fs/ext4/file.c

   1 /*
   2  *  linux/fs/ext4/file.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/file.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  ext4 fs regular file handling primitives
  16  *
  17  *  64-bit file support on 64-bit platforms by Jakub Jelinek
  18  *      (jj@sunsite.ms.mff.cuni.cz)
  19  */
  20
  21 #include <linux/time.h>
  22 #include <linux/fs.h>
  23 #include <linux/mount.h>
  24 #include <linux/path.h>
  25 #include <linux/dax.h>
  26 #include <linux/quotaops.h>
  27 #include <linux/pagevec.h>
  28 #include <linux/uio.h>
  29 #include "ext4.h"
  30 #include "ext4_jbd2.h"
  31 #include "xattr.h"
  32 #include "acl.h"
  33
  34 /*
  35  * Called when an inode is released. Note that this is different
  36  * from ext4_file_open: open gets called at every open, but release
  37  * gets called only when /all/ the files are closed.
  38  */
  39 static int ext4_release_file(struct inode *inode, struct file *filp)
  40 {
  41         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
  42                 ext4_alloc_da_blocks(inode);
  43                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
  44         }
  45         /* if we are the last writer on the inode, drop the block reservation */
  46         if ((filp->f_mode & FMODE_WRITE) &&
  47                         (atomic_read(&inode->i_writecount) == 1) &&
  48                         !EXT4_I(inode)->i_reserved_data_blocks)
  49         {
  50                 down_write(&EXT4_I(inode)->i_data_sem);
  51                 ext4_discard_preallocations(inode);
  52                 up_write(&EXT4_I(inode)->i_data_sem);
  53         }
  54         if (is_dx(inode) && filp->private_data)
  55                 ext4_htree_free_dir_info(filp->private_data);
  56
  57         return 0;
  58 }
  59
  60 static void ext4_unwritten_wait(struct inode *inode)
  61 {
  62         wait_queue_head_t *wq = ext4_ioend_wq(inode);
  63
  64         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
  65 }
  66
  67 /*
  68  * This tests whether the IO in question is block-aligned or not.
  69  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
  70  * are converted to written only after the IO is complete.  Until they are
  71  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
  72  * it needs to zero out portions of the start and/or end block.  If 2 AIO
  73  * threads are at work on the same unwritten block, they must be synchronized
  74  * or one thread will zero the other's data, causing corruption.
  75  */
  76 static int
  77 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
  78 {
  79         struct super_block *sb = inode->i_sb;
  80         int blockmask = sb->s_blocksize - 1;
  81
  82         if (pos >= i_size_read(inode))
  83                 return 0;
  84
  85         if ((pos | iov_iter_alignment(from)) & blockmask)
  86                 return 1;
  87
  88         return 0;
  89 }
  90
  91 static ssize_t
  92 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  93 {
  94         struct file *file = iocb->ki_filp;
  95         struct inode *inode = file_inode(iocb->ki_filp);
  96         struct mutex *aio_mutex = NULL;
  97         struct blk_plug plug;
  98         int o_direct = iocb->ki_flags & IOCB_DIRECT;
  99         int overwrite = 0;
 100         ssize_t ret;
 101
 102         /*
 103          * Unaligned direct AIO must be serialized; see comment above
 104          * In the case of O_APPEND, assume that we must always serialize
 105          */
 106         if (o_direct &&
 107             ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 108             !is_sync_kiocb(iocb) &&
 109             (iocb->ki_flags & IOCB_APPEND ||
 110              ext4_unaligned_aio(inode, from, iocb->ki_pos))) {
 111                 aio_mutex = ext4_aio_mutex(inode);
 112                 mutex_lock(aio_mutex);
 113                 ext4_unwritten_wait(inode);
 114         }
 115
 116         inode_lock(inode);
 117         ret = generic_write_checks(iocb, from);
 118         if (ret <= 0)
 119                 goto out;
 120
 121         /*
 122          * If we have encountered a bitmap-format file, the size limit
 123          * is smaller than s_maxbytes, which is for extent-mapped files.
