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_fs_info *fs_info,
 178                                             u64 bytenr, u32 blocksize)
 179 {
 180         return find_extent_buffer(&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)
 186 {
 187         return alloc_extent_buffer(fs_info, bytenr, fs_info->nodesize);
 188 }
 189
 190 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
 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(fs_info, bytenr, fs_info->nodesize);
 199         if (!(eb && btrfs_buffer_uptodate(eb, parent_transid)) &&
 200             !btrfs_map_block(fs_info, READ, bytenr, &length, &multi, 0,
 201                              NULL)) {
 202                 device = multi->stripes[0].dev;
 203                 device->total_ios++;
 204                 readahead(device->fd, multi->stripes[0].physical,
 205                                 fs_info->nodesize);
 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(struct btrfs_fs_info *fs_info, u64 bytenr,
 303                 u64 parent_transid)
 304 {
 305         int ret;
 306         struct extent_buffer *eb;
 307         u64 best_transid = 0;
 308         u32 sectorsize = fs_info->sectorsize;
 309         int mirror_num = 0;
 310         int good_mirror = 0;
 311         int num_copies;
 312         int ignore = 0;
 313
 314         /*
 315          * Don't even try to create tree block for unaligned tree block
 316          * bytenr.
 317          * Such unaligned tree block will free overlapping extent buffer,
 318          * causing use-after-free bugs for fuzzed images.
 319          */
 320         if (bytenr < sectorsize || !IS_ALIGNED(bytenr, sectorsize)) {
 321                 error("tree block bytenr %llu is not aligned to sectorsize %u",
 322                       bytenr, sectorsize);
 323                 return ERR_PTR(-EIO);
 324         }
 325
 326         eb = btrfs_find_create_tree_block(fs_info, bytenr);
 327         if (!eb)
 328                 return ERR_PTR(-ENOMEM);
 329
 330         if (btrfs_buffer_uptodate(eb, parent_transid))
 331                 return eb;
 332
 333         while (1) {
 334                 ret = read_whole_eb(fs_info, eb, mirror_num);
 335                 if (ret == 0 && csum_tree_block(fs_info, eb, 1) == 0 &&
 336                     check_tree_block(fs_info, eb) == 0 &&
 337                     verify_parent_transid(eb->tree, eb, parent_transid, ignore)
 338                     == 0) {
 339                         if (eb->flags & EXTENT_BAD_TRANSID &&
 340                             list_empty(&eb->recow)) {
 341                                 list_add_tail(&eb->recow,
 342                                               &fs_info->recow_ebs);
 343                                 eb->refs++;
 344                         }
 345                         btrfs_set_buffer_uptodate(eb);
 346                         return eb;
 347                 }
 348                 if (ignore) {
 349                         if (check_tree_block(fs_info, eb)) {
 350                                 if (!fs_info->suppress_check_block_errors)
 351                                         print_tree_block_error(fs_info, eb,
 352                                                 check_tree_block(fs_info, eb));
 353                         } else {
 354                                 if (!fs_info->suppress_check_block_errors)
 355                                         fprintf(stderr, "Csum didn't match\n");
 356                         }
 357                         ret = -EIO;
 358                         break;
 359                 }
 360                 num_copies = btrfs_num_copies(fs_info, eb->start, eb->len);
 361                 if (num_copies == 1) {
 362                         ignore = 1;
 363                         continue;
 364                 }
 365                 if (btrfs_header_generation(eb) > best_transid && mirror_num) {
 366                         best_transid = btrfs_header_generation(eb);
 367                         good_mirror = mirror_num;
 368                 }
 369                 mirror_num++;
 370                 if (mirror_num > num_copies) {
 371                         mirror_num = good_mirror;
 372                         ignore = 1;
 373                         continue;
 374                 }
 375         }
 376         free_extent_buffer(eb);
 377         return ERR_PTR(ret);
 378 }
 379
 380 int read_extent_data(struct btrfs_fs_info *fs_info, char *data, u64 logical,
 381                      u64 *len, int mirror)
 382 {
 383         u64 offset = 0;
 384         struct btrfs_multi_bio *multi = NULL;
 385         struct btrfs_device *device;
 386         int ret = 0;
 387         u64 max_len = *len;
 388
 389         ret = btrfs_map_block(fs_info, READ, logical, len, &multi, mirror,
 390                               NULL);
 391         if (ret) {
 392                 fprintf(stderr, "Couldn't map the block %llu\n",
 393                                 logical + offset);
 394                 goto err;
 395         }
 396         device = multi->stripes[0].dev;
 397
 398         if (*len > max_len)
 399                 *len = max_len;
 400         if (device->fd < 0) {
 401                 ret = -EIO;
 402                 goto err;
 403         }
 404
 405         ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
 406         if (ret != *len)
 407                 ret = -EIO;
 408         else
 409                 ret = 0;
 410 err:
 411         kfree(multi);
 412         return ret;
 413 }
 414
 415 int write_and_map_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
 416 {
 417         int ret;
 418         int dev_nr;
 419         u64 length;
 420         u64 *raid_map = NULL;
 421         struct btrfs_multi_bio *multi = NULL;
 422
 423         dev_nr = 0;
 424         length = eb->len;
 425         ret = btrfs_map_block(fs_info, WRITE, eb->start, &length,
 426                               &multi, 0, &raid_map);
 427
 428         if (raid_map) {
 429                 ret = write_raid56_with_parity(fs_info, eb, multi,
 430                                                length, raid_map);
 431                 BUG_ON(ret);
 432         } else while (dev_nr < multi->num_stripes) {
 433                 BUG_ON(ret);
 434                 eb->fd = multi->stripes[dev_nr].dev->fd;
 435                 eb->dev_bytenr = multi->stripes[dev_nr].physical;
 436                 multi->stripes[dev_nr].dev->total_ios++;
 437                 dev_nr++;
 438                 ret = write_extent_to_disk(eb);
 439                 BUG_ON(ret);
 440         }
 441         kfree(raid_map);
 442         kfree(multi);
 443         return 0;
 444 }
 445
 446 int write_tree_block(struct btrfs_trans_handle *trans,
 447                      struct btrfs_fs_info *fs_info,
 448                      struct extent_buffer *eb)
 449 {
 450         if (check_tree_block(fs_info, eb)) {
 451                 print_tree_block_error(fs_info, eb,
