blob: 826b164a4b5b1faa9719aefc9cceccec20eda479 [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{
Muthu Kumarb502bd12012-03-23 15:01:50 -0700578 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579}
580
581static struct file_system_type bd_type = {
582 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400583 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 .kill_sb = kill_anon_super,
585};
586
Tejun Heoa212b102015-05-22 17:13:33 -0400587struct super_block *blockdev_superblock __read_mostly;
588EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590void __init bdev_cache_init(void)
591{
592 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300593 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800596 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800597 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900598 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 err = register_filesystem(&bd_type);
600 if (err)
601 panic("Cannot register bdev pseudo-fs");
602 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (IS_ERR(bd_mnt))
604 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300605 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606}
607
608/*
609 * Most likely _very_ bad one - but then it's hardly critical for small
610 * /dev and can be fixed when somebody will need really large one.
611 * Keep in mind that it will be fed through icache hash function too.
612 */
613static inline unsigned long hash(dev_t dev)
614{
615 return MAJOR(dev)+MINOR(dev);
616}
617
618static int bdev_test(struct inode *inode, void *data)
619{
620 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
621}
622
623static int bdev_set(struct inode *inode, void *data)
624{
625 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
626 return 0;
627}
628
629static LIST_HEAD(all_bdevs);
630
631struct block_device *bdget(dev_t dev)
632{
633 struct block_device *bdev;
634 struct inode *inode;
635
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800636 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 bdev_test, bdev_set, &dev);
638
639 if (!inode)
640 return NULL;
641
642 bdev = &BDEV_I(inode)->bdev;
643
644 if (inode->i_state & I_NEW) {
645 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000646 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 bdev->bd_inode = inode;
648 bdev->bd_block_size = (1 << inode->i_blkbits);
649 bdev->bd_part_count = 0;
650 bdev->bd_invalidated = 0;
651 inode->i_mode = S_IFBLK;
652 inode->i_rdev = dev;
653 inode->i_bdev = bdev;
654 inode->i_data.a_ops = &def_blk_aops;
655 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 spin_lock(&bdev_lock);
657 list_add(&bdev->bd_list, &all_bdevs);
658 spin_unlock(&bdev_lock);
659 unlock_new_inode(inode);
660 }
661 return bdev;
662}
663
664EXPORT_SYMBOL(bdget);
665
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200666/**
667 * bdgrab -- Grab a reference to an already referenced block device
668 * @bdev: Block device to grab a reference to.
669 */
670struct block_device *bdgrab(struct block_device *bdev)
671{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400672 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200673 return bdev;
674}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700675EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200676
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677long nr_blockdev_pages(void)
678{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700679 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 long ret = 0;
681 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700682 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 ret += bdev->bd_inode->i_mapping->nrpages;
684 }
685 spin_unlock(&bdev_lock);
686 return ret;
687}
688
689void bdput(struct block_device *bdev)
690{
691 iput(bdev->bd_inode);
692}
693
694EXPORT_SYMBOL(bdput);
695
696static struct block_device *bd_acquire(struct inode *inode)
697{
698 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 spin_lock(&bdev_lock);
701 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700702 if (bdev) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100703 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 spin_unlock(&bdev_lock);
705 return bdev;
706 }
707 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700708
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 bdev = bdget(inode->i_rdev);
710 if (bdev) {
711 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700712 if (!inode->i_bdev) {
713 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400714 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700715 * and it's released in clear_inode() of inode.
716 * So, we can access it via ->i_mapping always
717 * without igrab().
718 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100719 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700720 inode->i_bdev = bdev;
721 inode->i_mapping = bdev->bd_inode->i_mapping;
722 list_add(&inode->i_devices, &bdev->bd_inodes);
723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 spin_unlock(&bdev_lock);
725 }
726 return bdev;
727}
728
729/* Call when you free inode */
730
731void bd_forget(struct inode *inode)
732{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700733 struct block_device *bdev = NULL;
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700736 if (!sb_is_blkdev_sb(inode->i_sb))
737 bdev = inode->i_bdev;
738 __bd_forget(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700740
741 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100742 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900745/**
746 * bd_may_claim - test whether a block device can be claimed
747 * @bdev: block device of interest
748 * @whole: whole block device containing @bdev, may equal @bdev
749 * @holder: holder trying to claim @bdev
750 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300751 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900752 *
753 * CONTEXT:
754 * spin_lock(&bdev_lock).
755 *
756 * RETURNS:
757 * %true if @bdev can be claimed, %false otherwise.
