blob: 4a181fcb51751dc2cbc8fda10930a47bc883380e [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>
Dan Williamsb0686262017-01-26 20:37:35 -080021#include <linux/dax.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/buffer_head.h>
Al Viroff01bb42011-09-16 02:31:11 -040023#include <linux/swap.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010024#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010025#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/mpage.h>
27#include <linux/mount.h>
28#include <linux/uio.h>
29#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070030#include <linux/log2.h>
Al Viroff01bb42011-09-16 02:31:11 -040031#include <linux/cleancache.h>
Matthew Wilcoxc94c2ac2015-09-08 14:58:40 -070032#include <linux/dax.h>
Dan Williamsacc93d32016-05-07 11:40:28 -070033#include <linux/badblocks.h>
Christoph Hellwig189ce2b2016-10-31 11:59:25 -060034#include <linux/task_io_accounting_ops.h>
Darrick J. Wong25f4c412016-10-11 13:51:11 -070035#include <linux/falloc.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080036#include <linux/uaccess.h>
David Howells07f3f05c2006-09-30 20:52:18 +020037#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39struct bdev_inode {
40 struct block_device bdev;
41 struct inode vfs_inode;
42};
43
Adrian Bunk4c54ac62008-02-18 13:48:31 +010044static const struct address_space_operations def_blk_aops;
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046static inline struct bdev_inode *BDEV_I(struct inode *inode)
47{
48 return container_of(inode, struct bdev_inode, vfs_inode);
49}
50
Geert Uytterhoevenff5053f2015-06-26 13:58:32 +020051struct block_device *I_BDEV(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
53 return &BDEV_I(inode)->bdev;
54}
Linus Torvalds1da177e2005-04-16 15:20:36 -070055EXPORT_SYMBOL(I_BDEV);
56
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070057static void bdev_write_inode(struct block_device *bdev)
Christoph Hellwig564f00f2015-01-14 10:42:33 +010058{
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070059 struct inode *inode = bdev->bd_inode;
60 int ret;
61
Christoph Hellwig564f00f2015-01-14 10:42:33 +010062 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070065 ret = write_inode_now(inode, true);
66 if (ret) {
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
71 }
Christoph Hellwig564f00f2015-01-14 10:42:33 +010072 spin_lock(&inode->i_lock);
73 }
74 spin_unlock(&inode->i_lock);
75}
76
Peter Zijlstraf9a14392007-05-06 14:49:55 -070077/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb42011-09-16 02:31:11 -040078void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079{
Al Viroff01bb42011-09-16 02:31:11 -040080 struct address_space *mapping = bdev->bd_inode->i_mapping;
81
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080082 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070083 return;
Al Viroff01bb42011-09-16 02:31:11 -040084
Peter Zijlstraf9a14392007-05-06 14:49:55 -070085 invalidate_bh_lrus();
Al Viroff01bb42011-09-16 02:31:11 -040086 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087}
Al Viroff01bb42011-09-16 02:31:11 -040088EXPORT_SYMBOL(kill_bdev);
89
90/* Invalidate clean unused buffers and pagecache. */
91void invalidate_bdev(struct block_device *bdev)
92{
93 struct address_space *mapping = bdev->bd_inode->i_mapping;
94
Andrey Ryabinina5f6a6a2017-05-03 14:56:02 -070095 if (mapping->nrpages) {
96 invalidate_bh_lrus();
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping, 0, -1);
99 }
Al Viroff01bb42011-09-16 02:31:11 -0400100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
102 */
Dan Magenheimer31677602011-09-21 11:56:28 -0400103 cleancache_invalidate_inode(mapping);
Al Viroff01bb42011-09-16 02:31:11 -0400104}
105EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107int set_blocksize(struct block_device *bdev, int size)
108{
109 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700110 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 return -EINVAL;
112
113 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400114 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 return -EINVAL;
116
117 /* Don't change the size if it is same as current */
118 if (bdev->bd_block_size != size) {
119 sync_blockdev(bdev);
120 bdev->bd_block_size = size;
121 bdev->bd_inode->i_blkbits = blksize_bits(size);
122 kill_bdev(bdev);
123 }
124 return 0;
125}
126
127EXPORT_SYMBOL(set_blocksize);
128
129int sb_set_blocksize(struct super_block *sb, int size)
130{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 if (set_blocksize(sb->s_bdev, size))
132 return 0;
133 /* If we get here, we know size is power of two
134 * and it's value is between 512 and PAGE_SIZE */
135 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800136 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 return sb->s_blocksize;
138}
139
140EXPORT_SYMBOL(sb_set_blocksize);
141
142int sb_min_blocksize(struct super_block *sb, int size)
143{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400144 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 if (size < minsize)
146 size = minsize;
147 return sb_set_blocksize(sb, size);
148}
149
150EXPORT_SYMBOL(sb_min_blocksize);
151
152static int
153blkdev_get_block(struct inode *inode, sector_t iblock,
154 struct buffer_head *bh, int create)
155{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 bh->b_bdev = I_BDEV(inode);
157 bh->b_blocknr = iblock;
158 set_buffer_mapped(bh);
159 return 0;
160}
161
Dan Williams4ebb16c2015-10-28 07:48:19 +0900162static struct inode *bdev_file_inode(struct file *file)
163{
164 return file->f_mapping->host;
165}
166
Jens Axboe78250c02016-11-17 17:50:47 +0100167static unsigned int dio_bio_write_op(struct kiocb *iocb)
168{
169 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
170
171 /* avoid the need for a I/O completion work item */
172 if (iocb->ki_flags & IOCB_DSYNC)
173 op |= REQ_FUA;
174 return op;
175}
176
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600177#define DIO_INLINE_BIO_VECS 4
178
179static void blkdev_bio_end_io_simple(struct bio *bio)
180{
181 struct task_struct *waiter = bio->bi_private;
182
183 WRITE_ONCE(bio->bi_private, NULL);
184 wake_up_process(waiter);
185}
186
187static ssize_t
188__blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
189 int nr_pages)
190{
191 struct file *file = iocb->ki_filp;
192 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Jens Axboe72ecad22016-11-16 23:11:42 -0700193 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600194 loff_t pos = iocb->ki_pos;
195 bool should_dirty = false;
196 struct bio bio;
197 ssize_t ret;
198 blk_qc_t qc;
199 int i;
200
Jens Axboe9a794fb2016-11-22 08:12:39 -0700201 if ((pos | iov_iter_alignment(iter)) &
202 (bdev_logical_block_size(bdev) - 1))
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600203 return -EINVAL;
204
Jens Axboe72ecad22016-11-16 23:11:42 -0700205 if (nr_pages <= DIO_INLINE_BIO_VECS)
206 vecs = inline_vecs;
207 else {
208 vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL);
209 if (!vecs)
210 return -ENOMEM;
211 }
212
Ming Lei3a83f462016-11-22 08:57:21 -0700213 bio_init(&bio, vecs, nr_pages);
Christoph Hellwig74d46992017-08-23 19:10:32 +0200214 bio_set_dev(&bio, bdev);
Damien Le Moal4d1a4762016-11-22 15:38:49 +0900215 bio.bi_iter.bi_sector = pos >> 9;
Jens Axboe45d06cf2017-06-27 11:01:22 -0600216 bio.bi_write_hint = iocb->ki_hint;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600217 bio.bi_private = current;
218 bio.bi_end_io = blkdev_bio_end_io_simple;
219
220 ret = bio_iov_iter_get_pages(&bio, iter);
221 if (unlikely(ret))
222 return ret;
223 ret = bio.bi_iter.bi_size;
224
225 if (iov_iter_rw(iter) == READ) {
Jens Axboe78250c02016-11-17 17:50:47 +0100226 bio.bi_opf = REQ_OP_READ;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600227 if (iter_is_iovec(iter))
228 should_dirty = true;
229 } else {
Jens Axboe78250c02016-11-17 17:50:47 +0100230 bio.bi_opf = dio_bio_write_op(iocb);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600231 task_io_account_write(ret);
232 }
233
234 qc = submit_bio(&bio);
235 for (;;) {
236 set_current_state(TASK_UNINTERRUPTIBLE);
237 if (!READ_ONCE(bio.bi_private))
238 break;
239 if (!(iocb->ki_flags & IOCB_HIPRI) ||
Christoph Hellwigea435e12017-11-02 21:29:54 +0300240 !blk_poll(bdev_get_queue(bdev), qc))
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600241 io_schedule();
242 }
243 __set_current_state(TASK_RUNNING);
244
245 bio_for_each_segment_all(bvec, &bio, i) {
246 if (should_dirty && !PageCompound(bvec->bv_page))
247 set_page_dirty_lock(bvec->bv_page);
248 put_page(bvec->bv_page);
249 }
250
