blob: 07e46b786500b5e3a5974fa2d468b5ea32131ee3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fcntl.h>
11#include <linux/slab.h>
12#include <linux/kmod.h>
13#include <linux/major.h>
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -070014#include <linux/device_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/highmem.h>
16#include <linux/blkdev.h>
Tejun Heo66114ca2015-05-22 17:13:32 -040017#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/module.h>
19#include <linux/blkpg.h>
Muthu Kumarb502bd12012-03-23 15:01:50 -070020#include <linux/magic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/buffer_head.h>
Al Viroff01bb482011-09-16 02:31:11 -040022#include <linux/swap.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010023#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010024#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/mpage.h>
26#include <linux/mount.h>
27#include <linux/uio.h>
28#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070029#include <linux/log2.h>
Al Viroff01bb482011-09-16 02:31:11 -040030#include <linux/cleancache.h>
Matthew Wilcoxc94c2ac2015-09-08 14:58:40 -070031#include <linux/dax.h>
Dan Williamsacc93d32016-05-07 11:40:28 -070032#include <linux/badblocks.h>
Darrick J. Wong25f4c412016-10-11 13:51:11 -070033#include <linux/falloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020035#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37struct bdev_inode {
38 struct block_device bdev;
39 struct inode vfs_inode;
40};
41
Adrian Bunk4c54ac62008-02-18 13:48:31 +010042static const struct address_space_operations def_blk_aops;
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044static inline struct bdev_inode *BDEV_I(struct inode *inode)
45{
46 return container_of(inode, struct bdev_inode, vfs_inode);
47}
48
Geert Uytterhoevenff5053f2015-06-26 13:58:32 +020049struct block_device *I_BDEV(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070050{
51 return &BDEV_I(inode)->bdev;
52}
Linus Torvalds1da177e2005-04-16 15:20:36 -070053EXPORT_SYMBOL(I_BDEV);
54
Toshi Kani2af3a812016-05-10 10:23:52 -060055void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
56{
57 struct va_format vaf;
58 va_list args;
59
60 va_start(args, fmt);
61 vaf.fmt = fmt;
62 vaf.va = &args;
63 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
64 va_end(args);
65}
66
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070067static void bdev_write_inode(struct block_device *bdev)
Christoph Hellwig564f00f2015-01-14 10:42:33 +010068{
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070069 struct inode *inode = bdev->bd_inode;
70 int ret;
71
Christoph Hellwig564f00f2015-01-14 10:42:33 +010072 spin_lock(&inode->i_lock);
73 while (inode->i_state & I_DIRTY) {
74 spin_unlock(&inode->i_lock);
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070075 ret = write_inode_now(inode, true);
76 if (ret) {
77 char name[BDEVNAME_SIZE];
78 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
79 "for block device %s (err=%d).\n",
80 bdevname(bdev, name), ret);
81 }
Christoph Hellwig564f00f2015-01-14 10:42:33 +010082 spin_lock(&inode->i_lock);
83 }
84 spin_unlock(&inode->i_lock);
85}
86
Peter Zijlstraf9a14392007-05-06 14:49:55 -070087/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb482011-09-16 02:31:11 -040088void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089{
Al Viroff01bb482011-09-16 02:31:11 -040090 struct address_space *mapping = bdev->bd_inode->i_mapping;
91
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080092 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070093 return;
Al Viroff01bb482011-09-16 02:31:11 -040094
Peter Zijlstraf9a14392007-05-06 14:49:55 -070095 invalidate_bh_lrus();
Al Viroff01bb482011-09-16 02:31:11 -040096 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097}
Al Viroff01bb482011-09-16 02:31:11 -040098EXPORT_SYMBOL(kill_bdev);
99
100/* Invalidate clean unused buffers and pagecache. */
101void invalidate_bdev(struct block_device *bdev)
102{
103 struct address_space *mapping = bdev->bd_inode->i_mapping;
104
Andrey Ryabinin945d0ec2017-05-03 14:56:02 -0700105 if (mapping->nrpages) {
106 invalidate_bh_lrus();
107 lru_add_drain_all(); /* make sure all lru add caches are flushed */
108 invalidate_mapping_pages(mapping, 0, -1);
109 }
Al Viroff01bb482011-09-16 02:31:11 -0400110 /* 99% of the time, we don't need to flush the cleancache on the bdev.
111 * But, for the strange corners, lets be cautious
112 */
Dan Magenheimer31677602011-09-21 11:56:28 -0400113 cleancache_invalidate_inode(mapping);
Al Viroff01bb482011-09-16 02:31:11 -0400114}
115EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
117int set_blocksize(struct block_device *bdev, int size)
118{
119 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700120 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 return -EINVAL;
122
123 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400124 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 return -EINVAL;
126
127 /* Don't change the size if it is same as current */
128 if (bdev->bd_block_size != size) {
129 sync_blockdev(bdev);
130 bdev->bd_block_size = size;
131 bdev->bd_inode->i_blkbits = blksize_bits(size);
132 kill_bdev(bdev);
133 }
134 return 0;
135}
136
137EXPORT_SYMBOL(set_blocksize);
138
139int sb_set_blocksize(struct super_block *sb, int size)
140{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 if (set_blocksize(sb->s_bdev, size))
142 return 0;
143 /* If we get here, we know size is power of two
144 * and it's value is between 512 and PAGE_SIZE */
145 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800146 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 return sb->s_blocksize;
148}
149
150EXPORT_SYMBOL(sb_set_blocksize);
151
152int sb_min_blocksize(struct super_block *sb, int size)
153{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400154 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (size < minsize)
156 size = minsize;
157 return sb_set_blocksize(sb, size);
158}
159
160EXPORT_SYMBOL(sb_min_blocksize);
161
162static int
163blkdev_get_block(struct inode *inode, sector_t iblock,
164 struct buffer_head *bh, int create)
165{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 bh->b_bdev = I_BDEV(inode);
167 bh->b_blocknr = iblock;
168 set_buffer_mapped(bh);
169 return 0;
170}
171
Dan Williams4ebb16c2015-10-28 07:48:19 +0900172static struct inode *bdev_file_inode(struct file *file)
173{
174 return file->f_mapping->host;
175}
176
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800177static ssize_t
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700178blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800179{
180 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +0900181 struct inode *inode = bdev_file_inode(file);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800182
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700183 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter,
Jens Axboefe0f07d2015-04-15 17:05:48 -0600184 blkdev_get_block, NULL, NULL,
185 DIO_SKIP_DIO_COUNT);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800186}
187
Jan Kara5cee5812009-04-27 16:43:51 +0200188int __sync_blockdev(struct block_device *bdev, int wait)
189{
190 if (!bdev)
191 return 0;
192 if (!wait)
193 return filemap_flush(bdev->bd_inode->i_mapping);
194 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
195}
196
Nick Piggin585d3bc2009-02-25 10:44:19 +0100197/*
198 * Write out and wait upon all the dirty data associated with a block
199 * device via its mapping. Does not take the superblock lock.
200 */
201int sync_blockdev(struct block_device *bdev)
202{
Jan Kara5cee5812009-04-27 16:43:51 +0200203 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100204}
205EXPORT_SYMBOL(sync_blockdev);
206
207/*
208 * Write out and wait upon all dirty data associated with this
209 * device. Filesystem data as well as the underlying block
210 * device. Takes the superblock lock.
