blob: cb936c90ae82eeb49fa92f4b6ecbe0da0d0f402a [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>
Tejun Heo66114ca2015-05-22 17:13:32 -040017#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/module.h>
19#include <linux/blkpg.h>
Muthu Kumarb502bd12012-03-23 15:01:50 -070020#include <linux/magic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/buffer_head.h>
Al Viroff01bb482011-09-16 02:31:11 -040022#include <linux/swap.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010023#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010024#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/mpage.h>
26#include <linux/mount.h>
27#include <linux/uio.h>
28#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070029#include <linux/log2.h>
Al Viroff01bb482011-09-16 02:31:11 -040030#include <linux/cleancache.h>
Matthew Wilcoxc94c2ac2015-09-08 14:58:40 -070031#include <linux/dax.h>
Dan Williamsacc93d32016-05-07 11:40:28 -070032#include <linux/badblocks.h>
Darrick J. Wong25f4c412016-10-11 13:51:11 -070033#include <linux/falloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020035#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37struct bdev_inode {
38 struct block_device bdev;
39 struct inode vfs_inode;
40};
41
Adrian Bunk4c54ac62008-02-18 13:48:31 +010042static const struct address_space_operations def_blk_aops;
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044static inline struct bdev_inode *BDEV_I(struct inode *inode)
45{
46 return container_of(inode, struct bdev_inode, vfs_inode);
47}
48
Geert Uytterhoevenff5053f2015-06-26 13:58:32 +020049struct block_device *I_BDEV(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070050{
51 return &BDEV_I(inode)->bdev;
52}
Linus Torvalds1da177e2005-04-16 15:20:36 -070053EXPORT_SYMBOL(I_BDEV);
54
Toshi Kani2af3a812016-05-10 10:23:52 -060055void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
56{
57 struct va_format vaf;
58 va_list args;
59
60 va_start(args, fmt);
61 vaf.fmt = fmt;
62 vaf.va = &args;
63 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
64 va_end(args);
65}
66
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070067static void bdev_write_inode(struct block_device *bdev)
Christoph Hellwig564f00f2015-01-14 10:42:33 +010068{
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070069 struct inode *inode = bdev->bd_inode;
70 int ret;
71
Christoph Hellwig564f00f2015-01-14 10:42:33 +010072 spin_lock(&inode->i_lock);
73 while (inode->i_state & I_DIRTY) {
74 spin_unlock(&inode->i_lock);
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070075 ret = write_inode_now(inode, true);
76 if (ret) {
77 char name[BDEVNAME_SIZE];
78 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
79 "for block device %s (err=%d).\n",
80 bdevname(bdev, name), ret);
81 }
Christoph Hellwig564f00f2015-01-14 10:42:33 +010082 spin_lock(&inode->i_lock);
83 }
84 spin_unlock(&inode->i_lock);
85}
86
Peter Zijlstraf9a14392007-05-06 14:49:55 -070087/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb482011-09-16 02:31:11 -040088void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089{
Al Viroff01bb482011-09-16 02:31:11 -040090 struct address_space *mapping = bdev->bd_inode->i_mapping;
91
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080092 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070093 return;
Al Viroff01bb482011-09-16 02:31:11 -040094
Peter Zijlstraf9a14392007-05-06 14:49:55 -070095 invalidate_bh_lrus();
Al Viroff01bb482011-09-16 02:31:11 -040096 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097}
Al Viroff01bb482011-09-16 02:31:11 -040098EXPORT_SYMBOL(kill_bdev);
99
100/* Invalidate clean unused buffers and pagecache. */
101void invalidate_bdev(struct block_device *bdev)
102{
103 struct address_space *mapping = bdev->bd_inode->i_mapping;
104
Andrey Ryabinin945d0ec2017-05-03 14:56:02 -0700105 if (mapping->nrpages) {
106 invalidate_bh_lrus();
107 lru_add_drain_all(); /* make sure all lru add caches are flushed */
108 invalidate_mapping_pages(mapping, 0, -1);
109 }
Al Viroff01bb482011-09-16 02:31:11 -0400110 /* 99% of the time, we don't need to flush the cleancache on the bdev.
111 * But, for the strange corners, lets be cautious
112 */
Dan Magenheimer31677602011-09-21 11:56:28 -0400113 cleancache_invalidate_inode(mapping);
Al Viroff01bb482011-09-16 02:31:11 -0400114}
115EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
117int set_blocksize(struct block_device *bdev, int size)
118{
119 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700120 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 return -EINVAL;
122
123 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400124 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 return -EINVAL;
126
127 /* Don't change the size if it is same as current */
128 if (bdev->bd_block_size != size) {
129 sync_blockdev(bdev);
130 bdev->bd_block_size = size;
131 bdev->bd_inode->i_blkbits = blksize_bits(size);
132 kill_bdev(bdev);
133 }
134 return 0;
135}
136
137EXPORT_SYMBOL(set_blocksize);
138
139int sb_set_blocksize(struct super_block *sb, int size)
140{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 if (set_blocksize(sb->s_bdev, size))
142 return 0;
143 /* If we get here, we know size is power of two
144 * and it's value is between 512 and PAGE_SIZE */
145 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800146 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 return sb->s_blocksize;
148}
149
150EXPORT_SYMBOL(sb_set_blocksize);
151
152int sb_min_blocksize(struct super_block *sb, int size)
153{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400154 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (size < minsize)
156 size = minsize;
157 return sb_set_blocksize(sb, size);
158}
159
160EXPORT_SYMBOL(sb_min_blocksize);
161
162static int
163blkdev_get_block(struct inode *inode, sector_t iblock,
164 struct buffer_head *bh, int create)
165{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 bh->b_bdev = I_BDEV(inode);
167 bh->b_blocknr = iblock;
168 set_buffer_mapped(bh);
169 return 0;
170}
171
Dan Williams4ebb16c2015-10-28 07:48:19 +0900172static struct inode *bdev_file_inode(struct file *file)
173{
174 return file->f_mapping->host;
175}
176
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800177static ssize_t
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700178blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800179{
180 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +0900181 struct inode *inode = bdev_file_inode(file);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800182
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700183 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter,
Jens Axboefe0f07d2015-04-15 17:05:48 -0600184 blkdev_get_block, NULL, NULL,
185 DIO_SKIP_DIO_COUNT);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800186}
187
Jan Kara5cee5812009-04-27 16:43:51 +0200188int __sync_blockdev(struct block_device *bdev, int wait)
189{
190 if (!bdev)
191 return 0;
192 if (!wait)
193 return filemap_flush(bdev->bd_inode->i_mapping);
194 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
195}
196
Nick Piggin585d3bc2009-02-25 10:44:19 +0100197/*
198 * Write out and wait upon all the dirty data associated with a block
199 * device via its mapping. Does not take the superblock lock.
200 */
201int sync_blockdev(struct block_device *bdev)
202{
Jan Kara5cee5812009-04-27 16:43:51 +0200203 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100204}
205EXPORT_SYMBOL(sync_blockdev);
206
207/*
208 * Write out and wait upon all dirty data associated with this
209 * device. Filesystem data as well as the underlying block
210 * device. Takes the superblock lock.
