blob: c25639e907bd21a194aecedc6c60a93fb4994375 [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
Johannes Weiner91b0abe2014-04-03 14:47:49 -070078 if (mapping->nrpages == 0 && mapping->nrshadows == 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
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800159static ssize_t
Omar Sandoval22c61862015-03-16 04:33:53 -0700160blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800161{
162 struct file *file = iocb->ki_filp;
163 struct inode *inode = file->f_mapping->host;
164
Matthew Wilcoxbbab37d2015-07-03 10:40:42 -0400165 if (IS_DAX(inode))
166 return dax_do_io(iocb, inode, iter, offset, blkdev_get_block,
167 NULL, DIO_SKIP_DIO_COUNT);
Omar Sandoval17f8c842015-03-16 04:33:50 -0700168 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset,
Jens Axboefe0f07d2015-04-15 17:05:48 -0600169 blkdev_get_block, NULL, NULL,
170 DIO_SKIP_DIO_COUNT);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800171}
172
Jan Kara5cee5812009-04-27 16:43:51 +0200173int __sync_blockdev(struct block_device *bdev, int wait)
174{
175 if (!bdev)
176 return 0;
177 if (!wait)
178 return filemap_flush(bdev->bd_inode->i_mapping);
179 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
180}
181
Nick Piggin585d3bc2009-02-25 10:44:19 +0100182/*
183 * Write out and wait upon all the dirty data associated with a block
184 * device via its mapping. Does not take the superblock lock.
185 */
186int sync_blockdev(struct block_device *bdev)
187{
Jan Kara5cee5812009-04-27 16:43:51 +0200188 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100189}
190EXPORT_SYMBOL(sync_blockdev);
191
192/*
193 * Write out and wait upon all dirty data associated with this
194 * device. Filesystem data as well as the underlying block
195 * device. Takes the superblock lock.
196 */
197int fsync_bdev(struct block_device *bdev)
198{
199 struct super_block *sb = get_super(bdev);
200 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200201 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100202 drop_super(sb);
203 return res;
204 }
205 return sync_blockdev(bdev);
206}
Al Viro47e44912009-04-01 07:07:16 -0400207EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100208
209/**
210 * freeze_bdev -- lock a filesystem and force it into a consistent state
211 * @bdev: blockdevice to lock
212 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100213 * If a superblock is found on this device, we take the s_umount semaphore
214 * on it to make sure nobody unmounts until the snapshot creation is done.
215 * The reference counter (bd_fsfreeze_count) guarantees that only the last
216 * unfreeze process can unfreeze the frozen filesystem actually when multiple
217 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
218 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
219 * actually.
220 */
221struct super_block *freeze_bdev(struct block_device *bdev)
222{
223 struct super_block *sb;
224 int error = 0;
225
226 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200227 if (++bdev->bd_fsfreeze_count > 1) {
228 /*
229 * We don't even need to grab a reference - the first call
230 * to freeze_bdev grab an active reference and only the last
231 * thaw_bdev drops it.
232 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100233 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200234 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100235 mutex_unlock(&bdev->bd_fsfreeze_mutex);
236 return sb;
237 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100238
Christoph Hellwig45042302009-08-03 23:28:35 +0200239 sb = get_active_super(bdev);
240 if (!sb)
241 goto out;
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600242 if (sb->s_op->freeze_super)
243 error = sb->s_op->freeze_super(sb);
244 else
245 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400246 if (error) {
247 deactivate_super(sb);
248 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200249 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400250 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100251 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400252 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200253 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100254 sync_blockdev(bdev);
255 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200256 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100257}
258EXPORT_SYMBOL(freeze_bdev);
259
260/**
261 * thaw_bdev -- unlock filesystem
262 * @bdev: blockdevice to unlock
263 * @sb: associated superblock
264 *
265 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
266 */
267int thaw_bdev(struct block_device *bdev, struct super_block *sb)
268{
Christoph Hellwig45042302009-08-03 23:28:35 +0200269 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100270
271 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200272 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400273 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100274
Christoph Hellwig45042302009-08-03 23:28:35 +0200275 error = 0;
276 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400277 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100278
Christoph Hellwig45042302009-08-03 23:28:35 +0200279 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400280 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200281
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600282 if (sb->s_op->thaw_super)
283 error = sb->s_op->thaw_super(sb);
284 else
285 error = thaw_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400286 if (error) {
287 bdev->bd_fsfreeze_count++;
288 mutex_unlock(&bdev->bd_fsfreeze_mutex);
289 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100290 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400291out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100292 mutex_unlock(&bdev->bd_fsfreeze_mutex);
293 return 0;
294}
295EXPORT_SYMBOL(thaw_bdev);
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
298{
299 return block_write_full_page(page, blkdev_get_block, wbc);
300}
301
302static int blkdev_readpage(struct file * file, struct page * page)
303{
304 return block_read_full_page(page, blkdev_get_block);
305}
306
Akinobu Mita447f05b2014-10-09 15:26:58 -0700307static int blkdev_readpages(struct file *file, struct address_space *mapping,
308 struct list_head *pages, unsigned nr_pages)
309{
310 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
311}
312
Nick Piggin6272b5a2007-10-16 01:25:04 -0700313static int blkdev_write_begin(struct file *file, struct address_space *mapping,
314 loff_t pos, unsigned len, unsigned flags,
315 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200317 return block_write_begin(mapping, pos, len, flags, pagep,
318 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319}
320
Nick Piggin6272b5a2007-10-16 01:25:04 -0700321static int blkdev_write_end(struct file *file, struct address_space *mapping,
322 loff_t pos, unsigned len, unsigned copied,
323 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700325 int ret;
326 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
327
328 unlock_page(page);
329 page_cache_release(page);
330
331 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}
333
334/*
335 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500336 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 * so we compute the size by hand (just as in block_read/write above)
338 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800339static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
341 struct inode *bd_inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 loff_t retval;
343
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800344 mutex_lock(&bd_inode->i_mutex);
Al Viro5d48f3a2013-06-23 21:34:45 +0400345 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800346 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return retval;
348}
349
Josef Bacik02c24a82011-07-16 20:44:56 -0400350int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400352 struct inode *bd_inode = filp->f_mapping->host;
353 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100354 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200355
356 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
357 if (error)
358 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100359
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400360 /*
361 * There is no need to serialise calls to blkdev_issue_flush with
362 * i_mutex and doing so causes performance issues with concurrent
363 * O_SYNC writers to a block device.
