disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #define _XOPEN_SOURCE 600
  20 #define __USE_XOPEN2K
  21 #define _GNU_SOURCE 1
  22 #include <stdio.h>
  23 #include <stdlib.h>
  24 #include <sys/types.h>
  25 #include <sys/stat.h>
  26 #include <fcntl.h>
  27 #include <unistd.h>
  28 #include "kerncompat.h"
  29 #include "radix-tree.h"
  30 #include "ctree.h"
  31 #include "disk-io.h"
  32 #include "volumes.h"
  33 #include "transaction.h"
  34 #include "crc32c.h"
  35 #include "utils.h"
  36 #include "print-tree.h"
  37
  38 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  39 {
  40
  41         struct btrfs_fs_devices *fs_devices;
  42         int ret = 1;
  43
  44         if (buf->start != btrfs_header_bytenr(buf))
  45                 return ret;
  46
  47         fs_devices = root->fs_info->fs_devices;
  48         while (fs_devices) {
  49                 if (!memcmp_extent_buffer(buf, fs_devices->fsid,
  50                                           (unsigned long)btrfs_header_fsid(buf),
  51                                           BTRFS_FSID_SIZE)) {
  52                         ret = 0;
  53                         break;
  54                 }
  55                 fs_devices = fs_devices->seed;
  56         }
  57         return ret;
  58 }
  59
  60 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
  61 {
  62         return crc32c(seed, data, len);
  63 }
  64
  65 void btrfs_csum_final(u32 crc, char *result)
  66 {
  67         *(__le32 *)result = ~cpu_to_le32(crc);
  68 }
  69
  70 int csum_tree_block_size(struct extent_buffer *buf, u16 csum_size,
  71                          int verify)
  72 {
  73         char *result;
  74         u32 len;
  75         u32 crc = ~(u32)0;
  76
  77         result = malloc(csum_size * sizeof(char));
  78         if (!result)
  79                 return 1;
  80
  81         len = buf->len - BTRFS_CSUM_SIZE;
  82         crc = crc32c(crc, buf->data + BTRFS_CSUM_SIZE, len);
  83         btrfs_csum_final(crc, result);
  84
  85         if (verify) {
  86                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
  87                         printk("checksum verify failed on %llu wanted %X "
  88                                "found %X\n", (unsigned long long)buf->start,
  89                                *((int *)result), *((char *)buf->data));
  90                         free(result);
  91                         return 1;
  92                 }
  93         } else {
  94                 write_extent_buffer(buf, result, 0, csum_size);
  95         }
  96         free(result);
  97         return 0;
  98 }
  99
 100 int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 101                     int verify)
 102 {
 103         u16 csum_size =
 104                 btrfs_super_csum_size(&root->fs_info->super_copy);
 105         return csum_tree_block_size(buf, csum_size, verify);
 106 }
 107
 108 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 109                                             u64 bytenr, u32 blocksize)
 110 {
 111         return find_extent_buffer(&root->fs_info->extent_cache,
 112                                   bytenr, blocksize);
 113 }
 114
 115 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 116                                                  u64 bytenr, u32 blocksize)
 117 {
 118         return alloc_extent_buffer(&root->fs_info->extent_cache, bytenr,
 119                                    blocksize);
 120 }
 121
 122 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 123                          u64 parent_transid)
 124 {
 125         int ret;
 126         int dev_nr;
 127         struct extent_buffer *eb;
 128         u64 length;
 129         struct btrfs_multi_bio *multi = NULL;
 130         struct btrfs_device *device;
 131
 132         eb = btrfs_find_tree_block(root, bytenr, blocksize);
 133         if (eb && btrfs_buffer_uptodate(eb, parent_transid)) {
 134                 free_extent_buffer(eb);
 135                 return 0;
 136         }
 137
 138         dev_nr = 0;
 139         length = blocksize;
 140         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
 141                               bytenr, &length, &multi, 0);
 142         BUG_ON(ret);
 143         device = multi->stripes[0].dev;
 144         device->total_ios++;
 145         blocksize = min(blocksize, (u32)(64 * 1024));
 146         readahead(device->fd, multi->stripes[0].physical, blocksize);
 147         kfree(multi);
 148         return 0;
 149 }
 150
 151 static int verify_parent_transid(struct extent_io_tree *io_tree,
 152                                  struct extent_buffer *eb, u64 parent_transid)
 153 {
 154         int ret;
 155
 156         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 157                 return 0;
 158
 159         if (extent_buffer_uptodate(eb) &&
 160             btrfs_header_generation(eb) == parent_transid) {
