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) {
 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 vfsmount *mnt = filp->f_path.mnt;
 223         struct path path;
 224         char buf[64], *cp;
 225
 226         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 227                      !(sb->s_flags & MS_RDONLY))) {
 228                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 229                 /*
 230                  * Sample where the filesystem has been mounted and
 231                  * store it in the superblock for sysadmin convenience
 232                  * when trying to sort through large numbers of block
 233                  * devices or filesystem images.
 234                  */
 235                 memset(buf, 0, sizeof(buf));
 236                 path.mnt = mnt;
 237                 path.dentry = mnt->mnt_root;
 238                 cp = d_path(&path, buf, sizeof(buf));
 239                 if (!IS_ERR(cp)) {
 240                         handle_t *handle;
 241                         int err;
 242
 243                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 244                         if (IS_ERR(handle))
 245                                 return PTR_ERR(handle);
 246                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 247                         if (err) {
 248                                 ext4_journal_stop(handle);
 249                                 return err;
 250                         }
 251                         strlcpy(sbi->s_es->s_last_mounted, cp,
 252                                 sizeof(sbi->s_es->s_last_mounted));
 253                         ext4_handle_dirty_super(handle, sb);
 254                         ext4_journal_stop(handle);
 255                 }
 256         }
 257         /*
 258          * Set up the jbd2_inode if we are opening the inode for
 259          * writing and the journal is present
 260          */
 261         if (filp->f_mode & FMODE_WRITE) {
 262                 int ret = ext4_inode_attach_jinode(inode);
 263                 if (ret < 0)
 264                         return ret;
 265         }
 266         return dquot_file_open(inode, filp);
 267 }
 268
 269 /*
 270  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 271  * file rather than ext4_ext_walk_space() because we can introduce
 272  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 273  * function.  When extent status tree has been fully implemented, it will
 274  * track all extent status for a file and we can directly use it to
 275  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 276  */
 277
 278 /*
 279  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 280  * lookup page cache to check whether or not there has some data between
 281  * [startoff, endoff] because, if this range contains an unwritten extent,
 282  * we determine this extent as a data or a hole according to whether the
 283  * page cache has data or not.
 284  */
 285 static int ext4_find_unwritten_pgoff(struct inode *inode,
 286                                      int whence,
 287                                      struct ext4_map_blocks *map,
 288                                      loff_t *offset)
 289 {
 290         struct pagevec pvec;
 291         unsigned int blkbits;
 292         pgoff_t index;
 293         pgoff_t end;
 294         loff_t endoff;
 295         loff_t startoff;
 296         loff_t lastoff;
 297         int found = 0;
 298
 299         blkbits = inode->i_sb->s_blocksize_bits;
 300         startoff = *offset;
 301         lastoff = startoff;
 302         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 303
 304         index = startoff >> PAGE_CACHE_SHIFT;
 305         end = endoff >> PAGE_CACHE_SHIFT;
 306
 307         pagevec_init(&pvec, 0);
 308         do {
 309                 int i, num;
 310                 unsigned long nr_pages;
 311
 312                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 313                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 314                                           (pgoff_t)num);
 315                 if (nr_pages == 0) {
 316                         if (whence == SEEK_DATA)
 317                                 break;
 318
 319                         BUG_ON(whence != SEEK_HOLE);
 320                         /*
 321                          * If this is the first time to go into the loop and
 322                          * offset is not beyond the end offset, it will be a
 323                          * hole at this offset
 324                          */
 325                         if (lastoff == startoff || lastoff < endoff)
 326                                 found = 1;
 327                         break;
 328                 }
 329
 330                 /*
 331                  * If this is the first time to go into the loop and
 332                  * offset is smaller than the first page offset, it will be a
 333                  * hole at this offset.
 334                  */
 335                 if (lastoff == startoff && whence == SEEK_HOLE &&
 336                     lastoff < page_offset(pvec.pages[0])) {
 337                         found = 1;
 338                         break;
 339                 }
 340
 341                 for (i = 0; i < nr_pages; i++) {
 342                         struct page *page = pvec.pages[i];
 343                         struct buffer_head *bh, *head;
 344
 345                         /*
 346                          * If the current offset is not beyond the end of given
 347                          * range, it will be a hole.
