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/module.h>
  19 #include "blk.h"
  20
  21 static inline struct inode *bdev_file_inode(struct file *file)
  22 {
  23         return file->f_mapping->host;
  24 }
  25
  26 static int blkdev_get_block(struct inode *inode, sector_t iblock,
  27                 struct buffer_head *bh, int create)
  28 {
  29         bh->b_bdev = I_BDEV(inode);
  30         bh->b_blocknr = iblock;
  31         set_buffer_mapped(bh);
  32         return 0;
  33 }
  34
  35 static unsigned int dio_bio_write_op(struct kiocb *iocb)
  36 {
  37         unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
  38
  39         /* avoid the need for a I/O completion work item */
  40         if (iocb->ki_flags & IOCB_DSYNC)
  41                 op |= REQ_FUA;
  42         return op;
  43 }
  44
  45 #define DIO_INLINE_BIO_VECS 4
  46
  47 static void blkdev_bio_end_io_simple(struct bio *bio)
  48 {
  49         struct task_struct *waiter = bio->bi_private;
  50
  51         WRITE_ONCE(bio->bi_private, NULL);
  52         blk_wake_io_task(waiter);
  53 }
  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 = iocb->ki_filp->private_data;
  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 ((pos | iov_iter_alignment(iter)) &
  66             (bdev_logical_block_size(bdev) - 1))
  67                 return -EINVAL;
  68
  69         if (nr_pages <= DIO_INLINE_BIO_VECS)
  70                 vecs = inline_vecs;
  71         else {
  72                 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
  73                                      GFP_KERNEL);
  74                 if (!vecs)
  75                         return -ENOMEM;
  76         }
  77
  78         bio_init(&bio, vecs, nr_pages);
  79         bio_set_dev(&bio, bdev);
  80         bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT;
  81         bio.bi_write_hint = iocb->ki_hint;
  82         bio.bi_private = current;
  83         bio.bi_end_io = blkdev_bio_end_io_simple;
  84         bio.bi_ioprio = iocb->ki_ioprio;
  85
  86         ret = bio_iov_iter_get_pages(&bio, iter);
  87         if (unlikely(ret))
  88                 goto out;
  89         ret = bio.bi_iter.bi_size;
  90
  91         if (iov_iter_rw(iter) == READ) {
  92                 bio.bi_opf = REQ_OP_READ;
  93                 if (iter_is_iovec(iter))
  94                         should_dirty = true;
  95         } else {
  96                 bio.bi_opf = dio_bio_write_op(iocb);
  97                 task_io_account_write(ret);
  98         }
  99         if (iocb->ki_flags & IOCB_NOWAIT)
 100                 bio.bi_opf |= REQ_NOWAIT;
 101         if (iocb->ki_flags & IOCB_HIPRI)
 102                 bio_set_polled(&bio, iocb);
 103
 104         submit_bio(&bio);
 105         for (;;) {
 106                 set_current_state(TASK_UNINTERRUPTIBLE);
 107                 if (!READ_ONCE(bio.bi_private))
 108                         break;
 109                 if (!(iocb->ki_flags & IOCB_HIPRI) || !bio_poll(&bio, NULL, 0))
 110                         blk_io_schedule();
 111         }
 112         __set_current_state(TASK_RUNNING);
 113
 114         bio_release_pages(&bio, should_dirty);
 115         if (unlikely(bio.bi_status))
 116                 ret = blk_status_to_errno(bio.bi_status);
 117
 118 out:
 119         if (vecs != inline_vecs)
 120                 kfree(vecs);
 121
 122         bio_uninit(&bio);
 123
 124         return ret;
 125 }
 126
 127 enum {
 128         DIO_SHOULD_DIRTY        = 1,
 129         DIO_IS_SYNC             = 2,
 130 };
 131
 132 struct blkdev_dio {
 133         union {
 134                 struct kiocb            *iocb;
 135                 struct task_struct      *waiter;
 136         };
 137         size_t                  size;
