blob: 98b37e77683d33759b3d6e7cf178100f5e2a6a75 [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
Jan Kara428ca582019-01-14 09:48:10 +0100117static void set_init_blocksize(struct block_device *bdev)
118{
119 unsigned bsize = bdev_logical_block_size(bdev);
120 loff_t size = i_size_read(bdev->bd_inode);
121
122 while (bsize < PAGE_SIZE) {
123 if (size & bsize)
124 break;
125 bsize <<= 1;
126 }
127 bdev->bd_block_size = bsize;
128 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
129}
130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131int set_blocksize(struct block_device *bdev, int size)
132{
133 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700134 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 return -EINVAL;
136
137 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400138 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return -EINVAL;
140
141 /* Don't change the size if it is same as current */
142 if (bdev->bd_block_size != size) {
143 sync_blockdev(bdev);
144 bdev->bd_block_size = size;
145 bdev->bd_inode->i_blkbits = blksize_bits(size);
146 kill_bdev(bdev);
147 }
148 return 0;
149}
150
151EXPORT_SYMBOL(set_blocksize);
152
153int sb_set_blocksize(struct super_block *sb, int size)
154{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (set_blocksize(sb->s_bdev, size))
156 return 0;
157 /* If we get here, we know size is power of two
158 * and it's value is between 512 and PAGE_SIZE */
159 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800160 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 return sb->s_blocksize;
162}
163
164EXPORT_SYMBOL(sb_set_blocksize);
165
166int sb_min_blocksize(struct super_block *sb, int size)
167{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400168 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 if (size < minsize)
170 size = minsize;
171 return sb_set_blocksize(sb, size);
172}
173
174EXPORT_SYMBOL(sb_min_blocksize);
175
176static int
177blkdev_get_block(struct inode *inode, sector_t iblock,
178 struct buffer_head *bh, int create)
179{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 bh->b_bdev = I_BDEV(inode);
181 bh->b_blocknr = iblock;
182 set_buffer_mapped(bh);
183 return 0;
184}
185
Dan Williams4ebb16c2015-10-28 07:48:19 +0900186static struct inode *bdev_file_inode(struct file *file)
187{
188 return file->f_mapping->host;
189}
190
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800191static ssize_t
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700192blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800193{
194 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +0900195 struct inode *inode = bdev_file_inode(file);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800196
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700197 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter,
Jens Axboefe0f07d2015-04-15 17:05:48 -0600198 blkdev_get_block, NULL, NULL,
199 DIO_SKIP_DIO_COUNT);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800200}
201
Jan Kara5cee5812009-04-27 16:43:51 +0200202int __sync_blockdev(struct block_device *bdev, int wait)
203{
204 if (!bdev)
205 return 0;
206 if (!wait)
207 return filemap_flush(bdev->bd_inode->i_mapping);
208 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
209}
210
Nick Piggin585d3bc2009-02-25 10:44:19 +0100211/*
212 * Write out and wait upon all the dirty data associated with a block
213 * device via its mapping. Does not take the superblock lock.
214 */
215int sync_blockdev(struct block_device *bdev)
216{
Jan Kara5cee5812009-04-27 16:43:51 +0200217 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100218}
219EXPORT_SYMBOL(sync_blockdev);
220
221/*
222 * Write out and wait upon all dirty data associated with this
223 * device. Filesystem data as well as the underlying block
224 * device. Takes the superblock lock.
225 */
226int fsync_bdev(struct block_device *bdev)
227{
228 struct super_block *sb = get_super(bdev);
229 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200230 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100231 drop_super(sb);
232 return res;
233 }
234 return sync_blockdev(bdev);
235}
Al Viro47e44912009-04-01 07:07:16 -0400236EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100237
238/**
239 * freeze_bdev -- lock a filesystem and force it into a consistent state
240 * @bdev: blockdevice to lock
241 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100242 * If a superblock is found on this device, we take the s_umount semaphore
243 * on it to make sure nobody unmounts until the snapshot creation is done.
244 * The reference counter (bd_fsfreeze_count) guarantees that only the last
245 * unfreeze process can unfreeze the frozen filesystem actually when multiple
246 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
247 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
248 * actually.
249 */
250struct super_block *freeze_bdev(struct block_device *bdev)
251{
252 struct super_block *sb;
253 int error = 0;
254
255 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200256 if (++bdev->bd_fsfreeze_count > 1) {
257 /*
258 * We don't even need to grab a reference - the first call
259 * to freeze_bdev grab an active reference and only the last
260 * thaw_bdev drops it.
261 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100262 sb = get_super(bdev);
Andrey Ryabinin5bb53c02016-08-23 18:55:31 +0300263 if (sb)
264 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
266 return sb;
267 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100268
Christoph Hellwig45042302009-08-03 23:28:35 +0200269 sb = get_active_super(bdev);
270 if (!sb)
271 goto out;
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600272 if (sb->s_op->freeze_super)
273 error = sb->s_op->freeze_super(sb);
274 else
275 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400276 if (error) {
277 deactivate_super(sb);
278 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200279 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400280 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100281 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400282 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200283 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100284 sync_blockdev(bdev);
285 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200286 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100287}
288EXPORT_SYMBOL(freeze_bdev);
289
290/**
291 * thaw_bdev -- unlock filesystem
292 * @bdev: blockdevice to unlock
293 * @sb: associated superblock
294 *
295 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
296 */
297int thaw_bdev(struct block_device *bdev, struct super_block *sb)
298{
Christoph Hellwig45042302009-08-03 23:28:35 +0200299 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100300
301 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200302 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400303 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100304
Christoph Hellwig45042302009-08-03 23:28:35 +0200305 error = 0;
306 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400307 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100308
Christoph Hellwig45042302009-08-03 23:28:35 +0200309 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400310 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200311
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600312 if (sb->s_op->thaw_super)
313 error = sb->s_op->thaw_super(sb);
314 else
315 error = thaw_super(sb);
Pierre Morel997198b2016-10-04 10:53:40 +0200316 if (error)
Josef Bacik18e9e512010-03-23 10:34:56 -0400317 bdev->bd_fsfreeze_count++;
Josef Bacik18e9e512010-03-23 10:34:56 -0400318out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100319 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Pierre Morel997198b2016-10-04 10:53:40 +0200320 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100321}
322EXPORT_SYMBOL(thaw_bdev);
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
325{
326 return block_write_full_page(page, blkdev_get_block, wbc);
327}
328
329static int blkdev_readpage(struct file * file, struct page * page)
330{
331 return block_read_full_page(page, blkdev_get_block);
332}
333
Akinobu Mita447f05b2014-10-09 15:26:58 -0700334static int blkdev_readpages(struct file *file, struct address_space *mapping,
335 struct list_head *pages, unsigned nr_pages)
336{
337 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
338}
339
Nick Piggin6272b5a2007-10-16 01:25:04 -0700340static int blkdev_write_begin(struct file *file, struct address_space *mapping,
341 loff_t pos, unsigned len, unsigned flags,
342 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200344 return block_write_begin(mapping, pos, len, flags, pagep,
345 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346}
347
Nick Piggin6272b5a2007-10-16 01:25:04 -0700348static int blkdev_write_end(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned copied,
350 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700352 int ret;
353 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
354
355 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300356 put_page(page);
Nick Piggin6272b5a2007-10-16 01:25:04 -0700357
358 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359}
360
361/*
362 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500363 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 * so we compute the size by hand (just as in block_read/write above)
365 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800366static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900368 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 loff_t retval;
370
Al Viro59551022016-01-22 15:40:57 -0500371 inode_lock(bd_inode);
Al Viro5d48f3a2013-06-23 21:34:45 +0400372 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Al Viro59551022016-01-22 15:40:57 -0500373 inode_unlock(bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 return retval;
375}
376
Josef Bacik02c24a82011-07-16 20:44:56 -0400377int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900379 struct inode *bd_inode = bdev_file_inode(filp);
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400380 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100381 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200382
383 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
384 if (error)
385 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100386
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400387 /*
388 * There is no need to serialise calls to blkdev_issue_flush with
389 * i_mutex and doing so causes performance issues with concurrent
390 * O_SYNC writers to a block device.
