btrfs-progs: convert: use on-stack path buffer in record_file_blocks
[platform/upstream/btrfs-progs.git] / 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 #if BTRFSCONVERT_EXT2
42 #include <ext2fs/ext2_fs.h>
43 #include <ext2fs/ext2fs.h>
44 #include <ext2fs/ext2_ext_attr.h>
45
46 #define INO_OFFSET (BTRFS_FIRST_FREE_OBJECTID - EXT2_ROOT_INO)
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 #endif
61
62 #define CONV_IMAGE_SUBVOL_OBJECTID BTRFS_FIRST_FREE_OBJECTID
63
64 struct task_ctx {
65         uint32_t max_copy_inodes;
66         uint32_t cur_copy_inodes;
67         struct task_info *info;
68 };
69
70 static void *print_copied_inodes(void *p)
71 {
72         struct task_ctx *priv = p;
73         const char work_indicator[] = { '.', 'o', 'O', 'o' };
74         uint32_t count = 0;
75
76         task_period_start(priv->info, 1000 /* 1s */);
77         while (1) {
78                 count++;
79                 printf("copy inodes [%c] [%10d/%10d]\r",
80                        work_indicator[count % 4], priv->cur_copy_inodes,
81                        priv->max_copy_inodes);
82                 fflush(stdout);
83                 task_period_wait(priv->info);
84         }
85
86         return NULL;
87 }
88
89 static int after_copied_inodes(void *p)
90 {
91         printf("\n");
92         fflush(stdout);
93
94         return 0;
95 }
96
97 struct btrfs_convert_context;
98 struct btrfs_convert_operations {
99         const char *name;
100         int (*open_fs)(struct btrfs_convert_context *cctx, const char *devname);
101         int (*read_used_space)(struct btrfs_convert_context *cctx);
102         int (*copy_inodes)(struct btrfs_convert_context *cctx,
103                          struct btrfs_root *root, int datacsum,
104                          int packing, int noxattr, struct task_ctx *p);
105         void (*close_fs)(struct btrfs_convert_context *cctx);
106         int (*check_state)(struct btrfs_convert_context *cctx);
107 };
108
109 static void init_convert_context(struct btrfs_convert_context *cctx)
110 {
111         cache_tree_init(&cctx->used);
112         cache_tree_init(&cctx->data_chunks);
113         cache_tree_init(&cctx->free);
114 }
115
116 static void clean_convert_context(struct btrfs_convert_context *cctx)
117 {
118         free_extent_cache_tree(&cctx->used);
119         free_extent_cache_tree(&cctx->data_chunks);
120         free_extent_cache_tree(&cctx->free);
121 }
122
123 static inline int copy_inodes(struct btrfs_convert_context *cctx,
124                               struct btrfs_root *root, int datacsum,
125                               int packing, int noxattr, struct task_ctx *p)
126 {
127         return cctx->convert_ops->copy_inodes(cctx, root, datacsum, packing,
128                                              noxattr, p);
129 }
130
131 static inline void convert_close_fs(struct btrfs_convert_context *cctx)
132 {
133         cctx->convert_ops->close_fs(cctx);
134 }
135
136 static inline int convert_check_state(struct btrfs_convert_context *cctx)
137 {
138         return cctx->convert_ops->check_state(cctx);
139 }
140
141 static int intersect_with_sb(u64 bytenr, u64 num_bytes)
142 {
143         int i;
144         u64 offset;
145
146         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
147                 offset = btrfs_sb_offset(i);
148                 offset &= ~((u64)BTRFS_STRIPE_LEN - 1);
149
150                 if (bytenr < offset + BTRFS_STRIPE_LEN &&
151                     bytenr + num_bytes > offset)
152                         return 1;
153         }
154         return 0;
155 }
156
157 static int convert_insert_dirent(struct btrfs_trans_handle *trans,
158                                  struct btrfs_root *root,
159                                  const char *name, size_t name_len,
160                                  u64 dir, u64 objectid,
161                                  u8 file_type, u64 index_cnt,
162                                  struct btrfs_inode_item *inode)
163 {
164         int ret;
165         u64 inode_size;
166         struct btrfs_key location = {
167                 .objectid = objectid,
168                 .offset = 0,
169                 .type = BTRFS_INODE_ITEM_KEY,
170         };
171
172         ret = btrfs_insert_dir_item(trans, root, name, name_len,
173                                     dir, &location, file_type, index_cnt);
174         if (ret)
175                 return ret;
176         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
177                                      objectid, dir, index_cnt);
178         if (ret)
179                 return ret;
180         inode_size = btrfs_stack_inode_size(inode) + name_len * 2;
181         btrfs_set_stack_inode_size(inode, inode_size);
182
183         return 0;
184 }
185
186 static int read_disk_extent(struct btrfs_root *root, u64 bytenr,
187                             u32 num_bytes, char *buffer)
188 {
189         int ret;
190         struct btrfs_fs_devices *fs_devs = root->fs_info->fs_devices;
191
192         ret = pread(fs_devs->latest_bdev, buffer, num_bytes, bytenr);
193         if (ret != num_bytes)
194                 goto fail;
195         ret = 0;
196 fail:
197         if (ret > 0)
198                 ret = -1;
199         return ret;
200 }
201
202 static int csum_disk_extent(struct btrfs_trans_handle *trans,
203                             struct btrfs_root *root,
204                             u64 disk_bytenr, u64 num_bytes)
205 {
206         u32 blocksize = root->sectorsize;
207         u64 offset;
208         char *buffer;
209         int ret = 0;
210
211         buffer = malloc(blocksize);
212         if (!buffer)
213                 return -ENOMEM;
214         for (offset = 0; offset < num_bytes; offset += blocksize) {
215                 ret = read_disk_extent(root, disk_bytenr + offset,
216                                         blocksize, buffer);
217                 if (ret)
218                         break;
219                 ret = btrfs_csum_file_block(trans,
220                                             root->fs_info->csum_root,
221                                             disk_bytenr + num_bytes,
222                                             disk_bytenr + offset,
223                                             buffer, blocksize);
224                 if (ret)
225                         break;
226         }
227         free(buffer);
228         return ret;
229 }
230
231 struct blk_iterate_data {
232         struct btrfs_trans_handle *trans;
233         struct btrfs_root *root;
234         struct btrfs_root *convert_root;
235         struct btrfs_inode_item *inode;
236         u64 convert_ino;
237         u64 objectid;
238         u64 first_block;
239         u64 disk_block;
240         u64 num_blocks;
241         u64 boundary;
242         int checksum;
243         int errcode;
244 };
245
246 static void init_blk_iterate_data(struct blk_iterate_data *data,
247                                   struct btrfs_trans_handle *trans,
248                                   struct btrfs_root *root,
249                                   struct btrfs_inode_item *inode,
250                                   u64 objectid, int checksum)
251 {
252         struct btrfs_key key;
253
254         data->trans             = trans;
255         data->root              = root;
256         data->inode             = inode;
257         data->objectid          = objectid;
258         data->first_block       = 0;
259         data->disk_block        = 0;
260         data->num_blocks        = 0;
261         data->boundary          = (u64)-1;
262         data->checksum          = checksum;
263         data->errcode           = 0;
264
265         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
266         key.type = BTRFS_ROOT_ITEM_KEY;
267         key.offset = (u64)-1;
268         data->convert_root = btrfs_read_fs_root(root->fs_info, &key);
269         /* Impossible as we just opened it before */
270         BUG_ON(!data->convert_root || IS_ERR(data->convert_root));
271         data->convert_ino = BTRFS_FIRST_FREE_OBJECTID + 1;
272 }
273
274 /*
275  * Record a file extent in original filesystem into btrfs one.
276  * The special point is, old disk_block can point to a reserved range.
277  * So here, we don't use disk_block directly but search convert_root
278  * to get the real disk_bytenr.
279  */
280 static int record_file_blocks(struct blk_iterate_data *data,
281                               u64 file_block, u64 disk_block, u64 num_blocks)
282 {
283         int ret = 0;
284         struct btrfs_root *root = data->root;
285         struct btrfs_root *convert_root = data->convert_root;
286         struct btrfs_path path;
287         u64 file_pos = file_block * root->sectorsize;
288         u64 old_disk_bytenr = disk_block * root->sectorsize;
289         u64 num_bytes = num_blocks * root->sectorsize;
290         u64 cur_off = old_disk_bytenr;
291
292         /* Hole, pass it to record_file_extent directly */
293         if (old_disk_bytenr == 0)
294                 return btrfs_record_file_extent(data->trans, root,
295                                 data->objectid, data->inode, file_pos, 0,
296                                 num_bytes);
297
298         btrfs_init_path(&path);
299
300         /*
301          * Search real disk bytenr from convert root
302          */
303         while (cur_off < old_disk_bytenr + num_bytes) {
304                 struct btrfs_key key;
305                 struct btrfs_file_extent_item *fi;
306                 struct extent_buffer *node;
307                 int slot;
308                 u64 extent_disk_bytenr;
309                 u64 extent_num_bytes;
310                 u64 real_disk_bytenr;
311                 u64 cur_len;
312
313                 key.objectid = data->convert_ino;
314                 key.type = BTRFS_EXTENT_DATA_KEY;
315                 key.offset = cur_off;
316
317                 ret = btrfs_search_slot(NULL, convert_root, &key, &path, 0, 0);
318                 if (ret < 0)
319                         break;
320                 if (ret > 0) {
321                         ret = btrfs_previous_item(convert_root, &path,
322                                                   data->convert_ino,
323                                                   BTRFS_EXTENT_DATA_KEY);
324                         if (ret < 0)
325                                 break;
326                         if (ret > 0) {
327                                 ret = -ENOENT;
328                                 break;
329                         }
330                 }
331                 node = path.nodes[0];
332                 slot = path.slots[0];
333                 btrfs_item_key_to_cpu(node, &key, slot);
334                 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY ||
335                        key.objectid != data->convert_ino ||
336                        key.offset > cur_off);
337                 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
338                 extent_disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
339                 extent_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
340                 BUG_ON(cur_off - key.offset >= extent_num_bytes);
341                 btrfs_release_path(&path);
342
343                 if (extent_disk_bytenr)
344                         real_disk_bytenr = cur_off - key.offset +
345                                            extent_disk_bytenr;
346                 else
347                         real_disk_bytenr = 0;
348                 cur_len = min(key.offset + extent_num_bytes,
349                               old_disk_bytenr + num_bytes) - cur_off;
350                 ret = btrfs_record_file_extent(data->trans, data->root,
351                                         data->objectid, data->inode, file_pos,
352                                         real_disk_bytenr, cur_len);
353                 if (ret < 0)
354                         break;
355                 cur_off += cur_len;
356                 file_pos += cur_len;
357
358                 /*
359                  * No need to care about csum
360                  * As every byte of old fs image is calculated for csum, no
361                  * need to waste CPU cycles now.
362                  */
363         }
364         btrfs_release_path(&path);
365         return ret;
366 }
367
368 static int block_iterate_proc(u64 disk_block, u64 file_block,
369                               struct blk_iterate_data *idata)
370 {
371         int ret = 0;
372         int sb_region;
373         int do_barrier;
374         struct btrfs_root *root = idata->root;
375         struct btrfs_block_group_cache *cache;
376         u64 bytenr = disk_block * root->sectorsize;
377
378         sb_region = intersect_with_sb(bytenr, root->sectorsize);
379         do_barrier = sb_region || disk_block >= idata->boundary;
380         if ((idata->num_blocks > 0 && do_barrier) ||
381             (file_block > idata->first_block + idata->num_blocks) ||
382             (disk_block != idata->disk_block + idata->num_blocks)) {
383                 if (idata->num_blocks > 0) {
384                         ret = record_file_blocks(idata, idata->first_block,
385                                                  idata->disk_block,
386                                                  idata->num_blocks);
387                         if (ret)
388                                 goto fail;
389                         idata->first_block += idata->num_blocks;
390                         idata->num_blocks = 0;
391                 }
392                 if (file_block > idata->first_block) {
393                         ret = record_file_blocks(idata, idata->first_block,
394                                         0, file_block - idata->first_block);
395                         if (ret)
396                                 goto fail;
397                 }
398
399                 if (sb_region) {
400                         bytenr += BTRFS_STRIPE_LEN - 1;
401                         bytenr &= ~((u64)BTRFS_STRIPE_LEN - 1);
402                 } else {
403                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
404                         BUG_ON(!cache);
405                         bytenr = cache->key.objectid + cache->key.offset;
406                 }
407
408                 idata->first_block = file_block;
409                 idata->disk_block = disk_block;
410                 idata->boundary = bytenr / root->sectorsize;
411         }
412         idata->num_blocks++;
413 fail:
414         return ret;
415 }
416
417 static int create_image_file_range(struct btrfs_trans_handle *trans,
418                                       struct btrfs_root *root,
419                                       struct cache_tree *used,
420                                       struct btrfs_inode_item *inode,
421                                       u64 ino, u64 bytenr, u64 *ret_len,
422                                       int datacsum)
423 {
424         struct cache_extent *cache;
425         struct btrfs_block_group_cache *bg_cache;
426         u64 len = *ret_len;
427         u64 disk_bytenr;
428         int i;
429         int ret;
430
431         if (bytenr != round_down(bytenr, root->sectorsize)) {
432                 error("bytenr not sectorsize aligned: %llu",
433                                 (unsigned long long)bytenr);
434                 return -EINVAL;
435         }
436         if (len != round_down(len, root->sectorsize)) {
437                 error("length not sectorsize aligned: %llu",
438                                 (unsigned long long)len);
439                 return -EINVAL;
440         }
441         len = min_t(u64, len, BTRFS_MAX_EXTENT_SIZE);
442
443         /*
444          * Skip sb ranges first
445          * [0, 1M), [sb_offset(1), +64K), [sb_offset(2), +64K].
