Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | #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 Emelyanov | 7db9cfd | 2008-06-05 22:46:27 -0700 | [diff] [blame] | 14 | #include <linux/device_cgroup.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 15 | #include <linux/highmem.h> |
| 16 | #include <linux/blkdev.h> |
Tejun Heo | 66114ca | 2015-05-22 17:13:32 -0400 | [diff] [blame] | 17 | #include <linux/backing-dev.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/module.h> |
| 19 | #include <linux/blkpg.h> |
Muthu Kumar | b502bd1 | 2012-03-23 15:01:50 -0700 | [diff] [blame] | 20 | #include <linux/magic.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | #include <linux/buffer_head.h> |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 22 | #include <linux/swap.h> |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 23 | #include <linux/pagevec.h> |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 24 | #include <linux/writeback.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/mpage.h> |
| 26 | #include <linux/mount.h> |
| 27 | #include <linux/uio.h> |
| 28 | #include <linux/namei.h> |
Vignesh Babu BM | 1368c4f | 2007-05-08 00:24:32 -0700 | [diff] [blame] | 29 | #include <linux/log2.h> |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 30 | #include <linux/cleancache.h> |
Matthew Wilcox | c94c2ac | 2015-09-08 14:58:40 -0700 | [diff] [blame] | 31 | #include <linux/dax.h> |
Dan Williams | acc93d3 | 2016-05-07 11:40:28 -0700 | [diff] [blame] | 32 | #include <linux/badblocks.h> |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 33 | #include <linux/task_io_accounting_ops.h> |
Darrick J. Wong | 25f4c41 | 2016-10-11 13:51:11 -0700 | [diff] [blame] | 34 | #include <linux/falloc.h> |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 35 | #include <linux/uaccess.h> |
David Howells | 07f3f05c | 2006-09-30 20:52:18 +0200 | [diff] [blame] | 36 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | |
| 38 | struct bdev_inode { |
| 39 | struct block_device bdev; |
| 40 | struct inode vfs_inode; |
| 41 | }; |
| 42 | |
Adrian Bunk | 4c54ac6 | 2008-02-18 13:48:31 +0100 | [diff] [blame] | 43 | static const struct address_space_operations def_blk_aops; |
| 44 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
| 46 | { |
| 47 | return container_of(inode, struct bdev_inode, vfs_inode); |
| 48 | } |
| 49 | |
Geert Uytterhoeven | ff5053f | 2015-06-26 13:58:32 +0200 | [diff] [blame] | 50 | struct block_device *I_BDEV(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | { |
| 52 | return &BDEV_I(inode)->bdev; |
| 53 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | EXPORT_SYMBOL(I_BDEV); |
| 55 | |
Toshi Kani | 2af3a81 | 2016-05-10 10:23:52 -0600 | [diff] [blame] | 56 | void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...) |
| 57 | { |
| 58 | struct va_format vaf; |
| 59 | va_list args; |
| 60 | |
| 61 | va_start(args, fmt); |
| 62 | vaf.fmt = fmt; |
| 63 | vaf.va = &args; |
| 64 | printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf); |
| 65 | va_end(args); |
| 66 | } |
| 67 | |
Vivek Goyal | dbd3ca5 | 2015-11-09 09:23:40 -0700 | [diff] [blame] | 68 | static void bdev_write_inode(struct block_device *bdev) |
Christoph Hellwig | 564f00f | 2015-01-14 10:42:33 +0100 | [diff] [blame] | 69 | { |
Vivek Goyal | dbd3ca5 | 2015-11-09 09:23:40 -0700 | [diff] [blame] | 70 | struct inode *inode = bdev->bd_inode; |
| 71 | int ret; |
| 72 | |
Christoph Hellwig | 564f00f | 2015-01-14 10:42:33 +0100 | [diff] [blame] | 73 | spin_lock(&inode->i_lock); |
| 74 | while (inode->i_state & I_DIRTY) { |
| 75 | spin_unlock(&inode->i_lock); |
Vivek Goyal | dbd3ca5 | 2015-11-09 09:23:40 -0700 | [diff] [blame] | 76 | ret = write_inode_now(inode, true); |
| 77 | if (ret) { |
| 78 | char name[BDEVNAME_SIZE]; |
| 79 | pr_warn_ratelimited("VFS: Dirty inode writeback failed " |
| 80 | "for block device %s (err=%d).\n", |
| 81 | bdevname(bdev, name), ret); |
| 82 | } |
Christoph Hellwig | 564f00f | 2015-01-14 10:42:33 +0100 | [diff] [blame] | 83 | spin_lock(&inode->i_lock); |
| 84 | } |
| 85 | spin_unlock(&inode->i_lock); |
| 86 | } |
| 87 | |
Peter Zijlstra | f9a1439 | 2007-05-06 14:49:55 -0700 | [diff] [blame] | 88 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 89 | void kill_bdev(struct block_device *bdev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | { |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 91 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
| 92 | |
Ross Zwisler | f9fe48b | 2016-01-22 15:10:40 -0800 | [diff] [blame] | 93 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
Peter Zijlstra | f9a1439 | 2007-05-06 14:49:55 -0700 | [diff] [blame] | 94 | return; |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 95 | |
Peter Zijlstra | f9a1439 | 2007-05-06 14:49:55 -0700 | [diff] [blame] | 96 | invalidate_bh_lrus(); |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 97 | truncate_inode_pages(mapping, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | } |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 99 | EXPORT_SYMBOL(kill_bdev); |
| 100 | |
| 101 | /* Invalidate clean unused buffers and pagecache. */ |
| 102 | void invalidate_bdev(struct block_device *bdev) |
| 103 | { |
| 104 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
| 105 | |
| 106 | if (mapping->nrpages == 0) |
| 107 | return; |
| 108 | |
| 109 | invalidate_bh_lrus(); |
| 110 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ |
| 111 | invalidate_mapping_pages(mapping, 0, -1); |
| 112 | /* 99% of the time, we don't need to flush the cleancache on the bdev. |
| 113 | * But, for the strange corners, lets be cautious |
| 114 | */ |
Dan Magenheimer | 3167760 | 2011-09-21 11:56:28 -0400 | [diff] [blame] | 115 | cleancache_invalidate_inode(mapping); |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 116 | } |
| 117 | EXPORT_SYMBOL(invalidate_bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | |
| 119 | int set_blocksize(struct block_device *bdev, int size) |
| 120 | { |
| 121 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ |
Vignesh Babu BM | 1368c4f | 2007-05-08 00:24:32 -0700 | [diff] [blame] | 122 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | return -EINVAL; |
| 124 | |
| 125 | /* Size cannot be smaller than the size supported by the device */ |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 126 | if (size < bdev_logical_block_size(bdev)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | return -EINVAL; |
| 128 | |
| 129 | /* Don't change the size if it is same as current */ |
| 130 | if (bdev->bd_block_size != size) { |
| 131 | sync_blockdev(bdev); |
| 132 | bdev->bd_block_size = size; |
| 133 | bdev->bd_inode->i_blkbits = blksize_bits(size); |
| 134 | kill_bdev(bdev); |
| 135 | } |
| 136 | return 0; |
| 137 | } |
| 138 | |
| 139 | EXPORT_SYMBOL(set_blocksize); |
| 140 | |
| 141 | int sb_set_blocksize(struct super_block *sb, int size) |
| 142 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 143 | if (set_blocksize(sb->s_bdev, size)) |
| 144 | return 0; |
| 145 | /* If we get here, we know size is power of two |
| 146 | * and it's value is between 512 and PAGE_SIZE */ |
| 147 | sb->s_blocksize = size; |
Coywolf Qi Hunt | 38885bd | 2006-03-24 03:18:05 -0800 | [diff] [blame] | 148 | sb->s_blocksize_bits = blksize_bits(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | return sb->s_blocksize; |
| 150 | } |
| 151 | |
| 152 | EXPORT_SYMBOL(sb_set_blocksize); |
| 153 | |
| 154 | int sb_min_blocksize(struct super_block *sb, int size) |
| 155 | { |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 156 | int minsize = bdev_logical_block_size(sb->s_bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | if (size < minsize) |
| 158 | size = minsize; |
| 159 | return sb_set_blocksize(sb, size); |
| 160 | } |
| 161 | |
| 162 | EXPORT_SYMBOL(sb_min_blocksize); |
| 163 | |
| 164 | static int |
| 165 | blkdev_get_block(struct inode *inode, sector_t iblock, |
| 166 | struct buffer_head *bh, int create) |
| 167 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | bh->b_bdev = I_BDEV(inode); |
| 169 | bh->b_blocknr = iblock; |
| 170 | set_buffer_mapped(bh); |
| 171 | return 0; |
| 172 | } |
| 173 | |
Dan Williams | 4ebb16c | 2015-10-28 07:48:19 +0900 | [diff] [blame] | 174 | static struct inode *bdev_file_inode(struct file *file) |
| 175 | { |
| 176 | return file->f_mapping->host; |
| 177 | } |
| 178 | |
Jens Axboe | 78250c0 | 2016-11-17 17:50:47 +0100 | [diff] [blame] | 179 | static unsigned int dio_bio_write_op(struct kiocb *iocb) |
| 180 | { |
| 181 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; |
| 182 | |
| 183 | /* avoid the need for a I/O completion work item */ |
| 184 | if (iocb->ki_flags & IOCB_DSYNC) |
| 185 | op |= REQ_FUA; |
| 186 | return op; |
| 187 | } |
| 188 | |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 189 | #define DIO_INLINE_BIO_VECS 4 |
| 190 | |
| 191 | static void blkdev_bio_end_io_simple(struct bio *bio) |
| 192 | { |
| 193 | struct task_struct *waiter = bio->bi_private; |
| 194 | |
| 195 | WRITE_ONCE(bio->bi_private, NULL); |
| 196 | wake_up_process(waiter); |
| 197 | } |
| 198 | |
| 199 | static ssize_t |
| 200 | __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter, |
| 201 | int nr_pages) |
| 202 | { |
| 203 | struct file *file = iocb->ki_filp; |
| 204 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
Jens Axboe | 72ecad2 | 2016-11-16 23:11:42 -0700 | [diff] [blame] | 205 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec; |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 206 | loff_t pos = iocb->ki_pos; |
| 207 | bool should_dirty = false; |
| 208 | struct bio bio; |
| 209 | ssize_t ret; |
| 210 | blk_qc_t qc; |
| 211 | int i; |
| 212 | |
Jens Axboe | 9a794fb | 2016-11-22 08:12:39 -0700 | [diff] [blame] | 213 | if ((pos | iov_iter_alignment(iter)) & |
| 214 | (bdev_logical_block_size(bdev) - 1)) |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 215 | return -EINVAL; |
| 216 | |
Jens Axboe | 72ecad2 | 2016-11-16 23:11:42 -0700 | [diff] [blame] | 217 | if (nr_pages <= DIO_INLINE_BIO_VECS) |
| 218 | vecs = inline_vecs; |
| 219 | else { |
| 220 | vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL); |
| 221 | if (!vecs) |
| 222 | return -ENOMEM; |
| 223 | } |
| 224 | |
Ming Lei | 3a83f46 | 2016-11-22 08:57:21 -0700 | [diff] [blame] | 225 | bio_init(&bio, vecs, nr_pages); |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 226 | bio.bi_bdev = bdev; |
Damien Le Moal | 4d1a476 | 2016-11-22 15:38:49 +0900 | [diff] [blame] | 227 | bio.bi_iter.bi_sector = pos >> 9; |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 228 | bio.bi_private = current; |
| 229 | bio.bi_end_io = blkdev_bio_end_io_simple; |
| 230 | |
| 231 | ret = bio_iov_iter_get_pages(&bio, iter); |
| 232 | if (unlikely(ret)) |
| 233 | return ret; |
| 234 | ret = bio.bi_iter.bi_size; |
| 235 | |
| 236 | if (iov_iter_rw(iter) == READ) { |
Jens Axboe | 78250c0 | 2016-11-17 17:50:47 +0100 | [diff] [blame] | 237 | bio.bi_opf = REQ_OP_READ; |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 238 | if (iter_is_iovec(iter)) |
| 239 | should_dirty = true; |
| 240 | } else { |
Jens Axboe | 78250c0 | 2016-11-17 17:50:47 +0100 | [diff] [blame] | 241 | bio.bi_opf = dio_bio_write_op(iocb); |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 242 | task_io_account_write(ret); |
| 243 | } |
| 244 | |
| 245 | qc = submit_bio(&bio); |
| 246 | for (;;) { |
| 247 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 248 | if (!READ_ONCE(bio.bi_private)) |
| 249 | break; |
| 250 | if (!(iocb->ki_flags & IOCB_HIPRI) || |
| 251 | !blk_mq_poll(bdev_get_queue(bdev), qc)) |
| 252 | io_schedule(); |
| 253 | } |
| 254 | __set_current_state(TASK_RUNNING); |
| 255 | |
| 256 | bio_for_each_segment_all(bvec, &bio, i) { |
| 257 | if (should_dirty && !PageCompound(bvec->bv_page)) |
| 258 | set_page_dirty_lock(bvec->bv_page); |
| 259 | put_page(bvec->bv_page); |
| 260 | } |
| 261 | |
Jens Axboe | 72ecad2 | 2016-11-16 23:11:42 -0700 | [diff] [blame] | 262 | if (vecs != inline_vecs) |
| 263 | kfree(vecs); |
| 264 | |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 265 | if (unlikely(bio.bi_error)) |
| 266 | return bio.bi_error; |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 267 | return ret; |
| 268 | } |
| 269 | |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 270 | struct blkdev_dio { |
| 271 | union { |
| 272 | struct kiocb *iocb; |
| 273 | struct task_struct *waiter; |
| 274 | }; |
| 275 | size_t size; |
| 276 | atomic_t ref; |
| 277 | bool multi_bio : 1; |
| 278 | bool should_dirty : 1; |
| 279 | bool is_sync : 1; |
| 280 | struct bio bio; |
| 281 | }; |
| 282 | |
| 283 | static struct bio_set *blkdev_dio_pool __read_mostly; |
| 284 | |
| 285 | static void blkdev_bio_end_io(struct bio *bio) |
| 286 | { |
| 287 | struct blkdev_dio *dio = bio->bi_private; |
| 288 | bool should_dirty = dio->should_dirty; |
| 289 | |
| 290 | if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) { |
| 291 | if (bio->bi_error && !dio->bio.bi_error) |
| 292 | dio->bio.bi_error = bio->bi_error; |
| 293 | } else { |
| 294 | if (!dio->is_sync) { |
| 295 | struct kiocb *iocb = dio->iocb; |
| 296 | ssize_t ret = dio->bio.bi_error; |
| 297 | |
| 298 | if (likely(!ret)) { |
| 299 | ret = dio->size; |
| 300 | iocb->ki_pos += ret; |
| 301 | } |
| 302 | |
| 303 | dio->iocb->ki_complete(iocb, ret, 0); |
| 304 | bio_put(&dio->bio); |
| 305 | } else { |
| 306 | struct task_struct *waiter = dio->waiter; |
| 307 | |
| 308 | WRITE_ONCE(dio->waiter, NULL); |
| 309 | wake_up_process(waiter); |
| 310 | } |
| 311 | } |
| 312 | |
| 313 | if (should_dirty) { |
| 314 | bio_check_pages_dirty(bio); |
| 315 | } else { |
| 316 | struct bio_vec *bvec; |
| 317 | int i; |
| 318 | |
| 319 | bio_for_each_segment_all(bvec, bio, i) |
| 320 | put_page(bvec->bv_page); |
| 321 | bio_put(bio); |
| 322 | } |
| 323 | } |
| 324 | |
Andrew Morton | b2e895d | 2007-02-03 01:14:01 -0800 | [diff] [blame] | 325 | static ssize_t |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 326 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages) |
Andrew Morton | b2e895d | 2007-02-03 01:14:01 -0800 | [diff] [blame] | 327 | { |
| 328 | struct file *file = iocb->ki_filp; |
Dan Williams | 4ebb16c | 2015-10-28 07:48:19 +0900 | [diff] [blame] | 329 | struct inode *inode = bdev_file_inode(file); |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 330 | struct block_device *bdev = I_BDEV(inode); |
