btrfs-convert.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 "kerncompat.h"
  20
  21 #include <sys/ioctl.h>
  22 #include <sys/mount.h>
  23 #include <stdio.h>
  24 #include <stdlib.h>
  25 #include <sys/types.h>
  26 #include <sys/stat.h>
  27 #include <fcntl.h>
  28 #include <unistd.h>
  29 #include <uuid/uuid.h>
  30 #include <linux/limits.h>
  31 #include <getopt.h>
  32
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "volumes.h"
  36 #include "transaction.h"
  37 #include "crc32c.h"
  38 #include "utils.h"
  39 #include "task-utils.h"
  40 #include <ext2fs/ext2_fs.h>
  41 #include <ext2fs/ext2fs.h>
  42 #include <ext2fs/ext2_ext_attr.h>
  43
  44 #define INO_OFFSET (BTRFS_FIRST_FREE_OBJECTID - EXT2_ROOT_INO)
  45 #define CONV_IMAGE_SUBVOL_OBJECTID BTRFS_FIRST_FREE_OBJECTID
  46
  47 /*
  48  * Compatibility code for e2fsprogs 1.41 which doesn't support RO compat flag
  49  * BIGALLOC.
  50  * Unlike normal RO compat flag, BIGALLOC affects how e2fsprogs check used
  51  * space, and btrfs-convert heavily relies on it.
  52  */
  53 #ifdef HAVE_OLD_E2FSPROGS
  54 #define EXT2FS_CLUSTER_RATIO(fs)        (1)
  55 #define EXT2_CLUSTERS_PER_GROUP(s)      (EXT2_BLOCKS_PER_GROUP(s))
  56 #define EXT2FS_B2C(fs, blk)             (blk)
  57 #endif
  58
  59 struct task_ctx {
  60         uint32_t max_copy_inodes;
  61         uint32_t cur_copy_inodes;
  62         struct task_info *info;
  63 };
  64
  65 static void *print_copied_inodes(void *p)
  66 {
  67         struct task_ctx *priv = p;
  68         const char work_indicator[] = { '.', 'o', 'O', 'o' };
  69         uint32_t count = 0;
  70
  71         task_period_start(priv->info, 1000 /* 1s */);
  72         while (1) {
  73                 count++;
  74                 printf("copy inodes [%c] [%10d/%10d]\r",
  75                        work_indicator[count % 4], priv->cur_copy_inodes,
  76                        priv->max_copy_inodes);
  77                 fflush(stdout);
  78                 task_period_wait(priv->info);
  79         }
  80
  81         return NULL;
  82 }
  83
  84 static int after_copied_inodes(void *p)
  85 {
  86         printf("\n");
  87         fflush(stdout);
  88
  89         return 0;
  90 }
  91
  92 struct btrfs_convert_context;
  93 struct btrfs_convert_operations {
  94         const char *name;
  95         int (*open_fs)(struct btrfs_convert_context *cctx, const char *devname);
  96         int (*read_used_space)(struct btrfs_convert_context *cctx);
  97         int (*copy_inodes)(struct btrfs_convert_context *cctx,
  98                          struct btrfs_root *root, int datacsum,
  99                          int packing, int noxattr, struct task_ctx *p);
 100         void (*close_fs)(struct btrfs_convert_context *cctx);
 101 };
 102
 103 static void init_convert_context(struct btrfs_convert_context *cctx)
 104 {
 105         cache_tree_init(&cctx->used);
 106         cache_tree_init(&cctx->data_chunks);
 107         cache_tree_init(&cctx->free);
 108 }
 109
 110 static void clean_convert_context(struct btrfs_convert_context *cctx)
 111 {
 112         free_extent_cache_tree(&cctx->used);
 113         free_extent_cache_tree(&cctx->data_chunks);
 114         free_extent_cache_tree(&cctx->free);
 115 }
 116
 117 static inline int copy_inodes(struct btrfs_convert_context *cctx,
 118                               struct btrfs_root *root, int datacsum,
 119                               int packing, int noxattr, struct task_ctx *p)
 120 {
 121         return cctx->convert_ops->copy_inodes(cctx, root, datacsum, packing,
 122                                              noxattr, p);
 123 }
 124
 125 static inline void convert_close_fs(struct btrfs_convert_context *cctx)
 126 {
 127         cctx->convert_ops->close_fs(cctx);
 128 }
 129
 130 /*
 131  * Open Ext2fs in readonly mode, read block allocation bitmap and
 132  * inode bitmap into memory.
 133  */
 134 static int ext2_open_fs(struct btrfs_convert_context *cctx, const char *name)
 135 {
 136         errcode_t ret;
 137         ext2_filsys ext2_fs;
 138         ext2_ino_t ino;
 139         u32 ro_feature;
 140
 141         ret = ext2fs_open(name, 0, 0, 0, unix_io_manager, &ext2_fs);
 142         if (ret) {
 143                 fprintf(stderr, "ext2fs_open: %s\n", error_message(ret));
 144                 return -1;
 145         }
 146         /*
 147          * We need to know exactly the used space, some RO compat flags like
 148          * BIGALLOC will affect how used space is present.
 149          * So we need manuall check any unsupported RO compat flags
 150          */
 151         ro_feature = ext2_fs->super->s_feature_ro_compat;
 152         if (ro_feature & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
 153                 error(
 154 "unsupported RO features detected: %x, abort convert to avoid possible corruption",
 155                       ro_feature & ~EXT2_LIB_FEATURE_COMPAT_SUPP);
 156                 goto fail;
 157         }
 158         ret = ext2fs_read_inode_bitmap(ext2_fs);
 159         if (ret) {
 160                 fprintf(stderr, "ext2fs_read_inode_bitmap: %s\n",
 161                         error_message(ret));
 162                 goto fail;
 163         }
 164         ret = ext2fs_read_block_bitmap(ext2_fs);
 165         if (ret) {
 166                 fprintf(stderr, "ext2fs_read_block_bitmap: %s\n",
 167                         error_message(ret));
 168                 goto fail;
 169         }
 170         /*
 171          * search each block group for a free inode. this set up
 172          * uninit block/inode bitmaps appropriately.
 173          */
 174         ino = 1;
 175         while (ino <= ext2_fs->super->s_inodes_count) {
 176                 ext2_ino_t foo;
 177                 ext2fs_new_inode(ext2_fs, ino, 0, NULL, &foo);
 178                 ino += EXT2_INODES_PER_GROUP(ext2_fs->super);
 179         }
 180
 181         if (!(ext2_fs->super->s_feature_incompat &
 182               EXT2_FEATURE_INCOMPAT_FILETYPE)) {
 183                 fprintf(stderr, "filetype feature is missing\n");
 184                 goto fail;
 185         }
 186
 187         cctx->fs_data = ext2_fs;
 188         cctx->blocksize = ext2_fs->blocksize;
 189         cctx->block_count = ext2_fs->super->s_blocks_count;
 190         cctx->total_bytes = ext2_fs->blocksize * ext2_fs->super->s_blocks_count;
 191         cctx->volume_name = strndup(ext2_fs->super->s_volume_name, 16);
 192         cctx->first_data_block = ext2_fs->super->s_first_data_block;
 193         cctx->inodes_count = ext2_fs->super->s_inodes_count;
 194         cctx->free_inodes_count = ext2_fs->super->s_free_inodes_count;
 195         return 0;
 196 fail:
 197         ext2fs_close(ext2_fs);
 198         return -1;
 199 }
 200
 201 static int __ext2_add_one_block(ext2_filsys fs, char *bitmap,
 202                                 unsigned long group_nr, struct cache_tree *used)
 203 {
 204         unsigned long offset;
 205         unsigned i;
 206         int ret = 0;
 207
 208         offset = fs->super->s_first_data_block;
 209         offset /= EXT2FS_CLUSTER_RATIO(fs);
 210         offset += group_nr * EXT2_CLUSTERS_PER_GROUP(fs->super);
 211         for (i = 0; i < EXT2_CLUSTERS_PER_GROUP(fs->super); i++) {
 212                 if (ext2fs_test_bit(i, bitmap)) {
 213                         u64 start;
 214
 215                         start = (i + offset) * EXT2FS_CLUSTER_RATIO(fs);
 216                         start *= fs->blocksize;
 217                         ret = add_merge_cache_extent(used, start,
 218                                                      fs->blocksize);
 219                         if (ret < 0)
 220                                 break;
 221                 }
 222         }
 223         return ret;
 224 }
 225
 226 /*
 227  * Read all used ext2 space into cctx->used cache tree
 228  */
 229 static int ext2_read_used_space(struct btrfs_convert_context *cctx)
 230 {
 231         ext2_filsys fs = (ext2_filsys)cctx->fs_data;
 232         blk64_t blk_itr = EXT2FS_B2C(fs, fs->super->s_first_data_block);
 233         struct cache_tree *used_tree = &cctx->used;
 234         char *block_bitmap = NULL;
 235         unsigned long i;
 236         int block_nbytes;
 237         int ret = 0;
 238
 239         block_nbytes = EXT2_CLUSTERS_PER_GROUP(fs->super) / 8;
 240         /* Shouldn't happen */
 241         BUG_ON(!fs->block_map);
 242
 243         block_bitmap = malloc(block_nbytes);
 244         if (!block_bitmap)
 245                 return -ENOMEM;
 246
 247         for (i = 0; i < fs->group_desc_count; i++) {
 248                 ret = ext2fs_get_block_bitmap_range(fs->block_map, blk_itr,
 249                                                 block_nbytes * 8, block_bitmap);
 250                 if (ret) {
 251                         error("fail to get bitmap from ext2, %s",
 252                               strerror(-ret));
 253                         break;
 254                 }
 255                 ret = __ext2_add_one_block(fs, block_bitmap, i, used_tree);
 256                 if (ret < 0) {
 257                         error("fail to build used space tree, %s",
 258                               strerror(-ret));
 259                         break;
 260                 }
 261                 blk_itr += EXT2_CLUSTERS_PER_GROUP(fs->super);
 262         }
 263
 264         free(block_bitmap);
 265         return ret;
 266 }
 267
 268 static void ext2_close_fs(struct btrfs_convert_context *cctx)
 269 {
 270         if (cctx->volume_name) {
 271                 free(cctx->volume_name);
 272                 cctx->volume_name = NULL;
 273         }
 274         ext2fs_close(cctx->fs_data);
 275 }
 276
 277 static int intersect_with_sb(u64 bytenr, u64 num_bytes)
 278 {
 279         int i;
 280         u64 offset;
 281
 282         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 283                 offset = btrfs_sb_offset(i);
 284                 offset &= ~((u64)BTRFS_STRIPE_LEN - 1);
 285
 286                 if (bytenr < offset + BTRFS_STRIPE_LEN &&
 287                     bytenr + num_bytes > offset)
 288                         return 1;
 289         }
 290         return 0;
 291 }
 292
 293 static int convert_insert_dirent(struct btrfs_trans_handle *trans,
 294                                  struct btrfs_root *root,
 295                                  const char *name, size_t name_len,
 296                                  u64 dir, u64 objectid,
 297                                  u8 file_type, u64 index_cnt,
 298                                  struct btrfs_inode_item *inode)
 299 {
 300         int ret;
 301         u64 inode_size;
 302         struct btrfs_key location = {
 303                 .objectid = objectid,
 304                 .offset = 0,
 305                 .type = BTRFS_INODE_ITEM_KEY,
 306         };
 307
 308         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 309                                     dir, &location, file_type, index_cnt);
 310         if (ret)
 311                 return ret;
 312         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 313                                      objectid, dir, index_cnt);
 314         if (ret)
 315                 return ret;
 316         inode_size = btrfs_stack_inode_size(inode) + name_len * 2;
 317         btrfs_set_stack_inode_size(inode, inode_size);
 318
 319         return 0;
 320 }
 321
 322 struct dir_iterate_data {
 323         struct btrfs_trans_handle *trans;
 324         struct btrfs_root *root;
 325         struct btrfs_inode_item *inode;
 326         u64 objectid;
 327         u64 index_cnt;
 328         u64 parent;
 329         int errcode;
 330 };
 331
 332 static u8 filetype_conversion_table[EXT2_FT_MAX] = {
 333         [EXT2_FT_UNKNOWN]       = BTRFS_FT_UNKNOWN,
 334         [EXT2_FT_REG_FILE]      = BTRFS_FT_REG_FILE,
 335         [EXT2_FT_DIR]           = BTRFS_FT_DIR,
 336         [EXT2_FT_CHRDEV]        = BTRFS_FT_CHRDEV,
 337         [EXT2_FT_BLKDEV]        = BTRFS_FT_BLKDEV,
 338         [EXT2_FT_FIFO]          = BTRFS_FT_FIFO,
 339         [EXT2_FT_SOCK]          = BTRFS_FT_SOCK,
 340         [EXT2_FT_SYMLINK]       = BTRFS_FT_SYMLINK,
 341 };
 342
 343 static int dir_iterate_proc(ext2_ino_t dir, int entry,
 344                             struct ext2_dir_entry *dirent,
 345                             int offset, int blocksize,
 346                             char *buf,void *priv_data)
 347 {
 348         int ret;
 349         int file_type;
 350         u64 objectid;
 351         char dotdot[] = "..";
 352         struct dir_iterate_data *idata = (struct dir_iterate_data *)priv_data;
 353         int name_len;
 354
 355         name_len = dirent->name_len & 0xFF;
 356
 357         objectid = dirent->inode + INO_OFFSET;
 358         if (!strncmp(dirent->name, dotdot, name_len)) {
 359                 if (name_len == 2) {
 360                         BUG_ON(idata->parent != 0);
