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/jbd2.h>
  24 #include <linux/mount.h>
  25 #include <linux/path.h>
  26 #include <linux/aio.h>
  27 #include <linux/quotaops.h>
  28 #include <linux/pagevec.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 = file->f_flags & O_DIRECT;
  99         int overwrite = 0;
 100         size_t length = iov_iter_count(from);
 101         ssize_t ret;
 102         loff_t pos = iocb->ki_pos;
 103
 104         /*
 105          * Unaligned direct AIO must be serialized; see comment above
 106          * In the case of O_APPEND, assume that we must always serialize
 107          */
 108         if (o_direct &&
 109             ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 110             !is_sync_kiocb(iocb) &&
 111             (file->f_flags & O_APPEND ||
 112              ext4_unaligned_aio(inode, from, pos))) {
 113                 aio_mutex = ext4_aio_mutex(inode);
 114                 mutex_lock(aio_mutex);
 115                 ext4_unwritten_wait(inode);
 116         }
 117
 118         mutex_lock(&inode->i_mutex);
 119         if (file->f_flags & O_APPEND)
 120                 iocb->ki_pos = pos = i_size_read(inode);
 121
 122         /*
 123          * If we have encountered a bitmap-format file, the size limit
 124          * is smaller than s_maxbytes, which is for extent-mapped files.
 125          */
 126         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 127                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 128
 129                 if ((pos > sbi->s_bitmap_maxbytes) ||
 130                     (pos == sbi->s_bitmap_maxbytes && length > 0)) {
 131                         mutex_unlock(&inode->i_mutex);
 132                         ret = -EFBIG;
 133                         goto errout;
 134                 }
 135
 136                 if (pos + length > sbi->s_bitmap_maxbytes)
 137                         iov_iter_truncate(from, sbi->s_bitmap_maxbytes - pos);
 138         }
 139
 140         if (o_direct) {
 141                 blk_start_plug(&plug);
 142
 143                 iocb->private = &overwrite;
 144
 145                 /* check whether we do a DIO overwrite or not */
 146                 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
 147                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 148                         struct ext4_map_blocks map;
 149                         unsigned int blkbits = inode->i_blkbits;
 150                         int err, len;
 151
 152                         map.m_lblk = pos >> blkbits;
 153                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 154                                 - map.m_lblk;
 155                         len = map.m_len;
 156
 157                         err = ext4_map_blocks(NULL, inode, &map, 0);
 158                         /*
 159                          * 'err==len' means that all of blocks has
 160                          * been preallocated no matter they are
 161                          * initialized or not.  For excluding
 162                          * unwritten extents, we need to check
 163                          * m_flags.  There are two conditions that
 164                          * indicate for initialized extents.  1) If we
 165                          * hit extent cache, EXT4_MAP_MAPPED flag is
 166                          * returned; 2) If we do a real lookup,
 167                          * non-flags are returned.  So we should check
 168                          * these two conditions.
 169                          */
 170                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 171                                 overwrite = 1;
 172                 }
 173         }
 174
 175         ret = __generic_file_write_iter(iocb, from);
 176         mutex_unlock(&inode->i_mutex);
 177
 178         if (ret > 0) {
 179                 ssize_t err;
 180
 181                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
 182                 if (err < 0)
 183                         ret = err;
 184         }
 185         if (o_direct)
 186                 blk_finish_plug(&plug);
 187
 188 errout:
 189         if (aio_mutex)
 190                 mutex_unlock(aio_mutex);
 191         return ret;
 192 }
 193
 194 static const struct vm_operations_struct ext4_file_vm_ops = {
 195         .fault          = filemap_fault,
 196         .map_pages      = filemap_map_pages,
 197         .page_mkwrite   = ext4_page_mkwrite,
 198         .remap_pages    = generic_file_remap_pages,
 199 };
 200
 201 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 202 {
 203         struct address_space *mapping = file->f_mapping;
 204
 205         if (!mapping->a_ops->readpage)
 206                 return -ENOEXEC;
 207         file_accessed(file);
 208         vma->vm_ops = &ext4_file_vm_ops;
 209         return 0;
 210 }
 211
 212 static int ext4_file_open(struct inode * inode, struct file * filp)
 213 {
 214         struct super_block *sb = inode->i_sb;
 215         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 216         struct vfsmount *mnt = filp->f_path.mnt;
 217         struct path path;
 218         char buf[64], *cp;
 219
 220         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 221                      !(sb->s_flags & MS_RDONLY))) {
 222                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 223                 /*
 224                  * Sample where the filesystem has been mounted and
 225                  * store it in the superblock for sysadmin convenience
 226                  * when trying to sort through large numbers of block
 227                  * devices or filesystem images.
