summaryrefslogtreecommitdiff
path: root/fs/block_dev.c (plain)
blob: f36aad28898ab4d4bb4d799e94562fbcf655e755
1/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
6 */
7
8#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fcntl.h>
11#include <linux/slab.h>
12#include <linux/kmod.h>
13#include <linux/major.h>
14#include <linux/device_cgroup.h>
15#include <linux/highmem.h>
16#include <linux/blkdev.h>
17#include <linux/backing-dev.h>
18#include <linux/module.h>
19#include <linux/blkpg.h>
20#include <linux/magic.h>
21#include <linux/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 <linux/badblocks.h>
33#include <linux/falloc.h>
34#include <asm/uaccess.h>
35#include "internal.h"
36
37struct bdev_inode {
38 struct block_device bdev;
39 struct inode vfs_inode;
40};
41
42static const struct address_space_operations def_blk_aops;
43
44static inline struct bdev_inode *BDEV_I(struct inode *inode)
45{
46 return container_of(inode, struct bdev_inode, vfs_inode);
47}
48
49struct block_device *I_BDEV(struct inode *inode)
50{
51 return &BDEV_I(inode)->bdev;
52}
53EXPORT_SYMBOL(I_BDEV);
54
55void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
56{
57 struct va_format vaf;
58 va_list args;
59
60 va_start(args, fmt);
61 vaf.fmt = fmt;
62 vaf.va = &args;
63 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
64 va_end(args);
65}
66
67static void bdev_write_inode(struct block_device *bdev)
68{
69 struct inode *inode = bdev->bd_inode;
70 int ret;
71
72 spin_lock(&inode->i_lock);
73 while (inode->i_state & I_DIRTY) {
74 spin_unlock(&inode->i_lock);
75 ret = write_inode_now(inode, true);
76 if (ret) {
77 char name[BDEVNAME_SIZE];
78 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
79 "for block device %s (err=%d).\n",
80 bdevname(bdev, name), ret);
81 }
82 spin_lock(&inode->i_lock);
83 }
84 spin_unlock(&inode->i_lock);
85}
86
87/* Kill _all_ buffers and pagecache , dirty or not.. */
88void kill_bdev(struct block_device *bdev)
89{
90 struct address_space *mapping = bdev->bd_inode->i_mapping;
91
92 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
93 return;
94
95 invalidate_bh_lrus();
96 truncate_inode_pages(mapping, 0);
97}
98EXPORT_SYMBOL(kill_bdev);
99
100/* Invalidate clean unused buffers and pagecache. */
101void invalidate_bdev(struct block_device *bdev)
102{
103 struct address_space *mapping = bdev->bd_inode->i_mapping;
104
105 if (mapping->nrpages) {
106 invalidate_bh_lrus();
107 lru_add_drain_all(); /* make sure all lru add caches are flushed */
108 invalidate_mapping_pages(mapping, 0, -1);
109 }
110 /* 99% of the time, we don't need to flush the cleancache on the bdev.
111 * But, for the strange corners, lets be cautious
112 */
113 cleancache_invalidate_inode(mapping);
114}
115EXPORT_SYMBOL(invalidate_bdev);
116
117static void set_init_blocksize(struct block_device *bdev)
118{
119 unsigned bsize = bdev_logical_block_size(bdev);
120 loff_t size = i_size_read(bdev->bd_inode);
121
122 while (bsize < PAGE_SIZE) {
123 if (size & bsize)
124 break;
125 bsize <<= 1;
126 }
127 bdev->bd_block_size = bsize;
128 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
129}
130
131int set_blocksize(struct block_device *bdev, int size)
132{
133 /* Size must be a power of two, and between 512 and PAGE_SIZE */
134 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
135 return -EINVAL;
136
137 /* Size cannot be smaller than the size supported by the device */
138 if (size < bdev_logical_block_size(bdev))
139 return -EINVAL;
140
141 /* Don't change the size if it is same as current */
142 if (bdev->bd_block_size != size) {
143 sync_blockdev(bdev);
144 bdev->bd_block_size = size;
145 bdev->bd_inode->i_blkbits = blksize_bits(size);
146 kill_bdev(bdev);
147 }
148 return 0;
149}
150
151EXPORT_SYMBOL(set_blocksize);
152
153int sb_set_blocksize(struct super_block *sb, int size)
154{
155 if (set_blocksize(sb->s_bdev, size))
156 return 0;
157 /* If we get here, we know size is power of two
158 * and it's value is between 512 and PAGE_SIZE */
159 sb->s_blocksize = size;
160 sb->s_blocksize_bits = blksize_bits(size);
161 return sb->s_blocksize;
162}
163
164EXPORT_SYMBOL(sb_set_blocksize);
165
166int sb_min_blocksize(struct super_block *sb, int size)
167{
168 int minsize = bdev_logical_block_size(sb->s_bdev);
169 if (size < minsize)
170 size = minsize;
171 return sb_set_blocksize(sb, size);
172}
173
174EXPORT_SYMBOL(sb_min_blocksize);
175
176static int
177blkdev_get_block(struct inode *inode, sector_t iblock,
178 struct buffer_head *bh, int create)
179{
180 bh->b_bdev = I_BDEV(inode);
181 bh->b_blocknr = iblock;
182 set_buffer_mapped(bh);
183 return 0;
184}
185
186static struct inode *bdev_file_inode(struct file *file)
187{
188 return file->f_mapping->host;
189}
190
191static ssize_t
192blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
193{
194 struct file *file = iocb->ki_filp;
195 struct inode *inode = bdev_file_inode(file);
196
197 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter,
198 blkdev_get_block, NULL, NULL,
199 DIO_SKIP_DIO_COUNT);
200}
201
202int __sync_blockdev(struct block_device *bdev, int wait)
203{
204 if (!bdev)
205 return 0;
206 if (!wait)
207 return filemap_flush(bdev->bd_inode->i_mapping);
208 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
209}
210
211/*
212 * Write out and wait upon all the dirty data associated with a block
213 * device via its mapping. Does not take the superblock lock.
214 */
215int sync_blockdev(struct block_device *bdev)
216{
217 return __sync_blockdev(bdev, 1);
218}
219EXPORT_SYMBOL(sync_blockdev);
220
221/*
222 * Write out and wait upon all dirty data associated with this
223 * device. Filesystem data as well as the underlying block
224 * device. Takes the superblock lock.
225 */
226int fsync_bdev(struct block_device *bdev)
227{
228 struct super_block *sb = get_super(bdev);
229 if (sb) {
230 int res = sync_filesystem(sb);
231 drop_super(sb);
232 return res;
233 }
234 return sync_blockdev(bdev);
235}
236EXPORT_SYMBOL(fsync_bdev);
237
238/**
239 * freeze_bdev -- lock a filesystem and force it into a consistent state
240 * @bdev: blockdevice to lock
241 *
242 * If a superblock is found on this device, we take the s_umount semaphore
243 * on it to make sure nobody unmounts until the snapshot creation is done.
244 * The reference counter (bd_fsfreeze_count) guarantees that only the last
245 * unfreeze process can unfreeze the frozen filesystem actually when multiple
246 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
247 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
248 * actually.
249 */
250struct super_block *freeze_bdev(struct block_device *bdev)
251{
252 struct super_block *sb;
253 int error = 0;
254
255 mutex_lock(&bdev->bd_fsfreeze_mutex);
256 if (++bdev->bd_fsfreeze_count > 1) {
257 /*
258 * We don't even need to grab a reference - the first call
259 * to freeze_bdev grab an active reference and only the last
260 * thaw_bdev drops it.