 361                         idata->parent = objectid;
 362                 }
 363                 return 0;
 364         }
 365         if (dirent->inode < EXT2_GOOD_OLD_FIRST_INO)
 366                 return 0;
 367
 368         file_type = dirent->name_len >> 8;
 369         BUG_ON(file_type > EXT2_FT_SYMLINK);
 370
 371         ret = convert_insert_dirent(idata->trans, idata->root, dirent->name,
 372                                     name_len, idata->objectid, objectid,
 373                                     filetype_conversion_table[file_type],
 374                                     idata->index_cnt, idata->inode);
 375         if (ret < 0) {
 376                 idata->errcode = ret;
 377                 return BLOCK_ABORT;
 378         }
 379
 380         idata->index_cnt++;
 381         return 0;
 382 }
 383
 384 static int create_dir_entries(struct btrfs_trans_handle *trans,
 385                               struct btrfs_root *root, u64 objectid,
 386                               struct btrfs_inode_item *btrfs_inode,
 387                               ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
 388 {
 389         int ret;
 390         errcode_t err;
 391         struct dir_iterate_data data = {
 392                 .trans          = trans,
 393                 .root           = root,
 394                 .inode          = btrfs_inode,
 395                 .objectid       = objectid,
 396                 .index_cnt      = 2,
 397                 .parent         = 0,
 398                 .errcode        = 0,
 399         };
 400
 401         err = ext2fs_dir_iterate2(ext2_fs, ext2_ino, 0, NULL,
 402                                   dir_iterate_proc, &data);
 403         if (err)
 404                 goto error;
 405         ret = data.errcode;
 406         if (ret == 0 && data.parent == objectid) {
 407                 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
 408                                              objectid, objectid, 0);
 409         }
 410         return ret;
 411 error:
 412         fprintf(stderr, "ext2fs_dir_iterate2: %s\n", error_message(err));
 413         return -1;
 414 }
 415
 416 static int read_disk_extent(struct btrfs_root *root, u64 bytenr,
 417                             u32 num_bytes, char *buffer)
 418 {
 419         int ret;
 420         struct btrfs_fs_devices *fs_devs = root->fs_info->fs_devices;
 421
 422         ret = pread(fs_devs->latest_bdev, buffer, num_bytes, bytenr);
 423         if (ret != num_bytes)
 424                 goto fail;
 425         ret = 0;
 426 fail:
 427         if (ret > 0)
 428                 ret = -1;
 429         return ret;
 430 }
 431
 432 static int csum_disk_extent(struct btrfs_trans_handle *trans,
 433                             struct btrfs_root *root,
 434                             u64 disk_bytenr, u64 num_bytes)
 435 {
 436         u32 blocksize = root->sectorsize;
 437         u64 offset;
 438         char *buffer;
 439         int ret = 0;
 440
 441         buffer = malloc(blocksize);
 442         if (!buffer)
 443                 return -ENOMEM;
 444         for (offset = 0; offset < num_bytes; offset += blocksize) {
 445                 ret = read_disk_extent(root, disk_bytenr + offset,
 446                                         blocksize, buffer);
 447                 if (ret)
 448                         break;
 449                 ret = btrfs_csum_file_block(trans,
 450                                             root->fs_info->csum_root,
 451                                             disk_bytenr + num_bytes,
 452                                             disk_bytenr + offset,
 453                                             buffer, blocksize);
 454                 if (ret)
 455                         break;
 456         }
 457         free(buffer);
 458         return ret;
 459 }
 460
 461 struct blk_iterate_data {
 462         struct btrfs_trans_handle *trans;
 463         struct btrfs_root *root;
 464         struct btrfs_root *convert_root;
 465         struct btrfs_inode_item *inode;
 466         u64 convert_ino;
 467         u64 objectid;
 468         u64 first_block;
 469         u64 disk_block;
 470         u64 num_blocks;
 471         u64 boundary;
 472         int checksum;
 473         int errcode;
 474 };
 475
 476 static void init_blk_iterate_data(struct blk_iterate_data *data,
 477                                   struct btrfs_trans_handle *trans,
 478                                   struct btrfs_root *root,
 479                                   struct btrfs_inode_item *inode,
 480                                   u64 objectid, int checksum)
 481 {
 482         struct btrfs_key key;
 483
 484         data->trans             = trans;
 485         data->root              = root;
 486         data->inode             = inode;
 487         data->objectid          = objectid;
 488         data->first_block       = 0;
 489         data->disk_block        = 0;
 490         data->num_blocks        = 0;
 491         data->boundary          = (u64)-1;
 492         data->checksum          = checksum;
 493         data->errcode           = 0;
 494
 495         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
 496         key.type = BTRFS_ROOT_ITEM_KEY;
 497         key.offset = (u64)-1;
 498         data->convert_root = btrfs_read_fs_root(root->fs_info, &key);
 499         /* Impossible as we just opened it before */
 500         BUG_ON(!data->convert_root || IS_ERR(data->convert_root));
 501         data->convert_ino = BTRFS_FIRST_FREE_OBJECTID + 1;
 502 }
 503
 504 /*
 505  * Record a file extent in original filesystem into btrfs one.
 506  * The special point is, old disk_block can point to a reserved range.
 507  * So here, we don't use disk_block directly but search convert_root
 508  * to get the real disk_bytenr.
 509  */
 510 static int record_file_blocks(struct blk_iterate_data *data,
 511                               u64 file_block, u64 disk_block, u64 num_blocks)
 512 {
 513         int ret = 0;
 514         struct btrfs_root *root = data->root;
 515         struct btrfs_root *convert_root = data->convert_root;
 516         struct btrfs_path *path;
 517         u64 file_pos = file_block * root->sectorsize;
 518         u64 old_disk_bytenr = disk_block * root->sectorsize;
 519         u64 num_bytes = num_blocks * root->sectorsize;
 520         u64 cur_off = old_disk_bytenr;
 521
 522         /* Hole, pass it to record_file_extent directly */
 523         if (old_disk_bytenr == 0)
 524                 return btrfs_record_file_extent(data->trans, root,
 525                                 data->objectid, data->inode, file_pos, 0,
 526                                 num_bytes);
 527
 528         path = btrfs_alloc_path();
 529         if (!path)
 530                 return -ENOMEM;
 531
 532         /*
 533          * Search real disk bytenr from convert root
 534          */
 535         while (cur_off < old_disk_bytenr + num_bytes) {
 536                 struct btrfs_key key;
 537                 struct btrfs_file_extent_item *fi;
 538                 struct extent_buffer *node;
 539                 int slot;
 540                 u64 extent_disk_bytenr;
 541                 u64 extent_num_bytes;
 542                 u64 real_disk_bytenr;
 543                 u64 cur_len;
 544
 545                 key.objectid = data->convert_ino;
 546                 key.type = BTRFS_EXTENT_DATA_KEY;
 547                 key.offset = cur_off;
 548
 549                 ret = btrfs_search_slot(NULL, convert_root, &key, path, 0, 0);
 550                 if (ret < 0)
 551                         break;
 552                 if (ret > 0) {
 553                         ret = btrfs_previous_item(convert_root, path,
 554                                                   data->convert_ino,
 555                                                   BTRFS_EXTENT_DATA_KEY);
 556                         if (ret < 0)
 557                                 break;
 558                         if (ret > 0) {
 559                                 ret = -ENOENT;
 560                                 break;
 561                         }
 562                 }
 563                 node = path->nodes[0];
 564                 slot = path->slots[0];
 565                 btrfs_item_key_to_cpu(node, &key, slot);
 566                 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY ||
 567                        key.objectid != data->convert_ino ||
 568                        key.offset > cur_off);
 569                 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
 570                 extent_disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
 571                 extent_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
 572                 BUG_ON(cur_off - key.offset >= extent_num_bytes);
 573                 btrfs_release_path(path);
 574
 575                 if (extent_disk_bytenr)
 576                         real_disk_bytenr = cur_off - key.offset +
 577                                            extent_disk_bytenr;
 578                 else
 579                         real_disk_bytenr = 0;
 580                 cur_len = min(key.offset + extent_num_bytes,
 581                               old_disk_bytenr + num_bytes) - cur_off;
 582                 ret = btrfs_record_file_extent(data->trans, data->root,
 583                                         data->objectid, data->inode, file_pos,
 584                                         real_disk_bytenr, cur_len);
 585                 if (ret < 0)
 586                         break;
 587                 cur_off += cur_len;
 588                 file_pos += cur_len;
 589
 590                 /*
 591                  * No need to care about csum
 592                  * As every byte of old fs image is calculated for csum, no
 593                  * need to waste CPU cycles now.
 594                  */
 595         }
 596         btrfs_free_path(path);
 597         return ret;
 598 }
 599
 600 static int block_iterate_proc(u64 disk_block, u64 file_block,
 601                               struct blk_iterate_data *idata)
 602 {
 603         int ret = 0;
 604         int sb_region;
 605         int do_barrier;
 606         struct btrfs_root *root = idata->root;
 607         struct btrfs_block_group_cache *cache;
 608         u64 bytenr = disk_block * root->sectorsize;
 609
 610         sb_region = intersect_with_sb(bytenr, root->sectorsize);
 611         do_barrier = sb_region || disk_block >= idata->boundary;
 612         if ((idata->num_blocks > 0 && do_barrier) ||
 613             (file_block > idata->first_block + idata->num_blocks) ||
 614             (disk_block != idata->disk_block + idata->num_blocks)) {
 615                 if (idata->num_blocks > 0) {
 616                         ret = record_file_blocks(idata, idata->first_block,
 617                                                  idata->disk_block,
 618                                                  idata->num_blocks);
 619                         if (ret)
 620                                 goto fail;
 621                         idata->first_block += idata->num_blocks;
 622                         idata->num_blocks = 0;
 623                 }
 624                 if (file_block > idata->first_block) {
 625                         ret = record_file_blocks(idata, idata->first_block,
 626                                         0, file_block - idata->first_block);
 627                         if (ret)
 628                                 goto fail;
 629                 }
 630
 631                 if (sb_region) {
 632                         bytenr += BTRFS_STRIPE_LEN - 1;
 633                         bytenr &= ~((u64)BTRFS_STRIPE_LEN - 1);
 634                 } else {
 635                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
 636                         BUG_ON(!cache);
 637                         bytenr = cache->key.objectid + cache->key.offset;
 638                 }
 639
 640                 idata->first_block = file_block;
 641                 idata->disk_block = disk_block;
 642                 idata->boundary = bytenr / root->sectorsize;
 643         }
 644         idata->num_blocks++;
 645 fail:
 646         return ret;
 647 }
 648
 649 static int __block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
 650                                 e2_blkcnt_t blockcnt, blk_t ref_block,
 651                                 int ref_offset, void *priv_data)
 652 {
 653         int ret;
 654         struct blk_iterate_data *idata;
 655         idata = (struct blk_iterate_data *)priv_data;
 656         ret = block_iterate_proc(*blocknr, blockcnt, idata);
 657         if (ret) {
 658                 idata->errcode = ret;
 659                 return BLOCK_ABORT;
 660         }
 661         return 0;
 662 }
 663
 664 /*
 665  * traverse file's data blocks, record these data blocks as file extents.