Christoph Hellwig | 64d656a | 2016-12-22 19:20:45 +0100 | [diff] [blame] | 331 | struct blk_plug plug; |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 332 | struct blkdev_dio *dio; |
| 333 | struct bio *bio; |
Christoph Hellwig | 690e532 | 2017-01-24 14:50:19 +0100 | [diff] [blame] | 334 | bool is_read = (iov_iter_rw(iter) == READ), is_sync; |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 335 | loff_t pos = iocb->ki_pos; |
| 336 | blk_qc_t qc = BLK_QC_T_NONE; |
| 337 | int ret; |
| 338 | |
Jens Axboe | 9a794fb | 2016-11-22 08:12:39 -0700 | [diff] [blame] | 339 | if ((pos | iov_iter_alignment(iter)) & |
| 340 | (bdev_logical_block_size(bdev) - 1)) |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 341 | return -EINVAL; |
| 342 | |
| 343 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool); |
| 344 | bio_get(bio); /* extra ref for the completion handler */ |
| 345 | |
| 346 | dio = container_of(bio, struct blkdev_dio, bio); |
Christoph Hellwig | 690e532 | 2017-01-24 14:50:19 +0100 | [diff] [blame] | 347 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 348 | if (dio->is_sync) |
| 349 | dio->waiter = current; |
| 350 | else |
| 351 | dio->iocb = iocb; |
| 352 | |
| 353 | dio->size = 0; |
| 354 | dio->multi_bio = false; |
| 355 | dio->should_dirty = is_read && (iter->type == ITER_IOVEC); |
| 356 | |
Christoph Hellwig | 64d656a | 2016-12-22 19:20:45 +0100 | [diff] [blame] | 357 | blk_start_plug(&plug); |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 358 | for (;;) { |
| 359 | bio->bi_bdev = bdev; |
Damien Le Moal | 4d1a476 | 2016-11-22 15:38:49 +0900 | [diff] [blame] | 360 | bio->bi_iter.bi_sector = pos >> 9; |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 361 | bio->bi_private = dio; |
| 362 | bio->bi_end_io = blkdev_bio_end_io; |
| 363 | |
| 364 | ret = bio_iov_iter_get_pages(bio, iter); |
| 365 | if (unlikely(ret)) { |
| 366 | bio->bi_error = ret; |
| 367 | bio_endio(bio); |
| 368 | break; |
| 369 | } |
| 370 | |
| 371 | if (is_read) { |
| 372 | bio->bi_opf = REQ_OP_READ; |
| 373 | if (dio->should_dirty) |
| 374 | bio_set_pages_dirty(bio); |
| 375 | } else { |
| 376 | bio->bi_opf = dio_bio_write_op(iocb); |
| 377 | task_io_account_write(bio->bi_iter.bi_size); |
| 378 | } |
| 379 | |
| 380 | dio->size += bio->bi_iter.bi_size; |
| 381 | pos += bio->bi_iter.bi_size; |
| 382 | |
| 383 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); |
| 384 | if (!nr_pages) { |
| 385 | qc = submit_bio(bio); |
| 386 | break; |
| 387 | } |
| 388 | |
| 389 | if (!dio->multi_bio) { |
| 390 | dio->multi_bio = true; |
| 391 | atomic_set(&dio->ref, 2); |
| 392 | } else { |
| 393 | atomic_inc(&dio->ref); |
| 394 | } |
| 395 | |
| 396 | submit_bio(bio); |
| 397 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
| 398 | } |
Christoph Hellwig | 64d656a | 2016-12-22 19:20:45 +0100 | [diff] [blame] | 399 | blk_finish_plug(&plug); |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 400 | |
Christoph Hellwig | 690e532 | 2017-01-24 14:50:19 +0100 | [diff] [blame] | 401 | if (!is_sync) |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 402 | return -EIOCBQUEUED; |
| 403 | |
| 404 | for (;;) { |
| 405 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 406 | if (!READ_ONCE(dio->waiter)) |
| 407 | break; |
| 408 | |
| 409 | if (!(iocb->ki_flags & IOCB_HIPRI) || |
| 410 | !blk_mq_poll(bdev_get_queue(bdev), qc)) |
| 411 | io_schedule(); |
| 412 | } |
| 413 | __set_current_state(TASK_RUNNING); |
| 414 | |
| 415 | ret = dio->bio.bi_error; |
Shaohua Li | 7a62a52 | 2016-12-13 19:07:55 -0800 | [diff] [blame] | 416 | if (likely(!ret)) |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 417 | ret = dio->size; |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 418 | |
| 419 | bio_put(&dio->bio); |
| 420 | return ret; |
| 421 | } |
| 422 | |
| 423 | static ssize_t |
| 424 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
| 425 | { |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 426 | int nr_pages; |
Andrew Morton | b2e895d | 2007-02-03 01:14:01 -0800 | [diff] [blame] | 427 | |
Jens Axboe | 72ecad2 | 2016-11-16 23:11:42 -0700 | [diff] [blame] | 428 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 429 | if (!nr_pages) |
| 430 | return 0; |
Jens Axboe | 72ecad2 | 2016-11-16 23:11:42 -0700 | [diff] [blame] | 431 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
Christoph Hellwig | 189ce2b | 2016-10-31 11:59:25 -0600 | [diff] [blame] | 432 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 433 | |
| 434 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); |
Andrew Morton | b2e895d | 2007-02-03 01:14:01 -0800 | [diff] [blame] | 435 | } |
| 436 | |
Christoph Hellwig | 542ff7b | 2016-11-16 23:14:22 -0700 | [diff] [blame] | 437 | static __init int blkdev_init(void) |
| 438 | { |
| 439 | blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio)); |
| 440 | if (!blkdev_dio_pool) |
| 441 | return -ENOMEM; |
| 442 | return 0; |
| 443 | } |
| 444 | module_init(blkdev_init); |
| 445 | |
Jan Kara | 5cee581 | 2009-04-27 16:43:51 +0200 | [diff] [blame] | 446 | int __sync_blockdev(struct block_device *bdev, int wait) |
| 447 | { |
| 448 | if (!bdev) |
| 449 | return 0; |
| 450 | if (!wait) |
| 451 | return filemap_flush(bdev->bd_inode->i_mapping); |
| 452 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); |
| 453 | } |
| 454 | |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 455 | /* |
| 456 | * Write out and wait upon all the dirty data associated with a block |
| 457 | * device via its mapping. Does not take the superblock lock. |
| 458 | */ |
| 459 | int sync_blockdev(struct block_device *bdev) |
| 460 | { |
Jan Kara | 5cee581 | 2009-04-27 16:43:51 +0200 | [diff] [blame] | 461 | return __sync_blockdev(bdev, 1); |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 462 | } |
| 463 | EXPORT_SYMBOL(sync_blockdev); |
| 464 | |
| 465 | /* |
| 466 | * Write out and wait upon all dirty data associated with this |
| 467 | * device. Filesystem data as well as the underlying block |
| 468 | * device. Takes the superblock lock. |
| 469 | */ |
| 470 | int fsync_bdev(struct block_device *bdev) |
| 471 | { |
| 472 | struct super_block *sb = get_super(bdev); |
| 473 | if (sb) { |
Jan Kara | 60b0680 | 2009-04-27 16:43:53 +0200 | [diff] [blame] | 474 | int res = sync_filesystem(sb); |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 475 | drop_super(sb); |
| 476 | return res; |
| 477 | } |
| 478 | return sync_blockdev(bdev); |
| 479 | } |
Al Viro | 47e4491 | 2009-04-01 07:07:16 -0400 | [diff] [blame] | 480 | EXPORT_SYMBOL(fsync_bdev); |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 481 | |
| 482 | /** |
| 483 | * freeze_bdev -- lock a filesystem and force it into a consistent state |
| 484 | * @bdev: blockdevice to lock |
| 485 | * |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 486 | * If a superblock is found on this device, we take the s_umount semaphore |
| 487 | * on it to make sure nobody unmounts until the snapshot creation is done. |
| 488 | * The reference counter (bd_fsfreeze_count) guarantees that only the last |
| 489 | * unfreeze process can unfreeze the frozen filesystem actually when multiple |
| 490 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and |
| 491 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze |
| 492 | * actually. |
| 493 | */ |
| 494 | struct super_block *freeze_bdev(struct block_device *bdev) |
| 495 | { |
| 496 | struct super_block *sb; |
| 497 | int error = 0; |
| 498 | |
| 499 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 500 | if (++bdev->bd_fsfreeze_count > 1) { |
| 501 | /* |
| 502 | * We don't even need to grab a reference - the first call |
| 503 | * to freeze_bdev grab an active reference and only the last |
| 504 | * thaw_bdev drops it. |
| 505 | */ |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 506 | sb = get_super(bdev); |
Andrey Ryabinin | 5bb53c0 | 2016-08-23 18:55:31 +0300 | [diff] [blame] | 507 | if (sb) |
| 508 | drop_super(sb); |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 509 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 510 | return sb; |
| 511 | } |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 512 | |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 513 | sb = get_active_super(bdev); |
| 514 | if (!sb) |
| 515 | goto out; |
Benjamin Marzinski | 48b6bca | 2014-11-13 20:42:03 -0600 | [diff] [blame] | 516 | if (sb->s_op->freeze_super) |
| 517 | error = sb->s_op->freeze_super(sb); |
| 518 | else |
| 519 | error = freeze_super(sb); |
Josef Bacik | 18e9e51 | 2010-03-23 10:34:56 -0400 | [diff] [blame] | 520 | if (error) { |
| 521 | deactivate_super(sb); |
| 522 | bdev->bd_fsfreeze_count--; |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 523 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
Josef Bacik | 18e9e51 | 2010-03-23 10:34:56 -0400 | [diff] [blame] | 524 | return ERR_PTR(error); |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 525 | } |
Josef Bacik | 18e9e51 | 2010-03-23 10:34:56 -0400 | [diff] [blame] | 526 | deactivate_super(sb); |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 527 | out: |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 528 | sync_blockdev(bdev); |
| 529 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
Christoph Hellwig | 4fadd7b | 2009-08-03 23:28:06 +0200 | [diff] [blame] | 530 | return sb; /* thaw_bdev releases s->s_umount */ |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 531 | } |
| 532 | EXPORT_SYMBOL(freeze_bdev); |
| 533 | |
| 534 | /** |
| 535 | * thaw_bdev -- unlock filesystem |
| 536 | * @bdev: blockdevice to unlock |
| 537 | * @sb: associated superblock |
| 538 | * |
| 539 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). |
| 540 | */ |
| 541 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) |
| 542 | { |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 543 | int error = -EINVAL; |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 544 | |
| 545 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 546 | if (!bdev->bd_fsfreeze_count) |
Josef Bacik | 18e9e51 | 2010-03-23 10:34:56 -0400 | [diff] [blame] | 547 | goto out; |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 548 | |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 549 | error = 0; |
| 550 | if (--bdev->bd_fsfreeze_count > 0) |
Josef Bacik | 18e9e51 | 2010-03-23 10:34:56 -0400 | [diff] [blame] | 551 | goto out; |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 552 | |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 553 | if (!sb) |
Josef Bacik | 18e9e51 | 2010-03-23 10:34:56 -0400 | [diff] [blame] | 554 | goto out; |
Christoph Hellwig | 4504230 | 2009-08-03 23:28:35 +0200 | [diff] [blame] | 555 | |
Benjamin Marzinski | 48b6bca | 2014-11-13 20:42:03 -0600 | [diff] [blame] | 556 | if (sb->s_op->thaw_super) |
| 557 | error = sb->s_op->thaw_super(sb); |
| 558 | else |
| 559 | error = thaw_super(sb); |
Pierre Morel | 997198b | 2016-10-04 10:53:40 +0200 | [diff] [blame] | 560 | if (error) |
Josef Bacik | 18e9e51 | 2010-03-23 10:34:56 -0400 | [diff] [blame] | 561 | bdev->bd_fsfreeze_count++; |
Josef Bacik | 18e9e51 | 2010-03-23 10:34:56 -0400 | [diff] [blame] | 562 | out: |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 563 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
Pierre Morel | 997198b | 2016-10-04 10:53:40 +0200 | [diff] [blame] | 564 | return error; |
Nick Piggin | 585d3bc | 2009-02-25 10:44:19 +0100 | [diff] [blame] | 565 | } |
| 566 | EXPORT_SYMBOL(thaw_bdev); |
| 567 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
| 569 | { |
| 570 | return block_write_full_page(page, blkdev_get_block, wbc); |
| 571 | } |
| 572 | |
| 573 | static int blkdev_readpage(struct file * file, struct page * page) |
| 574 | { |
| 575 | return block_read_full_page(page, blkdev_get_block); |
| 576 | } |
| 577 | |
Akinobu Mita | 447f05b | 2014-10-09 15:26:58 -0700 | [diff] [blame] | 578 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
| 579 | struct list_head *pages, unsigned nr_pages) |
| 580 | { |
| 581 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); |
| 582 | } |
| 583 | |
Nick Piggin | 6272b5a | 2007-10-16 01:25:04 -0700 | [diff] [blame] | 584 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
| 585 | loff_t pos, unsigned len, unsigned flags, |
| 586 | struct page **pagep, void **fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | { |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 588 | return block_write_begin(mapping, pos, len, flags, pagep, |
| 589 | blkdev_get_block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 590 | } |
| 591 | |
Nick Piggin | 6272b5a | 2007-10-16 01:25:04 -0700 | [diff] [blame] | 592 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
| 593 | loff_t pos, unsigned len, unsigned copied, |
| 594 | struct page *page, void *fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | { |
Nick Piggin | 6272b5a | 2007-10-16 01:25:04 -0700 | [diff] [blame] | 596 | int ret; |
| 597 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
| 598 | |
| 599 | unlock_page(page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 600 | put_page(page); |
Nick Piggin | 6272b5a | 2007-10-16 01:25:04 -0700 | [diff] [blame] | 601 | |
| 602 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | } |
| 604 | |
| 605 | /* |
| 606 | * private llseek: |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 607 | * for a block special file file_inode(file)->i_size is zero |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 608 | * so we compute the size by hand (just as in block_read/write above) |
| 609 | */ |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 610 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | { |
Dan Williams | 4ebb16c | 2015-10-28 07:48:19 +0900 | [diff] [blame] | 612 | struct inode *bd_inode = bdev_file_inode(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | loff_t retval; |
| 614 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 615 | inode_lock(bd_inode); |
Al Viro | 5d48f3a | 2013-06-23 21:34:45 +0400 | [diff] [blame] | 616 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 617 | inode_unlock(bd_inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | return retval; |
| 619 | } |
| 620 | |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 621 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 622 | { |
