blob: 58a4c1217fa87d60376771a2c315b8b02d0bff1c [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 Bunk4c54ac62008-02-18 13:48:31 +010044static const struct address_space_operations def_blk_aops;
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046static inline struct bdev_inode *BDEV_I(struct inode *inode)
47{
48 return container_of(inode, struct bdev_inode, vfs_inode);
49}
50
Geert Uytterhoevenff5053f2015-06-26 13:58:32 +020051struct block_device *I_BDEV(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
53 return &BDEV_I(inode)->bdev;
54}
Linus Torvalds1da177e2005-04-16 15:20:36 -070055EXPORT_SYMBOL(I_BDEV);
56
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070057static void bdev_write_inode(struct block_device *bdev)
Christoph Hellwig564f00f2015-01-14 10:42:33 +010058{
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070059 struct inode *inode = bdev->bd_inode;
60 int ret;
61
Christoph Hellwig564f00f2015-01-14 10:42:33 +010062 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
Vivek Goyaldbd3ca52015-11-09 09:23:40 -070065 ret = write_inode_now(inode, true);
66 if (ret) {
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
71 }
Christoph Hellwig564f00f2015-01-14 10:42:33 +010072 spin_lock(&inode->i_lock);
73 }
74 spin_unlock(&inode->i_lock);
75}
76
Peter Zijlstraf9a14392007-05-06 14:49:55 -070077/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb42011-09-16 02:31:11 -040078void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079{
Al Viroff01bb42011-09-16 02:31:11 -040080 struct address_space *mapping = bdev->bd_inode->i_mapping;
81
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080082 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070083 return;
Al Viroff01bb42011-09-16 02:31:11 -040084
Peter Zijlstraf9a14392007-05-06 14:49:55 -070085 invalidate_bh_lrus();
Al Viroff01bb42011-09-16 02:31:11 -040086 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087}
Al Viroff01bb42011-09-16 02:31:11 -040088EXPORT_SYMBOL(kill_bdev);
89
90/* Invalidate clean unused buffers and pagecache. */
91void invalidate_bdev(struct block_device *bdev)
92{
93 struct address_space *mapping = bdev->bd_inode->i_mapping;
94
Andrey Ryabinina5f6a6a2017-05-03 14:56:02 -070095 if (mapping->nrpages) {
96 invalidate_bh_lrus();
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping, 0, -1);
99 }
Al Viroff01bb42011-09-16 02:31:11 -0400100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
102 */
Dan Magenheimer31677602011-09-21 11:56:28 -0400103 cleancache_invalidate_inode(mapping);
Al Viroff01bb42011-09-16 02:31:11 -0400104}
105EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Jan Kara04906b22019-01-14 09:48:10 +0100107static void set_init_blocksize(struct block_device *bdev)
108{
109 unsigned bsize = bdev_logical_block_size(bdev);
110 loff_t size = i_size_read(bdev->bd_inode);
111
112 while (bsize < PAGE_SIZE) {
113 if (size & bsize)
114 break;
115 bsize <<= 1;
116 }
117 bdev->bd_block_size = bsize;
118 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
119}
120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121int set_blocksize(struct block_device *bdev, int size)
122{
123 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700124 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 return -EINVAL;
126
127 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400128 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 return -EINVAL;
130
131 /* Don't change the size if it is same as current */
132 if (bdev->bd_block_size != size) {
133 sync_blockdev(bdev);
134 bdev->bd_block_size = size;
135 bdev->bd_inode->i_blkbits = blksize_bits(size);
136 kill_bdev(bdev);
137 }
138 return 0;
139}
140
141EXPORT_SYMBOL(set_blocksize);
142
143int sb_set_blocksize(struct super_block *sb, int size)
144{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 if (set_blocksize(sb->s_bdev, size))
146 return 0;
147 /* If we get here, we know size is power of two
148 * and it's value is between 512 and PAGE_SIZE */
149 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800150 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 return sb->s_blocksize;
152}
153
154EXPORT_SYMBOL(sb_set_blocksize);
155
156int sb_min_blocksize(struct super_block *sb, int size)
157{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400158 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 if (size < minsize)
160 size = minsize;
161 return sb_set_blocksize(sb, size);
162}
163
164EXPORT_SYMBOL(sb_min_blocksize);
165
166static int
167blkdev_get_block(struct inode *inode, sector_t iblock,
168 struct buffer_head *bh, int create)
169{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 bh->b_bdev = I_BDEV(inode);
171 bh->b_blocknr = iblock;
172 set_buffer_mapped(bh);
173 return 0;
174}
175
Dan Williams4ebb16c2015-10-28 07:48:19 +0900176static struct inode *bdev_file_inode(struct file *file)
177{
178 return file->f_mapping->host;
179}
180
Jens Axboe78250c02016-11-17 17:50:47 +0100181static unsigned int dio_bio_write_op(struct kiocb *iocb)
182{
183 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
184
185 /* avoid the need for a I/O completion work item */
186 if (iocb->ki_flags & IOCB_DSYNC)
187 op |= REQ_FUA;
188 return op;
189}
190
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600191#define DIO_INLINE_BIO_VECS 4
192
193static void blkdev_bio_end_io_simple(struct bio *bio)
194{
195 struct task_struct *waiter = bio->bi_private;
196
197 WRITE_ONCE(bio->bi_private, NULL);
Jens Axboe06193172018-11-13 21:16:54 -0700198 blk_wake_io_task(waiter);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600199}
200
201static ssize_t
202__blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
203 int nr_pages)
204{
205 struct file *file = iocb->ki_filp;
206 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Jens Axboe72ecad22016-11-16 23:11:42 -0700207 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600208 loff_t pos = iocb->ki_pos;
209 bool should_dirty = false;
210 struct bio bio;
211 ssize_t ret;
212 blk_qc_t qc;
213 int i;
214
Jens Axboe9a794fb2016-11-22 08:12:39 -0700215 if ((pos | iov_iter_alignment(iter)) &
216 (bdev_logical_block_size(bdev) - 1))
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600217 return -EINVAL;
218
Jens Axboe72ecad22016-11-16 23:11:42 -0700219 if (nr_pages <= DIO_INLINE_BIO_VECS)
220 vecs = inline_vecs;
221 else {
Kees Cook6da2ec52018-06-12 13:55:00 -0700222 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
223 GFP_KERNEL);
Jens Axboe72ecad22016-11-16 23:11:42 -0700224 if (!vecs)
225 return -ENOMEM;
226 }
227
Ming Lei3a83f462016-11-22 08:57:21 -0700228 bio_init(&bio, vecs, nr_pages);
Christoph Hellwig74d46992017-08-23 19:10:32 +0200229 bio_set_dev(&bio, bdev);
Damien Le Moal4d1a4762016-11-22 15:38:49 +0900230 bio.bi_iter.bi_sector = pos >> 9;
Jens Axboe45d06cf2017-06-27 11:01:22 -0600231 bio.bi_write_hint = iocb->ki_hint;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600232 bio.bi_private = current;
233 bio.bi_end_io = blkdev_bio_end_io_simple;
Adam Manzanares074111c2018-05-22 10:52:20 -0700234 bio.bi_ioprio = iocb->ki_ioprio;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600235
236 ret = bio_iov_iter_get_pages(&bio, iter);
237 if (unlikely(ret))
Martin Wilck9362dd12018-07-25 23:15:08 +0200238 goto out;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600239 ret = bio.bi_iter.bi_size;
240
241 if (iov_iter_rw(iter) == READ) {
Jens Axboe78250c02016-11-17 17:50:47 +0100242 bio.bi_opf = REQ_OP_READ;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600243 if (iter_is_iovec(iter))
244 should_dirty = true;
245 } else {
Jens Axboe78250c02016-11-17 17:50:47 +0100246 bio.bi_opf = dio_bio_write_op(iocb);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600247 task_io_account_write(ret);
248 }
Jens Axboed1e36282018-08-29 10:36:56 -0600249 if (iocb->ki_flags & IOCB_HIPRI)
250 bio.bi_opf |= REQ_HIPRI;
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600251
252 qc = submit_bio(&bio);
253 for (;;) {
Linus Torvalds1ac5cd42019-01-02 10:46:03 -0800254 set_current_state(TASK_UNINTERRUPTIBLE);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600255 if (!READ_ONCE(bio.bi_private))
256 break;
257 if (!(iocb->ki_flags & IOCB_HIPRI) ||
Jens Axboe0a1b8b82018-11-26 08:24:43 -0700258 !blk_poll(bdev_get_queue(bdev), qc, true))
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600259 io_schedule();
260 }
261 __set_current_state(TASK_RUNNING);
262
263 bio_for_each_segment_all(bvec, &bio, i) {
264 if (should_dirty && !PageCompound(bvec->bv_page))
265 set_page_dirty_lock(bvec->bv_page);
266 put_page(bvec->bv_page);
267 }
268
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200269 if (unlikely(bio.bi_status))
Linus Torvaldsc6b1e362017-07-03 10:34:51 -0700270 ret = blk_status_to_errno(bio.bi_status);
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600271
Martin Wilck9362dd12018-07-25 23:15:08 +0200272out:
273 if (vecs != inline_vecs)
274 kfree(vecs);
275
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600276 bio_uninit(&bio);
277
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600278 return ret;
279}
280
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700281struct blkdev_dio {
282 union {
283 struct kiocb *iocb;
284 struct task_struct *waiter;
285 };
286 size_t size;
287 atomic_t ref;
288 bool multi_bio : 1;
289 bool should_dirty : 1;
290 bool is_sync : 1;
291 struct bio bio;
292};
293
Kent Overstreet52190f82018-05-20 18:25:55 -0400294static struct bio_set blkdev_dio_pool;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700295
296static void blkdev_bio_end_io(struct bio *bio)
297{
298 struct blkdev_dio *dio = bio->bi_private;
299 bool should_dirty = dio->should_dirty;
300
301 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200302 if (bio->bi_status && !dio->bio.bi_status)
