS |
758 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
759 | #endif | |
a5a79d00 | 760 | bdev->bd_bdi = &noop_backing_dev_info; |
a35afb83 | 761 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
762 | /* Initialize mutex for freeze. */ |
763 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
764 | } |
765 | ||
b57922d9 | 766 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
767 | { |
768 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
91b0abe3 | 769 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 770 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 771 | clear_inode(inode); |
1da177e4 | 772 | spin_lock(&bdev_lock); |
1da177e4 LT |
773 | list_del_init(&bdev->bd_list); |
774 | spin_unlock(&bdev_lock); | |
f759741d JK |
775 | /* Detach inode from wb early as bdi_put() may free bdi->wb */ |
776 | inode_detach_wb(inode); | |
a5a79d00 | 777 | if (bdev->bd_bdi != &noop_backing_dev_info) { |
b1d2dc56 | 778 | bdi_put(bdev->bd_bdi); |
a5a79d00 JK |
779 | bdev->bd_bdi = &noop_backing_dev_info; |
780 | } | |
1da177e4 LT |
781 | } |
782 | ||
ee9b6d61 | 783 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
784 | .statfs = simple_statfs, |
785 | .alloc_inode = bdev_alloc_inode, | |
786 | .destroy_inode = bdev_destroy_inode, | |
787 | .drop_inode = generic_delete_inode, | |
b57922d9 | 788 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
789 | }; |
790 | ||
51139ada AV |
791 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
792 | int flags, const char *dev_name, void *data) | |
1da177e4 | 793 | { |
3684aa70 SL |
794 | struct dentry *dent; |
795 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); | |
e9e5e3fa | 796 | if (!IS_ERR(dent)) |
3684aa70 SL |
797 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; |
798 | return dent; | |
1da177e4 LT |
799 | } |
800 | ||
801 | static struct file_system_type bd_type = { | |
802 | .name = "bdev", | |
51139ada | 803 | .mount = bd_mount, |
1da177e4 LT |
804 | .kill_sb = kill_anon_super, |
805 | }; | |
806 | ||
a212b105 TH |
807 | struct super_block *blockdev_superblock __read_mostly; |
808 | EXPORT_SYMBOL_GPL(blockdev_superblock); | |
1da177e4 LT |
809 | |
810 | void __init bdev_cache_init(void) | |
811 | { | |
812 | int err; | |
ace8577a | 813 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 814 | |
1da177e4 | 815 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 816 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
5d097056 | 817 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 818 | init_once); |
1da177e4 LT |
819 | err = register_filesystem(&bd_type); |
820 | if (err) | |
821 | panic("Cannot register bdev pseudo-fs"); | |
822 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
823 | if (IS_ERR(bd_mnt)) |
824 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 825 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
826 | } |
827 | ||
828 | /* | |
829 | * Most likely _very_ bad one - but then it's hardly critical for small | |
830 | * /dev and can be fixed when somebody will need really large one. | |
831 | * Keep in mind that it will be fed through icache hash function too. | |
832 | */ | |
833 | static inline unsigned long hash(dev_t dev) | |
834 | { | |
835 | return MAJOR(dev)+MINOR(dev); | |
836 | } | |
837 | ||
838 | static int bdev_test(struct inode *inode, void *data) | |
839 | { | |
840 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
841 | } | |
842 | ||
843 | static int bdev_set(struct inode *inode, void *data) | |
844 | { | |
845 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
846 | return 0; | |
847 | } | |
848 | ||
849 | static LIST_HEAD(all_bdevs); | |
850 | ||
f44f1ab5 JK |
851 | /* |
852 | * If there is a bdev inode for this device, unhash it so that it gets evicted | |
853 | * as soon as last inode reference is dropped. | |
854 | */ | |
855 | void bdev_unhash_inode(dev_t dev) | |
856 | { | |
857 | struct inode *inode; | |
858 | ||
859 | inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev); | |
860 | if (inode) { | |
861 | remove_inode_hash(inode); | |
862 | iput(inode); | |
863 | } | |
864 | } | |
865 | ||
1da177e4 LT |
866 | struct block_device *bdget(dev_t dev) |
867 | { | |
868 | struct block_device *bdev; | |
869 | struct inode *inode; | |
870 | ||
c2acf7b9 | 871 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
872 | bdev_test, bdev_set, &dev); |
873 | ||
874 | if (!inode) | |
875 | return NULL; | |
876 | ||
877 | bdev = &BDEV_I(inode)->bdev; | |
878 | ||
879 | if (inode->i_state & I_NEW) { | |
880 | bdev->bd_contains = NULL; | |
782b94cd | 881 | bdev->bd_super = NULL; |
1da177e4 | 882 | bdev->bd_inode = inode; |
93407472 | 883 | bdev->bd_block_size = i_blocksize(inode); |
1da177e4 LT |
884 | bdev->bd_part_count = 0; |
885 | bdev->bd_invalidated = 0; | |
886 | inode->i_mode = S_IFBLK; | |
887 | inode->i_rdev = dev; | |
888 | inode->i_bdev = bdev; | |
889 | inode->i_data.a_ops = &def_blk_aops; | |
890 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
1da177e4 LT |
891 | spin_lock(&bdev_lock); |
892 | list_add(&bdev->bd_list, &all_bdevs); | |
893 | spin_unlock(&bdev_lock); | |
894 | unlock_new_inode(inode); | |
895 | } | |
896 | return bdev; | |
897 | } | |
898 | ||
899 | EXPORT_SYMBOL(bdget); | |
900 | ||
dddac6a7 AJ |
901 | /** |
902 | * bdgrab -- Grab a reference to an already referenced block device | |
903 | * @bdev: Block device to grab a reference to. | |
904 | */ | |
905 | struct block_device *bdgrab(struct block_device *bdev) | |
906 | { | |
7de9c6ee | 907 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
908 | return bdev; |
909 | } | |
c1681bf8 | 910 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 911 | |
1da177e4 LT |
912 | long nr_blockdev_pages(void) |
913 | { | |
203a2935 | 914 | struct block_device *bdev; |
1da177e4 LT |
915 | long ret = 0; |
916 | spin_lock(&bdev_lock); | |
203a2935 | 917 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
918 | ret += bdev->bd_inode->i_mapping->nrpages; |
919 | } | |
920 | spin_unlock(&bdev_lock); | |
921 | return ret; | |
922 | } | |
923 | ||
924 | void bdput(struct block_device *bdev) | |
925 | { | |
926 | iput(bdev->bd_inode); | |
927 | } | |
928 | ||
929 | EXPORT_SYMBOL(bdput); | |
930 | ||
931 | static struct block_device *bd_acquire(struct inode *inode) | |
932 | { | |
933 | struct block_device *bdev; | |
09d967c6 | 934 | |
1da177e4 LT |
935 | spin_lock(&bdev_lock); |
936 | bdev = inode->i_bdev; | |
