fs/btrfs/free-space-tree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2015 Facebook.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/sched/mm.h>
   8 #include "ctree.h"
   9 #include "disk-io.h"
  10 #include "locking.h"
  11 #include "free-space-tree.h"
  12 #include "transaction.h"
  13
  14 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
  15                                         struct btrfs_fs_info *fs_info,
  16                                         struct btrfs_block_group_cache *block_group,
  17                                         struct btrfs_path *path);
  18
  19 void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache)
  20 {
  21         u32 bitmap_range;
  22         size_t bitmap_size;
  23         u64 num_bitmaps, total_bitmap_size;
  24
  25         /*
  26          * We convert to bitmaps when the disk space required for using extents
  27          * exceeds that required for using bitmaps.
  28          */
  29         bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
  30         num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1,
  31                               bitmap_range);
  32         bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
  33         total_bitmap_size = num_bitmaps * bitmap_size;
  34         cache->bitmap_high_thresh = div_u64(total_bitmap_size,
  35                                             sizeof(struct btrfs_item));
  36
  37         /*
  38          * We allow for a small buffer between the high threshold and low
  39          * threshold to avoid thrashing back and forth between the two formats.
  40          */
  41         if (cache->bitmap_high_thresh > 100)
  42                 cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
  43         else
  44                 cache->bitmap_low_thresh = 0;
  45 }
  46
  47 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
  48                                    struct btrfs_fs_info *fs_info,
  49                                    struct btrfs_block_group_cache *block_group,
  50                                    struct btrfs_path *path)
  51 {
  52         struct btrfs_root *root = fs_info->free_space_root;
  53         struct btrfs_free_space_info *info;
  54         struct btrfs_key key;
  55         struct extent_buffer *leaf;
  56         int ret;
  57
  58         key.objectid = block_group->key.objectid;
  59         key.type = BTRFS_FREE_SPACE_INFO_KEY;
  60         key.offset = block_group->key.offset;
  61
  62         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
  63         if (ret)
  64                 goto out;
  65
  66         leaf = path->nodes[0];
  67         info = btrfs_item_ptr(leaf, path->slots[0],
  68                               struct btrfs_free_space_info);
  69         btrfs_set_free_space_extent_count(leaf, info, 0);
  70         btrfs_set_free_space_flags(leaf, info, 0);
  71         btrfs_mark_buffer_dirty(leaf);
  72
  73         ret = 0;
  74 out:
  75         btrfs_release_path(path);
  76         return ret;
  77 }
  78
  79 struct btrfs_free_space_info *
  80 search_free_space_info(struct btrfs_trans_handle *trans,
  81                        struct btrfs_fs_info *fs_info,
  82                        struct btrfs_block_group_cache *block_group,
  83                        struct btrfs_path *path, int cow)
  84 {
  85         struct btrfs_root *root = fs_info->free_space_root;
  86         struct btrfs_key key;
  87         int ret;
  88
  89         key.objectid = block_group->key.objectid;
  90         key.type = BTRFS_FREE_SPACE_INFO_KEY;
  91         key.offset = block_group->key.offset;
  92
  93         ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
  94         if (ret < 0)
  95                 return ERR_PTR(ret);
  96         if (ret != 0) {
  97                 btrfs_warn(fs_info, "missing free space info for %llu",
  98                            block_group->key.objectid);
  99                 ASSERT(0);
 100                 return ERR_PTR(-ENOENT);
 101         }
 102
 103         return btrfs_item_ptr(path->nodes[0], path->slots[0],
 104                               struct btrfs_free_space_info);
 105 }
 106
 107 /*
 108  * btrfs_search_slot() but we're looking for the greatest key less than the
 109  * passed key.
 110  */
 111 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
 112                                   struct btrfs_root *root,
 113                                   struct btrfs_key *key, struct btrfs_path *p,
 114                                   int ins_len, int cow)
 115 {
 116         int ret;
 117
 118         ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
 119         if (ret < 0)
 120                 return ret;
 121
 122         if (ret == 0) {
 123                 ASSERT(0);
 124                 return -EIO;
 125         }
 126
 127         if (p->slots[0] == 0) {
 128                 ASSERT(0);
 129                 return -EIO;
 130         }
 131         p->slots[0]--;
 132
 133         return 0;
 134 }
 135
 136 static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
 137 {
 138         return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
 139 }
 140
 141 static u8 *alloc_bitmap(u32 bitmap_size)
 142 {
 143         u8 *ret;
 144         unsigned int nofs_flag;
 145
 146         /*
 147          * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
 148          * into the filesystem as the free space bitmap can be modified in the
 149          * critical section of a transaction commit.
 150          *
 151          * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
 152          * know that recursion is unsafe.
