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         inode_lock_shared(inode);
  41         /*
  42          * Recheck under inode lock - at this point we are sure it cannot
  43          * change anymore
  44          */
  45         if (!IS_DAX(inode)) {
  46                 inode_unlock_shared(inode);
  47                 /* Fallback to buffered IO in case we cannot support DAX */
  48                 return generic_file_read_iter(iocb, to);
  49         }
  50         ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
  51         inode_unlock_shared(inode);
  52
  53         file_accessed(iocb->ki_filp);
  54         return ret;
  55 }
  56 #endif
  57
  58 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
  59 {
  60         if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
  61                 return -EIO;
  62
  63         if (!iov_iter_count(to))
  64                 return 0; /* skip atime */
  65
  66 #ifdef CONFIG_FS_DAX
  67         if (IS_DAX(file_inode(iocb->ki_filp)))
  68                 return ext4_dax_read_iter(iocb, to);
  69 #endif
  70         return generic_file_read_iter(iocb, to);
  71 }
  72
  73 /*
  74  * Called when an inode is released. Note that this is different
  75  * from ext4_file_open: open gets called at every open, but release
  76  * gets called only when /all/ the files are closed.
  77  */
  78 static int ext4_release_file(struct inode *inode, struct file *filp)
  79 {
  80         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
  81                 ext4_alloc_da_blocks(inode);
  82                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
  83         }
  84         /* if we are the last writer on the inode, drop the block reservation */
  85         if ((filp->f_mode & FMODE_WRITE) &&
  86                         (atomic_read(&inode->i_writecount) == 1) &&
  87                         !EXT4_I(inode)->i_reserved_data_blocks)
  88         {
  89                 down_write(&EXT4_I(inode)->i_data_sem);
  90                 ext4_discard_preallocations(inode);
  91                 up_write(&EXT4_I(inode)->i_data_sem);
  92         }
  93         if (is_dx(inode) && filp->private_data)
  94                 ext4_htree_free_dir_info(filp->private_data);
  95
  96         return 0;
  97 }
  98
  99 static void ext4_unwritten_wait(struct inode *inode)
 100 {
 101         wait_queue_head_t *wq = ext4_ioend_wq(inode);
 102
 103         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
 104 }
 105
 106 /*
 107  * This tests whether the IO in question is block-aligned or not.
 108  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
 109  * are converted to written only after the IO is complete.  Until they are
 110  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
 111  * it needs to zero out portions of the start and/or end block.  If 2 AIO
 112  * threads are at work on the same unwritten block, they must be synchronized
 113  * or one thread will zero the other's data, causing corruption.
 114  */
 115 static int
 116 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
 117 {
 118         struct super_block *sb = inode->i_sb;
 119         int blockmask = sb->s_blocksize - 1;
 120
 121         if (pos >= i_size_read(inode))
 122                 return 0;
 123
 124         if ((pos | iov_iter_alignment(from)) & blockmask)
 125                 return 1;
 126
 127         return 0;
 128 }
 129
 130 /* Is IO overwriting allocated and initialized blocks? */
 131 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
 132 {
 133         struct ext4_map_blocks map;
 134         unsigned int blkbits = inode->i_blkbits;
 135         int err, blklen;
 136
 137         if (pos + len > i_size_read(inode))
 138                 return false;
 139
 140         map.m_lblk = pos >> blkbits;
 141         map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
 142         blklen = map.m_len;
 143
 144         err = ext4_map_blocks(NULL, inode, &map, 0);
 145         /*
 146          * 'err==len' means that all of the blocks have been preallocated,
 147          * regardless of whether they have been initialized or not. To exclude
 148          * unwritten extents, we need to check m_flags.
 149          */
 150         return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
 151 }
 152
 153 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
 154 {
 155         struct inode *inode = file_inode(iocb->ki_filp);
 156         ssize_t ret;
 157
 158         ret = generic_write_checks(iocb, from);
 159         if (ret <= 0)
 160                 return ret;
 161         /*
 162          * If we have encountered a bitmap-format file, the size limit
 163          * is smaller than s_maxbytes, which is for extent-mapped files.
