blob: 30b8d568203a0aa9c8724809478bc9af269ef4d4 [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 Viroff01bb42011-09-16 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 Viroff01bb42011-09-16 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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020034#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36struct bdev_inode {
37 struct block_device bdev;
38 struct inode vfs_inode;
39};
40
Adrian Bunk4c54ac62008-02-18 13:48:31 +010041static const struct address_space_operations def_blk_aops;
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043static inline struct bdev_inode *BDEV_I(struct inode *inode)
44{
45 return container_of(inode, struct bdev_inode, vfs_inode);
46}
47
Geert Uytterhoevenff5053f2015-06-26 13:58:32 +020048struct block_device *I_BDEV(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070049{
50 return &BDEV_I(inode)->bdev;
51}
Linus Torvalds1da177e2005-04-16 15:20:36 -070052EXPORT_SYMBOL(I_BDEV);
53
Toshi Kani2af3a812016-05-10 10:23:52 -060054void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
55{
56 struct va_format vaf;
57 va_list args;
58
59 va_start(args, fmt);
60 vaf.fmt = fmt;
61 vaf.va = &args;
62 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
63 va_end(args);
64}
65
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070066static void bdev_write_inode(struct block_device *bdev)
Christoph Hellwig564f00f2015-01-14 10:42:33 +010067{
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070068 struct inode *inode = bdev->bd_inode;
69 int ret;
70
Christoph Hellwig564f00f2015-01-14 10:42:33 +010071 spin_lock(&inode->i_lock);
72 while (inode->i_state & I_DIRTY) {
73 spin_unlock(&inode->i_lock);
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070074 ret = write_inode_now(inode, true);
75 if (ret) {
76 char name[BDEVNAME_SIZE];
77 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
78 "for block device %s (err=%d).\n",
79 bdevname(bdev, name), ret);
80 }
Christoph Hellwig564f00f2015-01-14 10:42:33 +010081 spin_lock(&inode->i_lock);
82 }
83 spin_unlock(&inode->i_lock);
84}
85
Peter Zijlstraf9a14392007-05-06 14:49:55 -070086/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb42011-09-16 02:31:11 -040087void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Al Viroff01bb42011-09-16 02:31:11 -040089 struct address_space *mapping = bdev->bd_inode->i_mapping;
90
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080091 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070092 return;
Al Viroff01bb42011-09-16 02:31:11 -040093
Peter Zijlstraf9a14392007-05-06 14:49:55 -070094 invalidate_bh_lrus();
Al Viroff01bb42011-09-16 02:31:11 -040095 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096}
Al Viroff01bb42011-09-16 02:31:11 -040097EXPORT_SYMBOL(kill_bdev);
98
99/* Invalidate clean unused buffers and pagecache. */
100void invalidate_bdev(struct block_device *bdev)
101{
102 struct address_space *mapping = bdev->bd_inode->i_mapping;
103
104 if (mapping->nrpages == 0)
105 return;
106
107 invalidate_bh_lrus();
108 lru_add_drain_all(); /* make sure all lru add caches are flushed */
109 invalidate_mapping_pages(mapping, 0, -1);
110 /* 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 Viroff01bb42011-09-16 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
Matthew Wilcoxbbab37d2015-07-03 10:40:42 -0400183 if (IS_DAX(inode))
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700184 return dax_do_io(iocb, inode, iter, blkdev_get_block,
Matthew Wilcoxbbab37d2015-07-03 10:40:42 -0400185 NULL, DIO_SKIP_DIO_COUNT);
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700186 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter,
Jens Axboefe0f07d2015-04-15 17:05:48 -0600187 blkdev_get_block, NULL, NULL,
188 DIO_SKIP_DIO_COUNT);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800189}
190
Jan Kara5cee5812009-04-27 16:43:51 +0200191int __sync_blockdev(struct block_device *bdev, int wait)
192{
193 if (!bdev)
194 return 0;
195 if (!wait)
196 return filemap_flush(bdev->bd_inode->i_mapping);
197 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
198}
199
Nick Piggin585d3bc2009-02-25 10:44:19 +0100200/*
201 * Write out and wait upon all the dirty data associated with a block
202 * device via its mapping. Does not take the superblock lock.
203 */
204int sync_blockdev(struct block_device *bdev)
205{
Jan Kara5cee5812009-04-27 16:43:51 +0200206 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100207}
208EXPORT_SYMBOL(sync_blockdev);
209
210/*
211 * Write out and wait upon all dirty data associated with this
212 * device. Filesystem data as well as the underlying block
213 * device. Takes the superblock lock.
214 */
215int fsync_bdev(struct block_device *bdev)
216{
217 struct super_block *sb = get_super(bdev);
218 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200219 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100220 drop_super(sb);
221 return res;
222 }
223 return sync_blockdev(bdev);
224}
Al Viro47e44912009-04-01 07:07:16 -0400225EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100226
227/**
228 * freeze_bdev -- lock a filesystem and force it into a consistent state
229 * @bdev: blockdevice to lock
230 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100231 * If a superblock is found on this device, we take the s_umount semaphore
232 * on it to make sure nobody unmounts until the snapshot creation is done.
233 * The reference counter (bd_fsfreeze_count) guarantees that only the last
234 * unfreeze process can unfreeze the frozen filesystem actually when multiple
235 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
236 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
237 * actually.
238 */
239struct super_block *freeze_bdev(struct block_device *bdev)
240{
241 struct super_block *sb;
242 int error = 0;
243
244 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200245 if (++bdev->bd_fsfreeze_count > 1) {
246 /*
247 * We don't even need to grab a reference - the first call
248 * to freeze_bdev grab an active reference and only the last
249 * thaw_bdev drops it.