364 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200365 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100366 if (error == -EOPNOTSUPP)
367 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400368
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100369 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700371EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700373/**
374 * bdev_read_page() - Start reading a page from a block device
375 * @bdev: The device to read the page from
376 * @sector: The offset on the device to read the page to (need not be aligned)
377 * @page: The page to read
378 *
379 * On entry, the page should be locked. It will be unlocked when the page
380 * has been read. If the block driver implements rw_page synchronously,
381 * that will be true on exit from this function, but it need not be.
382 *
383 * Errors returned by this function are usually "soft", eg out of memory, or
384 * queue full; callers should try a different route to read this page rather
385 * than propagate an error back up the stack.
386 *
387 * Return: negative errno if an error occurs, 0 if submission was successful.
388 */
389int bdev_read_page(struct block_device *bdev, sector_t sector,
390 struct page *page)
391{
392 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800393 int result = -EOPNOTSUPP;
394
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400395 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc92015-11-19 13:29:28 -0800396 return result;
397
398 result = blk_queue_enter(bdev->bd_queue, GFP_KERNEL);
399 if (result)
400 return result;
401 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
402 blk_queue_exit(bdev->bd_queue);
403 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700404}
405EXPORT_SYMBOL_GPL(bdev_read_page);
406
407/**
408 * bdev_write_page() - Start writing a page to a block device
409 * @bdev: The device to write the page to
410 * @sector: The offset on the device to write the page to (need not be aligned)
411 * @page: The page to write
412 * @wbc: The writeback_control for the write
413 *
414 * On entry, the page should be locked and not currently under writeback.
415 * On exit, if the write started successfully, the page will be unlocked and
416 * under writeback. If the write failed already (eg the driver failed to
417 * queue the page to the device), the page will still be locked. If the
418 * caller is a ->writepage implementation, it will need to unlock the page.
419 *
420 * Errors returned by this function are usually "soft", eg out of memory, or
421 * queue full; callers should try a different route to write this page rather
422 * than propagate an error back up the stack.
423 *
424 * Return: negative errno if an error occurs, 0 if submission was successful.
425 */
426int bdev_write_page(struct block_device *bdev, sector_t sector,
427 struct page *page, struct writeback_control *wbc)
428{
429 int result;
430 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
431 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800432
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400433 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700434 return -EOPNOTSUPP;
Dan Williams2e6edc92015-11-19 13:29:28 -0800435 result = blk_queue_enter(bdev->bd_queue, GFP_KERNEL);
436 if (result)
437 return result;
438
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700439 set_page_writeback(page);
440 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
441 if (result)
442 end_page_writeback(page);
443 else
444 unlock_page(page);
Dan Williams2e6edc92015-11-19 13:29:28 -0800445 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700446 return result;
447}
448EXPORT_SYMBOL_GPL(bdev_write_page);
449
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200450/**
451 * bdev_direct_access() - Get the address for directly-accessibly memory
452 * @bdev: The device containing the memory
453 * @sector: The offset within the device
454 * @addr: Where to put the address of the memory
455 * @pfn: The Page Frame Number for the memory
456 * @size: The number of bytes requested
457 *
458 * If a block device is made up of directly addressable memory, this function
459 * will tell the caller the PFN and the address of the memory. The address
460 * may be directly dereferenced within the kernel without the need to call
461 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
462 * page tables.
463 *
464 * Return: negative errno if an error occurs, otherwise the number of bytes
465 * accessible at this address.
466 */
467long bdev_direct_access(struct block_device *bdev, sector_t sector,
Ross Zwislere2e05392015-08-18 13:55:41 -0600468 void __pmem **addr, unsigned long *pfn, long size)
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200469{
470 long avail;
471 const struct block_device_operations *ops = bdev->bd_disk->fops;
472
Matthew Wilcox43c3dd02015-07-03 10:40:43 -0400473 /*
474 * The device driver is allowed to sleep, in order to make the
475 * memory directly accessible.
476 */
477 might_sleep();
478
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200479 if (size < 0)
480 return size;
481 if (!ops->direct_access)
482 return -EOPNOTSUPP;
483 if ((sector + DIV_ROUND_UP(size, 512)) >
484 part_nr_sects_read(bdev->bd_part))
485 return -ERANGE;
486 sector += get_start_sect(bdev);
487 if (sector % (PAGE_SIZE / 512))
488 return -EINVAL;
Dan Williamscb389b92015-08-07 17:41:00 -0400489 avail = ops->direct_access(bdev, sector, addr, pfn);
Matthew Wilcoxdd22f552015-01-07 18:05:34 +0200490 if (!avail)
491 return -ERANGE;
492 return min(avail, size);
493}
494EXPORT_SYMBOL_GPL(bdev_direct_access);
495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496/*
497 * pseudo-fs
498 */
499
500static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800501static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
503static struct inode *bdev_alloc_inode(struct super_block *sb)
504{
Christoph Lametere94b1762006-12-06 20:33:17 -0800505 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 if (!ei)
507 return NULL;
508 return &ei->vfs_inode;
509}
510
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100511static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100513 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 struct bdev_inode *bdi = BDEV_I(inode);
515
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 kmem_cache_free(bdev_cachep, bdi);
517}
518
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100519static void bdev_destroy_inode(struct inode *inode)
520{
521 call_rcu(&inode->i_rcu, bdev_i_callback);
522}
523
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700524static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
526 struct bdev_inode *ei = (struct bdev_inode *) foo;
527 struct block_device *bdev = &ei->bdev;
528
Christoph Lametera35afb82007-05-16 22:10:57 -0700529 memset(bdev, 0, sizeof(*bdev));
530 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700531 INIT_LIST_HEAD(&bdev->bd_inodes);
532 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100533#ifdef CONFIG_SYSFS
534 INIT_LIST_HEAD(&bdev->bd_holder_disks);
535#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700536 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800537 /* Initialize mutex for freeze. */
538 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539}
540
541static inline void __bd_forget(struct inode *inode)
542{
543 list_del_init(&inode->i_devices);
544 inode->i_bdev = NULL;
545 inode->i_mapping = &inode->i_data;
546}
547
Al Virob57922d2010-06-07 14:34:48 -0400548static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
550 struct block_device *bdev = &BDEV_I(inode)->bdev;
551 struct list_head *p;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700552 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400553 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200554 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 spin_lock(&bdev_lock);
556 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
557 __bd_forget(list_entry(p, struct inode, i_devices));
558 }
559 list_del_init(&bdev->bd_list);
560 spin_unlock(&bdev_lock);
561}
562
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800563static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 .statfs = simple_statfs,
565 .alloc_inode = bdev_alloc_inode,
566 .destroy_inode = bdev_destroy_inode,
567 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400568 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569};
570
Al Viro51139ad2010-07-25 23:47:46 +0400571static struct dentry *bd_mount(struct file_system_type *fs_type,
572 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
Muthu Kumarb502bd12012-03-23 15:01:50 -0700574 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575}
576
577static struct file_system_type bd_type = {
578 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400579 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 .kill_sb = kill_anon_super,
581};
582
Tejun Heoa212b102015-05-22 17:13:33 -0400583struct super_block *blockdev_superblock __read_mostly;
584EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
586void __init bdev_cache_init(void)
587{
588 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300589 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800592 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
593 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900594 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 err = register_filesystem(&bd_type);
596 if (err)
597 panic("Cannot register bdev pseudo-fs");
598 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 if (IS_ERR(bd_mnt))
600 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300601 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602}
603
604/*
605 * Most likely _very_ bad one - but then it's hardly critical for small
606 * /dev and can be fixed when somebody will need really large one.
