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 leafsize = btrfs_super_leafsize(fs_info->super_copy);
  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                                                     leafsize))
  68                 return BTRFS_BAD_NRITEMS;
  69
  70         fs_devices = fs_info->fs_devices;
  71         while (fs_devices) {
  72                 if (fs_info->ignore_fsid_mismatch ||
  73                     !memcmp_extent_buffer(buf, fs_devices->fsid,
  74                                           btrfs_header_fsid(),
  75                                           BTRFS_FSID_SIZE)) {
  76                         ret = 0;
  77                         break;
  78                 }
  79                 fs_devices = fs_devices->seed;
  80         }
  81         return ret;
  82 }
  83
  84 static void print_tree_block_error(struct btrfs_fs_info *fs_info,
  85                                 struct extent_buffer *eb,
  86                                 int err)
  87 {
  88         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = {'\0'};
  89         char found_uuid[BTRFS_UUID_UNPARSED_SIZE] = {'\0'};
  90         u8 buf[BTRFS_UUID_SIZE];
  91
  92         switch (err) {
  93         case BTRFS_BAD_FSID:
  94                 read_extent_buffer(eb, buf, btrfs_header_fsid(),
  95                                    BTRFS_UUID_SIZE);
  96                 uuid_unparse(buf, found_uuid);
  97                 uuid_unparse(fs_info->fsid, fs_uuid);
  98                 fprintf(stderr, "fsid mismatch, want=%s, have=%s\n",
  99                         fs_uuid, found_uuid);
 100                 break;
 101         case BTRFS_BAD_BYTENR:
 102                 fprintf(stderr, "bytenr mismatch, want=%llu, have=%llu\n",
 103                         eb->start, btrfs_header_bytenr(eb));
 104                 break;
 105         case BTRFS_BAD_LEVEL:
 106                 fprintf(stderr, "bad level, %u > %u\n",
 107                         btrfs_header_level(eb), BTRFS_MAX_LEVEL);
 108                 break;
 109         case BTRFS_BAD_NRITEMS:
 110                 fprintf(stderr, "invalid nr_items: %u\n",
 111                         btrfs_header_nritems(eb));
 112                 break;
 113         }
 114 }
 115
 116 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 117 {
 118         return crc32c(seed, data, len);
 119 }
 120
 121 void btrfs_csum_final(u32 crc, char *result)
 122 {
 123         *(__le32 *)result = ~cpu_to_le32(crc);
 124 }
 125
 126 static int __csum_tree_block_size(struct extent_buffer *buf, u16 csum_size,
 127                                   int verify, int silent)
 128 {
 129         char result[BTRFS_CSUM_SIZE];
 130         u32 len;
 131         u32 crc = ~(u32)0;
 132
 133         len = buf->len - BTRFS_CSUM_SIZE;
 134         crc = crc32c(crc, buf->data + BTRFS_CSUM_SIZE, len);
 135         btrfs_csum_final(crc, result);
 136
 137         if (verify) {
 138                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 139                         if (!silent)
 140                                 printk("checksum verify failed on %llu found %08X wanted %08X\n",
 141                                        (unsigned long long)buf->start,
 142                                        *((u32 *)result),
 143                                        *((u32*)(char *)buf->data));
 144                         return 1;
 145                 }
 146         } else {
 147                 write_extent_buffer(buf, result, 0, csum_size);
 148         }
 149         return 0;
 150 }
 151
 152 int csum_tree_block_size(struct extent_buffer *buf, u16 csum_size, int verify)
 153 {
 154         return __csum_tree_block_size(buf, csum_size, verify, 0);
 155 }
 156
 157 int verify_tree_block_csum_silent(struct extent_buffer *buf, u16 csum_size)
 158 {
 159         return __csum_tree_block_size(buf, csum_size, 1, 1);
 160 }
 161
 162 static int csum_tree_block_fs_info(struct btrfs_fs_info *fs_info,
 163                                    struct extent_buffer *buf, int verify)
 164 {
 165         u16 csum_size =
 166                 btrfs_super_csum_size(fs_info->super_copy);
 167         if (verify && fs_info->suppress_check_block_errors)
 168                 return verify_tree_block_csum_silent(buf, csum_size);
 169         return csum_tree_block_size(buf, csum_size, verify);
 170 }
 171
 172 int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 173                            int verify)
 174 {
 175         return csum_tree_block_fs_info(root->fs_info, buf, verify);
 176 }
 177
 178 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 179                                             u64 bytenr, u32 blocksize)
 180 {
 181         return find_extent_buffer(&root->fs_info->extent_cache,
 182                                   bytenr, blocksize);
 183 }
 184
 185 struct extent_buffer* btrfs_find_create_tree_block(
 186                 struct btrfs_fs_info *fs_info, u64 bytenr, u32 blocksize)
 187 {
 188         return alloc_extent_buffer(&fs_info->extent_cache, bytenr, blocksize);
 189 }
 190
 191 void readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 192                           u64 parent_transid)
 193 {
 194         struct extent_buffer *eb;
 195         u64 length;
 196         struct btrfs_multi_bio *multi = NULL;
 197         struct btrfs_device *device;
 198
 199         eb = btrfs_find_tree_block(root, bytenr, blocksize);
 200         if (!(eb && btrfs_buffer_uptodate(eb, parent_transid)) &&
 201             !btrfs_map_block(&root->fs_info->mapping_tree, READ,
 202                              bytenr, &length, &multi, 0, NULL)) {
 203                 device = multi->stripes[0].dev;
 204                 device->total_ios++;
 205                 blocksize = min(blocksize, (u32)(64 * 1024));
 206                 readahead(device->fd, multi->stripes[0].physical, blocksize);
 207         }
 208
 209         free_extent_buffer(eb);
 210         kfree(multi);
 211 }
 212
 213 static int verify_parent_transid(struct extent_io_tree *io_tree,
 214                                  struct extent_buffer *eb, u64 parent_transid,
 215                                  int ignore)
 216 {
 217         int ret;
 218
 219         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 220                 return 0;
 221
 222         if (extent_buffer_uptodate(eb) &&
 223             btrfs_header_generation(eb) == parent_transid) {
 224                 ret = 0;
 225                 goto out;
 226         }
 227         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 228                (unsigned long long)eb->start,
 229                (unsigned long long)parent_transid,
 230                (unsigned long long)btrfs_header_generation(eb));
 231         if (ignore) {
 232                 eb->flags |= EXTENT_BAD_TRANSID;
 233                 printk("Ignoring transid failure\n");
 234                 return 0;
 235         }
 236
 237         ret = 1;
 238 out:
 239         clear_extent_buffer_uptodate(io_tree, eb);
 240         return ret;
 241
 242 }
 243
 244
 245 int read_whole_eb(struct btrfs_fs_info *info, struct extent_buffer *eb, int mirror)
 246 {
 247         unsigned long offset = 0;
 248         struct btrfs_multi_bio *multi = NULL;
 249         struct btrfs_device *device;
 250         int ret = 0;
 251         u64 read_len;
 252         unsigned long bytes_left = eb->len;
 253
 254         while (bytes_left) {
 255                 read_len = bytes_left;
 256                 device = NULL;
 257
 258                 if (!info->on_restoring &&
 259                     eb->start != BTRFS_SUPER_INFO_OFFSET) {
 260                         ret = btrfs_map_block(&info->mapping_tree, READ,
 261                                               eb->start + offset, &read_len, &multi,
 262                                               mirror, NULL);
 263                         if (ret) {
 264                                 printk("Couldn't map the block %Lu\n", eb->start + offset);
 265                                 kfree(multi);
 266                                 return -EIO;
 267                         }
 268                         device = multi->stripes[0].dev;
 269
 270                         if (device->fd <= 0) {
 271                                 kfree(multi);
 272                                 return -EIO;
 273                         }
 274
 275                         eb->fd = device->fd;
 276                         device->total_ios++;
 277                         eb->dev_bytenr = multi->stripes[0].physical;
 278                         kfree(multi);