 164          */
 165         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 166                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 167
 168                 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
 169                         return -EFBIG;
 170                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 171         }
 172         return iov_iter_count(from);
 173 }
 174
 175 #ifdef CONFIG_FS_DAX
 176 static ssize_t
 177 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
 178 {
 179         struct inode *inode = file_inode(iocb->ki_filp);
 180         ssize_t ret;
 181
 182         inode_lock(inode);
 183         ret = ext4_write_checks(iocb, from);
 184         if (ret <= 0)
 185                 goto out;
 186         ret = file_remove_privs(iocb->ki_filp);
 187         if (ret)
 188                 goto out;
 189         ret = file_update_time(iocb->ki_filp);
 190         if (ret)
 191                 goto out;
 192
 193         ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
 194 out:
 195         inode_unlock(inode);
 196         if (ret > 0)
 197                 ret = generic_write_sync(iocb, ret);
 198         return ret;
 199 }
 200 #endif
 201
 202 static ssize_t
 203 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 204 {
 205         struct inode *inode = file_inode(iocb->ki_filp);
 206         int o_direct = iocb->ki_flags & IOCB_DIRECT;
 207         int unaligned_aio = 0;
 208         int overwrite = 0;
 209         ssize_t ret;
 210
 211         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 212                 return -EIO;
 213
 214 #ifdef CONFIG_FS_DAX
 215         if (IS_DAX(inode))
 216                 return ext4_dax_write_iter(iocb, from);
 217 #endif
 218
 219         inode_lock(inode);
 220         ret = ext4_write_checks(iocb, from);
 221         if (ret <= 0)
 222                 goto out;
 223
 224         /*
 225          * Unaligned direct AIO must be serialized among each other as zeroing
 226          * of partial blocks of two competing unaligned AIOs can result in data
 227          * corruption.
 228          */
 229         if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 230             !is_sync_kiocb(iocb) &&
 231             ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
 232                 unaligned_aio = 1;
 233                 ext4_unwritten_wait(inode);
 234         }
 235
 236         iocb->private = &overwrite;
 237         /* Check whether we do a DIO overwrite or not */
 238         if (o_direct && ext4_should_dioread_nolock(inode) && !unaligned_aio &&
 239             ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from)))
 240                 overwrite = 1;
 241
 242         ret = __generic_file_write_iter(iocb, from);
 243         inode_unlock(inode);
 244
 245         if (ret > 0)
 246                 ret = generic_write_sync(iocb, ret);
 247
 248         return ret;
 249
 250 out:
 251         inode_unlock(inode);
 252         return ret;
 253 }
 254
 255 #ifdef CONFIG_FS_DAX
 256 static int ext4_dax_huge_fault(struct vm_fault *vmf,
 257                 enum page_entry_size pe_size)
 258 {
 259         int result;
 260         handle_t *handle = NULL;
 261         struct inode *inode = file_inode(vmf->vma->vm_file);
 262         struct super_block *sb = inode->i_sb;
 263         bool write = vmf->flags & FAULT_FLAG_WRITE;
 264
 265         if (write) {
 266                 sb_start_pagefault(sb);
 267                 file_update_time(vmf->vma->vm_file);
 268                 down_read(&EXT4_I(inode)->i_mmap_sem);
 269                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 270                                                EXT4_DATA_TRANS_BLOCKS(sb));
 271         } else {
 272                 down_read(&EXT4_I(inode)->i_mmap_sem);
 273         }
 274         if (!IS_ERR(handle))
 275                 result = dax_iomap_fault(vmf, pe_size, &ext4_iomap_ops);
 276         else
 277                 result = VM_FAULT_SIGBUS;
 278         if (write) {
 279                 if (!IS_ERR(handle))
 280                         ext4_journal_stop(handle);
 281                 up_read(&EXT4_I(inode)->i_mmap_sem);
 282                 sb_end_pagefault(sb);
 283         } else {
 284                 up_read(&EXT4_I(inode)->i_mmap_sem);
 285         }
 286
 287         return result;
 288 }
 289
 290 static int ext4_dax_fault(struct vm_fault *vmf)
 291 {
 292         return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
 293 }
 294
 295 /*
 296  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
 297  * handler we check for races agaist truncate. Note that since we cycle through
 298  * i_mmap_sem, we are sure that also any hole punching that began before we
 299  * were called is finished by now and so if it included part of the file we
 300  * are working on, our pte will get unmapped and the check for pte_same() in
 301  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 302  * desired.