 666  */
 667 static int create_file_extents(struct btrfs_trans_handle *trans,
 668                                struct btrfs_root *root, u64 objectid,
 669                                struct btrfs_inode_item *btrfs_inode,
 670                                ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
 671                                int datacsum, int packing)
 672 {
 673         int ret;
 674         char *buffer = NULL;
 675         errcode_t err;
 676         u32 last_block;
 677         u32 sectorsize = root->sectorsize;
 678         u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
 679         struct blk_iterate_data data;
 680
 681         init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
 682                               datacsum);
 683
 684         err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
 685                                     NULL, __block_iterate_proc, &data);
 686         if (err)
 687                 goto error;
 688         ret = data.errcode;
 689         if (ret)
 690                 goto fail;
 691         if (packing && data.first_block == 0 && data.num_blocks > 0 &&
 692             inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 693                 u64 num_bytes = data.num_blocks * sectorsize;
 694                 u64 disk_bytenr = data.disk_block * sectorsize;
 695                 u64 nbytes;
 696
 697                 buffer = malloc(num_bytes);
 698                 if (!buffer)
 699                         return -ENOMEM;
 700                 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
 701                 if (ret)
 702                         goto fail;
 703                 if (num_bytes > inode_size)
 704                         num_bytes = inode_size;
 705                 ret = btrfs_insert_inline_extent(trans, root, objectid,
 706                                                  0, buffer, num_bytes);
 707                 if (ret)
 708                         goto fail;
 709                 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
 710                 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
 711         } else if (data.num_blocks > 0) {
 712                 ret = record_file_blocks(&data, data.first_block,
 713                                          data.disk_block, data.num_blocks);
 714                 if (ret)
 715                         goto fail;
 716         }
 717         data.first_block += data.num_blocks;
 718         last_block = (inode_size + sectorsize - 1) / sectorsize;
 719         if (last_block > data.first_block) {
 720                 ret = record_file_blocks(&data, data.first_block, 0,
 721                                          last_block - data.first_block);
 722         }
 723 fail:
 724         free(buffer);
 725         return ret;
 726 error:
 727         fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
 728         return -1;
 729 }
 730
 731 static int create_symbol_link(struct btrfs_trans_handle *trans,
 732                               struct btrfs_root *root, u64 objectid,
 733                               struct btrfs_inode_item *btrfs_inode,
 734                               ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
 735                               struct ext2_inode *ext2_inode)
 736 {
 737         int ret;
 738         char *pathname;
 739         u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
 740         if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
 741                 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
 742                 ret = create_file_extents(trans, root, objectid, btrfs_inode,
 743                                           ext2_fs, ext2_ino, 1, 1);
 744                 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
 745                 return ret;
 746         }
 747
 748         pathname = (char *)&(ext2_inode->i_block[0]);
 749         BUG_ON(pathname[inode_size] != 0);
 750         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 751                                          pathname, inode_size + 1);
 752         btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
 753         return ret;
 754 }
 755
 756 /*
 757  * Following xattr/acl related codes are based on codes in
 758  * fs/ext3/xattr.c and fs/ext3/acl.c
 759  */
 760 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
 761 #define EXT2_XATTR_BFIRST(ptr) \
 762         ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
 763 #define EXT2_XATTR_IHDR(inode) \
 764         ((struct ext2_ext_attr_header *) ((void *)(inode) + \
 765                 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
 766 #define EXT2_XATTR_IFIRST(inode) \
 767         ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
 768                 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
 769
 770 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
 771                                   const void *end)
 772 {
 773         struct ext2_ext_attr_entry *next;
 774
 775         while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
 776                 next = EXT2_EXT_ATTR_NEXT(entry);
 777                 if ((void *)next >= end)
 778                         return -EIO;
 779                 entry = next;
 780         }
 781         return 0;
 782 }
 783
 784 static int ext2_xattr_check_block(const char *buf, size_t size)
 785 {
 786         int error;
 787         struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
 788
 789         if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
 790             header->h_blocks != 1)
 791                 return -EIO;
 792         error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
 793         return error;
 794 }
 795
 796 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
 797                                   size_t size)
 798 {
 799         size_t value_size = entry->e_value_size;
 800
 801         if (entry->e_value_block != 0 || value_size > size ||
 802             entry->e_value_offs + value_size > size)
 803                 return -EIO;
 804         return 0;
 805 }
 806
 807 #define EXT2_ACL_VERSION        0x0001
 808
 809 /* 23.2.5 acl_tag_t values */
 810
 811 #define ACL_UNDEFINED_TAG       (0x00)
 812 #define ACL_USER_OBJ            (0x01)
 813 #define ACL_USER                (0x02)
 814 #define ACL_GROUP_OBJ           (0x04)
 815 #define ACL_GROUP               (0x08)
 816 #define ACL_MASK                (0x10)
 817 #define ACL_OTHER               (0x20)
 818
 819 /* 23.2.7 ACL qualifier constants */
 820
 821 #define ACL_UNDEFINED_ID        ((id_t)-1)
 822
 823 typedef struct {
 824         __le16          e_tag;
 825         __le16          e_perm;
 826         __le32          e_id;
 827 } ext2_acl_entry;
 828
 829 typedef struct {
 830         __le16          e_tag;
 831         __le16          e_perm;
 832 } ext2_acl_entry_short;
 833
 834 typedef struct {
 835         __le32          a_version;
 836 } ext2_acl_header;
 837
 838 static inline int ext2_acl_count(size_t size)
 839 {
 840         ssize_t s;
 841         size -= sizeof(ext2_acl_header);
 842         s = size - 4 * sizeof(ext2_acl_entry_short);
 843         if (s < 0) {
 844                 if (size % sizeof(ext2_acl_entry_short))
 845                         return -1;
 846                 return size / sizeof(ext2_acl_entry_short);
 847         } else {
 848                 if (s % sizeof(ext2_acl_entry))
 849                         return -1;
 850                 return s / sizeof(ext2_acl_entry) + 4;
 851         }
 852 }
 853
 854 #define ACL_EA_VERSION          0x0002
 855
 856 typedef struct {
 857         __le16          e_tag;
 858         __le16          e_perm;
 859         __le32          e_id;
 860 } acl_ea_entry;
 861
 862 typedef struct {
 863         __le32          a_version;
 864         acl_ea_entry    a_entries[0];
 865 } acl_ea_header;
 866
 867 static inline size_t acl_ea_size(int count)
 868 {
 869         return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
 870 }
 871
 872 static int ext2_acl_to_xattr(void *dst, const void *src,
 873                              size_t dst_size, size_t src_size)
 874 {
 875         int i, count;
 876         const void *end = src + src_size;
 877         acl_ea_header *ext_acl = (acl_ea_header *)dst;
 878         acl_ea_entry *dst_entry = ext_acl->a_entries;
 879         ext2_acl_entry *src_entry;
 880
 881         if (src_size < sizeof(ext2_acl_header))
 882                 goto fail;
 883         if (((ext2_acl_header *)src)->a_version !=
 884             cpu_to_le32(EXT2_ACL_VERSION))
 885                 goto fail;
 886         src += sizeof(ext2_acl_header);
 887         count = ext2_acl_count(src_size);
 888         if (count <= 0)
 889                 goto fail;
 890
 891         BUG_ON(dst_size < acl_ea_size(count));
 892         ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
 893         for (i = 0; i < count; i++, dst_entry++) {
 894                 src_entry = (ext2_acl_entry *)src;
 895                 if (src + sizeof(ext2_acl_entry_short) > end)
 896                         goto fail;
 897                 dst_entry->e_tag = src_entry->e_tag;
 898                 dst_entry->e_perm = src_entry->e_perm;
 899                 switch (le16_to_cpu(src_entry->e_tag)) {
 900                 case ACL_USER_OBJ:
 901                 case ACL_GROUP_OBJ:
 902                 case ACL_MASK:
 903                 case ACL_OTHER:
 904                         src += sizeof(ext2_acl_entry_short);
 905                         dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
 906                         break;
 907                 case ACL_USER:
 908                 case ACL_GROUP:
 909                         src += sizeof(ext2_acl_entry);
 910                         if (src > end)
 911                                 goto fail;
 912                         dst_entry->e_id = src_entry->e_id;
 913                         break;
 914                 default:
 915                         goto fail;
 916                 }
 917         }
 918         if (src != end)
 919                 goto fail;
 920         return 0;
 921 fail:
 922         return -EINVAL;
 923 }
 924
 925 static char *xattr_prefix_table[] = {
 926         [1] =   "user.",
 927         [2] =   "system.posix_acl_access",
 928         [3] =   "system.posix_acl_default",
 929         [4] =   "trusted.",
 930         [6] =   "security.",
 931 };
 932
 933 static int copy_single_xattr(struct btrfs_trans_handle *trans,
 934                              struct btrfs_root *root, u64 objectid,
 935                              struct ext2_ext_attr_entry *entry,
 936                              const void *data, u32 datalen)
 937 {
 938         int ret = 0;
 939         int name_len;
 940         int name_index;
 941         void *databuf = NULL;
 942         char namebuf[XATTR_NAME_MAX + 1];
 943
 944         name_index = entry->e_name_index;
 945         if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
 946             xattr_prefix_table[name_index] == NULL)
 947                 return -EOPNOTSUPP;
 948         name_len = strlen(xattr_prefix_table[name_index]) +
 949                    entry->e_name_len;
 950         if (name_len >= sizeof(namebuf))
 951                 return -ERANGE;
 952
 953         if (name_index == 2 || name_index == 3) {
 954                 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
 955                 databuf = malloc(bufsize);
 956                 if (!databuf)
 957                        return -ENOMEM;
 958                 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
 959                 if (ret)
 960                         goto out;
 961                 data = databuf;
 962                 datalen = bufsize;
 963         }
 964         strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
 965         strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
 966         if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
 967             sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
 968                 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
 969                         objectid - INO_OFFSET, name_len, namebuf);
 970                 goto out;
 971         }
 972         ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
 973                                       data, datalen, objectid);
 974 out:
 975         free(databuf);
 976         return ret;
 977 }
 978
 979 static int copy_extended_attrs(struct btrfs_trans_handle *trans,
 980                                struct btrfs_root *root, u64 objectid,
 981                                struct btrfs_inode_item *btrfs_inode,
 982                                ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
 983 {
 984         int ret = 0;
 985         int inline_ea = 0;
 986         errcode_t err;
 987         u32 datalen;
 988         u32 block_size = ext2_fs->blocksize;
 989         u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
 990         struct ext2_inode_large *ext2_inode;
 991         struct ext2_ext_attr_entry *entry;
 992         void *data;
 993         char *buffer = NULL;
 994         char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
 995
 996         if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
 997                 ext2_inode = (struct ext2_inode_large *)inode_buf;
 998         } else {
 999                 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
1000                 if (!ext2_inode)
1001                        return -ENOMEM;
1002         }
1003         err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
1004                                      inode_size);
1005         if (err) {
1006                 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
1007                         error_message(err));
1008                 ret = -1;
1009                 goto out;
1010         }
1011
1012         if (ext2_ino > ext2_fs->super->s_first_ino &&
1013             inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
1014                 if (EXT2_GOOD_OLD_INODE_SIZE +
1015                     ext2_inode->i_extra_isize > inode_size) {
1016                         ret = -EIO;
1017                         goto out;
1018                 }
1019                 if (ext2_inode->i_extra_isize != 0 &&
1020                     EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
1021                     EXT2_EXT_ATTR_MAGIC) {
1022                         inline_ea = 1;
1023                 }
1024         }
1025         if (inline_ea) {
1026                 int total;
1027                 void *end = (void *)ext2_inode + inode_size;
1028                 entry = EXT2_XATTR_IFIRST(ext2_inode);
1029                 total = end - (void *)entry;
1030                 ret = ext2_xattr_check_names(entry, end);
1031                 if (ret)
1032                         goto out;
1033                 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1034                         ret = ext2_xattr_check_entry(entry, total);
1035                         if (ret)
1036                                 goto out;
1037                         data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
1038                                 entry->e_value_offs;
1039                         datalen = entry->e_value_size;
1040                         ret = copy_single_xattr(trans, root, objectid,
1041                                                 entry, data, datalen);
1042                         if (ret)
1043                                 goto out;
1044                         entry = EXT2_EXT_ATTR_NEXT(entry);
1045                 }
1046         }
1047
1048         if (ext2_inode->i_file_acl == 0)
1049                 goto out;
1050
1051         buffer = malloc(block_size);
1052         if (!buffer) {
1053                 ret = -ENOMEM;
1054                 goto out;
1055         }
1056         err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
1057         if (err) {
1058                 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
1059                         error_message(err));
1060                 ret = -1;
1061                 goto out;
1062         }
1063         ret = ext2_xattr_check_block(buffer, block_size);
1064         if (ret)
1065                 goto out;
1066
1067         entry = EXT2_XATTR_BFIRST(buffer);
1068         while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1069                 ret = ext2_xattr_check_entry(entry, block_size);
1070                 if (ret)
1071                         goto out;
1072                 data = buffer + entry->e_value_offs;
1073                 datalen = entry->e_value_size;
1074                 ret = copy_single_xattr(trans, root, objectid,
1075                                         entry, data, datalen);
1076                 if (ret)
1077                         goto out;
1078                 entry = EXT2_EXT_ATTR_NEXT(entry);
1079         }
1080 out:
1081         free(buffer);
1082         if ((void *)ext2_inode != inode_buf)
1083                 free(ext2_inode);
1084         return ret;
1085 }
1086 #define MINORBITS       20
1087 #define MKDEV(ma, mi)   (((ma) << MINORBITS) | (mi))
1088
1089 static inline dev_t old_decode_dev(u16 val)
1090 {
1091         return MKDEV((val >> 8) & 255, val & 255);
1092 }
1093
1094 static inline dev_t new_decode_dev(u32 dev)
1095 {
1096         unsigned major = (dev & 0xfff00) >> 8;
1097         unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
1098         return MKDEV(major, minor);
1099 }
1100
1101 static int copy_inode_item(struct btrfs_inode_item *dst,
1102                            struct ext2_inode *src, u32 blocksize)
1103 {
1104         btrfs_set_stack_inode_generation(dst, 1);
1105         btrfs_set_stack_inode_sequence(dst, 0);
1106         btrfs_set_stack_inode_transid(dst, 1);
1107         btrfs_set_stack_inode_size(dst, src->i_size);
1108         btrfs_set_stack_inode_nbytes(dst, 0);
1109         btrfs_set_stack_inode_block_group(dst, 0);
1110         btrfs_set_stack_inode_nlink(dst, src->i_links_count);
1111         btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
1112         btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
1113         btrfs_set_stack_inode_mode(dst, src->i_mode);
1114         btrfs_set_stack_inode_rdev(dst, 0);
1115         btrfs_set_stack_inode_flags(dst, 0);
1116         btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
1117         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
1118         btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
1119         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
1120         btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
1121         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
1122         btrfs_set_stack_timespec_sec(&dst->otime, 0);
1123         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
1124
1125         if (S_ISDIR(src->i_mode)) {
1126                 btrfs_set_stack_inode_size(dst, 0);
1127                 btrfs_set_stack_inode_nlink(dst, 1);
1128         }
1129         if (S_ISREG(src->i_mode)) {
1130                 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
1131                                            (u64)src->i_size);
1132         }
1133         if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
1134             !S_ISLNK(src->i_mode)) {
1135                 if (src->i_block[0]) {
1136                         btrfs_set_stack_inode_rdev(dst,
1137                                 old_decode_dev(src->i_block[0]));
1138                 } else {
1139                         btrfs_set_stack_inode_rdev(dst,
1140                                 new_decode_dev(src->i_block[1]));
1141                 }
1142         }
1143         memset(&dst->reserved, 0, sizeof(dst->reserved));
1144
1145         return 0;
1146 }
1147
1148 /*
1149  * copy a single inode. do all the required works, such as cloning
1150  * inode item, creating file extents and creating directory entries.
