fs/ext4/file.c

   1 /*
   2  *  linux/fs/ext4/file.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/file.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  ext4 fs regular file handling primitives
  16  *
  17  *  64-bit file support on 64-bit platforms by Jakub Jelinek
  18  *      (jj@sunsite.ms.mff.cuni.cz)
  19  */
  20
  21 #include <linux/time.h>
  22 #include <linux/fs.h>
  23 #include <linux/mount.h>
  24 #include <linux/path.h>
  25 #include <linux/dax.h>
  26 #include <linux/quotaops.h>
  27 #include <linux/pagevec.h>
  28 #include <linux/uio.h>
  29 #include "ext4.h"
  30 #include "ext4_jbd2.h"
  31 #include "xattr.h"
  32 #include "acl.h"
  33
  34 /*
  35  * Called when an inode is released. Note that this is different
  36  * from ext4_file_open: open gets called at every open, but release
  37  * gets called only when /all/ the files are closed.
  38  */
  39 static int ext4_release_file(struct inode *inode, struct file *filp)
  40 {
  41         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
  42                 ext4_alloc_da_blocks(inode);
  43                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
  44         }
  45         /* if we are the last writer on the inode, drop the block reservation */
  46         if ((filp->f_mode & FMODE_WRITE) &&
  47                         (atomic_read(&inode->i_writecount) == 1) &&
  48                         !EXT4_I(inode)->i_reserved_data_blocks)
  49         {
  50                 down_write(&EXT4_I(inode)->i_data_sem);
  51                 ext4_discard_preallocations(inode);
  52                 up_write(&EXT4_I(inode)->i_data_sem);
  53         }
  54         if (is_dx(inode) && filp->private_data)
  55                 ext4_htree_free_dir_info(filp->private_data);
  56
  57         return 0;
  58 }
  59
  60 static void ext4_unwritten_wait(struct inode *inode)
  61 {
  62         wait_queue_head_t *wq = ext4_ioend_wq(inode);
  63
  64         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
  65 }
  66
  67 /*
  68  * This tests whether the IO in question is block-aligned or not.
  69  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
  70  * are converted to written only after the IO is complete.  Until they are
  71  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
  72  * it needs to zero out portions of the start and/or end block.  If 2 AIO
  73  * threads are at work on the same unwritten block, they must be synchronized
  74  * or one thread will zero the other's data, causing corruption.
  75  */
  76 static int
  77 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
  78 {
  79         struct super_block *sb = inode->i_sb;
  80         int blockmask = sb->s_blocksize - 1;
  81
  82         if (pos >= ALIGN(i_size_read(inode), sb->s_blocksize))
  83                 return 0;
  84
  85         if ((pos | iov_iter_alignment(from)) & blockmask)
  86                 return 1;
  87
  88         return 0;
  89 }
  90
  91 static ssize_t
  92 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  93 {
  94         struct inode *inode = file_inode(iocb->ki_filp);
  95         int o_direct = iocb->ki_flags & IOCB_DIRECT;
  96         int unaligned_aio = 0;
  97         int overwrite = 0;
  98         ssize_t ret;
  99
 100         inode_lock(inode);
 101         ret = generic_write_checks(iocb, from);
 102         if (ret <= 0)
 103                 goto out;
 104
 105         /*
 106          * Unaligned direct AIO must be serialized among each other as zeroing
 107          * of partial blocks of two competing unaligned AIOs can result in data
 108          * corruption.
 109          */
 110         if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 111             !is_sync_kiocb(iocb) &&
 112             ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
 113                 unaligned_aio = 1;
 114                 ext4_unwritten_wait(inode);
 115         }
 116
 117         /*
 118          * If we have encountered a bitmap-format file, the size limit
 119          * is smaller than s_maxbytes, which is for extent-mapped files.