 138         atomic_t                ref;
 139         unsigned int            flags;
 140         struct bio              bio ____cacheline_aligned_in_smp;
 141 };
 142
 143 static struct bio_set blkdev_dio_pool;
 144
 145 static void blkdev_bio_end_io(struct bio *bio)
 146 {
 147         struct blkdev_dio *dio = bio->bi_private;
 148         bool should_dirty = dio->flags & DIO_SHOULD_DIRTY;
 149
 150         if (bio->bi_status && !dio->bio.bi_status)
 151                 dio->bio.bi_status = bio->bi_status;
 152
 153         if (atomic_dec_and_test(&dio->ref)) {
 154                 if (!(dio->flags & DIO_IS_SYNC)) {
 155                         struct kiocb *iocb = dio->iocb;
 156                         ssize_t ret;
 157
 158                         WRITE_ONCE(iocb->private, NULL);
 159
 160                         if (likely(!dio->bio.bi_status)) {
 161                                 ret = dio->size;
 162                                 iocb->ki_pos += ret;
 163                         } else {
 164                                 ret = blk_status_to_errno(dio->bio.bi_status);
 165                         }
 166
 167                         dio->iocb->ki_complete(iocb, ret);
 168                         bio_put(&dio->bio);
 169                 } else {
 170                         struct task_struct *waiter = dio->waiter;
 171
 172                         WRITE_ONCE(dio->waiter, NULL);
 173                         blk_wake_io_task(waiter);
 174                 }
 175         }
 176
 177         if (should_dirty) {
 178                 bio_check_pages_dirty(bio);
 179         } else {
 180                 bio_release_pages(bio, false);
 181                 bio_put(bio);
 182         }
 183 }
 184
 185 static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 186                 unsigned int nr_pages)
 187 {
 188         struct block_device *bdev = iocb->ki_filp->private_data;
 189         struct blk_plug plug;
 190         struct blkdev_dio *dio;
 191         struct bio *bio;
 192         bool is_read = (iov_iter_rw(iter) == READ), is_sync;
 193         loff_t pos = iocb->ki_pos;
 194         int ret = 0;
 195
 196         if ((pos | iov_iter_alignment(iter)) &
 197             (bdev_logical_block_size(bdev) - 1))
 198                 return -EINVAL;
 199
 200         bio = bio_alloc_kiocb(iocb, nr_pages, &blkdev_dio_pool);
 201
 202         dio = container_of(bio, struct blkdev_dio, bio);
 203         atomic_set(&dio->ref, 1);
 204         /*
 205          * Grab an extra reference to ensure the dio structure which is embedded
 206          * into the first bio stays around.
 207          */
 208         bio_get(bio);
 209
 210         is_sync = is_sync_kiocb(iocb);
 211         if (is_sync) {
 212                 dio->flags = DIO_IS_SYNC;
 213                 dio->waiter = current;
 214         } else {
 215                 dio->flags = 0;
 216                 dio->iocb = iocb;
 217         }
 218
 219         dio->size = 0;
 220         if (is_read && iter_is_iovec(iter))
 221                 dio->flags |= DIO_SHOULD_DIRTY;
 222
 223         blk_start_plug(&plug);
 224
 225         for (;;) {
 226                 bio_set_dev(bio, bdev);
 227                 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
 228                 bio->bi_write_hint = iocb->ki_hint;
 229                 bio->bi_private = dio;
 230                 bio->bi_end_io = blkdev_bio_end_io;
 231                 bio->bi_ioprio = iocb->ki_ioprio;
 232
 233                 ret = bio_iov_iter_get_pages(bio, iter);
 234                 if (unlikely(ret)) {
 235                         bio->bi_status = BLK_STS_IOERR;
 236                         bio_endio(bio);
 237                         break;
 238                 }
 239
 240                 if (is_read) {
