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         struct inode *inode = file_inode(vmf->vma->vm_file);
 261         struct super_block *sb = inode->i_sb;
 262         bool write = vmf->flags & FAULT_FLAG_WRITE;
 263
 264         if (write) {
 265                 sb_start_pagefault(sb);
 266                 file_update_time(vmf->vma->vm_file);
 267         }
 268         down_read(&EXT4_I(inode)->i_mmap_sem);
 269         result = dax_iomap_fault(vmf, pe_size, &ext4_iomap_ops);
 270         up_read(&EXT4_I(inode)->i_mmap_sem);
 271         if (write)
 272                 sb_end_pagefault(sb);
 273
 274         return result;
 275 }
 276
 277 static int ext4_dax_fault(struct vm_fault *vmf)
 278 {
 279         return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
 280 }
 281
 282 /*
 283  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
 284  * handler we check for races agaist truncate. Note that since we cycle through
 285  * i_mmap_sem, we are sure that also any hole punching that began before we
 286  * were called is finished by now and so if it included part of the file we
 287  * are working on, our pte will get unmapped and the check for pte_same() in
 288  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 289  * desired.
 290  */
 291 static int ext4_dax_pfn_mkwrite(struct vm_fault *vmf)
 292 {
 293         struct inode *inode = file_inode(vmf->vma->vm_file);
 294         struct super_block *sb = inode->i_sb;
 295         loff_t size;
 296         int ret;
 297
 298         sb_start_pagefault(sb);
 299         file_update_time(vmf->vma->vm_file);
 300         down_read(&EXT4_I(inode)->i_mmap_sem);
 301         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 302         if (vmf->pgoff >= size)
 303                 ret = VM_FAULT_SIGBUS;
 304         else
 305                 ret = dax_pfn_mkwrite(vmf);
 306         up_read(&EXT4_I(inode)->i_mmap_sem);
 307         sb_end_pagefault(sb);
 308
 309         return ret;
 310 }
 311
 312 static const struct vm_operations_struct ext4_dax_vm_ops = {
 313         .fault          = ext4_dax_fault,
 314         .huge_fault     = ext4_dax_huge_fault,
 315         .page_mkwrite   = ext4_dax_fault,
 316         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 317 };
 318 #else
 319 #define ext4_dax_vm_ops ext4_file_vm_ops
 320 #endif
 321
 322 static const struct vm_operations_struct ext4_file_vm_ops = {
 323         .fault          = ext4_filemap_fault,
 324         .map_pages      = filemap_map_pages,
 325         .page_mkwrite   = ext4_page_mkwrite,
 326 };
 327
 328 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 329 {
 330         struct inode *inode = file->f_mapping->host;
 331
 332         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 333                 return -EIO;
 334
 335         if (ext4_encrypted_inode(inode)) {
 336                 int err = fscrypt_get_encryption_info(inode);
 337                 if (err)
 338                         return 0;
 339                 if (!fscrypt_has_encryption_key(inode))
 340                         return -ENOKEY;
 341         }
 342         file_accessed(file);
 343         if (IS_DAX(file_inode(file))) {
 344                 vma->vm_ops = &ext4_dax_vm_ops;
 345                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 346         } else {
 347                 vma->vm_ops = &ext4_file_vm_ops;
 348         }
 349         return 0;
 350 }
 351
 352 static int ext4_file_open(struct inode * inode, struct file * filp)
 353 {
 354         struct super_block *sb = inode->i_sb;
 355         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 356         struct vfsmount *mnt = filp->f_path.mnt;
 357         struct dentry *dir;
 358         struct path path;
 359         char buf[64], *cp;
 360         int ret;
 361
 362         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 363                 return -EIO;
 364
 365         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 366                      !(sb->s_flags & MS_RDONLY))) {
 367                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 368                 /*
 369                  * Sample where the filesystem has been mounted and
 370                  * store it in the superblock for sysadmin convenience
 371                  * when trying to sort through large numbers of block
 372                  * devices or filesystem images.
