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 = NULL;
 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, int compress_type)
 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         entry->compress_type = compress_type;
 193         if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 194                 set_bit(type, &entry->flags);
 195
 196         if (dio)
 197                 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
 198
 199         /* one ref for the tree */
 200         atomic_set(&entry->refs, 1);
 201         init_waitqueue_head(&entry->wait);
 202         INIT_LIST_HEAD(&entry->list);
 203         INIT_LIST_HEAD(&entry->root_extent_list);
 204
 205         spin_lock(&tree->lock);
 206         node = tree_insert(&tree->tree, file_offset,
 207                            &entry->rb_node);
 208         BUG_ON(node);
 209         spin_unlock(&tree->lock);
 210
 211         spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 212         list_add_tail(&entry->root_extent_list,
 213                       &BTRFS_I(inode)->root->fs_info->ordered_extents);
 214         spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 215
 216         BUG_ON(node);
 217         return 0;
 218 }
 219
 220 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 221                              u64 start, u64 len, u64 disk_len, int type)
 222 {
 223         return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 224                                           disk_len, type, 0,
 225                                           BTRFS_COMPRESS_NONE);
 226 }
 227
 228 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
 229                                  u64 start, u64 len, u64 disk_len, int type)
 230 {
 231         return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 232                                           disk_len, type, 1,
 233                                           BTRFS_COMPRESS_NONE);
 234 }
 235
 236 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
 237                                       u64 start, u64 len, u64 disk_len,
 238                                       int type, int compress_type)
 239 {
 240         return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 241                                           disk_len, type, 0,
 242                                           compress_type);
 243 }
 244
 245 /*
 246  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
 247  * when an ordered extent is finished.  If the list covers more than one
 248  * ordered extent, it is split across multiples.
 249  */
 250 int btrfs_add_ordered_sum(struct inode *inode,
 251                           struct btrfs_ordered_extent *entry,
 252                           struct btrfs_ordered_sum *sum)
 253 {
 254         struct btrfs_ordered_inode_tree *tree;
 255
 256         tree = &BTRFS_I(inode)->ordered_tree;
 257         spin_lock(&tree->lock);
 258         list_add_tail(&sum->list, &entry->list);
 259         spin_unlock(&tree->lock);
 260         return 0;
 261 }
 262
 263 /*
 264  * this is used to account for finished IO across a given range
 265  * of the file.  The IO may span ordered extents.  If
 266  * a given ordered_extent is completely done, 1 is returned, otherwise
 267  * 0.
 268  *
 269  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 270  * to make sure this function only returns 1 once for a given ordered extent.
 271  *
 272  * file_offset is updated to one byte past the range that is recorded as
 273  * complete.  This allows you to walk forward in the file.
 274  */
 275 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
 276                                    struct btrfs_ordered_extent **cached,
 277                                    u64 *file_offset, u64 io_size)
 278 {
 279         struct btrfs_ordered_inode_tree *tree;
 280         struct rb_node *node;
 281         struct btrfs_ordered_extent *entry = NULL;
 282         int ret;
 283         u64 dec_end;
 284         u64 dec_start;
 285         u64 to_dec;
 286
 287         tree = &BTRFS_I(inode)->ordered_tree;
 288         spin_lock(&tree->lock);
 289         node = tree_search(tree, *file_offset);
 290         if (!node) {
 291                 ret = 1;
 292                 goto out;
 293         }
 294
 295         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 296         if (!offset_in_entry(entry, *file_offset)) {
 297                 ret = 1;
 298                 goto out;
 299         }
 300
 301         dec_start = max(*file_offset, entry->file_offset);
 302         dec_end = min(*file_offset + io_size, entry->file_offset +
 303                       entry->len);
 304         *file_offset = dec_end;
 305         if (dec_start > dec_end) {
 306                 printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n",
 307                        (unsigned long long)dec_start,
 308                        (unsigned long long)dec_end);
 309         }
 310         to_dec = dec_end - dec_start;
 311         if (to_dec > entry->bytes_left) {
 312                 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
 313                        (unsigned long long)entry->bytes_left,
 314                        (unsigned long long)to_dec);
 315         }
 316         entry->bytes_left -= to_dec;
 317         if (entry->bytes_left == 0)
 318                 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 319         else
 320                 ret = 1;
 321 out:
 322         if (!ret && cached && entry) {
 323                 *cached = entry;
 324                 atomic_inc(&entry->refs);
 325         }
 326         spin_unlock(&tree->lock);
 327         return ret == 0;
 328 }
 329
 330 /*
 331  * this is used to account for finished IO across a given range
 332  * of the file.  The IO should not span ordered extents.  If
 333  * a given ordered_extent is completely done, 1 is returned, otherwise
 334  * 0.