 452                                 check_tree_block(fs_info, eb));
 453                 BUG();
 454         }
 455
 456         if (trans && !btrfs_buffer_uptodate(eb, trans->transid))
 457                 BUG();
 458
 459         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 460         csum_tree_block(fs_info, eb, 0);
 461
 462         return write_and_map_eb(fs_info, eb);
 463 }
 464
 465 void btrfs_setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
 466                       u64 objectid)
 467 {
 468         root->node = NULL;
 469         root->commit_root = NULL;
 470         root->ref_cows = 0;
 471         root->track_dirty = 0;
 472
 473         root->fs_info = fs_info;
 474         root->objectid = objectid;
 475         root->last_trans = 0;
 476         root->last_inode_alloc = 0;
 477
 478         INIT_LIST_HEAD(&root->dirty_list);
 479         INIT_LIST_HEAD(&root->orphan_data_extents);
 480         memset(&root->root_key, 0, sizeof(root->root_key));
 481         memset(&root->root_item, 0, sizeof(root->root_item));
 482         root->root_key.objectid = objectid;
 483 }
 484
 485 static int find_and_setup_root(struct btrfs_root *tree_root,
 486                                struct btrfs_fs_info *fs_info,
 487                                u64 objectid, struct btrfs_root *root)
 488 {
 489         int ret;
 490         u64 generation;
 491
 492         btrfs_setup_root(root, fs_info, objectid);
 493         ret = btrfs_find_last_root(tree_root, objectid,
 494                                    &root->root_item, &root->root_key);
 495         if (ret)
 496                 return ret;
 497
 498         generation = btrfs_root_generation(&root->root_item);
 499         root->node = read_tree_block(fs_info,
 500                         btrfs_root_bytenr(&root->root_item), generation);
 501         if (!extent_buffer_uptodate(root->node))
 502                 return -EIO;
 503
 504         return 0;
 505 }
 506
 507 static int find_and_setup_log_root(struct btrfs_root *tree_root,
 508                                struct btrfs_fs_info *fs_info,
 509                                struct btrfs_super_block *disk_super)
 510 {
 511         u64 blocknr = btrfs_super_log_root(disk_super);
 512         struct btrfs_root *log_root = malloc(sizeof(struct btrfs_root));
 513
 514         if (!log_root)
 515                 return -ENOMEM;
 516
 517         if (blocknr == 0) {
 518                 free(log_root);
 519                 return 0;
 520         }
 521
 522         btrfs_setup_root(log_root, fs_info,
 523                          BTRFS_TREE_LOG_OBJECTID);
 524
 525         log_root->node = read_tree_block(fs_info, blocknr,
 526                                      btrfs_super_generation(disk_super) + 1);
 527
 528         fs_info->log_root_tree = log_root;
 529
 530         if (!extent_buffer_uptodate(log_root->node)) {
 531                 free_extent_buffer(log_root->node);
 532                 free(log_root);
 533                 fs_info->log_root_tree = NULL;
 534                 return -EIO;
 535         }
 536
 537         return 0;
 538 }
 539
 540 int btrfs_free_fs_root(struct btrfs_root *root)
 541 {
 542         if (root->node)
 543                 free_extent_buffer(root->node);
 544         if (root->commit_root)
 545                 free_extent_buffer(root->commit_root);
 546         kfree(root);
 547         return 0;
 548 }
 549
 550 static void __free_fs_root(struct rb_node *node)
 551 {
 552         struct btrfs_root *root;
 553
 554         root = container_of(node, struct btrfs_root, rb_node);
 555         btrfs_free_fs_root(root);
 556 }
 557
 558 FREE_RB_BASED_TREE(fs_roots, __free_fs_root);
 559
 560 struct btrfs_root *btrfs_read_fs_root_no_cache(struct btrfs_fs_info *fs_info,
 561                                                struct btrfs_key *location)
 562 {
 563         struct btrfs_root *root;
 564         struct btrfs_root *tree_root = fs_info->tree_root;
 565         struct btrfs_path *path;
 566         struct extent_buffer *l;
 567         u64 generation;
 568         int ret = 0;
 569
 570         root = calloc(1, sizeof(*root));
 571         if (!root)
 572                 return ERR_PTR(-ENOMEM);
 573         if (location->offset == (u64)-1) {
 574                 ret = find_and_setup_root(tree_root, fs_info,
 575                                           location->objectid, root);
 576                 if (ret) {
 577                         free(root);
 578                         return ERR_PTR(ret);
 579                 }
 580                 goto insert;
 581         }
 582
 583         btrfs_setup_root(root, fs_info,
 584                          location->objectid);
 585
 586         path = btrfs_alloc_path();
 587         if (!path) {
 588                 free(root);
 589                 return ERR_PTR(-ENOMEM);
 590         }
 591
 592         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 593         if (ret != 0) {
 594                 if (ret > 0)
 595                         ret = -ENOENT;
 596                 goto out;
 597         }
 598         l = path->nodes[0];
 599         read_extent_buffer(l, &root->root_item,
 600                btrfs_item_ptr_offset(l, path->slots[0]),
 601                sizeof(root->root_item));
 602         memcpy(&root->root_key, location, sizeof(*location));
 603         ret = 0;
 604 out:
 605         btrfs_free_path(path);
 606         if (ret) {
 607                 free(root);
 608                 return ERR_PTR(ret);
 609         }
 610         generation = btrfs_root_generation(&root->root_item);
 611         root->node = read_tree_block(fs_info,
 612                         btrfs_root_bytenr(&root->root_item), generation);
 613         if (!extent_buffer_uptodate(root->node)) {
 614                 free(root);
 615                 return ERR_PTR(-EIO);
 616         }
 617 insert:
 618         root->ref_cows = 1;
 619         return root;
 620 }
 621
 622 static int btrfs_fs_roots_compare_objectids(struct rb_node *node,
 623                                             void *data)
 624 {
 625         u64 objectid = *((u64 *)data);
 626         struct btrfs_root *root;
 627
 628         root = rb_entry(node, struct btrfs_root, rb_node);
 629         if (objectid > root->objectid)
 630                 return 1;
 631         else if (objectid < root->objectid)
 632                 return -1;
 633         else
 634                 return 0;
 635 }
 636
 637 static int btrfs_fs_roots_compare_roots(struct rb_node *node1,
 638                                         struct rb_node *node2)
 639 {
 640         struct btrfs_root *root;
 641
 642         root = rb_entry(node2, struct btrfs_root, rb_node);
 643         return btrfs_fs_roots_compare_objectids(node1, (void *)&root->objectid);
 644 }
 645
 646 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 647                                       struct btrfs_key *location)
 648 {
 649         struct btrfs_root *root;
