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 <linux/slab.h>
  21 #include <linux/list_sort.h>
  22 #include "ctree.h"
  23 #include "transaction.h"
  24 #include "disk-io.h"
  25 #include "locking.h"
  26 #include "print-tree.h"
  27 #include "backref.h"
  28 #include "compat.h"
  29 #include "tree-log.h"
  30 #include "hash.h"
  31
  32 /* magic values for the inode_only field in btrfs_log_inode:
  33  *
  34  * LOG_INODE_ALL means to log everything
  35  * LOG_INODE_EXISTS means to log just enough to recreate the inode
  36  * during log replay
  37  */
  38 #define LOG_INODE_ALL 0
  39 #define LOG_INODE_EXISTS 1
  40
  41 /*
  42  * directory trouble cases
  43  *
  44  * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
  45  * log, we must force a full commit before doing an fsync of the directory
  46  * where the unlink was done.
  47  * ---> record transid of last unlink/rename per directory
  48  *
  49  * mkdir foo/some_dir
  50  * normal commit
  51  * rename foo/some_dir foo2/some_dir
  52  * mkdir foo/some_dir
  53  * fsync foo/some_dir/some_file
  54  *
  55  * The fsync above will unlink the original some_dir without recording
  56  * it in its new location (foo2).  After a crash, some_dir will be gone
  57  * unless the fsync of some_file forces a full commit
  58  *
  59  * 2) we must log any new names for any file or dir that is in the fsync
  60  * log. ---> check inode while renaming/linking.
  61  *
  62  * 2a) we must log any new names for any file or dir during rename
  63  * when the directory they are being removed from was logged.
  64  * ---> check inode and old parent dir during rename
  65  *
  66  *  2a is actually the more important variant.  With the extra logging
  67  *  a crash might unlink the old name without recreating the new one
  68  *
  69  * 3) after a crash, we must go through any directories with a link count
  70  * of zero and redo the rm -rf
  71  *
  72  * mkdir f1/foo
  73  * normal commit
  74  * rm -rf f1/foo
  75  * fsync(f1)
  76  *
  77  * The directory f1 was fully removed from the FS, but fsync was never
  78  * called on f1, only its parent dir.  After a crash the rm -rf must
  79  * be replayed.  This must be able to recurse down the entire
  80  * directory tree.  The inode link count fixup code takes care of the
  81  * ugly details.
  82  */
  83
  84 /*
  85  * stages for the tree walking.  The first
  86  * stage (0) is to only pin down the blocks we find
  87  * the second stage (1) is to make sure that all the inodes
  88  * we find in the log are created in the subvolume.
  89  *
  90  * The last stage is to deal with directories and links and extents
  91  * and all the other fun semantics
  92  */
  93 #define LOG_WALK_PIN_ONLY 0
  94 #define LOG_WALK_REPLAY_INODES 1
  95 #define LOG_WALK_REPLAY_ALL 2
  96
  97 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
  98                              struct btrfs_root *root, struct inode *inode,
  99                              int inode_only);
 100 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
 101                              struct btrfs_root *root,
 102                              struct btrfs_path *path, u64 objectid);
 103 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
 104                                        struct btrfs_root *root,
 105                                        struct btrfs_root *log,
 106                                        struct btrfs_path *path,
 107                                        u64 dirid, int del_all);
 108
 109 /*
 110  * tree logging is a special write ahead log used to make sure that
 111  * fsyncs and O_SYNCs can happen without doing full tree commits.
 112  *
 113  * Full tree commits are expensive because they require commonly
 114  * modified blocks to be recowed, creating many dirty pages in the
 115  * extent tree an 4x-6x higher write load than ext3.
 116  *
 117  * Instead of doing a tree commit on every fsync, we use the
 118  * key ranges and transaction ids to find items for a given file or directory
 119  * that have changed in this transaction.  Those items are copied into
 120  * a special tree (one per subvolume root), that tree is written to disk
 121  * and then the fsync is considered complete.
 122  *
 123  * After a crash, items are copied out of the log-tree back into the
 124  * subvolume tree.  Any file data extents found are recorded in the extent
 125  * allocation tree, and the log-tree freed.
 126  *
 127  * The log tree is read three times, once to pin down all the extents it is
 128  * using in ram and once, once to create all the inodes logged in the tree
 129  * and once to do all the other items.
 130  */
 131
 132 /*
 133  * start a sub transaction and setup the log tree
 134  * this increments the log tree writer count to make the people
 135  * syncing the tree wait for us to finish
 136  */
 137 static int start_log_trans(struct btrfs_trans_handle *trans,
 138                            struct btrfs_root *root)
 139 {
 140         int ret;
 141         int err = 0;
 142
 143         mutex_lock(&root->log_mutex);
 144         if (root->log_root) {
 145                 if (!root->log_start_pid) {
 146                         root->log_start_pid = current->pid;
 147                         root->log_multiple_pids = false;
 148                 } else if (root->log_start_pid != current->pid) {
 149                         root->log_multiple_pids = true;
 150                 }
 151
 152                 atomic_inc(&root->log_batch);
 153                 atomic_inc(&root->log_writers);
 154                 mutex_unlock(&root->log_mutex);
 155                 return 0;
 156         }
 157         root->log_multiple_pids = false;
 158         root->log_start_pid = current->pid;
 159         mutex_lock(&root->fs_info->tree_log_mutex);
 160         if (!root->fs_info->log_root_tree) {
 161                 ret = btrfs_init_log_root_tree(trans, root->fs_info);
 162                 if (ret)
 163                         err = ret;
 164         }
 165         if (err == 0 && !root->log_root) {
 166                 ret = btrfs_add_log_tree(trans, root);
 167                 if (ret)
 168                         err = ret;
 169         }
 170         mutex_unlock(&root->fs_info->tree_log_mutex);
 171         atomic_inc(&root->log_batch);
 172         atomic_inc(&root->log_writers);
 173         mutex_unlock(&root->log_mutex);
 174         return err;
 175 }
 176
 177 /*
 178  * returns 0 if there was a log transaction running and we were able
 179  * to join, or returns -ENOENT if there were not transactions
 180  * in progress
 181  */
 182 static int join_running_log_trans(struct btrfs_root *root)
 183 {
 184         int ret = -ENOENT;
 185
 186         smp_mb();
 187         if (!root->log_root)
 188                 return -ENOENT;
 189
 190         mutex_lock(&root->log_mutex);
 191         if (root->log_root) {
 192                 ret = 0;
 193                 atomic_inc(&root->log_writers);
 194         }
 195         mutex_unlock(&root->log_mutex);
 196         return ret;
 197 }
 198
 199 /*
 200  * This either makes the current running log transaction wait
 201  * until you call btrfs_end_log_trans() or it makes any future
 202  * log transactions wait until you call btrfs_end_log_trans()
 203  */
 204 int btrfs_pin_log_trans(struct btrfs_root *root)
 205 {
 206         int ret = -ENOENT;
 207
 208         mutex_lock(&root->log_mutex);
 209         atomic_inc(&root->log_writers);
 210         mutex_unlock(&root->log_mutex);
 211         return ret;
 212 }
 213
 214 /*
 215  * indicate we're done making changes to the log tree
 216  * and wake up anyone waiting to do a sync
 217  */
 218 void btrfs_end_log_trans(struct btrfs_root *root)
 219 {
 220         if (atomic_dec_and_test(&root->log_writers)) {
 221                 smp_mb();
 222                 if (waitqueue_active(&root->log_writer_wait))
 223                         wake_up(&root->log_writer_wait);
 224         }
 225 }
 226
 227
 228 /*
 229  * the walk control struct is used to pass state down the chain when
 230  * processing the log tree.  The stage field tells us which part
 231  * of the log tree processing we are currently doing.  The others
 232  * are state fields used for that specific part
 233  */
 234 struct walk_control {
 235         /* should we free the extent on disk when done?  This is used
 236          * at transaction commit time while freeing a log tree
 237          */
 238         int free;
 239
 240         /* should we write out the extent buffer?  This is used
 241          * while flushing the log tree to disk during a sync
 242          */
 243         int write;
 244
 245         /* should we wait for the extent buffer io to finish?  Also used
 246          * while flushing the log tree to disk for a sync
 247          */
 248         int wait;
 249
 250         /* pin only walk, we record which extents on disk belong to the
 251          * log trees
 252          */
 253         int pin;
 254
 255         /* what stage of the replay code we're currently in */
 256         int stage;
 257
 258         /* the root we are currently replaying */
 259         struct btrfs_root *replay_dest;
 260
 261         /* the trans handle for the current replay */
 262         struct btrfs_trans_handle *trans;
 263
 264         /* the function that gets used to process blocks we find in the
 265          * tree.  Note the extent_buffer might not be up to date when it is
 266          * passed in, and it must be checked or read if you need the data
 267          * inside it
 268          */
 269         int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
 270                             struct walk_control *wc, u64 gen);
 271 };
 272
 273 /*
 274  * process_func used to pin down extents, write them or wait on them
 275  */
 276 static int process_one_buffer(struct btrfs_root *log,
 277                               struct extent_buffer *eb,
 278                               struct walk_control *wc, u64 gen)
 279 {
 280         int ret = 0;
 281
 282         if (wc->pin)
 283                 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
 284                                                       eb->start, eb->len);
 285
 286         if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
 287                 if (wc->write)
 288                         btrfs_write_tree_block(eb);
 289                 if (wc->wait)
 290                         btrfs_wait_tree_block_writeback(eb);
 291         }
 292         return ret;
 293 }
 294
 295 /*
 296  * Item overwrite used by replay and tree logging.  eb, slot and key all refer
 297  * to the src data we are copying out.
 298  *
 299  * root is the tree we are copying into, and path is a scratch
 300  * path for use in this function (it should be released on entry and
 301  * will be released on exit).
 302  *
 303  * If the key is already in the destination tree the existing item is
 304  * overwritten.  If the existing item isn't big enough, it is extended.
 305  * If it is too large, it is truncated.
 306  *
 307  * If the key isn't in the destination yet, a new item is inserted.
 308  */
 309 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
 310                                    struct btrfs_root *root,
 311                                    struct btrfs_path *path,
 312                                    struct extent_buffer *eb, int slot,
 313                                    struct btrfs_key *key)
 314 {
 315         int ret;
 316         u32 item_size;
 317         u64 saved_i_size = 0;
 318         int save_old_i_size = 0;
 319         unsigned long src_ptr;
 320         unsigned long dst_ptr;
 321         int overwrite_root = 0;
 322         bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
 323
 324         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
 325                 overwrite_root = 1;
 326
 327         item_size = btrfs_item_size_nr(eb, slot);
 328         src_ptr = btrfs_item_ptr_offset(eb, slot);
 329
 330         /* look for the key in the destination tree */
 331         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
 332         if (ret < 0)
 333                 return ret;
 334
 335         if (ret == 0) {
 336                 char *src_copy;
 337                 char *dst_copy;
 338                 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
 339                                                   path->slots[0]);
 340                 if (dst_size != item_size)
 341                         goto insert;
 342
 343                 if (item_size == 0) {
 344                         btrfs_release_path(path);
 345                         return 0;
 346                 }
 347                 dst_copy = kmalloc(item_size, GFP_NOFS);
 348                 src_copy = kmalloc(item_size, GFP_NOFS);
 349                 if (!dst_copy || !src_copy) {
 350                         btrfs_release_path(path);
 351                         kfree(dst_copy);
 352                         kfree(src_copy);
 353                         return -ENOMEM;
 354                 }
 355
 356                 read_extent_buffer(eb, src_copy, src_ptr, item_size);
 357
 358                 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
 359                 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
 360                                    item_size);
 361                 ret = memcmp(dst_copy, src_copy, item_size);
 362
 363                 kfree(dst_copy);
 364                 kfree(src_copy);
 365                 /*
 366                  * they have the same contents, just return, this saves
 367                  * us from cowing blocks in the destination tree and doing
 368                  * extra writes that may not have been done by a previous
 369                  * sync
 370                  */
 371                 if (ret == 0) {
 372                         btrfs_release_path(path);
 373                         return 0;
 374                 }
 375
 376                 /*
 377                  * We need to load the old nbytes into the inode so when we
 378                  * replay the extents we've logged we get the right nbytes.
 379                  */
 380                 if (inode_item) {
 381                         struct btrfs_inode_item *item;
 382                         u64 nbytes;
 383
 384                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 385                                               struct btrfs_inode_item);
 386                         nbytes = btrfs_inode_nbytes(path->nodes[0], item);
 387                         item = btrfs_item_ptr(eb, slot,
 388                                               struct btrfs_inode_item);
 389                         btrfs_set_inode_nbytes(eb, item, nbytes);
 390                 }
 391         } else if (inode_item) {
 392                 struct btrfs_inode_item *item;
 393
 394                 /*
 395                  * New inode, set nbytes to 0 so that the nbytes comes out
 396                  * properly when we replay the extents.
 397                  */
 398                 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
 399                 btrfs_set_inode_nbytes(eb, item, 0);
 400         }
 401 insert:
 402         btrfs_release_path(path);
 403         /* try to insert the key into the destination tree */
 404         ret = btrfs_insert_empty_item(trans, root, path,
 405                                       key, item_size);
 406
 407         /* make sure any existing item is the correct size */
 408         if (ret == -EEXIST) {
 409                 u32 found_size;
 410                 found_size = btrfs_item_size_nr(path->nodes[0],
 411                                                 path->slots[0]);
 412                 if (found_size > item_size)
 413                         btrfs_truncate_item(root, path, item_size, 1);
 414                 else if (found_size < item_size)
 415                         btrfs_extend_item(root, path,
 416                                           item_size - found_size);
 417         } else if (ret) {
 418                 return ret;
 419         }
 420         dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
 421                                         path->slots[0]);
 422
 423         /* don't overwrite an existing inode if the generation number
 424          * was logged as zero.  This is done when the tree logging code
 425          * is just logging an inode to make sure it exists after recovery.
 426          *
 427          * Also, don't overwrite i_size on directories during replay.
 428          * log replay inserts and removes directory items based on the
 429          * state of the tree found in the subvolume, and i_size is modified
 430          * as it goes
 431          */
 432         if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
 433                 struct btrfs_inode_item *src_item;
 434                 struct btrfs_inode_item *dst_item;
 435
 436                 src_item = (struct btrfs_inode_item *)src_ptr;
 437                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 438
 439                 if (btrfs_inode_generation(eb, src_item) == 0)
 440                         goto no_copy;
 441
 442                 if (overwrite_root &&
 443                     S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
 444                     S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
 445                         save_old_i_size = 1;
 446                         saved_i_size = btrfs_inode_size(path->nodes[0],
 447                                                         dst_item);
 448                 }
 449         }
 450
 451         copy_extent_buffer(path->nodes[0], eb, dst_ptr,
 452                            src_ptr, item_size);
 453
 454         if (save_old_i_size) {
 455                 struct btrfs_inode_item *dst_item;
 456                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 457                 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
 458         }
 459
 460         /* make sure the generation is filled in */
 461         if (key->type == BTRFS_INODE_ITEM_KEY) {
 462                 struct btrfs_inode_item *dst_item;
 463                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 464                 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
 465                         btrfs_set_inode_generation(path->nodes[0], dst_item,
 466                                                    trans->transid);
 467                 }
 468         }
 469 no_copy:
 470         btrfs_mark_buffer_dirty(path->nodes[0]);
 471         btrfs_release_path(path);
 472         return 0;
 473 }
 474
 475 /*
 476  * simple helper to read an inode off the disk from a given root
 477  * This can only be called for subvolume roots and not for the log
 478  */
 479 static noinline struct inode *read_one_inode(struct btrfs_root *root,
 480                                              u64 objectid)
 481 {
 482         struct btrfs_key key;
 483         struct inode *inode;
 484
 485         key.objectid = objectid;
 486         key.type = BTRFS_INODE_ITEM_KEY;
 487         key.offset = 0;
 488         inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
 489         if (IS_ERR(inode)) {
 490                 inode = NULL;
 491         } else if (is_bad_inode(inode)) {
 492                 iput(inode);
 493                 inode = NULL;
 494         }
 495         return inode;
 496 }
 497
 498 /* replays a single extent in 'eb' at 'slot' with 'key' into the
 499  * subvolume 'root'.  path is released on entry and should be released
 500  * on exit.
 501  *
 502  * extents in the log tree have not been allocated out of the extent
 503  * tree yet.  So, this completes the allocation, taking a reference
 504  * as required if the extent already exists or creating a new extent
 505  * if it isn't in the extent allocation tree yet.
 506  *
 507  * The extent is inserted into the file, dropping any existing extents
 508  * from the file that overlap the new one.
