projects/linux·blob v4.5-rc3 · refs

1899 lines · 48.1 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/backing-dev.h>
18#include <linux/module.h>
19#include <linux/blkpg.h>
20#include <linux/magic.h>
21#include <linux/buffer_head.h>
22#include <linux/swap.h>
23#include <linux/pagevec.h>
24#include <linux/writeback.h>
25#include <linux/mpage.h>
26#include <linux/mount.h>
27#include <linux/uio.h>
28#include <linux/namei.h>
29#include <linux/log2.h>
30#include <linux/cleancache.h>
31#include <linux/dax.h>
32#include <asm/uaccess.h>
33#include "internal.h"
34
35struct bdev_inode {
36 struct block_device bdev;
37 struct inode vfs_inode;
38};
39
40static const struct address_space_operations def_blk_aops;
41
42static inline struct bdev_inode *BDEV_I(struct inode *inode)
43{
44 return container_of(inode, struct bdev_inode, vfs_inode);
45}
46
47struct block_device *I_BDEV(struct inode *inode)
48{
49 return &BDEV_I(inode)->bdev;
50}
51EXPORT_SYMBOL(I_BDEV);
52
53static void bdev_write_inode(struct block_device *bdev)
54{
55 struct inode *inode = bdev->bd_inode;
56 int ret;
57
58 spin_lock(&inode->i_lock);
59 while (inode->i_state & I_DIRTY) {
60 spin_unlock(&inode->i_lock);
61 ret = write_inode_now(inode, true);
62 if (ret) {
63 char name[BDEVNAME_SIZE];
64 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
65 "for block device %s (err=%d).\n",
66 bdevname(bdev, name), ret);
67 }
68 spin_lock(&inode->i_lock);
69 }
70 spin_unlock(&inode->i_lock);
71}
72
73/* Kill _all_ buffers and pagecache , dirty or not.. */
74void kill_bdev(struct block_device *bdev)
75{
76 struct address_space *mapping = bdev->bd_inode->i_mapping;
77
78 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
79 return;
80
81 invalidate_bh_lrus();
82 truncate_inode_pages(mapping, 0);
83}
84EXPORT_SYMBOL(kill_bdev);
85
86/* Invalidate clean unused buffers and pagecache. */
87void invalidate_bdev(struct block_device *bdev)
88{
89 struct address_space *mapping = bdev->bd_inode->i_mapping;
90
91 if (mapping->nrpages == 0)
92 return;
93
94 invalidate_bh_lrus();
95 lru_add_drain_all(); /* make sure all lru add caches are flushed */
96 invalidate_mapping_pages(mapping, 0, -1);
97 /* 99% of the time, we don't need to flush the cleancache on the bdev.
98 * But, for the strange corners, lets be cautious
99 */
100 cleancache_invalidate_inode(mapping);
101}
102EXPORT_SYMBOL(invalidate_bdev);
103
104int set_blocksize(struct block_device *bdev, int size)
105{
106 /* Size must be a power of two, and between 512 and PAGE_SIZE */
107 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
108 return -EINVAL;
109
110 /* Size cannot be smaller than the size supported by the device */
111 if (size < bdev_logical_block_size(bdev))
112 return -EINVAL;
113
114 /* Don't change the size if it is same as current */
115 if (bdev->bd_block_size != size) {
116 sync_blockdev(bdev);
117 bdev->bd_block_size = size;
118 bdev->bd_inode->i_blkbits = blksize_bits(size);
119 kill_bdev(bdev);
120 }
121 return 0;
122}
123
124EXPORT_SYMBOL(set_blocksize);
125
126int sb_set_blocksize(struct super_block *sb, int size)
127{
128 if (set_blocksize(sb->s_bdev, size))
129 return 0;
130 /* If we get here, we know size is power of two
131 * and it's value is between 512 and PAGE_SIZE */
132 sb->s_blocksize = size;
133 sb->s_blocksize_bits = blksize_bits(size);
134 return sb->s_blocksize;
135}
136
137EXPORT_SYMBOL(sb_set_blocksize);
138
139int sb_min_blocksize(struct super_block *sb, int size)
140{
141 int minsize = bdev_logical_block_size(sb->s_bdev);
142 if (size < minsize)
143 size = minsize;
144 return sb_set_blocksize(sb, size);
145}
146
147EXPORT_SYMBOL(sb_min_blocksize);
148
149static int
150blkdev_get_block(struct inode *inode, sector_t iblock,
151 struct buffer_head *bh, int create)
152{
153 bh->b_bdev = I_BDEV(inode);
154 bh->b_blocknr = iblock;
155 set_buffer_mapped(bh);
156 return 0;
157}
158
159static struct inode *bdev_file_inode(struct file *file)
160{
161 return file->f_mapping->host;
162}
163
164static ssize_t
165blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
166{
167 struct file *file = iocb->ki_filp;
168 struct inode *inode = bdev_file_inode(file);
169
170 if (IS_DAX(inode))
171 return dax_do_io(iocb, inode, iter, offset, blkdev_get_block,
172 NULL, DIO_SKIP_DIO_COUNT);
173 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset,
174 blkdev_get_block, NULL, NULL,
175 DIO_SKIP_DIO_COUNT);
176}
177
178int __sync_blockdev(struct block_device *bdev, int wait)
179{
180 if (!bdev)
181 return 0;
182 if (!wait)
183 return filemap_flush(bdev->bd_inode->i_mapping);
184 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
185}
186
187/*
188 * Write out and wait upon all the dirty data associated with a block
189 * device via its mapping. Does not take the superblock lock.
190 */
191int sync_blockdev(struct block_device *bdev)
192{
193 return __sync_blockdev(bdev, 1);
194}
195EXPORT_SYMBOL(sync_blockdev);
196
197/*
198 * Write out and wait upon all dirty data associated with this
199 * device. Filesystem data as well as the underlying block
200 * device. Takes the superblock lock.
201 */
202int fsync_bdev(struct block_device *bdev)
203{
204 struct super_block *sb = get_super(bdev);
205 if (sb) {
206 int res = sync_filesystem(sb);
207 drop_super(sb);
208 return res;
209 }
210 return sync_blockdev(bdev);
211}
212EXPORT_SYMBOL(fsync_bdev);
213
214/**
215 * freeze_bdev -- lock a filesystem and force it into a consistent state
216 * @bdev: blockdevice to lock
217 *
218 * If a superblock is found on this device, we take the s_umount semaphore
219 * on it to make sure nobody unmounts until the snapshot creation is done.
220 * The reference counter (bd_fsfreeze_count) guarantees that only the last
221 * unfreeze process can unfreeze the frozen filesystem actually when multiple
222 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
223 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
224 * actually.
225 */
226struct super_block *freeze_bdev(struct block_device *bdev)
227{
228 struct super_block *sb;
229 int error = 0;
230
231 mutex_lock(&bdev->bd_fsfreeze_mutex);
232 if (++bdev->bd_fsfreeze_count > 1) {
233 /*
234 * We don't even need to grab a reference - the first call
235 * to freeze_bdev grab an active reference and only the last
236 * thaw_bdev drops it.
