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/slab.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/writeback.h>
  22 #include <linux/pagevec.h>
  23 #include "ctree.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "extent_io.h"
  27
  28 static u64 entry_end(struct btrfs_ordered_extent *entry)
  29 {
  30         if (entry->file_offset + entry->len < entry->file_offset)
  31                 return (u64)-1;
  32         return entry->file_offset + entry->len;
  33 }
  34
  35 /* returns NULL if the insertion worked, or it returns the node it did find
  36  * in the tree
  37  */
  38 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
  39                                    struct rb_node *node)
  40 {
  41         struct rb_node **p = &root->rb_node;
  42         struct rb_node *parent = NULL;
  43         struct btrfs_ordered_extent *entry;
  44
  45         while (*p) {
  46                 parent = *p;
  47                 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
  48
  49                 if (file_offset < entry->file_offset)
  50                         p = &(*p)->rb_left;
  51                 else if (file_offset >= entry_end(entry))
  52                         p = &(*p)->rb_right;
  53                 else
  54                         return parent;
  55         }
  56
  57         rb_link_node(node, parent, p);
  58         rb_insert_color(node, root);
  59         return NULL;
  60 }
  61
  62 /*
  63  * look for a given offset in the tree, and if it can't be found return the
  64  * first lesser offset
  65  */
  66 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
  67                                      struct rb_node **prev_ret)
  68 {
  69         struct rb_node *n = root->rb_node;
  70         struct rb_node *prev = NULL;
  71         struct rb_node *test;
  72         struct btrfs_ordered_extent *entry;
  73         struct btrfs_ordered_extent *prev_entry = NULL;
  74
  75         while (n) {
  76                 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
  77                 prev = n;
  78                 prev_entry = entry;
  79
  80                 if (file_offset < entry->file_offset)
  81                         n = n->rb_left;
  82                 else if (file_offset >= entry_end(entry))
  83                         n = n->rb_right;
  84                 else
  85                         return n;
  86         }
  87         if (!prev_ret)
  88                 return NULL;
  89
  90         while (prev && file_offset >= entry_end(prev_entry)) {
  91                 test = rb_next(prev);
  92                 if (!test)
  93                         break;
  94                 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
  95                                       rb_node);
  96                 if (file_offset < entry_end(prev_entry))
  97                         break;
  98
  99                 prev = test;
 100         }
 101         if (prev)
 102                 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
 103                                       rb_node);
 104         while (prev && file_offset < entry_end(prev_entry)) {
 105                 test = rb_prev(prev);
 106                 if (!test)
 107                         break;
 108                 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
 109                                       rb_node);
 110                 prev = test;
 111         }
 112         *prev_ret = prev;
 113         return NULL;
 114 }
 115
 116 /*
 117  * helper to check if a given offset is inside a given entry
 118  */
 119 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
 120 {
 121         if (file_offset < entry->file_offset ||
 122             entry->file_offset + entry->len <= file_offset)
 123                 return 0;
 124         return 1;
 125 }
 126
 127 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
 128                           u64 len)
 129 {
 130         if (file_offset + len <= entry->file_offset ||
 131             entry->file_offset + entry->len <= file_offset)
 132                 return 0;
 133         return 1;
 134 }
 135
 136 /*
 137  * look find the first ordered struct that has this offset, otherwise
 138  * the first one less than this offset
 139  */
 140 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
 141                                           u64 file_offset)
 142 {
 143         struct rb_root *root = &tree->tree;
 144         struct rb_node *prev;
 145         struct rb_node *ret;
 146         struct btrfs_ordered_extent *entry;
 147
 148         if (tree->last) {
 149                 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
 150                                  rb_node);
 151                 if (offset_in_entry(entry, file_offset))
 152                         return tree->last;
 153         }
 154         ret = __tree_search(root, file_offset, &prev);
 155         if (!ret)
 156                 ret = prev;
 157         if (ret)
 158                 tree->last = ret;
 159         return ret;
 160 }
 161
 162 /* allocate and add a new ordered_extent into the per-inode tree.
 163  * file_offset is the logical offset in the file
 164  *
 165  * start is the disk block number of an extent already reserved in the
 166  * extent allocation tree
 167  *
 168  * len is the length of the extent
 169  *
 170  * The tree is given a single reference on the ordered extent that was
 171  * inserted.
