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                 real_disk_bytenr = cur_off - key.offset + extent_disk_bytenr;
 576                 cur_len = min(key.offset + extent_num_bytes,
 577                               old_disk_bytenr + num_bytes) - cur_off;
 578                 ret = btrfs_record_file_extent(data->trans, data->root,
 579                                         data->objectid, data->inode, file_pos,
 580                                         real_disk_bytenr, cur_len);
 581                 if (ret < 0)
 582                         break;
 583                 cur_off += cur_len;
 584                 file_pos += cur_len;
 585
 586                 /*
 587                  * No need to care about csum
 588                  * As every byte of old fs image is calculated for csum, no
 589                  * need to waste CPU cycles now.
 590                  */
 591         }
 592         btrfs_free_path(path);
 593         return ret;
 594 }
 595
 596 static int block_iterate_proc(u64 disk_block, u64 file_block,
 597                               struct blk_iterate_data *idata)
 598 {
 599         int ret = 0;
 600         int sb_region;
 601         int do_barrier;
 602         struct btrfs_root *root = idata->root;
 603         struct btrfs_block_group_cache *cache;
 604         u64 bytenr = disk_block * root->sectorsize;
 605
 606         sb_region = intersect_with_sb(bytenr, root->sectorsize);
 607         do_barrier = sb_region || disk_block >= idata->boundary;
 608         if ((idata->num_blocks > 0 && do_barrier) ||
 609             (file_block > idata->first_block + idata->num_blocks) ||
 610             (disk_block != idata->disk_block + idata->num_blocks)) {
 611                 if (idata->num_blocks > 0) {
 612                         ret = record_file_blocks(idata, idata->first_block,
 613                                                  idata->disk_block,
 614                                                  idata->num_blocks);
 615                         if (ret)
 616                                 goto fail;
 617                         idata->first_block += idata->num_blocks;
 618                         idata->num_blocks = 0;
 619                 }
 620                 if (file_block > idata->first_block) {
 621                         ret = record_file_blocks(idata, idata->first_block,
 622                                         0, file_block - idata->first_block);
 623                         if (ret)
 624                                 goto fail;
 625                 }
 626
 627                 if (sb_region) {
 628                         bytenr += BTRFS_STRIPE_LEN - 1;
 629                         bytenr &= ~((u64)BTRFS_STRIPE_LEN - 1);
 630                 } else {
 631                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
 632                         BUG_ON(!cache);
 633                         bytenr = cache->key.objectid + cache->key.offset;
 634                 }
 635
 636                 idata->first_block = file_block;
 637                 idata->disk_block = disk_block;
 638                 idata->boundary = bytenr / root->sectorsize;
 639         }
 640         idata->num_blocks++;
 641 fail:
 642         return ret;
 643 }
 644
 645 static int __block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
 646                                 e2_blkcnt_t blockcnt, blk_t ref_block,
 647                                 int ref_offset, void *priv_data)
 648 {
 649         int ret;
 650         struct blk_iterate_data *idata;
 651         idata = (struct blk_iterate_data *)priv_data;
 652         ret = block_iterate_proc(*blocknr, blockcnt, idata);
 653         if (ret) {
 654                 idata->errcode = ret;
 655                 return BLOCK_ABORT;
 656         }
 657         return 0;
 658 }
 659
 660 /*
 661  * traverse file's data blocks, record these data blocks as file extents.
 662  */
 663 static int create_file_extents(struct btrfs_trans_handle *trans,
 664                                struct btrfs_root *root, u64 objectid,
 665                                struct btrfs_inode_item *btrfs_inode,
 666                                ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
 667                                int datacsum, int packing)
 668 {
 669         int ret;
 670         char *buffer = NULL;
 671         errcode_t err;
 672         u32 last_block;
 673         u32 sectorsize = root->sectorsize;
 674         u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
 675         struct blk_iterate_data data;
 676
 677         init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
 678                               datacsum);
 679
 680         err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
 681                                     NULL, __block_iterate_proc, &data);
 682         if (err)
 683                 goto error;
 684         ret = data.errcode;
 685         if (ret)
 686                 goto fail;
 687         if (packing && data.first_block == 0 && data.num_blocks > 0 &&
 688             inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 689                 u64 num_bytes = data.num_blocks * sectorsize;
 690                 u64 disk_bytenr = data.disk_block * sectorsize;
 691                 u64 nbytes;
 692
 693                 buffer = malloc(num_bytes);
 694                 if (!buffer)
 695                         return -ENOMEM;
 696                 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
 697                 if (ret)
 698                         goto fail;
 699                 if (num_bytes > inode_size)
 700                         num_bytes = inode_size;
 701                 ret = btrfs_insert_inline_extent(trans, root, objectid,
 702                                                  0, buffer, num_bytes);
 703                 if (ret)
 704                         goto fail;
 705                 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
 706                 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
 707         } else if (data.num_blocks > 0) {
 708                 ret = record_file_blocks(&data, data.first_block,
 709                                          data.disk_block, data.num_blocks);
 710                 if (ret)
 711                         goto fail;
 712         }
 713         data.first_block += data.num_blocks;
 714         last_block = (inode_size + sectorsize - 1) / sectorsize;
 715         if (last_block > data.first_block) {
 716                 ret = record_file_blocks(&data, data.first_block, 0,
 717                                          last_block - data.first_block);
 718         }
 719 fail:
 720         free(buffer);
 721         return ret;
 722 error:
 723         fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
 724         return -1;
 725 }
 726
 727 static int create_symbol_link(struct btrfs_trans_handle *trans,
 728                               struct btrfs_root *root, u64 objectid,
 729                               struct btrfs_inode_item *btrfs_inode,
 730                               ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
 731                               struct ext2_inode *ext2_inode)
 732 {
 733         int ret;
 734         char *pathname;
 735         u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
 736         if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
 737                 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
 738                 ret = create_file_extents(trans, root, objectid, btrfs_inode,
 739                                           ext2_fs, ext2_ino, 1, 1);
 740                 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
 741                 return ret;
 742         }
 743
 744         pathname = (char *)&(ext2_inode->i_block[0]);
 745         BUG_ON(pathname[inode_size] != 0);
 746         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 747                                          pathname, inode_size + 1);
 748         btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
 749         return ret;
 750 }
 751
 752 /*
 753  * Following xattr/acl related codes are based on codes in
 754  * fs/ext3/xattr.c and fs/ext3/acl.c
 755  */
 756 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
 757 #define EXT2_XATTR_BFIRST(ptr) \
 758         ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
 759 #define EXT2_XATTR_IHDR(inode) \
 760         ((struct ext2_ext_attr_header *) ((void *)(inode) + \
 761                 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
 762 #define EXT2_XATTR_IFIRST(inode) \
 763         ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
 764                 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
 765
 766 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
 767                                   const void *end)
 768 {
 769         struct ext2_ext_attr_entry *next;
 770
 771         while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
 772                 next = EXT2_EXT_ATTR_NEXT(entry);
 773                 if ((void *)next >= end)
 774                         return -EIO;
 775                 entry = next;
 776         }
 777         return 0;
 778 }
 779
 780 static int ext2_xattr_check_block(const char *buf, size_t size)
 781 {
 782         int error;
 783         struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
 784
 785         if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
 786             header->h_blocks != 1)
 787                 return -EIO;
 788         error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
 789         return error;
 790 }
 791
 792 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
 793                                   size_t size)
 794 {
 795         size_t value_size = entry->e_value_size;
 796
 797         if (entry->e_value_block != 0 || value_size > size ||
 798             entry->e_value_offs + value_size > size)
 799                 return -EIO;
 800         return 0;
 801 }
 802
 803 #define EXT2_ACL_VERSION        0x0001
 804
 805 /* 23.2.5 acl_tag_t values */
 806
 807 #define ACL_UNDEFINED_TAG       (0x00)
 808 #define ACL_USER_OBJ            (0x01)
 809 #define ACL_USER                (0x02)
 810 #define ACL_GROUP_OBJ           (0x04)
 811 #define ACL_GROUP               (0x08)
 812 #define ACL_MASK                (0x10)
 813 #define ACL_OTHER               (0x20)
 814
 815 /* 23.2.7 ACL qualifier constants */
 816
 817 #define ACL_UNDEFINED_ID        ((id_t)-1)
 818
 819 typedef struct {
 820         __le16          e_tag;
 821         __le16          e_perm;
 822         __le32          e_id;
 823 } ext2_acl_entry;
 824
 825 typedef struct {
 826         __le16          e_tag;
 827         __le16          e_perm;
 828 } ext2_acl_entry_short;
 829
 830 typedef struct {
 831         __le32          a_version;
 832 } ext2_acl_header;
 833
 834 static inline int ext2_acl_count(size_t size)
 835 {
 836         ssize_t s;
 837         size -= sizeof(ext2_acl_header);
 838         s = size - 4 * sizeof(ext2_acl_entry_short);
 839         if (s < 0) {
 840                 if (size % sizeof(ext2_acl_entry_short))
 841                         return -1;
 842                 return size / sizeof(ext2_acl_entry_short);
 843         } else {
 844                 if (s % sizeof(ext2_acl_entry))
 845                         return -1;
 846                 return s / sizeof(ext2_acl_entry) + 4;
 847         }
 848 }
 849
 850 #define ACL_EA_VERSION          0x0002
 851
 852 typedef struct {
 853         __le16          e_tag;
 854         __le16          e_perm;
 855         __le32          e_id;
 856 } acl_ea_entry;
 857
 858 typedef struct {
 859         __le32          a_version;
 860         acl_ea_entry    a_entries[0];
 861 } acl_ea_header;
 862
 863 static inline size_t acl_ea_size(int count)
 864 {
 865         return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
 866 }
 867
 868 static int ext2_acl_to_xattr(void *dst, const void *src,
 869                              size_t dst_size, size_t src_size)
 870 {
 871         int i, count;
 872         const void *end = src + src_size;
 873         acl_ea_header *ext_acl = (acl_ea_header *)dst;
 874         acl_ea_entry *dst_entry = ext_acl->a_entries;
 875         ext2_acl_entry *src_entry;
 876
 877         if (src_size < sizeof(ext2_acl_header))
 878                 goto fail;
 879         if (((ext2_acl_header *)src)->a_version !=
 880             cpu_to_le32(EXT2_ACL_VERSION))
 881                 goto fail;
 882         src += sizeof(ext2_acl_header);
 883         count = ext2_acl_count(src_size);
 884         if (count <= 0)
 885                 goto fail;
 886
 887         BUG_ON(dst_size < acl_ea_size(count));
 888         ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
 889         for (i = 0; i < count; i++, dst_entry++) {
 890                 src_entry = (ext2_acl_entry *)src;
 891                 if (src + sizeof(ext2_acl_entry_short) > end)
 892                         goto fail;
 893                 dst_entry->e_tag = src_entry->e_tag;