237 */
238 sb = get_super(bdev);
239 drop_super(sb);
240 mutex_unlock(&bdev->bd_fsfreeze_mutex);
241 return sb;
242 }
243
244 sb = get_active_super(bdev);
245 if (!sb)
246 goto out;
247 if (sb->s_op->freeze_super)
248 error = sb->s_op->freeze_super(sb);
249 else
250 error = freeze_super(sb);
251 if (error) {
252 deactivate_super(sb);
253 bdev->bd_fsfreeze_count--;
254 mutex_unlock(&bdev->bd_fsfreeze_mutex);
255 return ERR_PTR(error);
256 }
257 deactivate_super(sb);
258 out:
259 sync_blockdev(bdev);
260 mutex_unlock(&bdev->bd_fsfreeze_mutex);
261 return sb; /* thaw_bdev releases s->s_umount */
262}
263EXPORT_SYMBOL(freeze_bdev);
264
265/**
266 * thaw_bdev -- unlock filesystem
267 * @bdev: blockdevice to unlock
268 * @sb: associated superblock
269 *
270 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
271 */
272int thaw_bdev(struct block_device *bdev, struct super_block *sb)
273{
274 int error = -EINVAL;
275
276 mutex_lock(&bdev->bd_fsfreeze_mutex);
277 if (!bdev->bd_fsfreeze_count)
278 goto out;
279
280 error = 0;
281 if (--bdev->bd_fsfreeze_count > 0)
282 goto out;
283
284 if (!sb)
285 goto out;
286
287 if (sb->s_op->thaw_super)
288 error = sb->s_op->thaw_super(sb);
289 else
290 error = thaw_super(sb);
291 if (error) {
292 bdev->bd_fsfreeze_count++;
293 mutex_unlock(&bdev->bd_fsfreeze_mutex);
294 return error;
295 }
296out:
297 mutex_unlock(&bdev->bd_fsfreeze_mutex);
298 return 0;
299}
300EXPORT_SYMBOL(thaw_bdev);
301
302static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
303{
304 return block_write_full_page(page, blkdev_get_block, wbc);
305}
306
307static int blkdev_readpage(struct file * file, struct page * page)
308{
309 return block_read_full_page(page, blkdev_get_block);
310}
311
312static int blkdev_readpages(struct file *file, struct address_space *mapping,
313 struct list_head *pages, unsigned nr_pages)
314{
315 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
316}
317
318static int blkdev_write_begin(struct file *file, struct address_space *mapping,
319 loff_t pos, unsigned len, unsigned flags,
320 struct page **pagep, void **fsdata)
321{
322 return block_write_begin(mapping, pos, len, flags, pagep,
323 blkdev_get_block);
324}
325
326static int blkdev_write_end(struct file *file, struct address_space *mapping,
327 loff_t pos, unsigned len, unsigned copied,
328 struct page *page, void *fsdata)
329{
330 int ret;
331 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
332
333 unlock_page(page);
334 page_cache_release(page);
335
336 return ret;
337}
338
339/*
340 * private llseek:
341 * for a block special file file_inode(file)->i_size is zero
342 * so we compute the size by hand (just as in block_read/write above)
343 */
344static loff_t block_llseek(struct file *file, loff_t offset, int whence)
345{
346 struct inode *bd_inode = bdev_file_inode(file);
347 loff_t retval;
348
349 inode_lock(bd_inode);
350 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
351 inode_unlock(bd_inode);
352 return retval;
353}
354
355int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
356{
357 struct inode *bd_inode = bdev_file_inode(filp);
358 struct block_device *bdev = I_BDEV(bd_inode);
359 int error;
360
361 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
362 if (error)
363 return error;
364
365 /*
366 * There is no need to serialise calls to blkdev_issue_flush with
367 * i_mutex and doing so causes performance issues with concurrent
368 * O_SYNC writers to a block device.
369 */
370 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
371 if (error == -EOPNOTSUPP)
372 error = 0;
373
374 return error;
375}
376EXPORT_SYMBOL(blkdev_fsync);
377
378/**
379 * bdev_read_page() - Start reading a page from a block device
380 * @bdev: The device to read the page from
381 * @sector: The offset on the device to read the page to (need not be aligned)
382 * @page: The page to read
383 *
384 * On entry, the page should be locked. It will be unlocked when the page
385 * has been read. If the block driver implements rw_page synchronously,
386 * that will be true on exit from this function, but it need not be.
387 *
388 * Errors returned by this function are usually "soft", eg out of memory, or
389 * queue full; callers should try a different route to read this page rather
390 * than propagate an error back up the stack.
391 *
392 * Return: negative errno if an error occurs, 0 if submission was successful.
393 */
394int bdev_read_page(struct block_device *bdev, sector_t sector,
395 struct page *page)
396{
397 const struct block_device_operations *ops = bdev->bd_disk->fops;
398 int result = -EOPNOTSUPP;
399
400 if (!ops->rw_page || bdev_get_integrity(bdev))
401 return result;
402
403 result = blk_queue_enter(bdev->bd_queue, false);
404 if (result)
405 return result;
406 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
407 blk_queue_exit(bdev->bd_queue);
408 return result;
409}
410EXPORT_SYMBOL_GPL(bdev_read_page);
411
412/**
413 * bdev_write_page() - Start writing a page to a block device
414 * @bdev: The device to write the page to
415 * @sector: The offset on the device to write the page to (need not be aligned)
416 * @page: The page to write
417 * @wbc: The writeback_control for the write
418 *
419 * On entry, the page should be locked and not currently under writeback.
420 * On exit, if the write started successfully, the page will be unlocked and
421 * under writeback. If the write failed already (eg the driver failed to
422 * queue the page to the device), the page will still be locked. If the
423 * caller is a ->writepage implementation, it will need to unlock the page.
424 *
425 * Errors returned by this function are usually "soft", eg out of memory, or
426 * queue full; callers should try a different route to write this page rather
427 * than propagate an error back up the stack.
428 *
429 * Return: negative errno if an error occurs, 0 if submission was successful.
430 */
431int bdev_write_page(struct block_device *bdev, sector_t sector,
432 struct page *page, struct writeback_control *wbc)
433{
434 int result;
435 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
436 const struct block_device_operations *ops = bdev->bd_disk->fops;
437
438 if (!ops->rw_page || bdev_get_integrity(bdev))
439 return -EOPNOTSUPP;
440 result = blk_queue_enter(bdev->bd_queue, false);
441 if (result)
442 return result;
443
444 set_page_writeback(page);
445 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
446 if (result)
447 end_page_writeback(page);
448 else
449 unlock_page(page);
450 blk_queue_exit(bdev->bd_queue);
451 return result;
452}
453EXPORT_SYMBOL_GPL(bdev_write_page);
454
455/**
456 * bdev_direct_access() - Get the address for directly-accessibly memory
457 * @bdev: The device containing the memory
458 * @dax: control and output parameters for ->direct_access
459 *
460 * If a block device is made up of directly addressable memory, this function
461 * will tell the caller the PFN and the address of the memory. The address
462 * may be directly dereferenced within the kernel without the need to call
463 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
464 * page tables.
465 *
466 * Return: negative errno if an error occurs, otherwise the number of bytes
467 * accessible at this address.
468 */
469long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
470{
471 sector_t sector = dax->sector;
472 long avail, size = dax->size;
473 const struct block_device_operations *ops = bdev->bd_disk->fops;
474
475 /*
476 * The device driver is allowed to sleep, in order to make the
477 * memory directly accessible.
