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