211 */
212int fsync_bdev(struct block_device *bdev)
213{
214 struct super_block *sb = get_super(bdev);
215 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200216 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100217 drop_super(sb);
218 return res;
219 }
220 return sync_blockdev(bdev);
221}
Al Viro47e44912009-04-01 07:07:16 -0400222EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100223
224/**
225 * freeze_bdev -- lock a filesystem and force it into a consistent state
226 * @bdev: blockdevice to lock
227 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100228 * If a superblock is found on this device, we take the s_umount semaphore
229 * on it to make sure nobody unmounts until the snapshot creation is done.
230 * The reference counter (bd_fsfreeze_count) guarantees that only the last
231 * unfreeze process can unfreeze the frozen filesystem actually when multiple
232 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
233 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
234 * actually.
235 */
236struct super_block *freeze_bdev(struct block_device *bdev)
237{
238 struct super_block *sb;
239 int error = 0;
240
241 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200242 if (++bdev->bd_fsfreeze_count > 1) {
243 /*
244 * We don't even need to grab a reference - the first call
245 * to freeze_bdev grab an active reference and only the last
246 * thaw_bdev drops it.
247 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100248 sb = get_super(bdev);
Andrey Ryabinin5bb53c02016-08-23 18:55:31 +0300249 if (sb)
250 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100251 mutex_unlock(&bdev->bd_fsfreeze_mutex);
252 return sb;
253 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100254
Christoph Hellwig45042302009-08-03 23:28:35 +0200255 sb = get_active_super(bdev);
256 if (!sb)
257 goto out;
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600258 if (sb->s_op->freeze_super)
259 error = sb->s_op->freeze_super(sb);
260 else
261 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400262 if (error) {
263 deactivate_super(sb);
264 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400266 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100267 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400268 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200269 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100270 sync_blockdev(bdev);
271 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200272 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100273}
274EXPORT_SYMBOL(freeze_bdev);
275
276/**
277 * thaw_bdev -- unlock filesystem
278 * @bdev: blockdevice to unlock
279 * @sb: associated superblock
280 *
281 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
282 */
283int thaw_bdev(struct block_device *bdev, struct super_block *sb)
284{
Christoph Hellwig45042302009-08-03 23:28:35 +0200285 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100286
287 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200288 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400289 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100290
Christoph Hellwig45042302009-08-03 23:28:35 +0200291 error = 0;
292 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400293 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100294
Christoph Hellwig45042302009-08-03 23:28:35 +0200295 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400296 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200297
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600298 if (sb->s_op->thaw_super)
299 error = sb->s_op->thaw_super(sb);
300 else
301 error = thaw_super(sb);
Pierre Morel997198b2016-10-04 10:53:40 +0200302 if (error)
Josef Bacik18e9e512010-03-23 10:34:56 -0400303 bdev->bd_fsfreeze_count++;
Josef Bacik18e9e512010-03-23 10:34:56 -0400304out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100305 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Pierre Morel997198b2016-10-04 10:53:40 +0200306 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100307}
308EXPORT_SYMBOL(thaw_bdev);
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
311{
312 return block_write_full_page(page, blkdev_get_block, wbc);
313}
314
315static int blkdev_readpage(struct file * file, struct page * page)
316{
317 return block_read_full_page(page, blkdev_get_block);
318}
319
Akinobu Mita447f05b2014-10-09 15:26:58 -0700320static int blkdev_readpages(struct file *file, struct address_space *mapping,
321 struct list_head *pages, unsigned nr_pages)
322{
323 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
324}
325
Nick Piggin6272b5a2007-10-16 01:25:04 -0700326static int blkdev_write_begin(struct file *file, struct address_space *mapping,
327 loff_t pos, unsigned len, unsigned flags,
328 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200330 return block_write_begin(mapping, pos, len, flags, pagep,
331 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}
333
Nick Piggin6272b5a2007-10-16 01:25:04 -0700334static int blkdev_write_end(struct file *file, struct address_space *mapping,
335 loff_t pos, unsigned len, unsigned copied,
336 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700338 int ret;
339 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
340
341 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300342 put_page(page);
Nick Piggin6272b5a2007-10-16 01:25:04 -0700343
344 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
347/*
348 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500349 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 * so we compute the size by hand (just as in block_read/write above)
351 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800352static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900354 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 loff_t retval;
356
Al Viro59551022016-01-22 15:40:57 -0500357 inode_lock(bd_inode);
Al Viro5d48f3a2013-06-23 21:34:45 +0400358 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Al Viro59551022016-01-22 15:40:57 -0500359 inode_unlock(bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 return retval;
361}
362
Josef Bacik02c24a82011-07-16 20:44:56 -0400363int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900365 struct inode *bd_inode = bdev_file_inode(filp);
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400366 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100367 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200368
369 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
370 if (error)
371 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100372
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400373 /*
374 * There is no need to serialise calls to blkdev_issue_flush with
375 * i_mutex and doing so causes performance issues with concurrent
376 * O_SYNC writers to a block device.
377 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200378 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100379 if (error == -EOPNOTSUPP)
380 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400381
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100382 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700384EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700386/**
387 * bdev_read_page() - Start reading a page from a block device
388 * @bdev: The device to read the page from
389 * @sector: The offset on the device to read the page to (need not be aligned)
390 * @page: The page to read
391 *
392 * On entry, the page should be locked. It will be unlocked when the page
393 * has been read. If the block driver implements rw_page synchronously,
394 * that will be true on exit from this function, but it need not be.
395 *
396 * Errors returned by this function are usually "soft", eg out of memory, or
397 * queue full; callers should try a different route to read this page rather
398 * than propagate an error back up the stack.
399 *
400 * Return: negative errno if an error occurs, 0 if submission was successful.
401 */
402int bdev_read_page(struct block_device *bdev, sector_t sector,
403 struct page *page)
404{
405 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800406 int result = -EOPNOTSUPP;
407
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400408 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc92015-11-19 13:29:28 -0800409 return result;
410
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100411 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800412 if (result)
413 return result;
Jens Axboec11f0c02016-08-05 08:11:04 -0600414 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800415 blk_queue_exit(bdev->bd_queue);
416 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700417}
418EXPORT_SYMBOL_GPL(bdev_read_page);
419
420/**
421 * bdev_write_page() - Start writing a page to a block device
422 * @bdev: The device to write the page to
423 * @sector: The offset on the device to write the page to (need not be aligned)
424 * @page: The page to write
425 * @wbc: The writeback_control for the write
426 *
427 * On entry, the page should be locked and not currently under writeback.
428 * On exit, if the write started successfully, the page will be unlocked and
429 * under writeback. If the write failed already (eg the driver failed to
430 * queue the page to the device), the page will still be locked. If the
431 * caller is a ->writepage implementation, it will need to unlock the page.
432 *
433 * Errors returned by this function are usually "soft", eg out of memory, or
434 * queue full; callers should try a different route to write this page rather
435 * than propagate an error back up the stack.
436 *
437 * Return: negative errno if an error occurs, 0 if submission was successful.