478 */
479 might_sleep();
480
481 if (size < 0)
482 return size;
483 if (!ops->direct_access)
484 return -EOPNOTSUPP;
485 if ((sector + DIV_ROUND_UP(size, 512)) >
486 part_nr_sects_read(bdev->bd_part))
487 return -ERANGE;
488 sector += get_start_sect(bdev);
489 if (sector % (PAGE_SIZE / 512))
490 return -EINVAL;
491 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn);
492 if (!avail)
493 return -ERANGE;
494 if (avail > 0 && avail & ~PAGE_MASK)
495 return -ENXIO;
496 return min(avail, size);
497}
498EXPORT_SYMBOL_GPL(bdev_direct_access);
499
500/*
501 * pseudo-fs
502 */
503
504static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
505static struct kmem_cache * bdev_cachep __read_mostly;
506
507static struct inode *bdev_alloc_inode(struct super_block *sb)
508{
509 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
510 if (!ei)
511 return NULL;
512 return &ei->vfs_inode;
513}
514
515static void bdev_i_callback(struct rcu_head *head)
516{
517 struct inode *inode = container_of(head, struct inode, i_rcu);
518 struct bdev_inode *bdi = BDEV_I(inode);
519
520 kmem_cache_free(bdev_cachep, bdi);
521}
522
523static void bdev_destroy_inode(struct inode *inode)
524{
525 call_rcu(&inode->i_rcu, bdev_i_callback);
526}
527
528static void init_once(void *foo)
529{
530 struct bdev_inode *ei = (struct bdev_inode *) foo;
531 struct block_device *bdev = &ei->bdev;
532
533 memset(bdev, 0, sizeof(*bdev));
534 mutex_init(&bdev->bd_mutex);
535 INIT_LIST_HEAD(&bdev->bd_inodes);
536 INIT_LIST_HEAD(&bdev->bd_list);
537#ifdef CONFIG_SYSFS
538 INIT_LIST_HEAD(&bdev->bd_holder_disks);
539#endif
540 inode_init_once(&ei->vfs_inode);
541 /* Initialize mutex for freeze. */
542 mutex_init(&bdev->bd_fsfreeze_mutex);
543}
544
545static inline void __bd_forget(struct inode *inode)
546{
547 list_del_init(&inode->i_devices);
548 inode->i_bdev = NULL;
549 inode->i_mapping = &inode->i_data;
550}
551
552static void bdev_evict_inode(struct inode *inode)
553{
554 struct block_device *bdev = &BDEV_I(inode)->bdev;
555 struct list_head *p;
556 truncate_inode_pages_final(&inode->i_data);
557 invalidate_inode_buffers(inode); /* is it needed here? */
558 clear_inode(inode);
559 spin_lock(&bdev_lock);
560 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
561 __bd_forget(list_entry(p, struct inode, i_devices));
562 }
563 list_del_init(&bdev->bd_list);
564 spin_unlock(&bdev_lock);
565}
566
567static const struct super_operations bdev_sops = {
568 .statfs = simple_statfs,
569 .alloc_inode = bdev_alloc_inode,
570 .destroy_inode = bdev_destroy_inode,
571 .drop_inode = generic_delete_inode,
572 .evict_inode = bdev_evict_inode,
573};
574
575static struct dentry *bd_mount(struct file_system_type *fs_type,
576 int flags, const char *dev_name, void *data)
577{
578 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
579}
580
581static struct file_system_type bd_type = {
582 .name = "bdev",
583 .mount = bd_mount,
584 .kill_sb = kill_anon_super,
585};
586
587struct super_block *blockdev_superblock __read_mostly;
588EXPORT_SYMBOL_GPL(blockdev_superblock);
589
590void __init bdev_cache_init(void)
591{
592 int err;
593 static struct vfsmount *bd_mnt;
594
595 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
596 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
597 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
598 init_once);
599 err = register_filesystem(&bd_type);
600 if (err)
601 panic("Cannot register bdev pseudo-fs");
602 bd_mnt = kern_mount(&bd_type);
603 if (IS_ERR(bd_mnt))
604 panic("Cannot create bdev pseudo-fs");
605 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
606}
607
608/*
609 * Most likely _very_ bad one - but then it's hardly critical for small
610 * /dev and can be fixed when somebody will need really large one.
611 * Keep in mind that it will be fed through icache hash function too.
612 */
613static inline unsigned long hash(dev_t dev)
614{
615 return MAJOR(dev)+MINOR(dev);
616}
617
618static int bdev_test(struct inode *inode, void *data)
619{
620 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
621}
622
623static int bdev_set(struct inode *inode, void *data)
624{
625 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
626 return 0;
627}
628
629static LIST_HEAD(all_bdevs);
630
631struct block_device *bdget(dev_t dev)
632{
633 struct block_device *bdev;
634 struct inode *inode;
635
636 inode = iget5_locked(blockdev_superblock, hash(dev),
637 bdev_test, bdev_set, &dev);
638
639 if (!inode)
640 return NULL;
641
642 bdev = &BDEV_I(inode)->bdev;
643
644 if (inode->i_state & I_NEW) {
645 bdev->bd_contains = NULL;
646 bdev->bd_super = NULL;
647 bdev->bd_inode = inode;
648 bdev->bd_block_size = (1 << inode->i_blkbits);
649 bdev->bd_part_count = 0;
650 bdev->bd_invalidated = 0;
651 inode->i_mode = S_IFBLK;
652 inode->i_rdev = dev;
653 inode->i_bdev = bdev;
654 inode->i_data.a_ops = &def_blk_aops;
655 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
656 spin_lock(&bdev_lock);
657 list_add(&bdev->bd_list, &all_bdevs);
658 spin_unlock(&bdev_lock);
659 unlock_new_inode(inode);
660 }
661 return bdev;
662}
663
664EXPORT_SYMBOL(bdget);
665
666/**
667 * bdgrab -- Grab a reference to an already referenced block device
668 * @bdev: Block device to grab a reference to.
669 */
670struct block_device *bdgrab(struct block_device *bdev)
671{
672 ihold(bdev->bd_inode);
673 return bdev;
674}
675EXPORT_SYMBOL(bdgrab);
676
677long nr_blockdev_pages(void)
678{
679 struct block_device *bdev;
680 long ret = 0;
681 spin_lock(&bdev_lock);
682 list_for_each_entry(bdev, &all_bdevs, bd_list) {
683 ret += bdev->bd_inode->i_mapping->nrpages;
684 }
685 spin_unlock(&bdev_lock);
686 return ret;
687}
688
689void bdput(struct block_device *bdev)
690{
691 iput(bdev->bd_inode);
692}
693
694EXPORT_SYMBOL(bdput);
695
696static struct block_device *bd_acquire(struct inode *inode)
697{
698 struct block_device *bdev;
699
700 spin_lock(&bdev_lock);
701 bdev = inode->i_bdev;
702 if (bdev) {
703 bdgrab(bdev);
704 spin_unlock(&bdev_lock);
705 return bdev;
706 }
707 spin_unlock(&bdev_lock);
708
709 bdev = bdget(inode->i_rdev);
710 if (bdev) {
711 spin_lock(&bdev_lock);
712 if (!inode->i_bdev) {
713 /*
714 * We take an additional reference to bd_inode,
715 * and it's released in clear_inode() of inode.
716 * So, we can access it via ->i_mapping always
717 * without igrab().
