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         bool overwrite = false;
 182
 183         inode_lock(inode);
 184         ret = ext4_write_checks(iocb, from);
 185         if (ret <= 0)
 186                 goto out;
 187         ret = file_remove_privs(iocb->ki_filp);
 188         if (ret)
 189                 goto out;
 190         ret = file_update_time(iocb->ki_filp);
 191         if (ret)
 192                 goto out;
 193
 194         if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
 195                 overwrite = true;
 196                 downgrade_write(&inode->i_rwsem);
 197         }
 198         ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
 199 out:
 200         if (!overwrite)
 201                 inode_unlock(inode);
 202         else
 203                 inode_unlock_shared(inode);
 204         if (ret > 0)
 205                 ret = generic_write_sync(iocb, ret);
 206         return ret;
 207 }
 208 #endif
 209
 210 static ssize_t
 211 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 212 {
 213         struct inode *inode = file_inode(iocb->ki_filp);
 214         int o_direct = iocb->ki_flags & IOCB_DIRECT;
 215         int unaligned_aio = 0;
 216         int overwrite = 0;
 217         ssize_t ret;
 218
 219         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 220                 return -EIO;
 221
 222 #ifdef CONFIG_FS_DAX
 223         if (IS_DAX(inode))
 224                 return ext4_dax_write_iter(iocb, from);
 225 #endif
 226
 227         inode_lock(inode);
 228         ret = ext4_write_checks(iocb, from);
 229         if (ret <= 0)
 230                 goto out;
 231
 232         /*
 233          * Unaligned direct AIO must be serialized among each other as zeroing
 234          * of partial blocks of two competing unaligned AIOs can result in data
 235          * corruption.
 236          */
 237         if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 238             !is_sync_kiocb(iocb) &&
 239             ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
 240                 unaligned_aio = 1;
 241                 ext4_unwritten_wait(inode);
 242         }
 243
 244         iocb->private = &overwrite;
 245         /* Check whether we do a DIO overwrite or not */
 246         if (o_direct && ext4_should_dioread_nolock(inode) && !unaligned_aio &&
 247             ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from)))
 248                 overwrite = 1;
 249
 250         ret = __generic_file_write_iter(iocb, from);
 251         inode_unlock(inode);
 252
 253         if (ret > 0)
 254                 ret = generic_write_sync(iocb, ret);
 255
 256         return ret;
 257
 258 out:
 259         inode_unlock(inode);
 260         return ret;
 261 }
 262
 263 #ifdef CONFIG_FS_DAX
 264 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 265 {
 266         int result;
 267         struct inode *inode = file_inode(vma->vm_file);
 268         struct super_block *sb = inode->i_sb;
 269         bool write = vmf->flags & FAULT_FLAG_WRITE;
 270
 271         if (write) {
 272                 sb_start_pagefault(sb);
 273                 file_update_time(vma->vm_file);
 274         }
 275         down_read(&EXT4_I(inode)->i_mmap_sem);
 276         result = dax_iomap_fault(vma, vmf, &ext4_iomap_ops);
 277         up_read(&EXT4_I(inode)->i_mmap_sem);
 278         if (write)
 279                 sb_end_pagefault(sb);
 280
 281         return result;
 282 }
 283
 284 static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
 285                                                 pmd_t *pmd, unsigned int flags)
 286 {
 287         int result;
 288         struct inode *inode = file_inode(vma->vm_file);
 289         struct super_block *sb = inode->i_sb;
 290         bool write = flags & FAULT_FLAG_WRITE;
 291
 292         if (write) {
 293                 sb_start_pagefault(sb);
 294                 file_update_time(vma->vm_file);
 295         }
 296         down_read(&EXT4_I(inode)->i_mmap_sem);
 297         result = dax_iomap_pmd_fault(vma, addr, pmd, flags,
 298                                      &ext4_iomap_ops);
 299         up_read(&EXT4_I(inode)->i_mmap_sem);
 300         if (write)
 301                 sb_end_pagefault(sb);
 302
 303         return result;
 304 }
 305
 306 /*
 307  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
 308  * handler we check for races agaist truncate. Note that since we cycle through
 309  * i_mmap_sem, we are sure that also any hole punching that began before we
 310  * were called is finished by now and so if it included part of the file we
 311  * are working on, our pte will get unmapped and the check for pte_same() in
 312  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 313  * desired.
