blob: c546cdce77e6df48ab48df9b84f3cb6c20c4d173 [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 Howells07f3f052006-09-30 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 Bunk4c54ac622008-02-18 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);
Jens Axboe06193172018-11-13 21:16:54 -0700184 blk_wake_io_task(waiter);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600185}
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 {
Kees Cook6da2ec52018-06-12 13:55:00 -0700208 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
209 GFP_KERNEL);
Jens Axboe72ecad22016-11-16 23:11:42 -0700210 if (!vecs)
211 return -ENOMEM;
212 }
213
Ming Lei3a83f462016-11-22 08:57:21 -0700214 bio_init(&bio, vecs, nr_pages);
Christoph Hellwig74d46992017-08-23 19:10:32 +0200215 bio_set_dev(&bio, bdev);
Damien Le Moal4d1a4762016-11-22 15:38:49 +0900216 bio.bi_iter.bi_sector = pos >> 9;
Jens Axboe45d06cf2017-06-27 11:01:22 -0600217 bio.bi_write_hint = iocb->ki_hint;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600218 bio.bi_private = current;
219 bio.bi_end_io = blkdev_bio_end_io_simple;
Adam Manzanares074111c2018-05-22 10:52:20 -0700220 bio.bi_ioprio = iocb->ki_ioprio;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600221
222 ret = bio_iov_iter_get_pages(&bio, iter);
223 if (unlikely(ret))
Martin Wilck9362dd12018-07-25 23:15:08 +0200224 goto out;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600225 ret = bio.bi_iter.bi_size;
226
227 if (iov_iter_rw(iter) == READ) {
Jens Axboe78250c02016-11-17 17:50:47 +0100228 bio.bi_opf = REQ_OP_READ;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600229 if (iter_is_iovec(iter))
230 should_dirty = true;
231 } else {
Jens Axboe78250c02016-11-17 17:50:47 +0100232 bio.bi_opf = dio_bio_write_op(iocb);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600233 task_io_account_write(ret);
234 }
Jens Axboed1e36282018-08-29 10:36:56 -0600235 if (iocb->ki_flags & IOCB_HIPRI)
236 bio.bi_opf |= REQ_HIPRI;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600237
238 qc = submit_bio(&bio);
239 for (;;) {
Linus Torvalds1ac5cd42019-01-02 10:46:03 -0800240 set_current_state(TASK_UNINTERRUPTIBLE);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600241 if (!READ_ONCE(bio.bi_private))
242 break;
243 if (!(iocb->ki_flags & IOCB_HIPRI) ||
Jens Axboe0a1b8b82018-11-26 08:24:43 -0700244 !blk_poll(bdev_get_queue(bdev), qc, true))
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600245 io_schedule();
246 }
247 __set_current_state(TASK_RUNNING);
248
249 bio_for_each_segment_all(bvec, &bio, i) {
250 if (should_dirty && !PageCompound(bvec->bv_page))
251 set_page_dirty_lock(bvec->bv_page);
252 put_page(bvec->bv_page);
253 }
254
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200255 if (unlikely(bio.bi_status))
Linus Torvaldsc6b1e362017-07-03 10:34:51 -0700256 ret = blk_status_to_errno(bio.bi_status);
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600257
Martin Wilck9362dd12018-07-25 23:15:08 +0200258out:
259 if (vecs != inline_vecs)
260 kfree(vecs);
261
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600262 bio_uninit(&bio);
263
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600264 return ret;
265}
266
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700267struct blkdev_dio {
268 union {
269 struct kiocb *iocb;
270 struct task_struct *waiter;
271 };
272 size_t size;
273 atomic_t ref;
274 bool multi_bio : 1;
275 bool should_dirty : 1;
276 bool is_sync : 1;
277 struct bio bio;
278};
279
Kent Overstreet52190f82018-05-20 18:25:55 -0400280static struct bio_set blkdev_dio_pool;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700281
282static void blkdev_bio_end_io(struct bio *bio)
283{
284 struct blkdev_dio *dio = bio->bi_private;
285 bool should_dirty = dio->should_dirty;
286
287 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200288 if (bio->bi_status && !dio->bio.bi_status)
289 dio->bio.bi_status = bio->bi_status;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700290 } else {
291 if (!dio->is_sync) {
292 struct kiocb *iocb = dio->iocb;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200293 ssize_t ret;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700294
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200295 if (likely(!dio->bio.bi_status)) {
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700296 ret = dio->size;
297 iocb->ki_pos += ret;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200298 } else {
299 ret = blk_status_to_errno(dio->bio.bi_status);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700300 }
301
302 dio->iocb->ki_complete(iocb, ret, 0);
Christoph Hellwig531724a2018-11-30 09:23:48 +0100303 if (dio->multi_bio)
304 bio_put(&dio->bio);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700305 } else {
306 struct task_struct *waiter = dio->waiter;
307
308 WRITE_ONCE(dio->waiter, NULL);
Jens Axboe06193172018-11-13 21:16:54 -0700309 blk_wake_io_task(waiter);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700310 }
311 }
312
313 if (should_dirty) {
314 bio_check_pages_dirty(bio);
315 } else {
316 struct bio_vec *bvec;
317 int i;
318
319 bio_for_each_segment_all(bvec, bio, i)
320 put_page(bvec->bv_page);
321 bio_put(bio);
322 }
323}
324
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800325static ssize_t
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700326__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800327{
328 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +0900329 struct inode *inode = bdev_file_inode(file);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700330 struct block_device *bdev = I_BDEV(inode);
Christoph Hellwig64d656a2016-12-22 19:20:45 +0100331 struct blk_plug plug;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700332 struct blkdev_dio *dio;
333 struct bio *bio;
Jens Axboecb700eb2018-11-15 19:56:53 -0700334 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
Christoph Hellwig690e5322017-01-24 14:50:19 +0100335 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700336 loff_t pos = iocb->ki_pos;
337 blk_qc_t qc = BLK_QC_T_NONE;
Christoph Hellwig36ffc6c2017-06-03 09:38:00 +0200338 int ret = 0;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700339
Jens Axboe9a794fb2016-11-22 08:12:39 -0700340 if ((pos | iov_iter_alignment(iter)) &
341 (bdev_logical_block_size(bdev) - 1))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700342 return -EINVAL;
343
Kent Overstreet52190f82018-05-20 18:25:55 -0400344 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700345
346 dio = container_of(bio, struct blkdev_dio, bio);
Christoph Hellwig690e5322017-01-24 14:50:19 +0100347 dio->is_sync = is_sync = is_sync_kiocb(iocb);
Christoph Hellwig531724a2018-11-30 09:23:48 +0100348 if (dio->is_sync) {
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700349 dio->waiter = current;
Christoph Hellwig531724a2018-11-30 09:23:48 +0100350 bio_get(bio);
351 } else {
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700352 dio->iocb = iocb;
Christoph Hellwig531724a2018-11-30 09:23:48 +0100353 }
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700354
355 dio->size = 0;
356 dio->multi_bio = false;
David Howells00e23702018-10-22 13:07:28 +0100357 dio->should_dirty = is_read && iter_is_iovec(iter);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700358
Jens Axboecb700eb2018-11-15 19:56:53 -0700359 /*
360 * Don't plug for HIPRI/polled IO, as those should go straight
361 * to issue
362 */
363 if (!is_poll)
364 blk_start_plug(&plug);
365
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700366 for (;;) {
Christoph Hellwig74d46992017-08-23 19:10:32 +0200367 bio_set_dev(bio, bdev);
Damien Le Moal4d1a4762016-11-22 15:38:49 +0900368 bio->bi_iter.bi_sector = pos >> 9;
Jens Axboe45d06cf2017-06-27 11:01:22 -0600369 bio->bi_write_hint = iocb->ki_hint;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700370 bio->bi_private = dio;
371 bio->bi_end_io = blkdev_bio_end_io;
Adam Manzanares074111c2018-05-22 10:52:20 -0700372 bio->bi_ioprio = iocb->ki_ioprio;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700373
374 ret = bio_iov_iter_get_pages(bio, iter);
375 if (unlikely(ret)) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200376 bio->bi_status = BLK_STS_IOERR;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700377 bio_endio(bio);
378 break;
379 }
380
381 if (is_read) {
382 bio->bi_opf = REQ_OP_READ;
383 if (dio->should_dirty)
384 bio_set_pages_dirty(bio);
385 } else {
386 bio->bi_opf = dio_bio_write_op(iocb);
387 task_io_account_write(bio->bi_iter.bi_size);
388 }
389
390 dio->size += bio->bi_iter.bi_size;
391 pos += bio->bi_iter.bi_size;
392
393 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
394 if (!nr_pages) {
Jens Axboed34513d2018-11-06 14:29:11 -0700395 if (iocb->ki_flags & IOCB_HIPRI)
396 bio->bi_opf |= REQ_HIPRI;
397
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700398 qc = submit_bio(bio);
399 break;
400 }
401
402 if (!dio->multi_bio) {
Christoph Hellwig531724a2018-11-30 09:23:48 +0100403 /*
404 * AIO needs an extra reference to ensure the dio
405 * structure which is embedded into the first bio
406 * stays around.