718 */
719 bdgrab(bdev);
720 inode->i_bdev = bdev;
721 inode->i_mapping = bdev->bd_inode->i_mapping;
722 list_add(&inode->i_devices, &bdev->bd_inodes);
723 }
724 spin_unlock(&bdev_lock);
725 }
726 return bdev;
727}
728
729/* Call when you free inode */
730
731void bd_forget(struct inode *inode)
732{
733 struct block_device *bdev = NULL;
734
735 spin_lock(&bdev_lock);
736 if (!sb_is_blkdev_sb(inode->i_sb))
737 bdev = inode->i_bdev;
738 __bd_forget(inode);
739 spin_unlock(&bdev_lock);
740
741 if (bdev)
742 bdput(bdev);
743}
744
745/**
746 * bd_may_claim - test whether a block device can be claimed
747 * @bdev: block device of interest
748 * @whole: whole block device containing @bdev, may equal @bdev
749 * @holder: holder trying to claim @bdev
750 *
751 * Test whether @bdev can be claimed by @holder.
752 *
753 * CONTEXT:
754 * spin_lock(&bdev_lock).
755 *
756 * RETURNS:
757 * %true if @bdev can be claimed, %false otherwise.
758 */
759static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
760 void *holder)
761{
762 if (bdev->bd_holder == holder)
763 return true; /* already a holder */
764 else if (bdev->bd_holder != NULL)
765 return false; /* held by someone else */
766 else if (bdev->bd_contains == bdev)
767 return true; /* is a whole device which isn't held */
768
769 else if (whole->bd_holder == bd_may_claim)
770 return true; /* is a partition of a device that is being partitioned */
771 else if (whole->bd_holder != NULL)
772 return false; /* is a partition of a held device */
773 else
774 return true; /* is a partition of an un-held device */
775}
776
777/**
778 * bd_prepare_to_claim - prepare to claim a block device
779 * @bdev: block device of interest
780 * @whole: the whole device containing @bdev, may equal @bdev
781 * @holder: holder trying to claim @bdev
782 *
783 * Prepare to claim @bdev. This function fails if @bdev is already
784 * claimed by another holder and waits if another claiming is in
785 * progress. This function doesn't actually claim. On successful
786 * return, the caller has ownership of bd_claiming and bd_holder[s].
787 *
788 * CONTEXT:
789 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
790 * it multiple times.
791 *
792 * RETURNS:
793 * 0 if @bdev can be claimed, -EBUSY otherwise.
794 */
795static int bd_prepare_to_claim(struct block_device *bdev,
796 struct block_device *whole, void *holder)
797{
798retry:
799 /* if someone else claimed, fail */
800 if (!bd_may_claim(bdev, whole, holder))
801 return -EBUSY;
802
803 /* if claiming is already in progress, wait for it to finish */
804 if (whole->bd_claiming) {
805 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
806 DEFINE_WAIT(wait);
807
808 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
809 spin_unlock(&bdev_lock);
810 schedule();
811 finish_wait(wq, &wait);
812 spin_lock(&bdev_lock);
813 goto retry;
814 }
815
816 /* yay, all mine */
817 return 0;
818}
819
820/**
821 * bd_start_claiming - start claiming a block device
822 * @bdev: block device of interest
823 * @holder: holder trying to claim @bdev
824 *
825 * @bdev is about to be opened exclusively. Check @bdev can be opened
826 * exclusively and mark that an exclusive open is in progress. Each
827 * successful call to this function must be matched with a call to
828 * either bd_finish_claiming() or bd_abort_claiming() (which do not
829 * fail).
830 *
831 * This function is used to gain exclusive access to the block device
832 * without actually causing other exclusive open attempts to fail. It
833 * should be used when the open sequence itself requires exclusive
834 * access but may subsequently fail.
835 *
836 * CONTEXT:
837 * Might sleep.
838 *
839 * RETURNS:
840 * Pointer to the block device containing @bdev on success, ERR_PTR()
841 * value on failure.
842 */
843static struct block_device *bd_start_claiming(struct block_device *bdev,
844 void *holder)
845{
846 struct gendisk *disk;
847 struct block_device *whole;
848 int partno, err;
849
850 might_sleep();
851
852 /*
853 * @bdev might not have been initialized properly yet, look up
854 * and grab the outer block device the hard way.
855 */
856 disk = get_gendisk(bdev->bd_dev, &partno);
857 if (!disk)
858 return ERR_PTR(-ENXIO);
859
860 /*
861 * Normally, @bdev should equal what's returned from bdget_disk()
862 * if partno is 0; however, some drivers (floppy) use multiple
863 * bdev's for the same physical device and @bdev may be one of the
864 * aliases. Keep @bdev if partno is 0. This means claimer
865 * tracking is broken for those devices but it has always been that
866 * way.
867 */
868 if (partno)
869 whole = bdget_disk(disk, 0);
870 else
871 whole = bdgrab(bdev);
872
873 module_put(disk->fops->owner);
874 put_disk(disk);
875 if (!whole)
876 return ERR_PTR(-ENOMEM);
877
878 /* prepare to claim, if successful, mark claiming in progress */
879 spin_lock(&bdev_lock);
880
881 err = bd_prepare_to_claim(bdev, whole, holder);
882 if (err == 0) {
883 whole->bd_claiming = holder;
884 spin_unlock(&bdev_lock);
885 return whole;
886 } else {
887 spin_unlock(&bdev_lock);
888 bdput(whole);
889 return ERR_PTR(err);
890 }
891}
892
893#ifdef CONFIG_SYSFS
894struct bd_holder_disk {
895 struct list_head list;
896 struct gendisk *disk;
897 int refcnt;
898};
899
900static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
901 struct gendisk *disk)
902{
903 struct bd_holder_disk *holder;
904
905 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
906 if (holder->disk == disk)
907 return holder;
908 return NULL;
909}
910
911static int add_symlink(struct kobject *from, struct kobject *to)
912{
913 return sysfs_create_link(from, to, kobject_name(to));
914}
915
916static void del_symlink(struct kobject *from, struct kobject *to)
917{
918 sysfs_remove_link(from, kobject_name(to));
919}
920
921/**
922 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
923 * @bdev: the claimed slave bdev
924 * @disk: the holding disk
925 *
926 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
927 *
928 * This functions creates the following sysfs symlinks.
929 *
930 * - from "slaves" directory of the holder @disk to the claimed @bdev
931 * - from "holders" directory of the @bdev to the holder @disk
932 *
933 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
934 * passed to bd_link_disk_holder(), then:
935 *
936 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
937 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
938 *
939 * The caller must have claimed @bdev before calling this function and
940 * ensure that both @bdev and @disk are valid during the creation and
941 * lifetime of these symlinks.
942 *
943 * CONTEXT:
944 * Might sleep.
945 *
946 * RETURNS:
947 * 0 on success, -errno on failure.
948 */
949int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
950{
951 struct bd_holder_disk *holder;
952 int ret = 0;
953
954 mutex_lock(&bdev->bd_mutex);
955
956 WARN_ON_ONCE(!bdev->bd_holder);
957
958 /* FIXME: remove the following once add_disk() handles errors */
959 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
960 goto out_unlock;
961
962 holder = bd_find_holder_disk(bdev, disk);
963 if (holder) {
964 holder->refcnt++;
965 goto out_unlock;
966 }
967
968 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
969 if (!holder) {
970 ret = -ENOMEM;
971 goto out_unlock;
972 }
973
974 INIT_LIST_HEAD(&holder->list);
975 holder->disk = disk;
976 holder->refcnt = 1;
977
978 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
979 if (ret)
980 goto out_free;
981
982 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
983 if (ret)
984 goto out_del;
985 /*
986 * bdev could be deleted beneath us which would implicitly destroy
987 * the holder directory. Hold on to it.