 153          */
 154         nofs_flag = memalloc_nofs_save();
 155         ret = kvzalloc(bitmap_size, GFP_KERNEL);
 156         memalloc_nofs_restore(nofs_flag);
 157         return ret;
 158 }
 159
 160 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
 161                                   struct btrfs_fs_info *fs_info,
 162                                   struct btrfs_block_group_cache *block_group,
 163                                   struct btrfs_path *path)
 164 {
 165         struct btrfs_root *root = fs_info->free_space_root;
 166         struct btrfs_free_space_info *info;
 167         struct btrfs_key key, found_key;
 168         struct extent_buffer *leaf;
 169         u8 *bitmap, *bitmap_cursor;
 170         u64 start, end;
 171         u64 bitmap_range, i;
 172         u32 bitmap_size, flags, expected_extent_count;
 173         u32 extent_count = 0;
 174         int done = 0, nr;
 175         int ret;
 176
 177         bitmap_size = free_space_bitmap_size(block_group->key.offset,
 178                                              fs_info->sectorsize);
 179         bitmap = alloc_bitmap(bitmap_size);
 180         if (!bitmap) {
 181                 ret = -ENOMEM;
 182                 goto out;
 183         }
 184
 185         start = block_group->key.objectid;
 186         end = block_group->key.objectid + block_group->key.offset;
 187
 188         key.objectid = end - 1;
 189         key.type = (u8)-1;
 190         key.offset = (u64)-1;
 191
 192         while (!done) {
 193                 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
 194                 if (ret)
 195                         goto out;
 196
 197                 leaf = path->nodes[0];
 198                 nr = 0;
 199                 path->slots[0]++;
 200                 while (path->slots[0] > 0) {
 201                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
 202
 203                         if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
 204                                 ASSERT(found_key.objectid == block_group->key.objectid);
 205                                 ASSERT(found_key.offset == block_group->key.offset);
 206                                 done = 1;
 207                                 break;
 208                         } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
 209                                 u64 first, last;
 210
 211                                 ASSERT(found_key.objectid >= start);
 212                                 ASSERT(found_key.objectid < end);
 213                                 ASSERT(found_key.objectid + found_key.offset <= end);
 214
 215                                 first = div_u64(found_key.objectid - start,
 216                                                 fs_info->sectorsize);
 217                                 last = div_u64(found_key.objectid + found_key.offset - start,
 218                                                fs_info->sectorsize);
 219                                 le_bitmap_set(bitmap, first, last - first);
 220
 221                                 extent_count++;
 222                                 nr++;
 223                                 path->slots[0]--;
 224                         } else {
 225                                 ASSERT(0);
 226                         }
 227                 }
 228
 229                 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
 230                 if (ret)
 231                         goto out;
 232                 btrfs_release_path(path);
 233         }
 234
 235         info = search_free_space_info(trans, fs_info, block_group, path, 1);
 236         if (IS_ERR(info)) {
 237                 ret = PTR_ERR(info);
 238                 goto out;
 239         }
 240         leaf = path->nodes[0];
 241         flags = btrfs_free_space_flags(leaf, info);
 242         flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
 243         btrfs_set_free_space_flags(leaf, info, flags);
 244         expected_extent_count = btrfs_free_space_extent_count(leaf, info);
 245         btrfs_mark_buffer_dirty(leaf);
 246         btrfs_release_path(path);
 247
 248         if (extent_count != expected_extent_count) {
 249                 btrfs_err(fs_info,
 250                           "incorrect extent count for %llu; counted %u, expected %u",
 251                           block_group->key.objectid, extent_count,
 252                           expected_extent_count);
 253                 ASSERT(0);
 254                 ret = -EIO;
 255                 goto out;
 256         }
 257
 258         bitmap_cursor = bitmap;
 259         bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
 260         i = start;
 261         while (i < end) {
 262                 unsigned long ptr;
 263                 u64 extent_size;
 264                 u32 data_size;
 265
 266                 extent_size = min(end - i, bitmap_range);
 267                 data_size = free_space_bitmap_size(extent_size,
 268                                                    fs_info->sectorsize);
 269
 270                 key.objectid = i;
 271                 key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
 272                 key.offset = extent_size;
 273
 274                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 275                                               data_size);
 276                 if (ret)
 277                         goto out;
 278
 279                 leaf = path->nodes[0];
 280                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
 281                 write_extent_buffer(leaf, bitmap_cursor, ptr,
 282                                     data_size);
 283                 btrfs_mark_buffer_dirty(leaf);
 284                 btrfs_release_path(path);
 285
 286                 i += extent_size;
 287                 bitmap_cursor += data_size;
 288         }
 289
 290         ret = 0;
 291 out:
 292         kvfree(bitmap);
 293         if (ret)
 294                 btrfs_abort_transaction(trans, ret);
 295         return ret;
 296 }
 297
 298 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
 299                                   struct btrfs_fs_info *fs_info,
 300                                   struct btrfs_block_group_cache *block_group,
 301                                   struct btrfs_path *path)
 302 {
 303         struct btrfs_root *root = fs_info->free_space_root;
 304         struct btrfs_free_space_info *info;
 305         struct btrfs_key key, found_key;
 306         struct extent_buffer *leaf;
 307         u8 *bitmap;
 308         u64 start, end;
 309         /* Initialize to silence GCC. */
 310         u64 extent_start = 0;
 311         u64 offset;
 312         u32 bitmap_size, flags, expected_extent_count;
 313         int prev_bit = 0, bit, bitnr;
 314         u32 extent_count = 0;
 315         int done = 0, nr;
 316         int ret;
 317
 318         bitmap_size = free_space_bitmap_size(block_group->key.offset,
 319                                              fs_info->sectorsize);
 320         bitmap = alloc_bitmap(bitmap_size);
 321         if (!bitmap) {
 322                 ret = -ENOMEM;
 323                 goto out;
 324         }
 325
 326         start = block_group->key.objectid;
 327         end = block_group->key.objectid + block_group->key.offset;
 328
 329         key.objectid = end - 1;
 330         key.type = (u8)-1;
 331         key.offset = (u64)-1;
 332
 333         while (!done) {
 334                 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
 335                 if (ret)
 336                         goto out;
 337
 338                 leaf = path->nodes[0];
 339                 nr = 0;
 340                 path->slots[0]++;
 341                 while (path->slots[0] > 0) {
 342                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
 343
 344                         if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
 345                                 ASSERT(found_key.objectid == block_group->key.objectid);
 346                                 ASSERT(found_key.offset == block_group->key.offset);
 347                                 done = 1;
 348                                 break;
 349                         } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
 350                                 unsigned long ptr;
 351                                 u8 *bitmap_cursor;
 352                                 u32 bitmap_pos, data_size;
 353
 354                                 ASSERT(found_key.objectid >= start);
 355                                 ASSERT(found_key.objectid < end);
 356                                 ASSERT(found_key.objectid + found_key.offset <= end);
 357
 358                                 bitmap_pos = div_u64(found_key.objectid - start,
