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 >= i_size_read(inode))
  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         inode_unlock(inode);
 167
 168         if (ret > 0)
 169                 ret = generic_write_sync(iocb, ret);
 170
 171         return ret;
 172
 173 out:
 174         inode_unlock(inode);
 175         return ret;
 176 }
 177
 178 #ifdef CONFIG_FS_DAX
 179 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 180 {
 181         int result;
 182         handle_t *handle = NULL;
 183         struct inode *inode = file_inode(vma->vm_file);
 184         struct super_block *sb = inode->i_sb;
 185         bool write = vmf->flags & FAULT_FLAG_WRITE;
 186
 187         if (write) {
 188                 sb_start_pagefault(sb);
 189                 file_update_time(vma->vm_file);
 190                 down_read(&EXT4_I(inode)->i_mmap_sem);
 191                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 192                                                 EXT4_DATA_TRANS_BLOCKS(sb));
 193         } else
 194                 down_read(&EXT4_I(inode)->i_mmap_sem);
 195
 196         if (IS_ERR(handle))
 197                 result = VM_FAULT_SIGBUS;
 198         else
 199                 result = dax_fault(vma, vmf, ext4_dax_get_block);
 200
 201         if (write) {
 202                 if (!IS_ERR(handle))
 203                         ext4_journal_stop(handle);
 204                 up_read(&EXT4_I(inode)->i_mmap_sem);
 205                 sb_end_pagefault(sb);
 206         } else
 207                 up_read(&EXT4_I(inode)->i_mmap_sem);
 208
 209         return result;
 210 }
 211
 212 static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
 213                                                 pmd_t *pmd, unsigned int flags)
 214 {
 215         int result;
 216         handle_t *handle = NULL;
 217         struct inode *inode = file_inode(vma->vm_file);
 218         struct super_block *sb = inode->i_sb;
 219         bool write = flags & FAULT_FLAG_WRITE;
 220
 221         if (write) {
 222                 sb_start_pagefault(sb);
 223                 file_update_time(vma->vm_file);
 224                 down_read(&EXT4_I(inode)->i_mmap_sem);
 225                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 226                                 ext4_chunk_trans_blocks(inode,
 227                                                         PMD_SIZE / PAGE_SIZE));
 228         } else
 229                 down_read(&EXT4_I(inode)->i_mmap_sem);
 230
 231         if (IS_ERR(handle))
 232                 result = VM_FAULT_SIGBUS;
 233         else
 234                 result = dax_pmd_fault(vma, addr, pmd, flags,
 235                                          ext4_dax_get_block);
 236
 237         if (write) {
 238                 if (!IS_ERR(handle))
 239                         ext4_journal_stop(handle);
 240                 up_read(&EXT4_I(inode)->i_mmap_sem);
 241                 sb_end_pagefault(sb);
 242         } else
 243                 up_read(&EXT4_I(inode)->i_mmap_sem);
 244
 245         return result;
 246 }
 247
 248 /*
 249  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
 250  * handler we check for races agaist truncate. Note that since we cycle through
 251  * i_mmap_sem, we are sure that also any hole punching that began before we
 252  * were called is finished by now and so if it included part of the file we
 253  * are working on, our pte will get unmapped and the check for pte_same() in
 254  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 255  * desired.
