fs/ext4/file.c

   1 /*
   2  *  linux/fs/ext4/file.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   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  *      ([email protected])
  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         if (!o_direct && (iocb->ki_flags & IOCB_NOWAIT))
 227                 return -EOPNOTSUPP;
 228
 229         if (!inode_trylock(inode)) {
 230                 if (iocb->ki_flags & IOCB_NOWAIT)
 231                         return -EAGAIN;
 232                 inode_lock(inode);
 233         }
 234
 235         ret = ext4_write_checks(iocb, from);
 236         if (ret <= 0)
 237                 goto out;
 238
 239         /*
 240          * Unaligned direct AIO must be serialized among each other as zeroing
 241          * of partial blocks of two competing unaligned AIOs can result in data
 242          * corruption.
 243          */
 244         if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 245             !is_sync_kiocb(iocb) &&
 246             ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
 247                 unaligned_aio = 1;
 248                 ext4_unwritten_wait(inode);
 249         }
 250
 251         iocb->private = &overwrite;
 252         /* Check whether we do a DIO overwrite or not */
 253         if (o_direct && !unaligned_aio) {
 254                 if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
 255                         if (ext4_should_dioread_nolock(inode))
 256                                 overwrite = 1;
 257                 } else if (iocb->ki_flags & IOCB_NOWAIT) {
 258                         ret = -EAGAIN;
 259                         goto out;
 260                 }
 261         }
 262
 263         ret = __generic_file_write_iter(iocb, from);
 264         inode_unlock(inode);
 265
 266         if (ret > 0)
 267                 ret = generic_write_sync(iocb, ret);
 268
 269         return ret;
 270
 271 out:
 272         inode_unlock(inode);
 273         return ret;
 274 }
 275
 276 #ifdef CONFIG_FS_DAX
 277 static int ext4_dax_huge_fault(struct vm_fault *vmf,
 278                 enum page_entry_size pe_size)
 279 {
 280         int result;
 281         handle_t *handle = NULL;
 282         struct inode *inode = file_inode(vmf->vma->vm_file);
 283         struct super_block *sb = inode->i_sb;
 284
 285         /*
 286          * We have to distinguish real writes from writes which will result in a
 287          * COW page; COW writes should *not* poke the journal (the file will not
 288          * be changed). Doing so would cause unintended failures when mounted
 289          * read-only.
 290          *
 291          * We check for VM_SHARED rather than vmf->cow_page since the latter is
 292          * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
 293          * other sizes, dax_iomap_fault will handle splitting / fallback so that
 294          * we eventually come back with a COW page.
 295          */
 296         bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
 297                 (vmf->vma->vm_flags & VM_SHARED);
 298
 299         if (write) {
 300                 sb_start_pagefault(sb);
 301                 file_update_time(vmf->vma->vm_file);
 302                 down_read(&EXT4_I(inode)->i_mmap_sem);
 303                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 304                                                EXT4_DATA_TRANS_BLOCKS(sb));
 305         } else {
 306                 down_read(&EXT4_I(inode)->i_mmap_sem);
 307         }
 308         if (!IS_ERR(handle))
 309                 result = dax_iomap_fault(vmf, pe_size, &ext4_iomap_ops);
 310         else
 311                 result = VM_FAULT_SIGBUS;
 312         if (write) {
 313                 if (!IS_ERR(handle))
 314                         ext4_journal_stop(handle);
 315                 up_read(&EXT4_I(inode)->i_mmap_sem);
 316                 sb_end_pagefault(sb);
 317         } else {
 318                 up_read(&EXT4_I(inode)->i_mmap_sem);
 319         }
 320
 321         return result;
 322 }
 323
 324 static int ext4_dax_fault(struct vm_fault *vmf)
 325 {
 326         return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
 327 }
 328
 329 static const struct vm_operations_struct ext4_dax_vm_ops = {
 330         .fault          = ext4_dax_fault,
 331         .huge_fault     = ext4_dax_huge_fault,
 332         .page_mkwrite   = ext4_dax_fault,
 333         .pfn_mkwrite    = ext4_dax_fault,
 334 };
 335 #else
 336 #define ext4_dax_vm_ops ext4_file_vm_ops
 337 #endif
 338
 339 static const struct vm_operations_struct ext4_file_vm_ops = {
 340         .fault          = ext4_filemap_fault,
 341         .map_pages      = filemap_map_pages,
 342         .page_mkwrite   = ext4_page_mkwrite,
 343 };
 344
 345 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 346 {
 347         struct inode *inode = file->f_mapping->host;
 348
 349         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 350                 return -EIO;
 351
 352         file_accessed(file);
 353         if (IS_DAX(file_inode(file))) {
 354                 vma->vm_ops = &ext4_dax_vm_ops;
 355                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 356         } else {
 357                 vma->vm_ops = &ext4_file_vm_ops;
 358         }
 359         return 0;
 360 }
 361
 362 static int ext4_file_open(struct inode * inode, struct file * filp)
 363 {
 364         struct super_block *sb = inode->i_sb;
 365         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 366         struct vfsmount *mnt = filp->f_path.mnt;
 367         struct dentry *dir;
 368         struct path path;
 369         char buf[64], *cp;
 370         int ret;
 371
 372         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 373                 return -EIO;
 374
 375         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 376                      !sb_rdonly(sb))) {
 377                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 378                 /*
 379                  * Sample where the filesystem has been mounted and
 380                  * store it in the superblock for sysadmin convenience
 381                  * when trying to sort through large numbers of block
 382                  * devices or filesystem images.