 359                                                      fs_info->sectorsize *
 360                                                      BITS_PER_BYTE);
 361                                 bitmap_cursor = bitmap + bitmap_pos;
 362                                 data_size = free_space_bitmap_size(found_key.offset,
 363                                                                    fs_info->sectorsize);
 364
 365                                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
 366                                 read_extent_buffer(leaf, bitmap_cursor, ptr,
 367                                                    data_size);
 368
 369                                 nr++;
 370                                 path->slots[0]--;
 371                         } else {
 372                                 ASSERT(0);
 373                         }
 374                 }
 375
 376                 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
 377                 if (ret)
 378                         goto out;
 379                 btrfs_release_path(path);
 380         }
 381
 382         info = search_free_space_info(trans, fs_info, block_group, path, 1);
 383         if (IS_ERR(info)) {
 384                 ret = PTR_ERR(info);
 385                 goto out;
 386         }
 387         leaf = path->nodes[0];
 388         flags = btrfs_free_space_flags(leaf, info);
 389         flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
 390         btrfs_set_free_space_flags(leaf, info, flags);
 391         expected_extent_count = btrfs_free_space_extent_count(leaf, info);
 392         btrfs_mark_buffer_dirty(leaf);
 393         btrfs_release_path(path);
 394
 395         offset = start;
 396         bitnr = 0;
 397         while (offset < end) {
 398                 bit = !!le_test_bit(bitnr, bitmap);
 399                 if (prev_bit == 0 && bit == 1) {
 400                         extent_start = offset;
 401                 } else if (prev_bit == 1 && bit == 0) {
 402                         key.objectid = extent_start;
 403                         key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
 404                         key.offset = offset - extent_start;
 405
 406                         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 407                         if (ret)
 408                                 goto out;
 409                         btrfs_release_path(path);
 410
 411                         extent_count++;
 412                 }
 413                 prev_bit = bit;
 414                 offset += fs_info->sectorsize;
 415                 bitnr++;
 416         }
 417         if (prev_bit == 1) {
 418                 key.objectid = extent_start;
 419                 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
 420                 key.offset = end - extent_start;
 421
 422                 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 423                 if (ret)
 424                         goto out;
 425                 btrfs_release_path(path);
 426
 427                 extent_count++;
 428         }
 429
 430         if (extent_count != expected_extent_count) {
 431                 btrfs_err(fs_info,
 432                           "incorrect extent count for %llu; counted %u, expected %u",
 433                           block_group->key.objectid, extent_count,
 434                           expected_extent_count);
 435                 ASSERT(0);
 436                 ret = -EIO;
 437                 goto out;
 438         }
 439
 440         ret = 0;
 441 out:
 442         kvfree(bitmap);
 443         if (ret)
 444                 btrfs_abort_transaction(trans, ret);
 445         return ret;
 446 }
 447
 448 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
 449                                           struct btrfs_fs_info *fs_info,
 450                                           struct btrfs_block_group_cache *block_group,
 451                                           struct btrfs_path *path,
 452                                           int new_extents)
 453 {
 454         struct btrfs_free_space_info *info;
 455         u32 flags;
 456         u32 extent_count;
 457         int ret = 0;
 458
 459         if (new_extents == 0)
 460                 return 0;
 461
 462         info = search_free_space_info(trans, fs_info, block_group, path, 1);
 463         if (IS_ERR(info)) {
 464                 ret = PTR_ERR(info);
 465                 goto out;
 466         }
 467         flags = btrfs_free_space_flags(path->nodes[0], info);
 468         extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
 469
 470         extent_count += new_extents;
 471         btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
 472         btrfs_mark_buffer_dirty(path->nodes[0]);
 473         btrfs_release_path(path);
 474
 475         if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
 476             extent_count > block_group->bitmap_high_thresh) {
 477                 ret = convert_free_space_to_bitmaps(trans, fs_info, block_group,
 478                                                     path);
 479         } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
 480                    extent_count < block_group->bitmap_low_thresh) {
 481                 ret = convert_free_space_to_extents(trans, fs_info, block_group,
 482                                                     path);
 483         }
 484
 485 out:
 486         return ret;
 487 }
 488
 489 int free_space_test_bit(struct btrfs_block_group_cache *block_group,
 490                         struct btrfs_path *path, u64 offset)
 491 {
 492         struct extent_buffer *leaf;
 493         struct btrfs_key key;
 494         u64 found_start, found_end;
 495         unsigned long ptr, i;
 496
 497         leaf = path->nodes[0];
 498         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 499         ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
 500
 501         found_start = key.objectid;
 502         found_end = key.objectid + key.offset;
 503         ASSERT(offset >= found_start && offset < found_end);
 504
 505         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
 506         i = div_u64(offset - found_start,
 507                     block_group->fs_info->sectorsize);
 508         return !!extent_buffer_test_bit(leaf, ptr, i);
 509 }
 510
 511 static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
 512                                 struct btrfs_path *path, u64 *start, u64 *size,
 513                                 int bit)
 514 {
 515         struct btrfs_fs_info *fs_info = block_group->fs_info;
 516         struct extent_buffer *leaf;
 517         struct btrfs_key key;
 518         u64 end = *start + *size;
 519         u64 found_start, found_end;
 520         unsigned long ptr, first, last;
 521
 522         leaf = path->nodes[0];
 523         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 524         ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
 525
 526         found_start = key.objectid;
 527         found_end = key.objectid + key.offset;
 528         ASSERT(*start >= found_start && *start < found_end);
 529         ASSERT(end > found_start);
 530
 531         if (end > found_end)
 532                 end = found_end;
 533
 534         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
 535         first = div_u64(*start - found_start, fs_info->sectorsize);
 536         last = div_u64(end - found_start, fs_info->sectorsize);
 537         if (bit)
 538                 extent_buffer_bitmap_set(leaf, ptr, first, last - first);
 539         else
 540                 extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
 541         btrfs_mark_buffer_dirty(leaf);
 542
 543         *size -= end - *start;
 544         *start = end;
 545 }
 546
 547 /*
 548  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
 549  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
 550  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
 551  * looking for.
 552  */
 553 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
 554                                   struct btrfs_root *root, struct btrfs_path *p)
 555 {
 556         struct btrfs_key key;
 557
 558         if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
 559                 p->slots[0]++;
 560                 return 0;
 561         }
 562
 563         btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
 564         btrfs_release_path(p);
 565
 566         key.objectid += key.offset;
 567         key.type = (u8)-1;
 568         key.offset = (u64)-1;
 569
 570         return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
 571 }
 572
 573 /*
 574  * If remove is 1, then we are removing free space, thus clearing bits in the
 575  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
 576  * the bitmap.