446          *
447          * Or we will insert a hole into current image file, and later
448          * migrate block will fail as there is already a file extent.
449          */
450         if (bytenr < 1024 * 1024) {
451                 *ret_len = 1024 * 1024 - bytenr;
452                 return 0;
453         }
454         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
455                 u64 cur = btrfs_sb_offset(i);
456
457                 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
458                         *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
459                         return 0;
460                 }
461         }
462         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
463                 u64 cur = btrfs_sb_offset(i);
464
465                 /*
466                  *      |--reserved--|
467                  * |----range-------|
468                  * May still need to go through file extent inserts
469                  */
470                 if (bytenr < cur && bytenr + len >= cur) {
471                         len = min_t(u64, len, cur - bytenr);
472                         break;
473                 }
474                 /*
475                  * |--reserved--|
476                  *      |---range---|
477                  * Drop out, no need to insert anything
478                  */
479                 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
480                         *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
481                         return 0;
482                 }
483         }
484
485         cache = search_cache_extent(used, bytenr);
486         if (cache) {
487                 if (cache->start <= bytenr) {
488                         /*
489                          * |///////Used///////|
490                          *      |<--insert--->|
491                          *      bytenr
492                          */
493                         len = min_t(u64, len, cache->start + cache->size -
494                                     bytenr);
495                         disk_bytenr = bytenr;
496                 } else {
497                         /*
498                          *              |//Used//|
499                          *  |<-insert-->|
500                          *  bytenr
501                          */
502                         len = min(len, cache->start - bytenr);
503                         disk_bytenr = 0;
504                         datacsum = 0;
505                 }
506         } else {
507                 /*
508                  * |//Used//|           |EOF
509                  *          |<-insert-->|
510                  *          bytenr
511                  */
512                 disk_bytenr = 0;
513                 datacsum = 0;
514         }
515
516         if (disk_bytenr) {
517                 /* Check if the range is in a data block group */
518                 bg_cache = btrfs_lookup_block_group(root->fs_info, bytenr);
519                 if (!bg_cache)
520                         return -ENOENT;
521                 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
522                         return -EINVAL;
523
524                 /* The extent should never cross block group boundary */
525                 len = min_t(u64, len, bg_cache->key.objectid +
526                             bg_cache->key.offset - bytenr);
527         }
528
529         if (len != round_down(len, root->sectorsize)) {
530                 error("remaining length not sectorsize aligned: %llu",
531                                 (unsigned long long)len);
532                 return -EINVAL;
533         }
534         ret = btrfs_record_file_extent(trans, root, ino, inode, bytenr,
535                                        disk_bytenr, len);
536         if (ret < 0)
537                 return ret;
538
539         if (datacsum)
540                 ret = csum_disk_extent(trans, root, bytenr, len);
541         *ret_len = len;
542         return ret;
543 }
544
545 /*
546  * Relocate old fs data in one reserved ranges
547  *
548  * Since all old fs data in reserved range is not covered by any chunk nor
549  * data extent, we don't need to handle any reference but add new
550  * extent/reference, which makes codes more clear
551  */
552 static int migrate_one_reserved_range(struct btrfs_trans_handle *trans,
553                                       struct btrfs_root *root,
554                                       struct cache_tree *used,
555                                       struct btrfs_inode_item *inode, int fd,
556                                       u64 ino, u64 start, u64 len, int datacsum)
557 {
558         u64 cur_off = start;
559         u64 cur_len = len;
560         u64 hole_start = start;
561         u64 hole_len;
562         struct cache_extent *cache;
563         struct btrfs_key key;
564         struct extent_buffer *eb;
565         int ret = 0;
566
567         while (cur_off < start + len) {
568                 cache = lookup_cache_extent(used, cur_off, cur_len);
569                 if (!cache)
570                         break;
571                 cur_off = max(cache->start, cur_off);
572                 cur_len = min(cache->start + cache->size, start + len) -
573                           cur_off;
574                 BUG_ON(cur_len < root->sectorsize);
575
576                 /* reserve extent for the data */
577                 ret = btrfs_reserve_extent(trans, root, cur_len, 0, 0, (u64)-1,
578                                            &key, 1);
579                 if (ret < 0)
580                         break;
581
582                 eb = malloc(sizeof(*eb) + cur_len);
583                 if (!eb) {
584                         ret = -ENOMEM;
585                         break;
586                 }
587
588                 ret = pread(fd, eb->data, cur_len, cur_off);
589                 if (ret < cur_len) {
590                         ret = (ret < 0 ? ret : -EIO);
591                         free(eb);
592                         break;
593                 }
594                 eb->start = key.objectid;
595                 eb->len = key.offset;
596
597                 /* Write the data */
598                 ret = write_and_map_eb(trans, root, eb);
599                 free(eb);
600                 if (ret < 0)
601                         break;
602
603                 /* Now handle extent item and file extent things */
604                 ret = btrfs_record_file_extent(trans, root, ino, inode, cur_off,
605                                                key.objectid, key.offset);
606                 if (ret < 0)
607                         break;
608                 /* Finally, insert csum items */
609                 if (datacsum)
610                         ret = csum_disk_extent(trans, root, key.objectid,
611                                                key.offset);
612
613                 /* Don't forget to insert hole */
614                 hole_len = cur_off - hole_start;
615                 if (hole_len) {
616                         ret = btrfs_record_file_extent(trans, root, ino, inode,
617                                         hole_start, 0, hole_len);
618                         if (ret < 0)
619                                 break;
620                 }
621
622                 cur_off += key.offset;
623                 hole_start = cur_off;
624                 cur_len = start + len - cur_off;
625         }
626         /* Last hole */
627         if (start + len - hole_start > 0)
628                 ret = btrfs_record_file_extent(trans, root, ino, inode,
629                                 hole_start, 0, start + len - hole_start);
630         return ret;
631 }
632
633 /*
634  * Relocate the used ext2 data in reserved ranges
635  * [0,1M)
636  * [btrfs_sb_offset(1), +BTRFS_STRIPE_LEN)
637  * [btrfs_sb_offset(2), +BTRFS_STRIPE_LEN)
638  */
639 static int migrate_reserved_ranges(struct btrfs_trans_handle *trans,
640                                    struct btrfs_root *root,
641                                    struct cache_tree *used,
642                                    struct btrfs_inode_item *inode, int fd,
643                                    u64 ino, u64 total_bytes, int datacsum)
644 {
645         u64 cur_off;
646         u64 cur_len;
647         int ret = 0;
648
649         /* 0 ~ 1M */
650         cur_off = 0;
651         cur_len = 1024 * 1024;
652         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
653                                          cur_off, cur_len, datacsum);
654         if (ret < 0)
655                 return ret;
656
657         /* second sb(fisrt sb is included in 0~1M) */
658         cur_off = btrfs_sb_offset(1);
659         cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
660         if (cur_off > total_bytes)
661                 return ret;
662         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
663                                          cur_off, cur_len, datacsum);
664         if (ret < 0)
665                 return ret;
666
667         /* Last sb */
668         cur_off = btrfs_sb_offset(2);
669         cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
670         if (cur_off > total_bytes)
671                 return ret;
672         ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
673                                          cur_off, cur_len, datacsum);
674         return ret;
675 }
676
677 /*
678  * Helper for expand and merge extent_cache for wipe_one_reserved_range() to
679  * handle wiping a range that exists in cache.
680  */
681 static int _expand_extent_cache(struct cache_tree *tree,
682                                 struct cache_extent *entry,
683                                 u64 min_stripe_size, int backward)
684 {
685         struct cache_extent *ce;
686         int diff;
687
688         if (entry->size >= min_stripe_size)
689                 return 0;
690         diff = min_stripe_size - entry->size;
691
692         if (backward) {
693                 ce = prev_cache_extent(entry);
694                 if (!ce)
695                         goto expand_back;
696                 if (ce->start + ce->size >= entry->start - diff) {
697                         /* Directly merge with previous extent */
698                         ce->size = entry->start + entry->size - ce->start;
699                         remove_cache_extent(tree, entry);
700                         free(entry);
701                         return 0;
702                 }
703 expand_back:
704                 /* No overlap, normal extent */
705                 if (entry->start < diff) {
706                         error("cannot find space for data chunk layout");
707                         return -ENOSPC;
708                 }
709                 entry->start -= diff;
710                 entry->size += diff;
711                 return 0;
712         }
713         ce = next_cache_extent(entry);
714         if (!ce)
715                 goto expand_after;
716         if (entry->start + entry->size + diff >= ce->start) {
717                 /* Directly merge with next extent */
718                 entry->size = ce->start + ce->size - entry->start;
719                 remove_cache_extent(tree, ce);
720                 free(ce);
721                 return 0;
722         }
723 expand_after:
724         entry->size += diff;
725         return 0;
726 }
727
728 /*
729  * Remove one reserve range from given cache tree
730  * if min_stripe_size is non-zero, it will ensure for split case,
731  * all its split cache extent is no smaller than @min_strip_size / 2.
732  */
733 static int wipe_one_reserved_range(struct cache_tree *tree,
734                                    u64 start, u64 len, u64 min_stripe_size,
735                                    int ensure_size)
736 {
737         struct cache_extent *cache;
738         int ret;
739
740         BUG_ON(ensure_size && min_stripe_size == 0);
741         /*
742          * The logical here is simplified to handle special cases only
743          * So we don't need to consider merge case for ensure_size
744          */
745         BUG_ON(min_stripe_size && (min_stripe_size < len * 2 ||
746                min_stripe_size / 2 < BTRFS_STRIPE_LEN));
747
748         /* Also, wipe range should already be aligned */
749         BUG_ON(start != round_down(start, BTRFS_STRIPE_LEN) ||
750                start + len != round_up(start + len, BTRFS_STRIPE_LEN));
751
752         min_stripe_size /= 2;
753
754         cache = lookup_cache_extent(tree, start, len);
755         if (!cache)
756                 return 0;
757
758         if (start <= cache->start) {
759                 /*
760                  *      |--------cache---------|
761                  * |-wipe-|
762                  */
763                 BUG_ON(start + len <= cache->start);
764
765                 /*
766                  * The wipe size is smaller than min_stripe_size / 2,
767                  * so the result length should still meet min_stripe_size
768                  * And no need to do alignment
769                  */
770                 cache->size -= (start + len - cache->start);
771                 if (cache->size == 0) {
772                         remove_cache_extent(tree, cache);
773                         free(cache);
774                         return 0;
775                 }
776
777                 BUG_ON(ensure_size && cache->size < min_stripe_size);
778
779                 cache->start = start + len;
780                 return 0;
781         } else if (start > cache->start && start + len < cache->start +
782                    cache->size) {
783                 /*
784                  * |-------cache-----|
785                  *      |-wipe-|
786                  */
787                 u64 old_start = cache->start;
788                 u64 old_len = cache->size;
789                 u64 insert_start = start + len;
790                 u64 insert_len;
791
792                 cache->size = start - cache->start;
793                 /* Expand the leading half part if needed */
794                 if (ensure_size && cache->size < min_stripe_size) {
795                         ret = _expand_extent_cache(tree, cache,
796                                         min_stripe_size, 1);
797                         if (ret < 0)
798                                 return ret;
799                 }
800
801                 /* And insert the new one */
802                 insert_len = old_start + old_len - start - len;
803                 ret = add_merge_cache_extent(tree, insert_start, insert_len);
804                 if (ret < 0)
805                         return ret;
806
807                 /* Expand the last half part if needed */
808                 if (ensure_size && insert_len < min_stripe_size) {
809                         cache = lookup_cache_extent(tree, insert_start,
810                                                     insert_len);
811                         if (!cache || cache->start != insert_start ||
812                             cache->size != insert_len)
813                                 return -ENOENT;
814                         ret = _expand_extent_cache(tree, cache,
815                                         min_stripe_size, 0);
816                 }
817
818                 return ret;
819         }
820         /*
821          * |----cache-----|
822          *              |--wipe-|
823          * Wipe len should be small enough and no need to expand the
824          * remaining extent
825          */
826         cache->size = start - cache->start;
827         BUG_ON(ensure_size && cache->size < min_stripe_size);
828         return 0;
829 }
830
831 /*
832  * Remove reserved ranges from given cache_tree
833  *
834  * It will remove the following ranges
835  * 1) 0~1M
836  * 2) 2nd superblock, +64K (make sure chunks are 64K aligned)
837  * 3) 3rd superblock, +64K
838  *
839  * @min_stripe must be given for safety check
840  * and if @ensure_size is given, it will ensure affected cache_extent will be
841  * larger than min_stripe_size
842  */
843 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
844                                 int ensure_size)
845 {
846         int ret;
847
848         ret = wipe_one_reserved_range(tree, 0, 1024 * 1024, min_stripe_size,
849                                       ensure_size);
850         if (ret < 0)
851                 return ret;
852         ret = wipe_one_reserved_range(tree, btrfs_sb_offset(1),
853                         BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
854         if (ret < 0)
855                 return ret;
856         ret = wipe_one_reserved_range(tree, btrfs_sb_offset(2),
857                         BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
858         return ret;
859 }
860
861 static int calculate_available_space(struct btrfs_convert_context *cctx)
862 {
863         struct cache_tree *used = &cctx->used;
864         struct cache_tree *data_chunks = &cctx->data_chunks;
865         struct cache_tree *free = &cctx->free;
866         struct cache_extent *cache;
867         u64 cur_off = 0;
868         /*
869          * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
870          * works without need to consider overlap
871          */
872         u64 min_stripe_size = 2 * 16 * 1024 * 1024;
873         int ret;
874
875         /* Calculate data_chunks */
876         for (cache = first_cache_extent(used); cache;
877              cache = next_cache_extent(cache)) {
878                 u64 cur_len;
879
880                 if (cache->start + cache->size < cur_off)
881                         continue;
882                 if (cache->start > cur_off + min_stripe_size)
883                         cur_off = cache->start;
884                 cur_len = max(cache->start + cache->size - cur_off,
885                               min_stripe_size);
886                 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
887                 if (ret < 0)
888                         goto out;
889                 cur_off += cur_len;
890         }
891         /*
892          * remove reserved ranges, so we won't ever bother relocating an old
893          * filesystem extent to other place.
