fs/btrfs/ordered-data.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/gfp.h>
  20 #include <linux/slab.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/writeback.h>
  23 #include <linux/pagevec.h>
  24 #include "ctree.h"
  25 #include "transaction.h"
  26 #include "btrfs_inode.h"
  27 #include "extent_io.h"
  28
  29 static u64 entry_end(struct btrfs_ordered_extent *entry)
  30 {
  31         if (entry->file_offset + entry->len < entry->file_offset)
  32                 return (u64)-1;
  33         return entry->file_offset + entry->len;
  34 }
  35
  36 /* returns NULL if the insertion worked, or it returns the node it did find
  37  * in the tree
  38  */
  39 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
  40                                    struct rb_node *node)
  41 {
  42         struct rb_node ** p = &root->rb_node;
  43         struct rb_node * parent = NULL;
  44         struct btrfs_ordered_extent *entry;
  45
  46         while(*p) {
  47                 parent = *p;
  48                 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
  49
  50                 if (file_offset < entry->file_offset)
  51                         p = &(*p)->rb_left;
  52                 else if (file_offset >= entry_end(entry))
  53                         p = &(*p)->rb_right;
  54                 else
  55                         return parent;
  56         }
  57
  58         rb_link_node(node, parent, p);
  59         rb_insert_color(node, root);
  60         return NULL;
  61 }
  62
  63 /*
  64  * look for a given offset in the tree, and if it can't be found return the
  65  * first lesser offset
  66  */
  67 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
  68                                      struct rb_node **prev_ret)
  69 {
  70         struct rb_node * n = root->rb_node;
  71         struct rb_node *prev = NULL;
  72         struct rb_node *test;
  73         struct btrfs_ordered_extent *entry;
  74         struct btrfs_ordered_extent *prev_entry = NULL;
  75
  76         while(n) {
  77                 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
  78                 prev = n;
  79                 prev_entry = entry;
  80
  81                 if (file_offset < entry->file_offset)
  82                         n = n->rb_left;
  83                 else if (file_offset >= entry_end(entry))
  84                         n = n->rb_right;
  85                 else
  86                         return n;
  87         }
  88         if (!prev_ret)
  89                 return NULL;
  90
  91         while(prev && file_offset >= entry_end(prev_entry)) {
  92                 test = rb_next(prev);
  93                 if (!test)
  94                         break;
  95                 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
  96                                       rb_node);
  97                 if (file_offset < entry_end(prev_entry))
  98                         break;
  99
 100                 prev = test;
 101         }
 102         if (prev)
 103                 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
 104                                       rb_node);
 105         while(prev && file_offset < entry_end(prev_entry)) {
 106                 test = rb_prev(prev);
 107                 if (!test)
 108                         break;
 109                 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
 110                                       rb_node);
 111                 prev = test;
 112         }
 113         *prev_ret = prev;
 114         return NULL;
 115 }
 116
 117 /*
 118  * helper to check if a given offset is inside a given entry
 119  */
 120 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
 121 {
 122         if (file_offset < entry->file_offset ||
 123             entry->file_offset + entry->len <= file_offset)
 124                 return 0;
 125         return 1;
 126 }
 127
 128 /*
 129  * look find the first ordered struct that has this offset, otherwise
 130  * the first one less than this offset
 131  */
 132 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
 133                                           u64 file_offset)
 134 {
 135         struct rb_root *root = &tree->tree;
 136         struct rb_node *prev;
 137         struct rb_node *ret;
 138         struct btrfs_ordered_extent *entry;
 139
 140         if (tree->last) {
 141                 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
 142                                  rb_node);
 143                 if (offset_in_entry(entry, file_offset))
 144                         return tree->last;
 145         }
 146         ret = __tree_search(root, file_offset, &prev);
 147         if (!ret)
 148                 ret = prev;
 149         if (ret)
 150                 tree->last = ret;
 151         return ret;
 152 }
 153
 154 /* allocate and add a new ordered_extent into the per-inode tree.