261 */
262 sb = get_super(bdev);
263 if (sb)
264 drop_super(sb);
265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
266 return sb;
267 }
268
269 sb = get_active_super(bdev);
270 if (!sb)
271 goto out;
272 if (sb->s_op->freeze_super)
273 error = sb->s_op->freeze_super(sb);
274 else
275 error = freeze_super(sb);
276 if (error) {
277 deactivate_super(sb);
278 bdev->bd_fsfreeze_count--;
279 mutex_unlock(&bdev->bd_fsfreeze_mutex);
280 return ERR_PTR(error);
281 }
282 deactivate_super(sb);
283 out:
284 sync_blockdev(bdev);
285 mutex_unlock(&bdev->bd_fsfreeze_mutex);
286 return sb; /* thaw_bdev releases s->s_umount */
287}
288EXPORT_SYMBOL(freeze_bdev);
289
290/**
291 * thaw_bdev -- unlock filesystem
292 * @bdev: blockdevice to unlock
293 * @sb: associated superblock
294 *
295 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
296 */
297int thaw_bdev(struct block_device *bdev, struct super_block *sb)
298{
299 int error = -EINVAL;
300
301 mutex_lock(&bdev->bd_fsfreeze_mutex);
302 if (!bdev->bd_fsfreeze_count)
303 goto out;
304
305 error = 0;
306 if (--bdev->bd_fsfreeze_count > 0)
307 goto out;
308
309 if (!sb)
310 goto out;
311
312 if (sb->s_op->thaw_super)
313 error = sb->s_op->thaw_super(sb);
314 else
315 error = thaw_super(sb);
316 if (error)
317 bdev->bd_fsfreeze_count++;
318out:
319 mutex_unlock(&bdev->bd_fsfreeze_mutex);
320 return error;
321}
322EXPORT_SYMBOL(thaw_bdev);
323
324static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
325{
326 return block_write_full_page(page, blkdev_get_block, wbc);
327}
328
329static int blkdev_readpage(struct file * file, struct page * page)
330{
331 return block_read_full_page(page, blkdev_get_block);
332}
333
334static int blkdev_readpages(struct file *file, struct address_space *mapping,
335 struct list_head *pages, unsigned nr_pages)
336{
337 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
338}
339
340static int blkdev_write_begin(struct file *file, struct address_space *mapping,
341 loff_t pos, unsigned len, unsigned flags,
342 struct page **pagep, void **fsdata)
343{
344 return block_write_begin(mapping, pos, len, flags, pagep,
345 blkdev_get_block);
346}
347
348static int blkdev_write_end(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned copied,
350 struct page *page, void *fsdata)
351{
352 int ret;
353 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
354
355 unlock_page(page);
356 put_page(page);
357
358 return ret;
359}
360
361/*
362 * private llseek:
363 * for a block special file file_inode(file)->i_size is zero
364 * so we compute the size by hand (just as in block_read/write above)
365 */
366static loff_t block_llseek(struct file *file, loff_t offset, int whence)
367{
368 struct inode *bd_inode = bdev_file_inode(file);
369 loff_t retval;
370
371 inode_lock(bd_inode);
372 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
373 inode_unlock(bd_inode);
374 return retval;
375}
376
377int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
378{
379 struct inode *bd_inode = bdev_file_inode(filp);
380 struct block_device *bdev = I_BDEV(bd_inode);
381 int error;
382
383 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
384 if (error)
385 return error;
386
387 /*
388 * There is no need to serialise calls to blkdev_issue_flush with
389 * i_mutex and doing so causes performance issues with concurrent
390 * O_SYNC writers to a block device.
391 */
392 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
393 if (error == -EOPNOTSUPP)
394 error = 0;
395
396 return error;
397}
398EXPORT_SYMBOL(blkdev_fsync);
399
400/**
401 * bdev_read_page() - Start reading a page from a block device
402 * @bdev: The device to read the page from
403 * @sector: The offset on the device to read the page to (need not be aligned)
404 * @page: The page to read
405 *
406 * On entry, the page should be locked. It will be unlocked when the page
407 * has been read. If the block driver implements rw_page synchronously,
408 * that will be true on exit from this function, but it need not be.
409 *
410 * Errors returned by this function are usually "soft", eg out of memory, or
411 * queue full; callers should try a different route to read this page rather
412 * than propagate an error back up the stack.
413 *
414 * Return: negative errno if an error occurs, 0 if submission was successful.
415 */
416int bdev_read_page(struct block_device *bdev, sector_t sector,
417 struct page *page)
418{
419 const struct block_device_operations *ops = bdev->bd_disk->fops;
420 int result = -EOPNOTSUPP;
421
422 if (!ops->rw_page || bdev_get_integrity(bdev))
423 return result;
424
425 result = blk_queue_enter(bdev->bd_queue, false);
426 if (result)
427 return result;
428 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
429 blk_queue_exit(bdev->bd_queue);
430 return result;
431}
432EXPORT_SYMBOL_GPL(bdev_read_page);
433
434/**
435 * bdev_write_page() - Start writing a page to a block device
436 * @bdev: The device to write the page to
437 * @sector: The offset on the device to write the page to (need not be aligned)
438 * @page: The page to write
439 * @wbc: The writeback_control for the write
440 *
441 * On entry, the page should be locked and not currently under writeback.
442 * On exit, if the write started successfully, the page will be unlocked and
443 * under writeback. If the write failed already (eg the driver failed to
444 * queue the page to the device), the page will still be locked. If the
445 * caller is a ->writepage implementation, it will need to unlock the page.
446 *
447 * Errors returned by this function are usually "soft", eg out of memory, or
448 * queue full; callers should try a different route to write this page rather
449 * than propagate an error back up the stack.
450 *
451 * Return: negative errno if an error occurs, 0 if submission was successful.
452 */
453int bdev_write_page(struct block_device *bdev, sector_t sector,
454 struct page *page, struct writeback_control *wbc)
455{
456 int result;
457 const struct block_device_operations *ops = bdev->bd_disk->fops;
458
459 if (!ops->rw_page || bdev_get_integrity(bdev))
460 return -EOPNOTSUPP;
461 result = blk_queue_enter(bdev->bd_queue, false);
462 if (result)
463 return result;
464
465 set_page_writeback(page);
466 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
467 if (result) {
468 end_page_writeback(page);
469 } else {
470 clean_page_buffers(page);
471 unlock_page(page);
472 }
473 blk_queue_exit(bdev->bd_queue);
474 return result;
475}
476EXPORT_SYMBOL_GPL(bdev_write_page);
477
478/**
479 * bdev_direct_access() - Get the address for directly-accessibly memory
480 * @bdev: The device containing the memory
481 * @dax: control and output parameters for ->direct_access
482 *
483 * If a block device is made up of directly addressable memory, this function
484 * will tell the caller the PFN and the address of the memory. The address
485 * may be directly dereferenced within the kernel without the need to call
486 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
487 * page tables.
488 *
489 * Return: negative errno if an error occurs, otherwise the number of bytes
490 * accessible at this address.
491 */
492long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
493{
494 sector_t sector = dax->sector;
495 long avail, size = dax->size;
496 const struct block_device_operations *ops = bdev->bd_disk->fops;
497
498 /*
499 * The device driver is allowed to sleep, in order to make the
500 * memory directly accessible.