391 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200392 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100393 if (error == -EOPNOTSUPP)
394 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400395
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100396 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700398EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700400/**
401 * bdev_read_page() - Start reading a page from a block device
402 * @bdev: The device to read the page from
403 * @sector: The offset on the device to read the page to (need not be aligned)
404 * @page: The page to read
405 *
406 * On entry, the page should be locked. It will be unlocked when the page
407 * has been read. If the block driver implements rw_page synchronously,
408 * that will be true on exit from this function, but it need not be.
409 *
410 * Errors returned by this function are usually "soft", eg out of memory, or
411 * queue full; callers should try a different route to read this page rather
412 * than propagate an error back up the stack.
413 *
414 * Return: negative errno if an error occurs, 0 if submission was successful.
415 */
416int bdev_read_page(struct block_device *bdev, sector_t sector,
417 struct page *page)
418{
419 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800420 int result = -EOPNOTSUPP;
421
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400422 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc92015-11-19 13:29:28 -0800423 return result;
424
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100425 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800426 if (result)
427 return result;
Jens Axboec11f0c02016-08-05 08:11:04 -0600428 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800429 blk_queue_exit(bdev->bd_queue);
430 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700431}
432EXPORT_SYMBOL_GPL(bdev_read_page);
433
434/**
435 * bdev_write_page() - Start writing a page to a block device
436 * @bdev: The device to write the page to
437 * @sector: The offset on the device to write the page to (need not be aligned)
438 * @page: The page to write
439 * @wbc: The writeback_control for the write
440 *
441 * On entry, the page should be locked and not currently under writeback.
442 * On exit, if the write started successfully, the page will be unlocked and
443 * under writeback. If the write failed already (eg the driver failed to
444 * queue the page to the device), the page will still be locked. If the
445 * caller is a ->writepage implementation, it will need to unlock the page.
446 *
447 * Errors returned by this function are usually "soft", eg out of memory, or
448 * queue full; callers should try a different route to write this page rather
449 * than propagate an error back up the stack.
450 *
451 * Return: negative errno if an error occurs, 0 if submission was successful.
452 */
453int bdev_write_page(struct block_device *bdev, sector_t sector,
454 struct page *page, struct writeback_control *wbc)
455{
456 int result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700457 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800458
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400459 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700460 return -EOPNOTSUPP;
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100461 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800462 if (result)
463 return result;
464
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700465 set_page_writeback(page);
Jens Axboec11f0c02016-08-05 08:11:04 -0600466 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
Matthew Wilcox133ca5c2017-10-13 15:58:15 -0700467 if (result) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700468 end_page_writeback(page);
Matthew Wilcox133ca5c2017-10-13 15:58:15 -0700469 } else {
470 clean_page_buffers(page);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700471 unlock_page(page);
Matthew Wilcox133ca5c2017-10-13 15:58:15 -0700472 }
Dan Williams2e6edc92015-11-19 13:29:28 -0800473 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700474 return result;
475}
476EXPORT_SYMBOL_GPL(bdev_write_page);
477
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200478/**
479 * bdev_direct_access() - Get the address for directly-accessibly memory
480 * @bdev: The device containing the memory
Dan Williamsb2e0d162016-01-15 16:55:59 -0800481 * @dax: control and output parameters for ->direct_access
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200482 *
483 * If a block device is made up of directly addressable memory, this function
484 * will tell the caller the PFN and the address of the memory. The address
485 * may be directly dereferenced within the kernel without the need to call
486 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
487 * page tables.
488 *
489 * Return: negative errno if an error occurs, otherwise the number of bytes
490 * accessible at this address.
491 */
Dan Williamsb2e0d162016-01-15 16:55:59 -0800492long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200493{
Dan Williamsb2e0d162016-01-15 16:55:59 -0800494 sector_t sector = dax->sector;
495 long avail, size = dax->size;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200496 const struct block_device_operations *ops = bdev->bd_disk->fops;
497
Matthew Wilcox43c3dd02015-07-03 10:40:43 -0400498 /*
499 * The device driver is allowed to sleep, in order to make the
500 * memory directly accessible.
501 */
502 might_sleep();
503
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200504 if (size < 0)
505 return size;
Toshi Kani163d4ba2016-06-23 17:05:50 -0400506 if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200507 return -EOPNOTSUPP;
508 if ((sector + DIV_ROUND_UP(size, 512)) >
509 part_nr_sects_read(bdev->bd_part))
510 return -ERANGE;
511 sector += get_start_sect(bdev);
512 if (sector % (PAGE_SIZE / 512))
513 return -EINVAL;
Dan Williams0a70bd42016-02-24 14:02:11 -0800514 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size);
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200515 if (!avail)
516 return -ERANGE;
Dan Williamsfe683ad2016-01-15 16:55:56 -0800517 if (avail > 0 && avail & ~PAGE_MASK)
518 return -ENXIO;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200519 return min(avail, size);
520}
521EXPORT_SYMBOL_GPL(bdev_direct_access);
522
Toshi Kani2d96afc2016-05-10 10:23:53 -0600523/**
524 * bdev_dax_supported() - Check if the device supports dax for filesystem
525 * @sb: The superblock of the device
526 * @blocksize: The block size of the device
527 *
528 * This is a library function for filesystems to check if the block device
529 * can be mounted with dax option.
530 *
531 * Return: negative errno if unsupported, 0 if supported.
