fs/ext4/file.c

   1 /*
   2  *  linux/fs/ext4/file.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   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  *      ([email protected])
  19  */
  20
  21 #include <linux/time.h>
  22 #include <linux/fs.h>
  23 #include <linux/mount.h>
  24 #include <linux/path.h>
  25 #include <linux/quotaops.h>
  26 #include <linux/pagevec.h>
  27 #include <linux/uio.h>
  28 #include "ext4.h"
  29 #include "ext4_jbd2.h"
  30 #include "xattr.h"
  31 #include "acl.h"
  32
  33 /*
  34  * Called when an inode is released. Note that this is different
  35  * from ext4_file_open: open gets called at every open, but release
  36  * gets called only when /all/ the files are closed.
  37  */
  38 static int ext4_release_file(struct inode *inode, struct file *filp)
  39 {
  40         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
  41                 ext4_alloc_da_blocks(inode);
  42                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
  43         }
  44         /* if we are the last writer on the inode, drop the block reservation */
  45         if ((filp->f_mode & FMODE_WRITE) &&
  46                         (atomic_read(&inode->i_writecount) == 1) &&
  47                         !EXT4_I(inode)->i_reserved_data_blocks)
  48         {
  49                 down_write(&EXT4_I(inode)->i_data_sem);
  50                 ext4_discard_preallocations(inode);
  51                 up_write(&EXT4_I(inode)->i_data_sem);
  52         }
  53         if (is_dx(inode) && filp->private_data)
  54                 ext4_htree_free_dir_info(filp->private_data);
  55
  56         return 0;
  57 }
  58
  59 static void ext4_unwritten_wait(struct inode *inode)
  60 {
  61         wait_queue_head_t *wq = ext4_ioend_wq(inode);
  62
  63         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
  64 }
  65
  66 /*
  67  * This tests whether the IO in question is block-aligned or not.
  68  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
  69  * are converted to written only after the IO is complete.  Until they are
  70  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
  71  * it needs to zero out portions of the start and/or end block.  If 2 AIO
  72  * threads are at work on the same unwritten block, they must be synchronized
  73  * or one thread will zero the other's data, causing corruption.
  74  */
  75 static int
  76 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
  77 {
  78         struct super_block *sb = inode->i_sb;
  79         int blockmask = sb->s_blocksize - 1;
  80
  81         if (pos >= i_size_read(inode))
  82                 return 0;
  83
  84         if ((pos | iov_iter_alignment(from)) & blockmask)
  85                 return 1;
  86
  87         return 0;
  88 }
  89
  90 static ssize_t
  91 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  92 {
  93         struct file *file = iocb->ki_filp;
  94         struct inode *inode = file_inode(iocb->ki_filp);
  95         struct mutex *aio_mutex = NULL;
  96         struct blk_plug plug;
  97         int o_direct = iocb->ki_flags & IOCB_DIRECT;
  98         int overwrite = 0;
  99         ssize_t ret;
 100
 101         /*
 102          * Unaligned direct AIO must be serialized; see comment above
 103          * In the case of O_APPEND, assume that we must always serialize
 104          */
 105         if (o_direct &&
 106             ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 107             !is_sync_kiocb(iocb) &&
 108             (iocb->ki_flags & IOCB_APPEND ||
 109              ext4_unaligned_aio(inode, from, iocb->ki_pos))) {
 110                 aio_mutex = ext4_aio_mutex(inode);
 111                 mutex_lock(aio_mutex);
 112                 ext4_unwritten_wait(inode);
 113         }
 114
 115         mutex_lock(&inode->i_mutex);
 116         ret = generic_write_checks(iocb, from);
 117         if (ret <= 0)
 118                 goto out;
 119
 120         /*
 121          * If we have encountered a bitmap-format file, the size limit
 122          * is smaller than s_maxbytes, which is for extent-mapped files.
