1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_update_interior.h"
7 #include "disk_groups.h"
12 #include "journal_io.h"
13 #include "journal_sb.h"
14 #include "journal_seq_blacklist.h"
22 #include <linux/backing-dev.h>
23 #include <linux/sort.h>
25 #include <trace/events/bcachefs.h>
27 const char * const bch2_sb_fields[] = {
28 #define x(name, nr) #name,
34 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
37 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
38 enum bch_sb_field_type type)
40 struct bch_sb_field *f;
42 /* XXX: need locking around superblock to access optional fields */
44 vstruct_for_each(sb, f)
45 if (le32_to_cpu(f->type) == type)
50 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
51 struct bch_sb_field *f,
54 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
55 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
57 BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
62 f = vstruct_last(sb->sb);
63 memset(f, 0, sizeof(u64) * u64s);
64 f->u64s = cpu_to_le32(u64s);
72 f->u64s = cpu_to_le32(u64s);
78 memmove(dst, src, vstruct_end(sb->sb) - src);
81 memset(src, 0, dst - src);
84 sb->sb->u64s = cpu_to_le32(sb_u64s);
86 return u64s ? f : NULL;
89 void bch2_sb_field_delete(struct bch_sb_handle *sb,
90 enum bch_sb_field_type type)
92 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
95 __bch2_sb_field_resize(sb, f, 0);
98 /* Superblock realloc/free: */
100 void bch2_free_super(struct bch_sb_handle *sb)
103 if (!IS_ERR_OR_NULL(sb->bdev))
104 blkdev_put(sb->bdev, sb->mode);
107 memset(sb, 0, sizeof(*sb));
110 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
112 size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
113 size_t new_buffer_size;
114 struct bch_sb *new_sb;
118 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
120 new_buffer_size = roundup_pow_of_two(new_bytes);
122 if (sb->sb && sb->buffer_size >= new_buffer_size)
125 if (sb->have_layout) {
126 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
128 if (new_bytes > max_bytes) {
129 pr_err("%pg: superblock too big: want %zu but have %llu",
130 sb->bdev, new_bytes, max_bytes);
131 return -BCH_ERR_ENOSPC_sb;
135 if (sb->buffer_size >= new_buffer_size && sb->sb)
138 if (dynamic_fault("bcachefs:add:super_realloc"))
142 unsigned nr_bvecs = DIV_ROUND_UP(new_buffer_size, PAGE_SIZE);
144 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
148 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
154 new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
159 sb->buffer_size = new_buffer_size;
164 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
165 enum bch_sb_field_type type,
168 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
169 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
170 ssize_t d = -old_u64s + u64s;
172 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
176 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
180 lockdep_assert_held(&c->sb_lock);
182 /* XXX: we're not checking that offline device have enough space */
184 for_each_online_member(ca, c, i) {
185 struct bch_sb_handle *sb = &ca->disk_sb;
187 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
188 percpu_ref_put(&ca->ref);
194 f = bch2_sb_field_get(sb->sb, type);
195 f = __bch2_sb_field_resize(sb, f, u64s);
197 f->type = cpu_to_le32(type);
201 /* Superblock validate: */
203 static inline void __bch2_sb_layout_size_assert(void)
205 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
208 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
210 u64 offset, prev_offset, max_sectors;
213 if (uuid_le_cmp(layout->magic, BCACHE_MAGIC) &&
214 uuid_le_cmp(layout->magic, BCHFS_MAGIC)) {
215 prt_printf(out, "Not a bcachefs superblock layout");
216 return -BCH_ERR_invalid_sb_layout;
219 if (layout->layout_type != 0) {
220 prt_printf(out, "Invalid superblock layout type %u",
221 layout->layout_type);
222 return -BCH_ERR_invalid_sb_layout_type;
225 if (!layout->nr_superblocks) {
226 prt_printf(out, "Invalid superblock layout: no superblocks");
227 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
230 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
231 prt_printf(out, "Invalid superblock layout: too many superblocks");
232 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