532 */
533int bdev_dax_supported(struct super_block *sb, int blocksize)
534{
535 struct blk_dax_ctl dax = {
536 .sector = 0,
537 .size = PAGE_SIZE,
538 };
539 int err;
540
541 if (blocksize != PAGE_SIZE) {
542 vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
543 return -EINVAL;
544 }
545
546 err = bdev_direct_access(sb->s_bdev, &dax);
547 if (err < 0) {
548 switch (err) {
549 case -EOPNOTSUPP:
550 vfs_msg(sb, KERN_ERR,
551 "error: device does not support dax");
552 break;
553 case -EINVAL:
554 vfs_msg(sb, KERN_ERR,
555 "error: unaligned partition for dax");
556 break;
557 default:
558 vfs_msg(sb, KERN_ERR,
559 "error: dax access failed (%d)", err);
560 }
561 return err;
562 }
563
564 return 0;
565}
566EXPORT_SYMBOL_GPL(bdev_dax_supported);
567
Toshi Kania8078b12016-05-10 10:23:57 -0600568/**
569 * bdev_dax_capable() - Return if the raw device is capable for dax
570 * @bdev: The device for raw block device access
571 */
572bool bdev_dax_capable(struct block_device *bdev)
573{
Toshi Kania8078b12016-05-10 10:23:57 -0600574 struct blk_dax_ctl dax = {
575 .size = PAGE_SIZE,
576 };
577
578 if (!IS_ENABLED(CONFIG_FS_DAX))
579 return false;
580
581 dax.sector = 0;
582 if (bdev_direct_access(bdev, &dax) < 0)
583 return false;
584
585 dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512);
586 if (bdev_direct_access(bdev, &dax) < 0)
587 return false;
588
Toshi Kania8078b12016-05-10 10:23:57 -0600589 return true;
590}
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592/*
593 * pseudo-fs
594 */
595
596static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800597static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599static struct inode *bdev_alloc_inode(struct super_block *sb)
600{
Christoph Lametere94b1762006-12-06 20:33:17 -0800601 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 if (!ei)
603 return NULL;
604 return &ei->vfs_inode;
605}
606
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100607static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100609 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 struct bdev_inode *bdi = BDEV_I(inode);
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 kmem_cache_free(bdev_cachep, bdi);
613}
614
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100615static void bdev_destroy_inode(struct inode *inode)
616{
617 call_rcu(&inode->i_rcu, bdev_i_callback);
618}
619
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700620static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
622 struct bdev_inode *ei = (struct bdev_inode *) foo;
623 struct block_device *bdev = &ei->bdev;
624
Christoph Lametera35afb82007-05-16 22:10:57 -0700625 memset(bdev, 0, sizeof(*bdev));
626 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700627 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100628#ifdef CONFIG_SYSFS
629 INIT_LIST_HEAD(&bdev->bd_holder_disks);
630#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700631 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800632 /* Initialize mutex for freeze. */
633 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634}
635
Al Virob57922d2010-06-07 14:34:48 -0400636static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
638 struct block_device *bdev = &BDEV_I(inode)->bdev;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700639 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400640 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200641 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 spin_lock(&bdev_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 list_del_init(&bdev->bd_list);
644 spin_unlock(&bdev_lock);
645}
646
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800647static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 .statfs = simple_statfs,
649 .alloc_inode = bdev_alloc_inode,
650 .destroy_inode = bdev_destroy_inode,
651 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400652 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653};
654
Al Viro51139ad2010-07-25 23:47:46 +0400655static struct dentry *bd_mount(struct file_system_type *fs_type,
656 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
Shaohua Li3684aa72016-02-22 15:27:40 -0700658 struct dentry *dent;
659 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Vegard Nossume9e5e3f2016-08-22 12:47:43 +0200660 if (!IS_ERR(dent))
Shaohua Li3684aa72016-02-22 15:27:40 -0700661 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
662 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665static struct file_system_type bd_type = {
666 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400667 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 .kill_sb = kill_anon_super,
669};
670
Tejun Heoa212b102015-05-22 17:13:33 -0400671struct super_block *blockdev_superblock __read_mostly;
672EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
674void __init bdev_cache_init(void)
675{
676 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300677 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800678
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800680 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800681 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900682 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 err = register_filesystem(&bd_type);
684 if (err)
685 panic("Cannot register bdev pseudo-fs");
686 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 if (IS_ERR(bd_mnt))
688 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300689 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
692/*
693 * Most likely _very_ bad one - but then it's hardly critical for small
694 * /dev and can be fixed when somebody will need really large one.
695 * Keep in mind that it will be fed through icache hash function too.
696 */
697static inline unsigned long hash(dev_t dev)
698{
699 return MAJOR(dev)+MINOR(dev);
700}
701
702static int bdev_test(struct inode *inode, void *data)
703{
704 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
705}
706
707static int bdev_set(struct inode *inode, void *data)
708{
709 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
710 return 0;
711}
712
713static LIST_HEAD(all_bdevs);
714
715struct block_device *bdget(dev_t dev)
716{
717 struct block_device *bdev;
718 struct inode *inode;
719
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800720 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 bdev_test, bdev_set, &dev);
722
723 if (!inode)
724 return NULL;
725
726 bdev = &BDEV_I(inode)->bdev;
727
728 if (inode->i_state & I_NEW) {
729 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000730 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 bdev->bd_inode = inode;
Fabian Frederick61604a22017-02-27 14:28:32 -0800732 bdev->bd_block_size = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 bdev->bd_part_count = 0;
734 bdev->bd_invalidated = 0;
735 inode->i_mode = S_IFBLK;
736 inode->i_rdev = dev;
737 inode->i_bdev = bdev;
738 inode->i_data.a_ops = &def_blk_aops;
739 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 spin_lock(&bdev_lock);
741 list_add(&bdev->bd_list, &all_bdevs);
742 spin_unlock(&bdev_lock);
743 unlock_new_inode(inode);
744 }
745 return bdev;
746}
747
748EXPORT_SYMBOL(bdget);
749
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200750/**
751 * bdgrab -- Grab a reference to an already referenced block device
752 * @bdev: Block device to grab a reference to.
753 */
754struct block_device *bdgrab(struct block_device *bdev)
755{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400756 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200757 return bdev;
758}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700759EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200760
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761long nr_blockdev_pages(void)
762{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700763 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 long ret = 0;
765 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700766 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 ret += bdev->bd_inode->i_mapping->nrpages;
768 }
769 spin_unlock(&bdev_lock);
770 return ret;
771}
772
773void bdput(struct block_device *bdev)
774{
775 iput(bdev->bd_inode);
776}
777
778EXPORT_SYMBOL(bdput);
779
780static struct block_device *bd_acquire(struct inode *inode)
781{
782 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 spin_lock(&bdev_lock);
785 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700786 if (bdev) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100787 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 spin_unlock(&bdev_lock);
789 return bdev;
790 }
791 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 bdev = bdget(inode->i_rdev);
794 if (bdev) {
795 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700796 if (!inode->i_bdev) {
797 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400798 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700799 * and it's released in clear_inode() of inode.
800 * So, we can access it via ->i_mapping always
801 * without igrab().