 383                  */
 384                 memset(buf, 0, sizeof(buf));
 385                 path.mnt = mnt;
 386                 path.dentry = mnt->mnt_root;
 387                 cp = d_path(&path, buf, sizeof(buf));
 388                 if (!IS_ERR(cp)) {
 389                         handle_t *handle;
 390                         int err;
 391
 392                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 393                         if (IS_ERR(handle))
 394                                 return PTR_ERR(handle);
 395                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 396                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 397                         if (err) {
 398                                 ext4_journal_stop(handle);
 399                                 return err;
 400                         }
 401                         strlcpy(sbi->s_es->s_last_mounted, cp,
 402                                 sizeof(sbi->s_es->s_last_mounted));
 403                         ext4_handle_dirty_super(handle, sb);
 404                         ext4_journal_stop(handle);
 405                 }
 406         }
 407         if (ext4_encrypted_inode(inode)) {
 408                 ret = fscrypt_get_encryption_info(inode);
 409                 if (ret)
 410                         return -EACCES;
 411                 if (!fscrypt_has_encryption_key(inode))
 412                         return -ENOKEY;
 413         }
 414
 415         dir = dget_parent(file_dentry(filp));
 416         if (ext4_encrypted_inode(d_inode(dir)) &&
 417                         !fscrypt_has_permitted_context(d_inode(dir), inode)) {
 418                 ext4_warning(inode->i_sb,
 419                              "Inconsistent encryption contexts: %lu/%lu",
 420                              (unsigned long) d_inode(dir)->i_ino,
 421                              (unsigned long) inode->i_ino);
 422                 dput(dir);
 423                 return -EPERM;
 424         }
 425         dput(dir);
 426         /*
 427          * Set up the jbd2_inode if we are opening the inode for
 428          * writing and the journal is present
 429          */
 430         if (filp->f_mode & FMODE_WRITE) {
 431                 ret = ext4_inode_attach_jinode(inode);
 432                 if (ret < 0)
 433                         return ret;
 434         }
 435
 436         filp->f_mode |= FMODE_NOWAIT;
 437         return dquot_file_open(inode, filp);
 438 }
 439
 440 /*
 441  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 442  * file rather than ext4_ext_walk_space() because we can introduce
 443  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 444  * function.  When extent status tree has been fully implemented, it will
 445  * track all extent status for a file and we can directly use it to
 446  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 447  */
 448
 449 /*
 450  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 451  * lookup page cache to check whether or not there has some data between
 452  * [startoff, endoff] because, if this range contains an unwritten extent,
 453  * we determine this extent as a data or a hole according to whether the
 454  * page cache has data or not.
