2 * Copyright (C) 2007 Oracle. All rights reserved.
3 * Copyright (C) 2008 Morey Roof. All rights reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
23 #include <sys/ioctl.h>
24 #include <sys/mount.h>
25 #include <sys/types.h>
27 #include <uuid/uuid.h>
32 #include <linux/loop.h>
33 #include <linux/major.h>
34 #include <linux/kdev_t.h>
36 #include <blkid/blkid.h>
38 #include <sys/statfs.h>
39 #include <linux/magic.h>
41 #include <sys/utsname.h>
42 #include <linux/version.h>
44 #include "kerncompat.h"
45 #include "radix-tree.h"
48 #include "transaction.h"
56 #define BLKDISCARD _IO(0x12,119)
59 static int btrfs_scan_done = 0;
61 static char argv0_buf[ARGV0_BUF_SIZE] = "btrfs";
63 static int rand_seed_initlized = 0;
64 static unsigned short rand_seed[3];
66 const char *get_argv0_buf(void)
71 void fixup_argv0(char **argv, const char *token)
73 int len = strlen(argv0_buf);
75 snprintf(argv0_buf + len, sizeof(argv0_buf) - len, " %s", token);
79 void set_argv0(char **argv)
81 strncpy(argv0_buf, argv[0], sizeof(argv0_buf));
82 argv0_buf[sizeof(argv0_buf) - 1] = 0;
85 int check_argc_exact(int nargs, int expected)
88 fprintf(stderr, "%s: too few arguments\n", argv0_buf);
90 fprintf(stderr, "%s: too many arguments\n", argv0_buf);
92 return nargs != expected;
95 int check_argc_min(int nargs, int expected)
97 if (nargs < expected) {
98 fprintf(stderr, "%s: too few arguments\n", argv0_buf);
105 int check_argc_max(int nargs, int expected)
107 if (nargs > expected) {
108 fprintf(stderr, "%s: too many arguments\n", argv0_buf);
117 * Discard the given range in one go
119 static int discard_range(int fd, u64 start, u64 len)
121 u64 range[2] = { start, len };
123 if (ioctl(fd, BLKDISCARD, &range) < 0)
129 * Discard blocks in the given range in 1G chunks, the process is interruptible
131 static int discard_blocks(int fd, u64 start, u64 len)
135 u64 chunk_size = min_t(u64, len, 1*1024*1024*1024);
138 ret = discard_range(fd, start, chunk_size);
148 static u64 reference_root_table[] = {
149 [1] = BTRFS_ROOT_TREE_OBJECTID,
150 [2] = BTRFS_EXTENT_TREE_OBJECTID,
151 [3] = BTRFS_CHUNK_TREE_OBJECTID,
152 [4] = BTRFS_DEV_TREE_OBJECTID,
153 [5] = BTRFS_FS_TREE_OBJECTID,
154 [6] = BTRFS_CSUM_TREE_OBJECTID,
157 int test_uuid_unique(char *fs_uuid)
160 blkid_dev_iterate iter = NULL;
161 blkid_dev dev = NULL;
162 blkid_cache cache = NULL;
164 if (blkid_get_cache(&cache, NULL) < 0) {
165 printf("ERROR: lblkid cache get failed\n");
168 blkid_probe_all(cache);
169 iter = blkid_dev_iterate_begin(cache);
170 blkid_dev_set_search(iter, "UUID", fs_uuid);
172 while (blkid_dev_next(iter, &dev) == 0) {
173 dev = blkid_verify(cache, dev);
180 blkid_dev_iterate_end(iter);
181 blkid_put_cache(cache);
187 * Reserve space from free_tree.
188 * The algorithm is very simple, find the first cache_extent with enough space
189 * and allocate from its beginning.
191 static int reserve_free_space(struct cache_tree *free_tree, u64 len,
194 struct cache_extent *cache;
197 ASSERT(ret_start != NULL);
198 cache = first_cache_extent(free_tree);
200 if (cache->size > len) {
202 *ret_start = cache->start;
205 if (cache->size == 0) {
206 remove_cache_extent(free_tree, cache);
213 cache = next_cache_extent(cache);
220 static inline int write_temp_super(int fd, struct btrfs_super_block *sb,
226 crc = btrfs_csum_data(NULL, (char *)sb + BTRFS_CSUM_SIZE, crc,
227 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
228 btrfs_csum_final(crc, &sb->csum[0]);
229 ret = pwrite(fd, sb, BTRFS_SUPER_INFO_SIZE, sb_bytenr);
230 if (ret < BTRFS_SUPER_INFO_SIZE)
231 ret = (ret < 0 ? -errno : -EIO);
238 * Setup temporary superblock at cfg->super_bynter
239 * Needed info are extracted from cfg, and root_bytenr, chunk_bytenr
241 * For now sys chunk array will be empty and dev_item is empty too.
242 * They will be re-initialized at temp chunk tree setup.
244 * The superblock signature is not valid, denotes a partially created
245 * filesystem, needs to be finalized.
247 static int setup_temp_super(int fd, struct btrfs_mkfs_config *cfg,
248 u64 root_bytenr, u64 chunk_bytenr)
250 unsigned char chunk_uuid[BTRFS_UUID_SIZE];
251 char super_buf[BTRFS_SUPER_INFO_SIZE];
252 struct btrfs_super_block *super = (struct btrfs_super_block *)super_buf;
255 memset(super_buf, 0, BTRFS_SUPER_INFO_SIZE);
256 cfg->num_bytes = round_down(cfg->num_bytes, cfg->sectorsize);
259 if (uuid_parse(cfg->fs_uuid, super->fsid) != 0) {
260 error("cound not parse UUID: %s", cfg->fs_uuid);
264 if (!test_uuid_unique(cfg->fs_uuid)) {
265 error("non-unique UUID: %s", cfg->fs_uuid);
270 uuid_generate(super->fsid);
271 uuid_unparse(super->fsid, cfg->fs_uuid);
273 uuid_generate(chunk_uuid);
274 uuid_unparse(chunk_uuid, cfg->chunk_uuid);
276 btrfs_set_super_bytenr(super, cfg->super_bytenr);
277 btrfs_set_super_num_devices(super, 1);
278 btrfs_set_super_magic(super, BTRFS_MAGIC_PARTIAL);
279 btrfs_set_super_generation(super, 1);
280 btrfs_set_super_root(super, root_bytenr);
281 btrfs_set_super_chunk_root(super, chunk_bytenr);
282 btrfs_set_super_total_bytes(super, cfg->num_bytes);
284 * Temporary filesystem will only have 6 tree roots:
285 * chunk tree, root tree, extent_tree, device tree, fs tree
288 btrfs_set_super_bytes_used(super, 6 * cfg->nodesize);
289 btrfs_set_super_sectorsize(super, cfg->sectorsize);
290 btrfs_set_super_leafsize(super, cfg->nodesize);
291 btrfs_set_super_nodesize(super, cfg->nodesize);
292 btrfs_set_super_stripesize(super, cfg->stripesize);
293 btrfs_set_super_csum_type(super, BTRFS_CSUM_TYPE_CRC32);
294 btrfs_set_super_chunk_root(super, chunk_bytenr);
295 btrfs_set_super_cache_generation(super, -1);
296 btrfs_set_super_incompat_flags(super, cfg->features);
298 __strncpy_null(super->label, cfg->label, BTRFS_LABEL_SIZE - 1);
300 /* Sys chunk array will be re-initialized at chunk tree init time */
301 super->sys_chunk_array_size = 0;
303 ret = write_temp_super(fd, super, cfg->super_bytenr);
309 * Setup an extent buffer for tree block.
311 static int setup_temp_extent_buffer(struct extent_buffer *buf,
312 struct btrfs_mkfs_config *cfg,
313 u64 bytenr, u64 owner)
315 unsigned char fsid[BTRFS_FSID_SIZE];
316 unsigned char chunk_uuid[BTRFS_UUID_SIZE];
319 ret = uuid_parse(cfg->fs_uuid, fsid);
322 ret = uuid_parse(cfg->chunk_uuid, chunk_uuid);
326 memset(buf->data, 0, cfg->nodesize);
327 buf->len = cfg->nodesize;
328 btrfs_set_header_bytenr(buf, bytenr);
329 btrfs_set_header_generation(buf, 1);
330 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
331 btrfs_set_header_owner(buf, owner);
332 btrfs_set_header_flags(buf, BTRFS_HEADER_FLAG_WRITTEN);
333 write_extent_buffer(buf, chunk_uuid, btrfs_header_chunk_tree_uuid(buf),
335 write_extent_buffer(buf, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
339 static inline int write_temp_extent_buffer(int fd, struct extent_buffer *buf,
344 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
346 /* Temporary extent buffer is always mapped 1:1 on disk */
347 ret = pwrite(fd, buf->data, buf->len, bytenr);
349 ret = (ret < 0 ? ret : -EIO);
356 * Insert a root item for temporary tree root
358 * Only used in make_btrfs_v2().
360 static void insert_temp_root_item(struct extent_buffer *buf,
361 struct btrfs_mkfs_config *cfg,
362 int *slot, u32 *itemoff, u64 objectid,
365 struct btrfs_root_item root_item;
366 struct btrfs_inode_item *inode_item;
367 struct btrfs_disk_key disk_key;
369 btrfs_set_header_nritems(buf, *slot + 1);
370 (*itemoff) -= sizeof(root_item);
371 memset(&root_item, 0, sizeof(root_item));
372 inode_item = &root_item.inode;
373 btrfs_set_stack_inode_generation(inode_item, 1);
374 btrfs_set_stack_inode_size(inode_item, 3);
375 btrfs_set_stack_inode_nlink(inode_item, 1);
376 btrfs_set_stack_inode_nbytes(inode_item, cfg->nodesize);
377 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
378 btrfs_set_root_refs(&root_item, 1);
379 btrfs_set_root_used(&root_item, cfg->nodesize);
380 btrfs_set_root_generation(&root_item, 1);
381 btrfs_set_root_bytenr(&root_item, bytenr);
383 memset(&disk_key, 0, sizeof(disk_key));
384 btrfs_set_disk_key_type(&disk_key, BTRFS_ROOT_ITEM_KEY);
385 btrfs_set_disk_key_objectid(&disk_key, objectid);
386 btrfs_set_disk_key_offset(&disk_key, 0);
388 btrfs_set_item_key(buf, &disk_key, *slot);
389 btrfs_set_item_offset(buf, btrfs_item_nr(*slot), *itemoff);
390 btrfs_set_item_size(buf, btrfs_item_nr(*slot), sizeof(root_item));
391 write_extent_buffer(buf, &root_item,
392 btrfs_item_ptr_offset(buf, *slot),
397 static int setup_temp_root_tree(int fd, struct btrfs_mkfs_config *cfg,
398 u64 root_bytenr, u64 extent_bytenr,
399 u64 dev_bytenr, u64 fs_bytenr, u64 csum_bytenr)
401 struct extent_buffer *buf = NULL;
402 u32 itemoff = __BTRFS_LEAF_DATA_SIZE(cfg->nodesize);
407 * Provided bytenr must in ascending order, or tree root will have a
410 if (!(root_bytenr < extent_bytenr && extent_bytenr < dev_bytenr &&
411 dev_bytenr < fs_bytenr && fs_bytenr < csum_bytenr)) {
412 error("bad tree bytenr order: "
413 "root < extent %llu < %llu, "
414 "extent < dev %llu < %llu, "
415 "dev < fs %llu < %llu, "
416 "fs < csum %llu < %llu",
417 (unsigned long long)root_bytenr,
418 (unsigned long long)extent_bytenr,
419 (unsigned long long)extent_bytenr,
420 (unsigned long long)dev_bytenr,
421 (unsigned long long)dev_bytenr,
422 (unsigned long long)fs_bytenr,
423 (unsigned long long)fs_bytenr,
424 (unsigned long long)csum_bytenr);
427 buf = malloc(sizeof(*buf) + cfg->nodesize);
431 ret = setup_temp_extent_buffer(buf, cfg, root_bytenr,
432 BTRFS_ROOT_TREE_OBJECTID);
436 insert_temp_root_item(buf, cfg, &slot, &itemoff,
437 BTRFS_EXTENT_TREE_OBJECTID, extent_bytenr);
438 insert_temp_root_item(buf, cfg, &slot, &itemoff,
439 BTRFS_DEV_TREE_OBJECTID, dev_bytenr);
440 insert_temp_root_item(buf, cfg, &slot, &itemoff,
441 BTRFS_FS_TREE_OBJECTID, fs_bytenr);
442 insert_temp_root_item(buf, cfg, &slot, &itemoff,
443 BTRFS_CSUM_TREE_OBJECTID, csum_bytenr);
445 ret = write_temp_extent_buffer(fd, buf, root_bytenr);
451 static int insert_temp_dev_item(int fd, struct extent_buffer *buf,
452 struct btrfs_mkfs_config *cfg,
453 int *slot, u32 *itemoff)
455 struct btrfs_disk_key disk_key;
456 struct btrfs_dev_item *dev_item;
457 char super_buf[BTRFS_SUPER_INFO_SIZE];
458 unsigned char dev_uuid[BTRFS_UUID_SIZE];
459 unsigned char fsid[BTRFS_FSID_SIZE];
460 struct btrfs_super_block *super = (struct btrfs_super_block *)super_buf;
463 ret = pread(fd, super_buf, BTRFS_SUPER_INFO_SIZE, cfg->super_bytenr);
464 if (ret < BTRFS_SUPER_INFO_SIZE) {
465 ret = (ret < 0 ? -errno : -EIO);
469 btrfs_set_header_nritems(buf, *slot + 1);
470 (*itemoff) -= sizeof(*dev_item);
471 /* setup device item 1, 0 is for replace case */
472 btrfs_set_disk_key_type(&disk_key, BTRFS_DEV_ITEM_KEY);
473 btrfs_set_disk_key_objectid(&disk_key, BTRFS_DEV_ITEMS_OBJECTID);
474 btrfs_set_disk_key_offset(&disk_key, 1);
475 btrfs_set_item_key(buf, &disk_key, *slot);
476 btrfs_set_item_offset(buf, btrfs_item_nr(*slot), *itemoff);
477 btrfs_set_item_size(buf, btrfs_item_nr(*slot), sizeof(*dev_item));
479 dev_item = btrfs_item_ptr(buf, *slot, struct btrfs_dev_item);
480 /* Generate device uuid */
481 uuid_generate(dev_uuid);
482 write_extent_buffer(buf, dev_uuid,
483 (unsigned long)btrfs_device_uuid(dev_item),
485 uuid_parse(cfg->fs_uuid, fsid);
486 write_extent_buffer(buf, fsid,
487 (unsigned long)btrfs_device_fsid(dev_item),
489 btrfs_set_device_id(buf, dev_item, 1);
490 btrfs_set_device_generation(buf, dev_item, 0);
491 btrfs_set_device_total_bytes(buf, dev_item, cfg->num_bytes);
493 * The number must match the initial SYSTEM and META chunk size
495 btrfs_set_device_bytes_used(buf, dev_item,
496 BTRFS_MKFS_SYSTEM_GROUP_SIZE +
497 BTRFS_CONVERT_META_GROUP_SIZE);
498 btrfs_set_device_io_align(buf, dev_item, cfg->sectorsize);
499 btrfs_set_device_io_width(buf, dev_item, cfg->sectorsize);
500 btrfs_set_device_sector_size(buf, dev_item, cfg->sectorsize);
501 btrfs_set_device_type(buf, dev_item, 0);
503 /* Super dev_item is not complete, copy the complete one to sb */
504 read_extent_buffer(buf, &super->dev_item, (unsigned long)dev_item,
506 ret = write_temp_super(fd, super, cfg->super_bytenr);
512 static int insert_temp_chunk_item(int fd, struct extent_buffer *buf,
513 struct btrfs_mkfs_config *cfg,
514 int *slot, u32 *itemoff, u64 start, u64 len,
517 struct btrfs_chunk *chunk;
518 struct btrfs_disk_key disk_key;
519 char super_buf[BTRFS_SUPER_INFO_SIZE];
520 struct btrfs_super_block *sb = (struct btrfs_super_block *)super_buf;
523 ret = pread(fd, super_buf, BTRFS_SUPER_INFO_SIZE,
525 if (ret < BTRFS_SUPER_INFO_SIZE) {
526 ret = (ret < 0 ? ret : -EIO);
530 btrfs_set_header_nritems(buf, *slot + 1);
531 (*itemoff) -= btrfs_chunk_item_size(1);
532 btrfs_set_disk_key_type(&disk_key, BTRFS_CHUNK_ITEM_KEY);
533 btrfs_set_disk_key_objectid(&disk_key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
534 btrfs_set_disk_key_offset(&disk_key, start);
535 btrfs_set_item_key(buf, &disk_key, *slot);
536 btrfs_set_item_offset(buf, btrfs_item_nr(*slot), *itemoff);
537 btrfs_set_item_size(buf, btrfs_item_nr(*slot),
538 btrfs_chunk_item_size(1));
540 chunk = btrfs_item_ptr(buf, *slot, struct btrfs_chunk);
541 btrfs_set_chunk_length(buf, chunk, len);
542 btrfs_set_chunk_owner(buf, chunk, BTRFS_EXTENT_TREE_OBJECTID);
543 btrfs_set_chunk_stripe_len(buf, chunk, 64 * 1024);
544 btrfs_set_chunk_type(buf, chunk, type);
545 btrfs_set_chunk_io_align(buf, chunk, cfg->sectorsize);
546 btrfs_set_chunk_io_width(buf, chunk, cfg->sectorsize);
547 btrfs_set_chunk_sector_size(buf, chunk, cfg->sectorsize);
548 btrfs_set_chunk_num_stripes(buf, chunk, 1);
549 /* TODO: Support DUP profile for system chunk */
550 btrfs_set_stripe_devid_nr(buf, chunk, 0, 1);
551 /* We are doing 1:1 mapping, so start is its dev offset */
552 btrfs_set_stripe_offset_nr(buf, chunk, 0, start);
553 write_extent_buffer(buf, &sb->dev_item.uuid,
554 (unsigned long)btrfs_stripe_dev_uuid_nr(chunk, 0),
559 * If it's system chunk, also copy it to super block.