Dan Williams | 4ebb16c | 2015-10-28 07:48:19 +0900 | [diff] [blame] | 623 | struct inode *bd_inode = bdev_file_inode(filp); |
Anton Blanchard | b8af67e | 2010-04-23 13:18:06 -0400 | [diff] [blame] | 624 | struct block_device *bdev = I_BDEV(bd_inode); |
Christoph Hellwig | ab0a973 | 2009-10-29 14:14:04 +0100 | [diff] [blame] | 625 | int error; |
Rafael J. Wysocki | da5aa86 | 2011-08-02 02:17:48 +0200 | [diff] [blame] | 626 | |
| 627 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); |
| 628 | if (error) |
| 629 | return error; |
Christoph Hellwig | ab0a973 | 2009-10-29 14:14:04 +0100 | [diff] [blame] | 630 | |
Anton Blanchard | b8af67e | 2010-04-23 13:18:06 -0400 | [diff] [blame] | 631 | /* |
| 632 | * There is no need to serialise calls to blkdev_issue_flush with |
| 633 | * i_mutex and doing so causes performance issues with concurrent |
| 634 | * O_SYNC writers to a block device. |
| 635 | */ |
Christoph Hellwig | dd3932e | 2010-09-16 20:51:46 +0200 | [diff] [blame] | 636 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
Christoph Hellwig | ab0a973 | 2009-10-29 14:14:04 +0100 | [diff] [blame] | 637 | if (error == -EOPNOTSUPP) |
| 638 | error = 0; |
Anton Blanchard | b8af67e | 2010-04-23 13:18:06 -0400 | [diff] [blame] | 639 | |
Christoph Hellwig | ab0a973 | 2009-10-29 14:14:04 +0100 | [diff] [blame] | 640 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | } |
Andrew Morton | b1dd3b2 | 2010-04-06 14:35:00 -0700 | [diff] [blame] | 642 | EXPORT_SYMBOL(blkdev_fsync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 644 | /** |
| 645 | * bdev_read_page() - Start reading a page from a block device |
| 646 | * @bdev: The device to read the page from |
| 647 | * @sector: The offset on the device to read the page to (need not be aligned) |
| 648 | * @page: The page to read |
| 649 | * |
| 650 | * On entry, the page should be locked. It will be unlocked when the page |
| 651 | * has been read. If the block driver implements rw_page synchronously, |
| 652 | * that will be true on exit from this function, but it need not be. |
| 653 | * |
| 654 | * Errors returned by this function are usually "soft", eg out of memory, or |
| 655 | * queue full; callers should try a different route to read this page rather |
| 656 | * than propagate an error back up the stack. |
| 657 | * |
| 658 | * Return: negative errno if an error occurs, 0 if submission was successful. |
| 659 | */ |
| 660 | int bdev_read_page(struct block_device *bdev, sector_t sector, |
| 661 | struct page *page) |
| 662 | { |
| 663 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
Dan Williams | 2e6edc9 | 2015-11-19 13:29:28 -0800 | [diff] [blame] | 664 | int result = -EOPNOTSUPP; |
| 665 | |
Vishal Verma | f68eb1e | 2015-05-12 13:48:53 -0400 | [diff] [blame] | 666 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
Dan Williams | 2e6edc9 | 2015-11-19 13:29:28 -0800 | [diff] [blame] | 667 | return result; |
| 668 | |
Christoph Hellwig | 6f3b0e8 | 2015-11-26 09:13:05 +0100 | [diff] [blame] | 669 | result = blk_queue_enter(bdev->bd_queue, false); |
Dan Williams | 2e6edc9 | 2015-11-19 13:29:28 -0800 | [diff] [blame] | 670 | if (result) |
| 671 | return result; |
Jens Axboe | c11f0c0 | 2016-08-05 08:11:04 -0600 | [diff] [blame] | 672 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false); |
Dan Williams | 2e6edc9 | 2015-11-19 13:29:28 -0800 | [diff] [blame] | 673 | blk_queue_exit(bdev->bd_queue); |
| 674 | return result; |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 675 | } |
| 676 | EXPORT_SYMBOL_GPL(bdev_read_page); |
| 677 | |
| 678 | /** |
| 679 | * bdev_write_page() - Start writing a page to a block device |
| 680 | * @bdev: The device to write the page to |
| 681 | * @sector: The offset on the device to write the page to (need not be aligned) |
| 682 | * @page: The page to write |
| 683 | * @wbc: The writeback_control for the write |
| 684 | * |
| 685 | * On entry, the page should be locked and not currently under writeback. |
| 686 | * On exit, if the write started successfully, the page will be unlocked and |
| 687 | * under writeback. If the write failed already (eg the driver failed to |
| 688 | * queue the page to the device), the page will still be locked. If the |
| 689 | * caller is a ->writepage implementation, it will need to unlock the page. |
| 690 | * |
| 691 | * Errors returned by this function are usually "soft", eg out of memory, or |
| 692 | * queue full; callers should try a different route to write this page rather |
| 693 | * than propagate an error back up the stack. |
| 694 | * |
| 695 | * Return: negative errno if an error occurs, 0 if submission was successful. |
| 696 | */ |
| 697 | int bdev_write_page(struct block_device *bdev, sector_t sector, |
| 698 | struct page *page, struct writeback_control *wbc) |
| 699 | { |
| 700 | int result; |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 701 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
Dan Williams | 2e6edc9 | 2015-11-19 13:29:28 -0800 | [diff] [blame] | 702 | |
Vishal Verma | f68eb1e | 2015-05-12 13:48:53 -0400 | [diff] [blame] | 703 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 704 | return -EOPNOTSUPP; |
Christoph Hellwig | 6f3b0e8 | 2015-11-26 09:13:05 +0100 | [diff] [blame] | 705 | result = blk_queue_enter(bdev->bd_queue, false); |
Dan Williams | 2e6edc9 | 2015-11-19 13:29:28 -0800 | [diff] [blame] | 706 | if (result) |
| 707 | return result; |
| 708 | |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 709 | set_page_writeback(page); |
Jens Axboe | c11f0c0 | 2016-08-05 08:11:04 -0600 | [diff] [blame] | 710 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true); |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 711 | if (result) |
| 712 | end_page_writeback(page); |
| 713 | else |
| 714 | unlock_page(page); |
Dan Williams | 2e6edc9 | 2015-11-19 13:29:28 -0800 | [diff] [blame] | 715 | blk_queue_exit(bdev->bd_queue); |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 716 | return result; |
| 717 | } |
| 718 | EXPORT_SYMBOL_GPL(bdev_write_page); |
| 719 | |
Matthew Wilcox | dd22f55 | 2015-01-07 18:05:34 +0200 | [diff] [blame] | 720 | /** |
| 721 | * bdev_direct_access() - Get the address for directly-accessibly memory |
| 722 | * @bdev: The device containing the memory |
Dan Williams | b2e0d16 | 2016-01-15 16:55:59 -0800 | [diff] [blame] | 723 | * @dax: control and output parameters for ->direct_access |
Matthew Wilcox | dd22f55 | 2015-01-07 18:05:34 +0200 | [diff] [blame] | 724 | * |
| 725 | * If a block device is made up of directly addressable memory, this function |
| 726 | * will tell the caller the PFN and the address of the memory. The address |
| 727 | * may be directly dereferenced within the kernel without the need to call |
| 728 | * ioremap(), kmap() or similar. The PFN is suitable for inserting into |
| 729 | * page tables. |
| 730 | * |
| 731 | * Return: negative errno if an error occurs, otherwise the number of bytes |
| 732 | * accessible at this address. |
| 733 | */ |
Dan Williams | b2e0d16 | 2016-01-15 16:55:59 -0800 | [diff] [blame] | 734 | long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax) |
Matthew Wilcox | dd22f55 | 2015-01-07 18:05:34 +0200 | [diff] [blame] | 735 | { |
Dan Williams | b2e0d16 | 2016-01-15 16:55:59 -0800 | [diff] [blame] | 736 | sector_t sector = dax->sector; |
| 737 | long avail, size = dax->size; |
Matthew Wilcox | dd22f55 | 2015-01-07 18:05:34 +0200 | [diff] [blame] | 738 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
| 739 | |
Matthew Wilcox | 43c3dd0 | 2015-07-03 10:40:43 -0400 | [diff] [blame] | 740 | /* |
| 741 | * The device driver is allowed to sleep, in order to make the |
| 742 | * memory directly accessible. |
| 743 | */ |
| 744 | might_sleep(); |
| 745 | |
Matthew Wilcox | dd22f55 | 2015-01-07 18:05:34 +0200 | [diff] [blame] | 746 | if (size < 0) |
| 747 | return size; |
Toshi Kani | 163d4ba | 2016-06-23 17:05:50 -0400 | [diff] [blame] | 748 | if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access) |
Matthew Wilcox | dd22f55 | 2015-01-07 18:05:34 +0200 | [diff] [blame] | 749 | return -EOPNOTSUPP; |
| 750 | if ((sector + DIV_ROUND_UP(size, 512)) > |
| 751 | part_nr_sects_read(bdev->bd_part)) |
| 752 | return -ERANGE; |
| 753 | sector += get_start_sect(bdev); |
| 754 | if (sector % (PAGE_SIZE / 512)) |
| 755 | return -EINVAL; |
Dan Williams | 0a70bd4 | 2016-02-24 14:02:11 -0800 | [diff] [blame] | 756 | avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size); |
Matthew Wilcox | dd22f55 | 2015-01-07 18:05:34 +0200 | [diff] [blame] | 757 | if (!avail) |
| 758 | return -ERANGE; |
Dan Williams | fe683ad | 2016-01-15 16:55:56 -0800 | [diff] [blame] | 759 | if (avail > 0 && avail & ~PAGE_MASK) |
| 760 | return -ENXIO; |
Matthew Wilcox | dd22f55 | 2015-01-07 18:05:34 +0200 | [diff] [blame] | 761 | return min(avail, size); |
| 762 | } |
| 763 | EXPORT_SYMBOL_GPL(bdev_direct_access); |
| 764 | |
Toshi Kani | 2d96afc | 2016-05-10 10:23:53 -0600 | [diff] [blame] | 765 | /** |
| 766 | * bdev_dax_supported() - Check if the device supports dax for filesystem |
| 767 | * @sb: The superblock of the device |
| 768 | * @blocksize: The block size of the device |
| 769 | * |
| 770 | * This is a library function for filesystems to check if the block device |
| 771 | * can be mounted with dax option. |
| 772 | * |
| 773 | * Return: negative errno if unsupported, 0 if supported. |
| 774 | */ |
| 775 | int bdev_dax_supported(struct super_block *sb, int blocksize) |
| 776 | { |
| 777 | struct blk_dax_ctl dax = { |
| 778 | .sector = 0, |
| 779 | .size = PAGE_SIZE, |
| 780 | }; |
| 781 | int err; |
| 782 | |
| 783 | if (blocksize != PAGE_SIZE) { |
| 784 | vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax"); |
| 785 | return -EINVAL; |
| 786 | } |
| 787 | |
| 788 | err = bdev_direct_access(sb->s_bdev, &dax); |
| 789 | if (err < 0) { |
| 790 | switch (err) { |
| 791 | case -EOPNOTSUPP: |
| 792 | vfs_msg(sb, KERN_ERR, |
| 793 | "error: device does not support dax"); |
| 794 | break; |
| 795 | case -EINVAL: |
| 796 | vfs_msg(sb, KERN_ERR, |
| 797 | "error: unaligned partition for dax"); |
| 798 | break; |
| 799 | default: |
| 800 | vfs_msg(sb, KERN_ERR, |
| 801 | "error: dax access failed (%d)", err); |
| 802 | } |
| 803 | return err; |
| 804 | } |
| 805 | |
| 806 | return 0; |
| 807 | } |
| 808 | EXPORT_SYMBOL_GPL(bdev_dax_supported); |
| 809 | |
Toshi Kani | a8078b1 | 2016-05-10 10:23:57 -0600 | [diff] [blame] | 810 | /** |
| 811 | * bdev_dax_capable() - Return if the raw device is capable for dax |
| 812 | * @bdev: The device for raw block device access |
| 813 | */ |
| 814 | bool bdev_dax_capable(struct block_device *bdev) |
| 815 | { |
Toshi Kani | a8078b1 | 2016-05-10 10:23:57 -0600 | [diff] [blame] | 816 | struct blk_dax_ctl dax = { |
| 817 | .size = PAGE_SIZE, |
| 818 | }; |
| 819 | |
| 820 | if (!IS_ENABLED(CONFIG_FS_DAX)) |
| 821 | return false; |
| 822 | |
| 823 | dax.sector = 0; |
| 824 | if (bdev_direct_access(bdev, &dax) < 0) |
| 825 | return false; |
| 826 | |
| 827 | dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512); |
| 828 | if (bdev_direct_access(bdev, &dax) < 0) |
| 829 | return false; |
| 830 | |
Toshi Kani | a8078b1 | 2016-05-10 10:23:57 -0600 | [diff] [blame] | 831 | return true; |
| 832 | } |
| 833 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | /* |
| 835 | * pseudo-fs |
| 836 | */ |
| 837 | |
| 838 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 839 | static struct kmem_cache * bdev_cachep __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 840 | |
| 841 | static struct inode *bdev_alloc_inode(struct super_block *sb) |
| 842 | { |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 843 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | if (!ei) |
| 845 | return NULL; |
| 846 | return &ei->vfs_inode; |
| 847 | } |
| 848 | |
Nick Piggin | fa0d7e3d | 2011-01-07 17:49:49 +1100 | [diff] [blame] | 849 | static void bdev_i_callback(struct rcu_head *head) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | { |
Nick Piggin | fa0d7e3d | 2011-01-07 17:49:49 +1100 | [diff] [blame] | 851 | struct inode *inode = container_of(head, struct inode, i_rcu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | struct bdev_inode *bdi = BDEV_I(inode); |
| 853 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 854 | kmem_cache_free(bdev_cachep, bdi); |
| 855 | } |
| 856 | |
Nick Piggin | fa0d7e3d | 2011-01-07 17:49:49 +1100 | [diff] [blame] | 857 | static void bdev_destroy_inode(struct inode *inode) |
| 858 | { |
| 859 | call_rcu(&inode->i_rcu, bdev_i_callback); |
| 860 | } |
| 861 | |
Alexey Dobriyan | 51cc506 | 2008-07-25 19:45:34 -0700 | [diff] [blame] | 862 | static void init_once(void *foo) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | { |
| 864 | struct bdev_inode *ei = (struct bdev_inode *) foo; |
| 865 | struct block_device *bdev = &ei->bdev; |
| 866 | |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 867 | memset(bdev, 0, sizeof(*bdev)); |
| 868 | mutex_init(&bdev->bd_mutex); |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 869 | INIT_LIST_HEAD(&bdev->bd_list); |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 870 | #ifdef CONFIG_SYSFS |
| 871 | INIT_LIST_HEAD(&bdev->bd_holder_disks); |
| 872 | #endif |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 873 | inode_init_once(&ei->vfs_inode); |
Takashi Sato | fcccf50 | 2009-01-09 16:40:59 -0800 | [diff] [blame] | 874 | /* Initialize mutex for freeze. */ |
| 875 | mutex_init(&bdev->bd_fsfreeze_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | } |
| 877 | |
Al Viro | b57922d | 2010-06-07 14:34:48 -0400 | [diff] [blame] | 878 | static void bdev_evict_inode(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 879 | { |
| 880 | struct block_device *bdev = &BDEV_I(inode)->bdev; |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 881 | truncate_inode_pages_final(&inode->i_data); |
Al Viro | b57922d | 2010-06-07 14:34:48 -0400 | [diff] [blame] | 882 | invalidate_inode_buffers(inode); /* is it needed here? */ |
Jan Kara | dbd5768 | 2012-05-03 14:48:02 +0200 | [diff] [blame] | 883 | clear_inode(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 884 | spin_lock(&bdev_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | list_del_init(&bdev->bd_list); |
| 886 | spin_unlock(&bdev_lock); |
| 887 | } |
| 888 | |
Josef 'Jeff' Sipek | ee9b6d6 | 2007-02-12 00:55:41 -0800 | [diff] [blame] | 889 | static const struct super_operations bdev_sops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | .statfs = simple_statfs, |