 373                  */
 374                 memset(buf, 0, sizeof(buf));
 375                 path.mnt = mnt;
 376                 path.dentry = mnt->mnt_root;
 377                 cp = d_path(&path, buf, sizeof(buf));
 378                 if (!IS_ERR(cp)) {
 379                         handle_t *handle;
 380                         int err;
 381
 382                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 383                         if (IS_ERR(handle))
 384                                 return PTR_ERR(handle);
 385                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 386                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 387                         if (err) {
 388                                 ext4_journal_stop(handle);
 389                                 return err;
 390                         }
 391                         strlcpy(sbi->s_es->s_last_mounted, cp,
 392                                 sizeof(sbi->s_es->s_last_mounted));
 393                         ext4_handle_dirty_super(handle, sb);
 394                         ext4_journal_stop(handle);
 395                 }
 396         }
 397         if (ext4_encrypted_inode(inode)) {
 398                 ret = fscrypt_get_encryption_info(inode);
 399                 if (ret)
 400                         return -EACCES;
 401                 if (!fscrypt_has_encryption_key(inode))
 402                         return -ENOKEY;
 403         }
 404
 405         dir = dget_parent(file_dentry(filp));
 406         if (ext4_encrypted_inode(d_inode(dir)) &&
 407                         !fscrypt_has_permitted_context(d_inode(dir), inode)) {
 408                 ext4_warning(inode->i_sb,
 409                              "Inconsistent encryption contexts: %lu/%lu",
 410                              (unsigned long) d_inode(dir)->i_ino,
 411                              (unsigned long) inode->i_ino);
 412                 dput(dir);
 413                 return -EPERM;
 414         }
 415         dput(dir);
 416         /*
 417          * Set up the jbd2_inode if we are opening the inode for
 418          * writing and the journal is present
 419          */
 420         if (filp->f_mode & FMODE_WRITE) {
 421                 ret = ext4_inode_attach_jinode(inode);
 422                 if (ret < 0)
 423                         return ret;
 424         }
 425         return dquot_file_open(inode, filp);
 426 }
 427
 428 /*
 429  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 430  * file rather than ext4_ext_walk_space() because we can introduce
 431  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 432  * function.  When extent status tree has been fully implemented, it will
 433  * track all extent status for a file and we can directly use it to
 434  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 435  */
 436
 437 /*
 438  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 439  * lookup page cache to check whether or not there has some data between
 440  * [startoff, endoff] because, if this range contains an unwritten extent,
 441  * we determine this extent as a data or a hole according to whether the
 442  * page cache has data or not.
 443  */
 444 static int ext4_find_unwritten_pgoff(struct inode *inode,
 445                                      int whence,
 446                                      ext4_lblk_t end_blk,
 447                                      loff_t *offset)
 448 {
 449         struct pagevec pvec;
 450         unsigned int blkbits;
 451         pgoff_t index;
 452         pgoff_t end;
 453         loff_t endoff;
 454         loff_t startoff;
 455         loff_t lastoff;
 456         int found = 0;
 457
 458         blkbits = inode->i_sb->s_blocksize_bits;
 459         startoff = *offset;
 460         lastoff = startoff;
 461         endoff = (loff_t)end_blk << blkbits;
 462
 463         index = startoff >> PAGE_SHIFT;
 464         end = endoff >> PAGE_SHIFT;
 465
 466         pagevec_init(&pvec, 0);
 467         do {
 468                 int i, num;
 469                 unsigned long nr_pages;
 470
 471                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 472                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 473                                           (pgoff_t)num);
 474                 if (nr_pages == 0) {
 475                         if (whence == SEEK_DATA)
 476                                 break;
 477
 478                         BUG_ON(whence != SEEK_HOLE);
 479                         /*
 480                          * If this is the first time to go into the loop and
 481                          * offset is not beyond the end offset, it will be a
 482                          * hole at this offset
 483                          */
 484                         if (lastoff == startoff || lastoff < endoff)
 485                                 found = 1;
 486                         break;
 487                 }
 488
 489                 /*
 490                  * If this is the first time to go into the loop and
 491                  * offset is smaller than the first page offset, it will be a
 492                  * hole at this offset.