561 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
564 cur = (char *)sb->sys_chunk_array + sb->sys_chunk_array_size;
565 memcpy(cur, &disk_key, sizeof(disk_key));
566 cur += sizeof(disk_key);
567 read_extent_buffer(buf, cur, (unsigned long int)chunk,
568 btrfs_chunk_item_size(1));
569 sb->sys_chunk_array_size += btrfs_chunk_item_size(1) +
572 ret = write_temp_super(fd, sb, cfg->super_bytenr);
577 static int setup_temp_chunk_tree(int fd, struct btrfs_mkfs_config *cfg,
578 u64 sys_chunk_start, u64 meta_chunk_start,
581 struct extent_buffer *buf = NULL;
582 u32 itemoff = __BTRFS_LEAF_DATA_SIZE(cfg->nodesize);
586 /* Must ensure SYS chunk starts before META chunk */
587 if (meta_chunk_start < sys_chunk_start) {
588 error("wrong chunk order: meta < system %llu < %llu",
589 (unsigned long long)meta_chunk_start,
590 (unsigned long long)sys_chunk_start);
593 buf = malloc(sizeof(*buf) + cfg->nodesize);
596 ret = setup_temp_extent_buffer(buf, cfg, chunk_bytenr,
597 BTRFS_CHUNK_TREE_OBJECTID);
601 ret = insert_temp_dev_item(fd, buf, cfg, &slot, &itemoff);
604 ret = insert_temp_chunk_item(fd, buf, cfg, &slot, &itemoff,
606 BTRFS_MKFS_SYSTEM_GROUP_SIZE,
607 BTRFS_BLOCK_GROUP_SYSTEM);
610 ret = insert_temp_chunk_item(fd, buf, cfg, &slot, &itemoff,
612 BTRFS_CONVERT_META_GROUP_SIZE,
613 BTRFS_BLOCK_GROUP_METADATA);
616 ret = write_temp_extent_buffer(fd, buf, chunk_bytenr);
623 static void insert_temp_dev_extent(struct extent_buffer *buf,
624 int *slot, u32 *itemoff, u64 start, u64 len)
626 struct btrfs_dev_extent *dev_extent;
627 struct btrfs_disk_key disk_key;
629 btrfs_set_header_nritems(buf, *slot + 1);
630 (*itemoff) -= sizeof(*dev_extent);
631 btrfs_set_disk_key_type(&disk_key, BTRFS_DEV_EXTENT_KEY);
632 btrfs_set_disk_key_objectid(&disk_key, 1);
633 btrfs_set_disk_key_offset(&disk_key, start);
634 btrfs_set_item_key(buf, &disk_key, *slot);
635 btrfs_set_item_offset(buf, btrfs_item_nr(*slot), *itemoff);
636 btrfs_set_item_size(buf, btrfs_item_nr(*slot), sizeof(*dev_extent));
638 dev_extent = btrfs_item_ptr(buf, *slot, struct btrfs_dev_extent);
639 btrfs_set_dev_extent_chunk_objectid(buf, dev_extent,
640 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
641 btrfs_set_dev_extent_length(buf, dev_extent, len);
642 btrfs_set_dev_extent_chunk_offset(buf, dev_extent, start);
643 btrfs_set_dev_extent_chunk_tree(buf, dev_extent,
644 BTRFS_CHUNK_TREE_OBJECTID);
648 static int setup_temp_dev_tree(int fd, struct btrfs_mkfs_config *cfg,
649 u64 sys_chunk_start, u64 meta_chunk_start,
652 struct extent_buffer *buf = NULL;
653 u32 itemoff = __BTRFS_LEAF_DATA_SIZE(cfg->nodesize);
657 /* Must ensure SYS chunk starts before META chunk */
658 if (meta_chunk_start < sys_chunk_start) {
659 error("wrong chunk order: meta < system %llu < %llu",
660 (unsigned long long)meta_chunk_start,
661 (unsigned long long)sys_chunk_start);
664 buf = malloc(sizeof(*buf) + cfg->nodesize);
667 ret = setup_temp_extent_buffer(buf, cfg, dev_bytenr,
668 BTRFS_DEV_TREE_OBJECTID);
671 insert_temp_dev_extent(buf, &slot, &itemoff, sys_chunk_start,
672 BTRFS_MKFS_SYSTEM_GROUP_SIZE);
673 insert_temp_dev_extent(buf, &slot, &itemoff, meta_chunk_start,
674 BTRFS_CONVERT_META_GROUP_SIZE);
675 ret = write_temp_extent_buffer(fd, buf, dev_bytenr);
681 static int setup_temp_fs_tree(int fd, struct btrfs_mkfs_config *cfg,
684 struct extent_buffer *buf = NULL;
687 buf = malloc(sizeof(*buf) + cfg->nodesize);
690 ret = setup_temp_extent_buffer(buf, cfg, fs_bytenr,
691 BTRFS_FS_TREE_OBJECTID);
695 * Temporary fs tree is completely empty.
697 ret = write_temp_extent_buffer(fd, buf, fs_bytenr);
703 static int setup_temp_csum_tree(int fd, struct btrfs_mkfs_config *cfg,
706 struct extent_buffer *buf = NULL;
709 buf = malloc(sizeof(*buf) + cfg->nodesize);
712 ret = setup_temp_extent_buffer(buf, cfg, csum_bytenr,
713 BTRFS_CSUM_TREE_OBJECTID);
717 * Temporary csum tree is completely empty.
719 ret = write_temp_extent_buffer(fd, buf, csum_bytenr);
726 * Insert one temporary extent item.
728 * NOTE: if skinny_metadata is not enabled, this function must be called
729 * after all other trees are initialized.
730 * Or fs without skinny-metadata will be screwed up.
732 static int insert_temp_extent_item(int fd, struct extent_buffer *buf,
733 struct btrfs_mkfs_config *cfg,
734 int *slot, u32 *itemoff, u64 bytenr,
737 struct extent_buffer *tmp;
738 struct btrfs_extent_item *ei;
739 struct btrfs_extent_inline_ref *iref;
740 struct btrfs_disk_key disk_key;
741 struct btrfs_disk_key tree_info_key;
742 struct btrfs_tree_block_info *info;
744 int skinny_metadata = cfg->features &
745 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA;
749 itemsize = sizeof(*ei) + sizeof(*iref);
751 itemsize = sizeof(*ei) + sizeof(*iref) +
752 sizeof(struct btrfs_tree_block_info);
754 btrfs_set_header_nritems(buf, *slot + 1);
755 *(itemoff) -= itemsize;
757 if (skinny_metadata) {
758 btrfs_set_disk_key_type(&disk_key, BTRFS_METADATA_ITEM_KEY);
759 btrfs_set_disk_key_offset(&disk_key, 0);
761 btrfs_set_disk_key_type(&disk_key, BTRFS_EXTENT_ITEM_KEY);
762 btrfs_set_disk_key_offset(&disk_key, cfg->nodesize);
764 btrfs_set_disk_key_objectid(&disk_key, bytenr);
766 btrfs_set_item_key(buf, &disk_key, *slot);
767 btrfs_set_item_offset(buf, btrfs_item_nr(*slot), *itemoff);
768 btrfs_set_item_size(buf, btrfs_item_nr(*slot), itemsize);
770 ei = btrfs_item_ptr(buf, *slot, struct btrfs_extent_item);
771 btrfs_set_extent_refs(buf, ei, 1);
772 btrfs_set_extent_generation(buf, ei, 1);
773 btrfs_set_extent_flags(buf, ei, BTRFS_EXTENT_FLAG_TREE_BLOCK);
775 if (skinny_metadata) {
776 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
778 info = (struct btrfs_tree_block_info *)(ei + 1);
779 iref = (struct btrfs_extent_inline_ref *)(info + 1);
781 btrfs_set_extent_inline_ref_type(buf, iref,
782 BTRFS_TREE_BLOCK_REF_KEY);
783 btrfs_set_extent_inline_ref_offset(buf, iref, ref_root);
790 * Lastly, check the tree block key by read the tree block
791 * Since we do 1:1 mapping for convert case, we can directly
792 * read the bytenr from disk
794 tmp = malloc(sizeof(*tmp) + cfg->nodesize);
797 ret = setup_temp_extent_buffer(tmp, cfg, bytenr, ref_root);
800 ret = pread(fd, tmp->data, cfg->nodesize, bytenr);
801 if (ret < cfg->nodesize) {
802 ret = (ret < 0 ? -errno : -EIO);
805 if (btrfs_header_nritems(tmp) == 0) {
806 btrfs_set_disk_key_type(&tree_info_key, 0);
807 btrfs_set_disk_key_objectid(&tree_info_key, 0);
808 btrfs_set_disk_key_offset(&tree_info_key, 0);
810 btrfs_item_key(tmp, &tree_info_key, 0);
812 btrfs_set_tree_block_key(buf, info, &tree_info_key);
819 static void insert_temp_block_group(struct extent_buffer *buf,
820 struct btrfs_mkfs_config *cfg,
821 int *slot, u32 *itemoff,
822 u64 bytenr, u64 len, u64 used, u64 flag)
824 struct btrfs_block_group_item bgi;
825 struct btrfs_disk_key disk_key;
827 btrfs_set_header_nritems(buf, *slot + 1);
828 (*itemoff) -= sizeof(bgi);
829 btrfs_set_disk_key_type(&disk_key, BTRFS_BLOCK_GROUP_ITEM_KEY);
830 btrfs_set_disk_key_objectid(&disk_key, bytenr);
831 btrfs_set_disk_key_offset(&disk_key, len);
832 btrfs_set_item_key(buf, &disk_key, *slot);
833 btrfs_set_item_offset(buf, btrfs_item_nr(*slot), *itemoff);
834 btrfs_set_item_size(buf, btrfs_item_nr(*slot), sizeof(bgi));
836 btrfs_set_block_group_flags(&bgi, flag);
837 btrfs_set_block_group_used(&bgi, used);
838 btrfs_set_block_group_chunk_objectid(&bgi,
839 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
840 write_extent_buffer(buf, &bgi, btrfs_item_ptr_offset(buf, *slot),
845 static int setup_temp_extent_tree(int fd, struct btrfs_mkfs_config *cfg,
846 u64 chunk_bytenr, u64 root_bytenr,
847 u64 extent_bytenr, u64 dev_bytenr,
848 u64 fs_bytenr, u64 csum_bytenr)
850 struct extent_buffer *buf = NULL;
851 u32 itemoff = __BTRFS_LEAF_DATA_SIZE(cfg->nodesize);
856 * We must ensure provided bytenr are in ascending order,
857 * or extent tree key order will be broken.