 241                         bio->bi_opf = REQ_OP_READ;
 242                         if (dio->flags & DIO_SHOULD_DIRTY)
 243                                 bio_set_pages_dirty(bio);
 244                 } else {
 245                         bio->bi_opf = dio_bio_write_op(iocb);
 246                         task_io_account_write(bio->bi_iter.bi_size);
 247                 }
 248                 if (iocb->ki_flags & IOCB_NOWAIT)
 249                         bio->bi_opf |= REQ_NOWAIT;
 250
 251                 dio->size += bio->bi_iter.bi_size;
 252                 pos += bio->bi_iter.bi_size;
 253
 254                 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS);
 255                 if (!nr_pages) {
 256                         submit_bio(bio);
 257                         break;
 258                 }
 259                 atomic_inc(&dio->ref);
 260                 submit_bio(bio);
 261                 bio = bio_alloc(GFP_KERNEL, nr_pages);
 262         }
 263
 264         blk_finish_plug(&plug);
 265
 266         if (!is_sync)
 267                 return -EIOCBQUEUED;
 268
 269         for (;;) {
 270                 set_current_state(TASK_UNINTERRUPTIBLE);
 271                 if (!READ_ONCE(dio->waiter))
 272                         break;
 273                 blk_io_schedule();
 274         }
 275         __set_current_state(TASK_RUNNING);
 276
 277         if (!ret)
 278                 ret = blk_status_to_errno(dio->bio.bi_status);
 279         if (likely(!ret))
 280                 ret = dio->size;
 281
 282         bio_put(&dio->bio);
 283         return ret;
 284 }
 285
 286 static void blkdev_bio_end_io_async(struct bio *bio)
 287 {
 288         struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio);
 289         struct kiocb *iocb = dio->iocb;
 290         ssize_t ret;
 291
 292         WRITE_ONCE(iocb->private, NULL);
 293
 294         if (likely(!bio->bi_status)) {
 295                 ret = dio->size;
 296                 iocb->ki_pos += ret;
 297         } else {
 298                 ret = blk_status_to_errno(bio->bi_status);
 299         }
 300
 301         iocb->ki_complete(iocb, ret);
 302
 303         if (dio->flags & DIO_SHOULD_DIRTY) {
 304                 bio_check_pages_dirty(bio);
 305         } else {
 306                 bio_release_pages(bio, false);
 307                 bio_put(bio);
 308         }
 309 }
 310
 311 static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
 312                                         struct iov_iter *iter,
 313                                         unsigned int nr_pages)
 314 {
 315         struct block_device *bdev = iocb->ki_filp->private_data;
 316         struct blkdev_dio *dio;
 317         struct bio *bio;
 318         loff_t pos = iocb->ki_pos;
 319         int ret = 0;
 320
 321         if ((pos | iov_iter_alignment(iter)) &
 322             (bdev_logical_block_size(bdev) - 1))
 323                 return -EINVAL;
 324
 325         bio = bio_alloc_kiocb(iocb, nr_pages, &blkdev_dio_pool);
 326         dio = container_of(bio, struct blkdev_dio, bio);
 327         dio->flags = 0;
 328         dio->iocb = iocb;
 329         bio_set_dev(bio, bdev);
 330         bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
 331         bio->bi_write_hint = iocb->ki_hint;
 332         bio->bi_end_io = blkdev_bio_end_io_async;
 333         bio->bi_ioprio = iocb->ki_ioprio;
 334
 335         if (iov_iter_is_bvec(iter)) {
 336                 /*
 337                  * Users don't rely on the iterator being in any particular
 338                  * state for async I/O returning -EIOCBQUEUED, hence we can
 339                  * avoid expensive iov_iter_advance(). Bypass
 340                  * bio_iov_iter_get_pages() and set the bvec directly.