 493                  */
 494                 if (lastoff == startoff && whence == SEEK_HOLE &&
 495                     lastoff < page_offset(pvec.pages[0])) {
 496                         found = 1;
 497                         break;
 498                 }
 499
 500                 for (i = 0; i < nr_pages; i++) {
 501                         struct page *page = pvec.pages[i];
 502                         struct buffer_head *bh, *head;
 503
 504                         /*
 505                          * If the current offset is not beyond the end of given
 506                          * range, it will be a hole.
 507                          */
 508                         if (lastoff < endoff && whence == SEEK_HOLE &&
 509                             page->index > end) {
 510                                 found = 1;
 511                                 *offset = lastoff;
 512                                 goto out;
 513                         }
 514
 515                         lock_page(page);
 516
 517                         if (unlikely(page->mapping != inode->i_mapping)) {
 518                                 unlock_page(page);
 519                                 continue;
 520                         }
 521
 522                         if (!page_has_buffers(page)) {
 523                                 unlock_page(page);
 524                                 continue;
 525                         }
 526
 527                         if (page_has_buffers(page)) {
 528                                 lastoff = page_offset(page);
 529                                 bh = head = page_buffers(page);
 530                                 do {
 531                                         if (buffer_uptodate(bh) ||
 532                                             buffer_unwritten(bh)) {
 533                                                 if (whence == SEEK_DATA)
 534                                                         found = 1;
 535                                         } else {
 536                                                 if (whence == SEEK_HOLE)
 537                                                         found = 1;
 538                                         }
 539                                         if (found) {
 540                                                 *offset = max_t(loff_t,
 541                                                         startoff, lastoff);
 542                                                 unlock_page(page);
 543                                                 goto out;
 544                                         }
 545                                         lastoff += bh->b_size;
 546                                         bh = bh->b_this_page;
 547                                 } while (bh != head);
 548                         }
 549
 550                         lastoff = page_offset(page) + PAGE_SIZE;
 551                         unlock_page(page);
 552                 }
 553
 554                 /*
 555                  * The no. of pages is less than our desired, that would be a
 556                  * hole in there.
 557                  */
 558                 if (nr_pages < num && whence == SEEK_HOLE) {
 559                         found = 1;
 560                         *offset = lastoff;
 561                         break;
 562                 }
 563
 564                 index = pvec.pages[i - 1]->index + 1;
 565                 pagevec_release(&pvec);
 566         } while (index <= end);
 567
 568 out:
 569         pagevec_release(&pvec);
 570         return found;
 571 }
 572
 573 /*
 574  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 575  */
 576 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 577 {
 578         struct inode *inode = file->f_mapping->host;
 579         struct extent_status es;
 580         ext4_lblk_t start, last, end;
 581         loff_t dataoff, isize;
 582         int blkbits;
 583         int ret;
 584
 585         inode_lock(inode);
 586
 587         isize = i_size_read(inode);
 588         if (offset >= isize) {
 589                 inode_unlock(inode);
 590                 return -ENXIO;
 591         }
 592
 593         blkbits = inode->i_sb->s_blocksize_bits;
 594         start = offset >> blkbits;
 595         last = start;
 596         end = isize >> blkbits;
 597         dataoff = offset;
 598
 599         do {
 600                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 601                 if (ret <= 0) {
 602                         /* No extent found -> no data */
 603                         if (ret == 0)
 604                                 ret = -ENXIO;
 605                         inode_unlock(inode);
 606                         return ret;
 607                 }
 608
 609                 last = es.es_lblk;
 610                 if (last != start)
 611                         dataoff = (loff_t)last << blkbits;
 612                 if (!ext4_es_is_unwritten(&es))
 613                         break;
 614
 615                 /*
 616                  * If there is a unwritten extent at this offset,
 617                  * it will be as a data or a hole according to page
 618                  * cache that has data or not.