 894                 dst_entry->e_perm = src_entry->e_perm;
 895                 switch (le16_to_cpu(src_entry->e_tag)) {
 896                 case ACL_USER_OBJ:
 897                 case ACL_GROUP_OBJ:
 898                 case ACL_MASK:
 899                 case ACL_OTHER:
 900                         src += sizeof(ext2_acl_entry_short);
 901                         dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
 902                         break;
 903                 case ACL_USER:
 904                 case ACL_GROUP:
 905                         src += sizeof(ext2_acl_entry);
 906                         if (src > end)
 907                                 goto fail;
 908                         dst_entry->e_id = src_entry->e_id;
 909                         break;
 910                 default:
 911                         goto fail;
 912                 }
 913         }
 914         if (src != end)
 915                 goto fail;
 916         return 0;
 917 fail:
 918         return -EINVAL;
 919 }
 920
 921 static char *xattr_prefix_table[] = {
 922         [1] =   "user.",
 923         [2] =   "system.posix_acl_access",
 924         [3] =   "system.posix_acl_default",
 925         [4] =   "trusted.",
 926         [6] =   "security.",
 927 };
 928
 929 static int copy_single_xattr(struct btrfs_trans_handle *trans,
 930                              struct btrfs_root *root, u64 objectid,
 931                              struct ext2_ext_attr_entry *entry,
 932                              const void *data, u32 datalen)
 933 {
 934         int ret = 0;
 935         int name_len;
 936         int name_index;
 937         void *databuf = NULL;
 938         char namebuf[XATTR_NAME_MAX + 1];
 939
 940         name_index = entry->e_name_index;
 941         if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
 942             xattr_prefix_table[name_index] == NULL)
 943                 return -EOPNOTSUPP;
 944         name_len = strlen(xattr_prefix_table[name_index]) +
 945                    entry->e_name_len;
 946         if (name_len >= sizeof(namebuf))
 947                 return -ERANGE;
 948
 949         if (name_index == 2 || name_index == 3) {
 950                 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
 951                 databuf = malloc(bufsize);
 952                 if (!databuf)
 953                        return -ENOMEM;
 954                 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
 955                 if (ret)
 956                         goto out;
 957                 data = databuf;
 958                 datalen = bufsize;
 959         }
 960         strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
 961         strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
 962         if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
 963             sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
 964                 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
 965                         objectid - INO_OFFSET, name_len, namebuf);
 966                 goto out;
 967         }
 968         ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
 969                                       data, datalen, objectid);
 970 out:
 971         free(databuf);
 972         return ret;
 973 }
 974
 975 static int copy_extended_attrs(struct btrfs_trans_handle *trans,
 976                                struct btrfs_root *root, u64 objectid,
 977                                struct btrfs_inode_item *btrfs_inode,
 978                                ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
 979 {
 980         int ret = 0;
 981         int inline_ea = 0;
 982         errcode_t err;
 983         u32 datalen;
 984         u32 block_size = ext2_fs->blocksize;
 985         u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
 986         struct ext2_inode_large *ext2_inode;
 987         struct ext2_ext_attr_entry *entry;
 988         void *data;
 989         char *buffer = NULL;
 990         char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
 991
 992         if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
 993                 ext2_inode = (struct ext2_inode_large *)inode_buf;
 994         } else {
 995                 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
 996                 if (!ext2_inode)
 997                        return -ENOMEM;
 998         }
 999         err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
1000                                      inode_size);
1001         if (err) {
1002                 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
1003                         error_message(err));
1004                 ret = -1;
1005                 goto out;
1006         }
1007
1008         if (ext2_ino > ext2_fs->super->s_first_ino &&
1009             inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
1010                 if (EXT2_GOOD_OLD_INODE_SIZE +
1011                     ext2_inode->i_extra_isize > inode_size) {
1012                         ret = -EIO;
1013                         goto out;
1014                 }
1015                 if (ext2_inode->i_extra_isize != 0 &&
1016                     EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
1017                     EXT2_EXT_ATTR_MAGIC) {
1018                         inline_ea = 1;
1019                 }
1020         }
1021         if (inline_ea) {
1022                 int total;
1023                 void *end = (void *)ext2_inode + inode_size;
1024                 entry = EXT2_XATTR_IFIRST(ext2_inode);
1025                 total = end - (void *)entry;
1026                 ret = ext2_xattr_check_names(entry, end);
1027                 if (ret)
1028                         goto out;
1029                 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1030                         ret = ext2_xattr_check_entry(entry, total);
1031                         if (ret)
1032                                 goto out;
1033                         data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
1034                                 entry->e_value_offs;
1035                         datalen = entry->e_value_size;
1036                         ret = copy_single_xattr(trans, root, objectid,
1037                                                 entry, data, datalen);
1038                         if (ret)
1039                                 goto out;
1040                         entry = EXT2_EXT_ATTR_NEXT(entry);
1041                 }
1042         }
1043
1044         if (ext2_inode->i_file_acl == 0)
1045                 goto out;
1046
1047         buffer = malloc(block_size);
1048         if (!buffer) {
1049                 ret = -ENOMEM;
1050                 goto out;
1051         }
1052         err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
1053         if (err) {
1054                 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
1055                         error_message(err));
1056                 ret = -1;
1057                 goto out;
1058         }
1059         ret = ext2_xattr_check_block(buffer, block_size);
1060         if (ret)
1061                 goto out;
1062
1063         entry = EXT2_XATTR_BFIRST(buffer);
1064         while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1065                 ret = ext2_xattr_check_entry(entry, block_size);
1066                 if (ret)
1067                         goto out;
1068                 data = buffer + entry->e_value_offs;
1069                 datalen = entry->e_value_size;
1070                 ret = copy_single_xattr(trans, root, objectid,
1071                                         entry, data, datalen);
1072                 if (ret)
1073                         goto out;
1074                 entry = EXT2_EXT_ATTR_NEXT(entry);
1075         }
1076 out:
1077         free(buffer);
1078         if ((void *)ext2_inode != inode_buf)
1079                 free(ext2_inode);
1080         return ret;
1081 }
1082 #define MINORBITS       20
1083 #define MKDEV(ma, mi)   (((ma) << MINORBITS) | (mi))
1084
1085 static inline dev_t old_decode_dev(u16 val)
1086 {
1087         return MKDEV((val >> 8) & 255, val & 255);
1088 }
1089
1090 static inline dev_t new_decode_dev(u32 dev)
1091 {
1092         unsigned major = (dev & 0xfff00) >> 8;
1093         unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
1094         return MKDEV(major, minor);
1095 }
1096
1097 static int copy_inode_item(struct btrfs_inode_item *dst,
1098                            struct ext2_inode *src, u32 blocksize)
1099 {
1100         btrfs_set_stack_inode_generation(dst, 1);
1101         btrfs_set_stack_inode_sequence(dst, 0);
1102         btrfs_set_stack_inode_transid(dst, 1);
1103         btrfs_set_stack_inode_size(dst, src->i_size);
1104         btrfs_set_stack_inode_nbytes(dst, 0);
1105         btrfs_set_stack_inode_block_group(dst, 0);
1106         btrfs_set_stack_inode_nlink(dst, src->i_links_count);
1107         btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
1108         btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
1109         btrfs_set_stack_inode_mode(dst, src->i_mode);
1110         btrfs_set_stack_inode_rdev(dst, 0);
1111         btrfs_set_stack_inode_flags(dst, 0);
1112         btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
1113         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
1114         btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
1115         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
1116         btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
1117         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
1118         btrfs_set_stack_timespec_sec(&dst->otime, 0);
1119         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
1120
1121         if (S_ISDIR(src->i_mode)) {
1122                 btrfs_set_stack_inode_size(dst, 0);
1123                 btrfs_set_stack_inode_nlink(dst, 1);
1124         }
1125         if (S_ISREG(src->i_mode)) {
1126                 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
1127                                            (u64)src->i_size);
1128         }
1129         if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
1130             !S_ISLNK(src->i_mode)) {
1131                 if (src->i_block[0]) {
1132                         btrfs_set_stack_inode_rdev(dst,
1133                                 old_decode_dev(src->i_block[0]));
1134                 } else {
1135                         btrfs_set_stack_inode_rdev(dst,
1136                                 new_decode_dev(src->i_block[1]));
1137                 }
1138         }
1139         memset(&dst->reserved, 0, sizeof(dst->reserved));
1140
1141         return 0;
1142 }
1143
1144 /*
1145  * copy a single inode. do all the required works, such as cloning
1146  * inode item, creating file extents and creating directory entries.
1147  */
1148 static int copy_single_inode(struct btrfs_trans_handle *trans,
1149                              struct btrfs_root *root, u64 objectid,
1150                              ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1151                              struct ext2_inode *ext2_inode,
1152                              int datacsum, int packing, int noxattr)
1153 {
1154         int ret;
1155         struct btrfs_inode_item btrfs_inode;
1156
1157         if (ext2_inode->i_links_count == 0)
1158                 return 0;
1159
1160         copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
1161         if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
1162                 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
1163                             BTRFS_INODE_NODATASUM;
1164                 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
1165         }
1166
1167         switch (ext2_inode->i_mode & S_IFMT) {
1168         case S_IFREG:
1169                 ret = create_file_extents(trans, root, objectid, &btrfs_inode,
1170                                         ext2_fs, ext2_ino, datacsum, packing);
1171                 break;
1172         case S_IFDIR:
1173                 ret = create_dir_entries(trans, root, objectid, &btrfs_inode,
1174                                          ext2_fs, ext2_ino);
1175                 break;
1176         case S_IFLNK:
1177                 ret = create_symbol_link(trans, root, objectid, &btrfs_inode,
1178                                          ext2_fs, ext2_ino, ext2_inode);
1179                 break;
1180         default:
1181                 ret = 0;
1182                 break;
1183         }
1184         if (ret)
1185                 return ret;
1186
1187         if (!noxattr) {
1188                 ret = copy_extended_attrs(trans, root, objectid, &btrfs_inode,
1189                                           ext2_fs, ext2_ino);
1190                 if (ret)
1191                         return ret;
1192         }
1193         return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
1194 }
1195
1196 /*
1197  * scan ext2's inode bitmap and copy all used inodes.