cccd9fb9 | 937 | if (bdev && !inode_unhashed(bdev->bd_inode)) { |
ed8a9d2c | 938 | bdgrab(bdev); |
1da177e4 LT |
939 | spin_unlock(&bdev_lock); |
940 | return bdev; | |
941 | } | |
942 | spin_unlock(&bdev_lock); | |
09d967c6 | 943 | |
cccd9fb9 JK |
944 | /* |
945 | * i_bdev references block device inode that was already shut down | |
946 | * (corresponding device got removed). Remove the reference and look | |
947 | * up block device inode again just in case new device got | |
948 | * reestablished under the same device number. | |
949 | */ | |
950 | if (bdev) | |
951 | bd_forget(inode); | |
952 | ||
1da177e4 LT |
953 | bdev = bdget(inode->i_rdev); |
954 | if (bdev) { | |
955 | spin_lock(&bdev_lock); | |
09d967c6 OH |
956 | if (!inode->i_bdev) { |
957 | /* | |
7de9c6ee | 958 | * We take an additional reference to bd_inode, |
09d967c6 OH |
959 | * and it's released in clear_inode() of inode. |
960 | * So, we can access it via ->i_mapping always | |
961 | * without igrab(). | |
962 | */ | |
ed8a9d2c | 963 | bdgrab(bdev); |
09d967c6 OH |
964 | inode->i_bdev = bdev; |
965 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
09d967c6 | 966 | } |
1da177e4 LT |
967 | spin_unlock(&bdev_lock); |
968 | } | |
969 | return bdev; | |
970 | } | |
971 | ||
972 | /* Call when you free inode */ | |
973 | ||
974 | void bd_forget(struct inode *inode) | |
975 | { | |
09d967c6 OH |
976 | struct block_device *bdev = NULL; |
977 | ||
1da177e4 | 978 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
979 | if (!sb_is_blkdev_sb(inode->i_sb)) |
980 | bdev = inode->i_bdev; | |
a4a4f943 AV |
981 | inode->i_bdev = NULL; |
982 | inode->i_mapping = &inode->i_data; | |
1da177e4 | 983 | spin_unlock(&bdev_lock); |
09d967c6 OH |
984 | |
985 | if (bdev) | |
ed8a9d2c | 986 | bdput(bdev); |
1da177e4 LT |
987 | } |
988 | ||
1a3cbbc5 TH |
989 | /** |
990 | * bd_may_claim - test whether a block device can be claimed | |
991 | * @bdev: block device of interest | |
992 | * @whole: whole block device containing @bdev, may equal @bdev | |
993 | * @holder: holder trying to claim @bdev | |
994 | * | |
25985edc | 995 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
996 | * |
997 | * CONTEXT: | |
998 | * spin_lock(&bdev_lock). | |
999 | * | |
1000 | * RETURNS: | |
1001 | * %true if @bdev can be claimed, %false otherwise. | |
1002 | */ | |
1003 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
1004 | void *holder) | |
1da177e4 | 1005 | { |
1da177e4 | 1006 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 1007 | return true; /* already a holder */ |
1da177e4 | 1008 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 1009 | return false; /* held by someone else */ |
bcc7f5b4 | 1010 | else if (whole == bdev) |
1a3cbbc5 | 1011 | return true; /* is a whole device which isn't held */ |
1da177e4 | 1012 | |
e525fd89 | 1013 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
1014 | return true; /* is a partition of a device that is being partitioned */ |
1015 | else if (whole->bd_holder != NULL) | |
1016 | return false; /* is a partition of a held device */ | |
1da177e4 | 1017 | else |
1a3cbbc5 TH |
1018 | return true; /* is a partition of an un-held device */ |
1019 | } | |
1020 | ||
6b4517a7 TH |
1021 | /** |
1022 | * bd_prepare_to_claim - prepare to claim a block device | |
1023 | * @bdev: block device of interest | |
1024 | * @whole: the whole device containing @bdev, may equal @bdev | |
1025 | * @holder: holder trying to claim @bdev | |
1026 | * | |
1027 | * Prepare to claim @bdev. This function fails if @bdev is already | |
1028 | * claimed by another holder and waits if another claiming is in | |
1029 | * progress. This function doesn't actually claim. On successful | |
1030 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
1031 | * | |
1032 | * CONTEXT: | |
1033 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
1034 | * it multiple times. | |
1035 | * | |
1036 | * RETURNS: | |
1037 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
1038 | */ | |
1039 | static int bd_prepare_to_claim(struct block_device *bdev, | |
1040 | struct block_device *whole, void *holder) | |
1041 | { | |
1042 | retry: | |
1043 | /* if someone else claimed, fail */ | |
1044 | if (!bd_may_claim(bdev, whole, holder)) | |
1045 | return -EBUSY; | |
1046 | ||
e75aa858 TH |
1047 | /* if claiming is already in progress, wait for it to finish */ |
1048 | if (whole->bd_claiming) { | |
6b4517a7 TH |
1049 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
1050 | DEFINE_WAIT(wait); | |
1051 | ||
1052 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
1053 | spin_unlock(&bdev_lock); | |
1054 | schedule(); | |
1055 | finish_wait(wq, &wait); | |
1056 | spin_lock(&bdev_lock); | |
1057 | goto retry; | |
1058 | } | |
1059 | ||
1060 | /* yay, all mine */ | |
1061 | return 0; | |
1062 | } | |
1063 | ||
1064 | /** | |
1065 | * bd_start_claiming - start claiming a block device | |
1066 | * @bdev: block device of interest | |
1067 | * @holder: holder trying to claim @bdev | |
1068 | * | |
1069 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
1070 | * exclusively and mark that an exclusive open is in progress. Each | |
1071 | * successful call to this function must be matched with a call to | |
b0018361 NP |
1072 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
1073 | * fail). | |
1074 | * | |
1075 | * This function is used to gain exclusive access to the block device | |
1076 | * without actually causing other exclusive open attempts to fail. It | |
1077 | * should be used when the open sequence itself requires exclusive | |
1078 | * access but may subsequently fail. | |
6b4517a7 TH |
1079 | * |
1080 | * CONTEXT: | |
1081 | * Might sleep. | |
1082 | * | |
1083 | * RETURNS: | |
1084 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
1085 | * value on failure. | |
1086 | */ | |
1087 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
1088 | void *holder) | |
1089 | { | |
1090 | struct gendisk *disk; | |
1091 | struct block_device *whole; | |
1092 | int partno, err; | |
1093 | ||
1094 | might_sleep(); | |
1095 | ||
1096 | /* | |
1097 | * @bdev might not have been initialized properly yet, look up | |
1098 | * and grab the outer block device the hard way. | |
1099 | */ | |
1100 | disk = get_gendisk(bdev->bd_dev, &partno); | |
1101 | if (!disk) | |
1102 | return ERR_PTR(-ENXIO); | |
1103 | ||
d4c208b8 TH |
1104 | /* |
1105 | * Normally, @bdev should equal what's returned from bdget_disk() | |