 577  */
 578 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
 579                                     struct btrfs_fs_info *fs_info,
 580                                     struct btrfs_block_group_cache *block_group,
 581                                     struct btrfs_path *path,
 582                                     u64 start, u64 size, int remove)
 583 {
 584         struct btrfs_root *root = fs_info->free_space_root;
 585         struct btrfs_key key;
 586         u64 end = start + size;
 587         u64 cur_start, cur_size;
 588         int prev_bit, next_bit;
 589         int new_extents;
 590         int ret;
 591
 592         /*
 593          * Read the bit for the block immediately before the extent of space if
 594          * that block is within the block group.
 595          */
 596         if (start > block_group->key.objectid) {
 597                 u64 prev_block = start - block_group->fs_info->sectorsize;
 598
 599                 key.objectid = prev_block;
 600                 key.type = (u8)-1;
 601                 key.offset = (u64)-1;
 602
 603                 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
 604                 if (ret)
 605                         goto out;
 606
 607                 prev_bit = free_space_test_bit(block_group, path, prev_block);
 608
 609                 /* The previous block may have been in the previous bitmap. */
 610                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 611                 if (start >= key.objectid + key.offset) {
 612                         ret = free_space_next_bitmap(trans, root, path);
 613                         if (ret)
 614                                 goto out;
 615                 }
 616         } else {
 617                 key.objectid = start;
 618                 key.type = (u8)-1;
 619                 key.offset = (u64)-1;
 620
 621                 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
 622                 if (ret)
 623                         goto out;
 624
 625                 prev_bit = -1;
 626         }
 627
 628         /*
 629          * Iterate over all of the bitmaps overlapped by the extent of space,
 630          * clearing/setting bits as required.
 631          */
 632         cur_start = start;
 633         cur_size = size;
 634         while (1) {
 635                 free_space_set_bits(block_group, path, &cur_start, &cur_size,
 636                                     !remove);
 637                 if (cur_size == 0)
 638                         break;
 639                 ret = free_space_next_bitmap(trans, root, path);
 640                 if (ret)
 641                         goto out;
 642         }
 643
 644         /*
 645          * Read the bit for the block immediately after the extent of space if
 646          * that block is within the block group.
 647          */
 648         if (end < block_group->key.objectid + block_group->key.offset) {
 649                 /* The next block may be in the next bitmap. */
 650                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 651                 if (end >= key.objectid + key.offset) {
 652                         ret = free_space_next_bitmap(trans, root, path);
 653                         if (ret)
 654                                 goto out;
 655                 }
 656
 657                 next_bit = free_space_test_bit(block_group, path, end);
 658         } else {
 659                 next_bit = -1;
 660         }
 661
 662         if (remove) {
 663                 new_extents = -1;
 664                 if (prev_bit == 1) {
 665                         /* Leftover on the left. */
 666                         new_extents++;
 667                 }
 668                 if (next_bit == 1) {
 669                         /* Leftover on the right. */
 670                         new_extents++;
 671                 }
 672         } else {
 673                 new_extents = 1;
 674                 if (prev_bit == 1) {
 675                         /* Merging with neighbor on the left. */
 676                         new_extents--;
 677                 }
 678                 if (next_bit == 1) {
 679                         /* Merging with neighbor on the right. */
 680                         new_extents--;
 681                 }
 682         }
 683
 684         btrfs_release_path(path);
 685         ret = update_free_space_extent_count(trans, fs_info, block_group, path,
 686                                              new_extents);
 687
 688 out:
 689         return ret;
 690 }
 691
 692 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
 693                                     struct btrfs_fs_info *fs_info,
 694                                     struct btrfs_block_group_cache *block_group,
 695                                     struct btrfs_path *path,
 696                                     u64 start, u64 size)
 697 {
 698         struct btrfs_root *root = fs_info->free_space_root;
 699         struct btrfs_key key;
 700         u64 found_start, found_end;
 701         u64 end = start + size;
 702         int new_extents = -1;
 703         int ret;
 704
 705         key.objectid = start;
 706         key.type = (u8)-1;
 707         key.offset = (u64)-1;
 708
 709         ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
 710         if (ret)
 711                 goto out;
 712
 713         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 714
 715         ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
 716
 717         found_start = key.objectid;
 718         found_end = key.objectid + key.offset;
 719         ASSERT(start >= found_start && end <= found_end);
 720
 721         /*
 722          * Okay, now that we've found the free space extent which contains the
 723          * free space that we are removing, there are four cases:
 724          *
 725          * 1. We're using the whole extent: delete the key we found and
 726          * decrement the free space extent count.
 727          * 2. We are using part of the extent starting at the beginning: delete
 728          * the key we found and insert a new key representing the leftover at
 729          * the end. There is no net change in the number of extents.
 730          * 3. We are using part of the extent ending at the end: delete the key
 731          * we found and insert a new key representing the leftover at the
 732          * beginning. There is no net change in the number of extents.
 733          * 4. We are using part of the extent in the middle: delete the key we
 734          * found and insert two new keys representing the leftovers on each
 735          * side. Where we used to have one extent, we now have two, so increment
 736          * the extent count. We may need to convert the block group to bitmaps
 737          * as a result.
