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         mutex_lock(&inode->i_mutex);
 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         mutex_unlock(&inode->i_mutex);
 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         mutex_unlock(&inode->i_mutex);
 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 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 266 {
 267         int err;
 268         struct inode *inode = file_inode(vma->vm_file);
 269
 270         sb_start_pagefault(inode->i_sb);
 271         file_update_time(vma->vm_file);
 272         down_read(&EXT4_I(inode)->i_mmap_sem);
 273         err = __dax_mkwrite(vma, vmf, ext4_dax_mmap_get_block, NULL);
 274         up_read(&EXT4_I(inode)->i_mmap_sem);
 275         sb_end_pagefault(inode->i_sb);
 276
 277         return err;
 278 }
 279
 280 /*
 281  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_mkwrite()
 282  * handler we check for races agaist truncate. Note that since we cycle through
 283  * i_mmap_sem, we are sure that also any hole punching that began before we
 284  * were called is finished by now and so if it included part of the file we
 285  * are working on, our pte will get unmapped and the check for pte_same() in
 286  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 287  * desired.
 288  */
 289 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
 290                                 struct vm_fault *vmf)
 291 {
 292         struct inode *inode = file_inode(vma->vm_file);
 293         struct super_block *sb = inode->i_sb;
 294         int ret = VM_FAULT_NOPAGE;
 295         loff_t size;
 296
 297         sb_start_pagefault(sb);
 298         file_update_time(vma->vm_file);
 299         down_read(&EXT4_I(inode)->i_mmap_sem);
 300         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 301         if (vmf->pgoff >= size)
 302                 ret = VM_FAULT_SIGBUS;
 303         up_read(&EXT4_I(inode)->i_mmap_sem);
 304         sb_end_pagefault(sb);
 305
 306         return ret;
 307 }
 308
 309 static const struct vm_operations_struct ext4_dax_vm_ops = {
 310         .fault          = ext4_dax_fault,
 311         .pmd_fault      = ext4_dax_pmd_fault,
 312         .page_mkwrite   = ext4_dax_mkwrite,
 313         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 314 };
 315 #else
 316 #define ext4_dax_vm_ops ext4_file_vm_ops
 317 #endif
 318
 319 static const struct vm_operations_struct ext4_file_vm_ops = {
 320         .fault          = ext4_filemap_fault,
 321         .map_pages      = filemap_map_pages,
 322         .page_mkwrite   = ext4_page_mkwrite,
 323 };
 324
 325 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 326 {
 327         struct inode *inode = file->f_mapping->host;
 328
 329         if (ext4_encrypted_inode(inode)) {
 330                 int err = ext4_get_encryption_info(inode);
 331                 if (err)
 332                         return 0;
 333                 if (ext4_encryption_info(inode) == NULL)
 334                         return -ENOKEY;
 335         }
 336         file_accessed(file);
 337         if (IS_DAX(file_inode(file))) {
 338                 vma->vm_ops = &ext4_dax_vm_ops;
 339                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 340         } else {
 341                 vma->vm_ops = &ext4_file_vm_ops;
 342         }
 343         return 0;
 344 }
 345
 346 static int ext4_file_open(struct inode * inode, struct file * filp)
 347 {
 348         struct super_block *sb = inode->i_sb;
 349         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 350         struct vfsmount *mnt = filp->f_path.mnt;
 351         struct path path;
 352         char buf[64], *cp;
 353         int ret;
 354
 355         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 356                      !(sb->s_flags & MS_RDONLY))) {
 357                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 358                 /*
 359                  * Sample where the filesystem has been mounted and
 360                  * store it in the superblock for sysadmin convenience
 361                  * when trying to sort through large numbers of block
 362                  * devices or filesystem images.