303 dio->bio.bi_status = bio->bi_status;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700304 } else {
305 if (!dio->is_sync) {
306 struct kiocb *iocb = dio->iocb;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200307 ssize_t ret;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700308
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200309 if (likely(!dio->bio.bi_status)) {
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700310 ret = dio->size;
311 iocb->ki_pos += ret;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200312 } else {
313 ret = blk_status_to_errno(dio->bio.bi_status);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700314 }
315
316 dio->iocb->ki_complete(iocb, ret, 0);
Christoph Hellwig531724a2018-11-30 09:23:48 +0100317 if (dio->multi_bio)
318 bio_put(&dio->bio);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700319 } else {
320 struct task_struct *waiter = dio->waiter;
321
322 WRITE_ONCE(dio->waiter, NULL);
Jens Axboe06193172018-11-13 21:16:54 -0700323 blk_wake_io_task(waiter);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700324 }
325 }
326
327 if (should_dirty) {
328 bio_check_pages_dirty(bio);
329 } else {
330 struct bio_vec *bvec;
331 int i;
332
333 bio_for_each_segment_all(bvec, bio, i)
334 put_page(bvec->bv_page);
335 bio_put(bio);
336 }
337}
338
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800339static ssize_t
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700340__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800341{
342 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +0900343 struct inode *inode = bdev_file_inode(file);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700344 struct block_device *bdev = I_BDEV(inode);
Christoph Hellwig64d656a2016-12-22 19:20:45 +0100345 struct blk_plug plug;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700346 struct blkdev_dio *dio;
347 struct bio *bio;
Jens Axboecb700eb2018-11-15 19:56:53 -0700348 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
Christoph Hellwig690e5322017-01-24 14:50:19 +0100349 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700350 loff_t pos = iocb->ki_pos;
351 blk_qc_t qc = BLK_QC_T_NONE;
Christoph Hellwig36ffc6c2017-06-03 09:38:00 +0200352 int ret = 0;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700353
Jens Axboe9a794fb2016-11-22 08:12:39 -0700354 if ((pos | iov_iter_alignment(iter)) &
355 (bdev_logical_block_size(bdev) - 1))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700356 return -EINVAL;
357
Kent Overstreet52190f82018-05-20 18:25:55 -0400358 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700359
360 dio = container_of(bio, struct blkdev_dio, bio);
Christoph Hellwig690e5322017-01-24 14:50:19 +0100361 dio->is_sync = is_sync = is_sync_kiocb(iocb);
Christoph Hellwig531724a2018-11-30 09:23:48 +0100362 if (dio->is_sync) {
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700363 dio->waiter = current;
Christoph Hellwig531724a2018-11-30 09:23:48 +0100364 bio_get(bio);
365 } else {
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700366 dio->iocb = iocb;
Christoph Hellwig531724a2018-11-30 09:23:48 +0100367 }
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700368
369 dio->size = 0;
370 dio->multi_bio = false;
David Howells00e23702018-10-22 13:07:28 +0100371 dio->should_dirty = is_read && iter_is_iovec(iter);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700372
Jens Axboecb700eb2018-11-15 19:56:53 -0700373 /*
374 * Don't plug for HIPRI/polled IO, as those should go straight
375 * to issue
376 */
377 if (!is_poll)
378 blk_start_plug(&plug);
379
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700380 for (;;) {
Christoph Hellwig74d46992017-08-23 19:10:32 +0200381 bio_set_dev(bio, bdev);
Damien Le Moal4d1a4762016-11-22 15:38:49 +0900382 bio->bi_iter.bi_sector = pos >> 9;
Jens Axboe45d06cf2017-06-27 11:01:22 -0600383 bio->bi_write_hint = iocb->ki_hint;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700384 bio->bi_private = dio;
385 bio->bi_end_io = blkdev_bio_end_io;
Adam Manzanares074111c2018-05-22 10:52:20 -0700386 bio->bi_ioprio = iocb->ki_ioprio;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700387
388 ret = bio_iov_iter_get_pages(bio, iter);
389 if (unlikely(ret)) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200390 bio->bi_status = BLK_STS_IOERR;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700391 bio_endio(bio);
392 break;
393 }
394
395 if (is_read) {
396 bio->bi_opf = REQ_OP_READ;
397 if (dio->should_dirty)
398 bio_set_pages_dirty(bio);
399 } else {
400 bio->bi_opf = dio_bio_write_op(iocb);
401 task_io_account_write(bio->bi_iter.bi_size);
402 }
403
404 dio->size += bio->bi_iter.bi_size;
405 pos += bio->bi_iter.bi_size;
406
407 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
408 if (!nr_pages) {
Jens Axboed34513d2018-11-06 14:29:11 -0700409 if (iocb->ki_flags & IOCB_HIPRI)
410 bio->bi_opf |= REQ_HIPRI;
411
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700412 qc = submit_bio(bio);
413 break;
414 }
415
416 if (!dio->multi_bio) {
Christoph Hellwig531724a2018-11-30 09:23:48 +0100417 /*
418 * AIO needs an extra reference to ensure the dio
419 * structure which is embedded into the first bio
420 * stays around.
421 */
422 if (!is_sync)
423 bio_get(bio);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700424 dio->multi_bio = true;
425 atomic_set(&dio->ref, 2);
426 } else {
427 atomic_inc(&dio->ref);
428 }
429
430 submit_bio(bio);
431 bio = bio_alloc(GFP_KERNEL, nr_pages);
432 }
Jens Axboecb700eb2018-11-15 19:56:53 -0700433
434 if (!is_poll)
435 blk_finish_plug(&plug);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700436
Christoph Hellwig690e5322017-01-24 14:50:19 +0100437 if (!is_sync)
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700438 return -EIOCBQUEUED;
439
440 for (;;) {
Linus Torvalds1ac5cd42019-01-02 10:46:03 -0800441 set_current_state(TASK_UNINTERRUPTIBLE);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700442 if (!READ_ONCE(dio->waiter))
443 break;
444
445 if (!(iocb->ki_flags & IOCB_HIPRI) ||
Jens Axboe0a1b8b82018-11-26 08:24:43 -0700446 !blk_poll(bdev_get_queue(bdev), qc, true))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700447 io_schedule();
448 }
449 __set_current_state(TASK_RUNNING);
450
Christoph Hellwig36ffc6c2017-06-03 09:38:00 +0200451 if (!ret)
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200452 ret = blk_status_to_errno(dio->bio.bi_status);
Shaohua Li7a62a522016-12-13 19:07:55 -0800453 if (likely(!ret))
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700454 ret = dio->size;
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700455
456 bio_put(&dio->bio);
457 return ret;
458}
459
460static ssize_t
461blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
462{
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600463 int nr_pages;
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800464
Jens Axboe72ecad22016-11-16 23:11:42 -0700465 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600466 if (!nr_pages)
467 return 0;
Jens Axboe72ecad22016-11-16 23:11:42 -0700468 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
Christoph Hellwig189ce2b2016-10-31 11:59:25 -0600469 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700470
471 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800472}
473
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700474static __init int blkdev_init(void)
475{
Kent Overstreet52190f82018-05-20 18:25:55 -0400476 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
Christoph Hellwig542ff7b2016-11-16 23:14:22 -0700477}
478module_init(blkdev_init);
479
Jan Kara5cee5812009-04-27 16:43:51 +0200480int __sync_blockdev(struct block_device *bdev, int wait)
481{
482 if (!bdev)
483 return 0;
484 if (!wait)
485 return filemap_flush(bdev->bd_inode->i_mapping);
486 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
487}
488
Nick Piggin585d3bc2009-02-25 10:44:19 +0100489/*
490 * Write out and wait upon all the dirty data associated with a block
491 * device via its mapping. Does not take the superblock lock.
492 */
493int sync_blockdev(struct block_device *bdev)
494{
Jan Kara5cee5812009-04-27 16:43:51 +0200495 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100496}
497EXPORT_SYMBOL(sync_blockdev);
498
499/*
500 * Write out and wait upon all dirty data associated with this
501 * device. Filesystem data as well as the underlying block
502 * device. Takes the superblock lock.
503 */
504int fsync_bdev(struct block_device *bdev)
505{
506 struct super_block *sb = get_super(bdev);
507 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200508 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100509 drop_super(sb);
510 return res;
511 }
512 return sync_blockdev(bdev);
513}
Al Viro47e44912009-04-01 07:07:16 -0400514EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100515
516/**
517 * freeze_bdev -- lock a filesystem and force it into a consistent state
518 * @bdev: blockdevice to lock
519 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100520 * If a superblock is found on this device, we take the s_umount semaphore
521 * on it to make sure nobody unmounts until the snapshot creation is done.
522 * The reference counter (bd_fsfreeze_count) guarantees that only the last
523 * unfreeze process can unfreeze the frozen filesystem actually when multiple
524 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
525 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
526 * actually.
527 */
528struct super_block *freeze_bdev(struct block_device *bdev)
529{
530 struct super_block *sb;
531 int error = 0;
532
533 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200534 if (++bdev->bd_fsfreeze_count > 1) {
535 /*
536 * We don't even need to grab a reference - the first call
537 * to freeze_bdev grab an active reference and only the last
538 * thaw_bdev drops it.
