block/fops.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 1991, 1992  Linus Torvalds
   4  * Copyright (C) 2001  Andrea Arcangeli <andrea@suse.de> SuSE
   5  * Copyright (C) 2016 - 2020 Christoph Hellwig
   6  */
   7 #include <linux/init.h>
   8 #include <linux/mm.h>
   9 #include <linux/blkdev.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/mpage.h>
  12 #include <linux/uio.h>
  13 #include <linux/namei.h>
  14 #include <linux/task_io_accounting_ops.h>
  15 #include <linux/falloc.h>
  16 #include <linux/suspend.h>
  17 #include <linux/fs.h>
  18 #include <linux/iomap.h>
  19 #include <linux/module.h>
  20 #include <linux/io_uring/cmd.h>
  21 #include "blk.h"
  22
  23 static inline struct inode *bdev_file_inode(struct file *file)
  24 {
  25         return file->f_mapping->host;
  26 }
  27
  28 static blk_opf_t dio_bio_write_op(struct kiocb *iocb)
  29 {
  30         blk_opf_t opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
  31
  32         /* avoid the need for a I/O completion work item */
  33         if (iocb_is_dsync(iocb))
  34                 opf |= REQ_FUA;
  35         return opf;
  36 }
  37
  38 static bool blkdev_dio_invalid(struct block_device *bdev, struct kiocb *iocb,
  39                                 struct iov_iter *iter)
  40 {
  41         return iocb->ki_pos & (bdev_logical_block_size(bdev) - 1) ||
  42                 !bdev_iter_is_aligned(bdev, iter);
  43 }
  44
  45 #define DIO_INLINE_BIO_VECS 4
  46
  47 static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb,
  48                 struct iov_iter *iter, struct block_device *bdev,
  49                 unsigned int nr_pages)
  50 {
  51         struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
  52         loff_t pos = iocb->ki_pos;
  53         bool should_dirty = false;
  54         struct bio bio;
  55         ssize_t ret;
  56
  57         if (nr_pages <= DIO_INLINE_BIO_VECS)
  58                 vecs = inline_vecs;
  59         else {
  60                 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
  61                                      GFP_KERNEL);
  62                 if (!vecs)
  63                         return -ENOMEM;
  64         }
  65
  66         if (iov_iter_rw(iter) == READ) {
  67                 bio_init(&bio, bdev, vecs, nr_pages, REQ_OP_READ);
  68                 if (user_backed_iter(iter))
  69                         should_dirty = true;
  70         } else {
  71                 bio_init(&bio, bdev, vecs, nr_pages, dio_bio_write_op(iocb));
  72         }
  73         bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT;
  74         bio.bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
  75         bio.bi_ioprio = iocb->ki_ioprio;
  76         if (iocb->ki_flags & IOCB_ATOMIC)
  77                 bio.bi_opf |= REQ_ATOMIC;
  78
  79         ret = bio_iov_iter_get_pages(&bio, iter);
  80         if (unlikely(ret))
  81                 goto out;
  82         ret = bio.bi_iter.bi_size;
  83
  84         if (iov_iter_rw(iter) == WRITE)
  85                 task_io_account_write(ret);
  86
  87         if (iocb->ki_flags & IOCB_NOWAIT)
  88                 bio.bi_opf |= REQ_NOWAIT;
  89
  90         submit_bio_wait(&bio);
  91
  92         bio_release_pages(&bio, should_dirty);
  93         if (unlikely(bio.bi_status))
  94                 ret = blk_status_to_errno(bio.bi_status);
  95
  96 out:
  97         if (vecs != inline_vecs)
  98                 kfree(vecs);
  99
 100         bio_uninit(&bio);
 101
 102         return ret;
 103 }
 104
 105 enum {
 106         DIO_SHOULD_DIRTY        = 1,
 107         DIO_IS_SYNC             = 2,
 108 };
 109
 110 struct blkdev_dio {
 111         union {
 112                 struct kiocb            *iocb;
 113                 struct task_struct      *waiter;
 114         };
 115         size_t                  size;
 116         atomic_t                ref;
 117         unsigned int            flags;
 118         struct bio              bio ____cacheline_aligned_in_smp;
 119 };
 120
 121 static struct bio_set blkdev_dio_pool;
 122
