Root/fs/block_dev.c

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/module.h>
18#include <linux/blkpg.h>
19#include <linux/buffer_head.h>
20#include <linux/pagevec.h>
21#include <linux/writeback.h>
22#include <linux/mpage.h>
23#include <linux/mount.h>
24#include <linux/uio.h>
25#include <linux/namei.h>
26#include <linux/log2.h>
27#include <linux/kmemleak.h>
28#include <asm/uaccess.h>
29#include "internal.h"
30
31struct bdev_inode {
32    struct block_device bdev;
33    struct inode vfs_inode;
34};
35
36static const struct address_space_operations def_blk_aops;
37
38static inline struct bdev_inode *BDEV_I(struct inode *inode)
39{
40    return container_of(inode, struct bdev_inode, vfs_inode);
41}
42
43inline struct block_device *I_BDEV(struct inode *inode)
44{
45    return &BDEV_I(inode)->bdev;
46}
47
48EXPORT_SYMBOL(I_BDEV);
49
50/*
51 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
52 * we need to move it onto the dirty list of @dst so that the inode is always
53 * on the right list.
54 */
55static void bdev_inode_switch_bdi(struct inode *inode,
56            struct backing_dev_info *dst)
57{
58    spin_lock(&inode_wb_list_lock);
59    spin_lock(&inode->i_lock);
60    inode->i_data.backing_dev_info = dst;
61    if (inode->i_state & I_DIRTY)
62        list_move(&inode->i_wb_list, &dst->wb.b_dirty);
63    spin_unlock(&inode->i_lock);
64    spin_unlock(&inode_wb_list_lock);
65}
66
67static sector_t max_block(struct block_device *bdev)
68{
69    sector_t retval = ~((sector_t)0);
70    loff_t sz = i_size_read(bdev->bd_inode);
71
72    if (sz) {
73        unsigned int size = block_size(bdev);
74        unsigned int sizebits = blksize_bits(size);
75        retval = (sz >> sizebits);
76    }
77    return retval;
78}
79
80/* Kill _all_ buffers and pagecache , dirty or not.. */
81static void kill_bdev(struct block_device *bdev)
82{
83    if (bdev->bd_inode->i_mapping->nrpages == 0)
84        return;
85    invalidate_bh_lrus();
86    truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
87}
88
89int set_blocksize(struct block_device *bdev, int size)
90{
91    /* Size must be a power of two, and between 512 and PAGE_SIZE */
92    if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
93        return -EINVAL;
94
95    /* Size cannot be smaller than the size supported by the device */
96    if (size < bdev_logical_block_size(bdev))
97        return -EINVAL;
98
99    /* Don't change the size if it is same as current */
100    if (bdev->bd_block_size != size) {
101        sync_blockdev(bdev);
102        bdev->bd_block_size = size;
103        bdev->bd_inode->i_blkbits = blksize_bits(size);
104        kill_bdev(bdev);
105    }
106    return 0;
107}
108
109EXPORT_SYMBOL(set_blocksize);
110
111int sb_set_blocksize(struct super_block *sb, int size)
112{
113    if (set_blocksize(sb->s_bdev, size))
114        return 0;
115    /* If we get here, we know size is power of two
116     * and it's value is between 512 and PAGE_SIZE */
117    sb->s_blocksize = size;
118    sb->s_blocksize_bits = blksize_bits(size);
119    return sb->s_blocksize;
120}
121
122EXPORT_SYMBOL(sb_set_blocksize);
123
124int sb_min_blocksize(struct super_block *sb, int size)
125{
126    int minsize = bdev_logical_block_size(sb->s_bdev);
127    if (size < minsize)
128        size = minsize;
129    return sb_set_blocksize(sb, size);
130}
131
132EXPORT_SYMBOL(sb_min_blocksize);
133
134static int
135blkdev_get_block(struct inode *inode, sector_t iblock,
136        struct buffer_head *bh, int create)
137{
138    if (iblock >= max_block(I_BDEV(inode))) {
139        if (create)
140            return -EIO;
141
142        /*
143         * for reads, we're just trying to fill a partial page.
144         * return a hole, they will have to call get_block again
145         * before they can fill it, and they will get -EIO at that
146         * time
147         */
148        return 0;
149    }
150    bh->b_bdev = I_BDEV(inode);
151    bh->b_blocknr = iblock;
152    set_buffer_mapped(bh);
153    return 0;
154}
155
156static int
157blkdev_get_blocks(struct inode *inode, sector_t iblock,
158        struct buffer_head *bh, int create)
159{
160    sector_t end_block = max_block(I_BDEV(inode));
161    unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
162
163    if ((iblock + max_blocks) > end_block) {
164        max_blocks = end_block - iblock;
165        if ((long)max_blocks <= 0) {
166            if (create)
167                return -EIO; /* write fully beyond EOF */
168            /*
169             * It is a read which is fully beyond EOF. We return
170             * a !buffer_mapped buffer
171             */
172            max_blocks = 0;
173        }
174    }
175
176    bh->b_bdev = I_BDEV(inode);
177    bh->b_blocknr = iblock;
178    bh->b_size = max_blocks << inode->i_blkbits;
179    if (max_blocks)
180        set_buffer_mapped(bh);
181    return 0;
182}
183
184static ssize_t
185blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
186            loff_t offset, unsigned long nr_segs)
187{
188    struct file *file = iocb->ki_filp;
189    struct inode *inode = file->f_mapping->host;
190
191    return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
192                    nr_segs, blkdev_get_blocks, NULL, NULL, 0);
193}
194
195int __sync_blockdev(struct block_device *bdev, int wait)
196{
197    if (!bdev)
198        return 0;
199    if (!wait)
200        return filemap_flush(bdev->bd_inode->i_mapping);
201    return filemap_write_and_wait(bdev->bd_inode->i_mapping);
202}
203
204/*
205 * Write out and wait upon all the dirty data associated with a block
206 * device via its mapping. Does not take the superblock lock.
207 */
208int sync_blockdev(struct block_device *bdev)
209{
210    return __sync_blockdev(bdev, 1);
211}
212EXPORT_SYMBOL(sync_blockdev);
213
214/*
215 * Write out and wait upon all dirty data associated with this
216 * device. Filesystem data as well as the underlying block
217 * device. Takes the superblock lock.
