projects/linux·blob v4.0-rc1 · refs

1786 lines · 44.9 KiB

1/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <[email protected]> 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/module.h>
18#include <linux/blkpg.h>
19#include <linux/magic.h>
20#include <linux/buffer_head.h>
21#include <linux/swap.h>
22#include <linux/pagevec.h>
23#include <linux/writeback.h>
24#include <linux/mpage.h>
25#include <linux/mount.h>
26#include <linux/uio.h>
27#include <linux/namei.h>
28#include <linux/log2.h>
29#include <linux/cleancache.h>
30#include <linux/aio.h>
31#include <asm/uaccess.h>
32#include "internal.h"
33
34struct bdev_inode {
35 struct block_device bdev;
36 struct inode vfs_inode;
37};
38
39static const struct address_space_operations def_blk_aops;
40
41static inline struct bdev_inode *BDEV_I(struct inode *inode)
42{
43 return container_of(inode, struct bdev_inode, vfs_inode);
44}
45
46inline struct block_device *I_BDEV(struct inode *inode)
47{
48 return &BDEV_I(inode)->bdev;
49}
50EXPORT_SYMBOL(I_BDEV);
51
52static void bdev_write_inode(struct inode *inode)
53{
54 spin_lock(&inode->i_lock);
55 while (inode->i_state & I_DIRTY) {
56 spin_unlock(&inode->i_lock);
57 WARN_ON_ONCE(write_inode_now(inode, true));
58 spin_lock(&inode->i_lock);
59 }
60 spin_unlock(&inode->i_lock);
61}
62
63/* Kill _all_ buffers and pagecache , dirty or not.. */
64void kill_bdev(struct block_device *bdev)
65{
66 struct address_space *mapping = bdev->bd_inode->i_mapping;
67
68 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
69 return;
70
71 invalidate_bh_lrus();
72 truncate_inode_pages(mapping, 0);
73}
74EXPORT_SYMBOL(kill_bdev);
75
76/* Invalidate clean unused buffers and pagecache. */
77void invalidate_bdev(struct block_device *bdev)
78{
79 struct address_space *mapping = bdev->bd_inode->i_mapping;
80
81 if (mapping->nrpages == 0)
82 return;
83
84 invalidate_bh_lrus();
85 lru_add_drain_all(); /* make sure all lru add caches are flushed */
86 invalidate_mapping_pages(mapping, 0, -1);
87 /* 99% of the time, we don't need to flush the cleancache on the bdev.
88 * But, for the strange corners, lets be cautious
89 */
90 cleancache_invalidate_inode(mapping);
91}
92EXPORT_SYMBOL(invalidate_bdev);
93
94int set_blocksize(struct block_device *bdev, int size)
95{
96 /* Size must be a power of two, and between 512 and PAGE_SIZE */
97 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
98 return -EINVAL;
99
100 /* Size cannot be smaller than the size supported by the device */
101 if (size < bdev_logical_block_size(bdev))
102 return -EINVAL;
103
104 /* Don't change the size if it is same as current */
105 if (bdev->bd_block_size != size) {
106 sync_blockdev(bdev);
107 bdev->bd_block_size = size;
108 bdev->bd_inode->i_blkbits = blksize_bits(size);
109 kill_bdev(bdev);
110 }
111 return 0;
112}
113
114EXPORT_SYMBOL(set_blocksize);
115
116int sb_set_blocksize(struct super_block *sb, int size)
117{
118 if (set_blocksize(sb->s_bdev, size))
119 return 0;
120 /* If we get here, we know size is power of two
121 * and it's value is between 512 and PAGE_SIZE */
122 sb->s_blocksize = size;
123 sb->s_blocksize_bits = blksize_bits(size);
124 return sb->s_blocksize;
125}
126
127EXPORT_SYMBOL(sb_set_blocksize);
128
129int sb_min_blocksize(struct super_block *sb, int size)
130{
131 int minsize = bdev_logical_block_size(sb->s_bdev);
132 if (size < minsize)
133 size = minsize;
134 return sb_set_blocksize(sb, size);
135}
136
137EXPORT_SYMBOL(sb_min_blocksize);
138
139static int
140blkdev_get_block(struct inode *inode, sector_t iblock,
141 struct buffer_head *bh, int create)
142{
143 bh->b_bdev = I_BDEV(inode);
144 bh->b_blocknr = iblock;
145 set_buffer_mapped(bh);
146 return 0;
147}
148
149static ssize_t
150blkdev_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
151 loff_t offset)
152{
153 struct file *file = iocb->ki_filp;
154 struct inode *inode = file->f_mapping->host;
155
156 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iter,
157 offset, blkdev_get_block,
158 NULL, NULL, 0);
159}
160
161int __sync_blockdev(struct block_device *bdev, int wait)
162{
163 if (!bdev)
164 return 0;
165 if (!wait)
166 return filemap_flush(bdev->bd_inode->i_mapping);
167 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
168}
169
170/*
171 * Write out and wait upon all the dirty data associated with a block
172 * device via its mapping. Does not take the superblock lock.
173 */
174int sync_blockdev(struct block_device *bdev)
175{
176 return __sync_blockdev(bdev, 1);
177}
178EXPORT_SYMBOL(sync_blockdev);
179
180/*
181 * Write out and wait upon all dirty data associated with this
182 * device. Filesystem data as well as the underlying block
183 * device. Takes the superblock lock.
184 */
185int fsync_bdev(struct block_device *bdev)
186{
187 struct super_block *sb = get_super(bdev);
188 if (sb) {
189 int res = sync_filesystem(sb);
190 drop_super(sb);
191 return res;
192 }
193 return sync_blockdev(bdev);
194}
195EXPORT_SYMBOL(fsync_bdev);
196
197/**
198 * freeze_bdev -- lock a filesystem and force it into a consistent state
199 * @bdev: blockdevice to lock
200 *
201 * If a superblock is found on this device, we take the s_umount semaphore
202 * on it to make sure nobody unmounts until the snapshot creation is done.
203 * The reference counter (bd_fsfreeze_count) guarantees that only the last
204 * unfreeze process can unfreeze the frozen filesystem actually when multiple
205 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
206 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
207 * actually.
208 */
209struct super_block *freeze_bdev(struct block_device *bdev)
210{
211 struct super_block *sb;
212 int error = 0;
213
214 mutex_lock(&bdev->bd_fsfreeze_mutex);
215 if (++bdev->bd_fsfreeze_count > 1) {
216 /*
217 * We don't even need to grab a reference - the first call
218 * to freeze_bdev grab an active reference and only the last
219 * thaw_bdev drops it.
