| 1 | |
| 2 | |
| 3 | * Copyright (C) 1991, 1992 Linus Torvalds
|
| 4 | * Copyright (C) 2001 Andrea Arcangeli <[email protected]> SuSE
|
| 5 | * Copyright (C) 2016 - 2020 Christoph Hellwig
|
| 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/buffer_head.h>
|
| 22 | #include <linux/swap.h>
|
| 23 | #include <linux/pagevec.h>
|
| 24 | #include <linux/writeback.h>
|
| 25 | #include <linux/mpage.h>
|
| 26 | #include <linux/mount.h>
|
| 27 | #include <linux/pseudo_fs.h>
|
| 28 | #include <linux/uio.h>
|
| 29 | #include <linux/namei.h>
|
| 30 | #include <linux/log2.h>
|
| 31 | #include <linux/cleancache.h>
|
| 32 | #include <linux/task_io_accounting_ops.h>
|
| 33 | #include <linux/falloc.h>
|
| 34 | #include <linux/part_stat.h>
|
| 35 | #include <linux/uaccess.h>
|
| 36 | #include <linux/suspend.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 |
|
| 78 | static 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 |
|
| 89 |
|
| 90 | void invalidate_bdev(struct block_device *bdev)
|
| 91 | {
|
| 92 | struct address_space *mapping = bdev->bd_inode->i_mapping;
|
| 93 |
|
| 94 | if (mapping->nrpages) {
|
| 95 | invalidate_bh_lrus();
|
| 96 | lru_add_drain_all();
|
| 97 | invalidate_mapping_pages(mapping, 0, -1);
|
| 98 | }
|
| 99 | |
| 100 | * But, for the strange corners, lets be cautious
|
| 101 |
|
| 102 | cleancache_invalidate_inode(mapping);
|
| 103 | }
|
| 104 | EXPORT_SYMBOL(invalidate_bdev);
|
| 105 |
|
| 106 | |
| 107 | * Drop all buffers & page cache for given bdev range. This function bails
|
| 108 | * with error if bdev has other exclusive owner (such as filesystem).
|
| 109 |
|
| 110 | int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
|
| 111 | loff_t lstart, loff_t lend)
|
| 112 | {
|
| 113 | |
| 114 | * If we don't hold exclusive handle for the device, upgrade to it
|
| 115 | * while we discard the buffer cache to avoid discarding buffers
|
| 116 | * under live filesystem.
|
| 117 |
|
| 118 | if (!(mode & FMODE_EXCL)) {
|
| 119 | int err = bd_prepare_to_claim(bdev, truncate_bdev_range);
|
| 120 | if (err)
|
| 121 | return err;
|
| 122 | }
|
| 123 |
|
| 124 | truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
|
| 125 | if (!(mode & FMODE_EXCL))
|
| 126 | bd_abort_claiming(bdev, truncate_bdev_range);
|
| 127 | return 0;
|
| 128 | }
|
| 129 | EXPORT_SYMBOL(truncate_bdev_range);
|
| 130 |
|
| 131 | static void set_init_blocksize(struct block_device *bdev)
|
| 132 | {
|
| 133 | unsigned int bsize = bdev_logical_block_size(bdev);
|
| 134 | loff_t size = i_size_read(bdev->bd_inode);
|
| 135 |
|
| 136 | while (bsize < PAGE_SIZE) {
|
| 137 | if (size & bsize)
|
| 138 | break;
|
| 139 | bsize <<= 1;
|
| 140 | }
|
| 141 | bdev->bd_inode->i_blkbits = blksize_bits(bsize);
|
| 142 | }
|
| 143 |
|
| 144 | int set_blocksize(struct block_device *bdev, int size)
|
| 145 | {
|
| 146 |
|
| 147 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
|
| 148 | return -EINVAL;
|
| 149 |
|
| 150 |
|
| 151 | if (size < bdev_logical_block_size(bdev))
|
| 152 | return -EINVAL;
|
| 153 |
|
| 154 |
|
| 155 | if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
|
| 156 | sync_blockdev(bdev);
|
| 157 | bdev->bd_inode->i_blkbits = blksize_bits(size);
|
| 158 | kill_bdev(bdev);
|
| 159 | }
|
| 160 | return 0;
|
| 161 | }
|
| 162 |
|
| 163 | EXPORT_SYMBOL(set_blocksize);
|
| 164 |
|
| 165 | int sb_set_blocksize(struct super_block *sb, int size)
|
| 166 | {
|
| 167 | if (set_blocksize(sb->s_bdev, size))
|
| 168 | return 0;
|
| 169 | |
| 170 | * and it's value is between 512 and PAGE_SIZE
|
| 171 | sb->s_blocksize = size;
|
| 172 | sb->s_blocksize_bits = blksize_bits(size);
|
| 173 | return sb->s_blocksize;
|
| 174 | }
|
| 175 |
|
| 176 | EXPORT_SYMBOL(sb_set_blocksize);
|
| 177 |
|
| 178 | int sb_min_blocksize(struct super_block *sb, int size)
|
| 179 | {
|
| 180 | int minsize = bdev_logical_block_size(sb->s_bdev);
|
| 181 | if (size < minsize)
|
| 182 | size = minsize;
|
| 183 | return sb_set_blocksize(sb, size);
|
| 184 | }
|
| 185 |
|
| 186 | EXPORT_SYMBOL(sb_min_blocksize);
|
| 187 |
|
| 188 | static int
|
| 189 | blkdev_get_block(struct inode *inode, sector_t iblock,
|
| 190 | struct buffer_head *bh, int create)
|
| 191 | {
|
| 192 | bh->b_bdev = I_BDEV(inode);
|
| 193 | bh->b_blocknr = iblock;
|
| 194 | set_buffer_mapped(bh);
|
| 195 | return 0;
|
| 196 | }
|
| 197 |
|
| 198 | static struct inode *bdev_file_inode(struct file *file)
|
| 199 | {
|
| 200 | return file->f_mapping->host;
|
| 201 | }
|
| 202 |
|
| 203 | static unsigned int dio_bio_write_op(struct kiocb *iocb)
|
| 204 | {
|
| 205 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
|
| 206 |
|
| 207 |
|
| 208 | if (iocb->ki_flags & IOCB_DSYNC)
|
| 209 | op |= REQ_FUA;
|
| 210 | return op;
|
| 211 | }
|
| 212 |
|
| 213 | #define DIO_INLINE_BIO_VECS 4
|
| 214 |
|
| 215 | static void blkdev_bio_end_io_simple(struct bio *bio)
|
| 216 | {
|
| 217 | struct task_struct *waiter = bio->bi_private;
|
| 218 |
|
| 219 | WRITE_ONCE(bio->bi_private, NULL);
|
| 220 | blk_wake_io_task(waiter);
|
| 221 | }
|
| 222 |
|
| 223 | static ssize_t
|
| 224 | __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
|
| 225 | int nr_pages)
|
| 226 | {
|
| 227 | struct file *file = iocb->ki_filp;
|
| 228 | struct block_device *bdev = I_BDEV(bdev_file_inode(file));
|
| 229 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
|
| 230 | loff_t pos = iocb->ki_pos;
|
| 231 | bool should_dirty = false;
|
| 232 | struct bio bio;
|
| 233 | ssize_t ret;
|
| 234 | blk_qc_t qc;
|
| 235 |
|
| 236 | if ((pos | iov_iter_alignment(iter)) &
|
| 237 | (bdev_logical_block_size(bdev) - 1))
|
| 238 | return -EINVAL;
|
| 239 |
|
| 240 | if (nr_pages <= DIO_INLINE_BIO_VECS)
|
| 241 | vecs = inline_vecs;
|
| 242 | else {
|
| 243 | vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
|
| 244 | GFP_KERNEL);
|
| 245 | if (!vecs)
|
| 246 | return -ENOMEM;
|
| 247 | }
|
| 248 |
|
| 249 | bio_init(&bio, vecs, nr_pages);
|
| 250 | bio_set_dev(&bio, bdev);
|
| 251 | bio.bi_iter.bi_sector = pos >> 9;
|
| 252 | bio.bi_write_hint = iocb->ki_hint;
|
| 253 | bio.bi_private = current;
|
| 254 | bio.bi_end_io = blkdev_bio_end_io_simple;
|
| 255 | bio.bi_ioprio = iocb->ki_ioprio;
|
| 256 |
|
| 257 | ret = bio_iov_iter_get_pages(&bio, iter);
|
| 258 | if (unlikely(ret))
|
| 259 | goto out;
|
| 260 | ret = bio.bi_iter.bi_size;
|
| 261 |
|
| 262 | if (iov_iter_rw(iter) == READ) {
|
| 263 | bio.bi_opf = REQ_OP_READ;
|
| 264 | if (iter_is_iovec(iter))
|
| 265 | should_dirty = true;
|
| 266 | } else {
|
| 267 | bio.bi_opf = dio_bio_write_op(iocb);
|
| 268 | task_io_account_write(ret);
|
| 269 | }
|
| 270 | if (iocb->ki_flags & IOCB_HIPRI)
|
| 271 | bio_set_polled(&bio, iocb);
|
| 272 |
|
| 273 | qc = submit_bio(&bio);
|
| 274 | for (;;) {
|
| 275 | set_current_state(TASK_UNINTERRUPTIBLE);
|
| 276 | if (!READ_ONCE(bio.bi_private))
|
| 277 | break;
|
| 278 | if (!(iocb->ki_flags & IOCB_HIPRI) ||
|
| 279 | !blk_poll(bdev_get_queue(bdev), qc, true))
|
| 280 | blk_io_schedule();
|
| 281 | }
|
| 282 | __set_current_state(TASK_RUNNING);
|
| 283 |
|
| 284 | bio_release_pages(&bio, should_dirty);
|
| 285 | if (unlikely(bio.bi_status))
|
| 286 | ret = blk_status_to_errno(bio.bi_status);
|
| 287 |
|
| 288 | out:
|
| 289 | if (vecs != inline_vecs)
|
| 290 | kfree(vecs);
|
| 291 |
|
| 292 | bio_uninit(&bio);
|
| 293 |
|
| 294 | return ret;
|
| 295 | }
|
| 296 |
|
| 297 | struct blkdev_dio {
|
| 298 | union {
|
| 299 | struct kiocb *iocb;
|
| 300 | struct task_struct *waiter;
|
| 301 | };
|
| 302 | size_t size;
|
| 303 | atomic_t ref;
|
| 304 | bool multi_bio : 1;