218 */
219int fsync_bdev(struct block_device *bdev)
220{
221    struct super_block *sb = get_super(bdev);
222    if (sb) {
223        int res = sync_filesystem(sb);
224        drop_super(sb);
225        return res;
226    }
227    return sync_blockdev(bdev);
228}
229EXPORT_SYMBOL(fsync_bdev);
230
231/**
232 * freeze_bdev -- lock a filesystem and force it into a consistent state
233 * @bdev: blockdevice to lock
234 *
235 * If a superblock is found on this device, we take the s_umount semaphore
236 * on it to make sure nobody unmounts until the snapshot creation is done.
237 * The reference counter (bd_fsfreeze_count) guarantees that only the last
238 * unfreeze process can unfreeze the frozen filesystem actually when multiple
239 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
240 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
241 * actually.
242 */
243struct super_block *freeze_bdev(struct block_device *bdev)
244{
245    struct super_block *sb;
246    int error = 0;
247
248    mutex_lock(&bdev->bd_fsfreeze_mutex);
249    if (++bdev->bd_fsfreeze_count > 1) {
250        /*
251         * We don't even need to grab a reference - the first call
252         * to freeze_bdev grab an active reference and only the last
253         * thaw_bdev drops it.
254         */
255        sb = get_super(bdev);
256        drop_super(sb);
257        mutex_unlock(&bdev->bd_fsfreeze_mutex);
258        return sb;
259    }
260
261    sb = get_active_super(bdev);
262    if (!sb)
263        goto out;
264    error = freeze_super(sb);
265    if (error) {
266        deactivate_super(sb);
267        bdev->bd_fsfreeze_count--;
268        mutex_unlock(&bdev->bd_fsfreeze_mutex);
269        return ERR_PTR(error);
270    }
271    deactivate_super(sb);
272 out:
273    sync_blockdev(bdev);
274    mutex_unlock(&bdev->bd_fsfreeze_mutex);
275    return sb; /* thaw_bdev releases s->s_umount */
276}
277EXPORT_SYMBOL(freeze_bdev);
278
279/**
280 * thaw_bdev -- unlock filesystem
281 * @bdev: blockdevice to unlock
282 * @sb: associated superblock
283 *
284 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
285 */
286int thaw_bdev(struct block_device *bdev, struct super_block *sb)
287{
288    int error = -EINVAL;
289
290    mutex_lock(&bdev->bd_fsfreeze_mutex);
291    if (!bdev->bd_fsfreeze_count)
292        goto out;
293
294    error = 0;
295    if (--bdev->bd_fsfreeze_count > 0)
296        goto out;
297
298    if (!sb)
299        goto out;
300
301    error = thaw_super(sb);
302    if (error) {
303        bdev->bd_fsfreeze_count++;
304        mutex_unlock(&bdev->bd_fsfreeze_mutex);
305        return error;
306    }
307out:
308    mutex_unlock(&bdev->bd_fsfreeze_mutex);
309    return 0;
310}
311EXPORT_SYMBOL(thaw_bdev);
312
313static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
314{
315    return block_write_full_page(page, blkdev_get_block, wbc);
316}
317
318static int blkdev_readpage(struct file * file, struct page * page)
319{
320    return block_read_full_page(page, blkdev_get_block);
321}
322
323static int blkdev_write_begin(struct file *file, struct address_space *mapping,
324            loff_t pos, unsigned len, unsigned flags,
325            struct page **pagep, void **fsdata)
326{
327    return block_write_begin(mapping, pos, len, flags, pagep,
328                 blkdev_get_block);
329}
330
331static int blkdev_write_end(struct file *file, struct address_space *mapping,
332            loff_t pos, unsigned len, unsigned copied,
333            struct page *page, void *fsdata)
334{
335    int ret;
336    ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
337
338    unlock_page(page);
339    page_cache_release(page);
340
341    return ret;
342}
343
344/*
345 * private llseek:
346 * for a block special file file->f_path.dentry->d_inode->i_size is zero
347 * so we compute the size by hand (just as in block_read/write above)
348 */
349static loff_t block_llseek(struct file *file, loff_t offset, int origin)
350{
351    struct inode *bd_inode = file->f_mapping->host;
352    loff_t size;
353    loff_t retval;
354
355    mutex_lock(&bd_inode->i_mutex);
356    size = i_size_read(bd_inode);
357
358    switch (origin) {
359        case 2:
360            offset += size;
361            break;
362        case 1:
363            offset += file->f_pos;
364    }
365    retval = -EINVAL;
366    if (offset >= 0 && offset <= size) {
367        if (offset != file->f_pos) {
368            file->f_pos = offset;
369        }
370        retval = offset;
371    }
372    mutex_unlock(&bd_inode->i_mutex);
373    return retval;
374}
375    
376int blkdev_fsync(struct file *filp, int datasync)
377{
378    struct inode *bd_inode = filp->f_mapping->host;
379    struct block_device *bdev = I_BDEV(bd_inode);
380    int error;
381
382    /*
383     * There is no need to serialise calls to blkdev_issue_flush with
384     * i_mutex and doing so causes performance issues with concurrent
385     * O_SYNC writers to a block device.
