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 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include <sys/types.h>
  22 #include <sys/stat.h>
  23 #include <fcntl.h>
  24 #include <unistd.h>
  25 #include <uuid/uuid.h>
  26 #include "kerncompat.h"
  27 #include "radix-tree.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "volumes.h"
  31 #include "transaction.h"
  32 #include "crc32c.h"
  33 #include "utils.h"
  34 #include "print-tree.h"
  35 #include "rbtree-utils.h"
  36
  37 /* specified errno for check_tree_block */
  38 #define BTRFS_BAD_BYTENR                (-1)
  39 #define BTRFS_BAD_FSID                  (-2)
  40 #define BTRFS_BAD_LEVEL                 (-3)
  41 #define BTRFS_BAD_NRITEMS               (-4)
  42
  43 /* Calculate max possible nritems for a leaf/node */
  44 static u32 max_nritems(u8 level, u32 nodesize)
  45 {
  46
  47         if (level == 0)
  48                 return ((nodesize - sizeof(struct btrfs_header)) /
  49                         sizeof(struct btrfs_item));
  50         return ((nodesize - sizeof(struct btrfs_header)) /
  51                 sizeof(struct btrfs_key_ptr));
  52 }
  53
  54 static int check_tree_block(struct btrfs_fs_info *fs_info,
  55                             struct extent_buffer *buf)
  56 {
  57
  58         struct btrfs_fs_devices *fs_devices;
  59         u32 nodesize = fs_info->nodesize;
  60         int ret = BTRFS_BAD_FSID;
  61
  62         if (buf->start != btrfs_header_bytenr(buf))
  63                 return BTRFS_BAD_BYTENR;
  64         if (btrfs_header_level(buf) >= BTRFS_MAX_LEVEL)
  65                 return BTRFS_BAD_LEVEL;
  66         if (btrfs_header_nritems(buf) > max_nritems(btrfs_header_level(buf),
  67                                                     nodesize))
  68                 return BTRFS_BAD_NRITEMS;
  69
  70         /* Only leaf can be empty */
  71         if (btrfs_header_nritems(buf) == 0 &&
  72             btrfs_header_level(buf) != 0)
  73                 return BTRFS_BAD_NRITEMS;
  74
  75         fs_devices = fs_info->fs_devices;
  76         while (fs_devices) {
  77                 if (fs_info->ignore_fsid_mismatch ||
  78                     !memcmp_extent_buffer(buf, fs_devices->fsid,
  79                                           btrfs_header_fsid(),
  80                                           BTRFS_FSID_SIZE)) {
  81                         ret = 0;
  82                         break;
  83                 }
  84                 fs_devices = fs_devices->seed;
  85         }
  86         return ret;
  87 }
  88
  89 static void print_tree_block_error(struct btrfs_fs_info *fs_info,
  90                                 struct extent_buffer *eb,
  91                                 int err)
  92 {
  93         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = {'\0'};
  94         char found_uuid[BTRFS_UUID_UNPARSED_SIZE] = {'\0'};
  95         u8 buf[BTRFS_UUID_SIZE];
  96
  97         switch (err) {
  98         case BTRFS_BAD_FSID:
  99                 read_extent_buffer(eb, buf, btrfs_header_fsid(),
 100                                    BTRFS_UUID_SIZE);
 101                 uuid_unparse(buf, found_uuid);
 102                 uuid_unparse(fs_info->fsid, fs_uuid);
 103                 fprintf(stderr, "fsid mismatch, want=%s, have=%s\n",
 104                         fs_uuid, found_uuid);
 105                 break;
 106         case BTRFS_BAD_BYTENR:
 107                 fprintf(stderr, "bytenr mismatch, want=%llu, have=%llu\n",
 108                         eb->start, btrfs_header_bytenr(eb));
 109                 break;
 110         case BTRFS_BAD_LEVEL:
 111                 fprintf(stderr, "bad level, %u > %u\n",
 112                         btrfs_header_level(eb), BTRFS_MAX_LEVEL);
 113                 break;
 114         case BTRFS_BAD_NRITEMS:
 115                 fprintf(stderr, "invalid nr_items: %u\n",
 116                         btrfs_header_nritems(eb));
 117                 break;
 118         }
 119 }
 120
 121 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 122 {
 123         return crc32c(seed, data, len);
 124 }
 125
 126 void btrfs_csum_final(u32 crc, u8 *result)
 127 {
 128         put_unaligned_le32(~crc, result);
 129 }
 130
 131 static int __csum_tree_block_size(struct extent_buffer *buf, u16 csum_size,
 132                                   int verify, int silent)
 133 {
 134         u8 result[BTRFS_CSUM_SIZE];
 135         u32 len;
 136         u32 crc = ~(u32)0;
 137
 138         len = buf->len - BTRFS_CSUM_SIZE;
 139         crc = crc32c(crc, buf->data + BTRFS_CSUM_SIZE, len);
 140         btrfs_csum_final(crc, result);
 141
 142         if (verify) {
 143                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 144                         if (!silent)
 145                                 printk("checksum verify failed on %llu found %08X wanted %08X\n",
 146                                        (unsigned long long)buf->start,
 147                                        *((u32 *)result),
 148                                        *((u32*)(char *)buf->data));
 149                         return 1;
 150                 }
 151         } else {
 152                 write_extent_buffer(buf, result, 0, csum_size);
 153         }
 154         return 0;
 155 }
 156
 157 int csum_tree_block_size(struct extent_buffer *buf, u16 csum_size, int verify)
 158 {
 159         return __csum_tree_block_size(buf, csum_size, verify, 0);
 160 }
 161
 162 int verify_tree_block_csum_silent(struct extent_buffer *buf, u16 csum_size)
 163 {
 164         return __csum_tree_block_size(buf, csum_size, 1, 1);
 165 }
 166
 167 int csum_tree_block(struct btrfs_fs_info *fs_info,
 168                     struct extent_buffer *buf, int verify)
 169 {
 170         u16 csum_size =
 171                 btrfs_super_csum_size(fs_info->super_copy);
 172         if (verify && fs_info->suppress_check_block_errors)
 173                 return verify_tree_block_csum_silent(buf, csum_size);
 174         return csum_tree_block_size(buf, csum_size, verify);
 175 }
 176
 177 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 178                                             u64 bytenr, u32 blocksize)
 179 {
 180         return find_extent_buffer(&root->fs_info->extent_cache,
 181                                   bytenr, blocksize);
 182 }
 183
 184 struct extent_buffer* btrfs_find_create_tree_block(
 185                 struct btrfs_fs_info *fs_info, u64 bytenr, u32 blocksize)
 186 {
 187         return alloc_extent_buffer(&fs_info->extent_cache, bytenr, blocksize);
 188 }
 189
 190 void readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 191                           u64 parent_transid)
 192 {
 193         struct extent_buffer *eb;
 194         u64 length;
 195         struct btrfs_multi_bio *multi = NULL;
 196         struct btrfs_device *device;
 197
 198         eb = btrfs_find_tree_block(root, bytenr, blocksize);
 199         if (!(eb && btrfs_buffer_uptodate(eb, parent_transid)) &&
 200             !btrfs_map_block(root->fs_info, READ, bytenr, &length, &multi, 0,
 201                              NULL)) {
 202                 device = multi->stripes[0].dev;
 203                 device->total_ios++;
 204                 blocksize = min(blocksize, (u32)SZ_64K);
 205                 readahead(device->fd, multi->stripes[0].physical, blocksize);
 206         }
 207
 208         free_extent_buffer(eb);
 209         kfree(multi);
 210 }
 211
 212 static int verify_parent_transid(struct extent_io_tree *io_tree,
 213                                  struct extent_buffer *eb, u64 parent_transid,
 214                                  int ignore)
 215 {
 216         int ret;
 217
 218         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 219                 return 0;
 220
 221         if (extent_buffer_uptodate(eb) &&
 222             btrfs_header_generation(eb) == parent_transid) {
 223                 ret = 0;
 224                 goto out;
 225         }
 226         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 227                (unsigned long long)eb->start,
 228                (unsigned long long)parent_transid,
 229                (unsigned long long)btrfs_header_generation(eb));
 230         if (ignore) {
 231                 eb->flags |= EXTENT_BAD_TRANSID;
 232                 printk("Ignoring transid failure\n");
 233                 return 0;
 234         }
 235
 236         ret = 1;
 237 out:
 238         clear_extent_buffer_uptodate(eb);
 239         return ret;
 240
 241 }
 242
 243
 244 int read_whole_eb(struct btrfs_fs_info *info, struct extent_buffer *eb, int mirror)
 245 {
 246         unsigned long offset = 0;
 247         struct btrfs_multi_bio *multi = NULL;
 248         struct btrfs_device *device;
 249         int ret = 0;
 250         u64 read_len;
 251         unsigned long bytes_left = eb->len;
 252
 253         while (bytes_left) {
 254                 read_len = bytes_left;
 255                 device = NULL;
 256
 257                 if (!info->on_restoring &&
 258                     eb->start != BTRFS_SUPER_INFO_OFFSET) {
 259                         ret = btrfs_map_block(info, READ, eb->start + offset,
 260                                               &read_len, &multi, mirror, NULL);
 261                         if (ret) {
 262                                 printk("Couldn't map the block %Lu\n", eb->start + offset);
 263                                 kfree(multi);
 264                                 return -EIO;
 265                         }
 266                         device = multi->stripes[0].dev;
 267
 268                         if (device->fd <= 0) {
 269                                 kfree(multi);
 270                                 return -EIO;
 271                         }
 272
 273                         eb->fd = device->fd;
 274                         device->total_ios++;
 275                         eb->dev_bytenr = multi->stripes[0].physical;
 276                         kfree(multi);
 277                         multi = NULL;
 278                 } else {
 279                         /* special case for restore metadump */
 280                         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
 281                                 if (device->devid == 1)
 282                                         break;
 283                         }
 284
 285                         eb->fd = device->fd;
 286                         eb->dev_bytenr = eb->start;
 287                         device->total_ios++;
 288                 }
 289
 290                 if (read_len > bytes_left)
 291                         read_len = bytes_left;
 292
 293                 ret = read_extent_from_disk(eb, offset, read_len);
 294                 if (ret)
 295                         return -EIO;
 296                 offset += read_len;
 297                 bytes_left -= read_len;
 298         }
 299         return 0;
 300 }
 301
 302 struct extent_buffer* read_tree_block(
 303                 struct btrfs_fs_info *fs_info, u64 bytenr, u32 blocksize,
 304                 u64 parent_transid)
 305 {
 306         int ret;
 307         struct extent_buffer *eb;
 308         u64 best_transid = 0;
 309         u32 sectorsize = fs_info->sectorsize;
 310         u32 nodesize = fs_info->nodesize;
 311         int mirror_num = 0;
 312         int good_mirror = 0;
 313         int num_copies;
 314         int ignore = 0;
 315
 316         /*
 317          * Don't even try to create tree block for unaligned tree block
 318          * bytenr.
