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  * The tree is given a single reference on the ordered extent that was
 163  * inserted.
 164  */
 165 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 166                              u64 start, u64 len, u64 disk_len, int type)
 167 {
 168         struct btrfs_ordered_inode_tree *tree;
 169         struct rb_node *node;
 170         struct btrfs_ordered_extent *entry;
 171
 172         tree = &BTRFS_I(inode)->ordered_tree;
 173         entry = kzalloc(sizeof(*entry), GFP_NOFS);
 174         if (!entry)
 175                 return -ENOMEM;
 176
 177         mutex_lock(&tree->mutex);
 178         entry->file_offset = file_offset;
 179         entry->start = start;
 180         entry->len = len;
 181         entry->disk_len = disk_len;
 182         entry->bytes_left = len;
 183         entry->inode = inode;
 184         if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 185                 set_bit(type, &entry->flags);
 186
 187         /* one ref for the tree */
 188         atomic_set(&entry->refs, 1);
 189         init_waitqueue_head(&entry->wait);
 190         INIT_LIST_HEAD(&entry->list);
 191         INIT_LIST_HEAD(&entry->root_extent_list);
 192
 193         node = tree_insert(&tree->tree, file_offset,
 194                            &entry->rb_node);
 195         BUG_ON(node);
 196
 197         spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 198         list_add_tail(&entry->root_extent_list,
 199                       &BTRFS_I(inode)->root->fs_info->ordered_extents);
 200         spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 201
 202         mutex_unlock(&tree->mutex);
 203         BUG_ON(node);
 204         return 0;
 205 }
 206
 207 /*
 208  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
 209  * when an ordered extent is finished.  If the list covers more than one
 210  * ordered extent, it is split across multiples.
 211  */
 212 int btrfs_add_ordered_sum(struct inode *inode,
 213                           struct btrfs_ordered_extent *entry,
 214                           struct btrfs_ordered_sum *sum)
 215 {
 216         struct btrfs_ordered_inode_tree *tree;
 217
 218         tree = &BTRFS_I(inode)->ordered_tree;
 219         mutex_lock(&tree->mutex);
 220         list_add_tail(&sum->list, &entry->list);
 221         mutex_unlock(&tree->mutex);
 222         return 0;
 223 }
 224
 225 /*
 226  * this is used to account for finished IO across a given range
 227  * of the file.  The IO should not span ordered extents.  If
 228  * a given ordered_extent is completely done, 1 is returned, otherwise
 229  * 0.
 230  *
 231  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 232  * to make sure this function only returns 1 once for a given ordered extent.
 233  */
 234 int btrfs_dec_test_ordered_pending(struct inode *inode,
 235                                    u64 file_offset, u64 io_size)
 236 {
 237         struct btrfs_ordered_inode_tree *tree;
 238         struct rb_node *node;
 239         struct btrfs_ordered_extent *entry;
 240         int ret;
 241
 242         tree = &BTRFS_I(inode)->ordered_tree;
 243         mutex_lock(&tree->mutex);
 244         node = tree_search(tree, file_offset);
 245         if (!node) {
 246                 ret = 1;
 247                 goto out;
 248         }
 249
 250         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 251         if (!offset_in_entry(entry, file_offset)) {
 252                 ret = 1;
 253                 goto out;
 254         }
 255
 256         if (io_size > entry->bytes_left) {
 257                 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
 258                        (unsigned long long)entry->bytes_left,
 259                        (unsigned long long)io_size);
 260         }
 261         entry->bytes_left -= io_size;
 262         if (entry->bytes_left == 0)
 263                 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 264         else
 265                 ret = 1;
 266 out:
 267         mutex_unlock(&tree->mutex);
 268         return ret == 0;
 269 }
 270
 271 /*
 272  * used to drop a reference on an ordered extent.  This will free
 273  * the extent if the last reference is dropped
 274  */
 275 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 276 {
 277         struct list_head *cur;
 278         struct btrfs_ordered_sum *sum;
 279
 280         if (atomic_dec_and_test(&entry->refs)) {
 281                 while (!list_empty(&entry->list)) {
 282                         cur = entry->list.next;
 283                         sum = list_entry(cur, struct btrfs_ordered_sum, list);
 284                         list_del(&sum->list);
 285                         kfree(sum);
 286                 }
 287                 kfree(entry);
 288         }
 289         return 0;
 290 }
 291
 292 /*
 293  * remove an ordered extent from the tree.  No references are dropped
 294  * and you must wake_up entry->wait.  You must hold the tree mutex
 295  * while you call this function.