 123 static void blkdev_bio_end_io(struct bio *bio)
 124 {
 125         struct blkdev_dio *dio = bio->bi_private;
 126         bool should_dirty = dio->flags & DIO_SHOULD_DIRTY;
 127
 128         if (bio->bi_status && !dio->bio.bi_status)
 129                 dio->bio.bi_status = bio->bi_status;
 130
 131         if (atomic_dec_and_test(&dio->ref)) {
 132                 if (!(dio->flags & DIO_IS_SYNC)) {
 133                         struct kiocb *iocb = dio->iocb;
 134                         ssize_t ret;
 135
 136                         WRITE_ONCE(iocb->private, NULL);
 137
 138                         if (likely(!dio->bio.bi_status)) {
 139                                 ret = dio->size;
 140                                 iocb->ki_pos += ret;
 141                         } else {
 142                                 ret = blk_status_to_errno(dio->bio.bi_status);
 143                         }
 144
 145                         dio->iocb->ki_complete(iocb, ret);
 146                         bio_put(&dio->bio);
 147                 } else {
 148                         struct task_struct *waiter = dio->waiter;
 149
 150                         WRITE_ONCE(dio->waiter, NULL);
 151                         blk_wake_io_task(waiter);
 152                 }
 153         }
 154
 155         if (should_dirty) {
 156                 bio_check_pages_dirty(bio);
 157         } else {
 158                 bio_release_pages(bio, false);
 159                 bio_put(bio);
 160         }
 161 }
 162
 163 static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 164                 struct block_device *bdev, unsigned int nr_pages)
 165 {
 166         struct blk_plug plug;
 167         struct blkdev_dio *dio;
 168         struct bio *bio;
 169         bool is_read = (iov_iter_rw(iter) == READ), is_sync;
 170         blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb);
 171         loff_t pos = iocb->ki_pos;
 172         int ret = 0;
 173
 174         if (iocb->ki_flags & IOCB_ALLOC_CACHE)
 175                 opf |= REQ_ALLOC_CACHE;
 176         bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
 177                                &blkdev_dio_pool);
 178         dio = container_of(bio, struct blkdev_dio, bio);
 179         atomic_set(&dio->ref, 1);
 180         /*
 181          * Grab an extra reference to ensure the dio structure which is embedded
 182          * into the first bio stays around.
 183          */
 184         bio_get(bio);
 185
 186         is_sync = is_sync_kiocb(iocb);
 187         if (is_sync) {
 188                 dio->flags = DIO_IS_SYNC;
 189                 dio->waiter = current;
 190         } else {
 191                 dio->flags = 0;
 192                 dio->iocb = iocb;
 193         }
 194
 195         dio->size = 0;
 196         if (is_read && user_backed_iter(iter))
 197                 dio->flags |= DIO_SHOULD_DIRTY;
 198
 199         blk_start_plug(&plug);
 200
 201         for (;;) {
 202                 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
 203                 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
 204                 bio->bi_private = dio;
 205                 bio->bi_end_io = blkdev_bio_end_io;
 206                 bio->bi_ioprio = iocb->ki_ioprio;
 207
 208                 ret = bio_iov_iter_get_pages(bio, iter);
 209                 if (unlikely(ret)) {
 210                         bio->bi_status = BLK_STS_IOERR;
 211                         bio_endio(bio);
 212                         break;
 213                 }
 214                 if (iocb->ki_flags & IOCB_NOWAIT) {
 215                         /*
 216                          * This is nonblocking IO, and we need to allocate
 217                          * another bio if we have data left to map. As we
 218                          * cannot guarantee that one of the sub bios will not
 219                          * fail getting issued FOR NOWAIT and as error results
 220                          * are coalesced across all of them, be safe and ask for
 221                          * a retry of this from blocking context.