 319          * Such unaligned tree block will free overlapping extent buffer,
 320          * causing use-after-free bugs for fuzzed images.
 321          */
 322         if (bytenr < sectorsize || !IS_ALIGNED(bytenr, sectorsize)) {
 323                 error("tree block bytenr %llu is not aligned to sectorsize %u",
 324                       bytenr, sectorsize);
 325                 return ERR_PTR(-EIO);
 326         }
 327         if (blocksize < nodesize || !IS_ALIGNED(blocksize, nodesize)) {
 328                 error("tree block size %u is not aligned to nodesize %u",
 329                       blocksize, nodesize);
 330                 return ERR_PTR(-EIO);
 331         }
 332
 333         eb = btrfs_find_create_tree_block(fs_info, bytenr, blocksize);
 334         if (!eb)
 335                 return ERR_PTR(-ENOMEM);
 336
 337         if (btrfs_buffer_uptodate(eb, parent_transid))
 338                 return eb;
 339
 340         while (1) {
 341                 ret = read_whole_eb(fs_info, eb, mirror_num);
 342                 if (ret == 0 && csum_tree_block(fs_info, eb, 1) == 0 &&
 343                     check_tree_block(fs_info, eb) == 0 &&
 344                     verify_parent_transid(eb->tree, eb, parent_transid, ignore)
 345                     == 0) {
 346                         if (eb->flags & EXTENT_BAD_TRANSID &&
 347                             list_empty(&eb->recow)) {
 348                                 list_add_tail(&eb->recow,
 349                                               &fs_info->recow_ebs);
 350                                 eb->refs++;
 351                         }
 352                         btrfs_set_buffer_uptodate(eb);
 353                         return eb;
 354                 }
 355                 if (ignore) {
 356                         if (check_tree_block(fs_info, eb)) {
 357                                 if (!fs_info->suppress_check_block_errors)
 358                                         print_tree_block_error(fs_info, eb,
 359                                                 check_tree_block(fs_info, eb));
 360                         } else {
 361                                 if (!fs_info->suppress_check_block_errors)
 362                                         fprintf(stderr, "Csum didn't match\n");
 363                         }
 364                         ret = -EIO;
 365                         break;
 366                 }
 367                 num_copies = btrfs_num_copies(fs_info, eb->start, eb->len);
 368                 if (num_copies == 1) {
 369                         ignore = 1;
 370                         continue;
 371                 }
 372                 if (btrfs_header_generation(eb) > best_transid && mirror_num) {
 373                         best_transid = btrfs_header_generation(eb);
 374                         good_mirror = mirror_num;
 375                 }
 376                 mirror_num++;
 377                 if (mirror_num > num_copies) {
 378                         mirror_num = good_mirror;
 379                         ignore = 1;
 380                         continue;
 381                 }
 382         }
 383         free_extent_buffer(eb);
 384         return ERR_PTR(ret);
 385 }
 386
 387 int read_extent_data(struct btrfs_root *root, char *data,
 388                            u64 logical, u64 *len, int mirror)
 389 {
 390         u64 offset = 0;
 391         struct btrfs_multi_bio *multi = NULL;
 392         struct btrfs_fs_info *info = root->fs_info;
 393         struct btrfs_device *device;
 394         int ret = 0;
 395         u64 max_len = *len;
 396
 397         ret = btrfs_map_block(info, READ, logical, len, &multi, mirror, NULL);
 398         if (ret) {
 399                 fprintf(stderr, "Couldn't map the block %llu\n",
 400                                 logical + offset);
 401                 goto err;
 402         }
 403         device = multi->stripes[0].dev;
 404
 405         if (device->fd <= 0)
 406                 goto err;
 407         if (*len > max_len)
 408                 *len = max_len;
 409
 410         ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
 411         if (ret != *len)
 412                 ret = -EIO;
 413         else
 414                 ret = 0;
 415 err:
 416         kfree(multi);
 417         return ret;
 418 }
 419
 420 int write_and_map_eb(struct btrfs_root *root, struct extent_buffer *eb)
 421 {
 422         int ret;
 423         int dev_nr;
 424         u64 length;
 425         u64 *raid_map = NULL;
 426         struct btrfs_multi_bio *multi = NULL;
 427
 428         dev_nr = 0;
 429         length = eb->len;
 430         ret = btrfs_map_block(root->fs_info, WRITE, eb->start, &length,
 431                               &multi, 0, &raid_map);
 432
 433         if (raid_map) {
 434                 ret = write_raid56_with_parity(root->fs_info, eb, multi,
 435                                                length, raid_map);
 436                 BUG_ON(ret);
 437         } else while (dev_nr < multi->num_stripes) {
 438                 BUG_ON(ret);
 439                 eb->fd = multi->stripes[dev_nr].dev->fd;
 440                 eb->dev_bytenr = multi->stripes[dev_nr].physical;
 441                 multi->stripes[dev_nr].dev->total_ios++;
 442                 dev_nr++;
 443                 ret = write_extent_to_disk(eb);
 444                 BUG_ON(ret);
 445         }
 446         kfree(raid_map);
 447         kfree(multi);
 448         return 0;
 449 }
 450
 451 int write_tree_block(struct btrfs_trans_handle *trans,
 452                      struct btrfs_root *root,
 453                      struct extent_buffer *eb)
 454 {
 455         if (check_tree_block(root->fs_info, eb)) {
 456                 print_tree_block_error(root->fs_info, eb,
 457                                 check_tree_block(root->fs_info, eb));
 458                 BUG();
 459         }
 460
 461         if (trans && !btrfs_buffer_uptodate(eb, trans->transid))
 462                 BUG();
 463
 464         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 465         csum_tree_block(root->fs_info, eb, 0);
 466
 467         return write_and_map_eb(root, eb);
 468 }
 469
 470 void btrfs_setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
 471                       u64 objectid)
 472 {
 473         root->node = NULL;
 474         root->commit_root = NULL;
 475         root->ref_cows = 0;
 476         root->track_dirty = 0;
 477
 478         root->fs_info = fs_info;
 479         root->objectid = objectid;
 480         root->last_trans = 0;
 481         root->last_inode_alloc = 0;
 482
 483         INIT_LIST_HEAD(&root->dirty_list);
 484         INIT_LIST_HEAD(&root->orphan_data_extents);
 485         memset(&root->root_key, 0, sizeof(root->root_key));
 486         memset(&root->root_item, 0, sizeof(root->root_item));
 487         root->root_key.objectid = objectid;
 488 }
 489
 490 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 491                                struct btrfs_root *root)
 492 {
 493         int ret;
 494         u64 old_root_bytenr;
 495         struct btrfs_root *tree_root = root->fs_info->tree_root;
 496
 497         btrfs_write_dirty_block_groups(trans, root);
 498         while(1) {
 499                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 500                 if (old_root_bytenr == root->node->start)
 501                         break;
 502                 btrfs_set_root_bytenr(&root->root_item,
 503                                        root->node->start);
 504                 btrfs_set_root_generation(&root->root_item,
 505                                           trans->transid);
 506                 root->root_item.level = btrfs_header_level(root->node);
 507                 ret = btrfs_update_root(trans, tree_root,
 508                                         &root->root_key,
 509                                         &root->root_item);
 510                 BUG_ON(ret);
 511                 btrfs_write_dirty_block_groups(trans, root);
 512         }
 513         return 0;
 514 }
 515
 516 static int commit_tree_roots(struct btrfs_trans_handle *trans,
 517                              struct btrfs_fs_info *fs_info)
 518 {
 519         struct btrfs_root *root;
 520         struct list_head *next;
 521         struct extent_buffer *eb;
 522         int ret;
 523
 524         if (fs_info->readonly)
 525                 return 0;
 526
 527         eb = fs_info->tree_root->node;
 528         extent_buffer_get(eb);
 529         ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
 530         free_extent_buffer(eb);
 531         if (ret)
 532                 return ret;
