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

TOMOYO Linux Cross Reference
Linux/fs/block_dev.c

Version: ~ [ linux-5.8-rc3 ] ~ [ linux-5.7.5 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.48 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.129 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.185 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.228 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.228 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ 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-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/magic.h>
 20 #include <linux/buffer_head.h>
 21 #include <linux/swap.h>
 22 #include <linux/pagevec.h>
 23 #include <linux/writeback.h>
 24 #include <linux/mpage.h>
 25 #include <linux/mount.h>
 26 #include <linux/uio.h>
 27 #include <linux/namei.h>
 28 #include <linux/log2.h>
 29 #include <linux/cleancache.h>
 30 #include <linux/aio.h>
 31 #include <asm/uaccess.h>
 32 #include "internal.h"
 33 
 34 struct bdev_inode {
 35         struct block_device bdev;
 36         struct inode vfs_inode;
 37 };
 38 
 39 static const struct address_space_operations def_blk_aops;
 40 
 41 static inline struct bdev_inode *BDEV_I(struct inode *inode)
 42 {
 43         return container_of(inode, struct bdev_inode, vfs_inode);
 44 }
 45 
 46 inline struct block_device *I_BDEV(struct inode *inode)
 47 {
 48         return &BDEV_I(inode)->bdev;
 49 }
 50 EXPORT_SYMBOL(I_BDEV);
 51 
 52 /*
 53  * Move the inode from its current bdi to a new bdi.  Make sure the inode
 54  * is clean before moving so that it doesn't linger on the old bdi.
 55  */
 56 static void bdev_inode_switch_bdi(struct inode *inode,
 57                         struct backing_dev_info *dst)
 58 {
 59         while (true) {
 60                 spin_lock(&inode->i_lock);
 61                 if (!(inode->i_state & I_DIRTY)) {
 62                         inode->i_data.backing_dev_info = dst;
 63                         spin_unlock(&inode->i_lock);
 64                         return;
 65                 }
 66                 spin_unlock(&inode->i_lock);
 67                 WARN_ON_ONCE(write_inode_now(inode, true));
 68         }
 69 }
 70 
 71 /* Kill _all_ buffers and pagecache , dirty or not.. */
 72 void kill_bdev(struct block_device *bdev)
 73 {
 74         struct address_space *mapping = bdev->bd_inode->i_mapping;
 75 
 76         if (mapping->nrpages == 0 && mapping->nrshadows == 0)
 77                 return;
 78 
 79         invalidate_bh_lrus();
 80         truncate_inode_pages(mapping, 0);
 81 }       
 82 EXPORT_SYMBOL(kill_bdev);
 83 
 84 /* Invalidate clean unused buffers and pagecache. */
 85 void invalidate_bdev(struct block_device *bdev)
 86 {
 87         struct address_space *mapping = bdev->bd_inode->i_mapping;
 88 
 89         if (mapping->nrpages == 0)
 90                 return;
 91 
 92         invalidate_bh_lrus();
 93         lru_add_drain_all();    /* make sure all lru add caches are flushed */
 94         invalidate_mapping_pages(mapping, 0, -1);
 95         /* 99% of the time, we don't need to flush the cleancache on the bdev.
 96          * But, for the strange corners, lets be cautious
 97          */
 98         cleancache_invalidate_inode(mapping);
 99 }
100 EXPORT_SYMBOL(invalidate_bdev);
101 
102 int set_blocksize(struct block_device *bdev, int size)
103 {
104         /* Size must be a power of two, and between 512 and PAGE_SIZE */
105         if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
106                 return -EINVAL;
107 
108         /* Size cannot be smaller than the size supported by the device */
109         if (size < bdev_logical_block_size(bdev))
110                 return -EINVAL;
111 
112         /* Don't change the size if it is same as current */
113         if (bdev->bd_block_size != size) {
114                 sync_blockdev(bdev);
115                 bdev->bd_block_size = size;
116                 bdev->bd_inode->i_blkbits = blksize_bits(size);
117                 kill_bdev(bdev);
118         }
119         return 0;
120 }
121 
122 EXPORT_SYMBOL(set_blocksize);
123 
124 int sb_set_blocksize(struct super_block *sb, int size)
125 {
126         if (set_blocksize(sb->s_bdev, size))
127                 return 0;
128         /* If we get here, we know size is power of two
129          * and it's value is between 512 and PAGE_SIZE */
130         sb->s_blocksize = size;
131         sb->s_blocksize_bits = blksize_bits(size);
132         return sb->s_blocksize;
133 }
134 
135 EXPORT_SYMBOL(sb_set_blocksize);
136 
137 int sb_min_blocksize(struct super_block *sb, int size)
138 {
139         int minsize = bdev_logical_block_size(sb->s_bdev);
140         if (size < minsize)
141                 size = minsize;
142         return sb_set_blocksize(sb, size);
143 }
144 
145 EXPORT_SYMBOL(sb_min_blocksize);
146 
147 static int
148 blkdev_get_block(struct inode *inode, sector_t iblock,
149                 struct buffer_head *bh, int create)
150 {
151         bh->b_bdev = I_BDEV(inode);
152         bh->b_blocknr = iblock;
153         set_buffer_mapped(bh);
154         return 0;
155 }
156 
157 static ssize_t
158 blkdev_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
159                         loff_t offset)
160 {
161         struct file *file = iocb->ki_filp;
162         struct inode *inode = file->f_mapping->host;
163 
164         return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iter,
165                                     offset, blkdev_get_block,
166                                     NULL, NULL, 0);
167 }
168 
169 int __sync_blockdev(struct block_device *bdev, int wait)
170 {
171         if (!bdev)
172                 return 0;
173         if (!wait)
174                 return filemap_flush(bdev->bd_inode->i_mapping);
175         return filemap_write_and_wait(bdev->bd_inode->i_mapping);
176 }
177 
178 /*
179  * Write out and wait upon all the dirty data associated with a block
180  * device via its mapping.  Does not take the superblock lock.
181  */
182 int sync_blockdev(struct block_device *bdev)
183 {
184         return __sync_blockdev(bdev, 1);
185 }
186 EXPORT_SYMBOL(sync_blockdev);
187 
188 /*
189  * Write out and wait upon all dirty data associated with this
190  * device.   Filesystem data as well as the underlying block
191  * device.  Takes the superblock lock.
192  */
193 int fsync_bdev(struct block_device *bdev)
194 {
195         struct super_block *sb = get_super(bdev);
196         if (sb) {
197                 int res = sync_filesystem(sb);
198                 drop_super(sb);
199                 return res;
200         }
201         return sync_blockdev(bdev);
202 }
203 EXPORT_SYMBOL(fsync_bdev);
204 
205 /**
206  * freeze_bdev  --  lock a filesystem and force it into a consistent state
207  * @bdev:       blockdevice to lock
208  *
209  * If a superblock is found on this device, we take the s_umount semaphore
210  * on it to make sure nobody unmounts until the snapshot creation is done.
211  * The reference counter (bd_fsfreeze_count) guarantees that only the last
212  * unfreeze process can unfreeze the frozen filesystem actually when multiple
213  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
214  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
215  * actually.
