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         if (!inode_trylock_shared(inode)) {
  41                 if (iocb->ki_flags & IOCB_NOWAIT)
  42                         return -EAGAIN;
  43                 inode_lock_shared(inode);
  44         }
  45         /*
  46          * Recheck under inode lock - at this point we are sure it cannot
  47          * change anymore
  48          */
  49         if (!IS_DAX(inode)) {
  50                 inode_unlock_shared(inode);
  51                 /* Fallback to buffered IO in case we cannot support DAX */
  52                 return generic_file_read_iter(iocb, to);
  53         }
  54         ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
  55         inode_unlock_shared(inode);
  56
  57         file_accessed(iocb->ki_filp);
  58         return ret;
  59 }
  60 #endif
  61
  62 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
  63 {
  64         if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
  65                 return -EIO;
  66
  67         if (!iov_iter_count(to))
  68                 return 0; /* skip atime */
  69
  70 #ifdef CONFIG_FS_DAX
  71         if (IS_DAX(file_inode(iocb->ki_filp)))
  72                 return ext4_dax_read_iter(iocb, to);
  73 #endif
  74         return generic_file_read_iter(iocb, to);
  75 }
  76
  77 /*
  78  * Called when an inode is released. Note that this is different
  79  * from ext4_file_open: open gets called at every open, but release
  80  * gets called only when /all/ the files are closed.
  81  */
  82 static int ext4_release_file(struct inode *inode, struct file *filp)
  83 {
  84         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
  85                 ext4_alloc_da_blocks(inode);
  86                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
  87         }
  88         /* if we are the last writer on the inode, drop the block reservation */
  89         if ((filp->f_mode & FMODE_WRITE) &&
  90                         (atomic_read(&inode->i_writecount) == 1) &&
  91                         !EXT4_I(inode)->i_reserved_data_blocks)
  92         {
  93                 down_write(&EXT4_I(inode)->i_data_sem);
  94                 ext4_discard_preallocations(inode);
  95                 up_write(&EXT4_I(inode)->i_data_sem);
  96         }
  97         if (is_dx(inode) && filp->private_data)
  98                 ext4_htree_free_dir_info(filp->private_data);
  99
 100         return 0;
 101 }
 102
 103 static void ext4_unwritten_wait(struct inode *inode)
 104 {
 105         wait_queue_head_t *wq = ext4_ioend_wq(inode);
 106
 107         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
 108 }
 109
 110 /*
 111  * This tests whether the IO in question is block-aligned or not.
 112  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
 113  * are converted to written only after the IO is complete.  Until they are
 114  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
 115  * it needs to zero out portions of the start and/or end block.  If 2 AIO
 116  * threads are at work on the same unwritten block, they must be synchronized
 117  * or one thread will zero the other's data, causing corruption.
 118  */
 119 static int
 120 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
 121 {
 122         struct super_block *sb = inode->i_sb;
 123         int blockmask = sb->s_blocksize - 1;
 124
 125         if (pos >= i_size_read(inode))
 126                 return 0;
 127
 128         if ((pos | iov_iter_alignment(from)) & blockmask)
 129                 return 1;
 130
 131         return 0;
 132 }
 133
 134 /* Is IO overwriting allocated and initialized blocks? */
 135 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
 136 {
 137         struct ext4_map_blocks map;
 138         unsigned int blkbits = inode->i_blkbits;
 139         int err, blklen;
 140
 141         if (pos + len > i_size_read(inode))
 142                 return false;
 143
 144         map.m_lblk = pos >> blkbits;
 145         map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
 146         blklen = map.m_len;
 147
 148         err = ext4_map_blocks(NULL, inode, &map, 0);
 149         /*
 150          * 'err==len' means that all of the blocks have been preallocated,
 151          * regardless of whether they have been initialized or not. To exclude
 152          * unwritten extents, we need to check m_flags.