 303  */
 304 static int ext4_dax_pfn_mkwrite(struct vm_fault *vmf)
 305 {
 306         struct inode *inode = file_inode(vmf->vma->vm_file);
 307         struct super_block *sb = inode->i_sb;
 308         loff_t size;
 309         int ret;
 310
 311         sb_start_pagefault(sb);
 312         file_update_time(vmf->vma->vm_file);
 313         down_read(&EXT4_I(inode)->i_mmap_sem);
 314         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 315         if (vmf->pgoff >= size)
 316                 ret = VM_FAULT_SIGBUS;
 317         else
 318                 ret = dax_pfn_mkwrite(vmf);
 319         up_read(&EXT4_I(inode)->i_mmap_sem);
 320         sb_end_pagefault(sb);
 321
 322         return ret;
 323 }
 324
 325 static const struct vm_operations_struct ext4_dax_vm_ops = {
 326         .fault          = ext4_dax_fault,
 327         .huge_fault     = ext4_dax_huge_fault,
 328         .page_mkwrite   = ext4_dax_fault,
 329         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 330 };
 331 #else
 332 #define ext4_dax_vm_ops ext4_file_vm_ops
 333 #endif
 334
 335 static const struct vm_operations_struct ext4_file_vm_ops = {
 336         .fault          = ext4_filemap_fault,
 337         .map_pages      = filemap_map_pages,
 338         .page_mkwrite   = ext4_page_mkwrite,
 339 };
 340
 341 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 342 {
 343         struct inode *inode = file->f_mapping->host;
 344
 345         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 346                 return -EIO;
 347
 348         if (ext4_encrypted_inode(inode)) {
 349                 int err = fscrypt_get_encryption_info(inode);
 350                 if (err)
 351                         return 0;
 352                 if (!fscrypt_has_encryption_key(inode))
 353                         return -ENOKEY;
 354         }
 355         file_accessed(file);
 356         if (IS_DAX(file_inode(file))) {
 357                 vma->vm_ops = &ext4_dax_vm_ops;
 358                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 359         } else {
 360                 vma->vm_ops = &ext4_file_vm_ops;
 361         }
 362         return 0;
 363 }
 364
 365 static int ext4_file_open(struct inode * inode, struct file * filp)
 366 {
 367         struct super_block *sb = inode->i_sb;
 368         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 369         struct vfsmount *mnt = filp->f_path.mnt;
 370         struct dentry *dir;
 371         struct path path;
 372         char buf[64], *cp;
 373         int ret;
 374
 375         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 376                 return -EIO;
 377
 378         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 379                      !(sb->s_flags & MS_RDONLY))) {
 380                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 381                 /*
 382                  * Sample where the filesystem has been mounted and
 383                  * store it in the superblock for sysadmin convenience
 384                  * when trying to sort through large numbers of block
 385                  * devices or filesystem images.