859 if (!(chunk_bytenr < root_bytenr && root_bytenr < extent_bytenr &&
860 extent_bytenr < dev_bytenr && dev_bytenr < fs_bytenr &&
861 fs_bytenr < csum_bytenr)) {
862 error("bad tree bytenr order: "
863 "chunk < root %llu < %llu, "
864 "root < extent %llu < %llu, "
865 "extent < dev %llu < %llu, "
866 "dev < fs %llu < %llu, "
867 "fs < csum %llu < %llu",
868 (unsigned long long)chunk_bytenr,
869 (unsigned long long)root_bytenr,
870 (unsigned long long)root_bytenr,
871 (unsigned long long)extent_bytenr,
872 (unsigned long long)extent_bytenr,
873 (unsigned long long)dev_bytenr,
874 (unsigned long long)dev_bytenr,
875 (unsigned long long)fs_bytenr,
876 (unsigned long long)fs_bytenr,
877 (unsigned long long)csum_bytenr);
880 buf = malloc(sizeof(*buf) + cfg->nodesize);
884 ret = setup_temp_extent_buffer(buf, cfg, extent_bytenr,
885 BTRFS_EXTENT_TREE_OBJECTID);
889 ret = insert_temp_extent_item(fd, buf, cfg, &slot, &itemoff,
890 chunk_bytenr, BTRFS_CHUNK_TREE_OBJECTID);
894 insert_temp_block_group(buf, cfg, &slot, &itemoff, chunk_bytenr,
895 BTRFS_MKFS_SYSTEM_GROUP_SIZE, cfg->nodesize,
896 BTRFS_BLOCK_GROUP_SYSTEM);
898 ret = insert_temp_extent_item(fd, buf, cfg, &slot, &itemoff,
899 root_bytenr, BTRFS_ROOT_TREE_OBJECTID);
903 /* 5 tree block used, root, extent, dev, fs and csum*/
904 insert_temp_block_group(buf, cfg, &slot, &itemoff, root_bytenr,
905 BTRFS_CONVERT_META_GROUP_SIZE, cfg->nodesize * 5,
906 BTRFS_BLOCK_GROUP_METADATA);
908 ret = insert_temp_extent_item(fd, buf, cfg, &slot, &itemoff,
909 extent_bytenr, BTRFS_EXTENT_TREE_OBJECTID);
912 ret = insert_temp_extent_item(fd, buf, cfg, &slot, &itemoff,
913 dev_bytenr, BTRFS_DEV_TREE_OBJECTID);
916 ret = insert_temp_extent_item(fd, buf, cfg, &slot, &itemoff,
917 fs_bytenr, BTRFS_FS_TREE_OBJECTID);
920 ret = insert_temp_extent_item(fd, buf, cfg, &slot, &itemoff,
921 csum_bytenr, BTRFS_CSUM_TREE_OBJECTID);
925 ret = write_temp_extent_buffer(fd, buf, extent_bytenr);
932 * Improved version of make_btrfs().
935 * 1) Do chunk allocation to avoid used data
936 * And after this function, extent type matches chunk type
937 * 2) Better structured code
938 * No super long hand written codes to initialized all tree blocks
939 * Split into small blocks and reuse codes.
940 * TODO: Reuse tree operation facilities by introducing new flags
942 static int make_convert_btrfs(int fd, struct btrfs_mkfs_config *cfg,
943 struct btrfs_convert_context *cctx)
945 struct cache_tree *free = &cctx->free;
946 struct cache_tree *used = &cctx->used;
948 u64 meta_chunk_start;
949 /* chunk tree bytenr, in system chunk */
951 /* metadata trees bytenr, in metadata chunk */
959 /* Shouldn't happen */
960 BUG_ON(cache_tree_empty(used));
963 * reserve space for temporary superblock first
964 * Here we allocate a little larger space, to keep later
965 * free space will be STRIPE_LEN aligned
967 ret = reserve_free_space(free, BTRFS_STRIPE_LEN,
973 * Then reserve system chunk space
974 * TODO: Change system group size depending on cctx->total_bytes.
975 * If using current 4M, it can only handle less than one TB for
976 * worst case and then run out of sys space.
978 ret = reserve_free_space(free, BTRFS_MKFS_SYSTEM_GROUP_SIZE,
982 ret = reserve_free_space(free, BTRFS_CONVERT_META_GROUP_SIZE,
988 * Allocated meta/sys chunks will be mapped 1:1 with device offset.
990 * Inside the allocated metadata chunk, the layout will be:
991 * | offset | contents |
992 * -------------------------------------
994 * | +nodesize | extent root |
995 * | +nodesize * 2 | device root |
996 * | +nodesize * 3 | fs tree |
997 * | +nodesize * 4 | csum tree |
998 * -------------------------------------
999 * Inside the allocated system chunk, the layout will be:
1000 * | offset | contents |
1001 * -------------------------------------
1002 * | +0 | chunk root |
1003 * -------------------------------------
1005 chunk_bytenr = sys_chunk_start;
1006 root_bytenr = meta_chunk_start;
1007 extent_bytenr = meta_chunk_start + cfg->nodesize;
1008 dev_bytenr = meta_chunk_start + cfg->nodesize * 2;
1009 fs_bytenr = meta_chunk_start + cfg->nodesize * 3;
1010 csum_bytenr = meta_chunk_start + cfg->nodesize * 4;
1012 ret = setup_temp_super(fd, cfg, root_bytenr, chunk_bytenr);
1016 ret = setup_temp_root_tree(fd, cfg, root_bytenr, extent_bytenr,
1017 dev_bytenr, fs_bytenr, csum_bytenr);
1020 ret = setup_temp_chunk_tree(fd, cfg, sys_chunk_start, meta_chunk_start,
1024 ret = setup_temp_dev_tree(fd, cfg, sys_chunk_start, meta_chunk_start,
1028 ret = setup_temp_fs_tree(fd, cfg, fs_bytenr);
1031 ret = setup_temp_csum_tree(fd, cfg, csum_bytenr);
1035 * Setup extent tree last, since it may need to read tree block key
1036 * for non-skinny metadata case.
1038 ret = setup_temp_extent_tree(fd, cfg, chunk_bytenr, root_bytenr,
1039 extent_bytenr, dev_bytenr, fs_bytenr,
1046 * @fs_uuid - if NULL, generates a UUID, returns back the new filesystem UUID
1048 * The superblock signature is not valid, denotes a partially created
1049 * filesystem, needs to be finalized.
1051 int make_btrfs(int fd, struct btrfs_mkfs_config *cfg,
1052 struct btrfs_convert_context *cctx)
1054 struct btrfs_super_block super;
1055 struct extent_buffer *buf;
1056 struct btrfs_root_item root_item;
1057 struct btrfs_disk_key disk_key;
1058 struct btrfs_extent_item *extent_item;
1059 struct btrfs_inode_item *inode_item;
1060 struct btrfs_chunk *chunk;
1061 struct btrfs_dev_item *dev_item;
1062 struct btrfs_dev_extent *dev_extent;
1063 u8 chunk_tree_uuid[BTRFS_UUID_SIZE];
1073 int skinny_metadata = !!(cfg->features &
1074 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1078 return make_convert_btrfs(fd, cfg, cctx);
1079 buf = malloc(sizeof(*buf) + max(cfg->sectorsize, cfg->nodesize));
1083 first_free = BTRFS_SUPER_INFO_OFFSET + cfg->sectorsize * 2 - 1;
1084 first_free &= ~((u64)cfg->sectorsize - 1);
1086 memset(&super, 0, sizeof(super));
1088 num_bytes = (cfg->num_bytes / cfg->sectorsize) * cfg->sectorsize;
1089 if (*cfg->fs_uuid) {
1090 if (uuid_parse(cfg->fs_uuid, super.fsid) != 0) {
1091 error("cannot not parse UUID: %s", cfg->fs_uuid);
1095 if (!test_uuid_unique(cfg->fs_uuid)) {
1096 error("non-unique UUID: %s", cfg->fs_uuid);
1101 uuid_generate(super.fsid);
1102 uuid_unparse(super.fsid, cfg->fs_uuid);
1104 uuid_generate(super.dev_item.uuid);
1105 uuid_generate(chunk_tree_uuid);
1107 btrfs_set_super_bytenr(&super, cfg->blocks[0]);
1108 btrfs_set_super_num_devices(&super, 1);
1109 btrfs_set_super_magic(&super, BTRFS_MAGIC_PARTIAL);
1110 btrfs_set_super_generation(&super, 1);
1111 btrfs_set_super_root(&super, cfg->blocks[1]);
1112 btrfs_set_super_chunk_root(&super, cfg->blocks[3]);
1113 btrfs_set_super_total_bytes(&super, num_bytes);
1114 btrfs_set_super_bytes_used(&super, 6 * cfg->nodesize);
1115 btrfs_set_super_sectorsize(&super, cfg->sectorsize);
1116 btrfs_set_super_leafsize(&super, cfg->nodesize);
1117 btrfs_set_super_nodesize(&super, cfg->nodesize);
1118 btrfs_set_super_stripesize(&super, cfg->stripesize);
1119 btrfs_set_super_csum_type(&super, BTRFS_CSUM_TYPE_CRC32);
1120 btrfs_set_super_chunk_root_generation(&super, 1);
1121 btrfs_set_super_cache_generation(&super, -1);
1122 btrfs_set_super_incompat_flags(&super, cfg->features);
1124 __strncpy_null(super.label, cfg->label, BTRFS_LABEL_SIZE - 1);
1126 /* create the tree of root objects */
1127 memset(buf->data, 0, cfg->nodesize);
1128 buf->len = cfg->nodesize;
1129 btrfs_set_header_bytenr(buf, cfg->blocks[1]);
1130 btrfs_set_header_nritems(buf, 4);
1131 btrfs_set_header_generation(buf, 1);
1132 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
1133 btrfs_set_header_owner(buf, BTRFS_ROOT_TREE_OBJECTID);
1134 write_extent_buffer(buf, super.fsid, btrfs_header_fsid(),
1137 write_extent_buffer(buf, chunk_tree_uuid,
1138 btrfs_header_chunk_tree_uuid(buf),
1141 /* create the items for the root tree */
1142 memset(&root_item, 0, sizeof(root_item));
1143 inode_item = &root_item.inode;
1144 btrfs_set_stack_inode_generation(inode_item, 1);
1145 btrfs_set_stack_inode_size(inode_item, 3);
1146 btrfs_set_stack_inode_nlink(inode_item, 1);
1147 btrfs_set_stack_inode_nbytes(inode_item, cfg->nodesize);
1148 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1149 btrfs_set_root_refs(&root_item, 1);
1150 btrfs_set_root_used(&root_item, cfg->nodesize);
1151 btrfs_set_root_generation(&root_item, 1);
1153 memset(&disk_key, 0, sizeof(disk_key));
1154 btrfs_set_disk_key_type(&disk_key, BTRFS_ROOT_ITEM_KEY);
1155 btrfs_set_disk_key_offset(&disk_key, 0);
1158 itemoff = __BTRFS_LEAF_DATA_SIZE(cfg->nodesize) - sizeof(root_item);
1159 btrfs_set_root_bytenr(&root_item, cfg->blocks[2]);
1160 btrfs_set_disk_key_objectid(&disk_key, BTRFS_EXTENT_TREE_OBJECTID);
1161 btrfs_set_item_key(buf, &disk_key, nritems);
1162 btrfs_set_item_offset(buf, btrfs_item_nr(nritems), itemoff);
1163 btrfs_set_item_size(buf, btrfs_item_nr(nritems),
1165 write_extent_buffer(buf, &root_item, btrfs_item_ptr_offset(buf,
1166 nritems), sizeof(root_item));
1169 itemoff = itemoff - sizeof(root_item);
1170 btrfs_set_root_bytenr(&root_item, cfg->blocks[4]);
1171 btrfs_set_disk_key_objectid(&disk_key, BTRFS_DEV_TREE_OBJECTID);
1172 btrfs_set_item_key(buf, &disk_key, nritems);
1173 btrfs_set_item_offset(buf, btrfs_item_nr(nritems), itemoff);
1174 btrfs_set_item_size(buf, btrfs_item_nr(nritems),
1176 write_extent_buffer(buf, &root_item,
1177 btrfs_item_ptr_offset(buf, nritems),
1181 itemoff = itemoff - sizeof(root_item);
1182 btrfs_set_root_bytenr(&root_item, cfg->blocks[5]);
1183 btrfs_set_disk_key_objectid(&disk_key, BTRFS_FS_TREE_OBJECTID);
1184 btrfs_set_item_key(buf, &disk_key, nritems);
1185 btrfs_set_item_offset(buf, btrfs_item_nr(nritems), itemoff);
1186 btrfs_set_item_size(buf, btrfs_item_nr(nritems),
1188 write_extent_buffer(buf, &root_item,
1189 btrfs_item_ptr_offset(buf, nritems),
1193 itemoff = itemoff - sizeof(root_item);
1194 btrfs_set_root_bytenr(&root_item, cfg->blocks[6]);
1195 btrfs_set_disk_key_objectid(&disk_key, BTRFS_CSUM_TREE_OBJECTID);
1196 btrfs_set_item_key(buf, &disk_key, nritems);
1197 btrfs_set_item_offset(buf, btrfs_item_nr(nritems), itemoff);
1198 btrfs_set_item_size(buf, btrfs_item_nr(nritems),
1200 write_extent_buffer(buf, &root_item,
1201 btrfs_item_ptr_offset(buf, nritems),
1206 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1207 ret = pwrite(fd, buf->data, cfg->nodesize, cfg->blocks[1]);
1208 if (ret != cfg->nodesize) {
1209 ret = (ret < 0 ? -errno : -EIO);
1213 /* create the items for the extent tree */
1214 memset(buf->data + sizeof(struct btrfs_header), 0,
1215 cfg->nodesize - sizeof(struct btrfs_header));
1217 itemoff = __BTRFS_LEAF_DATA_SIZE(cfg->nodesize);
1218 for (i = 1; i < 7; i++) {
1219 item_size = sizeof(struct btrfs_extent_item);
1220 if (!skinny_metadata)
1221 item_size += sizeof(struct btrfs_tree_block_info);
1223 if (cfg->blocks[i] < first_free) {
1224 error("block[%d] below first free: %llu < %llu",
1225 i, (unsigned long long)cfg->blocks[i],
1226 (unsigned long long)first_free);
1230 if (cfg->blocks[i] < cfg->blocks[i - 1]) {
1231 error("blocks %d and %d in reverse order: %llu < %llu",
1233 (unsigned long long)cfg->blocks[i],
1234 (unsigned long long)cfg->blocks[i - 1]);
