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 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 >= inode->i_size)
  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_dio_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->f_mapping->host;
 100         struct blk_plug plug;
 101         int unaligned_aio = 0;
 102         ssize_t ret;
 103         int overwrite = 0;
 104         size_t length = iov_length(iov, nr_segs);
 105
 106         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 107             !is_sync_kiocb(iocb))
 108                 unaligned_aio = ext4_unaligned_aio(inode, iov, nr_segs, pos);
 109
 110         /* Unaligned direct AIO must be serialized; see comment above */
 111         if (unaligned_aio) {
 112                 mutex_lock(ext4_aio_mutex(inode));
 113                 ext4_unwritten_wait(inode);
 114         }
 115
 116         BUG_ON(iocb->ki_pos != pos);
 117
 118         mutex_lock(&inode->i_mutex);
 119         blk_start_plug(&plug);
 120
 121         iocb->private = &overwrite;
 122
 123         /* check whether we do a DIO overwrite or not */
 124         if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
 125             !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 126                 struct ext4_map_blocks map;
 127                 unsigned int blkbits = inode->i_blkbits;
 128                 int err, len;
 129
 130                 map.m_lblk = pos >> blkbits;
 131                 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 132                         - map.m_lblk;
 133                 len = map.m_len;
 134
 135                 err = ext4_map_blocks(NULL, inode, &map, 0);
 136                 /*
 137                  * 'err==len' means that all of blocks has been preallocated no
 138                  * matter they are initialized or not.  For excluding
 139                  * uninitialized extents, we need to check m_flags.  There are
 140                  * two conditions that indicate for initialized extents.
 141                  * 1) If we hit extent cache, EXT4_MAP_MAPPED flag is returned;
 142                  * 2) If we do a real lookup, non-flags are returned.
 143                  * So we should check these two conditions.
 144                  */
 145                 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 146                         overwrite = 1;
 147         }
 148
 149         ret = __generic_file_aio_write(iocb, iov, nr_segs);
 150         mutex_unlock(&inode->i_mutex);
 151
 152         if (ret > 0) {
 153                 ssize_t err;
 154
 155                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
 156                 if (err < 0)
 157                         ret = err;
 158         }
 159         blk_finish_plug(&plug);
 160
 161         if (unaligned_aio)
 162                 mutex_unlock(ext4_aio_mutex(inode));
 163
 164         return ret;
 165 }
 166
 167 static ssize_t
 168 ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
 169                 unsigned long nr_segs, loff_t pos)
 170 {
 171         struct inode *inode = file_inode(iocb->ki_filp);
 172         ssize_t ret;
 173
 174         /*
 175          * If we have encountered a bitmap-format file, the size limit
 176          * is smaller than s_maxbytes, which is for extent-mapped files.
 177          */
 178
 179         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 180                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 181                 size_t length = iov_length(iov, nr_segs);
 182
 183                 if ((pos > sbi->s_bitmap_maxbytes ||
 184                     (pos == sbi->s_bitmap_maxbytes && length > 0)))
 185                         return -EFBIG;
 186
 187                 if (pos + length > sbi->s_bitmap_maxbytes) {
 188                         nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
 189                                               sbi->s_bitmap_maxbytes - pos);
 190                 }
 191         }
 192
 193         if (unlikely(iocb->ki_filp->f_flags & O_DIRECT))
 194                 ret = ext4_file_dio_write(iocb, iov, nr_segs, pos);
 195         else
 196                 ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
 197
 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                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 248                         if (err) {
 249                                 ext4_journal_stop(handle);
 250                                 return err;
 251                         }
 252                         strlcpy(sbi->s_es->s_last_mounted, cp,
 253                                 sizeof(sbi->s_es->s_last_mounted));
 254                         ext4_handle_dirty_super(handle, sb);
 255                         ext4_journal_stop(handle);
 256                 }
 257         }
 258         /*
 259          * Set up the jbd2_inode if we are opening the inode for
 260          * writing and the journal is present
 261          */
 262         if (filp->f_mode & FMODE_WRITE) {
 263                 int ret = ext4_inode_attach_jinode(inode);
 264                 if (ret < 0)
 265                         return ret;
 266         }
 267         return dquot_file_open(inode, filp);
 268 }
 269
 270 /*
 271  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 272  * file rather than ext4_ext_walk_space() because we can introduce
 273  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 274  * function.  When extent status tree has been fully implemented, it will
 275  * track all extent status for a file and we can directly use it to
 276  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 277  */
 278
 279 /*
 280  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 281  * lookup page cache to check whether or not there has some data between
 282  * [startoff, endoff] because, if this range contains an unwritten extent,
 283  * we determine this extent as a data or a hole according to whether the
 284  * page cache has data or not.