 256  */
 257 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
 258                                 struct vm_fault *vmf)
 259 {
 260         struct inode *inode = file_inode(vma->vm_file);
 261         struct super_block *sb = inode->i_sb;
 262         loff_t size;
 263         int ret;
 264
 265         sb_start_pagefault(sb);
 266         file_update_time(vma->vm_file);
 267         down_read(&EXT4_I(inode)->i_mmap_sem);
 268         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 269         if (vmf->pgoff >= size)
 270                 ret = VM_FAULT_SIGBUS;
 271         else
 272                 ret = dax_pfn_mkwrite(vma, vmf);
 273         up_read(&EXT4_I(inode)->i_mmap_sem);
 274         sb_end_pagefault(sb);
 275
 276         return ret;
 277 }
 278
 279 static const struct vm_operations_struct ext4_dax_vm_ops = {
 280         .fault          = ext4_dax_fault,
 281         .pmd_fault      = ext4_dax_pmd_fault,
 282         .page_mkwrite   = ext4_dax_fault,
 283         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 284 };
 285 #else
 286 #define ext4_dax_vm_ops ext4_file_vm_ops
 287 #endif
 288
 289 static const struct vm_operations_struct ext4_file_vm_ops = {
 290         .fault          = ext4_filemap_fault,
 291         .map_pages      = filemap_map_pages,
 292         .page_mkwrite   = ext4_page_mkwrite,
 293 };
 294
 295 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 296 {
 297         struct inode *inode = file->f_mapping->host;
 298
 299         if (ext4_encrypted_inode(inode)) {
 300                 int err = fscrypt_get_encryption_info(inode);
 301                 if (err)
 302                         return 0;
 303                 if (!fscrypt_has_encryption_key(inode))
 304                         return -ENOKEY;
 305         }
 306         file_accessed(file);
 307         if (IS_DAX(file_inode(file))) {
 308                 vma->vm_ops = &ext4_dax_vm_ops;
 309                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 310         } else {
 311                 vma->vm_ops = &ext4_file_vm_ops;
 312         }
 313         return 0;
 314 }
 315
 316 static int ext4_file_open(struct inode * inode, struct file * filp)
 317 {
 318         struct super_block *sb = inode->i_sb;
 319         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 320         struct vfsmount *mnt = filp->f_path.mnt;
 321         struct dentry *dir;
 322         struct path path;
 323         char buf[64], *cp;
 324         int ret;
 325
 326         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 327                      !(sb->s_flags & MS_RDONLY))) {
 328                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 329                 /*
 330                  * Sample where the filesystem has been mounted and
 331                  * store it in the superblock for sysadmin convenience
 332                  * when trying to sort through large numbers of block
 333                  * devices or filesystem images.
 334                  */
 335                 memset(buf, 0, sizeof(buf));
 336                 path.mnt = mnt;
 337                 path.dentry = mnt->mnt_root;
 338                 cp = d_path(&path, buf, sizeof(buf));
 339                 if (!IS_ERR(cp)) {
 340                         handle_t *handle;
 341                         int err;
 342
 343                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 344                         if (IS_ERR(handle))
 345                                 return PTR_ERR(handle);
 346                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 347                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 348                         if (err) {
 349                                 ext4_journal_stop(handle);
 350                                 return err;
 351                         }
 352                         strlcpy(sbi->s_es->s_last_mounted, cp,
 353                                 sizeof(sbi->s_es->s_last_mounted));
 354                         ext4_handle_dirty_super(handle, sb);
 355                         ext4_journal_stop(handle);
 356                 }
 357         }
 358         if (ext4_encrypted_inode(inode)) {
 359                 ret = fscrypt_get_encryption_info(inode);
 360                 if (ret)
 361                         return -EACCES;
 362                 if (!fscrypt_has_encryption_key(inode))
 363                         return -ENOKEY;
 364         }
 365
 366         dir = dget_parent(file_dentry(filp));
 367         if (ext4_encrypted_inode(d_inode(dir)) &&
 368                         !fscrypt_has_permitted_context(d_inode(dir), inode)) {
 369                 ext4_warning(inode->i_sb,
 370                              "Inconsistent encryption contexts: %lu/%lu",
 371                              (unsigned long) d_inode(dir)->i_ino,
 372                              (unsigned long) inode->i_ino);
 373                 dput(dir);
 374                 return -EPERM;
 375         }
 376         dput(dir);
 377         /*
 378          * Set up the jbd2_inode if we are opening the inode for
 379          * writing and the journal is present
 380          */
 381         if (filp->f_mode & FMODE_WRITE) {
 382                 ret = ext4_inode_attach_jinode(inode);
 383                 if (ret < 0)
 384                         return ret;
 385         }
 386         return dquot_file_open(inode, filp);
 387 }
 388
 389 /*
 390  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 391  * file rather than ext4_ext_walk_space() because we can introduce
 392  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 393  * function.  When extent status tree has been fully implemented, it will
 394  * track all extent status for a file and we can directly use it to
 395  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 396  */
 397
 398 /*
 399  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 400  * lookup page cache to check whether or not there has some data between
 401  * [startoff, endoff] because, if this range contains an unwritten extent,
 402  * we determine this extent as a data or a hole according to whether the
 403  * page cache has data or not.
