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 file *file = iocb->ki_filp;
  95         struct inode *inode = file_inode(iocb->ki_filp);
  96         struct blk_plug plug;
  97         int o_direct = iocb->ki_flags & IOCB_DIRECT;
  98         int unaligned_aio = 0;
  99         int overwrite = 0;
 100         ssize_t ret;
 101
 102         inode_lock(inode);
 103         ret = generic_write_checks(iocb, from);
 104         if (ret <= 0)
 105                 goto out;
 106
 107         /*
 108          * Unaligned direct AIO must be serialized among each other as zeroing
 109          * of partial blocks of two competing unaligned AIOs can result in data
 110          * corruption.
 111          */
 112         if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 113             !is_sync_kiocb(iocb) &&
 114             ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
 115                 unaligned_aio = 1;
 116                 ext4_unwritten_wait(inode);
 117         }
 118
 119         /*
 120          * If we have encountered a bitmap-format file, the size limit
 121          * is smaller than s_maxbytes, which is for extent-mapped files.
 122          */
 123         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 124                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 125
 126                 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
 127                         ret = -EFBIG;
 128                         goto out;
 129                 }
 130                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 131         }
 132
 133         iocb->private = &overwrite;
 134         if (o_direct) {
 135                 size_t length = iov_iter_count(from);
 136                 loff_t pos = iocb->ki_pos;
 137                 blk_start_plug(&plug);
 138
 139                 /* check whether we do a DIO overwrite or not */
 140                 if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
 141                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 142                         struct ext4_map_blocks map;
 143                         unsigned int blkbits = inode->i_blkbits;
 144                         int err, len;
 145
 146                         map.m_lblk = pos >> blkbits;
 147                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 148                                 - map.m_lblk;
 149                         len = map.m_len;
 150
 151                         err = ext4_map_blocks(NULL, inode, &map, 0);
 152                         /*
 153                          * 'err==len' means that all of blocks has
 154                          * been preallocated no matter they are
 155                          * initialized or not.  For excluding
 156                          * unwritten extents, we need to check
 157                          * m_flags.  There are two conditions that
 158                          * indicate for initialized extents.  1) If we
 159                          * hit extent cache, EXT4_MAP_MAPPED flag is
 160                          * returned; 2) If we do a real lookup,
 161                          * non-flags are returned.  So we should check
 162                          * these two conditions.
 163                          */
 164                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 165                                 overwrite = 1;
 166                 }
 167         }
 168
 169         ret = __generic_file_write_iter(iocb, from);
 170         inode_unlock(inode);
 171
 172         if (ret > 0) {
 173                 ssize_t err;
 174
 175                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
 176                 if (err < 0)
 177                         ret = err;
 178         }
 179         if (o_direct)
 180                 blk_finish_plug(&plug);
 181
 182         return ret;
 183
 184 out:
 185         inode_unlock(inode);
 186         return ret;
 187 }
 188
 189 #ifdef CONFIG_FS_DAX
 190 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 191 {
 192         int result;
 193         handle_t *handle = NULL;
 194         struct inode *inode = file_inode(vma->vm_file);
 195         struct super_block *sb = inode->i_sb;
 196         bool write = vmf->flags & FAULT_FLAG_WRITE;
 197
 198         if (write) {
 199                 sb_start_pagefault(sb);
 200                 file_update_time(vma->vm_file);
 201                 down_read(&EXT4_I(inode)->i_mmap_sem);
 202                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 203                                                 EXT4_DATA_TRANS_BLOCKS(sb));
 204         } else
 205                 down_read(&EXT4_I(inode)->i_mmap_sem);
 206
 207         if (IS_ERR(handle))
 208                 result = VM_FAULT_SIGBUS;
 209         else
 210                 result = __dax_fault(vma, vmf, ext4_dax_mmap_get_block, NULL);
 211
 212         if (write) {
 213                 if (!IS_ERR(handle))
 214                         ext4_journal_stop(handle);
 215                 up_read(&EXT4_I(inode)->i_mmap_sem);
 216                 sb_end_pagefault(sb);
 217         } else
 218                 up_read(&EXT4_I(inode)->i_mmap_sem);
 219
 220         return result;
 221 }
 222
 223 static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
 224                                                 pmd_t *pmd, unsigned int flags)
 225 {
 226         int result;
 227         handle_t *handle = NULL;
 228         struct inode *inode = file_inode(vma->vm_file);
 229         struct super_block *sb = inode->i_sb;
 230         bool write = flags & FAULT_FLAG_WRITE;
 231
 232         if (write) {
 233                 sb_start_pagefault(sb);
 234                 file_update_time(vma->vm_file);
 235                 down_read(&EXT4_I(inode)->i_mmap_sem);
 236                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 237                                 ext4_chunk_trans_blocks(inode,
 238                                                         PMD_SIZE / PAGE_SIZE));
 239         } else
 240                 down_read(&EXT4_I(inode)->i_mmap_sem);
 241
 242         if (IS_ERR(handle))
 243                 result = VM_FAULT_SIGBUS;
 244         else
 245                 result = __dax_pmd_fault(vma, addr, pmd, flags,
 246                                 ext4_dax_mmap_get_block, NULL);
 247
 248         if (write) {
 249                 if (!IS_ERR(handle))
 250                         ext4_journal_stop(handle);
 251                 up_read(&EXT4_I(inode)->i_mmap_sem);
 252                 sb_end_pagefault(sb);
 253         } else
 254                 up_read(&EXT4_I(inode)->i_mmap_sem);
 255
 256         return result;
 257 }
 258
 259 /*
 260  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
 261  * handler we check for races agaist truncate. Note that since we cycle through
 262  * i_mmap_sem, we are sure that also any hole punching that began before we
 263  * were called is finished by now and so if it included part of the file we
 264  * are working on, our pte will get unmapped and the check for pte_same() in
 265  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 266  * desired.