407 */
408 if (!is_sync)
409 bio_get(bio);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700410 dio->multi_bio = true;
411 atomic_set(&dio->ref, 2);
412 } else {
413 atomic_inc(&dio->ref);
414 }
415
416 submit_bio(bio);
417 bio = bio_alloc(GFP_KERNEL, nr_pages);
418 }
Jens Axboecb700eb2018-11-15 19:56:53 -0700419
420 if (!is_poll)
421 blk_finish_plug(&plug);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700422
Christoph Hellwig690e5322017-01-24 14:50:19 +0100423 if (!is_sync)
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700424 return -EIOCBQUEUED;
425
426 for (;;) {
Linus Torvalds1ac5cd42019-01-02 10:46:03 -0800427 set_current_state(TASK_UNINTERRUPTIBLE);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700428 if (!READ_ONCE(dio->waiter))
429 break;
430
431 if (!(iocb->ki_flags & IOCB_HIPRI) ||
Jens Axboe0a1b8b82018-11-26 08:24:43 -0700432 !blk_poll(bdev_get_queue(bdev), qc, true))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700433 io_schedule();
434 }
435 __set_current_state(TASK_RUNNING);
436
Christoph Hellwig36ffc6c2017-06-03 09:38:00 +0200437 if (!ret)
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200438 ret = blk_status_to_errno(dio->bio.bi_status);
Shaohua Li7a62a522016-12-13 19:07:55 -0800439 if (likely(!ret))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700440 ret = dio->size;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700441
442 bio_put(&dio->bio);
443 return ret;
444}
445
446static ssize_t
447blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
448{
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600449 int nr_pages;
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800450
Jens Axboe72ecad22016-11-16 23:11:42 -0700451 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600452 if (!nr_pages)
453 return 0;
Jens Axboe72ecad22016-11-16 23:11:42 -0700454 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600455 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700456
457 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800458}
459
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700460static __init int blkdev_init(void)
461{
Kent Overstreet52190f82018-05-20 18:25:55 -0400462 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700463}
464module_init(blkdev_init);
465
Jan Kara5cee5812009-04-27 16:43:51 +0200466int __sync_blockdev(struct block_device *bdev, int wait)
467{
468 if (!bdev)
469 return 0;
470 if (!wait)
471 return filemap_flush(bdev->bd_inode->i_mapping);
472 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
473}
474
Nick Piggin585d3bc2009-02-25 10:44:19 +0100475/*
476 * Write out and wait upon all the dirty data associated with a block
477 * device via its mapping. Does not take the superblock lock.
478 */
479int sync_blockdev(struct block_device *bdev)
480{
Jan Kara5cee5812009-04-27 16:43:51 +0200481 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100482}
483EXPORT_SYMBOL(sync_blockdev);
484
485/*
486 * Write out and wait upon all dirty data associated with this
487 * device. Filesystem data as well as the underlying block
488 * device. Takes the superblock lock.
489 */
490int fsync_bdev(struct block_device *bdev)
491{
492 struct super_block *sb = get_super(bdev);
493 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200494 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100495 drop_super(sb);
496 return res;
497 }
498 return sync_blockdev(bdev);
499}
Al Viro47e44912009-04-01 07:07:16 -0400500EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100501
502/**
503 * freeze_bdev -- lock a filesystem and force it into a consistent state
504 * @bdev: blockdevice to lock
505 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100506 * If a superblock is found on this device, we take the s_umount semaphore
507 * on it to make sure nobody unmounts until the snapshot creation is done.
508 * The reference counter (bd_fsfreeze_count) guarantees that only the last
509 * unfreeze process can unfreeze the frozen filesystem actually when multiple
510 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
511 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
512 * actually.
513 */
514struct super_block *freeze_bdev(struct block_device *bdev)
515{
516 struct super_block *sb;
517 int error = 0;
518
519 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200520 if (++bdev->bd_fsfreeze_count > 1) {
521 /*
522 * We don't even need to grab a reference - the first call
523 * to freeze_bdev grab an active reference and only the last
524 * thaw_bdev drops it.
525 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100526 sb = get_super(bdev);
Andrey Ryabinin5bb53c02016-08-23 18:55:31 +0300527 if (sb)
528 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100529 mutex_unlock(&bdev->bd_fsfreeze_mutex);
530 return sb;
531 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100532
Christoph Hellwig45042302009-08-03 23:28:35 +0200533 sb = get_active_super(bdev);
534 if (!sb)
535 goto out;
Benjamin Marzinski48b6bca62014-11-13 20:42:03 -0600536 if (sb->s_op->freeze_super)
537 error = sb->s_op->freeze_super(sb);
538 else
539 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400540 if (error) {
541 deactivate_super(sb);
542 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200543 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400544 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100545 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400546 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200547 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100548 sync_blockdev(bdev);
549 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200550 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100551}
552EXPORT_SYMBOL(freeze_bdev);
553
554/**
555 * thaw_bdev -- unlock filesystem
556 * @bdev: blockdevice to unlock
557 * @sb: associated superblock
558 *
559 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
560 */
561int thaw_bdev(struct block_device *bdev, struct super_block *sb)
562{
Christoph Hellwig45042302009-08-03 23:28:35 +0200563 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100564
565 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200566 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400567 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100568
Christoph Hellwig45042302009-08-03 23:28:35 +0200569 error = 0;
570 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400571 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100572
Christoph Hellwig45042302009-08-03 23:28:35 +0200573 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400574 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200575
Benjamin Marzinski48b6bca62014-11-13 20:42:03 -0600576 if (sb->s_op->thaw_super)
577 error = sb->s_op->thaw_super(sb);
578 else
579 error = thaw_super(sb);
Pierre Morel997198b2016-10-04 10:53:40 +0200580 if (error)
Josef Bacik18e9e512010-03-23 10:34:56 -0400581 bdev->bd_fsfreeze_count++;
Josef Bacik18e9e512010-03-23 10:34:56 -0400582out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100583 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Pierre Morel997198b2016-10-04 10:53:40 +0200584 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100585}
586EXPORT_SYMBOL(thaw_bdev);
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
589{
590 return block_write_full_page(page, blkdev_get_block, wbc);
591}
592
593static int blkdev_readpage(struct file * file, struct page * page)
594{
595 return block_read_full_page(page, blkdev_get_block);
596}
597
Akinobu Mita447f05b2014-10-09 15:26:58 -0700598static int blkdev_readpages(struct file *file, struct address_space *mapping,
599 struct list_head *pages, unsigned nr_pages)
600{
601 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
602}
603
Nick Piggin6272b5a2007-10-16 01:25:04 -0700604static int blkdev_write_begin(struct file *file, struct address_space *mapping,
605 loff_t pos, unsigned len, unsigned flags,
606 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200608 return block_write_begin(mapping, pos, len, flags, pagep,
609 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}
611
Nick Piggin6272b5a2007-10-16 01:25:04 -0700612static int blkdev_write_end(struct file *file, struct address_space *mapping,
613 loff_t pos, unsigned len, unsigned copied,
614 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700616 int ret;
617 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
618
619 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300620 put_page(page);
Nick Piggin6272b5a2007-10-16 01:25:04 -0700621
622 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623}
624
625/*
626 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500627 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 * so we compute the size by hand (just as in block_read/write above)
629 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800630static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900632 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 loff_t retval;
634
Al Viro59551022016-01-22 15:40:57 -0500635 inode_lock(bd_inode);
Al Viro5d48f3a2013-06-23 21:34:45 +0400636 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Al Viro59551022016-01-22 15:40:57 -0500637 inode_unlock(bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 return retval;
639}
640
Josef Bacik02c24a82011-07-16 20:44:56 -0400641int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900643 struct inode *bd_inode = bdev_file_inode(filp);
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400644 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100645 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200646
Jeff Layton372cf242017-07-06 07:02:28 -0400647 error = file_write_and_wait_range(filp, start, end);
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200648 if (error)
649 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100650
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400651 /*
652 * There is no need to serialise calls to blkdev_issue_flush with
653 * i_mutex and doing so causes performance issues with concurrent
654 * O_SYNC writers to a block device.