 533
 534         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 535                 next = fs_info->dirty_cowonly_roots.next;
 536                 list_del_init(next);
 537                 root = list_entry(next, struct btrfs_root, dirty_list);
 538                 update_cowonly_root(trans, root);
 539                 free_extent_buffer(root->commit_root);
 540                 root->commit_root = NULL;
 541         }
 542
 543         return 0;
 544 }
 545
 546 static int __commit_transaction(struct btrfs_trans_handle *trans,
 547                                 struct btrfs_root *root)
 548 {
 549         u64 start;
 550         u64 end;
 551         struct extent_buffer *eb;
 552         struct extent_io_tree *tree = &root->fs_info->extent_cache;
 553         int ret;
 554
 555         while(1) {
 556                 ret = find_first_extent_bit(tree, 0, &start, &end,
 557                                             EXTENT_DIRTY);
 558                 if (ret)
 559                         break;
 560                 while(start <= end) {
 561                         eb = find_first_extent_buffer(tree, start);
 562                         BUG_ON(!eb || eb->start != start);
 563                         ret = write_tree_block(trans, root, eb);
 564                         BUG_ON(ret);
 565                         start += eb->len;
 566                         clear_extent_buffer_dirty(eb);
 567                         free_extent_buffer(eb);
 568                 }
 569         }
 570         return 0;
 571 }
 572
 573 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 574                              struct btrfs_root *root)
 575 {
 576         u64 transid = trans->transid;
 577         int ret = 0;
 578         struct btrfs_fs_info *fs_info = root->fs_info;
 579
 580         if (root->commit_root == root->node)
 581                 goto commit_tree;
 582         if (root == root->fs_info->tree_root)
 583                 goto commit_tree;
 584         if (root == root->fs_info->chunk_root)
 585                 goto commit_tree;
 586
 587         free_extent_buffer(root->commit_root);
 588         root->commit_root = NULL;
 589
 590         btrfs_set_root_bytenr(&root->root_item, root->node->start);
 591         btrfs_set_root_generation(&root->root_item, trans->transid);
 592         root->root_item.level = btrfs_header_level(root->node);
 593         ret = btrfs_update_root(trans, root->fs_info->tree_root,
 594                                 &root->root_key, &root->root_item);
 595         BUG_ON(ret);
 596 commit_tree:
 597         ret = commit_tree_roots(trans, fs_info);
 598         BUG_ON(ret);
 599         ret = __commit_transaction(trans, root);
 600         BUG_ON(ret);
 601         write_ctree_super(trans, root);
 602         btrfs_finish_extent_commit(trans, fs_info->extent_root,
 603                                    &fs_info->pinned_extents);
 604         kfree(trans);
 605         free_extent_buffer(root->commit_root);
 606         root->commit_root = NULL;
 607         fs_info->running_transaction = NULL;
 608         fs_info->last_trans_committed = transid;
 609         return 0;
 610 }
 611
 612 static int find_and_setup_root(struct btrfs_root *tree_root,
 613                                struct btrfs_fs_info *fs_info,
 614                                u64 objectid, struct btrfs_root *root)
 615 {
 616         int ret;
 617         u32 blocksize;
 618         u64 generation;
 619
 620         btrfs_setup_root(root, fs_info, objectid);
 621         ret = btrfs_find_last_root(tree_root, objectid,
 622                                    &root->root_item, &root->root_key);
 623         if (ret)
 624                 return ret;
 625
 626         blocksize = fs_info->nodesize;
 627         generation = btrfs_root_generation(&root->root_item);
 628         root->node = read_tree_block(fs_info,
 629                         btrfs_root_bytenr(&root->root_item),
 630                         blocksize, generation);
 631         if (!extent_buffer_uptodate(root->node))
 632                 return -EIO;
 633
 634         return 0;
 635 }
 636
 637 static int find_and_setup_log_root(struct btrfs_root *tree_root,
 638                                struct btrfs_fs_info *fs_info,
 639                                struct btrfs_super_block *disk_super)
 640 {
 641         u32 blocksize;
 642         u64 blocknr = btrfs_super_log_root(disk_super);
 643         struct btrfs_root *log_root = malloc(sizeof(struct btrfs_root));
 644
 645         if (!log_root)
 646                 return -ENOMEM;
 647
 648         if (blocknr == 0) {
 649                 free(log_root);
 650                 return 0;
 651         }
 652
 653         blocksize = fs_info->nodesize;
 654
 655         btrfs_setup_root(log_root, fs_info,
 656                          BTRFS_TREE_LOG_OBJECTID);
 657
 658         log_root->node = read_tree_block(fs_info, blocknr,
 659                                      blocksize,
 660                                      btrfs_super_generation(disk_super) + 1);
 661
 662         fs_info->log_root_tree = log_root;
 663
 664         if (!extent_buffer_uptodate(log_root->node)) {
 665                 free_extent_buffer(log_root->node);
 666                 free(log_root);
 667                 fs_info->log_root_tree = NULL;
 668                 return -EIO;
 669         }
 670
 671         return 0;
 672 }
 673
 674 int btrfs_free_fs_root(struct btrfs_root *root)
 675 {
 676         if (root->node)
 677                 free_extent_buffer(root->node);
 678         if (root->commit_root)
 679                 free_extent_buffer(root->commit_root);
 680         kfree(root);
 681         return 0;
 682 }
 683
 684 static void __free_fs_root(struct rb_node *node)
 685 {
 686         struct btrfs_root *root;
 687
 688         root = container_of(node, struct btrfs_root, rb_node);
 689         btrfs_free_fs_root(root);
 690 }
 691
 692 FREE_RB_BASED_TREE(fs_roots, __free_fs_root);
 693
 694 struct btrfs_root *btrfs_read_fs_root_no_cache(struct btrfs_fs_info *fs_info,
 695                                                struct btrfs_key *location)
 696 {
 697         struct btrfs_root *root;
 698         struct btrfs_root *tree_root = fs_info->tree_root;
 699         struct btrfs_path *path;
 700         struct extent_buffer *l;
 701         u64 generation;
 702         u32 blocksize;
 703         int ret = 0;
 704
 705         root = calloc(1, sizeof(*root));
 706         if (!root)
 707                 return ERR_PTR(-ENOMEM);
 708         if (location->offset == (u64)-1) {
 709                 ret = find_and_setup_root(tree_root, fs_info,
 710                                           location->objectid, root);
 711                 if (ret) {
 712                         free(root);
 713                         return ERR_PTR(ret);
 714                 }
 715                 goto insert;
 716         }
 717
 718         btrfs_setup_root(root, fs_info,
 719                          location->objectid);
 720
 721         path = btrfs_alloc_path();
 722         if (!path) {
 723                 free(root);
 724                 return ERR_PTR(-ENOMEM);
 725         }
 726
 727         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 728         if (ret != 0) {
 729                 if (ret > 0)
 730                         ret = -ENOENT;
 731                 goto out;
 732         }
 733         l = path->nodes[0];
 734         read_extent_buffer(l, &root->root_item,
 735                btrfs_item_ptr_offset(l, path->slots[0]),
 736                sizeof(root->root_item));
 737         memcpy(&root->root_key, location, sizeof(*location));
 738         ret = 0;
 739 out:
 740         btrfs_free_path(path);
 741         if (ret) {
 742                 free(root);
 743                 return ERR_PTR(ret);
 744         }
 745         generation = btrfs_root_generation(&root->root_item);
 746         blocksize = fs_info->nodesize;
 747         root->node = read_tree_block(fs_info,
 748                         btrfs_root_bytenr(&root->root_item),
 749                         blocksize, generation);
 750         if (!extent_buffer_uptodate(root->node)) {
 751                 free(root);
 752                 return ERR_PTR(-EIO);
 753         }
 754 insert:
 755         root->ref_cows = 1;
 756         return root;
 757 }
 758
 759 static int btrfs_fs_roots_compare_objectids(struct rb_node *node,
 760                                             void *data)
 761 {
 762         u64 objectid = *((u64 *)data);
 763         struct btrfs_root *root;
 764
 765         root = rb_entry(node, struct btrfs_root, rb_node);