|
| 305 | bool should_dirty : 1;
|
| 306 | bool is_sync : 1;
|
| 307 | struct bio bio;
|
| 308 | };
|
| 309 |
|
| 310 | static struct bio_set blkdev_dio_pool;
|
| 311 |
|
| 312 | static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
|
| 313 | {
|
| 314 | struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
|
| 315 | struct request_queue *q = bdev_get_queue(bdev);
|
| 316 |
|
| 317 | return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
|
| 318 | }
|
| 319 |
|
| 320 | static void blkdev_bio_end_io(struct bio *bio)
|
| 321 | {
|
| 322 | struct blkdev_dio *dio = bio->bi_private;
|
| 323 | bool should_dirty = dio->should_dirty;
|
| 324 |
|
| 325 | if (bio->bi_status && !dio->bio.bi_status)
|
| 326 | dio->bio.bi_status = bio->bi_status;
|
| 327 |
|
| 328 | if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
|
| 329 | if (!dio->is_sync) {
|
| 330 | struct kiocb *iocb = dio->iocb;
|
| 331 | ssize_t ret;
|
| 332 |
|
| 333 | if (likely(!dio->bio.bi_status)) {
|
| 334 | ret = dio->size;
|
| 335 | iocb->ki_pos += ret;
|
| 336 | } else {
|
| 337 | ret = blk_status_to_errno(dio->bio.bi_status);
|
| 338 | }
|
| 339 |
|
| 340 | dio->iocb->ki_complete(iocb, ret, 0);
|
| 341 | if (dio->multi_bio)
|
| 342 | bio_put(&dio->bio);
|
| 343 | } else {
|
| 344 | struct task_struct *waiter = dio->waiter;
|
| 345 |
|
| 346 | WRITE_ONCE(dio->waiter, NULL);
|
| 347 | blk_wake_io_task(waiter);
|
| 348 | }
|
| 349 | }
|
| 350 |
|
| 351 | if (should_dirty) {
|
| 352 | bio_check_pages_dirty(bio);
|
| 353 | } else {
|
| 354 | bio_release_pages(bio, false);
|
| 355 | bio_put(bio);
|
| 356 | }
|
| 357 | }
|
| 358 |
|
| 359 | static ssize_t
|
| 360 | __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
|
| 361 | {
|
| 362 | struct file *file = iocb->ki_filp;
|
| 363 | struct inode *inode = bdev_file_inode(file);
|
| 364 | struct block_device *bdev = I_BDEV(inode);
|
| 365 | struct blk_plug plug;
|
| 366 | struct blkdev_dio *dio;
|
| 367 | struct bio *bio;
|
| 368 | bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
|
| 369 | bool is_read = (iov_iter_rw(iter) == READ), is_sync;
|
| 370 | loff_t pos = iocb->ki_pos;
|
| 371 | blk_qc_t qc = BLK_QC_T_NONE;
|
| 372 | int ret = 0;
|
| 373 |
|
| 374 | if ((pos | iov_iter_alignment(iter)) &
|
| 375 | (bdev_logical_block_size(bdev) - 1))
|
| 376 | return -EINVAL;
|
| 377 |
|
| 378 | bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
|
| 379 |
|
| 380 | dio = container_of(bio, struct blkdev_dio, bio);
|
| 381 | dio->is_sync = is_sync = is_sync_kiocb(iocb);
|
| 382 | if (dio->is_sync) {
|
| 383 | dio->waiter = current;
|
| 384 | bio_get(bio);
|
| 385 | } else {
|
| 386 | dio->iocb = iocb;
|
| 387 | }
|
| 388 |
|
| 389 | dio->size = 0;
|
| 390 | dio->multi_bio = false;
|
| 391 | dio->should_dirty = is_read && iter_is_iovec(iter);
|
| 392 |
|
| 393 | |
| 394 | * Don't plug for HIPRI/polled IO, as those should go straight
|
| 395 | * to issue
|
| 396 |
|
| 397 | if (!is_poll)
|
| 398 | blk_start_plug(&plug);
|
| 399 |
|
| 400 | for (;;) {
|
| 401 | bio_set_dev(bio, bdev);
|
| 402 | bio->bi_iter.bi_sector = pos >> 9;
|
| 403 | bio->bi_write_hint = iocb->ki_hint;
|
| 404 | bio->bi_private = dio;
|
| 405 | bio->bi_end_io = blkdev_bio_end_io;
|
| 406 | bio->bi_ioprio = iocb->ki_ioprio;
|
| 407 |
|
| 408 | ret = bio_iov_iter_get_pages(bio, iter);
|
| 409 | if (unlikely(ret)) {
|
| 410 | bio->bi_status = BLK_STS_IOERR;
|
| 411 | bio_endio(bio);
|
| 412 | break;
|
| 413 | }
|
| 414 |
|
| 415 | if (is_read) {
|
| 416 | bio->bi_opf = REQ_OP_READ;
|
| 417 | if (dio->should_dirty)
|
| 418 | bio_set_pages_dirty(bio);
|
| 419 | } else {
|
| 420 | bio->bi_opf = dio_bio_write_op(iocb);
|
| 421 | task_io_account_write(bio->bi_iter.bi_size);
|
| 422 | }
|
| 423 |
|
| 424 | dio->size += bio->bi_iter.bi_size;
|
| 425 | pos += bio->bi_iter.bi_size;
|
| 426 |
|
| 427 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
|
| 428 | if (!nr_pages) {
|
| 429 | bool polled = false;
|
| 430 |
|
| 431 | if (iocb->ki_flags & IOCB_HIPRI) {
|
| 432 | bio_set_polled(bio, iocb);
|
| 433 | polled = true;
|
| 434 | }
|
| 435 |
|
| 436 | qc = submit_bio(bio);
|
| 437 |
|
| 438 | if (polled)
|
| 439 | WRITE_ONCE(iocb->ki_cookie, qc);
|
| 440 | break;
|
| 441 | }
|
| 442 |
|
| 443 | if (!dio->multi_bio) {
|
| 444 | |
| 445 | * AIO needs an extra reference to ensure the dio
|
| 446 | * structure which is embedded into the first bio
|
| 447 | * stays around.
|
| 448 |
|
| 449 | if (!is_sync)
|
| 450 | bio_get(bio);
|
| 451 | dio->multi_bio = true;
|
| 452 | atomic_set(&dio->ref, 2);
|
| 453 | } else {
|
| 454 | atomic_inc(&dio->ref);
|
| 455 | }
|
| 456 |
|
| 457 | submit_bio(bio);
|
| 458 | bio = bio_alloc(GFP_KERNEL, nr_pages);
|
| 459 | }
|
| 460 |
|
| 461 | if (!is_poll)
|
| 462 | blk_finish_plug(&plug);
|
| 463 |
|
| 464 | if (!is_sync)
|
| 465 | return -EIOCBQUEUED;
|
| 466 |
|
| 467 | for (;;) {
|
| 468 | set_current_state(TASK_UNINTERRUPTIBLE);
|
| 469 | if (!READ_ONCE(dio->waiter))
|
| 470 | break;
|
| 471 |
|
| 472 | if (!(iocb->ki_flags & IOCB_HIPRI) ||
|
| 473 | !blk_poll(bdev_get_queue(bdev), qc, true))
|
| 474 | blk_io_schedule();
|
| 475 | }
|
| 476 | __set_current_state(TASK_RUNNING);
|
| 477 |
|
| 478 | if (!ret)
|
| 479 | ret = blk_status_to_errno(dio->bio.bi_status);
|
| 480 | if (likely(!ret))
|
| 481 | ret = dio->size;
|
| 482 |
|
| 483 | bio_put(&dio->bio);
|
| 484 | return ret;
|
| 485 | }
|
| 486 |
|
| 487 | static ssize_t
|
| 488 | blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
|
| 489 | {
|
| 490 | int nr_pages;
|
| 491 |
|
| 492 | nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
|
| 493 | if (!nr_pages)
|
| 494 | return 0;
|
| 495 | if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
|
| 496 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
|
| 497 |
|
| 498 | return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
|
| 499 | }
|
| 500 |
|
| 501 | static __init int blkdev_init(void)
|
| 502 | {
|
| 503 | return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
|
| 504 | }
|
| 505 | module_init(blkdev_init);
|
| 506 |
|
| 507 | int __sync_blockdev(struct block_device *bdev, int wait)
|
| 508 | {
|
| 509 | if (!bdev)
|
| 510 | return 0;
|
| 511 | if (!wait)
|
| 512 | return filemap_flush(bdev->bd_inode->i_mapping);
|
| 513 | return filemap_write_and_wait(bdev->bd_inode->i_mapping);
|
| 514 | }
|
| 515 |
|
| 516 | |
| 517 | * Write out and wait upon all the dirty data associated with a block
|
| 518 | * device via its mapping. Does not take the superblock lock.
|
| 519 |
|
| 520 | int sync_blockdev(struct block_device *bdev)
|
| 521 | {
|
| 522 | return __sync_blockdev(bdev, 1);
|
| 523 | }
|
| 524 | EXPORT_SYMBOL(sync_blockdev);
|
| 525 |
|
| 526 | |
| 527 | * Write out and wait upon all dirty data associated with this
|
| 528 | * device. Filesystem data as well as the underlying block
|
| 529 | * device. Takes the superblock lock.
|
| 530 |
|
| 531 | int fsync_bdev(struct block_device *bdev)
|
| 532 | {
|
| 533 | struct super_block *sb = get_super(bdev);
|
| 534 | if (sb) {
|
| 535 | int res = sync_filesystem(sb);
|
| 536 | drop_super(sb);
|
| 537 | return res;
|
| 538 | }
|
| 539 | return sync_blockdev(bdev);
|
| 540 | }
|
| 541 | EXPORT_SYMBOL(fsync_bdev);
|
| 542 |
|
| 543 | |
| 544 | * freeze_bdev -- lock a filesystem and force it into a consistent state
|
| 545 | * @bdev: blockdevice to lock
|
| 546 | *
|
| 547 | * If a superblock is found on this device, we take the s_umount semaphore
|
| 548 | * on it to make sure nobody unmounts until the snapshot creation is done.