 279                         multi = NULL;
 280                 } else {
 281                         /* special case for restore metadump */
 282                         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
 283                                 if (device->devid == 1)
 284                                         break;
 285                         }
 286
 287                         eb->fd = device->fd;
 288                         eb->dev_bytenr = eb->start;
 289                         device->total_ios++;
 290                 }
 291
 292                 if (read_len > bytes_left)
 293                         read_len = bytes_left;
 294
 295                 ret = read_extent_from_disk(eb, offset, read_len);
 296                 if (ret)
 297                         return -EIO;
 298                 offset += read_len;
 299                 bytes_left -= read_len;
 300         }
 301         return 0;
 302 }
 303
 304 struct extent_buffer* read_tree_block_fs_info(
 305                 struct btrfs_fs_info *fs_info, u64 bytenr, u32 blocksize,
 306                 u64 parent_transid)
 307 {
 308         int ret;
 309         struct extent_buffer *eb;
 310         u64 best_transid = 0;
 311         int mirror_num = 0;
 312         int good_mirror = 0;
 313         int num_copies;
 314         int ignore = 0;
 315
 316         eb = btrfs_find_create_tree_block(fs_info, bytenr, blocksize);
 317         if (!eb)
 318                 return ERR_PTR(-ENOMEM);
 319
 320         if (btrfs_buffer_uptodate(eb, parent_transid))
 321                 return eb;
 322
 323         while (1) {
 324                 ret = read_whole_eb(fs_info, eb, mirror_num);
 325                 if (ret == 0 && csum_tree_block_fs_info(fs_info, eb, 1) == 0 &&
 326                     check_tree_block(fs_info, eb) == 0 &&
 327                     verify_parent_transid(eb->tree, eb, parent_transid, ignore)
 328                     == 0) {
 329                         if (eb->flags & EXTENT_BAD_TRANSID &&
 330                             list_empty(&eb->recow)) {
 331                                 list_add_tail(&eb->recow,
 332                                               &fs_info->recow_ebs);
 333                                 eb->refs++;
 334                         }
 335                         btrfs_set_buffer_uptodate(eb);
 336                         return eb;
 337                 }
 338                 if (ignore) {
 339                         if (check_tree_block(fs_info, eb)) {
 340                                 if (!fs_info->suppress_check_block_errors)
 341                                         print_tree_block_error(fs_info, eb,
 342                                                 check_tree_block(fs_info, eb));
 343                         } else {
 344                                 if (!fs_info->suppress_check_block_errors)
 345                                         fprintf(stderr, "Csum didn't match\n");
 346                         }
 347                         ret = -EIO;
 348                         break;
 349                 }
 350                 num_copies = btrfs_num_copies(&fs_info->mapping_tree,
 351                                               eb->start, eb->len);
 352                 if (num_copies == 1) {
 353                         ignore = 1;
 354                         continue;
 355                 }
 356                 if (btrfs_header_generation(eb) > best_transid && mirror_num) {
 357                         best_transid = btrfs_header_generation(eb);
 358                         good_mirror = mirror_num;
 359                 }
 360                 mirror_num++;
 361                 if (mirror_num > num_copies) {
 362                         mirror_num = good_mirror;
 363                         ignore = 1;
 364                         continue;
 365                 }
 366         }
 367         free_extent_buffer(eb);
 368         return ERR_PTR(ret);
 369 }
 370
 371 int read_extent_data(struct btrfs_root *root, char *data,
 372                            u64 logical, u64 *len, int mirror)
 373 {
 374         u64 offset = 0;
 375         struct btrfs_multi_bio *multi = NULL;
 376         struct btrfs_fs_info *info = root->fs_info;
 377         struct btrfs_device *device;
 378         int ret = 0;
 379         u64 max_len = *len;
 380
 381         ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
 382                               &multi, mirror, NULL);
 383         if (ret) {
 384                 fprintf(stderr, "Couldn't map the block %llu\n",
 385                                 logical + offset);
 386                 goto err;
 387         }
 388         device = multi->stripes[0].dev;
 389
 390         if (device->fd <= 0)
 391                 goto err;
 392         if (*len > max_len)
 393                 *len = max_len;
 394
 395         ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
 396         if (ret != *len)
 397                 ret = -EIO;
 398         else
 399                 ret = 0;
 400 err:
 401         kfree(multi);
 402         return ret;
 403 }
 404
 405 int write_and_map_eb(struct btrfs_trans_handle *trans,
 406                      struct btrfs_root *root,
 407                      struct extent_buffer *eb)
 408 {
 409         int ret;
 410         int dev_nr;
 411         u64 length;
 412         u64 *raid_map = NULL;
 413         struct btrfs_multi_bio *multi = NULL;
 414
 415         dev_nr = 0;
 416         length = eb->len;
 417         ret = btrfs_map_block(&root->fs_info->mapping_tree, WRITE,
 418                               eb->start, &length, &multi, 0, &raid_map);
 419
 420         if (raid_map) {
 421                 ret = write_raid56_with_parity(root->fs_info, eb, multi,
 422                                                length, raid_map);
 423                 BUG_ON(ret);
 424         } else while (dev_nr < multi->num_stripes) {
 425                 BUG_ON(ret);
 426                 eb->fd = multi->stripes[dev_nr].dev->fd;
 427                 eb->dev_bytenr = multi->stripes[dev_nr].physical;
 428                 multi->stripes[dev_nr].dev->total_ios++;
 429                 dev_nr++;
 430                 ret = write_extent_to_disk(eb);
 431                 BUG_ON(ret);
 432         }
 433         kfree(raid_map);
 434         kfree(multi);
 435         return 0;
 436 }
 437
 438 int write_tree_block(struct btrfs_trans_handle *trans,
 439                      struct btrfs_root *root,
 440                      struct extent_buffer *eb)
 441 {
 442         if (check_tree_block(root->fs_info, eb)) {
 443                 print_tree_block_error(root->fs_info, eb,
 444                                 check_tree_block(root->fs_info, eb));
 445                 BUG();
 446         }
 447
 448         if (trans && !btrfs_buffer_uptodate(eb, trans->transid))
 449                 BUG();
 450
 451         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 452         csum_tree_block(root, eb, 0);
 453
 454         return write_and_map_eb(trans, root, eb);
 455 }
 456
 457 int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 458                         u32 stripesize, struct btrfs_root *root,
 459                         struct btrfs_fs_info *fs_info, u64 objectid)
 460 {
 461         root->node = NULL;
 462         root->commit_root = NULL;
 463         root->sectorsize = sectorsize;
 464         root->nodesize = nodesize;
 465         root->leafsize = leafsize;
 466         root->stripesize = stripesize;
 467         root->ref_cows = 0;
 468         root->track_dirty = 0;
 469
 470         root->fs_info = fs_info;
 471         root->objectid = objectid;
 472         root->last_trans = 0;
 473         root->highest_inode = 0;
 474         root->last_inode_alloc = 0;
 475
 476         INIT_LIST_HEAD(&root->dirty_list);
 477         INIT_LIST_HEAD(&root->orphan_data_extents);
 478         memset(&root->root_key, 0, sizeof(root->root_key));
 479         memset(&root->root_item, 0, sizeof(root->root_item));
 480         root->root_key.objectid = objectid;
 481         return 0;
 482 }
 483
 484 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 485                                struct btrfs_root *root)
 486 {
 487         int ret;
 488         u64 old_root_bytenr;
 489         struct btrfs_root *tree_root = root->fs_info->tree_root;
 490
 491         btrfs_write_dirty_block_groups(trans, root);
 492         while(1) {
 493                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 494                 if (old_root_bytenr == root->node->start)
 495                         break;