 766         if (objectid > root->objectid)
 767                 return 1;
 768         else if (objectid < root->objectid)
 769                 return -1;
 770         else
 771                 return 0;
 772 }
 773
 774 static int btrfs_fs_roots_compare_roots(struct rb_node *node1,
 775                                         struct rb_node *node2)
 776 {
 777         struct btrfs_root *root;
 778
 779         root = rb_entry(node2, struct btrfs_root, rb_node);
 780         return btrfs_fs_roots_compare_objectids(node1, (void *)&root->objectid);
 781 }
 782
 783 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 784                                       struct btrfs_key *location)
 785 {
 786         struct btrfs_root *root;
 787         struct rb_node *node;
 788         int ret;
 789         u64 objectid = location->objectid;
 790
 791         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 792                 return fs_info->tree_root;
 793         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 794                 return fs_info->extent_root;
 795         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 796                 return fs_info->chunk_root;
 797         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 798                 return fs_info->dev_root;
 799         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
 800                 return fs_info->csum_root;
 801         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
 802                 return fs_info->quota_enabled ? fs_info->quota_root :
 803                                 ERR_PTR(-ENOENT);
 804
 805         BUG_ON(location->objectid == BTRFS_TREE_RELOC_OBJECTID ||
 806                location->offset != (u64)-1);
 807
 808         node = rb_search(&fs_info->fs_root_tree, (void *)&objectid,
 809                          btrfs_fs_roots_compare_objectids, NULL);
 810         if (node)
 811                 return container_of(node, struct btrfs_root, rb_node);
 812
 813         root = btrfs_read_fs_root_no_cache(fs_info, location);
 814         if (IS_ERR(root))
 815                 return root;
 816
 817         ret = rb_insert(&fs_info->fs_root_tree, &root->rb_node,
 818                         btrfs_fs_roots_compare_roots);
 819         BUG_ON(ret);
 820         return root;
 821 }
 822
 823 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
 824 {
 825         if (fs_info->quota_root)
 826                 free(fs_info->quota_root);
 827
 828         free(fs_info->tree_root);
 829         free(fs_info->extent_root);
 830         free(fs_info->chunk_root);
 831         free(fs_info->dev_root);
 832         free(fs_info->csum_root);
 833         free(fs_info->free_space_root);
 834         free(fs_info->super_copy);
 835         free(fs_info->log_root_tree);
 836         free(fs_info);
 837 }
 838
 839 struct btrfs_fs_info *btrfs_new_fs_info(int writable, u64 sb_bytenr)
 840 {
 841         struct btrfs_fs_info *fs_info;
 842
 843         fs_info = calloc(1, sizeof(struct btrfs_fs_info));
 844         if (!fs_info)
 845                 return NULL;
 846
 847         fs_info->tree_root = calloc(1, sizeof(struct btrfs_root));
 848         fs_info->extent_root = calloc(1, sizeof(struct btrfs_root));
 849         fs_info->chunk_root = calloc(1, sizeof(struct btrfs_root));
 850         fs_info->dev_root = calloc(1, sizeof(struct btrfs_root));
 851         fs_info->csum_root = calloc(1, sizeof(struct btrfs_root));
 852         fs_info->quota_root = calloc(1, sizeof(struct btrfs_root));
 853         fs_info->free_space_root = calloc(1, sizeof(struct btrfs_root));
 854         fs_info->super_copy = calloc(1, BTRFS_SUPER_INFO_SIZE);
 855
 856         if (!fs_info->tree_root || !fs_info->extent_root ||
 857             !fs_info->chunk_root || !fs_info->dev_root ||
 858             !fs_info->csum_root || !fs_info->quota_root ||
 859             !fs_info->free_space_root || !fs_info->super_copy)
 860                 goto free_all;
 861
 862         extent_io_tree_init(&fs_info->extent_cache);
 863         extent_io_tree_init(&fs_info->free_space_cache);
 864         extent_io_tree_init(&fs_info->block_group_cache);
 865         extent_io_tree_init(&fs_info->pinned_extents);
 866         extent_io_tree_init(&fs_info->pending_del);
 867         extent_io_tree_init(&fs_info->extent_ins);
 868         fs_info->excluded_extents = NULL;
 869
 870         fs_info->fs_root_tree = RB_ROOT;
 871         cache_tree_init(&fs_info->mapping_tree.cache_tree);
 872
 873         mutex_init(&fs_info->fs_mutex);
 874         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
 875         INIT_LIST_HEAD(&fs_info->space_info);
 876         INIT_LIST_HEAD(&fs_info->recow_ebs);
 877
 878         if (!writable)
 879                 fs_info->readonly = 1;
 880
 881         fs_info->super_bytenr = sb_bytenr;
 882         fs_info->data_alloc_profile = (u64)-1;
 883         fs_info->metadata_alloc_profile = (u64)-1;
 884         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
 885         return fs_info;
 886 free_all:
 887         btrfs_free_fs_info(fs_info);
 888         return NULL;
 889 }
 890
 891 int btrfs_check_fs_compatibility(struct btrfs_super_block *sb,
 892                                  unsigned int flags)
 893 {
 894         u64 features;
 895
 896         features = btrfs_super_incompat_flags(sb) &
 897                    ~BTRFS_FEATURE_INCOMPAT_SUPP;
 898         if (features) {
 899                 printk("couldn't open because of unsupported "
 900                        "option features (%Lx).\n",
 901                        (unsigned long long)features);
 902                 return -ENOTSUP;
 903         }
 904
 905         features = btrfs_super_incompat_flags(sb);
 906         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
 907                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
 908                 btrfs_set_super_incompat_flags(sb, features);
 909         }
 910
 911         features = btrfs_super_compat_ro_flags(sb);
 912         if (flags & OPEN_CTREE_WRITES) {
 913                 if (flags & OPEN_CTREE_INVALIDATE_FST) {
 914                         /* Clear the FREE_SPACE_TREE_VALID bit on disk... */
 915                         features &= ~BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID;
 916                         btrfs_set_super_compat_ro_flags(sb, features);
 917                         /* ... and ignore the free space tree bit. */
 918                         features &= ~BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE;
 919                 }
 920                 if (features & ~BTRFS_FEATURE_COMPAT_RO_SUPP) {
 921                         printk("couldn't open RDWR because of unsupported "
 922                                "option features (%Lx).\n",
 923                                (unsigned long long)features);
 924                         return -ENOTSUP;
 925                 }
 926
 927         }
 928         return 0;
 929 }
 930
 931 static int find_best_backup_root(struct btrfs_super_block *super)
 932 {
 933         struct btrfs_root_backup *backup;
 934         u64 orig_gen = btrfs_super_generation(super);
 935         u64 gen = 0;
 936         int best_index = 0;
 937         int i;
 938
 939         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
 940                 backup = super->super_roots + i;
 941                 if (btrfs_backup_tree_root_gen(backup) != orig_gen &&
 942                     btrfs_backup_tree_root_gen(backup) > gen) {
 943                         best_index = i;
 944                         gen = btrfs_backup_tree_root_gen(backup);
 945                 }
 946         }
 947         return best_index;
 948 }
 949
 950 static int setup_root_or_create_block(struct btrfs_fs_info *fs_info,
 951                                       unsigned flags,
 952                                       struct btrfs_root *info_root,
 953                                       u64 objectid, char *str)
 954 {
 955         struct btrfs_super_block *sb = fs_info->super_copy;
 956         struct btrfs_root *root = fs_info->tree_root;
 957         u32 nodesize = btrfs_super_nodesize(sb);
 958         int ret;
 959
 960         ret = find_and_setup_root(root, fs_info, objectid, info_root);
 961         if (ret) {
 962                 printk("Couldn't setup %s tree\n", str);
 963                 if (!(flags & OPEN_CTREE_PARTIAL))
 964                         return -EIO;
 965                 /*
 966                  * Need a blank node here just so we don't screw up in the
 967                  * million of places that assume a root has a valid ->node
 968                  */