 153          */
 154         return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
 155 }
 156
 157 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
 158 {
 159         struct inode *inode = file_inode(iocb->ki_filp);
 160         ssize_t ret;
 161
 162         ret = generic_write_checks(iocb, from);
 163         if (ret <= 0)
 164                 return ret;
 165         /*
 166          * If we have encountered a bitmap-format file, the size limit
 167          * is smaller than s_maxbytes, which is for extent-mapped files.
 168          */
 169         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 170                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 171
 172                 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
 173                         return -EFBIG;
 174                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 175         }
 176         return iov_iter_count(from);
 177 }
 178
 179 #ifdef CONFIG_FS_DAX
 180 static ssize_t
 181 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
 182 {
 183         struct inode *inode = file_inode(iocb->ki_filp);
 184         ssize_t ret;
 185
 186         if (!inode_trylock(inode)) {
 187                 if (iocb->ki_flags & IOCB_NOWAIT)
 188                         return -EAGAIN;
 189                 inode_lock(inode);
 190         }
 191         ret = ext4_write_checks(iocb, from);
 192         if (ret <= 0)
 193                 goto out;
 194         ret = file_remove_privs(iocb->ki_filp);
 195         if (ret)
 196                 goto out;
 197         ret = file_update_time(iocb->ki_filp);
 198         if (ret)
 199                 goto out;
 200
 201         ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
 202 out:
 203         inode_unlock(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         if (!inode_trylock(inode)) {
 228                 if (iocb->ki_flags & IOCB_NOWAIT)
 229                         return -EAGAIN;
 230                 inode_lock(inode);
 231         }
 232
 233         ret = ext4_write_checks(iocb, from);
 234         if (ret <= 0)
 235                 goto out;
 236
 237         /*
 238          * Unaligned direct AIO must be serialized among each other as zeroing
 239          * of partial blocks of two competing unaligned AIOs can result in data
 240          * corruption.
 241          */
 242         if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 243             !is_sync_kiocb(iocb) &&
 244             ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
 245                 unaligned_aio = 1;
 246                 ext4_unwritten_wait(inode);
 247         }
 248
 249         iocb->private = &overwrite;
 250         /* Check whether we do a DIO overwrite or not */
 251         if (o_direct && !unaligned_aio) {
 252                 if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
 253                         if (ext4_should_dioread_nolock(inode))
 254                                 overwrite = 1;
 255                 } else if (iocb->ki_flags & IOCB_NOWAIT) {
 256                         ret = -EAGAIN;
 257                         goto out;
 258                 }
 259         }
 260
 261         ret = __generic_file_write_iter(iocb, from);
 262         inode_unlock(inode);
 263
 264         if (ret > 0)
 265                 ret = generic_write_sync(iocb, ret);
 266
 267         return ret;
 268
 269 out:
 270         inode_unlock(inode);
 271         return ret;
 272 }
 273
 274 #ifdef CONFIG_FS_DAX
 275 static int ext4_dax_huge_fault(struct vm_fault *vmf,
 276                 enum page_entry_size pe_size)
 277 {
 278         int result;
 279         handle_t *handle = NULL;
 280         struct inode *inode = file_inode(vmf->vma->vm_file);
 281         struct super_block *sb = inode->i_sb;
 282         bool write = vmf->flags & FAULT_FLAG_WRITE;
 283
 284         if (write) {
 285                 sb_start_pagefault(sb);
 286                 file_update_time(vmf->vma->vm_file);
 287                 down_read(&EXT4_I(inode)->i_mmap_sem);
 288                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 289                                                EXT4_DATA_TRANS_BLOCKS(sb));
 290         } else {
 291                 down_read(&EXT4_I(inode)->i_mmap_sem);
 292         }
 293         if (!IS_ERR(handle))
 294                 result = dax_iomap_fault(vmf, pe_size, &ext4_iomap_ops);
 295         else
 296                 result = VM_FAULT_SIGBUS;
 297         if (write) {
 298                 if (!IS_ERR(handle))
 299                         ext4_journal_stop(handle);
 300                 up_read(&EXT4_I(inode)->i_mmap_sem);
 301                 sb_end_pagefault(sb);
 302         } else {
 303                 up_read(&EXT4_I(inode)->i_mmap_sem);
 304         }
 305
 306         return result;
 307 }
 308
 309 static int ext4_dax_fault(struct vm_fault *vmf)
 310 {
 311         return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
 312 }
 313
 314 /*
 315  * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
 316  * handler we check for races agaist truncate. Note that since we cycle through
 317  * i_mmap_sem, we are sure that also any hole punching that began before we
 318  * were called is finished by now and so if it included part of the file we
 319  * are working on, our pte will get unmapped and the check for pte_same() in
 320  * wp_pfn_shared() fails. Thus fault gets retried and things work out as
 321  * desired.