1198  */
1199 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
1200                             struct btrfs_root *root,
1201                             int datacsum, int packing, int noxattr, struct task_ctx *p)
1202 {
1203         ext2_filsys ext2_fs = cctx->fs_data;
1204         int ret;
1205         errcode_t err;
1206         ext2_inode_scan ext2_scan;
1207         struct ext2_inode ext2_inode;
1208         ext2_ino_t ext2_ino;
1209         u64 objectid;
1210         struct btrfs_trans_handle *trans;
1211
1212         trans = btrfs_start_transaction(root, 1);
1213         if (!trans)
1214                 return -ENOMEM;
1215         err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
1216         if (err) {
1217                 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
1218                 return -1;
1219         }
1220         while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
1221                                              &ext2_inode))) {
1222                 /* no more inodes */
1223                 if (ext2_ino == 0)
1224                         break;
1225                 /* skip special inode in ext2fs */
1226                 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
1227                     ext2_ino != EXT2_ROOT_INO)
1228                         continue;
1229                 objectid = ext2_ino + INO_OFFSET;
1230                 ret = copy_single_inode(trans, root,
1231                                         objectid, ext2_fs, ext2_ino,
1232                                         &ext2_inode, datacsum, packing,
1233                                         noxattr);
1234                 p->cur_copy_inodes++;
1235                 if (ret)
1236                         return ret;
1237                 if (trans->blocks_used >= 4096) {
1238                         ret = btrfs_commit_transaction(trans, root);
1239                         BUG_ON(ret);
1240                         trans = btrfs_start_transaction(root, 1);
1241                         BUG_ON(!trans);
1242                 }
1243         }
1244         if (err) {
1245                 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
1246                 return -1;
1247         }
1248         ret = btrfs_commit_transaction(trans, root);
1249         BUG_ON(ret);
1250         ext2fs_close_inode_scan(ext2_scan);
1251
1252         return ret;
1253 }
1254
1255 static int create_image_file_range(struct btrfs_trans_handle *trans,
1256                                       struct btrfs_root *root,
1257                                       struct cache_tree *used,
1258                                       struct btrfs_inode_item *inode,
1259                                       u64 ino, u64 bytenr, u64 *ret_len,
1260                                       int datacsum)
1261 {
1262         struct cache_extent *cache;
1263         struct btrfs_block_group_cache *bg_cache;
1264         u64 len = *ret_len;
1265         u64 disk_bytenr;
1266         int i;
1267         int ret;
1268
1269         BUG_ON(bytenr != round_down(bytenr, root->sectorsize));
1270         BUG_ON(len != round_down(len, root->sectorsize));
1271         len = min_t(u64, len, BTRFS_MAX_EXTENT_SIZE);
1272
1273         /*
1274          * Skip sb ranges first
1275          * [0, 1M), [sb_offset(1), +64K), [sb_offset(2), +64K].
1276          *
1277          * Or we will insert a hole into current image file, and later
1278          * migrate block will fail as there is already a file extent.
1279          */
1280         if (bytenr < 1024 * 1024) {
1281                 *ret_len = 1024 * 1024 - bytenr;
1282                 return 0;
1283         }
1284         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1285                 u64 cur = btrfs_sb_offset(i);
1286
1287                 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
1288                         *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
1289                         return 0;
1290                 }
1291         }
1292         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1293                 u64 cur = btrfs_sb_offset(i);
1294
1295                 /*
1296                  *      |--reserved--|
1297                  * |----range-------|
1298                  * May still need to go through file extent inserts
1299                  */
1300                 if (bytenr < cur && bytenr + len >= cur) {
1301                         len = min_t(u64, len, cur - bytenr);
1302                         break;
1303                 }
1304                 /*
1305                  * |--reserved--|
1306                  *      |---range---|
1307                  * Drop out, no need to insert anything
1308                  */
1309                 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
1310                         *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
1311                         return 0;
1312                 }
1313         }
1314
1315         cache = search_cache_extent(used, bytenr);
1316         if (cache) {
1317                 if (cache->start <= bytenr) {
1318                         /*
1319                          * |///////Used///////|
1320                          *      |<--insert--->|
1321                          *      bytenr
1322                          */
1323                         len = min_t(u64, len, cache->start + cache->size -
1324                                     bytenr);
1325                         disk_bytenr = bytenr;
1326                 } else {
1327                         /*
1328                          *              |//Used//|
1329                          *  |<-insert-->|
1330                          *  bytenr
1331                          */
1332                         len = min(len, cache->start - bytenr);
1333                         disk_bytenr = 0;
1334                         datacsum = 0;
1335                 }
1336         } else {
1337                 /*
1338                  * |//Used//|           |EOF
1339                  *          |<-insert-->|
1340                  *          bytenr
1341                  */
1342                 disk_bytenr = 0;
1343                 datacsum = 0;
1344         }
1345
1346         if (disk_bytenr) {
1347                 /* Check if the range is in a data block group */
1348                 bg_cache = btrfs_lookup_block_group(root->fs_info, bytenr);
1349                 if (!bg_cache)
1350                         return -ENOENT;
1351                 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
1352                         return -EINVAL;
1353
1354                 /* The extent should never cross block group boundary */
1355                 len = min_t(u64, len, bg_cache->key.objectid +
1356                             bg_cache->key.offset - bytenr);
1357         }
1358
1359         BUG_ON(len != round_down(len, root->sectorsize));
1360         ret = btrfs_record_file_extent(trans, root, ino, inode, bytenr,
1361                                        disk_bytenr, len);
1362         if (ret < 0)
1363                 return ret;
1364
1365         if (datacsum)
1366                 ret = csum_disk_extent(trans, root, bytenr, len);
1367         *ret_len = len;
1368         return ret;
1369 }
1370
1371
1372 /*
1373  * Relocate old fs data in one reserved ranges
1374  *
1375  * Since all old fs data in reserved range is not covered by any chunk nor
1376  * data extent, we don't need to handle any reference but add new
1377  * extent/reference, which makes codes more clear
1378  */
1379 static int migrate_one_reserved_range(struct btrfs_trans_handle *trans,
1380                                       struct btrfs_root *root,
1381                                       struct cache_tree *used,
1382                                       struct btrfs_inode_item *inode, int fd,
1383                                       u64 ino, u64 start, u64 len, int datacsum)
1384 {
1385         u64 cur_off = start;
1386         u64 cur_len = len;
1387         struct cache_extent *cache;
1388         struct btrfs_key key;
1389         struct extent_buffer *eb;
1390         int ret = 0;
1391
1392         while (cur_off < start + len) {
1393                 cache = lookup_cache_extent(used, cur_off, cur_len);
1394                 if (!cache)
1395                         break;
1396                 cur_off = max(cache->start, cur_off);
1397                 cur_len = min(cache->start + cache->size, start + len) -
1398                           cur_off;
1399                 BUG_ON(cur_len < root->sectorsize);
1400
1401                 /* reserve extent for the data */
1402                 ret = btrfs_reserve_extent(trans, root, cur_len, 0, 0, (u64)-1,
1403                                            &key, 1);
1404                 if (ret < 0)
1405                         break;
1406
1407                 eb = malloc(sizeof(*eb) + cur_len);
1408                 if (!eb) {
1409                         ret = -ENOMEM;
1410                         break;
1411                 }
1412
1413                 ret = pread(fd, eb->data, cur_len, cur_off);
1414                 if (ret < cur_len) {
1415                         ret = (ret < 0 ? ret : -EIO);
1416                         free(eb);
1417                         break;
1418                 }
1419                 eb->start = key.objectid;
1420                 eb->len = key.offset;
1421
1422                 /* Write the data */
1423                 ret = write_and_map_eb(trans, root, eb);
1424                 free(eb);
1425                 if (ret < 0)
1426                         break;
1427
1428                 /* Now handle extent item and file extent things */
1429                 ret = btrfs_record_file_extent(trans, root, ino, inode, cur_off,
1430                                                key.objectid, key.offset);
1431                 if (ret < 0)
1432                         break;
1433                 /* Finally, insert csum items */
1434                 if (datacsum)
1435                         ret = csum_disk_extent(trans, root, key.objectid,
1436                                                key.offset);
1437
1438                 cur_off += key.offset;
1439                 cur_len = start + len - cur_off;
1440         }
1441         return ret;
1442 }
1443
1444 /*
1445  * Relocate the used ext2 data in reserved ranges
1446  * [0,1M)
1447  * [btrfs_sb_offset(1), +BTRFS_STRIPE_LEN)
1448  * [btrfs_sb_offset(2), +BTRFS_STRIPE_LEN)
1449  */
1450 static int migrate_reserved_ranges(struct btrfs_trans_handle *trans,
1451                                    struct btrfs_root *root,
1452                                    struct cache_tree *used,
1453                                    struct btrfs_inode_item *inode, int fd,
1454                                    u64 ino, u64 total_bytes, int datacsum)
1455 {
1456         u64 cur_off;
1457         u64 cur_len;
1458         int ret = 0;
1459
1460         /* 0 ~ 1M */
1461         cur_off = 0;
1462         cur_len = 1024 * 1024;
1463         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1464                                          cur_off, cur_len, datacsum);
1465         if (ret < 0)
1466                 return ret;
1467
1468         /* second sb(fisrt sb is included in 0~1M) */
1469         cur_off = btrfs_sb_offset(1);
1470         cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
1471         if (cur_off > total_bytes)
1472                 return ret;
1473         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1474                                          cur_off, cur_len, datacsum);
1475         if (ret < 0)
1476                 return ret;
1477
1478         /* Last sb */
1479         cur_off = btrfs_sb_offset(2);
1480         cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
1481         if (cur_off > total_bytes)
1482                 return ret;
1483         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1484                                          cur_off, cur_len, datacsum);
1485         return ret;
1486 }
1487
1488 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
1489                                 int ensure_size);
1490
1491 /*
1492  * Create the fs image file of old filesystem.
