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         entry->file_offset = file_offset;
 178         entry->start = start;
 179         entry->len = len;
 180         entry->disk_len = disk_len;
 181         entry->bytes_left = len;
 182         entry->inode = inode;
 183         if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 184                 set_bit(type, &entry->flags);
 185
 186         /* one ref for the tree */
 187         atomic_set(&entry->refs, 1);
 188         init_waitqueue_head(&entry->wait);
 189         INIT_LIST_HEAD(&entry->list);
 190         INIT_LIST_HEAD(&entry->root_extent_list);
 191
 192         spin_lock(&tree->lock);
 193         node = tree_insert(&tree->tree, file_offset,
 194                            &entry->rb_node);
 195         BUG_ON(node);
 196         spin_unlock(&tree->lock);
 197
 198         spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 199         list_add_tail(&entry->root_extent_list,
 200                       &BTRFS_I(inode)->root->fs_info->ordered_extents);
 201         spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 202
 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         spin_lock(&tree->lock);
 220         list_add_tail(&sum->list, &entry->list);
 221         spin_unlock(&tree->lock);
 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                                    struct btrfs_ordered_extent **cached,
 236                                    u64 file_offset, u64 io_size)
 237 {
 238         struct btrfs_ordered_inode_tree *tree;
 239         struct rb_node *node;
 240         struct btrfs_ordered_extent *entry = NULL;
 241         int ret;
 242
 243         tree = &BTRFS_I(inode)->ordered_tree;
 244         spin_lock(&tree->lock);
 245         node = tree_search(tree, file_offset);
 246         if (!node) {
 247                 ret = 1;
 248                 goto out;
 249         }
 250
 251         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 252         if (!offset_in_entry(entry, file_offset)) {
 253                 ret = 1;
 254                 goto out;
 255         }
 256
 257         if (io_size > entry->bytes_left) {
 258                 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
 259                        (unsigned long long)entry->bytes_left,
 260                        (unsigned long long)io_size);
 261         }
 262         entry->bytes_left -= io_size;
 263         if (entry->bytes_left == 0)
 264                 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 265         else
 266                 ret = 1;
 267 out:
 268         if (!ret && cached && entry) {
 269                 *cached = entry;
 270                 atomic_inc(&entry->refs);
 271         }
 272         spin_unlock(&tree->lock);
 273         return ret == 0;
 274 }
 275
 276 /*
 277  * used to drop a reference on an ordered extent.  This will free
 278  * the extent if the last reference is dropped
 279  */
 280 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 281 {
 282         struct list_head *cur;
 283         struct btrfs_ordered_sum *sum;
 284
 285         if (atomic_dec_and_test(&entry->refs)) {
 286                 while (!list_empty(&entry->list)) {
 287                         cur = entry->list.next;
 288                         sum = list_entry(cur, struct btrfs_ordered_sum, list);
 289                         list_del(&sum->list);
 290                         kfree(sum);
 291                 }
 292                 kfree(entry);
 293         }
 294         return 0;
 295 }
 296
 297 /*
 298  * remove an ordered extent from the tree.  No references are dropped
 299  * and you must wake_up entry->wait.  You must hold the tree lock
 300  * while you call this function.
 301  */
 302 static int __btrfs_remove_ordered_extent(struct inode *inode,
 303                                 struct btrfs_ordered_extent *entry)
 304 {
 305         struct btrfs_ordered_inode_tree *tree;
 306         struct rb_node *node;
 307
 308         tree = &BTRFS_I(inode)->ordered_tree;
 309         node = &entry->rb_node;
 310         rb_erase(node, &tree->tree);
 311         tree->last = NULL;
 312         set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 313
 314         spin_lock(&BTRFS_I(inode)->accounting_lock);
 315         BTRFS_I(inode)->outstanding_extents--;
 316         spin_unlock(&BTRFS_I(inode)->accounting_lock);
 317         btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
 318                                               inode, 1);
 319
 320         spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 321         list_del_init(&entry->root_extent_list);
 322
 323         /*
 324          * we have no more ordered extents for this inode and
 325          * no dirty pages.  We can safely remove it from the
 326          * list of ordered extents
 327          */
 328         if (RB_EMPTY_ROOT(&tree->tree) &&
 329             !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 330                 list_del_init(&BTRFS_I(inode)->ordered_operations);
 331         }
 332         spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 333
 334         return 0;
 335 }
 336
 337 /*
 338  * remove an ordered extent from the tree.  No references are dropped
 339  * but any waiters are woken.