1151  */
1152 static int copy_single_inode(struct btrfs_trans_handle *trans,
1153                              struct btrfs_root *root, u64 objectid,
1154                              ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1155                              struct ext2_inode *ext2_inode,
1156                              int datacsum, int packing, int noxattr)
1157 {
1158         int ret;
1159         struct btrfs_inode_item btrfs_inode;
1160
1161         if (ext2_inode->i_links_count == 0)
1162                 return 0;
1163
1164         copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
1165         if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
1166                 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
1167                             BTRFS_INODE_NODATASUM;
1168                 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
1169         }
1170
1171         switch (ext2_inode->i_mode & S_IFMT) {
1172         case S_IFREG:
1173                 ret = create_file_extents(trans, root, objectid, &btrfs_inode,
1174                                         ext2_fs, ext2_ino, datacsum, packing);
1175                 break;
1176         case S_IFDIR:
1177                 ret = create_dir_entries(trans, root, objectid, &btrfs_inode,
1178                                          ext2_fs, ext2_ino);
1179                 break;
1180         case S_IFLNK:
1181                 ret = create_symbol_link(trans, root, objectid, &btrfs_inode,
1182                                          ext2_fs, ext2_ino, ext2_inode);
1183                 break;
1184         default:
1185                 ret = 0;
1186                 break;
1187         }
1188         if (ret)
1189                 return ret;
1190
1191         if (!noxattr) {
1192                 ret = copy_extended_attrs(trans, root, objectid, &btrfs_inode,
1193                                           ext2_fs, ext2_ino);
1194                 if (ret)
1195                         return ret;
1196         }
1197         return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
1198 }
1199
1200 /*
1201  * scan ext2's inode bitmap and copy all used inodes.
1202  */
1203 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
1204                             struct btrfs_root *root,
1205                             int datacsum, int packing, int noxattr, struct task_ctx *p)
1206 {
1207         ext2_filsys ext2_fs = cctx->fs_data;
1208         int ret;
1209         errcode_t err;
1210         ext2_inode_scan ext2_scan;
1211         struct ext2_inode ext2_inode;
1212         ext2_ino_t ext2_ino;
1213         u64 objectid;
1214         struct btrfs_trans_handle *trans;
1215
1216         trans = btrfs_start_transaction(root, 1);
1217         if (!trans)
1218                 return -ENOMEM;
1219         err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
1220         if (err) {
1221                 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
1222                 return -1;
1223         }
1224         while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
1225                                              &ext2_inode))) {
1226                 /* no more inodes */
1227                 if (ext2_ino == 0)
1228                         break;
1229                 /* skip special inode in ext2fs */
1230                 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
1231                     ext2_ino != EXT2_ROOT_INO)
1232                         continue;
1233                 objectid = ext2_ino + INO_OFFSET;
1234                 ret = copy_single_inode(trans, root,
1235                                         objectid, ext2_fs, ext2_ino,
1236                                         &ext2_inode, datacsum, packing,
1237                                         noxattr);
1238                 p->cur_copy_inodes++;
1239                 if (ret)
1240                         return ret;
1241                 if (trans->blocks_used >= 4096) {
1242                         ret = btrfs_commit_transaction(trans, root);
1243                         BUG_ON(ret);
1244                         trans = btrfs_start_transaction(root, 1);
1245                         BUG_ON(!trans);
1246                 }
1247         }
1248         if (err) {
1249                 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
1250                 return -1;
1251         }
1252         ret = btrfs_commit_transaction(trans, root);
1253         BUG_ON(ret);
1254         ext2fs_close_inode_scan(ext2_scan);
1255
1256         return ret;
1257 }
1258
1259 static int create_image_file_range(struct btrfs_trans_handle *trans,
1260                                       struct btrfs_root *root,
1261                                       struct cache_tree *used,
1262                                       struct btrfs_inode_item *inode,
1263                                       u64 ino, u64 bytenr, u64 *ret_len,
1264                                       int datacsum)
1265 {
1266         struct cache_extent *cache;
1267         struct btrfs_block_group_cache *bg_cache;
1268         u64 len = *ret_len;
1269         u64 disk_bytenr;
1270         int i;
1271         int ret;
1272
1273         BUG_ON(bytenr != round_down(bytenr, root->sectorsize));
1274         BUG_ON(len != round_down(len, root->sectorsize));
1275         len = min_t(u64, len, BTRFS_MAX_EXTENT_SIZE);
1276
1277         /*
1278          * Skip sb ranges first
1279          * [0, 1M), [sb_offset(1), +64K), [sb_offset(2), +64K].
1280          *
1281          * Or we will insert a hole into current image file, and later
1282          * migrate block will fail as there is already a file extent.
1283          */
1284         if (bytenr < 1024 * 1024) {
1285                 *ret_len = 1024 * 1024 - bytenr;
1286                 return 0;
1287         }
1288         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1289                 u64 cur = btrfs_sb_offset(i);
1290
1291                 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
1292                         *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
1293                         return 0;
1294                 }
1295         }
1296         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1297                 u64 cur = btrfs_sb_offset(i);
1298
1299                 /*
1300                  *      |--reserved--|
1301                  * |----range-------|
1302                  * May still need to go through file extent inserts
1303                  */
1304                 if (bytenr < cur && bytenr + len >= cur) {
1305                         len = min_t(u64, len, cur - bytenr);
1306                         break;
1307                 }
1308                 /*
1309                  * |--reserved--|
1310                  *      |---range---|
1311                  * Drop out, no need to insert anything
1312                  */
1313                 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
1314                         *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
1315                         return 0;
1316                 }
1317         }
1318
1319         cache = search_cache_extent(used, bytenr);
1320         if (cache) {
1321                 if (cache->start <= bytenr) {
1322                         /*
1323                          * |///////Used///////|
1324                          *      |<--insert--->|
1325                          *      bytenr
1326                          */
1327                         len = min_t(u64, len, cache->start + cache->size -
1328                                     bytenr);
1329                         disk_bytenr = bytenr;
1330                 } else {
1331                         /*
1332                          *              |//Used//|
1333                          *  |<-insert-->|
1334                          *  bytenr
1335                          */
1336                         len = min(len, cache->start - bytenr);
1337                         disk_bytenr = 0;
1338                         datacsum = 0;
1339                 }
1340         } else {
1341                 /*
1342                  * |//Used//|           |EOF
1343                  *          |<-insert-->|
1344                  *          bytenr
1345                  */
1346                 disk_bytenr = 0;
1347                 datacsum = 0;
1348         }
1349
1350         if (disk_bytenr) {
1351                 /* Check if the range is in a data block group */
1352                 bg_cache = btrfs_lookup_block_group(root->fs_info, bytenr);
1353                 if (!bg_cache)
1354                         return -ENOENT;
1355                 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
1356                         return -EINVAL;
1357
1358                 /* The extent should never cross block group boundary */
1359                 len = min_t(u64, len, bg_cache->key.objectid +
1360                             bg_cache->key.offset - bytenr);
1361         }
1362
1363         BUG_ON(len != round_down(len, root->sectorsize));
1364         ret = btrfs_record_file_extent(trans, root, ino, inode, bytenr,
1365                                        disk_bytenr, len);
1366         if (ret < 0)
1367                 return ret;
1368
1369         if (datacsum)
1370                 ret = csum_disk_extent(trans, root, bytenr, len);
1371         *ret_len = len;
1372         return ret;
1373 }
1374
1375
1376 /*
1377  * Relocate old fs data in one reserved ranges
1378  *
1379  * Since all old fs data in reserved range is not covered by any chunk nor
1380  * data extent, we don't need to handle any reference but add new
1381  * extent/reference, which makes codes more clear
1382  */
1383 static int migrate_one_reserved_range(struct btrfs_trans_handle *trans,
1384                                       struct btrfs_root *root,
1385                                       struct cache_tree *used,
1386                                       struct btrfs_inode_item *inode, int fd,
1387                                       u64 ino, u64 start, u64 len, int datacsum)
1388 {
1389         u64 cur_off = start;
1390         u64 cur_len = len;
1391         u64 hole_start = start;
1392         u64 hole_len;
1393         struct cache_extent *cache;
1394         struct btrfs_key key;
1395         struct extent_buffer *eb;
1396         int ret = 0;
1397
1398         while (cur_off < start + len) {
1399                 cache = lookup_cache_extent(used, cur_off, cur_len);
1400                 if (!cache)
1401                         break;
1402                 cur_off = max(cache->start, cur_off);
1403                 cur_len = min(cache->start + cache->size, start + len) -
1404                           cur_off;
1405                 BUG_ON(cur_len < root->sectorsize);
1406
1407                 /* reserve extent for the data */
1408                 ret = btrfs_reserve_extent(trans, root, cur_len, 0, 0, (u64)-1,
1409                                            &key, 1);
1410                 if (ret < 0)
1411                         break;
1412
1413                 eb = malloc(sizeof(*eb) + cur_len);
1414                 if (!eb) {
1415                         ret = -ENOMEM;
1416                         break;
1417                 }
1418
1419                 ret = pread(fd, eb->data, cur_len, cur_off);
1420                 if (ret < cur_len) {
1421                         ret = (ret < 0 ? ret : -EIO);
1422                         free(eb);
1423                         break;
1424                 }
1425                 eb->start = key.objectid;
1426                 eb->len = key.offset;
1427
1428                 /* Write the data */
1429                 ret = write_and_map_eb(trans, root, eb);
1430                 free(eb);
1431                 if (ret < 0)
1432                         break;
1433
1434                 /* Now handle extent item and file extent things */
1435                 ret = btrfs_record_file_extent(trans, root, ino, inode, cur_off,
1436                                                key.objectid, key.offset);
1437                 if (ret < 0)
1438                         break;
1439                 /* Finally, insert csum items */
1440                 if (datacsum)
1441                         ret = csum_disk_extent(trans, root, key.objectid,
1442                                                key.offset);
1443
1444                 /* Don't forget to insert hole */
1445                 hole_len = cur_off - hole_start;
1446                 if (hole_len) {
1447                         ret = btrfs_record_file_extent(trans, root, ino, inode,
1448                                         hole_start, 0, hole_len);
1449                         if (ret < 0)
1450                                 break;
1451                 }
1452
1453                 cur_off += key.offset;
1454                 hole_start = cur_off;
1455                 cur_len = start + len - cur_off;
1456         }
1457         /* Last hole */
1458         if (start + len - hole_start > 0)
1459                 ret = btrfs_record_file_extent(trans, root, ino, inode,
1460                                 hole_start, 0, start + len - hole_start);
1461         return ret;
1462 }
1463
1464 /*
1465  * Relocate the used ext2 data in reserved ranges
1466  * [0,1M)
1467  * [btrfs_sb_offset(1), +BTRFS_STRIPE_LEN)
1468  * [btrfs_sb_offset(2), +BTRFS_STRIPE_LEN)
1469  */
1470 static int migrate_reserved_ranges(struct btrfs_trans_handle *trans,
1471                                    struct btrfs_root *root,
1472                                    struct cache_tree *used,
1473                                    struct btrfs_inode_item *inode, int fd,
1474                                    u64 ino, u64 total_bytes, int datacsum)
1475 {
1476         u64 cur_off;
1477         u64 cur_len;
1478         int ret = 0;
1479
1480         /* 0 ~ 1M */
1481         cur_off = 0;
1482         cur_len = 1024 * 1024;
1483         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1484                                          cur_off, cur_len, datacsum);
1485         if (ret < 0)
1486                 return ret;
1487
1488         /* second sb(fisrt sb is included in 0~1M) */
1489         cur_off = btrfs_sb_offset(1);
1490         cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
1491         if (cur_off > total_bytes)
1492                 return ret;
1493         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1494                                          cur_off, cur_len, datacsum);
1495         if (ret < 0)
1496                 return ret;
1497
1498         /* Last sb */
1499         cur_off = btrfs_sb_offset(2);
1500         cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
1501         if (cur_off > total_bytes)
1502                 return ret;
1503         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1504                                          cur_off, cur_len, datacsum);
1505         return ret;
1506 }
1507
1508 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
1509                                 int ensure_size);
1510
1511 /*
1512  * Create the fs image file of old filesystem.
