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
  41 #include <ext2fs/ext2_fs.h>
  42 #include <ext2fs/ext2fs.h>
  43 #include <ext2fs/ext2_ext_attr.h>
  44
  45 #define INO_OFFSET (BTRFS_FIRST_FREE_OBJECTID - EXT2_ROOT_INO)
  46 #define CONV_IMAGE_SUBVOL_OBJECTID BTRFS_FIRST_FREE_OBJECTID
  47
  48 /*
  49  * Compatibility code for e2fsprogs 1.41 which doesn't support RO compat flag
  50  * BIGALLOC.
  51  * Unlike normal RO compat flag, BIGALLOC affects how e2fsprogs check used
  52  * space, and btrfs-convert heavily relies on it.
  53  */
  54 #ifdef HAVE_OLD_E2FSPROGS
  55 #define EXT2FS_CLUSTER_RATIO(fs)        (1)
  56 #define EXT2_CLUSTERS_PER_GROUP(s)      (EXT2_BLOCKS_PER_GROUP(s))
  57 #define EXT2FS_B2C(fs, blk)             (blk)
  58 #endif
  59
  60 struct task_ctx {
  61         uint32_t max_copy_inodes;
  62         uint32_t cur_copy_inodes;
  63         struct task_info *info;
  64 };
  65
  66 static void *print_copied_inodes(void *p)
  67 {
  68         struct task_ctx *priv = p;
  69         const char work_indicator[] = { '.', 'o', 'O', 'o' };
  70         uint32_t count = 0;
  71
  72         task_period_start(priv->info, 1000 /* 1s */);
  73         while (1) {
  74                 count++;
  75                 printf("copy inodes [%c] [%10d/%10d]\r",
  76                        work_indicator[count % 4], priv->cur_copy_inodes,
  77                        priv->max_copy_inodes);
  78                 fflush(stdout);
  79                 task_period_wait(priv->info);
  80         }
  81
  82         return NULL;
  83 }
  84
  85 static int after_copied_inodes(void *p)
  86 {
  87         printf("\n");
  88         fflush(stdout);
  89
  90         return 0;
  91 }
  92
  93 struct btrfs_convert_context;
  94 struct btrfs_convert_operations {
  95         const char *name;
  96         int (*open_fs)(struct btrfs_convert_context *cctx, const char *devname);
  97         int (*read_used_space)(struct btrfs_convert_context *cctx);
  98         int (*copy_inodes)(struct btrfs_convert_context *cctx,
  99                          struct btrfs_root *root, int datacsum,
 100                          int packing, int noxattr, struct task_ctx *p);
 101         void (*close_fs)(struct btrfs_convert_context *cctx);
 102 };
 103
 104 static void init_convert_context(struct btrfs_convert_context *cctx)
 105 {
 106         cache_tree_init(&cctx->used);
 107         cache_tree_init(&cctx->data_chunks);
 108         cache_tree_init(&cctx->free);
 109 }
 110
 111 static void clean_convert_context(struct btrfs_convert_context *cctx)
 112 {
 113         free_extent_cache_tree(&cctx->used);
 114         free_extent_cache_tree(&cctx->data_chunks);
 115         free_extent_cache_tree(&cctx->free);
 116 }
 117
 118 static inline int copy_inodes(struct btrfs_convert_context *cctx,
 119                               struct btrfs_root *root, int datacsum,
 120                               int packing, int noxattr, struct task_ctx *p)
 121 {
 122         return cctx->convert_ops->copy_inodes(cctx, root, datacsum, packing,
 123                                              noxattr, p);
 124 }
 125
 126 static inline void convert_close_fs(struct btrfs_convert_context *cctx)
 127 {
 128         cctx->convert_ops->close_fs(cctx);
 129 }
 130
 131 static int intersect_with_sb(u64 bytenr, u64 num_bytes)
 132 {
 133         int i;
 134         u64 offset;
 135
 136         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 137                 offset = btrfs_sb_offset(i);
 138                 offset &= ~((u64)BTRFS_STRIPE_LEN - 1);
 139
 140                 if (bytenr < offset + BTRFS_STRIPE_LEN &&
 141                     bytenr + num_bytes > offset)
 142                         return 1;
 143         }
 144         return 0;
 145 }
 146
 147 static int convert_insert_dirent(struct btrfs_trans_handle *trans,
 148                                  struct btrfs_root *root,
 149                                  const char *name, size_t name_len,
 150                                  u64 dir, u64 objectid,
 151                                  u8 file_type, u64 index_cnt,
 152                                  struct btrfs_inode_item *inode)
 153 {
 154         int ret;
 155         u64 inode_size;
 156         struct btrfs_key location = {
 157                 .objectid = objectid,
 158                 .offset = 0,
 159                 .type = BTRFS_INODE_ITEM_KEY,
 160         };
 161
 162         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 163                                     dir, &location, file_type, index_cnt);
 164         if (ret)
 165                 return ret;
 166         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 167                                      objectid, dir, index_cnt);
 168         if (ret)
 169                 return ret;
 170         inode_size = btrfs_stack_inode_size(inode) + name_len * 2;
 171         btrfs_set_stack_inode_size(inode, inode_size);
 172
 173         return 0;
 174 }
 175
 176 static int read_disk_extent(struct btrfs_root *root, u64 bytenr,
 177                             u32 num_bytes, char *buffer)
 178 {
 179         int ret;
 180         struct btrfs_fs_devices *fs_devs = root->fs_info->fs_devices;
 181
 182         ret = pread(fs_devs->latest_bdev, buffer, num_bytes, bytenr);
 183         if (ret != num_bytes)
 184                 goto fail;
 185         ret = 0;
 186 fail:
 187         if (ret > 0)
 188                 ret = -1;
 189         return ret;
 190 }
 191
 192 static int csum_disk_extent(struct btrfs_trans_handle *trans,
 193                             struct btrfs_root *root,
 194                             u64 disk_bytenr, u64 num_bytes)
 195 {
 196         u32 blocksize = root->sectorsize;
 197         u64 offset;
 198         char *buffer;
 199         int ret = 0;
 200
 201         buffer = malloc(blocksize);
 202         if (!buffer)
 203                 return -ENOMEM;
 204         for (offset = 0; offset < num_bytes; offset += blocksize) {
 205                 ret = read_disk_extent(root, disk_bytenr + offset,
 206                                         blocksize, buffer);
 207                 if (ret)
 208                         break;
 209                 ret = btrfs_csum_file_block(trans,
 210                                             root->fs_info->csum_root,
 211                                             disk_bytenr + num_bytes,
 212                                             disk_bytenr + offset,
 213                                             buffer, blocksize);
 214                 if (ret)
 215                         break;
 216         }
 217         free(buffer);
 218         return ret;
 219 }
 220
 221 struct blk_iterate_data {
 222         struct btrfs_trans_handle *trans;
 223         struct btrfs_root *root;
 224         struct btrfs_root *convert_root;
 225         struct btrfs_inode_item *inode;
 226         u64 convert_ino;
 227         u64 objectid;
 228         u64 first_block;
 229         u64 disk_block;
 230         u64 num_blocks;
 231         u64 boundary;
 232         int checksum;
 233         int errcode;
 234 };
 235
 236 static void init_blk_iterate_data(struct blk_iterate_data *data,
 237                                   struct btrfs_trans_handle *trans,
 238                                   struct btrfs_root *root,
 239                                   struct btrfs_inode_item *inode,
 240                                   u64 objectid, int checksum)
 241 {
 242         struct btrfs_key key;
 243
 244         data->trans             = trans;
 245         data->root              = root;
 246         data->inode             = inode;
 247         data->objectid          = objectid;
 248         data->first_block       = 0;
 249         data->disk_block        = 0;
 250         data->num_blocks        = 0;
 251         data->boundary          = (u64)-1;
 252         data->checksum          = checksum;
 253         data->errcode           = 0;
 254
 255         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
 256         key.type = BTRFS_ROOT_ITEM_KEY;
 257         key.offset = (u64)-1;
 258         data->convert_root = btrfs_read_fs_root(root->fs_info, &key);
 259         /* Impossible as we just opened it before */
 260         BUG_ON(!data->convert_root || IS_ERR(data->convert_root));
 261         data->convert_ino = BTRFS_FIRST_FREE_OBJECTID + 1;
 262 }
 263
 264 /*
 265  * Record a file extent in original filesystem into btrfs one.
 266  * The special point is, old disk_block can point to a reserved range.
 267  * So here, we don't use disk_block directly but search convert_root
 268  * to get the real disk_bytenr.
 269  */
 270 static int record_file_blocks(struct blk_iterate_data *data,
 271                               u64 file_block, u64 disk_block, u64 num_blocks)
 272 {
 273         int ret = 0;
 274         struct btrfs_root *root = data->root;
 275         struct btrfs_root *convert_root = data->convert_root;
 276         struct btrfs_path *path;
 277         u64 file_pos = file_block * root->sectorsize;
 278         u64 old_disk_bytenr = disk_block * root->sectorsize;
 279         u64 num_bytes = num_blocks * root->sectorsize;
 280         u64 cur_off = old_disk_bytenr;
 281
 282         /* Hole, pass it to record_file_extent directly */
 283         if (old_disk_bytenr == 0)
 284                 return btrfs_record_file_extent(data->trans, root,
 285                                 data->objectid, data->inode, file_pos, 0,
 286                                 num_bytes);
 287
 288         path = btrfs_alloc_path();
 289         if (!path)
 290                 return -ENOMEM;
 291
 292         /*
 293          * Search real disk bytenr from convert root
 294          */
 295         while (cur_off < old_disk_bytenr + num_bytes) {
 296                 struct btrfs_key key;
 297                 struct btrfs_file_extent_item *fi;
 298                 struct extent_buffer *node;
 299                 int slot;
 300                 u64 extent_disk_bytenr;
 301                 u64 extent_num_bytes;
 302                 u64 real_disk_bytenr;
 303                 u64 cur_len;
 304
 305                 key.objectid = data->convert_ino;
 306                 key.type = BTRFS_EXTENT_DATA_KEY;
 307                 key.offset = cur_off;
 308
 309                 ret = btrfs_search_slot(NULL, convert_root, &key, path, 0, 0);
 310                 if (ret < 0)
 311                         break;
 312                 if (ret > 0) {
 313                         ret = btrfs_previous_item(convert_root, path,
 314                                                   data->convert_ino,
 315                                                   BTRFS_EXTENT_DATA_KEY);
 316                         if (ret < 0)
 317                                 break;
 318                         if (ret > 0) {
 319                                 ret = -ENOENT;
 320                                 break;
 321                         }
 322                 }
 323                 node = path->nodes[0];
 324                 slot = path->slots[0];
 325                 btrfs_item_key_to_cpu(node, &key, slot);
 326                 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY ||
 327                        key.objectid != data->convert_ino ||
 328                        key.offset > cur_off);
 329                 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
 330                 extent_disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
 331                 extent_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
 332                 BUG_ON(cur_off - key.offset >= extent_num_bytes);
 333                 btrfs_release_path(path);
 334
 335                 if (extent_disk_bytenr)
 336                         real_disk_bytenr = cur_off - key.offset +
 337                                            extent_disk_bytenr;
 338                 else
 339                         real_disk_bytenr = 0;
 340                 cur_len = min(key.offset + extent_num_bytes,
 341                               old_disk_bytenr + num_bytes) - cur_off;
 342                 ret = btrfs_record_file_extent(data->trans, data->root,
 343                                         data->objectid, data->inode, file_pos,
 344                                         real_disk_bytenr, cur_len);
 345                 if (ret < 0)
 346                         break;
 347                 cur_off += cur_len;
 348                 file_pos += cur_len;
 349
 350                 /*
 351                  * No need to care about csum
 352                  * As every byte of old fs image is calculated for csum, no
 353                  * need to waste CPU cycles now.