 340  */
 341 int btrfs_remove_ordered_extent(struct inode *inode,
 342                                 struct btrfs_ordered_extent *entry)
 343 {
 344         struct btrfs_ordered_inode_tree *tree;
 345         int ret;
 346
 347         tree = &BTRFS_I(inode)->ordered_tree;
 348         spin_lock(&tree->lock);
 349         ret = __btrfs_remove_ordered_extent(inode, entry);
 350         spin_unlock(&tree->lock);
 351         wake_up(&entry->wait);
 352
 353         return ret;
 354 }
 355
 356 /*
 357  * wait for all the ordered extents in a root.  This is done when balancing
 358  * space between drives.
 359  */
 360 int btrfs_wait_ordered_extents(struct btrfs_root *root,
 361                                int nocow_only, int delay_iput)
 362 {
 363         struct list_head splice;
 364         struct list_head *cur;
 365         struct btrfs_ordered_extent *ordered;
 366         struct inode *inode;
 367
 368         INIT_LIST_HEAD(&splice);
 369
 370         spin_lock(&root->fs_info->ordered_extent_lock);
 371         list_splice_init(&root->fs_info->ordered_extents, &splice);
 372         while (!list_empty(&splice)) {
 373                 cur = splice.next;
 374                 ordered = list_entry(cur, struct btrfs_ordered_extent,
 375                                      root_extent_list);
 376                 if (nocow_only &&
 377                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
 378                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
 379                         list_move(&ordered->root_extent_list,
 380                                   &root->fs_info->ordered_extents);
 381                         cond_resched_lock(&root->fs_info->ordered_extent_lock);
 382                         continue;
 383                 }
 384
 385                 list_del_init(&ordered->root_extent_list);
 386                 atomic_inc(&ordered->refs);
 387
 388                 /*
 389                  * the inode may be getting freed (in sys_unlink path).
 390                  */
 391                 inode = igrab(ordered->inode);
 392
 393                 spin_unlock(&root->fs_info->ordered_extent_lock);
 394
 395                 if (inode) {
 396                         btrfs_start_ordered_extent(inode, ordered, 1);
 397                         btrfs_put_ordered_extent(ordered);
 398                         if (delay_iput)
 399                                 btrfs_add_delayed_iput(inode);
 400                         else
 401                                 iput(inode);
 402                 } else {
 403                         btrfs_put_ordered_extent(ordered);
 404                 }
 405
 406                 spin_lock(&root->fs_info->ordered_extent_lock);
 407         }
 408         spin_unlock(&root->fs_info->ordered_extent_lock);
 409         return 0;
 410 }
 411
 412 /*
 413  * this is used during transaction commit to write all the inodes
 414  * added to the ordered operation list.  These files must be fully on
 415  * disk before the transaction commits.
 416  *
 417  * we have two modes here, one is to just start the IO via filemap_flush
 418  * and the other is to wait for all the io.  When we wait, we have an
 419  * extra check to make sure the ordered operation list really is empty
 420  * before we return
 421  */
 422 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
 423 {
 424         struct btrfs_inode *btrfs_inode;
 425         struct inode *inode;
 426         struct list_head splice;
 427
 428         INIT_LIST_HEAD(&splice);
 429
 430         mutex_lock(&root->fs_info->ordered_operations_mutex);
 431         spin_lock(&root->fs_info->ordered_extent_lock);
 432 again:
 433         list_splice_init(&root->fs_info->ordered_operations, &splice);
 434
 435         while (!list_empty(&splice)) {
 436                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
 437                                    ordered_operations);
 438
 439                 inode = &btrfs_inode->vfs_inode;
 440
 441                 list_del_init(&btrfs_inode->ordered_operations);
 442
 443                 /*
 444                  * the inode may be getting freed (in sys_unlink path).