1493  *
1494  * This is completely fs independent as we have cctx->used, only
1495  * need to create file extents pointing to all the positions.
1496  */
1497 static int create_image(struct btrfs_root *root,
1498                            struct btrfs_mkfs_config *cfg,
1499                            struct btrfs_convert_context *cctx, int fd,
1500                            u64 size, char *name, int datacsum)
1501 {
1502         struct btrfs_inode_item buf;
1503         struct btrfs_trans_handle *trans;
1504         struct btrfs_path *path = NULL;
1505         struct btrfs_key key;
1506         struct cache_extent *cache;
1507         struct cache_tree used_tmp;
1508         u64 cur;
1509         u64 ino;
1510         int ret;
1511
1512         trans = btrfs_start_transaction(root, 1);
1513         if (!trans)
1514                 return -ENOMEM;
1515
1516         cache_tree_init(&used_tmp);
1517
1518         ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
1519                                        &ino);
1520         if (ret < 0)
1521                 goto out;
1522         ret = btrfs_new_inode(trans, root, ino, 0600 | S_IFREG);
1523         if (ret < 0)
1524                 goto out;
1525         ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
1526                              strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
1527         if (ret < 0)
1528                 goto out;
1529
1530         path = btrfs_alloc_path();
1531         if (!path) {
1532                 ret = -ENOMEM;
1533                 goto out;
1534         }
1535         key.objectid = ino;
1536         key.type = BTRFS_INODE_ITEM_KEY;
1537         key.offset = 0;
1538
1539         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1540         if (ret) {
1541                 ret = (ret > 0 ? -ENOENT : ret);
1542                 goto out;
1543         }
1544         read_extent_buffer(path->nodes[0], &buf,
1545                         btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1546                         sizeof(buf));
1547         btrfs_release_path(path);
1548
1549         /*
1550          * Create a new used space cache, which doesn't contain the reserved
1551          * range
1552          */
1553         for (cache = first_cache_extent(&cctx->used); cache;
1554              cache = next_cache_extent(cache)) {
1555                 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1556                 if (ret < 0)
1557                         goto out;
1558         }
1559         ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1560         if (ret < 0)
1561                 goto out;
1562
1563         /*
1564          * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1565          * can't handle bytenr 0(will consider it as a hole)
1566          */
1567         cur = 1024 * 1024;
1568         while (cur < size) {
1569                 u64 len = size - cur;
1570
1571                 ret = create_image_file_range(trans, root, &used_tmp,
1572                                                 &buf, ino, cur, &len, datacsum);
1573                 if (ret < 0)
1574                         goto out;
1575                 cur += len;
1576         }
1577         /* Handle the reserved ranges */
1578         ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1579                                       cfg->num_bytes, datacsum);
1580
1581
1582         key.objectid = ino;
1583         key.type = BTRFS_INODE_ITEM_KEY;
1584         key.offset = 0;
1585         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1586         if (ret) {
1587                 ret = (ret > 0 ? -ENOENT : ret);
1588                 goto out;
1589         }
1590         btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1591         write_extent_buffer(path->nodes[0], &buf,
1592                         btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1593                         sizeof(buf));
1594 out:
1595         free_extent_cache_tree(&used_tmp);
1596         btrfs_free_path(path);
1597         btrfs_commit_transaction(trans, root);
1598         return ret;
1599 }
1600
1601 static struct btrfs_root * link_subvol(struct btrfs_root *root,
1602                 const char *base, u64 root_objectid)
1603 {
1604         struct btrfs_trans_handle *trans;
1605         struct btrfs_fs_info *fs_info = root->fs_info;
1606         struct btrfs_root *tree_root = fs_info->tree_root;
1607         struct btrfs_root *new_root = NULL;
1608         struct btrfs_path *path;
1609         struct btrfs_inode_item *inode_item;
1610         struct extent_buffer *leaf;
1611         struct btrfs_key key;
1612         u64 dirid = btrfs_root_dirid(&root->root_item);
1613         u64 index = 2;
1614         char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1615         int len;
1616         int i;
1617         int ret;
1618
1619         len = strlen(base);
1620         if (len == 0 || len > BTRFS_NAME_LEN)
1621                 return NULL;
1622
1623         path = btrfs_alloc_path();
1624         BUG_ON(!path);
1625
1626         key.objectid = dirid;
1627         key.type = BTRFS_DIR_INDEX_KEY;
1628         key.offset = (u64)-1;
1629
1630         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1631         BUG_ON(ret <= 0);
1632
1633         if (path->slots[0] > 0) {
1634                 path->slots[0]--;
1635                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1636                 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1637                         index = key.offset + 1;
1638         }
1639         btrfs_release_path(path);
1640
1641         trans = btrfs_start_transaction(root, 1);
1642         BUG_ON(!trans);
1643
1644         key.objectid = dirid;
1645         key.offset = 0;
1646         key.type =  BTRFS_INODE_ITEM_KEY;
1647
1648         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1649         BUG_ON(ret);
1650         leaf = path->nodes[0];
1651         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1652                                     struct btrfs_inode_item);
1653
1654         key.objectid = root_objectid;
1655         key.offset = (u64)-1;
1656         key.type = BTRFS_ROOT_ITEM_KEY;
1657
1658         memcpy(buf, base, len);
1659         for (i = 0; i < 1024; i++) {
1660                 ret = btrfs_insert_dir_item(trans, root, buf, len,
1661                                             dirid, &key, BTRFS_FT_DIR, index);
1662                 if (ret != -EEXIST)
1663                         break;
1664                 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1665                 if (len < 1 || len > BTRFS_NAME_LEN) {
1666                         ret = -EINVAL;
1667                         break;
1668                 }
1669         }
1670         if (ret)
1671                 goto fail;
1672
1673         btrfs_set_inode_size(leaf, inode_item, len * 2 +
1674                              btrfs_inode_size(leaf, inode_item));
1675         btrfs_mark_buffer_dirty(leaf);
1676         btrfs_release_path(path);
1677
1678         /* add the backref first */
1679         ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1680                                  BTRFS_ROOT_BACKREF_KEY,
1681                                  root->root_key.objectid,
1682                                  dirid, index, buf, len);
1683         BUG_ON(ret);
1684
1685         /* now add the forward ref */
1686         ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1687                                  BTRFS_ROOT_REF_KEY, root_objectid,
1688                                  dirid, index, buf, len);
1689
1690         ret = btrfs_commit_transaction(trans, root);
1691         BUG_ON(ret);
1692
1693         new_root = btrfs_read_fs_root(fs_info, &key);
1694         if (IS_ERR(new_root))
1695                 new_root = NULL;
1696 fail:
1697         btrfs_free_path(path);
1698         return new_root;
1699 }
1700
1701 static int create_subvol(struct btrfs_trans_handle *trans,
1702                          struct btrfs_root *root, u64 root_objectid)
1703 {
1704         struct extent_buffer *tmp;
1705         struct btrfs_root *new_root;
1706         struct btrfs_key key;
1707         struct btrfs_root_item root_item;
1708         int ret;
1709
1710         ret = btrfs_copy_root(trans, root, root->node, &tmp,
1711                               root_objectid);
1712         BUG_ON(ret);
1713
1714         memcpy(&root_item, &root->root_item, sizeof(root_item));
1715         btrfs_set_root_bytenr(&root_item, tmp->start);
1716         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
1717         btrfs_set_root_generation(&root_item, trans->transid);
1718         free_extent_buffer(tmp);
1719
1720         key.objectid = root_objectid;
1721         key.type = BTRFS_ROOT_ITEM_KEY;
1722         key.offset = trans->transid;
1723         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1724                                 &key, &root_item);
1725
1726         key.offset = (u64)-1;
1727         new_root = btrfs_read_fs_root(root->fs_info, &key);
1728         BUG_ON(!new_root || IS_ERR(new_root));
1729
1730         ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
1731         BUG_ON(ret);
1732
1733         return 0;
1734 }
1735
1736 /*
1737  * New make_btrfs() has handle system and meta chunks quite well.
1738  * So only need to add remaining data chunks.
1739  */
1740 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
1741                                           struct btrfs_fs_info *fs_info,
1742                                           struct btrfs_mkfs_config *cfg,
1743                                           struct btrfs_convert_context *cctx)
1744 {
1745         struct btrfs_root *extent_root = fs_info->extent_root;
1746         struct cache_tree *data_chunks = &cctx->data_chunks;
1747         struct cache_extent *cache;
1748         u64 max_chunk_size;
1749         int ret = 0;
1750
1751         /*
1752          * Don't create data chunk over 10% of the convert device
1753          * And for single chunk, don't create chunk larger than 1G.
1754          */
1755         max_chunk_size = cfg->num_bytes / 10;
1756         max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
1757         max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
1758
1759         for (cache = first_cache_extent(data_chunks); cache;
1760              cache = next_cache_extent(cache)) {
1761                 u64 cur = cache->start;
1762
1763                 while (cur < cache->start + cache->size) {
1764                         u64 len;
1765                         u64 cur_backup = cur;
1766
1767                         len = min(max_chunk_size,
1768                                   cache->start + cache->size - cur);
1769                         ret = btrfs_alloc_data_chunk(trans, extent_root,
1770                                         &cur_backup, len,
1771                                         BTRFS_BLOCK_GROUP_DATA, 1);
1772                         if (ret < 0)
1773                                 break;
1774                         ret = btrfs_make_block_group(trans, extent_root, 0,
1775                                         BTRFS_BLOCK_GROUP_DATA,
1776                                         BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1777                                         cur, len);
1778                         if (ret < 0)
1779                                 break;
1780                         cur += len;
1781                 }
1782         }
1783         return ret;
1784 }
1785
1786 /*
1787  * Init the temp btrfs to a operational status.
1788  *
1789  * It will fix the extent usage accounting(XXX: Do we really need?) and
1790  * insert needed data chunks, to ensure all old fs data extents are covered
1791  * by DATA chunks, preventing wrong chunks are allocated.
1792  *
1793  * And also create convert image subvolume and relocation tree.
1794  * (XXX: Not need again?)
1795  * But the convert image subvolume is *NOT* linked to fs tree yet.