 222                          */
 223                         if (unlikely(iov_iter_count(iter))) {
 224                                 bio_release_pages(bio, false);
 225                                 bio_clear_flag(bio, BIO_REFFED);
 226                                 bio_put(bio);
 227                                 blk_finish_plug(&plug);
 228                                 return -EAGAIN;
 229                         }
 230                         bio->bi_opf |= REQ_NOWAIT;
 231                 }
 232
 233                 if (is_read) {
 234                         if (dio->flags & DIO_SHOULD_DIRTY)
 235                                 bio_set_pages_dirty(bio);
 236                 } else {
 237                         task_io_account_write(bio->bi_iter.bi_size);
 238                 }
 239                 dio->size += bio->bi_iter.bi_size;
 240                 pos += bio->bi_iter.bi_size;
 241
 242                 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS);
 243                 if (!nr_pages) {
 244                         submit_bio(bio);
 245                         break;
 246                 }
 247                 atomic_inc(&dio->ref);
 248                 submit_bio(bio);
 249                 bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL);
 250         }
 251
 252         blk_finish_plug(&plug);
 253
 254         if (!is_sync)
 255                 return -EIOCBQUEUED;
 256
 257         for (;;) {
 258                 set_current_state(TASK_UNINTERRUPTIBLE);
 259                 if (!READ_ONCE(dio->waiter))
 260                         break;
 261                 blk_io_schedule();
 262         }
 263         __set_current_state(TASK_RUNNING);
 264
 265         if (!ret)
 266                 ret = blk_status_to_errno(dio->bio.bi_status);
 267         if (likely(!ret))
 268                 ret = dio->size;
 269
 270         bio_put(&dio->bio);
 271         return ret;
 272 }
 273
 274 static void blkdev_bio_end_io_async(struct bio *bio)
 275 {
 276         struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio);
 277         struct kiocb *iocb = dio->iocb;
 278         ssize_t ret;
 279
 280         WRITE_ONCE(iocb->private, NULL);
 281
 282         if (likely(!bio->bi_status)) {
 283                 ret = dio->size;
 284                 iocb->ki_pos += ret;
 285         } else {
 286                 ret = blk_status_to_errno(bio->bi_status);
 287         }
 288
 289         iocb->ki_complete(iocb, ret);
 290
 291         if (dio->flags & DIO_SHOULD_DIRTY) {
 292                 bio_check_pages_dirty(bio);
 293         } else {
 294                 bio_release_pages(bio, false);
 295                 bio_put(bio);
 296         }
 297 }
 298
 299 static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
 300                                         struct iov_iter *iter,
 301                                         struct block_device *bdev,
 302                                         unsigned int nr_pages)
 303 {
 304         bool is_read = iov_iter_rw(iter) == READ;
 305         blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb);
 306         struct blkdev_dio *dio;
 307         struct bio *bio;
 308         loff_t pos = iocb->ki_pos;
 309         int ret = 0;
 310
 311         if (iocb->ki_flags & IOCB_ALLOC_CACHE)
 312                 opf |= REQ_ALLOC_CACHE;
 313         bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
 314                                &blkdev_dio_pool);
 315         dio = container_of(bio, struct blkdev_dio, bio);
 316         dio->flags = 0;
 317         dio->iocb = iocb;
 318         bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
 319         bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
 320         bio->bi_end_io = blkdev_bio_end_io_async;
 321         bio->bi_ioprio = iocb->ki_ioprio;
 322
 323         if (iov_iter_is_bvec(iter)) {
 324                 /*
 325                  * Users don't rely on the iterator being in any particular
 326                  * state for async I/O returning -EIOCBQUEUED, hence we can
 327                  * avoid expensive iov_iter_advance(). Bypass
 328                  * bio_iov_iter_get_pages() and set the bvec directly.
