~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/block_dev.c

Version: ~ [ linux-5.2-rc1 ] ~ [ linux-5.1.2 ] ~ [ linux-5.0.16 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.43 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.119 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.176 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.179 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.139 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.67 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 
 31 struct bdev_inode {
 32         struct block_device bdev;
 33         struct inode vfs_inode;
 34 };
 35 
 36 static const struct address_space_operations def_blk_aops;
 37 
 38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
 39 {
 40         return container_of(inode, struct bdev_inode, vfs_inode);
 41 }
 42 
 43 inline struct block_device *I_BDEV(struct inode *inode)
 44 {
 45         return &BDEV_I(inode)->bdev;
 46 }
 47 
 48 EXPORT_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  */
 55 static void bdev_inode_switch_bdi(struct inode *inode,
 56                         struct backing_dev_info *dst)
 57 {
 58         bool wakeup_bdi = false;
 59 
 60         spin_lock(&inode_wb_list_lock);
 61         spin_lock(&inode->i_lock);
 62         inode->i_data.backing_dev_info = dst;
 63         if (inode->i_state & I_DIRTY) {
 64                 if (bdi_cap_writeback_dirty(dst) && !wb_has_dirty_io(&dst->wb))
 65                         wakeup_bdi = true;
 66                 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
 67         }
 68         spin_unlock(&inode->i_lock);
 69         spin_unlock(&inode_wb_list_lock);
 70 
 71         if (wakeup_bdi)
 72                 bdi_wakeup_thread_delayed(dst);
 73 }
 74 
 75 sector_t blkdev_max_block(struct block_device *bdev)
 76 {
 77         sector_t retval = ~((sector_t)0);
 78         loff_t sz = i_size_read(bdev->bd_inode);
 79 
 80         if (sz) {
 81                 unsigned int size = block_size(bdev);
 82                 unsigned int sizebits = blksize_bits(size);
 83                 retval = (sz >> sizebits);
 84         }
 85         return retval;
 86 }
 87 
 88 /* Kill _all_ buffers and pagecache , dirty or not.. */
 89 static void kill_bdev(struct block_device *bdev)
 90 {
 91         if (bdev->bd_inode->i_mapping->nrpages == 0)
 92                 return;
 93         invalidate_bh_lrus();
 94         truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
 95 }       
 96 
 97 int set_blocksize(struct block_device *bdev, int size)
 98 {
 99         /* Size must be a power of two, and between 512 and PAGE_SIZE */
100         if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
101                 return -EINVAL;
102 
103         /* Size cannot be smaller than the size supported by the device */
104         if (size < bdev_logical_block_size(bdev))
105                 return -EINVAL;
106 
107         /* Don't change the size if it is same as current */
108         if (bdev->bd_block_size != size) {
109                 sync_blockdev(bdev);
110                 bdev->bd_block_size = size;
111                 bdev->bd_inode->i_blkbits = blksize_bits(size);
112                 kill_bdev(bdev);
113         }
114         return 0;
115 }
116 
117 EXPORT_SYMBOL(set_blocksize);
118 
119 int sb_set_blocksize(struct super_block *sb, int size)
120 {
121         if (set_blocksize(sb->s_bdev, size))
122                 return 0;
123         /* If we get here, we know size is power of two
124          * and it's value is between 512 and PAGE_SIZE */
125         sb->s_blocksize = size;
126         sb->s_blocksize_bits = blksize_bits(size);
127         return sb->s_blocksize;
128 }
129 
130 EXPORT_SYMBOL(sb_set_blocksize);
131 
132 int sb_min_blocksize(struct super_block *sb, int size)
133 {
134         int minsize = bdev_logical_block_size(sb->s_bdev);
135         if (size < minsize)
136                 size = minsize;
137         return sb_set_blocksize(sb, size);
138 }
139 
140 EXPORT_SYMBOL(sb_min_blocksize);
141 
142 static int
143 blkdev_get_block(struct inode *inode, sector_t iblock,
144                 struct buffer_head *bh, int create)
145 {
146         if (iblock >= blkdev_max_block(I_BDEV(inode))) {
147                 if (create)
148                         return -EIO;
149 
150                 /*
151                  * for reads, we're just trying to fill a partial page.
152                  * return a hole, they will have to call get_block again
153                  * before they can fill it, and they will get -EIO at that
154                  * time
155                  */
156                 return 0;
157         }
158         bh->b_bdev = I_BDEV(inode);
159         bh->b_blocknr = iblock;
160         set_buffer_mapped(bh);
161         return 0;
162 }
163 
164 static int
165 blkdev_get_blocks(struct inode *inode, sector_t iblock,
166                 struct buffer_head *bh, int create)
167 {
168         sector_t end_block = blkdev_max_block(I_BDEV(inode));
169         unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
170 
171         if ((iblock + max_blocks) > end_block) {
172                 max_blocks = end_block - iblock;
173                 if ((long)max_blocks <= 0) {
174                         if (create)
175                                 return -EIO;    /* write fully beyond EOF */
176                         /*
177                          * It is a read which is fully beyond EOF.  We return
178                          * a !buffer_mapped buffer
179                          */
180                         max_blocks = 0;
181                 }
182         }
183 
184         bh->b_bdev = I_BDEV(inode);
185         bh->b_blocknr = iblock;
186         bh->b_size = max_blocks << inode->i_blkbits;
187         if (max_blocks)
188                 set_buffer_mapped(bh);
189         return 0;
190 }
191 
192 static ssize_t
193 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
194                         loff_t offset, unsigned long nr_segs)
195 {
196         struct file *file = iocb->ki_filp;
197         struct inode *inode = file->f_mapping->host;
198 
199         return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
200                                     nr_segs, blkdev_get_blocks, NULL, NULL, 0);
201 }
202 
203 int __sync_blockdev(struct block_device *bdev, int wait)
204 {
205         if (!bdev)
206                 return 0;
207         if (!wait)
208                 return filemap_flush(bdev->bd_inode->i_mapping);
209         return filemap_write_and_wait(bdev->bd_inode->i_mapping);
210 }
211 
212 /*
213  * Write out and wait upon all the dirty data associated with a block
214  * device via its mapping.  Does not take the superblock lock.
215  */
216 int sync_blockdev(struct block_device *bdev)
217 {
218         return __sync_blockdev(bdev, 1);
219 }
220 EXPORT_SYMBOL(sync_blockdev);
221 
222 /*
223  * Write out and wait upon all dirty data associated with this
224  * device.   Filesystem data as well as the underlying block
225  * device.  Takes the superblock lock.