 172  */
 173 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 174                                       u64 start, u64 len, u64 disk_len,
 175                                       int type, int dio)
 176 {
 177         struct btrfs_ordered_inode_tree *tree;
 178         struct rb_node *node;
 179         struct btrfs_ordered_extent *entry;
 180
 181         tree = &BTRFS_I(inode)->ordered_tree;
 182         entry = kzalloc(sizeof(*entry), GFP_NOFS);
 183         if (!entry)
 184                 return -ENOMEM;
 185
 186         entry->file_offset = file_offset;
 187         entry->start = start;
 188         entry->len = len;
 189         entry->disk_len = disk_len;
 190         entry->bytes_left = len;
 191         entry->inode = inode;
 192         if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 193                 set_bit(type, &entry->flags);
 194
 195         if (dio)
 196                 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
 197
 198         /* one ref for the tree */
 199         atomic_set(&entry->refs, 1);
 200         init_waitqueue_head(&entry->wait);
 201         INIT_LIST_HEAD(&entry->list);
 202         INIT_LIST_HEAD(&entry->root_extent_list);
 203
 204         spin_lock(&tree->lock);
 205         node = tree_insert(&tree->tree, file_offset,
 206                            &entry->rb_node);
 207         BUG_ON(node);
 208         spin_unlock(&tree->lock);
 209
 210         spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 211         list_add_tail(&entry->root_extent_list,
 212                       &BTRFS_I(inode)->root->fs_info->ordered_extents);
 213         spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 214
 215         BUG_ON(node);
 216         return 0;
 217 }
 218
 219 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 220                              u64 start, u64 len, u64 disk_len, int type)
 221 {
 222         return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 223                                           disk_len, type, 0);
 224 }
 225
 226 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
 227                                  u64 start, u64 len, u64 disk_len, int type)
 228 {
 229         return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 230                                           disk_len, type, 1);
 231 }
 232
 233 /*
 234  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
 235  * when an ordered extent is finished.  If the list covers more than one
 236  * ordered extent, it is split across multiples.
 237  */
 238 int btrfs_add_ordered_sum(struct inode *inode,
 239                           struct btrfs_ordered_extent *entry,
 240                           struct btrfs_ordered_sum *sum)
 241 {
 242         struct btrfs_ordered_inode_tree *tree;
 243
 244         tree = &BTRFS_I(inode)->ordered_tree;
 245         spin_lock(&tree->lock);
 246         list_add_tail(&sum->list, &entry->list);
 247         spin_unlock(&tree->lock);
 248         return 0;
 249 }
 250
 251 /*
 252  * this is used to account for finished IO across a given range
 253  * of the file.  The IO may span ordered extents.  If
 254  * a given ordered_extent is completely done, 1 is returned, otherwise
 255  * 0.
 256  *
 257  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 258  * to make sure this function only returns 1 once for a given ordered extent.
 259  *
 260  * file_offset is updated to one byte past the range that is recorded as
 261  * complete.  This allows you to walk forward in the file.
 262  */
 263 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
 264                                    struct btrfs_ordered_extent **cached,
 265                                    u64 *file_offset, u64 io_size)
 266 {
 267         struct btrfs_ordered_inode_tree *tree;
 268         struct rb_node *node;
 269         struct btrfs_ordered_extent *entry = NULL;
 270         int ret;
 271         u64 dec_end;
 272         u64 dec_start;
 273         u64 to_dec;
 274
 275         tree = &BTRFS_I(inode)->ordered_tree;
 276         spin_lock(&tree->lock);
 277         node = tree_search(tree, *file_offset);
 278         if (!node) {
 279                 ret = 1;
 280                 goto out;
 281         }
 282
 283         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 284         if (!offset_in_entry(entry, *file_offset)) {
 285                 ret = 1;
 286                 goto out;
 287         }
 288
 289         dec_start = max(*file_offset, entry->file_offset);
 290         dec_end = min(*file_offset + io_size, entry->file_offset +
 291                       entry->len);
 292         *file_offset = dec_end;
 293         if (dec_start > dec_end) {
 294                 printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n",
 295                        (unsigned long long)dec_start,
 296                        (unsigned long long)dec_end);
 297         }
 298         to_dec = dec_end - dec_start;
 299         if (to_dec > entry->bytes_left) {
 300                 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
 301                        (unsigned long long)entry->bytes_left,
 302                        (unsigned long long)to_dec);
 303         }
 304         entry->bytes_left -= to_dec;
 305         if (entry->bytes_left == 0)
 306                 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 307         else
 308                 ret = 1;
 309 out:
 310         if (!ret && cached && entry) {
 311                 *cached = entry;
 312                 atomic_inc(&entry->refs);
 313         }
 314         spin_unlock(&tree->lock);
 315         return ret == 0;
 316 }
 317
 318 /*
 319  * this is used to account for finished IO across a given range
 320  * of the file.  The IO should not span ordered extents.  If
 321  * a given ordered_extent is completely done, 1 is returned, otherwise
 322  * 0.
