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 | |
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> |
14 | #include <linux/device_cgroup.h> |
15 | #include <linux/highmem.h> |
16 | #include <linux/blkdev.h> |
17 | #include <linux/backing-dev.h> |
18 | #include <linux/module.h> |
19 | #include <linux/blkpg.h> |
20 | #include <linux/magic.h> |
21 | #include <linux/dax.h> |
22 | #include <linux/buffer_head.h> |
23 | #include <linux/swap.h> |
24 | #include <linux/pagevec.h> |
25 | #include <linux/writeback.h> |
26 | #include <linux/mpage.h> |
27 | #include <linux/mount.h> |
28 | #include <linux/uio.h> |
29 | #include <linux/namei.h> |
30 | #include <linux/log2.h> |
31 | #include <linux/cleancache.h> |
32 | #include <linux/dax.h> |
33 | #include <linux/badblocks.h> |
34 | #include <linux/task_io_accounting_ops.h> |
35 | #include <linux/falloc.h> |
36 | #include <linux/uaccess.h> |
37 | #include "internal.h" |
38 | |
39 | struct bdev_inode { |
40 | struct block_device bdev; |
41 | struct inode vfs_inode; |
42 | }; |
43 | |
44 | static const struct address_space_operations def_blk_aops; |
45 | |
46 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
47 | { |
48 | return container_of(inode, struct bdev_inode, vfs_inode); |
49 | } |
50 | |
51 | struct block_device *I_BDEV(struct inode *inode) |
52 | { |
53 | return &BDEV_I(inode)->bdev; |
54 | } |
55 | EXPORT_SYMBOL(I_BDEV); |
56 | |
57 | static void bdev_write_inode(struct block_device *bdev) |
58 | { |
59 | struct inode *inode = bdev->bd_inode; |
60 | int ret; |
61 | |
62 | spin_lock(&inode->i_lock); |
63 | while (inode->i_state & I_DIRTY) { |
64 | spin_unlock(&inode->i_lock); |
65 | ret = write_inode_now(inode, true); |
66 | if (ret) { |
67 | char name[BDEVNAME_SIZE]; |
68 | pr_warn_ratelimited("VFS: Dirty inode writeback failed " |
69 | "for block device %s (err=%d).\n" , |
70 | bdevname(bdev, name), ret); |
71 | } |
72 | spin_lock(&inode->i_lock); |
73 | } |
74 | spin_unlock(&inode->i_lock); |
75 | } |
76 | |
77 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
78 | void kill_bdev(struct block_device *bdev) |
79 | { |
80 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
81 | |
82 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
83 | return; |
84 | |
85 | invalidate_bh_lrus(); |
86 | truncate_inode_pages(mapping, 0); |
87 | } |
88 | EXPORT_SYMBOL(kill_bdev); |
89 | |
90 | /* Invalidate clean unused buffers and pagecache. */ |
91 | void invalidate_bdev(struct block_device *bdev) |
92 | { |
93 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
94 | |
95 | if (mapping->nrpages) { |
96 | invalidate_bh_lrus(); |
97 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ |
98 | invalidate_mapping_pages(mapping, 0, -1); |
99 | } |
100 | /* 99% of the time, we don't need to flush the cleancache on the bdev. |
101 | * But, for the strange corners, lets be cautious |
102 | */ |
103 | cleancache_invalidate_inode(mapping); |
104 | } |
105 | EXPORT_SYMBOL(invalidate_bdev); |
106 | |
107 | static void set_init_blocksize(struct block_device *bdev) |
108 | { |
109 | unsigned bsize = bdev_logical_block_size(bdev); |
110 | loff_t size = i_size_read(bdev->bd_inode); |
111 | |
112 | while (bsize < PAGE_SIZE) { |
113 | if (size & bsize) |
114 | break; |
115 | bsize <<= 1; |
116 | } |
117 | bdev->bd_block_size = bsize; |
118 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); |
119 | } |
120 | |
121 | int set_blocksize(struct block_device *bdev, int size) |
122 | { |
123 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ |
124 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
125 | return -EINVAL; |
126 | |
127 | /* Size cannot be smaller than the size supported by the device */ |
128 | if (size < bdev_logical_block_size(bdev)) |
129 | return -EINVAL; |
130 | |
131 | /* Don't change the size if it is same as current */ |
132 | if (bdev->bd_block_size != size) { |
133 | sync_blockdev(bdev); |
134 | bdev->bd_block_size = size; |
135 | bdev->bd_inode->i_blkbits = blksize_bits(size); |
136 | kill_bdev(bdev); |
137 | } |
138 | return 0; |
139 | } |
140 | |
141 | EXPORT_SYMBOL(set_blocksize); |
142 | |
143 | int sb_set_blocksize(struct super_block *sb, int size) |
144 | { |
145 | if (set_blocksize(sb->s_bdev, size)) |
146 | return 0; |
147 | /* If we get here, we know size is power of two |
148 | * and it's value is between 512 and PAGE_SIZE */ |
149 | sb->s_blocksize = size; |
150 | sb->s_blocksize_bits = blksize_bits(size); |
151 | return sb->s_blocksize; |
152 | } |
153 | |
154 | EXPORT_SYMBOL(sb_set_blocksize); |
155 | |
156 | int sb_min_blocksize(struct super_block *sb, int size) |
157 | { |
158 | int minsize = bdev_logical_block_size(sb->s_bdev); |
159 | if (size < minsize) |
160 | size = minsize; |
161 | return sb_set_blocksize(sb, size); |
162 | } |
163 | |
164 | EXPORT_SYMBOL(sb_min_blocksize); |
165 | |
166 | static int |
167 | blkdev_get_block(struct inode *inode, sector_t iblock, |
168 | struct buffer_head *bh, int create) |
169 | { |
170 | bh->b_bdev = I_BDEV(inode); |
171 | bh->b_blocknr = iblock; |
172 | set_buffer_mapped(bh); |
173 | return 0; |
174 | } |
175 | |
176 | static struct inode *bdev_file_inode(struct file *file) |
177 | { |
178 | return file->f_mapping->host; |
179 | } |
180 | |
181 | static unsigned int dio_bio_write_op(struct kiocb *iocb) |
182 | { |
183 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; |
184 | |
185 | /* avoid the need for a I/O completion work item */ |
186 | if (iocb->ki_flags & IOCB_DSYNC) |
187 | op |= REQ_FUA; |
188 | return op; |
189 | } |
190 | |
191 | #define DIO_INLINE_BIO_VECS 4 |
192 | |
193 | static void blkdev_bio_end_io_simple(struct bio *bio) |
194 | { |
195 | struct task_struct *waiter = bio->bi_private; |
196 | |
197 | WRITE_ONCE(bio->bi_private, NULL); |
198 | blk_wake_io_task(waiter); |
199 | } |
200 | |
201 | static ssize_t |
202 | __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter, |
203 | int nr_pages) |
204 | { |
205 | struct file *file = iocb->ki_filp; |
206 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
207 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec; |
208 | loff_t pos = iocb->ki_pos; |
209 | bool should_dirty = false; |
210 | struct bio bio; |
211 | ssize_t ret; |
212 | blk_qc_t qc; |
213 | int i; |
214 | struct bvec_iter_all iter_all; |
215 | |
216 | if ((pos | iov_iter_alignment(iter)) & |
217 | (bdev_logical_block_size(bdev) - 1)) |
218 | return -EINVAL; |
219 | |
220 | if (nr_pages <= DIO_INLINE_BIO_VECS) |
221 | vecs = inline_vecs; |
222 | else { |
223 | vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec), |
224 | GFP_KERNEL); |
225 | if (!vecs) |
226 | return -ENOMEM; |
227 | } |
228 | |
229 | bio_init(&bio, vecs, nr_pages); |
230 | bio_set_dev(&bio, bdev); |
231 | bio.bi_iter.bi_sector = pos >> 9; |
232 | bio.bi_write_hint = iocb->ki_hint; |
233 | bio.bi_private = current; |
234 | bio.bi_end_io = blkdev_bio_end_io_simple; |
235 | bio.bi_ioprio = iocb->ki_ioprio; |
236 | |
237 | ret = bio_iov_iter_get_pages(&bio, iter); |
238 | if (unlikely(ret)) |
239 | goto out; |
240 | ret = bio.bi_iter.bi_size; |
241 | |
242 | if (iov_iter_rw(iter) == READ) { |
243 | bio.bi_opf = REQ_OP_READ; |
244 | if (iter_is_iovec(iter)) |
245 | should_dirty = true; |
246 | } else { |
247 | bio.bi_opf = dio_bio_write_op(iocb); |
248 | task_io_account_write(ret); |
249 | } |
250 | if (iocb->ki_flags & IOCB_HIPRI) |
251 | bio_set_polled(&bio, iocb); |
252 | |
253 | qc = submit_bio(&bio); |
254 | for (;;) { |
255 | set_current_state(TASK_UNINTERRUPTIBLE); |
256 | if (!READ_ONCE(bio.bi_private)) |
257 | break; |
258 | if (!(iocb->ki_flags & IOCB_HIPRI) || |
259 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
260 | io_schedule(); |
261 | } |
262 | __set_current_state(TASK_RUNNING); |
263 | |
264 | bio_for_each_segment_all(bvec, &bio, i, iter_all) { |
265 | if (should_dirty && !PageCompound(bvec->bv_page)) |
266 | set_page_dirty_lock(bvec->bv_page); |
267 | put_page(bvec->bv_page); |
268 | } |
269 | |
270 | if (unlikely(bio.bi_status)) |
271 | ret = blk_status_to_errno(bio.bi_status); |
272 | |
273 | out: |
274 | if (vecs != inline_vecs) |
275 | kfree(vecs); |
276 | |
277 | bio_uninit(&bio); |
278 | |
279 | return ret; |
280 | } |
281 | |
282 | struct blkdev_dio { |
283 | union { |
284 | struct kiocb *iocb; |
285 | struct task_struct *waiter; |
286 | }; |
287 | size_t size; |
288 | atomic_t ref; |
289 | bool multi_bio : 1; |
290 | bool should_dirty : 1; |
291 | bool is_sync : 1; |
292 | struct bio bio; |
293 | }; |
294 | |
295 | static struct bio_set blkdev_dio_pool; |
296 | |
297 | static int blkdev_iopoll(struct kiocb *kiocb, bool wait) |
298 | { |
299 | struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host); |
300 | struct request_queue *q = bdev_get_queue(bdev); |
301 | |