 354                  */
 355         }
 356         btrfs_free_path(path);
 357         return ret;
 358 }
 359
 360 static int block_iterate_proc(u64 disk_block, u64 file_block,
 361                               struct blk_iterate_data *idata)
 362 {
 363         int ret = 0;
 364         int sb_region;
 365         int do_barrier;
 366         struct btrfs_root *root = idata->root;
 367         struct btrfs_block_group_cache *cache;
 368         u64 bytenr = disk_block * root->sectorsize;
 369
 370         sb_region = intersect_with_sb(bytenr, root->sectorsize);
 371         do_barrier = sb_region || disk_block >= idata->boundary;
 372         if ((idata->num_blocks > 0 && do_barrier) ||
 373             (file_block > idata->first_block + idata->num_blocks) ||
 374             (disk_block != idata->disk_block + idata->num_blocks)) {
 375                 if (idata->num_blocks > 0) {
 376                         ret = record_file_blocks(idata, idata->first_block,
 377                                                  idata->disk_block,
 378                                                  idata->num_blocks);
 379                         if (ret)
 380                                 goto fail;
 381                         idata->first_block += idata->num_blocks;
 382                         idata->num_blocks = 0;
 383                 }
 384                 if (file_block > idata->first_block) {
 385                         ret = record_file_blocks(idata, idata->first_block,
 386                                         0, file_block - idata->first_block);
 387                         if (ret)
 388                                 goto fail;
 389                 }
 390
 391                 if (sb_region) {
 392                         bytenr += BTRFS_STRIPE_LEN - 1;
 393                         bytenr &= ~((u64)BTRFS_STRIPE_LEN - 1);
 394                 } else {
 395                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
 396                         BUG_ON(!cache);
 397                         bytenr = cache->key.objectid + cache->key.offset;
 398                 }
 399
 400                 idata->first_block = file_block;
 401                 idata->disk_block = disk_block;
 402                 idata->boundary = bytenr / root->sectorsize;
 403         }
 404         idata->num_blocks++;
 405 fail:
 406         return ret;
 407 }
 408
 409 static int create_image_file_range(struct btrfs_trans_handle *trans,
 410                                       struct btrfs_root *root,
 411                                       struct cache_tree *used,
 412                                       struct btrfs_inode_item *inode,
 413                                       u64 ino, u64 bytenr, u64 *ret_len,
 414                                       int datacsum)
 415 {
 416         struct cache_extent *cache;
 417         struct btrfs_block_group_cache *bg_cache;
 418         u64 len = *ret_len;
 419         u64 disk_bytenr;
 420         int i;
 421         int ret;
 422
 423         BUG_ON(bytenr != round_down(bytenr, root->sectorsize));
 424         BUG_ON(len != round_down(len, root->sectorsize));
 425         len = min_t(u64, len, BTRFS_MAX_EXTENT_SIZE);
 426
 427         /*
 428          * Skip sb ranges first
 429          * [0, 1M), [sb_offset(1), +64K), [sb_offset(2), +64K].
 430          *
 431          * Or we will insert a hole into current image file, and later
 432          * migrate block will fail as there is already a file extent.
 433          */
 434         if (bytenr < 1024 * 1024) {
 435                 *ret_len = 1024 * 1024 - bytenr;
 436                 return 0;
 437         }
 438         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 439                 u64 cur = btrfs_sb_offset(i);
 440
 441                 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
 442                         *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
 443                         return 0;
 444                 }
 445         }
 446         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 447                 u64 cur = btrfs_sb_offset(i);
 448
 449                 /*
 450                  *      |--reserved--|
 451                  * |----range-------|
 452                  * May still need to go through file extent inserts
 453                  */
 454                 if (bytenr < cur && bytenr + len >= cur) {
 455                         len = min_t(u64, len, cur - bytenr);
 456                         break;
 457                 }
 458                 /*
 459                  * |--reserved--|
 460                  *      |---range---|
 461                  * Drop out, no need to insert anything
 462                  */
 463                 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
 464                         *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
 465                         return 0;
 466                 }
 467         }
 468
 469         cache = search_cache_extent(used, bytenr);
 470         if (cache) {
 471                 if (cache->start <= bytenr) {
 472                         /*
 473                          * |///////Used///////|
 474                          *      |<--insert--->|
 475                          *      bytenr
 476                          */
 477                         len = min_t(u64, len, cache->start + cache->size -
 478                                     bytenr);
 479                         disk_bytenr = bytenr;
 480                 } else {
 481                         /*
 482                          *              |//Used//|
 483                          *  |<-insert-->|
 484                          *  bytenr
 485                          */
 486                         len = min(len, cache->start - bytenr);
 487                         disk_bytenr = 0;
 488                         datacsum = 0;
 489                 }
 490         } else {
 491                 /*
 492                  * |//Used//|           |EOF
 493                  *          |<-insert-->|
 494                  *          bytenr
 495                  */
 496                 disk_bytenr = 0;
 497                 datacsum = 0;
 498         }
 499
 500         if (disk_bytenr) {
 501                 /* Check if the range is in a data block group */
 502                 bg_cache = btrfs_lookup_block_group(root->fs_info, bytenr);
 503                 if (!bg_cache)
 504                         return -ENOENT;
 505                 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
 506                         return -EINVAL;
 507
 508                 /* The extent should never cross block group boundary */
 509                 len = min_t(u64, len, bg_cache->key.objectid +
 510                             bg_cache->key.offset - bytenr);
 511         }
 512
 513         BUG_ON(len != round_down(len, root->sectorsize));
 514         ret = btrfs_record_file_extent(trans, root, ino, inode, bytenr,
 515                                        disk_bytenr, len);
 516         if (ret < 0)
 517                 return ret;
 518
 519         if (datacsum)
 520                 ret = csum_disk_extent(trans, root, bytenr, len);
 521         *ret_len = len;
 522         return ret;
 523 }
 524
 525 /*
 526  * Relocate old fs data in one reserved ranges
 527  *
 528  * Since all old fs data in reserved range is not covered by any chunk nor
 529  * data extent, we don't need to handle any reference but add new
 530  * extent/reference, which makes codes more clear
 531  */
 532 static int migrate_one_reserved_range(struct btrfs_trans_handle *trans,
 533                                       struct btrfs_root *root,
 534                                       struct cache_tree *used,
 535                                       struct btrfs_inode_item *inode, int fd,
 536                                       u64 ino, u64 start, u64 len, int datacsum)
 537 {
 538         u64 cur_off = start;
 539         u64 cur_len = len;
 540         u64 hole_start = start;
 541         u64 hole_len;
 542         struct cache_extent *cache;
 543         struct btrfs_key key;
 544         struct extent_buffer *eb;
 545         int ret = 0;
 546
 547         while (cur_off < start + len) {
 548                 cache = lookup_cache_extent(used, cur_off, cur_len);
 549                 if (!cache)
 550                         break;
 551                 cur_off = max(cache->start, cur_off);
 552                 cur_len = min(cache->start + cache->size, start + len) -
 553                           cur_off;
 554                 BUG_ON(cur_len < root->sectorsize);
 555
 556                 /* reserve extent for the data */
 557                 ret = btrfs_reserve_extent(trans, root, cur_len, 0, 0, (u64)-1,
 558                                            &key, 1);
 559                 if (ret < 0)
 560                         break;
 561
 562                 eb = malloc(sizeof(*eb) + cur_len);
 563                 if (!eb) {
 564                         ret = -ENOMEM;
 565                         break;
 566                 }
 567
 568                 ret = pread(fd, eb->data, cur_len, cur_off);
 569                 if (ret < cur_len) {
 570                         ret = (ret < 0 ? ret : -EIO);
 571                         free(eb);
 572                         break;
 573                 }
 574                 eb->start = key.objectid;
 575                 eb->len = key.offset;
 576
 577                 /* Write the data */
 578                 ret = write_and_map_eb(trans, root, eb);
 579                 free(eb);
 580                 if (ret < 0)
 581                         break;
 582
 583                 /* Now handle extent item and file extent things */
 584                 ret = btrfs_record_file_extent(trans, root, ino, inode, cur_off,
 585                                                key.objectid, key.offset);
 586                 if (ret < 0)
 587                         break;
 588                 /* Finally, insert csum items */
 589                 if (datacsum)
 590                         ret = csum_disk_extent(trans, root, key.objectid,
 591                                                key.offset);
 592
 593                 /* Don't forget to insert hole */
 594                 hole_len = cur_off - hole_start;
 595                 if (hole_len) {
 596                         ret = btrfs_record_file_extent(trans, root, ino, inode,
 597                                         hole_start, 0, hole_len);
 598                         if (ret < 0)
 599                                 break;
 600                 }
 601
 602                 cur_off += key.offset;
 603                 hole_start = cur_off;
 604                 cur_len = start + len - cur_off;
 605         }
 606         /* Last hole */
 607         if (start + len - hole_start > 0)
 608                 ret = btrfs_record_file_extent(trans, root, ino, inode,
 609                                 hole_start, 0, start + len - hole_start);
 610         return ret;
 611 }
 612
 613 /*
 614  * Relocate the used ext2 data in reserved ranges
 615  * [0,1M)
 616  * [btrfs_sb_offset(1), +BTRFS_STRIPE_LEN)
 617  * [btrfs_sb_offset(2), +BTRFS_STRIPE_LEN)
 618  */
 619 static int migrate_reserved_ranges(struct btrfs_trans_handle *trans,
 620                                    struct btrfs_root *root,
 621                                    struct cache_tree *used,
 622                                    struct btrfs_inode_item *inode, int fd,
 623                                    u64 ino, u64 total_bytes, int datacsum)
 624 {
 625         u64 cur_off;
 626         u64 cur_len;
 627         int ret = 0;
 628
 629         /* 0 ~ 1M */
 630         cur_off = 0;
 631         cur_len = 1024 * 1024;
 632         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
 633                                          cur_off, cur_len, datacsum);
 634         if (ret < 0)
 635                 return ret;
 636
 637         /* second sb(fisrt sb is included in 0~1M) */
 638         cur_off = btrfs_sb_offset(1);
 639         cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
 640         if (cur_off > total_bytes)
 641                 return ret;
 642         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
 643                                          cur_off, cur_len, datacsum);
 644         if (ret < 0)
 645                 return ret;
 646
 647         /* Last sb */
 648         cur_off = btrfs_sb_offset(2);
 649         cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
 650         if (cur_off > total_bytes)
 651                 return ret;
 652         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
 653                                          cur_off, cur_len, datacsum);
 654         return ret;
 655 }
 656
 657 /*
 658  * Helper for expand and merge extent_cache for wipe_one_reserved_range() to
 659  * handle wiping a range that exists in cache.
 660  */
 661 static int _expand_extent_cache(struct cache_tree *tree,
 662                                 struct cache_extent *entry,
 663                                 u64 min_stripe_size, int backward)
 664 {
 665         struct cache_extent *ce;
 666         int diff;
 667
 668         if (entry->size >= min_stripe_size)
 669                 return 0;
 670         diff = min_stripe_size - entry->size;
 671
 672         if (backward) {
 673                 ce = prev_cache_extent(entry);
 674                 if (!ce)
 675                         goto expand_back;
 676                 if (ce->start + ce->size >= entry->start - diff) {
 677                         /* Directly merge with previous extent */
 678                         ce->size = entry->start + entry->size - ce->start;
 679                         remove_cache_extent(tree, entry);
 680                         free(entry);
 681                         return 0;
 682                 }
 683 expand_back:
 684                 /* No overlap, normal extent */
 685                 if (entry->start < diff) {
 686                         error("cannot find space for data chunk layout");
 687                         return -ENOSPC;
 688                 }
 689                 entry->start -= diff;
 690                 entry->size += diff;
 691                 return 0;
 692         }
 693         ce = next_cache_extent(entry);
 694         if (!ce)
 695                 goto expand_after;
 696         if (entry->start + entry->size + diff >= ce->start) {
 697                 /* Directly merge with next extent */
 698                 entry->size = ce->start + ce->size - entry->start;
 699                 remove_cache_extent(tree, ce);
 700                 free(ce);
 701                 return 0;
 702         }
 703 expand_after:
 704         entry->size += diff;
 705         return 0;
 706 }
 707
 708 /*
 709  * Open Ext2fs in readonly mode, read block allocation bitmap and
 710  * inode bitmap into memory.
 711  */
 712 static int ext2_open_fs(struct btrfs_convert_context *cctx, const char *name)
 713 {
 714         errcode_t ret;
 715         ext2_filsys ext2_fs;
 716         ext2_ino_t ino;
 717         u32 ro_feature;
 718
 719         ret = ext2fs_open(name, 0, 0, 0, unix_io_manager, &ext2_fs);
 720         if (ret) {
 721                 fprintf(stderr, "ext2fs_open: %s\n", error_message(ret));
 722                 return -1;
 723         }
 724         /*
 725          * We need to know exactly the used space, some RO compat flags like
 726          * BIGALLOC will affect how used space is present.
 727          * So we need manuall check any unsupported RO compat flags
 728          */
 729         ro_feature = ext2_fs->super->s_feature_ro_compat;
 730         if (ro_feature & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
 731                 error(
 732 "unsupported RO features detected: %x, abort convert to avoid possible corruption",
 733                       ro_feature & ~EXT2_LIB_FEATURE_COMPAT_SUPP);
 734                 goto fail;
 735         }
 736         ret = ext2fs_read_inode_bitmap(ext2_fs);
 737         if (ret) {
 738                 fprintf(stderr, "ext2fs_read_inode_bitmap: %s\n",
 739                         error_message(ret));
 740                 goto fail;
 741         }
 742         ret = ext2fs_read_block_bitmap(ext2_fs);
 743         if (ret) {
 744                 fprintf(stderr, "ext2fs_read_block_bitmap: %s\n",
 745                         error_message(ret));
 746                 goto fail;
 747         }
 748         /*
 749          * search each block group for a free inode. this set up
 750          * uninit block/inode bitmaps appropriately.