211 */
212int fsync_bdev(struct block_device *bdev)
213{
214 struct super_block *sb = get_super(bdev);
215 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200216 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100217 drop_super(sb);
218 return res;
219 }
220 return sync_blockdev(bdev);
221}
Al Viro47e44912009-04-01 07:07:16 -0400222EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100223
224/**
225 * freeze_bdev -- lock a filesystem and force it into a consistent state
226 * @bdev: blockdevice to lock
227 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100228 * If a superblock is found on this device, we take the s_umount semaphore
229 * on it to make sure nobody unmounts until the snapshot creation is done.
230 * The reference counter (bd_fsfreeze_count) guarantees that only the last
231 * unfreeze process can unfreeze the frozen filesystem actually when multiple
232 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
233 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
234 * actually.
235 */
236struct super_block *freeze_bdev(struct block_device *bdev)
237{
238 struct super_block *sb;
239 int error = 0;
240
241 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200242 if (++bdev->bd_fsfreeze_count > 1) {
243 /*
244 * We don't even need to grab a reference - the first call
245 * to freeze_bdev grab an active reference and only the last
246 * thaw_bdev drops it.
247 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100248 sb = get_super(bdev);
Andrey Ryabinin5bb53c02016-08-23 18:55:31 +0300249 if (sb)
250 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100251 mutex_unlock(&bdev->bd_fsfreeze_mutex);
252 return sb;
253 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100254
Christoph Hellwig45042302009-08-03 23:28:35 +0200255 sb = get_active_super(bdev);
256 if (!sb)
257 goto out;
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600258 if (sb->s_op->freeze_super)
259 error = sb->s_op->freeze_super(sb);
260 else
261 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400262 if (error) {
263 deactivate_super(sb);
264 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400266 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100267 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400268 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200269 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100270 sync_blockdev(bdev);
271 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200272 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100273}
274EXPORT_SYMBOL(freeze_bdev);
275
276/**
277 * thaw_bdev -- unlock filesystem
278 * @bdev: blockdevice to unlock
279 * @sb: associated superblock
280 *
281 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
282 */
283int thaw_bdev(struct block_device *bdev, struct super_block *sb)
284{
Christoph Hellwig45042302009-08-03 23:28:35 +0200285 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100286
287 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200288 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400289 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100290
Christoph Hellwig45042302009-08-03 23:28:35 +0200291 error = 0;
292 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400293 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100294
Christoph Hellwig45042302009-08-03 23:28:35 +0200295 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400296 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200297
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600298 if (sb->s_op->thaw_super)
299 error = sb->s_op->thaw_super(sb);
300 else
301 error = thaw_super(sb);
Pierre Morel997198b2016-10-04 10:53:40 +0200302 if (error)
Josef Bacik18e9e512010-03-23 10:34:56 -0400303 bdev->bd_fsfreeze_count++;
Josef Bacik18e9e512010-03-23 10:34:56 -0400304out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100305 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Pierre Morel997198b2016-10-04 10:53:40 +0200306 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100307}
308EXPORT_SYMBOL(thaw_bdev);
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
311{
312 return block_write_full_page(page, blkdev_get_block, wbc);
313}
314
315static int blkdev_readpage(struct file * file, struct page * page)
316{
317 return block_read_full_page(page, blkdev_get_block);
318}
319
Akinobu Mita447f05b2014-10-09 15:26:58 -0700320static int blkdev_readpages(struct file *file, struct address_space *mapping,
321 struct list_head *pages, unsigned nr_pages)
322{
323 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
324}
325
Nick Piggin6272b5a2007-10-16 01:25:04 -0700326static int blkdev_write_begin(struct file *file, struct address_space *mapping,
327 loff_t pos, unsigned len, unsigned flags,
328 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200330 return block_write_begin(mapping, pos, len, flags, pagep,
331 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}
333
Nick Piggin6272b5a2007-10-16 01:25:04 -0700334static int blkdev_write_end(struct file *file, struct address_space *mapping,
335 loff_t pos, unsigned len, unsigned copied,
336 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700338 int ret;
339 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
340
341 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300342 put_page(page);
Nick Piggin6272b5a2007-10-16 01:25:04 -0700343
344 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
347/*
348 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500349 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 * so we compute the size by hand (just as in block_read/write above)
351 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800352static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900354 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 loff_t retval;
356
Al Viro59551022016-01-22 15:40:57 -0500357 inode_lock(bd_inode);
Al Viro5d48f3a2013-06-23 21:34:45 +0400358 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Al Viro59551022016-01-22 15:40:57 -0500359 inode_unlock(bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 return retval;
361}
362
Josef Bacik02c24a82011-07-16 20:44:56 -0400363int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900365 struct inode *bd_inode = bdev_file_inode(filp);
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400366 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100367 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200368
369 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
370 if (error)
371 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100372
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400373 /*
374 * There is no need to serialise calls to blkdev_issue_flush with
375 * i_mutex and doing so causes performance issues with concurrent
376 * O_SYNC writers to a block device.
377 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200378 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100379 if (error == -EOPNOTSUPP)
380 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400381
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100382 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700384EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700386/**
387 * bdev_read_page() - Start reading a page from a block device
388 * @bdev: The device to read the page from
389 * @sector: The offset on the device to read the page to (need not be aligned)
390 * @page: The page to read
391 *
392 * On entry, the page should be locked. It will be unlocked when the page
393 * has been read. If the block driver implements rw_page synchronously,
394 * that will be true on exit from this function, but it need not be.
395 *
396 * Errors returned by this function are usually "soft", eg out of memory, or
397 * queue full; callers should try a different route to read this page rather
398 * than propagate an error back up the stack.
399 *
400 * Return: negative errno if an error occurs, 0 if submission was successful.
401 */
402int bdev_read_page(struct block_device *bdev, sector_t sector,
403 struct page *page)
404{
405 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800406 int result = -EOPNOTSUPP;
407
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400408 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc92015-11-19 13:29:28 -0800409 return result;
410
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100411 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800412 if (result)
413 return result;
Jens Axboec11f0c02016-08-05 08:11:04 -0600414 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800415 blk_queue_exit(bdev->bd_queue);
416 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700417}
418EXPORT_SYMBOL_GPL(bdev_read_page);
419
420/**
421 * bdev_write_page() - Start writing a page to a block device
422 * @bdev: The device to write the page to
423 * @sector: The offset on the device to write the page to (need not be aligned)
424 * @page: The page to write
425 * @wbc: The writeback_control for the write
426 *
427 * On entry, the page should be locked and not currently under writeback.
428 * On exit, if the write started successfully, the page will be unlocked and
429 * under writeback. If the write failed already (eg the driver failed to
430 * queue the page to the device), the page will still be locked. If the
431 * caller is a ->writepage implementation, it will need to unlock the page.
432 *
433 * Errors returned by this function are usually "soft", eg out of memory, or
434 * queue full; callers should try a different route to write this page rather
435 * than propagate an error back up the stack.
436 *
437 * Return: negative errno if an error occurs, 0 if submission was successful.