386     */
387    mutex_unlock(&bd_inode->i_mutex);
388
389    error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
390    if (error == -EOPNOTSUPP)
391        error = 0;
392
393    mutex_lock(&bd_inode->i_mutex);
394
395    return error;
396}
397EXPORT_SYMBOL(blkdev_fsync);
398
399/*
400 * pseudo-fs
401 */
402
403static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
404static struct kmem_cache * bdev_cachep __read_mostly;
405
406static struct inode *bdev_alloc_inode(struct super_block *sb)
407{
408    struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
409    if (!ei)
410        return NULL;
411    return &ei->vfs_inode;
412}
413
414static void bdev_i_callback(struct rcu_head *head)
415{
416    struct inode *inode = container_of(head, struct inode, i_rcu);
417    struct bdev_inode *bdi = BDEV_I(inode);
418
419    INIT_LIST_HEAD(&inode->i_dentry);
420    kmem_cache_free(bdev_cachep, bdi);
421}
422
423static void bdev_destroy_inode(struct inode *inode)
424{
425    call_rcu(&inode->i_rcu, bdev_i_callback);
426}
427
428static void init_once(void *foo)
429{
430    struct bdev_inode *ei = (struct bdev_inode *) foo;
431    struct block_device *bdev = &ei->bdev;
432
433    memset(bdev, 0, sizeof(*bdev));
434    mutex_init(&bdev->bd_mutex);
435    INIT_LIST_HEAD(&bdev->bd_inodes);
436    INIT_LIST_HEAD(&bdev->bd_list);
437#ifdef CONFIG_SYSFS
438    INIT_LIST_HEAD(&bdev->bd_holder_disks);
439#endif
440    inode_init_once(&ei->vfs_inode);
441    /* Initialize mutex for freeze. */
442    mutex_init(&bdev->bd_fsfreeze_mutex);
443}
444
445static inline void __bd_forget(struct inode *inode)
446{
447    list_del_init(&inode->i_devices);
448    inode->i_bdev = NULL;
449    inode->i_mapping = &inode->i_data;
450}
451
452static void bdev_evict_inode(struct inode *inode)
453{
454    struct block_device *bdev = &BDEV_I(inode)->bdev;
455    struct list_head *p;
456    truncate_inode_pages(&inode->i_data, 0);
457    invalidate_inode_buffers(inode); /* is it needed here? */
458    end_writeback(inode);
459    spin_lock(&bdev_lock);
460    while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
461        __bd_forget(list_entry(p, struct inode, i_devices));
462    }
463    list_del_init(&bdev->bd_list);
464    spin_unlock(&bdev_lock);
465}
466
467static const struct super_operations bdev_sops = {
468    .statfs = simple_statfs,
469    .alloc_inode = bdev_alloc_inode,
470    .destroy_inode = bdev_destroy_inode,
471    .drop_inode = generic_delete_inode,
472    .evict_inode = bdev_evict_inode,
473};
474
475static struct dentry *bd_mount(struct file_system_type *fs_type,
476    int flags, const char *dev_name, void *data)
477{
478    return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
479}
480
481static struct file_system_type bd_type = {
482    .name = "bdev",
483    .mount = bd_mount,
484    .kill_sb = kill_anon_super,
485};
486
487struct super_block *blockdev_superblock __read_mostly;
488
489void __init bdev_cache_init(void)
490{
491    int err;
492    struct vfsmount *bd_mnt;
493
494    bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
495            0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
496                SLAB_MEM_SPREAD|SLAB_PANIC),
497            init_once);
498    err = register_filesystem(&bd_type);
499    if (err)
500        panic("Cannot register bdev pseudo-fs");
501    bd_mnt = kern_mount(&bd_type);
502    if (IS_ERR(bd_mnt))
503        panic("Cannot create bdev pseudo-fs");
504    /*
505     * This vfsmount structure is only used to obtain the
506     * blockdev_superblock, so tell kmemleak not to report it.
507     */
508    kmemleak_not_leak(bd_mnt);
509    blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
510}
511
512/*
513 * Most likely _very_ bad one - but then it's hardly critical for small
514 * /dev and can be fixed when somebody will need really large one.
515 * Keep in mind that it will be fed through icache hash function too.
516 */
517static inline unsigned long hash(dev_t dev)
518{
519    return MAJOR(dev)+MINOR(dev);
520}
521
522static int bdev_test(struct inode *inode, void *data)
523{
524    return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
525}
526
527static int bdev_set(struct inode *inode, void *data)
528{
529    BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
530    return 0;
531}
532
533static LIST_HEAD(all_bdevs);
534
535struct block_device *bdget(dev_t dev)
536{
537    struct block_device *bdev;
538    struct inode *inode;
539
540    inode = iget5_locked(blockdev_superblock, hash(dev),
541            bdev_test, bdev_set, &dev);
542
543    if (!inode)
544        return NULL;
545
546    bdev = &BDEV_I(inode)->bdev;
547
548    if (inode->i_state & I_NEW) {
549        bdev->bd_contains = NULL;
550        bdev->bd_inode = inode;
551        bdev->bd_block_size = (1 << inode->i_blkbits);
552        bdev->bd_part_count = 0;
553        bdev->bd_invalidated = 0;
554        inode->i_mode = S_IFBLK;
555        inode->i_rdev = dev;
556        inode->i_bdev = bdev;
557        inode->i_data.a_ops = &def_blk_aops;
558        mapping_set_gfp_mask(&inode->i_data, GFP_USER);
559        inode->i_data.backing_dev_info = &default_backing_dev_info;
560        spin_lock(&bdev_lock);
561        list_add(&bdev->bd_list, &all_bdevs);
562        spin_unlock(&bdev_lock);
563        unlock_new_inode(inode);
564    }
565    return bdev;
566}
567
568EXPORT_SYMBOL(bdget);
569
570/**
571 * bdgrab -- Grab a reference to an already referenced block device
572 * @bdev: Block device to grab a reference to.
573 */
574struct block_device *bdgrab(struct block_device *bdev)
575{
576    ihold(bdev->bd_inode);
577    return bdev;
578}
579
580long nr_blockdev_pages(void)
581{
582    struct block_device *bdev;
583    long ret = 0;
584    spin_lock(&bdev_lock);
585    list_for_each_entry(bdev, &all_bdevs, bd_list) {
586        ret += bdev->bd_inode->i_mapping->nrpages;
587    }
588    spin_unlock(&bdev_lock);
589    return ret;
590}
591
592void bdput(struct block_device *bdev)
593{
594    iput(bdev->bd_inode);
595}
596
597EXPORT_SYMBOL(bdput);
598 
599static struct block_device *bd_acquire(struct inode *inode)
600{
601    struct block_device *bdev;
602
603    spin_lock(&bdev_lock);
604    bdev = inode->i_bdev;
605    if (bdev) {
606        ihold(bdev->bd_inode);
607        spin_unlock(&bdev_lock);
608        return bdev;
609    }
610    spin_unlock(&bdev_lock);
611
612    bdev = bdget(inode->i_rdev);
613    if (bdev) {
614        spin_lock(&bdev_lock);
615        if (!inode->i_bdev) {
616            /*
617             * We take an additional reference to bd_inode,
618             * and it's released in clear_inode() of inode.
619             * So, we can access it via ->i_mapping always
620             * without igrab().
