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/quotaops.h>
  27 #include <linux/pagevec.h>
  28 #include <linux/uio.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 static 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, struct iov_iter *from, loff_t pos)
  78 {
  79         struct super_block *sb = inode->i_sb;
  80         int blockmask = sb->s_blocksize - 1;
  81
  82         if (pos >= i_size_read(inode))
  83                 return 0;
  84
  85         if ((pos | iov_iter_alignment(from)) & blockmask)
  86                 return 1;
  87
  88         return 0;
  89 }
  90
  91 static ssize_t
  92 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  93 {
  94         struct file *file = iocb->ki_filp;
  95         struct inode *inode = file_inode(iocb->ki_filp);
  96         struct mutex *aio_mutex = NULL;
  97         struct blk_plug plug;
  98         int o_direct = iocb->ki_flags & IOCB_DIRECT;
  99         int overwrite = 0;
 100         ssize_t ret;
 101
 102         /*
 103          * Unaligned direct AIO must be serialized; see comment above
 104          * In the case of O_APPEND, assume that we must always serialize
 105          */
 106         if (o_direct &&
 107             ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 108             !is_sync_kiocb(iocb) &&
 109             (iocb->ki_flags & IOCB_APPEND ||
 110              ext4_unaligned_aio(inode, from, iocb->ki_pos))) {
 111                 aio_mutex = ext4_aio_mutex(inode);
 112                 mutex_lock(aio_mutex);
 113                 ext4_unwritten_wait(inode);
 114         }
 115
 116         mutex_lock(&inode->i_mutex);
 117         ret = generic_write_checks(iocb, from);
 118         if (ret <= 0)
 119                 goto out;
 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 (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
 129                         ret = -EFBIG;
 130                         goto out;
 131                 }
 132                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 133         }
 134
 135         iocb->private = &overwrite;
 136         if (o_direct) {
 137                 size_t length = iov_iter_count(from);
 138                 loff_t pos = iocb->ki_pos;
 139                 blk_start_plug(&plug);
 140
 141                 /* check whether we do a DIO overwrite or not */
 142                 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
 143                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 144                         struct ext4_map_blocks map;
 145                         unsigned int blkbits = inode->i_blkbits;
 146                         int err, len;
 147
 148                         map.m_lblk = pos >> blkbits;
 149                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 150                                 - map.m_lblk;
 151                         len = map.m_len;
 152
 153                         err = ext4_map_blocks(NULL, inode, &map, 0);
 154                         /*
 155                          * 'err==len' means that all of blocks has
 156                          * been preallocated no matter they are
 157                          * initialized or not.  For excluding
 158                          * unwritten extents, we need to check
 159                          * m_flags.  There are two conditions that
 160                          * indicate for initialized extents.  1) If we
 161                          * hit extent cache, EXT4_MAP_MAPPED flag is
 162                          * returned; 2) If we do a real lookup,
 163                          * non-flags are returned.  So we should check
 164                          * these two conditions.
 165                          */
 166                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 167                                 overwrite = 1;
 168                 }
 169         }
 170
 171         ret = __generic_file_write_iter(iocb, from);
 172         mutex_unlock(&inode->i_mutex);
 173
 174         if (ret > 0) {
 175                 ssize_t err;
 176
 177                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
 178                 if (err < 0)
 179                         ret = err;
 180         }
 181         if (o_direct)
 182                 blk_finish_plug(&plug);
 183
 184         if (aio_mutex)
 185                 mutex_unlock(aio_mutex);
 186         return ret;
 187
 188 out:
 189         mutex_unlock(&inode->i_mutex);
 190         if (aio_mutex)
 191                 mutex_unlock(aio_mutex);
 192         return ret;
 193 }
 194
 195 #ifdef CONFIG_FS_DAX
 196 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 197 {
 198         return dax_fault(vma, vmf, ext4_get_block);
 199                                         /* Is this the right get_block? */
 200 }
 201
 202 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 203 {
 204         return dax_mkwrite(vma, vmf, ext4_get_block);
 205 }
 206
 207 static const struct vm_operations_struct ext4_dax_vm_ops = {
 208         .fault          = ext4_dax_fault,
 209         .page_mkwrite   = ext4_dax_mkwrite,
 210         .pfn_mkwrite    = dax_pfn_mkwrite,
 211 };
 212 #else
 213 #define ext4_dax_vm_ops ext4_file_vm_ops
 214 #endif
 215
 216 static const struct vm_operations_struct ext4_file_vm_ops = {
 217         .fault          = filemap_fault,
 218         .map_pages      = filemap_map_pages,
 219         .page_mkwrite   = ext4_page_mkwrite,
 220 };
 221
 222 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 223 {
 224         file_accessed(file);
 225         if (IS_DAX(file_inode(file))) {
 226                 vma->vm_ops = &ext4_dax_vm_ops;
 227                 vma->vm_flags |= VM_MIXEDMAP;
 228         } else {
 229                 vma->vm_ops = &ext4_file_vm_ops;
 230         }
 231         return 0;
 232 }
 233
 234 static int ext4_file_open(struct inode * inode, struct file * filp)
 235 {
 236         struct super_block *sb = inode->i_sb;
 237         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 238         struct vfsmount *mnt = filp->f_path.mnt;
 239         struct path path;
 240         char buf[64], *cp;
 241
 242         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 243                      !(sb->s_flags & MS_RDONLY))) {
 244                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 245                 /*
 246                  * Sample where the filesystem has been mounted and
 247                  * store it in the superblock for sysadmin convenience
 248                  * when trying to sort through large numbers of block
 249                  * devices or filesystem images.