 496                 btrfs_set_root_bytenr(&root->root_item,
 497                                        root->node->start);
 498                 btrfs_set_root_generation(&root->root_item,
 499                                           trans->transid);
 500                 root->root_item.level = btrfs_header_level(root->node);
 501                 ret = btrfs_update_root(trans, tree_root,
 502                                         &root->root_key,
 503                                         &root->root_item);
 504                 BUG_ON(ret);
 505                 btrfs_write_dirty_block_groups(trans, root);
 506         }
 507         return 0;
 508 }
 509
 510 static int commit_tree_roots(struct btrfs_trans_handle *trans,
 511                              struct btrfs_fs_info *fs_info)
 512 {
 513         struct btrfs_root *root;
 514         struct list_head *next;
 515         struct extent_buffer *eb;
 516         int ret;
 517
 518         if (fs_info->readonly)
 519                 return 0;
 520
 521         eb = fs_info->tree_root->node;
 522         extent_buffer_get(eb);
 523         ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
 524         free_extent_buffer(eb);
 525         if (ret)
 526                 return ret;
 527
 528         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 529                 next = fs_info->dirty_cowonly_roots.next;
 530                 list_del_init(next);
 531                 root = list_entry(next, struct btrfs_root, dirty_list);
 532                 update_cowonly_root(trans, root);
 533                 free_extent_buffer(root->commit_root);
 534                 root->commit_root = NULL;
 535         }
 536
 537         return 0;
 538 }
 539
 540 static int __commit_transaction(struct btrfs_trans_handle *trans,
 541                                 struct btrfs_root *root)
 542 {
 543         u64 start;
 544         u64 end;
 545         struct extent_buffer *eb;
 546         struct extent_io_tree *tree = &root->fs_info->extent_cache;
 547         int ret;
 548
 549         while(1) {
 550                 ret = find_first_extent_bit(tree, 0, &start, &end,
 551                                             EXTENT_DIRTY);
 552                 if (ret)
 553                         break;
 554                 while(start <= end) {
 555                         eb = find_first_extent_buffer(tree, start);
 556                         BUG_ON(!eb || eb->start != start);
 557                         ret = write_tree_block(trans, root, eb);
 558                         BUG_ON(ret);
 559                         start += eb->len;
 560                         clear_extent_buffer_dirty(eb);
 561                         free_extent_buffer(eb);
 562                 }
 563         }
 564         return 0;
 565 }
 566
 567 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 568                              struct btrfs_root *root)
 569 {
 570         u64 transid = trans->transid;
 571         int ret = 0;
 572         struct btrfs_fs_info *fs_info = root->fs_info;
 573
 574         if (root->commit_root == root->node)
 575                 goto commit_tree;
 576         if (root == root->fs_info->tree_root)
 577                 goto commit_tree;
 578         if (root == root->fs_info->chunk_root)
 579                 goto commit_tree;
 580
 581         free_extent_buffer(root->commit_root);
 582         root->commit_root = NULL;
 583
 584         btrfs_set_root_bytenr(&root->root_item, root->node->start);
 585         btrfs_set_root_generation(&root->root_item, trans->transid);
 586         root->root_item.level = btrfs_header_level(root->node);
 587         ret = btrfs_update_root(trans, root->fs_info->tree_root,
 588                                 &root->root_key, &root->root_item);
 589         BUG_ON(ret);
 590 commit_tree:
 591         ret = commit_tree_roots(trans, fs_info);
 592         BUG_ON(ret);
 593         ret = __commit_transaction(trans, root);
 594         BUG_ON(ret);
 595         write_ctree_super(trans, root);
 596         btrfs_finish_extent_commit(trans, fs_info->extent_root,
 597                                    &fs_info->pinned_extents);
 598         btrfs_free_transaction(root, trans);
 599         free_extent_buffer(root->commit_root);
 600         root->commit_root = NULL;
 601         fs_info->running_transaction = NULL;
 602         fs_info->last_trans_committed = transid;
 603         return 0;
 604 }
 605
 606 static int find_and_setup_root(struct btrfs_root *tree_root,
 607                                struct btrfs_fs_info *fs_info,
 608                                u64 objectid, struct btrfs_root *root)
 609 {
 610         int ret;
 611         u32 blocksize;
 612         u64 generation;
 613
 614         __setup_root(tree_root->nodesize, tree_root->leafsize,
 615                      tree_root->sectorsize, tree_root->stripesize,
 616                      root, fs_info, objectid);
 617         ret = btrfs_find_last_root(tree_root, objectid,
 618                                    &root->root_item, &root->root_key);
 619         if (ret)
 620                 return ret;
 621
 622         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 623         generation = btrfs_root_generation(&root->root_item);
 624         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 625                                      blocksize, generation);
 626         if (!extent_buffer_uptodate(root->node))
 627                 return -EIO;
 628
 629         return 0;
 630 }
 631
 632 static int find_and_setup_log_root(struct btrfs_root *tree_root,
 633                                struct btrfs_fs_info *fs_info,
 634                                struct btrfs_super_block *disk_super)
 635 {
 636         u32 blocksize;
 637         u64 blocknr = btrfs_super_log_root(disk_super);
 638         struct btrfs_root *log_root = malloc(sizeof(struct btrfs_root));
 639
 640         if (!log_root)
 641                 return -ENOMEM;
 642
 643         if (blocknr == 0) {
 644                 free(log_root);
 645                 return 0;
 646         }
 647
 648         blocksize = btrfs_level_size(tree_root,
 649                              btrfs_super_log_root_level(disk_super));
 650
 651         __setup_root(tree_root->nodesize, tree_root->leafsize,
 652                      tree_root->sectorsize, tree_root->stripesize,
 653                      log_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 654
 655         log_root->node = read_tree_block(tree_root, blocknr,
 656                                      blocksize,
 657                                      btrfs_super_generation(disk_super) + 1);
 658
 659         fs_info->log_root_tree = log_root;
 660
 661         if (!extent_buffer_uptodate(log_root->node)) {
 662                 free_extent_buffer(log_root->node);
 663                 free(log_root);
 664                 fs_info->log_root_tree = NULL;
 665                 return -EIO;
 666         }
 667
 668         return 0;
 669 }
 670
 671 int btrfs_free_fs_root(struct btrfs_root *root)
 672 {
 673         if (root->node)
 674                 free_extent_buffer(root->node);
 675         if (root->commit_root)
 676                 free_extent_buffer(root->commit_root);
 677         kfree(root);
 678         return 0;
 679 }
 680
 681 static void __free_fs_root(struct rb_node *node)
 682 {
 683         struct btrfs_root *root;
 684
 685         root = container_of(node, struct btrfs_root, rb_node);
 686         btrfs_free_fs_root(root);
 687 }
 688
 689 FREE_RB_BASED_TREE(fs_roots, __free_fs_root);
 690
 691 struct btrfs_root *btrfs_read_fs_root_no_cache(struct btrfs_fs_info *fs_info,
 692                                                struct btrfs_key *location)
 693 {
 694         struct btrfs_root *root;
 695         struct btrfs_root *tree_root = fs_info->tree_root;
 696         struct btrfs_path *path;
 697         struct extent_buffer *l;
 698         u64 generation;
 699         u32 blocksize;
 700         int ret = 0;
 701
 702         root = calloc(1, sizeof(*root));
 703         if (!root)
 704                 return ERR_PTR(-ENOMEM);
 705         if (location->offset == (u64)-1) {
 706                 ret = find_and_setup_root(tree_root, fs_info,
 707                                           location->objectid, root);
 708                 if (ret) {
 709                         free(root);
 710                         return ERR_PTR(ret);
 711                 }
 712                 goto insert;
 713         }
 714
 715         __setup_root(tree_root->nodesize, tree_root->leafsize,