438 */
439int bdev_write_page(struct block_device *bdev, sector_t sector,
440 struct page *page, struct writeback_control *wbc)
441{
442 int result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700443 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800444
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400445 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700446 return -EOPNOTSUPP;
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100447 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800448 if (result)
449 return result;
450
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700451 set_page_writeback(page);
Jens Axboec11f0c02016-08-05 08:11:04 -0600452 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
Matthew Wilcox133ca5c2017-10-13 15:58:15 -0700453 if (result) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700454 end_page_writeback(page);
Matthew Wilcox133ca5c2017-10-13 15:58:15 -0700455 } else {
456 clean_page_buffers(page);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700457 unlock_page(page);
Matthew Wilcox133ca5c2017-10-13 15:58:15 -0700458 }
Dan Williams2e6edc92015-11-19 13:29:28 -0800459 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700460 return result;
461}
462EXPORT_SYMBOL_GPL(bdev_write_page);
463
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200464/**
465 * bdev_direct_access() - Get the address for directly-accessibly memory
466 * @bdev: The device containing the memory
Dan Williamsb2e0d162016-01-15 16:55:59 -0800467 * @dax: control and output parameters for ->direct_access
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200468 *
469 * If a block device is made up of directly addressable memory, this function
470 * will tell the caller the PFN and the address of the memory. The address
471 * may be directly dereferenced within the kernel without the need to call
472 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
473 * page tables.
474 *
475 * Return: negative errno if an error occurs, otherwise the number of bytes
476 * accessible at this address.
477 */
Dan Williamsb2e0d162016-01-15 16:55:59 -0800478long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200479{
Dan Williamsb2e0d162016-01-15 16:55:59 -0800480 sector_t sector = dax->sector;
481 long avail, size = dax->size;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200482 const struct block_device_operations *ops = bdev->bd_disk->fops;
483
Matthew Wilcox43c3dd02015-07-03 10:40:43 -0400484 /*
485 * The device driver is allowed to sleep, in order to make the
486 * memory directly accessible.
487 */
488 might_sleep();
489
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200490 if (size < 0)
491 return size;
Toshi Kani163d4ba2016-06-23 17:05:50 -0400492 if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200493 return -EOPNOTSUPP;
494 if ((sector + DIV_ROUND_UP(size, 512)) >
495 part_nr_sects_read(bdev->bd_part))
496 return -ERANGE;
497 sector += get_start_sect(bdev);
498 if (sector % (PAGE_SIZE / 512))
499 return -EINVAL;
Dan Williams0a70bd42016-02-24 14:02:11 -0800500 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size);
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200501 if (!avail)
502 return -ERANGE;
Dan Williamsfe683ad2016-01-15 16:55:56 -0800503 if (avail > 0 && avail & ~PAGE_MASK)
504 return -ENXIO;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200505 return min(avail, size);
506}
507EXPORT_SYMBOL_GPL(bdev_direct_access);
508
Toshi Kani2d96afc2016-05-10 10:23:53 -0600509/**
510 * bdev_dax_supported() - Check if the device supports dax for filesystem
511 * @sb: The superblock of the device
512 * @blocksize: The block size of the device
513 *
514 * This is a library function for filesystems to check if the block device
515 * can be mounted with dax option.
516 *
517 * Return: negative errno if unsupported, 0 if supported.
518 */
519int bdev_dax_supported(struct super_block *sb, int blocksize)
520{
521 struct blk_dax_ctl dax = {
522 .sector = 0,
523 .size = PAGE_SIZE,
524 };
525 int err;
526
527 if (blocksize != PAGE_SIZE) {
528 vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
529 return -EINVAL;
530 }
531
532 err = bdev_direct_access(sb->s_bdev, &dax);
533 if (err < 0) {
534 switch (err) {
535 case -EOPNOTSUPP:
536 vfs_msg(sb, KERN_ERR,
537 "error: device does not support dax");
538 break;
539 case -EINVAL:
540 vfs_msg(sb, KERN_ERR,
541 "error: unaligned partition for dax");
542 break;
543 default:
544 vfs_msg(sb, KERN_ERR,
545 "error: dax access failed (%d)", err);
546 }
547 return err;
548 }
549
550 return 0;
551}
552EXPORT_SYMBOL_GPL(bdev_dax_supported);
553
Toshi Kania8078b12016-05-10 10:23:57 -0600554/**
555 * bdev_dax_capable() - Return if the raw device is capable for dax
556 * @bdev: The device for raw block device access
557 */
558bool bdev_dax_capable(struct block_device *bdev)
559{
Toshi Kania8078b12016-05-10 10:23:57 -0600560 struct blk_dax_ctl dax = {
561 .size = PAGE_SIZE,
562 };
563
564 if (!IS_ENABLED(CONFIG_FS_DAX))
565 return false;
566
567 dax.sector = 0;
568 if (bdev_direct_access(bdev, &dax) < 0)
569 return false;
570
571 dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512);
572 if (bdev_direct_access(bdev, &dax) < 0)
573 return false;
574
Toshi Kania8078b12016-05-10 10:23:57 -0600575 return true;
576}
577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578/*
579 * pseudo-fs
580 */
581
582static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800583static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585static struct inode *bdev_alloc_inode(struct super_block *sb)
586{
Christoph Lametere94b1762006-12-06 20:33:17 -0800587 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 if (!ei)
589 return NULL;
590 return &ei->vfs_inode;
591}
592
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100593static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100595 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 struct bdev_inode *bdi = BDEV_I(inode);
597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 kmem_cache_free(bdev_cachep, bdi);
599}
600
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100601static void bdev_destroy_inode(struct inode *inode)
602{
603 call_rcu(&inode->i_rcu, bdev_i_callback);
604}
605
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700606static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
608 struct bdev_inode *ei = (struct bdev_inode *) foo;
609 struct block_device *bdev = &ei->bdev;
610
Christoph Lametera35afb82007-05-16 22:10:57 -0700611 memset(bdev, 0, sizeof(*bdev));
612 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700613 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100614#ifdef CONFIG_SYSFS
615 INIT_LIST_HEAD(&bdev->bd_holder_disks);
616#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700617 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800618 /* Initialize mutex for freeze. */
619 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
Al Virob57922d2010-06-07 14:34:48 -0400622static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
624 struct block_device *bdev = &BDEV_I(inode)->bdev;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700625 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400626 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200627 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 spin_lock(&bdev_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 list_del_init(&bdev->bd_list);
630 spin_unlock(&bdev_lock);
631}
632
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800633static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 .statfs = simple_statfs,
635 .alloc_inode = bdev_alloc_inode,
636 .destroy_inode = bdev_destroy_inode,
637 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400638 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639};
640
Al Viro51139ad2010-07-25 23:47:46 +0400641static struct dentry *bd_mount(struct file_system_type *fs_type,
642 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
Shaohua Li3684aa72016-02-22 15:27:40 -0700644 struct dentry *dent;
645 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Vegard Nossume9e5e3f2016-08-22 12:47:43 +0200646 if (!IS_ERR(dent))
Shaohua Li3684aa72016-02-22 15:27:40 -0700647 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
648 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
651static struct file_system_type bd_type = {
652 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400653 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 .kill_sb = kill_anon_super,
655};
656
Tejun Heoa212b102015-05-22 17:13:33 -0400657struct super_block *blockdev_superblock __read_mostly;
658EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660void __init bdev_cache_init(void)
661{
662 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300663 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800666 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800667 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900668 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 err = register_filesystem(&bd_type);
670 if (err)
671 panic("Cannot register bdev pseudo-fs");
672 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (IS_ERR(bd_mnt))
674 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300675 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
678/*
679 * Most likely _very_ bad one - but then it's hardly critical for small
680 * /dev and can be fixed when somebody will need really large one.
681 * Keep in mind that it will be fed through icache hash function too.