1106 | * if partno is 0; however, some drivers (floppy) use multiple | |
1107 | * bdev's for the same physical device and @bdev may be one of the | |
1108 | * aliases. Keep @bdev if partno is 0. This means claimer | |
1109 | * tracking is broken for those devices but it has always been that | |
1110 | * way. | |
1111 | */ | |
1112 | if (partno) | |
1113 | whole = bdget_disk(disk, 0); | |
1114 | else | |
1115 | whole = bdgrab(bdev); | |
1116 | ||
cf342570 | 1117 | module_put(disk->fops->owner); |
6b4517a7 TH |
1118 | put_disk(disk); |
1119 | if (!whole) | |
1120 | return ERR_PTR(-ENOMEM); | |
1121 | ||
1122 | /* prepare to claim, if successful, mark claiming in progress */ | |
1123 | spin_lock(&bdev_lock); | |
1124 | ||
1125 | err = bd_prepare_to_claim(bdev, whole, holder); | |
1126 | if (err == 0) { | |
1127 | whole->bd_claiming = holder; | |
1128 | spin_unlock(&bdev_lock); | |
1129 | return whole; | |
1130 | } else { | |
1131 | spin_unlock(&bdev_lock); | |
1132 | bdput(whole); | |
1133 | return ERR_PTR(err); | |
1134 | } | |
1135 | } | |
1136 | ||
641dc636 | 1137 | #ifdef CONFIG_SYSFS |
49731baa TH |
1138 | struct bd_holder_disk { |
1139 | struct list_head list; | |
1140 | struct gendisk *disk; | |
1141 | int refcnt; | |
1142 | }; | |
1143 | ||
1144 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
1145 | struct gendisk *disk) | |
1146 | { | |
1147 | struct bd_holder_disk *holder; | |
1148 | ||
1149 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
1150 | if (holder->disk == disk) | |
1151 | return holder; | |
1152 | return NULL; | |
1153 | } | |
1154 | ||
4d7dd8fd | 1155 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 1156 | { |
4d7dd8fd | 1157 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
1158 | } |
1159 | ||
1160 | static void del_symlink(struct kobject *from, struct kobject *to) | |
1161 | { | |
641dc636 JN |
1162 | sysfs_remove_link(from, kobject_name(to)); |
1163 | } | |
1164 | ||
df6c0cd9 | 1165 | /** |
e09b457b TH |
1166 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
1167 | * @bdev: the claimed slave bdev | |
1168 | * @disk: the holding disk | |
df6c0cd9 | 1169 | * |
49731baa TH |
1170 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1171 | * | |
e09b457b | 1172 | * This functions creates the following sysfs symlinks. |
641dc636 | 1173 | * |
e09b457b TH |
1174 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1175 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 1176 | * |
e09b457b TH |
1177 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1178 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 1179 | * |
e09b457b TH |
1180 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1181 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 1182 | * |
e09b457b TH |
1183 | * The caller must have claimed @bdev before calling this function and |
1184 | * ensure that both @bdev and @disk are valid during the creation and | |
1185 | * lifetime of these symlinks. | |
641dc636 | 1186 | * |
e09b457b TH |
1187 | * CONTEXT: |
1188 | * Might sleep. | |
641dc636 | 1189 | * |
e09b457b TH |
1190 | * RETURNS: |
1191 | * 0 on success, -errno on failure. | |
641dc636 | 1192 | */ |
e09b457b | 1193 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 1194 | { |
49731baa | 1195 | struct bd_holder_disk *holder; |
e09b457b | 1196 | int ret = 0; |
641dc636 | 1197 | |
2e7b651d | 1198 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 1199 | |
49731baa | 1200 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 1201 | |
e09b457b TH |
1202 | /* FIXME: remove the following once add_disk() handles errors */ |
1203 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
1204 | goto out_unlock; | |
4e91672c | 1205 | |
49731baa TH |
1206 | holder = bd_find_holder_disk(bdev, disk); |
1207 | if (holder) { | |
1208 | holder->refcnt++; | |
e09b457b | 1209 | goto out_unlock; |
49731baa | 1210 | } |
641dc636 | 1211 | |
49731baa TH |
1212 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1213 | if (!holder) { | |
1214 | ret = -ENOMEM; | |
e09b457b TH |
1215 | goto out_unlock; |
1216 | } | |
641dc636 | 1217 | |
49731baa TH |
1218 | INIT_LIST_HEAD(&holder->list); |
1219 | holder->disk = disk; | |
1220 | holder->refcnt = 1; | |
1221 | ||
1222 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1223 | if (ret) | |
1224 | goto out_free; | |
1225 | ||
1226 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
1227 | if (ret) | |
1228 | goto out_del; | |
e7407d16 TH |
1229 | /* |
1230 | * bdev could be deleted beneath us which would implicitly destroy | |
1231 | * the holder directory. Hold on to it. | |
1232 | */ | |
1233 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
1234 | |
1235 | list_add(&holder->list, &bdev->bd_holder_disks); | |
1236 | goto out_unlock; | |
1237 | ||
1238 | out_del: | |
1239 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1240 | out_free: | |
1241 | kfree(holder); | |
e09b457b | 1242 | out_unlock: |
b4cf1b72 | 1243 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 1244 | return ret; |
641dc636 | 1245 | } |
e09b457b | 1246 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 1247 | |
49731baa TH |
1248 | /** |
1249 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
1250 | * @bdev: the calimed slave bdev | |
1251 | * @disk: the holding disk | |
1252 | * | |
1253 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
1254 | * | |
1255 | * CONTEXT: | |
1256 | * Might sleep. | |
1257 | */ | |
1258 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 1259 | { |
49731baa | 1260 | struct bd_holder_disk *holder; |
641dc636 | 1261 | |
49731baa | 1262 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 1263 | |
49731baa TH |
1264 | holder = bd_find_holder_disk(bdev, disk); |
1265 | ||
1266 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
1267 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
1268 | del_symlink(bdev->bd_part->holder_dir, | |
1269 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 1270 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
1271 | list_del_init(&holder->list); |
1272 | kfree(holder); | |
1273 | } | |
1274 | ||
1275 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 1276 | } |
49731baa | 1277 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 1278 | #endif |
1da177e4 | 1279 | |
56ade44b AP |
1280 | /** |
1281 | * flush_disk - invalidates all buffer-cache entries on a disk | |
1282 | * | |
1283 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 1284 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