250 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100251 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200252 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100253 mutex_unlock(&bdev->bd_fsfreeze_mutex);
254 return sb;
255 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100256
Christoph Hellwig45042302009-08-03 23:28:35 +0200257 sb = get_active_super(bdev);
258 if (!sb)
259 goto out;
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600260 if (sb->s_op->freeze_super)
261 error = sb->s_op->freeze_super(sb);
262 else
263 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400264 if (error) {
265 deactivate_super(sb);
266 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200267 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400268 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100269 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400270 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200271 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100272 sync_blockdev(bdev);
273 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200274 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100275}
276EXPORT_SYMBOL(freeze_bdev);
277
278/**
279 * thaw_bdev -- unlock filesystem
280 * @bdev: blockdevice to unlock
281 * @sb: associated superblock
282 *
283 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
284 */
285int thaw_bdev(struct block_device *bdev, struct super_block *sb)
286{
Christoph Hellwig45042302009-08-03 23:28:35 +0200287 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100288
289 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200290 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400291 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100292
Christoph Hellwig45042302009-08-03 23:28:35 +0200293 error = 0;
294 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400295 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100296
Christoph Hellwig45042302009-08-03 23:28:35 +0200297 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400298 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200299
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600300 if (sb->s_op->thaw_super)
301 error = sb->s_op->thaw_super(sb);
302 else
303 error = thaw_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400304 if (error) {
305 bdev->bd_fsfreeze_count++;
306 mutex_unlock(&bdev->bd_fsfreeze_mutex);
307 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100308 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400309out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100310 mutex_unlock(&bdev->bd_fsfreeze_mutex);
311 return 0;
312}
313EXPORT_SYMBOL(thaw_bdev);
314
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
316{
317 return block_write_full_page(page, blkdev_get_block, wbc);
318}
319
320static int blkdev_readpage(struct file * file, struct page * page)
321{
322 return block_read_full_page(page, blkdev_get_block);
323}
324
Akinobu Mita447f05b2014-10-09 15:26:58 -0700325static int blkdev_readpages(struct file *file, struct address_space *mapping,
326 struct list_head *pages, unsigned nr_pages)
327{
328 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
329}
330
Nick Piggin6272b5a2007-10-16 01:25:04 -0700331static int blkdev_write_begin(struct file *file, struct address_space *mapping,
332 loff_t pos, unsigned len, unsigned flags,
333 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200335 return block_write_begin(mapping, pos, len, flags, pagep,
336 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337}
338
Nick Piggin6272b5a2007-10-16 01:25:04 -0700339static int blkdev_write_end(struct file *file, struct address_space *mapping,
340 loff_t pos, unsigned len, unsigned copied,
341 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700343 int ret;
344 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
345
346 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300347 put_page(page);
Nick Piggin6272b5a2007-10-16 01:25:04 -0700348
349 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
352/*
353 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500354 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 * so we compute the size by hand (just as in block_read/write above)
356 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800357static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900359 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 loff_t retval;
361
Al Viro59551022016-01-22 15:40:57 -0500362 inode_lock(bd_inode);
Al Viro5d48f3a2013-06-23 21:34:45 +0400363 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Al Viro59551022016-01-22 15:40:57 -0500364 inode_unlock(bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 return retval;
366}
367
Josef Bacik02c24a82011-07-16 20:44:56 -0400368int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900370 struct inode *bd_inode = bdev_file_inode(filp);
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400371 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100372 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200373
374 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
375 if (error)
376 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100377
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400378 /*
379 * There is no need to serialise calls to blkdev_issue_flush with
380 * i_mutex and doing so causes performance issues with concurrent
381 * O_SYNC writers to a block device.
382 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200383 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100384 if (error == -EOPNOTSUPP)
385 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400386
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100387 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700389EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700391/**
392 * bdev_read_page() - Start reading a page from a block device
393 * @bdev: The device to read the page from
394 * @sector: The offset on the device to read the page to (need not be aligned)
395 * @page: The page to read
396 *
397 * On entry, the page should be locked. It will be unlocked when the page
398 * has been read. If the block driver implements rw_page synchronously,
399 * that will be true on exit from this function, but it need not be.
400 *
401 * Errors returned by this function are usually "soft", eg out of memory, or
402 * queue full; callers should try a different route to read this page rather
403 * than propagate an error back up the stack.
404 *
405 * Return: negative errno if an error occurs, 0 if submission was successful.
406 */
407int bdev_read_page(struct block_device *bdev, sector_t sector,
408 struct page *page)
409{
410 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc952015-11-19 13:29:28 -0800411 int result = -EOPNOTSUPP;
412
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400413 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc952015-11-19 13:29:28 -0800414 return result;
415
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100416 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc952015-11-19 13:29:28 -0800417 if (result)
418 return result;
419 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
420 blk_queue_exit(bdev->bd_queue);
421 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700422}
423EXPORT_SYMBOL_GPL(bdev_read_page);
424
425/**
426 * bdev_write_page() - Start writing a page to a block device
427 * @bdev: The device to write the page to
428 * @sector: The offset on the device to write the page to (need not be aligned)
429 * @page: The page to write
430 * @wbc: The writeback_control for the write
431 *
432 * On entry, the page should be locked and not currently under writeback.
433 * On exit, if the write started successfully, the page will be unlocked and
434 * under writeback. If the write failed already (eg the driver failed to
435 * queue the page to the device), the page will still be locked. If the
436 * caller is a ->writepage implementation, it will need to unlock the page.
437 *
438 * Errors returned by this function are usually "soft", eg out of memory, or
439 * queue full; callers should try a different route to write this page rather
440 * than propagate an error back up the stack.
441 *
442 * Return: negative errno if an error occurs, 0 if submission was successful.
443 */
444int bdev_write_page(struct block_device *bdev, sector_t sector,
445 struct page *page, struct writeback_control *wbc)
446{
447 int result;
448 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
449 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc952015-11-19 13:29:28 -0800450
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400451 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700452 return -EOPNOTSUPP;
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100453 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc952015-11-19 13:29:28 -0800454 if (result)
455 return result;
456
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700457 set_page_writeback(page);
458 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
459 if (result)
460 end_page_writeback(page);
461 else
462 unlock_page(page);
Dan Williams2e6edc952015-11-19 13:29:28 -0800463 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700464 return result;
465}
466EXPORT_SYMBOL_GPL(bdev_write_page);
467
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200468/**
469 * bdev_direct_access() - Get the address for directly-accessibly memory
470 * @bdev: The device containing the memory
Dan Williamsb2e0d162016-01-15 16:55:59 -0800471 * @dax: control and output parameters for ->direct_access
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200472 *
473 * If a block device is made up of directly addressable memory, this function
474 * will tell the caller the PFN and the address of the memory. The address
475 * may be directly dereferenced within the kernel without the need to call
476 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
477 * page tables.
478 *
479 * Return: negative errno if an error occurs, otherwise the number of bytes
480 * accessible at this address.
481 */
Dan Williamsb2e0d162016-01-15 16:55:59 -0800482long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200483{
Dan Williamsb2e0d162016-01-15 16:55:59 -0800484 sector_t sector = dax->sector;
485 long avail, size = dax->size;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200486 const struct block_device_operations *ops = bdev->bd_disk->fops;
487
Matthew Wilcox43c3dd02015-07-03 10:40:43 -0400488 /*
489 * The device driver is allowed to sleep, in order to make the
490 * memory directly accessible.