607 * Keep in mind that it will be fed through icache hash function too.
608 */
609static inline unsigned long hash(dev_t dev)
610{
611 return MAJOR(dev)+MINOR(dev);
612}
613
614static int bdev_test(struct inode *inode, void *data)
615{
616 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
617}
618
619static int bdev_set(struct inode *inode, void *data)
620{
621 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
622 return 0;
623}
624
625static LIST_HEAD(all_bdevs);
626
627struct block_device *bdget(dev_t dev)
628{
629 struct block_device *bdev;
630 struct inode *inode;
631
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800632 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 bdev_test, bdev_set, &dev);
634
635 if (!inode)
636 return NULL;
637
638 bdev = &BDEV_I(inode)->bdev;
639
640 if (inode->i_state & I_NEW) {
641 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000642 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 bdev->bd_inode = inode;
644 bdev->bd_block_size = (1 << inode->i_blkbits);
645 bdev->bd_part_count = 0;
646 bdev->bd_invalidated = 0;
647 inode->i_mode = S_IFBLK;
648 inode->i_rdev = dev;
649 inode->i_bdev = bdev;
650 inode->i_data.a_ops = &def_blk_aops;
651 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 spin_lock(&bdev_lock);
653 list_add(&bdev->bd_list, &all_bdevs);
654 spin_unlock(&bdev_lock);
655 unlock_new_inode(inode);
656 }
657 return bdev;
658}
659
660EXPORT_SYMBOL(bdget);
661
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200662/**
663 * bdgrab -- Grab a reference to an already referenced block device
664 * @bdev: Block device to grab a reference to.
665 */
666struct block_device *bdgrab(struct block_device *bdev)
667{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400668 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200669 return bdev;
670}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700671EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200672
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673long nr_blockdev_pages(void)
674{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700675 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 long ret = 0;
677 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700678 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 ret += bdev->bd_inode->i_mapping->nrpages;
680 }
681 spin_unlock(&bdev_lock);
682 return ret;
683}
684
685void bdput(struct block_device *bdev)
686{
687 iput(bdev->bd_inode);
688}
689
690EXPORT_SYMBOL(bdput);
691
692static struct block_device *bd_acquire(struct inode *inode)
693{
694 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 spin_lock(&bdev_lock);
697 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700698 if (bdev) {
Al Viro7de9c6ee2010-10-23 11:11:40 -0400699 ihold(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 spin_unlock(&bdev_lock);
701 return bdev;
702 }
703 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 bdev = bdget(inode->i_rdev);
706 if (bdev) {
707 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700708 if (!inode->i_bdev) {
709 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400710 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700711 * and it's released in clear_inode() of inode.
712 * So, we can access it via ->i_mapping always
713 * without igrab().
714 */
Al Viro7de9c6ee2010-10-23 11:11:40 -0400715 ihold(bdev->bd_inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700716 inode->i_bdev = bdev;
717 inode->i_mapping = bdev->bd_inode->i_mapping;
718 list_add(&inode->i_devices, &bdev->bd_inodes);
719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 spin_unlock(&bdev_lock);
721 }
722 return bdev;
723}
724
725/* Call when you free inode */
726
727void bd_forget(struct inode *inode)
728{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700729 struct block_device *bdev = NULL;
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700732 if (!sb_is_blkdev_sb(inode->i_sb))
733 bdev = inode->i_bdev;
734 __bd_forget(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700736
737 if (bdev)
738 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739}
740
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900741/**
742 * bd_may_claim - test whether a block device can be claimed
743 * @bdev: block device of interest
744 * @whole: whole block device containing @bdev, may equal @bdev
745 * @holder: holder trying to claim @bdev
746 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300747 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900748 *
749 * CONTEXT:
750 * spin_lock(&bdev_lock).
751 *
752 * RETURNS:
753 * %true if @bdev can be claimed, %false otherwise.
754 */
755static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
756 void *holder)
757{
758 if (bdev->bd_holder == holder)
759 return true; /* already a holder */
760 else if (bdev->bd_holder != NULL)
761 return false; /* held by someone else */
762 else if (bdev->bd_contains == bdev)
763 return true; /* is a whole device which isn't held */
764
Tejun Heoe525fd82010-11-13 11:55:17 +0100765 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900766 return true; /* is a partition of a device that is being partitioned */
767 else if (whole->bd_holder != NULL)
768 return false; /* is a partition of a held device */
769 else
770 return true; /* is a partition of an un-held device */
771}
772
773/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900774 * bd_prepare_to_claim - prepare to claim a block device
775 * @bdev: block device of interest
776 * @whole: the whole device containing @bdev, may equal @bdev
777 * @holder: holder trying to claim @bdev
778 *
779 * Prepare to claim @bdev. This function fails if @bdev is already
780 * claimed by another holder and waits if another claiming is in
781 * progress. This function doesn't actually claim. On successful
782 * return, the caller has ownership of bd_claiming and bd_holder[s].