 445                  */
 446                 inode = igrab(inode);
 447
 448                 if (!wait && inode) {
 449                         list_add_tail(&BTRFS_I(inode)->ordered_operations,
 450                               &root->fs_info->ordered_operations);
 451                 }
 452                 spin_unlock(&root->fs_info->ordered_extent_lock);
 453
 454                 if (inode) {
 455                         if (wait)
 456                                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
 457                         else
 458                                 filemap_flush(inode->i_mapping);
 459                         btrfs_add_delayed_iput(inode);
 460                 }
 461
 462                 cond_resched();
 463                 spin_lock(&root->fs_info->ordered_extent_lock);
 464         }
 465         if (wait && !list_empty(&root->fs_info->ordered_operations))
 466                 goto again;
 467
 468         spin_unlock(&root->fs_info->ordered_extent_lock);
 469         mutex_unlock(&root->fs_info->ordered_operations_mutex);
 470
 471         return 0;
 472 }
 473
 474 /*
 475  * Used to start IO or wait for a given ordered extent to finish.
 476  *
 477  * If wait is one, this effectively waits on page writeback for all the pages
 478  * in the extent, and it waits on the io completion code to insert
 479  * metadata into the btree corresponding to the extent
 480  */
 481 void btrfs_start_ordered_extent(struct inode *inode,
 482                                        struct btrfs_ordered_extent *entry,
 483                                        int wait)
 484 {
 485         u64 start = entry->file_offset;
 486         u64 end = start + entry->len - 1;
 487
 488         /*
 489          * pages in the range can be dirty, clean or writeback.  We
 490          * start IO on any dirty ones so the wait doesn't stall waiting
 491          * for pdflush to find them
 492          */
 493         filemap_fdatawrite_range(inode->i_mapping, start, end);
 494         if (wait) {
 495                 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
 496                                                  &entry->flags));
 497         }
 498 }
 499
 500 /*
 501  * Used to wait on ordered extents across a large range of bytes.
 502  */
 503 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 504 {
 505         u64 end;
 506         u64 orig_end;
 507         u64 wait_end;
 508         struct btrfs_ordered_extent *ordered;
 509         int found;
 510
 511         if (start + len < start) {
 512                 orig_end = INT_LIMIT(loff_t);
 513         } else {
 514                 orig_end = start + len - 1;
 515                 if (orig_end > INT_LIMIT(loff_t))
 516                         orig_end = INT_LIMIT(loff_t);
 517         }
 518         wait_end = orig_end;
 519 again:
 520         /* start IO across the range first to instantiate any delalloc
 521          * extents
 522          */
 523         filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 524
 525         /* The compression code will leave pages locked but return from
 526          * writepage without setting the page writeback.  Starting again
 527          * with WB_SYNC_ALL will end up waiting for the IO to actually start.
 528          */
 529         filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 530
 531         filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 532
 533         end = orig_end;
 534         found = 0;
 535         while (1) {
 536                 ordered = btrfs_lookup_first_ordered_extent(inode, end);
 537                 if (!ordered)
 538                         break;
 539                 if (ordered->file_offset > orig_end) {
 540                         btrfs_put_ordered_extent(ordered);
 541                         break;
 542                 }
 543                 if (ordered->file_offset + ordered->len < start) {
 544                         btrfs_put_ordered_extent(ordered);
 545                         break;
 546                 }
 547                 found++;
 548                 btrfs_start_ordered_extent(inode, ordered, 1);
 549                 end = ordered->file_offset;
 550                 btrfs_put_ordered_extent(ordered);
 551                 if (end == 0 || end == start)
 552                         break;
 553                 end--;
 554         }
 555         if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
 556                            EXTENT_DELALLOC, 0, NULL)) {
 557                 schedule_timeout(1);
 558                 goto again;
 559         }
 560         return 0;
 561 }
 562
 563 /*
 564  * find an ordered extent corresponding to file_offset.  return NULL if
 565  * nothing is found, otherwise take a reference on the extent and return it
 566  */
 567 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
 568                                                          u64 file_offset)
 569 {
 570         struct btrfs_ordered_inode_tree *tree;
 571         struct rb_node *node;
 572         struct btrfs_ordered_extent *entry = NULL;
 573
 574         tree = &BTRFS_I(inode)->ordered_tree;
 575         spin_lock(&tree->lock);
 576         node = tree_search(tree, file_offset);
 577         if (!node)
 578                 goto out;
 579
 580         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 581         if (!offset_in_entry(entry, file_offset))
 582                 entry = NULL;
 583         if (entry)
 584                 atomic_inc(&entry->refs);
 585 out:
 586         spin_unlock(&tree->lock);
 587         return entry;
 588 }
 589
 590 /*
 591  * lookup and return any extent before 'file_offset'.  NULL is returned
 592  * if none is found
 593  */
 594 struct btrfs_ordered_extent *
 595 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
 596 {
 597         struct btrfs_ordered_inode_tree *tree;
 598         struct rb_node *node;
 599         struct btrfs_ordered_extent *entry = NULL;
 600
 601         tree = &BTRFS_I(inode)->ordered_tree;
 602         spin_lock(&tree->lock);
 603         node = tree_search(tree, file_offset);
 604         if (!node)
 605                 goto out;
 606
 607         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 608         atomic_inc(&entry->refs);
 609 out:
 610         spin_unlock(&tree->lock);
 611         return entry;
 612 }
 613
 614 /*
 615  * After an extent is done, call this to conditionally update the on disk
 616  * i_size.  i_size is updated to cover any fully written part of the file.