|
| 549 | * The reference counter (bd_fsfreeze_count) guarantees that only the last
|
| 550 | * unfreeze process can unfreeze the frozen filesystem actually when multiple
|
| 551 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
|
| 552 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
|
| 553 | * actually.
|
| 554 |
|
| 555 | int freeze_bdev(struct block_device *bdev)
|
| 556 | {
|
| 557 | struct super_block *sb;
|
| 558 | int error = 0;
|
| 559 |
|
| 560 | mutex_lock(&bdev->bd_fsfreeze_mutex);
|
| 561 | if (++bdev->bd_fsfreeze_count > 1)
|
| 562 | goto done;
|
| 563 |
|
| 564 | sb = get_active_super(bdev);
|
| 565 | if (!sb)
|
| 566 | goto sync;
|
| 567 | if (sb->s_op->freeze_super)
|
| 568 | error = sb->s_op->freeze_super(sb);
|
| 569 | else
|
| 570 | error = freeze_super(sb);
|
| 571 | deactivate_super(sb);
|
| 572 |
|
| 573 | if (error) {
|
| 574 | bdev->bd_fsfreeze_count--;
|
| 575 | goto done;
|
| 576 | }
|
| 577 | bdev->bd_fsfreeze_sb = sb;
|
| 578 |
|
| 579 | sync:
|
| 580 | sync_blockdev(bdev);
|
| 581 | done:
|
| 582 | mutex_unlock(&bdev->bd_fsfreeze_mutex);
|
| 583 | return error;
|
| 584 | }
|
| 585 | EXPORT_SYMBOL(freeze_bdev);
|
| 586 |
|
| 587 | |
| 588 | * thaw_bdev -- unlock filesystem
|
| 589 | * @bdev: blockdevice to unlock
|
| 590 | *
|
| 591 | * Unlocks the filesystem and marks it writeable again after freeze_bdev().
|
| 592 |
|
| 593 | int thaw_bdev(struct block_device *bdev)
|
| 594 | {
|
| 595 | struct super_block *sb;
|
| 596 | int error = -EINVAL;
|
| 597 |
|
| 598 | mutex_lock(&bdev->bd_fsfreeze_mutex);
|
| 599 | if (!bdev->bd_fsfreeze_count)
|
| 600 | goto out;
|
| 601 |
|
| 602 | error = 0;
|
| 603 | if (--bdev->bd_fsfreeze_count > 0)
|
| 604 | goto out;
|
| 605 |
|
| 606 | sb = bdev->bd_fsfreeze_sb;
|
| 607 | if (!sb)
|
| 608 | goto out;
|
| 609 |
|
| 610 | if (sb->s_op->thaw_super)
|
| 611 | error = sb->s_op->thaw_super(sb);
|
| 612 | else
|
| 613 | error = thaw_super(sb);
|
| 614 | if (error)
|
| 615 | bdev->bd_fsfreeze_count++;
|
| 616 | else
|
| 617 | bdev->bd_fsfreeze_sb = NULL;
|
| 618 | out:
|
| 619 | mutex_unlock(&bdev->bd_fsfreeze_mutex);
|
| 620 | return error;
|
| 621 | }
|
| 622 | EXPORT_SYMBOL(thaw_bdev);
|
| 623 |
|
| 624 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
|
| 625 | {
|
| 626 | return block_write_full_page(page, blkdev_get_block, wbc);
|
| 627 | }
|
| 628 |
|
| 629 | static int blkdev_readpage(struct file * file, struct page * page)
|
| 630 | {
|
| 631 | return block_read_full_page(page, blkdev_get_block);
|
| 632 | }
|
| 633 |
|
| 634 | static void blkdev_readahead(struct readahead_control *rac)
|
| 635 | {
|
| 636 | mpage_readahead(rac, blkdev_get_block);
|
| 637 | }
|
| 638 |
|
| 639 | static int blkdev_write_begin(struct file *file, struct address_space *mapping,
|
| 640 | loff_t pos, unsigned len, unsigned flags,
|
| 641 | struct page **pagep, void **fsdata)
|
| 642 | {
|
| 643 | return block_write_begin(mapping, pos, len, flags, pagep,
|
| 644 | blkdev_get_block);
|
| 645 | }
|
| 646 |
|
| 647 | static int blkdev_write_end(struct file *file, struct address_space *mapping,
|
| 648 | loff_t pos, unsigned len, unsigned copied,
|
| 649 | struct page *page, void *fsdata)
|
| 650 | {
|
| 651 | int ret;
|
| 652 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
|
| 653 |
|
| 654 | unlock_page(page);
|
| 655 | put_page(page);
|
| 656 |
|
| 657 | return ret;
|
| 658 | }
|
| 659 |
|
| 660 | |
| 661 | * private llseek:
|
| 662 | * for a block special file file_inode(file)->i_size is zero
|
| 663 | * so we compute the size by hand (just as in block_read/write above)
|
| 664 |
|
| 665 | static loff_t block_llseek(struct file *file, loff_t offset, int whence)
|
| 666 | {
|
| 667 | struct inode *bd_inode = bdev_file_inode(file);
|
| 668 | loff_t retval;
|
| 669 |
|
| 670 | inode_lock(bd_inode);
|
| 671 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
|
| 672 | inode_unlock(bd_inode);
|
| 673 | return retval;
|
| 674 | }
|
| 675 |
|
| 676 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
|
| 677 | {
|
| 678 | struct inode *bd_inode = bdev_file_inode(filp);
|
| 679 | struct block_device *bdev = I_BDEV(bd_inode);
|
| 680 | int error;
|
| 681 |
|
| 682 | error = file_write_and_wait_range(filp, start, end);
|
| 683 | if (error)
|
| 684 | return error;
|
| 685 |
|
| 686 | |
| 687 | * There is no need to serialise calls to blkdev_issue_flush with
|
| 688 | * i_mutex and doing so causes performance issues with concurrent
|
| 689 | * O_SYNC writers to a block device.
|
| 690 |
|
| 691 | error = blkdev_issue_flush(bdev, GFP_KERNEL);
|
| 692 | if (error == -EOPNOTSUPP)
|
| 693 | error = 0;
|
| 694 |
|
| 695 | return error;
|
| 696 | }
|
| 697 | EXPORT_SYMBOL(blkdev_fsync);
|
| 698 |
|
| 699 | |
| 700 | * bdev_read_page() - Start reading a page from a block device
|
| 701 | * @bdev: The device to read the page from
|
| 702 | * @sector: The offset on the device to read the page to (need not be aligned)
|
| 703 | * @page: The page to read
|
| 704 | *
|
| 705 | * On entry, the page should be locked. It will be unlocked when the page
|
| 706 | * has been read. If the block driver implements rw_page synchronously,
|
| 707 | * that will be true on exit from this function, but it need not be.
|
| 708 | *
|
| 709 | * Errors returned by this function are usually "soft", eg out of memory, or
|
| 710 | * queue full; callers should try a different route to read this page rather
|
| 711 | * than propagate an error back up the stack.
|
| 712 | *
|
| 713 | * Return: negative errno if an error occurs, 0 if submission was successful.
|
| 714 |
|
| 715 | int bdev_read_page(struct block_device *bdev, sector_t sector,
|
| 716 | struct page *page)
|
| 717 | {
|
| 718 | const struct block_device_operations *ops = bdev->bd_disk->fops;
|
| 719 | int result = -EOPNOTSUPP;
|
| 720 |
|
| 721 | if (!ops->rw_page || bdev_get_integrity(bdev))
|
| 722 | return result;
|
| 723 |
|
| 724 | result = blk_queue_enter(bdev->bd_disk->queue, 0);
|
| 725 | if (result)
|
| 726 | return result;
|
| 727 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
|
| 728 | REQ_OP_READ);
|
| 729 | blk_queue_exit(bdev->bd_disk->queue);
|
| 730 | return result;
|
| 731 | }
|
| 732 |
|
| 733 | |
| 734 | * bdev_write_page() - Start writing a page to a block device
|
| 735 | * @bdev: The device to write the page to
|
| 736 | * @sector: The offset on the device to write the page to (need not be aligned)
|
| 737 | * @page: The page to write
|
| 738 | * @wbc: The writeback_control for the write
|
| 739 | *
|
| 740 | * On entry, the page should be locked and not currently under writeback.
|
| 741 | * On exit, if the write started successfully, the page will be unlocked and
|
| 742 | * under writeback. If the write failed already (eg the driver failed to
|
| 743 | * queue the page to the device), the page will still be locked. If the
|
| 744 | * caller is a ->writepage implementation, it will need to unlock the page.
|
| 745 | *
|
| 746 | * Errors returned by this function are usually "soft", eg out of memory, or
|
| 747 | * queue full; callers should try a different route to write this page rather
|
| 748 | * than propagate an error back up the stack.
|
| 749 | *
|
| 750 | * Return: negative errno if an error occurs, 0 if submission was successful.