655 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200656 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100657 if (error == -EOPNOTSUPP)
658 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400659
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100660 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700662EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700664/**
665 * bdev_read_page() - Start reading a page from a block device
666 * @bdev: The device to read the page from
667 * @sector: The offset on the device to read the page to (need not be aligned)
668 * @page: The page to read
669 *
670 * On entry, the page should be locked. It will be unlocked when the page
671 * has been read. If the block driver implements rw_page synchronously,
672 * that will be true on exit from this function, but it need not be.
673 *
674 * Errors returned by this function are usually "soft", eg out of memory, or
675 * queue full; callers should try a different route to read this page rather
676 * than propagate an error back up the stack.
677 *
678 * Return: negative errno if an error occurs, 0 if submission was successful.
679 */
680int bdev_read_page(struct block_device *bdev, sector_t sector,
681 struct page *page)
682{
683 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc952015-11-19 13:29:28 -0800684 int result = -EOPNOTSUPP;
685
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400686 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc952015-11-19 13:29:28 -0800687 return result;
688
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800689 result = blk_queue_enter(bdev->bd_queue, 0);
Dan Williams2e6edc952015-11-19 13:29:28 -0800690 if (result)
691 return result;
Tejun Heo3f289dc2018-07-18 04:47:36 -0700692 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
693 REQ_OP_READ);
Dan Williams2e6edc952015-11-19 13:29:28 -0800694 blk_queue_exit(bdev->bd_queue);
695 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700696}
697EXPORT_SYMBOL_GPL(bdev_read_page);
698
699/**
700 * bdev_write_page() - Start writing a page to a block device
701 * @bdev: The device to write the page to
702 * @sector: The offset on the device to write the page to (need not be aligned)
703 * @page: The page to write
704 * @wbc: The writeback_control for the write
705 *
706 * On entry, the page should be locked and not currently under writeback.
707 * On exit, if the write started successfully, the page will be unlocked and
708 * under writeback. If the write failed already (eg the driver failed to
709 * queue the page to the device), the page will still be locked. If the
710 * caller is a ->writepage implementation, it will need to unlock the page.
711 *
712 * Errors returned by this function are usually "soft", eg out of memory, or
713 * queue full; callers should try a different route to write this page rather
714 * than propagate an error back up the stack.
715 *
716 * Return: negative errno if an error occurs, 0 if submission was successful.
717 */
718int bdev_write_page(struct block_device *bdev, sector_t sector,
719 struct page *page, struct writeback_control *wbc)
720{
721 int result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700722 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc952015-11-19 13:29:28 -0800723
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400724 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700725 return -EOPNOTSUPP;
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800726 result = blk_queue_enter(bdev->bd_queue, 0);
Dan Williams2e6edc952015-11-19 13:29:28 -0800727 if (result)
728 return result;
729
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700730 set_page_writeback(page);
Tejun Heo3f289dc2018-07-18 04:47:36 -0700731 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
732 REQ_OP_WRITE);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700733 if (result) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700734 end_page_writeback(page);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700735 } else {
736 clean_page_buffers(page);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700737 unlock_page(page);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700738 }
Dan Williams2e6edc952015-11-19 13:29:28 -0800739 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700740 return result;
741}
742EXPORT_SYMBOL_GPL(bdev_write_page);
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744/*
745 * pseudo-fs
746 */
747
748static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800749static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
751static struct inode *bdev_alloc_inode(struct super_block *sb)
752{
Christoph Lametere94b1762006-12-06 20:33:17 -0800753 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 if (!ei)
755 return NULL;
756 return &ei->vfs_inode;
757}
758
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100759static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100761 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 struct bdev_inode *bdi = BDEV_I(inode);
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 kmem_cache_free(bdev_cachep, bdi);
765}
766
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100767static void bdev_destroy_inode(struct inode *inode)
768{
769 call_rcu(&inode->i_rcu, bdev_i_callback);
770}
771
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700772static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
774 struct bdev_inode *ei = (struct bdev_inode *) foo;
775 struct block_device *bdev = &ei->bdev;
776
Christoph Lametera35afb82007-05-16 22:10:57 -0700777 memset(bdev, 0, sizeof(*bdev));
778 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700779 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100780#ifdef CONFIG_SYSFS
781 INIT_LIST_HEAD(&bdev->bd_holder_disks);
782#endif
Jan Karaa5a79d02017-03-02 16:50:13 +0100783 bdev->bd_bdi = &noop_backing_dev_info;
Christoph Lametera35afb82007-05-16 22:10:57 -0700784 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800785 /* Initialize mutex for freeze. */
786 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787}
788
Al Virob57922d2010-06-07 14:34:48 -0400789static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
791 struct block_device *bdev = &BDEV_I(inode)->bdev;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700792 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400793 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200794 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 spin_lock(&bdev_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 list_del_init(&bdev->bd_list);
797 spin_unlock(&bdev_lock);
Jan Karaf7597412017-03-23 01:37:00 +0100798 /* Detach inode from wb early as bdi_put() may free bdi->wb */
799 inode_detach_wb(inode);
Jan Karaa5a79d02017-03-02 16:50:13 +0100800 if (bdev->bd_bdi != &noop_backing_dev_info) {
Jan Karab1d2dc562017-02-02 15:56:52 +0100801 bdi_put(bdev->bd_bdi);
Jan Karaa5a79d02017-03-02 16:50:13 +0100802 bdev->bd_bdi = &noop_backing_dev_info;
803 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800806static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 .statfs = simple_statfs,
808 .alloc_inode = bdev_alloc_inode,
809 .destroy_inode = bdev_destroy_inode,
810 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400811 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812};
813
Al Viro51139ad2010-07-25 23:47:46 +0400814static struct dentry *bd_mount(struct file_system_type *fs_type,
815 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
Shaohua Li3684aa72016-02-22 15:27:40 -0700817 struct dentry *dent;
818 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Vegard Nossume9e5e3f2016-08-22 12:47:43 +0200819 if (!IS_ERR(dent))
Shaohua Li3684aa72016-02-22 15:27:40 -0700820 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
821 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822}
823
824static struct file_system_type bd_type = {
825 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400826 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 .kill_sb = kill_anon_super,
828};
829
Tejun Heoa212b102015-05-22 17:13:33 -0400830struct super_block *blockdev_superblock __read_mostly;
831EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
833void __init bdev_cache_init(void)
834{
835 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300836 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800837
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800839 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800840 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900841 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 err = register_filesystem(&bd_type);
843 if (err)
844 panic("Cannot register bdev pseudo-fs");
845 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 if (IS_ERR(bd_mnt))
847 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300848 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849}
850
851/*
852 * Most likely _very_ bad one - but then it's hardly critical for small
853 * /dev and can be fixed when somebody will need really large one.
854 * Keep in mind that it will be fed through icache hash function too.
855 */
856static inline unsigned long hash(dev_t dev)
857{
858 return MAJOR(dev)+MINOR(dev);
859}
860
861static int bdev_test(struct inode *inode, void *data)
862{
863 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
864}
865
866static int bdev_set(struct inode *inode, void *data)
867{
868 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
869 return 0;
870}
871
872static LIST_HEAD(all_bdevs);
873
Jan Karaf44f1ab2017-02-02 15:56:49 +0100874/*
875 * If there is a bdev inode for this device, unhash it so that it gets evicted
876 * as soon as last inode reference is dropped.