 314  */
 315 static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
 316                                 struct vm_fault *vmf)
 317 {
 318         struct inode *inode = file_inode(vma->vm_file);
 319         struct super_block *sb = inode->i_sb;
 320         loff_t size;
 321         int ret;
 322
 323         sb_start_pagefault(sb);
 324         file_update_time(vma->vm_file);
 325         down_read(&EXT4_I(inode)->i_mmap_sem);
 326         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 327         if (vmf->pgoff >= size)
 328                 ret = VM_FAULT_SIGBUS;
 329         else
 330                 ret = dax_pfn_mkwrite(vma, vmf);
 331         up_read(&EXT4_I(inode)->i_mmap_sem);
 332         sb_end_pagefault(sb);
 333
 334         return ret;
 335 }
 336
 337 static const struct vm_operations_struct ext4_dax_vm_ops = {
 338         .fault          = ext4_dax_fault,
 339         .pmd_fault      = ext4_dax_pmd_fault,
 340         .page_mkwrite   = ext4_dax_fault,
 341         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 342 };
 343 #else
 344 #define ext4_dax_vm_ops ext4_file_vm_ops
 345 #endif
 346
 347 static const struct vm_operations_struct ext4_file_vm_ops = {
 348         .fault          = ext4_filemap_fault,
 349         .map_pages      = filemap_map_pages,
 350         .page_mkwrite   = ext4_page_mkwrite,
 351 };
 352
 353 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 354 {
 355         struct inode *inode = file->f_mapping->host;
 356
 357         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 358                 return -EIO;
 359
 360         if (ext4_encrypted_inode(inode)) {
 361                 int err = fscrypt_get_encryption_info(inode);
 362                 if (err)
 363                         return 0;
 364                 if (!fscrypt_has_encryption_key(inode))
 365                         return -ENOKEY;
 366         }
 367         file_accessed(file);
 368         if (IS_DAX(file_inode(file))) {
 369                 vma->vm_ops = &ext4_dax_vm_ops;
 370                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 371         } else {
 372                 vma->vm_ops = &ext4_file_vm_ops;
 373         }
 374         return 0;
 375 }
 376
 377 static int ext4_file_open(struct inode * inode, struct file * filp)
 378 {
 379         struct super_block *sb = inode->i_sb;
 380         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 381         struct vfsmount *mnt = filp->f_path.mnt;
 382         struct dentry *dir;
 383         struct path path;
 384         char buf[64], *cp;
 385         int ret;
 386
 387         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 388                 return -EIO;
 389
 390         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 391                      !(sb->s_flags & MS_RDONLY))) {
 392                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 393                 /*
 394                  * Sample where the filesystem has been mounted and
 395                  * store it in the superblock for sysadmin convenience
 396                  * when trying to sort through large numbers of block
 397                  * devices or filesystem images.
