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 #ifdef CONFIG_FS_DAX
  35 static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
  36 {
  37         struct inode *inode = file_inode(iocb->ki_filp);
  38         ssize_t ret;
  39
  40         if (!inode_trylock_shared(inode)) {
  41                 if (iocb->ki_flags & IOCB_NOWAIT)
  42                         return -EAGAIN;
  43                 inode_lock_shared(inode);
  44         }
  45         /*
  46          * Recheck under inode lock - at this point we are sure it cannot
  47          * change anymore
  48          */
  49         if (!IS_DAX(inode)) {
  50                 inode_unlock_shared(inode);
  51                 /* Fallback to buffered IO in case we cannot support DAX */
  52                 return generic_file_read_iter(iocb, to);
  53         }
  54         ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
  55         inode_unlock_shared(inode);
  56
  57         file_accessed(iocb->ki_filp);
  58         return ret;
  59 }
  60 #endif
  61
  62 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
  63 {
  64         if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
  65                 return -EIO;
  66
  67         if (!iov_iter_count(to))
  68                 return 0; /* skip atime */
  69
  70 #ifdef CONFIG_FS_DAX
  71         if (IS_DAX(file_inode(iocb->ki_filp)))
  72                 return ext4_dax_read_iter(iocb, to);
  73 #endif
  74         return generic_file_read_iter(iocb, to);
  75 }
  76
  77 /*
  78  * Called when an inode is released. Note that this is different
  79  * from ext4_file_open: open gets called at every open, but release
  80  * gets called only when /all/ the files are closed.
  81  */
  82 static int ext4_release_file(struct inode *inode, struct file *filp)
  83 {
  84         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
  85                 ext4_alloc_da_blocks(inode);
  86                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
  87         }
  88         /* if we are the last writer on the inode, drop the block reservation */
  89         if ((filp->f_mode & FMODE_WRITE) &&
  90                         (atomic_read(&inode->i_writecount) == 1) &&
  91                         !EXT4_I(inode)->i_reserved_data_blocks)
  92         {
  93                 down_write(&EXT4_I(inode)->i_data_sem);
  94                 ext4_discard_preallocations(inode);
  95                 up_write(&EXT4_I(inode)->i_data_sem);
  96         }
  97         if (is_dx(inode) && filp->private_data)
  98                 ext4_htree_free_dir_info(filp->private_data);
  99
 100         return 0;
 101 }
 102
 103 static void ext4_unwritten_wait(struct inode *inode)
 104 {
 105         wait_queue_head_t *wq = ext4_ioend_wq(inode);
 106
 107         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
 108 }
 109
 110 /*
 111  * This tests whether the IO in question is block-aligned or not.
 112  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
 113  * are converted to written only after the IO is complete.  Until they are
 114  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
 115  * it needs to zero out portions of the start and/or end block.  If 2 AIO
 116  * threads are at work on the same unwritten block, they must be synchronized
 117  * or one thread will zero the other's data, causing corruption.
 118  */
 119 static int
 120 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
 121 {
 122         struct super_block *sb = inode->i_sb;
 123         int blockmask = sb->s_blocksize - 1;
 124
 125         if (pos >= i_size_read(inode))
 126                 return 0;
 127
 128         if ((pos | iov_iter_alignment(from)) & blockmask)
 129                 return 1;
 130
 131         return 0;
 132 }
 133
 134 /* Is IO overwriting allocated and initialized blocks? */
 135 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
 136 {
 137         struct ext4_map_blocks map;
 138         unsigned int blkbits = inode->i_blkbits;
 139         int err, blklen;
 140
 141         if (pos + len > i_size_read(inode))
 142                 return false;
 143
 144         map.m_lblk = pos >> blkbits;
 145         map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
 146         blklen = map.m_len;
 147
 148         err = ext4_map_blocks(NULL, inode, &map, 0);
 149         /*
 150          * 'err==len' means that all of the blocks have been preallocated,
 151          * regardless of whether they have been initialized or not. To exclude
 152          * unwritten extents, we need to check m_flags.
 153          */
 154         return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
 155 }
 156
 157 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
 158 {
 159         struct inode *inode = file_inode(iocb->ki_filp);
 160         ssize_t ret;
 161
 162         ret = generic_write_checks(iocb, from);
 163         if (ret <= 0)
 164                 return ret;
 165         /*
 166          * If we have encountered a bitmap-format file, the size limit
 167          * is smaller than s_maxbytes, which is for extent-mapped files.