226  */
227 int fsync_bdev(struct block_device *bdev)
228 {
229         struct super_block *sb = get_super(bdev);
230         if (sb) {
231                 int res = sync_filesystem(sb);
232                 drop_super(sb);
233                 return res;
234         }
235         return sync_blockdev(bdev);
236 }
237 EXPORT_SYMBOL(fsync_bdev);
238 
239 /**
240  * freeze_bdev  --  lock a filesystem and force it into a consistent state
241  * @bdev:       blockdevice to lock
242  *
243  * If a superblock is found on this device, we take the s_umount semaphore
244  * on it to make sure nobody unmounts until the snapshot creation is done.
245  * The reference counter (bd_fsfreeze_count) guarantees that only the last
246  * unfreeze process can unfreeze the frozen filesystem actually when multiple
247  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
248  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
249  * actually.
250  */
251 struct super_block *freeze_bdev(struct block_device *bdev)
252 {
253         struct super_block *sb;
254         int error = 0;
255 
256         mutex_lock(&bdev->bd_fsfreeze_mutex);
257         if (++bdev->bd_fsfreeze_count > 1) {
258                 /*
259                  * We don't even need to grab a reference - the first call
260                  * to freeze_bdev grab an active reference and only the last
261                  * thaw_bdev drops it.
262                  */
263                 sb = get_super(bdev);
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         error = freeze_super(sb);
273         if (error) {
274                 deactivate_super(sb);
275                 bdev->bd_fsfreeze_count--;
276                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
277                 return ERR_PTR(error);
278         }
279         deactivate_super(sb);
280  out:
281         sync_blockdev(bdev);
282         mutex_unlock(&bdev->bd_fsfreeze_mutex);
283         return sb;      /* thaw_bdev releases s->s_umount */
284 }
285 EXPORT_SYMBOL(freeze_bdev);
286 
287 /**
288  * thaw_bdev  -- unlock filesystem
289  * @bdev:       blockdevice to unlock
290  * @sb:         associated superblock
291  *
292  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
293  */
294 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
295 {
296         int error = -EINVAL;
297 
298         mutex_lock(&bdev->bd_fsfreeze_mutex);
299         if (!bdev->bd_fsfreeze_count)
300                 goto out;
301 
302         error = 0;
303         if (--bdev->bd_fsfreeze_count > 0)
304                 goto out;
305 
306         if (!sb)
307                 goto out;
308 
309         error = thaw_super(sb);
310         if (error) {
311                 bdev->bd_fsfreeze_count++;
312                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
313                 return error;
314         }
315 out:
316         mutex_unlock(&bdev->bd_fsfreeze_mutex);
317         return 0;
318 }
319 EXPORT_SYMBOL(thaw_bdev);
320 
321 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
322 {
323         return block_write_full_page(page, blkdev_get_block, wbc);
324 }
325 
326 static int blkdev_readpage(struct file * file, struct page * page)
327 {
328         return block_read_full_page(page, blkdev_get_block);
329 }
330 
331 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
332                         loff_t pos, unsigned len, unsigned flags,
333                         struct page **pagep, void **fsdata)
334 {
335         return block_write_begin(mapping, pos, len, flags, pagep,
336                                  blkdev_get_block);
337 }
338 
339 static int blkdev_write_end(struct file *file, struct address_space *mapping,
340                         loff_t pos, unsigned len, unsigned copied,
341                         struct page *page, void *fsdata)
342 {
343         int ret;
344         ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
345 
346         unlock_page(page);
347         page_cache_release(page);
348 
349         return ret;
350 }
351 
352 /*
353  * private llseek:
354  * for a block special file file->f_path.dentry->d_inode->i_size is zero
355  * so we compute the size by hand (just as in block_read/write above)
356  */
357 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
358 {
359         struct inode *bd_inode = file->f_mapping->host;
360         loff_t size;
361         loff_t retval;
362 
363         mutex_lock(&bd_inode->i_mutex);
364         size = i_size_read(bd_inode);
365 
366         switch (origin) {
367                 case 2:
368                         offset += size;
369                         break;
370                 case 1:
371                         offset += file->f_pos;
372         }
373         retval = -EINVAL;
374         if (offset >= 0 && offset <= size) {
375                 if (offset != file->f_pos) {
376                         file->f_pos = offset;
377                 }
378                 retval = offset;
379         }
380         mutex_unlock(&bd_inode->i_mutex);
381         return retval;
382 }
383         
384 int blkdev_fsync(struct file *filp, int datasync)
385 {
386         struct inode *bd_inode = filp->f_mapping->host;
387         struct block_device *bdev = I_BDEV(bd_inode);
388         int error;
389 
390         /*
391          * There is no need to serialise calls to blkdev_issue_flush with
392          * i_mutex and doing so causes performance issues with concurrent
393          * O_SYNC writers to a block device.
394          */
395         mutex_unlock(&bd_inode->i_mutex);
396 
397         error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
398         if (error == -EOPNOTSUPP)
399                 error = 0;
400 
401         mutex_lock(&bd_inode->i_mutex);
402 
403         return error;
404 }
405 EXPORT_SYMBOL(blkdev_fsync);
406 
407 /*
408  * pseudo-fs
409  */
410 
411 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
412 static struct kmem_cache * bdev_cachep __read_mostly;
413 
414 static struct inode *bdev_alloc_inode(struct super_block *sb)
415 {
416         struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
417         if (!ei)
418                 return NULL;
419         return &ei->vfs_inode;
420 }
421 
422 static void bdev_i_callback(struct rcu_head *head)
423 {
424         struct inode *inode = container_of(head, struct inode, i_rcu);
425         struct bdev_inode *bdi = BDEV_I(inode);
426 
427         INIT_LIST_HEAD(&inode->i_dentry);
428         kmem_cache_free(bdev_cachep, bdi);
429 }
430 
431 static void bdev_destroy_inode(struct inode *inode)
432 {
433         call_rcu(&inode->i_rcu, bdev_i_callback);
434 }
435 
436 static void init_once(void *foo)
437 {
438         struct bdev_inode *ei = (struct bdev_inode *) foo;
439         struct block_device *bdev = &ei->bdev;
440 
441         memset(bdev, 0, sizeof(*bdev));
442         mutex_init(&bdev->bd_mutex);
443         INIT_LIST_HEAD(&bdev->bd_inodes);
444         INIT_LIST_HEAD(&bdev->bd_list);
445 #ifdef CONFIG_SYSFS
446         INIT_LIST_HEAD(&bdev->bd_holder_disks);
447 #endif
448         inode_init_once(&ei->vfs_inode);
449         /* Initialize mutex for freeze. */
450         mutex_init(&bdev->bd_fsfreeze_mutex);
451 }
452 
453 static inline void __bd_forget(struct inode *inode)
454 {
455         list_del_init(&inode->i_devices);
456         inode->i_bdev = NULL;
457         inode->i_mapping = &inode->i_data;
458 }
459 
460 static void bdev_evict_inode(struct inode *inode)
461 {
462         struct block_device *bdev = &BDEV_I(inode)->bdev;
463         struct list_head *p;
464         truncate_inode_pages(&inode->i_data, 0);
465         invalidate_inode_buffers(inode); /* is it needed here? */
466         end_writeback(inode);
467         spin_lock(&bdev_lock);
468         while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
469                 __bd_forget(list_entry(p, struct inode, i_devices));
470         }
471         list_del_init(&bdev->bd_list);
472         spin_unlock(&bdev_lock);
473 }
474 
475 static const struct super_operations bdev_sops = {
476         .statfs = simple_statfs,
477         .alloc_inode = bdev_alloc_inode,
478         .destroy_inode = bdev_destroy_inode,
479         .drop_inode = generic_delete_inode,
480         .evict_inode = bdev_evict_inode,
481 };
482 
483 static struct dentry *bd_mount(struct file_system_type *fs_type,
484         int flags, const char *dev_name, void *data)
485 {
486         return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
487 }
488 
489 static struct file_system_type bd_type = {
490         .name           = "bdev",
491         .mount          = bd_mount,
492         .kill_sb        = kill_anon_super,
493 };
494 
495 struct super_block *blockdev_superblock __read_mostly;
496 
497 void __init bdev_cache_init(void)
498 {
499         int err;
500         struct vfsmount *bd_mnt;
501 
502         bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
503                         0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
504                                 SLAB_MEM_SPREAD|SLAB_PANIC),
505                         init_once);
506         err = register_filesystem(&bd_type);
507         if (err)
508                 panic("Cannot register bdev pseudo-fs");
509         bd_mnt = kern_mount(&bd_type);
510         if (IS_ERR(bd_mnt))
511                 panic("Cannot create bdev pseudo-fs");
512         /*
513          * This vfsmount structure is only used to obtain the
514          * blockdev_superblock, so tell kmemleak not to report it.