758 */
759static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
760 void *holder)
761{
762 if (bdev->bd_holder == holder)
763 return true; /* already a holder */
764 else if (bdev->bd_holder != NULL)
765 return false; /* held by someone else */
766 else if (bdev->bd_contains == bdev)
767 return true; /* is a whole device which isn't held */
768
Tejun Heoe525fd82010-11-13 11:55:17 +0100769 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900770 return true; /* is a partition of a device that is being partitioned */
771 else if (whole->bd_holder != NULL)
772 return false; /* is a partition of a held device */
773 else
774 return true; /* is a partition of an un-held device */
775}
776
777/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900778 * bd_prepare_to_claim - prepare to claim a block device
779 * @bdev: block device of interest
780 * @whole: the whole device containing @bdev, may equal @bdev
781 * @holder: holder trying to claim @bdev
782 *
783 * Prepare to claim @bdev. This function fails if @bdev is already
784 * claimed by another holder and waits if another claiming is in
785 * progress. This function doesn't actually claim. On successful
786 * return, the caller has ownership of bd_claiming and bd_holder[s].
787 *
788 * CONTEXT:
789 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
790 * it multiple times.
791 *
792 * RETURNS:
793 * 0 if @bdev can be claimed, -EBUSY otherwise.
794 */
795static int bd_prepare_to_claim(struct block_device *bdev,
796 struct block_device *whole, void *holder)
797{
798retry:
799 /* if someone else claimed, fail */
800 if (!bd_may_claim(bdev, whole, holder))
801 return -EBUSY;
802
Tejun Heoe75aa852010-08-04 17:59:39 +0200803 /* if claiming is already in progress, wait for it to finish */
804 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900805 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
806 DEFINE_WAIT(wait);
807
808 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
809 spin_unlock(&bdev_lock);
810 schedule();
811 finish_wait(wq, &wait);
812 spin_lock(&bdev_lock);
813 goto retry;
814 }
815
816 /* yay, all mine */
817 return 0;
818}
819
820/**
821 * bd_start_claiming - start claiming a block device
822 * @bdev: block device of interest
823 * @holder: holder trying to claim @bdev
824 *
825 * @bdev is about to be opened exclusively. Check @bdev can be opened
826 * exclusively and mark that an exclusive open is in progress. Each
827 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000828 * either bd_finish_claiming() or bd_abort_claiming() (which do not
829 * fail).
830 *
831 * This function is used to gain exclusive access to the block device
832 * without actually causing other exclusive open attempts to fail. It
833 * should be used when the open sequence itself requires exclusive
834 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900835 *
836 * CONTEXT:
837 * Might sleep.
838 *
839 * RETURNS:
840 * Pointer to the block device containing @bdev on success, ERR_PTR()
841 * value on failure.
842 */
843static struct block_device *bd_start_claiming(struct block_device *bdev,
844 void *holder)
845{
846 struct gendisk *disk;
847 struct block_device *whole;
848 int partno, err;
849
850 might_sleep();
851
852 /*
853 * @bdev might not have been initialized properly yet, look up
854 * and grab the outer block device the hard way.
855 */
856 disk = get_gendisk(bdev->bd_dev, &partno);
857 if (!disk)
858 return ERR_PTR(-ENXIO);
859
Tejun Heod4c208b2011-06-13 12:45:48 +0200860 /*
861 * Normally, @bdev should equal what's returned from bdget_disk()
862 * if partno is 0; however, some drivers (floppy) use multiple
863 * bdev's for the same physical device and @bdev may be one of the
864 * aliases. Keep @bdev if partno is 0. This means claimer
865 * tracking is broken for those devices but it has always been that
866 * way.
867 */
868 if (partno)
869 whole = bdget_disk(disk, 0);
870 else
871 whole = bdgrab(bdev);
872
Nick Piggincf342572010-05-26 01:50:21 +1000873 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900874 put_disk(disk);
875 if (!whole)
876 return ERR_PTR(-ENOMEM);
877
878 /* prepare to claim, if successful, mark claiming in progress */
879 spin_lock(&bdev_lock);
880
881 err = bd_prepare_to_claim(bdev, whole, holder);
882 if (err == 0) {
883 whole->bd_claiming = holder;
884 spin_unlock(&bdev_lock);
885 return whole;
886 } else {
887 spin_unlock(&bdev_lock);
888 bdput(whole);
889 return ERR_PTR(err);
890 }
891}
892
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800893#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100894struct bd_holder_disk {
895 struct list_head list;
896 struct gendisk *disk;
897 int refcnt;
898};
899
900static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
901 struct gendisk *disk)
902{
903 struct bd_holder_disk *holder;
904
905 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
906 if (holder->disk == disk)
907 return holder;
908 return NULL;
909}
910
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700911static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800912{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700913 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800914}
915
916static void del_symlink(struct kobject *from, struct kobject *to)
917{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800918 sysfs_remove_link(from, kobject_name(to));
919}
920
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800921/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100922 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
923 * @bdev: the claimed slave bdev
924 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500925 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100926 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
927 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100928 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500929 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100930 * - from "slaves" directory of the holder @disk to the claimed @bdev
931 * - from "holders" directory of the @bdev to the holder @disk
932 *
933 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
934 * passed to bd_link_disk_holder(), then:
935 *
936 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
937 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
938 *
939 * The caller must have claimed @bdev before calling this function and
940 * ensure that both @bdev and @disk are valid during the creation and
941 * lifetime of these symlinks.