491 */
492 might_sleep();
493
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200494 if (size < 0)
495 return size;
496 if (!ops->direct_access)
497 return -EOPNOTSUPP;
498 if ((sector + DIV_ROUND_UP(size, 512)) >
499 part_nr_sects_read(bdev->bd_part))
500 return -ERANGE;
501 sector += get_start_sect(bdev);
502 if (sector % (PAGE_SIZE / 512))
503 return -EINVAL;
Dan Williams0a70bd42016-02-24 14:02:11 -0800504 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size);
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200505 if (!avail)
506 return -ERANGE;
Dan Williamsfe683ad2016-01-15 16:55:56 -0800507 if (avail > 0 && avail & ~PAGE_MASK)
508 return -ENXIO;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200509 return min(avail, size);
510}
511EXPORT_SYMBOL_GPL(bdev_direct_access);
512
Toshi Kani2d96afc2016-05-10 10:23:53 -0600513/**
514 * bdev_dax_supported() - Check if the device supports dax for filesystem
515 * @sb: The superblock of the device
516 * @blocksize: The block size of the device
517 *
518 * This is a library function for filesystems to check if the block device
519 * can be mounted with dax option.
520 *
521 * Return: negative errno if unsupported, 0 if supported.
522 */
523int bdev_dax_supported(struct super_block *sb, int blocksize)
524{
525 struct blk_dax_ctl dax = {
526 .sector = 0,
527 .size = PAGE_SIZE,
528 };
529 int err;
530
531 if (blocksize != PAGE_SIZE) {
532 vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
533 return -EINVAL;
534 }
535
536 err = bdev_direct_access(sb->s_bdev, &dax);
537 if (err < 0) {
538 switch (err) {
539 case -EOPNOTSUPP:
540 vfs_msg(sb, KERN_ERR,
541 "error: device does not support dax");
542 break;
543 case -EINVAL:
544 vfs_msg(sb, KERN_ERR,
545 "error: unaligned partition for dax");
546 break;
547 default:
548 vfs_msg(sb, KERN_ERR,
549 "error: dax access failed (%d)", err);
550 }
551 return err;
552 }
553
554 return 0;
555}
556EXPORT_SYMBOL_GPL(bdev_dax_supported);
557
Toshi Kania8078b12016-05-10 10:23:57 -0600558/**
559 * bdev_dax_capable() - Return if the raw device is capable for dax
560 * @bdev: The device for raw block device access
561 */
562bool bdev_dax_capable(struct block_device *bdev)
563{
Toshi Kania8078b12016-05-10 10:23:57 -0600564 struct blk_dax_ctl dax = {
565 .size = PAGE_SIZE,
566 };
567
568 if (!IS_ENABLED(CONFIG_FS_DAX))
569 return false;
570
571 dax.sector = 0;
572 if (bdev_direct_access(bdev, &dax) < 0)
573 return false;
574
575 dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512);
576 if (bdev_direct_access(bdev, &dax) < 0)
577 return false;
578
Toshi Kania8078b12016-05-10 10:23:57 -0600579 return true;
580}
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582/*
583 * pseudo-fs
584 */
585
586static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800587static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589static struct inode *bdev_alloc_inode(struct super_block *sb)
590{
Christoph Lametere94b1762006-12-06 20:33:17 -0800591 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 if (!ei)
593 return NULL;
594 return &ei->vfs_inode;
595}
596
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100597static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100599 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 struct bdev_inode *bdi = BDEV_I(inode);
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 kmem_cache_free(bdev_cachep, bdi);
603}
604
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100605static void bdev_destroy_inode(struct inode *inode)
606{
607 call_rcu(&inode->i_rcu, bdev_i_callback);
608}
609
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700610static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
612 struct bdev_inode *ei = (struct bdev_inode *) foo;
613 struct block_device *bdev = &ei->bdev;
614
Christoph Lametera35afb82007-05-16 22:10:57 -0700615 memset(bdev, 0, sizeof(*bdev));
616 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700617 INIT_LIST_HEAD(&bdev->bd_inodes);
618 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100619#ifdef CONFIG_SYSFS
620 INIT_LIST_HEAD(&bdev->bd_holder_disks);
621#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700622 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800623 /* Initialize mutex for freeze. */
624 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
627static inline void __bd_forget(struct inode *inode)
628{
629 list_del_init(&inode->i_devices);
630 inode->i_bdev = NULL;
631 inode->i_mapping = &inode->i_data;
632}
633
Al Virob57922d2010-06-07 14:34:48 -0400634static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
636 struct block_device *bdev = &BDEV_I(inode)->bdev;
637 struct list_head *p;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700638 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400639 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200640 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 spin_lock(&bdev_lock);
642 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
643 __bd_forget(list_entry(p, struct inode, i_devices));
644 }
645 list_del_init(&bdev->bd_list);
646 spin_unlock(&bdev_lock);
647}
648
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800649static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 .statfs = simple_statfs,
651 .alloc_inode = bdev_alloc_inode,
652 .destroy_inode = bdev_destroy_inode,
653 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400654 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655};
656
Al Viro51139ad2010-07-25 23:47:46 +0400657static struct dentry *bd_mount(struct file_system_type *fs_type,
658 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
Shaohua Li3684aa72016-02-22 15:27:40 -0700660 struct dentry *dent;
661 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
662 if (dent)
663 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
664 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
667static struct file_system_type bd_type = {
668 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400669 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 .kill_sb = kill_anon_super,
671};
672
Tejun Heoa212b102015-05-22 17:13:33 -0400673struct super_block *blockdev_superblock __read_mostly;
674EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
676void __init bdev_cache_init(void)
677{
678 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300679 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800682 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800683 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900684 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 err = register_filesystem(&bd_type);
686 if (err)
687 panic("Cannot register bdev pseudo-fs");
688 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 if (IS_ERR(bd_mnt))
690 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300691 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692}
693
694/*
695 * Most likely _very_ bad one - but then it's hardly critical for small
696 * /dev and can be fixed when somebody will need really large one.
697 * Keep in mind that it will be fed through icache hash function too.
698 */
699static inline unsigned long hash(dev_t dev)
700{
701 return MAJOR(dev)+MINOR(dev);
702}
703
704static int bdev_test(struct inode *inode, void *data)
705{
706 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
707}
708
709static int bdev_set(struct inode *inode, void *data)
710{
711 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
712 return 0;
713}
714
715static LIST_HEAD(all_bdevs);
716
717struct block_device *bdget(dev_t dev)
718{
719 struct block_device *bdev;
720 struct inode *inode;
721
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800722 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 bdev_test, bdev_set, &dev);
724
725 if (!inode)
726 return NULL;
727
728 bdev = &BDEV_I(inode)->bdev;
729
730 if (inode->i_state & I_NEW) {
731 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000732 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 bdev->bd_inode = inode;
734 bdev->bd_block_size = (1 << inode->i_blkbits);
735 bdev->bd_part_count = 0;
736 bdev->bd_invalidated = 0;
737 inode->i_mode = S_IFBLK;
738 inode->i_rdev = dev;
739 inode->i_bdev = bdev;
740 inode->i_data.a_ops = &def_blk_aops;
741 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 spin_lock(&bdev_lock);
743 list_add(&bdev->bd_list, &all_bdevs);
744 spin_unlock(&bdev_lock);
745 unlock_new_inode(inode);
746 }
747 return bdev;
748}
749
750EXPORT_SYMBOL(bdget);
751
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200752/**
753 * bdgrab -- Grab a reference to an already referenced block device
754 * @bdev: Block device to grab a reference to.