 296  */
 297 static int __btrfs_remove_ordered_extent(struct inode *inode,
 298                                 struct btrfs_ordered_extent *entry)
 299 {
 300         struct btrfs_ordered_inode_tree *tree;
 301         struct rb_node *node;
 302
 303         tree = &BTRFS_I(inode)->ordered_tree;
 304         node = &entry->rb_node;
 305         rb_erase(node, &tree->tree);
 306         tree->last = NULL;
 307         set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 308
 309         spin_lock(&BTRFS_I(inode)->accounting_lock);
 310         BTRFS_I(inode)->outstanding_extents--;
 311         spin_unlock(&BTRFS_I(inode)->accounting_lock);
 312         btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
 313                                               inode, 1);
 314
 315         spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 316         list_del_init(&entry->root_extent_list);
 317
 318         /*
 319          * we have no more ordered extents for this inode and
 320          * no dirty pages.  We can safely remove it from the
 321          * list of ordered extents
 322          */
 323         if (RB_EMPTY_ROOT(&tree->tree) &&
 324             !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 325                 list_del_init(&BTRFS_I(inode)->ordered_operations);
 326         }
 327         spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 328
 329         return 0;
 330 }
 331
 332 /*
 333  * remove an ordered extent from the tree.  No references are dropped
 334  * but any waiters are woken.
 335  */
 336 int btrfs_remove_ordered_extent(struct inode *inode,
 337                                 struct btrfs_ordered_extent *entry)
 338 {
 339         struct btrfs_ordered_inode_tree *tree;
 340         int ret;
 341
 342         tree = &BTRFS_I(inode)->ordered_tree;
 343         mutex_lock(&tree->mutex);
 344         ret = __btrfs_remove_ordered_extent(inode, entry);
 345         mutex_unlock(&tree->mutex);
 346         wake_up(&entry->wait);
 347
 348         return ret;
 349 }
 350
 351 /*
 352  * wait for all the ordered extents in a root.  This is done when balancing
 353  * space between drives.
 354  */
 355 int btrfs_wait_ordered_extents(struct btrfs_root *root,
 356                                int nocow_only, int delay_iput)
 357 {
 358         struct list_head splice;
 359         struct list_head *cur;
 360         struct btrfs_ordered_extent *ordered;
 361         struct inode *inode;
 362
 363         INIT_LIST_HEAD(&splice);
 364
 365         spin_lock(&root->fs_info->ordered_extent_lock);
 366         list_splice_init(&root->fs_info->ordered_extents, &splice);
 367         while (!list_empty(&splice)) {
 368                 cur = splice.next;
 369                 ordered = list_entry(cur, struct btrfs_ordered_extent,
 370                                      root_extent_list);
 371                 if (nocow_only &&
 372                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
 373                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
 374                         list_move(&ordered->root_extent_list,
 375                                   &root->fs_info->ordered_extents);
 376                         cond_resched_lock(&root->fs_info->ordered_extent_lock);
 377                         continue;
 378                 }
 379
 380                 list_del_init(&ordered->root_extent_list);
 381                 atomic_inc(&ordered->refs);
 382
 383                 /*
 384                  * the inode may be getting freed (in sys_unlink path).
 385                  */
 386                 inode = igrab(ordered->inode);
 387
 388                 spin_unlock(&root->fs_info->ordered_extent_lock);
 389
 390                 if (inode) {
 391                         btrfs_start_ordered_extent(inode, ordered, 1);
 392                         btrfs_put_ordered_extent(ordered);
 393                         if (delay_iput)
 394                                 btrfs_add_delayed_iput(inode);
 395                         else
 396                                 iput(inode);
 397                 } else {
 398                         btrfs_put_ordered_extent(ordered);
 399                 }
 400
 401                 spin_lock(&root->fs_info->ordered_extent_lock);
 402         }
 403         spin_unlock(&root->fs_info->ordered_extent_lock);
 404         return 0;
 405 }
 406
 407 /*
 408  * this is used during transaction commit to write all the inodes
 409  * added to the ordered operation list.  These files must be fully on
 410  * disk before the transaction commits.