1239 /* create extent item */
1240 itemoff -= item_size;
1241 btrfs_set_disk_key_objectid(&disk_key, cfg->blocks[i]);
1242 if (skinny_metadata) {
1243 btrfs_set_disk_key_type(&disk_key,
1244 BTRFS_METADATA_ITEM_KEY);
1245 btrfs_set_disk_key_offset(&disk_key, 0);
1247 btrfs_set_disk_key_type(&disk_key,
1248 BTRFS_EXTENT_ITEM_KEY);
1249 btrfs_set_disk_key_offset(&disk_key, cfg->nodesize);
1251 btrfs_set_item_key(buf, &disk_key, nritems);
1252 btrfs_set_item_offset(buf, btrfs_item_nr(nritems),
1254 btrfs_set_item_size(buf, btrfs_item_nr(nritems),
1256 extent_item = btrfs_item_ptr(buf, nritems,
1257 struct btrfs_extent_item);
1258 btrfs_set_extent_refs(buf, extent_item, 1);
1259 btrfs_set_extent_generation(buf, extent_item, 1);
1260 btrfs_set_extent_flags(buf, extent_item,
1261 BTRFS_EXTENT_FLAG_TREE_BLOCK);
1264 /* create extent ref */
1265 ref_root = reference_root_table[i];
1266 btrfs_set_disk_key_objectid(&disk_key, cfg->blocks[i]);
1267 btrfs_set_disk_key_offset(&disk_key, ref_root);
1268 btrfs_set_disk_key_type(&disk_key, BTRFS_TREE_BLOCK_REF_KEY);
1269 btrfs_set_item_key(buf, &disk_key, nritems);
1270 btrfs_set_item_offset(buf, btrfs_item_nr(nritems),
1272 btrfs_set_item_size(buf, btrfs_item_nr(nritems), 0);
1275 btrfs_set_header_bytenr(buf, cfg->blocks[2]);
1276 btrfs_set_header_owner(buf, BTRFS_EXTENT_TREE_OBJECTID);
1277 btrfs_set_header_nritems(buf, nritems);
1278 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1279 ret = pwrite(fd, buf->data, cfg->nodesize, cfg->blocks[2]);
1280 if (ret != cfg->nodesize) {
1281 ret = (ret < 0 ? -errno : -EIO);
1285 /* create the chunk tree */
1286 memset(buf->data + sizeof(struct btrfs_header), 0,
1287 cfg->nodesize - sizeof(struct btrfs_header));
1289 item_size = sizeof(*dev_item);
1290 itemoff = __BTRFS_LEAF_DATA_SIZE(cfg->nodesize) - item_size;
1292 /* first device 1 (there is no device 0) */
1293 btrfs_set_disk_key_objectid(&disk_key, BTRFS_DEV_ITEMS_OBJECTID);
1294 btrfs_set_disk_key_offset(&disk_key, 1);
1295 btrfs_set_disk_key_type(&disk_key, BTRFS_DEV_ITEM_KEY);
1296 btrfs_set_item_key(buf, &disk_key, nritems);
1297 btrfs_set_item_offset(buf, btrfs_item_nr(nritems), itemoff);
1298 btrfs_set_item_size(buf, btrfs_item_nr(nritems), item_size);
1300 dev_item = btrfs_item_ptr(buf, nritems, struct btrfs_dev_item);
1301 btrfs_set_device_id(buf, dev_item, 1);
1302 btrfs_set_device_generation(buf, dev_item, 0);
1303 btrfs_set_device_total_bytes(buf, dev_item, num_bytes);
1304 btrfs_set_device_bytes_used(buf, dev_item,
1305 BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1306 btrfs_set_device_io_align(buf, dev_item, cfg->sectorsize);
1307 btrfs_set_device_io_width(buf, dev_item, cfg->sectorsize);
1308 btrfs_set_device_sector_size(buf, dev_item, cfg->sectorsize);
1309 btrfs_set_device_type(buf, dev_item, 0);
1311 write_extent_buffer(buf, super.dev_item.uuid,
1312 (unsigned long)btrfs_device_uuid(dev_item),
1314 write_extent_buffer(buf, super.fsid,
1315 (unsigned long)btrfs_device_fsid(dev_item),
1317 read_extent_buffer(buf, &super.dev_item, (unsigned long)dev_item,
1321 item_size = btrfs_chunk_item_size(1);
1322 itemoff = itemoff - item_size;
1324 /* then we have chunk 0 */
1325 btrfs_set_disk_key_objectid(&disk_key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1326 btrfs_set_disk_key_offset(&disk_key, 0);
1327 btrfs_set_disk_key_type(&disk_key, BTRFS_CHUNK_ITEM_KEY);
1328 btrfs_set_item_key(buf, &disk_key, nritems);
1329 btrfs_set_item_offset(buf, btrfs_item_nr(nritems), itemoff);
1330 btrfs_set_item_size(buf, btrfs_item_nr(nritems), item_size);
1332 chunk = btrfs_item_ptr(buf, nritems, struct btrfs_chunk);
1333 btrfs_set_chunk_length(buf, chunk, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1334 btrfs_set_chunk_owner(buf, chunk, BTRFS_EXTENT_TREE_OBJECTID);
1335 btrfs_set_chunk_stripe_len(buf, chunk, 64 * 1024);
1336 btrfs_set_chunk_type(buf, chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1337 btrfs_set_chunk_io_align(buf, chunk, cfg->sectorsize);
1338 btrfs_set_chunk_io_width(buf, chunk, cfg->sectorsize);
1339 btrfs_set_chunk_sector_size(buf, chunk, cfg->sectorsize);
1340 btrfs_set_chunk_num_stripes(buf, chunk, 1);
1341 btrfs_set_stripe_devid_nr(buf, chunk, 0, 1);
1342 btrfs_set_stripe_offset_nr(buf, chunk, 0, 0);
1345 write_extent_buffer(buf, super.dev_item.uuid,
1346 (unsigned long)btrfs_stripe_dev_uuid(&chunk->stripe),
1349 /* copy the key for the chunk to the system array */
1350 ptr = super.sys_chunk_array;
1351 array_size = sizeof(disk_key);
1353 memcpy(ptr, &disk_key, sizeof(disk_key));
1354 ptr += sizeof(disk_key);
1356 /* copy the chunk to the system array */
1357 read_extent_buffer(buf, ptr, (unsigned long)chunk, item_size);
1358 array_size += item_size;
1360 btrfs_set_super_sys_array_size(&super, array_size);
1362 btrfs_set_header_bytenr(buf, cfg->blocks[3]);
1363 btrfs_set_header_owner(buf, BTRFS_CHUNK_TREE_OBJECTID);
1364 btrfs_set_header_nritems(buf, nritems);
1365 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1366 ret = pwrite(fd, buf->data, cfg->nodesize, cfg->blocks[3]);
1367 if (ret != cfg->nodesize) {
1368 ret = (ret < 0 ? -errno : -EIO);
1372 /* create the device tree */
1373 memset(buf->data + sizeof(struct btrfs_header), 0,
1374 cfg->nodesize - sizeof(struct btrfs_header));
1376 itemoff = __BTRFS_LEAF_DATA_SIZE(cfg->nodesize) -
1377 sizeof(struct btrfs_dev_extent);
1379 btrfs_set_disk_key_objectid(&disk_key, 1);
1380 btrfs_set_disk_key_offset(&disk_key, 0);
1381 btrfs_set_disk_key_type(&disk_key, BTRFS_DEV_EXTENT_KEY);
1382 btrfs_set_item_key(buf, &disk_key, nritems);
1383 btrfs_set_item_offset(buf, btrfs_item_nr(nritems), itemoff);
1384 btrfs_set_item_size(buf, btrfs_item_nr(nritems),
1385 sizeof(struct btrfs_dev_extent));
1386 dev_extent = btrfs_item_ptr(buf, nritems, struct btrfs_dev_extent);
1387 btrfs_set_dev_extent_chunk_tree(buf, dev_extent,
1388 BTRFS_CHUNK_TREE_OBJECTID);
1389 btrfs_set_dev_extent_chunk_objectid(buf, dev_extent,
1390 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1391 btrfs_set_dev_extent_chunk_offset(buf, dev_extent, 0);
1393 write_extent_buffer(buf, chunk_tree_uuid,
1394 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(dev_extent),
1397 btrfs_set_dev_extent_length(buf, dev_extent,
1398 BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1401 btrfs_set_header_bytenr(buf, cfg->blocks[4]);
1402 btrfs_set_header_owner(buf, BTRFS_DEV_TREE_OBJECTID);
1403 btrfs_set_header_nritems(buf, nritems);
1404 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1405 ret = pwrite(fd, buf->data, cfg->nodesize, cfg->blocks[4]);
1406 if (ret != cfg->nodesize) {
1407 ret = (ret < 0 ? -errno : -EIO);
1411 /* create the FS root */
1412 memset(buf->data + sizeof(struct btrfs_header), 0,
1413 cfg->nodesize - sizeof(struct btrfs_header));
1414 btrfs_set_header_bytenr(buf, cfg->blocks[5]);
1415 btrfs_set_header_owner(buf, BTRFS_FS_TREE_OBJECTID);
1416 btrfs_set_header_nritems(buf, 0);
1417 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1418 ret = pwrite(fd, buf->data, cfg->nodesize, cfg->blocks[5]);
1419 if (ret != cfg->nodesize) {
1420 ret = (ret < 0 ? -errno : -EIO);
1423 /* finally create the csum root */
1424 memset(buf->data + sizeof(struct btrfs_header), 0,
1425 cfg->nodesize - sizeof(struct btrfs_header));
1426 btrfs_set_header_bytenr(buf, cfg->blocks[6]);
1427 btrfs_set_header_owner(buf, BTRFS_CSUM_TREE_OBJECTID);
1428 btrfs_set_header_nritems(buf, 0);
1429 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1430 ret = pwrite(fd, buf->data, cfg->nodesize, cfg->blocks[6]);
1431 if (ret != cfg->nodesize) {
1432 ret = (ret < 0 ? -errno : -EIO);
1436 /* and write out the super block */
1437 memset(buf->data, 0, BTRFS_SUPER_INFO_SIZE);
1438 memcpy(buf->data, &super, sizeof(super));
1439 buf->len = BTRFS_SUPER_INFO_SIZE;
1440 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1441 ret = pwrite(fd, buf->data, BTRFS_SUPER_INFO_SIZE, cfg->blocks[0]);
1442 if (ret != BTRFS_SUPER_INFO_SIZE) {
1443 ret = (ret < 0 ? -errno : -EIO);
1454 #define VERSION_TO_STRING3(a,b,c) #a "." #b "." #c, KERNEL_VERSION(a,b,c)
1455 #define VERSION_TO_STRING2(a,b) #a "." #b, KERNEL_VERSION(a,b,0)
1458 * Feature stability status and versions: compat <= safe <= default
1460 static const struct btrfs_fs_feature {
1463 const char *sysfs_name;
1465 * Compatibility with kernel of given version. Filesystem can be
1468 const char *compat_str;
1471 * Considered safe for use, but is not on by default, even if the
1472 * kernel supports the feature.
1474 const char *safe_str;
1477 * Considered safe for use and will be turned on by default if
1478 * supported by the running kernel.
1480 const char *default_str;
1483 } mkfs_features[] = {
1484 { "mixed-bg", BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS,
1486 VERSION_TO_STRING3(2,6,37),
1487 VERSION_TO_STRING3(2,6,37),
1489 "mixed data and metadata block groups" },
1490 { "extref", BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF,
1492 VERSION_TO_STRING2(3,7),
1493 VERSION_TO_STRING2(3,12),
1494 VERSION_TO_STRING2(3,12),
1495 "increased hardlink limit per file to 65536" },
1496 { "raid56", BTRFS_FEATURE_INCOMPAT_RAID56,
1498 VERSION_TO_STRING2(3,9),
1501 "raid56 extended format" },
1502 { "skinny-metadata", BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA,
1504 VERSION_TO_STRING2(3,10),
1505 VERSION_TO_STRING2(3,18),
1506 VERSION_TO_STRING2(3,18),
1507 "reduced-size metadata extent refs" },
1508 { "no-holes", BTRFS_FEATURE_INCOMPAT_NO_HOLES,
1510 VERSION_TO_STRING2(3,14),
1511 VERSION_TO_STRING2(4,0),
1513 "no explicit hole extents for files" },
1514 /* Keep this one last */
1515 { "list-all", BTRFS_FEATURE_LIST_ALL, NULL }
1518 static int parse_one_fs_feature(const char *name, u64 *flags)
1523 for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1524 if (name[0] == '^' &&
1525 !strcmp(mkfs_features[i].name, name + 1)) {
1526 *flags &= ~ mkfs_features[i].flag;
1528 } else if (!strcmp(mkfs_features[i].name, name)) {
1529 *flags |= mkfs_features[i].flag;
1537 void btrfs_parse_features_to_string(char *buf, u64 flags)
1543 for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1544 if (flags & mkfs_features[i].flag) {
1547 strcat(buf, mkfs_features[i].name);
1552 void btrfs_process_fs_features(u64 flags)
1556 for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1557 if (flags & mkfs_features[i].flag) {
1558 printf("Turning ON incompat feature '%s': %s\n",
1559 mkfs_features[i].name,
1560 mkfs_features[i].desc);
1565 void btrfs_list_all_fs_features(u64 mask_disallowed)
1569 fprintf(stderr, "Filesystem features available:\n");
1570 for (i = 0; i < ARRAY_SIZE(mkfs_features) - 1; i++) {
1571 const struct btrfs_fs_feature *feat = &mkfs_features[i];
1573 if (feat->flag & mask_disallowed)
1575 fprintf(stderr, "%-20s- %s (0x%llx", feat->name, feat->desc,
1577 if (feat->compat_ver)
1578 fprintf(stderr, ", compat=%s", feat->compat_str);
1580 fprintf(stderr, ", safe=%s", feat->safe_str);
1581 if (feat->default_ver)
1582 fprintf(stderr, ", default=%s", feat->default_str);
1583 fprintf(stderr, ")\n");
1588 * Return NULL if all features were parsed fine, otherwise return the name of
1589 * the first unparsed.
