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, const struct iovec *iov,
  78                    unsigned long nr_segs, loff_t pos)
  79 {
  80         struct super_block *sb = inode->i_sb;
  81         int blockmask = sb->s_blocksize - 1;
  82         size_t count = iov_length(iov, nr_segs);
  83         loff_t final_size = pos + count;
  84
  85         if (pos >= i_size_read(inode))
  86                 return 0;
  87
  88         if ((pos & blockmask) || (final_size & blockmask))
  89                 return 1;
  90
  91         return 0;
  92 }
  93
  94 static ssize_t
  95 ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
  96                 unsigned long nr_segs, loff_t pos)
  97 {
  98         struct file *file = iocb->ki_filp;
  99         struct inode *inode = file_inode(iocb->ki_filp);
 100         struct mutex *aio_mutex = NULL;
 101         struct blk_plug plug;
 102         int o_direct = file->f_flags & O_DIRECT;
 103         int overwrite = 0;
 104         size_t length = iov_length(iov, nr_segs);
 105         ssize_t ret;
 106
 107         BUG_ON(iocb->ki_pos != pos);
 108
 109         /*
 110          * Unaligned direct AIO must be serialized; see comment above
 111          * In the case of O_APPEND, assume that we must always serialize
 112          */
 113         if (o_direct &&
 114             ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 115             !is_sync_kiocb(iocb) &&
 116             (file->f_flags & O_APPEND ||
 117              ext4_unaligned_aio(inode, iov, nr_segs, pos))) {
 118                 aio_mutex = ext4_aio_mutex(inode);
 119                 mutex_lock(aio_mutex);
 120                 ext4_unwritten_wait(inode);
 121         }
 122
 123         mutex_lock(&inode->i_mutex);
 124         if (file->f_flags & O_APPEND)
 125                 iocb->ki_pos = pos = i_size_read(inode);
 126
 127         /*
 128          * If we have encountered a bitmap-format file, the size limit
 129          * is smaller than s_maxbytes, which is for extent-mapped files.
 130          */
 131         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 132                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 133
 134                 if ((pos > sbi->s_bitmap_maxbytes) ||
 135                     (pos == sbi->s_bitmap_maxbytes && length > 0)) {
 136                         mutex_unlock(&inode->i_mutex);
 137                         ret = -EFBIG;
 138                         goto errout;
 139                 }
 140
 141                 if (pos + length > sbi->s_bitmap_maxbytes) {
 142                         nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
 143                                               sbi->s_bitmap_maxbytes - pos);
 144                 }
 145         }
 146
 147         if (o_direct) {
 148                 blk_start_plug(&plug);
 149
 150                 iocb->private = &overwrite;
 151
 152                 /* check whether we do a DIO overwrite or not */
 153                 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
 154                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 155                         struct ext4_map_blocks map;
 156                         unsigned int blkbits = inode->i_blkbits;
 157                         int err, len;
 158
 159                         map.m_lblk = pos >> blkbits;
 160                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 161                                 - map.m_lblk;
 162                         len = map.m_len;
 163
 164                         err = ext4_map_blocks(NULL, inode, &map, 0);
 165                         /*
 166                          * 'err==len' means that all of blocks has
 167                          * been preallocated no matter they are
 168                          * initialized or not.  For excluding
 169                          * unwritten extents, we need to check
 170                          * m_flags.  There are two conditions that
 171                          * indicate for initialized extents.  1) If we
 172                          * hit extent cache, EXT4_MAP_MAPPED flag is
 173                          * returned; 2) If we do a real lookup,
 174                          * non-flags are returned.  So we should check
 175                          * these two conditions.