 509  */
 510 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 511                                       struct btrfs_root *root,
 512                                       struct btrfs_path *path,
 513                                       struct extent_buffer *eb, int slot,
 514                                       struct btrfs_key *key)
 515 {
 516         int found_type;
 517         u64 extent_end;
 518         u64 start = key->offset;
 519         u64 nbytes = 0;
 520         struct btrfs_file_extent_item *item;
 521         struct inode *inode = NULL;
 522         unsigned long size;
 523         int ret = 0;
 524
 525         item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 526         found_type = btrfs_file_extent_type(eb, item);
 527
 528         if (found_type == BTRFS_FILE_EXTENT_REG ||
 529             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
 530                 nbytes = btrfs_file_extent_num_bytes(eb, item);
 531                 extent_end = start + nbytes;
 532
 533                 /*
 534                  * We don't add to the inodes nbytes if we are prealloc or a
 535                  * hole.
 536                  */
 537                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
 538                         nbytes = 0;
 539         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 540                 size = btrfs_file_extent_inline_len(eb, item);
 541                 nbytes = btrfs_file_extent_ram_bytes(eb, item);
 542                 extent_end = ALIGN(start + size, root->sectorsize);
 543         } else {
 544                 ret = 0;
 545                 goto out;
 546         }
 547
 548         inode = read_one_inode(root, key->objectid);
 549         if (!inode) {
 550                 ret = -EIO;
 551                 goto out;
 552         }
 553
 554         /*
 555          * first check to see if we already have this extent in the
 556          * file.  This must be done before the btrfs_drop_extents run
 557          * so we don't try to drop this extent.
 558          */
 559         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
 560                                        start, 0);
 561
 562         if (ret == 0 &&
 563             (found_type == BTRFS_FILE_EXTENT_REG ||
 564              found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
 565                 struct btrfs_file_extent_item cmp1;
 566                 struct btrfs_file_extent_item cmp2;
 567                 struct btrfs_file_extent_item *existing;
 568                 struct extent_buffer *leaf;
 569
 570                 leaf = path->nodes[0];
 571                 existing = btrfs_item_ptr(leaf, path->slots[0],
 572                                           struct btrfs_file_extent_item);
 573
 574                 read_extent_buffer(eb, &cmp1, (unsigned long)item,
 575                                    sizeof(cmp1));
 576                 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
 577                                    sizeof(cmp2));
 578
 579                 /*
 580                  * we already have a pointer to this exact extent,
 581                  * we don't have to do anything
 582                  */
 583                 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
 584                         btrfs_release_path(path);
 585                         goto out;
 586                 }
 587         }
 588         btrfs_release_path(path);
 589
 590         /* drop any overlapping extents */
 591         ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
 592         if (ret)
 593                 goto out;
 594
 595         if (found_type == BTRFS_FILE_EXTENT_REG ||
 596             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
 597                 u64 offset;
 598                 unsigned long dest_offset;
 599                 struct btrfs_key ins;
 600
 601                 ret = btrfs_insert_empty_item(trans, root, path, key,
 602                                               sizeof(*item));
 603                 if (ret)
 604                         goto out;
 605                 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
 606                                                     path->slots[0]);
 607                 copy_extent_buffer(path->nodes[0], eb, dest_offset,
 608                                 (unsigned long)item,  sizeof(*item));
 609
 610                 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
 611                 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
 612                 ins.type = BTRFS_EXTENT_ITEM_KEY;
 613                 offset = key->offset - btrfs_file_extent_offset(eb, item);
 614
 615                 if (ins.objectid > 0) {
 616                         u64 csum_start;
 617                         u64 csum_end;
 618                         LIST_HEAD(ordered_sums);
 619                         /*
 620                          * is this extent already allocated in the extent
 621                          * allocation tree?  If so, just add a reference
 622                          */
 623                         ret = btrfs_lookup_extent(root, ins.objectid,
 624                                                 ins.offset);
 625                         if (ret == 0) {
 626                                 ret = btrfs_inc_extent_ref(trans, root,
 627                                                 ins.objectid, ins.offset,
 628                                                 0, root->root_key.objectid,
 629                                                 key->objectid, offset, 0);
 630                                 if (ret)
 631                                         goto out;
 632                         } else {
 633                                 /*
 634                                  * insert the extent pointer in the extent
 635                                  * allocation tree
 636                                  */
 637                                 ret = btrfs_alloc_logged_file_extent(trans,
 638                                                 root, root->root_key.objectid,
 639                                                 key->objectid, offset, &ins);
 640                                 if (ret)
 641                                         goto out;
 642                         }
 643                         btrfs_release_path(path);
 644
 645                         if (btrfs_file_extent_compression(eb, item)) {
 646                                 csum_start = ins.objectid;
 647                                 csum_end = csum_start + ins.offset;
 648                         } else {
 649                                 csum_start = ins.objectid +
 650                                         btrfs_file_extent_offset(eb, item);
 651                                 csum_end = csum_start +
 652                                         btrfs_file_extent_num_bytes(eb, item);
 653                         }
 654
 655                         ret = btrfs_lookup_csums_range(root->log_root,
 656                                                 csum_start, csum_end - 1,
 657                                                 &ordered_sums, 0);
 658                         if (ret)
 659                                 goto out;
 660                         while (!list_empty(&ordered_sums)) {
 661                                 struct btrfs_ordered_sum *sums;
 662                                 sums = list_entry(ordered_sums.next,
 663                                                 struct btrfs_ordered_sum,
 664                                                 list);
 665                                 if (!ret)
 666                                         ret = btrfs_csum_file_blocks(trans,
 667                                                 root->fs_info->csum_root,
 668                                                 sums);
 669                                 list_del(&sums->list);
 670                                 kfree(sums);
 671                         }
 672                         if (ret)
 673                                 goto out;
 674                 } else {
 675                         btrfs_release_path(path);
 676                 }
 677         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 678                 /* inline extents are easy, we just overwrite them */
 679                 ret = overwrite_item(trans, root, path, eb, slot, key);
 680                 if (ret)
 681                         goto out;
 682         }
 683
 684         inode_add_bytes(inode, nbytes);
 685         ret = btrfs_update_inode(trans, root, inode);
 686 out:
 687         if (inode)
 688                 iput(inode);
 689         return ret;
 690 }
 691
 692 /*
 693  * when cleaning up conflicts between the directory names in the
 694  * subvolume, directory names in the log and directory names in the
 695  * inode back references, we may have to unlink inodes from directories.
 696  *
 697  * This is a helper function to do the unlink of a specific directory
 698  * item
 699  */
 700 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
 701                                       struct btrfs_root *root,
 702                                       struct btrfs_path *path,
 703                                       struct inode *dir,
 704                                       struct btrfs_dir_item *di)
 705 {
 706         struct inode *inode;
 707         char *name;
 708         int name_len;
 709         struct extent_buffer *leaf;
 710         struct btrfs_key location;
 711         int ret;
 712
 713         leaf = path->nodes[0];
 714
 715         btrfs_dir_item_key_to_cpu(leaf, di, &location);
 716         name_len = btrfs_dir_name_len(leaf, di);
 717         name = kmalloc(name_len, GFP_NOFS);
 718         if (!name)
 719                 return -ENOMEM;
 720
 721         read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
 722         btrfs_release_path(path);
 723
 724         inode = read_one_inode(root, location.objectid);
 725         if (!inode) {
 726                 ret = -EIO;
 727                 goto out;
 728         }
 729
 730         ret = link_to_fixup_dir(trans, root, path, location.objectid);
 731         if (ret)
 732                 goto out;
 733
 734         ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
 735         if (ret)
 736                 goto out;
 737         btrfs_run_delayed_items(trans, root);
 738 out:
 739         kfree(name);
 740         iput(inode);
 741         return ret;
 742 }
 743
 744 /*
 745  * helper function to see if a given name and sequence number found
 746  * in an inode back reference are already in a directory and correctly
 747  * point to this inode
 748  */
 749 static noinline int inode_in_dir(struct btrfs_root *root,
 750                                  struct btrfs_path *path,
 751                                  u64 dirid, u64 objectid, u64 index,
 752                                  const char *name, int name_len)
 753 {
 754         struct btrfs_dir_item *di;
 755         struct btrfs_key location;
 756         int match = 0;
 757
 758         di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
 759                                          index, name, name_len, 0);
 760         if (di && !IS_ERR(di)) {
 761                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 762                 if (location.objectid != objectid)
 763                         goto out;
 764         } else
 765                 goto out;
 766         btrfs_release_path(path);
 767
 768         di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
 769         if (di && !IS_ERR(di)) {
 770                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 771                 if (location.objectid != objectid)
 772                         goto out;
 773         } else
 774                 goto out;
 775         match = 1;
 776 out:
 777         btrfs_release_path(path);
 778         return match;
 779 }
 780
 781 /*
 782  * helper function to check a log tree for a named back reference in
 783  * an inode.  This is used to decide if a back reference that is
 784  * found in the subvolume conflicts with what we find in the log.
 785  *
 786  * inode backreferences may have multiple refs in a single item,
 787  * during replay we process one reference at a time, and we don't
 788  * want to delete valid links to a file from the subvolume if that
 789  * link is also in the log.
 790  */
 791 static noinline int backref_in_log(struct btrfs_root *log,
 792                                    struct btrfs_key *key,
 793                                    u64 ref_objectid,
 794                                    char *name, int namelen)
 795 {
 796         struct btrfs_path *path;
 797         struct btrfs_inode_ref *ref;
 798         unsigned long ptr;
 799         unsigned long ptr_end;
 800         unsigned long name_ptr;
 801         int found_name_len;
 802         int item_size;
 803         int ret;
 804         int match = 0;
 805
 806         path = btrfs_alloc_path();
 807         if (!path)
 808                 return -ENOMEM;
 809
 810         ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
 811         if (ret != 0)
 812                 goto out;
 813
 814         ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
 815
 816         if (key->type == BTRFS_INODE_EXTREF_KEY) {
 817                 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
 818                                                    name, namelen, NULL))
 819                         match = 1;
 820
 821                 goto out;
 822         }
 823
 824         item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
 825         ptr_end = ptr + item_size;
 826         while (ptr < ptr_end) {
 827                 ref = (struct btrfs_inode_ref *)ptr;
 828                 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
 829                 if (found_name_len == namelen) {
 830                         name_ptr = (unsigned long)(ref + 1);
 831                         ret = memcmp_extent_buffer(path->nodes[0], name,
 832                                                    name_ptr, namelen);
 833                         if (ret == 0) {
 834                                 match = 1;
 835                                 goto out;
 836                         }
 837                 }
 838                 ptr = (unsigned long)(ref + 1) + found_name_len;
 839         }
 840 out:
 841         btrfs_free_path(path);
 842         return match;
 843 }
 844
 845 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
 846                                   struct btrfs_root *root,
 847                                   struct btrfs_path *path,
 848                                   struct btrfs_root *log_root,
 849                                   struct inode *dir, struct inode *inode,
 850                                   struct extent_buffer *eb,
 851                                   u64 inode_objectid, u64 parent_objectid,
 852                                   u64 ref_index, char *name, int namelen,
 853                                   int *search_done)
 854 {
 855         int ret;
 856         char *victim_name;
 857         int victim_name_len;
 858         struct extent_buffer *leaf;
 859         struct btrfs_dir_item *di;
 860         struct btrfs_key search_key;
 861         struct btrfs_inode_extref *extref;
 862
 863 again:
 864         /* Search old style refs */
 865         search_key.objectid = inode_objectid;
 866         search_key.type = BTRFS_INODE_REF_KEY;
 867         search_key.offset = parent_objectid;
 868         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
 869         if (ret == 0) {
 870                 struct btrfs_inode_ref *victim_ref;
 871                 unsigned long ptr;
 872                 unsigned long ptr_end;
 873
 874                 leaf = path->nodes[0];
 875
 876                 /* are we trying to overwrite a back ref for the root directory
 877                  * if so, just jump out, we're done
 878                  */
 879                 if (search_key.objectid == search_key.offset)
 880                         return 1;
 881
 882                 /* check all the names in this back reference to see
 883                  * if they are in the log.  if so, we allow them to stay
 884                  * otherwise they must be unlinked as a conflict
 885                  */
 886                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
 887                 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
 888                 while (ptr < ptr_end) {
 889                         victim_ref = (struct btrfs_inode_ref *)ptr;
 890                         victim_name_len = btrfs_inode_ref_name_len(leaf,
 891                                                                    victim_ref);
 892                         victim_name = kmalloc(victim_name_len, GFP_NOFS);
 893                         if (!victim_name)
 894                                 return -ENOMEM;
 895
 896                         read_extent_buffer(leaf, victim_name,
 897                                            (unsigned long)(victim_ref + 1),
 898                                            victim_name_len);
 899
 900                         if (!backref_in_log(log_root, &search_key,
 901                                             parent_objectid,
 902                                             victim_name,
 903                                             victim_name_len)) {
 904                                 btrfs_inc_nlink(inode);
 905                                 btrfs_release_path(path);
 906
 907                                 ret = btrfs_unlink_inode(trans, root, dir,
 908                                                          inode, victim_name,
 909                                                          victim_name_len);
 910                                 kfree(victim_name);
 911                                 if (ret)
 912                                         return ret;
 913                                 btrfs_run_delayed_items(trans, root);
 914                                 *search_done = 1;
 915                                 goto again;
 916                         }
 917                         kfree(victim_name);
 918
 919                         ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
 920                 }
 921
 922                 /*
 923                  * NOTE: we have searched root tree and checked the
 924                  * coresponding ref, it does not need to check again.
 925                  */
 926                 *search_done = 1;
 927         }
 928         btrfs_release_path(path);
 929
 930         /* Same search but for extended refs */
 931         extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
 932                                            inode_objectid, parent_objectid, 0,
 933                                            0);
 934         if (!IS_ERR_OR_NULL(extref)) {
 935                 u32 item_size;
 936                 u32 cur_offset = 0;
 937                 unsigned long base;
 938                 struct inode *victim_parent;
 939
 940                 leaf = path->nodes[0];
 941
 942                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 943                 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
 944
 945                 while (cur_offset < item_size) {
 946                         extref = (struct btrfs_inode_extref *)(base + cur_offset);
 947
 948                         victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
 949
 950                         if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
 951                                 goto next;
 952
 953                         victim_name = kmalloc(victim_name_len, GFP_NOFS);
 954                         if (!victim_name)
 955                                 return -ENOMEM;
 956                         read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
 957                                            victim_name_len);
 958
 959                         search_key.objectid = inode_objectid;
 960                         search_key.type = BTRFS_INODE_EXTREF_KEY;
 961                         search_key.offset = btrfs_extref_hash(parent_objectid,
 962                                                               victim_name,
 963                                                               victim_name_len);
 964                         ret = 0;
 965                         if (!backref_in_log(log_root, &search_key,
 966                                             parent_objectid, victim_name,
 967                                             victim_name_len)) {
 968                                 ret = -ENOENT;
 969                                 victim_parent = read_one_inode(root,
 970                                                                parent_objectid);
 971                                 if (victim_parent) {
 972                                         btrfs_inc_nlink(inode);
 973                                         btrfs_release_path(path);
 974
 975                                         ret = btrfs_unlink_inode(trans, root,
 976                                                                  victim_parent,
 977                                                                  inode,
 978                                                                  victim_name,
 979                                                                  victim_name_len);
 980                                         btrfs_run_delayed_items(trans, root);
 981                                 }
 982                                 iput(victim_parent);
 983                                 kfree(victim_name);
 984                                 if (ret)
 985                                         return ret;
 986                                 *search_done = 1;
 987                                 goto again;
 988                         }
 989                         kfree(victim_name);
 990                         if (ret)
 991                                 return ret;
 992 next:
 993                         cur_offset += victim_name_len + sizeof(*extref);
 994                 }
 995                 *search_done = 1;
 996         }
 997         btrfs_release_path(path);
 998
 999         /* look for a conflicting sequence number */
1000         di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1001                                          ref_index, name, namelen, 0);
1002         if (di && !IS_ERR(di)) {
1003                 ret = drop_one_dir_item(trans, root, path, dir, di);
1004                 if (ret)
1005                         return ret;
1006         }
1007         btrfs_release_path(path);
1008
1009         /* look for a conflicing name */
1010         di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1011                                    name, namelen, 0);
1012         if (di && !IS_ERR(di)) {
1013                 ret = drop_one_dir_item(trans, root, path, dir, di);
1014                 if (ret)
1015                         return ret;
1016         }
1017         btrfs_release_path(path);
1018
1019         return 0;
1020 }
1021
1022 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1023                              u32 *namelen, char **name, u64 *index,
1024                              u64 *parent_objectid)
1025 {
1026         struct btrfs_inode_extref *extref;
1027
1028         extref = (struct btrfs_inode_extref *)ref_ptr;
1029
1030         *namelen = btrfs_inode_extref_name_len(eb, extref);
1031         *name = kmalloc(*namelen, GFP_NOFS);
1032         if (*name == NULL)
1033                 return -ENOMEM;
1034
1035         read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1036                            *namelen);
1037
1038         *index = btrfs_inode_extref_index(eb, extref);
1039         if (parent_objectid)
1040                 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1041
1042         return 0;
1043 }
1044
1045 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1046                           u32 *namelen, char **name, u64 *index)
1047 {
1048         struct btrfs_inode_ref *ref;
1049
1050         ref = (struct btrfs_inode_ref *)ref_ptr;
1051
1052         *namelen = btrfs_inode_ref_name_len(eb, ref);
1053         *name = kmalloc(*namelen, GFP_NOFS);
1054         if (*name == NULL)
1055                 return -ENOMEM;
1056
1057         read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1058
1059         *index = btrfs_inode_ref_index(eb, ref);
1060
1061         return 0;
1062 }
1063
1064 /*
1065  * replay one inode back reference item found in the log tree.