216  */
217 struct super_block *freeze_bdev(struct block_device *bdev)
218 {
219         struct super_block *sb;
220         int error = 0;
221 
222         mutex_lock(&bdev->bd_fsfreeze_mutex);
223         if (++bdev->bd_fsfreeze_count > 1) {
224                 /*
225                  * We don't even need to grab a reference - the first call
226                  * to freeze_bdev grab an active reference and only the last
227                  * thaw_bdev drops it.
228                  */
229                 sb = get_super(bdev);
230                 drop_super(sb);
231                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
232                 return sb;
233         }
234 
235         sb = get_active_super(bdev);
236         if (!sb)
237                 goto out;
238         if (sb->s_op->freeze_super)
239                 error = sb->s_op->freeze_super(sb);
240         else
241                 error = freeze_super(sb);
242         if (error) {
243                 deactivate_super(sb);
244                 bdev->bd_fsfreeze_count--;
245                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
246                 return ERR_PTR(error);
247         }
248         deactivate_super(sb);
249  out:
250         sync_blockdev(bdev);
251         mutex_unlock(&bdev->bd_fsfreeze_mutex);
252         return sb;      /* thaw_bdev releases s->s_umount */
253 }
254 EXPORT_SYMBOL(freeze_bdev);
255 
256 /**
257  * thaw_bdev  -- unlock filesystem
258  * @bdev:       blockdevice to unlock
259  * @sb:         associated superblock
260  *
261  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
262  */
263 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
264 {
265         int error = -EINVAL;
266 
267         mutex_lock(&bdev->bd_fsfreeze_mutex);
268         if (!bdev->bd_fsfreeze_count)
269                 goto out;
270 
271         error = 0;
272         if (--bdev->bd_fsfreeze_count > 0)
273                 goto out;
274 
275         if (!sb)
276                 goto out;
277 
278         if (sb->s_op->thaw_super)
279                 error = sb->s_op->thaw_super(sb);
280         else
281                 error = thaw_super(sb);
282         if (error) {
283                 bdev->bd_fsfreeze_count++;
284                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
285                 return error;
286         }
287 out:
288         mutex_unlock(&bdev->bd_fsfreeze_mutex);
289         return 0;
290 }
291 EXPORT_SYMBOL(thaw_bdev);
292 
293 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
294 {
295         return block_write_full_page(page, blkdev_get_block, wbc);
296 }
297 
298 static int blkdev_readpage(struct file * file, struct page * page)
299 {
300         return block_read_full_page(page, blkdev_get_block);
301 }
302 
303 static int blkdev_readpages(struct file *file, struct address_space *mapping,
304                         struct list_head *pages, unsigned nr_pages)
305 {
306         return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
307 }
308 
309 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
310                         loff_t pos, unsigned len, unsigned flags,
311                         struct page **pagep, void **fsdata)
312 {
313         return block_write_begin(mapping, pos, len, flags, pagep,
314                                  blkdev_get_block);
315 }
316 
317 static int blkdev_write_end(struct file *file, struct address_space *mapping,
318                         loff_t pos, unsigned len, unsigned copied,
319                         struct page *page, void *fsdata)
320 {
321         int ret;
322         ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
323 
324         unlock_page(page);
325         page_cache_release(page);
326 
327         return ret;
328 }
329 
330 /*
331  * private llseek:
332  * for a block special file file_inode(file)->i_size is zero
333  * so we compute the size by hand (just as in block_read/write above)
334  */
335 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
336 {
337         struct inode *bd_inode = file->f_mapping->host;
338         loff_t retval;
339 
340         mutex_lock(&bd_inode->i_mutex);
341         retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
342         mutex_unlock(&bd_inode->i_mutex);
343         return retval;
344 }
345         
346 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
347 {
348         struct inode *bd_inode = filp->f_mapping->host;
349         struct block_device *bdev = I_BDEV(bd_inode);
350         int error;
351         
352         error = filemap_write_and_wait_range(filp->f_mapping, start, end);
353         if (error)
354                 return error;
355 
356         /*
357          * There is no need to serialise calls to blkdev_issue_flush with
358          * i_mutex and doing so causes performance issues with concurrent
359          * O_SYNC writers to a block device.
360          */
361         error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
362         if (error == -EOPNOTSUPP)
363                 error = 0;
364 
365         return error;
366 }
367 EXPORT_SYMBOL(blkdev_fsync);
368 
369 /**
370  * bdev_read_page() - Start reading a page from a block device
371  * @bdev: The device to read the page from
372  * @sector: The offset on the device to read the page to (need not be aligned)
373  * @page: The page to read
374  *
375  * On entry, the page should be locked.  It will be unlocked when the page
376  * has been read.  If the block driver implements rw_page synchronously,
377  * that will be true on exit from this function, but it need not be.
378  *
379  * Errors returned by this function are usually "soft", eg out of memory, or
380  * queue full; callers should try a different route to read this page rather
381  * than propagate an error back up the stack.
382  *
383  * Return: negative errno if an error occurs, 0 if submission was successful.
384  */
385 int bdev_read_page(struct block_device *bdev, sector_t sector,
386                         struct page *page)
387 {
388         const struct block_device_operations *ops = bdev->bd_disk->fops;
389         if (!ops->rw_page)
390                 return -EOPNOTSUPP;
391         return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
392 }
393 EXPORT_SYMBOL_GPL(bdev_read_page);
394 
395 /**
396  * bdev_write_page() - Start writing a page to a block device
397  * @bdev: The device to write the page to
398  * @sector: The offset on the device to write the page to (need not be aligned)
399  * @page: The page to write
400  * @wbc: The writeback_control for the write
401  *
402  * On entry, the page should be locked and not currently under writeback.
403  * On exit, if the write started successfully, the page will be unlocked and
404  * under writeback.  If the write failed already (eg the driver failed to
405  * queue the page to the device), the page will still be locked.  If the
406  * caller is a ->writepage implementation, it will need to unlock the page.
407  *
408  * Errors returned by this function are usually "soft", eg out of memory, or
409  * queue full; callers should try a different route to write this page rather
410  * than propagate an error back up the stack.
411  *
412  * Return: negative errno if an error occurs, 0 if submission was successful.