515          */
516         kmemleak_not_leak(bd_mnt);
517         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
518 }
519 
520 /*
521  * Most likely _very_ bad one - but then it's hardly critical for small
522  * /dev and can be fixed when somebody will need really large one.
523  * Keep in mind that it will be fed through icache hash function too.
524  */
525 static inline unsigned long hash(dev_t dev)
526 {
527         return MAJOR(dev)+MINOR(dev);
528 }
529 
530 static int bdev_test(struct inode *inode, void *data)
531 {
532         return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
533 }
534 
535 static int bdev_set(struct inode *inode, void *data)
536 {
537         BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
538         return 0;
539 }
540 
541 static LIST_HEAD(all_bdevs);
542 
543 struct block_device *bdget(dev_t dev)
544 {
545         struct block_device *bdev;
546         struct inode *inode;
547 
548         inode = iget5_locked(blockdev_superblock, hash(dev),
549                         bdev_test, bdev_set, &dev);
550 
551         if (!inode)
552                 return NULL;
553 
554         bdev = &BDEV_I(inode)->bdev;
555 
556         if (inode->i_state & I_NEW) {
557                 bdev->bd_contains = NULL;
558                 bdev->bd_inode = inode;
559                 bdev->bd_block_size = (1 << inode->i_blkbits);
560                 bdev->bd_part_count = 0;
561                 bdev->bd_invalidated = 0;
562                 inode->i_mode = S_IFBLK;
563                 inode->i_rdev = dev;
564                 inode->i_bdev = bdev;
565                 inode->i_data.a_ops = &def_blk_aops;
566                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
567                 inode->i_data.backing_dev_info = &default_backing_dev_info;
568                 spin_lock(&bdev_lock);
569                 list_add(&bdev->bd_list, &all_bdevs);
570                 spin_unlock(&bdev_lock);
571                 unlock_new_inode(inode);
572         }
573         return bdev;
574 }
575 
576 EXPORT_SYMBOL(bdget);
577 
578 /**
579  * bdgrab -- Grab a reference to an already referenced block device
580  * @bdev:       Block device to grab a reference to.
581  */
582 struct block_device *bdgrab(struct block_device *bdev)
583 {
584         ihold(bdev->bd_inode);
585         return bdev;
586 }
587 EXPORT_SYMBOL(bdgrab);
588 
589 long nr_blockdev_pages(void)
590 {
591         struct block_device *bdev;
592         long ret = 0;
593         spin_lock(&bdev_lock);
594         list_for_each_entry(bdev, &all_bdevs, bd_list) {
595                 ret += bdev->bd_inode->i_mapping->nrpages;
596         }
597         spin_unlock(&bdev_lock);
598         return ret;
599 }
600 
601 void bdput(struct block_device *bdev)
602 {
603         iput(bdev->bd_inode);
604 }
605 
606 EXPORT_SYMBOL(bdput);
607  
608 static struct block_device *bd_acquire(struct inode *inode)
609 {
610         struct block_device *bdev;
611 
612         spin_lock(&bdev_lock);
613         bdev = inode->i_bdev;
614         if (bdev) {
615                 ihold(bdev->bd_inode);
616                 spin_unlock(&bdev_lock);
617                 return bdev;
618         }
619         spin_unlock(&bdev_lock);
620 
621         bdev = bdget(inode->i_rdev);
622         if (bdev) {
623                 spin_lock(&bdev_lock);
624                 if (!inode->i_bdev) {
625                         /*
626                          * We take an additional reference to bd_inode,
627                          * and it's released in clear_inode() of inode.
628                          * So, we can access it via ->i_mapping always
629                          * without igrab().
630                          */
631                         ihold(bdev->bd_inode);
632                         inode->i_bdev = bdev;
633                         inode->i_mapping = bdev->bd_inode->i_mapping;
634                         list_add(&inode->i_devices, &bdev->bd_inodes);
635                 }
636                 spin_unlock(&bdev_lock);
637         }
638         return bdev;
639 }
640 
641 /* Call when you free inode */
642 
643 void bd_forget(struct inode *inode)
644 {
645         struct block_device *bdev = NULL;
646 
647         spin_lock(&bdev_lock);
648         if (inode->i_bdev) {
649                 if (!sb_is_blkdev_sb(inode->i_sb))
650                         bdev = inode->i_bdev;
651                 __bd_forget(inode);
652         }
653         spin_unlock(&bdev_lock);
654 
655         if (bdev)
656                 iput(bdev->bd_inode);
657 }
658 
659 /**
660  * bd_may_claim - test whether a block device can be claimed
661  * @bdev: block device of interest
662  * @whole: whole block device containing @bdev, may equal @bdev
663  * @holder: holder trying to claim @bdev
664  *
665  * Test whether @bdev can be claimed by @holder.