1591 char* btrfs_parse_fs_features(char *namelist, u64 *flags)
1594 char *save_ptr = NULL; /* Satisfy static checkers */
1596 for (this_char = strtok_r(namelist, ",", &save_ptr);
1598 this_char = strtok_r(NULL, ",", &save_ptr)) {
1599 if (parse_one_fs_feature(this_char, flags))
1606 void print_kernel_version(FILE *stream, u32 version)
1610 v[0] = version & 0xFF;
1611 v[1] = (version >> 8) & 0xFF;
1612 v[2] = version >> 16;
1613 fprintf(stream, "%u.%u", v[2], v[1]);
1615 fprintf(stream, ".%u", v[0]);
1618 u32 get_running_kernel_version(void)
1620 struct utsname utsbuf;
1622 char *saveptr = NULL;
1626 if (strcmp(utsbuf.sysname, "Linux") != 0) {
1627 error("unsupported system: %s", utsbuf.sysname);
1631 tmp = strchr(utsbuf.release, '-');
1635 tmp = strtok_r(utsbuf.release, ".", &saveptr);
1636 if (!string_is_numerical(tmp))
1638 version = atoi(tmp) << 16;
1639 tmp = strtok_r(NULL, ".", &saveptr);
1640 if (!string_is_numerical(tmp))
1642 version |= atoi(tmp) << 8;
1643 tmp = strtok_r(NULL, ".", &saveptr);
1645 if (!string_is_numerical(tmp))
1647 version |= atoi(tmp);
1653 u64 btrfs_device_size(int fd, struct stat *st)
1656 if (S_ISREG(st->st_mode)) {
1659 if (!S_ISBLK(st->st_mode)) {
1662 if (ioctl(fd, BLKGETSIZE64, &size) >= 0) {
1668 static int zero_blocks(int fd, off_t start, size_t len)
1670 char *buf = malloc(len);
1676 memset(buf, 0, len);
1677 written = pwrite(fd, buf, len, start);
1684 #define ZERO_DEV_BYTES (2 * 1024 * 1024)
1686 /* don't write outside the device by clamping the region to the device size */
1687 static int zero_dev_clamped(int fd, off_t start, ssize_t len, u64 dev_size)
1689 off_t end = max(start, start + len);
1692 /* and don't overwrite the disk labels on sparc */
1693 start = max(start, 1024);
1694 end = max(end, 1024);
1697 start = min_t(u64, start, dev_size);
1698 end = min_t(u64, end, dev_size);
1700 return zero_blocks(fd, start, end - start);
1703 int btrfs_add_to_fsid(struct btrfs_trans_handle *trans,
1704 struct btrfs_root *root, int fd, const char *path,
1705 u64 device_total_bytes, u32 io_width, u32 io_align,
1708 struct btrfs_super_block *disk_super;
1709 struct btrfs_super_block *super = root->fs_info->super_copy;
1710 struct btrfs_device *device;
1711 struct btrfs_dev_item *dev_item;
1717 device_total_bytes = (device_total_bytes / sectorsize) * sectorsize;
1719 device = calloc(1, sizeof(*device));
1724 buf = calloc(1, sectorsize);
1730 disk_super = (struct btrfs_super_block *)buf;
1731 dev_item = &disk_super->dev_item;
1733 uuid_generate(device->uuid);
1736 device->io_width = io_width;
1737 device->io_align = io_align;
1738 device->sector_size = sectorsize;
1740 device->writeable = 1;
1741 device->total_bytes = device_total_bytes;
1742 device->bytes_used = 0;
1743 device->total_ios = 0;
1744 device->dev_root = root->fs_info->dev_root;
1745 device->name = strdup(path);
1746 if (!device->name) {
1751 INIT_LIST_HEAD(&device->dev_list);
1752 ret = btrfs_add_device(trans, root, device);
1756 fs_total_bytes = btrfs_super_total_bytes(super) + device_total_bytes;
1757 btrfs_set_super_total_bytes(super, fs_total_bytes);
1759 num_devs = btrfs_super_num_devices(super) + 1;
1760 btrfs_set_super_num_devices(super, num_devs);
1762 memcpy(disk_super, super, sizeof(*disk_super));
1764 btrfs_set_super_bytenr(disk_super, BTRFS_SUPER_INFO_OFFSET);
1765 btrfs_set_stack_device_id(dev_item, device->devid);
1766 btrfs_set_stack_device_type(dev_item, device->type);
1767 btrfs_set_stack_device_io_align(dev_item, device->io_align);
1768 btrfs_set_stack_device_io_width(dev_item, device->io_width);
1769 btrfs_set_stack_device_sector_size(dev_item, device->sector_size);
1770 btrfs_set_stack_device_total_bytes(dev_item, device->total_bytes);
1771 btrfs_set_stack_device_bytes_used(dev_item, device->bytes_used);
1772 memcpy(&dev_item->uuid, device->uuid, BTRFS_UUID_SIZE);
1774 ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
1775 BUG_ON(ret != sectorsize);
1778 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1779 device->fs_devices = root->fs_info->fs_devices;
1788 static int btrfs_wipe_existing_sb(int fd)
1790 const char *off = NULL;
1795 blkid_probe pr = NULL;
1797 pr = blkid_new_probe();
1801 if (blkid_probe_set_device(pr, fd, 0, 0)) {
1806 ret = blkid_probe_lookup_value(pr, "SBMAGIC_OFFSET", &off, NULL);
1808 ret = blkid_probe_lookup_value(pr, "SBMAGIC", NULL, &len);
1810 if (ret || len == 0 || off == NULL) {
1812 * If lookup fails, the probe did not find any values, eg. for
1813 * a file image or a loop device. Soft error.
1819 offset = strtoll(off, NULL, 10);
1820 if (len > sizeof(buf))
1823 memset(buf, 0, len);
1824 ret = pwrite(fd, buf, len, offset);
1826 error("cannot wipe existing superblock: %s", strerror(errno));
1828 } else if (ret != len) {
1829 error("cannot wipe existing superblock: wrote %d of %zd", ret, len);
1835 blkid_free_probe(pr);
1839 int btrfs_prepare_device(int fd, const char *file, u64 *block_count_ret,
1840 u64 max_block_count, unsigned opflags)
1846 ret = fstat(fd, &st);
1848 error("unable to stat %s: %s", file, strerror(errno));
1852 block_count = btrfs_device_size(fd, &st);
1853 if (block_count == 0) {
1854 error("unable to determine size of %s", file);
1857 if (max_block_count)
1858 block_count = min(block_count, max_block_count);
1860 if (opflags & PREP_DEVICE_DISCARD) {
1862 * We intentionally ignore errors from the discard ioctl. It
1863 * is not necessary for the mkfs functionality but just an
1866 if (discard_range(fd, 0, 0) == 0) {
1867 if (opflags & PREP_DEVICE_VERBOSE)
1868 printf("Performing full device TRIM (%s) ...\n",
1869 pretty_size(block_count));
1870 discard_blocks(fd, 0, block_count);
1874 ret = zero_dev_clamped(fd, 0, ZERO_DEV_BYTES, block_count);
1875 for (i = 0 ; !ret && i < BTRFS_SUPER_MIRROR_MAX; i++)
1876 ret = zero_dev_clamped(fd, btrfs_sb_offset(i),
1877 BTRFS_SUPER_INFO_SIZE, block_count);
1878 if (!ret && (opflags & PREP_DEVICE_ZERO_END))
1879 ret = zero_dev_clamped(fd, block_count - ZERO_DEV_BYTES,
1880 ZERO_DEV_BYTES, block_count);
1883 error("failed to zero device '%s': %s", file, strerror(-ret));
1887 ret = btrfs_wipe_existing_sb(fd);
1889 error("cannot wipe superblocks on %s", file);
1893 *block_count_ret = block_count;
1897 int btrfs_make_root_dir(struct btrfs_trans_handle *trans,
1898 struct btrfs_root *root, u64 objectid)
1901 struct btrfs_inode_item inode_item;
1902 time_t now = time(NULL);
1904 memset(&inode_item, 0, sizeof(inode_item));
1905 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
1906 btrfs_set_stack_inode_size(&inode_item, 0);
1907 btrfs_set_stack_inode_nlink(&inode_item, 1);
1908 btrfs_set_stack_inode_nbytes(&inode_item, root->nodesize);
1909 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
1910 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
1911 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
1912 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
1913 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
1914 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
1915 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
1916 btrfs_set_stack_timespec_sec(&inode_item.otime, now);
1917 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
1919 if (root->fs_info->tree_root == root)
1920 btrfs_set_super_root_dir(root->fs_info->super_copy, objectid);
1922 ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
1926 ret = btrfs_insert_inode_ref(trans, root, "..", 2, objectid, objectid, 0);
1930 btrfs_set_root_dirid(&root->root_item, objectid);
1937 * checks if a path is a block device node
1938 * Returns negative errno on failure, otherwise
1939 * returns 1 for blockdev, 0 for not-blockdev
1941 int is_block_device(const char *path)
1943 struct stat statbuf;
1945 if (stat(path, &statbuf) < 0)
1948 return !!S_ISBLK(statbuf.st_mode);
1952 * check if given path is a mount point
1953 * return 1 if yes. 0 if no. -1 for error
1955 int is_mount_point(const char *path)
1961 f = setmntent("/proc/self/mounts", "r");
1965 while ((mnt = getmntent(f)) != NULL) {
1966 if (strcmp(mnt->mnt_dir, path))
1975 static int is_reg_file(const char *path)
1977 struct stat statbuf;
1979 if (stat(path, &statbuf) < 0)
1981 return S_ISREG(statbuf.st_mode);
1985 * This function checks if the given input parameter is
1987 * return <0 : some error in the given input
1988 * return BTRFS_ARG_UNKNOWN: unknown input
1989 * return BTRFS_ARG_UUID: given input is uuid
1990 * return BTRFS_ARG_MNTPOINT: given input is path
1991 * return BTRFS_ARG_REG: given input is regular file
1992 * return BTRFS_ARG_BLKDEV: given input is block device
1994 int check_arg_type(const char *input)
1997 char path[PATH_MAX];
2002 if (realpath(input, path)) {
2003 if (is_block_device(path) == 1)
2004 return BTRFS_ARG_BLKDEV;
2006 if (is_mount_point(path) == 1)
2007 return BTRFS_ARG_MNTPOINT;
2009 if (is_reg_file(path))
2010 return BTRFS_ARG_REG;
2012 return BTRFS_ARG_UNKNOWN;
2015 if (strlen(input) == (BTRFS_UUID_UNPARSED_SIZE - 1) &&
2016 !uuid_parse(input, uuid))
2017 return BTRFS_ARG_UUID;
2019 return BTRFS_ARG_UNKNOWN;
2023 * Find the mount point for a mounted device.
2024 * On success, returns 0 with mountpoint in *mp.
2025 * On failure, returns -errno (not mounted yields -EINVAL)
2026 * Is noisy on failures, expects to be given a mounted device.
2028 int get_btrfs_mount(const char *dev, char *mp, size_t mp_size)
2033 ret = is_block_device(dev);
2036 error("not a block device: %s", dev);
2039 error("cannot check %s: %s", dev, strerror(-ret));
2044 fd = open(dev, O_RDONLY);
2047 error("cannot open %s: %s", dev, strerror(errno));
2051 ret = check_mounted_where(fd, dev, mp, mp_size, NULL);
2054 } else { /* mounted, all good */
2064 * Given a pathname, return a filehandle to:
2065 * the original pathname or,
2066 * if the pathname is a mounted btrfs device, to its mountpoint.
2068 * On error, return -1, errno should be set.
2070 int open_path_or_dev_mnt(const char *path, DIR **dirstream, int verbose)
2075 if (is_block_device(path)) {
2076 ret = get_btrfs_mount(path, mp, sizeof(mp));
2078 /* not a mounted btrfs dev */
2079 error_on(verbose, "'%s' is not a mounted btrfs device",
2084 ret = open_file_or_dir(mp, dirstream);
2085 error_on(verbose && ret < 0, "can't access '%s': %s",
2086 path, strerror(errno));
2088 ret = btrfs_open_dir(path, dirstream, 1);
2095 * Do the following checks before calling open_file_or_dir():
2096 * 1: path is in a btrfs filesystem
2097 * 2: path is a directory
2099 int btrfs_open_dir(const char *path, DIR **dirstream, int verbose)
2105 if (statfs(path, &stfs) != 0) {
2106 error_on(verbose, "cannot access '%s': %s", path,
2111 if (stfs.f_type != BTRFS_SUPER_MAGIC) {
2112 error_on(verbose, "not a btrfs filesystem: %s", path);
2116 if (stat(path, &st) != 0) {
2117 error_on(verbose, "cannot access '%s': %s", path,
2122 if (!S_ISDIR(st.st_mode)) {
2123 error_on(verbose, "not a directory: %s", path);
2127 ret = open_file_or_dir(path, dirstream);
2129 error_on(verbose, "cannot access '%s': %s", path,
2136 /* checks if a device is a loop device */
2137 static int is_loop_device (const char* device) {
2138 struct stat statbuf;
2140 if(stat(device, &statbuf) < 0)
2143 return (S_ISBLK(statbuf.st_mode) &&
2144 MAJOR(statbuf.st_rdev) == LOOP_MAJOR);
2148 * Takes a loop device path (e.g. /dev/loop0) and returns
2149 * the associated file (e.g. /images/my_btrfs.img) using
2152 static int resolve_loop_device_with_loopdev(const char* loop_dev, char* loop_file)
2156 struct loop_info64 lo64;
2158 fd = open(loop_dev, O_RDONLY | O_NONBLOCK);
2161 ret = ioctl(fd, LOOP_GET_STATUS64, &lo64);
2167 memcpy(loop_file, lo64.lo_file_name, sizeof(lo64.lo_file_name));
2168 loop_file[sizeof(lo64.lo_file_name)] = 0;
2176 /* Takes a loop device path (e.g. /dev/loop0) and returns
2177 * the associated file (e.g. /images/my_btrfs.img) */
2178 static int resolve_loop_device(const char* loop_dev, char* loop_file,
2185 char real_loop_dev[PATH_MAX];
2187 if (!realpath(loop_dev, real_loop_dev))
2189 snprintf(p, PATH_MAX, "/sys/block/%s/loop/backing_file", strrchr(real_loop_dev, '/'));
2190 if (!(f = fopen(p, "r"))) {
2191 if (errno == ENOENT)
2193 * It's possibly a partitioned loop device, which is
2194 * resolvable with loopdev API.