1285 | * |
1286 | * Invalidates all buffer-cache entries on a disk. It should be called | |
1287 | * when a disk has been changed -- either by a media change or online | |
1288 | * resize. | |
1289 | */ | |
93b270f7 | 1290 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 1291 | { |
93b270f7 | 1292 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b | 1293 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
424081f3 DM |
1294 | "resized disk %s\n", |
1295 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); | |
56ade44b AP |
1296 | } |
1297 | ||
1298 | if (!bdev->bd_disk) | |
1299 | return; | |
d27769ec | 1300 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
1301 | bdev->bd_invalidated = 1; |
1302 | } | |
1303 | ||
c3279d14 | 1304 | /** |
57d1b536 | 1305 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
1306 | * @disk: struct gendisk to check |
1307 | * @bdev: struct bdev to adjust. | |
1308 | * | |
1309 | * This routine checks to see if the bdev size does not match the disk size | |
1310 | * and adjusts it if it differs. | |
1311 | */ | |
1312 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
1313 | { | |
1314 | loff_t disk_size, bdev_size; | |
1315 | ||
1316 | disk_size = (loff_t)get_capacity(disk) << 9; | |
1317 | bdev_size = i_size_read(bdev->bd_inode); | |
1318 | if (disk_size != bdev_size) { | |
c3279d14 AP |
1319 | printk(KERN_INFO |
1320 | "%s: detected capacity change from %lld to %lld\n", | |
424081f3 | 1321 | disk->disk_name, bdev_size, disk_size); |
c3279d14 | 1322 | i_size_write(bdev->bd_inode, disk_size); |
93b270f7 | 1323 | flush_disk(bdev, false); |
c3279d14 AP |
1324 | } |
1325 | } | |
1326 | EXPORT_SYMBOL(check_disk_size_change); | |
1327 | ||
0c002c2f | 1328 | /** |
57d1b536 | 1329 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
1330 | * @disk: struct gendisk to be revalidated |
1331 | * | |
1332 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
1333 | * call-backs. It is used to do common pre and post operations needed | |
1334 | * for all revalidate_disk operations. | |
1335 | */ | |
1336 | int revalidate_disk(struct gendisk *disk) | |
1337 | { | |
c3279d14 | 1338 | struct block_device *bdev; |
0c002c2f AP |
1339 | int ret = 0; |
1340 | ||
1341 | if (disk->fops->revalidate_disk) | |
1342 | ret = disk->fops->revalidate_disk(disk); | |
c3279d14 AP |
1343 | bdev = bdget_disk(disk, 0); |
1344 | if (!bdev) | |
1345 | return ret; | |
1346 | ||
1347 | mutex_lock(&bdev->bd_mutex); | |
1348 | check_disk_size_change(disk, bdev); | |
7630b661 | 1349 | bdev->bd_invalidated = 0; |
c3279d14 AP |
1350 | mutex_unlock(&bdev->bd_mutex); |
1351 | bdput(bdev); | |
0c002c2f AP |
1352 | return ret; |
1353 | } | |
1354 | EXPORT_SYMBOL(revalidate_disk); | |
1355 | ||
1da177e4 LT |
1356 | /* |
1357 | * This routine checks whether a removable media has been changed, | |
1358 | * and invalidates all buffer-cache-entries in that case. This | |
1359 | * is a relatively slow routine, so we have to try to minimize using | |
1360 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
1361 | * is the best way of combining speed and utility, I think. | |
1362 | * People changing diskettes in the middle of an operation deserve | |
1363 | * to lose :-) | |
1364 | */ | |
1365 | int check_disk_change(struct block_device *bdev) | |
1366 | { | |
1367 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 1368 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 1369 | unsigned int events; |
1da177e4 | 1370 | |
77ea887e TH |
1371 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1372 | DISK_EVENT_EJECT_REQUEST); | |
1373 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1374 | return 0; |
1375 | ||
93b270f7 | 1376 | flush_disk(bdev, true); |
1da177e4 LT |
1377 | if (bdops->revalidate_disk) |
1378 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1379 | return 1; |
1380 | } | |
1381 | ||
1382 | EXPORT_SYMBOL(check_disk_change); | |
1383 | ||
1384 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1385 | { | |
e1defc4f | 1386 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1387 | |
5955102c | 1388 | inode_lock(bdev->bd_inode); |
d646a02a | 1389 | i_size_write(bdev->bd_inode, size); |
5955102c | 1390 | inode_unlock(bdev->bd_inode); |
09cbfeaf | 1391 | while (bsize < PAGE_SIZE) { |
1da177e4 LT |
1392 | if (size & bsize) |
1393 | break; | |
1394 | bsize <<= 1; | |
1395 | } | |
1396 | bdev->bd_block_size = bsize; | |
1397 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1398 | } | |
1399 | EXPORT_SYMBOL(bd_set_size); | |
1400 | ||
4385bab1 | 1401 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1402 | |
6d740cd5 PZ |
1403 | /* |
1404 | * bd_mutex locking: | |
1405 | * | |
1406 | * mutex_lock(part->bd_mutex) | |
1407 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1408 | */ | |
1409 | ||
572c4892 | 1410 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1411 | { |
1da177e4 | 1412 | struct gendisk *disk; |
523e1d39 | 1413 | struct module *owner; |
7db9cfd3 | 1414 | int ret; |
cf771cb5 | 1415 | int partno; |
fe6e9c1f AV |
1416 | int perm = 0; |
1417 | ||
572c4892 | 1418 | if (mode & FMODE_READ) |
fe6e9c1f | 1419 | perm |= MAY_READ; |
572c4892 | 1420 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1421 | perm |= MAY_WRITE; |
1422 | /* | |
1423 | * hooks: /n/, see "layering violations". | |
1424 | */ | |
b7300b78 CW |
1425 | if (!for_part) { |
1426 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1427 | if (ret != 0) { | |
1428 | bdput(bdev); | |
1429 | return ret; | |
1430 | } | |
82666020 | 1431 | } |
7db9cfd3 | 1432 | |
d3374825 | 1433 | restart: |
0762b8bd | 1434 | |
89f97496 | 1435 | ret = -ENXIO; |
cf771cb5 | 1436 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1437 | if (!disk) |
6e9624b8 | 1438 | goto out; |
523e1d39 | 1439 | owner = disk->fops->owner; |
1da177e4 | 1440 | |
69e02c59 | 1441 | disk_block_events(disk); |
6796bf54 | 1442 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1443 | if (!bdev->bd_openers) { |
1444 | bdev->bd_disk = disk; | |
87192a2a | 1445 | bdev->bd_queue = disk->queue; |
1da177e4 | 1446 | bdev->bd_contains = bdev; |
03cdadb0 | 1447 | |
cf771cb5 | 1448 | if (!partno) { |
89f97496 TH |
1449 | ret = -ENXIO; |
1450 | bdev->bd_part = disk_get_part(disk, partno); | |
1451 | if (!bdev->bd_part) | |
1452 | goto out_clear; | |
1453 | ||
1196f8b8 | 1454 | ret = 0; |
1da177e4 | 1455 | if (disk->fops->open) { |