 168          */
 169         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 170                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 171
 172                 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
 173                         return -EFBIG;
 174                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 175         }
 176         return iov_iter_count(from);
 177 }
 178
 179 #ifdef CONFIG_FS_DAX
 180 static ssize_t
 181 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
 182 {
 183         struct inode *inode = file_inode(iocb->ki_filp);
 184         ssize_t ret;
 185
 186         if (!inode_trylock(inode)) {
 187                 if (iocb->ki_flags & IOCB_NOWAIT)
 188                         return -EAGAIN;
 189                 inode_lock(inode);
 190         }
 191         ret = ext4_write_checks(iocb, from);
 192         if (ret <= 0)
 193                 goto out;
 194         ret = file_remove_privs(iocb->ki_filp);
 195         if (ret)
 196                 goto out;
 197         ret = file_update_time(iocb->ki_filp);
 198         if (ret)
 199                 goto out;
 200
 201         ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
 202 out:
 203         inode_unlock(inode);
 204         if (ret > 0)
 205                 ret = generic_write_sync(iocb, ret);
 206         return ret;
 207 }
 208 #endif
 209
 210 static ssize_t
 211 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 212 {
 213         struct inode *inode = file_inode(iocb->ki_filp);
 214         int o_direct = iocb->ki_flags & IOCB_DIRECT;
 215         int unaligned_aio = 0;
 216         int overwrite = 0;
 217         ssize_t ret;
 218
 219         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 220                 return -EIO;
 221
 222 #ifdef CONFIG_FS_DAX
 223         if (IS_DAX(inode))
 224                 return ext4_dax_write_iter(iocb, from);
 225 #endif
 226
 227         if (!inode_trylock(inode)) {
 228                 if (iocb->ki_flags & IOCB_NOWAIT)
 229                         return -EAGAIN;
 230                 inode_lock(inode);
 231         }
 232
 233         ret = ext4_write_checks(iocb, from);
 234         if (ret <= 0)
 235                 goto out;
 236
 237         /*
 238          * Unaligned direct AIO must be serialized among each other as zeroing
 239          * of partial blocks of two competing unaligned AIOs can result in data
 240          * corruption.
 241          */
 242         if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 243             !is_sync_kiocb(iocb) &&
 244             ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
 245                 unaligned_aio = 1;
 246                 ext4_unwritten_wait(inode);
 247         }
 248
 249         iocb->private = &overwrite;
 250         /* Check whether we do a DIO overwrite or not */
 251         if (o_direct && !unaligned_aio) {
 252                 if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
 253                         if (ext4_should_dioread_nolock(inode))
 254                                 overwrite = 1;
 255                 } else if (iocb->ki_flags & IOCB_NOWAIT) {
 256                         ret = -EAGAIN;
 257                         goto out;
 258                 }
 259         }
 260
 261         ret = __generic_file_write_iter(iocb, from);
 262         inode_unlock(inode);
 263
 264         if (ret > 0)
 265                 ret = generic_write_sync(iocb, ret);
 266
 267         return ret;
 268
 269 out:
 270         inode_unlock(inode);
 271         return ret;
 272 }
 273
 274 #ifdef CONFIG_FS_DAX
 275 static int ext4_dax_huge_fault(struct vm_fault *vmf,
 276                 enum page_entry_size pe_size)
 277 {
 278         int result;
 279         handle_t *handle = NULL;
 280         struct inode *inode = file_inode(vmf->vma->vm_file);
 281         struct super_block *sb = inode->i_sb;
 282         bool write = vmf->flags & FAULT_FLAG_WRITE;
 283
 284         if (write) {
 285                 sb_start_pagefault(sb);
 286                 file_update_time(vmf->vma->vm_file);
 287                 down_read(&EXT4_I(inode)->i_mmap_sem);
 288                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 289                                                EXT4_DATA_TRANS_BLOCKS(sb));
 290         } else {
 291                 down_read(&EXT4_I(inode)->i_mmap_sem);
 292         }
 293         if (!IS_ERR(handle))
 294                 result = dax_iomap_fault(vmf, pe_size, &ext4_iomap_ops);
 295         else
 296                 result = VM_FAULT_SIGBUS;
 297         if (write) {
 298                 if (!IS_ERR(handle))
 299                         ext4_journal_stop(handle);
 300                 up_read(&EXT4_I(inode)->i_mmap_sem);
 301                 sb_end_pagefault(sb);
 302         } else {
 303                 up_read(&EXT4_I(inode)->i_mmap_sem);
 304         }
 305
 306         return result;
 307 }
 308
 309 static int ext4_dax_fault(struct vm_fault *vmf)
 310 {
 311         return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
 312 }
 313
 314 static const struct vm_operations_struct ext4_dax_vm_ops = {
 315         .fault          = ext4_dax_fault,
 316         .huge_fault     = ext4_dax_huge_fault,