 751          */
 752         ino = 1;
 753         while (ino <= ext2_fs->super->s_inodes_count) {
 754                 ext2_ino_t foo;
 755                 ext2fs_new_inode(ext2_fs, ino, 0, NULL, &foo);
 756                 ino += EXT2_INODES_PER_GROUP(ext2_fs->super);
 757         }
 758
 759         if (!(ext2_fs->super->s_feature_incompat &
 760               EXT2_FEATURE_INCOMPAT_FILETYPE)) {
 761                 fprintf(stderr, "filetype feature is missing\n");
 762                 goto fail;
 763         }
 764
 765         cctx->fs_data = ext2_fs;
 766         cctx->blocksize = ext2_fs->blocksize;
 767         cctx->block_count = ext2_fs->super->s_blocks_count;
 768         cctx->total_bytes = ext2_fs->blocksize * ext2_fs->super->s_blocks_count;
 769         cctx->volume_name = strndup(ext2_fs->super->s_volume_name, 16);
 770         cctx->first_data_block = ext2_fs->super->s_first_data_block;
 771         cctx->inodes_count = ext2_fs->super->s_inodes_count;
 772         cctx->free_inodes_count = ext2_fs->super->s_free_inodes_count;
 773         return 0;
 774 fail:
 775         ext2fs_close(ext2_fs);
 776         return -1;
 777 }
 778
 779 static int __ext2_add_one_block(ext2_filsys fs, char *bitmap,
 780                                 unsigned long group_nr, struct cache_tree *used)
 781 {
 782         unsigned long offset;
 783         unsigned i;
 784         int ret = 0;
 785
 786         offset = fs->super->s_first_data_block;
 787         offset /= EXT2FS_CLUSTER_RATIO(fs);
 788         offset += group_nr * EXT2_CLUSTERS_PER_GROUP(fs->super);
 789         for (i = 0; i < EXT2_CLUSTERS_PER_GROUP(fs->super); i++) {
 790                 if (ext2fs_test_bit(i, bitmap)) {
 791                         u64 start;
 792
 793                         start = (i + offset) * EXT2FS_CLUSTER_RATIO(fs);
 794                         start *= fs->blocksize;
 795                         ret = add_merge_cache_extent(used, start,
 796                                                      fs->blocksize);
 797                         if (ret < 0)
 798                                 break;
 799                 }
 800         }
 801         return ret;
 802 }
 803
 804 /*
 805  * Read all used ext2 space into cctx->used cache tree
 806  */
 807 static int ext2_read_used_space(struct btrfs_convert_context *cctx)
 808 {
 809         ext2_filsys fs = (ext2_filsys)cctx->fs_data;
 810         blk64_t blk_itr = EXT2FS_B2C(fs, fs->super->s_first_data_block);
 811         struct cache_tree *used_tree = &cctx->used;
 812         char *block_bitmap = NULL;
 813         unsigned long i;
 814         int block_nbytes;
 815         int ret = 0;
 816
 817         block_nbytes = EXT2_CLUSTERS_PER_GROUP(fs->super) / 8;
 818         /* Shouldn't happen */
 819         BUG_ON(!fs->block_map);
 820
 821         block_bitmap = malloc(block_nbytes);
 822         if (!block_bitmap)
 823                 return -ENOMEM;
 824
 825         for (i = 0; i < fs->group_desc_count; i++) {
 826                 ret = ext2fs_get_block_bitmap_range(fs->block_map, blk_itr,
 827                                                 block_nbytes * 8, block_bitmap);
 828                 if (ret) {
 829                         error("fail to get bitmap from ext2, %s",
 830                               strerror(-ret));
 831                         break;
 832                 }
 833                 ret = __ext2_add_one_block(fs, block_bitmap, i, used_tree);
 834                 if (ret < 0) {
 835                         error("fail to build used space tree, %s",
 836                               strerror(-ret));
 837                         break;
 838                 }
 839                 blk_itr += EXT2_CLUSTERS_PER_GROUP(fs->super);
 840         }
 841
 842         free(block_bitmap);
 843         return ret;
 844 }
 845
 846 static void ext2_close_fs(struct btrfs_convert_context *cctx)
 847 {
 848         if (cctx->volume_name) {
 849                 free(cctx->volume_name);
 850                 cctx->volume_name = NULL;
 851         }
 852         ext2fs_close(cctx->fs_data);
 853 }
 854
 855 struct dir_iterate_data {
 856         struct btrfs_trans_handle *trans;
 857         struct btrfs_root *root;
 858         struct btrfs_inode_item *inode;
 859         u64 objectid;
 860         u64 index_cnt;
 861         u64 parent;
 862         int errcode;
 863 };
 864
 865 static u8 ext2_filetype_conversion_table[EXT2_FT_MAX] = {
 866         [EXT2_FT_UNKNOWN]       = BTRFS_FT_UNKNOWN,
 867         [EXT2_FT_REG_FILE]      = BTRFS_FT_REG_FILE,
 868         [EXT2_FT_DIR]           = BTRFS_FT_DIR,
 869         [EXT2_FT_CHRDEV]        = BTRFS_FT_CHRDEV,
 870         [EXT2_FT_BLKDEV]        = BTRFS_FT_BLKDEV,
 871         [EXT2_FT_FIFO]          = BTRFS_FT_FIFO,
 872         [EXT2_FT_SOCK]          = BTRFS_FT_SOCK,
 873         [EXT2_FT_SYMLINK]       = BTRFS_FT_SYMLINK,
 874 };
 875
 876 static int ext2_dir_iterate_proc(ext2_ino_t dir, int entry,
 877                             struct ext2_dir_entry *dirent,
 878                             int offset, int blocksize,
 879                             char *buf,void *priv_data)
 880 {
 881         int ret;
 882         int file_type;
 883         u64 objectid;
 884         char dotdot[] = "..";
 885         struct dir_iterate_data *idata = (struct dir_iterate_data *)priv_data;
 886         int name_len;
 887
 888         name_len = dirent->name_len & 0xFF;
 889
 890         objectid = dirent->inode + INO_OFFSET;
 891         if (!strncmp(dirent->name, dotdot, name_len)) {
 892                 if (name_len == 2) {
 893                         BUG_ON(idata->parent != 0);
 894                         idata->parent = objectid;
 895                 }
 896                 return 0;
 897         }
 898         if (dirent->inode < EXT2_GOOD_OLD_FIRST_INO)
 899                 return 0;
 900
 901         file_type = dirent->name_len >> 8;
 902         BUG_ON(file_type > EXT2_FT_SYMLINK);
 903
 904         ret = convert_insert_dirent(idata->trans, idata->root, dirent->name,
 905                                     name_len, idata->objectid, objectid,
 906                                     ext2_filetype_conversion_table[file_type],
 907                                     idata->index_cnt, idata->inode);
 908         if (ret < 0) {
 909                 idata->errcode = ret;
 910                 return BLOCK_ABORT;
 911         }
 912
 913         idata->index_cnt++;
 914         return 0;
 915 }
 916
 917 static int ext2_create_dir_entries(struct btrfs_trans_handle *trans,
 918                               struct btrfs_root *root, u64 objectid,
 919                               struct btrfs_inode_item *btrfs_inode,
 920                               ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
 921 {
 922         int ret;
 923         errcode_t err;
 924         struct dir_iterate_data data = {
 925                 .trans          = trans,
 926                 .root           = root,
 927                 .inode          = btrfs_inode,
 928                 .objectid       = objectid,
 929                 .index_cnt      = 2,
 930                 .parent         = 0,
 931                 .errcode        = 0,
 932         };
 933
 934         err = ext2fs_dir_iterate2(ext2_fs, ext2_ino, 0, NULL,
 935                                   ext2_dir_iterate_proc, &data);
 936         if (err)
 937                 goto error;
 938         ret = data.errcode;
 939         if (ret == 0 && data.parent == objectid) {
 940                 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
 941                                              objectid, objectid, 0);
 942         }
 943         return ret;
 944 error:
 945         fprintf(stderr, "ext2fs_dir_iterate2: %s\n", error_message(err));
 946         return -1;
 947 }
 948
 949 static int ext2_block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
 950                                 e2_blkcnt_t blockcnt, blk_t ref_block,
 951                                 int ref_offset, void *priv_data)
 952 {
 953         int ret;
 954         struct blk_iterate_data *idata;
 955         idata = (struct blk_iterate_data *)priv_data;
 956         ret = block_iterate_proc(*blocknr, blockcnt, idata);
 957         if (ret) {
 958                 idata->errcode = ret;
 959                 return BLOCK_ABORT;
 960         }
 961         return 0;
 962 }
 963
 964 /*
 965  * traverse file's data blocks, record these data blocks as file extents.
 966  */
 967 static int ext2_create_file_extents(struct btrfs_trans_handle *trans,
 968                                struct btrfs_root *root, u64 objectid,
 969                                struct btrfs_inode_item *btrfs_inode,
 970                                ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
 971                                int datacsum, int packing)
 972 {
 973         int ret;
 974         char *buffer = NULL;
 975         errcode_t err;
 976         u32 last_block;
 977         u32 sectorsize = root->sectorsize;
 978         u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
 979         struct blk_iterate_data data;
 980
 981         init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
 982                               datacsum);
 983
 984         err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
 985                                     NULL, ext2_block_iterate_proc, &data);
 986         if (err)
 987                 goto error;
 988         ret = data.errcode;
 989         if (ret)
 990                 goto fail;
 991         if (packing && data.first_block == 0 && data.num_blocks > 0 &&
 992             inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 993                 u64 num_bytes = data.num_blocks * sectorsize;
 994                 u64 disk_bytenr = data.disk_block * sectorsize;
 995                 u64 nbytes;
 996
 997                 buffer = malloc(num_bytes);
 998                 if (!buffer)
 999                         return -ENOMEM;
1000                 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
1001                 if (ret)
1002                         goto fail;
1003                 if (num_bytes > inode_size)
1004                         num_bytes = inode_size;
1005                 ret = btrfs_insert_inline_extent(trans, root, objectid,
1006                                                  0, buffer, num_bytes);
1007                 if (ret)
1008                         goto fail;
1009                 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
1010                 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
1011         } else if (data.num_blocks > 0) {
1012                 ret = record_file_blocks(&data, data.first_block,
1013                                          data.disk_block, data.num_blocks);
1014                 if (ret)
1015                         goto fail;
1016         }
1017         data.first_block += data.num_blocks;
1018         last_block = (inode_size + sectorsize - 1) / sectorsize;
1019         if (last_block > data.first_block) {
1020                 ret = record_file_blocks(&data, data.first_block, 0,
1021                                          last_block - data.first_block);
1022         }
1023 fail:
1024         free(buffer);
1025         return ret;
1026 error:
1027         fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
1028         return -1;
1029 }
1030
1031 static int ext2_create_symbol_link(struct btrfs_trans_handle *trans,
1032                               struct btrfs_root *root, u64 objectid,
1033                               struct btrfs_inode_item *btrfs_inode,
1034                               ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1035                               struct ext2_inode *ext2_inode)
1036 {
1037         int ret;
1038         char *pathname;
1039         u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1040         if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
1041                 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
1042                 ret = ext2_create_file_extents(trans, root, objectid,
1043                                 btrfs_inode, ext2_fs, ext2_ino, 1, 1);
1044                 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
1045                 return ret;
1046         }
1047
1048         pathname = (char *)&(ext2_inode->i_block[0]);
1049         BUG_ON(pathname[inode_size] != 0);
1050         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
1051                                          pathname, inode_size + 1);
1052         btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
1053         return ret;
1054 }
1055
1056 /*
1057  * Following xattr/acl related codes are based on codes in
1058  * fs/ext3/xattr.c and fs/ext3/acl.c
1059  */
1060 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
1061 #define EXT2_XATTR_BFIRST(ptr) \
1062         ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
1063 #define EXT2_XATTR_IHDR(inode) \
1064         ((struct ext2_ext_attr_header *) ((void *)(inode) + \
1065                 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
1066 #define EXT2_XATTR_IFIRST(inode) \
1067         ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
1068                 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
1069
1070 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
1071                                   const void *end)
1072 {
1073         struct ext2_ext_attr_entry *next;
1074
1075         while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1076                 next = EXT2_EXT_ATTR_NEXT(entry);
1077                 if ((void *)next >= end)
1078                         return -EIO;
1079                 entry = next;
1080         }
1081         return 0;
1082 }
1083
1084 static int ext2_xattr_check_block(const char *buf, size_t size)
1085 {
1086         int error;
1087         struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
1088
1089         if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
1090             header->h_blocks != 1)
1091                 return -EIO;
1092         error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
1093         return error;
1094 }
1095
1096 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
1097                                   size_t size)
1098 {
1099         size_t value_size = entry->e_value_size;
1100
1101         if (entry->e_value_block != 0 || value_size > size ||
1102             entry->e_value_offs + value_size > size)
1103                 return -EIO;
1104         return 0;
1105 }
1106
1107 #define EXT2_ACL_VERSION        0x0001
1108
1109 /* 23.2.5 acl_tag_t values */
1110
1111 #define ACL_UNDEFINED_TAG       (0x00)
1112 #define ACL_USER_OBJ            (0x01)
1113 #define ACL_USER                (0x02)
1114 #define ACL_GROUP_OBJ           (0x04)
1115 #define ACL_GROUP               (0x08)
1116 #define ACL_MASK                (0x10)
1117 #define ACL_OTHER               (0x20)
1118
1119 /* 23.2.7 ACL qualifier constants */
1120
1121 #define ACL_UNDEFINED_ID        ((id_t)-1)
1122
1123 typedef struct {
1124         __le16          e_tag;
1125         __le16          e_perm;
1126         __le32          e_id;
1127 } ext2_acl_entry;
1128
1129 typedef struct {
1130         __le16          e_tag;
1131         __le16          e_perm;
