fs/btrfs/tree-log.c

   1 /*
   2  * Copyright (C) 2008 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/sched.h>
  20 #include "ctree.h"
  21 #include "transaction.h"
  22 #include "disk-io.h"
  23 #include "locking.h"
  24 #include "print-tree.h"
  25 #include "compat.h"
  26
  27 /* magic values for the inode_only field in btrfs_log_inode:
  28  *
  29  * LOG_INODE_ALL means to log everything
  30  * LOG_INODE_EXISTS means to log just enough to recreate the inode
  31  * during log replay
  32  */
  33 #define LOG_INODE_ALL 0
  34 #define LOG_INODE_EXISTS 1
  35
  36 /*
  37  * stages for the tree walking.  The first
  38  * stage (0) is to only pin down the blocks we find
  39  * the second stage (1) is to make sure that all the inodes
  40  * we find in the log are created in the subvolume.
  41  *
  42  * The last stage is to deal with directories and links and extents
  43  * and all the other fun semantics
  44  */
  45 #define LOG_WALK_PIN_ONLY 0
  46 #define LOG_WALK_REPLAY_INODES 1
  47 #define LOG_WALK_REPLAY_ALL 2
  48
  49 static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
  50                              struct btrfs_root *root, struct inode *inode,
  51                              int inode_only);
  52
  53 /*
  54  * tree logging is a special write ahead log used to make sure that
  55  * fsyncs and O_SYNCs can happen without doing full tree commits.
  56  *
  57  * Full tree commits are expensive because they require commonly
  58  * modified blocks to be recowed, creating many dirty pages in the
  59  * extent tree an 4x-6x higher write load than ext3.
  60  *
  61  * Instead of doing a tree commit on every fsync, we use the
  62  * key ranges and transaction ids to find items for a given file or directory
  63  * that have changed in this transaction.  Those items are copied into
  64  * a special tree (one per subvolume root), that tree is written to disk
  65  * and then the fsync is considered complete.
  66  *
  67  * After a crash, items are copied out of the log-tree back into the
  68  * subvolume tree.  Any file data extents found are recorded in the extent
  69  * allocation tree, and the log-tree freed.
  70  *
  71  * The log tree is read three times, once to pin down all the extents it is
  72  * using in ram and once, once to create all the inodes logged in the tree
  73  * and once to do all the other items.
  74  */
  75
  76 /*
  77  * btrfs_add_log_tree adds a new per-subvolume log tree into the
  78  * tree of log tree roots.  This must be called with a tree log transaction
  79  * running (see start_log_trans).
  80  */
  81 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
  82                       struct btrfs_root *root)
  83 {
  84         struct btrfs_key key;
  85         struct btrfs_root_item root_item;
  86         struct btrfs_inode_item *inode_item;
  87         struct extent_buffer *leaf;
  88         struct btrfs_root *new_root = root;
  89         int ret;
  90         u64 objectid = root->root_key.objectid;
  91
  92         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
  93                                       BTRFS_TREE_LOG_OBJECTID,
  94                                       trans->transid, 0, 0, 0);
  95         if (IS_ERR(leaf)) {
  96                 ret = PTR_ERR(leaf);
  97                 return ret;
  98         }
  99
 100         btrfs_set_header_nritems(leaf, 0);
 101         btrfs_set_header_level(leaf, 0);
 102         btrfs_set_header_bytenr(leaf, leaf->start);
 103         btrfs_set_header_generation(leaf, trans->transid);
 104         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
 105
 106         write_extent_buffer(leaf, root->fs_info->fsid,
 107                             (unsigned long)btrfs_header_fsid(leaf),
 108                             BTRFS_FSID_SIZE);
 109         btrfs_mark_buffer_dirty(leaf);
 110
 111         inode_item = &root_item.inode;
 112         memset(inode_item, 0, sizeof(*inode_item));
 113         inode_item->generation = cpu_to_le64(1);
 114         inode_item->size = cpu_to_le64(3);
 115         inode_item->nlink = cpu_to_le32(1);
 116         inode_item->nbytes = cpu_to_le64(root->leafsize);
 117         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 118
 119         btrfs_set_root_bytenr(&root_item, leaf->start);
 120         btrfs_set_root_generation(&root_item, trans->transid);
 121         btrfs_set_root_level(&root_item, 0);
 122         btrfs_set_root_refs(&root_item, 0);
 123         btrfs_set_root_used(&root_item, 0);
 124
 125         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 126         root_item.drop_level = 0;
 127
 128         btrfs_tree_unlock(leaf);
 129         free_extent_buffer(leaf);
 130         leaf = NULL;
 131
 132         btrfs_set_root_dirid(&root_item, 0);
 133
 134         key.objectid = BTRFS_TREE_LOG_OBJECTID;
 135         key.offset = objectid;
 136         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 137         ret = btrfs_insert_root(trans, root->fs_info->log_root_tree, &key,
 138                                 &root_item);
 139         if (ret)
 140                 goto fail;
 141
 142         new_root = btrfs_read_fs_root_no_radix(root->fs_info->log_root_tree,
 143                                                &key);
 144         BUG_ON(!new_root);
 145
 146         WARN_ON(root->log_root);
 147         root->log_root = new_root;
 148
 149         /*
 150          * log trees do not get reference counted because they go away
 151          * before a real commit is actually done.  They do store pointers
 152          * to file data extents, and those reference counts still get
 153          * updated (along with back refs to the log tree).
 154          */
 155         new_root->ref_cows = 0;
 156         new_root->last_trans = trans->transid;
 157 fail:
 158         return ret;
 159 }
 160
 161 /*
 162  * start a sub transaction and setup the log tree
 163  * this increments the log tree writer count to make the people
 164  * syncing the tree wait for us to finish
 165  */
 166 static int start_log_trans(struct btrfs_trans_handle *trans,
 167                            struct btrfs_root *root)
 168 {
 169         int ret;
 170         mutex_lock(&root->fs_info->tree_log_mutex);
 171         if (!root->fs_info->log_root_tree) {
 172                 ret = btrfs_init_log_root_tree(trans, root->fs_info);
 173                 BUG_ON(ret);
 174         }
 175         if (!root->log_root) {
 176                 ret = btrfs_add_log_tree(trans, root);
 177                 BUG_ON(ret);
 178         }
 179         atomic_inc(&root->fs_info->tree_log_writers);
 180         root->fs_info->tree_log_batch++;
 181         mutex_unlock(&root->fs_info->tree_log_mutex);
 182         return 0;
 183 }
 184
 185 /*
 186  * returns 0 if there was a log transaction running and we were able
 187  * to join, or returns -ENOENT if there were not transactions
 188  * in progress
 189  */
 190 static int join_running_log_trans(struct btrfs_root *root)
 191 {
 192         int ret = -ENOENT;
 193
 194         smp_mb();
 195         if (!root->log_root)
 196                 return -ENOENT;
 197
 198         mutex_lock(&root->fs_info->tree_log_mutex);
 199         if (root->log_root) {
 200                 ret = 0;
 201                 atomic_inc(&root->fs_info->tree_log_writers);
 202                 root->fs_info->tree_log_batch++;
 203         }
 204         mutex_unlock(&root->fs_info->tree_log_mutex);
 205         return ret;
 206 }
 207
 208 /*
 209  * indicate we're done making changes to the log tree
 210  * and wake up anyone waiting to do a sync
 211  */
 212 static int end_log_trans(struct btrfs_root *root)
 213 {
 214         atomic_dec(&root->fs_info->tree_log_writers);
 215         smp_mb();
 216         if (waitqueue_active(&root->fs_info->tree_log_wait))
 217                 wake_up(&root->fs_info->tree_log_wait);
 218         return 0;
 219 }
 220
 221
 222 /*
 223  * the walk control struct is used to pass state down the chain when
 224  * processing the log tree.  The stage field tells us which part
 225  * of the log tree processing we are currently doing.  The others
 226  * are state fields used for that specific part
 227  */
 228 struct walk_control {
 229         /* should we free the extent on disk when done?  This is used
 230          * at transaction commit time while freeing a log tree
 231          */
 232         int free;
 233
 234         /* should we write out the extent buffer?  This is used
 235          * while flushing the log tree to disk during a sync
 236          */
 237         int write;
 238
 239         /* should we wait for the extent buffer io to finish?  Also used
 240          * while flushing the log tree to disk for a sync
 241          */
 242         int wait;
 243
 244         /* pin only walk, we record which extents on disk belong to the
 245          * log trees
 246          */
 247         int pin;
 248
 249         /* what stage of the replay code we're currently in */
 250         int stage;
 251
 252         /* the root we are currently replaying */
 253         struct btrfs_root *replay_dest;
 254
 255         /* the trans handle for the current replay */
 256         struct btrfs_trans_handle *trans;
 257
 258         /* the function that gets used to process blocks we find in the
 259          * tree.  Note the extent_buffer might not be up to date when it is
 260          * passed in, and it must be checked or read if you need the data
 261          * inside it
 262          */
 263         int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
 264                             struct walk_control *wc, u64 gen);
 265 };
 266
 267 /*
 268  * process_func used to pin down extents, write them or wait on them
 269  */
 270 static int process_one_buffer(struct btrfs_root *log,
 271                               struct extent_buffer *eb,
 272                               struct walk_control *wc, u64 gen)
 273 {
 274         if (wc->pin) {
 275                 mutex_lock(&log->fs_info->pinned_mutex);
 276                 btrfs_update_pinned_extents(log->fs_info->extent_root,
 277                                             eb->start, eb->len, 1);
 278                 mutex_unlock(&log->fs_info->pinned_mutex);
 279         }
 280
 281         if (btrfs_buffer_uptodate(eb, gen)) {
 282                 if (wc->write)
 283                         btrfs_write_tree_block(eb);
 284                 if (wc->wait)
 285                         btrfs_wait_tree_block_writeback(eb);
 286         }
 287         return 0;
 288 }
 289
 290 /*
 291  * Item overwrite used by replay and tree logging.  eb, slot and key all refer
 292  * to the src data we are copying out.
 293  *
 294  * root is the tree we are copying into, and path is a scratch
 295  * path for use in this function (it should be released on entry and
 296  * will be released on exit).
 297  *
 298  * If the key is already in the destination tree the existing item is
 299  * overwritten.  If the existing item isn't big enough, it is extended.
 300  * If it is too large, it is truncated.
 301  *
 302  * If the key isn't in the destination yet, a new item is inserted.
 303  */
 304 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
 305                                    struct btrfs_root *root,
 306                                    struct btrfs_path *path,
 307                                    struct extent_buffer *eb, int slot,
 308                                    struct btrfs_key *key)
 309 {
 310         int ret;
 311         u32 item_size;
 312         u64 saved_i_size = 0;
 313         int save_old_i_size = 0;
 314         unsigned long src_ptr;
 315         unsigned long dst_ptr;
 316         int overwrite_root = 0;
 317
 318         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
 319                 overwrite_root = 1;
 320
 321         item_size = btrfs_item_size_nr(eb, slot);
 322         src_ptr = btrfs_item_ptr_offset(eb, slot);
 323
 324         /* look for the key in the destination tree */
 325         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
 326         if (ret == 0) {
 327                 char *src_copy;
 328                 char *dst_copy;
 329                 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
 330                                                   path->slots[0]);
 331                 if (dst_size != item_size)
 332                         goto insert;
 333
 334                 if (item_size == 0) {
 335                         btrfs_release_path(root, path);
 336                         return 0;
 337                 }
 338                 dst_copy = kmalloc(item_size, GFP_NOFS);
 339                 src_copy = kmalloc(item_size, GFP_NOFS);
 340
 341                 read_extent_buffer(eb, src_copy, src_ptr, item_size);
 342
 343                 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
 344                 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
 345                                    item_size);
 346                 ret = memcmp(dst_copy, src_copy, item_size);
 347
 348                 kfree(dst_copy);
 349                 kfree(src_copy);
 350                 /*
 351                  * they have the same contents, just return, this saves
 352                  * us from cowing blocks in the destination tree and doing
 353                  * extra writes that may not have been done by a previous
 354                  * sync
 355                  */
 356                 if (ret == 0) {
 357                         btrfs_release_path(root, path);
 358                         return 0;
 359                 }
 360
 361         }
 362 insert:
 363         btrfs_release_path(root, path);
 364         /* try to insert the key into the destination tree */
 365         ret = btrfs_insert_empty_item(trans, root, path,
 366                                       key, item_size);
 367
 368         /* make sure any existing item is the correct size */
 369         if (ret == -EEXIST) {
 370                 u32 found_size;
 371                 found_size = btrfs_item_size_nr(path->nodes[0],
 372                                                 path->slots[0]);
 373                 if (found_size > item_size) {
 374                         btrfs_truncate_item(trans, root, path, item_size, 1);
 375                 } else if (found_size < item_size) {
 376                         ret = btrfs_del_item(trans, root,
 377                                              path);
 378                         BUG_ON(ret);
 379
 380                         btrfs_release_path(root, path);
 381                         ret = btrfs_insert_empty_item(trans,
 382                                   root, path, key, item_size);
 383                         BUG_ON(ret);
 384                 }
 385         } else if (ret) {
 386                 BUG();
 387         }
 388         dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
 389                                         path->slots[0]);
 390
 391         /* don't overwrite an existing inode if the generation number
 392          * was logged as zero.  This is done when the tree logging code
 393          * is just logging an inode to make sure it exists after recovery.