539 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100540 sb = get_super(bdev);
Andrey Ryabinin5bb53c02016-08-23 18:55:31 +0300541 if (sb)
542 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100543 mutex_unlock(&bdev->bd_fsfreeze_mutex);
544 return sb;
545 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100546
Christoph Hellwig45042302009-08-03 23:28:35 +0200547 sb = get_active_super(bdev);
548 if (!sb)
549 goto out;
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600550 if (sb->s_op->freeze_super)
551 error = sb->s_op->freeze_super(sb);
552 else
553 error = freeze_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -0400554 if (error) {
555 deactivate_super(sb);
556 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200557 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400558 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100559 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400560 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200561 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100562 sync_blockdev(bdev);
563 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200564 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100565}
566EXPORT_SYMBOL(freeze_bdev);
567
568/**
569 * thaw_bdev -- unlock filesystem
570 * @bdev: blockdevice to unlock
571 * @sb: associated superblock
572 *
573 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
574 */
575int thaw_bdev(struct block_device *bdev, struct super_block *sb)
576{
Christoph Hellwig45042302009-08-03 23:28:35 +0200577 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100578
579 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200580 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400581 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100582
Christoph Hellwig45042302009-08-03 23:28:35 +0200583 error = 0;
584 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400585 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100586
Christoph Hellwig45042302009-08-03 23:28:35 +0200587 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400588 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200589
Benjamin Marzinski48b6bca2014-11-13 20:42:03 -0600590 if (sb->s_op->thaw_super)
591 error = sb->s_op->thaw_super(sb);
592 else
593 error = thaw_super(sb);
Pierre Morel997198b2016-10-04 10:53:40 +0200594 if (error)
Josef Bacik18e9e512010-03-23 10:34:56 -0400595 bdev->bd_fsfreeze_count++;
Josef Bacik18e9e512010-03-23 10:34:56 -0400596out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100597 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Pierre Morel997198b2016-10-04 10:53:40 +0200598 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100599}
600EXPORT_SYMBOL(thaw_bdev);
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
603{
604 return block_write_full_page(page, blkdev_get_block, wbc);
605}
606
607static int blkdev_readpage(struct file * file, struct page * page)
608{
609 return block_read_full_page(page, blkdev_get_block);
610}
611
Akinobu Mita447f05b2014-10-09 15:26:58 -0700612static int blkdev_readpages(struct file *file, struct address_space *mapping,
613 struct list_head *pages, unsigned nr_pages)
614{
615 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
616}
617
Nick Piggin6272b5a2007-10-16 01:25:04 -0700618static int blkdev_write_begin(struct file *file, struct address_space *mapping,
619 loff_t pos, unsigned len, unsigned flags,
620 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200622 return block_write_begin(mapping, pos, len, flags, pagep,
623 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624}
625
Nick Piggin6272b5a2007-10-16 01:25:04 -0700626static int blkdev_write_end(struct file *file, struct address_space *mapping,
627 loff_t pos, unsigned len, unsigned copied,
628 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700630 int ret;
631 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
632
633 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300634 put_page(page);
Nick Piggin6272b5a2007-10-16 01:25:04 -0700635
636 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638
639/*
640 * private llseek:
Al Viro496ad9a2013-01-23 17:07:38 -0500641 * for a block special file file_inode(file)->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 * so we compute the size by hand (just as in block_read/write above)
643 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800644static loff_t block_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900646 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 loff_t retval;
648
Al Viro59551022016-01-22 15:40:57 -0500649 inode_lock(bd_inode);
Al Viro5d48f3a2013-06-23 21:34:45 +0400650 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
Al Viro59551022016-01-22 15:40:57 -0500651 inode_unlock(bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 return retval;
653}
654
Josef Bacik02c24a82011-07-16 20:44:56 -0400655int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Dan Williams4ebb16c2015-10-28 07:48:19 +0900657 struct inode *bd_inode = bdev_file_inode(filp);
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400658 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100659 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200660
Jeff Layton372cf242017-07-06 07:02:28 -0400661 error = file_write_and_wait_range(filp, start, end);
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200662 if (error)
663 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100664
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400665 /*
666 * There is no need to serialise calls to blkdev_issue_flush with
667 * i_mutex and doing so causes performance issues with concurrent
668 * O_SYNC writers to a block device.
669 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200670 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100671 if (error == -EOPNOTSUPP)
672 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400673
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100674 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700676EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700678/**
679 * bdev_read_page() - Start reading a page from a block device
680 * @bdev: The device to read the page from
681 * @sector: The offset on the device to read the page to (need not be aligned)
682 * @page: The page to read
683 *
684 * On entry, the page should be locked. It will be unlocked when the page
685 * has been read. If the block driver implements rw_page synchronously,
686 * that will be true on exit from this function, but it need not be.
687 *
688 * Errors returned by this function are usually "soft", eg out of memory, or
689 * queue full; callers should try a different route to read this page rather
690 * than propagate an error back up the stack.
691 *
692 * Return: negative errno if an error occurs, 0 if submission was successful.
693 */
694int bdev_read_page(struct block_device *bdev, sector_t sector,
695 struct page *page)
696{
697 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc952015-11-19 13:29:28 -0800698 int result = -EOPNOTSUPP;
699
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400700 if (!ops->rw_page || bdev_get_integrity(bdev))
Dan Williams2e6edc952015-11-19 13:29:28 -0800701 return result;
702
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800703 result = blk_queue_enter(bdev->bd_queue, 0);
Dan Williams2e6edc952015-11-19 13:29:28 -0800704 if (result)
705 return result;
Tejun Heo3f289dc2018-07-18 04:47:36 -0700706 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
707 REQ_OP_READ);
Dan Williams2e6edc952015-11-19 13:29:28 -0800708 blk_queue_exit(bdev->bd_queue);
709 return result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700710}
711EXPORT_SYMBOL_GPL(bdev_read_page);
712
713/**
714 * bdev_write_page() - Start writing a page to a block device
715 * @bdev: The device to write the page to
716 * @sector: The offset on the device to write the page to (need not be aligned)
717 * @page: The page to write
718 * @wbc: The writeback_control for the write
719 *
720 * On entry, the page should be locked and not currently under writeback.
721 * On exit, if the write started successfully, the page will be unlocked and
722 * under writeback. If the write failed already (eg the driver failed to
723 * queue the page to the device), the page will still be locked. If the
724 * caller is a ->writepage implementation, it will need to unlock the page.
725 *
726 * Errors returned by this function are usually "soft", eg out of memory, or
727 * queue full; callers should try a different route to write this page rather
728 * than propagate an error back up the stack.
729 *
730 * Return: negative errno if an error occurs, 0 if submission was successful.
731 */
732int bdev_write_page(struct block_device *bdev, sector_t sector,
733 struct page *page, struct writeback_control *wbc)
734{
735 int result;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700736 const struct block_device_operations *ops = bdev->bd_disk->fops;
Dan Williams2e6edc952015-11-19 13:29:28 -0800737
Vishal Vermaf68eb1e2015-05-12 13:48:53 -0400738 if (!ops->rw_page || bdev_get_integrity(bdev))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700739 return -EOPNOTSUPP;
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800740 result = blk_queue_enter(bdev->bd_queue, 0);
Dan Williams2e6edc952015-11-19 13:29:28 -0800741 if (result)
742 return result;
743
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700744 set_page_writeback(page);
Tejun Heo3f289dc2018-07-18 04:47:36 -0700745 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
746 REQ_OP_WRITE);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700747 if (result) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700748 end_page_writeback(page);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700749 } else {
750 clean_page_buffers(page);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700751 unlock_page(page);
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700752 }
Dan Williams2e6edc952015-11-19 13:29:28 -0800753 blk_queue_exit(bdev->bd_queue);
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700754 return result;
755}
756EXPORT_SYMBOL_GPL(bdev_write_page);
757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758/*
759 * pseudo-fs
760 */
761
762static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800763static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765static struct inode *bdev_alloc_inode(struct super_block *sb)
766{
Christoph Lametere94b1762006-12-06 20:33:17 -0800767 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 if (!ei)
769 return NULL;
770 return &ei->vfs_inode;
771}
772
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100773static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100775 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 struct bdev_inode *bdi = BDEV_I(inode);
777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 kmem_cache_free(bdev_cachep, bdi);
779}
780
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100781static void bdev_destroy_inode(struct inode *inode)
782{
783 call_rcu(&inode->i_rcu, bdev_i_callback);
784}
785
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700786static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
788 struct bdev_inode *ei = (struct bdev_inode *) foo;
789 struct block_device *bdev = &ei->bdev;
790
Christoph Lametera35afb82007-05-16 22:10:57 -0700791 memset(bdev, 0, sizeof(*bdev));
792 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700793 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100794#ifdef CONFIG_SYSFS
795 INIT_LIST_HEAD(&bdev->bd_holder_disks);
796#endif
Jan Karaa5a79d02017-03-02 16:50:13 +0100797 bdev->bd_bdi = &noop_backing_dev_info;
Christoph Lametera35afb82007-05-16 22:10:57 -0700798 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800799 /* Initialize mutex for freeze. */
800 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801}
802
Al Virob57922d2010-06-07 14:34:48 -0400803static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804{
805 struct block_device *bdev = &BDEV_I(inode)->bdev;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700806 truncate_inode_pages_final(&inode->i_data);
Al Virob57922d2010-06-07 14:34:48 -0400807 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200808 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 spin_lock(&bdev_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 list_del_init(&bdev->bd_list);
811 spin_unlock(&bdev_lock);
Jan Karaf7597412017-03-23 01:37:00 +0100812 /* Detach inode from wb early as bdi_put() may free bdi->wb */
813 inode_detach_wb(inode);
Jan Karaa5a79d02017-03-02 16:50:13 +0100814 if (bdev->bd_bdi != &noop_backing_dev_info) {
Jan Karab1d2dc562017-02-02 15:56:52 +0100815 bdi_put(bdev->bd_bdi);
Jan Karaa5a79d02017-03-02 16:50:13 +0100816 bdev->bd_bdi = &noop_backing_dev_info;
817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818}
819
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800820static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 .statfs = simple_statfs,
822 .alloc_inode = bdev_alloc_inode,
823 .destroy_inode = bdev_destroy_inode,
824 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400825 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826};
827
Al Viro51139ad2010-07-25 23:47:46 +0400828static struct dentry *bd_mount(struct file_system_type *fs_type,
829 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
Shaohua Li3684aa72016-02-22 15:27:40 -0700831 struct dentry *dent;
832 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Vegard Nossume9e5e3f2016-08-22 12:47:43 +0200833 if (!IS_ERR(dent))
Shaohua Li3684aa72016-02-22 15:27:40 -0700834 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
835 return dent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836}
837
838static struct file_system_type bd_type = {
839 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400840 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 .kill_sb = kill_anon_super,
842};
843
Tejun Heoa212b102015-05-22 17:13:33 -0400844struct super_block *blockdev_superblock __read_mostly;
845EXPORT_SYMBOL_GPL(blockdev_superblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
847void __init bdev_cache_init(void)
848{
849 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300850 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800853 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
Vladimir Davydov5d097052016-01-14 15:18:21 -0800854 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900855 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 err = register_filesystem(&bd_type);
857 if (err)
858 panic("Cannot register bdev pseudo-fs");
859 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 if (IS_ERR(bd_mnt))
861 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300862 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863}
864
865/*
866 * Most likely _very_ bad one - but then it's hardly critical for small
867 * /dev and can be fixed when somebody will need really large one.