438 */
439int bdev_write_page(struct block_device *bdev, sector_t sector,
440 struct page *page, struct writeback_control *wbc)
441{
442 int result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700443 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800444
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400445 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700446 return -EOPNOTSUPP;
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100447 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800448 if (result)
449 return result;
450
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700451 set_page_writeback(page);
Jens Axboec11f0c02016-08-05 08:11:04 -0600452 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700453 if (result)
454 end_page_writeback(page);
455 else
456 unlock_page(page);
Dan Williams2e6edc92015-11-19 13:29:28 -0800457 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700458 return result;
459}
460EXPORT_SYMBOL_GPL(bdev_write_page);
461
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200462/**
463 * bdev_direct_access() - Get the address for directly-accessibly memory
464 * @bdev: The device containing the memory
Dan Williamsb2e0d162016-01-15 16:55:59 -0800465 * @dax: control and output parameters for ->direct_access
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200466 *
467 * If a block device is made up of directly addressable memory, this function
468 * will tell the caller the PFN and the address of the memory. The address
469 * may be directly dereferenced within the kernel without the need to call
470 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
471 * page tables.
472 *
473 * Return: negative errno if an error occurs, otherwise the number of bytes
474 * accessible at this address.
475 */
Dan Williamsb2e0d162016-01-15 16:55:59 -0800476long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200477{
Dan Williamsb2e0d162016-01-15 16:55:59 -0800478 sector_t sector = dax->sector;
479 long avail, size = dax->size;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200480 const struct block_device_operations *ops = bdev->bd_disk->fops;
481
Matthew Wilcox43c3dd02015-07-03 10:40:43 -0400482 /*
483 * The device driver is allowed to sleep, in order to make the
484 * memory directly accessible.
485 */
486 might_sleep();
487
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200488 if (size < 0)
489 return size;
Toshi Kani163d4ba2016-06-23 17:05:50 -0400490 if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200491 return -EOPNOTSUPP;
492 if ((sector + DIV_ROUND_UP(size, 512)) >
493 part_nr_sects_read(bdev->bd_part))
494 return -ERANGE;
495 sector += get_start_sect(bdev);
496 if (sector % (PAGE_SIZE / 512))
497 return -EINVAL;
Dan Williams0a70bd42016-02-24 14:02:11 -0800498 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size);
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200499 if (!avail)
500 return -ERANGE;
Dan Williamsfe683ad2016-01-15 16:55:56 -0800501 if (avail > 0 && avail & ~PAGE_MASK)
502 return -ENXIO;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200503 return min(avail, size);
504}
505EXPORT_SYMBOL_GPL(bdev_direct_access);
506
Toshi Kani2d96afc2016-05-10 10:23:53 -0600507/**
508 * bdev_dax_supported() - Check if the device supports dax for filesystem
509 * @sb: The superblock of the device
510 * @blocksize: The block size of the device
511 *
512 * This is a library function for filesystems to check if the block device
513 * can be mounted with dax option.
514 *
515 * Return: negative errno if unsupported, 0 if supported.
516 */
517int bdev_dax_supported(struct super_block *sb, int blocksize)
518{
519 struct blk_dax_ctl dax = {
520 .sector = 0,
521 .size = PAGE_SIZE,
522 };
523 int err;
524
525 if (blocksize != PAGE_SIZE) {
526 vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
527 return -EINVAL;
528 }
529
530 err = bdev_direct_access(sb->s_bdev, &dax);
531 if (err < 0) {
532 switch (err) {
533 case -EOPNOTSUPP:
534 vfs_msg(sb, KERN_ERR,
535 "error: device does not support dax");
536 break;
537 case -EINVAL:
538 vfs_msg(sb, KERN_ERR,
539 "error: unaligned partition for dax");
540 break;
541 default:
542 vfs_msg(sb, KERN_ERR,
543 "error: dax access failed (%d)", err);
544 }
545 return err;
546 }
547
548 return 0;
549}
550EXPORT_SYMBOL_GPL(bdev_dax_supported);
551
Toshi Kania8078b12016-05-10 10:23:57 -0600552/**
553 * bdev_dax_capable() - Return if the raw device is capable for dax
554 * @bdev: The device for raw block device access
555 */
556bool bdev_dax_capable(struct block_device *bdev)
557{
Toshi Kania8078b12016-05-10 10:23:57 -0600558 struct blk_dax_ctl dax = {
559 .size = PAGE_SIZE,
560 };
561
562 if (!IS_ENABLED(CONFIG_FS_DAX))
563 return false;
564
565 dax.sector = 0;
566 if (bdev_direct_access(bdev, &dax) < 0)
567 return false;
568
569 dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512);
570 if (bdev_direct_access(bdev, &dax) < 0)
571 return false;
572
Toshi Kania8078b12016-05-10 10:23:57 -0600573 return true;
574}
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576/*
577 * pseudo-fs
578 */
579
580static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800581static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583static struct inode *bdev_alloc_inode(struct super_block *sb)
584{
Christoph Lametere94b1762006-12-06 20:33:17 -0800585 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (!ei)
587 return NULL;
588 return &ei->vfs_inode;
589}
590
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100591static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100593 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 struct bdev_inode *bdi = BDEV_I(inode);
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 kmem_cache_free(bdev_cachep, bdi);
597}
598
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100599static void bdev_destroy_inode(struct inode *inode)
600{
601 call_rcu(&inode->i_rcu, bdev_i_callback);
602}
603
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700604static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
606 struct bdev_inode *ei = (struct bdev_inode *) foo;
607 struct block_device *bdev = &ei->bdev;
608
Christoph Lametera35afb82007-05-16 22:10:57 -0700609 memset(bdev, 0, sizeof(*bdev));
610 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700611 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100612#ifdef CONFIG_SYSFS
613 INIT_LIST_HEAD(&bdev->bd_holder_disks);
614#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700615 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800616 /* Initialize mutex for freeze. */
617 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
619
Al Virob57922d2010-06-07 14:34:48 -0400620static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
622 struct block_device *bdev = &BDEV_I(inode)->bdev;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700623 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400624 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200625 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 spin_lock(&bdev_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 list_del_init(&bdev->bd_list);
628 spin_unlock(&bdev_lock);
629}
630
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800631static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 .statfs = simple_statfs,
633 .alloc_inode = bdev_alloc_inode,
634 .destroy_inode = bdev_destroy_inode,
635 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400636 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637};
638
Al Viro51139ad2010-07-25 23:47:46 +0400639static struct dentry *bd_mount(struct file_system_type *fs_type,
640 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Shaohua Li3684aa72016-02-22 15:27:40 -0700642 struct dentry *dent;
643 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Vegard Nossume9e5e3f2016-08-22 12:47:43 +0200644 if (!IS_ERR(dent))
Shaohua Li3684aa72016-02-22 15:27:40 -0700645 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
646 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
649static struct file_system_type bd_type = {
650 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400651 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 .kill_sb = kill_anon_super,
653};
654
Tejun Heoa212b102015-05-22 17:13:33 -0400655struct super_block *blockdev_superblock __read_mostly;
656EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658void __init bdev_cache_init(void)
659{
660 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300661 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800664 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800665 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900666 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 err = register_filesystem(&bd_type);
668 if (err)
669 panic("Cannot register bdev pseudo-fs");
670 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 if (IS_ERR(bd_mnt))
672 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300673 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
676/*
677 * Most likely _very_ bad one - but then it's hardly critical for small
678 * /dev and can be fixed when somebody will need really large one.
679 * Keep in mind that it will be fed through icache hash function too.