 335  *
 336  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 337  * to make sure this function only returns 1 once for a given ordered extent.
 338  */
 339 int btrfs_dec_test_ordered_pending(struct inode *inode,
 340                                    struct btrfs_ordered_extent **cached,
 341                                    u64 file_offset, u64 io_size)
 342 {
 343         struct btrfs_ordered_inode_tree *tree;
 344         struct rb_node *node;
 345         struct btrfs_ordered_extent *entry = NULL;
 346         int ret;
 347
 348         tree = &BTRFS_I(inode)->ordered_tree;
 349         spin_lock(&tree->lock);
 350         node = tree_search(tree, file_offset);
 351         if (!node) {
 352                 ret = 1;
 353                 goto out;
 354         }
 355
 356         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 357         if (!offset_in_entry(entry, file_offset)) {
 358                 ret = 1;
 359                 goto out;
 360         }
 361
 362         if (io_size > entry->bytes_left) {
 363                 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
 364                        (unsigned long long)entry->bytes_left,
 365                        (unsigned long long)io_size);
 366         }
 367         entry->bytes_left -= io_size;
 368         if (entry->bytes_left == 0)
 369                 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 370         else
 371                 ret = 1;
 372 out:
 373         if (!ret && cached && entry) {
 374                 *cached = entry;
 375                 atomic_inc(&entry->refs);
 376         }
 377         spin_unlock(&tree->lock);
 378         return ret == 0;
 379 }
 380
 381 /*
 382  * used to drop a reference on an ordered extent.  This will free
 383  * the extent if the last reference is dropped
 384  */
 385 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 386 {
 387         struct list_head *cur;
 388         struct btrfs_ordered_sum *sum;
 389
 390         if (atomic_dec_and_test(&entry->refs)) {
 391                 while (!list_empty(&entry->list)) {
 392                         cur = entry->list.next;
 393                         sum = list_entry(cur, struct btrfs_ordered_sum, list);
 394                         list_del(&sum->list);
 395                         kfree(sum);
 396                 }
 397                 kfree(entry);
 398         }
 399         return 0;
 400 }
 401
 402 /*
 403  * remove an ordered extent from the tree.  No references are dropped
 404  * and you must wake_up entry->wait.  You must hold the tree lock
 405  * while you call this function.
 406  */
 407 static int __btrfs_remove_ordered_extent(struct inode *inode,
 408                                 struct btrfs_ordered_extent *entry)
 409 {
 410         struct btrfs_ordered_inode_tree *tree;
 411         struct btrfs_root *root = BTRFS_I(inode)->root;
 412         struct rb_node *node;
 413
 414         tree = &BTRFS_I(inode)->ordered_tree;
 415         node = &entry->rb_node;
 416         rb_erase(node, &tree->tree);
 417         tree->last = NULL;
 418         set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 419
 420         spin_lock(&root->fs_info->ordered_extent_lock);
 421         list_del_init(&entry->root_extent_list);
 422
 423         /*
 424          * we have no more ordered extents for this inode and
 425          * no dirty pages.  We can safely remove it from the
 426          * list of ordered extents
 427          */
 428         if (RB_EMPTY_ROOT(&tree->tree) &&
 429             !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 430                 list_del_init(&BTRFS_I(inode)->ordered_operations);
 431         }
 432         spin_unlock(&root->fs_info->ordered_extent_lock);
 433
 434         return 0;
 435 }
 436
 437 /*
 438  * remove an ordered extent from the tree.  No references are dropped
 439  * but any waiters are woken.
 440  */
 441 int btrfs_remove_ordered_extent(struct inode *inode,
 442                                 struct btrfs_ordered_extent *entry)
 443 {
 444         struct btrfs_ordered_inode_tree *tree;
 445         int ret;
 446
 447         tree = &BTRFS_I(inode)->ordered_tree;
 448         spin_lock(&tree->lock);
 449         ret = __btrfs_remove_ordered_extent(inode, entry);
 450         spin_unlock(&tree->lock);
 451         wake_up(&entry->wait);
 452
 453         return ret;
 454 }
 455
 456 /*
 457  * wait for all the ordered extents in a root.  This is done when balancing
 458  * space between drives.