 716                      tree_root->sectorsize, tree_root->stripesize,
 717                      root, fs_info, location->objectid);
 718
 719         path = btrfs_alloc_path();
 720         BUG_ON(!path);
 721         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 722         if (ret != 0) {
 723                 if (ret > 0)
 724                         ret = -ENOENT;
 725                 goto out;
 726         }
 727         l = path->nodes[0];
 728         read_extent_buffer(l, &root->root_item,
 729                btrfs_item_ptr_offset(l, path->slots[0]),
 730                sizeof(root->root_item));
 731         memcpy(&root->root_key, location, sizeof(*location));
 732         ret = 0;
 733 out:
 734         btrfs_free_path(path);
 735         if (ret) {
 736                 free(root);
 737                 return ERR_PTR(ret);
 738         }
 739         generation = btrfs_root_generation(&root->root_item);
 740         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 741         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 742                                      blocksize, generation);
 743         if (!extent_buffer_uptodate(root->node)) {
 744                 free(root);
 745                 return ERR_PTR(-EIO);
 746         }
 747 insert:
 748         root->ref_cows = 1;
 749         return root;
 750 }
 751
 752 static int btrfs_fs_roots_compare_objectids(struct rb_node *node,
 753                                             void *data)
 754 {
 755         u64 objectid = *((u64 *)data);
 756         struct btrfs_root *root;
 757
 758         root = rb_entry(node, struct btrfs_root, rb_node);
 759         if (objectid > root->objectid)
 760                 return 1;
 761         else if (objectid < root->objectid)
 762                 return -1;
 763         else
 764                 return 0;
 765 }
 766
 767 static int btrfs_fs_roots_compare_roots(struct rb_node *node1,
 768                                         struct rb_node *node2)
 769 {
 770         struct btrfs_root *root;
 771
 772         root = rb_entry(node2, struct btrfs_root, rb_node);
 773         return btrfs_fs_roots_compare_objectids(node1, (void *)&root->objectid);
 774 }
 775
 776 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 777                                       struct btrfs_key *location)
 778 {
 779         struct btrfs_root *root;
 780         struct rb_node *node;
 781         int ret;
 782         u64 objectid = location->objectid;
 783
 784         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 785                 return fs_info->tree_root;
 786         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 787                 return fs_info->extent_root;
 788         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 789                 return fs_info->chunk_root;
 790         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 791                 return fs_info->dev_root;
 792         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
 793                 return fs_info->csum_root;
 794         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
 795                 return fs_info->quota_root;
 796
 797         BUG_ON(location->objectid == BTRFS_TREE_RELOC_OBJECTID ||
 798                location->offset != (u64)-1);
 799
 800         node = rb_search(&fs_info->fs_root_tree, (void *)&objectid,
 801                          btrfs_fs_roots_compare_objectids, NULL);
 802         if (node)
 803                 return container_of(node, struct btrfs_root, rb_node);
 804
 805         root = btrfs_read_fs_root_no_cache(fs_info, location);
 806         if (IS_ERR(root))
 807                 return root;
 808
 809         ret = rb_insert(&fs_info->fs_root_tree, &root->rb_node,
 810                         btrfs_fs_roots_compare_roots);
 811         BUG_ON(ret);
 812         return root;
 813 }
 814
 815 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
 816 {
 817         free(fs_info->tree_root);
 818         free(fs_info->extent_root);
 819         free(fs_info->chunk_root);
 820         free(fs_info->dev_root);
 821         free(fs_info->csum_root);
 822         free(fs_info->quota_root);
 823         free(fs_info->free_space_root);
 824         free(fs_info->super_copy);
 825         free(fs_info->log_root_tree);
 826         free(fs_info);
 827 }
 828
 829 struct btrfs_fs_info *btrfs_new_fs_info(int writable, u64 sb_bytenr)
 830 {
 831         struct btrfs_fs_info *fs_info;
 832
 833         fs_info = calloc(1, sizeof(struct btrfs_fs_info));
 834         if (!fs_info)
 835                 return NULL;
 836
 837         fs_info->tree_root = calloc(1, sizeof(struct btrfs_root));
 838         fs_info->extent_root = calloc(1, sizeof(struct btrfs_root));
 839         fs_info->chunk_root = calloc(1, sizeof(struct btrfs_root));
 840         fs_info->dev_root = calloc(1, sizeof(struct btrfs_root));
 841         fs_info->csum_root = calloc(1, sizeof(struct btrfs_root));
 842         fs_info->quota_root = calloc(1, sizeof(struct btrfs_root));
 843         fs_info->free_space_root = calloc(1, sizeof(struct btrfs_root));
 844         fs_info->super_copy = calloc(1, BTRFS_SUPER_INFO_SIZE);
 845
 846         if (!fs_info->tree_root || !fs_info->extent_root ||
 847             !fs_info->chunk_root || !fs_info->dev_root ||
 848             !fs_info->csum_root || !fs_info->quota_root ||
 849             !fs_info->free_space_root || !fs_info->super_copy)
 850                 goto free_all;
 851
 852         extent_io_tree_init(&fs_info->extent_cache);
 853         extent_io_tree_init(&fs_info->free_space_cache);
 854         extent_io_tree_init(&fs_info->block_group_cache);
 855         extent_io_tree_init(&fs_info->pinned_extents);
 856         extent_io_tree_init(&fs_info->pending_del);
 857         extent_io_tree_init(&fs_info->extent_ins);
 858         fs_info->excluded_extents = NULL;
 859
 860         fs_info->fs_root_tree = RB_ROOT;
 861         cache_tree_init(&fs_info->mapping_tree.cache_tree);
 862
 863         mutex_init(&fs_info->fs_mutex);
 864         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
 865         INIT_LIST_HEAD(&fs_info->space_info);
 866         INIT_LIST_HEAD(&fs_info->recow_ebs);
 867
 868         if (!writable)
 869                 fs_info->readonly = 1;
 870
 871         fs_info->super_bytenr = sb_bytenr;
 872         fs_info->data_alloc_profile = (u64)-1;
 873         fs_info->metadata_alloc_profile = (u64)-1;
 874         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
 875         return fs_info;
 876 free_all:
 877         btrfs_free_fs_info(fs_info);
 878         return NULL;
 879 }
 880
 881 int btrfs_check_fs_compatibility(struct btrfs_super_block *sb, int writable)
 882 {
 883         u64 features;
 884
 885         features = btrfs_super_incompat_flags(sb) &
 886                    ~BTRFS_FEATURE_INCOMPAT_SUPP;
 887         if (features) {
 888                 printk("couldn't open because of unsupported "
 889                        "option features (%Lx).\n",
 890                        (unsigned long long)features);
 891                 return -ENOTSUP;
 892         }
 893
 894         features = btrfs_super_incompat_flags(sb);
 895         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
 896                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
 897                 btrfs_set_super_incompat_flags(sb, features);
 898         }
 899
 900         features = btrfs_super_compat_ro_flags(sb) &
 901                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
 902         if (writable && features) {
 903                 printk("couldn't open RDWR because of unsupported "
 904                        "option features (%Lx).\n",
 905                        (unsigned long long)features);
 906                 return -ENOTSUP;
 907         }
 908         return 0;
 909 }
 910
 911 static int find_best_backup_root(struct btrfs_super_block *super)
 912 {
 913         struct btrfs_root_backup *backup;
 914         u64 orig_gen = btrfs_super_generation(super);
 915         u64 gen = 0;
 916         int best_index = 0;
 917         int i;
 918
 919         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
 920                 backup = super->super_roots + i;
 921                 if (btrfs_backup_tree_root_gen(backup) != orig_gen &&
 922                     btrfs_backup_tree_root_gen(backup) > gen) {
 923                         best_index = i;
 924                         gen = btrfs_backup_tree_root_gen(backup);
 925                 }
 926         }
 927         return best_index;
