| 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 | |
| 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> |
| 14 | #include <linux/device_cgroup.h> |
| 15 | #include <linux/highmem.h> |
| 16 | #include <linux/blkdev.h> |
| 17 | #include <linux/backing-dev.h> |
| 18 | #include <linux/module.h> |
| 19 | #include <linux/blkpg.h> |
| 20 | #include <linux/magic.h> |
| 21 | #include <linux/dax.h> |
| 22 | #include <linux/buffer_head.h> |
| 23 | #include <linux/swap.h> |
| 24 | #include <linux/pagevec.h> |
| 25 | #include <linux/writeback.h> |
| 26 | #include <linux/mpage.h> |
| 27 | #include <linux/mount.h> |
| 28 | #include <linux/uio.h> |
| 29 | #include <linux/namei.h> |
| 30 | #include <linux/log2.h> |
| 31 | #include <linux/cleancache.h> |
| 32 | #include <linux/dax.h> |
| 33 | #include <linux/badblocks.h> |
| 34 | #include <linux/task_io_accounting_ops.h> |
| 35 | #include <linux/falloc.h> |
| 36 | #include <linux/uaccess.h> |
| 37 | #include "internal.h" |
| 38 | |
| 39 | struct bdev_inode { |
| 40 | struct block_device bdev; |
| 41 | struct inode vfs_inode; |
| 42 | }; |
| 43 | |
| 44 | static const struct address_space_operations def_blk_aops; |
| 45 | |
| 46 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
| 47 | { |
| 48 | return container_of(inode, struct bdev_inode, vfs_inode); |
| 49 | } |
| 50 | |
| 51 | struct block_device *I_BDEV(struct inode *inode) |
| 52 | { |
| 53 | return &BDEV_I(inode)->bdev; |
| 54 | } |
| 55 | EXPORT_SYMBOL(I_BDEV); |
| 56 | |
| 57 | static void bdev_write_inode(struct block_device *bdev) |
| 58 | { |
| 59 | struct inode *inode = bdev->bd_inode; |
| 60 | int ret; |
| 61 | |
| 62 | spin_lock(&inode->i_lock); |
| 63 | while (inode->i_state & I_DIRTY) { |
| 64 | spin_unlock(&inode->i_lock); |
| 65 | ret = write_inode_now(inode, true); |
| 66 | if (ret) { |
| 67 | char name[BDEVNAME_SIZE]; |
| 68 | pr_warn_ratelimited("VFS: Dirty inode writeback failed " |
| 69 | "for block device %s (err=%d).\n", |
| 70 | bdevname(bdev, name), ret); |
| 71 | } |
| 72 | spin_lock(&inode->i_lock); |
| 73 | } |
| 74 | spin_unlock(&inode->i_lock); |
| 75 | } |
| 76 | |
| 77 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
| 78 | void kill_bdev(struct block_device *bdev) |
| 79 | { |
| 80 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
| 81 | |
| 82 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
| 83 | return; |
| 84 | |
| 85 | invalidate_bh_lrus(); |
| 86 | truncate_inode_pages(mapping, 0); |
| 87 | } |
| 88 | EXPORT_SYMBOL(kill_bdev); |
| 89 | |
| 90 | /* Invalidate clean unused buffers and pagecache. */ |
| 91 | void invalidate_bdev(struct block_device *bdev) |
| 92 | { |
| 93 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
| 94 | |
| 95 | if (mapping->nrpages) { |
| 96 | invalidate_bh_lrus(); |
| 97 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ |
| 98 | invalidate_mapping_pages(mapping, 0, -1); |
| 99 | } |
| 100 | /* 99% of the time, we don't need to flush the cleancache on the bdev. |
| 101 | * But, for the strange corners, lets be cautious |
| 102 | */ |
| 103 | cleancache_invalidate_inode(mapping); |
| 104 | } |
| 105 | EXPORT_SYMBOL(invalidate_bdev); |
| 106 | |
| 107 | int set_blocksize(struct block_device *bdev, int size) |
| 108 | { |
| 109 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ |
| 110 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
| 111 | return -EINVAL; |
| 112 | |
| 113 | /* Size cannot be smaller than the size supported by the device */ |
| 114 | if (size < bdev_logical_block_size(bdev)) |
| 115 | return -EINVAL; |
| 116 | |
| 117 | /* Don't change the size if it is same as current */ |
| 118 | if (bdev->bd_block_size != size) { |
| 119 | sync_blockdev(bdev); |
| 120 | bdev->bd_block_size = size; |
| 121 | bdev->bd_inode->i_blkbits = blksize_bits(size); |
| 122 | kill_bdev(bdev); |
| 123 | } |
| 124 | return 0; |
| 125 | } |
| 126 | |
| 127 | EXPORT_SYMBOL(set_blocksize); |
| 128 | |
| 129 | int sb_set_blocksize(struct super_block *sb, int size) |
| 130 | { |
| 131 | if (set_blocksize(sb->s_bdev, size)) |
| 132 | return 0; |
| 133 | /* If we get here, we know size is power of two |
| 134 | * and it's value is between 512 and PAGE_SIZE */ |
| 135 | sb->s_blocksize = size; |
| 136 | sb->s_blocksize_bits = blksize_bits(size); |
| 137 | return sb->s_blocksize; |
| 138 | } |
| 139 | |
| 140 | EXPORT_SYMBOL(sb_set_blocksize); |
| 141 | |
| 142 | int sb_min_blocksize(struct super_block *sb, int size) |
| 143 | { |
| 144 | int minsize = bdev_logical_block_size(sb->s_bdev); |
| 145 | if (size < minsize) |
| 146 | size = minsize; |
| 147 | return sb_set_blocksize(sb, size); |
| 148 | } |
| 149 | |
| 150 | EXPORT_SYMBOL(sb_min_blocksize); |
| 151 | |
| 152 | static int |
| 153 | blkdev_get_block(struct inode *inode, sector_t iblock, |
| 154 | struct buffer_head *bh, int create) |
| 155 | { |
| 156 | bh->b_bdev = I_BDEV(inode); |
| 157 | bh->b_blocknr = iblock; |
| 158 | set_buffer_mapped(bh); |
| 159 | return 0; |
| 160 | } |
| 161 | |
| 162 | static struct inode *bdev_file_inode(struct file *file) |
| 163 | { |
| 164 | return file->f_mapping->host; |
| 165 | } |
| 166 | |
| 167 | static unsigned int dio_bio_write_op(struct kiocb *iocb) |
| 168 | { |
| 169 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; |
| 170 | |
| 171 | /* avoid the need for a I/O completion work item */ |
| 172 | if (iocb->ki_flags & IOCB_DSYNC) |
| 173 | op |= REQ_FUA; |
| 174 | return op; |
| 175 | } |
| 176 | |
| 177 | #define DIO_INLINE_BIO_VECS 4 |
| 178 | |
| 179 | static void blkdev_bio_end_io_simple(struct bio *bio) |
| 180 | { |
| 181 | struct task_struct *waiter = bio->bi_private; |
| 182 | |
| 183 | WRITE_ONCE(bio->bi_private, NULL); |
| 184 | wake_up_process(waiter); |
| 185 | } |
| 186 | |
| 187 | static ssize_t |
| 188 | __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter, |
| 189 | int nr_pages) |
| 190 | { |
| 191 | struct file *file = iocb->ki_filp; |
| 192 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
| 193 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec; |
| 194 | loff_t pos = iocb->ki_pos; |
| 195 | bool should_dirty = false; |
| 196 | struct bio bio; |
| 197 | ssize_t ret; |
| 198 | blk_qc_t qc; |
| 199 | int i; |
| 200 | |
| 201 | if ((pos | iov_iter_alignment(iter)) & |
| 202 | (bdev_logical_block_size(bdev) - 1)) |
| 203 | return -EINVAL; |
| 204 | |
| 205 | if (nr_pages <= DIO_INLINE_BIO_VECS) |
| 206 | vecs = inline_vecs; |
| 207 | else { |
| 208 | vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL); |
| 209 | if (!vecs) |
| 210 | return -ENOMEM; |
| 211 | } |
| 212 | |
| 213 | bio_init(&bio, vecs, nr_pages); |
| 214 | bio_set_dev(&bio, bdev); |
| 215 | bio.bi_iter.bi_sector = pos >> 9; |
| 216 | bio.bi_write_hint = iocb->ki_hint; |
| 217 | bio.bi_private = current; |
| 218 | bio.bi_end_io = blkdev_bio_end_io_simple; |
| 219 | |
| 220 | ret = bio_iov_iter_get_pages(&bio, iter); |
| 221 | if (unlikely(ret)) |
| 222 | return ret; |
| 223 | ret = bio.bi_iter.bi_size; |
| 224 | |
| 225 | if (iov_iter_rw(iter) == READ) { |
| 226 | bio.bi_opf = REQ_OP_READ; |
| 227 | if (iter_is_iovec(iter)) |
| 228 | should_dirty = true; |
| 229 | } else { |
| 230 | bio.bi_opf = dio_bio_write_op(iocb); |
| 231 | task_io_account_write(ret); |
| 232 | } |
| 233 | |
| 234 | qc = submit_bio(&bio); |
| 235 | for (;;) { |
| 236 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 237 | if (!READ_ONCE(bio.bi_private)) |
| 238 | break; |
| 239 | if (!(iocb->ki_flags & IOCB_HIPRI) || |
| 240 | !blk_poll(bdev_get_queue(bdev), qc)) |
| 241 | io_schedule(); |
| 242 | } |
| 243 | __set_current_state(TASK_RUNNING); |
| 244 | |
| 245 | bio_for_each_segment_all(bvec, &bio, i) { |
| 246 | if (should_dirty && !PageCompound(bvec->bv_page)) |
| 247 | set_page_dirty_lock(bvec->bv_page); |
| 248 | put_page(bvec->bv_page); |
| 249 | } |
| 250 | |
| 251 | if (vecs != inline_vecs) |
| 252 | kfree(vecs); |
| 253 | |
| 254 | if (unlikely(bio.bi_status)) |
| 255 | ret = blk_status_to_errno(bio.bi_status); |
| 256 | |
| 257 | bio_uninit(&bio); |
| 258 | |
| 259 | return ret; |
| 260 | } |
| 261 | |
| 262 | struct blkdev_dio { |
| 263 | union { |
| 264 | struct kiocb *iocb; |
| 265 | struct task_struct *waiter; |
| 266 | }; |
| 267 | size_t size; |
| 268 | atomic_t ref; |
| 269 | bool multi_bio : 1; |
| 270 | bool should_dirty : 1; |
| 271 | bool is_sync : 1; |
| 272 | struct bio bio; |
| 273 | }; |
| 274 | |
| 275 | static struct bio_set *blkdev_dio_pool __read_mostly; |
| 276 | |
| 277 | static void blkdev_bio_end_io(struct bio *bio) |
| 278 | { |
| 279 | struct blkdev_dio *dio = bio->bi_private; |
| 280 | bool should_dirty = dio->should_dirty; |
| 281 | |
| 282 | if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) { |
| 283 | if (bio->bi_status && !dio->bio.bi_status) |
| 284 | dio->bio.bi_status = bio->bi_status; |
| 285 | } else { |
| 286 | if (!dio->is_sync) { |
| 287 | struct kiocb *iocb = dio->iocb; |
| 288 | ssize_t ret; |
| 289 | |
| 290 | if (likely(!dio->bio.bi_status)) { |
| 291 | ret = dio->size; |