|
| 751 |
|
| 752 | int bdev_write_page(struct block_device *bdev, sector_t sector,
|
| 753 | struct page *page, struct writeback_control *wbc)
|
| 754 | {
|
| 755 | int result;
|
| 756 | const struct block_device_operations *ops = bdev->bd_disk->fops;
|
| 757 |
|
| 758 | if (!ops->rw_page || bdev_get_integrity(bdev))
|
| 759 | return -EOPNOTSUPP;
|
| 760 | result = blk_queue_enter(bdev->bd_disk->queue, 0);
|
| 761 | if (result)
|
| 762 | return result;
|
| 763 |
|
| 764 | set_page_writeback(page);
|
| 765 | result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
|
| 766 | REQ_OP_WRITE);
|
| 767 | if (result) {
|
| 768 | end_page_writeback(page);
|
| 769 | } else {
|
| 770 | clean_page_buffers(page);
|
| 771 | unlock_page(page);
|
| 772 | }
|
| 773 | blk_queue_exit(bdev->bd_disk->queue);
|
| 774 | return result;
|
| 775 | }
|
| 776 |
|
| 777 | |
| 778 | * pseudo-fs
|
| 779 |
|
| 780 |
|
| 781 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
|
| 782 | static struct kmem_cache * bdev_cachep __read_mostly;
|
| 783 |
|
| 784 | static struct inode *bdev_alloc_inode(struct super_block *sb)
|
| 785 | {
|
| 786 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
|
| 787 |
|
| 788 | if (!ei)
|
| 789 | return NULL;
|
| 790 | memset(&ei->bdev, 0, sizeof(ei->bdev));
|
| 791 | ei->bdev.bd_bdi = &noop_backing_dev_info;
|
| 792 | return &ei->vfs_inode;
|
| 793 | }
|
| 794 |
|
| 795 | static void bdev_free_inode(struct inode *inode)
|
| 796 | {
|
| 797 | struct block_device *bdev = I_BDEV(inode);
|
| 798 |
|
| 799 | free_percpu(bdev->bd_stats);
|
| 800 | kfree(bdev->bd_meta_info);
|
| 801 |
|
| 802 | kmem_cache_free(bdev_cachep, BDEV_I(inode));
|
| 803 | }
|
| 804 |
|
| 805 | static void init_once(void *data)
|
| 806 | {
|
| 807 | struct bdev_inode *ei = data;
|
| 808 |
|
| 809 | inode_init_once(&ei->vfs_inode);
|
| 810 | }
|
| 811 |
|
| 812 | static void bdev_evict_inode(struct inode *inode)
|
| 813 | {
|
| 814 | struct block_device *bdev = &BDEV_I(inode)->bdev;
|
| 815 | truncate_inode_pages_final(&inode->i_data);
|
| 816 | invalidate_inode_buffers(inode);
|
| 817 | clear_inode(inode);
|
| 818 |
|
| 819 | inode_detach_wb(inode);
|
| 820 | if (bdev->bd_bdi != &noop_backing_dev_info) {
|
| 821 | bdi_put(bdev->bd_bdi);
|
| 822 | bdev->bd_bdi = &noop_backing_dev_info;
|
| 823 | }
|
| 824 | }
|
| 825 |
|
| 826 | static const struct super_operations bdev_sops = {
|
| 827 | .statfs = simple_statfs,
|
| 828 | .alloc_inode = bdev_alloc_inode,
|
| 829 | .free_inode = bdev_free_inode,
|
| 830 | .drop_inode = generic_delete_inode,
|
| 831 | .evict_inode = bdev_evict_inode,
|
| 832 | };
|
| 833 |
|
| 834 | static int bd_init_fs_context(struct fs_context *fc)
|
| 835 | {
|
| 836 | struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
|
| 837 | if (!ctx)
|
| 838 | return -ENOMEM;
|
| 839 | fc->s_iflags |= SB_I_CGROUPWB;
|
| 840 | ctx->ops = &bdev_sops;
|
| 841 | return 0;
|
| 842 | }
|
| 843 |
|
| 844 | static struct file_system_type bd_type = {
|
| 845 | .name = "bdev",
|
| 846 | .init_fs_context = bd_init_fs_context,
|
| 847 | .kill_sb = kill_anon_super,
|
| 848 | };
|
| 849 |
|
| 850 | struct super_block *blockdev_superblock __read_mostly;
|
| 851 | EXPORT_SYMBOL_GPL(blockdev_superblock);
|
| 852 |
|
| 853 | void __init bdev_cache_init(void)
|
| 854 | {
|
| 855 | int err;
|
| 856 | static struct vfsmount *bd_mnt;
|
| 857 |
|
| 858 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
|
| 859 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
|
| 860 | SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
|
| 861 | init_once);
|
| 862 | err = register_filesystem(&bd_type);
|
| 863 | if (err)
|
| 864 | panic("Cannot register bdev pseudo-fs");
|
| 865 | bd_mnt = kern_mount(&bd_type);
|
| 866 | if (IS_ERR(bd_mnt))
|
| 867 | panic("Cannot create bdev pseudo-fs");
|
| 868 | blockdev_superblock = bd_mnt->mnt_sb;
|
| 869 | }
|
| 870 |
|
| 871 | struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
|
| 872 | {
|
| 873 | struct block_device *bdev;
|
| 874 | struct inode *inode;
|
| 875 |
|
| 876 | inode = new_inode(blockdev_superblock);
|
| 877 | if (!inode)
|
| 878 | return NULL;
|
| 879 | inode->i_mode = S_IFBLK;
|
| 880 | inode->i_rdev = 0;
|
| 881 | inode->i_data.a_ops = &def_blk_aops;
|
| 882 | mapping_set_gfp_mask(&inode->i_data, GFP_USER);
|
| 883 |
|
| 884 | bdev = I_BDEV(inode);
|
| 885 | mutex_init(&bdev->bd_mutex);
|
| 886 | mutex_init(&bdev->bd_fsfreeze_mutex);
|
| 887 | spin_lock_init(&bdev->bd_size_lock);
|
| 888 | bdev->bd_disk = disk;
|
| 889 | bdev->bd_partno = partno;
|
| 890 | bdev->bd_inode = inode;
|
| 891 | #ifdef CONFIG_SYSFS
|
| 892 | INIT_LIST_HEAD(&bdev->bd_holder_disks);
|
| 893 | #endif
|
| 894 | bdev->bd_stats = alloc_percpu(struct disk_stats);
|
| 895 | if (!bdev->bd_stats) {
|
| 896 | iput(inode);
|
| 897 | return NULL;
|
| 898 | }
|
| 899 | return bdev;
|
| 900 | }
|
| 901 |
|
| 902 | void bdev_add(struct block_device *bdev, dev_t dev)
|
| 903 | {
|
| 904 | bdev->bd_dev = dev;
|
| 905 | bdev->bd_inode->i_rdev = dev;
|
| 906 | bdev->bd_inode->i_ino = dev;
|
| 907 | insert_inode_hash(bdev->bd_inode);
|
| 908 | }
|
| 909 |
|
| 910 | static struct block_device *bdget(dev_t dev)
|
| 911 | {
|
| 912 | struct inode *inode;
|
| 913 |
|
| 914 | inode = ilookup(blockdev_superblock, dev);
|
| 915 | if (!inode)
|
| 916 | return NULL;
|
| 917 | return &BDEV_I(inode)->bdev;
|
| 918 | }
|
| 919 |
|
| 920 | |
| 921 | * bdgrab -- Grab a reference to an already referenced block device
|
| 922 | * @bdev: Block device to grab a reference to.
|
| 923 | *
|
| 924 | * Returns the block_device with an additional reference when successful,
|
| 925 | * or NULL if the inode is already beeing freed.
|
| 926 |
|
| 927 | struct block_device *bdgrab(struct block_device *bdev)
|
| 928 | {
|
| 929 | if (!igrab(bdev->bd_inode))
|
| 930 | return NULL;
|
| 931 | return bdev;
|
| 932 | }
|
| 933 | EXPORT_SYMBOL(bdgrab);
|
| 934 |
|
| 935 | long nr_blockdev_pages(void)
|
| 936 | {
|
| 937 | struct inode *inode;
|
| 938 | long ret = 0;
|
| 939 |
|
| 940 | spin_lock(&blockdev_superblock->s_inode_list_lock);
|
| 941 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
|
| 942 | ret += inode->i_mapping->nrpages;
|
| 943 | spin_unlock(&blockdev_superblock->s_inode_list_lock);
|
| 944 |
|
| 945 | return ret;
|
| 946 | }
|
| 947 |
|
| 948 | void bdput(struct block_device *bdev)
|
| 949 | {
|
| 950 | iput(bdev->bd_inode);
|
| 951 | }
|
| 952 | EXPORT_SYMBOL(bdput);
|
| 953 |
|
| 954 | |
| 955 | * bd_may_claim - test whether a block device can be claimed
|
| 956 | * @bdev: block device of interest
|
| 957 | * @whole: whole block device containing @bdev, may equal @bdev
|
| 958 | * @holder: holder trying to claim @bdev
|
| 959 | *
|
| 960 | * Test whether @bdev can be claimed by @holder.
|
| 961 | *
|
| 962 | * CONTEXT:
|
| 963 | * spin_lock(&bdev_lock).
|
| 964 | *
|
| 965 | * RETURNS:
|
| 966 | * %true if @bdev can be claimed, %false otherwise.
|
| 967 |
|
| 968 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
|
| 969 | void *holder)
|
| 970 | {
|
| 971 | if (bdev->bd_holder == holder)
|
| 972 | return true;
|
| 973 | else if (bdev->bd_holder != NULL)
|
| 974 | return false;
|
| 975 | else if (whole == bdev)
|
| 976 | return true;
|
| 977 |
|
| 978 | else if (whole->bd_holder == bd_may_claim)
|
| 979 | return true;
|
| 980 | else if (whole->bd_holder != NULL)
|
| 981 | return false;
|
| 982 | else
|
| 983 | return true;
|
| 984 | }
|
| 985 |
|
| 986 | |
| 987 | * bd_prepare_to_claim - claim a block device
|
| 988 | * @bdev: block device of interest
|
| 989 | * @holder: holder trying to claim @bdev
|
| 990 | *
|
| 991 | * Claim @bdev. This function fails if @bdev is already claimed by another
|
| 992 | * holder and waits if another claiming is in progress. return, the caller
|
| 993 | * has ownership of bd_claiming and bd_holder[s].
|
| 994 | *
|
| 995 | * RETURNS:
|
| 996 | * 0 if @bdev can be claimed, -EBUSY otherwise.
|
| 997 |
|
| 998 | int bd_prepare_to_claim(struct block_device *bdev, void *holder)
|
| 999 | {
|
| 1000 | struct block_device *whole = bdev_whole(bdev);
|
| 1001 |
|
| 1002 | if (WARN_ON_ONCE(!holder))
|
| 1003 | return -EINVAL;
|
| 1004 | retry:
|
| 1005 | spin_lock(&bdev_lock);
|
| 1006 |
|
| 1007 | if (!bd_may_claim(bdev, whole, holder)) {
|
| 1008 | spin_unlock(&bdev_lock);
|
| 1009 | return -EBUSY;
|
| 1010 | }
|
| 1011 |
|
| 1012 |
|
| 1013 | if (whole->bd_claiming) {
|
| 1014 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
|
| 1015 | DEFINE_WAIT(wait);
|
| 1016 |
|
| 1017 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
|
| 1018 | spin_unlock(&bdev_lock);
|
| 1019 | schedule();
|
| 1020 | finish_wait(wq, &wait);
|
| 1021 | goto retry;
|
| 1022 | }
|
| 1023 |
|
| 1024 |
|
| 1025 | whole->bd_claiming = holder;
|
| 1026 | spin_unlock(&bdev_lock);
|
| 1027 | return 0;
|
| 1028 | }
|
| 1029 | EXPORT_SYMBOL_GPL(bd_prepare_to_claim);
|
| 1030 |
|
| 1031 | static void bd_clear_claiming(struct block_device *whole, void *holder)
|
| 1032 | {
|
| 1033 | lockdep_assert_held(&bdev_lock);
|
| 1034 |
|
| 1035 | BUG_ON(whole->bd_claiming != holder);
|
| 1036 | whole->bd_claiming = NULL;
|
| 1037 | wake_up_bit(&whole->bd_claiming, 0);
|
| 1038 | }
|
| 1039 |
|
| 1040 | |
| 1041 | * bd_finish_claiming - finish claiming of a block device
|
| 1042 | * @bdev: block device of interest
|
| 1043 | * @holder: holder that has claimed @bdev
|
| 1044 | *
|
| 1045 | * Finish exclusive open of a block device. Mark the device as exlusively
|
| 1046 | * open by the holder and wake up all waiters for exclusive open to finish.