 348                          */
 349                         if (lastoff < endoff && whence == SEEK_HOLE &&
 350                             page->index > end) {
 351                                 found = 1;
 352                                 *offset = lastoff;
 353                                 goto out;
 354                         }
 355
 356                         lock_page(page);
 357
 358                         if (unlikely(page->mapping != inode->i_mapping)) {
 359                                 unlock_page(page);
 360                                 continue;
 361                         }
 362
 363                         if (!page_has_buffers(page)) {
 364                                 unlock_page(page);
 365                                 continue;
 366                         }
 367
 368                         if (page_has_buffers(page)) {
 369                                 lastoff = page_offset(page);
 370                                 bh = head = page_buffers(page);
 371                                 do {
 372                                         if (buffer_uptodate(bh) ||
 373                                             buffer_unwritten(bh)) {
 374                                                 if (whence == SEEK_DATA)
 375                                                         found = 1;
 376                                         } else {
 377                                                 if (whence == SEEK_HOLE)
 378                                                         found = 1;
 379                                         }
 380                                         if (found) {
 381                                                 *offset = max_t(loff_t,
 382                                                         startoff, lastoff);
 383                                                 unlock_page(page);
 384                                                 goto out;
 385                                         }
 386                                         lastoff += bh->b_size;
 387                                         bh = bh->b_this_page;
 388                                 } while (bh != head);
 389                         }
 390
 391                         lastoff = page_offset(page) + PAGE_SIZE;
 392                         unlock_page(page);
 393                 }
 394
 395                 /*
 396                  * The no. of pages is less than our desired, that would be a
 397                  * hole in there.
 398                  */
 399                 if (nr_pages < num && whence == SEEK_HOLE) {
 400                         found = 1;
 401                         *offset = lastoff;
 402                         break;
 403                 }
 404
 405                 index = pvec.pages[i - 1]->index + 1;
 406                 pagevec_release(&pvec);
 407         } while (index <= end);
 408
 409 out:
 410         pagevec_release(&pvec);
 411         return found;
 412 }
 413
 414 /*
 415  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 416  */
 417 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 418 {
 419         struct inode *inode = file->f_mapping->host;
 420         struct ext4_map_blocks map;
 421         struct extent_status es;
 422         ext4_lblk_t start, last, end;
 423         loff_t dataoff, isize;
 424         int blkbits;
 425         int ret = 0;
 426
 427         mutex_lock(&inode->i_mutex);
 428
 429         isize = i_size_read(inode);
 430         if (offset >= isize) {
 431                 mutex_unlock(&inode->i_mutex);
 432                 return -ENXIO;
 433         }
 434
 435         blkbits = inode->i_sb->s_blocksize_bits;
 436         start = offset >> blkbits;
 437         last = start;
 438         end = isize >> blkbits;
 439         dataoff = offset;
 440
 441         do {
 442                 map.m_lblk = last;
 443                 map.m_len = end - last + 1;
 444                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 445                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 446                         if (last != start)
 447                                 dataoff = (loff_t)last << blkbits;
 448                         break;
 449                 }
 450
 451                 /*
 452                  * If there is a delay extent at this offset,
 453                  * it will be as a data.
 454                  */
 455                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 456                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 457                         if (last != start)
 458                                 dataoff = (loff_t)last << blkbits;
 459                         break;
 460                 }
 461
 462                 /*
 463                  * If there is a unwritten extent at this offset,
 464                  * it will be as a data or a hole according to page
 465                  * cache that has data or not.