413  */
414 int bdev_write_page(struct block_device *bdev, sector_t sector,
415                         struct page *page, struct writeback_control *wbc)
416 {
417         int result;
418         int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
419         const struct block_device_operations *ops = bdev->bd_disk->fops;
420         if (!ops->rw_page)
421                 return -EOPNOTSUPP;
422         set_page_writeback(page);
423         result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
424         if (result)
425                 end_page_writeback(page);
426         else
427                 unlock_page(page);
428         return result;
429 }
430 EXPORT_SYMBOL_GPL(bdev_write_page);
431 
432 /*
433  * pseudo-fs
434  */
435 
436 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
437 static struct kmem_cache * bdev_cachep __read_mostly;
438 
439 static struct inode *bdev_alloc_inode(struct super_block *sb)
440 {
441         struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
442         if (!ei)
443                 return NULL;
444         return &ei->vfs_inode;
445 }
446 
447 static void bdev_i_callback(struct rcu_head *head)
448 {
449         struct inode *inode = container_of(head, struct inode, i_rcu);
450         struct bdev_inode *bdi = BDEV_I(inode);
451 
452         kmem_cache_free(bdev_cachep, bdi);
453 }
454 
455 static void bdev_destroy_inode(struct inode *inode)
456 {
457         call_rcu(&inode->i_rcu, bdev_i_callback);
458 }
459 
460 static void init_once(void *foo)
461 {
462         struct bdev_inode *ei = (struct bdev_inode *) foo;
463         struct block_device *bdev = &ei->bdev;
464 
465         memset(bdev, 0, sizeof(*bdev));
466         mutex_init(&bdev->bd_mutex);
467         INIT_LIST_HEAD(&bdev->bd_inodes);
468         INIT_LIST_HEAD(&bdev->bd_list);
469 #ifdef CONFIG_SYSFS
470         INIT_LIST_HEAD(&bdev->bd_holder_disks);
471 #endif
472         inode_init_once(&ei->vfs_inode);
473         /* Initialize mutex for freeze. */
474         mutex_init(&bdev->bd_fsfreeze_mutex);
475 }
476 
477 static inline void __bd_forget(struct inode *inode)
478 {
479         list_del_init(&inode->i_devices);
480         inode->i_bdev = NULL;
481         inode->i_mapping = &inode->i_data;
482 }
483 
484 static void bdev_evict_inode(struct inode *inode)
485 {
486         struct block_device *bdev = &BDEV_I(inode)->bdev;
487         struct list_head *p;
488         truncate_inode_pages_final(&inode->i_data);
489         invalidate_inode_buffers(inode); /* is it needed here? */
490         clear_inode(inode);
491         spin_lock(&bdev_lock);
492         while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
493                 __bd_forget(list_entry(p, struct inode, i_devices));
494         }
495         list_del_init(&bdev->bd_list);
496         spin_unlock(&bdev_lock);
497 }
498 
499 static const struct super_operations bdev_sops = {
500         .statfs = simple_statfs,
501         .alloc_inode = bdev_alloc_inode,
502         .destroy_inode = bdev_destroy_inode,
503         .drop_inode = generic_delete_inode,
504         .evict_inode = bdev_evict_inode,
505 };
506 
507 static struct dentry *bd_mount(struct file_system_type *fs_type,
508         int flags, const char *dev_name, void *data)
509 {
510         return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
511 }
512 
513 static struct file_system_type bd_type = {
514         .name           = "bdev",
515         .mount          = bd_mount,
516         .kill_sb        = kill_anon_super,
517 };
518 
519 static struct super_block *blockdev_superblock __read_mostly;
520 
521 void __init bdev_cache_init(void)
522 {
523         int err;
524         static struct vfsmount *bd_mnt;
525 
526         bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
527                         0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
528                                 SLAB_MEM_SPREAD|SLAB_PANIC),
529                         init_once);
530         err = register_filesystem(&bd_type);
531         if (err)
532                 panic("Cannot register bdev pseudo-fs");
533         bd_mnt = kern_mount(&bd_type);
534         if (IS_ERR(bd_mnt))
535                 panic("Cannot create bdev pseudo-fs");
536         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
537 }
538 
539 /*
540  * Most likely _very_ bad one - but then it's hardly critical for small
541  * /dev and can be fixed when somebody will need really large one.
542  * Keep in mind that it will be fed through icache hash function too.
543  */
544 static inline unsigned long hash(dev_t dev)
545 {
546         return MAJOR(dev)+MINOR(dev);
547 }
548 
549 static int bdev_test(struct inode *inode, void *data)
550 {
551         return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
552 }
553 
554 static int bdev_set(struct inode *inode, void *data)
555 {
556         BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
557         return 0;
558 }
559 
560 static LIST_HEAD(all_bdevs);
561 
562 struct block_device *bdget(dev_t dev)
563 {
564         struct block_device *bdev;
565         struct inode *inode;
566 
567         inode = iget5_locked(blockdev_superblock, hash(dev),
568                         bdev_test, bdev_set, &dev);
569 
570         if (!inode)
571                 return NULL;
572 
573         bdev = &BDEV_I(inode)->bdev;
574 
575         if (inode->i_state & I_NEW) {
576                 bdev->bd_contains = NULL;
577                 bdev->bd_super = NULL;
578                 bdev->bd_inode = inode;
579                 bdev->bd_block_size = (1 << inode->i_blkbits);
580                 bdev->bd_part_count = 0;
581                 bdev->bd_invalidated = 0;
582                 inode->i_mode = S_IFBLK;
583                 inode->i_rdev = dev;
584                 inode->i_bdev = bdev;
585                 inode->i_data.a_ops = &def_blk_aops;
586                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
587                 inode->i_data.backing_dev_info = &default_backing_dev_info;
588                 spin_lock(&bdev_lock);
589                 list_add(&bdev->bd_list, &all_bdevs);
590                 spin_unlock(&bdev_lock);
591                 unlock_new_inode(inode);
592         }
593         return bdev;
594 }
595 
596 EXPORT_SYMBOL(bdget);
597 
598 /**
599  * bdgrab -- Grab a reference to an already referenced block device
600  * @bdev:       Block device to grab a reference to.
601  */
602 struct block_device *bdgrab(struct block_device *bdev)
603 {
604         ihold(bdev->bd_inode);
605         return bdev;
606 }
607 EXPORT_SYMBOL(bdgrab);
608 
609 long nr_blockdev_pages(void)
610 {
611         struct block_device *bdev;
612         long ret = 0;
613         spin_lock(&bdev_lock);
614         list_for_each_entry(bdev, &all_bdevs, bd_list) {
615                 ret += bdev->bd_inode->i_mapping->nrpages;
616         }
617         spin_unlock(&bdev_lock);
618         return ret;
619 }
620 
621 void bdput(struct block_device *bdev)
622 {
623         iput(bdev->bd_inode);
624 }
625 
626 EXPORT_SYMBOL(bdput);
627  
628 static struct block_device *bd_acquire(struct inode *inode)
629 {
630         struct block_device *bdev;
631 
632         spin_lock(&bdev_lock);
633         bdev = inode->i_bdev;
634         if (bdev) {
635                 ihold(bdev->bd_inode);
636                 spin_unlock(&bdev_lock);
637                 return bdev;
638         }
639         spin_unlock(&bdev_lock);
640 
641         bdev = bdget(inode->i_rdev);
642         if (bdev) {
643                 spin_lock(&bdev_lock);
644                 if (!inode->i_bdev) {
645                         /*
646                          * We take an additional reference to bd_inode,
647                          * and it's released in clear_inode() of inode.
648                          * So, we can access it via ->i_mapping always
649                          * without igrab().
650                          */
651                         ihold(bdev->bd_inode);
652                         inode->i_bdev = bdev;
653                         inode->i_mapping = bdev->bd_inode->i_mapping;
654                         list_add(&inode->i_devices, &bdev->bd_inodes);
655                 }
656                 spin_unlock(&bdev_lock);
657         }
658         return bdev;
659 }
660 
661 int sb_is_blkdev_sb(struct super_block *sb)
662 {
663         return sb == blockdev_superblock;
664 }
665 
666 /* Call when you free inode */
667 
668 void bd_forget(struct inode *inode)
669 {
670         struct block_device *bdev = NULL;
671 
672         spin_lock(&bdev_lock);
673         if (!sb_is_blkdev_sb(inode->i_sb))
674                 bdev = inode->i_bdev;
675         __bd_forget(inode);
676         spin_unlock(&bdev_lock);
677 
678         if (bdev)
679                 iput(bdev->bd_inode);
680 }
681 
682 /**
683  * bd_may_claim - test whether a block device can be claimed
684  * @bdev: block device of interest
685  * @whole: whole block device containing @bdev, may equal @bdev
686  * @holder: holder trying to claim @bdev
687  *
688  * Test whether @bdev can be claimed by @holder.
689  *
690  * CONTEXT:
691  * spin_lock(&bdev_lock).
692  *
693  * RETURNS:
694  * %true if @bdev can be claimed, %false otherwise.
695  */
696 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
697                          void *holder)
698 {
699         if (bdev->bd_holder == holder)
700                 return true;     /* already a holder */
701         else if (bdev->bd_holder != NULL)
702                 return false;    /* held by someone else */
703         else if (bdev->bd_contains == bdev)
704                 return true;     /* is a whole device which isn't held */
705 
706         else if (whole->bd_holder == bd_may_claim)
707                 return true;     /* is a partition of a device that is being partitioned */
708         else if (whole->bd_holder != NULL)
709                 return false;    /* is a partition of a held device */
710         else
711                 return true;     /* is a partition of an un-held device */
712 }
713 
714 /**
715  * bd_prepare_to_claim - prepare to claim a block device
716  * @bdev: block device of interest
717  * @whole: the whole device containing @bdev, may equal @bdev
718  * @holder: holder trying to claim @bdev
719  *
720  * Prepare to claim @bdev.  This function fails if @bdev is already
721  * claimed by another holder and waits if another claiming is in
722  * progress.  This function doesn't actually claim.  On successful
723  * return, the caller has ownership of bd_claiming and bd_holder[s].
724  *
725  * CONTEXT:
726  * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
727  * it multiple times.
728  *
729  * RETURNS:
730  * 0 if @bdev can be claimed, -EBUSY otherwise.
731  */
732 static int bd_prepare_to_claim(struct block_device *bdev,
733                                struct block_device *whole, void *holder)
734 {
735 retry:
736         /* if someone else claimed, fail */
737         if (!bd_may_claim(bdev, whole, holder))
738                 return -EBUSY;
739 
740         /* if claiming is already in progress, wait for it to finish */
741         if (whole->bd_claiming) {
742                 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
743                 DEFINE_WAIT(wait);
744 
745                 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
746                 spin_unlock(&bdev_lock);
747                 schedule();
748                 finish_wait(wq, &wait);
749                 spin_lock(&bdev_lock);
750                 goto retry;
751         }
752 
753         /* yay, all mine */
754         return 0;
755 }
756 
757 /**
758  * bd_start_claiming - start claiming a block device
759  * @bdev: block device of interest
760  * @holder: holder trying to claim @bdev
761  *
762  * @bdev is about to be opened exclusively.  Check @bdev can be opened
763  * exclusively and mark that an exclusive open is in progress.  Each
764  * successful call to this function must be matched with a call to
765  * either bd_finish_claiming() or bd_abort_claiming() (which do not
766  * fail).
767  *
768  * This function is used to gain exclusive access to the block device
769  * without actually causing other exclusive open attempts to fail. It
770  * should be used when the open sequence itself requires exclusive
771  * access but may subsequently fail.
772  *
773  * CONTEXT:
774  * Might sleep.
775  *
776  * RETURNS:
777  * Pointer to the block device containing @bdev on success, ERR_PTR()
778  * value on failure.
779  */
780 static struct block_device *bd_start_claiming(struct block_device *bdev,
781                                               void *holder)
782 {
783         struct gendisk *disk;
784         struct block_device *whole;
785         int partno, err;
786 
787         might_sleep();
788 
789         /*
790          * @bdev might not have been initialized properly yet, look up
791          * and grab the outer block device the hard way.
792          */
793         disk = get_gendisk(bdev->bd_dev, &partno);
794         if (!disk)
795                 return ERR_PTR(-ENXIO);
796 
797         /*
798          * Normally, @bdev should equal what's returned from bdget_disk()
799          * if partno is 0; however, some drivers (floppy) use multiple
800          * bdev's for the same physical device and @bdev may be one of the
801          * aliases.  Keep @bdev if partno is 0.  This means claimer
802          * tracking is broken for those devices but it has always been that
803          * way.
804          */
805         if (partno)
806                 whole = bdget_disk(disk, 0);
807         else
808                 whole = bdgrab(bdev);
809 
810         module_put(disk->fops->owner);
811         put_disk(disk);
812         if (!whole)
813                 return ERR_PTR(-ENOMEM);
814 
815         /* prepare to claim, if successful, mark claiming in progress */
816         spin_lock(&bdev_lock);
817 
818         err = bd_prepare_to_claim(bdev, whole, holder);
819         if (err == 0) {
820                 whole->bd_claiming = holder;
821                 spin_unlock(&bdev_lock);
822                 return whole;
823         } else {
824                 spin_unlock(&bdev_lock);
825                 bdput(whole);
826                 return ERR_PTR(err);
827         }
828 }
829 
830 #ifdef CONFIG_SYSFS
831 struct bd_holder_disk {
832         struct list_head        list;
833         struct gendisk          *disk;
834         int                     refcnt;
835 };
836 
837 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
838                                                   struct gendisk *disk)
839 {
840         struct bd_holder_disk *holder;
841 
842         list_for_each_entry(holder, &bdev->bd_holder_disks, list)
843                 if (holder->disk == disk)
844                         return holder;
845         return NULL;
846 }
847 
848 static int add_symlink(struct kobject *from, struct kobject *to)
849 {
850         return sysfs_create_link(from, to, kobject_name(to));
851 }
852 
853 static void del_symlink(struct kobject *from, struct kobject *to)
854 {
855         sysfs_remove_link(from, kobject_name(to));
856 }
857 
858 /**
859  * bd_link_disk_holder - create symlinks between holding disk and slave bdev
860  * @bdev: the claimed slave bdev
861  * @disk: the holding disk
862  *
863  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
864  *
865  * This functions creates the following sysfs symlinks.
866  *
867  * - from "slaves" directory of the holder @disk to the claimed @bdev
868  * - from "holders" directory of the @bdev to the holder @disk
869  *
870  * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
871  * passed to bd_link_disk_holder(), then:
872  *
873  *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
874  *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
875  *
876  * The caller must have claimed @bdev before calling this function and
877  * ensure that both @bdev and @disk are valid during the creation and
878  * lifetime of these symlinks.
879  *
880  * CONTEXT:
881  * Might sleep.
882  *
883  * RETURNS:
884  * 0 on success, -errno on failure.
885  */
886 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
887 {
888         struct bd_holder_disk *holder;
889         int ret = 0;
890 
891         mutex_lock(&bdev->bd_mutex);
892 
893         WARN_ON_ONCE(!bdev->bd_holder);
894 
895         /* FIXME: remove the following once add_disk() handles errors */
896         if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
897                 goto out_unlock;
898 
899         holder = bd_find_holder_disk(bdev, disk);
900         if (holder) {
901                 holder->refcnt++;
902                 goto out_unlock;
903         }
904 
905         holder = kzalloc(sizeof(*holder), GFP_KERNEL);
906         if (!holder) {
907                 ret = -ENOMEM;
908                 goto out_unlock;
909         }
910 
911         INIT_LIST_HEAD(&holder->list);
912         holder->disk = disk;
913         holder->refcnt = 1;
914 
915         ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
916         if (ret)
917                 goto out_free;
918 
919         ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
920         if (ret)
921                 goto out_del;
922         /*
923          * bdev could be deleted beneath us which would implicitly destroy
924          * the holder directory.  Hold on to it.
925          */
926         kobject_get(bdev->bd_part->holder_dir);
927 
928         list_add(&holder->list, &bdev->bd_holder_disks);
929         goto out_unlock;
930 
931 out_del:
932         del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
933 out_free:
934         kfree(holder);
935 out_unlock:
936         mutex_unlock(&bdev->bd_mutex);
937         return ret;
938 }
939 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
940 
941 /**
942  * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
943  * @bdev: the calimed slave bdev
944  * @disk: the holding disk
945  *
946  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
947  *
948  * CONTEXT:
949  * Might sleep.
950  */
951 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
952 {
953         struct bd_holder_disk *holder;
954 
955         mutex_lock(&bdev->bd_mutex);
956 
957         holder = bd_find_holder_disk(bdev, disk);
958 
959         if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
960                 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
961                 del_symlink(bdev->bd_part->holder_dir,
962                             &disk_to_dev(disk)->kobj);
963                 kobject_put(bdev->bd_part->holder_dir);
964                 list_del_init(&holder->list);
965                 kfree(holder);
966         }
967 
968         mutex_unlock(&bdev->bd_mutex);
969 }
970 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
971 #endif
972 
973 /**
974  * flush_disk - invalidates all buffer-cache entries on a disk
975  *
976  * @bdev:      struct block device to be flushed
977  * @kill_dirty: flag to guide handling of dirty inodes
978  *
979  * Invalidates all buffer-cache entries on a disk. It should be called
980  * when a disk has been changed -- either by a media change or online
981  * resize.
982  */
983 static void flush_disk(struct block_device *bdev, bool kill_dirty)
984 {
985         if (__invalidate_device(bdev, kill_dirty)) {
986                 char name[BDEVNAME_SIZE] = "";
987 
988                 if (bdev->bd_disk)
989                         disk_name(bdev->bd_disk, 0, name);
990                 printk(KERN_WARNING "VFS: busy inodes on changed media or "
991                        "resized disk %s\n", name);
992         }
993 
994         if (!bdev->bd_disk)
995                 return;
996         if (disk_part_scan_enabled(bdev->bd_disk))
997                 bdev->bd_invalidated = 1;
998 }
999 
1000 /**
1001  * check_disk_size_change - checks for disk size change and adjusts bdev size.
1002  * @disk: struct gendisk to check
1003  * @bdev: struct bdev to adjust.
1004  *
1005  * This routine checks to see if the bdev size does not match the disk size
1006  * and adjusts it if it differs.
1007  */
1008 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1009 {
1010         loff_t disk_size, bdev_size;
1011 
1012         disk_size = (loff_t)get_capacity(disk) << 9;
1013         bdev_size = i_size_read(bdev->bd_inode);
1014         if (disk_size != bdev_size) {
1015                 char name[BDEVNAME_SIZE];
1016 
1017                 disk_name(disk, 0, name);
1018                 printk(KERN_INFO
1019                        "%s: detected capacity change from %lld to %lld\n",
1020                        name, bdev_size, disk_size);
1021                 i_size_write(bdev->bd_inode, disk_size);
1022                 flush_disk(bdev, false);
1023         }
1024 }
1025 EXPORT_SYMBOL(check_disk_size_change);
1026 
1027 /**
1028  * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1029  * @disk: struct gendisk to be revalidated
1030  *
1031  * This routine is a wrapper for lower-level driver's revalidate_disk
1032  * call-backs.  It is used to do common pre and post operations needed
1033  * for all revalidate_disk operations.
1034  */
1035 int revalidate_disk(struct gendisk *disk)
1036 {
1037         struct block_device *bdev;
1038         int ret = 0;
1039 
1040         if (disk->fops->revalidate_disk)
1041                 ret = disk->fops->revalidate_disk(disk);
1042 
1043         bdev = bdget_disk(disk, 0);
1044         if (!bdev)
1045                 return ret;
1046 
1047         mutex_lock(&bdev->bd_mutex);
1048         check_disk_size_change(disk, bdev);
1049         bdev->bd_invalidated = 0;
1050         mutex_unlock(&bdev->bd_mutex);
1051         bdput(bdev);
1052         return ret;
1053 }
1054 EXPORT_SYMBOL(revalidate_disk);
1055 
1056 /*
1057  * This routine checks whether a removable media has been changed,
1058  * and invalidates all buffer-cache-entries in that case. This
1059  * is a relatively slow routine, so we have to try to minimize using
1060  * it. Thus it is called only upon a 'mount' or 'open'. This
1061  * is the best way of combining speed and utility, I think.
1062  * People changing diskettes in the middle of an operation deserve
1063  * to lose :-)
1064  */
1065 int check_disk_change(struct block_device *bdev)
1066 {
1067         struct gendisk *disk = bdev->bd_disk;
1068         const struct block_device_operations *bdops = disk->fops;
1069         unsigned int events;
1070 
1071         events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1072                                    DISK_EVENT_EJECT_REQUEST);
1073         if (!(events & DISK_EVENT_MEDIA_CHANGE))
1074                 return 0;
1075 
1076         flush_disk(bdev, true);
1077         if (bdops->revalidate_disk)
1078                 bdops->revalidate_disk(bdev->bd_disk);
1079         return 1;
1080 }
1081 
1082 EXPORT_SYMBOL(check_disk_change);
1083 
1084 void bd_set_size(struct block_device *bdev, loff_t size)
1085 {
1086         unsigned bsize = bdev_logical_block_size(bdev);
1087 
1088         mutex_lock(&bdev->bd_inode->i_mutex);
1089         i_size_write(bdev->bd_inode, size);
1090         mutex_unlock(&bdev->bd_inode->i_mutex);
1091         while (bsize < PAGE_CACHE_SIZE) {
1092                 if (size & bsize)
1093                         break;
1094                 bsize <<= 1;
1095         }
1096         bdev->bd_block_size = bsize;
1097         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1098 }
1099 EXPORT_SYMBOL(bd_set_size);
1100 
1101 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1102 
1103 /*
1104  * bd_mutex locking:
1105  *
1106  *  mutex_lock(part->bd_mutex)
1107  *    mutex_lock_nested(whole->bd_mutex, 1)
1108  */
1109 
1110 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1111 {
1112         struct gendisk *disk;
1113         struct module *owner;
1114         int ret;
1115         int partno;
1116         int perm = 0;
1117 
1118         if (mode & FMODE_READ)
1119                 perm |= MAY_READ;
1120         if (mode & FMODE_WRITE)
1121                 perm |= MAY_WRITE;
1122         /*
1123          * hooks: /n/, see "layering violations".
1124          */
1125         if (!for_part) {
1126                 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1127                 if (ret != 0) {
1128                         bdput(bdev);
1129                         return ret;
1130                 }
1131         }
1132 
1133  restart:
1134 
1135         ret = -ENXIO;
1136         disk = get_gendisk(bdev->bd_dev, &partno);
1137         if (!disk)
1138                 goto out;
1139         owner = disk->fops->owner;
1140 
1141         disk_block_events(disk);
1142         mutex_lock_nested(&bdev->bd_mutex, for_part);
1143         if (!bdev->bd_openers) {
1144                 bdev->bd_disk = disk;
1145                 bdev->bd_queue = disk->queue;
1146                 bdev->bd_contains = bdev;
1147                 if (!partno) {
1148                         struct backing_dev_info *bdi;
1149 
1150                         ret = -ENXIO;
1151                         bdev->bd_part = disk_get_part(disk, partno);
1152                         if (!bdev->bd_part)
1153                                 goto out_clear;
1154 
1155                         ret = 0;
1156                         if (disk->fops->open) {
1157                                 ret = disk->fops->open(bdev, mode);
1158                                 if (ret == -ERESTARTSYS) {
1159                                         /* Lost a race with 'disk' being
1160                                          * deleted, try again.
1161                                          * See md.c
1162                                          */
1163                                         disk_put_part(bdev->bd_part);
1164                                         bdev->bd_part = NULL;
1165                                         bdev->bd_disk = NULL;
1166                                         bdev->bd_queue = NULL;
1167                                         mutex_unlock(&bdev->bd_mutex);
1168                                         disk_unblock_events(disk);
1169                                         put_disk(disk);
1170                                         module_put(owner);
1171                                         goto restart;
1172                                 }
1173                         }
1174 
1175                         if (!ret) {
1176                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1177                                 bdi = blk_get_backing_dev_info(bdev);
1178                                 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1179                         }
1180 
1181                         /*
1182                          * If the device is invalidated, rescan partition
1183                          * if open succeeded or failed with -ENOMEDIUM.
1184                          * The latter is necessary to prevent ghost
1185                          * partitions on a removed medium.
1186                          */
1187                         if (bdev->bd_invalidated) {
1188                                 if (!ret)
1189                                         rescan_partitions(disk, bdev);
1190                                 else if (ret == -ENOMEDIUM)
1191                                         invalidate_partitions(disk, bdev);
1192                         }
1193                         if (ret)
1194                                 goto out_clear;
1195                 } else {
1196                         struct block_device *whole;
1197                         whole = bdget_disk(disk, 0);
1198                         ret = -ENOMEM;
1199                         if (!whole)
1200                                 goto out_clear;
1201                         BUG_ON(for_part);
1202                         ret = __blkdev_get(whole, mode, 1);
1203                         if (ret)
1204                                 goto out_clear;
1205                         bdev->bd_contains = whole;
1206                         bdev_inode_switch_bdi(bdev->bd_inode,
1207                                 whole->bd_inode->i_data.backing_dev_info);
1208                         bdev->bd_part = disk_get_part(disk, partno);
1209                         if (!(disk->flags & GENHD_FL_UP) ||
1210                             !bdev->bd_part || !bdev->bd_part->nr_sects) {
1211                                 ret = -ENXIO;
1212                                 goto out_clear;
1213                         }
1214                         bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1215                 }
1216         } else {
1217                 if (bdev->bd_contains == bdev) {
1218                         ret = 0;
1219                         if (bdev->bd_disk->fops->open)
1220                                 ret = bdev->bd_disk->fops->open(bdev, mode);
1221                         /* the same as first opener case, read comment there */
1222                         if (bdev->bd_invalidated) {
1223                                 if (!ret)
1224                                         rescan_partitions(bdev->bd_disk, bdev);
1225                                 else if (ret == -ENOMEDIUM)
1226                                         invalidate_partitions(bdev->bd_disk, bdev);
1227                         }
1228                         if (ret)
1229                                 goto out_unlock_bdev;
1230                 }
1231                 /* only one opener holds refs to the module and disk */
1232                 put_disk(disk);
1233                 module_put(owner);
1234         }
1235         bdev->bd_openers++;
1236         if (for_part)
1237                 bdev->bd_part_count++;
1238         mutex_unlock(&bdev->bd_mutex);
1239         disk_unblock_events(disk);
1240         return 0;
1241 
1242  out_clear:
1243         disk_put_part(bdev->bd_part);
1244         bdev->bd_disk = NULL;
1245         bdev->bd_part = NULL;
1246         bdev->bd_queue = NULL;
1247         bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1248         if (bdev != bdev->bd_contains)
1249                 __blkdev_put(bdev->bd_contains, mode, 1);
1250         bdev->bd_contains = NULL;
1251  out_unlock_bdev:
1252         mutex_unlock(&bdev->bd_mutex);
1253         disk_unblock_events(disk);
1254         put_disk(disk);
1255         module_put(owner);
1256  out:
1257         bdput(bdev);
1258 
1259         return ret;
1260 }
1261 
1262 /**
1263  * blkdev_get - open a block device
1264  * @bdev: block_device to open
1265  * @mode: FMODE_* mask
1266  * @holder: exclusive holder identifier
1267  *
1268  * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1269  * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1270  * @holder is invalid.  Exclusive opens may nest for the same @holder.
1271  *
1272  * On success, the reference count of @bdev is unchanged.  On failure,
1273  * @bdev is put.
1274  *
1275  * CONTEXT:
1276  * Might sleep.
1277  *
1278  * RETURNS:
1279  * 0 on success, -errno on failure.
1280  */
1281 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1282 {
1283         struct block_device *whole = NULL;
1284         int res;
1285 
1286         WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1287 
1288         if ((mode & FMODE_EXCL) && holder) {
1289                 whole = bd_start_claiming(bdev, holder);
1290                 if (IS_ERR(whole)) {
1291                         bdput(bdev);
1292                         return PTR_ERR(whole);
1293                 }
1294         }
1295 
1296         res = __blkdev_get(bdev, mode, 0);
1297 
1298         if (whole) {
1299                 struct gendisk *disk = whole->bd_disk;
1300 
1301                 /* finish claiming */
1302                 mutex_lock(&bdev->bd_mutex);
1303                 spin_lock(&bdev_lock);
1304 
1305                 if (!res) {
1306                         BUG_ON(!bd_may_claim(bdev, whole, holder));
1307                         /*
1308                          * Note that for a whole device bd_holders
1309                          * will be incremented twice, and bd_holder
1310                          * will be set to bd_may_claim before being
1311                          * set to holder
1312                          */
1313                         whole->bd_holders++;
1314                         whole->bd_holder = bd_may_claim;
1315                         bdev->bd_holders++;
1316                         bdev->bd_holder = holder;
1317                 }
1318 
1319                 /* tell others that we're done */
1320                 BUG_ON(whole->bd_claiming != holder);
1321                 whole->bd_claiming = NULL;
1322                 wake_up_bit(&whole->bd_claiming, 0);
1323 
1324                 spin_unlock(&bdev_lock);
1325 
1326                 /*
1327                  * Block event polling for write claims if requested.  Any
1328                  * write holder makes the write_holder state stick until
1329                  * all are released.  This is good enough and tracking
1330                  * individual writeable reference is too fragile given the
1331                  * way @mode is used in blkdev_get/put().
1332                  */
1333                 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1334                     (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1335                         bdev->bd_write_holder = true;
1336                         disk_block_events(disk);
1337                 }
1338 
1339                 mutex_unlock(&bdev->bd_mutex);
1340                 bdput(whole);
1341         }
1342 
1343         return res;
1344 }
1345 EXPORT_SYMBOL(blkdev_get);
1346 
1347 /**
1348  * blkdev_get_by_path - open a block device by name
1349  * @path: path to the block device to open
1350  * @mode: FMODE_* mask
1351  * @holder: exclusive holder identifier
1352  *
1353  * Open the blockdevice described by the device file at @path.  @mode
1354  * and @holder are identical to blkdev_get().
1355  *
1356  * On success, the returned block_device has reference count of one.
1357  *
1358  * CONTEXT:
1359  * Might sleep.
1360  *
1361  * RETURNS:
1362  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1363  */
1364 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1365                                         void *holder)
1366 {
1367         struct block_device *bdev;
1368         int err;
1369 
1370         bdev = lookup_bdev(path);
1371         if (IS_ERR(bdev))
1372                 return bdev;
1373 
1374         err = blkdev_get(bdev, mode, holder);
1375         if (err)
1376                 return ERR_PTR(err);
1377 
1378         if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1379                 blkdev_put(bdev, mode);
1380                 return ERR_PTR(-EACCES);
1381         }
1382 
1383         return bdev;
1384 }
1385 EXPORT_SYMBOL(blkdev_get_by_path);
1386 
1387 /**
1388  * blkdev_get_by_dev - open a block device by device number
1389  * @dev: device number of block device to open
1390  * @mode: FMODE_* mask
1391  * @holder: exclusive holder identifier
1392  *
1393  * Open the blockdevice described by device number @dev.  @mode and
1394  * @holder are identical to blkdev_get().
1395  *
1396  * Use it ONLY if you really do not have anything better - i.e. when
1397  * you are behind a truly sucky interface and all you are given is a
1398  * device number.  _Never_ to be used for internal purposes.  If you
1399  * ever need it - reconsider your API.
1400  *
1401  * On success, the returned block_device has reference count of one.
1402  *
1403  * CONTEXT:
1404  * Might sleep.
1405  *
1406  * RETURNS:
1407  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1408  */
1409 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1410 {
1411         struct block_device *bdev;
1412         int err;
1413 
1414         bdev = bdget(dev);
1415         if (!bdev)
1416                 return ERR_PTR(-ENOMEM);
1417 
1418         err = blkdev_get(bdev, mode, holder);
1419         if (err)
1420                 return ERR_PTR(err);
1421 
1422         return bdev;
1423 }
1424 EXPORT_SYMBOL(blkdev_get_by_dev);
1425 
1426 static int blkdev_open(struct inode * inode, struct file * filp)
1427 {
1428         struct block_device *bdev;
1429 
1430         /*
1431          * Preserve backwards compatibility and allow large file access
1432          * even if userspace doesn't ask for it explicitly. Some mkfs
1433          * binary needs it. We might want to drop this workaround
1434          * during an unstable branch.
1435          */
1436         filp->f_flags |= O_LARGEFILE;
1437 
1438         if (filp->f_flags & O_NDELAY)
1439                 filp->f_mode |= FMODE_NDELAY;
1440         if (filp->f_flags & O_EXCL)
1441                 filp->f_mode |= FMODE_EXCL;
1442         if ((filp->f_flags & O_ACCMODE) == 3)
1443                 filp->f_mode |= FMODE_WRITE_IOCTL;
1444 
1445         bdev = bd_acquire(inode);
1446         if (bdev == NULL)
1447                 return -ENOMEM;
1448 
1449         filp->f_mapping = bdev->bd_inode->i_mapping;
1450 
1451         return blkdev_get(bdev, filp->f_mode, filp);
1452 }
1453 
1454 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1455 {
1456         struct gendisk *disk = bdev->bd_disk;
1457         struct block_device *victim = NULL;
1458 
1459         mutex_lock_nested(&bdev->bd_mutex, for_part);
1460         if (for_part)
1461                 bdev->bd_part_count--;
1462 
1463         if (!--bdev->bd_openers) {
1464                 WARN_ON_ONCE(bdev->bd_holders);
1465                 sync_blockdev(bdev);
1466                 kill_bdev(bdev);
1467                 /* ->release can cause the old bdi to disappear,
1468                  * so must switch it out first
1469                  */
1470                 bdev_inode_switch_bdi(bdev->bd_inode,
1471                                         &default_backing_dev_info);
1472         }
1473         if (bdev->bd_contains == bdev) {
1474                 if (disk->fops->release)
1475                         disk->fops->release(disk, mode);
1476         }
1477         if (!bdev->bd_openers) {
1478                 struct module *owner = disk->fops->owner;
1479 
1480                 disk_put_part(bdev->bd_part);
1481                 bdev->bd_part = NULL;
1482                 bdev->bd_disk = NULL;
1483                 if (bdev != bdev->bd_contains)
1484                         victim = bdev->bd_contains;
1485                 bdev->bd_contains = NULL;
1486 
1487                 put_disk(disk);
1488                 module_put(owner);
1489         }
1490         mutex_unlock(&bdev->bd_mutex);
1491         bdput(bdev);
1492         if (victim)
1493                 __blkdev_put(victim, mode, 1);
1494 }
1495 
1496 void blkdev_put(struct block_device *bdev, fmode_t mode)
1497 {
1498         mutex_lock(&bdev->bd_mutex);
1499 
1500         if (mode & FMODE_EXCL) {
1501                 bool bdev_free;
1502 
1503                 /*
1504                  * Release a claim on the device.  The holder fields
1505                  * are protected with bdev_lock.  bd_mutex is to
1506                  * synchronize disk_holder unlinking.
1507                  */
1508                 spin_lock(&bdev_lock);
1509 
1510                 WARN_ON_ONCE(--bdev->bd_holders < 0);
1511                 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1512 
1513                 /* bd_contains might point to self, check in a separate step */
1514                 if ((bdev_free = !bdev->bd_holders))
1515                         bdev->bd_holder = NULL;
1516                 if (!bdev->bd_contains->bd_holders)
1517                         bdev->bd_contains->bd_holder = NULL;
1518 
1519                 spin_unlock(&bdev_lock);
1520 
1521                 /*
1522                  * If this was the last claim, remove holder link and
1523                  * unblock evpoll if it was a write holder.
1524                  */
1525                 if (bdev_free && bdev->bd_write_holder) {
1526                         disk_unblock_events(bdev->bd_disk);
1527                         bdev->bd_write_holder = false;
1528                 }
1529         }
1530 
1531         /*
1532          * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1533          * event.  This is to ensure detection of media removal commanded
1534          * from userland - e.g. eject(1).