| 891 | .alloc_inode = bdev_alloc_inode, |
| 892 | .destroy_inode = bdev_destroy_inode, |
| 893 | .drop_inode = generic_delete_inode, |
Al Viro | b57922d | 2010-06-07 14:34:48 -0400 | [diff] [blame] | 894 | .evict_inode = bdev_evict_inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 895 | }; |
| 896 | |
Al Viro | 51139ad | 2010-07-25 23:47:46 +0400 | [diff] [blame] | 897 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
| 898 | int flags, const char *dev_name, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 899 | { |
Shaohua Li | 3684aa7 | 2016-02-22 15:27:40 -0700 | [diff] [blame] | 900 | struct dentry *dent; |
| 901 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); |
Vegard Nossum | e9e5e3f | 2016-08-22 12:47:43 +0200 | [diff] [blame] | 902 | if (!IS_ERR(dent)) |
Shaohua Li | 3684aa7 | 2016-02-22 15:27:40 -0700 | [diff] [blame] | 903 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; |
| 904 | return dent; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 905 | } |
| 906 | |
| 907 | static struct file_system_type bd_type = { |
| 908 | .name = "bdev", |
Al Viro | 51139ad | 2010-07-25 23:47:46 +0400 | [diff] [blame] | 909 | .mount = bd_mount, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 910 | .kill_sb = kill_anon_super, |
| 911 | }; |
| 912 | |
Tejun Heo | a212b10 | 2015-05-22 17:13:33 -0400 | [diff] [blame] | 913 | struct super_block *blockdev_superblock __read_mostly; |
| 914 | EXPORT_SYMBOL_GPL(blockdev_superblock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 915 | |
| 916 | void __init bdev_cache_init(void) |
| 917 | { |
| 918 | int err; |
Sergey Senozhatsky | ace8577 | 2012-01-10 02:43:59 +0300 | [diff] [blame] | 919 | static struct vfsmount *bd_mnt; |
Denis ChengRq | c2acf7b | 2008-12-01 14:34:56 -0800 | [diff] [blame] | 920 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
Paul Jackson | fffb60f | 2006-03-24 03:16:06 -0800 | [diff] [blame] | 922 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
Vladimir Davydov | 5d09705 | 2016-01-14 15:18:21 -0800 | [diff] [blame] | 923 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 924 | init_once); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | err = register_filesystem(&bd_type); |
| 926 | if (err) |
| 927 | panic("Cannot register bdev pseudo-fs"); |
| 928 | bd_mnt = kern_mount(&bd_type); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | if (IS_ERR(bd_mnt)) |
| 930 | panic("Cannot create bdev pseudo-fs"); |
Sergey Senozhatsky | ace8577 | 2012-01-10 02:43:59 +0300 | [diff] [blame] | 931 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 932 | } |
| 933 | |
| 934 | /* |
| 935 | * Most likely _very_ bad one - but then it's hardly critical for small |
| 936 | * /dev and can be fixed when somebody will need really large one. |
| 937 | * Keep in mind that it will be fed through icache hash function too. |
| 938 | */ |
| 939 | static inline unsigned long hash(dev_t dev) |
| 940 | { |
| 941 | return MAJOR(dev)+MINOR(dev); |
| 942 | } |
| 943 | |
| 944 | static int bdev_test(struct inode *inode, void *data) |
| 945 | { |
| 946 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; |
| 947 | } |
| 948 | |
| 949 | static int bdev_set(struct inode *inode, void *data) |
| 950 | { |
| 951 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; |
| 952 | return 0; |
| 953 | } |
| 954 | |
| 955 | static LIST_HEAD(all_bdevs); |
| 956 | |
| 957 | struct block_device *bdget(dev_t dev) |
| 958 | { |
| 959 | struct block_device *bdev; |
| 960 | struct inode *inode; |
| 961 | |
Denis ChengRq | c2acf7b | 2008-12-01 14:34:56 -0800 | [diff] [blame] | 962 | inode = iget5_locked(blockdev_superblock, hash(dev), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 963 | bdev_test, bdev_set, &dev); |
| 964 | |
| 965 | if (!inode) |
| 966 | return NULL; |
| 967 | |
| 968 | bdev = &BDEV_I(inode)->bdev; |
| 969 | |
| 970 | if (inode->i_state & I_NEW) { |
| 971 | bdev->bd_contains = NULL; |
Lachlan McIlroy | 782b94c | 2011-06-30 11:01:45 +1000 | [diff] [blame] | 972 | bdev->bd_super = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | bdev->bd_inode = inode; |
| 974 | bdev->bd_block_size = (1 << inode->i_blkbits); |
| 975 | bdev->bd_part_count = 0; |
| 976 | bdev->bd_invalidated = 0; |
| 977 | inode->i_mode = S_IFBLK; |
| 978 | inode->i_rdev = dev; |
| 979 | inode->i_bdev = bdev; |
| 980 | inode->i_data.a_ops = &def_blk_aops; |
| 981 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | spin_lock(&bdev_lock); |
| 983 | list_add(&bdev->bd_list, &all_bdevs); |
| 984 | spin_unlock(&bdev_lock); |
| 985 | unlock_new_inode(inode); |
| 986 | } |
| 987 | return bdev; |
| 988 | } |
| 989 | |
| 990 | EXPORT_SYMBOL(bdget); |
| 991 | |
Alan Jenkins | dddac6a | 2009-07-29 21:07:55 +0200 | [diff] [blame] | 992 | /** |
| 993 | * bdgrab -- Grab a reference to an already referenced block device |
| 994 | * @bdev: Block device to grab a reference to. |
| 995 | */ |
| 996 | struct block_device *bdgrab(struct block_device *bdev) |
| 997 | { |
Al Viro | 7de9c6ee | 2010-10-23 11:11:40 -0400 | [diff] [blame] | 998 | ihold(bdev->bd_inode); |
Alan Jenkins | dddac6a | 2009-07-29 21:07:55 +0200 | [diff] [blame] | 999 | return bdev; |
| 1000 | } |
Anatol Pomozov | c1681bf | 2013-04-01 09:47:56 -0700 | [diff] [blame] | 1001 | EXPORT_SYMBOL(bdgrab); |
Alan Jenkins | dddac6a | 2009-07-29 21:07:55 +0200 | [diff] [blame] | 1002 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1003 | long nr_blockdev_pages(void) |
| 1004 | { |
Matthias Kaehlcke | 203a293 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1005 | struct block_device *bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1006 | long ret = 0; |
| 1007 | spin_lock(&bdev_lock); |
Matthias Kaehlcke | 203a293 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1008 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | ret += bdev->bd_inode->i_mapping->nrpages; |
| 1010 | } |
| 1011 | spin_unlock(&bdev_lock); |
| 1012 | return ret; |
| 1013 | } |
| 1014 | |
| 1015 | void bdput(struct block_device *bdev) |
| 1016 | { |
| 1017 | iput(bdev->bd_inode); |
| 1018 | } |
| 1019 | |
| 1020 | EXPORT_SYMBOL(bdput); |
| 1021 | |
| 1022 | static struct block_device *bd_acquire(struct inode *inode) |
| 1023 | { |
| 1024 | struct block_device *bdev; |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1025 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | spin_lock(&bdev_lock); |
| 1027 | bdev = inode->i_bdev; |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1028 | if (bdev) { |
Ilya Dryomov | ed8a9d2c | 2015-11-20 22:18:43 +0100 | [diff] [blame] | 1029 | bdgrab(bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1030 | spin_unlock(&bdev_lock); |
| 1031 | return bdev; |
| 1032 | } |
| 1033 | spin_unlock(&bdev_lock); |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1034 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1035 | bdev = bdget(inode->i_rdev); |
| 1036 | if (bdev) { |
| 1037 | spin_lock(&bdev_lock); |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1038 | if (!inode->i_bdev) { |
| 1039 | /* |
Al Viro | 7de9c6ee | 2010-10-23 11:11:40 -0400 | [diff] [blame] | 1040 | * We take an additional reference to bd_inode, |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1041 | * and it's released in clear_inode() of inode. |
| 1042 | * So, we can access it via ->i_mapping always |
| 1043 | * without igrab(). |
| 1044 | */ |
Ilya Dryomov | ed8a9d2c | 2015-11-20 22:18:43 +0100 | [diff] [blame] | 1045 | bdgrab(bdev); |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1046 | inode->i_bdev = bdev; |
| 1047 | inode->i_mapping = bdev->bd_inode->i_mapping; |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1048 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1049 | spin_unlock(&bdev_lock); |
| 1050 | } |
| 1051 | return bdev; |
| 1052 | } |
| 1053 | |
| 1054 | /* Call when you free inode */ |
| 1055 | |
| 1056 | void bd_forget(struct inode *inode) |
| 1057 | { |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1058 | struct block_device *bdev = NULL; |
| 1059 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | spin_lock(&bdev_lock); |
Yan Hong | b4ea2ea | 2013-04-30 15:26:47 -0700 | [diff] [blame] | 1061 | if (!sb_is_blkdev_sb(inode->i_sb)) |
| 1062 | bdev = inode->i_bdev; |
Al Viro | a4a4f94 | 2016-07-19 13:16:52 -0400 | [diff] [blame] | 1063 | inode->i_bdev = NULL; |
| 1064 | inode->i_mapping = &inode->i_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1065 | spin_unlock(&bdev_lock); |
OGAWA Hirofumi | 09d967c | 2006-06-22 14:47:21 -0700 | [diff] [blame] | 1066 | |
| 1067 | if (bdev) |
Ilya Dryomov | ed8a9d2c | 2015-11-20 22:18:43 +0100 | [diff] [blame] | 1068 | bdput(bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | } |
| 1070 | |
Tejun Heo | 1a3cbbc | 2010-04-07 18:52:29 +0900 | [diff] [blame] | 1071 | /** |
| 1072 | * bd_may_claim - test whether a block device can be claimed |
| 1073 | * @bdev: block device of interest |
| 1074 | * @whole: whole block device containing @bdev, may equal @bdev |
| 1075 | * @holder: holder trying to claim @bdev |
| 1076 | * |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1077 | * Test whether @bdev can be claimed by @holder. |
Tejun Heo | 1a3cbbc | 2010-04-07 18:52:29 +0900 | [diff] [blame] | 1078 | * |
| 1079 | * CONTEXT: |
| 1080 | * spin_lock(&bdev_lock). |
| 1081 | * |
| 1082 | * RETURNS: |
| 1083 | * %true if @bdev can be claimed, %false otherwise. |
| 1084 | */ |
| 1085 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, |
| 1086 | void *holder) |
| 1087 | { |
| 1088 | if (bdev->bd_holder == holder) |
| 1089 | return true; /* already a holder */ |
| 1090 | else if (bdev->bd_holder != NULL) |
| 1091 | return false; /* held by someone else */ |
NeilBrown | bcc7f5b | 2016-12-12 08:21:51 -0700 | [diff] [blame] | 1092 | else if (whole == bdev) |
Tejun Heo | 1a3cbbc | 2010-04-07 18:52:29 +0900 | [diff] [blame] | 1093 | return true; /* is a whole device which isn't held */ |
| 1094 | |
Tejun Heo | e525fd8 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1095 | else if (whole->bd_holder == bd_may_claim) |
Tejun Heo | 1a3cbbc | 2010-04-07 18:52:29 +0900 | [diff] [blame] | 1096 | return true; /* is a partition of a device that is being partitioned */ |
| 1097 | else if (whole->bd_holder != NULL) |
| 1098 | return false; /* is a partition of a held device */ |
| 1099 | else |
| 1100 | return true; /* is a partition of an un-held device */ |
| 1101 | } |
| 1102 | |
| 1103 | /** |
Tejun Heo | 6b4517a | 2010-04-07 18:53:59 +0900 | [diff] [blame] | 1104 | * bd_prepare_to_claim - prepare to claim a block device |
| 1105 | * @bdev: block device of interest |
| 1106 | * @whole: the whole device containing @bdev, may equal @bdev |
| 1107 | * @holder: holder trying to claim @bdev |
| 1108 | * |
| 1109 | * Prepare to claim @bdev. This function fails if @bdev is already |
| 1110 | * claimed by another holder and waits if another claiming is in |
| 1111 | * progress. This function doesn't actually claim. On successful |
| 1112 | * return, the caller has ownership of bd_claiming and bd_holder[s]. |
| 1113 | * |
| 1114 | * CONTEXT: |
| 1115 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab |
| 1116 | * it multiple times. |
| 1117 | * |
| 1118 | * RETURNS: |
| 1119 | * 0 if @bdev can be claimed, -EBUSY otherwise. |
| 1120 | */ |
| 1121 | static int bd_prepare_to_claim(struct block_device *bdev, |
| 1122 | struct block_device *whole, void *holder) |
| 1123 | { |
| 1124 | retry: |
| 1125 | /* if someone else claimed, fail */ |
| 1126 | if (!bd_may_claim(bdev, whole, holder)) |
| 1127 | return -EBUSY; |
| 1128 | |
Tejun Heo | e75aa85 | 2010-08-04 17:59:39 +0200 | [diff] [blame] | 1129 | /* if claiming is already in progress, wait for it to finish */ |
| 1130 | if (whole->bd_claiming) { |
Tejun Heo | 6b4517a | 2010-04-07 18:53:59 +0900 | [diff] [blame] | 1131 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
| 1132 | DEFINE_WAIT(wait); |
| 1133 | |
| 1134 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); |
| 1135 | spin_unlock(&bdev_lock); |
| 1136 | schedule(); |
| 1137 | finish_wait(wq, &wait); |
| 1138 | spin_lock(&bdev_lock); |
| 1139 | goto retry; |
| 1140 | } |
| 1141 | |
| 1142 | /* yay, all mine */ |
| 1143 | return 0; |
| 1144 | } |
| 1145 | |
| 1146 | /** |
| 1147 | * bd_start_claiming - start claiming a block device |
| 1148 | * @bdev: block device of interest |
| 1149 | * @holder: holder trying to claim @bdev |
| 1150 | * |
| 1151 | * @bdev is about to be opened exclusively. Check @bdev can be opened |
| 1152 | * exclusively and mark that an exclusive open is in progress. Each |
| 1153 | * successful call to this function must be matched with a call to |
Nick Piggin | b001836 | 2010-05-26 01:51:19 +1000 | [diff] [blame] | 1154 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
| 1155 | * fail). |
| 1156 | * |
| 1157 | * This function is used to gain exclusive access to the block device |
| 1158 | * without actually causing other exclusive open attempts to fail. It |
| 1159 | * should be used when the open sequence itself requires exclusive |
| 1160 | * access but may subsequently fail. |
Tejun Heo | 6b4517a | 2010-04-07 18:53:59 +0900 | [diff] [blame] | 1161 | * |
| 1162 | * CONTEXT: |
| 1163 | * Might sleep. |
| 1164 | * |
| 1165 | * RETURNS: |
| 1166 | * Pointer to the block device containing @bdev on success, ERR_PTR() |
| 1167 | * value on failure. |
| 1168 | */ |
| 1169 | static struct block_device *bd_start_claiming(struct block_device *bdev, |
| 1170 | void *holder) |
| 1171 | { |
| 1172 | struct gendisk *disk; |
| 1173 | struct block_device *whole; |
| 1174 | int partno, err; |
| 1175 | |
| 1176 | might_sleep(); |
| 1177 | |
| 1178 | /* |
| 1179 | * @bdev might not have been initialized properly yet, look up |
| 1180 | * and grab the outer block device the hard way. |
| 1181 | */ |
| 1182 | disk = get_gendisk(bdev->bd_dev, &partno); |
| 1183 | if (!disk) |
| 1184 | return ERR_PTR(-ENXIO); |
| 1185 | |
Tejun Heo | d4c208b | 2011-06-13 12:45:48 +0200 | [diff] [blame] | 1186 | /* |
| 1187 | * Normally, @bdev should equal what's returned from bdget_disk() |
| 1188 | * if partno is 0; however, some drivers (floppy) use multiple |
| 1189 | * bdev's for the same physical device and @bdev may be one of the |
| 1190 | * aliases. Keep @bdev if partno is 0. This means claimer |