501 */
502 might_sleep();
503
504 if (size < 0)
505 return size;
506 if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access)
507 return -EOPNOTSUPP;
508 if ((sector + DIV_ROUND_UP(size, 512)) >
509 part_nr_sects_read(bdev->bd_part))
510 return -ERANGE;
511 sector += get_start_sect(bdev);
512 if (sector % (PAGE_SIZE / 512))
513 return -EINVAL;
514 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size);
515 if (!avail)
516 return -ERANGE;
517 if (avail > 0 && avail & ~PAGE_MASK)
518 return -ENXIO;
519 return min(avail, size);
520}
521EXPORT_SYMBOL_GPL(bdev_direct_access);
522
523/**
524 * bdev_dax_supported() - Check if the device supports dax for filesystem
525 * @sb: The superblock of the device
526 * @blocksize: The block size of the device
527 *
528 * This is a library function for filesystems to check if the block device
529 * can be mounted with dax option.
530 *
531 * Return: negative errno if unsupported, 0 if supported.
532 */
533int bdev_dax_supported(struct super_block *sb, int blocksize)
534{
535 struct blk_dax_ctl dax = {
536 .sector = 0,
537 .size = PAGE_SIZE,
538 };
539 int err;
540
541 if (blocksize != PAGE_SIZE) {
542 vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
543 return -EINVAL;
544 }
545
546 err = bdev_direct_access(sb->s_bdev, &dax);
547 if (err < 0) {
548 switch (err) {
549 case -EOPNOTSUPP:
550 vfs_msg(sb, KERN_ERR,
551 "error: device does not support dax");
552 break;
553 case -EINVAL:
554 vfs_msg(sb, KERN_ERR,
555 "error: unaligned partition for dax");
556 break;
557 default:
558 vfs_msg(sb, KERN_ERR,
559 "error: dax access failed (%d)", err);
560 }
561 return err;
562 }
563
564 return 0;
565}
566EXPORT_SYMBOL_GPL(bdev_dax_supported);
567
568/**
569 * bdev_dax_capable() - Return if the raw device is capable for dax
570 * @bdev: The device for raw block device access
571 */
572bool bdev_dax_capable(struct block_device *bdev)
573{
574 struct blk_dax_ctl dax = {
575 .size = PAGE_SIZE,
576 };
577
578 if (!IS_ENABLED(CONFIG_FS_DAX))
579 return false;
580
581 dax.sector = 0;
582 if (bdev_direct_access(bdev, &dax) < 0)
583 return false;
584
585 dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512);
586 if (bdev_direct_access(bdev, &dax) < 0)
587 return false;
588
589 return true;
590}
591
592/*
593 * pseudo-fs
594 */
595
596static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
597static struct kmem_cache * bdev_cachep __read_mostly;
598
599static struct inode *bdev_alloc_inode(struct super_block *sb)
600{
601 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
602 if (!ei)
603 return NULL;
604 return &ei->vfs_inode;
605}
606
607static void bdev_i_callback(struct rcu_head *head)
608{
609 struct inode *inode = container_of(head, struct inode, i_rcu);
610 struct bdev_inode *bdi = BDEV_I(inode);
611
612 kmem_cache_free(bdev_cachep, bdi);
613}
614
615static void bdev_destroy_inode(struct inode *inode)
616{
617 call_rcu(&inode->i_rcu, bdev_i_callback);
618}
619
620static void init_once(void *foo)
621{
622 struct bdev_inode *ei = (struct bdev_inode *) foo;
623 struct block_device *bdev = &ei->bdev;
624
625 memset(bdev, 0, sizeof(*bdev));
626 mutex_init(&bdev->bd_mutex);
627 INIT_LIST_HEAD(&bdev->bd_list);
628#ifdef CONFIG_SYSFS
629 INIT_LIST_HEAD(&bdev->bd_holder_disks);
630#endif
631 inode_init_once(&ei->vfs_inode);
632 /* Initialize mutex for freeze. */
633 mutex_init(&bdev->bd_fsfreeze_mutex);
634}
635
636static void bdev_evict_inode(struct inode *inode)
637{
638 struct block_device *bdev = &BDEV_I(inode)->bdev;
639 truncate_inode_pages_final(&inode->i_data);
640 invalidate_inode_buffers(inode); /* is it needed here? */
641 clear_inode(inode);
642 spin_lock(&bdev_lock);
643 list_del_init(&bdev->bd_list);
644 spin_unlock(&bdev_lock);
645}
646
647static const struct super_operations bdev_sops = {
648 .statfs = simple_statfs,
649 .alloc_inode = bdev_alloc_inode,
650 .destroy_inode = bdev_destroy_inode,
651 .drop_inode = generic_delete_inode,
652 .evict_inode = bdev_evict_inode,
653};
654
655static struct dentry *bd_mount(struct file_system_type *fs_type,
656 int flags, const char *dev_name, void *data)
657{
658 struct dentry *dent;
659 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
660 if (!IS_ERR(dent))
661 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
662 return dent;
663}
664
665static struct file_system_type bd_type = {
666 .name = "bdev",
667 .mount = bd_mount,
668 .kill_sb = kill_anon_super,
669};
670
671struct super_block *blockdev_superblock __read_mostly;
672EXPORT_SYMBOL_GPL(blockdev_superblock);
673
674void __init bdev_cache_init(void)
675{
676 int err;
677 static struct vfsmount *bd_mnt;
678
679 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
680 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
681 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
682 init_once);
683 err = register_filesystem(&bd_type);
684 if (err)
685 panic("Cannot register bdev pseudo-fs");
686 bd_mnt = kern_mount(&bd_type);
687 if (IS_ERR(bd_mnt))
688 panic("Cannot create bdev pseudo-fs");
689 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
690}
691
692/*
693 * Most likely _very_ bad one - but then it's hardly critical for small
694 * /dev and can be fixed when somebody will need really large one.
695 * Keep in mind that it will be fed through icache hash function too.
696 */
697static inline unsigned long hash(dev_t dev)
698{
699 return MAJOR(dev)+MINOR(dev);
700}
701
702static int bdev_test(struct inode *inode, void *data)
703{
704 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
705}
706
707static int bdev_set(struct inode *inode, void *data)
708{
709 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
710 return 0;
711}
712
713static LIST_HEAD(all_bdevs);
714
715struct block_device *bdget(dev_t dev)
716{
717 struct block_device *bdev;
718 struct inode *inode;
719
720 inode = iget5_locked(blockdev_superblock, hash(dev),
721 bdev_test, bdev_set, &dev);
722
723 if (!inode)
724 return NULL;
725
726 bdev = &BDEV_I(inode)->bdev;
727
728 if (inode->i_state & I_NEW) {
729 bdev->bd_contains = NULL;
730 bdev->bd_super = NULL;
731 bdev->bd_inode = inode;
732 bdev->bd_block_size = i_blocksize(inode);
733 bdev->bd_part_count = 0;
734 bdev->bd_invalidated = 0;
735 inode->i_mode = S_IFBLK;
736 inode->i_rdev = dev;
737 inode->i_bdev = bdev;
738 inode->i_data.a_ops = &def_blk_aops;
739 #ifdef CONFIG_AMLOGIC_CMA
740 mapping_set_gfp_mask(&inode->i_data, GFP_USER | __GFP_BDEV);
741 #else
742 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
743 #endif /* CONFIG_AMLOGIC_CMA */
744 spin_lock(&bdev_lock);
745 list_add(&bdev->bd_list, &all_bdevs);
746 spin_unlock(&bdev_lock);
747 unlock_new_inode(inode);
748 }
749 return bdev;
750}
751
752EXPORT_SYMBOL(bdget);
753
754/**
755 * bdgrab -- Grab a reference to an already referenced block device
756 * @bdev: Block device to grab a reference to.