868 * Keep in mind that it will be fed through icache hash function too.
869 */
870static inline unsigned long hash(dev_t dev)
871{
872 return MAJOR(dev)+MINOR(dev);
873}
874
875static int bdev_test(struct inode *inode, void *data)
876{
877 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
878}
879
880static int bdev_set(struct inode *inode, void *data)
881{
882 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
883 return 0;
884}
885
886static LIST_HEAD(all_bdevs);
887
Jan Karaf44f1ab2017-02-02 15:56:49 +0100888/*
889 * If there is a bdev inode for this device, unhash it so that it gets evicted
890 * as soon as last inode reference is dropped.
891 */
892void bdev_unhash_inode(dev_t dev)
893{
894 struct inode *inode;
895
896 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
897 if (inode) {
898 remove_inode_hash(inode);
899 iput(inode);
900 }
901}
902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903struct block_device *bdget(dev_t dev)
904{
905 struct block_device *bdev;
906 struct inode *inode;
907
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800908 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 bdev_test, bdev_set, &dev);
910
911 if (!inode)
912 return NULL;
913
914 bdev = &BDEV_I(inode)->bdev;
915
916 if (inode->i_state & I_NEW) {
917 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000918 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 bdev->bd_inode = inode;
Fabian Frederick93407472017-02-27 14:28:32 -0800920 bdev->bd_block_size = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 bdev->bd_part_count = 0;
922 bdev->bd_invalidated = 0;
923 inode->i_mode = S_IFBLK;
924 inode->i_rdev = dev;
925 inode->i_bdev = bdev;
926 inode->i_data.a_ops = &def_blk_aops;
927 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 spin_lock(&bdev_lock);
929 list_add(&bdev->bd_list, &all_bdevs);
930 spin_unlock(&bdev_lock);
931 unlock_new_inode(inode);
932 }
933 return bdev;
934}
935
936EXPORT_SYMBOL(bdget);
937
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200938/**
939 * bdgrab -- Grab a reference to an already referenced block device
940 * @bdev: Block device to grab a reference to.
941 */
942struct block_device *bdgrab(struct block_device *bdev)
943{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400944 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200945 return bdev;
946}
Anatol Pomozovc1681bf2013-04-01 09:47:56 -0700947EXPORT_SYMBOL(bdgrab);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949long nr_blockdev_pages(void)
950{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700951 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 long ret = 0;
953 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700954 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 ret += bdev->bd_inode->i_mapping->nrpages;
956 }
957 spin_unlock(&bdev_lock);
958 return ret;
959}
960
961void bdput(struct block_device *bdev)
962{
963 iput(bdev->bd_inode);
964}
965
966EXPORT_SYMBOL(bdput);
967
968static struct block_device *bd_acquire(struct inode *inode)
969{
970 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 spin_lock(&bdev_lock);
973 bdev = inode->i_bdev;
Jan Karacccd9fb2017-02-21 18:09:48 +0100974 if (bdev && !inode_unhashed(bdev->bd_inode)) {
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +0100975 bdgrab(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 spin_unlock(&bdev_lock);
977 return bdev;
978 }
979 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700980
Jan Karacccd9fb2017-02-21 18:09:48 +0100981 /*
982 * i_bdev references block device inode that was already shut down
983 * (corresponding device got removed). Remove the reference and look
984 * up block device inode again just in case new device got
985 * reestablished under the same device number.
986 */
987 if (bdev)
988 bd_forget(inode);
989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 bdev = bdget(inode->i_rdev);
991 if (bdev) {
992 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700993 if (!inode->i_bdev) {
994 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400995 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700996 * and it's released in clear_inode() of inode.
997 * So, we can access it via ->i_mapping always
998 * without igrab().
999 */
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +01001000 bdgrab(bdev);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -07001001 inode->i_bdev = bdev;
1002 inode->i_mapping = bdev->bd_inode->i_mapping;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -07001003 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 spin_unlock(&bdev_lock);
1005 }
1006 return bdev;
1007}
1008
1009/* Call when you free inode */
1010
1011void bd_forget(struct inode *inode)
1012{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -07001013 struct block_device *bdev = NULL;
1014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 spin_lock(&bdev_lock);
Yan Hongb4ea2ea2013-04-30 15:26:47 -07001016 if (!sb_is_blkdev_sb(inode->i_sb))
1017 bdev = inode->i_bdev;
Al Viroa4a4f942016-07-19 13:16:52 -04001018 inode->i_bdev = NULL;
1019 inode->i_mapping = &inode->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -07001021
1022 if (bdev)
Ilya Dryomoved8a9d2c2015-11-20 22:18:43 +01001023 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024}
1025
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001026/**
1027 * bd_may_claim - test whether a block device can be claimed
1028 * @bdev: block device of interest
1029 * @whole: whole block device containing @bdev, may equal @bdev
1030 * @holder: holder trying to claim @bdev
1031 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001032 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001033 *
1034 * CONTEXT:
1035 * spin_lock(&bdev_lock).
1036 *
1037 * RETURNS:
1038 * %true if @bdev can be claimed, %false otherwise.
1039 */
1040static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1041 void *holder)
1042{
1043 if (bdev->bd_holder == holder)
1044 return true; /* already a holder */
1045 else if (bdev->bd_holder != NULL)
1046 return false; /* held by someone else */
NeilBrownbcc7f5b2016-12-12 08:21:51 -07001047 else if (whole == bdev)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001048 return true; /* is a whole device which isn't held */
1049
Tejun Heoe525fd82010-11-13 11:55:17 +01001050 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +09001051 return true; /* is a partition of a device that is being partitioned */
1052 else if (whole->bd_holder != NULL)
1053 return false; /* is a partition of a held device */
1054 else
1055 return true; /* is a partition of an un-held device */
1056}
1057
1058/**
Tejun Heo6b4517a2010-04-07 18:53:59 +09001059 * bd_prepare_to_claim - prepare to claim a block device
1060 * @bdev: block device of interest
1061 * @whole: the whole device containing @bdev, may equal @bdev
1062 * @holder: holder trying to claim @bdev
1063 *
1064 * Prepare to claim @bdev. This function fails if @bdev is already
1065 * claimed by another holder and waits if another claiming is in
1066 * progress. This function doesn't actually claim. On successful
1067 * return, the caller has ownership of bd_claiming and bd_holder[s].
1068 *
1069 * CONTEXT:
1070 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1071 * it multiple times.
1072 *
1073 * RETURNS:
1074 * 0 if @bdev can be claimed, -EBUSY otherwise.
1075 */
1076static int bd_prepare_to_claim(struct block_device *bdev,
1077 struct block_device *whole, void *holder)
1078{
1079retry:
1080 /* if someone else claimed, fail */
1081 if (!bd_may_claim(bdev, whole, holder))
1082 return -EBUSY;
1083
Tejun Heoe75aa852010-08-04 17:59:39 +02001084 /* if claiming is already in progress, wait for it to finish */
1085 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +09001086 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1087 DEFINE_WAIT(wait);
1088
1089 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1090 spin_unlock(&bdev_lock);
1091 schedule();
1092 finish_wait(wq, &wait);
1093 spin_lock(&bdev_lock);
1094 goto retry;
1095 }
1096
1097 /* yay, all mine */
1098 return 0;
1099}
1100
Jan Kara560e7cb2018-02-26 13:01:42 +01001101static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1102{
1103 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1104
1105 if (!disk)
1106 return NULL;
1107 /*
1108 * Now that we hold gendisk reference we make sure bdev we looked up is
1109 * not stale. If it is, it means device got removed and created before
1110 * we looked up gendisk and we fail open in such case. Associating
1111 * unhashed bdev with newly created gendisk could lead to two bdevs
1112 * (and thus two independent caches) being associated with one device
1113 * which is bad.