 161                 ret = 0;
 162                 goto out;
 163         }
 164         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 165                (unsigned long long)eb->start,
 166                (unsigned long long)parent_transid,
 167                (unsigned long long)btrfs_header_generation(eb));
 168         ret = 1;
 169 out:
 170         clear_extent_buffer_uptodate(io_tree, eb);
 171         return ret;
 172
 173 }
 174
 175
 176 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 177                                      u32 blocksize, u64 parent_transid)
 178 {
 179         int ret;
 180         int dev_nr;
 181         struct extent_buffer *eb;
 182         u64 length;
 183         struct btrfs_multi_bio *multi = NULL;
 184         struct btrfs_device *device;
 185         int mirror_num = 0;
 186         int num_copies;
 187
 188         eb = btrfs_find_create_tree_block(root, bytenr, blocksize);
 189         if (!eb)
 190                 return NULL;
 191
 192         if (btrfs_buffer_uptodate(eb, parent_transid))
 193                 return eb;
 194
 195         dev_nr = 0;
 196         length = blocksize;
 197         while (1) {
 198                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
 199                                       eb->start, &length, &multi, mirror_num);
 200                 BUG_ON(ret);
 201                 device = multi->stripes[0].dev;
 202                 eb->fd = device->fd;
 203                 device->total_ios++;
 204                 eb->dev_bytenr = multi->stripes[0].physical;
 205                 kfree(multi);
 206                 ret = read_extent_from_disk(eb);
 207                 if (ret == 0 && check_tree_block(root, eb) == 0 &&
 208                     csum_tree_block(root, eb, 1) == 0 &&
 209                     verify_parent_transid(eb->tree, eb, parent_transid) == 0) {
 210                         btrfs_set_buffer_uptodate(eb);
 211                         return eb;
 212                 }
 213                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 214                                               eb->start, eb->len);
 215                 if (num_copies == 1) {
 216                         break;
 217                 }
 218                 mirror_num++;
 219                 if (mirror_num > num_copies) {
 220                         break;
 221                 }
 222         }
 223         free_extent_buffer(eb);
 224         return NULL;
 225 }
 226
 227 int write_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 228                      struct extent_buffer *eb)
 229 {
 230         int ret;
 231         int dev_nr;
 232         u64 length;
 233         struct btrfs_multi_bio *multi = NULL;
 234
 235         if (check_tree_block(root, eb))
 236                 BUG();
 237         if (!btrfs_buffer_uptodate(eb, trans->transid))
 238                 BUG();
 239
 240         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 241         csum_tree_block(root, eb, 0);
 242
 243         dev_nr = 0;
 244         length = eb->len;
 245         ret = btrfs_map_block(&root->fs_info->mapping_tree, WRITE,
 246                               eb->start, &length, &multi, 0);
 247
 248         while(dev_nr < multi->num_stripes) {
 249                 BUG_ON(ret);
 250                 eb->fd = multi->stripes[dev_nr].dev->fd;
 251                 eb->dev_bytenr = multi->stripes[dev_nr].physical;
 252                 multi->stripes[dev_nr].dev->total_ios++;
 253                 dev_nr++;
 254                 ret = write_extent_to_disk(eb);
 255                 BUG_ON(ret);
 256         }
 257         kfree(multi);
 258         return 0;
 259 }
 260
 261 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 262                         u32 stripesize, struct btrfs_root *root,
 263                         struct btrfs_fs_info *fs_info, u64 objectid)
 264 {
 265         root->node = NULL;
 266         root->commit_root = NULL;
 267         root->sectorsize = sectorsize;
 268         root->nodesize = nodesize;
 269         root->leafsize = leafsize;
 270         root->stripesize = stripesize;
 271         root->ref_cows = 0;
 272         root->track_dirty = 0;
 273
 274         root->fs_info = fs_info;
 275         root->objectid = objectid;
 276         root->last_trans = 0;
 277         root->highest_inode = 0;
 278         root->last_inode_alloc = 0;
 279
 280         INIT_LIST_HEAD(&root->dirty_list);
 281         memset(&root->root_key, 0, sizeof(root->root_key));
 282         memset(&root->root_item, 0, sizeof(root->root_item));
 283         root->root_key.objectid = objectid;
 284         return 0;
 285 }
 286
 287 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 288                                struct btrfs_root *root)
 289 {
 290         int ret;
 291         u64 old_root_bytenr;
 292         struct btrfs_root *tree_root = root->fs_info->tree_root;