682 */
683static inline unsigned long hash(dev_t dev)
684{
685 return MAJOR(dev)+MINOR(dev);
686}
687
688static int bdev_test(struct inode *inode, void *data)
689{
690 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
691}
692
693static int bdev_set(struct inode *inode, void *data)
694{
695 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
696 return 0;
697}
698
699static LIST_HEAD(all_bdevs);
700
701struct block_device *bdget(dev_t dev)
702{
703 struct block_device *bdev;
704 struct inode *inode;
705
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800706 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 bdev_test, bdev_set, &dev);
708
709 if (!inode)
710 return NULL;
711
712 bdev = &BDEV_I(inode)->bdev;
713
714 if (inode->i_state & I_NEW) {
715 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000716 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 bdev->bd_inode = inode;
Fabian Frederick61604a22017-02-27 14:28:32 -0800718 bdev->bd_block_size = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 bdev->bd_part_count = 0;
720 bdev->bd_invalidated = 0;
721 inode->i_mode = S_IFBLK;
722 inode->i_rdev = dev;
723 inode->i_bdev = bdev;
724 inode->i_data.a_ops = &def_blk_aops;
725 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 spin_lock(&bdev_lock);
727 list_add(&bdev->bd_list, &all_bdevs);
728 spin_unlock(&bdev_lock);
729 unlock_new_inode(inode);
730 }
731 return bdev;
732}
733
734EXPORT_SYMBOL(bdget);
735
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200736/**
737 * bdgrab -- Grab a reference to an already referenced block device
738 * @bdev: Block device to grab a reference to.
739 */
740struct block_device *bdgrab(struct block_device *bdev)
741{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400742 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200743 return bdev;
744}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700745EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747long nr_blockdev_pages(void)
748{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700749 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 long ret = 0;
751 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700752 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 ret += bdev->bd_inode->i_mapping->nrpages;
754 }
755 spin_unlock(&bdev_lock);
756 return ret;
757}
758
759void bdput(struct block_device *bdev)
760{
761 iput(bdev->bd_inode);
762}
763
764EXPORT_SYMBOL(bdput);
765
766static struct block_device *bd_acquire(struct inode *inode)
767{
768 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700769
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 spin_lock(&bdev_lock);
771 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700772 if (bdev) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100773 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 spin_unlock(&bdev_lock);
775 return bdev;
776 }
777 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 bdev = bdget(inode->i_rdev);
780 if (bdev) {
781 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700782 if (!inode->i_bdev) {
783 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400784 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700785 * and it's released in clear_inode() of inode.
786 * So, we can access it via ->i_mapping always
787 * without igrab().
788 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100789 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700790 inode->i_bdev = bdev;
791 inode->i_mapping = bdev->bd_inode->i_mapping;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 spin_unlock(&bdev_lock);
794 }
795 return bdev;
796}
797
798/* Call when you free inode */
799
800void bd_forget(struct inode *inode)
801{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700802 struct block_device *bdev = NULL;
803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700805 if (!sb_is_blkdev_sb(inode->i_sb))
806 bdev = inode->i_bdev;
Al Viroa4a4f942016-07-19 13:16:52 -0400807 inode->i_bdev = NULL;
808 inode->i_mapping = &inode->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700810
811 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100812 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813}
814
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900815/**
816 * bd_may_claim - test whether a block device can be claimed
817 * @bdev: block device of interest
818 * @whole: whole block device containing @bdev, may equal @bdev
819 * @holder: holder trying to claim @bdev
820 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300821 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900822 *
823 * CONTEXT:
824 * spin_lock(&bdev_lock).
825 *
826 * RETURNS:
827 * %true if @bdev can be claimed, %false otherwise.
828 */
829static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
830 void *holder)
831{
832 if (bdev->bd_holder == holder)
833 return true; /* already a holder */
834 else if (bdev->bd_holder != NULL)
835 return false; /* held by someone else */
NeilBrowncfa2d652016-12-12 08:21:51 -0700836 else if (whole == bdev)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900837 return true; /* is a whole device which isn't held */
838
Tejun Heoe525fd82010-11-13 11:55:17 +0100839 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900840 return true; /* is a partition of a device that is being partitioned */
841 else if (whole->bd_holder != NULL)
842 return false; /* is a partition of a held device */
843 else
844 return true; /* is a partition of an un-held device */
845}
846
847/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900848 * bd_prepare_to_claim - prepare to claim a block device
849 * @bdev: block device of interest
850 * @whole: the whole device containing @bdev, may equal @bdev
851 * @holder: holder trying to claim @bdev
852 *
853 * Prepare to claim @bdev. This function fails if @bdev is already
854 * claimed by another holder and waits if another claiming is in
855 * progress. This function doesn't actually claim. On successful
856 * return, the caller has ownership of bd_claiming and bd_holder[s].
857 *
858 * CONTEXT:
859 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
860 * it multiple times.
861 *
862 * RETURNS:
863 * 0 if @bdev can be claimed, -EBUSY otherwise.
864 */
865static int bd_prepare_to_claim(struct block_device *bdev,
866 struct block_device *whole, void *holder)
867{
868retry:
869 /* if someone else claimed, fail */
870 if (!bd_may_claim(bdev, whole, holder))
871 return -EBUSY;
872
Tejun Heoe75aa852010-08-04 17:59:39 +0200873 /* if claiming is already in progress, wait for it to finish */
874 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900875 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
876 DEFINE_WAIT(wait);
877
878 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
879 spin_unlock(&bdev_lock);
880 schedule();
881 finish_wait(wq, &wait);
882 spin_lock(&bdev_lock);
883 goto retry;
884 }
885
886 /* yay, all mine */
887 return 0;
888}
889
890/**
891 * bd_start_claiming - start claiming a block device
892 * @bdev: block device of interest
893 * @holder: holder trying to claim @bdev
894 *
895 * @bdev is about to be opened exclusively. Check @bdev can be opened
896 * exclusively and mark that an exclusive open is in progress. Each
897 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000898 * either bd_finish_claiming() or bd_abort_claiming() (which do not
899 * fail).
900 *
901 * This function is used to gain exclusive access to the block device
902 * without actually causing other exclusive open attempts to fail. It
903 * should be used when the open sequence itself requires exclusive
904 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900905 *
906 * CONTEXT:
907 * Might sleep.
908 *
909 * RETURNS:
910 * Pointer to the block device containing @bdev on success, ERR_PTR()
911 * value on failure.
912 */
913static struct block_device *bd_start_claiming(struct block_device *bdev,
914 void *holder)
915{
916 struct gendisk *disk;
917 struct block_device *whole;
918 int partno, err;
919
920 might_sleep();
921
922 /*
923 * @bdev might not have been initialized properly yet, look up
924 * and grab the outer block device the hard way.
925 */
926 disk = get_gendisk(bdev->bd_dev, &partno);
927 if (!disk)
928 return ERR_PTR(-ENXIO);
929
Tejun Heod4c208b2011-06-13 12:45:48 +0200930 /*
931 * Normally, @bdev should equal what's returned from bdget_disk()
932 * if partno is 0; however, some drivers (floppy) use multiple
933 * bdev's for the same physical device and @bdev may be one of the
934 * aliases. Keep @bdev if partno is 0. This means claimer
935 * tracking is broken for those devices but it has always been that
936 * way.