877 */
878void bdev_unhash_inode(dev_t dev)
879{
880 struct inode *inode;
881
882 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
883 if (inode) {
884 remove_inode_hash(inode);
885 iput(inode);
886 }
887}
888
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889struct block_device *bdget(dev_t dev)
890{
891 struct block_device *bdev;
892 struct inode *inode;
893
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800894 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 bdev_test, bdev_set, &dev);
896
897 if (!inode)
898 return NULL;
899
900 bdev = &BDEV_I(inode)->bdev;
901
902 if (inode->i_state & I_NEW) {
903 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000904 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 bdev->bd_inode = inode;
Fabian Frederick93407472017-02-27 14:28:32 -0800906 bdev->bd_block_size = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 bdev->bd_part_count = 0;
908 bdev->bd_invalidated = 0;
909 inode->i_mode = S_IFBLK;
910 inode->i_rdev = dev;
911 inode->i_bdev = bdev;
912 inode->i_data.a_ops = &def_blk_aops;
913 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 spin_lock(&bdev_lock);
915 list_add(&bdev->bd_list, &all_bdevs);
916 spin_unlock(&bdev_lock);
917 unlock_new_inode(inode);
918 }
919 return bdev;
920}
921
922EXPORT_SYMBOL(bdget);
923
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200924/**
925 * bdgrab -- Grab a reference to an already referenced block device
926 * @bdev: Block device to grab a reference to.
927 */
928struct block_device *bdgrab(struct block_device *bdev)
929{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400930 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200931 return bdev;
932}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700933EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200934
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935long nr_blockdev_pages(void)
936{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700937 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 long ret = 0;
939 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700940 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 ret += bdev->bd_inode->i_mapping->nrpages;
942 }
943 spin_unlock(&bdev_lock);
944 return ret;
945}
946
947void bdput(struct block_device *bdev)
948{
949 iput(bdev->bd_inode);
950}
951
952EXPORT_SYMBOL(bdput);
953
954static struct block_device *bd_acquire(struct inode *inode)
955{
956 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700957
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 spin_lock(&bdev_lock);
959 bdev = inode->i_bdev;
Jan Karacccd9fb2017-02-21 18:09:48 +0100960 if (bdev && !inode_unhashed(bdev->bd_inode)) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100961 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 spin_unlock(&bdev_lock);
963 return bdev;
964 }
965 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700966
Jan Karacccd9fb2017-02-21 18:09:48 +0100967 /*
968 * i_bdev references block device inode that was already shut down
969 * (corresponding device got removed). Remove the reference and look
970 * up block device inode again just in case new device got
971 * reestablished under the same device number.
972 */
973 if (bdev)
974 bd_forget(inode);
975
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 bdev = bdget(inode->i_rdev);
977 if (bdev) {
978 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700979 if (!inode->i_bdev) {
980 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400981 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700982 * and it's released in clear_inode() of inode.
983 * So, we can access it via ->i_mapping always
984 * without igrab().
985 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100986 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700987 inode->i_bdev = bdev;
988 inode->i_mapping = bdev->bd_inode->i_mapping;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 spin_unlock(&bdev_lock);
991 }
992 return bdev;
993}
994
995/* Call when you free inode */
996
997void bd_forget(struct inode *inode)
998{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700999 struct block_device *bdev = NULL;
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -07001002 if (!sb_is_blkdev_sb(inode->i_sb))
1003 bdev = inode->i_bdev;
Al Viroa4a4f942016-07-19 13:16:52 -04001004 inode->i_bdev = NULL;
1005 inode->i_mapping = &inode->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -07001007
1008 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +01001009 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010}
1011
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001012/**
1013 * bd_may_claim - test whether a block device can be claimed
1014 * @bdev: block device of interest
1015 * @whole: whole block device containing @bdev, may equal @bdev
1016 * @holder: holder trying to claim @bdev
1017 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001018 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001019 *
1020 * CONTEXT:
1021 * spin_lock(&bdev_lock).
1022 *
1023 * RETURNS:
1024 * %true if @bdev can be claimed, %false otherwise.
1025 */
1026static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1027 void *holder)
1028{
1029 if (bdev->bd_holder == holder)
1030 return true; /* already a holder */
1031 else if (bdev->bd_holder != NULL)
1032 return false; /* held by someone else */
NeilBrownbcc7f5b2016-12-12 08:21:51 -07001033 else if (whole == bdev)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001034 return true; /* is a whole device which isn't held */
1035
Tejun Heoe525fd82010-11-13 11:55:17 +01001036 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001037 return true; /* is a partition of a device that is being partitioned */
1038 else if (whole->bd_holder != NULL)
1039 return false; /* is a partition of a held device */
1040 else
1041 return true; /* is a partition of an un-held device */
1042}
1043
1044/**
Tejun Heo6b4517a2010-04-07 18:53:59 +09001045 * bd_prepare_to_claim - prepare to claim a block device
1046 * @bdev: block device of interest
1047 * @whole: the whole device containing @bdev, may equal @bdev
1048 * @holder: holder trying to claim @bdev
1049 *
1050 * Prepare to claim @bdev. This function fails if @bdev is already
1051 * claimed by another holder and waits if another claiming is in
1052 * progress. This function doesn't actually claim. On successful
1053 * return, the caller has ownership of bd_claiming and bd_holder[s].
1054 *
1055 * CONTEXT:
1056 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1057 * it multiple times.
1058 *
1059 * RETURNS:
1060 * 0 if @bdev can be claimed, -EBUSY otherwise.
1061 */
1062static int bd_prepare_to_claim(struct block_device *bdev,
1063 struct block_device *whole, void *holder)
1064{
1065retry:
1066 /* if someone else claimed, fail */
1067 if (!bd_may_claim(bdev, whole, holder))
1068 return -EBUSY;
1069
Tejun Heoe75aa852010-08-04 17:59:39 +02001070 /* if claiming is already in progress, wait for it to finish */
1071 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +09001072 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1073 DEFINE_WAIT(wait);
1074
1075 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1076 spin_unlock(&bdev_lock);
1077 schedule();
1078 finish_wait(wq, &wait);
1079 spin_lock(&bdev_lock);
1080 goto retry;
1081 }
1082
1083 /* yay, all mine */
1084 return 0;
1085}
1086
Jan Kara560e7cb2018-02-26 13:01:42 +01001087static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1088{
1089 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1090
1091 if (!disk)
1092 return NULL;
1093 /*
1094 * Now that we hold gendisk reference we make sure bdev we looked up is
1095 * not stale. If it is, it means device got removed and created before
1096 * we looked up gendisk and we fail open in such case. Associating
1097 * unhashed bdev with newly created gendisk could lead to two bdevs
1098 * (and thus two independent caches) being associated with one device
1099 * which is bad.
1100 */
1101 if (inode_unhashed(bdev->bd_inode)) {
1102 put_disk_and_module(disk);
1103 return NULL;
1104 }
1105 return disk;
1106}
1107
Tejun Heo6b4517a2010-04-07 18:53:59 +09001108/**
1109 * bd_start_claiming - start claiming a block device
1110 * @bdev: block device of interest
1111 * @holder: holder trying to claim @bdev
1112 *
1113 * @bdev is about to be opened exclusively. Check @bdev can be opened
1114 * exclusively and mark that an exclusive open is in progress. Each
1115 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +10001116 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1117 * fail).
1118 *
1119 * This function is used to gain exclusive access to the block device
1120 * without actually causing other exclusive open attempts to fail. It
1121 * should be used when the open sequence itself requires exclusive
1122 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +09001123 *
1124 * CONTEXT:
1125 * Might sleep.
1126 *
1127 * RETURNS:
1128 * Pointer to the block device containing @bdev on success, ERR_PTR()
1129 * value on failure.
1130 */
1131static struct block_device *bd_start_claiming(struct block_device *bdev,
1132 void *holder)
1133{
1134 struct gendisk *disk;
1135 struct block_device *whole;
1136 int partno, err;
1137
1138 might_sleep();
1139
1140 /*
1141 * @bdev might not have been initialized properly yet, look up
1142 * and grab the outer block device the hard way.