Jens Axboe72ecad22016-11-16 23:11:42 -0700251 if (vecs != inline_vecs)
252 kfree(vecs);
253
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200254 if (unlikely(bio.bi_status))
Linus Torvaldsc6b1e362017-07-03 10:34:51 -0700255 ret = blk_status_to_errno(bio.bi_status);
Jens Axboe9ae3b3f2017-06-28 15:30:13 -0600256
257 bio_uninit(&bio);
258
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600259 return ret;
260}
261
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700262struct blkdev_dio {
263 union {
264 struct kiocb *iocb;
265 struct task_struct *waiter;
266 };
267 size_t size;
268 atomic_t ref;
269 bool multi_bio : 1;
270 bool should_dirty : 1;
271 bool is_sync : 1;
272 struct bio bio;
273};
274
275static struct bio_set *blkdev_dio_pool __read_mostly;
276
277static void blkdev_bio_end_io(struct bio *bio)
278{
279 struct blkdev_dio *dio = bio->bi_private;
280 bool should_dirty = dio->should_dirty;
281
282 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200283 if (bio->bi_status && !dio->bio.bi_status)
284 dio->bio.bi_status = bio->bi_status;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700285 } else {
286 if (!dio->is_sync) {
287 struct kiocb *iocb = dio->iocb;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200288 ssize_t ret;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700289
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200290 if (likely(!dio->bio.bi_status)) {
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700291 ret = dio->size;
292 iocb->ki_pos += ret;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200293 } else {
294 ret = blk_status_to_errno(dio->bio.bi_status);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700295 }
296
297 dio->iocb->ki_complete(iocb, ret, 0);
298 bio_put(&dio->bio);
299 } else {
300 struct task_struct *waiter = dio->waiter;
301
302 WRITE_ONCE(dio->waiter, NULL);
303 wake_up_process(waiter);
304 }
305 }
306
307 if (should_dirty) {
308 bio_check_pages_dirty(bio);
309 } else {
310 struct bio_vec *bvec;
311 int i;
312
313 bio_for_each_segment_all(bvec, bio, i)
314 put_page(bvec->bv_page);
315 bio_put(bio);
316 }
317}
318
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800319static ssize_t
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700320__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800321{
322 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +0900323 struct inode *inode = bdev_file_inode(file);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700324 struct block_device *bdev = I_BDEV(inode);
Christoph Hellwig64d656a2016-12-22 19:20:45 +0100325 struct blk_plug plug;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700326 struct blkdev_dio *dio;
327 struct bio *bio;
Christoph Hellwig690e5322017-01-24 14:50:19 +0100328 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700329 loff_t pos = iocb->ki_pos;
330 blk_qc_t qc = BLK_QC_T_NONE;
Christoph Hellwig36ffc6c2017-06-03 09:38:00 +0200331 int ret = 0;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700332
Jens Axboe9a794fb2016-11-22 08:12:39 -0700333 if ((pos | iov_iter_alignment(iter)) &
334 (bdev_logical_block_size(bdev) - 1))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700335 return -EINVAL;
336
337 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool);
338 bio_get(bio); /* extra ref for the completion handler */
339
340 dio = container_of(bio, struct blkdev_dio, bio);
Christoph Hellwig690e5322017-01-24 14:50:19 +0100341 dio->is_sync = is_sync = is_sync_kiocb(iocb);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700342 if (dio->is_sync)
343 dio->waiter = current;
344 else
345 dio->iocb = iocb;
346
347 dio->size = 0;
348 dio->multi_bio = false;
349 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
350
Christoph Hellwig64d656a2016-12-22 19:20:45 +0100351 blk_start_plug(&plug);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700352 for (;;) {
Christoph Hellwig74d46992017-08-23 19:10:32 +0200353 bio_set_dev(bio, bdev);
Damien Le Moal4d1a4762016-11-22 15:38:49 +0900354 bio->bi_iter.bi_sector = pos >> 9;
Jens Axboe45d06cf2017-06-27 11:01:22 -0600355 bio->bi_write_hint = iocb->ki_hint;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700356 bio->bi_private = dio;
357 bio->bi_end_io = blkdev_bio_end_io;
358
359 ret = bio_iov_iter_get_pages(bio, iter);
360 if (unlikely(ret)) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200361 bio->bi_status = BLK_STS_IOERR;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700362 bio_endio(bio);
363 break;
364 }
365
366 if (is_read) {
367 bio->bi_opf = REQ_OP_READ;
368 if (dio->should_dirty)
369 bio_set_pages_dirty(bio);
370 } else {
371 bio->bi_opf = dio_bio_write_op(iocb);
372 task_io_account_write(bio->bi_iter.bi_size);
373 }
374
375 dio->size += bio->bi_iter.bi_size;
376 pos += bio->bi_iter.bi_size;
377
378 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
379 if (!nr_pages) {
380 qc = submit_bio(bio);
381 break;
382 }
383
384 if (!dio->multi_bio) {
385 dio->multi_bio = true;
386 atomic_set(&dio->ref, 2);
387 } else {
388 atomic_inc(&dio->ref);
389 }
390
391 submit_bio(bio);
392 bio = bio_alloc(GFP_KERNEL, nr_pages);
393 }
Christoph Hellwig64d656a2016-12-22 19:20:45 +0100394 blk_finish_plug(&plug);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700395
Christoph Hellwig690e5322017-01-24 14:50:19 +0100396 if (!is_sync)
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700397 return -EIOCBQUEUED;
398
399 for (;;) {
400 set_current_state(TASK_UNINTERRUPTIBLE);
401 if (!READ_ONCE(dio->waiter))
402 break;
403
404 if (!(iocb->ki_flags & IOCB_HIPRI) ||
Christoph Hellwigea435e12017-11-02 21:29:54 +0300405 !blk_poll(bdev_get_queue(bdev), qc))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700406 io_schedule();
407 }
408 __set_current_state(TASK_RUNNING);
409
Christoph Hellwig36ffc6c2017-06-03 09:38:00 +0200410 if (!ret)
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200411 ret = blk_status_to_errno(dio->bio.bi_status);
Shaohua Li7a62a522016-12-13 19:07:55 -0800412 if (likely(!ret))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700413 ret = dio->size;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700414
415 bio_put(&dio->bio);
416 return ret;
417}
418
419static ssize_t
420blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
421{
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600422 int nr_pages;
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800423
Jens Axboe72ecad22016-11-16 23:11:42 -0700424 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600425 if (!nr_pages)
426 return 0;
Jens Axboe72ecad22016-11-16 23:11:42 -0700427 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600428 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700429
430 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800431}
432
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700433static __init int blkdev_init(void)
434{
NeilBrown011067b2017-06-18 14:38:57 +1000435 blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700436 if (!blkdev_dio_pool)
437 return -ENOMEM;
438 return 0;
439}
440module_init(blkdev_init);
441
Jan Kara5cee5812009-04-27 16:43:51 +0200442int __sync_blockdev(struct block_device *bdev, int wait)
443{
444 if (!bdev)
445 return 0;
446 if (!wait)
447 return filemap_flush(bdev->bd_inode->i_mapping);
448 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
449}
450
Nick Piggin585d3bc2009-02-25 10:44:19 +0100451/*
452 * Write out and wait upon all the dirty data associated with a block
453 * device via its mapping. Does not take the superblock lock.
454 */
455int sync_blockdev(struct block_device *bdev)
456{
Jan Kara5cee5812009-04-27 16:43:51 +0200457 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100458}
459EXPORT_SYMBOL(sync_blockdev);
460
461/*
462 * Write out and wait upon all dirty data associated with this
463 * device. Filesystem data as well as the underlying block
464 * device. Takes the superblock lock.
465 */
466int fsync_bdev(struct block_device *bdev)
467{
468 struct super_block *sb = get_super(bdev);
469 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200470 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100471 drop_super(sb);
472 return res;
473 }
474 return sync_blockdev(bdev);
475}
Al Viro47e44912009-04-01 07:07:16 -0400476EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100477
478/**
479 * freeze_bdev -- lock a filesystem and force it into a consistent state
480 * @bdev: blockdevice to lock
481 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100482 * If a superblock is found on this device, we take the s_umount semaphore
483 * on it to make sure nobody unmounts until the snapshot creation is done.