942 *
943 * CONTEXT:
944 * Might sleep.
945 *
946 * RETURNS:
947 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500948 */
Tejun Heoe09b4572010-11-13 11:55:17 +0100949int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500950{
Tejun Heo49731ba2011-01-14 18:43:57 +0100951 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100952 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800953
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800954 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500955
Tejun Heo49731ba2011-01-14 18:43:57 +0100956 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -0700957
Tejun Heoe09b4572010-11-13 11:55:17 +0100958 /* FIXME: remove the following once add_disk() handles errors */
959 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
960 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -0700961
Tejun Heo49731ba2011-01-14 18:43:57 +0100962 holder = bd_find_holder_disk(bdev, disk);
963 if (holder) {
964 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +0100965 goto out_unlock;
966 }
967
Tejun Heo49731ba2011-01-14 18:43:57 +0100968 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
969 if (!holder) {
970 ret = -ENOMEM;
971 goto out_unlock;
972 }
973
974 INIT_LIST_HEAD(&holder->list);
975 holder->disk = disk;
976 holder->refcnt = 1;
977
978 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
979 if (ret)
980 goto out_free;
981
982 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
983 if (ret)
984 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +0100985 /*
986 * bdev could be deleted beneath us which would implicitly destroy
987 * the holder directory. Hold on to it.
988 */
989 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100990
991 list_add(&holder->list, &bdev->bd_holder_disks);
992 goto out_unlock;
993
994out_del:
995 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
996out_free:
997 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +0100998out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800999 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001000 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001001}
Tejun Heoe09b4572010-11-13 11:55:17 +01001002EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001003
Tejun Heo49731ba2011-01-14 18:43:57 +01001004/**
1005 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1006 * @bdev: the calimed slave bdev
1007 * @disk: the holding disk
1008 *
1009 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1010 *
1011 * CONTEXT:
1012 * Might sleep.
1013 */
1014void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001015{
Tejun Heo49731ba2011-01-14 18:43:57 +01001016 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001017
Tejun Heo49731ba2011-01-14 18:43:57 +01001018 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001019
Tejun Heo49731ba2011-01-14 18:43:57 +01001020 holder = bd_find_holder_disk(bdev, disk);
1021
1022 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1023 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1024 del_symlink(bdev->bd_part->holder_dir,
1025 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001026 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001027 list_del_init(&holder->list);
1028 kfree(holder);
1029 }
1030
1031 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001032}
Tejun Heo49731ba2011-01-14 18:43:57 +01001033EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001034#endif
1035
Andrew Patterson0c002c22008-09-04 14:27:20 -06001036/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001037 * flush_disk - invalidates all buffer-cache entries on a disk
1038 *
1039 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001040 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001041 *
1042 * Invalidates all buffer-cache entries on a disk. It should be called
1043 * when a disk has been changed -- either by a media change or online
1044 * resize.
1045 */
NeilBrown93b270f2011-02-24 17:25:47 +11001046static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001047{
NeilBrown93b270f2011-02-24 17:25:47 +11001048 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001049 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001050 "resized disk %s\n",
1051 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001052 }
1053
1054 if (!bdev->bd_disk)
1055 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001056 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001057 bdev->bd_invalidated = 1;
1058}
1059
1060/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001061 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001062 * @disk: struct gendisk to check
1063 * @bdev: struct bdev to adjust.
1064 *
1065 * This routine checks to see if the bdev size does not match the disk size
1066 * and adjusts it if it differs.
1067 */
1068void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1069{
1070 loff_t disk_size, bdev_size;
1071
1072 disk_size = (loff_t)get_capacity(disk) << 9;
1073 bdev_size = i_size_read(bdev->bd_inode);
1074 if (disk_size != bdev_size) {
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001075 printk(KERN_INFO
1076 "%s: detected capacity change from %lld to %lld\n",
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001077 disk->disk_name, bdev_size, disk_size);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001078 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001079 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001080 }
1081}
1082EXPORT_SYMBOL(check_disk_size_change);
1083
1084/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001085 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001086 * @disk: struct gendisk to be revalidated
1087 *
1088 * This routine is a wrapper for lower-level driver's revalidate_disk
1089 * call-backs. It is used to do common pre and post operations needed
1090 * for all revalidate_disk operations.
1091 */
1092int revalidate_disk(struct gendisk *disk)
1093{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001094 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001095 int ret = 0;
1096
1097 if (disk->fops->revalidate_disk)
1098 ret = disk->fops->revalidate_disk(disk);
Martin K. Petersen25520d52015-10-21 13:19:49 -04001099 blk_integrity_revalidate(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001100 bdev = bdget_disk(disk, 0);
1101 if (!bdev)
1102 return ret;
1103
1104 mutex_lock(&bdev->bd_mutex);
1105 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001106 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001107 mutex_unlock(&bdev->bd_mutex);
1108 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001109 return ret;
1110}
1111EXPORT_SYMBOL(revalidate_disk);
1112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113/*
1114 * This routine checks whether a removable media has been changed,
1115 * and invalidates all buffer-cache-entries in that case. This
1116 * is a relatively slow routine, so we have to try to minimize using
1117 * it. Thus it is called only upon a 'mount' or 'open'. This
1118 * is the best way of combining speed and utility, I think.