666  *
667  * CONTEXT:
668  * spin_lock(&bdev_lock).
669  *
670  * RETURNS:
671  * %true if @bdev can be claimed, %false otherwise.
672  */
673 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
674                          void *holder)
675 {
676         if (bdev->bd_holder == holder)
677                 return true;     /* already a holder */
678         else if (bdev->bd_holder != NULL)
679                 return false;    /* held by someone else */
680         else if (bdev->bd_contains == bdev)
681                 return true;     /* is a whole device which isn't held */
682 
683         else if (whole->bd_holder == bd_may_claim)
684                 return true;     /* is a partition of a device that is being partitioned */
685         else if (whole->bd_holder != NULL)
686                 return false;    /* is a partition of a held device */
687         else
688                 return true;     /* is a partition of an un-held device */
689 }
690 
691 /**
692  * bd_prepare_to_claim - prepare to claim a block device
693  * @bdev: block device of interest
694  * @whole: the whole device containing @bdev, may equal @bdev
695  * @holder: holder trying to claim @bdev
696  *
697  * Prepare to claim @bdev.  This function fails if @bdev is already
698  * claimed by another holder and waits if another claiming is in
699  * progress.  This function doesn't actually claim.  On successful
700  * return, the caller has ownership of bd_claiming and bd_holder[s].
701  *
702  * CONTEXT:
703  * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
704  * it multiple times.
705  *
706  * RETURNS:
707  * 0 if @bdev can be claimed, -EBUSY otherwise.
708  */
709 static int bd_prepare_to_claim(struct block_device *bdev,
710                                struct block_device *whole, void *holder)
711 {
712 retry:
713         /* if someone else claimed, fail */
714         if (!bd_may_claim(bdev, whole, holder))
715                 return -EBUSY;
716 
717         /* if claiming is already in progress, wait for it to finish */
718         if (whole->bd_claiming) {
719                 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
720                 DEFINE_WAIT(wait);
721 
722                 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
723                 spin_unlock(&bdev_lock);
724                 schedule();
725                 finish_wait(wq, &wait);
726                 spin_lock(&bdev_lock);
727                 goto retry;
728         }
729 
730         /* yay, all mine */
731         return 0;
732 }
733 
734 /**
735  * bd_start_claiming - start claiming a block device
736  * @bdev: block device of interest
737  * @holder: holder trying to claim @bdev
738  *
739  * @bdev is about to be opened exclusively.  Check @bdev can be opened
740  * exclusively and mark that an exclusive open is in progress.  Each
741  * successful call to this function must be matched with a call to
742  * either bd_finish_claiming() or bd_abort_claiming() (which do not
743  * fail).
744  *
745  * This function is used to gain exclusive access to the block device
746  * without actually causing other exclusive open attempts to fail. It
747  * should be used when the open sequence itself requires exclusive
748  * access but may subsequently fail.
749  *
750  * CONTEXT:
751  * Might sleep.
752  *
753  * RETURNS:
754  * Pointer to the block device containing @bdev on success, ERR_PTR()
755  * value on failure.
756  */
757 static struct block_device *bd_start_claiming(struct block_device *bdev,
758                                               void *holder)
759 {
760         struct gendisk *disk;
761         struct block_device *whole;
762         int partno, err;
763 
764         might_sleep();
765 
766         /*
767          * @bdev might not have been initialized properly yet, look up
768          * and grab the outer block device the hard way.
769          */
770         disk = get_gendisk(bdev->bd_dev, &partno);
771         if (!disk)
772                 return ERR_PTR(-ENXIO);
773 
774         /*
775          * Normally, @bdev should equal what's returned from bdget_disk()
776          * if partno is 0; however, some drivers (floppy) use multiple
777          * bdev's for the same physical device and @bdev may be one of the
778          * aliases.  Keep @bdev if partno is 0.  This means claimer
779          * tracking is broken for those devices but it has always been that
780          * way.
781          */
782         if (partno)
783                 whole = bdget_disk(disk, 0);
784         else
785                 whole = bdgrab(bdev);
786 
787         module_put(disk->fops->owner);
788         put_disk(disk);
789         if (!whole)
790                 return ERR_PTR(-ENOMEM);
791 
792         /* prepare to claim, if successful, mark claiming in progress */
793         spin_lock(&bdev_lock);
794 
795         err = bd_prepare_to_claim(bdev, whole, holder);
796         if (err == 0) {
797                 whole->bd_claiming = holder;
798                 spin_unlock(&bdev_lock);
799                 return whole;
800         } else {
801                 spin_unlock(&bdev_lock);
802                 bdput(whole);
803                 return ERR_PTR(err);
804         }
805 }
806 
807 #ifdef CONFIG_SYSFS
808 struct bd_holder_disk {
809         struct list_head        list;
810         struct gendisk          *disk;
811         int                     refcnt;
812 };
813 
814 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
815                                                   struct gendisk *disk)
816 {
817         struct bd_holder_disk *holder;
818 
819         list_for_each_entry(holder, &bdev->bd_holder_disks, list)
820                 if (holder->disk == disk)
821                         return holder;
822         return NULL;
823 }
824 
825 static int add_symlink(struct kobject *from, struct kobject *to)
826 {
827         return sysfs_create_link(from, to, kobject_name(to));
828 }
829 
830 static void del_symlink(struct kobject *from, struct kobject *to)
831 {
832         sysfs_remove_link(from, kobject_name(to));
833 }
834 
835 /**
836  * bd_link_disk_holder - create symlinks between holding disk and slave bdev
837  * @bdev: the claimed slave bdev
838  * @disk: the holding disk
839  *
840  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
841  *
842  * This functions creates the following sysfs symlinks.
843  *
844  * - from "slaves" directory of the holder @disk to the claimed @bdev
845  * - from "holders" directory of the @bdev to the holder @disk
846  *
847  * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
848  * passed to bd_link_disk_holder(), then:
849  *
850  *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
851  *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
852  *
853  * The caller must have claimed @bdev before calling this function and
854  * ensure that both @bdev and @disk are valid during the creation and
855  * lifetime of these symlinks.
856  *
857  * CONTEXT:
858  * Might sleep.
859  *
860  * RETURNS:
861  * 0 on success, -errno on failure.
