blob: 3172c4e2f50255e1cb1fb426c1723a6b7b493bb2 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020033#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35struct bdev_inode {
36 struct block_device bdev;
37 struct inode vfs_inode;
38};
39
Adrian Bunk4c54ac62008-02-18 13:48:31 +010040static const struct address_space_operations def_blk_aops;
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042static inline struct bdev_inode *BDEV_I(struct inode *inode)
43{
44 return container_of(inode, struct bdev_inode, vfs_inode);
45}
46
Geert Uytterhoevenff5053f2015-06-26 13:58:32 +020047struct block_device *I_BDEV(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070048{
49 return &BDEV_I(inode)->bdev;
50}
Linus Torvalds1da177e2005-04-16 15:20:36 -070051EXPORT_SYMBOL(I_BDEV);
52
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070053static void bdev_write_inode(struct block_device *bdev)
Christoph Hellwig564f00f2015-01-14 10:42:33 +010054{
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070055 struct inode *inode = bdev->bd_inode;
56 int ret;
57
Christoph Hellwig564f00f2015-01-14 10:42:33 +010058 spin_lock(&inode->i_lock);
59 while (inode->i_state & I_DIRTY) {
60 spin_unlock(&inode->i_lock);
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070061 ret = write_inode_now(inode, true);
62 if (ret) {
63 char name[BDEVNAME_SIZE];
64 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
65 "for block device %s (err=%d).\n",
66 bdevname(bdev, name), ret);
67 }
Christoph Hellwig564f00f2015-01-14 10:42:33 +010068 spin_lock(&inode->i_lock);
69 }
70 spin_unlock(&inode->i_lock);
71}
72
Peter Zijlstraf9a14392007-05-06 14:49:55 -070073/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb482011-09-16 02:31:11 -040074void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Al Viroff01bb482011-09-16 02:31:11 -040076 struct address_space *mapping = bdev->bd_inode->i_mapping;
77
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080078 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070079 return;
Al Viroff01bb482011-09-16 02:31:11 -040080
Peter Zijlstraf9a14392007-05-06 14:49:55 -070081 invalidate_bh_lrus();
Al Viroff01bb482011-09-16 02:31:11 -040082 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083}
Al Viroff01bb482011-09-16 02:31:11 -040084EXPORT_SYMBOL(kill_bdev);
85
86/* Invalidate clean unused buffers and pagecache. */
87void invalidate_bdev(struct block_device *bdev)
88{
89 struct address_space *mapping = bdev->bd_inode->i_mapping;
90
91 if (mapping->nrpages == 0)
92 return;
93
94 invalidate_bh_lrus();
95 lru_add_drain_all(); /* make sure all lru add caches are flushed */
96 invalidate_mapping_pages(mapping, 0, -1);
97 /* 99% of the time, we don't need to flush the cleancache on the bdev.
98 * But, for the strange corners, lets be cautious
99 */
Dan Magenheimer31677602011-09-21 11:56:28 -0400100 cleancache_invalidate_inode(mapping);
Al Viroff01bb482011-09-16 02:31:11 -0400101}
102EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104int set_blocksize(struct block_device *bdev, int size)
105{
106 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700107 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 return -EINVAL;
109
110 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400111 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 return -EINVAL;
113
114 /* Don't change the size if it is same as current */
115 if (bdev->bd_block_size != size) {
116 sync_blockdev(bdev);
117 bdev->bd_block_size = size;
118 bdev->bd_inode->i_blkbits = blksize_bits(size);
119 kill_bdev(bdev);
120 }
121 return 0;
122}
123
124EXPORT_SYMBOL(set_blocksize);
125
126int sb_set_blocksize(struct super_block *sb, int size)
127{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 if (set_blocksize(sb->s_bdev, size))
129 return 0;
130 /* If we get here, we know size is power of two
131 * and it's value is between 512 and PAGE_SIZE */
132 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800133 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 return sb->s_blocksize;
135}
136
137EXPORT_SYMBOL(sb_set_blocksize);
138
139int sb_min_blocksize(struct super_block *sb, int size)
140{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400141 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 if (size < minsize)
143 size = minsize;
144 return sb_set_blocksize(sb, size);
145}
146
147EXPORT_SYMBOL(sb_min_blocksize);
148
149static int
150blkdev_get_block(struct inode *inode, sector_t iblock,
151 struct buffer_head *bh, int create)
152{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 bh->b_bdev = I_BDEV(inode);
154 bh->b_blocknr = iblock;
155 set_buffer_mapped(bh);
156 return 0;
157}
158
Dan Williams4ebb16c2015-10-28 07:48:19 +0900159static struct inode *bdev_file_inode(struct file *file)
160{
161 return file->f_mapping->host;
162}
163
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800164static ssize_t
Omar Sandoval22c61862015-03-16 04:33:53 -0700165blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800166{
167 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +0900168 struct inode *inode = bdev_file_inode(file);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800169
Matthew Wilcoxbbab37d2015-07-03 10:40:42 -0400170 if (IS_DAX(inode))
171 return dax_do_io(iocb, inode, iter, offset, blkdev_get_block,
172 NULL, DIO_SKIP_DIO_COUNT);
Omar Sandoval17f8c842015-03-16 04:33:50 -0700173 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset,
Jens Axboefe0f07d2015-04-15 17:05:48 -0600174 blkdev_get_block, NULL, NULL,
175 DIO_SKIP_DIO_COUNT);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800176}
177
Jan Kara5cee5812009-04-27 16:43:51 +0200178int __sync_blockdev(struct block_device *bdev, int wait)
179{
180 if (!bdev)
181 return 0;
182 if (!wait)
183 return filemap_flush(bdev->bd_inode->i_mapping);
184 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
185}
186
Nick Piggin585d3bc2009-02-25 10:44:19 +0100187/*
188 * Write out and wait upon all the dirty data associated with a block
189 * device via its mapping. Does not take the superblock lock.
190 */
191int sync_blockdev(struct block_device *bdev)
192{
Jan Kara5cee5812009-04-27 16:43:51 +0200193 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100194}
195EXPORT_SYMBOL(sync_blockdev);
196
197/*
198 * Write out and wait upon all dirty data associated with this
199 * device. Filesystem data as well as the underlying block
200 * device. Takes the superblock lock.
201 */
202int fsync_bdev(struct block_device *bdev)
203{
204 struct super_block *sb = get_super(bdev);
205 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200206 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100207 drop_super(sb);
208 return res;
209 }
210 return sync_blockdev(bdev);
211}
Al Viro47e44912009-04-01 07:07:16 -0400212EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100213
214/**
215 * freeze_bdev -- lock a filesystem and force it into a consistent state
216 * @bdev: blockdevice to lock
217 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100218 * If a superblock is found on this device, we take the s_umount semaphore
219 * on it to make sure nobody unmounts until the snapshot creation is done.
220 * The reference counter (bd_fsfreeze_count) guarantees that only the last
221 * unfreeze process can unfreeze the frozen filesystem actually when multiple
222 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
223 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
224 * actually.