1143 */
Jan Kara560e7cb2018-02-26 13:01:42 +01001144 disk = bdev_get_gendisk(bdev, &partno);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001145 if (!disk)
1146 return ERR_PTR(-ENXIO);
1147
Tejun Heod4c208b2011-06-13 12:45:48 +02001148 /*
1149 * Normally, @bdev should equal what's returned from bdget_disk()
1150 * if partno is 0; however, some drivers (floppy) use multiple
1151 * bdev's for the same physical device and @bdev may be one of the
1152 * aliases. Keep @bdev if partno is 0. This means claimer
1153 * tracking is broken for those devices but it has always been that
1154 * way.
1155 */
1156 if (partno)
1157 whole = bdget_disk(disk, 0);
1158 else
1159 whole = bdgrab(bdev);
1160
Jan Kara9df6c292018-02-26 13:01:39 +01001161 put_disk_and_module(disk);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001162 if (!whole)
1163 return ERR_PTR(-ENOMEM);
1164
1165 /* prepare to claim, if successful, mark claiming in progress */
1166 spin_lock(&bdev_lock);
1167
1168 err = bd_prepare_to_claim(bdev, whole, holder);
1169 if (err == 0) {
1170 whole->bd_claiming = holder;
1171 spin_unlock(&bdev_lock);
1172 return whole;
1173 } else {
1174 spin_unlock(&bdev_lock);
1175 bdput(whole);
1176 return ERR_PTR(err);
1177 }
1178}
1179
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001180#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +01001181struct bd_holder_disk {
1182 struct list_head list;
1183 struct gendisk *disk;
1184 int refcnt;
1185};
1186
1187static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1188 struct gendisk *disk)
1189{
1190 struct bd_holder_disk *holder;
1191
1192 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1193 if (holder->disk == disk)
1194 return holder;
1195 return NULL;
1196}
1197
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -07001198static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001199{
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -07001200 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001201}
1202
1203static void del_symlink(struct kobject *from, struct kobject *to)
1204{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001205 sysfs_remove_link(from, kobject_name(to));
1206}
1207
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001208/**
Tejun Heoe09b4572010-11-13 11:55:17 +01001209 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1210 * @bdev: the claimed slave bdev
1211 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001212 *
Tejun Heo49731ba2011-01-14 18:43:57 +01001213 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1214 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001215 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001216 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001217 * - from "slaves" directory of the holder @disk to the claimed @bdev
1218 * - from "holders" directory of the @bdev to the holder @disk
1219 *
1220 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1221 * passed to bd_link_disk_holder(), then:
1222 *
1223 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1224 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1225 *
1226 * The caller must have claimed @bdev before calling this function and
1227 * ensure that both @bdev and @disk are valid during the creation and
1228 * lifetime of these symlinks.
1229 *
1230 * CONTEXT:
1231 * Might sleep.
1232 *
1233 * RETURNS:
1234 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001235 */
Tejun Heoe09b4572010-11-13 11:55:17 +01001236int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001237{
Tejun Heo49731ba2011-01-14 18:43:57 +01001238 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001239 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001240
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001241 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001242
Tejun Heo49731ba2011-01-14 18:43:57 +01001243 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -07001244
Tejun Heoe09b4572010-11-13 11:55:17 +01001245 /* FIXME: remove the following once add_disk() handles errors */
1246 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1247 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -07001248
Tejun Heo49731ba2011-01-14 18:43:57 +01001249 holder = bd_find_holder_disk(bdev, disk);
1250 if (holder) {
1251 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +01001252 goto out_unlock;
1253 }
1254
Tejun Heo49731ba2011-01-14 18:43:57 +01001255 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1256 if (!holder) {
1257 ret = -ENOMEM;
1258 goto out_unlock;
1259 }
1260
1261 INIT_LIST_HEAD(&holder->list);
1262 holder->disk = disk;
1263 holder->refcnt = 1;
1264
1265 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1266 if (ret)
1267 goto out_free;
1268
1269 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1270 if (ret)
1271 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +01001272 /*
1273 * bdev could be deleted beneath us which would implicitly destroy
1274 * the holder directory. Hold on to it.
1275 */
1276 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001277
1278 list_add(&holder->list, &bdev->bd_holder_disks);
1279 goto out_unlock;
1280
1281out_del:
1282 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1283out_free:
1284 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001285out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001286 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001287 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001288}
Tejun Heoe09b4572010-11-13 11:55:17 +01001289EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001290
Tejun Heo49731ba2011-01-14 18:43:57 +01001291/**
1292 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1293 * @bdev: the calimed slave bdev
1294 * @disk: the holding disk
1295 *
1296 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1297 *
1298 * CONTEXT:
1299 * Might sleep.
1300 */
1301void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001302{
Tejun Heo49731ba2011-01-14 18:43:57 +01001303 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001304
Tejun Heo49731ba2011-01-14 18:43:57 +01001305 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001306
Tejun Heo49731ba2011-01-14 18:43:57 +01001307 holder = bd_find_holder_disk(bdev, disk);
1308
1309 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1310 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1311 del_symlink(bdev->bd_part->holder_dir,
1312 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001313 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001314 list_del_init(&holder->list);
1315 kfree(holder);
1316 }
1317
1318 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001319}
Tejun Heo49731ba2011-01-14 18:43:57 +01001320EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001321#endif
1322
Andrew Patterson0c002c22008-09-04 14:27:20 -06001323/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001324 * flush_disk - invalidates all buffer-cache entries on a disk
1325 *
1326 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001327 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001328 *
1329 * Invalidates all buffer-cache entries on a disk. It should be called
1330 * when a disk has been changed -- either by a media change or online
1331 * resize.
1332 */
NeilBrown93b270f2011-02-24 17:25:47 +11001333static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001334{
NeilBrown93b270f2011-02-24 17:25:47 +11001335 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001336 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001337 "resized disk %s\n",
1338 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001339 }
1340
1341 if (!bdev->bd_disk)
1342 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001343 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001344 bdev->bd_invalidated = 1;
1345}
1346
1347/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001348 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001349 * @disk: struct gendisk to check
1350 * @bdev: struct bdev to adjust.
Christoph Hellwig5afb7832018-05-29 16:42:59 +02001351 * @verbose: if %true log a message about a size change if there is any
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001352 *
1353 * This routine checks to see if the bdev size does not match the disk size
shunki-fujita849cf552018-04-05 16:20:07 -07001354 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1355 * are freed.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001356 */
Christoph Hellwig5afb7832018-05-29 16:42:59 +02001357void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1358 bool verbose)
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001359{
1360 loff_t disk_size, bdev_size;
1361
1362 disk_size = (loff_t)get_capacity(disk) << 9;
1363 bdev_size = i_size_read(bdev->bd_inode);
1364 if (disk_size != bdev_size) {
Christoph Hellwig5afb7832018-05-29 16:42:59 +02001365 if (verbose) {
1366 printk(KERN_INFO
1367 "%s: detected capacity change from %lld to %lld\n",
1368 disk->disk_name, bdev_size, disk_size);
1369 }
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001370 i_size_write(bdev->bd_inode, disk_size);
shunki-fujita849cf552018-04-05 16:20:07 -07001371 if (bdev_size > disk_size)
1372 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001373 }
1374}
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001375
1376/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001377 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001378 * @disk: struct gendisk to be revalidated
1379 *
1380 * This routine is a wrapper for lower-level driver's revalidate_disk
1381 * call-backs. It is used to do common pre and post operations needed
1382 * for all revalidate_disk operations.
1383 */
1384int revalidate_disk(struct gendisk *disk)
1385{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001386 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001387 int ret = 0;
1388
1389 if (disk->fops->revalidate_disk)
1390 ret = disk->fops->revalidate_disk(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001391 bdev = bdget_disk(disk, 0);
1392 if (!bdev)
1393 return ret;
1394
1395 mutex_lock(&bdev->bd_mutex);
Christoph Hellwig5afb7832018-05-29 16:42:59 +02001396 check_disk_size_change(disk, bdev, ret == 0);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001397 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001398 mutex_unlock(&bdev->bd_mutex);
1399 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001400 return ret;
1401}
1402EXPORT_SYMBOL(revalidate_disk);
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404/*
1405 * This routine checks whether a removable media has been changed,
1406 * and invalidates all buffer-cache-entries in that case. This
1407 * is a relatively slow routine, so we have to try to minimize using
1408 * it. Thus it is called only upon a 'mount' or 'open'. This
1409 * is the best way of combining speed and utility, I think.