937 */
938 if (partno)
939 whole = bdget_disk(disk, 0);
940 else
941 whole = bdgrab(bdev);
942
Nick Piggincf342572010-05-26 01:50:21 +1000943 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900944 put_disk(disk);
945 if (!whole)
946 return ERR_PTR(-ENOMEM);
947
948 /* prepare to claim, if successful, mark claiming in progress */
949 spin_lock(&bdev_lock);
950
951 err = bd_prepare_to_claim(bdev, whole, holder);
952 if (err == 0) {
953 whole->bd_claiming = holder;
954 spin_unlock(&bdev_lock);
955 return whole;
956 } else {
957 spin_unlock(&bdev_lock);
958 bdput(whole);
959 return ERR_PTR(err);
960 }
961}
962
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800963#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100964struct bd_holder_disk {
965 struct list_head list;
966 struct gendisk *disk;
967 int refcnt;
968};
969
970static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
971 struct gendisk *disk)
972{
973 struct bd_holder_disk *holder;
974
975 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
976 if (holder->disk == disk)
977 return holder;
978 return NULL;
979}
980
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700981static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800982{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700983 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800984}
985
986static void del_symlink(struct kobject *from, struct kobject *to)
987{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800988 sysfs_remove_link(from, kobject_name(to));
989}
990
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800991/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100992 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
993 * @bdev: the claimed slave bdev
994 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500995 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100996 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
997 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100998 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500999 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001000 * - from "slaves" directory of the holder @disk to the claimed @bdev
1001 * - from "holders" directory of the @bdev to the holder @disk
1002 *
1003 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1004 * passed to bd_link_disk_holder(), then:
1005 *
1006 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1007 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1008 *
1009 * The caller must have claimed @bdev before calling this function and
1010 * ensure that both @bdev and @disk are valid during the creation and
1011 * lifetime of these symlinks.
1012 *
1013 * CONTEXT:
1014 * Might sleep.
1015 *
1016 * RETURNS:
1017 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001018 */
Tejun Heoe09b4572010-11-13 11:55:17 +01001019int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001020{
Tejun Heo49731ba2011-01-14 18:43:57 +01001021 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001022 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001023
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001024 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001025
Tejun Heo49731ba2011-01-14 18:43:57 +01001026 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -07001027
Tejun Heoe09b4572010-11-13 11:55:17 +01001028 /* FIXME: remove the following once add_disk() handles errors */
1029 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1030 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -07001031
Tejun Heo49731ba2011-01-14 18:43:57 +01001032 holder = bd_find_holder_disk(bdev, disk);
1033 if (holder) {
1034 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +01001035 goto out_unlock;
1036 }
1037
Tejun Heo49731ba2011-01-14 18:43:57 +01001038 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1039 if (!holder) {
1040 ret = -ENOMEM;
1041 goto out_unlock;
1042 }
1043
1044 INIT_LIST_HEAD(&holder->list);
1045 holder->disk = disk;
1046 holder->refcnt = 1;
1047
1048 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1049 if (ret)
1050 goto out_free;
1051
1052 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1053 if (ret)
1054 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +01001055 /*
1056 * bdev could be deleted beneath us which would implicitly destroy
1057 * the holder directory. Hold on to it.
1058 */
1059 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001060
1061 list_add(&holder->list, &bdev->bd_holder_disks);
1062 goto out_unlock;
1063
1064out_del:
1065 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1066out_free:
1067 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001068out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001069 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001070 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001071}
Tejun Heoe09b4572010-11-13 11:55:17 +01001072EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001073
Tejun Heo49731ba2011-01-14 18:43:57 +01001074/**
1075 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1076 * @bdev: the calimed slave bdev
1077 * @disk: the holding disk
1078 *
1079 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1080 *
1081 * CONTEXT:
1082 * Might sleep.
1083 */
1084void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001085{
Tejun Heo49731ba2011-01-14 18:43:57 +01001086 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001087
Tejun Heo49731ba2011-01-14 18:43:57 +01001088 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001089
Tejun Heo49731ba2011-01-14 18:43:57 +01001090 holder = bd_find_holder_disk(bdev, disk);
1091
1092 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1093 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1094 del_symlink(bdev->bd_part->holder_dir,
1095 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001096 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001097 list_del_init(&holder->list);
1098 kfree(holder);
1099 }
1100
1101 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001102}
Tejun Heo49731ba2011-01-14 18:43:57 +01001103EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001104#endif
1105
Andrew Patterson0c002c22008-09-04 14:27:20 -06001106/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001107 * flush_disk - invalidates all buffer-cache entries on a disk
1108 *
1109 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001110 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001111 *
1112 * Invalidates all buffer-cache entries on a disk. It should be called
1113 * when a disk has been changed -- either by a media change or online
1114 * resize.
1115 */
NeilBrown93b270f2011-02-24 17:25:47 +11001116static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001117{
NeilBrown93b270f2011-02-24 17:25:47 +11001118 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001119 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001120 "resized disk %s\n",
1121 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001122 }
1123
1124 if (!bdev->bd_disk)
1125 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001126 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001127 bdev->bd_invalidated = 1;
1128}
1129
1130/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001131 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001132 * @disk: struct gendisk to check
1133 * @bdev: struct bdev to adjust.
1134 *
1135 * This routine checks to see if the bdev size does not match the disk size
1136 * and adjusts it if it differs.
1137 */
1138void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1139{
1140 loff_t disk_size, bdev_size;
1141
1142 disk_size = (loff_t)get_capacity(disk) << 9;
1143 bdev_size = i_size_read(bdev->bd_inode);
1144 if (disk_size != bdev_size) {
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001145 printk(KERN_INFO
1146 "%s: detected capacity change from %lld to %lld\n",
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001147 disk->disk_name, bdev_size, disk_size);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001148 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001149 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001150 }
1151}
1152EXPORT_SYMBOL(check_disk_size_change);
1153
1154/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001155 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001156 * @disk: struct gendisk to be revalidated
1157 *
1158 * This routine is a wrapper for lower-level driver's revalidate_disk
1159 * call-backs. It is used to do common pre and post operations needed
1160 * for all revalidate_disk operations.
1161 */
1162int revalidate_disk(struct gendisk *disk)
1163{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001164 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001165 int ret = 0;
1166
1167 if (disk->fops->revalidate_disk)
1168 ret = disk->fops->revalidate_disk(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001169 bdev = bdget_disk(disk, 0);
1170 if (!bdev)
1171 return ret;
1172
1173 mutex_lock(&bdev->bd_mutex);
1174 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001175 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001176 mutex_unlock(&bdev->bd_mutex);
1177 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001178 return ret;
1179}
1180EXPORT_SYMBOL(revalidate_disk);
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182/*
1183 * This routine checks whether a removable media has been changed,
1184 * and invalidates all buffer-cache-entries in that case. This
1185 * is a relatively slow routine, so we have to try to minimize using
1186 * it. Thus it is called only upon a 'mount' or 'open'. This
1187 * is the best way of combining speed and utility, I think.