 411  *
 412  * we have two modes here, one is to just start the IO via filemap_flush
 413  * and the other is to wait for all the io.  When we wait, we have an
 414  * extra check to make sure the ordered operation list really is empty
 415  * before we return
 416  */
 417 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
 418 {
 419         struct btrfs_inode *btrfs_inode;
 420         struct inode *inode;
 421         struct list_head splice;
 422
 423         INIT_LIST_HEAD(&splice);
 424
 425         mutex_lock(&root->fs_info->ordered_operations_mutex);
 426         spin_lock(&root->fs_info->ordered_extent_lock);
 427 again:
 428         list_splice_init(&root->fs_info->ordered_operations, &splice);
 429
 430         while (!list_empty(&splice)) {
 431                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
 432                                    ordered_operations);
 433
 434                 inode = &btrfs_inode->vfs_inode;
 435
 436                 list_del_init(&btrfs_inode->ordered_operations);
 437
 438                 /*
 439                  * the inode may be getting freed (in sys_unlink path).
 440                  */
 441                 inode = igrab(inode);
 442
 443                 if (!wait && inode) {
 444                         list_add_tail(&BTRFS_I(inode)->ordered_operations,
 445                               &root->fs_info->ordered_operations);
 446                 }
 447                 spin_unlock(&root->fs_info->ordered_extent_lock);
 448
 449                 if (inode) {
 450                         if (wait)
 451                                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
 452                         else
 453                                 filemap_flush(inode->i_mapping);
 454                         btrfs_add_delayed_iput(inode);
 455                 }
 456
 457                 cond_resched();
 458                 spin_lock(&root->fs_info->ordered_extent_lock);
 459         }
 460         if (wait && !list_empty(&root->fs_info->ordered_operations))
 461                 goto again;
 462
 463         spin_unlock(&root->fs_info->ordered_extent_lock);
 464         mutex_unlock(&root->fs_info->ordered_operations_mutex);
 465
 466         return 0;
 467 }
 468
 469 /*
 470  * Used to start IO or wait for a given ordered extent to finish.
 471  *
 472  * If wait is one, this effectively waits on page writeback for all the pages
 473  * in the extent, and it waits on the io completion code to insert
 474  * metadata into the btree corresponding to the extent
 475  */
 476 void btrfs_start_ordered_extent(struct inode *inode,
 477                                        struct btrfs_ordered_extent *entry,
 478                                        int wait)
 479 {
 480         u64 start = entry->file_offset;
 481         u64 end = start + entry->len - 1;
 482
 483         /*
 484          * pages in the range can be dirty, clean or writeback.  We
 485          * start IO on any dirty ones so the wait doesn't stall waiting
 486          * for pdflush to find them
 487          */
 488         filemap_fdatawrite_range(inode->i_mapping, start, end);
 489         if (wait) {
 490                 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
 491                                                  &entry->flags));
 492         }
 493 }
 494
 495 /*
 496  * Used to wait on ordered extents across a large range of bytes.
 497  */
 498 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 499 {
 500         u64 end;
 501         u64 orig_end;
 502         u64 wait_end;
 503         struct btrfs_ordered_extent *ordered;
 504         int found;
 505
 506         if (start + len < start) {
 507                 orig_end = INT_LIMIT(loff_t);
 508         } else {
 509                 orig_end = start + len - 1;
 510                 if (orig_end > INT_LIMIT(loff_t))
 511                         orig_end = INT_LIMIT(loff_t);
 512         }
 513         wait_end = orig_end;
 514 again:
 515         /* start IO across the range first to instantiate any delalloc
 516          * extents
 517          */
 518         filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 519
 520         /* The compression code will leave pages locked but return from
 521          * writepage without setting the page writeback.  Starting again
 522          * with WB_SYNC_ALL will end up waiting for the IO to actually start.