 317         .page_mkwrite   = ext4_dax_fault,
 318         .pfn_mkwrite    = ext4_dax_fault,
 319 };
 320 #else
 321 #define ext4_dax_vm_ops ext4_file_vm_ops
 322 #endif
 323
 324 static const struct vm_operations_struct ext4_file_vm_ops = {
 325         .fault          = ext4_filemap_fault,
 326         .map_pages      = filemap_map_pages,
 327         .page_mkwrite   = ext4_page_mkwrite,
 328 };
 329
 330 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 331 {
 332         struct inode *inode = file->f_mapping->host;
 333
 334         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 335                 return -EIO;
 336
 337         file_accessed(file);
 338         if (IS_DAX(file_inode(file))) {
 339                 vma->vm_ops = &ext4_dax_vm_ops;
 340                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 341         } else {
 342                 vma->vm_ops = &ext4_file_vm_ops;
 343         }
 344         return 0;
 345 }
 346
 347 static int ext4_file_open(struct inode * inode, struct file * filp)
 348 {
 349         struct super_block *sb = inode->i_sb;
 350         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 351         struct vfsmount *mnt = filp->f_path.mnt;
 352         struct dentry *dir;
 353         struct path path;
 354         char buf[64], *cp;
 355         int ret;
 356
 357         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 358                 return -EIO;
 359
 360         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 361                      !(sb->s_flags & MS_RDONLY))) {
 362                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 363                 /*
 364                  * Sample where the filesystem has been mounted and
 365                  * store it in the superblock for sysadmin convenience
 366                  * when trying to sort through large numbers of block
 367                  * devices or filesystem images.
 368                  */
 369                 memset(buf, 0, sizeof(buf));
 370                 path.mnt = mnt;
 371                 path.dentry = mnt->mnt_root;
 372                 cp = d_path(&path, buf, sizeof(buf));
 373                 if (!IS_ERR(cp)) {
 374                         handle_t *handle;
 375                         int err;
 376
 377                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 378                         if (IS_ERR(handle))
 379                                 return PTR_ERR(handle);
 380                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 381                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 382                         if (err) {
 383                                 ext4_journal_stop(handle);
 384                                 return err;
 385                         }
 386                         strlcpy(sbi->s_es->s_last_mounted, cp,
 387                                 sizeof(sbi->s_es->s_last_mounted));
 388                         ext4_handle_dirty_super(handle, sb);
 389                         ext4_journal_stop(handle);
 390                 }
 391         }
 392         if (ext4_encrypted_inode(inode)) {
 393                 ret = fscrypt_get_encryption_info(inode);
 394                 if (ret)
 395                         return -EACCES;
 396                 if (!fscrypt_has_encryption_key(inode))
 397                         return -ENOKEY;
 398         }
 399
 400         dir = dget_parent(file_dentry(filp));
 401         if (ext4_encrypted_inode(d_inode(dir)) &&
 402                         !fscrypt_has_permitted_context(d_inode(dir), inode)) {
 403                 ext4_warning(inode->i_sb,
 404                              "Inconsistent encryption contexts: %lu/%lu",
 405                              (unsigned long) d_inode(dir)->i_ino,
 406                              (unsigned long) inode->i_ino);
 407                 dput(dir);
 408                 return -EPERM;
 409         }
 410         dput(dir);
 411         /*
 412          * Set up the jbd2_inode if we are opening the inode for
 413          * writing and the journal is present
 414          */
 415         if (filp->f_mode & FMODE_WRITE) {
 416                 ret = ext4_inode_attach_jinode(inode);
 417                 if (ret < 0)
 418                         return ret;
 419         }
 420
 421         /* Set the flags to support nowait AIO */
 422         filp->f_mode |= FMODE_AIO_NOWAIT;
 423
 424         return dquot_file_open(inode, filp);
 425 }
 426
 427 /*
 428  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 429  * file rather than ext4_ext_walk_space() because we can introduce
 430  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 431  * function.  When extent status tree has been fully implemented, it will
 432  * track all extent status for a file and we can directly use it to
 433  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 434  */
 435
 436 /*
 437  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 438  * lookup page cache to check whether or not there has some data between
 439  * [startoff, endoff] because, if this range contains an unwritten extent,
 440  * we determine this extent as a data or a hole according to whether the
 441  * page cache has data or not.