|
| 1047 |
|
| 1048 | static void bd_finish_claiming(struct block_device *bdev, void *holder)
|
| 1049 | {
|
| 1050 | struct block_device *whole = bdev_whole(bdev);
|
| 1051 |
|
| 1052 | spin_lock(&bdev_lock);
|
| 1053 | BUG_ON(!bd_may_claim(bdev, whole, holder));
|
| 1054 | |
| 1055 | * Note that for a whole device bd_holders will be incremented twice,
|
| 1056 | * and bd_holder will be set to bd_may_claim before being set to holder
|
| 1057 |
|
| 1058 | whole->bd_holders++;
|
| 1059 | whole->bd_holder = bd_may_claim;
|
| 1060 | bdev->bd_holders++;
|
| 1061 | bdev->bd_holder = holder;
|
| 1062 | bd_clear_claiming(whole, holder);
|
| 1063 | spin_unlock(&bdev_lock);
|
| 1064 | }
|
| 1065 |
|
| 1066 | |
| 1067 | * bd_abort_claiming - abort claiming of a block device
|
| 1068 | * @bdev: block device of interest
|
| 1069 | * @holder: holder that has claimed @bdev
|
| 1070 | *
|
| 1071 | * Abort claiming of a block device when the exclusive open failed. This can be
|
| 1072 | * also used when exclusive open is not actually desired and we just needed
|
| 1073 | * to block other exclusive openers for a while.
|
| 1074 |
|
| 1075 | void bd_abort_claiming(struct block_device *bdev, void *holder)
|
| 1076 | {
|
| 1077 | spin_lock(&bdev_lock);
|
| 1078 | bd_clear_claiming(bdev_whole(bdev), holder);
|
| 1079 | spin_unlock(&bdev_lock);
|
| 1080 | }
|
| 1081 | EXPORT_SYMBOL(bd_abort_claiming);
|
| 1082 |
|
| 1083 | #ifdef CONFIG_SYSFS
|
| 1084 | struct bd_holder_disk {
|
| 1085 | struct list_head list;
|
| 1086 | struct gendisk *disk;
|
| 1087 | int refcnt;
|
| 1088 | };
|
| 1089 |
|
| 1090 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
|
| 1091 | struct gendisk *disk)
|
| 1092 | {
|
| 1093 | struct bd_holder_disk *holder;
|
| 1094 |
|
| 1095 | list_for_each_entry(holder, &bdev->bd_holder_disks, list)
|
| 1096 | if (holder->disk == disk)
|
| 1097 | return holder;
|
| 1098 | return NULL;
|
| 1099 | }
|
| 1100 |
|
| 1101 | static int add_symlink(struct kobject *from, struct kobject *to)
|
| 1102 | {
|
| 1103 | return sysfs_create_link(from, to, kobject_name(to));
|
| 1104 | }
|
| 1105 |
|
| 1106 | static void del_symlink(struct kobject *from, struct kobject *to)
|
| 1107 | {
|
| 1108 | sysfs_remove_link(from, kobject_name(to));
|
| 1109 | }
|
| 1110 |
|
| 1111 | |
| 1112 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev
|
| 1113 | * @bdev: the claimed slave bdev
|
| 1114 | * @disk: the holding disk
|
| 1115 | *
|
| 1116 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
|
| 1117 | *
|
| 1118 | * This functions creates the following sysfs symlinks.
|
| 1119 | *
|
| 1120 | * - from "slaves" directory of the holder @disk to the claimed @bdev
|
| 1121 | * - from "holders" directory of the @bdev to the holder @disk
|
| 1122 | *
|
| 1123 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
|
| 1124 | * passed to bd_link_disk_holder(), then:
|
| 1125 | *
|
| 1126 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda
|
| 1127 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
|
| 1128 | *
|
| 1129 | * The caller must have claimed @bdev before calling this function and
|
| 1130 | * ensure that both @bdev and @disk are valid during the creation and
|
| 1131 | * lifetime of these symlinks.
|
| 1132 | *
|
| 1133 | * CONTEXT:
|
| 1134 | * Might sleep.
|
| 1135 | *
|
| 1136 | * RETURNS:
|
| 1137 | * 0 on success, -errno on failure.
|
| 1138 |
|
| 1139 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
|
| 1140 | {
|
| 1141 | struct bd_holder_disk *holder;
|
| 1142 | int ret = 0;
|
| 1143 |
|
| 1144 | mutex_lock(&bdev->bd_mutex);
|
| 1145 |
|
| 1146 | WARN_ON_ONCE(!bdev->bd_holder);
|
| 1147 |
|
| 1148 |
|
| 1149 | if (WARN_ON(!disk->slave_dir || !bdev->bd_holder_dir))
|
| 1150 | goto out_unlock;
|
| 1151 |
|
| 1152 | holder = bd_find_holder_disk(bdev, disk);
|
| 1153 | if (holder) {
|
| 1154 | holder->refcnt++;
|
| 1155 | goto out_unlock;
|
| 1156 | }
|
| 1157 |
|
| 1158 | holder = kzalloc(sizeof(*holder), GFP_KERNEL);
|
| 1159 | if (!holder) {
|
| 1160 | ret = -ENOMEM;
|
| 1161 | goto out_unlock;
|
| 1162 | }
|
| 1163 |
|
| 1164 | INIT_LIST_HEAD(&holder->list);
|
| 1165 | holder->disk = disk;
|
| 1166 | holder->refcnt = 1;
|
| 1167 |
|
| 1168 | ret = add_symlink(disk->slave_dir, bdev_kobj(bdev));
|
| 1169 | if (ret)
|
| 1170 | goto out_free;
|
| 1171 |
|
| 1172 | ret = add_symlink(bdev->bd_holder_dir, &disk_to_dev(disk)->kobj);
|
| 1173 | if (ret)
|
| 1174 | goto out_del;
|
| 1175 | |
| 1176 | * bdev could be deleted beneath us which would implicitly destroy
|
| 1177 | * the holder directory. Hold on to it.
|
| 1178 |
|
| 1179 | kobject_get(bdev->bd_holder_dir);
|
| 1180 |
|
| 1181 | list_add(&holder->list, &bdev->bd_holder_disks);
|
| 1182 | goto out_unlock;
|
| 1183 |
|
| 1184 | out_del:
|
| 1185 | del_symlink(disk->slave_dir, bdev_kobj(bdev));
|
| 1186 | out_free:
|
| 1187 | kfree(holder);
|
| 1188 | out_unlock:
|
| 1189 | mutex_unlock(&bdev->bd_mutex);
|
| 1190 | return ret;
|
| 1191 | }
|
| 1192 | EXPORT_SYMBOL_GPL(bd_link_disk_holder);
|
| 1193 |
|
| 1194 | |
| 1195 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
|
| 1196 | * @bdev: the calimed slave bdev
|
| 1197 | * @disk: the holding disk
|
| 1198 | *
|
| 1199 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
|
| 1200 | *
|
| 1201 | * CONTEXT:
|
| 1202 | * Might sleep.
|
| 1203 |
|
| 1204 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
|
| 1205 | {
|
| 1206 | struct bd_holder_disk *holder;
|
| 1207 |
|
| 1208 | mutex_lock(&bdev->bd_mutex);
|
| 1209 |
|
| 1210 | holder = bd_find_holder_disk(bdev, disk);
|
| 1211 |
|
| 1212 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
|
| 1213 | del_symlink(disk->slave_dir, bdev_kobj(bdev));
|
| 1214 | del_symlink(bdev->bd_holder_dir, &disk_to_dev(disk)->kobj);
|
| 1215 | kobject_put(bdev->bd_holder_dir);
|
| 1216 | list_del_init(&holder->list);
|
| 1217 | kfree(holder);
|
| 1218 | }
|
| 1219 |
|
| 1220 | mutex_unlock(&bdev->bd_mutex);
|
| 1221 | }
|
| 1222 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
|
| 1223 | #endif
|
| 1224 |
|
| 1225 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
|
| 1226 |
|
| 1227 | int bdev_disk_changed(struct block_device *bdev, bool invalidate)
|
| 1228 | {
|
| 1229 | struct gendisk *disk = bdev->bd_disk;
|
| 1230 | int ret;
|
| 1231 |
|
| 1232 | lockdep_assert_held(&bdev->bd_mutex);
|
| 1233 |
|
| 1234 | clear_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
|
| 1235 |
|
| 1236 | rescan:
|
| 1237 | ret = blk_drop_partitions(bdev);
|
| 1238 | if (ret)
|
| 1239 | return ret;
|
| 1240 |
|
| 1241 | |
| 1242 | * Historically we only set the capacity to zero for devices that
|
| 1243 | * support partitions (independ of actually having partitions created).
|
| 1244 | * Doing that is rather inconsistent, but changing it broke legacy
|
| 1245 | * udisks polling for legacy ide-cdrom devices. Use the crude check
|
| 1246 | * below to get the sane behavior for most device while not breaking
|
| 1247 | * userspace for this particular setup.