 523          */
 524         filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 525
 526         filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 527
 528         end = orig_end;
 529         found = 0;
 530         while (1) {
 531                 ordered = btrfs_lookup_first_ordered_extent(inode, end);
 532                 if (!ordered)
 533                         break;
 534                 if (ordered->file_offset > orig_end) {
 535                         btrfs_put_ordered_extent(ordered);
 536                         break;
 537                 }
 538                 if (ordered->file_offset + ordered->len < start) {
 539                         btrfs_put_ordered_extent(ordered);
 540                         break;
 541                 }
 542                 found++;
 543                 btrfs_start_ordered_extent(inode, ordered, 1);
 544                 end = ordered->file_offset;
 545                 btrfs_put_ordered_extent(ordered);
 546                 if (end == 0 || end == start)
 547                         break;
 548                 end--;
 549         }
 550         if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
 551                            EXTENT_DELALLOC, 0, NULL)) {
 552                 schedule_timeout(1);
 553                 goto again;
 554         }
 555         return 0;
 556 }
 557
 558 /*
 559  * find an ordered extent corresponding to file_offset.  return NULL if
 560  * nothing is found, otherwise take a reference on the extent and return it
 561  */
 562 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
 563                                                          u64 file_offset)
 564 {
 565         struct btrfs_ordered_inode_tree *tree;
 566         struct rb_node *node;
 567         struct btrfs_ordered_extent *entry = NULL;
 568
 569         tree = &BTRFS_I(inode)->ordered_tree;
 570         mutex_lock(&tree->mutex);
 571         node = tree_search(tree, file_offset);
 572         if (!node)
 573                 goto out;
 574
 575         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 576         if (!offset_in_entry(entry, file_offset))
 577                 entry = NULL;
 578         if (entry)
 579                 atomic_inc(&entry->refs);
 580 out:
 581         mutex_unlock(&tree->mutex);
 582         return entry;
 583 }
 584
 585 /*
 586  * lookup and return any extent before 'file_offset'.  NULL is returned
 587  * if none is found
 588  */
 589 struct btrfs_ordered_extent *
 590 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
 591 {
 592         struct btrfs_ordered_inode_tree *tree;
 593         struct rb_node *node;
 594         struct btrfs_ordered_extent *entry = NULL;
 595
 596         tree = &BTRFS_I(inode)->ordered_tree;
 597         mutex_lock(&tree->mutex);
 598         node = tree_search(tree, file_offset);
 599         if (!node)
 600                 goto out;
 601
 602         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 603         atomic_inc(&entry->refs);
 604 out:
 605         mutex_unlock(&tree->mutex);
 606         return entry;
 607 }
 608
 609 /*
 610  * After an extent is done, call this to conditionally update the on disk
 611  * i_size.  i_size is updated to cover any fully written part of the file.
 612  */
 613 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
 614                                 struct btrfs_ordered_extent *ordered)
 615 {
 616         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 617         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 618         u64 disk_i_size;
 619         u64 new_i_size;
 620         u64 i_size_test;
 621         u64 i_size = i_size_read(inode);
 622         struct rb_node *node;
 623         struct rb_node *prev = NULL;
 624         struct btrfs_ordered_extent *test;
 625         int ret = 1;
 626
 627         if (ordered)
 628                 offset = entry_end(ordered);
 629         else
 630                 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
 631
 632         mutex_lock(&tree->mutex);
 633         disk_i_size = BTRFS_I(inode)->disk_i_size;
 634
 635         /* truncate file */
 636         if (disk_i_size > i_size) {
 637                 BTRFS_I(inode)->disk_i_size = i_size;
 638                 ret = 0;
 639                 goto out;
 640         }
 641
 642         /*
 643          * if the disk i_size is already at the inode->i_size, or
 644          * this ordered extent is inside the disk i_size, we're done
 645          */
 646         if (disk_i_size == i_size || offset <= disk_i_size) {
 647                 goto out;
 648         }
 649
 650         /*
 651          * we can't update the disk_isize if there are delalloc bytes
 652          * between disk_i_size and  this ordered extent
 653          */
 654         if (test_range_bit(io_tree, disk_i_size, offset - 1,
 655                            EXTENT_DELALLOC, 0, NULL)) {
 656                 goto out;
 657         }
 658         /*
 659          * walk backward from this ordered extent to disk_i_size.
 660          * if we find an ordered extent then we can't update disk i_size
 661          * yet
 662          */
 663         if (ordered) {
 664                 node = rb_prev(&ordered->rb_node);
 665         } else {
 666                 prev = tree_search(tree, offset);
 667                 /*
 668                  * we insert file extents without involving ordered struct,
 669                  * so there should be no ordered struct cover this offset
 670                  */
 671                 if (prev) {
 672                         test = rb_entry(prev, struct btrfs_ordered_extent,
 673                                         rb_node);
 674                         BUG_ON(offset_in_entry(test, offset));
 675                 }
 676                 node = prev;
 677         }
 678         while (node) {
 679                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 680                 if (test->file_offset + test->len <= disk_i_size)
 681                         break;
 682                 if (test->file_offset >= i_size)
 683                         break;
 684                 if (test->file_offset >= disk_i_size)
 685                         goto out;
 686                 node = rb_prev(node);
 687         }
 688         new_i_size = min_t(u64, offset, i_size);
 689
 690         /*
 691          * at this point, we know we can safely update i_size to at least
 692          * the offset from this ordered extent.  But, we need to
 693          * walk forward and see if ios from higher up in the file have
 694          * finished.