894          */
895         ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
896         if (ret < 0)
897                 goto out;
898
899         cur_off = 0;
900         /*
901          * Calculate free space
902          * Always round up the start bytenr, to avoid metadata extent corss
903          * stripe boundary, as later mkfs_convert() won't have all the extent
904          * allocation check
905          */
906         for (cache = first_cache_extent(data_chunks); cache;
907              cache = next_cache_extent(cache)) {
908                 if (cache->start < cur_off)
909                         continue;
910                 if (cache->start > cur_off) {
911                         u64 insert_start;
912                         u64 len;
913
914                         len = cache->start - round_up(cur_off,
915                                                       BTRFS_STRIPE_LEN);
916                         insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
917
918                         ret = add_merge_cache_extent(free, insert_start, len);
919                         if (ret < 0)
920                                 goto out;
921                 }
922                 cur_off = cache->start + cache->size;
923         }
924         /* Don't forget the last range */
925         if (cctx->total_bytes > cur_off) {
926                 u64 len = cctx->total_bytes - cur_off;
927                 u64 insert_start;
928
929                 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
930
931                 ret = add_merge_cache_extent(free, insert_start, len);
932                 if (ret < 0)
933                         goto out;
934         }
935
936         /* Remove reserved bytes */
937         ret = wipe_reserved_ranges(free, min_stripe_size, 0);
938 out:
939         return ret;
940 }
941
942 /*
943  * Read used space, and since we have the used space,
944  * calcuate data_chunks and free for later mkfs
945  */
946 static int convert_read_used_space(struct btrfs_convert_context *cctx)
947 {
948         int ret;
949
950         ret = cctx->convert_ops->read_used_space(cctx);
951         if (ret)
952                 return ret;
953
954         ret = calculate_available_space(cctx);
955         return ret;
956 }
957
958 /*
959  * Create the fs image file of old filesystem.
960  *
961  * This is completely fs independent as we have cctx->used, only
962  * need to create file extents pointing to all the positions.
963  */
964 static int create_image(struct btrfs_root *root,
965                            struct btrfs_mkfs_config *cfg,
966                            struct btrfs_convert_context *cctx, int fd,
967                            u64 size, char *name, int datacsum)
968 {
969         struct btrfs_inode_item buf;
970         struct btrfs_trans_handle *trans;
971         struct btrfs_path *path = NULL;
972         struct btrfs_key key;
973         struct cache_extent *cache;
974         struct cache_tree used_tmp;
975         u64 cur;
976         u64 ino;
977         u64 flags = BTRFS_INODE_READONLY;
978         int ret;
979
980         if (!datacsum)
981                 flags |= BTRFS_INODE_NODATASUM;
982
983         trans = btrfs_start_transaction(root, 1);
984         if (!trans)
985                 return -ENOMEM;
986
987         cache_tree_init(&used_tmp);
988
989         ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
990                                        &ino);
991         if (ret < 0)
992                 goto out;
993         ret = btrfs_new_inode(trans, root, ino, 0400 | S_IFREG);
994         if (ret < 0)
995                 goto out;
996         ret = btrfs_change_inode_flags(trans, root, ino, flags);
997         if (ret < 0)
998                 goto out;
999         ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
1000                              strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
1001         if (ret < 0)
1002                 goto out;
1003
1004         path = btrfs_alloc_path();
1005         if (!path) {
1006                 ret = -ENOMEM;
1007                 goto out;
1008         }
1009         key.objectid = ino;
1010         key.type = BTRFS_INODE_ITEM_KEY;
1011         key.offset = 0;
1012
1013         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1014         if (ret) {
1015                 ret = (ret > 0 ? -ENOENT : ret);
1016                 goto out;
1017         }
1018         read_extent_buffer(path->nodes[0], &buf,
1019                         btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1020                         sizeof(buf));
1021         btrfs_release_path(path);
1022
1023         /*
1024          * Create a new used space cache, which doesn't contain the reserved
1025          * range
1026          */
1027         for (cache = first_cache_extent(&cctx->used); cache;
1028              cache = next_cache_extent(cache)) {
1029                 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1030                 if (ret < 0)
1031                         goto out;
1032         }
1033         ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1034         if (ret < 0)
1035                 goto out;
1036
1037         /*
1038          * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1039          * can't handle bytenr 0(will consider it as a hole)
1040          */
1041         cur = 1024 * 1024;
1042         while (cur < size) {
1043                 u64 len = size - cur;
1044
1045                 ret = create_image_file_range(trans, root, &used_tmp,
1046                                                 &buf, ino, cur, &len, datacsum);
1047                 if (ret < 0)
1048                         goto out;
1049                 cur += len;
1050         }
1051         /* Handle the reserved ranges */
1052         ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1053                                       cfg->num_bytes, datacsum);
1054
1055
1056         key.objectid = ino;
1057         key.type = BTRFS_INODE_ITEM_KEY;
1058         key.offset = 0;
1059         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1060         if (ret) {
1061                 ret = (ret > 0 ? -ENOENT : ret);
1062                 goto out;
1063         }
1064         btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1065         write_extent_buffer(path->nodes[0], &buf,
1066                         btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1067                         sizeof(buf));
1068 out:
1069         free_extent_cache_tree(&used_tmp);
1070         btrfs_free_path(path);
1071         btrfs_commit_transaction(trans, root);
1072         return ret;
1073 }
1074
1075 static struct btrfs_root* link_subvol(struct btrfs_root *root,
1076                 const char *base, u64 root_objectid)
1077 {
1078         struct btrfs_trans_handle *trans;
1079         struct btrfs_fs_info *fs_info = root->fs_info;
1080         struct btrfs_root *tree_root = fs_info->tree_root;
1081         struct btrfs_root *new_root = NULL;
1082         struct btrfs_path *path;
1083         struct btrfs_inode_item *inode_item;
1084         struct extent_buffer *leaf;
1085         struct btrfs_key key;
1086         u64 dirid = btrfs_root_dirid(&root->root_item);
1087         u64 index = 2;
1088         char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1089         int len;
1090         int i;
1091         int ret;
1092
1093         len = strlen(base);
1094         if (len == 0 || len > BTRFS_NAME_LEN)
1095                 return NULL;
1096
1097         path = btrfs_alloc_path();
1098         if (!path)
1099                 return NULL;
1100
1101         key.objectid = dirid;
1102         key.type = BTRFS_DIR_INDEX_KEY;
1103         key.offset = (u64)-1;
1104
1105         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1106         if (ret <= 0) {
1107                 error("search for DIR_INDEX dirid %llu failed: %d",
1108                                 (unsigned long long)dirid, ret);
1109                 goto fail;
1110         }
1111
1112         if (path->slots[0] > 0) {
1113                 path->slots[0]--;
1114                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1115                 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1116                         index = key.offset + 1;
1117         }
1118         btrfs_release_path(path);
1119
1120         trans = btrfs_start_transaction(root, 1);
1121         if (!trans) {
1122                 error("unable to start transaction");
1123                 goto fail;
1124         }
1125
1126         key.objectid = dirid;
1127         key.offset = 0;
1128         key.type =  BTRFS_INODE_ITEM_KEY;
1129
1130         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1131         if (ret) {
1132                 error("search for INODE_ITEM %llu failed: %d",
1133                                 (unsigned long long)dirid, ret);
1134                 goto fail;
1135         }
1136         leaf = path->nodes[0];
1137         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1138                                     struct btrfs_inode_item);
1139
1140         key.objectid = root_objectid;
1141         key.offset = (u64)-1;
1142         key.type = BTRFS_ROOT_ITEM_KEY;
1143
1144         memcpy(buf, base, len);
1145         for (i = 0; i < 1024; i++) {
1146                 ret = btrfs_insert_dir_item(trans, root, buf, len,
1147                                             dirid, &key, BTRFS_FT_DIR, index);
1148                 if (ret != -EEXIST)
1149                         break;
1150                 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1151                 if (len < 1 || len > BTRFS_NAME_LEN) {
1152                         ret = -EINVAL;
1153                         break;
1154                 }
1155         }
1156         if (ret)
1157                 goto fail;
1158
1159         btrfs_set_inode_size(leaf, inode_item, len * 2 +
1160                              btrfs_inode_size(leaf, inode_item));
1161         btrfs_mark_buffer_dirty(leaf);
1162         btrfs_release_path(path);
1163
1164         /* add the backref first */
1165         ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1166                                  BTRFS_ROOT_BACKREF_KEY,
1167                                  root->root_key.objectid,
1168                                  dirid, index, buf, len);
1169         if (ret) {
1170                 error("unable to add root backref for %llu: %d",
1171                                 root->root_key.objectid, ret);
1172                 goto fail;
1173         }
1174
1175         /* now add the forward ref */
1176         ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1177                                  BTRFS_ROOT_REF_KEY, root_objectid,
1178                                  dirid, index, buf, len);
1179         if (ret) {
1180                 error("unable to add root ref for %llu: %d",
1181                                 root->root_key.objectid, ret);
1182                 goto fail;
1183         }
1184
1185         ret = btrfs_commit_transaction(trans, root);
1186         if (ret) {
1187                 error("transaction commit failed: %d", ret);
1188                 goto fail;
1189         }
1190
1191         new_root = btrfs_read_fs_root(fs_info, &key);
1192         if (IS_ERR(new_root)) {
1193                 error("unable to fs read root: %lu", PTR_ERR(new_root));
1194                 new_root = NULL;
1195         }
1196 fail:
1197         btrfs_free_path(path);
1198         return new_root;
1199 }
1200
1201 static int create_subvol(struct btrfs_trans_handle *trans,
1202                          struct btrfs_root *root, u64 root_objectid)
1203 {
1204         struct extent_buffer *tmp;
1205         struct btrfs_root *new_root;
1206         struct btrfs_key key;
1207         struct btrfs_root_item root_item;
1208         int ret;
1209
1210         ret = btrfs_copy_root(trans, root, root->node, &tmp,
1211                               root_objectid);
1212         if (ret)
1213                 return ret;
1214
1215         memcpy(&root_item, &root->root_item, sizeof(root_item));
1216         btrfs_set_root_bytenr(&root_item, tmp->start);
1217         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
1218         btrfs_set_root_generation(&root_item, trans->transid);
1219         free_extent_buffer(tmp);
1220
1221         key.objectid = root_objectid;
1222         key.type = BTRFS_ROOT_ITEM_KEY;
1223         key.offset = trans->transid;
1224         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1225                                 &key, &root_item);
1226
1227         key.offset = (u64)-1;
1228         new_root = btrfs_read_fs_root(root->fs_info, &key);
1229         if (!new_root || IS_ERR(new_root)) {
1230                 error("unable to fs read root: %lu", PTR_ERR(new_root));
1231                 return PTR_ERR(new_root);
1232         }
1233
1234         ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
1235
1236         return ret;
1237 }
1238
1239 /*
1240  * New make_btrfs() has handle system and meta chunks quite well.
1241  * So only need to add remaining data chunks.
1242  */
1243 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
1244                                           struct btrfs_fs_info *fs_info,
1245                                           struct btrfs_mkfs_config *cfg,
1246                                           struct btrfs_convert_context *cctx)
1247 {
1248         struct btrfs_root *extent_root = fs_info->extent_root;
1249         struct cache_tree *data_chunks = &cctx->data_chunks;
1250         struct cache_extent *cache;
1251         u64 max_chunk_size;
1252         int ret = 0;
1253
1254         /*
1255          * Don't create data chunk over 10% of the convert device
1256          * And for single chunk, don't create chunk larger than 1G.