 120          */
 121         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 122                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 123
 124                 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
 125                         ret = -EFBIG;
 126                         goto out;
 127                 }
 128                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 129         }
 130
 131         iocb->private = &overwrite;
 132         if (o_direct) {
 133                 size_t length = iov_iter_count(from);
 134                 loff_t pos = iocb->ki_pos;
 135
 136                 /* check whether we do a DIO overwrite or not */
 137                 if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
 138                     pos + length <= i_size_read(inode)) {
 139                         struct ext4_map_blocks map;
 140                         unsigned int blkbits = inode->i_blkbits;
 141                         int err, len;
 142
 143                         map.m_lblk = pos >> blkbits;
 144                         map.m_len = EXT4_MAX_BLOCKS(length, pos, blkbits);
 145                         len = map.m_len;
 146
 147                         err = ext4_map_blocks(NULL, inode, &map, 0);
 148                         /*
 149                          * 'err==len' means that all of blocks has
 150                          * been preallocated no matter they are
 151                          * initialized or not.  For excluding
 152                          * unwritten extents, we need to check
 153                          * m_flags.  There are two conditions that
 154                          * indicate for initialized extents.  1) If we
 155                          * hit extent cache, EXT4_MAP_MAPPED flag is
 156                          * returned; 2) If we do a real lookup,
 157                          * non-flags are returned.  So we should check
 158                          * these two conditions.
 159                          */
 160                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 161                                 overwrite = 1;
 162                 }
 163         }
 164
 165         ret = __generic_file_write_iter(iocb, from);
 166         /*
 167          * Unaligned direct AIO must be the only IO in flight. Otherwise
 168          * overlapping aligned IO after unaligned might result in data
 169          * corruption.
 170          */
 171         if (ret == -EIOCBQUEUED && unaligned_aio)
 172                 ext4_unwritten_wait(inode);
 173         inode_unlock(inode);
 174
 175         if (ret > 0)
 176                 ret = generic_write_sync(iocb, ret);
 177
 178         return ret;
 179
 180 out:
 181         inode_unlock(inode);
 182         return ret;
 183 }
 184
 185 #ifdef CONFIG_FS_DAX
 186 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 187 {
 188         int result;
 189         handle_t *handle = NULL;
 190         struct inode *inode = file_inode(vma->vm_file);
 191         struct super_block *sb = inode->i_sb;
 192         bool write = vmf->flags & FAULT_FLAG_WRITE;
 193
 194         if (write) {
 195                 sb_start_pagefault(sb);
 196                 file_update_time(vma->vm_file);
 197                 down_read(&EXT4_I(inode)->i_mmap_sem);
 198                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 199                                                 EXT4_DATA_TRANS_BLOCKS(sb));
 200         } else
 201                 down_read(&EXT4_I(inode)->i_mmap_sem);
 202
 203         if (IS_ERR(handle))
 204                 result = VM_FAULT_SIGBUS;
 205         else
 206                 result = dax_fault(vma, vmf, ext4_dax_get_block);
 207
 208         if (write) {
 209                 if (!IS_ERR(handle))
 210                         ext4_journal_stop(handle);
 211                 up_read(&EXT4_I(inode)->i_mmap_sem);
 212                 sb_end_pagefault(sb);
 213         } else
 214                 up_read(&EXT4_I(inode)->i_mmap_sem);
 215
 216         return result;
 217 }
 218
 219 static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
 220                                                 pmd_t *pmd, unsigned int flags)
 221 {
 222         int result;
 223         handle_t *handle = NULL;
 224         struct inode *inode = file_inode(vma->vm_file);
 225         struct super_block *sb = inode->i_sb;
 226         bool write = flags & FAULT_FLAG_WRITE;
 227
 228         if (write) {
 229                 sb_start_pagefault(sb);
 230                 file_update_time(vma->vm_file);
 231                 down_read(&EXT4_I(inode)->i_mmap_sem);
 232                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 233                                 ext4_chunk_trans_blocks(inode,
 234                                                         PMD_SIZE / PAGE_SIZE));
 235         } else
 236                 down_read(&EXT4_I(inode)->i_mmap_sem);
 237
 238         if (IS_ERR(handle))
 239                 result = VM_FAULT_SIGBUS;
 240         else
 241                 result = dax_pmd_fault(vma, addr, pmd, flags,
 242                                          ext4_dax_get_block);
 243
 244         if (write) {
 245                 if (!IS_ERR(handle))
 246                         ext4_journal_stop(handle);
 247                 up_read(&EXT4_I(inode)->i_mmap_sem);
 248                 sb_end_pagefault(sb);
 249         } else
 250                 up_read(&EXT4_I(inode)->i_mmap_sem);
 251
 252         return result;
 253 }
 254
 255 /*
 256  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
 257  * handler we check for races agaist truncate. Note that since we cycle through
 258  * i_mmap_sem, we are sure that also any hole punching that began before we
 259  * were called is finished by now and so if it included part of the file we
 260  * are working on, our pte will get unmapped and the check for pte_same() in
 261  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 262  * desired.