 363                  */
 364                 memset(buf, 0, sizeof(buf));
 365                 path.mnt = mnt;
 366                 path.dentry = mnt->mnt_root;
 367                 cp = d_path(&path, buf, sizeof(buf));
 368                 if (!IS_ERR(cp)) {
 369                         handle_t *handle;
 370                         int err;
 371
 372                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 373                         if (IS_ERR(handle))
 374                                 return PTR_ERR(handle);
 375                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 376                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 377                         if (err) {
 378                                 ext4_journal_stop(handle);
 379                                 return err;
 380                         }
 381                         strlcpy(sbi->s_es->s_last_mounted, cp,
 382                                 sizeof(sbi->s_es->s_last_mounted));
 383                         ext4_handle_dirty_super(handle, sb);
 384                         ext4_journal_stop(handle);
 385                 }
 386         }
 387         if (ext4_encrypted_inode(inode)) {
 388                 ret = ext4_get_encryption_info(inode);
 389                 if (ret)
 390                         return -EACCES;
 391                 if (ext4_encryption_info(inode) == NULL)
 392                         return -ENOKEY;
 393         }
 394         /*
 395          * Set up the jbd2_inode if we are opening the inode for
 396          * writing and the journal is present
 397          */
 398         if (filp->f_mode & FMODE_WRITE) {
 399                 ret = ext4_inode_attach_jinode(inode);
 400                 if (ret < 0)
 401                         return ret;
 402         }
 403         return dquot_file_open(inode, filp);
 404 }
 405
 406 /*
 407  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 408  * file rather than ext4_ext_walk_space() because we can introduce
 409  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 410  * function.  When extent status tree has been fully implemented, it will
 411  * track all extent status for a file and we can directly use it to
 412  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 413  */
 414
 415 /*
 416  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 417  * lookup page cache to check whether or not there has some data between
 418  * [startoff, endoff] because, if this range contains an unwritten extent,
 419  * we determine this extent as a data or a hole according to whether the
 420  * page cache has data or not.
 421  */
 422 static int ext4_find_unwritten_pgoff(struct inode *inode,
 423                                      int whence,
 424                                      struct ext4_map_blocks *map,
 425                                      loff_t *offset)
 426 {
 427         struct pagevec pvec;
 428         unsigned int blkbits;
 429         pgoff_t index;
 430         pgoff_t end;
 431         loff_t endoff;
 432         loff_t startoff;
 433         loff_t lastoff;
 434         int found = 0;
 435
 436         blkbits = inode->i_sb->s_blocksize_bits;
 437         startoff = *offset;
 438         lastoff = startoff;
 439         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 440
 441         index = startoff >> PAGE_CACHE_SHIFT;
 442         end = endoff >> PAGE_CACHE_SHIFT;
 443
 444         pagevec_init(&pvec, 0);
 445         do {
 446                 int i, num;
 447                 unsigned long nr_pages;
 448
 449                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 450                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 451                                           (pgoff_t)num);
 452                 if (nr_pages == 0) {
 453                         if (whence == SEEK_DATA)
 454                                 break;
 455
 456                         BUG_ON(whence != SEEK_HOLE);
 457                         /*
 458                          * If this is the first time to go into the loop and
 459                          * offset is not beyond the end offset, it will be a
 460                          * hole at this offset
 461                          */
 462                         if (lastoff == startoff || lastoff < endoff)
 463                                 found = 1;
 464                         break;
 465                 }
 466
 467                 /*
 468                  * If this is the first time to go into the loop and
 469                  * offset is smaller than the first page offset, it will be a
 470                  * hole at this offset.
 471                  */
 472                 if (lastoff == startoff && whence == SEEK_HOLE &&
 473                     lastoff < page_offset(pvec.pages[0])) {
 474                         found = 1;
 475                         break;
 476                 }
 477
 478                 for (i = 0; i < nr_pages; i++) {
 479                         struct page *page = pvec.pages[i];
 480                         struct buffer_head *bh, *head;
 481
 482                         /*
 483                          * If the current offset is not beyond the end of given
 484                          * range, it will be a hole.