757 */
758struct block_device *bdgrab(struct block_device *bdev)
759{
760 ihold(bdev->bd_inode);
761 return bdev;
762}
763EXPORT_SYMBOL(bdgrab);
764
765long nr_blockdev_pages(void)
766{
767 struct block_device *bdev;
768 long ret = 0;
769 spin_lock(&bdev_lock);
770 list_for_each_entry(bdev, &all_bdevs, bd_list) {
771 ret += bdev->bd_inode->i_mapping->nrpages;
772 }
773 spin_unlock(&bdev_lock);
774 return ret;
775}
776
777void bdput(struct block_device *bdev)
778{
779 iput(bdev->bd_inode);
780}
781
782EXPORT_SYMBOL(bdput);
783
784static struct block_device *bd_acquire(struct inode *inode)
785{
786 struct block_device *bdev;
787
788 spin_lock(&bdev_lock);
789 bdev = inode->i_bdev;
790 if (bdev) {
791 bdgrab(bdev);
792 spin_unlock(&bdev_lock);
793 return bdev;
794 }
795 spin_unlock(&bdev_lock);
796
797 bdev = bdget(inode->i_rdev);
798 if (bdev) {
799 spin_lock(&bdev_lock);
800 if (!inode->i_bdev) {
801 /*
802 * We take an additional reference to bd_inode,
803 * and it's released in clear_inode() of inode.
804 * So, we can access it via ->i_mapping always
805 * without igrab().
806 */
807 bdgrab(bdev);
808 inode->i_bdev = bdev;
809 inode->i_mapping = bdev->bd_inode->i_mapping;
810 }
811 spin_unlock(&bdev_lock);
812 }
813 return bdev;
814}
815
816/* Call when you free inode */
817
818void bd_forget(struct inode *inode)
819{
820 struct block_device *bdev = NULL;
821
822 spin_lock(&bdev_lock);
823 if (!sb_is_blkdev_sb(inode->i_sb))
824 bdev = inode->i_bdev;
825 inode->i_bdev = NULL;
826 inode->i_mapping = &inode->i_data;
827 spin_unlock(&bdev_lock);
828
829 if (bdev)
830 bdput(bdev);
831}
832
833/**
834 * bd_may_claim - test whether a block device can be claimed
835 * @bdev: block device of interest
836 * @whole: whole block device containing @bdev, may equal @bdev
837 * @holder: holder trying to claim @bdev
838 *
839 * Test whether @bdev can be claimed by @holder.
840 *
841 * CONTEXT:
842 * spin_lock(&bdev_lock).
843 *
844 * RETURNS:
845 * %true if @bdev can be claimed, %false otherwise.
846 */
847static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
848 void *holder)
849{
850 if (bdev->bd_holder == holder)
851 return true; /* already a holder */
852 else if (bdev->bd_holder != NULL)
853 return false; /* held by someone else */
854 else if (whole == bdev)
855 return true; /* is a whole device which isn't held */
856
857 else if (whole->bd_holder == bd_may_claim)
858 return true; /* is a partition of a device that is being partitioned */
859 else if (whole->bd_holder != NULL)
860 return false; /* is a partition of a held device */
861 else
862 return true; /* is a partition of an un-held device */
863}
864
865/**
866 * bd_prepare_to_claim - prepare to claim a block device
867 * @bdev: block device of interest
868 * @whole: the whole device containing @bdev, may equal @bdev
869 * @holder: holder trying to claim @bdev
870 *
871 * Prepare to claim @bdev. This function fails if @bdev is already
872 * claimed by another holder and waits if another claiming is in
873 * progress. This function doesn't actually claim. On successful
874 * return, the caller has ownership of bd_claiming and bd_holder[s].
875 *
876 * CONTEXT:
877 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
878 * it multiple times.
879 *
880 * RETURNS:
881 * 0 if @bdev can be claimed, -EBUSY otherwise.
882 */
883static int bd_prepare_to_claim(struct block_device *bdev,
884 struct block_device *whole, void *holder)
885{
886retry:
887 /* if someone else claimed, fail */
888 if (!bd_may_claim(bdev, whole, holder))
889 return -EBUSY;
890
891 /* if claiming is already in progress, wait for it to finish */
892 if (whole->bd_claiming) {
893 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
894 DEFINE_WAIT(wait);
895
896 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
897 spin_unlock(&bdev_lock);
898 schedule();
899 finish_wait(wq, &wait);
900 spin_lock(&bdev_lock);
901 goto retry;
902 }
903
904 /* yay, all mine */
905 return 0;
906}
907
908/**
909 * bd_start_claiming - start claiming a block device
910 * @bdev: block device of interest
911 * @holder: holder trying to claim @bdev
912 *
913 * @bdev is about to be opened exclusively. Check @bdev can be opened
914 * exclusively and mark that an exclusive open is in progress. Each
915 * successful call to this function must be matched with a call to
916 * either bd_finish_claiming() or bd_abort_claiming() (which do not
917 * fail).
918 *
919 * This function is used to gain exclusive access to the block device
920 * without actually causing other exclusive open attempts to fail. It
921 * should be used when the open sequence itself requires exclusive
922 * access but may subsequently fail.
923 *
924 * CONTEXT:
925 * Might sleep.
926 *
927 * RETURNS:
928 * Pointer to the block device containing @bdev on success, ERR_PTR()
929 * value on failure.
930 */
931static struct block_device *bd_start_claiming(struct block_device *bdev,
932 void *holder)
933{
934 struct gendisk *disk;
935 struct block_device *whole;
936 int partno, err;
937
938 might_sleep();
939
940 /*
941 * @bdev might not have been initialized properly yet, look up
942 * and grab the outer block device the hard way.
943 */
944 disk = get_gendisk(bdev->bd_dev, &partno);
945 if (!disk)
946 return ERR_PTR(-ENXIO);
947
948 /*
949 * Normally, @bdev should equal what's returned from bdget_disk()
950 * if partno is 0; however, some drivers (floppy) use multiple
951 * bdev's for the same physical device and @bdev may be one of the
952 * aliases. Keep @bdev if partno is 0. This means claimer
953 * tracking is broken for those devices but it has always been that
954 * way.
955 */
956 if (partno)
957 whole = bdget_disk(disk, 0);
958 else
959 whole = bdgrab(bdev);
960
961 module_put(disk->fops->owner);
962 put_disk(disk);
963 if (!whole)
964 return ERR_PTR(-ENOMEM);
965
966 /* prepare to claim, if successful, mark claiming in progress */
967 spin_lock(&bdev_lock);
968
969 err = bd_prepare_to_claim(bdev, whole, holder);
970 if (err == 0) {
971 whole->bd_claiming = holder;
972 spin_unlock(&bdev_lock);
973 return whole;
974 } else {
975 spin_unlock(&bdev_lock);
976 bdput(whole);
977 return ERR_PTR(err);
978 }
979}
980
981#ifdef CONFIG_SYSFS
982struct bd_holder_disk {
983 struct list_head list;
984 struct gendisk *disk;
985 int refcnt;
986};
987
988static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
989 struct gendisk *disk)
990{
991 struct bd_holder_disk *holder;
992
993 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
994 if (holder->disk == disk)
995 return holder;
996 return NULL;
997}
998
999static int add_symlink(struct kobject *from, struct kobject *to)
1000{
1001 return sysfs_create_link(from, to, kobject_name(to));
1002}
1003
1004static void del_symlink(struct kobject *from, struct kobject *to)
1005{
1006 sysfs_remove_link(from, kobject_name(to));
1007}
1008
1009/**
1010 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1011 * @bdev: the claimed slave bdev
1012 * @disk: the holding disk
1013 *
1014 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1015 *
1016 * This functions creates the following sysfs symlinks.
1017 *
1018 * - from "slaves" directory of the holder @disk to the claimed @bdev
1019 * - from "holders" directory of the @bdev to the holder @disk
1020 *
1021 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1022 * passed to bd_link_disk_holder(), then:
1023 *
1024 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1025 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1026 *
1027 * The caller must have claimed @bdev before calling this function and
1028 * ensure that both @bdev and @disk are valid during the creation and
1029 * lifetime of these symlinks.