572c4892 | 1456 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1457 | if (ret == -ERESTARTSYS) { |
1458 | /* Lost a race with 'disk' being | |
1459 | * deleted, try again. | |
1460 | * See md.c | |
1461 | */ | |
1462 | disk_put_part(bdev->bd_part); | |
1463 | bdev->bd_part = NULL; | |
d3374825 | 1464 | bdev->bd_disk = NULL; |
87192a2a | 1465 | bdev->bd_queue = NULL; |
d3374825 | 1466 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1467 | disk_unblock_events(disk); |
69e02c59 | 1468 | put_disk(disk); |
523e1d39 | 1469 | module_put(owner); |
d3374825 N |
1470 | goto restart; |
1471 | } | |
1da177e4 | 1472 | } |
7e69723f | 1473 | |
22375701 | 1474 | if (!ret) |
7e69723f | 1475 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
7e69723f | 1476 | |
1196f8b8 TH |
1477 | /* |
1478 | * If the device is invalidated, rescan partition | |
1479 | * if open succeeded or failed with -ENOMEDIUM. | |
1480 | * The latter is necessary to prevent ghost | |
1481 | * partitions on a removed medium. | |
1482 | */ | |
fe316bf2 JN |
1483 | if (bdev->bd_invalidated) { |
1484 | if (!ret) | |
1485 | rescan_partitions(disk, bdev); | |
1486 | else if (ret == -ENOMEDIUM) | |
1487 | invalidate_partitions(disk, bdev); | |
1488 | } | |
5a023cdb | 1489 | |
1196f8b8 TH |
1490 | if (ret) |
1491 | goto out_clear; | |
1da177e4 | 1492 | } else { |
1da177e4 LT |
1493 | struct block_device *whole; |
1494 | whole = bdget_disk(disk, 0); | |
1495 | ret = -ENOMEM; | |
1496 | if (!whole) | |
0762b8bd | 1497 | goto out_clear; |
37be4124 | 1498 | BUG_ON(for_part); |
572c4892 | 1499 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1500 | if (ret) |
0762b8bd | 1501 | goto out_clear; |
1da177e4 | 1502 | bdev->bd_contains = whole; |
89f97496 | 1503 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1504 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1505 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1506 | ret = -ENXIO; |
0762b8bd | 1507 | goto out_clear; |
1da177e4 | 1508 | } |
89f97496 | 1509 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1da177e4 | 1510 | } |
03e26279 JK |
1511 | |
1512 | if (bdev->bd_bdi == &noop_backing_dev_info) | |
1513 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info); | |
1da177e4 | 1514 | } else { |
1da177e4 | 1515 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1516 | ret = 0; |
1517 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1518 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1519 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1520 | if (bdev->bd_invalidated) { |
1521 | if (!ret) | |
1522 | rescan_partitions(bdev->bd_disk, bdev); | |
1523 | else if (ret == -ENOMEDIUM) | |
1524 | invalidate_partitions(bdev->bd_disk, bdev); | |
1525 | } | |
1196f8b8 TH |
1526 | if (ret) |
1527 | goto out_unlock_bdev; | |
1da177e4 | 1528 | } |
69e02c59 | 1529 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1530 | put_disk(disk); |
523e1d39 | 1531 | module_put(owner); |
1da177e4 LT |
1532 | } |
1533 | bdev->bd_openers++; | |
37be4124 N |
1534 | if (for_part) |
1535 | bdev->bd_part_count++; | |
c039e313 | 1536 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1537 | disk_unblock_events(disk); |
1da177e4 LT |
1538 | return 0; |
1539 | ||
0762b8bd | 1540 | out_clear: |
89f97496 | 1541 | disk_put_part(bdev->bd_part); |
1da177e4 | 1542 | bdev->bd_disk = NULL; |
0762b8bd | 1543 | bdev->bd_part = NULL; |
87192a2a | 1544 | bdev->bd_queue = NULL; |
1da177e4 | 1545 | if (bdev != bdev->bd_contains) |
572c4892 | 1546 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1547 | bdev->bd_contains = NULL; |
0762b8bd | 1548 | out_unlock_bdev: |
c039e313 | 1549 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1550 | disk_unblock_events(disk); |
0762b8bd | 1551 | put_disk(disk); |
523e1d39 | 1552 | module_put(owner); |
4345caba | 1553 | out: |
0762b8bd TH |
1554 | bdput(bdev); |
1555 | ||
1da177e4 LT |
1556 | return ret; |
1557 | } | |
1558 | ||
d4d77629 TH |
1559 | /** |
1560 | * blkdev_get - open a block device | |
1561 | * @bdev: block_device to open | |
1562 | * @mode: FMODE_* mask | |
1563 | * @holder: exclusive holder identifier | |
1564 | * | |
1565 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1566 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1567 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1568 | * | |
1569 | * On success, the reference count of @bdev is unchanged. On failure, | |
1570 | * @bdev is put. | |
1571 | * | |
1572 | * CONTEXT: | |
1573 | * Might sleep. | |
1574 | * | |
1575 | * RETURNS: | |
1576 | * 0 on success, -errno on failure. | |
1577 | */ | |
e525fd89 | 1578 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1579 | { |
e525fd89 TH |
1580 | struct block_device *whole = NULL; |
1581 | int res; | |
1582 | ||
1583 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1584 | ||
1585 | if ((mode & FMODE_EXCL) && holder) { | |
1586 | whole = bd_start_claiming(bdev, holder); | |
1587 | if (IS_ERR(whole)) { | |
1588 | bdput(bdev); | |
1589 | return PTR_ERR(whole); | |
1590 | } | |
1591 | } | |
1592 | ||
1593 | res = __blkdev_get(bdev, mode, 0); | |
1594 | ||
1595 | if (whole) { | |
d4dc210f TH |
1596 | struct gendisk *disk = whole->bd_disk; |
1597 | ||
6a027eff | 1598 | /* finish claiming */ |
77ea887e | 1599 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1600 | spin_lock(&bdev_lock); |
1601 | ||
77ea887e | 1602 | if (!res) { |
6a027eff TH |
1603 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1604 | /* | |
1605 | * Note that for a whole device bd_holders | |
1606 | * will be incremented twice, and bd_holder | |
1607 | * will be set to bd_may_claim before being | |
1608 | * set to holder | |
1609 | */ | |
1610 | whole->bd_holders++; | |
1611 | whole->bd_holder = bd_may_claim; | |
1612 | bdev->bd_holders++; | |
1613 | bdev->bd_holder = holder; | |
1614 | } | |
1615 | ||
1616 | /* tell others that we're done */ | |
1617 | BUG_ON(whole->bd_claiming != holder); | |
1618 | whole->bd_claiming = NULL; | |
1619 | wake_up_bit(&whole->bd_claiming, 0); | |
1620 | ||
1621 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1622 | |
1623 | /* | |
d4dc210f TH |
1624 | * Block event polling for write claims if requested. Any |
1625 | * write holder makes the write_holder state stick until | |
1626 | * all are released. This is good enough and tracking | |
1627 | * individual writeable reference is too fragile given the | |
1628 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1629 | */ |