1066  * eb, slot and key refer to the buffer and key found in the log tree.
1067  * root is the destination we are replaying into, and path is for temp
1068  * use by this function.  (it should be released on return).
1069  */
1070 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1071                                   struct btrfs_root *root,
1072                                   struct btrfs_root *log,
1073                                   struct btrfs_path *path,
1074                                   struct extent_buffer *eb, int slot,
1075                                   struct btrfs_key *key)
1076 {
1077         struct inode *dir;
1078         struct inode *inode;
1079         unsigned long ref_ptr;
1080         unsigned long ref_end;
1081         char *name;
1082         int namelen;
1083         int ret;
1084         int search_done = 0;
1085         int log_ref_ver = 0;
1086         u64 parent_objectid;
1087         u64 inode_objectid;
1088         u64 ref_index = 0;
1089         int ref_struct_size;
1090
1091         ref_ptr = btrfs_item_ptr_offset(eb, slot);
1092         ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1093
1094         if (key->type == BTRFS_INODE_EXTREF_KEY) {
1095                 struct btrfs_inode_extref *r;
1096
1097                 ref_struct_size = sizeof(struct btrfs_inode_extref);
1098                 log_ref_ver = 1;
1099                 r = (struct btrfs_inode_extref *)ref_ptr;
1100                 parent_objectid = btrfs_inode_extref_parent(eb, r);
1101         } else {
1102                 ref_struct_size = sizeof(struct btrfs_inode_ref);
1103                 parent_objectid = key->offset;
1104         }
1105         inode_objectid = key->objectid;
1106
1107         /*
1108          * it is possible that we didn't log all the parent directories
1109          * for a given inode.  If we don't find the dir, just don't
1110          * copy the back ref in.  The link count fixup code will take
1111          * care of the rest
1112          */
1113         dir = read_one_inode(root, parent_objectid);
1114         if (!dir)
1115                 return -ENOENT;
1116
1117         inode = read_one_inode(root, inode_objectid);
1118         if (!inode) {
1119                 iput(dir);
1120                 return -EIO;
1121         }
1122
1123         while (ref_ptr < ref_end) {
1124                 if (log_ref_ver) {
1125                         ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1126                                                 &ref_index, &parent_objectid);
1127                         /*
1128                          * parent object can change from one array
1129                          * item to another.
1130                          */
1131                         if (!dir)
1132                                 dir = read_one_inode(root, parent_objectid);
1133                         if (!dir)
1134                                 return -ENOENT;
1135                 } else {
1136                         ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1137                                              &ref_index);
1138                 }
1139                 if (ret)
1140                         return ret;
1141
1142                 /* if we already have a perfect match, we're done */
1143                 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1144                                   ref_index, name, namelen)) {
1145                         /*
1146                          * look for a conflicting back reference in the
1147                          * metadata. if we find one we have to unlink that name
1148                          * of the file before we add our new link.  Later on, we
1149                          * overwrite any existing back reference, and we don't
1150                          * want to create dangling pointers in the directory.
1151                          */
1152
1153                         if (!search_done) {
1154                                 ret = __add_inode_ref(trans, root, path, log,
1155                                                       dir, inode, eb,
1156                                                       inode_objectid,
1157                                                       parent_objectid,
1158                                                       ref_index, name, namelen,
1159                                                       &search_done);
1160                                 if (ret == 1) {
1161                                         ret = 0;
1162                                         goto out;
1163                                 }
1164                                 if (ret)
1165                                         goto out;
1166                         }
1167
1168                         /* insert our name */
1169                         ret = btrfs_add_link(trans, dir, inode, name, namelen,
1170                                              0, ref_index);
1171                         if (ret)
1172                                 goto out;
1173
1174                         btrfs_update_inode(trans, root, inode);
1175                 }
1176
1177                 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1178                 kfree(name);
1179                 if (log_ref_ver) {
1180                         iput(dir);
1181                         dir = NULL;
1182                 }
1183         }
1184
1185         /* finally write the back reference in the inode */
1186         ret = overwrite_item(trans, root, path, eb, slot, key);
1187 out:
1188         btrfs_release_path(path);
1189         iput(dir);
1190         iput(inode);
1191         return ret;
1192 }
1193
1194 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1195                               struct btrfs_root *root, u64 offset)
1196 {
1197         int ret;
1198         ret = btrfs_find_orphan_item(root, offset);
1199         if (ret > 0)
1200                 ret = btrfs_insert_orphan_item(trans, root, offset);
1201         return ret;
1202 }
1203
1204 static int count_inode_extrefs(struct btrfs_root *root,
1205                                struct inode *inode, struct btrfs_path *path)
1206 {
1207         int ret = 0;
1208         int name_len;
1209         unsigned int nlink = 0;
1210         u32 item_size;
1211         u32 cur_offset = 0;
1212         u64 inode_objectid = btrfs_ino(inode);
1213         u64 offset = 0;
1214         unsigned long ptr;
1215         struct btrfs_inode_extref *extref;
1216         struct extent_buffer *leaf;
1217
1218         while (1) {
1219                 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1220                                             &extref, &offset);
1221                 if (ret)
1222                         break;
1223
1224                 leaf = path->nodes[0];
1225                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1226                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1227
1228                 while (cur_offset < item_size) {
1229                         extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1230                         name_len = btrfs_inode_extref_name_len(leaf, extref);
1231
1232                         nlink++;
1233
1234                         cur_offset += name_len + sizeof(*extref);
1235                 }
1236
1237                 offset++;
1238                 btrfs_release_path(path);
1239         }
1240         btrfs_release_path(path);
1241
1242         if (ret < 0)
1243                 return ret;
1244         return nlink;
1245 }
1246
1247 static int count_inode_refs(struct btrfs_root *root,
1248                                struct inode *inode, struct btrfs_path *path)
1249 {
1250         int ret;
1251         struct btrfs_key key;
1252         unsigned int nlink = 0;
1253         unsigned long ptr;
1254         unsigned long ptr_end;
1255         int name_len;
1256         u64 ino = btrfs_ino(inode);
1257
1258         key.objectid = ino;
1259         key.type = BTRFS_INODE_REF_KEY;
1260         key.offset = (u64)-1;
1261
1262         while (1) {
1263                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1264                 if (ret < 0)
1265                         break;
1266                 if (ret > 0) {
1267                         if (path->slots[0] == 0)
1268                                 break;
1269                         path->slots[0]--;
1270                 }
1271                 btrfs_item_key_to_cpu(path->nodes[0], &key,
1272                                       path->slots[0]);
1273                 if (key.objectid != ino ||
1274                     key.type != BTRFS_INODE_REF_KEY)
1275                         break;
1276                 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1277                 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1278                                                    path->slots[0]);
1279                 while (ptr < ptr_end) {
1280                         struct btrfs_inode_ref *ref;
1281
1282                         ref = (struct btrfs_inode_ref *)ptr;
1283                         name_len = btrfs_inode_ref_name_len(path->nodes[0],
1284                                                             ref);
1285                         ptr = (unsigned long)(ref + 1) + name_len;
1286                         nlink++;
1287                 }
1288
1289                 if (key.offset == 0)
1290                         break;
1291                 key.offset--;
1292                 btrfs_release_path(path);
1293         }
1294         btrfs_release_path(path);
1295
1296         return nlink;
1297 }
1298
1299 /*
1300  * There are a few corners where the link count of the file can't
1301  * be properly maintained during replay.  So, instead of adding
1302  * lots of complexity to the log code, we just scan the backrefs
1303  * for any file that has been through replay.
1304  *
1305  * The scan will update the link count on the inode to reflect the
1306  * number of back refs found.  If it goes down to zero, the iput
1307  * will free the inode.
1308  */
1309 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1310                                            struct btrfs_root *root,
1311                                            struct inode *inode)
1312 {
1313         struct btrfs_path *path;
1314         int ret;
1315         u64 nlink = 0;
1316         u64 ino = btrfs_ino(inode);
1317
1318         path = btrfs_alloc_path();
1319         if (!path)
1320                 return -ENOMEM;
1321
1322         ret = count_inode_refs(root, inode, path);
1323         if (ret < 0)
1324                 goto out;
1325
1326         nlink = ret;
1327
1328         ret = count_inode_extrefs(root, inode, path);
1329         if (ret == -ENOENT)
1330                 ret = 0;
1331
1332         if (ret < 0)
1333                 goto out;
1334
1335         nlink += ret;
1336
1337         ret = 0;
1338
1339         if (nlink != inode->i_nlink) {
1340                 set_nlink(inode, nlink);
1341                 btrfs_update_inode(trans, root, inode);
1342         }
1343         BTRFS_I(inode)->index_cnt = (u64)-1;
1344
1345         if (inode->i_nlink == 0) {
1346                 if (S_ISDIR(inode->i_mode)) {
1347                         ret = replay_dir_deletes(trans, root, NULL, path,
1348                                                  ino, 1);
1349                         if (ret)
1350                                 goto out;
1351                 }
1352                 ret = insert_orphan_item(trans, root, ino);
1353         }
1354
1355 out:
1356         btrfs_free_path(path);
1357         return ret;
1358 }
1359
1360 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1361                                             struct btrfs_root *root,
1362                                             struct btrfs_path *path)
1363 {
1364         int ret;
1365         struct btrfs_key key;
1366         struct inode *inode;
1367
1368         key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1369         key.type = BTRFS_ORPHAN_ITEM_KEY;
1370         key.offset = (u64)-1;
1371         while (1) {
1372                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1373                 if (ret < 0)
1374                         break;
1375
1376                 if (ret == 1) {
1377                         if (path->slots[0] == 0)
1378                                 break;
1379                         path->slots[0]--;
1380                 }
1381
1382                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1383                 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1384                     key.type != BTRFS_ORPHAN_ITEM_KEY)
1385                         break;
1386
1387                 ret = btrfs_del_item(trans, root, path);
1388                 if (ret)
1389                         goto out;
1390
1391                 btrfs_release_path(path);
1392                 inode = read_one_inode(root, key.offset);
1393                 if (!inode)
1394                         return -EIO;
1395
1396                 ret = fixup_inode_link_count(trans, root, inode);
1397                 iput(inode);
1398                 if (ret)
1399                         goto out;
1400
1401                 /*
1402                  * fixup on a directory may create new entries,
1403                  * make sure we always look for the highset possible
1404                  * offset
1405                  */
1406                 key.offset = (u64)-1;
1407         }
1408         ret = 0;
1409 out:
1410         btrfs_release_path(path);
1411         return ret;
1412 }
1413
1414
1415 /*
1416  * record a given inode in the fixup dir so we can check its link
1417  * count when replay is done.  The link count is incremented here
1418  * so the inode won't go away until we check it
1419  */
1420 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1421                                       struct btrfs_root *root,
1422                                       struct btrfs_path *path,
1423                                       u64 objectid)
1424 {
1425         struct btrfs_key key;
1426         int ret = 0;
1427         struct inode *inode;
1428
1429         inode = read_one_inode(root, objectid);
1430         if (!inode)
1431                 return -EIO;
1432
1433         key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1434         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1435         key.offset = objectid;
1436
1437         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1438
1439         btrfs_release_path(path);
1440         if (ret == 0) {
1441                 if (!inode->i_nlink)
1442                         set_nlink(inode, 1);
1443                 else
1444                         btrfs_inc_nlink(inode);
1445                 ret = btrfs_update_inode(trans, root, inode);
1446         } else if (ret == -EEXIST) {
1447                 ret = 0;
1448         } else {
1449                 BUG(); /* Logic Error */
1450         }
1451         iput(inode);
1452
1453         return ret;
1454 }
1455
1456 /*
1457  * when replaying the log for a directory, we only insert names
1458  * for inodes that actually exist.  This means an fsync on a directory
1459  * does not implicitly fsync all the new files in it
1460  */
1461 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1462                                     struct btrfs_root *root,
1463                                     struct btrfs_path *path,
1464                                     u64 dirid, u64 index,
1465                                     char *name, int name_len, u8 type,
1466                                     struct btrfs_key *location)
1467 {
1468         struct inode *inode;
1469         struct inode *dir;
1470         int ret;
1471
1472         inode = read_one_inode(root, location->objectid);
1473         if (!inode)
1474                 return -ENOENT;
1475
1476         dir = read_one_inode(root, dirid);
1477         if (!dir) {
1478                 iput(inode);
1479                 return -EIO;
1480         }
1481         ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1482
1483         /* FIXME, put inode into FIXUP list */
1484
1485         iput(inode);
1486         iput(dir);
1487         return ret;
1488 }
1489
1490 /*
1491  * take a single entry in a log directory item and replay it into
1492  * the subvolume.
1493  *
1494  * if a conflicting item exists in the subdirectory already,
1495  * the inode it points to is unlinked and put into the link count
1496  * fix up tree.
1497  *
1498  * If a name from the log points to a file or directory that does
1499  * not exist in the FS, it is skipped.  fsyncs on directories
1500  * do not force down inodes inside that directory, just changes to the
1501  * names or unlinks in a directory.
1502  */
1503 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1504                                     struct btrfs_root *root,
1505                                     struct btrfs_path *path,
1506                                     struct extent_buffer *eb,
1507                                     struct btrfs_dir_item *di,
1508                                     struct btrfs_key *key)
1509 {
1510         char *name;
1511         int name_len;
1512         struct btrfs_dir_item *dst_di;
1513         struct btrfs_key found_key;
1514         struct btrfs_key log_key;
1515         struct inode *dir;
1516         u8 log_type;
1517         int exists;
1518         int ret = 0;
1519
1520         dir = read_one_inode(root, key->objectid);
1521         if (!dir)
1522                 return -EIO;
1523
1524         name_len = btrfs_dir_name_len(eb, di);
1525         name = kmalloc(name_len, GFP_NOFS);
1526         if (!name)
1527                 return -ENOMEM;
1528
1529         log_type = btrfs_dir_type(eb, di);
1530         read_extent_buffer(eb, name, (unsigned long)(di + 1),
1531                    name_len);
1532
1533         btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1534         exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1535         if (exists == 0)
1536                 exists = 1;
1537         else
1538                 exists = 0;
1539         btrfs_release_path(path);
1540
1541         if (key->type == BTRFS_DIR_ITEM_KEY) {
1542                 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1543                                        name, name_len, 1);
1544         } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1545                 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1546                                                      key->objectid,
1547                                                      key->offset, name,
1548                                                      name_len, 1);
1549         } else {
1550                 /* Corruption */
1551                 ret = -EINVAL;
1552                 goto out;
1553         }
1554         if (IS_ERR_OR_NULL(dst_di)) {
1555                 /* we need a sequence number to insert, so we only
1556                  * do inserts for the BTRFS_DIR_INDEX_KEY types
1557                  */
1558                 if (key->type != BTRFS_DIR_INDEX_KEY)
1559                         goto out;
1560                 goto insert;
1561         }
1562
1563         btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1564         /* the existing item matches the logged item */
1565         if (found_key.objectid == log_key.objectid &&
1566             found_key.type == log_key.type &&
1567             found_key.offset == log_key.offset &&
1568             btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1569                 goto out;
1570         }
1571
1572         /*
1573          * don't drop the conflicting directory entry if the inode
1574          * for the new entry doesn't exist
1575          */
1576         if (!exists)
1577                 goto out;
1578
1579         ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1580         if (ret)
1581                 goto out;
1582
1583         if (key->type == BTRFS_DIR_INDEX_KEY)
1584                 goto insert;
1585 out:
1586         btrfs_release_path(path);
1587         kfree(name);
1588         iput(dir);
1589         return ret;
1590
1591 insert:
1592         btrfs_release_path(path);
1593         ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1594                               name, name_len, log_type, &log_key);
1595         if (ret && ret != -ENOENT)
1596                 goto out;
1597         ret = 0;
1598         goto out;
1599 }
1600
1601 /*
1602  * find all the names in a directory item and reconcile them into
1603  * the subvolume.  Only BTRFS_DIR_ITEM_KEY types will have more than
1604  * one name in a directory item, but the same code gets used for
1605  * both directory index types
1606  */
1607 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1608                                         struct btrfs_root *root,
1609                                         struct btrfs_path *path,
1610                                         struct extent_buffer *eb, int slot,
1611                                         struct btrfs_key *key)
1612 {
1613         int ret;
1614         u32 item_size = btrfs_item_size_nr(eb, slot);
1615         struct btrfs_dir_item *di;
1616         int name_len;
1617         unsigned long ptr;
1618         unsigned long ptr_end;
1619
1620         ptr = btrfs_item_ptr_offset(eb, slot);
1621         ptr_end = ptr + item_size;
1622         while (ptr < ptr_end) {
1623                 di = (struct btrfs_dir_item *)ptr;
1624                 if (verify_dir_item(root, eb, di))
1625                         return -EIO;
1626                 name_len = btrfs_dir_name_len(eb, di);
1627                 ret = replay_one_name(trans, root, path, eb, di, key);
1628                 if (ret)
1629                         return ret;
1630                 ptr = (unsigned long)(di + 1);
1631                 ptr += name_len;
1632         }
1633         return 0;
1634 }
1635
1636 /*
1637  * directory replay has two parts.  There are the standard directory
1638  * items in the log copied from the subvolume, and range items
1639  * created in the log while the subvolume was logged.