 650         struct rb_node *node;
 651         int ret;
 652         u64 objectid = location->objectid;
 653
 654         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 655                 return fs_info->tree_root;
 656         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 657                 return fs_info->extent_root;
 658         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 659                 return fs_info->chunk_root;
 660         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 661                 return fs_info->dev_root;
 662         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
 663                 return fs_info->csum_root;
 664         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
 665                 return fs_info->quota_enabled ? fs_info->quota_root :
 666                                 ERR_PTR(-ENOENT);
 667
 668         BUG_ON(location->objectid == BTRFS_TREE_RELOC_OBJECTID ||
 669                location->offset != (u64)-1);
 670
 671         node = rb_search(&fs_info->fs_root_tree, (void *)&objectid,
 672                          btrfs_fs_roots_compare_objectids, NULL);
 673         if (node)
 674                 return container_of(node, struct btrfs_root, rb_node);
 675
 676         root = btrfs_read_fs_root_no_cache(fs_info, location);
 677         if (IS_ERR(root))
 678                 return root;
 679
 680         ret = rb_insert(&fs_info->fs_root_tree, &root->rb_node,
 681                         btrfs_fs_roots_compare_roots);
 682         BUG_ON(ret);
 683         return root;
 684 }
 685
 686 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
 687 {
 688         if (fs_info->quota_root)
 689                 free(fs_info->quota_root);
 690
 691         free(fs_info->tree_root);
 692         free(fs_info->extent_root);
 693         free(fs_info->chunk_root);
 694         free(fs_info->dev_root);
 695         free(fs_info->csum_root);
 696         free(fs_info->free_space_root);
 697         free(fs_info->super_copy);
 698         free(fs_info->log_root_tree);
 699         free(fs_info);
 700 }
 701
 702 struct btrfs_fs_info *btrfs_new_fs_info(int writable, u64 sb_bytenr)
 703 {
 704         struct btrfs_fs_info *fs_info;
 705
 706         fs_info = calloc(1, sizeof(struct btrfs_fs_info));
 707         if (!fs_info)
 708                 return NULL;
 709
 710         fs_info->tree_root = calloc(1, sizeof(struct btrfs_root));
 711         fs_info->extent_root = calloc(1, sizeof(struct btrfs_root));
 712         fs_info->chunk_root = calloc(1, sizeof(struct btrfs_root));
 713         fs_info->dev_root = calloc(1, sizeof(struct btrfs_root));
 714         fs_info->csum_root = calloc(1, sizeof(struct btrfs_root));
 715         fs_info->quota_root = calloc(1, sizeof(struct btrfs_root));
 716         fs_info->free_space_root = calloc(1, sizeof(struct btrfs_root));
 717         fs_info->super_copy = calloc(1, BTRFS_SUPER_INFO_SIZE);
 718
 719         if (!fs_info->tree_root || !fs_info->extent_root ||
 720             !fs_info->chunk_root || !fs_info->dev_root ||
 721             !fs_info->csum_root || !fs_info->quota_root ||
 722             !fs_info->free_space_root || !fs_info->super_copy)
 723                 goto free_all;
 724
 725         extent_io_tree_init(&fs_info->extent_cache);
 726         extent_io_tree_init(&fs_info->free_space_cache);
 727         extent_io_tree_init(&fs_info->block_group_cache);
 728         extent_io_tree_init(&fs_info->pinned_extents);
 729         extent_io_tree_init(&fs_info->pending_del);
 730         extent_io_tree_init(&fs_info->extent_ins);
 731         fs_info->excluded_extents = NULL;
 732
 733         fs_info->fs_root_tree = RB_ROOT;
 734         cache_tree_init(&fs_info->mapping_tree.cache_tree);
 735
 736         mutex_init(&fs_info->fs_mutex);
 737         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
 738         INIT_LIST_HEAD(&fs_info->space_info);
 739         INIT_LIST_HEAD(&fs_info->recow_ebs);
 740
 741         if (!writable)
 742                 fs_info->readonly = 1;
 743
 744         fs_info->super_bytenr = sb_bytenr;
 745         fs_info->data_alloc_profile = (u64)-1;
 746         fs_info->metadata_alloc_profile = (u64)-1;
 747         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
 748         return fs_info;
 749 free_all:
 750         btrfs_free_fs_info(fs_info);
 751         return NULL;
 752 }
 753
 754 int btrfs_check_fs_compatibility(struct btrfs_super_block *sb,
 755                                  unsigned int flags)
 756 {
 757         u64 features;
 758
 759         features = btrfs_super_incompat_flags(sb) &
 760                    ~BTRFS_FEATURE_INCOMPAT_SUPP;
 761         if (features) {
 762                 printk("couldn't open because of unsupported "
 763                        "option features (%Lx).\n",
 764                        (unsigned long long)features);
 765                 return -ENOTSUP;
 766         }
 767
 768         features = btrfs_super_incompat_flags(sb);
 769         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
 770                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
 771                 btrfs_set_super_incompat_flags(sb, features);
 772         }
 773
 774         features = btrfs_super_compat_ro_flags(sb);
 775         if (flags & OPEN_CTREE_WRITES) {
 776                 if (flags & OPEN_CTREE_INVALIDATE_FST) {
 777                         /* Clear the FREE_SPACE_TREE_VALID bit on disk... */
 778                         features &= ~BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID;
 779                         btrfs_set_super_compat_ro_flags(sb, features);
 780                         /* ... and ignore the free space tree bit. */
 781                         features &= ~BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE;
 782                 }
 783                 if (features & ~BTRFS_FEATURE_COMPAT_RO_SUPP) {
 784                         printk("couldn't open RDWR because of unsupported "
 785                                "option features (%Lx).\n",
 786                                (unsigned long long)features);
 787                         return -ENOTSUP;
 788                 }
 789
 790         }
 791         return 0;
 792 }
 793
 794 static int find_best_backup_root(struct btrfs_super_block *super)
 795 {
 796         struct btrfs_root_backup *backup;
 797         u64 orig_gen = btrfs_super_generation(super);
 798         u64 gen = 0;
 799         int best_index = 0;
 800         int i;
 801
 802         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
 803                 backup = super->super_roots + i;
 804                 if (btrfs_backup_tree_root_gen(backup) != orig_gen &&
 805                     btrfs_backup_tree_root_gen(backup) > gen) {
 806                         best_index = i;
 807                         gen = btrfs_backup_tree_root_gen(backup);
 808                 }
 809         }
 810         return best_index;
 811 }