1796  */
1797 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
1798                          struct btrfs_convert_context *cctx, int datacsum,
1799                          int packing, int noxattr)
1800 {
1801         struct btrfs_key location;
1802         struct btrfs_trans_handle *trans;
1803         struct btrfs_fs_info *fs_info = root->fs_info;
1804         int ret;
1805
1806         /*
1807          * Don't alloc any metadata/system chunk, as we don't want
1808          * any meta/sys chunk allcated before all data chunks are inserted.
1809          * Or we screw up the chunk layout just like the old implement.
1810          */
1811         fs_info->avoid_sys_chunk_alloc = 1;
1812         fs_info->avoid_meta_chunk_alloc = 1;
1813         trans = btrfs_start_transaction(root, 1);
1814         BUG_ON(!trans);
1815         ret = btrfs_fix_block_accounting(trans, root);
1816         if (ret)
1817                 goto err;
1818         ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
1819         if (ret)
1820                 goto err;
1821         ret = btrfs_make_root_dir(trans, fs_info->tree_root,
1822                                   BTRFS_ROOT_TREE_DIR_OBJECTID);
1823         if (ret)
1824                 goto err;
1825         memcpy(&location, &root->root_key, sizeof(location));
1826         location.offset = (u64)-1;
1827         ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
1828                                 btrfs_super_root_dir(fs_info->super_copy),
1829                                 &location, BTRFS_FT_DIR, 0);
1830         if (ret)
1831                 goto err;
1832         ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
1833                                 location.objectid,
1834                                 btrfs_super_root_dir(fs_info->super_copy), 0);
1835         if (ret)
1836                 goto err;
1837         btrfs_set_root_dirid(&fs_info->fs_root->root_item,
1838                              BTRFS_FIRST_FREE_OBJECTID);
1839
1840         /* subvol for fs image file */
1841         ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
1842         if (ret < 0)
1843                 goto err;
1844         /* subvol for data relocation tree */
1845         ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1846         if (ret < 0)
1847                 goto err;
1848
1849         ret = btrfs_commit_transaction(trans, root);
1850         fs_info->avoid_sys_chunk_alloc = 0;
1851         fs_info->avoid_meta_chunk_alloc = 0;
1852 err:
1853         return ret;
1854 }
1855
1856 /*
1857  * Migrate super block to its default position and zero 0 ~ 16k
1858  */
1859 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
1860 {
1861         int ret;
1862         struct extent_buffer *buf;
1863         struct btrfs_super_block *super;
1864         u32 len;
1865         u32 bytenr;
1866
1867         BUG_ON(sectorsize < sizeof(*super));
1868         buf = malloc(sizeof(*buf) + sectorsize);
1869         if (!buf)
1870                 return -ENOMEM;
1871
1872         buf->len = sectorsize;
1873         ret = pread(fd, buf->data, sectorsize, old_bytenr);
1874         if (ret != sectorsize)
1875                 goto fail;
1876
1877         super = (struct btrfs_super_block *)buf->data;
1878         BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
1879         btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
1880
1881         csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1882         ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
1883         if (ret != sectorsize)
1884                 goto fail;
1885
1886         ret = fsync(fd);
1887         if (ret)
1888                 goto fail;
1889
1890         memset(buf->data, 0, sectorsize);
1891         for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
1892                 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
1893                 if (len > sectorsize)
1894                         len = sectorsize;
1895                 ret = pwrite(fd, buf->data, len, bytenr);
1896                 if (ret != len) {
1897                         fprintf(stderr, "unable to zero fill device\n");
1898                         break;
1899                 }
1900                 bytenr += len;
1901         }
1902         ret = 0;
1903         fsync(fd);
1904 fail:
1905         free(buf);
1906         if (ret > 0)
1907                 ret = -1;
1908         return ret;
1909 }
1910
1911 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
1912 {
1913         struct btrfs_chunk *chunk;
1914         struct btrfs_disk_key *key;
1915         u32 sectorsize = btrfs_super_sectorsize(super);
1916
1917         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1918         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1919                                        sizeof(struct btrfs_disk_key));
1920
1921         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1922         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1923         btrfs_set_disk_key_offset(key, 0);
1924
1925         btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
1926         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1927         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1928         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1929         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1930         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1931         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1932         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1933         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1934         chunk->stripe.devid = super->dev_item.devid;
1935         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1936         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1937         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1938         return 0;
1939 }
1940
1941 static const struct btrfs_convert_operations ext2_convert_ops = {
1942         .name                   = "ext2",
1943         .open_fs                = ext2_open_fs,
1944         .read_used_space        = ext2_read_used_space,
1945         .copy_inodes            = ext2_copy_inodes,
1946         .close_fs               = ext2_close_fs,
1947 };
1948
1949 static const struct btrfs_convert_operations *convert_operations[] = {
1950         &ext2_convert_ops,
1951 };
1952
1953 static int convert_open_fs(const char *devname,
1954                            struct btrfs_convert_context *cctx)
1955 {
1956         int i;
1957
1958         memset(cctx, 0, sizeof(*cctx));
1959
1960         for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
1961                 int ret = convert_operations[i]->open_fs(cctx, devname);
1962
1963                 if (ret == 0) {
1964                         cctx->convert_ops = convert_operations[i];
1965                         return ret;
1966                 }
1967         }
1968
1969         fprintf(stderr, "No file system found to convert.\n");
1970         return -1;
1971 }
1972
1973 /*
1974  * Helper for expand and merge extent_cache for wipe_one_reserved_range() to
1975  * handle wiping a range that exists in cache.
1976  */
1977 static int _expand_extent_cache(struct cache_tree *tree,
1978                                 struct cache_extent *entry,
1979                                 u64 min_stripe_size, int backward)
1980 {
1981         struct cache_extent *ce;
1982         int diff;
1983
1984         if (entry->size >= min_stripe_size)
1985                 return 0;
1986         diff = min_stripe_size - entry->size;
1987
1988         if (backward) {
1989                 ce = prev_cache_extent(entry);
1990                 if (!ce)
1991                         goto expand_back;
1992                 if (ce->start + ce->size >= entry->start - diff) {
1993                         /* Directly merge with previous extent */
1994                         ce->size = entry->start + entry->size - ce->start;
1995                         remove_cache_extent(tree, entry);
1996                         free(entry);
1997                         return 0;
1998                 }
1999 expand_back:
2000                 /* No overlap, normal extent */
2001                 if (entry->start < diff) {
2002                         error("cannot find space for data chunk layout");
2003                         return -ENOSPC;
2004                 }
2005                 entry->start -= diff;
2006                 entry->size += diff;
2007                 return 0;
2008         }
2009         ce = next_cache_extent(entry);
2010         if (!ce)
2011                 goto expand_after;
2012         if (entry->start + entry->size + diff >= ce->start) {
2013                 /* Directly merge with next extent */
2014                 entry->size = ce->start + ce->size - entry->start;
2015                 remove_cache_extent(tree, ce);
2016                 free(ce);
2017                 return 0;
2018         }
2019 expand_after:
2020         entry->size += diff;
2021         return 0;
2022 }
2023
2024 /*
2025  * Remove one reserve range from given cache tree
2026  * if min_stripe_size is non-zero, it will ensure for split case,
2027  * all its split cache extent is no smaller than @min_strip_size / 2.
2028  */
2029 static int wipe_one_reserved_range(struct cache_tree *tree,
2030                                    u64 start, u64 len, u64 min_stripe_size,
2031                                    int ensure_size)
2032 {
2033         struct cache_extent *cache;
2034         int ret;
2035
2036         BUG_ON(ensure_size && min_stripe_size == 0);
2037         /*
2038          * The logical here is simplified to handle special cases only
2039          * So we don't need to consider merge case for ensure_size
2040          */
2041         BUG_ON(min_stripe_size && (min_stripe_size < len * 2 ||
2042                min_stripe_size / 2 < BTRFS_STRIPE_LEN));
2043
2044         /* Also, wipe range should already be aligned */
2045         BUG_ON(start != round_down(start, BTRFS_STRIPE_LEN) ||
2046                start + len != round_up(start + len, BTRFS_STRIPE_LEN));
2047
2048         min_stripe_size /= 2;
2049
2050         cache = lookup_cache_extent(tree, start, len);
2051         if (!cache)
2052                 return 0;
2053
2054         if (start <= cache->start) {
2055                 /*
2056                  *      |--------cache---------|
2057                  * |-wipe-|
2058                  */
2059                 BUG_ON(start + len <= cache->start);
2060
2061                 /*
2062                  * The wipe size is smaller than min_stripe_size / 2,
2063                  * so the result length should still meet min_stripe_size
2064                  * And no need to do alignment
2065                  */
2066                 cache->size -= (start + len - cache->start);
2067                 if (cache->size == 0) {
2068                         remove_cache_extent(tree, cache);
2069                         free(cache);
2070                         return 0;
2071                 }
2072
2073                 BUG_ON(ensure_size && cache->size < min_stripe_size);
2074
2075                 cache->start = start + len;
2076                 return 0;
2077         } else if (start > cache->start && start + len < cache->start +
2078                    cache->size) {
2079                 /*
2080                  * |-------cache-----|
2081                  *      |-wipe-|
2082                  */
2083                 u64 old_start = cache->start;
2084                 u64 old_len = cache->size;
2085                 u64 insert_start = start + len;
2086                 u64 insert_len;
2087
2088                 cache->size = start - cache->start;
2089                 /* Expand the leading half part if needed */
2090                 if (ensure_size && cache->size < min_stripe_size) {
2091                         ret = _expand_extent_cache(tree, cache,
2092                                         min_stripe_size, 1);
2093                         if (ret < 0)
2094                                 return ret;
2095                 }
2096
2097                 /* And insert the new one */
2098                 insert_len = old_start + old_len - start - len;
2099                 ret = add_merge_cache_extent(tree, insert_start, insert_len);
2100                 if (ret < 0)
2101                         return ret;
2102
2103                 /* Expand the last half part if needed */
2104                 if (ensure_size && insert_len < min_stripe_size) {
2105                         cache = lookup_cache_extent(tree, insert_start,
2106                                                     insert_len);
2107                         if (!cache || cache->start != insert_start ||
2108                             cache->size != insert_len)
2109                                 return -ENOENT;
2110                         ret = _expand_extent_cache(tree, cache,
2111                                         min_stripe_size, 0);
2112                 }
2113
2114                 return ret;
2115         }
2116         /*
2117          * |----cache-----|
2118          *              |--wipe-|
2119          * Wipe len should be small enough and no need to expand the
2120          * remaining extent
2121          */
2122         cache->size = start - cache->start;
2123         BUG_ON(ensure_size && cache->size < min_stripe_size);
2124         return 0;
2125 }
2126
2127 /*
2128  * Remove reserved ranges from given cache_tree
2129  *
2130  * It will remove the following ranges
2131  * 1) 0~1M
2132  * 2) 2nd superblock, +64K (make sure chunks are 64K aligned)
2133  * 3) 3rd superblock, +64K
2134  *
2135  * @min_stripe must be given for safety check
2136  * and if @ensure_size is given, it will ensure affected cache_extent will be
2137  * larger than min_stripe_size
2138  */
2139 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
2140                                 int ensure_size)
2141 {
2142         int ret;
2143
2144         ret = wipe_one_reserved_range(tree, 0, 1024 * 1024, min_stripe_size,
2145                                       ensure_size);
2146         if (ret < 0)
2147                 return ret;
2148         ret = wipe_one_reserved_range(tree, btrfs_sb_offset(1),
2149                         BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
2150         if (ret < 0)
2151                 return ret;
2152         ret = wipe_one_reserved_range(tree, btrfs_sb_offset(2),
2153                         BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
2154         return ret;
2155 }
2156
2157 static int calculate_available_space(struct btrfs_convert_context *cctx)
2158 {
2159         struct cache_tree *used = &cctx->used;
2160         struct cache_tree *data_chunks = &cctx->data_chunks;
2161         struct cache_tree *free = &cctx->free;
2162         struct cache_extent *cache;
2163         u64 cur_off = 0;
2164         /*
2165          * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
2166          * works without need to consider overlap
2167          */
2168         u64 min_stripe_size = 2 * 16 * 1024 * 1024;
2169         int ret;
2170
2171         /* Calculate data_chunks */
2172         for (cache = first_cache_extent(used); cache;
2173              cache = next_cache_extent(cache)) {
2174                 u64 cur_len;
2175
2176                 if (cache->start + cache->size < cur_off)
2177                         continue;
2178                 if (cache->start > cur_off + min_stripe_size)
2179                         cur_off = cache->start;
2180                 cur_len = max(cache->start + cache->size - cur_off,
2181                               min_stripe_size);
2182                 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
2183                 if (ret < 0)
2184                         goto out;
2185                 cur_off += cur_len;
2186         }
2187         /*
2188          * remove reserved ranges, so we won't ever bother relocating an old
2189          * filesystem extent to other place.