755 */
756struct block_device *bdgrab(struct block_device *bdev)
757{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400758 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200759 return bdev;
760}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700761EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763long nr_blockdev_pages(void)
764{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700765 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 long ret = 0;
767 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700768 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 ret += bdev->bd_inode->i_mapping->nrpages;
770 }
771 spin_unlock(&bdev_lock);
772 return ret;
773}
774
775void bdput(struct block_device *bdev)
776{
777 iput(bdev->bd_inode);
778}
779
780EXPORT_SYMBOL(bdput);
781
782static struct block_device *bd_acquire(struct inode *inode)
783{
784 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 spin_lock(&bdev_lock);
787 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700788 if (bdev) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100789 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 spin_unlock(&bdev_lock);
791 return bdev;
792 }
793 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700794
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 bdev = bdget(inode->i_rdev);
796 if (bdev) {
797 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700798 if (!inode->i_bdev) {
799 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400800 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700801 * and it's released in clear_inode() of inode.
802 * So, we can access it via ->i_mapping always
803 * without igrab().
804 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100805 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700806 inode->i_bdev = bdev;
807 inode->i_mapping = bdev->bd_inode->i_mapping;
808 list_add(&inode->i_devices, &bdev->bd_inodes);
809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 spin_unlock(&bdev_lock);
811 }
812 return bdev;
813}
814
815/* Call when you free inode */
816
817void bd_forget(struct inode *inode)
818{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700819 struct block_device *bdev = NULL;
820
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700822 if (!sb_is_blkdev_sb(inode->i_sb))
823 bdev = inode->i_bdev;
824 __bd_forget(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700826
827 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100828 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829}
830
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900831/**
832 * bd_may_claim - test whether a block device can be claimed
833 * @bdev: block device of interest
834 * @whole: whole block device containing @bdev, may equal @bdev
835 * @holder: holder trying to claim @bdev
836 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300837 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900838 *
839 * CONTEXT:
840 * spin_lock(&bdev_lock).
841 *
842 * RETURNS:
843 * %true if @bdev can be claimed, %false otherwise.
844 */
845static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
846 void *holder)
847{
848 if (bdev->bd_holder == holder)
849 return true; /* already a holder */
850 else if (bdev->bd_holder != NULL)
851 return false; /* held by someone else */
852 else if (bdev->bd_contains == bdev)
853 return true; /* is a whole device which isn't held */
854
Tejun Heoe525fd82010-11-13 11:55:17 +0100855 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900856 return true; /* is a partition of a device that is being partitioned */
857 else if (whole->bd_holder != NULL)
858 return false; /* is a partition of a held device */
859 else
860 return true; /* is a partition of an un-held device */
861}
862
863/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900864 * bd_prepare_to_claim - prepare to claim a block device
865 * @bdev: block device of interest
866 * @whole: the whole device containing @bdev, may equal @bdev
867 * @holder: holder trying to claim @bdev
868 *
869 * Prepare to claim @bdev. This function fails if @bdev is already
870 * claimed by another holder and waits if another claiming is in
871 * progress. This function doesn't actually claim. On successful
872 * return, the caller has ownership of bd_claiming and bd_holder[s].
873 *
874 * CONTEXT:
875 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
876 * it multiple times.
877 *
878 * RETURNS:
879 * 0 if @bdev can be claimed, -EBUSY otherwise.
880 */
881static int bd_prepare_to_claim(struct block_device *bdev,
882 struct block_device *whole, void *holder)
883{
884retry:
885 /* if someone else claimed, fail */
886 if (!bd_may_claim(bdev, whole, holder))
887 return -EBUSY;
888
Tejun Heoe75aa852010-08-04 17:59:39 +0200889 /* if claiming is already in progress, wait for it to finish */
890 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900891 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
892 DEFINE_WAIT(wait);
893
894 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
895 spin_unlock(&bdev_lock);
896 schedule();
897 finish_wait(wq, &wait);
898 spin_lock(&bdev_lock);
899 goto retry;
900 }
901
902 /* yay, all mine */
903 return 0;
904}
905
906/**
907 * bd_start_claiming - start claiming a block device
908 * @bdev: block device of interest
909 * @holder: holder trying to claim @bdev
910 *
911 * @bdev is about to be opened exclusively. Check @bdev can be opened
912 * exclusively and mark that an exclusive open is in progress. Each
913 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000914 * either bd_finish_claiming() or bd_abort_claiming() (which do not
915 * fail).
916 *
917 * This function is used to gain exclusive access to the block device
918 * without actually causing other exclusive open attempts to fail. It
919 * should be used when the open sequence itself requires exclusive
920 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900921 *
922 * CONTEXT:
923 * Might sleep.
924 *
925 * RETURNS:
926 * Pointer to the block device containing @bdev on success, ERR_PTR()
927 * value on failure.
928 */
929static struct block_device *bd_start_claiming(struct block_device *bdev,
930 void *holder)
931{
932 struct gendisk *disk;
933 struct block_device *whole;
934 int partno, err;
935
936 might_sleep();
937
938 /*
939 * @bdev might not have been initialized properly yet, look up
940 * and grab the outer block device the hard way.
941 */
942 disk = get_gendisk(bdev->bd_dev, &partno);
943 if (!disk)
944 return ERR_PTR(-ENXIO);
945
Tejun Heod4c208b2011-06-13 12:45:48 +0200946 /*
947 * Normally, @bdev should equal what's returned from bdget_disk()
948 * if partno is 0; however, some drivers (floppy) use multiple
949 * bdev's for the same physical device and @bdev may be one of the
950 * aliases. Keep @bdev if partno is 0. This means claimer
951 * tracking is broken for those devices but it has always been that
952 * way.