220 */
221 sb = get_super(bdev);
222 drop_super(sb);
223 mutex_unlock(&bdev->bd_fsfreeze_mutex);
224 return sb;
225 }
226
227 sb = get_active_super(bdev);
228 if (!sb)
229 goto out;
230 if (sb->s_op->freeze_super)
231 error = sb->s_op->freeze_super(sb);
232 else
233 error = freeze_super(sb);
234 if (error) {
235 deactivate_super(sb);
236 bdev->bd_fsfreeze_count--;
237 mutex_unlock(&bdev->bd_fsfreeze_mutex);
238 return ERR_PTR(error);
239 }
240 deactivate_super(sb);
241 out:
242 sync_blockdev(bdev);
243 mutex_unlock(&bdev->bd_fsfreeze_mutex);
244 return sb; /* thaw_bdev releases s->s_umount */
245}
246EXPORT_SYMBOL(freeze_bdev);
247
248/**
249 * thaw_bdev -- unlock filesystem
250 * @bdev: blockdevice to unlock
251 * @sb: associated superblock
252 *
253 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
254 */
255int thaw_bdev(struct block_device *bdev, struct super_block *sb)
256{
257 int error = -EINVAL;
258
259 mutex_lock(&bdev->bd_fsfreeze_mutex);
260 if (!bdev->bd_fsfreeze_count)
261 goto out;
262
263 error = 0;
264 if (--bdev->bd_fsfreeze_count > 0)
265 goto out;
266
267 if (!sb)
268 goto out;
269
270 if (sb->s_op->thaw_super)
271 error = sb->s_op->thaw_super(sb);
272 else
273 error = thaw_super(sb);
274 if (error) {
275 bdev->bd_fsfreeze_count++;
276 mutex_unlock(&bdev->bd_fsfreeze_mutex);
277 return error;
278 }
279out:
280 mutex_unlock(&bdev->bd_fsfreeze_mutex);
281 return 0;
282}
283EXPORT_SYMBOL(thaw_bdev);
284
285static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
286{
287 return block_write_full_page(page, blkdev_get_block, wbc);
288}
289
290static int blkdev_readpage(struct file * file, struct page * page)
291{
292 return block_read_full_page(page, blkdev_get_block);
293}
294
295static int blkdev_readpages(struct file *file, struct address_space *mapping,
296 struct list_head *pages, unsigned nr_pages)
297{
298 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
299}
300
301static int blkdev_write_begin(struct file *file, struct address_space *mapping,
302 loff_t pos, unsigned len, unsigned flags,
303 struct page **pagep, void **fsdata)
304{
305 return block_write_begin(mapping, pos, len, flags, pagep,
306 blkdev_get_block);
307}
308
309static int blkdev_write_end(struct file *file, struct address_space *mapping,
310 loff_t pos, unsigned len, unsigned copied,
311 struct page *page, void *fsdata)
312{
313 int ret;
314 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
315
316 unlock_page(page);
317 page_cache_release(page);
318
319 return ret;
320}
321
322/*
323 * private llseek:
324 * for a block special file file_inode(file)->i_size is zero
325 * so we compute the size by hand (just as in block_read/write above)
326 */
327static loff_t block_llseek(struct file *file, loff_t offset, int whence)
328{
329 struct inode *bd_inode = file->f_mapping->host;
330 loff_t retval;
331
332 mutex_lock(&bd_inode->i_mutex);
333 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
334 mutex_unlock(&bd_inode->i_mutex);
335 return retval;
336}
337
338int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
339{
340 struct inode *bd_inode = filp->f_mapping->host;
341 struct block_device *bdev = I_BDEV(bd_inode);
342 int error;
343
344 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
345 if (error)
346 return error;
347
348 /*
349 * There is no need to serialise calls to blkdev_issue_flush with
350 * i_mutex and doing so causes performance issues with concurrent
351 * O_SYNC writers to a block device.
352 */
353 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
354 if (error == -EOPNOTSUPP)
355 error = 0;
356
357 return error;
358}
359EXPORT_SYMBOL(blkdev_fsync);
360
361/**
362 * bdev_read_page() - Start reading a page from a block device
363 * @bdev: The device to read the page from
364 * @sector: The offset on the device to read the page to (need not be aligned)
365 * @page: The page to read
366 *
367 * On entry, the page should be locked. It will be unlocked when the page
368 * has been read. If the block driver implements rw_page synchronously,
369 * that will be true on exit from this function, but it need not be.
370 *
371 * Errors returned by this function are usually "soft", eg out of memory, or
372 * queue full; callers should try a different route to read this page rather
373 * than propagate an error back up the stack.
374 *
375 * Return: negative errno if an error occurs, 0 if submission was successful.
376 */
377int bdev_read_page(struct block_device *bdev, sector_t sector,
378 struct page *page)
379{
380 const struct block_device_operations *ops = bdev->bd_disk->fops;
381 if (!ops->rw_page)
382 return -EOPNOTSUPP;
383 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
384}
385EXPORT_SYMBOL_GPL(bdev_read_page);
386
387/**
388 * bdev_write_page() - Start writing a page to a block device
389 * @bdev: The device to write the page to
390 * @sector: The offset on the device to write the page to (need not be aligned)
391 * @page: The page to write
392 * @wbc: The writeback_control for the write
393 *
394 * On entry, the page should be locked and not currently under writeback.
395 * On exit, if the write started successfully, the page will be unlocked and
396 * under writeback. If the write failed already (eg the driver failed to
397 * queue the page to the device), the page will still be locked. If the
398 * caller is a ->writepage implementation, it will need to unlock the page.
399 *
400 * Errors returned by this function are usually "soft", eg out of memory, or
401 * queue full; callers should try a different route to write this page rather
402 * than propagate an error back up the stack.
403 *
404 * Return: negative errno if an error occurs, 0 if submission was successful.
405 */
406int bdev_write_page(struct block_device *bdev, sector_t sector,
407 struct page *page, struct writeback_control *wbc)
408{
409 int result;
410 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
411 const struct block_device_operations *ops = bdev->bd_disk->fops;
412 if (!ops->rw_page)
413 return -EOPNOTSUPP;
414 set_page_writeback(page);
415 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
416 if (result)
417 end_page_writeback(page);
418 else
419 unlock_page(page);
420 return result;
421}
422EXPORT_SYMBOL_GPL(bdev_write_page);
423
424/**
425 * bdev_direct_access() - Get the address for directly-accessibly memory
426 * @bdev: The device containing the memory
427 * @sector: The offset within the device
428 * @addr: Where to put the address of the memory
429 * @pfn: The Page Frame Number for the memory
430 * @size: The number of bytes requested
431 *
432 * If a block device is made up of directly addressable memory, this function
433 * will tell the caller the PFN and the address of the memory. The address
434 * may be directly dereferenced within the kernel without the need to call
435 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
436 * page tables.
437 *
438 * Return: negative errno if an error occurs, otherwise the number of bytes
439 * accessible at this address.