 124          */
 125         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 126                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 127
 128                 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
 129                         ret = -EFBIG;
 130                         goto out;
 131                 }
 132                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 133         }
 134
 135         iocb->private = &overwrite;
 136         if (o_direct) {
 137                 size_t length = iov_iter_count(from);
 138                 loff_t pos = iocb->ki_pos;
 139                 blk_start_plug(&plug);
 140
 141                 /* check whether we do a DIO overwrite or not */
 142                 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
 143                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 144                         struct ext4_map_blocks map;
 145                         unsigned int blkbits = inode->i_blkbits;
 146                         int err, len;
 147
 148                         map.m_lblk = pos >> blkbits;
 149                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 150                                 - map.m_lblk;
 151                         len = map.m_len;
 152
 153                         err = ext4_map_blocks(NULL, inode, &map, 0);
 154                         /*
 155                          * 'err==len' means that all of blocks has
 156                          * been preallocated no matter they are
 157                          * initialized or not.  For excluding
 158                          * unwritten extents, we need to check
 159                          * m_flags.  There are two conditions that
 160                          * indicate for initialized extents.  1) If we
 161                          * hit extent cache, EXT4_MAP_MAPPED flag is
 162                          * returned; 2) If we do a real lookup,
 163                          * non-flags are returned.  So we should check
 164                          * these two conditions.
 165                          */
 166                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 167                                 overwrite = 1;
 168                 }
 169         }
 170
 171         ret = __generic_file_write_iter(iocb, from);
 172         inode_unlock(inode);
 173
 174         if (ret > 0) {
 175                 ssize_t err;
 176
 177                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
 178                 if (err < 0)
 179                         ret = err;
 180         }
 181         if (o_direct)
 182                 blk_finish_plug(&plug);
 183
 184         if (aio_mutex)
 185                 mutex_unlock(aio_mutex);
 186         return ret;
 187
 188 out:
 189         inode_unlock(inode);
 190         if (aio_mutex)
 191                 mutex_unlock(aio_mutex);
 192         return ret;
 193 }
 194
 195 #ifdef CONFIG_FS_DAX
 196 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 197 {
 198         int result;
 199         handle_t *handle = NULL;
 200         struct inode *inode = file_inode(vma->vm_file);
 201         struct super_block *sb = inode->i_sb;
 202         bool write = vmf->flags & FAULT_FLAG_WRITE;
 203
 204         if (write) {
 205                 sb_start_pagefault(sb);
 206                 file_update_time(vma->vm_file);
 207                 down_read(&EXT4_I(inode)->i_mmap_sem);
 208                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 209                                                 EXT4_DATA_TRANS_BLOCKS(sb));
 210         } else
 211                 down_read(&EXT4_I(inode)->i_mmap_sem);
 212
 213         if (IS_ERR(handle))
 214                 result = VM_FAULT_SIGBUS;
 215         else
 216                 result = __dax_fault(vma, vmf, ext4_dax_mmap_get_block, NULL);
 217
 218         if (write) {
 219                 if (!IS_ERR(handle))
 220                         ext4_journal_stop(handle);
 221                 up_read(&EXT4_I(inode)->i_mmap_sem);
 222                 sb_end_pagefault(sb);
 223         } else
 224                 up_read(&EXT4_I(inode)->i_mmap_sem);
 225
 226         return result;
 227 }
 228
 229 static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
 230                                                 pmd_t *pmd, unsigned int flags)
 231 {
 232         int result;
 233         handle_t *handle = NULL;
 234         struct inode *inode = file_inode(vma->vm_file);
 235         struct super_block *sb = inode->i_sb;
 236         bool write = flags & FAULT_FLAG_WRITE;
 237
 238         if (write) {
 239                 sb_start_pagefault(sb);
 240                 file_update_time(vma->vm_file);
 241                 down_read(&EXT4_I(inode)->i_mmap_sem);
 242                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 243                                 ext4_chunk_trans_blocks(inode,
 244                                                         PMD_SIZE / PAGE_SIZE));
 245         } else
 246                 down_read(&EXT4_I(inode)->i_mmap_sem);
 247
 248         if (IS_ERR(handle))
 249                 result = VM_FAULT_SIGBUS;
 250         else
 251                 result = __dax_pmd_fault(vma, addr, pmd, flags,
 252                                 ext4_dax_mmap_get_block, NULL);
 253
 254         if (write) {
 255                 if (!IS_ERR(handle))
 256                         ext4_journal_stop(handle);
 257                 up_read(&EXT4_I(inode)->i_mmap_sem);
 258                 sb_end_pagefault(sb);
 259         } else
 260                 up_read(&EXT4_I(inode)->i_mmap_sem);
 261
 262         return result;
 263 }
 264
 265 /*
 266  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
 267  * handler we check for races agaist truncate. Note that since we cycle through
 268  * i_mmap_sem, we are sure that also any hole punching that began before we
 269  * were called is finished by now and so if it included part of the file we
 270  * are working on, our pte will get unmapped and the check for pte_same() in
 271  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 272  * desired.