1030 *
1031 * CONTEXT:
1032 * Might sleep.
1033 *
1034 * RETURNS:
1035 * 0 on success, -errno on failure.
1036 */
1037int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1038{
1039 struct bd_holder_disk *holder;
1040 int ret = 0;
1041
1042 mutex_lock(&bdev->bd_mutex);
1043
1044 WARN_ON_ONCE(!bdev->bd_holder);
1045
1046 /* FIXME: remove the following once add_disk() handles errors */
1047 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1048 goto out_unlock;
1049
1050 holder = bd_find_holder_disk(bdev, disk);
1051 if (holder) {
1052 holder->refcnt++;
1053 goto out_unlock;
1054 }
1055
1056 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1057 if (!holder) {
1058 ret = -ENOMEM;
1059 goto out_unlock;
1060 }
1061
1062 INIT_LIST_HEAD(&holder->list);
1063 holder->disk = disk;
1064 holder->refcnt = 1;
1065
1066 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1067 if (ret)
1068 goto out_free;
1069
1070 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1071 if (ret)
1072 goto out_del;
1073 /*
1074 * bdev could be deleted beneath us which would implicitly destroy
1075 * the holder directory. Hold on to it.
1076 */
1077 kobject_get(bdev->bd_part->holder_dir);
1078
1079 list_add(&holder->list, &bdev->bd_holder_disks);
1080 goto out_unlock;
1081
1082out_del:
1083 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1084out_free:
1085 kfree(holder);
1086out_unlock:
1087 mutex_unlock(&bdev->bd_mutex);
1088 return ret;
1089}
1090EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1091
1092/**
1093 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1094 * @bdev: the calimed slave bdev
1095 * @disk: the holding disk
1096 *
1097 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1098 *
1099 * CONTEXT:
1100 * Might sleep.
1101 */
1102void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1103{
1104 struct bd_holder_disk *holder;
1105
1106 mutex_lock(&bdev->bd_mutex);
1107
1108 holder = bd_find_holder_disk(bdev, disk);
1109
1110 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1111 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1112 del_symlink(bdev->bd_part->holder_dir,
1113 &disk_to_dev(disk)->kobj);
1114 kobject_put(bdev->bd_part->holder_dir);
1115 list_del_init(&holder->list);
1116 kfree(holder);
1117 }
1118
1119 mutex_unlock(&bdev->bd_mutex);
1120}
1121EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1122#endif
1123
1124/**
1125 * flush_disk - invalidates all buffer-cache entries on a disk
1126 *
1127 * @bdev: struct block device to be flushed
1128 * @kill_dirty: flag to guide handling of dirty inodes
1129 *
1130 * Invalidates all buffer-cache entries on a disk. It should be called
1131 * when a disk has been changed -- either by a media change or online
1132 * resize.
1133 */
1134static void flush_disk(struct block_device *bdev, bool kill_dirty)
1135{
1136 if (__invalidate_device(bdev, kill_dirty)) {
1137 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1138 "resized disk %s\n",
1139 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1140 }
1141
1142 if (!bdev->bd_disk)
1143 return;
1144 if (disk_part_scan_enabled(bdev->bd_disk))
1145 bdev->bd_invalidated = 1;
1146}
1147
1148/**
1149 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1150 * @disk: struct gendisk to check
1151 * @bdev: struct bdev to adjust.
1152 *
1153 * This routine checks to see if the bdev size does not match the disk size
1154 * and adjusts it if it differs.
1155 */
1156void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1157{
1158 loff_t disk_size, bdev_size;
1159
1160 disk_size = (loff_t)get_capacity(disk) << 9;
1161 bdev_size = i_size_read(bdev->bd_inode);
1162 if (disk_size != bdev_size) {
1163 printk(KERN_INFO
1164 "%s: detected capacity change from %lld to %lld\n",
1165 disk->disk_name, bdev_size, disk_size);
1166 i_size_write(bdev->bd_inode, disk_size);
1167 flush_disk(bdev, false);
1168 }
1169}
1170EXPORT_SYMBOL(check_disk_size_change);
1171
1172/**
1173 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1174 * @disk: struct gendisk to be revalidated
1175 *
1176 * This routine is a wrapper for lower-level driver's revalidate_disk
1177 * call-backs. It is used to do common pre and post operations needed
1178 * for all revalidate_disk operations.
1179 */
1180int revalidate_disk(struct gendisk *disk)
1181{
1182 struct block_device *bdev;
1183 int ret = 0;
1184
1185 if (disk->fops->revalidate_disk)
1186 ret = disk->fops->revalidate_disk(disk);
1187 bdev = bdget_disk(disk, 0);
1188 if (!bdev)
1189 return ret;
1190
1191 mutex_lock(&bdev->bd_mutex);
1192 check_disk_size_change(disk, bdev);
1193 bdev->bd_invalidated = 0;
1194 mutex_unlock(&bdev->bd_mutex);
1195 bdput(bdev);
1196 return ret;
1197}
1198EXPORT_SYMBOL(revalidate_disk);
1199
1200/*
1201 * This routine checks whether a removable media has been changed,
1202 * and invalidates all buffer-cache-entries in that case. This
1203 * is a relatively slow routine, so we have to try to minimize using
1204 * it. Thus it is called only upon a 'mount' or 'open'. This
1205 * is the best way of combining speed and utility, I think.
1206 * People changing diskettes in the middle of an operation deserve
1207 * to lose :-)
1208 */
1209int check_disk_change(struct block_device *bdev)
1210{
1211 struct gendisk *disk = bdev->bd_disk;
1212 const struct block_device_operations *bdops = disk->fops;
1213 unsigned int events;
1214
1215 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1216 DISK_EVENT_EJECT_REQUEST);
1217 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1218 return 0;
1219
1220 flush_disk(bdev, true);
1221 if (bdops->revalidate_disk)
1222 bdops->revalidate_disk(bdev->bd_disk);
1223 return 1;
1224}
1225
1226EXPORT_SYMBOL(check_disk_change);
1227
1228void bd_set_size(struct block_device *bdev, loff_t size)
1229{
1230 inode_lock(bdev->bd_inode);
1231 i_size_write(bdev->bd_inode, size);
1232 inode_unlock(bdev->bd_inode);
1233}
1234EXPORT_SYMBOL(bd_set_size);
1235
1236static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1237
1238/*
1239 * bd_mutex locking:
1240 *
1241 * mutex_lock(part->bd_mutex)
1242 * mutex_lock_nested(whole->bd_mutex, 1)
1243 */
1244
1245static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1246{
1247 struct gendisk *disk;
1248 struct module *owner;
1249 int ret;
1250 int partno;
1251 int perm = 0;
1252
1253 if (mode & FMODE_READ)
1254 perm |= MAY_READ;
1255 if (mode & FMODE_WRITE)
1256 perm |= MAY_WRITE;
1257 /*
1258 * hooks: /n/, see "layering violations".
1259 */
1260 if (!for_part) {
1261 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1262 if (ret != 0) {
1263 bdput(bdev);
1264 return ret;
1265 }
1266 }
1267
1268 restart:
1269
1270 ret = -ENXIO;
1271 disk = get_gendisk(bdev->bd_dev, &partno);
1272 if (!disk)
1273 goto out;
1274 owner = disk->fops->owner;
1275
1276 disk_block_events(disk);
1277 mutex_lock_nested(&bdev->bd_mutex, for_part);
1278 if (!bdev->bd_openers) {
1279 bdev->bd_disk = disk;
1280 bdev->bd_queue = disk->queue;
1281 bdev->bd_contains = bdev;
1282
1283 if (!partno) {
1284 ret = -ENXIO;
1285 bdev->bd_part = disk_get_part(disk, partno);
1286 if (!bdev->bd_part)
1287 goto out_clear;
1288
1289 ret = 0;
1290 if (disk->fops->open) {
1291 ret = disk->fops->open(bdev, mode);
1292 if (ret == -ERESTARTSYS) {
1293 /* Lost a race with 'disk' being
1294 * deleted, try again.
