blob: 11c061b17e6eabb6a290997425a10cdcfc55ab28 [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
Jan Kara3385bb12017-03-02 16:50:13 +0100631 bdev->bd_bdi = &noop_backing_dev_info;
Christoph Lametera35afb82007-05-16 22:10:57 -0700632 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800633 /* Initialize mutex for freeze. */
634 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
Al Virob57922d2010-06-07 14:34:48 -0400637static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638{
639 struct block_device *bdev = &BDEV_I(inode)->bdev;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700640 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400641 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200642 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 spin_lock(&bdev_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 list_del_init(&bdev->bd_list);
645 spin_unlock(&bdev_lock);
Jan Kara90b65ff2017-03-23 01:37:00 +0100646 /* Detach inode from wb early as bdi_put() may free bdi->wb */
647 inode_detach_wb(inode);
Jan Kara3385bb12017-03-02 16:50:13 +0100648 if (bdev->bd_bdi != &noop_backing_dev_info) {
Jan Kara84972a92017-02-02 15:56:52 +0100649 bdi_put(bdev->bd_bdi);
Jan Kara3385bb12017-03-02 16:50:13 +0100650 bdev->bd_bdi = &noop_backing_dev_info;
651 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652}
653
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800654static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 .statfs = simple_statfs,
656 .alloc_inode = bdev_alloc_inode,
657 .destroy_inode = bdev_destroy_inode,
658 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400659 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660};
661
Al Viro51139ad2010-07-25 23:47:46 +0400662static struct dentry *bd_mount(struct file_system_type *fs_type,
663 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Shaohua Li3684aa72016-02-22 15:27:40 -0700665 struct dentry *dent;
666 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Vegard Nossume9e5e3f2016-08-22 12:47:43 +0200667 if (!IS_ERR(dent))
Shaohua Li3684aa72016-02-22 15:27:40 -0700668 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
669 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670}
671
672static struct file_system_type bd_type = {
673 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400674 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 .kill_sb = kill_anon_super,
676};
677
Tejun Heoa212b102015-05-22 17:13:33 -0400678struct super_block *blockdev_superblock __read_mostly;
679EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
681void __init bdev_cache_init(void)
682{
683 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300684 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800687 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800688 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900689 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 err = register_filesystem(&bd_type);
691 if (err)
692 panic("Cannot register bdev pseudo-fs");
693 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 if (IS_ERR(bd_mnt))
695 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300696 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697}
698
699/*
700 * Most likely _very_ bad one - but then it's hardly critical for small
701 * /dev and can be fixed when somebody will need really large one.
702 * Keep in mind that it will be fed through icache hash function too.
703 */
704static inline unsigned long hash(dev_t dev)
705{
706 return MAJOR(dev)+MINOR(dev);
707}
708
709static int bdev_test(struct inode *inode, void *data)
710{
711 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
712}
713
714static int bdev_set(struct inode *inode, void *data)
715{
716 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
717 return 0;
718}
719
720static LIST_HEAD(all_bdevs);
721
Jan Kara9507b392017-02-02 15:56:49 +0100722/*
723 * If there is a bdev inode for this device, unhash it so that it gets evicted
724 * as soon as last inode reference is dropped.
725 */
726void bdev_unhash_inode(dev_t dev)
727{
728 struct inode *inode;
729
730 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
731 if (inode) {
732 remove_inode_hash(inode);
733 iput(inode);
734 }
735}
736
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737struct block_device *bdget(dev_t dev)
738{
739 struct block_device *bdev;
740 struct inode *inode;
741
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800742 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 bdev_test, bdev_set, &dev);
744
745 if (!inode)
746 return NULL;
747
748 bdev = &BDEV_I(inode)->bdev;
749
750 if (inode->i_state & I_NEW) {
751 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000752 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 bdev->bd_inode = inode;
Fabian Frederick61604a22017-02-27 14:28:32 -0800754 bdev->bd_block_size = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 bdev->bd_part_count = 0;
756 bdev->bd_invalidated = 0;
757 inode->i_mode = S_IFBLK;
758 inode->i_rdev = dev;
759 inode->i_bdev = bdev;
760 inode->i_data.a_ops = &def_blk_aops;
761 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 spin_lock(&bdev_lock);
763 list_add(&bdev->bd_list, &all_bdevs);
764 spin_unlock(&bdev_lock);
765 unlock_new_inode(inode);
766 }
767 return bdev;
768}
769
770EXPORT_SYMBOL(bdget);
771
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200772/**
773 * bdgrab -- Grab a reference to an already referenced block device
774 * @bdev: Block device to grab a reference to.