 155  * file_offset is the logical offset in the file
 156  *
 157  * start is the disk block number of an extent already reserved in the
 158  * extent allocation tree
 159  *
 160  * len is the length of the extent
 161  *
 162  * This also sets the EXTENT_ORDERED bit on the range in the inode.
 163  *
 164  * The tree is given a single reference on the ordered extent that was
 165  * inserted.
 166  */
 167 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 168                              u64 start, u64 len, u64 disk_len, int type)
 169 {
 170         struct btrfs_ordered_inode_tree *tree;
 171         struct rb_node *node;
 172         struct btrfs_ordered_extent *entry;
 173
 174         tree = &BTRFS_I(inode)->ordered_tree;
 175         entry = kzalloc(sizeof(*entry), GFP_NOFS);
 176         if (!entry)
 177                 return -ENOMEM;
 178
 179         mutex_lock(&tree->mutex);
 180         entry->file_offset = file_offset;
 181         entry->start = start;
 182         entry->len = len;
 183         entry->disk_len = disk_len;
 184         entry->inode = inode;
 185         if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 186                 set_bit(type, &entry->flags);
 187
 188         /* one ref for the tree */
 189         atomic_set(&entry->refs, 1);
 190         init_waitqueue_head(&entry->wait);
 191         INIT_LIST_HEAD(&entry->list);
 192         INIT_LIST_HEAD(&entry->root_extent_list);
 193
 194         node = tree_insert(&tree->tree, file_offset,
 195                            &entry->rb_node);
 196         if (node) {
 197                 printk("warning dup entry from add_ordered_extent\n");
 198                 BUG();
 199         }
 200         set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
 201                            entry_end(entry) - 1, GFP_NOFS);
 202
 203         spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 204         list_add_tail(&entry->root_extent_list,
 205                       &BTRFS_I(inode)->root->fs_info->ordered_extents);
 206         spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 207
 208         mutex_unlock(&tree->mutex);
 209         BUG_ON(node);
 210         return 0;
 211 }
 212
 213 /*
 214  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
 215  * when an ordered extent is finished.  If the list covers more than one
 216  * ordered extent, it is split across multiples.
 217  */
 218 int btrfs_add_ordered_sum(struct inode *inode,
 219                           struct btrfs_ordered_extent *entry,
 220                           struct btrfs_ordered_sum *sum)
 221 {
 222         struct btrfs_ordered_inode_tree *tree;
 223
 224         tree = &BTRFS_I(inode)->ordered_tree;
 225         mutex_lock(&tree->mutex);
 226         list_add_tail(&sum->list, &entry->list);
 227         mutex_unlock(&tree->mutex);
 228         return 0;
 229 }
 230
 231 /*
 232  * this is used to account for finished IO across a given range
 233  * of the file.  The IO should not span ordered extents.  If
 234  * a given ordered_extent is completely done, 1 is returned, otherwise
 235  * 0.
 236  *
 237  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 238  * to make sure this function only returns 1 once for a given ordered extent.