235 max_sectors = 1 << layout->sb_max_size_bits;
237 prev_offset = le64_to_cpu(layout->sb_offset[0]);
239 for (i = 1; i < layout->nr_superblocks; i++) {
240 offset = le64_to_cpu(layout->sb_offset[i]);
242 if (offset < prev_offset + max_sectors) {
243 prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
244 " (sb %u ends at %llu next starts at %llu",
245 i - 1, prev_offset + max_sectors, offset);
246 return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
248 prev_offset = offset;
254 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
257 struct bch_sb *sb = disk_sb->sb;
258 struct bch_sb_field *f;
259 struct bch_sb_field_members *mi;
260 enum bch_opt_id opt_id;
261 u32 version, version_min;
265 version = le16_to_cpu(sb->version);
266 version_min = version >= bcachefs_metadata_version_bkey_renumber
267 ? le16_to_cpu(sb->version_min)
270 if (version >= bcachefs_metadata_version_max) {
271 prt_printf(out, "Unsupported superblock version %u (min %u, max %u)",
272 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
273 return -BCH_ERR_invalid_sb_version;
276 if (version_min < bcachefs_metadata_version_min) {
277 prt_printf(out, "Unsupported superblock version %u (min %u, max %u)",
278 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
279 return -BCH_ERR_invalid_sb_version;
282 if (version_min > version) {
283 prt_printf(out, "Bad minimum version %u, greater than version field %u",
284 version_min, version);
285 return -BCH_ERR_invalid_sb_version;
288 if (sb->features[1] ||
289 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
290 prt_printf(out, "Filesystem has incompatible features");
291 return -BCH_ERR_invalid_sb_features;
294 block_size = le16_to_cpu(sb->block_size);
296 if (block_size > PAGE_SECTORS) {
297 prt_printf(out, "Block size too big (got %u, max %u)",
298 block_size, PAGE_SECTORS);
299 return -BCH_ERR_invalid_sb_block_size;
302 if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
303 prt_printf(out, "Bad user UUID (got zeroes)");
304 return -BCH_ERR_invalid_sb_uuid;
307 if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
308 prt_printf(out, "Bad intenal UUID (got zeroes)");
309 return -BCH_ERR_invalid_sb_uuid;
312 if (!sb->nr_devices ||
313 sb->nr_devices > BCH_SB_MEMBERS_MAX) {
314 prt_printf(out, "Bad number of member devices %u (max %u)",
315 sb->nr_devices, BCH_SB_MEMBERS_MAX);
316 return -BCH_ERR_invalid_sb_too_many_members;
319 if (sb->dev_idx >= sb->nr_devices) {
320 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
321 sb->dev_idx, sb->nr_devices);
322 return -BCH_ERR_invalid_sb_dev_idx;
325 if (!sb->time_precision ||
326 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
327 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
328 le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
329 return -BCH_ERR_invalid_sb_time_precision;
334 * Been seeing a bug where these are getting inexplicably
335 * zeroed, so we'r now validating them, but we have to be
336 * careful not to preven people's filesystems from mounting:
338 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
339 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
340 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
341 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
344 for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
345 const struct bch_option *opt = bch2_opt_table + opt_id;
347 if (opt->get_sb != BCH2_NO_SB_OPT) {
348 u64 v = bch2_opt_from_sb(sb, opt_id);
350 prt_printf(out, "Invalid option ");
351 ret = bch2_opt_validate(opt, v, out);
359 /* validate layout */
360 ret = validate_sb_layout(&sb->layout, out);
364 vstruct_for_each(sb, f) {
366 prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
367 le32_to_cpu(f->type));
368 return -BCH_ERR_invalid_sb_field_size;
371 if (vstruct_next(f) > vstruct_last(sb)) {
372 prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
373 le32_to_cpu(f->type));
374 return -BCH_ERR_invalid_sb_field_size;
378 /* members must be validated first: */
379 mi = bch2_sb_get_members(sb);
381 prt_printf(out, "Invalid superblock: member info area missing");
382 return -BCH_ERR_invalid_sb_members_missing;
385 ret = bch2_sb_field_validate(sb, &mi->field, out);
389 vstruct_for_each(sb, f) {