484 * The reference counter (bd_fsfreeze_count) guarantees that only the last
485 * unfreeze process can unfreeze the frozen filesystem actually when multiple
486 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
487 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
488 * actually.
489 */
490struct super_block *freeze_bdev(struct block_device *bdev)
491{
492 struct super_block *sb;
493 int error = 0;
494
495 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200496 if (++bdev->bd_fsfreeze_count > 1) {
497 /*
498 * We don't even need to grab a reference - the first call
499 * to freeze_bdev grab an active reference and only the last
500 * thaw_bdev drops it.
501 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100502 sb = get_super(bdev);
Andrey Ryabinin5bb53c02016-08-23 18:55:31 +0300503 if (sb)
504 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100505 mutex_unlock(&bdev->bd_fsfreeze_mutex);
506 return sb;
507 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100508
Christoph Hellwig45042302009-08-03 23:28:35 +0200509 sb = get_active_super(bdev);
510 if (!sb)
511 goto out;
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600512 if (sb->s_op->freeze_super)
513 error = sb->s_op->freeze_super(sb);
514 else
515 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400516 if (error) {
517 deactivate_super(sb);
518 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200519 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400520 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100521 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400522 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200523 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100524 sync_blockdev(bdev);
525 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200526 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100527}
528EXPORT_SYMBOL(freeze_bdev);
529
530/**
531 * thaw_bdev -- unlock filesystem
532 * @bdev: blockdevice to unlock
533 * @sb: associated superblock
534 *
535 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
536 */
537int thaw_bdev(struct block_device *bdev, struct super_block *sb)
538{
Christoph Hellwig45042302009-08-03 23:28:35 +0200539 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100540
541 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200542 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400543 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100544
Christoph Hellwig45042302009-08-03 23:28:35 +0200545 error = 0;
546 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400547 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100548
Christoph Hellwig45042302009-08-03 23:28:35 +0200549 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400550 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200551
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600552 if (sb->s_op->thaw_super)
553 error = sb->s_op->thaw_super(sb);
554 else
555 error = thaw_super(sb);
Pierre Morel997198b2016-10-04 10:53:40 +0200556 if (error)
Josef Bacik18e9e512010-03-23 10:34:56 -0400557 bdev->bd_fsfreeze_count++;
Josef Bacik18e9e512010-03-23 10:34:56 -0400558out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100559 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Pierre Morel997198b2016-10-04 10:53:40 +0200560 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100561}
562EXPORT_SYMBOL(thaw_bdev);
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
565{
566 return block_write_full_page(page, blkdev_get_block, wbc);
567}
568
569static int blkdev_readpage(struct file * file, struct page * page)
570{
571 return block_read_full_page(page, blkdev_get_block);
572}
573
Akinobu Mita447f05b2014-10-09 15:26:58 -0700574static int blkdev_readpages(struct file *file, struct address_space *mapping,
575 struct list_head *pages, unsigned nr_pages)
576{
577 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
578}
579
Nick Piggin6272b5a2007-10-16 01:25:04 -0700580static int blkdev_write_begin(struct file *file, struct address_space *mapping,
581 loff_t pos, unsigned len, unsigned flags,
582 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200584 return block_write_begin(mapping, pos, len, flags, pagep,
585 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
Nick Piggin6272b5a2007-10-16 01:25:04 -0700588static int blkdev_write_end(struct file *file, struct address_space *mapping,
589 loff_t pos, unsigned len, unsigned copied,
590 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700592 int ret;
593 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
594
595 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300596 put_page(page);
Nick Piggin6272b5a2007-10-16 01:25:04 -0700597
598 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
601/*
602 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500603 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 * so we compute the size by hand (just as in block_read/write above)
605 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800606static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900608 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 loff_t retval;
610
Al Viro59551022016-01-22 15:40:57 -0500611 inode_lock(bd_inode);
Al Viro5d48f3a2013-06-23 21:34:45 +0400612 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Al Viro59551022016-01-22 15:40:57 -0500613 inode_unlock(bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 return retval;
615}
616
Josef Bacik02c24a82011-07-16 20:44:56 -0400617int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900619 struct inode *bd_inode = bdev_file_inode(filp);
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400620 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100621 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200622
Jeff Layton372cf242017-07-06 07:02:28 -0400623 error = file_write_and_wait_range(filp, start, end);
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200624 if (error)
625 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100626
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400627 /*
628 * There is no need to serialise calls to blkdev_issue_flush with
629 * i_mutex and doing so causes performance issues with concurrent
630 * O_SYNC writers to a block device.
631 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200632 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100633 if (error == -EOPNOTSUPP)
634 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400635
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100636 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700638EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700640/**
641 * bdev_read_page() - Start reading a page from a block device
642 * @bdev: The device to read the page from
643 * @sector: The offset on the device to read the page to (need not be aligned)
644 * @page: The page to read
645 *
646 * On entry, the page should be locked. It will be unlocked when the page
647 * has been read. If the block driver implements rw_page synchronously,
648 * that will be true on exit from this function, but it need not be.
649 *
650 * Errors returned by this function are usually "soft", eg out of memory, or
651 * queue full; callers should try a different route to read this page rather
652 * than propagate an error back up the stack.
653 *
654 * Return: negative errno if an error occurs, 0 if submission was successful.
655 */
656int bdev_read_page(struct block_device *bdev, sector_t sector,
657 struct page *page)
658{
659 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800660 int result = -EOPNOTSUPP;
661
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400662 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc92015-11-19 13:29:28 -0800663 return result;
664
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800665 result = blk_queue_enter(bdev->bd_queue, 0);
Dan Williams2e6edc92015-11-19 13:29:28 -0800666 if (result)
667 return result;
Jens Axboec11f0c02016-08-05 08:11:04 -0600668 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
Dan Williams2e6edc92015-11-19 13:29:28 -0800669 blk_queue_exit(bdev->bd_queue);
670 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700671}
672EXPORT_SYMBOL_GPL(bdev_read_page);
673
674/**
675 * bdev_write_page() - Start writing a page to a block device
676 * @bdev: The device to write the page to
677 * @sector: The offset on the device to write the page to (need not be aligned)
678 * @page: The page to write
679 * @wbc: The writeback_control for the write
680 *
681 * On entry, the page should be locked and not currently under writeback.
682 * On exit, if the write started successfully, the page will be unlocked and
683 * under writeback. If the write failed already (eg the driver failed to
684 * queue the page to the device), the page will still be locked. If the
685 * caller is a ->writepage implementation, it will need to unlock the page.
686 *
687 * Errors returned by this function are usually "soft", eg out of memory, or
688 * queue full; callers should try a different route to write this page rather
689 * than propagate an error back up the stack.
690 *
691 * Return: negative errno if an error occurs, 0 if submission was successful.