440 */
441long bdev_direct_access(struct block_device *bdev, sector_t sector,
442 void **addr, unsigned long *pfn, long size)
443{
444 long avail;
445 const struct block_device_operations *ops = bdev->bd_disk->fops;
446
447 if (size < 0)
448 return size;
449 if (!ops->direct_access)
450 return -EOPNOTSUPP;
451 if ((sector + DIV_ROUND_UP(size, 512)) >
452 part_nr_sects_read(bdev->bd_part))
453 return -ERANGE;
454 sector += get_start_sect(bdev);
455 if (sector % (PAGE_SIZE / 512))
456 return -EINVAL;
457 avail = ops->direct_access(bdev, sector, addr, pfn, size);
458 if (!avail)
459 return -ERANGE;
460 return min(avail, size);
461}
462EXPORT_SYMBOL_GPL(bdev_direct_access);
463
464/*
465 * pseudo-fs
466 */
467
468static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
469static struct kmem_cache * bdev_cachep __read_mostly;
470
471static struct inode *bdev_alloc_inode(struct super_block *sb)
472{
473 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
474 if (!ei)
475 return NULL;
476 return &ei->vfs_inode;
477}
478
479static void bdev_i_callback(struct rcu_head *head)
480{
481 struct inode *inode = container_of(head, struct inode, i_rcu);
482 struct bdev_inode *bdi = BDEV_I(inode);
483
484 kmem_cache_free(bdev_cachep, bdi);
485}
486
487static void bdev_destroy_inode(struct inode *inode)
488{
489 call_rcu(&inode->i_rcu, bdev_i_callback);
490}
491
492static void init_once(void *foo)
493{
494 struct bdev_inode *ei = (struct bdev_inode *) foo;
495 struct block_device *bdev = &ei->bdev;
496
497 memset(bdev, 0, sizeof(*bdev));
498 mutex_init(&bdev->bd_mutex);
499 INIT_LIST_HEAD(&bdev->bd_inodes);
500 INIT_LIST_HEAD(&bdev->bd_list);
501#ifdef CONFIG_SYSFS
502 INIT_LIST_HEAD(&bdev->bd_holder_disks);
503#endif
504 inode_init_once(&ei->vfs_inode);
505 /* Initialize mutex for freeze. */
506 mutex_init(&bdev->bd_fsfreeze_mutex);
507}
508
509static inline void __bd_forget(struct inode *inode)
510{
511 list_del_init(&inode->i_devices);
512 inode->i_bdev = NULL;
513 inode->i_mapping = &inode->i_data;
514}
515
516static void bdev_evict_inode(struct inode *inode)
517{
518 struct block_device *bdev = &BDEV_I(inode)->bdev;
519 struct list_head *p;
520 truncate_inode_pages_final(&inode->i_data);
521 invalidate_inode_buffers(inode); /* is it needed here? */
522 clear_inode(inode);
523 spin_lock(&bdev_lock);
524 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
525 __bd_forget(list_entry(p, struct inode, i_devices));
526 }
527 list_del_init(&bdev->bd_list);
528 spin_unlock(&bdev_lock);
529}
530
531static const struct super_operations bdev_sops = {
532 .statfs = simple_statfs,
533 .alloc_inode = bdev_alloc_inode,
534 .destroy_inode = bdev_destroy_inode,
535 .drop_inode = generic_delete_inode,
536 .evict_inode = bdev_evict_inode,
537};
538
539static struct dentry *bd_mount(struct file_system_type *fs_type,
540 int flags, const char *dev_name, void *data)
541{
542 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
543}
544
545static struct file_system_type bd_type = {
546 .name = "bdev",
547 .mount = bd_mount,
548 .kill_sb = kill_anon_super,
549};
550
551static struct super_block *blockdev_superblock __read_mostly;
552
553void __init bdev_cache_init(void)
554{
555 int err;
556 static struct vfsmount *bd_mnt;
557
558 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
559 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
560 SLAB_MEM_SPREAD|SLAB_PANIC),
561 init_once);
562 err = register_filesystem(&bd_type);
563 if (err)
564 panic("Cannot register bdev pseudo-fs");
565 bd_mnt = kern_mount(&bd_type);
566 if (IS_ERR(bd_mnt))
567 panic("Cannot create bdev pseudo-fs");
568 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
569}
570
571/*
572 * Most likely _very_ bad one - but then it's hardly critical for small
573 * /dev and can be fixed when somebody will need really large one.
574 * Keep in mind that it will be fed through icache hash function too.
575 */
576static inline unsigned long hash(dev_t dev)
577{
578 return MAJOR(dev)+MINOR(dev);
579}
580
581static int bdev_test(struct inode *inode, void *data)
582{
583 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
584}
585
586static int bdev_set(struct inode *inode, void *data)
587{
588 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
589 return 0;
590}
591
592static LIST_HEAD(all_bdevs);
593
594struct block_device *bdget(dev_t dev)
595{
596 struct block_device *bdev;
597 struct inode *inode;
598
599 inode = iget5_locked(blockdev_superblock, hash(dev),
600 bdev_test, bdev_set, &dev);
601
602 if (!inode)
603 return NULL;
604
605 bdev = &BDEV_I(inode)->bdev;
606
607 if (inode->i_state & I_NEW) {
608 bdev->bd_contains = NULL;
609 bdev->bd_super = NULL;
610 bdev->bd_inode = inode;
611 bdev->bd_block_size = (1 << inode->i_blkbits);
612 bdev->bd_part_count = 0;
613 bdev->bd_invalidated = 0;
614 inode->i_mode = S_IFBLK;
615 inode->i_rdev = dev;
616 inode->i_bdev = bdev;
617 inode->i_data.a_ops = &def_blk_aops;
618 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
619 spin_lock(&bdev_lock);
620 list_add(&bdev->bd_list, &all_bdevs);
621 spin_unlock(&bdev_lock);
622 unlock_new_inode(inode);
623 }
624 return bdev;
625}
626
627EXPORT_SYMBOL(bdget);
628
629/**
630 * bdgrab -- Grab a reference to an already referenced block device
631 * @bdev: Block device to grab a reference to.
632 */
633struct block_device *bdgrab(struct block_device *bdev)
634{
635 ihold(bdev->bd_inode);
636 return bdev;
637}
638EXPORT_SYMBOL(bdgrab);
639
640long nr_blockdev_pages(void)
641{
642 struct block_device *bdev;
643 long ret = 0;
644 spin_lock(&bdev_lock);
645 list_for_each_entry(bdev, &all_bdevs, bd_list) {
646 ret += bdev->bd_inode->i_mapping->nrpages;
647 }
648 spin_unlock(&bdev_lock);
649 return ret;
650}
651
652void bdput(struct block_device *bdev)
653{
654 iput(bdev->bd_inode);
655}
656
657EXPORT_SYMBOL(bdput);
658
659static struct block_device *bd_acquire(struct inode *inode)
660{
661 struct block_device *bdev;
662
663 spin_lock(&bdev_lock);
664 bdev = inode->i_bdev;
665 if (bdev) {
666 ihold(bdev->bd_inode);
667 spin_unlock(&bdev_lock);
668 return bdev;
669 }
670 spin_unlock(&bdev_lock);
671
672 bdev = bdget(inode->i_rdev);
673 if (bdev) {
674 spin_lock(&bdev_lock);
675 if (!inode->i_bdev) {
676 /*
677 * We take an additional reference to bd_inode,
678 * and it's released in clear_inode() of inode.