 228                  */
 229                 memset(buf, 0, sizeof(buf));
 230                 path.mnt = mnt;
 231                 path.dentry = mnt->mnt_root;
 232                 cp = d_path(&path, buf, sizeof(buf));
 233                 if (!IS_ERR(cp)) {
 234                         handle_t *handle;
 235                         int err;
 236
 237                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 238                         if (IS_ERR(handle))
 239                                 return PTR_ERR(handle);
 240                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 241                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 242                         if (err) {
 243                                 ext4_journal_stop(handle);
 244                                 return err;
 245                         }
 246                         strlcpy(sbi->s_es->s_last_mounted, cp,
 247                                 sizeof(sbi->s_es->s_last_mounted));
 248                         ext4_handle_dirty_super(handle, sb);
 249                         ext4_journal_stop(handle);
 250                 }
 251         }
 252         /*
 253          * Set up the jbd2_inode if we are opening the inode for
 254          * writing and the journal is present
 255          */
 256         if (filp->f_mode & FMODE_WRITE) {
 257                 int ret = ext4_inode_attach_jinode(inode);
 258                 if (ret < 0)
 259                         return ret;
 260         }
 261         return dquot_file_open(inode, filp);
 262 }
 263
 264 /*
 265  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 266  * file rather than ext4_ext_walk_space() because we can introduce
 267  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 268  * function.  When extent status tree has been fully implemented, it will
 269  * track all extent status for a file and we can directly use it to
 270  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 271  */
 272
 273 /*
 274  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 275  * lookup page cache to check whether or not there has some data between
 276  * [startoff, endoff] because, if this range contains an unwritten extent,
 277  * we determine this extent as a data or a hole according to whether the
 278  * page cache has data or not.
 279  */
 280 static int ext4_find_unwritten_pgoff(struct inode *inode,
 281                                      int whence,
 282                                      struct ext4_map_blocks *map,
 283                                      loff_t *offset)
 284 {
 285         struct pagevec pvec;
 286         unsigned int blkbits;
 287         pgoff_t index;
 288         pgoff_t end;
 289         loff_t endoff;
 290         loff_t startoff;
 291         loff_t lastoff;
 292         int found = 0;
 293
 294         blkbits = inode->i_sb->s_blocksize_bits;
 295         startoff = *offset;
 296         lastoff = startoff;
 297         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 298
 299         index = startoff >> PAGE_CACHE_SHIFT;
 300         end = endoff >> PAGE_CACHE_SHIFT;
 301
 302         pagevec_init(&pvec, 0);
 303         do {
 304                 int i, num;
 305                 unsigned long nr_pages;
 306
 307                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 308                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 309                                           (pgoff_t)num);
 310                 if (nr_pages == 0) {
 311                         if (whence == SEEK_DATA)
 312                                 break;
 313
 314                         BUG_ON(whence != SEEK_HOLE);
 315                         /*
 316                          * If this is the first time to go into the loop and
 317                          * offset is not beyond the end offset, it will be a
 318                          * hole at this offset
 319                          */
 320                         if (lastoff == startoff || lastoff < endoff)
 321                                 found = 1;
 322                         break;
 323                 }
 324
 325                 /*
 326                  * If this is the first time to go into the loop and
 327                  * offset is smaller than the first page offset, it will be a
 328                  * hole at this offset.
 329                  */
 330                 if (lastoff == startoff && whence == SEEK_HOLE &&
 331                     lastoff < page_offset(pvec.pages[0])) {
 332                         found = 1;
 333                         break;
 334                 }
 335
 336                 for (i = 0; i < nr_pages; i++) {
 337                         struct page *page = pvec.pages[i];
 338                         struct buffer_head *bh, *head;
 339
 340                         /*
 341                          * If the current offset is not beyond the end of given
 342                          * range, it will be a hole.