2196 return resolve_loop_device_with_loopdev(loop_dev, loop_file);
2200 snprintf(fmt, 20, "%%%i[^\n]", max_len-1);
2201 ret = fscanf(f, fmt, loop_file);
2210 * Checks whether a and b are identical or device
2211 * files associated with the same block device
2213 static int is_same_blk_file(const char* a, const char* b)
2215 struct stat st_buf_a, st_buf_b;
2216 char real_a[PATH_MAX];
2217 char real_b[PATH_MAX];
2219 if (!realpath(a, real_a))
2220 strncpy_null(real_a, a);
2222 if (!realpath(b, real_b))
2223 strncpy_null(real_b, b);
2225 /* Identical path? */
2226 if (strcmp(real_a, real_b) == 0)
2229 if (stat(a, &st_buf_a) < 0 || stat(b, &st_buf_b) < 0) {
2230 if (errno == ENOENT)
2235 /* Same blockdevice? */
2236 if (S_ISBLK(st_buf_a.st_mode) && S_ISBLK(st_buf_b.st_mode) &&
2237 st_buf_a.st_rdev == st_buf_b.st_rdev) {
2242 if (st_buf_a.st_dev == st_buf_b.st_dev &&
2243 st_buf_a.st_ino == st_buf_b.st_ino) {
2250 /* checks if a and b are identical or device
2251 * files associated with the same block device or
2252 * if one file is a loop device that uses the other
2255 static int is_same_loop_file(const char* a, const char* b)
2257 char res_a[PATH_MAX];
2258 char res_b[PATH_MAX];
2259 const char* final_a = NULL;
2260 const char* final_b = NULL;
2263 /* Resolve a if it is a loop device */
2264 if((ret = is_loop_device(a)) < 0) {
2269 ret = resolve_loop_device(a, res_a, sizeof(res_a));
2280 /* Resolve b if it is a loop device */
2281 if ((ret = is_loop_device(b)) < 0) {
2286 ret = resolve_loop_device(b, res_b, sizeof(res_b));
2297 return is_same_blk_file(final_a, final_b);
2300 /* Checks if a file exists and is a block or regular file*/
2301 static int is_existing_blk_or_reg_file(const char* filename)
2305 if(stat(filename, &st_buf) < 0) {
2312 return (S_ISBLK(st_buf.st_mode) || S_ISREG(st_buf.st_mode));
2315 /* Checks if a file is used (directly or indirectly via a loop device)
2316 * by a device in fs_devices
2318 static int blk_file_in_dev_list(struct btrfs_fs_devices* fs_devices,
2322 struct list_head *head;
2323 struct list_head *cur;
2324 struct btrfs_device *device;
2326 head = &fs_devices->devices;
2327 list_for_each(cur, head) {
2328 device = list_entry(cur, struct btrfs_device, dev_list);
2330 if((ret = is_same_loop_file(device->name, file)))
2338 * Resolve a pathname to a device mapper node to /dev/mapper/<name>
2339 * Returns NULL on invalid input or malloc failure; Other failures
2340 * will be handled by the caller using the input pathame.
2342 char *canonicalize_dm_name(const char *ptname)
2346 char path[PATH_MAX], name[PATH_MAX], *res = NULL;
2348 if (!ptname || !*ptname)
2351 snprintf(path, sizeof(path), "/sys/block/%s/dm/name", ptname);
2352 if (!(f = fopen(path, "r")))
2355 /* read <name>\n from sysfs */
2356 if (fgets(name, sizeof(name), f) && (sz = strlen(name)) > 1) {
2357 name[sz - 1] = '\0';
2358 snprintf(path, sizeof(path), "/dev/mapper/%s", name);
2360 if (access(path, F_OK) == 0)
2368 * Resolve a pathname to a canonical device node, e.g. /dev/sda1 or
2369 * to a device mapper pathname.
2370 * Returns NULL on invalid input or malloc failure; Other failures
2371 * will be handled by the caller using the input pathame.
2373 char *canonicalize_path(const char *path)
2375 char *canonical, *p;
2377 if (!path || !*path)
2380 canonical = realpath(path, NULL);
2382 return strdup(path);
2383 p = strrchr(canonical, '/');
2384 if (p && strncmp(p, "/dm-", 4) == 0 && isdigit(*(p + 4))) {
2385 char *dm = canonicalize_dm_name(p + 1);
2396 * returns 1 if the device was mounted, < 0 on error or 0 if everything
2397 * is safe to continue.
2399 int check_mounted(const char* file)
2404 fd = open(file, O_RDONLY);
2406 error("mount check: cannot open %s: %s", file,
2411 ret = check_mounted_where(fd, file, NULL, 0, NULL);
2417 int check_mounted_where(int fd, const char *file, char *where, int size,
2418 struct btrfs_fs_devices **fs_dev_ret)
2423 struct btrfs_fs_devices *fs_devices_mnt = NULL;
2427 /* scan the initial device */
2428 ret = btrfs_scan_one_device(fd, file, &fs_devices_mnt,
2429 &total_devs, BTRFS_SUPER_INFO_OFFSET, SBREAD_DEFAULT);
2430 is_btrfs = (ret >= 0);
2432 /* scan other devices */
2433 if (is_btrfs && total_devs > 1) {
2434 ret = btrfs_scan_devices();
2439 /* iterate over the list of currently mounted filesystems */
2440 if ((f = setmntent ("/proc/self/mounts", "r")) == NULL)
2443 while ((mnt = getmntent (f)) != NULL) {
2445 if(strcmp(mnt->mnt_type, "btrfs") != 0)
2448 ret = blk_file_in_dev_list(fs_devices_mnt, mnt->mnt_fsname);
2450 /* ignore entries in the mount table that are not
2451 associated with a file*/
2452 if((ret = is_existing_blk_or_reg_file(mnt->mnt_fsname)) < 0)
2453 goto out_mntloop_err;
2457 ret = is_same_loop_file(file, mnt->mnt_fsname);
2461 goto out_mntloop_err;
2466 /* Did we find an entry in mnt table? */
2467 if (mnt && size && where) {
2468 strncpy(where, mnt->mnt_dir, size);
2472 *fs_dev_ret = fs_devices_mnt;
2474 ret = (mnt != NULL);
2482 struct pending_dir {
2483 struct list_head list;
2484 char name[PATH_MAX];
2487 int btrfs_register_one_device(const char *fname)
2489 struct btrfs_ioctl_vol_args args;
2493 fd = open("/dev/btrfs-control", O_RDWR);
2496 "failed to open /dev/btrfs-control, skipping device registration: %s",
2500 memset(&args, 0, sizeof(args));
2501 strncpy_null(args.name, fname);
2502 ret = ioctl(fd, BTRFS_IOC_SCAN_DEV, &args);
2504 error("device scan failed on '%s': %s", fname,
2513 * Register all devices in the fs_uuid list created in the user
2514 * space. Ensure btrfs_scan_devices() is called before this func.
2516 int btrfs_register_all_devices(void)
2520 struct btrfs_fs_devices *fs_devices;
2521 struct btrfs_device *device;
2522 struct list_head *all_uuids;
2524 all_uuids = btrfs_scanned_uuids();
2526 list_for_each_entry(fs_devices, all_uuids, list) {
2527 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2529 err = btrfs_register_one_device(device->name);
2539 int btrfs_device_already_in_root(struct btrfs_root *root, int fd,
2542 struct btrfs_super_block *disk_super;
2546 buf = malloc(BTRFS_SUPER_INFO_SIZE);
2551 ret = pread(fd, buf, BTRFS_SUPER_INFO_SIZE, super_offset);
2552 if (ret != BTRFS_SUPER_INFO_SIZE)
2556 disk_super = (struct btrfs_super_block *)buf;
2558 * Accept devices from the same filesystem, allow partially created
2561 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC &&
2562 btrfs_super_magic(disk_super) != BTRFS_MAGIC_PARTIAL)
2565 if (!memcmp(disk_super->fsid, root->fs_info->super_copy->fsid,
2575 * Note: this function uses a static per-thread buffer. Do not call this
2576 * function more than 10 times within one argument list!
2578 const char *pretty_size_mode(u64 size, unsigned mode)
2580 static __thread int ps_index = 0;
2581 static __thread char ps_array[10][32];
2584 ret = ps_array[ps_index];
2587 (void)pretty_size_snprintf(size, ret, 32, mode);
2592 static const char* unit_suffix_binary[] =
2593 { "B", "KiB", "MiB", "GiB", "TiB", "PiB", "EiB"};
2594 static const char* unit_suffix_decimal[] =
2595 { "B", "kB", "MB", "GB", "TB", "PB", "EB"};
2597 int pretty_size_snprintf(u64 size, char *str, size_t str_size, unsigned unit_mode)
2603 const char** suffix = NULL;
2609 if ((unit_mode & ~UNITS_MODE_MASK) == UNITS_RAW) {
2610 snprintf(str, str_size, "%llu", size);
2614 if ((unit_mode & ~UNITS_MODE_MASK) == UNITS_BINARY) {
2617 suffix = unit_suffix_binary;
2618 } else if ((unit_mode & ~UNITS_MODE_MASK) == UNITS_DECIMAL) {
2621 suffix = unit_suffix_decimal;
2626 fprintf(stderr, "INTERNAL ERROR: unknown unit base, mode %d\n",
2634 switch (unit_mode & UNITS_MODE_MASK) {
2635 case UNITS_TBYTES: base *= mult; num_divs++;
2636 case UNITS_GBYTES: base *= mult; num_divs++;
2637 case UNITS_MBYTES: base *= mult; num_divs++;
2638 case UNITS_KBYTES: num_divs++;
2645 while (size >= mult) {
2651 * If the value is smaller than base, we didn't do any
2652 * division, in that case, base should be 1, not original
2653 * base, or the unit will be wrong
2659 if (num_divs >= ARRAY_SIZE(unit_suffix_binary)) {
2661 printf("INTERNAL ERROR: unsupported unit suffix, index %d\n",
2666 fraction = (float)last_size / base;
2668 return snprintf(str, str_size, "%.2f%s", fraction, suffix[num_divs]);
2672 * __strncpy_null - strncpy with null termination
2673 * @dest: the target array
2674 * @src: the source string
2675 * @n: maximum bytes to copy (size of *dest)
2677 * Like strncpy, but ensures destination is null-terminated.
2679 * Copies the string pointed to by src, including the terminating null
2680 * byte ('\0'), to the buffer pointed to by dest, up to a maximum
2681 * of n bytes. Then ensure that dest is null-terminated.
2683 char *__strncpy_null(char *dest, const char *src, size_t n)
2685 strncpy(dest, src, n);
2692 * Checks to make sure that the label matches our requirements.
2694 0 if everything is safe and usable
2695 -1 if the label is too long
2697 static int check_label(const char *input)
2699 int len = strlen(input);
2701 if (len > BTRFS_LABEL_SIZE - 1) {
2702 error("label %s is too long (max %d)", input,
2703 BTRFS_LABEL_SIZE - 1);
2710 static int set_label_unmounted(const char *dev, const char *label)
2712 struct btrfs_trans_handle *trans;
2713 struct btrfs_root *root;
2716 ret = check_mounted(dev);
2718 error("checking mount status of %s failed: %d", dev, ret);
2722 error("device %s is mounted, use mount point", dev);
2726 /* Open the super_block at the default location
2727 * and as read-write.
2729 root = open_ctree(dev, 0, OPEN_CTREE_WRITES);
2730 if (!root) /* errors are printed by open_ctree() */
2733 trans = btrfs_start_transaction(root, 1);
2734 __strncpy_null(root->fs_info->super_copy->label, label, BTRFS_LABEL_SIZE - 1);
2736 btrfs_commit_transaction(trans, root);
2738 /* Now we close it since we are done. */
2743 static int set_label_mounted(const char *mount_path, const char *labelp)
2746 char label[BTRFS_LABEL_SIZE];
2748 fd = open(mount_path, O_RDONLY | O_NOATIME);
2750 error("unable to access %s: %s", mount_path, strerror(errno));
2754 memset(label, 0, sizeof(label));
2755 __strncpy_null(label, labelp, BTRFS_LABEL_SIZE - 1);
2756 if (ioctl(fd, BTRFS_IOC_SET_FSLABEL, label) < 0) {
2757 error("unable to set label of %s: %s", mount_path,
2767 int get_label_unmounted(const char *dev, char *label)
2769 struct btrfs_root *root;
2772 ret = check_mounted(dev);
2774 error("checking mount status of %s failed: %d", dev, ret);
2778 /* Open the super_block at the default location
2781 root = open_ctree(dev, 0, 0);
2785 __strncpy_null(label, root->fs_info->super_copy->label,
2786 BTRFS_LABEL_SIZE - 1);
2788 /* Now we close it since we are done. */
2794 * If a partition is mounted, try to get the filesystem label via its
2795 * mounted path rather than device. Return the corresponding error
2796 * the user specified the device path.