 329                  */
 330                 bio_iov_bvec_set(bio, iter);
 331         } else {
 332                 ret = bio_iov_iter_get_pages(bio, iter);
 333                 if (unlikely(ret)) {
 334                         bio_put(bio);
 335                         return ret;
 336                 }
 337         }
 338         dio->size = bio->bi_iter.bi_size;
 339
 340         if (is_read) {
 341                 if (user_backed_iter(iter)) {
 342                         dio->flags |= DIO_SHOULD_DIRTY;
 343                         bio_set_pages_dirty(bio);
 344                 }
 345         } else {
 346                 task_io_account_write(bio->bi_iter.bi_size);
 347         }
 348
 349         if (iocb->ki_flags & IOCB_ATOMIC)
 350                 bio->bi_opf |= REQ_ATOMIC;
 351
 352         if (iocb->ki_flags & IOCB_NOWAIT)
 353                 bio->bi_opf |= REQ_NOWAIT;
 354
 355         if (iocb->ki_flags & IOCB_HIPRI) {
 356                 bio->bi_opf |= REQ_POLLED;
 357                 submit_bio(bio);
 358                 WRITE_ONCE(iocb->private, bio);
 359         } else {
 360                 submit_bio(bio);
 361         }
 362         return -EIOCBQUEUED;
 363 }
 364
 365 static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 366 {
 367         struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
 368         unsigned int nr_pages;
 369
 370         if (!iov_iter_count(iter))
 371                 return 0;
 372
 373         if (blkdev_dio_invalid(bdev, iocb, iter))
 374                 return -EINVAL;
 375
 376         nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
 377         if (likely(nr_pages <= BIO_MAX_VECS)) {
 378                 if (is_sync_kiocb(iocb))
 379                         return __blkdev_direct_IO_simple(iocb, iter, bdev,
 380                                                         nr_pages);
 381                 return __blkdev_direct_IO_async(iocb, iter, bdev, nr_pages);
 382         } else if (iocb->ki_flags & IOCB_ATOMIC) {
 383                 return -EINVAL;
 384         }
 385         return __blkdev_direct_IO(iocb, iter, bdev, bio_max_segs(nr_pages));
 386 }
 387
 388 static int blkdev_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 389                 unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
 390 {
 391         struct block_device *bdev = I_BDEV(inode);
 392         loff_t isize = i_size_read(inode);
 393
 394         if (offset >= isize)
 395                 return -EIO;
 396
 397         iomap->bdev = bdev;
 398         iomap->offset = ALIGN_DOWN(offset, bdev_logical_block_size(bdev));
 399         iomap->type = IOMAP_MAPPED;
 400         iomap->addr = iomap->offset;
 401         iomap->length = isize - iomap->offset;
 402         iomap->flags |= IOMAP_F_BUFFER_HEAD; /* noop for !CONFIG_BUFFER_HEAD */
 403         return 0;
 404 }
 405
 406 static const struct iomap_ops blkdev_iomap_ops = {
 407         .iomap_begin            = blkdev_iomap_begin,
 408 };
 409
 410 #ifdef CONFIG_BUFFER_HEAD
 411 static int blkdev_get_block(struct inode *inode, sector_t iblock,
 412                 struct buffer_head *bh, int create)
 413 {
 414         bh->b_bdev = I_BDEV(inode);
 415         bh->b_blocknr = iblock;
 416         set_buffer_mapped(bh);
 417         return 0;
 418 }
 419
 420 /*
 421  * We cannot call mpage_writepages() as it does not take the buffer lock.
 422  * We must use block_write_full_folio() directly which holds the buffer
 423  * lock.  The buffer lock provides the synchronisation with writeback
 424  * that filesystems rely on when they use the blockdev's mapping.