 323  *
 324  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 325  * to make sure this function only returns 1 once for a given ordered extent.
 326  */
 327 int btrfs_dec_test_ordered_pending(struct inode *inode,
 328                                    struct btrfs_ordered_extent **cached,
 329                                    u64 file_offset, u64 io_size)
 330 {
 331         struct btrfs_ordered_inode_tree *tree;
 332         struct rb_node *node;
 333         struct btrfs_ordered_extent *entry = NULL;
 334         int ret;
 335
 336         tree = &BTRFS_I(inode)->ordered_tree;
 337         spin_lock(&tree->lock);
 338         node = tree_search(tree, file_offset);
 339         if (!node) {
 340                 ret = 1;
 341                 goto out;
 342         }
 343
 344         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 345         if (!offset_in_entry(entry, file_offset)) {
 346                 ret = 1;
 347                 goto out;
 348         }
 349
 350         if (io_size > entry->bytes_left) {
 351                 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
 352                        (unsigned long long)entry->bytes_left,
 353                        (unsigned long long)io_size);
 354         }
 355         entry->bytes_left -= io_size;
 356         if (entry->bytes_left == 0)
 357                 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 358         else
 359                 ret = 1;
 360 out:
 361         if (!ret && cached && entry) {
 362                 *cached = entry;
 363                 atomic_inc(&entry->refs);
 364         }
 365         spin_unlock(&tree->lock);
 366         return ret == 0;
 367 }
 368
 369 /*
 370  * used to drop a reference on an ordered extent.  This will free
 371  * the extent if the last reference is dropped
 372  */
 373 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 374 {
 375         struct list_head *cur;
 376         struct btrfs_ordered_sum *sum;
 377
 378         if (atomic_dec_and_test(&entry->refs)) {
 379                 while (!list_empty(&entry->list)) {
 380                         cur = entry->list.next;
 381                         sum = list_entry(cur, struct btrfs_ordered_sum, list);
 382                         list_del(&sum->list);
 383                         kfree(sum);
 384                 }
 385                 kfree(entry);
 386         }
 387         return 0;
 388 }
 389
 390 /*
 391  * remove an ordered extent from the tree.  No references are dropped
 392  * and you must wake_up entry->wait.  You must hold the tree lock
 393  * while you call this function.
 394  */
 395 static int __btrfs_remove_ordered_extent(struct inode *inode,
 396                                 struct btrfs_ordered_extent *entry)
 397 {
 398         struct btrfs_ordered_inode_tree *tree;
 399         struct btrfs_root *root = BTRFS_I(inode)->root;
 400         struct rb_node *node;
 401
 402         tree = &BTRFS_I(inode)->ordered_tree;
 403         node = &entry->rb_node;
 404         rb_erase(node, &tree->tree);
 405         tree->last = NULL;
 406         set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 407
 408         spin_lock(&root->fs_info->ordered_extent_lock);
 409         list_del_init(&entry->root_extent_list);
 410
 411         /*
 412          * we have no more ordered extents for this inode and
 413          * no dirty pages.  We can safely remove it from the
 414          * list of ordered extents
 415          */
 416         if (RB_EMPTY_ROOT(&tree->tree) &&
 417             !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 418                 list_del_init(&BTRFS_I(inode)->ordered_operations);
 419         }
 420         spin_unlock(&root->fs_info->ordered_extent_lock);
 421
 422         return 0;
 423 }
 424
 425 /*
 426  * remove an ordered extent from the tree.  No references are dropped
 427  * but any waiters are woken.