692 */
693int bdev_write_page(struct block_device *bdev, sector_t sector,
694 struct page *page, struct writeback_control *wbc)
695{
696 int result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700697 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc92015-11-19 13:29:28 -0800698
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400699 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700700 return -EOPNOTSUPP;
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800701 result = blk_queue_enter(bdev->bd_queue, 0);
Dan Williams2e6edc92015-11-19 13:29:28 -0800702 if (result)
703 return result;
704
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700705 set_page_writeback(page);
Jens Axboec11f0c02016-08-05 08:11:04 -0600706 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700707 if (result) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700708 end_page_writeback(page);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700709 } else {
710 clean_page_buffers(page);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700711 unlock_page(page);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700712 }
Dan Williams2e6edc92015-11-19 13:29:28 -0800713 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700714 return result;
715}
716EXPORT_SYMBOL_GPL(bdev_write_page);
717
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718/*
719 * pseudo-fs
720 */
721
722static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800723static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
725static struct inode *bdev_alloc_inode(struct super_block *sb)
726{
Christoph Lametere94b1762006-12-06 20:33:17 -0800727 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 if (!ei)
729 return NULL;
730 return &ei->vfs_inode;
731}
732
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100733static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100735 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 struct bdev_inode *bdi = BDEV_I(inode);
737
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 kmem_cache_free(bdev_cachep, bdi);
739}
740
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100741static void bdev_destroy_inode(struct inode *inode)
742{
743 call_rcu(&inode->i_rcu, bdev_i_callback);
744}
745
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700746static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
748 struct bdev_inode *ei = (struct bdev_inode *) foo;
749 struct block_device *bdev = &ei->bdev;
750
Christoph Lametera35afb82007-05-16 22:10:57 -0700751 memset(bdev, 0, sizeof(*bdev));
752 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700753 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100754#ifdef CONFIG_SYSFS
755 INIT_LIST_HEAD(&bdev->bd_holder_disks);
756#endif
Jan Karaa5a79d02017-03-02 16:50:13 +0100757 bdev->bd_bdi = &noop_backing_dev_info;
Christoph Lametera35afb82007-05-16 22:10:57 -0700758 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800759 /* Initialize mutex for freeze. */
760 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
Al Virob57922d2010-06-07 14:34:48 -0400763static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764{
765 struct block_device *bdev = &BDEV_I(inode)->bdev;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700766 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400767 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200768 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 spin_lock(&bdev_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 list_del_init(&bdev->bd_list);
771 spin_unlock(&bdev_lock);
Jan Karaf7597412017-03-23 01:37:00 +0100772 /* Detach inode from wb early as bdi_put() may free bdi->wb */
773 inode_detach_wb(inode);
Jan Karaa5a79d02017-03-02 16:50:13 +0100774 if (bdev->bd_bdi != &noop_backing_dev_info) {
Jan Karab1d2dc562017-02-02 15:56:52 +0100775 bdi_put(bdev->bd_bdi);
Jan Karaa5a79d02017-03-02 16:50:13 +0100776 bdev->bd_bdi = &noop_backing_dev_info;
777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800780static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 .statfs = simple_statfs,
782 .alloc_inode = bdev_alloc_inode,
783 .destroy_inode = bdev_destroy_inode,
784 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400785 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786};
787
Al Viro51139ad2010-07-25 23:47:46 +0400788static struct dentry *bd_mount(struct file_system_type *fs_type,
789 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
Shaohua Li3684aa72016-02-22 15:27:40 -0700791 struct dentry *dent;
792 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Vegard Nossume9e5e3f2016-08-22 12:47:43 +0200793 if (!IS_ERR(dent))
Shaohua Li3684aa72016-02-22 15:27:40 -0700794 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
795 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
797
798static struct file_system_type bd_type = {
799 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400800 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 .kill_sb = kill_anon_super,
802};
803
Tejun Heoa212b102015-05-22 17:13:33 -0400804struct super_block *blockdev_superblock __read_mostly;
805EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
807void __init bdev_cache_init(void)
808{
809 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300810 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800811
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800813 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800814 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900815 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 err = register_filesystem(&bd_type);
817 if (err)
818 panic("Cannot register bdev pseudo-fs");
819 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 if (IS_ERR(bd_mnt))
821 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300822 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823}
824
825/*
826 * Most likely _very_ bad one - but then it's hardly critical for small
827 * /dev and can be fixed when somebody will need really large one.
828 * Keep in mind that it will be fed through icache hash function too.
829 */
830static inline unsigned long hash(dev_t dev)
831{
832 return MAJOR(dev)+MINOR(dev);
833}
834
835static int bdev_test(struct inode *inode, void *data)
836{
837 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
838}
839
840static int bdev_set(struct inode *inode, void *data)
841{
842 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
843 return 0;
844}
845
846static LIST_HEAD(all_bdevs);
847
Jan Karaf44f1ab2017-02-02 15:56:49 +0100848/*
849 * If there is a bdev inode for this device, unhash it so that it gets evicted
850 * as soon as last inode reference is dropped.
851 */
852void bdev_unhash_inode(dev_t dev)
853{
854 struct inode *inode;
855
856 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
857 if (inode) {
858 remove_inode_hash(inode);
859 iput(inode);
860 }
861}
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863struct block_device *bdget(dev_t dev)
864{
865 struct block_device *bdev;
866 struct inode *inode;
867
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800868 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 bdev_test, bdev_set, &dev);
870
871 if (!inode)
872 return NULL;
873
874 bdev = &BDEV_I(inode)->bdev;
875
876 if (inode->i_state & I_NEW) {
877 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000878 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 bdev->bd_inode = inode;
Fabian Frederick93407472017-02-27 14:28:32 -0800880 bdev->bd_block_size = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 bdev->bd_part_count = 0;
882 bdev->bd_invalidated = 0;
883 inode->i_mode = S_IFBLK;
884 inode->i_rdev = dev;
885 inode->i_bdev = bdev;
886 inode->i_data.a_ops = &def_blk_aops;
887 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 spin_lock(&bdev_lock);
889 list_add(&bdev->bd_list, &all_bdevs);
890 spin_unlock(&bdev_lock);
891 unlock_new_inode(inode);
892 }
893 return bdev;
894}
895
896EXPORT_SYMBOL(bdget);
897
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200898/**
899 * bdgrab -- Grab a reference to an already referenced block device
900 * @bdev: Block device to grab a reference to.
901 */
902struct block_device *bdgrab(struct block_device *bdev)
903{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400904 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200905 return bdev;
906}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700907EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909long nr_blockdev_pages(void)
910{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700911 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 long ret = 0;
913 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700914 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 ret += bdev->bd_inode->i_mapping->nrpages;
916 }
917 spin_unlock(&bdev_lock);
918 return ret;
919}
920
921void bdput(struct block_device *bdev)
922{
923 iput(bdev->bd_inode);
924}
925
926EXPORT_SYMBOL(bdput);
927
928static struct block_device *bd_acquire(struct inode *inode)
929{
930 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 spin_lock(&bdev_lock);
933 bdev = inode->i_bdev;
Jan Karacccd9fb2017-02-21 18:09:48 +0100934 if (bdev && !inode_unhashed(bdev->bd_inode)) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100935 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 spin_unlock(&bdev_lock);
937 return bdev;
938 }
939 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700940
Jan Karacccd9fb2017-02-21 18:09:48 +0100941 /*
942 * i_bdev references block device inode that was already shut down
943 * (corresponding device got removed). Remove the reference and look
944 * up block device inode again just in case new device got
945 * reestablished under the same device number.
946 */
947 if (bdev)
948 bd_forget(inode);
949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 bdev = bdget(inode->i_rdev);
951 if (bdev) {
952 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700953 if (!inode->i_bdev) {
954 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400955 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700956 * and it's released in clear_inode() of inode.
957 * So, we can access it via ->i_mapping always
958 * without igrab().
959 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100960 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700961 inode->i_bdev = bdev;
962 inode->i_mapping = bdev->bd_inode->i_mapping;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 spin_unlock(&bdev_lock);
965 }
966 return bdev;
967}
968
969/* Call when you free inode */
970
971void bd_forget(struct inode *inode)
972{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700973 struct block_device *bdev = NULL;
974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -0700976 if (!sb_is_blkdev_sb(inode->i_sb))
977 bdev = inode->i_bdev;
Al Viroa4a4f942016-07-19 13:16:52 -0400978 inode->i_bdev = NULL;
979 inode->i_mapping = &inode->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700981
982 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100983 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984}
985
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900986/**
987 * bd_may_claim - test whether a block device can be claimed
988 * @bdev: block device of interest
989 * @whole: whole block device containing @bdev, may equal @bdev
990 * @holder: holder trying to claim @bdev
991 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300992 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900993 *
994 * CONTEXT:
995 * spin_lock(&bdev_lock).
996 *
997 * RETURNS:
998 * %true if @bdev can be claimed, %false otherwise.
999 */
1000static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1001 void *holder)
1002{
1003 if (bdev->bd_holder == holder)
1004 return true; /* already a holder */
1005 else if (bdev->bd_holder != NULL)
1006 return false; /* held by someone else */
NeilBrownbcc7f5b2016-12-12 08:21:51 -07001007 else if (whole == bdev)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001008 return true; /* is a whole device which isn't held */
1009
Tejun Heoe525fd82010-11-13 11:55:17 +01001010 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001011 return true; /* is a partition of a device that is being partitioned */
1012 else if (whole->bd_holder != NULL)
1013 return false; /* is a partition of a held device */
1014 else
1015 return true; /* is a partition of an un-held device */
1016}
1017
1018/**
Tejun Heo6b4517a2010-04-07 18:53:59 +09001019 * bd_prepare_to_claim - prepare to claim a block device
1020 * @bdev: block device of interest
1021 * @whole: the whole device containing @bdev, may equal @bdev
1022 * @holder: holder trying to claim @bdev
1023 *
1024 * Prepare to claim @bdev. This function fails if @bdev is already
1025 * claimed by another holder and waits if another claiming is in
1026 * progress. This function doesn't actually claim. On successful
1027 * return, the caller has ownership of bd_claiming and bd_holder[s].
