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/mount.h>
  24 #include <linux/path.h>
  25 #include <linux/aio.h>
  26 #include <linux/quotaops.h>
  27 #include <linux/pagevec.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 = io_is_direct(file);
  98         int overwrite = 0;
  99         size_t length = iov_iter_count(from);
 100         ssize_t ret;
 101         loff_t pos = iocb->ki_pos;
 102
 103         /*
 104          * Unaligned direct AIO must be serialized; see comment above
 105          * In the case of O_APPEND, assume that we must always serialize
 106          */
 107         if (o_direct &&
 108             ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 109             !is_sync_kiocb(iocb) &&
 110             (file->f_flags & O_APPEND ||
 111              ext4_unaligned_aio(inode, from, pos))) {
 112                 aio_mutex = ext4_aio_mutex(inode);
 113                 mutex_lock(aio_mutex);
 114                 ext4_unwritten_wait(inode);
 115         }
 116
 117         mutex_lock(&inode->i_mutex);
 118         if (file->f_flags & O_APPEND)
 119                 iocb->ki_pos = pos = i_size_read(inode);
 120
 121         /*
 122          * If we have encountered a bitmap-format file, the size limit
 123          * is smaller than s_maxbytes, which is for extent-mapped files.
 124          */
 125         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 126                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 127
 128                 if ((pos > sbi->s_bitmap_maxbytes) ||
 129                     (pos == sbi->s_bitmap_maxbytes && length > 0)) {
 130                         mutex_unlock(&inode->i_mutex);
 131                         ret = -EFBIG;
 132                         goto errout;
 133                 }
 134
 135                 if (pos + length > sbi->s_bitmap_maxbytes)
 136                         iov_iter_truncate(from, sbi->s_bitmap_maxbytes - pos);
 137         }
 138
 139         iocb->private = &overwrite;
 140         if (o_direct) {
 141                 blk_start_plug(&plug);
 142
 143
 144                 /* check whether we do a DIO overwrite or not */
 145                 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
 146                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 147                         struct ext4_map_blocks map;
 148                         unsigned int blkbits = inode->i_blkbits;
 149                         int err, len;
 150
 151                         map.m_lblk = pos >> blkbits;
 152                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 153                                 - map.m_lblk;
 154                         len = map.m_len;
 155
 156                         err = ext4_map_blocks(NULL, inode, &map, 0);
 157                         /*
 158                          * 'err==len' means that all of blocks has
 159                          * been preallocated no matter they are
 160                          * initialized or not.  For excluding
 161                          * unwritten extents, we need to check
 162                          * m_flags.  There are two conditions that
 163                          * indicate for initialized extents.  1) If we
 164                          * hit extent cache, EXT4_MAP_MAPPED flag is
 165                          * returned; 2) If we do a real lookup,
 166                          * non-flags are returned.  So we should check
 167                          * these two conditions.
 168                          */
 169                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 170                                 overwrite = 1;
 171                 }
 172         }
 173
 174         ret = __generic_file_write_iter(iocb, from);
 175         mutex_unlock(&inode->i_mutex);
 176
 177         if (ret > 0) {
 178                 ssize_t err;
 179
 180                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
 181                 if (err < 0)
 182                         ret = err;
 183         }
 184         if (o_direct)
 185                 blk_finish_plug(&plug);
 186
 187 errout:
 188         if (aio_mutex)
 189                 mutex_unlock(aio_mutex);
 190         return ret;
 191 }
 192
 193 #ifdef CONFIG_FS_DAX
 194 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 195 {
 196         return dax_fault(vma, vmf, ext4_get_block);
 197                                         /* Is this the right get_block? */
 198 }
 199
 200 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 201 {
 202         return dax_mkwrite(vma, vmf, ext4_get_block);
 203 }
 204
 205 static const struct vm_operations_struct ext4_dax_vm_ops = {
 206         .fault          = ext4_dax_fault,
 207         .page_mkwrite   = ext4_dax_mkwrite,
 208 };
 209 #else
 210 #define ext4_dax_vm_ops ext4_file_vm_ops
 211 #endif
 212
 213 static const struct vm_operations_struct ext4_file_vm_ops = {
 214         .fault          = filemap_fault,
 215         .map_pages      = filemap_map_pages,
 216         .page_mkwrite   = ext4_page_mkwrite,
 217 };
 218
 219 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 220 {
 221         file_accessed(file);
 222         if (IS_DAX(file_inode(file))) {
 223                 vma->vm_ops = &ext4_dax_vm_ops;
 224                 vma->vm_flags |= VM_MIXEDMAP;
 225         } else {
 226                 vma->vm_ops = &ext4_file_vm_ops;
 227         }
 228         return 0;
 229 }
 230
 231 static int ext4_file_open(struct inode * inode, struct file * filp)
 232 {
 233         struct super_block *sb = inode->i_sb;
 234         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 235         struct vfsmount *mnt = filp->f_path.mnt;
 236         struct path path;
 237         char buf[64], *cp;
 238
 239         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 240                      !(sb->s_flags & MS_RDONLY))) {
 241                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 242                 /*
 243                  * Sample where the filesystem has been mounted and
 244                  * store it in the superblock for sysadmin convenience
 245                  * when trying to sort through large numbers of block
 246                  * devices or filesystem images.