 250                  */
 251                 memset(buf, 0, sizeof(buf));
 252                 path.mnt = mnt;
 253                 path.dentry = mnt->mnt_root;
 254                 cp = d_path(&path, buf, sizeof(buf));
 255                 if (!IS_ERR(cp)) {
 256                         handle_t *handle;
 257                         int err;
 258
 259                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 260                         if (IS_ERR(handle))
 261                                 return PTR_ERR(handle);
 262                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 263                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 264                         if (err) {
 265                                 ext4_journal_stop(handle);
 266                                 return err;
 267                         }
 268                         strlcpy(sbi->s_es->s_last_mounted, cp,
 269                                 sizeof(sbi->s_es->s_last_mounted));
 270                         ext4_handle_dirty_super(handle, sb);
 271                         ext4_journal_stop(handle);
 272                 }
 273         }
 274         /*
 275          * Set up the jbd2_inode if we are opening the inode for
 276          * writing and the journal is present
 277          */
 278         if (filp->f_mode & FMODE_WRITE) {
 279                 int ret = ext4_inode_attach_jinode(inode);
 280                 if (ret < 0)
 281                         return ret;
 282         }
 283         return dquot_file_open(inode, filp);
 284 }
 285
 286 /*
 287  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
 288  * file rather than ext4_ext_walk_space() because we can introduce
 289  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
 290  * function.  When extent status tree has been fully implemented, it will
 291  * track all extent status for a file and we can directly use it to
 292  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
 293  */
 294
 295 /*
 296  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
 297  * lookup page cache to check whether or not there has some data between
 298  * [startoff, endoff] because, if this range contains an unwritten extent,
 299  * we determine this extent as a data or a hole according to whether the
 300  * page cache has data or not.
 301  */
 302 static int ext4_find_unwritten_pgoff(struct inode *inode,
 303                                      int whence,
 304                                      struct ext4_map_blocks *map,
 305                                      loff_t *offset)
 306 {
 307         struct pagevec pvec;
 308         unsigned int blkbits;
 309         pgoff_t index;
 310         pgoff_t end;
 311         loff_t endoff;
 312         loff_t startoff;
 313         loff_t lastoff;
 314         int found = 0;
 315
 316         blkbits = inode->i_sb->s_blocksize_bits;
 317         startoff = *offset;
 318         lastoff = startoff;
 319         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
 320
 321         index = startoff >> PAGE_CACHE_SHIFT;
 322         end = endoff >> PAGE_CACHE_SHIFT;
 323
 324         pagevec_init(&pvec, 0);
 325         do {
 326                 int i, num;
 327                 unsigned long nr_pages;
 328
 329                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
 330                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
 331                                           (pgoff_t)num);
 332                 if (nr_pages == 0) {
 333                         if (whence == SEEK_DATA)
 334                                 break;
 335
 336                         BUG_ON(whence != SEEK_HOLE);
 337                         /*
 338                          * If this is the first time to go into the loop and
 339                          * offset is not beyond the end offset, it will be a
 340                          * hole at this offset
 341                          */
 342                         if (lastoff == startoff || lastoff < endoff)
 343                                 found = 1;
 344                         break;
 345                 }
 346
 347                 /*
 348                  * If this is the first time to go into the loop and
 349                  * offset is smaller than the first page offset, it will be a
 350                  * hole at this offset.