802 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100803 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700804 inode->i_bdev = bdev;
805 inode->i_mapping = bdev->bd_inode->i_mapping;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 spin_unlock(&bdev_lock);
808 }
809 return bdev;
810}
811
812/* Call when you free inode */
813
814void bd_forget(struct inode *inode)
815{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700816 struct block_device *bdev = NULL;
817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700819 if (!sb_is_blkdev_sb(inode->i_sb))
820 bdev = inode->i_bdev;
Al Viroa4a4f942016-07-19 13:16:52 -0400821 inode->i_bdev = NULL;
822 inode->i_mapping = &inode->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700824
825 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100826 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827}
828
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900829/**
830 * bd_may_claim - test whether a block device can be claimed
831 * @bdev: block device of interest
832 * @whole: whole block device containing @bdev, may equal @bdev
833 * @holder: holder trying to claim @bdev
834 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300835 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900836 *
837 * CONTEXT:
838 * spin_lock(&bdev_lock).
839 *
840 * RETURNS:
841 * %true if @bdev can be claimed, %false otherwise.
842 */
843static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
844 void *holder)
845{
846 if (bdev->bd_holder == holder)
847 return true; /* already a holder */
848 else if (bdev->bd_holder != NULL)
849 return false; /* held by someone else */
NeilBrowncfa2d652016-12-12 08:21:51 -0700850 else if (whole == bdev)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900851 return true; /* is a whole device which isn't held */
852
Tejun Heoe525fd82010-11-13 11:55:17 +0100853 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900854 return true; /* is a partition of a device that is being partitioned */
855 else if (whole->bd_holder != NULL)
856 return false; /* is a partition of a held device */
857 else
858 return true; /* is a partition of an un-held device */
859}
860
861/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900862 * bd_prepare_to_claim - prepare to claim a block device
863 * @bdev: block device of interest
864 * @whole: the whole device containing @bdev, may equal @bdev
865 * @holder: holder trying to claim @bdev
866 *
867 * Prepare to claim @bdev. This function fails if @bdev is already
868 * claimed by another holder and waits if another claiming is in
869 * progress. This function doesn't actually claim. On successful
870 * return, the caller has ownership of bd_claiming and bd_holder[s].
871 *
872 * CONTEXT:
873 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
874 * it multiple times.
875 *
876 * RETURNS:
877 * 0 if @bdev can be claimed, -EBUSY otherwise.
878 */
879static int bd_prepare_to_claim(struct block_device *bdev,
880 struct block_device *whole, void *holder)
881{
882retry:
883 /* if someone else claimed, fail */
884 if (!bd_may_claim(bdev, whole, holder))
885 return -EBUSY;
886
Tejun Heoe75aa852010-08-04 17:59:39 +0200887 /* if claiming is already in progress, wait for it to finish */
888 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900889 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
890 DEFINE_WAIT(wait);
891
892 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
893 spin_unlock(&bdev_lock);
894 schedule();
895 finish_wait(wq, &wait);
896 spin_lock(&bdev_lock);
897 goto retry;
898 }
899
900 /* yay, all mine */
901 return 0;
902}
903
904/**
905 * bd_start_claiming - start claiming a block device
906 * @bdev: block device of interest
907 * @holder: holder trying to claim @bdev
908 *
909 * @bdev is about to be opened exclusively. Check @bdev can be opened
910 * exclusively and mark that an exclusive open is in progress. Each
911 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000912 * either bd_finish_claiming() or bd_abort_claiming() (which do not
913 * fail).
914 *
915 * This function is used to gain exclusive access to the block device
916 * without actually causing other exclusive open attempts to fail. It
917 * should be used when the open sequence itself requires exclusive
918 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900919 *
920 * CONTEXT:
921 * Might sleep.
922 *
923 * RETURNS:
924 * Pointer to the block device containing @bdev on success, ERR_PTR()
925 * value on failure.
926 */
927static struct block_device *bd_start_claiming(struct block_device *bdev,
928 void *holder)
929{
930 struct gendisk *disk;
931 struct block_device *whole;
932 int partno, err;
933
934 might_sleep();
935
936 /*
937 * @bdev might not have been initialized properly yet, look up
938 * and grab the outer block device the hard way.
939 */
940 disk = get_gendisk(bdev->bd_dev, &partno);
941 if (!disk)
942 return ERR_PTR(-ENXIO);
943
Tejun Heod4c208b2011-06-13 12:45:48 +0200944 /*
945 * Normally, @bdev should equal what's returned from bdget_disk()
946 * if partno is 0; however, some drivers (floppy) use multiple
947 * bdev's for the same physical device and @bdev may be one of the
948 * aliases. Keep @bdev if partno is 0. This means claimer
949 * tracking is broken for those devices but it has always been that
950 * way.
951 */
952 if (partno)
953 whole = bdget_disk(disk, 0);
954 else
955 whole = bdgrab(bdev);
956
Nick Piggincf342572010-05-26 01:50:21 +1000957 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900958 put_disk(disk);
959 if (!whole)
960 return ERR_PTR(-ENOMEM);
961
962 /* prepare to claim, if successful, mark claiming in progress */
963 spin_lock(&bdev_lock);
964
965 err = bd_prepare_to_claim(bdev, whole, holder);
966 if (err == 0) {
967 whole->bd_claiming = holder;
968 spin_unlock(&bdev_lock);
969 return whole;
970 } else {
971 spin_unlock(&bdev_lock);
972 bdput(whole);
973 return ERR_PTR(err);
974 }
975}
976
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800977#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100978struct bd_holder_disk {
979 struct list_head list;
980 struct gendisk *disk;
981 int refcnt;
982};
983
984static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
985 struct gendisk *disk)
986{
987 struct bd_holder_disk *holder;
988
989 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
990 if (holder->disk == disk)
991 return holder;
992 return NULL;
993}
994
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700995static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800996{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700997 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800998}
999
1000static void del_symlink(struct kobject *from, struct kobject *to)
1001{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001002 sysfs_remove_link(from, kobject_name(to));
1003}
1004
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001005/**
Tejun Heoe09b4572010-11-13 11:55:17 +01001006 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1007 * @bdev: the claimed slave bdev
1008 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001009 *
Tejun Heo49731ba2011-01-14 18:43:57 +01001010 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1011 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001012 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001013 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001014 * - from "slaves" directory of the holder @disk to the claimed @bdev
1015 * - from "holders" directory of the @bdev to the holder @disk
1016 *
1017 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1018 * passed to bd_link_disk_holder(), then:
1019 *
1020 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1021 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1022 *
1023 * The caller must have claimed @bdev before calling this function and
1024 * ensure that both @bdev and @disk are valid during the creation and
1025 * lifetime of these symlinks.
1026 *
1027 * CONTEXT:
1028 * Might sleep.
1029 *
1030 * RETURNS:
1031 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001032 */
Tejun Heoe09b4572010-11-13 11:55:17 +01001033int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001034{
Tejun Heo49731ba2011-01-14 18:43:57 +01001035 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001036 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001037
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001038 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001039
Tejun Heo49731ba2011-01-14 18:43:57 +01001040 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -07001041
Tejun Heoe09b4572010-11-13 11:55:17 +01001042 /* FIXME: remove the following once add_disk() handles errors */
1043 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1044 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -07001045
Tejun Heo49731ba2011-01-14 18:43:57 +01001046 holder = bd_find_holder_disk(bdev, disk);
1047 if (holder) {
1048 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +01001049 goto out_unlock;
1050 }
1051
Tejun Heo49731ba2011-01-14 18:43:57 +01001052 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1053 if (!holder) {
1054 ret = -ENOMEM;
1055 goto out_unlock;
1056 }
1057
1058 INIT_LIST_HEAD(&holder->list);
1059 holder->disk = disk;
1060 holder->refcnt = 1;
1061
1062 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1063 if (ret)
1064 goto out_free;
1065
1066 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1067 if (ret)
1068 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +01001069 /*
1070 * bdev could be deleted beneath us which would implicitly destroy
1071 * the holder directory. Hold on to it.