1640  *
1641  * The range items tell us which parts of the key space the log
1642  * is authoritative for.  During replay, if a key in the subvolume
1643  * directory is in a logged range item, but not actually in the log
1644  * that means it was deleted from the directory before the fsync
1645  * and should be removed.
1646  */
1647 static noinline int find_dir_range(struct btrfs_root *root,
1648                                    struct btrfs_path *path,
1649                                    u64 dirid, int key_type,
1650                                    u64 *start_ret, u64 *end_ret)
1651 {
1652         struct btrfs_key key;
1653         u64 found_end;
1654         struct btrfs_dir_log_item *item;
1655         int ret;
1656         int nritems;
1657
1658         if (*start_ret == (u64)-1)
1659                 return 1;
1660
1661         key.objectid = dirid;
1662         key.type = key_type;
1663         key.offset = *start_ret;
1664
1665         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1666         if (ret < 0)
1667                 goto out;
1668         if (ret > 0) {
1669                 if (path->slots[0] == 0)
1670                         goto out;
1671                 path->slots[0]--;
1672         }
1673         if (ret != 0)
1674                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1675
1676         if (key.type != key_type || key.objectid != dirid) {
1677                 ret = 1;
1678                 goto next;
1679         }
1680         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1681                               struct btrfs_dir_log_item);
1682         found_end = btrfs_dir_log_end(path->nodes[0], item);
1683
1684         if (*start_ret >= key.offset && *start_ret <= found_end) {
1685                 ret = 0;
1686                 *start_ret = key.offset;
1687                 *end_ret = found_end;
1688                 goto out;
1689         }
1690         ret = 1;
1691 next:
1692         /* check the next slot in the tree to see if it is a valid item */
1693         nritems = btrfs_header_nritems(path->nodes[0]);
1694         if (path->slots[0] >= nritems) {
1695                 ret = btrfs_next_leaf(root, path);
1696                 if (ret)
1697                         goto out;
1698         } else {
1699                 path->slots[0]++;
1700         }
1701
1702         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1703
1704         if (key.type != key_type || key.objectid != dirid) {
1705                 ret = 1;
1706                 goto out;
1707         }
1708         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1709                               struct btrfs_dir_log_item);
1710         found_end = btrfs_dir_log_end(path->nodes[0], item);
1711         *start_ret = key.offset;
1712         *end_ret = found_end;
1713         ret = 0;
1714 out:
1715         btrfs_release_path(path);
1716         return ret;
1717 }
1718
1719 /*
1720  * this looks for a given directory item in the log.  If the directory
1721  * item is not in the log, the item is removed and the inode it points
1722  * to is unlinked
1723  */
1724 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1725                                       struct btrfs_root *root,
1726                                       struct btrfs_root *log,
1727                                       struct btrfs_path *path,
1728                                       struct btrfs_path *log_path,
1729                                       struct inode *dir,
1730                                       struct btrfs_key *dir_key)
1731 {
1732         int ret;
1733         struct extent_buffer *eb;
1734         int slot;
1735         u32 item_size;
1736         struct btrfs_dir_item *di;
1737         struct btrfs_dir_item *log_di;
1738         int name_len;
1739         unsigned long ptr;
1740         unsigned long ptr_end;
1741         char *name;
1742         struct inode *inode;
1743         struct btrfs_key location;
1744
1745 again:
1746         eb = path->nodes[0];
1747         slot = path->slots[0];
1748         item_size = btrfs_item_size_nr(eb, slot);
1749         ptr = btrfs_item_ptr_offset(eb, slot);
1750         ptr_end = ptr + item_size;
1751         while (ptr < ptr_end) {
1752                 di = (struct btrfs_dir_item *)ptr;
1753                 if (verify_dir_item(root, eb, di)) {
1754                         ret = -EIO;
1755                         goto out;
1756                 }
1757
1758                 name_len = btrfs_dir_name_len(eb, di);
1759                 name = kmalloc(name_len, GFP_NOFS);
1760                 if (!name) {
1761                         ret = -ENOMEM;
1762                         goto out;
1763                 }
1764                 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1765                                   name_len);
1766                 log_di = NULL;
1767                 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1768                         log_di = btrfs_lookup_dir_item(trans, log, log_path,
1769                                                        dir_key->objectid,
1770                                                        name, name_len, 0);
1771                 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1772                         log_di = btrfs_lookup_dir_index_item(trans, log,
1773                                                      log_path,
1774                                                      dir_key->objectid,
1775                                                      dir_key->offset,
1776                                                      name, name_len, 0);
1777                 }
1778                 if (IS_ERR_OR_NULL(log_di)) {
1779                         btrfs_dir_item_key_to_cpu(eb, di, &location);
1780                         btrfs_release_path(path);
1781                         btrfs_release_path(log_path);
1782                         inode = read_one_inode(root, location.objectid);
1783                         if (!inode) {
1784                                 kfree(name);
1785                                 return -EIO;
1786                         }
1787
1788                         ret = link_to_fixup_dir(trans, root,
1789                                                 path, location.objectid);
1790                         if (ret) {
1791                                 kfree(name);
1792                                 iput(inode);
1793                                 goto out;
1794                         }
1795
1796                         btrfs_inc_nlink(inode);
1797                         ret = btrfs_unlink_inode(trans, root, dir, inode,
1798                                                  name, name_len);
1799                         if (!ret)
1800                                 btrfs_run_delayed_items(trans, root);
1801                         kfree(name);
1802                         iput(inode);
1803                         if (ret)
1804                                 goto out;
1805
1806                         /* there might still be more names under this key
1807                          * check and repeat if required
1808                          */
1809                         ret = btrfs_search_slot(NULL, root, dir_key, path,
1810                                                 0, 0);
1811                         if (ret == 0)
1812                                 goto again;
1813                         ret = 0;
1814                         goto out;
1815                 }
1816                 btrfs_release_path(log_path);
1817                 kfree(name);
1818
1819                 ptr = (unsigned long)(di + 1);
1820                 ptr += name_len;
1821         }
1822         ret = 0;
1823 out:
1824         btrfs_release_path(path);
1825         btrfs_release_path(log_path);
1826         return ret;
1827 }
1828
1829 /*
1830  * deletion replay happens before we copy any new directory items
1831  * out of the log or out of backreferences from inodes.  It
1832  * scans the log to find ranges of keys that log is authoritative for,
1833  * and then scans the directory to find items in those ranges that are
1834  * not present in the log.
1835  *
1836  * Anything we don't find in the log is unlinked and removed from the
1837  * directory.
1838  */
1839 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1840                                        struct btrfs_root *root,
1841                                        struct btrfs_root *log,
1842                                        struct btrfs_path *path,
1843                                        u64 dirid, int del_all)
1844 {
1845         u64 range_start;
1846         u64 range_end;
1847         int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1848         int ret = 0;
1849         struct btrfs_key dir_key;
1850         struct btrfs_key found_key;
1851         struct btrfs_path *log_path;
1852         struct inode *dir;
1853
1854         dir_key.objectid = dirid;
1855         dir_key.type = BTRFS_DIR_ITEM_KEY;
1856         log_path = btrfs_alloc_path();
1857         if (!log_path)
1858                 return -ENOMEM;
1859
1860         dir = read_one_inode(root, dirid);
1861         /* it isn't an error if the inode isn't there, that can happen
1862          * because we replay the deletes before we copy in the inode item
1863          * from the log
1864          */
1865         if (!dir) {
1866                 btrfs_free_path(log_path);
1867                 return 0;
1868         }
1869 again:
1870         range_start = 0;
1871         range_end = 0;
1872         while (1) {
1873                 if (del_all)
1874                         range_end = (u64)-1;
1875                 else {
1876                         ret = find_dir_range(log, path, dirid, key_type,
1877                                              &range_start, &range_end);
1878                         if (ret != 0)
1879                                 break;
1880                 }
1881
1882                 dir_key.offset = range_start;
1883                 while (1) {
1884                         int nritems;
1885                         ret = btrfs_search_slot(NULL, root, &dir_key, path,
1886                                                 0, 0);
1887                         if (ret < 0)
1888                                 goto out;
1889
1890                         nritems = btrfs_header_nritems(path->nodes[0]);
1891                         if (path->slots[0] >= nritems) {
1892                                 ret = btrfs_next_leaf(root, path);
1893                                 if (ret)
1894                                         break;
1895                         }
1896                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1897                                               path->slots[0]);
1898                         if (found_key.objectid != dirid ||
1899                             found_key.type != dir_key.type)
1900                                 goto next_type;
1901
1902                         if (found_key.offset > range_end)
1903                                 break;
1904
1905                         ret = check_item_in_log(trans, root, log, path,
1906                                                 log_path, dir,
1907                                                 &found_key);
1908                         if (ret)
1909                                 goto out;
1910                         if (found_key.offset == (u64)-1)
1911                                 break;
1912                         dir_key.offset = found_key.offset + 1;
1913                 }
1914                 btrfs_release_path(path);
1915                 if (range_end == (u64)-1)
1916                         break;
1917                 range_start = range_end + 1;
1918         }
1919
1920 next_type:
1921         ret = 0;
1922         if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1923                 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1924                 dir_key.type = BTRFS_DIR_INDEX_KEY;
1925                 btrfs_release_path(path);
1926                 goto again;
1927         }
1928 out:
1929         btrfs_release_path(path);
1930         btrfs_free_path(log_path);
1931         iput(dir);
1932         return ret;
1933 }
1934
1935 /*
1936  * the process_func used to replay items from the log tree.  This
1937  * gets called in two different stages.  The first stage just looks
1938  * for inodes and makes sure they are all copied into the subvolume.
1939  *
1940  * The second stage copies all the other item types from the log into
1941  * the subvolume.  The two stage approach is slower, but gets rid of
1942  * lots of complexity around inodes referencing other inodes that exist
1943  * only in the log (references come from either directory items or inode
1944  * back refs).
1945  */
1946 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1947                              struct walk_control *wc, u64 gen)
1948 {
1949         int nritems;
1950         struct btrfs_path *path;
1951         struct btrfs_root *root = wc->replay_dest;
1952         struct btrfs_key key;
1953         int level;
1954         int i;
1955         int ret;
1956
1957         ret = btrfs_read_buffer(eb, gen);
1958         if (ret)
1959                 return ret;
1960
1961         level = btrfs_header_level(eb);
1962
1963         if (level != 0)
1964                 return 0;
1965
1966         path = btrfs_alloc_path();
1967         if (!path)
1968                 return -ENOMEM;
1969
1970         nritems = btrfs_header_nritems(eb);
1971         for (i = 0; i < nritems; i++) {
1972                 btrfs_item_key_to_cpu(eb, &key, i);
1973
1974                 /* inode keys are done during the first stage */
1975                 if (key.type == BTRFS_INODE_ITEM_KEY &&
1976                     wc->stage == LOG_WALK_REPLAY_INODES) {
1977                         struct btrfs_inode_item *inode_item;
1978                         u32 mode;
1979
1980                         inode_item = btrfs_item_ptr(eb, i,
1981                                             struct btrfs_inode_item);
1982                         mode = btrfs_inode_mode(eb, inode_item);
1983                         if (S_ISDIR(mode)) {
1984                                 ret = replay_dir_deletes(wc->trans,
1985                                          root, log, path, key.objectid, 0);
1986                                 if (ret)
1987                                         break;
1988                         }
1989                         ret = overwrite_item(wc->trans, root, path,
1990                                              eb, i, &key);
1991                         if (ret)
1992                                 break;
1993
1994                         /* for regular files, make sure corresponding
1995                          * orhpan item exist. extents past the new EOF
1996                          * will be truncated later by orphan cleanup.
1997                          */
1998                         if (S_ISREG(mode)) {
1999                                 ret = insert_orphan_item(wc->trans, root,
2000                                                          key.objectid);
2001                                 if (ret)
2002                                         break;
2003                         }
2004
2005                         ret = link_to_fixup_dir(wc->trans, root,
2006                                                 path, key.objectid);
2007                         if (ret)
2008                                 break;
2009                 }
2010                 if (wc->stage < LOG_WALK_REPLAY_ALL)
2011                         continue;
2012
2013                 /* these keys are simply copied */
2014                 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2015                         ret = overwrite_item(wc->trans, root, path,
2016                                              eb, i, &key);
2017                         if (ret)
2018                                 break;
2019                 } else if (key.type == BTRFS_INODE_REF_KEY) {
2020                         ret = add_inode_ref(wc->trans, root, log, path,
2021                                             eb, i, &key);
2022                         if (ret && ret != -ENOENT)
2023                                 break;
2024                         ret = 0;
2025                 } else if (key.type == BTRFS_INODE_EXTREF_KEY) {
2026                         ret = add_inode_ref(wc->trans, root, log, path,
2027                                             eb, i, &key);
2028                         if (ret && ret != -ENOENT)
2029                                 break;
2030                         ret = 0;
2031                 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2032                         ret = replay_one_extent(wc->trans, root, path,
2033                                                 eb, i, &key);
2034                         if (ret)
2035                                 break;
2036                 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
2037                            key.type == BTRFS_DIR_INDEX_KEY) {
2038                         ret = replay_one_dir_item(wc->trans, root, path,
2039                                                   eb, i, &key);
2040                         if (ret)
2041                                 break;
2042                 }
2043         }
2044         btrfs_free_path(path);
2045         return ret;
2046 }
2047
2048 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2049                                    struct btrfs_root *root,
2050                                    struct btrfs_path *path, int *level,
2051                                    struct walk_control *wc)
2052 {
2053         u64 root_owner;
2054         u64 bytenr;
2055         u64 ptr_gen;
2056         struct extent_buffer *next;
2057         struct extent_buffer *cur;
2058         struct extent_buffer *parent;
2059         u32 blocksize;
2060         int ret = 0;
2061
2062         WARN_ON(*level < 0);
2063         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2064
2065         while (*level > 0) {
2066                 WARN_ON(*level < 0);
2067                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2068                 cur = path->nodes[*level];
2069
2070                 if (btrfs_header_level(cur) != *level)
2071                         WARN_ON(1);
2072
2073                 if (path->slots[*level] >=
2074                     btrfs_header_nritems(cur))
2075                         break;
2076
2077                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2078                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2079                 blocksize = btrfs_level_size(root, *level - 1);
2080
2081                 parent = path->nodes[*level];
2082                 root_owner = btrfs_header_owner(parent);
2083
2084                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
2085                 if (!next)
2086                         return -ENOMEM;
2087
2088                 if (*level == 1) {
2089                         ret = wc->process_func(root, next, wc, ptr_gen);
2090                         if (ret) {
2091                                 free_extent_buffer(next);
2092                                 return ret;
2093                         }
2094
2095                         path->slots[*level]++;
2096                         if (wc->free) {
2097                                 ret = btrfs_read_buffer(next, ptr_gen);
2098                                 if (ret) {
2099                                         free_extent_buffer(next);
2100                                         return ret;
2101                                 }
2102
2103                                 btrfs_tree_lock(next);
2104                                 btrfs_set_lock_blocking(next);
2105                                 clean_tree_block(trans, root, next);
2106                                 btrfs_wait_tree_block_writeback(next);
2107                                 btrfs_tree_unlock(next);
2108
2109                                 WARN_ON(root_owner !=
2110                                         BTRFS_TREE_LOG_OBJECTID);
2111                                 ret = btrfs_free_and_pin_reserved_extent(root,
2112                                                          bytenr, blocksize);
2113                                 if (ret) {
2114                                         free_extent_buffer(next);
2115                                         return ret;
2116                                 }
2117                         }
2118                         free_extent_buffer(next);
2119                         continue;
2120                 }
2121                 ret = btrfs_read_buffer(next, ptr_gen);
2122                 if (ret) {
2123                         free_extent_buffer(next);
2124                         return ret;
2125                 }
2126
2127                 WARN_ON(*level <= 0);
2128                 if (path->nodes[*level-1])
2129                         free_extent_buffer(path->nodes[*level-1]);
2130                 path->nodes[*level-1] = next;
2131                 *level = btrfs_header_level(next);
2132                 path->slots[*level] = 0;
2133                 cond_resched();
2134         }
2135         WARN_ON(*level < 0);
2136         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2137
2138         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2139
2140         cond_resched();
2141         return 0;
2142 }
2143
2144 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2145                                  struct btrfs_root *root,
2146                                  struct btrfs_path *path, int *level,
2147                                  struct walk_control *wc)
2148 {
2149         u64 root_owner;
2150         int i;
2151         int slot;
2152         int ret;
2153
2154         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2155                 slot = path->slots[i];
2156                 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2157                         path->slots[i]++;
2158                         *level = i;
2159                         WARN_ON(*level == 0);
2160                         return 0;
2161                 } else {
2162                         struct extent_buffer *parent;
2163                         if (path->nodes[*level] == root->node)
2164                                 parent = path->nodes[*level];
2165                         else
2166                                 parent = path->nodes[*level + 1];
2167
2168                         root_owner = btrfs_header_owner(parent);
2169                         ret = wc->process_func(root, path->nodes[*level], wc,
2170                                  btrfs_header_generation(path->nodes[*level]));
2171                         if (ret)
2172                                 return ret;
2173
2174                         if (wc->free) {
2175                                 struct extent_buffer *next;
2176
2177                                 next = path->nodes[*level];
2178
2179                                 btrfs_tree_lock(next);
2180                                 btrfs_set_lock_blocking(next);
2181                                 clean_tree_block(trans, root, next);
2182                                 btrfs_wait_tree_block_writeback(next);
2183                                 btrfs_tree_unlock(next);
2184
2185                                 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2186                                 ret = btrfs_free_and_pin_reserved_extent(root,
2187                                                 path->nodes[*level]->start,
2188                                                 path->nodes[*level]->len);
2189                                 if (ret)
2190                                         return ret;
2191                         }
2192                         free_extent_buffer(path->nodes[*level]);
2193                         path->nodes[*level] = NULL;
2194                         *level = i + 1;
2195                 }
2196         }
2197         return 1;
2198 }
2199
2200 /*
2201  * drop the reference count on the tree rooted at 'snap'.  This traverses
2202  * the tree freeing any blocks that have a ref count of zero after being
2203  * decremented.