679 * So, we can access it via ->i_mapping always
680 * without igrab().
681 */
682 ihold(bdev->bd_inode);
683 inode->i_bdev = bdev;
684 inode->i_mapping = bdev->bd_inode->i_mapping;
685 list_add(&inode->i_devices, &bdev->bd_inodes);
686 }
687 spin_unlock(&bdev_lock);
688 }
689 return bdev;
690}
691
692int sb_is_blkdev_sb(struct super_block *sb)
693{
694 return sb == blockdev_superblock;
695}
696
697/* Call when you free inode */
698
699void bd_forget(struct inode *inode)
700{
701 struct block_device *bdev = NULL;
702
703 spin_lock(&bdev_lock);
704 if (!sb_is_blkdev_sb(inode->i_sb))
705 bdev = inode->i_bdev;
706 __bd_forget(inode);
707 spin_unlock(&bdev_lock);
708
709 if (bdev)
710 iput(bdev->bd_inode);
711}
712
713/**
714 * bd_may_claim - test whether a block device can be claimed
715 * @bdev: block device of interest
716 * @whole: whole block device containing @bdev, may equal @bdev
717 * @holder: holder trying to claim @bdev
718 *
719 * Test whether @bdev can be claimed by @holder.
720 *
721 * CONTEXT:
722 * spin_lock(&bdev_lock).
723 *
724 * RETURNS:
725 * %true if @bdev can be claimed, %false otherwise.
726 */
727static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
728 void *holder)
729{
730 if (bdev->bd_holder == holder)
731 return true; /* already a holder */
732 else if (bdev->bd_holder != NULL)
733 return false; /* held by someone else */
734 else if (bdev->bd_contains == bdev)
735 return true; /* is a whole device which isn't held */
736
737 else if (whole->bd_holder == bd_may_claim)
738 return true; /* is a partition of a device that is being partitioned */
739 else if (whole->bd_holder != NULL)
740 return false; /* is a partition of a held device */
741 else
742 return true; /* is a partition of an un-held device */
743}
744
745/**
746 * bd_prepare_to_claim - prepare to claim a block device
747 * @bdev: block device of interest
748 * @whole: the whole device containing @bdev, may equal @bdev
749 * @holder: holder trying to claim @bdev
750 *
751 * Prepare to claim @bdev. This function fails if @bdev is already
752 * claimed by another holder and waits if another claiming is in
753 * progress. This function doesn't actually claim. On successful
754 * return, the caller has ownership of bd_claiming and bd_holder[s].
755 *
756 * CONTEXT:
757 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
758 * it multiple times.
759 *
760 * RETURNS:
761 * 0 if @bdev can be claimed, -EBUSY otherwise.
762 */
763static int bd_prepare_to_claim(struct block_device *bdev,
764 struct block_device *whole, void *holder)
765{
766retry:
767 /* if someone else claimed, fail */
768 if (!bd_may_claim(bdev, whole, holder))
769 return -EBUSY;
770
771 /* if claiming is already in progress, wait for it to finish */
772 if (whole->bd_claiming) {
773 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
774 DEFINE_WAIT(wait);
775
776 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
777 spin_unlock(&bdev_lock);
778 schedule();
779 finish_wait(wq, &wait);
780 spin_lock(&bdev_lock);
781 goto retry;
782 }
783
784 /* yay, all mine */
785 return 0;
786}
787
788/**
789 * bd_start_claiming - start claiming a block device
790 * @bdev: block device of interest
791 * @holder: holder trying to claim @bdev
792 *
793 * @bdev is about to be opened exclusively. Check @bdev can be opened
794 * exclusively and mark that an exclusive open is in progress. Each
795 * successful call to this function must be matched with a call to
796 * either bd_finish_claiming() or bd_abort_claiming() (which do not
797 * fail).
798 *
799 * This function is used to gain exclusive access to the block device
800 * without actually causing other exclusive open attempts to fail. It
801 * should be used when the open sequence itself requires exclusive
802 * access but may subsequently fail.
803 *
804 * CONTEXT:
805 * Might sleep.
806 *
807 * RETURNS:
808 * Pointer to the block device containing @bdev on success, ERR_PTR()
809 * value on failure.
810 */
811static struct block_device *bd_start_claiming(struct block_device *bdev,
812 void *holder)
813{
814 struct gendisk *disk;
815 struct block_device *whole;
816 int partno, err;
817
818 might_sleep();
819
820 /*
821 * @bdev might not have been initialized properly yet, look up
822 * and grab the outer block device the hard way.
823 */
824 disk = get_gendisk(bdev->bd_dev, &partno);
825 if (!disk)
826 return ERR_PTR(-ENXIO);
827
828 /*
829 * Normally, @bdev should equal what's returned from bdget_disk()
830 * if partno is 0; however, some drivers (floppy) use multiple
831 * bdev's for the same physical device and @bdev may be one of the
832 * aliases. Keep @bdev if partno is 0. This means claimer
833 * tracking is broken for those devices but it has always been that
834 * way.
835 */
836 if (partno)
837 whole = bdget_disk(disk, 0);
838 else
839 whole = bdgrab(bdev);
840
841 module_put(disk->fops->owner);
842 put_disk(disk);
843 if (!whole)
844 return ERR_PTR(-ENOMEM);
845
846 /* prepare to claim, if successful, mark claiming in progress */
847 spin_lock(&bdev_lock);
848
849 err = bd_prepare_to_claim(bdev, whole, holder);
850 if (err == 0) {
851 whole->bd_claiming = holder;
852 spin_unlock(&bdev_lock);
853 return whole;
854 } else {
855 spin_unlock(&bdev_lock);
856 bdput(whole);
857 return ERR_PTR(err);
858 }
859}
860
861#ifdef CONFIG_SYSFS
862struct bd_holder_disk {
863 struct list_head list;
864 struct gendisk *disk;
865 int refcnt;
866};
867
868static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
869 struct gendisk *disk)
870{
871 struct bd_holder_disk *holder;
872
873 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
874 if (holder->disk == disk)
875 return holder;
876 return NULL;
877}
878
879static int add_symlink(struct kobject *from, struct kobject *to)
880{
881 return sysfs_create_link(from, to, kobject_name(to));
882}
883
884static void del_symlink(struct kobject *from, struct kobject *to)
885{
886 sysfs_remove_link(from, kobject_name(to));
887}
888
889/**
890 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
891 * @bdev: the claimed slave bdev
892 * @disk: the holding disk
893 *
894 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
895 *
896 * This functions creates the following sysfs symlinks.