1513  *
1514  * This is completely fs independent as we have cctx->used, only
1515  * need to create file extents pointing to all the positions.
1516  */
1517 static int create_image(struct btrfs_root *root,
1518                            struct btrfs_mkfs_config *cfg,
1519                            struct btrfs_convert_context *cctx, int fd,
1520                            u64 size, char *name, int datacsum)
1521 {
1522         struct btrfs_inode_item buf;
1523         struct btrfs_trans_handle *trans;
1524         struct btrfs_path *path = NULL;
1525         struct btrfs_key key;
1526         struct cache_extent *cache;
1527         struct cache_tree used_tmp;
1528         u64 cur;
1529         u64 ino;
1530         u64 flags = BTRFS_INODE_READONLY;
1531         int ret;
1532
1533         if (!datacsum)
1534                 flags |= BTRFS_INODE_NODATASUM;
1535
1536         trans = btrfs_start_transaction(root, 1);
1537         if (!trans)
1538                 return -ENOMEM;
1539
1540         cache_tree_init(&used_tmp);
1541
1542         ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
1543                                        &ino);
1544         if (ret < 0)
1545                 goto out;
1546         ret = btrfs_new_inode(trans, root, ino, 0400 | S_IFREG);
1547         if (ret < 0)
1548                 goto out;
1549         ret = btrfs_change_inode_flags(trans, root, ino, flags);
1550         if (ret < 0)
1551                 goto out;
1552         ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
1553                              strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
1554         if (ret < 0)
1555                 goto out;
1556
1557         path = btrfs_alloc_path();
1558         if (!path) {
1559                 ret = -ENOMEM;
1560                 goto out;
1561         }
1562         key.objectid = ino;
1563         key.type = BTRFS_INODE_ITEM_KEY;
1564         key.offset = 0;
1565
1566         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1567         if (ret) {
1568                 ret = (ret > 0 ? -ENOENT : ret);
1569                 goto out;
1570         }
1571         read_extent_buffer(path->nodes[0], &buf,
1572                         btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1573                         sizeof(buf));
1574         btrfs_release_path(path);
1575
1576         /*
1577          * Create a new used space cache, which doesn't contain the reserved
1578          * range
1579          */
1580         for (cache = first_cache_extent(&cctx->used); cache;
1581              cache = next_cache_extent(cache)) {
1582                 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1583                 if (ret < 0)
1584                         goto out;
1585         }
1586         ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1587         if (ret < 0)
1588                 goto out;
1589
1590         /*
1591          * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1592          * can't handle bytenr 0(will consider it as a hole)
1593          */
1594         cur = 1024 * 1024;
1595         while (cur < size) {
1596                 u64 len = size - cur;
1597
1598                 ret = create_image_file_range(trans, root, &used_tmp,
1599                                                 &buf, ino, cur, &len, datacsum);
1600                 if (ret < 0)
1601                         goto out;
1602                 cur += len;
1603         }
1604         /* Handle the reserved ranges */
1605         ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1606                                       cfg->num_bytes, datacsum);
1607
1608
1609         key.objectid = ino;
1610         key.type = BTRFS_INODE_ITEM_KEY;
1611         key.offset = 0;
1612         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1613         if (ret) {
1614                 ret = (ret > 0 ? -ENOENT : ret);
1615                 goto out;
1616         }
1617         btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1618         write_extent_buffer(path->nodes[0], &buf,
1619                         btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1620                         sizeof(buf));
1621 out:
1622         free_extent_cache_tree(&used_tmp);
1623         btrfs_free_path(path);
1624         btrfs_commit_transaction(trans, root);
1625         return ret;
1626 }
1627
1628 static struct btrfs_root * link_subvol(struct btrfs_root *root,
1629                 const char *base, u64 root_objectid)
1630 {
1631         struct btrfs_trans_handle *trans;
1632         struct btrfs_fs_info *fs_info = root->fs_info;
1633         struct btrfs_root *tree_root = fs_info->tree_root;
1634         struct btrfs_root *new_root = NULL;
1635         struct btrfs_path *path;
1636         struct btrfs_inode_item *inode_item;
1637         struct extent_buffer *leaf;
1638         struct btrfs_key key;
1639         u64 dirid = btrfs_root_dirid(&root->root_item);
1640         u64 index = 2;
1641         char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1642         int len;
1643         int i;
1644         int ret;
1645
1646         len = strlen(base);
1647         if (len == 0 || len > BTRFS_NAME_LEN)
1648                 return NULL;
1649
1650         path = btrfs_alloc_path();
1651         BUG_ON(!path);
1652
1653         key.objectid = dirid;
1654         key.type = BTRFS_DIR_INDEX_KEY;
1655         key.offset = (u64)-1;
1656
1657         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1658         BUG_ON(ret <= 0);
1659
1660         if (path->slots[0] > 0) {
1661                 path->slots[0]--;
1662                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1663                 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1664                         index = key.offset + 1;
1665         }
1666         btrfs_release_path(path);
1667
1668         trans = btrfs_start_transaction(root, 1);
1669         BUG_ON(!trans);
1670
1671         key.objectid = dirid;
1672         key.offset = 0;
1673         key.type =  BTRFS_INODE_ITEM_KEY;
1674
1675         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1676         BUG_ON(ret);
1677         leaf = path->nodes[0];
1678         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1679                                     struct btrfs_inode_item);
1680
1681         key.objectid = root_objectid;
1682         key.offset = (u64)-1;
1683         key.type = BTRFS_ROOT_ITEM_KEY;
1684
1685         memcpy(buf, base, len);
1686         for (i = 0; i < 1024; i++) {
1687                 ret = btrfs_insert_dir_item(trans, root, buf, len,
1688                                             dirid, &key, BTRFS_FT_DIR, index);
1689                 if (ret != -EEXIST)
1690                         break;
1691                 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1692                 if (len < 1 || len > BTRFS_NAME_LEN) {
1693                         ret = -EINVAL;
1694                         break;
1695                 }
1696         }
1697         if (ret)
1698                 goto fail;
1699
1700         btrfs_set_inode_size(leaf, inode_item, len * 2 +
1701                              btrfs_inode_size(leaf, inode_item));
1702         btrfs_mark_buffer_dirty(leaf);
1703         btrfs_release_path(path);
1704
1705         /* add the backref first */
1706         ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1707                                  BTRFS_ROOT_BACKREF_KEY,
1708                                  root->root_key.objectid,
1709                                  dirid, index, buf, len);
1710         BUG_ON(ret);
1711
1712         /* now add the forward ref */
1713         ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1714                                  BTRFS_ROOT_REF_KEY, root_objectid,
1715                                  dirid, index, buf, len);
1716
1717         ret = btrfs_commit_transaction(trans, root);
1718         BUG_ON(ret);
1719
1720         new_root = btrfs_read_fs_root(fs_info, &key);
1721         if (IS_ERR(new_root))
1722                 new_root = NULL;
1723 fail:
1724         btrfs_free_path(path);
1725         return new_root;
1726 }
1727
1728 static int create_subvol(struct btrfs_trans_handle *trans,
1729                          struct btrfs_root *root, u64 root_objectid)
1730 {
1731         struct extent_buffer *tmp;
1732         struct btrfs_root *new_root;
1733         struct btrfs_key key;
1734         struct btrfs_root_item root_item;
1735         int ret;
1736
1737         ret = btrfs_copy_root(trans, root, root->node, &tmp,
1738                               root_objectid);
1739         BUG_ON(ret);
1740
1741         memcpy(&root_item, &root->root_item, sizeof(root_item));
1742         btrfs_set_root_bytenr(&root_item, tmp->start);
1743         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
1744         btrfs_set_root_generation(&root_item, trans->transid);
1745         free_extent_buffer(tmp);
1746
1747         key.objectid = root_objectid;
1748         key.type = BTRFS_ROOT_ITEM_KEY;
1749         key.offset = trans->transid;
1750         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1751                                 &key, &root_item);
1752
1753         key.offset = (u64)-1;
1754         new_root = btrfs_read_fs_root(root->fs_info, &key);
1755         BUG_ON(!new_root || IS_ERR(new_root));
1756
1757         ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
1758         BUG_ON(ret);
1759
1760         return 0;
1761 }
1762
1763 /*
1764  * New make_btrfs() has handle system and meta chunks quite well.
1765  * So only need to add remaining data chunks.
1766  */
1767 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
1768                                           struct btrfs_fs_info *fs_info,
1769                                           struct btrfs_mkfs_config *cfg,
1770                                           struct btrfs_convert_context *cctx)
1771 {
1772         struct btrfs_root *extent_root = fs_info->extent_root;
1773         struct cache_tree *data_chunks = &cctx->data_chunks;
1774         struct cache_extent *cache;
1775         u64 max_chunk_size;
1776         int ret = 0;
1777
1778         /*
1779          * Don't create data chunk over 10% of the convert device
1780          * And for single chunk, don't create chunk larger than 1G.
1781          */
1782         max_chunk_size = cfg->num_bytes / 10;
1783         max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
1784         max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
1785
1786         for (cache = first_cache_extent(data_chunks); cache;
1787              cache = next_cache_extent(cache)) {
1788                 u64 cur = cache->start;
1789
1790                 while (cur < cache->start + cache->size) {
1791                         u64 len;
1792                         u64 cur_backup = cur;
1793
1794                         len = min(max_chunk_size,
1795                                   cache->start + cache->size - cur);
1796                         ret = btrfs_alloc_data_chunk(trans, extent_root,
1797                                         &cur_backup, len,
1798                                         BTRFS_BLOCK_GROUP_DATA, 1);
1799                         if (ret < 0)
1800                                 break;
1801                         ret = btrfs_make_block_group(trans, extent_root, 0,
1802                                         BTRFS_BLOCK_GROUP_DATA,
1803                                         BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1804                                         cur, len);
1805                         if (ret < 0)
1806                                 break;
1807                         cur += len;
1808                 }
1809         }
1810         return ret;
1811 }
1812
1813 /*
1814  * Init the temp btrfs to a operational status.
1815  *
1816  * It will fix the extent usage accounting(XXX: Do we really need?) and
1817  * insert needed data chunks, to ensure all old fs data extents are covered
1818  * by DATA chunks, preventing wrong chunks are allocated.
1819  *
1820  * And also create convert image subvolume and relocation tree.
1821  * (XXX: Not need again?)
1822  * But the convert image subvolume is *NOT* linked to fs tree yet.
1823  */
1824 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
1825                          struct btrfs_convert_context *cctx, int datacsum,
1826                          int packing, int noxattr)
1827 {
1828         struct btrfs_key location;
1829         struct btrfs_trans_handle *trans;
1830         struct btrfs_fs_info *fs_info = root->fs_info;
1831         int ret;
1832
1833         /*
1834          * Don't alloc any metadata/system chunk, as we don't want
1835          * any meta/sys chunk allcated before all data chunks are inserted.
1836          * Or we screw up the chunk layout just like the old implement.