1119 * People changing diskettes in the middle of an operation deserve
1120 * to lose :-)
1121 */
1122int check_disk_change(struct block_device *bdev)
1123{
1124 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001125 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001126 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
Tejun Heo77ea8872010-12-08 20:57:37 +01001128 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1129 DISK_EVENT_EJECT_REQUEST);
1130 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 return 0;
1132
NeilBrown93b270f2011-02-24 17:25:47 +11001133 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 if (bdops->revalidate_disk)
1135 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 return 1;
1137}
1138
1139EXPORT_SYMBOL(check_disk_change);
1140
1141void bd_set_size(struct block_device *bdev, loff_t size)
1142{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001143 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Al Viro59551022016-01-22 15:40:57 -05001145 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001146 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001147 inode_unlock(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 while (bsize < PAGE_CACHE_SIZE) {
1149 if (size & bsize)
1150 break;
1151 bsize <<= 1;
1152 }
1153 bdev->bd_block_size = bsize;
1154 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1155}
1156EXPORT_SYMBOL(bd_set_size);
1157
Al Viro4385bab2013-05-05 22:11:03 -04001158static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001159
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001160/*
1161 * bd_mutex locking:
1162 *
1163 * mutex_lock(part->bd_mutex)
1164 * mutex_lock_nested(whole->bd_mutex, 1)
1165 */
1166
Al Viro572c4892007-10-08 13:24:05 -04001167static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001170 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001171 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001172 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001173 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
Al Viro572c4892007-10-08 13:24:05 -04001175 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001176 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001177 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001178 perm |= MAY_WRITE;
1179 /*
1180 * hooks: /n/, see "layering violations".
1181 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001182 if (!for_part) {
1183 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1184 if (ret != 0) {
1185 bdput(bdev);
1186 return ret;
1187 }
Al Viro82666022008-08-01 05:32:04 -04001188 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001189
NeilBrownd3374822009-01-09 08:31:10 +11001190 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001191
Tejun Heo89f97492008-11-05 10:21:06 +01001192 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001193 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001194 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001195 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001196 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Tejun Heo69e02c52011-03-09 19:54:27 +01001198 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001199 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 if (!bdev->bd_openers) {
1201 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001202 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 bdev->bd_contains = bdev;
Dan Williams03cdadb2016-02-26 15:19:43 -08001204 if (IS_ENABLED(CONFIG_BLK_DEV_DAX) && disk->fops->direct_access)
1205 bdev->bd_inode->i_flags = S_DAX;
1206 else
1207 bdev->bd_inode->i_flags = 0;
1208
Tejun Heocf771cb2008-09-03 09:01:09 +02001209 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001210 ret = -ENXIO;
1211 bdev->bd_part = disk_get_part(disk, partno);
1212 if (!bdev->bd_part)
1213 goto out_clear;
1214
Tejun Heo1196f8b2011-04-21 20:54:45 +02001215 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001217 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001218 if (ret == -ERESTARTSYS) {
1219 /* Lost a race with 'disk' being
1220 * deleted, try again.
1221 * See md.c
1222 */
1223 disk_put_part(bdev->bd_part);
1224 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001225 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001226 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001227 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001228 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001229 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001230 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001231 goto restart;
1232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 }
Tejun Heo7e697232011-05-23 13:26:07 +02001234
Dan Williams5a023cd2015-11-30 10:20:29 -08001235 if (!ret) {
Tejun Heo7e697232011-05-23 13:26:07 +02001236 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Dan Williams5a023cd2015-11-30 10:20:29 -08001237 if (!blkdev_dax_capable(bdev))
1238 bdev->bd_inode->i_flags &= ~S_DAX;
1239 }
Tejun Heo7e697232011-05-23 13:26:07 +02001240
Tejun Heo1196f8b2011-04-21 20:54:45 +02001241 /*
1242 * If the device is invalidated, rescan partition
1243 * if open succeeded or failed with -ENOMEDIUM.
1244 * The latter is necessary to prevent ghost
1245 * partitions on a removed medium.