988 */
989 kobject_get(bdev->bd_part->holder_dir);
990
991 list_add(&holder->list, &bdev->bd_holder_disks);
992 goto out_unlock;
993
994out_del:
995 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
996out_free:
997 kfree(holder);
998out_unlock:
999 mutex_unlock(&bdev->bd_mutex);
1000 return ret;
1001}
1002EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1003
1004/**
1005 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1006 * @bdev: the calimed slave bdev
1007 * @disk: the holding disk
1008 *
1009 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1010 *
1011 * CONTEXT:
1012 * Might sleep.
1013 */
1014void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1015{
1016 struct bd_holder_disk *holder;
1017
1018 mutex_lock(&bdev->bd_mutex);
1019
1020 holder = bd_find_holder_disk(bdev, disk);
1021
1022 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1023 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1024 del_symlink(bdev->bd_part->holder_dir,
1025 &disk_to_dev(disk)->kobj);
1026 kobject_put(bdev->bd_part->holder_dir);
1027 list_del_init(&holder->list);
1028 kfree(holder);
1029 }
1030
1031 mutex_unlock(&bdev->bd_mutex);
1032}
1033EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1034#endif
1035
1036/**
1037 * flush_disk - invalidates all buffer-cache entries on a disk
1038 *
1039 * @bdev: struct block device to be flushed
1040 * @kill_dirty: flag to guide handling of dirty inodes
1041 *
1042 * Invalidates all buffer-cache entries on a disk. It should be called
1043 * when a disk has been changed -- either by a media change or online
1044 * resize.
1045 */
1046static void flush_disk(struct block_device *bdev, bool kill_dirty)
1047{
1048 if (__invalidate_device(bdev, kill_dirty)) {
1049 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1050 "resized disk %s\n",
1051 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1052 }
1053
1054 if (!bdev->bd_disk)
1055 return;
1056 if (disk_part_scan_enabled(bdev->bd_disk))
1057 bdev->bd_invalidated = 1;
1058}
1059
1060/**
1061 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1062 * @disk: struct gendisk to check
1063 * @bdev: struct bdev to adjust.
1064 *
1065 * This routine checks to see if the bdev size does not match the disk size
1066 * and adjusts it if it differs.
1067 */
1068void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1069{
1070 loff_t disk_size, bdev_size;
1071
1072 disk_size = (loff_t)get_capacity(disk) << 9;
1073 bdev_size = i_size_read(bdev->bd_inode);
1074 if (disk_size != bdev_size) {
1075 printk(KERN_INFO
1076 "%s: detected capacity change from %lld to %lld\n",
1077 disk->disk_name, bdev_size, disk_size);
1078 i_size_write(bdev->bd_inode, disk_size);
1079 flush_disk(bdev, false);
1080 }
1081}
1082EXPORT_SYMBOL(check_disk_size_change);
1083
1084/**
1085 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1086 * @disk: struct gendisk to be revalidated
1087 *
1088 * This routine is a wrapper for lower-level driver's revalidate_disk
1089 * call-backs. It is used to do common pre and post operations needed
1090 * for all revalidate_disk operations.
1091 */
1092int revalidate_disk(struct gendisk *disk)
1093{
1094 struct block_device *bdev;
1095 int ret = 0;
1096
1097 if (disk->fops->revalidate_disk)
1098 ret = disk->fops->revalidate_disk(disk);
1099 blk_integrity_revalidate(disk);
1100 bdev = bdget_disk(disk, 0);
1101 if (!bdev)
1102 return ret;
1103
1104 mutex_lock(&bdev->bd_mutex);
1105 check_disk_size_change(disk, bdev);
1106 bdev->bd_invalidated = 0;
1107 mutex_unlock(&bdev->bd_mutex);
1108 bdput(bdev);
1109 return ret;
1110}
1111EXPORT_SYMBOL(revalidate_disk);
1112
1113/*
1114 * This routine checks whether a removable media has been changed,
1115 * and invalidates all buffer-cache-entries in that case. This
1116 * is a relatively slow routine, so we have to try to minimize using
1117 * it. Thus it is called only upon a 'mount' or 'open'. This
1118 * is the best way of combining speed and utility, I think.
1119 * People changing diskettes in the middle of an operation deserve
1120 * to lose :-)
1121 */
1122int check_disk_change(struct block_device *bdev)
1123{
1124 struct gendisk *disk = bdev->bd_disk;
1125 const struct block_device_operations *bdops = disk->fops;
1126 unsigned int events;
1127
1128 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1129 DISK_EVENT_EJECT_REQUEST);
1130 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1131 return 0;
1132
1133 flush_disk(bdev, true);
1134 if (bdops->revalidate_disk)
1135 bdops->revalidate_disk(bdev->bd_disk);
1136 return 1;
1137}
1138
1139EXPORT_SYMBOL(check_disk_change);
1140
1141void bd_set_size(struct block_device *bdev, loff_t size)
1142{
1143 unsigned bsize = bdev_logical_block_size(bdev);
1144
1145 inode_lock(bdev->bd_inode);
1146 i_size_write(bdev->bd_inode, size);
1147 inode_unlock(bdev->bd_inode);
1148 while (bsize < PAGE_CACHE_SIZE) {
1149 if (size & bsize)
1150 break;
1151 bsize <<= 1;
1152 }
1153 bdev->bd_block_size = bsize;
1154 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1155}
1156EXPORT_SYMBOL(bd_set_size);
1157
1158static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1159
1160/*
1161 * bd_mutex locking:
1162 *
1163 * mutex_lock(part->bd_mutex)
1164 * mutex_lock_nested(whole->bd_mutex, 1)
1165 */
1166
1167static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1168{
1169 struct gendisk *disk;
1170 struct module *owner;
1171 int ret;
1172 int partno;
1173 int perm = 0;
1174
1175 if (mode & FMODE_READ)
1176 perm |= MAY_READ;
1177 if (mode & FMODE_WRITE)
1178 perm |= MAY_WRITE;
1179 /*
1180 * hooks: /n/, see "layering violations".
1181 */
1182 if (!for_part) {
1183 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1184 if (ret != 0) {
1185 bdput(bdev);
1186 return ret;
1187 }
1188 }
1189
1190 restart:
1191
1192 ret = -ENXIO;
1193 disk = get_gendisk(bdev->bd_dev, &partno);
1194 if (!disk)
1195 goto out;
1196 owner = disk->fops->owner;
1197
1198 disk_block_events(disk);
1199 mutex_lock_nested(&bdev->bd_mutex, for_part);
1200 if (!bdev->bd_openers) {
1201 bdev->bd_disk = disk;
1202 bdev->bd_queue = disk->queue;
1203 bdev->bd_contains = bdev;
1204 bdev->bd_inode->i_flags = disk->fops->direct_access ? S_DAX : 0;
1205 if (!partno) {
1206 ret = -ENXIO;
1207 bdev->bd_part = disk_get_part(disk, partno);
1208 if (!bdev->bd_part)
1209 goto out_clear;
1210
1211 ret = 0;
1212 if (disk->fops->open) {
1213 ret = disk->fops->open(bdev, mode);
1214 if (ret == -ERESTARTSYS) {
1215 /* Lost a race with 'disk' being
1216 * deleted, try again.