 273  */
 274 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
 275                                 struct vm_fault *vmf)
 276 {
 277         struct inode *inode = file_inode(vma->vm_file);
 278         struct super_block *sb = inode->i_sb;
 279         loff_t size;
 280         int ret;
 281
 282         sb_start_pagefault(sb);
 283         file_update_time(vma->vm_file);
 284         down_read(&EXT4_I(inode)->i_mmap_sem);
 285         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 286         if (vmf->pgoff >= size)
 287                 ret = VM_FAULT_SIGBUS;
 288         else
 289                 ret = dax_pfn_mkwrite(vma, vmf);
 290         up_read(&EXT4_I(inode)->i_mmap_sem);
 291         sb_end_pagefault(sb);
 292
 293         return ret;
 294 }
 295
 296 static const struct vm_operations_struct ext4_dax_vm_ops = {
 297         .fault          = ext4_dax_fault,
 298         .pmd_fault      = ext4_dax_pmd_fault,
 299         .page_mkwrite   = ext4_dax_fault,
 300         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 301 };
 302 #else
 303 #define ext4_dax_vm_ops ext4_file_vm_ops
 304 #endif
 305
 306 static const struct vm_operations_struct ext4_file_vm_ops = {
 307         .fault          = ext4_filemap_fault,
 308         .map_pages      = filemap_map_pages,
 309         .page_mkwrite   = ext4_page_mkwrite,
 310 };
 311
 312 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 313 {
 314         struct inode *inode = file->f_mapping->host;
 315
 316         if (ext4_encrypted_inode(inode)) {
 317                 int err = ext4_get_encryption_info(inode);
 318                 if (err)
 319                         return 0;
 320                 if (ext4_encryption_info(inode) == NULL)
 321                         return -ENOKEY;
 322         }
 323         file_accessed(file);
 324         if (IS_DAX(file_inode(file))) {
 325                 vma->vm_ops = &ext4_dax_vm_ops;
 326                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 327         } else {
 328                 vma->vm_ops = &ext4_file_vm_ops;
 329         }
 330         return 0;
 331 }
 332
 333 static int ext4_file_open(struct inode * inode, struct file * filp)
 334 {
 335         struct super_block *sb = inode->i_sb;
 336         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 337         struct vfsmount *mnt = filp->f_path.mnt;
 338         struct inode *dir = filp->f_path.dentry->d_parent->d_inode;
 339         struct path path;
 340         char buf[64], *cp;
 341         int ret;
 342
 343         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 344                      !(sb->s_flags & MS_RDONLY))) {
 345                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 346                 /*
 347                  * Sample where the filesystem has been mounted and
 348                  * store it in the superblock for sysadmin convenience
 349                  * when trying to sort through large numbers of block
 350                  * devices or filesystem images.