775 */
776struct block_device *bdgrab(struct block_device *bdev)
777{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400778 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200779 return bdev;
780}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700781EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783long nr_blockdev_pages(void)
784{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700785 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 long ret = 0;
787 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700788 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 ret += bdev->bd_inode->i_mapping->nrpages;
790 }
791 spin_unlock(&bdev_lock);
792 return ret;
793}
794
795void bdput(struct block_device *bdev)
796{
797 iput(bdev->bd_inode);
798}
799
800EXPORT_SYMBOL(bdput);
801
802static struct block_device *bd_acquire(struct inode *inode)
803{
804 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 spin_lock(&bdev_lock);
807 bdev = inode->i_bdev;
Jan Karac1d5f2a2017-02-21 18:09:48 +0100808 if (bdev && !inode_unhashed(bdev->bd_inode)) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100809 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 spin_unlock(&bdev_lock);
811 return bdev;
812 }
813 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700814
Jan Karac1d5f2a2017-02-21 18:09:48 +0100815 /*
816 * i_bdev references block device inode that was already shut down
817 * (corresponding device got removed). Remove the reference and look
818 * up block device inode again just in case new device got
819 * reestablished under the same device number.
820 */
821 if (bdev)
822 bd_forget(inode);
823
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 bdev = bdget(inode->i_rdev);
825 if (bdev) {
826 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700827 if (!inode->i_bdev) {
828 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400829 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700830 * and it's released in clear_inode() of inode.
831 * So, we can access it via ->i_mapping always
832 * without igrab().
833 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100834 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700835 inode->i_bdev = bdev;
836 inode->i_mapping = bdev->bd_inode->i_mapping;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700837 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 spin_unlock(&bdev_lock);
839 }
840 return bdev;
841}
842
843/* Call when you free inode */
844
845void bd_forget(struct inode *inode)
846{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700847 struct block_device *bdev = NULL;
848
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700850 if (!sb_is_blkdev_sb(inode->i_sb))
851 bdev = inode->i_bdev;
Al Viroa4a4f942016-07-19 13:16:52 -0400852 inode->i_bdev = NULL;
853 inode->i_mapping = &inode->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700855
856 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100857 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858}
859
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900860/**
861 * bd_may_claim - test whether a block device can be claimed
862 * @bdev: block device of interest
863 * @whole: whole block device containing @bdev, may equal @bdev
864 * @holder: holder trying to claim @bdev
865 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300866 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900867 *
868 * CONTEXT:
869 * spin_lock(&bdev_lock).
870 *
871 * RETURNS:
872 * %true if @bdev can be claimed, %false otherwise.
873 */
874static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
875 void *holder)
876{
877 if (bdev->bd_holder == holder)
878 return true; /* already a holder */
879 else if (bdev->bd_holder != NULL)
880 return false; /* held by someone else */
NeilBrowncfa2d652016-12-12 08:21:51 -0700881 else if (whole == bdev)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900882 return true; /* is a whole device which isn't held */
883
Tejun Heoe525fd82010-11-13 11:55:17 +0100884 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900885 return true; /* is a partition of a device that is being partitioned */
886 else if (whole->bd_holder != NULL)
887 return false; /* is a partition of a held device */
888 else
889 return true; /* is a partition of an un-held device */
890}
891
892/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900893 * bd_prepare_to_claim - prepare to claim a block device
894 * @bdev: block device of interest
895 * @whole: the whole device containing @bdev, may equal @bdev
896 * @holder: holder trying to claim @bdev
897 *
898 * Prepare to claim @bdev. This function fails if @bdev is already
899 * claimed by another holder and waits if another claiming is in
900 * progress. This function doesn't actually claim. On successful
901 * return, the caller has ownership of bd_claiming and bd_holder[s].
902 *
903 * CONTEXT:
904 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
905 * it multiple times.
906 *
907 * RETURNS:
908 * 0 if @bdev can be claimed, -EBUSY otherwise.
909 */
910static int bd_prepare_to_claim(struct block_device *bdev,
911 struct block_device *whole, void *holder)
912{
913retry:
914 /* if someone else claimed, fail */
915 if (!bd_may_claim(bdev, whole, holder))
916 return -EBUSY;
917
Tejun Heoe75aa852010-08-04 17:59:39 +0200918 /* if claiming is already in progress, wait for it to finish */
919 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900920 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
921 DEFINE_WAIT(wait);
922
923 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
924 spin_unlock(&bdev_lock);
925 schedule();
926 finish_wait(wq, &wait);
927 spin_lock(&bdev_lock);
928 goto retry;
929 }
930
931 /* yay, all mine */
932 return 0;
933}
934
935/**
936 * bd_start_claiming - start claiming a block device
937 * @bdev: block device of interest
938 * @holder: holder trying to claim @bdev
939 *
940 * @bdev is about to be opened exclusively. Check @bdev can be opened
941 * exclusively and mark that an exclusive open is in progress. Each
942 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000943 * either bd_finish_claiming() or bd_abort_claiming() (which do not
944 * fail).
945 *
946 * This function is used to gain exclusive access to the block device
947 * without actually causing other exclusive open attempts to fail. It
948 * should be used when the open sequence itself requires exclusive
949 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900950 *
951 * CONTEXT:
952 * Might sleep.
953 *
954 * RETURNS:
955 * Pointer to the block device containing @bdev on success, ERR_PTR()
956 * value on failure.
957 */
958static struct block_device *bd_start_claiming(struct block_device *bdev,
959 void *holder)
960{
961 struct gendisk *disk;
962 struct block_device *whole;
963 int partno, err;
964
965 might_sleep();
966
967 /*
968 * @bdev might not have been initialized properly yet, look up
969 * and grab the outer block device the hard way.