 969                 info_root->node =
 970                         btrfs_find_create_tree_block(fs_info, 0, nodesize);
 971                 if (!info_root->node)
 972                         return -ENOMEM;
 973                 clear_extent_buffer_uptodate(info_root->node);
 974         }
 975
 976         return 0;
 977 }
 978
 979 int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr,
 980                           unsigned flags)
 981 {
 982         struct btrfs_super_block *sb = fs_info->super_copy;
 983         struct btrfs_root *root;
 984         struct btrfs_key key;
 985         u64 generation;
 986         u32 blocksize;
 987         int ret;
 988
 989         root = fs_info->tree_root;
 990         btrfs_setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
 991         blocksize = fs_info->nodesize;
 992         generation = btrfs_super_generation(sb);
 993
 994         if (!root_tree_bytenr && !(flags & OPEN_CTREE_BACKUP_ROOT)) {
 995                 root_tree_bytenr = btrfs_super_root(sb);
 996         } else if (flags & OPEN_CTREE_BACKUP_ROOT) {
 997                 struct btrfs_root_backup *backup;
 998                 int index = find_best_backup_root(sb);
 999                 if (index >= BTRFS_NUM_BACKUP_ROOTS) {
1000                         fprintf(stderr, "Invalid backup root number\n");
1001                         return -EIO;
1002                 }
1003                 backup = fs_info->super_copy->super_roots + index;
1004                 root_tree_bytenr = btrfs_backup_tree_root(backup);
1005                 generation = btrfs_backup_tree_root_gen(backup);
1006         }
1007
1008         root->node = read_tree_block(fs_info, root_tree_bytenr, blocksize,
1009                                      generation);
1010         if (!extent_buffer_uptodate(root->node)) {
1011                 fprintf(stderr, "Couldn't read tree root\n");
1012                 return -EIO;
1013         }
1014
1015         ret = setup_root_or_create_block(fs_info, flags, fs_info->extent_root,
1016                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
1017         if (ret)
1018                 return ret;
1019         fs_info->extent_root->track_dirty = 1;
1020
1021         ret = find_and_setup_root(root, fs_info, BTRFS_DEV_TREE_OBJECTID,
1022                                   fs_info->dev_root);
1023         if (ret) {
1024                 printk("Couldn't setup device tree\n");
1025                 return -EIO;
1026         }
1027         fs_info->dev_root->track_dirty = 1;
1028
1029         ret = setup_root_or_create_block(fs_info, flags, fs_info->csum_root,
1030                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
1031         if (ret)
1032                 return ret;
1033         fs_info->csum_root->track_dirty = 1;
1034
1035         ret = find_and_setup_root(root, fs_info, BTRFS_QUOTA_TREE_OBJECTID,
1036                                   fs_info->quota_root);
1037         if (ret) {
1038                 free(fs_info->quota_root);
1039                 fs_info->quota_root = NULL;
1040         } else {
1041                 fs_info->quota_enabled = 1;
1042         }
1043
1044         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
1045                 ret = find_and_setup_root(root, fs_info, BTRFS_FREE_SPACE_TREE_OBJECTID,
1046                                           fs_info->free_space_root);
1047                 if (ret) {
1048                         printk("Couldn't read free space tree\n");
1049                         return -EIO;
1050                 }
1051                 fs_info->free_space_root->track_dirty = 1;
1052         }
1053
1054         ret = find_and_setup_log_root(root, fs_info, sb);
1055         if (ret) {
1056                 printk("Couldn't setup log root tree\n");
1057                 if (!(flags & OPEN_CTREE_PARTIAL))
1058                         return -EIO;
1059         }
1060
1061         fs_info->generation = generation;
1062         fs_info->last_trans_committed = generation;
1063         if (extent_buffer_uptodate(fs_info->extent_root->node) &&
1064             !(flags & OPEN_CTREE_NO_BLOCK_GROUPS))
1065                 btrfs_read_block_groups(fs_info->tree_root);
1066
1067         key.objectid = BTRFS_FS_TREE_OBJECTID;
1068         key.type = BTRFS_ROOT_ITEM_KEY;
1069         key.offset = (u64)-1;
1070         fs_info->fs_root = btrfs_read_fs_root(fs_info, &key);
1071
1072         if (IS_ERR(fs_info->fs_root))
1073                 return -EIO;
1074         return 0;
1075 }
1076
1077 void btrfs_release_all_roots(struct btrfs_fs_info *fs_info)
1078 {
1079         if (fs_info->free_space_root)
1080                 free_extent_buffer(fs_info->free_space_root->node);
1081         if (fs_info->quota_root)
1082                 free_extent_buffer(fs_info->quota_root->node);
1083         if (fs_info->csum_root)
1084                 free_extent_buffer(fs_info->csum_root->node);
1085         if (fs_info->dev_root)
1086                 free_extent_buffer(fs_info->dev_root->node);
1087         if (fs_info->extent_root)
1088                 free_extent_buffer(fs_info->extent_root->node);
1089         if (fs_info->tree_root)
1090                 free_extent_buffer(fs_info->tree_root->node);
1091         if (fs_info->log_root_tree)
1092                 free_extent_buffer(fs_info->log_root_tree->node);
1093         if (fs_info->chunk_root)
1094                 free_extent_buffer(fs_info->chunk_root->node);
1095 }
1096
1097 static void free_map_lookup(struct cache_extent *ce)
1098 {
1099         struct map_lookup *map;
1100
1101         map = container_of(ce, struct map_lookup, ce);
1102         kfree(map);
1103 }
1104
1105 FREE_EXTENT_CACHE_BASED_TREE(mapping_cache, free_map_lookup);
1106
1107 void btrfs_cleanup_all_caches(struct btrfs_fs_info *fs_info)
1108 {
1109         while (!list_empty(&fs_info->recow_ebs)) {
1110                 struct extent_buffer *eb;
1111                 eb = list_first_entry(&fs_info->recow_ebs,
1112                                       struct extent_buffer, recow);
1113                 list_del_init(&eb->recow);
1114                 free_extent_buffer(eb);
1115         }
1116         free_mapping_cache_tree(&fs_info->mapping_tree.cache_tree);
1117         extent_io_tree_cleanup(&fs_info->extent_cache);
1118         extent_io_tree_cleanup(&fs_info->free_space_cache);
1119         extent_io_tree_cleanup(&fs_info->block_group_cache);
1120         extent_io_tree_cleanup(&fs_info->pinned_extents);
1121         extent_io_tree_cleanup(&fs_info->pending_del);
1122         extent_io_tree_cleanup(&fs_info->extent_ins);
1123 }
1124
1125 int btrfs_scan_fs_devices(int fd, const char *path,
1126                           struct btrfs_fs_devices **fs_devices,
1127                           u64 sb_bytenr, unsigned sbflags,
1128                           int skip_devices)
1129 {
1130         u64 total_devs;
1131         u64 dev_size;
1132         off_t seek_ret;
1133         int ret;
1134         if (!sb_bytenr)
1135                 sb_bytenr = BTRFS_SUPER_INFO_OFFSET;
1136
1137         seek_ret = lseek(fd, 0, SEEK_END);
1138         if (seek_ret < 0)
1139                 return -errno;
1140
1141         dev_size = seek_ret;
1142         lseek(fd, 0, SEEK_SET);
1143         if (sb_bytenr > dev_size) {
1144                 error("superblock bytenr %llu is larger than device size %llu",
1145                                 (unsigned long long)sb_bytenr,
1146                                 (unsigned long long)dev_size);
1147                 return -EINVAL;
1148         }
1149
1150         ret = btrfs_scan_one_device(fd, path, fs_devices,
1151                                     &total_devs, sb_bytenr, sbflags);
1152         if (ret) {
1153                 fprintf(stderr, "No valid Btrfs found on %s\n", path);
1154                 return ret;
1155         }
1156
1157         if (!skip_devices && total_devs != 1) {
1158                 ret = btrfs_scan_devices();
1159                 if (ret)
1160                         return ret;
1161         }
1162         return 0;
1163 }
1164
1165 int btrfs_setup_chunk_tree_and_device_map(struct btrfs_fs_info *fs_info,
1166                                           u64 chunk_root_bytenr)
1167 {
1168         struct btrfs_super_block *sb = fs_info->super_copy;
1169         u64 generation;
1170         int ret;
1171
1172         btrfs_setup_root(fs_info->chunk_root, fs_info,
1173                         BTRFS_CHUNK_TREE_OBJECTID);
1174
1175         ret = btrfs_read_sys_array(fs_info->chunk_root);
1176         if (ret)
1177                 return ret;
1178
1179         generation = btrfs_super_chunk_root_generation(sb);
1180