 398                  */
 399                 memset(buf, 0, sizeof(buf));
 400                 path.mnt = mnt;
 401                 path.dentry = mnt->mnt_root;
 402                 cp = d_path(&path, buf, sizeof(buf));
 403                 if (!IS_ERR(cp)) {
 404                         handle_t *handle;
 405                         int err;
 406
 407                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 408                         if (IS_ERR(handle))
 409                                 return PTR_ERR(handle);
 410                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 411                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 412                         if (err) {
 413                                 ext4_journal_stop(handle);
 414                                 return err;
 415                         }
 416                         strlcpy(sbi->s_es->s_last_mounted, cp,
 417                                 sizeof(sbi->s_es->s_last_mounted));
 418                         ext4_handle_dirty_super(handle, sb);
 419                         ext4_journal_stop(handle);
 420                 }
 421         }
 422         if (ext4_encrypted_inode(inode)) {
 423                 ret = fscrypt_get_encryption_info(inode);
 424                 if (ret)
 425                         return -EACCES;
 426                 if (!fscrypt_has_encryption_key(inode))
 427                         return -ENOKEY;
 428         }
 429
 430         dir = dget_parent(file_dentry(filp));
 431         if (ext4_encrypted_inode(d_inode(dir)) &&
 432                         !fscrypt_has_permitted_context(d_inode(dir), inode)) {
 433                 ext4_warning(inode->i_sb,
 434                              "Inconsistent encryption contexts: %lu/%lu",
 435                              (unsigned long) d_inode(dir)->i_ino,
 436                              (unsigned long) inode->i_ino);
 437                 dput(dir);
 438                 return -EPERM;
 439         }
 440         dput(dir);
 441         /*
 442          * Set up the jbd2_inode if we are opening the inode for
 443          * writing and the journal is present
 444          */
 445         if (filp->f_mode & FMODE_WRITE) {
 446                 ret = ext4_inode_attach_jinode(inode);
 447                 if (ret < 0)
 448                         return ret;
 449         }
 450         return dquot_file_open(inode, filp);
 451 }
 452
 453 /*
 454  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 455  * file rather than ext4_ext_walk_space() because we can introduce
 456  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 457  * function.  When extent status tree has been fully implemented, it will
 458  * track all extent status for a file and we can directly use it to
 459  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 460  */
 461
 462 /*
 463  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 464  * lookup page cache to check whether or not there has some data between
 465  * [startoff, endoff] because, if this range contains an unwritten extent,
 466  * we determine this extent as a data or a hole according to whether the
 467  * page cache has data or not.
 468  */
 469 static int ext4_find_unwritten_pgoff(struct inode *inode,
 470                                      int whence,
 471                                      ext4_lblk_t end_blk,
 472                                      loff_t *offset)
 473 {
 474         struct pagevec pvec;
 475         unsigned int blkbits;
 476         pgoff_t index;
 477         pgoff_t end;
 478         loff_t endoff;
 479         loff_t startoff;
 480         loff_t lastoff;
 481         int found = 0;
 482
 483         blkbits = inode->i_sb->s_blocksize_bits;
 484         startoff = *offset;
 485         lastoff = startoff;
 486         endoff = (loff_t)end_blk << blkbits;
 487
 488         index = startoff >> PAGE_SHIFT;
 489         end = endoff >> PAGE_SHIFT;
 490
 491         pagevec_init(&pvec, 0);
 492         do {
 493                 int i, num;
 494                 unsigned long nr_pages;
 495
 496                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 497                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 498                                           (pgoff_t)num);
 499                 if (nr_pages == 0) {
 500                         if (whence == SEEK_DATA)
 501                                 break;
 502
 503                         BUG_ON(whence != SEEK_HOLE);
 504                         /*
 505                          * If this is the first time to go into the loop and
 506                          * offset is not beyond the end offset, it will be a
 507                          * hole at this offset
 508                          */
 509                         if (lastoff == startoff || lastoff < endoff)
 510                                 found = 1;
 511                         break;
 512                 }
 513
 514                 /*
 515                  * If this is the first time to go into the loop and
 516                  * offset is smaller than the first page offset, it will be a
 517                  * hole at this offset.
 518                  */
 519                 if (lastoff == startoff && whence == SEEK_HOLE &&
 520                     lastoff < page_offset(pvec.pages[0])) {
 521                         found = 1;
 522                         break;
 523                 }
 524
 525                 for (i = 0; i < nr_pages; i++) {
 526                         struct page *page = pvec.pages[i];
 527                         struct buffer_head *bh, *head;
 528
 529                         /*
 530                          * If the current offset is not beyond the end of given
 531                          * range, it will be a hole.