 343                          */
 344                         if (lastoff < endoff && whence == SEEK_HOLE &&
 345                             page->index > end) {
 346                                 found = 1;
 347                                 *offset = lastoff;
 348                                 goto out;
 349                         }
 350
 351                         lock_page(page);
 352
 353                         if (unlikely(page->mapping != inode->i_mapping)) {
 354                                 unlock_page(page);
 355                                 continue;
 356                         }
 357
 358                         if (!page_has_buffers(page)) {
 359                                 unlock_page(page);
 360                                 continue;
 361                         }
 362
 363                         if (page_has_buffers(page)) {
 364                                 lastoff = page_offset(page);
 365                                 bh = head = page_buffers(page);
 366                                 do {
 367                                         if (buffer_uptodate(bh) ||
 368                                             buffer_unwritten(bh)) {
 369                                                 if (whence == SEEK_DATA)
 370                                                         found = 1;
 371                                         } else {
 372                                                 if (whence == SEEK_HOLE)
 373                                                         found = 1;
 374                                         }
 375                                         if (found) {
 376                                                 *offset = max_t(loff_t,
 377                                                         startoff, lastoff);
 378                                                 unlock_page(page);
 379                                                 goto out;
 380                                         }
 381                                         lastoff += bh->b_size;
 382                                         bh = bh->b_this_page;
 383                                 } while (bh != head);
 384                         }
 385
 386                         lastoff = page_offset(page) + PAGE_SIZE;
 387                         unlock_page(page);
 388                 }
 389
 390                 /*
 391                  * The no. of pages is less than our desired, that would be a
 392                  * hole in there.
 393                  */
 394                 if (nr_pages < num && whence == SEEK_HOLE) {
 395                         found = 1;
 396                         *offset = lastoff;
 397                         break;
 398                 }
 399
 400                 index = pvec.pages[i - 1]->index + 1;
 401                 pagevec_release(&pvec);
 402         } while (index <= end);
 403
 404 out:
 405         pagevec_release(&pvec);
 406         return found;
 407 }
 408
 409 /*
 410  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 411  */
 412 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 413 {
 414         struct inode *inode = file->f_mapping->host;
 415         struct ext4_map_blocks map;
 416         struct extent_status es;
 417         ext4_lblk_t start, last, end;
 418         loff_t dataoff, isize;
 419         int blkbits;
 420         int ret = 0;
 421
 422         mutex_lock(&inode->i_mutex);
 423
 424         isize = i_size_read(inode);
 425         if (offset >= isize) {
 426                 mutex_unlock(&inode->i_mutex);
 427                 return -ENXIO;
 428         }
 429
 430         blkbits = inode->i_sb->s_blocksize_bits;
 431         start = offset >> blkbits;
 432         last = start;
 433         end = isize >> blkbits;
 434         dataoff = offset;
 435
 436         do {
 437                 map.m_lblk = last;
 438                 map.m_len = end - last + 1;
 439                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 440                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 441                         if (last != start)
 442                                 dataoff = (loff_t)last << blkbits;
 443                         break;
 444                 }
 445
 446                 /*
 447                  * If there is a delay extent at this offset,
 448                  * it will be as a data.
 449                  */
 450                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 451                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 452                         if (last != start)
 453                                 dataoff = (loff_t)last << blkbits;
 454                         break;
 455                 }
 456
 457                 /*
 458                  * If there is a unwritten extent at this offset,
 459                  * it will be as a data or a hole according to page
 460                  * cache that has data or not.