 293
 294         btrfs_write_dirty_block_groups(trans, root);
 295         while(1) {
 296                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 297                 if (old_root_bytenr == root->node->start)
 298                         break;
 299                 btrfs_set_root_bytenr(&root->root_item,
 300                                        root->node->start);
 301                 btrfs_set_root_generation(&root->root_item,
 302                                           trans->transid);
 303                 root->root_item.level = btrfs_header_level(root->node);
 304                 ret = btrfs_update_root(trans, tree_root,
 305                                         &root->root_key,
 306                                         &root->root_item);
 307                 BUG_ON(ret);
 308                 btrfs_write_dirty_block_groups(trans, root);
 309         }
 310         return 0;
 311 }
 312
 313 static int commit_tree_roots(struct btrfs_trans_handle *trans,
 314                              struct btrfs_fs_info *fs_info)
 315 {
 316         struct btrfs_root *root;
 317         struct list_head *next;
 318         struct extent_buffer *eb;
 319
 320         if (fs_info->readonly)
 321                 return 0;
 322
 323         eb = fs_info->tree_root->node;
 324         extent_buffer_get(eb);
 325         btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
 326         free_extent_buffer(eb);
 327
 328         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 329                 next = fs_info->dirty_cowonly_roots.next;
 330                 list_del_init(next);
 331                 root = list_entry(next, struct btrfs_root, dirty_list);
 332                 update_cowonly_root(trans, root);
 333         }
 334         return 0;
 335 }
 336
 337 static int __commit_transaction(struct btrfs_trans_handle *trans,
 338                                 struct btrfs_root *root)
 339 {
 340         u64 start;
 341         u64 end;
 342         struct extent_buffer *eb;
 343         struct extent_io_tree *tree = &root->fs_info->extent_cache;
 344         int ret;
 345
 346         while(1) {
 347                 ret = find_first_extent_bit(tree, 0, &start, &end,
 348                                             EXTENT_DIRTY);
 349                 if (ret)
 350                         break;
 351                 while(start <= end) {
 352                         eb = find_first_extent_buffer(tree, start);
 353                         BUG_ON(!eb || eb->start != start);
 354                         ret = write_tree_block(trans, root, eb);
 355                         BUG_ON(ret);
 356                         start += eb->len;
 357                         clear_extent_buffer_dirty(eb);
 358                         free_extent_buffer(eb);
 359                 }
 360         }
 361         return 0;
 362 }
 363
 364 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 365                              struct btrfs_root *root)
 366 {
 367         int ret = 0;
 368         struct btrfs_fs_info *fs_info = root->fs_info;
 369
 370         if (root->commit_root == root->node)
 371                 goto commit_tree;
 372
 373         free_extent_buffer(root->commit_root);
 374         root->commit_root = NULL;
 375
 376         btrfs_set_root_bytenr(&root->root_item, root->node->start);
 377         btrfs_set_root_generation(&root->root_item, trans->transid);
 378         root->root_item.level = btrfs_header_level(root->node);
 379         ret = btrfs_update_root(trans, root->fs_info->tree_root,
 380                                 &root->root_key, &root->root_item);
 381         BUG_ON(ret);
 382 commit_tree:
 383         ret = commit_tree_roots(trans, fs_info);
 384         BUG_ON(ret);
 385         ret = __commit_transaction(trans, root);
 386         BUG_ON(ret);
 387         write_ctree_super(trans, root);
 388         btrfs_finish_extent_commit(trans, fs_info->extent_root,
 389                                    &fs_info->pinned_extents);
 390         btrfs_free_transaction(root, trans);
 391         free_extent_buffer(root->commit_root);
 392         root->commit_root = NULL;
 393         fs_info->running_transaction = NULL;
 394         return 0;
 395 }
 396
 397 static int find_and_setup_root(struct btrfs_root *tree_root,
 398                                struct btrfs_fs_info *fs_info,
 399                                u64 objectid, struct btrfs_root *root)
 400 {
 401         int ret;
 402         u32 blocksize;
 403         u64 generation;
 404
 405         __setup_root(tree_root->nodesize, tree_root->leafsize,
 406                      tree_root->sectorsize, tree_root->stripesize,
 407                      root, fs_info, objectid);
 408         ret = btrfs_find_last_root(tree_root, objectid,
 409                                    &root->root_item, &root->root_key);
 410         BUG_ON(ret);
 411