2190          */
2191         ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
2192         if (ret < 0)
2193                 goto out;
2194
2195         cur_off = 0;
2196         /*
2197          * Calculate free space
2198          * Always round up the start bytenr, to avoid metadata extent corss
2199          * stripe boundary, as later mkfs_convert() won't have all the extent
2200          * allocation check
2201          */
2202         for (cache = first_cache_extent(data_chunks); cache;
2203              cache = next_cache_extent(cache)) {
2204                 if (cache->start < cur_off)
2205                         continue;
2206                 if (cache->start > cur_off) {
2207                         u64 insert_start;
2208                         u64 len;
2209
2210                         len = cache->start - round_up(cur_off,
2211                                                       BTRFS_STRIPE_LEN);
2212                         insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2213
2214                         ret = add_merge_cache_extent(free, insert_start, len);
2215                         if (ret < 0)
2216                                 goto out;
2217                 }
2218                 cur_off = cache->start + cache->size;
2219         }
2220         /* Don't forget the last range */
2221         if (cctx->total_bytes > cur_off) {
2222                 u64 len = cctx->total_bytes - cur_off;
2223                 u64 insert_start;
2224
2225                 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2226
2227                 ret = add_merge_cache_extent(free, insert_start, len);
2228                 if (ret < 0)
2229                         goto out;
2230         }
2231
2232         /* Remove reserved bytes */
2233         ret = wipe_reserved_ranges(free, min_stripe_size, 0);
2234 out:
2235         return ret;
2236 }
2237 /*
2238  * Read used space, and since we have the used space,
2239  * calcuate data_chunks and free for later mkfs
2240  */
2241 static int convert_read_used_space(struct btrfs_convert_context *cctx)
2242 {
2243         int ret;
2244
2245         ret = cctx->convert_ops->read_used_space(cctx);
2246         if (ret)
2247                 return ret;
2248
2249         ret = calculate_available_space(cctx);
2250         return ret;
2251 }
2252
2253 static int do_convert(const char *devname, int datacsum, int packing,
2254                 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2255                 int progress, u64 features)
2256 {
2257         int ret;
2258         int fd = -1;
2259         int is_btrfs = 0;
2260         u32 blocksize;
2261         u64 total_bytes;
2262         struct btrfs_root *root;
2263         struct btrfs_root *image_root;
2264         struct btrfs_convert_context cctx;
2265         struct btrfs_key key;
2266         char *subvol_name = NULL;
2267         struct task_ctx ctx;
2268         char features_buf[64];
2269         struct btrfs_mkfs_config mkfs_cfg;
2270
2271         init_convert_context(&cctx);
2272         ret = convert_open_fs(devname, &cctx);
2273         if (ret)
2274                 goto fail;
2275         ret = convert_read_used_space(&cctx);
2276         if (ret)
2277                 goto fail;
2278
2279         blocksize = cctx.blocksize;
2280         total_bytes = (u64)blocksize * (u64)cctx.block_count;
2281         if (blocksize < 4096) {
2282                 fprintf(stderr, "block size is too small\n");
2283                 goto fail;
2284         }
2285         if (btrfs_check_nodesize(nodesize, blocksize, features))
2286                 goto fail;
2287         fd = open(devname, O_RDWR);
2288         if (fd < 0) {
2289                 fprintf(stderr, "unable to open %s\n", devname);
2290                 goto fail;
2291         }
2292         btrfs_parse_features_to_string(features_buf, features);
2293         if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2294                 strcat(features_buf, " (default)");
2295
2296         printf("create btrfs filesystem:\n");
2297         printf("\tblocksize: %u\n", blocksize);
2298         printf("\tnodesize:  %u\n", nodesize);
2299         printf("\tfeatures:  %s\n", features_buf);
2300
2301         mkfs_cfg.label = cctx.volume_name;
2302         mkfs_cfg.num_bytes = total_bytes;
2303         mkfs_cfg.nodesize = nodesize;
2304         mkfs_cfg.sectorsize = blocksize;
2305         mkfs_cfg.stripesize = blocksize;
2306         mkfs_cfg.features = features;
2307         /* New convert need these space */
2308         mkfs_cfg.fs_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2309         mkfs_cfg.chunk_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2310         *(mkfs_cfg.fs_uuid) = '\0';
2311         *(mkfs_cfg.chunk_uuid) = '\0';
2312
2313         ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2314         if (ret) {
2315                 fprintf(stderr, "unable to create initial ctree: %s\n",
2316                         strerror(-ret));
2317                 goto fail;
2318         }
2319
2320         root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2321                              OPEN_CTREE_WRITES);
2322         if (!root) {
2323                 fprintf(stderr, "unable to open ctree\n");
2324                 goto fail;
2325         }
2326         ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2327         if (ret) {
2328                 fprintf(stderr, "unable to setup the root tree\n");
2329                 goto fail;
2330         }
2331
2332         printf("creating %s image file.\n", cctx.convert_ops->name);
2333         ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2334         if (ret < 0) {
2335                 fprintf(stderr, "error allocating subvolume name: %s_saved\n",
2336                         cctx.convert_ops->name);
2337                 goto fail;
2338         }
2339         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2340         key.offset = (u64)-1;
2341         key.type = BTRFS_ROOT_ITEM_KEY;
2342         image_root = btrfs_read_fs_root(root->fs_info, &key);
2343         if (!image_root) {
2344                 fprintf(stderr, "unable to create subvol\n");
2345                 goto fail;
2346         }
2347         ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2348                               mkfs_cfg.num_bytes, "image", datacsum);
2349         if (ret) {
2350                 fprintf(stderr, "error during create_image %d\n", ret);
2351                 goto fail;
2352         }
2353
2354         printf("creating btrfs metadata.\n");
2355         ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2356         ctx.cur_copy_inodes = 0;
2357
2358         if (progress) {
2359                 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2360                                      &ctx);
2361                 task_start(ctx.info);
2362         }
2363         ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2364         if (ret) {
2365                 fprintf(stderr, "error during copy_inodes %d\n", ret);
2366                 goto fail;
2367         }
2368         if (progress) {
2369                 task_stop(ctx.info);
2370                 task_deinit(ctx.info);
2371         }
2372
2373         image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2374
2375         free(subvol_name);
2376
2377         memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2378         if (copylabel == 1) {
2379                 __strncpy_null(root->fs_info->super_copy->label,
2380                                 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2381                 fprintf(stderr, "copy label '%s'\n",
2382                                 root->fs_info->super_copy->label);
2383         } else if (copylabel == -1) {
2384                 strcpy(root->fs_info->super_copy->label, fslabel);
2385                 fprintf(stderr, "set label to '%s'\n", fslabel);
2386         }
2387
2388         ret = close_ctree(root);
2389         if (ret) {
2390                 fprintf(stderr, "error during close_ctree %d\n", ret);
2391                 goto fail;
2392         }
2393         convert_close_fs(&cctx);
2394         clean_convert_context(&cctx);
2395
2396         /*
2397          * If this step succeed, we get a mountable btrfs. Otherwise
2398          * the source fs is left unchanged.
2399          */
2400         ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2401         if (ret) {
2402                 fprintf(stderr, "unable to migrate super block\n");
2403                 goto fail;
2404         }
2405         is_btrfs = 1;
2406
2407         root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2408         if (!root) {
2409                 fprintf(stderr, "unable to open ctree\n");
2410                 goto fail;
2411         }
2412         close(fd);
2413
2414         printf("conversion complete.\n");
2415         return 0;
2416 fail:
2417         clean_convert_context(&cctx);
2418         if (fd != -1)
2419                 close(fd);
2420         if (is_btrfs)
2421                 fprintf(stderr,
2422                         "WARNING: an error occurred during chunk mapping fixup, filesystem mountable but not finalized\n");
2423         else
2424                 fprintf(stderr, "conversion aborted\n");
2425         return -1;
2426 }
2427
2428 /*
2429  * Check if a non 1:1 mapped chunk can be rolled back.
