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, &iocb->ki_pos);
 150         mutex_unlock(&inode->i_mutex);
 151
 152         if (ret > 0 || ret == -EIOCBQUEUED) {
 153                 ssize_t err;
 154
 155                 err = generic_write_sync(file, pos, ret);
 156                 if (err < 0 && ret > 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         .page_mkwrite   = ext4_page_mkwrite,
 204         .remap_pages    = generic_file_remap_pages,
 205 };
 206
 207 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 208 {
 209         struct address_space *mapping = file->f_mapping;
 210
 211         if (!mapping->a_ops->readpage)
 212                 return -ENOEXEC;
 213         file_accessed(file);
 214         vma->vm_ops = &ext4_file_vm_ops;
 215         return 0;
 216 }
 217
 218 static int ext4_file_open(struct inode * inode, struct file * filp)
 219 {
 220         struct super_block *sb = inode->i_sb;
 221         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 222         struct ext4_inode_info *ei = EXT4_I(inode);
 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 (sbi->s_journal && !ei->jinode && (filp->f_mode & FMODE_WRITE)) {
 263                 struct jbd2_inode *jinode = jbd2_alloc_inode(GFP_KERNEL);
 264
 265                 spin_lock(&inode->i_lock);
 266                 if (!ei->jinode) {
 267                         if (!jinode) {
 268                                 spin_unlock(&inode->i_lock);
 269                                 return -ENOMEM;
 270                         }
 271                         ei->jinode = jinode;
 272                         jbd2_journal_init_jbd_inode(ei->jinode, inode);
 273                         jinode = NULL;
 274                 }
 275                 spin_unlock(&inode->i_lock);
 276                 if (unlikely(jinode != NULL))
 277                         jbd2_free_inode(jinode);
 278         }
 279         return dquot_file_open(inode, filp);
 280 }
 281
 282 /*
 283  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 284  * file rather than ext4_ext_walk_space() because we can introduce
 285  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 286  * function.  When extent status tree has been fully implemented, it will
 287  * track all extent status for a file and we can directly use it to
 288  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 289  */
 290
 291 /*
 292  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 293  * lookup page cache to check whether or not there has some data between
 294  * [startoff, endoff] because, if this range contains an unwritten extent,
 295  * we determine this extent as a data or a hole according to whether the
 296  * page cache has data or not.
 297  */
 298 static int ext4_find_unwritten_pgoff(struct inode *inode,
 299                                      int whence,
 300                                      struct ext4_map_blocks *map,
 301                                      loff_t *offset)
 302 {
 303         struct pagevec pvec;
 304         unsigned int blkbits;
 305         pgoff_t index;
 306         pgoff_t end;
 307         loff_t endoff;
 308         loff_t startoff;
 309         loff_t lastoff;
 310         int found = 0;
 311
 312         blkbits = inode->i_sb->s_blocksize_bits;
 313         startoff = *offset;
 314         lastoff = startoff;
 315         endoff = (map->m_lblk + map->m_len) << blkbits;
 316
 317         index = startoff >> PAGE_CACHE_SHIFT;
 318         end = endoff >> PAGE_CACHE_SHIFT;
 319
 320         pagevec_init(&pvec, 0);
 321         do {
 322                 int i, num;
 323                 unsigned long nr_pages;
 324
 325                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 326                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 327                                           (pgoff_t)num);
 328                 if (nr_pages == 0) {
 329                         if (whence == SEEK_DATA)
 330                                 break;
 331
 332                         BUG_ON(whence != SEEK_HOLE);
 333                         /*
 334                          * If this is the first time to go into the loop and
 335                          * offset is not beyond the end offset, it will be a
 336                          * hole at this offset
 337                          */
 338                         if (lastoff == startoff || lastoff < endoff)
 339                                 found = 1;
 340                         break;
 341                 }
 342
 343                 /*
 344                  * If this is the first time to go into the loop and
 345                  * offset is smaller than the first page offset, it will be a
 346                  * hole at this offset.
 347                  */
 348                 if (lastoff == startoff && whence == SEEK_HOLE &&
 349                     lastoff < page_offset(pvec.pages[0])) {
 350                         found = 1;
 351                         break;
 352                 }
 353
 354                 for (i = 0; i < nr_pages; i++) {
 355                         struct page *page = pvec.pages[i];
 356                         struct buffer_head *bh, *head;
 357
 358                         /*
 359                          * If the current offset is not beyond the end of given
 360                          * range, it will be a hole.