1132 } ext2_acl_entry_short;
1133
1134 typedef struct {
1135         __le32          a_version;
1136 } ext2_acl_header;
1137
1138 static inline int ext2_acl_count(size_t size)
1139 {
1140         ssize_t s;
1141         size -= sizeof(ext2_acl_header);
1142         s = size - 4 * sizeof(ext2_acl_entry_short);
1143         if (s < 0) {
1144                 if (size % sizeof(ext2_acl_entry_short))
1145                         return -1;
1146                 return size / sizeof(ext2_acl_entry_short);
1147         } else {
1148                 if (s % sizeof(ext2_acl_entry))
1149                         return -1;
1150                 return s / sizeof(ext2_acl_entry) + 4;
1151         }
1152 }
1153
1154 #define ACL_EA_VERSION          0x0002
1155
1156 typedef struct {
1157         __le16          e_tag;
1158         __le16          e_perm;
1159         __le32          e_id;
1160 } acl_ea_entry;
1161
1162 typedef struct {
1163         __le32          a_version;
1164         acl_ea_entry    a_entries[0];
1165 } acl_ea_header;
1166
1167 static inline size_t acl_ea_size(int count)
1168 {
1169         return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
1170 }
1171
1172 static int ext2_acl_to_xattr(void *dst, const void *src,
1173                              size_t dst_size, size_t src_size)
1174 {
1175         int i, count;
1176         const void *end = src + src_size;
1177         acl_ea_header *ext_acl = (acl_ea_header *)dst;
1178         acl_ea_entry *dst_entry = ext_acl->a_entries;
1179         ext2_acl_entry *src_entry;
1180
1181         if (src_size < sizeof(ext2_acl_header))
1182                 goto fail;
1183         if (((ext2_acl_header *)src)->a_version !=
1184             cpu_to_le32(EXT2_ACL_VERSION))
1185                 goto fail;
1186         src += sizeof(ext2_acl_header);
1187         count = ext2_acl_count(src_size);
1188         if (count <= 0)
1189                 goto fail;
1190
1191         BUG_ON(dst_size < acl_ea_size(count));
1192         ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
1193         for (i = 0; i < count; i++, dst_entry++) {
1194                 src_entry = (ext2_acl_entry *)src;
1195                 if (src + sizeof(ext2_acl_entry_short) > end)
1196                         goto fail;
1197                 dst_entry->e_tag = src_entry->e_tag;
1198                 dst_entry->e_perm = src_entry->e_perm;
1199                 switch (le16_to_cpu(src_entry->e_tag)) {
1200                 case ACL_USER_OBJ:
1201                 case ACL_GROUP_OBJ:
1202                 case ACL_MASK:
1203                 case ACL_OTHER:
1204                         src += sizeof(ext2_acl_entry_short);
1205                         dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
1206                         break;
1207                 case ACL_USER:
1208                 case ACL_GROUP:
1209                         src += sizeof(ext2_acl_entry);
1210                         if (src > end)
1211                                 goto fail;
1212                         dst_entry->e_id = src_entry->e_id;
1213                         break;
1214                 default:
1215                         goto fail;
1216                 }
1217         }
1218         if (src != end)
1219                 goto fail;
1220         return 0;
1221 fail:
1222         return -EINVAL;
1223 }
1224
1225 static char *xattr_prefix_table[] = {
1226         [1] =   "user.",
1227         [2] =   "system.posix_acl_access",
1228         [3] =   "system.posix_acl_default",
1229         [4] =   "trusted.",
1230         [6] =   "security.",
1231 };
1232
1233 static int ext2_copy_single_xattr(struct btrfs_trans_handle *trans,
1234                              struct btrfs_root *root, u64 objectid,
1235                              struct ext2_ext_attr_entry *entry,
1236                              const void *data, u32 datalen)
1237 {
1238         int ret = 0;
1239         int name_len;
1240         int name_index;
1241         void *databuf = NULL;
1242         char namebuf[XATTR_NAME_MAX + 1];
1243
1244         name_index = entry->e_name_index;
1245         if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
1246             xattr_prefix_table[name_index] == NULL)
1247                 return -EOPNOTSUPP;
1248         name_len = strlen(xattr_prefix_table[name_index]) +
1249                    entry->e_name_len;
1250         if (name_len >= sizeof(namebuf))
1251                 return -ERANGE;
1252
1253         if (name_index == 2 || name_index == 3) {
1254                 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
1255                 databuf = malloc(bufsize);
1256                 if (!databuf)
1257                        return -ENOMEM;
1258                 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
1259                 if (ret)
1260                         goto out;
1261                 data = databuf;
1262                 datalen = bufsize;
1263         }
1264         strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
1265         strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
1266         if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
1267             sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
1268                 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
1269                         objectid - INO_OFFSET, name_len, namebuf);
1270                 goto out;
1271         }
1272         ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
1273                                       data, datalen, objectid);
1274 out:
1275         free(databuf);
1276         return ret;
1277 }
1278
1279 static int ext2_copy_extended_attrs(struct btrfs_trans_handle *trans,
1280                                struct btrfs_root *root, u64 objectid,
1281                                struct btrfs_inode_item *btrfs_inode,
1282                                ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
1283 {
1284         int ret = 0;
1285         int inline_ea = 0;
1286         errcode_t err;
1287         u32 datalen;
1288         u32 block_size = ext2_fs->blocksize;
1289         u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
1290         struct ext2_inode_large *ext2_inode;
1291         struct ext2_ext_attr_entry *entry;
1292         void *data;
1293         char *buffer = NULL;
1294         char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
1295
1296         if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
1297                 ext2_inode = (struct ext2_inode_large *)inode_buf;
1298         } else {
1299                 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
1300                 if (!ext2_inode)
1301                        return -ENOMEM;
1302         }
1303         err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
1304                                      inode_size);
1305         if (err) {
1306                 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
1307                         error_message(err));
1308                 ret = -1;
1309                 goto out;
1310         }
1311
1312         if (ext2_ino > ext2_fs->super->s_first_ino &&
1313             inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
1314                 if (EXT2_GOOD_OLD_INODE_SIZE +
1315                     ext2_inode->i_extra_isize > inode_size) {
1316                         ret = -EIO;
1317                         goto out;
1318                 }
1319                 if (ext2_inode->i_extra_isize != 0 &&
1320                     EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
1321                     EXT2_EXT_ATTR_MAGIC) {
1322                         inline_ea = 1;
1323                 }
1324         }
1325         if (inline_ea) {
1326                 int total;
1327                 void *end = (void *)ext2_inode + inode_size;
1328                 entry = EXT2_XATTR_IFIRST(ext2_inode);
1329                 total = end - (void *)entry;
1330                 ret = ext2_xattr_check_names(entry, end);
1331                 if (ret)
1332                         goto out;
1333                 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1334                         ret = ext2_xattr_check_entry(entry, total);
1335                         if (ret)
1336                                 goto out;
1337                         data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
1338                                 entry->e_value_offs;
1339                         datalen = entry->e_value_size;
1340                         ret = ext2_copy_single_xattr(trans, root, objectid,
1341                                                 entry, data, datalen);
1342                         if (ret)
1343                                 goto out;
1344                         entry = EXT2_EXT_ATTR_NEXT(entry);
1345                 }
1346         }
1347
1348         if (ext2_inode->i_file_acl == 0)
1349                 goto out;
1350
1351         buffer = malloc(block_size);
1352         if (!buffer) {
1353                 ret = -ENOMEM;
1354                 goto out;
1355         }
1356         err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
1357         if (err) {
1358                 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
1359                         error_message(err));
1360                 ret = -1;
1361                 goto out;
1362         }
1363         ret = ext2_xattr_check_block(buffer, block_size);
1364         if (ret)
1365                 goto out;
1366
1367         entry = EXT2_XATTR_BFIRST(buffer);
1368         while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1369                 ret = ext2_xattr_check_entry(entry, block_size);
1370                 if (ret)
1371                         goto out;
1372                 data = buffer + entry->e_value_offs;
1373                 datalen = entry->e_value_size;
1374                 ret = ext2_copy_single_xattr(trans, root, objectid,
1375                                         entry, data, datalen);
1376                 if (ret)
1377                         goto out;
1378                 entry = EXT2_EXT_ATTR_NEXT(entry);
1379         }
1380 out:
1381         free(buffer);
1382         if ((void *)ext2_inode != inode_buf)
1383                 free(ext2_inode);
1384         return ret;
1385 }
1386 #define MINORBITS       20
1387 #define MKDEV(ma, mi)   (((ma) << MINORBITS) | (mi))
1388
1389 static inline dev_t old_decode_dev(u16 val)
1390 {
1391         return MKDEV((val >> 8) & 255, val & 255);
1392 }
1393
1394 static inline dev_t new_decode_dev(u32 dev)
1395 {
1396         unsigned major = (dev & 0xfff00) >> 8;
1397         unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
1398         return MKDEV(major, minor);
1399 }
1400
1401 static void ext2_copy_inode_item(struct btrfs_inode_item *dst,
1402                            struct ext2_inode *src, u32 blocksize)
1403 {
1404         btrfs_set_stack_inode_generation(dst, 1);
1405         btrfs_set_stack_inode_sequence(dst, 0);
1406         btrfs_set_stack_inode_transid(dst, 1);
1407         btrfs_set_stack_inode_size(dst, src->i_size);
1408         btrfs_set_stack_inode_nbytes(dst, 0);
1409         btrfs_set_stack_inode_block_group(dst, 0);
1410         btrfs_set_stack_inode_nlink(dst, src->i_links_count);
1411         btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
1412         btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
1413         btrfs_set_stack_inode_mode(dst, src->i_mode);
1414         btrfs_set_stack_inode_rdev(dst, 0);
1415         btrfs_set_stack_inode_flags(dst, 0);
1416         btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
1417         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
1418         btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
1419         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
1420         btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
1421         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
1422         btrfs_set_stack_timespec_sec(&dst->otime, 0);
1423         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
1424
1425         if (S_ISDIR(src->i_mode)) {
1426                 btrfs_set_stack_inode_size(dst, 0);
1427                 btrfs_set_stack_inode_nlink(dst, 1);
1428         }
1429         if (S_ISREG(src->i_mode)) {
1430                 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
1431                                            (u64)src->i_size);
1432         }
1433         if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
1434             !S_ISLNK(src->i_mode)) {
1435                 if (src->i_block[0]) {
1436                         btrfs_set_stack_inode_rdev(dst,
1437                                 old_decode_dev(src->i_block[0]));
1438                 } else {
1439                         btrfs_set_stack_inode_rdev(dst,
1440                                 new_decode_dev(src->i_block[1]));
1441                 }
1442         }
1443         memset(&dst->reserved, 0, sizeof(dst->reserved));
1444 }
1445
1446 /*
1447  * copy a single inode. do all the required works, such as cloning
1448  * inode item, creating file extents and creating directory entries.
1449  */
1450 static int ext2_copy_single_inode(struct btrfs_trans_handle *trans,
1451                              struct btrfs_root *root, u64 objectid,
1452                              ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1453                              struct ext2_inode *ext2_inode,
1454                              int datacsum, int packing, int noxattr)
1455 {
1456         int ret;
1457         struct btrfs_inode_item btrfs_inode;
1458
1459         if (ext2_inode->i_links_count == 0)
1460                 return 0;
1461
1462         ext2_copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
1463         if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
1464                 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
1465                             BTRFS_INODE_NODATASUM;
1466                 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
1467         }
1468
1469         switch (ext2_inode->i_mode & S_IFMT) {
1470         case S_IFREG:
1471                 ret = ext2_create_file_extents(trans, root, objectid,
1472                         &btrfs_inode, ext2_fs, ext2_ino, datacsum, packing);
1473                 break;
1474         case S_IFDIR:
1475                 ret = ext2_create_dir_entries(trans, root, objectid,
1476                                 &btrfs_inode, ext2_fs, ext2_ino);
1477                 break;
1478         case S_IFLNK:
1479                 ret = ext2_create_symbol_link(trans, root, objectid,
1480                                 &btrfs_inode, ext2_fs, ext2_ino, ext2_inode);
1481                 break;
1482         default:
1483                 ret = 0;
1484                 break;
1485         }
1486         if (ret)
1487                 return ret;
1488
1489         if (!noxattr) {
1490                 ret = ext2_copy_extended_attrs(trans, root, objectid,
1491                                 &btrfs_inode, ext2_fs, ext2_ino);
1492                 if (ret)
1493                         return ret;
1494         }
1495         return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
1496 }
1497
1498 /*
1499  * scan ext2's inode bitmap and copy all used inodes.