225 */
226struct super_block *freeze_bdev(struct block_device *bdev)
227{
228 struct super_block *sb;
229 int error = 0;
230
231 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200232 if (++bdev->bd_fsfreeze_count > 1) {
233 /*
234 * We don't even need to grab a reference - the first call
235 * to freeze_bdev grab an active reference and only the last
236 * thaw_bdev drops it.
237 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100238 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200239 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100240 mutex_unlock(&bdev->bd_fsfreeze_mutex);
241 return sb;
242 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100243
Christoph Hellwig45042302009-08-03 23:28:35 +0200244 sb = get_active_super(bdev);
245 if (!sb)
246 goto out;
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600247 if (sb->s_op->freeze_super)
248 error = sb->s_op->freeze_super(sb);
249 else
250 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400251 if (error) {
252 deactivate_super(sb);
253 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200254 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400255 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100256 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400257 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200258 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100259 sync_blockdev(bdev);
260 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200261 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100262}
263EXPORT_SYMBOL(freeze_bdev);
264
265/**
266 * thaw_bdev -- unlock filesystem
267 * @bdev: blockdevice to unlock
268 * @sb: associated superblock
269 *
270 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
271 */
272int thaw_bdev(struct block_device *bdev, struct super_block *sb)
273{
Christoph Hellwig45042302009-08-03 23:28:35 +0200274 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100275
276 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200277 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400278 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100279
Christoph Hellwig45042302009-08-03 23:28:35 +0200280 error = 0;
281 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400282 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100283
Christoph Hellwig45042302009-08-03 23:28:35 +0200284 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400285 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200286
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600287 if (sb->s_op->thaw_super)
288 error = sb->s_op->thaw_super(sb);
289 else
290 error = thaw_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400291 if (error) {
292 bdev->bd_fsfreeze_count++;
293 mutex_unlock(&bdev->bd_fsfreeze_mutex);
294 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100295 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400296out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100297 mutex_unlock(&bdev->bd_fsfreeze_mutex);
298 return 0;
299}
300EXPORT_SYMBOL(thaw_bdev);
301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
303{
304 return block_write_full_page(page, blkdev_get_block, wbc);
305}
306
307static int blkdev_readpage(struct file * file, struct page * page)
308{
309 return block_read_full_page(page, blkdev_get_block);
310}
311
Akinobu Mita447f05b2014-10-09 15:26:58 -0700312static int blkdev_readpages(struct file *file, struct address_space *mapping,
313 struct list_head *pages, unsigned nr_pages)
314{
315 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
316}
317
Nick Piggin6272b5a2007-10-16 01:25:04 -0700318static int blkdev_write_begin(struct file *file, struct address_space *mapping,
319 loff_t pos, unsigned len, unsigned flags,
320 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200322 return block_write_begin(mapping, pos, len, flags, pagep,
323 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324}
325
Nick Piggin6272b5a2007-10-16 01:25:04 -0700326static int blkdev_write_end(struct file *file, struct address_space *mapping,
327 loff_t pos, unsigned len, unsigned copied,
328 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700330 int ret;
331 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
332
333 unlock_page(page);
334 page_cache_release(page);
335
336 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337}
338
339/*
340 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500341 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 * so we compute the size by hand (just as in block_read/write above)
343 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800344static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900346 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 loff_t retval;
348
Al Viro59551022016-01-22 15:40:57 -0500349 inode_lock(bd_inode);
Al Viro5d48f3a2013-06-23 21:34:45 +0400350 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Al Viro59551022016-01-22 15:40:57 -0500351 inode_unlock(bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 return retval;
353}
354
Josef Bacik02c24a82011-07-16 20:44:56 -0400355int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900357 struct inode *bd_inode = bdev_file_inode(filp);
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400358 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100359 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200360
361 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
362 if (error)
363 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100364
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400365 /*
366 * There is no need to serialise calls to blkdev_issue_flush with
367 * i_mutex and doing so causes performance issues with concurrent
368 * O_SYNC writers to a block device.
369 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200370 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100371 if (error == -EOPNOTSUPP)
372 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400373
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100374 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700376EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700378/**
379 * bdev_read_page() - Start reading a page from a block device
380 * @bdev: The device to read the page from
381 * @sector: The offset on the device to read the page to (need not be aligned)
382 * @page: The page to read
383 *
384 * On entry, the page should be locked. It will be unlocked when the page
385 * has been read. If the block driver implements rw_page synchronously,
386 * that will be true on exit from this function, but it need not be.
387 *
388 * Errors returned by this function are usually "soft", eg out of memory, or
389 * queue full; callers should try a different route to read this page rather
390 * than propagate an error back up the stack.
391 *
392 * Return: negative errno if an error occurs, 0 if submission was successful.
393 */
394int bdev_read_page(struct block_device *bdev, sector_t sector,
395 struct page *page)
396{
397 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800398 int result = -EOPNOTSUPP;
399
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400400 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc92015-11-19 13:29:28 -0800401 return result;
402
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100403 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800404 if (result)
405 return result;
406 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
407 blk_queue_exit(bdev->bd_queue);
408 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700409}
410EXPORT_SYMBOL_GPL(bdev_read_page);
411
412/**
413 * bdev_write_page() - Start writing a page to a block device
414 * @bdev: The device to write the page to
415 * @sector: The offset on the device to write the page to (need not be aligned)
416 * @page: The page to write
417 * @wbc: The writeback_control for the write
418 *
419 * On entry, the page should be locked and not currently under writeback.
420 * On exit, if the write started successfully, the page will be unlocked and
421 * under writeback. If the write failed already (eg the driver failed to
422 * queue the page to the device), the page will still be locked. If the
423 * caller is a ->writepage implementation, it will need to unlock the page.
424 *
425 * Errors returned by this function are usually "soft", eg out of memory, or
426 * queue full; callers should try a different route to write this page rather
427 * than propagate an error back up the stack.
428 *
429 * Return: negative errno if an error occurs, 0 if submission was successful.
430 */
431int bdev_write_page(struct block_device *bdev, sector_t sector,
432 struct page *page, struct writeback_control *wbc)
433{
434 int result;
435 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
436 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800437
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400438 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700439 return -EOPNOTSUPP;
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100440 result = blk_queue_enter(bdev->bd_queue, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800441 if (result)
442 return result;
443
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700444 set_page_writeback(page);
445 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
446 if (result)
447 end_page_writeback(page);
448 else
449 unlock_page(page);
Dan Williams2e6edc92015-11-19 13:29:28 -0800450 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700451 return result;
452}
453EXPORT_SYMBOL_GPL(bdev_write_page);
454
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200455/**
456 * bdev_direct_access() - Get the address for directly-accessibly memory
457 * @bdev: The device containing the memory
Dan Williamsb2e0d162016-01-15 16:55:59 -0800458 * @dax: control and output parameters for ->direct_access
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200459 *
460 * If a block device is made up of directly addressable memory, this function
461 * will tell the caller the PFN and the address of the memory. The address
462 * may be directly dereferenced within the kernel without the need to call
463 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
464 * page tables.
465 *
466 * Return: negative errno if an error occurs, otherwise the number of bytes
467 * accessible at this address.