1410 * People changing diskettes in the middle of an operation deserve
1411 * to lose :-)
1412 */
1413int check_disk_change(struct block_device *bdev)
1414{
1415 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001416 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001417 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Tejun Heo77ea8872010-12-08 20:57:37 +01001419 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1420 DISK_EVENT_EJECT_REQUEST);
1421 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 return 0;
1423
NeilBrown93b270f2011-02-24 17:25:47 +11001424 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 if (bdops->revalidate_disk)
1426 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 return 1;
1428}
1429
1430EXPORT_SYMBOL(check_disk_change);
1431
1432void bd_set_size(struct block_device *bdev, loff_t size)
1433{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001434 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Al Viro59551022016-01-22 15:40:57 -05001436 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001437 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001438 inode_unlock(bdev->bd_inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001439 while (bsize < PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (size & bsize)
1441 break;
1442 bsize <<= 1;
1443 }
1444 bdev->bd_block_size = bsize;
1445 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1446}
1447EXPORT_SYMBOL(bd_set_size);
1448
Al Viro4385bab2013-05-05 22:11:03 -04001449static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001450
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001451/*
1452 * bd_mutex locking:
1453 *
1454 * mutex_lock(part->bd_mutex)
1455 * mutex_lock_nested(whole->bd_mutex, 1)
1456 */
1457
Al Viro572c4892007-10-08 13:24:05 -04001458static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001461 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001462 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001463 int perm = 0;
Jan Kara89736652018-02-26 13:01:40 +01001464 bool first_open = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
Al Viro572c4892007-10-08 13:24:05 -04001466 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001467 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001468 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001469 perm |= MAY_WRITE;
1470 /*
1471 * hooks: /n/, see "layering violations".
1472 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001473 if (!for_part) {
1474 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1475 if (ret != 0) {
1476 bdput(bdev);
1477 return ret;
1478 }
Al Viro82666022008-08-01 05:32:04 -04001479 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001480
NeilBrownd3374822009-01-09 08:31:10 +11001481 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001482
Tejun Heo89f97492008-11-05 10:21:06 +01001483 ret = -ENXIO;
Jan Kara560e7cb2018-02-26 13:01:42 +01001484 disk = bdev_get_gendisk(bdev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001485 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001486 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Tejun Heo69e02c52011-03-09 19:54:27 +01001488 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001489 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 if (!bdev->bd_openers) {
Jan Kara89736652018-02-26 13:01:40 +01001491 first_open = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001493 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 bdev->bd_contains = bdev;
Christoph Hellwigc2ee0702017-08-23 19:10:31 +02001495 bdev->bd_partno = partno;
Dan Williams03cdadb2016-02-26 15:19:43 -08001496
Tejun Heocf771cb2008-09-03 09:01:09 +02001497 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001498 ret = -ENXIO;
1499 bdev->bd_part = disk_get_part(disk, partno);
1500 if (!bdev->bd_part)
1501 goto out_clear;
1502
Tejun Heo1196f8b2011-04-21 20:54:45 +02001503 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001505 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001506 if (ret == -ERESTARTSYS) {
1507 /* Lost a race with 'disk' being
1508 * deleted, try again.
1509 * See md.c
1510 */
1511 disk_put_part(bdev->bd_part);
1512 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001513 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001514 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001515 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001516 disk_unblock_events(disk);
Jan Kara9df6c292018-02-26 13:01:39 +01001517 put_disk_and_module(disk);
NeilBrownd3374822009-01-09 08:31:10 +11001518 goto restart;
1519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 }
Tejun Heo7e697232011-05-23 13:26:07 +02001521
Christoph Hellwig22375702016-09-14 11:56:13 +02001522 if (!ret)
Tejun Heo7e697232011-05-23 13:26:07 +02001523 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Tejun Heo7e697232011-05-23 13:26:07 +02001524
Tejun Heo1196f8b2011-04-21 20:54:45 +02001525 /*
1526 * If the device is invalidated, rescan partition
1527 * if open succeeded or failed with -ENOMEDIUM.
1528 * The latter is necessary to prevent ghost
1529 * partitions on a removed medium.
1530 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001531 if (bdev->bd_invalidated) {
1532 if (!ret)
1533 rescan_partitions(disk, bdev);
1534 else if (ret == -ENOMEDIUM)
1535 invalidate_partitions(disk, bdev);
1536 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001537
Tejun Heo1196f8b2011-04-21 20:54:45 +02001538 if (ret)
1539 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 struct block_device *whole;
1542 whole = bdget_disk(disk, 0);
1543 ret = -ENOMEM;
1544 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001545 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001546 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001547 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001549 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001551 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001552 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001553 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001555 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 }
Tejun Heo89f97492008-11-05 10:21:06 +01001557 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 }
Jan Kara03e26272017-03-23 01:36:53 +01001559
1560 if (bdev->bd_bdi == &noop_backing_dev_info)
1561 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001564 ret = 0;
1565 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001566 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001567 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001568 if (bdev->bd_invalidated) {
1569 if (!ret)
1570 rescan_partitions(bdev->bd_disk, bdev);
1571 else if (ret == -ENOMEDIUM)
1572 invalidate_partitions(bdev->bd_disk, bdev);
1573 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001574 if (ret)
1575 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 }
1577 }
1578 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001579 if (for_part)
1580 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001581 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001582 disk_unblock_events(disk);
Jan Kara89736652018-02-26 13:01:40 +01001583 /* only one opener holds refs to the module and disk */
1584 if (!first_open)
1585 put_disk_and_module(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 return 0;
1587
Tejun Heo0762b8b2008-08-25 19:56:12 +09001588 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001589 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001591 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001592 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001594 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001596 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001597 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001598 disk_unblock_events(disk);
Jan Kara9df6c292018-02-26 13:01:39 +01001599 put_disk_and_module(disk);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001600 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001601 bdput(bdev);
1602
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 return ret;
1604}
1605
Tejun Heod4d77622010-11-13 11:55:18 +01001606/**
1607 * blkdev_get - open a block device
1608 * @bdev: block_device to open
1609 * @mode: FMODE_* mask
1610 * @holder: exclusive holder identifier
1611 *
1612 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1613 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1614 * @holder is invalid. Exclusive opens may nest for the same @holder.
1615 *
1616 * On success, the reference count of @bdev is unchanged. On failure,
1617 * @bdev is put.
1618 *
1619 * CONTEXT:
1620 * Might sleep.
1621 *
1622 * RETURNS:
1623 * 0 on success, -errno on failure.
1624 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001625int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
Tejun Heoe525fd82010-11-13 11:55:17 +01001627 struct block_device *whole = NULL;
1628 int res;
1629
1630 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1631
1632 if ((mode & FMODE_EXCL) && holder) {
1633 whole = bd_start_claiming(bdev, holder);
1634 if (IS_ERR(whole)) {
1635 bdput(bdev);
1636 return PTR_ERR(whole);
1637 }
1638 }
1639
1640 res = __blkdev_get(bdev, mode, 0);
1641
1642 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001643 struct gendisk *disk = whole->bd_disk;
1644
Tejun Heo6a027ef2010-11-13 11:55:17 +01001645 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001646 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001647 spin_lock(&bdev_lock);
1648
Tejun Heo77ea8872010-12-08 20:57:37 +01001649 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001650 BUG_ON(!bd_may_claim(bdev, whole, holder));
1651 /*
1652 * Note that for a whole device bd_holders
1653 * will be incremented twice, and bd_holder
1654 * will be set to bd_may_claim before being
1655 * set to holder
1656 */
1657 whole->bd_holders++;
1658 whole->bd_holder = bd_may_claim;
1659 bdev->bd_holders++;
1660 bdev->bd_holder = holder;
1661 }
1662
1663 /* tell others that we're done */
1664 BUG_ON(whole->bd_claiming != holder);
1665 whole->bd_claiming = NULL;
1666 wake_up_bit(&whole->bd_claiming, 0);
1667
1668 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001669
1670 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001671 * Block event polling for write claims if requested. Any
1672 * write holder makes the write_holder state stick until
1673 * all are released. This is good enough and tracking
1674 * individual writeable reference is too fragile given the
1675 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001676 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001677 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1678 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001679 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001680 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001681 }
1682
1683 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001684 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001685 }
1686
1687 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001688}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689EXPORT_SYMBOL(blkdev_get);
1690
Tejun Heod4d77622010-11-13 11:55:18 +01001691/**
1692 * blkdev_get_by_path - open a block device by name
1693 * @path: path to the block device to open
1694 * @mode: FMODE_* mask
1695 * @holder: exclusive holder identifier
1696 *
1697 * Open the blockdevice described by the device file at @path. @mode
1698 * and @holder are identical to blkdev_get().