| 1191 | * tracking is broken for those devices but it has always been that |
| 1192 | * way. |
| 1193 | */ |
| 1194 | if (partno) |
| 1195 | whole = bdget_disk(disk, 0); |
| 1196 | else |
| 1197 | whole = bdgrab(bdev); |
| 1198 | |
Nick Piggin | cf34257 | 2010-05-26 01:50:21 +1000 | [diff] [blame] | 1199 | module_put(disk->fops->owner); |
Tejun Heo | 6b4517a | 2010-04-07 18:53:59 +0900 | [diff] [blame] | 1200 | put_disk(disk); |
| 1201 | if (!whole) |
| 1202 | return ERR_PTR(-ENOMEM); |
| 1203 | |
| 1204 | /* prepare to claim, if successful, mark claiming in progress */ |
| 1205 | spin_lock(&bdev_lock); |
| 1206 | |
| 1207 | err = bd_prepare_to_claim(bdev, whole, holder); |
| 1208 | if (err == 0) { |
| 1209 | whole->bd_claiming = holder; |
| 1210 | spin_unlock(&bdev_lock); |
| 1211 | return whole; |
| 1212 | } else { |
| 1213 | spin_unlock(&bdev_lock); |
| 1214 | bdput(whole); |
| 1215 | return ERR_PTR(err); |
| 1216 | } |
| 1217 | } |
| 1218 | |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1219 | #ifdef CONFIG_SYSFS |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1220 | struct bd_holder_disk { |
| 1221 | struct list_head list; |
| 1222 | struct gendisk *disk; |
| 1223 | int refcnt; |
| 1224 | }; |
| 1225 | |
| 1226 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, |
| 1227 | struct gendisk *disk) |
| 1228 | { |
| 1229 | struct bd_holder_disk *holder; |
| 1230 | |
| 1231 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) |
| 1232 | if (holder->disk == disk) |
| 1233 | return holder; |
| 1234 | return NULL; |
| 1235 | } |
| 1236 | |
Andrew Morton | 4d7dd8fd | 2006-09-29 01:58:56 -0700 | [diff] [blame] | 1237 | static int add_symlink(struct kobject *from, struct kobject *to) |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1238 | { |
Andrew Morton | 4d7dd8fd | 2006-09-29 01:58:56 -0700 | [diff] [blame] | 1239 | return sysfs_create_link(from, to, kobject_name(to)); |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1240 | } |
| 1241 | |
| 1242 | static void del_symlink(struct kobject *from, struct kobject *to) |
| 1243 | { |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1244 | sysfs_remove_link(from, kobject_name(to)); |
| 1245 | } |
| 1246 | |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1247 | /** |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1248 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
| 1249 | * @bdev: the claimed slave bdev |
| 1250 | * @disk: the holding disk |
Jun'ichi Nomura | df6c0cd | 2006-10-30 16:23:56 -0500 | [diff] [blame] | 1251 | * |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1252 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
| 1253 | * |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1254 | * This functions creates the following sysfs symlinks. |
Jun'ichi Nomura | df6c0cd | 2006-10-30 16:23:56 -0500 | [diff] [blame] | 1255 | * |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1256 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
| 1257 | * - from "holders" directory of the @bdev to the holder @disk |
| 1258 | * |
| 1259 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
| 1260 | * passed to bd_link_disk_holder(), then: |
| 1261 | * |
| 1262 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
| 1263 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 |
| 1264 | * |
| 1265 | * The caller must have claimed @bdev before calling this function and |
| 1266 | * ensure that both @bdev and @disk are valid during the creation and |
| 1267 | * lifetime of these symlinks. |
| 1268 | * |
| 1269 | * CONTEXT: |
| 1270 | * Might sleep. |
| 1271 | * |
| 1272 | * RETURNS: |
| 1273 | * 0 on success, -errno on failure. |
Jun'ichi Nomura | df6c0cd | 2006-10-30 16:23:56 -0500 | [diff] [blame] | 1274 | */ |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1275 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
Jun'ichi Nomura | df6c0cd | 2006-10-30 16:23:56 -0500 | [diff] [blame] | 1276 | { |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1277 | struct bd_holder_disk *holder; |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1278 | int ret = 0; |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1279 | |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1280 | mutex_lock(&bdev->bd_mutex); |
Jun'ichi Nomura | df6c0cd | 2006-10-30 16:23:56 -0500 | [diff] [blame] | 1281 | |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1282 | WARN_ON_ONCE(!bdev->bd_holder); |
Johannes Weiner | 4e91672 | 2007-07-15 23:41:25 -0700 | [diff] [blame] | 1283 | |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1284 | /* FIXME: remove the following once add_disk() handles errors */ |
| 1285 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) |
| 1286 | goto out_unlock; |
Johannes Weiner | 4e91672 | 2007-07-15 23:41:25 -0700 | [diff] [blame] | 1287 | |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1288 | holder = bd_find_holder_disk(bdev, disk); |
| 1289 | if (holder) { |
| 1290 | holder->refcnt++; |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1291 | goto out_unlock; |
| 1292 | } |
| 1293 | |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1294 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
| 1295 | if (!holder) { |
| 1296 | ret = -ENOMEM; |
| 1297 | goto out_unlock; |
| 1298 | } |
| 1299 | |
| 1300 | INIT_LIST_HEAD(&holder->list); |
| 1301 | holder->disk = disk; |
| 1302 | holder->refcnt = 1; |
| 1303 | |
| 1304 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 1305 | if (ret) |
| 1306 | goto out_free; |
| 1307 | |
| 1308 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); |
| 1309 | if (ret) |
| 1310 | goto out_del; |
Tejun Heo | e7407d1 | 2011-02-24 09:56:32 +0100 | [diff] [blame] | 1311 | /* |
| 1312 | * bdev could be deleted beneath us which would implicitly destroy |
| 1313 | * the holder directory. Hold on to it. |
| 1314 | */ |
| 1315 | kobject_get(bdev->bd_part->holder_dir); |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1316 | |
| 1317 | list_add(&holder->list, &bdev->bd_holder_disks); |
| 1318 | goto out_unlock; |
| 1319 | |
| 1320 | out_del: |
| 1321 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 1322 | out_free: |
| 1323 | kfree(holder); |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1324 | out_unlock: |
Jun'ichi Nomura | b4cf1b7 | 2006-03-27 01:18:00 -0800 | [diff] [blame] | 1325 | mutex_unlock(&bdev->bd_mutex); |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1326 | return ret; |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1327 | } |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1328 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1329 | |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1330 | /** |
| 1331 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() |
| 1332 | * @bdev: the calimed slave bdev |
| 1333 | * @disk: the holding disk |
| 1334 | * |
| 1335 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
| 1336 | * |
| 1337 | * CONTEXT: |
| 1338 | * Might sleep. |
| 1339 | */ |
| 1340 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1341 | { |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1342 | struct bd_holder_disk *holder; |
Tejun Heo | e09b457 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1343 | |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1344 | mutex_lock(&bdev->bd_mutex); |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1345 | |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1346 | holder = bd_find_holder_disk(bdev, disk); |
| 1347 | |
| 1348 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { |
| 1349 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 1350 | del_symlink(bdev->bd_part->holder_dir, |
| 1351 | &disk_to_dev(disk)->kobj); |
Tejun Heo | e7407d1 | 2011-02-24 09:56:32 +0100 | [diff] [blame] | 1352 | kobject_put(bdev->bd_part->holder_dir); |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1353 | list_del_init(&holder->list); |
| 1354 | kfree(holder); |
| 1355 | } |
| 1356 | |
| 1357 | mutex_unlock(&bdev->bd_mutex); |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1358 | } |
Tejun Heo | 49731ba | 2011-01-14 18:43:57 +0100 | [diff] [blame] | 1359 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
Jun'ichi Nomura | 641dc63 | 2006-03-27 01:17:57 -0800 | [diff] [blame] | 1360 | #endif |
| 1361 | |
Andrew Patterson | 0c002c2 | 2008-09-04 14:27:20 -0600 | [diff] [blame] | 1362 | /** |
Andrew Patterson | 56ade44 | 2008-09-04 14:27:40 -0600 | [diff] [blame] | 1363 | * flush_disk - invalidates all buffer-cache entries on a disk |
| 1364 | * |
| 1365 | * @bdev: struct block device to be flushed |
Randy Dunlap | e6eb5ce | 2011-02-26 10:54:00 -0800 | [diff] [blame] | 1366 | * @kill_dirty: flag to guide handling of dirty inodes |
Andrew Patterson | 56ade44 | 2008-09-04 14:27:40 -0600 | [diff] [blame] | 1367 | * |
| 1368 | * Invalidates all buffer-cache entries on a disk. It should be called |
| 1369 | * when a disk has been changed -- either by a media change or online |
| 1370 | * resize. |
| 1371 | */ |
NeilBrown | 93b270f | 2011-02-24 17:25:47 +1100 | [diff] [blame] | 1372 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
Andrew Patterson | 56ade44 | 2008-09-04 14:27:40 -0600 | [diff] [blame] | 1373 | { |
NeilBrown | 93b270f | 2011-02-24 17:25:47 +1100 | [diff] [blame] | 1374 | if (__invalidate_device(bdev, kill_dirty)) { |
Andrew Patterson | 56ade44 | 2008-09-04 14:27:40 -0600 | [diff] [blame] | 1375 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
Dmitry Monakhov | 424081f | 2015-04-13 16:31:34 +0400 | [diff] [blame] | 1376 | "resized disk %s\n", |
| 1377 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); |
Andrew Patterson | 56ade44 | 2008-09-04 14:27:40 -0600 | [diff] [blame] | 1378 | } |
| 1379 | |
| 1380 | if (!bdev->bd_disk) |
| 1381 | return; |
Tejun Heo | d27769e | 2011-08-23 20:01:04 +0200 | [diff] [blame] | 1382 | if (disk_part_scan_enabled(bdev->bd_disk)) |
Andrew Patterson | 56ade44 | 2008-09-04 14:27:40 -0600 | [diff] [blame] | 1383 | bdev->bd_invalidated = 1; |
| 1384 | } |
| 1385 | |
| 1386 | /** |
Randy Dunlap | 57d1b53 | 2008-10-09 10:42:38 +0200 | [diff] [blame] | 1387 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
Andrew Patterson | c3279d1 | 2008-09-04 14:27:25 -0600 | [diff] [blame] | 1388 | * @disk: struct gendisk to check |
| 1389 | * @bdev: struct bdev to adjust. |
| 1390 | * |
| 1391 | * This routine checks to see if the bdev size does not match the disk size |
| 1392 | * and adjusts it if it differs. |
| 1393 | */ |
| 1394 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) |
| 1395 | { |
| 1396 | loff_t disk_size, bdev_size; |
| 1397 | |
| 1398 | disk_size = (loff_t)get_capacity(disk) << 9; |
| 1399 | bdev_size = i_size_read(bdev->bd_inode); |
| 1400 | if (disk_size != bdev_size) { |
Andrew Patterson | c3279d1 | 2008-09-04 14:27:25 -0600 | [diff] [blame] | 1401 | printk(KERN_INFO |
| 1402 | "%s: detected capacity change from %lld to %lld\n", |
Dmitry Monakhov | 424081f | 2015-04-13 16:31:34 +0400 | [diff] [blame] | 1403 | disk->disk_name, bdev_size, disk_size); |
Andrew Patterson | c3279d1 | 2008-09-04 14:27:25 -0600 | [diff] [blame] | 1404 | i_size_write(bdev->bd_inode, disk_size); |
NeilBrown | 93b270f | 2011-02-24 17:25:47 +1100 | [diff] [blame] | 1405 | flush_disk(bdev, false); |
Andrew Patterson | c3279d1 | 2008-09-04 14:27:25 -0600 | [diff] [blame] | 1406 | } |
| 1407 | } |
| 1408 | EXPORT_SYMBOL(check_disk_size_change); |
| 1409 | |
| 1410 | /** |
Randy Dunlap | 57d1b53 | 2008-10-09 10:42:38 +0200 | [diff] [blame] | 1411 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
Andrew Patterson | 0c002c2 | 2008-09-04 14:27:20 -0600 | [diff] [blame] | 1412 | * @disk: struct gendisk to be revalidated |
| 1413 | * |
| 1414 | * This routine is a wrapper for lower-level driver's revalidate_disk |
| 1415 | * call-backs. It is used to do common pre and post operations needed |
| 1416 | * for all revalidate_disk operations. |
| 1417 | */ |
| 1418 | int revalidate_disk(struct gendisk *disk) |
| 1419 | { |
Andrew Patterson | c3279d1 | 2008-09-04 14:27:25 -0600 | [diff] [blame] | 1420 | struct block_device *bdev; |
Andrew Patterson | 0c002c2 | 2008-09-04 14:27:20 -0600 | [diff] [blame] | 1421 | int ret = 0; |
| 1422 | |
| 1423 | if (disk->fops->revalidate_disk) |
| 1424 | ret = disk->fops->revalidate_disk(disk); |
Martin K. Petersen | 25520d5 | 2015-10-21 13:19:49 -0400 | [diff] [blame] | 1425 | blk_integrity_revalidate(disk); |
Andrew Patterson | c3279d1 | 2008-09-04 14:27:25 -0600 | [diff] [blame] | 1426 | bdev = bdget_disk(disk, 0); |
| 1427 | if (!bdev) |
| 1428 | return ret; |
| 1429 | |
| 1430 | mutex_lock(&bdev->bd_mutex); |
| 1431 | check_disk_size_change(disk, bdev); |
MITSUNARI Shigeo | 7630b66 | 2013-02-21 16:42:01 -0800 | [diff] [blame] | 1432 | bdev->bd_invalidated = 0; |
Andrew Patterson | c3279d1 | 2008-09-04 14:27:25 -0600 | [diff] [blame] | 1433 | mutex_unlock(&bdev->bd_mutex); |
| 1434 | bdput(bdev); |
Andrew Patterson | 0c002c2 | 2008-09-04 14:27:20 -0600 | [diff] [blame] | 1435 | return ret; |
| 1436 | } |
| 1437 | EXPORT_SYMBOL(revalidate_disk); |
| 1438 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1439 | /* |
| 1440 | * This routine checks whether a removable media has been changed, |
| 1441 | * and invalidates all buffer-cache-entries in that case. This |
| 1442 | * is a relatively slow routine, so we have to try to minimize using |
| 1443 | * it. Thus it is called only upon a 'mount' or 'open'. This |
| 1444 | * is the best way of combining speed and utility, I think. |
| 1445 | * People changing diskettes in the middle of an operation deserve |
| 1446 | * to lose :-) |
| 1447 | */ |
| 1448 | int check_disk_change(struct block_device *bdev) |
| 1449 | { |
| 1450 | struct gendisk *disk = bdev->bd_disk; |
Alexey Dobriyan | 83d5cde | 2009-09-21 17:01:13 -0700 | [diff] [blame] | 1451 | const struct block_device_operations *bdops = disk->fops; |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1452 | unsigned int events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1454 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
| 1455 | DISK_EVENT_EJECT_REQUEST); |
| 1456 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1457 | return 0; |
| 1458 | |
NeilBrown | 93b270f | 2011-02-24 17:25:47 +1100 | [diff] [blame] | 1459 | flush_disk(bdev, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1460 | if (bdops->revalidate_disk) |