 394          *
 395          * Also, don't overwrite i_size on directories during replay.
 396          * log replay inserts and removes directory items based on the
 397          * state of the tree found in the subvolume, and i_size is modified
 398          * as it goes
 399          */
 400         if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
 401                 struct btrfs_inode_item *src_item;
 402                 struct btrfs_inode_item *dst_item;
 403
 404                 src_item = (struct btrfs_inode_item *)src_ptr;
 405                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 406
 407                 if (btrfs_inode_generation(eb, src_item) == 0)
 408                         goto no_copy;
 409
 410                 if (overwrite_root &&
 411                     S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
 412                     S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
 413                         save_old_i_size = 1;
 414                         saved_i_size = btrfs_inode_size(path->nodes[0],
 415                                                         dst_item);
 416                 }
 417         }
 418
 419         copy_extent_buffer(path->nodes[0], eb, dst_ptr,
 420                            src_ptr, item_size);
 421
 422         if (save_old_i_size) {
 423                 struct btrfs_inode_item *dst_item;
 424                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 425                 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
 426         }
 427
 428         /* make sure the generation is filled in */
 429         if (key->type == BTRFS_INODE_ITEM_KEY) {
 430                 struct btrfs_inode_item *dst_item;
 431                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 432                 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
 433                         btrfs_set_inode_generation(path->nodes[0], dst_item,
 434                                                    trans->transid);
 435                 }
 436         }
 437
 438         if (overwrite_root &&
 439             key->type == BTRFS_EXTENT_DATA_KEY) {
 440                 int extent_type;
 441                 struct btrfs_file_extent_item *fi;
 442
 443                 fi = (struct btrfs_file_extent_item *)dst_ptr;
 444                 extent_type = btrfs_file_extent_type(path->nodes[0], fi);
 445                 if (extent_type == BTRFS_FILE_EXTENT_REG) {
 446                         struct btrfs_key ins;
 447                         ins.objectid = btrfs_file_extent_disk_bytenr(
 448                                                         path->nodes[0], fi);
 449                         ins.offset = btrfs_file_extent_disk_num_bytes(
 450                                                         path->nodes[0], fi);
 451                         ins.type = BTRFS_EXTENT_ITEM_KEY;
 452
 453                         /*
 454                          * is this extent already allocated in the extent
 455                          * allocation tree?  If so, just add a reference
 456                          */
 457                         ret = btrfs_lookup_extent(root, ins.objectid,
 458                                                   ins.offset);
 459                         if (ret == 0) {
 460                                 ret = btrfs_inc_extent_ref(trans, root,
 461                                                 ins.objectid, ins.offset,
 462                                                 path->nodes[0]->start,
 463                                                 root->root_key.objectid,
 464                                                 trans->transid, key->objectid);
 465                         } else {
 466                                 /*
 467                                  * insert the extent pointer in the extent
 468                                  * allocation tree
 469                                  */
 470                                 ret = btrfs_alloc_logged_extent(trans, root,
 471                                                 path->nodes[0]->start,
 472                                                 root->root_key.objectid,
 473                                                 trans->transid, key->objectid,
 474                                                 &ins);
 475                                 BUG_ON(ret);
 476                         }
 477                 }
 478         }
 479 no_copy:
 480         btrfs_mark_buffer_dirty(path->nodes[0]);
 481         btrfs_release_path(root, path);
 482         return 0;
 483 }
 484
 485 /*
 486  * simple helper to read an inode off the disk from a given root
 487  * This can only be called for subvolume roots and not for the log
 488  */
 489 static noinline struct inode *read_one_inode(struct btrfs_root *root,
 490                                              u64 objectid)
 491 {
 492         struct inode *inode;
 493         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
 494         if (inode->i_state & I_NEW) {
 495                 BTRFS_I(inode)->root = root;
 496                 BTRFS_I(inode)->location.objectid = objectid;
 497                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
 498                 BTRFS_I(inode)->location.offset = 0;
 499                 btrfs_read_locked_inode(inode);
 500                 unlock_new_inode(inode);
 501
 502         }
 503         if (is_bad_inode(inode)) {
 504                 iput(inode);
 505                 inode = NULL;
 506         }
 507         return inode;
 508 }
 509
 510 /* replays a single extent in 'eb' at 'slot' with 'key' into the
 511  * subvolume 'root'.  path is released on entry and should be released
 512  * on exit.
 513  *
 514  * extents in the log tree have not been allocated out of the extent
 515  * tree yet.  So, this completes the allocation, taking a reference
 516  * as required if the extent already exists or creating a new extent
 517  * if it isn't in the extent allocation tree yet.
 518  *
 519  * The extent is inserted into the file, dropping any existing extents
 520  * from the file that overlap the new one.
 521  */
 522 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 523                                       struct btrfs_root *root,
 524                                       struct btrfs_path *path,
 525                                       struct extent_buffer *eb, int slot,
 526                                       struct btrfs_key *key)
 527 {
 528         int found_type;
 529         u64 mask = root->sectorsize - 1;
 530         u64 extent_end;
 531         u64 alloc_hint;
 532         u64 start = key->offset;
 533         struct btrfs_file_extent_item *item;
 534         struct inode *inode = NULL;
 535         unsigned long size;
 536         int ret = 0;
 537
 538         item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 539         found_type = btrfs_file_extent_type(eb, item);
 540
 541         if (found_type == BTRFS_FILE_EXTENT_REG)
 542                 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
 543         else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 544                 size = btrfs_file_extent_inline_len(eb, item);
 545                 extent_end = (start + size + mask) & ~mask;
 546         } else {
 547                 ret = 0;
 548                 goto out;
 549         }
 550
 551         inode = read_one_inode(root, key->objectid);
 552         if (!inode) {
 553                 ret = -EIO;
 554                 goto out;
 555         }
 556
 557         /*
 558          * first check to see if we already have this extent in the
 559          * file.  This must be done before the btrfs_drop_extents run
 560          * so we don't try to drop this extent.
 561          */
 562         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 563                                        start, 0);
 564
 565         if (ret == 0 && found_type == BTRFS_FILE_EXTENT_REG) {
 566                 struct btrfs_file_extent_item cmp1;
 567                 struct btrfs_file_extent_item cmp2;
 568                 struct btrfs_file_extent_item *existing;
 569                 struct extent_buffer *leaf;
 570
 571                 leaf = path->nodes[0];
 572                 existing = btrfs_item_ptr(leaf, path->slots[0],
 573                                           struct btrfs_file_extent_item);
 574
 575                 read_extent_buffer(eb, &cmp1, (unsigned long)item,
 576                                    sizeof(cmp1));
 577                 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
 578                                    sizeof(cmp2));
 579
 580                 /*
 581                  * we already have a pointer to this exact extent,
 582                  * we don't have to do anything
 583                  */
 584                 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
 585                         btrfs_release_path(root, path);
 586                         goto out;
 587                 }
 588         }
 589         btrfs_release_path(root, path);
 590
 591         /* drop any overlapping extents */
 592         ret = btrfs_drop_extents(trans, root, inode,
 593                          start, extent_end, start, &alloc_hint);
 594         BUG_ON(ret);
 595
 596         /* insert the extent */
 597         ret = overwrite_item(trans, root, path, eb, slot, key);
 598         BUG_ON(ret);
 599
 600         /* btrfs_drop_extents changes i_bytes & i_blocks, update it here */
 601         inode_add_bytes(inode, extent_end - start);
 602         btrfs_update_inode(trans, root, inode);
 603 out:
 604         if (inode)
 605                 iput(inode);
 606         return ret;
 607 }
 608
 609 /*
 610  * when cleaning up conflicts between the directory names in the
 611  * subvolume, directory names in the log and directory names in the
 612  * inode back references, we may have to unlink inodes from directories.
 613  *
 614  * This is a helper function to do the unlink of a specific directory
 615  * item
 616  */
 617 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
 618                                       struct btrfs_root *root,
 619                                       struct btrfs_path *path,
 620                                       struct inode *dir,
 621                                       struct btrfs_dir_item *di)
 622 {
 623         struct inode *inode;
 624         char *name;
 625         int name_len;
 626         struct extent_buffer *leaf;
 627         struct btrfs_key location;
 628         int ret;
 629
 630         leaf = path->nodes[0];
 631
 632         btrfs_dir_item_key_to_cpu(leaf, di, &location);
 633         name_len = btrfs_dir_name_len(leaf, di);
 634         name = kmalloc(name_len, GFP_NOFS);
 635         read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
 636         btrfs_release_path(root, path);
 637
 638         inode = read_one_inode(root, location.objectid);
 639         BUG_ON(!inode);
 640
 641         btrfs_inc_nlink(inode);
 642         ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
 643         kfree(name);
 644
 645         iput(inode);
 646         return ret;
 647 }
 648
 649 /*
 650  * helper function to see if a given name and sequence number found
 651  * in an inode back reference are already in a directory and correctly
 652  * point to this inode
 653  */
 654 static noinline int inode_in_dir(struct btrfs_root *root,
 655                                  struct btrfs_path *path,
 656                                  u64 dirid, u64 objectid, u64 index,
 657                                  const char *name, int name_len)
 658 {
 659         struct btrfs_dir_item *di;
 660         struct btrfs_key location;
 661         int match = 0;
 662
 663         di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
 664                                          index, name, name_len, 0);
 665         if (di && !IS_ERR(di)) {
 666                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 667                 if (location.objectid != objectid)
 668                         goto out;
 669         } else
 670                 goto out;
 671         btrfs_release_path(root, path);
 672
 673         di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
 674         if (di && !IS_ERR(di)) {
 675                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 676                 if (location.objectid != objectid)
 677                         goto out;
 678         } else
 679                 goto out;
 680         match = 1;
 681 out:
 682         btrfs_release_path(root, path);
 683         return match;
 684 }
 685
 686 /*
 687  * helper function to check a log tree for a named back reference in
 688  * an inode.  This is used to decide if a back reference that is
 689  * found in the subvolume conflicts with what we find in the log.
 690  *
 691  * inode backreferences may have multiple refs in a single item,
 692  * during replay we process one reference at a time, and we don't
 693  * want to delete valid links to a file from the subvolume if that
 694  * link is also in the log.