 532                          */
 533                         if (lastoff < endoff && whence == SEEK_HOLE &&
 534                             page->index > end) {
 535                                 found = 1;
 536                                 *offset = lastoff;
 537                                 goto out;
 538                         }
 539
 540                         lock_page(page);
 541
 542                         if (unlikely(page->mapping != inode->i_mapping)) {
 543                                 unlock_page(page);
 544                                 continue;
 545                         }
 546
 547                         if (!page_has_buffers(page)) {
 548                                 unlock_page(page);
 549                                 continue;
 550                         }
 551
 552                         if (page_has_buffers(page)) {
 553                                 lastoff = page_offset(page);
 554                                 bh = head = page_buffers(page);
 555                                 do {
 556                                         if (buffer_uptodate(bh) ||
 557                                             buffer_unwritten(bh)) {
 558                                                 if (whence == SEEK_DATA)
 559                                                         found = 1;
 560                                         } else {
 561                                                 if (whence == SEEK_HOLE)
 562                                                         found = 1;
 563                                         }
 564                                         if (found) {
 565                                                 *offset = max_t(loff_t,
 566                                                         startoff, lastoff);
 567                                                 unlock_page(page);
 568                                                 goto out;
 569                                         }
 570                                         lastoff += bh->b_size;
 571                                         bh = bh->b_this_page;
 572                                 } while (bh != head);
 573                         }
 574
 575                         lastoff = page_offset(page) + PAGE_SIZE;
 576                         unlock_page(page);
 577                 }
 578
 579                 /*
 580                  * The no. of pages is less than our desired, that would be a
 581                  * hole in there.
 582                  */
 583                 if (nr_pages < num && whence == SEEK_HOLE) {
 584                         found = 1;
 585                         *offset = lastoff;
 586                         break;
 587                 }
 588
 589                 index = pvec.pages[i - 1]->index + 1;
 590                 pagevec_release(&pvec);
 591         } while (index <= end);
 592
 593 out:
 594         pagevec_release(&pvec);
 595         return found;
 596 }
 597
 598 /*
 599  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 600  */
 601 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 602 {
 603         struct inode *inode = file->f_mapping->host;
 604         struct extent_status es;
 605         ext4_lblk_t start, last, end;
 606         loff_t dataoff, isize;
 607         int blkbits;
 608         int ret;
 609
 610         inode_lock(inode);
 611
 612         isize = i_size_read(inode);
 613         if (offset >= isize) {
 614                 inode_unlock(inode);
 615                 return -ENXIO;
 616         }
 617
 618         blkbits = inode->i_sb->s_blocksize_bits;
 619         start = offset >> blkbits;
 620         last = start;
 621         end = isize >> blkbits;
 622         dataoff = offset;
 623
 624         do {
 625                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 626                 if (ret <= 0) {
 627                         /* No extent found -> no data */
 628                         if (ret == 0)
 629                                 ret = -ENXIO;
 630                         inode_unlock(inode);
 631                         return ret;
 632                 }
 633
 634                 last = es.es_lblk;
 635                 if (last != start)
 636                         dataoff = (loff_t)last << blkbits;
 637                 if (!ext4_es_is_unwritten(&es))
 638                         break;
 639
 640                 /*
 641                  * If there is a unwritten extent at this offset,
 642                  * it will be as a data or a hole according to page
 643                  * cache that has data or not.