1246 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001247 if (bdev->bd_invalidated) {
1248 if (!ret)
1249 rescan_partitions(disk, bdev);
1250 else if (ret == -ENOMEDIUM)
1251 invalidate_partitions(disk, bdev);
1252 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001253
Tejun Heo1196f8b2011-04-21 20:54:45 +02001254 if (ret)
1255 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 struct block_device *whole;
1258 whole = bdget_disk(disk, 0);
1259 ret = -ENOMEM;
1260 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001261 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001262 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001263 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001265 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001267 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001268 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001269 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001271 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 }
Tejun Heo89f97492008-11-05 10:21:06 +01001273 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Dan Williams5a023cd2015-11-30 10:20:29 -08001274 if (!blkdev_dax_capable(bdev))
Jeff Moyerf0b2e562015-08-14 16:15:32 -04001275 bdev->bd_inode->i_flags &= ~S_DAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 }
1277 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001279 ret = 0;
1280 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001281 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001282 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001283 if (bdev->bd_invalidated) {
1284 if (!ret)
1285 rescan_partitions(bdev->bd_disk, bdev);
1286 else if (ret == -ENOMEDIUM)
1287 invalidate_partitions(bdev->bd_disk, bdev);
1288 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001289 if (ret)
1290 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001292 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001293 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001294 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 }
1296 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001297 if (for_part)
1298 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001299 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001300 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 return 0;
1302
Tejun Heo0762b8b2008-08-25 19:56:12 +09001303 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001304 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001306 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001307 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001309 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001311 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001312 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001313 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001314 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001315 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001316 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001317 bdput(bdev);
1318
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 return ret;
1320}
1321
Tejun Heod4d77622010-11-13 11:55:18 +01001322/**
1323 * blkdev_get - open a block device
1324 * @bdev: block_device to open
1325 * @mode: FMODE_* mask
1326 * @holder: exclusive holder identifier
1327 *
1328 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1329 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1330 * @holder is invalid. Exclusive opens may nest for the same @holder.
1331 *
1332 * On success, the reference count of @bdev is unchanged. On failure,
1333 * @bdev is put.
1334 *
1335 * CONTEXT:
1336 * Might sleep.
1337 *
1338 * RETURNS:
1339 * 0 on success, -errno on failure.
1340 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001341int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342{
Tejun Heoe525fd82010-11-13 11:55:17 +01001343 struct block_device *whole = NULL;
1344 int res;
1345
1346 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1347
1348 if ((mode & FMODE_EXCL) && holder) {
1349 whole = bd_start_claiming(bdev, holder);
1350 if (IS_ERR(whole)) {
1351 bdput(bdev);
1352 return PTR_ERR(whole);
1353 }
1354 }
1355
1356 res = __blkdev_get(bdev, mode, 0);
1357
1358 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001359 struct gendisk *disk = whole->bd_disk;
1360
Tejun Heo6a027ef2010-11-13 11:55:17 +01001361 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001362 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001363 spin_lock(&bdev_lock);
1364
Tejun Heo77ea8872010-12-08 20:57:37 +01001365 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001366 BUG_ON(!bd_may_claim(bdev, whole, holder));
1367 /*
1368 * Note that for a whole device bd_holders
1369 * will be incremented twice, and bd_holder
1370 * will be set to bd_may_claim before being
1371 * set to holder
1372 */
1373 whole->bd_holders++;
1374 whole->bd_holder = bd_may_claim;
1375 bdev->bd_holders++;
1376 bdev->bd_holder = holder;
1377 }
1378
1379 /* tell others that we're done */
1380 BUG_ON(whole->bd_claiming != holder);
1381 whole->bd_claiming = NULL;
1382 wake_up_bit(&whole->bd_claiming, 0);
1383
1384 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001385
1386 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001387 * Block event polling for write claims if requested. Any
1388 * write holder makes the write_holder state stick until
1389 * all are released. This is good enough and tracking
1390 * individual writeable reference is too fragile given the
1391 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001392 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001393 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1394 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001395 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001396 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001397 }
1398
1399 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001400 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001401 }
1402
1403 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001404}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405EXPORT_SYMBOL(blkdev_get);
1406
Tejun Heod4d77622010-11-13 11:55:18 +01001407/**
1408 * blkdev_get_by_path - open a block device by name
1409 * @path: path to the block device to open
1410 * @mode: FMODE_* mask
1411 * @holder: exclusive holder identifier
1412 *
1413 * Open the blockdevice described by the device file at @path. @mode
1414 * and @holder are identical to blkdev_get().
1415 *
1416 * On success, the returned block_device has reference count of one.
1417 *
1418 * CONTEXT:
1419 * Might sleep.
1420 *
1421 * RETURNS:
1422 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1423 */
1424struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1425 void *holder)
1426{
1427 struct block_device *bdev;
1428 int err;
1429
1430 bdev = lookup_bdev(path);
1431 if (IS_ERR(bdev))
1432 return bdev;
1433
1434 err = blkdev_get(bdev, mode, holder);
1435 if (err)
1436 return ERR_PTR(err);
1437
Chuck Ebberte51900f2011-02-16 18:11:53 -05001438 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1439 blkdev_put(bdev, mode);
1440 return ERR_PTR(-EACCES);
1441 }
1442
Tejun Heod4d77622010-11-13 11:55:18 +01001443 return bdev;
1444}
1445EXPORT_SYMBOL(blkdev_get_by_path);
1446
1447/**
1448 * blkdev_get_by_dev - open a block device by device number
1449 * @dev: device number of block device to open
1450 * @mode: FMODE_* mask
1451 * @holder: exclusive holder identifier
1452 *
1453 * Open the blockdevice described by device number @dev. @mode and
1454 * @holder are identical to blkdev_get().