970 */
971 disk = get_gendisk(bdev->bd_dev, &partno);
972 if (!disk)
973 return ERR_PTR(-ENXIO);
974
Tejun Heod4c208b2011-06-13 12:45:48 +0200975 /*
976 * Normally, @bdev should equal what's returned from bdget_disk()
977 * if partno is 0; however, some drivers (floppy) use multiple
978 * bdev's for the same physical device and @bdev may be one of the
979 * aliases. Keep @bdev if partno is 0. This means claimer
980 * tracking is broken for those devices but it has always been that
981 * way.
982 */
983 if (partno)
984 whole = bdget_disk(disk, 0);
985 else
986 whole = bdgrab(bdev);
987
Nick Piggincf342572010-05-26 01:50:21 +1000988 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900989 put_disk(disk);
990 if (!whole)
991 return ERR_PTR(-ENOMEM);
992
993 /* prepare to claim, if successful, mark claiming in progress */
994 spin_lock(&bdev_lock);
995
996 err = bd_prepare_to_claim(bdev, whole, holder);
997 if (err == 0) {
998 whole->bd_claiming = holder;
999 spin_unlock(&bdev_lock);
1000 return whole;
1001 } else {
1002 spin_unlock(&bdev_lock);
1003 bdput(whole);
1004 return ERR_PTR(err);
1005 }
1006}
1007
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001008#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +01001009struct bd_holder_disk {
1010 struct list_head list;
1011 struct gendisk *disk;
1012 int refcnt;
1013};
1014
1015static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1016 struct gendisk *disk)
1017{
1018 struct bd_holder_disk *holder;
1019
1020 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1021 if (holder->disk == disk)
1022 return holder;
1023 return NULL;
1024}
1025
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001026static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001027{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001028 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001029}
1030
1031static void del_symlink(struct kobject *from, struct kobject *to)
1032{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001033 sysfs_remove_link(from, kobject_name(to));
1034}
1035
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001036/**
Tejun Heoe09b4572010-11-13 11:55:17 +01001037 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1038 * @bdev: the claimed slave bdev
1039 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001040 *
Tejun Heo49731ba2011-01-14 18:43:57 +01001041 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1042 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001043 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001044 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001045 * - from "slaves" directory of the holder @disk to the claimed @bdev
1046 * - from "holders" directory of the @bdev to the holder @disk
1047 *
1048 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1049 * passed to bd_link_disk_holder(), then:
1050 *
1051 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1052 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1053 *
1054 * The caller must have claimed @bdev before calling this function and
1055 * ensure that both @bdev and @disk are valid during the creation and
1056 * lifetime of these symlinks.
1057 *
1058 * CONTEXT:
1059 * Might sleep.
1060 *
1061 * RETURNS:
1062 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001063 */
Tejun Heoe09b4572010-11-13 11:55:17 +01001064int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001065{
Tejun Heo49731ba2011-01-14 18:43:57 +01001066 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001067 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001068
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001069 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001070
Tejun Heo49731ba2011-01-14 18:43:57 +01001071 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -07001072
Tejun Heoe09b4572010-11-13 11:55:17 +01001073 /* FIXME: remove the following once add_disk() handles errors */
1074 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1075 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -07001076
Tejun Heo49731ba2011-01-14 18:43:57 +01001077 holder = bd_find_holder_disk(bdev, disk);
1078 if (holder) {
1079 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +01001080 goto out_unlock;
1081 }
1082
Tejun Heo49731ba2011-01-14 18:43:57 +01001083 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1084 if (!holder) {
1085 ret = -ENOMEM;
1086 goto out_unlock;
1087 }
1088
1089 INIT_LIST_HEAD(&holder->list);
1090 holder->disk = disk;
1091 holder->refcnt = 1;
1092
1093 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1094 if (ret)
1095 goto out_free;
1096
1097 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1098 if (ret)
1099 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +01001100 /*
1101 * bdev could be deleted beneath us which would implicitly destroy
1102 * the holder directory. Hold on to it.
1103 */
1104 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001105
1106 list_add(&holder->list, &bdev->bd_holder_disks);
1107 goto out_unlock;
1108
1109out_del:
1110 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1111out_free:
1112 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001113out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001114 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001115 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001116}
Tejun Heoe09b4572010-11-13 11:55:17 +01001117EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001118
Tejun Heo49731ba2011-01-14 18:43:57 +01001119/**
1120 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1121 * @bdev: the calimed slave bdev
1122 * @disk: the holding disk
1123 *
1124 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1125 *
1126 * CONTEXT:
1127 * Might sleep.
1128 */
1129void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001130{
Tejun Heo49731ba2011-01-14 18:43:57 +01001131 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001132
Tejun Heo49731ba2011-01-14 18:43:57 +01001133 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001134
Tejun Heo49731ba2011-01-14 18:43:57 +01001135 holder = bd_find_holder_disk(bdev, disk);
1136
1137 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1138 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1139 del_symlink(bdev->bd_part->holder_dir,
1140 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001141 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001142 list_del_init(&holder->list);
1143 kfree(holder);
1144 }
1145
1146 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001147}
Tejun Heo49731ba2011-01-14 18:43:57 +01001148EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001149#endif
1150
Andrew Patterson0c002c22008-09-04 14:27:20 -06001151/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001152 * flush_disk - invalidates all buffer-cache entries on a disk
1153 *
1154 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001155 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001156 *
1157 * Invalidates all buffer-cache entries on a disk. It should be called
1158 * when a disk has been changed -- either by a media change or online
1159 * resize.