 176                          */
 177                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 178                                 overwrite = 1;
 179                 }
 180         }
 181
 182         ret = __generic_file_aio_write(iocb, iov, nr_segs);
 183         mutex_unlock(&inode->i_mutex);
 184
 185         if (ret > 0) {
 186                 ssize_t err;
 187
 188                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
 189                 if (err < 0)
 190                         ret = err;
 191         }
 192         if (o_direct)
 193                 blk_finish_plug(&plug);
 194
 195 errout:
 196         if (aio_mutex)
 197                 mutex_unlock(aio_mutex);
 198         return ret;
 199 }
 200
 201 static const struct vm_operations_struct ext4_file_vm_ops = {
 202         .fault          = filemap_fault,
 203         .map_pages      = filemap_map_pages,
 204         .page_mkwrite   = ext4_page_mkwrite,
 205         .remap_pages    = generic_file_remap_pages,
 206 };
 207
 208 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 209 {
 210         struct address_space *mapping = file->f_mapping;
 211
 212         if (!mapping->a_ops->readpage)
 213                 return -ENOEXEC;
 214         file_accessed(file);
 215         vma->vm_ops = &ext4_file_vm_ops;
 216         return 0;
 217 }
 218
 219 static int ext4_file_open(struct inode * inode, struct file * filp)
 220 {
 221         struct super_block *sb = inode->i_sb;
 222         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 223         struct vfsmount *mnt = filp->f_path.mnt;
 224         struct path path;
 225         char buf[64], *cp;
 226
 227         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 228                      !(sb->s_flags & MS_RDONLY))) {
 229                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 230                 /*
 231                  * Sample where the filesystem has been mounted and
 232                  * store it in the superblock for sysadmin convenience
 233                  * when trying to sort through large numbers of block
 234                  * devices or filesystem images.
 235                  */
 236                 memset(buf, 0, sizeof(buf));
 237                 path.mnt = mnt;
 238                 path.dentry = mnt->mnt_root;
 239                 cp = d_path(&path, buf, sizeof(buf));
 240                 if (!IS_ERR(cp)) {
 241                         handle_t *handle;
 242                         int err;
 243
 244                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 245                         if (IS_ERR(handle))
 246                                 return PTR_ERR(handle);
 247                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 248                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 249                         if (err) {
 250                                 ext4_journal_stop(handle);
 251                                 return err;
 252                         }
 253                         strlcpy(sbi->s_es->s_last_mounted, cp,
 254                                 sizeof(sbi->s_es->s_last_mounted));
 255                         ext4_handle_dirty_super(handle, sb);
 256                         ext4_journal_stop(handle);
 257                 }
 258         }
 259         /*
 260          * Set up the jbd2_inode if we are opening the inode for
 261          * writing and the journal is present
 262          */
 263         if (filp->f_mode & FMODE_WRITE) {
 264                 int ret = ext4_inode_attach_jinode(inode);
 265                 if (ret < 0)
 266                         return ret;
 267         }
 268         return dquot_file_open(inode, filp);
 269 }
 270
 271 /*
 272  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 273  * file rather than ext4_ext_walk_space() because we can introduce
 274  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 275  * function.  When extent status tree has been fully implemented, it will
 276  * track all extent status for a file and we can directly use it to
 277  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 278  */
 279
 280 /*
 281  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 282  * lookup page cache to check whether or not there has some data between
 283  * [startoff, endoff] because, if this range contains an unwritten extent,
 284  * we determine this extent as a data or a hole according to whether the
 285  * page cache has data or not.
 286  */
 287 static int ext4_find_unwritten_pgoff(struct inode *inode,
 288                                      int whence,
 289                                      struct ext4_map_blocks *map,
 290                                      loff_t *offset)
 291 {
 292         struct pagevec pvec;
 293         unsigned int blkbits;
 294         pgoff_t index;
 295         pgoff_t end;
 296         loff_t endoff;
 297         loff_t startoff;
 298         loff_t lastoff;
 299         int found = 0;
 300
 301         blkbits = inode->i_sb->s_blocksize_bits;
 302         startoff = *offset;
 303         lastoff = startoff;
 304         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 305
 306         index = startoff >> PAGE_CACHE_SHIFT;
 307         end = endoff >> PAGE_CACHE_SHIFT;
 308
 309         pagevec_init(&pvec, 0);
 310         do {
 311                 int i, num;
 312                 unsigned long nr_pages;
 313
 314                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 315                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 316                                           (pgoff_t)num);
 317                 if (nr_pages == 0) {
 318                         if (whence == SEEK_DATA)
 319                                 break;
 320
 321                         BUG_ON(whence != SEEK_HOLE);
 322                         /*
 323                          * If this is the first time to go into the loop and
 324                          * offset is not beyond the end offset, it will be a
 325                          * hole at this offset
 326                          */
 327                         if (lastoff == startoff || lastoff < endoff)
 328                                 found = 1;
 329                         break;
 330                 }
 331
 332                 /*
 333                  * If this is the first time to go into the loop and
 334                  * offset is smaller than the first page offset, it will be a
 335                  * hole at this offset.