 738          */
 739
 740         /* Delete the existing key (cases 1-4). */
 741         ret = btrfs_del_item(trans, root, path);
 742         if (ret)
 743                 goto out;
 744
 745         /* Add a key for leftovers at the beginning (cases 3 and 4). */
 746         if (start > found_start) {
 747                 key.objectid = found_start;
 748                 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
 749                 key.offset = start - found_start;
 750
 751                 btrfs_release_path(path);
 752                 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 753                 if (ret)
 754                         goto out;
 755                 new_extents++;
 756         }
 757
 758         /* Add a key for leftovers at the end (cases 2 and 4). */
 759         if (end < found_end) {
 760                 key.objectid = end;
 761                 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
 762                 key.offset = found_end - end;
 763
 764                 btrfs_release_path(path);
 765                 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 766                 if (ret)
 767                         goto out;
 768                 new_extents++;
 769         }
 770
 771         btrfs_release_path(path);
 772         ret = update_free_space_extent_count(trans, fs_info, block_group, path,
 773                                              new_extents);
 774
 775 out:
 776         return ret;
 777 }
 778
 779 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
 780                                   struct btrfs_fs_info *fs_info,
 781                                   struct btrfs_block_group_cache *block_group,
 782                                   struct btrfs_path *path, u64 start, u64 size)
 783 {
 784         struct btrfs_free_space_info *info;
 785         u32 flags;
 786         int ret;
 787
 788         if (block_group->needs_free_space) {
 789                 ret = __add_block_group_free_space(trans, fs_info, block_group,
 790                                                    path);
 791                 if (ret)
 792                         return ret;
 793         }
 794
 795         info = search_free_space_info(NULL, fs_info, block_group, path, 0);
 796         if (IS_ERR(info))
 797                 return PTR_ERR(info);
 798         flags = btrfs_free_space_flags(path->nodes[0], info);
 799         btrfs_release_path(path);
 800
 801         if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
 802                 return modify_free_space_bitmap(trans, fs_info, block_group,
 803                                                 path, start, size, 1);
 804         } else {
 805                 return remove_free_space_extent(trans, fs_info, block_group,
 806                                                 path, start, size);
 807         }
 808 }
 809
 810 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
 811                                 struct btrfs_fs_info *fs_info,
 812                                 u64 start, u64 size)
 813 {
 814         struct btrfs_block_group_cache *block_group;
 815         struct btrfs_path *path;
 816         int ret;
 817
 818         if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
 819                 return 0;
 820
 821         path = btrfs_alloc_path();
 822         if (!path) {
 823                 ret = -ENOMEM;
 824                 goto out;
 825         }
 826
 827         block_group = btrfs_lookup_block_group(fs_info, start);
 828         if (!block_group) {
 829                 ASSERT(0);
 830                 ret = -ENOENT;
 831                 goto out;
 832         }
 833
 834         mutex_lock(&block_group->free_space_lock);
 835         ret = __remove_from_free_space_tree(trans, fs_info, block_group, path,
 836                                             start, size);
 837         mutex_unlock(&block_group->free_space_lock);
 838
 839         btrfs_put_block_group(block_group);
 840 out:
 841         btrfs_free_path(path);
 842         if (ret)
 843                 btrfs_abort_transaction(trans, ret);
 844         return ret;
 845 }
 846
 847 static int add_free_space_extent(struct btrfs_trans_handle *trans,
 848                                  struct btrfs_fs_info *fs_info,
 849                                  struct btrfs_block_group_cache *block_group,
 850                                  struct btrfs_path *path,
 851                                  u64 start, u64 size)
 852 {
 853         struct btrfs_root *root = fs_info->free_space_root;
 854         struct btrfs_key key, new_key;
 855         u64 found_start, found_end;
 856         u64 end = start + size;
 857         int new_extents = 1;
 858         int ret;
 859
 860         /*
 861          * We are adding a new extent of free space, but we need to merge
 862          * extents. There are four cases here:
 863          *
 864          * 1. The new extent does not have any immediate neighbors to merge
 865          * with: add the new key and increment the free space extent count. We
 866          * may need to convert the block group to bitmaps as a result.
 867          * 2. The new extent has an immediate neighbor before it: remove the
 868          * previous key and insert a new key combining both of them. There is no
 869          * net change in the number of extents.
 870          * 3. The new extent has an immediate neighbor after it: remove the next
 871          * key and insert a new key combining both of them. There is no net
 872          * change in the number of extents.
 873          * 4. The new extent has immediate neighbors on both sides: remove both
 874          * of the keys and insert a new key combining all of them. Where we used
 875          * to have two extents, we now have one, so decrement the extent count.
 876          */
 877
 878         new_key.objectid = start;
 879         new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
 880         new_key.offset = size;
 881
 882         /* Search for a neighbor on the left. */
 883         if (start == block_group->key.objectid)
 884                 goto right;
 885         key.objectid = start - 1;
 886         key.type = (u8)-1;
 887         key.offset = (u64)-1;
 888
 889         ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
 890         if (ret)
 891                 goto out;
 892
 893         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 894
 895         if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
 896                 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
 897                 btrfs_release_path(path);
 898                 goto right;
 899         }
 900
 901         found_start = key.objectid;
 902         found_end = key.objectid + key.offset;
 903         ASSERT(found_start >= block_group->key.objectid &&
 904                found_end > block_group->key.objectid);
 905         ASSERT(found_start < start && found_end <= start);
 906
 907         /*
 908          * Delete the neighbor on the left and absorb it into the new key (cases
 909          * 2 and 4).
 910          */
 911         if (found_end == start) {
 912                 ret = btrfs_del_item(trans, root, path);
 913                 if (ret)
 914                         goto out;
 915                 new_key.objectid = found_start;
 916                 new_key.offset += key.offset;
 917                 new_extents--;
 918         }
 919         btrfs_release_path(path);
 920
 921 right:
 922         /* Search for a neighbor on the right. */
 923         if (end == block_group->key.objectid + block_group->key.offset)
 924                 goto insert;
 925         key.objectid = end;
 926         key.type = (u8)-1;
 927         key.offset = (u64)-1;
 928
 929         ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
 930         if (ret)
 931                 goto out;
 932
 933         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 934
 935         if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
 936                 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
 937                 btrfs_release_path(path);
 938                 goto insert;
 939         }
 940
 941         found_start = key.objectid;
 942         found_end = key.objectid + key.offset;
 943         ASSERT(found_start >= block_group->key.objectid &&
 944                found_end > block_group->key.objectid);
 945         ASSERT((found_start < start && found_end <= start) ||
 946                (found_start >= end && found_end > end));
 947
 948         /*
 949          * Delete the neighbor on the right and absorb it into the new key
 950          * (cases 3 and 4).