302 | return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait); |
303 | } |
304 | |
305 | static void blkdev_bio_end_io(struct bio *bio) |
306 | { |
307 | struct blkdev_dio *dio = bio->bi_private; |
308 | bool should_dirty = dio->should_dirty; |
309 | |
310 | if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) { |
311 | if (bio->bi_status && !dio->bio.bi_status) |
312 | dio->bio.bi_status = bio->bi_status; |
313 | } else { |
314 | if (!dio->is_sync) { |
315 | struct kiocb *iocb = dio->iocb; |
316 | ssize_t ret; |
317 | |
318 | if (likely(!dio->bio.bi_status)) { |
319 | ret = dio->size; |
320 | iocb->ki_pos += ret; |
321 | } else { |
322 | ret = blk_status_to_errno(dio->bio.bi_status); |
323 | } |
324 | |
325 | dio->iocb->ki_complete(iocb, ret, 0); |
326 | if (dio->multi_bio) |
327 | bio_put(&dio->bio); |
328 | } else { |
329 | struct task_struct *waiter = dio->waiter; |
330 | |
331 | WRITE_ONCE(dio->waiter, NULL); |
332 | blk_wake_io_task(waiter); |
333 | } |
334 | } |
335 | |
336 | if (should_dirty) { |
337 | bio_check_pages_dirty(bio); |
338 | } else { |
339 | if (!bio_flagged(bio, BIO_NO_PAGE_REF)) { |
340 | struct bvec_iter_all iter_all; |
341 | struct bio_vec *bvec; |
342 | int i; |
343 | |
344 | bio_for_each_segment_all(bvec, bio, i, iter_all) |
345 | put_page(bvec->bv_page); |
346 | } |
347 | bio_put(bio); |
348 | } |
349 | } |
350 | |
351 | static ssize_t |
352 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages) |
353 | { |
354 | struct file *file = iocb->ki_filp; |
355 | struct inode *inode = bdev_file_inode(file); |
356 | struct block_device *bdev = I_BDEV(inode); |
357 | struct blk_plug plug; |
358 | struct blkdev_dio *dio; |
359 | struct bio *bio; |
360 | bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0; |
361 | bool is_read = (iov_iter_rw(iter) == READ), is_sync; |
362 | loff_t pos = iocb->ki_pos; |
363 | blk_qc_t qc = BLK_QC_T_NONE; |
364 | int ret = 0; |
365 | |
366 | if ((pos | iov_iter_alignment(iter)) & |
367 | (bdev_logical_block_size(bdev) - 1)) |
368 | return -EINVAL; |
369 | |
370 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool); |
371 | |
372 | dio = container_of(bio, struct blkdev_dio, bio); |
373 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
374 | if (dio->is_sync) { |
375 | dio->waiter = current; |
376 | bio_get(bio); |
377 | } else { |
378 | dio->iocb = iocb; |
379 | } |
380 | |
381 | dio->size = 0; |
382 | dio->multi_bio = false; |
383 | dio->should_dirty = is_read && iter_is_iovec(iter); |
384 | |
385 | /* |
386 | * Don't plug for HIPRI/polled IO, as those should go straight |
387 | * to issue |
388 | */ |
389 | if (!is_poll) |
390 | blk_start_plug(&plug); |
391 | |
392 | for (;;) { |
393 | bio_set_dev(bio, bdev); |
394 | bio->bi_iter.bi_sector = pos >> 9; |
395 | bio->bi_write_hint = iocb->ki_hint; |
396 | bio->bi_private = dio; |
397 | bio->bi_end_io = blkdev_bio_end_io; |
398 | bio->bi_ioprio = iocb->ki_ioprio; |
399 | |
400 | ret = bio_iov_iter_get_pages(bio, iter); |
401 | if (unlikely(ret)) { |
402 | bio->bi_status = BLK_STS_IOERR; |
403 | bio_endio(bio); |
404 | break; |
405 | } |
406 | |
407 | if (is_read) { |
408 | bio->bi_opf = REQ_OP_READ; |
409 | if (dio->should_dirty) |
410 | bio_set_pages_dirty(bio); |
411 | } else { |
412 | bio->bi_opf = dio_bio_write_op(iocb); |
413 | task_io_account_write(bio->bi_iter.bi_size); |
414 | } |
415 | |
416 | dio->size += bio->bi_iter.bi_size; |
417 | pos += bio->bi_iter.bi_size; |
418 | |
419 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); |
420 | if (!nr_pages) { |
421 | bool polled = false; |
422 | |
423 | if (iocb->ki_flags & IOCB_HIPRI) { |
424 | bio_set_polled(bio, iocb); |
425 | polled = true; |
426 | } |
427 | |
428 | qc = submit_bio(bio); |
429 | |
430 | if (polled) |
431 | WRITE_ONCE(iocb->ki_cookie, qc); |
432 | break; |
433 | } |
434 | |
435 | if (!dio->multi_bio) { |
436 | /* |
437 | * AIO needs an extra reference to ensure the dio |
438 | * structure which is embedded into the first bio |
439 | * stays around. |
440 | */ |
441 | if (!is_sync) |
442 | bio_get(bio); |
443 | dio->multi_bio = true; |
444 | atomic_set(&dio->ref, 2); |
445 | } else { |
446 | atomic_inc(&dio->ref); |
447 | } |
448 | |
449 | submit_bio(bio); |
450 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
451 | } |
452 | |
453 | if (!is_poll) |
454 | blk_finish_plug(&plug); |
455 | |
456 | if (!is_sync) |
457 | return -EIOCBQUEUED; |
458 | |
459 | for (;;) { |
460 | set_current_state(TASK_UNINTERRUPTIBLE); |
461 | if (!READ_ONCE(dio->waiter)) |
462 | break; |
463 | |
464 | if (!(iocb->ki_flags & IOCB_HIPRI) || |
465 | !blk_poll(bdev_get_queue(bdev), qc, true)) |
466 | io_schedule(); |
467 | } |
468 | __set_current_state(TASK_RUNNING); |
469 | |
470 | if (!ret) |
471 | ret = blk_status_to_errno(dio->bio.bi_status); |
472 | if (likely(!ret)) |
473 | ret = dio->size; |
474 | |
475 | bio_put(&dio->bio); |
476 | return ret; |
477 | } |
478 | |
479 | static ssize_t |
480 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
481 | { |
482 | int nr_pages; |
483 | |
484 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
485 | if (!nr_pages) |
486 | return 0; |
487 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
488 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
489 | |
490 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); |
491 | } |
492 | |
493 | static __init int blkdev_init(void) |
494 | { |
495 | return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS); |
496 | } |
497 | module_init(blkdev_init); |
498 | |
499 | int __sync_blockdev(struct block_device *bdev, int wait) |
500 | { |
501 | if (!bdev) |
502 | return 0; |
503 | if (!wait) |
504 | return filemap_flush(bdev->bd_inode->i_mapping); |
505 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); |
506 | } |
507 | |
508 | /* |
509 | * Write out and wait upon all the dirty data associated with a block |
510 | * device via its mapping. Does not take the superblock lock. |
511 | */ |
512 | int sync_blockdev(struct block_device *bdev) |
513 | { |
514 | return __sync_blockdev(bdev, 1); |
515 | } |
516 | EXPORT_SYMBOL(sync_blockdev); |
517 | |
518 | /* |
519 | * Write out and wait upon all dirty data associated with this |
520 | * device. Filesystem data as well as the underlying block |
521 | * device. Takes the superblock lock. |
522 | */ |
523 | int fsync_bdev(struct block_device *bdev) |
524 | { |
525 | struct super_block *sb = get_super(bdev); |
526 | if (sb) { |
527 | int res = sync_filesystem(sb); |
528 | drop_super(sb); |
529 | return res; |
530 | } |
531 | return sync_blockdev(bdev); |
532 | } |
533 | EXPORT_SYMBOL(fsync_bdev); |
534 | |
535 | /** |
536 | * freeze_bdev -- lock a filesystem and force it into a consistent state |
537 | * @bdev: blockdevice to lock |
538 | * |
539 | * If a superblock is found on this device, we take the s_umount semaphore |
540 | * on it to make sure nobody unmounts until the snapshot creation is done. |
541 | * The reference counter (bd_fsfreeze_count) guarantees that only the last |
542 | * unfreeze process can unfreeze the frozen filesystem actually when multiple |
543 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and |
544 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze |
545 | * actually. |
546 | */ |
547 | struct super_block *freeze_bdev(struct block_device *bdev) |
548 | { |
549 | struct super_block *sb; |
550 | int error = 0; |
551 | |
552 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
553 | if (++bdev->bd_fsfreeze_count > 1) { |
554 | /* |
555 | * We don't even need to grab a reference - the first call |
556 | * to freeze_bdev grab an active reference and only the last |
557 | * thaw_bdev drops it. |
558 | */ |
559 | sb = get_super(bdev); |
560 | if (sb) |
561 | drop_super(sb); |
562 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
563 | return sb; |
564 | } |
565 | |
566 | sb = get_active_super(bdev); |
567 | if (!sb) |
568 | goto out; |
569 | if (sb->s_op->freeze_super) |
570 | error = sb->s_op->freeze_super(sb); |
571 | else |
572 | error = freeze_super(sb); |
573 | if (error) { |
574 | deactivate_super(sb); |
575 | bdev->bd_fsfreeze_count--; |
576 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
577 | return ERR_PTR(error); |
578 | } |
579 | deactivate_super(sb); |
580 | out: |
581 | sync_blockdev(bdev); |
582 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
583 | return sb; /* thaw_bdev releases s->s_umount */ |
584 | } |
585 | EXPORT_SYMBOL(freeze_bdev); |
586 | |
587 | /** |
588 | * thaw_bdev -- unlock filesystem |
589 | * @bdev: blockdevice to unlock |
590 | * @sb: associated superblock |
591 | * |
592 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). |
593 | */ |
594 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) |
595 | { |
596 | int error = -EINVAL; |
597 | |
598 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