 336                  */
 337                 if (lastoff == startoff && whence == SEEK_HOLE &&
 338                     lastoff < page_offset(pvec.pages[0])) {
 339                         found = 1;
 340                         break;
 341                 }
 342
 343                 for (i = 0; i < nr_pages; i++) {
 344                         struct page *page = pvec.pages[i];
 345                         struct buffer_head *bh, *head;
 346
 347                         /*
 348                          * If the current offset is not beyond the end of given
 349                          * range, it will be a hole.
 350                          */
 351                         if (lastoff < endoff && whence == SEEK_HOLE &&
 352                             page->index > end) {
 353                                 found = 1;
 354                                 *offset = lastoff;
 355                                 goto out;
 356                         }
 357
 358                         lock_page(page);
 359
 360                         if (unlikely(page->mapping != inode->i_mapping)) {
 361                                 unlock_page(page);
 362                                 continue;
 363                         }
 364
 365                         if (!page_has_buffers(page)) {
 366                                 unlock_page(page);
 367                                 continue;
 368                         }
 369
 370                         if (page_has_buffers(page)) {
 371                                 lastoff = page_offset(page);
 372                                 bh = head = page_buffers(page);
 373                                 do {
 374                                         if (buffer_uptodate(bh) ||
 375                                             buffer_unwritten(bh)) {
 376                                                 if (whence == SEEK_DATA)
 377                                                         found = 1;
 378                                         } else {
 379                                                 if (whence == SEEK_HOLE)
 380                                                         found = 1;
 381                                         }
 382                                         if (found) {
 383                                                 *offset = max_t(loff_t,
 384                                                         startoff, lastoff);
 385                                                 unlock_page(page);
 386                                                 goto out;
 387                                         }
 388                                         lastoff += bh->b_size;
 389                                         bh = bh->b_this_page;
 390                                 } while (bh != head);
 391                         }
 392
 393                         lastoff = page_offset(page) + PAGE_SIZE;
 394                         unlock_page(page);
 395                 }
 396
 397                 /*
 398                  * The no. of pages is less than our desired, that would be a
 399                  * hole in there.
 400                  */
 401                 if (nr_pages < num && whence == SEEK_HOLE) {
 402                         found = 1;
 403                         *offset = lastoff;
 404                         break;
 405                 }
 406
 407                 index = pvec.pages[i - 1]->index + 1;
 408                 pagevec_release(&pvec);
 409         } while (index <= end);
 410
 411 out:
 412         pagevec_release(&pvec);
 413         return found;
 414 }
 415
 416 /*
 417  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 418  */
 419 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 420 {
 421         struct inode *inode = file->f_mapping->host;
 422         struct ext4_map_blocks map;
 423         struct extent_status es;
 424         ext4_lblk_t start, last, end;
 425         loff_t dataoff, isize;
 426         int blkbits;
 427         int ret = 0;
 428
 429         mutex_lock(&inode->i_mutex);
 430
 431         isize = i_size_read(inode);
 432         if (offset >= isize) {
 433                 mutex_unlock(&inode->i_mutex);
 434                 return -ENXIO;
 435         }
 436
 437         blkbits = inode->i_sb->s_blocksize_bits;
 438         start = offset >> blkbits;
 439         last = start;
 440         end = isize >> blkbits;
 441         dataoff = offset;
 442
 443         do {
 444                 map.m_lblk = last;
 445                 map.m_len = end - last + 1;
 446                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 447                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 448                         if (last != start)
 449                                 dataoff = (loff_t)last << blkbits;
 450                         break;
 451                 }
 452
 453                 /*
 454                  * If there is a delay extent at this offset,
 455                  * it will be as a data.
 456                  */
 457                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 458                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 459                         if (last != start)
 460                                 dataoff = (loff_t)last << blkbits;
 461                         break;
 462                 }
 463
 464                 /*
 465                  * If there is a unwritten extent at this offset,
 466                  * it will be as a data or a hole according to page
 467                  * cache that has data or not.