 695          */
 696         if (ordered) {
 697                 node = rb_next(&ordered->rb_node);
 698         } else {
 699                 if (prev)
 700                         node = rb_next(prev);
 701                 else
 702                         node = rb_first(&tree->tree);
 703         }
 704         i_size_test = 0;
 705         if (node) {
 706                 /*
 707                  * do we have an area where IO might have finished
 708                  * between our ordered extent and the next one.
 709                  */
 710                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 711                 if (test->file_offset > offset)
 712                         i_size_test = test->file_offset;
 713         } else {
 714                 i_size_test = i_size;
 715         }
 716
 717         /*
 718          * i_size_test is the end of a region after this ordered
 719          * extent where there are no ordered extents.  As long as there
 720          * are no delalloc bytes in this area, it is safe to update
 721          * disk_i_size to the end of the region.
 722          */
 723         if (i_size_test > offset &&
 724             !test_range_bit(io_tree, offset, i_size_test - 1,
 725                             EXTENT_DELALLOC, 0, NULL)) {
 726                 new_i_size = min_t(u64, i_size_test, i_size);
 727         }
 728         BTRFS_I(inode)->disk_i_size = new_i_size;
 729         ret = 0;
 730 out:
 731         /*
 732          * we need to remove the ordered extent with the tree lock held
 733          * so that other people calling this function don't find our fully
 734          * processed ordered entry and skip updating the i_size
 735          */
 736         if (ordered)
 737                 __btrfs_remove_ordered_extent(inode, ordered);
 738         mutex_unlock(&tree->mutex);
 739         if (ordered)
 740                 wake_up(&ordered->wait);
 741         return ret;
 742 }
 743
 744 /*
 745  * search the ordered extents for one corresponding to 'offset' and
 746  * try to find a checksum.  This is used because we allow pages to
 747  * be reclaimed before their checksum is actually put into the btree
 748  */
 749 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
 750                            u32 *sum)
 751 {
 752         struct btrfs_ordered_sum *ordered_sum;
 753         struct btrfs_sector_sum *sector_sums;
 754         struct btrfs_ordered_extent *ordered;
 755         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 756         unsigned long num_sectors;
 757         unsigned long i;
 758         u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
 759         int ret = 1;
 760
 761         ordered = btrfs_lookup_ordered_extent(inode, offset);
 762         if (!ordered)
 763                 return 1;
 764
 765         mutex_lock(&tree->mutex);
 766         list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
 767                 if (disk_bytenr >= ordered_sum->bytenr) {
 768                         num_sectors = ordered_sum->len / sectorsize;
 769                         sector_sums = ordered_sum->sums;
 770                         for (i = 0; i < num_sectors; i++) {
 771                                 if (sector_sums[i].bytenr == disk_bytenr) {
 772                                         *sum = sector_sums[i].sum;
 773                                         ret = 0;
 774                                         goto out;
 775                                 }
 776                         }
 777                 }
 778         }
 779 out:
 780         mutex_unlock(&tree->mutex);
 781         btrfs_put_ordered_extent(ordered);
 782         return ret;
 783 }
 784
 785
 786 /*
 787  * add a given inode to the list of inodes that must be fully on
 788  * disk before a transaction commit finishes.
 789  *
 790  * This basically gives us the ext3 style data=ordered mode, and it is mostly
 791  * used to make sure renamed files are fully on disk.
 792  *
 793  * It is a noop if the inode is already fully on disk.
 794  *
 795  * If trans is not null, we'll do a friendly check for a transaction that
 796  * is already flushing things and force the IO down ourselves.
 797  */
 798 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
 799                                 struct btrfs_root *root,
 800                                 struct inode *inode)
 801 {
 802         u64 last_mod;
 803
 804         last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
 805
 806         /*
 807          * if this file hasn't been changed since the last transaction
 808          * commit, we can safely return without doing anything
 809          */
 810         if (last_mod < root->fs_info->last_trans_committed)
 811                 return 0;
 812
 813         /*
 814          * the transaction is already committing.  Just start the IO and
 815          * don't bother with all of this list nonsense
 816          */
 817         if (trans && root->fs_info->running_transaction->blocked) {
 818                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
 819                 return 0;
 820         }
 821
 822         spin_lock(&root->fs_info->ordered_extent_lock);
 823         if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
 824                 list_add_tail(&BTRFS_I(inode)->ordered_operations,
 825                               &root->fs_info->ordered_operations);
 826         }
 827         spin_unlock(&root->fs_info->ordered_extent_lock);
 828
 829         return 0;
 830 }