 928 }
 929
 930 static int setup_root_or_create_block(struct btrfs_fs_info *fs_info,
 931                                       enum btrfs_open_ctree_flags flags,
 932                                       struct btrfs_root *info_root,
 933                                       u64 objectid, char *str)
 934 {
 935         struct btrfs_super_block *sb = fs_info->super_copy;
 936         struct btrfs_root *root = fs_info->tree_root;
 937         u32 leafsize = btrfs_super_leafsize(sb);
 938         int ret;
 939
 940         ret = find_and_setup_root(root, fs_info, objectid, info_root);
 941         if (ret) {
 942                 printk("Couldn't setup %s tree\n", str);
 943                 if (!(flags & OPEN_CTREE_PARTIAL))
 944                         return -EIO;
 945                 /*
 946                  * Need a blank node here just so we don't screw up in the
 947                  * million of places that assume a root has a valid ->node
 948                  */
 949                 info_root->node =
 950                         btrfs_find_create_tree_block(fs_info, 0, leafsize);
 951                 if (!info_root->node)
 952                         return -ENOMEM;
 953                 clear_extent_buffer_uptodate(NULL, info_root->node);
 954         }
 955
 956         return 0;
 957 }
 958
 959 int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr,
 960                           enum btrfs_open_ctree_flags flags)
 961 {
 962         struct btrfs_super_block *sb = fs_info->super_copy;
 963         struct btrfs_root *root;
 964         struct btrfs_key key;
 965         u32 sectorsize;
 966         u32 nodesize;
 967         u32 leafsize;
 968         u32 stripesize;
 969         u64 generation;
 970         u32 blocksize;
 971         int ret;
 972
 973         nodesize = btrfs_super_nodesize(sb);
 974         leafsize = btrfs_super_leafsize(sb);
 975         sectorsize = btrfs_super_sectorsize(sb);
 976         stripesize = btrfs_super_stripesize(sb);
 977
 978         root = fs_info->tree_root;
 979         __setup_root(nodesize, leafsize, sectorsize, stripesize,
 980                      root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
 981         blocksize = btrfs_level_size(root, btrfs_super_root_level(sb));
 982         generation = btrfs_super_generation(sb);
 983
 984         if (!root_tree_bytenr && !(flags & OPEN_CTREE_BACKUP_ROOT)) {
 985                 root_tree_bytenr = btrfs_super_root(sb);
 986         } else if (flags & OPEN_CTREE_BACKUP_ROOT) {
 987                 struct btrfs_root_backup *backup;
 988                 int index = find_best_backup_root(sb);
 989                 if (index >= BTRFS_NUM_BACKUP_ROOTS) {
 990                         fprintf(stderr, "Invalid backup root number\n");
 991                         return -EIO;
 992                 }
 993                 backup = fs_info->super_copy->super_roots + index;
 994                 root_tree_bytenr = btrfs_backup_tree_root(backup);
 995                 generation = btrfs_backup_tree_root_gen(backup);
 996         }
 997
 998         root->node = read_tree_block(root, root_tree_bytenr, blocksize,
 999                                      generation);
1000         if (!extent_buffer_uptodate(root->node)) {
1001                 fprintf(stderr, "Couldn't read tree root\n");
1002                 return -EIO;
1003         }
1004
1005         ret = setup_root_or_create_block(fs_info, flags, fs_info->extent_root,
1006                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
1007         if (ret)
1008                 return ret;
1009         fs_info->extent_root->track_dirty = 1;
1010
1011         ret = find_and_setup_root(root, fs_info, BTRFS_DEV_TREE_OBJECTID,
1012                                   fs_info->dev_root);
1013         if (ret) {
1014                 printk("Couldn't setup device tree\n");
1015                 return -EIO;
1016         }
1017         fs_info->dev_root->track_dirty = 1;
1018
1019         ret = setup_root_or_create_block(fs_info, flags, fs_info->csum_root,
1020                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
1021         if (ret)
1022                 return ret;
1023         fs_info->csum_root->track_dirty = 1;
1024
1025         ret = find_and_setup_root(root, fs_info, BTRFS_QUOTA_TREE_OBJECTID,
1026                                   fs_info->quota_root);
1027         if (ret == 0)
1028                 fs_info->quota_enabled = 1;
1029
1030         if (btrfs_fs_compat_ro(fs_info, BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
1031                 ret = find_and_setup_root(root, fs_info, BTRFS_FREE_SPACE_TREE_OBJECTID,
1032                                           fs_info->free_space_root);
1033                 if (ret) {
1034                         printk("Couldn't read free space tree\n");
1035                         return -EIO;
1036                 }
1037                 fs_info->free_space_root->track_dirty = 1;
1038         }
1039
1040         ret = find_and_setup_log_root(root, fs_info, sb);
1041         if (ret) {
1042                 printk("Couldn't setup log root tree\n");
1043                 if (!(flags & OPEN_CTREE_PARTIAL))
1044                         return -EIO;
1045         }
1046
1047         fs_info->generation = generation;
1048         fs_info->last_trans_committed = generation;
1049         if (extent_buffer_uptodate(fs_info->extent_root->node) &&
1050             !(flags & OPEN_CTREE_NO_BLOCK_GROUPS))
1051                 btrfs_read_block_groups(fs_info->tree_root);
1052
1053         key.objectid = BTRFS_FS_TREE_OBJECTID;
1054         key.type = BTRFS_ROOT_ITEM_KEY;
1055         key.offset = (u64)-1;
1056         fs_info->fs_root = btrfs_read_fs_root(fs_info, &key);
1057
1058         if (IS_ERR(fs_info->fs_root))
1059                 return -EIO;
1060         return 0;
1061 }
1062
1063 void btrfs_release_all_roots(struct btrfs_fs_info *fs_info)
1064 {
1065         if (fs_info->free_space_root)
1066                 free_extent_buffer(fs_info->free_space_root->node);
1067         if (fs_info->quota_root)
1068                 free_extent_buffer(fs_info->quota_root->node);
1069         if (fs_info->csum_root)
1070                 free_extent_buffer(fs_info->csum_root->node);
1071         if (fs_info->dev_root)
1072                 free_extent_buffer(fs_info->dev_root->node);
1073         if (fs_info->extent_root)
1074                 free_extent_buffer(fs_info->extent_root->node);
1075         if (fs_info->tree_root)
1076                 free_extent_buffer(fs_info->tree_root->node);
1077         if (fs_info->log_root_tree)
1078                 free_extent_buffer(fs_info->log_root_tree->node);
1079         if (fs_info->chunk_root)
1080                 free_extent_buffer(fs_info->chunk_root->node);
1081 }
1082
1083 static void free_map_lookup(struct cache_extent *ce)
1084 {
1085         struct map_lookup *map;
1086
1087         map = container_of(ce, struct map_lookup, ce);
1088         kfree(map);
1089 }
1090
1091 FREE_EXTENT_CACHE_BASED_TREE(mapping_cache, free_map_lookup);
1092
1093 void btrfs_cleanup_all_caches(struct btrfs_fs_info *fs_info)
1094 {
1095         while (!list_empty(&fs_info->recow_ebs)) {
1096                 struct extent_buffer *eb;
1097                 eb = list_first_entry(&fs_info->recow_ebs,
1098                                       struct extent_buffer, recow);
1099                 list_del_init(&eb->recow);
1100                 free_extent_buffer(eb);
1101         }
1102         free_mapping_cache_tree(&fs_info->mapping_tree.cache_tree);
1103         extent_io_tree_cleanup(&fs_info->extent_cache);
1104         extent_io_tree_cleanup(&fs_info->free_space_cache);
1105         extent_io_tree_cleanup(&fs_info->block_group_cache);
1106         extent_io_tree_cleanup(&fs_info->pinned_extents);
1107         extent_io_tree_cleanup(&fs_info->pending_del);
1108         extent_io_tree_cleanup(&fs_info->extent_ins);
1109 }
1110
1111 int btrfs_scan_fs_devices(int fd, const char *path,
1112                           struct btrfs_fs_devices **fs_devices,
1113                           u64 sb_bytenr, int super_recover,
1114                           int skip_devices)
1115 {
1116         u64 total_devs;
1117         u64 dev_size;
1118         off_t seek_ret;
1119         int ret;
1120         if (!sb_bytenr)
1121                 sb_bytenr = BTRFS_SUPER_INFO_OFFSET;
1122
1123         seek_ret = lseek(fd, 0, SEEK_END);
1124         if (seek_ret < 0)
1125                 return -errno;
1126
1127         dev_size = seek_ret;
1128         lseek(fd, 0, SEEK_SET);