 459  */
 460 int btrfs_wait_ordered_extents(struct btrfs_root *root,
 461                                int nocow_only, int delay_iput)
 462 {
 463         struct list_head splice;
 464         struct list_head *cur;
 465         struct btrfs_ordered_extent *ordered;
 466         struct inode *inode;
 467
 468         INIT_LIST_HEAD(&splice);
 469
 470         spin_lock(&root->fs_info->ordered_extent_lock);
 471         list_splice_init(&root->fs_info->ordered_extents, &splice);
 472         while (!list_empty(&splice)) {
 473                 cur = splice.next;
 474                 ordered = list_entry(cur, struct btrfs_ordered_extent,
 475                                      root_extent_list);
 476                 if (nocow_only &&
 477                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
 478                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
 479                         list_move(&ordered->root_extent_list,
 480                                   &root->fs_info->ordered_extents);
 481                         cond_resched_lock(&root->fs_info->ordered_extent_lock);
 482                         continue;
 483                 }
 484
 485                 list_del_init(&ordered->root_extent_list);
 486                 atomic_inc(&ordered->refs);
 487
 488                 /*
 489                  * the inode may be getting freed (in sys_unlink path).
 490                  */
 491                 inode = igrab(ordered->inode);
 492
 493                 spin_unlock(&root->fs_info->ordered_extent_lock);
 494
 495                 if (inode) {
 496                         btrfs_start_ordered_extent(inode, ordered, 1);
 497                         btrfs_put_ordered_extent(ordered);
 498                         if (delay_iput)
 499                                 btrfs_add_delayed_iput(inode);
 500                         else
 501                                 iput(inode);
 502                 } else {
 503                         btrfs_put_ordered_extent(ordered);
 504                 }
 505
 506                 spin_lock(&root->fs_info->ordered_extent_lock);
 507         }
 508         spin_unlock(&root->fs_info->ordered_extent_lock);
 509         return 0;
 510 }
 511
 512 /*
 513  * this is used during transaction commit to write all the inodes
 514  * added to the ordered operation list.  These files must be fully on
 515  * disk before the transaction commits.
 516  *
 517  * we have two modes here, one is to just start the IO via filemap_flush
 518  * and the other is to wait for all the io.  When we wait, we have an
 519  * extra check to make sure the ordered operation list really is empty
 520  * before we return
 521  */
 522 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
 523 {
 524         struct btrfs_inode *btrfs_inode;
 525         struct inode *inode;
 526         struct list_head splice;
 527
 528         INIT_LIST_HEAD(&splice);
 529
 530         mutex_lock(&root->fs_info->ordered_operations_mutex);
 531         spin_lock(&root->fs_info->ordered_extent_lock);
 532 again:
 533         list_splice_init(&root->fs_info->ordered_operations, &splice);
 534
 535         while (!list_empty(&splice)) {
 536                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
 537                                    ordered_operations);
 538
 539                 inode = &btrfs_inode->vfs_inode;
 540
 541                 list_del_init(&btrfs_inode->ordered_operations);
 542
 543                 /*
 544                  * the inode may be getting freed (in sys_unlink path).
 545                  */
 546                 inode = igrab(inode);
 547
 548                 if (!wait && inode) {
 549                         list_add_tail(&BTRFS_I(inode)->ordered_operations,
 550                               &root->fs_info->ordered_operations);
 551                 }
 552                 spin_unlock(&root->fs_info->ordered_extent_lock);
 553
 554                 if (inode) {
 555                         if (wait)
 556                                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
 557                         else
 558                                 filemap_flush(inode->i_mapping);
 559                         btrfs_add_delayed_iput(inode);
 560                 }
 561
 562                 cond_resched();
 563                 spin_lock(&root->fs_info->ordered_extent_lock);
 564         }
 565         if (wait && !list_empty(&root->fs_info->ordered_operations))
 566                 goto again;
 567
 568         spin_unlock(&root->fs_info->ordered_extent_lock);
 569         mutex_unlock(&root->fs_info->ordered_operations_mutex);
 570
 571         return 0;
 572 }
 573
 574 /*
 575  * Used to start IO or wait for a given ordered extent to finish.