599 | if (!bdev->bd_fsfreeze_count) |
600 | goto out; |
601 | |
602 | error = 0; |
603 | if (--bdev->bd_fsfreeze_count > 0) |
604 | goto out; |
605 | |
606 | if (!sb) |
607 | goto out; |
608 | |
609 | if (sb->s_op->thaw_super) |
610 | error = sb->s_op->thaw_super(sb); |
611 | else |
612 | error = thaw_super(sb); |
613 | if (error) |
614 | bdev->bd_fsfreeze_count++; |
615 | out: |
616 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
617 | return error; |
618 | } |
619 | EXPORT_SYMBOL(thaw_bdev); |
620 | |
621 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
622 | { |
623 | return block_write_full_page(page, blkdev_get_block, wbc); |
624 | } |
625 | |
626 | static int blkdev_readpage(struct file * file, struct page * page) |
627 | { |
628 | return block_read_full_page(page, blkdev_get_block); |
629 | } |
630 | |
631 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
632 | struct list_head *pages, unsigned nr_pages) |
633 | { |
634 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); |
635 | } |
636 | |
637 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
638 | loff_t pos, unsigned len, unsigned flags, |
639 | struct page **pagep, void **fsdata) |
640 | { |
641 | return block_write_begin(mapping, pos, len, flags, pagep, |
642 | blkdev_get_block); |
643 | } |
644 | |
645 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
646 | loff_t pos, unsigned len, unsigned copied, |
647 | struct page *page, void *fsdata) |
648 | { |
649 | int ret; |
650 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
651 | |
652 | unlock_page(page); |
653 | put_page(page); |
654 | |
655 | return ret; |
656 | } |
657 | |
658 | /* |
659 | * private llseek: |
660 | * for a block special file file_inode(file)->i_size is zero |
661 | * so we compute the size by hand (just as in block_read/write above) |
662 | */ |
663 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
664 | { |
665 | struct inode *bd_inode = bdev_file_inode(file); |
666 | loff_t retval; |
667 | |
668 | inode_lock(bd_inode); |
669 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
670 | inode_unlock(bd_inode); |
671 | return retval; |
672 | } |
673 | |
674 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
675 | { |
676 | struct inode *bd_inode = bdev_file_inode(filp); |
677 | struct block_device *bdev = I_BDEV(bd_inode); |
678 | int error; |
679 | |
680 | error = file_write_and_wait_range(filp, start, end); |
681 | if (error) |
682 | return error; |
683 | |
684 | /* |
685 | * There is no need to serialise calls to blkdev_issue_flush with |
686 | * i_mutex and doing so causes performance issues with concurrent |
687 | * O_SYNC writers to a block device. |
688 | */ |
689 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
690 | if (error == -EOPNOTSUPP) |
691 | error = 0; |
692 | |
693 | return error; |
694 | } |
695 | EXPORT_SYMBOL(blkdev_fsync); |
696 | |
697 | /** |
698 | * bdev_read_page() - Start reading a page from a block device |
699 | * @bdev: The device to read the page from |
700 | * @sector: The offset on the device to read the page to (need not be aligned) |
701 | * @page: The page to read |
702 | * |
703 | * On entry, the page should be locked. It will be unlocked when the page |
704 | * has been read. If the block driver implements rw_page synchronously, |
705 | * that will be true on exit from this function, but it need not be. |
706 | * |
707 | * Errors returned by this function are usually "soft", eg out of memory, or |
708 | * queue full; callers should try a different route to read this page rather |
709 | * than propagate an error back up the stack. |
710 | * |
711 | * Return: negative errno if an error occurs, 0 if submission was successful. |
712 | */ |
713 | int bdev_read_page(struct block_device *bdev, sector_t sector, |
714 | struct page *page) |
715 | { |
716 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
717 | int result = -EOPNOTSUPP; |
718 | |
719 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
720 | return result; |
721 | |
722 | result = blk_queue_enter(bdev->bd_queue, 0); |
723 | if (result) |
724 | return result; |
725 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
726 | REQ_OP_READ); |
727 | blk_queue_exit(bdev->bd_queue); |
728 | return result; |
729 | } |
730 | EXPORT_SYMBOL_GPL(bdev_read_page); |
731 | |
732 | /** |
733 | * bdev_write_page() - Start writing a page to a block device |
734 | * @bdev: The device to write the page to |
735 | * @sector: The offset on the device to write the page to (need not be aligned) |
736 | * @page: The page to write |
737 | * @wbc: The writeback_control for the write |
738 | * |
739 | * On entry, the page should be locked and not currently under writeback. |
740 | * On exit, if the write started successfully, the page will be unlocked and |
741 | * under writeback. If the write failed already (eg the driver failed to |
742 | * queue the page to the device), the page will still be locked. If the |
743 | * caller is a ->writepage implementation, it will need to unlock the page. |
744 | * |
745 | * Errors returned by this function are usually "soft", eg out of memory, or |
746 | * queue full; callers should try a different route to write this page rather |
747 | * than propagate an error back up the stack. |
748 | * |
749 | * Return: negative errno if an error occurs, 0 if submission was successful. |
750 | */ |
751 | int bdev_write_page(struct block_device *bdev, sector_t sector, |
752 | struct page *page, struct writeback_control *wbc) |
753 | { |
754 | int result; |
755 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
756 | |
757 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
758 | return -EOPNOTSUPP; |
759 | result = blk_queue_enter(bdev->bd_queue, 0); |
760 | if (result) |
761 | return result; |
762 | |
763 | set_page_writeback(page); |
764 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, |
765 | REQ_OP_WRITE); |
766 | if (result) { |
767 | end_page_writeback(page); |
768 | } else { |
769 | clean_page_buffers(page); |
770 | unlock_page(page); |
771 | } |
772 | blk_queue_exit(bdev->bd_queue); |
773 | return result; |
774 | } |
775 | EXPORT_SYMBOL_GPL(bdev_write_page); |
776 | |
777 | /* |
778 | * pseudo-fs |
779 | */ |
780 | |
781 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); |
782 | static struct kmem_cache * bdev_cachep __read_mostly; |
783 | |
784 | static struct inode *bdev_alloc_inode(struct super_block *sb) |
785 | { |
786 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
787 | if (!ei) |
788 | return NULL; |
789 | return &ei->vfs_inode; |
790 | } |
791 | |
792 | static void bdev_i_callback(struct rcu_head *head) |
793 | { |
794 | struct inode *inode = container_of(head, struct inode, i_rcu); |
795 | struct bdev_inode *bdi = BDEV_I(inode); |
796 | |
797 | kmem_cache_free(bdev_cachep, bdi); |
798 | } |
799 | |
800 | static void bdev_destroy_inode(struct inode *inode) |
801 | { |
802 | call_rcu(&inode->i_rcu, bdev_i_callback); |
803 | } |
804 | |
805 | static void init_once(void *foo) |
806 | { |
807 | struct bdev_inode *ei = (struct bdev_inode *) foo; |
808 | struct block_device *bdev = &ei->bdev; |
809 | |
810 | memset(bdev, 0, sizeof(*bdev)); |
811 | mutex_init(&bdev->bd_mutex); |
812 | INIT_LIST_HEAD(&bdev->bd_list); |
813 | #ifdef CONFIG_SYSFS |
814 | INIT_LIST_HEAD(&bdev->bd_holder_disks); |
815 | #endif |
816 | bdev->bd_bdi = &noop_backing_dev_info; |
817 | inode_init_once(&ei->vfs_inode); |
818 | /* Initialize mutex for freeze. */ |
819 | mutex_init(&bdev->bd_fsfreeze_mutex); |
820 | } |
821 | |
822 | static void bdev_evict_inode(struct inode *inode) |
823 | { |
824 | struct block_device *bdev = &BDEV_I(inode)->bdev; |
825 | truncate_inode_pages_final(&inode->i_data); |
826 | invalidate_inode_buffers(inode); /* is it needed here? */ |
827 | clear_inode(inode); |
828 | spin_lock(&bdev_lock); |
829 | list_del_init(&bdev->bd_list); |
830 | spin_unlock(&bdev_lock); |
831 | /* Detach inode from wb early as bdi_put() may free bdi->wb */ |
832 | inode_detach_wb(inode); |
833 | if (bdev->bd_bdi != &noop_backing_dev_info) { |
834 | bdi_put(bdev->bd_bdi); |
835 | bdev->bd_bdi = &noop_backing_dev_info; |
836 | } |
837 | } |
838 | |
839 | static const struct super_operations bdev_sops = { |
840 | .statfs = simple_statfs, |
841 | .alloc_inode = bdev_alloc_inode, |
842 | .destroy_inode = bdev_destroy_inode, |
843 | .drop_inode = generic_delete_inode, |
844 | .evict_inode = bdev_evict_inode, |
845 | }; |
846 | |
847 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
848 | int flags, const char *dev_name, void *data) |
849 | { |
850 | struct dentry *dent; |
851 | dent = mount_pseudo(fs_type, "bdev:" , &bdev_sops, NULL, BDEVFS_MAGIC); |
852 | if (!IS_ERR(dent)) |
853 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; |
854 | return dent; |
855 | } |
856 | |