| 292 | iocb->ki_pos += ret; |
| 293 | } else { |
| 294 | ret = blk_status_to_errno(dio->bio.bi_status); |
| 295 | } |
| 296 | |
| 297 | dio->iocb->ki_complete(iocb, ret, 0); |
| 298 | bio_put(&dio->bio); |
| 299 | } else { |
| 300 | struct task_struct *waiter = dio->waiter; |
| 301 | |
| 302 | WRITE_ONCE(dio->waiter, NULL); |
| 303 | wake_up_process(waiter); |
| 304 | } |
| 305 | } |
| 306 | |
| 307 | if (should_dirty) { |
| 308 | bio_check_pages_dirty(bio); |
| 309 | } else { |
| 310 | struct bio_vec *bvec; |
| 311 | int i; |
| 312 | |
| 313 | bio_for_each_segment_all(bvec, bio, i) |
| 314 | put_page(bvec->bv_page); |
| 315 | bio_put(bio); |
| 316 | } |
| 317 | } |
| 318 | |
| 319 | static ssize_t |
| 320 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages) |
| 321 | { |
| 322 | struct file *file = iocb->ki_filp; |
| 323 | struct inode *inode = bdev_file_inode(file); |
| 324 | struct block_device *bdev = I_BDEV(inode); |
| 325 | struct blk_plug plug; |
| 326 | struct blkdev_dio *dio; |
| 327 | struct bio *bio; |
| 328 | bool is_read = (iov_iter_rw(iter) == READ), is_sync; |
| 329 | loff_t pos = iocb->ki_pos; |
| 330 | blk_qc_t qc = BLK_QC_T_NONE; |
| 331 | int ret = 0; |
| 332 | |
| 333 | if ((pos | iov_iter_alignment(iter)) & |
| 334 | (bdev_logical_block_size(bdev) - 1)) |
| 335 | return -EINVAL; |
| 336 | |
| 337 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool); |
| 338 | bio_get(bio); /* extra ref for the completion handler */ |
| 339 | |
| 340 | dio = container_of(bio, struct blkdev_dio, bio); |
| 341 | dio->is_sync = is_sync = is_sync_kiocb(iocb); |
| 342 | if (dio->is_sync) |
| 343 | dio->waiter = current; |
| 344 | else |
| 345 | dio->iocb = iocb; |
| 346 | |
| 347 | dio->size = 0; |
| 348 | dio->multi_bio = false; |
| 349 | dio->should_dirty = is_read && (iter->type == ITER_IOVEC); |
| 350 | |
| 351 | blk_start_plug(&plug); |
| 352 | for (;;) { |
| 353 | bio_set_dev(bio, bdev); |
| 354 | bio->bi_iter.bi_sector = pos >> 9; |
| 355 | bio->bi_write_hint = iocb->ki_hint; |
| 356 | bio->bi_private = dio; |
| 357 | bio->bi_end_io = blkdev_bio_end_io; |
| 358 | |
| 359 | ret = bio_iov_iter_get_pages(bio, iter); |
| 360 | if (unlikely(ret)) { |
| 361 | bio->bi_status = BLK_STS_IOERR; |
| 362 | bio_endio(bio); |
| 363 | break; |
| 364 | } |
| 365 | |
| 366 | if (is_read) { |
| 367 | bio->bi_opf = REQ_OP_READ; |
| 368 | if (dio->should_dirty) |
| 369 | bio_set_pages_dirty(bio); |
| 370 | } else { |
| 371 | bio->bi_opf = dio_bio_write_op(iocb); |
| 372 | task_io_account_write(bio->bi_iter.bi_size); |
| 373 | } |
| 374 | |
| 375 | dio->size += bio->bi_iter.bi_size; |
| 376 | pos += bio->bi_iter.bi_size; |
| 377 | |
| 378 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES); |
| 379 | if (!nr_pages) { |
| 380 | qc = submit_bio(bio); |
| 381 | break; |
| 382 | } |
| 383 | |
| 384 | if (!dio->multi_bio) { |
| 385 | dio->multi_bio = true; |
| 386 | atomic_set(&dio->ref, 2); |
| 387 | } else { |
| 388 | atomic_inc(&dio->ref); |
| 389 | } |
| 390 | |
| 391 | submit_bio(bio); |
| 392 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
| 393 | } |
| 394 | blk_finish_plug(&plug); |
| 395 | |
| 396 | if (!is_sync) |
| 397 | return -EIOCBQUEUED; |
| 398 | |
| 399 | for (;;) { |
| 400 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 401 | if (!READ_ONCE(dio->waiter)) |
| 402 | break; |
| 403 | |
| 404 | if (!(iocb->ki_flags & IOCB_HIPRI) || |
| 405 | !blk_poll(bdev_get_queue(bdev), qc)) |
| 406 | io_schedule(); |
| 407 | } |
| 408 | __set_current_state(TASK_RUNNING); |
| 409 | |
| 410 | if (!ret) |
| 411 | ret = blk_status_to_errno(dio->bio.bi_status); |
| 412 | if (likely(!ret)) |
| 413 | ret = dio->size; |
| 414 | |
| 415 | bio_put(&dio->bio); |
| 416 | return ret; |
| 417 | } |
| 418 | |
| 419 | static ssize_t |
| 420 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
| 421 | { |
| 422 | int nr_pages; |
| 423 | |
| 424 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1); |
| 425 | if (!nr_pages) |
| 426 | return 0; |
| 427 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES) |
| 428 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); |
| 429 | |
| 430 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES)); |
| 431 | } |
| 432 | |
| 433 | static __init int blkdev_init(void) |
| 434 | { |
| 435 | blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS); |
| 436 | if (!blkdev_dio_pool) |
| 437 | return -ENOMEM; |
| 438 | return 0; |
| 439 | } |
| 440 | module_init(blkdev_init); |
| 441 | |
| 442 | int __sync_blockdev(struct block_device *bdev, int wait) |
| 443 | { |
| 444 | if (!bdev) |
| 445 | return 0; |
| 446 | if (!wait) |
| 447 | return filemap_flush(bdev->bd_inode->i_mapping); |
| 448 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); |
| 449 | } |
| 450 | |
| 451 | /* |
| 452 | * Write out and wait upon all the dirty data associated with a block |
| 453 | * device via its mapping. Does not take the superblock lock. |
| 454 | */ |
| 455 | int sync_blockdev(struct block_device *bdev) |
| 456 | { |
| 457 | return __sync_blockdev(bdev, 1); |
| 458 | } |
| 459 | EXPORT_SYMBOL(sync_blockdev); |
| 460 | |
| 461 | /* |
| 462 | * Write out and wait upon all dirty data associated with this |
| 463 | * device. Filesystem data as well as the underlying block |
| 464 | * device. Takes the superblock lock. |
| 465 | */ |
| 466 | int fsync_bdev(struct block_device *bdev) |
| 467 | { |
| 468 | struct super_block *sb = get_super(bdev); |
| 469 | if (sb) { |
| 470 | int res = sync_filesystem(sb); |
| 471 | drop_super(sb); |
| 472 | return res; |
| 473 | } |
| 474 | return sync_blockdev(bdev); |
| 475 | } |
| 476 | EXPORT_SYMBOL(fsync_bdev); |
| 477 | |
| 478 | /** |
| 479 | * freeze_bdev -- lock a filesystem and force it into a consistent state |
| 480 | * @bdev: blockdevice to lock |
| 481 | * |
| 482 | * If a superblock is found on this device, we take the s_umount semaphore |
| 483 | * on it to make sure nobody unmounts until the snapshot creation is done. |
| 484 | * The reference counter (bd_fsfreeze_count) guarantees that only the last |
| 485 | * unfreeze process can unfreeze the frozen filesystem actually when multiple |
| 486 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and |
| 487 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze |
| 488 | * actually. |
| 489 | */ |
| 490 | struct super_block *freeze_bdev(struct block_device *bdev) |
| 491 | { |
| 492 | struct super_block *sb; |
| 493 | int error = 0; |
| 494 | |
| 495 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
| 496 | if (++bdev->bd_fsfreeze_count > 1) { |
| 497 | /* |
| 498 | * We don't even need to grab a reference - the first call |
| 499 | * to freeze_bdev grab an active reference and only the last |
| 500 | * thaw_bdev drops it. |
| 501 | */ |
| 502 | sb = get_super(bdev); |
| 503 | if (sb) |
| 504 | drop_super(sb); |
| 505 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 506 | return sb; |
| 507 | } |
| 508 | |
| 509 | sb = get_active_super(bdev); |
| 510 | if (!sb) |
| 511 | goto out; |
| 512 | if (sb->s_op->freeze_super) |
| 513 | error = sb->s_op->freeze_super(sb); |
| 514 | else |
| 515 | error = freeze_super(sb); |
| 516 | if (error) { |
| 517 | deactivate_super(sb); |
| 518 | bdev->bd_fsfreeze_count--; |
| 519 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 520 | return ERR_PTR(error); |
| 521 | } |
| 522 | deactivate_super(sb); |
| 523 | out: |
| 524 | sync_blockdev(bdev); |
| 525 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 526 | return sb; /* thaw_bdev releases s->s_umount */ |
| 527 | } |
| 528 | EXPORT_SYMBOL(freeze_bdev); |
| 529 | |
| 530 | /** |
| 531 | * thaw_bdev -- unlock filesystem |
| 532 | * @bdev: blockdevice to unlock |
| 533 | * @sb: associated superblock |
| 534 | * |
| 535 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). |
| 536 | */ |
| 537 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) |
| 538 | { |
| 539 | int error = -EINVAL; |
| 540 | |
| 541 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
| 542 | if (!bdev->bd_fsfreeze_count) |
| 543 | goto out; |
| 544 | |
| 545 | error = 0; |
| 546 | if (--bdev->bd_fsfreeze_count > 0) |
| 547 | goto out; |
| 548 | |
| 549 | if (!sb) |
| 550 | goto out; |
| 551 | |
| 552 | if (sb->s_op->thaw_super) |
| 553 | error = sb->s_op->thaw_super(sb); |
| 554 | else |
| 555 | error = thaw_super(sb); |
| 556 | if (error) |
| 557 | bdev->bd_fsfreeze_count++; |
| 558 | out: |
| 559 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 560 | return error; |
| 561 | } |
| 562 | EXPORT_SYMBOL(thaw_bdev); |
| 563 | |
| 564 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
| 565 | { |
| 566 | return block_write_full_page(page, blkdev_get_block, wbc); |
| 567 | } |
| 568 | |
| 569 | static int blkdev_readpage(struct file * file, struct page * page) |
| 570 | { |
| 571 | return block_read_full_page(page, blkdev_get_block); |
| 572 | } |
| 573 | |
| 574 | static int blkdev_readpages(struct file *file, struct address_space *mapping, |
| 575 | struct list_head *pages, unsigned nr_pages) |