783 *
784 * CONTEXT:
785 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
786 * it multiple times.
787 *
788 * RETURNS:
789 * 0 if @bdev can be claimed, -EBUSY otherwise.
790 */
791static int bd_prepare_to_claim(struct block_device *bdev,
792 struct block_device *whole, void *holder)
793{
794retry:
795 /* if someone else claimed, fail */
796 if (!bd_may_claim(bdev, whole, holder))
797 return -EBUSY;
798
Tejun Heoe75aa852010-08-04 17:59:39 +0200799 /* if claiming is already in progress, wait for it to finish */
800 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900801 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
802 DEFINE_WAIT(wait);
803
804 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
805 spin_unlock(&bdev_lock);
806 schedule();
807 finish_wait(wq, &wait);
808 spin_lock(&bdev_lock);
809 goto retry;
810 }
811
812 /* yay, all mine */
813 return 0;
814}
815
816/**
817 * bd_start_claiming - start claiming a block device
818 * @bdev: block device of interest
819 * @holder: holder trying to claim @bdev
820 *
821 * @bdev is about to be opened exclusively. Check @bdev can be opened
822 * exclusively and mark that an exclusive open is in progress. Each
823 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000824 * either bd_finish_claiming() or bd_abort_claiming() (which do not
825 * fail).
826 *
827 * This function is used to gain exclusive access to the block device
828 * without actually causing other exclusive open attempts to fail. It
829 * should be used when the open sequence itself requires exclusive
830 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900831 *
832 * CONTEXT:
833 * Might sleep.
834 *
835 * RETURNS:
836 * Pointer to the block device containing @bdev on success, ERR_PTR()
837 * value on failure.
838 */
839static struct block_device *bd_start_claiming(struct block_device *bdev,
840 void *holder)
841{
842 struct gendisk *disk;
843 struct block_device *whole;
844 int partno, err;
845
846 might_sleep();
847
848 /*
849 * @bdev might not have been initialized properly yet, look up
850 * and grab the outer block device the hard way.
851 */
852 disk = get_gendisk(bdev->bd_dev, &partno);
853 if (!disk)
854 return ERR_PTR(-ENXIO);
855
Tejun Heod4c208b2011-06-13 12:45:48 +0200856 /*
857 * Normally, @bdev should equal what's returned from bdget_disk()
858 * if partno is 0; however, some drivers (floppy) use multiple
859 * bdev's for the same physical device and @bdev may be one of the
860 * aliases. Keep @bdev if partno is 0. This means claimer
861 * tracking is broken for those devices but it has always been that
862 * way.
863 */
864 if (partno)
865 whole = bdget_disk(disk, 0);
866 else
867 whole = bdgrab(bdev);
868
Nick Piggincf342572010-05-26 01:50:21 +1000869 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900870 put_disk(disk);
871 if (!whole)
872 return ERR_PTR(-ENOMEM);
873
874 /* prepare to claim, if successful, mark claiming in progress */
875 spin_lock(&bdev_lock);
876
877 err = bd_prepare_to_claim(bdev, whole, holder);
878 if (err == 0) {
879 whole->bd_claiming = holder;
880 spin_unlock(&bdev_lock);
881 return whole;
882 } else {
883 spin_unlock(&bdev_lock);
884 bdput(whole);
885 return ERR_PTR(err);
886 }
887}
888
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800889#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100890struct bd_holder_disk {
891 struct list_head list;
892 struct gendisk *disk;
893 int refcnt;
894};
895
896static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
897 struct gendisk *disk)
898{
899 struct bd_holder_disk *holder;
900
901 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
902 if (holder->disk == disk)
903 return holder;
904 return NULL;
905}
906
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700907static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800908{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700909 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800910}
911
912static void del_symlink(struct kobject *from, struct kobject *to)
913{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800914 sysfs_remove_link(from, kobject_name(to));
915}
916
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800917/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100918 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
919 * @bdev: the claimed slave bdev
920 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500921 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100922 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
923 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100924 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500925 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100926 * - from "slaves" directory of the holder @disk to the claimed @bdev
927 * - from "holders" directory of the @bdev to the holder @disk
928 *
929 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
930 * passed to bd_link_disk_holder(), then:
931 *
932 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
933 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
934 *
935 * The caller must have claimed @bdev before calling this function and
936 * ensure that both @bdev and @disk are valid during the creation and
937 * lifetime of these symlinks.
938 *
939 * CONTEXT:
940 * Might sleep.
941 *
942 * RETURNS:
943 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500944 */
Tejun Heoe09b4572010-11-13 11:55:17 +0100945int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500946{
Tejun Heo49731ba2011-01-14 18:43:57 +0100947 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100948 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800949
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800950 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500951
Tejun Heo49731ba2011-01-14 18:43:57 +0100952 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -0700953
Tejun Heoe09b4572010-11-13 11:55:17 +0100954 /* FIXME: remove the following once add_disk() handles errors */
955 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
956 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -0700957
Tejun Heo49731ba2011-01-14 18:43:57 +0100958 holder = bd_find_holder_disk(bdev, disk);
959 if (holder) {
960 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +0100961 goto out_unlock;
962 }
963
Tejun Heo49731ba2011-01-14 18:43:57 +0100964 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
965 if (!holder) {
966 ret = -ENOMEM;
967 goto out_unlock;
968 }
969
970 INIT_LIST_HEAD(&holder->list);
971 holder->disk = disk;
972 holder->refcnt = 1;
973
974 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
975 if (ret)
976 goto out_free;
977
978 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
979 if (ret)
980 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +0100981 /*
982 * bdev could be deleted beneath us which would implicitly destroy
983 * the holder directory. Hold on to it.
984 */
985 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100986
987 list_add(&holder->list, &bdev->bd_holder_disks);
988 goto out_unlock;
989
990out_del:
991 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
992out_free:
993 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +0100994out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800995 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +0100996 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800997}
Tejun Heoe09b4572010-11-13 11:55:17 +0100998EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800999
Tejun Heo49731ba2011-01-14 18:43:57 +01001000/**
1001 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1002 * @bdev: the calimed slave bdev
1003 * @disk: the holding disk
1004 *
1005 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1006 *
1007 * CONTEXT:
1008 * Might sleep.
