9 #include <sys/sysmacros.h>
10 #include <sys/types.h>
14 #include <linux/fiemap.h>
16 #include <linux/stat.h>
18 #include <uuid/uuid.h>
22 #include "libbcachefs.h"
24 #include <linux/dcache.h>
25 #include <linux/generic-radix-tree.h>
26 #include <linux/xattr.h>
27 #include "libbcachefs/bcachefs.h"
28 #include "libbcachefs/alloc_background.h"
29 #include "libbcachefs/alloc_foreground.h"
30 #include "libbcachefs/btree_update.h"
31 #include "libbcachefs/buckets.h"
32 #include "libbcachefs/dirent.h"
33 #include "libbcachefs/errcode.h"
34 #include "libbcachefs/fs-common.h"
35 #include "libbcachefs/inode.h"
36 #include "libbcachefs/io_write.h"
37 #include "libbcachefs/replicas.h"
38 #include "libbcachefs/str_hash.h"
39 #include "libbcachefs/super.h"
40 #include "libbcachefs/xattr.h"
42 /* XXX cut and pasted from fsck.c */
43 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
45 static char *dev_t_to_path(dev_t dev)
47 char link[PATH_MAX], *p;
50 char *sysfs_dev = mprintf("/sys/dev/block/%u:%u",
51 major(dev), minor(dev));
52 ret = readlink(sysfs_dev, link, sizeof(link));
55 if (ret < 0 || ret >= sizeof(link))
56 die("readlink error while looking up block device: %m");
60 p = strrchr(link, '/');
62 die("error looking up device name");
65 return mprintf("/dev/%s", p);
68 static bool path_is_fs_root(const char *path)
70 char *line = NULL, *p, *mount;
75 f = fopen("/proc/self/mountinfo", "r");
77 die("Error getting mount information");
79 while (getline(&line, &n, f) != -1) {
82 strsep(&p, " "); /* mount id */
83 strsep(&p, " "); /* parent id */
84 strsep(&p, " "); /* dev */
85 strsep(&p, " "); /* root */
86 mount = strsep(&p, " ");
89 if (mount && !strcmp(path, mount))
100 static void mark_unreserved_space(struct bch_fs *c, ranges extents)
102 struct bch_dev *ca = c->devs[0];
103 struct hole_iter iter;
106 for_each_hole(iter, extents, bucket_to_sector(ca, ca->mi.nbuckets) << 9, i) {
109 if (i.start == i.end)
112 b = sector_to_bucket(ca, i.start >> 9);
114 set_bit(b, ca->buckets_nouse);
116 } while (bucket_to_sector(ca, b) << 9 < i.end);
120 static void update_inode(struct bch_fs *c,
121 struct bch_inode_unpacked *inode)
123 struct bkey_inode_buf packed;
126 bch2_inode_pack(&packed, inode);
127 packed.inode.k.p.snapshot = U32_MAX;
128 ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed.inode.k_i,
131 die("error updating inode: %s", bch2_err_str(ret));
134 static void create_link(struct bch_fs *c,
135 struct bch_inode_unpacked *parent,
136 const char *name, u64 inum, mode_t mode)
138 struct qstr qstr = QSTR(name);
139 struct bch_inode_unpacked parent_u;
140 struct bch_inode_unpacked inode;
142 int ret = bch2_trans_do(c, NULL, NULL, 0,
143 bch2_link_trans(trans,
144 (subvol_inum) { 1, parent->bi_inum }, &parent_u,
145 (subvol_inum) { 1, inum }, &inode, &qstr));
147 die("error creating hardlink: %s", bch2_err_str(ret));
150 static struct bch_inode_unpacked create_file(struct bch_fs *c,
151 struct bch_inode_unpacked *parent,
153 uid_t uid, gid_t gid,
154 mode_t mode, dev_t rdev)
156 struct qstr qstr = QSTR(name);
157 struct bch_inode_unpacked new_inode;
159 bch2_inode_init_early(c, &new_inode);