1257          */
1258         max_chunk_size = cfg->num_bytes / 10;
1259         max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
1260         max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
1261
1262         for (cache = first_cache_extent(data_chunks); cache;
1263              cache = next_cache_extent(cache)) {
1264                 u64 cur = cache->start;
1265
1266                 while (cur < cache->start + cache->size) {
1267                         u64 len;
1268                         u64 cur_backup = cur;
1269
1270                         len = min(max_chunk_size,
1271                                   cache->start + cache->size - cur);
1272                         ret = btrfs_alloc_data_chunk(trans, extent_root,
1273                                         &cur_backup, len,
1274                                         BTRFS_BLOCK_GROUP_DATA, 1);
1275                         if (ret < 0)
1276                                 break;
1277                         ret = btrfs_make_block_group(trans, extent_root, 0,
1278                                         BTRFS_BLOCK_GROUP_DATA,
1279                                         BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1280                                         cur, len);
1281                         if (ret < 0)
1282                                 break;
1283                         cur += len;
1284                 }
1285         }
1286         return ret;
1287 }
1288
1289 /*
1290  * Init the temp btrfs to a operational status.
1291  *
1292  * It will fix the extent usage accounting(XXX: Do we really need?) and
1293  * insert needed data chunks, to ensure all old fs data extents are covered
1294  * by DATA chunks, preventing wrong chunks are allocated.
1295  *
1296  * And also create convert image subvolume and relocation tree.
1297  * (XXX: Not need again?)
1298  * But the convert image subvolume is *NOT* linked to fs tree yet.
1299  */
1300 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
1301                          struct btrfs_convert_context *cctx, int datacsum,
1302                          int packing, int noxattr)
1303 {
1304         struct btrfs_key location;
1305         struct btrfs_trans_handle *trans;
1306         struct btrfs_fs_info *fs_info = root->fs_info;
1307         int ret;
1308
1309         /*
1310          * Don't alloc any metadata/system chunk, as we don't want
1311          * any meta/sys chunk allcated before all data chunks are inserted.
1312          * Or we screw up the chunk layout just like the old implement.
1313          */
1314         fs_info->avoid_sys_chunk_alloc = 1;
1315         fs_info->avoid_meta_chunk_alloc = 1;
1316         trans = btrfs_start_transaction(root, 1);
1317         if (!trans) {
1318                 error("unable to start transaction");
1319                 ret = -EINVAL;
1320                 goto err;
1321         }
1322         ret = btrfs_fix_block_accounting(trans, root);
1323         if (ret)
1324                 goto err;
1325         ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
1326         if (ret)
1327                 goto err;
1328         ret = btrfs_make_root_dir(trans, fs_info->tree_root,
1329                                   BTRFS_ROOT_TREE_DIR_OBJECTID);
1330         if (ret)
1331                 goto err;
1332         memcpy(&location, &root->root_key, sizeof(location));
1333         location.offset = (u64)-1;
1334         ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
1335                                 btrfs_super_root_dir(fs_info->super_copy),
1336                                 &location, BTRFS_FT_DIR, 0);
1337         if (ret)
1338                 goto err;
1339         ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
1340                                 location.objectid,
1341                                 btrfs_super_root_dir(fs_info->super_copy), 0);
1342         if (ret)
1343                 goto err;
1344         btrfs_set_root_dirid(&fs_info->fs_root->root_item,
1345                              BTRFS_FIRST_FREE_OBJECTID);
1346
1347         /* subvol for fs image file */
1348         ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
1349         if (ret < 0) {
1350                 error("failed to create subvolume image root: %d", ret);
1351                 goto err;
1352         }
1353         /* subvol for data relocation tree */
1354         ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1355         if (ret < 0) {
1356                 error("failed to create DATA_RELOC root: %d", ret);
1357                 goto err;
1358         }
1359
1360         ret = btrfs_commit_transaction(trans, root);
1361         fs_info->avoid_sys_chunk_alloc = 0;
1362         fs_info->avoid_meta_chunk_alloc = 0;
1363 err:
1364         return ret;
1365 }
1366
1367 /*
1368  * Migrate super block to its default position and zero 0 ~ 16k
1369  */
1370 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
1371 {
1372         int ret;
1373         struct extent_buffer *buf;
1374         struct btrfs_super_block *super;
1375         u32 len;
1376         u32 bytenr;
1377
1378         buf = malloc(sizeof(*buf) + sectorsize);
1379         if (!buf)
1380                 return -ENOMEM;
1381
1382         buf->len = sectorsize;
1383         ret = pread(fd, buf->data, sectorsize, old_bytenr);
1384         if (ret != sectorsize)
1385                 goto fail;
1386
1387         super = (struct btrfs_super_block *)buf->data;
1388         BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
1389         btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
1390
1391         csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1392         ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
1393         if (ret != sectorsize)
1394                 goto fail;
1395
1396         ret = fsync(fd);
1397         if (ret)
1398                 goto fail;
1399
1400         memset(buf->data, 0, sectorsize);
1401         for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
1402                 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
1403                 if (len > sectorsize)
1404                         len = sectorsize;
1405                 ret = pwrite(fd, buf->data, len, bytenr);
1406                 if (ret != len) {
1407                         fprintf(stderr, "unable to zero fill device\n");
1408                         break;
1409                 }
1410                 bytenr += len;
1411         }
1412         ret = 0;
1413         fsync(fd);
1414 fail:
1415         free(buf);
1416         if (ret > 0)
1417                 ret = -1;
1418         return ret;
1419 }
1420
1421 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
1422 {
1423         struct btrfs_chunk *chunk;
1424         struct btrfs_disk_key *key;
1425         u32 sectorsize = btrfs_super_sectorsize(super);
1426
1427         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1428         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1429                                        sizeof(struct btrfs_disk_key));
1430
1431         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1432         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1433         btrfs_set_disk_key_offset(key, 0);
1434
1435         btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
1436         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1437         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1438         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1439         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1440         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1441         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1442         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1443         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1444         chunk->stripe.devid = super->dev_item.devid;
1445         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1446         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1447         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1448         return 0;
1449 }
1450
1451 #if BTRFSCONVERT_EXT2
1452
1453 /*
1454  * Open Ext2fs in readonly mode, read block allocation bitmap and
1455  * inode bitmap into memory.
1456  */
1457 static int ext2_open_fs(struct btrfs_convert_context *cctx, const char *name)
1458 {
1459         errcode_t ret;
1460         ext2_filsys ext2_fs;
1461         ext2_ino_t ino;
1462         u32 ro_feature;
1463
1464         ret = ext2fs_open(name, 0, 0, 0, unix_io_manager, &ext2_fs);
1465         if (ret) {
1466                 fprintf(stderr, "ext2fs_open: %s\n", error_message(ret));
1467                 return -1;
1468         }
1469         /*
1470          * We need to know exactly the used space, some RO compat flags like
1471          * BIGALLOC will affect how used space is present.
1472          * So we need manuall check any unsupported RO compat flags
1473          */
1474         ro_feature = ext2_fs->super->s_feature_ro_compat;
1475         if (ro_feature & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
1476                 error(
1477 "unsupported RO features detected: %x, abort convert to avoid possible corruption",
1478                       ro_feature & ~EXT2_LIB_FEATURE_COMPAT_SUPP);
1479                 goto fail;
1480         }
1481         ret = ext2fs_read_inode_bitmap(ext2_fs);
1482         if (ret) {
1483                 fprintf(stderr, "ext2fs_read_inode_bitmap: %s\n",
1484                         error_message(ret));
1485                 goto fail;
1486         }
1487         ret = ext2fs_read_block_bitmap(ext2_fs);
1488         if (ret) {
1489                 fprintf(stderr, "ext2fs_read_block_bitmap: %s\n",
1490                         error_message(ret));
1491                 goto fail;
1492         }
1493         /*
1494          * search each block group for a free inode. this set up
1495          * uninit block/inode bitmaps appropriately.
1496          */
1497         ino = 1;
1498         while (ino <= ext2_fs->super->s_inodes_count) {
1499                 ext2_ino_t foo;
1500                 ext2fs_new_inode(ext2_fs, ino, 0, NULL, &foo);
1501                 ino += EXT2_INODES_PER_GROUP(ext2_fs->super);
1502         }
1503
1504         if (!(ext2_fs->super->s_feature_incompat &
1505               EXT2_FEATURE_INCOMPAT_FILETYPE)) {
1506                 error("filetype feature is missing");
1507                 goto fail;
1508         }
1509
1510         cctx->fs_data = ext2_fs;
1511         cctx->blocksize = ext2_fs->blocksize;
1512         cctx->block_count = ext2_fs->super->s_blocks_count;
1513         cctx->total_bytes = ext2_fs->blocksize * ext2_fs->super->s_blocks_count;
1514         cctx->volume_name = strndup(ext2_fs->super->s_volume_name, 16);
1515         cctx->first_data_block = ext2_fs->super->s_first_data_block;
1516         cctx->inodes_count = ext2_fs->super->s_inodes_count;
1517         cctx->free_inodes_count = ext2_fs->super->s_free_inodes_count;
1518         return 0;
1519 fail:
1520         ext2fs_close(ext2_fs);
1521         return -1;
1522 }
1523
1524 static int __ext2_add_one_block(ext2_filsys fs, char *bitmap,
1525                                 unsigned long group_nr, struct cache_tree *used)
1526 {
1527         unsigned long offset;
1528         unsigned i;
1529         int ret = 0;
1530
1531         offset = fs->super->s_first_data_block;
1532         offset /= EXT2FS_CLUSTER_RATIO(fs);
1533         offset += group_nr * EXT2_CLUSTERS_PER_GROUP(fs->super);
1534         for (i = 0; i < EXT2_CLUSTERS_PER_GROUP(fs->super); i++) {
1535                 if (ext2fs_test_bit(i, bitmap)) {
1536                         u64 start;
1537
1538                         start = (i + offset) * EXT2FS_CLUSTER_RATIO(fs);
1539                         start *= fs->blocksize;
1540                         ret = add_merge_cache_extent(used, start,
1541                                                      fs->blocksize);
1542                         if (ret < 0)
1543                                 break;
1544                 }
1545         }
1546         return ret;
1547 }
1548
1549 /*
1550  * Read all used ext2 space into cctx->used cache tree
1551  */
1552 static int ext2_read_used_space(struct btrfs_convert_context *cctx)
1553 {
1554         ext2_filsys fs = (ext2_filsys)cctx->fs_data;
1555         blk64_t blk_itr = EXT2FS_B2C(fs, fs->super->s_first_data_block);
1556         struct cache_tree *used_tree = &cctx->used;
1557         char *block_bitmap = NULL;
1558         unsigned long i;
1559         int block_nbytes;
1560         int ret = 0;
1561
1562         block_nbytes = EXT2_CLUSTERS_PER_GROUP(fs->super) / 8;
1563         /* Shouldn't happen */
1564         BUG_ON(!fs->block_map);
1565
1566         block_bitmap = malloc(block_nbytes);
1567         if (!block_bitmap)
1568                 return -ENOMEM;
1569
1570         for (i = 0; i < fs->group_desc_count; i++) {
1571                 ret = ext2fs_get_block_bitmap_range(fs->block_map, blk_itr,
1572                                                 block_nbytes * 8, block_bitmap);
1573                 if (ret) {
1574                         error("fail to get bitmap from ext2, %s",
1575                               strerror(-ret));
1576                         break;
1577                 }
1578                 ret = __ext2_add_one_block(fs, block_bitmap, i, used_tree);
1579                 if (ret < 0) {
1580                         error("fail to build used space tree, %s",
1581                               strerror(-ret));
1582                         break;
1583                 }
1584                 blk_itr += EXT2_CLUSTERS_PER_GROUP(fs->super);
1585         }
1586
1587         free(block_bitmap);
1588         return ret;
1589 }
1590
1591 static void ext2_close_fs(struct btrfs_convert_context *cctx)
1592 {
1593         if (cctx->volume_name) {
1594                 free(cctx->volume_name);
1595                 cctx->volume_name = NULL;
1596         }
1597         ext2fs_close(cctx->fs_data);
1598 }
1599
1600 struct dir_iterate_data {
1601         struct btrfs_trans_handle *trans;
1602         struct btrfs_root *root;
1603         struct btrfs_inode_item *inode;
1604         u64 objectid;
1605         u64 index_cnt;
1606         u64 parent;
1607         int errcode;
1608 };
1609
1610 static u8 ext2_filetype_conversion_table[EXT2_FT_MAX] = {
1611         [EXT2_FT_UNKNOWN]       = BTRFS_FT_UNKNOWN,
1612         [EXT2_FT_REG_FILE]      = BTRFS_FT_REG_FILE,
1613         [EXT2_FT_DIR]           = BTRFS_FT_DIR,
1614         [EXT2_FT_CHRDEV]        = BTRFS_FT_CHRDEV,
1615         [EXT2_FT_BLKDEV]        = BTRFS_FT_BLKDEV,
1616         [EXT2_FT_FIFO]          = BTRFS_FT_FIFO,
1617         [EXT2_FT_SOCK]          = BTRFS_FT_SOCK,
1618         [EXT2_FT_SYMLINK]       = BTRFS_FT_SYMLINK,
1619 };
1620
1621 static int ext2_dir_iterate_proc(ext2_ino_t dir, int entry,
1622                             struct ext2_dir_entry *dirent,
1623                             int offset, int blocksize,
1624                             char *buf,void *priv_data)
1625 {
1626         int ret;
1627         int file_type;
1628         u64 objectid;
1629         char dotdot[] = "..";
1630         struct dir_iterate_data *idata = (struct dir_iterate_data *)priv_data;
1631         int name_len;
1632
1633         name_len = dirent->name_len & 0xFF;
1634
1635         objectid = dirent->inode + INO_OFFSET;
1636         if (!strncmp(dirent->name, dotdot, name_len)) {
1637                 if (name_len == 2) {
1638                         BUG_ON(idata->parent != 0);
1639                         idata->parent = objectid;
1640                 }
1641                 return 0;
1642         }
1643         if (dirent->inode < EXT2_GOOD_OLD_FIRST_INO)
1644                 return 0;
1645
1646         file_type = dirent->name_len >> 8;
1647         BUG_ON(file_type > EXT2_FT_SYMLINK);
1648
1649         ret = convert_insert_dirent(idata->trans, idata->root, dirent->name,
1650                                     name_len, idata->objectid, objectid,
1651                                     ext2_filetype_conversion_table[file_type],
1652                                     idata->index_cnt, idata->inode);
1653         if (ret < 0) {
1654                 idata->errcode = ret;
1655                 return BLOCK_ABORT;
1656         }
1657
1658         idata->index_cnt++;
1659         return 0;
1660 }
1661
1662 static int ext2_create_dir_entries(struct btrfs_trans_handle *trans,
1663                               struct btrfs_root *root, u64 objectid,
1664                               struct btrfs_inode_item *btrfs_inode,
1665                               ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
1666 {
1667         int ret;
1668         errcode_t err;
1669         struct dir_iterate_data data = {
1670                 .trans          = trans,
1671                 .root           = root,
1672                 .inode          = btrfs_inode,
1673                 .objectid       = objectid,
1674                 .index_cnt      = 2,
1675                 .parent         = 0,
1676                 .errcode        = 0,
1677         };
1678
1679         err = ext2fs_dir_iterate2(ext2_fs, ext2_ino, 0, NULL,
1680                                   ext2_dir_iterate_proc, &data);
1681         if (err)
1682                 goto error;
1683         ret = data.errcode;
1684         if (ret == 0 && data.parent == objectid) {
1685                 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
1686                                              objectid, objectid, 0);
1687         }
1688         return ret;
1689 error:
1690         fprintf(stderr, "ext2fs_dir_iterate2: %s\n", error_message(err));
1691         return -1;
1692 }
1693
1694 static int ext2_block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
1695                                 e2_blkcnt_t blockcnt, blk_t ref_block,
1696                                 int ref_offset, void *priv_data)
1697 {
1698         int ret;
1699         struct blk_iterate_data *idata;
1700         idata = (struct blk_iterate_data *)priv_data;
1701         ret = block_iterate_proc(*blocknr, blockcnt, idata);
1702         if (ret) {
1703                 idata->errcode = ret;
1704                 return BLOCK_ABORT;
1705         }
1706         return 0;
1707 }
1708
1709 /*
1710  * traverse file's data blocks, record these data blocks as file extents.