2204  */
2205 static int walk_log_tree(struct btrfs_trans_handle *trans,
2206                          struct btrfs_root *log, struct walk_control *wc)
2207 {
2208         int ret = 0;
2209         int wret;
2210         int level;
2211         struct btrfs_path *path;
2212         int orig_level;
2213
2214         path = btrfs_alloc_path();
2215         if (!path)
2216                 return -ENOMEM;
2217
2218         level = btrfs_header_level(log->node);
2219         orig_level = level;
2220         path->nodes[level] = log->node;
2221         extent_buffer_get(log->node);
2222         path->slots[level] = 0;
2223
2224         while (1) {
2225                 wret = walk_down_log_tree(trans, log, path, &level, wc);
2226                 if (wret > 0)
2227                         break;
2228                 if (wret < 0) {
2229                         ret = wret;
2230                         goto out;
2231                 }
2232
2233                 wret = walk_up_log_tree(trans, log, path, &level, wc);
2234                 if (wret > 0)
2235                         break;
2236                 if (wret < 0) {
2237                         ret = wret;
2238                         goto out;
2239                 }
2240         }
2241
2242         /* was the root node processed? if not, catch it here */
2243         if (path->nodes[orig_level]) {
2244                 ret = wc->process_func(log, path->nodes[orig_level], wc,
2245                          btrfs_header_generation(path->nodes[orig_level]));
2246                 if (ret)
2247                         goto out;
2248                 if (wc->free) {
2249                         struct extent_buffer *next;
2250
2251                         next = path->nodes[orig_level];
2252
2253                         btrfs_tree_lock(next);
2254                         btrfs_set_lock_blocking(next);
2255                         clean_tree_block(trans, log, next);
2256                         btrfs_wait_tree_block_writeback(next);
2257                         btrfs_tree_unlock(next);
2258
2259                         WARN_ON(log->root_key.objectid !=
2260                                 BTRFS_TREE_LOG_OBJECTID);
2261                         ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2262                                                          next->len);
2263                         if (ret)
2264                                 goto out;
2265                 }
2266         }
2267
2268 out:
2269         btrfs_free_path(path);
2270         return ret;
2271 }
2272
2273 /*
2274  * helper function to update the item for a given subvolumes log root
2275  * in the tree of log roots
2276  */
2277 static int update_log_root(struct btrfs_trans_handle *trans,
2278                            struct btrfs_root *log)
2279 {
2280         int ret;
2281
2282         if (log->log_transid == 1) {
2283                 /* insert root item on the first sync */
2284                 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2285                                 &log->root_key, &log->root_item);
2286         } else {
2287                 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2288                                 &log->root_key, &log->root_item);
2289         }
2290         return ret;
2291 }
2292
2293 static int wait_log_commit(struct btrfs_trans_handle *trans,
2294                            struct btrfs_root *root, unsigned long transid)
2295 {
2296         DEFINE_WAIT(wait);
2297         int index = transid % 2;
2298
2299         /*
2300          * we only allow two pending log transactions at a time,
2301          * so we know that if ours is more than 2 older than the
2302          * current transaction, we're done
2303          */
2304         do {
2305                 prepare_to_wait(&root->log_commit_wait[index],
2306                                 &wait, TASK_UNINTERRUPTIBLE);
2307                 mutex_unlock(&root->log_mutex);
2308
2309                 if (root->fs_info->last_trans_log_full_commit !=
2310                     trans->transid && root->log_transid < transid + 2 &&
2311                     atomic_read(&root->log_commit[index]))
2312                         schedule();
2313
2314                 finish_wait(&root->log_commit_wait[index], &wait);
2315                 mutex_lock(&root->log_mutex);
2316         } while (root->fs_info->last_trans_log_full_commit !=
2317                  trans->transid && root->log_transid < transid + 2 &&
2318                  atomic_read(&root->log_commit[index]));
2319         return 0;
2320 }
2321
2322 static void wait_for_writer(struct btrfs_trans_handle *trans,
2323                             struct btrfs_root *root)
2324 {
2325         DEFINE_WAIT(wait);
2326         while (root->fs_info->last_trans_log_full_commit !=
2327                trans->transid && atomic_read(&root->log_writers)) {
2328                 prepare_to_wait(&root->log_writer_wait,
2329                                 &wait, TASK_UNINTERRUPTIBLE);
2330                 mutex_unlock(&root->log_mutex);
2331                 if (root->fs_info->last_trans_log_full_commit !=
2332                     trans->transid && atomic_read(&root->log_writers))
2333                         schedule();
2334                 mutex_lock(&root->log_mutex);
2335                 finish_wait(&root->log_writer_wait, &wait);
2336         }
2337 }
2338
2339 /*
2340  * btrfs_sync_log does sends a given tree log down to the disk and
2341  * updates the super blocks to record it.  When this call is done,
2342  * you know that any inodes previously logged are safely on disk only
2343  * if it returns 0.
2344  *
2345  * Any other return value means you need to call btrfs_commit_transaction.
2346  * Some of the edge cases for fsyncing directories that have had unlinks
2347  * or renames done in the past mean that sometimes the only safe
2348  * fsync is to commit the whole FS.  When btrfs_sync_log returns -EAGAIN,
2349  * that has happened.
2350  */
2351 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2352                    struct btrfs_root *root)
2353 {
2354         int index1;
2355         int index2;
2356         int mark;
2357         int ret;
2358         struct btrfs_root *log = root->log_root;
2359         struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2360         unsigned long log_transid = 0;
2361
2362         mutex_lock(&root->log_mutex);
2363         log_transid = root->log_transid;
2364         index1 = root->log_transid % 2;
2365         if (atomic_read(&root->log_commit[index1])) {
2366                 wait_log_commit(trans, root, root->log_transid);
2367                 mutex_unlock(&root->log_mutex);
2368                 return 0;
2369         }
2370         atomic_set(&root->log_commit[index1], 1);
2371
2372         /* wait for previous tree log sync to complete */
2373         if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2374                 wait_log_commit(trans, root, root->log_transid - 1);
2375         while (1) {
2376                 int batch = atomic_read(&root->log_batch);
2377                 /* when we're on an ssd, just kick the log commit out */
2378                 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
2379                         mutex_unlock(&root->log_mutex);
2380                         schedule_timeout_uninterruptible(1);
2381                         mutex_lock(&root->log_mutex);
2382                 }
2383                 wait_for_writer(trans, root);
2384                 if (batch == atomic_read(&root->log_batch))
2385                         break;
2386         }
2387
2388         /* bail out if we need to do a full commit */
2389         if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2390                 ret = -EAGAIN;
2391                 btrfs_free_logged_extents(log, log_transid);
2392                 mutex_unlock(&root->log_mutex);
2393                 goto out;
2394         }
2395
2396         if (log_transid % 2 == 0)
2397                 mark = EXTENT_DIRTY;
2398         else
2399                 mark = EXTENT_NEW;
2400
2401         /* we start IO on  all the marked extents here, but we don't actually
2402          * wait for them until later.
2403          */
2404         ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2405         if (ret) {
2406                 btrfs_abort_transaction(trans, root, ret);
2407                 btrfs_free_logged_extents(log, log_transid);
2408                 mutex_unlock(&root->log_mutex);
2409                 goto out;
2410         }
2411
2412         btrfs_set_root_node(&log->root_item, log->node);
2413
2414         root->log_transid++;
2415         log->log_transid = root->log_transid;
2416         root->log_start_pid = 0;
2417         smp_mb();
2418         /*
2419          * IO has been started, blocks of the log tree have WRITTEN flag set
2420          * in their headers. new modifications of the log will be written to
2421          * new positions. so it's safe to allow log writers to go in.
2422          */
2423         mutex_unlock(&root->log_mutex);
2424
2425         mutex_lock(&log_root_tree->log_mutex);
2426         atomic_inc(&log_root_tree->log_batch);
2427         atomic_inc(&log_root_tree->log_writers);
2428         mutex_unlock(&log_root_tree->log_mutex);
2429
2430         ret = update_log_root(trans, log);
2431
2432         mutex_lock(&log_root_tree->log_mutex);
2433         if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2434                 smp_mb();
2435                 if (waitqueue_active(&log_root_tree->log_writer_wait))
2436                         wake_up(&log_root_tree->log_writer_wait);
2437         }
2438
2439         if (ret) {
2440                 if (ret != -ENOSPC) {
2441                         btrfs_abort_transaction(trans, root, ret);
2442                         mutex_unlock(&log_root_tree->log_mutex);
2443                         goto out;
2444                 }
2445                 root->fs_info->last_trans_log_full_commit = trans->transid;
2446                 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2447                 btrfs_free_logged_extents(log, log_transid);
2448                 mutex_unlock(&log_root_tree->log_mutex);
2449                 ret = -EAGAIN;
2450                 goto out;
2451         }
2452
2453         index2 = log_root_tree->log_transid % 2;
2454         if (atomic_read(&log_root_tree->log_commit[index2])) {
2455                 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2456                 wait_log_commit(trans, log_root_tree,
2457                                 log_root_tree->log_transid);
2458                 btrfs_free_logged_extents(log, log_transid);
2459                 mutex_unlock(&log_root_tree->log_mutex);
2460                 ret = 0;
2461                 goto out;
2462         }
2463         atomic_set(&log_root_tree->log_commit[index2], 1);
2464
2465         if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2466                 wait_log_commit(trans, log_root_tree,
2467                                 log_root_tree->log_transid - 1);
2468         }
2469
2470         wait_for_writer(trans, log_root_tree);
2471
2472         /*
2473          * now that we've moved on to the tree of log tree roots,
2474          * check the full commit flag again
2475          */
2476         if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2477                 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2478                 btrfs_free_logged_extents(log, log_transid);
2479                 mutex_unlock(&log_root_tree->log_mutex);
2480                 ret = -EAGAIN;
2481                 goto out_wake_log_root;
2482         }
2483
2484         ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2485                                 &log_root_tree->dirty_log_pages,
2486                                 EXTENT_DIRTY | EXTENT_NEW);
2487         if (ret) {
2488                 btrfs_abort_transaction(trans, root, ret);
2489                 btrfs_free_logged_extents(log, log_transid);
2490                 mutex_unlock(&log_root_tree->log_mutex);
2491                 goto out_wake_log_root;
2492         }
2493         btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2494         btrfs_wait_logged_extents(log, log_transid);
2495
2496         btrfs_set_super_log_root(root->fs_info->super_for_commit,
2497                                 log_root_tree->node->start);
2498         btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2499                                 btrfs_header_level(log_root_tree->node));
2500
2501         log_root_tree->log_transid++;
2502         smp_mb();
2503
2504         mutex_unlock(&log_root_tree->log_mutex);
2505
2506         /*
2507          * nobody else is going to jump in and write the the ctree
2508          * super here because the log_commit atomic below is protecting
2509          * us.  We must be called with a transaction handle pinning
2510          * the running transaction open, so a full commit can't hop
2511          * in and cause problems either.
2512          */
2513         btrfs_scrub_pause_super(root);
2514         ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2515         btrfs_scrub_continue_super(root);
2516         if (ret) {
2517                 btrfs_abort_transaction(trans, root, ret);
2518                 goto out_wake_log_root;
2519         }
2520
2521         mutex_lock(&root->log_mutex);
2522         if (root->last_log_commit < log_transid)
2523                 root->last_log_commit = log_transid;
2524         mutex_unlock(&root->log_mutex);
2525
2526 out_wake_log_root:
2527         atomic_set(&log_root_tree->log_commit[index2], 0);
2528         smp_mb();
2529         if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2530                 wake_up(&log_root_tree->log_commit_wait[index2]);
2531 out:
2532         atomic_set(&root->log_commit[index1], 0);
2533         smp_mb();
2534         if (waitqueue_active(&root->log_commit_wait[index1]))
2535                 wake_up(&root->log_commit_wait[index1]);
2536         return ret;
2537 }
2538
2539 static void free_log_tree(struct btrfs_trans_handle *trans,
2540                           struct btrfs_root *log)
2541 {
2542         int ret;
2543         u64 start;
2544         u64 end;
2545         struct walk_control wc = {
2546                 .free = 1,
2547                 .process_func = process_one_buffer
2548         };
2549
2550         if (trans) {
2551                 ret = walk_log_tree(trans, log, &wc);
2552
2553                 /* I don't think this can happen but just in case */
2554                 if (ret)
2555                         btrfs_abort_transaction(trans, log, ret);
2556         }
2557
2558         while (1) {
2559                 ret = find_first_extent_bit(&log->dirty_log_pages,
2560                                 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2561                                 NULL);
2562                 if (ret)
2563                         break;
2564
2565                 clear_extent_bits(&log->dirty_log_pages, start, end,
2566                                   EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2567         }
2568
2569         /*
2570          * We may have short-circuited the log tree with the full commit logic
2571          * and left ordered extents on our list, so clear these out to keep us
2572          * from leaking inodes and memory.
2573          */
2574         btrfs_free_logged_extents(log, 0);
2575         btrfs_free_logged_extents(log, 1);
2576
2577         free_extent_buffer(log->node);
2578         kfree(log);
2579 }
2580
2581 /*
2582  * free all the extents used by the tree log.  This should be called
2583  * at commit time of the full transaction
2584  */
2585 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2586 {
2587         if (root->log_root) {
2588                 free_log_tree(trans, root->log_root);
2589                 root->log_root = NULL;
2590         }
2591         return 0;
2592 }
2593
2594 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2595                              struct btrfs_fs_info *fs_info)
2596 {
2597         if (fs_info->log_root_tree) {
2598                 free_log_tree(trans, fs_info->log_root_tree);
2599                 fs_info->log_root_tree = NULL;
2600         }
2601         return 0;
2602 }
2603
2604 /*
2605  * If both a file and directory are logged, and unlinks or renames are
2606  * mixed in, we have a few interesting corners:
2607  *
2608  * create file X in dir Y
2609  * link file X to X.link in dir Y
2610  * fsync file X
2611  * unlink file X but leave X.link
2612  * fsync dir Y
2613  *
2614  * After a crash we would expect only X.link to exist.  But file X
2615  * didn't get fsync'd again so the log has back refs for X and X.link.