 412         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 413         generation = btrfs_root_generation(&root->root_item);
 414         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 415                                      blocksize, generation);
 416         BUG_ON(!root->node);
 417         return 0;
 418 }
 419
 420 static int find_and_setup_log_root(struct btrfs_root *tree_root,
 421                                struct btrfs_fs_info *fs_info,
 422                                struct btrfs_super_block *disk_super)
 423 {
 424         u32 blocksize;
 425         u64 blocknr = btrfs_super_log_root(disk_super);
 426         struct btrfs_root *log_root = malloc(sizeof(struct btrfs_root));
 427
 428         if (blocknr == 0)
 429                 return 0;
 430
 431         blocksize = btrfs_level_size(tree_root,
 432                              btrfs_super_log_root_level(disk_super));
 433
 434         __setup_root(tree_root->nodesize, tree_root->leafsize,
 435                      tree_root->sectorsize, tree_root->stripesize,
 436                      log_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 437
 438         log_root->node = read_tree_block(tree_root, blocknr,
 439                                      blocksize,
 440                                      btrfs_super_generation(disk_super) + 1);
 441
 442         fs_info->log_root_tree = log_root;
 443         BUG_ON(!log_root->node);
 444         return 0;
 445 }
 446
 447
 448 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info,
 449                        struct btrfs_root *root)
 450 {
 451         if (root->node)
 452                 free_extent_buffer(root->node);
 453         if (root->commit_root)
 454                 free_extent_buffer(root->commit_root);
 455         kfree(root);
 456         return 0;
 457 }
 458
 459 static int free_fs_roots(struct btrfs_fs_info *fs_info)
 460 {
 461         struct cache_extent *cache;
 462         struct btrfs_root *root;
 463
 464         while (1) {
 465                 cache = find_first_cache_extent(&fs_info->fs_root_cache, 0);
 466                 if (!cache)
 467                         break;
 468                 root = container_of(cache, struct btrfs_root, cache);
 469                 remove_cache_extent(&fs_info->fs_root_cache, cache);
 470                 btrfs_free_fs_root(fs_info, root);
 471         }
 472         return 0;
 473 }
 474
 475 struct btrfs_root *btrfs_read_fs_root_no_cache(struct btrfs_fs_info *fs_info,
 476                                                struct btrfs_key *location)
 477 {
 478         struct btrfs_root *root;
 479         struct btrfs_root *tree_root = fs_info->tree_root;
 480         struct btrfs_path *path;
 481         struct extent_buffer *l;
 482         u64 generation;
 483         u32 blocksize;
 484         int ret = 0;
 485
 486         root = malloc(sizeof(*root));
 487         if (!root)
 488                 return ERR_PTR(-ENOMEM);
 489         memset(root, 0, sizeof(*root));
 490         if (location->offset == (u64)-1) {
 491                 ret = find_and_setup_root(tree_root, fs_info,
 492                                           location->objectid, root);
 493                 if (ret) {
 494                         free(root);
 495                         return ERR_PTR(ret);
 496                 }
 497                 goto insert;
 498         }
 499
 500         __setup_root(tree_root->nodesize, tree_root->leafsize,
 501                      tree_root->sectorsize, tree_root->stripesize,
 502                      root, fs_info, location->objectid);
 503
 504         path = btrfs_alloc_path();
 505         BUG_ON(!path);
 506         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 507         if (ret != 0) {
 508                 if (ret > 0)
 509                         ret = -ENOENT;
 510                 goto out;
 511         }
 512         l = path->nodes[0];
 513         read_extent_buffer(l, &root->root_item,
 514                btrfs_item_ptr_offset(l, path->slots[0]),
 515                sizeof(root->root_item));
 516         memcpy(&root->root_key, location, sizeof(*location));
 517         ret = 0;
 518 out:
 519         btrfs_release_path(root, path);
 520         btrfs_free_path(path);
 521         if (ret) {
 522                 free(root);
 523                 return ERR_PTR(ret);
 524         }
 525         generation = btrfs_root_generation(&root->root_item);
 526         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 527         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 528                                      blocksize, generation);
 529         BUG_ON(!root->node);
 530 insert:
 531         root->ref_cows = 1;
 532         return root;
 533 }
 534
 535 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 536                                       struct btrfs_key *location)
 537 {
 538         struct btrfs_root *root;