1009 */
1010void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001011{
Tejun Heo49731ba2011-01-14 18:43:57 +01001012 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001013
Tejun Heo49731ba2011-01-14 18:43:57 +01001014 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001015
Tejun Heo49731ba2011-01-14 18:43:57 +01001016 holder = bd_find_holder_disk(bdev, disk);
1017
1018 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1019 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1020 del_symlink(bdev->bd_part->holder_dir,
1021 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001022 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001023 list_del_init(&holder->list);
1024 kfree(holder);
1025 }
1026
1027 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001028}
Tejun Heo49731ba2011-01-14 18:43:57 +01001029EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001030#endif
1031
Andrew Patterson0c002c22008-09-04 14:27:20 -06001032/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001033 * flush_disk - invalidates all buffer-cache entries on a disk
1034 *
1035 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001036 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001037 *
1038 * Invalidates all buffer-cache entries on a disk. It should be called
1039 * when a disk has been changed -- either by a media change or online
1040 * resize.
1041 */
NeilBrown93b270f2011-02-24 17:25:47 +11001042static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001043{
NeilBrown93b270f2011-02-24 17:25:47 +11001044 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001045 char name[BDEVNAME_SIZE] = "";
1046
1047 if (bdev->bd_disk)
1048 disk_name(bdev->bd_disk, 0, name);
1049 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1050 "resized disk %s\n", name);
1051 }
1052
1053 if (!bdev->bd_disk)
1054 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001055 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001056 bdev->bd_invalidated = 1;
1057}
1058
1059/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001060 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001061 * @disk: struct gendisk to check
1062 * @bdev: struct bdev to adjust.
1063 *
1064 * This routine checks to see if the bdev size does not match the disk size
1065 * and adjusts it if it differs.
1066 */
1067void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1068{
1069 loff_t disk_size, bdev_size;
1070
1071 disk_size = (loff_t)get_capacity(disk) << 9;
1072 bdev_size = i_size_read(bdev->bd_inode);
1073 if (disk_size != bdev_size) {
1074 char name[BDEVNAME_SIZE];
1075
1076 disk_name(disk, 0, name);
1077 printk(KERN_INFO
1078 "%s: detected capacity change from %lld to %lld\n",
1079 name, bdev_size, disk_size);
1080 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001081 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001082 }
1083}
1084EXPORT_SYMBOL(check_disk_size_change);
1085
1086/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001087 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001088 * @disk: struct gendisk to be revalidated
1089 *
1090 * This routine is a wrapper for lower-level driver's revalidate_disk
1091 * call-backs. It is used to do common pre and post operations needed
1092 * for all revalidate_disk operations.
1093 */
1094int revalidate_disk(struct gendisk *disk)
1095{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001096 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001097 int ret = 0;
1098
1099 if (disk->fops->revalidate_disk)
1100 ret = disk->fops->revalidate_disk(disk);
Martin K. Petersen25520d52015-10-21 13:19:49 -04001101 blk_integrity_revalidate(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001102 bdev = bdget_disk(disk, 0);
1103 if (!bdev)
1104 return ret;
1105
1106 mutex_lock(&bdev->bd_mutex);
1107 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001108 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001109 mutex_unlock(&bdev->bd_mutex);
1110 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001111 return ret;
1112}
1113EXPORT_SYMBOL(revalidate_disk);
1114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115/*
1116 * This routine checks whether a removable media has been changed,
1117 * and invalidates all buffer-cache-entries in that case. This
1118 * is a relatively slow routine, so we have to try to minimize using
1119 * it. Thus it is called only upon a 'mount' or 'open'. This
1120 * is the best way of combining speed and utility, I think.
1121 * People changing diskettes in the middle of an operation deserve
1122 * to lose :-)
1123 */
1124int check_disk_change(struct block_device *bdev)
1125{
1126 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001127 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001128 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129
Tejun Heo77ea8872010-12-08 20:57:37 +01001130 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1131 DISK_EVENT_EJECT_REQUEST);
1132 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return 0;
1134
NeilBrown93b270f2011-02-24 17:25:47 +11001135 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 if (bdops->revalidate_disk)
1137 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 return 1;
1139}
1140
1141EXPORT_SYMBOL(check_disk_change);
1142
1143void bd_set_size(struct block_device *bdev, loff_t size)
1144{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001145 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Guo Chaod646a022013-02-21 15:16:42 -08001147 mutex_lock(&bdev->bd_inode->i_mutex);
1148 i_size_write(bdev->bd_inode, size);
1149 mutex_unlock(&bdev->bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 while (bsize < PAGE_CACHE_SIZE) {
1151 if (size & bsize)
1152 break;
1153 bsize <<= 1;
1154 }
1155 bdev->bd_block_size = bsize;
1156 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1157}
1158EXPORT_SYMBOL(bd_set_size);
1159
Al Viro4385bab2013-05-05 22:11:03 -04001160static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001161
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001162/*
1163 * bd_mutex locking:
1164 *
1165 * mutex_lock(part->bd_mutex)
1166 * mutex_lock_nested(whole->bd_mutex, 1)
1167 */
1168
Al Viro572c4892007-10-08 13:24:05 -04001169static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001172 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001173 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001174 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001175 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Al Viro572c4892007-10-08 13:24:05 -04001177 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001178 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001179 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001180 perm |= MAY_WRITE;
1181 /*
1182 * hooks: /n/, see "layering violations".