 455  */
 456 static int ext4_find_unwritten_pgoff(struct inode *inode,
 457                                      int whence,
 458                                      ext4_lblk_t end_blk,
 459                                      loff_t *offset)
 460 {
 461         struct pagevec pvec;
 462         unsigned int blkbits;
 463         pgoff_t index;
 464         pgoff_t end;
 465         loff_t endoff;
 466         loff_t startoff;
 467         loff_t lastoff;
 468         int found = 0;
 469
 470         blkbits = inode->i_sb->s_blocksize_bits;
 471         startoff = *offset;
 472         lastoff = startoff;
 473         endoff = (loff_t)end_blk << blkbits;
 474
 475         index = startoff >> PAGE_SHIFT;
 476         end = (endoff - 1) >> PAGE_SHIFT;
 477
 478         pagevec_init(&pvec, 0);
 479         do {
 480                 int i;
 481                 unsigned long nr_pages;
 482
 483                 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
 484                                         &index, end);
 485                 if (nr_pages == 0)
 486                         break;
 487
 488                 for (i = 0; i < nr_pages; i++) {
 489                         struct page *page = pvec.pages[i];
 490                         struct buffer_head *bh, *head;
 491
 492                         /*
 493                          * If current offset is smaller than the page offset,
 494                          * there is a hole at this offset.
 495                          */
 496                         if (whence == SEEK_HOLE && lastoff < endoff &&
 497                             lastoff < page_offset(pvec.pages[i])) {
 498                                 found = 1;
 499                                 *offset = lastoff;
 500                                 goto out;
 501                         }
 502
 503                         lock_page(page);
 504
 505                         if (unlikely(page->mapping != inode->i_mapping)) {
 506                                 unlock_page(page);
 507                                 continue;
 508                         }
 509
 510                         if (!page_has_buffers(page)) {
 511                                 unlock_page(page);
 512                                 continue;
 513                         }
 514
 515                         if (page_has_buffers(page)) {
 516                                 lastoff = page_offset(page);
 517                                 bh = head = page_buffers(page);
 518                                 do {
 519                                         if (lastoff + bh->b_size <= startoff)
 520                                                 goto next;
 521                                         if (buffer_uptodate(bh) ||
 522                                             buffer_unwritten(bh)) {
 523                                                 if (whence == SEEK_DATA)
 524                                                         found = 1;
 525                                         } else {
 526                                                 if (whence == SEEK_HOLE)
 527                                                         found = 1;
 528                                         }
 529                                         if (found) {
 530                                                 *offset = max_t(loff_t,
 531                                                         startoff, lastoff);
 532                                                 unlock_page(page);
 533                                                 goto out;
 534                                         }
 535 next:
 536                                         lastoff += bh->b_size;
 537                                         bh = bh->b_this_page;
 538                                 } while (bh != head);
 539                         }
 540
 541                         lastoff = page_offset(page) + PAGE_SIZE;
 542                         unlock_page(page);
 543                 }
 544
 545                 pagevec_release(&pvec);
 546         } while (index <= end);
 547
 548         /* There are no pages upto endoff - that would be a hole in there. */
 549         if (whence == SEEK_HOLE && lastoff < endoff) {
 550                 found = 1;
 551                 *offset = lastoff;
 552         }
 553 out:
 554         pagevec_release(&pvec);
 555         return found;
 556 }
 557
 558 /*
 559  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 560  */
 561 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 562 {
 563         struct inode *inode = file->f_mapping->host;
 564         struct extent_status es;
 565         ext4_lblk_t start, last, end;
 566         loff_t dataoff, isize;
 567         int blkbits;
 568         int ret;
 569
 570         inode_lock(inode);
 571
 572         isize = i_size_read(inode);
 573         if (offset < 0 || offset >= isize) {
 574                 inode_unlock(inode);
 575                 return -ENXIO;
 576         }
 577
 578         blkbits = inode->i_sb->s_blocksize_bits;
 579         start = offset >> blkbits;
 580         last = start;
 581         end = isize >> blkbits;
 582         dataoff = offset;
 583
 584         do {
 585                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 586                 if (ret <= 0) {
 587                         /* No extent found -> no data */
 588                         if (ret == 0)
 589                                 ret = -ENXIO;
 590                         inode_unlock(inode);
 591                         return ret;
 592                 }
 593
 594                 last = es.es_lblk;
 595                 if (last != start)
 596                         dataoff = (loff_t)last << blkbits;
 597                 if (!ext4_es_is_unwritten(&es))
 598                         break;
 599
 600                 /*
 601                  * If there is a unwritten extent at this offset,
 602                  * it will be as a data or a hole according to page
 603                  * cache that has data or not.