| 576 | { |
| 577 | return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block); |
| 578 | } |
| 579 | |
| 580 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
| 581 | loff_t pos, unsigned len, unsigned flags, |
| 582 | struct page **pagep, void **fsdata) |
| 583 | { |
| 584 | return block_write_begin(mapping, pos, len, flags, pagep, |
| 585 | blkdev_get_block); |
| 586 | } |
| 587 | |
| 588 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
| 589 | loff_t pos, unsigned len, unsigned copied, |
| 590 | struct page *page, void *fsdata) |
| 591 | { |
| 592 | int ret; |
| 593 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
| 594 | |
| 595 | unlock_page(page); |
| 596 | put_page(page); |
| 597 | |
| 598 | return ret; |
| 599 | } |
| 600 | |
| 601 | /* |
| 602 | * private llseek: |
| 603 | * for a block special file file_inode(file)->i_size is zero |
| 604 | * so we compute the size by hand (just as in block_read/write above) |
| 605 | */ |
| 606 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
| 607 | { |
| 608 | struct inode *bd_inode = bdev_file_inode(file); |
| 609 | loff_t retval; |
| 610 | |
| 611 | inode_lock(bd_inode); |
| 612 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
| 613 | inode_unlock(bd_inode); |
| 614 | return retval; |
| 615 | } |
| 616 | |
| 617 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
| 618 | { |
| 619 | struct inode *bd_inode = bdev_file_inode(filp); |
| 620 | struct block_device *bdev = I_BDEV(bd_inode); |
| 621 | int error; |
| 622 | |
| 623 | error = file_write_and_wait_range(filp, start, end); |
| 624 | if (error) |
| 625 | return error; |
| 626 | |
| 627 | /* |
| 628 | * There is no need to serialise calls to blkdev_issue_flush with |
| 629 | * i_mutex and doing so causes performance issues with concurrent |
| 630 | * O_SYNC writers to a block device. |
| 631 | */ |
| 632 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
| 633 | if (error == -EOPNOTSUPP) |
| 634 | error = 0; |
| 635 | |
| 636 | return error; |
| 637 | } |
| 638 | EXPORT_SYMBOL(blkdev_fsync); |
| 639 | |
| 640 | /** |
| 641 | * bdev_read_page() - Start reading a page from a block device |
| 642 | * @bdev: The device to read the page from |
| 643 | * @sector: The offset on the device to read the page to (need not be aligned) |
| 644 | * @page: The page to read |
| 645 | * |
| 646 | * On entry, the page should be locked. It will be unlocked when the page |
| 647 | * has been read. If the block driver implements rw_page synchronously, |
| 648 | * that will be true on exit from this function, but it need not be. |
| 649 | * |
| 650 | * Errors returned by this function are usually "soft", eg out of memory, or |
| 651 | * queue full; callers should try a different route to read this page rather |
| 652 | * than propagate an error back up the stack. |
| 653 | * |
| 654 | * Return: negative errno if an error occurs, 0 if submission was successful. |
| 655 | */ |
| 656 | int bdev_read_page(struct block_device *bdev, sector_t sector, |
| 657 | struct page *page) |
| 658 | { |
| 659 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
| 660 | int result = -EOPNOTSUPP; |
| 661 | |
| 662 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
| 663 | return result; |
| 664 | |
| 665 | result = blk_queue_enter(bdev->bd_queue, 0); |
| 666 | if (result) |
| 667 | return result; |
| 668 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false); |
| 669 | blk_queue_exit(bdev->bd_queue); |
| 670 | return result; |
| 671 | } |
| 672 | EXPORT_SYMBOL_GPL(bdev_read_page); |
| 673 | |
| 674 | /** |
| 675 | * bdev_write_page() - Start writing a page to a block device |
| 676 | * @bdev: The device to write the page to |
| 677 | * @sector: The offset on the device to write the page to (need not be aligned) |
| 678 | * @page: The page to write |
| 679 | * @wbc: The writeback_control for the write |
| 680 | * |
| 681 | * On entry, the page should be locked and not currently under writeback. |
| 682 | * On exit, if the write started successfully, the page will be unlocked and |
| 683 | * under writeback. If the write failed already (eg the driver failed to |
| 684 | * queue the page to the device), the page will still be locked. If the |
| 685 | * caller is a ->writepage implementation, it will need to unlock the page. |
| 686 | * |
| 687 | * Errors returned by this function are usually "soft", eg out of memory, or |
| 688 | * queue full; callers should try a different route to write this page rather |
| 689 | * than propagate an error back up the stack. |
| 690 | * |
| 691 | * Return: negative errno if an error occurs, 0 if submission was successful. |
| 692 | */ |
| 693 | int bdev_write_page(struct block_device *bdev, sector_t sector, |
| 694 | struct page *page, struct writeback_control *wbc) |
| 695 | { |
| 696 | int result; |
| 697 | const struct block_device_operations *ops = bdev->bd_disk->fops; |
| 698 | |
| 699 | if (!ops->rw_page || bdev_get_integrity(bdev)) |
| 700 | return -EOPNOTSUPP; |
| 701 | result = blk_queue_enter(bdev->bd_queue, 0); |
| 702 | if (result) |
| 703 | return result; |
| 704 | |
| 705 | set_page_writeback(page); |
| 706 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true); |
| 707 | if (result) { |
| 708 | end_page_writeback(page); |
| 709 | } else { |
| 710 | clean_page_buffers(page); |
| 711 | unlock_page(page); |
| 712 | } |
| 713 | blk_queue_exit(bdev->bd_queue); |
| 714 | return result; |
| 715 | } |
| 716 | EXPORT_SYMBOL_GPL(bdev_write_page); |
| 717 | |
| 718 | /* |
| 719 | * pseudo-fs |
| 720 | */ |
| 721 | |
| 722 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); |
| 723 | static struct kmem_cache * bdev_cachep __read_mostly; |
| 724 | |
| 725 | static struct inode *bdev_alloc_inode(struct super_block *sb) |
| 726 | { |
| 727 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
| 728 | if (!ei) |
| 729 | return NULL; |
| 730 | return &ei->vfs_inode; |
| 731 | } |
| 732 | |
| 733 | static void bdev_i_callback(struct rcu_head *head) |
| 734 | { |
| 735 | struct inode *inode = container_of(head, struct inode, i_rcu); |
| 736 | struct bdev_inode *bdi = BDEV_I(inode); |
| 737 | |
| 738 | kmem_cache_free(bdev_cachep, bdi); |
| 739 | } |
| 740 | |
| 741 | static void bdev_destroy_inode(struct inode *inode) |
| 742 | { |
| 743 | call_rcu(&inode->i_rcu, bdev_i_callback); |
| 744 | } |
| 745 | |
| 746 | static void init_once(void *foo) |
| 747 | { |
| 748 | struct bdev_inode *ei = (struct bdev_inode *) foo; |
| 749 | struct block_device *bdev = &ei->bdev; |
| 750 | |
| 751 | memset(bdev, 0, sizeof(*bdev)); |
| 752 | mutex_init(&bdev->bd_mutex); |
| 753 | INIT_LIST_HEAD(&bdev->bd_list); |
| 754 | #ifdef CONFIG_SYSFS |
| 755 | INIT_LIST_HEAD(&bdev->bd_holder_disks); |
| 756 | #endif |
| 757 | bdev->bd_bdi = &noop_backing_dev_info; |
| 758 | inode_init_once(&ei->vfs_inode); |
| 759 | /* Initialize mutex for freeze. */ |
| 760 | mutex_init(&bdev->bd_fsfreeze_mutex); |
| 761 | } |
| 762 | |
| 763 | static void bdev_evict_inode(struct inode *inode) |
| 764 | { |
| 765 | struct block_device *bdev = &BDEV_I(inode)->bdev; |
| 766 | truncate_inode_pages_final(&inode->i_data); |
| 767 | invalidate_inode_buffers(inode); /* is it needed here? */ |
| 768 | clear_inode(inode); |
| 769 | spin_lock(&bdev_lock); |
| 770 | list_del_init(&bdev->bd_list); |
| 771 | spin_unlock(&bdev_lock); |
| 772 | /* Detach inode from wb early as bdi_put() may free bdi->wb */ |
| 773 | inode_detach_wb(inode); |
| 774 | if (bdev->bd_bdi != &noop_backing_dev_info) { |
| 775 | bdi_put(bdev->bd_bdi); |
| 776 | bdev->bd_bdi = &noop_backing_dev_info; |
| 777 | } |
| 778 | } |
| 779 | |
| 780 | static const struct super_operations bdev_sops = { |
| 781 | .statfs = simple_statfs, |
| 782 | .alloc_inode = bdev_alloc_inode, |
| 783 | .destroy_inode = bdev_destroy_inode, |
| 784 | .drop_inode = generic_delete_inode, |
| 785 | .evict_inode = bdev_evict_inode, |
| 786 | }; |
| 787 | |
| 788 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
| 789 | int flags, const char *dev_name, void *data) |
| 790 | { |
| 791 | struct dentry *dent; |
| 792 | dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); |
| 793 | if (!IS_ERR(dent)) |
| 794 | dent->d_sb->s_iflags |= SB_I_CGROUPWB; |
| 795 | return dent; |
| 796 | } |
| 797 | |
| 798 | static struct file_system_type bd_type = { |
| 799 | .name = "bdev", |
| 800 | .mount = bd_mount, |
| 801 | .kill_sb = kill_anon_super, |
| 802 | }; |
| 803 | |
| 804 | struct super_block *blockdev_superblock __read_mostly; |
| 805 | EXPORT_SYMBOL_GPL(blockdev_superblock); |
| 806 | |
| 807 | void __init bdev_cache_init(void) |
| 808 | { |
| 809 | int err; |
| 810 | static struct vfsmount *bd_mnt; |
| 811 | |
| 812 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
| 813 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
| 814 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC), |
| 815 | init_once); |
| 816 | err = register_filesystem(&bd_type); |
| 817 | if (err) |
| 818 | panic("Cannot register bdev pseudo-fs"); |
| 819 | bd_mnt = kern_mount(&bd_type); |