1028 *
1029 * CONTEXT:
1030 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1031 * it multiple times.
1032 *
1033 * RETURNS:
1034 * 0 if @bdev can be claimed, -EBUSY otherwise.
1035 */
1036static int bd_prepare_to_claim(struct block_device *bdev,
1037 struct block_device *whole, void *holder)
1038{
1039retry:
1040 /* if someone else claimed, fail */
1041 if (!bd_may_claim(bdev, whole, holder))
1042 return -EBUSY;
1043
Tejun Heoe75aa852010-08-04 17:59:39 +02001044 /* if claiming is already in progress, wait for it to finish */
1045 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +09001046 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1047 DEFINE_WAIT(wait);
1048
1049 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1050 spin_unlock(&bdev_lock);
1051 schedule();
1052 finish_wait(wq, &wait);
1053 spin_lock(&bdev_lock);
1054 goto retry;
1055 }
1056
1057 /* yay, all mine */
1058 return 0;
1059}
1060
1061/**
1062 * bd_start_claiming - start claiming a block device
1063 * @bdev: block device of interest
1064 * @holder: holder trying to claim @bdev
1065 *
1066 * @bdev is about to be opened exclusively. Check @bdev can be opened
1067 * exclusively and mark that an exclusive open is in progress. Each
1068 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +10001069 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1070 * fail).
1071 *
1072 * This function is used to gain exclusive access to the block device
1073 * without actually causing other exclusive open attempts to fail. It
1074 * should be used when the open sequence itself requires exclusive
1075 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +09001076 *
1077 * CONTEXT:
1078 * Might sleep.
1079 *
1080 * RETURNS:
1081 * Pointer to the block device containing @bdev on success, ERR_PTR()
1082 * value on failure.
1083 */
1084static struct block_device *bd_start_claiming(struct block_device *bdev,
1085 void *holder)
1086{
1087 struct gendisk *disk;
1088 struct block_device *whole;
1089 int partno, err;
1090
1091 might_sleep();
1092
1093 /*
1094 * @bdev might not have been initialized properly yet, look up
1095 * and grab the outer block device the hard way.
1096 */
1097 disk = get_gendisk(bdev->bd_dev, &partno);
1098 if (!disk)
1099 return ERR_PTR(-ENXIO);
1100
Tejun Heod4c208b2011-06-13 12:45:48 +02001101 /*
1102 * Normally, @bdev should equal what's returned from bdget_disk()
1103 * if partno is 0; however, some drivers (floppy) use multiple
1104 * bdev's for the same physical device and @bdev may be one of the
1105 * aliases. Keep @bdev if partno is 0. This means claimer
1106 * tracking is broken for those devices but it has always been that
1107 * way.
1108 */
1109 if (partno)
1110 whole = bdget_disk(disk, 0);
1111 else
1112 whole = bdgrab(bdev);
1113
Nick Piggincf342572010-05-26 01:50:21 +10001114 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001115 put_disk(disk);
1116 if (!whole)
1117 return ERR_PTR(-ENOMEM);
1118
1119 /* prepare to claim, if successful, mark claiming in progress */
1120 spin_lock(&bdev_lock);
1121
1122 err = bd_prepare_to_claim(bdev, whole, holder);
1123 if (err == 0) {
1124 whole->bd_claiming = holder;
1125 spin_unlock(&bdev_lock);
1126 return whole;
1127 } else {
1128 spin_unlock(&bdev_lock);
1129 bdput(whole);
1130 return ERR_PTR(err);
1131 }
1132}
1133
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001134#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +01001135struct bd_holder_disk {
1136 struct list_head list;
1137 struct gendisk *disk;
1138 int refcnt;
1139};
1140
1141static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1142 struct gendisk *disk)
1143{
1144 struct bd_holder_disk *holder;
1145
1146 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1147 if (holder->disk == disk)
1148 return holder;
1149 return NULL;
1150}
1151
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -07001152static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001153{
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -07001154 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001155}
1156
1157static void del_symlink(struct kobject *from, struct kobject *to)
1158{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001159 sysfs_remove_link(from, kobject_name(to));
1160}
1161
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001162/**
Tejun Heoe09b4572010-11-13 11:55:17 +01001163 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1164 * @bdev: the claimed slave bdev
1165 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001166 *
Tejun Heo49731ba2011-01-14 18:43:57 +01001167 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1168 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001169 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001170 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001171 * - from "slaves" directory of the holder @disk to the claimed @bdev
1172 * - from "holders" directory of the @bdev to the holder @disk
1173 *
1174 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1175 * passed to bd_link_disk_holder(), then:
1176 *
1177 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1178 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1179 *
1180 * The caller must have claimed @bdev before calling this function and
1181 * ensure that both @bdev and @disk are valid during the creation and
1182 * lifetime of these symlinks.
1183 *
1184 * CONTEXT:
1185 * Might sleep.
1186 *
1187 * RETURNS:
1188 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001189 */
Tejun Heoe09b4572010-11-13 11:55:17 +01001190int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001191{
Tejun Heo49731ba2011-01-14 18:43:57 +01001192 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001193 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001194
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001195 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001196
Tejun Heo49731ba2011-01-14 18:43:57 +01001197 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -07001198
Tejun Heoe09b4572010-11-13 11:55:17 +01001199 /* FIXME: remove the following once add_disk() handles errors */
1200 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1201 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -07001202
Tejun Heo49731ba2011-01-14 18:43:57 +01001203 holder = bd_find_holder_disk(bdev, disk);
1204 if (holder) {
1205 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +01001206 goto out_unlock;
1207 }
1208
Tejun Heo49731ba2011-01-14 18:43:57 +01001209 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1210 if (!holder) {
1211 ret = -ENOMEM;
1212 goto out_unlock;
1213 }
1214
1215 INIT_LIST_HEAD(&holder->list);
1216 holder->disk = disk;
1217 holder->refcnt = 1;
1218
1219 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1220 if (ret)
1221 goto out_free;
1222
1223 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1224 if (ret)
1225 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +01001226 /*
1227 * bdev could be deleted beneath us which would implicitly destroy
1228 * the holder directory. Hold on to it.
1229 */
1230 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001231
1232 list_add(&holder->list, &bdev->bd_holder_disks);
1233 goto out_unlock;
1234
1235out_del:
1236 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1237out_free:
1238 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001239out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001240 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001241 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001242}
Tejun Heoe09b4572010-11-13 11:55:17 +01001243EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001244
Tejun Heo49731ba2011-01-14 18:43:57 +01001245/**
1246 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1247 * @bdev: the calimed slave bdev
1248 * @disk: the holding disk
1249 *
1250 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1251 *
1252 * CONTEXT:
1253 * Might sleep.
1254 */
1255void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001256{
Tejun Heo49731ba2011-01-14 18:43:57 +01001257 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001258
Tejun Heo49731ba2011-01-14 18:43:57 +01001259 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001260
Tejun Heo49731ba2011-01-14 18:43:57 +01001261 holder = bd_find_holder_disk(bdev, disk);
1262
1263 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1264 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1265 del_symlink(bdev->bd_part->holder_dir,
1266 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001267 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001268 list_del_init(&holder->list);
1269 kfree(holder);
1270 }
1271
1272 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001273}
Tejun Heo49731ba2011-01-14 18:43:57 +01001274EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001275#endif
1276
Andrew Patterson0c002c22008-09-04 14:27:20 -06001277/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001278 * flush_disk - invalidates all buffer-cache entries on a disk
1279 *
1280 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001281 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001282 *
1283 * Invalidates all buffer-cache entries on a disk. It should be called
1284 * when a disk has been changed -- either by a media change or online
1285 * resize.
1286 */
NeilBrown93b270f2011-02-24 17:25:47 +11001287static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001288{
NeilBrown93b270f2011-02-24 17:25:47 +11001289 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001290 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001291 "resized disk %s\n",
1292 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001293 }
1294
1295 if (!bdev->bd_disk)
1296 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001297 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001298 bdev->bd_invalidated = 1;
1299}
1300
1301/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001302 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001303 * @disk: struct gendisk to check
1304 * @bdev: struct bdev to adjust.