468 */
Dan Williamsb2e0d162016-01-15 16:55:59 -0800469long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200470{
Dan Williamsb2e0d162016-01-15 16:55:59 -0800471 sector_t sector = dax->sector;
472 long avail, size = dax->size;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200473 const struct block_device_operations *ops = bdev->bd_disk->fops;
474
Matthew Wilcox43c3dd02015-07-03 10:40:43 -0400475 /*
476 * The device driver is allowed to sleep, in order to make the
477 * memory directly accessible.
478 */
479 might_sleep();
480
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200481 if (size < 0)
482 return size;
483 if (!ops->direct_access)
484 return -EOPNOTSUPP;
485 if ((sector + DIV_ROUND_UP(size, 512)) >
486 part_nr_sects_read(bdev->bd_part))
487 return -ERANGE;
488 sector += get_start_sect(bdev);
489 if (sector % (PAGE_SIZE / 512))
490 return -EINVAL;
Dan Williamsb2e0d162016-01-15 16:55:59 -0800491 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn);
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200492 if (!avail)
493 return -ERANGE;
Dan Williamsfe683ad2016-01-15 16:55:56 -0800494 if (avail > 0 && avail & ~PAGE_MASK)
495 return -ENXIO;
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200496 return min(avail, size);
497}
498EXPORT_SYMBOL_GPL(bdev_direct_access);
499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500/*
501 * pseudo-fs
502 */
503
504static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800505static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
507static struct inode *bdev_alloc_inode(struct super_block *sb)
508{
Christoph Lametere94b1762006-12-06 20:33:17 -0800509 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 if (!ei)
511 return NULL;
512 return &ei->vfs_inode;
513}
514
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100515static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100517 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 struct bdev_inode *bdi = BDEV_I(inode);
519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 kmem_cache_free(bdev_cachep, bdi);
521}
522
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100523static void bdev_destroy_inode(struct inode *inode)
524{
525 call_rcu(&inode->i_rcu, bdev_i_callback);
526}
527
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700528static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529{
530 struct bdev_inode *ei = (struct bdev_inode *) foo;
531 struct block_device *bdev = &ei->bdev;
532
Christoph Lametera35afb82007-05-16 22:10:57 -0700533 memset(bdev, 0, sizeof(*bdev));
534 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700535 INIT_LIST_HEAD(&bdev->bd_inodes);
536 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100537#ifdef CONFIG_SYSFS
538 INIT_LIST_HEAD(&bdev->bd_holder_disks);
539#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700540 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800541 /* Initialize mutex for freeze. */
542 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543}
544
545static inline void __bd_forget(struct inode *inode)
546{
547 list_del_init(&inode->i_devices);
548 inode->i_bdev = NULL;
549 inode->i_mapping = &inode->i_data;
550}
551
Al Virob57922d2010-06-07 14:34:48 -0400552static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553{
554 struct block_device *bdev = &BDEV_I(inode)->bdev;
555 struct list_head *p;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700556 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400557 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200558 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 spin_lock(&bdev_lock);
560 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
561 __bd_forget(list_entry(p, struct inode, i_devices));
562 }
563 list_del_init(&bdev->bd_list);
564 spin_unlock(&bdev_lock);
565}
566
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800567static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 .statfs = simple_statfs,
569 .alloc_inode = bdev_alloc_inode,
570 .destroy_inode = bdev_destroy_inode,
571 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400572 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573};
574
Al Viro51139ad2010-07-25 23:47:46 +0400575static struct dentry *bd_mount(struct file_system_type *fs_type,
576 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
Shaohua Li3684aa72016-02-22 15:27:40 -0700578 struct dentry *dent;
579 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
580 if (dent)
581 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
582 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583}
584
585static struct file_system_type bd_type = {
586 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400587 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 .kill_sb = kill_anon_super,
589};
590
Tejun Heoa212b102015-05-22 17:13:33 -0400591struct super_block *blockdev_superblock __read_mostly;
592EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594void __init bdev_cache_init(void)
595{
596 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300597 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800600 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800601 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900602 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 err = register_filesystem(&bd_type);
604 if (err)
605 panic("Cannot register bdev pseudo-fs");
606 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 if (IS_ERR(bd_mnt))
608 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300609 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}
611
612/*
613 * Most likely _very_ bad one - but then it's hardly critical for small
614 * /dev and can be fixed when somebody will need really large one.
615 * Keep in mind that it will be fed through icache hash function too.
616 */
617static inline unsigned long hash(dev_t dev)
618{
619 return MAJOR(dev)+MINOR(dev);
620}
621
622static int bdev_test(struct inode *inode, void *data)
623{
624 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
625}
626
627static int bdev_set(struct inode *inode, void *data)
628{
629 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
630 return 0;
631}
632
633static LIST_HEAD(all_bdevs);
634
635struct block_device *bdget(dev_t dev)
636{
637 struct block_device *bdev;
638 struct inode *inode;
639
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800640 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 bdev_test, bdev_set, &dev);
642
643 if (!inode)
644 return NULL;
645
646 bdev = &BDEV_I(inode)->bdev;
647
648 if (inode->i_state & I_NEW) {
649 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000650 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 bdev->bd_inode = inode;
652 bdev->bd_block_size = (1 << inode->i_blkbits);
653 bdev->bd_part_count = 0;
654 bdev->bd_invalidated = 0;
655 inode->i_mode = S_IFBLK;
656 inode->i_rdev = dev;
657 inode->i_bdev = bdev;
658 inode->i_data.a_ops = &def_blk_aops;
659 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 spin_lock(&bdev_lock);
661 list_add(&bdev->bd_list, &all_bdevs);
662 spin_unlock(&bdev_lock);
663 unlock_new_inode(inode);
664 }
665 return bdev;
666}
667
668EXPORT_SYMBOL(bdget);
669
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200670/**
671 * bdgrab -- Grab a reference to an already referenced block device
672 * @bdev: Block device to grab a reference to.
673 */
674struct block_device *bdgrab(struct block_device *bdev)
675{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400676 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200677 return bdev;
678}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700679EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681long nr_blockdev_pages(void)
682{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700683 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 long ret = 0;
685 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700686 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 ret += bdev->bd_inode->i_mapping->nrpages;
688 }
689 spin_unlock(&bdev_lock);
690 return ret;
691}
692
693void bdput(struct block_device *bdev)
694{
695 iput(bdev->bd_inode);
696}
697
698EXPORT_SYMBOL(bdput);
699
700static struct block_device *bd_acquire(struct inode *inode)
701{
702 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 spin_lock(&bdev_lock);
705 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700706 if (bdev) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100707 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 spin_unlock(&bdev_lock);
709 return bdev;
710 }
711 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 bdev = bdget(inode->i_rdev);
714 if (bdev) {
715 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700716 if (!inode->i_bdev) {
717 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400718 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700719 * and it's released in clear_inode() of inode.
720 * So, we can access it via ->i_mapping always
721 * without igrab().