 576  *
 577  * If wait is one, this effectively waits on page writeback for all the pages
 578  * in the extent, and it waits on the io completion code to insert
 579  * metadata into the btree corresponding to the extent
 580  */
 581 void btrfs_start_ordered_extent(struct inode *inode,
 582                                        struct btrfs_ordered_extent *entry,
 583                                        int wait)
 584 {
 585         u64 start = entry->file_offset;
 586         u64 end = start + entry->len - 1;
 587
 588         /*
 589          * pages in the range can be dirty, clean or writeback.  We
 590          * start IO on any dirty ones so the wait doesn't stall waiting
 591          * for pdflush to find them
 592          */
 593         if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
 594                 filemap_fdatawrite_range(inode->i_mapping, start, end);
 595         if (wait) {
 596                 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
 597                                                  &entry->flags));
 598         }
 599 }
 600
 601 /*
 602  * Used to wait on ordered extents across a large range of bytes.
 603  */
 604 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 605 {
 606         u64 end;
 607         u64 orig_end;
 608         struct btrfs_ordered_extent *ordered;
 609         int found;
 610
 611         if (start + len < start) {
 612                 orig_end = INT_LIMIT(loff_t);
 613         } else {
 614                 orig_end = start + len - 1;
 615                 if (orig_end > INT_LIMIT(loff_t))
 616                         orig_end = INT_LIMIT(loff_t);
 617         }
 618 again:
 619         /* start IO across the range first to instantiate any delalloc
 620          * extents
 621          */
 622         filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 623
 624         /* The compression code will leave pages locked but return from
 625          * writepage without setting the page writeback.  Starting again
 626          * with WB_SYNC_ALL will end up waiting for the IO to actually start.
 627          */
 628         filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 629
 630         filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 631
 632         end = orig_end;
 633         found = 0;
 634         while (1) {
 635                 ordered = btrfs_lookup_first_ordered_extent(inode, end);
 636                 if (!ordered)
 637                         break;
 638                 if (ordered->file_offset > orig_end) {
 639                         btrfs_put_ordered_extent(ordered);
 640                         break;
 641                 }
 642                 if (ordered->file_offset + ordered->len < start) {
 643                         btrfs_put_ordered_extent(ordered);
 644                         break;
 645                 }
 646                 found++;
 647                 btrfs_start_ordered_extent(inode, ordered, 1);
 648                 end = ordered->file_offset;
 649                 btrfs_put_ordered_extent(ordered);
 650                 if (end == 0 || end == start)
 651                         break;
 652                 end--;
 653         }
 654         if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
 655                            EXTENT_DELALLOC, 0, NULL)) {
 656                 schedule_timeout(1);
 657                 goto again;
 658         }
 659         return 0;
 660 }
 661
 662 /*
 663  * find an ordered extent corresponding to file_offset.  return NULL if
 664  * nothing is found, otherwise take a reference on the extent and return it
 665  */
 666 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
 667                                                          u64 file_offset)
 668 {
 669         struct btrfs_ordered_inode_tree *tree;
 670         struct rb_node *node;
 671         struct btrfs_ordered_extent *entry = NULL;
 672
 673         tree = &BTRFS_I(inode)->ordered_tree;
 674         spin_lock(&tree->lock);
 675         node = tree_search(tree, file_offset);
 676         if (!node)
 677                 goto out;
 678
 679         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 680         if (!offset_in_entry(entry, file_offset))
 681                 entry = NULL;
 682         if (entry)
 683                 atomic_inc(&entry->refs);
 684 out:
 685         spin_unlock(&tree->lock);
 686         return entry;
 687 }
 688
 689 /* Since the DIO code tries to lock a wide area we need to look for any ordered
 690  * extents that exist in the range, rather than just the start of the range.