1295 * See md.c
1296 */
1297 disk_put_part(bdev->bd_part);
1298 bdev->bd_part = NULL;
1299 bdev->bd_disk = NULL;
1300 bdev->bd_queue = NULL;
1301 mutex_unlock(&bdev->bd_mutex);
1302 disk_unblock_events(disk);
1303 put_disk(disk);
1304 module_put(owner);
1305 goto restart;
1306 }
1307 }
1308
1309 if (!ret) {
1310 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1311 set_init_blocksize(bdev);
1312 }
1313
1314 /*
1315 * If the device is invalidated, rescan partition
1316 * if open succeeded or failed with -ENOMEDIUM.
1317 * The latter is necessary to prevent ghost
1318 * partitions on a removed medium.
1319 */
1320 if (bdev->bd_invalidated) {
1321 if (!ret)
1322 rescan_partitions(disk, bdev);
1323 else if (ret == -ENOMEDIUM)
1324 invalidate_partitions(disk, bdev);
1325 }
1326
1327 if (ret)
1328 goto out_clear;
1329 } else {
1330 struct block_device *whole;
1331 whole = bdget_disk(disk, 0);
1332 ret = -ENOMEM;
1333 if (!whole)
1334 goto out_clear;
1335 BUG_ON(for_part);
1336 ret = __blkdev_get(whole, mode, 1);
1337 if (ret)
1338 goto out_clear;
1339 bdev->bd_contains = whole;
1340 bdev->bd_part = disk_get_part(disk, partno);
1341 if (!(disk->flags & GENHD_FL_UP) ||
1342 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1343 ret = -ENXIO;
1344 goto out_clear;
1345 }
1346 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1347 set_init_blocksize(bdev);
1348 }
1349 } else {
1350 if (bdev->bd_contains == bdev) {
1351 ret = 0;
1352 if (bdev->bd_disk->fops->open)
1353 ret = bdev->bd_disk->fops->open(bdev, mode);
1354 /* the same as first opener case, read comment there */
1355 if (bdev->bd_invalidated) {
1356 if (!ret)
1357 rescan_partitions(bdev->bd_disk, bdev);
1358 else if (ret == -ENOMEDIUM)
1359 invalidate_partitions(bdev->bd_disk, bdev);
1360 }
1361 if (ret)
1362 goto out_unlock_bdev;
1363 }
1364 /* only one opener holds refs to the module and disk */
1365 put_disk(disk);
1366 module_put(owner);
1367 }
1368 bdev->bd_openers++;
1369 if (for_part)
1370 bdev->bd_part_count++;
1371 mutex_unlock(&bdev->bd_mutex);
1372 disk_unblock_events(disk);
1373 return 0;
1374
1375 out_clear:
1376 disk_put_part(bdev->bd_part);
1377 bdev->bd_disk = NULL;
1378 bdev->bd_part = NULL;
1379 bdev->bd_queue = NULL;
1380 if (bdev != bdev->bd_contains)
1381 __blkdev_put(bdev->bd_contains, mode, 1);
1382 bdev->bd_contains = NULL;
1383 out_unlock_bdev:
1384 mutex_unlock(&bdev->bd_mutex);
1385 disk_unblock_events(disk);
1386 put_disk(disk);
1387 module_put(owner);
1388 out:
1389 bdput(bdev);
1390
1391 return ret;
1392}
1393
1394/**
1395 * blkdev_get - open a block device
1396 * @bdev: block_device to open
1397 * @mode: FMODE_* mask
1398 * @holder: exclusive holder identifier
1399 *
1400 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1401 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1402 * @holder is invalid. Exclusive opens may nest for the same @holder.
1403 *
1404 * On success, the reference count of @bdev is unchanged. On failure,
1405 * @bdev is put.
1406 *
1407 * CONTEXT:
1408 * Might sleep.
1409 *
1410 * RETURNS:
1411 * 0 on success, -errno on failure.
1412 */
1413int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1414{
1415 struct block_device *whole = NULL;
1416 int res;
1417
1418 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1419
1420 if ((mode & FMODE_EXCL) && holder) {
1421 whole = bd_start_claiming(bdev, holder);
1422 if (IS_ERR(whole)) {
1423 bdput(bdev);
1424 return PTR_ERR(whole);
1425 }
1426 }
1427
1428 res = __blkdev_get(bdev, mode, 0);
1429
1430 if (whole) {
1431 struct gendisk *disk = whole->bd_disk;
1432
1433 /* finish claiming */
1434 mutex_lock(&bdev->bd_mutex);
1435 spin_lock(&bdev_lock);
1436
1437 if (!res) {
1438 BUG_ON(!bd_may_claim(bdev, whole, holder));
1439 /*
1440 * Note that for a whole device bd_holders
1441 * will be incremented twice, and bd_holder
1442 * will be set to bd_may_claim before being
1443 * set to holder
1444 */
1445 whole->bd_holders++;
1446 whole->bd_holder = bd_may_claim;
1447 bdev->bd_holders++;
1448 bdev->bd_holder = holder;
1449 }
1450
1451 /* tell others that we're done */
1452 BUG_ON(whole->bd_claiming != holder);
1453 whole->bd_claiming = NULL;
1454 wake_up_bit(&whole->bd_claiming, 0);
1455
1456 spin_unlock(&bdev_lock);
1457
1458 /*
1459 * Block event polling for write claims if requested. Any
1460 * write holder makes the write_holder state stick until
1461 * all are released. This is good enough and tracking
1462 * individual writeable reference is too fragile given the
1463 * way @mode is used in blkdev_get/put().
1464 */
1465 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1466 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1467 bdev->bd_write_holder = true;
1468 disk_block_events(disk);
1469 }
1470
1471 mutex_unlock(&bdev->bd_mutex);
1472 bdput(whole);
1473 }
1474
1475 return res;
1476}
1477EXPORT_SYMBOL(blkdev_get);
1478
1479/**
1480 * blkdev_get_by_path - open a block device by name
1481 * @path: path to the block device to open
1482 * @mode: FMODE_* mask
1483 * @holder: exclusive holder identifier
1484 *
1485 * Open the blockdevice described by the device file at @path. @mode
1486 * and @holder are identical to blkdev_get().
1487 *
1488 * On success, the returned block_device has reference count of one.
1489 *
1490 * CONTEXT:
1491 * Might sleep.
1492 *
1493 * RETURNS:
1494 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1495 */
1496struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1497 void *holder)
1498{
1499 struct block_device *bdev;
1500 int err;
1501
1502 bdev = lookup_bdev(path);
1503 if (IS_ERR(bdev))
1504 return bdev;
1505
1506 err = blkdev_get(bdev, mode, holder);
1507 if (err)
1508 return ERR_PTR(err);
1509
1510 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1511 blkdev_put(bdev, mode);
1512 return ERR_PTR(-EACCES);
1513 }
1514
1515 return bdev;
1516}
1517EXPORT_SYMBOL(blkdev_get_by_path);
1518
1519/**
1520 * blkdev_get_by_dev - open a block device by device number
1521 * @dev: device number of block device to open
1522 * @mode: FMODE_* mask
1523 * @holder: exclusive holder identifier
1524 *
1525 * Open the blockdevice described by device number @dev. @mode and
1526 * @holder are identical to blkdev_get().