 461                  */
 462                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 463                         int unwritten;
 464                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 465                                                               &map, &dataoff);
 466                         if (unwritten)
 467                                 break;
 468                 }
 469
 470                 last++;
 471                 dataoff = (loff_t)last << blkbits;
 472         } while (last <= end);
 473
 474         mutex_unlock(&inode->i_mutex);
 475
 476         if (dataoff > isize)
 477                 return -ENXIO;
 478
 479         return vfs_setpos(file, dataoff, maxsize);
 480 }
 481
 482 /*
 483  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 484  */
 485 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 486 {
 487         struct inode *inode = file->f_mapping->host;
 488         struct ext4_map_blocks map;
 489         struct extent_status es;
 490         ext4_lblk_t start, last, end;
 491         loff_t holeoff, isize;
 492         int blkbits;
 493         int ret = 0;
 494
 495         mutex_lock(&inode->i_mutex);
 496
 497         isize = i_size_read(inode);
 498         if (offset >= isize) {
 499                 mutex_unlock(&inode->i_mutex);
 500                 return -ENXIO;
 501         }
 502
 503         blkbits = inode->i_sb->s_blocksize_bits;
 504         start = offset >> blkbits;
 505         last = start;
 506         end = isize >> blkbits;
 507         holeoff = offset;
 508
 509         do {
 510                 map.m_lblk = last;
 511                 map.m_len = end - last + 1;
 512                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 513                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 514                         last += ret;
 515                         holeoff = (loff_t)last << blkbits;
 516                         continue;
 517                 }
 518
 519                 /*
 520                  * If there is a delay extent at this offset,
 521                  * we will skip this extent.
 522                  */
 523                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 524                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 525                         last = es.es_lblk + es.es_len;
 526                         holeoff = (loff_t)last << blkbits;
 527                         continue;
 528                 }
 529
 530                 /*
 531                  * If there is a unwritten extent at this offset,
 532                  * it will be as a data or a hole according to page
 533                  * cache that has data or not.
 534                  */
 535                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 536                         int unwritten;
 537                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 538                                                               &map, &holeoff);
 539                         if (!unwritten) {
 540                                 last += ret;
 541                                 holeoff = (loff_t)last << blkbits;
 542                                 continue;
 543                         }
 544                 }
 545
 546                 /* find a hole */
 547                 break;
 548         } while (last <= end);
 549
 550         mutex_unlock(&inode->i_mutex);
 551
 552         if (holeoff > isize)
 553                 holeoff = isize;
 554
 555         return vfs_setpos(file, holeoff, maxsize);
 556 }
 557
 558 /*
 559  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 560  * by calling generic_file_llseek_size() with the appropriate maxbytes
 561  * value for each.
 562  */
 563 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 564 {
 565         struct inode *inode = file->f_mapping->host;
 566         loff_t maxbytes;
 567
 568         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 569                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 570         else
 571                 maxbytes = inode->i_sb->s_maxbytes;
 572
 573         switch (whence) {
 574         case SEEK_SET:
 575         case SEEK_CUR:
 576         case SEEK_END:
 577                 return generic_file_llseek_size(file, offset, whence,
 578                                                 maxbytes, i_size_read(inode));
 579         case SEEK_DATA:
 580                 return ext4_seek_data(file, offset, maxbytes);
 581         case SEEK_HOLE:
 582                 return ext4_seek_hole(file, offset, maxbytes);
 583         }
 584
 585         return -EINVAL;
 586 }
 587
 588 const struct file_operations ext4_file_operations = {
 589         .llseek         = ext4_llseek,
 590         .read           = new_sync_read,
 591         .write          = new_sync_write,
 592         .read_iter      = generic_file_read_iter,
 593         .write_iter     = ext4_file_write_iter,
 594         .unlocked_ioctl = ext4_ioctl,
 595 #ifdef CONFIG_COMPAT
 596         .compat_ioctl   = ext4_compat_ioctl,
 597 #endif
 598         .mmap           = ext4_file_mmap,
 599         .open           = ext4_file_open,
 600         .release        = ext4_release_file,
 601         .fsync          = ext4_sync_file,
 602         .splice_read    = generic_file_splice_read,
 603         .splice_write   = iter_file_splice_write,
 604         .fallocate      = ext4_fallocate,
 605 };
 606
 607 const struct inode_operations ext4_file_inode_operations = {
 608         .setattr        = ext4_setattr,
 609         .getattr        = ext4_getattr,
 610         .setxattr       = generic_setxattr,
 611         .getxattr       = generic_getxattr,
 612         .listxattr      = ext4_listxattr,
 613         .removexattr    = generic_removexattr,
 614         .get_acl        = ext4_get_acl,
 615         .set_acl        = ext4_set_acl,
 616         .fiemap         = ext4_fiemap,
 617 };
 618