 539         struct cache_extent *cache;
 540         int ret;
 541
 542         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 543                 return fs_info->tree_root;
 544         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 545                 return fs_info->extent_root;
 546         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 547                 return fs_info->chunk_root;
 548         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 549                 return fs_info->dev_root;
 550         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
 551                 return fs_info->csum_root;
 552
 553         BUG_ON(location->objectid == BTRFS_TREE_RELOC_OBJECTID ||
 554                location->offset != (u64)-1);
 555
 556         cache = find_cache_extent(&fs_info->fs_root_cache,
 557                                   location->objectid, 1);
 558         if (cache)
 559                 return container_of(cache, struct btrfs_root, cache);
 560
 561         root = btrfs_read_fs_root_no_cache(fs_info, location);
 562         if (IS_ERR(root))
 563                 return root;
 564
 565         root->cache.start = location->objectid;
 566         root->cache.size = 1;
 567         ret = insert_existing_cache_extent(&fs_info->fs_root_cache,
 568                                            &root->cache);
 569         BUG_ON(ret);
 570         return root;
 571 }
 572
 573 struct btrfs_root *open_ctree(const char *filename, u64 sb_bytenr, int writes)
 574 {
 575         int fp;
 576         struct btrfs_root *root;
 577         int flags = O_CREAT | O_RDWR;
 578
 579         if (!writes)
 580                 flags = O_RDONLY;
 581
 582         fp = open(filename, flags, 0600);
 583         if (fp < 0) {
 584                 fprintf (stderr, "Could not open %s\n", filename);
 585                 return NULL;
 586         }
 587         root = open_ctree_fd(fp, filename, sb_bytenr, writes);
 588         close(fp);
 589
 590         return root;
 591 }
 592
 593 struct btrfs_root *open_ctree_fd(int fp, const char *path, u64 sb_bytenr,
 594                                  int writes)
 595 {
 596         u32 sectorsize;
 597         u32 nodesize;
 598         u32 leafsize;
 599         u32 blocksize;
 600         u32 stripesize;
 601         u64 generation;
 602         struct btrfs_key key;
 603         struct btrfs_root *tree_root = malloc(sizeof(struct btrfs_root));
 604         struct btrfs_root *extent_root = malloc(sizeof(struct btrfs_root));
 605         struct btrfs_root *chunk_root = malloc(sizeof(struct btrfs_root));
 606         struct btrfs_root *dev_root = malloc(sizeof(struct btrfs_root));
 607         struct btrfs_root *csum_root = malloc(sizeof(struct btrfs_root));
 608         struct btrfs_fs_info *fs_info = malloc(sizeof(*fs_info));
 609         int ret;
 610         struct btrfs_super_block *disk_super;
 611         struct btrfs_fs_devices *fs_devices = NULL;
 612         u64 total_devs;
 613         u64 features;
 614
 615         if (sb_bytenr == 0)
 616                 sb_bytenr = BTRFS_SUPER_INFO_OFFSET;
 617
 618         ret = btrfs_scan_one_device(fp, path, &fs_devices,
 619                                     &total_devs, sb_bytenr);
 620
 621         if (ret) {
 622                 fprintf(stderr, "No valid Btrfs found on %s\n", path);
 623                 return NULL;
 624         }
 625
 626         if (total_devs != 1) {
 627                 ret = btrfs_scan_for_fsid(fs_devices, total_devs, 1);
 628                 BUG_ON(ret);
 629         }
 630
 631         memset(fs_info, 0, sizeof(*fs_info));
 632         fs_info->tree_root = tree_root;
 633         fs_info->extent_root = extent_root;
 634         fs_info->chunk_root = chunk_root;
 635         fs_info->dev_root = dev_root;
 636         fs_info->csum_root = csum_root;
 637
 638         if (!writes)
 639                 fs_info->readonly = 1;
 640
 641         extent_io_tree_init(&fs_info->extent_cache);
 642         extent_io_tree_init(&fs_info->free_space_cache);
 643         extent_io_tree_init(&fs_info->block_group_cache);
 644         extent_io_tree_init(&fs_info->pinned_extents);
 645         extent_io_tree_init(&fs_info->pending_del);
 646         extent_io_tree_init(&fs_info->extent_ins);
 647         cache_tree_init(&fs_info->fs_root_cache);
 648
 649         cache_tree_init(&fs_info->mapping_tree.cache_tree);
 650
 651         mutex_init(&fs_info->fs_mutex);
 652         fs_info->fs_devices = fs_devices;
 653         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
 654         INIT_LIST_HEAD(&fs_info->space_info);
 655
 656         __setup_root(4096, 4096, 4096, 4096, tree_root,
 657                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 658
 659         if (writes)
 660                 ret = btrfs_open_devices(fs_devices, O_RDWR);
 661         else
 662                 ret = btrfs_open_devices(fs_devices, O_RDONLY);