 263  */
 264 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
 265                                 struct vm_fault *vmf)
 266 {
 267         struct inode *inode = file_inode(vma->vm_file);
 268         struct super_block *sb = inode->i_sb;
 269         loff_t size;
 270         int ret;
 271
 272         sb_start_pagefault(sb);
 273         file_update_time(vma->vm_file);
 274         down_read(&EXT4_I(inode)->i_mmap_sem);
 275         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 276         if (vmf->pgoff >= size)
 277                 ret = VM_FAULT_SIGBUS;
 278         else
 279                 ret = dax_pfn_mkwrite(vma, vmf);
 280         up_read(&EXT4_I(inode)->i_mmap_sem);
 281         sb_end_pagefault(sb);
 282
 283         return ret;
 284 }
 285
 286 static const struct vm_operations_struct ext4_dax_vm_ops = {
 287         .fault          = ext4_dax_fault,
 288         .pmd_fault      = ext4_dax_pmd_fault,
 289         .page_mkwrite   = ext4_dax_fault,
 290         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 291 };
 292 #else
 293 #define ext4_dax_vm_ops ext4_file_vm_ops
 294 #endif
 295
 296 static const struct vm_operations_struct ext4_file_vm_ops = {
 297         .fault          = ext4_filemap_fault,
 298         .map_pages      = filemap_map_pages,
 299         .page_mkwrite   = ext4_page_mkwrite,
 300 };
 301
 302 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 303 {
 304         struct inode *inode = file->f_mapping->host;
 305
 306         if (ext4_encrypted_inode(inode)) {
 307                 int err = fscrypt_get_encryption_info(inode);
 308                 if (err)
 309                         return 0;
 310                 if (!fscrypt_has_encryption_key(inode))
 311                         return -ENOKEY;
 312         }
 313         file_accessed(file);
 314         if (IS_DAX(file_inode(file))) {
 315                 vma->vm_ops = &ext4_dax_vm_ops;
 316                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 317         } else {
 318                 vma->vm_ops = &ext4_file_vm_ops;
 319         }
 320         return 0;
 321 }
 322
 323 static int ext4_file_open(struct inode * inode, struct file * filp)
 324 {
 325         struct super_block *sb = inode->i_sb;
 326         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 327         struct vfsmount *mnt = filp->f_path.mnt;
 328         struct dentry *dir;
 329         struct path path;
 330         char buf[64], *cp;
 331         int ret;
 332
 333         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 334                      !(sb->s_flags & MS_RDONLY))) {
 335                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 336                 /*
 337                  * Sample where the filesystem has been mounted and
 338                  * store it in the superblock for sysadmin convenience
 339                  * when trying to sort through large numbers of block
 340                  * devices or filesystem images.