 267  */
 268 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
 269                                 struct vm_fault *vmf)
 270 {
 271         struct inode *inode = file_inode(vma->vm_file);
 272         struct super_block *sb = inode->i_sb;
 273         loff_t size;
 274         int ret;
 275
 276         sb_start_pagefault(sb);
 277         file_update_time(vma->vm_file);
 278         down_read(&EXT4_I(inode)->i_mmap_sem);
 279         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 280         if (vmf->pgoff >= size)
 281                 ret = VM_FAULT_SIGBUS;
 282         else
 283                 ret = dax_pfn_mkwrite(vma, vmf);
 284         up_read(&EXT4_I(inode)->i_mmap_sem);
 285         sb_end_pagefault(sb);
 286
 287         return ret;
 288 }
 289
 290 static const struct vm_operations_struct ext4_dax_vm_ops = {
 291         .fault          = ext4_dax_fault,
 292         .pmd_fault      = ext4_dax_pmd_fault,
 293         .page_mkwrite   = ext4_dax_fault,
 294         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 295 };
 296 #else
 297 #define ext4_dax_vm_ops ext4_file_vm_ops
 298 #endif
 299
 300 static const struct vm_operations_struct ext4_file_vm_ops = {
 301         .fault          = ext4_filemap_fault,
 302         .map_pages      = filemap_map_pages,
 303         .page_mkwrite   = ext4_page_mkwrite,
 304 };
 305
 306 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 307 {
 308         struct inode *inode = file->f_mapping->host;
 309
 310         if (ext4_encrypted_inode(inode)) {
 311                 int err = ext4_get_encryption_info(inode);
 312                 if (err)
 313                         return 0;
 314                 if (ext4_encryption_info(inode) == NULL)
 315                         return -ENOKEY;
 316         }
 317         file_accessed(file);
 318         if (IS_DAX(file_inode(file))) {
 319                 vma->vm_ops = &ext4_dax_vm_ops;
 320                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 321         } else {
 322                 vma->vm_ops = &ext4_file_vm_ops;
 323         }
 324         return 0;
 325 }
 326
 327 static int ext4_file_open(struct inode * inode, struct file * filp)
 328 {
 329         struct super_block *sb = inode->i_sb;
 330         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 331         struct vfsmount *mnt = filp->f_path.mnt;
 332         struct inode *dir = filp->f_path.dentry->d_parent->d_inode;
 333         struct path path;
 334         char buf[64], *cp;
 335         int ret;
 336
 337         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 338                      !(sb->s_flags & MS_RDONLY))) {
 339                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 340                 /*
 341                  * Sample where the filesystem has been mounted and
 342                  * store it in the superblock for sysadmin convenience
 343                  * when trying to sort through large numbers of block
 344                  * devices or filesystem images.