1527 *
1528 * Use it ONLY if you really do not have anything better - i.e. when
1529 * you are behind a truly sucky interface and all you are given is a
1530 * device number. _Never_ to be used for internal purposes. If you
1531 * ever need it - reconsider your API.
1532 *
1533 * On success, the returned block_device has reference count of one.
1534 *
1535 * CONTEXT:
1536 * Might sleep.
1537 *
1538 * RETURNS:
1539 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1540 */
1541struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1542{
1543 struct block_device *bdev;
1544 int err;
1545
1546 bdev = bdget(dev);
1547 if (!bdev)
1548 return ERR_PTR(-ENOMEM);
1549
1550 err = blkdev_get(bdev, mode, holder);
1551 if (err)
1552 return ERR_PTR(err);
1553
1554 return bdev;
1555}
1556EXPORT_SYMBOL(blkdev_get_by_dev);
1557
1558static int blkdev_open(struct inode * inode, struct file * filp)
1559{
1560 struct block_device *bdev;
1561
1562 /*
1563 * Preserve backwards compatibility and allow large file access
1564 * even if userspace doesn't ask for it explicitly. Some mkfs
1565 * binary needs it. We might want to drop this workaround
1566 * during an unstable branch.
1567 */
1568 filp->f_flags |= O_LARGEFILE;
1569
1570 if (filp->f_flags & O_NDELAY)
1571 filp->f_mode |= FMODE_NDELAY;
1572 if (filp->f_flags & O_EXCL)
1573 filp->f_mode |= FMODE_EXCL;
1574 if ((filp->f_flags & O_ACCMODE) == 3)
1575 filp->f_mode |= FMODE_WRITE_IOCTL;
1576
1577 bdev = bd_acquire(inode);
1578 if (bdev == NULL)
1579 return -ENOMEM;
1580
1581 filp->f_mapping = bdev->bd_inode->i_mapping;
1582
1583 return blkdev_get(bdev, filp->f_mode, filp);
1584}
1585
1586static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1587{
1588 struct gendisk *disk = bdev->bd_disk;
1589 struct block_device *victim = NULL;
1590
1591 mutex_lock_nested(&bdev->bd_mutex, for_part);
1592 if (for_part)
1593 bdev->bd_part_count--;
1594
1595 if (!--bdev->bd_openers) {
1596 WARN_ON_ONCE(bdev->bd_holders);
1597 sync_blockdev(bdev);
1598 kill_bdev(bdev);
1599
1600 bdev_write_inode(bdev);
1601 /*
1602 * Detaching bdev inode from its wb in __destroy_inode()
1603 * is too late: the queue which embeds its bdi (along with
1604 * root wb) can be gone as soon as we put_disk() below.
1605 */
1606 inode_detach_wb(bdev->bd_inode);
1607 }
1608 if (bdev->bd_contains == bdev) {
1609 if (disk->fops->release)
1610 disk->fops->release(disk, mode);
1611 }
1612 if (!bdev->bd_openers) {
1613 struct module *owner = disk->fops->owner;
1614
1615 disk_put_part(bdev->bd_part);
1616 bdev->bd_part = NULL;
1617 bdev->bd_disk = NULL;
1618 if (bdev != bdev->bd_contains)
1619 victim = bdev->bd_contains;
1620 bdev->bd_contains = NULL;
1621
1622 put_disk(disk);
1623 module_put(owner);
1624 }
1625 mutex_unlock(&bdev->bd_mutex);
1626 bdput(bdev);
1627 if (victim)
1628 __blkdev_put(victim, mode, 1);
1629}
1630
1631void blkdev_put(struct block_device *bdev, fmode_t mode)
1632{
1633 mutex_lock(&bdev->bd_mutex);
1634
1635 if (mode & FMODE_EXCL) {
1636 bool bdev_free;
1637
1638 /*
1639 * Release a claim on the device. The holder fields
1640 * are protected with bdev_lock. bd_mutex is to
1641 * synchronize disk_holder unlinking.
1642 */
1643 spin_lock(&bdev_lock);
1644
1645 WARN_ON_ONCE(--bdev->bd_holders < 0);
1646 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1647
1648 /* bd_contains might point to self, check in a separate step */
1649 if ((bdev_free = !bdev->bd_holders))
1650 bdev->bd_holder = NULL;
1651 if (!bdev->bd_contains->bd_holders)
1652 bdev->bd_contains->bd_holder = NULL;
1653
1654 spin_unlock(&bdev_lock);
1655
1656 /*
1657 * If this was the last claim, remove holder link and
1658 * unblock evpoll if it was a write holder.
1659 */
1660 if (bdev_free && bdev->bd_write_holder) {
1661 disk_unblock_events(bdev->bd_disk);
1662 bdev->bd_write_holder = false;
1663 }
1664 }
1665
1666 /*
1667 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1668 * event. This is to ensure detection of media removal commanded
1669 * from userland - e.g. eject(1).
1670 */
1671 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1672
1673 mutex_unlock(&bdev->bd_mutex);
1674
1675 __blkdev_put(bdev, mode, 0);
1676}
1677EXPORT_SYMBOL(blkdev_put);
1678
1679static int blkdev_close(struct inode * inode, struct file * filp)
1680{
1681 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1682 blkdev_put(bdev, filp->f_mode);
1683 return 0;
1684}
1685
1686static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1687{
1688 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1689 fmode_t mode = file->f_mode;
1690
1691 /*
1692 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1693 * to updated it before every ioctl.
1694 */
1695 if (file->f_flags & O_NDELAY)
1696 mode |= FMODE_NDELAY;
1697 else
1698 mode &= ~FMODE_NDELAY;
1699
1700 return blkdev_ioctl(bdev, mode, cmd, arg);
1701}
1702
1703/*
1704 * Write data to the block device. Only intended for the block device itself
1705 * and the raw driver which basically is a fake block device.
1706 *
1707 * Does not take i_mutex for the write and thus is not for general purpose
1708 * use.
1709 */
1710ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1711{
1712 struct file *file = iocb->ki_filp;
1713 struct inode *bd_inode = bdev_file_inode(file);
1714 loff_t size = i_size_read(bd_inode);
1715 struct blk_plug plug;
1716 ssize_t ret;
1717
1718 if (bdev_read_only(I_BDEV(bd_inode)))
1719 return -EPERM;
1720
1721 if (!iov_iter_count(from))
1722 return 0;
1723
1724 if (iocb->ki_pos >= size)
1725 return -ENOSPC;
1726
1727 iov_iter_truncate(from, size - iocb->ki_pos);
1728
1729 blk_start_plug(&plug);
1730 ret = __generic_file_write_iter(iocb, from);
1731 if (ret > 0)
1732 ret = generic_write_sync(iocb, ret);
1733 blk_finish_plug(&plug);
1734 return ret;
1735}
1736EXPORT_SYMBOL_GPL(blkdev_write_iter);
1737
1738ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1739{
1740 struct file *file = iocb->ki_filp;
1741 struct inode *bd_inode = bdev_file_inode(file);
1742 loff_t size = i_size_read(bd_inode);
1743 loff_t pos = iocb->ki_pos;
1744
1745 if (pos >= size)
1746 return 0;
1747
1748 size -= pos;
1749 iov_iter_truncate(to, size);
1750 return generic_file_read_iter(iocb, to);
1751}
1752EXPORT_SYMBOL_GPL(blkdev_read_iter);
1753
1754/*
1755 * Try to release a page associated with block device when the system
1756 * is under memory pressure.