1699 *
1700 * On success, the returned block_device has reference count of one.
1701 *
1702 * CONTEXT:
1703 * Might sleep.
1704 *
1705 * RETURNS:
1706 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1707 */
1708struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1709 void *holder)
1710{
1711 struct block_device *bdev;
1712 int err;
1713
1714 bdev = lookup_bdev(path);
1715 if (IS_ERR(bdev))
1716 return bdev;
1717
1718 err = blkdev_get(bdev, mode, holder);
1719 if (err)
1720 return ERR_PTR(err);
1721
Chuck Ebberte51900f2011-02-16 18:11:53 -05001722 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1723 blkdev_put(bdev, mode);
1724 return ERR_PTR(-EACCES);
1725 }
1726
Tejun Heod4d77622010-11-13 11:55:18 +01001727 return bdev;
1728}
1729EXPORT_SYMBOL(blkdev_get_by_path);
1730
1731/**
1732 * blkdev_get_by_dev - open a block device by device number
1733 * @dev: device number of block device to open
1734 * @mode: FMODE_* mask
1735 * @holder: exclusive holder identifier
1736 *
1737 * Open the blockdevice described by device number @dev. @mode and
1738 * @holder are identical to blkdev_get().
1739 *
1740 * Use it ONLY if you really do not have anything better - i.e. when
1741 * you are behind a truly sucky interface and all you are given is a
1742 * device number. _Never_ to be used for internal purposes. If you
1743 * ever need it - reconsider your API.
1744 *
1745 * On success, the returned block_device has reference count of one.
1746 *
1747 * CONTEXT:
1748 * Might sleep.
1749 *
1750 * RETURNS:
1751 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1752 */
1753struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1754{
1755 struct block_device *bdev;
1756 int err;
1757
1758 bdev = bdget(dev);
1759 if (!bdev)
1760 return ERR_PTR(-ENOMEM);
1761
1762 err = blkdev_get(bdev, mode, holder);
1763 if (err)
1764 return ERR_PTR(err);
1765
1766 return bdev;
1767}
1768EXPORT_SYMBOL(blkdev_get_by_dev);
1769
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770static int blkdev_open(struct inode * inode, struct file * filp)
1771{
1772 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774 /*
1775 * Preserve backwards compatibility and allow large file access
1776 * even if userspace doesn't ask for it explicitly. Some mkfs
1777 * binary needs it. We might want to drop this workaround
1778 * during an unstable branch.
1779 */
1780 filp->f_flags |= O_LARGEFILE;
1781
Christoph Hellwigc35fc7a2017-08-29 16:13:21 +02001782 filp->f_mode |= FMODE_NOWAIT;
1783
Al Viro572c4892007-10-08 13:24:05 -04001784 if (filp->f_flags & O_NDELAY)
1785 filp->f_mode |= FMODE_NDELAY;
1786 if (filp->f_flags & O_EXCL)
1787 filp->f_mode |= FMODE_EXCL;
1788 if ((filp->f_flags & O_ACCMODE) == 3)
1789 filp->f_mode |= FMODE_WRITE_IOCTL;
1790
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001792 if (bdev == NULL)
1793 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
Al Viro572c4892007-10-08 13:24:05 -04001795 filp->f_mapping = bdev->bd_inode->i_mapping;
Jeff Layton5660e132017-07-06 07:02:25 -04001796 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
Al Viro572c4892007-10-08 13:24:05 -04001797
Tejun Heoe525fd82010-11-13 11:55:17 +01001798 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799}
1800
Al Viro4385bab2013-05-05 22:11:03 -04001801static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001802{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001803 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001804 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001805
NeilBrown6796bf52006-12-08 02:36:16 -08001806 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001807 if (for_part)
1808 bdev->bd_part_count--;
1809
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001810 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001811 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001812 sync_blockdev(bdev);
1813 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001814
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001815 bdev_write_inode(bdev);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001816 }
1817 if (bdev->bd_contains == bdev) {
1818 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001819 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001820 }
1821 if (!bdev->bd_openers) {
Tejun Heo0762b8b2008-08-25 19:56:12 +09001822 disk_put_part(bdev->bd_part);
1823 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001824 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001825 if (bdev != bdev->bd_contains)
1826 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001827 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001828
Jan Kara9df6c292018-02-26 13:01:39 +01001829 put_disk_and_module(disk);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001830 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001831 mutex_unlock(&bdev->bd_mutex);
1832 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001833 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001834 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001835}
1836
Al Viro4385bab2013-05-05 22:11:03 -04001837void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001838{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001839 mutex_lock(&bdev->bd_mutex);
1840
Tejun Heoe525fd82010-11-13 11:55:17 +01001841 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001842 bool bdev_free;
1843
1844 /*
1845 * Release a claim on the device. The holder fields
1846 * are protected with bdev_lock. bd_mutex is to
1847 * synchronize disk_holder unlinking.
1848 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001849 spin_lock(&bdev_lock);
1850
1851 WARN_ON_ONCE(--bdev->bd_holders < 0);
1852 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1853
1854 /* bd_contains might point to self, check in a separate step */
1855 if ((bdev_free = !bdev->bd_holders))
1856 bdev->bd_holder = NULL;
1857 if (!bdev->bd_contains->bd_holders)
1858 bdev->bd_contains->bd_holder = NULL;
1859
1860 spin_unlock(&bdev_lock);
1861
Tejun Heo77ea8872010-12-08 20:57:37 +01001862 /*
1863 * If this was the last claim, remove holder link and
1864 * unblock evpoll if it was a write holder.
1865 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001866 if (bdev_free && bdev->bd_write_holder) {
1867 disk_unblock_events(bdev->bd_disk);
1868 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001869 }
Tejun Heo69362172011-03-09 19:54:27 +01001870 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001871
Tejun Heo85ef06d2011-07-01 16:17:47 +02001872 /*
1873 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1874 * event. This is to ensure detection of media removal commanded
1875 * from userland - e.g. eject(1).
1876 */
1877 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1878
1879 mutex_unlock(&bdev->bd_mutex);
1880
Al Viro4385bab2013-05-05 22:11:03 -04001881 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001882}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001883EXPORT_SYMBOL(blkdev_put);
1884
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885static int blkdev_close(struct inode * inode, struct file * filp)
1886{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001887 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001888 blkdev_put(bdev, filp->f_mode);
1889 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890}
1891
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001892static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001894 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001895 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001896
1897 /*
1898 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1899 * to updated it before every ioctl.
1900 */
Al Viro56b26ad2008-09-19 03:17:36 -04001901 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001902 mode |= FMODE_NDELAY;
1903 else
1904 mode &= ~FMODE_NDELAY;
1905
Al Viro56b26ad2008-09-19 03:17:36 -04001906 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907}
1908
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001909/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001910 * Write data to the block device. Only intended for the block device itself
1911 * and the raw driver which basically is a fake block device.
1912 *
1913 * Does not take i_mutex for the write and thus is not for general purpose
1914 * use.