 951          */
 952         if (found_start == end) {
 953                 ret = btrfs_del_item(trans, root, path);
 954                 if (ret)
 955                         goto out;
 956                 new_key.offset += key.offset;
 957                 new_extents--;
 958         }
 959         btrfs_release_path(path);
 960
 961 insert:
 962         /* Insert the new key (cases 1-4). */
 963         ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
 964         if (ret)
 965                 goto out;
 966
 967         btrfs_release_path(path);
 968         ret = update_free_space_extent_count(trans, fs_info, block_group, path,
 969                                              new_extents);
 970
 971 out:
 972         return ret;
 973 }
 974
 975 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
 976                              struct btrfs_fs_info *fs_info,
 977                              struct btrfs_block_group_cache *block_group,
 978                              struct btrfs_path *path, u64 start, u64 size)
 979 {
 980         struct btrfs_free_space_info *info;
 981         u32 flags;
 982         int ret;
 983
 984         if (block_group->needs_free_space) {
 985                 ret = __add_block_group_free_space(trans, fs_info, block_group,
 986                                                    path);
 987                 if (ret)
 988                         return ret;
 989         }
 990
 991         info = search_free_space_info(NULL, fs_info, block_group, path, 0);
 992         if (IS_ERR(info))
 993                 return PTR_ERR(info);
 994         flags = btrfs_free_space_flags(path->nodes[0], info);
 995         btrfs_release_path(path);
 996
 997         if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
 998                 return modify_free_space_bitmap(trans, fs_info, block_group,
 999                                                 path, start, size, 0);
1000         } else {
1001                 return add_free_space_extent(trans, fs_info, block_group, path,
1002                                              start, size);
1003         }
1004 }
1005
1006 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1007                            struct btrfs_fs_info *fs_info,
1008                            u64 start, u64 size)
1009 {
1010         struct btrfs_block_group_cache *block_group;
1011         struct btrfs_path *path;
1012         int ret;
1013
1014         if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1015                 return 0;
1016
1017         path = btrfs_alloc_path();
1018         if (!path) {
1019                 ret = -ENOMEM;
1020                 goto out;
1021         }
1022
1023         block_group = btrfs_lookup_block_group(fs_info, start);
1024         if (!block_group) {
1025                 ASSERT(0);
1026                 ret = -ENOENT;
1027                 goto out;
1028         }
1029
1030         mutex_lock(&block_group->free_space_lock);
1031         ret = __add_to_free_space_tree(trans, fs_info, block_group, path, start,
1032                                        size);
1033         mutex_unlock(&block_group->free_space_lock);
1034
1035         btrfs_put_block_group(block_group);
1036 out:
1037         btrfs_free_path(path);
1038         if (ret)
1039                 btrfs_abort_transaction(trans, ret);
1040         return ret;
1041 }
1042
1043 /*
1044  * Populate the free space tree by walking the extent tree. Operations on the
1045  * extent tree that happen as a result of writes to the free space tree will go
1046  * through the normal add/remove hooks.
1047  */
1048 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1049                                     struct btrfs_fs_info *fs_info,
1050                                     struct btrfs_block_group_cache *block_group)
1051 {
1052         struct btrfs_root *extent_root = fs_info->extent_root;
1053         struct btrfs_path *path, *path2;
1054         struct btrfs_key key;
1055         u64 start, end;
1056         int ret;
1057
1058         path = btrfs_alloc_path();
1059         if (!path)
1060                 return -ENOMEM;
1061         path->reada = READA_FORWARD;
1062
1063         path2 = btrfs_alloc_path();
1064         if (!path2) {
1065                 btrfs_free_path(path);
1066                 return -ENOMEM;
1067         }
1068
1069         ret = add_new_free_space_info(trans, fs_info, block_group, path2);
1070         if (ret)
1071                 goto out;
1072
1073         mutex_lock(&block_group->free_space_lock);
1074
1075         /*
1076          * Iterate through all of the extent and metadata items in this block
1077          * group, adding the free space between them and the free space at the
1078          * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1079          * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1080          * contained in.
1081          */
1082         key.objectid = block_group->key.objectid;
1083         key.type = BTRFS_EXTENT_ITEM_KEY;
1084         key.offset = 0;
1085
1086         ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1087         if (ret < 0)
1088                 goto out_locked;
1089         ASSERT(ret == 0);
1090
1091         start = block_group->key.objectid;
1092         end = block_group->key.objectid + block_group->key.offset;
1093         while (1) {
1094                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1095
1096                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1097                     key.type == BTRFS_METADATA_ITEM_KEY) {
1098                         if (key.objectid >= end)
1099                                 break;
1100
1101                         if (start < key.objectid) {
1102                                 ret = __add_to_free_space_tree(trans, fs_info,
1103                                                                block_group,
1104                                                                path2, start,
1105                                                                key.objectid -
1106                                                                start);
1107                                 if (ret)
1108                                         goto out_locked;
1109                         }
1110                         start = key.objectid;
1111                         if (key.type == BTRFS_METADATA_ITEM_KEY)
1112                                 start += fs_info->nodesize;
1113                         else
1114                                 start += key.offset;
1115                 } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1116                         if (key.objectid != block_group->key.objectid)
1117                                 break;
1118                 }
1119
1120                 ret = btrfs_next_item(extent_root, path);
1121                 if (ret < 0)
1122                         goto out_locked;
1123                 if (ret)
1124                         break;
1125         }
1126         if (start < end) {
1127                 ret = __add_to_free_space_tree(trans, fs_info, block_group,
1128                                                path2, start, end - start);
1129                 if (ret)
1130                         goto out_locked;
1131         }
1132
1133         ret = 0;
1134 out_locked:
1135         mutex_unlock(&block_group->free_space_lock);
1136 out:
1137         btrfs_free_path(path2);
1138         btrfs_free_path(path);
1139         return ret;
1140 }
1141
1142 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1143 {
1144         struct btrfs_trans_handle *trans;
1145         struct btrfs_root *tree_root = fs_info->tree_root;
1146         struct btrfs_root *free_space_root;
1147         struct btrfs_block_group_cache *block_group;
1148         struct rb_node *node;
1149         int ret;
1150
1151         trans = btrfs_start_transaction(tree_root, 0);
1152         if (IS_ERR(trans))
1153                 return PTR_ERR(trans);
1154
1155         set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1156         free_space_root = btrfs_create_tree(trans, fs_info,