1711  */
1712 static int ext2_create_file_extents(struct btrfs_trans_handle *trans,
1713                                struct btrfs_root *root, u64 objectid,
1714                                struct btrfs_inode_item *btrfs_inode,
1715                                ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1716                                int datacsum, int packing)
1717 {
1718         int ret;
1719         char *buffer = NULL;
1720         errcode_t err;
1721         u32 last_block;
1722         u32 sectorsize = root->sectorsize;
1723         u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1724         struct blk_iterate_data data;
1725
1726         init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
1727                               datacsum);
1728
1729         err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
1730                                     NULL, ext2_block_iterate_proc, &data);
1731         if (err)
1732                 goto error;
1733         ret = data.errcode;
1734         if (ret)
1735                 goto fail;
1736         if (packing && data.first_block == 0 && data.num_blocks > 0 &&
1737             inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
1738                 u64 num_bytes = data.num_blocks * sectorsize;
1739                 u64 disk_bytenr = data.disk_block * sectorsize;
1740                 u64 nbytes;
1741
1742                 buffer = malloc(num_bytes);
1743                 if (!buffer)
1744                         return -ENOMEM;
1745                 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
1746                 if (ret)
1747                         goto fail;
1748                 if (num_bytes > inode_size)
1749                         num_bytes = inode_size;
1750                 ret = btrfs_insert_inline_extent(trans, root, objectid,
1751                                                  0, buffer, num_bytes);
1752                 if (ret)
1753                         goto fail;
1754                 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
1755                 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
1756         } else if (data.num_blocks > 0) {
1757                 ret = record_file_blocks(&data, data.first_block,
1758                                          data.disk_block, data.num_blocks);
1759                 if (ret)
1760                         goto fail;
1761         }
1762         data.first_block += data.num_blocks;
1763         last_block = (inode_size + sectorsize - 1) / sectorsize;
1764         if (last_block > data.first_block) {
1765                 ret = record_file_blocks(&data, data.first_block, 0,
1766                                          last_block - data.first_block);
1767         }
1768 fail:
1769         free(buffer);
1770         return ret;
1771 error:
1772         fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
1773         return -1;
1774 }
1775
1776 static int ext2_create_symbol_link(struct btrfs_trans_handle *trans,
1777                               struct btrfs_root *root, u64 objectid,
1778                               struct btrfs_inode_item *btrfs_inode,
1779                               ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1780                               struct ext2_inode *ext2_inode)
1781 {
1782         int ret;
1783         char *pathname;
1784         u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1785         if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
1786                 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
1787                 ret = ext2_create_file_extents(trans, root, objectid,
1788                                 btrfs_inode, ext2_fs, ext2_ino, 1, 1);
1789                 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
1790                 return ret;
1791         }
1792
1793         pathname = (char *)&(ext2_inode->i_block[0]);
1794         BUG_ON(pathname[inode_size] != 0);
1795         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
1796                                          pathname, inode_size + 1);
1797         btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
1798         return ret;
1799 }
1800
1801 /*
1802  * Following xattr/acl related codes are based on codes in
1803  * fs/ext3/xattr.c and fs/ext3/acl.c
1804  */
1805 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
1806 #define EXT2_XATTR_BFIRST(ptr) \
1807         ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
1808 #define EXT2_XATTR_IHDR(inode) \
1809         ((struct ext2_ext_attr_header *) ((void *)(inode) + \
1810                 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
1811 #define EXT2_XATTR_IFIRST(inode) \
1812         ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
1813                 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
1814
1815 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
1816                                   const void *end)
1817 {
1818         struct ext2_ext_attr_entry *next;
1819
1820         while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1821                 next = EXT2_EXT_ATTR_NEXT(entry);
1822                 if ((void *)next >= end)
1823                         return -EIO;
1824                 entry = next;
1825         }
1826         return 0;
1827 }
1828
1829 static int ext2_xattr_check_block(const char *buf, size_t size)
1830 {
1831         int error;
1832         struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
1833
1834         if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
1835             header->h_blocks != 1)
1836                 return -EIO;
1837         error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
1838         return error;
1839 }
1840
1841 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
1842                                   size_t size)
1843 {
1844         size_t value_size = entry->e_value_size;
1845
1846         if (entry->e_value_block != 0 || value_size > size ||
1847             entry->e_value_offs + value_size > size)
1848                 return -EIO;
1849         return 0;
1850 }
1851
1852 #define EXT2_ACL_VERSION        0x0001
1853
1854 /* 23.2.5 acl_tag_t values */
1855
1856 #define ACL_UNDEFINED_TAG       (0x00)
1857 #define ACL_USER_OBJ            (0x01)
1858 #define ACL_USER                (0x02)
1859 #define ACL_GROUP_OBJ           (0x04)
1860 #define ACL_GROUP               (0x08)
1861 #define ACL_MASK                (0x10)
1862 #define ACL_OTHER               (0x20)
1863
1864 /* 23.2.7 ACL qualifier constants */
1865
1866 #define ACL_UNDEFINED_ID        ((id_t)-1)
1867
1868 typedef struct {
1869         __le16          e_tag;
1870         __le16          e_perm;
1871         __le32          e_id;
1872 } ext2_acl_entry;
1873
1874 typedef struct {
1875         __le16          e_tag;
1876         __le16          e_perm;
1877 } ext2_acl_entry_short;
1878
1879 typedef struct {
1880         __le32          a_version;
1881 } ext2_acl_header;
1882
1883 static inline int ext2_acl_count(size_t size)
1884 {
1885         ssize_t s;
1886         size -= sizeof(ext2_acl_header);
1887         s = size - 4 * sizeof(ext2_acl_entry_short);
1888         if (s < 0) {
1889                 if (size % sizeof(ext2_acl_entry_short))
1890                         return -1;
1891                 return size / sizeof(ext2_acl_entry_short);
1892         } else {
1893                 if (s % sizeof(ext2_acl_entry))
1894                         return -1;
1895                 return s / sizeof(ext2_acl_entry) + 4;
1896         }
1897 }
1898
1899 #define ACL_EA_VERSION          0x0002
1900
1901 typedef struct {
1902         __le16          e_tag;
1903         __le16          e_perm;
1904         __le32          e_id;
1905 } acl_ea_entry;
1906
1907 typedef struct {
1908         __le32          a_version;
1909         acl_ea_entry    a_entries[0];
1910 } acl_ea_header;
1911
1912 static inline size_t acl_ea_size(int count)
1913 {
1914         return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
1915 }
1916
1917 static int ext2_acl_to_xattr(void *dst, const void *src,
1918                              size_t dst_size, size_t src_size)
1919 {
1920         int i, count;
1921         const void *end = src + src_size;
1922         acl_ea_header *ext_acl = (acl_ea_header *)dst;
1923         acl_ea_entry *dst_entry = ext_acl->a_entries;
1924         ext2_acl_entry *src_entry;
1925
1926         if (src_size < sizeof(ext2_acl_header))
1927                 goto fail;
1928         if (((ext2_acl_header *)src)->a_version !=
1929             cpu_to_le32(EXT2_ACL_VERSION))
1930                 goto fail;
1931         src += sizeof(ext2_acl_header);
1932         count = ext2_acl_count(src_size);
1933         if (count <= 0)
1934                 goto fail;
1935
1936         BUG_ON(dst_size < acl_ea_size(count));
1937         ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
1938         for (i = 0; i < count; i++, dst_entry++) {
1939                 src_entry = (ext2_acl_entry *)src;
1940                 if (src + sizeof(ext2_acl_entry_short) > end)
1941                         goto fail;
1942                 dst_entry->e_tag = src_entry->e_tag;
1943                 dst_entry->e_perm = src_entry->e_perm;
1944                 switch (le16_to_cpu(src_entry->e_tag)) {
1945                 case ACL_USER_OBJ:
1946                 case ACL_GROUP_OBJ:
1947                 case ACL_MASK:
1948                 case ACL_OTHER:
1949                         src += sizeof(ext2_acl_entry_short);
1950                         dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
1951                         break;
1952                 case ACL_USER:
1953                 case ACL_GROUP:
1954                         src += sizeof(ext2_acl_entry);
1955                         if (src > end)
1956                                 goto fail;
1957                         dst_entry->e_id = src_entry->e_id;
1958                         break;
1959                 default:
1960                         goto fail;
1961                 }
1962         }
1963         if (src != end)
1964                 goto fail;
1965         return 0;
1966 fail:
1967         return -EINVAL;
1968 }
1969
1970 static char *xattr_prefix_table[] = {
1971         [1] =   "user.",
1972         [2] =   "system.posix_acl_access",
1973         [3] =   "system.posix_acl_default",
1974         [4] =   "trusted.",
1975         [6] =   "security.",
1976 };
1977
1978 static int ext2_copy_single_xattr(struct btrfs_trans_handle *trans,
1979                              struct btrfs_root *root, u64 objectid,
1980                              struct ext2_ext_attr_entry *entry,
1981                              const void *data, u32 datalen)
1982 {
1983         int ret = 0;
1984         int name_len;
1985         int name_index;
1986         void *databuf = NULL;
1987         char namebuf[XATTR_NAME_MAX + 1];
1988
1989         name_index = entry->e_name_index;
1990         if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
1991             xattr_prefix_table[name_index] == NULL)
1992                 return -EOPNOTSUPP;
1993         name_len = strlen(xattr_prefix_table[name_index]) +
1994                    entry->e_name_len;
1995         if (name_len >= sizeof(namebuf))
1996                 return -ERANGE;
1997
1998         if (name_index == 2 || name_index == 3) {
1999                 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
2000                 databuf = malloc(bufsize);
2001                 if (!databuf)
2002                        return -ENOMEM;
2003                 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
2004                 if (ret)
2005                         goto out;
2006                 data = databuf;
2007                 datalen = bufsize;
2008         }