 386                  */
 387                 memset(buf, 0, sizeof(buf));
 388                 path.mnt = mnt;
 389                 path.dentry = mnt->mnt_root;
 390                 cp = d_path(&path, buf, sizeof(buf));
 391                 if (!IS_ERR(cp)) {
 392                         handle_t *handle;
 393                         int err;
 394
 395                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 396                         if (IS_ERR(handle))
 397                                 return PTR_ERR(handle);
 398                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 399                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 400                         if (err) {
 401                                 ext4_journal_stop(handle);
 402                                 return err;
 403                         }
 404                         strlcpy(sbi->s_es->s_last_mounted, cp,
 405                                 sizeof(sbi->s_es->s_last_mounted));
 406                         ext4_handle_dirty_super(handle, sb);
 407                         ext4_journal_stop(handle);
 408                 }
 409         }
 410         if (ext4_encrypted_inode(inode)) {
 411                 ret = fscrypt_get_encryption_info(inode);
 412                 if (ret)
 413                         return -EACCES;
 414                 if (!fscrypt_has_encryption_key(inode))
 415                         return -ENOKEY;
 416         }
 417
 418         dir = dget_parent(file_dentry(filp));
 419         if (ext4_encrypted_inode(d_inode(dir)) &&
 420                         !fscrypt_has_permitted_context(d_inode(dir), inode)) {
 421                 ext4_warning(inode->i_sb,
 422                              "Inconsistent encryption contexts: %lu/%lu",
 423                              (unsigned long) d_inode(dir)->i_ino,
 424                              (unsigned long) inode->i_ino);
 425                 dput(dir);
 426                 return -EPERM;
 427         }
 428         dput(dir);
 429         /*
 430          * Set up the jbd2_inode if we are opening the inode for
 431          * writing and the journal is present
 432          */
 433         if (filp->f_mode & FMODE_WRITE) {
 434                 ret = ext4_inode_attach_jinode(inode);
 435                 if (ret < 0)
 436                         return ret;
 437         }
 438         return dquot_file_open(inode, filp);
 439 }
 440
 441 /*
 442  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 443  * file rather than ext4_ext_walk_space() because we can introduce
 444  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 445  * function.  When extent status tree has been fully implemented, it will
 446  * track all extent status for a file and we can directly use it to
 447  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 448  */
 449
 450 /*
 451  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 452  * lookup page cache to check whether or not there has some data between
 453  * [startoff, endoff] because, if this range contains an unwritten extent,
 454  * we determine this extent as a data or a hole according to whether the
 455  * page cache has data or not.
 456  */
 457 static int ext4_find_unwritten_pgoff(struct inode *inode,
 458                                      int whence,
 459                                      ext4_lblk_t end_blk,
 460                                      loff_t *offset)
 461 {
 462         struct pagevec pvec;
 463         unsigned int blkbits;
 464         pgoff_t index;
 465         pgoff_t end;
 466         loff_t endoff;
 467         loff_t startoff;
 468         loff_t lastoff;
 469         int found = 0;
 470
 471         blkbits = inode->i_sb->s_blocksize_bits;
 472         startoff = *offset;
 473         lastoff = startoff;
 474         endoff = (loff_t)end_blk << blkbits;
 475
 476         index = startoff >> PAGE_SHIFT;
 477         end = endoff >> PAGE_SHIFT;
 478
 479         pagevec_init(&pvec, 0);
 480         do {
 481                 int i, num;
 482                 unsigned long nr_pages;
 483
 484                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 485                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 486                                           (pgoff_t)num);
 487                 if (nr_pages == 0) {
 488                         if (whence == SEEK_DATA)
 489                                 break;
 490
 491                         BUG_ON(whence != SEEK_HOLE);
 492                         /*
 493                          * If this is the first time to go into the loop and
 494                          * offset is not beyond the end offset, it will be a
 495                          * hole at this offset
 496                          */
 497                         if (lastoff == startoff || lastoff < endoff)
 498                                 found = 1;
 499                         break;
 500                 }
 501
 502                 /*
 503                  * If this is the first time to go into the loop and
 504                  * offset is smaller than the first page offset, it will be a
 505                  * hole at this offset.