161 int ret = bch2_trans_do(c, NULL, NULL, 0,
162 bch2_create_trans(trans,
163 (subvol_inum) { 1, parent->bi_inum }, parent,
165 uid, gid, mode, rdev, NULL, NULL,
166 (subvol_inum) {}, 0));
168 die("error creating %s: %s", name, bch2_err_str(ret));
173 #define for_each_xattr_handler(handlers, handler) \
175 for ((handler) = *(handlers)++; \
177 (handler) = *(handlers)++)
179 static const struct xattr_handler *xattr_resolve_name(char **name)
181 const struct xattr_handler **handlers = bch2_xattr_handlers;
182 const struct xattr_handler *handler;
184 for_each_xattr_handler(handlers, handler) {
187 n = strcmp_prefix(*name, xattr_prefix(handler));
189 if (!handler->prefix ^ !*n) {
192 return ERR_PTR(-EINVAL);
198 return ERR_PTR(-EOPNOTSUPP);
201 static void copy_times(struct bch_fs *c, struct bch_inode_unpacked *dst,
204 dst->bi_atime = timespec_to_bch2_time(c, src->st_atim);
205 dst->bi_mtime = timespec_to_bch2_time(c, src->st_mtim);
206 dst->bi_ctime = timespec_to_bch2_time(c, src->st_ctim);
209 static void copy_xattrs(struct bch_fs *c, struct bch_inode_unpacked *dst,
212 struct bch_hash_info hash_info = bch2_hash_info_init(c, dst);
214 char attrs[XATTR_LIST_MAX];
215 ssize_t attrs_size = llistxattr(src, attrs, sizeof(attrs));
217 die("listxattr error: %m");
221 attr < attrs + attrs_size;
223 next = attr + strlen(attr) + 1;
225 char val[XATTR_SIZE_MAX];
226 ssize_t val_size = lgetxattr(src, attr, val, sizeof(val));
229 die("error getting xattr val: %m");
231 const struct xattr_handler *h = xattr_resolve_name(&attr);
232 struct bch_inode_unpacked inode_u;
234 int ret = bch2_trans_do(c, NULL, NULL, 0,
235 bch2_xattr_set(trans,
236 (subvol_inum) { 1, dst->bi_inum },
237 &inode_u, &hash_info, attr,
238 val, val_size, h->flags, 0));
240 die("error creating xattr: %s", bch2_err_str(ret));
244 #define WRITE_DATA_BUF (1 << 20)
246 static char buf[WRITE_DATA_BUF] __aligned(PAGE_SIZE);
248 static void write_data(struct bch_fs *c,
249 struct bch_inode_unpacked *dst_inode,
250 u64 dst_offset, void *buf, size_t len)
252 struct bch_write_op op;
253 struct bio_vec bv[WRITE_DATA_BUF / PAGE_SIZE];
255 BUG_ON(dst_offset & (block_bytes(c) - 1));
256 BUG_ON(len & (block_bytes(c) - 1));
257 BUG_ON(len > WRITE_DATA_BUF);
259 bio_init(&op.wbio.bio, NULL, bv, ARRAY_SIZE(bv), 0);
260 bch2_bio_map(&op.wbio.bio, buf, len);
262 bch2_write_op_init(&op, c, bch2_opts_to_inode_opts(c->opts));
263 op.write_point = writepoint_hashed(0);
266 op.pos = SPOS(dst_inode->bi_inum, dst_offset >> 9, U32_MAX);
267 op.flags |= BCH_WRITE_SYNC;
269 int ret = bch2_disk_reservation_get(c, &op.res, len >> 9,
270 c->opts.data_replicas, 0);
272 die("error reserving space in new filesystem: %s", bch2_err_str(ret));
274 closure_call(&op.cl, bch2_write, NULL, NULL);
276 BUG_ON(!(op.flags & BCH_WRITE_DONE));
277 dst_inode->bi_sectors += len >> 9;
280 die("write error: %s", bch2_err_str(op.error));
283 static void copy_data(struct bch_fs *c,
284 struct bch_inode_unpacked *dst_inode,
285 int src_fd, u64 start, u64 end)
287 while (start < end) {
288 unsigned len = min_t(u64, end - start, sizeof(buf));
289 unsigned pad = round_up(len, block_bytes(c)) - len;
291 xpread(src_fd, buf, len, start);