390 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
393 ret = bch2_sb_field_validate(sb, f, out);
403 static void bch2_sb_update(struct bch_fs *c)
405 struct bch_sb *src = c->disk_sb.sb;
406 struct bch_sb_field_members *mi = bch2_sb_get_members(src);
410 lockdep_assert_held(&c->sb_lock);
412 c->sb.uuid = src->uuid;
413 c->sb.user_uuid = src->user_uuid;
414 c->sb.version = le16_to_cpu(src->version);
415 c->sb.version_min = le16_to_cpu(src->version_min);
416 c->sb.nr_devices = src->nr_devices;
417 c->sb.clean = BCH_SB_CLEAN(src);
418 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
420 c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
421 c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
423 /* XXX this is wrong, we need a 96 or 128 bit integer type */
424 c->sb.time_base_lo = div_u64(le64_to_cpu(src->time_base_lo),
425 c->sb.nsec_per_time_unit);
426 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
428 c->sb.features = le64_to_cpu(src->features[0]);
429 c->sb.compat = le64_to_cpu(src->compat[0]);
431 for_each_member_device(ca, c, i)
432 ca->mi = bch2_mi_to_cpu(mi->members + i);
435 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
437 struct bch_sb_field *src_f, *dst_f;
438 struct bch_sb *dst = dst_handle->sb;
441 dst->version = src->version;
442 dst->version_min = src->version_min;
444 dst->uuid = src->uuid;
445 dst->user_uuid = src->user_uuid;
446 memcpy(dst->label, src->label, sizeof(dst->label));
448 dst->block_size = src->block_size;
449 dst->nr_devices = src->nr_devices;
451 dst->time_base_lo = src->time_base_lo;
452 dst->time_base_hi = src->time_base_hi;
453 dst->time_precision = src->time_precision;
455 memcpy(dst->flags, src->flags, sizeof(dst->flags));
456 memcpy(dst->features, src->features, sizeof(dst->features));
457 memcpy(dst->compat, src->compat, sizeof(dst->compat));
459 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
462 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
465 src_f = bch2_sb_field_get(src, i);
466 dst_f = bch2_sb_field_get(dst, i);
468 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
469 (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
471 int ret = bch2_sb_realloc(dst_handle, le32_to_cpu(dst_handle->sb->u64s) + d);
475 dst = dst_handle->sb;
476 dst_f = bch2_sb_field_get(dst, i);
479 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
480 src_f ? le32_to_cpu(src_f->u64s) : 0);
483 memcpy(dst_f, src_f, vstruct_bytes(src_f));
489 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
493 lockdep_assert_held(&c->sb_lock);
495 ret = bch2_sb_realloc(&c->disk_sb, 0) ?:
496 __copy_super(&c->disk_sb, src) ?:
497 bch2_sb_replicas_to_cpu_replicas(c) ?:
498 bch2_sb_disk_groups_to_cpu(c);
506 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
508 return __copy_super(&ca->disk_sb, c->disk_sb.sb);
511 /* read superblock: */
513 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
515 struct bch_csum csum;
516 u32 version, version_min;
520 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
521 sb->bio->bi_iter.bi_sector = offset;
522 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
524 ret = submit_bio_wait(sb->bio);
526 prt_printf(err, "IO error: %i", ret);
530 if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC) &&
531 uuid_le_cmp(sb->sb->magic, BCHFS_MAGIC)) {
532 prt_printf(err, "Not a bcachefs superblock");
533 return -BCH_ERR_invalid_sb_magic;
536 version = le16_to_cpu(sb->sb->version);
537 version_min = version >= bcachefs_metadata_version_bkey_renumber
538 ? le16_to_cpu(sb->sb->version_min)
541 if (version >= bcachefs_metadata_version_max) {
542 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
543 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
544 return -BCH_ERR_invalid_sb_version;
547 if (version_min < bcachefs_metadata_version_min) {
548 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
549 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
550 return -BCH_ERR_invalid_sb_version;
553 bytes = vstruct_bytes(sb->sb);
555 if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
556 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
557 bytes, 512UL << sb->sb->layout.sb_max_size_bits);
558 return -BCH_ERR_invalid_sb_too_big;