1455 *
1456 * Use it ONLY if you really do not have anything better - i.e. when
1457 * you are behind a truly sucky interface and all you are given is a
1458 * device number. _Never_ to be used for internal purposes. If you
1459 * ever need it - reconsider your API.
1460 *
1461 * On success, the returned block_device has reference count of one.
1462 *
1463 * CONTEXT:
1464 * Might sleep.
1465 *
1466 * RETURNS:
1467 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1468 */
1469struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1470{
1471 struct block_device *bdev;
1472 int err;
1473
1474 bdev = bdget(dev);
1475 if (!bdev)
1476 return ERR_PTR(-ENOMEM);
1477
1478 err = blkdev_get(bdev, mode, holder);
1479 if (err)
1480 return ERR_PTR(err);
1481
1482 return bdev;
1483}
1484EXPORT_SYMBOL(blkdev_get_by_dev);
1485
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486static int blkdev_open(struct inode * inode, struct file * filp)
1487{
1488 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
1490 /*
1491 * Preserve backwards compatibility and allow large file access
1492 * even if userspace doesn't ask for it explicitly. Some mkfs
1493 * binary needs it. We might want to drop this workaround
1494 * during an unstable branch.
1495 */
1496 filp->f_flags |= O_LARGEFILE;
1497
Al Viro572c4892007-10-08 13:24:05 -04001498 if (filp->f_flags & O_NDELAY)
1499 filp->f_mode |= FMODE_NDELAY;
1500 if (filp->f_flags & O_EXCL)
1501 filp->f_mode |= FMODE_EXCL;
1502 if ((filp->f_flags & O_ACCMODE) == 3)
1503 filp->f_mode |= FMODE_WRITE_IOCTL;
1504
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001506 if (bdev == NULL)
1507 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Al Viro572c4892007-10-08 13:24:05 -04001509 filp->f_mapping = bdev->bd_inode->i_mapping;
1510
Tejun Heoe525fd82010-11-13 11:55:17 +01001511 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512}
1513
Al Viro4385bab2013-05-05 22:11:03 -04001514static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001515{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001516 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001517 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001518
NeilBrown6796bf52006-12-08 02:36:16 -08001519 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001520 if (for_part)
1521 bdev->bd_part_count--;
1522
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001523 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001524 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001525 sync_blockdev(bdev);
1526 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001527
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001528 bdev_write_inode(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001529 /*
1530 * Detaching bdev inode from its wb in __destroy_inode()
1531 * is too late: the queue which embeds its bdi (along with
1532 * root wb) can be gone as soon as we put_disk() below.
1533 */
1534 inode_detach_wb(bdev->bd_inode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001535 }
1536 if (bdev->bd_contains == bdev) {
1537 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001538 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001539 }
1540 if (!bdev->bd_openers) {
1541 struct module *owner = disk->fops->owner;
1542
Tejun Heo0762b8b2008-08-25 19:56:12 +09001543 disk_put_part(bdev->bd_part);
1544 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001545 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001546 if (bdev != bdev->bd_contains)
1547 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001548 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001549
1550 put_disk(disk);
1551 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001552 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001553 mutex_unlock(&bdev->bd_mutex);
1554 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001555 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001556 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001557}
1558
Al Viro4385bab2013-05-05 22:11:03 -04001559void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001560{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001561 mutex_lock(&bdev->bd_mutex);
1562
Tejun Heoe525fd82010-11-13 11:55:17 +01001563 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001564 bool bdev_free;
1565
1566 /*
1567 * Release a claim on the device. The holder fields
1568 * are protected with bdev_lock. bd_mutex is to
1569 * synchronize disk_holder unlinking.
1570 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001571 spin_lock(&bdev_lock);
1572
1573 WARN_ON_ONCE(--bdev->bd_holders < 0);
1574 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1575
1576 /* bd_contains might point to self, check in a separate step */
1577 if ((bdev_free = !bdev->bd_holders))
1578 bdev->bd_holder = NULL;
1579 if (!bdev->bd_contains->bd_holders)
1580 bdev->bd_contains->bd_holder = NULL;
1581
1582 spin_unlock(&bdev_lock);
1583
Tejun Heo77ea8872010-12-08 20:57:37 +01001584 /*
1585 * If this was the last claim, remove holder link and
1586 * unblock evpoll if it was a write holder.
1587 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001588 if (bdev_free && bdev->bd_write_holder) {
1589 disk_unblock_events(bdev->bd_disk);
1590 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001591 }
Tejun Heo69362172011-03-09 19:54:27 +01001592 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001593
Tejun Heo85ef06d2011-07-01 16:17:47 +02001594 /*
1595 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1596 * event. This is to ensure detection of media removal commanded
1597 * from userland - e.g. eject(1).