 123          */
 124         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 125                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 126
 127                 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
 128                         ret = -EFBIG;
 129                         goto out;
 130                 }
 131                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 132         }
 133
 134         iocb->private = &overwrite;
 135         if (o_direct) {
 136                 size_t length = iov_iter_count(from);
 137                 loff_t pos = iocb->ki_pos;
 138                 blk_start_plug(&plug);
 139
 140                 /* check whether we do a DIO overwrite or not */
 141                 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
 142                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 143                         struct ext4_map_blocks map;
 144                         unsigned int blkbits = inode->i_blkbits;
 145                         int err, len;
 146
 147                         map.m_lblk = pos >> blkbits;
 148                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 149                                 - map.m_lblk;
 150                         len = map.m_len;
 151
 152                         err = ext4_map_blocks(NULL, inode, &map, 0);
 153                         /*
 154                          * 'err==len' means that all of blocks has
 155                          * been preallocated no matter they are
 156                          * initialized or not.  For excluding
 157                          * unwritten extents, we need to check
 158                          * m_flags.  There are two conditions that
 159                          * indicate for initialized extents.  1) If we
 160                          * hit extent cache, EXT4_MAP_MAPPED flag is
 161                          * returned; 2) If we do a real lookup,
 162                          * non-flags are returned.  So we should check
 163                          * these two conditions.
 164                          */
 165                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 166                                 overwrite = 1;
 167                 }
 168         }
 169
 170         ret = __generic_file_write_iter(iocb, from);
 171         mutex_unlock(&inode->i_mutex);
 172
 173         if (ret > 0) {
 174                 ssize_t err;
 175
 176                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
 177                 if (err < 0)
 178                         ret = err;
 179         }
 180         if (o_direct)
 181                 blk_finish_plug(&plug);
 182
 183         if (aio_mutex)
 184                 mutex_unlock(aio_mutex);
 185         return ret;
 186
 187 out:
 188         mutex_unlock(&inode->i_mutex);
 189         if (aio_mutex)
 190                 mutex_unlock(aio_mutex);
 191         return ret;
 192 }
 193
 194 #ifdef CONFIG_FS_DAX
 195 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 196 {
 197         return dax_fault(vma, vmf, ext4_get_block);
 198                                         /* Is this the right get_block? */
 199 }
 200
 201 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 202 {
 203         return dax_mkwrite(vma, vmf, ext4_get_block);
 204 }
 205
 206 static const struct vm_operations_struct ext4_dax_vm_ops = {
 207         .fault          = ext4_dax_fault,
 208         .page_mkwrite   = ext4_dax_mkwrite,
 209         .pfn_mkwrite    = dax_pfn_mkwrite,
 210 };
 211 #else
 212 #define ext4_dax_vm_ops ext4_file_vm_ops
 213 #endif
 214
 215 static const struct vm_operations_struct ext4_file_vm_ops = {
 216         .fault          = filemap_fault,
 217         .map_pages      = filemap_map_pages,
 218         .page_mkwrite   = ext4_page_mkwrite,
 219 };
 220
 221 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 222 {
 223         struct inode *inode = file->f_mapping->host;
 224
 225         if (ext4_encrypted_inode(inode)) {
 226                 int err = ext4_get_encryption_info(inode);
 227                 if (err)
 228                         return 0;
 229                 if (ext4_encryption_info(inode) == NULL)
 230                         return -ENOKEY;
 231         }
 232         file_accessed(file);
 233         if (IS_DAX(file_inode(file))) {
 234                 vma->vm_ops = &ext4_dax_vm_ops;
 235                 vma->vm_flags |= VM_MIXEDMAP;
 236         } else {
 237                 vma->vm_ops = &ext4_file_vm_ops;
 238         }
 239         return 0;
 240 }
 241
 242 static int ext4_file_open(struct inode * inode, struct file * filp)
 243 {
 244         struct super_block *sb = inode->i_sb;
 245         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 246         struct vfsmount *mnt = filp->f_path.mnt;
 247         struct path path;
 248         char buf[64], *cp;
 249         int ret;
 250
 251         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 252                      !(sb->s_flags & MS_RDONLY))) {
 253                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 254                 /*
 255                  * Sample where the filesystem has been mounted and
 256                  * store it in the superblock for sysadmin convenience
 257                  * when trying to sort through large numbers of block
 258                  * devices or filesystem images.