 341                  */
 342                 memset(buf, 0, sizeof(buf));
 343                 path.mnt = mnt;
 344                 path.dentry = mnt->mnt_root;
 345                 cp = d_path(&path, buf, sizeof(buf));
 346                 if (!IS_ERR(cp)) {
 347                         handle_t *handle;
 348                         int err;
 349
 350                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 351                         if (IS_ERR(handle))
 352                                 return PTR_ERR(handle);
 353                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 354                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 355                         if (err) {
 356                                 ext4_journal_stop(handle);
 357                                 return err;
 358                         }
 359                         strlcpy(sbi->s_es->s_last_mounted, cp,
 360                                 sizeof(sbi->s_es->s_last_mounted));
 361                         ext4_handle_dirty_super(handle, sb);
 362                         ext4_journal_stop(handle);
 363                 }
 364         }
 365         if (ext4_encrypted_inode(inode)) {
 366                 ret = fscrypt_get_encryption_info(inode);
 367                 if (ret)
 368                         return -EACCES;
 369                 if (!fscrypt_has_encryption_key(inode))
 370                         return -ENOKEY;
 371         }
 372
 373         dir = dget_parent(file_dentry(filp));
 374         if (ext4_encrypted_inode(d_inode(dir)) &&
 375                         !fscrypt_has_permitted_context(d_inode(dir), inode)) {
 376                 ext4_warning(inode->i_sb,
 377                              "Inconsistent encryption contexts: %lu/%lu",
 378                              (unsigned long) d_inode(dir)->i_ino,
 379                              (unsigned long) inode->i_ino);
 380                 dput(dir);
 381                 return -EPERM;
 382         }
 383         dput(dir);
 384         /*
 385          * Set up the jbd2_inode if we are opening the inode for
 386          * writing and the journal is present
 387          */
 388         if (filp->f_mode & FMODE_WRITE) {
 389                 ret = ext4_inode_attach_jinode(inode);
 390                 if (ret < 0)
 391                         return ret;
 392         }
 393         return dquot_file_open(inode, filp);
 394 }
 395
 396 /*
 397  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 398  * file rather than ext4_ext_walk_space() because we can introduce
 399  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 400  * function.  When extent status tree has been fully implemented, it will
 401  * track all extent status for a file and we can directly use it to
 402  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 403  */
 404
 405 /*
 406  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 407  * lookup page cache to check whether or not there has some data between
 408  * [startoff, endoff] because, if this range contains an unwritten extent,
 409  * we determine this extent as a data or a hole according to whether the
 410  * page cache has data or not.
 411  */
 412 static int ext4_find_unwritten_pgoff(struct inode *inode,
 413                                      int whence,
 414                                      ext4_lblk_t end_blk,
 415                                      loff_t *offset)
 416 {
 417         struct pagevec pvec;
 418         unsigned int blkbits;
 419         pgoff_t index;
 420         pgoff_t end;
 421         loff_t endoff;
 422         loff_t startoff;
 423         loff_t lastoff;
 424         int found = 0;
 425
 426         blkbits = inode->i_sb->s_blocksize_bits;
 427         startoff = *offset;
 428         lastoff = startoff;
 429         endoff = (loff_t)end_blk << blkbits;
 430
 431         index = startoff >> PAGE_SHIFT;
 432         end = endoff >> PAGE_SHIFT;
 433
 434         pagevec_init(&pvec, 0);
 435         do {
 436                 int i, num;
 437                 unsigned long nr_pages;
 438
 439                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE - 1) + 1;
 440                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 441                                           (pgoff_t)num);
 442                 if (nr_pages == 0)
 443                         break;
 444
 445                 for (i = 0; i < nr_pages; i++) {
 446                         struct page *page = pvec.pages[i];
 447                         struct buffer_head *bh, *head;
 448
 449                         /*
 450                          * If current offset is smaller than the page offset,
 451                          * there is a hole at this offset.