1305 *
1306 * This routine checks to see if the bdev size does not match the disk size
1307 * and adjusts it if it differs.
1308 */
1309void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1310{
1311 loff_t disk_size, bdev_size;
1312
1313 disk_size = (loff_t)get_capacity(disk) << 9;
1314 bdev_size = i_size_read(bdev->bd_inode);
1315 if (disk_size != bdev_size) {
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001316 printk(KERN_INFO
1317 "%s: detected capacity change from %lld to %lld\n",
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001318 disk->disk_name, bdev_size, disk_size);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001319 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001320 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001321 }
1322}
1323EXPORT_SYMBOL(check_disk_size_change);
1324
1325/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001326 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001327 * @disk: struct gendisk to be revalidated
1328 *
1329 * This routine is a wrapper for lower-level driver's revalidate_disk
1330 * call-backs. It is used to do common pre and post operations needed
1331 * for all revalidate_disk operations.
1332 */
1333int revalidate_disk(struct gendisk *disk)
1334{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001335 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001336 int ret = 0;
1337
1338 if (disk->fops->revalidate_disk)
1339 ret = disk->fops->revalidate_disk(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001340 bdev = bdget_disk(disk, 0);
1341 if (!bdev)
1342 return ret;
1343
1344 mutex_lock(&bdev->bd_mutex);
1345 check_disk_size_change(disk, bdev);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001346 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001347 mutex_unlock(&bdev->bd_mutex);
1348 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001349 return ret;
1350}
1351EXPORT_SYMBOL(revalidate_disk);
1352
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353/*
1354 * This routine checks whether a removable media has been changed,
1355 * and invalidates all buffer-cache-entries in that case. This
1356 * is a relatively slow routine, so we have to try to minimize using
1357 * it. Thus it is called only upon a 'mount' or 'open'. This
1358 * is the best way of combining speed and utility, I think.
1359 * People changing diskettes in the middle of an operation deserve
1360 * to lose :-)
1361 */
1362int check_disk_change(struct block_device *bdev)
1363{
1364 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001365 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001366 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Tejun Heo77ea8872010-12-08 20:57:37 +01001368 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1369 DISK_EVENT_EJECT_REQUEST);
1370 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 return 0;
1372
NeilBrown93b270f2011-02-24 17:25:47 +11001373 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (bdops->revalidate_disk)
1375 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 return 1;
1377}
1378
1379EXPORT_SYMBOL(check_disk_change);
1380
1381void bd_set_size(struct block_device *bdev, loff_t size)
1382{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001383 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
Al Viro59551022016-01-22 15:40:57 -05001385 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001386 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001387 inode_unlock(bdev->bd_inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001388 while (bsize < PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 if (size & bsize)
1390 break;
1391 bsize <<= 1;
1392 }
1393 bdev->bd_block_size = bsize;
1394 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1395}
1396EXPORT_SYMBOL(bd_set_size);
1397
Al Viro4385bab2013-05-05 22:11:03 -04001398static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001399
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001400/*
1401 * bd_mutex locking:
1402 *
1403 * mutex_lock(part->bd_mutex)
1404 * mutex_lock_nested(whole->bd_mutex, 1)
1405 */
1406
Al Viro572c4892007-10-08 13:24:05 -04001407static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001410 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001411 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001412 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001413 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Al Viro572c4892007-10-08 13:24:05 -04001415 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001416 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001417 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001418 perm |= MAY_WRITE;
1419 /*
1420 * hooks: /n/, see "layering violations".
1421 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001422 if (!for_part) {
1423 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1424 if (ret != 0) {
1425 bdput(bdev);
1426 return ret;
1427 }
Al Viro82666022008-08-01 05:32:04 -04001428 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001429
NeilBrownd3374822009-01-09 08:31:10 +11001430 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001431
Tejun Heo89f97492008-11-05 10:21:06 +01001432 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001433 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001434 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001435 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001436 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Tejun Heo69e02c52011-03-09 19:54:27 +01001438 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001439 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (!bdev->bd_openers) {
1441 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001442 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 bdev->bd_contains = bdev;
Christoph Hellwigc2ee0702017-08-23 19:10:31 +02001444 bdev->bd_partno = partno;
Dan Williams03cdadb2016-02-26 15:19:43 -08001445
Tejun Heocf771cb2008-09-03 09:01:09 +02001446 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001447 ret = -ENXIO;
1448 bdev->bd_part = disk_get_part(disk, partno);
1449 if (!bdev->bd_part)
1450 goto out_clear;
1451
Tejun Heo1196f8b2011-04-21 20:54:45 +02001452 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001454 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001455 if (ret == -ERESTARTSYS) {
1456 /* Lost a race with 'disk' being
1457 * deleted, try again.
1458 * See md.c
1459 */
1460 disk_put_part(bdev->bd_part);
1461 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001462 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001463 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001464 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001465 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001466 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001467 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001468 goto restart;
1469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 }
Tejun Heo7e697232011-05-23 13:26:07 +02001471
Christoph Hellwig22375702016-09-14 11:56:13 +02001472 if (!ret)
Tejun Heo7e697232011-05-23 13:26:07 +02001473 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Tejun Heo7e697232011-05-23 13:26:07 +02001474
Tejun Heo1196f8b2011-04-21 20:54:45 +02001475 /*
1476 * If the device is invalidated, rescan partition
1477 * if open succeeded or failed with -ENOMEDIUM.
1478 * The latter is necessary to prevent ghost
1479 * partitions on a removed medium.
1480 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001481 if (bdev->bd_invalidated) {
1482 if (!ret)
1483 rescan_partitions(disk, bdev);
1484 else if (ret == -ENOMEDIUM)
1485 invalidate_partitions(disk, bdev);
1486 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001487
Tejun Heo1196f8b2011-04-21 20:54:45 +02001488 if (ret)
1489 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 struct block_device *whole;
1492 whole = bdget_disk(disk, 0);
1493 ret = -ENOMEM;
1494 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001495 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001496 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001497 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001499 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001501 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001502 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001503 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001505 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 }
Tejun Heo89f97492008-11-05 10:21:06 +01001507 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
Jan Kara03e26272017-03-23 01:36:53 +01001509
1510 if (bdev->bd_bdi == &noop_backing_dev_info)
1511 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001514 ret = 0;
1515 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001516 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001517 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001518 if (bdev->bd_invalidated) {
1519 if (!ret)
1520 rescan_partitions(bdev->bd_disk, bdev);
1521 else if (ret == -ENOMEDIUM)
1522 invalidate_partitions(bdev->bd_disk, bdev);
1523 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001524 if (ret)
1525 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001527 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001528 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001529 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 }
1531 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001532 if (for_part)
1533 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001534 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001535 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 return 0;
1537
Tejun Heo0762b8b2008-08-25 19:56:12 +09001538 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001539 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001541 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001542 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001544 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001546 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001547 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001548 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001549 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001550 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001551 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001552 bdput(bdev);
1553
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 return ret;
1555}
1556
Tejun Heod4d77622010-11-13 11:55:18 +01001557/**
1558 * blkdev_get - open a block device
1559 * @bdev: block_device to open
1560 * @mode: FMODE_* mask
1561 * @holder: exclusive holder identifier
1562 *
1563 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1564 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1565 * @holder is invalid. Exclusive opens may nest for the same @holder.
1566 *
1567 * On success, the reference count of @bdev is unchanged. On failure,
1568 * @bdev is put.
1569 *
1570 * CONTEXT:
1571 * Might sleep.
1572 *
1573 * RETURNS:
1574 * 0 on success, -errno on failure.