| 1461 | bdops->revalidate_disk(bdev->bd_disk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1462 | return 1; |
| 1463 | } |
| 1464 | |
| 1465 | EXPORT_SYMBOL(check_disk_change); |
| 1466 | |
| 1467 | void bd_set_size(struct block_device *bdev, loff_t size) |
| 1468 | { |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 1469 | unsigned bsize = bdev_logical_block_size(bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1470 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1471 | inode_lock(bdev->bd_inode); |
Guo Chao | d646a02 | 2013-02-21 15:16:42 -0800 | [diff] [blame] | 1472 | i_size_write(bdev->bd_inode, size); |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1473 | inode_unlock(bdev->bd_inode); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1474 | while (bsize < PAGE_SIZE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1475 | if (size & bsize) |
| 1476 | break; |
| 1477 | bsize <<= 1; |
| 1478 | } |
| 1479 | bdev->bd_block_size = bsize; |
| 1480 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); |
| 1481 | } |
| 1482 | EXPORT_SYMBOL(bd_set_size); |
| 1483 | |
Al Viro | 4385bab | 2013-05-05 22:11:03 -0400 | [diff] [blame] | 1484 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1485 | |
Peter Zijlstra | 6d740cd | 2007-02-20 13:58:18 -0800 | [diff] [blame] | 1486 | /* |
| 1487 | * bd_mutex locking: |
| 1488 | * |
| 1489 | * mutex_lock(part->bd_mutex) |
| 1490 | * mutex_lock_nested(whole->bd_mutex, 1) |
| 1491 | */ |
| 1492 | |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1493 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1494 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | struct gendisk *disk; |
Tejun Heo | 523e1d3 | 2011-10-19 14:31:07 +0200 | [diff] [blame] | 1496 | struct module *owner; |
Pavel Emelyanov | 7db9cfd | 2008-06-05 22:46:27 -0700 | [diff] [blame] | 1497 | int ret; |
Tejun Heo | cf771cb | 2008-09-03 09:01:09 +0200 | [diff] [blame] | 1498 | int partno; |
Al Viro | fe6e9c1 | 2008-06-23 08:30:55 -0400 | [diff] [blame] | 1499 | int perm = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1501 | if (mode & FMODE_READ) |
Al Viro | fe6e9c1 | 2008-06-23 08:30:55 -0400 | [diff] [blame] | 1502 | perm |= MAY_READ; |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1503 | if (mode & FMODE_WRITE) |
Al Viro | fe6e9c1 | 2008-06-23 08:30:55 -0400 | [diff] [blame] | 1504 | perm |= MAY_WRITE; |
| 1505 | /* |
| 1506 | * hooks: /n/, see "layering violations". |
| 1507 | */ |
Chris Wright | b7300b7 | 2010-08-10 18:02:55 -0700 | [diff] [blame] | 1508 | if (!for_part) { |
| 1509 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); |
| 1510 | if (ret != 0) { |
| 1511 | bdput(bdev); |
| 1512 | return ret; |
| 1513 | } |
Al Viro | 8266602 | 2008-08-01 05:32:04 -0400 | [diff] [blame] | 1514 | } |
Pavel Emelyanov | 7db9cfd | 2008-06-05 22:46:27 -0700 | [diff] [blame] | 1515 | |
NeilBrown | d337482 | 2009-01-09 08:31:10 +1100 | [diff] [blame] | 1516 | restart: |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1517 | |
Tejun Heo | 89f9749 | 2008-11-05 10:21:06 +0100 | [diff] [blame] | 1518 | ret = -ENXIO; |
Tejun Heo | cf771cb | 2008-09-03 09:01:09 +0200 | [diff] [blame] | 1519 | disk = get_gendisk(bdev->bd_dev, &partno); |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1520 | if (!disk) |
Arnd Bergmann | 6e9624b | 2010-08-07 18:25:34 +0200 | [diff] [blame] | 1521 | goto out; |
Tejun Heo | 523e1d3 | 2011-10-19 14:31:07 +0200 | [diff] [blame] | 1522 | owner = disk->fops->owner; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1523 | |
Tejun Heo | 69e02c5 | 2011-03-09 19:54:27 +0100 | [diff] [blame] | 1524 | disk_block_events(disk); |
NeilBrown | 6796bf5 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1525 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1526 | if (!bdev->bd_openers) { |
| 1527 | bdev->bd_disk = disk; |
Andi Kleen | 87192a2 | 2012-01-12 17:20:34 -0800 | [diff] [blame] | 1528 | bdev->bd_queue = disk->queue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1529 | bdev->bd_contains = bdev; |
Dan Williams | 03cdadb | 2016-02-26 15:19:43 -0800 | [diff] [blame] | 1530 | |
Tejun Heo | cf771cb | 2008-09-03 09:01:09 +0200 | [diff] [blame] | 1531 | if (!partno) { |
Tejun Heo | 89f9749 | 2008-11-05 10:21:06 +0100 | [diff] [blame] | 1532 | ret = -ENXIO; |
| 1533 | bdev->bd_part = disk_get_part(disk, partno); |
| 1534 | if (!bdev->bd_part) |
| 1535 | goto out_clear; |
| 1536 | |
Tejun Heo | 1196f8b | 2011-04-21 20:54:45 +0200 | [diff] [blame] | 1537 | ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1538 | if (disk->fops->open) { |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1539 | ret = disk->fops->open(bdev, mode); |
NeilBrown | d337482 | 2009-01-09 08:31:10 +1100 | [diff] [blame] | 1540 | if (ret == -ERESTARTSYS) { |
| 1541 | /* Lost a race with 'disk' being |
| 1542 | * deleted, try again. |
| 1543 | * See md.c |
| 1544 | */ |
| 1545 | disk_put_part(bdev->bd_part); |
| 1546 | bdev->bd_part = NULL; |
NeilBrown | d337482 | 2009-01-09 08:31:10 +1100 | [diff] [blame] | 1547 | bdev->bd_disk = NULL; |
Andi Kleen | 87192a2 | 2012-01-12 17:20:34 -0800 | [diff] [blame] | 1548 | bdev->bd_queue = NULL; |
NeilBrown | d337482 | 2009-01-09 08:31:10 +1100 | [diff] [blame] | 1549 | mutex_unlock(&bdev->bd_mutex); |
Tejun Heo | 69e02c5 | 2011-03-09 19:54:27 +0100 | [diff] [blame] | 1550 | disk_unblock_events(disk); |
Tejun Heo | 69e02c5 | 2011-03-09 19:54:27 +0100 | [diff] [blame] | 1551 | put_disk(disk); |
Tejun Heo | 523e1d3 | 2011-10-19 14:31:07 +0200 | [diff] [blame] | 1552 | module_put(owner); |
NeilBrown | d337482 | 2009-01-09 08:31:10 +1100 | [diff] [blame] | 1553 | goto restart; |
| 1554 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1555 | } |
Tejun Heo | 7e69723 | 2011-05-23 13:26:07 +0200 | [diff] [blame] | 1556 | |
Christoph Hellwig | 2237570 | 2016-09-14 11:56:13 +0200 | [diff] [blame] | 1557 | if (!ret) |
Tejun Heo | 7e69723 | 2011-05-23 13:26:07 +0200 | [diff] [blame] | 1558 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
Tejun Heo | 7e69723 | 2011-05-23 13:26:07 +0200 | [diff] [blame] | 1559 | |
Tejun Heo | 1196f8b | 2011-04-21 20:54:45 +0200 | [diff] [blame] | 1560 | /* |
| 1561 | * If the device is invalidated, rescan partition |
| 1562 | * if open succeeded or failed with -ENOMEDIUM. |
| 1563 | * The latter is necessary to prevent ghost |
| 1564 | * partitions on a removed medium. |
| 1565 | */ |
Jun'ichi Nomura | fe316bf | 2012-03-02 10:38:33 +0100 | [diff] [blame] | 1566 | if (bdev->bd_invalidated) { |
| 1567 | if (!ret) |
| 1568 | rescan_partitions(disk, bdev); |
| 1569 | else if (ret == -ENOMEDIUM) |
| 1570 | invalidate_partitions(disk, bdev); |
| 1571 | } |
Dan Williams | 5a023cd | 2015-11-30 10:20:29 -0800 | [diff] [blame] | 1572 | |
Tejun Heo | 1196f8b | 2011-04-21 20:54:45 +0200 | [diff] [blame] | 1573 | if (ret) |
| 1574 | goto out_clear; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1575 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1576 | struct block_device *whole; |
| 1577 | whole = bdget_disk(disk, 0); |
| 1578 | ret = -ENOMEM; |
| 1579 | if (!whole) |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1580 | goto out_clear; |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1581 | BUG_ON(for_part); |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1582 | ret = __blkdev_get(whole, mode, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1583 | if (ret) |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1584 | goto out_clear; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1585 | bdev->bd_contains = whole; |
Tejun Heo | 89f9749 | 2008-11-05 10:21:06 +0100 | [diff] [blame] | 1586 | bdev->bd_part = disk_get_part(disk, partno); |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 1587 | if (!(disk->flags & GENHD_FL_UP) || |
Tejun Heo | 89f9749 | 2008-11-05 10:21:06 +0100 | [diff] [blame] | 1588 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1589 | ret = -ENXIO; |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1590 | goto out_clear; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1591 | } |
Tejun Heo | 89f9749 | 2008-11-05 10:21:06 +0100 | [diff] [blame] | 1592 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1593 | } |
| 1594 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | if (bdev->bd_contains == bdev) { |
Tejun Heo | 1196f8b | 2011-04-21 20:54:45 +0200 | [diff] [blame] | 1596 | ret = 0; |
| 1597 | if (bdev->bd_disk->fops->open) |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1598 | ret = bdev->bd_disk->fops->open(bdev, mode); |
Tejun Heo | 1196f8b | 2011-04-21 20:54:45 +0200 | [diff] [blame] | 1599 | /* the same as first opener case, read comment there */ |
Jun'ichi Nomura | fe316bf | 2012-03-02 10:38:33 +0100 | [diff] [blame] | 1600 | if (bdev->bd_invalidated) { |
| 1601 | if (!ret) |
| 1602 | rescan_partitions(bdev->bd_disk, bdev); |
| 1603 | else if (ret == -ENOMEDIUM) |
| 1604 | invalidate_partitions(bdev->bd_disk, bdev); |
| 1605 | } |
Tejun Heo | 1196f8b | 2011-04-21 20:54:45 +0200 | [diff] [blame] | 1606 | if (ret) |
| 1607 | goto out_unlock_bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1608 | } |
Tejun Heo | 69e02c5 | 2011-03-09 19:54:27 +0100 | [diff] [blame] | 1609 | /* only one opener holds refs to the module and disk */ |
Tejun Heo | 69e02c5 | 2011-03-09 19:54:27 +0100 | [diff] [blame] | 1610 | put_disk(disk); |
Tejun Heo | 523e1d3 | 2011-10-19 14:31:07 +0200 | [diff] [blame] | 1611 | module_put(owner); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1612 | } |
| 1613 | bdev->bd_openers++; |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1614 | if (for_part) |
| 1615 | bdev->bd_part_count++; |
Arjan van de Ven | c039e31 | 2006-03-23 03:00:28 -0800 | [diff] [blame] | 1616 | mutex_unlock(&bdev->bd_mutex); |
Tejun Heo | 69e02c5 | 2011-03-09 19:54:27 +0100 | [diff] [blame] | 1617 | disk_unblock_events(disk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1618 | return 0; |
| 1619 | |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1620 | out_clear: |
Tejun Heo | 89f9749 | 2008-11-05 10:21:06 +0100 | [diff] [blame] | 1621 | disk_put_part(bdev->bd_part); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1622 | bdev->bd_disk = NULL; |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1623 | bdev->bd_part = NULL; |
Andi Kleen | 87192a2 | 2012-01-12 17:20:34 -0800 | [diff] [blame] | 1624 | bdev->bd_queue = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1625 | if (bdev != bdev->bd_contains) |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1626 | __blkdev_put(bdev->bd_contains, mode, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1627 | bdev->bd_contains = NULL; |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1628 | out_unlock_bdev: |
Arjan van de Ven | c039e31 | 2006-03-23 03:00:28 -0800 | [diff] [blame] | 1629 | mutex_unlock(&bdev->bd_mutex); |
Tejun Heo | 69e02c5 | 2011-03-09 19:54:27 +0100 | [diff] [blame] | 1630 | disk_unblock_events(disk); |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1631 | put_disk(disk); |
Tejun Heo | 523e1d3 | 2011-10-19 14:31:07 +0200 | [diff] [blame] | 1632 | module_put(owner); |
Dan Carpenter | 4345cab | 2011-03-19 13:53:31 +0100 | [diff] [blame] | 1633 | out: |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1634 | bdput(bdev); |
| 1635 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1636 | return ret; |
| 1637 | } |
| 1638 | |
Tejun Heo | d4d7762 | 2010-11-13 11:55:18 +0100 | [diff] [blame] | 1639 | /** |
| 1640 | * blkdev_get - open a block device |
| 1641 | * @bdev: block_device to open |
| 1642 | * @mode: FMODE_* mask |
| 1643 | * @holder: exclusive holder identifier |
| 1644 | * |
| 1645 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is |
| 1646 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL |
| 1647 | * @holder is invalid. Exclusive opens may nest for the same @holder. |
| 1648 | * |
| 1649 | * On success, the reference count of @bdev is unchanged. On failure, |
| 1650 | * @bdev is put. |
| 1651 | * |
| 1652 | * CONTEXT: |
| 1653 | * Might sleep. |
| 1654 | * |
| 1655 | * RETURNS: |
| 1656 | * 0 on success, -errno on failure. |
| 1657 | */ |
Tejun Heo | e525fd8 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1658 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1659 | { |
Tejun Heo | e525fd8 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1660 | struct block_device *whole = NULL; |
| 1661 | int res; |
| 1662 | |
| 1663 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); |
| 1664 | |
| 1665 | if ((mode & FMODE_EXCL) && holder) { |
| 1666 | whole = bd_start_claiming(bdev, holder); |
| 1667 | if (IS_ERR(whole)) { |
| 1668 | bdput(bdev); |
| 1669 | return PTR_ERR(whole); |
| 1670 | } |
| 1671 | } |
| 1672 | |
| 1673 | res = __blkdev_get(bdev, mode, 0); |
| 1674 | |
| 1675 | if (whole) { |
Tejun Heo | d4dc210 | 2011-04-21 20:54:46 +0200 | [diff] [blame] | 1676 | struct gendisk *disk = whole->bd_disk; |
| 1677 | |
Tejun Heo | 6a027ef | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1678 | /* finish claiming */ |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1679 | mutex_lock(&bdev->bd_mutex); |
Tejun Heo | 6a027ef | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1680 | spin_lock(&bdev_lock); |
| 1681 | |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1682 | if (!res) { |
Tejun Heo | 6a027ef | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1683 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
| 1684 | /* |
| 1685 | * Note that for a whole device bd_holders |
| 1686 | * will be incremented twice, and bd_holder |
| 1687 | * will be set to bd_may_claim before being |
| 1688 | * set to holder |
| 1689 | */ |
| 1690 | whole->bd_holders++; |
| 1691 | whole->bd_holder = bd_may_claim; |
| 1692 | bdev->bd_holders++; |
| 1693 | bdev->bd_holder = holder; |
| 1694 | } |
| 1695 | |
| 1696 | /* tell others that we're done */ |
| 1697 | BUG_ON(whole->bd_claiming != holder); |
| 1698 | whole->bd_claiming = NULL; |
| 1699 | wake_up_bit(&whole->bd_claiming, 0); |
| 1700 | |
| 1701 | spin_unlock(&bdev_lock); |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1702 | |
| 1703 | /* |
Tejun Heo | d4dc210 | 2011-04-21 20:54:46 +0200 | [diff] [blame] | 1704 | * Block event polling for write claims if requested. Any |
| 1705 | * write holder makes the write_holder state stick until |
| 1706 | * all are released. This is good enough and tracking |