897 *
898 * - from "slaves" directory of the holder @disk to the claimed @bdev
899 * - from "holders" directory of the @bdev to the holder @disk
900 *
901 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
902 * passed to bd_link_disk_holder(), then:
903 *
904 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
905 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
906 *
907 * The caller must have claimed @bdev before calling this function and
908 * ensure that both @bdev and @disk are valid during the creation and
909 * lifetime of these symlinks.
910 *
911 * CONTEXT:
912 * Might sleep.
913 *
914 * RETURNS:
915 * 0 on success, -errno on failure.
916 */
917int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
918{
919 struct bd_holder_disk *holder;
920 int ret = 0;
921
922 mutex_lock(&bdev->bd_mutex);
923
924 WARN_ON_ONCE(!bdev->bd_holder);
925
926 /* FIXME: remove the following once add_disk() handles errors */
927 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
928 goto out_unlock;
929
930 holder = bd_find_holder_disk(bdev, disk);
931 if (holder) {
932 holder->refcnt++;
933 goto out_unlock;
934 }
935
936 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
937 if (!holder) {
938 ret = -ENOMEM;
939 goto out_unlock;
940 }
941
942 INIT_LIST_HEAD(&holder->list);
943 holder->disk = disk;
944 holder->refcnt = 1;
945
946 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
947 if (ret)
948 goto out_free;
949
950 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
951 if (ret)
952 goto out_del;
953 /*
954 * bdev could be deleted beneath us which would implicitly destroy
955 * the holder directory. Hold on to it.
956 */
957 kobject_get(bdev->bd_part->holder_dir);
958
959 list_add(&holder->list, &bdev->bd_holder_disks);
960 goto out_unlock;
961
962out_del:
963 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
964out_free:
965 kfree(holder);
966out_unlock:
967 mutex_unlock(&bdev->bd_mutex);
968 return ret;
969}
970EXPORT_SYMBOL_GPL(bd_link_disk_holder);
971
972/**
973 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
974 * @bdev: the calimed slave bdev
975 * @disk: the holding disk
976 *
977 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
978 *
979 * CONTEXT:
980 * Might sleep.
981 */
982void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
983{
984 struct bd_holder_disk *holder;
985
986 mutex_lock(&bdev->bd_mutex);
987
988 holder = bd_find_holder_disk(bdev, disk);
989
990 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
991 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
992 del_symlink(bdev->bd_part->holder_dir,
993 &disk_to_dev(disk)->kobj);
994 kobject_put(bdev->bd_part->holder_dir);
995 list_del_init(&holder->list);
996 kfree(holder);
997 }
998
999 mutex_unlock(&bdev->bd_mutex);
1000}
1001EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1002#endif
1003
1004/**
1005 * flush_disk - invalidates all buffer-cache entries on a disk
1006 *
1007 * @bdev: struct block device to be flushed
1008 * @kill_dirty: flag to guide handling of dirty inodes
1009 *
1010 * Invalidates all buffer-cache entries on a disk. It should be called
1011 * when a disk has been changed -- either by a media change or online
1012 * resize.
1013 */
1014static void flush_disk(struct block_device *bdev, bool kill_dirty)
1015{
1016 if (__invalidate_device(bdev, kill_dirty)) {
1017 char name[BDEVNAME_SIZE] = "";
1018
1019 if (bdev->bd_disk)
1020 disk_name(bdev->bd_disk, 0, name);
1021 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1022 "resized disk %s\n", name);
1023 }
1024
1025 if (!bdev->bd_disk)
1026 return;
1027 if (disk_part_scan_enabled(bdev->bd_disk))
1028 bdev->bd_invalidated = 1;
1029}
1030
1031/**
1032 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1033 * @disk: struct gendisk to check
1034 * @bdev: struct bdev to adjust.
1035 *
1036 * This routine checks to see if the bdev size does not match the disk size
1037 * and adjusts it if it differs.
1038 */
1039void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1040{
1041 loff_t disk_size, bdev_size;
1042
1043 disk_size = (loff_t)get_capacity(disk) << 9;
1044 bdev_size = i_size_read(bdev->bd_inode);
1045 if (disk_size != bdev_size) {
1046 char name[BDEVNAME_SIZE];
1047
1048 disk_name(disk, 0, name);
1049 printk(KERN_INFO
1050 "%s: detected capacity change from %lld to %lld\n",
1051 name, bdev_size, disk_size);
1052 i_size_write(bdev->bd_inode, disk_size);
1053 flush_disk(bdev, false);
1054 }
1055}
1056EXPORT_SYMBOL(check_disk_size_change);
1057
1058/**
1059 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1060 * @disk: struct gendisk to be revalidated
1061 *
1062 * This routine is a wrapper for lower-level driver's revalidate_disk
1063 * call-backs. It is used to do common pre and post operations needed
1064 * for all revalidate_disk operations.
1065 */
1066int revalidate_disk(struct gendisk *disk)
1067{
1068 struct block_device *bdev;
1069 int ret = 0;
1070
1071 if (disk->fops->revalidate_disk)
1072 ret = disk->fops->revalidate_disk(disk);
1073
1074 bdev = bdget_disk(disk, 0);
1075 if (!bdev)
1076 return ret;
1077
1078 mutex_lock(&bdev->bd_mutex);
1079 check_disk_size_change(disk, bdev);
1080 bdev->bd_invalidated = 0;
1081 mutex_unlock(&bdev->bd_mutex);
1082 bdput(bdev);
1083 return ret;
1084}
1085EXPORT_SYMBOL(revalidate_disk);
1086
1087/*
1088 * This routine checks whether a removable media has been changed,
1089 * and invalidates all buffer-cache-entries in that case. This
1090 * is a relatively slow routine, so we have to try to minimize using
1091 * it. Thus it is called only upon a 'mount' or 'open'. This
1092 * is the best way of combining speed and utility, I think.
1093 * People changing diskettes in the middle of an operation deserve
1094 * to lose :-)
1095 */
1096int check_disk_change(struct block_device *bdev)
1097{
1098 struct gendisk *disk = bdev->bd_disk;
1099 const struct block_device_operations *bdops = disk->fops;
1100 unsigned int events;
1101
1102 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1103 DISK_EVENT_EJECT_REQUEST);
1104 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1105 return 0;
1106
1107 flush_disk(bdev, true);
1108 if (bdops->revalidate_disk)
1109 bdops->revalidate_disk(bdev->bd_disk);
1110 return 1;
1111}
1112
1113EXPORT_SYMBOL(check_disk_change);
1114
1115void bd_set_size(struct block_device *bdev, loff_t size)
1116{
1117 unsigned bsize = bdev_logical_block_size(bdev);
1118
1119 mutex_lock(&bdev->bd_inode->i_mutex);
1120 i_size_write(bdev->bd_inode, size);
1121 mutex_unlock(&bdev->bd_inode->i_mutex);
1122 while (bsize < PAGE_CACHE_SIZE) {
1123 if (size & bsize)
1124 break;
1125 bsize <<= 1;
1126 }
1127 bdev->bd_block_size = bsize;
1128 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1129}
1130EXPORT_SYMBOL(bd_set_size);
1131
1132static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1133
1134/*
1135 * bd_mutex locking:
1136 *
1137 * mutex_lock(part->bd_mutex)
1138 * mutex_lock_nested(whole->bd_mutex, 1)
1139 */
1140
1141static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1142{
1143 struct gendisk *disk;
1144 struct module *owner;
1145 int ret;
1146 int partno;
1147 int perm = 0;
1148
1149 if (mode & FMODE_READ)
1150 perm |= MAY_READ;
1151 if (mode & FMODE_WRITE)
1152 perm |= MAY_WRITE;
1153 /*
1154 * hooks: /n/, see "layering violations".