 812
 813 static int setup_root_or_create_block(struct btrfs_fs_info *fs_info,
 814                                       unsigned flags,
 815                                       struct btrfs_root *info_root,
 816                                       u64 objectid, char *str)
 817 {
 818         struct btrfs_root *root = fs_info->tree_root;
 819         int ret;
 820
 821         ret = find_and_setup_root(root, fs_info, objectid, info_root);
 822         if (ret) {
 823                 printk("Couldn't setup %s tree\n", str);
 824                 if (!(flags & OPEN_CTREE_PARTIAL))
 825                         return -EIO;
 826                 /*
 827                  * Need a blank node here just so we don't screw up in the
 828                  * million of places that assume a root has a valid ->node
 829                  */
 830                 info_root->node =
 831                         btrfs_find_create_tree_block(fs_info, 0);
 832                 if (!info_root->node)
 833                         return -ENOMEM;
 834                 clear_extent_buffer_uptodate(info_root->node);
 835         }
 836
 837         return 0;
 838 }
 839
 840 int btrfs_setup_all_roots(struct btrfs_fs_info *fs_info, u64 root_tree_bytenr,
 841                           unsigned flags)
 842 {
 843         struct btrfs_super_block *sb = fs_info->super_copy;
 844         struct btrfs_root *root;
 845         struct btrfs_key key;
 846         u64 generation;
 847         int ret;
 848
 849         root = fs_info->tree_root;
 850         btrfs_setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
 851         generation = btrfs_super_generation(sb);
 852
 853         if (!root_tree_bytenr && !(flags & OPEN_CTREE_BACKUP_ROOT)) {
 854                 root_tree_bytenr = btrfs_super_root(sb);
 855         } else if (flags & OPEN_CTREE_BACKUP_ROOT) {
 856                 struct btrfs_root_backup *backup;
 857                 int index = find_best_backup_root(sb);
 858                 if (index >= BTRFS_NUM_BACKUP_ROOTS) {
 859                         fprintf(stderr, "Invalid backup root number\n");
 860                         return -EIO;
 861                 }
 862                 backup = fs_info->super_copy->super_roots + index;
 863                 root_tree_bytenr = btrfs_backup_tree_root(backup);
 864                 generation = btrfs_backup_tree_root_gen(backup);
 865         }
 866
 867         root->node = read_tree_block(fs_info, root_tree_bytenr, generation);
 868         if (!extent_buffer_uptodate(root->node)) {
 869                 fprintf(stderr, "Couldn't read tree root\n");
 870                 return -EIO;
 871         }
 872
 873         ret = setup_root_or_create_block(fs_info, flags, fs_info->extent_root,
 874                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
 875         if (ret)
 876                 return ret;
 877         fs_info->extent_root->track_dirty = 1;
 878
 879         ret = find_and_setup_root(root, fs_info, BTRFS_DEV_TREE_OBJECTID,
 880                                   fs_info->dev_root);
 881         if (ret) {
 882                 printk("Couldn't setup device tree\n");
 883                 return -EIO;
 884         }
 885         fs_info->dev_root->track_dirty = 1;
 886
 887         ret = setup_root_or_create_block(fs_info, flags, fs_info->csum_root,
 888                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
 889         if (ret)
 890                 return ret;
 891         fs_info->csum_root->track_dirty = 1;
 892
 893         ret = find_and_setup_root(root, fs_info, BTRFS_QUOTA_TREE_OBJECTID,
 894                                   fs_info->quota_root);
 895         if (ret) {
 896                 free(fs_info->quota_root);
 897                 fs_info->quota_root = NULL;
 898         } else {
 899                 fs_info->quota_enabled = 1;
 900         }
 901
 902         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
 903                 ret = find_and_setup_root(root, fs_info, BTRFS_FREE_SPACE_TREE_OBJECTID,
 904                                           fs_info->free_space_root);
 905                 if (ret) {
 906                         printk("Couldn't read free space tree\n");
 907                         return -EIO;
 908                 }
 909                 fs_info->free_space_root->track_dirty = 1;
 910         }
 911
 912         ret = find_and_setup_log_root(root, fs_info, sb);
 913         if (ret) {
 914                 printk("Couldn't setup log root tree\n");
 915                 if (!(flags & OPEN_CTREE_PARTIAL))
 916                         return -EIO;
 917         }
 918
 919         fs_info->generation = generation;
 920         fs_info->last_trans_committed = generation;
 921         if (extent_buffer_uptodate(fs_info->extent_root->node) &&
 922             !(flags & OPEN_CTREE_NO_BLOCK_GROUPS)) {
 923                 ret = btrfs_read_block_groups(fs_info->tree_root);
 924                 /*
 925                  * If we don't find any blockgroups (ENOENT) we're either
 926                  * restoring or creating the filesystem, where it's expected,
 927                  * anything else is error
 928                  */
 929                 if (ret != -ENOENT)
 930                         return -EIO;
 931         }
 932
 933         key.objectid = BTRFS_FS_TREE_OBJECTID;
 934         key.type = BTRFS_ROOT_ITEM_KEY;
 935         key.offset = (u64)-1;
 936         fs_info->fs_root = btrfs_read_fs_root(fs_info, &key);
 937
 938         if (IS_ERR(fs_info->fs_root))
 939                 return -EIO;
 940         return 0;
 941 }
 942
 943 void btrfs_release_all_roots(struct btrfs_fs_info *fs_info)
 944 {
 945         if (fs_info->free_space_root)
 946                 free_extent_buffer(fs_info->free_space_root->node);
 947         if (fs_info->quota_root)
 948                 free_extent_buffer(fs_info->quota_root->node);
 949         if (fs_info->csum_root)
 950                 free_extent_buffer(fs_info->csum_root->node);
 951         if (fs_info->dev_root)
 952                 free_extent_buffer(fs_info->dev_root->node);
 953         if (fs_info->extent_root)
 954                 free_extent_buffer(fs_info->extent_root->node);
 955         if (fs_info->tree_root)
 956                 free_extent_buffer(fs_info->tree_root->node);
 957         if (fs_info->log_root_tree)
 958                 free_extent_buffer(fs_info->log_root_tree->node);
 959         if (fs_info->chunk_root)
 960                 free_extent_buffer(fs_info->chunk_root->node);
 961 }
 962
 963 static void free_map_lookup(struct cache_extent *ce)
 964 {
 965         struct map_lookup *map;
 966
 967         map = container_of(ce, struct map_lookup, ce);
 968         kfree(map);
 969 }
 970
 971 FREE_EXTENT_CACHE_BASED_TREE(mapping_cache, free_map_lookup);
 972
 973 void btrfs_cleanup_all_caches(struct btrfs_fs_info *fs_info)
 974 {
 975         while (!list_empty(&fs_info->recow_ebs)) {
 976                 struct extent_buffer *eb;