| 1707 | * individual writeable reference is too fragile given the |
| 1708 | * way @mode is used in blkdev_get/put(). |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1709 | */ |
Tejun Heo | 4c49ff3 | 2011-06-01 08:27:41 +0200 | [diff] [blame] | 1710 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
| 1711 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1712 | bdev->bd_write_holder = true; |
Tejun Heo | d4dc210 | 2011-04-21 20:54:46 +0200 | [diff] [blame] | 1713 | disk_block_events(disk); |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1714 | } |
| 1715 | |
| 1716 | mutex_unlock(&bdev->bd_mutex); |
Tejun Heo | 6a027ef | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1717 | bdput(whole); |
Tejun Heo | e525fd8 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1718 | } |
| 1719 | |
| 1720 | return res; |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1721 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1722 | EXPORT_SYMBOL(blkdev_get); |
| 1723 | |
Tejun Heo | d4d7762 | 2010-11-13 11:55:18 +0100 | [diff] [blame] | 1724 | /** |
| 1725 | * blkdev_get_by_path - open a block device by name |
| 1726 | * @path: path to the block device to open |
| 1727 | * @mode: FMODE_* mask |
| 1728 | * @holder: exclusive holder identifier |
| 1729 | * |
| 1730 | * Open the blockdevice described by the device file at @path. @mode |
| 1731 | * and @holder are identical to blkdev_get(). |
| 1732 | * |
| 1733 | * On success, the returned block_device has reference count of one. |
| 1734 | * |
| 1735 | * CONTEXT: |
| 1736 | * Might sleep. |
| 1737 | * |
| 1738 | * RETURNS: |
| 1739 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
| 1740 | */ |
| 1741 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, |
| 1742 | void *holder) |
| 1743 | { |
| 1744 | struct block_device *bdev; |
| 1745 | int err; |
| 1746 | |
| 1747 | bdev = lookup_bdev(path); |
| 1748 | if (IS_ERR(bdev)) |
| 1749 | return bdev; |
| 1750 | |
| 1751 | err = blkdev_get(bdev, mode, holder); |
| 1752 | if (err) |
| 1753 | return ERR_PTR(err); |
| 1754 | |
Chuck Ebbert | e51900f | 2011-02-16 18:11:53 -0500 | [diff] [blame] | 1755 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
| 1756 | blkdev_put(bdev, mode); |
| 1757 | return ERR_PTR(-EACCES); |
| 1758 | } |
| 1759 | |
Tejun Heo | d4d7762 | 2010-11-13 11:55:18 +0100 | [diff] [blame] | 1760 | return bdev; |
| 1761 | } |
| 1762 | EXPORT_SYMBOL(blkdev_get_by_path); |
| 1763 | |
| 1764 | /** |
| 1765 | * blkdev_get_by_dev - open a block device by device number |
| 1766 | * @dev: device number of block device to open |
| 1767 | * @mode: FMODE_* mask |
| 1768 | * @holder: exclusive holder identifier |
| 1769 | * |
| 1770 | * Open the blockdevice described by device number @dev. @mode and |
| 1771 | * @holder are identical to blkdev_get(). |
| 1772 | * |
| 1773 | * Use it ONLY if you really do not have anything better - i.e. when |
| 1774 | * you are behind a truly sucky interface and all you are given is a |
| 1775 | * device number. _Never_ to be used for internal purposes. If you |
| 1776 | * ever need it - reconsider your API. |
| 1777 | * |
| 1778 | * On success, the returned block_device has reference count of one. |
| 1779 | * |
| 1780 | * CONTEXT: |
| 1781 | * Might sleep. |
| 1782 | * |
| 1783 | * RETURNS: |
| 1784 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
| 1785 | */ |
| 1786 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) |
| 1787 | { |
| 1788 | struct block_device *bdev; |
| 1789 | int err; |
| 1790 | |
| 1791 | bdev = bdget(dev); |
| 1792 | if (!bdev) |
| 1793 | return ERR_PTR(-ENOMEM); |
| 1794 | |
| 1795 | err = blkdev_get(bdev, mode, holder); |
| 1796 | if (err) |
| 1797 | return ERR_PTR(err); |
| 1798 | |
| 1799 | return bdev; |
| 1800 | } |
| 1801 | EXPORT_SYMBOL(blkdev_get_by_dev); |
| 1802 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1803 | static int blkdev_open(struct inode * inode, struct file * filp) |
| 1804 | { |
| 1805 | struct block_device *bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1806 | |
| 1807 | /* |
| 1808 | * Preserve backwards compatibility and allow large file access |
| 1809 | * even if userspace doesn't ask for it explicitly. Some mkfs |
| 1810 | * binary needs it. We might want to drop this workaround |
| 1811 | * during an unstable branch. |
| 1812 | */ |
| 1813 | filp->f_flags |= O_LARGEFILE; |
| 1814 | |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1815 | if (filp->f_flags & O_NDELAY) |
| 1816 | filp->f_mode |= FMODE_NDELAY; |
| 1817 | if (filp->f_flags & O_EXCL) |
| 1818 | filp->f_mode |= FMODE_EXCL; |
| 1819 | if ((filp->f_flags & O_ACCMODE) == 3) |
| 1820 | filp->f_mode |= FMODE_WRITE_IOCTL; |
| 1821 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1822 | bdev = bd_acquire(inode); |
Pavel Emelianov | 6a2aae0 | 2006-10-28 10:38:33 -0700 | [diff] [blame] | 1823 | if (bdev == NULL) |
| 1824 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1825 | |
Al Viro | 572c489 | 2007-10-08 13:24:05 -0400 | [diff] [blame] | 1826 | filp->f_mapping = bdev->bd_inode->i_mapping; |
| 1827 | |
Tejun Heo | e525fd8 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1828 | return blkdev_get(bdev, filp->f_mode, filp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1829 | } |
| 1830 | |
Al Viro | 4385bab | 2013-05-05 22:11:03 -0400 | [diff] [blame] | 1831 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1832 | { |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1833 | struct gendisk *disk = bdev->bd_disk; |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1834 | struct block_device *victim = NULL; |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1835 | |
NeilBrown | 6796bf5 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1836 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1837 | if (for_part) |
| 1838 | bdev->bd_part_count--; |
| 1839 | |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1840 | if (!--bdev->bd_openers) { |
Tejun Heo | 6a027ef | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1841 | WARN_ON_ONCE(bdev->bd_holders); |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1842 | sync_blockdev(bdev); |
| 1843 | kill_bdev(bdev); |
Ilya Dryomov | 43d1c0e | 2015-11-20 22:22:34 +0100 | [diff] [blame] | 1844 | |
Vivek Goyal | dbd3ca5 | 2015-11-09 09:23:40 -0700 | [diff] [blame] | 1845 | bdev_write_inode(bdev); |
Ilya Dryomov | 43d1c0e | 2015-11-20 22:22:34 +0100 | [diff] [blame] | 1846 | /* |
| 1847 | * Detaching bdev inode from its wb in __destroy_inode() |
| 1848 | * is too late: the queue which embeds its bdi (along with |
| 1849 | * root wb) can be gone as soon as we put_disk() below. |
| 1850 | */ |
| 1851 | inode_detach_wb(bdev->bd_inode); |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1852 | } |
| 1853 | if (bdev->bd_contains == bdev) { |
| 1854 | if (disk->fops->release) |
Al Viro | db2a144 | 2013-05-05 21:52:57 -0400 | [diff] [blame] | 1855 | disk->fops->release(disk, mode); |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1856 | } |
| 1857 | if (!bdev->bd_openers) { |
| 1858 | struct module *owner = disk->fops->owner; |
| 1859 | |
Tejun Heo | 0762b8b | 2008-08-25 19:56:12 +0900 | [diff] [blame] | 1860 | disk_put_part(bdev->bd_part); |
| 1861 | bdev->bd_part = NULL; |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1862 | bdev->bd_disk = NULL; |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1863 | if (bdev != bdev->bd_contains) |
| 1864 | victim = bdev->bd_contains; |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1865 | bdev->bd_contains = NULL; |
Tejun Heo | 523e1d3 | 2011-10-19 14:31:07 +0200 | [diff] [blame] | 1866 | |
| 1867 | put_disk(disk); |
| 1868 | module_put(owner); |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1869 | } |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1870 | mutex_unlock(&bdev->bd_mutex); |
| 1871 | bdput(bdev); |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1872 | if (victim) |
Al Viro | 9a1c354 | 2008-02-22 20:40:24 -0500 | [diff] [blame] | 1873 | __blkdev_put(victim, mode, 1); |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1874 | } |
| 1875 | |
Al Viro | 4385bab | 2013-05-05 22:11:03 -0400 | [diff] [blame] | 1876 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1877 | { |
Tejun Heo | 85ef06d | 2011-07-01 16:17:47 +0200 | [diff] [blame] | 1878 | mutex_lock(&bdev->bd_mutex); |
| 1879 | |
Tejun Heo | e525fd8 | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1880 | if (mode & FMODE_EXCL) { |
Tejun Heo | 6a027ef | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1881 | bool bdev_free; |
| 1882 | |
| 1883 | /* |
| 1884 | * Release a claim on the device. The holder fields |
| 1885 | * are protected with bdev_lock. bd_mutex is to |
| 1886 | * synchronize disk_holder unlinking. |
| 1887 | */ |
Tejun Heo | 6a027ef | 2010-11-13 11:55:17 +0100 | [diff] [blame] | 1888 | spin_lock(&bdev_lock); |
| 1889 | |
| 1890 | WARN_ON_ONCE(--bdev->bd_holders < 0); |
| 1891 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); |
| 1892 | |
| 1893 | /* bd_contains might point to self, check in a separate step */ |
| 1894 | if ((bdev_free = !bdev->bd_holders)) |
| 1895 | bdev->bd_holder = NULL; |
| 1896 | if (!bdev->bd_contains->bd_holders) |
| 1897 | bdev->bd_contains->bd_holder = NULL; |
| 1898 | |
| 1899 | spin_unlock(&bdev_lock); |
| 1900 | |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1901 | /* |
| 1902 | * If this was the last claim, remove holder link and |
| 1903 | * unblock evpoll if it was a write holder. |
| 1904 | */ |
Tejun Heo | 85ef06d | 2011-07-01 16:17:47 +0200 | [diff] [blame] | 1905 | if (bdev_free && bdev->bd_write_holder) { |
| 1906 | disk_unblock_events(bdev->bd_disk); |
| 1907 | bdev->bd_write_holder = false; |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1908 | } |
Tejun Heo | 6936217 | 2011-03-09 19:54:27 +0100 | [diff] [blame] | 1909 | } |
Tejun Heo | 77ea887 | 2010-12-08 20:57:37 +0100 | [diff] [blame] | 1910 | |
Tejun Heo | 85ef06d | 2011-07-01 16:17:47 +0200 | [diff] [blame] | 1911 | /* |
| 1912 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE |
| 1913 | * event. This is to ensure detection of media removal commanded |
| 1914 | * from userland - e.g. eject(1). |
| 1915 | */ |
| 1916 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); |
| 1917 | |
| 1918 | mutex_unlock(&bdev->bd_mutex); |
| 1919 | |
Al Viro | 4385bab | 2013-05-05 22:11:03 -0400 | [diff] [blame] | 1920 | __blkdev_put(bdev, mode, 0); |
NeilBrown | 37be412 | 2006-12-08 02:36:16 -0800 | [diff] [blame] | 1921 | } |
Peter Zijlstra | 2e7b651 | 2006-12-08 02:36:13 -0800 | [diff] [blame] | 1922 | EXPORT_SYMBOL(blkdev_put); |
| 1923 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1924 | static int blkdev_close(struct inode * inode, struct file * filp) |
| 1925 | { |
Dan Williams | 4ebb16c | 2015-10-28 07:48:19 +0900 | [diff] [blame] | 1926 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
Al Viro | 4385bab | 2013-05-05 22:11:03 -0400 | [diff] [blame] | 1927 | blkdev_put(bdev, filp->f_mode); |
| 1928 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1929 | } |
| 1930 | |
Arnd Bergmann | bb93e3a | 2005-06-23 00:10:15 -0700 | [diff] [blame] | 1931 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1932 | { |
Dan Williams | 4ebb16c | 2015-10-28 07:48:19 +0900 | [diff] [blame] | 1933 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
Al Viro | 56b26ad | 2008-09-19 03:17:36 -0400 | [diff] [blame] | 1934 | fmode_t mode = file->f_mode; |
Christoph Hellwig | fd4ce1a | 2008-11-05 14:58:42 +0100 | [diff] [blame] | 1935 | |
| 1936 | /* |
| 1937 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have |
| 1938 | * to updated it before every ioctl. |
| 1939 | */ |
Al Viro | 56b26ad | 2008-09-19 03:17:36 -0400 | [diff] [blame] | 1940 | if (file->f_flags & O_NDELAY) |
Christoph Hellwig | fd4ce1a | 2008-11-05 14:58:42 +0100 | [diff] [blame] | 1941 | mode |= FMODE_NDELAY; |
| 1942 | else |
| 1943 | mode &= ~FMODE_NDELAY; |
| 1944 | |
Al Viro | 56b26ad | 2008-09-19 03:17:36 -0400 | [diff] [blame] | 1945 | return blkdev_ioctl(bdev, mode, cmd, arg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1946 | } |
| 1947 | |
Theodore Ts'o | 87d8fe1 | 2009-01-03 09:47:09 -0500 | [diff] [blame] | 1948 | /* |
Christoph Hellwig | eef9938 | 2009-08-20 17:43:41 +0200 | [diff] [blame] | 1949 | * Write data to the block device. Only intended for the block device itself |
| 1950 | * and the raw driver which basically is a fake block device. |
| 1951 | * |
| 1952 | * Does not take i_mutex for the write and thus is not for general purpose |
| 1953 | * use. |
| 1954 | */ |
Al Viro | 1456c0a | 2014-04-03 03:21:50 -0400 | [diff] [blame] | 1955 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
Christoph Hellwig | eef9938 | 2009-08-20 17:43:41 +0200 | [diff] [blame] | 1956 | { |
| 1957 | struct file *file = iocb->ki_filp; |
Dan Williams | 4ebb16c | 2015-10-28 07:48:19 +0900 | [diff] [blame] | 1958 | struct inode *bd_inode = bdev_file_inode(file); |
Al Viro | 7ec7b94 | 2015-04-07 11:35:14 -0400 | [diff] [blame] | 1959 | loff_t size = i_size_read(bd_inode); |
Jianpeng Ma | 53362a0 | 2012-08-02 09:50:39 +0200 | [diff] [blame] | 1960 | struct blk_plug plug; |
Christoph Hellwig | eef9938 | 2009-08-20 17:43:41 +0200 | [diff] [blame] | 1961 | ssize_t ret; |
Al Viro | 5f380c7 | 2015-04-07 11:28:12 -0400 | [diff] [blame] | 1962 | |
Al Viro | 7ec7b94 | 2015-04-07 11:35:14 -0400 | [diff] [blame] | 1963 | if (bdev_read_only(I_BDEV(bd_inode))) |
| 1964 | return -EPERM; |
Al Viro | 5f380c7 | 2015-04-07 11:28:12 -0400 | [diff] [blame] | 1965 | |
Al Viro | 7ec7b94 | 2015-04-07 11:35:14 -0400 | [diff] [blame] | 1966 | if (!iov_iter_count(from)) |
Al Viro | 5f380c7 | 2015-04-07 11:28:12 -0400 | [diff] [blame] | 1967 | return 0; |
| 1968 | |
Al Viro | 7ec7b94 | 2015-04-07 11:35:14 -0400 | [diff] [blame] | 1969 | if (iocb->ki_pos >= size) |
| 1970 | return -ENOSPC; |
| 1971 | |
| 1972 | iov_iter_truncate(from, size - iocb->ki_pos); |
Christoph Hellwig | eef9938 | 2009-08-20 17:43:41 +0200 | [diff] [blame] | 1973 | |
Jianpeng Ma | 53362a0 | 2012-08-02 09:50:39 +0200 | [diff] [blame] | 1974 | blk_start_plug(&plug); |
Al Viro | 1456c0a | 2014-04-03 03:21:50 -0400 | [diff] [blame] | 1975 | ret = __generic_file_write_iter(iocb, from); |
Christoph Hellwig | e259221 | 2016-04-07 08:52:01 -0700 | [diff] [blame] | 1976 | if (ret > 0) |
| 1977 | ret = generic_write_sync(iocb, ret); |