857 | static struct file_system_type bd_type = { |
858 | .name = "bdev" , |
859 | .mount = bd_mount, |
860 | .kill_sb = kill_anon_super, |
861 | }; |
862 | |
863 | struct super_block *blockdev_superblock __read_mostly; |
864 | EXPORT_SYMBOL_GPL(blockdev_superblock); |
865 | |
866 | void __init bdev_cache_init(void) |
867 | { |
868 | int err; |
869 | static struct vfsmount *bd_mnt; |
870 | |
871 | bdev_cachep = kmem_cache_create("bdev_cache" , sizeof(struct bdev_inode), |
872 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
873 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
874 | init_once); |
875 | err = register_filesystem(&bd_type); |
876 | if (err) |
877 | panic("Cannot register bdev pseudo-fs" ); |
878 | bd_mnt = kern_mount(&bd_type); |
879 | if (IS_ERR(bd_mnt)) |
880 | panic("Cannot create bdev pseudo-fs" ); |
881 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
882 | } |
883 | |
884 | /* |
885 | * Most likely _very_ bad one - but then it's hardly critical for small |
886 | * /dev and can be fixed when somebody will need really large one. |
887 | * Keep in mind that it will be fed through icache hash function too. |
888 | */ |
889 | static inline unsigned long hash(dev_t dev) |
890 | { |
891 | return MAJOR(dev)+MINOR(dev); |
892 | } |
893 | |
894 | static int bdev_test(struct inode *inode, void *data) |
895 | { |
896 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; |
897 | } |
898 | |
899 | static int bdev_set(struct inode *inode, void *data) |
900 | { |
901 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; |
902 | return 0; |
903 | } |
904 | |
905 | static LIST_HEAD(all_bdevs); |
906 | |
907 | /* |
908 | * If there is a bdev inode for this device, unhash it so that it gets evicted |
909 | * as soon as last inode reference is dropped. |
910 | */ |
911 | void bdev_unhash_inode(dev_t dev) |
912 | { |
913 | struct inode *inode; |
914 | |
915 | inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev); |
916 | if (inode) { |
917 | remove_inode_hash(inode); |
918 | iput(inode); |
919 | } |
920 | } |
921 | |
922 | struct block_device *bdget(dev_t dev) |
923 | { |
924 | struct block_device *bdev; |
925 | struct inode *inode; |
926 | |
927 | inode = iget5_locked(blockdev_superblock, hash(dev), |
928 | bdev_test, bdev_set, &dev); |
929 | |
930 | if (!inode) |
931 | return NULL; |
932 | |
933 | bdev = &BDEV_I(inode)->bdev; |
934 | |
935 | if (inode->i_state & I_NEW) { |
936 | bdev->bd_contains = NULL; |
937 | bdev->bd_super = NULL; |
938 | bdev->bd_inode = inode; |
939 | bdev->bd_block_size = i_blocksize(inode); |
940 | bdev->bd_part_count = 0; |
941 | bdev->bd_invalidated = 0; |
942 | inode->i_mode = S_IFBLK; |
943 | inode->i_rdev = dev; |
944 | inode->i_bdev = bdev; |
945 | inode->i_data.a_ops = &def_blk_aops; |
946 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); |
947 | spin_lock(&bdev_lock); |
948 | list_add(&bdev->bd_list, &all_bdevs); |
949 | spin_unlock(&bdev_lock); |
950 | unlock_new_inode(inode); |
951 | } |
952 | return bdev; |
953 | } |
954 | |
955 | EXPORT_SYMBOL(bdget); |
956 | |
957 | /** |
958 | * bdgrab -- Grab a reference to an already referenced block device |
959 | * @bdev: Block device to grab a reference to. |
960 | */ |
961 | struct block_device *bdgrab(struct block_device *bdev) |
962 | { |
963 | ihold(bdev->bd_inode); |
964 | return bdev; |
965 | } |
966 | EXPORT_SYMBOL(bdgrab); |
967 | |
968 | long nr_blockdev_pages(void) |
969 | { |
970 | struct block_device *bdev; |
971 | long ret = 0; |
972 | spin_lock(&bdev_lock); |
973 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
974 | ret += bdev->bd_inode->i_mapping->nrpages; |
975 | } |
976 | spin_unlock(&bdev_lock); |
977 | return ret; |
978 | } |
979 | |
980 | void bdput(struct block_device *bdev) |
981 | { |
982 | iput(bdev->bd_inode); |
983 | } |
984 | |
985 | EXPORT_SYMBOL(bdput); |
986 | |
987 | static struct block_device *bd_acquire(struct inode *inode) |
988 | { |
989 | struct block_device *bdev; |
990 | |
991 | spin_lock(&bdev_lock); |
992 | bdev = inode->i_bdev; |
993 | if (bdev && !inode_unhashed(bdev->bd_inode)) { |
994 | bdgrab(bdev); |
995 | spin_unlock(&bdev_lock); |
996 | return bdev; |
997 | } |
998 | spin_unlock(&bdev_lock); |
999 | |
1000 | /* |
1001 | * i_bdev references block device inode that was already shut down |
1002 | * (corresponding device got removed). Remove the reference and look |
1003 | * up block device inode again just in case new device got |
1004 | * reestablished under the same device number. |
1005 | */ |
1006 | if (bdev) |
1007 | bd_forget(inode); |
1008 | |
1009 | bdev = bdget(inode->i_rdev); |
1010 | if (bdev) { |
1011 | spin_lock(&bdev_lock); |
1012 | if (!inode->i_bdev) { |
1013 | /* |
1014 | * We take an additional reference to bd_inode, |
1015 | * and it's released in clear_inode() of inode. |
1016 | * So, we can access it via ->i_mapping always |
1017 | * without igrab(). |
1018 | */ |
1019 | bdgrab(bdev); |
1020 | inode->i_bdev = bdev; |
1021 | inode->i_mapping = bdev->bd_inode->i_mapping; |
1022 | } |
1023 | spin_unlock(&bdev_lock); |
1024 | } |
1025 | return bdev; |
1026 | } |
1027 | |
1028 | /* Call when you free inode */ |
1029 | |
1030 | void bd_forget(struct inode *inode) |
1031 | { |
1032 | struct block_device *bdev = NULL; |
1033 | |
1034 | spin_lock(&bdev_lock); |
1035 | if (!sb_is_blkdev_sb(inode->i_sb)) |
1036 | bdev = inode->i_bdev; |
1037 | inode->i_bdev = NULL; |
1038 | inode->i_mapping = &inode->i_data; |
1039 | spin_unlock(&bdev_lock); |
1040 | |
1041 | if (bdev) |
1042 | bdput(bdev); |
1043 | } |
1044 | |
1045 | /** |
1046 | * bd_may_claim - test whether a block device can be claimed |
1047 | * @bdev: block device of interest |
1048 | * @whole: whole block device containing @bdev, may equal @bdev |
1049 | * @holder: holder trying to claim @bdev |
1050 | * |
1051 | * Test whether @bdev can be claimed by @holder. |
1052 | * |
1053 | * CONTEXT: |
1054 | * spin_lock(&bdev_lock). |
1055 | * |
1056 | * RETURNS: |
1057 | * %true if @bdev can be claimed, %false otherwise. |
1058 | */ |
1059 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, |
1060 | void *holder) |
1061 | { |
1062 | if (bdev->bd_holder == holder) |
1063 | return true; /* already a holder */ |
1064 | else if (bdev->bd_holder != NULL) |
1065 | return false; /* held by someone else */ |
1066 | else if (whole == bdev) |
1067 | return true; /* is a whole device which isn't held */ |
1068 | |
1069 | else if (whole->bd_holder == bd_may_claim) |
1070 | return true; /* is a partition of a device that is being partitioned */ |
1071 | else if (whole->bd_holder != NULL) |
1072 | return false; /* is a partition of a held device */ |
1073 | else |
1074 | return true; /* is a partition of an un-held device */ |
1075 | } |
1076 | |
1077 | /** |
1078 | * bd_prepare_to_claim - prepare to claim a block device |
1079 | * @bdev: block device of interest |
1080 | * @whole: the whole device containing @bdev, may equal @bdev |
1081 | * @holder: holder trying to claim @bdev |
1082 | * |
1083 | * Prepare to claim @bdev. This function fails if @bdev is already |
1084 | * claimed by another holder and waits if another claiming is in |
1085 | * progress. This function doesn't actually claim. On successful |
1086 | * return, the caller has ownership of bd_claiming and bd_holder[s]. |
1087 | * |
1088 | * CONTEXT: |
1089 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab |
1090 | * it multiple times. |
1091 | * |
1092 | * RETURNS: |
1093 | * 0 if @bdev can be claimed, -EBUSY otherwise. |
1094 | */ |
1095 | static int bd_prepare_to_claim(struct block_device *bdev, |
1096 | struct block_device *whole, void *holder) |
1097 | { |
1098 | retry: |
1099 | /* if someone else claimed, fail */ |
1100 | if (!bd_may_claim(bdev, whole, holder)) |
1101 | return -EBUSY; |
1102 | |
1103 | /* if claiming is already in progress, wait for it to finish */ |
1104 | if (whole->bd_claiming) { |
1105 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
1106 | DEFINE_WAIT(wait); |
1107 | |
1108 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); |
1109 | spin_unlock(&bdev_lock); |
1110 | schedule(); |
1111 | finish_wait(wq, &wait); |
1112 | spin_lock(&bdev_lock); |
1113 | goto retry; |
1114 | } |
1115 | |
1116 | /* yay, all mine */ |
1117 | return 0; |
1118 | } |
1119 | |
1120 | static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno) |
1121 | { |
1122 | struct gendisk *disk = get_gendisk(bdev->bd_dev, partno); |
1123 | |
1124 | if (!disk) |
1125 | return NULL; |
1126 | /* |
1127 | * Now that we hold gendisk reference we make sure bdev we looked up is |
1128 | * not stale. If it is, it means device got removed and created before |
1129 | * we looked up gendisk and we fail open in such case. Associating |
1130 | * unhashed bdev with newly created gendisk could lead to two bdevs |