 691  */
 692 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
 693                                                         u64 file_offset,
 694                                                         u64 len)
 695 {
 696         struct btrfs_ordered_inode_tree *tree;
 697         struct rb_node *node;
 698         struct btrfs_ordered_extent *entry = NULL;
 699
 700         tree = &BTRFS_I(inode)->ordered_tree;
 701         spin_lock(&tree->lock);
 702         node = tree_search(tree, file_offset);
 703         if (!node) {
 704                 node = tree_search(tree, file_offset + len);
 705                 if (!node)
 706                         goto out;
 707         }
 708
 709         while (1) {
 710                 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 711                 if (range_overlaps(entry, file_offset, len))
 712                         break;
 713
 714                 if (entry->file_offset >= file_offset + len) {
 715                         entry = NULL;
 716                         break;
 717                 }
 718                 entry = NULL;
 719                 node = rb_next(node);
 720                 if (!node)
 721                         break;
 722         }
 723 out:
 724         if (entry)
 725                 atomic_inc(&entry->refs);
 726         spin_unlock(&tree->lock);
 727         return entry;
 728 }
 729
 730 /*
 731  * lookup and return any extent before 'file_offset'.  NULL is returned
 732  * if none is found
 733  */
 734 struct btrfs_ordered_extent *
 735 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
 736 {
 737         struct btrfs_ordered_inode_tree *tree;
 738         struct rb_node *node;
 739         struct btrfs_ordered_extent *entry = NULL;
 740
 741         tree = &BTRFS_I(inode)->ordered_tree;
 742         spin_lock(&tree->lock);
 743         node = tree_search(tree, file_offset);
 744         if (!node)
 745                 goto out;
 746
 747         entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 748         atomic_inc(&entry->refs);
 749 out:
 750         spin_unlock(&tree->lock);
 751         return entry;
 752 }
 753
 754 /*
 755  * After an extent is done, call this to conditionally update the on disk
 756  * i_size.  i_size is updated to cover any fully written part of the file.
 757  */
 758 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
 759                                 struct btrfs_ordered_extent *ordered)
 760 {
 761         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 762         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 763         u64 disk_i_size;
 764         u64 new_i_size;
 765         u64 i_size_test;
 766         u64 i_size = i_size_read(inode);
 767         struct rb_node *node;
 768         struct rb_node *prev = NULL;
 769         struct btrfs_ordered_extent *test;
 770         int ret = 1;
 771
 772         if (ordered)
 773                 offset = entry_end(ordered);
 774         else
 775                 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
 776
 777         spin_lock(&tree->lock);
 778         disk_i_size = BTRFS_I(inode)->disk_i_size;
 779
 780         /* truncate file */
 781         if (disk_i_size > i_size) {
 782                 BTRFS_I(inode)->disk_i_size = i_size;
 783                 ret = 0;
 784                 goto out;
 785         }
 786
 787         /*
 788          * if the disk i_size is already at the inode->i_size, or
 789          * this ordered extent is inside the disk i_size, we're done
 790          */
 791         if (disk_i_size == i_size || offset <= disk_i_size) {
 792                 goto out;
 793         }
 794
 795         /*
 796          * we can't update the disk_isize if there are delalloc bytes
 797          * between disk_i_size and  this ordered extent
 798          */
 799         if (test_range_bit(io_tree, disk_i_size, offset - 1,
 800                            EXTENT_DELALLOC, 0, NULL)) {
 801                 goto out;
 802         }
 803         /*
 804          * walk backward from this ordered extent to disk_i_size.
 805          * if we find an ordered extent then we can't update disk i_size
 806          * yet
 807          */
 808         if (ordered) {
 809                 node = rb_prev(&ordered->rb_node);
 810         } else {
 811                 prev = tree_search(tree, offset);
 812                 /*
 813                  * we insert file extents without involving ordered struct,
 814                  * so there should be no ordered struct cover this offset
 815                  */
 816                 if (prev) {
 817                         test = rb_entry(prev, struct btrfs_ordered_extent,
 818                                         rb_node);
 819                         BUG_ON(offset_in_entry(test, offset));
 820                 }
 821                 node = prev;
 822         }
 823         while (node) {
 824                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 825                 if (test->file_offset + test->len <= disk_i_size)
 826                         break;
 827                 if (test->file_offset >= i_size)
 828                         break;
 829                 if (test->file_offset >= disk_i_size)
 830                         goto out;
 831                 node = rb_prev(node);
 832         }
 833         new_i_size = min_t(u64, offset, i_size);
 834
 835         /*
 836          * at this point, we know we can safely update i_size to at least
 837          * the offset from this ordered extent.  But, we need to
 838          * walk forward and see if ios from higher up in the file have
 839          * finished.