|
| 1248 |
|
| 1249 | if (invalidate) {
|
| 1250 | if (disk_part_scan_enabled(disk) ||
|
| 1251 | !(disk->flags & GENHD_FL_REMOVABLE))
|
| 1252 | set_capacity(disk, 0);
|
| 1253 | } else {
|
| 1254 | if (disk->fops->revalidate_disk)
|
| 1255 | disk->fops->revalidate_disk(disk);
|
| 1256 | }
|
| 1257 |
|
| 1258 | if (get_capacity(disk)) {
|
| 1259 | ret = blk_add_partitions(disk, bdev);
|
| 1260 | if (ret == -EAGAIN)
|
| 1261 | goto rescan;
|
| 1262 | } else if (invalidate) {
|
| 1263 | |
| 1264 | * Tell userspace that the media / partition table may have
|
| 1265 | * changed.
|
| 1266 |
|
| 1267 | kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
|
| 1268 | }
|
| 1269 |
|
| 1270 | return ret;
|
| 1271 | }
|
| 1272 | |
| 1273 | * Only exported for for loop and dasd for historic reasons. Don't use in new
|
| 1274 | * code!
|
| 1275 |
|
| 1276 | EXPORT_SYMBOL_GPL(bdev_disk_changed);
|
| 1277 |
|
| 1278 | |
| 1279 | * bd_mutex locking:
|
| 1280 | *
|
| 1281 | * mutex_lock(part->bd_mutex)
|
| 1282 | * mutex_lock_nested(whole->bd_mutex, 1)
|
| 1283 |
|
| 1284 | static int __blkdev_get(struct block_device *bdev, fmode_t mode)
|
| 1285 | {
|
| 1286 | struct gendisk *disk = bdev->bd_disk;
|
| 1287 | int ret = 0;
|
| 1288 |
|
| 1289 | if (!bdev->bd_openers) {
|
| 1290 | if (!bdev_is_partition(bdev)) {
|
| 1291 | ret = 0;
|
| 1292 | if (disk->fops->open)
|
| 1293 | ret = disk->fops->open(bdev, mode);
|
| 1294 |
|
| 1295 | if (!ret)
|
| 1296 | set_init_blocksize(bdev);
|
| 1297 |
|
| 1298 | |
| 1299 | * If the device is invalidated, rescan partition
|
| 1300 | * if open succeeded or failed with -ENOMEDIUM.
|
| 1301 | * The latter is necessary to prevent ghost
|
| 1302 | * partitions on a removed medium.
|
| 1303 |
|
| 1304 | if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
|
| 1305 | (!ret || ret == -ENOMEDIUM))
|
| 1306 | bdev_disk_changed(bdev, ret == -ENOMEDIUM);
|
| 1307 |
|
| 1308 | if (ret)
|
| 1309 | return ret;
|
| 1310 | } else {
|
| 1311 | struct block_device *whole = bdgrab(disk->part0);
|
| 1312 |
|
| 1313 | mutex_lock_nested(&whole->bd_mutex, 1);
|
| 1314 | ret = __blkdev_get(whole, mode);
|
| 1315 | if (ret) {
|
| 1316 | mutex_unlock(&whole->bd_mutex);
|
| 1317 | bdput(whole);
|
| 1318 | return ret;
|
| 1319 | }
|
| 1320 | whole->bd_part_count++;
|
| 1321 | mutex_unlock(&whole->bd_mutex);
|
| 1322 |
|
| 1323 | if (!(disk->flags & GENHD_FL_UP) ||
|
| 1324 | !bdev_nr_sectors(bdev)) {
|
| 1325 | __blkdev_put(whole, mode, 1);
|
| 1326 | bdput(whole);
|
| 1327 | return -ENXIO;
|
| 1328 | }
|
| 1329 | set_init_blocksize(bdev);
|
| 1330 | }
|
| 1331 |
|
| 1332 | if (bdev->bd_bdi == &noop_backing_dev_info)
|
| 1333 | bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
|
| 1334 | } else {
|
| 1335 | if (!bdev_is_partition(bdev)) {
|
| 1336 | if (bdev->bd_disk->fops->open)
|
| 1337 | ret = bdev->bd_disk->fops->open(bdev, mode);
|
| 1338 |
|
| 1339 | if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
|
| 1340 | (!ret || ret == -ENOMEDIUM))
|
| 1341 | bdev_disk_changed(bdev, ret == -ENOMEDIUM);
|
| 1342 | if (ret)
|
| 1343 | return ret;
|
| 1344 | }
|
| 1345 | }
|
| 1346 | bdev->bd_openers++;
|
| 1347 | return 0;
|
| 1348 | }
|
| 1349 |
|
| 1350 | struct block_device *blkdev_get_no_open(dev_t dev)
|
| 1351 | {
|
| 1352 | struct block_device *bdev;
|
| 1353 | struct gendisk *disk;
|
| 1354 |
|
| 1355 | down_read(&bdev_lookup_sem);
|
| 1356 | bdev = bdget(dev);
|
| 1357 | if (!bdev) {
|
| 1358 | up_read(&bdev_lookup_sem);
|
| 1359 | blk_request_module(dev);
|
| 1360 | down_read(&bdev_lookup_sem);
|
| 1361 |
|
| 1362 | bdev = bdget(dev);
|
| 1363 | if (!bdev)
|
| 1364 | goto unlock;
|
| 1365 | }
|
| 1366 |
|
| 1367 | disk = bdev->bd_disk;
|
| 1368 | if (!kobject_get_unless_zero(&disk_to_dev(disk)->kobj))
|
| 1369 | goto bdput;
|
| 1370 | if ((disk->flags & (GENHD_FL_UP | GENHD_FL_HIDDEN)) != GENHD_FL_UP)
|
| 1371 | goto put_disk;
|
| 1372 | if (!try_module_get(bdev->bd_disk->fops->owner))
|
| 1373 | goto put_disk;
|
| 1374 | up_read(&bdev_lookup_sem);
|
| 1375 | return bdev;
|
| 1376 | put_disk:
|
| 1377 | put_disk(disk);
|
| 1378 | bdput:
|
| 1379 | bdput(bdev);
|
| 1380 | unlock:
|
| 1381 | up_read(&bdev_lookup_sem);
|
| 1382 | return NULL;
|
| 1383 | }
|
| 1384 |
|
| 1385 | void blkdev_put_no_open(struct block_device *bdev)
|
| 1386 | {
|
| 1387 | module_put(bdev->bd_disk->fops->owner);
|
| 1388 | put_disk(bdev->bd_disk);
|
| 1389 | bdput(bdev);
|
| 1390 | }
|
| 1391 |
|
| 1392 | |
| 1393 | * blkdev_get_by_dev - open a block device by device number
|
| 1394 | * @dev: device number of block device to open
|
| 1395 | * @mode: FMODE_* mask
|
| 1396 | * @holder: exclusive holder identifier
|
| 1397 | *
|
| 1398 | * Open the block device described by device number @dev. If @mode includes
|
| 1399 | * %FMODE_EXCL, the block device is opened with exclusive access. Specifying
|
| 1400 | * %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may nest for
|
| 1401 | * the same @holder.
|
| 1402 | *
|
| 1403 | * Use this interface ONLY if you really do not have anything better - i.e. when
|
| 1404 | * you are behind a truly sucky interface and all you are given is a device
|
| 1405 | * number. Everything else should use blkdev_get_by_path().
|
| 1406 | *
|
| 1407 | * CONTEXT:
|
| 1408 | * Might sleep.
|
| 1409 | *
|
| 1410 | * RETURNS:
|
| 1411 | * Reference to the block_device on success, ERR_PTR(-errno) on failure.
|
| 1412 |
|
| 1413 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
|
| 1414 | {
|
| 1415 | bool unblock_events = true;
|
| 1416 | struct block_device *bdev;
|
| 1417 | struct gendisk *disk;
|
| 1418 | int ret;
|
| 1419 |
|
| 1420 | ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
|
| 1421 | MAJOR(dev), MINOR(dev),
|
| 1422 | ((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) |
|
| 1423 | ((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0));
|
| 1424 | if (ret)
|
| 1425 | return ERR_PTR(ret);
|
| 1426 |
|
| 1427 | |
| 1428 | * If we lost a race with 'disk' being deleted, try again. See md.c.
|
| 1429 |
|
| 1430 | retry:
|
| 1431 | bdev = blkdev_get_no_open(dev);
|
| 1432 | if (!bdev)
|
| 1433 | return ERR_PTR(-ENXIO);
|
| 1434 | disk = bdev->bd_disk;
|
| 1435 |
|
| 1436 | if (mode & FMODE_EXCL) {
|
| 1437 | ret = bd_prepare_to_claim(bdev, holder);
|
| 1438 | if (ret)
|
| 1439 | goto put_blkdev;
|
| 1440 | }
|
| 1441 |
|
| 1442 | disk_block_events(disk);
|
| 1443 |
|
| 1444 | mutex_lock(&bdev->bd_mutex);
|
| 1445 | ret =__blkdev_get(bdev, mode);
|
| 1446 | if (ret)
|
| 1447 | goto abort_claiming;
|
| 1448 | if (mode & FMODE_EXCL) {
|
| 1449 | bd_finish_claiming(bdev, holder);
|
| 1450 |
|
| 1451 | |
| 1452 | * Block event polling for write claims if requested. Any write
|
| 1453 | * holder makes the write_holder state stick until all are
|
| 1454 | * released. This is good enough and tracking individual
|
| 1455 | * writeable reference is too fragile given the way @mode is
|
| 1456 | * used in blkdev_get/put().
|
| 1457 |
|
| 1458 | if ((mode & FMODE_WRITE) && !bdev->bd_write_holder &&
|
| 1459 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
|
| 1460 | bdev->bd_write_holder = true;
|
| 1461 | unblock_events = false;
|
| 1462 | }
|
| 1463 | }
|
| 1464 | mutex_unlock(&bdev->bd_mutex);
|
| 1465 |
|
| 1466 | if (unblock_events)
|
| 1467 | disk_unblock_events(disk);
|
| 1468 | return bdev;
|
| 1469 |
|
| 1470 | abort_claiming:
|
| 1471 | if (mode & FMODE_EXCL)
|
| 1472 | bd_abort_claiming(bdev, holder);
|
| 1473 | mutex_unlock(&bdev->bd_mutex);
|
| 1474 | disk_unblock_events(disk);
|
| 1475 | put_blkdev:
|
| 1476 | blkdev_put_no_open(bdev);
|
| 1477 | if (ret == -ERESTARTSYS)
|
| 1478 | goto retry;
|
| 1479 | return ERR_PTR(ret);
|
| 1480 | }
|
| 1481 | EXPORT_SYMBOL(blkdev_get_by_dev);
|
| 1482 |
|
| 1483 | |
| 1484 | * blkdev_get_by_path - open a block device by name
|
| 1485 | * @path: path to the block device to open
|
| 1486 | * @mode: FMODE_* mask
|
| 1487 | * @holder: exclusive holder identifier
|
| 1488 | *
|
| 1489 | * Open the block device described by the device file at @path. If @mode
|
| 1490 | * includes %FMODE_EXCL, the block device is opened with exclusive access.