1114 */
1115 if (inode_unhashed(bdev->bd_inode)) {
1116 put_disk_and_module(disk);
1117 return NULL;
1118 }
1119 return disk;
1120}
1121
Tejun Heo6b4517a2010-04-07 18:53:59 +09001122/**
1123 * bd_start_claiming - start claiming a block device
1124 * @bdev: block device of interest
1125 * @holder: holder trying to claim @bdev
1126 *
1127 * @bdev is about to be opened exclusively. Check @bdev can be opened
1128 * exclusively and mark that an exclusive open is in progress. Each
1129 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +10001130 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1131 * fail).
1132 *
1133 * This function is used to gain exclusive access to the block device
1134 * without actually causing other exclusive open attempts to fail. It
1135 * should be used when the open sequence itself requires exclusive
1136 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +09001137 *
1138 * CONTEXT:
1139 * Might sleep.
1140 *
1141 * RETURNS:
1142 * Pointer to the block device containing @bdev on success, ERR_PTR()
1143 * value on failure.
1144 */
1145static struct block_device *bd_start_claiming(struct block_device *bdev,
1146 void *holder)
1147{
1148 struct gendisk *disk;
1149 struct block_device *whole;
1150 int partno, err;
1151
1152 might_sleep();
1153
1154 /*
1155 * @bdev might not have been initialized properly yet, look up
1156 * and grab the outer block device the hard way.
1157 */
Jan Kara560e7cb2018-02-26 13:01:42 +01001158 disk = bdev_get_gendisk(bdev, &partno);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001159 if (!disk)
1160 return ERR_PTR(-ENXIO);
1161
Tejun Heod4c208b2011-06-13 12:45:48 +02001162 /*
1163 * Normally, @bdev should equal what's returned from bdget_disk()
1164 * if partno is 0; however, some drivers (floppy) use multiple
1165 * bdev's for the same physical device and @bdev may be one of the
1166 * aliases. Keep @bdev if partno is 0. This means claimer
1167 * tracking is broken for those devices but it has always been that
1168 * way.
1169 */
1170 if (partno)
1171 whole = bdget_disk(disk, 0);
1172 else
1173 whole = bdgrab(bdev);
1174
Jan Kara9df6c292018-02-26 13:01:39 +01001175 put_disk_and_module(disk);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001176 if (!whole)
1177 return ERR_PTR(-ENOMEM);
1178
1179 /* prepare to claim, if successful, mark claiming in progress */
1180 spin_lock(&bdev_lock);
1181
1182 err = bd_prepare_to_claim(bdev, whole, holder);
1183 if (err == 0) {
1184 whole->bd_claiming = holder;
1185 spin_unlock(&bdev_lock);
1186 return whole;
1187 } else {
1188 spin_unlock(&bdev_lock);
1189 bdput(whole);
1190 return ERR_PTR(err);
1191 }
1192}
1193
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001194#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +01001195struct bd_holder_disk {
1196 struct list_head list;
1197 struct gendisk *disk;
1198 int refcnt;
1199};
1200
1201static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1202 struct gendisk *disk)
1203{
1204 struct bd_holder_disk *holder;
1205
1206 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1207 if (holder->disk == disk)
1208 return holder;
1209 return NULL;
1210}
1211
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -07001212static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001213{
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -07001214 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001215}
1216
1217static void del_symlink(struct kobject *from, struct kobject *to)
1218{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001219 sysfs_remove_link(from, kobject_name(to));
1220}
1221
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001222/**
Tejun Heoe09b4572010-11-13 11:55:17 +01001223 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1224 * @bdev: the claimed slave bdev
1225 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001226 *
Tejun Heo49731ba2011-01-14 18:43:57 +01001227 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1228 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001229 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001230 *
Tejun Heoe09b4572010-11-13 11:55:17 +01001231 * - from "slaves" directory of the holder @disk to the claimed @bdev
1232 * - from "holders" directory of the @bdev to the holder @disk
1233 *
1234 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1235 * passed to bd_link_disk_holder(), then:
1236 *
1237 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1238 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1239 *
1240 * The caller must have claimed @bdev before calling this function and
1241 * ensure that both @bdev and @disk are valid during the creation and
1242 * lifetime of these symlinks.
1243 *
1244 * CONTEXT:
1245 * Might sleep.
1246 *
1247 * RETURNS:
1248 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001249 */
Tejun Heoe09b4572010-11-13 11:55:17 +01001250int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001251{
Tejun Heo49731ba2011-01-14 18:43:57 +01001252 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001253 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001254
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001255 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001256
Tejun Heo49731ba2011-01-14 18:43:57 +01001257 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -07001258
Tejun Heoe09b4572010-11-13 11:55:17 +01001259 /* FIXME: remove the following once add_disk() handles errors */
1260 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1261 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -07001262
Tejun Heo49731ba2011-01-14 18:43:57 +01001263 holder = bd_find_holder_disk(bdev, disk);
1264 if (holder) {
1265 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +01001266 goto out_unlock;
1267 }
1268
Tejun Heo49731ba2011-01-14 18:43:57 +01001269 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1270 if (!holder) {
1271 ret = -ENOMEM;
1272 goto out_unlock;
1273 }
1274
1275 INIT_LIST_HEAD(&holder->list);
1276 holder->disk = disk;
1277 holder->refcnt = 1;
1278
1279 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1280 if (ret)
1281 goto out_free;
1282
1283 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1284 if (ret)
1285 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +01001286 /*
1287 * bdev could be deleted beneath us which would implicitly destroy
1288 * the holder directory. Hold on to it.
1289 */
1290 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001291
1292 list_add(&holder->list, &bdev->bd_holder_disks);
1293 goto out_unlock;
1294
1295out_del:
1296 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1297out_free:
1298 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +01001299out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001300 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +01001301 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001302}
Tejun Heoe09b4572010-11-13 11:55:17 +01001303EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001304
Tejun Heo49731ba2011-01-14 18:43:57 +01001305/**
1306 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1307 * @bdev: the calimed slave bdev
1308 * @disk: the holding disk
1309 *
1310 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1311 *
1312 * CONTEXT:
1313 * Might sleep.
1314 */
1315void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001316{
Tejun Heo49731ba2011-01-14 18:43:57 +01001317 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +01001318
Tejun Heo49731ba2011-01-14 18:43:57 +01001319 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001320
Tejun Heo49731ba2011-01-14 18:43:57 +01001321 holder = bd_find_holder_disk(bdev, disk);
1322
1323 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1324 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1325 del_symlink(bdev->bd_part->holder_dir,
1326 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +01001327 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +01001328 list_del_init(&holder->list);
1329 kfree(holder);
1330 }
1331
1332 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001333}
Tejun Heo49731ba2011-01-14 18:43:57 +01001334EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001335#endif
1336
Andrew Patterson0c002c22008-09-04 14:27:20 -06001337/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001338 * flush_disk - invalidates all buffer-cache entries on a disk
1339 *
1340 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -08001341 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -06001342 *
1343 * Invalidates all buffer-cache entries on a disk. It should be called
1344 * when a disk has been changed -- either by a media change or online
1345 * resize.
1346 */
NeilBrown93b270f2011-02-24 17:25:47 +11001347static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -06001348{
NeilBrown93b270f2011-02-24 17:25:47 +11001349 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -06001350 printk(KERN_WARNING "VFS: busy inodes on changed media or "
Dmitry Monakhov424081f2015-04-13 16:31:34 +04001351 "resized disk %s\n",
1352 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
Andrew Patterson56ade442008-09-04 14:27:40 -06001353 }
1354
1355 if (!bdev->bd_disk)
1356 return;
Tejun Heod27769e2011-08-23 20:01:04 +02001357 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -06001358 bdev->bd_invalidated = 1;
1359}
1360
1361/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001362 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001363 * @disk: struct gendisk to check
1364 * @bdev: struct bdev to adjust.
Christoph Hellwig5afb7832018-05-29 16:42:59 +02001365 * @verbose: if %true log a message about a size change if there is any
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001366 *
1367 * This routine checks to see if the bdev size does not match the disk size
shunki-fujita849cf552018-04-05 16:20:07 -07001368 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1369 * are freed.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001370 */
Christoph Hellwig5afb7832018-05-29 16:42:59 +02001371void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1372 bool verbose)
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001373{
1374 loff_t disk_size, bdev_size;
1375
1376 disk_size = (loff_t)get_capacity(disk) << 9;
1377 bdev_size = i_size_read(bdev->bd_inode);
1378 if (disk_size != bdev_size) {
Christoph Hellwig5afb7832018-05-29 16:42:59 +02001379 if (verbose) {
1380 printk(KERN_INFO
1381 "%s: detected capacity change from %lld to %lld\n",
1382 disk->disk_name, bdev_size, disk_size);
1383 }
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001384 i_size_write(bdev->bd_inode, disk_size);
shunki-fujita849cf552018-04-05 16:20:07 -07001385 if (bdev_size > disk_size)
1386 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001387 }
1388}
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001389
1390/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001391 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001392 * @disk: struct gendisk to be revalidated
1393 *
1394 * This routine is a wrapper for lower-level driver's revalidate_disk
1395 * call-backs. It is used to do common pre and post operations needed
1396 * for all revalidate_disk operations.