2616  *
2617  * We solve this by removing directory entries and inode backrefs from the
2618  * log when a file that was logged in the current transaction is
2619  * unlinked.  Any later fsync will include the updated log entries, and
2620  * we'll be able to reconstruct the proper directory items from backrefs.
2621  *
2622  * This optimizations allows us to avoid relogging the entire inode
2623  * or the entire directory.
2624  */
2625 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2626                                  struct btrfs_root *root,
2627                                  const char *name, int name_len,
2628                                  struct inode *dir, u64 index)
2629 {
2630         struct btrfs_root *log;
2631         struct btrfs_dir_item *di;
2632         struct btrfs_path *path;
2633         int ret;
2634         int err = 0;
2635         int bytes_del = 0;
2636         u64 dir_ino = btrfs_ino(dir);
2637
2638         if (BTRFS_I(dir)->logged_trans < trans->transid)
2639                 return 0;
2640
2641         ret = join_running_log_trans(root);
2642         if (ret)
2643                 return 0;
2644
2645         mutex_lock(&BTRFS_I(dir)->log_mutex);
2646
2647         log = root->log_root;
2648         path = btrfs_alloc_path();
2649         if (!path) {
2650                 err = -ENOMEM;
2651                 goto out_unlock;
2652         }
2653
2654         di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2655                                    name, name_len, -1);
2656         if (IS_ERR(di)) {
2657                 err = PTR_ERR(di);
2658                 goto fail;
2659         }
2660         if (di) {
2661                 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2662                 bytes_del += name_len;
2663                 if (ret) {
2664                         err = ret;
2665                         goto fail;
2666                 }
2667         }
2668         btrfs_release_path(path);
2669         di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2670                                          index, name, name_len, -1);
2671         if (IS_ERR(di)) {
2672                 err = PTR_ERR(di);
2673                 goto fail;
2674         }
2675         if (di) {
2676                 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2677                 bytes_del += name_len;
2678                 if (ret) {
2679                         err = ret;
2680                         goto fail;
2681                 }
2682         }
2683
2684         /* update the directory size in the log to reflect the names
2685          * we have removed
2686          */
2687         if (bytes_del) {
2688                 struct btrfs_key key;
2689
2690                 key.objectid = dir_ino;
2691                 key.offset = 0;
2692                 key.type = BTRFS_INODE_ITEM_KEY;
2693                 btrfs_release_path(path);
2694
2695                 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2696                 if (ret < 0) {
2697                         err = ret;
2698                         goto fail;
2699                 }
2700                 if (ret == 0) {
2701                         struct btrfs_inode_item *item;
2702                         u64 i_size;
2703
2704                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2705                                               struct btrfs_inode_item);
2706                         i_size = btrfs_inode_size(path->nodes[0], item);
2707                         if (i_size > bytes_del)
2708                                 i_size -= bytes_del;
2709                         else
2710                                 i_size = 0;
2711                         btrfs_set_inode_size(path->nodes[0], item, i_size);
2712                         btrfs_mark_buffer_dirty(path->nodes[0]);
2713                 } else
2714                         ret = 0;
2715                 btrfs_release_path(path);
2716         }
2717 fail:
2718         btrfs_free_path(path);
2719 out_unlock:
2720         mutex_unlock(&BTRFS_I(dir)->log_mutex);
2721         if (ret == -ENOSPC) {
2722                 root->fs_info->last_trans_log_full_commit = trans->transid;
2723                 ret = 0;
2724         } else if (ret < 0)
2725                 btrfs_abort_transaction(trans, root, ret);
2726
2727         btrfs_end_log_trans(root);
2728
2729         return err;
2730 }
2731
2732 /* see comments for btrfs_del_dir_entries_in_log */
2733 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2734                                struct btrfs_root *root,
2735                                const char *name, int name_len,
2736                                struct inode *inode, u64 dirid)
2737 {
2738         struct btrfs_root *log;
2739         u64 index;
2740         int ret;
2741
2742         if (BTRFS_I(inode)->logged_trans < trans->transid)
2743                 return 0;
2744
2745         ret = join_running_log_trans(root);
2746         if (ret)
2747                 return 0;
2748         log = root->log_root;
2749         mutex_lock(&BTRFS_I(inode)->log_mutex);
2750
2751         ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2752                                   dirid, &index);
2753         mutex_unlock(&BTRFS_I(inode)->log_mutex);
2754         if (ret == -ENOSPC) {
2755                 root->fs_info->last_trans_log_full_commit = trans->transid;
2756                 ret = 0;
2757         } else if (ret < 0 && ret != -ENOENT)
2758                 btrfs_abort_transaction(trans, root, ret);
2759         btrfs_end_log_trans(root);
2760
2761         return ret;
2762 }
2763
2764 /*
2765  * creates a range item in the log for 'dirid'.  first_offset and
2766  * last_offset tell us which parts of the key space the log should
2767  * be considered authoritative for.
2768  */
2769 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2770                                        struct btrfs_root *log,
2771                                        struct btrfs_path *path,
2772                                        int key_type, u64 dirid,
2773                                        u64 first_offset, u64 last_offset)
2774 {
2775         int ret;
2776         struct btrfs_key key;
2777         struct btrfs_dir_log_item *item;
2778
2779         key.objectid = dirid;
2780         key.offset = first_offset;
2781         if (key_type == BTRFS_DIR_ITEM_KEY)
2782                 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2783         else
2784                 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2785         ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2786         if (ret)
2787                 return ret;
2788
2789         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2790                               struct btrfs_dir_log_item);
2791         btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2792         btrfs_mark_buffer_dirty(path->nodes[0]);
2793         btrfs_release_path(path);
2794         return 0;
2795 }
2796
2797 /*
2798  * log all the items included in the current transaction for a given
2799  * directory.  This also creates the range items in the log tree required
2800  * to replay anything deleted before the fsync
2801  */
2802 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2803                           struct btrfs_root *root, struct inode *inode,
2804                           struct btrfs_path *path,
2805                           struct btrfs_path *dst_path, int key_type,
2806                           u64 min_offset, u64 *last_offset_ret)
2807 {
2808         struct btrfs_key min_key;
2809         struct btrfs_key max_key;
2810         struct btrfs_root *log = root->log_root;
2811         struct extent_buffer *src;
2812         int err = 0;
2813         int ret;
2814         int i;
2815         int nritems;
2816         u64 first_offset = min_offset;
2817         u64 last_offset = (u64)-1;
2818         u64 ino = btrfs_ino(inode);
2819
2820         log = root->log_root;
2821         max_key.objectid = ino;
2822         max_key.offset = (u64)-1;
2823         max_key.type = key_type;
2824
2825         min_key.objectid = ino;
2826         min_key.type = key_type;
2827         min_key.offset = min_offset;
2828
2829         path->keep_locks = 1;
2830
2831         ret = btrfs_search_forward(root, &min_key, &max_key,
2832                                    path, trans->transid);
2833
2834         /*
2835          * we didn't find anything from this transaction, see if there
2836          * is anything at all
2837          */
2838         if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2839                 min_key.objectid = ino;
2840                 min_key.type = key_type;
2841                 min_key.offset = (u64)-1;
2842                 btrfs_release_path(path);
2843                 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2844                 if (ret < 0) {
2845                         btrfs_release_path(path);
2846                         return ret;
2847                 }
2848                 ret = btrfs_previous_item(root, path, ino, key_type);
2849
2850                 /* if ret == 0 there are items for this type,
2851                  * create a range to tell us the last key of this type.
2852                  * otherwise, there are no items in this directory after
2853                  * *min_offset, and we create a range to indicate that.
2854                  */
2855                 if (ret == 0) {
2856                         struct btrfs_key tmp;
2857                         btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2858                                               path->slots[0]);
2859                         if (key_type == tmp.type)
2860                                 first_offset = max(min_offset, tmp.offset) + 1;
2861                 }
2862                 goto done;
2863         }
2864
2865         /* go backward to find any previous key */
2866         ret = btrfs_previous_item(root, path, ino, key_type);
2867         if (ret == 0) {
2868                 struct btrfs_key tmp;
2869                 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2870                 if (key_type == tmp.type) {
2871                         first_offset = tmp.offset;
2872                         ret = overwrite_item(trans, log, dst_path,
2873                                              path->nodes[0], path->slots[0],
2874                                              &tmp);
2875                         if (ret) {
2876                                 err = ret;
2877                                 goto done;
2878                         }
2879                 }
2880         }
2881         btrfs_release_path(path);
2882
2883         /* find the first key from this transaction again */
2884         ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2885         if (ret != 0) {
2886                 WARN_ON(1);
2887                 goto done;
2888         }
2889
2890         /*
2891          * we have a block from this transaction, log every item in it
2892          * from our directory
2893          */
2894         while (1) {
2895                 struct btrfs_key tmp;
2896                 src = path->nodes[0];
2897                 nritems = btrfs_header_nritems(src);
2898                 for (i = path->slots[0]; i < nritems; i++) {
2899                         btrfs_item_key_to_cpu(src, &min_key, i);
2900
2901                         if (min_key.objectid != ino || min_key.type != key_type)
2902                                 goto done;
2903                         ret = overwrite_item(trans, log, dst_path, src, i,
2904                                              &min_key);
2905                         if (ret) {
2906                                 err = ret;
2907                                 goto done;
2908                         }
2909                 }
2910                 path->slots[0] = nritems;
2911
2912                 /*
2913                  * look ahead to the next item and see if it is also
2914                  * from this directory and from this transaction
2915                  */
2916                 ret = btrfs_next_leaf(root, path);
2917                 if (ret == 1) {
2918                         last_offset = (u64)-1;
2919                         goto done;
2920                 }
2921                 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2922                 if (tmp.objectid != ino || tmp.type != key_type) {
2923                         last_offset = (u64)-1;
2924                         goto done;
2925                 }
2926                 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2927                         ret = overwrite_item(trans, log, dst_path,
2928                                              path->nodes[0], path->slots[0],
2929                                              &tmp);
2930                         if (ret)
2931                                 err = ret;
2932                         else
2933                                 last_offset = tmp.offset;
2934                         goto done;
2935                 }
2936         }
2937 done:
2938         btrfs_release_path(path);
2939         btrfs_release_path(dst_path);
2940
2941         if (err == 0) {
2942                 *last_offset_ret = last_offset;
2943                 /*
2944                  * insert the log range keys to indicate where the log
2945                  * is valid
2946                  */
2947                 ret = insert_dir_log_key(trans, log, path, key_type,
2948                                          ino, first_offset, last_offset);
2949                 if (ret)
2950                         err = ret;
2951         }
2952         return err;
2953 }
2954
2955 /*
2956  * logging directories is very similar to logging inodes, We find all the items
2957  * from the current transaction and write them to the log.
2958  *
2959  * The recovery code scans the directory in the subvolume, and if it finds a
2960  * key in the range logged that is not present in the log tree, then it means
2961  * that dir entry was unlinked during the transaction.
2962  *
2963  * In order for that scan to work, we must include one key smaller than
2964  * the smallest logged by this transaction and one key larger than the largest
2965  * key logged by this transaction.
2966  */
2967 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2968                           struct btrfs_root *root, struct inode *inode,
2969                           struct btrfs_path *path,
2970                           struct btrfs_path *dst_path)
2971 {
2972         u64 min_key;
2973         u64 max_key;
2974         int ret;
2975         int key_type = BTRFS_DIR_ITEM_KEY;
2976
2977 again:
2978         min_key = 0;
2979         max_key = 0;
2980         while (1) {
2981                 ret = log_dir_items(trans, root, inode, path,
2982                                     dst_path, key_type, min_key,
2983                                     &max_key);
2984                 if (ret)
2985                         return ret;
2986                 if (max_key == (u64)-1)
2987                         break;
2988                 min_key = max_key + 1;
2989         }
2990
2991         if (key_type == BTRFS_DIR_ITEM_KEY) {
2992                 key_type = BTRFS_DIR_INDEX_KEY;
2993                 goto again;
2994         }
2995         return 0;
2996 }
2997
2998 /*
2999  * a helper function to drop items from the log before we relog an
3000  * inode.  max_key_type indicates the highest item type to remove.
3001  * This cannot be run for file data extents because it does not
3002  * free the extents they point to.
3003  */
3004 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3005                                   struct btrfs_root *log,
3006                                   struct btrfs_path *path,
3007                                   u64 objectid, int max_key_type)
3008 {
3009         int ret;
3010         struct btrfs_key key;
3011         struct btrfs_key found_key;
3012         int start_slot;
3013
3014         key.objectid = objectid;
3015         key.type = max_key_type;
3016         key.offset = (u64)-1;
3017
3018         while (1) {
3019                 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3020                 BUG_ON(ret == 0); /* Logic error */
3021                 if (ret < 0)
3022                         break;
3023
3024                 if (path->slots[0] == 0)
3025                         break;
3026
3027                 path->slots[0]--;
3028                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3029                                       path->slots[0]);
3030
3031                 if (found_key.objectid != objectid)
3032                         break;
3033
3034                 found_key.offset = 0;
3035                 found_key.type = 0;
3036                 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3037                                        &start_slot);
3038
3039                 ret = btrfs_del_items(trans, log, path, start_slot,
3040                                       path->slots[0] - start_slot + 1);
3041                 /*
3042                  * If start slot isn't 0 then we don't need to re-search, we've
3043                  * found the last guy with the objectid in this tree.