 977                 eb = list_first_entry(&fs_info->recow_ebs,
 978                                       struct extent_buffer, recow);
 979                 list_del_init(&eb->recow);
 980                 free_extent_buffer(eb);
 981         }
 982         free_mapping_cache_tree(&fs_info->mapping_tree.cache_tree);
 983         extent_io_tree_cleanup(&fs_info->extent_cache);
 984         extent_io_tree_cleanup(&fs_info->free_space_cache);
 985         extent_io_tree_cleanup(&fs_info->block_group_cache);
 986         extent_io_tree_cleanup(&fs_info->pinned_extents);
 987         extent_io_tree_cleanup(&fs_info->pending_del);
 988         extent_io_tree_cleanup(&fs_info->extent_ins);
 989 }
 990
 991 int btrfs_scan_fs_devices(int fd, const char *path,
 992                           struct btrfs_fs_devices **fs_devices,
 993                           u64 sb_bytenr, unsigned sbflags,
 994                           int skip_devices)
 995 {
 996         u64 total_devs;
 997         u64 dev_size;
 998         off_t seek_ret;
 999         int ret;
1000         if (!sb_bytenr)
1001                 sb_bytenr = BTRFS_SUPER_INFO_OFFSET;
1002
1003         seek_ret = lseek(fd, 0, SEEK_END);
1004         if (seek_ret < 0)
1005                 return -errno;
1006
1007         dev_size = seek_ret;
1008         lseek(fd, 0, SEEK_SET);
1009         if (sb_bytenr > dev_size) {
1010                 error("superblock bytenr %llu is larger than device size %llu",
1011                                 (unsigned long long)sb_bytenr,
1012                                 (unsigned long long)dev_size);
1013                 return -EINVAL;
1014         }
1015
1016         ret = btrfs_scan_one_device(fd, path, fs_devices,
1017                                     &total_devs, sb_bytenr, sbflags);
1018         if (ret) {
1019                 fprintf(stderr, "No valid Btrfs found on %s\n", path);
1020                 return ret;
1021         }
1022
1023         if (!skip_devices && total_devs != 1) {
1024                 ret = btrfs_scan_devices();
1025                 if (ret)
1026                         return ret;
1027         }
1028         return 0;
1029 }
1030
1031 int btrfs_setup_chunk_tree_and_device_map(struct btrfs_fs_info *fs_info,
1032                                           u64 chunk_root_bytenr)
1033 {
1034         struct btrfs_super_block *sb = fs_info->super_copy;
1035         u64 generation;
1036         int ret;
1037
1038         btrfs_setup_root(fs_info->chunk_root, fs_info,
1039                         BTRFS_CHUNK_TREE_OBJECTID);
1040
1041         ret = btrfs_read_sys_array(fs_info);
1042         if (ret)
1043                 return ret;
1044
1045         generation = btrfs_super_chunk_root_generation(sb);
1046
1047         if (chunk_root_bytenr && !IS_ALIGNED(chunk_root_bytenr,
1048                                             fs_info->sectorsize)) {
1049                 warning("chunk_root_bytenr %llu is unaligned to %u, ignore it",
1050                         chunk_root_bytenr, fs_info->sectorsize);
1051                 chunk_root_bytenr = 0;
1052         }
1053
1054         if (!chunk_root_bytenr)
1055                 chunk_root_bytenr = btrfs_super_chunk_root(sb);
1056         else
1057                 generation = 0;
1058
1059         fs_info->chunk_root->node = read_tree_block(fs_info,
1060                                                     chunk_root_bytenr,
1061                                                     generation);
1062         if (!extent_buffer_uptodate(fs_info->chunk_root->node)) {
1063                 if (fs_info->ignore_chunk_tree_error) {
1064                         warning("cannot read chunk root, continue anyway");
1065                         fs_info->chunk_root = NULL;
1066                         return 0;
1067                 } else {
1068                         error("cannot read chunk root");
1069                         return -EIO;
1070                 }
1071         }
1072
1073         if (!(btrfs_super_flags(sb) & BTRFS_SUPER_FLAG_METADUMP)) {
1074                 ret = btrfs_read_chunk_tree(fs_info);
1075                 if (ret) {
1076                         fprintf(stderr, "Couldn't read chunk tree\n");
1077                         return ret;
1078                 }
1079         }
1080         return 0;
1081 }
1082
1083 static struct btrfs_fs_info *__open_ctree_fd(int fp, const char *path,
1084                                              u64 sb_bytenr,
1085                                              u64 root_tree_bytenr,
1086                                              u64 chunk_root_bytenr,
1087                                              unsigned flags)
1088 {
1089         struct btrfs_fs_info *fs_info;
1090         struct btrfs_super_block *disk_super;
1091         struct btrfs_fs_devices *fs_devices = NULL;
1092         struct extent_buffer *eb;
1093         int ret;
1094         int oflags;
1095         unsigned sbflags = SBREAD_DEFAULT;
1096
1097         if (sb_bytenr == 0)
1098                 sb_bytenr = BTRFS_SUPER_INFO_OFFSET;
1099
1100         /* try to drop all the caches */
1101         if (posix_fadvise(fp, 0, 0, POSIX_FADV_DONTNEED))
1102                 fprintf(stderr, "Warning, could not drop caches\n");
1103
1104         fs_info = btrfs_new_fs_info(flags & OPEN_CTREE_WRITES, sb_bytenr);
1105         if (!fs_info) {
1106                 fprintf(stderr, "Failed to allocate memory for fs_info\n");
1107                 return NULL;
1108         }
1109         if (flags & OPEN_CTREE_RESTORE)
1110                 fs_info->on_restoring = 1;
1111         if (flags & OPEN_CTREE_SUPPRESS_CHECK_BLOCK_ERRORS)
1112                 fs_info->suppress_check_block_errors = 1;
1113         if (flags & OPEN_CTREE_IGNORE_FSID_MISMATCH)
1114                 fs_info->ignore_fsid_mismatch = 1;
1115         if (flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR)
1116                 fs_info->ignore_chunk_tree_error = 1;
1117
1118         if ((flags & OPEN_CTREE_RECOVER_SUPER)
1119              && (flags & OPEN_CTREE_TEMPORARY_SUPER)) {
1120                 fprintf(stderr,
1121         "cannot open a filesystem with temporary super block for recovery");
1122                 goto out;
1123         }
1124
1125         if (flags & OPEN_CTREE_TEMPORARY_SUPER)
1126                 sbflags = SBREAD_TEMPORARY;
1127
1128         ret = btrfs_scan_fs_devices(fp, path, &fs_devices, sb_bytenr, sbflags,
1129                         (flags & OPEN_CTREE_NO_DEVICES));
1130         if (ret)
1131                 goto out;
1132
1133         fs_info->fs_devices = fs_devices;
1134         if (flags & OPEN_CTREE_WRITES)
1135                 oflags = O_RDWR;
1136         else
1137                 oflags = O_RDONLY;
1138
1139         if (flags & OPEN_CTREE_EXCLUSIVE)
1140                 oflags |= O_EXCL;
1141
1142         ret = btrfs_open_devices(fs_devices, oflags);
1143         if (ret)
1144                 goto out;
1145