1397 */
1398int revalidate_disk(struct gendisk *disk)
1399{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001400 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001401 int ret = 0;
1402
1403 if (disk->fops->revalidate_disk)
1404 ret = disk->fops->revalidate_disk(disk);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001405 bdev = bdget_disk(disk, 0);
1406 if (!bdev)
1407 return ret;
1408
1409 mutex_lock(&bdev->bd_mutex);
Christoph Hellwig5afb7832018-05-29 16:42:59 +02001410 check_disk_size_change(disk, bdev, ret == 0);
MITSUNARI Shigeo7630b662013-02-21 16:42:01 -08001411 bdev->bd_invalidated = 0;
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001412 mutex_unlock(&bdev->bd_mutex);
1413 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001414 return ret;
1415}
1416EXPORT_SYMBOL(revalidate_disk);
1417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418/*
1419 * This routine checks whether a removable media has been changed,
1420 * and invalidates all buffer-cache-entries in that case. This
1421 * is a relatively slow routine, so we have to try to minimize using
1422 * it. Thus it is called only upon a 'mount' or 'open'. This
1423 * is the best way of combining speed and utility, I think.
1424 * People changing diskettes in the middle of an operation deserve
1425 * to lose :-)
1426 */
1427int check_disk_change(struct block_device *bdev)
1428{
1429 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001430 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001431 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Tejun Heo77ea8872010-12-08 20:57:37 +01001433 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1434 DISK_EVENT_EJECT_REQUEST);
1435 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 return 0;
1437
NeilBrown93b270f2011-02-24 17:25:47 +11001438 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 if (bdops->revalidate_disk)
1440 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 return 1;
1442}
1443
1444EXPORT_SYMBOL(check_disk_change);
1445
1446void bd_set_size(struct block_device *bdev, loff_t size)
1447{
Al Viro59551022016-01-22 15:40:57 -05001448 inode_lock(bdev->bd_inode);
Guo Chaod646a022013-02-21 15:16:42 -08001449 i_size_write(bdev->bd_inode, size);
Al Viro59551022016-01-22 15:40:57 -05001450 inode_unlock(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451}
1452EXPORT_SYMBOL(bd_set_size);
1453
Al Viro4385bab2013-05-05 22:11:03 -04001454static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001455
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001456/*
1457 * bd_mutex locking:
1458 *
1459 * mutex_lock(part->bd_mutex)
1460 * mutex_lock_nested(whole->bd_mutex, 1)
1461 */
1462
Al Viro572c4892007-10-08 13:24:05 -04001463static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001466 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001467 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001468 int perm = 0;
Jan Kara89736652018-02-26 13:01:40 +01001469 bool first_open = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
Al Viro572c4892007-10-08 13:24:05 -04001471 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001472 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001473 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001474 perm |= MAY_WRITE;
1475 /*
1476 * hooks: /n/, see "layering violations".
1477 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001478 if (!for_part) {
1479 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1480 if (ret != 0) {
1481 bdput(bdev);
1482 return ret;
1483 }
Al Viro82666022008-08-01 05:32:04 -04001484 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001485
NeilBrownd3374822009-01-09 08:31:10 +11001486 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001487
Tejun Heo89f97492008-11-05 10:21:06 +01001488 ret = -ENXIO;
Jan Kara560e7cb2018-02-26 13:01:42 +01001489 disk = bdev_get_gendisk(bdev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001490 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001491 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Tejun Heo69e02c52011-03-09 19:54:27 +01001493 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001494 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 if (!bdev->bd_openers) {
Jan Kara89736652018-02-26 13:01:40 +01001496 first_open = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001498 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 bdev->bd_contains = bdev;
Christoph Hellwigc2ee0702017-08-23 19:10:31 +02001500 bdev->bd_partno = partno;
Dan Williams03cdadb2016-02-26 15:19:43 -08001501
Tejun Heocf771cb2008-09-03 09:01:09 +02001502 if (!partno) {
Tejun Heo89f97492008-11-05 10:21:06 +01001503 ret = -ENXIO;
1504 bdev->bd_part = disk_get_part(disk, partno);
1505 if (!bdev->bd_part)
1506 goto out_clear;
1507
Tejun Heo1196f8b2011-04-21 20:54:45 +02001508 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001510 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001511 if (ret == -ERESTARTSYS) {
1512 /* Lost a race with 'disk' being
1513 * deleted, try again.
1514 * See md.c
1515 */
1516 disk_put_part(bdev->bd_part);
1517 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001518 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001519 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001520 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001521 disk_unblock_events(disk);
Jan Kara9df6c292018-02-26 13:01:39 +01001522 put_disk_and_module(disk);
NeilBrownd3374822009-01-09 08:31:10 +11001523 goto restart;
1524 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 }
Tejun Heo7e697232011-05-23 13:26:07 +02001526
Jan Kara04906b22019-01-14 09:48:10 +01001527 if (!ret) {
Tejun Heo7e697232011-05-23 13:26:07 +02001528 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
Jan Kara04906b22019-01-14 09:48:10 +01001529 set_init_blocksize(bdev);
1530 }
Tejun Heo7e697232011-05-23 13:26:07 +02001531
Tejun Heo1196f8b2011-04-21 20:54:45 +02001532 /*
1533 * If the device is invalidated, rescan partition
1534 * if open succeeded or failed with -ENOMEDIUM.
1535 * The latter is necessary to prevent ghost
1536 * partitions on a removed medium.
1537 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001538 if (bdev->bd_invalidated) {
1539 if (!ret)
1540 rescan_partitions(disk, bdev);
1541 else if (ret == -ENOMEDIUM)
1542 invalidate_partitions(disk, bdev);
1543 }
Dan Williams5a023cd2015-11-30 10:20:29 -08001544
Tejun Heo1196f8b2011-04-21 20:54:45 +02001545 if (ret)
1546 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 struct block_device *whole;
1549 whole = bdget_disk(disk, 0);
1550 ret = -ENOMEM;
1551 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001552 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001553 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001554 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001556 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 bdev->bd_contains = whole;
Tejun Heo89f97492008-11-05 10:21:06 +01001558 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001559 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001560 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001562 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 }
Tejun Heo89f97492008-11-05 10:21:06 +01001564 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Jan Kara04906b22019-01-14 09:48:10 +01001565 set_init_blocksize(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 }
Jan Kara03e26272017-03-23 01:36:53 +01001567
1568 if (bdev->bd_bdi == &noop_backing_dev_info)
1569 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001572 ret = 0;
1573 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001574 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001575 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001576 if (bdev->bd_invalidated) {
1577 if (!ret)
1578 rescan_partitions(bdev->bd_disk, bdev);
1579 else if (ret == -ENOMEDIUM)
1580 invalidate_partitions(bdev->bd_disk, bdev);
1581 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001582 if (ret)
1583 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 }
1585 }
1586 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001587 if (for_part)
1588 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001589 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001590 disk_unblock_events(disk);
Jan Kara89736652018-02-26 13:01:40 +01001591 /* only one opener holds refs to the module and disk */
1592 if (!first_open)
1593 put_disk_and_module(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 return 0;
1595
Tejun Heo0762b8b2008-08-25 19:56:12 +09001596 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001597 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001599 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001600 bdev->bd_queue = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001602 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001604 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001605 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001606 disk_unblock_events(disk);
Jan Kara9df6c292018-02-26 13:01:39 +01001607 put_disk_and_module(disk);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001608 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001609 bdput(bdev);
1610
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 return ret;
1612}
1613
Tejun Heod4d77622010-11-13 11:55:18 +01001614/**
1615 * blkdev_get - open a block device
1616 * @bdev: block_device to open
1617 * @mode: FMODE_* mask
1618 * @holder: exclusive holder identifier
1619 *
1620 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1621 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1622 * @holder is invalid. Exclusive opens may nest for the same @holder.
1623 *
1624 * On success, the reference count of @bdev is unchanged. On failure,
1625 * @bdev is put.
1626 *
1627 * CONTEXT:
1628 * Might sleep.
1629 *
1630 * RETURNS:
1631 * 0 on success, -errno on failure.
1632 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001633int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634{
Tejun Heoe525fd82010-11-13 11:55:17 +01001635 struct block_device *whole = NULL;
1636 int res;
1637
1638 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1639
1640 if ((mode & FMODE_EXCL) && holder) {
1641 whole = bd_start_claiming(bdev, holder);
1642 if (IS_ERR(whole)) {
1643 bdput(bdev);
1644 return PTR_ERR(whole);
1645 }
1646 }
1647
1648 res = __blkdev_get(bdev, mode, 0);
1649
1650 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001651 struct gendisk *disk = whole->bd_disk;
1652
Tejun Heo6a027ef2010-11-13 11:55:17 +01001653 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001654 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001655 spin_lock(&bdev_lock);
1656
Tejun Heo77ea8872010-12-08 20:57:37 +01001657 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001658 BUG_ON(!bd_may_claim(bdev, whole, holder));
1659 /*
1660 * Note that for a whole device bd_holders
1661 * will be incremented twice, and bd_holder
1662 * will be set to bd_may_claim before being
1663 * set to holder
1664 */
1665 whole->bd_holders++;
1666 whole->bd_holder = bd_may_claim;
1667 bdev->bd_holders++;
1668 bdev->bd_holder = holder;
1669 }
1670
1671 /* tell others that we're done */
1672 BUG_ON(whole->bd_claiming != holder);
1673 whole->bd_claiming = NULL;
1674 wake_up_bit(&whole->bd_claiming, 0);
1675
1676 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001677
1678 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001679 * Block event polling for write claims if requested. Any
1680 * write holder makes the write_holder state stick until
1681 * all are released. This is good enough and tracking
1682 * individual writeable reference is too fragile given the
1683 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001684 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001685 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1686 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001687 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001688 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001689 }
1690
1691 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001692 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001693 }
1694
1695 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001696}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697EXPORT_SYMBOL(blkdev_get);
1698
Tejun Heod4d77622010-11-13 11:55:18 +01001699/**
1700 * blkdev_get_by_path - open a block device by name
1701 * @path: path to the block device to open
1702 * @mode: FMODE_* mask
1703 * @holder: exclusive holder identifier
1704 *
1705 * Open the blockdevice described by the device file at @path. @mode
1706 * and @holder are identical to blkdev_get().