1188 * People changing diskettes in the middle of an operation deserve
1189 * to lose :-)
1190 */
1191int check_disk_change(struct block_device *bdev)
1192{
1193 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001194 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001195 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Tejun Heo77ea8872010-12-08 20:57:37 +01001197 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1198 DISK_EVENT_EJECT_REQUEST);
1199 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 return 0;
1201
NeilBrown93b270f2011-02-24 17:25:47 +11001202 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 if (bdops->revalidate_disk)
1204 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 return 1;
1206}
1207
1208EXPORT_SYMBOL(check_disk_change);
1209
1210void bd_set_size(struct block_device *bdev, loff_t size)
1211{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001212 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Al Viro59551022016-01-22 15:40:57 -05001214 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001215 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001216 inode_unlock(bdev->bd_inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001217 while (bsize < PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 if (size & bsize)
1219 break;
1220 bsize <<= 1;
1221 }
1222 bdev->bd_block_size = bsize;
1223 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1224}
1225EXPORT_SYMBOL(bd_set_size);
1226
Al Viro4385bab2013-05-05 22:11:03 -04001227static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001228
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001229/*
1230 * bd_mutex locking:
1231 *
1232 * mutex_lock(part->bd_mutex)
1233 * mutex_lock_nested(whole->bd_mutex, 1)
1234 */
1235
Al Viro572c4892007-10-08 13:24:05 -04001236static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001239 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001240 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001241 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001242 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
Al Viro572c4892007-10-08 13:24:05 -04001244 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001245 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001246 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001247 perm |= MAY_WRITE;
1248 /*
1249 * hooks: /n/, see "layering violations".
1250 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001251 if (!for_part) {
1252 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1253 if (ret != 0) {
1254 bdput(bdev);
1255 return ret;
1256 }
Al Viro82666022008-08-01 05:32:04 -04001257 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001258
NeilBrownd3374822009-01-09 08:31:10 +11001259 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001260
Tejun Heo89f97492008-11-05 10:21:06 +01001261 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001262 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001263 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001264 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001265 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Tejun Heo69e02c52011-03-09 19:54:27 +01001267 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001268 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 if (!bdev->bd_openers) {
1270 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001271 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 bdev->bd_contains = bdev;
Dan Williams03cdadb2016-02-26 15:19:43 -08001273
Tejun Heocf771cb2008-09-03 09:01:09 +02001274 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001275 ret = -ENXIO;
1276 bdev->bd_part = disk_get_part(disk, partno);
1277 if (!bdev->bd_part)
1278 goto out_clear;
1279
Tejun Heo1196f8b2011-04-21 20:54:45 +02001280 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001282 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001283 if (ret == -ERESTARTSYS) {
1284 /* Lost a race with 'disk' being
1285 * deleted, try again.
1286 * See md.c
1287 */
1288 disk_put_part(bdev->bd_part);
1289 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001290 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001291 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001292 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001293 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001294 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001295 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001296 goto restart;
1297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 }
Tejun Heo7e697232011-05-23 13:26:07 +02001299
Christoph Hellwig22375702016-09-14 11:56:13 +02001300 if (!ret)
Tejun Heo7e697232011-05-23 13:26:07 +02001301 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Tejun Heo7e697232011-05-23 13:26:07 +02001302
Tejun Heo1196f8b2011-04-21 20:54:45 +02001303 /*
1304 * If the device is invalidated, rescan partition
1305 * if open succeeded or failed with -ENOMEDIUM.
1306 * The latter is necessary to prevent ghost
1307 * partitions on a removed medium.
1308 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001309 if (bdev->bd_invalidated) {
1310 if (!ret)
1311 rescan_partitions(disk, bdev);
1312 else if (ret == -ENOMEDIUM)
1313 invalidate_partitions(disk, bdev);
1314 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001315
Tejun Heo1196f8b2011-04-21 20:54:45 +02001316 if (ret)
1317 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 struct block_device *whole;
1320 whole = bdget_disk(disk, 0);
1321 ret = -ENOMEM;
1322 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001323 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001324 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001325 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001327 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001329 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001330 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001331 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001333 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 }
Tejun Heo89f97492008-11-05 10:21:06 +01001335 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 }
1337 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001339 ret = 0;
1340 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001341 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001342 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001343 if (bdev->bd_invalidated) {
1344 if (!ret)
1345 rescan_partitions(bdev->bd_disk, bdev);
1346 else if (ret == -ENOMEDIUM)
1347 invalidate_partitions(bdev->bd_disk, bdev);
1348 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001349 if (ret)
1350 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001352 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001353 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001354 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 }
1356 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001357 if (for_part)
1358 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001359 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001360 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 return 0;
1362
Tejun Heo0762b8b2008-08-25 19:56:12 +09001363 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001364 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001366 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001367 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001369 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001371 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001372 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001373 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001374 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001375 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001376 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001377 bdput(bdev);
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 return ret;
1380}
1381
Tejun Heod4d77622010-11-13 11:55:18 +01001382/**
1383 * blkdev_get - open a block device
1384 * @bdev: block_device to open
1385 * @mode: FMODE_* mask
1386 * @holder: exclusive holder identifier
1387 *
1388 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1389 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1390 * @holder is invalid. Exclusive opens may nest for the same @holder.
1391 *
1392 * On success, the reference count of @bdev is unchanged. On failure,
1393 * @bdev is put.
1394 *
1395 * CONTEXT:
1396 * Might sleep.
1397 *
1398 * RETURNS:
1399 * 0 on success, -errno on failure.
1400 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001401int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402{
Tejun Heoe525fd82010-11-13 11:55:17 +01001403 struct block_device *whole = NULL;
1404 int res;
1405
1406 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1407
1408 if ((mode & FMODE_EXCL) && holder) {
1409 whole = bd_start_claiming(bdev, holder);
1410 if (IS_ERR(whole)) {
1411 bdput(bdev);
1412 return PTR_ERR(whole);
1413 }
1414 }
1415
1416 res = __blkdev_get(bdev, mode, 0);
1417
1418 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001419 struct gendisk *disk = whole->bd_disk;
1420
Tejun Heo6a027ef2010-11-13 11:55:17 +01001421 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001422 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001423 spin_lock(&bdev_lock);
1424
Tejun Heo77ea8872010-12-08 20:57:37 +01001425 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001426 BUG_ON(!bd_may_claim(bdev, whole, holder));
1427 /*
1428 * Note that for a whole device bd_holders
1429 * will be incremented twice, and bd_holder
1430 * will be set to bd_may_claim before being
1431 * set to holder
1432 */
1433 whole->bd_holders++;
1434 whole->bd_holder = bd_may_claim;
1435 bdev->bd_holders++;
1436 bdev->bd_holder = holder;
1437 }
1438
1439 /* tell others that we're done */
1440 BUG_ON(whole->bd_claiming != holder);
1441 whole->bd_claiming = NULL;
1442 wake_up_bit(&whole->bd_claiming, 0);
1443
1444 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001445
1446 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001447 * Block event polling for write claims if requested. Any
1448 * write holder makes the write_holder state stick until
1449 * all are released. This is good enough and tracking
1450 * individual writeable reference is too fragile given the
1451 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001452 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001453 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1454 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001455 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001456 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001457 }
1458
1459 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001460 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001461 }
1462
1463 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001464}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465EXPORT_SYMBOL(blkdev_get);
1466
Tejun Heod4d77622010-11-13 11:55:18 +01001467/**
1468 * blkdev_get_by_path - open a block device by name
1469 * @path: path to the block device to open
1470 * @mode: FMODE_* mask
1471 * @holder: exclusive holder identifier
1472 *
1473 * Open the blockdevice described by the device file at @path. @mode
1474 * and @holder are identical to blkdev_get().
1475 *
1476 * On success, the returned block_device has reference count of one.
1477 *
1478 * CONTEXT:
1479 * Might sleep.