1837          */
1838         fs_info->avoid_sys_chunk_alloc = 1;
1839         fs_info->avoid_meta_chunk_alloc = 1;
1840         trans = btrfs_start_transaction(root, 1);
1841         BUG_ON(!trans);
1842         ret = btrfs_fix_block_accounting(trans, root);
1843         if (ret)
1844                 goto err;
1845         ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
1846         if (ret)
1847                 goto err;
1848         ret = btrfs_make_root_dir(trans, fs_info->tree_root,
1849                                   BTRFS_ROOT_TREE_DIR_OBJECTID);
1850         if (ret)
1851                 goto err;
1852         memcpy(&location, &root->root_key, sizeof(location));
1853         location.offset = (u64)-1;
1854         ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
1855                                 btrfs_super_root_dir(fs_info->super_copy),
1856                                 &location, BTRFS_FT_DIR, 0);
1857         if (ret)
1858                 goto err;
1859         ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
1860                                 location.objectid,
1861                                 btrfs_super_root_dir(fs_info->super_copy), 0);
1862         if (ret)
1863                 goto err;
1864         btrfs_set_root_dirid(&fs_info->fs_root->root_item,
1865                              BTRFS_FIRST_FREE_OBJECTID);
1866
1867         /* subvol for fs image file */
1868         ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
1869         if (ret < 0)
1870                 goto err;
1871         /* subvol for data relocation tree */
1872         ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1873         if (ret < 0)
1874                 goto err;
1875
1876         ret = btrfs_commit_transaction(trans, root);
1877         fs_info->avoid_sys_chunk_alloc = 0;
1878         fs_info->avoid_meta_chunk_alloc = 0;
1879 err:
1880         return ret;
1881 }
1882
1883 /*
1884  * Migrate super block to its default position and zero 0 ~ 16k
1885  */
1886 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
1887 {
1888         int ret;
1889         struct extent_buffer *buf;
1890         struct btrfs_super_block *super;
1891         u32 len;
1892         u32 bytenr;
1893
1894         BUG_ON(sectorsize < sizeof(*super));
1895         buf = malloc(sizeof(*buf) + sectorsize);
1896         if (!buf)
1897                 return -ENOMEM;
1898
1899         buf->len = sectorsize;
1900         ret = pread(fd, buf->data, sectorsize, old_bytenr);
1901         if (ret != sectorsize)
1902                 goto fail;
1903
1904         super = (struct btrfs_super_block *)buf->data;
1905         BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
1906         btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
1907
1908         csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1909         ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
1910         if (ret != sectorsize)
1911                 goto fail;
1912
1913         ret = fsync(fd);
1914         if (ret)
1915                 goto fail;
1916
1917         memset(buf->data, 0, sectorsize);
1918         for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
1919                 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
1920                 if (len > sectorsize)
1921                         len = sectorsize;
1922                 ret = pwrite(fd, buf->data, len, bytenr);
1923                 if (ret != len) {
1924                         fprintf(stderr, "unable to zero fill device\n");
1925                         break;
1926                 }
1927                 bytenr += len;
1928         }
1929         ret = 0;
1930         fsync(fd);
1931 fail:
1932         free(buf);
1933         if (ret > 0)
1934                 ret = -1;
1935         return ret;
1936 }
1937
1938 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
1939 {
1940         struct btrfs_chunk *chunk;
1941         struct btrfs_disk_key *key;
1942         u32 sectorsize = btrfs_super_sectorsize(super);
1943
1944         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1945         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1946                                        sizeof(struct btrfs_disk_key));
1947
1948         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1949         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1950         btrfs_set_disk_key_offset(key, 0);
1951
1952         btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
1953         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1954         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1955         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1956         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1957         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1958         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1959         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1960         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1961         chunk->stripe.devid = super->dev_item.devid;
1962         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1963         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1964         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1965         return 0;
1966 }
1967
1968 static const struct btrfs_convert_operations ext2_convert_ops = {
1969         .name                   = "ext2",
1970         .open_fs                = ext2_open_fs,
1971         .read_used_space        = ext2_read_used_space,
1972         .copy_inodes            = ext2_copy_inodes,
1973         .close_fs               = ext2_close_fs,
1974 };
1975
1976 static const struct btrfs_convert_operations *convert_operations[] = {
1977         &ext2_convert_ops,
1978 };
1979
1980 static int convert_open_fs(const char *devname,
1981                            struct btrfs_convert_context *cctx)
1982 {
1983         int i;
1984
1985         memset(cctx, 0, sizeof(*cctx));
1986
1987         for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
1988                 int ret = convert_operations[i]->open_fs(cctx, devname);
1989
1990                 if (ret == 0) {
1991                         cctx->convert_ops = convert_operations[i];
1992                         return ret;
1993                 }
1994         }
1995
1996         fprintf(stderr, "No file system found to convert.\n");
1997         return -1;
1998 }
1999
2000 /*
2001  * Helper for expand and merge extent_cache for wipe_one_reserved_range() to
2002  * handle wiping a range that exists in cache.
2003  */
2004 static int _expand_extent_cache(struct cache_tree *tree,
2005                                 struct cache_extent *entry,
2006                                 u64 min_stripe_size, int backward)
2007 {
2008         struct cache_extent *ce;
2009         int diff;
2010
2011         if (entry->size >= min_stripe_size)
2012                 return 0;
2013         diff = min_stripe_size - entry->size;
2014
2015         if (backward) {
2016                 ce = prev_cache_extent(entry);
2017                 if (!ce)
2018                         goto expand_back;
2019                 if (ce->start + ce->size >= entry->start - diff) {
2020                         /* Directly merge with previous extent */
2021                         ce->size = entry->start + entry->size - ce->start;
2022                         remove_cache_extent(tree, entry);
2023                         free(entry);
2024                         return 0;
2025                 }
2026 expand_back:
2027                 /* No overlap, normal extent */
2028                 if (entry->start < diff) {
2029                         error("cannot find space for data chunk layout");
2030                         return -ENOSPC;
2031                 }
2032                 entry->start -= diff;
2033                 entry->size += diff;
2034                 return 0;
2035         }
2036         ce = next_cache_extent(entry);
2037         if (!ce)
2038                 goto expand_after;
2039         if (entry->start + entry->size + diff >= ce->start) {
2040                 /* Directly merge with next extent */
2041                 entry->size = ce->start + ce->size - entry->start;
2042                 remove_cache_extent(tree, ce);
2043                 free(ce);
2044                 return 0;
2045         }
2046 expand_after:
2047         entry->size += diff;
2048         return 0;
2049 }
2050
2051 /*
2052  * Remove one reserve range from given cache tree
2053  * if min_stripe_size is non-zero, it will ensure for split case,
2054  * all its split cache extent is no smaller than @min_strip_size / 2.
2055  */
2056 static int wipe_one_reserved_range(struct cache_tree *tree,
2057                                    u64 start, u64 len, u64 min_stripe_size,
2058                                    int ensure_size)
2059 {
2060         struct cache_extent *cache;
2061         int ret;
2062
2063         BUG_ON(ensure_size && min_stripe_size == 0);
2064         /*
2065          * The logical here is simplified to handle special cases only
2066          * So we don't need to consider merge case for ensure_size
2067          */
2068         BUG_ON(min_stripe_size && (min_stripe_size < len * 2 ||
2069                min_stripe_size / 2 < BTRFS_STRIPE_LEN));
2070
2071         /* Also, wipe range should already be aligned */
2072         BUG_ON(start != round_down(start, BTRFS_STRIPE_LEN) ||
2073                start + len != round_up(start + len, BTRFS_STRIPE_LEN));
2074
2075         min_stripe_size /= 2;
2076
2077         cache = lookup_cache_extent(tree, start, len);
2078         if (!cache)
2079                 return 0;
2080
2081         if (start <= cache->start) {
2082                 /*
2083                  *      |--------cache---------|
2084                  * |-wipe-|
2085                  */
2086                 BUG_ON(start + len <= cache->start);
2087
2088                 /*
2089                  * The wipe size is smaller than min_stripe_size / 2,
2090                  * so the result length should still meet min_stripe_size
2091                  * And no need to do alignment
2092                  */
2093                 cache->size -= (start + len - cache->start);
2094                 if (cache->size == 0) {
2095                         remove_cache_extent(tree, cache);
2096                         free(cache);
2097                         return 0;
2098                 }
2099
2100                 BUG_ON(ensure_size && cache->size < min_stripe_size);
2101
2102                 cache->start = start + len;
2103                 return 0;
2104         } else if (start > cache->start && start + len < cache->start +
2105                    cache->size) {
2106                 /*
2107                  * |-------cache-----|
2108                  *      |-wipe-|
2109                  */
2110                 u64 old_start = cache->start;
2111                 u64 old_len = cache->size;
2112                 u64 insert_start = start + len;
2113                 u64 insert_len;
2114
2115                 cache->size = start - cache->start;
2116                 /* Expand the leading half part if needed */
2117                 if (ensure_size && cache->size < min_stripe_size) {
2118                         ret = _expand_extent_cache(tree, cache,
2119                                         min_stripe_size, 1);
2120                         if (ret < 0)
2121                                 return ret;
2122                 }
2123
2124                 /* And insert the new one */
2125                 insert_len = old_start + old_len - start - len;
2126                 ret = add_merge_cache_extent(tree, insert_start, insert_len);
2127                 if (ret < 0)
2128                         return ret;
2129
2130                 /* Expand the last half part if needed */
2131                 if (ensure_size && insert_len < min_stripe_size) {
2132                         cache = lookup_cache_extent(tree, insert_start,
2133                                                     insert_len);
2134                         if (!cache || cache->start != insert_start ||
2135                             cache->size != insert_len)
2136                                 return -ENOENT;
2137                         ret = _expand_extent_cache(tree, cache,
2138                                         min_stripe_size, 0);
2139                 }
2140
2141                 return ret;
2142         }
2143         /*
2144          * |----cache-----|
2145          *              |--wipe-|
2146          * Wipe len should be small enough and no need to expand the
2147          * remaining extent
2148          */
2149         cache->size = start - cache->start;
2150         BUG_ON(ensure_size && cache->size < min_stripe_size);
2151         return 0;
2152 }
2153
2154 /*
2155  * Remove reserved ranges from given cache_tree
2156  *
2157  * It will remove the following ranges
2158  * 1) 0~1M
2159  * 2) 2nd superblock, +64K (make sure chunks are 64K aligned)
2160  * 3) 3rd superblock, +64K
2161  *
2162  * @min_stripe must be given for safety check
2163  * and if @ensure_size is given, it will ensure affected cache_extent will be
2164  * larger than min_stripe_size
2165  */
2166 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
2167                                 int ensure_size)
2168 {
2169         int ret;
2170
2171         ret = wipe_one_reserved_range(tree, 0, 1024 * 1024, min_stripe_size,
2172                                       ensure_size);
2173         if (ret < 0)
2174                 return ret;
2175         ret = wipe_one_reserved_range(tree, btrfs_sb_offset(1),
2176                         BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
2177         if (ret < 0)
2178                 return ret;
2179         ret = wipe_one_reserved_range(tree, btrfs_sb_offset(2),
2180                         BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
2181         return ret;
2182 }
2183
2184 static int calculate_available_space(struct btrfs_convert_context *cctx)
2185 {
2186         struct cache_tree *used = &cctx->used;
2187         struct cache_tree *data_chunks = &cctx->data_chunks;
2188         struct cache_tree *free = &cctx->free;
2189         struct cache_extent *cache;
2190         u64 cur_off = 0;
2191         /*
2192          * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
2193          * works without need to consider overlap
2194          */
2195         u64 min_stripe_size = 2 * 16 * 1024 * 1024;
2196         int ret;
2197
2198         /* Calculate data_chunks */
2199         for (cache = first_cache_extent(used); cache;
2200              cache = next_cache_extent(cache)) {
2201                 u64 cur_len;
2202
2203                 if (cache->start + cache->size < cur_off)
2204                         continue;
2205                 if (cache->start > cur_off + min_stripe_size)
2206                         cur_off = cache->start;
2207                 cur_len = max(cache->start + cache->size - cur_off,
2208                               min_stripe_size);
2209                 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
2210                 if (ret < 0)
2211                         goto out;
2212                 cur_off += cur_len;
2213         }
2214         /*
2215          * remove reserved ranges, so we won't ever bother relocating an old
2216          * filesystem extent to other place.