1155 */
1156 if (!for_part) {
1157 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1158 if (ret != 0) {
1159 bdput(bdev);
1160 return ret;
1161 }
1162 }
1163
1164 restart:
1165
1166 ret = -ENXIO;
1167 disk = get_gendisk(bdev->bd_dev, &partno);
1168 if (!disk)
1169 goto out;
1170 owner = disk->fops->owner;
1171
1172 disk_block_events(disk);
1173 mutex_lock_nested(&bdev->bd_mutex, for_part);
1174 if (!bdev->bd_openers) {
1175 bdev->bd_disk = disk;
1176 bdev->bd_queue = disk->queue;
1177 bdev->bd_contains = bdev;
1178 if (!partno) {
1179 ret = -ENXIO;
1180 bdev->bd_part = disk_get_part(disk, partno);
1181 if (!bdev->bd_part)
1182 goto out_clear;
1183
1184 ret = 0;
1185 if (disk->fops->open) {
1186 ret = disk->fops->open(bdev, mode);
1187 if (ret == -ERESTARTSYS) {
1188 /* Lost a race with 'disk' being
1189 * deleted, try again.
1190 * See md.c
1191 */
1192 disk_put_part(bdev->bd_part);
1193 bdev->bd_part = NULL;
1194 bdev->bd_disk = NULL;
1195 bdev->bd_queue = NULL;
1196 mutex_unlock(&bdev->bd_mutex);
1197 disk_unblock_events(disk);
1198 put_disk(disk);
1199 module_put(owner);
1200 goto restart;
1201 }
1202 }
1203
1204 if (!ret)
1205 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1206
1207 /*
1208 * If the device is invalidated, rescan partition
1209 * if open succeeded or failed with -ENOMEDIUM.
1210 * The latter is necessary to prevent ghost
1211 * partitions on a removed medium.
1212 */
1213 if (bdev->bd_invalidated) {
1214 if (!ret)
1215 rescan_partitions(disk, bdev);
1216 else if (ret == -ENOMEDIUM)
1217 invalidate_partitions(disk, bdev);
1218 }
1219 if (ret)
1220 goto out_clear;
1221 } else {
1222 struct block_device *whole;
1223 whole = bdget_disk(disk, 0);
1224 ret = -ENOMEM;
1225 if (!whole)
1226 goto out_clear;
1227 BUG_ON(for_part);
1228 ret = __blkdev_get(whole, mode, 1);
1229 if (ret)
1230 goto out_clear;
1231 bdev->bd_contains = whole;
1232 bdev->bd_part = disk_get_part(disk, partno);
1233 if (!(disk->flags & GENHD_FL_UP) ||
1234 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1235 ret = -ENXIO;
1236 goto out_clear;
1237 }
1238 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1239 }
1240 } else {
1241 if (bdev->bd_contains == bdev) {
1242 ret = 0;
1243 if (bdev->bd_disk->fops->open)
1244 ret = bdev->bd_disk->fops->open(bdev, mode);
1245 /* the same as first opener case, read comment there */
1246 if (bdev->bd_invalidated) {
1247 if (!ret)
1248 rescan_partitions(bdev->bd_disk, bdev);
1249 else if (ret == -ENOMEDIUM)
1250 invalidate_partitions(bdev->bd_disk, bdev);
1251 }
1252 if (ret)
1253 goto out_unlock_bdev;
1254 }
1255 /* only one opener holds refs to the module and disk */
1256 put_disk(disk);
1257 module_put(owner);
1258 }
1259 bdev->bd_openers++;
1260 if (for_part)
1261 bdev->bd_part_count++;
1262 mutex_unlock(&bdev->bd_mutex);
1263 disk_unblock_events(disk);
1264 return 0;
1265
1266 out_clear:
1267 disk_put_part(bdev->bd_part);
1268 bdev->bd_disk = NULL;
1269 bdev->bd_part = NULL;
1270 bdev->bd_queue = NULL;
1271 if (bdev != bdev->bd_contains)
1272 __blkdev_put(bdev->bd_contains, mode, 1);
1273 bdev->bd_contains = NULL;
1274 out_unlock_bdev:
1275 mutex_unlock(&bdev->bd_mutex);
1276 disk_unblock_events(disk);
1277 put_disk(disk);
1278 module_put(owner);
1279 out:
1280 bdput(bdev);
1281
1282 return ret;
1283}
1284
1285/**
1286 * blkdev_get - open a block device
1287 * @bdev: block_device to open
1288 * @mode: FMODE_* mask
1289 * @holder: exclusive holder identifier
1290 *
1291 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1292 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1293 * @holder is invalid. Exclusive opens may nest for the same @holder.
1294 *
1295 * On success, the reference count of @bdev is unchanged. On failure,
1296 * @bdev is put.
1297 *
1298 * CONTEXT:
1299 * Might sleep.
1300 *
1301 * RETURNS:
1302 * 0 on success, -errno on failure.
1303 */
1304int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1305{
1306 struct block_device *whole = NULL;
1307 int res;
1308
1309 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1310
1311 if ((mode & FMODE_EXCL) && holder) {
1312 whole = bd_start_claiming(bdev, holder);
1313 if (IS_ERR(whole)) {
1314 bdput(bdev);
1315 return PTR_ERR(whole);
1316 }
1317 }
1318
1319 res = __blkdev_get(bdev, mode, 0);
1320
1321 if (whole) {
1322 struct gendisk *disk = whole->bd_disk;
1323
1324 /* finish claiming */
1325 mutex_lock(&bdev->bd_mutex);
1326 spin_lock(&bdev_lock);
1327
1328 if (!res) {
1329 BUG_ON(!bd_may_claim(bdev, whole, holder));
1330 /*
1331 * Note that for a whole device bd_holders
1332 * will be incremented twice, and bd_holder
1333 * will be set to bd_may_claim before being
1334 * set to holder
1335 */
1336 whole->bd_holders++;
1337 whole->bd_holder = bd_may_claim;
1338 bdev->bd_holders++;
1339 bdev->bd_holder = holder;
1340 }
1341
1342 /* tell others that we're done */
1343 BUG_ON(whole->bd_claiming != holder);
1344 whole->bd_claiming = NULL;
1345 wake_up_bit(&whole->bd_claiming, 0);
1346
1347 spin_unlock(&bdev_lock);
1348
1349 /*
1350 * Block event polling for write claims if requested. Any
1351 * write holder makes the write_holder state stick until
1352 * all are released. This is good enough and tracking
1353 * individual writeable reference is too fragile given the
1354 * way @mode is used in blkdev_get/put().