953 */
954 if (partno)
955 whole = bdget_disk(disk, 0);
956 else
957 whole = bdgrab(bdev);
958
Nick Piggincf342572010-05-26 01:50:21 +1000959 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900960 put_disk(disk);
961 if (!whole)
962 return ERR_PTR(-ENOMEM);
963
964 /* prepare to claim, if successful, mark claiming in progress */
965 spin_lock(&bdev_lock);
966
967 err = bd_prepare_to_claim(bdev, whole, holder);
968 if (err == 0) {
969 whole->bd_claiming = holder;
970 spin_unlock(&bdev_lock);
971 return whole;
972 } else {
973 spin_unlock(&bdev_lock);
974 bdput(whole);
975 return ERR_PTR(err);
976 }
977}
978
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800979#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100980struct bd_holder_disk {
981 struct list_head list;
982 struct gendisk *disk;
983 int refcnt;
984};
985
986static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
987 struct gendisk *disk)
988{
989 struct bd_holder_disk *holder;
990
991 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
992 if (holder->disk == disk)
993 return holder;
994 return NULL;
995}
996
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700997static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800998{
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700999 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001000}
1001
1002static void del_symlink(struct kobject *from, struct kobject *to)
1003{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001004 sysfs_remove_link(from, kobject_name(to));
1005}
1006
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001007/**
Tejun Heoe09b4572010-11-13 11:55:17 +01001008 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1009 * @bdev: the claimed slave bdev
1010 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001011 *
Tejun Heo49731ba2011-01-14 18:43:57 +01001012 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1013 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001014 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001015 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001016 * - from "slaves" directory of the holder @disk to the claimed @bdev
1017 * - from "holders" directory of the @bdev to the holder @disk
1018 *
1019 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1020 * passed to bd_link_disk_holder(), then:
1021 *
1022 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1023 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1024 *
1025 * The caller must have claimed @bdev before calling this function and
1026 * ensure that both @bdev and @disk are valid during the creation and
1027 * lifetime of these symlinks.
1028 *
1029 * CONTEXT:
1030 * Might sleep.
1031 *
1032 * RETURNS:
1033 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001034 */
Tejun Heoe09b4572010-11-13 11:55:17 +01001035int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001036{
Tejun Heo49731ba2011-01-14 18:43:57 +01001037 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001038 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001039
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001040 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001041
Tejun Heo49731ba2011-01-14 18:43:57 +01001042 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -07001043
Tejun Heoe09b4572010-11-13 11:55:17 +01001044 /* FIXME: remove the following once add_disk() handles errors */
1045 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1046 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -07001047
Tejun Heo49731ba2011-01-14 18:43:57 +01001048 holder = bd_find_holder_disk(bdev, disk);
1049 if (holder) {
1050 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +01001051 goto out_unlock;
1052 }
1053
Tejun Heo49731ba2011-01-14 18:43:57 +01001054 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1055 if (!holder) {
1056 ret = -ENOMEM;
1057 goto out_unlock;
1058 }
1059
1060 INIT_LIST_HEAD(&holder->list);
1061 holder->disk = disk;
1062 holder->refcnt = 1;
1063
1064 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1065 if (ret)
1066 goto out_free;
1067
1068 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1069 if (ret)
1070 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +01001071 /*
1072 * bdev could be deleted beneath us which would implicitly destroy
1073 * the holder directory. Hold on to it.
1074 */
1075 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001076
1077 list_add(&holder->list, &bdev->bd_holder_disks);
1078 goto out_unlock;
1079
1080out_del:
1081 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1082out_free:
1083 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001084out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001085 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001086 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001087}
Tejun Heoe09b4572010-11-13 11:55:17 +01001088EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001089
Tejun Heo49731ba2011-01-14 18:43:57 +01001090/**
1091 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1092 * @bdev: the calimed slave bdev
1093 * @disk: the holding disk
1094 *
1095 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1096 *
1097 * CONTEXT:
1098 * Might sleep.
1099 */
1100void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001101{
Tejun Heo49731ba2011-01-14 18:43:57 +01001102 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001103
Tejun Heo49731ba2011-01-14 18:43:57 +01001104 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001105
Tejun Heo49731ba2011-01-14 18:43:57 +01001106 holder = bd_find_holder_disk(bdev, disk);
1107
1108 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1109 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1110 del_symlink(bdev->bd_part->holder_dir,
1111 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001112 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001113 list_del_init(&holder->list);
1114 kfree(holder);
1115 }
1116
1117 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001118}
Tejun Heo49731ba2011-01-14 18:43:57 +01001119EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001120#endif
1121
Andrew Patterson0c002c22008-09-04 14:27:20 -06001122/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001123 * flush_disk - invalidates all buffer-cache entries on a disk
1124 *
1125 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001126 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001127 *
1128 * Invalidates all buffer-cache entries on a disk. It should be called
1129 * when a disk has been changed -- either by a media change or online
1130 * resize.
1131 */
NeilBrown93b270f2011-02-24 17:25:47 +11001132static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001133{
NeilBrown93b270f2011-02-24 17:25:47 +11001134 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001135 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001136 "resized disk %s\n",
1137 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001138 }
1139
1140 if (!bdev->bd_disk)
1141 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001142 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001143 bdev->bd_invalidated = 1;
1144}
1145
1146/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001147 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001148 * @disk: struct gendisk to check
1149 * @bdev: struct bdev to adjust.
1150 *
1151 * This routine checks to see if the bdev size does not match the disk size
1152 * and adjusts it if it differs.
1153 */
1154void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1155{
1156 loff_t disk_size, bdev_size;
1157
1158 disk_size = (loff_t)get_capacity(disk) << 9;
1159 bdev_size = i_size_read(bdev->bd_inode);
1160 if (disk_size != bdev_size) {
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001161 printk(KERN_INFO
1162 "%s: detected capacity change from %lld to %lld\n",
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001163 disk->disk_name, bdev_size, disk_size);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001164 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001165 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001166 }
1167}
1168EXPORT_SYMBOL(check_disk_size_change);
1169
1170/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001171 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001172 * @disk: struct gendisk to be revalidated
1173 *
1174 * This routine is a wrapper for lower-level driver's revalidate_disk
1175 * call-backs. It is used to do common pre and post operations needed
1176 * for all revalidate_disk operations.
1177 */
1178int revalidate_disk(struct gendisk *disk)
1179{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001180 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001181 int ret = 0;
1182
1183 if (disk->fops->revalidate_disk)
1184 ret = disk->fops->revalidate_disk(disk);
Martin K. Petersen25520d52015-10-21 13:19:49 -04001185 blk_integrity_revalidate(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001186 bdev = bdget_disk(disk, 0);
1187 if (!bdev)
1188 return ret;
1189
1190 mutex_lock(&bdev->bd_mutex);
1191 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001192 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001193 mutex_unlock(&bdev->bd_mutex);
1194 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001195 return ret;
1196}
1197EXPORT_SYMBOL(revalidate_disk);
1198
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199/*
1200 * This routine checks whether a removable media has been changed,
1201 * and invalidates all buffer-cache-entries in that case. This
1202 * is a relatively slow routine, so we have to try to minimize using
1203 * it. Thus it is called only upon a 'mount' or 'open'. This
1204 * is the best way of combining speed and utility, I think.