 617  */
 618 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
 619                                 struct btrfs_ordered_extent *ordered)
 620 {
 621         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 622         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 623         u64 disk_i_size;
 624         u64 new_i_size;
 625         u64 i_size_test;
 626         u64 i_size = i_size_read(inode);
 627         struct rb_node *node;
 628         struct rb_node *prev = NULL;
 629         struct btrfs_ordered_extent *test;
 630         int ret = 1;
 631
 632         if (ordered)
 633                 offset = entry_end(ordered);
 634         else
 635                 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
 636
 637         spin_lock(&tree->lock);
 638         disk_i_size = BTRFS_I(inode)->disk_i_size;
 639
 640         /* truncate file */
 641         if (disk_i_size > i_size) {
 642                 BTRFS_I(inode)->disk_i_size = i_size;
 643                 ret = 0;
 644                 goto out;
 645         }
 646
 647         /*
 648          * if the disk i_size is already at the inode->i_size, or
 649          * this ordered extent is inside the disk i_size, we're done
 650          */
 651         if (disk_i_size == i_size || offset <= disk_i_size) {
 652                 goto out;
 653         }
 654
 655         /*
 656          * we can't update the disk_isize if there are delalloc bytes
 657          * between disk_i_size and  this ordered extent
 658          */
 659         if (test_range_bit(io_tree, disk_i_size, offset - 1,
 660                            EXTENT_DELALLOC, 0, NULL)) {
 661                 goto out;
 662         }
 663         /*
 664          * walk backward from this ordered extent to disk_i_size.
 665          * if we find an ordered extent then we can't update disk i_size
 666          * yet
 667          */
 668         if (ordered) {
 669                 node = rb_prev(&ordered->rb_node);
 670         } else {
 671                 prev = tree_search(tree, offset);
 672                 /*
 673                  * we insert file extents without involving ordered struct,
 674                  * so there should be no ordered struct cover this offset
 675                  */
 676                 if (prev) {
 677                         test = rb_entry(prev, struct btrfs_ordered_extent,
 678                                         rb_node);
 679                         BUG_ON(offset_in_entry(test, offset));
 680                 }
 681                 node = prev;
 682         }
 683         while (node) {
 684                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 685                 if (test->file_offset + test->len <= disk_i_size)
 686                         break;
 687                 if (test->file_offset >= i_size)
 688                         break;
 689                 if (test->file_offset >= disk_i_size)
 690                         goto out;
 691                 node = rb_prev(node);
 692         }
 693         new_i_size = min_t(u64, offset, i_size);
 694
 695         /*
 696          * at this point, we know we can safely update i_size to at least
 697          * the offset from this ordered extent.  But, we need to
 698          * walk forward and see if ios from higher up in the file have
 699          * finished.
 700          */
 701         if (ordered) {
 702                 node = rb_next(&ordered->rb_node);
 703         } else {
 704                 if (prev)
 705                         node = rb_next(prev);
 706                 else
 707                         node = rb_first(&tree->tree);
 708         }
 709         i_size_test = 0;
 710         if (node) {
 711                 /*
 712                  * do we have an area where IO might have finished
 713                  * between our ordered extent and the next one.