1072 */
1073 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001074
1075 list_add(&holder->list, &bdev->bd_holder_disks);
1076 goto out_unlock;
1077
1078out_del:
1079 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1080out_free:
1081 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001082out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001083 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001084 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001085}
Tejun Heoe09b4572010-11-13 11:55:17 +01001086EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001087
Tejun Heo49731ba2011-01-14 18:43:57 +01001088/**
1089 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1090 * @bdev: the calimed slave bdev
1091 * @disk: the holding disk
1092 *
1093 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1094 *
1095 * CONTEXT:
1096 * Might sleep.
1097 */
1098void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001099{
Tejun Heo49731ba2011-01-14 18:43:57 +01001100 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001101
Tejun Heo49731ba2011-01-14 18:43:57 +01001102 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001103
Tejun Heo49731ba2011-01-14 18:43:57 +01001104 holder = bd_find_holder_disk(bdev, disk);
1105
1106 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1107 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1108 del_symlink(bdev->bd_part->holder_dir,
1109 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001110 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001111 list_del_init(&holder->list);
1112 kfree(holder);
1113 }
1114
1115 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001116}
Tejun Heo49731ba2011-01-14 18:43:57 +01001117EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001118#endif
1119
Andrew Patterson0c002c22008-09-04 14:27:20 -06001120/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001121 * flush_disk - invalidates all buffer-cache entries on a disk
1122 *
1123 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001124 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001125 *
1126 * Invalidates all buffer-cache entries on a disk. It should be called
1127 * when a disk has been changed -- either by a media change or online
1128 * resize.
1129 */
NeilBrown93b270f2011-02-24 17:25:47 +11001130static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001131{
NeilBrown93b270f2011-02-24 17:25:47 +11001132 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001133 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001134 "resized disk %s\n",
1135 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001136 }
1137
1138 if (!bdev->bd_disk)
1139 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001140 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001141 bdev->bd_invalidated = 1;
1142}
1143
1144/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001145 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001146 * @disk: struct gendisk to check
1147 * @bdev: struct bdev to adjust.
1148 *
1149 * This routine checks to see if the bdev size does not match the disk size
1150 * and adjusts it if it differs.
1151 */
1152void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1153{
1154 loff_t disk_size, bdev_size;
1155
1156 disk_size = (loff_t)get_capacity(disk) << 9;
1157 bdev_size = i_size_read(bdev->bd_inode);
1158 if (disk_size != bdev_size) {
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001159 printk(KERN_INFO
1160 "%s: detected capacity change from %lld to %lld\n",
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001161 disk->disk_name, bdev_size, disk_size);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001162 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001163 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001164 }
1165}
1166EXPORT_SYMBOL(check_disk_size_change);
1167
1168/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001169 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001170 * @disk: struct gendisk to be revalidated
1171 *
1172 * This routine is a wrapper for lower-level driver's revalidate_disk
1173 * call-backs. It is used to do common pre and post operations needed
1174 * for all revalidate_disk operations.
1175 */
1176int revalidate_disk(struct gendisk *disk)
1177{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001178 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001179 int ret = 0;
1180
1181 if (disk->fops->revalidate_disk)
1182 ret = disk->fops->revalidate_disk(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001183 bdev = bdget_disk(disk, 0);
1184 if (!bdev)
1185 return ret;
1186
1187 mutex_lock(&bdev->bd_mutex);
1188 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001189 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001190 mutex_unlock(&bdev->bd_mutex);
1191 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001192 return ret;
1193}
1194EXPORT_SYMBOL(revalidate_disk);
1195
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196/*
1197 * This routine checks whether a removable media has been changed,
1198 * and invalidates all buffer-cache-entries in that case. This
1199 * is a relatively slow routine, so we have to try to minimize using
1200 * it. Thus it is called only upon a 'mount' or 'open'. This
1201 * is the best way of combining speed and utility, I think.
1202 * People changing diskettes in the middle of an operation deserve
1203 * to lose :-)
1204 */
1205int check_disk_change(struct block_device *bdev)
1206{
1207 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001208 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001209 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
Tejun Heo77ea8872010-12-08 20:57:37 +01001211 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1212 DISK_EVENT_EJECT_REQUEST);
1213 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 return 0;
1215
NeilBrown93b270f2011-02-24 17:25:47 +11001216 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (bdops->revalidate_disk)
1218 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 return 1;
1220}
1221
1222EXPORT_SYMBOL(check_disk_change);
1223
1224void bd_set_size(struct block_device *bdev, loff_t size)
1225{
Al Viro59551022016-01-22 15:40:57 -05001226 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001227 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001228 inode_unlock(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229}
1230EXPORT_SYMBOL(bd_set_size);
1231
Al Viro4385bab2013-05-05 22:11:03 -04001232static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001233
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001234/*
1235 * bd_mutex locking:
1236 *
1237 * mutex_lock(part->bd_mutex)
1238 * mutex_lock_nested(whole->bd_mutex, 1)
1239 */
1240
Al Viro572c4892007-10-08 13:24:05 -04001241static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001244 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001245 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001246 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001247 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
Al Viro572c4892007-10-08 13:24:05 -04001249 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001250 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001251 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001252 perm |= MAY_WRITE;
1253 /*
1254 * hooks: /n/, see "layering violations".
1255 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001256 if (!for_part) {
1257 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
Jason Yana54b15a2020-06-16 20:16:55 +08001258 if (ret != 0)
Chris Wrightb7300b72010-08-10 18:02:55 -07001259 return ret;
Al Viro82666022008-08-01 05:32:04 -04001260 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001261
NeilBrownd3374822009-01-09 08:31:10 +11001262 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001263
Tejun Heo89f97492008-11-05 10:21:06 +01001264 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001265 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001266 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001267 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001268 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
Tejun Heo69e02c52011-03-09 19:54:27 +01001270 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001271 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 if (!bdev->bd_openers) {
1273 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001274 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 bdev->bd_contains = bdev;
Dan Williams03cdadb2016-02-26 15:19:43 -08001276
Tejun Heocf771cb2008-09-03 09:01:09 +02001277 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001278 ret = -ENXIO;
1279 bdev->bd_part = disk_get_part(disk, partno);
1280 if (!bdev->bd_part)
1281 goto out_clear;
1282
Tejun Heo1196f8b2011-04-21 20:54:45 +02001283 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001285 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001286 if (ret == -ERESTARTSYS) {
1287 /* Lost a race with 'disk' being
1288 * deleted, try again.