1146         disk_super = fs_info->super_copy;
1147         if (flags & OPEN_CTREE_RECOVER_SUPER)
1148                 ret = btrfs_read_dev_super(fs_devices->latest_bdev, disk_super,
1149                                 sb_bytenr, SBREAD_RECOVER);
1150         else
1151                 ret = btrfs_read_dev_super(fp, disk_super, sb_bytenr,
1152                                 sbflags);
1153         if (ret) {
1154                 printk("No valid btrfs found\n");
1155                 goto out_devices;
1156         }
1157
1158         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID &&
1159             !fs_info->ignore_fsid_mismatch) {
1160                 fprintf(stderr, "ERROR: Filesystem UUID change in progress\n");
1161                 goto out_devices;
1162         }
1163
1164         memcpy(fs_info->fsid, &disk_super->fsid, BTRFS_FSID_SIZE);
1165         fs_info->sectorsize = btrfs_super_sectorsize(disk_super);
1166         fs_info->nodesize = btrfs_super_nodesize(disk_super);
1167         fs_info->stripesize = btrfs_super_stripesize(disk_super);
1168
1169         ret = btrfs_check_fs_compatibility(fs_info->super_copy, flags);
1170         if (ret)
1171                 goto out_devices;
1172
1173         ret = btrfs_setup_chunk_tree_and_device_map(fs_info, chunk_root_bytenr);
1174         if (ret)
1175                 goto out_chunk;
1176
1177         /* Chunk tree root is unable to read, return directly */
1178         if (!fs_info->chunk_root)
1179                 return fs_info;
1180
1181         eb = fs_info->chunk_root->node;
1182         read_extent_buffer(eb, fs_info->chunk_tree_uuid,
1183                            btrfs_header_chunk_tree_uuid(eb),
1184                            BTRFS_UUID_SIZE);
1185
1186         ret = btrfs_setup_all_roots(fs_info, root_tree_bytenr, flags);
1187         if (ret && !(flags & __OPEN_CTREE_RETURN_CHUNK_ROOT) &&
1188             !fs_info->ignore_chunk_tree_error)
1189                 goto out_chunk;
1190
1191         return fs_info;
1192
1193 out_chunk:
1194         btrfs_release_all_roots(fs_info);
1195         btrfs_cleanup_all_caches(fs_info);
1196 out_devices:
1197         btrfs_close_devices(fs_devices);
1198 out:
1199         btrfs_free_fs_info(fs_info);
1200         return NULL;
1201 }
1202
1203 struct btrfs_fs_info *open_ctree_fs_info(const char *filename,
1204                                          u64 sb_bytenr, u64 root_tree_bytenr,
1205                                          u64 chunk_root_bytenr,
1206                                          unsigned flags)
1207 {
1208         int fp;
1209         int ret;
1210         struct btrfs_fs_info *info;
1211         int oflags = O_RDWR;
1212         struct stat st;
1213
1214         ret = stat(filename, &st);
1215         if (ret < 0) {
1216                 error("cannot stat '%s': %m", filename);
1217                 return NULL;
1218         }
1219         if (!(((st.st_mode & S_IFMT) == S_IFREG) || ((st.st_mode & S_IFMT) == S_IFBLK))) {
1220                 error("not a regular file or block device: %s", filename);
1221                 return NULL;
1222         }
1223
1224         if (!(flags & OPEN_CTREE_WRITES))
1225                 oflags = O_RDONLY;
1226
1227         fp = open(filename, oflags);
1228         if (fp < 0) {
1229                 error("cannot open '%s': %m", filename);
1230                 return NULL;
1231         }
1232         info = __open_ctree_fd(fp, filename, sb_bytenr, root_tree_bytenr,
1233                                chunk_root_bytenr, flags);
1234         close(fp);
1235         return info;
1236 }
1237
1238 struct btrfs_root *open_ctree(const char *filename, u64 sb_bytenr,
1239                               unsigned flags)
1240 {
1241         struct btrfs_fs_info *info;
1242
1243         /* This flags may not return fs_info with any valid root */
1244         BUG_ON(flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR);
1245         info = open_ctree_fs_info(filename, sb_bytenr, 0, 0, flags);
1246         if (!info)
1247                 return NULL;
1248         if (flags & __OPEN_CTREE_RETURN_CHUNK_ROOT)
1249                 return info->chunk_root;
1250         return info->fs_root;
1251 }
1252
1253 struct btrfs_root *open_ctree_fd(int fp, const char *path, u64 sb_bytenr,
1254                                  unsigned flags)
1255 {
1256         struct btrfs_fs_info *info;
1257
1258         /* This flags may not return fs_info with any valid root */
1259         if (flags & OPEN_CTREE_IGNORE_CHUNK_TREE_ERROR) {
1260                 error("invalid open_ctree flags: 0x%llx",
1261                                 (unsigned long long)flags);
1262                 return NULL;
1263         }
1264         info = __open_ctree_fd(fp, path, sb_bytenr, 0, 0, flags);
1265         if (!info)
1266                 return NULL;
1267         if (flags & __OPEN_CTREE_RETURN_CHUNK_ROOT)
1268                 return info->chunk_root;
1269         return info->fs_root;
1270 }
1271
1272 /*
1273  * Check if the super is valid:
1274  * - nodesize/sectorsize - minimum, maximum, alignment
1275  * - tree block starts   - alignment
1276  * - number of devices   - something sane
1277  * - sys array size      - maximum
1278  */
1279 static int check_super(struct btrfs_super_block *sb, unsigned sbflags)
1280 {
1281         u8 result[BTRFS_CSUM_SIZE];
1282         u32 crc;
1283         u16 csum_type;
1284         int csum_size;
1285
1286         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
1287                 if (btrfs_super_magic(sb) == BTRFS_MAGIC_TEMPORARY) {
1288                         if (!(sbflags & SBREAD_TEMPORARY)) {
1289                                 error("superblock magic doesn't match");
1290                                 return -EIO;
1291                         }
1292                 }
1293         }
1294
1295         csum_type = btrfs_super_csum_type(sb);
1296         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
1297                 error("unsupported checksum algorithm %u", csum_type);
1298                 return -EIO;
1299         }
1300         csum_size = btrfs_csum_sizes[csum_type];
1301
1302         crc = ~(u32)0;
1303         crc = btrfs_csum_data((char *)sb + BTRFS_CSUM_SIZE, crc,
1304                               BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1305         btrfs_csum_final(crc, result);
1306
1307         if (memcmp(result, sb->csum, csum_size)) {
1308                 error("superblock checksum mismatch");
1309                 return -EIO;
1310         }
1311         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
1312                 error("tree_root level too big: %d >= %d",
1313                         btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
1314                 goto error_out;
1315         }
1316         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
1317                 error("chunk_root level too big: %d >= %d",