 259                  */
 260                 memset(buf, 0, sizeof(buf));
 261                 path.mnt = mnt;
 262                 path.dentry = mnt->mnt_root;
 263                 cp = d_path(&path, buf, sizeof(buf));
 264                 if (!IS_ERR(cp)) {
 265                         handle_t *handle;
 266                         int err;
 267
 268                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 269                         if (IS_ERR(handle))
 270                                 return PTR_ERR(handle);
 271                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 272                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 273                         if (err) {
 274                                 ext4_journal_stop(handle);
 275                                 return err;
 276                         }
 277                         strlcpy(sbi->s_es->s_last_mounted, cp,
 278                                 sizeof(sbi->s_es->s_last_mounted));
 279                         ext4_handle_dirty_super(handle, sb);
 280                         ext4_journal_stop(handle);
 281                 }
 282         }
 283         if (ext4_encrypted_inode(inode)) {
 284                 ret = ext4_get_encryption_info(inode);
 285                 if (ret)
 286                         return -EACCES;
 287                 if (ext4_encryption_info(inode) == NULL)
 288                         return -ENOKEY;
 289         }
 290         /*
 291          * Set up the jbd2_inode if we are opening the inode for
 292          * writing and the journal is present
 293          */
 294         if (filp->f_mode & FMODE_WRITE) {
 295                 ret = ext4_inode_attach_jinode(inode);
 296                 if (ret < 0)
 297                         return ret;
 298         }
 299         return dquot_file_open(inode, filp);
 300 }
 301
 302 /*
 303  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 304  * file rather than ext4_ext_walk_space() because we can introduce
 305  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 306  * function.  When extent status tree has been fully implemented, it will
 307  * track all extent status for a file and we can directly use it to
 308  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 309  */
 310
 311 /*
 312  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 313  * lookup page cache to check whether or not there has some data between
 314  * [startoff, endoff] because, if this range contains an unwritten extent,
 315  * we determine this extent as a data or a hole according to whether the
 316  * page cache has data or not.
 317  */
 318 static int ext4_find_unwritten_pgoff(struct inode *inode,
 319                                      int whence,
 320                                      struct ext4_map_blocks *map,
 321                                      loff_t *offset)
 322 {
 323         struct pagevec pvec;
 324         unsigned int blkbits;
 325         pgoff_t index;
 326         pgoff_t end;
 327         loff_t endoff;
 328         loff_t startoff;
 329         loff_t lastoff;
 330         int found = 0;
 331
 332         blkbits = inode->i_sb->s_blocksize_bits;
 333         startoff = *offset;
 334         lastoff = startoff;
 335         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 336
 337         index = startoff >> PAGE_CACHE_SHIFT;
 338         end = endoff >> PAGE_CACHE_SHIFT;
 339
 340         pagevec_init(&pvec, 0);
 341         do {
 342                 int i, num;
 343                 unsigned long nr_pages;
 344
 345                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 346                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 347                                           (pgoff_t)num);
 348                 if (nr_pages == 0) {
 349                         if (whence == SEEK_DATA)
 350                                 break;
 351
 352                         BUG_ON(whence != SEEK_HOLE);
 353                         /*
 354                          * If this is the first time to go into the loop and
 355                          * offset is not beyond the end offset, it will be a
 356                          * hole at this offset
 357                          */
 358                         if (lastoff == startoff || lastoff < endoff)
 359                                 found = 1;
 360                         break;
 361                 }
 362
 363                 /*
 364                  * If this is the first time to go into the loop and
 365                  * offset is smaller than the first page offset, it will be a
 366                  * hole at this offset.