1707 *
1708 * On success, the returned block_device has reference count of one.
1709 *
1710 * CONTEXT:
1711 * Might sleep.
1712 *
1713 * RETURNS:
1714 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1715 */
1716struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1717 void *holder)
1718{
1719 struct block_device *bdev;
1720 int err;
1721
1722 bdev = lookup_bdev(path);
1723 if (IS_ERR(bdev))
1724 return bdev;
1725
1726 err = blkdev_get(bdev, mode, holder);
1727 if (err)
1728 return ERR_PTR(err);
1729
Chuck Ebberte51900f2011-02-16 18:11:53 -05001730 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1731 blkdev_put(bdev, mode);
1732 return ERR_PTR(-EACCES);
1733 }
1734
Tejun Heod4d77622010-11-13 11:55:18 +01001735 return bdev;
1736}
1737EXPORT_SYMBOL(blkdev_get_by_path);
1738
1739/**
1740 * blkdev_get_by_dev - open a block device by device number
1741 * @dev: device number of block device to open
1742 * @mode: FMODE_* mask
1743 * @holder: exclusive holder identifier
1744 *
1745 * Open the blockdevice described by device number @dev. @mode and
1746 * @holder are identical to blkdev_get().
1747 *
1748 * Use it ONLY if you really do not have anything better - i.e. when
1749 * you are behind a truly sucky interface and all you are given is a
1750 * device number. _Never_ to be used for internal purposes. If you
1751 * ever need it - reconsider your API.
1752 *
1753 * On success, the returned block_device has reference count of one.
1754 *
1755 * CONTEXT:
1756 * Might sleep.
1757 *
1758 * RETURNS:
1759 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1760 */
1761struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1762{
1763 struct block_device *bdev;
1764 int err;
1765
1766 bdev = bdget(dev);
1767 if (!bdev)
1768 return ERR_PTR(-ENOMEM);
1769
1770 err = blkdev_get(bdev, mode, holder);
1771 if (err)
1772 return ERR_PTR(err);
1773
1774 return bdev;
1775}
1776EXPORT_SYMBOL(blkdev_get_by_dev);
1777
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778static int blkdev_open(struct inode * inode, struct file * filp)
1779{
1780 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
1782 /*
1783 * Preserve backwards compatibility and allow large file access
1784 * even if userspace doesn't ask for it explicitly. Some mkfs
1785 * binary needs it. We might want to drop this workaround
1786 * during an unstable branch.
1787 */
1788 filp->f_flags |= O_LARGEFILE;
1789
Christoph Hellwigc35fc7a2017-08-29 16:13:21 +02001790 filp->f_mode |= FMODE_NOWAIT;
1791
Al Viro572c4892007-10-08 13:24:05 -04001792 if (filp->f_flags & O_NDELAY)
1793 filp->f_mode |= FMODE_NDELAY;
1794 if (filp->f_flags & O_EXCL)
1795 filp->f_mode |= FMODE_EXCL;
1796 if ((filp->f_flags & O_ACCMODE) == 3)
1797 filp->f_mode |= FMODE_WRITE_IOCTL;
1798
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001800 if (bdev == NULL)
1801 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
Al Viro572c4892007-10-08 13:24:05 -04001803 filp->f_mapping = bdev->bd_inode->i_mapping;
Jeff Layton5660e132017-07-06 07:02:25 -04001804 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
Al Viro572c4892007-10-08 13:24:05 -04001805
Tejun Heoe525fd82010-11-13 11:55:17 +01001806 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807}
1808
Al Viro4385bab2013-05-05 22:11:03 -04001809static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001810{
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001811 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001812 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001813
NeilBrown6796bf52006-12-08 02:36:16 -08001814 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001815 if (for_part)
1816 bdev->bd_part_count--;
1817
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001818 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001819 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001820 sync_blockdev(bdev);
1821 kill_bdev(bdev);
Ilya Dryomov43d1c0e2015-11-20 22:22:34 +01001822
Vivek Goyaldbd3ca52015-11-09 09:23:40 -07001823 bdev_write_inode(bdev);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001824 }
1825 if (bdev->bd_contains == bdev) {
1826 if (disk->fops->release)
Al Virodb2a1442013-05-05 21:52:57 -04001827 disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001828 }
1829 if (!bdev->bd_openers) {
Tejun Heo0762b8b2008-08-25 19:56:12 +09001830 disk_put_part(bdev->bd_part);
1831 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001832 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001833 if (bdev != bdev->bd_contains)
1834 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001835 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001836
Jan Kara9df6c292018-02-26 13:01:39 +01001837 put_disk_and_module(disk);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001838 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001839 mutex_unlock(&bdev->bd_mutex);
1840 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001841 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001842 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001843}
1844
Al Viro4385bab2013-05-05 22:11:03 -04001845void blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001846{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001847 mutex_lock(&bdev->bd_mutex);
1848
Tejun Heoe525fd82010-11-13 11:55:17 +01001849 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001850 bool bdev_free;
1851
1852 /*
1853 * Release a claim on the device. The holder fields
1854 * are protected with bdev_lock. bd_mutex is to
1855 * synchronize disk_holder unlinking.
1856 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001857 spin_lock(&bdev_lock);
1858
1859 WARN_ON_ONCE(--bdev->bd_holders < 0);
1860 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1861
1862 /* bd_contains might point to self, check in a separate step */
1863 if ((bdev_free = !bdev->bd_holders))
1864 bdev->bd_holder = NULL;
1865 if (!bdev->bd_contains->bd_holders)
1866 bdev->bd_contains->bd_holder = NULL;
1867
1868 spin_unlock(&bdev_lock);
1869
Tejun Heo77ea8872010-12-08 20:57:37 +01001870 /*
1871 * If this was the last claim, remove holder link and
1872 * unblock evpoll if it was a write holder.
1873 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001874 if (bdev_free && bdev->bd_write_holder) {
1875 disk_unblock_events(bdev->bd_disk);
1876 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001877 }
Tejun Heo69362172011-03-09 19:54:27 +01001878 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001879
Tejun Heo85ef06d2011-07-01 16:17:47 +02001880 /*
1881 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1882 * event. This is to ensure detection of media removal commanded
1883 * from userland - e.g. eject(1).
1884 */
1885 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1886
1887 mutex_unlock(&bdev->bd_mutex);
1888
Al Viro4385bab2013-05-05 22:11:03 -04001889 __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001890}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001891EXPORT_SYMBOL(blkdev_put);
1892
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893static int blkdev_close(struct inode * inode, struct file * filp)
1894{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001895 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
Al Viro4385bab2013-05-05 22:11:03 -04001896 blkdev_put(bdev, filp->f_mode);
1897 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898}
1899
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001900static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901{
Dan Williams4ebb16c2015-10-28 07:48:19 +09001902 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Al Viro56b26ad2008-09-19 03:17:36 -04001903 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001904
1905 /*
1906 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1907 * to updated it before every ioctl.
1908 */
Al Viro56b26ad2008-09-19 03:17:36 -04001909 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001910 mode |= FMODE_NDELAY;
1911 else
1912 mode &= ~FMODE_NDELAY;
1913
Al Viro56b26ad2008-09-19 03:17:36 -04001914 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915}
1916
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001917/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001918 * Write data to the block device. Only intended for the block device itself
1919 * and the raw driver which basically is a fake block device.
1920 *
1921 * Does not take i_mutex for the write and thus is not for general purpose
1922 * use.