4c49ff3f TH |
1630 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1631 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1632 | bdev->bd_write_holder = true; |
d4dc210f | 1633 | disk_block_events(disk); |
77ea887e TH |
1634 | } |
1635 | ||
1636 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1637 | bdput(whole); |
e525fd89 TH |
1638 | } |
1639 | ||
1640 | return res; | |
37be4124 | 1641 | } |
1da177e4 LT |
1642 | EXPORT_SYMBOL(blkdev_get); |
1643 | ||
d4d77629 TH |
1644 | /** |
1645 | * blkdev_get_by_path - open a block device by name | |
1646 | * @path: path to the block device to open | |
1647 | * @mode: FMODE_* mask | |
1648 | * @holder: exclusive holder identifier | |
1649 | * | |
1650 | * Open the blockdevice described by the device file at @path. @mode | |
1651 | * and @holder are identical to blkdev_get(). | |
1652 | * | |
1653 | * On success, the returned block_device has reference count of one. | |
1654 | * | |
1655 | * CONTEXT: | |
1656 | * Might sleep. | |
1657 | * | |
1658 | * RETURNS: | |
1659 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1660 | */ | |
1661 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1662 | void *holder) | |
1663 | { | |
1664 | struct block_device *bdev; | |
1665 | int err; | |
1666 | ||
1667 | bdev = lookup_bdev(path); | |
1668 | if (IS_ERR(bdev)) | |
1669 | return bdev; | |
1670 | ||
1671 | err = blkdev_get(bdev, mode, holder); | |
1672 | if (err) | |
1673 | return ERR_PTR(err); | |
1674 | ||
e51900f7 CE |
1675 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1676 | blkdev_put(bdev, mode); | |
1677 | return ERR_PTR(-EACCES); | |
1678 | } | |
1679 | ||
d4d77629 TH |
1680 | return bdev; |
1681 | } | |
1682 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1683 | ||
1684 | /** | |
1685 | * blkdev_get_by_dev - open a block device by device number | |
1686 | * @dev: device number of block device to open | |
1687 | * @mode: FMODE_* mask | |
1688 | * @holder: exclusive holder identifier | |
1689 | * | |
1690 | * Open the blockdevice described by device number @dev. @mode and | |
1691 | * @holder are identical to blkdev_get(). | |
1692 | * | |
1693 | * Use it ONLY if you really do not have anything better - i.e. when | |
1694 | * you are behind a truly sucky interface and all you are given is a | |
1695 | * device number. _Never_ to be used for internal purposes. If you | |
1696 | * ever need it - reconsider your API. | |
1697 | * | |
1698 | * On success, the returned block_device has reference count of one. | |
1699 | * | |
1700 | * CONTEXT: | |
1701 | * Might sleep. | |
1702 | * | |
1703 | * RETURNS: | |
1704 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1705 | */ | |
1706 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1707 | { | |
1708 | struct block_device *bdev; | |
1709 | int err; | |
1710 | ||
1711 | bdev = bdget(dev); | |
1712 | if (!bdev) | |
1713 | return ERR_PTR(-ENOMEM); | |
1714 | ||
1715 | err = blkdev_get(bdev, mode, holder); | |
1716 | if (err) | |
1717 | return ERR_PTR(err); | |
1718 | ||
1719 | return bdev; | |
1720 | } | |
1721 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1722 | ||
1da177e4 LT |
1723 | static int blkdev_open(struct inode * inode, struct file * filp) |
1724 | { | |
1725 | struct block_device *bdev; | |
1da177e4 LT |
1726 | |
1727 | /* | |
1728 | * Preserve backwards compatibility and allow large file access | |
1729 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1730 | * binary needs it. We might want to drop this workaround | |
1731 | * during an unstable branch. | |
1732 | */ | |
1733 | filp->f_flags |= O_LARGEFILE; | |
1734 | ||
572c4892 AV |
1735 | if (filp->f_flags & O_NDELAY) |
1736 | filp->f_mode |= FMODE_NDELAY; | |
1737 | if (filp->f_flags & O_EXCL) | |
1738 | filp->f_mode |= FMODE_EXCL; | |
1739 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1740 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1741 | ||
1da177e4 | 1742 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1743 | if (bdev == NULL) |
1744 | return -ENOMEM; | |
1da177e4 | 1745 | |
572c4892 AV |
1746 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1747 | ||
e525fd89 | 1748 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1749 | } |
1750 | ||
4385bab1 | 1751 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1752 | { |
2e7b651d | 1753 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1754 | struct block_device *victim = NULL; |
2e7b651d | 1755 | |
6796bf54 | 1756 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1757 | if (for_part) |
1758 | bdev->bd_part_count--; | |
1759 | ||
2e7b651d | 1760 | if (!--bdev->bd_openers) { |
6a027eff | 1761 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1762 | sync_blockdev(bdev); |
1763 | kill_bdev(bdev); | |
43d1c0eb ID |
1764 | |
1765 | bdev_write_inode(bdev); | |
2e7b651d PZ |
1766 | } |
1767 | if (bdev->bd_contains == bdev) { | |
1768 | if (disk->fops->release) | |
db2a144b | 1769 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1770 | } |
1771 | if (!bdev->bd_openers) { | |
1772 | struct module *owner = disk->fops->owner; | |
1773 | ||
0762b8bd TH |
1774 | disk_put_part(bdev->bd_part); |
1775 | bdev->bd_part = NULL; | |
2e7b651d | 1776 | bdev->bd_disk = NULL; |
37be4124 N |
1777 | if (bdev != bdev->bd_contains) |
1778 | victim = bdev->bd_contains; | |
2e7b651d | 1779 | bdev->bd_contains = NULL; |
523e1d39 TH |
1780 | |
1781 | put_disk(disk); | |
1782 | module_put(owner); | |
2e7b651d | 1783 | } |
2e7b651d PZ |
1784 | mutex_unlock(&bdev->bd_mutex); |
1785 | bdput(bdev); | |
37be4124 | 1786 | if (victim) |
9a1c3542 | 1787 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1788 | } |
1789 | ||
4385bab1 | 1790 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1791 | { |
85ef06d1 TH |
1792 | mutex_lock(&bdev->bd_mutex); |
1793 | ||
e525fd89 | 1794 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1795 | bool bdev_free; |
1796 | ||
1797 | /* | |
1798 | * Release a claim on the device. The holder fields | |
1799 | * are protected with bdev_lock. bd_mutex is to | |
1800 | * synchronize disk_holder unlinking. | |
1801 | */ | |
6a027eff TH |
1802 | spin_lock(&bdev_lock); |
1803 | ||
1804 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1805 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1806 | ||
1807 | /* bd_contains might point to self, check in a separate step */ | |
1808 | if ((bdev_free = !bdev->bd_holders)) | |
1809 | bdev->bd_holder = NULL; | |