 404  */
 405 static int ext4_find_unwritten_pgoff(struct inode *inode,
 406                                      int whence,
 407                                      ext4_lblk_t end_blk,
 408                                      loff_t *offset)
 409 {
 410         struct pagevec pvec;
 411         unsigned int blkbits;
 412         pgoff_t index;
 413         pgoff_t end;
 414         loff_t endoff;
 415         loff_t startoff;
 416         loff_t lastoff;
 417         int found = 0;
 418
 419         blkbits = inode->i_sb->s_blocksize_bits;
 420         startoff = *offset;
 421         lastoff = startoff;
 422         endoff = (loff_t)end_blk << blkbits;
 423
 424         index = startoff >> PAGE_SHIFT;
 425         end = endoff >> PAGE_SHIFT;
 426
 427         pagevec_init(&pvec, 0);
 428         do {
 429                 int i, num;
 430                 unsigned long nr_pages;
 431
 432                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 433                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 434                                           (pgoff_t)num);
 435                 if (nr_pages == 0) {
 436                         if (whence == SEEK_DATA)
 437                                 break;
 438
 439                         BUG_ON(whence != SEEK_HOLE);
 440                         /*
 441                          * If this is the first time to go into the loop and
 442                          * offset is not beyond the end offset, it will be a
 443                          * hole at this offset
 444                          */
 445                         if (lastoff == startoff || lastoff < endoff)
 446                                 found = 1;
 447                         break;
 448                 }
 449
 450                 /*
 451                  * If this is the first time to go into the loop and
 452                  * offset is smaller than the first page offset, it will be a
 453                  * hole at this offset.
 454                  */
 455                 if (lastoff == startoff && whence == SEEK_HOLE &&
 456                     lastoff < page_offset(pvec.pages[0])) {
 457                         found = 1;
 458                         break;
 459                 }
 460
 461                 for (i = 0; i < nr_pages; i++) {
 462                         struct page *page = pvec.pages[i];
 463                         struct buffer_head *bh, *head;
 464
 465                         /*
 466                          * If the current offset is not beyond the end of given
 467                          * range, it will be a hole.
 468                          */
 469                         if (lastoff < endoff && whence == SEEK_HOLE &&
 470                             page->index > end) {
 471                                 found = 1;
 472                                 *offset = lastoff;
 473                                 goto out;
 474                         }
 475
 476                         lock_page(page);
 477
 478                         if (unlikely(page->mapping != inode->i_mapping)) {
 479                                 unlock_page(page);
 480                                 continue;
 481                         }
 482
 483                         if (!page_has_buffers(page)) {
 484                                 unlock_page(page);
 485                                 continue;
 486                         }
 487
 488                         if (page_has_buffers(page)) {
 489                                 lastoff = page_offset(page);
 490                                 bh = head = page_buffers(page);
 491                                 do {
 492                                         if (buffer_uptodate(bh) ||
 493                                             buffer_unwritten(bh)) {
 494                                                 if (whence == SEEK_DATA)
 495                                                         found = 1;
 496                                         } else {
 497                                                 if (whence == SEEK_HOLE)
 498                                                         found = 1;
 499                                         }
 500                                         if (found) {
 501                                                 *offset = max_t(loff_t,
 502                                                         startoff, lastoff);
 503                                                 unlock_page(page);
 504                                                 goto out;
 505                                         }
 506                                         lastoff += bh->b_size;
 507                                         bh = bh->b_this_page;
 508                                 } while (bh != head);
 509                         }
 510
 511                         lastoff = page_offset(page) + PAGE_SIZE;
 512                         unlock_page(page);
 513                 }
 514
 515                 /*
 516                  * The no. of pages is less than our desired, that would be a
 517                  * hole in there.
 518                  */
 519                 if (nr_pages < num && whence == SEEK_HOLE) {
 520                         found = 1;
 521                         *offset = lastoff;
 522                         break;
 523                 }
 524
 525                 index = pvec.pages[i - 1]->index + 1;
 526                 pagevec_release(&pvec);
 527         } while (index <= end);
 528
 529 out:
 530         pagevec_release(&pvec);
 531         return found;
 532 }
 533
 534 /*
 535  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 536  */
 537 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 538 {
 539         struct inode *inode = file->f_mapping->host;
 540         struct extent_status es;
 541         ext4_lblk_t start, last, end;
 542         loff_t dataoff, isize;
 543         int blkbits;
 544         int ret;
 545
 546         inode_lock(inode);
 547
 548         isize = i_size_read(inode);
 549         if (offset >= isize) {
 550                 inode_unlock(inode);
 551                 return -ENXIO;
 552         }
 553
 554         blkbits = inode->i_sb->s_blocksize_bits;
 555         start = offset >> blkbits;
 556         last = start;
 557         end = isize >> blkbits;
 558         dataoff = offset;
 559
 560         do {
 561                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 562                 if (ret <= 0) {
 563                         /* No extent found -> no data */
 564                         if (ret == 0)
 565                                 ret = -ENXIO;
 566                         inode_unlock(inode);
 567                         return ret;
 568                 }
 569
 570                 last = es.es_lblk;
 571                 if (last != start)
 572                         dataoff = (loff_t)last << blkbits;
 573                 if (!ext4_es_is_unwritten(&es))
 574                         break;
 575
 576                 /*
 577                  * If there is a unwritten extent at this offset,
 578                  * it will be as a data or a hole according to page
 579                  * cache that has data or not.