862  */
863 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
864 {
865         struct bd_holder_disk *holder;
866         int ret = 0;
867 
868         mutex_lock(&bdev->bd_mutex);
869 
870         WARN_ON_ONCE(!bdev->bd_holder);
871 
872         /* FIXME: remove the following once add_disk() handles errors */
873         if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
874                 goto out_unlock;
875 
876         holder = bd_find_holder_disk(bdev, disk);
877         if (holder) {
878                 holder->refcnt++;
879                 goto out_unlock;
880         }
881 
882         holder = kzalloc(sizeof(*holder), GFP_KERNEL);
883         if (!holder) {
884                 ret = -ENOMEM;
885                 goto out_unlock;
886         }
887 
888         INIT_LIST_HEAD(&holder->list);
889         holder->disk = disk;
890         holder->refcnt = 1;
891 
892         ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
893         if (ret)
894                 goto out_free;
895 
896         ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
897         if (ret)
898                 goto out_del;
899         /*
900          * bdev could be deleted beneath us which would implicitly destroy
901          * the holder directory.  Hold on to it.
902          */
903         kobject_get(bdev->bd_part->holder_dir);
904 
905         list_add(&holder->list, &bdev->bd_holder_disks);
906         goto out_unlock;
907 
908 out_del:
909         del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
910 out_free:
911         kfree(holder);
912 out_unlock:
913         mutex_unlock(&bdev->bd_mutex);
914         return ret;
915 }
916 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
917 
918 /**
919  * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
920  * @bdev: the calimed slave bdev
921  * @disk: the holding disk
922  *
923  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
924  *
925  * CONTEXT:
926  * Might sleep.
927  */
928 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
929 {
930         struct bd_holder_disk *holder;
931 
932         mutex_lock(&bdev->bd_mutex);
933 
934         holder = bd_find_holder_disk(bdev, disk);
935 
936         if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
937                 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
938                 del_symlink(bdev->bd_part->holder_dir,
939                             &disk_to_dev(disk)->kobj);
940                 kobject_put(bdev->bd_part->holder_dir);
941                 list_del_init(&holder->list);
942                 kfree(holder);
943         }
944 
945         mutex_unlock(&bdev->bd_mutex);
946 }
947 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
948 #endif
949 
950 /**
951  * flush_disk - invalidates all buffer-cache entries on a disk
952  *
953  * @bdev:      struct block device to be flushed
954  * @kill_dirty: flag to guide handling of dirty inodes
955  *
956  * Invalidates all buffer-cache entries on a disk. It should be called
957  * when a disk has been changed -- either by a media change or online
958  * resize.
959  */
960 static void flush_disk(struct block_device *bdev, bool kill_dirty)
961 {
962         if (__invalidate_device(bdev, kill_dirty)) {
963                 char name[BDEVNAME_SIZE] = "";
964 
965                 if (bdev->bd_disk)
966                         disk_name(bdev->bd_disk, 0, name);
967                 printk(KERN_WARNING "VFS: busy inodes on changed media or "
968                        "resized disk %s\n", name);
969         }
970 
971         if (!bdev->bd_disk)
972                 return;
973         if (disk_partitionable(bdev->bd_disk))
974                 bdev->bd_invalidated = 1;
975 }
976 
977 /**
978  * check_disk_size_change - checks for disk size change and adjusts bdev size.
979  * @disk: struct gendisk to check
980  * @bdev: struct bdev to adjust.
981  *
982  * This routine checks to see if the bdev size does not match the disk size
983  * and adjusts it if it differs.
984  */
985 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
986 {
987         loff_t disk_size, bdev_size;
988 
989         disk_size = (loff_t)get_capacity(disk) << 9;
990         bdev_size = i_size_read(bdev->bd_inode);
991         if (disk_size != bdev_size) {
992                 char name[BDEVNAME_SIZE];
993 
994                 disk_name(disk, 0, name);
995                 printk(KERN_INFO
996                        "%s: detected capacity change from %lld to %lld\n",
997                        name, bdev_size, disk_size);
998                 i_size_write(bdev->bd_inode, disk_size);
999                 flush_disk(bdev, false);
1000         }
1001 }
1002 EXPORT_SYMBOL(check_disk_size_change);
1003 
1004 /**
1005  * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1006  * @disk: struct gendisk to be revalidated
1007  *
1008  * This routine is a wrapper for lower-level driver's revalidate_disk
1009  * call-backs.  It is used to do common pre and post operations needed
1010  * for all revalidate_disk operations.
1011  */
1012 int revalidate_disk(struct gendisk *disk)
1013 {
1014         struct block_device *bdev;
1015         int ret = 0;
1016 
1017         if (disk->fops->revalidate_disk)
1018                 ret = disk->fops->revalidate_disk(disk);
1019 
1020         bdev = bdget_disk(disk, 0);
1021         if (!bdev)
1022                 return ret;
1023 
1024         mutex_lock(&bdev->bd_mutex);
1025         check_disk_size_change(disk, bdev);
1026         mutex_unlock(&bdev->bd_mutex);
1027         bdput(bdev);
1028         return ret;
1029 }
1030 EXPORT_SYMBOL(revalidate_disk);
1031 
1032 /*
1033  * This routine checks whether a removable media has been changed,
1034  * and invalidates all buffer-cache-entries in that case. This
1035  * is a relatively slow routine, so we have to try to minimize using
1036  * it. Thus it is called only upon a 'mount' or 'open'. This
1037  * is the best way of combining speed and utility, I think.
1038  * People changing diskettes in the middle of an operation deserve
1039  * to lose :-)
1040  */
1041 int check_disk_change(struct block_device *bdev)
1042 {
1043         struct gendisk *disk = bdev->bd_disk;
1044         const struct block_device_operations *bdops = disk->fops;
1045         unsigned int events;
1046 
1047         events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1048                                    DISK_EVENT_EJECT_REQUEST);
1049         if (!(events & DISK_EVENT_MEDIA_CHANGE))
1050                 return 0;
1051 
1052         flush_disk(bdev, true);
1053         if (bdops->revalidate_disk)
1054                 bdops->revalidate_disk(bdev->bd_disk);
1055         return 1;
1056 }
1057 
1058 EXPORT_SYMBOL(check_disk_change);
1059 
1060 void bd_set_size(struct block_device *bdev, loff_t size)
1061 {
1062         unsigned bsize = bdev_logical_block_size(bdev);
1063 
1064         mutex_lock(&bdev->bd_inode->i_mutex);
1065         i_size_write(bdev->bd_inode, size);
1066         mutex_unlock(&bdev->bd_inode->i_mutex);
1067         while (bsize < PAGE_CACHE_SIZE) {
1068                 if (size & bsize)
1069                         break;
1070                 bsize <<= 1;
1071         }
1072         bdev->bd_block_size = bsize;
1073         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1074 }
1075 EXPORT_SYMBOL(bd_set_size);
1076 
1077 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1078 
1079 /*
1080  * bd_mutex locking:
1081  *
1082  *  mutex_lock(part->bd_mutex)
1083  *    mutex_lock_nested(whole->bd_mutex, 1)
1084  */
1085 
1086 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1087 {
1088         struct gendisk *disk;
1089         struct module *owner;
1090         int ret;
1091         int partno;
1092         int perm = 0;
1093 
1094         if (mode & FMODE_READ)
1095                 perm |= MAY_READ;
1096         if (mode & FMODE_WRITE)
1097                 perm |= MAY_WRITE;
1098         /*
1099          * hooks: /n/, see "layering violations".