1129         if (sb_bytenr > dev_size) {
1130                 fprintf(stderr, "Superblock bytenr is larger than device size\n");
1131                 return -EINVAL;
1132         }
1133
1134         ret = btrfs_scan_one_device(fd, path, fs_devices,
1135                                     &total_devs, sb_bytenr, super_recover);
1136         if (ret) {
1137                 fprintf(stderr, "No valid Btrfs found on %s\n", path);
1138                 return ret;
1139         }
1140
1141         if (!skip_devices && total_devs != 1) {
1142                 ret = btrfs_scan_lblkid();
1143                 if (ret)
1144                         return ret;
1145         }
1146         return 0;
1147 }
1148
1149 int btrfs_setup_chunk_tree_and_device_map(struct btrfs_fs_info *fs_info,
1150                                           u64 chunk_root_bytenr)
1151 {
1152         struct btrfs_super_block *sb = fs_info->super_copy;
1153         u32 sectorsize;
1154         u32 nodesize;
1155         u32 leafsize;
1156         u32 blocksize;
1157         u32 stripesize;
1158         u64 generation;
1159         int ret;
1160
1161         nodesize = btrfs_super_nodesize(sb);
1162         leafsize = btrfs_super_leafsize(sb);
1163         sectorsize = btrfs_super_sectorsize(sb);
1164         stripesize = btrfs_super_stripesize(sb);
1165
1166         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1167                      fs_info->chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1168
1169         ret = btrfs_read_sys_array(fs_info->chunk_root);
1170         if (ret)
1171                 return ret;
1172
1173         blocksize = btrfs_level_size(fs_info->chunk_root,
1174                                      btrfs_super_chunk_root_level(sb));
1175         generation = btrfs_super_chunk_root_generation(sb);
1176
1177         if (chunk_root_bytenr && !IS_ALIGNED(chunk_root_bytenr,
1178                                             btrfs_super_sectorsize(sb))) {
1179                 warning("chunk_root_bytenr %llu is unaligned to %u, ignore it",
1180                         chunk_root_bytenr, btrfs_super_sectorsize(sb));
1181                 chunk_root_bytenr = 0;
1182         }
1183
1184         if (!chunk_root_bytenr)
1185                 chunk_root_bytenr = btrfs_super_chunk_root(sb);
1186         else
1187                 generation = 0;
1188
1189         fs_info->chunk_root->node = read_tree_block(fs_info->chunk_root,
1190                                                     chunk_root_bytenr,
1191                                                     blocksize, generation);
1192         if (!extent_buffer_uptodate(fs_info->chunk_root->node)) {
1193                 if (fs_info->ignore_chunk_tree_error) {
1194                         warning("cannot read chunk root, continue anyway");
1195                         fs_info->chunk_root = NULL;
1196                         return 0;
1197                 } else {
1198                         error("cannot read chunk root");
1199                         return -EIO;
1200                 }
1201         }
1202
1203         if (!(btrfs_super_flags(sb) & BTRFS_SUPER_FLAG_METADUMP)) {
1204                 ret = btrfs_read_chunk_tree(fs_info->chunk_root);
1205                 if (ret) {
1206                         fprintf(stderr, "Couldn't read chunk tree\n");
1207                         return ret;
1208                 }
1209         }
1210         return 0;
1211 }
1212
1213 static struct btrfs_fs_info *__open_ctree_fd(int fp, const char *path,
1214                                              u64 sb_bytenr,
1215                                              u64 root_tree_bytenr,
1216                                              u64 chunk_root_bytenr,
1217                                              enum btrfs_open_ctree_flags flags)
1218 {
1219         struct btrfs_fs_info *fs_info;
1220         struct btrfs_super_block *disk_super;
1221         struct btrfs_fs_devices *fs_devices = NULL;
1222         struct extent_buffer *eb;
1223         int ret;
1224         int oflags;
1225
1226         if (sb_bytenr == 0)
1227                 sb_bytenr = BTRFS_SUPER_INFO_OFFSET;
1228
1229         /* try to drop all the caches */
1230         if (posix_fadvise(fp, 0, 0, POSIX_FADV_DONTNEED))
1231                 fprintf(stderr, "Warning, could not drop caches\n");
1232
1233         fs_info = btrfs_new_fs_info(flags & OPEN_CTREE_WRITES, sb_bytenr);
1234         if (!fs_info) {
1235                 fprintf(stderr, "Failed to allocate memory for fs_info\n");
1236                 return NULL;
1237         }
1238         if (flags & OPEN_CTREE_RESTORE)
1239                 fs_info->on_restoring = 1;
1240         if (flags & OPEN_CTREE_SUPPRESS_CHECK_BLOCK_ERRORS)
1241                 fs_info->suppress_check_block_errors = 1;
1242         if (flags & OPEN_CTREE_IGNORE_FSID_MISMATCH)
1243                 fs_info->ignore_fsid_mismatch = 1;
1244         if (flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR)
1245                 fs_info->ignore_chunk_tree_error = 1;
1246
1247         ret = btrfs_scan_fs_devices(fp, path, &fs_devices, sb_bytenr,
1248                                     (flags & OPEN_CTREE_RECOVER_SUPER),
1249                                     (flags & OPEN_CTREE_NO_DEVICES));
1250         if (ret)
1251                 goto out;
1252
1253         fs_info->fs_devices = fs_devices;
1254         if (flags & OPEN_CTREE_WRITES)
1255                 oflags = O_RDWR;
1256         else
1257                 oflags = O_RDONLY;
1258
1259         if (flags & OPEN_CTREE_EXCLUSIVE)
1260                 oflags |= O_EXCL;
1261
1262         ret = btrfs_open_devices(fs_devices, oflags);
1263         if (ret)
1264                 goto out;
1265
1266         disk_super = fs_info->super_copy;
1267         if (!(flags & OPEN_CTREE_RECOVER_SUPER))
1268                 ret = btrfs_read_dev_super(fs_devices->latest_bdev,
1269                                            disk_super, sb_bytenr, 1);
1270         else
1271                 ret = btrfs_read_dev_super(fp, disk_super, sb_bytenr, 0);
1272         if (ret) {
1273                 printk("No valid btrfs found\n");
1274                 goto out_devices;
1275         }
1276
1277         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID &&
1278             !fs_info->ignore_fsid_mismatch) {
1279                 fprintf(stderr, "ERROR: Filesystem UUID change in progress\n");
1280                 goto out_devices;
1281         }
1282
1283         memcpy(fs_info->fsid, &disk_super->fsid, BTRFS_FSID_SIZE);
1284
1285         ret = btrfs_check_fs_compatibility(fs_info->super_copy,
1286                                            flags & OPEN_CTREE_WRITES);
1287         if (ret)
1288                 goto out_devices;
1289
1290         ret = btrfs_setup_chunk_tree_and_device_map(fs_info, chunk_root_bytenr);
1291         if (ret)
1292                 goto out_chunk;
1293
1294         /* Chunk tree root is unable to read, return directly */
1295         if (!fs_info->chunk_root)
1296                 return fs_info;
1297
1298         eb = fs_info->chunk_root->node;
1299         read_extent_buffer(eb, fs_info->chunk_tree_uuid,
1300                            btrfs_header_chunk_tree_uuid(eb),
1301                            BTRFS_UUID_SIZE);
1302
1303         ret = btrfs_setup_all_roots(fs_info, root_tree_bytenr, flags);
1304         if (ret && !(flags & __OPEN_CTREE_RETURN_CHUNK_ROOT) &&
1305             !fs_info->ignore_chunk_tree_error)
1306                 goto out_chunk;
1307
1308         return fs_info;
1309
1310 out_chunk:
1311         btrfs_release_all_roots(fs_info);
1312         btrfs_cleanup_all_caches(fs_info);
1313 out_devices:
1314         btrfs_close_devices(fs_devices);
1315 out:
1316         btrfs_free_fs_info(fs_info);
1317         return NULL;
1318 }
1319
1320 struct btrfs_fs_info *open_ctree_fs_info(const char *filename,
1321                                          u64 sb_bytenr, u64 root_tree_bytenr,
1322                                          u64 chunk_root_bytenr,
1323                                          enum btrfs_open_ctree_flags flags)
1324 {
1325         int fp;
1326         struct btrfs_fs_info *info;
1327         int oflags = O_CREAT | O_RDWR;
1328
1329         if (!(flags & OPEN_CTREE_WRITES))
1330                 oflags = O_RDONLY;
1331
1332         fp = open(filename, oflags, 0600);
1333         if (fp < 0) {
1334                 fprintf (stderr, "Could not open %s\n", filename);
1335                 return NULL;
1336         }
1337         info = __open_ctree_fd(fp, filename, sb_bytenr, root_tree_bytenr,