1500  */
1501 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
1502                             struct btrfs_root *root,
1503                             int datacsum, int packing, int noxattr, struct task_ctx *p)
1504 {
1505         ext2_filsys ext2_fs = cctx->fs_data;
1506         int ret;
1507         errcode_t err;
1508         ext2_inode_scan ext2_scan;
1509         struct ext2_inode ext2_inode;
1510         ext2_ino_t ext2_ino;
1511         u64 objectid;
1512         struct btrfs_trans_handle *trans;
1513
1514         trans = btrfs_start_transaction(root, 1);
1515         if (!trans)
1516                 return -ENOMEM;
1517         err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
1518         if (err) {
1519                 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
1520                 return -1;
1521         }
1522         while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
1523                                              &ext2_inode))) {
1524                 /* no more inodes */
1525                 if (ext2_ino == 0)
1526                         break;
1527                 /* skip special inode in ext2fs */
1528                 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
1529                     ext2_ino != EXT2_ROOT_INO)
1530                         continue;
1531                 objectid = ext2_ino + INO_OFFSET;
1532                 ret = ext2_copy_single_inode(trans, root,
1533                                         objectid, ext2_fs, ext2_ino,
1534                                         &ext2_inode, datacsum, packing,
1535                                         noxattr);
1536                 p->cur_copy_inodes++;
1537                 if (ret)
1538                         return ret;
1539                 if (trans->blocks_used >= 4096) {
1540                         ret = btrfs_commit_transaction(trans, root);
1541                         BUG_ON(ret);
1542                         trans = btrfs_start_transaction(root, 1);
1543                         BUG_ON(!trans);
1544                 }
1545         }
1546         if (err) {
1547                 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
1548                 return -1;
1549         }
1550         ret = btrfs_commit_transaction(trans, root);
1551         BUG_ON(ret);
1552         ext2fs_close_inode_scan(ext2_scan);
1553
1554         return ret;
1555 }
1556
1557 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
1558                                 int ensure_size);
1559
1560 /*
1561  * Create the fs image file of old filesystem.
1562  *
1563  * This is completely fs independent as we have cctx->used, only
1564  * need to create file extents pointing to all the positions.
1565  */
1566 static int create_image(struct btrfs_root *root,
1567                            struct btrfs_mkfs_config *cfg,
1568                            struct btrfs_convert_context *cctx, int fd,
1569                            u64 size, char *name, int datacsum)
1570 {
1571         struct btrfs_inode_item buf;
1572         struct btrfs_trans_handle *trans;
1573         struct btrfs_path *path = NULL;
1574         struct btrfs_key key;
1575         struct cache_extent *cache;
1576         struct cache_tree used_tmp;
1577         u64 cur;
1578         u64 ino;
1579         u64 flags = BTRFS_INODE_READONLY;
1580         int ret;
1581
1582         if (!datacsum)
1583                 flags |= BTRFS_INODE_NODATASUM;
1584
1585         trans = btrfs_start_transaction(root, 1);
1586         if (!trans)
1587                 return -ENOMEM;
1588
1589         cache_tree_init(&used_tmp);
1590
1591         ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
1592                                        &ino);
1593         if (ret < 0)
1594                 goto out;
1595         ret = btrfs_new_inode(trans, root, ino, 0400 | S_IFREG);
1596         if (ret < 0)
1597                 goto out;
1598         ret = btrfs_change_inode_flags(trans, root, ino, flags);
1599         if (ret < 0)
1600                 goto out;
1601         ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
1602                              strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
1603         if (ret < 0)
1604                 goto out;
1605
1606         path = btrfs_alloc_path();
1607         if (!path) {
1608                 ret = -ENOMEM;
1609                 goto out;
1610         }
1611         key.objectid = ino;
1612         key.type = BTRFS_INODE_ITEM_KEY;
1613         key.offset = 0;
1614
1615         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1616         if (ret) {
1617                 ret = (ret > 0 ? -ENOENT : ret);
1618                 goto out;
1619         }
1620         read_extent_buffer(path->nodes[0], &buf,
1621                         btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1622                         sizeof(buf));
1623         btrfs_release_path(path);
1624
1625         /*
1626          * Create a new used space cache, which doesn't contain the reserved
1627          * range
1628          */
1629         for (cache = first_cache_extent(&cctx->used); cache;
1630              cache = next_cache_extent(cache)) {
1631                 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1632                 if (ret < 0)
1633                         goto out;
1634         }
1635         ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1636         if (ret < 0)
1637                 goto out;
1638
1639         /*
1640          * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1641          * can't handle bytenr 0(will consider it as a hole)
1642          */
1643         cur = 1024 * 1024;
1644         while (cur < size) {
1645                 u64 len = size - cur;
1646
1647                 ret = create_image_file_range(trans, root, &used_tmp,
1648                                                 &buf, ino, cur, &len, datacsum);
1649                 if (ret < 0)
1650                         goto out;
1651                 cur += len;
1652         }
1653         /* Handle the reserved ranges */
1654         ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1655                                       cfg->num_bytes, datacsum);
1656
1657
1658         key.objectid = ino;
1659         key.type = BTRFS_INODE_ITEM_KEY;
1660         key.offset = 0;
1661         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1662         if (ret) {
1663                 ret = (ret > 0 ? -ENOENT : ret);
1664                 goto out;
1665         }
1666         btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1667         write_extent_buffer(path->nodes[0], &buf,
1668                         btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1669                         sizeof(buf));
1670 out:
1671         free_extent_cache_tree(&used_tmp);
1672         btrfs_free_path(path);
1673         btrfs_commit_transaction(trans, root);
1674         return ret;
1675 }
1676
1677 static struct btrfs_root * link_subvol(struct btrfs_root *root,
1678                 const char *base, u64 root_objectid)
1679 {
1680         struct btrfs_trans_handle *trans;
1681         struct btrfs_fs_info *fs_info = root->fs_info;
1682         struct btrfs_root *tree_root = fs_info->tree_root;
1683         struct btrfs_root *new_root = NULL;
1684         struct btrfs_path *path;
1685         struct btrfs_inode_item *inode_item;
1686         struct extent_buffer *leaf;
1687         struct btrfs_key key;
1688         u64 dirid = btrfs_root_dirid(&root->root_item);
1689         u64 index = 2;
1690         char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1691         int len;
1692         int i;
1693         int ret;
1694
1695         len = strlen(base);
1696         if (len == 0 || len > BTRFS_NAME_LEN)
1697                 return NULL;
1698
1699         path = btrfs_alloc_path();
1700         BUG_ON(!path);
1701
1702         key.objectid = dirid;
1703         key.type = BTRFS_DIR_INDEX_KEY;
1704         key.offset = (u64)-1;
1705
1706         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1707         BUG_ON(ret <= 0);
1708
1709         if (path->slots[0] > 0) {
1710                 path->slots[0]--;
1711                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1712                 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1713                         index = key.offset + 1;
1714         }
1715         btrfs_release_path(path);
1716
1717         trans = btrfs_start_transaction(root, 1);
1718         BUG_ON(!trans);
1719
1720         key.objectid = dirid;
1721         key.offset = 0;
1722         key.type =  BTRFS_INODE_ITEM_KEY;
1723
1724         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1725         BUG_ON(ret);
1726         leaf = path->nodes[0];
1727         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1728                                     struct btrfs_inode_item);
1729
1730         key.objectid = root_objectid;
1731         key.offset = (u64)-1;
1732         key.type = BTRFS_ROOT_ITEM_KEY;
1733
1734         memcpy(buf, base, len);
1735         for (i = 0; i < 1024; i++) {
1736                 ret = btrfs_insert_dir_item(trans, root, buf, len,
1737                                             dirid, &key, BTRFS_FT_DIR, index);
1738                 if (ret != -EEXIST)
1739                         break;
1740                 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1741                 if (len < 1 || len > BTRFS_NAME_LEN) {
1742                         ret = -EINVAL;
1743                         break;
1744                 }
1745         }
1746         if (ret)
1747                 goto fail;
1748
1749         btrfs_set_inode_size(leaf, inode_item, len * 2 +
1750                              btrfs_inode_size(leaf, inode_item));
1751         btrfs_mark_buffer_dirty(leaf);
1752         btrfs_release_path(path);
1753
1754         /* add the backref first */
1755         ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1756                                  BTRFS_ROOT_BACKREF_KEY,
1757                                  root->root_key.objectid,
1758                                  dirid, index, buf, len);
1759         BUG_ON(ret);
1760
1761         /* now add the forward ref */
1762         ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1763                                  BTRFS_ROOT_REF_KEY, root_objectid,
1764                                  dirid, index, buf, len);
1765
1766         ret = btrfs_commit_transaction(trans, root);
1767         BUG_ON(ret);
1768
1769         new_root = btrfs_read_fs_root(fs_info, &key);
1770         if (IS_ERR(new_root))
1771                 new_root = NULL;
1772 fail:
1773         btrfs_free_path(path);
1774         return new_root;
1775 }
1776
1777 static int create_subvol(struct btrfs_trans_handle *trans,
1778                          struct btrfs_root *root, u64 root_objectid)
1779 {
1780         struct extent_buffer *tmp;
1781         struct btrfs_root *new_root;
1782         struct btrfs_key key;
1783         struct btrfs_root_item root_item;
1784         int ret;
1785
1786         ret = btrfs_copy_root(trans, root, root->node, &tmp,
1787                               root_objectid);
1788         BUG_ON(ret);
1789
1790         memcpy(&root_item, &root->root_item, sizeof(root_item));
1791         btrfs_set_root_bytenr(&root_item, tmp->start);
1792         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
1793         btrfs_set_root_generation(&root_item, trans->transid);
1794         free_extent_buffer(tmp);
1795
1796         key.objectid = root_objectid;
1797         key.type = BTRFS_ROOT_ITEM_KEY;
1798         key.offset = trans->transid;
1799         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1800                                 &key, &root_item);
1801
1802         key.offset = (u64)-1;
1803         new_root = btrfs_read_fs_root(root->fs_info, &key);
1804         BUG_ON(!new_root || IS_ERR(new_root));
1805
1806         ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
1807         BUG_ON(ret);
1808
1809         return 0;
1810 }
1811
1812 /*
1813  * New make_btrfs() has handle system and meta chunks quite well.
1814  * So only need to add remaining data chunks.
1815  */
1816 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
1817                                           struct btrfs_fs_info *fs_info,
1818                                           struct btrfs_mkfs_config *cfg,
1819                                           struct btrfs_convert_context *cctx)
1820 {
1821         struct btrfs_root *extent_root = fs_info->extent_root;
1822         struct cache_tree *data_chunks = &cctx->data_chunks;
1823         struct cache_extent *cache;
1824         u64 max_chunk_size;
1825         int ret = 0;
1826
1827         /*
1828          * Don't create data chunk over 10% of the convert device
1829          * And for single chunk, don't create chunk larger than 1G.
1830          */
1831         max_chunk_size = cfg->num_bytes / 10;
1832         max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
1833         max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
1834
1835         for (cache = first_cache_extent(data_chunks); cache;
1836              cache = next_cache_extent(cache)) {
1837                 u64 cur = cache->start;
1838
1839                 while (cur < cache->start + cache->size) {
1840                         u64 len;
1841                         u64 cur_backup = cur;
1842
1843                         len = min(max_chunk_size,
1844                                   cache->start + cache->size - cur);
1845                         ret = btrfs_alloc_data_chunk(trans, extent_root,
1846                                         &cur_backup, len,
1847                                         BTRFS_BLOCK_GROUP_DATA, 1);
1848                         if (ret < 0)
1849                                 break;
1850                         ret = btrfs_make_block_group(trans, extent_root, 0,
1851                                         BTRFS_BLOCK_GROUP_DATA,
1852                                         BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1853                                         cur, len);
1854                         if (ret < 0)
1855                                 break;
1856                         cur += len;
1857                 }
1858         }
1859         return ret;
1860 }
1861
1862 /*
1863  * Init the temp btrfs to a operational status.
1864  *
1865  * It will fix the extent usage accounting(XXX: Do we really need?) and
1866  * insert needed data chunks, to ensure all old fs data extents are covered
1867  * by DATA chunks, preventing wrong chunks are allocated.
1868  *
1869  * And also create convert image subvolume and relocation tree.
1870  * (XXX: Not need again?)
1871  * But the convert image subvolume is *NOT* linked to fs tree yet.
1872  */
1873 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
1874                          struct btrfs_convert_context *cctx, int datacsum,
1875                          int packing, int noxattr)
1876 {
1877         struct btrfs_key location;
1878         struct btrfs_trans_handle *trans;
1879         struct btrfs_fs_info *fs_info = root->fs_info;
1880         int ret;
1881
1882         /*
1883          * Don't alloc any metadata/system chunk, as we don't want
1884          * any meta/sys chunk allcated before all data chunks are inserted.
1885          * Or we screw up the chunk layout just like the old implement.