 428  */
 429 int btrfs_remove_ordered_extent(struct inode *inode,
 430                                 struct btrfs_ordered_extent *entry)
 431 {
 432         struct btrfs_ordered_inode_tree *tree;
 433         int ret;
 434
 435         tree = &BTRFS_I(inode)->ordered_tree;
 436         spin_lock(&tree->lock);
 437         ret = __btrfs_remove_ordered_extent(inode, entry);
 438         spin_unlock(&tree->lock);
 439         wake_up(&entry->wait);
 440
 441         return ret;
 442 }
 443
 444 /*
 445  * wait for all the ordered extents in a root.  This is done when balancing
 446  * space between drives.
 447  */
 448 int btrfs_wait_ordered_extents(struct btrfs_root *root,
 449                                int nocow_only, int delay_iput)
 450 {
 451         struct list_head splice;
 452         struct list_head *cur;
 453         struct btrfs_ordered_extent *ordered;
 454         struct inode *inode;
 455
 456         INIT_LIST_HEAD(&splice);
 457
 458         spin_lock(&root->fs_info->ordered_extent_lock);
 459         list_splice_init(&root->fs_info->ordered_extents, &splice);
 460         while (!list_empty(&splice)) {
 461                 cur = splice.next;
 462                 ordered = list_entry(cur, struct btrfs_ordered_extent,
 463                                      root_extent_list);
 464                 if (nocow_only &&
 465                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
 466                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
 467                         list_move(&ordered->root_extent_list,
 468                                   &root->fs_info->ordered_extents);
 469                         cond_resched_lock(&root->fs_info->ordered_extent_lock);
 470                         continue;
 471                 }
 472
 473                 list_del_init(&ordered->root_extent_list);
 474                 atomic_inc(&ordered->refs);
 475
 476                 /*
 477                  * the inode may be getting freed (in sys_unlink path).
 478                  */
 479                 inode = igrab(ordered->inode);
 480
 481                 spin_unlock(&root->fs_info->ordered_extent_lock);
 482
 483                 if (inode) {
 484                         btrfs_start_ordered_extent(inode, ordered, 1);
 485                         btrfs_put_ordered_extent(ordered);
 486                         if (delay_iput)
 487                                 btrfs_add_delayed_iput(inode);
 488                         else
 489                                 iput(inode);
 490                 } else {
 491                         btrfs_put_ordered_extent(ordered);
 492                 }
 493
 494                 spin_lock(&root->fs_info->ordered_extent_lock);
 495         }
 496         spin_unlock(&root->fs_info->ordered_extent_lock);
 497         return 0;
 498 }
 499
 500 /*
 501  * this is used during transaction commit to write all the inodes
 502  * added to the ordered operation list.  These files must be fully on
 503  * disk before the transaction commits.
 504  *
 505  * we have two modes here, one is to just start the IO via filemap_flush
 506  * and the other is to wait for all the io.  When we wait, we have an
 507  * extra check to make sure the ordered operation list really is empty
 508  * before we return
 509  */
 510 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
 511 {
 512         struct btrfs_inode *btrfs_inode;
 513         struct inode *inode;
 514         struct list_head splice;
 515
 516         INIT_LIST_HEAD(&splice);
 517
 518         mutex_lock(&root->fs_info->ordered_operations_mutex);
 519         spin_lock(&root->fs_info->ordered_extent_lock);
 520 again:
 521         list_splice_init(&root->fs_info->ordered_operations, &splice);
 522
 523         while (!list_empty(&splice)) {
 524                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
 525                                    ordered_operations);
 526
 527                 inode = &btrfs_inode->vfs_inode;
 528
 529                 list_del_init(&btrfs_inode->ordered_operations);
 530
 531                 /*
 532                  * the inode may be getting freed (in sys_unlink path).
 533                  */
 534                 inode = igrab(inode);
 535
 536                 if (!wait && inode) {
 537                         list_add_tail(&BTRFS_I(inode)->ordered_operations,
 538                               &root->fs_info->ordered_operations);
 539                 }
 540                 spin_unlock(&root->fs_info->ordered_extent_lock);
 541
 542                 if (inode) {
 543                         if (wait)
 544                                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
 545                         else
 546                                 filemap_flush(inode->i_mapping);
 547                         btrfs_add_delayed_iput(inode);
 548                 }
 549
 550                 cond_resched();
 551                 spin_lock(&root->fs_info->ordered_extent_lock);
 552         }
 553         if (wait && !list_empty(&root->fs_info->ordered_operations))
 554                 goto again;
 555
 556         spin_unlock(&root->fs_info->ordered_extent_lock);
 557         mutex_unlock(&root->fs_info->ordered_operations_mutex);
 558
 559         return 0;
 560 }
 561
 562 /*
 563  * Used to start IO or wait for a given ordered extent to finish.