1217 * See md.c
1218 */
1219 disk_put_part(bdev->bd_part);
1220 bdev->bd_part = NULL;
1221 bdev->bd_disk = NULL;
1222 bdev->bd_queue = NULL;
1223 mutex_unlock(&bdev->bd_mutex);
1224 disk_unblock_events(disk);
1225 put_disk(disk);
1226 module_put(owner);
1227 goto restart;
1228 }
1229 }
1230
1231 if (!ret) {
1232 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1233 if (!blkdev_dax_capable(bdev))
1234 bdev->bd_inode->i_flags &= ~S_DAX;
1235 }
1236
1237 /*
1238 * If the device is invalidated, rescan partition
1239 * if open succeeded or failed with -ENOMEDIUM.
1240 * The latter is necessary to prevent ghost
1241 * partitions on a removed medium.
1242 */
1243 if (bdev->bd_invalidated) {
1244 if (!ret)
1245 rescan_partitions(disk, bdev);
1246 else if (ret == -ENOMEDIUM)
1247 invalidate_partitions(disk, bdev);
1248 }
1249
1250 if (ret)
1251 goto out_clear;
1252 } else {
1253 struct block_device *whole;
1254 whole = bdget_disk(disk, 0);
1255 ret = -ENOMEM;
1256 if (!whole)
1257 goto out_clear;
1258 BUG_ON(for_part);
1259 ret = __blkdev_get(whole, mode, 1);
1260 if (ret)
1261 goto out_clear;
1262 bdev->bd_contains = whole;
1263 bdev->bd_part = disk_get_part(disk, partno);
1264 if (!(disk->flags & GENHD_FL_UP) ||
1265 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1266 ret = -ENXIO;
1267 goto out_clear;
1268 }
1269 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1270 if (!blkdev_dax_capable(bdev))
1271 bdev->bd_inode->i_flags &= ~S_DAX;
1272 }
1273 } else {
1274 if (bdev->bd_contains == bdev) {
1275 ret = 0;
1276 if (bdev->bd_disk->fops->open)
1277 ret = bdev->bd_disk->fops->open(bdev, mode);
1278 /* the same as first opener case, read comment there */
1279 if (bdev->bd_invalidated) {
1280 if (!ret)
1281 rescan_partitions(bdev->bd_disk, bdev);
1282 else if (ret == -ENOMEDIUM)
1283 invalidate_partitions(bdev->bd_disk, bdev);
1284 }
1285 if (ret)
1286 goto out_unlock_bdev;
1287 }
1288 /* only one opener holds refs to the module and disk */
1289 put_disk(disk);
1290 module_put(owner);
1291 }
1292 bdev->bd_openers++;
1293 if (for_part)
1294 bdev->bd_part_count++;
1295 mutex_unlock(&bdev->bd_mutex);
1296 disk_unblock_events(disk);
1297 return 0;
1298
1299 out_clear:
1300 disk_put_part(bdev->bd_part);
1301 bdev->bd_disk = NULL;
1302 bdev->bd_part = NULL;
1303 bdev->bd_queue = NULL;
1304 if (bdev != bdev->bd_contains)
1305 __blkdev_put(bdev->bd_contains, mode, 1);
1306 bdev->bd_contains = NULL;
1307 out_unlock_bdev:
1308 mutex_unlock(&bdev->bd_mutex);
1309 disk_unblock_events(disk);
1310 put_disk(disk);
1311 module_put(owner);
1312 out:
1313 bdput(bdev);
1314
1315 return ret;
1316}
1317
1318/**
1319 * blkdev_get - open a block device
1320 * @bdev: block_device to open
1321 * @mode: FMODE_* mask
1322 * @holder: exclusive holder identifier
1323 *
1324 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1325 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1326 * @holder is invalid. Exclusive opens may nest for the same @holder.
1327 *
1328 * On success, the reference count of @bdev is unchanged. On failure,
1329 * @bdev is put.
1330 *
1331 * CONTEXT:
1332 * Might sleep.
1333 *
1334 * RETURNS:
1335 * 0 on success, -errno on failure.
1336 */
1337int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1338{
1339 struct block_device *whole = NULL;
1340 int res;
1341
1342 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1343
1344 if ((mode & FMODE_EXCL) && holder) {
1345 whole = bd_start_claiming(bdev, holder);
1346 if (IS_ERR(whole)) {
1347 bdput(bdev);
1348 return PTR_ERR(whole);
1349 }
1350 }
1351
1352 res = __blkdev_get(bdev, mode, 0);
1353
1354 if (whole) {
1355 struct gendisk *disk = whole->bd_disk;
1356
1357 /* finish claiming */
1358 mutex_lock(&bdev->bd_mutex);
1359 spin_lock(&bdev_lock);
1360
1361 if (!res) {
1362 BUG_ON(!bd_may_claim(bdev, whole, holder));
1363 /*
1364 * Note that for a whole device bd_holders
1365 * will be incremented twice, and bd_holder
1366 * will be set to bd_may_claim before being
1367 * set to holder
1368 */
1369 whole->bd_holders++;
1370 whole->bd_holder = bd_may_claim;
1371 bdev->bd_holders++;
1372 bdev->bd_holder = holder;
1373 }
1374
1375 /* tell others that we're done */
1376 BUG_ON(whole->bd_claiming != holder);
1377 whole->bd_claiming = NULL;
1378 wake_up_bit(&whole->bd_claiming, 0);
1379
1380 spin_unlock(&bdev_lock);
1381
1382 /*
1383 * Block event polling for write claims if requested. Any
1384 * write holder makes the write_holder state stick until
1385 * all are released. This is good enough and tracking
1386 * individual writeable reference is too fragile given the
1387 * way @mode is used in blkdev_get/put().
1388 */
1389 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1390 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1391 bdev->bd_write_holder = true;
1392 disk_block_events(disk);
1393 }
1394
1395 mutex_unlock(&bdev->bd_mutex);
1396 bdput(whole);
1397 }
1398
1399 return res;
1400}
1401EXPORT_SYMBOL(blkdev_get);
1402
1403/**
1404 * blkdev_get_by_path - open a block device by name
1405 * @path: path to the block device to open
1406 * @mode: FMODE_* mask
1407 * @holder: exclusive holder identifier
1408 *
1409 * Open the blockdevice described by the device file at @path. @mode
1410 * and @holder are identical to blkdev_get().
1411 *
1412 * On success, the returned block_device has reference count of one.
1413 *
1414 * CONTEXT:
1415 * Might sleep.
1416 *
1417 * RETURNS:
1418 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1419 */
1420struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1421 void *holder)
1422{
1423 struct block_device *bdev;
1424 int err;
1425
1426 bdev = lookup_bdev(path);
1427 if (IS_ERR(bdev))
1428 return bdev;
1429
1430 err = blkdev_get(bdev, mode, holder);
1431 if (err)
1432 return ERR_PTR(err);
1433
1434 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1435 blkdev_put(bdev, mode);
1436 return ERR_PTR(-EACCES);
1437 }
1438
1439 return bdev;
1440}
1441EXPORT_SYMBOL(blkdev_get_by_path);
1442
1443/**
1444 * blkdev_get_by_dev - open a block device by device number
1445 * @dev: device number of block device to open
1446 * @mode: FMODE_* mask
1447 * @holder: exclusive holder identifier
1448 *
1449 * Open the blockdevice described by device number @dev. @mode and
1450 * @holder are identical to blkdev_get().