2009         strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
2010         strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
2011         if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
2012             sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
2013                 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
2014                         objectid - INO_OFFSET, name_len, namebuf);
2015                 goto out;
2016         }
2017         ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
2018                                       data, datalen, objectid);
2019 out:
2020         free(databuf);
2021         return ret;
2022 }
2023
2024 static int ext2_copy_extended_attrs(struct btrfs_trans_handle *trans,
2025                                struct btrfs_root *root, u64 objectid,
2026                                struct btrfs_inode_item *btrfs_inode,
2027                                ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
2028 {
2029         int ret = 0;
2030         int inline_ea = 0;
2031         errcode_t err;
2032         u32 datalen;
2033         u32 block_size = ext2_fs->blocksize;
2034         u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
2035         struct ext2_inode_large *ext2_inode;
2036         struct ext2_ext_attr_entry *entry;
2037         void *data;
2038         char *buffer = NULL;
2039         char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
2040
2041         if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
2042                 ext2_inode = (struct ext2_inode_large *)inode_buf;
2043         } else {
2044                 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
2045                 if (!ext2_inode)
2046                        return -ENOMEM;
2047         }
2048         err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
2049                                      inode_size);
2050         if (err) {
2051                 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
2052                         error_message(err));
2053                 ret = -1;
2054                 goto out;
2055         }
2056
2057         if (ext2_ino > ext2_fs->super->s_first_ino &&
2058             inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
2059                 if (EXT2_GOOD_OLD_INODE_SIZE +
2060                     ext2_inode->i_extra_isize > inode_size) {
2061                         ret = -EIO;
2062                         goto out;
2063                 }
2064                 if (ext2_inode->i_extra_isize != 0 &&
2065                     EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
2066                     EXT2_EXT_ATTR_MAGIC) {
2067                         inline_ea = 1;
2068                 }
2069         }
2070         if (inline_ea) {
2071                 int total;
2072                 void *end = (void *)ext2_inode + inode_size;
2073                 entry = EXT2_XATTR_IFIRST(ext2_inode);
2074                 total = end - (void *)entry;
2075                 ret = ext2_xattr_check_names(entry, end);
2076                 if (ret)
2077                         goto out;
2078                 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
2079                         ret = ext2_xattr_check_entry(entry, total);
2080                         if (ret)
2081                                 goto out;
2082                         data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
2083                                 entry->e_value_offs;
2084                         datalen = entry->e_value_size;
2085                         ret = ext2_copy_single_xattr(trans, root, objectid,
2086                                                 entry, data, datalen);
2087                         if (ret)
2088                                 goto out;
2089                         entry = EXT2_EXT_ATTR_NEXT(entry);
2090                 }
2091         }
2092
2093         if (ext2_inode->i_file_acl == 0)
2094                 goto out;
2095
2096         buffer = malloc(block_size);
2097         if (!buffer) {
2098                 ret = -ENOMEM;
2099                 goto out;
2100         }
2101         err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
2102         if (err) {
2103                 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
2104                         error_message(err));
2105                 ret = -1;
2106                 goto out;
2107         }
2108         ret = ext2_xattr_check_block(buffer, block_size);
2109         if (ret)
2110                 goto out;
2111
2112         entry = EXT2_XATTR_BFIRST(buffer);
2113         while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
2114                 ret = ext2_xattr_check_entry(entry, block_size);
2115                 if (ret)
2116                         goto out;
2117                 data = buffer + entry->e_value_offs;
2118                 datalen = entry->e_value_size;
2119                 ret = ext2_copy_single_xattr(trans, root, objectid,
2120                                         entry, data, datalen);
2121                 if (ret)
2122                         goto out;
2123                 entry = EXT2_EXT_ATTR_NEXT(entry);
2124         }
2125 out:
2126         free(buffer);
2127         if ((void *)ext2_inode != inode_buf)
2128                 free(ext2_inode);
2129         return ret;
2130 }
2131 #define MINORBITS       20
2132 #define MKDEV(ma, mi)   (((ma) << MINORBITS) | (mi))
2133
2134 static inline dev_t old_decode_dev(u16 val)
2135 {
2136         return MKDEV((val >> 8) & 255, val & 255);
2137 }
2138
2139 static inline dev_t new_decode_dev(u32 dev)
2140 {
2141         unsigned major = (dev & 0xfff00) >> 8;
2142         unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
2143         return MKDEV(major, minor);
2144 }
2145
2146 static void ext2_copy_inode_item(struct btrfs_inode_item *dst,
2147                            struct ext2_inode *src, u32 blocksize)
2148 {
2149         btrfs_set_stack_inode_generation(dst, 1);
2150         btrfs_set_stack_inode_sequence(dst, 0);
2151         btrfs_set_stack_inode_transid(dst, 1);
2152         btrfs_set_stack_inode_size(dst, src->i_size);
2153         btrfs_set_stack_inode_nbytes(dst, 0);
2154         btrfs_set_stack_inode_block_group(dst, 0);
2155         btrfs_set_stack_inode_nlink(dst, src->i_links_count);
2156         btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
2157         btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
2158         btrfs_set_stack_inode_mode(dst, src->i_mode);
2159         btrfs_set_stack_inode_rdev(dst, 0);
2160         btrfs_set_stack_inode_flags(dst, 0);
2161         btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
2162         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
2163         btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
2164         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
2165         btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
2166         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
2167         btrfs_set_stack_timespec_sec(&dst->otime, 0);
2168         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
2169
2170         if (S_ISDIR(src->i_mode)) {
2171                 btrfs_set_stack_inode_size(dst, 0);
2172                 btrfs_set_stack_inode_nlink(dst, 1);
2173         }
2174         if (S_ISREG(src->i_mode)) {
2175                 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
2176                                            (u64)src->i_size);
2177         }
2178         if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
2179             !S_ISLNK(src->i_mode)) {
2180                 if (src->i_block[0]) {
2181                         btrfs_set_stack_inode_rdev(dst,
2182                                 old_decode_dev(src->i_block[0]));
2183                 } else {
2184                         btrfs_set_stack_inode_rdev(dst,
2185                                 new_decode_dev(src->i_block[1]));
2186                 }
2187         }
2188         memset(&dst->reserved, 0, sizeof(dst->reserved));
2189 }
2190 static int ext2_check_state(struct btrfs_convert_context *cctx)
2191 {
2192         ext2_filsys fs = cctx->fs_data;
2193
2194         if (!(fs->super->s_state & EXT2_VALID_FS))
2195                 return 1;
2196         else if (fs->super->s_state & EXT2_ERROR_FS)
2197                 return 1;
2198         else
2199                 return 0;
2200 }
2201
2202 /* EXT2_*_FL to BTRFS_INODE_FLAG_* stringification helper */
2203 #define COPY_ONE_EXT2_FLAG(flags, ext2_inode, name) ({                  \
2204         if (ext2_inode->i_flags & EXT2_##name##_FL)                     \
2205                 flags |= BTRFS_INODE_##name;                            \
2206 })
2207
2208 /*
2209  * Convert EXT2_*_FL to corresponding BTRFS_INODE_* flags
2210  *
2211  * Only a subset of EXT_*_FL is supported in btrfs.
2212  */
2213 static void ext2_convert_inode_flags(struct btrfs_inode_item *dst,
2214                                      struct ext2_inode *src)
2215 {
2216         u64 flags = 0;
2217
2218         COPY_ONE_EXT2_FLAG(flags, src, APPEND);
2219         COPY_ONE_EXT2_FLAG(flags, src, SYNC);
2220         COPY_ONE_EXT2_FLAG(flags, src, IMMUTABLE);
2221         COPY_ONE_EXT2_FLAG(flags, src, NODUMP);
2222         COPY_ONE_EXT2_FLAG(flags, src, NOATIME);
2223         COPY_ONE_EXT2_FLAG(flags, src, DIRSYNC);
2224         btrfs_set_stack_inode_flags(dst, flags);
2225 }
2226
2227 /*
2228  * copy a single inode. do all the required works, such as cloning
2229  * inode item, creating file extents and creating directory entries.
2230  */
2231 static int ext2_copy_single_inode(struct btrfs_trans_handle *trans,
2232                              struct btrfs_root *root, u64 objectid,
2233                              ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
2234                              struct ext2_inode *ext2_inode,
2235                              int datacsum, int packing, int noxattr)
2236 {
2237         int ret;
2238         struct btrfs_inode_item btrfs_inode;
2239
2240         if (ext2_inode->i_links_count == 0)
2241                 return 0;
2242
2243         ext2_copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
2244         if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
2245                 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
2246                             BTRFS_INODE_NODATASUM;
2247                 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
2248         }
2249         ext2_convert_inode_flags(&btrfs_inode, ext2_inode);
2250
2251         switch (ext2_inode->i_mode & S_IFMT) {
2252         case S_IFREG:
2253                 ret = ext2_create_file_extents(trans, root, objectid,
2254                         &btrfs_inode, ext2_fs, ext2_ino, datacsum, packing);
2255                 break;
2256         case S_IFDIR:
2257                 ret = ext2_create_dir_entries(trans, root, objectid,
2258                                 &btrfs_inode, ext2_fs, ext2_ino);
2259                 break;
2260         case S_IFLNK:
2261                 ret = ext2_create_symbol_link(trans, root, objectid,
2262                                 &btrfs_inode, ext2_fs, ext2_ino, ext2_inode);
2263                 break;
2264         default:
2265                 ret = 0;
2266                 break;
2267         }
2268         if (ret)
2269                 return ret;
2270
2271         if (!noxattr) {
2272                 ret = ext2_copy_extended_attrs(trans, root, objectid,
2273                                 &btrfs_inode, ext2_fs, ext2_ino);
2274                 if (ret)
2275                         return ret;
2276         }
2277         return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
2278 }
2279
2280 /*
2281  * scan ext2's inode bitmap and copy all used inodes.