1355 */
1356 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1357 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1358 bdev->bd_write_holder = true;
1359 disk_block_events(disk);
1360 }
1361
1362 mutex_unlock(&bdev->bd_mutex);
1363 bdput(whole);
1364 }
1365
1366 return res;
1367}
1368EXPORT_SYMBOL(blkdev_get);
1369
1370/**
1371 * blkdev_get_by_path - open a block device by name
1372 * @path: path to the block device to open
1373 * @mode: FMODE_* mask
1374 * @holder: exclusive holder identifier
1375 *
1376 * Open the blockdevice described by the device file at @path. @mode
1377 * and @holder are identical to blkdev_get().
1378 *
1379 * On success, the returned block_device has reference count of one.
1380 *
1381 * CONTEXT:
1382 * Might sleep.
1383 *
1384 * RETURNS:
1385 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1386 */
1387struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1388 void *holder)
1389{
1390 struct block_device *bdev;
1391 int err;
1392
1393 bdev = lookup_bdev(path);
1394 if (IS_ERR(bdev))
1395 return bdev;
1396
1397 err = blkdev_get(bdev, mode, holder);
1398 if (err)
1399 return ERR_PTR(err);
1400
1401 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1402 blkdev_put(bdev, mode);
1403 return ERR_PTR(-EACCES);
1404 }
1405
1406 return bdev;
1407}
1408EXPORT_SYMBOL(blkdev_get_by_path);
1409
1410/**
1411 * blkdev_get_by_dev - open a block device by device number
1412 * @dev: device number of block device to open
1413 * @mode: FMODE_* mask
1414 * @holder: exclusive holder identifier
1415 *
1416 * Open the blockdevice described by device number @dev. @mode and
1417 * @holder are identical to blkdev_get().
1418 *
1419 * Use it ONLY if you really do not have anything better - i.e. when
1420 * you are behind a truly sucky interface and all you are given is a
1421 * device number. _Never_ to be used for internal purposes. If you
1422 * ever need it - reconsider your API.
1423 *
1424 * On success, the returned block_device has reference count of one.
1425 *
1426 * CONTEXT:
1427 * Might sleep.
1428 *
1429 * RETURNS:
1430 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1431 */
1432struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1433{
1434 struct block_device *bdev;
1435 int err;
1436
1437 bdev = bdget(dev);
1438 if (!bdev)
1439 return ERR_PTR(-ENOMEM);
1440
1441 err = blkdev_get(bdev, mode, holder);
1442 if (err)
1443 return ERR_PTR(err);
1444
1445 return bdev;
1446}
1447EXPORT_SYMBOL(blkdev_get_by_dev);
1448
1449static int blkdev_open(struct inode * inode, struct file * filp)
1450{
1451 struct block_device *bdev;
1452
1453 /*
1454 * Preserve backwards compatibility and allow large file access
1455 * even if userspace doesn't ask for it explicitly. Some mkfs
1456 * binary needs it. We might want to drop this workaround
1457 * during an unstable branch.
1458 */
1459 filp->f_flags |= O_LARGEFILE;
1460
1461 if (filp->f_flags & O_NDELAY)
1462 filp->f_mode |= FMODE_NDELAY;
1463 if (filp->f_flags & O_EXCL)
1464 filp->f_mode |= FMODE_EXCL;
1465 if ((filp->f_flags & O_ACCMODE) == 3)
1466 filp->f_mode |= FMODE_WRITE_IOCTL;
1467
1468 bdev = bd_acquire(inode);
1469 if (bdev == NULL)
1470 return -ENOMEM;
1471
1472 filp->f_mapping = bdev->bd_inode->i_mapping;
1473
1474 return blkdev_get(bdev, filp->f_mode, filp);
1475}
1476
1477static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1478{
1479 struct gendisk *disk = bdev->bd_disk;
1480 struct block_device *victim = NULL;
1481
1482 mutex_lock_nested(&bdev->bd_mutex, for_part);
1483 if (for_part)
1484 bdev->bd_part_count--;
1485
1486 if (!--bdev->bd_openers) {
1487 WARN_ON_ONCE(bdev->bd_holders);
1488 sync_blockdev(bdev);
1489 kill_bdev(bdev);
1490 /*
1491 * ->release can cause the queue to disappear, so flush all
1492 * dirty data before.
1493 */
1494 bdev_write_inode(bdev->bd_inode);
1495 }
1496 if (bdev->bd_contains == bdev) {
1497 if (disk->fops->release)
1498 disk->fops->release(disk, mode);
1499 }
1500 if (!bdev->bd_openers) {
1501 struct module *owner = disk->fops->owner;
1502
1503 disk_put_part(bdev->bd_part);
1504 bdev->bd_part = NULL;
1505 bdev->bd_disk = NULL;
1506 if (bdev != bdev->bd_contains)
1507 victim = bdev->bd_contains;
1508 bdev->bd_contains = NULL;
1509
1510 put_disk(disk);
1511 module_put(owner);
1512 }
1513 mutex_unlock(&bdev->bd_mutex);
1514 bdput(bdev);
1515 if (victim)
1516 __blkdev_put(victim, mode, 1);
1517}
1518
1519void blkdev_put(struct block_device *bdev, fmode_t mode)
1520{
1521 mutex_lock(&bdev->bd_mutex);
1522
1523 if (mode & FMODE_EXCL) {
1524 bool bdev_free;
1525
1526 /*
1527 * Release a claim on the device. The holder fields
1528 * are protected with bdev_lock. bd_mutex is to
1529 * synchronize disk_holder unlinking.
1530 */
1531 spin_lock(&bdev_lock);
1532
1533 WARN_ON_ONCE(--bdev->bd_holders < 0);
1534 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1535
1536 /* bd_contains might point to self, check in a separate step */
1537 if ((bdev_free = !bdev->bd_holders))
1538 bdev->bd_holder = NULL;
1539 if (!bdev->bd_contains->bd_holders)
1540 bdev->bd_contains->bd_holder = NULL;
1541
1542 spin_unlock(&bdev_lock);
1543
1544 /*
1545 * If this was the last claim, remove holder link and
1546 * unblock evpoll if it was a write holder.
1547 */
1548 if (bdev_free && bdev->bd_write_holder) {
1549 disk_unblock_events(bdev->bd_disk);
1550 bdev->bd_write_holder = false;
1551 }
1552 }
1553
1554 /*
1555 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1556 * event. This is to ensure detection of media removal commanded
1557 * from userland - e.g. eject(1).