2798 int get_label_mounted(const char *mount_path, char *labelp)
2800 char label[BTRFS_LABEL_SIZE];
2804 fd = open(mount_path, O_RDONLY | O_NOATIME);
2806 error("unable to access %s: %s", mount_path, strerror(errno));
2810 memset(label, '\0', sizeof(label));
2811 ret = ioctl(fd, BTRFS_IOC_GET_FSLABEL, label);
2813 if (errno != ENOTTY)
2814 error("unable to get label of %s: %s", mount_path,
2821 __strncpy_null(labelp, label, BTRFS_LABEL_SIZE - 1);
2826 int get_label(const char *btrfs_dev, char *label)
2830 ret = is_existing_blk_or_reg_file(btrfs_dev);
2832 ret = get_label_mounted(btrfs_dev, label);
2834 ret = get_label_unmounted(btrfs_dev, label);
2839 int set_label(const char *btrfs_dev, const char *label)
2843 if (check_label(label))
2846 ret = is_existing_blk_or_reg_file(btrfs_dev);
2848 ret = set_label_mounted(btrfs_dev, label);
2850 ret = set_label_unmounted(btrfs_dev, label);
2856 * A not-so-good version fls64. No fascinating optimization since
2857 * no one except parse_size use it
2859 static int fls64(u64 x)
2863 for (i = 0; i <64; i++)
2864 if (x << i & (1ULL << 63))
2869 u64 parse_size(char *s)
2877 error("size value is empty");
2881 error("size value '%s' is less equal than 0", s);
2884 ret = strtoull(s, &endptr, 10);
2886 error("size value '%s' is invalid", s);
2889 if (endptr[0] && endptr[1]) {
2890 error("illegal suffix contains character '%c' in wrong position",
2895 * strtoll returns LLONG_MAX when overflow, if this happens,
2896 * need to call strtoull to get the real size
2898 if (errno == ERANGE && ret == ULLONG_MAX) {
2899 error("size value '%s' is too large for u64", s);
2903 c = tolower(endptr[0]);
2926 error("unknown size descriptor '%c'", c);
2930 /* Check whether ret * mult overflow */
2931 if (fls64(ret) + fls64(mult) - 1 > 64) {
2932 error("size value '%s' is too large for u64", s);
2939 u64 parse_qgroupid(const char *p)
2941 char *s = strchr(p, '/');
2942 const char *ptr_src_end = p + strlen(p);
2943 char *ptr_parse_end = NULL;
2952 /* Numeric format like '0/257' is the primary case */
2954 id = strtoull(p, &ptr_parse_end, 10);
2955 if (ptr_parse_end != ptr_src_end)
2959 level = strtoull(p, &ptr_parse_end, 10);
2960 if (ptr_parse_end != s)
2963 id = strtoull(s + 1, &ptr_parse_end, 10);
2964 if (ptr_parse_end != ptr_src_end)
2967 return (level << BTRFS_QGROUP_LEVEL_SHIFT) | id;
2970 /* Path format like subv at 'my_subvol' is the fallback case */
2971 ret = test_issubvolume(p);
2972 if (ret < 0 || !ret)
2974 fd = open(p, O_RDONLY);
2977 ret = lookup_path_rootid(fd, &id);
2979 error("failed to lookup root id: %s", strerror(-ret));
2986 error("invalid qgroupid or subvolume path: %s", p);
2990 int open_file_or_dir3(const char *fname, DIR **dirstream, int open_flags)
2996 ret = stat(fname, &st);
3000 if (S_ISDIR(st.st_mode)) {
3001 *dirstream = opendir(fname);
3004 fd = dirfd(*dirstream);
3005 } else if (S_ISREG(st.st_mode) || S_ISLNK(st.st_mode)) {
3006 fd = open(fname, open_flags);
3009 * we set this on purpose, in case the caller output
3010 * strerror(errno) as success
3018 closedir(*dirstream);
3025 int open_file_or_dir(const char *fname, DIR **dirstream)
3027 return open_file_or_dir3(fname, dirstream, O_RDWR);
3030 void close_file_or_dir(int fd, DIR *dirstream)
3033 closedir(dirstream);
3038 int get_device_info(int fd, u64 devid,
3039 struct btrfs_ioctl_dev_info_args *di_args)
3043 di_args->devid = devid;
3044 memset(&di_args->uuid, '\0', sizeof(di_args->uuid));
3046 ret = ioctl(fd, BTRFS_IOC_DEV_INFO, di_args);
3047 return ret < 0 ? -errno : 0;
3050 static u64 find_max_device_id(struct btrfs_ioctl_search_args *search_args,
3053 struct btrfs_dev_item *dev_item;
3054 char *buf = search_args->buf;
3056 buf += (nr_items - 1) * (sizeof(struct btrfs_ioctl_search_header)
3057 + sizeof(struct btrfs_dev_item));
3058 buf += sizeof(struct btrfs_ioctl_search_header);
3060 dev_item = (struct btrfs_dev_item *)buf;
3062 return btrfs_stack_device_id(dev_item);
3065 static int search_chunk_tree_for_fs_info(int fd,
3066 struct btrfs_ioctl_fs_info_args *fi_args)
3070 u64 start_devid = 1;
3071 struct btrfs_ioctl_search_args search_args;
3072 struct btrfs_ioctl_search_key *search_key = &search_args.key;
3074 fi_args->num_devices = 0;
3076 max_items = BTRFS_SEARCH_ARGS_BUFSIZE
3077 / (sizeof(struct btrfs_ioctl_search_header)
3078 + sizeof(struct btrfs_dev_item));
3080 search_key->tree_id = BTRFS_CHUNK_TREE_OBJECTID;
3081 search_key->min_objectid = BTRFS_DEV_ITEMS_OBJECTID;
3082 search_key->max_objectid = BTRFS_DEV_ITEMS_OBJECTID;
3083 search_key->min_type = BTRFS_DEV_ITEM_KEY;
3084 search_key->max_type = BTRFS_DEV_ITEM_KEY;
3085 search_key->min_transid = 0;
3086 search_key->max_transid = (u64)-1;
3087 search_key->nr_items = max_items;
3088 search_key->max_offset = (u64)-1;
3091 search_key->min_offset = start_devid;
3093 ret = ioctl(fd, BTRFS_IOC_TREE_SEARCH, &search_args);
3097 fi_args->num_devices += (u64)search_key->nr_items;
3099 if (search_key->nr_items == max_items) {
3100 start_devid = find_max_device_id(&search_args,
3101 search_key->nr_items) + 1;
3105 /* get the lastest max_id to stay consistent with the num_devices */
3106 if (search_key->nr_items == 0)
3108 * last tree_search returns an empty buf, use the devid of
3109 * the last dev_item of the previous tree_search
3111 fi_args->max_id = start_devid - 1;
3113 fi_args->max_id = find_max_device_id(&search_args,
3114 search_key->nr_items);
3120 * For a given path, fill in the ioctl fs_ and info_ args.
3121 * If the path is a btrfs mountpoint, fill info for all devices.
3122 * If the path is a btrfs device, fill in only that device.
3124 * The path provided must be either on a mounted btrfs fs,
3125 * or be a mounted btrfs device.
3127 * Returns 0 on success, or a negative errno.
3129 int get_fs_info(const char *path, struct btrfs_ioctl_fs_info_args *fi_args,
3130 struct btrfs_ioctl_dev_info_args **di_ret)
3137 struct btrfs_fs_devices *fs_devices_mnt = NULL;
3138 struct btrfs_ioctl_dev_info_args *di_args;
3139 struct btrfs_ioctl_dev_info_args tmp;
3141 DIR *dirstream = NULL;
3143 memset(fi_args, 0, sizeof(*fi_args));
3145 if (is_block_device(path) == 1) {
3146 struct btrfs_super_block *disk_super;
3147 char buf[BTRFS_SUPER_INFO_SIZE];
3149 /* Ensure it's mounted, then set path to the mountpoint */
3150 fd = open(path, O_RDONLY);
3153 error("cannot open %s: %s", path, strerror(errno));
3156 ret = check_mounted_where(fd, path, mp, sizeof(mp),
3165 /* Only fill in this one device */
3166 fi_args->num_devices = 1;
3168 disk_super = (struct btrfs_super_block *)buf;
3169 ret = btrfs_read_dev_super(fd, disk_super,
3170 BTRFS_SUPER_INFO_OFFSET, 0);
3175 last_devid = btrfs_stack_device_id(&disk_super->dev_item);
3176 fi_args->max_id = last_devid;
3178 memcpy(fi_args->fsid, fs_devices_mnt->fsid, BTRFS_FSID_SIZE);
3182 /* at this point path must not be for a block device */
3183 fd = open_file_or_dir(path, &dirstream);
3189 /* fill in fi_args if not just a single device */
3190 if (fi_args->num_devices != 1) {
3191 ret = ioctl(fd, BTRFS_IOC_FS_INFO, fi_args);
3198 * The fs_args->num_devices does not include seed devices
3200 ret = search_chunk_tree_for_fs_info(fd, fi_args);
3205 * search_chunk_tree_for_fs_info() will lacks the devid 0
3206 * so manual probe for it here.
3208 ret = get_device_info(fd, 0, &tmp);
3210 fi_args->num_devices++;
3213 if (last_devid == 0)
3218 if (!fi_args->num_devices)
3221 di_args = *di_ret = malloc((fi_args->num_devices) * sizeof(*di_args));
3228 memcpy(di_args, &tmp, sizeof(tmp));
3229 for (; last_devid <= fi_args->max_id; last_devid++) {
3230 ret = get_device_info(fd, last_devid, &di_args[ndevs]);
3239 * only when the only dev we wanted to find is not there then
3240 * let any error be returned
3242 if (fi_args->num_devices != 1) {
3248 close_file_or_dir(fd, dirstream);
3252 #define isoctal(c) (((c) & ~7) == '0')
3254 static inline void translate(char *f, char *t)
3256 while (*f != '\0') {
3258 isoctal(f[1]) && isoctal(f[2]) && isoctal(f[3])) {
3259 *t++ = 64*(f[1] & 7) + 8*(f[2] & 7) + (f[3] & 7);
3269 * Checks if the swap device.
3270 * Returns 1 if swap device, < 0 on error or 0 if not swap device.
3272 static int is_swap_device(const char *file)
3283 if (stat(file, &st_buf) < 0)
3285 if (S_ISBLK(st_buf.st_mode))
3286 dev = st_buf.st_rdev;
3287 else if (S_ISREG(st_buf.st_mode)) {
3288 dev = st_buf.st_dev;
3289 ino = st_buf.st_ino;
3293 if ((f = fopen("/proc/swaps", "r")) == NULL)
3296 /* skip the first line */
3297 if (fgets(tmp, sizeof(tmp), f) == NULL)
3300 while (fgets(tmp, sizeof(tmp), f) != NULL) {
3301 if ((cp = strchr(tmp, ' ')) != NULL)
3303 if ((cp = strchr(tmp, '\t')) != NULL)
3305 translate(tmp, buf);
3306 if (stat(buf, &st_buf) != 0)
3308 if (S_ISBLK(st_buf.st_mode)) {
3309 if (dev == st_buf.st_rdev) {
3313 } else if (S_ISREG(st_buf.st_mode)) {
3314 if (dev == st_buf.st_dev && ino == st_buf.st_ino) {
3328 * Check for existing filesystem or partition table on device.
3330 * 1 for existing fs or partition
3331 * 0 for nothing found
3332 * -1 for internal error
3334 static int check_overwrite(const char *device)
3337 blkid_probe pr = NULL;
3341 if (!device || !*device)
3344 ret = -1; /* will reset on success of all setup calls */
3346 pr = blkid_new_probe_from_filename(device);
3350 size = blkid_probe_get_size(pr);
3354 /* nothing to overwrite on a 0-length device */
3360 ret = blkid_probe_enable_partitions(pr, 1);
3364 ret = blkid_do_fullprobe(pr);
3369 * Blkid returns 1 for nothing found and 0 when it finds a signature,
3370 * but we want the exact opposite, so reverse the return value here.
3372 * In addition print some useful diagnostics about what actually is
3380 if (!blkid_probe_lookup_value(pr, "TYPE", &type, NULL)) {
3382 "%s appears to contain an existing "
3383 "filesystem (%s).\n", device, type);
3384 } else if (!blkid_probe_lookup_value(pr, "PTTYPE", &type, NULL)) {
3386 "%s appears to contain a partition "
3387 "table (%s).\n", device, type);
3390 "%s appears to contain something weird "
3391 "according to blkid\n", device);
3397 blkid_free_probe(pr);
3400 "probe of %s failed, cannot detect "
3401 "existing filesystem.\n", device);
3405 static int group_profile_devs_min(u64 flag)
3407 switch (flag & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3408 case 0: /* single */
3409 case BTRFS_BLOCK_GROUP_DUP:
3411 case BTRFS_BLOCK_GROUP_RAID0:
3412 case BTRFS_BLOCK_GROUP_RAID1:
3413 case BTRFS_BLOCK_GROUP_RAID5:
3415 case BTRFS_BLOCK_GROUP_RAID6:
3417 case BTRFS_BLOCK_GROUP_RAID10:
3424 int test_num_disk_vs_raid(u64 metadata_profile, u64 data_profile,
3425 u64 dev_cnt, int mixed, int ssd)
3428 u64 profile = metadata_profile | data_profile;
3433 allowed |= BTRFS_BLOCK_GROUP_RAID10;
3435 allowed |= BTRFS_BLOCK_GROUP_RAID6;
3437 allowed |= BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3438 BTRFS_BLOCK_GROUP_RAID5;
3440 allowed |= BTRFS_BLOCK_GROUP_DUP;
3443 if (dev_cnt > 1 && profile & BTRFS_BLOCK_GROUP_DUP) {
3444 warning("DUP is not recommended on filesystem with multiple devices");
3446 if (metadata_profile & ~allowed) {
3448 "ERROR: unable to create FS with metadata profile %s "
3449 "(have %llu devices but %d devices are required)\n",
3450 btrfs_group_profile_str(metadata_profile), dev_cnt,
3451 group_profile_devs_min(metadata_profile));
3454 if (data_profile & ~allowed) {
3456 "ERROR: unable to create FS with data profile %s "
3457 "(have %llu devices but %d devices are required)\n",
3458 btrfs_group_profile_str(data_profile), dev_cnt,
3459 group_profile_devs_min(data_profile));
3463 if (dev_cnt == 3 && profile & BTRFS_BLOCK_GROUP_RAID6) {
3464 warning("RAID6 is not recommended on filesystem with 3 devices only");
3466 if (dev_cnt == 2 && profile & BTRFS_BLOCK_GROUP_RAID5) {
3467 warning("RAID5 is not recommended on filesystem with 2 devices only");
3469 warning_on(!mixed && (data_profile & BTRFS_BLOCK_GROUP_DUP) && ssd,
3470 "DUP may not actually lead to 2 copies on the device, see manual page");
3475 int group_profile_max_safe_loss(u64 flags)
3477 switch (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3478 case 0: /* single */
3479 case BTRFS_BLOCK_GROUP_DUP:
3480 case BTRFS_BLOCK_GROUP_RAID0:
3482 case BTRFS_BLOCK_GROUP_RAID1:
3483 case BTRFS_BLOCK_GROUP_RAID5:
3484 case BTRFS_BLOCK_GROUP_RAID10:
3486 case BTRFS_BLOCK_GROUP_RAID6:
3494 * Check if a device is suitable for btrfs
3496 * 1: something is wrong, an error is printed
3499 int test_dev_for_mkfs(const char *file, int force_overwrite)
3504 ret = is_swap_device(file);
3506 error("checking status of %s: %s", file, strerror(-ret));
3510 error("%s is a swap device", file);
3513 if (!force_overwrite) {
3514 if (check_overwrite(file)) {
3515 error("use the -f option to force overwrite of %s",
3520 ret = check_mounted(file);
3522 error("cannot check mount status of %s: %s", file,
3527 error("%s is mounted", file);