1886          */
1887         fs_info->avoid_sys_chunk_alloc = 1;
1888         fs_info->avoid_meta_chunk_alloc = 1;
1889         trans = btrfs_start_transaction(root, 1);
1890         BUG_ON(!trans);
1891         ret = btrfs_fix_block_accounting(trans, root);
1892         if (ret)
1893                 goto err;
1894         ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
1895         if (ret)
1896                 goto err;
1897         ret = btrfs_make_root_dir(trans, fs_info->tree_root,
1898                                   BTRFS_ROOT_TREE_DIR_OBJECTID);
1899         if (ret)
1900                 goto err;
1901         memcpy(&location, &root->root_key, sizeof(location));
1902         location.offset = (u64)-1;
1903         ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
1904                                 btrfs_super_root_dir(fs_info->super_copy),
1905                                 &location, BTRFS_FT_DIR, 0);
1906         if (ret)
1907                 goto err;
1908         ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
1909                                 location.objectid,
1910                                 btrfs_super_root_dir(fs_info->super_copy), 0);
1911         if (ret)
1912                 goto err;
1913         btrfs_set_root_dirid(&fs_info->fs_root->root_item,
1914                              BTRFS_FIRST_FREE_OBJECTID);
1915
1916         /* subvol for fs image file */
1917         ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
1918         if (ret < 0)
1919                 goto err;
1920         /* subvol for data relocation tree */
1921         ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1922         if (ret < 0)
1923                 goto err;
1924
1925         ret = btrfs_commit_transaction(trans, root);
1926         fs_info->avoid_sys_chunk_alloc = 0;
1927         fs_info->avoid_meta_chunk_alloc = 0;
1928 err:
1929         return ret;
1930 }
1931
1932 /*
1933  * Migrate super block to its default position and zero 0 ~ 16k
1934  */
1935 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
1936 {
1937         int ret;
1938         struct extent_buffer *buf;
1939         struct btrfs_super_block *super;
1940         u32 len;
1941         u32 bytenr;
1942
1943         BUG_ON(sectorsize < sizeof(*super));
1944         buf = malloc(sizeof(*buf) + sectorsize);
1945         if (!buf)
1946                 return -ENOMEM;
1947
1948         buf->len = sectorsize;
1949         ret = pread(fd, buf->data, sectorsize, old_bytenr);
1950         if (ret != sectorsize)
1951                 goto fail;
1952
1953         super = (struct btrfs_super_block *)buf->data;
1954         BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
1955         btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
1956
1957         csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1958         ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
1959         if (ret != sectorsize)
1960                 goto fail;
1961
1962         ret = fsync(fd);
1963         if (ret)
1964                 goto fail;
1965
1966         memset(buf->data, 0, sectorsize);
1967         for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
1968                 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
1969                 if (len > sectorsize)
1970                         len = sectorsize;
1971                 ret = pwrite(fd, buf->data, len, bytenr);
1972                 if (ret != len) {
1973                         fprintf(stderr, "unable to zero fill device\n");
1974                         break;
1975                 }
1976                 bytenr += len;
1977         }
1978         ret = 0;
1979         fsync(fd);
1980 fail:
1981         free(buf);
1982         if (ret > 0)
1983                 ret = -1;
1984         return ret;
1985 }
1986
1987 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
1988 {
1989         struct btrfs_chunk *chunk;
1990         struct btrfs_disk_key *key;
1991         u32 sectorsize = btrfs_super_sectorsize(super);
1992
1993         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1994         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1995                                        sizeof(struct btrfs_disk_key));
1996
1997         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1998         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1999         btrfs_set_disk_key_offset(key, 0);
2000
2001         btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
2002         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
2003         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
2004         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
2005         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
2006         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
2007         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
2008         btrfs_set_stack_chunk_num_stripes(chunk, 1);
2009         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
2010         chunk->stripe.devid = super->dev_item.devid;
2011         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
2012         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
2013         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
2014         return 0;
2015 }
2016
2017 static const struct btrfs_convert_operations ext2_convert_ops = {
2018         .name                   = "ext2",
2019         .open_fs                = ext2_open_fs,
2020         .read_used_space        = ext2_read_used_space,
2021         .copy_inodes            = ext2_copy_inodes,
2022         .close_fs               = ext2_close_fs,
2023 };
2024
2025 static const struct btrfs_convert_operations *convert_operations[] = {
2026         &ext2_convert_ops,
2027 };
2028
2029 static int convert_open_fs(const char *devname,
2030                            struct btrfs_convert_context *cctx)
2031 {
2032         int i;
2033
2034         memset(cctx, 0, sizeof(*cctx));
2035
2036         for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
2037                 int ret = convert_operations[i]->open_fs(cctx, devname);
2038
2039                 if (ret == 0) {
2040                         cctx->convert_ops = convert_operations[i];
2041                         return ret;
2042                 }
2043         }
2044
2045         fprintf(stderr, "No file system found to convert.\n");
2046         return -1;
2047 }
2048
2049 /*
2050  * Remove one reserve range from given cache tree
2051  * if min_stripe_size is non-zero, it will ensure for split case,
2052  * all its split cache extent is no smaller than @min_strip_size / 2.
2053  */
2054 static int wipe_one_reserved_range(struct cache_tree *tree,
2055                                    u64 start, u64 len, u64 min_stripe_size,
2056                                    int ensure_size)
2057 {
2058         struct cache_extent *cache;
2059         int ret;
2060
2061         BUG_ON(ensure_size && min_stripe_size == 0);
2062         /*
2063          * The logical here is simplified to handle special cases only
2064          * So we don't need to consider merge case for ensure_size
2065          */
2066         BUG_ON(min_stripe_size && (min_stripe_size < len * 2 ||
2067                min_stripe_size / 2 < BTRFS_STRIPE_LEN));
2068
2069         /* Also, wipe range should already be aligned */
2070         BUG_ON(start != round_down(start, BTRFS_STRIPE_LEN) ||
2071                start + len != round_up(start + len, BTRFS_STRIPE_LEN));
2072
2073         min_stripe_size /= 2;
2074
2075         cache = lookup_cache_extent(tree, start, len);
2076         if (!cache)
2077                 return 0;
2078
2079         if (start <= cache->start) {
2080                 /*
2081                  *      |--------cache---------|
2082                  * |-wipe-|
2083                  */
2084                 BUG_ON(start + len <= cache->start);
2085
2086                 /*
2087                  * The wipe size is smaller than min_stripe_size / 2,
2088                  * so the result length should still meet min_stripe_size
2089                  * And no need to do alignment
2090                  */
2091                 cache->size -= (start + len - cache->start);
2092                 if (cache->size == 0) {
2093                         remove_cache_extent(tree, cache);
2094                         free(cache);
2095                         return 0;
2096                 }
2097
2098                 BUG_ON(ensure_size && cache->size < min_stripe_size);
2099
2100                 cache->start = start + len;
2101                 return 0;
2102         } else if (start > cache->start && start + len < cache->start +
2103                    cache->size) {
2104                 /*
2105                  * |-------cache-----|
2106                  *      |-wipe-|
2107                  */
2108                 u64 old_start = cache->start;
2109                 u64 old_len = cache->size;
2110                 u64 insert_start = start + len;
2111                 u64 insert_len;
2112
2113                 cache->size = start - cache->start;
2114                 /* Expand the leading half part if needed */
2115                 if (ensure_size && cache->size < min_stripe_size) {
2116                         ret = _expand_extent_cache(tree, cache,
2117                                         min_stripe_size, 1);
2118                         if (ret < 0)
2119                                 return ret;
2120                 }
2121
2122                 /* And insert the new one */
2123                 insert_len = old_start + old_len - start - len;
2124                 ret = add_merge_cache_extent(tree, insert_start, insert_len);
2125                 if (ret < 0)
2126                         return ret;
2127
2128                 /* Expand the last half part if needed */
2129                 if (ensure_size && insert_len < min_stripe_size) {
2130                         cache = lookup_cache_extent(tree, insert_start,
2131                                                     insert_len);
2132                         if (!cache || cache->start != insert_start ||
2133                             cache->size != insert_len)
2134                                 return -ENOENT;
2135                         ret = _expand_extent_cache(tree, cache,
2136                                         min_stripe_size, 0);
2137                 }
2138
2139                 return ret;
2140         }
2141         /*
2142          * |----cache-----|
2143          *              |--wipe-|
2144          * Wipe len should be small enough and no need to expand the
2145          * remaining extent
2146          */
2147         cache->size = start - cache->start;
2148         BUG_ON(ensure_size && cache->size < min_stripe_size);
2149         return 0;
2150 }
2151
2152 /*
2153  * Remove reserved ranges from given cache_tree
2154  *
2155  * It will remove the following ranges
2156  * 1) 0~1M
2157  * 2) 2nd superblock, +64K (make sure chunks are 64K aligned)
2158  * 3) 3rd superblock, +64K
2159  *
2160  * @min_stripe must be given for safety check
2161  * and if @ensure_size is given, it will ensure affected cache_extent will be
2162  * larger than min_stripe_size
2163  */
2164 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
2165                                 int ensure_size)
2166 {
2167         int ret;
2168
2169         ret = wipe_one_reserved_range(tree, 0, 1024 * 1024, min_stripe_size,
2170                                       ensure_size);
2171         if (ret < 0)
2172                 return ret;
2173         ret = wipe_one_reserved_range(tree, btrfs_sb_offset(1),
2174                         BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
2175         if (ret < 0)
2176                 return ret;
2177         ret = wipe_one_reserved_range(tree, btrfs_sb_offset(2),
2178                         BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
2179         return ret;
2180 }
2181
2182 static int calculate_available_space(struct btrfs_convert_context *cctx)
2183 {
2184         struct cache_tree *used = &cctx->used;
2185         struct cache_tree *data_chunks = &cctx->data_chunks;
2186         struct cache_tree *free = &cctx->free;
2187         struct cache_extent *cache;
2188         u64 cur_off = 0;
2189         /*
2190          * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
2191          * works without need to consider overlap
2192          */
2193         u64 min_stripe_size = 2 * 16 * 1024 * 1024;
2194         int ret;
2195
2196         /* Calculate data_chunks */
2197         for (cache = first_cache_extent(used); cache;
2198              cache = next_cache_extent(cache)) {
2199                 u64 cur_len;
2200
2201                 if (cache->start + cache->size < cur_off)
2202                         continue;
2203                 if (cache->start > cur_off + min_stripe_size)
2204                         cur_off = cache->start;
2205                 cur_len = max(cache->start + cache->size - cur_off,
2206                               min_stripe_size);
2207                 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
2208                 if (ret < 0)
2209                         goto out;
2210                 cur_off += cur_len;
2211         }
2212         /*
2213          * remove reserved ranges, so we won't ever bother relocating an old
2214          * filesystem extent to other place.