1181         if (chunk_root_bytenr && !IS_ALIGNED(chunk_root_bytenr,
1182                                             fs_info->sectorsize)) {
1183                 warning("chunk_root_bytenr %llu is unaligned to %u, ignore it",
1184                         chunk_root_bytenr, fs_info->sectorsize);
1185                 chunk_root_bytenr = 0;
1186         }
1187
1188         if (!chunk_root_bytenr)
1189                 chunk_root_bytenr = btrfs_super_chunk_root(sb);
1190         else
1191                 generation = 0;
1192
1193         fs_info->chunk_root->node = read_tree_block(fs_info,
1194                                                     chunk_root_bytenr,
1195                                                     fs_info->nodesize,
1196                                                     generation);
1197         if (!extent_buffer_uptodate(fs_info->chunk_root->node)) {
1198                 if (fs_info->ignore_chunk_tree_error) {
1199                         warning("cannot read chunk root, continue anyway");
1200                         fs_info->chunk_root = NULL;
1201                         return 0;
1202                 } else {
1203                         error("cannot read chunk root");
1204                         return -EIO;
1205                 }
1206         }
1207
1208         if (!(btrfs_super_flags(sb) & BTRFS_SUPER_FLAG_METADUMP)) {
1209                 ret = btrfs_read_chunk_tree(fs_info->chunk_root);
1210                 if (ret) {
1211                         fprintf(stderr, "Couldn't read chunk tree\n");
1212                         return ret;
1213                 }
1214         }
1215         return 0;
1216 }
1217
1218 static struct btrfs_fs_info *__open_ctree_fd(int fp, const char *path,
1219                                              u64 sb_bytenr,
1220                                              u64 root_tree_bytenr,
1221                                              u64 chunk_root_bytenr,
1222                                              unsigned flags)
1223 {
1224         struct btrfs_fs_info *fs_info;
1225         struct btrfs_super_block *disk_super;
1226         struct btrfs_fs_devices *fs_devices = NULL;
1227         struct extent_buffer *eb;
1228         int ret;
1229         int oflags;
1230         unsigned sbflags = SBREAD_DEFAULT;
1231
1232         if (sb_bytenr == 0)
1233                 sb_bytenr = BTRFS_SUPER_INFO_OFFSET;
1234
1235         /* try to drop all the caches */
1236         if (posix_fadvise(fp, 0, 0, POSIX_FADV_DONTNEED))
1237                 fprintf(stderr, "Warning, could not drop caches\n");
1238
1239         fs_info = btrfs_new_fs_info(flags & OPEN_CTREE_WRITES, sb_bytenr);
1240         if (!fs_info) {
1241                 fprintf(stderr, "Failed to allocate memory for fs_info\n");
1242                 return NULL;
1243         }
1244         if (flags & OPEN_CTREE_RESTORE)
1245                 fs_info->on_restoring = 1;
1246         if (flags & OPEN_CTREE_SUPPRESS_CHECK_BLOCK_ERRORS)
1247                 fs_info->suppress_check_block_errors = 1;
1248         if (flags & OPEN_CTREE_IGNORE_FSID_MISMATCH)
1249                 fs_info->ignore_fsid_mismatch = 1;
1250         if (flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR)
1251                 fs_info->ignore_chunk_tree_error = 1;
1252
1253         if ((flags & OPEN_CTREE_RECOVER_SUPER)
1254              && (flags & OPEN_CTREE_FS_PARTIAL)) {
1255                 fprintf(stderr,
1256                     "cannot open a partially created filesystem for recovery");
1257                 goto out;
1258         }
1259
1260         if (flags & OPEN_CTREE_FS_PARTIAL)
1261                 sbflags = SBREAD_PARTIAL;
1262
1263         ret = btrfs_scan_fs_devices(fp, path, &fs_devices, sb_bytenr, sbflags,
1264                         (flags & OPEN_CTREE_NO_DEVICES));
1265         if (ret)
1266                 goto out;
1267
1268         fs_info->fs_devices = fs_devices;
1269         if (flags & OPEN_CTREE_WRITES)
1270                 oflags = O_RDWR;
1271         else
1272                 oflags = O_RDONLY;
1273
1274         if (flags & OPEN_CTREE_EXCLUSIVE)
1275                 oflags |= O_EXCL;
1276
1277         ret = btrfs_open_devices(fs_devices, oflags);
1278         if (ret)
1279                 goto out;
1280
1281         disk_super = fs_info->super_copy;
1282         if (flags & OPEN_CTREE_RECOVER_SUPER)
1283                 ret = btrfs_read_dev_super(fs_devices->latest_bdev, disk_super,
1284                                 sb_bytenr, SBREAD_RECOVER);
1285         else
1286                 ret = btrfs_read_dev_super(fp, disk_super, sb_bytenr,
1287                                 sbflags);
1288         if (ret) {
1289                 printk("No valid btrfs found\n");
1290                 goto out_devices;
1291         }
1292
1293         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID &&
1294             !fs_info->ignore_fsid_mismatch) {
1295                 fprintf(stderr, "ERROR: Filesystem UUID change in progress\n");
1296                 goto out_devices;
1297         }
1298
1299         memcpy(fs_info->fsid, &disk_super->fsid, BTRFS_FSID_SIZE);
1300         fs_info->sectorsize = btrfs_super_sectorsize(disk_super);
1301         fs_info->nodesize = btrfs_super_nodesize(disk_super);
1302         fs_info->stripesize = btrfs_super_stripesize(disk_super);
1303
1304         ret = btrfs_check_fs_compatibility(fs_info->super_copy, flags);
1305         if (ret)
1306                 goto out_devices;
1307
1308         ret = btrfs_setup_chunk_tree_and_device_map(fs_info, chunk_root_bytenr);
1309         if (ret)
1310                 goto out_chunk;
1311
1312         /* Chunk tree root is unable to read, return directly */
1313         if (!fs_info->chunk_root)
1314                 return fs_info;
1315
1316         eb = fs_info->chunk_root->node;
1317         read_extent_buffer(eb, fs_info->chunk_tree_uuid,
1318                            btrfs_header_chunk_tree_uuid(eb),
1319                            BTRFS_UUID_SIZE);
1320
1321         ret = btrfs_setup_all_roots(fs_info, root_tree_bytenr, flags);
1322         if (ret && !(flags & __OPEN_CTREE_RETURN_CHUNK_ROOT) &&
1323             !fs_info->ignore_chunk_tree_error)
1324                 goto out_chunk;
1325
1326         return fs_info;
1327
1328 out_chunk:
1329         btrfs_release_all_roots(fs_info);
1330         btrfs_cleanup_all_caches(fs_info);
1331 out_devices:
1332         btrfs_close_devices(fs_devices);
1333 out:
1334         btrfs_free_fs_info(fs_info);
1335         return NULL;
1336 }
1337
1338 struct btrfs_fs_info *open_ctree_fs_info(const char *filename,
1339                                          u64 sb_bytenr, u64 root_tree_bytenr,
1340                                          u64 chunk_root_bytenr,
1341                                          unsigned flags)
1342 {
1343         int fp;
1344         int ret;
1345         struct btrfs_fs_info *info;
1346         int oflags = O_RDWR;
1347         struct stat st;
1348
1349         ret = stat(filename, &st);
1350         if (ret < 0) {
1351                 error("cannot stat '%s': %s", filename, strerror(errno));
1352                 return NULL;
1353         }
1354         if (!(((st.st_mode & S_IFMT) == S_IFREG) || ((st.st_mode & S_IFMT) == S_IFBLK))) {
1355                 error("not a regular file or block device: %s", filename);
1356                 return NULL;
1357         }
1358
1359         if (!(flags & OPEN_CTREE_WRITES))
1360                 oflags = O_RDONLY;
1361
1362         fp = open(filename, oflags);
1363         if (fp < 0) {
1364                 error("cannot open '%s': %s", filename, strerror(errno));
1365                 return NULL;
1366         }
1367         info = __open_ctree_fd(fp, filename, sb_bytenr, root_tree_bytenr,
1368                                chunk_root_bytenr, flags);
1369         close(fp);
1370         return info;
1371 }
1372
1373 struct btrfs_root *open_ctree(const char *filename, u64 sb_bytenr,
1374                               unsigned flags)
1375 {
1376         struct btrfs_fs_info *info;
1377
1378         /* This flags may not return fs_info with any valid root */
1379         BUG_ON(flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR);
1380         info = open_ctree_fs_info(filename, sb_bytenr, 0, 0, flags);
1381         if (!info)
1382                 return NULL;
1383         if (flags & __OPEN_CTREE_RETURN_CHUNK_ROOT)
1384                 return info->chunk_root;
1385         return info->fs_root;
1386 }
1387
1388 struct btrfs_root *open_ctree_fd(int fp, const char *path, u64 sb_bytenr,
1389                                  unsigned flags)
1390 {