2217          */
2218         ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
2219         if (ret < 0)
2220                 goto out;
2221
2222         cur_off = 0;
2223         /*
2224          * Calculate free space
2225          * Always round up the start bytenr, to avoid metadata extent corss
2226          * stripe boundary, as later mkfs_convert() won't have all the extent
2227          * allocation check
2228          */
2229         for (cache = first_cache_extent(data_chunks); cache;
2230              cache = next_cache_extent(cache)) {
2231                 if (cache->start < cur_off)
2232                         continue;
2233                 if (cache->start > cur_off) {
2234                         u64 insert_start;
2235                         u64 len;
2236
2237                         len = cache->start - round_up(cur_off,
2238                                                       BTRFS_STRIPE_LEN);
2239                         insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2240
2241                         ret = add_merge_cache_extent(free, insert_start, len);
2242                         if (ret < 0)
2243                                 goto out;
2244                 }
2245                 cur_off = cache->start + cache->size;
2246         }
2247         /* Don't forget the last range */
2248         if (cctx->total_bytes > cur_off) {
2249                 u64 len = cctx->total_bytes - cur_off;
2250                 u64 insert_start;
2251
2252                 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2253
2254                 ret = add_merge_cache_extent(free, insert_start, len);
2255                 if (ret < 0)
2256                         goto out;
2257         }
2258
2259         /* Remove reserved bytes */
2260         ret = wipe_reserved_ranges(free, min_stripe_size, 0);
2261 out:
2262         return ret;
2263 }
2264 /*
2265  * Read used space, and since we have the used space,
2266  * calcuate data_chunks and free for later mkfs
2267  */
2268 static int convert_read_used_space(struct btrfs_convert_context *cctx)
2269 {
2270         int ret;
2271
2272         ret = cctx->convert_ops->read_used_space(cctx);
2273         if (ret)
2274                 return ret;
2275
2276         ret = calculate_available_space(cctx);
2277         return ret;
2278 }
2279
2280 static int do_convert(const char *devname, int datacsum, int packing,
2281                 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2282                 int progress, u64 features)
2283 {
2284         int ret;
2285         int fd = -1;
2286         int is_btrfs = 0;
2287         u32 blocksize;
2288         u64 total_bytes;
2289         struct btrfs_root *root;
2290         struct btrfs_root *image_root;
2291         struct btrfs_convert_context cctx;
2292         struct btrfs_key key;
2293         char *subvol_name = NULL;
2294         struct task_ctx ctx;
2295         char features_buf[64];
2296         struct btrfs_mkfs_config mkfs_cfg;
2297
2298         init_convert_context(&cctx);
2299         ret = convert_open_fs(devname, &cctx);
2300         if (ret)
2301                 goto fail;
2302         ret = convert_read_used_space(&cctx);
2303         if (ret)
2304                 goto fail;
2305
2306         blocksize = cctx.blocksize;
2307         total_bytes = (u64)blocksize * (u64)cctx.block_count;
2308         if (blocksize < 4096) {
2309                 fprintf(stderr, "block size is too small\n");
2310                 goto fail;
2311         }
2312         if (btrfs_check_nodesize(nodesize, blocksize, features))
2313                 goto fail;
2314         fd = open(devname, O_RDWR);
2315         if (fd < 0) {
2316                 fprintf(stderr, "unable to open %s\n", devname);
2317                 goto fail;
2318         }
2319         btrfs_parse_features_to_string(features_buf, features);
2320         if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2321                 strcat(features_buf, " (default)");
2322
2323         printf("create btrfs filesystem:\n");
2324         printf("\tblocksize: %u\n", blocksize);
2325         printf("\tnodesize:  %u\n", nodesize);
2326         printf("\tfeatures:  %s\n", features_buf);
2327
2328         mkfs_cfg.label = cctx.volume_name;
2329         mkfs_cfg.num_bytes = total_bytes;
2330         mkfs_cfg.nodesize = nodesize;
2331         mkfs_cfg.sectorsize = blocksize;
2332         mkfs_cfg.stripesize = blocksize;
2333         mkfs_cfg.features = features;
2334         /* New convert need these space */
2335         mkfs_cfg.fs_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2336         mkfs_cfg.chunk_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2337         *(mkfs_cfg.fs_uuid) = '\0';
2338         *(mkfs_cfg.chunk_uuid) = '\0';
2339
2340         ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2341         if (ret) {
2342                 fprintf(stderr, "unable to create initial ctree: %s\n",
2343                         strerror(-ret));
2344                 goto fail;
2345         }
2346
2347         root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2348                              OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2349         if (!root) {
2350                 fprintf(stderr, "unable to open ctree\n");
2351                 goto fail;
2352         }
2353         ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2354         if (ret) {
2355                 fprintf(stderr, "unable to setup the root tree\n");
2356                 goto fail;
2357         }
2358
2359         printf("creating %s image file.\n", cctx.convert_ops->name);
2360         ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2361         if (ret < 0) {
2362                 fprintf(stderr, "error allocating subvolume name: %s_saved\n",
2363                         cctx.convert_ops->name);
2364                 goto fail;
2365         }
2366         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2367         key.offset = (u64)-1;
2368         key.type = BTRFS_ROOT_ITEM_KEY;
2369         image_root = btrfs_read_fs_root(root->fs_info, &key);
2370         if (!image_root) {
2371                 fprintf(stderr, "unable to create subvol\n");
2372                 goto fail;
2373         }
2374         ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2375                               mkfs_cfg.num_bytes, "image", datacsum);
2376         if (ret) {
2377                 fprintf(stderr, "error during create_image %d\n", ret);
2378                 goto fail;
2379         }
2380
2381         printf("creating btrfs metadata.\n");
2382         ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2383         ctx.cur_copy_inodes = 0;
2384
2385         if (progress) {
2386                 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2387                                      &ctx);
2388                 task_start(ctx.info);
2389         }
2390         ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2391         if (ret) {
2392                 fprintf(stderr, "error during copy_inodes %d\n", ret);
2393                 goto fail;
2394         }
2395         if (progress) {
2396                 task_stop(ctx.info);
2397                 task_deinit(ctx.info);
2398         }
2399
2400         image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2401
2402         free(subvol_name);
2403
2404         memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2405         if (copylabel == 1) {
2406                 __strncpy_null(root->fs_info->super_copy->label,
2407                                 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2408                 fprintf(stderr, "copy label '%s'\n",
2409                                 root->fs_info->super_copy->label);
2410         } else if (copylabel == -1) {
2411                 strcpy(root->fs_info->super_copy->label, fslabel);
2412                 fprintf(stderr, "set label to '%s'\n", fslabel);
2413         }
2414
2415         ret = close_ctree(root);
2416         if (ret) {
2417                 fprintf(stderr, "error during close_ctree %d\n", ret);
2418                 goto fail;
2419         }
2420         convert_close_fs(&cctx);
2421         clean_convert_context(&cctx);
2422
2423         /*
2424          * If this step succeed, we get a mountable btrfs. Otherwise
2425          * the source fs is left unchanged.
2426          */
2427         ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2428         if (ret) {
2429                 fprintf(stderr, "unable to migrate super block\n");
2430                 goto fail;
2431         }
2432         is_btrfs = 1;
2433
2434         root = open_ctree_fd(fd, devname, 0,
2435                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2436         if (!root) {
2437                 fprintf(stderr, "unable to open ctree\n");
2438                 goto fail;
2439         }
2440         root->fs_info->finalize_on_close = 1;
2441         close_ctree(root);
2442         close(fd);
2443
2444         printf("conversion complete.\n");
2445         return 0;
2446 fail:
2447         clean_convert_context(&cctx);
2448         if (fd != -1)
2449                 close(fd);
2450         if (is_btrfs)
2451                 fprintf(stderr,
2452                         "WARNING: an error occurred during chunk mapping fixup, filesystem mountable but not finalized\n");
2453         else
2454                 fprintf(stderr, "conversion aborted\n");
2455         return -1;
2456 }
2457
2458 /*
2459  * Check if a non 1:1 mapped chunk can be rolled back.
2460  * For new convert, it's OK while for old convert it's not.
2461  */
2462 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2463 {
2464         struct btrfs_block_group_cache *bg;
2465         struct btrfs_key key;
2466         struct btrfs_path path;
2467         struct btrfs_root *extent_root = fs_info->extent_root;
2468         u64 bg_start;
2469         u64 bg_end;
2470         int ret;
2471
2472         bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2473         if (!bg)
2474                 return -ENOENT;
2475         bg_start = bg->key.objectid;
2476         bg_end = bg->key.objectid + bg->key.offset;
2477
2478         key.objectid = bg_end;
2479         key.type = BTRFS_METADATA_ITEM_KEY;
2480         key.offset = 0;
2481         btrfs_init_path(&path);
2482
2483         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2484         if (ret < 0)
2485                 return ret;
2486
2487         while (1) {
2488                 struct btrfs_extent_item *ei;
2489
2490                 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2491                 if (ret > 0) {
2492                         ret = 0;
2493                         break;
2494                 }
2495                 if (ret < 0)
2496                         break;
2497
2498                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2499                 if (key.type == BTRFS_METADATA_ITEM_KEY)
2500                         continue;
2501                 /* Now it's EXTENT_ITEM_KEY only */
2502                 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2503                                     struct btrfs_extent_item);
2504                 /*
2505                  * Found data extent, means this is old convert must follow 1:1
2506                  * mapping.
2507                  */
2508                 if (btrfs_extent_flags(path.nodes[0], ei)
2509                                 & BTRFS_EXTENT_FLAG_DATA) {
2510                         ret = -EINVAL;
2511                         break;
2512                 }
2513         }
2514         btrfs_release_path(&path);
2515         return ret;
2516 }
2517
2518 static int may_rollback(struct btrfs_root *root)
2519 {
2520         struct btrfs_fs_info *info = root->fs_info;
2521         struct btrfs_multi_bio *multi = NULL;
2522         u64 bytenr;
2523         u64 length;
2524         u64 physical;
2525         u64 total_bytes;
2526         int num_stripes;
2527         int ret;
2528
2529         if (btrfs_super_num_devices(info->super_copy) != 1)
2530                 goto fail;
2531
2532         bytenr = BTRFS_SUPER_INFO_OFFSET;
2533         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2534
2535         while (1) {
2536                 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2537                                       &length, &multi, 0, NULL);
2538                 if (ret) {
2539                         if (ret == -ENOENT) {
2540                                 /* removed block group at the tail */
2541                                 if (length == (u64)-1)
2542                                         break;
2543
2544                                 /* removed block group in the middle */
2545                                 goto next;
2546                         }
2547                         goto fail;
2548                 }
2549
2550                 num_stripes = multi->num_stripes;
2551                 physical = multi->stripes[0].physical;
2552                 kfree(multi);
2553
2554                 if (num_stripes != 1) {
2555                         error("num stripes for bytenr %llu is not 1", bytenr);
2556                         goto fail;
2557                 }
2558
2559                 /*
2560                  * Extra check for new convert, as metadata chunk from new
2561                  * convert is much more free than old convert, it doesn't need
2562                  * to do 1:1 mapping.
2563                  */
2564                 if (physical != bytenr) {
2565                         /*
2566                          * Check if it's a metadata chunk and has only metadata
2567                          * extent.
2568                          */
2569                         ret = may_rollback_chunk(info, bytenr);
2570                         if (ret < 0)
2571                                 goto fail;
2572                 }
2573 next:
2574                 bytenr += length;
2575                 if (bytenr >= total_bytes)
2576                         break;
2577         }
2578         return 0;
2579 fail:
2580         return -1;
2581 }
2582
2583 static int do_rollback(const char *devname)
2584 {
2585         int fd = -1;
2586         int ret;
2587         int i;
2588         struct btrfs_root *root;
2589         struct btrfs_root *image_root;
2590         struct btrfs_root *chunk_root;
2591         struct btrfs_dir_item *dir;
2592         struct btrfs_inode_item *inode;
2593         struct btrfs_file_extent_item *fi;
2594         struct btrfs_trans_handle *trans;
2595         struct extent_buffer *leaf;
2596         struct btrfs_block_group_cache *cache1;
2597         struct btrfs_block_group_cache *cache2;
2598         struct btrfs_key key;
2599         struct btrfs_path path;
2600         struct extent_io_tree io_tree;
2601         char *buf = NULL;
2602         char *name;
2603         u64 bytenr;
2604         u64 num_bytes;
2605         u64 root_dir;
2606         u64 objectid;
2607         u64 offset;
2608         u64 start;
2609         u64 end;
2610         u64 sb_bytenr;
2611         u64 first_free;
2612         u64 total_bytes;
2613         u32 sectorsize;
2614
2615         extent_io_tree_init(&io_tree);
2616
2617         fd = open(devname, O_RDWR);
2618         if (fd < 0) {
2619                 fprintf(stderr, "unable to open %s\n", devname);
2620                 goto fail;
2621         }
2622         root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2623         if (!root) {
2624                 fprintf(stderr, "unable to open ctree\n");
2625                 goto fail;
2626         }
2627         ret = may_rollback(root);
2628         if (ret < 0) {
2629                 fprintf(stderr, "unable to do rollback\n");
2630                 goto fail;
2631         }
2632
2633         sectorsize = root->sectorsize;
2634         buf = malloc(sectorsize);
2635         if (!buf) {
2636                 fprintf(stderr, "unable to allocate memory\n");
2637                 goto fail;
2638         }
2639
2640         btrfs_init_path(&path);
2641
2642         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2643         key.type = BTRFS_ROOT_BACKREF_KEY;
2644         key.offset = BTRFS_FS_TREE_OBJECTID;
2645         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2646                                 0);
2647         btrfs_release_path(&path);
2648         if (ret > 0) {
2649                 fprintf(stderr,
2650                 "ERROR: unable to convert ext2 image subvolume, is it deleted?\n");
2651                 goto fail;
2652         } else if (ret < 0) {
2653                 fprintf(stderr,
2654                         "ERROR: unable to open ext2_saved, id=%llu: %s\n",
2655                         (unsigned long long)key.objectid, strerror(-ret));
2656                 goto fail;
2657         }
2658
2659         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2660         key.type = BTRFS_ROOT_ITEM_KEY;
2661         key.offset = (u64)-1;
2662         image_root = btrfs_read_fs_root(root->fs_info, &key);
2663         if (!image_root || IS_ERR(image_root)) {
2664                 fprintf(stderr, "unable to open subvol %llu\n",
2665                         (unsigned long long)key.objectid);
2666                 goto fail;
2667         }
2668
2669         name = "image";
2670         root_dir = btrfs_root_dirid(&root->root_item);
2671         dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2672                                    root_dir, name, strlen(name), 0);
2673         if (!dir || IS_ERR(dir)) {
2674                 fprintf(stderr, "unable to find file %s\n", name);
2675                 goto fail;
2676         }
2677         leaf = path.nodes[0];
2678         btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2679         btrfs_release_path(&path);
2680
2681         objectid = key.objectid;
2682
2683         ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2684         if (ret) {
2685                 fprintf(stderr, "unable to find inode item\n");
2686                 goto fail;
2687         }
2688         leaf = path.nodes[0];
2689         inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2690         total_bytes = btrfs_inode_size(leaf, inode);
2691         btrfs_release_path(&path);
2692
2693         key.objectid = objectid;
2694         key.offset = 0;
2695         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2696         ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2697         if (ret != 0) {
2698                 fprintf(stderr, "unable to find first file extent\n");
2699                 btrfs_release_path(&path);
2700                 goto fail;
2701         }
2702
2703         /* build mapping tree for the relocated blocks */
2704         for (offset = 0; offset < total_bytes; ) {
2705                 leaf = path.nodes[0];
2706                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2707                         ret = btrfs_next_leaf(root, &path);
2708                         if (ret != 0)
2709                                 break;
2710                         continue;
2711                 }
2712
2713                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2714                 if (key.objectid != objectid || key.offset != offset ||
2715                     btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2716                         break;
2717
2718                 fi = btrfs_item_ptr(leaf, path.slots[0],
2719                                     struct btrfs_file_extent_item);
2720                 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2721                         break;
2722                 if (btrfs_file_extent_compression(leaf, fi) ||
2723                     btrfs_file_extent_encryption(leaf, fi) ||
2724                     btrfs_file_extent_other_encoding(leaf, fi))
2725                         break;
2726
2727                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2728                 /* skip holes and direct mapped extents */
2729                 if (bytenr == 0 || bytenr == offset)
2730                         goto next_extent;
2731
2732                 bytenr += btrfs_file_extent_offset(leaf, fi);
2733                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2734
2735                 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2736                 cache2 = btrfs_lookup_block_group(root->fs_info,
2737                                                   offset + num_bytes - 1);
2738                 /*
2739                  * Here we must take consideration of old and new convert
2740                  * behavior.