3044                  */
3045                 if (ret || start_slot != 0)
3046                         break;
3047                 btrfs_release_path(path);
3048         }
3049         btrfs_release_path(path);
3050         if (ret > 0)
3051                 ret = 0;
3052         return ret;
3053 }
3054
3055 static void fill_inode_item(struct btrfs_trans_handle *trans,
3056                             struct extent_buffer *leaf,
3057                             struct btrfs_inode_item *item,
3058                             struct inode *inode, int log_inode_only)
3059 {
3060         struct btrfs_map_token token;
3061
3062         btrfs_init_map_token(&token);
3063
3064         if (log_inode_only) {
3065                 /* set the generation to zero so the recover code
3066                  * can tell the difference between an logging
3067                  * just to say 'this inode exists' and a logging
3068                  * to say 'update this inode with these values'
3069                  */
3070                 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3071                 btrfs_set_token_inode_size(leaf, item, 0, &token);
3072         } else {
3073                 btrfs_set_token_inode_generation(leaf, item,
3074                                                  BTRFS_I(inode)->generation,
3075                                                  &token);
3076                 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3077         }
3078
3079         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3080         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3081         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3082         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3083
3084         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3085                                      inode->i_atime.tv_sec, &token);
3086         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3087                                       inode->i_atime.tv_nsec, &token);
3088
3089         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3090                                      inode->i_mtime.tv_sec, &token);
3091         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3092                                       inode->i_mtime.tv_nsec, &token);
3093
3094         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3095                                      inode->i_ctime.tv_sec, &token);
3096         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3097                                       inode->i_ctime.tv_nsec, &token);
3098
3099         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3100                                      &token);
3101
3102         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3103         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3104         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3105         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3106         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3107 }
3108
3109 static int log_inode_item(struct btrfs_trans_handle *trans,
3110                           struct btrfs_root *log, struct btrfs_path *path,
3111                           struct inode *inode)
3112 {
3113         struct btrfs_inode_item *inode_item;
3114         struct btrfs_key key;
3115         int ret;
3116
3117         memcpy(&key, &BTRFS_I(inode)->location, sizeof(key));
3118         ret = btrfs_insert_empty_item(trans, log, path, &key,
3119                                       sizeof(*inode_item));
3120         if (ret && ret != -EEXIST)
3121                 return ret;
3122         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3123                                     struct btrfs_inode_item);
3124         fill_inode_item(trans, path->nodes[0], inode_item, inode, 0);
3125         btrfs_release_path(path);
3126         return 0;
3127 }
3128
3129 static noinline int copy_items(struct btrfs_trans_handle *trans,
3130                                struct inode *inode,
3131                                struct btrfs_path *dst_path,
3132                                struct extent_buffer *src,
3133                                int start_slot, int nr, int inode_only)
3134 {
3135         unsigned long src_offset;
3136         unsigned long dst_offset;
3137         struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3138         struct btrfs_file_extent_item *extent;
3139         struct btrfs_inode_item *inode_item;
3140         int ret;
3141         struct btrfs_key *ins_keys;
3142         u32 *ins_sizes;
3143         char *ins_data;
3144         int i;
3145         struct list_head ordered_sums;
3146         int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3147
3148         INIT_LIST_HEAD(&ordered_sums);
3149
3150         ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3151                            nr * sizeof(u32), GFP_NOFS);
3152         if (!ins_data)
3153                 return -ENOMEM;
3154
3155         ins_sizes = (u32 *)ins_data;
3156         ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3157
3158         for (i = 0; i < nr; i++) {
3159                 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3160                 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3161         }
3162         ret = btrfs_insert_empty_items(trans, log, dst_path,
3163                                        ins_keys, ins_sizes, nr);
3164         if (ret) {
3165                 kfree(ins_data);
3166                 return ret;
3167         }
3168
3169         for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3170                 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3171                                                    dst_path->slots[0]);
3172
3173                 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3174
3175                 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3176                         inode_item = btrfs_item_ptr(dst_path->nodes[0],
3177                                                     dst_path->slots[0],
3178                                                     struct btrfs_inode_item);
3179                         fill_inode_item(trans, dst_path->nodes[0], inode_item,
3180                                         inode, inode_only == LOG_INODE_EXISTS);
3181                 } else {
3182                         copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3183                                            src_offset, ins_sizes[i]);
3184                 }
3185
3186                 /* take a reference on file data extents so that truncates
3187                  * or deletes of this inode don't have to relog the inode
3188                  * again
3189                  */
3190                 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY &&
3191                     !skip_csum) {
3192                         int found_type;
3193                         extent = btrfs_item_ptr(src, start_slot + i,
3194                                                 struct btrfs_file_extent_item);
3195
3196                         if (btrfs_file_extent_generation(src, extent) < trans->transid)
3197                                 continue;
3198
3199                         found_type = btrfs_file_extent_type(src, extent);
3200                         if (found_type == BTRFS_FILE_EXTENT_REG) {
3201                                 u64 ds, dl, cs, cl;
3202                                 ds = btrfs_file_extent_disk_bytenr(src,
3203                                                                 extent);
3204                                 /* ds == 0 is a hole */
3205                                 if (ds == 0)
3206                                         continue;
3207
3208                                 dl = btrfs_file_extent_disk_num_bytes(src,
3209                                                                 extent);
3210                                 cs = btrfs_file_extent_offset(src, extent);
3211                                 cl = btrfs_file_extent_num_bytes(src,
3212                                                                 extent);
3213                                 if (btrfs_file_extent_compression(src,
3214                                                                   extent)) {
3215                                         cs = 0;
3216                                         cl = dl;
3217                                 }
3218
3219                                 ret = btrfs_lookup_csums_range(
3220                                                 log->fs_info->csum_root,
3221                                                 ds + cs, ds + cs + cl - 1,
3222                                                 &ordered_sums, 0);
3223                                 if (ret) {
3224                                         btrfs_release_path(dst_path);
3225                                         kfree(ins_data);
3226                                         return ret;
3227                                 }
3228                         }
3229                 }
3230         }
3231
3232         btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3233         btrfs_release_path(dst_path);
3234         kfree(ins_data);
3235
3236         /*
3237          * we have to do this after the loop above to avoid changing the
3238          * log tree while trying to change the log tree.
3239          */
3240         ret = 0;
3241         while (!list_empty(&ordered_sums)) {
3242                 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3243                                                    struct btrfs_ordered_sum,
3244                                                    list);
3245                 if (!ret)
3246                         ret = btrfs_csum_file_blocks(trans, log, sums);
3247                 list_del(&sums->list);
3248                 kfree(sums);
3249         }
3250         return ret;
3251 }
3252
3253 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3254 {
3255         struct extent_map *em1, *em2;
3256
3257         em1 = list_entry(a, struct extent_map, list);
3258         em2 = list_entry(b, struct extent_map, list);
3259
3260         if (em1->start < em2->start)
3261                 return -1;
3262         else if (em1->start > em2->start)
3263                 return 1;
3264         return 0;
3265 }
3266
3267 static int log_one_extent(struct btrfs_trans_handle *trans,
3268                           struct inode *inode, struct btrfs_root *root,
3269                           struct extent_map *em, struct btrfs_path *path)
3270 {
3271         struct btrfs_root *log = root->log_root;
3272         struct btrfs_file_extent_item *fi;
3273         struct extent_buffer *leaf;
3274         struct btrfs_ordered_extent *ordered;
3275         struct list_head ordered_sums;
3276         struct btrfs_map_token token;
3277         struct btrfs_key key;
3278         u64 mod_start = em->mod_start;
3279         u64 mod_len = em->mod_len;
3280         u64 csum_offset;
3281         u64 csum_len;
3282         u64 extent_offset = em->start - em->orig_start;
3283         u64 block_len;
3284         int ret;
3285         int index = log->log_transid % 2;
3286         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3287
3288         ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
3289                                    em->start + em->len, NULL, 0);
3290         if (ret)
3291                 return ret;
3292
3293         INIT_LIST_HEAD(&ordered_sums);
3294         btrfs_init_map_token(&token);
3295         key.objectid = btrfs_ino(inode);
3296         key.type = BTRFS_EXTENT_DATA_KEY;
3297         key.offset = em->start;
3298
3299         ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*fi));
3300         if (ret)
3301                 return ret;
3302         leaf = path->nodes[0];
3303         fi = btrfs_item_ptr(leaf, path->slots[0],
3304                             struct btrfs_file_extent_item);
3305
3306         btrfs_set_token_file_extent_generation(leaf, fi, em->generation,
3307                                                &token);
3308         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3309                 skip_csum = true;
3310                 btrfs_set_token_file_extent_type(leaf, fi,
3311                                                  BTRFS_FILE_EXTENT_PREALLOC,
3312                                                  &token);
3313         } else {
3314                 btrfs_set_token_file_extent_type(leaf, fi,
3315                                                  BTRFS_FILE_EXTENT_REG,
3316                                                  &token);
3317                 if (em->block_start == EXTENT_MAP_HOLE)
3318                         skip_csum = true;
3319         }
3320
3321         block_len = max(em->block_len, em->orig_block_len);
3322         if (em->compress_type != BTRFS_COMPRESS_NONE) {
3323                 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3324                                                         em->block_start,
3325                                                         &token);
3326                 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3327                                                            &token);
3328         } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
3329                 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3330                                                         em->block_start -
3331                                                         extent_offset, &token);
3332                 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3333                                                            &token);
3334         } else {
3335                 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
3336                 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
3337                                                            &token);
3338         }
3339
3340         btrfs_set_token_file_extent_offset(leaf, fi,
3341                                            em->start - em->orig_start,
3342                                            &token);
3343         btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
3344         btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
3345         btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
3346                                                 &token);
3347         btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
3348         btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
3349         btrfs_mark_buffer_dirty(leaf);
3350
3351         btrfs_release_path(path);
3352         if (ret) {
3353                 return ret;
3354         }
3355
3356         if (skip_csum)
3357                 return 0;
3358
3359         if (em->compress_type) {
3360                 csum_offset = 0;
3361                 csum_len = block_len;
3362         }
3363
3364         /*
3365          * First check and see if our csums are on our outstanding ordered
3366          * extents.
3367          */
3368 again:
3369         spin_lock_irq(&log->log_extents_lock[index]);
3370         list_for_each_entry(ordered, &log->logged_list[index], log_list) {
3371                 struct btrfs_ordered_sum *sum;
3372
3373                 if (!mod_len)
3374                         break;
3375
3376                 if (ordered->inode != inode)
3377                         continue;
3378
3379                 if (ordered->file_offset + ordered->len <= mod_start ||
3380                     mod_start + mod_len <= ordered->file_offset)
3381                         continue;
3382
3383                 /*
3384                  * We are going to copy all the csums on this ordered extent, so
3385                  * go ahead and adjust mod_start and mod_len in case this
3386                  * ordered extent has already been logged.
3387                  */
3388                 if (ordered->file_offset > mod_start) {
3389                         if (ordered->file_offset + ordered->len >=
3390                             mod_start + mod_len)
3391                                 mod_len = ordered->file_offset - mod_start;
3392                         /*
3393                          * If we have this case
3394                          *
3395                          * |--------- logged extent ---------|
3396                          *       |----- ordered extent ----|
3397                          *
3398                          * Just don't mess with mod_start and mod_len, we'll
3399                          * just end up logging more csums than we need and it
3400                          * will be ok.
3401                          */
3402                 } else {
3403                         if (ordered->file_offset + ordered->len <
3404                             mod_start + mod_len) {
3405                                 mod_len = (mod_start + mod_len) -
3406                                         (ordered->file_offset + ordered->len);
3407                                 mod_start = ordered->file_offset +
3408                                         ordered->len;
3409                         } else {
3410                                 mod_len = 0;
3411                         }
3412                 }
3413
3414                 /*
3415                  * To keep us from looping for the above case of an ordered
3416                  * extent that falls inside of the logged extent.
3417                  */
3418                 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3419                                      &ordered->flags))
3420                         continue;
3421                 atomic_inc(&ordered->refs);
3422                 spin_unlock_irq(&log->log_extents_lock[index]);
3423                 /*
3424                  * we've dropped the lock, we must either break or
3425                  * start over after this.
3426                  */
3427
3428                 wait_event(ordered->wait, ordered->csum_bytes_left == 0);
3429
3430                 list_for_each_entry(sum, &ordered->list, list) {
3431                         ret = btrfs_csum_file_blocks(trans, log, sum);
3432                         if (ret) {
3433                                 btrfs_put_ordered_extent(ordered);
3434                                 goto unlocked;
3435                         }
3436                 }
3437                 btrfs_put_ordered_extent(ordered);
3438                 goto again;
3439
3440         }
3441         spin_unlock_irq(&log->log_extents_lock[index]);
3442 unlocked:
3443
3444         if (!mod_len || ret)
3445                 return ret;
3446
3447         csum_offset = mod_start - em->start;
3448         csum_len = mod_len;
3449
3450         /* block start is already adjusted for the file extent offset. */
3451         ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
3452                                        em->block_start + csum_offset,
3453                                        em->block_start + csum_offset +
3454                                        csum_len - 1, &ordered_sums, 0);
3455         if (ret)
3456                 return ret;
3457
3458         while (!list_empty(&ordered_sums)) {
3459                 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3460                                                    struct btrfs_ordered_sum,
3461                                                    list);
3462                 if (!ret)
3463                         ret = btrfs_csum_file_blocks(trans, log, sums);
3464                 list_del(&sums->list);
3465                 kfree(sums);
3466         }
3467
3468         return ret;
3469 }
3470
3471 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
3472                                      struct btrfs_root *root,
3473                                      struct inode *inode,
3474                                      struct btrfs_path *path)
3475 {
3476         struct extent_map *em, *n;
3477         struct list_head extents;
3478         struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3479         u64 test_gen;
3480         int ret = 0;
3481         int num = 0;
3482
3483         INIT_LIST_HEAD(&extents);
3484
3485         write_lock(&tree->lock);
3486         test_gen = root->fs_info->last_trans_committed;
3487
3488         list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
3489                 list_del_init(&em->list);
3490
3491                 /*
3492                  * Just an arbitrary number, this can be really CPU intensive
3493                  * once we start getting a lot of extents, and really once we
3494                  * have a bunch of extents we just want to commit since it will
3495                  * be faster.
3496                  */
3497                 if (++num > 32768) {
3498                         list_del_init(&tree->modified_extents);
3499                         ret = -EFBIG;
3500                         goto process;
3501                 }
3502
3503                 if (em->generation <= test_gen)
3504                         continue;
3505                 /* Need a ref to keep it from getting evicted from cache */
3506                 atomic_inc(&em->refs);
3507                 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
3508                 list_add_tail(&em->list, &extents);
3509                 num++;
3510         }
3511
3512         list_sort(NULL, &extents, extent_cmp);
3513
3514 process:
3515         while (!list_empty(&extents)) {
3516                 em = list_entry(extents.next, struct extent_map, list);
3517
3518                 list_del_init(&em->list);
3519
3520                 /*
3521                  * If we had an error we just need to delete everybody from our
3522                  * private list.
3523                  */
3524                 if (ret) {
3525                         clear_em_logging(tree, em);
3526                         free_extent_map(em);
3527                         continue;
3528                 }
3529
3530                 write_unlock(&tree->lock);
3531
3532                 ret = log_one_extent(trans, inode, root, em, path);
3533                 write_lock(&tree->lock);
3534                 clear_em_logging(tree, em);
3535                 free_extent_map(em);
3536         }
3537         WARN_ON(!list_empty(&extents));
3538         write_unlock(&tree->lock);
3539
3540         btrfs_release_path(path);
3541         return ret;
3542 }
3543
3544 /* log a single inode in the tree log.
3545  * At least one parent directory for this inode must exist in the tree
3546  * or be logged already.
3547  *
3548  * Any items from this inode changed by the current transaction are copied
3549  * to the log tree.  An extra reference is taken on any extents in this
3550  * file, allowing us to avoid a whole pile of corner cases around logging
3551  * blocks that have been removed from the tree.
3552  *
3553  * See LOG_INODE_ALL and related defines for a description of what inode_only
3554  * does.
3555  *
3556  * This handles both files and directories.
3557  */
3558 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
3559                              struct btrfs_root *root, struct inode *inode,
3560                              int inode_only)
3561 {
3562         struct btrfs_path *path;
3563         struct btrfs_path *dst_path;
3564         struct btrfs_key min_key;
3565         struct btrfs_key max_key;
3566         struct btrfs_root *log = root->log_root;
3567         struct extent_buffer *src = NULL;
3568         int err = 0;
3569         int ret;
3570         int nritems;
3571         int ins_start_slot = 0;
3572         int ins_nr;
3573         bool fast_search = false;
3574         u64 ino = btrfs_ino(inode);
3575
3576         path = btrfs_alloc_path();
3577         if (!path)
3578                 return -ENOMEM;
3579         dst_path = btrfs_alloc_path();
3580         if (!dst_path) {
3581                 btrfs_free_path(path);
3582                 return -ENOMEM;
3583         }
3584
3585         min_key.objectid = ino;
3586         min_key.type = BTRFS_INODE_ITEM_KEY;
3587         min_key.offset = 0;
3588
3589         max_key.objectid = ino;
3590
3591
3592         /* today the code can only do partial logging of directories */
3593         if (S_ISDIR(inode->i_mode) ||
3594             (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3595                        &BTRFS_I(inode)->runtime_flags) &&
3596              inode_only == LOG_INODE_EXISTS))
3597                 max_key.type = BTRFS_XATTR_ITEM_KEY;
3598         else
3599                 max_key.type = (u8)-1;
3600         max_key.offset = (u64)-1;
3601
3602         /* Only run delayed items if we are a dir or a new file */
3603         if (S_ISDIR(inode->i_mode) ||
3604             BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) {
3605                 ret = btrfs_commit_inode_delayed_items(trans, inode);
3606                 if (ret) {
3607                         btrfs_free_path(path);
3608                         btrfs_free_path(dst_path);
3609                         return ret;
3610                 }
3611         }
3612
3613         mutex_lock(&BTRFS_I(inode)->log_mutex);
3614
3615         btrfs_get_logged_extents(log, inode);
3616
3617         /*
3618          * a brute force approach to making sure we get the most uptodate
3619          * copies of everything.