1289 * See md.c
1290 */
1291 disk_put_part(bdev->bd_part);
1292 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001293 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001294 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001295 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001296 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001297 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001298 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001299 goto restart;
1300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 }
Tejun Heo7e697232011-05-23 13:26:07 +02001302
Jan Kara428ca582019-01-14 09:48:10 +01001303 if (!ret) {
Tejun Heo7e697232011-05-23 13:26:07 +02001304 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Jan Kara428ca582019-01-14 09:48:10 +01001305 set_init_blocksize(bdev);
1306 }
Tejun Heo7e697232011-05-23 13:26:07 +02001307
Tejun Heo1196f8b2011-04-21 20:54:45 +02001308 /*
1309 * If the device is invalidated, rescan partition
1310 * if open succeeded or failed with -ENOMEDIUM.
1311 * The latter is necessary to prevent ghost
1312 * partitions on a removed medium.
1313 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001314 if (bdev->bd_invalidated) {
1315 if (!ret)
1316 rescan_partitions(disk, bdev);
1317 else if (ret == -ENOMEDIUM)
1318 invalidate_partitions(disk, bdev);
1319 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001320
Tejun Heo1196f8b2011-04-21 20:54:45 +02001321 if (ret)
1322 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 struct block_device *whole;
1325 whole = bdget_disk(disk, 0);
1326 ret = -ENOMEM;
1327 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001328 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001329 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001330 ret = __blkdev_get(whole, mode, 1);
Jason Yana54b15a2020-06-16 20:16:55 +08001331 if (ret) {
1332 bdput(whole);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001333 goto out_clear;
Jason Yana54b15a2020-06-16 20:16:55 +08001334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001336 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001337 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001338 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001340 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 }
Tejun Heo89f97492008-11-05 10:21:06 +01001342 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Jan Kara428ca582019-01-14 09:48:10 +01001343 set_init_blocksize(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 }
1345 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001347 ret = 0;
1348 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001349 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001350 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001351 if (bdev->bd_invalidated) {
1352 if (!ret)
1353 rescan_partitions(bdev->bd_disk, bdev);
1354 else if (ret == -ENOMEDIUM)
1355 invalidate_partitions(bdev->bd_disk, bdev);
1356 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001357 if (ret)
1358 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001360 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001361 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001362 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001365 if (for_part)
1366 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001367 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001368 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 return 0;
1370
Tejun Heo0762b8b2008-08-25 19:56:12 +09001371 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001372 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001374 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001375 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001377 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001379 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001380 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001381 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001382 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001383 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001384 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 return ret;
1387}
1388
Tejun Heod4d77622010-11-13 11:55:18 +01001389/**
1390 * blkdev_get - open a block device
1391 * @bdev: block_device to open
1392 * @mode: FMODE_* mask
1393 * @holder: exclusive holder identifier
1394 *
1395 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1396 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1397 * @holder is invalid. Exclusive opens may nest for the same @holder.
1398 *
1399 * On success, the reference count of @bdev is unchanged. On failure,
1400 * @bdev is put.
1401 *
1402 * CONTEXT:
1403 * Might sleep.
1404 *
1405 * RETURNS:
1406 * 0 on success, -errno on failure.
1407 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001408int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409{
Tejun Heoe525fd82010-11-13 11:55:17 +01001410 struct block_device *whole = NULL;
1411 int res;
1412
1413 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1414
1415 if ((mode & FMODE_EXCL) && holder) {
1416 whole = bd_start_claiming(bdev, holder);
1417 if (IS_ERR(whole)) {
1418 bdput(bdev);
1419 return PTR_ERR(whole);
1420 }
1421 }
1422
1423 res = __blkdev_get(bdev, mode, 0);
1424
1425 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001426 struct gendisk *disk = whole->bd_disk;
1427
Tejun Heo6a027ef2010-11-13 11:55:17 +01001428 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001429 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001430 spin_lock(&bdev_lock);
1431
Tejun Heo77ea8872010-12-08 20:57:37 +01001432 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001433 BUG_ON(!bd_may_claim(bdev, whole, holder));
1434 /*
1435 * Note that for a whole device bd_holders
1436 * will be incremented twice, and bd_holder
1437 * will be set to bd_may_claim before being
1438 * set to holder
1439 */
1440 whole->bd_holders++;
1441 whole->bd_holder = bd_may_claim;
1442 bdev->bd_holders++;
1443 bdev->bd_holder = holder;
1444 }
1445
1446 /* tell others that we're done */
1447 BUG_ON(whole->bd_claiming != holder);
1448 whole->bd_claiming = NULL;
1449 wake_up_bit(&whole->bd_claiming, 0);
1450
1451 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001452
1453 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001454 * Block event polling for write claims if requested. Any
1455 * write holder makes the write_holder state stick until
1456 * all are released. This is good enough and tracking
1457 * individual writeable reference is too fragile given the
1458 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001459 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001460 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1461 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001462 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001463 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001464 }
1465
1466 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001467 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001468 }
1469
Jason Yana54b15a2020-06-16 20:16:55 +08001470 if (res)
1471 bdput(bdev);
1472
Tejun Heoe525fd82010-11-13 11:55:17 +01001473 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001474}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475EXPORT_SYMBOL(blkdev_get);
1476
Tejun Heod4d77622010-11-13 11:55:18 +01001477/**
1478 * blkdev_get_by_path - open a block device by name
1479 * @path: path to the block device to open
1480 * @mode: FMODE_* mask
1481 * @holder: exclusive holder identifier
1482 *
1483 * Open the blockdevice described by the device file at @path. @mode
1484 * and @holder are identical to blkdev_get().
1485 *
1486 * On success, the returned block_device has reference count of one.
1487 *
1488 * CONTEXT:
1489 * Might sleep.
1490 *
1491 * RETURNS:
1492 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1493 */
1494struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1495 void *holder)
1496{
1497 struct block_device *bdev;
1498 int err;
1499
1500 bdev = lookup_bdev(path);
1501 if (IS_ERR(bdev))
1502 return bdev;
1503
1504 err = blkdev_get(bdev, mode, holder);
1505 if (err)
1506 return ERR_PTR(err);
1507
Chuck Ebberte51900f2011-02-16 18:11:53 -05001508 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1509 blkdev_put(bdev, mode);
1510 return ERR_PTR(-EACCES);
1511 }
1512
Tejun Heod4d77622010-11-13 11:55:18 +01001513 return bdev;
1514}
1515EXPORT_SYMBOL(blkdev_get_by_path);
1516
1517/**
1518 * blkdev_get_by_dev - open a block device by device number
1519 * @dev: device number of block device to open
1520 * @mode: FMODE_* mask
1521 * @holder: exclusive holder identifier
1522 *
1523 * Open the blockdevice described by device number @dev. @mode and
1524 * @holder are identical to blkdev_get().
1525 *
1526 * Use it ONLY if you really do not have anything better - i.e. when
1527 * you are behind a truly sucky interface and all you are given is a
1528 * device number. _Never_ to be used for internal purposes. If you
1529 * ever need it - reconsider your API.