 695  */
 696 static noinline int backref_in_log(struct btrfs_root *log,
 697                                    struct btrfs_key *key,
 698                                    char *name, int namelen)
 699 {
 700         struct btrfs_path *path;
 701         struct btrfs_inode_ref *ref;
 702         unsigned long ptr;
 703         unsigned long ptr_end;
 704         unsigned long name_ptr;
 705         int found_name_len;
 706         int item_size;
 707         int ret;
 708         int match = 0;
 709
 710         path = btrfs_alloc_path();
 711         ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
 712         if (ret != 0)
 713                 goto out;
 714
 715         item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
 716         ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
 717         ptr_end = ptr + item_size;
 718         while (ptr < ptr_end) {
 719                 ref = (struct btrfs_inode_ref *)ptr;
 720                 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
 721                 if (found_name_len == namelen) {
 722                         name_ptr = (unsigned long)(ref + 1);
 723                         ret = memcmp_extent_buffer(path->nodes[0], name,
 724                                                    name_ptr, namelen);
 725                         if (ret == 0) {
 726                                 match = 1;
 727                                 goto out;
 728                         }
 729                 }
 730                 ptr = (unsigned long)(ref + 1) + found_name_len;
 731         }
 732 out:
 733         btrfs_free_path(path);
 734         return match;
 735 }
 736
 737
 738 /*
 739  * replay one inode back reference item found in the log tree.
 740  * eb, slot and key refer to the buffer and key found in the log tree.
 741  * root is the destination we are replaying into, and path is for temp
 742  * use by this function.  (it should be released on return).
 743  */
 744 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
 745                                   struct btrfs_root *root,
 746                                   struct btrfs_root *log,
 747                                   struct btrfs_path *path,
 748                                   struct extent_buffer *eb, int slot,
 749                                   struct btrfs_key *key)
 750 {
 751         struct inode *dir;
 752         int ret;
 753         struct btrfs_key location;
 754         struct btrfs_inode_ref *ref;
 755         struct btrfs_dir_item *di;
 756         struct inode *inode;
 757         char *name;
 758         int namelen;
 759         unsigned long ref_ptr;
 760         unsigned long ref_end;
 761
 762         location.objectid = key->objectid;
 763         location.type = BTRFS_INODE_ITEM_KEY;
 764         location.offset = 0;
 765
 766         /*
 767          * it is possible that we didn't log all the parent directories
 768          * for a given inode.  If we don't find the dir, just don't
 769          * copy the back ref in.  The link count fixup code will take
 770          * care of the rest
 771          */
 772         dir = read_one_inode(root, key->offset);
 773         if (!dir)
 774                 return -ENOENT;
 775
 776         inode = read_one_inode(root, key->objectid);
 777         BUG_ON(!dir);
 778
 779         ref_ptr = btrfs_item_ptr_offset(eb, slot);
 780         ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
 781
 782 again:
 783         ref = (struct btrfs_inode_ref *)ref_ptr;
 784
 785         namelen = btrfs_inode_ref_name_len(eb, ref);
 786         name = kmalloc(namelen, GFP_NOFS);
 787         BUG_ON(!name);
 788
 789         read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
 790
 791         /* if we already have a perfect match, we're done */
 792         if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
 793                          btrfs_inode_ref_index(eb, ref),
 794                          name, namelen)) {
 795                 goto out;
 796         }
 797
 798         /*
 799          * look for a conflicting back reference in the metadata.
 800          * if we find one we have to unlink that name of the file
 801          * before we add our new link.  Later on, we overwrite any
 802          * existing back reference, and we don't want to create
 803          * dangling pointers in the directory.
 804          */
 805 conflict_again:
 806         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
 807         if (ret == 0) {
 808                 char *victim_name;
 809                 int victim_name_len;
 810                 struct btrfs_inode_ref *victim_ref;
 811                 unsigned long ptr;
 812                 unsigned long ptr_end;
 813                 struct extent_buffer *leaf = path->nodes[0];
 814
 815                 /* are we trying to overwrite a back ref for the root directory
 816                  * if so, just jump out, we're done
 817                  */
 818                 if (key->objectid == key->offset)
 819                         goto out_nowrite;
 820
 821                 /* check all the names in this back reference to see
 822                  * if they are in the log.  if so, we allow them to stay
 823                  * otherwise they must be unlinked as a conflict
 824                  */
 825                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
 826                 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
 827                 while(ptr < ptr_end) {
 828                         victim_ref = (struct btrfs_inode_ref *)ptr;
 829                         victim_name_len = btrfs_inode_ref_name_len(leaf,
 830                                                                    victim_ref);
 831                         victim_name = kmalloc(victim_name_len, GFP_NOFS);
 832                         BUG_ON(!victim_name);
 833
 834                         read_extent_buffer(leaf, victim_name,
 835                                            (unsigned long)(victim_ref + 1),
 836                                            victim_name_len);
 837
 838                         if (!backref_in_log(log, key, victim_name,
 839                                             victim_name_len)) {
 840                                 btrfs_inc_nlink(inode);
 841                                 btrfs_release_path(root, path);
 842                                 ret = btrfs_unlink_inode(trans, root, dir,
 843                                                          inode, victim_name,
 844                                                          victim_name_len);
 845                                 kfree(victim_name);
 846                                 btrfs_release_path(root, path);
 847                                 goto conflict_again;
 848                         }
 849                         kfree(victim_name);
 850                         ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
 851                 }
 852                 BUG_ON(ret);
 853         }
 854         btrfs_release_path(root, path);
 855
 856         /* look for a conflicting sequence number */
 857         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 858                                          btrfs_inode_ref_index(eb, ref),
 859                                          name, namelen, 0);
 860         if (di && !IS_ERR(di)) {
 861                 ret = drop_one_dir_item(trans, root, path, dir, di);
 862                 BUG_ON(ret);
 863         }
 864         btrfs_release_path(root, path);
 865
 866
 867         /* look for a conflicting name */
 868         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 869                                    name, namelen, 0);
 870         if (di && !IS_ERR(di)) {
 871                 ret = drop_one_dir_item(trans, root, path, dir, di);
 872                 BUG_ON(ret);
 873         }
 874         btrfs_release_path(root, path);
 875
 876         /* insert our name */
 877         ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
 878                              btrfs_inode_ref_index(eb, ref));
 879         BUG_ON(ret);
 880
 881         btrfs_update_inode(trans, root, inode);
 882
 883 out:
 884         ref_ptr = (unsigned long)(ref + 1) + namelen;
 885         kfree(name);
 886         if (ref_ptr < ref_end)
 887                 goto again;
 888
 889         /* finally write the back reference in the inode */
 890         ret = overwrite_item(trans, root, path, eb, slot, key);
 891         BUG_ON(ret);
 892
 893 out_nowrite:
 894         btrfs_release_path(root, path);
 895         iput(dir);
 896         iput(inode);
 897         return 0;
 898 }
 899
 900 /*
 901  * replay one csum item from the log tree into the subvolume 'root'
 902  * eb, slot and key all refer to the log tree
 903  * path is for temp use by this function and should be released on return
 904  *
 905  * This copies the checksums out of the log tree and inserts them into
 906  * the subvolume.  Any existing checksums for this range in the file
 907  * are overwritten, and new items are added where required.
 908  *
 909  * We keep this simple by reusing the btrfs_ordered_sum code from
 910  * the data=ordered mode.  This basically means making a copy
 911  * of all the checksums in ram, which we have to do anyway for kmap
 912  * rules.
 913  *
 914  * The copy is then sent down to btrfs_csum_file_blocks, which
 915  * does all the hard work of finding existing items in the file
 916  * or adding new ones.
 917  */
 918 static noinline int replay_one_csum(struct btrfs_trans_handle *trans,
 919                                       struct btrfs_root *root,
 920                                       struct btrfs_path *path,
 921                                       struct extent_buffer *eb, int slot,
 922                                       struct btrfs_key *key)
 923 {
 924         int ret;
 925         u32 item_size = btrfs_item_size_nr(eb, slot);
 926         u64 cur_offset;
 927         unsigned long file_bytes;
 928         struct btrfs_ordered_sum *sums;
 929         struct btrfs_sector_sum *sector_sum;
 930         struct inode *inode;
 931         unsigned long ptr;
 932
 933         file_bytes = (item_size / BTRFS_CRC32_SIZE) * root->sectorsize;
 934         inode = read_one_inode(root, key->objectid);
 935         if (!inode) {
 936                 return -EIO;
 937         }
 938
 939         sums = kzalloc(btrfs_ordered_sum_size(root, file_bytes), GFP_NOFS);
 940         if (!sums) {
 941                 iput(inode);
 942                 return -ENOMEM;
 943         }
 944
 945         INIT_LIST_HEAD(&sums->list);
 946         sums->len = file_bytes;
 947         sums->file_offset = key->offset;
 948
 949         /*
 950          * copy all the sums into the ordered sum struct
 951          */
 952         sector_sum = sums->sums;
 953         cur_offset = key->offset;
 954         ptr = btrfs_item_ptr_offset(eb, slot);
 955         while(item_size > 0) {
 956                 sector_sum->offset = cur_offset;
 957                 read_extent_buffer(eb, &sector_sum->sum, ptr, BTRFS_CRC32_SIZE);
 958                 sector_sum++;
 959                 item_size -= BTRFS_CRC32_SIZE;
 960                 ptr += BTRFS_CRC32_SIZE;
 961                 cur_offset += root->sectorsize;
 962         }
 963
 964         /* let btrfs_csum_file_blocks add them into the file */
 965         ret = btrfs_csum_file_blocks(trans, root, inode, sums);
 966         BUG_ON(ret);
 967         kfree(sums);
 968         iput(inode);
 969
 970         return 0;
 971 }
 972 /*
 973  * There are a few corners where the link count of the file can't
 974  * be properly maintained during replay.  So, instead of adding
 975  * lots of complexity to the log code, we just scan the backrefs
 976  * for any file that has been through replay.
 977  *
 978  * The scan will update the link count on the inode to reflect the
 979  * number of back refs found.  If it goes down to zero, the iput
 980  * will free the inode.
 981  */
 982 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
 983                                            struct btrfs_root *root,
 984                                            struct inode *inode)
 985 {
 986         struct btrfs_path *path;
 987         int ret;
 988         struct btrfs_key key;
 989         u64 nlink = 0;
 990         unsigned long ptr;
 991         unsigned long ptr_end;
 992         int name_len;
 993
 994         key.objectid = inode->i_ino;
 995         key.type = BTRFS_INODE_REF_KEY;
 996         key.offset = (u64)-1;
 997
 998         path = btrfs_alloc_path();
 999
1000         while(1) {
1001                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1002                 if (ret < 0)
1003                         break;
1004                 if (ret > 0) {
1005                         if (path->slots[0] == 0)
1006                                 break;
1007                         path->slots[0]--;
1008                 }
1009                 btrfs_item_key_to_cpu(path->nodes[0], &key,
1010                                       path->slots[0]);
1011                 if (key.objectid != inode->i_ino ||
1012                     key.type != BTRFS_INODE_REF_KEY)
1013                         break;
1014                 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1015                 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1016                                                    path->slots[0]);
1017                 while(ptr < ptr_end) {
1018                         struct btrfs_inode_ref *ref;
1019
1020                         ref = (struct btrfs_inode_ref *)ptr;
1021                         name_len = btrfs_inode_ref_name_len(path->nodes[0],
1022                                                             ref);
1023                         ptr = (unsigned long)(ref + 1) + name_len;
1024                         nlink++;
1025                 }
1026
1027                 if (key.offset == 0)
1028                         break;
1029                 key.offset--;
1030                 btrfs_release_path(root, path);
1031         }
1032         btrfs_free_path(path);
1033         if (nlink != inode->i_nlink) {
1034                 inode->i_nlink = nlink;
1035                 btrfs_update_inode(trans, root, inode);
1036         }
1037         BTRFS_I(inode)->index_cnt = (u64)-1;
1038
1039         return 0;
1040 }
1041
1042 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1043                                             struct btrfs_root *root,
1044                                             struct btrfs_path *path)
1045 {
1046         int ret;
1047         struct btrfs_key key;
1048         struct inode *inode;
1049
1050         key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1051         key.type = BTRFS_ORPHAN_ITEM_KEY;
1052         key.offset = (u64)-1;
1053         while(1) {
1054                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1055                 if (ret < 0)
1056                         break;
1057
1058                 if (ret == 1) {
1059                         if (path->slots[0] == 0)
1060                                 break;
1061                         path->slots[0]--;
1062                 }
1063
1064                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1065                 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1066                     key.type != BTRFS_ORPHAN_ITEM_KEY)
1067                         break;
1068
1069                 ret = btrfs_del_item(trans, root, path);
1070                 BUG_ON(ret);
1071
1072                 btrfs_release_path(root, path);
1073                 inode = read_one_inode(root, key.offset);
1074                 BUG_ON(!inode);
1075
1076                 ret = fixup_inode_link_count(trans, root, inode);
1077                 BUG_ON(ret);
1078
1079                 iput(inode);
1080
1081                 if (key.offset == 0)
1082                         break;
1083                 key.offset--;
1084         }
1085         btrfs_release_path(root, path);
1086         return 0;
1087 }
1088
1089
1090 /*
1091  * record a given inode in the fixup dir so we can check its link
1092  * count when replay is done.  The link count is incremented here
1093  * so the inode won't go away until we check it
1094  */
1095 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1096                                       struct btrfs_root *root,
1097                                       struct btrfs_path *path,
1098                                       u64 objectid)
1099 {
1100         struct btrfs_key key;
1101         int ret = 0;
1102         struct inode *inode;
1103
1104         inode = read_one_inode(root, objectid);
1105         BUG_ON(!inode);
1106
1107         key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1108         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1109         key.offset = objectid;
1110
1111         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1112
1113         btrfs_release_path(root, path);
1114         if (ret == 0) {
1115                 btrfs_inc_nlink(inode);
1116                 btrfs_update_inode(trans, root, inode);
1117         } else if (ret == -EEXIST) {
1118                 ret = 0;
1119         } else {
1120                 BUG();
1121         }
1122         iput(inode);
1123
1124         return ret;
1125 }
1126
1127 /*
1128  * when replaying the log for a directory, we only insert names
1129  * for inodes that actually exist.  This means an fsync on a directory
1130  * does not implicitly fsync all the new files in it
1131  */
1132 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1133                                     struct btrfs_root *root,
1134                                     struct btrfs_path *path,
1135                                     u64 dirid, u64 index,
1136                                     char *name, int name_len, u8 type,
1137                                     struct btrfs_key *location)
1138 {
1139         struct inode *inode;
1140         struct inode *dir;
1141         int ret;
1142
1143         inode = read_one_inode(root, location->objectid);
1144         if (!inode)
1145                 return -ENOENT;
1146
1147         dir = read_one_inode(root, dirid);
1148         if (!dir) {
1149                 iput(inode);
1150                 return -EIO;
1151         }
1152         ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1153
1154         /* FIXME, put inode into FIXUP list */
1155
1156         iput(inode);
1157         iput(dir);
1158         return ret;
1159 }
1160
1161 /*
1162  * take a single entry in a log directory item and replay it into
1163  * the subvolume.