292 memset(buf + len, 0, pad);
294 write_data(c, dst_inode, start, buf, len + pad);
299 static void link_data(struct bch_fs *c, struct bch_inode_unpacked *dst,
300 u64 logical, u64 physical, u64 length)
302 struct bch_dev *ca = c->devs[0];
304 BUG_ON(logical & (block_bytes(c) - 1));
305 BUG_ON(physical & (block_bytes(c) - 1));
306 BUG_ON(length & (block_bytes(c) - 1));
312 BUG_ON(physical + length > bucket_to_sector(ca, ca->mi.nbuckets));
315 struct bkey_i_extent *e;
316 BKEY_PADDED_ONSTACK(k, BKEY_EXTENT_VAL_U64s_MAX) k;
317 u64 b = sector_to_bucket(ca, physical);
318 struct disk_reservation res;
322 sectors = min(ca->mi.bucket_size -
323 (physical & (ca->mi.bucket_size - 1)),
326 e = bkey_extent_init(&k.k);
327 e->k.p.inode = dst->bi_inum;
328 e->k.p.offset = logical + sectors;
329 e->k.p.snapshot = U32_MAX;
331 bch2_bkey_append_ptr(&e->k_i, (struct bch_extent_ptr) {
334 .gen = *bucket_gen(ca, b),
337 ret = bch2_disk_reservation_get(c, &res, sectors, 1,
338 BCH_DISK_RESERVATION_NOFAIL);
340 die("error reserving space in new filesystem: %s",
343 ret = bch2_btree_insert(c, BTREE_ID_extents, &e->k_i, &res, 0);
345 die("btree insert error %s", bch2_err_str(ret));
347 bch2_disk_reservation_put(c, &res);
349 dst->bi_sectors += sectors;
356 static void copy_link(struct bch_fs *c, struct bch_inode_unpacked *dst,
359 ssize_t ret = readlink(src, buf, sizeof(buf));
361 die("readlink error: %m");
363 write_data(c, dst, 0, buf, round_up(ret, block_bytes(c)));
366 static void copy_file(struct bch_fs *c, struct bch_inode_unpacked *dst,
367 int src_fd, u64 src_size,
368 char *src_path, ranges *extents)
370 struct fiemap_iter iter;
371 struct fiemap_extent e;
373 fiemap_for_each(src_fd, iter, e)
374 if (e.fe_flags & FIEMAP_EXTENT_UNKNOWN) {
379 fiemap_for_each(src_fd, iter, e) {
380 if ((e.fe_logical & (block_bytes(c) - 1)) ||
381 (e.fe_length & (block_bytes(c) - 1)))
382 die("Unaligned extent in %s - can't handle", src_path);
384 if (e.fe_flags & (FIEMAP_EXTENT_UNKNOWN|
385 FIEMAP_EXTENT_ENCODED|
386 FIEMAP_EXTENT_NOT_ALIGNED|
387 FIEMAP_EXTENT_DATA_INLINE)) {
388 copy_data(c, dst, src_fd, e.fe_logical,
389 min(src_size - e.fe_logical,
395 * if the data is below 1 MB, copy it so it doesn't conflict
396 * with bcachefs's potentially larger superblock:
398 if (e.fe_physical < 1 << 20) {
399 copy_data(c, dst, src_fd, e.fe_logical,
400 min(src_size - e.fe_logical,
405 if ((e.fe_physical & (block_bytes(c) - 1)))
406 die("Unaligned extent in %s - can't handle", src_path);
408 range_add(extents, e.fe_physical, e.fe_length);
409 link_data(c, dst, e.fe_logical, e.fe_physical, e.fe_length);
413 struct copy_fs_state {
417 GENRADIX(u64) hardlinks;
421 static void copy_dir(struct copy_fs_state *s,
423 struct bch_inode_unpacked *dst,
424 int src_fd, const char *src_path)
426 DIR *dir = fdopendir(src_fd);
429 while ((errno = 0), (d = readdir(dir))) {
430 struct bch_inode_unpacked inode;
434 die("chdir error: %m");
437 xfstatat(src_fd, d->d_name, AT_SYMLINK_NOFOLLOW);
439 if (!strcmp(d->d_name, ".") ||
440 !strcmp(d->d_name, "..") ||
441 !strcmp(d->d_name, "lost+found") ||
442 stat.st_ino == s->bcachefs_inum)