 322  */
 323 static int ext4_dax_pfn_mkwrite(struct vm_fault *vmf)
 324 {
 325         struct inode *inode = file_inode(vmf->vma->vm_file);
 326         struct super_block *sb = inode->i_sb;
 327         loff_t size;
 328         int ret;
 329
 330         sb_start_pagefault(sb);
 331         file_update_time(vmf->vma->vm_file);
 332         down_read(&EXT4_I(inode)->i_mmap_sem);
 333         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 334         if (vmf->pgoff >= size)
 335                 ret = VM_FAULT_SIGBUS;
 336         else
 337                 ret = dax_pfn_mkwrite(vmf);
 338         up_read(&EXT4_I(inode)->i_mmap_sem);
 339         sb_end_pagefault(sb);
 340
 341         return ret;
 342 }
 343
 344 static const struct vm_operations_struct ext4_dax_vm_ops = {
 345         .fault          = ext4_dax_fault,
 346         .huge_fault     = ext4_dax_huge_fault,
 347         .page_mkwrite   = ext4_dax_fault,
 348         .pfn_mkwrite    = ext4_dax_pfn_mkwrite,
 349 };
 350 #else
 351 #define ext4_dax_vm_ops ext4_file_vm_ops
 352 #endif
 353
 354 static const struct vm_operations_struct ext4_file_vm_ops = {
 355         .fault          = ext4_filemap_fault,
 356         .map_pages      = filemap_map_pages,
 357         .page_mkwrite   = ext4_page_mkwrite,
 358 };
 359
 360 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 361 {
 362         struct inode *inode = file->f_mapping->host;
 363
 364         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 365                 return -EIO;
 366
 367         if (ext4_encrypted_inode(inode)) {
 368                 int err = fscrypt_get_encryption_info(inode);
 369                 if (err)
 370                         return 0;
 371                 if (!fscrypt_has_encryption_key(inode))
 372                         return -ENOKEY;
 373         }
 374         file_accessed(file);
 375         if (IS_DAX(file_inode(file))) {
 376                 vma->vm_ops = &ext4_dax_vm_ops;
 377                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
 378         } else {
 379                 vma->vm_ops = &ext4_file_vm_ops;
 380         }
 381         return 0;
 382 }
 383
 384 static int ext4_file_open(struct inode * inode, struct file * filp)
 385 {
 386         struct super_block *sb = inode->i_sb;
 387         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 388         struct vfsmount *mnt = filp->f_path.mnt;
 389         struct dentry *dir;
 390         struct path path;
 391         char buf[64], *cp;
 392         int ret;
 393
 394         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 395                 return -EIO;
 396
 397         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 398                      !(sb->s_flags & MS_RDONLY))) {
 399                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 400                 /*
 401                  * Sample where the filesystem has been mounted and
 402                  * store it in the superblock for sysadmin convenience
 403                  * when trying to sort through large numbers of block
 404                  * devices or filesystem images.