1183 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001184 if (!for_part) {
1185 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1186 if (ret != 0) {
1187 bdput(bdev);
1188 return ret;
1189 }
Al Viro82666022008-08-01 05:32:04 -04001190 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001191
NeilBrownd3374822009-01-09 08:31:10 +11001192 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001193
Tejun Heo89f97492008-11-05 10:21:06 +01001194 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001195 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001196 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001197 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001198 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
Tejun Heo69e02c52011-03-09 19:54:27 +01001200 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001201 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 if (!bdev->bd_openers) {
1203 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001204 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 bdev->bd_contains = bdev;
Matthew Wilcoxbbab37d2015-07-03 10:40:42 -04001206 bdev->bd_inode->i_flags = disk->fops->direct_access ? S_DAX : 0;
Tejun Heocf771cb2008-09-03 09:01:09 +02001207 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001208 ret = -ENXIO;
1209 bdev->bd_part = disk_get_part(disk, partno);
1210 if (!bdev->bd_part)
1211 goto out_clear;
1212
Tejun Heo1196f8b2011-04-21 20:54:45 +02001213 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001215 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001216 if (ret == -ERESTARTSYS) {
1217 /* Lost a race with 'disk' being
1218 * deleted, try again.
1219 * See md.c
1220 */
1221 disk_put_part(bdev->bd_part);
1222 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001223 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001224 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001225 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001226 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001227 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001228 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001229 goto restart;
1230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 }
Tejun Heo7e697232011-05-23 13:26:07 +02001232
Christoph Hellwigb83ae6d2015-01-14 10:42:37 +01001233 if (!ret)
Tejun Heo7e697232011-05-23 13:26:07 +02001234 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Tejun Heo7e697232011-05-23 13:26:07 +02001235
Tejun Heo1196f8b2011-04-21 20:54:45 +02001236 /*
1237 * If the device is invalidated, rescan partition
1238 * if open succeeded or failed with -ENOMEDIUM.
1239 * The latter is necessary to prevent ghost
1240 * partitions on a removed medium.
1241 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001242 if (bdev->bd_invalidated) {
1243 if (!ret)
1244 rescan_partitions(disk, bdev);
1245 else if (ret == -ENOMEDIUM)
1246 invalidate_partitions(disk, bdev);
1247 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001248 if (ret)
1249 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 struct block_device *whole;
1252 whole = bdget_disk(disk, 0);
1253 ret = -ENOMEM;
1254 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001255 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001256 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001257 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001259 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001261 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001262 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001263 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001265 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 }
Tejun Heo89f97492008-11-05 10:21:06 +01001267 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Jeff Moyerf0b2e562015-08-14 16:15:32 -04001268 /*
1269 * If the partition is not aligned on a page
1270 * boundary, we can't do dax I/O to it.
1271 */
1272 if ((bdev->bd_part->start_sect % (PAGE_SIZE / 512)) ||
1273 (bdev->bd_part->nr_sects % (PAGE_SIZE / 512)))
1274 bdev->bd_inode->i_flags &= ~S_DAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 }
1276 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001278 ret = 0;
1279 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001280 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001281 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001282 if (bdev->bd_invalidated) {
1283 if (!ret)
1284 rescan_partitions(bdev->bd_disk, bdev);
1285 else if (ret == -ENOMEDIUM)
1286 invalidate_partitions(bdev->bd_disk, bdev);
1287 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001288 if (ret)
1289 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001291 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001292 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001293 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
1295 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001296 if (for_part)
1297 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001298 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001299 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 return 0;
1301
Tejun Heo0762b8b2008-08-25 19:56:12 +09001302 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001303 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001305 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001306 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001308 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001310 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001311 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001312 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001313 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001314 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001315 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001316 bdput(bdev);
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 return ret;
1319}
1320
Tejun Heod4d77622010-11-13 11:55:18 +01001321/**
1322 * blkdev_get - open a block device
1323 * @bdev: block_device to open
1324 * @mode: FMODE_* mask
1325 * @holder: exclusive holder identifier
1326 *
1327 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1328 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1329 * @holder is invalid. Exclusive opens may nest for the same @holder.
1330 *
1331 * On success, the reference count of @bdev is unchanged. On failure,
1332 * @bdev is put.
1333 *
1334 * CONTEXT:
1335 * Might sleep.
1336 *
1337 * RETURNS:
1338 * 0 on success, -errno on failure.
1339 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001340int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
Tejun Heoe525fd82010-11-13 11:55:17 +01001342 struct block_device *whole = NULL;
1343 int res;
1344
1345 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1346
1347 if ((mode & FMODE_EXCL) && holder) {
1348 whole = bd_start_claiming(bdev, holder);
1349 if (IS_ERR(whole)) {
1350 bdput(bdev);
1351 return PTR_ERR(whole);
1352 }
1353 }
1354
1355 res = __blkdev_get(bdev, mode, 0);
1356
1357 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001358 struct gendisk *disk = whole->bd_disk;
1359
Tejun Heo6a027ef2010-11-13 11:55:17 +01001360 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001361 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001362 spin_lock(&bdev_lock);
1363
Tejun Heo77ea8872010-12-08 20:57:37 +01001364 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001365 BUG_ON(!bd_may_claim(bdev, whole, holder));
1366 /*
1367 * Note that for a whole device bd_holders
1368 * will be incremented twice, and bd_holder
1369 * will be set to bd_may_claim before being
1370 * set to holder
1371 */
1372 whole->bd_holders++;
1373 whole->bd_holder = bd_may_claim;
1374 bdev->bd_holders++;
1375 bdev->bd_holder = holder;
1376 }
1377
1378 /* tell others that we're done */
1379 BUG_ON(whole->bd_claiming != holder);
1380 whole->bd_claiming = NULL;
1381 wake_up_bit(&whole->bd_claiming, 0);
1382
1383 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001384
1385 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001386 * Block event polling for write claims if requested. Any
1387 * write holder makes the write_holder state stick until
1388 * all are released. This is good enough and tracking
1389 * individual writeable reference is too fragile given the
1390 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001391 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001392 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1393 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001394 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001395 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001396 }
1397
1398 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001399 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001400 }
1401
1402 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001403}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404EXPORT_SYMBOL(blkdev_get);
1405
Tejun Heod4d77622010-11-13 11:55:18 +01001406/**
1407 * blkdev_get_by_path - open a block device by name
1408 * @path: path to the block device to open
1409 * @mode: FMODE_* mask
1410 * @holder: exclusive holder identifier
1411 *
1412 * Open the blockdevice described by the device file at @path. @mode
1413 * and @holder are identical to blkdev_get().
1414 *
1415 * On success, the returned block_device has reference count of one.
1416 *
1417 * CONTEXT:
1418 * Might sleep.