 239  */
 240 int btrfs_dec_test_ordered_pending(struct inode *inode,
 241                                    u64 file_offset, u64 io_size)
 242 {
 243         struct btrfs_ordered_inode_tree *tree;
 244         struct rb_node *node;
 245         struct btrfs_ordered_extent *entry;
 246         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 247         int ret;
 248
 249         tree = &BTRFS_I(inode)->ordered_tree;
 250         mutex_lock(&tree->mutex);
 251         clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
 252                              GFP_NOFS);
 253         node = tree_search(tree, file_offset);
 254         if (!node) {
 255                 ret = 1;
 256                 goto out;
 257         }
 258
 259         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 260         if (!offset_in_entry(entry, file_offset)) {
 261                 ret = 1;
 262                 goto out;
 263         }
 264
 265         ret = test_range_bit(io_tree, entry->file_offset,
 266                              entry->file_offset + entry->len - 1,
 267                              EXTENT_ORDERED, 0);
 268         if (ret == 0)
 269                 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 270 out:
 271         mutex_unlock(&tree->mutex);
 272         return ret == 0;
 273 }
 274
 275 /*
 276  * used to drop a reference on an ordered extent.  This will free
 277  * the extent if the last reference is dropped
 278  */
 279 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 280 {
 281         struct list_head *cur;
 282         struct btrfs_ordered_sum *sum;
 283
 284         if (atomic_dec_and_test(&entry->refs)) {
 285                 while(!list_empty(&entry->list)) {
 286                         cur = entry->list.next;
 287                         sum = list_entry(cur, struct btrfs_ordered_sum, list);
 288                         list_del(&sum->list);
 289                         kfree(sum);
 290                 }
 291                 kfree(entry);
 292         }
 293         return 0;
 294 }
 295
 296 /*
 297  * remove an ordered extent from the tree.  No references are dropped
 298  * but, anyone waiting on this extent is woken up.
 299  */
 300 int btrfs_remove_ordered_extent(struct inode *inode,
 301                                 struct btrfs_ordered_extent *entry)
 302 {
 303         struct btrfs_ordered_inode_tree *tree;
 304         struct rb_node *node;
 305
 306         tree = &BTRFS_I(inode)->ordered_tree;
 307         mutex_lock(&tree->mutex);
 308         node = &entry->rb_node;
 309         rb_erase(node, &tree->tree);
 310         tree->last = NULL;
 311         set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 312
 313         spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 314         list_del_init(&entry->root_extent_list);
 315         spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 316
 317         mutex_unlock(&tree->mutex);
 318         wake_up(&entry->wait);
 319         return 0;
 320 }
 321
 322 /*
 323  * wait for all the ordered extents in a root.  This is done when balancing
 324  * space between drives.
 325  */
 326 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
 327 {
 328         struct list_head splice;
 329         struct list_head *cur;
 330         struct btrfs_ordered_extent *ordered;
 331         struct inode *inode;
 332
 333         INIT_LIST_HEAD(&splice);
 334
 335         spin_lock(&root->fs_info->ordered_extent_lock);
 336         list_splice_init(&root->fs_info->ordered_extents, &splice);
 337         while (!list_empty(&splice)) {
 338                 cur = splice.next;
 339                 ordered = list_entry(cur, struct btrfs_ordered_extent,
 340                                      root_extent_list);
 341                 if (nocow_only &&
 342                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
 343                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
 344                         list_move(&ordered->root_extent_list,
 345                                   &root->fs_info->ordered_extents);
 346                         cond_resched_lock(&root->fs_info->ordered_extent_lock);
 347                         continue;
 348                 }
 349
 350                 list_del_init(&ordered->root_extent_list);
 351                 atomic_inc(&ordered->refs);
 352
 353                 /*
 354                  * the inode may be getting freed (in sys_unlink path).
 355                  */
 356                 inode = igrab(ordered->inode);
 357
 358                 spin_unlock(&root->fs_info->ordered_extent_lock);
 359
 360                 if (inode) {
 361                         btrfs_start_ordered_extent(inode, ordered, 1);
 362                         btrfs_put_ordered_extent(ordered);
 363                         iput(inode);
 364                 } else {
 365                         btrfs_put_ordered_extent(ordered);
 366                 }
 367
 368                 spin_lock(&root->fs_info->ordered_extent_lock);
 369         }
 370         spin_unlock(&root->fs_info->ordered_extent_lock);
 371         return 0;
 372 }
 373
 374 /*
 375  * Used to start IO or wait for a given ordered extent to finish.