1157                                             BTRFS_FREE_SPACE_TREE_OBJECTID);
1158         if (IS_ERR(free_space_root)) {
1159                 ret = PTR_ERR(free_space_root);
1160                 goto abort;
1161         }
1162         fs_info->free_space_root = free_space_root;
1163
1164         node = rb_first(&fs_info->block_group_cache_tree);
1165         while (node) {
1166                 block_group = rb_entry(node, struct btrfs_block_group_cache,
1167                                        cache_node);
1168                 ret = populate_free_space_tree(trans, fs_info, block_group);
1169                 if (ret)
1170                         goto abort;
1171                 node = rb_next(node);
1172         }
1173
1174         btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1175         btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1176         clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1177
1178         return btrfs_commit_transaction(trans);
1179
1180 abort:
1181         clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1182         btrfs_abort_transaction(trans, ret);
1183         btrfs_end_transaction(trans);
1184         return ret;
1185 }
1186
1187 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1188                                  struct btrfs_root *root)
1189 {
1190         struct btrfs_path *path;
1191         struct btrfs_key key;
1192         int nr;
1193         int ret;
1194
1195         path = btrfs_alloc_path();
1196         if (!path)
1197                 return -ENOMEM;
1198
1199         path->leave_spinning = 1;
1200
1201         key.objectid = 0;
1202         key.type = 0;
1203         key.offset = 0;
1204
1205         while (1) {
1206                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1207                 if (ret < 0)
1208                         goto out;
1209
1210                 nr = btrfs_header_nritems(path->nodes[0]);
1211                 if (!nr)
1212                         break;
1213
1214                 path->slots[0] = 0;
1215                 ret = btrfs_del_items(trans, root, path, 0, nr);
1216                 if (ret)
1217                         goto out;
1218
1219                 btrfs_release_path(path);
1220         }
1221
1222         ret = 0;
1223 out:
1224         btrfs_free_path(path);
1225         return ret;
1226 }
1227
1228 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1229 {
1230         struct btrfs_trans_handle *trans;
1231         struct btrfs_root *tree_root = fs_info->tree_root;
1232         struct btrfs_root *free_space_root = fs_info->free_space_root;
1233         int ret;
1234
1235         trans = btrfs_start_transaction(tree_root, 0);
1236         if (IS_ERR(trans))
1237                 return PTR_ERR(trans);
1238
1239         btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1240         btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1241         fs_info->free_space_root = NULL;
1242
1243         ret = clear_free_space_tree(trans, free_space_root);
1244         if (ret)
1245                 goto abort;
1246
1247         ret = btrfs_del_root(trans, fs_info, &free_space_root->root_key);
1248         if (ret)
1249                 goto abort;
1250
1251         list_del(&free_space_root->dirty_list);
1252
1253         btrfs_tree_lock(free_space_root->node);
1254         clean_tree_block(fs_info, free_space_root->node);
1255         btrfs_tree_unlock(free_space_root->node);
1256         btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1257                               0, 1);
1258
1259         free_extent_buffer(free_space_root->node);
1260         free_extent_buffer(free_space_root->commit_root);
1261         kfree(free_space_root);
1262
1263         return btrfs_commit_transaction(trans);
1264
1265 abort:
1266         btrfs_abort_transaction(trans, ret);
1267         btrfs_end_transaction(trans);
1268         return ret;
1269 }
1270
1271 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1272                                         struct btrfs_fs_info *fs_info,
1273                                         struct btrfs_block_group_cache *block_group,
1274                                         struct btrfs_path *path)
1275 {
1276         int ret;
1277
1278         block_group->needs_free_space = 0;
1279
1280         ret = add_new_free_space_info(trans, fs_info, block_group, path);
1281         if (ret)
1282                 return ret;
1283
1284         return __add_to_free_space_tree(trans, fs_info, block_group, path,
1285                                         block_group->key.objectid,
1286                                         block_group->key.offset);
1287 }
1288
1289 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1290                                struct btrfs_fs_info *fs_info,
1291                                struct btrfs_block_group_cache *block_group)
1292 {
1293         struct btrfs_path *path = NULL;
1294         int ret = 0;
1295
1296         if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1297                 return 0;
1298
1299         mutex_lock(&block_group->free_space_lock);
1300         if (!block_group->needs_free_space)
1301                 goto out;
1302
1303         path = btrfs_alloc_path();
1304         if (!path) {
1305                 ret = -ENOMEM;
1306                 goto out;
1307         }
1308
1309         ret = __add_block_group_free_space(trans, fs_info, block_group, path);
1310
1311 out:
1312         btrfs_free_path(path);
1313         mutex_unlock(&block_group->free_space_lock);
1314         if (ret)
1315                 btrfs_abort_transaction(trans, ret);
1316         return ret;
1317 }
1318
1319 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1320                                   struct btrfs_fs_info *fs_info,
1321                                   struct btrfs_block_group_cache *block_group)
1322 {
1323         struct btrfs_root *root = fs_info->free_space_root;
1324         struct btrfs_path *path;
1325         struct btrfs_key key, found_key;
1326         struct extent_buffer *leaf;
1327         u64 start, end;
1328         int done = 0, nr;
1329         int ret;
1330
1331         if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1332                 return 0;
1333
1334         if (block_group->needs_free_space) {
1335                 /* We never added this block group to the free space tree. */
1336                 return 0;
1337         }
1338
1339         path = btrfs_alloc_path();
1340         if (!path) {
1341                 ret = -ENOMEM;
1342                 goto out;
1343         }
1344
1345         start = block_group->key.objectid;
1346         end = block_group->key.objectid + block_group->key.offset;
1347
1348         key.objectid = end - 1;
1349         key.type = (u8)-1;
1350         key.offset = (u64)-1;
1351
1352         while (!done) {
1353                 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1354                 if (ret)
1355                         goto out;
1356
1357                 leaf = path->nodes[0];
1358                 nr = 0;
1359                 path->slots[0]++;
1360                 while (path->slots[0] > 0) {
1361                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1362
1363                         if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1364                                 ASSERT(found_key.objectid == block_group->key.objectid);
1365                                 ASSERT(found_key.offset == block_group->key.offset);
1366                                 done = 1;
1367                                 nr++;
1368                                 path->slots[0]--;
1369                                 break;
1370                         } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1371                                    found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1372                                 ASSERT(found_key.objectid >= start);
1373                                 ASSERT(found_key.objectid < end);
1374                                 ASSERT(found_key.objectid + found_key.offset <= end);
1375                                 nr++;