 564  *
 565  * If wait is one, this effectively waits on page writeback for all the pages
 566  * in the extent, and it waits on the io completion code to insert
 567  * metadata into the btree corresponding to the extent
 568  */
 569 void btrfs_start_ordered_extent(struct inode *inode,
 570                                        struct btrfs_ordered_extent *entry,
 571                                        int wait)
 572 {
 573         u64 start = entry->file_offset;
 574         u64 end = start + entry->len - 1;
 575
 576         /*
 577          * pages in the range can be dirty, clean or writeback.  We
 578          * start IO on any dirty ones so the wait doesn't stall waiting
 579          * for pdflush to find them
 580          */
 581         if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
 582                 filemap_fdatawrite_range(inode->i_mapping, start, end);
 583         if (wait) {
 584                 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
 585                                                  &entry->flags));
 586         }
 587 }
 588
 589 /*
 590  * Used to wait on ordered extents across a large range of bytes.
 591  */
 592 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 593 {
 594         u64 end;
 595         u64 orig_end;
 596         struct btrfs_ordered_extent *ordered;
 597         int found;
 598
 599         if (start + len < start) {
 600                 orig_end = INT_LIMIT(loff_t);
 601         } else {
 602                 orig_end = start + len - 1;
 603                 if (orig_end > INT_LIMIT(loff_t))
 604                         orig_end = INT_LIMIT(loff_t);
 605         }
 606 again:
 607         /* start IO across the range first to instantiate any delalloc
 608          * extents
 609          */
 610         filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 611
 612         /* The compression code will leave pages locked but return from
 613          * writepage without setting the page writeback.  Starting again
 614          * with WB_SYNC_ALL will end up waiting for the IO to actually start.
 615          */
 616         filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 617
 618         filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 619
 620         end = orig_end;
 621         found = 0;
 622         while (1) {
 623                 ordered = btrfs_lookup_first_ordered_extent(inode, end);
 624                 if (!ordered)
 625                         break;
 626                 if (ordered->file_offset > orig_end) {
 627                         btrfs_put_ordered_extent(ordered);
 628                         break;
 629                 }
 630                 if (ordered->file_offset + ordered->len < start) {
 631                         btrfs_put_ordered_extent(ordered);
 632                         break;
 633                 }
 634                 found++;
 635                 btrfs_start_ordered_extent(inode, ordered, 1);
 636                 end = ordered->file_offset;
 637                 btrfs_put_ordered_extent(ordered);
 638                 if (end == 0 || end == start)
 639                         break;
 640                 end--;
 641         }
 642         if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
 643                            EXTENT_DELALLOC, 0, NULL)) {
 644                 schedule_timeout(1);
 645                 goto again;
 646         }
 647         return 0;
 648 }
 649
 650 /*
 651  * find an ordered extent corresponding to file_offset.  return NULL if
 652  * nothing is found, otherwise take a reference on the extent and return it
 653  */
 654 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
 655                                                          u64 file_offset)
 656 {
 657         struct btrfs_ordered_inode_tree *tree;
 658         struct rb_node *node;
 659         struct btrfs_ordered_extent *entry = NULL;
 660
 661         tree = &BTRFS_I(inode)->ordered_tree;
 662         spin_lock(&tree->lock);
 663         node = tree_search(tree, file_offset);
 664         if (!node)
 665                 goto out;
 666
 667         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 668         if (!offset_in_entry(entry, file_offset))
 669                 entry = NULL;
 670         if (entry)
 671                 atomic_inc(&entry->refs);
 672 out:
 673         spin_unlock(&tree->lock);
 674         return entry;
 675 }
 676
 677 /* Since the DIO code tries to lock a wide area we need to look for any ordered
 678  * extents that exist in the range, rather than just the start of the range.