1100          */
1101         if (!for_part) {
1102                 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1103                 if (ret != 0) {
1104                         bdput(bdev);
1105                         return ret;
1106                 }
1107         }
1108 
1109  restart:
1110 
1111         ret = -ENXIO;
1112         disk = get_gendisk(bdev->bd_dev, &partno);
1113         if (!disk)
1114                 goto out;
1115         owner = disk->fops->owner;
1116 
1117         disk_block_events(disk);
1118         mutex_lock_nested(&bdev->bd_mutex, for_part);
1119         if (!bdev->bd_openers) {
1120                 bdev->bd_disk = disk;
1121                 bdev->bd_contains = bdev;
1122                 if (!partno) {
1123                         struct backing_dev_info *bdi;
1124 
1125                         ret = -ENXIO;
1126                         bdev->bd_part = disk_get_part(disk, partno);
1127                         if (!bdev->bd_part)
1128                                 goto out_clear;
1129 
1130                         ret = 0;
1131                         if (disk->fops->open) {
1132                                 ret = disk->fops->open(bdev, mode);
1133                                 if (ret == -ERESTARTSYS) {
1134                                         /* Lost a race with 'disk' being
1135                                          * deleted, try again.
1136                                          * See md.c
1137                                          */
1138                                         disk_put_part(bdev->bd_part);
1139                                         bdev->bd_part = NULL;
1140                                         bdev->bd_disk = NULL;
1141                                         mutex_unlock(&bdev->bd_mutex);
1142                                         disk_unblock_events(disk);
1143                                         put_disk(disk);
1144                                         module_put(owner);
1145                                         goto restart;
1146                                 }
1147                         }
1148 
1149                         if (!ret && !bdev->bd_openers) {
1150                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1151                                 bdi = blk_get_backing_dev_info(bdev);
1152                                 if (bdi == NULL)
1153                                         bdi = &default_backing_dev_info;
1154                                 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1155                         }
1156 
1157                         /*
1158                          * If the device is invalidated, rescan partition
1159                          * if open succeeded or failed with -ENOMEDIUM.
1160                          * The latter is necessary to prevent ghost
1161                          * partitions on a removed medium.
1162                          */
1163                         if (bdev->bd_invalidated) {
1164                                 if (!ret)
1165                                         rescan_partitions(disk, bdev);
1166                                 else if (ret == -ENOMEDIUM)
1167                                         invalidate_partitions(disk, bdev);
1168                         }
1169                         if (ret)
1170                                 goto out_clear;
1171                 } else {
1172                         struct block_device *whole;
1173                         whole = bdget_disk(disk, 0);
1174                         ret = -ENOMEM;
1175                         if (!whole)
1176                                 goto out_clear;
1177                         BUG_ON(for_part);
1178                         ret = __blkdev_get(whole, mode, 1);
1179                         if (ret)
1180                                 goto out_clear;
1181                         bdev->bd_contains = whole;
1182                         bdev_inode_switch_bdi(bdev->bd_inode,
1183                                 whole->bd_inode->i_data.backing_dev_info);
1184                         bdev->bd_part = disk_get_part(disk, partno);
1185                         if (!(disk->flags & GENHD_FL_UP) ||
1186                             !bdev->bd_part || !bdev->bd_part->nr_sects) {
1187                                 ret = -ENXIO;
1188                                 goto out_clear;
1189                         }
1190                         bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1191                 }
1192         } else {
1193                 if (bdev->bd_contains == bdev) {
1194                         ret = 0;
1195                         if (bdev->bd_disk->fops->open)
1196                                 ret = bdev->bd_disk->fops->open(bdev, mode);
1197                         /* the same as first opener case, read comment there */
1198                         if (bdev->bd_invalidated) {
1199                                 if (!ret)
1200                                         rescan_partitions(bdev->bd_disk, bdev);
1201                                 else if (ret == -ENOMEDIUM)
1202                                         invalidate_partitions(bdev->bd_disk, bdev);
1203                         }
1204                         if (ret)
1205                                 goto out_unlock_bdev;
1206                 }
1207                 /* only one opener holds refs to the module and disk */
1208                 put_disk(disk);
1209                 module_put(owner);
1210         }
1211         bdev->bd_openers++;
1212         if (for_part)
1213                 bdev->bd_part_count++;
1214         mutex_unlock(&bdev->bd_mutex);
1215         disk_unblock_events(disk);
1216         return 0;
1217 
1218  out_clear:
1219         disk_put_part(bdev->bd_part);
1220         bdev->bd_disk = NULL;
1221         bdev->bd_part = NULL;
1222         bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1223         if (bdev != bdev->bd_contains)
1224                 __blkdev_put(bdev->bd_contains, mode, 1);
1225         bdev->bd_contains = NULL;
1226  out_unlock_bdev:
1227         mutex_unlock(&bdev->bd_mutex);
1228         disk_unblock_events(disk);
1229         put_disk(disk);
1230         module_put(owner);
1231  out:
1232         bdput(bdev);
1233 
1234         return ret;
1235 }
1236 
1237 /**
1238  * blkdev_get - open a block device
1239  * @bdev: block_device to open
1240  * @mode: FMODE_* mask
1241  * @holder: exclusive holder identifier
1242  *
1243  * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1244  * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1245  * @holder is invalid.  Exclusive opens may nest for the same @holder.
1246  *
1247  * On success, the reference count of @bdev is unchanged.  On failure,
1248  * @bdev is put.
1249  *
1250  * CONTEXT:
1251  * Might sleep.
1252  *
1253  * RETURNS:
1254  * 0 on success, -errno on failure.