1131 | * (and thus two independent caches) being associated with one device |
1132 | * which is bad. |
1133 | */ |
1134 | if (inode_unhashed(bdev->bd_inode)) { |
1135 | put_disk_and_module(disk); |
1136 | return NULL; |
1137 | } |
1138 | return disk; |
1139 | } |
1140 | |
1141 | /** |
1142 | * bd_start_claiming - start claiming a block device |
1143 | * @bdev: block device of interest |
1144 | * @holder: holder trying to claim @bdev |
1145 | * |
1146 | * @bdev is about to be opened exclusively. Check @bdev can be opened |
1147 | * exclusively and mark that an exclusive open is in progress. Each |
1148 | * successful call to this function must be matched with a call to |
1149 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
1150 | * fail). |
1151 | * |
1152 | * This function is used to gain exclusive access to the block device |
1153 | * without actually causing other exclusive open attempts to fail. It |
1154 | * should be used when the open sequence itself requires exclusive |
1155 | * access but may subsequently fail. |
1156 | * |
1157 | * CONTEXT: |
1158 | * Might sleep. |
1159 | * |
1160 | * RETURNS: |
1161 | * Pointer to the block device containing @bdev on success, ERR_PTR() |
1162 | * value on failure. |
1163 | */ |
1164 | static struct block_device *bd_start_claiming(struct block_device *bdev, |
1165 | void *holder) |
1166 | { |
1167 | struct gendisk *disk; |
1168 | struct block_device *whole; |
1169 | int partno, err; |
1170 | |
1171 | might_sleep(); |
1172 | |
1173 | /* |
1174 | * @bdev might not have been initialized properly yet, look up |
1175 | * and grab the outer block device the hard way. |
1176 | */ |
1177 | disk = bdev_get_gendisk(bdev, &partno); |
1178 | if (!disk) |
1179 | return ERR_PTR(-ENXIO); |
1180 | |
1181 | /* |
1182 | * Normally, @bdev should equal what's returned from bdget_disk() |
1183 | * if partno is 0; however, some drivers (floppy) use multiple |
1184 | * bdev's for the same physical device and @bdev may be one of the |
1185 | * aliases. Keep @bdev if partno is 0. This means claimer |
1186 | * tracking is broken for those devices but it has always been that |
1187 | * way. |
1188 | */ |
1189 | if (partno) |
1190 | whole = bdget_disk(disk, 0); |
1191 | else |
1192 | whole = bdgrab(bdev); |
1193 | |
1194 | put_disk_and_module(disk); |
1195 | if (!whole) |
1196 | return ERR_PTR(-ENOMEM); |
1197 | |
1198 | /* prepare to claim, if successful, mark claiming in progress */ |
1199 | spin_lock(&bdev_lock); |
1200 | |
1201 | err = bd_prepare_to_claim(bdev, whole, holder); |
1202 | if (err == 0) { |
1203 | whole->bd_claiming = holder; |
1204 | spin_unlock(&bdev_lock); |
1205 | return whole; |
1206 | } else { |
1207 | spin_unlock(&bdev_lock); |
1208 | bdput(whole); |
1209 | return ERR_PTR(err); |
1210 | } |
1211 | } |
1212 | |
1213 | #ifdef CONFIG_SYSFS |
1214 | struct bd_holder_disk { |
1215 | struct list_head list; |
1216 | struct gendisk *disk; |
1217 | int refcnt; |
1218 | }; |
1219 | |
1220 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, |
1221 | struct gendisk *disk) |
1222 | { |
1223 | struct bd_holder_disk *holder; |
1224 | |
1225 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) |
1226 | if (holder->disk == disk) |
1227 | return holder; |
1228 | return NULL; |
1229 | } |
1230 | |
1231 | static int add_symlink(struct kobject *from, struct kobject *to) |
1232 | { |
1233 | return sysfs_create_link(from, to, kobject_name(to)); |
1234 | } |
1235 | |
1236 | static void del_symlink(struct kobject *from, struct kobject *to) |
1237 | { |
1238 | sysfs_remove_link(from, kobject_name(to)); |
1239 | } |
1240 | |
1241 | /** |
1242 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
1243 | * @bdev: the claimed slave bdev |
1244 | * @disk: the holding disk |
1245 | * |
1246 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1247 | * |
1248 | * This functions creates the following sysfs symlinks. |
1249 | * |
1250 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
1251 | * - from "holders" directory of the @bdev to the holder @disk |
1252 | * |
1253 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
1254 | * passed to bd_link_disk_holder(), then: |
1255 | * |
1256 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
1257 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 |
1258 | * |
1259 | * The caller must have claimed @bdev before calling this function and |
1260 | * ensure that both @bdev and @disk are valid during the creation and |
1261 | * lifetime of these symlinks. |
1262 | * |
1263 | * CONTEXT: |
1264 | * Might sleep. |
1265 | * |
1266 | * RETURNS: |
1267 | * 0 on success, -errno on failure. |
1268 | */ |
1269 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
1270 | { |
1271 | struct bd_holder_disk *holder; |
1272 | int ret = 0; |
1273 | |
1274 | mutex_lock(&bdev->bd_mutex); |
1275 | |
1276 | WARN_ON_ONCE(!bdev->bd_holder); |
1277 | |
1278 | /* FIXME: remove the following once add_disk() handles errors */ |
1279 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) |
1280 | goto out_unlock; |
1281 | |
1282 | holder = bd_find_holder_disk(bdev, disk); |
1283 | if (holder) { |
1284 | holder->refcnt++; |
1285 | goto out_unlock; |
1286 | } |
1287 | |
1288 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
1289 | if (!holder) { |
1290 | ret = -ENOMEM; |
1291 | goto out_unlock; |
1292 | } |
1293 | |
1294 | INIT_LIST_HEAD(&holder->list); |
1295 | holder->disk = disk; |
1296 | holder->refcnt = 1; |
1297 | |
1298 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
1299 | if (ret) |
1300 | goto out_free; |
1301 | |
1302 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); |
1303 | if (ret) |
1304 | goto out_del; |
1305 | /* |
1306 | * bdev could be deleted beneath us which would implicitly destroy |
1307 | * the holder directory. Hold on to it. |
1308 | */ |
1309 | kobject_get(bdev->bd_part->holder_dir); |
1310 | |
1311 | list_add(&holder->list, &bdev->bd_holder_disks); |
1312 | goto out_unlock; |
1313 | |
1314 | out_del: |
1315 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
1316 | out_free: |
1317 | kfree(holder); |
1318 | out_unlock: |
1319 | mutex_unlock(&bdev->bd_mutex); |
1320 | return ret; |
1321 | } |
1322 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
1323 | |
1324 | /** |
1325 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() |
1326 | * @bdev: the calimed slave bdev |
1327 | * @disk: the holding disk |
1328 | * |
1329 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
1330 | * |
1331 | * CONTEXT: |
1332 | * Might sleep. |
1333 | */ |
1334 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) |
1335 | { |
1336 | struct bd_holder_disk *holder; |
1337 | |
1338 | mutex_lock(&bdev->bd_mutex); |
1339 | |
1340 | holder = bd_find_holder_disk(bdev, disk); |
1341 | |
1342 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { |
1343 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
1344 | del_symlink(bdev->bd_part->holder_dir, |
1345 | &disk_to_dev(disk)->kobj); |
1346 | kobject_put(bdev->bd_part->holder_dir); |
1347 | list_del_init(&holder->list); |
1348 | kfree(holder); |
1349 | } |
1350 | |
1351 | mutex_unlock(&bdev->bd_mutex); |
1352 | } |
1353 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
1354 | #endif |
1355 | |
1356 | /** |
1357 | * flush_disk - invalidates all buffer-cache entries on a disk |
1358 | * |
1359 | * @bdev: struct block device to be flushed |
1360 | * @kill_dirty: flag to guide handling of dirty inodes |
1361 | * |
1362 | * Invalidates all buffer-cache entries on a disk. It should be called |
1363 | * when a disk has been changed -- either by a media change or online |
1364 | * resize. |
1365 | */ |
1366 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
1367 | { |
1368 | if (__invalidate_device(bdev, kill_dirty)) { |
1369 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
1370 | "resized disk %s\n" , |
1371 | bdev->bd_disk ? bdev->bd_disk->disk_name : "" ); |
1372 | } |
1373 | |
1374 | if (!bdev->bd_disk) |
1375 | return; |
1376 | if (disk_part_scan_enabled(bdev->bd_disk)) |
1377 | bdev->bd_invalidated = 1; |
1378 | } |
1379 | |
1380 | /** |
1381 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
1382 | * @disk: struct gendisk to check |
1383 | * @bdev: struct bdev to adjust. |
1384 | * @verbose: if %true log a message about a size change if there is any |
1385 | * |
1386 | * This routine checks to see if the bdev size does not match the disk size |
1387 | * and adjusts it if it differs. When shrinking the bdev size, its all caches |
1388 | * are freed. |
1389 | */ |
1390 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev, |
1391 | bool verbose) |
1392 | { |
1393 | loff_t disk_size, bdev_size; |
1394 | |
1395 | disk_size = (loff_t)get_capacity(disk) << 9; |
1396 | bdev_size = i_size_read(bdev->bd_inode); |