1598 */
1599 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1600
1601 mutex_unlock(&bdev->bd_mutex);
1602
Al Viro4385bab2013-05-05 22:11:03 -04001603 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001604}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001605EXPORT_SYMBOL(blkdev_put);
1606
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607static int blkdev_close(struct inode * inode, struct file * filp)
1608{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001609 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001610 blkdev_put(bdev, filp->f_mode);
1611 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612}
1613
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001614static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001616 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001617 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001618
1619 /*
1620 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1621 * to updated it before every ioctl.
1622 */
Al Viro56b26ad2008-09-19 03:17:36 -04001623 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001624 mode |= FMODE_NDELAY;
1625 else
1626 mode &= ~FMODE_NDELAY;
1627
Al Viro56b26ad2008-09-19 03:17:36 -04001628 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629}
1630
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001631/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001632 * Write data to the block device. Only intended for the block device itself
1633 * and the raw driver which basically is a fake block device.
1634 *
1635 * Does not take i_mutex for the write and thus is not for general purpose
1636 * use.
1637 */
Al Viro1456c0a2014-04-03 03:21:50 -04001638ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001639{
1640 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001641 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001642 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001643 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001644 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001645
Al Viro7ec7b942015-04-07 11:35:14 -04001646 if (bdev_read_only(I_BDEV(bd_inode)))
1647 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001648
Al Viro7ec7b942015-04-07 11:35:14 -04001649 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001650 return 0;
1651
Al Viro7ec7b942015-04-07 11:35:14 -04001652 if (iocb->ki_pos >= size)
1653 return -ENOSPC;
1654
1655 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001656
Jianpeng Ma53362a02012-08-02 09:50:39 +02001657 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001658 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwig02afc272013-09-04 15:04:40 +02001659 if (ret > 0) {
Christoph Hellwigeef99382009-08-20 17:43:41 +02001660 ssize_t err;
Al Viro1456c0a2014-04-03 03:21:50 -04001661 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
Dan Carpenter45d4f852014-04-03 14:47:17 -07001662 if (err < 0)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001663 ret = err;
1664 }
Jianpeng Ma53362a02012-08-02 09:50:39 +02001665 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001666 return ret;
1667}
Al Viro1456c0a2014-04-03 03:21:50 -04001668EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001669
David Jefferyb2de5252014-09-29 10:21:10 -04001670ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001671{
1672 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001673 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001674 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001675 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001676
1677 if (pos >= size)
1678 return 0;
1679
1680 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001681 iov_iter_truncate(to, size);
1682 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001683}
David Jefferyb2de5252014-09-29 10:21:10 -04001684EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001685
Christoph Hellwigeef99382009-08-20 17:43:41 +02001686/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001687 * Try to release a page associated with block device when the system
1688 * is under memory pressure.
1689 */
1690static int blkdev_releasepage(struct page *page, gfp_t wait)
1691{
1692 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1693
1694 if (super && super->s_op->bdev_try_to_free_page)
1695 return super->s_op->bdev_try_to_free_page(super, page, wait);
1696
1697 return try_to_free_buffers(page);
1698}
1699
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001700static int blkdev_writepages(struct address_space *mapping,
1701 struct writeback_control *wbc)
1702{
1703 if (dax_mapping(mapping)) {
1704 struct block_device *bdev = I_BDEV(mapping->host);
1705
1706 return dax_writeback_mapping_range(mapping, bdev, wbc);
1707 }
1708 return generic_writepages(mapping, wbc);
1709}
1710
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001711static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001713 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001715 .write_begin = blkdev_write_begin,
1716 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001717 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001718 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001720 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721};
1722
Dan Williams5a023cd2015-11-30 10:20:29 -08001723#ifdef CONFIG_FS_DAX
1724/*
1725 * In the raw block case we do not need to contend with truncation nor
1726 * unwritten file extents. Without those concerns there is no need for
1727 * additional locking beyond the mmap_sem context that these routines
1728 * are already executing under.
1729 *
1730 * Note, there is no protection if the block device is dynamically
1731 * resized (partition grow/shrink) during a fault. A stable block device
1732 * size is already not enforced in the blkdev_direct_IO path.
1733 *
1734 * For DAX, it is the responsibility of the block device driver to
1735 * ensure the whole-disk device size is stable while requests are in
1736 * flight.
1737 *
1738 * Finally, unlike the filemap_page_mkwrite() case there is no
1739 * filesystem superblock to sync against freezing. We still include a
1740 * pfn_mkwrite callback for dax drivers to receive write fault
1741 * notifications.