445 char *child_path = mprintf("%s/%s", src_path, d->d_name);
447 if (stat.st_dev != s->dev)
448 die("%s does not have correct st_dev!", child_path);
450 u64 *dst_inum = S_ISREG(stat.st_mode)
451 ? genradix_ptr_alloc(&s->hardlinks, stat.st_ino, GFP_KERNEL)
454 if (dst_inum && *dst_inum) {
455 create_link(c, dst, d->d_name, *dst_inum, S_IFREG);
459 inode = create_file(c, dst, d->d_name,
460 stat.st_uid, stat.st_gid,
461 stat.st_mode, stat.st_rdev);
464 *dst_inum = inode.bi_inum;
466 copy_times(c, &inode, &stat);
467 copy_xattrs(c, &inode, d->d_name);
471 switch (mode_to_type(stat.st_mode)) {
473 fd = xopen(d->d_name, O_RDONLY|O_NOATIME);
474 copy_dir(s, c, &inode, fd, child_path);
478 inode.bi_size = stat.st_size;
480 fd = xopen(d->d_name, O_RDONLY|O_NOATIME);
481 copy_file(c, &inode, fd, stat.st_size,
482 child_path, &s->extents);
486 inode.bi_size = stat.st_size;
488 copy_link(c, &inode, d->d_name);
495 /* nothing else to copy for these: */
501 update_inode(c, &inode);
507 die("readdir error: %m");
510 static ranges reserve_new_fs_space(const char *file_path, unsigned block_size,
511 u64 size, u64 *bcachefs_inum, dev_t dev,
515 ? open(file_path, O_RDWR|O_CREAT, 0600)
516 : open(file_path, O_RDWR|O_CREAT|O_EXCL, 0600);
518 die("Error creating %s for bcachefs metadata: %m",
521 struct stat statbuf = xfstat(fd);
523 if (statbuf.st_dev != dev)
524 die("bcachefs file has incorrect device");
526 *bcachefs_inum = statbuf.st_ino;
528 if (fallocate(fd, 0, 0, size))
529 die("Error reserving space for bcachefs metadata: %m");
533 struct fiemap_iter iter;
534 struct fiemap_extent e;
535 ranges extents = { 0 };
537 fiemap_for_each(fd, iter, e) {
538 if (e.fe_flags & (FIEMAP_EXTENT_UNKNOWN|
539 FIEMAP_EXTENT_ENCODED|
540 FIEMAP_EXTENT_NOT_ALIGNED|
541 FIEMAP_EXTENT_DATA_INLINE))
542 die("Unable to continue: metadata file not fully mapped");
544 if ((e.fe_physical & (block_size - 1)) ||
545 (e.fe_length & (block_size - 1)))
546 die("Unable to continue: unaligned extents in metadata file");
548 range_add(&extents, e.fe_physical, e.fe_length);
552 ranges_sort_merge(&extents);
556 static void reserve_old_fs_space(struct bch_fs *c,
557 struct bch_inode_unpacked *root_inode,
560 struct bch_dev *ca = c->devs[0];
561 struct bch_inode_unpacked dst;
562 struct hole_iter iter;
565 dst = create_file(c, root_inode, "old_migrated_filesystem",
566 0, 0, S_IFREG|0400, 0);
567 dst.bi_size = bucket_to_sector(ca, ca->mi.nbuckets) << 9;
569 ranges_sort_merge(extents);
571 for_each_hole(iter, *extents, bucket_to_sector(ca, ca->mi.nbuckets) << 9, i)
572 link_data(c, &dst, i.start, i.start, i.end - i.start);
574 update_inode(c, &dst);
577 static void copy_fs(struct bch_fs *c, int src_fd, const char *src_path,
578 u64 bcachefs_inum, ranges *extents)
582 struct bch_inode_unpacked root_inode;
583 int ret = bch2_inode_find_by_inum(c, (subvol_inum) { 1, BCACHEFS_ROOT_INO },
586 die("error looking up root directory: %s", bch2_err_str(ret));
589 die("chdir error: %m");
591 struct stat stat = xfstat(src_fd);
592 copy_times(c, &root_inode, &stat);
593 copy_xattrs(c, &root_inode, ".");
595 struct copy_fs_state s = {
596 .bcachefs_inum = bcachefs_inum,