1376                                 path->slots[0]--;
1377                         } else {
1378                                 ASSERT(0);
1379                         }
1380                 }
1381
1382                 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1383                 if (ret)
1384                         goto out;
1385                 btrfs_release_path(path);
1386         }
1387
1388         ret = 0;
1389 out:
1390         btrfs_free_path(path);
1391         if (ret)
1392                 btrfs_abort_transaction(trans, ret);
1393         return ret;
1394 }
1395
1396 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1397                                    struct btrfs_path *path,
1398                                    u32 expected_extent_count)
1399 {
1400         struct btrfs_block_group_cache *block_group;
1401         struct btrfs_fs_info *fs_info;
1402         struct btrfs_root *root;
1403         struct btrfs_key key;
1404         int prev_bit = 0, bit;
1405         /* Initialize to silence GCC. */
1406         u64 extent_start = 0;
1407         u64 end, offset;
1408         u64 total_found = 0;
1409         u32 extent_count = 0;
1410         int ret;
1411
1412         block_group = caching_ctl->block_group;
1413         fs_info = block_group->fs_info;
1414         root = fs_info->free_space_root;
1415
1416         end = block_group->key.objectid + block_group->key.offset;
1417
1418         while (1) {
1419                 ret = btrfs_next_item(root, path);
1420                 if (ret < 0)
1421                         goto out;
1422                 if (ret)
1423                         break;
1424
1425                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1426
1427                 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1428                         break;
1429
1430                 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1431                 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1432
1433                 caching_ctl->progress = key.objectid;
1434
1435                 offset = key.objectid;
1436                 while (offset < key.objectid + key.offset) {
1437                         bit = free_space_test_bit(block_group, path, offset);
1438                         if (prev_bit == 0 && bit == 1) {
1439                                 extent_start = offset;
1440                         } else if (prev_bit == 1 && bit == 0) {
1441                                 total_found += add_new_free_space(block_group,
1442                                                                   fs_info,
1443                                                                   extent_start,
1444                                                                   offset);
1445                                 if (total_found > CACHING_CTL_WAKE_UP) {
1446                                         total_found = 0;
1447                                         wake_up(&caching_ctl->wait);
1448                                 }
1449                                 extent_count++;
1450                         }
1451                         prev_bit = bit;
1452                         offset += fs_info->sectorsize;
1453                 }
1454         }
1455         if (prev_bit == 1) {
1456                 total_found += add_new_free_space(block_group, fs_info,
1457                                                   extent_start, end);
1458                 extent_count++;
1459         }
1460
1461         if (extent_count != expected_extent_count) {
1462                 btrfs_err(fs_info,
1463                           "incorrect extent count for %llu; counted %u, expected %u",
1464                           block_group->key.objectid, extent_count,
1465                           expected_extent_count);
1466                 ASSERT(0);
1467                 ret = -EIO;
1468                 goto out;
1469         }
1470
1471         caching_ctl->progress = (u64)-1;
1472
1473         ret = 0;
1474 out:
1475         return ret;
1476 }
1477
1478 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1479                                    struct btrfs_path *path,
1480                                    u32 expected_extent_count)
1481 {
1482         struct btrfs_block_group_cache *block_group;
1483         struct btrfs_fs_info *fs_info;
1484         struct btrfs_root *root;
1485         struct btrfs_key key;
1486         u64 end;
1487         u64 total_found = 0;
1488         u32 extent_count = 0;
1489         int ret;
1490
1491         block_group = caching_ctl->block_group;
1492         fs_info = block_group->fs_info;
1493         root = fs_info->free_space_root;
1494
1495         end = block_group->key.objectid + block_group->key.offset;
1496
1497         while (1) {
1498                 ret = btrfs_next_item(root, path);
1499                 if (ret < 0)
1500                         goto out;
1501                 if (ret)
1502                         break;
1503
1504                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1505
1506                 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1507                         break;
1508
1509                 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1510                 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1511
1512                 caching_ctl->progress = key.objectid;
1513
1514                 total_found += add_new_free_space(block_group, fs_info,
1515                                                   key.objectid,
1516                                                   key.objectid + key.offset);
1517                 if (total_found > CACHING_CTL_WAKE_UP) {
1518                         total_found = 0;
1519                         wake_up(&caching_ctl->wait);
1520                 }
1521                 extent_count++;
1522         }
1523
1524         if (extent_count != expected_extent_count) {
1525                 btrfs_err(fs_info,
1526                           "incorrect extent count for %llu; counted %u, expected %u",
1527                           block_group->key.objectid, extent_count,
1528                           expected_extent_count);
1529                 ASSERT(0);
1530                 ret = -EIO;
1531                 goto out;
1532         }
1533
1534         caching_ctl->progress = (u64)-1;
1535
1536         ret = 0;
1537 out:
1538         return ret;
1539 }
1540
1541 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1542 {
1543         struct btrfs_block_group_cache *block_group;
1544         struct btrfs_fs_info *fs_info;
1545         struct btrfs_free_space_info *info;
1546         struct btrfs_path *path;
1547         u32 extent_count, flags;
1548         int ret;
1549
1550         block_group = caching_ctl->block_group;
1551         fs_info = block_group->fs_info;
1552
1553         path = btrfs_alloc_path();
1554         if (!path)
1555                 return -ENOMEM;
1556
1557         /*
1558          * Just like caching_thread() doesn't want to deadlock on the extent
1559          * tree, we don't want to deadlock on the free space tree.
1560          */
1561         path->skip_locking = 1;
1562         path->search_commit_root = 1;
1563         path->reada = READA_FORWARD;
1564
1565         info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1566         if (IS_ERR(info)) {
1567                 ret = PTR_ERR(info);
1568                 goto out;
1569         }
1570         extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1571         flags = btrfs_free_space_flags(path->nodes[0], info);
1572
1573         /*
1574          * We left path pointing to the free space info item, so now
1575          * load_free_space_foo can just iterate through the free space tree from
1576          * there.
1577          */
1578         if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1579                 ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1580         else
1581                 ret = load_free_space_extents(caching_ctl, path, extent_count);
1582
1583 out:
1584         btrfs_free_path(path);
1585         return ret;
1586 }