 361                          */
 362                         if (lastoff < endoff && whence == SEEK_HOLE &&
 363                             page->index > end) {
 364                                 found = 1;
 365                                 *offset = lastoff;
 366                                 goto out;
 367                         }
 368
 369                         lock_page(page);
 370
 371                         if (unlikely(page->mapping != inode->i_mapping)) {
 372                                 unlock_page(page);
 373                                 continue;
 374                         }
 375
 376                         if (!page_has_buffers(page)) {
 377                                 unlock_page(page);
 378                                 continue;
 379                         }
 380
 381                         if (page_has_buffers(page)) {
 382                                 lastoff = page_offset(page);
 383                                 bh = head = page_buffers(page);
 384                                 do {
 385                                         if (buffer_uptodate(bh) ||
 386                                             buffer_unwritten(bh)) {
 387                                                 if (whence == SEEK_DATA)
 388                                                         found = 1;
 389                                         } else {
 390                                                 if (whence == SEEK_HOLE)
 391                                                         found = 1;
 392                                         }
 393                                         if (found) {
 394                                                 *offset = max_t(loff_t,
 395                                                         startoff, lastoff);
 396                                                 unlock_page(page);
 397                                                 goto out;
 398                                         }
 399                                         lastoff += bh->b_size;
 400                                         bh = bh->b_this_page;
 401                                 } while (bh != head);
 402                         }
 403
 404                         lastoff = page_offset(page) + PAGE_SIZE;
 405                         unlock_page(page);
 406                 }
 407
 408                 /*
 409                  * The no. of pages is less than our desired, that would be a
 410                  * hole in there.
 411                  */
 412                 if (nr_pages < num && whence == SEEK_HOLE) {
 413                         found = 1;
 414                         *offset = lastoff;
 415                         break;
 416                 }
 417
 418                 index = pvec.pages[i - 1]->index + 1;
 419                 pagevec_release(&pvec);
 420         } while (index <= end);
 421
 422 out:
 423         pagevec_release(&pvec);
 424         return found;
 425 }
 426
 427 /*
 428  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 429  */
 430 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 431 {
 432         struct inode *inode = file->f_mapping->host;
 433         struct ext4_map_blocks map;
 434         struct extent_status es;
 435         ext4_lblk_t start, last, end;
 436         loff_t dataoff, isize;
 437         int blkbits;
 438         int ret = 0;
 439
 440         mutex_lock(&inode->i_mutex);
 441
 442         isize = i_size_read(inode);
 443         if (offset >= isize) {
 444                 mutex_unlock(&inode->i_mutex);
 445                 return -ENXIO;
 446         }
 447
 448         blkbits = inode->i_sb->s_blocksize_bits;
 449         start = offset >> blkbits;
 450         last = start;
 451         end = isize >> blkbits;
 452         dataoff = offset;
 453
 454         do {
 455                 map.m_lblk = last;
 456                 map.m_len = end - last + 1;
 457                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 458                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 459                         if (last != start)
 460                                 dataoff = last << blkbits;
 461                         break;
 462                 }
 463
 464                 /*
 465                  * If there is a delay extent at this offset,
 466                  * it will be as a data.
 467                  */
 468                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 469                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 470                         if (last != start)
 471                                 dataoff = last << blkbits;
 472                         break;
 473                 }
 474
 475                 /*
 476                  * If there is a unwritten extent at this offset,
 477                  * it will be as a data or a hole according to page
 478                  * cache that has data or not.