1338                                chunk_root_bytenr, flags);
1339         close(fp);
1340         return info;
1341 }
1342
1343 struct btrfs_root *open_ctree(const char *filename, u64 sb_bytenr,
1344                               enum btrfs_open_ctree_flags flags)
1345 {
1346         struct btrfs_fs_info *info;
1347
1348         /* This flags may not return fs_info with any valid root */
1349         BUG_ON(flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR);
1350         info = open_ctree_fs_info(filename, sb_bytenr, 0, 0, flags);
1351         if (!info)
1352                 return NULL;
1353         if (flags & __OPEN_CTREE_RETURN_CHUNK_ROOT)
1354                 return info->chunk_root;
1355         return info->fs_root;
1356 }
1357
1358 struct btrfs_root *open_ctree_fd(int fp, const char *path, u64 sb_bytenr,
1359                                  enum btrfs_open_ctree_flags flags)
1360 {
1361         struct btrfs_fs_info *info;
1362
1363         /* This flags may not return fs_info with any valid root */
1364         BUG_ON(flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR);
1365         info = __open_ctree_fd(fp, path, sb_bytenr, 0, 0, flags);
1366         if (!info)
1367                 return NULL;
1368         if (flags & __OPEN_CTREE_RETURN_CHUNK_ROOT)
1369                 return info->chunk_root;
1370         return info->fs_root;
1371 }
1372
1373 /*
1374  * Check if the super is valid:
1375  * - nodesize/sectorsize - minimum, maximum, alignment
1376  * - tree block starts   - alignment
1377  * - number of devices   - something sane
1378  * - sys array size      - maximum
1379  */
1380 static int check_super(struct btrfs_super_block *sb)
1381 {
1382         char result[BTRFS_CSUM_SIZE];
1383         u32 crc;
1384         u16 csum_type;
1385         int csum_size;
1386
1387         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
1388                 fprintf(stderr, "ERROR: superblock magic doesn't match\n");
1389                 return -EIO;
1390         }
1391
1392         csum_type = btrfs_super_csum_type(sb);
1393         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
1394                 fprintf(stderr, "ERROR: unsupported checksum algorithm %u\n",
1395                         csum_type);
1396                 return -EIO;
1397         }
1398         csum_size = btrfs_csum_sizes[csum_type];
1399
1400         crc = ~(u32)0;
1401         crc = btrfs_csum_data(NULL, (char *)sb + BTRFS_CSUM_SIZE, crc,
1402                               BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1403         btrfs_csum_final(crc, result);
1404
1405         if (memcmp(result, sb->csum, csum_size)) {
1406                 fprintf(stderr, "ERROR: superblock checksum mismatch\n");
1407                 return -EIO;
1408         }
1409         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
1410                 fprintf(stderr, "ERROR: tree_root level too big: %d >= %d\n",
1411                         btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
1412                 return -EIO;
1413         }
1414         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
1415                 fprintf(stderr, "ERROR: chunk_root level too big: %d >= %d\n",
1416                         btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
1417                 return -EIO;
1418         }
1419         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
1420                 fprintf(stderr, "ERROR: log_root level too big: %d >= %d\n",
1421                         btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
1422                 return -EIO;
1423         }
1424
1425         if (!IS_ALIGNED(btrfs_super_root(sb), 4096)) {
1426                 fprintf(stderr, "ERROR: tree_root block unaligned: %llu\n",
1427                         btrfs_super_root(sb));
1428                 return -EIO;
1429         }
1430         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096)) {
1431                 fprintf(stderr, "ERROR: chunk_root block unaligned: %llu\n",
1432                         btrfs_super_chunk_root(sb));
1433                 return -EIO;
1434         }
1435         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096)) {
1436                 fprintf(stderr, "ERROR: log_root block unaligned: %llu\n",
1437                         btrfs_super_log_root(sb));
1438                 return -EIO;
1439         }
1440         if (btrfs_super_nodesize(sb) < 4096) {
1441                 fprintf(stderr, "ERROR: nodesize too small: %u < 4096\n",
1442                         btrfs_super_nodesize(sb));
1443                 return -EIO;
1444         }
1445         if (!IS_ALIGNED(btrfs_super_nodesize(sb), 4096)) {
1446                 fprintf(stderr, "ERROR: nodesize unaligned: %u\n",
1447                         btrfs_super_nodesize(sb));
1448                 return -EIO;
1449         }
1450         if (btrfs_super_sectorsize(sb) < 4096) {
1451                 fprintf(stderr, "ERROR: sectorsize too small: %u < 4096\n",
1452                         btrfs_super_sectorsize(sb));
1453                 return -EIO;
1454         }
1455         if (!IS_ALIGNED(btrfs_super_sectorsize(sb), 4096)) {
1456                 fprintf(stderr, "ERROR: sectorsize unaligned: %u\n",
1457                         btrfs_super_sectorsize(sb));
1458                 return -EIO;
1459         }
1460
1461         if (memcmp(sb->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
1462                 char fsid[BTRFS_UUID_UNPARSED_SIZE];
1463                 char dev_fsid[BTRFS_UUID_UNPARSED_SIZE];
1464
1465                 uuid_unparse(sb->fsid, fsid);
1466                 uuid_unparse(sb->dev_item.fsid, dev_fsid);
1467                 printk(KERN_ERR
1468                         "ERROR: dev_item UUID does not match fsid: %s != %s\n",
1469                         dev_fsid, fsid);
1470                 return -EIO;
1471         }
1472
1473         /*
1474          * Hint to catch really bogus numbers, bitflips or so
1475          */
1476         if (btrfs_super_num_devices(sb) > (1UL << 31)) {
1477                 fprintf(stderr, "WARNING: suspicious number of devices: %llu\n",
1478                         btrfs_super_num_devices(sb));
1479         }
1480
1481         if (btrfs_super_num_devices(sb) == 0) {
1482                 fprintf(stderr, "ERROR: number of devices is 0\n");
1483                 return -EIO;
1484         }
1485
1486         /*
1487          * Obvious sys_chunk_array corruptions, it must hold at least one key
1488          * and one chunk
1489          */
1490         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
1491                 fprintf(stderr, "BTRFS: system chunk array too big %u > %u\n",
1492                         btrfs_super_sys_array_size(sb),
1493                         BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
1494                 return -EIO;
1495         }
1496         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
1497                         + sizeof(struct btrfs_chunk)) {
1498                 fprintf(stderr, "BTRFS: system chunk array too small %u < %lu\n",
1499                         btrfs_super_sys_array_size(sb),
1500                         sizeof(struct btrfs_disk_key) +
1501                         sizeof(struct btrfs_chunk));
1502                 return -EIO;
1503         }
1504
1505         return 0;
1506 }
1507
1508 int btrfs_read_dev_super(int fd, struct btrfs_super_block *sb, u64 sb_bytenr,
1509                          int super_recover)
1510 {
1511         u8 fsid[BTRFS_FSID_SIZE];
1512         int fsid_is_initialized = 0;
1513         char tmp[BTRFS_SUPER_INFO_SIZE];
1514         struct btrfs_super_block *buf = (struct btrfs_super_block *)tmp;
1515         int i;
1516         int ret;
1517         int max_super = super_recover ? BTRFS_SUPER_MIRROR_MAX : 1;
1518         u64 transid = 0;
1519         u64 bytenr;
1520
1521         if (sb_bytenr != BTRFS_SUPER_INFO_OFFSET) {
1522                 ret = pread64(fd, buf, BTRFS_SUPER_INFO_SIZE, sb_bytenr);
1523                 if (ret < BTRFS_SUPER_INFO_SIZE)
1524                         return -1;
1525
1526                 if (btrfs_super_bytenr(buf) != sb_bytenr)
1527                         return -1;
1528
1529                 if (check_super(buf))
1530                         return -1;
1531                 memcpy(sb, buf, BTRFS_SUPER_INFO_SIZE);
1532                 return 0;
1533         }
1534
1535         /*
1536         * we would like to check all the supers, but that would make
1537         * a btrfs mount succeed after a mkfs from a different FS.