1164  *
1165  * if a conflicting item exists in the subdirectory already,
1166  * the inode it points to is unlinked and put into the link count
1167  * fix up tree.
1168  *
1169  * If a name from the log points to a file or directory that does
1170  * not exist in the FS, it is skipped.  fsyncs on directories
1171  * do not force down inodes inside that directory, just changes to the
1172  * names or unlinks in a directory.
1173  */
1174 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1175                                     struct btrfs_root *root,
1176                                     struct btrfs_path *path,
1177                                     struct extent_buffer *eb,
1178                                     struct btrfs_dir_item *di,
1179                                     struct btrfs_key *key)
1180 {
1181         char *name;
1182         int name_len;
1183         struct btrfs_dir_item *dst_di;
1184         struct btrfs_key found_key;
1185         struct btrfs_key log_key;
1186         struct inode *dir;
1187         u8 log_type;
1188         int exists;
1189         int ret;
1190
1191         dir = read_one_inode(root, key->objectid);
1192         BUG_ON(!dir);
1193
1194         name_len = btrfs_dir_name_len(eb, di);
1195         name = kmalloc(name_len, GFP_NOFS);
1196         log_type = btrfs_dir_type(eb, di);
1197         read_extent_buffer(eb, name, (unsigned long)(di + 1),
1198                    name_len);
1199
1200         btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1201         exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1202         if (exists == 0)
1203                 exists = 1;
1204         else
1205                 exists = 0;
1206         btrfs_release_path(root, path);
1207
1208         if (key->type == BTRFS_DIR_ITEM_KEY) {
1209                 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1210                                        name, name_len, 1);
1211         }
1212         else if (key->type == BTRFS_DIR_INDEX_KEY) {
1213                 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1214                                                      key->objectid,
1215                                                      key->offset, name,
1216                                                      name_len, 1);
1217         } else {
1218                 BUG();
1219         }
1220         if (!dst_di || IS_ERR(dst_di)) {
1221                 /* we need a sequence number to insert, so we only
1222                  * do inserts for the BTRFS_DIR_INDEX_KEY types
1223                  */
1224                 if (key->type != BTRFS_DIR_INDEX_KEY)
1225                         goto out;
1226                 goto insert;
1227         }
1228
1229         btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1230         /* the existing item matches the logged item */
1231         if (found_key.objectid == log_key.objectid &&
1232             found_key.type == log_key.type &&
1233             found_key.offset == log_key.offset &&
1234             btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1235                 goto out;
1236         }
1237
1238         /*
1239          * don't drop the conflicting directory entry if the inode
1240          * for the new entry doesn't exist
1241          */
1242         if (!exists)
1243                 goto out;
1244
1245         ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1246         BUG_ON(ret);
1247
1248         if (key->type == BTRFS_DIR_INDEX_KEY)
1249                 goto insert;
1250 out:
1251         btrfs_release_path(root, path);
1252         kfree(name);
1253         iput(dir);
1254         return 0;
1255
1256 insert:
1257         btrfs_release_path(root, path);
1258         ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1259                               name, name_len, log_type, &log_key);
1260
1261         if (ret && ret != -ENOENT)
1262                 BUG();
1263         goto out;
1264 }
1265
1266 /*
1267  * find all the names in a directory item and reconcile them into
1268  * the subvolume.  Only BTRFS_DIR_ITEM_KEY types will have more than
1269  * one name in a directory item, but the same code gets used for
1270  * both directory index types
1271  */
1272 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1273                                         struct btrfs_root *root,
1274                                         struct btrfs_path *path,
1275                                         struct extent_buffer *eb, int slot,
1276                                         struct btrfs_key *key)
1277 {
1278         int ret;
1279         u32 item_size = btrfs_item_size_nr(eb, slot);
1280         struct btrfs_dir_item *di;
1281         int name_len;
1282         unsigned long ptr;
1283         unsigned long ptr_end;
1284
1285         ptr = btrfs_item_ptr_offset(eb, slot);
1286         ptr_end = ptr + item_size;
1287         while(ptr < ptr_end) {
1288                 di = (struct btrfs_dir_item *)ptr;
1289                 name_len = btrfs_dir_name_len(eb, di);
1290                 ret = replay_one_name(trans, root, path, eb, di, key);
1291                 BUG_ON(ret);
1292                 ptr = (unsigned long)(di + 1);
1293                 ptr += name_len;
1294         }
1295         return 0;
1296 }
1297
1298 /*
1299  * directory replay has two parts.  There are the standard directory
1300  * items in the log copied from the subvolume, and range items
1301  * created in the log while the subvolume was logged.
1302  *
1303  * The range items tell us which parts of the key space the log
1304  * is authoritative for.  During replay, if a key in the subvolume
1305  * directory is in a logged range item, but not actually in the log
1306  * that means it was deleted from the directory before the fsync
1307  * and should be removed.
1308  */
1309 static noinline int find_dir_range(struct btrfs_root *root,
1310                                    struct btrfs_path *path,
1311                                    u64 dirid, int key_type,
1312                                    u64 *start_ret, u64 *end_ret)
1313 {
1314         struct btrfs_key key;
1315         u64 found_end;
1316         struct btrfs_dir_log_item *item;
1317         int ret;
1318         int nritems;
1319
1320         if (*start_ret == (u64)-1)
1321                 return 1;
1322
1323         key.objectid = dirid;
1324         key.type = key_type;
1325         key.offset = *start_ret;
1326
1327         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1328         if (ret < 0)
1329                 goto out;
1330         if (ret > 0) {
1331                 if (path->slots[0] == 0)
1332                         goto out;
1333                 path->slots[0]--;
1334         }
1335         if (ret != 0)
1336                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1337
1338         if (key.type != key_type || key.objectid != dirid) {
1339                 ret = 1;
1340                 goto next;
1341         }
1342         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1343                               struct btrfs_dir_log_item);
1344         found_end = btrfs_dir_log_end(path->nodes[0], item);
1345
1346         if (*start_ret >= key.offset && *start_ret <= found_end) {
1347                 ret = 0;
1348                 *start_ret = key.offset;
1349                 *end_ret = found_end;
1350                 goto out;
1351         }
1352         ret = 1;
1353 next:
1354         /* check the next slot in the tree to see if it is a valid item */
1355         nritems = btrfs_header_nritems(path->nodes[0]);
1356         if (path->slots[0] >= nritems) {
1357                 ret = btrfs_next_leaf(root, path);
1358                 if (ret)
1359                         goto out;
1360         } else {
1361                 path->slots[0]++;
1362         }
1363
1364         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1365
1366         if (key.type != key_type || key.objectid != dirid) {
1367                 ret = 1;
1368                 goto out;
1369         }
1370         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1371                               struct btrfs_dir_log_item);
1372         found_end = btrfs_dir_log_end(path->nodes[0], item);
1373         *start_ret = key.offset;
1374         *end_ret = found_end;
1375         ret = 0;
1376 out:
1377         btrfs_release_path(root, path);
1378         return ret;
1379 }
1380
1381 /*
1382  * this looks for a given directory item in the log.  If the directory
1383  * item is not in the log, the item is removed and the inode it points
1384  * to is unlinked
1385  */
1386 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1387                                       struct btrfs_root *root,
1388                                       struct btrfs_root *log,
1389                                       struct btrfs_path *path,
1390                                       struct btrfs_path *log_path,
1391                                       struct inode *dir,
1392                                       struct btrfs_key *dir_key)
1393 {
1394         int ret;
1395         struct extent_buffer *eb;
1396         int slot;
1397         u32 item_size;
1398         struct btrfs_dir_item *di;
1399         struct btrfs_dir_item *log_di;
1400         int name_len;
1401         unsigned long ptr;
1402         unsigned long ptr_end;
1403         char *name;
1404         struct inode *inode;
1405         struct btrfs_key location;
1406
1407 again:
1408         eb = path->nodes[0];
1409         slot = path->slots[0];
1410         item_size = btrfs_item_size_nr(eb, slot);
1411         ptr = btrfs_item_ptr_offset(eb, slot);
1412         ptr_end = ptr + item_size;
1413         while(ptr < ptr_end) {
1414                 di = (struct btrfs_dir_item *)ptr;
1415                 name_len = btrfs_dir_name_len(eb, di);
1416                 name = kmalloc(name_len, GFP_NOFS);
1417                 if (!name) {
1418                         ret = -ENOMEM;
1419                         goto out;
1420                 }
1421                 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1422                                   name_len);
1423                 log_di = NULL;
1424                 if (dir_key->type == BTRFS_DIR_ITEM_KEY) {
1425                         log_di = btrfs_lookup_dir_item(trans, log, log_path,
1426                                                        dir_key->objectid,
1427                                                        name, name_len, 0);
1428                 } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) {
1429                         log_di = btrfs_lookup_dir_index_item(trans, log,
1430                                                      log_path,
1431                                                      dir_key->objectid,
1432                                                      dir_key->offset,
1433                                                      name, name_len, 0);
1434                 }
1435                 if (!log_di || IS_ERR(log_di)) {
1436                         btrfs_dir_item_key_to_cpu(eb, di, &location);
1437                         btrfs_release_path(root, path);
1438                         btrfs_release_path(log, log_path);
1439                         inode = read_one_inode(root, location.objectid);
1440                         BUG_ON(!inode);
1441
1442                         ret = link_to_fixup_dir(trans, root,
1443                                                 path, location.objectid);
1444                         BUG_ON(ret);
1445                         btrfs_inc_nlink(inode);
1446                         ret = btrfs_unlink_inode(trans, root, dir, inode,
1447                                                  name, name_len);
1448                         BUG_ON(ret);
1449                         kfree(name);
1450                         iput(inode);
1451
1452                         /* there might still be more names under this key
1453                          * check and repeat if required
1454                          */
1455                         ret = btrfs_search_slot(NULL, root, dir_key, path,
1456                                                 0, 0);
1457                         if (ret == 0)
1458                                 goto again;
1459                         ret = 0;
1460                         goto out;
1461                 }
1462                 btrfs_release_path(log, log_path);
1463                 kfree(name);
1464
1465                 ptr = (unsigned long)(di + 1);
1466                 ptr += name_len;
1467         }
1468         ret = 0;
1469 out:
1470         btrfs_release_path(root, path);
1471         btrfs_release_path(log, log_path);
1472         return ret;
1473 }
1474
1475 /*
1476  * deletion replay happens before we copy any new directory items
1477  * out of the log or out of backreferences from inodes.  It
1478  * scans the log to find ranges of keys that log is authoritative for,
1479  * and then scans the directory to find items in those ranges that are
1480  * not present in the log.