 679  */
 680 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
 681                                                         u64 file_offset,
 682                                                         u64 len)
 683 {
 684         struct btrfs_ordered_inode_tree *tree;
 685         struct rb_node *node;
 686         struct btrfs_ordered_extent *entry = NULL;
 687
 688         tree = &BTRFS_I(inode)->ordered_tree;
 689         spin_lock(&tree->lock);
 690         node = tree_search(tree, file_offset);
 691         if (!node) {
 692                 node = tree_search(tree, file_offset + len);
 693                 if (!node)
 694                         goto out;
 695         }
 696
 697         while (1) {
 698                 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 699                 if (range_overlaps(entry, file_offset, len))
 700                         break;
 701
 702                 if (entry->file_offset >= file_offset + len) {
 703                         entry = NULL;
 704                         break;
 705                 }
 706                 entry = NULL;
 707                 node = rb_next(node);
 708                 if (!node)
 709                         break;
 710         }
 711 out:
 712         if (entry)
 713                 atomic_inc(&entry->refs);
 714         spin_unlock(&tree->lock);
 715         return entry;
 716 }
 717
 718 /*
 719  * lookup and return any extent before 'file_offset'.  NULL is returned
 720  * if none is found
 721  */
 722 struct btrfs_ordered_extent *
 723 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
 724 {
 725         struct btrfs_ordered_inode_tree *tree;
 726         struct rb_node *node;
 727         struct btrfs_ordered_extent *entry = NULL;
 728
 729         tree = &BTRFS_I(inode)->ordered_tree;
 730         spin_lock(&tree->lock);
 731         node = tree_search(tree, file_offset);
 732         if (!node)
 733                 goto out;
 734
 735         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 736         atomic_inc(&entry->refs);
 737 out:
 738         spin_unlock(&tree->lock);
 739         return entry;
 740 }
 741
 742 /*
 743  * After an extent is done, call this to conditionally update the on disk
 744  * i_size.  i_size is updated to cover any fully written part of the file.
 745  */
 746 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
 747                                 struct btrfs_ordered_extent *ordered)
 748 {
 749         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 750         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 751         u64 disk_i_size;
 752         u64 new_i_size;
 753         u64 i_size_test;
 754         u64 i_size = i_size_read(inode);
 755         struct rb_node *node;
 756         struct rb_node *prev = NULL;
 757         struct btrfs_ordered_extent *test;
 758         int ret = 1;
 759
 760         if (ordered)
 761                 offset = entry_end(ordered);
 762         else
 763                 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
 764
 765         spin_lock(&tree->lock);
 766         disk_i_size = BTRFS_I(inode)->disk_i_size;
 767
 768         /* truncate file */
 769         if (disk_i_size > i_size) {
 770                 BTRFS_I(inode)->disk_i_size = i_size;
 771                 ret = 0;
 772                 goto out;
 773         }
 774
 775         /*
 776          * if the disk i_size is already at the inode->i_size, or
 777          * this ordered extent is inside the disk i_size, we're done
 778          */
 779         if (disk_i_size == i_size || offset <= disk_i_size) {
 780                 goto out;
 781         }
 782
 783         /*
 784          * we can't update the disk_isize if there are delalloc bytes
 785          * between disk_i_size and  this ordered extent
 786          */
 787         if (test_range_bit(io_tree, disk_i_size, offset - 1,
 788                            EXTENT_DELALLOC, 0, NULL)) {
 789                 goto out;
 790         }
 791         /*
 792          * walk backward from this ordered extent to disk_i_size.
 793          * if we find an ordered extent then we can't update disk i_size
 794          * yet
 795          */
 796         if (ordered) {
 797                 node = rb_prev(&ordered->rb_node);
 798         } else {
 799                 prev = tree_search(tree, offset);
 800                 /*
 801                  * we insert file extents without involving ordered struct,
 802                  * so there should be no ordered struct cover this offset
 803                  */
 804                 if (prev) {
 805                         test = rb_entry(prev, struct btrfs_ordered_extent,
 806                                         rb_node);
 807                         BUG_ON(offset_in_entry(test, offset));
 808                 }
 809                 node = prev;
 810         }
 811         while (node) {
 812                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 813                 if (test->file_offset + test->len <= disk_i_size)
 814                         break;
 815                 if (test->file_offset >= i_size)
 816                         break;
 817                 if (test->file_offset >= disk_i_size)
 818                         goto out;
 819                 node = rb_prev(node);
 820         }
 821         new_i_size = min_t(u64, offset, i_size);
 822
 823         /*
 824          * at this point, we know we can safely update i_size to at least
 825          * the offset from this ordered extent.  But, we need to
 826          * walk forward and see if ios from higher up in the file have
 827          * finished.