1160 */
NeilBrown93b270f2011-02-24 17:25:47 +11001161static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001162{
NeilBrown93b270f2011-02-24 17:25:47 +11001163 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001164 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001165 "resized disk %s\n",
1166 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001167 }
1168
1169 if (!bdev->bd_disk)
1170 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001171 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001172 bdev->bd_invalidated = 1;
1173}
1174
1175/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001176 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001177 * @disk: struct gendisk to check
1178 * @bdev: struct bdev to adjust.
1179 *
1180 * This routine checks to see if the bdev size does not match the disk size
1181 * and adjusts it if it differs.
1182 */
1183void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1184{
1185 loff_t disk_size, bdev_size;
1186
1187 disk_size = (loff_t)get_capacity(disk) << 9;
1188 bdev_size = i_size_read(bdev->bd_inode);
1189 if (disk_size != bdev_size) {
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001190 printk(KERN_INFO
1191 "%s: detected capacity change from %lld to %lld\n",
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001192 disk->disk_name, bdev_size, disk_size);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001193 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001194 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001195 }
1196}
1197EXPORT_SYMBOL(check_disk_size_change);
1198
1199/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001200 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001201 * @disk: struct gendisk to be revalidated
1202 *
1203 * This routine is a wrapper for lower-level driver's revalidate_disk
1204 * call-backs. It is used to do common pre and post operations needed
1205 * for all revalidate_disk operations.
1206 */
1207int revalidate_disk(struct gendisk *disk)
1208{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001209 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001210 int ret = 0;
1211
1212 if (disk->fops->revalidate_disk)
1213 ret = disk->fops->revalidate_disk(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001214 bdev = bdget_disk(disk, 0);
1215 if (!bdev)
1216 return ret;
1217
1218 mutex_lock(&bdev->bd_mutex);
1219 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001220 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001221 mutex_unlock(&bdev->bd_mutex);
1222 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001223 return ret;
1224}
1225EXPORT_SYMBOL(revalidate_disk);
1226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227/*
1228 * This routine checks whether a removable media has been changed,
1229 * and invalidates all buffer-cache-entries in that case. This
1230 * is a relatively slow routine, so we have to try to minimize using
1231 * it. Thus it is called only upon a 'mount' or 'open'. This
1232 * is the best way of combining speed and utility, I think.
1233 * People changing diskettes in the middle of an operation deserve
1234 * to lose :-)
1235 */
1236int check_disk_change(struct block_device *bdev)
1237{
1238 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001239 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001240 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Tejun Heo77ea8872010-12-08 20:57:37 +01001242 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1243 DISK_EVENT_EJECT_REQUEST);
1244 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 return 0;
1246
NeilBrown93b270f2011-02-24 17:25:47 +11001247 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 if (bdops->revalidate_disk)
1249 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 return 1;
1251}
1252
1253EXPORT_SYMBOL(check_disk_change);
1254
1255void bd_set_size(struct block_device *bdev, loff_t size)
1256{
Al Viro59551022016-01-22 15:40:57 -05001257 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001258 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001259 inode_unlock(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
1261EXPORT_SYMBOL(bd_set_size);
1262
Al Viro4385bab2013-05-05 22:11:03 -04001263static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001264
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001265/*
1266 * bd_mutex locking:
1267 *
1268 * mutex_lock(part->bd_mutex)
1269 * mutex_lock_nested(whole->bd_mutex, 1)
1270 */
1271
Al Viro572c4892007-10-08 13:24:05 -04001272static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001275 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001276 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001277 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001278 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
Al Viro572c4892007-10-08 13:24:05 -04001280 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001281 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001282 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001283 perm |= MAY_WRITE;
1284 /*
1285 * hooks: /n/, see "layering violations".
1286 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001287 if (!for_part) {
1288 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1289 if (ret != 0) {
1290 bdput(bdev);
1291 return ret;
1292 }
Al Viro82666022008-08-01 05:32:04 -04001293 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001294
NeilBrownd3374822009-01-09 08:31:10 +11001295 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001296
Tejun Heo89f97492008-11-05 10:21:06 +01001297 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001298 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001299 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001300 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001301 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Tejun Heo69e02c52011-03-09 19:54:27 +01001303 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001304 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (!bdev->bd_openers) {
1306 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001307 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 bdev->bd_contains = bdev;
Dan Williams03cdadb2016-02-26 15:19:43 -08001309
Tejun Heocf771cb2008-09-03 09:01:09 +02001310 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001311 ret = -ENXIO;
1312 bdev->bd_part = disk_get_part(disk, partno);
1313 if (!bdev->bd_part)
1314 goto out_clear;
1315
Tejun Heo1196f8b2011-04-21 20:54:45 +02001316 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001318 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001319 if (ret == -ERESTARTSYS) {
1320 /* Lost a race with 'disk' being
1321 * deleted, try again.