602 copy_dir(&s, c, &root_inode, src_fd, src_path);
604 reserve_old_fs_space(c, &root_inode, &s.extents);
606 update_inode(c, &root_inode);
608 darray_exit(&s.extents);
609 genradix_free(&s.hardlinks);
612 static void find_superblock_space(ranges extents,
613 struct format_opts opts,
614 struct dev_opts *dev)
616 darray_for_each(extents, i) {
617 u64 start = round_up(max(256ULL << 10, i->start),
618 dev->bucket_size << 9);
619 u64 end = round_down(i->end,
620 dev->bucket_size << 9);
622 /* Need space for two superblocks: */
623 if (start + (opts.superblock_size << 9) * 2 <= end) {
624 dev->sb_offset = start >> 9;
625 dev->sb_end = dev->sb_offset + opts.superblock_size * 2;
630 die("Couldn't find a valid location for superblock");
633 static void migrate_usage(void)
635 puts("bcachefs migrate - migrate an existing filesystem to bcachefs\n"
636 "Usage: bcachefs migrate [OPTION]...\n"
639 " -f fs Root of filesystem to migrate(s)\n"
640 " --encrypted Enable whole filesystem encryption (chacha20/poly1305)\n"
641 " --no_passphrase Don't encrypt master encryption key\n"
642 " -F Force, even if metadata file already exists\n"
643 " -h Display this help and exit\n"
644 "Report bugs to <linux-bcachefs@vger.kernel.org>");
647 static const struct option migrate_opts[] = {
648 { "encrypted", no_argument, NULL, 'e' },
649 { "no_passphrase", no_argument, NULL, 'p' },
653 static int migrate_fs(const char *fs_path,
654 struct bch_opt_strs fs_opt_strs,
655 struct bch_opts fs_opts,
656 struct format_opts format_opts,
659 if (!path_is_fs_root(fs_path))
660 die("%s is not a filysestem root", fs_path);
662 int fs_fd = xopen(fs_path, O_RDONLY|O_NOATIME);
663 struct stat stat = xfstat(fs_fd);
665 if (!S_ISDIR(stat.st_mode))
666 die("%s is not a directory", fs_path);
668 struct dev_opts dev = dev_opts_default();
670 dev.path = dev_t_to_path(stat.st_dev);
671 dev.bdev = blkdev_get_by_path(dev.path, BLK_OPEN_READ|BLK_OPEN_WRITE, &dev, NULL);
673 opt_set(fs_opts, block_size, get_blocksize(dev.bdev->bd_buffered_fd));
675 char *file_path = mprintf("%s/bcachefs", fs_path);
676 printf("Creating new filesystem on %s in space reserved at %s\n",
677 dev.path, file_path);
679 dev.size = get_size(dev.bdev->bd_buffered_fd);
680 dev.bucket_size = bch2_pick_bucket_size(fs_opts, &dev);
681 dev.nbuckets = dev.size / dev.bucket_size;
683 bch2_check_bucket_size(fs_opts, &dev);
686 ranges extents = reserve_new_fs_space(file_path,
687 fs_opts.block_size >> 9,
688 get_size(dev.bdev->bd_buffered_fd) / 5,
689 &bcachefs_inum, stat.st_dev, force);
691 find_superblock_space(extents, format_opts, &dev);
693 struct bch_sb *sb = bch2_format(fs_opt_strs,
694 fs_opts,format_opts, &dev, 1);
695 u64 sb_offset = le64_to_cpu(sb->layout.sb_offset[0]);
697 if (format_opts.passphrase)
698 bch2_add_key(sb, "user", "user", format_opts.passphrase);
702 struct bch_opts opts = bch2_opts_empty();
703 struct bch_fs *c = NULL;
704 char *path[1] = { dev.path };
706 opt_set(opts, sb, sb_offset);
707 opt_set(opts, nostart, true);
708 opt_set(opts, noexcl, true);
709 opt_set(opts, buckets_nouse, true);
711 c = bch2_fs_open(path, 1, opts);
713 die("Error opening new filesystem: %s", bch2_err_str(PTR_ERR(c)));
