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