1322 * See md.c
1323 */
1324 disk_put_part(bdev->bd_part);
1325 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001326 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001327 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001328 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001329 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001330 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001331 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001332 goto restart;
1333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 }
Tejun Heo7e697232011-05-23 13:26:07 +02001335
Jan Kara428ca582019-01-14 09:48:10 +01001336 if (!ret) {
Tejun Heo7e697232011-05-23 13:26:07 +02001337 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Jan Kara428ca582019-01-14 09:48:10 +01001338 set_init_blocksize(bdev);
1339 }
Tejun Heo7e697232011-05-23 13:26:07 +02001340
Tejun Heo1196f8b2011-04-21 20:54:45 +02001341 /*
1342 * If the device is invalidated, rescan partition
1343 * if open succeeded or failed with -ENOMEDIUM.
1344 * The latter is necessary to prevent ghost
1345 * partitions on a removed medium.
1346 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001347 if (bdev->bd_invalidated) {
1348 if (!ret)
1349 rescan_partitions(disk, bdev);
1350 else if (ret == -ENOMEDIUM)
1351 invalidate_partitions(disk, bdev);
1352 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001353
Tejun Heo1196f8b2011-04-21 20:54:45 +02001354 if (ret)
1355 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 struct block_device *whole;
1358 whole = bdget_disk(disk, 0);
1359 ret = -ENOMEM;
1360 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001361 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001362 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001363 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001365 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001367 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001368 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001369 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001371 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 }
Tejun Heo89f97492008-11-05 10:21:06 +01001373 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Jan Kara428ca582019-01-14 09:48:10 +01001374 set_init_blocksize(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 }
Jan Kara9179eba2017-03-23 01:36:53 +01001376
1377 if (bdev->bd_bdi == &noop_backing_dev_info)
1378 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001381 ret = 0;
1382 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001383 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001384 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001385 if (bdev->bd_invalidated) {
1386 if (!ret)
1387 rescan_partitions(bdev->bd_disk, bdev);
1388 else if (ret == -ENOMEDIUM)
1389 invalidate_partitions(bdev->bd_disk, bdev);
1390 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001391 if (ret)
1392 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001394 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001395 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001396 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 }
1398 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001399 if (for_part)
1400 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001401 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001402 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 return 0;
1404
Tejun Heo0762b8b2008-08-25 19:56:12 +09001405 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001406 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001408 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001409 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001411 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001413 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001414 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001415 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001416 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001417 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001418 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001419 bdput(bdev);
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 return ret;
1422}
1423
Tejun Heod4d77622010-11-13 11:55:18 +01001424/**
1425 * blkdev_get - open a block device
1426 * @bdev: block_device to open
1427 * @mode: FMODE_* mask
1428 * @holder: exclusive holder identifier
1429 *
1430 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1431 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1432 * @holder is invalid. Exclusive opens may nest for the same @holder.
1433 *
1434 * On success, the reference count of @bdev is unchanged. On failure,
1435 * @bdev is put.
1436 *
1437 * CONTEXT:
1438 * Might sleep.
1439 *
1440 * RETURNS:
1441 * 0 on success, -errno on failure.
1442 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001443int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
Tejun Heoe525fd82010-11-13 11:55:17 +01001445 struct block_device *whole = NULL;
1446 int res;
1447
1448 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1449
1450 if ((mode & FMODE_EXCL) && holder) {
1451 whole = bd_start_claiming(bdev, holder);
1452 if (IS_ERR(whole)) {
1453 bdput(bdev);
1454 return PTR_ERR(whole);
1455 }
1456 }
1457
1458 res = __blkdev_get(bdev, mode, 0);
1459
1460 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001461 struct gendisk *disk = whole->bd_disk;
1462
Tejun Heo6a027ef2010-11-13 11:55:17 +01001463 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001464 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001465 spin_lock(&bdev_lock);
1466
Tejun Heo77ea8872010-12-08 20:57:37 +01001467 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001468 BUG_ON(!bd_may_claim(bdev, whole, holder));
1469 /*
1470 * Note that for a whole device bd_holders
1471 * will be incremented twice, and bd_holder
1472 * will be set to bd_may_claim before being
1473 * set to holder
1474 */
1475 whole->bd_holders++;
1476 whole->bd_holder = bd_may_claim;
1477 bdev->bd_holders++;
1478 bdev->bd_holder = holder;
1479 }
1480
1481 /* tell others that we're done */
1482 BUG_ON(whole->bd_claiming != holder);
1483 whole->bd_claiming = NULL;
1484 wake_up_bit(&whole->bd_claiming, 0);
1485
1486 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001487
1488 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001489 * Block event polling for write claims if requested. Any
1490 * write holder makes the write_holder state stick until
1491 * all are released. This is good enough and tracking
1492 * individual writeable reference is too fragile given the
1493 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001494 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001495 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1496 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001497 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001498 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001499 }
1500
1501 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001502 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001503 }
1504
1505 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001506}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507EXPORT_SYMBOL(blkdev_get);
1508
Tejun Heod4d77622010-11-13 11:55:18 +01001509/**
1510 * blkdev_get_by_path - open a block device by name
1511 * @path: path to the block device to open
1512 * @mode: FMODE_* mask
1513 * @holder: exclusive holder identifier
1514 *
1515 * Open the blockdevice described by the device file at @path. @mode
1516 * and @holder are identical to blkdev_get().