 485                          */
 486                         if (lastoff < endoff && whence == SEEK_HOLE &&
 487                             page->index > end) {
 488                                 found = 1;
 489                                 *offset = lastoff;
 490                                 goto out;
 491                         }
 492
 493                         lock_page(page);
 494
 495                         if (unlikely(page->mapping != inode->i_mapping)) {
 496                                 unlock_page(page);
 497                                 continue;
 498                         }
 499
 500                         if (!page_has_buffers(page)) {
 501                                 unlock_page(page);
 502                                 continue;
 503                         }
 504
 505                         if (page_has_buffers(page)) {
 506                                 lastoff = page_offset(page);
 507                                 bh = head = page_buffers(page);
 508                                 do {
 509                                         if (buffer_uptodate(bh) ||
 510                                             buffer_unwritten(bh)) {
 511                                                 if (whence == SEEK_DATA)
 512                                                         found = 1;
 513                                         } else {
 514                                                 if (whence == SEEK_HOLE)
 515                                                         found = 1;
 516                                         }
 517                                         if (found) {
 518                                                 *offset = max_t(loff_t,
 519                                                         startoff, lastoff);
 520                                                 unlock_page(page);
 521                                                 goto out;
 522                                         }
 523                                         lastoff += bh->b_size;
 524                                         bh = bh->b_this_page;
 525                                 } while (bh != head);
 526                         }
 527
 528                         lastoff = page_offset(page) + PAGE_SIZE;
 529                         unlock_page(page);
 530                 }
 531
 532                 /*
 533                  * The no. of pages is less than our desired, that would be a
 534                  * hole in there.
 535                  */
 536                 if (nr_pages < num && whence == SEEK_HOLE) {
 537                         found = 1;
 538                         *offset = lastoff;
 539                         break;
 540                 }
 541
 542                 index = pvec.pages[i - 1]->index + 1;
 543                 pagevec_release(&pvec);
 544         } while (index <= end);
 545
 546 out:
 547         pagevec_release(&pvec);
 548         return found;
 549 }
 550
 551 /*
 552  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 553  */
 554 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 555 {
 556         struct inode *inode = file->f_mapping->host;
 557         struct ext4_map_blocks map;
 558         struct extent_status es;
 559         ext4_lblk_t start, last, end;
 560         loff_t dataoff, isize;
 561         int blkbits;
 562         int ret = 0;
 563
 564         mutex_lock(&inode->i_mutex);
 565
 566         isize = i_size_read(inode);
 567         if (offset >= isize) {
 568                 mutex_unlock(&inode->i_mutex);
 569                 return -ENXIO;
 570         }
 571
 572         blkbits = inode->i_sb->s_blocksize_bits;
 573         start = offset >> blkbits;
 574         last = start;
 575         end = isize >> blkbits;
 576         dataoff = offset;
 577
 578         do {
 579                 map.m_lblk = last;
 580                 map.m_len = end - last + 1;
 581                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 582                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 583                         if (last != start)
 584                                 dataoff = (loff_t)last << blkbits;
 585                         break;
 586                 }
 587
 588                 /*
 589                  * If there is a delay extent at this offset,
 590                  * it will be as a data.
 591                  */
 592                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 593                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 594                         if (last != start)
 595                                 dataoff = (loff_t)last << blkbits;
 596                         break;
 597                 }
 598
 599                 /*
 600                  * If there is a unwritten extent at this offset,
 601                  * it will be as a data or a hole according to page
 602                  * cache that has data or not.