 376  *
 377  * If wait is one, this effectively waits on page writeback for all the pages
 378  * in the extent, and it waits on the io completion code to insert
 379  * metadata into the btree corresponding to the extent
 380  */
 381 void btrfs_start_ordered_extent(struct inode *inode,
 382                                        struct btrfs_ordered_extent *entry,
 383                                        int wait)
 384 {
 385         u64 start = entry->file_offset;
 386         u64 end = start + entry->len - 1;
 387
 388         /*
 389          * pages in the range can be dirty, clean or writeback.  We
 390          * start IO on any dirty ones so the wait doesn't stall waiting
 391          * for pdflush to find them
 392          */
 393         btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_ALL);
 394         if (wait) {
 395                 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
 396                                                  &entry->flags));
 397         }
 398 }
 399
 400 /*
 401  * Used to wait on ordered extents across a large range of bytes.
 402  */
 403 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 404 {
 405         u64 end;
 406         u64 orig_end;
 407         u64 wait_end;
 408         struct btrfs_ordered_extent *ordered;
 409
 410         if (start + len < start) {
 411                 orig_end = INT_LIMIT(loff_t);
 412         } else {
 413                 orig_end = start + len - 1;
 414                 if (orig_end > INT_LIMIT(loff_t))
 415                         orig_end = INT_LIMIT(loff_t);
 416         }
 417         wait_end = orig_end;
 418 again:
 419         /* start IO across the range first to instantiate any delalloc
 420          * extents
 421          */
 422         btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
 423
 424         /* The compression code will leave pages locked but return from
 425          * writepage without setting the page writeback.  Starting again
 426          * with WB_SYNC_ALL will end up waiting for the IO to actually start.
 427          */
 428         btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL);
 429
 430         btrfs_wait_on_page_writeback_range(inode->i_mapping,
 431                                            start >> PAGE_CACHE_SHIFT,
 432                                            orig_end >> PAGE_CACHE_SHIFT);
 433
 434         end = orig_end;
 435         while(1) {
 436                 ordered = btrfs_lookup_first_ordered_extent(inode, end);
 437                 if (!ordered) {
 438                         break;
 439                 }
 440                 if (ordered->file_offset > orig_end) {
 441                         btrfs_put_ordered_extent(ordered);
 442                         break;
 443                 }
 444                 if (ordered->file_offset + ordered->len < start) {
 445                         btrfs_put_ordered_extent(ordered);
 446                         break;
 447                 }
 448                 btrfs_start_ordered_extent(inode, ordered, 1);
 449                 end = ordered->file_offset;
 450                 btrfs_put_ordered_extent(ordered);
 451                 if (end == 0 || end == start)
 452                         break;
 453                 end--;
 454         }
 455         if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
 456                            EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
 457                 schedule_timeout(1);
 458                 goto again;
 459         }
 460         return 0;
 461 }
 462
 463 /*
 464  * find an ordered extent corresponding to file_offset.  return NULL if
 465  * nothing is found, otherwise take a reference on the extent and return it
 466  */
 467 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
 468                                                          u64 file_offset)
 469 {
 470         struct btrfs_ordered_inode_tree *tree;
 471         struct rb_node *node;
 472         struct btrfs_ordered_extent *entry = NULL;
 473
 474         tree = &BTRFS_I(inode)->ordered_tree;
 475         mutex_lock(&tree->mutex);
 476         node = tree_search(tree, file_offset);
 477         if (!node)
 478                 goto out;
 479
 480         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 481         if (!offset_in_entry(entry, file_offset))
 482                 entry = NULL;
 483         if (entry)
 484                 atomic_inc(&entry->refs);
 485 out:
 486         mutex_unlock(&tree->mutex);
 487         return entry;
 488 }
 489
 490 /*
 491  * lookup and return any extent before 'file_offset'.  NULL is returned
 492  * if none is found
 493  */
 494 struct btrfs_ordered_extent *
 495 btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset)
 496 {
 497         struct btrfs_ordered_inode_tree *tree;
 498         struct rb_node *node;
 499         struct btrfs_ordered_extent *entry = NULL;
 500
 501         tree = &BTRFS_I(inode)->ordered_tree;
 502         mutex_lock(&tree->mutex);
 503         node = tree_search(tree, file_offset);
 504         if (!node)
 505                 goto out;
 506
 507         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 508         atomic_inc(&entry->refs);
 509 out:
 510         mutex_unlock(&tree->mutex);
 511         return entry;
 512 }
 513
 514 /*
 515  * After an extent is done, call this to conditionally update the on disk
 516  * i_size.  i_size is updated to cover any fully written part of the file.