 840          */
 841         if (ordered) {
 842                 node = rb_next(&ordered->rb_node);
 843         } else {
 844                 if (prev)
 845                         node = rb_next(prev);
 846                 else
 847                         node = rb_first(&tree->tree);
 848         }
 849         i_size_test = 0;
 850         if (node) {
 851                 /*
 852                  * do we have an area where IO might have finished
 853                  * between our ordered extent and the next one.
 854                  */
 855                 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 856                 if (test->file_offset > offset)
 857                         i_size_test = test->file_offset;
 858         } else {
 859                 i_size_test = i_size;
 860         }
 861
 862         /*
 863          * i_size_test is the end of a region after this ordered
 864          * extent where there are no ordered extents.  As long as there
 865          * are no delalloc bytes in this area, it is safe to update
 866          * disk_i_size to the end of the region.
 867          */
 868         if (i_size_test > offset &&
 869             !test_range_bit(io_tree, offset, i_size_test - 1,
 870                             EXTENT_DELALLOC, 0, NULL)) {
 871                 new_i_size = min_t(u64, i_size_test, i_size);
 872         }
 873         BTRFS_I(inode)->disk_i_size = new_i_size;
 874         ret = 0;
 875 out:
 876         /*
 877          * we need to remove the ordered extent with the tree lock held
 878          * so that other people calling this function don't find our fully
 879          * processed ordered entry and skip updating the i_size
 880          */
 881         if (ordered)
 882                 __btrfs_remove_ordered_extent(inode, ordered);
 883         spin_unlock(&tree->lock);
 884         if (ordered)
 885                 wake_up(&ordered->wait);
 886         return ret;
 887 }
 888
 889 /*
 890  * search the ordered extents for one corresponding to 'offset' and
 891  * try to find a checksum.  This is used because we allow pages to
 892  * be reclaimed before their checksum is actually put into the btree
 893  */
 894 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
 895                            u32 *sum)
 896 {
 897         struct btrfs_ordered_sum *ordered_sum;
 898         struct btrfs_sector_sum *sector_sums;
 899         struct btrfs_ordered_extent *ordered;
 900         struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 901         unsigned long num_sectors;
 902         unsigned long i;
 903         u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
 904         int ret = 1;
 905
 906         ordered = btrfs_lookup_ordered_extent(inode, offset);
 907         if (!ordered)
 908                 return 1;
 909
 910         spin_lock(&tree->lock);
 911         list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
 912                 if (disk_bytenr >= ordered_sum->bytenr) {
 913                         num_sectors = ordered_sum->len / sectorsize;
 914                         sector_sums = ordered_sum->sums;
 915                         for (i = 0; i < num_sectors; i++) {
 916                                 if (sector_sums[i].bytenr == disk_bytenr) {
 917                                         *sum = sector_sums[i].sum;
 918                                         ret = 0;
 919                                         goto out;
 920                                 }
 921                         }
 922                 }
 923         }
 924 out:
 925         spin_unlock(&tree->lock);
 926         btrfs_put_ordered_extent(ordered);
 927         return ret;
 928 }
 929
 930
 931 /*
 932  * add a given inode to the list of inodes that must be fully on
 933  * disk before a transaction commit finishes.
 934  *
 935  * This basically gives us the ext3 style data=ordered mode, and it is mostly
 936  * used to make sure renamed files are fully on disk.
 937  *
 938  * It is a noop if the inode is already fully on disk.
 939  *
 940  * If trans is not null, we'll do a friendly check for a transaction that
 941  * is already flushing things and force the IO down ourselves.
 942  */
 943 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
 944                                 struct btrfs_root *root,
 945                                 struct inode *inode)
 946 {
 947         u64 last_mod;
 948
 949         last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
 950
 951         /*
 952          * if this file hasn't been changed since the last transaction
 953          * commit, we can safely return without doing anything
 954          */
 955         if (last_mod < root->fs_info->last_trans_committed)
 956                 return 0;
 957
 958         /*
 959          * the transaction is already committing.  Just start the IO and
 960          * don't bother with all of this list nonsense
 961          */
 962         if (trans && root->fs_info->running_transaction->blocked) {
 963                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
 964                 return 0;
 965         }
 966
 967         spin_lock(&root->fs_info->ordered_extent_lock);
 968         if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
 969                 list_add_tail(&BTRFS_I(inode)->ordered_operations,
 970                               &root->fs_info->ordered_operations);
 971         }
 972         spin_unlock(&root->fs_info->ordered_extent_lock);
 973
 974         return 0;
 975 }