 351                  */
 352                 memset(buf, 0, sizeof(buf));
 353                 path.mnt = mnt;
 354                 path.dentry = mnt->mnt_root;
 355                 cp = d_path(&path, buf, sizeof(buf));
 356                 if (!IS_ERR(cp)) {
 357                         handle_t *handle;
 358                         int err;
 359
 360                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 361                         if (IS_ERR(handle))
 362                                 return PTR_ERR(handle);
 363                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 364                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 365                         if (err) {
 366                                 ext4_journal_stop(handle);
 367                                 return err;
 368                         }
 369                         strlcpy(sbi->s_es->s_last_mounted, cp,
 370                                 sizeof(sbi->s_es->s_last_mounted));
 371                         ext4_handle_dirty_super(handle, sb);
 372                         ext4_journal_stop(handle);
 373                 }
 374         }
 375         if (ext4_encrypted_inode(inode)) {
 376                 ret = ext4_get_encryption_info(inode);
 377                 if (ret)
 378                         return -EACCES;
 379                 if (ext4_encryption_info(inode) == NULL)
 380                         return -ENOKEY;
 381         }
 382         if (ext4_encrypted_inode(dir) &&
 383             !ext4_is_child_context_consistent_with_parent(dir, inode)) {
 384                 ext4_warning(inode->i_sb,
 385                              "Inconsistent encryption contexts: %lu/%lu\n",
 386                              (unsigned long) dir->i_ino,
 387                              (unsigned long) inode->i_ino);
 388                 return -EPERM;
 389         }
 390         /*
 391          * Set up the jbd2_inode if we are opening the inode for
 392          * writing and the journal is present
 393          */
 394         if (filp->f_mode & FMODE_WRITE) {
 395                 ret = ext4_inode_attach_jinode(inode);
 396                 if (ret < 0)
 397                         return ret;
 398         }
 399         return dquot_file_open(inode, filp);
 400 }
 401
 402 /*
 403  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 404  * file rather than ext4_ext_walk_space() because we can introduce
 405  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 406  * function.  When extent status tree has been fully implemented, it will
 407  * track all extent status for a file and we can directly use it to
 408  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 409  */
 410
 411 /*
 412  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 413  * lookup page cache to check whether or not there has some data between
 414  * [startoff, endoff] because, if this range contains an unwritten extent,
 415  * we determine this extent as a data or a hole according to whether the
 416  * page cache has data or not.
 417  */
 418 static int ext4_find_unwritten_pgoff(struct inode *inode,
 419                                      int whence,
 420                                      struct ext4_map_blocks *map,
 421                                      loff_t *offset)
 422 {
 423         struct pagevec pvec;
 424         unsigned int blkbits;
 425         pgoff_t index;
 426         pgoff_t end;
 427         loff_t endoff;
 428         loff_t startoff;
 429         loff_t lastoff;
 430         int found = 0;
 431
 432         blkbits = inode->i_sb->s_blocksize_bits;
 433         startoff = *offset;
 434         lastoff = startoff;
 435         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 436
 437         index = startoff >> PAGE_CACHE_SHIFT;
 438         end = endoff >> PAGE_CACHE_SHIFT;
 439
 440         pagevec_init(&pvec, 0);
 441         do {
 442                 int i, num;
 443                 unsigned long nr_pages;
 444
 445                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 446                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 447                                           (pgoff_t)num);
 448                 if (nr_pages == 0) {
 449                         if (whence == SEEK_DATA)
 450                                 break;
 451
 452                         BUG_ON(whence != SEEK_HOLE);
 453                         /*
 454                          * If this is the first time to go into the loop and
 455                          * offset is not beyond the end offset, it will be a
 456                          * hole at this offset
 457                          */
 458                         if (lastoff == startoff || lastoff < endoff)
 459                                 found = 1;
 460                         break;
 461                 }
 462
 463                 /*
 464                  * If this is the first time to go into the loop and
 465                  * offset is smaller than the first page offset, it will be a
 466                  * hole at this offset.