 517  */
 518 int btrfs_ordered_update_i_size(struct inode *inode,
 519                                 struct btrfs_ordered_extent *ordered)
 520 {
 521         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 522         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 523         u64 disk_i_size;
 524         u64 new_i_size;
 525         u64 i_size_test;
 526         struct rb_node *node;
 527         struct btrfs_ordered_extent *test;
 528
 529         mutex_lock(&tree->mutex);
 530         disk_i_size = BTRFS_I(inode)->disk_i_size;
 531
 532         /*
 533          * if the disk i_size is already at the inode->i_size, or
 534          * this ordered extent is inside the disk i_size, we're done
 535          */
 536         if (disk_i_size >= inode->i_size ||
 537             ordered->file_offset + ordered->len <= disk_i_size) {
 538                 goto out;
 539         }
 540
 541         /*
 542          * we can't update the disk_isize if there are delalloc bytes
 543          * between disk_i_size and  this ordered extent
 544          */
 545         if (test_range_bit(io_tree, disk_i_size,
 546                            ordered->file_offset + ordered->len - 1,
 547                            EXTENT_DELALLOC, 0)) {
 548                 goto out;
 549         }
 550         /*
 551          * walk backward from this ordered extent to disk_i_size.
 552          * if we find an ordered extent then we can't update disk i_size
 553          * yet
 554          */
 555         node = &ordered->rb_node;
 556         while(1) {
 557                 node = rb_prev(node);
 558                 if (!node)
 559                         break;
 560                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 561                 if (test->file_offset + test->len <= disk_i_size)
 562                         break;
 563                 if (test->file_offset >= inode->i_size)
 564                         break;
 565                 if (test->file_offset >= disk_i_size)
 566                         goto out;
 567         }
 568         new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
 569
 570         /*
 571          * at this point, we know we can safely update i_size to at least
 572          * the offset from this ordered extent.  But, we need to
 573          * walk forward and see if ios from higher up in the file have
 574          * finished.
 575          */
 576         node = rb_next(&ordered->rb_node);
 577         i_size_test = 0;
 578         if (node) {
 579                 /*
 580                  * do we have an area where IO might have finished
 581                  * between our ordered extent and the next one.
 582                  */
 583                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 584                 if (test->file_offset > entry_end(ordered)) {
 585                         i_size_test = test->file_offset;
 586                 }
 587         } else {
 588                 i_size_test = i_size_read(inode);
 589         }
 590
 591         /*
 592          * i_size_test is the end of a region after this ordered
 593          * extent where there are no ordered extents.  As long as there
 594          * are no delalloc bytes in this area, it is safe to update
 595          * disk_i_size to the end of the region.