 351                  */
 352                 if (lastoff == startoff && whence == SEEK_HOLE &&
 353                     lastoff < page_offset(pvec.pages[0])) {
 354                         found = 1;
 355                         break;
 356                 }
 357
 358                 for (i = 0; i < nr_pages; i++) {
 359                         struct page *page = pvec.pages[i];
 360                         struct buffer_head *bh, *head;
 361
 362                         /*
 363                          * If the current offset is not beyond the end of given
 364                          * range, it will be a hole.
 365                          */
 366                         if (lastoff < endoff && whence == SEEK_HOLE &&
 367                             page->index > end) {
 368                                 found = 1;
 369                                 *offset = lastoff;
 370                                 goto out;
 371                         }
 372
 373                         lock_page(page);
 374
 375                         if (unlikely(page->mapping != inode->i_mapping)) {
 376                                 unlock_page(page);
 377                                 continue;
 378                         }
 379
 380                         if (!page_has_buffers(page)) {
 381                                 unlock_page(page);
 382                                 continue;
 383                         }
 384
 385                         if (page_has_buffers(page)) {
 386                                 lastoff = page_offset(page);
 387                                 bh = head = page_buffers(page);
 388                                 do {
 389                                         if (buffer_uptodate(bh) ||
 390                                             buffer_unwritten(bh)) {
 391                                                 if (whence == SEEK_DATA)
 392                                                         found = 1;
 393                                         } else {
 394                                                 if (whence == SEEK_HOLE)
 395                                                         found = 1;
 396                                         }
 397                                         if (found) {
 398                                                 *offset = max_t(loff_t,
 399                                                         startoff, lastoff);
 400                                                 unlock_page(page);
 401                                                 goto out;
 402                                         }
 403                                         lastoff += bh->b_size;
 404                                         bh = bh->b_this_page;
 405                                 } while (bh != head);
 406                         }
 407
 408                         lastoff = page_offset(page) + PAGE_SIZE;
 409                         unlock_page(page);
 410                 }
 411
 412                 /*
 413                  * The no. of pages is less than our desired, that would be a
 414                  * hole in there.
 415                  */
 416                 if (nr_pages < num && whence == SEEK_HOLE) {
 417                         found = 1;
 418                         *offset = lastoff;
 419                         break;
 420                 }
 421
 422                 index = pvec.pages[i - 1]->index + 1;
 423                 pagevec_release(&pvec);
 424         } while (index <= end);
 425
 426 out:
 427         pagevec_release(&pvec);
 428         return found;
 429 }
 430
 431 /*
 432  * ext4_seek_data() retrieves the offset for SEEK_DATA.
 433  */
 434 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 435 {
 436         struct inode *inode = file->f_mapping->host;
 437         struct ext4_map_blocks map;
 438         struct extent_status es;
 439         ext4_lblk_t start, last, end;
 440         loff_t dataoff, isize;
 441         int blkbits;
 442         int ret = 0;
 443
 444         mutex_lock(&inode->i_mutex);
 445
 446         isize = i_size_read(inode);
 447         if (offset >= isize) {
 448                 mutex_unlock(&inode->i_mutex);
 449                 return -ENXIO;
 450         }
 451
 452         blkbits = inode->i_sb->s_blocksize_bits;
 453         start = offset >> blkbits;
 454         last = start;
 455         end = isize >> blkbits;
 456         dataoff = offset;
 457
 458         do {
 459                 map.m_lblk = last;
 460                 map.m_len = end - last + 1;
 461                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 462                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 463                         if (last != start)
 464                                 dataoff = (loff_t)last << blkbits;
 465                         break;
 466                 }
 467
 468                 /*
 469                  * If there is a delay extent at this offset,
 470                  * it will be as a data.
 471                  */
 472                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 473                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 474                         if (last != start)
 475                                 dataoff = (loff_t)last << blkbits;
 476                         break;
 477                 }
 478
 479                 /*
 480                  * If there is a unwritten extent at this offset,
 481                  * it will be as a data or a hole according to page
 482                  * cache that has data or not.