1480 *
1481 * RETURNS:
1482 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1483 */
1484struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1485 void *holder)
1486{
1487 struct block_device *bdev;
1488 int err;
1489
1490 bdev = lookup_bdev(path);
1491 if (IS_ERR(bdev))
1492 return bdev;
1493
1494 err = blkdev_get(bdev, mode, holder);
1495 if (err)
1496 return ERR_PTR(err);
1497
Chuck Ebberte51900f2011-02-16 18:11:53 -05001498 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1499 blkdev_put(bdev, mode);
1500 return ERR_PTR(-EACCES);
1501 }
1502
Tejun Heod4d77622010-11-13 11:55:18 +01001503 return bdev;
1504}
1505EXPORT_SYMBOL(blkdev_get_by_path);
1506
1507/**
1508 * blkdev_get_by_dev - open a block device by device number
1509 * @dev: device number of block device to open
1510 * @mode: FMODE_* mask
1511 * @holder: exclusive holder identifier
1512 *
1513 * Open the blockdevice described by device number @dev. @mode and
1514 * @holder are identical to blkdev_get().
1515 *
1516 * Use it ONLY if you really do not have anything better - i.e. when
1517 * you are behind a truly sucky interface and all you are given is a
1518 * device number. _Never_ to be used for internal purposes. If you
1519 * ever need it - reconsider your API.
1520 *
1521 * On success, the returned block_device has reference count of one.
1522 *
1523 * CONTEXT:
1524 * Might sleep.
1525 *
1526 * RETURNS:
1527 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1528 */
1529struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1530{
1531 struct block_device *bdev;
1532 int err;
1533
1534 bdev = bdget(dev);
1535 if (!bdev)
1536 return ERR_PTR(-ENOMEM);
1537
1538 err = blkdev_get(bdev, mode, holder);
1539 if (err)
1540 return ERR_PTR(err);
1541
1542 return bdev;
1543}
1544EXPORT_SYMBOL(blkdev_get_by_dev);
1545
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546static int blkdev_open(struct inode * inode, struct file * filp)
1547{
1548 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 /*
1551 * Preserve backwards compatibility and allow large file access
1552 * even if userspace doesn't ask for it explicitly. Some mkfs
1553 * binary needs it. We might want to drop this workaround
1554 * during an unstable branch.
1555 */
1556 filp->f_flags |= O_LARGEFILE;
1557
Al Viro572c4892007-10-08 13:24:05 -04001558 if (filp->f_flags & O_NDELAY)
1559 filp->f_mode |= FMODE_NDELAY;
1560 if (filp->f_flags & O_EXCL)
1561 filp->f_mode |= FMODE_EXCL;
1562 if ((filp->f_flags & O_ACCMODE) == 3)
1563 filp->f_mode |= FMODE_WRITE_IOCTL;
1564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001566 if (bdev == NULL)
1567 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Al Viro572c4892007-10-08 13:24:05 -04001569 filp->f_mapping = bdev->bd_inode->i_mapping;
1570
Tejun Heoe525fd82010-11-13 11:55:17 +01001571 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572}
1573
Al Viro4385bab2013-05-05 22:11:03 -04001574static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001575{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001576 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001577 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001578
NeilBrown6796bf52006-12-08 02:36:16 -08001579 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001580 if (for_part)
1581 bdev->bd_part_count--;
1582
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001583 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001584 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001585 sync_blockdev(bdev);
1586 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001587
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001588 bdev_write_inode(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001589 /*
1590 * Detaching bdev inode from its wb in __destroy_inode()
1591 * is too late: the queue which embeds its bdi (along with
1592 * root wb) can be gone as soon as we put_disk() below.
1593 */
1594 inode_detach_wb(bdev->bd_inode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001595 }
1596 if (bdev->bd_contains == bdev) {
1597 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001598 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001599 }
1600 if (!bdev->bd_openers) {
1601 struct module *owner = disk->fops->owner;
1602
Tejun Heo0762b8b2008-08-25 19:56:12 +09001603 disk_put_part(bdev->bd_part);
1604 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001605 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001606 if (bdev != bdev->bd_contains)
1607 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001608 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001609
1610 put_disk(disk);
1611 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001612 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001613 mutex_unlock(&bdev->bd_mutex);
1614 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001615 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001616 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001617}
1618
Al Viro4385bab2013-05-05 22:11:03 -04001619void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001620{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001621 mutex_lock(&bdev->bd_mutex);
1622
Tejun Heoe525fd82010-11-13 11:55:17 +01001623 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001624 bool bdev_free;
1625
1626 /*
1627 * Release a claim on the device. The holder fields
1628 * are protected with bdev_lock. bd_mutex is to
1629 * synchronize disk_holder unlinking.
1630 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001631 spin_lock(&bdev_lock);
1632
1633 WARN_ON_ONCE(--bdev->bd_holders < 0);
1634 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1635
1636 /* bd_contains might point to self, check in a separate step */
1637 if ((bdev_free = !bdev->bd_holders))
1638 bdev->bd_holder = NULL;
1639 if (!bdev->bd_contains->bd_holders)
1640 bdev->bd_contains->bd_holder = NULL;
1641
1642 spin_unlock(&bdev_lock);
1643
Tejun Heo77ea8872010-12-08 20:57:37 +01001644 /*
1645 * If this was the last claim, remove holder link and
1646 * unblock evpoll if it was a write holder.
1647 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001648 if (bdev_free && bdev->bd_write_holder) {
1649 disk_unblock_events(bdev->bd_disk);
1650 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001651 }
Tejun Heo69362172011-03-09 19:54:27 +01001652 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001653
Tejun Heo85ef06d2011-07-01 16:17:47 +02001654 /*
1655 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1656 * event. This is to ensure detection of media removal commanded
1657 * from userland - e.g. eject(1).
1658 */
1659 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1660
1661 mutex_unlock(&bdev->bd_mutex);
1662
Al Viro4385bab2013-05-05 22:11:03 -04001663 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001664}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001665EXPORT_SYMBOL(blkdev_put);
1666
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667static int blkdev_close(struct inode * inode, struct file * filp)
1668{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001669 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001670 blkdev_put(bdev, filp->f_mode);
1671 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672}
1673
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001674static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001676 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001677 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001678
1679 /*
1680 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1681 * to updated it before every ioctl.
1682 */
Al Viro56b26ad2008-09-19 03:17:36 -04001683 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001684 mode |= FMODE_NDELAY;
1685 else
1686 mode &= ~FMODE_NDELAY;
1687
Al Viro56b26ad2008-09-19 03:17:36 -04001688 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689}
1690
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001691/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001692 * Write data to the block device. Only intended for the block device itself
1693 * and the raw driver which basically is a fake block device.
1694 *
1695 * Does not take i_mutex for the write and thus is not for general purpose
1696 * use.