 425  */
 426 static int blkdev_writepages(struct address_space *mapping,
 427                 struct writeback_control *wbc)
 428 {
 429         struct blk_plug plug;
 430         int err;
 431
 432         blk_start_plug(&plug);
 433         err = write_cache_pages(mapping, wbc, block_write_full_folio,
 434                         blkdev_get_block);
 435         blk_finish_plug(&plug);
 436
 437         return err;
 438 }
 439
 440 static int blkdev_read_folio(struct file *file, struct folio *folio)
 441 {
 442         return block_read_full_folio(folio, blkdev_get_block);
 443 }
 444
 445 static void blkdev_readahead(struct readahead_control *rac)
 446 {
 447         mpage_readahead(rac, blkdev_get_block);
 448 }
 449
 450 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 451                 loff_t pos, unsigned len, struct folio **foliop, void **fsdata)
 452 {
 453         return block_write_begin(mapping, pos, len, foliop, blkdev_get_block);
 454 }
 455
 456 static int blkdev_write_end(struct file *file, struct address_space *mapping,
 457                 loff_t pos, unsigned len, unsigned copied, struct folio *folio,
 458                 void *fsdata)
 459 {
 460         int ret;
 461         ret = block_write_end(file, mapping, pos, len, copied, folio, fsdata);
 462
 463         folio_unlock(folio);
 464         folio_put(folio);
 465
 466         return ret;
 467 }
 468
 469 const struct address_space_operations def_blk_aops = {
 470         .dirty_folio    = block_dirty_folio,
 471         .invalidate_folio = block_invalidate_folio,
 472         .read_folio     = blkdev_read_folio,
 473         .readahead      = blkdev_readahead,
 474         .writepages     = blkdev_writepages,
 475         .write_begin    = blkdev_write_begin,
 476         .write_end      = blkdev_write_end,
 477         .migrate_folio  = buffer_migrate_folio_norefs,
 478         .is_dirty_writeback = buffer_check_dirty_writeback,
 479 };
 480 #else /* CONFIG_BUFFER_HEAD */
 481 static int blkdev_read_folio(struct file *file, struct folio *folio)
 482 {
 483         return iomap_read_folio(folio, &blkdev_iomap_ops);
 484 }
 485
 486 static void blkdev_readahead(struct readahead_control *rac)
 487 {
 488         iomap_readahead(rac, &blkdev_iomap_ops);
 489 }
 490
 491 static int blkdev_map_blocks(struct iomap_writepage_ctx *wpc,
 492                 struct inode *inode, loff_t offset, unsigned int len)
 493 {
 494         loff_t isize = i_size_read(inode);
 495
 496         if (WARN_ON_ONCE(offset >= isize))
 497                 return -EIO;
 498         if (offset >= wpc->iomap.offset &&
 499             offset < wpc->iomap.offset + wpc->iomap.length)
 500                 return 0;
 501         return blkdev_iomap_begin(inode, offset, isize - offset,
 502                                   IOMAP_WRITE, &wpc->iomap, NULL);
 503 }
 504
 505 static const struct iomap_writeback_ops blkdev_writeback_ops = {
 506         .map_blocks             = blkdev_map_blocks,
 507 };
 508
 509 static int blkdev_writepages(struct address_space *mapping,
 510                 struct writeback_control *wbc)
 511 {
 512         struct iomap_writepage_ctx wpc = { };
 513
 514         return iomap_writepages(mapping, wbc, &wpc, &blkdev_writeback_ops);
 515 }
 516
 517 const struct address_space_operations def_blk_aops = {
 518         .dirty_folio    = filemap_dirty_folio,
 519         .release_folio          = iomap_release_folio,
 520         .invalidate_folio       = iomap_invalidate_folio,
 521         .read_folio             = blkdev_read_folio,
 522         .readahead              = blkdev_readahead,
 523         .writepages             = blkdev_writepages,
 524         .is_partially_uptodate  = iomap_is_partially_uptodate,
 525         .error_remove_folio     = generic_error_remove_folio,
 526         .migrate_folio          = filemap_migrate_folio,
 527 };
 528 #endif /* CONFIG_BUFFER_HEAD */
 529
 530 /*
 531  * for a block special file file_inode(file)->i_size is zero
 532  * so we compute the size by hand (just as in block_read/write above)
 533  */
 534 static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence)
 535 {
 536         struct inode *bd_inode = bdev_file_inode(file);
 537         loff_t retval;
 538
 539         inode_lock(bd_inode);
 540         retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
 541         inode_unlock(bd_inode);
 542         return retval;
 543 }
 544
 545 static int blkdev_fsync(struct file *filp, loff_t start, loff_t end,
 546                 int datasync)
 547 {
 548         struct block_device *bdev = I_BDEV(filp->f_mapping->host);
 549         int error;
 550
 551         error = file_write_and_wait_range(filp, start, end);
 552         if (error)
 553                 return error;
 554
 555         /*
 556          * There is no need to serialise calls to blkdev_issue_flush with
 557          * i_mutex and doing so causes performance issues with concurrent
 558          * O_SYNC writers to a block device.