3620          */
3621         if (S_ISDIR(inode->i_mode)) {
3622                 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
3623
3624                 if (inode_only == LOG_INODE_EXISTS)
3625                         max_key_type = BTRFS_XATTR_ITEM_KEY;
3626                 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
3627         } else {
3628                 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3629                                        &BTRFS_I(inode)->runtime_flags)) {
3630                         clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3631                                   &BTRFS_I(inode)->runtime_flags);
3632                         ret = btrfs_truncate_inode_items(trans, log,
3633                                                          inode, 0, 0);
3634                 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3635                                               &BTRFS_I(inode)->runtime_flags)) {
3636                         if (inode_only == LOG_INODE_ALL)
3637                                 fast_search = true;
3638                         max_key.type = BTRFS_XATTR_ITEM_KEY;
3639                         ret = drop_objectid_items(trans, log, path, ino,
3640                                                   max_key.type);
3641                 } else {
3642                         if (inode_only == LOG_INODE_ALL)
3643                                 fast_search = true;
3644                         ret = log_inode_item(trans, log, dst_path, inode);
3645                         if (ret) {
3646                                 err = ret;
3647                                 goto out_unlock;
3648                         }
3649                         goto log_extents;
3650                 }
3651
3652         }
3653         if (ret) {
3654                 err = ret;
3655                 goto out_unlock;
3656         }
3657         path->keep_locks = 1;
3658
3659         while (1) {
3660                 ins_nr = 0;
3661                 ret = btrfs_search_forward(root, &min_key, &max_key,
3662                                            path, trans->transid);
3663                 if (ret != 0)
3664                         break;
3665 again:
3666                 /* note, ins_nr might be > 0 here, cleanup outside the loop */
3667                 if (min_key.objectid != ino)
3668                         break;
3669                 if (min_key.type > max_key.type)
3670                         break;
3671
3672                 src = path->nodes[0];
3673                 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
3674                         ins_nr++;
3675                         goto next_slot;
3676                 } else if (!ins_nr) {
3677                         ins_start_slot = path->slots[0];
3678                         ins_nr = 1;
3679                         goto next_slot;
3680                 }
3681
3682                 ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
3683                                  ins_nr, inode_only);
3684                 if (ret) {
3685                         err = ret;
3686                         goto out_unlock;
3687                 }
3688                 ins_nr = 1;
3689                 ins_start_slot = path->slots[0];
3690 next_slot:
3691
3692                 nritems = btrfs_header_nritems(path->nodes[0]);
3693                 path->slots[0]++;
3694                 if (path->slots[0] < nritems) {
3695                         btrfs_item_key_to_cpu(path->nodes[0], &min_key,
3696                                               path->slots[0]);
3697                         goto again;
3698                 }
3699                 if (ins_nr) {
3700                         ret = copy_items(trans, inode, dst_path, src,
3701                                          ins_start_slot,
3702                                          ins_nr, inode_only);
3703                         if (ret) {
3704                                 err = ret;
3705                                 goto out_unlock;
3706                         }
3707                         ins_nr = 0;
3708                 }
3709                 btrfs_release_path(path);
3710
3711                 if (min_key.offset < (u64)-1)
3712                         min_key.offset++;
3713                 else if (min_key.type < (u8)-1)
3714                         min_key.type++;
3715                 else if (min_key.objectid < (u64)-1)
3716                         min_key.objectid++;
3717                 else
3718                         break;
3719         }
3720         if (ins_nr) {
3721                 ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
3722                                  ins_nr, inode_only);
3723                 if (ret) {
3724                         err = ret;
3725                         goto out_unlock;
3726                 }
3727                 ins_nr = 0;
3728         }
3729
3730 log_extents:
3731         btrfs_release_path(path);
3732         btrfs_release_path(dst_path);
3733         if (fast_search) {
3734                 ret = btrfs_log_changed_extents(trans, root, inode, dst_path);
3735                 if (ret) {
3736                         err = ret;
3737                         goto out_unlock;
3738                 }
3739         } else {
3740                 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3741                 struct extent_map *em, *n;
3742
3743                 write_lock(&tree->lock);
3744                 list_for_each_entry_safe(em, n, &tree->modified_extents, list)
3745                         list_del_init(&em->list);
3746                 write_unlock(&tree->lock);
3747         }
3748
3749         if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
3750                 ret = log_directory_changes(trans, root, inode, path, dst_path);
3751                 if (ret) {
3752                         err = ret;
3753                         goto out_unlock;
3754                 }
3755         }
3756         BTRFS_I(inode)->logged_trans = trans->transid;
3757         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
3758 out_unlock:
3759         if (err)
3760                 btrfs_free_logged_extents(log, log->log_transid);
3761         mutex_unlock(&BTRFS_I(inode)->log_mutex);
3762
3763         btrfs_free_path(path);
3764         btrfs_free_path(dst_path);
3765         return err;
3766 }
3767
3768 /*
3769  * follow the dentry parent pointers up the chain and see if any
3770  * of the directories in it require a full commit before they can
3771  * be logged.  Returns zero if nothing special needs to be done or 1 if
3772  * a full commit is required.
3773  */
3774 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
3775                                                struct inode *inode,
3776                                                struct dentry *parent,
3777                                                struct super_block *sb,
3778                                                u64 last_committed)
3779 {
3780         int ret = 0;
3781         struct btrfs_root *root;
3782         struct dentry *old_parent = NULL;
3783
3784         /*
3785          * for regular files, if its inode is already on disk, we don't
3786          * have to worry about the parents at all.  This is because
3787          * we can use the last_unlink_trans field to record renames
3788          * and other fun in this file.
3789          */
3790         if (S_ISREG(inode->i_mode) &&
3791             BTRFS_I(inode)->generation <= last_committed &&
3792             BTRFS_I(inode)->last_unlink_trans <= last_committed)
3793                         goto out;
3794
3795         if (!S_ISDIR(inode->i_mode)) {
3796                 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3797                         goto out;
3798                 inode = parent->d_inode;
3799         }
3800
3801         while (1) {
3802                 BTRFS_I(inode)->logged_trans = trans->transid;
3803                 smp_mb();
3804
3805                 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
3806                         root = BTRFS_I(inode)->root;
3807
3808                         /*
3809                          * make sure any commits to the log are forced
3810                          * to be full commits
3811                          */
3812                         root->fs_info->last_trans_log_full_commit =
3813                                 trans->transid;
3814                         ret = 1;
3815                         break;
3816                 }
3817
3818                 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3819                         break;
3820
3821                 if (IS_ROOT(parent))
3822                         break;
3823
3824                 parent = dget_parent(parent);
3825                 dput(old_parent);
3826                 old_parent = parent;
3827                 inode = parent->d_inode;
3828
3829         }
3830         dput(old_parent);
3831 out:
3832         return ret;
3833 }
3834
3835 /*
3836  * helper function around btrfs_log_inode to make sure newly created
3837  * parent directories also end up in the log.  A minimal inode and backref
3838  * only logging is done of any parent directories that are older than
3839  * the last committed transaction
3840  */
3841 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3842                                   struct btrfs_root *root, struct inode *inode,
3843                                   struct dentry *parent, int exists_only)
3844 {
3845         int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
3846         struct super_block *sb;
3847         struct dentry *old_parent = NULL;
3848         int ret = 0;
3849         u64 last_committed = root->fs_info->last_trans_committed;
3850
3851         sb = inode->i_sb;
3852
3853         if (btrfs_test_opt(root, NOTREELOG)) {
3854                 ret = 1;
3855                 goto end_no_trans;
3856         }
3857
3858         if (root->fs_info->last_trans_log_full_commit >
3859             root->fs_info->last_trans_committed) {
3860                 ret = 1;
3861                 goto end_no_trans;
3862         }
3863
3864         if (root != BTRFS_I(inode)->root ||
3865             btrfs_root_refs(&root->root_item) == 0) {
3866                 ret = 1;
3867                 goto end_no_trans;
3868         }
3869
3870         ret = check_parent_dirs_for_sync(trans, inode, parent,
3871                                          sb, last_committed);
3872         if (ret)
3873                 goto end_no_trans;
3874
3875         if (btrfs_inode_in_log(inode, trans->transid)) {
3876                 ret = BTRFS_NO_LOG_SYNC;
3877                 goto end_no_trans;
3878         }
3879
3880         ret = start_log_trans(trans, root);
3881         if (ret)
3882                 goto end_trans;
3883
3884         ret = btrfs_log_inode(trans, root, inode, inode_only);
3885         if (ret)
3886                 goto end_trans;
3887
3888         /*
3889          * for regular files, if its inode is already on disk, we don't
3890          * have to worry about the parents at all.  This is because
3891          * we can use the last_unlink_trans field to record renames
3892          * and other fun in this file.
3893          */
3894         if (S_ISREG(inode->i_mode) &&
3895             BTRFS_I(inode)->generation <= last_committed &&
3896             BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3897                 ret = 0;
3898                 goto end_trans;
3899         }
3900
3901         inode_only = LOG_INODE_EXISTS;
3902         while (1) {
3903                 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3904                         break;
3905
3906                 inode = parent->d_inode;
3907                 if (root != BTRFS_I(inode)->root)
3908                         break;
3909
3910                 if (BTRFS_I(inode)->generation >
3911                     root->fs_info->last_trans_committed) {
3912                         ret = btrfs_log_inode(trans, root, inode, inode_only);
3913                         if (ret)
3914                                 goto end_trans;
3915                 }
3916                 if (IS_ROOT(parent))
3917                         break;
3918
3919                 parent = dget_parent(parent);
3920                 dput(old_parent);
3921                 old_parent = parent;
3922         }
3923         ret = 0;
3924 end_trans:
3925         dput(old_parent);
3926         if (ret < 0) {
3927                 root->fs_info->last_trans_log_full_commit = trans->transid;
3928                 ret = 1;
3929         }
3930         btrfs_end_log_trans(root);
3931 end_no_trans:
3932         return ret;
3933 }
3934
3935 /*
3936  * it is not safe to log dentry if the chunk root has added new
3937  * chunks.  This returns 0 if the dentry was logged, and 1 otherwise.
3938  * If this returns 1, you must commit the transaction to safely get your
3939  * data on disk.
3940  */
3941 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3942                           struct btrfs_root *root, struct dentry *dentry)
3943 {
3944         struct dentry *parent = dget_parent(dentry);
3945         int ret;
3946
3947         ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3948         dput(parent);
3949
3950         return ret;
3951 }
3952
3953 /*
3954  * should be called during mount to recover any replay any log trees
3955  * from the FS
3956  */
3957 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3958 {
3959         int ret;
3960         struct btrfs_path *path;
3961         struct btrfs_trans_handle *trans;
3962         struct btrfs_key key;
3963         struct btrfs_key found_key;
3964         struct btrfs_key tmp_key;
3965         struct btrfs_root *log;
3966         struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3967         struct walk_control wc = {
3968                 .process_func = process_one_buffer,
3969                 .stage = 0,
3970         };
3971
3972         path = btrfs_alloc_path();
3973         if (!path)
3974                 return -ENOMEM;
3975
3976         fs_info->log_root_recovering = 1;
3977
3978         trans = btrfs_start_transaction(fs_info->tree_root, 0);
3979         if (IS_ERR(trans)) {
3980                 ret = PTR_ERR(trans);
3981                 goto error;
3982         }
3983
3984         wc.trans = trans;
3985         wc.pin = 1;
3986
3987         ret = walk_log_tree(trans, log_root_tree, &wc);
3988         if (ret) {
3989                 btrfs_error(fs_info, ret, "Failed to pin buffers while "
3990                             "recovering log root tree.");
3991                 goto error;
3992         }
3993
3994 again:
3995         key.objectid = BTRFS_TREE_LOG_OBJECTID;
3996         key.offset = (u64)-1;
3997         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3998
3999         while (1) {
4000                 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
4001
4002                 if (ret < 0) {
4003                         btrfs_error(fs_info, ret,
4004                                     "Couldn't find tree log root.");
4005                         goto error;
4006                 }
4007                 if (ret > 0) {
4008                         if (path->slots[0] == 0)
4009                                 break;
4010                         path->slots[0]--;
4011                 }
4012                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4013                                       path->slots[0]);
4014                 btrfs_release_path(path);
4015                 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4016                         break;
4017
4018                 log = btrfs_read_fs_root_no_radix(log_root_tree,
4019                                                   &found_key);
4020                 if (IS_ERR(log)) {
4021                         ret = PTR_ERR(log);
4022                         btrfs_error(fs_info, ret,
4023                                     "Couldn't read tree log root.");
4024                         goto error;
4025                 }
4026
4027                 tmp_key.objectid = found_key.offset;
4028                 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
4029                 tmp_key.offset = (u64)-1;
4030
4031                 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
4032                 if (IS_ERR(wc.replay_dest)) {
4033                         ret = PTR_ERR(wc.replay_dest);
4034                         free_extent_buffer(log->node);
4035                         free_extent_buffer(log->commit_root);
4036                         kfree(log);
4037                         btrfs_error(fs_info, ret, "Couldn't read target root "
4038                                     "for tree log recovery.");
4039                         goto error;
4040                 }
4041
4042                 wc.replay_dest->log_root = log;
4043                 btrfs_record_root_in_trans(trans, wc.replay_dest);
4044                 ret = walk_log_tree(trans, log, &wc);
4045
4046                 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
4047                         ret = fixup_inode_link_counts(trans, wc.replay_dest,
4048                                                       path);
4049                 }
4050
4051                 key.offset = found_key.offset - 1;
4052                 wc.replay_dest->log_root = NULL;
4053                 free_extent_buffer(log->node);
4054                 free_extent_buffer(log->commit_root);
4055                 kfree(log);
4056
4057                 if (ret)
4058                         goto error;
4059
4060                 if (found_key.offset == 0)
4061                         break;
4062         }
4063         btrfs_release_path(path);
4064
4065         /* step one is to pin it all, step two is to replay just inodes */
4066         if (wc.pin) {
4067                 wc.pin = 0;
4068                 wc.process_func = replay_one_buffer;
4069                 wc.stage = LOG_WALK_REPLAY_INODES;
4070                 goto again;
4071         }
4072         /* step three is to replay everything */
4073         if (wc.stage < LOG_WALK_REPLAY_ALL) {
4074                 wc.stage++;
4075                 goto again;
4076         }
4077
4078         btrfs_free_path(path);
4079
4080         /* step 4: commit the transaction, which also unpins the blocks */
4081         ret = btrfs_commit_transaction(trans, fs_info->tree_root);
4082         if (ret)
4083                 return ret;
4084
4085         free_extent_buffer(log_root_tree->node);
4086         log_root_tree->log_root = NULL;
4087         fs_info->log_root_recovering = 0;
4088         kfree(log_root_tree);
4089
4090         return 0;
4091 error:
4092         if (wc.trans)
4093                 btrfs_end_transaction(wc.trans, fs_info->tree_root);
4094         btrfs_free_path(path);
4095         return ret;
4096 }
4097
4098 /*
4099  * there are some corner cases where we want to force a full
4100  * commit instead of allowing a directory to be logged.
4101  *
4102  * They revolve around files there were unlinked from the directory, and
4103  * this function updates the parent directory so that a full commit is
4104  * properly done if it is fsync'd later after the unlinks are done.
4105  */
4106 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
4107                              struct inode *dir, struct inode *inode,
4108                              int for_rename)
4109 {
4110         /*
4111          * when we're logging a file, if it hasn't been renamed
4112          * or unlinked, and its inode is fully committed on disk,
4113          * we don't have to worry about walking up the directory chain
4114          * to log its parents.
4115          *
4116          * So, we use the last_unlink_trans field to put this transid
4117          * into the file.  When the file is logged we check it and
4118          * don't log the parents if the file is fully on disk.
4119          */
4120         if (S_ISREG(inode->i_mode))
4121                 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4122
4123         /*
4124          * if this directory was already logged any new
4125          * names for this file/dir will get recorded
4126          */
4127         smp_mb();
4128         if (BTRFS_I(dir)->logged_trans == trans->transid)
4129                 return;
4130
4131         /*
4132          * if the inode we're about to unlink was logged,
4133          * the log will be properly updated for any new names
4134          */
4135         if (BTRFS_I(inode)->logged_trans == trans->transid)
4136                 return;
4137
4138         /*
4139          * when renaming files across directories, if the directory
4140          * there we're unlinking from gets fsync'd later on, there's
4141          * no way to find the destination directory later and fsync it
4142          * properly.  So, we have to be conservative and force commits
4143          * so the new name gets discovered.
4144          */
4145         if (for_rename)
4146                 goto record;
4147
4148         /* we can safely do the unlink without any special recording */
4149         return;
4150
4151 record:
4152         BTRFS_I(dir)->last_unlink_trans = trans->transid;
4153 }
4154
4155 /*
4156  * Call this after adding a new name for a file and it will properly
4157  * update the log to reflect the new name.
4158  *
4159  * It will return zero if all goes well, and it will return 1 if a
4160  * full transaction commit is required.
4161  */
4162 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
4163                         struct inode *inode, struct inode *old_dir,
4164                         struct dentry *parent)
4165 {
4166         struct btrfs_root * root = BTRFS_I(inode)->root;
4167
4168         /*
4169          * this will force the logging code to walk the dentry chain
4170          * up for the file
4171          */
4172         if (S_ISREG(inode->i_mode))
4173                 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4174
4175         /*
4176          * if this inode hasn't been logged and directory we're renaming it
4177          * from hasn't been logged, we don't need to log it
4178          */
4179         if (BTRFS_I(inode)->logged_trans <=
4180             root->fs_info->last_trans_committed &&
4181             (!old_dir || BTRFS_I(old_dir)->logged_trans <=
4182                     root->fs_info->last_trans_committed))
4183                 return 0;
4184
4185         return btrfs_log_inode_parent(trans, root, inode, parent, 1);
4186 }
4187