722 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100723 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700724 inode->i_bdev = bdev;
725 inode->i_mapping = bdev->bd_inode->i_mapping;
726 list_add(&inode->i_devices, &bdev->bd_inodes);
727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 spin_unlock(&bdev_lock);
729 }
730 return bdev;
731}
732
733/* Call when you free inode */
734
735void bd_forget(struct inode *inode)
736{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700737 struct block_device *bdev = NULL;
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700740 if (!sb_is_blkdev_sb(inode->i_sb))
741 bdev = inode->i_bdev;
742 __bd_forget(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700744
745 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100746 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
748
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900749/**
750 * bd_may_claim - test whether a block device can be claimed
751 * @bdev: block device of interest
752 * @whole: whole block device containing @bdev, may equal @bdev
753 * @holder: holder trying to claim @bdev
754 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300755 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900756 *
757 * CONTEXT:
758 * spin_lock(&bdev_lock).
759 *
760 * RETURNS:
761 * %true if @bdev can be claimed, %false otherwise.
762 */
763static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
764 void *holder)
765{
766 if (bdev->bd_holder == holder)
767 return true; /* already a holder */
768 else if (bdev->bd_holder != NULL)
769 return false; /* held by someone else */
770 else if (bdev->bd_contains == bdev)
771 return true; /* is a whole device which isn't held */
772
Tejun Heoe525fd82010-11-13 11:55:17 +0100773 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900774 return true; /* is a partition of a device that is being partitioned */
775 else if (whole->bd_holder != NULL)
776 return false; /* is a partition of a held device */
777 else
778 return true; /* is a partition of an un-held device */
779}
780
781/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900782 * bd_prepare_to_claim - prepare to claim a block device
783 * @bdev: block device of interest
784 * @whole: the whole device containing @bdev, may equal @bdev
785 * @holder: holder trying to claim @bdev
786 *
787 * Prepare to claim @bdev. This function fails if @bdev is already
788 * claimed by another holder and waits if another claiming is in
789 * progress. This function doesn't actually claim. On successful
790 * return, the caller has ownership of bd_claiming and bd_holder[s].
791 *
792 * CONTEXT:
793 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
794 * it multiple times.
795 *
796 * RETURNS:
797 * 0 if @bdev can be claimed, -EBUSY otherwise.
798 */
799static int bd_prepare_to_claim(struct block_device *bdev,
800 struct block_device *whole, void *holder)
801{
802retry:
803 /* if someone else claimed, fail */
804 if (!bd_may_claim(bdev, whole, holder))
805 return -EBUSY;
806
Tejun Heoe75aa852010-08-04 17:59:39 +0200807 /* if claiming is already in progress, wait for it to finish */
808 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900809 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
810 DEFINE_WAIT(wait);
811
812 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
813 spin_unlock(&bdev_lock);
814 schedule();
815 finish_wait(wq, &wait);
816 spin_lock(&bdev_lock);
817 goto retry;
818 }
819
820 /* yay, all mine */
821 return 0;
822}
823
824/**
825 * bd_start_claiming - start claiming a block device
826 * @bdev: block device of interest
827 * @holder: holder trying to claim @bdev
828 *
829 * @bdev is about to be opened exclusively. Check @bdev can be opened
830 * exclusively and mark that an exclusive open is in progress. Each
831 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000832 * either bd_finish_claiming() or bd_abort_claiming() (which do not
833 * fail).
834 *
835 * This function is used to gain exclusive access to the block device
836 * without actually causing other exclusive open attempts to fail. It
837 * should be used when the open sequence itself requires exclusive
838 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900839 *
840 * CONTEXT:
841 * Might sleep.
842 *
843 * RETURNS:
844 * Pointer to the block device containing @bdev on success, ERR_PTR()
845 * value on failure.
846 */
847static struct block_device *bd_start_claiming(struct block_device *bdev,
848 void *holder)
849{
850 struct gendisk *disk;
851 struct block_device *whole;
852 int partno, err;
853
854 might_sleep();
855
856 /*
857 * @bdev might not have been initialized properly yet, look up
858 * and grab the outer block device the hard way.
859 */
860 disk = get_gendisk(bdev->bd_dev, &partno);
861 if (!disk)
862 return ERR_PTR(-ENXIO);
863
Tejun Heod4c208b2011-06-13 12:45:48 +0200864 /*
865 * Normally, @bdev should equal what's returned from bdget_disk()
866 * if partno is 0; however, some drivers (floppy) use multiple
867 * bdev's for the same physical device and @bdev may be one of the
868 * aliases. Keep @bdev if partno is 0. This means claimer
869 * tracking is broken for those devices but it has always been that
870 * way.
871 */
872 if (partno)
873 whole = bdget_disk(disk, 0);
874 else
875 whole = bdgrab(bdev);
876
Nick Piggincf342572010-05-26 01:50:21 +1000877 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900878 put_disk(disk);
879 if (!whole)
880 return ERR_PTR(-ENOMEM);
881
882 /* prepare to claim, if successful, mark claiming in progress */
883 spin_lock(&bdev_lock);
884
885 err = bd_prepare_to_claim(bdev, whole, holder);
886 if (err == 0) {
887 whole->bd_claiming = holder;
888 spin_unlock(&bdev_lock);
889 return whole;
890 } else {
891 spin_unlock(&bdev_lock);
892 bdput(whole);
893 return ERR_PTR(err);
894 }
895}
896
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800897#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100898struct bd_holder_disk {
899 struct list_head list;
900 struct gendisk *disk;
901 int refcnt;
902};
903
904static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
905 struct gendisk *disk)
906{
907 struct bd_holder_disk *holder;
908
909 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
910 if (holder->disk == disk)
911 return holder;
912 return NULL;
913}
914
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700915static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800916{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700917 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800918}
919
920static void del_symlink(struct kobject *from, struct kobject *to)
921{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800922 sysfs_remove_link(from, kobject_name(to));
923}
924
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800925/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100926 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
927 * @bdev: the claimed slave bdev
928 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500929 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100930 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
931 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100932 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500933 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100934 * - from "slaves" directory of the holder @disk to the claimed @bdev
935 * - from "holders" directory of the @bdev to the holder @disk
936 *
937 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
938 * passed to bd_link_disk_holder(), then:
939 *
940 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
941 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
942 *
943 * The caller must have claimed @bdev before calling this function and
944 * ensure that both @bdev and @disk are valid during the creation and
945 * lifetime of these symlinks.
946 *
947 * CONTEXT:
948 * Might sleep.
949 *
950 * RETURNS:
951 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500952 */
Tejun Heoe09b4572010-11-13 11:55:17 +0100953int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500954{
Tejun Heo49731ba2011-01-14 18:43:57 +0100955 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100956 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800957
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800958 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500959
Tejun Heo49731ba2011-01-14 18:43:57 +0100960 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -0700961
Tejun Heoe09b4572010-11-13 11:55:17 +0100962 /* FIXME: remove the following once add_disk() handles errors */
963 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
964 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -0700965
Tejun Heo49731ba2011-01-14 18:43:57 +0100966 holder = bd_find_holder_disk(bdev, disk);
967 if (holder) {
968 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +0100969 goto out_unlock;
970 }
971
Tejun Heo49731ba2011-01-14 18:43:57 +0100972 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
973 if (!holder) {
974 ret = -ENOMEM;
975 goto out_unlock;
976 }
977
978 INIT_LIST_HEAD(&holder->list);
979 holder->disk = disk;
980 holder->refcnt = 1;
981
982 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
983 if (ret)
984 goto out_free;
985
986 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
987 if (ret)
988 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +0100989 /*
990 * bdev could be deleted beneath us which would implicitly destroy
991 * the holder directory. Hold on to it.