 285  */
 286 static int ext4_find_unwritten_pgoff(struct inode *inode,
 287                                      int whence,
 288                                      struct ext4_map_blocks *map,
 289                                      loff_t *offset)
 290 {
 291         struct pagevec pvec;
 292         unsigned int blkbits;
 293         pgoff_t index;
 294         pgoff_t end;
 295         loff_t endoff;
 296         loff_t startoff;
 297         loff_t lastoff;
 298         int found = 0;
 299
 300         blkbits = inode->i_sb->s_blocksize_bits;
 301         startoff = *offset;
 302         lastoff = startoff;
 303         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 304
 305         index = startoff >> PAGE_CACHE_SHIFT;
 306         end = endoff >> PAGE_CACHE_SHIFT;
 307
 308         pagevec_init(&pvec, 0);
 309         do {
 310                 int i, num;
 311                 unsigned long nr_pages;
 312
 313                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 314                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 315                                           (pgoff_t)num);
 316                 if (nr_pages == 0) {
 317                         if (whence == SEEK_DATA)
 318                                 break;
 319
 320                         BUG_ON(whence != SEEK_HOLE);
 321                         /*
 322                          * If this is the first time to go into the loop and
 323                          * offset is not beyond the end offset, it will be a
 324                          * hole at this offset
 325                          */
 326                         if (lastoff == startoff || lastoff < endoff)
 327                                 found = 1;
 328                         break;
 329                 }
 330
 331                 /*
 332                  * If this is the first time to go into the loop and
 333                  * offset is smaller than the first page offset, it will be a
 334                  * hole at this offset.
 335                  */
 336                 if (lastoff == startoff && whence == SEEK_HOLE &&
 337                     lastoff < page_offset(pvec.pages[0])) {
 338                         found = 1;
 339                         break;
 340                 }
 341
 342                 for (i = 0; i < nr_pages; i++) {
 343                         struct page *page = pvec.pages[i];
 344                         struct buffer_head *bh, *head;
 345
 346                         /*
 347                          * If the current offset is not beyond the end of given
 348                          * range, it will be a hole.
 349                          */
 350                         if (lastoff < endoff && whence == SEEK_HOLE &&
 351                             page->index > end) {
 352                                 found = 1;
 353                                 *offset = lastoff;
 354                                 goto out;
 355                         }
 356
 357                         lock_page(page);
 358
 359                         if (unlikely(page->mapping != inode->i_mapping)) {
 360                                 unlock_page(page);
 361                                 continue;
 362                         }
 363
 364                         if (!page_has_buffers(page)) {
 365                                 unlock_page(page);
 366                                 continue;
 367                         }
 368
 369                         if (page_has_buffers(page)) {
 370                                 lastoff = page_offset(page);
 371                                 bh = head = page_buffers(page);
 372                                 do {
 373                                         if (buffer_uptodate(bh) ||
 374                                             buffer_unwritten(bh)) {
 375                                                 if (whence == SEEK_DATA)
 376                                                         found = 1;
 377                                         } else {
 378                                                 if (whence == SEEK_HOLE)
 379                                                         found = 1;
 380                                         }
 381                                         if (found) {
 382                                                 *offset = max_t(loff_t,
 383                                                         startoff, lastoff);
 384                                                 unlock_page(page);
 385                                                 goto out;
 386                                         }
 387                                         lastoff += bh->b_size;
 388                                         bh = bh->b_this_page;
 389                                 } while (bh != head);
 390                         }
 391
 392                         lastoff = page_offset(page) + PAGE_SIZE;
 393                         unlock_page(page);
 394                 }
 395
 396                 /*
 397                  * The no. of pages is less than our desired, that would be a
 398                  * hole in there.
 399                  */
 400                 if (nr_pages < num && whence == SEEK_HOLE) {
 401                         found = 1;
 402                         *offset = lastoff;
 403                         break;
 404                 }
 405
 406                 index = pvec.pages[i - 1]->index + 1;
 407                 pagevec_release(&pvec);
 408         } while (index <= end);
 409
 410 out:
 411         pagevec_release(&pvec);
 412         return found;
 413 }
 414
 415 /*
 416  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 417  */
 418 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 419 {
 420         struct inode *inode = file->f_mapping->host;
 421         struct ext4_map_blocks map;
 422         struct extent_status es;
 423         ext4_lblk_t start, last, end;
 424         loff_t dataoff, isize;
 425         int blkbits;
 426         int ret = 0;
 427
 428         mutex_lock(&inode->i_mutex);
 429
 430         isize = i_size_read(inode);
 431         if (offset >= isize) {
 432                 mutex_unlock(&inode->i_mutex);
 433                 return -ENXIO;
 434         }
 435
 436         blkbits = inode->i_sb->s_blocksize_bits;
 437         start = offset >> blkbits;
 438         last = start;
 439         end = isize >> blkbits;
 440         dataoff = offset;
 441
 442         do {
 443                 map.m_lblk = last;
 444                 map.m_len = end - last + 1;
 445                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 446                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 447                         if (last != start)
 448                                 dataoff = (loff_t)last << blkbits;
 449                         break;
 450                 }
 451
 452                 /*
 453                  * If there is a delay extent at this offset,
 454                  * it will be as a data.