 405                  */
 406                 memset(buf, 0, sizeof(buf));
 407                 path.mnt = mnt;
 408                 path.dentry = mnt->mnt_root;
 409                 cp = d_path(&path, buf, sizeof(buf));
 410                 if (!IS_ERR(cp)) {
 411                         handle_t *handle;
 412                         int err;
 413
 414                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 415                         if (IS_ERR(handle))
 416                                 return PTR_ERR(handle);
 417                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 418                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 419                         if (err) {
 420                                 ext4_journal_stop(handle);
 421                                 return err;
 422                         }
 423                         strlcpy(sbi->s_es->s_last_mounted, cp,
 424                                 sizeof(sbi->s_es->s_last_mounted));
 425                         ext4_handle_dirty_super(handle, sb);
 426                         ext4_journal_stop(handle);
 427                 }
 428         }
 429         if (ext4_encrypted_inode(inode)) {
 430                 ret = fscrypt_get_encryption_info(inode);
 431                 if (ret)
 432                         return -EACCES;
 433                 if (!fscrypt_has_encryption_key(inode))
 434                         return -ENOKEY;
 435         }
 436
 437         dir = dget_parent(file_dentry(filp));
 438         if (ext4_encrypted_inode(d_inode(dir)) &&
 439                         !fscrypt_has_permitted_context(d_inode(dir), inode)) {
 440                 ext4_warning(inode->i_sb,
 441                              "Inconsistent encryption contexts: %lu/%lu",
 442                              (unsigned long) d_inode(dir)->i_ino,
 443                              (unsigned long) inode->i_ino);
 444                 dput(dir);
 445                 return -EPERM;
 446         }
 447         dput(dir);
 448         /*
 449          * Set up the jbd2_inode if we are opening the inode for
 450          * writing and the journal is present
 451          */
 452         if (filp->f_mode & FMODE_WRITE) {
 453                 ret = ext4_inode_attach_jinode(inode);
 454                 if (ret < 0)
 455                         return ret;
 456         }
 457
 458         /* Set the flags to support nowait AIO */
 459         filp->f_mode |= FMODE_AIO_NOWAIT;
 460
 461         return dquot_file_open(inode, filp);
 462 }
 463
 464 /*
 465  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 466  * file rather than ext4_ext_walk_space() because we can introduce
 467  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 468  * function.  When extent status tree has been fully implemented, it will
 469  * track all extent status for a file and we can directly use it to
 470  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 471  */
 472
 473 /*
 474  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 475  * lookup page cache to check whether or not there has some data between
 476  * [startoff, endoff] because, if this range contains an unwritten extent,
 477  * we determine this extent as a data or a hole according to whether the
 478  * page cache has data or not.
 479  */
 480 static int ext4_find_unwritten_pgoff(struct inode *inode,
 481                                      int whence,
 482                                      ext4_lblk_t end_blk,
 483                                      loff_t *offset)
 484 {
 485         struct pagevec pvec;
 486         unsigned int blkbits;
 487         pgoff_t index;
 488         pgoff_t end;
 489         loff_t endoff;
 490         loff_t startoff;
 491         loff_t lastoff;
 492         int found = 0;
 493
 494         blkbits = inode->i_sb->s_blocksize_bits;
 495         startoff = *offset;
 496         lastoff = startoff;
 497         endoff = (loff_t)end_blk << blkbits;
 498
 499         index = startoff >> PAGE_SHIFT;
 500         end = (endoff - 1) >> PAGE_SHIFT;
 501
 502         pagevec_init(&pvec, 0);
 503         do {
 504                 int i, num;
 505                 unsigned long nr_pages;
 506
 507                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE - 1) + 1;
 508                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 509                                           (pgoff_t)num);
 510                 if (nr_pages == 0)
 511                         break;
 512
 513                 for (i = 0; i < nr_pages; i++) {
 514                         struct page *page = pvec.pages[i];
 515                         struct buffer_head *bh, *head;
 516
 517                         /*
 518                          * If current offset is smaller than the page offset,
 519                          * there is a hole at this offset.