 506                  */
 507                 if (lastoff == startoff && whence == SEEK_HOLE &&
 508                     lastoff < page_offset(pvec.pages[0])) {
 509                         found = 1;
 510                         break;
 511                 }
 512
 513                 for (i = 0; i < nr_pages; i++) {
 514                         struct page *page = pvec.pages[i];
 515                         struct buffer_head *bh, *head;
 516
 517                         /*
 518                          * If the current offset is not beyond the end of given
 519                          * range, it will be a hole.
 520                          */
 521                         if (lastoff < endoff && whence == SEEK_HOLE &&
 522                             page->index > end) {
 523                                 found = 1;
 524                                 *offset = lastoff;
 525                                 goto out;
 526                         }
 527
 528                         lock_page(page);
 529
 530                         if (unlikely(page->mapping != inode->i_mapping)) {
 531                                 unlock_page(page);
 532                                 continue;
 533                         }
 534
 535                         if (!page_has_buffers(page)) {
 536                                 unlock_page(page);
 537                                 continue;
 538                         }
 539
 540                         if (page_has_buffers(page)) {
 541                                 lastoff = page_offset(page);
 542                                 bh = head = page_buffers(page);
 543                                 do {
 544                                         if (buffer_uptodate(bh) ||
 545                                             buffer_unwritten(bh)) {
 546                                                 if (whence == SEEK_DATA)
 547                                                         found = 1;
 548                                         } else {
 549                                                 if (whence == SEEK_HOLE)
 550                                                         found = 1;
 551                                         }
 552                                         if (found) {
 553                                                 *offset = max_t(loff_t,
 554                                                         startoff, lastoff);
 555                                                 unlock_page(page);
 556                                                 goto out;
 557                                         }
 558                                         lastoff += bh->b_size;
 559                                         bh = bh->b_this_page;
 560                                 } while (bh != head);
 561                         }
 562
 563                         lastoff = page_offset(page) + PAGE_SIZE;
 564                         unlock_page(page);
 565                 }
 566
 567                 /*
 568                  * The no. of pages is less than our desired, that would be a
 569                  * hole in there.
 570                  */
 571                 if (nr_pages < num && whence == SEEK_HOLE) {
 572                         found = 1;
 573                         *offset = lastoff;
 574                         break;
 575                 }
 576
 577                 index = pvec.pages[i - 1]->index + 1;
 578                 pagevec_release(&pvec);
 579         } while (index <= end);
 580
 581 out:
 582         pagevec_release(&pvec);
 583         return found;
 584 }
 585
 586 /*
 587  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 588  */
 589 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 590 {
 591         struct inode *inode = file->f_mapping->host;
 592         struct extent_status es;
 593         ext4_lblk_t start, last, end;
 594         loff_t dataoff, isize;
 595         int blkbits;
 596         int ret;
 597
 598         inode_lock(inode);
 599
 600         isize = i_size_read(inode);
 601         if (offset >= isize) {
 602                 inode_unlock(inode);
 603                 return -ENXIO;
 604         }
 605
 606         blkbits = inode->i_sb->s_blocksize_bits;
 607         start = offset >> blkbits;
 608         last = start;
 609         end = isize >> blkbits;
 610         dataoff = offset;
 611
 612         do {
 613                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 614                 if (ret <= 0) {
 615                         /* No extent found -> no data */
 616                         if (ret == 0)
 617                                 ret = -ENXIO;
 618                         inode_unlock(inode);
 619                         return ret;
 620                 }
 621
 622                 last = es.es_lblk;
 623                 if (last != start)
 624                         dataoff = (loff_t)last << blkbits;
 625                 if (!ext4_es_is_unwritten(&es))
 626                         break;
 627
 628                 /*
 629                  * If there is a unwritten extent at this offset,
 630                  * it will be as a data or a hole according to page
 631                  * cache that has data or not.