1923 */
Al Viro1456c0a2014-04-03 03:21:50 -04001924ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
Christoph Hellwigeef99382009-08-20 17:43:41 +02001925{
1926 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001927 struct inode *bd_inode = bdev_file_inode(file);
Al Viro7ec7b942015-04-07 11:35:14 -04001928 loff_t size = i_size_read(bd_inode);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001929 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001930 ssize_t ret;
Al Viro5f380c72015-04-07 11:28:12 -04001931
Al Viro7ec7b942015-04-07 11:35:14 -04001932 if (bdev_read_only(I_BDEV(bd_inode)))
1933 return -EPERM;
Al Viro5f380c72015-04-07 11:28:12 -04001934
Al Viro7ec7b942015-04-07 11:35:14 -04001935 if (!iov_iter_count(from))
Al Viro5f380c72015-04-07 11:28:12 -04001936 return 0;
1937
Al Viro7ec7b942015-04-07 11:35:14 -04001938 if (iocb->ki_pos >= size)
1939 return -ENOSPC;
1940
Christoph Hellwigc35fc7a2017-08-29 16:13:21 +02001941 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1942 return -EOPNOTSUPP;
1943
Al Viro7ec7b942015-04-07 11:35:14 -04001944 iov_iter_truncate(from, size - iocb->ki_pos);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001945
Jianpeng Ma53362a02012-08-02 09:50:39 +02001946 blk_start_plug(&plug);
Al Viro1456c0a2014-04-03 03:21:50 -04001947 ret = __generic_file_write_iter(iocb, from);
Christoph Hellwige2592212016-04-07 08:52:01 -07001948 if (ret > 0)
1949 ret = generic_write_sync(iocb, ret);
Jianpeng Ma53362a02012-08-02 09:50:39 +02001950 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001951 return ret;
1952}
Al Viro1456c0a2014-04-03 03:21:50 -04001953EXPORT_SYMBOL_GPL(blkdev_write_iter);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001954
David Jefferyb2de5252014-09-29 10:21:10 -04001955ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001956{
1957 struct file *file = iocb->ki_filp;
Dan Williams4ebb16c2015-10-28 07:48:19 +09001958 struct inode *bd_inode = bdev_file_inode(file);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001959 loff_t size = i_size_read(bd_inode);
Al Viroa8860382014-04-02 20:02:21 -04001960 loff_t pos = iocb->ki_pos;
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001961
1962 if (pos >= size)
1963 return 0;
1964
1965 size -= pos;
Al Viroa8860382014-04-02 20:02:21 -04001966 iov_iter_truncate(to, size);
1967 return generic_file_read_iter(iocb, to);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001968}
David Jefferyb2de5252014-09-29 10:21:10 -04001969EXPORT_SYMBOL_GPL(blkdev_read_iter);
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001970
Christoph Hellwigeef99382009-08-20 17:43:41 +02001971/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001972 * Try to release a page associated with block device when the system
1973 * is under memory pressure.
1974 */
1975static int blkdev_releasepage(struct page *page, gfp_t wait)
1976{
1977 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1978
1979 if (super && super->s_op->bdev_try_to_free_page)
1980 return super->s_op->bdev_try_to_free_page(super, page, wait);
1981
1982 return try_to_free_buffers(page);
1983}
1984
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001985static int blkdev_writepages(struct address_space *mapping,
1986 struct writeback_control *wbc)
1987{
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001988 return generic_writepages(mapping, wbc);
1989}
1990
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001991static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 .readpage = blkdev_readpage,
Akinobu Mita447f05b2014-10-09 15:26:58 -07001993 .readpages = blkdev_readpages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001995 .write_begin = blkdev_write_begin,
1996 .write_end = blkdev_write_end,
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001997 .writepages = blkdev_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001998 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 .direct_IO = blkdev_direct_IO,
Jan Kara88dbcbb2018-12-28 00:39:16 -08002000 .migratepage = buffer_migrate_page_norefs,
Mel Gormanb4597222013-07-03 15:02:05 -07002001 .is_dirty_writeback = buffer_check_dirty_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002};
2003
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002004#define BLKDEV_FALLOC_FL_SUPPORTED \
2005 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2006 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2007
2008static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2009 loff_t len)
2010{
2011 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002012 struct address_space *mapping;
2013 loff_t end = start + len - 1;
2014 loff_t isize;
2015 int error;
2016
2017 /* Fail if we don't recognize the flags. */
2018 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2019 return -EOPNOTSUPP;
2020
2021 /* Don't go off the end of the device. */
2022 isize = i_size_read(bdev->bd_inode);
2023 if (start >= isize)
2024 return -EINVAL;
2025 if (end >= isize) {
2026 if (mode & FALLOC_FL_KEEP_SIZE) {
2027 len = isize - start;
2028 end = start + len - 1;
2029 } else
2030 return -EINVAL;
2031 }
2032
2033 /*
2034 * Don't allow IO that isn't aligned to logical block size.
2035 */
2036 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2037 return -EINVAL;
2038
2039 /* Invalidate the page cache, including dirty pages. */
2040 mapping = bdev->bd_inode->i_mapping;
2041 truncate_inode_pages_range(mapping, start, end);
2042
2043 switch (mode) {
2044 case FALLOC_FL_ZERO_RANGE:
2045 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2046 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
Christoph Hellwigee472d82017-04-05 19:21:08 +02002047 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002048 break;
2049 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
Christoph Hellwig34045122017-04-05 19:21:11 +02002050 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2051 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002052 break;
2053 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002054 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2055 GFP_KERNEL, 0);
2056 break;
2057 default:
2058 return -EOPNOTSUPP;
2059 }
2060 if (error)
2061 return error;
2062
2063 /*
2064 * Invalidate again; if someone wandered in and dirtied a page,
2065 * the caller will be given -EBUSY. The third argument is
2066 * inclusive, so the rounding here is safe.
2067 */
2068 return invalidate_inode_pages2_range(mapping,
2069 start >> PAGE_SHIFT,
2070 end >> PAGE_SHIFT);
2071}
2072
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002073const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 .open = blkdev_open,
2075 .release = blkdev_close,
2076 .llseek = block_llseek,
Al Viroa8860382014-04-02 20:02:21 -04002077 .read_iter = blkdev_read_iter,
Al Viro1456c0a2014-04-03 03:21:50 -04002078 .write_iter = blkdev_write_iter,
Dan Williamsacc93d32016-05-07 11:40:28 -07002079 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07002080 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07002081 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082#ifdef CONFIG_COMPAT
2083 .compat_ioctl = compat_blkdev_ioctl,
2084#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08002085 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04002086 .splice_write = iter_file_splice_write,
Darrick J. Wong25f4c412016-10-11 13:51:11 -07002087 .fallocate = blkdev_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088};
2089
2090int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2091{
2092 int res;
2093 mm_segment_t old_fs = get_fs();
2094 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04002095 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 set_fs(old_fs);
2097 return res;
2098}
2099
2100EXPORT_SYMBOL(ioctl_by_bdev);
2101
2102/**
2103 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08002104 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02002106 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 * namespace if possible and return it. Return ERR_PTR(error)
2108 * otherwise.
2109 */
Al Viro421748e2008-08-02 01:04:36 -04002110struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111{
2112 struct block_device *bdev;
2113 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04002114 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 int error;
2116
Al Viro421748e2008-08-02 01:04:36 -04002117 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 return ERR_PTR(-EINVAL);
2119
Al Viro421748e2008-08-02 01:04:36 -04002120 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 if (error)
2122 return ERR_PTR(error);
2123
David Howellsbb6687342015-03-17 22:26:21 +00002124 inode = d_backing_inode(path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 error = -ENOTBLK;
2126 if (!S_ISBLK(inode->i_mode))
2127 goto fail;
2128 error = -EACCES;
Eric W. Biedermana2982cc2016-06-09 15:34:02 -05002129 if (!may_open_dev(&path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 goto fail;
2131 error = -ENOMEM;
2132 bdev = bd_acquire(inode);
2133 if (!bdev)
2134 goto fail;
2135out:
Al Viro421748e2008-08-02 01:04:36 -04002136 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 return bdev;
2138fail:
2139 bdev = ERR_PTR(error);
2140 goto out;
2141}
Al Virod5686b42008-08-01 05:00:11 -04002142EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143
NeilBrown93b270f2011-02-24 17:25:47 +11002144int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01002145{
2146 struct super_block *sb = get_super(bdev);
2147 int res = 0;
2148
2149 if (sb) {
2150 /*
2151 * no need to lock the super, get_super holds the
2152 * read mutex so the filesystem cannot go away
2153 * under us (->put_super runs with the write lock
2154 * hold).
2155 */
2156 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11002157 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01002158 drop_super(sb);
2159 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07002160 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01002161 return res;
2162}
2163EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002164
2165void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2166{
2167 struct inode *inode, *old_inode = NULL;
2168
Dave Chinner74278da2015-03-04 12:37:22 -05002169 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002170 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2171 struct address_space *mapping = inode->i_mapping;
Rabin Vincentaf309222016-12-01 09:18:28 +01002172 struct block_device *bdev;
Jan Kara5c0d6b62012-07-03 16:45:31 +02002173
2174 spin_lock(&inode->i_lock);
2175 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2176 mapping->nrpages == 0) {
2177 spin_unlock(&inode->i_lock);
2178 continue;
2179 }
2180 __iget(inode);
2181 spin_unlock(&inode->i_lock);
Dave Chinner74278da2015-03-04 12:37:22 -05002182 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002183 /*
2184 * We hold a reference to 'inode' so it couldn't have been
2185 * removed from s_inodes list while we dropped the
Dave Chinner74278da2015-03-04 12:37:22 -05002186 * s_inode_list_lock We cannot iput the inode now as we can
Jan Kara5c0d6b62012-07-03 16:45:31 +02002187 * be holding the last reference and we cannot iput it under
Dave Chinner74278da2015-03-04 12:37:22 -05002188 * s_inode_list_lock. So we keep the reference and iput it
Jan Kara5c0d6b62012-07-03 16:45:31 +02002189 * later.
2190 */
2191 iput(old_inode);
2192 old_inode = inode;
Rabin Vincentaf309222016-12-01 09:18:28 +01002193 bdev = I_BDEV(inode);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002194
Rabin Vincentaf309222016-12-01 09:18:28 +01002195 mutex_lock(&bdev->bd_mutex);
2196 if (bdev->bd_openers)
2197 func(bdev, arg);
2198 mutex_unlock(&bdev->bd_mutex);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002199
Dave Chinner74278da2015-03-04 12:37:22 -05002200 spin_lock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002201 }
Dave Chinner74278da2015-03-04 12:37:22 -05002202 spin_unlock(&blockdev_superblock->s_inode_list_lock);
Jan Kara5c0d6b62012-07-03 16:45:31 +02002203 iput(old_inode);
2204}