 580                  */
 581                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 582                                               es.es_lblk + es.es_len, &dataoff))
 583                         break;
 584                 last += es.es_len;
 585                 dataoff = (loff_t)last << blkbits;
 586                 cond_resched();
 587         } while (last <= end);
 588
 589         inode_unlock(inode);
 590
 591         if (dataoff > isize)
 592                 return -ENXIO;
 593
 594         return vfs_setpos(file, dataoff, maxsize);
 595 }
 596
 597 /*
 598  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 599  */
 600 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 601 {
 602         struct inode *inode = file->f_mapping->host;
 603         struct extent_status es;
 604         ext4_lblk_t start, last, end;
 605         loff_t holeoff, isize;
 606         int blkbits;
 607         int ret;
 608
 609         inode_lock(inode);
 610
 611         isize = i_size_read(inode);
 612         if (offset >= isize) {
 613                 inode_unlock(inode);
 614                 return -ENXIO;
 615         }
 616
 617         blkbits = inode->i_sb->s_blocksize_bits;
 618         start = offset >> blkbits;
 619         last = start;
 620         end = isize >> blkbits;
 621         holeoff = offset;
 622
 623         do {
 624                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 625                 if (ret < 0) {
 626                         inode_unlock(inode);
 627                         return ret;
 628                 }
 629                 /* Found a hole? */
 630                 if (ret == 0 || es.es_lblk > last) {
 631                         if (last != start)
 632                                 holeoff = (loff_t)last << blkbits;
 633                         break;
 634                 }
 635                 /*
 636                  * If there is a unwritten extent at this offset,
 637                  * it will be as a data or a hole according to page
 638                  * cache that has data or not.
 639                  */
 640                 if (ext4_es_is_unwritten(&es) &&
 641                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 642                                               last + es.es_len, &holeoff))
 643                         break;
 644
 645                 last += es.es_len;
 646                 holeoff = (loff_t)last << blkbits;
 647                 cond_resched();
 648         } while (last <= end);
 649
 650         inode_unlock(inode);
 651
 652         if (holeoff > isize)
 653                 holeoff = isize;
 654
 655         return vfs_setpos(file, holeoff, maxsize);
 656 }
 657
 658 /*
 659  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 660  * by calling generic_file_llseek_size() with the appropriate maxbytes
 661  * value for each.
 662  */
 663 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 664 {
 665         struct inode *inode = file->f_mapping->host;
 666         loff_t maxbytes;
 667
 668         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 669                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 670         else
 671                 maxbytes = inode->i_sb->s_maxbytes;
 672
 673         switch (whence) {
 674         case SEEK_SET:
 675         case SEEK_CUR:
 676         case SEEK_END:
 677                 return generic_file_llseek_size(file, offset, whence,
 678                                                 maxbytes, i_size_read(inode));
 679         case SEEK_DATA:
 680                 return ext4_seek_data(file, offset, maxbytes);
 681         case SEEK_HOLE:
 682                 return ext4_seek_hole(file, offset, maxbytes);
 683         }
 684
 685         return -EINVAL;
 686 }
 687
 688 const struct file_operations ext4_file_operations = {
 689         .llseek         = ext4_llseek,
 690         .read_iter      = generic_file_read_iter,
 691         .write_iter     = ext4_file_write_iter,
 692         .unlocked_ioctl = ext4_ioctl,
 693 #ifdef CONFIG_COMPAT
 694         .compat_ioctl   = ext4_compat_ioctl,
 695 #endif
 696         .mmap           = ext4_file_mmap,
 697         .open           = ext4_file_open,
 698         .release        = ext4_release_file,
 699         .fsync          = ext4_sync_file,
 700         .get_unmapped_area = thp_get_unmapped_area,
 701         .splice_read    = generic_file_splice_read,
 702         .splice_write   = iter_file_splice_write,
 703         .fallocate      = ext4_fallocate,
 704 };
 705
 706 const struct inode_operations ext4_file_inode_operations = {
 707         .setattr        = ext4_setattr,
 708         .getattr        = ext4_getattr,
 709         .listxattr      = ext4_listxattr,
 710         .get_acl        = ext4_get_acl,
 711         .set_acl        = ext4_set_acl,
 712         .fiemap         = ext4_fiemap,
 713 };
 714