1530 *
1531 * On success, the returned block_device has reference count of one.
1532 *
1533 * CONTEXT:
1534 * Might sleep.
1535 *
1536 * RETURNS:
1537 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1538 */
1539struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1540{
1541 struct block_device *bdev;
1542 int err;
1543
1544 bdev = bdget(dev);
1545 if (!bdev)
1546 return ERR_PTR(-ENOMEM);
1547
1548 err = blkdev_get(bdev, mode, holder);
1549 if (err)
1550 return ERR_PTR(err);
1551
1552 return bdev;
1553}
1554EXPORT_SYMBOL(blkdev_get_by_dev);
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556static int blkdev_open(struct inode * inode, struct file * filp)
1557{
1558 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
1560 /*
1561 * Preserve backwards compatibility and allow large file access
1562 * even if userspace doesn't ask for it explicitly. Some mkfs
1563 * binary needs it. We might want to drop this workaround
1564 * during an unstable branch.
1565 */
1566 filp->f_flags |= O_LARGEFILE;
1567
Al Viro572c4892007-10-08 13:24:05 -04001568 if (filp->f_flags & O_NDELAY)
1569 filp->f_mode |= FMODE_NDELAY;
1570 if (filp->f_flags & O_EXCL)
1571 filp->f_mode |= FMODE_EXCL;
1572 if ((filp->f_flags & O_ACCMODE) == 3)
1573 filp->f_mode |= FMODE_WRITE_IOCTL;
1574
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001576 if (bdev == NULL)
1577 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Al Viro572c4892007-10-08 13:24:05 -04001579 filp->f_mapping = bdev->bd_inode->i_mapping;
1580
Tejun Heoe525fd82010-11-13 11:55:17 +01001581 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582}
1583
Al Viro4385bab2013-05-05 22:11:03 -04001584static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001585{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001586 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001587 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001588
Douglas Anderson789789a2020-03-24 14:48:27 -07001589 /*
1590 * Sync early if it looks like we're the last one. If someone else
1591 * opens the block device between now and the decrement of bd_openers
1592 * then we did a sync that we didn't need to, but that's not the end
1593 * of the world and we want to avoid long (could be several minute)
1594 * syncs while holding the mutex.
1595 */
1596 if (bdev->bd_openers == 1)
1597 sync_blockdev(bdev);
1598
NeilBrown6796bf52006-12-08 02:36:16 -08001599 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001600 if (for_part)
1601 bdev->bd_part_count--;
1602
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001603 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001604 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001605 sync_blockdev(bdev);
1606 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001607
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001608 bdev_write_inode(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001609 /*
1610 * Detaching bdev inode from its wb in __destroy_inode()
1611 * is too late: the queue which embeds its bdi (along with
1612 * root wb) can be gone as soon as we put_disk() below.
1613 */
1614 inode_detach_wb(bdev->bd_inode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001615 }
1616 if (bdev->bd_contains == bdev) {
1617 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001618 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001619 }
1620 if (!bdev->bd_openers) {
1621 struct module *owner = disk->fops->owner;
1622
Tejun Heo0762b8b2008-08-25 19:56:12 +09001623 disk_put_part(bdev->bd_part);
1624 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001625 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001626 if (bdev != bdev->bd_contains)
1627 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001628 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001629
1630 put_disk(disk);
1631 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001632 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001633 mutex_unlock(&bdev->bd_mutex);
1634 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001635 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001636 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001637}
1638
Al Viro4385bab2013-05-05 22:11:03 -04001639void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001640{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001641 mutex_lock(&bdev->bd_mutex);
1642
Tejun Heoe525fd82010-11-13 11:55:17 +01001643 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001644 bool bdev_free;
1645
1646 /*
1647 * Release a claim on the device. The holder fields
1648 * are protected with bdev_lock. bd_mutex is to
1649 * synchronize disk_holder unlinking.
1650 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001651 spin_lock(&bdev_lock);
1652
1653 WARN_ON_ONCE(--bdev->bd_holders < 0);
1654 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1655
1656 /* bd_contains might point to self, check in a separate step */
1657 if ((bdev_free = !bdev->bd_holders))
1658 bdev->bd_holder = NULL;
1659 if (!bdev->bd_contains->bd_holders)
1660 bdev->bd_contains->bd_holder = NULL;
1661
1662 spin_unlock(&bdev_lock);
1663
Tejun Heo77ea8872010-12-08 20:57:37 +01001664 /*
1665 * If this was the last claim, remove holder link and
1666 * unblock evpoll if it was a write holder.
1667 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001668 if (bdev_free && bdev->bd_write_holder) {
1669 disk_unblock_events(bdev->bd_disk);
1670 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001671 }
Tejun Heo69362172011-03-09 19:54:27 +01001672 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001673
Tejun Heo85ef06d2011-07-01 16:17:47 +02001674 /*
1675 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1676 * event. This is to ensure detection of media removal commanded
1677 * from userland - e.g. eject(1).
1678 */
1679 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1680
1681 mutex_unlock(&bdev->bd_mutex);
1682
Al Viro4385bab2013-05-05 22:11:03 -04001683 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001684}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001685EXPORT_SYMBOL(blkdev_put);
1686
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687static int blkdev_close(struct inode * inode, struct file * filp)
1688{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001689 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001690 blkdev_put(bdev, filp->f_mode);
1691 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692}
1693
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001694static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001696 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001697 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001698
1699 /*
1700 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1701 * to updated it before every ioctl.
1702 */
Al Viro56b26ad2008-09-19 03:17:36 -04001703 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001704 mode |= FMODE_NDELAY;
1705 else
1706 mode &= ~FMODE_NDELAY;
1707
Al Viro56b26ad2008-09-19 03:17:36 -04001708 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709}
1710
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001711/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001712 * Write data to the block device. Only intended for the block device itself
1713 * and the raw driver which basically is a fake block device.
1714 *
1715 * Does not take i_mutex for the write and thus is not for general purpose
1716 * use.
1717 */
Al Viro1456c0a2014-04-03 03:21:50 -04001718ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001719{
1720 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001721 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001722 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001723 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001724 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001725
Al Viro7ec7b942015-04-07 11:35:14 -04001726 if (bdev_read_only(I_BDEV(bd_inode)))
1727 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001728
Al Viro7ec7b942015-04-07 11:35:14 -04001729 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001730 return 0;
1731
Al Viro7ec7b942015-04-07 11:35:14 -04001732 if (iocb->ki_pos >= size)
1733 return -ENOSPC;
1734
1735 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001736
Jianpeng Ma53362a02012-08-02 09:50:39 +02001737 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001738 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwige2592212016-04-07 08:52:01 -07001739 if (ret > 0)
1740 ret = generic_write_sync(iocb, ret);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001741 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001742 return ret;
1743}
Al Viro1456c0a2014-04-03 03:21:50 -04001744EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001745
David Jefferyb2de5252014-09-29 10:21:10 -04001746ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001747{
1748 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001749 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001750 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001751 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001752
1753 if (pos >= size)
1754 return 0;
1755
1756 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001757 iov_iter_truncate(to, size);
1758 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001759}
David Jefferyb2de5252014-09-29 10:21:10 -04001760EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001761
Christoph Hellwigeef99382009-08-20 17:43:41 +02001762/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001763 * Try to release a page associated with block device when the system
1764 * is under memory pressure.