1742 */
1743static int blkdev_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1744{
1745 return __dax_fault(vma, vmf, blkdev_get_block, NULL);
1746}
1747
Ross Zwisler9c5a05bc2016-02-05 15:36:08 -08001748static int blkdev_dax_pfn_mkwrite(struct vm_area_struct *vma,
1749 struct vm_fault *vmf)
1750{
1751 return dax_pfn_mkwrite(vma, vmf);
1752}
1753
Dan Williams5a023cd2015-11-30 10:20:29 -08001754static int blkdev_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
1755 pmd_t *pmd, unsigned int flags)
1756{
1757 return __dax_pmd_fault(vma, addr, pmd, flags, blkdev_get_block, NULL);
1758}
1759
Dan Williams5a023cd2015-11-30 10:20:29 -08001760static const struct vm_operations_struct blkdev_dax_vm_ops = {
Dan Williams5a023cd2015-11-30 10:20:29 -08001761 .fault = blkdev_dax_fault,
1762 .pmd_fault = blkdev_dax_pmd_fault,
Ross Zwisler9c5a05bc2016-02-05 15:36:08 -08001763 .pfn_mkwrite = blkdev_dax_pfn_mkwrite,
Dan Williams5a023cd2015-11-30 10:20:29 -08001764};
1765
1766static const struct vm_operations_struct blkdev_default_vm_ops = {
Dan Williams5a023cd2015-11-30 10:20:29 -08001767 .fault = filemap_fault,
1768 .map_pages = filemap_map_pages,
1769};
1770
1771static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
1772{
1773 struct inode *bd_inode = bdev_file_inode(file);
Dan Williams5a023cd2015-11-30 10:20:29 -08001774
1775 file_accessed(file);
Dan Williams5a023cd2015-11-30 10:20:29 -08001776 if (IS_DAX(bd_inode)) {
1777 vma->vm_ops = &blkdev_dax_vm_ops;
1778 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1779 } else {
1780 vma->vm_ops = &blkdev_default_vm_ops;
1781 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001782
1783 return 0;
1784}
1785#else
1786#define blkdev_mmap generic_file_mmap
1787#endif
1788
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001789const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 .open = blkdev_open,
1791 .release = blkdev_close,
1792 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001793 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001794 .write_iter = blkdev_write_iter,
Dan Williams5a023cd2015-11-30 10:20:29 -08001795 .mmap = blkdev_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001796 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001797 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798#ifdef CONFIG_COMPAT
1799 .compat_ioctl = compat_blkdev_ioctl,
1800#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001801 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001802 .splice_write = iter_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803};
1804
1805int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1806{
1807 int res;
1808 mm_segment_t old_fs = get_fs();
1809 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001810 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 set_fs(old_fs);
1812 return res;
1813}
1814
1815EXPORT_SYMBOL(ioctl_by_bdev);
1816
1817/**
1818 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001819 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001821 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 * namespace if possible and return it. Return ERR_PTR(error)
1823 * otherwise.
1824 */
Al Viro421748e2008-08-02 01:04:36 -04001825struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826{
1827 struct block_device *bdev;
1828 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001829 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 int error;
1831
Al Viro421748e2008-08-02 01:04:36 -04001832 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 return ERR_PTR(-EINVAL);
1834
Al Viro421748e2008-08-02 01:04:36 -04001835 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 if (error)
1837 return ERR_PTR(error);
1838
David Howellsbb6687342015-03-17 22:26:21 +00001839 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 error = -ENOTBLK;
1841 if (!S_ISBLK(inode->i_mode))
1842 goto fail;
1843 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001844 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 goto fail;
1846 error = -ENOMEM;
1847 bdev = bd_acquire(inode);
1848 if (!bdev)
1849 goto fail;
1850out:
Al Viro421748e2008-08-02 01:04:36 -04001851 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 return bdev;
1853fail:
1854 bdev = ERR_PTR(error);
1855 goto out;
1856}
Al Virod5686b42008-08-01 05:00:11 -04001857EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
NeilBrown93b270f2011-02-24 17:25:47 +11001859int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001860{
1861 struct super_block *sb = get_super(bdev);
1862 int res = 0;
1863
1864 if (sb) {
1865 /*
1866 * no need to lock the super, get_super holds the
1867 * read mutex so the filesystem cannot go away
1868 * under us (->put_super runs with the write lock
1869 * hold).
1870 */
1871 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001872 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001873 drop_super(sb);
1874 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001875 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001876 return res;
1877}
1878EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001879
1880void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1881{
1882 struct inode *inode, *old_inode = NULL;
1883
Dave Chinner74278da2015-03-04 12:37:22 -05001884 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001885 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1886 struct address_space *mapping = inode->i_mapping;
1887
1888 spin_lock(&inode->i_lock);
1889 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1890 mapping->nrpages == 0) {
1891 spin_unlock(&inode->i_lock);
1892 continue;
1893 }
1894 __iget(inode);
1895 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05001896 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001897 /*
1898 * We hold a reference to 'inode' so it couldn't have been
1899 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05001900 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02001901 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05001902 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02001903 * later.
1904 */
1905 iput(old_inode);
1906 old_inode = inode;
1907
1908 func(I_BDEV(inode), arg);
1909
Dave Chinner74278da2015-03-04 12:37:22 -05001910 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001911 }
Dave Chinner74278da2015-03-04 12:37:22 -05001912 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001913 iput(old_inode);
1914}