680 */
681static inline unsigned long hash(dev_t dev)
682{
683 return MAJOR(dev)+MINOR(dev);
684}
685
686static int bdev_test(struct inode *inode, void *data)
687{
688 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
689}
690
691static int bdev_set(struct inode *inode, void *data)
692{
693 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
694 return 0;
695}
696
697static LIST_HEAD(all_bdevs);
698
699struct block_device *bdget(dev_t dev)
700{
701 struct block_device *bdev;
702 struct inode *inode;
703
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800704 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 bdev_test, bdev_set, &dev);
706
707 if (!inode)
708 return NULL;
709
710 bdev = &BDEV_I(inode)->bdev;
711
712 if (inode->i_state & I_NEW) {
713 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000714 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 bdev->bd_inode = inode;
Fabian Frederick61604a22017-02-27 14:28:32 -0800716 bdev->bd_block_size = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 bdev->bd_part_count = 0;
718 bdev->bd_invalidated = 0;
719 inode->i_mode = S_IFBLK;
720 inode->i_rdev = dev;
721 inode->i_bdev = bdev;
722 inode->i_data.a_ops = &def_blk_aops;
723 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 spin_lock(&bdev_lock);
725 list_add(&bdev->bd_list, &all_bdevs);
726 spin_unlock(&bdev_lock);
727 unlock_new_inode(inode);
728 }
729 return bdev;
730}
731
732EXPORT_SYMBOL(bdget);
733
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200734/**
735 * bdgrab -- Grab a reference to an already referenced block device
736 * @bdev: Block device to grab a reference to.
737 */
738struct block_device *bdgrab(struct block_device *bdev)
739{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400740 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200741 return bdev;
742}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700743EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200744
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745long nr_blockdev_pages(void)
746{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700747 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 long ret = 0;
749 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700750 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 ret += bdev->bd_inode->i_mapping->nrpages;
752 }
753 spin_unlock(&bdev_lock);
754 return ret;
755}
756
757void bdput(struct block_device *bdev)
758{
759 iput(bdev->bd_inode);
760}
761
762EXPORT_SYMBOL(bdput);
763
764static struct block_device *bd_acquire(struct inode *inode)
765{
766 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700767
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 spin_lock(&bdev_lock);
769 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700770 if (bdev) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100771 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 spin_unlock(&bdev_lock);
773 return bdev;
774 }
775 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 bdev = bdget(inode->i_rdev);
778 if (bdev) {
779 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700780 if (!inode->i_bdev) {
781 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400782 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700783 * and it's released in clear_inode() of inode.
784 * So, we can access it via ->i_mapping always
785 * without igrab().
786 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100787 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700788 inode->i_bdev = bdev;
789 inode->i_mapping = bdev->bd_inode->i_mapping;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 spin_unlock(&bdev_lock);
792 }
793 return bdev;
794}
795
796/* Call when you free inode */
797
798void bd_forget(struct inode *inode)
799{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700800 struct block_device *bdev = NULL;
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700803 if (!sb_is_blkdev_sb(inode->i_sb))
804 bdev = inode->i_bdev;
Al Viroa4a4f942016-07-19 13:16:52 -0400805 inode->i_bdev = NULL;
806 inode->i_mapping = &inode->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700808
809 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100810 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
812
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900813/**
814 * bd_may_claim - test whether a block device can be claimed
815 * @bdev: block device of interest
816 * @whole: whole block device containing @bdev, may equal @bdev
817 * @holder: holder trying to claim @bdev
818 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300819 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900820 *
821 * CONTEXT:
822 * spin_lock(&bdev_lock).
823 *
824 * RETURNS:
825 * %true if @bdev can be claimed, %false otherwise.
826 */
827static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
828 void *holder)
829{
830 if (bdev->bd_holder == holder)
831 return true; /* already a holder */
832 else if (bdev->bd_holder != NULL)
833 return false; /* held by someone else */
NeilBrowncfa2d652016-12-12 08:21:51 -0700834 else if (whole == bdev)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900835 return true; /* is a whole device which isn't held */
836
Tejun Heoe525fd82010-11-13 11:55:17 +0100837 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900838 return true; /* is a partition of a device that is being partitioned */
839 else if (whole->bd_holder != NULL)
840 return false; /* is a partition of a held device */
841 else
842 return true; /* is a partition of an un-held device */
843}
844
845/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900846 * bd_prepare_to_claim - prepare to claim a block device
847 * @bdev: block device of interest
848 * @whole: the whole device containing @bdev, may equal @bdev
849 * @holder: holder trying to claim @bdev
850 *
851 * Prepare to claim @bdev. This function fails if @bdev is already
852 * claimed by another holder and waits if another claiming is in
853 * progress. This function doesn't actually claim. On successful
854 * return, the caller has ownership of bd_claiming and bd_holder[s].
855 *
856 * CONTEXT:
857 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
858 * it multiple times.
859 *
860 * RETURNS:
861 * 0 if @bdev can be claimed, -EBUSY otherwise.
862 */
863static int bd_prepare_to_claim(struct block_device *bdev,
864 struct block_device *whole, void *holder)
865{
866retry:
867 /* if someone else claimed, fail */
868 if (!bd_may_claim(bdev, whole, holder))
869 return -EBUSY;
870
Tejun Heoe75aa852010-08-04 17:59:39 +0200871 /* if claiming is already in progress, wait for it to finish */
872 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900873 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
874 DEFINE_WAIT(wait);
875
876 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
877 spin_unlock(&bdev_lock);
878 schedule();
879 finish_wait(wq, &wait);
880 spin_lock(&bdev_lock);
881 goto retry;
882 }
883
884 /* yay, all mine */
885 return 0;
886}
887
888/**
889 * bd_start_claiming - start claiming a block device
890 * @bdev: block device of interest
891 * @holder: holder trying to claim @bdev
892 *
893 * @bdev is about to be opened exclusively. Check @bdev can be opened
894 * exclusively and mark that an exclusive open is in progress. Each
895 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000896 * either bd_finish_claiming() or bd_abort_claiming() (which do not
897 * fail).
898 *
899 * This function is used to gain exclusive access to the block device
900 * without actually causing other exclusive open attempts to fail. It
901 * should be used when the open sequence itself requires exclusive
902 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900903 *
904 * CONTEXT:
905 * Might sleep.
906 *
907 * RETURNS:
908 * Pointer to the block device containing @bdev on success, ERR_PTR()
909 * value on failure.
910 */
911static struct block_device *bd_start_claiming(struct block_device *bdev,
912 void *holder)
913{
914 struct gendisk *disk;
915 struct block_device *whole;
916 int partno, err;
917
918 might_sleep();
919
920 /*
921 * @bdev might not have been initialized properly yet, look up
922 * and grab the outer block device the hard way.
923 */
924 disk = get_gendisk(bdev->bd_dev, &partno);
925 if (!disk)
926 return ERR_PTR(-ENXIO);
927
Tejun Heod4c208b2011-06-13 12:45:48 +0200928 /*
929 * Normally, @bdev should equal what's returned from bdget_disk()
930 * if partno is 0; however, some drivers (floppy) use multiple
931 * bdev's for the same physical device and @bdev may be one of the
932 * aliases. Keep @bdev if partno is 0. This means claimer
933 * tracking is broken for those devices but it has always been that
934 * way.