621             */
622            ihold(bdev->bd_inode);
623            inode->i_bdev = bdev;
624            inode->i_mapping = bdev->bd_inode->i_mapping;
625            list_add(&inode->i_devices, &bdev->bd_inodes);
626        }
627        spin_unlock(&bdev_lock);
628    }
629    return bdev;
630}
631
632/* Call when you free inode */
633
634void bd_forget(struct inode *inode)
635{
636    struct block_device *bdev = NULL;
637
638    spin_lock(&bdev_lock);
639    if (inode->i_bdev) {
640        if (!sb_is_blkdev_sb(inode->i_sb))
641            bdev = inode->i_bdev;
642        __bd_forget(inode);
643    }
644    spin_unlock(&bdev_lock);
645
646    if (bdev)
647        iput(bdev->bd_inode);
648}
649
650/**
651 * bd_may_claim - test whether a block device can be claimed
652 * @bdev: block device of interest
653 * @whole: whole block device containing @bdev, may equal @bdev
654 * @holder: holder trying to claim @bdev
655 *
656 * Test whether @bdev can be claimed by @holder.
657 *
658 * CONTEXT:
659 * spin_lock(&bdev_lock).
660 *
661 * RETURNS:
662 * %true if @bdev can be claimed, %false otherwise.
663 */
664static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
665             void *holder)
666{
667    if (bdev->bd_holder == holder)
668        return true; /* already a holder */
669    else if (bdev->bd_holder != NULL)
670        return false; /* held by someone else */
671    else if (bdev->bd_contains == bdev)
672        return true; /* is a whole device which isn't held */
673
674    else if (whole->bd_holder == bd_may_claim)
675        return true; /* is a partition of a device that is being partitioned */
676    else if (whole->bd_holder != NULL)
677        return false; /* is a partition of a held device */
678    else
679        return true; /* is a partition of an un-held device */
680}
681
682/**
683 * bd_prepare_to_claim - prepare to claim a block device
684 * @bdev: block device of interest
685 * @whole: the whole device containing @bdev, may equal @bdev
686 * @holder: holder trying to claim @bdev
687 *
688 * Prepare to claim @bdev. This function fails if @bdev is already
689 * claimed by another holder and waits if another claiming is in
690 * progress. This function doesn't actually claim. On successful
691 * return, the caller has ownership of bd_claiming and bd_holder[s].
692 *
693 * CONTEXT:
694 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
695 * it multiple times.
696 *
697 * RETURNS:
698 * 0 if @bdev can be claimed, -EBUSY otherwise.
699 */
700static int bd_prepare_to_claim(struct block_device *bdev,
701                   struct block_device *whole, void *holder)
702{
703retry:
704    /* if someone else claimed, fail */
705    if (!bd_may_claim(bdev, whole, holder))
706        return -EBUSY;
707
708    /* if claiming is already in progress, wait for it to finish */
709    if (whole->bd_claiming) {
710        wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
711        DEFINE_WAIT(wait);
712
713        prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
714        spin_unlock(&bdev_lock);
715        schedule();
716        finish_wait(wq, &wait);
717        spin_lock(&bdev_lock);
718        goto retry;
719    }
720
721    /* yay, all mine */
722    return 0;
723}
724
725/**
726 * bd_start_claiming - start claiming a block device
727 * @bdev: block device of interest
728 * @holder: holder trying to claim @bdev
729 *
730 * @bdev is about to be opened exclusively. Check @bdev can be opened
731 * exclusively and mark that an exclusive open is in progress. Each
732 * successful call to this function must be matched with a call to
733 * either bd_finish_claiming() or bd_abort_claiming() (which do not
734 * fail).
735 *
736 * This function is used to gain exclusive access to the block device
737 * without actually causing other exclusive open attempts to fail. It
738 * should be used when the open sequence itself requires exclusive
739 * access but may subsequently fail.
740 *
741 * CONTEXT:
742 * Might sleep.
743 *
744 * RETURNS:
745 * Pointer to the block device containing @bdev on success, ERR_PTR()
746 * value on failure.
747 */
748static struct block_device *bd_start_claiming(struct block_device *bdev,
749                          void *holder)
750{
751    struct gendisk *disk;
752    struct block_device *whole;
753    int partno, err;
754
755    might_sleep();
756
757    /*
758     * @bdev might not have been initialized properly yet, look up
759     * and grab the outer block device the hard way.
760     */
761    disk = get_gendisk(bdev->bd_dev, &partno);
762    if (!disk)
763        return ERR_PTR(-ENXIO);
764
765    /*
766     * Normally, @bdev should equal what's returned from bdget_disk()
767     * if partno is 0; however, some drivers (floppy) use multiple
768     * bdev's for the same physical device and @bdev may be one of the
769     * aliases. Keep @bdev if partno is 0. This means claimer
770     * tracking is broken for those devices but it has always been that
771     * way.
772     */
773    if (partno)
774        whole = bdget_disk(disk, 0);
775    else
776        whole = bdgrab(bdev);
777
778    module_put(disk->fops->owner);
779    put_disk(disk);
780    if (!whole)
781        return ERR_PTR(-ENOMEM);
782
783    /* prepare to claim, if successful, mark claiming in progress */
784    spin_lock(&bdev_lock);
785
786    err = bd_prepare_to_claim(bdev, whole, holder);
787    if (err == 0) {
788        whole->bd_claiming = holder;
789        spin_unlock(&bdev_lock);
790        return whole;
791    } else {
792        spin_unlock(&bdev_lock);
793        bdput(whole);
794        return ERR_PTR(err);
795    }
796}
797
798#ifdef CONFIG_SYSFS
799struct bd_holder_disk {
800    struct list_head list;
801    struct gendisk *disk;
802    int refcnt;
803};
804
805static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
806                          struct gendisk *disk)
807{
808    struct bd_holder_disk *holder;
809
810    list_for_each_entry(holder, &bdev->bd_holder_disks, list)
811        if (holder->disk == disk)
812            return holder;
813    return NULL;
814}
815
816static int add_symlink(struct kobject *from, struct kobject *to)
817{
818    return sysfs_create_link(from, to, kobject_name(to));
819}
820
821static void del_symlink(struct kobject *from, struct kobject *to)
822{
823    sysfs_remove_link(from, kobject_name(to));
824}
825
826/**
827 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
828 * @bdev: the claimed slave bdev
829 * @disk: the holding disk
830 *
831 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
832 *
833 * This functions creates the following sysfs symlinks.
834 *
835 * - from "slaves" directory of the holder @disk to the claimed @bdev
836 * - from "holders" directory of the @bdev to the holder @disk
837 *
838 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
839 * passed to bd_link_disk_holder(), then:
840 *
841 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
842 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
843 *
844 * The caller must have claimed @bdev before calling this function and
845 * ensure that both @bdev and @disk are valid during the creation and
846 * lifetime of these symlinks.
847 *
848 * CONTEXT:
849 * Might sleep.