 520                          */
 521                         if (whence == SEEK_HOLE && lastoff < endoff &&
 522                             lastoff < page_offset(pvec.pages[i])) {
 523                                 found = 1;
 524                                 *offset = lastoff;
 525                                 goto out;
 526                         }
 527
 528                         if (page->index > end)
 529                                 goto out;
 530
 531                         lock_page(page);
 532
 533                         if (unlikely(page->mapping != inode->i_mapping)) {
 534                                 unlock_page(page);
 535                                 continue;
 536                         }
 537
 538                         if (!page_has_buffers(page)) {
 539                                 unlock_page(page);
 540                                 continue;
 541                         }
 542
 543                         if (page_has_buffers(page)) {
 544                                 lastoff = page_offset(page);
 545                                 bh = head = page_buffers(page);
 546                                 do {
 547                                         if (buffer_uptodate(bh) ||
 548                                             buffer_unwritten(bh)) {
 549                                                 if (whence == SEEK_DATA)
 550                                                         found = 1;
 551                                         } else {
 552                                                 if (whence == SEEK_HOLE)
 553                                                         found = 1;
 554                                         }
 555                                         if (found) {
 556                                                 *offset = max_t(loff_t,
 557                                                         startoff, lastoff);
 558                                                 unlock_page(page);
 559                                                 goto out;
 560                                         }
 561                                         lastoff += bh->b_size;
 562                                         bh = bh->b_this_page;
 563                                 } while (bh != head);
 564                         }
 565
 566                         lastoff = page_offset(page) + PAGE_SIZE;
 567                         unlock_page(page);
 568                 }
 569
 570                 /* The no. of pages is less than our desired, we are done. */
 571                 if (nr_pages < num)
 572                         break;
 573
 574                 index = pvec.pages[i - 1]->index + 1;
 575                 pagevec_release(&pvec);
 576         } while (index <= end);
 577
 578         if (whence == SEEK_HOLE && lastoff < endoff) {
 579                 found = 1;
 580                 *offset = lastoff;
 581         }
 582 out:
 583         pagevec_release(&pvec);
 584         return found;
 585 }
 586
 587 /*
 588  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 589  */
 590 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 591 {
 592         struct inode *inode = file->f_mapping->host;
 593         struct extent_status es;
 594         ext4_lblk_t start, last, end;
 595         loff_t dataoff, isize;
 596         int blkbits;
 597         int ret;
 598
 599         inode_lock(inode);
 600
 601         isize = i_size_read(inode);
 602         if (offset >= isize) {
 603                 inode_unlock(inode);
 604                 return -ENXIO;
 605         }
 606
 607         blkbits = inode->i_sb->s_blocksize_bits;
 608         start = offset >> blkbits;
 609         last = start;
 610         end = isize >> blkbits;
 611         dataoff = offset;
 612
 613         do {
 614                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 615                 if (ret <= 0) {
 616                         /* No extent found -> no data */
 617                         if (ret == 0)
 618                                 ret = -ENXIO;
 619                         inode_unlock(inode);
 620                         return ret;
 621                 }
 622
 623                 last = es.es_lblk;
 624                 if (last != start)
 625                         dataoff = (loff_t)last << blkbits;
 626                 if (!ext4_es_is_unwritten(&es))
 627                         break;
 628
 629                 /*
 630                  * If there is a unwritten extent at this offset,
 631                  * it will be as a data or a hole according to page
 632                  * cache that has data or not.