935 */
936 if (partno)
937 whole = bdget_disk(disk, 0);
938 else
939 whole = bdgrab(bdev);
940
Nick Piggincf342572010-05-26 01:50:21 +1000941 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900942 put_disk(disk);
943 if (!whole)
944 return ERR_PTR(-ENOMEM);
945
946 /* prepare to claim, if successful, mark claiming in progress */
947 spin_lock(&bdev_lock);
948
949 err = bd_prepare_to_claim(bdev, whole, holder);
950 if (err == 0) {
951 whole->bd_claiming = holder;
952 spin_unlock(&bdev_lock);
953 return whole;
954 } else {
955 spin_unlock(&bdev_lock);
956 bdput(whole);
957 return ERR_PTR(err);
958 }
959}
960
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800961#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100962struct bd_holder_disk {
963 struct list_head list;
964 struct gendisk *disk;
965 int refcnt;
966};
967
968static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
969 struct gendisk *disk)
970{
971 struct bd_holder_disk *holder;
972
973 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
974 if (holder->disk == disk)
975 return holder;
976 return NULL;
977}
978
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700979static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800980{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700981 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800982}
983
984static void del_symlink(struct kobject *from, struct kobject *to)
985{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800986 sysfs_remove_link(from, kobject_name(to));
987}
988
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800989/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100990 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
991 * @bdev: the claimed slave bdev
992 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500993 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100994 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
995 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100996 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500997 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100998 * - from "slaves" directory of the holder @disk to the claimed @bdev
999 * - from "holders" directory of the @bdev to the holder @disk
1000 *
1001 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1002 * passed to bd_link_disk_holder(), then:
1003 *
1004 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1005 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1006 *
1007 * The caller must have claimed @bdev before calling this function and
1008 * ensure that both @bdev and @disk are valid during the creation and
1009 * lifetime of these symlinks.
1010 *
1011 * CONTEXT:
1012 * Might sleep.
1013 *
1014 * RETURNS:
1015 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001016 */
Tejun Heoe09b4572010-11-13 11:55:17 +01001017int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001018{
Tejun Heo49731ba2011-01-14 18:43:57 +01001019 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001020 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001021
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001022 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001023
Tejun Heo49731ba2011-01-14 18:43:57 +01001024 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -07001025
Tejun Heoe09b4572010-11-13 11:55:17 +01001026 /* FIXME: remove the following once add_disk() handles errors */
1027 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1028 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -07001029
Tejun Heo49731ba2011-01-14 18:43:57 +01001030 holder = bd_find_holder_disk(bdev, disk);
1031 if (holder) {
1032 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +01001033 goto out_unlock;
1034 }
1035
Tejun Heo49731ba2011-01-14 18:43:57 +01001036 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1037 if (!holder) {
1038 ret = -ENOMEM;
1039 goto out_unlock;
1040 }
1041
1042 INIT_LIST_HEAD(&holder->list);
1043 holder->disk = disk;
1044 holder->refcnt = 1;
1045
1046 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1047 if (ret)
1048 goto out_free;
1049
1050 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1051 if (ret)
1052 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +01001053 /*
1054 * bdev could be deleted beneath us which would implicitly destroy
1055 * the holder directory. Hold on to it.
1056 */
1057 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001058
1059 list_add(&holder->list, &bdev->bd_holder_disks);
1060 goto out_unlock;
1061
1062out_del:
1063 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1064out_free:
1065 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001066out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001067 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001068 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001069}
Tejun Heoe09b4572010-11-13 11:55:17 +01001070EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001071
Tejun Heo49731ba2011-01-14 18:43:57 +01001072/**
1073 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1074 * @bdev: the calimed slave bdev
1075 * @disk: the holding disk
1076 *
1077 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1078 *
1079 * CONTEXT:
1080 * Might sleep.
1081 */
1082void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001083{
Tejun Heo49731ba2011-01-14 18:43:57 +01001084 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001085
Tejun Heo49731ba2011-01-14 18:43:57 +01001086 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001087
Tejun Heo49731ba2011-01-14 18:43:57 +01001088 holder = bd_find_holder_disk(bdev, disk);
1089
1090 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1091 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1092 del_symlink(bdev->bd_part->holder_dir,
1093 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001094 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001095 list_del_init(&holder->list);
1096 kfree(holder);
1097 }
1098
1099 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001100}
Tejun Heo49731ba2011-01-14 18:43:57 +01001101EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001102#endif
1103
Andrew Patterson0c002c22008-09-04 14:27:20 -06001104/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001105 * flush_disk - invalidates all buffer-cache entries on a disk
1106 *
1107 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001108 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001109 *
1110 * Invalidates all buffer-cache entries on a disk. It should be called
1111 * when a disk has been changed -- either by a media change or online
1112 * resize.
1113 */
NeilBrown93b270f2011-02-24 17:25:47 +11001114static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001115{
NeilBrown93b270f2011-02-24 17:25:47 +11001116 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001117 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001118 "resized disk %s\n",
1119 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001120 }
1121
1122 if (!bdev->bd_disk)
1123 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001124 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001125 bdev->bd_invalidated = 1;
1126}
1127
1128/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001129 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001130 * @disk: struct gendisk to check
1131 * @bdev: struct bdev to adjust.
1132 *
1133 * This routine checks to see if the bdev size does not match the disk size
1134 * and adjusts it if it differs.
1135 */
1136void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1137{
1138 loff_t disk_size, bdev_size;
1139
1140 disk_size = (loff_t)get_capacity(disk) << 9;
1141 bdev_size = i_size_read(bdev->bd_inode);
1142 if (disk_size != bdev_size) {
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001143 printk(KERN_INFO
1144 "%s: detected capacity change from %lld to %lld\n",
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001145 disk->disk_name, bdev_size, disk_size);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001146 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001147 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001148 }
1149}
1150EXPORT_SYMBOL(check_disk_size_change);
1151
1152/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001153 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001154 * @disk: struct gendisk to be revalidated
1155 *
1156 * This routine is a wrapper for lower-level driver's revalidate_disk
1157 * call-backs. It is used to do common pre and post operations needed
1158 * for all revalidate_disk operations.
1159 */
1160int revalidate_disk(struct gendisk *disk)
1161{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001162 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001163 int ret = 0;
1164
1165 if (disk->fops->revalidate_disk)
1166 ret = disk->fops->revalidate_disk(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001167 bdev = bdget_disk(disk, 0);
1168 if (!bdev)
1169 return ret;
1170
1171 mutex_lock(&bdev->bd_mutex);
1172 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001173 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001174 mutex_unlock(&bdev->bd_mutex);
1175 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001176 return ret;
1177}
1178EXPORT_SYMBOL(revalidate_disk);
1179
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180/*
1181 * This routine checks whether a removable media has been changed,
1182 * and invalidates all buffer-cache-entries in that case. This
1183 * is a relatively slow routine, so we have to try to minimize using
1184 * it. Thus it is called only upon a 'mount' or 'open'. This
1185 * is the best way of combining speed and utility, I think.