850 *
851 * RETURNS:
852 * 0 on success, -errno on failure.
853 */
854int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
855{
856    struct bd_holder_disk *holder;
857    int ret = 0;
858
859    mutex_lock(&bdev->bd_mutex);
860
861    WARN_ON_ONCE(!bdev->bd_holder);
862
863    /* FIXME: remove the following once add_disk() handles errors */
864    if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
865        goto out_unlock;
866
867    holder = bd_find_holder_disk(bdev, disk);
868    if (holder) {
869        holder->refcnt++;
870        goto out_unlock;
871    }
872
873    holder = kzalloc(sizeof(*holder), GFP_KERNEL);
874    if (!holder) {
875        ret = -ENOMEM;
876        goto out_unlock;
877    }
878
879    INIT_LIST_HEAD(&holder->list);
880    holder->disk = disk;
881    holder->refcnt = 1;
882
883    ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
884    if (ret)
885        goto out_free;
886
887    ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
888    if (ret)
889        goto out_del;
890    /*
891     * bdev could be deleted beneath us which would implicitly destroy
892     * the holder directory. Hold on to it.
893     */
894    kobject_get(bdev->bd_part->holder_dir);
895
896    list_add(&holder->list, &bdev->bd_holder_disks);
897    goto out_unlock;
898
899out_del:
900    del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
901out_free:
902    kfree(holder);
903out_unlock:
904    mutex_unlock(&bdev->bd_mutex);
905    return ret;
906}
907EXPORT_SYMBOL_GPL(bd_link_disk_holder);
908
909/**
910 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
911 * @bdev: the calimed slave bdev
912 * @disk: the holding disk
913 *
914 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
915 *
916 * CONTEXT:
917 * Might sleep.
918 */
919void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
920{
921    struct bd_holder_disk *holder;
922
923    mutex_lock(&bdev->bd_mutex);
924
925    holder = bd_find_holder_disk(bdev, disk);
926
927    if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
928        del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
929        del_symlink(bdev->bd_part->holder_dir,
930                &disk_to_dev(disk)->kobj);
931        kobject_put(bdev->bd_part->holder_dir);
932        list_del_init(&holder->list);
933        kfree(holder);
934    }
935
936    mutex_unlock(&bdev->bd_mutex);
937}
938EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
939#endif
940
941/**
942 * flush_disk - invalidates all buffer-cache entries on a disk
943 *
944 * @bdev: struct block device to be flushed
945 * @kill_dirty: flag to guide handling of dirty inodes
946 *
947 * Invalidates all buffer-cache entries on a disk. It should be called
948 * when a disk has been changed -- either by a media change or online
949 * resize.
950 */
951static void flush_disk(struct block_device *bdev, bool kill_dirty)
952{
953    if (__invalidate_device(bdev, kill_dirty)) {
954        char name[BDEVNAME_SIZE] = "";
955
956        if (bdev->bd_disk)
957            disk_name(bdev->bd_disk, 0, name);
958        printk(KERN_WARNING "VFS: busy inodes on changed media or "
959               "resized disk %s\n", name);
960    }
961
962    if (!bdev->bd_disk)
963        return;
964    if (disk_partitionable(bdev->bd_disk))
965        bdev->bd_invalidated = 1;
966}
967
968/**
969 * check_disk_size_change - checks for disk size change and adjusts bdev size.
970 * @disk: struct gendisk to check
971 * @bdev: struct bdev to adjust.
972 *
973 * This routine checks to see if the bdev size does not match the disk size
974 * and adjusts it if it differs.
975 */
976void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
977{
978    loff_t disk_size, bdev_size;
979
980    disk_size = (loff_t)get_capacity(disk) << 9;
981    bdev_size = i_size_read(bdev->bd_inode);
982    if (disk_size != bdev_size) {
983        char name[BDEVNAME_SIZE];
984
985        disk_name(disk, 0, name);
986        printk(KERN_INFO
987               "%s: detected capacity change from %lld to %lld\n",
988               name, bdev_size, disk_size);
989        i_size_write(bdev->bd_inode, disk_size);
990        flush_disk(bdev, false);
991    }
992}
993EXPORT_SYMBOL(check_disk_size_change);
994
995/**
996 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
997 * @disk: struct gendisk to be revalidated
998 *
999 * This routine is a wrapper for lower-level driver's revalidate_disk
1000 * call-backs. It is used to do common pre and post operations needed
1001 * for all revalidate_disk operations.
1002 */
1003int revalidate_disk(struct gendisk *disk)
1004{
1005    struct block_device *bdev;
1006    int ret = 0;
1007
1008    if (disk->fops->revalidate_disk)
1009        ret = disk->fops->revalidate_disk(disk);
1010
1011    bdev = bdget_disk(disk, 0);
1012    if (!bdev)
1013        return ret;
1014
1015    mutex_lock(&bdev->bd_mutex);
1016    check_disk_size_change(disk, bdev);
1017    mutex_unlock(&bdev->bd_mutex);
1018    bdput(bdev);
1019    return ret;
1020}
1021EXPORT_SYMBOL(revalidate_disk);
1022
1023/*
1024 * This routine checks whether a removable media has been changed,
1025 * and invalidates all buffer-cache-entries in that case. This
1026 * is a relatively slow routine, so we have to try to minimize using
1027 * it. Thus it is called only upon a 'mount' or 'open'. This
1028 * is the best way of combining speed and utility, I think.
1029 * People changing diskettes in the middle of an operation deserve
1030 * to lose :-)
1031 */
1032int check_disk_change(struct block_device *bdev)
1033{
1034    struct gendisk *disk = bdev->bd_disk;
1035    const struct block_device_operations *bdops = disk->fops;
1036    unsigned int events;
1037
1038    events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1039                   DISK_EVENT_EJECT_REQUEST);
1040    if (!(events & DISK_EVENT_MEDIA_CHANGE))
1041        return 0;
1042
1043    flush_disk(bdev, true);
1044    if (bdops->revalidate_disk)
1045        bdops->revalidate_disk(bdev->bd_disk);
1046    return 1;
1047}
1048
1049EXPORT_SYMBOL(check_disk_change);
1050
1051void bd_set_size(struct block_device *bdev, loff_t size)
1052{
1053    unsigned bsize = bdev_logical_block_size(bdev);
1054
1055    bdev->bd_inode->i_size = size;
1056    while (bsize < PAGE_CACHE_SIZE) {
1057        if (size & bsize)
1058            break;
1059        bsize <<= 1;
1060    }
1061    bdev->bd_block_size = bsize;
1062    bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1063}
1064EXPORT_SYMBOL(bd_set_size);
1065
1066static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1067
1068/*
1069 * bd_mutex locking:
1070 *
1071 * mutex_lock(part->bd_mutex)
1072 * mutex_lock_nested(whole->bd_mutex, 1)
1073 */
1074
1075static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1076{
1077    struct gendisk *disk;
1078    int ret;
1079    int partno;
1080    int perm = 0;
1081
1082    if (mode & FMODE_READ)
1083        perm |= MAY_READ;
1084    if (mode & FMODE_WRITE)
1085        perm |= MAY_WRITE;
1086    /*
1087     * hooks: /n/, see "layering violations".