 341                  */
 342                 bio_iov_bvec_set(bio, iter);
 343         } else {
 344                 ret = bio_iov_iter_get_pages(bio, iter);
 345                 if (unlikely(ret)) {
 346                         bio_put(bio);
 347                         return ret;
 348                 }
 349         }
 350         dio->size = bio->bi_iter.bi_size;
 351
 352         if (iov_iter_rw(iter) == READ) {
 353                 bio->bi_opf = REQ_OP_READ;
 354                 if (iter_is_iovec(iter)) {
 355                         dio->flags |= DIO_SHOULD_DIRTY;
 356                         bio_set_pages_dirty(bio);
 357                 }
 358         } else {
 359                 bio->bi_opf = dio_bio_write_op(iocb);
 360                 task_io_account_write(bio->bi_iter.bi_size);
 361         }
 362
 363         if (iocb->ki_flags & IOCB_HIPRI) {
 364                 bio->bi_opf |= REQ_POLLED | REQ_NOWAIT;
 365                 submit_bio(bio);
 366                 WRITE_ONCE(iocb->private, bio);
 367         } else {
 368                 if (iocb->ki_flags & IOCB_NOWAIT)
 369                         bio->bi_opf |= REQ_NOWAIT;
 370                 submit_bio(bio);
 371         }
 372         return -EIOCBQUEUED;
 373 }
 374
 375 static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 376 {
 377         unsigned int nr_pages;
 378
 379         if (!iov_iter_count(iter))
 380                 return 0;
 381
 382         nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
 383         if (likely(nr_pages <= BIO_MAX_VECS)) {
 384                 if (is_sync_kiocb(iocb))
 385                         return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
 386                 return __blkdev_direct_IO_async(iocb, iter, nr_pages);
 387         }
 388         return __blkdev_direct_IO(iocb, iter, bio_max_segs(nr_pages));
 389 }
 390
 391 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 392 {
 393         return block_write_full_page(page, blkdev_get_block, wbc);
 394 }
 395
 396 static int blkdev_readpage(struct file * file, struct page * page)
 397 {
 398         return block_read_full_page(page, blkdev_get_block);
 399 }
 400
 401 static void blkdev_readahead(struct readahead_control *rac)
 402 {
 403         mpage_readahead(rac, blkdev_get_block);
 404 }
 405
 406 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 407                 loff_t pos, unsigned len, unsigned flags, struct page **pagep,
 408                 void **fsdata)
 409 {
 410         return block_write_begin(mapping, pos, len, flags, pagep,
 411                                  blkdev_get_block);
 412 }
 413
 414 static int blkdev_write_end(struct file *file, struct address_space *mapping,
 415                 loff_t pos, unsigned len, unsigned copied, struct page *page,
 416                 void *fsdata)
 417 {
 418         int ret;
 419         ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 420
 421         unlock_page(page);
 422         put_page(page);
 423
 424         return ret;
 425 }
 426
 427 static int blkdev_writepages(struct address_space *mapping,
 428                              struct writeback_control *wbc)
 429 {
 430         return generic_writepages(mapping, wbc);
 431 }
 432
 433 const struct address_space_operations def_blk_aops = {
 434         .set_page_dirty = __set_page_dirty_buffers,
 435         .readpage       = blkdev_readpage,
 436         .readahead      = blkdev_readahead,
 437         .writepage      = blkdev_writepage,
 438         .write_begin    = blkdev_write_begin,
 439         .write_end      = blkdev_write_end,
 440         .writepages     = blkdev_writepages,
 441         .direct_IO      = blkdev_direct_IO,
 442         .migratepage    = buffer_migrate_page_norefs,
 443         .is_dirty_writeback = buffer_check_dirty_writeback,
 444 };
 445
 446 /*
 447  * for a block special file file_inode(file)->i_size is zero
 448  * so we compute the size by hand (just as in block_read/write above)
 449  */
 450 static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence)
 451 {
 452         struct inode *bd_inode = bdev_file_inode(file);
 453         loff_t retval;
 454
 455         inode_lock(bd_inode);
 456         retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
 457         inode_unlock(bd_inode);
 458         return retval;
 459 }
 460
 461 static int blkdev_fsync(struct file *filp, loff_t start, loff_t end,
 462                 int datasync)
 463 {
 464         struct block_device *bdev = filp->private_data;
 465         int error;
 466
 467         error = file_write_and_wait_range(filp, start, end);
 468         if (error)
 469                 return error;
 470
 471         /*
 472          * There is no need to serialise calls to blkdev_issue_flush with
 473          * i_mutex and doing so causes performance issues with concurrent
 474          * O_SYNC writers to a block device.