2282  */
2283 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
2284                             struct btrfs_root *root,
2285                             int datacsum, int packing, int noxattr, struct task_ctx *p)
2286 {
2287         ext2_filsys ext2_fs = cctx->fs_data;
2288         int ret;
2289         errcode_t err;
2290         ext2_inode_scan ext2_scan;
2291         struct ext2_inode ext2_inode;
2292         ext2_ino_t ext2_ino;
2293         u64 objectid;
2294         struct btrfs_trans_handle *trans;
2295
2296         trans = btrfs_start_transaction(root, 1);
2297         if (!trans)
2298                 return -ENOMEM;
2299         err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
2300         if (err) {
2301                 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
2302                 return -1;
2303         }
2304         while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
2305                                              &ext2_inode))) {
2306                 /* no more inodes */
2307                 if (ext2_ino == 0)
2308                         break;
2309                 /* skip special inode in ext2fs */
2310                 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
2311                     ext2_ino != EXT2_ROOT_INO)
2312                         continue;
2313                 objectid = ext2_ino + INO_OFFSET;
2314                 ret = ext2_copy_single_inode(trans, root,
2315                                         objectid, ext2_fs, ext2_ino,
2316                                         &ext2_inode, datacsum, packing,
2317                                         noxattr);
2318                 p->cur_copy_inodes++;
2319                 if (ret)
2320                         return ret;
2321                 if (trans->blocks_used >= 4096) {
2322                         ret = btrfs_commit_transaction(trans, root);
2323                         BUG_ON(ret);
2324                         trans = btrfs_start_transaction(root, 1);
2325                         BUG_ON(!trans);
2326                 }
2327         }
2328         if (err) {
2329                 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
2330                 return -1;
2331         }
2332         ret = btrfs_commit_transaction(trans, root);
2333         BUG_ON(ret);
2334         ext2fs_close_inode_scan(ext2_scan);
2335
2336         return ret;
2337 }
2338
2339 static const struct btrfs_convert_operations ext2_convert_ops = {
2340         .name                   = "ext2",
2341         .open_fs                = ext2_open_fs,
2342         .read_used_space        = ext2_read_used_space,
2343         .copy_inodes            = ext2_copy_inodes,
2344         .close_fs               = ext2_close_fs,
2345         .check_state            = ext2_check_state,
2346 };
2347
2348 #endif
2349
2350 static const struct btrfs_convert_operations *convert_operations[] = {
2351 #if BTRFSCONVERT_EXT2
2352         &ext2_convert_ops,
2353 #endif
2354 };
2355
2356 static int convert_open_fs(const char *devname,
2357                            struct btrfs_convert_context *cctx)
2358 {
2359         int i;
2360
2361         memset(cctx, 0, sizeof(*cctx));
2362
2363         for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
2364                 int ret = convert_operations[i]->open_fs(cctx, devname);
2365
2366                 if (ret == 0) {
2367                         cctx->convert_ops = convert_operations[i];
2368                         return ret;
2369                 }
2370         }
2371
2372         error("no file system found to convert");
2373         return -1;
2374 }
2375
2376 static int do_convert(const char *devname, int datacsum, int packing,
2377                 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2378                 int progress, u64 features)
2379 {
2380         int ret;
2381         int fd = -1;
2382         u32 blocksize;
2383         u64 total_bytes;
2384         struct btrfs_root *root;
2385         struct btrfs_root *image_root;
2386         struct btrfs_convert_context cctx;
2387         struct btrfs_key key;
2388         char *subvol_name = NULL;
2389         struct task_ctx ctx;
2390         char features_buf[64];
2391         struct btrfs_mkfs_config mkfs_cfg;
2392
2393         init_convert_context(&cctx);
2394         ret = convert_open_fs(devname, &cctx);
2395         if (ret)
2396                 goto fail;
2397         ret = convert_check_state(&cctx);
2398         if (ret)
2399                 warning(
2400                 "source filesystem is not clean, running filesystem check is recommended");
2401         ret = convert_read_used_space(&cctx);
2402         if (ret)
2403                 goto fail;
2404
2405         blocksize = cctx.blocksize;
2406         total_bytes = (u64)blocksize * (u64)cctx.block_count;
2407         if (blocksize < 4096) {
2408                 error("block size is too small: %u < 4096", blocksize);
2409                 goto fail;
2410         }
2411         if (btrfs_check_nodesize(nodesize, blocksize, features))
2412                 goto fail;
2413         fd = open(devname, O_RDWR);
2414         if (fd < 0) {
2415                 error("unable to open %s: %s", devname, strerror(errno));
2416                 goto fail;
2417         }
2418         btrfs_parse_features_to_string(features_buf, features);
2419         if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2420                 strcat(features_buf, " (default)");
2421
2422         printf("create btrfs filesystem:\n");
2423         printf("\tblocksize: %u\n", blocksize);
2424         printf("\tnodesize:  %u\n", nodesize);
2425         printf("\tfeatures:  %s\n", features_buf);
2426
2427         mkfs_cfg.label = cctx.volume_name;
2428         mkfs_cfg.num_bytes = total_bytes;
2429         mkfs_cfg.nodesize = nodesize;
2430         mkfs_cfg.sectorsize = blocksize;
2431         mkfs_cfg.stripesize = blocksize;
2432         mkfs_cfg.features = features;
2433         /* New convert need these space */
2434         memset(mkfs_cfg.chunk_uuid, 0, BTRFS_UUID_UNPARSED_SIZE);
2435         memset(mkfs_cfg.fs_uuid, 0, BTRFS_UUID_UNPARSED_SIZE);
2436
2437         ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2438         if (ret) {
2439                 error("unable to create initial ctree: %s", strerror(-ret));
2440                 goto fail;
2441         }
2442
2443         root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2444                              OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2445         if (!root) {
2446                 error("unable to open ctree");
2447                 goto fail;
2448         }
2449         ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2450         if (ret) {
2451                 error("unable to setup the root tree: %d", ret);
2452                 goto fail;
2453         }
2454
2455         printf("creating %s image file\n", cctx.convert_ops->name);
2456         ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2457         if (ret < 0) {
2458                 error("memory allocation failure for subvolume name: %s_saved",
2459                         cctx.convert_ops->name);
2460                 goto fail;
2461         }
2462         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2463         key.offset = (u64)-1;
2464         key.type = BTRFS_ROOT_ITEM_KEY;
2465         image_root = btrfs_read_fs_root(root->fs_info, &key);
2466         if (!image_root) {
2467                 error("unable to create image subvolume");
2468                 goto fail;
2469         }
2470         ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2471                               mkfs_cfg.num_bytes, "image", datacsum);
2472         if (ret) {
2473                 error("failed to create %s/image: %d", subvol_name, ret);
2474                 goto fail;
2475         }
2476
2477         printf("creating btrfs metadata");
2478         ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2479         ctx.cur_copy_inodes = 0;
2480
2481         if (progress) {
2482                 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2483                                      &ctx);
2484                 task_start(ctx.info);
2485         }
2486         ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2487         if (ret) {
2488                 error("error during copy_inodes %d", ret);
2489                 goto fail;
2490         }
2491         if (progress) {
2492                 task_stop(ctx.info);
2493                 task_deinit(ctx.info);
2494         }
2495
2496         image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2497         if (!image_root) {
2498                 error("unable to link subvolume %s", subvol_name);
2499                 goto fail;
2500         }
2501
2502         free(subvol_name);
2503
2504         memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2505         if (copylabel == 1) {
2506                 __strncpy_null(root->fs_info->super_copy->label,
2507                                 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2508                 printf("copy label '%s'\n", root->fs_info->super_copy->label);
2509         } else if (copylabel == -1) {
2510                 strcpy(root->fs_info->super_copy->label, fslabel);
2511                 printf("set label to '%s'\n", fslabel);
2512         }
2513
2514         ret = close_ctree(root);
2515         if (ret) {
2516                 error("close_ctree failed: %d", ret);
2517                 goto fail;
2518         }
2519         convert_close_fs(&cctx);
2520         clean_convert_context(&cctx);
2521
2522         /*
2523          * If this step succeed, we get a mountable btrfs. Otherwise
2524          * the source fs is left unchanged.
2525          */
2526         ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2527         if (ret) {
2528                 error("unable to migrate super block: %d", ret);
2529                 goto fail;
2530         }
2531
2532         root = open_ctree_fd(fd, devname, 0,
2533                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2534         if (!root) {
2535                 error("unable to open ctree for finalization");
2536                 goto fail;
2537         }
2538         root->fs_info->finalize_on_close = 1;
2539         close_ctree(root);
2540         close(fd);
2541
2542         printf("conversion complete");
2543         return 0;
2544 fail:
2545         clean_convert_context(&cctx);
2546         if (fd != -1)
2547                 close(fd);
2548         warning(
2549 "an error occurred during conversion, filesystem is partially created but not finalized and not mountable");
2550         return -1;
2551 }
2552
2553 /*
2554  * Check if a non 1:1 mapped chunk can be rolled back.
2555  * For new convert, it's OK while for old convert it's not.
2556  */
2557 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2558 {
2559         struct btrfs_block_group_cache *bg;
2560         struct btrfs_key key;
2561         struct btrfs_path path;
2562         struct btrfs_root *extent_root = fs_info->extent_root;
2563         u64 bg_start;
2564         u64 bg_end;
2565         int ret;
2566
2567         bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2568         if (!bg)
2569                 return -ENOENT;
2570         bg_start = bg->key.objectid;
2571         bg_end = bg->key.objectid + bg->key.offset;
2572
2573         key.objectid = bg_end;
2574         key.type = BTRFS_METADATA_ITEM_KEY;
2575         key.offset = 0;
2576         btrfs_init_path(&path);
2577
2578         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2579         if (ret < 0)
2580                 return ret;
2581
2582         while (1) {
2583                 struct btrfs_extent_item *ei;
2584
2585                 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2586                 if (ret > 0) {
2587                         ret = 0;
2588                         break;
2589                 }
2590                 if (ret < 0)
2591                         break;
2592
2593                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2594                 if (key.type == BTRFS_METADATA_ITEM_KEY)
2595                         continue;
2596                 /* Now it's EXTENT_ITEM_KEY only */
2597                 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2598                                     struct btrfs_extent_item);
2599                 /*
2600                  * Found data extent, means this is old convert must follow 1:1
2601                  * mapping.
2602                  */
2603                 if (btrfs_extent_flags(path.nodes[0], ei)
2604                                 & BTRFS_EXTENT_FLAG_DATA) {
2605                         ret = -EINVAL;
2606                         break;
2607                 }
2608         }
2609         btrfs_release_path(&path);
2610         return ret;
2611 }
2612
2613 static int may_rollback(struct btrfs_root *root)
2614 {
2615         struct btrfs_fs_info *info = root->fs_info;
2616         struct btrfs_multi_bio *multi = NULL;
2617         u64 bytenr;
2618         u64 length;
2619         u64 physical;
2620         u64 total_bytes;
2621         int num_stripes;
2622         int ret;
2623
2624         if (btrfs_super_num_devices(info->super_copy) != 1)
2625                 goto fail;
2626
2627         bytenr = BTRFS_SUPER_INFO_OFFSET;
2628         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2629
2630         while (1) {
2631                 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2632                                       &length, &multi, 0, NULL);
2633                 if (ret) {
2634                         if (ret == -ENOENT) {
2635                                 /* removed block group at the tail */
2636                                 if (length == (u64)-1)
2637                                         break;
2638
2639                                 /* removed block group in the middle */
2640                                 goto next;
2641                         }
2642                         goto fail;
2643                 }
2644
2645                 num_stripes = multi->num_stripes;
2646                 physical = multi->stripes[0].physical;
2647                 free(multi);
2648
2649                 if (num_stripes != 1) {
2650                         error("num stripes for bytenr %llu is not 1", bytenr);
2651                         goto fail;
2652                 }
2653
2654                 /*
2655                  * Extra check for new convert, as metadata chunk from new
2656                  * convert is much more free than old convert, it doesn't need
2657                  * to do 1:1 mapping.
2658                  */
2659                 if (physical != bytenr) {
2660                         /*
2661                          * Check if it's a metadata chunk and has only metadata
2662                          * extent.
2663                          */
2664                         ret = may_rollback_chunk(info, bytenr);
2665                         if (ret < 0)
2666                                 goto fail;
2667                 }
2668 next:
2669                 bytenr += length;
2670                 if (bytenr >= total_bytes)
2671                         break;
2672         }
2673         return 0;
2674 fail:
2675         return -1;
2676 }
2677
2678 static int do_rollback(const char *devname)
2679 {
2680         int fd = -1;
2681         int ret;
2682         int i;
2683         struct btrfs_root *root;
2684         struct btrfs_root *image_root;
2685         struct btrfs_root *chunk_root;
2686         struct btrfs_dir_item *dir;
2687         struct btrfs_inode_item *inode;
2688         struct btrfs_file_extent_item *fi;
2689         struct btrfs_trans_handle *trans;
2690         struct extent_buffer *leaf;
2691         struct btrfs_block_group_cache *cache1;
2692         struct btrfs_block_group_cache *cache2;
2693         struct btrfs_key key;
2694         struct btrfs_path path;
2695         struct extent_io_tree io_tree;
2696         char *buf = NULL;
2697         char *name;
2698         u64 bytenr;
2699         u64 num_bytes;
2700         u64 root_dir;
2701         u64 objectid;
2702         u64 offset;
2703         u64 start;
2704         u64 end;
2705         u64 sb_bytenr;
2706         u64 first_free;
2707         u64 total_bytes;
2708         u32 sectorsize;
2709
2710         extent_io_tree_init(&io_tree);
2711
2712         fd = open(devname, O_RDWR);
2713         if (fd < 0) {
2714                 error("unable to open %s: %s", devname, strerror(errno));
2715                 goto fail;
2716         }
2717         root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2718         if (!root) {
2719                 error("unable to open ctree");
2720                 goto fail;
2721         }
2722         ret = may_rollback(root);
2723         if (ret < 0) {
2724                 error("unable to do rollback: %d", ret);
2725                 goto fail;
2726         }
2727
2728         sectorsize = root->sectorsize;
2729         buf = malloc(sectorsize);
2730         if (!buf) {
2731                 error("unable to allocate memory");
2732                 goto fail;
2733         }
2734
2735         btrfs_init_path(&path);
2736
2737         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2738         key.type = BTRFS_ROOT_BACKREF_KEY;
2739         key.offset = BTRFS_FS_TREE_OBJECTID;
2740         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2741                                 0);
2742         btrfs_release_path(&path);
2743         if (ret > 0) {
2744                 error("unable to convert ext2 image subvolume, is it deleted?");
2745                 goto fail;
2746         } else if (ret < 0) {
2747                 error("unable to open ext2_saved, id %llu: %s",
2748                         (unsigned long long)key.objectid, strerror(-ret));
2749                 goto fail;
2750         }
2751
2752         key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2753         key.type = BTRFS_ROOT_ITEM_KEY;
2754         key.offset = (u64)-1;
2755         image_root = btrfs_read_fs_root(root->fs_info, &key);
2756         if (!image_root || IS_ERR(image_root)) {
2757                 error("unable to open subvolume %llu: %ld",
2758                         (unsigned long long)key.objectid, PTR_ERR(image_root));
2759                 goto fail;
2760         }
2761
2762         name = "image";
2763         root_dir = btrfs_root_dirid(&root->root_item);
2764         dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2765                                    root_dir, name, strlen(name), 0);
2766         if (!dir || IS_ERR(dir)) {
2767                 error("unable to find file %s: %ld", name, PTR_ERR(dir));
2768                 goto fail;
2769         }
2770         leaf = path.nodes[0];
2771         btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2772         btrfs_release_path(&path);
2773
2774         objectid = key.objectid;
2775
2776         ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2777         if (ret) {
2778                 error("unable to find inode item: %d", ret);
2779                 goto fail;
2780         }
2781         leaf = path.nodes[0];
2782         inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2783         total_bytes = btrfs_inode_size(leaf, inode);
2784         btrfs_release_path(&path);
2785
2786         key.objectid = objectid;
2787         key.offset = 0;
2788         key.type = BTRFS_EXTENT_DATA_KEY;
2789         ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2790         if (ret != 0) {
2791                 error("unable to find first file extent: %d", ret);
2792                 btrfs_release_path(&path);
2793                 goto fail;
2794         }
2795
2796         /* build mapping tree for the relocated blocks */
2797         for (offset = 0; offset < total_bytes; ) {
2798                 leaf = path.nodes[0];
2799                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2800                         ret = btrfs_next_leaf(root, &path);
2801                         if (ret != 0)
2802                                 break;  
2803                         continue;
2804                 }
2805
2806                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2807                 if (key.objectid != objectid || key.offset != offset ||
2808                     key.type != BTRFS_EXTENT_DATA_KEY)
2809                         break;
2810
2811                 fi = btrfs_item_ptr(leaf, path.slots[0],
2812                                     struct btrfs_file_extent_item);
2813                 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2814                         break;
2815                 if (btrfs_file_extent_compression(leaf, fi) ||
2816                     btrfs_file_extent_encryption(leaf, fi) ||
2817                     btrfs_file_extent_other_encoding(leaf, fi))
2818                         break;
2819
2820                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2821                 /* skip holes and direct mapped extents */
2822                 if (bytenr == 0 || bytenr == offset)
2823                         goto next_extent;
2824
2825                 bytenr += btrfs_file_extent_offset(leaf, fi);
2826                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2827
2828                 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2829                 cache2 = btrfs_lookup_block_group(root->fs_info,
2830                                                   offset + num_bytes - 1);
2831                 /*
2832                  * Here we must take consideration of old and new convert
2833                  * behavior.