 442  */
 443 static int ext4_find_unwritten_pgoff(struct inode *inode,
 444                                      int whence,
 445                                      ext4_lblk_t end_blk,
 446                                      loff_t *offset)
 447 {
 448         struct pagevec pvec;
 449         unsigned int blkbits;
 450         pgoff_t index;
 451         pgoff_t end;
 452         loff_t endoff;
 453         loff_t startoff;
 454         loff_t lastoff;
 455         int found = 0;
 456
 457         blkbits = inode->i_sb->s_blocksize_bits;
 458         startoff = *offset;
 459         lastoff = startoff;
 460         endoff = (loff_t)end_blk << blkbits;
 461
 462         index = startoff >> PAGE_SHIFT;
 463         end = (endoff - 1) >> PAGE_SHIFT;
 464
 465         pagevec_init(&pvec, 0);
 466         do {
 467                 int i, num;
 468                 unsigned long nr_pages;
 469
 470                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE - 1) + 1;
 471                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, &index,
 472                                           (pgoff_t)num);
 473                 if (nr_pages == 0)
 474                         break;
 475
 476                 for (i = 0; i < nr_pages; i++) {
 477                         struct page *page = pvec.pages[i];
 478                         struct buffer_head *bh, *head;
 479
 480                         /*
 481                          * If current offset is smaller than the page offset,
 482                          * there is a hole at this offset.
 483                          */
 484                         if (whence == SEEK_HOLE && lastoff < endoff &&
 485                             lastoff < page_offset(pvec.pages[i])) {
 486                                 found = 1;
 487                                 *offset = lastoff;
 488                                 goto out;
 489                         }
 490
 491                         if (page->index > end)
 492                                 goto out;
 493
 494                         lock_page(page);
 495
 496                         if (unlikely(page->mapping != inode->i_mapping)) {
 497                                 unlock_page(page);
 498                                 continue;
 499                         }
 500
 501                         if (!page_has_buffers(page)) {
 502                                 unlock_page(page);
 503                                 continue;
 504                         }
 505
 506                         if (page_has_buffers(page)) {
 507                                 lastoff = page_offset(page);
 508                                 bh = head = page_buffers(page);
 509                                 do {
 510                                         if (lastoff + bh->b_size <= startoff)
 511                                                 goto next;
 512                                         if (buffer_uptodate(bh) ||
 513                                             buffer_unwritten(bh)) {
 514                                                 if (whence == SEEK_DATA)
 515                                                         found = 1;
 516                                         } else {
 517                                                 if (whence == SEEK_HOLE)
 518                                                         found = 1;
 519                                         }
 520                                         if (found) {
 521                                                 *offset = max_t(loff_t,
 522                                                         startoff, lastoff);
 523                                                 unlock_page(page);
 524                                                 goto out;
 525                                         }
 526 next:
 527                                         lastoff += bh->b_size;
 528                                         bh = bh->b_this_page;
 529                                 } while (bh != head);
 530                         }
 531
 532                         lastoff = page_offset(page) + PAGE_SIZE;
 533                         unlock_page(page);
 534                 }
 535
 536                 /* The no. of pages is less than our desired, we are done. */
 537                 if (nr_pages < num)
 538                         break;
 539                 pagevec_release(&pvec);
 540         } while (index <= end);
 541
 542         if (whence == SEEK_HOLE && lastoff < endoff) {
 543                 found = 1;
 544                 *offset = lastoff;
 545         }
 546 out:
 547         pagevec_release(&pvec);
 548         return found;
 549 }
 550
 551 /*
 552  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 553  */
 554 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 555 {
 556         struct inode *inode = file->f_mapping->host;
 557         struct extent_status es;
 558         ext4_lblk_t start, last, end;
 559         loff_t dataoff, isize;
 560         int blkbits;
 561         int ret;
 562
 563         inode_lock(inode);
 564
 565         isize = i_size_read(inode);
 566         if (offset >= isize) {
 567                 inode_unlock(inode);
 568                 return -ENXIO;
 569         }
 570
 571         blkbits = inode->i_sb->s_blocksize_bits;
 572         start = offset >> blkbits;
 573         last = start;
 574         end = isize >> blkbits;
 575         dataoff = offset;
 576
 577         do {
 578                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 579                 if (ret <= 0) {
 580                         /* No extent found -> no data */
 581                         if (ret == 0)
 582                                 ret = -ENXIO;
 583                         inode_unlock(inode);
 584                         return ret;
 585                 }
 586
 587                 last = es.es_lblk;
 588                 if (last != start)
 589                         dataoff = (loff_t)last << blkbits;
 590                 if (!ext4_es_is_unwritten(&es))
 591                         break;
 592
 593                 /*
 594                  * If there is a unwritten extent at this offset,
 595                  * it will be as a data or a hole according to page
 596                  * cache that has data or not.