1451 *
1452 * Use it ONLY if you really do not have anything better - i.e. when
1453 * you are behind a truly sucky interface and all you are given is a
1454 * device number. _Never_ to be used for internal purposes. If you
1455 * ever need it - reconsider your API.
1456 *
1457 * On success, the returned block_device has reference count of one.
1458 *
1459 * CONTEXT:
1460 * Might sleep.
1461 *
1462 * RETURNS:
1463 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1464 */
1465struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1466{
1467 struct block_device *bdev;
1468 int err;
1469
1470 bdev = bdget(dev);
1471 if (!bdev)
1472 return ERR_PTR(-ENOMEM);
1473
1474 err = blkdev_get(bdev, mode, holder);
1475 if (err)
1476 return ERR_PTR(err);
1477
1478 return bdev;
1479}
1480EXPORT_SYMBOL(blkdev_get_by_dev);
1481
1482static int blkdev_open(struct inode * inode, struct file * filp)
1483{
1484 struct block_device *bdev;
1485
1486 /*
1487 * Preserve backwards compatibility and allow large file access
1488 * even if userspace doesn't ask for it explicitly. Some mkfs
1489 * binary needs it. We might want to drop this workaround
1490 * during an unstable branch.
1491 */
1492 filp->f_flags |= O_LARGEFILE;
1493
1494 if (filp->f_flags & O_NDELAY)
1495 filp->f_mode |= FMODE_NDELAY;
1496 if (filp->f_flags & O_EXCL)
1497 filp->f_mode |= FMODE_EXCL;
1498 if ((filp->f_flags & O_ACCMODE) == 3)
1499 filp->f_mode |= FMODE_WRITE_IOCTL;
1500
1501 bdev = bd_acquire(inode);
1502 if (bdev == NULL)
1503 return -ENOMEM;
1504
1505 filp->f_mapping = bdev->bd_inode->i_mapping;
1506
1507 return blkdev_get(bdev, filp->f_mode, filp);
1508}
1509
1510static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1511{
1512 struct gendisk *disk = bdev->bd_disk;
1513 struct block_device *victim = NULL;
1514
1515 mutex_lock_nested(&bdev->bd_mutex, for_part);
1516 if (for_part)
1517 bdev->bd_part_count--;
1518
1519 if (!--bdev->bd_openers) {
1520 WARN_ON_ONCE(bdev->bd_holders);
1521 sync_blockdev(bdev);
1522 kill_bdev(bdev);
1523
1524 bdev_write_inode(bdev);
1525 /*
1526 * Detaching bdev inode from its wb in __destroy_inode()
1527 * is too late: the queue which embeds its bdi (along with
1528 * root wb) can be gone as soon as we put_disk() below.
1529 */
1530 inode_detach_wb(bdev->bd_inode);
1531 }
1532 if (bdev->bd_contains == bdev) {
1533 if (disk->fops->release)
1534 disk->fops->release(disk, mode);
1535 }
1536 if (!bdev->bd_openers) {
1537 struct module *owner = disk->fops->owner;
1538
1539 disk_put_part(bdev->bd_part);
1540 bdev->bd_part = NULL;
1541 bdev->bd_disk = NULL;
1542 if (bdev != bdev->bd_contains)
1543 victim = bdev->bd_contains;
1544 bdev->bd_contains = NULL;
1545
1546 put_disk(disk);
1547 module_put(owner);
1548 }
1549 mutex_unlock(&bdev->bd_mutex);
1550 bdput(bdev);
1551 if (victim)
1552 __blkdev_put(victim, mode, 1);
1553}
1554
1555void blkdev_put(struct block_device *bdev, fmode_t mode)
1556{
1557 mutex_lock(&bdev->bd_mutex);
1558
1559 if (mode & FMODE_EXCL) {
1560 bool bdev_free;
1561
1562 /*
1563 * Release a claim on the device. The holder fields
1564 * are protected with bdev_lock. bd_mutex is to
1565 * synchronize disk_holder unlinking.
1566 */
1567 spin_lock(&bdev_lock);
1568
1569 WARN_ON_ONCE(--bdev->bd_holders < 0);
1570 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1571
1572 /* bd_contains might point to self, check in a separate step */
1573 if ((bdev_free = !bdev->bd_holders))
1574 bdev->bd_holder = NULL;
1575 if (!bdev->bd_contains->bd_holders)
1576 bdev->bd_contains->bd_holder = NULL;
1577
1578 spin_unlock(&bdev_lock);
1579
1580 /*
1581 * If this was the last claim, remove holder link and
1582 * unblock evpoll if it was a write holder.
1583 */
1584 if (bdev_free && bdev->bd_write_holder) {
1585 disk_unblock_events(bdev->bd_disk);
1586 bdev->bd_write_holder = false;
1587 }
1588 }
1589
1590 /*
1591 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1592 * event. This is to ensure detection of media removal commanded
1593 * from userland - e.g. eject(1).
1594 */
1595 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1596
1597 mutex_unlock(&bdev->bd_mutex);
1598
1599 __blkdev_put(bdev, mode, 0);
1600}
1601EXPORT_SYMBOL(blkdev_put);
1602
1603static int blkdev_close(struct inode * inode, struct file * filp)
1604{
1605 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1606 blkdev_put(bdev, filp->f_mode);
1607 return 0;
1608}
1609
1610static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1611{
1612 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1613 fmode_t mode = file->f_mode;
1614
1615 /*
1616 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1617 * to updated it before every ioctl.
1618 */
1619 if (file->f_flags & O_NDELAY)
1620 mode |= FMODE_NDELAY;
1621 else
1622 mode &= ~FMODE_NDELAY;
1623
1624 return blkdev_ioctl(bdev, mode, cmd, arg);
1625}
1626
1627/*
1628 * Write data to the block device. Only intended for the block device itself
1629 * and the raw driver which basically is a fake block device.
1630 *
1631 * Does not take i_mutex for the write and thus is not for general purpose
1632 * use.
1633 */
1634ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1635{
1636 struct file *file = iocb->ki_filp;
1637 struct inode *bd_inode = bdev_file_inode(file);
1638 loff_t size = i_size_read(bd_inode);
1639 struct blk_plug plug;
1640 ssize_t ret;
1641
1642 if (bdev_read_only(I_BDEV(bd_inode)))
1643 return -EPERM;
1644
1645 if (!iov_iter_count(from))
1646 return 0;
1647
1648 if (iocb->ki_pos >= size)
1649 return -ENOSPC;
1650
1651 iov_iter_truncate(from, size - iocb->ki_pos);
1652
1653 blk_start_plug(&plug);
1654 ret = __generic_file_write_iter(iocb, from);
1655 if (ret > 0) {
1656 ssize_t err;
1657 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1658 if (err < 0)
1659 ret = err;
1660 }
1661 blk_finish_plug(&plug);
1662 return ret;
1663}
1664EXPORT_SYMBOL_GPL(blkdev_write_iter);
1665
1666ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1667{
1668 struct file *file = iocb->ki_filp;
1669 struct inode *bd_inode = bdev_file_inode(file);
1670 loff_t size = i_size_read(bd_inode);
1671 loff_t pos = iocb->ki_pos;
1672
1673 if (pos >= size)
1674 return 0;
1675
1676 size -= pos;
1677 iov_iter_truncate(to, size);
1678 return generic_file_read_iter(iocb, to);
1679}
1680EXPORT_SYMBOL_GPL(blkdev_read_iter);
1681
1682/*
1683 * Try to release a page associated with block device when the system
1684 * is under memory pressure.