 559          */
 560         error = blkdev_issue_flush(bdev);
 561         if (error == -EOPNOTSUPP)
 562                 error = 0;
 563
 564         return error;
 565 }
 566
 567 /**
 568  * file_to_blk_mode - get block open flags from file flags
 569  * @file: file whose open flags should be converted
 570  *
 571  * Look at file open flags and generate corresponding block open flags from
 572  * them. The function works both for file just being open (e.g. during ->open
 573  * callback) and for file that is already open. This is actually non-trivial
 574  * (see comment in the function).
 575  */
 576 blk_mode_t file_to_blk_mode(struct file *file)
 577 {
 578         blk_mode_t mode = 0;
 579
 580         if (file->f_mode & FMODE_READ)
 581                 mode |= BLK_OPEN_READ;
 582         if (file->f_mode & FMODE_WRITE)
 583                 mode |= BLK_OPEN_WRITE;
 584         /*
 585          * do_dentry_open() clears O_EXCL from f_flags, use file->private_data
 586          * to determine whether the open was exclusive for already open files.
 587          */
 588         if (file->private_data)
 589                 mode |= BLK_OPEN_EXCL;
 590         else if (file->f_flags & O_EXCL)
 591                 mode |= BLK_OPEN_EXCL;
 592         if (file->f_flags & O_NDELAY)
 593                 mode |= BLK_OPEN_NDELAY;
 594
 595         /*
 596          * If all bits in O_ACCMODE set (aka O_RDWR | O_WRONLY), the floppy
 597          * driver has historically allowed ioctls as if the file was opened for
 598          * writing, but does not allow and actual reads or writes.
 599          */
 600         if ((file->f_flags & O_ACCMODE) == (O_RDWR | O_WRONLY))
 601                 mode |= BLK_OPEN_WRITE_IOCTL;
 602
 603         return mode;
 604 }
 605
 606 static int blkdev_open(struct inode *inode, struct file *filp)
 607 {
 608         struct block_device *bdev;
 609         blk_mode_t mode;
 610         int ret;
 611
 612         mode = file_to_blk_mode(filp);
 613         /* Use the file as the holder. */
 614         if (mode & BLK_OPEN_EXCL)
 615                 filp->private_data = filp;
 616         ret = bdev_permission(inode->i_rdev, mode, filp->private_data);
 617         if (ret)
 618                 return ret;
 619
 620         bdev = blkdev_get_no_open(inode->i_rdev);
 621         if (!bdev)
 622                 return -ENXIO;
 623
 624         if (bdev_can_atomic_write(bdev))
 625                 filp->f_mode |= FMODE_CAN_ATOMIC_WRITE;
 626
 627         ret = bdev_open(bdev, mode, filp->private_data, NULL, filp);
 628         if (ret)
 629                 blkdev_put_no_open(bdev);
 630         return ret;
 631 }
 632
 633 static int blkdev_release(struct inode *inode, struct file *filp)
 634 {
 635         bdev_release(filp);
 636         return 0;
 637 }
 638
 639 static ssize_t
 640 blkdev_direct_write(struct kiocb *iocb, struct iov_iter *from)
 641 {
 642         size_t count = iov_iter_count(from);
 643         ssize_t written;
 644
 645         written = kiocb_invalidate_pages(iocb, count);
 646         if (written) {
 647                 if (written == -EBUSY)
 648                         return 0;
 649                 return written;
 650         }
 651
 652         written = blkdev_direct_IO(iocb, from);
 653         if (written > 0) {
 654                 kiocb_invalidate_post_direct_write(iocb, count);
 655                 iocb->ki_pos += written;
 656                 count -= written;
 657         }
 658         if (written != -EIOCBQUEUED)
 659                 iov_iter_revert(from, count - iov_iter_count(from));
 660         return written;
 661 }
 662
 663 static ssize_t blkdev_buffered_write(struct kiocb *iocb, struct iov_iter *from)
 664 {
 665         return iomap_file_buffered_write(iocb, from, &blkdev_iomap_ops, NULL);
 666 }
 667
 668 /*
 669  * Write data to the block device.  Only intended for the block device itself
 670  * and the raw driver which basically is a fake block device.