 603                  */
 604                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 605                         int unwritten;
 606                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 607                                                               &map, &dataoff);
 608                         if (unwritten)
 609                                 break;
 610                 }
 611
 612                 last++;
 613                 dataoff = (loff_t)last << blkbits;
 614         } while (last <= end);
 615
 616         mutex_unlock(&inode->i_mutex);
 617
 618         if (dataoff > isize)
 619                 return -ENXIO;
 620
 621         return vfs_setpos(file, dataoff, maxsize);
 622 }
 623
 624 /*
 625  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 626  */
 627 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 628 {
 629         struct inode *inode = file->f_mapping->host;
 630         struct ext4_map_blocks map;
 631         struct extent_status es;
 632         ext4_lblk_t start, last, end;
 633         loff_t holeoff, isize;
 634         int blkbits;
 635         int ret = 0;
 636
 637         mutex_lock(&inode->i_mutex);
 638
 639         isize = i_size_read(inode);
 640         if (offset >= isize) {
 641                 mutex_unlock(&inode->i_mutex);
 642                 return -ENXIO;
 643         }
 644
 645         blkbits = inode->i_sb->s_blocksize_bits;
 646         start = offset >> blkbits;
 647         last = start;
 648         end = isize >> blkbits;
 649         holeoff = offset;
 650
 651         do {
 652                 map.m_lblk = last;
 653                 map.m_len = end - last + 1;
 654                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 655                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 656                         last += ret;
 657                         holeoff = (loff_t)last << blkbits;
 658                         continue;
 659                 }
 660
 661                 /*
 662                  * If there is a delay extent at this offset,
 663                  * we will skip this extent.
 664                  */
 665                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 666                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 667                         last = es.es_lblk + es.es_len;
 668                         holeoff = (loff_t)last << blkbits;
 669                         continue;
 670                 }
 671
 672                 /*
 673                  * If there is a unwritten extent at this offset,
 674                  * it will be as a data or a hole according to page
 675                  * cache that has data or not.
 676                  */
 677                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 678                         int unwritten;
 679                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 680                                                               &map, &holeoff);
 681                         if (!unwritten) {
 682                                 last += ret;
 683                                 holeoff = (loff_t)last << blkbits;
 684                                 continue;
 685                         }
 686                 }
 687
 688                 /* find a hole */
 689                 break;
 690         } while (last <= end);
 691
 692         mutex_unlock(&inode->i_mutex);
 693
 694         if (holeoff > isize)
 695                 holeoff = isize;
 696
 697         return vfs_setpos(file, holeoff, maxsize);
 698 }
 699
 700 /*
 701  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 702  * by calling generic_file_llseek_size() with the appropriate maxbytes
 703  * value for each.
 704  */
 705 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 706 {
 707         struct inode *inode = file->f_mapping->host;
 708         loff_t maxbytes;
 709
 710         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 711                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 712         else
 713                 maxbytes = inode->i_sb->s_maxbytes;
 714
 715         switch (whence) {
 716         case SEEK_SET:
 717         case SEEK_CUR:
 718         case SEEK_END:
 719                 return generic_file_llseek_size(file, offset, whence,
 720                                                 maxbytes, i_size_read(inode));
 721         case SEEK_DATA:
 722                 return ext4_seek_data(file, offset, maxbytes);
 723         case SEEK_HOLE:
 724                 return ext4_seek_hole(file, offset, maxbytes);
 725         }
 726
 727         return -EINVAL;
 728 }
 729
 730 const struct file_operations ext4_file_operations = {
 731         .llseek         = ext4_llseek,
 732         .read_iter      = generic_file_read_iter,
 733         .write_iter     = ext4_file_write_iter,
 734         .unlocked_ioctl = ext4_ioctl,
 735 #ifdef CONFIG_COMPAT
 736         .compat_ioctl   = ext4_compat_ioctl,
 737 #endif
 738         .mmap           = ext4_file_mmap,
 739         .open           = ext4_file_open,
 740         .release        = ext4_release_file,
 741         .fsync          = ext4_sync_file,
 742         .splice_read    = generic_file_splice_read,
 743         .splice_write   = iter_file_splice_write,
 744         .fallocate      = ext4_fallocate,
 745 };
 746
 747 const struct inode_operations ext4_file_inode_operations = {
 748         .setattr        = ext4_setattr,
 749         .getattr        = ext4_getattr,
 750         .setxattr       = generic_setxattr,
 751         .getxattr       = generic_getxattr,
 752         .listxattr      = ext4_listxattr,
 753         .removexattr    = generic_removexattr,
 754         .get_acl        = ext4_get_acl,
 755         .set_acl        = ext4_set_acl,
 756         .fiemap         = ext4_fiemap,
 757 };
 758