715 mark_unreserved_space(c, extents);
717 int ret = bch2_fs_start(c);
719 die("Error starting new filesystem: %s", bch2_err_str(ret));
721 copy_fs(c, fs_fd, fs_path, bcachefs_inum, &extents);
725 printf("Migrate complete, running fsck:\n");
726 opt_set(opts, nostart, false);
727 opt_set(opts, nochanges, true);
729 c = bch2_fs_open(path, 1, opts);
731 die("Error opening new filesystem: %s", bch2_err_str(PTR_ERR(c)));
734 printf("fsck complete\n");
736 printf("To mount the new filesystem, run\n"
737 " mount -t bcachefs -o sb=%llu %s dir\n"
739 "After verifying that the new filesystem is correct, to create a\n"
740 "superblock at the default offset and finish the migration run\n"
741 " bcachefs migrate-superblock -d %s -o %llu\n"
743 "The new filesystem will have a file at /old_migrated_filestem\n"
744 "referencing all disk space that might be used by the existing\n"
745 "filesystem. That file can be deleted once the old filesystem is\n"
746 "no longer needed (and should be deleted prior to running\n"
747 "bcachefs migrate-superblock)\n",
748 sb_offset, dev.path, dev.path, sb_offset);
752 int cmd_migrate(int argc, char *argv[])
754 struct format_opts format_opts = format_opts_default();
755 char *fs_path = NULL;
756 bool no_passphrase = false, force = false;
759 struct bch_opt_strs fs_opt_strs =
760 bch2_cmdline_opts_get(&argc, argv, OPT_FORMAT);
761 struct bch_opts fs_opts = bch2_parse_opts(fs_opt_strs);
763 while ((opt = getopt_long(argc, argv, "f:Fh",
764 migrate_opts, NULL)) != -1)
770 format_opts.encrypted = true;
773 no_passphrase = true;
784 die("Please specify a filesystem to migrate");
786 if (format_opts.encrypted && !no_passphrase)
787 format_opts.passphrase = read_passphrase_twice("Enter passphrase: ");
789 int ret = migrate_fs(fs_path,
793 bch2_opt_strs_free(&fs_opt_strs);
797 static void migrate_superblock_usage(void)
799 puts("bcachefs migrate-superblock - create default superblock after migrating\n"
800 "Usage: bcachefs migrate-superblock [OPTION]...\n"
803 " -d device Device to create superblock for\n"
804 " -o offset Offset of existing superblock\n"
805 " -h Display this help and exit\n"
806 "Report bugs to <linux-bcachefs@vger.kernel.org>");
809 int cmd_migrate_superblock(int argc, char *argv[])
815 while ((opt = getopt(argc, argv, "d:o:h")) != -1)
821 ret = kstrtou64(optarg, 10, &offset);
823 die("Invalid offset");
826 migrate_superblock_usage();
831 die("Please specify a device");
834 die("Please specify offset of existing superblock");
836 int fd = xopen(dev, O_RDWR);
837 struct bch_sb *sb = __bch2_super_read(fd, offset);
839 if (sb->layout.nr_superblocks >= ARRAY_SIZE(sb->layout.sb_offset))
840 die("Can't add superblock: no space left in superblock layout");
843 for (i = 0; i < sb->layout.nr_superblocks; i++)
844 if (le64_to_cpu(sb->layout.sb_offset[i]) == BCH_SB_SECTOR)
845 die("Superblock layout already has default superblock");
847 memmove(&sb->layout.sb_offset[1],
848 &sb->layout.sb_offset[0],
849 sb->layout.nr_superblocks * sizeof(u64));
850 sb->layout.nr_superblocks++;
852 sb->layout.sb_offset[0] = cpu_to_le64(BCH_SB_SECTOR);
854 bch2_super_write(fd, sb);