 247                  */
 248                 memset(buf, 0, sizeof(buf));
 249                 path.mnt = mnt;
 250                 path.dentry = mnt->mnt_root;
 251                 cp = d_path(&path, buf, sizeof(buf));
 252                 if (!IS_ERR(cp)) {
 253                         handle_t *handle;
 254                         int err;
 255
 256                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 257                         if (IS_ERR(handle))
 258                                 return PTR_ERR(handle);
 259                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 260                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 261                         if (err) {
 262                                 ext4_journal_stop(handle);
 263                                 return err;
 264                         }
 265                         strlcpy(sbi->s_es->s_last_mounted, cp,
 266                                 sizeof(sbi->s_es->s_last_mounted));
 267                         ext4_handle_dirty_super(handle, sb);
 268                         ext4_journal_stop(handle);
 269                 }
 270         }
 271         /*
 272          * Set up the jbd2_inode if we are opening the inode for
 273          * writing and the journal is present
 274          */
 275         if (filp->f_mode & FMODE_WRITE) {
 276                 int ret = ext4_inode_attach_jinode(inode);
 277                 if (ret < 0)
 278                         return ret;
 279         }
 280         return dquot_file_open(inode, filp);
 281 }
 282
 283 /*
 284  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 285  * file rather than ext4_ext_walk_space() because we can introduce
 286  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 287  * function.  When extent status tree has been fully implemented, it will
 288  * track all extent status for a file and we can directly use it to
 289  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 290  */
 291
 292 /*
 293  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 294  * lookup page cache to check whether or not there has some data between
 295  * [startoff, endoff] because, if this range contains an unwritten extent,
 296  * we determine this extent as a data or a hole according to whether the
 297  * page cache has data or not.
 298  */
 299 static int ext4_find_unwritten_pgoff(struct inode *inode,
 300                                      int whence,
 301                                      struct ext4_map_blocks *map,
 302                                      loff_t *offset)
 303 {
 304         struct pagevec pvec;
 305         unsigned int blkbits;
 306         pgoff_t index;
 307         pgoff_t end;
 308         loff_t endoff;
 309         loff_t startoff;
 310         loff_t lastoff;
 311         int found = 0;
 312
 313         blkbits = inode->i_sb->s_blocksize_bits;
 314         startoff = *offset;
 315         lastoff = startoff;
 316         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 317
 318         index = startoff >> PAGE_CACHE_SHIFT;
 319         end = endoff >> PAGE_CACHE_SHIFT;
 320
 321         pagevec_init(&pvec, 0);
 322         do {
 323                 int i, num;
 324                 unsigned long nr_pages;
 325
 326                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 327                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 328                                           (pgoff_t)num);
 329                 if (nr_pages == 0) {
 330                         if (whence == SEEK_DATA)
 331                                 break;
 332
 333                         BUG_ON(whence != SEEK_HOLE);
 334                         /*
 335                          * If this is the first time to go into the loop and
 336                          * offset is not beyond the end offset, it will be a
 337                          * hole at this offset
 338                          */
 339                         if (lastoff == startoff || lastoff < endoff)
 340                                 found = 1;
 341                         break;
 342                 }
 343
 344                 /*
 345                  * If this is the first time to go into the loop and
 346                  * offset is smaller than the first page offset, it will be a
 347                  * hole at this offset.
 348                  */
 349                 if (lastoff == startoff && whence == SEEK_HOLE &&
 350                     lastoff < page_offset(pvec.pages[0])) {
 351                         found = 1;
 352                         break;
 353                 }
 354
 355                 for (i = 0; i < nr_pages; i++) {
 356                         struct page *page = pvec.pages[i];
 357                         struct buffer_head *bh, *head;
 358
 359                         /*
 360                          * If the current offset is not beyond the end of given
 361                          * range, it will be a hole.