1481  *
1482  * Anything we don't find in the log is unlinked and removed from the
1483  * directory.
1484  */
1485 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1486                                        struct btrfs_root *root,
1487                                        struct btrfs_root *log,
1488                                        struct btrfs_path *path,
1489                                        u64 dirid)
1490 {
1491         u64 range_start;
1492         u64 range_end;
1493         int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1494         int ret = 0;
1495         struct btrfs_key dir_key;
1496         struct btrfs_key found_key;
1497         struct btrfs_path *log_path;
1498         struct inode *dir;
1499
1500         dir_key.objectid = dirid;
1501         dir_key.type = BTRFS_DIR_ITEM_KEY;
1502         log_path = btrfs_alloc_path();
1503         if (!log_path)
1504                 return -ENOMEM;
1505
1506         dir = read_one_inode(root, dirid);
1507         /* it isn't an error if the inode isn't there, that can happen
1508          * because we replay the deletes before we copy in the inode item
1509          * from the log
1510          */
1511         if (!dir) {
1512                 btrfs_free_path(log_path);
1513                 return 0;
1514         }
1515 again:
1516         range_start = 0;
1517         range_end = 0;
1518         while(1) {
1519                 ret = find_dir_range(log, path, dirid, key_type,
1520                                      &range_start, &range_end);
1521                 if (ret != 0)
1522                         break;
1523
1524                 dir_key.offset = range_start;
1525                 while(1) {
1526                         int nritems;
1527                         ret = btrfs_search_slot(NULL, root, &dir_key, path,
1528                                                 0, 0);
1529                         if (ret < 0)
1530                                 goto out;
1531
1532                         nritems = btrfs_header_nritems(path->nodes[0]);
1533                         if (path->slots[0] >= nritems) {
1534                                 ret = btrfs_next_leaf(root, path);
1535                                 if (ret)
1536                                         break;
1537                         }
1538                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1539                                               path->slots[0]);
1540                         if (found_key.objectid != dirid ||
1541                             found_key.type != dir_key.type)
1542                                 goto next_type;
1543
1544                         if (found_key.offset > range_end)
1545                                 break;
1546
1547                         ret = check_item_in_log(trans, root, log, path,
1548                                                 log_path, dir, &found_key);
1549                         BUG_ON(ret);
1550                         if (found_key.offset == (u64)-1)
1551                                 break;
1552                         dir_key.offset = found_key.offset + 1;
1553                 }
1554                 btrfs_release_path(root, path);
1555                 if (range_end == (u64)-1)
1556                         break;
1557                 range_start = range_end + 1;
1558         }
1559
1560 next_type:
1561         ret = 0;
1562         if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1563                 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1564                 dir_key.type = BTRFS_DIR_INDEX_KEY;
1565                 btrfs_release_path(root, path);
1566                 goto again;
1567         }
1568 out:
1569         btrfs_release_path(root, path);
1570         btrfs_free_path(log_path);
1571         iput(dir);
1572         return ret;
1573 }
1574
1575 /*
1576  * the process_func used to replay items from the log tree.  This
1577  * gets called in two different stages.  The first stage just looks
1578  * for inodes and makes sure they are all copied into the subvolume.
1579  *
1580  * The second stage copies all the other item types from the log into
1581  * the subvolume.  The two stage approach is slower, but gets rid of
1582  * lots of complexity around inodes referencing other inodes that exist
1583  * only in the log (references come from either directory items or inode
1584  * back refs).
1585  */
1586 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1587                              struct walk_control *wc, u64 gen)
1588 {
1589         int nritems;
1590         struct btrfs_path *path;
1591         struct btrfs_root *root = wc->replay_dest;
1592         struct btrfs_key key;
1593         u32 item_size;
1594         int level;
1595         int i;
1596         int ret;
1597
1598         btrfs_read_buffer(eb, gen);
1599
1600         level = btrfs_header_level(eb);
1601
1602         if (level != 0)
1603                 return 0;
1604
1605         path = btrfs_alloc_path();
1606         BUG_ON(!path);
1607
1608         nritems = btrfs_header_nritems(eb);
1609         for (i = 0; i < nritems; i++) {
1610                 btrfs_item_key_to_cpu(eb, &key, i);
1611                 item_size = btrfs_item_size_nr(eb, i);
1612
1613                 /* inode keys are done during the first stage */
1614                 if (key.type == BTRFS_INODE_ITEM_KEY &&
1615                     wc->stage == LOG_WALK_REPLAY_INODES) {
1616                         struct inode *inode;
1617                         struct btrfs_inode_item *inode_item;
1618                         u32 mode;
1619
1620                         inode_item = btrfs_item_ptr(eb, i,
1621                                             struct btrfs_inode_item);
1622                         mode = btrfs_inode_mode(eb, inode_item);
1623                         if (S_ISDIR(mode)) {
1624                                 ret = replay_dir_deletes(wc->trans,
1625                                          root, log, path, key.objectid);
1626                                 BUG_ON(ret);
1627                         }
1628                         ret = overwrite_item(wc->trans, root, path,
1629                                              eb, i, &key);
1630                         BUG_ON(ret);
1631
1632                         /* for regular files, truncate away
1633                          * extents past the new EOF
1634                          */
1635                         if (S_ISREG(mode)) {
1636                                 inode = read_one_inode(root,
1637                                                        key.objectid);
1638                                 BUG_ON(!inode);
1639
1640                                 ret = btrfs_truncate_inode_items(wc->trans,
1641                                         root, inode, inode->i_size,
1642                                         BTRFS_EXTENT_DATA_KEY);
1643                                 BUG_ON(ret);
1644                                 iput(inode);
1645                         }
1646                         ret = link_to_fixup_dir(wc->trans, root,
1647                                                 path, key.objectid);
1648                         BUG_ON(ret);
1649                 }
1650                 if (wc->stage < LOG_WALK_REPLAY_ALL)
1651                         continue;
1652
1653                 /* these keys are simply copied */
1654                 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1655                         ret = overwrite_item(wc->trans, root, path,
1656                                              eb, i, &key);
1657                         BUG_ON(ret);
1658                 } else if (key.type == BTRFS_INODE_REF_KEY) {
1659                         ret = add_inode_ref(wc->trans, root, log, path,
1660                                             eb, i, &key);
1661                         BUG_ON(ret && ret != -ENOENT);
1662                 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1663                         ret = replay_one_extent(wc->trans, root, path,
1664                                                 eb, i, &key);
1665                         BUG_ON(ret);
1666                 } else if (key.type == BTRFS_CSUM_ITEM_KEY) {
1667                         ret = replay_one_csum(wc->trans, root, path,
1668                                               eb, i, &key);
1669                         BUG_ON(ret);
1670                 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1671                            key.type == BTRFS_DIR_INDEX_KEY) {
1672                         ret = replay_one_dir_item(wc->trans, root, path,
1673                                                   eb, i, &key);
1674                         BUG_ON(ret);
1675                 }
1676         }
1677         btrfs_free_path(path);
1678         return 0;
1679 }
1680
1681 static int noinline walk_down_log_tree(struct btrfs_trans_handle *trans,
1682                                    struct btrfs_root *root,
1683                                    struct btrfs_path *path, int *level,
1684                                    struct walk_control *wc)
1685 {
1686         u64 root_owner;
1687         u64 root_gen;
1688         u64 bytenr;
1689         u64 ptr_gen;
1690         struct extent_buffer *next;
1691         struct extent_buffer *cur;
1692         struct extent_buffer *parent;
1693         u32 blocksize;
1694         int ret = 0;
1695
1696         WARN_ON(*level < 0);
1697         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1698
1699         while(*level > 0) {
1700                 WARN_ON(*level < 0);
1701                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1702                 cur = path->nodes[*level];
1703
1704                 if (btrfs_header_level(cur) != *level)
1705                         WARN_ON(1);
1706
1707                 if (path->slots[*level] >=
1708                     btrfs_header_nritems(cur))
1709                         break;
1710
1711                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1712                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1713                 blocksize = btrfs_level_size(root, *level - 1);
1714
1715                 parent = path->nodes[*level];
1716                 root_owner = btrfs_header_owner(parent);
1717                 root_gen = btrfs_header_generation(parent);
1718
1719                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1720
1721                 wc->process_func(root, next, wc, ptr_gen);
1722
1723                 if (*level == 1) {
1724                         path->slots[*level]++;
1725                         if (wc->free) {
1726                                 btrfs_read_buffer(next, ptr_gen);
1727
1728                                 btrfs_tree_lock(next);
1729                                 clean_tree_block(trans, root, next);
1730                                 btrfs_wait_tree_block_writeback(next);
1731                                 btrfs_tree_unlock(next);
1732
1733                                 ret = btrfs_drop_leaf_ref(trans, root, next);
1734                                 BUG_ON(ret);
1735
1736                                 WARN_ON(root_owner !=
1737                                         BTRFS_TREE_LOG_OBJECTID);
1738                                 ret = btrfs_free_reserved_extent(root,
1739                                                          bytenr, blocksize);
1740                                 BUG_ON(ret);
1741                         }
1742                         free_extent_buffer(next);
1743                         continue;
1744                 }
1745                 btrfs_read_buffer(next, ptr_gen);
1746
1747                 WARN_ON(*level <= 0);
1748                 if (path->nodes[*level-1])
1749                         free_extent_buffer(path->nodes[*level-1]);
1750                 path->nodes[*level-1] = next;
1751                 *level = btrfs_header_level(next);
1752                 path->slots[*level] = 0;
1753                 cond_resched();
1754         }
1755         WARN_ON(*level < 0);
1756         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1757
1758         if (path->nodes[*level] == root->node) {
1759                 parent = path->nodes[*level];
1760         } else {
1761                 parent = path->nodes[*level + 1];
1762         }
1763         bytenr = path->nodes[*level]->start;
1764
1765         blocksize = btrfs_level_size(root, *level);
1766         root_owner = btrfs_header_owner(parent);
1767         root_gen = btrfs_header_generation(parent);
1768
1769         wc->process_func(root, path->nodes[*level], wc,
1770                          btrfs_header_generation(path->nodes[*level]));
1771
1772         if (wc->free) {
1773                 next = path->nodes[*level];
1774                 btrfs_tree_lock(next);
1775                 clean_tree_block(trans, root, next);
1776                 btrfs_wait_tree_block_writeback(next);
1777                 btrfs_tree_unlock(next);
1778
1779                 if (*level == 0) {
1780                         ret = btrfs_drop_leaf_ref(trans, root, next);
1781                         BUG_ON(ret);
1782                 }
1783                 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1784                 ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
1785                 BUG_ON(ret);
1786         }
1787         free_extent_buffer(path->nodes[*level]);
1788         path->nodes[*level] = NULL;
1789         *level += 1;
1790
1791         cond_resched();
1792         return 0;
1793 }
1794
1795 static int noinline walk_up_log_tree(struct btrfs_trans_handle *trans,
1796                                  struct btrfs_root *root,
1797                                  struct btrfs_path *path, int *level,
1798                                  struct walk_control *wc)
1799 {
1800         u64 root_owner;
1801         u64 root_gen;
1802         int i;
1803         int slot;
1804         int ret;
1805
1806         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1807                 slot = path->slots[i];
1808                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1809                         struct extent_buffer *node;
1810                         node = path->nodes[i];
1811                         path->slots[i]++;
1812                         *level = i;
1813                         WARN_ON(*level == 0);
1814                         return 0;
1815                 } else {
1816                         struct extent_buffer *parent;
1817                         if (path->nodes[*level] == root->node)
1818                                 parent = path->nodes[*level];
1819                         else
1820                                 parent = path->nodes[*level + 1];
1821
1822                         root_owner = btrfs_header_owner(parent);
1823                         root_gen = btrfs_header_generation(parent);
1824                         wc->process_func(root, path->nodes[*level], wc,
1825                                  btrfs_header_generation(path->nodes[*level]));
1826                         if (wc->free) {
1827                                 struct extent_buffer *next;
1828
1829                                 next = path->nodes[*level];
1830
1831                                 btrfs_tree_lock(next);
1832                                 clean_tree_block(trans, root, next);
1833                                 btrfs_wait_tree_block_writeback(next);
1834                                 btrfs_tree_unlock(next);
1835
1836                                 if (*level == 0) {
1837                                         ret = btrfs_drop_leaf_ref(trans, root,
1838                                                                   next);
1839                                         BUG_ON(ret);
1840                                 }
1841
1842                                 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1843                                 ret = btrfs_free_reserved_extent(root,
1844                                                 path->nodes[*level]->start,
1845                                                 path->nodes[*level]->len);
1846                                 BUG_ON(ret);
1847                         }
1848                         free_extent_buffer(path->nodes[*level]);
1849                         path->nodes[*level] = NULL;
1850                         *level = i + 1;
1851                 }
1852         }
1853         return 1;
1854 }
1855
1856 /*
1857  * drop the reference count on the tree rooted at 'snap'.  This traverses
1858  * the tree freeing any blocks that have a ref count of zero after being
1859  * decremented.