1397 | if (disk_size != bdev_size) { |
1398 | if (verbose) { |
1399 | printk(KERN_INFO |
1400 | "%s: detected capacity change from %lld to %lld\n" , |
1401 | disk->disk_name, bdev_size, disk_size); |
1402 | } |
1403 | i_size_write(bdev->bd_inode, disk_size); |
1404 | if (bdev_size > disk_size) |
1405 | flush_disk(bdev, false); |
1406 | } |
1407 | } |
1408 | |
1409 | /** |
1410 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
1411 | * @disk: struct gendisk to be revalidated |
1412 | * |
1413 | * This routine is a wrapper for lower-level driver's revalidate_disk |
1414 | * call-backs. It is used to do common pre and post operations needed |
1415 | * for all revalidate_disk operations. |
1416 | */ |
1417 | int revalidate_disk(struct gendisk *disk) |
1418 | { |
1419 | struct block_device *bdev; |
1420 | int ret = 0; |
1421 | |
1422 | if (disk->fops->revalidate_disk) |
1423 | ret = disk->fops->revalidate_disk(disk); |
1424 | bdev = bdget_disk(disk, 0); |
1425 | if (!bdev) |
1426 | return ret; |
1427 | |
1428 | mutex_lock(&bdev->bd_mutex); |
1429 | check_disk_size_change(disk, bdev, ret == 0); |
1430 | bdev->bd_invalidated = 0; |
1431 | mutex_unlock(&bdev->bd_mutex); |
1432 | bdput(bdev); |
1433 | return ret; |
1434 | } |
1435 | EXPORT_SYMBOL(revalidate_disk); |
1436 | |
1437 | /* |
1438 | * This routine checks whether a removable media has been changed, |
1439 | * and invalidates all buffer-cache-entries in that case. This |
1440 | * is a relatively slow routine, so we have to try to minimize using |
1441 | * it. Thus it is called only upon a 'mount' or 'open'. This |
1442 | * is the best way of combining speed and utility, I think. |
1443 | * People changing diskettes in the middle of an operation deserve |
1444 | * to lose :-) |
1445 | */ |
1446 | int check_disk_change(struct block_device *bdev) |
1447 | { |
1448 | struct gendisk *disk = bdev->bd_disk; |
1449 | const struct block_device_operations *bdops = disk->fops; |
1450 | unsigned int events; |
1451 | |
1452 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
1453 | DISK_EVENT_EJECT_REQUEST); |
1454 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) |
1455 | return 0; |
1456 | |
1457 | flush_disk(bdev, true); |
1458 | if (bdops->revalidate_disk) |
1459 | bdops->revalidate_disk(bdev->bd_disk); |
1460 | return 1; |
1461 | } |
1462 | |
1463 | EXPORT_SYMBOL(check_disk_change); |
1464 | |
1465 | void bd_set_size(struct block_device *bdev, loff_t size) |
1466 | { |
1467 | inode_lock(bdev->bd_inode); |
1468 | i_size_write(bdev->bd_inode, size); |
1469 | inode_unlock(bdev->bd_inode); |
1470 | } |
1471 | EXPORT_SYMBOL(bd_set_size); |
1472 | |
1473 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
1474 | |
1475 | /* |
1476 | * bd_mutex locking: |
1477 | * |
1478 | * mutex_lock(part->bd_mutex) |
1479 | * mutex_lock_nested(whole->bd_mutex, 1) |
1480 | */ |
1481 | |
1482 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1483 | { |
1484 | struct gendisk *disk; |
1485 | int ret; |
1486 | int partno; |
1487 | int perm = 0; |
1488 | bool first_open = false; |
1489 | |
1490 | if (mode & FMODE_READ) |
1491 | perm |= MAY_READ; |
1492 | if (mode & FMODE_WRITE) |
1493 | perm |= MAY_WRITE; |
1494 | /* |
1495 | * hooks: /n/, see "layering violations". |
1496 | */ |
1497 | if (!for_part) { |
1498 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); |
1499 | if (ret != 0) { |
1500 | bdput(bdev); |
1501 | return ret; |
1502 | } |
1503 | } |
1504 | |
1505 | restart: |
1506 | |
1507 | ret = -ENXIO; |
1508 | disk = bdev_get_gendisk(bdev, &partno); |
1509 | if (!disk) |
1510 | goto out; |
1511 | |
1512 | disk_block_events(disk); |
1513 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1514 | if (!bdev->bd_openers) { |
1515 | first_open = true; |
1516 | bdev->bd_disk = disk; |
1517 | bdev->bd_queue = disk->queue; |
1518 | bdev->bd_contains = bdev; |
1519 | bdev->bd_partno = partno; |
1520 | |
1521 | if (!partno) { |
1522 | ret = -ENXIO; |
1523 | bdev->bd_part = disk_get_part(disk, partno); |
1524 | if (!bdev->bd_part) |
1525 | goto out_clear; |
1526 | |
1527 | ret = 0; |
1528 | if (disk->fops->open) { |
1529 | ret = disk->fops->open(bdev, mode); |
1530 | if (ret == -ERESTARTSYS) { |
1531 | /* Lost a race with 'disk' being |
1532 | * deleted, try again. |
1533 | * See md.c |
1534 | */ |
1535 | disk_put_part(bdev->bd_part); |
1536 | bdev->bd_part = NULL; |
1537 | bdev->bd_disk = NULL; |
1538 | bdev->bd_queue = NULL; |
1539 | mutex_unlock(&bdev->bd_mutex); |
1540 | disk_unblock_events(disk); |
1541 | put_disk_and_module(disk); |
1542 | goto restart; |
1543 | } |
1544 | } |
1545 | |
1546 | if (!ret) { |
1547 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
1548 | set_init_blocksize(bdev); |
1549 | } |
1550 | |
1551 | /* |
1552 | * If the device is invalidated, rescan partition |
1553 | * if open succeeded or failed with -ENOMEDIUM. |
1554 | * The latter is necessary to prevent ghost |
1555 | * partitions on a removed medium. |
1556 | */ |
1557 | if (bdev->bd_invalidated) { |
1558 | if (!ret) |
1559 | rescan_partitions(disk, bdev); |
1560 | else if (ret == -ENOMEDIUM) |
1561 | invalidate_partitions(disk, bdev); |
1562 | } |
1563 | |
1564 | if (ret) |
1565 | goto out_clear; |
1566 | } else { |
1567 | struct block_device *whole; |
1568 | whole = bdget_disk(disk, 0); |
1569 | ret = -ENOMEM; |
1570 | if (!whole) |
1571 | goto out_clear; |
1572 | BUG_ON(for_part); |
1573 | ret = __blkdev_get(whole, mode, 1); |
1574 | if (ret) |
1575 | goto out_clear; |
1576 | bdev->bd_contains = whole; |
1577 | bdev->bd_part = disk_get_part(disk, partno); |
1578 | if (!(disk->flags & GENHD_FL_UP) || |
1579 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1580 | ret = -ENXIO; |
1581 | goto out_clear; |
1582 | } |
1583 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1584 | set_init_blocksize(bdev); |
1585 | } |
1586 | |
1587 | if (bdev->bd_bdi == &noop_backing_dev_info) |
1588 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info); |
1589 | } else { |
1590 | if (bdev->bd_contains == bdev) { |
1591 | ret = 0; |
1592 | if (bdev->bd_disk->fops->open) |
1593 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1594 | /* the same as first opener case, read comment there */ |
1595 | if (bdev->bd_invalidated) { |
1596 | if (!ret) |
1597 | rescan_partitions(bdev->bd_disk, bdev); |
1598 | else if (ret == -ENOMEDIUM) |
1599 | invalidate_partitions(bdev->bd_disk, bdev); |
1600 | } |
1601 | if (ret) |
1602 | goto out_unlock_bdev; |
1603 | } |
1604 | } |
1605 | bdev->bd_openers++; |
1606 | if (for_part) |
1607 | bdev->bd_part_count++; |
1608 | mutex_unlock(&bdev->bd_mutex); |
1609 | disk_unblock_events(disk); |
1610 | /* only one opener holds refs to the module and disk */ |
1611 | if (!first_open) |
1612 | put_disk_and_module(disk); |
1613 | return 0; |
1614 | |
1615 | out_clear: |
1616 | disk_put_part(bdev->bd_part); |
1617 | bdev->bd_disk = NULL; |
1618 | bdev->bd_part = NULL; |
1619 | bdev->bd_queue = NULL; |
1620 | if (bdev != bdev->bd_contains) |
1621 | __blkdev_put(bdev->bd_contains, mode, 1); |
1622 | bdev->bd_contains = NULL; |
1623 | out_unlock_bdev: |
1624 | mutex_unlock(&bdev->bd_mutex); |
1625 | disk_unblock_events(disk); |
1626 | put_disk_and_module(disk); |
1627 | out: |
1628 | bdput(bdev); |
1629 | |
1630 | return ret; |
1631 | } |
1632 | |
1633 | /** |
1634 | * blkdev_get - open a block device |
1635 | * @bdev: block_device to open |
1636 | * @mode: FMODE_* mask |
1637 | * @holder: exclusive holder identifier |
1638 | * |
1639 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is |
1640 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL |
1641 | * @holder is invalid. Exclusive opens may nest for the same @holder. |
1642 | * |
1643 | * On success, the reference count of @bdev is unchanged. On failure, |
1644 | * @bdev is put. |
1645 | * |
1646 | * CONTEXT: |
1647 | * Might sleep. |
1648 | * |
1649 | * RETURNS: |
1650 | * 0 on success, -errno on failure. |
1651 | */ |
1652 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1653 | { |
1654 | struct block_device *whole = NULL; |
1655 | int res; |
1656 | |
1657 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); |
1658 | |
1659 | if ((mode & FMODE_EXCL) && holder) { |
1660 | whole = bd_start_claiming(bdev, holder); |
1661 | if (IS_ERR(whole)) { |
1662 | bdput(bdev); |
1663 | return PTR_ERR(whole); |
1664 | } |
1665 | } |
1666 | |
1667 | res = __blkdev_get(bdev, mode, 0); |
1668 | |
1669 | if (whole) { |
1670 | struct gendisk *disk = whole->bd_disk; |
1671 | |
1672 | /* finish claiming */ |
1673 | mutex_lock(&bdev->bd_mutex); |
1674 | spin_lock(&bdev_lock); |
1675 | |
1676 | if (!res) { |
1677 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1678 | /* |
1679 | * Note that for a whole device bd_holders |
1680 | * will be incremented twice, and bd_holder |
1681 | * will be set to bd_may_claim before being |