| 820 | if (IS_ERR(bd_mnt)) |
| 821 | panic("Cannot create bdev pseudo-fs"); |
| 822 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
| 823 | } |
| 824 | |
| 825 | /* |
| 826 | * Most likely _very_ bad one - but then it's hardly critical for small |
| 827 | * /dev and can be fixed when somebody will need really large one. |
| 828 | * Keep in mind that it will be fed through icache hash function too. |
| 829 | */ |
| 830 | static inline unsigned long hash(dev_t dev) |
| 831 | { |
| 832 | return MAJOR(dev)+MINOR(dev); |
| 833 | } |
| 834 | |
| 835 | static int bdev_test(struct inode *inode, void *data) |
| 836 | { |
| 837 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; |
| 838 | } |
| 839 | |
| 840 | static int bdev_set(struct inode *inode, void *data) |
| 841 | { |
| 842 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; |
| 843 | return 0; |
| 844 | } |
| 845 | |
| 846 | static LIST_HEAD(all_bdevs); |
| 847 | |
| 848 | /* |
| 849 | * If there is a bdev inode for this device, unhash it so that it gets evicted |
| 850 | * as soon as last inode reference is dropped. |
| 851 | */ |
| 852 | void bdev_unhash_inode(dev_t dev) |
| 853 | { |
| 854 | struct inode *inode; |
| 855 | |
| 856 | inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev); |
| 857 | if (inode) { |
| 858 | remove_inode_hash(inode); |
| 859 | iput(inode); |
| 860 | } |
| 861 | } |
| 862 | |
| 863 | struct block_device *bdget(dev_t dev) |
| 864 | { |
| 865 | struct block_device *bdev; |
| 866 | struct inode *inode; |
| 867 | |
| 868 | inode = iget5_locked(blockdev_superblock, hash(dev), |
| 869 | bdev_test, bdev_set, &dev); |
| 870 | |
| 871 | if (!inode) |
| 872 | return NULL; |
| 873 | |
| 874 | bdev = &BDEV_I(inode)->bdev; |
| 875 | |
| 876 | if (inode->i_state & I_NEW) { |
| 877 | bdev->bd_contains = NULL; |
| 878 | bdev->bd_super = NULL; |
| 879 | bdev->bd_inode = inode; |
| 880 | bdev->bd_block_size = i_blocksize(inode); |
| 881 | bdev->bd_part_count = 0; |
| 882 | bdev->bd_invalidated = 0; |
| 883 | inode->i_mode = S_IFBLK; |
| 884 | inode->i_rdev = dev; |
| 885 | inode->i_bdev = bdev; |
| 886 | inode->i_data.a_ops = &def_blk_aops; |
| 887 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); |
| 888 | spin_lock(&bdev_lock); |
| 889 | list_add(&bdev->bd_list, &all_bdevs); |
| 890 | spin_unlock(&bdev_lock); |
| 891 | unlock_new_inode(inode); |
| 892 | } |
| 893 | return bdev; |
| 894 | } |
| 895 | |
| 896 | EXPORT_SYMBOL(bdget); |
| 897 | |
| 898 | /** |
| 899 | * bdgrab -- Grab a reference to an already referenced block device |
| 900 | * @bdev: Block device to grab a reference to. |
| 901 | */ |
| 902 | struct block_device *bdgrab(struct block_device *bdev) |
| 903 | { |
| 904 | ihold(bdev->bd_inode); |
| 905 | return bdev; |
| 906 | } |
| 907 | EXPORT_SYMBOL(bdgrab); |
| 908 | |
| 909 | long nr_blockdev_pages(void) |
| 910 | { |
| 911 | struct block_device *bdev; |
| 912 | long ret = 0; |
| 913 | spin_lock(&bdev_lock); |
| 914 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
| 915 | ret += bdev->bd_inode->i_mapping->nrpages; |
| 916 | } |
| 917 | spin_unlock(&bdev_lock); |
| 918 | return ret; |
| 919 | } |
| 920 | |
| 921 | void bdput(struct block_device *bdev) |
| 922 | { |
| 923 | iput(bdev->bd_inode); |
| 924 | } |
| 925 | |
| 926 | EXPORT_SYMBOL(bdput); |
| 927 | |
| 928 | static struct block_device *bd_acquire(struct inode *inode) |
| 929 | { |
| 930 | struct block_device *bdev; |
| 931 | |
| 932 | spin_lock(&bdev_lock); |
| 933 | bdev = inode->i_bdev; |
| 934 | if (bdev && !inode_unhashed(bdev->bd_inode)) { |
| 935 | bdgrab(bdev); |
| 936 | spin_unlock(&bdev_lock); |
| 937 | return bdev; |
| 938 | } |
| 939 | spin_unlock(&bdev_lock); |
| 940 | |
| 941 | /* |
| 942 | * i_bdev references block device inode that was already shut down |
| 943 | * (corresponding device got removed). Remove the reference and look |
| 944 | * up block device inode again just in case new device got |
| 945 | * reestablished under the same device number. |
| 946 | */ |
| 947 | if (bdev) |
| 948 | bd_forget(inode); |
| 949 | |
| 950 | bdev = bdget(inode->i_rdev); |
| 951 | if (bdev) { |
| 952 | spin_lock(&bdev_lock); |
| 953 | if (!inode->i_bdev) { |
| 954 | /* |
| 955 | * We take an additional reference to bd_inode, |
| 956 | * and it's released in clear_inode() of inode. |
| 957 | * So, we can access it via ->i_mapping always |
| 958 | * without igrab(). |
| 959 | */ |
| 960 | bdgrab(bdev); |
| 961 | inode->i_bdev = bdev; |
| 962 | inode->i_mapping = bdev->bd_inode->i_mapping; |
| 963 | } |
| 964 | spin_unlock(&bdev_lock); |
| 965 | } |
| 966 | return bdev; |
| 967 | } |
| 968 | |
| 969 | /* Call when you free inode */ |
| 970 | |
| 971 | void bd_forget(struct inode *inode) |
| 972 | { |
| 973 | struct block_device *bdev = NULL; |
| 974 | |
| 975 | spin_lock(&bdev_lock); |
| 976 | if (!sb_is_blkdev_sb(inode->i_sb)) |
| 977 | bdev = inode->i_bdev; |
| 978 | inode->i_bdev = NULL; |
| 979 | inode->i_mapping = &inode->i_data; |
| 980 | spin_unlock(&bdev_lock); |
| 981 | |
| 982 | if (bdev) |
| 983 | bdput(bdev); |
| 984 | } |
| 985 | |
| 986 | /** |
| 987 | * bd_may_claim - test whether a block device can be claimed |
| 988 | * @bdev: block device of interest |
| 989 | * @whole: whole block device containing @bdev, may equal @bdev |
| 990 | * @holder: holder trying to claim @bdev |
| 991 | * |
| 992 | * Test whether @bdev can be claimed by @holder. |
| 993 | * |
| 994 | * CONTEXT: |
| 995 | * spin_lock(&bdev_lock). |
| 996 | * |
| 997 | * RETURNS: |
| 998 | * %true if @bdev can be claimed, %false otherwise. |
| 999 | */ |
| 1000 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, |
| 1001 | void *holder) |
| 1002 | { |
| 1003 | if (bdev->bd_holder == holder) |
| 1004 | return true; /* already a holder */ |
| 1005 | else if (bdev->bd_holder != NULL) |
| 1006 | return false; /* held by someone else */ |
| 1007 | else if (whole == bdev) |
| 1008 | return true; /* is a whole device which isn't held */ |
| 1009 | |
| 1010 | else if (whole->bd_holder == bd_may_claim) |
| 1011 | return true; /* is a partition of a device that is being partitioned */ |
| 1012 | else if (whole->bd_holder != NULL) |
| 1013 | return false; /* is a partition of a held device */ |
| 1014 | else |
| 1015 | return true; /* is a partition of an un-held device */ |
| 1016 | } |
| 1017 | |
| 1018 | /** |
| 1019 | * bd_prepare_to_claim - prepare to claim a block device |
| 1020 | * @bdev: block device of interest |
| 1021 | * @whole: the whole device containing @bdev, may equal @bdev |
| 1022 | * @holder: holder trying to claim @bdev |
| 1023 | * |
| 1024 | * Prepare to claim @bdev. This function fails if @bdev is already |
| 1025 | * claimed by another holder and waits if another claiming is in |
| 1026 | * progress. This function doesn't actually claim. On successful |
| 1027 | * return, the caller has ownership of bd_claiming and bd_holder[s]. |
| 1028 | * |
| 1029 | * CONTEXT: |
| 1030 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab |
| 1031 | * it multiple times. |
| 1032 | * |
| 1033 | * RETURNS: |
| 1034 | * 0 if @bdev can be claimed, -EBUSY otherwise. |
| 1035 | */ |
| 1036 | static int bd_prepare_to_claim(struct block_device *bdev, |
| 1037 | struct block_device *whole, void *holder) |
| 1038 | { |
| 1039 | retry: |
| 1040 | /* if someone else claimed, fail */ |
| 1041 | if (!bd_may_claim(bdev, whole, holder)) |
| 1042 | return -EBUSY; |
| 1043 | |
| 1044 | /* if claiming is already in progress, wait for it to finish */ |
| 1045 | if (whole->bd_claiming) { |
| 1046 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
| 1047 | DEFINE_WAIT(wait); |
| 1048 | |
| 1049 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); |
| 1050 | spin_unlock(&bdev_lock); |
| 1051 | schedule(); |
| 1052 | finish_wait(wq, &wait); |
| 1053 | spin_lock(&bdev_lock); |
| 1054 | goto retry; |
| 1055 | } |
| 1056 | |
| 1057 | /* yay, all mine */ |
| 1058 | return 0; |
| 1059 | } |
| 1060 | |
| 1061 | /** |
| 1062 | * bd_start_claiming - start claiming a block device |
| 1063 | * @bdev: block device of interest |
| 1064 | * @holder: holder trying to claim @bdev |
| 1065 | * |
| 1066 | * @bdev is about to be opened exclusively. Check @bdev can be opened |
| 1067 | * exclusively and mark that an exclusive open is in progress. Each |
| 1068 | * successful call to this function must be matched with a call to |
| 1069 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
| 1070 | * fail). |
| 1071 | * |
| 1072 | * This function is used to gain exclusive access to the block device |
| 1073 | * without actually causing other exclusive open attempts to fail. It |
| 1074 | * should be used when the open sequence itself requires exclusive |
| 1075 | * access but may subsequently fail. |
| 1076 | * |
| 1077 | * CONTEXT: |
| 1078 | * Might sleep. |
| 1079 | * |
| 1080 | * RETURNS: |
| 1081 | * Pointer to the block device containing @bdev on success, ERR_PTR() |
| 1082 | * value on failure. |
| 1083 | */ |