1186 * People changing diskettes in the middle of an operation deserve
1187 * to lose :-)
1188 */
1189int check_disk_change(struct block_device *bdev)
1190{
1191 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001192 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001193 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Tejun Heo77ea8872010-12-08 20:57:37 +01001195 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1196 DISK_EVENT_EJECT_REQUEST);
1197 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 return 0;
1199
NeilBrown93b270f2011-02-24 17:25:47 +11001200 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 if (bdops->revalidate_disk)
1202 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return 1;
1204}
1205
1206EXPORT_SYMBOL(check_disk_change);
1207
1208void bd_set_size(struct block_device *bdev, loff_t size)
1209{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001210 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
Al Viro59551022016-01-22 15:40:57 -05001212 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001213 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001214 inode_unlock(bdev->bd_inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001215 while (bsize < PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if (size & bsize)
1217 break;
1218 bsize <<= 1;
1219 }
1220 bdev->bd_block_size = bsize;
1221 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1222}
1223EXPORT_SYMBOL(bd_set_size);
1224
Al Viro4385bab2013-05-05 22:11:03 -04001225static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001226
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001227/*
1228 * bd_mutex locking:
1229 *
1230 * mutex_lock(part->bd_mutex)
1231 * mutex_lock_nested(whole->bd_mutex, 1)
1232 */
1233
Al Viro572c4892007-10-08 13:24:05 -04001234static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001237 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001238 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001239 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001240 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Al Viro572c4892007-10-08 13:24:05 -04001242 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001243 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001244 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001245 perm |= MAY_WRITE;
1246 /*
1247 * hooks: /n/, see "layering violations".
1248 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001249 if (!for_part) {
1250 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1251 if (ret != 0) {
1252 bdput(bdev);
1253 return ret;
1254 }
Al Viro82666022008-08-01 05:32:04 -04001255 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001256
NeilBrownd3374822009-01-09 08:31:10 +11001257 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001258
Tejun Heo89f97492008-11-05 10:21:06 +01001259 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001260 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001261 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001262 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001263 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Tejun Heo69e02c52011-03-09 19:54:27 +01001265 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001266 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (!bdev->bd_openers) {
1268 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001269 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 bdev->bd_contains = bdev;
Dan Williams03cdadb2016-02-26 15:19:43 -08001271
Tejun Heocf771cb2008-09-03 09:01:09 +02001272 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001273 ret = -ENXIO;
1274 bdev->bd_part = disk_get_part(disk, partno);
1275 if (!bdev->bd_part)
1276 goto out_clear;
1277
Tejun Heo1196f8b2011-04-21 20:54:45 +02001278 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001280 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001281 if (ret == -ERESTARTSYS) {
1282 /* Lost a race with 'disk' being
1283 * deleted, try again.
1284 * See md.c
1285 */
1286 disk_put_part(bdev->bd_part);
1287 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001288 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001289 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001290 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001291 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001292 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001293 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001294 goto restart;
1295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 }
Tejun Heo7e697232011-05-23 13:26:07 +02001297
Christoph Hellwig22375702016-09-14 11:56:13 +02001298 if (!ret)
Tejun Heo7e697232011-05-23 13:26:07 +02001299 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Tejun Heo7e697232011-05-23 13:26:07 +02001300
Tejun Heo1196f8b2011-04-21 20:54:45 +02001301 /*
1302 * If the device is invalidated, rescan partition
1303 * if open succeeded or failed with -ENOMEDIUM.
1304 * The latter is necessary to prevent ghost
1305 * partitions on a removed medium.
1306 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001307 if (bdev->bd_invalidated) {
1308 if (!ret)
1309 rescan_partitions(disk, bdev);
1310 else if (ret == -ENOMEDIUM)
1311 invalidate_partitions(disk, bdev);
1312 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001313
Tejun Heo1196f8b2011-04-21 20:54:45 +02001314 if (ret)
1315 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 struct block_device *whole;
1318 whole = bdget_disk(disk, 0);
1319 ret = -ENOMEM;
1320 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001321 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001322 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001323 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001325 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001327 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001328 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001329 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001331 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
Tejun Heo89f97492008-11-05 10:21:06 +01001333 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 }
1335 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001337 ret = 0;
1338 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001339 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001340 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001341 if (bdev->bd_invalidated) {
1342 if (!ret)
1343 rescan_partitions(bdev->bd_disk, bdev);
1344 else if (ret == -ENOMEDIUM)
1345 invalidate_partitions(bdev->bd_disk, bdev);
1346 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001347 if (ret)
1348 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001350 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001351 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001352 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 }
1354 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001355 if (for_part)
1356 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001357 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001358 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 return 0;
1360
Tejun Heo0762b8b2008-08-25 19:56:12 +09001361 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001362 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001364 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001365 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001367 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001369 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001370 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001371 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001372 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001373 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001374 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001375 bdput(bdev);
1376
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 return ret;
1378}
1379
Tejun Heod4d77622010-11-13 11:55:18 +01001380/**
1381 * blkdev_get - open a block device
1382 * @bdev: block_device to open
1383 * @mode: FMODE_* mask
1384 * @holder: exclusive holder identifier
1385 *
1386 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1387 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1388 * @holder is invalid. Exclusive opens may nest for the same @holder.
1389 *
1390 * On success, the reference count of @bdev is unchanged. On failure,
1391 * @bdev is put.
1392 *
1393 * CONTEXT:
1394 * Might sleep.
1395 *
1396 * RETURNS:
1397 * 0 on success, -errno on failure.
1398 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001399int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
Tejun Heoe525fd82010-11-13 11:55:17 +01001401 struct block_device *whole = NULL;
1402 int res;
1403
1404 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1405
1406 if ((mode & FMODE_EXCL) && holder) {
1407 whole = bd_start_claiming(bdev, holder);
1408 if (IS_ERR(whole)) {
1409 bdput(bdev);
1410 return PTR_ERR(whole);
1411 }
1412 }
1413
1414 res = __blkdev_get(bdev, mode, 0);
1415
1416 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001417 struct gendisk *disk = whole->bd_disk;
1418
Tejun Heo6a027ef2010-11-13 11:55:17 +01001419 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001420 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001421 spin_lock(&bdev_lock);
1422
Tejun Heo77ea8872010-12-08 20:57:37 +01001423 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001424 BUG_ON(!bd_may_claim(bdev, whole, holder));
1425 /*
1426 * Note that for a whole device bd_holders
1427 * will be incremented twice, and bd_holder
1428 * will be set to bd_may_claim before being
1429 * set to holder
1430 */
1431 whole->bd_holders++;
1432 whole->bd_holder = bd_may_claim;
1433 bdev->bd_holders++;
1434 bdev->bd_holder = holder;
1435 }
1436
1437 /* tell others that we're done */
1438 BUG_ON(whole->bd_claiming != holder);
1439 whole->bd_claiming = NULL;
1440 wake_up_bit(&whole->bd_claiming, 0);
1441
1442 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001443
1444 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001445 * Block event polling for write claims if requested. Any
1446 * write holder makes the write_holder state stick until
1447 * all are released. This is good enough and tracking
1448 * individual writeable reference is too fragile given the
1449 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001450 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001451 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1452 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001453 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001454 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001455 }
1456
1457 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001458 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001459 }
1460
1461 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001462}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463EXPORT_SYMBOL(blkdev_get);
1464
Tejun Heod4d77622010-11-13 11:55:18 +01001465/**
1466 * blkdev_get_by_path - open a block device by name
1467 * @path: path to the block device to open
1468 * @mode: FMODE_* mask
1469 * @holder: exclusive holder identifier
1470 *
1471 * Open the blockdevice described by the device file at @path. @mode
1472 * and @holder are identical to blkdev_get().
1473 *
1474 * On success, the returned block_device has reference count of one.
1475 *
1476 * CONTEXT:
1477 * Might sleep.