 367                  */
 368                 if (lastoff == startoff && whence == SEEK_HOLE &&
 369                     lastoff < page_offset(pvec.pages[0])) {
 370                         found = 1;
 371                         break;
 372                 }
 373
 374                 for (i = 0; i < nr_pages; i++) {
 375                         struct page *page = pvec.pages[i];
 376                         struct buffer_head *bh, *head;
 377
 378                         /*
 379                          * If the current offset is not beyond the end of given
 380                          * range, it will be a hole.
 381                          */
 382                         if (lastoff < endoff && whence == SEEK_HOLE &&
 383                             page->index > end) {
 384                                 found = 1;
 385                                 *offset = lastoff;
 386                                 goto out;
 387                         }
 388
 389                         lock_page(page);
 390
 391                         if (unlikely(page->mapping != inode->i_mapping)) {
 392                                 unlock_page(page);
 393                                 continue;
 394                         }
 395
 396                         if (!page_has_buffers(page)) {
 397                                 unlock_page(page);
 398                                 continue;
 399                         }
 400
 401                         if (page_has_buffers(page)) {
 402                                 lastoff = page_offset(page);
 403                                 bh = head = page_buffers(page);
 404                                 do {
 405                                         if (buffer_uptodate(bh) ||
 406                                             buffer_unwritten(bh)) {
 407                                                 if (whence == SEEK_DATA)
 408                                                         found = 1;
 409                                         } else {
 410                                                 if (whence == SEEK_HOLE)
 411                                                         found = 1;
 412                                         }
 413                                         if (found) {
 414                                                 *offset = max_t(loff_t,
 415                                                         startoff, lastoff);
 416                                                 unlock_page(page);
 417                                                 goto out;
 418                                         }
 419                                         lastoff += bh->b_size;
 420                                         bh = bh->b_this_page;
 421                                 } while (bh != head);
 422                         }
 423
 424                         lastoff = page_offset(page) + PAGE_SIZE;
 425                         unlock_page(page);
 426                 }
 427
 428                 /*
 429                  * The no. of pages is less than our desired, that would be a
 430                  * hole in there.
 431                  */
 432                 if (nr_pages < num && whence == SEEK_HOLE) {
 433                         found = 1;
 434                         *offset = lastoff;
 435                         break;
 436                 }
 437
 438                 index = pvec.pages[i - 1]->index + 1;
 439                 pagevec_release(&pvec);
 440         } while (index <= end);
 441
 442 out:
 443         pagevec_release(&pvec);
 444         return found;
 445 }
 446
 447 /*
 448  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 449  */
 450 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 451 {
 452         struct inode *inode = file->f_mapping->host;
 453         struct ext4_map_blocks map;
 454         struct extent_status es;
 455         ext4_lblk_t start, last, end;
 456         loff_t dataoff, isize;
 457         int blkbits;
 458         int ret = 0;
 459
 460         mutex_lock(&inode->i_mutex);
 461
 462         isize = i_size_read(inode);
 463         if (offset >= isize) {
 464                 mutex_unlock(&inode->i_mutex);
 465                 return -ENXIO;
 466         }
 467
 468         blkbits = inode->i_sb->s_blocksize_bits;
 469         start = offset >> blkbits;
 470         last = start;
 471         end = isize >> blkbits;
 472         dataoff = offset;
 473
 474         do {
 475                 map.m_lblk = last;
 476                 map.m_len = end - last + 1;
 477                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 478                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 479                         if (last != start)
 480                                 dataoff = (loff_t)last << blkbits;
 481                         break;
 482                 }
 483
 484                 /*
 485                  * If there is a delay extent at this offset,
 486                  * it will be as a data.