Jianpeng Ma | 53362a0 | 2012-08-02 09:50:39 +0200 | [diff] [blame] | 1978 | blk_finish_plug(&plug); |
Christoph Hellwig | eef9938 | 2009-08-20 17:43:41 +0200 | [diff] [blame] | 1979 | return ret; |
| 1980 | } |
Al Viro | 1456c0a | 2014-04-03 03:21:50 -0400 | [diff] [blame] | 1981 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
Christoph Hellwig | eef9938 | 2009-08-20 17:43:41 +0200 | [diff] [blame] | 1982 | |
David Jeffery | b2de525 | 2014-09-29 10:21:10 -0400 | [diff] [blame] | 1983 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
Linus Torvalds | 684c9aa | 2012-12-07 16:48:39 -0800 | [diff] [blame] | 1984 | { |
| 1985 | struct file *file = iocb->ki_filp; |
Dan Williams | 4ebb16c | 2015-10-28 07:48:19 +0900 | [diff] [blame] | 1986 | struct inode *bd_inode = bdev_file_inode(file); |
Linus Torvalds | 684c9aa | 2012-12-07 16:48:39 -0800 | [diff] [blame] | 1987 | loff_t size = i_size_read(bd_inode); |
Al Viro | a886038 | 2014-04-02 20:02:21 -0400 | [diff] [blame] | 1988 | loff_t pos = iocb->ki_pos; |
Linus Torvalds | 684c9aa | 2012-12-07 16:48:39 -0800 | [diff] [blame] | 1989 | |
| 1990 | if (pos >= size) |
| 1991 | return 0; |
| 1992 | |
| 1993 | size -= pos; |
Al Viro | a886038 | 2014-04-02 20:02:21 -0400 | [diff] [blame] | 1994 | iov_iter_truncate(to, size); |
| 1995 | return generic_file_read_iter(iocb, to); |
Linus Torvalds | 684c9aa | 2012-12-07 16:48:39 -0800 | [diff] [blame] | 1996 | } |
David Jeffery | b2de525 | 2014-09-29 10:21:10 -0400 | [diff] [blame] | 1997 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
Linus Torvalds | 684c9aa | 2012-12-07 16:48:39 -0800 | [diff] [blame] | 1998 | |
Christoph Hellwig | eef9938 | 2009-08-20 17:43:41 +0200 | [diff] [blame] | 1999 | /* |
Theodore Ts'o | 87d8fe1 | 2009-01-03 09:47:09 -0500 | [diff] [blame] | 2000 | * Try to release a page associated with block device when the system |
| 2001 | * is under memory pressure. |
| 2002 | */ |
| 2003 | static int blkdev_releasepage(struct page *page, gfp_t wait) |
| 2004 | { |
| 2005 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; |
| 2006 | |
| 2007 | if (super && super->s_op->bdev_try_to_free_page) |
| 2008 | return super->s_op->bdev_try_to_free_page(super, page, wait); |
| 2009 | |
| 2010 | return try_to_free_buffers(page); |
| 2011 | } |
| 2012 | |
Ross Zwisler | 7f6d5b5 | 2016-02-26 15:19:55 -0800 | [diff] [blame] | 2013 | static int blkdev_writepages(struct address_space *mapping, |
| 2014 | struct writeback_control *wbc) |
| 2015 | { |
| 2016 | if (dax_mapping(mapping)) { |
| 2017 | struct block_device *bdev = I_BDEV(mapping->host); |
| 2018 | |
| 2019 | return dax_writeback_mapping_range(mapping, bdev, wbc); |
| 2020 | } |
| 2021 | return generic_writepages(mapping, wbc); |
| 2022 | } |
| 2023 | |
Adrian Bunk | 4c54ac6 | 2008-02-18 13:48:31 +0100 | [diff] [blame] | 2024 | static const struct address_space_operations def_blk_aops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2025 | .readpage = blkdev_readpage, |
Akinobu Mita | 447f05b | 2014-10-09 15:26:58 -0700 | [diff] [blame] | 2026 | .readpages = blkdev_readpages, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2027 | .writepage = blkdev_writepage, |
Nick Piggin | 6272b5a | 2007-10-16 01:25:04 -0700 | [diff] [blame] | 2028 | .write_begin = blkdev_write_begin, |
| 2029 | .write_end = blkdev_write_end, |
Ross Zwisler | 7f6d5b5 | 2016-02-26 15:19:55 -0800 | [diff] [blame] | 2030 | .writepages = blkdev_writepages, |
Theodore Ts'o | 87d8fe1 | 2009-01-03 09:47:09 -0500 | [diff] [blame] | 2031 | .releasepage = blkdev_releasepage, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2032 | .direct_IO = blkdev_direct_IO, |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 2033 | .is_dirty_writeback = buffer_check_dirty_writeback, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2034 | }; |
| 2035 | |
Darrick J. Wong | 25f4c41 | 2016-10-11 13:51:11 -0700 | [diff] [blame] | 2036 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
| 2037 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ |
| 2038 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) |
| 2039 | |
| 2040 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, |
| 2041 | loff_t len) |
| 2042 | { |
| 2043 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
| 2044 | struct request_queue *q = bdev_get_queue(bdev); |
| 2045 | struct address_space *mapping; |
| 2046 | loff_t end = start + len - 1; |
| 2047 | loff_t isize; |
| 2048 | int error; |
| 2049 | |
| 2050 | /* Fail if we don't recognize the flags. */ |
| 2051 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) |
| 2052 | return -EOPNOTSUPP; |
| 2053 | |
| 2054 | /* Don't go off the end of the device. */ |
| 2055 | isize = i_size_read(bdev->bd_inode); |
| 2056 | if (start >= isize) |
| 2057 | return -EINVAL; |
| 2058 | if (end >= isize) { |
| 2059 | if (mode & FALLOC_FL_KEEP_SIZE) { |
| 2060 | len = isize - start; |
| 2061 | end = start + len - 1; |
| 2062 | } else |
| 2063 | return -EINVAL; |
| 2064 | } |
| 2065 | |
| 2066 | /* |
| 2067 | * Don't allow IO that isn't aligned to logical block size. |
| 2068 | */ |
| 2069 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) |
| 2070 | return -EINVAL; |
| 2071 | |
| 2072 | /* Invalidate the page cache, including dirty pages. */ |
| 2073 | mapping = bdev->bd_inode->i_mapping; |
| 2074 | truncate_inode_pages_range(mapping, start, end); |
| 2075 | |
| 2076 | switch (mode) { |
| 2077 | case FALLOC_FL_ZERO_RANGE: |
| 2078 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: |
| 2079 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
| 2080 | GFP_KERNEL, false); |
| 2081 | break; |
| 2082 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: |
| 2083 | /* Only punch if the device can do zeroing discard. */ |
| 2084 | if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data) |
| 2085 | return -EOPNOTSUPP; |
| 2086 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
| 2087 | GFP_KERNEL, 0); |
| 2088 | break; |
| 2089 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: |
| 2090 | if (!blk_queue_discard(q)) |
| 2091 | return -EOPNOTSUPP; |
| 2092 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
| 2093 | GFP_KERNEL, 0); |
| 2094 | break; |
| 2095 | default: |
| 2096 | return -EOPNOTSUPP; |
| 2097 | } |
| 2098 | if (error) |
| 2099 | return error; |
| 2100 | |
| 2101 | /* |
| 2102 | * Invalidate again; if someone wandered in and dirtied a page, |
| 2103 | * the caller will be given -EBUSY. The third argument is |
| 2104 | * inclusive, so the rounding here is safe. |
| 2105 | */ |
| 2106 | return invalidate_inode_pages2_range(mapping, |
| 2107 | start >> PAGE_SHIFT, |
| 2108 | end >> PAGE_SHIFT); |
| 2109 | } |
| 2110 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 2111 | const struct file_operations def_blk_fops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2112 | .open = blkdev_open, |
| 2113 | .release = blkdev_close, |
| 2114 | .llseek = block_llseek, |
Al Viro | a886038 | 2014-04-02 20:02:21 -0400 | [diff] [blame] | 2115 | .read_iter = blkdev_read_iter, |
Al Viro | 1456c0a | 2014-04-03 03:21:50 -0400 | [diff] [blame] | 2116 | .write_iter = blkdev_write_iter, |
Dan Williams | acc93d3 | 2016-05-07 11:40:28 -0700 | [diff] [blame] | 2117 | .mmap = generic_file_mmap, |
Andrew Morton | b1dd3b2 | 2010-04-06 14:35:00 -0700 | [diff] [blame] | 2118 | .fsync = blkdev_fsync, |
Arnd Bergmann | bb93e3a | 2005-06-23 00:10:15 -0700 | [diff] [blame] | 2119 | .unlocked_ioctl = block_ioctl, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2120 | #ifdef CONFIG_COMPAT |
| 2121 | .compat_ioctl = compat_blkdev_ioctl, |
| 2122 | #endif |
Linus Torvalds | 1e8b333 | 2012-11-29 10:49:50 -0800 | [diff] [blame] | 2123 | .splice_read = generic_file_splice_read, |
Al Viro | 8d02076 | 2014-04-05 04:27:08 -0400 | [diff] [blame] | 2124 | .splice_write = iter_file_splice_write, |
Darrick J. Wong | 25f4c41 | 2016-10-11 13:51:11 -0700 | [diff] [blame] | 2125 | .fallocate = blkdev_fallocate, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2126 | }; |
| 2127 | |
| 2128 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) |
| 2129 | { |
| 2130 | int res; |
| 2131 | mm_segment_t old_fs = get_fs(); |
| 2132 | set_fs(KERNEL_DS); |
Al Viro | 56b26ad | 2008-09-19 03:17:36 -0400 | [diff] [blame] | 2133 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2134 | set_fs(old_fs); |
| 2135 | return res; |
| 2136 | } |
| 2137 | |
| 2138 | EXPORT_SYMBOL(ioctl_by_bdev); |
| 2139 | |
| 2140 | /** |
| 2141 | * lookup_bdev - lookup a struct block_device by name |
Randy Dunlap | 94e2959 | 2009-01-06 14:41:15 -0800 | [diff] [blame] | 2142 | * @pathname: special file representing the block device |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2143 | * |
Randy Dunlap | 57d1b53 | 2008-10-09 10:42:38 +0200 | [diff] [blame] | 2144 | * Get a reference to the blockdevice at @pathname in the current |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2145 | * namespace if possible and return it. Return ERR_PTR(error) |
| 2146 | * otherwise. |
| 2147 | */ |
Al Viro | 421748e | 2008-08-02 01:04:36 -0400 | [diff] [blame] | 2148 | struct block_device *lookup_bdev(const char *pathname) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2149 | { |
| 2150 | struct block_device *bdev; |
| 2151 | struct inode *inode; |
Al Viro | 421748e | 2008-08-02 01:04:36 -0400 | [diff] [blame] | 2152 | struct path path; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2153 | int error; |
| 2154 | |
Al Viro | 421748e | 2008-08-02 01:04:36 -0400 | [diff] [blame] | 2155 | if (!pathname || !*pathname) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2156 | return ERR_PTR(-EINVAL); |
| 2157 | |
Al Viro | 421748e | 2008-08-02 01:04:36 -0400 | [diff] [blame] | 2158 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2159 | if (error) |
| 2160 | return ERR_PTR(error); |
| 2161 | |
David Howells | bb668734 | 2015-03-17 22:26:21 +0000 | [diff] [blame] | 2162 | inode = d_backing_inode(path.dentry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2163 | error = -ENOTBLK; |
| 2164 | if (!S_ISBLK(inode->i_mode)) |
| 2165 | goto fail; |
| 2166 | error = -EACCES; |
Eric W. Biederman | a2982cc | 2016-06-09 15:34:02 -0500 | [diff] [blame] | 2167 | if (!may_open_dev(&path)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | goto fail; |
| 2169 | error = -ENOMEM; |
| 2170 | bdev = bd_acquire(inode); |
| 2171 | if (!bdev) |
| 2172 | goto fail; |
| 2173 | out: |
Al Viro | 421748e | 2008-08-02 01:04:36 -0400 | [diff] [blame] | 2174 | path_put(&path); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2175 | return bdev; |
| 2176 | fail: |
| 2177 | bdev = ERR_PTR(error); |
| 2178 | goto out; |
| 2179 | } |
Al Viro | d5686b4 | 2008-08-01 05:00:11 -0400 | [diff] [blame] | 2180 | EXPORT_SYMBOL(lookup_bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2181 | |
NeilBrown | 93b270f | 2011-02-24 17:25:47 +1100 | [diff] [blame] | 2182 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
David Howells | b71e8a4 | 2006-08-29 19:06:11 +0100 | [diff] [blame] | 2183 | { |
| 2184 | struct super_block *sb = get_super(bdev); |
| 2185 | int res = 0; |
| 2186 | |
| 2187 | if (sb) { |
| 2188 | /* |
| 2189 | * no need to lock the super, get_super holds the |
| 2190 | * read mutex so the filesystem cannot go away |
| 2191 | * under us (->put_super runs with the write lock |
| 2192 | * hold). |
| 2193 | */ |
| 2194 | shrink_dcache_sb(sb); |
NeilBrown | 93b270f | 2011-02-24 17:25:47 +1100 | [diff] [blame] | 2195 | res = invalidate_inodes(sb, kill_dirty); |
David Howells | b71e8a4 | 2006-08-29 19:06:11 +0100 | [diff] [blame] | 2196 | drop_super(sb); |
| 2197 | } |
Peter Zijlstra | f98393a | 2007-05-06 14:49:54 -0700 | [diff] [blame] | 2198 | invalidate_bdev(bdev); |
David Howells | b71e8a4 | 2006-08-29 19:06:11 +0100 | [diff] [blame] | 2199 | return res; |
| 2200 | } |
| 2201 | EXPORT_SYMBOL(__invalidate_device); |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2202 | |
| 2203 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) |
| 2204 | { |
| 2205 | struct inode *inode, *old_inode = NULL; |
| 2206 | |
Dave Chinner | 74278da | 2015-03-04 12:37:22 -0500 | [diff] [blame] | 2207 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2208 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
| 2209 | struct address_space *mapping = inode->i_mapping; |
Rabin Vincent | af30922 | 2016-12-01 09:18:28 +0100 | [diff] [blame] | 2210 | struct block_device *bdev; |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2211 | |
| 2212 | spin_lock(&inode->i_lock); |
| 2213 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || |
| 2214 | mapping->nrpages == 0) { |
| 2215 | spin_unlock(&inode->i_lock); |
| 2216 | continue; |
| 2217 | } |
| 2218 | __iget(inode); |
| 2219 | spin_unlock(&inode->i_lock); |
Dave Chinner | 74278da | 2015-03-04 12:37:22 -0500 | [diff] [blame] | 2220 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2221 | /* |
| 2222 | * We hold a reference to 'inode' so it couldn't have been |
| 2223 | * removed from s_inodes list while we dropped the |
Dave Chinner | 74278da | 2015-03-04 12:37:22 -0500 | [diff] [blame] | 2224 | * s_inode_list_lock We cannot iput the inode now as we can |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2225 | * be holding the last reference and we cannot iput it under |
Dave Chinner | 74278da | 2015-03-04 12:37:22 -0500 | [diff] [blame] | 2226 | * s_inode_list_lock. So we keep the reference and iput it |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2227 | * later. |
| 2228 | */ |
| 2229 | iput(old_inode); |
| 2230 | old_inode = inode; |
Rabin Vincent | af30922 | 2016-12-01 09:18:28 +0100 | [diff] [blame] | 2231 | bdev = I_BDEV(inode); |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2232 | |
Rabin Vincent | af30922 | 2016-12-01 09:18:28 +0100 | [diff] [blame] | 2233 | mutex_lock(&bdev->bd_mutex); |
| 2234 | if (bdev->bd_openers) |
| 2235 | func(bdev, arg); |
| 2236 | mutex_unlock(&bdev->bd_mutex); |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2237 | |
Dave Chinner | 74278da | 2015-03-04 12:37:22 -0500 | [diff] [blame] | 2238 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2239 | } |
Dave Chinner | 74278da | 2015-03-04 12:37:22 -0500 | [diff] [blame] | 2240 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
Jan Kara | 5c0d6b6 | 2012-07-03 16:45:31 +0200 | [diff] [blame] | 2241 | iput(old_inode); |
| 2242 | } |