 455                  */
 456                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 457                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 458                         if (last != start)
 459                                 dataoff = (loff_t)last << blkbits;
 460                         break;
 461                 }
 462
 463                 /*
 464                  * If there is a unwritten extent at this offset,
 465                  * it will be as a data or a hole according to page
 466                  * cache that has data or not.
 467                  */
 468                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 469                         int unwritten;
 470                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 471                                                               &map, &dataoff);
 472                         if (unwritten)
 473                                 break;
 474                 }
 475
 476                 last++;
 477                 dataoff = (loff_t)last << blkbits;
 478         } while (last <= end);
 479
 480         mutex_unlock(&inode->i_mutex);
 481
 482         if (dataoff > isize)
 483                 return -ENXIO;
 484
 485         return vfs_setpos(file, dataoff, maxsize);
 486 }
 487
 488 /*
 489  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 490  */
 491 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 492 {
 493         struct inode *inode = file->f_mapping->host;
 494         struct ext4_map_blocks map;
 495         struct extent_status es;
 496         ext4_lblk_t start, last, end;
 497         loff_t holeoff, isize;
 498         int blkbits;
 499         int ret = 0;
 500
 501         mutex_lock(&inode->i_mutex);
 502
 503         isize = i_size_read(inode);
 504         if (offset >= isize) {
 505                 mutex_unlock(&inode->i_mutex);
 506                 return -ENXIO;
 507         }
 508
 509         blkbits = inode->i_sb->s_blocksize_bits;
 510         start = offset >> blkbits;
 511         last = start;
 512         end = isize >> blkbits;
 513         holeoff = offset;
 514
 515         do {
 516                 map.m_lblk = last;
 517                 map.m_len = end - last + 1;
 518                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 519                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 520                         last += ret;
 521                         holeoff = (loff_t)last << blkbits;
 522                         continue;
 523                 }
 524
 525                 /*
 526                  * If there is a delay extent at this offset,
 527                  * we will skip this extent.
 528                  */
 529                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 530                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 531                         last = es.es_lblk + es.es_len;
 532                         holeoff = (loff_t)last << blkbits;
 533                         continue;
 534                 }
 535
 536                 /*
 537                  * If there is a unwritten extent at this offset,
 538                  * it will be as a data or a hole according to page
 539                  * cache that has data or not.
 540                  */
 541                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 542                         int unwritten;
 543                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 544                                                               &map, &holeoff);
 545                         if (!unwritten) {
 546                                 last += ret;
 547                                 holeoff = (loff_t)last << blkbits;
 548                                 continue;
 549                         }
 550                 }
 551
 552                 /* find a hole */
 553                 break;
 554         } while (last <= end);
 555
 556         mutex_unlock(&inode->i_mutex);
 557
 558         if (holeoff > isize)
 559                 holeoff = isize;
 560
 561         return vfs_setpos(file, holeoff, maxsize);
 562 }
 563
 564 /*
 565  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 566  * by calling generic_file_llseek_size() with the appropriate maxbytes
 567  * value for each.
 568  */
 569 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 570 {
 571         struct inode *inode = file->f_mapping->host;
 572         loff_t maxbytes;
 573
 574         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 575                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 576         else
 577                 maxbytes = inode->i_sb->s_maxbytes;
 578
 579         switch (whence) {
 580         case SEEK_SET:
 581         case SEEK_CUR:
 582         case SEEK_END:
 583                 return generic_file_llseek_size(file, offset, whence,
 584                                                 maxbytes, i_size_read(inode));
 585         case SEEK_DATA:
 586                 return ext4_seek_data(file, offset, maxbytes);
 587         case SEEK_HOLE:
 588                 return ext4_seek_hole(file, offset, maxbytes);
 589         }
 590
 591         return -EINVAL;
 592 }
 593
 594 const struct file_operations ext4_file_operations = {
 595         .llseek         = ext4_llseek,
 596         .read           = do_sync_read,
 597         .write          = do_sync_write,
 598         .aio_read       = generic_file_aio_read,
 599         .aio_write      = ext4_file_write,
 600         .unlocked_ioctl = ext4_ioctl,
 601 #ifdef CONFIG_COMPAT
 602         .compat_ioctl   = ext4_compat_ioctl,
 603 #endif
 604         .mmap           = ext4_file_mmap,
 605         .open           = ext4_file_open,
 606         .release        = ext4_release_file,
 607         .fsync          = ext4_sync_file,
 608         .splice_read    = generic_file_splice_read,
 609         .splice_write   = generic_file_splice_write,
 610         .fallocate      = ext4_fallocate,
 611 };
 612
 613 const struct inode_operations ext4_file_inode_operations = {
 614         .setattr        = ext4_setattr,
 615         .getattr        = ext4_getattr,
 616         .setxattr       = generic_setxattr,
 617         .getxattr       = generic_getxattr,
 618         .listxattr      = ext4_listxattr,
 619         .removexattr    = generic_removexattr,
 620         .get_acl        = ext4_get_acl,
 621         .set_acl        = ext4_set_acl,
 622         .fiemap         = ext4_fiemap,
 623 };
 624