2741                  * For old convert case, sign, there is no consist chunk type
2742                  * that will cover the extent. META/DATA/SYS are all possible.
2743                  * Just ensure relocate one is in SYS chunk.
2744                  * For new convert case, they are all covered by DATA chunk.
2745                  *
2746                  * So, there is not valid chunk type check for it now.
2747                  */
2748                 if (cache1 != cache2)
2749                         break;
2750
2751                 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2752                                 EXTENT_LOCKED, GFP_NOFS);
2753                 set_state_private(&io_tree, offset, bytenr);
2754 next_extent:
2755                 offset += btrfs_file_extent_num_bytes(leaf, fi);
2756                 path.slots[0]++;
2757         }
2758         btrfs_release_path(&path);
2759
2760         if (offset < total_bytes) {
2761                 fprintf(stderr, "unable to build extent mapping\n");
2762                 fprintf(stderr, "converted filesystem after balance is unable to rollback\n");
2763                 goto fail;
2764         }
2765
2766         first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2767         first_free &= ~((u64)sectorsize - 1);
2768         /* backup for extent #0 should exist */
2769         if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2770                 fprintf(stderr, "no backup for the first extent\n");
2771                 goto fail;
2772         }
2773         /* force no allocation from system block group */
2774         root->fs_info->system_allocs = -1;
2775         trans = btrfs_start_transaction(root, 1);
2776         BUG_ON(!trans);
2777         /*
2778          * recow the whole chunk tree, this will remove all chunk tree blocks
2779          * from system block group
2780          */
2781         chunk_root = root->fs_info->chunk_root;
2782         memset(&key, 0, sizeof(key));
2783         while (1) {
2784                 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2785                 if (ret < 0)
2786                         break;
2787
2788                 ret = btrfs_next_leaf(chunk_root, &path);
2789                 if (ret)
2790                         break;
2791
2792                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2793                 btrfs_release_path(&path);
2794         }
2795         btrfs_release_path(&path);
2796
2797         offset = 0;
2798         num_bytes = 0;
2799         while(1) {
2800                 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2801                 if (!cache1)
2802                         break;
2803
2804                 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2805                         num_bytes += btrfs_block_group_used(&cache1->item);
2806
2807                 offset = cache1->key.objectid + cache1->key.offset;
2808         }
2809         /* only extent #0 left in system block group? */
2810         if (num_bytes > first_free) {
2811                 fprintf(stderr, "unable to empty system block group\n");
2812                 goto fail;
2813         }
2814         /* create a system chunk that maps the whole device */
2815         ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2816         if (ret) {
2817                 fprintf(stderr, "unable to update system chunk\n");
2818                 goto fail;
2819         }
2820
2821         ret = btrfs_commit_transaction(trans, root);
2822         BUG_ON(ret);
2823
2824         ret = close_ctree(root);
2825         if (ret) {
2826                 fprintf(stderr, "error during close_ctree %d\n", ret);
2827                 goto fail;
2828         }
2829
2830         /* zero btrfs super block mirrors */
2831         memset(buf, 0, sectorsize);
2832         for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2833                 bytenr = btrfs_sb_offset(i);
2834                 if (bytenr >= total_bytes)
2835                         break;
2836                 ret = pwrite(fd, buf, sectorsize, bytenr);
2837                 if (ret != sectorsize) {
2838                         fprintf(stderr,
2839                                 "error during zeroing superblock %d: %d\n",
2840                                 i, ret);
2841                         goto fail;
2842                 }
2843         }
2844
2845         sb_bytenr = (u64)-1;
2846         /* copy all relocated blocks back */
2847         while(1) {
2848                 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2849                                             EXTENT_LOCKED);
2850                 if (ret)
2851                         break;
2852
2853                 ret = get_state_private(&io_tree, start, &bytenr);
2854                 BUG_ON(ret);
2855
2856                 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2857                                   GFP_NOFS);
2858
2859                 while (start <= end) {
2860                         if (start == BTRFS_SUPER_INFO_OFFSET) {
2861                                 sb_bytenr = bytenr;
2862                                 goto next_sector;
2863                         }
2864                         ret = pread(fd, buf, sectorsize, bytenr);
2865                         if (ret < 0) {
2866                                 fprintf(stderr, "error during pread %d\n", ret);
2867                                 goto fail;
2868                         }
2869                         BUG_ON(ret != sectorsize);
2870                         ret = pwrite(fd, buf, sectorsize, start);
2871                         if (ret < 0) {
2872                                 fprintf(stderr, "error during pwrite %d\n", ret);
2873                                 goto fail;
2874                         }
2875                         BUG_ON(ret != sectorsize);
2876 next_sector:
2877                         start += sectorsize;
2878                         bytenr += sectorsize;
2879                 }
2880         }
2881
2882         ret = fsync(fd);
2883         if (ret) {
2884                 fprintf(stderr, "error during fsync %d\n", ret);
2885                 goto fail;
2886         }
2887         /*
2888          * finally, overwrite btrfs super block.
2889          */
2890         ret = pread(fd, buf, sectorsize, sb_bytenr);
2891         if (ret < 0) {
2892                 fprintf(stderr, "error during pread %d\n", ret);
2893                 goto fail;
2894         }
2895         BUG_ON(ret != sectorsize);
2896         ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2897         if (ret < 0) {
2898                 fprintf(stderr, "error during pwrite %d\n", ret);
2899                 goto fail;
2900         }
2901         BUG_ON(ret != sectorsize);
2902         ret = fsync(fd);
2903         if (ret) {
2904                 fprintf(stderr, "error during fsync %d\n", ret);
2905                 goto fail;
2906         }
2907
2908         close(fd);
2909         free(buf);
2910         extent_io_tree_cleanup(&io_tree);
2911         printf("rollback complete.\n");
2912         return 0;
2913
2914 fail:
2915         if (fd != -1)
2916                 close(fd);
2917         free(buf);
2918         fprintf(stderr, "rollback aborted.\n");
2919         return -1;
2920 }
2921
2922 static void print_usage(void)
2923 {
2924         printf("usage: btrfs-convert [options] device\n");
2925         printf("options:\n");
2926         printf("\t-d|--no-datasum        disable data checksum, sets NODATASUM\n");
2927         printf("\t-i|--no-xattr          ignore xattrs and ACLs\n");
2928         printf("\t-n|--no-inline         disable inlining of small files to metadata\n");
2929         printf("\t-N|--nodesize SIZE     set filesystem metadata nodesize\n");
2930         printf("\t-r|--rollback          roll back to the original filesystem\n");
2931         printf("\t-l|--label LABEL       set filesystem label\n");
2932         printf("\t-L|--copy-label        use label from converted filesystem\n");
2933         printf("\t-p|--progress          show converting progress (default)\n");
2934         printf("\t-O|--features LIST     comma separated list of filesystem features\n");
2935         printf("\t--no-progress          show only overview, not the detailed progress\n");
2936 }
2937
2938 int main(int argc, char *argv[])
2939 {
2940         int ret;
2941         int packing = 1;
2942         int noxattr = 0;
2943         int datacsum = 1;
2944         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
2945                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
2946         int rollback = 0;
2947         int copylabel = 0;
2948         int usage_error = 0;
2949         int progress = 1;
2950         char *file;
2951         char fslabel[BTRFS_LABEL_SIZE];
2952         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
2953
2954         while(1) {
2955                 enum { GETOPT_VAL_NO_PROGRESS = 256 };
2956                 static const struct option long_options[] = {
2957                         { "no-progress", no_argument, NULL,
2958                                 GETOPT_VAL_NO_PROGRESS },
2959                         { "no-datasum", no_argument, NULL, 'd' },
2960                         { "no-inline", no_argument, NULL, 'n' },
2961                         { "no-xattr", no_argument, NULL, 'i' },
2962                         { "rollback", no_argument, NULL, 'r' },
2963                         { "features", required_argument, NULL, 'O' },
2964                         { "progress", no_argument, NULL, 'p' },
2965                         { "label", required_argument, NULL, 'l' },
2966                         { "copy-label", no_argument, NULL, 'L' },
2967                         { "nodesize", required_argument, NULL, 'N' },
2968                         { "help", no_argument, NULL, GETOPT_VAL_HELP},
2969                         { NULL, 0, NULL, 0 }
2970                 };
2971                 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
2972
2973                 if (c < 0)
2974                         break;
2975                 switch(c) {
2976                         case 'd':
2977                                 datacsum = 0;
2978                                 break;
2979                         case 'i':
2980                                 noxattr = 1;
2981                                 break;
2982                         case 'n':
2983                                 packing = 0;
2984                                 break;
2985                         case 'N':
2986                                 nodesize = parse_size(optarg);
2987                                 break;
2988                         case 'r':
2989                                 rollback = 1;
2990                                 break;
2991                         case 'l':
2992                                 copylabel = -1;
2993                                 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
2994                                         fprintf(stderr,
2995                                 "WARNING: label too long, trimmed to %d bytes\n",
2996                                                 BTRFS_LABEL_SIZE - 1);
2997                                 }
2998                                 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
2999                                 break;
3000                         case 'L':
3001                                 copylabel = 1;
3002                                 break;
3003                         case 'p':
3004                                 progress = 1;
3005                                 break;
3006                         case 'O': {
3007                                 char *orig = strdup(optarg);
3008                                 char *tmp = orig;
3009
3010                                 tmp = btrfs_parse_fs_features(tmp, &features);
3011                                 if (tmp) {
3012                                         fprintf(stderr,
3013                                                 "Unrecognized filesystem feature '%s'\n",
3014                                                         tmp);
3015                                         free(orig);
3016                                         exit(1);
3017                                 }
3018                                 free(orig);
3019                                 if (features & BTRFS_FEATURE_LIST_ALL) {
3020                                         btrfs_list_all_fs_features(
3021                                                 ~BTRFS_CONVERT_ALLOWED_FEATURES);
3022                                         exit(0);
3023                                 }
3024                                 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
3025                                         char buf[64];
3026
3027                                         btrfs_parse_features_to_string(buf,
3028                                                 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
3029                                         fprintf(stderr,
3030                                                 "ERROR: features not allowed for convert: %s\n",
3031                                                 buf);
3032                                         exit(1);
3033                                 }
3034
3035                                 break;
3036                                 }
3037                         case GETOPT_VAL_NO_PROGRESS:
3038                                 progress = 0;
3039                                 break;
3040                         case GETOPT_VAL_HELP:
3041                         default:
3042                                 print_usage();
3043                                 return c != GETOPT_VAL_HELP;
3044                 }
3045         }
3046         set_argv0(argv);
3047         if (check_argc_exact(argc - optind, 1)) {
3048                 print_usage();
3049                 return 1;
3050         }
3051
3052         if (rollback && (!datacsum || noxattr || !packing)) {
3053                 fprintf(stderr,
3054                         "Usage error: -d, -i, -n options do not apply to rollback\n");
3055                 usage_error++;
3056         }
3057
3058         if (usage_error) {
3059                 print_usage();
3060                 return 1;
3061         }
3062
3063         file = argv[optind];
3064         ret = check_mounted(file);
3065         if (ret < 0) {
3066                 fprintf(stderr, "Could not check mount status: %s\n",
3067                         strerror(-ret));
3068                 return 1;
3069         } else if (ret) {
3070                 fprintf(stderr, "%s is mounted\n", file);
3071                 return 1;
3072         }
3073
3074         if (rollback) {
3075                 ret = do_rollback(file);
3076         } else {
3077                 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3078                                 copylabel, fslabel, progress, features);
3079         }
3080         if (ret)
3081                 return 1;
3082         return 0;
3083 }