|
| 1491 | * Specifying %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may
|
| 1492 | * nest for the same @holder.
|
| 1493 | *
|
| 1494 | * CONTEXT:
|
| 1495 | * Might sleep.
|
| 1496 | *
|
| 1497 | * RETURNS:
|
| 1498 | * Reference to the block_device on success, ERR_PTR(-errno) on failure.
|
| 1499 |
|
| 1500 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
|
| 1501 | void *holder)
|
| 1502 | {
|
| 1503 | struct block_device *bdev;
|
| 1504 | dev_t dev;
|
| 1505 | int error;
|
| 1506 |
|
| 1507 | error = lookup_bdev(path, &dev);
|
| 1508 | if (error)
|
| 1509 | return ERR_PTR(error);
|
| 1510 |
|
| 1511 | bdev = blkdev_get_by_dev(dev, mode, holder);
|
| 1512 | if (!IS_ERR(bdev) && (mode & FMODE_WRITE) && bdev_read_only(bdev)) {
|
| 1513 | blkdev_put(bdev, mode);
|
| 1514 | return ERR_PTR(-EACCES);
|
| 1515 | }
|
| 1516 |
|
| 1517 | return bdev;
|
| 1518 | }
|
| 1519 | EXPORT_SYMBOL(blkdev_get_by_path);
|
| 1520 |
|
| 1521 | static int blkdev_open(struct inode * inode, struct file * filp)
|
| 1522 | {
|
| 1523 | struct block_device *bdev;
|
| 1524 |
|
| 1525 | |
| 1526 | * Preserve backwards compatibility and allow large file access
|
| 1527 | * even if userspace doesn't ask for it explicitly. Some mkfs
|
| 1528 | * binary needs it. We might want to drop this workaround
|
| 1529 | * during an unstable branch.
|
| 1530 |
|
| 1531 | filp->f_flags |= O_LARGEFILE;
|
| 1532 |
|
| 1533 | filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
|
| 1534 |
|
| 1535 | if (filp->f_flags & O_NDELAY)
|
| 1536 | filp->f_mode |= FMODE_NDELAY;
|
| 1537 | if (filp->f_flags & O_EXCL)
|
| 1538 | filp->f_mode |= FMODE_EXCL;
|
| 1539 | if ((filp->f_flags & O_ACCMODE) == 3)
|
| 1540 | filp->f_mode |= FMODE_WRITE_IOCTL;
|
| 1541 |
|
| 1542 | bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
|
| 1543 | if (IS_ERR(bdev))
|
| 1544 | return PTR_ERR(bdev);
|
| 1545 | filp->f_mapping = bdev->bd_inode->i_mapping;
|
| 1546 | filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
|
| 1547 | return 0;
|
| 1548 | }
|
| 1549 |
|
| 1550 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
|
| 1551 | {
|
| 1552 | struct gendisk *disk = bdev->bd_disk;
|
| 1553 | struct block_device *victim = NULL;
|
| 1554 |
|
| 1555 | |
| 1556 | * Sync early if it looks like we're the last one. If someone else
|
| 1557 | * opens the block device between now and the decrement of bd_openers
|
| 1558 | * then we did a sync that we didn't need to, but that's not the end
|
| 1559 | * of the world and we want to avoid long (could be several minute)
|
| 1560 | * syncs while holding the mutex.
|
| 1561 |
|
| 1562 | if (bdev->bd_openers == 1)
|
| 1563 | sync_blockdev(bdev);
|
| 1564 |
|
| 1565 | mutex_lock_nested(&bdev->bd_mutex, for_part);
|
| 1566 | if (for_part)
|
| 1567 | bdev->bd_part_count--;
|
| 1568 |
|
| 1569 | if (!--bdev->bd_openers) {
|
| 1570 | WARN_ON_ONCE(bdev->bd_holders);
|
| 1571 | sync_blockdev(bdev);
|
| 1572 | kill_bdev(bdev);
|
| 1573 | bdev_write_inode(bdev);
|
| 1574 | if (bdev_is_partition(bdev))
|
| 1575 | victim = bdev_whole(bdev);
|
| 1576 | }
|
| 1577 |
|
| 1578 | if (!bdev_is_partition(bdev) && disk->fops->release)
|
| 1579 | disk->fops->release(disk, mode);
|
| 1580 | mutex_unlock(&bdev->bd_mutex);
|
| 1581 | if (victim) {
|
| 1582 | __blkdev_put(victim, mode, 1);
|
| 1583 | bdput(victim);
|
| 1584 | }
|
| 1585 | }
|
| 1586 |
|
| 1587 | void blkdev_put(struct block_device *bdev, fmode_t mode)
|
| 1588 | {
|
| 1589 | struct gendisk *disk = bdev->bd_disk;
|
| 1590 |
|
| 1591 | mutex_lock(&bdev->bd_mutex);
|
| 1592 |
|
| 1593 | if (mode & FMODE_EXCL) {
|
| 1594 | struct block_device *whole = bdev_whole(bdev);
|
| 1595 | bool bdev_free;
|
| 1596 |
|
| 1597 | |
| 1598 | * Release a claim on the device. The holder fields
|
| 1599 | * are protected with bdev_lock. bd_mutex is to
|
| 1600 | * synchronize disk_holder unlinking.
|
| 1601 |
|
| 1602 | spin_lock(&bdev_lock);
|
| 1603 |
|
| 1604 | WARN_ON_ONCE(--bdev->bd_holders < 0);
|
| 1605 | WARN_ON_ONCE(--whole->bd_holders < 0);
|
| 1606 |
|
| 1607 | if ((bdev_free = !bdev->bd_holders))
|
| 1608 | bdev->bd_holder = NULL;
|
| 1609 | if (!whole->bd_holders)
|
| 1610 | whole->bd_holder = NULL;
|
| 1611 |
|
| 1612 | spin_unlock(&bdev_lock);
|
| 1613 |
|
| 1614 | |
| 1615 | * If this was the last claim, remove holder link and
|
| 1616 | * unblock evpoll if it was a write holder.
|
| 1617 |
|
| 1618 | if (bdev_free && bdev->bd_write_holder) {
|
| 1619 | disk_unblock_events(disk);
|
| 1620 | bdev->bd_write_holder = false;
|
| 1621 | }
|
| 1622 | }
|
| 1623 |
|
| 1624 | |
| 1625 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE
|
| 1626 | * event. This is to ensure detection of media removal commanded
|
| 1627 | * from userland - e.g. eject(1).
|
| 1628 |
|
| 1629 | disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
|
| 1630 | mutex_unlock(&bdev->bd_mutex);
|
| 1631 |
|
| 1632 | __blkdev_put(bdev, mode, 0);
|
| 1633 | blkdev_put_no_open(bdev);
|
| 1634 | }
|
| 1635 | EXPORT_SYMBOL(blkdev_put);
|
| 1636 |
|
| 1637 | static int blkdev_close(struct inode * inode, struct file * filp)
|
| 1638 | {
|
| 1639 | struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
|
| 1640 | blkdev_put(bdev, filp->f_mode);
|
| 1641 | return 0;
|
| 1642 | }
|
| 1643 |
|
| 1644 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
|
| 1645 | {
|
| 1646 | struct block_device *bdev = I_BDEV(bdev_file_inode(file));
|
| 1647 | fmode_t mode = file->f_mode;
|
| 1648 |
|
| 1649 | |
| 1650 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
|
| 1651 | * to updated it before every ioctl.
|
| 1652 |
|
| 1653 | if (file->f_flags & O_NDELAY)
|
| 1654 | mode |= FMODE_NDELAY;
|
| 1655 | else
|
| 1656 | mode &= ~FMODE_NDELAY;
|
| 1657 |
|
| 1658 | return blkdev_ioctl(bdev, mode, cmd, arg);
|
| 1659 | }
|
| 1660 |
|
| 1661 | |
| 1662 | * Write data to the block device. Only intended for the block device itself
|
| 1663 | * and the raw driver which basically is a fake block device.
|
| 1664 | *
|
| 1665 | * Does not take i_mutex for the write and thus is not for general purpose
|
| 1666 | * use.
|
| 1667 |
|
| 1668 | ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
|
| 1669 | {
|
| 1670 | struct file *file = iocb->ki_filp;
|
| 1671 | struct inode *bd_inode = bdev_file_inode(file);
|
| 1672 | loff_t size = i_size_read(bd_inode);
|
| 1673 | struct blk_plug plug;
|
| 1674 | ssize_t ret;
|
| 1675 |
|
| 1676 | if (bdev_read_only(I_BDEV(bd_inode)))
|
| 1677 | return -EPERM;
|
| 1678 |
|
| 1679 | if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
|
| 1680 | return -ETXTBSY;
|
| 1681 |
|
| 1682 | if (!iov_iter_count(from))
|
| 1683 | return 0;
|
| 1684 |
|
| 1685 | if (iocb->ki_pos >= size)
|
| 1686 | return -ENOSPC;
|
| 1687 |
|
| 1688 | if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
|
| 1689 | return -EOPNOTSUPP;
|
| 1690 |
|
| 1691 | iov_iter_truncate(from, size - iocb->ki_pos);
|
| 1692 |
|
| 1693 | blk_start_plug(&plug);
|
| 1694 | ret = __generic_file_write_iter(iocb, from);
|
| 1695 | if (ret > 0)
|
| 1696 | ret = generic_write_sync(iocb, ret);
|
| 1697 | blk_finish_plug(&plug);
|
| 1698 | return ret;
|
| 1699 | }
|
| 1700 | EXPORT_SYMBOL_GPL(blkdev_write_iter);
|
| 1701 |
|
| 1702 | ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
|
| 1703 | {
|
| 1704 | struct file *file = iocb->ki_filp;
|
| 1705 | struct inode *bd_inode = bdev_file_inode(file);
|
| 1706 | loff_t size = i_size_read(bd_inode);
|
| 1707 | loff_t pos = iocb->ki_pos;
|
| 1708 |
|
| 1709 | if (pos >= size)
|
| 1710 | return 0;
|
| 1711 |
|
| 1712 | size -= pos;
|
| 1713 | iov_iter_truncate(to, size);
|
| 1714 | return generic_file_read_iter(iocb, to);
|
| 1715 | }
|
| 1716 | EXPORT_SYMBOL_GPL(blkdev_read_iter);
|
| 1717 |
|
| 1718 | |
| 1719 | * Try to release a page associated with block device when the system
|
| 1720 | * is under memory pressure.