 362                          */
 363                         if (lastoff < endoff && whence == SEEK_HOLE &&
 364                             page->index > end) {
 365                                 found = 1;
 366                                 *offset = lastoff;
 367                                 goto out;
 368                         }
 369
 370                         lock_page(page);
 371
 372                         if (unlikely(page->mapping != inode->i_mapping)) {
 373                                 unlock_page(page);
 374                                 continue;
 375                         }
 376
 377                         if (!page_has_buffers(page)) {
 378                                 unlock_page(page);
 379                                 continue;
 380                         }
 381
 382                         if (page_has_buffers(page)) {
 383                                 lastoff = page_offset(page);
 384                                 bh = head = page_buffers(page);
 385                                 do {
 386                                         if (buffer_uptodate(bh) ||
 387                                             buffer_unwritten(bh)) {
 388                                                 if (whence == SEEK_DATA)
 389                                                         found = 1;
 390                                         } else {
 391                                                 if (whence == SEEK_HOLE)
 392                                                         found = 1;
 393                                         }
 394                                         if (found) {
 395                                                 *offset = max_t(loff_t,
 396                                                         startoff, lastoff);
 397                                                 unlock_page(page);
 398                                                 goto out;
 399                                         }
 400                                         lastoff += bh->b_size;
 401                                         bh = bh->b_this_page;
 402                                 } while (bh != head);
 403                         }
 404
 405                         lastoff = page_offset(page) + PAGE_SIZE;
 406                         unlock_page(page);
 407                 }
 408
 409                 /*
 410                  * The no. of pages is less than our desired, that would be a
 411                  * hole in there.
 412                  */
 413                 if (nr_pages < num && whence == SEEK_HOLE) {
 414                         found = 1;
 415                         *offset = lastoff;
 416                         break;
 417                 }
 418
 419                 index = pvec.pages[i - 1]->index + 1;
 420                 pagevec_release(&pvec);
 421         } while (index <= end);
 422
 423 out:
 424         pagevec_release(&pvec);
 425         return found;
 426 }
 427
 428 /*
 429  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 430  */
 431 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 432 {
 433         struct inode *inode = file->f_mapping->host;
 434         struct ext4_map_blocks map;
 435         struct extent_status es;
 436         ext4_lblk_t start, last, end;
 437         loff_t dataoff, isize;
 438         int blkbits;
 439         int ret = 0;
 440
 441         mutex_lock(&inode->i_mutex);
 442
 443         isize = i_size_read(inode);
 444         if (offset >= isize) {
 445                 mutex_unlock(&inode->i_mutex);
 446                 return -ENXIO;
 447         }
 448
 449         blkbits = inode->i_sb->s_blocksize_bits;
 450         start = offset >> blkbits;
 451         last = start;
 452         end = isize >> blkbits;
 453         dataoff = offset;
 454
 455         do {
 456                 map.m_lblk = last;
 457                 map.m_len = end - last + 1;
 458                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 459                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 460                         if (last != start)
 461                                 dataoff = (loff_t)last << blkbits;
 462                         break;
 463                 }
 464
 465                 /*
 466                  * If there is a delay extent at this offset,
 467                  * it will be as a data.
 468                  */
 469                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 470                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 471                         if (last != start)
 472                                 dataoff = (loff_t)last << blkbits;
 473                         break;
 474                 }
 475
 476                 /*
 477                  * If there is a unwritten extent at this offset,
 478                  * it will be as a data or a hole according to page
 479                  * cache that has data or not.
 480                  */
 481                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 482                         int unwritten;
 483                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 484                                                               &map, &dataoff);
 485                         if (unwritten)
 486                                 break;
 487                 }
 488
 489                 last++;
 490                 dataoff = (loff_t)last << blkbits;
 491         } while (last <= end);
 492
 493         mutex_unlock(&inode->i_mutex);
 494
 495         if (dataoff > isize)
 496                 return -ENXIO;
 497
 498         return vfs_setpos(file, dataoff, maxsize);
 499 }
 500
 501 /*
 502  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 503  */
 504 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 505 {
 506         struct inode *inode = file->f_mapping->host;
 507         struct ext4_map_blocks map;
 508         struct extent_status es;
 509         ext4_lblk_t start, last, end;
 510         loff_t holeoff, isize;
 511         int blkbits;
 512         int ret = 0;
 513
 514         mutex_lock(&inode->i_mutex);
 515
 516         isize = i_size_read(inode);
 517         if (offset >= isize) {
 518                 mutex_unlock(&inode->i_mutex);
 519                 return -ENXIO;
 520         }
 521
 522         blkbits = inode->i_sb->s_blocksize_bits;
 523         start = offset >> blkbits;
 524         last = start;
 525         end = isize >> blkbits;
 526         holeoff = offset;
 527
 528         do {
 529                 map.m_lblk = last;
 530                 map.m_len = end - last + 1;
 531                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 532                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 533                         last += ret;
 534                         holeoff = (loff_t)last << blkbits;
 535                         continue;
 536                 }
 537
 538                 /*
 539                  * If there is a delay extent at this offset,
 540                  * we will skip this extent.