1915 */
Al Viro1456c0a2014-04-03 03:21:50 -04001916ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001917{
1918 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001919 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001920 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001921 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001922 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001923
Al Viro7ec7b942015-04-07 11:35:14 -04001924 if (bdev_read_only(I_BDEV(bd_inode)))
1925 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001926
Al Viro7ec7b942015-04-07 11:35:14 -04001927 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001928 return 0;
1929
Al Viro7ec7b942015-04-07 11:35:14 -04001930 if (iocb->ki_pos >= size)
1931 return -ENOSPC;
1932
Christoph Hellwigc35fc7a2017-08-29 16:13:21 +02001933 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1934 return -EOPNOTSUPP;
1935
Al Viro7ec7b942015-04-07 11:35:14 -04001936 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001937
Jianpeng Ma53362a02012-08-02 09:50:39 +02001938 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001939 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwige2592212016-04-07 08:52:01 -07001940 if (ret > 0)
1941 ret = generic_write_sync(iocb, ret);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001942 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001943 return ret;
1944}
Al Viro1456c0a2014-04-03 03:21:50 -04001945EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001946
David Jefferyb2de5252014-09-29 10:21:10 -04001947ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001948{
1949 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001950 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001951 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001952 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001953
1954 if (pos >= size)
1955 return 0;
1956
1957 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001958 iov_iter_truncate(to, size);
1959 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001960}
David Jefferyb2de5252014-09-29 10:21:10 -04001961EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001962
Christoph Hellwigeef99382009-08-20 17:43:41 +02001963/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001964 * Try to release a page associated with block device when the system
1965 * is under memory pressure.
1966 */
1967static int blkdev_releasepage(struct page *page, gfp_t wait)
1968{
1969 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1970
1971 if (super && super->s_op->bdev_try_to_free_page)
1972 return super->s_op->bdev_try_to_free_page(super, page, wait);
1973
1974 return try_to_free_buffers(page);
1975}
1976
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001977static int blkdev_writepages(struct address_space *mapping,
1978 struct writeback_control *wbc)
1979{
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001980 return generic_writepages(mapping, wbc);
1981}
1982
Adrian Bunk4c54ac622008-02-18 13:48:31 +01001983static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001985 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001987 .write_begin = blkdev_write_begin,
1988 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001989 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001990 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 .direct_IO = blkdev_direct_IO,
Jan Kara88dbcbb2018-12-28 00:39:16 -08001992 .migratepage = buffer_migrate_page_norefs,
Mel Gormanb4597222013-07-03 15:02:05 -07001993 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994};
1995
Darrick J. Wong25f4c412016-10-11 13:51:11 -07001996#define BLKDEV_FALLOC_FL_SUPPORTED \
1997 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1998 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1999
2000static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2001 loff_t len)
2002{
2003 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002004 struct address_space *mapping;
2005 loff_t end = start + len - 1;
2006 loff_t isize;
2007 int error;
2008
2009 /* Fail if we don't recognize the flags. */
2010 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2011 return -EOPNOTSUPP;
2012
2013 /* Don't go off the end of the device. */
2014 isize = i_size_read(bdev->bd_inode);
2015 if (start >= isize)
2016 return -EINVAL;
2017 if (end >= isize) {
2018 if (mode & FALLOC_FL_KEEP_SIZE) {
2019 len = isize - start;
2020 end = start + len - 1;
2021 } else
2022 return -EINVAL;
2023 }
2024
2025 /*
2026 * Don't allow IO that isn't aligned to logical block size.
2027 */
2028 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2029 return -EINVAL;
2030
2031 /* Invalidate the page cache, including dirty pages. */
2032 mapping = bdev->bd_inode->i_mapping;
2033 truncate_inode_pages_range(mapping, start, end);
2034
2035 switch (mode) {
2036 case FALLOC_FL_ZERO_RANGE:
2037 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2038 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
Christoph Hellwigee472d82017-04-05 19:21:08 +02002039 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002040 break;
2041 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
Christoph Hellwig34045122017-04-05 19:21:11 +02002042 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2043 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002044 break;
2045 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002046 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2047 GFP_KERNEL, 0);
2048 break;
2049 default:
2050 return -EOPNOTSUPP;
2051 }
2052 if (error)
2053 return error;
2054
2055 /*
2056 * Invalidate again; if someone wandered in and dirtied a page,
2057 * the caller will be given -EBUSY. The third argument is
2058 * inclusive, so the rounding here is safe.
2059 */
2060 return invalidate_inode_pages2_range(mapping,
2061 start >> PAGE_SHIFT,
2062 end >> PAGE_SHIFT);
2063}
2064
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002065const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 .open = blkdev_open,
2067 .release = blkdev_close,
2068 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04002069 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04002070 .write_iter = blkdev_write_iter,
Dan Williamsacc93d32016-05-07 11:40:28 -07002071 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07002072 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07002073 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074#ifdef CONFIG_COMPAT
2075 .compat_ioctl = compat_blkdev_ioctl,
2076#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08002077 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04002078 .splice_write = iter_file_splice_write,
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002079 .fallocate = blkdev_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080};
2081
2082int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2083{
2084 int res;
2085 mm_segment_t old_fs = get_fs();
2086 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04002087 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 set_fs(old_fs);
2089 return res;
2090}
2091
2092EXPORT_SYMBOL(ioctl_by_bdev);
2093
2094/**
2095 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08002096 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02002098 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 * namespace if possible and return it. Return ERR_PTR(error)
2100 * otherwise.
2101 */
Al Viro421748e2008-08-02 01:04:36 -04002102struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103{
2104 struct block_device *bdev;
2105 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04002106 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 int error;
2108
Al Viro421748e2008-08-02 01:04:36 -04002109 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 return ERR_PTR(-EINVAL);
2111
Al Viro421748e2008-08-02 01:04:36 -04002112 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 if (error)
2114 return ERR_PTR(error);
2115
David Howellsbb6687342015-03-17 22:26:21 +00002116 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 error = -ENOTBLK;
2118 if (!S_ISBLK(inode->i_mode))
2119 goto fail;
2120 error = -EACCES;
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002121 if (!may_open_dev(&path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 goto fail;
2123 error = -ENOMEM;
2124 bdev = bd_acquire(inode);
2125 if (!bdev)
2126 goto fail;
2127out:
Al Viro421748e2008-08-02 01:04:36 -04002128 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 return bdev;
2130fail:
2131 bdev = ERR_PTR(error);
2132 goto out;
2133}
Al Virod5686b42008-08-01 05:00:11 -04002134EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
NeilBrown93b270f2011-02-24 17:25:47 +11002136int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01002137{
2138 struct super_block *sb = get_super(bdev);
2139 int res = 0;
2140
2141 if (sb) {
2142 /*
2143 * no need to lock the super, get_super holds the
2144 * read mutex so the filesystem cannot go away
2145 * under us (->put_super runs with the write lock
2146 * hold).
2147 */
2148 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11002149 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01002150 drop_super(sb);
2151 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07002152 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01002153 return res;
2154}
2155EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002156
2157void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2158{
2159 struct inode *inode, *old_inode = NULL;
2160
Dave Chinner74278da92015-03-04 12:37:22 -05002161 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002162 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2163 struct address_space *mapping = inode->i_mapping;
Rabin Vincentaf309222016-12-01 09:18:28 +01002164 struct block_device *bdev;
Jan Kara5c0d6b62012-07-03 16:45:31 +02002165
2166 spin_lock(&inode->i_lock);
2167 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2168 mapping->nrpages == 0) {
2169 spin_unlock(&inode->i_lock);
2170 continue;
2171 }
2172 __iget(inode);
2173 spin_unlock(&inode->i_lock);
Dave Chinner74278da92015-03-04 12:37:22 -05002174 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002175 /*
2176 * We hold a reference to 'inode' so it couldn't have been
2177 * removed from s_inodes list while we dropped the
Dave Chinner74278da92015-03-04 12:37:22 -05002178 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02002179 * be holding the last reference and we cannot iput it under
Dave Chinner74278da92015-03-04 12:37:22 -05002180 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02002181 * later.
2182 */
2183 iput(old_inode);
2184 old_inode = inode;
Rabin Vincentaf309222016-12-01 09:18:28 +01002185 bdev = I_BDEV(inode);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002186
Rabin Vincentaf309222016-12-01 09:18:28 +01002187 mutex_lock(&bdev->bd_mutex);
2188 if (bdev->bd_openers)
2189 func(bdev, arg);
2190 mutex_unlock(&bdev->bd_mutex);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002191
Dave Chinner74278da92015-03-04 12:37:22 -05002192 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002193 }
Dave Chinner74278da92015-03-04 12:37:22 -05002194 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002195 iput(old_inode);
2196}