| 1084 | static struct block_device *bd_start_claiming(struct block_device *bdev, |
| 1085 | void *holder) |
| 1086 | { |
| 1087 | struct gendisk *disk; |
| 1088 | struct block_device *whole; |
| 1089 | int partno, err; |
| 1090 | |
| 1091 | might_sleep(); |
| 1092 | |
| 1093 | /* |
| 1094 | * @bdev might not have been initialized properly yet, look up |
| 1095 | * and grab the outer block device the hard way. |
| 1096 | */ |
| 1097 | disk = get_gendisk(bdev->bd_dev, &partno); |
| 1098 | if (!disk) |
| 1099 | return ERR_PTR(-ENXIO); |
| 1100 | |
| 1101 | /* |
| 1102 | * Normally, @bdev should equal what's returned from bdget_disk() |
| 1103 | * if partno is 0; however, some drivers (floppy) use multiple |
| 1104 | * bdev's for the same physical device and @bdev may be one of the |
| 1105 | * aliases. Keep @bdev if partno is 0. This means claimer |
| 1106 | * tracking is broken for those devices but it has always been that |
| 1107 | * way. |
| 1108 | */ |
| 1109 | if (partno) |
| 1110 | whole = bdget_disk(disk, 0); |
| 1111 | else |
| 1112 | whole = bdgrab(bdev); |
| 1113 | |
| 1114 | put_disk_and_module(disk); |
| 1115 | if (!whole) |
| 1116 | return ERR_PTR(-ENOMEM); |
| 1117 | |
| 1118 | /* prepare to claim, if successful, mark claiming in progress */ |
| 1119 | spin_lock(&bdev_lock); |
| 1120 | |
| 1121 | err = bd_prepare_to_claim(bdev, whole, holder); |
| 1122 | if (err == 0) { |
| 1123 | whole->bd_claiming = holder; |
| 1124 | spin_unlock(&bdev_lock); |
| 1125 | return whole; |
| 1126 | } else { |
| 1127 | spin_unlock(&bdev_lock); |
| 1128 | bdput(whole); |
| 1129 | return ERR_PTR(err); |
| 1130 | } |
| 1131 | } |
| 1132 | |
| 1133 | #ifdef CONFIG_SYSFS |
| 1134 | struct bd_holder_disk { |
| 1135 | struct list_head list; |
| 1136 | struct gendisk *disk; |
| 1137 | int refcnt; |
| 1138 | }; |
| 1139 | |
| 1140 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, |
| 1141 | struct gendisk *disk) |
| 1142 | { |
| 1143 | struct bd_holder_disk *holder; |
| 1144 | |
| 1145 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) |
| 1146 | if (holder->disk == disk) |
| 1147 | return holder; |
| 1148 | return NULL; |
| 1149 | } |
| 1150 | |
| 1151 | static int add_symlink(struct kobject *from, struct kobject *to) |
| 1152 | { |
| 1153 | return sysfs_create_link(from, to, kobject_name(to)); |
| 1154 | } |
| 1155 | |
| 1156 | static void del_symlink(struct kobject *from, struct kobject *to) |
| 1157 | { |
| 1158 | sysfs_remove_link(from, kobject_name(to)); |
| 1159 | } |
| 1160 | |
| 1161 | /** |
| 1162 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
| 1163 | * @bdev: the claimed slave bdev |
| 1164 | * @disk: the holding disk |
| 1165 | * |
| 1166 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
| 1167 | * |
| 1168 | * This functions creates the following sysfs symlinks. |
| 1169 | * |
| 1170 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
| 1171 | * - from "holders" directory of the @bdev to the holder @disk |
| 1172 | * |
| 1173 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
| 1174 | * passed to bd_link_disk_holder(), then: |
| 1175 | * |
| 1176 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
| 1177 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 |
| 1178 | * |
| 1179 | * The caller must have claimed @bdev before calling this function and |
| 1180 | * ensure that both @bdev and @disk are valid during the creation and |
| 1181 | * lifetime of these symlinks. |
| 1182 | * |
| 1183 | * CONTEXT: |
| 1184 | * Might sleep. |
| 1185 | * |
| 1186 | * RETURNS: |
| 1187 | * 0 on success, -errno on failure. |
| 1188 | */ |
| 1189 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
| 1190 | { |
| 1191 | struct bd_holder_disk *holder; |
| 1192 | int ret = 0; |
| 1193 | |
| 1194 | mutex_lock(&bdev->bd_mutex); |
| 1195 | |
| 1196 | WARN_ON_ONCE(!bdev->bd_holder); |
| 1197 | |
| 1198 | /* FIXME: remove the following once add_disk() handles errors */ |
| 1199 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) |
| 1200 | goto out_unlock; |
| 1201 | |
| 1202 | holder = bd_find_holder_disk(bdev, disk); |
| 1203 | if (holder) { |
| 1204 | holder->refcnt++; |
| 1205 | goto out_unlock; |
| 1206 | } |
| 1207 | |
| 1208 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
| 1209 | if (!holder) { |
| 1210 | ret = -ENOMEM; |
| 1211 | goto out_unlock; |
| 1212 | } |
| 1213 | |
| 1214 | INIT_LIST_HEAD(&holder->list); |
| 1215 | holder->disk = disk; |
| 1216 | holder->refcnt = 1; |
| 1217 | |
| 1218 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 1219 | if (ret) |
| 1220 | goto out_free; |
| 1221 | |
| 1222 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); |
| 1223 | if (ret) |
| 1224 | goto out_del; |
| 1225 | /* |
| 1226 | * bdev could be deleted beneath us which would implicitly destroy |
| 1227 | * the holder directory. Hold on to it. |
| 1228 | */ |
| 1229 | kobject_get(bdev->bd_part->holder_dir); |
| 1230 | |
| 1231 | list_add(&holder->list, &bdev->bd_holder_disks); |
| 1232 | goto out_unlock; |
| 1233 | |
| 1234 | out_del: |
| 1235 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 1236 | out_free: |
| 1237 | kfree(holder); |
| 1238 | out_unlock: |
| 1239 | mutex_unlock(&bdev->bd_mutex); |
| 1240 | return ret; |
| 1241 | } |
| 1242 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
| 1243 | |
| 1244 | /** |
| 1245 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() |
| 1246 | * @bdev: the calimed slave bdev |
| 1247 | * @disk: the holding disk |
| 1248 | * |
| 1249 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
| 1250 | * |
| 1251 | * CONTEXT: |
| 1252 | * Might sleep. |
| 1253 | */ |
| 1254 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) |
| 1255 | { |
| 1256 | struct bd_holder_disk *holder; |
| 1257 | |
| 1258 | mutex_lock(&bdev->bd_mutex); |
| 1259 | |
| 1260 | holder = bd_find_holder_disk(bdev, disk); |
| 1261 | |
| 1262 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { |
| 1263 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
| 1264 | del_symlink(bdev->bd_part->holder_dir, |
| 1265 | &disk_to_dev(disk)->kobj); |
| 1266 | kobject_put(bdev->bd_part->holder_dir); |
| 1267 | list_del_init(&holder->list); |
| 1268 | kfree(holder); |
| 1269 | } |
| 1270 | |
| 1271 | mutex_unlock(&bdev->bd_mutex); |
| 1272 | } |
| 1273 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
| 1274 | #endif |
| 1275 | |
| 1276 | /** |
| 1277 | * flush_disk - invalidates all buffer-cache entries on a disk |
| 1278 | * |
| 1279 | * @bdev: struct block device to be flushed |
| 1280 | * @kill_dirty: flag to guide handling of dirty inodes |
| 1281 | * |
| 1282 | * Invalidates all buffer-cache entries on a disk. It should be called |
| 1283 | * when a disk has been changed -- either by a media change or online |
| 1284 | * resize. |
| 1285 | */ |
| 1286 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
| 1287 | { |
| 1288 | if (__invalidate_device(bdev, kill_dirty)) { |
| 1289 | printk(KERN_WARNING "VFS: busy inodes on changed media or " |
| 1290 | "resized disk %s\n", |
| 1291 | bdev->bd_disk ? bdev->bd_disk->disk_name : ""); |
| 1292 | } |
| 1293 | |
| 1294 | if (!bdev->bd_disk) |
| 1295 | return; |
| 1296 | if (disk_part_scan_enabled(bdev->bd_disk)) |
| 1297 | bdev->bd_invalidated = 1; |
| 1298 | } |
| 1299 | |
| 1300 | /** |
| 1301 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
| 1302 | * @disk: struct gendisk to check |
| 1303 | * @bdev: struct bdev to adjust. |
| 1304 | * |
| 1305 | * This routine checks to see if the bdev size does not match the disk size |
| 1306 | * and adjusts it if it differs. |
| 1307 | */ |
| 1308 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) |
| 1309 | { |
| 1310 | loff_t disk_size, bdev_size; |
| 1311 | |
| 1312 | disk_size = (loff_t)get_capacity(disk) << 9; |
| 1313 | bdev_size = i_size_read(bdev->bd_inode); |
| 1314 | if (disk_size != bdev_size) { |
| 1315 | printk(KERN_INFO |
| 1316 | "%s: detected capacity change from %lld to %lld\n", |
| 1317 | disk->disk_name, bdev_size, disk_size); |
| 1318 | i_size_write(bdev->bd_inode, disk_size); |
| 1319 | flush_disk(bdev, false); |
| 1320 | } |
| 1321 | } |
| 1322 | EXPORT_SYMBOL(check_disk_size_change); |
| 1323 | |
| 1324 | /** |
| 1325 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
| 1326 | * @disk: struct gendisk to be revalidated |
| 1327 | * |
| 1328 | * This routine is a wrapper for lower-level driver's revalidate_disk |
| 1329 | * call-backs. It is used to do common pre and post operations needed |
| 1330 | * for all revalidate_disk operations. |
| 1331 | */ |
| 1332 | int revalidate_disk(struct gendisk *disk) |
| 1333 | { |
| 1334 | struct block_device *bdev; |
| 1335 | int ret = 0; |
| 1336 | |
| 1337 | if (disk->fops->revalidate_disk) |
| 1338 | ret = disk->fops->revalidate_disk(disk); |
| 1339 | bdev = bdget_disk(disk, 0); |
| 1340 | if (!bdev) |
| 1341 | return ret; |
| 1342 | |
| 1343 | mutex_lock(&bdev->bd_mutex); |
| 1344 | check_disk_size_change(disk, bdev); |
| 1345 | bdev->bd_invalidated = 0; |
| 1346 | mutex_unlock(&bdev->bd_mutex); |
| 1347 | bdput(bdev); |