 345                  */
 346                 memset(buf, 0, sizeof(buf));
 347                 path.mnt = mnt;
 348                 path.dentry = mnt->mnt_root;
 349                 cp = d_path(&path, buf, sizeof(buf));
 350                 if (!IS_ERR(cp)) {
 351                         handle_t *handle;
 352                         int err;
 353
 354                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 355                         if (IS_ERR(handle))
 356                                 return PTR_ERR(handle);
 357                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 358                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 359                         if (err) {
 360                                 ext4_journal_stop(handle);
 361                                 return err;
 362                         }
 363                         strlcpy(sbi->s_es->s_last_mounted, cp,
 364                                 sizeof(sbi->s_es->s_last_mounted));
 365                         ext4_handle_dirty_super(handle, sb);
 366                         ext4_journal_stop(handle);
 367                 }
 368         }
 369         if (ext4_encrypted_inode(inode)) {
 370                 ret = ext4_get_encryption_info(inode);
 371                 if (ret)
 372                         return -EACCES;
 373                 if (ext4_encryption_info(inode) == NULL)
 374                         return -ENOKEY;
 375         }
 376         if (ext4_encrypted_inode(dir) &&
 377             !ext4_is_child_context_consistent_with_parent(dir, inode)) {
 378                 ext4_warning(inode->i_sb,
 379                              "Inconsistent encryption contexts: %lu/%lu\n",
 380                              (unsigned long) dir->i_ino,
 381                              (unsigned long) inode->i_ino);
 382                 return -EPERM;
 383         }
 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_CACHE_SHIFT;
 432         end = endoff >> PAGE_CACHE_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);
 440                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 441                                           (pgoff_t)num);
 442                 if (nr_pages == 0) {
 443                         if (whence == SEEK_DATA)
 444                                 break;
 445
 446                         BUG_ON(whence != SEEK_HOLE);
 447                         /*
 448                          * If this is the first time to go into the loop and
 449                          * offset is not beyond the end offset, it will be a
 450                          * hole at this offset
 451                          */
 452                         if (lastoff == startoff || lastoff < endoff)
 453                                 found = 1;
 454                         break;
 455                 }
 456
 457                 /*
 458                  * If this is the first time to go into the loop and
 459                  * offset is smaller than the first page offset, it will be a
 460                  * hole at this offset.
 461                  */
 462                 if (lastoff == startoff && whence == SEEK_HOLE &&
 463                     lastoff < page_offset(pvec.pages[0])) {
 464                         found = 1;
 465                         break;
 466                 }
 467
 468                 for (i = 0; i < nr_pages; i++) {
 469                         struct page *page = pvec.pages[i];
 470                         struct buffer_head *bh, *head;
 471
 472                         /*
 473                          * If the current offset is not beyond the end of given
 474                          * range, it will be a hole.
 475                          */
 476                         if (lastoff < endoff && whence == SEEK_HOLE &&
 477                             page->index > end) {
 478                                 found = 1;
 479                                 *offset = lastoff;
 480                                 goto out;
 481                         }
 482
 483                         lock_page(page);
 484
 485                         if (unlikely(page->mapping != inode->i_mapping)) {
 486                                 unlock_page(page);
 487                                 continue;
 488                         }
 489
 490                         if (!page_has_buffers(page)) {
 491                                 unlock_page(page);
 492                                 continue;
 493                         }
 494
 495                         if (page_has_buffers(page)) {
 496                                 lastoff = page_offset(page);
 497                                 bh = head = page_buffers(page);
 498                                 do {
 499                                         if (buffer_uptodate(bh) ||
 500                                             buffer_unwritten(bh)) {
 501                                                 if (whence == SEEK_DATA)
 502                                                         found = 1;
 503                                         } else {
 504                                                 if (whence == SEEK_HOLE)
 505                                                         found = 1;
 506                                         }
 507                                         if (found) {
 508                                                 *offset = max_t(loff_t,
 509                                                         startoff, lastoff);
 510                                                 unlock_page(page);
 511                                                 goto out;
 512                                         }
 513                                         lastoff += bh->b_size;
 514                                         bh = bh->b_this_page;
 515                                 } while (bh != head);
 516                         }
 517
 518                         lastoff = page_offset(page) + PAGE_SIZE;
 519                         unlock_page(page);
 520                 }
 521
 522                 /*
 523                  * The no. of pages is less than our desired, that would be a
 524                  * hole in there.
 525                  */
 526                 if (nr_pages < num && whence == SEEK_HOLE) {
 527                         found = 1;
 528                         *offset = lastoff;
 529                         break;
 530                 }
 531
 532                 index = pvec.pages[i - 1]->index + 1;
 533                 pagevec_release(&pvec);
 534         } while (index <= end);
 535
 536 out:
 537         pagevec_release(&pvec);
 538         return found;
 539 }
 540
 541 /*
 542  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 543  */
 544 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 545 {
 546         struct inode *inode = file->f_mapping->host;
 547         struct extent_status es;
 548         ext4_lblk_t start, last, end;
 549         loff_t dataoff, isize;
 550         int blkbits;
 551         int ret;
 552
 553         inode_lock(inode);
 554
 555         isize = i_size_read(inode);
 556         if (offset >= isize) {
 557                 inode_unlock(inode);
 558                 return -ENXIO;
 559         }
 560
 561         blkbits = inode->i_sb->s_blocksize_bits;
 562         start = offset >> blkbits;
 563         last = start;
 564         end = isize >> blkbits;
 565         dataoff = offset;
 566
 567         do {
 568                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 569                 if (ret <= 0) {
 570                         /* No extent found -> no data */
 571                         if (ret == 0)
 572                                 ret = -ENXIO;
 573                         inode_unlock(inode);
 574                         return ret;
 575                 }
 576
 577                 last = es.es_lblk;
 578                 if (last != start)
 579                         dataoff = (loff_t)last << blkbits;
 580                 if (!ext4_es_is_unwritten(&es))
 581                         break;
 582
 583                 /*
 584                  * If there is a unwritten extent at this offset,
 585                  * it will be as a data or a hole according to page
 586                  * cache that has data or not.