1575 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001576int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577{
Tejun Heoe525fd82010-11-13 11:55:17 +01001578 struct block_device *whole = NULL;
1579 int res;
1580
1581 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1582
1583 if ((mode & FMODE_EXCL) && holder) {
1584 whole = bd_start_claiming(bdev, holder);
1585 if (IS_ERR(whole)) {
1586 bdput(bdev);
1587 return PTR_ERR(whole);
1588 }
1589 }
1590
1591 res = __blkdev_get(bdev, mode, 0);
1592
1593 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001594 struct gendisk *disk = whole->bd_disk;
1595
Tejun Heo6a027ef2010-11-13 11:55:17 +01001596 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001597 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001598 spin_lock(&bdev_lock);
1599
Tejun Heo77ea8872010-12-08 20:57:37 +01001600 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001601 BUG_ON(!bd_may_claim(bdev, whole, holder));
1602 /*
1603 * Note that for a whole device bd_holders
1604 * will be incremented twice, and bd_holder
1605 * will be set to bd_may_claim before being
1606 * set to holder
1607 */
1608 whole->bd_holders++;
1609 whole->bd_holder = bd_may_claim;
1610 bdev->bd_holders++;
1611 bdev->bd_holder = holder;
1612 }
1613
1614 /* tell others that we're done */
1615 BUG_ON(whole->bd_claiming != holder);
1616 whole->bd_claiming = NULL;
1617 wake_up_bit(&whole->bd_claiming, 0);
1618
1619 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001620
1621 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001622 * Block event polling for write claims if requested. Any
1623 * write holder makes the write_holder state stick until
1624 * all are released. This is good enough and tracking
1625 * individual writeable reference is too fragile given the
1626 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001627 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001628 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1629 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001630 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001631 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001632 }
1633
1634 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001635 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001636 }
1637
1638 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001639}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640EXPORT_SYMBOL(blkdev_get);
1641
Tejun Heod4d77622010-11-13 11:55:18 +01001642/**
1643 * blkdev_get_by_path - open a block device by name
1644 * @path: path to the block device to open
1645 * @mode: FMODE_* mask
1646 * @holder: exclusive holder identifier
1647 *
1648 * Open the blockdevice described by the device file at @path. @mode
1649 * and @holder are identical to blkdev_get().
1650 *
1651 * On success, the returned block_device has reference count of one.
1652 *
1653 * CONTEXT:
1654 * Might sleep.
1655 *
1656 * RETURNS:
1657 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1658 */
1659struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1660 void *holder)
1661{
1662 struct block_device *bdev;
1663 int err;
1664
1665 bdev = lookup_bdev(path);
1666 if (IS_ERR(bdev))
1667 return bdev;
1668
1669 err = blkdev_get(bdev, mode, holder);
1670 if (err)
1671 return ERR_PTR(err);
1672
Chuck Ebberte51900f2011-02-16 18:11:53 -05001673 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1674 blkdev_put(bdev, mode);
1675 return ERR_PTR(-EACCES);
1676 }
1677
Tejun Heod4d77622010-11-13 11:55:18 +01001678 return bdev;
1679}
1680EXPORT_SYMBOL(blkdev_get_by_path);
1681
1682/**
1683 * blkdev_get_by_dev - open a block device by device number
1684 * @dev: device number of block device to open
1685 * @mode: FMODE_* mask
1686 * @holder: exclusive holder identifier
1687 *
1688 * Open the blockdevice described by device number @dev. @mode and
1689 * @holder are identical to blkdev_get().
1690 *
1691 * Use it ONLY if you really do not have anything better - i.e. when
1692 * you are behind a truly sucky interface and all you are given is a
1693 * device number. _Never_ to be used for internal purposes. If you
1694 * ever need it - reconsider your API.
1695 *
1696 * On success, the returned block_device has reference count of one.
1697 *
1698 * CONTEXT:
1699 * Might sleep.
1700 *
1701 * RETURNS:
1702 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1703 */
1704struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1705{
1706 struct block_device *bdev;
1707 int err;
1708
1709 bdev = bdget(dev);
1710 if (!bdev)
1711 return ERR_PTR(-ENOMEM);
1712
1713 err = blkdev_get(bdev, mode, holder);
1714 if (err)
1715 return ERR_PTR(err);
1716
1717 return bdev;
1718}
1719EXPORT_SYMBOL(blkdev_get_by_dev);
1720
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721static int blkdev_open(struct inode * inode, struct file * filp)
1722{
1723 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725 /*
1726 * Preserve backwards compatibility and allow large file access
1727 * even if userspace doesn't ask for it explicitly. Some mkfs
1728 * binary needs it. We might want to drop this workaround
1729 * during an unstable branch.
1730 */
1731 filp->f_flags |= O_LARGEFILE;
1732
Christoph Hellwigc35fc7a2017-08-29 16:13:21 +02001733 filp->f_mode |= FMODE_NOWAIT;
1734
Al Viro572c4892007-10-08 13:24:05 -04001735 if (filp->f_flags & O_NDELAY)
1736 filp->f_mode |= FMODE_NDELAY;
1737 if (filp->f_flags & O_EXCL)
1738 filp->f_mode |= FMODE_EXCL;
1739 if ((filp->f_flags & O_ACCMODE) == 3)
1740 filp->f_mode |= FMODE_WRITE_IOCTL;
1741
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001743 if (bdev == NULL)
1744 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
Al Viro572c4892007-10-08 13:24:05 -04001746 filp->f_mapping = bdev->bd_inode->i_mapping;
Jeff Layton5660e132017-07-06 07:02:25 -04001747 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
Al Viro572c4892007-10-08 13:24:05 -04001748
Tejun Heoe525fd82010-11-13 11:55:17 +01001749 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750}
1751
Al Viro4385bab2013-05-05 22:11:03 -04001752static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001753{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001754 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001755 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001756
NeilBrown6796bf52006-12-08 02:36:16 -08001757 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001758 if (for_part)
1759 bdev->bd_part_count--;
1760
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001761 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001762 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001763 sync_blockdev(bdev);
1764 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001765
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001766 bdev_write_inode(bdev);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001767 }
1768 if (bdev->bd_contains == bdev) {
1769 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001770 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001771 }
1772 if (!bdev->bd_openers) {
1773 struct module *owner = disk->fops->owner;
1774
Tejun Heo0762b8b2008-08-25 19:56:12 +09001775 disk_put_part(bdev->bd_part);
1776 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001777 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001778 if (bdev != bdev->bd_contains)
1779 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001780 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001781
1782 put_disk(disk);
1783 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001784 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001785 mutex_unlock(&bdev->bd_mutex);
1786 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001787 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001788 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001789}
1790
Al Viro4385bab2013-05-05 22:11:03 -04001791void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001792{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001793 mutex_lock(&bdev->bd_mutex);
1794
Tejun Heoe525fd82010-11-13 11:55:17 +01001795 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001796 bool bdev_free;
1797
1798 /*
1799 * Release a claim on the device. The holder fields
1800 * are protected with bdev_lock. bd_mutex is to
1801 * synchronize disk_holder unlinking.
1802 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001803 spin_lock(&bdev_lock);
1804
1805 WARN_ON_ONCE(--bdev->bd_holders < 0);
1806 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1807
1808 /* bd_contains might point to self, check in a separate step */
1809 if ((bdev_free = !bdev->bd_holders))
1810 bdev->bd_holder = NULL;
1811 if (!bdev->bd_contains->bd_holders)
1812 bdev->bd_contains->bd_holder = NULL;
1813
1814 spin_unlock(&bdev_lock);
1815
Tejun Heo77ea8872010-12-08 20:57:37 +01001816 /*
1817 * If this was the last claim, remove holder link and
1818 * unblock evpoll if it was a write holder.
1819 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001820 if (bdev_free && bdev->bd_write_holder) {
1821 disk_unblock_events(bdev->bd_disk);
1822 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001823 }
Tejun Heo69362172011-03-09 19:54:27 +01001824 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001825
Tejun Heo85ef06d2011-07-01 16:17:47 +02001826 /*
1827 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1828 * event. This is to ensure detection of media removal commanded
1829 * from userland - e.g. eject(1).
1830 */
1831 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1832
1833 mutex_unlock(&bdev->bd_mutex);
1834
Al Viro4385bab2013-05-05 22:11:03 -04001835 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001836}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001837EXPORT_SYMBOL(blkdev_put);
1838
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839static int blkdev_close(struct inode * inode, struct file * filp)
1840{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001841 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001842 blkdev_put(bdev, filp->f_mode);
1843 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844}
1845
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001846static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001848 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001849 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001850
1851 /*
1852 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1853 * to updated it before every ioctl.
1854 */
Al Viro56b26ad2008-09-19 03:17:36 -04001855 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001856 mode |= FMODE_NDELAY;
1857 else
1858 mode &= ~FMODE_NDELAY;
1859
Al Viro56b26ad2008-09-19 03:17:36 -04001860 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861}
1862
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001863/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001864 * Write data to the block device. Only intended for the block device itself
1865 * and the raw driver which basically is a fake block device.
1866 *
1867 * Does not take i_mutex for the write and thus is not for general purpose
1868 * use.
