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