 632                  */
 633                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 634                                               es.es_lblk + es.es_len, &dataoff))
 635                         break;
 636                 last += es.es_len;
 637                 dataoff = (loff_t)last << blkbits;
 638                 cond_resched();
 639         } while (last <= end);
 640
 641         inode_unlock(inode);
 642
 643         if (dataoff > isize)
 644                 return -ENXIO;
 645
 646         return vfs_setpos(file, dataoff, maxsize);
 647 }
 648
 649 /*
 650  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 651  */
 652 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 653 {
 654         struct inode *inode = file->f_mapping->host;
 655         struct extent_status es;
 656         ext4_lblk_t start, last, end;
 657         loff_t holeoff, isize;
 658         int blkbits;
 659         int ret;
 660
 661         inode_lock(inode);
 662
 663         isize = i_size_read(inode);
 664         if (offset >= isize) {
 665                 inode_unlock(inode);
 666                 return -ENXIO;
 667         }
 668
 669         blkbits = inode->i_sb->s_blocksize_bits;
 670         start = offset >> blkbits;
 671         last = start;
 672         end = isize >> blkbits;
 673         holeoff = offset;
 674
 675         do {
 676                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 677                 if (ret < 0) {
 678                         inode_unlock(inode);
 679                         return ret;
 680                 }
 681                 /* Found a hole? */
 682                 if (ret == 0 || es.es_lblk > last) {
 683                         if (last != start)
 684                                 holeoff = (loff_t)last << blkbits;
 685                         break;
 686                 }
 687                 /*
 688                  * If there is a unwritten extent at this offset,
 689                  * it will be as a data or a hole according to page
 690                  * cache that has data or not.
 691                  */
 692                 if (ext4_es_is_unwritten(&es) &&
 693                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 694                                               last + es.es_len, &holeoff))
 695                         break;
 696
 697                 last += es.es_len;
 698                 holeoff = (loff_t)last << blkbits;
 699                 cond_resched();
 700         } while (last <= end);
 701
 702         inode_unlock(inode);
 703
 704         if (holeoff > isize)
 705                 holeoff = isize;
 706
 707         return vfs_setpos(file, holeoff, maxsize);
 708 }
 709
 710 /*
 711  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 712  * by calling generic_file_llseek_size() with the appropriate maxbytes
 713  * value for each.
 714  */
 715 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 716 {
 717         struct inode *inode = file->f_mapping->host;
 718         loff_t maxbytes;
 719
 720         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 721                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 722         else
 723                 maxbytes = inode->i_sb->s_maxbytes;
 724
 725         switch (whence) {
 726         case SEEK_SET:
 727         case SEEK_CUR:
 728         case SEEK_END:
 729                 return generic_file_llseek_size(file, offset, whence,
 730                                                 maxbytes, i_size_read(inode));
 731         case SEEK_DATA:
 732                 return ext4_seek_data(file, offset, maxbytes);
 733         case SEEK_HOLE:
 734                 return ext4_seek_hole(file, offset, maxbytes);
 735         }
 736
 737         return -EINVAL;
 738 }
 739
 740 const struct file_operations ext4_file_operations = {
 741         .llseek         = ext4_llseek,
 742         .read_iter      = ext4_file_read_iter,
 743         .write_iter     = ext4_file_write_iter,
 744         .unlocked_ioctl = ext4_ioctl,
 745 #ifdef CONFIG_COMPAT
 746         .compat_ioctl   = ext4_compat_ioctl,
 747 #endif
 748         .mmap           = ext4_file_mmap,
 749         .open           = ext4_file_open,
 750         .release        = ext4_release_file,
 751         .fsync          = ext4_sync_file,
 752         .get_unmapped_area = thp_get_unmapped_area,
 753         .splice_read    = generic_file_splice_read,
 754         .splice_write   = iter_file_splice_write,
 755         .fallocate      = ext4_fallocate,
 756 };
 757
 758 const struct inode_operations ext4_file_inode_operations = {
 759         .setattr        = ext4_setattr,
 760         .getattr        = ext4_file_getattr,
 761         .listxattr      = ext4_listxattr,
 762         .get_acl        = ext4_get_acl,
 763         .set_acl        = ext4_set_acl,
 764         .fiemap         = ext4_fiemap,
 765 };
 766