3530 /* check if the device is busy */
3531 fd = open(file, O_RDWR|O_EXCL);
3533 error("unable to open %s: %s", file, strerror(errno));
3536 if (fstat(fd, &st)) {
3537 error("unable to stat %s: %s", file, strerror(errno));
3541 if (!S_ISBLK(st.st_mode)) {
3542 error("%s is not a block device", file);
3550 int btrfs_scan_devices(void)
3555 struct btrfs_fs_devices *tmp_devices;
3556 blkid_dev_iterate iter = NULL;
3557 blkid_dev dev = NULL;
3558 blkid_cache cache = NULL;
3559 char path[PATH_MAX];
3561 if (btrfs_scan_done)
3564 if (blkid_get_cache(&cache, NULL) < 0) {
3565 error("blkid cache get failed");
3568 blkid_probe_all(cache);
3569 iter = blkid_dev_iterate_begin(cache);
3570 blkid_dev_set_search(iter, "TYPE", "btrfs");
3571 while (blkid_dev_next(iter, &dev) == 0) {
3572 dev = blkid_verify(cache, dev);
3575 /* if we are here its definitely a btrfs disk*/
3576 strncpy_null(path, blkid_dev_devname(dev));
3578 fd = open(path, O_RDONLY);
3580 error("cannot open %s: %s", path, strerror(errno));
3583 ret = btrfs_scan_one_device(fd, path, &tmp_devices,
3584 &num_devices, BTRFS_SUPER_INFO_OFFSET,
3587 error("cannot scan %s: %s", path, strerror(-ret));
3594 blkid_dev_iterate_end(iter);
3595 blkid_put_cache(cache);
3597 btrfs_scan_done = 1;
3602 int is_vol_small(const char *file)
3609 fd = open(file, O_RDONLY);
3612 if (fstat(fd, &st) < 0) {
3617 size = btrfs_device_size(fd, &st);
3622 if (size < BTRFS_MKFS_SMALL_VOLUME_SIZE) {
3632 * This reads a line from the stdin and only returns non-zero if the
3633 * first whitespace delimited token is a case insensitive match with yes
3636 int ask_user(const char *question)
3638 char buf[30] = {0,};
3639 char *saveptr = NULL;
3642 printf("%s [y/N]: ", question);
3644 return fgets(buf, sizeof(buf) - 1, stdin) &&
3645 (answer = strtok_r(buf, " \t\n\r", &saveptr)) &&
3646 (!strcasecmp(answer, "yes") || !strcasecmp(answer, "y"));
3651 * - file or directory return the containing tree root id
3652 * - subvolume return its own tree id
3653 * - BTRFS_EMPTY_SUBVOL_DIR_OBJECTID (directory with ino == 2) the result is
3654 * undefined and function returns -1
3656 int lookup_path_rootid(int fd, u64 *rootid)
3658 struct btrfs_ioctl_ino_lookup_args args;
3661 memset(&args, 0, sizeof(args));
3663 args.objectid = BTRFS_FIRST_FREE_OBJECTID;
3665 ret = ioctl(fd, BTRFS_IOC_INO_LOOKUP, &args);
3669 *rootid = args.treeid;
3675 * return 0 if a btrfs mount point is found
3676 * return 1 if a mount point is found but not btrfs
3677 * return <0 if something goes wrong
3679 int find_mount_root(const char *path, char **mount_root)
3687 int longest_matchlen = 0;
3688 char *longest_match = NULL;
3690 fd = open(path, O_RDONLY | O_NOATIME);
3695 mnttab = setmntent("/proc/self/mounts", "r");
3699 while ((ent = getmntent(mnttab))) {
3700 len = strlen(ent->mnt_dir);
3701 if (strncmp(ent->mnt_dir, path, len) == 0) {
3702 /* match found and use the latest match */
3703 if (longest_matchlen <= len) {
3704 free(longest_match);
3705 longest_matchlen = len;
3706 longest_match = strdup(ent->mnt_dir);
3707 not_btrfs = strcmp(ent->mnt_type, "btrfs");
3716 free(longest_match);
3721 *mount_root = realpath(longest_match, NULL);
3725 free(longest_match);
3729 int test_minimum_size(const char *file, u32 nodesize)
3732 struct stat statbuf;
3734 fd = open(file, O_RDONLY);
3737 if (stat(file, &statbuf) < 0) {
3741 if (btrfs_device_size(fd, &statbuf) < btrfs_min_dev_size(nodesize)) {
3751 * Test if path is a directory
3753 * 0 - path exists but it is not a directory
3754 * 1 - path exists and it is a directory
3757 int test_isdir(const char *path)
3762 ret = stat(path, &st);
3766 return !!S_ISDIR(st.st_mode);
3769 void units_set_mode(unsigned *units, unsigned mode)
3771 unsigned base = *units & UNITS_MODE_MASK;
3773 *units = base | mode;
3776 void units_set_base(unsigned *units, unsigned base)
3778 unsigned mode = *units & ~UNITS_MODE_MASK;
3780 *units = base | mode;
3783 int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
3787 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3788 if (!path->nodes[level])
3790 if (path->slots[level] + 1 >=
3791 btrfs_header_nritems(path->nodes[level]))
3794 btrfs_item_key_to_cpu(path->nodes[level], key,
3795 path->slots[level] + 1);
3797 btrfs_node_key_to_cpu(path->nodes[level], key,
3798 path->slots[level] + 1);
3804 const char* btrfs_group_type_str(u64 flag)
3806 u64 mask = BTRFS_BLOCK_GROUP_TYPE_MASK |
3807 BTRFS_SPACE_INFO_GLOBAL_RSV;
3809 switch (flag & mask) {
3810 case BTRFS_BLOCK_GROUP_DATA:
3812 case BTRFS_BLOCK_GROUP_SYSTEM:
3814 case BTRFS_BLOCK_GROUP_METADATA:
3816 case BTRFS_BLOCK_GROUP_DATA|BTRFS_BLOCK_GROUP_METADATA:
3817 return "Data+Metadata";
3818 case BTRFS_SPACE_INFO_GLOBAL_RSV:
3819 return "GlobalReserve";
3825 const char* btrfs_group_profile_str(u64 flag)
3827 switch (flag & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3830 case BTRFS_BLOCK_GROUP_RAID0:
3832 case BTRFS_BLOCK_GROUP_RAID1:
3834 case BTRFS_BLOCK_GROUP_RAID5:
3836 case BTRFS_BLOCK_GROUP_RAID6:
3838 case BTRFS_BLOCK_GROUP_DUP:
3840 case BTRFS_BLOCK_GROUP_RAID10:
3847 u64 disk_size(const char *path)
3851 if (statfs(path, &sfs) < 0)
3854 return sfs.f_bsize * sfs.f_blocks;
3857 u64 get_partition_size(const char *dev)
3860 int fd = open(dev, O_RDONLY);
3864 if (ioctl(fd, BLKGETSIZE64, &result) < 0) {
3874 * Check if the BTRFS_IOC_TREE_SEARCH_V2 ioctl is supported on a given
3875 * filesystem, opened at fd
3877 int btrfs_tree_search2_ioctl_supported(int fd)
3879 struct btrfs_ioctl_search_args_v2 *args2;
3880 struct btrfs_ioctl_search_key *sk;
3881 int args2_size = 1024;
3882 char args2_buf[args2_size];
3885 args2 = (struct btrfs_ioctl_search_args_v2 *)args2_buf;
3889 * Search for the extent tree item in the root tree.
3891 sk->tree_id = BTRFS_ROOT_TREE_OBJECTID;
3892 sk->min_objectid = BTRFS_EXTENT_TREE_OBJECTID;
3893 sk->max_objectid = BTRFS_EXTENT_TREE_OBJECTID;
3894 sk->min_type = BTRFS_ROOT_ITEM_KEY;
3895 sk->max_type = BTRFS_ROOT_ITEM_KEY;
3897 sk->max_offset = (u64)-1;
3898 sk->min_transid = 0;
3899 sk->max_transid = (u64)-1;
3901 args2->buf_size = args2_size - sizeof(struct btrfs_ioctl_search_args_v2);
3902 ret = ioctl(fd, BTRFS_IOC_TREE_SEARCH_V2, args2);
3903 if (ret == -EOPNOTSUPP)
3910 int btrfs_check_nodesize(u32 nodesize, u32 sectorsize, u64 features)
3912 if (nodesize < sectorsize) {
3913 error("illegal nodesize %u (smaller than %u)",
3914 nodesize, sectorsize);
3916 } else if (nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3917 error("illegal nodesize %u (larger than %u)",
3918 nodesize, BTRFS_MAX_METADATA_BLOCKSIZE);
3920 } else if (nodesize & (sectorsize - 1)) {
3921 error("illegal nodesize %u (not aligned to %u)",
3922 nodesize, sectorsize);
3924 } else if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS &&
3925 nodesize != sectorsize) {
3926 error("illegal nodesize %u (not equal to %u for mixed block group)",
3927 nodesize, sectorsize);
3934 * Copy a path argument from SRC to DEST and check the SRC length if it's at
3935 * most PATH_MAX and fits into DEST. DESTLEN is supposed to be exact size of
3937 * The destination buffer is zero terminated.
3938 * Return < 0 for error, 0 otherwise.
3940 int arg_copy_path(char *dest, const char *src, int destlen)
3942 size_t len = strlen(src);
3944 if (len >= PATH_MAX || len >= destlen)
3945 return -ENAMETOOLONG;
3947 __strncpy_null(dest, src, destlen);
3952 unsigned int get_unit_mode_from_arg(int *argc, char *argv[], int df_mode)
3954 unsigned int unit_mode = UNITS_DEFAULT;
3958 for (arg_i = 0; arg_i < *argc; arg_i++) {
3959 if (!strcmp(argv[arg_i], "--"))
3962 if (!strcmp(argv[arg_i], "--raw")) {
3963 unit_mode = UNITS_RAW;
3967 if (!strcmp(argv[arg_i], "--human-readable")) {
3968 unit_mode = UNITS_HUMAN_BINARY;
3973 if (!strcmp(argv[arg_i], "--iec")) {
3974 units_set_mode(&unit_mode, UNITS_BINARY);
3978 if (!strcmp(argv[arg_i], "--si")) {
3979 units_set_mode(&unit_mode, UNITS_DECIMAL);
3984 if (!strcmp(argv[arg_i], "--kbytes")) {
3985 units_set_base(&unit_mode, UNITS_KBYTES);
3989 if (!strcmp(argv[arg_i], "--mbytes")) {
3990 units_set_base(&unit_mode, UNITS_MBYTES);
3994 if (!strcmp(argv[arg_i], "--gbytes")) {
3995 units_set_base(&unit_mode, UNITS_GBYTES);
3999 if (!strcmp(argv[arg_i], "--tbytes")) {
4000 units_set_base(&unit_mode, UNITS_TBYTES);
4008 if (!strcmp(argv[arg_i], "-b")) {
4009 unit_mode = UNITS_RAW;
4013 if (!strcmp(argv[arg_i], "-h")) {
4014 unit_mode = UNITS_HUMAN_BINARY;
4018 if (!strcmp(argv[arg_i], "-H")) {
4019 unit_mode = UNITS_HUMAN_DECIMAL;
4023 if (!strcmp(argv[arg_i], "-k")) {
4024 units_set_base(&unit_mode, UNITS_KBYTES);
4028 if (!strcmp(argv[arg_i], "-m")) {
4029 units_set_base(&unit_mode, UNITS_MBYTES);
4033 if (!strcmp(argv[arg_i], "-g")) {
4034 units_set_base(&unit_mode, UNITS_GBYTES);
4038 if (!strcmp(argv[arg_i], "-t")) {
4039 units_set_base(&unit_mode, UNITS_TBYTES);
4045 for (arg_i = 0, arg_end = 0; arg_i < *argc; arg_i++) {
4048 argv[arg_end] = argv[arg_i];
4057 int string_is_numerical(const char *str)
4061 if (!(*str >= '0' && *str <= '9'))
4063 while (*str >= '0' && *str <= '9')
4071 * Preprocess @argv with getopt_long to reorder options and consume the "--"
4073 * Unknown short and long options are reported, optionally the @usage is printed
4076 void clean_args_no_options(int argc, char *argv[], const char * const *usagestr)
4078 static const struct option long_options[] = {
4083 int c = getopt_long(argc, argv, "", long_options, NULL);
4097 * Same as clean_args_no_options but pass through arguments that could look
4098 * like short options. Eg. reisze which takes a negative resize argument like
4101 * This accepts only two forms:
4102 * - "-- option1 option2 ..."
4103 * - "option1 option2 ..."
4105 void clean_args_no_options_relaxed(int argc, char *argv[], const char * const *usagestr)
4110 if (strcmp(argv[1], "--") == 0)
4114 /* Subvolume helper functions */
4116 * test if name is a correct subvolume name
4117 * this function return
4118 * 0-> name is not a correct subvolume name
4119 * 1-> name is a correct subvolume name
4121 int test_issubvolname(const char *name)
4123 return name[0] != '\0' && !strchr(name, '/') &&
4124 strcmp(name, ".") && strcmp(name, "..");
4128 * Test if path is a subvolume
4130 * 0 - path exists but it is not a subvolume
4131 * 1 - path exists and it is a subvolume
4134 int test_issubvolume(const char *path)
4140 res = stat(path, &st);
4144 if (st.st_ino != BTRFS_FIRST_FREE_OBJECTID || !S_ISDIR(st.st_mode))
4147 res = statfs(path, &stfs);
4151 return (int)stfs.f_type == BTRFS_SUPER_MAGIC;
4154 const char *subvol_strip_mountpoint(const char *mnt, const char *full_path)
4156 int len = strlen(mnt);
4160 if (mnt[len - 1] != '/')
4163 return full_path + len;
4170 * 1: Error; and error info printed to the terminal. Fixme.
4171 * 2: If the fullpath is root tree instead of subvol tree
4173 int get_subvol_info(const char *fullpath, struct root_info *get_ri)
4180 const char *svpath = NULL;
4181 DIR *dirstream1 = NULL;
4182 DIR *dirstream2 = NULL;
4184 ret = test_issubvolume(fullpath);
4188 error("not a subvolume: %s", fullpath);
4192 ret = find_mount_root(fullpath, &mnt);
4196 error("%s doesn't belong to btrfs mount point", fullpath);
4200 svpath = subvol_strip_mountpoint(mnt, fullpath);
4202 fd = btrfs_open_dir(fullpath, &dirstream1, 1);
4206 ret = btrfs_list_get_path_rootid(fd, &sv_id);
4210 mntfd = btrfs_open_dir(mnt, &dirstream2, 1);
4214 memset(get_ri, 0, sizeof(*get_ri));
4215 get_ri->root_id = sv_id;
4217 if (sv_id == BTRFS_FS_TREE_OBJECTID)
4218 ret = btrfs_get_toplevel_subvol(mntfd, get_ri);
4220 ret = btrfs_get_subvol(mntfd, get_ri);
4222 error("can't find '%s': %d", svpath, ret);
4225 close_file_or_dir(mntfd, dirstream2);
4226 close_file_or_dir(fd, dirstream1);
4232 void init_rand_seed(u64 seed)
4236 /* only use the last 48 bits */
4237 for (i = 0; i < 3; i++) {
4238 rand_seed[i] = (unsigned short)(seed ^ (unsigned short)(-1));
4241 rand_seed_initlized = 1;
4244 static void __init_seed(void)
4250 if(rand_seed_initlized)
4252 /* Use urandom as primary seed source. */
4253 fd = open("/dev/urandom", O_RDONLY);
4255 ret = read(fd, rand_seed, sizeof(rand_seed));
4257 if (ret < sizeof(rand_seed))
4261 /* Use time and pid as fallback seed */
4262 warning("failed to read /dev/urandom, use time and pid as random seed");
4263 gettimeofday(&tv, 0);
4264 rand_seed[0] = getpid() ^ (tv.tv_sec & 0xFFFF);
4265 rand_seed[1] = getppid() ^ (tv.tv_usec & 0xFFFF);
4266 rand_seed[2] = (tv.tv_sec ^ tv.tv_usec) >> 16;
4268 rand_seed_initlized = 1;
4275 * Don't use nrand48, its range is [0,2^31) The highest bit will alwasy
4276 * be 0. Use jrand48 to include the highest bit.
4278 return (u32)jrand48(rand_seed);
4281 unsigned int rand_range(unsigned int upper)
4285 * Use the full 48bits to mod, which would be more uniformly
4288 return (unsigned int)(jrand48(rand_seed) % upper);