561 if (bytes > sb->buffer_size) {
562 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
567 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
568 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
569 return -BCH_ERR_invalid_sb_csum_type;
572 /* XXX: verify MACs */
573 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
574 null_nonce(), sb->sb);
576 if (bch2_crc_cmp(csum, sb->sb->csum)) {
577 prt_printf(err, "bad checksum");
578 return -BCH_ERR_invalid_sb_csum;
581 sb->seq = le64_to_cpu(sb->sb->seq);
586 int bch2_read_super(const char *path, struct bch_opts *opts,
587 struct bch_sb_handle *sb)
589 u64 offset = opt_get(*opts, sb);
590 struct bch_sb_layout layout;
591 struct printbuf err = PRINTBUF;
595 pr_verbose_init(*opts, "");
597 memset(sb, 0, sizeof(*sb));
598 sb->mode = FMODE_READ;
601 if (!opt_get(*opts, noexcl))
602 sb->mode |= FMODE_EXCL;
604 if (!opt_get(*opts, nochanges))
605 sb->mode |= FMODE_WRITE;
607 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
608 if (IS_ERR(sb->bdev) &&
609 PTR_ERR(sb->bdev) == -EACCES &&
610 opt_get(*opts, read_only)) {
611 sb->mode &= ~FMODE_WRITE;
613 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
614 if (!IS_ERR(sb->bdev))
615 opt_set(*opts, nochanges, true);
618 if (IS_ERR(sb->bdev)) {
619 ret = PTR_ERR(sb->bdev);
623 ret = bch2_sb_realloc(sb, 0);
625 prt_printf(&err, "error allocating memory for superblock");
629 if (bch2_fs_init_fault("read_super")) {
630 prt_printf(&err, "dynamic fault");
635 ret = read_one_super(sb, offset, &err);
639 if (opt_defined(*opts, sb))
642 printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
644 printbuf_reset(&err);
647 * Error reading primary superblock - read location of backup
650 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
651 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
653 * use sb buffer to read layout, since sb buffer is page aligned but
656 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
658 ret = submit_bio_wait(sb->bio);
660 prt_printf(&err, "IO error: %i", ret);
664 memcpy(&layout, sb->sb, sizeof(layout));
665 ret = validate_sb_layout(&layout, &err);
669 for (i = layout.sb_offset;
670 i < layout.sb_offset + layout.nr_superblocks; i++) {
671 offset = le64_to_cpu(*i);
673 if (offset == opt_get(*opts, sb))
676 ret = read_one_super(sb, offset, &err);
684 if (le16_to_cpu(sb->sb->block_size) << 9 <
685 bdev_logical_block_size(sb->bdev)) {
686 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
687 le16_to_cpu(sb->sb->block_size) << 9,
688 bdev_logical_block_size(sb->bdev));
689 ret = -BCH_ERR_block_size_too_small;
694 sb->have_layout = true;
696 ret = bch2_sb_validate(sb, &err, READ);
698 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
703 pr_verbose_init(*opts, "ret %i", ret);
707 printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
714 /* write superblock: */
716 static void write_super_endio(struct bio *bio)
718 struct bch_dev *ca = bio->bi_private;
720 /* XXX: return errors directly */
722 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
723 bch2_blk_status_to_str(bio->bi_status)))
724 ca->sb_write_error = 1;
726 closure_put(&ca->fs->sb_write);
727 percpu_ref_put(&ca->io_ref);
730 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
732 struct bch_sb *sb = ca->disk_sb.sb;
733 struct bio *bio = ca->disk_sb.bio;
735 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
736 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
737 bio->bi_end_io = write_super_endio;
738 bio->bi_private = ca;
739 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
741 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
744 percpu_ref_get(&ca->io_ref);
745 closure_bio_submit(bio, &c->sb_write);
748 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
750 struct bch_sb *sb = ca->disk_sb.sb;
751 struct bio *bio = ca->disk_sb.bio;
753 sb->offset = sb->layout.sb_offset[idx];
755 SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
756 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
759 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
760 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
761 bio->bi_end_io = write_super_endio;
762 bio->bi_private = ca;
763 bch2_bio_map(bio, sb,
764 roundup((size_t) vstruct_bytes(sb),