1810 | if (!bdev->bd_contains->bd_holders) | |
1811 | bdev->bd_contains->bd_holder = NULL; | |
1812 | ||
1813 | spin_unlock(&bdev_lock); | |
1814 | ||
77ea887e TH |
1815 | /* |
1816 | * If this was the last claim, remove holder link and | |
1817 | * unblock evpoll if it was a write holder. | |
1818 | */ | |
85ef06d1 TH |
1819 | if (bdev_free && bdev->bd_write_holder) { |
1820 | disk_unblock_events(bdev->bd_disk); | |
1821 | bdev->bd_write_holder = false; | |
77ea887e | 1822 | } |
6936217c | 1823 | } |
77ea887e | 1824 | |
85ef06d1 TH |
1825 | /* |
1826 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1827 | * event. This is to ensure detection of media removal commanded | |
1828 | * from userland - e.g. eject(1). | |
1829 | */ | |
1830 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1831 | ||
1832 | mutex_unlock(&bdev->bd_mutex); | |
1833 | ||
4385bab1 | 1834 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1835 | } |
2e7b651d PZ |
1836 | EXPORT_SYMBOL(blkdev_put); |
1837 | ||
1da177e4 LT |
1838 | static int blkdev_close(struct inode * inode, struct file * filp) |
1839 | { | |
4ebb16ca | 1840 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
4385bab1 AV |
1841 | blkdev_put(bdev, filp->f_mode); |
1842 | return 0; | |
1da177e4 LT |
1843 | } |
1844 | ||
bb93e3a5 | 1845 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1846 | { |
4ebb16ca | 1847 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
56b26add | 1848 | fmode_t mode = file->f_mode; |
fd4ce1ac CH |
1849 | |
1850 | /* | |
1851 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1852 | * to updated it before every ioctl. | |
1853 | */ | |
56b26add | 1854 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1855 | mode |= FMODE_NDELAY; |
1856 | else | |
1857 | mode &= ~FMODE_NDELAY; | |
1858 | ||
56b26add | 1859 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1860 | } |
1861 | ||
eef99380 CH |
1862 | /* |
1863 | * Write data to the block device. Only intended for the block device itself | |
1864 | * and the raw driver which basically is a fake block device. | |
1865 | * | |
1866 | * Does not take i_mutex for the write and thus is not for general purpose | |
1867 | * use. | |
1868 | */ | |
1456c0a8 | 1869 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
eef99380 CH |
1870 | { |
1871 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1872 | struct inode *bd_inode = bdev_file_inode(file); |
7ec7b94a | 1873 | loff_t size = i_size_read(bd_inode); |
53362a05 | 1874 | struct blk_plug plug; |
eef99380 | 1875 | ssize_t ret; |
5f380c7f | 1876 | |
7ec7b94a AV |
1877 | if (bdev_read_only(I_BDEV(bd_inode))) |
1878 | return -EPERM; | |
5f380c7f | 1879 | |
7ec7b94a | 1880 | if (!iov_iter_count(from)) |
5f380c7f AV |
1881 | return 0; |
1882 | ||
7ec7b94a AV |
1883 | if (iocb->ki_pos >= size) |
1884 | return -ENOSPC; | |
1885 | ||
1886 | iov_iter_truncate(from, size - iocb->ki_pos); | |
eef99380 | 1887 | |
53362a05 | 1888 | blk_start_plug(&plug); |
1456c0a8 | 1889 | ret = __generic_file_write_iter(iocb, from); |
e2592217 CH |
1890 | if (ret > 0) |
1891 | ret = generic_write_sync(iocb, ret); | |
53362a05 | 1892 | blk_finish_plug(&plug); |
eef99380 CH |
1893 | return ret; |
1894 | } | |
1456c0a8 | 1895 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
eef99380 | 1896 | |
b2de525f | 1897 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
684c9aae LT |
1898 | { |
1899 | struct file *file = iocb->ki_filp; | |
4ebb16ca | 1900 | struct inode *bd_inode = bdev_file_inode(file); |
684c9aae | 1901 | loff_t size = i_size_read(bd_inode); |
a886038b | 1902 | loff_t pos = iocb->ki_pos; |
684c9aae LT |
1903 | |
1904 | if (pos >= size) | |
1905 | return 0; | |
1906 | ||
1907 | size -= pos; | |
a886038b AV |
1908 | iov_iter_truncate(to, size); |
1909 | return generic_file_read_iter(iocb, to); | |
684c9aae | 1910 | } |
b2de525f | 1911 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
684c9aae | 1912 | |
87d8fe1e TT |
1913 | /* |
1914 | * Try to release a page associated with block device when the system | |
1915 | * is under memory pressure. | |
1916 | */ | |
1917 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1918 | { | |
1919 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1920 | ||
1921 | if (super && super->s_op->bdev_try_to_free_page) | |
1922 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1923 | ||
1924 | return try_to_free_buffers(page); | |
1925 | } | |
1926 | ||
7f6d5b52 RZ |
1927 | static int blkdev_writepages(struct address_space *mapping, |
1928 | struct writeback_control *wbc) | |
1929 | { | |
1930 | if (dax_mapping(mapping)) { | |
1931 | struct block_device *bdev = I_BDEV(mapping->host); | |
1932 | ||
1933 | return dax_writeback_mapping_range(mapping, bdev, wbc); | |
1934 | } | |
1935 | return generic_writepages(mapping, wbc); | |
1936 | } | |
1937 | ||
4c54ac62 | 1938 | static const struct address_space_operations def_blk_aops = { |
1da177e4 | 1939 | .readpage = blkdev_readpage, |
447f05bb | 1940 | .readpages = blkdev_readpages, |
1da177e4 | 1941 | .writepage = blkdev_writepage, |
6272b5a5 NP |
1942 | .write_begin = blkdev_write_begin, |
1943 | .write_end = blkdev_write_end, | |
7f6d5b52 | 1944 | .writepages = blkdev_writepages, |
87d8fe1e | 1945 | .releasepage = blkdev_releasepage, |
1da177e4 | 1946 | .direct_IO = blkdev_direct_IO, |
b4597226 | 1947 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1948 | }; |
1949 | ||
25f4c414 DW |
1950 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
1951 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
1952 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) | |
1953 | ||
1954 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, | |
1955 | loff_t len) | |
1956 | { | |
1957 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); | |
25f4c414 DW |
1958 | struct address_space *mapping; |
1959 | loff_t end = start + len - 1; | |
1960 | loff_t isize; | |
1961 | int error; | |
1962 | ||
1963 | /* Fail if we don't recognize the flags. */ | |
1964 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) | |
1965 | return -EOPNOTSUPP; | |
1966 | ||
1967 | /* Don't go off the end of the device. */ | |
1968 | isize = i_size_read(bdev->bd_inode); | |
1969 | if (start >= isize) | |
1970 | return -EINVAL; | |
1971 | if (end >= isize) { | |
1972 | if (mode & FALLOC_FL_KEEP_SIZE) { | |
1973 | len = isize - start; | |
1974 | end = start + len - 1; | |
1975 | } else | |
1976 | return -EINVAL; | |