2215          */
2216         ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
2217         if (ret < 0)
2218                 goto out;
2219
2220         cur_off = 0;
2221         /*
2222          * Calculate free space
2223          * Always round up the start bytenr, to avoid metadata extent corss
2224          * stripe boundary, as later mkfs_convert() won't have all the extent
2225          * allocation check
2226          */
2227         for (cache = first_cache_extent(data_chunks); cache;
2228              cache = next_cache_extent(cache)) {
2229                 if (cache->start < cur_off)
2230                         continue;
2231                 if (cache->start > cur_off) {
2232                         u64 insert_start;
2233                         u64 len;
2234
2235                         len = cache->start - round_up(cur_off,
2236                                                       BTRFS_STRIPE_LEN);
2237                         insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2238
2239                         ret = add_merge_cache_extent(free, insert_start, len);
2240                         if (ret < 0)
2241                                 goto out;
2242                 }
2243                 cur_off = cache->start + cache->size;
2244         }
2245         /* Don't forget the last range */
2246         if (cctx->total_bytes > cur_off) {
2247                 u64 len = cctx->total_bytes - cur_off;
2248                 u64 insert_start;
2249
2250                 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2251
2252                 ret = add_merge_cache_extent(free, insert_start, len);
2253                 if (ret < 0)
2254                         goto out;
2255         }
2256
2257         /* Remove reserved bytes */
2258         ret = wipe_reserved_ranges(free, min_stripe_size, 0);
2259 out:
2260         return ret;
2261 }
2262 /*
2263  * Read used space, and since we have the used space,
2264  * calcuate data_chunks and free for later mkfs
2265  */
2266 static int convert_read_used_space(struct btrfs_convert_context *cctx)
2267 {
2268         int ret;
2269
2270         ret = cctx->convert_ops->read_used_space(cctx);
2271         if (ret)
2272                 return ret;
2273
2274         ret = calculate_available_space(cctx);
2275         return ret;
2276 }
2277
2278 static int do_convert(const char *devname, int datacsum, int packing,
2279                 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2280                 int progress, u64 features)
2281 {
2282         int ret;
2283         int fd = -1;
2284         int is_btrfs = 0;
2285         u32 blocksize;
2286         u64 total_bytes;
2287         struct btrfs_root *root;
2288         struct btrfs_root *image_root;
2289         struct btrfs_convert_context cctx;
2290         struct btrfs_key key;
2291         char *subvol_name = NULL;
2292         struct task_ctx ctx;
2293         char features_buf[64];
2294         struct btrfs_mkfs_config mkfs_cfg;
2295
2296         init_convert_context(&cctx);
2297         ret = convert_open_fs(devname, &cctx);
2298         if (ret)
2299                 goto fail;
2300         ret = convert_read_used_space(&cctx);
2301         if (ret)
2302                 goto fail;
2303
2304         blocksize = cctx.blocksize;
2305         total_bytes = (u64)blocksize * (u64)cctx.block_count;
2306         if (blocksize < 4096) {
2307                 fprintf(stderr, "block size is too small\n");
2308                 goto fail;
2309         }
2310         if (btrfs_check_nodesize(nodesize, blocksize, features))
2311                 goto fail;
2312         fd = open(devname, O_RDWR);
2313         if (fd < 0) {
2314                 fprintf(stderr, "unable to open %s\n", devname);
2315                 goto fail;
2316         }
2317         btrfs_parse_features_to_string(features_buf, features);
2318         if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2319                 strcat(features_buf, " (default)");
2320
2321         printf("create btrfs filesystem:\n");
2322         printf("\tblocksize: %u\n", blocksize);
2323         printf("\tnodesize:  %u\n", nodesize);
2324         printf("\tfeatures:  %s\n", features_buf);
2325
2326         mkfs_cfg.label = cctx.volume_name;
2327         mkfs_cfg.num_bytes = total_bytes;
2328         mkfs_cfg.nodesize = nodesize;
2329         mkfs_cfg.sectorsize = blocksize;
2330         mkfs_cfg.stripesize = blocksize;
2331         mkfs_cfg.features = features;
2332         /* New convert need these space */
2333         mkfs_cfg.fs_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2334         mkfs_cfg.chunk_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2335         *(mkfs_cfg.fs_uuid) = '\0';
2336         *(mkfs_cfg.chunk_uuid) = '\0';
2337
2338         ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2339         if (ret) {
2340                 fprintf(stderr, "unable to create initial ctree: %s\n",
2341                         strerror(-ret));
2342                 goto fail;
2343         }
2344
2345         root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2346                              OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2347         if (!root) {
2348                 fprintf(stderr, "unable to open ctree\n");
2349                 goto fail;
2350         }
2351         ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2352         if (ret) {
2353                 fprintf(stderr, "unable to setup the root tree\n");
2354                 goto fail;
2355         }
2356
2357         printf("creating %s image file.\n", cctx.convert_ops->name);
2358         ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2359         if (ret < 0) {
2360                 fprintf(stderr, "error allocating subvolume name: %s_saved\n",
2361                         cctx.convert_ops->name);
2362                 goto fail;
2363         }
2364         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2365         key.offset = (u64)-1;
2366         key.type = BTRFS_ROOT_ITEM_KEY;
2367         image_root = btrfs_read_fs_root(root->fs_info, &key);
2368         if (!image_root) {
2369                 fprintf(stderr, "unable to create subvol\n");
2370                 goto fail;
2371         }
2372         ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2373                               mkfs_cfg.num_bytes, "image", datacsum);
2374         if (ret) {
2375                 fprintf(stderr, "error during create_image %d\n", ret);
2376                 goto fail;
2377         }
2378
2379         printf("creating btrfs metadata.\n");
2380         ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2381         ctx.cur_copy_inodes = 0;
2382
2383         if (progress) {
2384                 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2385                                      &ctx);
2386                 task_start(ctx.info);
2387         }
2388         ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2389         if (ret) {
2390                 fprintf(stderr, "error during copy_inodes %d\n", ret);
2391                 goto fail;
2392         }
2393         if (progress) {
2394                 task_stop(ctx.info);
2395                 task_deinit(ctx.info);
2396         }
2397
2398         image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2399
2400         free(subvol_name);
2401
2402         memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2403         if (copylabel == 1) {
2404                 __strncpy_null(root->fs_info->super_copy->label,
2405                                 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2406                 fprintf(stderr, "copy label '%s'\n",
2407                                 root->fs_info->super_copy->label);
2408         } else if (copylabel == -1) {
2409                 strcpy(root->fs_info->super_copy->label, fslabel);
2410                 fprintf(stderr, "set label to '%s'\n", fslabel);
2411         }
2412
2413         ret = close_ctree(root);
2414         if (ret) {
2415                 fprintf(stderr, "error during close_ctree %d\n", ret);
2416                 goto fail;
2417         }
2418         convert_close_fs(&cctx);
2419         clean_convert_context(&cctx);
2420
2421         /*
2422          * If this step succeed, we get a mountable btrfs. Otherwise
2423          * the source fs is left unchanged.
2424          */
2425         ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2426         if (ret) {
2427                 fprintf(stderr, "unable to migrate super block\n");
2428                 goto fail;
2429         }
2430         is_btrfs = 1;
2431
2432         root = open_ctree_fd(fd, devname, 0,
2433                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2434         if (!root) {
2435                 fprintf(stderr, "unable to open ctree\n");
2436                 goto fail;
2437         }
2438         root->fs_info->finalize_on_close = 1;
2439         close_ctree(root);
2440         close(fd);
2441
2442         printf("conversion complete.\n");
2443         return 0;
2444 fail:
2445         clean_convert_context(&cctx);
2446         if (fd != -1)
2447                 close(fd);
2448         if (is_btrfs)
2449                 fprintf(stderr,
2450                         "WARNING: an error occurred during chunk mapping fixup, filesystem mountable but not finalized\n");
2451         else
2452                 fprintf(stderr, "conversion aborted\n");
2453         return -1;
2454 }
2455
2456 /*
2457  * Check if a non 1:1 mapped chunk can be rolled back.
2458  * For new convert, it's OK while for old convert it's not.
2459  */
2460 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2461 {
2462         struct btrfs_block_group_cache *bg;
2463         struct btrfs_key key;
2464         struct btrfs_path path;
2465         struct btrfs_root *extent_root = fs_info->extent_root;
2466         u64 bg_start;
2467         u64 bg_end;
2468         int ret;
2469
2470         bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2471         if (!bg)
2472                 return -ENOENT;
2473         bg_start = bg->key.objectid;
2474         bg_end = bg->key.objectid + bg->key.offset;
2475
2476         key.objectid = bg_end;
2477         key.type = BTRFS_METADATA_ITEM_KEY;
2478         key.offset = 0;
2479         btrfs_init_path(&path);
2480
2481         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2482         if (ret < 0)
2483                 return ret;
2484
2485         while (1) {
2486                 struct btrfs_extent_item *ei;
2487
2488                 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2489                 if (ret > 0) {
2490                         ret = 0;
2491                         break;
2492                 }
2493                 if (ret < 0)
2494                         break;
2495
2496                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2497                 if (key.type == BTRFS_METADATA_ITEM_KEY)
2498                         continue;
2499                 /* Now it's EXTENT_ITEM_KEY only */
2500                 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2501                                     struct btrfs_extent_item);
2502                 /*
2503                  * Found data extent, means this is old convert must follow 1:1
2504                  * mapping.
2505                  */
2506                 if (btrfs_extent_flags(path.nodes[0], ei)
2507                                 & BTRFS_EXTENT_FLAG_DATA) {
2508                         ret = -EINVAL;
2509                         break;
2510                 }
2511         }
2512         btrfs_release_path(&path);
2513         return ret;
2514 }
2515
2516 static int may_rollback(struct btrfs_root *root)
2517 {
2518         struct btrfs_fs_info *info = root->fs_info;
2519         struct btrfs_multi_bio *multi = NULL;
2520         u64 bytenr;
2521         u64 length;
2522         u64 physical;
2523         u64 total_bytes;
2524         int num_stripes;
2525         int ret;
2526
2527         if (btrfs_super_num_devices(info->super_copy) != 1)
2528                 goto fail;
2529
2530         bytenr = BTRFS_SUPER_INFO_OFFSET;
2531         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2532
2533         while (1) {
2534                 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2535                                       &length, &multi, 0, NULL);
2536                 if (ret) {
2537                         if (ret == -ENOENT) {
2538                                 /* removed block group at the tail */
2539                                 if (length == (u64)-1)
2540                                         break;
2541
2542                                 /* removed block group in the middle */
2543                                 goto next;
2544                         }
2545                         goto fail;
2546                 }
2547
2548                 num_stripes = multi->num_stripes;
2549                 physical = multi->stripes[0].physical;
2550                 kfree(multi);
2551
2552                 if (num_stripes != 1) {
2553                         error("num stripes for bytenr %llu is not 1", bytenr);
2554                         goto fail;
2555                 }
2556
2557                 /*
2558                  * Extra check for new convert, as metadata chunk from new
2559                  * convert is much more free than old convert, it doesn't need
2560                  * to do 1:1 mapping.
2561                  */
2562                 if (physical != bytenr) {
2563                         /*
2564                          * Check if it's a metadata chunk and has only metadata
2565                          * extent.
2566                          */
2567                         ret = may_rollback_chunk(info, bytenr);
2568                         if (ret < 0)
2569                                 goto fail;
2570                 }
2571 next:
2572                 bytenr += length;
2573                 if (bytenr >= total_bytes)
2574                         break;
2575         }
2576         return 0;
2577 fail:
2578         return -1;
2579 }
2580
2581 static int do_rollback(const char *devname)
2582 {
2583         int fd = -1;
2584         int ret;
2585         int i;
2586         struct btrfs_root *root;
2587         struct btrfs_root *image_root;
2588         struct btrfs_root *chunk_root;
2589         struct btrfs_dir_item *dir;
2590         struct btrfs_inode_item *inode;
2591         struct btrfs_file_extent_item *fi;
2592         struct btrfs_trans_handle *trans;
2593         struct extent_buffer *leaf;
2594         struct btrfs_block_group_cache *cache1;
2595         struct btrfs_block_group_cache *cache2;
2596         struct btrfs_key key;
2597         struct btrfs_path path;
2598         struct extent_io_tree io_tree;
2599         char *buf = NULL;
2600         char *name;
2601         u64 bytenr;
2602         u64 num_bytes;
2603         u64 root_dir;
2604         u64 objectid;
2605         u64 offset;
2606         u64 start;
2607         u64 end;
2608         u64 sb_bytenr;
2609         u64 first_free;
2610         u64 total_bytes;
2611         u32 sectorsize;
2612
2613         extent_io_tree_init(&io_tree);
2614
2615         fd = open(devname, O_RDWR);
2616         if (fd < 0) {
2617                 fprintf(stderr, "unable to open %s\n", devname);
2618                 goto fail;
2619         }
2620         root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2621         if (!root) {
2622                 fprintf(stderr, "unable to open ctree\n");
2623                 goto fail;
2624         }
2625         ret = may_rollback(root);
2626         if (ret < 0) {
2627                 fprintf(stderr, "unable to do rollback\n");
2628                 goto fail;
2629         }
2630
2631         sectorsize = root->sectorsize;
2632         buf = malloc(sectorsize);
2633         if (!buf) {
2634                 fprintf(stderr, "unable to allocate memory\n");
2635                 goto fail;
2636         }
2637
2638         btrfs_init_path(&path);
2639
2640         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2641         key.type = BTRFS_ROOT_BACKREF_KEY;
2642         key.offset = BTRFS_FS_TREE_OBJECTID;
2643         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2644                                 0);
2645         btrfs_release_path(&path);
2646         if (ret > 0) {
2647                 fprintf(stderr,
2648                 "ERROR: unable to convert ext2 image subvolume, is it deleted?\n");
2649                 goto fail;
2650         } else if (ret < 0) {
2651                 fprintf(stderr,
2652                         "ERROR: unable to open ext2_saved, id=%llu: %s\n",
2653                         (unsigned long long)key.objectid, strerror(-ret));
2654                 goto fail;
2655         }
2656
2657         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2658         key.type = BTRFS_ROOT_ITEM_KEY;
2659         key.offset = (u64)-1;
2660         image_root = btrfs_read_fs_root(root->fs_info, &key);
2661         if (!image_root || IS_ERR(image_root)) {
2662                 fprintf(stderr, "unable to open subvol %llu\n",
2663                         (unsigned long long)key.objectid);
2664                 goto fail;
2665         }
2666
2667         name = "image";
2668         root_dir = btrfs_root_dirid(&root->root_item);
2669         dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2670                                    root_dir, name, strlen(name), 0);
2671         if (!dir || IS_ERR(dir)) {
2672                 fprintf(stderr, "unable to find file %s\n", name);
2673                 goto fail;
2674         }
2675         leaf = path.nodes[0];
2676         btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2677         btrfs_release_path(&path);
2678
2679         objectid = key.objectid;
2680
2681         ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2682         if (ret) {
2683                 fprintf(stderr, "unable to find inode item\n");
2684                 goto fail;
2685         }
2686         leaf = path.nodes[0];
2687         inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2688         total_bytes = btrfs_inode_size(leaf, inode);
2689         btrfs_release_path(&path);
2690
2691         key.objectid = objectid;
2692         key.offset = 0;
2693         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2694         ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2695         if (ret != 0) {
2696                 fprintf(stderr, "unable to find first file extent\n");
2697                 btrfs_release_path(&path);
2698                 goto fail;
2699         }
2700
2701         /* build mapping tree for the relocated blocks */
2702         for (offset = 0; offset < total_bytes; ) {
2703                 leaf = path.nodes[0];
2704                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2705                         ret = btrfs_next_leaf(root, &path);
2706                         if (ret != 0)
2707                                 break;
2708                         continue;
2709                 }
2710
2711                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2712                 if (key.objectid != objectid || key.offset != offset ||
2713                     btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2714                         break;
2715
2716                 fi = btrfs_item_ptr(leaf, path.slots[0],
2717                                     struct btrfs_file_extent_item);
2718                 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2719                         break;
2720                 if (btrfs_file_extent_compression(leaf, fi) ||
2721                     btrfs_file_extent_encryption(leaf, fi) ||
2722                     btrfs_file_extent_other_encoding(leaf, fi))
2723                         break;
2724
2725                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2726                 /* skip holes and direct mapped extents */
2727                 if (bytenr == 0 || bytenr == offset)
2728                         goto next_extent;
2729
2730                 bytenr += btrfs_file_extent_offset(leaf, fi);
2731                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2732
2733                 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2734                 cache2 = btrfs_lookup_block_group(root->fs_info,
2735                                                   offset + num_bytes - 1);
2736                 /*
2737                  * Here we must take consideration of old and new convert
2738                  * behavior.