| 1348 | return ret; |
| 1349 | } |
| 1350 | EXPORT_SYMBOL(revalidate_disk); |
| 1351 | |
| 1352 | /* |
| 1353 | * This routine checks whether a removable media has been changed, |
| 1354 | * and invalidates all buffer-cache-entries in that case. This |
| 1355 | * is a relatively slow routine, so we have to try to minimize using |
| 1356 | * it. Thus it is called only upon a 'mount' or 'open'. This |
| 1357 | * is the best way of combining speed and utility, I think. |
| 1358 | * People changing diskettes in the middle of an operation deserve |
| 1359 | * to lose :-) |
| 1360 | */ |
| 1361 | int check_disk_change(struct block_device *bdev) |
| 1362 | { |
| 1363 | struct gendisk *disk = bdev->bd_disk; |
| 1364 | const struct block_device_operations *bdops = disk->fops; |
| 1365 | unsigned int events; |
| 1366 | |
| 1367 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
| 1368 | DISK_EVENT_EJECT_REQUEST); |
| 1369 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) |
| 1370 | return 0; |
| 1371 | |
| 1372 | flush_disk(bdev, true); |
| 1373 | if (bdops->revalidate_disk) |
| 1374 | bdops->revalidate_disk(bdev->bd_disk); |
| 1375 | return 1; |
| 1376 | } |
| 1377 | |
| 1378 | EXPORT_SYMBOL(check_disk_change); |
| 1379 | |
| 1380 | void bd_set_size(struct block_device *bdev, loff_t size) |
| 1381 | { |
| 1382 | unsigned bsize = bdev_logical_block_size(bdev); |
| 1383 | |
| 1384 | inode_lock(bdev->bd_inode); |
| 1385 | i_size_write(bdev->bd_inode, size); |
| 1386 | inode_unlock(bdev->bd_inode); |
| 1387 | while (bsize < PAGE_SIZE) { |
| 1388 | if (size & bsize) |
| 1389 | break; |
| 1390 | bsize <<= 1; |
| 1391 | } |
| 1392 | bdev->bd_block_size = bsize; |
| 1393 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); |
| 1394 | } |
| 1395 | EXPORT_SYMBOL(bd_set_size); |
| 1396 | |
| 1397 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
| 1398 | |
| 1399 | /* |
| 1400 | * bd_mutex locking: |
| 1401 | * |
| 1402 | * mutex_lock(part->bd_mutex) |
| 1403 | * mutex_lock_nested(whole->bd_mutex, 1) |
| 1404 | */ |
| 1405 | |
| 1406 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
| 1407 | { |
| 1408 | struct gendisk *disk; |
| 1409 | int ret; |
| 1410 | int partno; |
| 1411 | int perm = 0; |
| 1412 | bool first_open = false; |
| 1413 | |
| 1414 | if (mode & FMODE_READ) |
| 1415 | perm |= MAY_READ; |
| 1416 | if (mode & FMODE_WRITE) |
| 1417 | perm |= MAY_WRITE; |
| 1418 | /* |
| 1419 | * hooks: /n/, see "layering violations". |
| 1420 | */ |
| 1421 | if (!for_part) { |
| 1422 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); |
| 1423 | if (ret != 0) { |
| 1424 | bdput(bdev); |
| 1425 | return ret; |
| 1426 | } |
| 1427 | } |
| 1428 | |
| 1429 | restart: |
| 1430 | |
| 1431 | ret = -ENXIO; |
| 1432 | disk = get_gendisk(bdev->bd_dev, &partno); |
| 1433 | if (!disk) |
| 1434 | goto out; |
| 1435 | |
| 1436 | disk_block_events(disk); |
| 1437 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
| 1438 | if (!bdev->bd_openers) { |
| 1439 | first_open = true; |
| 1440 | bdev->bd_disk = disk; |
| 1441 | bdev->bd_queue = disk->queue; |
| 1442 | bdev->bd_contains = bdev; |
| 1443 | bdev->bd_partno = partno; |
| 1444 | |
| 1445 | if (!partno) { |
| 1446 | ret = -ENXIO; |
| 1447 | bdev->bd_part = disk_get_part(disk, partno); |
| 1448 | if (!bdev->bd_part) |
| 1449 | goto out_clear; |
| 1450 | |
| 1451 | ret = 0; |
| 1452 | if (disk->fops->open) { |
| 1453 | ret = disk->fops->open(bdev, mode); |
| 1454 | if (ret == -ERESTARTSYS) { |
| 1455 | /* Lost a race with 'disk' being |
| 1456 | * deleted, try again. |
| 1457 | * See md.c |
| 1458 | */ |
| 1459 | disk_put_part(bdev->bd_part); |
| 1460 | bdev->bd_part = NULL; |
| 1461 | bdev->bd_disk = NULL; |
| 1462 | bdev->bd_queue = NULL; |
| 1463 | mutex_unlock(&bdev->bd_mutex); |
| 1464 | disk_unblock_events(disk); |
| 1465 | put_disk_and_module(disk); |
| 1466 | goto restart; |
| 1467 | } |
| 1468 | } |
| 1469 | |
| 1470 | if (!ret) |
| 1471 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
| 1472 | |
| 1473 | /* |
| 1474 | * If the device is invalidated, rescan partition |
| 1475 | * if open succeeded or failed with -ENOMEDIUM. |
| 1476 | * The latter is necessary to prevent ghost |
| 1477 | * partitions on a removed medium. |
| 1478 | */ |
| 1479 | if (bdev->bd_invalidated) { |
| 1480 | if (!ret) |
| 1481 | rescan_partitions(disk, bdev); |
| 1482 | else if (ret == -ENOMEDIUM) |
| 1483 | invalidate_partitions(disk, bdev); |
| 1484 | } |
| 1485 | |
| 1486 | if (ret) |
| 1487 | goto out_clear; |
| 1488 | } else { |
| 1489 | struct block_device *whole; |
| 1490 | whole = bdget_disk(disk, 0); |
| 1491 | ret = -ENOMEM; |
| 1492 | if (!whole) |
| 1493 | goto out_clear; |
| 1494 | BUG_ON(for_part); |
| 1495 | ret = __blkdev_get(whole, mode, 1); |
| 1496 | if (ret) |
| 1497 | goto out_clear; |
| 1498 | bdev->bd_contains = whole; |
| 1499 | bdev->bd_part = disk_get_part(disk, partno); |
| 1500 | if (!(disk->flags & GENHD_FL_UP) || |
| 1501 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
| 1502 | ret = -ENXIO; |
| 1503 | goto out_clear; |
| 1504 | } |
| 1505 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
| 1506 | } |
| 1507 | |
| 1508 | if (bdev->bd_bdi == &noop_backing_dev_info) |
| 1509 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info); |
| 1510 | } else { |
| 1511 | if (bdev->bd_contains == bdev) { |
| 1512 | ret = 0; |
| 1513 | if (bdev->bd_disk->fops->open) |
| 1514 | ret = bdev->bd_disk->fops->open(bdev, mode); |
| 1515 | /* the same as first opener case, read comment there */ |
| 1516 | if (bdev->bd_invalidated) { |
| 1517 | if (!ret) |
| 1518 | rescan_partitions(bdev->bd_disk, bdev); |
| 1519 | else if (ret == -ENOMEDIUM) |
| 1520 | invalidate_partitions(bdev->bd_disk, bdev); |
| 1521 | } |
| 1522 | if (ret) |
| 1523 | goto out_unlock_bdev; |
| 1524 | } |
| 1525 | } |
| 1526 | bdev->bd_openers++; |
| 1527 | if (for_part) |
| 1528 | bdev->bd_part_count++; |
| 1529 | mutex_unlock(&bdev->bd_mutex); |
| 1530 | disk_unblock_events(disk); |
| 1531 | /* only one opener holds refs to the module and disk */ |
| 1532 | if (!first_open) |
| 1533 | put_disk_and_module(disk); |
| 1534 | return 0; |
| 1535 | |
| 1536 | out_clear: |
| 1537 | disk_put_part(bdev->bd_part); |
| 1538 | bdev->bd_disk = NULL; |
| 1539 | bdev->bd_part = NULL; |
| 1540 | bdev->bd_queue = NULL; |
| 1541 | if (bdev != bdev->bd_contains) |
| 1542 | __blkdev_put(bdev->bd_contains, mode, 1); |
| 1543 | bdev->bd_contains = NULL; |
| 1544 | out_unlock_bdev: |
| 1545 | mutex_unlock(&bdev->bd_mutex); |
| 1546 | disk_unblock_events(disk); |
| 1547 | put_disk_and_module(disk); |
| 1548 | out: |
| 1549 | bdput(bdev); |
| 1550 | |
| 1551 | return ret; |
| 1552 | } |
| 1553 | |
| 1554 | /** |
| 1555 | * blkdev_get - open a block device |
| 1556 | * @bdev: block_device to open |
| 1557 | * @mode: FMODE_* mask |
| 1558 | * @holder: exclusive holder identifier |
| 1559 | * |
| 1560 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is |
| 1561 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL |
| 1562 | * @holder is invalid. Exclusive opens may nest for the same @holder. |
| 1563 | * |
| 1564 | * On success, the reference count of @bdev is unchanged. On failure, |
| 1565 | * @bdev is put. |
| 1566 | * |
| 1567 | * CONTEXT: |
| 1568 | * Might sleep. |
| 1569 | * |
| 1570 | * RETURNS: |
| 1571 | * 0 on success, -errno on failure. |
| 1572 | */ |
| 1573 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
| 1574 | { |
| 1575 | struct block_device *whole = NULL; |
| 1576 | int res; |
| 1577 | |
| 1578 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); |
| 1579 | |
| 1580 | if ((mode & FMODE_EXCL) && holder) { |
| 1581 | whole = bd_start_claiming(bdev, holder); |
| 1582 | if (IS_ERR(whole)) { |
| 1583 | bdput(bdev); |
| 1584 | return PTR_ERR(whole); |
| 1585 | } |
| 1586 | } |
| 1587 | |
| 1588 | res = __blkdev_get(bdev, mode, 0); |
| 1589 | |
| 1590 | if (whole) { |
| 1591 | struct gendisk *disk = whole->bd_disk; |
| 1592 | |
| 1593 | /* finish claiming */ |
| 1594 | mutex_lock(&bdev->bd_mutex); |
| 1595 | spin_lock(&bdev_lock); |
| 1596 | |
| 1597 | if (!res) { |
| 1598 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
| 1599 | /* |
| 1600 | * Note that for a whole device bd_holders |
| 1601 | * will be incremented twice, and bd_holder |
| 1602 | * will be set to bd_may_claim before being |
| 1603 | * set to holder |
| 1604 | */ |
| 1605 | whole->bd_holders++; |
| 1606 | whole->bd_holder = bd_may_claim; |
| 1607 | bdev->bd_holders++; |
| 1608 | bdev->bd_holder = holder; |
| 1609 | } |
| 1610 | |
| 1611 | /* tell others that we're done */ |
| 1612 | BUG_ON(whole->bd_claiming != holder); |
| 1613 | whole->bd_claiming = NULL; |
| 1614 | wake_up_bit(&whole->bd_claiming, 0); |
| 1615 | |
| 1616 | spin_unlock(&bdev_lock); |
| 1617 | |
| 1618 | /* |
| 1619 | * Block event polling for write claims if requested. Any |
| 1620 | * write holder makes the write_holder state stick until |
| 1621 | * all are released. This is good enough and tracking |