 475          */
 476         error = blkdev_issue_flush(bdev);
 477         if (error == -EOPNOTSUPP)
 478                 error = 0;
 479
 480         return error;
 481 }
 482
 483 static int blkdev_open(struct inode *inode, struct file *filp)
 484 {
 485         struct block_device *bdev;
 486
 487         /*
 488          * Preserve backwards compatibility and allow large file access
 489          * even if userspace doesn't ask for it explicitly. Some mkfs
 490          * binary needs it. We might want to drop this workaround
 491          * during an unstable branch.
 492          */
 493         filp->f_flags |= O_LARGEFILE;
 494         filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
 495
 496         if (filp->f_flags & O_NDELAY)
 497                 filp->f_mode |= FMODE_NDELAY;
 498         if (filp->f_flags & O_EXCL)
 499                 filp->f_mode |= FMODE_EXCL;
 500         if ((filp->f_flags & O_ACCMODE) == 3)
 501                 filp->f_mode |= FMODE_WRITE_IOCTL;
 502
 503         bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
 504         if (IS_ERR(bdev))
 505                 return PTR_ERR(bdev);
 506
 507         filp->private_data = bdev;
 508         filp->f_mapping = bdev->bd_inode->i_mapping;
 509         filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
 510         return 0;
 511 }
 512
 513 static int blkdev_close(struct inode *inode, struct file *filp)
 514 {
 515         struct block_device *bdev = filp->private_data;
 516
 517         blkdev_put(bdev, filp->f_mode);
 518         return 0;
 519 }
 520
 521 /*
 522  * Write data to the block device.  Only intended for the block device itself
 523  * and the raw driver which basically is a fake block device.
 524  *
 525  * Does not take i_mutex for the write and thus is not for general purpose
 526  * use.
 527  */
 528 static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
 529 {
 530         struct block_device *bdev = iocb->ki_filp->private_data;
 531         struct inode *bd_inode = bdev->bd_inode;
 532         loff_t size = bdev_nr_bytes(bdev);
 533         struct blk_plug plug;
 534         size_t shorted = 0;
 535         ssize_t ret;
 536
 537         if (bdev_read_only(bdev))
 538                 return -EPERM;
 539
 540         if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
 541                 return -ETXTBSY;
 542
 543         if (!iov_iter_count(from))
 544                 return 0;
 545
 546         if (iocb->ki_pos >= size)
 547                 return -ENOSPC;
 548
 549         if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
 550                 return -EOPNOTSUPP;
 551
 552         size -= iocb->ki_pos;
 553         if (iov_iter_count(from) > size) {
 554                 shorted = iov_iter_count(from) - size;
 555                 iov_iter_truncate(from, size);
 556         }
 557
 558         blk_start_plug(&plug);
 559         ret = __generic_file_write_iter(iocb, from);
 560         if (ret > 0)
 561                 ret = generic_write_sync(iocb, ret);
 562         iov_iter_reexpand(from, iov_iter_count(from) + shorted);
 563         blk_finish_plug(&plug);
 564         return ret;
 565 }
 566
 567 static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
 568 {
 569         struct block_device *bdev = iocb->ki_filp->private_data;
 570         loff_t size = bdev_nr_bytes(bdev);
 571         loff_t pos = iocb->ki_pos;
 572         size_t shorted = 0;
 573         ssize_t ret = 0;
 574         size_t count;
 575
 576         if (unlikely(pos + iov_iter_count(to) > size)) {
 577                 if (pos >= size)
 578                         return 0;
 579                 size -= pos;
 580                 shorted = iov_iter_count(to) - size;
 581                 iov_iter_truncate(to, size);
 582         }
 583
 584         count = iov_iter_count(to);
 585         if (!count)
 586                 goto reexpand; /* skip atime */
 587
 588         if (iocb->ki_flags & IOCB_DIRECT) {
 589                 struct address_space *mapping = iocb->ki_filp->f_mapping;
 590
 591                 if (iocb->ki_flags & IOCB_NOWAIT) {
 592                         if (filemap_range_needs_writeback(mapping, pos,
 593                                                           pos + count - 1)) {
 594                                 ret = -EAGAIN;
 595                                 goto reexpand;
 596                         }
 597                 } else {
 598                         ret = filemap_write_and_wait_range(mapping, pos,
 599                                                            pos + count - 1);
 600                         if (ret < 0)