 597                  */
 598                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 599                                               es.es_lblk + es.es_len, &dataoff))
 600                         break;
 601                 last += es.es_len;
 602                 dataoff = (loff_t)last << blkbits;
 603                 cond_resched();
 604         } while (last <= end);
 605
 606         inode_unlock(inode);
 607
 608         if (dataoff > isize)
 609                 return -ENXIO;
 610
 611         return vfs_setpos(file, dataoff, maxsize);
 612 }
 613
 614 /*
 615  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 616  */
 617 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 618 {
 619         struct inode *inode = file->f_mapping->host;
 620         struct extent_status es;
 621         ext4_lblk_t start, last, end;
 622         loff_t holeoff, isize;
 623         int blkbits;
 624         int ret;
 625
 626         inode_lock(inode);
 627
 628         isize = i_size_read(inode);
 629         if (offset >= isize) {
 630                 inode_unlock(inode);
 631                 return -ENXIO;
 632         }
 633
 634         blkbits = inode->i_sb->s_blocksize_bits;
 635         start = offset >> blkbits;
 636         last = start;
 637         end = isize >> blkbits;
 638         holeoff = offset;
 639
 640         do {
 641                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 642                 if (ret < 0) {
 643                         inode_unlock(inode);
 644                         return ret;
 645                 }
 646                 /* Found a hole? */
 647                 if (ret == 0 || es.es_lblk > last) {
 648                         if (last != start)
 649                                 holeoff = (loff_t)last << blkbits;
 650                         break;
 651                 }
 652                 /*
 653                  * If there is a unwritten extent at this offset,
 654                  * it will be as a data or a hole according to page
 655                  * cache that has data or not.
 656                  */
 657                 if (ext4_es_is_unwritten(&es) &&
 658                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 659                                               last + es.es_len, &holeoff))
 660                         break;
 661
 662                 last += es.es_len;
 663                 holeoff = (loff_t)last << blkbits;
 664                 cond_resched();
 665         } while (last <= end);
 666
 667         inode_unlock(inode);
 668
 669         if (holeoff > isize)
 670                 holeoff = isize;
 671
 672         return vfs_setpos(file, holeoff, maxsize);
 673 }
 674
 675 /*
 676  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 677  * by calling generic_file_llseek_size() with the appropriate maxbytes
 678  * value for each.
 679  */
 680 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 681 {
 682         struct inode *inode = file->f_mapping->host;
 683         loff_t maxbytes;
 684
 685         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 686                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 687         else
 688                 maxbytes = inode->i_sb->s_maxbytes;
 689
 690         switch (whence) {
 691         case SEEK_SET:
 692         case SEEK_CUR:
 693         case SEEK_END:
 694                 return generic_file_llseek_size(file, offset, whence,
 695                                                 maxbytes, i_size_read(inode));
 696         case SEEK_DATA:
 697                 return ext4_seek_data(file, offset, maxbytes);
 698         case SEEK_HOLE:
 699                 return ext4_seek_hole(file, offset, maxbytes);
 700         }
 701
 702         return -EINVAL;
 703 }
 704
 705 const struct file_operations ext4_file_operations = {
 706         .llseek         = ext4_llseek,
 707         .read_iter      = ext4_file_read_iter,
 708         .write_iter     = ext4_file_write_iter,
 709         .unlocked_ioctl = ext4_ioctl,
 710 #ifdef CONFIG_COMPAT
 711         .compat_ioctl   = ext4_compat_ioctl,
 712 #endif
 713         .mmap           = ext4_file_mmap,
 714         .open           = ext4_file_open,
 715         .release        = ext4_release_file,
 716         .fsync          = ext4_sync_file,
 717         .get_unmapped_area = thp_get_unmapped_area,
 718         .splice_read    = generic_file_splice_read,
 719         .splice_write   = iter_file_splice_write,
 720         .fallocate      = ext4_fallocate,
 721 };
 722
 723 const struct inode_operations ext4_file_inode_operations = {
 724         .setattr        = ext4_setattr,
 725         .getattr        = ext4_file_getattr,
 726         .listxattr      = ext4_listxattr,
 727         .get_acl        = ext4_get_acl,
 728         .set_acl        = ext4_set_acl,
 729         .fiemap         = ext4_fiemap,
 730 };
 731