 479                  */
 480                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 481                         int unwritten;
 482                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 483                                                               &map, &dataoff);
 484                         if (unwritten)
 485                                 break;
 486                 }
 487
 488                 last++;
 489                 dataoff = last << blkbits;
 490         } while (last <= end);
 491
 492         mutex_unlock(&inode->i_mutex);
 493
 494         if (dataoff > isize)
 495                 return -ENXIO;
 496
 497         if (dataoff < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
 498                 return -EINVAL;
 499         if (dataoff > maxsize)
 500                 return -EINVAL;
 501
 502         if (dataoff != file->f_pos) {
 503                 file->f_pos = dataoff;
 504                 file->f_version = 0;
 505         }
 506
 507         return dataoff;
 508 }
 509
 510 /*
 511  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 512  */
 513 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 514 {
 515         struct inode *inode = file->f_mapping->host;
 516         struct ext4_map_blocks map;
 517         struct extent_status es;
 518         ext4_lblk_t start, last, end;
 519         loff_t holeoff, isize;
 520         int blkbits;
 521         int ret = 0;
 522
 523         mutex_lock(&inode->i_mutex);
 524
 525         isize = i_size_read(inode);
 526         if (offset >= isize) {
 527                 mutex_unlock(&inode->i_mutex);
 528                 return -ENXIO;
 529         }
 530
 531         blkbits = inode->i_sb->s_blocksize_bits;
 532         start = offset >> blkbits;
 533         last = start;
 534         end = isize >> blkbits;
 535         holeoff = offset;
 536
 537         do {
 538                 map.m_lblk = last;
 539                 map.m_len = end - last + 1;
 540                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 541                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 542                         last += ret;
 543                         holeoff = last << blkbits;
 544                         continue;
 545                 }
 546
 547                 /*
 548                  * If there is a delay extent at this offset,
 549                  * we will skip this extent.
 550                  */
 551                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 552                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 553                         last = es.es_lblk + es.es_len;
 554                         holeoff = last << blkbits;
 555                         continue;
 556                 }
 557
 558                 /*
 559                  * If there is a unwritten extent at this offset,
 560                  * it will be as a data or a hole according to page
 561                  * cache that has data or not.
 562                  */
 563                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 564                         int unwritten;
 565                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 566                                                               &map, &holeoff);
 567                         if (!unwritten) {
 568                                 last += ret;
 569                                 holeoff = last << blkbits;
 570                                 continue;
 571                         }
 572                 }
 573
 574                 /* find a hole */
 575                 break;
 576         } while (last <= end);
 577
 578         mutex_unlock(&inode->i_mutex);
 579
 580         if (holeoff > isize)
 581                 holeoff = isize;
 582
 583         if (holeoff < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
 584                 return -EINVAL;
 585         if (holeoff > maxsize)
 586                 return -EINVAL;
 587
 588         if (holeoff != file->f_pos) {
 589                 file->f_pos = holeoff;
 590                 file->f_version = 0;
 591         }
 592
 593         return holeoff;
 594 }
 595
 596 /*
 597  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 598  * by calling generic_file_llseek_size() with the appropriate maxbytes
 599  * value for each.
 600  */
 601 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 602 {
 603         struct inode *inode = file->f_mapping->host;
 604         loff_t maxbytes;
 605
 606         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 607                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 608         else
 609                 maxbytes = inode->i_sb->s_maxbytes;
 610
 611         switch (whence) {
 612         case SEEK_SET:
 613         case SEEK_CUR:
 614         case SEEK_END:
 615                 return generic_file_llseek_size(file, offset, whence,
 616                                                 maxbytes, i_size_read(inode));
 617         case SEEK_DATA:
 618                 return ext4_seek_data(file, offset, maxbytes);
 619         case SEEK_HOLE:
 620                 return ext4_seek_hole(file, offset, maxbytes);
 621         }
 622
 623         return -EINVAL;
 624 }
 625
 626 const struct file_operations ext4_file_operations = {
 627         .llseek         = ext4_llseek,
 628         .read           = do_sync_read,
 629         .write          = do_sync_write,
 630         .aio_read       = generic_file_aio_read,
 631         .aio_write      = ext4_file_write,
 632         .unlocked_ioctl = ext4_ioctl,
 633 #ifdef CONFIG_COMPAT
 634         .compat_ioctl   = ext4_compat_ioctl,
 635 #endif
 636         .mmap           = ext4_file_mmap,
 637         .open           = ext4_file_open,
 638         .release        = ext4_release_file,
 639         .fsync          = ext4_sync_file,
 640         .splice_read    = generic_file_splice_read,
 641         .splice_write   = generic_file_splice_write,
 642         .fallocate      = ext4_fallocate,
 643 };
 644
 645 const struct inode_operations ext4_file_inode_operations = {
 646         .setattr        = ext4_setattr,
 647         .getattr        = ext4_getattr,
 648         .setxattr       = generic_setxattr,
 649         .getxattr       = generic_getxattr,
 650         .listxattr      = ext4_listxattr,
 651         .removexattr    = generic_removexattr,
 652         .get_acl        = ext4_get_acl,
 653         .fiemap         = ext4_fiemap,
 654 };
 655