 467                  */
 468                 if (lastoff == startoff && whence == SEEK_HOLE &&
 469                     lastoff < page_offset(pvec.pages[0])) {
 470                         found = 1;
 471                         break;
 472                 }
 473
 474                 for (i = 0; i < nr_pages; i++) {
 475                         struct page *page = pvec.pages[i];
 476                         struct buffer_head *bh, *head;
 477
 478                         /*
 479                          * If the current offset is not beyond the end of given
 480                          * range, it will be a hole.
 481                          */
 482                         if (lastoff < endoff && whence == SEEK_HOLE &&
 483                             page->index > end) {
 484                                 found = 1;
 485                                 *offset = lastoff;
 486                                 goto out;
 487                         }
 488
 489                         lock_page(page);
 490
 491                         if (unlikely(page->mapping != inode->i_mapping)) {
 492                                 unlock_page(page);
 493                                 continue;
 494                         }
 495
 496                         if (!page_has_buffers(page)) {
 497                                 unlock_page(page);
 498                                 continue;
 499                         }
 500
 501                         if (page_has_buffers(page)) {
 502                                 lastoff = page_offset(page);
 503                                 bh = head = page_buffers(page);
 504                                 do {
 505                                         if (buffer_uptodate(bh) ||
 506                                             buffer_unwritten(bh)) {
 507                                                 if (whence == SEEK_DATA)
 508                                                         found = 1;
 509                                         } else {
 510                                                 if (whence == SEEK_HOLE)
 511                                                         found = 1;
 512                                         }
 513                                         if (found) {
 514                                                 *offset = max_t(loff_t,
 515                                                         startoff, lastoff);
 516                                                 unlock_page(page);
 517                                                 goto out;
 518                                         }
 519                                         lastoff += bh->b_size;
 520                                         bh = bh->b_this_page;
 521                                 } while (bh != head);
 522                         }
 523
 524                         lastoff = page_offset(page) + PAGE_SIZE;
 525                         unlock_page(page);
 526                 }
 527
 528                 /*
 529                  * The no. of pages is less than our desired, that would be a
 530                  * hole in there.
 531                  */
 532                 if (nr_pages < num && whence == SEEK_HOLE) {
 533                         found = 1;
 534                         *offset = lastoff;
 535                         break;
 536                 }
 537
 538                 index = pvec.pages[i - 1]->index + 1;
 539                 pagevec_release(&pvec);
 540         } while (index <= end);
 541
 542 out:
 543         pagevec_release(&pvec);
 544         return found;
 545 }
 546
 547 /*
 548  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 549  */
 550 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 551 {
 552         struct inode *inode = file->f_mapping->host;
 553         struct ext4_map_blocks map;
 554         struct extent_status es;
 555         ext4_lblk_t start, last, end;
 556         loff_t dataoff, isize;
 557         int blkbits;
 558         int ret = 0;
 559
 560         inode_lock(inode);
 561
 562         isize = i_size_read(inode);
 563         if (offset >= isize) {
 564                 inode_unlock(inode);
 565                 return -ENXIO;
 566         }
 567
 568         blkbits = inode->i_sb->s_blocksize_bits;
 569         start = offset >> blkbits;
 570         last = start;
 571         end = isize >> blkbits;
 572         dataoff = offset;
 573
 574         do {
 575                 map.m_lblk = last;
 576                 map.m_len = end - last + 1;
 577                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 578                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 579                         if (last != start)
 580                                 dataoff = (loff_t)last << blkbits;
 581                         break;
 582                 }
 583
 584                 /*
 585                  * If there is a delay extent at this offset,
 586                  * it will be as a data.
 587                  */
 588                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 589                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 590                         if (last != start)
 591                                 dataoff = (loff_t)last << blkbits;
 592                         break;
 593                 }
 594
 595                 /*
 596                  * If there is a unwritten extent at this offset,
 597                  * it will be as a data or a hole according to page
 598                  * cache that has data or not.