1318                         btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
1319                 goto error_out;
1320         }
1321         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
1322                 error("log_root level too big: %d >= %d",
1323                         btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
1324                 goto error_out;
1325         }
1326
1327         if (!IS_ALIGNED(btrfs_super_root(sb), 4096)) {
1328                 error("tree_root block unaligned: %llu", btrfs_super_root(sb));
1329                 goto error_out;
1330         }
1331         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096)) {
1332                 error("chunk_root block unaligned: %llu",
1333                         btrfs_super_chunk_root(sb));
1334                 goto error_out;
1335         }
1336         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096)) {
1337                 error("log_root block unaligned: %llu",
1338                         btrfs_super_log_root(sb));
1339                 goto error_out;
1340         }
1341         if (btrfs_super_nodesize(sb) < 4096) {
1342                 error("nodesize too small: %u < 4096",
1343                         btrfs_super_nodesize(sb));
1344                 goto error_out;
1345         }
1346         if (!IS_ALIGNED(btrfs_super_nodesize(sb), 4096)) {
1347                 error("nodesize unaligned: %u", btrfs_super_nodesize(sb));
1348                 goto error_out;
1349         }
1350         if (btrfs_super_sectorsize(sb) < 4096) {
1351                 error("sectorsize too small: %u < 4096",
1352                         btrfs_super_sectorsize(sb));
1353                 goto error_out;
1354         }
1355         if (!IS_ALIGNED(btrfs_super_sectorsize(sb), 4096)) {
1356                 error("sectorsize unaligned: %u", btrfs_super_sectorsize(sb));
1357                 goto error_out;
1358         }
1359         if (btrfs_super_total_bytes(sb) == 0) {
1360                 error("invalid total_bytes 0");
1361                 goto error_out;
1362         }
1363         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
1364                 error("invalid bytes_used %llu", btrfs_super_bytes_used(sb));
1365                 goto error_out;
1366         }
1367         if ((btrfs_super_stripesize(sb) != 4096)
1368                 && (btrfs_super_stripesize(sb) != btrfs_super_sectorsize(sb))) {
1369                 error("invalid stripesize %u", btrfs_super_stripesize(sb));
1370                 goto error_out;
1371         }
1372
1373         if (memcmp(sb->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
1374                 char fsid[BTRFS_UUID_UNPARSED_SIZE];
1375                 char dev_fsid[BTRFS_UUID_UNPARSED_SIZE];
1376
1377                 uuid_unparse(sb->fsid, fsid);
1378                 uuid_unparse(sb->dev_item.fsid, dev_fsid);
1379                 error("dev_item UUID does not match fsid: %s != %s",
1380                         dev_fsid, fsid);
1381                 goto error_out;
1382         }
1383
1384         /*
1385          * Hint to catch really bogus numbers, bitflips or so
1386          */
1387         if (btrfs_super_num_devices(sb) > (1UL << 31)) {
1388                 warning("suspicious number of devices: %llu",
1389                         btrfs_super_num_devices(sb));
1390         }
1391
1392         if (btrfs_super_num_devices(sb) == 0) {
1393                 error("number of devices is 0");
1394                 goto error_out;
1395         }
1396
1397         /*
1398          * Obvious sys_chunk_array corruptions, it must hold at least one key
1399          * and one chunk
1400          */
1401         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
1402                 error("system chunk array too big %u > %u",
1403                       btrfs_super_sys_array_size(sb),
1404                       BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
1405                 goto error_out;
1406         }
1407         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
1408                         + sizeof(struct btrfs_chunk)) {
1409                 error("system chunk array too small %u < %zu",
1410                       btrfs_super_sys_array_size(sb),
1411                       sizeof(struct btrfs_disk_key) +
1412                       sizeof(struct btrfs_chunk));
1413                 goto error_out;
1414         }
1415
1416         return 0;
1417
1418 error_out:
1419         error("superblock checksum matches but it has invalid members");
1420         return -EIO;
1421 }
1422
1423 /*
1424  * btrfs_read_dev_super - read a valid superblock from a block device
1425  * @fd:         file descriptor of the device
1426  * @sb:         buffer where the superblock is going to be read in
1427  * @sb_bytenr:  offset of the particular superblock copy we want
1428  * @sbflags:    flags controlling how the superblock is read
1429  *
1430  * This function is used by various btrfs comands to obtain a valid superblock.
1431  *
1432  * It's mode of operation is controlled by the @sb_bytenr and @sbdflags
1433  * parameters. If SBREAD_RECOVER flag is set and @sb_bytenr is
1434  * BTRFS_SUPER_INFO_OFFSET then the function reads all 3 superblock copies and
1435  * returns the newest one. If SBREAD_RECOVER is not set then only a single
1436  * copy is read, which one is decided by @sb_bytenr. If @sb_bytenr !=
1437  * BTRFS_SUPER_INFO_OFFSET then the @sbflags is effectively ignored and only a
1438  * single copy is read.
1439  */
1440 int btrfs_read_dev_super(int fd, struct btrfs_super_block *sb, u64 sb_bytenr,
1441                          unsigned sbflags)
1442 {
1443         u8 fsid[BTRFS_FSID_SIZE];
1444         int fsid_is_initialized = 0;
1445         char tmp[BTRFS_SUPER_INFO_SIZE];
1446         struct btrfs_super_block *buf = (struct btrfs_super_block *)tmp;
1447         int i;
1448         int ret;
1449         int max_super = sbflags & SBREAD_RECOVER ? BTRFS_SUPER_MIRROR_MAX : 1;
1450         u64 transid = 0;
1451         u64 bytenr;
1452
1453         if (sb_bytenr != BTRFS_SUPER_INFO_OFFSET) {
1454                 ret = pread64(fd, buf, BTRFS_SUPER_INFO_SIZE, sb_bytenr);
1455                 /* real error */
1456                 if (ret < 0)
1457                         return -errno;
1458
1459                 /* Not large enough sb, return -ENOENT instead of normal -EIO */
1460                 if (ret < BTRFS_SUPER_INFO_SIZE)
1461                         return -ENOENT;
1462
1463                 if (btrfs_super_bytenr(buf) != sb_bytenr)
1464                         return -EIO;
1465
1466                 ret = check_super(buf, sbflags);
1467                 if (ret < 0)
1468                         return ret;
1469                 memcpy(sb, buf, BTRFS_SUPER_INFO_SIZE);
1470                 return 0;
1471         }
1472
1473         /*
1474         * we would like to check all the supers, but that would make
1475         * a btrfs mount succeed after a mkfs from a different FS.