1391         struct btrfs_fs_info *info;
1392
1393         /* This flags may not return fs_info with any valid root */
1394         if (flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR) {
1395                 error("invalid open_ctree flags: 0x%llx",
1396                                 (unsigned long long)flags);
1397                 return NULL;
1398         }
1399         info = __open_ctree_fd(fp, path, sb_bytenr, 0, 0, flags);
1400         if (!info)
1401                 return NULL;
1402         if (flags & __OPEN_CTREE_RETURN_CHUNK_ROOT)
1403                 return info->chunk_root;
1404         return info->fs_root;
1405 }
1406
1407 /*
1408  * Check if the super is valid:
1409  * - nodesize/sectorsize - minimum, maximum, alignment
1410  * - tree block starts   - alignment
1411  * - number of devices   - something sane
1412  * - sys array size      - maximum
1413  */
1414 static int check_super(struct btrfs_super_block *sb, unsigned sbflags)
1415 {
1416         u8 result[BTRFS_CSUM_SIZE];
1417         u32 crc;
1418         u16 csum_type;
1419         int csum_size;
1420
1421         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
1422                 if (btrfs_super_magic(sb) == BTRFS_MAGIC_PARTIAL) {
1423                         if (!(sbflags & SBREAD_PARTIAL)) {
1424                                 error("superblock magic doesn't match");
1425                                 return -EIO;
1426                         }
1427                 }
1428         }
1429
1430         csum_type = btrfs_super_csum_type(sb);
1431         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
1432                 error("unsupported checksum algorithm %u", csum_type);
1433                 return -EIO;
1434         }
1435         csum_size = btrfs_csum_sizes[csum_type];
1436
1437         crc = ~(u32)0;
1438         crc = btrfs_csum_data((char *)sb + BTRFS_CSUM_SIZE, crc,
1439                               BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1440         btrfs_csum_final(crc, result);
1441
1442         if (memcmp(result, sb->csum, csum_size)) {
1443                 error("superblock checksum mismatch");
1444                 return -EIO;
1445         }
1446         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
1447                 error("tree_root level too big: %d >= %d",
1448                         btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
1449                 goto error_out;
1450         }
1451         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
1452                 error("chunk_root level too big: %d >= %d",
1453                         btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
1454                 goto error_out;
1455         }
1456         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
1457                 error("log_root level too big: %d >= %d",
1458                         btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
1459                 goto error_out;
1460         }
1461
1462         if (!IS_ALIGNED(btrfs_super_root(sb), 4096)) {
1463                 error("tree_root block unaligned: %llu", btrfs_super_root(sb));
1464                 goto error_out;
1465         }
1466         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096)) {
1467                 error("chunk_root block unaligned: %llu",
1468                         btrfs_super_chunk_root(sb));
1469                 goto error_out;
1470         }
1471         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096)) {
1472                 error("log_root block unaligned: %llu",
1473                         btrfs_super_log_root(sb));
1474                 goto error_out;
1475         }
1476         if (btrfs_super_nodesize(sb) < 4096) {
1477                 error("nodesize too small: %u < 4096",
1478                         btrfs_super_nodesize(sb));
1479                 goto error_out;
1480         }
1481         if (!IS_ALIGNED(btrfs_super_nodesize(sb), 4096)) {
1482                 error("nodesize unaligned: %u", btrfs_super_nodesize(sb));
1483                 goto error_out;
1484         }
1485         if (btrfs_super_sectorsize(sb) < 4096) {
1486                 error("sectorsize too small: %u < 4096",
1487                         btrfs_super_sectorsize(sb));
1488                 goto error_out;
1489         }
1490         if (!IS_ALIGNED(btrfs_super_sectorsize(sb), 4096)) {
1491                 error("sectorsize unaligned: %u", btrfs_super_sectorsize(sb));
1492                 goto error_out;
1493         }
1494         if (btrfs_super_total_bytes(sb) == 0) {
1495                 error("invalid total_bytes 0");
1496                 goto error_out;
1497         }
1498         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
1499                 error("invalid bytes_used %llu", btrfs_super_bytes_used(sb));
1500                 goto error_out;
1501         }
1502         if ((btrfs_super_stripesize(sb) != 4096)
1503                 && (btrfs_super_stripesize(sb) != btrfs_super_sectorsize(sb))) {
1504                 error("invalid stripesize %u", btrfs_super_stripesize(sb));
1505                 goto error_out;
1506         }
1507
1508         if (memcmp(sb->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
1509                 char fsid[BTRFS_UUID_UNPARSED_SIZE];
1510                 char dev_fsid[BTRFS_UUID_UNPARSED_SIZE];
1511
1512                 uuid_unparse(sb->fsid, fsid);
1513                 uuid_unparse(sb->dev_item.fsid, dev_fsid);
1514                 error("dev_item UUID does not match fsid: %s != %s",
1515                         dev_fsid, fsid);
1516                 goto error_out;
1517         }
1518
1519         /*
1520          * Hint to catch really bogus numbers, bitflips or so
1521          */
1522         if (btrfs_super_num_devices(sb) > (1UL << 31)) {
1523                 warning("suspicious number of devices: %llu",
1524                         btrfs_super_num_devices(sb));
1525         }
1526
1527         if (btrfs_super_num_devices(sb) == 0) {
1528                 error("number of devices is 0");
1529                 goto error_out;
1530         }
1531
1532         /*
1533          * Obvious sys_chunk_array corruptions, it must hold at least one key
1534          * and one chunk
1535          */
1536         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
1537                 error("system chunk array too big %u > %u",
1538                       btrfs_super_sys_array_size(sb),
1539                       BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
1540                 goto error_out;
1541         }
1542         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
1543                         + sizeof(struct btrfs_chunk)) {
1544                 error("system chunk array too small %u < %zu",
1545                       btrfs_super_sys_array_size(sb),
1546                       sizeof(struct btrfs_disk_key) +
1547                       sizeof(struct btrfs_chunk));
1548                 goto error_out;
1549         }
1550
1551         return 0;
1552
1553 error_out:
1554         error("superblock checksum matches but it has invalid members");
1555         return -EIO;
1556 }
1557
1558 int btrfs_read_dev_super(int fd, struct btrfs_super_block *sb, u64 sb_bytenr,
1559                          unsigned sbflags)
1560 {
1561         u8 fsid[BTRFS_FSID_SIZE];
1562         int fsid_is_initialized = 0;
1563         char tmp[BTRFS_SUPER_INFO_SIZE];
1564         struct btrfs_super_block *buf = (struct btrfs_super_block *)tmp;
1565         int i;
1566         int ret;
1567         int max_super = sbflags & SBREAD_RECOVER ? BTRFS_SUPER_MIRROR_MAX : 1;
1568         u64 transid = 0;
1569         u64 bytenr;
1570
1571         if (sb_bytenr != BTRFS_SUPER_INFO_OFFSET) {
1572                 ret = pread64(fd, buf, BTRFS_SUPER_INFO_SIZE, sb_bytenr);
1573                 /* real error */
1574                 if (ret < 0)
1575                         return -errno;
1576
1577                 /* Not large enough sb, return -ENOENT instead of normal -EIO */
1578                 if (ret < BTRFS_SUPER_INFO_SIZE)
1579                         return -ENOENT;
1580
1581                 if (btrfs_super_bytenr(buf) != sb_bytenr)
1582                         return -EIO;
1583
1584                 ret = check_super(buf, sbflags);
1585                 if (ret < 0)
1586                         return ret;
1587                 memcpy(sb, buf, BTRFS_SUPER_INFO_SIZE);
1588                 return 0;
1589         }
1590
1591         /*
1592         * we would like to check all the supers, but that would make
1593         * a btrfs mount succeed after a mkfs from a different FS.