1255  */
1256 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1257 {
1258         struct block_device *whole = NULL;
1259         int res;
1260 
1261         WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1262 
1263         if ((mode & FMODE_EXCL) && holder) {
1264                 whole = bd_start_claiming(bdev, holder);
1265                 if (IS_ERR(whole)) {
1266                         bdput(bdev);
1267                         return PTR_ERR(whole);
1268                 }
1269         }
1270 
1271         res = __blkdev_get(bdev, mode, 0);
1272 
1273         if (whole) {
1274                 struct gendisk *disk = whole->bd_disk;
1275 
1276                 /* finish claiming */
1277                 mutex_lock(&bdev->bd_mutex);
1278                 spin_lock(&bdev_lock);
1279 
1280                 if (!res) {
1281                         BUG_ON(!bd_may_claim(bdev, whole, holder));
1282                         /*
1283                          * Note that for a whole device bd_holders
1284                          * will be incremented twice, and bd_holder
1285                          * will be set to bd_may_claim before being
1286                          * set to holder
1287                          */
1288                         whole->bd_holders++;
1289                         whole->bd_holder = bd_may_claim;
1290                         bdev->bd_holders++;
1291                         bdev->bd_holder = holder;
1292                 }
1293 
1294                 /* tell others that we're done */
1295                 BUG_ON(whole->bd_claiming != holder);
1296                 whole->bd_claiming = NULL;
1297                 wake_up_bit(&whole->bd_claiming, 0);
1298 
1299                 spin_unlock(&bdev_lock);
1300 
1301                 /*
1302                  * Block event polling for write claims if requested.  Any
1303                  * write holder makes the write_holder state stick until
1304                  * all are released.  This is good enough and tracking
1305                  * individual writeable reference is too fragile given the
1306                  * way @mode is used in blkdev_get/put().
1307                  */
1308                 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1309                     (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1310                         bdev->bd_write_holder = true;
1311                         disk_block_events(disk);
1312                 }
1313 
1314                 mutex_unlock(&bdev->bd_mutex);
1315                 bdput(whole);
1316         }
1317 
1318         return res;
1319 }
1320 EXPORT_SYMBOL(blkdev_get);
1321 
1322 /**
1323  * blkdev_get_by_path - open a block device by name
1324  * @path: path to the block device to open
1325  * @mode: FMODE_* mask
1326  * @holder: exclusive holder identifier
1327  *
1328  * Open the blockdevice described by the device file at @path.  @mode
1329  * and @holder are identical to blkdev_get().
1330  *
1331  * On success, the returned block_device has reference count of one.
1332  *
1333  * CONTEXT:
1334  * Might sleep.
1335  *
1336  * RETURNS:
1337  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1338  */
1339 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1340                                         void *holder)
1341 {
1342         struct block_device *bdev;
1343         int err;
1344 
1345         bdev = lookup_bdev(path);
1346         if (IS_ERR(bdev))
1347                 return bdev;
1348 
1349         err = blkdev_get(bdev, mode, holder);
1350         if (err)
1351                 return ERR_PTR(err);
1352 
1353         if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1354                 blkdev_put(bdev, mode);
1355                 return ERR_PTR(-EACCES);
1356         }
1357 
1358         return bdev;
1359 }
1360 EXPORT_SYMBOL(blkdev_get_by_path);
1361 
1362 /**
1363  * blkdev_get_by_dev - open a block device by device number
1364  * @dev: device number of block device to open
1365  * @mode: FMODE_* mask
1366  * @holder: exclusive holder identifier
1367  *
1368  * Open the blockdevice described by device number @dev.  @mode and
1369  * @holder are identical to blkdev_get().
1370  *
1371  * Use it ONLY if you really do not have anything better - i.e. when
1372  * you are behind a truly sucky interface and all you are given is a
1373  * device number.  _Never_ to be used for internal purposes.  If you
1374  * ever need it - reconsider your API.
1375  *
1376  * On success, the returned block_device has reference count of one.
1377  *
1378  * CONTEXT:
1379  * Might sleep.
1380  *
1381  * RETURNS:
1382  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1383  */
1384 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1385 {
1386         struct block_device *bdev;
1387         int err;
1388 
1389         bdev = bdget(dev);
1390         if (!bdev)
1391                 return ERR_PTR(-ENOMEM);
1392 
1393         err = blkdev_get(bdev, mode, holder);
1394         if (err)
1395                 return ERR_PTR(err);
1396 
1397         return bdev;
1398 }
1399 EXPORT_SYMBOL(blkdev_get_by_dev);
1400 
1401 static int blkdev_open(struct inode * inode, struct file * filp)
1402 {
1403         struct block_device *bdev;
1404 
1405         /*
1406          * Preserve backwards compatibility and allow large file access
1407          * even if userspace doesn't ask for it explicitly. Some mkfs
1408          * binary needs it. We might want to drop this workaround
1409          * during an unstable branch.
1410          */
1411         filp->f_flags |= O_LARGEFILE;
1412 
1413         if (filp->f_flags & O_NDELAY)
1414                 filp->f_mode |= FMODE_NDELAY;
1415         if (filp->f_flags & O_EXCL)
1416                 filp->f_mode |= FMODE_EXCL;
1417         if ((filp->f_flags & O_ACCMODE) == 3)
1418                 filp->f_mode |= FMODE_WRITE_IOCTL;
1419 
1420         bdev = bd_acquire(inode);
1421         if (bdev == NULL)
1422                 return -ENOMEM;
1423 
1424         filp->f_mapping = bdev->bd_inode->i_mapping;
1425 
1426         return blkdev_get(bdev, filp->f_mode, filp);
1427 }
1428 
1429 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1430 {
1431         int ret = 0;
1432         struct gendisk *disk = bdev->bd_disk;
1433         struct block_device *victim = NULL;
1434 
1435         mutex_lock_nested(&bdev->bd_mutex, for_part);
1436         if (for_part)
1437                 bdev->bd_part_count--;
1438 
1439         if (!--bdev->bd_openers) {
1440                 WARN_ON_ONCE(bdev->bd_holders);
1441                 sync_blockdev(bdev);
1442                 kill_bdev(bdev);
1443                 /* ->release can cause the old bdi to disappear,
1444                  * so must switch it out first
1445                  */
1446                 bdev_inode_switch_bdi(bdev->bd_inode,
1447                                         &default_backing_dev_info);
1448         }
1449         if (bdev->bd_contains == bdev) {
1450                 if (disk->fops->release)
1451                         ret = disk->fops->release(disk, mode);
1452         }
1453         if (!bdev->bd_openers) {
1454                 struct module *owner = disk->fops->owner;
1455 
1456                 disk_put_part(bdev->bd_part);
1457                 bdev->bd_part = NULL;
1458                 bdev->bd_disk = NULL;
1459                 if (bdev != bdev->bd_contains)
1460                         victim = bdev->bd_contains;
1461                 bdev->bd_contains = NULL;
1462 
1463                 put_disk(disk);
1464                 module_put(owner);
1465         }
1466         mutex_unlock(&bdev->bd_mutex);
1467         bdput(bdev);
1468         if (victim)
1469                 __blkdev_put(victim, mode, 1);
1470         return ret;
1471 }
1472 
1473 int blkdev_put(struct block_device *bdev, fmode_t mode)
1474 {
1475         if (mode & FMODE_EXCL) {
1476                 bool bdev_free;
1477 
1478                 /*
1479                  * Release a claim on the device.  The holder fields
1480                  * are protected with bdev_lock.  bd_mutex is to
1481                  * synchronize disk_holder unlinking.