1682 | * set to holder |
1683 | */ |
1684 | whole->bd_holders++; |
1685 | whole->bd_holder = bd_may_claim; |
1686 | bdev->bd_holders++; |
1687 | bdev->bd_holder = holder; |
1688 | } |
1689 | |
1690 | /* tell others that we're done */ |
1691 | BUG_ON(whole->bd_claiming != holder); |
1692 | whole->bd_claiming = NULL; |
1693 | wake_up_bit(&whole->bd_claiming, 0); |
1694 | |
1695 | spin_unlock(&bdev_lock); |
1696 | |
1697 | /* |
1698 | * Block event polling for write claims if requested. Any |
1699 | * write holder makes the write_holder state stick until |
1700 | * all are released. This is good enough and tracking |
1701 | * individual writeable reference is too fragile given the |
1702 | * way @mode is used in blkdev_get/put(). |
1703 | */ |
1704 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1705 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { |
1706 | bdev->bd_write_holder = true; |
1707 | disk_block_events(disk); |
1708 | } |
1709 | |
1710 | mutex_unlock(&bdev->bd_mutex); |
1711 | bdput(whole); |
1712 | } |
1713 | |
1714 | return res; |
1715 | } |
1716 | EXPORT_SYMBOL(blkdev_get); |
1717 | |
1718 | /** |
1719 | * blkdev_get_by_path - open a block device by name |
1720 | * @path: path to the block device to open |
1721 | * @mode: FMODE_* mask |
1722 | * @holder: exclusive holder identifier |
1723 | * |
1724 | * Open the blockdevice described by the device file at @path. @mode |
1725 | * and @holder are identical to blkdev_get(). |
1726 | * |
1727 | * On success, the returned block_device has reference count of one. |
1728 | * |
1729 | * CONTEXT: |
1730 | * Might sleep. |
1731 | * |
1732 | * RETURNS: |
1733 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
1734 | */ |
1735 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, |
1736 | void *holder) |
1737 | { |
1738 | struct block_device *bdev; |
1739 | int err; |
1740 | |
1741 | bdev = lookup_bdev(path); |
1742 | if (IS_ERR(bdev)) |
1743 | return bdev; |
1744 | |
1745 | err = blkdev_get(bdev, mode, holder); |
1746 | if (err) |
1747 | return ERR_PTR(err); |
1748 | |
1749 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1750 | blkdev_put(bdev, mode); |
1751 | return ERR_PTR(-EACCES); |
1752 | } |
1753 | |
1754 | return bdev; |
1755 | } |
1756 | EXPORT_SYMBOL(blkdev_get_by_path); |
1757 | |
1758 | /** |
1759 | * blkdev_get_by_dev - open a block device by device number |
1760 | * @dev: device number of block device to open |
1761 | * @mode: FMODE_* mask |
1762 | * @holder: exclusive holder identifier |
1763 | * |
1764 | * Open the blockdevice described by device number @dev. @mode and |
1765 | * @holder are identical to blkdev_get(). |
1766 | * |
1767 | * Use it ONLY if you really do not have anything better - i.e. when |
1768 | * you are behind a truly sucky interface and all you are given is a |
1769 | * device number. _Never_ to be used for internal purposes. If you |
1770 | * ever need it - reconsider your API. |
1771 | * |
1772 | * On success, the returned block_device has reference count of one. |
1773 | * |
1774 | * CONTEXT: |
1775 | * Might sleep. |
1776 | * |
1777 | * RETURNS: |
1778 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
1779 | */ |
1780 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) |
1781 | { |
1782 | struct block_device *bdev; |
1783 | int err; |
1784 | |
1785 | bdev = bdget(dev); |
1786 | if (!bdev) |
1787 | return ERR_PTR(-ENOMEM); |
1788 | |
1789 | err = blkdev_get(bdev, mode, holder); |
1790 | if (err) |
1791 | return ERR_PTR(err); |
1792 | |
1793 | return bdev; |
1794 | } |
1795 | EXPORT_SYMBOL(blkdev_get_by_dev); |
1796 | |
1797 | static int blkdev_open(struct inode * inode, struct file * filp) |
1798 | { |
1799 | struct block_device *bdev; |
1800 | |
1801 | /* |
1802 | * Preserve backwards compatibility and allow large file access |
1803 | * even if userspace doesn't ask for it explicitly. Some mkfs |
1804 | * binary needs it. We might want to drop this workaround |
1805 | * during an unstable branch. |
1806 | */ |
1807 | filp->f_flags |= O_LARGEFILE; |
1808 | |
1809 | filp->f_mode |= FMODE_NOWAIT; |
1810 | |
1811 | if (filp->f_flags & O_NDELAY) |
1812 | filp->f_mode |= FMODE_NDELAY; |
1813 | if (filp->f_flags & O_EXCL) |
1814 | filp->f_mode |= FMODE_EXCL; |
1815 | if ((filp->f_flags & O_ACCMODE) == 3) |
1816 | filp->f_mode |= FMODE_WRITE_IOCTL; |
1817 | |
1818 | bdev = bd_acquire(inode); |
1819 | if (bdev == NULL) |
1820 | return -ENOMEM; |
1821 | |
1822 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1823 | filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping); |
1824 | |
1825 | return blkdev_get(bdev, filp->f_mode, filp); |
1826 | } |
1827 | |
1828 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
1829 | { |
1830 | struct gendisk *disk = bdev->bd_disk; |
1831 | struct block_device *victim = NULL; |
1832 | |
1833 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1834 | if (for_part) |
1835 | bdev->bd_part_count--; |
1836 | |
1837 | if (!--bdev->bd_openers) { |
1838 | WARN_ON_ONCE(bdev->bd_holders); |
1839 | sync_blockdev(bdev); |
1840 | kill_bdev(bdev); |
1841 | |
1842 | bdev_write_inode(bdev); |
1843 | } |
1844 | if (bdev->bd_contains == bdev) { |
1845 | if (disk->fops->release) |
1846 | disk->fops->release(disk, mode); |
1847 | } |
1848 | if (!bdev->bd_openers) { |
1849 | disk_put_part(bdev->bd_part); |
1850 | bdev->bd_part = NULL; |
1851 | bdev->bd_disk = NULL; |
1852 | if (bdev != bdev->bd_contains) |
1853 | victim = bdev->bd_contains; |
1854 | bdev->bd_contains = NULL; |
1855 | |
1856 | put_disk_and_module(disk); |
1857 | } |
1858 | mutex_unlock(&bdev->bd_mutex); |
1859 | bdput(bdev); |
1860 | if (victim) |
1861 | __blkdev_put(victim, mode, 1); |
1862 | } |
1863 | |
1864 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
1865 | { |
1866 | mutex_lock(&bdev->bd_mutex); |
1867 | |
1868 | if (mode & FMODE_EXCL) { |
1869 | bool bdev_free; |
1870 | |
1871 | /* |
1872 | * Release a claim on the device. The holder fields |
1873 | * are protected with bdev_lock. bd_mutex is to |
1874 | * synchronize disk_holder unlinking. |
1875 | */ |
1876 | spin_lock(&bdev_lock); |
1877 | |
1878 | WARN_ON_ONCE(--bdev->bd_holders < 0); |
1879 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); |
1880 | |
1881 | /* bd_contains might point to self, check in a separate step */ |
1882 | if ((bdev_free = !bdev->bd_holders)) |
1883 | bdev->bd_holder = NULL; |
1884 | if (!bdev->bd_contains->bd_holders) |
1885 | bdev->bd_contains->bd_holder = NULL; |
1886 | |
1887 | spin_unlock(&bdev_lock); |
1888 | |
1889 | /* |
1890 | * If this was the last claim, remove holder link and |
1891 | * unblock evpoll if it was a write holder. |
1892 | */ |
1893 | if (bdev_free && bdev->bd_write_holder) { |
1894 | disk_unblock_events(bdev->bd_disk); |
1895 | bdev->bd_write_holder = false; |
1896 | } |
1897 | } |
1898 | |
1899 | /* |
1900 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE |
1901 | * event. This is to ensure detection of media removal commanded |
1902 | * from userland - e.g. eject(1). |
1903 | */ |
1904 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); |
1905 | |
1906 | mutex_unlock(&bdev->bd_mutex); |
1907 | |
1908 | __blkdev_put(bdev, mode, 0); |
1909 | } |
1910 | EXPORT_SYMBOL(blkdev_put); |
1911 | |
1912 | static int blkdev_close(struct inode * inode, struct file * filp) |
1913 | { |
1914 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
1915 | blkdev_put(bdev, filp->f_mode); |
1916 | return 0; |
1917 | } |
1918 | |
1919 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1920 | { |
1921 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
1922 | fmode_t mode = file->f_mode; |
1923 | |
1924 | /* |
1925 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have |
1926 | * to updated it before every ioctl. |
1927 | */ |
1928 | if (file->f_flags & O_NDELAY) |
1929 | mode |= FMODE_NDELAY; |
1930 | else |
1931 | mode &= ~FMODE_NDELAY; |
1932 | |
1933 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1934 | } |
1935 | |
1936 | /* |
1937 | * Write data to the block device. Only intended for the block device itself |
1938 | * and the raw driver which basically is a fake block device. |
1939 | * |
1940 | * Does not take i_mutex for the write and thus is not for general purpose |
1941 | * use. |
1942 | */ |
1943 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1944 | { |
1945 | struct file *file = iocb->ki_filp; |
1946 | struct inode *bd_inode = bdev_file_inode(file); |
1947 | loff_t size = i_size_read(bd_inode); |
1948 | struct blk_plug plug; |
1949 | ssize_t ret; |
1950 | |
1951 | if (bdev_read_only(I_BDEV(bd_inode))) |
1952 | return -EPERM; |
1953 | |
1954 | if (!iov_iter_count(from)) |
1955 | return 0; |
1956 | |
1957 | if (iocb->ki_pos >= size) |
1958 | return -ENOSPC; |
1959 | |