1517 *
1518 * On success, the returned block_device has reference count of one.
1519 *
1520 * CONTEXT:
1521 * Might sleep.
1522 *
1523 * RETURNS:
1524 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1525 */
1526struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1527 void *holder)
1528{
1529 struct block_device *bdev;
1530 int err;
1531
1532 bdev = lookup_bdev(path);
1533 if (IS_ERR(bdev))
1534 return bdev;
1535
1536 err = blkdev_get(bdev, mode, holder);
1537 if (err)
1538 return ERR_PTR(err);
1539
Chuck Ebberte51900f2011-02-16 18:11:53 -05001540 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1541 blkdev_put(bdev, mode);
1542 return ERR_PTR(-EACCES);
1543 }
1544
Tejun Heod4d77622010-11-13 11:55:18 +01001545 return bdev;
1546}
1547EXPORT_SYMBOL(blkdev_get_by_path);
1548
1549/**
1550 * blkdev_get_by_dev - open a block device by device number
1551 * @dev: device number of block device to open
1552 * @mode: FMODE_* mask
1553 * @holder: exclusive holder identifier
1554 *
1555 * Open the blockdevice described by device number @dev. @mode and
1556 * @holder are identical to blkdev_get().
1557 *
1558 * Use it ONLY if you really do not have anything better - i.e. when
1559 * you are behind a truly sucky interface and all you are given is a
1560 * device number. _Never_ to be used for internal purposes. If you
1561 * ever need it - reconsider your API.
1562 *
1563 * On success, the returned block_device has reference count of one.
1564 *
1565 * CONTEXT:
1566 * Might sleep.
1567 *
1568 * RETURNS:
1569 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1570 */
1571struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1572{
1573 struct block_device *bdev;
1574 int err;
1575
1576 bdev = bdget(dev);
1577 if (!bdev)
1578 return ERR_PTR(-ENOMEM);
1579
1580 err = blkdev_get(bdev, mode, holder);
1581 if (err)
1582 return ERR_PTR(err);
1583
1584 return bdev;
1585}
1586EXPORT_SYMBOL(blkdev_get_by_dev);
1587
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588static int blkdev_open(struct inode * inode, struct file * filp)
1589{
1590 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
1592 /*
1593 * Preserve backwards compatibility and allow large file access
1594 * even if userspace doesn't ask for it explicitly. Some mkfs
1595 * binary needs it. We might want to drop this workaround
1596 * during an unstable branch.
1597 */
1598 filp->f_flags |= O_LARGEFILE;
1599
Al Viro572c4892007-10-08 13:24:05 -04001600 if (filp->f_flags & O_NDELAY)
1601 filp->f_mode |= FMODE_NDELAY;
1602 if (filp->f_flags & O_EXCL)
1603 filp->f_mode |= FMODE_EXCL;
1604 if ((filp->f_flags & O_ACCMODE) == 3)
1605 filp->f_mode |= FMODE_WRITE_IOCTL;
1606
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001608 if (bdev == NULL)
1609 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
Al Viro572c4892007-10-08 13:24:05 -04001611 filp->f_mapping = bdev->bd_inode->i_mapping;
1612
Tejun Heoe525fd82010-11-13 11:55:17 +01001613 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614}
1615
Al Viro4385bab2013-05-05 22:11:03 -04001616static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001617{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001618 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001619 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001620
NeilBrown6796bf52006-12-08 02:36:16 -08001621 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001622 if (for_part)
1623 bdev->bd_part_count--;
1624
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001625 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001626 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001627 sync_blockdev(bdev);
1628 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001629
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001630 bdev_write_inode(bdev);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001631 }
1632 if (bdev->bd_contains == bdev) {
1633 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001634 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001635 }
1636 if (!bdev->bd_openers) {
1637 struct module *owner = disk->fops->owner;
1638
Tejun Heo0762b8b2008-08-25 19:56:12 +09001639 disk_put_part(bdev->bd_part);
1640 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001641 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001642 if (bdev != bdev->bd_contains)
1643 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001644 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001645
1646 put_disk(disk);
1647 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001648 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001649 mutex_unlock(&bdev->bd_mutex);
1650 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001651 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001652 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001653}
1654
Al Viro4385bab2013-05-05 22:11:03 -04001655void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001656{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001657 mutex_lock(&bdev->bd_mutex);
1658
Tejun Heoe525fd82010-11-13 11:55:17 +01001659 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001660 bool bdev_free;
1661
1662 /*
1663 * Release a claim on the device. The holder fields
1664 * are protected with bdev_lock. bd_mutex is to
1665 * synchronize disk_holder unlinking.