 599                  */
 600                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 601                         int unwritten;
 602                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 603                                                               &map, &dataoff);
 604                         if (unwritten)
 605                                 break;
 606                 }
 607
 608                 last++;
 609                 dataoff = (loff_t)last << blkbits;
 610         } while (last <= end);
 611
 612         inode_unlock(inode);
 613
 614         if (dataoff > isize)
 615                 return -ENXIO;
 616
 617         return vfs_setpos(file, dataoff, maxsize);
 618 }
 619
 620 /*
 621  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 622  */
 623 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 624 {
 625         struct inode *inode = file->f_mapping->host;
 626         struct ext4_map_blocks map;
 627         struct extent_status es;
 628         ext4_lblk_t start, last, end;
 629         loff_t holeoff, isize;
 630         int blkbits;
 631         int ret = 0;
 632
 633         inode_lock(inode);
 634
 635         isize = i_size_read(inode);
 636         if (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                 map.m_lblk = last;
 649                 map.m_len = end - last + 1;
 650                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 651                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 652                         last += ret;
 653                         holeoff = (loff_t)last << blkbits;
 654                         continue;
 655                 }
 656
 657                 /*
 658                  * If there is a delay extent at this offset,
 659                  * we will skip this extent.
 660                  */
 661                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 662                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 663                         last = es.es_lblk + es.es_len;
 664                         holeoff = (loff_t)last << blkbits;
 665                         continue;
 666                 }
 667
 668                 /*
 669                  * If there is a unwritten extent at this offset,
 670                  * it will be as a data or a hole according to page
 671                  * cache that has data or not.
 672                  */
 673                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 674                         int unwritten;
 675                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 676                                                               &map, &holeoff);
 677                         if (!unwritten) {
 678                                 last += ret;
 679                                 holeoff = (loff_t)last << blkbits;
 680                                 continue;
 681                         }
 682                 }
 683
 684                 /* find a hole */
 685                 break;
 686         } while (last <= end);
 687
 688         inode_unlock(inode);
 689
 690         if (holeoff > isize)
 691                 holeoff = isize;
 692
 693         return vfs_setpos(file, holeoff, maxsize);
 694 }
 695
 696 /*
 697  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 698  * by calling generic_file_llseek_size() with the appropriate maxbytes
 699  * value for each.
 700  */
 701 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 702 {
 703         struct inode *inode = file->f_mapping->host;
 704         loff_t maxbytes;
 705
 706         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 707                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 708         else
 709                 maxbytes = inode->i_sb->s_maxbytes;
 710
 711         switch (whence) {
 712         case SEEK_SET:
 713         case SEEK_CUR:
 714         case SEEK_END:
 715                 return generic_file_llseek_size(file, offset, whence,
 716                                                 maxbytes, i_size_read(inode));
 717         case SEEK_DATA:
 718                 return ext4_seek_data(file, offset, maxbytes);
 719         case SEEK_HOLE:
 720                 return ext4_seek_hole(file, offset, maxbytes);
 721         }
 722
 723         return -EINVAL;
 724 }
 725
 726 const struct file_operations ext4_file_operations = {
 727         .llseek         = ext4_llseek,
 728         .read_iter      = generic_file_read_iter,
 729         .write_iter     = ext4_file_write_iter,
 730         .unlocked_ioctl = ext4_ioctl,
 731 #ifdef CONFIG_COMPAT
 732         .compat_ioctl   = ext4_compat_ioctl,
 733 #endif
 734         .mmap           = ext4_file_mmap,
 735         .open           = ext4_file_open,
 736         .release        = ext4_release_file,
 737         .fsync          = ext4_sync_file,
 738         .splice_read    = generic_file_splice_read,
 739         .splice_write   = iter_file_splice_write,
 740         .fallocate      = ext4_fallocate,
 741 };
 742
 743 const struct inode_operations ext4_file_inode_operations = {
 744         .setattr        = ext4_setattr,
 745         .getattr        = ext4_getattr,
 746         .setxattr       = generic_setxattr,
 747         .getxattr       = generic_getxattr,
 748         .listxattr      = ext4_listxattr,
 749         .removexattr    = generic_removexattr,
 750         .get_acl        = ext4_get_acl,
 751         .set_acl        = ext4_set_acl,
 752         .fiemap         = ext4_fiemap,
 753 };
 754