1960 | if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) |
1961 | return -EOPNOTSUPP; |
1962 | |
1963 | iov_iter_truncate(from, size - iocb->ki_pos); |
1964 | |
1965 | blk_start_plug(&plug); |
1966 | ret = __generic_file_write_iter(iocb, from); |
1967 | if (ret > 0) |
1968 | ret = generic_write_sync(iocb, ret); |
1969 | blk_finish_plug(&plug); |
1970 | return ret; |
1971 | } |
1972 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
1973 | |
1974 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1975 | { |
1976 | struct file *file = iocb->ki_filp; |
1977 | struct inode *bd_inode = bdev_file_inode(file); |
1978 | loff_t size = i_size_read(bd_inode); |
1979 | loff_t pos = iocb->ki_pos; |
1980 | |
1981 | if (pos >= size) |
1982 | return 0; |
1983 | |
1984 | size -= pos; |
1985 | iov_iter_truncate(to, size); |
1986 | return generic_file_read_iter(iocb, to); |
1987 | } |
1988 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
1989 | |
1990 | /* |
1991 | * Try to release a page associated with block device when the system |
1992 | * is under memory pressure. |
1993 | */ |
1994 | static int blkdev_releasepage(struct page *page, gfp_t wait) |
1995 | { |
1996 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; |
1997 | |
1998 | if (super && super->s_op->bdev_try_to_free_page) |
1999 | return super->s_op->bdev_try_to_free_page(super, page, wait); |
2000 | |
2001 | return try_to_free_buffers(page); |
2002 | } |
2003 | |
2004 | static int blkdev_writepages(struct address_space *mapping, |
2005 | struct writeback_control *wbc) |
2006 | { |
2007 | return generic_writepages(mapping, wbc); |
2008 | } |
2009 | |
2010 | static const struct address_space_operations def_blk_aops = { |
2011 | .readpage = blkdev_readpage, |
2012 | .readpages = blkdev_readpages, |
2013 | .writepage = blkdev_writepage, |
2014 | .write_begin = blkdev_write_begin, |
2015 | .write_end = blkdev_write_end, |
2016 | .writepages = blkdev_writepages, |
2017 | .releasepage = blkdev_releasepage, |
2018 | .direct_IO = blkdev_direct_IO, |
2019 | .migratepage = buffer_migrate_page_norefs, |
2020 | .is_dirty_writeback = buffer_check_dirty_writeback, |
2021 | }; |
2022 | |
2023 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
2024 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ |
2025 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) |
2026 | |
2027 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, |
2028 | loff_t len) |
2029 | { |
2030 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
2031 | struct address_space *mapping; |
2032 | loff_t end = start + len - 1; |
2033 | loff_t isize; |
2034 | int error; |
2035 | |
2036 | /* Fail if we don't recognize the flags. */ |
2037 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) |
2038 | return -EOPNOTSUPP; |
2039 | |
2040 | /* Don't go off the end of the device. */ |
2041 | isize = i_size_read(bdev->bd_inode); |
2042 | if (start >= isize) |
2043 | return -EINVAL; |
2044 | if (end >= isize) { |
2045 | if (mode & FALLOC_FL_KEEP_SIZE) { |
2046 | len = isize - start; |
2047 | end = start + len - 1; |
2048 | } else |
2049 | return -EINVAL; |
2050 | } |
2051 | |
2052 | /* |
2053 | * Don't allow IO that isn't aligned to logical block size. |
2054 | */ |
2055 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) |
2056 | return -EINVAL; |
2057 | |
2058 | /* Invalidate the page cache, including dirty pages. */ |
2059 | mapping = bdev->bd_inode->i_mapping; |
2060 | truncate_inode_pages_range(mapping, start, end); |
2061 | |
2062 | switch (mode) { |
2063 | case FALLOC_FL_ZERO_RANGE: |
2064 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: |
2065 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
2066 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP); |
2067 | break; |
2068 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: |
2069 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
2070 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK); |
2071 | break; |
2072 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: |
2073 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
2074 | GFP_KERNEL, 0); |
2075 | break; |
2076 | default: |
2077 | return -EOPNOTSUPP; |
2078 | } |
2079 | if (error) |
2080 | return error; |
2081 | |
2082 | /* |
2083 | * Invalidate again; if someone wandered in and dirtied a page, |
2084 | * the caller will be given -EBUSY. The third argument is |
2085 | * inclusive, so the rounding here is safe. |
2086 | */ |
2087 | return invalidate_inode_pages2_range(mapping, |
2088 | start >> PAGE_SHIFT, |
2089 | end >> PAGE_SHIFT); |
2090 | } |
2091 | |
2092 | const struct file_operations def_blk_fops = { |
2093 | .open = blkdev_open, |
2094 | .release = blkdev_close, |
2095 | .llseek = block_llseek, |
2096 | .read_iter = blkdev_read_iter, |
2097 | .write_iter = blkdev_write_iter, |
2098 | .iopoll = blkdev_iopoll, |
2099 | .mmap = generic_file_mmap, |
2100 | .fsync = blkdev_fsync, |
2101 | .unlocked_ioctl = block_ioctl, |
2102 | #ifdef CONFIG_COMPAT |
2103 | .compat_ioctl = compat_blkdev_ioctl, |
2104 | #endif |
2105 | .splice_read = generic_file_splice_read, |
2106 | .splice_write = iter_file_splice_write, |
2107 | .fallocate = blkdev_fallocate, |
2108 | }; |
2109 | |
2110 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) |
2111 | { |
2112 | int res; |
2113 | mm_segment_t old_fs = get_fs(); |
2114 | set_fs(KERNEL_DS); |
2115 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
2116 | set_fs(old_fs); |
2117 | return res; |
2118 | } |
2119 | |
2120 | EXPORT_SYMBOL(ioctl_by_bdev); |
2121 | |
2122 | /** |
2123 | * lookup_bdev - lookup a struct block_device by name |
2124 | * @pathname: special file representing the block device |
2125 | * |
2126 | * Get a reference to the blockdevice at @pathname in the current |
2127 | * namespace if possible and return it. Return ERR_PTR(error) |
2128 | * otherwise. |
2129 | */ |
2130 | struct block_device *lookup_bdev(const char *pathname) |
2131 | { |
2132 | struct block_device *bdev; |
2133 | struct inode *inode; |
2134 | struct path path; |
2135 | int error; |
2136 | |
2137 | if (!pathname || !*pathname) |
2138 | return ERR_PTR(-EINVAL); |
2139 | |
2140 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
2141 | if (error) |
2142 | return ERR_PTR(error); |
2143 | |
2144 | inode = d_backing_inode(path.dentry); |
2145 | error = -ENOTBLK; |
2146 | if (!S_ISBLK(inode->i_mode)) |
2147 | goto fail; |
2148 | error = -EACCES; |
2149 | if (!may_open_dev(&path)) |
2150 | goto fail; |
2151 | error = -ENOMEM; |
2152 | bdev = bd_acquire(inode); |
2153 | if (!bdev) |
2154 | goto fail; |
2155 | out: |
2156 | path_put(&path); |
2157 | return bdev; |
2158 | fail: |
2159 | bdev = ERR_PTR(error); |
2160 | goto out; |
2161 | } |
2162 | EXPORT_SYMBOL(lookup_bdev); |
2163 | |
2164 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
2165 | { |
2166 | struct super_block *sb = get_super(bdev); |
2167 | int res = 0; |
2168 | |
2169 | if (sb) { |
2170 | /* |
2171 | * no need to lock the super, get_super holds the |
2172 | * read mutex so the filesystem cannot go away |
2173 | * under us (->put_super runs with the write lock |
2174 | * hold). |
2175 | */ |
2176 | shrink_dcache_sb(sb); |
2177 | res = invalidate_inodes(sb, kill_dirty); |
2178 | drop_super(sb); |
2179 | } |
2180 | invalidate_bdev(bdev); |
2181 | return res; |
2182 | } |
2183 | EXPORT_SYMBOL(__invalidate_device); |
2184 | |
2185 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) |
2186 | { |
2187 | struct inode *inode, *old_inode = NULL; |
2188 | |
2189 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
2190 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
2191 | struct address_space *mapping = inode->i_mapping; |
2192 | struct block_device *bdev; |
2193 | |
2194 | spin_lock(&inode->i_lock); |
2195 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || |
2196 | mapping->nrpages == 0) { |
2197 | spin_unlock(&inode->i_lock); |
2198 | continue; |
2199 | } |
2200 | __iget(inode); |
2201 | spin_unlock(&inode->i_lock); |
2202 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
2203 | /* |
2204 | * We hold a reference to 'inode' so it couldn't have been |
2205 | * removed from s_inodes list while we dropped the |
2206 | * s_inode_list_lock We cannot iput the inode now as we can |
2207 | * be holding the last reference and we cannot iput it under |
2208 | * s_inode_list_lock. So we keep the reference and iput it |
2209 | * later. |
2210 | */ |
2211 | iput(old_inode); |
2212 | old_inode = inode; |
2213 | bdev = I_BDEV(inode); |
2214 | |
2215 | mutex_lock(&bdev->bd_mutex); |
2216 | if (bdev->bd_openers) |
2217 | func(bdev, arg); |
2218 | mutex_unlock(&bdev->bd_mutex); |
2219 | |
2220 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
2221 | } |
2222 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
2223 | iput(old_inode); |
2224 | } |
2225 | |