1869 */
Al Viro1456c0a2014-04-03 03:21:50 -04001870ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001871{
1872 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001873 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001874 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001875 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001876 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001877
Al Viro7ec7b942015-04-07 11:35:14 -04001878 if (bdev_read_only(I_BDEV(bd_inode)))
1879 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001880
Al Viro7ec7b942015-04-07 11:35:14 -04001881 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001882 return 0;
1883
Al Viro7ec7b942015-04-07 11:35:14 -04001884 if (iocb->ki_pos >= size)
1885 return -ENOSPC;
1886
Christoph Hellwigc35fc7a2017-08-29 16:13:21 +02001887 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1888 return -EOPNOTSUPP;
1889
Al Viro7ec7b942015-04-07 11:35:14 -04001890 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001891
Jianpeng Ma53362a02012-08-02 09:50:39 +02001892 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001893 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwige2592212016-04-07 08:52:01 -07001894 if (ret > 0)
1895 ret = generic_write_sync(iocb, ret);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001896 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001897 return ret;
1898}
Al Viro1456c0a2014-04-03 03:21:50 -04001899EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001900
David Jefferyb2de5252014-09-29 10:21:10 -04001901ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001902{
1903 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001904 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001905 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001906 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001907
1908 if (pos >= size)
1909 return 0;
1910
1911 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001912 iov_iter_truncate(to, size);
1913 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001914}
David Jefferyb2de5252014-09-29 10:21:10 -04001915EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001916
Christoph Hellwigeef99382009-08-20 17:43:41 +02001917/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001918 * Try to release a page associated with block device when the system
1919 * is under memory pressure.
1920 */
1921static int blkdev_releasepage(struct page *page, gfp_t wait)
1922{
1923 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1924
1925 if (super && super->s_op->bdev_try_to_free_page)
1926 return super->s_op->bdev_try_to_free_page(super, page, wait);
1927
1928 return try_to_free_buffers(page);
1929}
1930
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001931static int blkdev_writepages(struct address_space *mapping,
1932 struct writeback_control *wbc)
1933{
1934 if (dax_mapping(mapping)) {
1935 struct block_device *bdev = I_BDEV(mapping->host);
1936
1937 return dax_writeback_mapping_range(mapping, bdev, wbc);
1938 }
1939 return generic_writepages(mapping, wbc);
1940}
1941
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001942static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001944 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001946 .write_begin = blkdev_write_begin,
1947 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001948 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001949 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 .direct_IO = blkdev_direct_IO,
Mel Gormanb4597222013-07-03 15:02:05 -07001951 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952};
1953
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001954#define BLKDEV_FALLOC_FL_SUPPORTED \
1955 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1956 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1957
1958static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1959 loff_t len)
1960{
1961 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001962 struct address_space *mapping;
1963 loff_t end = start + len - 1;
1964 loff_t isize;
1965 int error;
1966
1967 /* Fail if we don't recognize the flags. */
1968 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1969 return -EOPNOTSUPP;
1970
1971 /* Don't go off the end of the device. */
1972 isize = i_size_read(bdev->bd_inode);
1973 if (start >= isize)
1974 return -EINVAL;
1975 if (end >= isize) {
1976 if (mode & FALLOC_FL_KEEP_SIZE) {
1977 len = isize - start;
1978 end = start + len - 1;
1979 } else
1980 return -EINVAL;
1981 }
1982
1983 /*
1984 * Don't allow IO that isn't aligned to logical block size.
1985 */
1986 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1987 return -EINVAL;
1988
1989 /* Invalidate the page cache, including dirty pages. */
1990 mapping = bdev->bd_inode->i_mapping;
1991 truncate_inode_pages_range(mapping, start, end);
1992
1993 switch (mode) {
1994 case FALLOC_FL_ZERO_RANGE:
1995 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1996 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
Christoph Hellwigee472d82017-04-05 19:21:08 +02001997 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001998 break;
1999 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
Christoph Hellwig34045122017-04-05 19:21:11 +02002000 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2001 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002002 break;
2003 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002004 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2005 GFP_KERNEL, 0);
2006 break;
2007 default:
2008 return -EOPNOTSUPP;
2009 }
2010 if (error)
2011 return error;
2012
2013 /*
2014 * Invalidate again; if someone wandered in and dirtied a page,
2015 * the caller will be given -EBUSY. The third argument is
2016 * inclusive, so the rounding here is safe.
2017 */
2018 return invalidate_inode_pages2_range(mapping,
2019 start >> PAGE_SHIFT,
2020 end >> PAGE_SHIFT);
2021}
2022
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002023const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 .open = blkdev_open,
2025 .release = blkdev_close,
2026 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04002027 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04002028 .write_iter = blkdev_write_iter,
Dan Williamsacc93d32016-05-07 11:40:28 -07002029 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07002030 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07002031 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032#ifdef CONFIG_COMPAT
2033 .compat_ioctl = compat_blkdev_ioctl,
2034#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08002035 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04002036 .splice_write = iter_file_splice_write,
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002037 .fallocate = blkdev_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038};
2039
2040int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2041{
2042 int res;
2043 mm_segment_t old_fs = get_fs();
2044 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04002045 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 set_fs(old_fs);
2047 return res;
2048}
2049
2050EXPORT_SYMBOL(ioctl_by_bdev);
2051
2052/**
2053 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08002054 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02002056 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 * namespace if possible and return it. Return ERR_PTR(error)
2058 * otherwise.
2059 */
Al Viro421748e2008-08-02 01:04:36 -04002060struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061{
2062 struct block_device *bdev;
2063 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04002064 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 int error;
2066
Al Viro421748e2008-08-02 01:04:36 -04002067 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 return ERR_PTR(-EINVAL);
2069
Al Viro421748e2008-08-02 01:04:36 -04002070 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 if (error)
2072 return ERR_PTR(error);
2073
David Howellsbb6687342015-03-17 22:26:21 +00002074 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 error = -ENOTBLK;
2076 if (!S_ISBLK(inode->i_mode))
2077 goto fail;
2078 error = -EACCES;
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002079 if (!may_open_dev(&path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 goto fail;
2081 error = -ENOMEM;
2082 bdev = bd_acquire(inode);
2083 if (!bdev)
2084 goto fail;
2085out:
Al Viro421748e2008-08-02 01:04:36 -04002086 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 return bdev;
2088fail:
2089 bdev = ERR_PTR(error);
2090 goto out;
2091}
Al Virod5686b42008-08-01 05:00:11 -04002092EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
NeilBrown93b270f2011-02-24 17:25:47 +11002094int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01002095{
2096 struct super_block *sb = get_super(bdev);
2097 int res = 0;
2098
2099 if (sb) {
2100 /*
2101 * no need to lock the super, get_super holds the
2102 * read mutex so the filesystem cannot go away
2103 * under us (->put_super runs with the write lock
2104 * hold).
2105 */
2106 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11002107 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01002108 drop_super(sb);
2109 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07002110 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01002111 return res;
2112}
2113EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002114
2115void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2116{
2117 struct inode *inode, *old_inode = NULL;
2118
Dave Chinner74278da2015-03-04 12:37:22 -05002119 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002120 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2121 struct address_space *mapping = inode->i_mapping;
Rabin Vincentaf309222016-12-01 09:18:28 +01002122 struct block_device *bdev;
Jan Kara5c0d6b62012-07-03 16:45:31 +02002123
2124 spin_lock(&inode->i_lock);
2125 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2126 mapping->nrpages == 0) {
2127 spin_unlock(&inode->i_lock);
2128 continue;
2129 }
2130 __iget(inode);
2131 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05002132 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002133 /*
2134 * We hold a reference to 'inode' so it couldn't have been
2135 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05002136 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02002137 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05002138 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02002139 * later.
2140 */
2141 iput(old_inode);
2142 old_inode = inode;
Rabin Vincentaf309222016-12-01 09:18:28 +01002143 bdev = I_BDEV(inode);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002144
Rabin Vincentaf309222016-12-01 09:18:28 +01002145 mutex_lock(&bdev->bd_mutex);
2146 if (bdev->bd_openers)
2147 func(bdev, arg);
2148 mutex_unlock(&bdev->bd_mutex);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002149
Dave Chinner74278da2015-03-04 12:37:22 -05002150 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002151 }
Dave Chinner74278da2015-03-04 12:37:22 -05002152 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002153 iput(old_inode);
2154}