 633                  */
 634                 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 635                                               es.es_lblk + es.es_len, &dataoff))
 636                         break;
 637                 last += es.es_len;
 638                 dataoff = (loff_t)last << blkbits;
 639                 cond_resched();
 640         } while (last <= end);
 641
 642         inode_unlock(inode);
 643
 644         if (dataoff > isize)
 645                 return -ENXIO;
 646
 647         return vfs_setpos(file, dataoff, maxsize);
 648 }
 649
 650 /*
 651  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 652  */
 653 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 654 {
 655         struct inode *inode = file->f_mapping->host;
 656         struct extent_status es;
 657         ext4_lblk_t start, last, end;
 658         loff_t holeoff, isize;
 659         int blkbits;
 660         int ret;
 661
 662         inode_lock(inode);
 663
 664         isize = i_size_read(inode);
 665         if (offset >= isize) {
 666                 inode_unlock(inode);
 667                 return -ENXIO;
 668         }
 669
 670         blkbits = inode->i_sb->s_blocksize_bits;
 671         start = offset >> blkbits;
 672         last = start;
 673         end = isize >> blkbits;
 674         holeoff = offset;
 675
 676         do {
 677                 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
 678                 if (ret < 0) {
 679                         inode_unlock(inode);
 680                         return ret;
 681                 }
 682                 /* Found a hole? */
 683                 if (ret == 0 || es.es_lblk > last) {
 684                         if (last != start)
 685                                 holeoff = (loff_t)last << blkbits;
 686                         break;
 687                 }
 688                 /*
 689                  * If there is a unwritten extent at this offset,
 690                  * it will be as a data or a hole according to page
 691                  * cache that has data or not.
 692                  */
 693                 if (ext4_es_is_unwritten(&es) &&
 694                     ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 695                                               last + es.es_len, &holeoff))
 696                         break;
 697
 698                 last += es.es_len;
 699                 holeoff = (loff_t)last << blkbits;
 700                 cond_resched();
 701         } while (last <= end);
 702
 703         inode_unlock(inode);
 704
 705         if (holeoff > isize)
 706                 holeoff = isize;
 707
 708         return vfs_setpos(file, holeoff, maxsize);
 709 }
 710
 711 /*
 712  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 713  * by calling generic_file_llseek_size() with the appropriate maxbytes
 714  * value for each.
 715  */
 716 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 717 {
 718         struct inode *inode = file->f_mapping->host;
 719         loff_t maxbytes;
 720
 721         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 722                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 723         else
 724                 maxbytes = inode->i_sb->s_maxbytes;
 725
 726         switch (whence) {
 727         case SEEK_SET:
 728         case SEEK_CUR:
 729         case SEEK_END:
 730                 return generic_file_llseek_size(file, offset, whence,
 731                                                 maxbytes, i_size_read(inode));
 732         case SEEK_DATA:
 733                 return ext4_seek_data(file, offset, maxbytes);
 734         case SEEK_HOLE:
 735                 return ext4_seek_hole(file, offset, maxbytes);
 736         }
 737
 738         return -EINVAL;
 739 }
 740
 741 const struct file_operations ext4_file_operations = {
 742         .llseek         = ext4_llseek,
 743         .read_iter      = ext4_file_read_iter,
 744         .write_iter     = ext4_file_write_iter,
 745         .unlocked_ioctl = ext4_ioctl,
 746 #ifdef CONFIG_COMPAT
 747         .compat_ioctl   = ext4_compat_ioctl,
 748 #endif
 749         .mmap           = ext4_file_mmap,
 750         .open           = ext4_file_open,
 751         .release        = ext4_release_file,
 752         .fsync          = ext4_sync_file,
 753         .get_unmapped_area = thp_get_unmapped_area,
 754         .splice_read    = generic_file_splice_read,
 755         .splice_write   = iter_file_splice_write,
 756         .fallocate      = ext4_fallocate,
 757 };
 758
 759 const struct inode_operations ext4_file_inode_operations = {
 760         .setattr        = ext4_setattr,
 761         .getattr        = ext4_file_getattr,
 762         .listxattr      = ext4_listxattr,
 763         .get_acl        = ext4_get_acl,
 764         .set_acl        = ext4_set_acl,
 765         .fiemap         = ext4_fiemap,
 766 };
 767