 466                  */
 467                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 468                         int unwritten;
 469                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 470                                                               &map, &dataoff);
 471                         if (unwritten)
 472                                 break;
 473                 }
 474
 475                 last++;
 476                 dataoff = (loff_t)last << blkbits;
 477         } while (last <= end);
 478
 479         mutex_unlock(&inode->i_mutex);
 480
 481         if (dataoff > isize)
 482                 return -ENXIO;
 483
 484         return vfs_setpos(file, dataoff, maxsize);
 485 }
 486
 487 /*
 488  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 489  */
 490 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 491 {
 492         struct inode *inode = file->f_mapping->host;
 493         struct ext4_map_blocks map;
 494         struct extent_status es;
 495         ext4_lblk_t start, last, end;
 496         loff_t holeoff, isize;
 497         int blkbits;
 498         int ret = 0;
 499
 500         mutex_lock(&inode->i_mutex);
 501
 502         isize = i_size_read(inode);
 503         if (offset >= isize) {
 504                 mutex_unlock(&inode->i_mutex);
 505                 return -ENXIO;
 506         }
 507
 508         blkbits = inode->i_sb->s_blocksize_bits;
 509         start = offset >> blkbits;
 510         last = start;
 511         end = isize >> blkbits;
 512         holeoff = offset;
 513
 514         do {
 515                 map.m_lblk = last;
 516                 map.m_len = end - last + 1;
 517                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 518                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 519                         last += ret;
 520                         holeoff = (loff_t)last << blkbits;
 521                         continue;
 522                 }
 523
 524                 /*
 525                  * If there is a delay extent at this offset,
 526                  * we will skip this extent.
 527                  */
 528                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 529                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 530                         last = es.es_lblk + es.es_len;
 531                         holeoff = (loff_t)last << blkbits;
 532                         continue;
 533                 }
 534
 535                 /*
 536                  * If there is a unwritten extent at this offset,
 537                  * it will be as a data or a hole according to page
 538                  * cache that has data or not.
 539                  */
 540                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 541                         int unwritten;
 542                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 543                                                               &map, &holeoff);
 544                         if (!unwritten) {
 545                                 last += ret;
 546                                 holeoff = (loff_t)last << blkbits;
 547                                 continue;
 548                         }
 549                 }
 550
 551                 /* find a hole */
 552                 break;
 553         } while (last <= end);
 554
 555         mutex_unlock(&inode->i_mutex);
 556
 557         if (holeoff > isize)
 558                 holeoff = isize;
 559
 560         return vfs_setpos(file, holeoff, maxsize);
 561 }
 562
 563 /*
 564  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 565  * by calling generic_file_llseek_size() with the appropriate maxbytes
 566  * value for each.
 567  */
 568 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 569 {
 570         struct inode *inode = file->f_mapping->host;
 571         loff_t maxbytes;
 572
 573         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 574                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 575         else
 576                 maxbytes = inode->i_sb->s_maxbytes;
 577
 578         switch (whence) {
 579         case SEEK_SET:
 580         case SEEK_CUR:
 581         case SEEK_END:
 582                 return generic_file_llseek_size(file, offset, whence,
 583                                                 maxbytes, i_size_read(inode));
 584         case SEEK_DATA:
 585                 return ext4_seek_data(file, offset, maxbytes);
 586         case SEEK_HOLE:
 587                 return ext4_seek_hole(file, offset, maxbytes);
 588         }
 589
 590         return -EINVAL;
 591 }
 592
 593 const struct file_operations ext4_file_operations = {
 594         .llseek         = ext4_llseek,
 595         .read           = do_sync_read,
 596         .write          = do_sync_write,
 597         .aio_read       = generic_file_aio_read,
 598         .aio_write      = ext4_file_write,
 599         .unlocked_ioctl = ext4_ioctl,
 600 #ifdef CONFIG_COMPAT
 601         .compat_ioctl   = ext4_compat_ioctl,
 602 #endif
 603         .mmap           = ext4_file_mmap,
 604         .open           = ext4_file_open,
 605         .release        = ext4_release_file,
 606         .fsync          = ext4_sync_file,
 607         .splice_read    = generic_file_splice_read,
 608         .splice_write   = generic_file_splice_write,
 609         .fallocate      = ext4_fallocate,
 610 };
 611
 612 const struct inode_operations ext4_file_inode_operations = {
 613         .setattr        = ext4_setattr,
 614         .getattr        = ext4_getattr,
 615         .setxattr       = generic_setxattr,
 616         .getxattr       = generic_getxattr,
 617         .listxattr      = ext4_listxattr,
 618         .removexattr    = generic_removexattr,
 619         .get_acl        = ext4_get_acl,
 620         .fiemap         = ext4_fiemap,
 621 };
 622