1205 * People changing diskettes in the middle of an operation deserve
1206 * to lose :-)
1207 */
1208int check_disk_change(struct block_device *bdev)
1209{
1210 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001211 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001212 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Tejun Heo77ea8872010-12-08 20:57:37 +01001214 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1215 DISK_EVENT_EJECT_REQUEST);
1216 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 return 0;
1218
NeilBrown93b270f2011-02-24 17:25:47 +11001219 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 if (bdops->revalidate_disk)
1221 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 return 1;
1223}
1224
1225EXPORT_SYMBOL(check_disk_change);
1226
1227void bd_set_size(struct block_device *bdev, loff_t size)
1228{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001229 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Al Viro59551022016-01-22 15:40:57 -05001231 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001232 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001233 inode_unlock(bdev->bd_inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001234 while (bsize < PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (size & bsize)
1236 break;
1237 bsize <<= 1;
1238 }
1239 bdev->bd_block_size = bsize;
1240 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1241}
1242EXPORT_SYMBOL(bd_set_size);
1243
Al Viro4385bab2013-05-05 22:11:03 -04001244static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001245
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001246/*
1247 * bd_mutex locking:
1248 *
1249 * mutex_lock(part->bd_mutex)
1250 * mutex_lock_nested(whole->bd_mutex, 1)
1251 */
1252
Al Viro572c4892007-10-08 13:24:05 -04001253static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001256 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001257 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001258 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001259 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Al Viro572c4892007-10-08 13:24:05 -04001261 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001262 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001263 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001264 perm |= MAY_WRITE;
1265 /*
1266 * hooks: /n/, see "layering violations".
1267 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001268 if (!for_part) {
1269 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1270 if (ret != 0) {
1271 bdput(bdev);
1272 return ret;
1273 }
Al Viro82666022008-08-01 05:32:04 -04001274 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001275
NeilBrownd3374822009-01-09 08:31:10 +11001276 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001277
Tejun Heo89f97492008-11-05 10:21:06 +01001278 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001279 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001280 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001281 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001282 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Tejun Heo69e02c52011-03-09 19:54:27 +01001284 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001285 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 if (!bdev->bd_openers) {
1287 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001288 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 bdev->bd_contains = bdev;
Dan Williams03cdadb2016-02-26 15:19:43 -08001290 if (IS_ENABLED(CONFIG_BLK_DEV_DAX) && disk->fops->direct_access)
1291 bdev->bd_inode->i_flags = S_DAX;
1292 else
1293 bdev->bd_inode->i_flags = 0;
1294
Tejun Heocf771cb2008-09-03 09:01:09 +02001295 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001296 ret = -ENXIO;
1297 bdev->bd_part = disk_get_part(disk, partno);
1298 if (!bdev->bd_part)
1299 goto out_clear;
1300
Tejun Heo1196f8b2011-04-21 20:54:45 +02001301 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001303 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001304 if (ret == -ERESTARTSYS) {
1305 /* Lost a race with 'disk' being
1306 * deleted, try again.
1307 * See md.c
1308 */
1309 disk_put_part(bdev->bd_part);
1310 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001311 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001312 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001313 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001314 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001315 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001316 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001317 goto restart;
1318 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 }
Tejun Heo7e697232011-05-23 13:26:07 +02001320
Dan Williams5a023cd2015-11-30 10:20:29 -08001321 if (!ret) {
Tejun Heo7e697232011-05-23 13:26:07 +02001322 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Toshi Kania8078b12016-05-10 10:23:57 -06001323 if (!bdev_dax_capable(bdev))
Dan Williams5a023cd2015-11-30 10:20:29 -08001324 bdev->bd_inode->i_flags &= ~S_DAX;
1325 }
Tejun Heo7e697232011-05-23 13:26:07 +02001326
Tejun Heo1196f8b2011-04-21 20:54:45 +02001327 /*
1328 * If the device is invalidated, rescan partition
1329 * if open succeeded or failed with -ENOMEDIUM.
1330 * The latter is necessary to prevent ghost
1331 * partitions on a removed medium.
1332 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001333 if (bdev->bd_invalidated) {
1334 if (!ret)
1335 rescan_partitions(disk, bdev);
1336 else if (ret == -ENOMEDIUM)
1337 invalidate_partitions(disk, bdev);
1338 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001339
Tejun Heo1196f8b2011-04-21 20:54:45 +02001340 if (ret)
1341 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 struct block_device *whole;
1344 whole = bdget_disk(disk, 0);
1345 ret = -ENOMEM;
1346 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001347 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001348 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001349 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001351 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001353 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001354 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001355 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001357 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 }
Tejun Heo89f97492008-11-05 10:21:06 +01001359 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Toshi Kania8078b12016-05-10 10:23:57 -06001360 if (!bdev_dax_capable(bdev))
Jeff Moyerf0b2e562015-08-14 16:15:32 -04001361 bdev->bd_inode->i_flags &= ~S_DAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 }
1363 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001365 ret = 0;
1366 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001367 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001368 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001369 if (bdev->bd_invalidated) {
1370 if (!ret)
1371 rescan_partitions(bdev->bd_disk, bdev);
1372 else if (ret == -ENOMEDIUM)
1373 invalidate_partitions(bdev->bd_disk, bdev);
1374 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001375 if (ret)
1376 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001378 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001379 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001380 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 }
1382 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001383 if (for_part)
1384 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001385 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001386 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 return 0;
1388
Tejun Heo0762b8b2008-08-25 19:56:12 +09001389 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001390 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001392 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001393 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001395 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001397 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001398 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001399 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001400 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001401 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001402 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001403 bdput(bdev);
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 return ret;
1406}
1407
Tejun Heod4d77622010-11-13 11:55:18 +01001408/**
1409 * blkdev_get - open a block device
1410 * @bdev: block_device to open
1411 * @mode: FMODE_* mask
1412 * @holder: exclusive holder identifier
1413 *
1414 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1415 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1416 * @holder is invalid. Exclusive opens may nest for the same @holder.
1417 *
1418 * On success, the reference count of @bdev is unchanged. On failure,
1419 * @bdev is put.
1420 *
1421 * CONTEXT:
1422 * Might sleep.
1423 *
1424 * RETURNS:
1425 * 0 on success, -errno on failure.