1088     */
1089    if (!for_part) {
1090        ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1091        if (ret != 0) {
1092            bdput(bdev);
1093            return ret;
1094        }
1095    }
1096
1097 restart:
1098
1099    ret = -ENXIO;
1100    disk = get_gendisk(bdev->bd_dev, &partno);
1101    if (!disk)
1102        goto out;
1103
1104    disk_block_events(disk);
1105    mutex_lock_nested(&bdev->bd_mutex, for_part);
1106    if (!bdev->bd_openers) {
1107        bdev->bd_disk = disk;
1108        bdev->bd_contains = bdev;
1109        if (!partno) {
1110            struct backing_dev_info *bdi;
1111
1112            ret = -ENXIO;
1113            bdev->bd_part = disk_get_part(disk, partno);
1114            if (!bdev->bd_part)
1115                goto out_clear;
1116
1117            ret = 0;
1118            if (disk->fops->open) {
1119                ret = disk->fops->open(bdev, mode);
1120                if (ret == -ERESTARTSYS) {
1121                    /* Lost a race with 'disk' being
1122                     * deleted, try again.
1123                     * See md.c
1124                     */
1125                    disk_put_part(bdev->bd_part);
1126                    bdev->bd_part = NULL;
1127                    bdev->bd_disk = NULL;
1128                    mutex_unlock(&bdev->bd_mutex);
1129                    disk_unblock_events(disk);
1130                    module_put(disk->fops->owner);
1131                    put_disk(disk);
1132                    goto restart;
1133                }
1134            }
1135
1136            if (!ret && !bdev->bd_openers) {
1137                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1138                bdi = blk_get_backing_dev_info(bdev);
1139                if (bdi == NULL)
1140                    bdi = &default_backing_dev_info;
1141                bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1142            }
1143
1144            /*
1145             * If the device is invalidated, rescan partition
1146             * if open succeeded or failed with -ENOMEDIUM.
1147             * The latter is necessary to prevent ghost
1148             * partitions on a removed medium.
1149             */
1150            if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1151                rescan_partitions(disk, bdev);
1152            if (ret)
1153                goto out_clear;
1154        } else {
1155            struct block_device *whole;
1156            whole = bdget_disk(disk, 0);
1157            ret = -ENOMEM;
1158            if (!whole)
1159                goto out_clear;
1160            BUG_ON(for_part);
1161            ret = __blkdev_get(whole, mode, 1);
1162            if (ret)
1163                goto out_clear;
1164            bdev->bd_contains = whole;
1165            bdev_inode_switch_bdi(bdev->bd_inode,
1166                whole->bd_inode->i_data.backing_dev_info);
1167            bdev->bd_part = disk_get_part(disk, partno);
1168            if (!(disk->flags & GENHD_FL_UP) ||
1169                !bdev->bd_part || !bdev->bd_part->nr_sects) {
1170                ret = -ENXIO;
1171                goto out_clear;
1172            }
1173            bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1174        }
1175    } else {
1176        if (bdev->bd_contains == bdev) {
1177            ret = 0;
1178            if (bdev->bd_disk->fops->open)
1179                ret = bdev->bd_disk->fops->open(bdev, mode);
1180            /* the same as first opener case, read comment there */
1181            if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1182                rescan_partitions(bdev->bd_disk, bdev);
1183            if (ret)
1184                goto out_unlock_bdev;
1185        }
1186        /* only one opener holds refs to the module and disk */
1187        module_put(disk->fops->owner);
1188        put_disk(disk);
1189    }
1190    bdev->bd_openers++;
1191    if (for_part)
1192        bdev->bd_part_count++;
1193    mutex_unlock(&bdev->bd_mutex);
1194    disk_unblock_events(disk);
1195    return 0;
1196
1197 out_clear:
1198    disk_put_part(bdev->bd_part);
1199    bdev->bd_disk = NULL;
1200    bdev->bd_part = NULL;
1201    bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1202    if (bdev != bdev->bd_contains)
1203        __blkdev_put(bdev->bd_contains, mode, 1);
1204    bdev->bd_contains = NULL;
1205 out_unlock_bdev:
1206    mutex_unlock(&bdev->bd_mutex);
1207    disk_unblock_events(disk);
1208    module_put(disk->fops->owner);
1209    put_disk(disk);
1210 out:
1211    bdput(bdev);
1212
1213    return ret;
1214}
1215
1216/**
1217 * blkdev_get - open a block device
1218 * @bdev: block_device to open
1219 * @mode: FMODE_* mask
1220 * @holder: exclusive holder identifier
1221 *
1222 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1223 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1224 * @holder is invalid. Exclusive opens may nest for the same @holder.
1225 *
1226 * On success, the reference count of @bdev is unchanged. On failure,
1227 * @bdev is put.
1228 *
1229 * CONTEXT:
1230 * Might sleep.
1231 *
1232 * RETURNS:
1233 * 0 on success, -errno on failure.