1594         * So, we need to add a special mount option to scan for
1595         * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1596         */
1597
1598         for (i = 0; i < max_super; i++) {
1599                 bytenr = btrfs_sb_offset(i);
1600                 ret = pread64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1601                 if (ret < BTRFS_SUPER_INFO_SIZE)
1602                         break;
1603
1604                 if (btrfs_super_bytenr(buf) != bytenr )
1605                         continue;
1606                 /* if magic is NULL, the device was removed */
1607                 if (btrfs_super_magic(buf) == 0 && i == 0)
1608                         break;
1609                 if (check_super(buf, sbflags))
1610                         continue;
1611
1612                 if (!fsid_is_initialized) {
1613                         memcpy(fsid, buf->fsid, sizeof(fsid));
1614                         fsid_is_initialized = 1;
1615                 } else if (memcmp(fsid, buf->fsid, sizeof(fsid))) {
1616                         /*
1617                          * the superblocks (the original one and
1618                          * its backups) contain data of different
1619                          * filesystems -> the super cannot be trusted
1620                          */
1621                         continue;
1622                 }
1623
1624                 if (btrfs_super_generation(buf) > transid) {
1625                         memcpy(sb, buf, BTRFS_SUPER_INFO_SIZE);
1626                         transid = btrfs_super_generation(buf);
1627                 }
1628         }
1629
1630         return transid > 0 ? 0 : -1;
1631 }
1632
1633 static int write_dev_supers(struct btrfs_root *root,
1634                             struct btrfs_super_block *sb,
1635                             struct btrfs_device *device)
1636 {
1637         u64 bytenr;
1638         u32 crc;
1639         int i, ret;
1640
1641         if (root->fs_info->super_bytenr != BTRFS_SUPER_INFO_OFFSET) {
1642                 btrfs_set_super_bytenr(sb, root->fs_info->super_bytenr);
1643                 crc = ~(u32)0;
1644                 crc = btrfs_csum_data((char *)sb + BTRFS_CSUM_SIZE, crc,
1645                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1646                 btrfs_csum_final(crc, &sb->csum[0]);
1647
1648                 /*
1649                  * super_copy is BTRFS_SUPER_INFO_SIZE bytes and is
1650                  * zero filled, we can use it directly
1651                  */
1652                 ret = pwrite64(device->fd, root->fs_info->super_copy,
1653                                 BTRFS_SUPER_INFO_SIZE,
1654                                 root->fs_info->super_bytenr);
1655                 if (ret != BTRFS_SUPER_INFO_SIZE)
1656                         goto write_err;
1657                 return 0;
1658         }
1659
1660         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1661                 bytenr = btrfs_sb_offset(i);
1662                 if (bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
1663                         break;
1664
1665                 btrfs_set_super_bytenr(sb, bytenr);
1666
1667                 crc = ~(u32)0;
1668                 crc = btrfs_csum_data((char *)sb + BTRFS_CSUM_SIZE, crc,
1669                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1670                 btrfs_csum_final(crc, &sb->csum[0]);
1671
1672                 /*
1673                  * super_copy is BTRFS_SUPER_INFO_SIZE bytes and is
1674                  * zero filled, we can use it directly
1675                  */
1676                 ret = pwrite64(device->fd, root->fs_info->super_copy,
1677                                 BTRFS_SUPER_INFO_SIZE, bytenr);
1678                 if (ret != BTRFS_SUPER_INFO_SIZE)
1679                         goto write_err;
1680         }
1681
1682         return 0;
1683
1684 write_err:
1685         if (ret > 0)
1686                 fprintf(stderr, "WARNING: failed to write all sb data\n");
1687         else
1688                 fprintf(stderr, "WARNING: failed to write sb: %s\n",
1689                         strerror(errno));
1690         return ret;
1691 }
1692
1693 int write_all_supers(struct btrfs_root *root)
1694 {
1695         struct list_head *cur;
1696         struct list_head *head = &root->fs_info->fs_devices->devices;
1697         struct btrfs_device *dev;
1698         struct btrfs_super_block *sb;
1699         struct btrfs_dev_item *dev_item;
1700         int ret;
1701         u64 flags;
1702
1703         sb = root->fs_info->super_copy;
1704         dev_item = &sb->dev_item;
1705         list_for_each(cur, head) {
1706                 dev = list_entry(cur, struct btrfs_device, dev_list);
1707                 if (!dev->writeable)
1708                         continue;
1709
1710                 btrfs_set_stack_device_generation(dev_item, 0);
1711                 btrfs_set_stack_device_type(dev_item, dev->type);
1712                 btrfs_set_stack_device_id(dev_item, dev->devid);
1713                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1714                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1715                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1716                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1717                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1718                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1719                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1720
1721                 flags = btrfs_super_flags(sb);
1722                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1723
1724                 ret = write_dev_supers(root, sb, dev);
1725                 BUG_ON(ret);
1726         }
1727         return 0;
1728 }
1729
1730 int write_ctree_super(struct btrfs_trans_handle *trans,
1731                       struct btrfs_root *root)
1732 {
1733         int ret;
1734         struct btrfs_root *tree_root = root->fs_info->tree_root;
1735         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1736
1737         if (root->fs_info->readonly)
1738                 return 0;
1739
1740         btrfs_set_super_generation(root->fs_info->super_copy,
1741                                    trans->transid);
1742         btrfs_set_super_root(root->fs_info->super_copy,
1743                              tree_root->node->start);
1744         btrfs_set_super_root_level(root->fs_info->super_copy,
1745                                    btrfs_header_level(tree_root->node));
1746         btrfs_set_super_chunk_root(root->fs_info->super_copy,
1747                                    chunk_root->node->start);
1748         btrfs_set_super_chunk_root_level(root->fs_info->super_copy,
1749                                          btrfs_header_level(chunk_root->node));
1750         btrfs_set_super_chunk_root_generation(root->fs_info->super_copy,
1751                                 btrfs_header_generation(chunk_root->node));
1752
1753         ret = write_all_supers(root);
1754         if (ret)
1755                 fprintf(stderr, "failed to write new super block err %d\n", ret);
1756         return ret;
1757 }
1758
1759 int close_ctree_fs_info(struct btrfs_fs_info *fs_info)
1760 {
1761         int ret;
1762         struct btrfs_trans_handle *trans;
1763         struct btrfs_root *root = fs_info->tree_root;
1764
1765         if (fs_info->last_trans_committed !=
1766             fs_info->generation) {
1767                 BUG_ON(!root);
1768                 trans = btrfs_start_transaction(root, 1);
1769                 btrfs_commit_transaction(trans, root);
1770                 trans = btrfs_start_transaction(root, 1);
1771                 ret = commit_tree_roots(trans, fs_info);
1772                 BUG_ON(ret);
1773                 ret = __commit_transaction(trans, root);
1774                 BUG_ON(ret);
1775                 write_ctree_super(trans, root);
1776                 kfree(trans);
1777         }
1778
1779         if (fs_info->finalize_on_close) {
1780                 btrfs_set_super_magic(fs_info->super_copy, BTRFS_MAGIC);
1781                 root->fs_info->finalize_on_close = 0;
1782                 ret = write_all_supers(root);
1783                 if (ret)
1784                         fprintf(stderr,
1785                                 "failed to write new super block err %d\n", ret);
1786         }
1787         btrfs_free_block_groups(fs_info);
1788
1789         free_fs_roots_tree(&fs_info->fs_root_tree);
1790
1791         btrfs_release_all_roots(fs_info);
1792         ret = btrfs_close_devices(fs_info->fs_devices);
1793         btrfs_cleanup_all_caches(fs_info);
1794         btrfs_free_fs_info(fs_info);
1795         return ret;
1796 }
1797
1798 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1799                      struct extent_buffer *eb)
1800 {
1801         return clear_extent_buffer_dirty(eb);
1802 }
1803
1804 void btrfs_mark_buffer_dirty(struct extent_buffer *eb)
1805 {
1806         set_extent_buffer_dirty(eb);
1807 }
1808
1809 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1810 {
1811         int ret;
1812
1813         ret = extent_buffer_uptodate(buf);
1814         if (!ret)
1815                 return ret;
1816
1817         ret = verify_parent_transid(buf->tree, buf, parent_transid, 1);
1818         return !ret;
1819 }
1820
1821 int btrfs_set_buffer_uptodate(struct extent_buffer *eb)
1822 {
1823         return set_extent_buffer_uptodate(eb);
1824 }