 604                  */
 605                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 606                                               es.es_lblk + es.es_len, &dataoff))
 607                         break;
 608                 last += es.es_len;
 609                 dataoff = (loff_t)last << blkbits;
 610                 cond_resched();
 611         } while (last <= end);
 612
 613         inode_unlock(inode);
 614
 615         if (dataoff > isize)
 616                 return -ENXIO;
 617
 618         return vfs_setpos(file, dataoff, maxsize);
 619 }
 620
 621 /*
 622  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 623  */
 624 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 625 {
 626         struct inode *inode = file->f_mapping->host;
 627         struct extent_status es;
 628         ext4_lblk_t start, last, end;
 629         loff_t holeoff, isize;
 630         int blkbits;
 631         int ret;
 632
 633         inode_lock(inode);
 634
 635         isize = i_size_read(inode);
 636         if (offset < 0 || offset >= isize) {
 637                 inode_unlock(inode);
 638                 return -ENXIO;
 639         }
 640
 641         blkbits = inode->i_sb->s_blocksize_bits;
 642         start = offset >> blkbits;
 643         last = start;
 644         end = isize >> blkbits;
 645         holeoff = offset;
 646
 647         do {
 648                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 649                 if (ret < 0) {
 650                         inode_unlock(inode);
 651                         return ret;
 652                 }
 653                 /* Found a hole? */
 654                 if (ret == 0 || es.es_lblk > last) {
 655                         if (last != start)
 656                                 holeoff = (loff_t)last << blkbits;
 657                         break;
 658                 }
 659                 /*
 660                  * If there is a unwritten extent at this offset,
 661                  * it will be as a data or a hole according to page
 662                  * cache that has data or not.
 663                  */
 664                 if (ext4_es_is_unwritten(&es) &&
 665                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 666                                               last + es.es_len, &holeoff))
 667                         break;
 668
 669                 last += es.es_len;
 670                 holeoff = (loff_t)last << blkbits;
 671                 cond_resched();
 672         } while (last <= end);
 673
 674         inode_unlock(inode);
 675
 676         if (holeoff > isize)
 677                 holeoff = isize;
 678
 679         return vfs_setpos(file, holeoff, maxsize);
 680 }
 681
 682 /*
 683  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 684  * by calling generic_file_llseek_size() with the appropriate maxbytes
 685  * value for each.
 686  */
 687 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 688 {
 689         struct inode *inode = file->f_mapping->host;
 690         loff_t maxbytes;
 691
 692         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 693                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 694         else
 695                 maxbytes = inode->i_sb->s_maxbytes;
 696
 697         switch (whence) {
 698         case SEEK_SET:
 699         case SEEK_CUR:
 700         case SEEK_END:
 701                 return generic_file_llseek_size(file, offset, whence,
 702                                                 maxbytes, i_size_read(inode));
 703         case SEEK_DATA:
 704                 return ext4_seek_data(file, offset, maxbytes);
 705         case SEEK_HOLE:
 706                 return ext4_seek_hole(file, offset, maxbytes);
 707         }
 708
 709         return -EINVAL;
 710 }
 711
 712 const struct file_operations ext4_file_operations = {
 713         .llseek         = ext4_llseek,
 714         .read_iter      = ext4_file_read_iter,
 715         .write_iter     = ext4_file_write_iter,
 716         .unlocked_ioctl = ext4_ioctl,
 717 #ifdef CONFIG_COMPAT
 718         .compat_ioctl   = ext4_compat_ioctl,
 719 #endif
 720         .mmap           = ext4_file_mmap,
 721         .open           = ext4_file_open,
 722         .release        = ext4_release_file,
 723         .fsync          = ext4_sync_file,
 724         .get_unmapped_area = thp_get_unmapped_area,
 725         .splice_read    = generic_file_splice_read,
 726         .splice_write   = iter_file_splice_write,
 727         .fallocate      = ext4_fallocate,
 728 };
 729
 730 const struct inode_operations ext4_file_inode_operations = {
 731         .setattr        = ext4_setattr,
 732         .getattr        = ext4_file_getattr,
 733         .listxattr      = ext4_listxattr,
 734         .get_acl        = ext4_get_acl,
 735         .set_acl        = ext4_set_acl,
 736         .fiemap         = ext4_fiemap,
 737 };
 738