1765 */
1766static int blkdev_releasepage(struct page *page, gfp_t wait)
1767{
1768 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1769
1770 if (super && super->s_op->bdev_try_to_free_page)
1771 return super->s_op->bdev_try_to_free_page(super, page, wait);
1772
1773 return try_to_free_buffers(page);
1774}
1775
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001776static int blkdev_writepages(struct address_space *mapping,
1777 struct writeback_control *wbc)
1778{
1779 if (dax_mapping(mapping)) {
1780 struct block_device *bdev = I_BDEV(mapping->host);
1781
1782 return dax_writeback_mapping_range(mapping, bdev, wbc);
1783 }
1784 return generic_writepages(mapping, wbc);
1785}
1786
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001787static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001789 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001791 .write_begin = blkdev_write_begin,
1792 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001793 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001794 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001796 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797};
1798
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001799#define BLKDEV_FALLOC_FL_SUPPORTED \
1800 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1801 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1802
1803static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1804 loff_t len)
1805{
1806 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1807 struct request_queue *q = bdev_get_queue(bdev);
1808 struct address_space *mapping;
1809 loff_t end = start + len - 1;
1810 loff_t isize;
1811 int error;
1812
1813 /* Fail if we don't recognize the flags. */
1814 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1815 return -EOPNOTSUPP;
1816
1817 /* Don't go off the end of the device. */
1818 isize = i_size_read(bdev->bd_inode);
1819 if (start >= isize)
1820 return -EINVAL;
1821 if (end >= isize) {
1822 if (mode & FALLOC_FL_KEEP_SIZE) {
1823 len = isize - start;
1824 end = start + len - 1;
1825 } else
1826 return -EINVAL;
1827 }
1828
1829 /*
1830 * Don't allow IO that isn't aligned to logical block size.
1831 */
1832 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1833 return -EINVAL;
1834
1835 /* Invalidate the page cache, including dirty pages. */
1836 mapping = bdev->bd_inode->i_mapping;
1837 truncate_inode_pages_range(mapping, start, end);
1838
1839 switch (mode) {
1840 case FALLOC_FL_ZERO_RANGE:
1841 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1842 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1843 GFP_KERNEL, false);
1844 break;
1845 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1846 /* Only punch if the device can do zeroing discard. */
1847 if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data)
1848 return -EOPNOTSUPP;
1849 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1850 GFP_KERNEL, 0);
1851 break;
1852 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1853 if (!blk_queue_discard(q))
1854 return -EOPNOTSUPP;
1855 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1856 GFP_KERNEL, 0);
1857 break;
1858 default:
1859 return -EOPNOTSUPP;
1860 }
1861 if (error)
1862 return error;
1863
1864 /*
1865 * Invalidate again; if someone wandered in and dirtied a page,
1866 * the caller will be given -EBUSY. The third argument is
1867 * inclusive, so the rounding here is safe.
1868 */
1869 return invalidate_inode_pages2_range(mapping,
1870 start >> PAGE_SHIFT,
1871 end >> PAGE_SHIFT);
1872}
1873
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001874const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 .open = blkdev_open,
1876 .release = blkdev_close,
1877 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001878 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001879 .write_iter = blkdev_write_iter,
Dan Williamsacc93d32016-05-07 11:40:28 -07001880 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001881 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001882 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883#ifdef CONFIG_COMPAT
1884 .compat_ioctl = compat_blkdev_ioctl,
1885#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001886 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001887 .splice_write = iter_file_splice_write,
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001888 .fallocate = blkdev_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889};
1890
1891int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1892{
1893 int res;
1894 mm_segment_t old_fs = get_fs();
1895 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001896 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 set_fs(old_fs);
1898 return res;
1899}
1900
1901EXPORT_SYMBOL(ioctl_by_bdev);
1902
1903/**
1904 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001905 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001907 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 * namespace if possible and return it. Return ERR_PTR(error)
1909 * otherwise.
1910 */
Al Viro421748e2008-08-02 01:04:36 -04001911struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912{
1913 struct block_device *bdev;
1914 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001915 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 int error;
1917
Al Viro421748e2008-08-02 01:04:36 -04001918 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 return ERR_PTR(-EINVAL);
1920
Al Viro421748e2008-08-02 01:04:36 -04001921 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 if (error)
1923 return ERR_PTR(error);
1924
David Howellsbb6687342015-03-17 22:26:21 +00001925 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 error = -ENOTBLK;
1927 if (!S_ISBLK(inode->i_mode))
1928 goto fail;
1929 error = -EACCES;
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05001930 if (!may_open_dev(&path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 goto fail;
1932 error = -ENOMEM;
1933 bdev = bd_acquire(inode);
1934 if (!bdev)
1935 goto fail;
1936out:
Al Viro421748e2008-08-02 01:04:36 -04001937 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 return bdev;
1939fail:
1940 bdev = ERR_PTR(error);
1941 goto out;
1942}
Al Virod5686b42008-08-01 05:00:11 -04001943EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
NeilBrown93b270f2011-02-24 17:25:47 +11001945int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001946{
1947 struct super_block *sb = get_super(bdev);
1948 int res = 0;
1949
1950 if (sb) {
1951 /*
1952 * no need to lock the super, get_super holds the
1953 * read mutex so the filesystem cannot go away
1954 * under us (->put_super runs with the write lock
1955 * hold).
1956 */
1957 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001958 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001959 drop_super(sb);
1960 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001961 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001962 return res;
1963}
1964EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001965
1966void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1967{
1968 struct inode *inode, *old_inode = NULL;
1969
Dave Chinner74278da2015-03-04 12:37:22 -05001970 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001971 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1972 struct address_space *mapping = inode->i_mapping;
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001973 struct block_device *bdev;
Jan Kara5c0d6b62012-07-03 16:45:31 +02001974
1975 spin_lock(&inode->i_lock);
1976 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1977 mapping->nrpages == 0) {
1978 spin_unlock(&inode->i_lock);
1979 continue;
1980 }
1981 __iget(inode);
1982 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05001983 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001984 /*
1985 * We hold a reference to 'inode' so it couldn't have been
1986 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05001987 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02001988 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05001989 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02001990 * later.
1991 */
1992 iput(old_inode);
1993 old_inode = inode;
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001994 bdev = I_BDEV(inode);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001995
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001996 mutex_lock(&bdev->bd_mutex);
1997 if (bdev->bd_openers)
1998 func(bdev, arg);
1999 mutex_unlock(&bdev->bd_mutex);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002000
Dave Chinner74278da2015-03-04 12:37:22 -05002001 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002002 }
Dave Chinner74278da2015-03-04 12:37:22 -05002003 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002004 iput(old_inode);
2005}