992 */
993 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100994
995 list_add(&holder->list, &bdev->bd_holder_disks);
996 goto out_unlock;
997
998out_del:
999 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1000out_free:
1001 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001002out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001003 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001004 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001005}
Tejun Heoe09b4572010-11-13 11:55:17 +01001006EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001007
Tejun Heo49731ba2011-01-14 18:43:57 +01001008/**
1009 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1010 * @bdev: the calimed slave bdev
1011 * @disk: the holding disk
1012 *
1013 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1014 *
1015 * CONTEXT:
1016 * Might sleep.
1017 */
1018void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001019{
Tejun Heo49731ba2011-01-14 18:43:57 +01001020 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001021
Tejun Heo49731ba2011-01-14 18:43:57 +01001022 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001023
Tejun Heo49731ba2011-01-14 18:43:57 +01001024 holder = bd_find_holder_disk(bdev, disk);
1025
1026 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1027 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1028 del_symlink(bdev->bd_part->holder_dir,
1029 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001030 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001031 list_del_init(&holder->list);
1032 kfree(holder);
1033 }
1034
1035 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001036}
Tejun Heo49731ba2011-01-14 18:43:57 +01001037EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001038#endif
1039
Andrew Patterson0c002c22008-09-04 14:27:20 -06001040/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001041 * flush_disk - invalidates all buffer-cache entries on a disk
1042 *
1043 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001044 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001045 *
1046 * Invalidates all buffer-cache entries on a disk. It should be called
1047 * when a disk has been changed -- either by a media change or online
1048 * resize.
1049 */
NeilBrown93b270f2011-02-24 17:25:47 +11001050static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001051{
NeilBrown93b270f2011-02-24 17:25:47 +11001052 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001053 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001054 "resized disk %s\n",
1055 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001056 }
1057
1058 if (!bdev->bd_disk)
1059 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001060 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001061 bdev->bd_invalidated = 1;
1062}
1063
1064/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001065 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001066 * @disk: struct gendisk to check
1067 * @bdev: struct bdev to adjust.
1068 *
1069 * This routine checks to see if the bdev size does not match the disk size
1070 * and adjusts it if it differs.
1071 */
1072void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1073{
1074 loff_t disk_size, bdev_size;
1075
1076 disk_size = (loff_t)get_capacity(disk) << 9;
1077 bdev_size = i_size_read(bdev->bd_inode);
1078 if (disk_size != bdev_size) {
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001079 printk(KERN_INFO
1080 "%s: detected capacity change from %lld to %lld\n",
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001081 disk->disk_name, bdev_size, disk_size);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001082 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001083 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001084 }
1085}
1086EXPORT_SYMBOL(check_disk_size_change);
1087
1088/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001089 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001090 * @disk: struct gendisk to be revalidated
1091 *
1092 * This routine is a wrapper for lower-level driver's revalidate_disk
1093 * call-backs. It is used to do common pre and post operations needed
1094 * for all revalidate_disk operations.
1095 */
1096int revalidate_disk(struct gendisk *disk)
1097{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001098 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001099 int ret = 0;
1100
1101 if (disk->fops->revalidate_disk)
1102 ret = disk->fops->revalidate_disk(disk);
Martin K. Petersen25520d52015-10-21 13:19:49 -04001103 blk_integrity_revalidate(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001104 bdev = bdget_disk(disk, 0);
1105 if (!bdev)
1106 return ret;
1107
1108 mutex_lock(&bdev->bd_mutex);
1109 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001110 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001111 mutex_unlock(&bdev->bd_mutex);
1112 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001113 return ret;
1114}
1115EXPORT_SYMBOL(revalidate_disk);
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117/*
1118 * This routine checks whether a removable media has been changed,
1119 * and invalidates all buffer-cache-entries in that case. This
1120 * is a relatively slow routine, so we have to try to minimize using
1121 * it. Thus it is called only upon a 'mount' or 'open'. This
1122 * is the best way of combining speed and utility, I think.
1123 * People changing diskettes in the middle of an operation deserve
1124 * to lose :-)
1125 */
1126int check_disk_change(struct block_device *bdev)
1127{
1128 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001129 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001130 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Tejun Heo77ea8872010-12-08 20:57:37 +01001132 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1133 DISK_EVENT_EJECT_REQUEST);
1134 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 return 0;
1136
NeilBrown93b270f2011-02-24 17:25:47 +11001137 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 if (bdops->revalidate_disk)
1139 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 return 1;
1141}
1142
1143EXPORT_SYMBOL(check_disk_change);
1144
1145void bd_set_size(struct block_device *bdev, loff_t size)
1146{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001147 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Al Viro59551022016-01-22 15:40:57 -05001149 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001150 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001151 inode_unlock(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 while (bsize < PAGE_CACHE_SIZE) {
1153 if (size & bsize)
1154 break;
1155 bsize <<= 1;
1156 }
1157 bdev->bd_block_size = bsize;
1158 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1159}
1160EXPORT_SYMBOL(bd_set_size);
1161
Al Viro4385bab2013-05-05 22:11:03 -04001162static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001163
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001164/*
1165 * bd_mutex locking:
1166 *
1167 * mutex_lock(part->bd_mutex)
1168 * mutex_lock_nested(whole->bd_mutex, 1)
1169 */
1170
Al Viro572c4892007-10-08 13:24:05 -04001171static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001174 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001175 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001176 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001177 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
Al Viro572c4892007-10-08 13:24:05 -04001179 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001180 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001181 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001182 perm |= MAY_WRITE;
1183 /*
1184 * hooks: /n/, see "layering violations".
1185 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001186 if (!for_part) {
1187 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1188 if (ret != 0) {
1189 bdput(bdev);
1190 return ret;
1191 }
Al Viro82666022008-08-01 05:32:04 -04001192 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001193
NeilBrownd3374822009-01-09 08:31:10 +11001194 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001195
Tejun Heo89f97492008-11-05 10:21:06 +01001196 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001197 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001198 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001199 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001200 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
Tejun Heo69e02c52011-03-09 19:54:27 +01001202 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001203 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 if (!bdev->bd_openers) {
1205 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001206 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 bdev->bd_contains = bdev;
Dan Williams03cdadb2016-02-26 15:19:43 -08001208 if (IS_ENABLED(CONFIG_BLK_DEV_DAX) && disk->fops->direct_access)
1209 bdev->bd_inode->i_flags = S_DAX;
1210 else
1211 bdev->bd_inode->i_flags = 0;
1212
Tejun Heocf771cb2008-09-03 09:01:09 +02001213 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001214 ret = -ENXIO;
1215 bdev->bd_part = disk_get_part(disk, partno);
1216 if (!bdev->bd_part)
1217 goto out_clear;
1218
Tejun Heo1196f8b2011-04-21 20:54:45 +02001219 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001221 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001222 if (ret == -ERESTARTSYS) {
1223 /* Lost a race with 'disk' being
1224 * deleted, try again.