|
| 1721 |
|
| 1722 | static int blkdev_releasepage(struct page *page, gfp_t wait)
|
| 1723 | {
|
| 1724 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
|
| 1725 |
|
| 1726 | if (super && super->s_op->bdev_try_to_free_page)
|
| 1727 | return super->s_op->bdev_try_to_free_page(super, page, wait);
|
| 1728 |
|
| 1729 | return try_to_free_buffers(page);
|
| 1730 | }
|
| 1731 |
|
| 1732 | static int blkdev_writepages(struct address_space *mapping,
|
| 1733 | struct writeback_control *wbc)
|
| 1734 | {
|
| 1735 | return generic_writepages(mapping, wbc);
|
| 1736 | }
|
| 1737 |
|
| 1738 | static const struct address_space_operations def_blk_aops = {
|
| 1739 | .readpage = blkdev_readpage,
|
| 1740 | .readahead = blkdev_readahead,
|
| 1741 | .writepage = blkdev_writepage,
|
| 1742 | .write_begin = blkdev_write_begin,
|
| 1743 | .write_end = blkdev_write_end,
|
| 1744 | .writepages = blkdev_writepages,
|
| 1745 | .releasepage = blkdev_releasepage,
|
| 1746 | .direct_IO = blkdev_direct_IO,
|
| 1747 | .migratepage = buffer_migrate_page_norefs,
|
| 1748 | .is_dirty_writeback = buffer_check_dirty_writeback,
|
| 1749 | };
|
| 1750 |
|
| 1751 | #define BLKDEV_FALLOC_FL_SUPPORTED \
|
| 1752 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
|
| 1753 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
|
| 1754 |
|
| 1755 | static long blkdev_fallocate(struct file *file, int mode, loff_t start,
|
| 1756 | loff_t len)
|
| 1757 | {
|
| 1758 | struct block_device *bdev = I_BDEV(bdev_file_inode(file));
|
| 1759 | loff_t end = start + len - 1;
|
| 1760 | loff_t isize;
|
| 1761 | int error;
|
| 1762 |
|
| 1763 |
|
| 1764 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
|
| 1765 | return -EOPNOTSUPP;
|
| 1766 |
|
| 1767 |
|
| 1768 | isize = i_size_read(bdev->bd_inode);
|
| 1769 | if (start >= isize)
|
| 1770 | return -EINVAL;
|
| 1771 | if (end >= isize) {
|
| 1772 | if (mode & FALLOC_FL_KEEP_SIZE) {
|
| 1773 | len = isize - start;
|
| 1774 | end = start + len - 1;
|
| 1775 | } else
|
| 1776 | return -EINVAL;
|
| 1777 | }
|
| 1778 |
|
| 1779 | |
| 1780 | * Don't allow IO that isn't aligned to logical block size.
|
| 1781 |
|
| 1782 | if ((start | len) & (bdev_logical_block_size(bdev) - 1))
|
| 1783 | return -EINVAL;
|
| 1784 |
|
| 1785 |
|
| 1786 | error = truncate_bdev_range(bdev, file->f_mode, start, end);
|
| 1787 | if (error)
|
| 1788 | return error;
|
| 1789 |
|
| 1790 | switch (mode) {
|
| 1791 | case FALLOC_FL_ZERO_RANGE:
|
| 1792 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
|
| 1793 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
|
| 1794 | GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
|
| 1795 | break;
|
| 1796 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
|
| 1797 | error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
|
| 1798 | GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
|
| 1799 | break;
|
| 1800 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
|
| 1801 | error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
|
| 1802 | GFP_KERNEL, 0);
|
| 1803 | break;
|
| 1804 | default:
|
| 1805 | return -EOPNOTSUPP;
|
| 1806 | }
|
| 1807 | if (error)
|
| 1808 | return error;
|
| 1809 |
|
| 1810 | |
| 1811 | * Invalidate again; if someone wandered in and dirtied a page,
|
| 1812 | * the caller will be given -EBUSY. The third argument is
|
| 1813 | * inclusive, so the rounding here is safe.
|
| 1814 |
|
| 1815 | return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
|
| 1816 | start >> PAGE_SHIFT,
|
| 1817 | end >> PAGE_SHIFT);
|
| 1818 | }
|
| 1819 |
|
| 1820 | const struct file_operations def_blk_fops = {
|
| 1821 | .open = blkdev_open,
|
| 1822 | .release = blkdev_close,
|
| 1823 | .llseek = block_llseek,
|
| 1824 | .read_iter = blkdev_read_iter,
|
| 1825 | .write_iter = blkdev_write_iter,
|
| 1826 | .iopoll = blkdev_iopoll,
|
| 1827 | .mmap = generic_file_mmap,
|
| 1828 | .fsync = blkdev_fsync,
|
| 1829 | .unlocked_ioctl = block_ioctl,
|
| 1830 | #ifdef CONFIG_COMPAT
|
| 1831 | .compat_ioctl = compat_blkdev_ioctl,
|
| 1832 | #endif
|
| 1833 | .splice_read = generic_file_splice_read,
|
| 1834 | .splice_write = iter_file_splice_write,
|
| 1835 | .fallocate = blkdev_fallocate,
|
| 1836 | };
|
| 1837 |
|
| 1838 | |
| 1839 | * lookup_bdev - lookup a struct block_device by name
|
| 1840 | * @pathname: special file representing the block device
|
| 1841 | * @dev: return value of the block device's dev_t
|
| 1842 | *
|
| 1843 | * Get a reference to the blockdevice at @pathname in the current
|
| 1844 | * namespace if possible and return it. Return ERR_PTR(error)
|
| 1845 | * otherwise.
|
| 1846 |
|
| 1847 | int lookup_bdev(const char *pathname, dev_t *dev)
|
| 1848 | {
|
| 1849 | struct inode *inode;
|
| 1850 | struct path path;
|
| 1851 | int error;
|
| 1852 |
|
| 1853 | if (!pathname || !*pathname)
|
| 1854 | return -EINVAL;
|
| 1855 |
|
| 1856 | error = kern_path(pathname, LOOKUP_FOLLOW, &path);
|
| 1857 | if (error)
|
| 1858 | return error;
|
| 1859 |
|
| 1860 | inode = d_backing_inode(path.dentry);
|
| 1861 | error = -ENOTBLK;
|
| 1862 | if (!S_ISBLK(inode->i_mode))
|
| 1863 | goto out_path_put;
|
| 1864 | error = -EACCES;
|
| 1865 | if (!may_open_dev(&path))
|
| 1866 | goto out_path_put;
|
| 1867 |
|
| 1868 | *dev = inode->i_rdev;
|
| 1869 | error = 0;
|
| 1870 | out_path_put:
|
| 1871 | path_put(&path);
|
| 1872 | return error;
|
| 1873 | }
|
| 1874 | EXPORT_SYMBOL(lookup_bdev);
|
| 1875 |
|
| 1876 | int __invalidate_device(struct block_device *bdev, bool kill_dirty)
|
| 1877 | {
|
| 1878 | struct super_block *sb = get_super(bdev);
|
| 1879 | int res = 0;
|
| 1880 |
|
| 1881 | if (sb) {
|
| 1882 | |
| 1883 | * no need to lock the super, get_super holds the
|
| 1884 | * read mutex so the filesystem cannot go away
|
| 1885 | * under us (->put_super runs with the write lock
|
| 1886 | * hold).
|
| 1887 |
|
| 1888 | shrink_dcache_sb(sb);
|
| 1889 | res = invalidate_inodes(sb, kill_dirty);
|
| 1890 | drop_super(sb);
|
| 1891 | }
|
| 1892 | invalidate_bdev(bdev);
|
| 1893 | return res;
|
| 1894 | }
|
| 1895 | EXPORT_SYMBOL(__invalidate_device);
|
| 1896 |
|
| 1897 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
|
| 1898 | {
|
| 1899 | struct inode *inode, *old_inode = NULL;
|
| 1900 |
|
| 1901 | spin_lock(&blockdev_superblock->s_inode_list_lock);
|
| 1902 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
|
| 1903 | struct address_space *mapping = inode->i_mapping;
|
| 1904 | struct block_device *bdev;
|
| 1905 |
|
| 1906 | spin_lock(&inode->i_lock);
|
| 1907 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
|
| 1908 | mapping->nrpages == 0) {
|
| 1909 | spin_unlock(&inode->i_lock);
|
| 1910 | continue;
|
| 1911 | }
|
| 1912 | __iget(inode);
|
| 1913 | spin_unlock(&inode->i_lock);
|
| 1914 | spin_unlock(&blockdev_superblock->s_inode_list_lock);
|
| 1915 | |
| 1916 | * We hold a reference to 'inode' so it couldn't have been
|
| 1917 | * removed from s_inodes list while we dropped the
|
| 1918 | * s_inode_list_lock We cannot iput the inode now as we can
|
| 1919 | * be holding the last reference and we cannot iput it under
|
| 1920 | * s_inode_list_lock. So we keep the reference and iput it
|
| 1921 | * later.
|
| 1922 |
|
| 1923 | iput(old_inode);
|
| 1924 | old_inode = inode;
|
| 1925 | bdev = I_BDEV(inode);
|
| 1926 |
|
| 1927 | mutex_lock(&bdev->bd_mutex);
|
| 1928 | if (bdev->bd_openers)
|
| 1929 | func(bdev, arg);
|
| 1930 | mutex_unlock(&bdev->bd_mutex);
|
| 1931 |
|
| 1932 | spin_lock(&blockdev_superblock->s_inode_list_lock);
|
| 1933 | }
|
| 1934 | spin_unlock(&blockdev_superblock->s_inode_list_lock);
|
| 1935 | iput(old_inode);
|
| 1936 | }
|