 452                          */
 453                         if (whence == SEEK_HOLE && lastoff < endoff &&
 454                             lastoff < page_offset(pvec.pages[i])) {
 455                                 found = 1;
 456                                 *offset = lastoff;
 457                                 goto out;
 458                         }
 459
 460                         if (page->index > end)
 461                                 goto out;
 462
 463                         lock_page(page);
 464
 465                         if (unlikely(page->mapping != inode->i_mapping)) {
 466                                 unlock_page(page);
 467                                 continue;
 468                         }
 469
 470                         if (!page_has_buffers(page)) {
 471                                 unlock_page(page);
 472                                 continue;
 473                         }
 474
 475                         if (page_has_buffers(page)) {
 476                                 lastoff = page_offset(page);
 477                                 bh = head = page_buffers(page);
 478                                 do {
 479                                         if (lastoff + bh->b_size <= startoff)
 480                                                 goto next;
 481                                         if (buffer_uptodate(bh) ||
 482                                             buffer_unwritten(bh)) {
 483                                                 if (whence == SEEK_DATA)
 484                                                         found = 1;
 485                                         } else {
 486                                                 if (whence == SEEK_HOLE)
 487                                                         found = 1;
 488                                         }
 489                                         if (found) {
 490                                                 *offset = max_t(loff_t,
 491                                                         startoff, lastoff);
 492                                                 unlock_page(page);
 493                                                 goto out;
 494                                         }
 495 next:
 496                                         lastoff += bh->b_size;
 497                                         bh = bh->b_this_page;
 498                                 } while (bh != head);
 499                         }
 500
 501                         lastoff = page_offset(page) + PAGE_SIZE;
 502                         unlock_page(page);
 503                 }
 504
 505                 /* The no. of pages is less than our desired, we are done. */
 506                 if (nr_pages < num)
 507                         break;
 508
 509                 index = pvec.pages[i - 1]->index + 1;
 510                 pagevec_release(&pvec);
 511         } while (index <= end);
 512
 513         if (whence == SEEK_HOLE && lastoff < endoff) {
 514                 found = 1;
 515                 *offset = lastoff;
 516         }
 517 out:
 518         pagevec_release(&pvec);
 519         return found;
 520 }
 521
 522 /*
 523  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 524  */
 525 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 526 {
 527         struct inode *inode = file->f_mapping->host;
 528         struct extent_status es;
 529         ext4_lblk_t start, last, end;
 530         loff_t dataoff, isize;
 531         int blkbits;
 532         int ret;
 533
 534         inode_lock(inode);
 535
 536         isize = i_size_read(inode);
 537         if (offset < 0 || offset >= isize) {
 538                 inode_unlock(inode);
 539                 return -ENXIO;
 540         }
 541
 542         blkbits = inode->i_sb->s_blocksize_bits;
 543         start = offset >> blkbits;
 544         last = start;
 545         end = isize >> blkbits;
 546         dataoff = offset;
 547
 548         do {
 549                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 550                 if (ret <= 0) {
 551                         /* No extent found -> no data */
 552                         if (ret == 0)
 553                                 ret = -ENXIO;
 554                         inode_unlock(inode);
 555                         return ret;
 556                 }
 557
 558                 last = es.es_lblk;
 559                 if (last != start)
 560                         dataoff = (loff_t)last << blkbits;
 561                 if (!ext4_es_is_unwritten(&es))
 562                         break;
 563
 564                 /*
 565                  * If there is a unwritten extent at this offset,
 566                  * it will be as a data or a hole according to page
 567                  * cache that has data or not.