1558 */
1559 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1560
1561 mutex_unlock(&bdev->bd_mutex);
1562
1563 __blkdev_put(bdev, mode, 0);
1564}
1565EXPORT_SYMBOL(blkdev_put);
1566
1567static int blkdev_close(struct inode * inode, struct file * filp)
1568{
1569 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1570 blkdev_put(bdev, filp->f_mode);
1571 return 0;
1572}
1573
1574static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1575{
1576 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1577 fmode_t mode = file->f_mode;
1578
1579 /*
1580 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1581 * to updated it before every ioctl.
1582 */
1583 if (file->f_flags & O_NDELAY)
1584 mode |= FMODE_NDELAY;
1585 else
1586 mode &= ~FMODE_NDELAY;
1587
1588 return blkdev_ioctl(bdev, mode, cmd, arg);
1589}
1590
1591/*
1592 * Write data to the block device. Only intended for the block device itself
1593 * and the raw driver which basically is a fake block device.
1594 *
1595 * Does not take i_mutex for the write and thus is not for general purpose
1596 * use.
1597 */
1598ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1599{
1600 struct file *file = iocb->ki_filp;
1601 struct blk_plug plug;
1602 ssize_t ret;
1603
1604 blk_start_plug(&plug);
1605 ret = __generic_file_write_iter(iocb, from);
1606 if (ret > 0) {
1607 ssize_t err;
1608 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1609 if (err < 0)
1610 ret = err;
1611 }
1612 blk_finish_plug(&plug);
1613 return ret;
1614}
1615EXPORT_SYMBOL_GPL(blkdev_write_iter);
1616
1617ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1618{
1619 struct file *file = iocb->ki_filp;
1620 struct inode *bd_inode = file->f_mapping->host;
1621 loff_t size = i_size_read(bd_inode);
1622 loff_t pos = iocb->ki_pos;
1623
1624 if (pos >= size)
1625 return 0;
1626
1627 size -= pos;
1628 iov_iter_truncate(to, size);
1629 return generic_file_read_iter(iocb, to);
1630}
1631EXPORT_SYMBOL_GPL(blkdev_read_iter);
1632
1633/*
1634 * Try to release a page associated with block device when the system
1635 * is under memory pressure.
1636 */
1637static int blkdev_releasepage(struct page *page, gfp_t wait)
1638{
1639 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1640
1641 if (super && super->s_op->bdev_try_to_free_page)
1642 return super->s_op->bdev_try_to_free_page(super, page, wait);
1643
1644 return try_to_free_buffers(page);
1645}
1646
1647static const struct address_space_operations def_blk_aops = {
1648 .readpage = blkdev_readpage,
1649 .readpages = blkdev_readpages,
1650 .writepage = blkdev_writepage,
1651 .write_begin = blkdev_write_begin,
1652 .write_end = blkdev_write_end,
1653 .writepages = generic_writepages,
1654 .releasepage = blkdev_releasepage,
1655 .direct_IO = blkdev_direct_IO,
1656 .is_dirty_writeback = buffer_check_dirty_writeback,
1657};
1658
1659const struct file_operations def_blk_fops = {
1660 .open = blkdev_open,
1661 .release = blkdev_close,
1662 .llseek = block_llseek,
1663 .read = new_sync_read,
1664 .write = new_sync_write,
1665 .read_iter = blkdev_read_iter,
1666 .write_iter = blkdev_write_iter,
1667 .mmap = generic_file_mmap,
1668 .fsync = blkdev_fsync,
1669 .unlocked_ioctl = block_ioctl,
1670#ifdef CONFIG_COMPAT
1671 .compat_ioctl = compat_blkdev_ioctl,
1672#endif
1673 .splice_read = generic_file_splice_read,
1674 .splice_write = iter_file_splice_write,
1675};
1676
1677int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1678{
1679 int res;
1680 mm_segment_t old_fs = get_fs();
1681 set_fs(KERNEL_DS);
1682 res = blkdev_ioctl(bdev, 0, cmd, arg);
1683 set_fs(old_fs);
1684 return res;
1685}
1686
1687EXPORT_SYMBOL(ioctl_by_bdev);
1688
1689/**
1690 * lookup_bdev - lookup a struct block_device by name
1691 * @pathname: special file representing the block device
1692 *
1693 * Get a reference to the blockdevice at @pathname in the current
1694 * namespace if possible and return it. Return ERR_PTR(error)
1695 * otherwise.
1696 */
1697struct block_device *lookup_bdev(const char *pathname)
1698{
1699 struct block_device *bdev;
1700 struct inode *inode;
1701 struct path path;
1702 int error;
1703
1704 if (!pathname || !*pathname)
1705 return ERR_PTR(-EINVAL);
1706
1707 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1708 if (error)
1709 return ERR_PTR(error);
1710
1711 inode = path.dentry->d_inode;
1712 error = -ENOTBLK;
1713 if (!S_ISBLK(inode->i_mode))
1714 goto fail;
1715 error = -EACCES;
1716 if (path.mnt->mnt_flags & MNT_NODEV)
1717 goto fail;
1718 error = -ENOMEM;
1719 bdev = bd_acquire(inode);
1720 if (!bdev)
1721 goto fail;
1722out:
1723 path_put(&path);
1724 return bdev;
1725fail:
1726 bdev = ERR_PTR(error);
1727 goto out;
1728}
1729EXPORT_SYMBOL(lookup_bdev);
1730
1731int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1732{
1733 struct super_block *sb = get_super(bdev);
1734 int res = 0;
1735
1736 if (sb) {
1737 /*
1738 * no need to lock the super, get_super holds the
1739 * read mutex so the filesystem cannot go away
1740 * under us (->put_super runs with the write lock
1741 * hold).
1742 */
1743 shrink_dcache_sb(sb);
1744 res = invalidate_inodes(sb, kill_dirty);
1745 drop_super(sb);
1746 }
1747 invalidate_bdev(bdev);
1748 return res;
1749}
1750EXPORT_SYMBOL(__invalidate_device);
1751
1752void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1753{
1754 struct inode *inode, *old_inode = NULL;
1755
1756 spin_lock(&inode_sb_list_lock);
1757 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1758 struct address_space *mapping = inode->i_mapping;
1759
1760 spin_lock(&inode->i_lock);
1761 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1762 mapping->nrpages == 0) {
1763 spin_unlock(&inode->i_lock);
1764 continue;
1765 }
1766 __iget(inode);
1767 spin_unlock(&inode->i_lock);
1768 spin_unlock(&inode_sb_list_lock);
1769 /*
1770 * We hold a reference to 'inode' so it couldn't have been
1771 * removed from s_inodes list while we dropped the
1772 * inode_sb_list_lock. We cannot iput the inode now as we can
1773 * be holding the last reference and we cannot iput it under
1774 * inode_sb_list_lock. So we keep the reference and iput it
1775 * later.
1776 */
1777 iput(old_inode);
1778 old_inode = inode;
1779
1780 func(I_BDEV(inode), arg);
1781
1782 spin_lock(&inode_sb_list_lock);
1783 }
1784 spin_unlock(&inode_sb_list_lock);
1785 iput(old_inode);
1786}