 671  *
 672  * Does not take i_mutex for the write and thus is not for general purpose
 673  * use.
 674  */
 675 static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
 676 {
 677         struct file *file = iocb->ki_filp;
 678         struct inode *bd_inode = bdev_file_inode(file);
 679         struct block_device *bdev = I_BDEV(bd_inode);
 680         bool atomic = iocb->ki_flags & IOCB_ATOMIC;
 681         loff_t size = bdev_nr_bytes(bdev);
 682         size_t shorted = 0;
 683         ssize_t ret;
 684
 685         if (bdev_read_only(bdev))
 686                 return -EPERM;
 687
 688         if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
 689                 return -ETXTBSY;
 690
 691         if (!iov_iter_count(from))
 692                 return 0;
 693
 694         if (iocb->ki_pos >= size)
 695                 return -ENOSPC;
 696
 697         if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
 698                 return -EOPNOTSUPP;
 699
 700         if (atomic) {
 701                 ret = generic_atomic_write_valid(iocb, from);
 702                 if (ret)
 703                         return ret;
 704         }
 705
 706         size -= iocb->ki_pos;
 707         if (iov_iter_count(from) > size) {
 708                 if (atomic)
 709                         return -EINVAL;
 710                 shorted = iov_iter_count(from) - size;
 711                 iov_iter_truncate(from, size);
 712         }
 713
 714         ret = file_update_time(file);
 715         if (ret)
 716                 return ret;
 717
 718         if (iocb->ki_flags & IOCB_DIRECT) {
 719                 ret = blkdev_direct_write(iocb, from);
 720                 if (ret >= 0 && iov_iter_count(from))
 721                         ret = direct_write_fallback(iocb, from, ret,
 722                                         blkdev_buffered_write(iocb, from));
 723         } else {
 724                 ret = blkdev_buffered_write(iocb, from);
 725         }
 726
 727         if (ret > 0)
 728                 ret = generic_write_sync(iocb, ret);
 729         iov_iter_reexpand(from, iov_iter_count(from) + shorted);
 730         return ret;
 731 }
 732
 733 static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
 734 {
 735         struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
 736         loff_t size = bdev_nr_bytes(bdev);
 737         loff_t pos = iocb->ki_pos;
 738         size_t shorted = 0;
 739         ssize_t ret = 0;
 740         size_t count;
 741
 742         if (unlikely(pos + iov_iter_count(to) > size)) {
 743                 if (pos >= size)
 744                         return 0;
 745                 size -= pos;
 746                 shorted = iov_iter_count(to) - size;
 747                 iov_iter_truncate(to, size);
 748         }
 749
 750         count = iov_iter_count(to);
 751         if (!count)
 752                 goto reexpand; /* skip atime */
 753
 754         if (iocb->ki_flags & IOCB_DIRECT) {
 755                 ret = kiocb_write_and_wait(iocb, count);
 756                 if (ret < 0)
 757                         goto reexpand;
 758                 file_accessed(iocb->ki_filp);
 759
 760                 ret = blkdev_direct_IO(iocb, to);
 761                 if (ret >= 0) {
 762                         iocb->ki_pos += ret;
 763                         count -= ret;
 764                 }
 765                 iov_iter_revert(to, count - iov_iter_count(to));
 766                 if (ret < 0 || !count)
 767                         goto reexpand;
 768         }
 769
 770         ret = filemap_read(iocb, to, ret);
 771
 772 reexpand:
 773         if (unlikely(shorted))
 774                 iov_iter_reexpand(to, iov_iter_count(to) + shorted);
 775         return ret;
 776 }
 777
 778 #define BLKDEV_FALLOC_FL_SUPPORTED                                      \
 779                 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |           \
 780                  FALLOC_FL_ZERO_RANGE)
 781
 782 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