 596          */
 597         if (i_size_test > entry_end(ordered) &&
 598             !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
 599                            EXTENT_DELALLOC, 0)) {
 600                 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
 601         }
 602         BTRFS_I(inode)->disk_i_size = new_i_size;
 603 out:
 604         mutex_unlock(&tree->mutex);
 605         return 0;
 606 }
 607
 608 /*
 609  * search the ordered extents for one corresponding to 'offset' and
 610  * try to find a checksum.  This is used because we allow pages to
 611  * be reclaimed before their checksum is actually put into the btree
 612  */
 613 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u32 *sum)
 614 {
 615         struct btrfs_ordered_sum *ordered_sum;
 616         struct btrfs_sector_sum *sector_sums;
 617         struct btrfs_ordered_extent *ordered;
 618         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 619         struct list_head *cur;
 620         unsigned long num_sectors;
 621         unsigned long i;
 622         u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
 623         int ret = 1;
 624
 625         ordered = btrfs_lookup_ordered_extent(inode, offset);
 626         if (!ordered)
 627                 return 1;
 628
 629         mutex_lock(&tree->mutex);
 630         list_for_each_prev(cur, &ordered->list) {
 631                 ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
 632                 if (offset >= ordered_sum->file_offset) {
 633                         num_sectors = ordered_sum->len / sectorsize;
 634                         sector_sums = ordered_sum->sums;
 635                         for (i = 0; i < num_sectors; i++) {
 636                                 if (sector_sums[i].offset == offset) {
 637                                         *sum = sector_sums[i].sum;
 638                                         ret = 0;
 639                                         goto out;
 640                                 }
 641                         }
 642                 }
 643         }
 644 out:
 645         mutex_unlock(&tree->mutex);
 646         btrfs_put_ordered_extent(ordered);
 647         return ret;
 648 }
 649
 650
 651 /**
 652  * taken from mm/filemap.c because it isn't exported
 653  *
 654  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 655  * @mapping:    address space structure to write
 656  * @start:      offset in bytes where the range starts
 657  * @end:        offset in bytes where the range ends (inclusive)
 658  * @sync_mode:  enable synchronous operation
 659  *
 660  * Start writeback against all of a mapping's dirty pages that lie
 661  * within the byte offsets <start, end> inclusive.
 662  *
 663  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 664  * opposed to a regular memory cleansing writeback.  The difference between
 665  * these two operations is that if a dirty page/buffer is encountered, it must
 666  * be waited upon, and not just skipped over.
 667  */
 668 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
 669                            loff_t end, int sync_mode)
 670 {
 671         struct writeback_control wbc = {
 672                 .sync_mode = sync_mode,
 673                 .nr_to_write = mapping->nrpages * 2,
 674                 .range_start = start,
 675                 .range_end = end,
 676                 .for_writepages = 1,
 677         };
 678         return btrfs_writepages(mapping, &wbc);
 679 }
 680
 681 /**
 682  * taken from mm/filemap.c because it isn't exported
 683  *
 684  * wait_on_page_writeback_range - wait for writeback to complete
 685  * @mapping:    target address_space
 686  * @start:      beginning page index
 687  * @end:        ending page index
 688  *
 689  * Wait for writeback to complete against pages indexed by start->end
 690  * inclusive
 691  */
 692 int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
 693                                        pgoff_t start, pgoff_t end)
 694 {
 695         struct pagevec pvec;
 696         int nr_pages;
 697         int ret = 0;
 698         pgoff_t index;
 699
 700         if (end < start)
 701                 return 0;
 702
 703         pagevec_init(&pvec, 0);
 704         index = start;
 705         while ((index <= end) &&
 706                         (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 707                         PAGECACHE_TAG_WRITEBACK,
 708                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
 709                 unsigned i;
 710
 711                 for (i = 0; i < nr_pages; i++) {
 712                         struct page *page = pvec.pages[i];
 713
 714                         /* until radix tree lookup accepts end_index */
 715                         if (page->index > end)
 716                                 continue;
 717
 718                         wait_on_page_writeback(page);
 719                         if (PageError(page))
 720                                 ret = -EIO;
 721                 }
 722                 pagevec_release(&pvec);
 723                 cond_resched();
 724         }
 725
 726         /* Check for outstanding write errors */
 727         if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 728                 ret = -ENOSPC;
 729         if (test_and_clear_bit(AS_EIO, &mapping->flags))
 730                 ret = -EIO;
 731
 732         return ret;
 733 }