 468                  */
 469                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 470                         int unwritten;
 471                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 472                                                               &map, &dataoff);
 473                         if (unwritten)
 474                                 break;
 475                 }
 476
 477                 last++;
 478                 dataoff = (loff_t)last << blkbits;
 479         } while (last <= end);
 480
 481         mutex_unlock(&inode->i_mutex);
 482
 483         if (dataoff > isize)
 484                 return -ENXIO;
 485
 486         return vfs_setpos(file, dataoff, maxsize);
 487 }
 488
 489 /*
 490  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 491  */
 492 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 493 {
 494         struct inode *inode = file->f_mapping->host;
 495         struct ext4_map_blocks map;
 496         struct extent_status es;
 497         ext4_lblk_t start, last, end;
 498         loff_t holeoff, isize;
 499         int blkbits;
 500         int ret = 0;
 501
 502         mutex_lock(&inode->i_mutex);
 503
 504         isize = i_size_read(inode);
 505         if (offset >= isize) {
 506                 mutex_unlock(&inode->i_mutex);
 507                 return -ENXIO;
 508         }
 509
 510         blkbits = inode->i_sb->s_blocksize_bits;
 511         start = offset >> blkbits;
 512         last = start;
 513         end = isize >> blkbits;
 514         holeoff = offset;
 515
 516         do {
 517                 map.m_lblk = last;
 518                 map.m_len = end - last + 1;
 519                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 520                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 521                         last += ret;
 522                         holeoff = (loff_t)last << blkbits;
 523                         continue;
 524                 }
 525
 526                 /*
 527                  * If there is a delay extent at this offset,
 528                  * we will skip this extent.
 529                  */
 530                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 531                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 532                         last = es.es_lblk + es.es_len;
 533                         holeoff = (loff_t)last << blkbits;
 534                         continue;
 535                 }
 536
 537                 /*
 538                  * If there is a unwritten extent at this offset,
 539                  * it will be as a data or a hole according to page
 540                  * cache that has data or not.
 541                  */
 542                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 543                         int unwritten;
 544                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 545                                                               &map, &holeoff);
 546                         if (!unwritten) {
 547                                 last += ret;
 548                                 holeoff = (loff_t)last << blkbits;
 549                                 continue;
 550                         }
 551                 }
 552
 553                 /* find a hole */
 554                 break;
 555         } while (last <= end);
 556
 557         mutex_unlock(&inode->i_mutex);
 558
 559         if (holeoff > isize)
 560                 holeoff = isize;
 561
 562         return vfs_setpos(file, holeoff, maxsize);
 563 }
 564
 565 /*
 566  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 567  * by calling generic_file_llseek_size() with the appropriate maxbytes
 568  * value for each.
 569  */
 570 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 571 {
 572         struct inode *inode = file->f_mapping->host;
 573         loff_t maxbytes;
 574
 575         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 576                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 577         else
 578                 maxbytes = inode->i_sb->s_maxbytes;
 579
 580         switch (whence) {
 581         case SEEK_SET:
 582         case SEEK_CUR:
 583         case SEEK_END:
 584                 return generic_file_llseek_size(file, offset, whence,
 585                                                 maxbytes, i_size_read(inode));
 586         case SEEK_DATA:
 587                 return ext4_seek_data(file, offset, maxbytes);
 588         case SEEK_HOLE:
 589                 return ext4_seek_hole(file, offset, maxbytes);
 590         }
 591
 592         return -EINVAL;
 593 }
 594
 595 const struct file_operations ext4_file_operations = {
 596         .llseek         = ext4_llseek,
 597         .read           = do_sync_read,
 598         .write          = do_sync_write,
 599         .aio_read       = generic_file_aio_read,
 600         .aio_write      = ext4_file_write,
 601         .unlocked_ioctl = ext4_ioctl,
 602 #ifdef CONFIG_COMPAT
 603         .compat_ioctl   = ext4_compat_ioctl,
 604 #endif
 605         .mmap           = ext4_file_mmap,
 606         .open           = ext4_file_open,
 607         .release        = ext4_release_file,
 608         .fsync          = ext4_sync_file,
 609         .splice_read    = generic_file_splice_read,
 610         .splice_write   = generic_file_splice_write,
 611         .fallocate      = ext4_fallocate,
 612 };
 613
 614 const struct inode_operations ext4_file_inode_operations = {
 615         .setattr        = ext4_setattr,
 616         .getattr        = ext4_getattr,
 617         .setxattr       = generic_setxattr,
 618         .getxattr       = generic_getxattr,
 619         .listxattr      = ext4_listxattr,
 620         .removexattr    = generic_removexattr,
 621         .get_acl        = ext4_get_acl,
 622         .set_acl        = ext4_set_acl,
 623         .fiemap         = ext4_fiemap,
 624 };
 625