1482                  */
1483                 mutex_lock(&bdev->bd_mutex);
1484                 spin_lock(&bdev_lock);
1485 
1486                 WARN_ON_ONCE(--bdev->bd_holders < 0);
1487                 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1488 
1489                 /* bd_contains might point to self, check in a separate step */
1490                 if ((bdev_free = !bdev->bd_holders))
1491                         bdev->bd_holder = NULL;
1492                 if (!bdev->bd_contains->bd_holders)
1493                         bdev->bd_contains->bd_holder = NULL;
1494 
1495                 spin_unlock(&bdev_lock);
1496 
1497                 /*
1498                  * If this was the last claim, remove holder link and
1499                  * unblock evpoll if it was a write holder.
1500                  */
1501                 if (bdev_free) {
1502                         if (bdev->bd_write_holder) {
1503                                 disk_unblock_events(bdev->bd_disk);
1504                                 disk_check_events(bdev->bd_disk);
1505                                 bdev->bd_write_holder = false;
1506                         }
1507                 }
1508 
1509                 mutex_unlock(&bdev->bd_mutex);
1510         }
1511 
1512         return __blkdev_put(bdev, mode, 0);
1513 }
1514 EXPORT_SYMBOL(blkdev_put);
1515 
1516 static int blkdev_close(struct inode * inode, struct file * filp)
1517 {
1518         struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1519 
1520         return blkdev_put(bdev, filp->f_mode);
1521 }
1522 
1523 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1524 {
1525         struct block_device *bdev = I_BDEV(file->f_mapping->host);
1526         fmode_t mode = file->f_mode;
1527 
1528         /*
1529          * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1530          * to updated it before every ioctl.
1531          */
1532         if (file->f_flags & O_NDELAY)
1533                 mode |= FMODE_NDELAY;
1534         else
1535                 mode &= ~FMODE_NDELAY;
1536 
1537         return blkdev_ioctl(bdev, mode, cmd, arg);
1538 }
1539 
1540 /*
1541  * Write data to the block device.  Only intended for the block device itself
1542  * and the raw driver which basically is a fake block device.
1543  *
1544  * Does not take i_mutex for the write and thus is not for general purpose
1545  * use.
1546  */
1547 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1548                          unsigned long nr_segs, loff_t pos)
1549 {
1550         struct file *file = iocb->ki_filp;
1551         ssize_t ret;
1552 
1553         BUG_ON(iocb->ki_pos != pos);
1554 
1555         ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1556         if (ret > 0 || ret == -EIOCBQUEUED) {
1557                 ssize_t err;
1558 
1559                 err = generic_write_sync(file, pos, ret);
1560                 if (err < 0 && ret > 0)
1561                         ret = err;
1562         }
1563         return ret;
1564 }
1565 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1566 
1567 /*
1568  * Try to release a page associated with block device when the system
1569  * is under memory pressure.
1570  */
1571 static int blkdev_releasepage(struct page *page, gfp_t wait)
1572 {
1573         struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1574 
1575         if (super && super->s_op->bdev_try_to_free_page)
1576                 return super->s_op->bdev_try_to_free_page(super, page, wait);
1577 
1578         return try_to_free_buffers(page);
1579 }
1580 
1581 static const struct address_space_operations def_blk_aops = {
1582         .readpage       = blkdev_readpage,
1583         .writepage      = blkdev_writepage,
1584         .write_begin    = blkdev_write_begin,
1585         .write_end      = blkdev_write_end,
1586         .writepages     = generic_writepages,
1587         .releasepage    = blkdev_releasepage,
1588         .direct_IO      = blkdev_direct_IO,
1589 };
1590 
1591 const struct file_operations def_blk_fops = {
1592         .open           = blkdev_open,
1593         .release        = blkdev_close,
1594         .llseek         = block_llseek,
1595         .read           = do_sync_read,
1596         .write          = do_sync_write,
1597         .aio_read       = generic_file_aio_read,
1598         .aio_write      = blkdev_aio_write,
1599         .mmap           = generic_file_mmap,
1600         .fsync          = blkdev_fsync,
1601         .unlocked_ioctl = block_ioctl,
1602 #ifdef CONFIG_COMPAT
1603         .compat_ioctl   = compat_blkdev_ioctl,
1604 #endif
1605         .splice_read    = generic_file_splice_read,
1606         .splice_write   = generic_file_splice_write,
1607 };
1608 
1609 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1610 {
1611         int res;
1612         mm_segment_t old_fs = get_fs();
1613         set_fs(KERNEL_DS);
1614         res = blkdev_ioctl(bdev, 0, cmd, arg);
1615         set_fs(old_fs);
1616         return res;
1617 }
1618 
1619 EXPORT_SYMBOL(ioctl_by_bdev);
1620 
1621 /**
1622  * lookup_bdev  - lookup a struct block_device by name
1623  * @pathname:   special file representing the block device
1624  *
1625  * Get a reference to the blockdevice at @pathname in the current
1626  * namespace if possible and return it.  Return ERR_PTR(error)
1627  * otherwise.
1628  */
1629 struct block_device *lookup_bdev(const char *pathname)
1630 {
1631         struct block_device *bdev;
1632         struct inode *inode;
1633         struct path path;
1634         int error;
1635 
1636         if (!pathname || !*pathname)
1637                 return ERR_PTR(-EINVAL);
1638 
1639         error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1640         if (error)
1641                 return ERR_PTR(error);
1642 
1643         inode = path.dentry->d_inode;
1644         error = -ENOTBLK;
1645         if (!S_ISBLK(inode->i_mode))
1646                 goto fail;
1647         error = -EACCES;
1648         if (path.mnt->mnt_flags & MNT_NODEV)
1649                 goto fail;
1650         error = -ENOMEM;
1651         bdev = bd_acquire(inode);
1652         if (!bdev)
1653                 goto fail;
1654 out:
1655         path_put(&path);
1656         return bdev;
1657 fail:
1658         bdev = ERR_PTR(error);
1659         goto out;
1660 }
1661 EXPORT_SYMBOL(lookup_bdev);
1662 
1663 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1664 {
1665         struct super_block *sb = get_super(bdev);
1666         int res = 0;
1667 
1668         if (sb) {
1669                 /*
1670                  * no need to lock the super, get_super holds the
1671                  * read mutex so the filesystem cannot go away
1672                  * under us (->put_super runs with the write lock
1673                  * hold).
1674                  */
1675                 shrink_dcache_sb(sb);
1676                 res = invalidate_inodes(sb, kill_dirty);
1677                 drop_super(sb);
1678         }
1679         invalidate_bdev(bdev);
1680         return res;
1681 }
1682 EXPORT_SYMBOL(__invalidate_device);
1683 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp