blob: 6d7274619bf916c2dcf0d7744ba8d888d948d711 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#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>
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -070014#include <linux/device_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/highmem.h>
16#include <linux/blkdev.h>
17#include <linux/module.h>
18#include <linux/blkpg.h>
Muthu Kumarb502bd12012-03-23 15:01:50 -070019#include <linux/magic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/buffer_head.h>
Al Viroff01bb42011-09-16 02:31:11 -040021#include <linux/swap.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010022#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010023#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/mpage.h>
25#include <linux/mount.h>
26#include <linux/uio.h>
27#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070028#include <linux/log2.h>
Al Viroff01bb42011-09-16 02:31:11 -040029#include <linux/cleancache.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070030#include <linux/aio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <asm/uaccess.h>
David Howells07f3f05c2006-09-30 20:52:18 +020032#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34struct bdev_inode {
35 struct block_device bdev;
36 struct inode vfs_inode;
37};
38
Adrian Bunk4c54ac62008-02-18 13:48:31 +010039static const struct address_space_operations def_blk_aops;
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041static inline struct bdev_inode *BDEV_I(struct inode *inode)
42{
43 return container_of(inode, struct bdev_inode, vfs_inode);
44}
45
46inline struct block_device *I_BDEV(struct inode *inode)
47{
48 return &BDEV_I(inode)->bdev;
49}
Linus Torvalds1da177e2005-04-16 15:20:36 -070050EXPORT_SYMBOL(I_BDEV);
51
Dave Chinnera5491e02010-10-21 11:49:26 +110052/*
Christoph Hellwigf758eea2011-04-21 18:19:44 -060053 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
54 * need to move it onto the dirty list of @dst so that the inode is always on
55 * the right list.
Dave Chinnera5491e02010-10-21 11:49:26 +110056 */
57static void bdev_inode_switch_bdi(struct inode *inode,
58 struct backing_dev_info *dst)
59{
Christoph Hellwigf758eea2011-04-21 18:19:44 -060060 struct backing_dev_info *old = inode->i_data.backing_dev_info;
Jan Karaa5faeaf2013-06-28 16:04:02 +020061 bool wakeup_bdi = false;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060062
63 if (unlikely(dst == old)) /* deadlock avoidance */
64 return;
65 bdi_lock_two(&old->wb, &dst->wb);
Dave Chinner250df6e2011-03-22 22:23:36 +110066 spin_lock(&inode->i_lock);
Dave Chinnera5491e02010-10-21 11:49:26 +110067 inode->i_data.backing_dev_info = dst;
Jan Karaa5faeaf2013-06-28 16:04:02 +020068 if (inode->i_state & I_DIRTY) {
69 if (bdi_cap_writeback_dirty(dst) && !wb_has_dirty_io(&dst->wb))
70 wakeup_bdi = true;
Nick Piggin7ccf19a2010-10-21 11:49:30 +110071 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
Jan Karaa5faeaf2013-06-28 16:04:02 +020072 }
Dave Chinner250df6e2011-03-22 22:23:36 +110073 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -060074 spin_unlock(&old->wb.list_lock);
75 spin_unlock(&dst->wb.list_lock);
Jan Karaa5faeaf2013-06-28 16:04:02 +020076
77 if (wakeup_bdi)
78 bdi_wakeup_thread_delayed(dst);
Dave Chinnera5491e02010-10-21 11:49:26 +110079}
80
Peter Zijlstraf9a14392007-05-06 14:49:55 -070081/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb42011-09-16 02:31:11 -040082void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083{
Al Viroff01bb42011-09-16 02:31:11 -040084 struct address_space *mapping = bdev->bd_inode->i_mapping;
85
Johannes Weiner91b0abe2014-04-03 14:47:49 -070086 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070087 return;
Al Viroff01bb42011-09-16 02:31:11 -040088
Peter Zijlstraf9a14392007-05-06 14:49:55 -070089 invalidate_bh_lrus();
Al Viroff01bb42011-09-16 02:31:11 -040090 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091}
Al Viroff01bb42011-09-16 02:31:11 -040092EXPORT_SYMBOL(kill_bdev);
93
94/* Invalidate clean unused buffers and pagecache. */
95void invalidate_bdev(struct block_device *bdev)
96{
97 struct address_space *mapping = bdev->bd_inode->i_mapping;
98
99 if (mapping->nrpages == 0)
100 return;
101
102 invalidate_bh_lrus();
103 lru_add_drain_all(); /* make sure all lru add caches are flushed */
104 invalidate_mapping_pages(mapping, 0, -1);
105 /* 99% of the time, we don't need to flush the cleancache on the bdev.
106 * But, for the strange corners, lets be cautious
107 */
Dan Magenheimer31677602011-09-21 11:56:28 -0400108 cleancache_invalidate_inode(mapping);
Al Viroff01bb42011-09-16 02:31:11 -0400109}
110EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
112int set_blocksize(struct block_device *bdev, int size)
113{
114 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700115 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 return -EINVAL;
117
118 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400119 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 return -EINVAL;
121
122 /* Don't change the size if it is same as current */
123 if (bdev->bd_block_size != size) {
124 sync_blockdev(bdev);
125 bdev->bd_block_size = size;
126 bdev->bd_inode->i_blkbits = blksize_bits(size);
127 kill_bdev(bdev);
128 }
129 return 0;
130}
131
132EXPORT_SYMBOL(set_blocksize);
133
134int sb_set_blocksize(struct super_block *sb, int size)
135{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 if (set_blocksize(sb->s_bdev, size))
137 return 0;
138 /* If we get here, we know size is power of two
139 * and it's value is between 512 and PAGE_SIZE */
140 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800141 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 return sb->s_blocksize;
143}
144
145EXPORT_SYMBOL(sb_set_blocksize);
146
147int sb_min_blocksize(struct super_block *sb, int size)
148{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400149 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 if (size < minsize)
151 size = minsize;
152 return sb_set_blocksize(sb, size);
153}
154
155EXPORT_SYMBOL(sb_min_blocksize);
156
157static int
158blkdev_get_block(struct inode *inode, sector_t iblock,
159 struct buffer_head *bh, int create)
160{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 bh->b_bdev = I_BDEV(inode);
162 bh->b_blocknr = iblock;
163 set_buffer_mapped(bh);
164 return 0;
165}
166
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800167static ssize_t
Al Virod8d3d942014-03-04 21:27:34 -0500168blkdev_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
169 loff_t offset)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800170{
171 struct file *file = iocb->ki_filp;
172 struct inode *inode = file->f_mapping->host;
173
Al Viro31b14032014-03-05 01:33:16 -0500174 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iter,
175 offset, blkdev_get_block,
Al Virod8d3d942014-03-04 21:27:34 -0500176 NULL, NULL, 0);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800177}
178
Jan Kara5cee5812009-04-27 16:43:51 +0200179int __sync_blockdev(struct block_device *bdev, int wait)
180{
181 if (!bdev)
182 return 0;
183 if (!wait)
184 return filemap_flush(bdev->bd_inode->i_mapping);
185 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
186}
187
Nick Piggin585d3bc2009-02-25 10:44:19 +0100188/*
189 * Write out and wait upon all the dirty data associated with a block
190 * device via its mapping. Does not take the superblock lock.
191 */
192int sync_blockdev(struct block_device *bdev)
193{
Jan Kara5cee5812009-04-27 16:43:51 +0200194 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100195}
196EXPORT_SYMBOL(sync_blockdev);
197
198/*
199 * Write out and wait upon all dirty data associated with this
200 * device. Filesystem data as well as the underlying block
201 * device. Takes the superblock lock.
202 */
203int fsync_bdev(struct block_device *bdev)
204{
205 struct super_block *sb = get_super(bdev);
206 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200207 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100208 drop_super(sb);
209 return res;
210 }
211 return sync_blockdev(bdev);
212}
Al Viro47e44912009-04-01 07:07:16 -0400213EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100214
215/**
216 * freeze_bdev -- lock a filesystem and force it into a consistent state
217 * @bdev: blockdevice to lock
218 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100219 * If a superblock is found on this device, we take the s_umount semaphore
220 * on it to make sure nobody unmounts until the snapshot creation is done.
221 * The reference counter (bd_fsfreeze_count) guarantees that only the last
222 * unfreeze process can unfreeze the frozen filesystem actually when multiple
223 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
224 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
225 * actually.
226 */
227struct super_block *freeze_bdev(struct block_device *bdev)
228{
229 struct super_block *sb;
230 int error = 0;
231
232 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200233 if (++bdev->bd_fsfreeze_count > 1) {
234 /*
235 * We don't even need to grab a reference - the first call
236 * to freeze_bdev grab an active reference and only the last
237 * thaw_bdev drops it.
238 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100239 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200240 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100241 mutex_unlock(&bdev->bd_fsfreeze_mutex);
242 return sb;
243 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100244
Christoph Hellwig45042302009-08-03 23:28:35 +0200245 sb = get_active_super(bdev);
246 if (!sb)
247 goto out;
Josef Bacik18e9e512010-03-23 10:34:56 -0400248 error = freeze_super(sb);
249 if (error) {
250 deactivate_super(sb);
251 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200252 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400253 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100254 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400255 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200256 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100257 sync_blockdev(bdev);
258 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200259 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100260}
261EXPORT_SYMBOL(freeze_bdev);
262
263/**
264 * thaw_bdev -- unlock filesystem
265 * @bdev: blockdevice to unlock
266 * @sb: associated superblock
267 *
268 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
269 */
270int thaw_bdev(struct block_device *bdev, struct super_block *sb)
271{
Christoph Hellwig45042302009-08-03 23:28:35 +0200272 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100273
274 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200275 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400276 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100277
Christoph Hellwig45042302009-08-03 23:28:35 +0200278 error = 0;
279 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400280 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100281
Christoph Hellwig45042302009-08-03 23:28:35 +0200282 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400283 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200284
Josef Bacik18e9e512010-03-23 10:34:56 -0400285 error = thaw_super(sb);
286 if (error) {
287 bdev->bd_fsfreeze_count++;
288 mutex_unlock(&bdev->bd_fsfreeze_mutex);
289 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100290 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400291out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100292 mutex_unlock(&bdev->bd_fsfreeze_mutex);
293 return 0;
294}
295EXPORT_SYMBOL(thaw_bdev);
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
298{
299 return block_write_full_page(page, blkdev_get_block, wbc);
300}
301
302static int blkdev_readpage(struct file * file, struct page * page)
303{
304 return block_read_full_page(page, blkdev_get_block);
305}
306
Nick Piggin6272b5a2007-10-16 01:25:04 -0700307static int blkdev_write_begin(struct file *file, struct address_space *mapping,
308 loff_t pos, unsigned len, unsigned flags,
309 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200311 return block_write_begin(mapping, pos, len, flags, pagep,
312 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313}
314
Nick Piggin6272b5a2007-10-16 01:25:04 -0700315static int blkdev_write_end(struct file *file, struct address_space *mapping,
316 loff_t pos, unsigned len, unsigned copied,
317 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700319 int ret;
320 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
321
322 unlock_page(page);
323 page_cache_release(page);
324
325 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
328/*
329 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500330 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 * so we compute the size by hand (just as in block_read/write above)
332 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800333static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 struct inode *bd_inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 loff_t retval;
337
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800338 mutex_lock(&bd_inode->i_mutex);
Al Viro5d48f3a2013-06-23 21:34:45 +0400339 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800340 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 return retval;
342}
343
Josef Bacik02c24a82011-07-16 20:44:56 -0400344int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400346 struct inode *bd_inode = filp->f_mapping->host;
347 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100348 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200349
350 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
351 if (error)
352 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100353
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400354 /*
355 * There is no need to serialise calls to blkdev_issue_flush with
356 * i_mutex and doing so causes performance issues with concurrent
357 * O_SYNC writers to a block device.
358 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200359 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100360 if (error == -EOPNOTSUPP)
361 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400362
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100363 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700365EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700367/**
368 * bdev_read_page() - Start reading a page from a block device
369 * @bdev: The device to read the page from
370 * @sector: The offset on the device to read the page to (need not be aligned)
371 * @page: The page to read
372 *
373 * On entry, the page should be locked. It will be unlocked when the page
374 * has been read. If the block driver implements rw_page synchronously,
375 * that will be true on exit from this function, but it need not be.
376 *
377 * Errors returned by this function are usually "soft", eg out of memory, or
378 * queue full; callers should try a different route to read this page rather
379 * than propagate an error back up the stack.
380 *
381 * Return: negative errno if an error occurs, 0 if submission was successful.
382 */
383int bdev_read_page(struct block_device *bdev, sector_t sector,
384 struct page *page)
385{
386 const struct block_device_operations *ops = bdev->bd_disk->fops;
387 if (!ops->rw_page)
388 return -EOPNOTSUPP;
389 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
390}
391EXPORT_SYMBOL_GPL(bdev_read_page);
392
393/**
394 * bdev_write_page() - Start writing a page to a block device
395 * @bdev: The device to write the page to
396 * @sector: The offset on the device to write the page to (need not be aligned)
397 * @page: The page to write
398 * @wbc: The writeback_control for the write
399 *
400 * On entry, the page should be locked and not currently under writeback.
401 * On exit, if the write started successfully, the page will be unlocked and
402 * under writeback. If the write failed already (eg the driver failed to
403 * queue the page to the device), the page will still be locked. If the
404 * caller is a ->writepage implementation, it will need to unlock the page.
405 *
406 * Errors returned by this function are usually "soft", eg out of memory, or
407 * queue full; callers should try a different route to write this page rather
408 * than propagate an error back up the stack.
409 *
410 * Return: negative errno if an error occurs, 0 if submission was successful.
411 */
412int bdev_write_page(struct block_device *bdev, sector_t sector,
413 struct page *page, struct writeback_control *wbc)
414{
415 int result;
416 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
417 const struct block_device_operations *ops = bdev->bd_disk->fops;
418 if (!ops->rw_page)
419 return -EOPNOTSUPP;
420 set_page_writeback(page);
421 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
422 if (result)
423 end_page_writeback(page);
424 else
425 unlock_page(page);
426 return result;
427}
428EXPORT_SYMBOL_GPL(bdev_write_page);
429
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430/*
431 * pseudo-fs
432 */
433
434static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800435static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437static struct inode *bdev_alloc_inode(struct super_block *sb)
438{
Christoph Lametere94b1762006-12-06 20:33:17 -0800439 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 if (!ei)
441 return NULL;
442 return &ei->vfs_inode;
443}
444
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100445static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100447 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 struct bdev_inode *bdi = BDEV_I(inode);
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 kmem_cache_free(bdev_cachep, bdi);
451}
452
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100453static void bdev_destroy_inode(struct inode *inode)
454{
455 call_rcu(&inode->i_rcu, bdev_i_callback);
456}
457
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700458static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
460 struct bdev_inode *ei = (struct bdev_inode *) foo;
461 struct block_device *bdev = &ei->bdev;
462
Christoph Lametera35afb82007-05-16 22:10:57 -0700463 memset(bdev, 0, sizeof(*bdev));
464 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700465 INIT_LIST_HEAD(&bdev->bd_inodes);
466 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100467#ifdef CONFIG_SYSFS
468 INIT_LIST_HEAD(&bdev->bd_holder_disks);
469#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700470 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800471 /* Initialize mutex for freeze. */
472 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
475static inline void __bd_forget(struct inode *inode)
476{
477 list_del_init(&inode->i_devices);
478 inode->i_bdev = NULL;
479 inode->i_mapping = &inode->i_data;
480}
481
Al Virob57922d2010-06-07 14:34:48 -0400482static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
484 struct block_device *bdev = &BDEV_I(inode)->bdev;
485 struct list_head *p;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700486 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400487 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200488 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 spin_lock(&bdev_lock);
490 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
491 __bd_forget(list_entry(p, struct inode, i_devices));
492 }
493 list_del_init(&bdev->bd_list);
494 spin_unlock(&bdev_lock);
495}
496
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800497static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 .statfs = simple_statfs,
499 .alloc_inode = bdev_alloc_inode,
500 .destroy_inode = bdev_destroy_inode,
501 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400502 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503};
504
Al Viro51139ad2010-07-25 23:47:46 +0400505static struct dentry *bd_mount(struct file_system_type *fs_type,
506 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
Muthu Kumarb502bd12012-03-23 15:01:50 -0700508 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509}
510
511static struct file_system_type bd_type = {
512 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400513 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 .kill_sb = kill_anon_super,
515};
516
Al Virof47ec3f2011-11-21 21:15:42 -0500517static struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
519void __init bdev_cache_init(void)
520{
521 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300522 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800523
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800525 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
526 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900527 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 err = register_filesystem(&bd_type);
529 if (err)
530 panic("Cannot register bdev pseudo-fs");
531 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 if (IS_ERR(bd_mnt))
533 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300534 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
536
537/*
538 * Most likely _very_ bad one - but then it's hardly critical for small
539 * /dev and can be fixed when somebody will need really large one.
540 * Keep in mind that it will be fed through icache hash function too.
541 */
542static inline unsigned long hash(dev_t dev)
543{
544 return MAJOR(dev)+MINOR(dev);
545}
546
547static int bdev_test(struct inode *inode, void *data)
548{
549 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
550}
551
552static int bdev_set(struct inode *inode, void *data)
553{
554 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
555 return 0;
556}
557
558static LIST_HEAD(all_bdevs);
559
560struct block_device *bdget(dev_t dev)
561{
562 struct block_device *bdev;
563 struct inode *inode;
564
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800565 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 bdev_test, bdev_set, &dev);
567
568 if (!inode)
569 return NULL;
570
571 bdev = &BDEV_I(inode)->bdev;
572
573 if (inode->i_state & I_NEW) {
574 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000575 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 bdev->bd_inode = inode;
577 bdev->bd_block_size = (1 << inode->i_blkbits);
578 bdev->bd_part_count = 0;
579 bdev->bd_invalidated = 0;
580 inode->i_mode = S_IFBLK;
581 inode->i_rdev = dev;
582 inode->i_bdev = bdev;
583 inode->i_data.a_ops = &def_blk_aops;
584 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
585 inode->i_data.backing_dev_info = &default_backing_dev_info;
586 spin_lock(&bdev_lock);
587 list_add(&bdev->bd_list, &all_bdevs);
588 spin_unlock(&bdev_lock);
589 unlock_new_inode(inode);
590 }
591 return bdev;
592}
593
594EXPORT_SYMBOL(bdget);
595
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200596/**
597 * bdgrab -- Grab a reference to an already referenced block device
598 * @bdev: Block device to grab a reference to.
599 */
600struct block_device *bdgrab(struct block_device *bdev)
601{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400602 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200603 return bdev;
604}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700605EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607long nr_blockdev_pages(void)
608{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700609 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 long ret = 0;
611 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700612 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 ret += bdev->bd_inode->i_mapping->nrpages;
614 }
615 spin_unlock(&bdev_lock);
616 return ret;
617}
618
619void bdput(struct block_device *bdev)
620{
621 iput(bdev->bd_inode);
622}
623
624EXPORT_SYMBOL(bdput);
625
626static struct block_device *bd_acquire(struct inode *inode)
627{
628 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 spin_lock(&bdev_lock);
631 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700632 if (bdev) {
Al Viro7de9c6ee2010-10-23 11:11:40 -0400633 ihold(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 spin_unlock(&bdev_lock);
635 return bdev;
636 }
637 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 bdev = bdget(inode->i_rdev);
640 if (bdev) {
641 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700642 if (!inode->i_bdev) {
643 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400644 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700645 * and it's released in clear_inode() of inode.
646 * So, we can access it via ->i_mapping always
647 * without igrab().
648 */
Al Viro7de9c6ee2010-10-23 11:11:40 -0400649 ihold(bdev->bd_inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700650 inode->i_bdev = bdev;
651 inode->i_mapping = bdev->bd_inode->i_mapping;
652 list_add(&inode->i_devices, &bdev->bd_inodes);
653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 spin_unlock(&bdev_lock);
655 }
656 return bdev;
657}
658
Jan Karaa8855992013-07-09 22:36:45 +0800659int sb_is_blkdev_sb(struct super_block *sb)
Al Virof47ec3f2011-11-21 21:15:42 -0500660{
661 return sb == blockdev_superblock;
662}
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664/* Call when you free inode */
665
666void bd_forget(struct inode *inode)
667{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700668 struct block_device *bdev = NULL;
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700671 if (!sb_is_blkdev_sb(inode->i_sb))
672 bdev = inode->i_bdev;
673 __bd_forget(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700675
676 if (bdev)
677 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678}
679
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900680/**
681 * bd_may_claim - test whether a block device can be claimed
682 * @bdev: block device of interest
683 * @whole: whole block device containing @bdev, may equal @bdev
684 * @holder: holder trying to claim @bdev
685 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300686 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900687 *
688 * CONTEXT:
689 * spin_lock(&bdev_lock).
690 *
691 * RETURNS:
692 * %true if @bdev can be claimed, %false otherwise.
693 */
694static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
695 void *holder)
696{
697 if (bdev->bd_holder == holder)
698 return true; /* already a holder */
699 else if (bdev->bd_holder != NULL)
700 return false; /* held by someone else */
701 else if (bdev->bd_contains == bdev)
702 return true; /* is a whole device which isn't held */
703
Tejun Heoe525fd82010-11-13 11:55:17 +0100704 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900705 return true; /* is a partition of a device that is being partitioned */
706 else if (whole->bd_holder != NULL)
707 return false; /* is a partition of a held device */
708 else
709 return true; /* is a partition of an un-held device */
710}
711
712/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900713 * bd_prepare_to_claim - prepare to claim a block device
714 * @bdev: block device of interest
715 * @whole: the whole device containing @bdev, may equal @bdev
716 * @holder: holder trying to claim @bdev
717 *
718 * Prepare to claim @bdev. This function fails if @bdev is already
719 * claimed by another holder and waits if another claiming is in
720 * progress. This function doesn't actually claim. On successful
721 * return, the caller has ownership of bd_claiming and bd_holder[s].
722 *
723 * CONTEXT:
724 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
725 * it multiple times.
726 *
727 * RETURNS:
728 * 0 if @bdev can be claimed, -EBUSY otherwise.
729 */
730static int bd_prepare_to_claim(struct block_device *bdev,
731 struct block_device *whole, void *holder)
732{
733retry:
734 /* if someone else claimed, fail */
735 if (!bd_may_claim(bdev, whole, holder))
736 return -EBUSY;
737
Tejun Heoe75aa852010-08-04 17:59:39 +0200738 /* if claiming is already in progress, wait for it to finish */
739 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900740 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
741 DEFINE_WAIT(wait);
742
743 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
744 spin_unlock(&bdev_lock);
745 schedule();
746 finish_wait(wq, &wait);
747 spin_lock(&bdev_lock);
748 goto retry;
749 }
750
751 /* yay, all mine */
752 return 0;
753}
754
755/**
756 * bd_start_claiming - start claiming a block device
757 * @bdev: block device of interest
758 * @holder: holder trying to claim @bdev
759 *
760 * @bdev is about to be opened exclusively. Check @bdev can be opened
761 * exclusively and mark that an exclusive open is in progress. Each
762 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000763 * either bd_finish_claiming() or bd_abort_claiming() (which do not
764 * fail).
765 *
766 * This function is used to gain exclusive access to the block device
767 * without actually causing other exclusive open attempts to fail. It
768 * should be used when the open sequence itself requires exclusive
769 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900770 *
771 * CONTEXT:
772 * Might sleep.
773 *
774 * RETURNS:
775 * Pointer to the block device containing @bdev on success, ERR_PTR()
776 * value on failure.
777 */
778static struct block_device *bd_start_claiming(struct block_device *bdev,
779 void *holder)
780{
781 struct gendisk *disk;
782 struct block_device *whole;
783 int partno, err;
784
785 might_sleep();
786
787 /*
788 * @bdev might not have been initialized properly yet, look up
789 * and grab the outer block device the hard way.
790 */
791 disk = get_gendisk(bdev->bd_dev, &partno);
792 if (!disk)
793 return ERR_PTR(-ENXIO);
794
Tejun Heod4c208b2011-06-13 12:45:48 +0200795 /*
796 * Normally, @bdev should equal what's returned from bdget_disk()
797 * if partno is 0; however, some drivers (floppy) use multiple
798 * bdev's for the same physical device and @bdev may be one of the
799 * aliases. Keep @bdev if partno is 0. This means claimer
800 * tracking is broken for those devices but it has always been that
801 * way.
802 */
803 if (partno)
804 whole = bdget_disk(disk, 0);
805 else
806 whole = bdgrab(bdev);
807
Nick Piggincf342572010-05-26 01:50:21 +1000808 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900809 put_disk(disk);
810 if (!whole)
811 return ERR_PTR(-ENOMEM);
812
813 /* prepare to claim, if successful, mark claiming in progress */
814 spin_lock(&bdev_lock);
815
816 err = bd_prepare_to_claim(bdev, whole, holder);
817 if (err == 0) {
818 whole->bd_claiming = holder;
819 spin_unlock(&bdev_lock);
820 return whole;
821 } else {
822 spin_unlock(&bdev_lock);
823 bdput(whole);
824 return ERR_PTR(err);
825 }
826}
827
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800828#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100829struct bd_holder_disk {
830 struct list_head list;
831 struct gendisk *disk;
832 int refcnt;
833};
834
835static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
836 struct gendisk *disk)
837{
838 struct bd_holder_disk *holder;
839
840 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
841 if (holder->disk == disk)
842 return holder;
843 return NULL;
844}
845
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700846static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800847{
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700848 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800849}
850
851static void del_symlink(struct kobject *from, struct kobject *to)
852{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800853 sysfs_remove_link(from, kobject_name(to));
854}
855
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800856/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100857 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
858 * @bdev: the claimed slave bdev
859 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500860 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100861 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
862 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100863 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500864 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100865 * - from "slaves" directory of the holder @disk to the claimed @bdev
866 * - from "holders" directory of the @bdev to the holder @disk
867 *
868 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
869 * passed to bd_link_disk_holder(), then:
870 *
871 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
872 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
873 *
874 * The caller must have claimed @bdev before calling this function and
875 * ensure that both @bdev and @disk are valid during the creation and
876 * lifetime of these symlinks.
877 *
878 * CONTEXT:
879 * Might sleep.
880 *
881 * RETURNS:
882 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500883 */
Tejun Heoe09b4572010-11-13 11:55:17 +0100884int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500885{
Tejun Heo49731ba2011-01-14 18:43:57 +0100886 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100887 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800888
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800889 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500890
Tejun Heo49731ba2011-01-14 18:43:57 +0100891 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -0700892
Tejun Heoe09b4572010-11-13 11:55:17 +0100893 /* FIXME: remove the following once add_disk() handles errors */
894 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
895 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -0700896
Tejun Heo49731ba2011-01-14 18:43:57 +0100897 holder = bd_find_holder_disk(bdev, disk);
898 if (holder) {
899 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +0100900 goto out_unlock;
901 }
902
Tejun Heo49731ba2011-01-14 18:43:57 +0100903 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
904 if (!holder) {
905 ret = -ENOMEM;
906 goto out_unlock;
907 }
908
909 INIT_LIST_HEAD(&holder->list);
910 holder->disk = disk;
911 holder->refcnt = 1;
912
913 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
914 if (ret)
915 goto out_free;
916
917 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
918 if (ret)
919 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +0100920 /*
921 * bdev could be deleted beneath us which would implicitly destroy
922 * the holder directory. Hold on to it.
923 */
924 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100925
926 list_add(&holder->list, &bdev->bd_holder_disks);
927 goto out_unlock;
928
929out_del:
930 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
931out_free:
932 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +0100933out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800934 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +0100935 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800936}
Tejun Heoe09b4572010-11-13 11:55:17 +0100937EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800938
Tejun Heo49731ba2011-01-14 18:43:57 +0100939/**
940 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
941 * @bdev: the calimed slave bdev
942 * @disk: the holding disk
943 *
944 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
945 *
946 * CONTEXT:
947 * Might sleep.
948 */
949void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800950{
Tejun Heo49731ba2011-01-14 18:43:57 +0100951 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100952
Tejun Heo49731ba2011-01-14 18:43:57 +0100953 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800954
Tejun Heo49731ba2011-01-14 18:43:57 +0100955 holder = bd_find_holder_disk(bdev, disk);
956
957 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
958 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
959 del_symlink(bdev->bd_part->holder_dir,
960 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +0100961 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100962 list_del_init(&holder->list);
963 kfree(holder);
964 }
965
966 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800967}
Tejun Heo49731ba2011-01-14 18:43:57 +0100968EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800969#endif
970
Andrew Patterson0c002c22008-09-04 14:27:20 -0600971/**
Andrew Patterson56ade442008-09-04 14:27:40 -0600972 * flush_disk - invalidates all buffer-cache entries on a disk
973 *
974 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -0800975 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -0600976 *
977 * Invalidates all buffer-cache entries on a disk. It should be called
978 * when a disk has been changed -- either by a media change or online
979 * resize.
980 */
NeilBrown93b270f2011-02-24 17:25:47 +1100981static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -0600982{
NeilBrown93b270f2011-02-24 17:25:47 +1100983 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -0600984 char name[BDEVNAME_SIZE] = "";
985
986 if (bdev->bd_disk)
987 disk_name(bdev->bd_disk, 0, name);
988 printk(KERN_WARNING "VFS: busy inodes on changed media or "
989 "resized disk %s\n", name);
990 }
991
992 if (!bdev->bd_disk)
993 return;
Tejun Heod27769e2011-08-23 20:01:04 +0200994 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -0600995 bdev->bd_invalidated = 1;
996}
997
998/**
Randy Dunlap57d1b532008-10-09 10:42:38 +0200999 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001000 * @disk: struct gendisk to check
1001 * @bdev: struct bdev to adjust.
1002 *
1003 * This routine checks to see if the bdev size does not match the disk size
1004 * and adjusts it if it differs.
1005 */
1006void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1007{
1008 loff_t disk_size, bdev_size;
1009
1010 disk_size = (loff_t)get_capacity(disk) << 9;
1011 bdev_size = i_size_read(bdev->bd_inode);
1012 if (disk_size != bdev_size) {
1013 char name[BDEVNAME_SIZE];
1014
1015 disk_name(disk, 0, name);
1016 printk(KERN_INFO
1017 "%s: detected capacity change from %lld to %lld\n",
1018 name, bdev_size, disk_size);
1019 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001020 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001021 }
1022}
1023EXPORT_SYMBOL(check_disk_size_change);
1024
1025/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001026 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001027 * @disk: struct gendisk to be revalidated
1028 *
1029 * This routine is a wrapper for lower-level driver's revalidate_disk
1030 * call-backs. It is used to do common pre and post operations needed
1031 * for all revalidate_disk operations.
1032 */
1033int revalidate_disk(struct gendisk *disk)
1034{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001035 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001036 int ret = 0;
1037
1038 if (disk->fops->revalidate_disk)
1039 ret = disk->fops->revalidate_disk(disk);
1040
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001041 bdev = bdget_disk(disk, 0);
1042 if (!bdev)
1043 return ret;
1044
1045 mutex_lock(&bdev->bd_mutex);
1046 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001047 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001048 mutex_unlock(&bdev->bd_mutex);
1049 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001050 return ret;
1051}
1052EXPORT_SYMBOL(revalidate_disk);
1053
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054/*
1055 * This routine checks whether a removable media has been changed,
1056 * and invalidates all buffer-cache-entries in that case. This
1057 * is a relatively slow routine, so we have to try to minimize using
1058 * it. Thus it is called only upon a 'mount' or 'open'. This
1059 * is the best way of combining speed and utility, I think.
1060 * People changing diskettes in the middle of an operation deserve
1061 * to lose :-)
1062 */
1063int check_disk_change(struct block_device *bdev)
1064{
1065 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001066 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001067 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
Tejun Heo77ea8872010-12-08 20:57:37 +01001069 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1070 DISK_EVENT_EJECT_REQUEST);
1071 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 return 0;
1073
NeilBrown93b270f2011-02-24 17:25:47 +11001074 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 if (bdops->revalidate_disk)
1076 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 return 1;
1078}
1079
1080EXPORT_SYMBOL(check_disk_change);
1081
1082void bd_set_size(struct block_device *bdev, loff_t size)
1083{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001084 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
Guo Chaod646a022013-02-21 15:16:42 -08001086 mutex_lock(&bdev->bd_inode->i_mutex);
1087 i_size_write(bdev->bd_inode, size);
1088 mutex_unlock(&bdev->bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 while (bsize < PAGE_CACHE_SIZE) {
1090 if (size & bsize)
1091 break;
1092 bsize <<= 1;
1093 }
1094 bdev->bd_block_size = bsize;
1095 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1096}
1097EXPORT_SYMBOL(bd_set_size);
1098
Al Viro4385bab2013-05-05 22:11:03 -04001099static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001100
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001101/*
1102 * bd_mutex locking:
1103 *
1104 * mutex_lock(part->bd_mutex)
1105 * mutex_lock_nested(whole->bd_mutex, 1)
1106 */
1107
Al Viro572c4892007-10-08 13:24:05 -04001108static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001111 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001112 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001113 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001114 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
Al Viro572c4892007-10-08 13:24:05 -04001116 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001117 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001118 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001119 perm |= MAY_WRITE;
1120 /*
1121 * hooks: /n/, see "layering violations".
1122 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001123 if (!for_part) {
1124 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1125 if (ret != 0) {
1126 bdput(bdev);
1127 return ret;
1128 }
Al Viro82666022008-08-01 05:32:04 -04001129 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001130
NeilBrownd3374822009-01-09 08:31:10 +11001131 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001132
Tejun Heo89f97492008-11-05 10:21:06 +01001133 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001134 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001135 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001136 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001137 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Tejun Heo69e02c52011-03-09 19:54:27 +01001139 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001140 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 if (!bdev->bd_openers) {
1142 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001143 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001145 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001147
1148 ret = -ENXIO;
1149 bdev->bd_part = disk_get_part(disk, partno);
1150 if (!bdev->bd_part)
1151 goto out_clear;
1152
Tejun Heo1196f8b2011-04-21 20:54:45 +02001153 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001155 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001156 if (ret == -ERESTARTSYS) {
1157 /* Lost a race with 'disk' being
1158 * deleted, try again.
1159 * See md.c
1160 */
1161 disk_put_part(bdev->bd_part);
1162 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001163 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001164 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001165 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001166 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001167 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001168 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001169 goto restart;
1170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 }
Tejun Heo7e697232011-05-23 13:26:07 +02001172
Guo Chaode331272013-02-21 15:16:43 -08001173 if (!ret) {
Tejun Heo7e697232011-05-23 13:26:07 +02001174 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1175 bdi = blk_get_backing_dev_info(bdev);
1176 if (bdi == NULL)
1177 bdi = &default_backing_dev_info;
1178 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1179 }
1180
Tejun Heo1196f8b2011-04-21 20:54:45 +02001181 /*
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 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001187 if (bdev->bd_invalidated) {
1188 if (!ret)
1189 rescan_partitions(disk, bdev);
1190 else if (ret == -ENOMEDIUM)
1191 invalidate_partitions(disk, bdev);
1192 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001193 if (ret)
1194 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 struct block_device *whole;
1197 whole = bdget_disk(disk, 0);
1198 ret = -ENOMEM;
1199 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001200 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001201 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001202 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001204 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 bdev->bd_contains = whole;
Dave Chinnera5491e02010-10-21 11:49:26 +11001206 bdev_inode_switch_bdi(bdev->bd_inode,
1207 whole->bd_inode->i_data.backing_dev_info);
Tejun Heo89f97492008-11-05 10:21:06 +01001208 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001209 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001210 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001212 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 }
Tejun Heo89f97492008-11-05 10:21:06 +01001214 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001218 ret = 0;
1219 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001220 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001221 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001222 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 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001228 if (ret)
1229 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001231 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001232 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001233 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 }
1235 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001236 if (for_part)
1237 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001238 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001239 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 return 0;
1241
Tejun Heo0762b8b2008-08-25 19:56:12 +09001242 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001243 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001245 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001246 bdev->bd_queue = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001247 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001249 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001251 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001252 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001253 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001254 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001255 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001256 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001257 bdput(bdev);
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 return ret;
1260}
1261
Tejun Heod4d77622010-11-13 11:55:18 +01001262/**
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 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001281int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282{
Tejun Heoe525fd82010-11-13 11:55:17 +01001283 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) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001299 struct gendisk *disk = whole->bd_disk;
1300
Tejun Heo6a027ef2010-11-13 11:55:17 +01001301 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001302 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001303 spin_lock(&bdev_lock);
1304
Tejun Heo77ea8872010-12-08 20:57:37 +01001305 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001306 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);
Tejun Heo77ea8872010-12-08 20:57:37 +01001325
1326 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001327 * 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().
Tejun Heo77ea8872010-12-08 20:57:37 +01001332 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001333 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1334 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001335 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001336 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001337 }
1338
1339 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001340 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001341 }
1342
1343 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001344}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345EXPORT_SYMBOL(blkdev_get);
1346
Tejun Heod4d77622010-11-13 11:55:18 +01001347/**
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 */
1364struct 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
Chuck Ebberte51900f2011-02-16 18:11:53 -05001378 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1379 blkdev_put(bdev, mode);
1380 return ERR_PTR(-EACCES);
1381 }
1382
Tejun Heod4d77622010-11-13 11:55:18 +01001383 return bdev;
1384}
1385EXPORT_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 */
1409struct 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}
1424EXPORT_SYMBOL(blkdev_get_by_dev);
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426static int blkdev_open(struct inode * inode, struct file * filp)
1427{
1428 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
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
Al Viro572c4892007-10-08 13:24:05 -04001438 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
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001446 if (bdev == NULL)
1447 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Al Viro572c4892007-10-08 13:24:05 -04001449 filp->f_mapping = bdev->bd_inode->i_mapping;
1450
Tejun Heoe525fd82010-11-13 11:55:17 +01001451 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453
Al Viro4385bab2013-05-05 22:11:03 -04001454static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001455{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001456 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001457 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001458
NeilBrown6796bf52006-12-08 02:36:16 -08001459 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001460 if (for_part)
1461 bdev->bd_part_count--;
1462
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001463 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001464 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001465 sync_blockdev(bdev);
1466 kill_bdev(bdev);
NeilBrown94007752011-09-10 17:20:21 +10001467 /* ->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);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001472 }
1473 if (bdev->bd_contains == bdev) {
1474 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001475 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001476 }
1477 if (!bdev->bd_openers) {
1478 struct module *owner = disk->fops->owner;
1479
Tejun Heo0762b8b2008-08-25 19:56:12 +09001480 disk_put_part(bdev->bd_part);
1481 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001482 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001483 if (bdev != bdev->bd_contains)
1484 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001485 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001486
1487 put_disk(disk);
1488 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001489 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001490 mutex_unlock(&bdev->bd_mutex);
1491 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001492 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001493 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001494}
1495
Al Viro4385bab2013-05-05 22:11:03 -04001496void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001497{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001498 mutex_lock(&bdev->bd_mutex);
1499
Tejun Heoe525fd82010-11-13 11:55:17 +01001500 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001501 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 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001508 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
Tejun Heo77ea8872010-12-08 20:57:37 +01001521 /*
1522 * If this was the last claim, remove holder link and
1523 * unblock evpoll if it was a write holder.
1524 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001525 if (bdev_free && bdev->bd_write_holder) {
1526 disk_unblock_events(bdev->bd_disk);
1527 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001528 }
Tejun Heo69362172011-03-09 19:54:27 +01001529 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001530
Tejun Heo85ef06d2011-07-01 16:17:47 +02001531 /*
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
Al Viro4385bab2013-05-05 22:11:03 -04001540 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001541}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001542EXPORT_SYMBOL(blkdev_put);
1543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544static int blkdev_close(struct inode * inode, struct file * filp)
1545{
1546 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
Al Viro4385bab2013-05-05 22:11:03 -04001547 blkdev_put(bdev, filp->f_mode);
1548 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549}
1550
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001551static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552{
Al Viro56b26ad2008-09-19 03:17:36 -04001553 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1554 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001555
1556 /*
1557 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1558 * to updated it before every ioctl.
1559 */
Al Viro56b26ad2008-09-19 03:17:36 -04001560 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001561 mode |= FMODE_NDELAY;
1562 else
1563 mode &= ~FMODE_NDELAY;
1564
Al Viro56b26ad2008-09-19 03:17:36 -04001565 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566}
1567
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001568/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001569 * 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 */
Al Viro1456c0a2014-04-03 03:21:50 -04001575ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001576{
1577 struct file *file = iocb->ki_filp;
Jianpeng Ma53362a02012-08-02 09:50:39 +02001578 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001579 ssize_t ret;
1580
Jianpeng Ma53362a02012-08-02 09:50:39 +02001581 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001582 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwig02afc272013-09-04 15:04:40 +02001583 if (ret > 0) {
Christoph Hellwigeef99382009-08-20 17:43:41 +02001584 ssize_t err;
Al Viro1456c0a2014-04-03 03:21:50 -04001585 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
Dan Carpenter45d4f852014-04-03 14:47:17 -07001586 if (err < 0)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001587 ret = err;
1588 }
Jianpeng Ma53362a02012-08-02 09:50:39 +02001589 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001590 return ret;
1591}
Al Viro1456c0a2014-04-03 03:21:50 -04001592EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001593
Al Viroa8860382014-04-02 20:02:21 -04001594static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001595{
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);
Al Viroa8860382014-04-02 20:02:21 -04001599 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001600
1601 if (pos >= size)
1602 return 0;
1603
1604 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001605 iov_iter_truncate(to, size);
1606 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001607}
1608
Christoph Hellwigeef99382009-08-20 17:43:41 +02001609/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001610 * Try to release a page associated with block device when the system
1611 * is under memory pressure.
1612 */
1613static int blkdev_releasepage(struct page *page, gfp_t wait)
1614{
1615 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1616
1617 if (super && super->s_op->bdev_try_to_free_page)
1618 return super->s_op->bdev_try_to_free_page(super, page, wait);
1619
1620 return try_to_free_buffers(page);
1621}
1622
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001623static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 .readpage = blkdev_readpage,
1625 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001626 .write_begin = blkdev_write_begin,
1627 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001629 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001631 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632};
1633
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001634const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 .open = blkdev_open,
1636 .release = blkdev_close,
1637 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001638 .read = new_sync_read,
Al Viro1456c0a2014-04-03 03:21:50 -04001639 .write = new_sync_write,
Al Viroa8860382014-04-02 20:02:21 -04001640 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001641 .write_iter = blkdev_write_iter,
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001642 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001643 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001644 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645#ifdef CONFIG_COMPAT
1646 .compat_ioctl = compat_blkdev_ioctl,
1647#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001648 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001649 .splice_write = iter_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650};
1651
1652int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1653{
1654 int res;
1655 mm_segment_t old_fs = get_fs();
1656 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001657 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 set_fs(old_fs);
1659 return res;
1660}
1661
1662EXPORT_SYMBOL(ioctl_by_bdev);
1663
1664/**
1665 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001666 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001668 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 * namespace if possible and return it. Return ERR_PTR(error)
1670 * otherwise.
1671 */
Al Viro421748e2008-08-02 01:04:36 -04001672struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673{
1674 struct block_device *bdev;
1675 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001676 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 int error;
1678
Al Viro421748e2008-08-02 01:04:36 -04001679 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 return ERR_PTR(-EINVAL);
1681
Al Viro421748e2008-08-02 01:04:36 -04001682 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 if (error)
1684 return ERR_PTR(error);
1685
Al Viro421748e2008-08-02 01:04:36 -04001686 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 error = -ENOTBLK;
1688 if (!S_ISBLK(inode->i_mode))
1689 goto fail;
1690 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001691 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 goto fail;
1693 error = -ENOMEM;
1694 bdev = bd_acquire(inode);
1695 if (!bdev)
1696 goto fail;
1697out:
Al Viro421748e2008-08-02 01:04:36 -04001698 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 return bdev;
1700fail:
1701 bdev = ERR_PTR(error);
1702 goto out;
1703}
Al Virod5686b42008-08-01 05:00:11 -04001704EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
NeilBrown93b270f2011-02-24 17:25:47 +11001706int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001707{
1708 struct super_block *sb = get_super(bdev);
1709 int res = 0;
1710
1711 if (sb) {
1712 /*
1713 * no need to lock the super, get_super holds the
1714 * read mutex so the filesystem cannot go away
1715 * under us (->put_super runs with the write lock
1716 * hold).
1717 */
1718 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001719 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001720 drop_super(sb);
1721 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001722 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001723 return res;
1724}
1725EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001726
1727void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1728{
1729 struct inode *inode, *old_inode = NULL;
1730
1731 spin_lock(&inode_sb_list_lock);
1732 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1733 struct address_space *mapping = inode->i_mapping;
1734
1735 spin_lock(&inode->i_lock);
1736 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1737 mapping->nrpages == 0) {
1738 spin_unlock(&inode->i_lock);
1739 continue;
1740 }
1741 __iget(inode);
1742 spin_unlock(&inode->i_lock);
1743 spin_unlock(&inode_sb_list_lock);
1744 /*
1745 * We hold a reference to 'inode' so it couldn't have been
1746 * removed from s_inodes list while we dropped the
1747 * inode_sb_list_lock. We cannot iput the inode now as we can
1748 * be holding the last reference and we cannot iput it under
1749 * inode_sb_list_lock. So we keep the reference and iput it
1750 * later.
1751 */
1752 iput(old_inode);
1753 old_inode = inode;
1754
1755 func(I_BDEV(inode), arg);
1756
1757 spin_lock(&inode_sb_list_lock);
1758 }
1759 spin_unlock(&inode_sb_list_lock);
1760 iput(old_inode);
1761}