1225 * See md.c
1226 */
1227 disk_put_part(bdev->bd_part);
1228 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001229 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001230 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001231 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001232 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001233 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001234 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001235 goto restart;
1236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 }
Tejun Heo7e697232011-05-23 13:26:07 +02001238
Dan Williams5a023cd2015-11-30 10:20:29 -08001239 if (!ret) {
Tejun Heo7e697232011-05-23 13:26:07 +02001240 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Dan Williams5a023cd2015-11-30 10:20:29 -08001241 if (!blkdev_dax_capable(bdev))
1242 bdev->bd_inode->i_flags &= ~S_DAX;
1243 }
Tejun Heo7e697232011-05-23 13:26:07 +02001244
Tejun Heo1196f8b2011-04-21 20:54:45 +02001245 /*
1246 * If the device is invalidated, rescan partition
1247 * if open succeeded or failed with -ENOMEDIUM.
1248 * The latter is necessary to prevent ghost
1249 * partitions on a removed medium.
1250 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001251 if (bdev->bd_invalidated) {
1252 if (!ret)
1253 rescan_partitions(disk, bdev);
1254 else if (ret == -ENOMEDIUM)
1255 invalidate_partitions(disk, bdev);
1256 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001257
Tejun Heo1196f8b2011-04-21 20:54:45 +02001258 if (ret)
1259 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 struct block_device *whole;
1262 whole = bdget_disk(disk, 0);
1263 ret = -ENOMEM;
1264 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001265 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001266 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001267 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001269 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001271 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001272 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001273 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001275 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 }
Tejun Heo89f97492008-11-05 10:21:06 +01001277 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Dan Williams5a023cd2015-11-30 10:20:29 -08001278 if (!blkdev_dax_capable(bdev))
Jeff Moyerf0b2e562015-08-14 16:15:32 -04001279 bdev->bd_inode->i_flags &= ~S_DAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 }
1281 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001283 ret = 0;
1284 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001285 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001286 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001287 if (bdev->bd_invalidated) {
1288 if (!ret)
1289 rescan_partitions(bdev->bd_disk, bdev);
1290 else if (ret == -ENOMEDIUM)
1291 invalidate_partitions(bdev->bd_disk, bdev);
1292 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001293 if (ret)
1294 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001296 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001297 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001298 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 }
1300 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001301 if (for_part)
1302 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001303 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001304 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 return 0;
1306
Tejun Heo0762b8b2008-08-25 19:56:12 +09001307 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001308 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001310 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001311 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001313 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001315 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001316 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001317 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001318 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001319 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001320 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001321 bdput(bdev);
1322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 return ret;
1324}
1325
Tejun Heod4d77622010-11-13 11:55:18 +01001326/**
1327 * blkdev_get - open a block device
1328 * @bdev: block_device to open
1329 * @mode: FMODE_* mask
1330 * @holder: exclusive holder identifier
1331 *
1332 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1333 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1334 * @holder is invalid. Exclusive opens may nest for the same @holder.
1335 *
1336 * On success, the reference count of @bdev is unchanged. On failure,
1337 * @bdev is put.
1338 *
1339 * CONTEXT:
1340 * Might sleep.
1341 *
1342 * RETURNS:
1343 * 0 on success, -errno on failure.
1344 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001345int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346{
Tejun Heoe525fd82010-11-13 11:55:17 +01001347 struct block_device *whole = NULL;
1348 int res;
1349
1350 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1351
1352 if ((mode & FMODE_EXCL) && holder) {
1353 whole = bd_start_claiming(bdev, holder);
1354 if (IS_ERR(whole)) {
1355 bdput(bdev);
1356 return PTR_ERR(whole);
1357 }
1358 }
1359
1360 res = __blkdev_get(bdev, mode, 0);
1361
1362 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001363 struct gendisk *disk = whole->bd_disk;
1364
Tejun Heo6a027ef2010-11-13 11:55:17 +01001365 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001366 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001367 spin_lock(&bdev_lock);
1368
Tejun Heo77ea8872010-12-08 20:57:37 +01001369 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001370 BUG_ON(!bd_may_claim(bdev, whole, holder));
1371 /*
1372 * Note that for a whole device bd_holders
1373 * will be incremented twice, and bd_holder
1374 * will be set to bd_may_claim before being
1375 * set to holder
1376 */
1377 whole->bd_holders++;
1378 whole->bd_holder = bd_may_claim;
1379 bdev->bd_holders++;
1380 bdev->bd_holder = holder;
1381 }
1382
1383 /* tell others that we're done */
1384 BUG_ON(whole->bd_claiming != holder);
1385 whole->bd_claiming = NULL;
1386 wake_up_bit(&whole->bd_claiming, 0);
1387
1388 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001389
1390 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001391 * Block event polling for write claims if requested. Any
1392 * write holder makes the write_holder state stick until
1393 * all are released. This is good enough and tracking
1394 * individual writeable reference is too fragile given the
1395 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001396 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001397 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1398 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001399 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001400 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001401 }
1402
1403 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001404 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001405 }
1406
1407 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001408}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409EXPORT_SYMBOL(blkdev_get);
1410
Tejun Heod4d77622010-11-13 11:55:18 +01001411/**
1412 * blkdev_get_by_path - open a block device by name
1413 * @path: path to the block device to open
1414 * @mode: FMODE_* mask
1415 * @holder: exclusive holder identifier
1416 *
1417 * Open the blockdevice described by the device file at @path. @mode
1418 * and @holder are identical to blkdev_get().
1419 *
1420 * On success, the returned block_device has reference count of one.
1421 *
1422 * CONTEXT:
1423 * Might sleep.
1424 *
1425 * RETURNS:
1426 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1427 */
1428struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1429 void *holder)
1430{
1431 struct block_device *bdev;
1432 int err;
1433
1434 bdev = lookup_bdev(path);
1435 if (IS_ERR(bdev))
1436 return bdev;
1437
1438 err = blkdev_get(bdev, mode, holder);
1439 if (err)
1440 return ERR_PTR(err);
1441
Chuck Ebberte51900f2011-02-16 18:11:53 -05001442 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1443 blkdev_put(bdev, mode);
1444 return ERR_PTR(-EACCES);
1445 }
1446
Tejun Heod4d77622010-11-13 11:55:18 +01001447 return bdev;
1448}
1449EXPORT_SYMBOL(blkdev_get_by_path);
1450
1451/**
1452 * blkdev_get_by_dev - open a block device by device number
1453 * @dev: device number of block device to open
1454 * @mode: FMODE_* mask
1455 * @holder: exclusive holder identifier
1456 *
1457 * Open the blockdevice described by device number @dev. @mode and
1458 * @holder are identical to blkdev_get().
1459 *
1460 * Use it ONLY if you really do not have anything better - i.e. when
1461 * you are behind a truly sucky interface and all you are given is a
1462 * device number. _Never_ to be used for internal purposes. If you
1463 * ever need it - reconsider your API.
1464 *
1465 * On success, the returned block_device has reference count of one.