| 1622 | * individual writeable reference is too fragile given the |
| 1623 | * way @mode is used in blkdev_get/put(). |
| 1624 | */ |
| 1625 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
| 1626 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { |
| 1627 | bdev->bd_write_holder = true; |
| 1628 | disk_block_events(disk); |
| 1629 | } |
| 1630 | |
| 1631 | mutex_unlock(&bdev->bd_mutex); |
| 1632 | bdput(whole); |
| 1633 | } |
| 1634 | |
| 1635 | return res; |
| 1636 | } |
| 1637 | EXPORT_SYMBOL(blkdev_get); |
| 1638 | |
| 1639 | /** |
| 1640 | * blkdev_get_by_path - open a block device by name |
| 1641 | * @path: path to the block device to open |
| 1642 | * @mode: FMODE_* mask |
| 1643 | * @holder: exclusive holder identifier |
| 1644 | * |
| 1645 | * Open the blockdevice described by the device file at @path. @mode |
| 1646 | * and @holder are identical to blkdev_get(). |
| 1647 | * |
| 1648 | * On success, the returned block_device has reference count of one. |
| 1649 | * |
| 1650 | * CONTEXT: |
| 1651 | * Might sleep. |
| 1652 | * |
| 1653 | * RETURNS: |
| 1654 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
| 1655 | */ |
| 1656 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, |
| 1657 | void *holder) |
| 1658 | { |
| 1659 | struct block_device *bdev; |
| 1660 | int err; |
| 1661 | |
| 1662 | bdev = lookup_bdev(path); |
| 1663 | if (IS_ERR(bdev)) |
| 1664 | return bdev; |
| 1665 | |
| 1666 | err = blkdev_get(bdev, mode, holder); |
| 1667 | if (err) |
| 1668 | return ERR_PTR(err); |
| 1669 | |
| 1670 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
| 1671 | blkdev_put(bdev, mode); |
| 1672 | return ERR_PTR(-EACCES); |
| 1673 | } |
| 1674 | |
| 1675 | return bdev; |
| 1676 | } |
| 1677 | EXPORT_SYMBOL(blkdev_get_by_path); |
| 1678 | |
| 1679 | /** |
| 1680 | * blkdev_get_by_dev - open a block device by device number |
| 1681 | * @dev: device number of block device to open |
| 1682 | * @mode: FMODE_* mask |
| 1683 | * @holder: exclusive holder identifier |
| 1684 | * |
| 1685 | * Open the blockdevice described by device number @dev. @mode and |
| 1686 | * @holder are identical to blkdev_get(). |
| 1687 | * |
| 1688 | * Use it ONLY if you really do not have anything better - i.e. when |
| 1689 | * you are behind a truly sucky interface and all you are given is a |
| 1690 | * device number. _Never_ to be used for internal purposes. If you |
| 1691 | * ever need it - reconsider your API. |
| 1692 | * |
| 1693 | * On success, the returned block_device has reference count of one. |
| 1694 | * |
| 1695 | * CONTEXT: |
| 1696 | * Might sleep. |
| 1697 | * |
| 1698 | * RETURNS: |
| 1699 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. |
| 1700 | */ |
| 1701 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) |
| 1702 | { |
| 1703 | struct block_device *bdev; |
| 1704 | int err; |
| 1705 | |
| 1706 | bdev = bdget(dev); |
| 1707 | if (!bdev) |
| 1708 | return ERR_PTR(-ENOMEM); |
| 1709 | |
| 1710 | err = blkdev_get(bdev, mode, holder); |
| 1711 | if (err) |
| 1712 | return ERR_PTR(err); |
| 1713 | |
| 1714 | return bdev; |
| 1715 | } |
| 1716 | EXPORT_SYMBOL(blkdev_get_by_dev); |
| 1717 | |
| 1718 | static int blkdev_open(struct inode * inode, struct file * filp) |
| 1719 | { |
| 1720 | struct block_device *bdev; |
| 1721 | |
| 1722 | /* |
| 1723 | * Preserve backwards compatibility and allow large file access |
| 1724 | * even if userspace doesn't ask for it explicitly. Some mkfs |
| 1725 | * binary needs it. We might want to drop this workaround |
| 1726 | * during an unstable branch. |
| 1727 | */ |
| 1728 | filp->f_flags |= O_LARGEFILE; |
| 1729 | |
| 1730 | filp->f_mode |= FMODE_NOWAIT; |
| 1731 | |
| 1732 | if (filp->f_flags & O_NDELAY) |
| 1733 | filp->f_mode |= FMODE_NDELAY; |
| 1734 | if (filp->f_flags & O_EXCL) |
| 1735 | filp->f_mode |= FMODE_EXCL; |
| 1736 | if ((filp->f_flags & O_ACCMODE) == 3) |
| 1737 | filp->f_mode |= FMODE_WRITE_IOCTL; |
| 1738 | |
| 1739 | bdev = bd_acquire(inode); |
| 1740 | if (bdev == NULL) |
| 1741 | return -ENOMEM; |
| 1742 | |
| 1743 | filp->f_mapping = bdev->bd_inode->i_mapping; |
| 1744 | filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping); |
| 1745 | |
| 1746 | return blkdev_get(bdev, filp->f_mode, filp); |
| 1747 | } |
| 1748 | |
| 1749 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
| 1750 | { |
| 1751 | struct gendisk *disk = bdev->bd_disk; |
| 1752 | struct block_device *victim = NULL; |
| 1753 | |
| 1754 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
| 1755 | if (for_part) |
| 1756 | bdev->bd_part_count--; |
| 1757 | |
| 1758 | if (!--bdev->bd_openers) { |
| 1759 | WARN_ON_ONCE(bdev->bd_holders); |
| 1760 | sync_blockdev(bdev); |
| 1761 | kill_bdev(bdev); |
| 1762 | |
| 1763 | bdev_write_inode(bdev); |
| 1764 | } |
| 1765 | if (bdev->bd_contains == bdev) { |
| 1766 | if (disk->fops->release) |
| 1767 | disk->fops->release(disk, mode); |
| 1768 | } |
| 1769 | if (!bdev->bd_openers) { |
| 1770 | disk_put_part(bdev->bd_part); |
| 1771 | bdev->bd_part = NULL; |
| 1772 | bdev->bd_disk = NULL; |
| 1773 | if (bdev != bdev->bd_contains) |
| 1774 | victim = bdev->bd_contains; |
| 1775 | bdev->bd_contains = NULL; |
| 1776 | |
| 1777 | put_disk_and_module(disk); |
| 1778 | } |
| 1779 | mutex_unlock(&bdev->bd_mutex); |
| 1780 | bdput(bdev); |
| 1781 | if (victim) |
| 1782 | __blkdev_put(victim, mode, 1); |
| 1783 | } |
| 1784 | |
| 1785 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
| 1786 | { |
| 1787 | mutex_lock(&bdev->bd_mutex); |
| 1788 | |
| 1789 | if (mode & FMODE_EXCL) { |
| 1790 | bool bdev_free; |
| 1791 | |
| 1792 | /* |
| 1793 | * Release a claim on the device. The holder fields |
| 1794 | * are protected with bdev_lock. bd_mutex is to |
| 1795 | * synchronize disk_holder unlinking. |
| 1796 | */ |
| 1797 | spin_lock(&bdev_lock); |
| 1798 | |
| 1799 | WARN_ON_ONCE(--bdev->bd_holders < 0); |
| 1800 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); |
| 1801 | |
| 1802 | /* bd_contains might point to self, check in a separate step */ |
| 1803 | if ((bdev_free = !bdev->bd_holders)) |
| 1804 | bdev->bd_holder = NULL; |
| 1805 | if (!bdev->bd_contains->bd_holders) |
| 1806 | bdev->bd_contains->bd_holder = NULL; |
| 1807 | |
| 1808 | spin_unlock(&bdev_lock); |
| 1809 | |
| 1810 | /* |
| 1811 | * If this was the last claim, remove holder link and |
| 1812 | * unblock evpoll if it was a write holder. |
| 1813 | */ |
| 1814 | if (bdev_free && bdev->bd_write_holder) { |
| 1815 | disk_unblock_events(bdev->bd_disk); |
| 1816 | bdev->bd_write_holder = false; |
| 1817 | } |
| 1818 | } |
| 1819 | |
| 1820 | /* |
| 1821 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE |
| 1822 | * event. This is to ensure detection of media removal commanded |
| 1823 | * from userland - e.g. eject(1). |
| 1824 | */ |
| 1825 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); |
| 1826 | |
| 1827 | mutex_unlock(&bdev->bd_mutex); |
| 1828 | |
| 1829 | __blkdev_put(bdev, mode, 0); |
| 1830 | } |
| 1831 | EXPORT_SYMBOL(blkdev_put); |
| 1832 | |
| 1833 | static int blkdev_close(struct inode * inode, struct file * filp) |
| 1834 | { |
| 1835 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp)); |
| 1836 | blkdev_put(bdev, filp->f_mode); |
| 1837 | return 0; |
| 1838 | } |
| 1839 | |
| 1840 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
| 1841 | { |
| 1842 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
| 1843 | fmode_t mode = file->f_mode; |
| 1844 | |
| 1845 | /* |
| 1846 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have |
| 1847 | * to updated it before every ioctl. |
| 1848 | */ |
| 1849 | if (file->f_flags & O_NDELAY) |
| 1850 | mode |= FMODE_NDELAY; |
| 1851 | else |
| 1852 | mode &= ~FMODE_NDELAY; |
| 1853 | |
| 1854 | return blkdev_ioctl(bdev, mode, cmd, arg); |
| 1855 | } |
| 1856 | |
| 1857 | /* |
| 1858 | * Write data to the block device. Only intended for the block device itself |
| 1859 | * and the raw driver which basically is a fake block device. |
| 1860 | * |
| 1861 | * Does not take i_mutex for the write and thus is not for general purpose |
| 1862 | * use. |
| 1863 | */ |
| 1864 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) |
| 1865 | { |
| 1866 | struct file *file = iocb->ki_filp; |
| 1867 | struct inode *bd_inode = bdev_file_inode(file); |
| 1868 | loff_t size = i_size_read(bd_inode); |
| 1869 | struct blk_plug plug; |
| 1870 | ssize_t ret; |
| 1871 | |
| 1872 | if (bdev_read_only(I_BDEV(bd_inode))) |
| 1873 | return -EPERM; |
| 1874 | |
| 1875 | if (!iov_iter_count(from)) |
| 1876 | return 0; |
| 1877 | |
| 1878 | if (iocb->ki_pos >= size) |
| 1879 | return -ENOSPC; |
| 1880 | |
| 1881 | if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) |
| 1882 | return -EOPNOTSUPP; |
| 1883 | |
| 1884 | iov_iter_truncate(from, size - iocb->ki_pos); |
| 1885 | |
| 1886 | blk_start_plug(&plug); |
| 1887 | ret = __generic_file_write_iter(iocb, from); |
| 1888 | if (ret > 0) |
| 1889 | ret = generic_write_sync(iocb, ret); |
| 1890 | blk_finish_plug(&plug); |
| 1891 | return ret; |
| 1892 | } |