1535          */
1536         disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1537 
1538         mutex_unlock(&bdev->bd_mutex);
1539 
1540         __blkdev_put(bdev, mode, 0);
1541 }
1542 EXPORT_SYMBOL(blkdev_put);
1543 
1544 static int blkdev_close(struct inode * inode, struct file * filp)
1545 {
1546         struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1547         blkdev_put(bdev, filp->f_mode);
1548         return 0;
1549 }
1550 
1551 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1552 {
1553         struct block_device *bdev = I_BDEV(file->f_mapping->host);
1554         fmode_t mode = file->f_mode;
1555 
1556         /*
1557          * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1558          * to updated it before every ioctl.
1559          */
1560         if (file->f_flags & O_NDELAY)
1561                 mode |= FMODE_NDELAY;
1562         else
1563                 mode &= ~FMODE_NDELAY;
1564 
1565         return blkdev_ioctl(bdev, mode, cmd, arg);
1566 }
1567 
1568 /*
1569  * Write data to the block device.  Only intended for the block device itself
1570  * and the raw driver which basically is a fake block device.
1571  *
1572  * Does not take i_mutex for the write and thus is not for general purpose
1573  * use.
1574  */
1575 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1576 {
1577         struct file *file = iocb->ki_filp;
1578         struct blk_plug plug;
1579         ssize_t ret;
1580 
1581         blk_start_plug(&plug);
1582         ret = __generic_file_write_iter(iocb, from);
1583         if (ret > 0) {
1584                 ssize_t err;
1585                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1586                 if (err < 0)
1587                         ret = err;
1588         }
1589         blk_finish_plug(&plug);
1590         return ret;
1591 }
1592 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1593 
1594 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1595 {
1596         struct file *file = iocb->ki_filp;
1597         struct inode *bd_inode = file->f_mapping->host;
1598         loff_t size = i_size_read(bd_inode);
1599         loff_t pos = iocb->ki_pos;
1600 
1601         if (pos >= size)
1602                 return 0;
1603 
1604         size -= pos;
1605         iov_iter_truncate(to, size);
1606         return generic_file_read_iter(iocb, to);
1607 }
1608 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1609 
1610 /*
1611  * Try to release a page associated with block device when the system
1612  * is under memory pressure.
1613  */
1614 static int blkdev_releasepage(struct page *page, gfp_t wait)
1615 {
1616         struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1617 
1618         if (super && super->s_op->bdev_try_to_free_page)
1619                 return super->s_op->bdev_try_to_free_page(super, page, wait);
1620 
1621         return try_to_free_buffers(page);
1622 }
1623 
1624 static const struct address_space_operations def_blk_aops = {
1625         .readpage       = blkdev_readpage,
1626         .readpages      = blkdev_readpages,
1627         .writepage      = blkdev_writepage,
1628         .write_begin    = blkdev_write_begin,
1629         .write_end      = blkdev_write_end,
1630         .writepages     = generic_writepages,
1631         .releasepage    = blkdev_releasepage,
1632         .direct_IO      = blkdev_direct_IO,
1633         .is_dirty_writeback = buffer_check_dirty_writeback,
1634 };
1635 
1636 const struct file_operations def_blk_fops = {
1637         .open           = blkdev_open,
1638         .release        = blkdev_close,
1639         .llseek         = block_llseek,
1640         .read           = new_sync_read,
1641         .write          = new_sync_write,
1642         .read_iter      = blkdev_read_iter,
1643         .write_iter     = blkdev_write_iter,
1644         .mmap           = generic_file_mmap,
1645         .fsync          = blkdev_fsync,
1646         .unlocked_ioctl = block_ioctl,
1647 #ifdef CONFIG_COMPAT
1648         .compat_ioctl   = compat_blkdev_ioctl,
1649 #endif
1650         .splice_read    = generic_file_splice_read,
1651         .splice_write   = iter_file_splice_write,
1652 };
1653 
1654 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1655 {
1656         int res;
1657         mm_segment_t old_fs = get_fs();
1658         set_fs(KERNEL_DS);
1659         res = blkdev_ioctl(bdev, 0, cmd, arg);
1660         set_fs(old_fs);
1661         return res;
1662 }
1663 
1664 EXPORT_SYMBOL(ioctl_by_bdev);
1665 
1666 /**
1667  * lookup_bdev  - lookup a struct block_device by name
1668  * @pathname:   special file representing the block device
1669  *
1670  * Get a reference to the blockdevice at @pathname in the current
1671  * namespace if possible and return it.  Return ERR_PTR(error)
1672  * otherwise.
1673  */
1674 struct block_device *lookup_bdev(const char *pathname)
1675 {
1676         struct block_device *bdev;
1677         struct inode *inode;
1678         struct path path;
1679         int error;
1680 
1681         if (!pathname || !*pathname)
1682                 return ERR_PTR(-EINVAL);
1683 
1684         error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1685         if (error)
1686                 return ERR_PTR(error);
1687 
1688         inode = path.dentry->d_inode;
1689         error = -ENOTBLK;
1690         if (!S_ISBLK(inode->i_mode))
1691                 goto fail;
1692         error = -EACCES;
1693         if (path.mnt->mnt_flags & MNT_NODEV)
1694                 goto fail;
1695         error = -ENOMEM;
1696         bdev = bd_acquire(inode);
1697         if (!bdev)
1698                 goto fail;
1699 out:
1700         path_put(&path);
1701         return bdev;
1702 fail:
1703         bdev = ERR_PTR(error);
1704         goto out;
1705 }
1706 EXPORT_SYMBOL(lookup_bdev);
1707 
1708 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1709 {
1710         struct super_block *sb = get_super(bdev);
1711         int res = 0;
1712 
1713         if (sb) {
1714                 /*
1715                  * no need to lock the super, get_super holds the
1716                  * read mutex so the filesystem cannot go away
1717                  * under us (->put_super runs with the write lock
1718                  * hold).
1719                  */
1720                 shrink_dcache_sb(sb);
1721                 res = invalidate_inodes(sb, kill_dirty);
1722                 drop_super(sb);
1723         }
1724         invalidate_bdev(bdev);
1725         return res;
1726 }
1727 EXPORT_SYMBOL(__invalidate_device);
1728 
1729 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1730 {
1731         struct inode *inode, *old_inode = NULL;
1732 
1733         spin_lock(&inode_sb_list_lock);
1734         list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1735                 struct address_space *mapping = inode->i_mapping;
1736 
1737                 spin_lock(&inode->i_lock);
1738                 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1739                     mapping->nrpages == 0) {
1740                         spin_unlock(&inode->i_lock);
1741                         continue;
1742                 }
1743                 __iget(inode);
1744                 spin_unlock(&inode->i_lock);
1745                 spin_unlock(&inode_sb_list_lock);
1746                 /*
1747                  * We hold a reference to 'inode' so it couldn't have been
1748                  * removed from s_inodes list while we dropped the
1749                  * inode_sb_list_lock.  We cannot iput the inode now as we can
1750                  * be holding the last reference and we cannot iput it under
1751                  * inode_sb_list_lock. So we keep the reference and iput it
1752                  * later.
1753                  */
1754                 iput(old_inode);
1755                 old_inode = inode;
1756 
1757                 func(I_BDEV(inode), arg);
1758 
1759                 spin_lock(&inode_sb_list_lock);
1760         }
1761         spin_unlock(&inode_sb_list_lock);
1762         iput(old_inode);
1763 }
1764 

~ [ 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