2834                  * For old convert case, sign, there is no consist chunk type
2835                  * that will cover the extent. META/DATA/SYS are all possible.
2836                  * Just ensure relocate one is in SYS chunk.
2837                  * For new convert case, they are all covered by DATA chunk.
2838                  *
2839                  * So, there is not valid chunk type check for it now.
2840                  */
2841                 if (cache1 != cache2)
2842                         break;
2843
2844                 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2845                                 EXTENT_LOCKED, GFP_NOFS);
2846                 set_state_private(&io_tree, offset, bytenr);
2847 next_extent:
2848                 offset += btrfs_file_extent_num_bytes(leaf, fi);
2849                 path.slots[0]++;
2850         }
2851         btrfs_release_path(&path);
2852
2853         if (offset < total_bytes) {
2854                 error("unable to build extent mapping (offset %llu, total_bytes %llu)",
2855                                 (unsigned long long)offset,
2856                                 (unsigned long long)total_bytes);
2857                 error("converted filesystem after balance is unable to rollback");
2858                 goto fail;
2859         }
2860
2861         first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2862         first_free &= ~((u64)sectorsize - 1);
2863         /* backup for extent #0 should exist */
2864         if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2865                 error("no backup for the first extent");
2866                 goto fail;
2867         }
2868         /* force no allocation from system block group */
2869         root->fs_info->system_allocs = -1;
2870         trans = btrfs_start_transaction(root, 1);
2871         if (!trans) {
2872                 error("unable to start transaction");
2873                 goto fail;
2874         }
2875         /*
2876          * recow the whole chunk tree, this will remove all chunk tree blocks
2877          * from system block group
2878          */
2879         chunk_root = root->fs_info->chunk_root;
2880         memset(&key, 0, sizeof(key));
2881         while (1) {
2882                 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2883                 if (ret < 0)
2884                         break;
2885
2886                 ret = btrfs_next_leaf(chunk_root, &path);
2887                 if (ret)
2888                         break;
2889
2890                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2891                 btrfs_release_path(&path);
2892         }
2893         btrfs_release_path(&path);
2894
2895         offset = 0;
2896         num_bytes = 0;
2897         while(1) {
2898                 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2899                 if (!cache1)
2900                         break;
2901
2902                 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2903                         num_bytes += btrfs_block_group_used(&cache1->item);
2904
2905                 offset = cache1->key.objectid + cache1->key.offset;
2906         }
2907         /* only extent #0 left in system block group? */
2908         if (num_bytes > first_free) {
2909                 error(
2910         "unable to empty system block group (num_bytes %llu, first_free %llu",
2911                                 (unsigned long long)num_bytes,
2912                                 (unsigned long long)first_free);
2913                 goto fail;
2914         }
2915         /* create a system chunk that maps the whole device */
2916         ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2917         if (ret) {
2918                 error("unable to update system chunk: %d", ret);
2919                 goto fail;
2920         }
2921
2922         ret = btrfs_commit_transaction(trans, root);
2923         if (ret) {
2924                 error("transaction commit failed: %d", ret);
2925                 goto fail;
2926         }
2927
2928         ret = close_ctree(root);
2929         if (ret) {
2930                 error("close_ctree failed: %d", ret);
2931                 goto fail;
2932         }
2933
2934         /* zero btrfs super block mirrors */
2935         memset(buf, 0, sectorsize);
2936         for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2937                 bytenr = btrfs_sb_offset(i);
2938                 if (bytenr >= total_bytes)
2939                         break;
2940                 ret = pwrite(fd, buf, sectorsize, bytenr);
2941                 if (ret != sectorsize) {
2942                         error("zeroing superblock mirror %d failed: %d",
2943                                         i, ret);
2944                         goto fail;
2945                 }
2946         }
2947
2948         sb_bytenr = (u64)-1;
2949         /* copy all relocated blocks back */
2950         while(1) {
2951                 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2952                                             EXTENT_LOCKED);
2953                 if (ret)
2954                         break;
2955
2956                 ret = get_state_private(&io_tree, start, &bytenr);
2957                 BUG_ON(ret);
2958
2959                 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2960                                   GFP_NOFS);
2961
2962                 while (start <= end) {
2963                         if (start == BTRFS_SUPER_INFO_OFFSET) {
2964                                 sb_bytenr = bytenr;
2965                                 goto next_sector;
2966                         }
2967                         ret = pread(fd, buf, sectorsize, bytenr);
2968                         if (ret < 0) {
2969                                 error("reading superblock at %llu failed: %d",
2970                                                 (unsigned long long)bytenr, ret);
2971                                 goto fail;
2972                         }
2973                         BUG_ON(ret != sectorsize);
2974                         ret = pwrite(fd, buf, sectorsize, start);
2975                         if (ret < 0) {
2976                                 error("writing superblock at %llu failed: %d",
2977                                                 (unsigned long long)start, ret);
2978                                 goto fail;
2979                         }
2980                         BUG_ON(ret != sectorsize);
2981 next_sector:
2982                         start += sectorsize;
2983                         bytenr += sectorsize;
2984                 }
2985         }
2986
2987         ret = fsync(fd);
2988         if (ret < 0) {
2989                 error("fsync failed: %s", strerror(errno));
2990                 goto fail;
2991         }
2992         /*
2993          * finally, overwrite btrfs super block.
2994          */
2995         ret = pread(fd, buf, sectorsize, sb_bytenr);
2996         if (ret < 0) {
2997                 error("reading primary superblock failed: %s",
2998                                 strerror(errno));
2999                 goto fail;
3000         }
3001         BUG_ON(ret != sectorsize);
3002         ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
3003         if (ret < 0) {
3004                 error("writing primary superblock failed: %s",
3005                                 strerror(errno));
3006                 goto fail;
3007         }
3008         BUG_ON(ret != sectorsize);
3009         ret = fsync(fd);
3010         if (ret < 0) {
3011                 error("fsync failed: %s", strerror(errno));
3012                 goto fail;
3013         }
3014
3015         close(fd);
3016         free(buf);
3017         extent_io_tree_cleanup(&io_tree);
3018         printf("rollback complete\n");
3019         return 0;
3020
3021 fail:
3022         if (fd != -1)
3023                 close(fd);
3024         free(buf);
3025         error("rollback aborted");
3026         return -1;
3027 }
3028
3029 static void print_usage(void)
3030 {
3031         printf("usage: btrfs-convert [options] device\n");
3032         printf("options:\n");
3033         printf("\t-d|--no-datasum        disable data checksum, sets NODATASUM\n");
3034         printf("\t-i|--no-xattr          ignore xattrs and ACLs\n");
3035         printf("\t-n|--no-inline         disable inlining of small files to metadata\n");
3036         printf("\t-N|--nodesize SIZE     set filesystem metadata nodesize\n");
3037         printf("\t-r|--rollback          roll back to the original filesystem\n");
3038         printf("\t-l|--label LABEL       set filesystem label\n");
3039         printf("\t-L|--copy-label        use label from converted filesystem\n");
3040         printf("\t-p|--progress          show converting progress (default)\n");
3041         printf("\t-O|--features LIST     comma separated list of filesystem features\n");
3042         printf("\t--no-progress          show only overview, not the detailed progress\n");
3043         printf("\n");
3044         printf("Supported filesystems:\n");
3045         printf("\text2/3/4: %s\n", BTRFSCONVERT_EXT2 ? "yes" : "no");
3046 }
3047
3048 int main(int argc, char *argv[])
3049 {
3050         int ret;
3051         int packing = 1;
3052         int noxattr = 0;
3053         int datacsum = 1;
3054         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
3055                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
3056         int rollback = 0;
3057         int copylabel = 0;
3058         int usage_error = 0;
3059         int progress = 1;
3060         char *file;
3061         char fslabel[BTRFS_LABEL_SIZE];
3062         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
3063
3064         while(1) {
3065                 enum { GETOPT_VAL_NO_PROGRESS = 256 };
3066                 static const struct option long_options[] = {
3067                         { "no-progress", no_argument, NULL,
3068                                 GETOPT_VAL_NO_PROGRESS },
3069                         { "no-datasum", no_argument, NULL, 'd' },
3070                         { "no-inline", no_argument, NULL, 'n' },
3071                         { "no-xattr", no_argument, NULL, 'i' },
3072                         { "rollback", no_argument, NULL, 'r' },
3073                         { "features", required_argument, NULL, 'O' },
3074                         { "progress", no_argument, NULL, 'p' },
3075                         { "label", required_argument, NULL, 'l' },
3076                         { "copy-label", no_argument, NULL, 'L' },
3077                         { "nodesize", required_argument, NULL, 'N' },
3078                         { "help", no_argument, NULL, GETOPT_VAL_HELP},
3079                         { NULL, 0, NULL, 0 }
3080                 };
3081                 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
3082
3083                 if (c < 0)
3084                         break;
3085                 switch(c) {
3086                         case 'd':
3087                                 datacsum = 0;
3088                                 break;
3089                         case 'i':
3090                                 noxattr = 1;
3091                                 break;
3092                         case 'n':
3093                                 packing = 0;
3094                                 break;
3095                         case 'N':
3096                                 nodesize = parse_size(optarg);
3097                                 break;
3098                         case 'r':
3099                                 rollback = 1;
3100                                 break;
3101                         case 'l':
3102                                 copylabel = -1;
3103                                 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
3104                                         warning(
3105                                         "label too long, trimmed to %d bytes",
3106                                                 BTRFS_LABEL_SIZE - 1);
3107                                 }
3108                                 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
3109                                 break;
3110                         case 'L':
3111                                 copylabel = 1;
3112                                 break;
3113                         case 'p':
3114                                 progress = 1;
3115                                 break;
3116                         case 'O': {
3117                                 char *orig = strdup(optarg);
3118                                 char *tmp = orig;
3119
3120                                 tmp = btrfs_parse_fs_features(tmp, &features);
3121                                 if (tmp) {
3122                                         error("unrecognized filesystem feature: %s",
3123                                                         tmp);
3124                                         free(orig);
3125                                         exit(1);
3126                                 }
3127                                 free(orig);
3128                                 if (features & BTRFS_FEATURE_LIST_ALL) {
3129                                         btrfs_list_all_fs_features(
3130                                                 ~BTRFS_CONVERT_ALLOWED_FEATURES);
3131                                         exit(0);
3132                                 }
3133                                 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
3134                                         char buf[64];
3135
3136                                         btrfs_parse_features_to_string(buf,
3137                                                 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
3138                                         error("features not allowed for convert: %s",
3139                                                 buf);
3140                                         exit(1);
3141                                 }
3142
3143                                 break;
3144                                 }
3145                         case GETOPT_VAL_NO_PROGRESS:
3146                                 progress = 0;
3147                                 break;
3148                         case GETOPT_VAL_HELP:
3149                         default:
3150                                 print_usage();
3151                                 return c != GETOPT_VAL_HELP;
3152                 }
3153         }
3154         set_argv0(argv);
3155         if (check_argc_exact(argc - optind, 1)) {
3156                 print_usage();
3157                 return 1;
3158         }
3159
3160         if (rollback && (!datacsum || noxattr || !packing)) {
3161                 fprintf(stderr,
3162                         "Usage error: -d, -i, -n options do not apply to rollback\n");
3163                 usage_error++;
3164         }
3165
3166         if (usage_error) {
3167                 print_usage();
3168                 return 1;
3169         }
3170
3171         file = argv[optind];
3172         ret = check_mounted(file);
3173         if (ret < 0) {
3174                 error("could not check mount status: %s", strerror(-ret));
3175                 return 1;
3176         } else if (ret) {
3177                 error("%s is mounted", file);
3178                 return 1;
3179         }
3180
3181         if (rollback) {
3182                 ret = do_rollback(file);
3183         } else {
3184                 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3185                                 copylabel, fslabel, progress, features);
3186         }
3187         if (ret)
3188                 return 1;
3189         return 0;
3190 }