1234 */
1235int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1236{
1237    struct block_device *whole = NULL;
1238    int res;
1239
1240    WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1241
1242    if ((mode & FMODE_EXCL) && holder) {
1243        whole = bd_start_claiming(bdev, holder);
1244        if (IS_ERR(whole)) {
1245            bdput(bdev);
1246            return PTR_ERR(whole);
1247        }
1248    }
1249
1250    res = __blkdev_get(bdev, mode, 0);
1251
1252    if (whole) {
1253        struct gendisk *disk = whole->bd_disk;
1254
1255        /* finish claiming */
1256        mutex_lock(&bdev->bd_mutex);
1257        spin_lock(&bdev_lock);
1258
1259        if (!res) {
1260            BUG_ON(!bd_may_claim(bdev, whole, holder));
1261            /*
1262             * Note that for a whole device bd_holders
1263             * will be incremented twice, and bd_holder
1264             * will be set to bd_may_claim before being
1265             * set to holder
1266             */
1267            whole->bd_holders++;
1268            whole->bd_holder = bd_may_claim;
1269            bdev->bd_holders++;
1270            bdev->bd_holder = holder;
1271        }
1272
1273        /* tell others that we're done */
1274        BUG_ON(whole->bd_claiming != holder);
1275        whole->bd_claiming = NULL;
1276        wake_up_bit(&whole->bd_claiming, 0);
1277
1278        spin_unlock(&bdev_lock);
1279
1280        /*
1281         * Block event polling for write claims if requested. Any
1282         * write holder makes the write_holder state stick until
1283         * all are released. This is good enough and tracking
1284         * individual writeable reference is too fragile given the
1285         * way @mode is used in blkdev_get/put().
1286         */
1287        if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1288            (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1289            bdev->bd_write_holder = true;
1290            disk_block_events(disk);
1291        }
1292
1293        mutex_unlock(&bdev->bd_mutex);
1294        bdput(whole);
1295    }
1296
1297    return res;
1298}
1299EXPORT_SYMBOL(blkdev_get);
1300
1301/**
1302 * blkdev_get_by_path - open a block device by name
1303 * @path: path to the block device to open
1304 * @mode: FMODE_* mask
1305 * @holder: exclusive holder identifier
1306 *
1307 * Open the blockdevice described by the device file at @path. @mode
1308 * and @holder are identical to blkdev_get().
1309 *
1310 * On success, the returned block_device has reference count of one.
1311 *
1312 * CONTEXT:
1313 * Might sleep.
1314 *
1315 * RETURNS:
1316 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1317 */
1318struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1319                    void *holder)
1320{
1321    struct block_device *bdev;
1322    int err;
1323
1324    bdev = lookup_bdev(path);
1325    if (IS_ERR(bdev))
1326        return bdev;
1327
1328    err = blkdev_get(bdev, mode, holder);
1329    if (err)
1330        return ERR_PTR(err);
1331
1332    if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1333        blkdev_put(bdev, mode);
1334        return ERR_PTR(-EACCES);
1335    }
1336
1337    return bdev;
1338}
1339EXPORT_SYMBOL(blkdev_get_by_path);
1340
1341/**
1342 * blkdev_get_by_dev - open a block device by device number
1343 * @dev: device number of block device to open
1344 * @mode: FMODE_* mask
1345 * @holder: exclusive holder identifier
1346 *
1347 * Open the blockdevice described by device number @dev. @mode and
1348 * @holder are identical to blkdev_get().
1349 *
1350 * Use it ONLY if you really do not have anything better - i.e. when
1351 * you are behind a truly sucky interface and all you are given is a
1352 * device number. _Never_ to be used for internal purposes. If you
1353 * ever need it - reconsider your API.
1354 *
1355 * On success, the returned block_device has reference count of one.
1356 *
1357 * CONTEXT:
1358 * Might sleep.
1359 *
1360 * RETURNS:
1361 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1362 */
1363struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1364{
1365    struct block_device *bdev;
1366    int err;
1367
1368    bdev = bdget(dev);
1369    if (!bdev)
1370        return ERR_PTR(-ENOMEM);
1371
1372    err = blkdev_get(bdev, mode, holder);
1373    if (err)
1374        return ERR_PTR(err);
1375
1376    return bdev;
1377}
1378EXPORT_SYMBOL(blkdev_get_by_dev);
1379
1380static int blkdev_open(struct inode * inode, struct file * filp)
1381{
1382    struct block_device *bdev;
1383
1384    /*
1385     * Preserve backwards compatibility and allow large file access
1386     * even if userspace doesn't ask for it explicitly. Some mkfs
1387     * binary needs it. We might want to drop this workaround
1388     * during an unstable branch.
1389     */
1390    filp->f_flags |= O_LARGEFILE;
1391
1392    if (filp->f_flags & O_NDELAY)
1393        filp->f_mode |= FMODE_NDELAY;
1394    if (filp->f_flags & O_EXCL)
1395        filp->f_mode |= FMODE_EXCL;
1396    if ((filp->f_flags & O_ACCMODE) == 3)
1397        filp->f_mode |= FMODE_WRITE_IOCTL;
1398
1399    bdev = bd_acquire(inode);
1400    if (bdev == NULL)
1401        return -ENOMEM;
1402
1403    filp->f_mapping = bdev->bd_inode->i_mapping;
1404
1405    return blkdev_get(bdev, filp->f_mode, filp);
1406}
1407
1408static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1409{
1410    int ret = 0;
1411    struct gendisk *disk = bdev->bd_disk;
1412    struct block_device *victim = NULL;
1413
1414    mutex_lock_nested(&bdev->bd_mutex, for_part);
1415    if (for_part)
1416        bdev->bd_part_count--;
1417
1418    if (!--bdev->bd_openers) {
1419        WARN_ON_ONCE(bdev->bd_holders);
1420        sync_blockdev(bdev);
1421        kill_bdev(bdev);
1422    }
1423    if (bdev->bd_contains == bdev) {
1424        if (disk->fops->release)
1425            ret = disk->fops->release(disk, mode);
1426    }
1427    if (!bdev->bd_openers) {
1428        struct module *owner = disk->fops->owner;
1429
1430        put_disk(disk);
1431        module_put(owner);
1432        disk_put_part(bdev->bd_part);
1433        bdev->bd_part = NULL;
1434        bdev->bd_disk = NULL;
1435        bdev_inode_switch_bdi(bdev->bd_inode,
1436                    &default_backing_dev_info);
1437        if (bdev != bdev->bd_contains)
1438            victim = bdev->bd_contains;
1439        bdev->bd_contains = NULL;
1440    }
1441    mutex_unlock(&bdev->bd_mutex);
1442    bdput(bdev);
1443    if (victim)
1444        __blkdev_put(victim, mode, 1);
1445    return ret;
1446}
1447
1448int blkdev_put(struct block_device *bdev, fmode_t mode)
1449{
1450    if (mode & FMODE_EXCL) {
1451        bool bdev_free;
1452
1453        /*
1454         * Release a claim on the device. The holder fields
1455         * are protected with bdev_lock. bd_mutex is to
1456         * synchronize disk_holder unlinking.