1478 *
1479 * RETURNS:
1480 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1481 */
1482struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1483 void *holder)
1484{
1485 struct block_device *bdev;
1486 int err;
1487
1488 bdev = lookup_bdev(path);
1489 if (IS_ERR(bdev))
1490 return bdev;
1491
1492 err = blkdev_get(bdev, mode, holder);
1493 if (err)
1494 return ERR_PTR(err);
1495
Chuck Ebberte51900f2011-02-16 18:11:53 -05001496 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1497 blkdev_put(bdev, mode);
1498 return ERR_PTR(-EACCES);
1499 }
1500
Tejun Heod4d77622010-11-13 11:55:18 +01001501 return bdev;
1502}
1503EXPORT_SYMBOL(blkdev_get_by_path);
1504
1505/**
1506 * blkdev_get_by_dev - open a block device by device number
1507 * @dev: device number of block device to open
1508 * @mode: FMODE_* mask
1509 * @holder: exclusive holder identifier
1510 *
1511 * Open the blockdevice described by device number @dev. @mode and
1512 * @holder are identical to blkdev_get().
1513 *
1514 * Use it ONLY if you really do not have anything better - i.e. when
1515 * you are behind a truly sucky interface and all you are given is a
1516 * device number. _Never_ to be used for internal purposes. If you
1517 * ever need it - reconsider your API.
1518 *
1519 * On success, the returned block_device has reference count of one.
1520 *
1521 * CONTEXT:
1522 * Might sleep.
1523 *
1524 * RETURNS:
1525 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1526 */
1527struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1528{
1529 struct block_device *bdev;
1530 int err;
1531
1532 bdev = bdget(dev);
1533 if (!bdev)
1534 return ERR_PTR(-ENOMEM);
1535
1536 err = blkdev_get(bdev, mode, holder);
1537 if (err)
1538 return ERR_PTR(err);
1539
1540 return bdev;
1541}
1542EXPORT_SYMBOL(blkdev_get_by_dev);
1543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544static int blkdev_open(struct inode * inode, struct file * filp)
1545{
1546 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 /*
1549 * Preserve backwards compatibility and allow large file access
1550 * even if userspace doesn't ask for it explicitly. Some mkfs
1551 * binary needs it. We might want to drop this workaround
1552 * during an unstable branch.
1553 */
1554 filp->f_flags |= O_LARGEFILE;
1555
Al Viro572c4892007-10-08 13:24:05 -04001556 if (filp->f_flags & O_NDELAY)
1557 filp->f_mode |= FMODE_NDELAY;
1558 if (filp->f_flags & O_EXCL)
1559 filp->f_mode |= FMODE_EXCL;
1560 if ((filp->f_flags & O_ACCMODE) == 3)
1561 filp->f_mode |= FMODE_WRITE_IOCTL;
1562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001564 if (bdev == NULL)
1565 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Al Viro572c4892007-10-08 13:24:05 -04001567 filp->f_mapping = bdev->bd_inode->i_mapping;
1568
Tejun Heoe525fd82010-11-13 11:55:17 +01001569 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570}
1571
Al Viro4385bab2013-05-05 22:11:03 -04001572static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001573{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001574 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001575 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001576
NeilBrown6796bf52006-12-08 02:36:16 -08001577 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001578 if (for_part)
1579 bdev->bd_part_count--;
1580
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001581 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001582 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001583 sync_blockdev(bdev);
1584 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001585
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001586 bdev_write_inode(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001587 /*
1588 * Detaching bdev inode from its wb in __destroy_inode()
1589 * is too late: the queue which embeds its bdi (along with
1590 * root wb) can be gone as soon as we put_disk() below.
1591 */
1592 inode_detach_wb(bdev->bd_inode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001593 }
1594 if (bdev->bd_contains == bdev) {
1595 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001596 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001597 }
1598 if (!bdev->bd_openers) {
1599 struct module *owner = disk->fops->owner;
1600
Tejun Heo0762b8b2008-08-25 19:56:12 +09001601 disk_put_part(bdev->bd_part);
1602 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001603 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001604 if (bdev != bdev->bd_contains)
1605 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001606 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001607
1608 put_disk(disk);
1609 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001610 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001611 mutex_unlock(&bdev->bd_mutex);
1612 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001613 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001614 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001615}
1616
Al Viro4385bab2013-05-05 22:11:03 -04001617void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001618{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001619 mutex_lock(&bdev->bd_mutex);
1620
Tejun Heoe525fd82010-11-13 11:55:17 +01001621 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001622 bool bdev_free;
1623
1624 /*
1625 * Release a claim on the device. The holder fields
1626 * are protected with bdev_lock. bd_mutex is to
1627 * synchronize disk_holder unlinking.
1628 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001629 spin_lock(&bdev_lock);
1630
1631 WARN_ON_ONCE(--bdev->bd_holders < 0);
1632 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1633
1634 /* bd_contains might point to self, check in a separate step */
1635 if ((bdev_free = !bdev->bd_holders))
1636 bdev->bd_holder = NULL;
1637 if (!bdev->bd_contains->bd_holders)
1638 bdev->bd_contains->bd_holder = NULL;
1639
1640 spin_unlock(&bdev_lock);
1641
Tejun Heo77ea8872010-12-08 20:57:37 +01001642 /*
1643 * If this was the last claim, remove holder link and
1644 * unblock evpoll if it was a write holder.
1645 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001646 if (bdev_free && bdev->bd_write_holder) {
1647 disk_unblock_events(bdev->bd_disk);
1648 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001649 }
Tejun Heo69362172011-03-09 19:54:27 +01001650 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001651
Tejun Heo85ef06d2011-07-01 16:17:47 +02001652 /*
1653 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1654 * event. This is to ensure detection of media removal commanded
1655 * from userland - e.g. eject(1).
1656 */
1657 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1658
1659 mutex_unlock(&bdev->bd_mutex);
1660
Al Viro4385bab2013-05-05 22:11:03 -04001661 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001662}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001663EXPORT_SYMBOL(blkdev_put);
1664
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665static int blkdev_close(struct inode * inode, struct file * filp)
1666{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001667 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001668 blkdev_put(bdev, filp->f_mode);
1669 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670}
1671
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001672static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001674 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001675 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001676
1677 /*
1678 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1679 * to updated it before every ioctl.
1680 */
Al Viro56b26ad2008-09-19 03:17:36 -04001681 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001682 mode |= FMODE_NDELAY;
1683 else
1684 mode &= ~FMODE_NDELAY;
1685
Al Viro56b26ad2008-09-19 03:17:36 -04001686 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687}
1688
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001689/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001690 * Write data to the block device. Only intended for the block device itself
1691 * and the raw driver which basically is a fake block device.
1692 *
1693 * Does not take i_mutex for the write and thus is not for general purpose
1694 * use.
1695 */
Al Viro1456c0a2014-04-03 03:21:50 -04001696ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001697{
1698 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001699 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001700 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001701 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001702 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001703
Al Viro7ec7b942015-04-07 11:35:14 -04001704 if (bdev_read_only(I_BDEV(bd_inode)))
1705 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001706
Al Viro7ec7b942015-04-07 11:35:14 -04001707 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001708 return 0;
1709
Al Viro7ec7b942015-04-07 11:35:14 -04001710 if (iocb->ki_pos >= size)
1711 return -ENOSPC;
1712
1713 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001714
Jianpeng Ma53362a02012-08-02 09:50:39 +02001715 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001716 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwige2592212016-04-07 08:52:01 -07001717 if (ret > 0)
1718 ret = generic_write_sync(iocb, ret);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001719 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001720 return ret;
1721}
Al Viro1456c0a2014-04-03 03:21:50 -04001722EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001723
David Jefferyb2de5252014-09-29 10:21:10 -04001724ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001725{
1726 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001727 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001728 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001729 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001730
1731 if (pos >= size)
1732 return 0;
1733
1734 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001735 iov_iter_truncate(to, size);
1736 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001737}
David Jefferyb2de5252014-09-29 10:21:10 -04001738EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001739
Christoph Hellwigeef99382009-08-20 17:43:41 +02001740/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001741 * Try to release a page associated with block device when the system
1742 * is under memory pressure.