1977 | } | |
1978 | ||
1979 | /* | |
1980 | * Don't allow IO that isn't aligned to logical block size. | |
1981 | */ | |
1982 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) | |
1983 | return -EINVAL; | |
1984 | ||
1985 | /* Invalidate the page cache, including dirty pages. */ | |
1986 | mapping = bdev->bd_inode->i_mapping; | |
1987 | truncate_inode_pages_range(mapping, start, end); | |
1988 | ||
1989 | switch (mode) { | |
1990 | case FALLOC_FL_ZERO_RANGE: | |
1991 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: | |
1992 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, | |
ee472d83 | 1993 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP); |
25f4c414 DW |
1994 | break; |
1995 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: | |
34045129 CH |
1996 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
1997 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK); | |
25f4c414 DW |
1998 | break; |
1999 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: | |
25f4c414 DW |
2000 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
2001 | GFP_KERNEL, 0); | |
2002 | break; | |
2003 | default: | |
2004 | return -EOPNOTSUPP; | |
2005 | } | |
2006 | if (error) | |
2007 | return error; | |
2008 | ||
2009 | /* | |
2010 | * Invalidate again; if someone wandered in and dirtied a page, | |
2011 | * the caller will be given -EBUSY. The third argument is | |
2012 | * inclusive, so the rounding here is safe. | |
2013 | */ | |
2014 | return invalidate_inode_pages2_range(mapping, | |
2015 | start >> PAGE_SHIFT, | |
2016 | end >> PAGE_SHIFT); | |
2017 | } | |
2018 | ||
4b6f5d20 | 2019 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
2020 | .open = blkdev_open, |
2021 | .release = blkdev_close, | |
2022 | .llseek = block_llseek, | |
a886038b | 2023 | .read_iter = blkdev_read_iter, |
1456c0a8 | 2024 | .write_iter = blkdev_write_iter, |
acc93d30 | 2025 | .mmap = generic_file_mmap, |
b1dd3b28 | 2026 | .fsync = blkdev_fsync, |
bb93e3a5 | 2027 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
2028 | #ifdef CONFIG_COMPAT |
2029 | .compat_ioctl = compat_blkdev_ioctl, | |
2030 | #endif | |
1e8b3332 | 2031 | .splice_read = generic_file_splice_read, |
8d020765 | 2032 | .splice_write = iter_file_splice_write, |
25f4c414 | 2033 | .fallocate = blkdev_fallocate, |
1da177e4 LT |
2034 | }; |
2035 | ||
2036 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
2037 | { | |
2038 | int res; | |
2039 | mm_segment_t old_fs = get_fs(); | |
2040 | set_fs(KERNEL_DS); | |
56b26add | 2041 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
2042 | set_fs(old_fs); |
2043 | return res; | |
2044 | } | |
2045 | ||
2046 | EXPORT_SYMBOL(ioctl_by_bdev); | |
2047 | ||
2048 | /** | |
2049 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 2050 | * @pathname: special file representing the block device |
1da177e4 | 2051 | * |
57d1b536 | 2052 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
2053 | * namespace if possible and return it. Return ERR_PTR(error) |
2054 | * otherwise. | |
2055 | */ | |
421748ec | 2056 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
2057 | { |
2058 | struct block_device *bdev; | |
2059 | struct inode *inode; | |
421748ec | 2060 | struct path path; |
1da177e4 LT |
2061 | int error; |
2062 | ||
421748ec | 2063 | if (!pathname || !*pathname) |
1da177e4 LT |
2064 | return ERR_PTR(-EINVAL); |
2065 | ||
421748ec | 2066 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
2067 | if (error) |
2068 | return ERR_PTR(error); | |
2069 | ||
bb668734 | 2070 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
2071 | error = -ENOTBLK; |
2072 | if (!S_ISBLK(inode->i_mode)) | |
2073 | goto fail; | |
2074 | error = -EACCES; | |
a2982cc9 | 2075 | if (!may_open_dev(&path)) |
1da177e4 LT |
2076 | goto fail; |
2077 | error = -ENOMEM; | |
2078 | bdev = bd_acquire(inode); | |
2079 | if (!bdev) | |
2080 | goto fail; | |
2081 | out: | |
421748ec | 2082 | path_put(&path); |
1da177e4 LT |
2083 | return bdev; |
2084 | fail: | |
2085 | bdev = ERR_PTR(error); | |
2086 | goto out; | |
2087 | } | |
d5686b44 | 2088 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 2089 | |
93b270f7 | 2090 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
2091 | { |
2092 | struct super_block *sb = get_super(bdev); | |
2093 | int res = 0; | |
2094 | ||
2095 | if (sb) { | |
2096 | /* | |
2097 | * no need to lock the super, get_super holds the | |
2098 | * read mutex so the filesystem cannot go away | |
2099 | * under us (->put_super runs with the write lock | |
2100 | * hold). | |
2101 | */ | |
2102 | shrink_dcache_sb(sb); | |
93b270f7 | 2103 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
2104 | drop_super(sb); |
2105 | } | |
f98393a6 | 2106 | invalidate_bdev(bdev); |
b71e8a4c DH |
2107 | return res; |
2108 | } | |
2109 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
2110 | |
2111 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
2112 | { | |
2113 | struct inode *inode, *old_inode = NULL; | |
2114 | ||
74278da9 | 2115 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2116 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
2117 | struct address_space *mapping = inode->i_mapping; | |
af309226 | 2118 | struct block_device *bdev; |
5c0d6b60 JK |
2119 | |
2120 | spin_lock(&inode->i_lock); | |
2121 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
2122 | mapping->nrpages == 0) { | |
2123 | spin_unlock(&inode->i_lock); | |
2124 | continue; | |
2125 | } | |
2126 | __iget(inode); | |
2127 | spin_unlock(&inode->i_lock); | |
74278da9 | 2128 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2129 | /* |
2130 | * We hold a reference to 'inode' so it couldn't have been | |
2131 | * removed from s_inodes list while we dropped the | |
74278da9 | 2132 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 2133 | * be holding the last reference and we cannot iput it under |
74278da9 | 2134 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
2135 | * later. |
2136 | */ | |
2137 | iput(old_inode); | |
2138 | old_inode = inode; | |
af309226 | 2139 | bdev = I_BDEV(inode); |
5c0d6b60 | 2140 | |
af309226 RV |
2141 | mutex_lock(&bdev->bd_mutex); |
2142 | if (bdev->bd_openers) | |
2143 | func(bdev, arg); | |
2144 | mutex_unlock(&bdev->bd_mutex); | |
5c0d6b60 | 2145 | |
74278da9 | 2146 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 2147 | } |
74278da9 | 2148 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
2149 | iput(old_inode); |
2150 | } |