 663         BUG_ON(ret);
 664
 665         fs_info->super_bytenr = sb_bytenr;
 666         disk_super = &fs_info->super_copy;
 667         ret = btrfs_read_dev_super(fs_devices->latest_bdev,
 668                                    disk_super, sb_bytenr);
 669         if (ret) {
 670                 printk("No valid btrfs found\n");
 671                 BUG_ON(1);
 672         }
 673
 674         memcpy(fs_info->fsid, &disk_super->fsid, BTRFS_FSID_SIZE);
 675
 676
 677         features = btrfs_super_incompat_flags(disk_super) &
 678                    ~BTRFS_FEATURE_INCOMPAT_SUPP;
 679         if (features) {
 680                 printk("couldn't open because of unsupported "
 681                        "option features (%Lx).\n", features);
 682                 BUG_ON(1);
 683         }
 684
 685         features = btrfs_super_incompat_flags(disk_super);
 686         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
 687                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
 688                 btrfs_set_super_incompat_flags(disk_super, features);
 689         }
 690
 691         features = btrfs_super_compat_ro_flags(disk_super) &
 692                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
 693         if (writes && features) {
 694                 printk("couldn't open RDWR because of unsupported "
 695                        "option features (%Lx).\n", features);
 696                 BUG_ON(1);
 697         }
 698
 699         nodesize = btrfs_super_nodesize(disk_super);
 700         leafsize = btrfs_super_leafsize(disk_super);
 701         sectorsize = btrfs_super_sectorsize(disk_super);
 702         stripesize = btrfs_super_stripesize(disk_super);
 703         tree_root->nodesize = nodesize;
 704         tree_root->leafsize = leafsize;
 705         tree_root->sectorsize = sectorsize;
 706         tree_root->stripesize = stripesize;
 707
 708         ret = btrfs_read_sys_array(tree_root);
 709         BUG_ON(ret);
 710         blocksize = btrfs_level_size(tree_root,
 711                                      btrfs_super_chunk_root_level(disk_super));
 712         generation = btrfs_super_chunk_root_generation(disk_super);
 713
 714         __setup_root(nodesize, leafsize, sectorsize, stripesize,
 715                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
 716
 717         chunk_root->node = read_tree_block(chunk_root,
 718                                            btrfs_super_chunk_root(disk_super),
 719                                            blocksize, generation);
 720
 721         BUG_ON(!chunk_root->node);
 722
 723         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
 724                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
 725                  BTRFS_UUID_SIZE);
 726
 727         if (!(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)) {
 728                 ret = btrfs_read_chunk_tree(chunk_root);
 729                 BUG_ON(ret);
 730         }
 731
 732         blocksize = btrfs_level_size(tree_root,
 733                                      btrfs_super_root_level(disk_super));
 734         generation = btrfs_super_generation(disk_super);
 735
 736         tree_root->node = read_tree_block(tree_root,
 737                                           btrfs_super_root(disk_super),
 738                                           blocksize, generation);
 739         BUG_ON(!tree_root->node);
 740         ret = find_and_setup_root(tree_root, fs_info,
 741                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 742         BUG_ON(ret);
 743         extent_root->track_dirty = 1;
 744
 745         ret = find_and_setup_root(tree_root, fs_info,
 746                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
 747         BUG_ON(ret);
 748         dev_root->track_dirty = 1;
 749
 750         ret = find_and_setup_root(tree_root, fs_info,
 751                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
 752         BUG_ON(ret);
 753         csum_root->track_dirty = 1;
 754
 755         BUG_ON(ret);
 756
 757         find_and_setup_log_root(tree_root, fs_info, disk_super);
 758
 759         fs_info->generation = generation + 1;
 760         btrfs_read_block_groups(fs_info->tree_root);
 761
 762         key.objectid = BTRFS_FS_TREE_OBJECTID;
 763         key.type = BTRFS_ROOT_ITEM_KEY;
 764         key.offset = (u64)-1;
 765         fs_info->fs_root = btrfs_read_fs_root(fs_info, &key);
 766
 767         fs_info->data_alloc_profile = (u64)-1;
 768         fs_info->metadata_alloc_profile = (u64)-1;
 769         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
 770
 771         return fs_info->fs_root;
 772 }
 773
 774 int btrfs_read_dev_super(int fd, struct btrfs_super_block *sb, u64 sb_bytenr)
 775 {
 776         u8 fsid[BTRFS_FSID_SIZE];
 777         struct btrfs_super_block buf;
 778         int i;
 779         int ret;