2430  * For new convert, it's OK while for old convert it's not.
2431  */
2432 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2433 {
2434         struct btrfs_block_group_cache *bg;
2435         struct btrfs_key key;
2436         struct btrfs_path path;
2437         struct btrfs_root *extent_root = fs_info->extent_root;
2438         u64 bg_start;
2439         u64 bg_end;
2440         int ret;
2441
2442         bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2443         if (!bg)
2444                 return -ENOENT;
2445         bg_start = bg->key.objectid;
2446         bg_end = bg->key.objectid + bg->key.offset;
2447
2448         key.objectid = bg_end;
2449         key.type = BTRFS_METADATA_ITEM_KEY;
2450         key.offset = 0;
2451         btrfs_init_path(&path);
2452
2453         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2454         if (ret < 0)
2455                 return ret;
2456
2457         while (1) {
2458                 struct btrfs_extent_item *ei;
2459
2460                 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2461                 if (ret > 0) {
2462                         ret = 0;
2463                         break;
2464                 }
2465                 if (ret < 0)
2466                         break;
2467
2468                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2469                 if (key.type == BTRFS_METADATA_ITEM_KEY)
2470                         continue;
2471                 /* Now it's EXTENT_ITEM_KEY only */
2472                 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2473                                     struct btrfs_extent_item);
2474                 /*
2475                  * Found data extent, means this is old convert must follow 1:1
2476                  * mapping.
2477                  */
2478                 if (btrfs_extent_flags(path.nodes[0], ei)
2479                                 & BTRFS_EXTENT_FLAG_DATA) {
2480                         ret = -EINVAL;
2481                         break;
2482                 }
2483         }
2484         btrfs_release_path(&path);
2485         return ret;
2486 }
2487
2488 static int may_rollback(struct btrfs_root *root)
2489 {
2490         struct btrfs_fs_info *info = root->fs_info;
2491         struct btrfs_multi_bio *multi = NULL;
2492         u64 bytenr;
2493         u64 length;
2494         u64 physical;
2495         u64 total_bytes;
2496         int num_stripes;
2497         int ret;
2498
2499         if (btrfs_super_num_devices(info->super_copy) != 1)
2500                 goto fail;
2501
2502         bytenr = BTRFS_SUPER_INFO_OFFSET;
2503         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2504
2505         while (1) {
2506                 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2507                                       &length, &multi, 0, NULL);
2508                 if (ret) {
2509                         if (ret == -ENOENT) {
2510                                 /* removed block group at the tail */
2511                                 if (length == (u64)-1)
2512                                         break;
2513
2514                                 /* removed block group in the middle */
2515                                 goto next;
2516                         }
2517                         goto fail;
2518                 }
2519
2520                 num_stripes = multi->num_stripes;
2521                 physical = multi->stripes[0].physical;
2522                 kfree(multi);
2523
2524                 if (num_stripes != 1) {
2525                         error("num stripes for bytenr %llu is not 1", bytenr);
2526                         goto fail;
2527                 }
2528
2529                 /*
2530                  * Extra check for new convert, as metadata chunk from new
2531                  * convert is much more free than old convert, it doesn't need
2532                  * to do 1:1 mapping.
2533                  */
2534                 if (physical != bytenr) {
2535                         /*
2536                          * Check if it's a metadata chunk and has only metadata
2537                          * extent.
2538                          */
2539                         ret = may_rollback_chunk(info, bytenr);
2540                         if (ret < 0)
2541                                 goto fail;
2542                 }
2543 next:
2544                 bytenr += length;
2545                 if (bytenr >= total_bytes)
2546                         break;
2547         }
2548         return 0;
2549 fail:
2550         return -1;
2551 }
2552
2553 static int do_rollback(const char *devname)
2554 {
2555         int fd = -1;
2556         int ret;
2557         int i;
2558         struct btrfs_root *root;
2559         struct btrfs_root *image_root;
2560         struct btrfs_root *chunk_root;
2561         struct btrfs_dir_item *dir;
2562         struct btrfs_inode_item *inode;
2563         struct btrfs_file_extent_item *fi;
2564         struct btrfs_trans_handle *trans;
2565         struct extent_buffer *leaf;
2566         struct btrfs_block_group_cache *cache1;
2567         struct btrfs_block_group_cache *cache2;
2568         struct btrfs_key key;
2569         struct btrfs_path path;
2570         struct extent_io_tree io_tree;
2571         char *buf = NULL;
2572         char *name;
2573         u64 bytenr;
2574         u64 num_bytes;
2575         u64 root_dir;
2576         u64 objectid;
2577         u64 offset;
2578         u64 start;
2579         u64 end;
2580         u64 sb_bytenr;
2581         u64 first_free;
2582         u64 total_bytes;
2583         u32 sectorsize;
2584
2585         extent_io_tree_init(&io_tree);
2586
2587         fd = open(devname, O_RDWR);
2588         if (fd < 0) {
2589                 fprintf(stderr, "unable to open %s\n", devname);
2590                 goto fail;
2591         }
2592         root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2593         if (!root) {
2594                 fprintf(stderr, "unable to open ctree\n");
2595                 goto fail;
2596         }
2597         ret = may_rollback(root);
2598         if (ret < 0) {
2599                 fprintf(stderr, "unable to do rollback\n");
2600                 goto fail;
2601         }
2602
2603         sectorsize = root->sectorsize;
2604         buf = malloc(sectorsize);
2605         if (!buf) {
2606                 fprintf(stderr, "unable to allocate memory\n");
2607                 goto fail;
2608         }
2609
2610         btrfs_init_path(&path);
2611
2612         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2613         key.type = BTRFS_ROOT_BACKREF_KEY;
2614         key.offset = BTRFS_FS_TREE_OBJECTID;
2615         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2616                                 0);
2617         btrfs_release_path(&path);
2618         if (ret > 0) {
2619                 fprintf(stderr,
2620                 "ERROR: unable to convert ext2 image subvolume, is it deleted?\n");
2621                 goto fail;
2622         } else if (ret < 0) {
2623                 fprintf(stderr,
2624                         "ERROR: unable to open ext2_saved, id=%llu: %s\n",
2625                         (unsigned long long)key.objectid, strerror(-ret));
2626                 goto fail;
2627         }
2628
2629         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2630         key.type = BTRFS_ROOT_ITEM_KEY;
2631         key.offset = (u64)-1;
2632         image_root = btrfs_read_fs_root(root->fs_info, &key);
2633         if (!image_root || IS_ERR(image_root)) {
2634                 fprintf(stderr, "unable to open subvol %llu\n",
2635                         (unsigned long long)key.objectid);
2636                 goto fail;
2637         }
2638
2639         name = "image";
2640         root_dir = btrfs_root_dirid(&root->root_item);
2641         dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2642                                    root_dir, name, strlen(name), 0);
2643         if (!dir || IS_ERR(dir)) {
2644                 fprintf(stderr, "unable to find file %s\n", name);
2645                 goto fail;
2646         }
2647         leaf = path.nodes[0];
2648         btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2649         btrfs_release_path(&path);
2650
2651         objectid = key.objectid;
2652
2653         ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2654         if (ret) {
2655                 fprintf(stderr, "unable to find inode item\n");
2656                 goto fail;
2657         }
2658         leaf = path.nodes[0];
2659         inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2660         total_bytes = btrfs_inode_size(leaf, inode);
2661         btrfs_release_path(&path);
2662
2663         key.objectid = objectid;
2664         key.offset = 0;
2665         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2666         ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2667         if (ret != 0) {
2668                 fprintf(stderr, "unable to find first file extent\n");
2669                 btrfs_release_path(&path);
2670                 goto fail;
2671         }
2672
2673         /* build mapping tree for the relocated blocks */
2674         for (offset = 0; offset < total_bytes; ) {
2675                 leaf = path.nodes[0];
2676                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2677                         ret = btrfs_next_leaf(root, &path);
2678                         if (ret != 0)
2679                                 break;
2680                         continue;
2681                 }
2682
2683                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2684                 if (key.objectid != objectid || key.offset != offset ||
2685                     btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2686                         break;
2687
2688                 fi = btrfs_item_ptr(leaf, path.slots[0],
2689                                     struct btrfs_file_extent_item);
2690                 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2691                         break;
2692                 if (btrfs_file_extent_compression(leaf, fi) ||
2693                     btrfs_file_extent_encryption(leaf, fi) ||
2694                     btrfs_file_extent_other_encoding(leaf, fi))
2695                         break;
2696
2697                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2698                 /* skip holes and direct mapped extents */
2699                 if (bytenr == 0 || bytenr == offset)
2700                         goto next_extent;
2701
2702                 bytenr += btrfs_file_extent_offset(leaf, fi);
2703                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2704
2705                 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2706                 cache2 = btrfs_lookup_block_group(root->fs_info,
2707                                                   offset + num_bytes - 1);
2708                 /*
2709                  * Here we must take consideration of old and new convert
2710                  * behavior.
2711                  * For old convert case, sign, there is no consist chunk type
2712                  * that will cover the extent. META/DATA/SYS are all possible.
2713                  * Just ensure relocate one is in SYS chunk.
2714                  * For new convert case, they are all covered by DATA chunk.
2715                  *
2716                  * So, there is not valid chunk type check for it now.