1860  */
1861 static int walk_log_tree(struct btrfs_trans_handle *trans,
1862                          struct btrfs_root *log, struct walk_control *wc)
1863 {
1864         int ret = 0;
1865         int wret;
1866         int level;
1867         struct btrfs_path *path;
1868         int i;
1869         int orig_level;
1870
1871         path = btrfs_alloc_path();
1872         BUG_ON(!path);
1873
1874         level = btrfs_header_level(log->node);
1875         orig_level = level;
1876         path->nodes[level] = log->node;
1877         extent_buffer_get(log->node);
1878         path->slots[level] = 0;
1879
1880         while(1) {
1881                 wret = walk_down_log_tree(trans, log, path, &level, wc);
1882                 if (wret > 0)
1883                         break;
1884                 if (wret < 0)
1885                         ret = wret;
1886
1887                 wret = walk_up_log_tree(trans, log, path, &level, wc);
1888                 if (wret > 0)
1889                         break;
1890                 if (wret < 0)
1891                         ret = wret;
1892         }
1893
1894         /* was the root node processed? if not, catch it here */
1895         if (path->nodes[orig_level]) {
1896                 wc->process_func(log, path->nodes[orig_level], wc,
1897                          btrfs_header_generation(path->nodes[orig_level]));
1898                 if (wc->free) {
1899                         struct extent_buffer *next;
1900
1901                         next = path->nodes[orig_level];
1902
1903                         btrfs_tree_lock(next);
1904                         clean_tree_block(trans, log, next);
1905                         btrfs_wait_tree_block_writeback(next);
1906                         btrfs_tree_unlock(next);
1907
1908                         if (orig_level == 0) {
1909                                 ret = btrfs_drop_leaf_ref(trans, log,
1910                                                           next);
1911                                 BUG_ON(ret);
1912                         }
1913                         WARN_ON(log->root_key.objectid !=
1914                                 BTRFS_TREE_LOG_OBJECTID);
1915                         ret = btrfs_free_reserved_extent(log, next->start,
1916                                                          next->len);
1917                         BUG_ON(ret);
1918                 }
1919         }
1920
1921         for (i = 0; i <= orig_level; i++) {
1922                 if (path->nodes[i]) {
1923                         free_extent_buffer(path->nodes[i]);
1924                         path->nodes[i] = NULL;
1925                 }
1926         }
1927         btrfs_free_path(path);
1928         if (wc->free)
1929                 free_extent_buffer(log->node);
1930         return ret;
1931 }
1932
1933 int wait_log_commit(struct btrfs_root *log)
1934 {
1935         DEFINE_WAIT(wait);
1936         u64 transid = log->fs_info->tree_log_transid;
1937
1938         do {
1939                 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1940                                 TASK_UNINTERRUPTIBLE);
1941                 mutex_unlock(&log->fs_info->tree_log_mutex);
1942                 if (atomic_read(&log->fs_info->tree_log_commit))
1943                         schedule();
1944                 finish_wait(&log->fs_info->tree_log_wait, &wait);
1945                 mutex_lock(&log->fs_info->tree_log_mutex);
1946         } while(transid == log->fs_info->tree_log_transid &&
1947                 atomic_read(&log->fs_info->tree_log_commit));
1948         return 0;
1949 }
1950
1951 /*
1952  * btrfs_sync_log does sends a given tree log down to the disk and
1953  * updates the super blocks to record it.  When this call is done,
1954  * you know that any inodes previously logged are safely on disk
1955  */
1956 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1957                    struct btrfs_root *root)
1958 {
1959         int ret;
1960         unsigned long batch;
1961         struct btrfs_root *log = root->log_root;
1962
1963         mutex_lock(&log->fs_info->tree_log_mutex);
1964         if (atomic_read(&log->fs_info->tree_log_commit)) {
1965                 wait_log_commit(log);
1966                 goto out;
1967         }
1968         atomic_set(&log->fs_info->tree_log_commit, 1);
1969
1970         while(1) {
1971                 batch = log->fs_info->tree_log_batch;
1972                 mutex_unlock(&log->fs_info->tree_log_mutex);
1973                 schedule_timeout_uninterruptible(1);
1974                 mutex_lock(&log->fs_info->tree_log_mutex);
1975
1976                 while(atomic_read(&log->fs_info->tree_log_writers)) {
1977                         DEFINE_WAIT(wait);
1978                         prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1979                                         TASK_UNINTERRUPTIBLE);
1980                         mutex_unlock(&log->fs_info->tree_log_mutex);
1981                         if (atomic_read(&log->fs_info->tree_log_writers))
1982                                 schedule();
1983                         mutex_lock(&log->fs_info->tree_log_mutex);
1984                         finish_wait(&log->fs_info->tree_log_wait, &wait);
1985                 }
1986                 if (batch == log->fs_info->tree_log_batch)
1987                         break;
1988         }
1989
1990         ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
1991         BUG_ON(ret);
1992         ret = btrfs_write_and_wait_marked_extents(root->fs_info->log_root_tree,
1993                                &root->fs_info->log_root_tree->dirty_log_pages);
1994         BUG_ON(ret);
1995
1996         btrfs_set_super_log_root(&root->fs_info->super_for_commit,
1997                                  log->fs_info->log_root_tree->node->start);
1998         btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
1999                        btrfs_header_level(log->fs_info->log_root_tree->node));
2000
2001         write_ctree_super(trans, log->fs_info->tree_root);
2002         log->fs_info->tree_log_transid++;
2003         log->fs_info->tree_log_batch = 0;
2004         atomic_set(&log->fs_info->tree_log_commit, 0);
2005         smp_mb();
2006         if (waitqueue_active(&log->fs_info->tree_log_wait))
2007                 wake_up(&log->fs_info->tree_log_wait);
2008 out:
2009         mutex_unlock(&log->fs_info->tree_log_mutex);
2010         return 0;
2011
2012 }
2013
2014 /* * free all the extents used by the tree log.  This should be called
2015  * at commit time of the full transaction
2016  */
2017 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2018 {
2019         int ret;
2020         struct btrfs_root *log;
2021         struct key;
2022         u64 start;
2023         u64 end;
2024         struct walk_control wc = {
2025                 .free = 1,
2026                 .process_func = process_one_buffer
2027         };
2028
2029         if (!root->log_root)
2030                 return 0;
2031
2032         log = root->log_root;
2033         ret = walk_log_tree(trans, log, &wc);
2034         BUG_ON(ret);
2035
2036         while(1) {
2037                 ret = find_first_extent_bit(&log->dirty_log_pages,
2038                                     0, &start, &end, EXTENT_DIRTY);
2039                 if (ret)
2040                         break;
2041
2042                 clear_extent_dirty(&log->dirty_log_pages,
2043                                    start, end, GFP_NOFS);
2044         }
2045
2046         log = root->log_root;
2047         ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2048                              &log->root_key);
2049         BUG_ON(ret);
2050         root->log_root = NULL;
2051         kfree(root->log_root);
2052         return 0;
2053 }
2054
2055 /*
2056  * helper function to update the item for a given subvolumes log root
2057  * in the tree of log roots
2058  */
2059 static int update_log_root(struct btrfs_trans_handle *trans,
2060                            struct btrfs_root *log)
2061 {
2062         u64 bytenr = btrfs_root_bytenr(&log->root_item);
2063         int ret;
2064
2065         if (log->node->start == bytenr)
2066                 return 0;
2067
2068         btrfs_set_root_bytenr(&log->root_item, log->node->start);
2069         btrfs_set_root_generation(&log->root_item, trans->transid);
2070         btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
2071         ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2072                                 &log->root_key, &log->root_item);
2073         BUG_ON(ret);
2074         return ret;
2075 }
2076
2077 /*
2078  * If both a file and directory are logged, and unlinks or renames are
2079  * mixed in, we have a few interesting corners:
2080  *
2081  * create file X in dir Y
2082  * link file X to X.link in dir Y
2083  * fsync file X
2084  * unlink file X but leave X.link
2085  * fsync dir Y
2086  *
2087  * After a crash we would expect only X.link to exist.  But file X
2088  * didn't get fsync'd again so the log has back refs for X and X.link.
2089  *
2090  * We solve this by removing directory entries and inode backrefs from the
2091  * log when a file that was logged in the current transaction is
2092  * unlinked.  Any later fsync will include the updated log entries, and
2093  * we'll be able to reconstruct the proper directory items from backrefs.
2094  *
2095  * This optimizations allows us to avoid relogging the entire inode
2096  * or the entire directory.
2097  */
2098 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2099                                  struct btrfs_root *root,
2100                                  const char *name, int name_len,
2101                                  struct inode *dir, u64 index)
2102 {
2103         struct btrfs_root *log;
2104         struct btrfs_dir_item *di;
2105         struct btrfs_path *path;
2106         int ret;
2107         int bytes_del = 0;
2108
2109         if (BTRFS_I(dir)->logged_trans < trans->transid)
2110                 return 0;
2111
2112         ret = join_running_log_trans(root);
2113         if (ret)
2114                 return 0;
2115
2116         mutex_lock(&BTRFS_I(dir)->log_mutex);
2117
2118         log = root->log_root;
2119         path = btrfs_alloc_path();
2120         di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2121                                    name, name_len, -1);
2122         if (di && !IS_ERR(di)) {
2123                 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2124                 bytes_del += name_len;
2125                 BUG_ON(ret);
2126         }
2127         btrfs_release_path(log, path);
2128         di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2129                                          index, name, name_len, -1);
2130         if (di && !IS_ERR(di)) {
2131                 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2132                 bytes_del += name_len;
2133                 BUG_ON(ret);
2134         }
2135
2136         /* update the directory size in the log to reflect the names
2137          * we have removed
2138          */
2139         if (bytes_del) {
2140                 struct btrfs_key key;
2141
2142                 key.objectid = dir->i_ino;
2143                 key.offset = 0;
2144                 key.type = BTRFS_INODE_ITEM_KEY;
2145                 btrfs_release_path(log, path);
2146
2147                 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2148                 if (ret == 0) {
2149                         struct btrfs_inode_item *item;
2150                         u64 i_size;
2151
2152                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2153                                               struct btrfs_inode_item);
2154                         i_size = btrfs_inode_size(path->nodes[0], item);
2155                         if (i_size > bytes_del)
2156                                 i_size -= bytes_del;
2157                         else
2158                                 i_size = 0;
2159                         btrfs_set_inode_size(path->nodes[0], item, i_size);
2160                         btrfs_mark_buffer_dirty(path->nodes[0]);
2161                 } else
2162                         ret = 0;
2163                 btrfs_release_path(log, path);
2164         }
2165
2166         btrfs_free_path(path);
2167         mutex_unlock(&BTRFS_I(dir)->log_mutex);
2168         end_log_trans(root);
2169
2170         return 0;
2171 }
2172
2173 /* see comments for btrfs_del_dir_entries_in_log */
2174 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2175                                struct btrfs_root *root,
2176                                const char *name, int name_len,
2177                                struct inode *inode, u64 dirid)
2178 {
2179         struct btrfs_root *log;
2180         u64 index;
2181         int ret;
2182
2183         if (BTRFS_I(inode)->logged_trans < trans->transid)
2184                 return 0;
2185
2186         ret = join_running_log_trans(root);
2187         if (ret)
2188                 return 0;
2189         log = root->log_root;
2190         mutex_lock(&BTRFS_I(inode)->log_mutex);
2191
2192         ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2193                                   dirid, &index);
2194         mutex_unlock(&BTRFS_I(inode)->log_mutex);
2195         end_log_trans(root);
2196
2197         return ret;
2198 }
2199
2200 /*
2201  * creates a range item in the log for 'dirid'.  first_offset and
2202  * last_offset tell us which parts of the key space the log should
2203  * be considered authoritative for.