 644                  */
 645                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 646                                               es.es_lblk + es.es_len, &dataoff))
 647                         break;
 648                 last += es.es_len;
 649                 dataoff = (loff_t)last << blkbits;
 650                 cond_resched();
 651         } while (last <= end);
 652
 653         inode_unlock(inode);
 654
 655         if (dataoff > isize)
 656                 return -ENXIO;
 657
 658         return vfs_setpos(file, dataoff, maxsize);
 659 }
 660
 661 /*
 662  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 663  */
 664 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 665 {
 666         struct inode *inode = file->f_mapping->host;
 667         struct extent_status es;
 668         ext4_lblk_t start, last, end;
 669         loff_t holeoff, isize;
 670         int blkbits;
 671         int ret;
 672
 673         inode_lock(inode);
 674
 675         isize = i_size_read(inode);
 676         if (offset >= isize) {
 677                 inode_unlock(inode);
 678                 return -ENXIO;
 679         }
 680
 681         blkbits = inode->i_sb->s_blocksize_bits;
 682         start = offset >> blkbits;
 683         last = start;
 684         end = isize >> blkbits;
 685         holeoff = offset;
 686
 687         do {
 688                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 689                 if (ret < 0) {
 690                         inode_unlock(inode);
 691                         return ret;
 692                 }
 693                 /* Found a hole? */
 694                 if (ret == 0 || es.es_lblk > last) {
 695                         if (last != start)
 696                                 holeoff = (loff_t)last << blkbits;
 697                         break;
 698                 }
 699                 /*
 700                  * If there is a unwritten extent at this offset,
 701                  * it will be as a data or a hole according to page
 702                  * cache that has data or not.
 703                  */
 704                 if (ext4_es_is_unwritten(&es) &&
 705                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 706                                               last + es.es_len, &holeoff))
 707                         break;
 708
 709                 last += es.es_len;
 710                 holeoff = (loff_t)last << blkbits;
 711                 cond_resched();
 712         } while (last <= end);
 713
 714         inode_unlock(inode);
 715
 716         if (holeoff > isize)
 717                 holeoff = isize;
 718
 719         return vfs_setpos(file, holeoff, maxsize);
 720 }
 721
 722 /*
 723  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 724  * by calling generic_file_llseek_size() with the appropriate maxbytes
 725  * value for each.
 726  */
 727 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 728 {
 729         struct inode *inode = file->f_mapping->host;
 730         loff_t maxbytes;
 731
 732         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 733                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 734         else
 735                 maxbytes = inode->i_sb->s_maxbytes;
 736
 737         switch (whence) {
 738         case SEEK_SET:
 739         case SEEK_CUR:
 740         case SEEK_END:
 741                 return generic_file_llseek_size(file, offset, whence,
 742                                                 maxbytes, i_size_read(inode));
 743         case SEEK_DATA:
 744                 return ext4_seek_data(file, offset, maxbytes);
 745         case SEEK_HOLE:
 746                 return ext4_seek_hole(file, offset, maxbytes);
 747         }
 748
 749         return -EINVAL;
 750 }
 751
 752 const struct file_operations ext4_file_operations = {
 753         .llseek         = ext4_llseek,
 754         .read_iter      = ext4_file_read_iter,
 755         .write_iter     = ext4_file_write_iter,
 756         .unlocked_ioctl = ext4_ioctl,
 757 #ifdef CONFIG_COMPAT
 758         .compat_ioctl   = ext4_compat_ioctl,
 759 #endif
 760         .mmap           = ext4_file_mmap,
 761         .open           = ext4_file_open,
 762         .release        = ext4_release_file,
 763         .fsync          = ext4_sync_file,
 764         .get_unmapped_area = thp_get_unmapped_area,
 765         .splice_read    = generic_file_splice_read,
 766         .splice_write   = iter_file_splice_write,
 767         .fallocate      = ext4_fallocate,
 768 };
 769
 770 const struct inode_operations ext4_file_inode_operations = {
 771         .setattr        = ext4_setattr,
 772         .getattr        = ext4_getattr,
 773         .listxattr      = ext4_listxattr,
 774         .get_acl        = ext4_get_acl,
 775         .set_acl        = ext4_set_acl,
 776         .fiemap         = ext4_fiemap,
 777 };
 778