 541                  */
 542                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 543                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 544                         last = es.es_lblk + es.es_len;
 545                         holeoff = (loff_t)last << blkbits;
 546                         continue;
 547                 }
 548
 549                 /*
 550                  * If there is a unwritten extent at this offset,
 551                  * it will be as a data or a hole according to page
 552                  * cache that has data or not.
 553                  */
 554                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 555                         int unwritten;
 556                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 557                                                               &map, &holeoff);
 558                         if (!unwritten) {
 559                                 last += ret;
 560                                 holeoff = (loff_t)last << blkbits;
 561                                 continue;
 562                         }
 563                 }
 564
 565                 /* find a hole */
 566                 break;
 567         } while (last <= end);
 568
 569         mutex_unlock(&inode->i_mutex);
 570
 571         if (holeoff > isize)
 572                 holeoff = isize;
 573
 574         return vfs_setpos(file, holeoff, maxsize);
 575 }
 576
 577 /*
 578  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 579  * by calling generic_file_llseek_size() with the appropriate maxbytes
 580  * value for each.
 581  */
 582 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 583 {
 584         struct inode *inode = file->f_mapping->host;
 585         loff_t maxbytes;
 586
 587         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 588                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 589         else
 590                 maxbytes = inode->i_sb->s_maxbytes;
 591
 592         switch (whence) {
 593         case SEEK_SET:
 594         case SEEK_CUR:
 595         case SEEK_END:
 596                 return generic_file_llseek_size(file, offset, whence,
 597                                                 maxbytes, i_size_read(inode));
 598         case SEEK_DATA:
 599                 return ext4_seek_data(file, offset, maxbytes);
 600         case SEEK_HOLE:
 601                 return ext4_seek_hole(file, offset, maxbytes);
 602         }
 603
 604         return -EINVAL;
 605 }
 606
 607 const struct file_operations ext4_file_operations = {
 608         .llseek         = ext4_llseek,
 609         .read           = new_sync_read,
 610         .write          = new_sync_write,
 611         .read_iter      = generic_file_read_iter,
 612         .write_iter     = ext4_file_write_iter,
 613         .unlocked_ioctl = ext4_ioctl,
 614 #ifdef CONFIG_COMPAT
 615         .compat_ioctl   = ext4_compat_ioctl,
 616 #endif
 617         .mmap           = ext4_file_mmap,
 618         .open           = ext4_file_open,
 619         .release        = ext4_release_file,
 620         .fsync          = ext4_sync_file,
 621         .splice_read    = generic_file_splice_read,
 622         .splice_write   = iter_file_splice_write,
 623         .fallocate      = ext4_fallocate,
 624 };
 625
 626 #ifdef CONFIG_FS_DAX
 627 const struct file_operations ext4_dax_file_operations = {
 628         .llseek         = ext4_llseek,
 629         .read           = new_sync_read,
 630         .write          = new_sync_write,
 631         .read_iter      = generic_file_read_iter,
 632         .write_iter     = ext4_file_write_iter,
 633         .unlocked_ioctl = ext4_ioctl,
 634 #ifdef CONFIG_COMPAT
 635         .compat_ioctl   = ext4_compat_ioctl,
 636 #endif
 637         .mmap           = ext4_file_mmap,
 638         .open           = ext4_file_open,
 639         .release        = ext4_release_file,
 640         .fsync          = ext4_sync_file,
 641         /* Splice not yet supported with DAX */
 642         .fallocate      = ext4_fallocate,
 643 };
 644 #endif
 645
 646 const struct inode_operations ext4_file_inode_operations = {
 647         .setattr        = ext4_setattr,
 648         .getattr        = ext4_getattr,
 649         .setxattr       = generic_setxattr,
 650         .getxattr       = generic_getxattr,
 651         .listxattr      = ext4_listxattr,
 652         .removexattr    = generic_removexattr,
 653         .get_acl        = ext4_get_acl,
 654         .set_acl        = ext4_set_acl,
 655         .fiemap         = ext4_fiemap,
 656 };
 657