1697 */
Al Viro1456c0a2014-04-03 03:21:50 -04001698ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001699{
1700 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001701 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001702 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001703 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001704 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001705
Al Viro7ec7b942015-04-07 11:35:14 -04001706 if (bdev_read_only(I_BDEV(bd_inode)))
1707 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001708
Al Viro7ec7b942015-04-07 11:35:14 -04001709 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001710 return 0;
1711
Al Viro7ec7b942015-04-07 11:35:14 -04001712 if (iocb->ki_pos >= size)
1713 return -ENOSPC;
1714
1715 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001716
Jianpeng Ma53362a02012-08-02 09:50:39 +02001717 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001718 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwige2592212016-04-07 08:52:01 -07001719 if (ret > 0)
1720 ret = generic_write_sync(iocb, ret);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001721 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001722 return ret;
1723}
Al Viro1456c0a2014-04-03 03:21:50 -04001724EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001725
David Jefferyb2de5252014-09-29 10:21:10 -04001726ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001727{
1728 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001729 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001730 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001731 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001732
1733 if (pos >= size)
1734 return 0;
1735
1736 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001737 iov_iter_truncate(to, size);
1738 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001739}
David Jefferyb2de5252014-09-29 10:21:10 -04001740EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001741
Christoph Hellwigeef99382009-08-20 17:43:41 +02001742/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001743 * Try to release a page associated with block device when the system
1744 * is under memory pressure.
1745 */
1746static int blkdev_releasepage(struct page *page, gfp_t wait)
1747{
1748 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1749
1750 if (super && super->s_op->bdev_try_to_free_page)
1751 return super->s_op->bdev_try_to_free_page(super, page, wait);
1752
1753 return try_to_free_buffers(page);
1754}
1755
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001756static int blkdev_writepages(struct address_space *mapping,
1757 struct writeback_control *wbc)
1758{
1759 if (dax_mapping(mapping)) {
1760 struct block_device *bdev = I_BDEV(mapping->host);
1761
1762 return dax_writeback_mapping_range(mapping, bdev, wbc);
1763 }
1764 return generic_writepages(mapping, wbc);
1765}
1766
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001767static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001769 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001771 .write_begin = blkdev_write_begin,
1772 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001773 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001774 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001776 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777};
1778
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001779#define BLKDEV_FALLOC_FL_SUPPORTED \
1780 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1781 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1782
1783static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1784 loff_t len)
1785{
1786 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1787 struct request_queue *q = bdev_get_queue(bdev);
1788 struct address_space *mapping;
1789 loff_t end = start + len - 1;
1790 loff_t isize;
1791 int error;
1792
1793 /* Fail if we don't recognize the flags. */
1794 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1795 return -EOPNOTSUPP;
1796
1797 /* Don't go off the end of the device. */
1798 isize = i_size_read(bdev->bd_inode);
1799 if (start >= isize)
1800 return -EINVAL;
1801 if (end >= isize) {
1802 if (mode & FALLOC_FL_KEEP_SIZE) {
1803 len = isize - start;
1804 end = start + len - 1;
1805 } else
1806 return -EINVAL;
1807 }
1808
1809 /*
1810 * Don't allow IO that isn't aligned to logical block size.
1811 */
1812 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1813 return -EINVAL;
1814
1815 /* Invalidate the page cache, including dirty pages. */
1816 mapping = bdev->bd_inode->i_mapping;
1817 truncate_inode_pages_range(mapping, start, end);
1818
1819 switch (mode) {
1820 case FALLOC_FL_ZERO_RANGE:
1821 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1822 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1823 GFP_KERNEL, false);
1824 break;
1825 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1826 /* Only punch if the device can do zeroing discard. */
1827 if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data)
1828 return -EOPNOTSUPP;
1829 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1830 GFP_KERNEL, 0);
1831 break;
1832 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1833 if (!blk_queue_discard(q))
1834 return -EOPNOTSUPP;
1835 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1836 GFP_KERNEL, 0);
1837 break;
1838 default:
1839 return -EOPNOTSUPP;
1840 }
1841 if (error)
1842 return error;
1843
1844 /*
1845 * Invalidate again; if someone wandered in and dirtied a page,
1846 * the caller will be given -EBUSY. The third argument is
1847 * inclusive, so the rounding here is safe.
1848 */
1849 return invalidate_inode_pages2_range(mapping,
1850 start >> PAGE_SHIFT,
1851 end >> PAGE_SHIFT);
1852}
1853
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001854const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 .open = blkdev_open,
1856 .release = blkdev_close,
1857 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001858 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001859 .write_iter = blkdev_write_iter,
Dan Williamsacc93d32016-05-07 11:40:28 -07001860 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001861 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001862 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863#ifdef CONFIG_COMPAT
1864 .compat_ioctl = compat_blkdev_ioctl,
1865#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001866 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001867 .splice_write = iter_file_splice_write,
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001868 .fallocate = blkdev_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869};
1870
1871int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1872{
1873 int res;
1874 mm_segment_t old_fs = get_fs();
1875 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001876 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 set_fs(old_fs);
1878 return res;
1879}
1880
1881EXPORT_SYMBOL(ioctl_by_bdev);
1882
1883/**
1884 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001885 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001887 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 * namespace if possible and return it. Return ERR_PTR(error)
1889 * otherwise.
1890 */
Al Viro421748e2008-08-02 01:04:36 -04001891struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892{
1893 struct block_device *bdev;
1894 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001895 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 int error;
1897
Al Viro421748e2008-08-02 01:04:36 -04001898 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 return ERR_PTR(-EINVAL);
1900
Al Viro421748e2008-08-02 01:04:36 -04001901 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 if (error)
1903 return ERR_PTR(error);
1904
David Howellsbb6687342015-03-17 22:26:21 +00001905 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 error = -ENOTBLK;
1907 if (!S_ISBLK(inode->i_mode))
1908 goto fail;
1909 error = -EACCES;
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05001910 if (!may_open_dev(&path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 goto fail;
1912 error = -ENOMEM;
1913 bdev = bd_acquire(inode);
1914 if (!bdev)
1915 goto fail;
1916out:
Al Viro421748e2008-08-02 01:04:36 -04001917 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 return bdev;
1919fail:
1920 bdev = ERR_PTR(error);
1921 goto out;
1922}
Al Virod5686b42008-08-01 05:00:11 -04001923EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
NeilBrown93b270f2011-02-24 17:25:47 +11001925int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001926{
1927 struct super_block *sb = get_super(bdev);
1928 int res = 0;
1929
1930 if (sb) {
1931 /*
1932 * no need to lock the super, get_super holds the
1933 * read mutex so the filesystem cannot go away
1934 * under us (->put_super runs with the write lock
1935 * hold).
1936 */
1937 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001938 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001939 drop_super(sb);
1940 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001941 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001942 return res;
1943}
1944EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001945
1946void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1947{
1948 struct inode *inode, *old_inode = NULL;
1949
Dave Chinner74278da2015-03-04 12:37:22 -05001950 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001951 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1952 struct address_space *mapping = inode->i_mapping;
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001953 struct block_device *bdev;
Jan Kara5c0d6b62012-07-03 16:45:31 +02001954
1955 spin_lock(&inode->i_lock);
1956 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1957 mapping->nrpages == 0) {
1958 spin_unlock(&inode->i_lock);
1959 continue;
1960 }
1961 __iget(inode);
1962 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05001963 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001964 /*
1965 * We hold a reference to 'inode' so it couldn't have been
1966 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05001967 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02001968 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05001969 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02001970 * later.
1971 */
1972 iput(old_inode);
1973 old_inode = inode;
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001974 bdev = I_BDEV(inode);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001975
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001976 mutex_lock(&bdev->bd_mutex);
1977 if (bdev->bd_openers)
1978 func(bdev, arg);
1979 mutex_unlock(&bdev->bd_mutex);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001980
Dave Chinner74278da2015-03-04 12:37:22 -05001981 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001982 }
Dave Chinner74278da2015-03-04 12:37:22 -05001983 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001984 iput(old_inode);
1985}