| 1893 | EXPORT_SYMBOL_GPL(blkdev_write_iter); |
| 1894 | |
| 1895 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) |
| 1896 | { |
| 1897 | struct file *file = iocb->ki_filp; |
| 1898 | struct inode *bd_inode = bdev_file_inode(file); |
| 1899 | loff_t size = i_size_read(bd_inode); |
| 1900 | loff_t pos = iocb->ki_pos; |
| 1901 | |
| 1902 | if (pos >= size) |
| 1903 | return 0; |
| 1904 | |
| 1905 | size -= pos; |
| 1906 | iov_iter_truncate(to, size); |
| 1907 | return generic_file_read_iter(iocb, to); |
| 1908 | } |
| 1909 | EXPORT_SYMBOL_GPL(blkdev_read_iter); |
| 1910 | |
| 1911 | /* |
| 1912 | * Try to release a page associated with block device when the system |
| 1913 | * is under memory pressure. |
| 1914 | */ |
| 1915 | static int blkdev_releasepage(struct page *page, gfp_t wait) |
| 1916 | { |
| 1917 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; |
| 1918 | |
| 1919 | if (super && super->s_op->bdev_try_to_free_page) |
| 1920 | return super->s_op->bdev_try_to_free_page(super, page, wait); |
| 1921 | |
| 1922 | return try_to_free_buffers(page); |
| 1923 | } |
| 1924 | |
| 1925 | static int blkdev_writepages(struct address_space *mapping, |
| 1926 | struct writeback_control *wbc) |
| 1927 | { |
| 1928 | if (dax_mapping(mapping)) { |
| 1929 | struct block_device *bdev = I_BDEV(mapping->host); |
| 1930 | |
| 1931 | return dax_writeback_mapping_range(mapping, bdev, wbc); |
| 1932 | } |
| 1933 | return generic_writepages(mapping, wbc); |
| 1934 | } |
| 1935 | |
| 1936 | static const struct address_space_operations def_blk_aops = { |
| 1937 | .readpage = blkdev_readpage, |
| 1938 | .readpages = blkdev_readpages, |
| 1939 | .writepage = blkdev_writepage, |
| 1940 | .write_begin = blkdev_write_begin, |
| 1941 | .write_end = blkdev_write_end, |
| 1942 | .writepages = blkdev_writepages, |
| 1943 | .releasepage = blkdev_releasepage, |
| 1944 | .direct_IO = blkdev_direct_IO, |
| 1945 | .is_dirty_writeback = buffer_check_dirty_writeback, |
| 1946 | }; |
| 1947 | |
| 1948 | #define BLKDEV_FALLOC_FL_SUPPORTED \ |
| 1949 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ |
| 1950 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) |
| 1951 | |
| 1952 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, |
| 1953 | loff_t len) |
| 1954 | { |
| 1955 | struct block_device *bdev = I_BDEV(bdev_file_inode(file)); |
| 1956 | struct address_space *mapping; |
| 1957 | loff_t end = start + len - 1; |
| 1958 | loff_t isize; |
| 1959 | int error; |
| 1960 | |
| 1961 | /* Fail if we don't recognize the flags. */ |
| 1962 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) |
| 1963 | return -EOPNOTSUPP; |
| 1964 | |
| 1965 | /* Don't go off the end of the device. */ |
| 1966 | isize = i_size_read(bdev->bd_inode); |
| 1967 | if (start >= isize) |
| 1968 | return -EINVAL; |
| 1969 | if (end >= isize) { |
| 1970 | if (mode & FALLOC_FL_KEEP_SIZE) { |
| 1971 | len = isize - start; |
| 1972 | end = start + len - 1; |
| 1973 | } else |
| 1974 | return -EINVAL; |
| 1975 | } |
| 1976 | |
| 1977 | /* |
| 1978 | * Don't allow IO that isn't aligned to logical block size. |
| 1979 | */ |
| 1980 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) |
| 1981 | return -EINVAL; |
| 1982 | |
| 1983 | /* Invalidate the page cache, including dirty pages. */ |
| 1984 | mapping = bdev->bd_inode->i_mapping; |
| 1985 | truncate_inode_pages_range(mapping, start, end); |
| 1986 | |
| 1987 | switch (mode) { |
| 1988 | case FALLOC_FL_ZERO_RANGE: |
| 1989 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: |
| 1990 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
| 1991 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP); |
| 1992 | break; |
| 1993 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: |
| 1994 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, |
| 1995 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK); |
| 1996 | break; |
| 1997 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: |
| 1998 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9, |
| 1999 | GFP_KERNEL, 0); |
| 2000 | break; |
| 2001 | default: |
| 2002 | return -EOPNOTSUPP; |
| 2003 | } |
| 2004 | if (error) |
| 2005 | return error; |
| 2006 | |
| 2007 | /* |
| 2008 | * Invalidate again; if someone wandered in and dirtied a page, |
| 2009 | * the caller will be given -EBUSY. The third argument is |
| 2010 | * inclusive, so the rounding here is safe. |
| 2011 | */ |
| 2012 | return invalidate_inode_pages2_range(mapping, |
| 2013 | start >> PAGE_SHIFT, |
| 2014 | end >> PAGE_SHIFT); |
| 2015 | } |
| 2016 | |
| 2017 | const struct file_operations def_blk_fops = { |
| 2018 | .open = blkdev_open, |
| 2019 | .release = blkdev_close, |
| 2020 | .llseek = block_llseek, |
| 2021 | .read_iter = blkdev_read_iter, |
| 2022 | .write_iter = blkdev_write_iter, |
| 2023 | .mmap = generic_file_mmap, |
| 2024 | .fsync = blkdev_fsync, |
| 2025 | .unlocked_ioctl = block_ioctl, |
| 2026 | #ifdef CONFIG_COMPAT |
| 2027 | .compat_ioctl = compat_blkdev_ioctl, |
| 2028 | #endif |
| 2029 | .splice_read = generic_file_splice_read, |
| 2030 | .splice_write = iter_file_splice_write, |
| 2031 | .fallocate = blkdev_fallocate, |
| 2032 | }; |
| 2033 | |
| 2034 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) |
| 2035 | { |
| 2036 | int res; |
| 2037 | mm_segment_t old_fs = get_fs(); |
| 2038 | set_fs(KERNEL_DS); |
| 2039 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
| 2040 | set_fs(old_fs); |
| 2041 | return res; |
| 2042 | } |
| 2043 | |
| 2044 | EXPORT_SYMBOL(ioctl_by_bdev); |
| 2045 | |
| 2046 | /** |
| 2047 | * lookup_bdev - lookup a struct block_device by name |
| 2048 | * @pathname: special file representing the block device |
| 2049 | * |
| 2050 | * Get a reference to the blockdevice at @pathname in the current |
| 2051 | * namespace if possible and return it. Return ERR_PTR(error) |
| 2052 | * otherwise. |
| 2053 | */ |
| 2054 | struct block_device *lookup_bdev(const char *pathname) |
| 2055 | { |
| 2056 | struct block_device *bdev; |
| 2057 | struct inode *inode; |
| 2058 | struct path path; |
| 2059 | int error; |
| 2060 | |
| 2061 | if (!pathname || !*pathname) |
| 2062 | return ERR_PTR(-EINVAL); |
| 2063 | |
| 2064 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
| 2065 | if (error) |
| 2066 | return ERR_PTR(error); |
| 2067 | |
| 2068 | inode = d_backing_inode(path.dentry); |
| 2069 | error = -ENOTBLK; |
| 2070 | if (!S_ISBLK(inode->i_mode)) |
| 2071 | goto fail; |
| 2072 | error = -EACCES; |
| 2073 | if (!may_open_dev(&path)) |
| 2074 | goto fail; |
| 2075 | error = -ENOMEM; |
| 2076 | bdev = bd_acquire(inode); |
| 2077 | if (!bdev) |
| 2078 | goto fail; |
| 2079 | out: |
| 2080 | path_put(&path); |
| 2081 | return bdev; |
| 2082 | fail: |
| 2083 | bdev = ERR_PTR(error); |
| 2084 | goto out; |
| 2085 | } |
| 2086 | EXPORT_SYMBOL(lookup_bdev); |
| 2087 | |
| 2088 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
| 2089 | { |
| 2090 | struct super_block *sb = get_super(bdev); |
| 2091 | int res = 0; |
| 2092 | |
| 2093 | if (sb) { |
| 2094 | /* |
| 2095 | * no need to lock the super, get_super holds the |
| 2096 | * read mutex so the filesystem cannot go away |
| 2097 | * under us (->put_super runs with the write lock |
| 2098 | * hold). |
| 2099 | */ |
| 2100 | shrink_dcache_sb(sb); |
| 2101 | res = invalidate_inodes(sb, kill_dirty); |
| 2102 | drop_super(sb); |
| 2103 | } |
| 2104 | invalidate_bdev(bdev); |
| 2105 | return res; |
| 2106 | } |
| 2107 | EXPORT_SYMBOL(__invalidate_device); |
| 2108 | |
| 2109 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) |
| 2110 | { |
| 2111 | struct inode *inode, *old_inode = NULL; |
| 2112 | |
| 2113 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
| 2114 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
| 2115 | struct address_space *mapping = inode->i_mapping; |
| 2116 | struct block_device *bdev; |
| 2117 | |
| 2118 | spin_lock(&inode->i_lock); |
| 2119 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || |
| 2120 | mapping->nrpages == 0) { |
| 2121 | spin_unlock(&inode->i_lock); |
| 2122 | continue; |
| 2123 | } |
| 2124 | __iget(inode); |
| 2125 | spin_unlock(&inode->i_lock); |
| 2126 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
| 2127 | /* |
| 2128 | * We hold a reference to 'inode' so it couldn't have been |
| 2129 | * removed from s_inodes list while we dropped the |
| 2130 | * s_inode_list_lock We cannot iput the inode now as we can |
| 2131 | * be holding the last reference and we cannot iput it under |
| 2132 | * s_inode_list_lock. So we keep the reference and iput it |
| 2133 | * later. |
| 2134 | */ |
| 2135 | iput(old_inode); |
| 2136 | old_inode = inode; |
| 2137 | bdev = I_BDEV(inode); |
| 2138 | |
| 2139 | mutex_lock(&bdev->bd_mutex); |
| 2140 | if (bdev->bd_openers) |
| 2141 | func(bdev, arg); |
| 2142 | mutex_unlock(&bdev->bd_mutex); |
| 2143 | |
| 2144 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
| 2145 | } |
| 2146 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
| 2147 | iput(old_inode); |
| 2148 | } |