 780         u64 transid = 0;
 781         u64 bytenr;
 782
 783         if (sb_bytenr != BTRFS_SUPER_INFO_OFFSET) {
 784                 ret = pread64(fd, &buf, sizeof(buf), sb_bytenr);
 785                 if (ret < sizeof(buf))
 786                         return -1;
 787
 788                 if (btrfs_super_bytenr(&buf) != sb_bytenr ||
 789                     strncmp((char *)(&buf.magic), BTRFS_MAGIC,
 790                             sizeof(buf.magic)))
 791                         return -1;
 792
 793                 memcpy(sb, &buf, sizeof(*sb));
 794                 return 0;
 795         }
 796
 797         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 798                 bytenr = btrfs_sb_offset(i);
 799                 ret = pread64(fd, &buf, sizeof(buf), bytenr);
 800                 if (ret < sizeof(buf))
 801                         break;
 802
 803                 if (btrfs_super_bytenr(&buf) != bytenr ||
 804                     strncmp((char *)(&buf.magic), BTRFS_MAGIC,
 805                             sizeof(buf.magic)))
 806                         continue;
 807
 808                 if (i == 0)
 809                         memcpy(fsid, buf.fsid, sizeof(fsid));
 810                 else if (memcmp(fsid, buf.fsid, sizeof(fsid)))
 811                         continue;
 812
 813                 if (btrfs_super_generation(&buf) > transid) {
 814                         memcpy(sb, &buf, sizeof(*sb));
 815                         transid = btrfs_super_generation(&buf);
 816                 }
 817         }
 818
 819         return transid > 0 ? 0 : -1;
 820 }
 821
 822 int write_dev_supers(struct btrfs_root *root, struct btrfs_super_block *sb,
 823                      struct btrfs_device *device)
 824 {
 825         u64 bytenr;
 826         u32 crc;
 827         int i, ret;
 828
 829         if (root->fs_info->super_bytenr != BTRFS_SUPER_INFO_OFFSET) {
 830                 btrfs_set_super_bytenr(sb, root->fs_info->super_bytenr);
 831
 832                 crc = ~(u32)0;
 833                 crc = btrfs_csum_data(NULL, (char *)sb + BTRFS_CSUM_SIZE, crc,
 834                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 835                 btrfs_csum_final(crc, (char *)&sb->csum[0]);
 836
 837                 ret = pwrite64(device->fd, sb, BTRFS_SUPER_INFO_SIZE,
 838                                root->fs_info->super_bytenr);
 839                 BUG_ON(ret != BTRFS_SUPER_INFO_SIZE);
 840                 return 0;
 841         }
 842
 843         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 844                 bytenr = btrfs_sb_offset(i);
 845                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
 846                         break;
 847
 848                 btrfs_set_super_bytenr(sb, bytenr);
 849
 850                 crc = ~(u32)0;
 851                 crc = btrfs_csum_data(NULL, (char *)sb + BTRFS_CSUM_SIZE, crc,
 852                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 853                 btrfs_csum_final(crc, (char *)&sb->csum[0]);
 854
 855                 ret = pwrite64(device->fd, sb, BTRFS_SUPER_INFO_SIZE, bytenr);
 856                 BUG_ON(ret != BTRFS_SUPER_INFO_SIZE);
 857         }
 858         return 0;
 859 }
 860
 861 int write_all_supers(struct btrfs_root *root)
 862 {
 863         struct list_head *cur;
 864         struct list_head *head = &root->fs_info->fs_devices->devices;
 865         struct btrfs_device *dev;
 866         struct btrfs_super_block *sb;
 867         struct btrfs_dev_item *dev_item;
 868         int ret;
 869         u64 flags;
 870
 871         sb = &root->fs_info->super_copy;
 872         dev_item = &sb->dev_item;
 873         list_for_each(cur, head) {
 874                 dev = list_entry(cur, struct btrfs_device, dev_list);
 875                 if (!dev->writeable)
 876                         continue;
 877
 878                 btrfs_set_stack_device_generation(dev_item, 0);
 879                 btrfs_set_stack_device_type(dev_item, dev->type);
 880                 btrfs_set_stack_device_id(dev_item, dev->devid);
 881                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
 882                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
 883                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
 884                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
 885                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
 886                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
 887                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
 888
 889                 flags = btrfs_super_flags(sb);
 890                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
 891
 892                 ret = write_dev_supers(root, sb, dev);
 893                 BUG_ON(ret);
 894         }
 895         return 0;
 896 }
 897