1757 */
1758static int blkdev_releasepage(struct page *page, gfp_t wait)
1759{
1760 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1761
1762 if (super && super->s_op->bdev_try_to_free_page)
1763 return super->s_op->bdev_try_to_free_page(super, page, wait);
1764
1765 return try_to_free_buffers(page);
1766}
1767
1768static int blkdev_writepages(struct address_space *mapping,
1769 struct writeback_control *wbc)
1770{
1771 if (dax_mapping(mapping)) {
1772 struct block_device *bdev = I_BDEV(mapping->host);
1773
1774 return dax_writeback_mapping_range(mapping, bdev, wbc);
1775 }
1776 return generic_writepages(mapping, wbc);
1777}
1778
1779static const struct address_space_operations def_blk_aops = {
1780 .readpage = blkdev_readpage,
1781 .readpages = blkdev_readpages,
1782 .writepage = blkdev_writepage,
1783 .write_begin = blkdev_write_begin,
1784 .write_end = blkdev_write_end,
1785 .writepages = blkdev_writepages,
1786 .releasepage = blkdev_releasepage,
1787 .direct_IO = blkdev_direct_IO,
1788 .is_dirty_writeback = buffer_check_dirty_writeback,
1789};
1790
1791#define BLKDEV_FALLOC_FL_SUPPORTED \
1792 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1793 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1794
1795static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1796 loff_t len)
1797{
1798 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1799 struct request_queue *q = bdev_get_queue(bdev);
1800 struct address_space *mapping;
1801 loff_t end = start + len - 1;
1802 loff_t isize;
1803 int error;
1804
1805 /* Fail if we don't recognize the flags. */
1806 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1807 return -EOPNOTSUPP;
1808
1809 /* Don't go off the end of the device. */
1810 isize = i_size_read(bdev->bd_inode);
1811 if (start >= isize)
1812 return -EINVAL;
1813 if (end >= isize) {
1814 if (mode & FALLOC_FL_KEEP_SIZE) {
1815 len = isize - start;
1816 end = start + len - 1;
1817 } else
1818 return -EINVAL;
1819 }
1820
1821 /*
1822 * Don't allow IO that isn't aligned to logical block size.
1823 */
1824 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1825 return -EINVAL;
1826
1827 /* Invalidate the page cache, including dirty pages. */
1828 mapping = bdev->bd_inode->i_mapping;
1829 truncate_inode_pages_range(mapping, start, end);
1830
1831 switch (mode) {
1832 case FALLOC_FL_ZERO_RANGE:
1833 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1834 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1835 GFP_KERNEL, false);
1836 break;
1837 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1838 /* Only punch if the device can do zeroing discard. */
1839 if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data)
1840 return -EOPNOTSUPP;
1841 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1842 GFP_KERNEL, 0);
1843 break;
1844 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1845 if (!blk_queue_discard(q))
1846 return -EOPNOTSUPP;
1847 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1848 GFP_KERNEL, 0);
1849 break;
1850 default:
1851 return -EOPNOTSUPP;
1852 }
1853 if (error)
1854 return error;
1855
1856 /*
1857 * Invalidate again; if someone wandered in and dirtied a page,
1858 * the caller will be given -EBUSY. The third argument is
1859 * inclusive, so the rounding here is safe.
1860 */
1861 return invalidate_inode_pages2_range(mapping,
1862 start >> PAGE_SHIFT,
1863 end >> PAGE_SHIFT);
1864}
1865
1866const struct file_operations def_blk_fops = {
1867 .open = blkdev_open,
1868 .release = blkdev_close,
1869 .llseek = block_llseek,
1870 .read_iter = blkdev_read_iter,
1871 .write_iter = blkdev_write_iter,
1872 .mmap = generic_file_mmap,
1873 .fsync = blkdev_fsync,
1874 .unlocked_ioctl = block_ioctl,
1875#ifdef CONFIG_COMPAT
1876 .compat_ioctl = compat_blkdev_ioctl,
1877#endif
1878 .splice_read = generic_file_splice_read,
1879 .splice_write = iter_file_splice_write,
1880 .fallocate = blkdev_fallocate,
1881};
1882
1883int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1884{
1885 int res;
1886 mm_segment_t old_fs = get_fs();
1887 set_fs(KERNEL_DS);
1888 res = blkdev_ioctl(bdev, 0, cmd, arg);
1889 set_fs(old_fs);
1890 return res;
1891}
1892
1893EXPORT_SYMBOL(ioctl_by_bdev);
1894
1895/**
1896 * lookup_bdev - lookup a struct block_device by name
1897 * @pathname: special file representing the block device
1898 *
1899 * Get a reference to the blockdevice at @pathname in the current
1900 * namespace if possible and return it. Return ERR_PTR(error)
1901 * otherwise.
1902 */
1903struct block_device *lookup_bdev(const char *pathname)
1904{
1905 struct block_device *bdev;
1906 struct inode *inode;
1907 struct path path;
1908 int error;
1909
1910 if (!pathname || !*pathname)
1911 return ERR_PTR(-EINVAL);
1912
1913 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1914 if (error)
1915 return ERR_PTR(error);
1916
1917 inode = d_backing_inode(path.dentry);
1918 error = -ENOTBLK;
1919 if (!S_ISBLK(inode->i_mode))
1920 goto fail;
1921 error = -EACCES;
1922 if (!may_open_dev(&path))
1923 goto fail;
1924 error = -ENOMEM;
1925 bdev = bd_acquire(inode);
1926 if (!bdev)
1927 goto fail;
1928out:
1929 path_put(&path);
1930 return bdev;
1931fail:
1932 bdev = ERR_PTR(error);
1933 goto out;
1934}
1935EXPORT_SYMBOL(lookup_bdev);
1936
1937int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1938{
1939 struct super_block *sb = get_super(bdev);
1940 int res = 0;
1941
1942 if (sb) {
1943 /*
1944 * no need to lock the super, get_super holds the
1945 * read mutex so the filesystem cannot go away
1946 * under us (->put_super runs with the write lock
1947 * hold).
1948 */
1949 shrink_dcache_sb(sb);
1950 res = invalidate_inodes(sb, kill_dirty);
1951 drop_super(sb);
1952 }
1953 invalidate_bdev(bdev);
1954 return res;
1955}
1956EXPORT_SYMBOL(__invalidate_device);
1957
1958void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1959{
1960 struct inode *inode, *old_inode = NULL;
1961
1962 spin_lock(&blockdev_superblock->s_inode_list_lock);
1963 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1964 struct address_space *mapping = inode->i_mapping;
1965 struct block_device *bdev;
1966
1967 spin_lock(&inode->i_lock);
1968 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1969 mapping->nrpages == 0) {
1970 spin_unlock(&inode->i_lock);
1971 continue;
1972 }
1973 __iget(inode);
1974 spin_unlock(&inode->i_lock);
1975 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1976 /*
1977 * We hold a reference to 'inode' so it couldn't have been
1978 * removed from s_inodes list while we dropped the
1979 * s_inode_list_lock We cannot iput the inode now as we can
1980 * be holding the last reference and we cannot iput it under
1981 * s_inode_list_lock. So we keep the reference and iput it
1982 * later.
1983 */
1984 iput(old_inode);
1985 old_inode = inode;
1986 bdev = I_BDEV(inode);
1987
1988 mutex_lock(&bdev->bd_mutex);
1989 if (bdev->bd_openers)
1990 func(bdev, arg);
1991 mutex_unlock(&bdev->bd_mutex);
1992
1993 spin_lock(&blockdev_superblock->s_inode_list_lock);
1994 }
1995 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1996 iput(old_inode);
1997}
1998
generated by cgit at 2024-05-18 11:48:16 (GMT)