2739                  * For old convert case, sign, there is no consist chunk type
2740                  * that will cover the extent. META/DATA/SYS are all possible.
2741                  * Just ensure relocate one is in SYS chunk.
2742                  * For new convert case, they are all covered by DATA chunk.
2743                  *
2744                  * So, there is not valid chunk type check for it now.
2745                  */
2746                 if (cache1 != cache2)
2747                         break;
2748
2749                 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2750                                 EXTENT_LOCKED, GFP_NOFS);
2751                 set_state_private(&io_tree, offset, bytenr);
2752 next_extent:
2753                 offset += btrfs_file_extent_num_bytes(leaf, fi);
2754                 path.slots[0]++;
2755         }
2756         btrfs_release_path(&path);
2757
2758         if (offset < total_bytes) {
2759                 fprintf(stderr, "unable to build extent mapping\n");
2760                 fprintf(stderr, "converted filesystem after balance is unable to rollback\n");
2761                 goto fail;
2762         }
2763
2764         first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2765         first_free &= ~((u64)sectorsize - 1);
2766         /* backup for extent #0 should exist */
2767         if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2768                 fprintf(stderr, "no backup for the first extent\n");
2769                 goto fail;
2770         }
2771         /* force no allocation from system block group */
2772         root->fs_info->system_allocs = -1;
2773         trans = btrfs_start_transaction(root, 1);
2774         BUG_ON(!trans);
2775         /*
2776          * recow the whole chunk tree, this will remove all chunk tree blocks
2777          * from system block group
2778          */
2779         chunk_root = root->fs_info->chunk_root;
2780         memset(&key, 0, sizeof(key));
2781         while (1) {
2782                 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2783                 if (ret < 0)
2784                         break;
2785
2786                 ret = btrfs_next_leaf(chunk_root, &path);
2787                 if (ret)
2788                         break;
2789
2790                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2791                 btrfs_release_path(&path);
2792         }
2793         btrfs_release_path(&path);
2794
2795         offset = 0;
2796         num_bytes = 0;
2797         while(1) {
2798                 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2799                 if (!cache1)
2800                         break;
2801
2802                 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2803                         num_bytes += btrfs_block_group_used(&cache1->item);
2804
2805                 offset = cache1->key.objectid + cache1->key.offset;
2806         }
2807         /* only extent #0 left in system block group? */
2808         if (num_bytes > first_free) {
2809                 fprintf(stderr, "unable to empty system block group\n");
2810                 goto fail;
2811         }
2812         /* create a system chunk that maps the whole device */
2813         ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2814         if (ret) {
2815                 fprintf(stderr, "unable to update system chunk\n");
2816                 goto fail;
2817         }
2818
2819         ret = btrfs_commit_transaction(trans, root);
2820         BUG_ON(ret);
2821
2822         ret = close_ctree(root);
2823         if (ret) {
2824                 fprintf(stderr, "error during close_ctree %d\n", ret);
2825                 goto fail;
2826         }
2827
2828         /* zero btrfs super block mirrors */
2829         memset(buf, 0, sectorsize);
2830         for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2831                 bytenr = btrfs_sb_offset(i);
2832                 if (bytenr >= total_bytes)
2833                         break;
2834                 ret = pwrite(fd, buf, sectorsize, bytenr);
2835                 if (ret != sectorsize) {
2836                         fprintf(stderr,
2837                                 "error during zeroing superblock %d: %d\n",
2838                                 i, ret);
2839                         goto fail;
2840                 }
2841         }
2842
2843         sb_bytenr = (u64)-1;
2844         /* copy all relocated blocks back */
2845         while(1) {
2846                 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2847                                             EXTENT_LOCKED);
2848                 if (ret)
2849                         break;
2850
2851                 ret = get_state_private(&io_tree, start, &bytenr);
2852                 BUG_ON(ret);
2853
2854                 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2855                                   GFP_NOFS);
2856
2857                 while (start <= end) {
2858                         if (start == BTRFS_SUPER_INFO_OFFSET) {
2859                                 sb_bytenr = bytenr;
2860                                 goto next_sector;
2861                         }
2862                         ret = pread(fd, buf, sectorsize, bytenr);
2863                         if (ret < 0) {
2864                                 fprintf(stderr, "error during pread %d\n", ret);
2865                                 goto fail;
2866                         }
2867                         BUG_ON(ret != sectorsize);
2868                         ret = pwrite(fd, buf, sectorsize, start);
2869                         if (ret < 0) {
2870                                 fprintf(stderr, "error during pwrite %d\n", ret);
2871                                 goto fail;
2872                         }
2873                         BUG_ON(ret != sectorsize);
2874 next_sector:
2875                         start += sectorsize;
2876                         bytenr += sectorsize;
2877                 }
2878         }
2879
2880         ret = fsync(fd);
2881         if (ret) {
2882                 fprintf(stderr, "error during fsync %d\n", ret);
2883                 goto fail;
2884         }
2885         /*
2886          * finally, overwrite btrfs super block.
2887          */
2888         ret = pread(fd, buf, sectorsize, sb_bytenr);
2889         if (ret < 0) {
2890                 fprintf(stderr, "error during pread %d\n", ret);
2891                 goto fail;
2892         }
2893         BUG_ON(ret != sectorsize);
2894         ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2895         if (ret < 0) {
2896                 fprintf(stderr, "error during pwrite %d\n", ret);
2897                 goto fail;
2898         }
2899         BUG_ON(ret != sectorsize);
2900         ret = fsync(fd);
2901         if (ret) {
2902                 fprintf(stderr, "error during fsync %d\n", ret);
2903                 goto fail;
2904         }
2905
2906         close(fd);
2907         free(buf);
2908         extent_io_tree_cleanup(&io_tree);
2909         printf("rollback complete.\n");
2910         return 0;
2911
2912 fail:
2913         if (fd != -1)
2914                 close(fd);
2915         free(buf);
2916         fprintf(stderr, "rollback aborted.\n");
2917         return -1;
2918 }
2919
2920 static void print_usage(void)
2921 {
2922         printf("usage: btrfs-convert [options] device\n");
2923         printf("options:\n");
2924         printf("\t-d|--no-datasum        disable data checksum, sets NODATASUM\n");
2925         printf("\t-i|--no-xattr          ignore xattrs and ACLs\n");
2926         printf("\t-n|--no-inline         disable inlining of small files to metadata\n");
2927         printf("\t-N|--nodesize SIZE     set filesystem metadata nodesize\n");
2928         printf("\t-r|--rollback          roll back to the original filesystem\n");
2929         printf("\t-l|--label LABEL       set filesystem label\n");
2930         printf("\t-L|--copy-label        use label from converted filesystem\n");
2931         printf("\t-p|--progress          show converting progress (default)\n");
2932         printf("\t-O|--features LIST     comma separated list of filesystem features\n");
2933         printf("\t--no-progress          show only overview, not the detailed progress\n");
2934 }
2935
2936 int main(int argc, char *argv[])
2937 {
2938         int ret;
2939         int packing = 1;
2940         int noxattr = 0;
2941         int datacsum = 1;
2942         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
2943                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
2944         int rollback = 0;
2945         int copylabel = 0;
2946         int usage_error = 0;
2947         int progress = 1;
2948         char *file;
2949         char fslabel[BTRFS_LABEL_SIZE];
2950         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
2951
2952         while(1) {
2953                 enum { GETOPT_VAL_NO_PROGRESS = 256 };
2954                 static const struct option long_options[] = {
2955                         { "no-progress", no_argument, NULL,
2956                                 GETOPT_VAL_NO_PROGRESS },
2957                         { "no-datasum", no_argument, NULL, 'd' },
2958                         { "no-inline", no_argument, NULL, 'n' },
2959                         { "no-xattr", no_argument, NULL, 'i' },
2960                         { "rollback", no_argument, NULL, 'r' },
2961                         { "features", required_argument, NULL, 'O' },
2962                         { "progress", no_argument, NULL, 'p' },
2963                         { "label", required_argument, NULL, 'l' },
2964                         { "copy-label", no_argument, NULL, 'L' },
2965                         { "nodesize", required_argument, NULL, 'N' },
2966                         { "help", no_argument, NULL, GETOPT_VAL_HELP},
2967                         { NULL, 0, NULL, 0 }
2968                 };
2969                 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
2970
2971                 if (c < 0)
2972                         break;
2973                 switch(c) {
2974                         case 'd':
2975                                 datacsum = 0;
2976                                 break;
2977                         case 'i':
2978                                 noxattr = 1;
2979                                 break;
2980                         case 'n':
2981                                 packing = 0;
2982                                 break;
2983                         case 'N':
2984                                 nodesize = parse_size(optarg);
2985                                 break;
2986                         case 'r':
2987                                 rollback = 1;
2988                                 break;
2989                         case 'l':
2990                                 copylabel = -1;
2991                                 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
2992                                         fprintf(stderr,
2993                                 "WARNING: label too long, trimmed to %d bytes\n",
2994                                                 BTRFS_LABEL_SIZE - 1);
2995                                 }
2996                                 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
2997                                 break;
2998                         case 'L':
2999                                 copylabel = 1;
3000                                 break;
3001                         case 'p':
3002                                 progress = 1;
3003                                 break;
3004                         case 'O': {
3005                                 char *orig = strdup(optarg);
3006                                 char *tmp = orig;
3007
3008                                 tmp = btrfs_parse_fs_features(tmp, &features);
3009                                 if (tmp) {
3010                                         fprintf(stderr,
3011                                                 "Unrecognized filesystem feature '%s'\n",
3012                                                         tmp);
3013                                         free(orig);
3014                                         exit(1);
3015                                 }
3016                                 free(orig);
3017                                 if (features & BTRFS_FEATURE_LIST_ALL) {
3018                                         btrfs_list_all_fs_features(
3019                                                 ~BTRFS_CONVERT_ALLOWED_FEATURES);
3020                                         exit(0);
3021                                 }
3022                                 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
3023                                         char buf[64];
3024
3025                                         btrfs_parse_features_to_string(buf,
3026                                                 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
3027                                         fprintf(stderr,
3028                                                 "ERROR: features not allowed for convert: %s\n",
3029                                                 buf);
3030                                         exit(1);
3031                                 }
3032
3033                                 break;
3034                                 }
3035                         case GETOPT_VAL_NO_PROGRESS:
3036                                 progress = 0;
3037                                 break;
3038                         case GETOPT_VAL_HELP:
3039                         default:
3040                                 print_usage();
3041                                 return c != GETOPT_VAL_HELP;
3042                 }
3043         }
3044         set_argv0(argv);
3045         if (check_argc_exact(argc - optind, 1)) {
3046                 print_usage();
3047                 return 1;
3048         }
3049
3050         if (rollback && (!datacsum || noxattr || !packing)) {
3051                 fprintf(stderr,
3052                         "Usage error: -d, -i, -n options do not apply to rollback\n");
3053                 usage_error++;
3054         }
3055
3056         if (usage_error) {
3057                 print_usage();
3058                 return 1;
3059         }
3060
3061         file = argv[optind];
3062         ret = check_mounted(file);
3063         if (ret < 0) {
3064                 fprintf(stderr, "Could not check mount status: %s\n",
3065                         strerror(-ret));
3066                 return 1;
3067         } else if (ret) {
3068                 fprintf(stderr, "%s is mounted\n", file);
3069                 return 1;
3070         }
3071
3072         if (rollback) {
3073                 ret = do_rollback(file);
3074         } else {
3075                 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3076                                 copylabel, fslabel, progress, features);
3077         }
3078         if (ret)
3079                 return 1;
3080         return 0;
3081 }