 487                  */
 488                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 489                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 490                         if (last != start)
 491                                 dataoff = (loff_t)last << blkbits;
 492                         break;
 493                 }
 494
 495                 /*
 496                  * If there is a unwritten extent at this offset,
 497                  * it will be as a data or a hole according to page
 498                  * cache that has data or not.
 499                  */
 500                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 501                         int unwritten;
 502                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 503                                                               &map, &dataoff);
 504                         if (unwritten)
 505                                 break;
 506                 }
 507
 508                 last++;
 509                 dataoff = (loff_t)last << blkbits;
 510         } while (last <= end);
 511
 512         mutex_unlock(&inode->i_mutex);
 513
 514         if (dataoff > isize)
 515                 return -ENXIO;
 516
 517         return vfs_setpos(file, dataoff, maxsize);
 518 }
 519
 520 /*
 521  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 522  */
 523 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 524 {
 525         struct inode *inode = file->f_mapping->host;
 526         struct ext4_map_blocks map;
 527         struct extent_status es;
 528         ext4_lblk_t start, last, end;
 529         loff_t holeoff, isize;
 530         int blkbits;
 531         int ret = 0;
 532
 533         mutex_lock(&inode->i_mutex);
 534
 535         isize = i_size_read(inode);
 536         if (offset >= isize) {
 537                 mutex_unlock(&inode->i_mutex);
 538                 return -ENXIO;
 539         }
 540
 541         blkbits = inode->i_sb->s_blocksize_bits;
 542         start = offset >> blkbits;
 543         last = start;
 544         end = isize >> blkbits;
 545         holeoff = offset;
 546
 547         do {
 548                 map.m_lblk = last;
 549                 map.m_len = end - last + 1;
 550                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 551                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 552                         last += ret;
 553                         holeoff = (loff_t)last << blkbits;
 554                         continue;
 555                 }
 556
 557                 /*
 558                  * If there is a delay extent at this offset,
 559                  * we will skip this extent.
 560                  */
 561                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 562                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 563                         last = es.es_lblk + es.es_len;
 564                         holeoff = (loff_t)last << blkbits;
 565                         continue;
 566                 }
 567
 568                 /*
 569                  * If there is a unwritten extent at this offset,
 570                  * it will be as a data or a hole according to page
 571                  * cache that has data or not.
 572                  */
 573                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 574                         int unwritten;
 575                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 576                                                               &map, &holeoff);
 577                         if (!unwritten) {
 578                                 last += ret;
 579                                 holeoff = (loff_t)last << blkbits;
 580                                 continue;
 581                         }
 582                 }
 583
 584                 /* find a hole */
 585                 break;
 586         } while (last <= end);
 587
 588         mutex_unlock(&inode->i_mutex);
 589
 590         if (holeoff > isize)
 591                 holeoff = isize;
 592
 593         return vfs_setpos(file, holeoff, maxsize);
 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_iter      = generic_file_read_iter,
 629         .write_iter     = ext4_file_write_iter,
 630         .unlocked_ioctl = ext4_ioctl,
 631 #ifdef CONFIG_COMPAT
 632         .compat_ioctl   = ext4_compat_ioctl,
 633 #endif
 634         .mmap           = ext4_file_mmap,
 635         .open           = ext4_file_open,
 636         .release        = ext4_release_file,
 637         .fsync          = ext4_sync_file,
 638         .splice_read    = generic_file_splice_read,
 639         .splice_write   = iter_file_splice_write,
 640         .fallocate      = ext4_fallocate,
 641 };
 642
 643 const struct inode_operations ext4_file_inode_operations = {
 644         .setattr        = ext4_setattr,
 645         .getattr        = ext4_getattr,
 646         .setxattr       = generic_setxattr,
 647         .getxattr       = generic_getxattr,
 648         .listxattr      = ext4_listxattr,
 649         .removexattr    = generic_removexattr,
 650         .get_acl        = ext4_get_acl,
 651         .set_acl        = ext4_set_acl,
 652         .fiemap         = ext4_fiemap,
 653 };
 654