1419 *
1420 * RETURNS:
1421 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1422 */
1423struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1424 void *holder)
1425{
1426 struct block_device *bdev;
1427 int err;
1428
1429 bdev = lookup_bdev(path);
1430 if (IS_ERR(bdev))
1431 return bdev;
1432
1433 err = blkdev_get(bdev, mode, holder);
1434 if (err)
1435 return ERR_PTR(err);
1436
Chuck Ebberte51900f2011-02-16 18:11:53 -05001437 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1438 blkdev_put(bdev, mode);
1439 return ERR_PTR(-EACCES);
1440 }
1441
Tejun Heod4d77622010-11-13 11:55:18 +01001442 return bdev;
1443}
1444EXPORT_SYMBOL(blkdev_get_by_path);
1445
1446/**
1447 * blkdev_get_by_dev - open a block device by device number
1448 * @dev: device number of block device to open
1449 * @mode: FMODE_* mask
1450 * @holder: exclusive holder identifier
1451 *
1452 * Open the blockdevice described by device number @dev. @mode and
1453 * @holder are identical to blkdev_get().
1454 *
1455 * Use it ONLY if you really do not have anything better - i.e. when
1456 * you are behind a truly sucky interface and all you are given is a
1457 * device number. _Never_ to be used for internal purposes. If you
1458 * ever need it - reconsider your API.
1459 *
1460 * On success, the returned block_device has reference count of one.
1461 *
1462 * CONTEXT:
1463 * Might sleep.
1464 *
1465 * RETURNS:
1466 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1467 */
1468struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1469{
1470 struct block_device *bdev;
1471 int err;
1472
1473 bdev = bdget(dev);
1474 if (!bdev)
1475 return ERR_PTR(-ENOMEM);
1476
1477 err = blkdev_get(bdev, mode, holder);
1478 if (err)
1479 return ERR_PTR(err);
1480
1481 return bdev;
1482}
1483EXPORT_SYMBOL(blkdev_get_by_dev);
1484
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485static int blkdev_open(struct inode * inode, struct file * filp)
1486{
1487 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
1489 /*
1490 * Preserve backwards compatibility and allow large file access
1491 * even if userspace doesn't ask for it explicitly. Some mkfs
1492 * binary needs it. We might want to drop this workaround
1493 * during an unstable branch.
1494 */
1495 filp->f_flags |= O_LARGEFILE;
1496
Al Viro572c4892007-10-08 13:24:05 -04001497 if (filp->f_flags & O_NDELAY)
1498 filp->f_mode |= FMODE_NDELAY;
1499 if (filp->f_flags & O_EXCL)
1500 filp->f_mode |= FMODE_EXCL;
1501 if ((filp->f_flags & O_ACCMODE) == 3)
1502 filp->f_mode |= FMODE_WRITE_IOCTL;
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001505 if (bdev == NULL)
1506 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Al Viro572c4892007-10-08 13:24:05 -04001508 filp->f_mapping = bdev->bd_inode->i_mapping;
1509
Tejun Heoe525fd82010-11-13 11:55:17 +01001510 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511}
1512
Al Viro4385bab2013-05-05 22:11:03 -04001513static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001514{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001515 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001516 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001517
NeilBrown6796bf52006-12-08 02:36:16 -08001518 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001519 if (for_part)
1520 bdev->bd_part_count--;
1521
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001522 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001523 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001524 sync_blockdev(bdev);
1525 kill_bdev(bdev);
Christoph Hellwig564f00f2015-01-14 10:42:33 +01001526 /*
1527 * ->release can cause the queue to disappear, so flush all
1528 * dirty data before.
NeilBrown94007752011-09-10 17:20:21 +10001529 */
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001530 bdev_write_inode(bdev);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001531 }
1532 if (bdev->bd_contains == bdev) {
1533 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001534 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001535 }
1536 if (!bdev->bd_openers) {
1537 struct module *owner = disk->fops->owner;
1538
Tejun Heo0762b8b2008-08-25 19:56:12 +09001539 disk_put_part(bdev->bd_part);
1540 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001541 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001542 if (bdev != bdev->bd_contains)
1543 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001544 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001545
1546 put_disk(disk);
1547 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001548 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001549 mutex_unlock(&bdev->bd_mutex);
1550 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001551 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001552 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001553}
1554
Al Viro4385bab2013-05-05 22:11:03 -04001555void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001556{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001557 mutex_lock(&bdev->bd_mutex);
1558
Tejun Heoe525fd82010-11-13 11:55:17 +01001559 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001560 bool bdev_free;
1561
1562 /*
1563 * Release a claim on the device. The holder fields
1564 * are protected with bdev_lock. bd_mutex is to
1565 * synchronize disk_holder unlinking.
1566 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001567 spin_lock(&bdev_lock);
1568
1569 WARN_ON_ONCE(--bdev->bd_holders < 0);
1570 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1571
1572 /* bd_contains might point to self, check in a separate step */
1573 if ((bdev_free = !bdev->bd_holders))
1574 bdev->bd_holder = NULL;
1575 if (!bdev->bd_contains->bd_holders)
1576 bdev->bd_contains->bd_holder = NULL;
1577
1578 spin_unlock(&bdev_lock);
1579
Tejun Heo77ea8872010-12-08 20:57:37 +01001580 /*
1581 * If this was the last claim, remove holder link and
1582 * unblock evpoll if it was a write holder.
1583 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001584 if (bdev_free && bdev->bd_write_holder) {
1585 disk_unblock_events(bdev->bd_disk);
1586 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001587 }
Tejun Heo69362172011-03-09 19:54:27 +01001588 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001589
Tejun Heo85ef06d2011-07-01 16:17:47 +02001590 /*
1591 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1592 * event. This is to ensure detection of media removal commanded
1593 * from userland - e.g. eject(1).
1594 */
1595 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1596
1597 mutex_unlock(&bdev->bd_mutex);
1598
Al Viro4385bab2013-05-05 22:11:03 -04001599 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001600}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001601EXPORT_SYMBOL(blkdev_put);
1602
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603static int blkdev_close(struct inode * inode, struct file * filp)
1604{
1605 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
Al Viro4385bab2013-05-05 22:11:03 -04001606 blkdev_put(bdev, filp->f_mode);
1607 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608}
1609
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001610static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611{
Al Viro56b26ad2008-09-19 03:17:36 -04001612 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1613 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001614
1615 /*
1616 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1617 * to updated it before every ioctl.