1666 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001667 spin_lock(&bdev_lock);
1668
1669 WARN_ON_ONCE(--bdev->bd_holders < 0);
1670 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1671
1672 /* bd_contains might point to self, check in a separate step */
1673 if ((bdev_free = !bdev->bd_holders))
1674 bdev->bd_holder = NULL;
1675 if (!bdev->bd_contains->bd_holders)
1676 bdev->bd_contains->bd_holder = NULL;
1677
1678 spin_unlock(&bdev_lock);
1679
Tejun Heo77ea8872010-12-08 20:57:37 +01001680 /*
1681 * If this was the last claim, remove holder link and
1682 * unblock evpoll if it was a write holder.
1683 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001684 if (bdev_free && bdev->bd_write_holder) {
1685 disk_unblock_events(bdev->bd_disk);
1686 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001687 }
Tejun Heo69362172011-03-09 19:54:27 +01001688 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001689
Tejun Heo85ef06d2011-07-01 16:17:47 +02001690 /*
1691 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1692 * event. This is to ensure detection of media removal commanded
1693 * from userland - e.g. eject(1).
1694 */
1695 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1696
1697 mutex_unlock(&bdev->bd_mutex);
1698
Al Viro4385bab2013-05-05 22:11:03 -04001699 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001700}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001701EXPORT_SYMBOL(blkdev_put);
1702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703static int blkdev_close(struct inode * inode, struct file * filp)
1704{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001705 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001706 blkdev_put(bdev, filp->f_mode);
1707 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708}
1709
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001710static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001712 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001713 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001714
1715 /*
1716 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1717 * to updated it before every ioctl.
1718 */
Al Viro56b26ad2008-09-19 03:17:36 -04001719 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001720 mode |= FMODE_NDELAY;
1721 else
1722 mode &= ~FMODE_NDELAY;
1723
Al Viro56b26ad2008-09-19 03:17:36 -04001724 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725}
1726
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001727/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001728 * Write data to the block device. Only intended for the block device itself
1729 * and the raw driver which basically is a fake block device.
1730 *
1731 * Does not take i_mutex for the write and thus is not for general purpose
1732 * use.
1733 */
Al Viro1456c0a2014-04-03 03:21:50 -04001734ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001735{
1736 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001737 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001738 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001739 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001740 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001741
Al Viro7ec7b942015-04-07 11:35:14 -04001742 if (bdev_read_only(I_BDEV(bd_inode)))
1743 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001744
Al Viro7ec7b942015-04-07 11:35:14 -04001745 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001746 return 0;
1747
Al Viro7ec7b942015-04-07 11:35:14 -04001748 if (iocb->ki_pos >= size)
1749 return -ENOSPC;
1750
1751 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001752
Jianpeng Ma53362a02012-08-02 09:50:39 +02001753 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001754 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwige2592212016-04-07 08:52:01 -07001755 if (ret > 0)
1756 ret = generic_write_sync(iocb, ret);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001757 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001758 return ret;
1759}
Al Viro1456c0a2014-04-03 03:21:50 -04001760EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001761
David Jefferyb2de5252014-09-29 10:21:10 -04001762ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001763{
1764 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001765 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001766 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001767 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001768
1769 if (pos >= size)
1770 return 0;
1771
1772 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001773 iov_iter_truncate(to, size);
1774 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001775}
David Jefferyb2de5252014-09-29 10:21:10 -04001776EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001777
Christoph Hellwigeef99382009-08-20 17:43:41 +02001778/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001779 * Try to release a page associated with block device when the system
1780 * is under memory pressure.
1781 */
1782static int blkdev_releasepage(struct page *page, gfp_t wait)
1783{
1784 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1785
1786 if (super && super->s_op->bdev_try_to_free_page)
1787 return super->s_op->bdev_try_to_free_page(super, page, wait);
1788
1789 return try_to_free_buffers(page);
1790}
1791
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001792static int blkdev_writepages(struct address_space *mapping,
1793 struct writeback_control *wbc)
1794{
1795 if (dax_mapping(mapping)) {
1796 struct block_device *bdev = I_BDEV(mapping->host);
1797
1798 return dax_writeback_mapping_range(mapping, bdev, wbc);
1799 }
1800 return generic_writepages(mapping, wbc);
1801}
1802
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001803static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001805 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001807 .write_begin = blkdev_write_begin,
1808 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001809 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001810 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001812 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813};
1814
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001815#define BLKDEV_FALLOC_FL_SUPPORTED \
1816 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1817 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1818
1819static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1820 loff_t len)
1821{
1822 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1823 struct request_queue *q = bdev_get_queue(bdev);
1824 struct address_space *mapping;
1825 loff_t end = start + len - 1;
1826 loff_t isize;
1827 int error;
1828
1829 /* Fail if we don't recognize the flags. */
1830 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1831 return -EOPNOTSUPP;
1832
1833 /* Don't go off the end of the device. */
1834 isize = i_size_read(bdev->bd_inode);
1835 if (start >= isize)
1836 return -EINVAL;
1837 if (end >= isize) {
1838 if (mode & FALLOC_FL_KEEP_SIZE) {
1839 len = isize - start;
1840 end = start + len - 1;
1841 } else
1842 return -EINVAL;
1843 }
1844
1845 /*
1846 * Don't allow IO that isn't aligned to logical block size.