2717                  */
2718                 if (cache1 != cache2)
2719                         break;
2720
2721                 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2722                                 EXTENT_LOCKED, GFP_NOFS);
2723                 set_state_private(&io_tree, offset, bytenr);
2724 next_extent:
2725                 offset += btrfs_file_extent_num_bytes(leaf, fi);
2726                 path.slots[0]++;
2727         }
2728         btrfs_release_path(&path);
2729
2730         if (offset < total_bytes) {
2731                 fprintf(stderr, "unable to build extent mapping\n");
2732                 fprintf(stderr, "converted filesystem after balance is unable to rollback\n");
2733                 goto fail;
2734         }
2735
2736         first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2737         first_free &= ~((u64)sectorsize - 1);
2738         /* backup for extent #0 should exist */
2739         if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2740                 fprintf(stderr, "no backup for the first extent\n");
2741                 goto fail;
2742         }
2743         /* force no allocation from system block group */
2744         root->fs_info->system_allocs = -1;
2745         trans = btrfs_start_transaction(root, 1);
2746         BUG_ON(!trans);
2747         /*
2748          * recow the whole chunk tree, this will remove all chunk tree blocks
2749          * from system block group
2750          */
2751         chunk_root = root->fs_info->chunk_root;
2752         memset(&key, 0, sizeof(key));
2753         while (1) {
2754                 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2755                 if (ret < 0)
2756                         break;
2757
2758                 ret = btrfs_next_leaf(chunk_root, &path);
2759                 if (ret)
2760                         break;
2761
2762                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2763                 btrfs_release_path(&path);
2764         }
2765         btrfs_release_path(&path);
2766
2767         offset = 0;
2768         num_bytes = 0;
2769         while(1) {
2770                 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2771                 if (!cache1)
2772                         break;
2773
2774                 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2775                         num_bytes += btrfs_block_group_used(&cache1->item);
2776
2777                 offset = cache1->key.objectid + cache1->key.offset;
2778         }
2779         /* only extent #0 left in system block group? */
2780         if (num_bytes > first_free) {
2781                 fprintf(stderr, "unable to empty system block group\n");
2782                 goto fail;
2783         }
2784         /* create a system chunk that maps the whole device */
2785         ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2786         if (ret) {
2787                 fprintf(stderr, "unable to update system chunk\n");
2788                 goto fail;
2789         }
2790
2791         ret = btrfs_commit_transaction(trans, root);
2792         BUG_ON(ret);
2793
2794         ret = close_ctree(root);
2795         if (ret) {
2796                 fprintf(stderr, "error during close_ctree %d\n", ret);
2797                 goto fail;
2798         }
2799
2800         /* zero btrfs super block mirrors */
2801         memset(buf, 0, sectorsize);
2802         for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2803                 bytenr = btrfs_sb_offset(i);
2804                 if (bytenr >= total_bytes)
2805                         break;
2806                 ret = pwrite(fd, buf, sectorsize, bytenr);
2807                 if (ret != sectorsize) {
2808                         fprintf(stderr,
2809                                 "error during zeroing superblock %d: %d\n",
2810                                 i, ret);
2811                         goto fail;
2812                 }
2813         }
2814
2815         sb_bytenr = (u64)-1;
2816         /* copy all relocated blocks back */
2817         while(1) {
2818                 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2819                                             EXTENT_LOCKED);
2820                 if (ret)
2821                         break;
2822
2823                 ret = get_state_private(&io_tree, start, &bytenr);
2824                 BUG_ON(ret);
2825
2826                 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2827                                   GFP_NOFS);
2828
2829                 while (start <= end) {
2830                         if (start == BTRFS_SUPER_INFO_OFFSET) {
2831                                 sb_bytenr = bytenr;
2832                                 goto next_sector;
2833                         }
2834                         ret = pread(fd, buf, sectorsize, bytenr);
2835                         if (ret < 0) {
2836                                 fprintf(stderr, "error during pread %d\n", ret);
2837                                 goto fail;
2838                         }
2839                         BUG_ON(ret != sectorsize);
2840                         ret = pwrite(fd, buf, sectorsize, start);
2841                         if (ret < 0) {
2842                                 fprintf(stderr, "error during pwrite %d\n", ret);
2843                                 goto fail;
2844                         }
2845                         BUG_ON(ret != sectorsize);
2846 next_sector:
2847                         start += sectorsize;
2848                         bytenr += sectorsize;
2849                 }
2850         }
2851
2852         ret = fsync(fd);
2853         if (ret) {
2854                 fprintf(stderr, "error during fsync %d\n", ret);
2855                 goto fail;
2856         }
2857         /*
2858          * finally, overwrite btrfs super block.
2859          */
2860         ret = pread(fd, buf, sectorsize, sb_bytenr);
2861         if (ret < 0) {
2862                 fprintf(stderr, "error during pread %d\n", ret);
2863                 goto fail;
2864         }
2865         BUG_ON(ret != sectorsize);
2866         ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2867         if (ret < 0) {
2868                 fprintf(stderr, "error during pwrite %d\n", ret);
2869                 goto fail;
2870         }
2871         BUG_ON(ret != sectorsize);
2872         ret = fsync(fd);
2873         if (ret) {
2874                 fprintf(stderr, "error during fsync %d\n", ret);
2875                 goto fail;
2876         }
2877
2878         close(fd);
2879         free(buf);
2880         extent_io_tree_cleanup(&io_tree);
2881         printf("rollback complete.\n");
2882         return 0;
2883
2884 fail:
2885         if (fd != -1)
2886                 close(fd);
2887         free(buf);
2888         fprintf(stderr, "rollback aborted.\n");
2889         return -1;
2890 }
2891
2892 static void print_usage(void)
2893 {
2894         printf("usage: btrfs-convert [options] device\n");
2895         printf("options:\n");
2896         printf("\t-d|--no-datasum        disable data checksum, sets NODATASUM\n");
2897         printf("\t-i|--no-xattr          ignore xattrs and ACLs\n");
2898         printf("\t-n|--no-inline         disable inlining of small files to metadata\n");
2899         printf("\t-N|--nodesize SIZE     set filesystem metadata nodesize\n");
2900         printf("\t-r|--rollback          roll back to the original filesystem\n");
2901         printf("\t-l|--label LABEL       set filesystem label\n");
2902         printf("\t-L|--copy-label        use label from converted filesystem\n");
2903         printf("\t-p|--progress          show converting progress (default)\n");
2904         printf("\t-O|--features LIST     comma separated list of filesystem features\n");
2905         printf("\t--no-progress          show only overview, not the detailed progress\n");
2906 }
2907
2908 int main(int argc, char *argv[])
2909 {
2910         int ret;
2911         int packing = 1;
2912         int noxattr = 0;
2913         int datacsum = 1;
2914         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
2915                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
2916         int rollback = 0;
2917         int copylabel = 0;
2918         int usage_error = 0;
2919         int progress = 1;
2920         char *file;
2921         char fslabel[BTRFS_LABEL_SIZE];
2922         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
2923
2924         while(1) {
2925                 enum { GETOPT_VAL_NO_PROGRESS = 256 };
2926                 static const struct option long_options[] = {
2927                         { "no-progress", no_argument, NULL,
2928                                 GETOPT_VAL_NO_PROGRESS },
2929                         { "no-datasum", no_argument, NULL, 'd' },
2930                         { "no-inline", no_argument, NULL, 'n' },
2931                         { "no-xattr", no_argument, NULL, 'i' },
2932                         { "rollback", no_argument, NULL, 'r' },
2933                         { "features", required_argument, NULL, 'O' },
2934                         { "progress", no_argument, NULL, 'p' },
2935                         { "label", required_argument, NULL, 'l' },
2936                         { "copy-label", no_argument, NULL, 'L' },
2937                         { "nodesize", required_argument, NULL, 'N' },
2938                         { "help", no_argument, NULL, GETOPT_VAL_HELP},
2939                         { NULL, 0, NULL, 0 }
2940                 };
2941                 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
2942
2943                 if (c < 0)
2944                         break;
2945                 switch(c) {
2946                         case 'd':
2947                                 datacsum = 0;
2948                                 break;
2949                         case 'i':
2950                                 noxattr = 1;
2951                                 break;
2952                         case 'n':
2953                                 packing = 0;
2954                                 break;
2955                         case 'N':
2956                                 nodesize = parse_size(optarg);
2957                                 break;
2958                         case 'r':
2959                                 rollback = 1;
2960                                 break;
2961                         case 'l':
2962                                 copylabel = -1;
2963                                 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
2964                                         fprintf(stderr,
2965                                 "WARNING: label too long, trimmed to %d bytes\n",
2966                                                 BTRFS_LABEL_SIZE - 1);
2967                                 }
2968                                 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
2969                                 break;
2970                         case 'L':
2971                                 copylabel = 1;
2972                                 break;
2973                         case 'p':
2974                                 progress = 1;
2975                                 break;
2976                         case 'O': {
2977                                 char *orig = strdup(optarg);
2978                                 char *tmp = orig;
2979
2980                                 tmp = btrfs_parse_fs_features(tmp, &features);
2981                                 if (tmp) {
2982                                         fprintf(stderr,
2983                                                 "Unrecognized filesystem feature '%s'\n",
2984                                                         tmp);
2985                                         free(orig);
2986                                         exit(1);
2987                                 }
2988                                 free(orig);
2989                                 if (features & BTRFS_FEATURE_LIST_ALL) {
2990                                         btrfs_list_all_fs_features(
2991                                                 ~BTRFS_CONVERT_ALLOWED_FEATURES);
2992                                         exit(0);
2993                                 }
2994                                 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
2995                                         char buf[64];
2996
2997                                         btrfs_parse_features_to_string(buf,
2998                                                 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
2999                                         fprintf(stderr,
3000                                                 "ERROR: features not allowed for convert: %s\n",
3001                                                 buf);
3002                                         exit(1);
3003                                 }
3004
3005                                 break;
3006                                 }
3007                         case GETOPT_VAL_NO_PROGRESS:
3008                                 progress = 0;
3009                                 break;
3010                         case GETOPT_VAL_HELP:
3011                         default:
3012                                 print_usage();
3013                                 return c != GETOPT_VAL_HELP;
3014                 }
3015         }
3016         set_argv0(argv);
3017         if (check_argc_exact(argc - optind, 1)) {
3018                 print_usage();
3019                 return 1;
3020         }
3021
3022         if (rollback && (!datacsum || noxattr || !packing)) {
3023                 fprintf(stderr,
3024                         "Usage error: -d, -i, -n options do not apply to rollback\n");
3025                 usage_error++;
3026         }
3027
3028         if (usage_error) {
3029                 print_usage();
3030                 return 1;
3031         }
3032
3033         file = argv[optind];
3034         ret = check_mounted(file);
3035         if (ret < 0) {
3036                 fprintf(stderr, "Could not check mount status: %s\n",
3037                         strerror(-ret));
3038                 return 1;
3039         } else if (ret) {
3040                 fprintf(stderr, "%s is mounted\n", file);
3041                 return 1;
3042         }
3043
3044         if (rollback) {
3045                 ret = do_rollback(file);
3046         } else {
3047                 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3048                                 copylabel, fslabel, progress, features);
3049         }
3050         if (ret)
3051                 return 1;
3052         return 0;
3053 }