1476         * So, we need to add a special mount option to scan for
1477         * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1478         */
1479
1480         for (i = 0; i < max_super; i++) {
1481                 bytenr = btrfs_sb_offset(i);
1482                 ret = pread64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1483                 if (ret < BTRFS_SUPER_INFO_SIZE)
1484                         break;
1485
1486                 if (btrfs_super_bytenr(buf) != bytenr )
1487                         continue;
1488                 /* if magic is NULL, the device was removed */
1489                 if (btrfs_super_magic(buf) == 0 && i == 0)
1490                         break;
1491                 if (check_super(buf, sbflags))
1492                         continue;
1493
1494                 if (!fsid_is_initialized) {
1495                         memcpy(fsid, buf->fsid, sizeof(fsid));
1496                         fsid_is_initialized = 1;
1497                 } else if (memcmp(fsid, buf->fsid, sizeof(fsid))) {
1498                         /*
1499                          * the superblocks (the original one and
1500                          * its backups) contain data of different
1501                          * filesystems -> the super cannot be trusted
1502                          */
1503                         continue;
1504                 }
1505
1506                 if (btrfs_super_generation(buf) > transid) {
1507                         memcpy(sb, buf, BTRFS_SUPER_INFO_SIZE);
1508                         transid = btrfs_super_generation(buf);
1509                 }
1510         }
1511
1512         return transid > 0 ? 0 : -1;
1513 }
1514
1515 static int write_dev_supers(struct btrfs_fs_info *fs_info,
1516                             struct btrfs_super_block *sb,
1517                             struct btrfs_device *device)
1518 {
1519         u64 bytenr;
1520         u32 crc;
1521         int i, ret;
1522
1523         if (fs_info->super_bytenr != BTRFS_SUPER_INFO_OFFSET) {
1524                 btrfs_set_super_bytenr(sb, fs_info->super_bytenr);
1525                 crc = ~(u32)0;
1526                 crc = btrfs_csum_data((char *)sb + BTRFS_CSUM_SIZE, crc,
1527                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1528                 btrfs_csum_final(crc, &sb->csum[0]);
1529
1530                 /*
1531                  * super_copy is BTRFS_SUPER_INFO_SIZE bytes and is
1532                  * zero filled, we can use it directly
1533                  */
1534                 ret = pwrite64(device->fd, fs_info->super_copy,
1535                                 BTRFS_SUPER_INFO_SIZE,
1536                                 fs_info->super_bytenr);
1537                 if (ret != BTRFS_SUPER_INFO_SIZE)
1538                         goto write_err;
1539                 return 0;
1540         }
1541
1542         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1543                 bytenr = btrfs_sb_offset(i);
1544                 if (bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
1545                         break;
1546
1547                 btrfs_set_super_bytenr(sb, bytenr);
1548
1549                 crc = ~(u32)0;
1550                 crc = btrfs_csum_data((char *)sb + BTRFS_CSUM_SIZE, crc,
1551                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1552                 btrfs_csum_final(crc, &sb->csum[0]);
1553
1554                 /*
1555                  * super_copy is BTRFS_SUPER_INFO_SIZE bytes and is
1556                  * zero filled, we can use it directly
1557                  */
1558                 ret = pwrite64(device->fd, fs_info->super_copy,
1559                                 BTRFS_SUPER_INFO_SIZE, bytenr);
1560                 if (ret != BTRFS_SUPER_INFO_SIZE)
1561                         goto write_err;
1562         }
1563
1564         return 0;
1565
1566 write_err:
1567         if (ret > 0)
1568                 fprintf(stderr, "WARNING: failed to write all sb data\n");
1569         else
1570                 fprintf(stderr, "WARNING: failed to write sb: %m\n");
1571         return ret;
1572 }
1573
1574 int write_all_supers(struct btrfs_fs_info *fs_info)
1575 {
1576         struct list_head *head = &fs_info->fs_devices->devices;
1577         struct btrfs_device *dev;
1578         struct btrfs_super_block *sb;
1579         struct btrfs_dev_item *dev_item;
1580         int ret;
1581         u64 flags;
1582
1583         sb = fs_info->super_copy;
1584         dev_item = &sb->dev_item;
1585         list_for_each_entry(dev, head, dev_list) {
1586                 if (!dev->writeable)
1587                         continue;
1588
1589                 btrfs_set_stack_device_generation(dev_item, 0);
1590                 btrfs_set_stack_device_type(dev_item, dev->type);
1591                 btrfs_set_stack_device_id(dev_item, dev->devid);
1592                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1593                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1594                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1595                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1596                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1597                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1598                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1599
1600                 flags = btrfs_super_flags(sb);
1601                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1602
1603                 ret = write_dev_supers(fs_info, sb, dev);
1604                 BUG_ON(ret);
1605         }
1606         return 0;
1607 }
1608
1609 int write_ctree_super(struct btrfs_trans_handle *trans,
1610                       struct btrfs_fs_info *fs_info)
1611 {
1612         int ret;
1613         struct btrfs_root *tree_root = fs_info->tree_root;
1614         struct btrfs_root *chunk_root = fs_info->chunk_root;
1615
1616         if (fs_info->readonly)
1617                 return 0;
1618
1619         btrfs_set_super_generation(fs_info->super_copy,
1620                                    trans->transid);
1621         btrfs_set_super_root(fs_info->super_copy,
1622                              tree_root->node->start);
1623         btrfs_set_super_root_level(fs_info->super_copy,
1624                                    btrfs_header_level(tree_root->node));
1625         btrfs_set_super_chunk_root(fs_info->super_copy,
1626                                    chunk_root->node->start);
1627         btrfs_set_super_chunk_root_level(fs_info->super_copy,
1628                                          btrfs_header_level(chunk_root->node));
1629         btrfs_set_super_chunk_root_generation(fs_info->super_copy,
1630                                 btrfs_header_generation(chunk_root->node));
1631
1632         ret = write_all_supers(fs_info);
1633         if (ret)
1634                 fprintf(stderr, "failed to write new super block err %d\n", ret);
1635         return ret;
1636 }
1637
1638 int close_ctree_fs_info(struct btrfs_fs_info *fs_info)
1639 {
1640         int ret;
1641         int err = 0;
1642         struct btrfs_trans_handle *trans;
1643         struct btrfs_root *root = fs_info->tree_root;
1644
1645         if (fs_info->last_trans_committed !=
1646             fs_info->generation) {
1647                 BUG_ON(!root);
1648                 trans = btrfs_start_transaction(root, 1);
1649                 if (IS_ERR(trans)) {
1650                         err = PTR_ERR(trans);
1651                         goto skip_commit;
1652                 }
1653                 btrfs_commit_transaction(trans, root);
1654                 trans = btrfs_start_transaction(root, 1);
1655                 BUG_ON(IS_ERR(trans));
1656                 ret = commit_tree_roots(trans, fs_info);
1657                 BUG_ON(ret);
1658                 ret = __commit_transaction(trans, root);
1659                 BUG_ON(ret);
1660                 write_ctree_super(trans, fs_info);
1661                 kfree(trans);
1662         }
1663
1664         if (fs_info->finalize_on_close) {
1665                 btrfs_set_super_magic(fs_info->super_copy, BTRFS_MAGIC);
1666                 root->fs_info->finalize_on_close = 0;
1667                 ret = write_all_supers(fs_info);
1668                 if (ret)
1669                         fprintf(stderr,
1670                                 "failed to write new super block err %d\n", ret);
1671         }
1672
1673 skip_commit:
1674         btrfs_free_block_groups(fs_info);
1675
1676         free_fs_roots_tree(&fs_info->fs_root_tree);
1677
1678         btrfs_release_all_roots(fs_info);
1679         ret = btrfs_close_devices(fs_info->fs_devices);
1680         btrfs_cleanup_all_caches(fs_info);
1681         btrfs_free_fs_info(fs_info);
1682         if (!err)
1683                 err = ret;
1684         return err;
1685 }
1686
1687 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1688                      struct extent_buffer *eb)
1689 {
1690         return clear_extent_buffer_dirty(eb);
1691 }
1692
1693 void btrfs_mark_buffer_dirty(struct extent_buffer *eb)
1694 {
1695         set_extent_buffer_dirty(eb);
1696 }
1697
1698 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1699 {
1700         int ret;
1701
1702         ret = extent_buffer_uptodate(buf);
1703         if (!ret)
1704                 return ret;
1705
1706         ret = verify_parent_transid(buf->tree, buf, parent_transid, 1);
1707         return !ret;
1708 }
1709
1710 int btrfs_set_buffer_uptodate(struct extent_buffer *eb)
1711 {
1712         return set_extent_buffer_uptodate(eb);
1713 }