 619                  */
 620                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 621                                               es.es_lblk + es.es_len, &dataoff))
 622                         break;
 623                 last += es.es_len;
 624                 dataoff = (loff_t)last << blkbits;
 625                 cond_resched();
 626         } while (last <= end);
 627
 628         inode_unlock(inode);
 629
 630         if (dataoff > isize)
 631                 return -ENXIO;
 632
 633         return vfs_setpos(file, dataoff, maxsize);
 634 }
 635
 636 /*
 637  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 638  */
 639 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 640 {
 641         struct inode *inode = file->f_mapping->host;
 642         struct extent_status es;
 643         ext4_lblk_t start, last, end;
 644         loff_t holeoff, isize;
 645         int blkbits;
 646         int ret;
 647
 648         inode_lock(inode);
 649
 650         isize = i_size_read(inode);
 651         if (offset >= isize) {
 652                 inode_unlock(inode);
 653                 return -ENXIO;
 654         }
 655
 656         blkbits = inode->i_sb->s_blocksize_bits;
 657         start = offset >> blkbits;
 658         last = start;
 659         end = isize >> blkbits;
 660         holeoff = offset;
 661
 662         do {
 663                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 664                 if (ret < 0) {
 665                         inode_unlock(inode);
 666                         return ret;
 667                 }
 668                 /* Found a hole? */
 669                 if (ret == 0 || es.es_lblk > last) {
 670                         if (last != start)
 671                                 holeoff = (loff_t)last << blkbits;
 672                         break;
 673                 }
 674                 /*
 675                  * If there is a unwritten extent at this offset,
 676                  * it will be as a data or a hole according to page
 677                  * cache that has data or not.
 678                  */
 679                 if (ext4_es_is_unwritten(&es) &&
 680                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 681                                               last + es.es_len, &holeoff))
 682                         break;
 683
 684                 last += es.es_len;
 685                 holeoff = (loff_t)last << blkbits;
 686                 cond_resched();
 687         } while (last <= end);
 688
 689         inode_unlock(inode);
 690
 691         if (holeoff > isize)
 692                 holeoff = isize;
 693
 694         return vfs_setpos(file, holeoff, maxsize);
 695 }
 696
 697 /*
 698  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 699  * by calling generic_file_llseek_size() with the appropriate maxbytes
 700  * value for each.
 701  */
 702 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 703 {
 704         struct inode *inode = file->f_mapping->host;
 705         loff_t maxbytes;
 706
 707         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 708                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 709         else
 710                 maxbytes = inode->i_sb->s_maxbytes;
 711
 712         switch (whence) {
 713         case SEEK_SET:
 714         case SEEK_CUR:
 715         case SEEK_END:
 716                 return generic_file_llseek_size(file, offset, whence,
 717                                                 maxbytes, i_size_read(inode));
 718         case SEEK_DATA:
 719                 return ext4_seek_data(file, offset, maxbytes);
 720         case SEEK_HOLE:
 721                 return ext4_seek_hole(file, offset, maxbytes);
 722         }
 723
 724         return -EINVAL;
 725 }
 726
 727 const struct file_operations ext4_file_operations = {
 728         .llseek         = ext4_llseek,
 729         .read_iter      = ext4_file_read_iter,
 730         .write_iter     = ext4_file_write_iter,
 731         .unlocked_ioctl = ext4_ioctl,
 732 #ifdef CONFIG_COMPAT
 733         .compat_ioctl   = ext4_compat_ioctl,
 734 #endif
 735         .mmap           = ext4_file_mmap,
 736         .open           = ext4_file_open,
 737         .release        = ext4_release_file,
 738         .fsync          = ext4_sync_file,
 739         .get_unmapped_area = thp_get_unmapped_area,
 740         .splice_read    = generic_file_splice_read,
 741         .splice_write   = iter_file_splice_write,
 742         .fallocate      = ext4_fallocate,
 743 };
 744
 745 const struct inode_operations ext4_file_inode_operations = {
 746         .setattr        = ext4_setattr,
 747         .getattr        = ext4_file_getattr,
 748         .listxattr      = ext4_listxattr,
 749         .get_acl        = ext4_get_acl,
 750         .set_acl        = ext4_set_acl,
 751         .fiemap         = ext4_fiemap,
 752 };
 753