1426 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001427int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428{
Tejun Heoe525fd82010-11-13 11:55:17 +01001429 struct block_device *whole = NULL;
1430 int res;
1431
1432 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1433
1434 if ((mode & FMODE_EXCL) && holder) {
1435 whole = bd_start_claiming(bdev, holder);
1436 if (IS_ERR(whole)) {
1437 bdput(bdev);
1438 return PTR_ERR(whole);
1439 }
1440 }
1441
1442 res = __blkdev_get(bdev, mode, 0);
1443
1444 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001445 struct gendisk *disk = whole->bd_disk;
1446
Tejun Heo6a027ef2010-11-13 11:55:17 +01001447 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001448 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001449 spin_lock(&bdev_lock);
1450
Tejun Heo77ea8872010-12-08 20:57:37 +01001451 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001452 BUG_ON(!bd_may_claim(bdev, whole, holder));
1453 /*
1454 * Note that for a whole device bd_holders
1455 * will be incremented twice, and bd_holder
1456 * will be set to bd_may_claim before being
1457 * set to holder
1458 */
1459 whole->bd_holders++;
1460 whole->bd_holder = bd_may_claim;
1461 bdev->bd_holders++;
1462 bdev->bd_holder = holder;
1463 }
1464
1465 /* tell others that we're done */
1466 BUG_ON(whole->bd_claiming != holder);
1467 whole->bd_claiming = NULL;
1468 wake_up_bit(&whole->bd_claiming, 0);
1469
1470 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001471
1472 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001473 * Block event polling for write claims if requested. Any
1474 * write holder makes the write_holder state stick until
1475 * all are released. This is good enough and tracking
1476 * individual writeable reference is too fragile given the
1477 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001478 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001479 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1480 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001481 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001482 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001483 }
1484
1485 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001486 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001487 }
1488
1489 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001490}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491EXPORT_SYMBOL(blkdev_get);
1492
Tejun Heod4d77622010-11-13 11:55:18 +01001493/**
1494 * blkdev_get_by_path - open a block device by name
1495 * @path: path to the block device to open
1496 * @mode: FMODE_* mask
1497 * @holder: exclusive holder identifier
1498 *
1499 * Open the blockdevice described by the device file at @path. @mode
1500 * and @holder are identical to blkdev_get().
1501 *
1502 * On success, the returned block_device has reference count of one.
1503 *
1504 * CONTEXT:
1505 * Might sleep.
1506 *
1507 * RETURNS:
1508 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1509 */
1510struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1511 void *holder)
1512{
1513 struct block_device *bdev;
1514 int err;
1515
1516 bdev = lookup_bdev(path);
1517 if (IS_ERR(bdev))
1518 return bdev;
1519
1520 err = blkdev_get(bdev, mode, holder);
1521 if (err)
1522 return ERR_PTR(err);
1523
Chuck Ebberte51900f2011-02-16 18:11:53 -05001524 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1525 blkdev_put(bdev, mode);
1526 return ERR_PTR(-EACCES);
1527 }
1528
Tejun Heod4d77622010-11-13 11:55:18 +01001529 return bdev;
1530}
1531EXPORT_SYMBOL(blkdev_get_by_path);
1532
1533/**
1534 * blkdev_get_by_dev - open a block device by device number
1535 * @dev: device number of block device to open
1536 * @mode: FMODE_* mask
1537 * @holder: exclusive holder identifier
1538 *
1539 * Open the blockdevice described by device number @dev. @mode and
1540 * @holder are identical to blkdev_get().
1541 *
1542 * Use it ONLY if you really do not have anything better - i.e. when
1543 * you are behind a truly sucky interface and all you are given is a
1544 * device number. _Never_ to be used for internal purposes. If you
1545 * ever need it - reconsider your API.
1546 *
1547 * On success, the returned block_device has reference count of one.
1548 *
1549 * CONTEXT:
1550 * Might sleep.
1551 *
1552 * RETURNS:
1553 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1554 */
1555struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1556{
1557 struct block_device *bdev;
1558 int err;
1559
1560 bdev = bdget(dev);
1561 if (!bdev)
1562 return ERR_PTR(-ENOMEM);
1563
1564 err = blkdev_get(bdev, mode, holder);
1565 if (err)
1566 return ERR_PTR(err);
1567
1568 return bdev;
1569}
1570EXPORT_SYMBOL(blkdev_get_by_dev);
1571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572static int blkdev_open(struct inode * inode, struct file * filp)
1573{
1574 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576 /*
1577 * Preserve backwards compatibility and allow large file access
1578 * even if userspace doesn't ask for it explicitly. Some mkfs
1579 * binary needs it. We might want to drop this workaround
1580 * during an unstable branch.
1581 */
1582 filp->f_flags |= O_LARGEFILE;
1583
Al Viro572c4892007-10-08 13:24:05 -04001584 if (filp->f_flags & O_NDELAY)
1585 filp->f_mode |= FMODE_NDELAY;
1586 if (filp->f_flags & O_EXCL)
1587 filp->f_mode |= FMODE_EXCL;
1588 if ((filp->f_flags & O_ACCMODE) == 3)
1589 filp->f_mode |= FMODE_WRITE_IOCTL;
1590
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001592 if (bdev == NULL)
1593 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
Al Viro572c4892007-10-08 13:24:05 -04001595 filp->f_mapping = bdev->bd_inode->i_mapping;
1596
Tejun Heoe525fd82010-11-13 11:55:17 +01001597 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598}
1599
Al Viro4385bab2013-05-05 22:11:03 -04001600static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001601{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001602 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001603 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001604
NeilBrown6796bf52006-12-08 02:36:16 -08001605 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001606 if (for_part)
1607 bdev->bd_part_count--;
1608
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001609 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001610 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001611 sync_blockdev(bdev);
1612 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001613
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001614 bdev_write_inode(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001615 /*
1616 * Detaching bdev inode from its wb in __destroy_inode()
1617 * is too late: the queue which embeds its bdi (along with
1618 * root wb) can be gone as soon as we put_disk() below.
1619 */
1620 inode_detach_wb(bdev->bd_inode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001621 }
1622 if (bdev->bd_contains == bdev) {
1623 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001624 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001625 }
1626 if (!bdev->bd_openers) {
1627 struct module *owner = disk->fops->owner;
1628
Tejun Heo0762b8b2008-08-25 19:56:12 +09001629 disk_put_part(bdev->bd_part);
1630 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001631 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001632 if (bdev != bdev->bd_contains)
1633 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001634 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001635
1636 put_disk(disk);
1637 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001638 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001639 mutex_unlock(&bdev->bd_mutex);
1640 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001641 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001642 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001643}
1644
Al Viro4385bab2013-05-05 22:11:03 -04001645void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001646{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001647 mutex_lock(&bdev->bd_mutex);
1648
Tejun Heoe525fd82010-11-13 11:55:17 +01001649 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001650 bool bdev_free;
1651
1652 /*
1653 * Release a claim on the device. The holder fields
1654 * are protected with bdev_lock. bd_mutex is to
1655 * synchronize disk_holder unlinking.
1656 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001657 spin_lock(&bdev_lock);
1658
1659 WARN_ON_ONCE(--bdev->bd_holders < 0);
1660 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1661
1662 /* bd_contains might point to self, check in a separate step */
1663 if ((bdev_free = !bdev->bd_holders))
1664 bdev->bd_holder = NULL;
1665 if (!bdev->bd_contains->bd_holders)
1666 bdev->bd_contains->bd_holder = NULL;
1667
1668 spin_unlock(&bdev_lock);
1669
Tejun Heo77ea8872010-12-08 20:57:37 +01001670 /*
1671 * If this was the last claim, remove holder link and
1672 * unblock evpoll if it was a write holder.