1743 */
1744static int blkdev_releasepage(struct page *page, gfp_t wait)
1745{
1746 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1747
1748 if (super && super->s_op->bdev_try_to_free_page)
1749 return super->s_op->bdev_try_to_free_page(super, page, wait);
1750
1751 return try_to_free_buffers(page);
1752}
1753
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001754static int blkdev_writepages(struct address_space *mapping,
1755 struct writeback_control *wbc)
1756{
1757 if (dax_mapping(mapping)) {
1758 struct block_device *bdev = I_BDEV(mapping->host);
1759
1760 return dax_writeback_mapping_range(mapping, bdev, wbc);
1761 }
1762 return generic_writepages(mapping, wbc);
1763}
1764
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001765static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001767 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001769 .write_begin = blkdev_write_begin,
1770 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001771 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001772 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001774 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775};
1776
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001777#define BLKDEV_FALLOC_FL_SUPPORTED \
1778 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1779 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1780
1781static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1782 loff_t len)
1783{
1784 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1785 struct request_queue *q = bdev_get_queue(bdev);
1786 struct address_space *mapping;
1787 loff_t end = start + len - 1;
1788 loff_t isize;
1789 int error;
1790
1791 /* Fail if we don't recognize the flags. */
1792 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1793 return -EOPNOTSUPP;
1794
1795 /* Don't go off the end of the device. */
1796 isize = i_size_read(bdev->bd_inode);
1797 if (start >= isize)
1798 return -EINVAL;
1799 if (end >= isize) {
1800 if (mode & FALLOC_FL_KEEP_SIZE) {
1801 len = isize - start;
1802 end = start + len - 1;
1803 } else
1804 return -EINVAL;
1805 }
1806
1807 /*
1808 * Don't allow IO that isn't aligned to logical block size.
1809 */
1810 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1811 return -EINVAL;
1812
1813 /* Invalidate the page cache, including dirty pages. */
1814 mapping = bdev->bd_inode->i_mapping;
1815 truncate_inode_pages_range(mapping, start, end);
1816
1817 switch (mode) {
1818 case FALLOC_FL_ZERO_RANGE:
1819 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1820 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1821 GFP_KERNEL, false);
1822 break;
1823 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1824 /* Only punch if the device can do zeroing discard. */
1825 if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data)
1826 return -EOPNOTSUPP;
1827 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1828 GFP_KERNEL, 0);
1829 break;
1830 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1831 if (!blk_queue_discard(q))
1832 return -EOPNOTSUPP;
1833 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1834 GFP_KERNEL, 0);
1835 break;
1836 default:
1837 return -EOPNOTSUPP;
1838 }
1839 if (error)
1840 return error;
1841
1842 /*
1843 * Invalidate again; if someone wandered in and dirtied a page,
1844 * the caller will be given -EBUSY. The third argument is
1845 * inclusive, so the rounding here is safe.
1846 */
1847 return invalidate_inode_pages2_range(mapping,
1848 start >> PAGE_SHIFT,
1849 end >> PAGE_SHIFT);
1850}
1851
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001852const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 .open = blkdev_open,
1854 .release = blkdev_close,
1855 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001856 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001857 .write_iter = blkdev_write_iter,
Dan Williamsacc93d32016-05-07 11:40:28 -07001858 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001859 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001860 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861#ifdef CONFIG_COMPAT
1862 .compat_ioctl = compat_blkdev_ioctl,
1863#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001864 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001865 .splice_write = iter_file_splice_write,
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001866 .fallocate = blkdev_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867};
1868
1869int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1870{
1871 int res;
1872 mm_segment_t old_fs = get_fs();
1873 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001874 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 set_fs(old_fs);
1876 return res;
1877}
1878
1879EXPORT_SYMBOL(ioctl_by_bdev);
1880
1881/**
1882 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001883 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001885 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 * namespace if possible and return it. Return ERR_PTR(error)
1887 * otherwise.
1888 */
Al Viro421748e2008-08-02 01:04:36 -04001889struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890{
1891 struct block_device *bdev;
1892 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001893 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 int error;
1895
Al Viro421748e2008-08-02 01:04:36 -04001896 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 return ERR_PTR(-EINVAL);
1898
Al Viro421748e2008-08-02 01:04:36 -04001899 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 if (error)
1901 return ERR_PTR(error);
1902
David Howellsbb6687342015-03-17 22:26:21 +00001903 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 error = -ENOTBLK;
1905 if (!S_ISBLK(inode->i_mode))
1906 goto fail;
1907 error = -EACCES;
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05001908 if (!may_open_dev(&path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 goto fail;
1910 error = -ENOMEM;
1911 bdev = bd_acquire(inode);
1912 if (!bdev)
1913 goto fail;
1914out:
Al Viro421748e2008-08-02 01:04:36 -04001915 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 return bdev;
1917fail:
1918 bdev = ERR_PTR(error);
1919 goto out;
1920}
Al Virod5686b42008-08-01 05:00:11 -04001921EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
NeilBrown93b270f2011-02-24 17:25:47 +11001923int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001924{
1925 struct super_block *sb = get_super(bdev);
1926 int res = 0;
1927
1928 if (sb) {
1929 /*
1930 * no need to lock the super, get_super holds the
1931 * read mutex so the filesystem cannot go away
1932 * under us (->put_super runs with the write lock
1933 * hold).
1934 */
1935 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001936 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001937 drop_super(sb);
1938 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001939 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001940 return res;
1941}
1942EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001943
1944void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1945{
1946 struct inode *inode, *old_inode = NULL;
1947
Dave Chinner74278da2015-03-04 12:37:22 -05001948 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001949 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1950 struct address_space *mapping = inode->i_mapping;
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001951 struct block_device *bdev;
Jan Kara5c0d6b62012-07-03 16:45:31 +02001952
1953 spin_lock(&inode->i_lock);
1954 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1955 mapping->nrpages == 0) {
1956 spin_unlock(&inode->i_lock);
1957 continue;
1958 }
1959 __iget(inode);
1960 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05001961 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001962 /*
1963 * We hold a reference to 'inode' so it couldn't have been
1964 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05001965 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02001966 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05001967 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02001968 * later.
1969 */
1970 iput(old_inode);
1971 old_inode = inode;
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001972 bdev = I_BDEV(inode);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001973
Rabin Vincent11aa5c12016-12-01 09:18:28 +01001974 mutex_lock(&bdev->bd_mutex);
1975 if (bdev->bd_openers)
1976 func(bdev, arg);
1977 mutex_unlock(&bdev->bd_mutex);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001978
Dave Chinner74278da2015-03-04 12:37:22 -05001979 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001980 }
Dave Chinner74278da2015-03-04 12:37:22 -05001981 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001982 iput(old_inode);
1983}