1847 */
1848 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1849 return -EINVAL;
1850
1851 /* Invalidate the page cache, including dirty pages. */
1852 mapping = bdev->bd_inode->i_mapping;
1853 truncate_inode_pages_range(mapping, start, end);
1854
1855 switch (mode) {
1856 case FALLOC_FL_ZERO_RANGE:
1857 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1858 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1859 GFP_KERNEL, false);
1860 break;
1861 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1862 /* Only punch if the device can do zeroing discard. */
1863 if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data)
1864 return -EOPNOTSUPP;
1865 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1866 GFP_KERNEL, 0);
1867 break;
1868 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1869 if (!blk_queue_discard(q))
1870 return -EOPNOTSUPP;
1871 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1872 GFP_KERNEL, 0);
1873 break;
1874 default:
1875 return -EOPNOTSUPP;
1876 }
1877 if (error)
1878 return error;
1879
1880 /*
1881 * Invalidate again; if someone wandered in and dirtied a page,
1882 * the caller will be given -EBUSY. The third argument is
1883 * inclusive, so the rounding here is safe.
1884 */
1885 return invalidate_inode_pages2_range(mapping,
1886 start >> PAGE_SHIFT,
1887 end >> PAGE_SHIFT);
1888}
1889
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001890const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 .open = blkdev_open,
1892 .release = blkdev_close,
1893 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001894 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001895 .write_iter = blkdev_write_iter,
Dan Williamsacc93d32016-05-07 11:40:28 -07001896 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001897 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001898 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899#ifdef CONFIG_COMPAT
1900 .compat_ioctl = compat_blkdev_ioctl,
1901#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001902 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001903 .splice_write = iter_file_splice_write,
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001904 .fallocate = blkdev_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905};
1906
1907int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1908{
1909 int res;
1910 mm_segment_t old_fs = get_fs();
1911 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001912 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 set_fs(old_fs);
1914 return res;
1915}
1916
1917EXPORT_SYMBOL(ioctl_by_bdev);
1918
1919/**
1920 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001921 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001923 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 * namespace if possible and return it. Return ERR_PTR(error)
1925 * otherwise.
1926 */
Al Viro421748e2008-08-02 01:04:36 -04001927struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928{
1929 struct block_device *bdev;
1930 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001931 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 int error;
1933
Al Viro421748e2008-08-02 01:04:36 -04001934 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 return ERR_PTR(-EINVAL);
1936
Al Viro421748e2008-08-02 01:04:36 -04001937 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 if (error)
1939 return ERR_PTR(error);
1940
David Howellsbb6687342015-03-17 22:26:21 +00001941 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 error = -ENOTBLK;
1943 if (!S_ISBLK(inode->i_mode))
1944 goto fail;
1945 error = -EACCES;
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05001946 if (!may_open_dev(&path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 goto fail;
1948 error = -ENOMEM;
1949 bdev = bd_acquire(inode);
1950 if (!bdev)
1951 goto fail;
1952out:
Al Viro421748e2008-08-02 01:04:36 -04001953 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return bdev;
1955fail:
1956 bdev = ERR_PTR(error);
1957 goto out;
1958}
Al Virod5686b42008-08-01 05:00:11 -04001959EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
NeilBrown93b270f2011-02-24 17:25:47 +11001961int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001962{
1963 struct super_block *sb = get_super(bdev);
1964 int res = 0;
1965
1966 if (sb) {
1967 /*
1968 * no need to lock the super, get_super holds the
1969 * read mutex so the filesystem cannot go away
1970 * under us (->put_super runs with the write lock
1971 * hold).
1972 */
1973 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001974 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001975 drop_super(sb);
1976 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001977 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001978 return res;
1979}
1980EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001981
1982void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1983{
1984 struct inode *inode, *old_inode = NULL;
1985
Dave Chinner74278da2015-03-04 12:37:22 -05001986 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001987 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1988 struct address_space *mapping = inode->i_mapping;
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001989 struct block_device *bdev;
Jan Kara5c0d6b62012-07-03 16:45:31 +02001990
1991 spin_lock(&inode->i_lock);
1992 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1993 mapping->nrpages == 0) {
1994 spin_unlock(&inode->i_lock);
1995 continue;
1996 }
1997 __iget(inode);
1998 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05001999 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002000 /*
2001 * We hold a reference to 'inode' so it couldn't have been
2002 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05002003 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02002004 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05002005 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02002006 * later.
2007 */
2008 iput(old_inode);
2009 old_inode = inode;
Rabin Vincent11aa5c12016-12-01 09:18:28 +01002010 bdev = I_BDEV(inode);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002011
Rabin Vincent11aa5c12016-12-01 09:18:28 +01002012 mutex_lock(&bdev->bd_mutex);
2013 if (bdev->bd_openers)
2014 func(bdev, arg);
2015 mutex_unlock(&bdev->bd_mutex);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002016
Dave Chinner74278da2015-03-04 12:37:22 -05002017 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002018 }
Dave Chinner74278da2015-03-04 12:37:22 -05002019 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002020 iput(old_inode);
2021}