 783                              loff_t len)
 784 {
 785         struct inode *inode = bdev_file_inode(file);
 786         struct block_device *bdev = I_BDEV(inode);
 787         loff_t end = start + len - 1;
 788         loff_t isize;
 789         int error;
 790
 791         /* Fail if we don't recognize the flags. */
 792         if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
 793                 return -EOPNOTSUPP;
 794
 795         /* Don't go off the end of the device. */
 796         isize = bdev_nr_bytes(bdev);
 797         if (start >= isize)
 798                 return -EINVAL;
 799         if (end >= isize) {
 800                 if (mode & FALLOC_FL_KEEP_SIZE) {
 801                         len = isize - start;
 802                         end = start + len - 1;
 803                 } else
 804                         return -EINVAL;
 805         }
 806
 807         /*
 808          * Don't allow IO that isn't aligned to logical block size.
 809          */
 810         if ((start | len) & (bdev_logical_block_size(bdev) - 1))
 811                 return -EINVAL;
 812
 813         filemap_invalidate_lock(inode->i_mapping);
 814
 815         /*
 816          * Invalidate the page cache, including dirty pages, for valid
 817          * de-allocate mode calls to fallocate().
 818          */
 819         switch (mode) {
 820         case FALLOC_FL_ZERO_RANGE:
 821         case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
 822                 error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end);
 823                 if (error)
 824                         goto fail;
 825
 826                 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
 827                                              len >> SECTOR_SHIFT, GFP_KERNEL,
 828                                              BLKDEV_ZERO_NOUNMAP);
 829                 break;
 830         case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
 831                 error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end);
 832                 if (error)
 833                         goto fail;
 834
 835                 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
 836                                              len >> SECTOR_SHIFT, GFP_KERNEL,
 837                                              BLKDEV_ZERO_NOFALLBACK);
 838                 break;
 839         default:
 840                 error = -EOPNOTSUPP;
 841         }
 842
 843  fail:
 844         filemap_invalidate_unlock(inode->i_mapping);
 845         return error;
 846 }
 847
 848 static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
 849 {
 850         struct inode *bd_inode = bdev_file_inode(file);
 851
 852         if (bdev_read_only(I_BDEV(bd_inode)))
 853                 return generic_file_readonly_mmap(file, vma);
 854
 855         return generic_file_mmap(file, vma);
 856 }
 857
 858 const struct file_operations def_blk_fops = {
 859         .open           = blkdev_open,
 860         .release        = blkdev_release,
 861         .llseek         = blkdev_llseek,
 862         .read_iter      = blkdev_read_iter,
 863         .write_iter     = blkdev_write_iter,
 864         .iopoll         = iocb_bio_iopoll,
 865         .mmap           = blkdev_mmap,
 866         .fsync          = blkdev_fsync,
 867         .unlocked_ioctl = blkdev_ioctl,
 868 #ifdef CONFIG_COMPAT
 869         .compat_ioctl   = compat_blkdev_ioctl,
 870 #endif
 871         .splice_read    = filemap_splice_read,
 872         .splice_write   = iter_file_splice_write,
 873         .fallocate      = blkdev_fallocate,
 874         .uring_cmd      = blkdev_uring_cmd,
 875         .fop_flags      = FOP_BUFFER_RASYNC,
 876 };
 877
 878 static __init int blkdev_init(void)
 879 {
 880         return bioset_init(&blkdev_dio_pool, 4,
 881                                 offsetof(struct blkdev_dio, bio),
 882                                 BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE);
 883 }
 884 module_init(blkdev_init);