 587                  */
 588                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 589                                               es.es_lblk + es.es_len, &dataoff))
 590                         break;
 591                 last += es.es_len;
 592                 dataoff = (loff_t)last << blkbits;
 593                 cond_resched();
 594         } while (last <= end);
 595
 596         inode_unlock(inode);
 597
 598         if (dataoff > isize)
 599                 return -ENXIO;
 600
 601         return vfs_setpos(file, dataoff, maxsize);
 602 }
 603
 604 /*
 605  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 606  */
 607 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 608 {
 609         struct inode *inode = file->f_mapping->host;
 610         struct extent_status es;
 611         ext4_lblk_t start, last, end;
 612         loff_t holeoff, isize;
 613         int blkbits;
 614         int ret;
 615
 616         inode_lock(inode);
 617
 618         isize = i_size_read(inode);
 619         if (offset >= isize) {
 620                 inode_unlock(inode);
 621                 return -ENXIO;
 622         }
 623
 624         blkbits = inode->i_sb->s_blocksize_bits;
 625         start = offset >> blkbits;
 626         last = start;
 627         end = isize >> blkbits;
 628         holeoff = offset;
 629
 630         do {
 631                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 632                 if (ret < 0) {
 633                         inode_unlock(inode);
 634                         return ret;
 635                 }
 636                 /* Found a hole? */
 637                 if (ret == 0 || es.es_lblk > last) {
 638                         if (last != start)
 639                                 holeoff = (loff_t)last << blkbits;
 640                         break;
 641                 }
 642                 /*
 643                  * If there is a unwritten extent at this offset,
 644                  * it will be as a data or a hole according to page
 645                  * cache that has data or not.
 646                  */
 647                 if (ext4_es_is_unwritten(&es) &&
 648                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 649                                               last + es.es_len, &holeoff))
 650                         break;
 651
 652                 last += es.es_len;
 653                 holeoff = (loff_t)last << blkbits;
 654                 cond_resched();
 655         } while (last <= end);
 656
 657         inode_unlock(inode);
 658
 659         if (holeoff > isize)
 660                 holeoff = isize;
 661
 662         return vfs_setpos(file, holeoff, maxsize);
 663 }
 664
 665 /*
 666  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 667  * by calling generic_file_llseek_size() with the appropriate maxbytes
 668  * value for each.
 669  */
 670 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 671 {
 672         struct inode *inode = file->f_mapping->host;
 673         loff_t maxbytes;
 674
 675         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 676                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 677         else
 678                 maxbytes = inode->i_sb->s_maxbytes;
 679
 680         switch (whence) {
 681         case SEEK_SET:
 682         case SEEK_CUR:
 683         case SEEK_END:
 684                 return generic_file_llseek_size(file, offset, whence,
 685                                                 maxbytes, i_size_read(inode));
 686         case SEEK_DATA:
 687                 return ext4_seek_data(file, offset, maxbytes);
 688         case SEEK_HOLE:
 689                 return ext4_seek_hole(file, offset, maxbytes);
 690         }
 691
 692         return -EINVAL;
 693 }
 694
 695 const struct file_operations ext4_file_operations = {
 696         .llseek         = ext4_llseek,
 697         .read_iter      = generic_file_read_iter,
 698         .write_iter     = ext4_file_write_iter,
 699         .unlocked_ioctl = ext4_ioctl,
 700 #ifdef CONFIG_COMPAT
 701         .compat_ioctl   = ext4_compat_ioctl,
 702 #endif
 703         .mmap           = ext4_file_mmap,
 704         .open           = ext4_file_open,
 705         .release        = ext4_release_file,
 706         .fsync          = ext4_sync_file,
 707         .splice_read    = generic_file_splice_read,
 708         .splice_write   = iter_file_splice_write,
 709         .fallocate      = ext4_fallocate,
 710 };
 711
 712 const struct inode_operations ext4_file_inode_operations = {
 713         .setattr        = ext4_setattr,
 714         .getattr        = ext4_getattr,
 715         .setxattr       = generic_setxattr,
 716         .getxattr       = generic_getxattr,
 717         .listxattr      = ext4_listxattr,
 718         .removexattr    = generic_removexattr,
 719         .get_acl        = ext4_get_acl,
 720         .set_acl        = ext4_set_acl,
 721         .fiemap         = ext4_fiemap,
 722 };
 723