1466 *
1467 * CONTEXT:
1468 * Might sleep.
1469 *
1470 * RETURNS:
1471 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1472 */
1473struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1474{
1475 struct block_device *bdev;
1476 int err;
1477
1478 bdev = bdget(dev);
1479 if (!bdev)
1480 return ERR_PTR(-ENOMEM);
1481
1482 err = blkdev_get(bdev, mode, holder);
1483 if (err)
1484 return ERR_PTR(err);
1485
1486 return bdev;
1487}
1488EXPORT_SYMBOL(blkdev_get_by_dev);
1489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490static int blkdev_open(struct inode * inode, struct file * filp)
1491{
1492 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
1494 /*
1495 * Preserve backwards compatibility and allow large file access
1496 * even if userspace doesn't ask for it explicitly. Some mkfs
1497 * binary needs it. We might want to drop this workaround
1498 * during an unstable branch.
1499 */
1500 filp->f_flags |= O_LARGEFILE;
1501
Al Viro572c4892007-10-08 13:24:05 -04001502 if (filp->f_flags & O_NDELAY)
1503 filp->f_mode |= FMODE_NDELAY;
1504 if (filp->f_flags & O_EXCL)
1505 filp->f_mode |= FMODE_EXCL;
1506 if ((filp->f_flags & O_ACCMODE) == 3)
1507 filp->f_mode |= FMODE_WRITE_IOCTL;
1508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001510 if (bdev == NULL)
1511 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
Al Viro572c4892007-10-08 13:24:05 -04001513 filp->f_mapping = bdev->bd_inode->i_mapping;
1514
Tejun Heoe525fd82010-11-13 11:55:17 +01001515 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516}
1517
Al Viro4385bab2013-05-05 22:11:03 -04001518static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001519{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001520 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001521 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001522
NeilBrown6796bf52006-12-08 02:36:16 -08001523 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001524 if (for_part)
1525 bdev->bd_part_count--;
1526
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001527 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001528 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001529 sync_blockdev(bdev);
1530 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001531
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001532 bdev_write_inode(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001533 /*
1534 * Detaching bdev inode from its wb in __destroy_inode()
1535 * is too late: the queue which embeds its bdi (along with
1536 * root wb) can be gone as soon as we put_disk() below.
1537 */
1538 inode_detach_wb(bdev->bd_inode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001539 }
1540 if (bdev->bd_contains == bdev) {
1541 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001542 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001543 }
1544 if (!bdev->bd_openers) {
1545 struct module *owner = disk->fops->owner;
1546
Tejun Heo0762b8b2008-08-25 19:56:12 +09001547 disk_put_part(bdev->bd_part);
1548 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001549 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001550 if (bdev != bdev->bd_contains)
1551 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001552 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001553
1554 put_disk(disk);
1555 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001556 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001557 mutex_unlock(&bdev->bd_mutex);
1558 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001559 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001560 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001561}
1562
Al Viro4385bab2013-05-05 22:11:03 -04001563void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001564{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001565 mutex_lock(&bdev->bd_mutex);
1566
Tejun Heoe525fd82010-11-13 11:55:17 +01001567 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001568 bool bdev_free;
1569
1570 /*
1571 * Release a claim on the device. The holder fields
1572 * are protected with bdev_lock. bd_mutex is to
1573 * synchronize disk_holder unlinking.
1574 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001575 spin_lock(&bdev_lock);
1576
1577 WARN_ON_ONCE(--bdev->bd_holders < 0);
1578 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1579
1580 /* bd_contains might point to self, check in a separate step */
1581 if ((bdev_free = !bdev->bd_holders))
1582 bdev->bd_holder = NULL;
1583 if (!bdev->bd_contains->bd_holders)
1584 bdev->bd_contains->bd_holder = NULL;
1585
1586 spin_unlock(&bdev_lock);
1587
Tejun Heo77ea8872010-12-08 20:57:37 +01001588 /*
1589 * If this was the last claim, remove holder link and
1590 * unblock evpoll if it was a write holder.
1591 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001592 if (bdev_free && bdev->bd_write_holder) {
1593 disk_unblock_events(bdev->bd_disk);
1594 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001595 }
Tejun Heo69362172011-03-09 19:54:27 +01001596 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001597
Tejun Heo85ef06d2011-07-01 16:17:47 +02001598 /*
1599 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1600 * event. This is to ensure detection of media removal commanded
1601 * from userland - e.g. eject(1).
1602 */
1603 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1604
1605 mutex_unlock(&bdev->bd_mutex);
1606
Al Viro4385bab2013-05-05 22:11:03 -04001607 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001608}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001609EXPORT_SYMBOL(blkdev_put);
1610
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611static int blkdev_close(struct inode * inode, struct file * filp)
1612{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001613 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001614 blkdev_put(bdev, filp->f_mode);
1615 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616}
1617
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001618static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001620 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001621 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001622
1623 /*
1624 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1625 * to updated it before every ioctl.
1626 */
Al Viro56b26ad2008-09-19 03:17:36 -04001627 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001628 mode |= FMODE_NDELAY;
1629 else
1630 mode &= ~FMODE_NDELAY;
1631
Al Viro56b26ad2008-09-19 03:17:36 -04001632 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633}
1634
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001635/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001636 * Write data to the block device. Only intended for the block device itself
1637 * and the raw driver which basically is a fake block device.
1638 *
1639 * Does not take i_mutex for the write and thus is not for general purpose
1640 * use.
1641 */
Al Viro1456c0a2014-04-03 03:21:50 -04001642ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001643{
1644 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001645 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001646 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001647 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001648 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001649
Al Viro7ec7b942015-04-07 11:35:14 -04001650 if (bdev_read_only(I_BDEV(bd_inode)))
1651 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001652
Al Viro7ec7b942015-04-07 11:35:14 -04001653 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001654 return 0;
1655
Al Viro7ec7b942015-04-07 11:35:14 -04001656 if (iocb->ki_pos >= size)
1657 return -ENOSPC;
1658
1659 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001660
Jianpeng Ma53362a02012-08-02 09:50:39 +02001661 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001662 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwig02afc272013-09-04 15:04:40 +02001663 if (ret > 0) {
Christoph Hellwigeef99382009-08-20 17:43:41 +02001664 ssize_t err;
Al Viro1456c0a2014-04-03 03:21:50 -04001665 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
Dan Carpenter45d4f852014-04-03 14:47:17 -07001666 if (err < 0)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001667 ret = err;
1668 }
Jianpeng Ma53362a02012-08-02 09:50:39 +02001669 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001670 return ret;
1671}
Al Viro1456c0a2014-04-03 03:21:50 -04001672EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001673
David Jefferyb2de5252014-09-29 10:21:10 -04001674ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001675{
1676 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001677 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001678 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001679 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001680
1681 if (pos >= size)
1682 return 0;
1683
1684 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001685 iov_iter_truncate(to, size);
1686 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001687}
David Jefferyb2de5252014-09-29 10:21:10 -04001688EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001689
Christoph Hellwigeef99382009-08-20 17:43:41 +02001690/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001691 * Try to release a page associated with block device when the system
1692 * is under memory pressure.