 828          */
 829         if (ordered) {
 830                 node = rb_next(&ordered->rb_node);
 831         } else {
 832                 if (prev)
 833                         node = rb_next(prev);
 834                 else
 835                         node = rb_first(&tree->tree);
 836         }
 837         i_size_test = 0;
 838         if (node) {
 839                 /*
 840                  * do we have an area where IO might have finished
 841                  * between our ordered extent and the next one.
 842                  */
 843                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 844                 if (test->file_offset > offset)
 845                         i_size_test = test->file_offset;
 846         } else {
 847                 i_size_test = i_size;
 848         }
 849
 850         /*
 851          * i_size_test is the end of a region after this ordered
 852          * extent where there are no ordered extents.  As long as there
 853          * are no delalloc bytes in this area, it is safe to update
 854          * disk_i_size to the end of the region.
 855          */
 856         if (i_size_test > offset &&
 857             !test_range_bit(io_tree, offset, i_size_test - 1,
 858                             EXTENT_DELALLOC, 0, NULL)) {
 859                 new_i_size = min_t(u64, i_size_test, i_size);
 860         }
 861         BTRFS_I(inode)->disk_i_size = new_i_size;
 862         ret = 0;
 863 out:
 864         /*
 865          * we need to remove the ordered extent with the tree lock held
 866          * so that other people calling this function don't find our fully
 867          * processed ordered entry and skip updating the i_size
 868          */
 869         if (ordered)
 870                 __btrfs_remove_ordered_extent(inode, ordered);
 871         spin_unlock(&tree->lock);
 872         if (ordered)
 873                 wake_up(&ordered->wait);
 874         return ret;
 875 }
 876
 877 /*
 878  * search the ordered extents for one corresponding to 'offset' and
 879  * try to find a checksum.  This is used because we allow pages to
 880  * be reclaimed before their checksum is actually put into the btree
 881  */
 882 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
 883                            u32 *sum)
 884 {
 885         struct btrfs_ordered_sum *ordered_sum;
 886         struct btrfs_sector_sum *sector_sums;
 887         struct btrfs_ordered_extent *ordered;
 888         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 889         unsigned long num_sectors;
 890         unsigned long i;
 891         u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
 892         int ret = 1;
 893
 894         ordered = btrfs_lookup_ordered_extent(inode, offset);
 895         if (!ordered)
 896                 return 1;
 897
 898         spin_lock(&tree->lock);
 899         list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
 900                 if (disk_bytenr >= ordered_sum->bytenr) {
 901                         num_sectors = ordered_sum->len / sectorsize;
 902                         sector_sums = ordered_sum->sums;
 903                         for (i = 0; i < num_sectors; i++) {
 904                                 if (sector_sums[i].bytenr == disk_bytenr) {
 905                                         *sum = sector_sums[i].sum;
 906                                         ret = 0;
 907                                         goto out;
 908                                 }
 909                         }
 910                 }
 911         }
 912 out:
 913         spin_unlock(&tree->lock);
 914         btrfs_put_ordered_extent(ordered);
 915         return ret;
 916 }
 917
 918
 919 /*
 920  * add a given inode to the list of inodes that must be fully on
 921  * disk before a transaction commit finishes.
 922  *
 923  * This basically gives us the ext3 style data=ordered mode, and it is mostly
 924  * used to make sure renamed files are fully on disk.
 925  *
 926  * It is a noop if the inode is already fully on disk.
 927  *
 928  * If trans is not null, we'll do a friendly check for a transaction that
 929  * is already flushing things and force the IO down ourselves.
 930  */
 931 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
 932                                 struct btrfs_root *root,
 933                                 struct inode *inode)
 934 {
 935         u64 last_mod;
 936
 937         last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
 938
 939         /*
 940          * if this file hasn't been changed since the last transaction
 941          * commit, we can safely return without doing anything
 942          */
 943         if (last_mod < root->fs_info->last_trans_committed)
 944                 return 0;
 945
 946         /*
 947          * the transaction is already committing.  Just start the IO and
 948          * don't bother with all of this list nonsense
 949          */
 950         if (trans && root->fs_info->running_transaction->blocked) {
 951                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
 952                 return 0;
 953         }
 954
 955         spin_lock(&root->fs_info->ordered_extent_lock);
 956         if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
 957                 list_add_tail(&BTRFS_I(inode)->ordered_operations,
 958                               &root->fs_info->ordered_operations);
 959         }
 960         spin_unlock(&root->fs_info->ordered_extent_lock);
 961
 962         return 0;
 963 }