2204  */
2205 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2206                                        struct btrfs_root *log,
2207                                        struct btrfs_path *path,
2208                                        int key_type, u64 dirid,
2209                                        u64 first_offset, u64 last_offset)
2210 {
2211         int ret;
2212         struct btrfs_key key;
2213         struct btrfs_dir_log_item *item;
2214
2215         key.objectid = dirid;
2216         key.offset = first_offset;
2217         if (key_type == BTRFS_DIR_ITEM_KEY)
2218                 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2219         else
2220                 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2221         ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2222         BUG_ON(ret);
2223
2224         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2225                               struct btrfs_dir_log_item);
2226         btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2227         btrfs_mark_buffer_dirty(path->nodes[0]);
2228         btrfs_release_path(log, path);
2229         return 0;
2230 }
2231
2232 /*
2233  * log all the items included in the current transaction for a given
2234  * directory.  This also creates the range items in the log tree required
2235  * to replay anything deleted before the fsync
2236  */
2237 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2238                           struct btrfs_root *root, struct inode *inode,
2239                           struct btrfs_path *path,
2240                           struct btrfs_path *dst_path, int key_type,
2241                           u64 min_offset, u64 *last_offset_ret)
2242 {
2243         struct btrfs_key min_key;
2244         struct btrfs_key max_key;
2245         struct btrfs_root *log = root->log_root;
2246         struct extent_buffer *src;
2247         int ret;
2248         int i;
2249         int nritems;
2250         u64 first_offset = min_offset;
2251         u64 last_offset = (u64)-1;
2252
2253         log = root->log_root;
2254         max_key.objectid = inode->i_ino;
2255         max_key.offset = (u64)-1;
2256         max_key.type = key_type;
2257
2258         min_key.objectid = inode->i_ino;
2259         min_key.type = key_type;
2260         min_key.offset = min_offset;
2261
2262         path->keep_locks = 1;
2263
2264         ret = btrfs_search_forward(root, &min_key, &max_key,
2265                                    path, 0, trans->transid);
2266
2267         /*
2268          * we didn't find anything from this transaction, see if there
2269          * is anything at all
2270          */
2271         if (ret != 0 || min_key.objectid != inode->i_ino ||
2272             min_key.type != key_type) {
2273                 min_key.objectid = inode->i_ino;
2274                 min_key.type = key_type;
2275                 min_key.offset = (u64)-1;
2276                 btrfs_release_path(root, path);
2277                 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2278                 if (ret < 0) {
2279                         btrfs_release_path(root, path);
2280                         return ret;
2281                 }
2282                 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2283
2284                 /* if ret == 0 there are items for this type,
2285                  * create a range to tell us the last key of this type.
2286                  * otherwise, there are no items in this directory after
2287                  * *min_offset, and we create a range to indicate that.
2288                  */
2289                 if (ret == 0) {
2290                         struct btrfs_key tmp;
2291                         btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2292                                               path->slots[0]);
2293                         if (key_type == tmp.type) {
2294                                 first_offset = max(min_offset, tmp.offset) + 1;
2295                         }
2296                 }
2297                 goto done;
2298         }
2299
2300         /* go backward to find any previous key */
2301         ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2302         if (ret == 0) {
2303                 struct btrfs_key tmp;
2304                 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2305                 if (key_type == tmp.type) {
2306                         first_offset = tmp.offset;
2307                         ret = overwrite_item(trans, log, dst_path,
2308                                              path->nodes[0], path->slots[0],
2309                                              &tmp);
2310                 }
2311         }
2312         btrfs_release_path(root, path);
2313
2314         /* find the first key from this transaction again */
2315         ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2316         if (ret != 0) {
2317                 WARN_ON(1);
2318                 goto done;
2319         }
2320
2321         /*
2322          * we have a block from this transaction, log every item in it
2323          * from our directory
2324          */
2325         while(1) {
2326                 struct btrfs_key tmp;
2327                 src = path->nodes[0];
2328                 nritems = btrfs_header_nritems(src);
2329                 for (i = path->slots[0]; i < nritems; i++) {
2330                         btrfs_item_key_to_cpu(src, &min_key, i);
2331
2332                         if (min_key.objectid != inode->i_ino ||
2333                             min_key.type != key_type)
2334                                 goto done;
2335                         ret = overwrite_item(trans, log, dst_path, src, i,
2336                                              &min_key);
2337                         BUG_ON(ret);
2338                 }
2339                 path->slots[0] = nritems;
2340
2341                 /*
2342                  * look ahead to the next item and see if it is also
2343                  * from this directory and from this transaction
2344                  */
2345                 ret = btrfs_next_leaf(root, path);
2346                 if (ret == 1) {
2347                         last_offset = (u64)-1;
2348                         goto done;
2349                 }
2350                 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2351                 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2352                         last_offset = (u64)-1;
2353                         goto done;
2354                 }
2355                 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2356                         ret = overwrite_item(trans, log, dst_path,
2357                                              path->nodes[0], path->slots[0],
2358                                              &tmp);
2359
2360                         BUG_ON(ret);
2361                         last_offset = tmp.offset;
2362                         goto done;
2363                 }
2364         }
2365 done:
2366         *last_offset_ret = last_offset;
2367         btrfs_release_path(root, path);
2368         btrfs_release_path(log, dst_path);
2369
2370         /* insert the log range keys to indicate where the log is valid */
2371         ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2372                                  first_offset, last_offset);
2373         BUG_ON(ret);
2374         return 0;
2375 }
2376
2377 /*
2378  * logging directories is very similar to logging inodes, We find all the items
2379  * from the current transaction and write them to the log.
2380  *
2381  * The recovery code scans the directory in the subvolume, and if it finds a
2382  * key in the range logged that is not present in the log tree, then it means
2383  * that dir entry was unlinked during the transaction.
2384  *
2385  * In order for that scan to work, we must include one key smaller than
2386  * the smallest logged by this transaction and one key larger than the largest
2387  * key logged by this transaction.
2388  */
2389 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2390                           struct btrfs_root *root, struct inode *inode,
2391                           struct btrfs_path *path,
2392                           struct btrfs_path *dst_path)
2393 {
2394         u64 min_key;
2395         u64 max_key;
2396         int ret;
2397         int key_type = BTRFS_DIR_ITEM_KEY;
2398
2399 again:
2400         min_key = 0;
2401         max_key = 0;
2402         while(1) {
2403                 ret = log_dir_items(trans, root, inode, path,
2404                                     dst_path, key_type, min_key,
2405                                     &max_key);
2406                 BUG_ON(ret);
2407                 if (max_key == (u64)-1)
2408                         break;
2409                 min_key = max_key + 1;
2410         }
2411
2412         if (key_type == BTRFS_DIR_ITEM_KEY) {
2413                 key_type = BTRFS_DIR_INDEX_KEY;
2414                 goto again;
2415         }
2416         return 0;
2417 }
2418
2419 /*
2420  * a helper function to drop items from the log before we relog an
2421  * inode.  max_key_type indicates the highest item type to remove.
2422  * This cannot be run for file data extents because it does not
2423  * free the extents they point to.
2424  */
2425 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2426                                   struct btrfs_root *log,
2427                                   struct btrfs_path *path,
2428                                   u64 objectid, int max_key_type)
2429 {
2430         int ret;
2431         struct btrfs_key key;
2432         struct btrfs_key found_key;
2433
2434         key.objectid = objectid;
2435         key.type = max_key_type;
2436         key.offset = (u64)-1;
2437
2438         while(1) {
2439                 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2440
2441                 if (ret != 1)
2442                         break;
2443
2444                 if (path->slots[0] == 0)
2445                         break;
2446
2447                 path->slots[0]--;
2448                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2449                                       path->slots[0]);
2450
2451                 if (found_key.objectid != objectid)
2452                         break;
2453
2454                 ret = btrfs_del_item(trans, log, path);
2455                 BUG_ON(ret);
2456                 btrfs_release_path(log, path);
2457         }
2458         btrfs_release_path(log, path);
2459         return 0;
2460 }
2461
2462 static noinline int copy_items(struct btrfs_trans_handle *trans,
2463                                struct btrfs_root *log,
2464                                struct btrfs_path *dst_path,
2465                                struct extent_buffer *src,
2466                                int start_slot, int nr, int inode_only)
2467 {
2468         unsigned long src_offset;
2469         unsigned long dst_offset;
2470         struct btrfs_file_extent_item *extent;
2471         struct btrfs_inode_item *inode_item;
2472         int ret;
2473         struct btrfs_key *ins_keys;
2474         u32 *ins_sizes;
2475         char *ins_data;
2476         int i;
2477
2478         ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2479                            nr * sizeof(u32), GFP_NOFS);
2480         ins_sizes = (u32 *)ins_data;
2481         ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2482
2483         for (i = 0; i < nr; i++) {
2484                 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2485                 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2486         }
2487         ret = btrfs_insert_empty_items(trans, log, dst_path,
2488                                        ins_keys, ins_sizes, nr);
2489         BUG_ON(ret);
2490
2491         for (i = 0; i < nr; i++) {
2492                 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2493                                                    dst_path->slots[0]);
2494
2495                 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2496
2497                 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2498                                    src_offset, ins_sizes[i]);
2499
2500                 if (inode_only == LOG_INODE_EXISTS &&
2501                     ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2502                         inode_item = btrfs_item_ptr(dst_path->nodes[0],
2503                                                     dst_path->slots[0],
2504                                                     struct btrfs_inode_item);
2505                         btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2506
2507                         /* set the generation to zero so the recover code
2508                          * can tell the difference between an logging
2509                          * just to say 'this inode exists' and a logging
2510                          * to say 'update this inode with these values'
2511                          */
2512                         btrfs_set_inode_generation(dst_path->nodes[0],
2513                                                    inode_item, 0);
2514                 }
2515                 /* take a reference on file data extents so that truncates
2516                  * or deletes of this inode don't have to relog the inode
2517                  * again
2518                  */
2519                 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2520                         int found_type;
2521                         extent = btrfs_item_ptr(src, start_slot + i,
2522                                                 struct btrfs_file_extent_item);
2523
2524                         found_type = btrfs_file_extent_type(src, extent);
2525                         if (found_type == BTRFS_FILE_EXTENT_REG) {
2526                                 u64 ds = btrfs_file_extent_disk_bytenr(src,
2527                                                                    extent);
2528                                 u64 dl = btrfs_file_extent_disk_num_bytes(src,
2529                                                                       extent);
2530                                 /* ds == 0 is a hole */
2531                                 if (ds != 0) {
2532                                         ret = btrfs_inc_extent_ref(trans, log,
2533                                                    ds, dl,
2534                                                    dst_path->nodes[0]->start,
2535                                                    BTRFS_TREE_LOG_OBJECTID,
2536                                                    trans->transid,
2537                                                    ins_keys[i].objectid);
2538                                         BUG_ON(ret);
2539                                 }
2540                         }
2541                 }
2542                 dst_path->slots[0]++;
2543         }
2544
2545         btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2546         btrfs_release_path(log, dst_path);
2547         kfree(ins_data);
2548         return 0;
2549 }
2550
2551 /* log a single inode in the tree log.