 601                                 goto reexpand;
 602                 }
 603
 604                 file_accessed(iocb->ki_filp);
 605
 606                 ret = blkdev_direct_IO(iocb, to);
 607                 if (ret >= 0) {
 608                         iocb->ki_pos += ret;
 609                         count -= ret;
 610                 }
 611                 iov_iter_revert(to, count - iov_iter_count(to));
 612                 if (ret < 0 || !count)
 613                         goto reexpand;
 614         }
 615
 616         ret = filemap_read(iocb, to, ret);
 617
 618 reexpand:
 619         if (unlikely(shorted))
 620                 iov_iter_reexpand(to, iov_iter_count(to) + shorted);
 621         return ret;
 622 }
 623
 624 #define BLKDEV_FALLOC_FL_SUPPORTED                                      \
 625                 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |           \
 626                  FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
 627
 628 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
 629                              loff_t len)
 630 {
 631         struct inode *inode = bdev_file_inode(file);
 632         struct block_device *bdev = I_BDEV(inode);
 633         loff_t end = start + len - 1;
 634         loff_t isize;
 635         int error;
 636
 637         /* Fail if we don't recognize the flags. */
 638         if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
 639                 return -EOPNOTSUPP;
 640
 641         /* Don't go off the end of the device. */
 642         isize = bdev_nr_bytes(bdev);
 643         if (start >= isize)
 644                 return -EINVAL;
 645         if (end >= isize) {
 646                 if (mode & FALLOC_FL_KEEP_SIZE) {
 647                         len = isize - start;
 648                         end = start + len - 1;
 649                 } else
 650                         return -EINVAL;
 651         }
 652
 653         /*
 654          * Don't allow IO that isn't aligned to logical block size.
 655          */
 656         if ((start | len) & (bdev_logical_block_size(bdev) - 1))
 657                 return -EINVAL;
 658
 659         filemap_invalidate_lock(inode->i_mapping);
 660
 661         /* Invalidate the page cache, including dirty pages. */
 662         error = truncate_bdev_range(bdev, file->f_mode, start, end);
 663         if (error)
 664                 goto fail;
 665
 666         switch (mode) {
 667         case FALLOC_FL_ZERO_RANGE:
 668         case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
 669                 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
 670                                              len >> SECTOR_SHIFT, GFP_KERNEL,
 671                                              BLKDEV_ZERO_NOUNMAP);
 672                 break;
 673         case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
 674                 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
 675                                              len >> SECTOR_SHIFT, GFP_KERNEL,
 676                                              BLKDEV_ZERO_NOFALLBACK);
 677                 break;
 678         case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
 679                 error = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
 680                                              len >> SECTOR_SHIFT, GFP_KERNEL, 0);
 681                 break;
 682         default:
 683                 error = -EOPNOTSUPP;
 684         }
 685
 686  fail:
 687         filemap_invalidate_unlock(inode->i_mapping);
 688         return error;
 689 }
 690
 691 const struct file_operations def_blk_fops = {
 692         .open           = blkdev_open,
 693         .release        = blkdev_close,
 694         .llseek         = blkdev_llseek,
 695         .read_iter      = blkdev_read_iter,
 696         .write_iter     = blkdev_write_iter,
 697         .iopoll         = iocb_bio_iopoll,
 698         .mmap           = generic_file_mmap,
 699         .fsync          = blkdev_fsync,
 700         .unlocked_ioctl = blkdev_ioctl,
 701 #ifdef CONFIG_COMPAT
 702         .compat_ioctl   = compat_blkdev_ioctl,
 703 #endif
 704         .splice_read    = generic_file_splice_read,
 705         .splice_write   = iter_file_splice_write,
 706         .fallocate      = blkdev_fallocate,
 707 };
 708
 709 static __init int blkdev_init(void)
 710 {
 711         return bioset_init(&blkdev_dio_pool, 4,
 712                                 offsetof(struct blkdev_dio, bio),
 713                                 BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE);
 714 }
 715 module_init(blkdev_init);