1618 */
Al Viro56b26ad2008-09-19 03:17:36 -04001619 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001620 mode |= FMODE_NDELAY;
1621 else
1622 mode &= ~FMODE_NDELAY;
1623
Al Viro56b26ad2008-09-19 03:17:36 -04001624 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
1626
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001627/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001628 * Write data to the block device. Only intended for the block device itself
1629 * and the raw driver which basically is a fake block device.
1630 *
1631 * Does not take i_mutex for the write and thus is not for general purpose
1632 * use.
1633 */
Al Viro1456c0a2014-04-03 03:21:50 -04001634ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001635{
1636 struct file *file = iocb->ki_filp;
Al Viro7ec7b942015-04-07 11:35:14 -04001637 struct inode *bd_inode = file->f_mapping->host;
1638 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001639 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001640 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001641
Al Viro7ec7b942015-04-07 11:35:14 -04001642 if (bdev_read_only(I_BDEV(bd_inode)))
1643 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001644
Al Viro7ec7b942015-04-07 11:35:14 -04001645 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001646 return 0;
1647
Al Viro7ec7b942015-04-07 11:35:14 -04001648 if (iocb->ki_pos >= size)
1649 return -ENOSPC;
1650
1651 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001652
Jianpeng Ma53362a02012-08-02 09:50:39 +02001653 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001654 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwig02afc272013-09-04 15:04:40 +02001655 if (ret > 0) {
Christoph Hellwigeef99382009-08-20 17:43:41 +02001656 ssize_t err;
Al Viro1456c0a2014-04-03 03:21:50 -04001657 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
Dan Carpenter45d4f852014-04-03 14:47:17 -07001658 if (err < 0)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001659 ret = err;
1660 }
Jianpeng Ma53362a02012-08-02 09:50:39 +02001661 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001662 return ret;
1663}
Al Viro1456c0a2014-04-03 03:21:50 -04001664EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001665
David Jefferyb2de5252014-09-29 10:21:10 -04001666ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001667{
1668 struct file *file = iocb->ki_filp;
1669 struct inode *bd_inode = file->f_mapping->host;
1670 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001671 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001672
1673 if (pos >= size)
1674 return 0;
1675
1676 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001677 iov_iter_truncate(to, size);
1678 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001679}
David Jefferyb2de5252014-09-29 10:21:10 -04001680EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001681
Christoph Hellwigeef99382009-08-20 17:43:41 +02001682/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001683 * Try to release a page associated with block device when the system
1684 * is under memory pressure.
1685 */
1686static int blkdev_releasepage(struct page *page, gfp_t wait)
1687{
1688 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1689
1690 if (super && super->s_op->bdev_try_to_free_page)
1691 return super->s_op->bdev_try_to_free_page(super, page, wait);
1692
1693 return try_to_free_buffers(page);
1694}
1695
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001696static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001698 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001700 .write_begin = blkdev_write_begin,
1701 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001703 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001705 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706};
1707
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001708const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 .open = blkdev_open,
1710 .release = blkdev_close,
1711 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04001712 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04001713 .write_iter = blkdev_write_iter,
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001714 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001715 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001716 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717#ifdef CONFIG_COMPAT
1718 .compat_ioctl = compat_blkdev_ioctl,
1719#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001720 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001721 .splice_write = iter_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722};
1723
1724int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1725{
1726 int res;
1727 mm_segment_t old_fs = get_fs();
1728 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001729 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 set_fs(old_fs);
1731 return res;
1732}
1733
1734EXPORT_SYMBOL(ioctl_by_bdev);
1735
1736/**
1737 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001738 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001740 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 * namespace if possible and return it. Return ERR_PTR(error)
1742 * otherwise.
1743 */
Al Viro421748e2008-08-02 01:04:36 -04001744struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745{
1746 struct block_device *bdev;
1747 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001748 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 int error;
1750
Al Viro421748e2008-08-02 01:04:36 -04001751 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 return ERR_PTR(-EINVAL);
1753
Al Viro421748e2008-08-02 01:04:36 -04001754 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 if (error)
1756 return ERR_PTR(error);
1757
David Howellsbb6687342015-03-17 22:26:21 +00001758 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 error = -ENOTBLK;
1760 if (!S_ISBLK(inode->i_mode))
1761 goto fail;
1762 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001763 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 goto fail;
1765 error = -ENOMEM;
1766 bdev = bd_acquire(inode);
1767 if (!bdev)
1768 goto fail;
1769out:
Al Viro421748e2008-08-02 01:04:36 -04001770 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 return bdev;
1772fail:
1773 bdev = ERR_PTR(error);
1774 goto out;
1775}
Al Virod5686b42008-08-01 05:00:11 -04001776EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
NeilBrown93b270f2011-02-24 17:25:47 +11001778int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001779{
1780 struct super_block *sb = get_super(bdev);
1781 int res = 0;
1782
1783 if (sb) {
1784 /*
1785 * no need to lock the super, get_super holds the
1786 * read mutex so the filesystem cannot go away
1787 * under us (->put_super runs with the write lock
1788 * hold).
1789 */
1790 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001791 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001792 drop_super(sb);
1793 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001794 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001795 return res;
1796}
1797EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001798
1799void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1800{
1801 struct inode *inode, *old_inode = NULL;
1802
Dave Chinner74278da2015-03-04 12:37:22 -05001803 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001804 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1805 struct address_space *mapping = inode->i_mapping;
1806
1807 spin_lock(&inode->i_lock);
1808 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1809 mapping->nrpages == 0) {
1810 spin_unlock(&inode->i_lock);
1811 continue;
1812 }
1813 __iget(inode);
1814 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05001815 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001816 /*
1817 * We hold a reference to 'inode' so it couldn't have been
1818 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05001819 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02001820 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05001821 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02001822 * later.
1823 */
1824 iput(old_inode);
1825 old_inode = inode;
1826
1827 func(I_BDEV(inode), arg);
1828
Dave Chinner74278da2015-03-04 12:37:22 -05001829 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001830 }
Dave Chinner74278da2015-03-04 12:37:22 -05001831 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001832 iput(old_inode);
1833}