1457         */
1458        mutex_lock(&bdev->bd_mutex);
1459        spin_lock(&bdev_lock);
1460
1461        WARN_ON_ONCE(--bdev->bd_holders < 0);
1462        WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1463
1464        /* bd_contains might point to self, check in a separate step */
1465        if ((bdev_free = !bdev->bd_holders))
1466            bdev->bd_holder = NULL;
1467        if (!bdev->bd_contains->bd_holders)
1468            bdev->bd_contains->bd_holder = NULL;
1469
1470        spin_unlock(&bdev_lock);
1471
1472        /*
1473         * If this was the last claim, remove holder link and
1474         * unblock evpoll if it was a write holder.
1475         */
1476        if (bdev_free) {
1477            if (bdev->bd_write_holder) {
1478                disk_unblock_events(bdev->bd_disk);
1479                disk_check_events(bdev->bd_disk);
1480                bdev->bd_write_holder = false;
1481            }
1482        }
1483
1484        mutex_unlock(&bdev->bd_mutex);
1485    }
1486
1487    return __blkdev_put(bdev, mode, 0);
1488}
1489EXPORT_SYMBOL(blkdev_put);
1490
1491static int blkdev_close(struct inode * inode, struct file * filp)
1492{
1493    struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1494
1495    return blkdev_put(bdev, filp->f_mode);
1496}
1497
1498static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1499{
1500    struct block_device *bdev = I_BDEV(file->f_mapping->host);
1501    fmode_t mode = file->f_mode;
1502
1503    /*
1504     * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1505     * to updated it before every ioctl.
1506     */
1507    if (file->f_flags & O_NDELAY)
1508        mode |= FMODE_NDELAY;
1509    else
1510        mode &= ~FMODE_NDELAY;
1511
1512    return blkdev_ioctl(bdev, mode, cmd, arg);
1513}
1514
1515/*
1516 * Write data to the block device. Only intended for the block device itself
1517 * and the raw driver which basically is a fake block device.
1518 *
1519 * Does not take i_mutex for the write and thus is not for general purpose
1520 * use.
1521 */
1522ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1523             unsigned long nr_segs, loff_t pos)
1524{
1525    struct file *file = iocb->ki_filp;
1526    ssize_t ret;
1527
1528    BUG_ON(iocb->ki_pos != pos);
1529
1530    ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1531    if (ret > 0 || ret == -EIOCBQUEUED) {
1532        ssize_t err;
1533
1534        err = generic_write_sync(file, pos, ret);
1535        if (err < 0 && ret > 0)
1536            ret = err;
1537    }
1538    return ret;
1539}
1540EXPORT_SYMBOL_GPL(blkdev_aio_write);
1541
1542/*
1543 * Try to release a page associated with block device when the system
1544 * is under memory pressure.
1545 */
1546static int blkdev_releasepage(struct page *page, gfp_t wait)
1547{
1548    struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1549
1550    if (super && super->s_op->bdev_try_to_free_page)
1551        return super->s_op->bdev_try_to_free_page(super, page, wait);
1552
1553    return try_to_free_buffers(page);
1554}
1555
1556static const struct address_space_operations def_blk_aops = {
1557    .readpage = blkdev_readpage,
1558    .writepage = blkdev_writepage,
1559    .write_begin = blkdev_write_begin,
1560    .write_end = blkdev_write_end,
1561    .writepages = generic_writepages,
1562    .releasepage = blkdev_releasepage,
1563    .direct_IO = blkdev_direct_IO,
1564};
1565
1566const struct file_operations def_blk_fops = {
1567    .open = blkdev_open,
1568    .release = blkdev_close,
1569    .llseek = block_llseek,
1570    .read = do_sync_read,
1571    .write = do_sync_write,
1572      .aio_read = generic_file_aio_read,
1573    .aio_write = blkdev_aio_write,
1574    .mmap = generic_file_mmap,
1575    .fsync = blkdev_fsync,
1576    .unlocked_ioctl = block_ioctl,
1577#ifdef CONFIG_COMPAT
1578    .compat_ioctl = compat_blkdev_ioctl,
1579#endif
1580    .splice_read = generic_file_splice_read,
1581    .splice_write = generic_file_splice_write,
1582};
1583
1584int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1585{
1586    int res;
1587    mm_segment_t old_fs = get_fs();
1588    set_fs(KERNEL_DS);
1589    res = blkdev_ioctl(bdev, 0, cmd, arg);
1590    set_fs(old_fs);
1591    return res;
1592}
1593
1594EXPORT_SYMBOL(ioctl_by_bdev);
1595
1596/**
1597 * lookup_bdev - lookup a struct block_device by name
1598 * @pathname: special file representing the block device
1599 *
1600 * Get a reference to the blockdevice at @pathname in the current
1601 * namespace if possible and return it. Return ERR_PTR(error)
1602 * otherwise.
1603 */
1604struct block_device *lookup_bdev(const char *pathname)
1605{
1606    struct block_device *bdev;
1607    struct inode *inode;
1608    struct path path;
1609    int error;
1610
1611    if (!pathname || !*pathname)
1612        return ERR_PTR(-EINVAL);
1613
1614    error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1615    if (error)
1616        return ERR_PTR(error);
1617
1618    inode = path.dentry->d_inode;
1619    error = -ENOTBLK;
1620    if (!S_ISBLK(inode->i_mode))
1621        goto fail;
1622    error = -EACCES;
1623    if (path.mnt->mnt_flags & MNT_NODEV)
1624        goto fail;
1625    error = -ENOMEM;
1626    bdev = bd_acquire(inode);
1627    if (!bdev)
1628        goto fail;
1629out:
1630    path_put(&path);
1631    return bdev;
1632fail:
1633    bdev = ERR_PTR(error);
1634    goto out;
1635}
1636EXPORT_SYMBOL(lookup_bdev);
1637
1638int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1639{
1640    struct super_block *sb = get_super(bdev);
1641    int res = 0;
1642
1643    if (sb) {
1644        /*
1645         * no need to lock the super, get_super holds the
1646         * read mutex so the filesystem cannot go away
1647         * under us (->put_super runs with the write lock
1648         * hold).
1649         */
1650        shrink_dcache_sb(sb);
1651        res = invalidate_inodes(sb, kill_dirty);
1652        drop_super(sb);
1653    }
1654    invalidate_bdev(bdev);
1655    return res;
1656}
1657EXPORT_SYMBOL(__invalidate_device);
1658

Archive Download this file



interactive