 898 int write_ctree_super(struct btrfs_trans_handle *trans,
 899                       struct btrfs_root *root)
 900 {
 901         int ret;
 902         struct btrfs_root *tree_root = root->fs_info->tree_root;
 903         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 904
 905         if (root->fs_info->readonly)
 906                 return 0;
 907
 908         btrfs_set_super_generation(&root->fs_info->super_copy,
 909                                    trans->transid);
 910         btrfs_set_super_root(&root->fs_info->super_copy,
 911                              tree_root->node->start);
 912         btrfs_set_super_root_level(&root->fs_info->super_copy,
 913                                    btrfs_header_level(tree_root->node));
 914         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 915                                    chunk_root->node->start);
 916         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 917                                          btrfs_header_level(chunk_root->node));
 918         btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy,
 919                                 btrfs_header_generation(chunk_root->node));
 920
 921         ret = write_all_supers(root);
 922         if (ret)
 923                 fprintf(stderr, "failed to write new super block err %d\n", ret);
 924         return ret;
 925 }
 926
 927 static int close_all_devices(struct btrfs_fs_info *fs_info)
 928 {
 929         struct list_head *list;
 930         struct list_head *next;
 931         struct btrfs_device *device;
 932
 933         return 0;
 934
 935         list = &fs_info->fs_devices->devices;
 936         list_for_each(next, list) {
 937                 device = list_entry(next, struct btrfs_device, dev_list);
 938                 close(device->fd);
 939         }
 940         return 0;
 941 }
 942
 943 int close_ctree(struct btrfs_root *root)
 944 {
 945         int ret;
 946         struct btrfs_trans_handle *trans;
 947         struct btrfs_fs_info *fs_info = root->fs_info;
 948
 949         trans = btrfs_start_transaction(root, 1);
 950         btrfs_commit_transaction(trans, root);
 951         trans = btrfs_start_transaction(root, 1);
 952         ret = commit_tree_roots(trans, fs_info);
 953         BUG_ON(ret);
 954         ret = __commit_transaction(trans, root);
 955         BUG_ON(ret);
 956         write_ctree_super(trans, root);
 957         btrfs_free_transaction(root, trans);
 958         btrfs_free_block_groups(fs_info);
 959
 960         free_fs_roots(fs_info);
 961
 962         if (fs_info->extent_root->node)
 963                 free_extent_buffer(fs_info->extent_root->node);
 964         if (fs_info->tree_root->node)
 965                 free_extent_buffer(fs_info->tree_root->node);
 966         if (fs_info->chunk_root->node)
 967                 free_extent_buffer(fs_info->chunk_root->node);
 968         if (fs_info->dev_root->node)
 969                 free_extent_buffer(fs_info->dev_root->node);
 970         if (fs_info->csum_root->node)
 971                 free_extent_buffer(fs_info->csum_root->node);
 972
 973         if (fs_info->log_root_tree) {
 974                 if (fs_info->log_root_tree->node)
 975                         free_extent_buffer(fs_info->log_root_tree->node);
 976                 free(fs_info->log_root_tree);
 977         }
 978
 979         close_all_devices(fs_info);
 980         extent_io_tree_cleanup(&fs_info->extent_cache);
 981         extent_io_tree_cleanup(&fs_info->free_space_cache);
 982         extent_io_tree_cleanup(&fs_info->block_group_cache);
 983         extent_io_tree_cleanup(&fs_info->pinned_extents);
 984         extent_io_tree_cleanup(&fs_info->pending_del);
 985         extent_io_tree_cleanup(&fs_info->extent_ins);
 986
 987         free(fs_info->tree_root);
 988         free(fs_info->extent_root);
 989         free(fs_info->chunk_root);
 990         free(fs_info->dev_root);
 991         free(fs_info->csum_root);
 992         free(fs_info);
 993
 994         return 0;
 995 }
 996
 997 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 998                      struct extent_buffer *eb)
 999 {
1000         return clear_extent_buffer_dirty(eb);
1001 }
1002
1003 int wait_on_tree_block_writeback(struct btrfs_root *root,
1004                                  struct extent_buffer *eb)
1005 {
1006         return 0;
1007 }
1008
1009 void btrfs_mark_buffer_dirty(struct extent_buffer *eb)
1010 {
1011         set_extent_buffer_dirty(eb);
1012 }
1013
1014 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1015 {
1016         int ret;
1017
1018         ret = extent_buffer_uptodate(buf);
1019         if (!ret)
1020                 return ret;
1021
1022         ret = verify_parent_transid(buf->tree, buf, parent_transid);
1023         return !ret;
1024 }
1025
1026 int btrfs_set_buffer_uptodate(struct extent_buffer *eb)
1027 {
1028         return set_extent_buffer_uptodate(eb);
1029 }