1693 */
1694static int blkdev_releasepage(struct page *page, gfp_t wait)
1695{
1696 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1697
1698 if (super && super->s_op->bdev_try_to_free_page)
1699 return super->s_op->bdev_try_to_free_page(super, page, wait);
1700
1701 return try_to_free_buffers(page);
1702}
1703
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001704static int blkdev_writepages(struct address_space *mapping,
1705 struct writeback_control *wbc)
1706{
1707 if (dax_mapping(mapping)) {
1708 struct block_device *bdev = I_BDEV(mapping->host);
1709
1710 return dax_writeback_mapping_range(mapping, bdev, wbc);
1711 }
1712 return generic_writepages(mapping, wbc);
1713}
1714
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001715static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001717 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001719 .write_begin = blkdev_write_begin,
1720 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001721 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001722 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001724 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725};
1726
Dan Williams5a023cd2015-11-30 10:20:29 -08001727#ifdef CONFIG_FS_DAX
1728/*
1729 * In the raw block case we do not need to contend with truncation nor
1730 * unwritten file extents. Without those concerns there is no need for
1731 * additional locking beyond the mmap_sem context that these routines
1732 * are already executing under.
1733 *
1734 * Note, there is no protection if the block device is dynamically
1735 * resized (partition grow/shrink) during a fault. A stable block device
1736 * size is already not enforced in the blkdev_direct_IO path.
1737 *
1738 * For DAX, it is the responsibility of the block device driver to
1739 * ensure the whole-disk device size is stable while requests are in
1740 * flight.
1741 *
1742 * Finally, unlike the filemap_page_mkwrite() case there is no
1743 * filesystem superblock to sync against freezing. We still include a
1744 * pfn_mkwrite callback for dax drivers to receive write fault
1745 * notifications.
1746 */
1747static int blkdev_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1748{
1749 return __dax_fault(vma, vmf, blkdev_get_block, NULL);
1750}
1751
Ross Zwisler9c5a05bc2016-02-05 15:36:08 -08001752static int blkdev_dax_pfn_mkwrite(struct vm_area_struct *vma,
1753 struct vm_fault *vmf)
1754{
1755 return dax_pfn_mkwrite(vma, vmf);
1756}
1757
Dan Williams5a023cd2015-11-30 10:20:29 -08001758static int blkdev_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
1759 pmd_t *pmd, unsigned int flags)
1760{
1761 return __dax_pmd_fault(vma, addr, pmd, flags, blkdev_get_block, NULL);
1762}
1763
Dan Williams5a023cd2015-11-30 10:20:29 -08001764static const struct vm_operations_struct blkdev_dax_vm_ops = {
Dan Williams5a023cd2015-11-30 10:20:29 -08001765 .fault = blkdev_dax_fault,
1766 .pmd_fault = blkdev_dax_pmd_fault,
Ross Zwisler9c5a05bc2016-02-05 15:36:08 -08001767 .pfn_mkwrite = blkdev_dax_pfn_mkwrite,
Dan Williams5a023cd2015-11-30 10:20:29 -08001768};
1769
1770static const struct vm_operations_struct blkdev_default_vm_ops = {
Dan Williams5a023cd2015-11-30 10:20:29 -08001771 .fault = filemap_fault,
1772 .map_pages = filemap_map_pages,
1773};
1774
1775static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
1776{
1777 struct inode *bd_inode = bdev_file_inode(file);
Dan Williams5a023cd2015-11-30 10:20:29 -08001778
1779 file_accessed(file);
Dan Williams5a023cd2015-11-30 10:20:29 -08001780 if (IS_DAX(bd_inode)) {
1781 vma->vm_ops = &blkdev_dax_vm_ops;
1782 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1783 } else {
1784 vma->vm_ops = &blkdev_default_vm_ops;
1785 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001786
1787 return 0;
1788}
1789#else
1790#define blkdev_mmap generic_file_mmap
1791#endif
1792
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001793const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 .open = blkdev_open,
1795 .release = blkdev_close,
1796 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001797 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001798 .write_iter = blkdev_write_iter,
Dan Williams5a023cd2015-11-30 10:20:29 -08001799 .mmap = blkdev_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001800 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001801 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802#ifdef CONFIG_COMPAT
1803 .compat_ioctl = compat_blkdev_ioctl,
1804#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001805 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001806 .splice_write = iter_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807};
1808
1809int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1810{
1811 int res;
1812 mm_segment_t old_fs = get_fs();
1813 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001814 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 set_fs(old_fs);
1816 return res;
1817}
1818
1819EXPORT_SYMBOL(ioctl_by_bdev);
1820
1821/**
1822 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001823 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001825 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 * namespace if possible and return it. Return ERR_PTR(error)
1827 * otherwise.
1828 */
Al Viro421748e2008-08-02 01:04:36 -04001829struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830{
1831 struct block_device *bdev;
1832 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001833 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 int error;
1835
Al Viro421748e2008-08-02 01:04:36 -04001836 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 return ERR_PTR(-EINVAL);
1838
Al Viro421748e2008-08-02 01:04:36 -04001839 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 if (error)
1841 return ERR_PTR(error);
1842
David Howellsbb6687342015-03-17 22:26:21 +00001843 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 error = -ENOTBLK;
1845 if (!S_ISBLK(inode->i_mode))
1846 goto fail;
1847 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001848 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 goto fail;
1850 error = -ENOMEM;
1851 bdev = bd_acquire(inode);
1852 if (!bdev)
1853 goto fail;
1854out:
Al Viro421748e2008-08-02 01:04:36 -04001855 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 return bdev;
1857fail:
1858 bdev = ERR_PTR(error);
1859 goto out;
1860}
Al Virod5686b42008-08-01 05:00:11 -04001861EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
NeilBrown93b270f2011-02-24 17:25:47 +11001863int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001864{
1865 struct super_block *sb = get_super(bdev);
1866 int res = 0;
1867
1868 if (sb) {
1869 /*
1870 * no need to lock the super, get_super holds the
1871 * read mutex so the filesystem cannot go away
1872 * under us (->put_super runs with the write lock
1873 * hold).
1874 */
1875 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001876 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001877 drop_super(sb);
1878 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001879 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001880 return res;
1881}
1882EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001883
1884void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1885{
1886 struct inode *inode, *old_inode = NULL;
1887
Dave Chinner74278da2015-03-04 12:37:22 -05001888 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001889 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1890 struct address_space *mapping = inode->i_mapping;
1891
1892 spin_lock(&inode->i_lock);
1893 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1894 mapping->nrpages == 0) {
1895 spin_unlock(&inode->i_lock);
1896 continue;
1897 }
1898 __iget(inode);
1899 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05001900 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001901 /*
1902 * We hold a reference to 'inode' so it couldn't have been
1903 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05001904 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02001905 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05001906 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02001907 * later.
1908 */
1909 iput(old_inode);
1910 old_inode = inode;
1911
1912 func(I_BDEV(inode), arg);
1913
Dave Chinner74278da2015-03-04 12:37:22 -05001914 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001915 }
Dave Chinner74278da2015-03-04 12:37:22 -05001916 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001917 iput(old_inode);
1918}