2552  * At least one parent directory for this inode must exist in the tree
2553  * or be logged already.
2554  *
2555  * Any items from this inode changed by the current transaction are copied
2556  * to the log tree.  An extra reference is taken on any extents in this
2557  * file, allowing us to avoid a whole pile of corner cases around logging
2558  * blocks that have been removed from the tree.
2559  *
2560  * See LOG_INODE_ALL and related defines for a description of what inode_only
2561  * does.
2562  *
2563  * This handles both files and directories.
2564  */
2565 static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
2566                              struct btrfs_root *root, struct inode *inode,
2567                              int inode_only)
2568 {
2569         struct btrfs_path *path;
2570         struct btrfs_path *dst_path;
2571         struct btrfs_key min_key;
2572         struct btrfs_key max_key;
2573         struct btrfs_root *log = root->log_root;
2574         struct extent_buffer *src = NULL;
2575         u32 size;
2576         int ret;
2577         int nritems;
2578         int ins_start_slot = 0;
2579         int ins_nr;
2580
2581         log = root->log_root;
2582
2583         path = btrfs_alloc_path();
2584         dst_path = btrfs_alloc_path();
2585
2586         min_key.objectid = inode->i_ino;
2587         min_key.type = BTRFS_INODE_ITEM_KEY;
2588         min_key.offset = 0;
2589
2590         max_key.objectid = inode->i_ino;
2591         if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2592                 max_key.type = BTRFS_XATTR_ITEM_KEY;
2593         else
2594                 max_key.type = (u8)-1;
2595         max_key.offset = (u64)-1;
2596
2597         /*
2598          * if this inode has already been logged and we're in inode_only
2599          * mode, we don't want to delete the things that have already
2600          * been written to the log.
2601          *
2602          * But, if the inode has been through an inode_only log,
2603          * the logged_trans field is not set.  This allows us to catch
2604          * any new names for this inode in the backrefs by logging it
2605          * again
2606          */
2607         if (inode_only == LOG_INODE_EXISTS &&
2608             BTRFS_I(inode)->logged_trans == trans->transid) {
2609                 btrfs_free_path(path);
2610                 btrfs_free_path(dst_path);
2611                 goto out;
2612         }
2613         mutex_lock(&BTRFS_I(inode)->log_mutex);
2614
2615         /*
2616          * a brute force approach to making sure we get the most uptodate
2617          * copies of everything.
2618          */
2619         if (S_ISDIR(inode->i_mode)) {
2620                 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2621
2622                 if (inode_only == LOG_INODE_EXISTS)
2623                         max_key_type = BTRFS_XATTR_ITEM_KEY;
2624                 ret = drop_objectid_items(trans, log, path,
2625                                           inode->i_ino, max_key_type);
2626         } else {
2627                 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2628         }
2629         BUG_ON(ret);
2630         path->keep_locks = 1;
2631
2632         while(1) {
2633                 ins_nr = 0;
2634                 ret = btrfs_search_forward(root, &min_key, &max_key,
2635                                            path, 0, trans->transid);
2636                 if (ret != 0)
2637                         break;
2638 again:
2639                 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2640                 if (min_key.objectid != inode->i_ino)
2641                         break;
2642                 if (min_key.type > max_key.type)
2643                         break;
2644
2645                 src = path->nodes[0];
2646                 size = btrfs_item_size_nr(src, path->slots[0]);
2647                 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2648                         ins_nr++;
2649                         goto next_slot;
2650                 } else if (!ins_nr) {
2651                         ins_start_slot = path->slots[0];
2652                         ins_nr = 1;
2653                         goto next_slot;
2654                 }
2655
2656                 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2657                                  ins_nr, inode_only);
2658                 BUG_ON(ret);
2659                 ins_nr = 1;
2660                 ins_start_slot = path->slots[0];
2661 next_slot:
2662
2663                 nritems = btrfs_header_nritems(path->nodes[0]);
2664                 path->slots[0]++;
2665                 if (path->slots[0] < nritems) {
2666                         btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2667                                               path->slots[0]);
2668                         goto again;
2669                 }
2670                 if (ins_nr) {
2671                         ret = copy_items(trans, log, dst_path, src,
2672                                          ins_start_slot,
2673                                          ins_nr, inode_only);
2674                         BUG_ON(ret);
2675                         ins_nr = 0;
2676                 }
2677                 btrfs_release_path(root, path);
2678
2679                 if (min_key.offset < (u64)-1)
2680                         min_key.offset++;
2681                 else if (min_key.type < (u8)-1)
2682                         min_key.type++;
2683                 else if (min_key.objectid < (u64)-1)
2684                         min_key.objectid++;
2685                 else
2686                         break;
2687         }
2688         if (ins_nr) {
2689                 ret = copy_items(trans, log, dst_path, src,
2690                                  ins_start_slot,
2691                                  ins_nr, inode_only);
2692                 BUG_ON(ret);
2693                 ins_nr = 0;
2694         }
2695         WARN_ON(ins_nr);
2696         if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2697                 btrfs_release_path(root, path);
2698                 btrfs_release_path(log, dst_path);
2699                 BTRFS_I(inode)->log_dirty_trans = 0;
2700                 ret = log_directory_changes(trans, root, inode, path, dst_path);
2701                 BUG_ON(ret);
2702         }
2703         BTRFS_I(inode)->logged_trans = trans->transid;
2704         mutex_unlock(&BTRFS_I(inode)->log_mutex);
2705
2706         btrfs_free_path(path);
2707         btrfs_free_path(dst_path);
2708
2709         mutex_lock(&root->fs_info->tree_log_mutex);
2710         ret = update_log_root(trans, log);
2711         BUG_ON(ret);
2712         mutex_unlock(&root->fs_info->tree_log_mutex);
2713 out:
2714         return 0;
2715 }
2716
2717 int btrfs_log_inode(struct btrfs_trans_handle *trans,
2718                     struct btrfs_root *root, struct inode *inode,
2719                     int inode_only)
2720 {
2721         int ret;
2722
2723         start_log_trans(trans, root);
2724         ret = __btrfs_log_inode(trans, root, inode, inode_only);
2725         end_log_trans(root);
2726         return ret;
2727 }
2728
2729 /*
2730  * helper function around btrfs_log_inode to make sure newly created
2731  * parent directories also end up in the log.  A minimal inode and backref
2732  * only logging is done of any parent directories that are older than
2733  * the last committed transaction
2734  */
2735 int btrfs_log_dentry(struct btrfs_trans_handle *trans,
2736                     struct btrfs_root *root, struct dentry *dentry)
2737 {
2738         int inode_only = LOG_INODE_ALL;
2739         struct super_block *sb;
2740         int ret;
2741
2742         start_log_trans(trans, root);
2743         sb = dentry->d_inode->i_sb;
2744         while(1) {
2745                 ret = __btrfs_log_inode(trans, root, dentry->d_inode,
2746                                         inode_only);
2747                 BUG_ON(ret);
2748                 inode_only = LOG_INODE_EXISTS;
2749
2750                 dentry = dentry->d_parent;
2751                 if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb)
2752                         break;
2753
2754                 if (BTRFS_I(dentry->d_inode)->generation <=
2755                     root->fs_info->last_trans_committed)
2756                         break;
2757         }
2758         end_log_trans(root);
2759         return 0;
2760 }
2761
2762 /*
2763  * it is not safe to log dentry if the chunk root has added new
2764  * chunks.  This returns 0 if the dentry was logged, and 1 otherwise.
2765  * If this returns 1, you must commit the transaction to safely get your
2766  * data on disk.
2767  */
2768 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2769                           struct btrfs_root *root, struct dentry *dentry)
2770 {
2771         u64 gen;
2772         gen = root->fs_info->last_trans_new_blockgroup;
2773         if (gen > root->fs_info->last_trans_committed)
2774                 return 1;
2775         else
2776                 return btrfs_log_dentry(trans, root, dentry);
2777 }
2778
2779 /*
2780  * should be called during mount to recover any replay any log trees
2781  * from the FS
2782  */
2783 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
2784 {
2785         int ret;
2786         struct btrfs_path *path;
2787         struct btrfs_trans_handle *trans;
2788         struct btrfs_key key;
2789         struct btrfs_key found_key;
2790         struct btrfs_key tmp_key;
2791         struct btrfs_root *log;
2792         struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
2793         u64 highest_inode;
2794         struct walk_control wc = {
2795                 .process_func = process_one_buffer,
2796                 .stage = 0,
2797         };
2798
2799         fs_info->log_root_recovering = 1;
2800         path = btrfs_alloc_path();
2801         BUG_ON(!path);
2802
2803         trans = btrfs_start_transaction(fs_info->tree_root, 1);
2804
2805         wc.trans = trans;
2806         wc.pin = 1;
2807
2808         walk_log_tree(trans, log_root_tree, &wc);
2809
2810 again:
2811         key.objectid = BTRFS_TREE_LOG_OBJECTID;
2812         key.offset = (u64)-1;
2813         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2814
2815         while(1) {
2816                 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
2817                 if (ret < 0)
2818                         break;
2819                 if (ret > 0) {
2820                         if (path->slots[0] == 0)
2821                                 break;
2822                         path->slots[0]--;
2823                 }
2824                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2825                                       path->slots[0]);
2826                 btrfs_release_path(log_root_tree, path);
2827                 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
2828                         break;
2829
2830                 log = btrfs_read_fs_root_no_radix(log_root_tree,
2831                                                   &found_key);
2832                 BUG_ON(!log);
2833
2834
2835                 tmp_key.objectid = found_key.offset;
2836                 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
2837                 tmp_key.offset = (u64)-1;
2838
2839                 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
2840
2841                 BUG_ON(!wc.replay_dest);
2842
2843                 btrfs_record_root_in_trans(wc.replay_dest);
2844                 ret = walk_log_tree(trans, log, &wc);
2845                 BUG_ON(ret);
2846
2847                 if (wc.stage == LOG_WALK_REPLAY_ALL) {
2848                         ret = fixup_inode_link_counts(trans, wc.replay_dest,
2849                                                       path);
2850                         BUG_ON(ret);
2851                 }
2852                 ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
2853                 if (ret == 0) {
2854                         wc.replay_dest->highest_inode = highest_inode;
2855                         wc.replay_dest->last_inode_alloc = highest_inode;
2856                 }
2857
2858                 key.offset = found_key.offset - 1;
2859                 free_extent_buffer(log->node);
2860                 kfree(log);
2861
2862                 if (found_key.offset == 0)
2863                         break;
2864         }
2865         btrfs_release_path(log_root_tree, path);
2866
2867         /* step one is to pin it all, step two is to replay just inodes */
2868         if (wc.pin) {
2869                 wc.pin = 0;
2870                 wc.process_func = replay_one_buffer;
2871                 wc.stage = LOG_WALK_REPLAY_INODES;
2872                 goto again;
2873         }
2874         /* step three is to replay everything */
2875         if (wc.stage < LOG_WALK_REPLAY_ALL) {
2876                 wc.stage++;
2877                 goto again;
2878         }
2879
2880         btrfs_free_path(path);
2881
2882         free_extent_buffer(log_root_tree->node);
2883         log_root_tree->log_root = NULL;
2884         fs_info->log_root_recovering = 0;
2885
2886         /* step 4: commit the transaction, which also unpins the blocks */
2887         btrfs_commit_transaction(trans, fs_info->tree_root);
2888
2889         kfree(log_root_tree);
2890         return 0;
2891 }