 714                  */
 715                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 716                 if (test->file_offset > offset)
 717                         i_size_test = test->file_offset;
 718         } else {
 719                 i_size_test = i_size;
 720         }
 721
 722         /*
 723          * i_size_test is the end of a region after this ordered
 724          * extent where there are no ordered extents.  As long as there
 725          * are no delalloc bytes in this area, it is safe to update
 726          * disk_i_size to the end of the region.
 727          */
 728         if (i_size_test > offset &&
 729             !test_range_bit(io_tree, offset, i_size_test - 1,
 730                             EXTENT_DELALLOC, 0, NULL)) {
 731                 new_i_size = min_t(u64, i_size_test, i_size);
 732         }
 733         BTRFS_I(inode)->disk_i_size = new_i_size;
 734         ret = 0;
 735 out:
 736         /*
 737          * we need to remove the ordered extent with the tree lock held
 738          * so that other people calling this function don't find our fully
 739          * processed ordered entry and skip updating the i_size
 740          */
 741         if (ordered)
 742                 __btrfs_remove_ordered_extent(inode, ordered);
 743         spin_unlock(&tree->lock);
 744         if (ordered)
 745                 wake_up(&ordered->wait);
 746         return ret;
 747 }
 748
 749 /*
 750  * search the ordered extents for one corresponding to 'offset' and
 751  * try to find a checksum.  This is used because we allow pages to
 752  * be reclaimed before their checksum is actually put into the btree
 753  */
 754 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
 755                            u32 *sum)
 756 {
 757         struct btrfs_ordered_sum *ordered_sum;
 758         struct btrfs_sector_sum *sector_sums;
 759         struct btrfs_ordered_extent *ordered;
 760         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 761         unsigned long num_sectors;
 762         unsigned long i;
 763         u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
 764         int ret = 1;
 765
 766         ordered = btrfs_lookup_ordered_extent(inode, offset);
 767         if (!ordered)
 768                 return 1;
 769
 770         spin_lock(&tree->lock);
 771         list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
 772                 if (disk_bytenr >= ordered_sum->bytenr) {
 773                         num_sectors = ordered_sum->len / sectorsize;
 774                         sector_sums = ordered_sum->sums;
 775                         for (i = 0; i < num_sectors; i++) {
 776                                 if (sector_sums[i].bytenr == disk_bytenr) {
 777                                         *sum = sector_sums[i].sum;
 778                                         ret = 0;
 779                                         goto out;
 780                                 }
 781                         }
 782                 }
 783         }
 784 out:
 785         spin_unlock(&tree->lock);
 786         btrfs_put_ordered_extent(ordered);
 787         return ret;
 788 }
 789
 790
 791 /*
 792  * add a given inode to the list of inodes that must be fully on
 793  * disk before a transaction commit finishes.
 794  *
 795  * This basically gives us the ext3 style data=ordered mode, and it is mostly
 796  * used to make sure renamed files are fully on disk.
 797  *
 798  * It is a noop if the inode is already fully on disk.
 799  *
 800  * If trans is not null, we'll do a friendly check for a transaction that
 801  * is already flushing things and force the IO down ourselves.
 802  */
 803 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
 804                                 struct btrfs_root *root,
 805                                 struct inode *inode)
 806 {
 807         u64 last_mod;
 808
 809         last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
 810
 811         /*
 812          * if this file hasn't been changed since the last transaction
 813          * commit, we can safely return without doing anything
 814          */
 815         if (last_mod < root->fs_info->last_trans_committed)
 816                 return 0;
 817
 818         /*
 819          * the transaction is already committing.  Just start the IO and
 820          * don't bother with all of this list nonsense
 821          */
 822         if (trans && root->fs_info->running_transaction->blocked) {
 823                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
 824                 return 0;
 825         }
 826
 827         spin_lock(&root->fs_info->ordered_extent_lock);
 828         if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
 829                 list_add_tail(&BTRFS_I(inode)->ordered_operations,
 830                               &root->fs_info->ordered_operations);
 831         }
 832         spin_unlock(&root->fs_info->ordered_extent_lock);
 833
 834         return 0;
 835 }