1673 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001674 if (bdev_free && bdev->bd_write_holder) {
1675 disk_unblock_events(bdev->bd_disk);
1676 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001677 }
Tejun Heo69362172011-03-09 19:54:27 +01001678 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001679
Tejun Heo85ef06d2011-07-01 16:17:47 +02001680 /*
1681 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1682 * event. This is to ensure detection of media removal commanded
1683 * from userland - e.g. eject(1).
1684 */
1685 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1686
1687 mutex_unlock(&bdev->bd_mutex);
1688
Al Viro4385bab2013-05-05 22:11:03 -04001689 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001690}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001691EXPORT_SYMBOL(blkdev_put);
1692
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693static int blkdev_close(struct inode * inode, struct file * filp)
1694{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001695 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001696 blkdev_put(bdev, filp->f_mode);
1697 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698}
1699
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001700static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001702 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001703 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001704
1705 /*
1706 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1707 * to updated it before every ioctl.
1708 */
Al Viro56b26ad2008-09-19 03:17:36 -04001709 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001710 mode |= FMODE_NDELAY;
1711 else
1712 mode &= ~FMODE_NDELAY;
1713
Al Viro56b26ad2008-09-19 03:17:36 -04001714 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715}
1716
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001717/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001718 * Write data to the block device. Only intended for the block device itself
1719 * and the raw driver which basically is a fake block device.
1720 *
1721 * Does not take i_mutex for the write and thus is not for general purpose
1722 * use.
1723 */
Al Viro1456c0a2014-04-03 03:21:50 -04001724ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001725{
1726 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001727 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001728 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001729 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001730 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001731
Al Viro7ec7b942015-04-07 11:35:14 -04001732 if (bdev_read_only(I_BDEV(bd_inode)))
1733 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001734
Al Viro7ec7b942015-04-07 11:35:14 -04001735 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001736 return 0;
1737
Al Viro7ec7b942015-04-07 11:35:14 -04001738 if (iocb->ki_pos >= size)
1739 return -ENOSPC;
1740
1741 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001742
Jianpeng Ma53362a02012-08-02 09:50:39 +02001743 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001744 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwige2592212016-04-07 08:52:01 -07001745 if (ret > 0)
1746 ret = generic_write_sync(iocb, ret);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001747 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001748 return ret;
1749}
Al Viro1456c0a2014-04-03 03:21:50 -04001750EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001751
David Jefferyb2de5252014-09-29 10:21:10 -04001752ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001753{
1754 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001755 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001756 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001757 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001758
1759 if (pos >= size)
1760 return 0;
1761
1762 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001763 iov_iter_truncate(to, size);
1764 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001765}
David Jefferyb2de5252014-09-29 10:21:10 -04001766EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001767
Christoph Hellwigeef99382009-08-20 17:43:41 +02001768/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001769 * Try to release a page associated with block device when the system
1770 * is under memory pressure.
1771 */
1772static int blkdev_releasepage(struct page *page, gfp_t wait)
1773{
1774 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1775
1776 if (super && super->s_op->bdev_try_to_free_page)
1777 return super->s_op->bdev_try_to_free_page(super, page, wait);
1778
1779 return try_to_free_buffers(page);
1780}
1781
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001782static int blkdev_writepages(struct address_space *mapping,
1783 struct writeback_control *wbc)
1784{
1785 if (dax_mapping(mapping)) {
1786 struct block_device *bdev = I_BDEV(mapping->host);
1787
1788 return dax_writeback_mapping_range(mapping, bdev, wbc);
1789 }
1790 return generic_writepages(mapping, wbc);
1791}
1792
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001793static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001795 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001797 .write_begin = blkdev_write_begin,
1798 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001799 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001800 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001802 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803};
1804
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001805const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 .open = blkdev_open,
1807 .release = blkdev_close,
1808 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001809 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001810 .write_iter = blkdev_write_iter,
Dan Williamsacc93d32016-05-07 11:40:28 -07001811 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001812 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001813 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814#ifdef CONFIG_COMPAT
1815 .compat_ioctl = compat_blkdev_ioctl,
1816#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001817 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001818 .splice_write = iter_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819};
1820
1821int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1822{
1823 int res;
1824 mm_segment_t old_fs = get_fs();
1825 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001826 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 set_fs(old_fs);
1828 return res;
1829}
1830
1831EXPORT_SYMBOL(ioctl_by_bdev);
1832
1833/**
1834 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001835 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001837 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 * namespace if possible and return it. Return ERR_PTR(error)
1839 * otherwise.
1840 */
Al Viro421748e2008-08-02 01:04:36 -04001841struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842{
1843 struct block_device *bdev;
1844 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001845 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 int error;
1847
Al Viro421748e2008-08-02 01:04:36 -04001848 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 return ERR_PTR(-EINVAL);
1850
Al Viro421748e2008-08-02 01:04:36 -04001851 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 if (error)
1853 return ERR_PTR(error);
1854
David Howellsbb6687342015-03-17 22:26:21 +00001855 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 error = -ENOTBLK;
1857 if (!S_ISBLK(inode->i_mode))
1858 goto fail;
1859 error = -EACCES;
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05001860 if (!may_open_dev(&path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 goto fail;
1862 error = -ENOMEM;
1863 bdev = bd_acquire(inode);
1864 if (!bdev)
1865 goto fail;
1866out:
Al Viro421748e2008-08-02 01:04:36 -04001867 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 return bdev;
1869fail:
1870 bdev = ERR_PTR(error);
1871 goto out;
1872}
Al Virod5686b42008-08-01 05:00:11 -04001873EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874
NeilBrown93b270f2011-02-24 17:25:47 +11001875int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001876{
1877 struct super_block *sb = get_super(bdev);
1878 int res = 0;
1879
1880 if (sb) {
1881 /*
1882 * no need to lock the super, get_super holds the
1883 * read mutex so the filesystem cannot go away
1884 * under us (->put_super runs with the write lock
1885 * hold).
1886 */
1887 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001888 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001889 drop_super(sb);
1890 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001891 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001892 return res;
1893}
1894EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001895
1896void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1897{
1898 struct inode *inode, *old_inode = NULL;
1899
Dave Chinner74278da2015-03-04 12:37:22 -05001900 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001901 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1902 struct address_space *mapping = inode->i_mapping;
1903
1904 spin_lock(&inode->i_lock);
1905 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1906 mapping->nrpages == 0) {
1907 spin_unlock(&inode->i_lock);
1908 continue;
1909 }
1910 __iget(inode);
1911 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05001912 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001913 /*
1914 * We hold a reference to 'inode' so it couldn't have been
1915 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05001916 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02001917 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05001918 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02001919 * later.
1920 */
1921 iput(old_inode);
1922 old_inode = inode;
1923
1924 func(I_BDEV(inode), arg);
1925
Dave Chinner74278da2015-03-04 12:37:22 -05001926 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001927 }
Dave Chinner74278da2015-03-04 12:37:22 -05001928 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001929 iput(old_inode);
1930}