1685 */
1686static int blkdev_releasepage(struct page *page, gfp_t wait)
1687{
1688 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1689
1690 if (super && super->s_op->bdev_try_to_free_page)
1691 return super->s_op->bdev_try_to_free_page(super, page, wait);
1692
1693 return try_to_free_buffers(page);
1694}
1695
1696static const struct address_space_operations def_blk_aops = {
1697 .readpage = blkdev_readpage,
1698 .readpages = blkdev_readpages,
1699 .writepage = blkdev_writepage,
1700 .write_begin = blkdev_write_begin,
1701 .write_end = blkdev_write_end,
1702 .writepages = generic_writepages,
1703 .releasepage = blkdev_releasepage,
1704 .direct_IO = blkdev_direct_IO,
1705 .is_dirty_writeback = buffer_check_dirty_writeback,
1706};
1707
1708#ifdef CONFIG_FS_DAX
1709/*
1710 * In the raw block case we do not need to contend with truncation nor
1711 * unwritten file extents. Without those concerns there is no need for
1712 * additional locking beyond the mmap_sem context that these routines
1713 * are already executing under.
1714 *
1715 * Note, there is no protection if the block device is dynamically
1716 * resized (partition grow/shrink) during a fault. A stable block device
1717 * size is already not enforced in the blkdev_direct_IO path.
1718 *
1719 * For DAX, it is the responsibility of the block device driver to
1720 * ensure the whole-disk device size is stable while requests are in
1721 * flight.
1722 *
1723 * Finally, unlike the filemap_page_mkwrite() case there is no
1724 * filesystem superblock to sync against freezing. We still include a
1725 * pfn_mkwrite callback for dax drivers to receive write fault
1726 * notifications.
1727 */
1728static int blkdev_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1729{
1730 return __dax_fault(vma, vmf, blkdev_get_block, NULL);
1731}
1732
1733static int blkdev_dax_pfn_mkwrite(struct vm_area_struct *vma,
1734 struct vm_fault *vmf)
1735{
1736 return dax_pfn_mkwrite(vma, vmf);
1737}
1738
1739static int blkdev_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
1740 pmd_t *pmd, unsigned int flags)
1741{
1742 return __dax_pmd_fault(vma, addr, pmd, flags, blkdev_get_block, NULL);
1743}
1744
1745static const struct vm_operations_struct blkdev_dax_vm_ops = {
1746 .fault = blkdev_dax_fault,
1747 .pmd_fault = blkdev_dax_pmd_fault,
1748 .pfn_mkwrite = blkdev_dax_pfn_mkwrite,
1749};
1750
1751static const struct vm_operations_struct blkdev_default_vm_ops = {
1752 .fault = filemap_fault,
1753 .map_pages = filemap_map_pages,
1754};
1755
1756static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
1757{
1758 struct inode *bd_inode = bdev_file_inode(file);
1759
1760 file_accessed(file);
1761 if (IS_DAX(bd_inode)) {
1762 vma->vm_ops = &blkdev_dax_vm_ops;
1763 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1764 } else {
1765 vma->vm_ops = &blkdev_default_vm_ops;
1766 }
1767
1768 return 0;
1769}
1770#else
1771#define blkdev_mmap generic_file_mmap
1772#endif
1773
1774const struct file_operations def_blk_fops = {
1775 .open = blkdev_open,
1776 .release = blkdev_close,
1777 .llseek = block_llseek,
1778 .read_iter = blkdev_read_iter,
1779 .write_iter = blkdev_write_iter,
1780 .mmap = blkdev_mmap,
1781 .fsync = blkdev_fsync,
1782 .unlocked_ioctl = block_ioctl,
1783#ifdef CONFIG_COMPAT
1784 .compat_ioctl = compat_blkdev_ioctl,
1785#endif
1786 .splice_read = generic_file_splice_read,
1787 .splice_write = iter_file_splice_write,
1788};
1789
1790int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1791{
1792 int res;
1793 mm_segment_t old_fs = get_fs();
1794 set_fs(KERNEL_DS);
1795 res = blkdev_ioctl(bdev, 0, cmd, arg);
1796 set_fs(old_fs);
1797 return res;
1798}
1799
1800EXPORT_SYMBOL(ioctl_by_bdev);
1801
1802/**
1803 * lookup_bdev - lookup a struct block_device by name
1804 * @pathname: special file representing the block device
1805 *
1806 * Get a reference to the blockdevice at @pathname in the current
1807 * namespace if possible and return it. Return ERR_PTR(error)
1808 * otherwise.
1809 */
1810struct block_device *lookup_bdev(const char *pathname)
1811{
1812 struct block_device *bdev;
1813 struct inode *inode;
1814 struct path path;
1815 int error;
1816
1817 if (!pathname || !*pathname)
1818 return ERR_PTR(-EINVAL);
1819
1820 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1821 if (error)
1822 return ERR_PTR(error);
1823
1824 inode = d_backing_inode(path.dentry);
1825 error = -ENOTBLK;
1826 if (!S_ISBLK(inode->i_mode))
1827 goto fail;
1828 error = -EACCES;
1829 if (path.mnt->mnt_flags & MNT_NODEV)
1830 goto fail;
1831 error = -ENOMEM;
1832 bdev = bd_acquire(inode);
1833 if (!bdev)
1834 goto fail;
1835out:
1836 path_put(&path);
1837 return bdev;
1838fail:
1839 bdev = ERR_PTR(error);
1840 goto out;
1841}
1842EXPORT_SYMBOL(lookup_bdev);
1843
1844int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1845{
1846 struct super_block *sb = get_super(bdev);
1847 int res = 0;
1848
1849 if (sb) {
1850 /*
1851 * no need to lock the super, get_super holds the
1852 * read mutex so the filesystem cannot go away
1853 * under us (->put_super runs with the write lock
1854 * hold).
1855 */
1856 shrink_dcache_sb(sb);
1857 res = invalidate_inodes(sb, kill_dirty);
1858 drop_super(sb);
1859 }
1860 invalidate_bdev(bdev);
1861 return res;
1862}
1863EXPORT_SYMBOL(__invalidate_device);
1864
1865void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1866{
1867 struct inode *inode, *old_inode = NULL;
1868
1869 spin_lock(&blockdev_superblock->s_inode_list_lock);
1870 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1871 struct address_space *mapping = inode->i_mapping;
1872
1873 spin_lock(&inode->i_lock);
1874 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1875 mapping->nrpages == 0) {
1876 spin_unlock(&inode->i_lock);
1877 continue;
1878 }
1879 __iget(inode);
1880 spin_unlock(&inode->i_lock);
1881 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1882 /*
1883 * We hold a reference to 'inode' so it couldn't have been
1884 * removed from s_inodes list while we dropped the
1885 * s_inode_list_lock We cannot iput the inode now as we can
1886 * be holding the last reference and we cannot iput it under
1887 * s_inode_list_lock. So we keep the reference and iput it
1888 * later.
1889 */
1890 iput(old_inode);
1891 old_inode = inode;
1892
1893 func(I_BDEV(inode), arg);
1894
1895 spin_lock(&blockdev_superblock->s_inode_list_lock);
1896 }
1897 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1898 iput(old_inode);
1899}