 568                  */
 569                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 570                                               es.es_lblk + es.es_len, &dataoff))
 571                         break;
 572                 last += es.es_len;
 573                 dataoff = (loff_t)last << blkbits;
 574                 cond_resched();
 575         } while (last <= end);
 576
 577         inode_unlock(inode);
 578
 579         if (dataoff > isize)
 580                 return -ENXIO;
 581
 582         return vfs_setpos(file, dataoff, maxsize);
 583 }
 584
 585 /*
 586  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 587  */
 588 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 589 {
 590         struct inode *inode = file->f_mapping->host;
 591         struct extent_status es;
 592         ext4_lblk_t start, last, end;
 593         loff_t holeoff, isize;
 594         int blkbits;
 595         int ret;
 596
 597         inode_lock(inode);
 598
 599         isize = i_size_read(inode);
 600         if (offset < 0 || offset >= isize) {
 601                 inode_unlock(inode);
 602                 return -ENXIO;
 603         }
 604
 605         blkbits = inode->i_sb->s_blocksize_bits;
 606         start = offset >> blkbits;
 607         last = start;
 608         end = isize >> blkbits;
 609         holeoff = offset;
 610
 611         do {
 612                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 613                 if (ret < 0) {
 614                         inode_unlock(inode);
 615                         return ret;
 616                 }
 617                 /* Found a hole? */
 618                 if (ret == 0 || es.es_lblk > last) {
 619                         if (last != start)
 620                                 holeoff = (loff_t)last << blkbits;
 621                         break;
 622                 }
 623                 /*
 624                  * If there is a unwritten extent at this offset,
 625                  * it will be as a data or a hole according to page
 626                  * cache that has data or not.
 627                  */
 628                 if (ext4_es_is_unwritten(&es) &&
 629                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 630                                               last + es.es_len, &holeoff))
 631                         break;
 632
 633                 last += es.es_len;
 634                 holeoff = (loff_t)last << blkbits;
 635                 cond_resched();
 636         } while (last <= end);
 637
 638         inode_unlock(inode);
 639
 640         if (holeoff > isize)
 641                 holeoff = isize;
 642
 643         return vfs_setpos(file, holeoff, maxsize);
 644 }
 645
 646 /*
 647  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 648  * by calling generic_file_llseek_size() with the appropriate maxbytes
 649  * value for each.
 650  */
 651 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 652 {
 653         struct inode *inode = file->f_mapping->host;
 654         loff_t maxbytes;
 655
 656         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 657                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 658         else
 659                 maxbytes = inode->i_sb->s_maxbytes;
 660
 661         switch (whence) {
 662         case SEEK_SET:
 663         case SEEK_CUR:
 664         case SEEK_END:
 665                 return generic_file_llseek_size(file, offset, whence,
 666                                                 maxbytes, i_size_read(inode));
 667         case SEEK_DATA:
 668                 return ext4_seek_data(file, offset, maxbytes);
 669         case SEEK_HOLE:
 670                 return ext4_seek_hole(file, offset, maxbytes);
 671         }
 672
 673         return -EINVAL;
 674 }
 675
 676 const struct file_operations ext4_file_operations = {
 677         .llseek         = ext4_llseek,
 678         .read_iter      = generic_file_read_iter,
 679         .write_iter     = ext4_file_write_iter,
 680         .unlocked_ioctl = ext4_ioctl,
 681 #ifdef CONFIG_COMPAT
 682         .compat_ioctl   = ext4_compat_ioctl,
 683 #endif
 684         .mmap           = ext4_file_mmap,
 685         .open           = ext4_file_open,
 686         .release        = ext4_release_file,
 687         .fsync          = ext4_sync_file,
 688         .get_unmapped_area = thp_get_unmapped_area,
 689         .splice_read    = generic_file_splice_read,
 690         .splice_write   = iter_file_splice_write,
 691         .fallocate      = ext4_fallocate,
 692 };
 693
 694 const struct inode_operations ext4_file_inode_operations = {
 695         .setattr        = ext4_setattr,
 696         .getattr        = ext4_getattr,
 697         .listxattr      = ext4_listxattr,
 698         .get_acl        = ext4_get_acl,
 699         .set_acl        = ext4_set_acl,
 700         .fiemap         = ext4_fiemap,
 701 };
 702