1538         * So, we need to add a special mount option to scan for
1539         * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1540         */
1541
1542         for (i = 0; i < max_super; i++) {
1543                 bytenr = btrfs_sb_offset(i);
1544                 ret = pread64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1545                 if (ret < BTRFS_SUPER_INFO_SIZE)
1546                         break;
1547
1548                 if (btrfs_super_bytenr(buf) != bytenr )
1549                         continue;
1550                 /* if magic is NULL, the device was removed */
1551                 if (btrfs_super_magic(buf) == 0 && i == 0)
1552                         break;
1553                 if (check_super(buf))
1554                         continue;
1555
1556                 if (!fsid_is_initialized) {
1557                         memcpy(fsid, buf->fsid, sizeof(fsid));
1558                         fsid_is_initialized = 1;
1559                 } else if (memcmp(fsid, buf->fsid, sizeof(fsid))) {
1560                         /*
1561                          * the superblocks (the original one and
1562                          * its backups) contain data of different
1563                          * filesystems -> the super cannot be trusted
1564                          */
1565                         continue;
1566                 }
1567
1568                 if (btrfs_super_generation(buf) > transid) {
1569                         memcpy(sb, buf, BTRFS_SUPER_INFO_SIZE);
1570                         transid = btrfs_super_generation(buf);
1571                 }
1572         }
1573
1574         return transid > 0 ? 0 : -1;
1575 }
1576
1577 static int write_dev_supers(struct btrfs_root *root,
1578                             struct btrfs_super_block *sb,
1579                             struct btrfs_device *device)
1580 {
1581         u64 bytenr;
1582         u32 crc;
1583         int i, ret;
1584
1585         if (root->fs_info->super_bytenr != BTRFS_SUPER_INFO_OFFSET) {
1586                 btrfs_set_super_bytenr(sb, root->fs_info->super_bytenr);
1587                 crc = ~(u32)0;
1588                 crc = btrfs_csum_data(NULL, (char *)sb + BTRFS_CSUM_SIZE, crc,
1589                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1590                 btrfs_csum_final(crc, (char *)&sb->csum[0]);
1591
1592                 /*
1593                  * super_copy is BTRFS_SUPER_INFO_SIZE bytes and is
1594                  * zero filled, we can use it directly
1595                  */
1596                 ret = pwrite64(device->fd, root->fs_info->super_copy,
1597                                 BTRFS_SUPER_INFO_SIZE,
1598                                 root->fs_info->super_bytenr);
1599                 if (ret != BTRFS_SUPER_INFO_SIZE)
1600                         goto write_err;
1601                 return 0;
1602         }
1603
1604         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1605                 bytenr = btrfs_sb_offset(i);
1606                 if (bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
1607                         break;
1608
1609                 btrfs_set_super_bytenr(sb, bytenr);
1610
1611                 crc = ~(u32)0;
1612                 crc = btrfs_csum_data(NULL, (char *)sb + BTRFS_CSUM_SIZE, crc,
1613                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1614                 btrfs_csum_final(crc, (char *)&sb->csum[0]);
1615
1616                 /*
1617                  * super_copy is BTRFS_SUPER_INFO_SIZE bytes and is
1618                  * zero filled, we can use it directly
1619                  */
1620                 ret = pwrite64(device->fd, root->fs_info->super_copy,
1621                                 BTRFS_SUPER_INFO_SIZE, bytenr);
1622                 if (ret != BTRFS_SUPER_INFO_SIZE)
1623                         goto write_err;
1624         }
1625
1626         return 0;
1627
1628 write_err:
1629         if (ret > 0)
1630                 fprintf(stderr, "WARNING: failed to write all sb data\n");
1631         else
1632                 fprintf(stderr, "WARNING: failed to write sb: %s\n",
1633                         strerror(errno));
1634         return ret;
1635 }
1636
1637 int write_all_supers(struct btrfs_root *root)
1638 {
1639         struct list_head *cur;
1640         struct list_head *head = &root->fs_info->fs_devices->devices;
1641         struct btrfs_device *dev;
1642         struct btrfs_super_block *sb;
1643         struct btrfs_dev_item *dev_item;
1644         int ret;
1645         u64 flags;
1646
1647         sb = root->fs_info->super_copy;
1648         dev_item = &sb->dev_item;
1649         list_for_each(cur, head) {
1650                 dev = list_entry(cur, struct btrfs_device, dev_list);
1651                 if (!dev->writeable)
1652                         continue;
1653
1654                 btrfs_set_stack_device_generation(dev_item, 0);
1655                 btrfs_set_stack_device_type(dev_item, dev->type);
1656                 btrfs_set_stack_device_id(dev_item, dev->devid);
1657                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1658                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1659                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1660                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1661                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1662                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1663                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1664
1665                 flags = btrfs_super_flags(sb);
1666                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1667
1668                 ret = write_dev_supers(root, sb, dev);
1669                 BUG_ON(ret);
1670         }
1671         return 0;
1672 }
1673
1674 int write_ctree_super(struct btrfs_trans_handle *trans,
1675                       struct btrfs_root *root)
1676 {
1677         int ret;
1678         struct btrfs_root *tree_root = root->fs_info->tree_root;
1679         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1680
1681         if (root->fs_info->readonly)
1682                 return 0;
1683
1684         btrfs_set_super_generation(root->fs_info->super_copy,
1685                                    trans->transid);
1686         btrfs_set_super_root(root->fs_info->super_copy,
1687                              tree_root->node->start);
1688         btrfs_set_super_root_level(root->fs_info->super_copy,
1689                                    btrfs_header_level(tree_root->node));
1690         btrfs_set_super_chunk_root(root->fs_info->super_copy,
1691                                    chunk_root->node->start);
1692         btrfs_set_super_chunk_root_level(root->fs_info->super_copy,
1693                                          btrfs_header_level(chunk_root->node));
1694         btrfs_set_super_chunk_root_generation(root->fs_info->super_copy,
1695                                 btrfs_header_generation(chunk_root->node));
1696
1697         ret = write_all_supers(root);
1698         if (ret)
1699                 fprintf(stderr, "failed to write new super block err %d\n", ret);
1700         return ret;
1701 }
1702
1703 int close_ctree_fs_info(struct btrfs_fs_info *fs_info)
1704 {
1705         int ret;
1706         struct btrfs_trans_handle *trans;
1707         struct btrfs_root *root = fs_info->tree_root;
1708
1709         if (fs_info->last_trans_committed !=
1710             fs_info->generation) {
1711                 BUG_ON(!root);
1712                 trans = btrfs_start_transaction(root, 1);
1713                 btrfs_commit_transaction(trans, root);
1714                 trans = btrfs_start_transaction(root, 1);
1715                 ret = commit_tree_roots(trans, fs_info);
1716                 BUG_ON(ret);
1717                 ret = __commit_transaction(trans, root);
1718                 BUG_ON(ret);
1719                 write_ctree_super(trans, root);
1720                 btrfs_free_transaction(root, trans);
1721         }
1722         btrfs_free_block_groups(fs_info);
1723
1724         free_fs_roots_tree(&fs_info->fs_root_tree);
1725
1726         btrfs_release_all_roots(fs_info);
1727         btrfs_close_devices(fs_info->fs_devices);
1728         btrfs_cleanup_all_caches(fs_info);
1729         btrfs_free_fs_info(fs_info);
1730         return 0;
1731 }
1732
1733 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1734                      struct extent_buffer *eb)
1735 {
1736         return clear_extent_buffer_dirty(eb);
1737 }
1738
1739 int wait_on_tree_block_writeback(struct btrfs_root *root,
1740                                  struct extent_buffer *eb)
1741 {
1742         return 0;
1743 }
1744
1745 void btrfs_mark_buffer_dirty(struct extent_buffer *eb)
1746 {
1747         set_extent_buffer_dirty(eb);
1748 }
1749
1750 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1751 {
1752         int ret;
1753
1754         ret = extent_buffer_uptodate(buf);
1755         if (!ret)
1756                 return ret;
1757
1758         ret = verify_parent_transid(buf->tree, buf, parent_transid, 1);
1759         return !ret;
1760 }
1761
1762 int btrfs_set_buffer_uptodate(struct extent_buffer *eb)
1763 {
1764         return set_extent_buffer_uptodate(eb);
1765 }