 483                  */
 484                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 485                         int unwritten;
 486                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
 487                                                               &map, &dataoff);
 488                         if (unwritten)
 489                                 break;
 490                 }
 491
 492                 last++;
 493                 dataoff = (loff_t)last << blkbits;
 494         } while (last <= end);
 495
 496         mutex_unlock(&inode->i_mutex);
 497
 498         if (dataoff > isize)
 499                 return -ENXIO;
 500
 501         return vfs_setpos(file, dataoff, maxsize);
 502 }
 503
 504 /*
 505  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
 506  */
 507 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 508 {
 509         struct inode *inode = file->f_mapping->host;
 510         struct ext4_map_blocks map;
 511         struct extent_status es;
 512         ext4_lblk_t start, last, end;
 513         loff_t holeoff, isize;
 514         int blkbits;
 515         int ret = 0;
 516
 517         mutex_lock(&inode->i_mutex);
 518
 519         isize = i_size_read(inode);
 520         if (offset >= isize) {
 521                 mutex_unlock(&inode->i_mutex);
 522                 return -ENXIO;
 523         }
 524
 525         blkbits = inode->i_sb->s_blocksize_bits;
 526         start = offset >> blkbits;
 527         last = start;
 528         end = isize >> blkbits;
 529         holeoff = offset;
 530
 531         do {
 532                 map.m_lblk = last;
 533                 map.m_len = end - last + 1;
 534                 ret = ext4_map_blocks(NULL, inode, &map, 0);
 535                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
 536                         last += ret;
 537                         holeoff = (loff_t)last << blkbits;
 538                         continue;
 539                 }
 540
 541                 /*
 542                  * If there is a delay extent at this offset,
 543                  * we will skip this extent.
 544                  */
 545                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
 546                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
 547                         last = es.es_lblk + es.es_len;
 548                         holeoff = (loff_t)last << blkbits;
 549                         continue;
 550                 }
 551
 552                 /*
 553                  * If there is a unwritten extent at this offset,
 554                  * it will be as a data or a hole according to page
 555                  * cache that has data or not.
 556                  */
 557                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
 558                         int unwritten;
 559                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
 560                                                               &map, &holeoff);
 561                         if (!unwritten) {
 562                                 last += ret;
 563                                 holeoff = (loff_t)last << blkbits;
 564                                 continue;
 565                         }
 566                 }
 567
 568                 /* find a hole */
 569                 break;
 570         } while (last <= end);
 571
 572         mutex_unlock(&inode->i_mutex);
 573
 574         if (holeoff > isize)
 575                 holeoff = isize;
 576
 577         return vfs_setpos(file, holeoff, maxsize);
 578 }
 579
 580 /*
 581  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 582  * by calling generic_file_llseek_size() with the appropriate maxbytes
 583  * value for each.
 584  */
 585 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 586 {
 587         struct inode *inode = file->f_mapping->host;
 588         loff_t maxbytes;
 589
 590         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 591                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 592         else
 593                 maxbytes = inode->i_sb->s_maxbytes;
 594
 595         switch (whence) {
 596         case SEEK_SET:
 597         case SEEK_CUR:
 598         case SEEK_END:
 599                 return generic_file_llseek_size(file, offset, whence,
 600                                                 maxbytes, i_size_read(inode));
 601         case SEEK_DATA:
 602                 return ext4_seek_data(file, offset, maxbytes);
 603         case SEEK_HOLE:
 604                 return ext4_seek_hole(file, offset, maxbytes);
 605         }
 606
 607         return -EINVAL;
 608 }
 609
 610 const struct file_operations ext4_file_operations = {
 611         .llseek         = ext4_llseek,
 612         .read_iter      = generic_file_read_iter,
 613         .write_iter     = ext4_file_write_iter,
 614         .unlocked_ioctl = ext4_ioctl,
 615 #ifdef CONFIG_COMPAT
 616         .compat_ioctl   = ext4_compat_ioctl,
 617 #endif
 618         .mmap           = ext4_file_mmap,
 619         .open           = ext4_file_open,
 620         .release        = ext4_release_file,
 621         .fsync          = ext4_sync_file,
 622         .splice_read    = generic_file_splice_read,
 623         .splice_write   = iter_file_splice_write,
 624         .fallocate      = ext4_fallocate,
 625 };
 626
 627 const struct inode_operations ext4_file_inode_operations = {
 628         .setattr        = ext4_setattr,
 629         .getattr        = ext4_getattr,
 630         .setxattr       = generic_setxattr,
 631         .getxattr       = generic_getxattr,
 632         .listxattr      = ext4_listxattr,
 633         .removexattr    = generic_removexattr,
 634         .get_acl        = ext4_get_acl,
 635         .set_acl        = ext4_set_acl,
 636         .fiemap         = ext4_fiemap,
 637 };
 638