765 bdev_logical_block_size(ca->disk_sb.bdev)));
767 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
770 percpu_ref_get(&ca->io_ref);
771 closure_bio_submit(bio, &c->sb_write);
774 int bch2_write_super(struct bch_fs *c)
776 struct closure *cl = &c->sb_write;
778 struct printbuf err = PRINTBUF;
779 unsigned i, sb = 0, nr_wrote;
780 struct bch_devs_mask sb_written;
781 bool wrote, can_mount_without_written, can_mount_with_written;
782 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
785 trace_and_count(c, write_super, c, _RET_IP_);
787 if (c->opts.very_degraded)
788 degraded_flags |= BCH_FORCE_IF_LOST;
790 lockdep_assert_held(&c->sb_lock);
792 closure_init_stack(cl);
793 memset(&sb_written, 0, sizeof(sb_written));
795 if (c->opts.version_upgrade) {
796 c->disk_sb.sb->magic = BCHFS_MAGIC;
797 c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
800 le64_add_cpu(&c->disk_sb.sb->seq, 1);
802 if (test_bit(BCH_FS_ERROR, &c->flags))
803 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
804 if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
805 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
807 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
809 bch2_sb_counters_from_cpu(c);
811 for_each_online_member(ca, c, i)
812 bch2_sb_from_fs(c, ca);
814 for_each_online_member(ca, c, i) {
815 printbuf_reset(&err);
817 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
819 bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
820 percpu_ref_put(&ca->io_ref);
825 if (c->opts.nochanges)
829 * Defer writing the superblock until filesystem initialization is
830 * complete - don't write out a partly initialized superblock:
832 if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
835 for_each_online_member(ca, c, i) {
836 __set_bit(ca->dev_idx, sb_written.d);
837 ca->sb_write_error = 0;
840 for_each_online_member(ca, c, i)
841 read_back_super(c, ca);
844 for_each_online_member(ca, c, i) {
845 if (ca->sb_write_error)
848 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
849 bch2_fs_fatal_error(c,
850 "Superblock write was silently dropped! (seq %llu expected %llu)",
851 le64_to_cpu(ca->sb_read_scratch->seq),
853 percpu_ref_put(&ca->io_ref);
854 ret = -BCH_ERR_erofs_sb_err;
858 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
859 bch2_fs_fatal_error(c,
860 "Superblock modified by another process (seq %llu expected %llu)",
861 le64_to_cpu(ca->sb_read_scratch->seq),
863 percpu_ref_put(&ca->io_ref);
864 ret = -BCH_ERR_erofs_sb_err;
871 for_each_online_member(ca, c, i)
872 if (!ca->sb_write_error &&
873 sb < ca->disk_sb.sb->layout.nr_superblocks) {
874 write_one_super(c, ca, sb);
881 for_each_online_member(ca, c, i) {
882 if (ca->sb_write_error)
883 __clear_bit(ca->dev_idx, sb_written.d);
885 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
888 nr_wrote = dev_mask_nr(&sb_written);
890 can_mount_with_written =
891 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
893 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
894 sb_written.d[i] = ~sb_written.d[i];
896 can_mount_without_written =
897 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
900 * If we would be able to mount _without_ the devices we successfully
901 * wrote superblocks to, we weren't able to write to enough devices:
903 * Exception: if we can mount without the successes because we haven't
904 * written anything (new filesystem), we continue if we'd be able to
905 * mount with the devices we did successfully write to:
907 if (bch2_fs_fatal_err_on(!nr_wrote ||
908 !can_mount_with_written ||
909 (can_mount_without_written &&
910 !can_mount_with_written), c,
911 "Unable to write superblock to sufficient devices (from %ps)",
915 /* Make new options visible after they're persistent: */
921 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
923 mutex_lock(&c->sb_lock);
924 if (!(c->sb.features & (1ULL << feat))) {
925 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
929 mutex_unlock(&c->sb_lock);
932 /* BCH_SB_FIELD_members: */
934 static int bch2_sb_members_validate(struct bch_sb *sb,
935 struct bch_sb_field *f,
936 struct printbuf *err)
938 struct bch_sb_field_members *mi = field_to_type(f, members);
941 if ((void *) (mi->members + sb->nr_devices) >
942 vstruct_end(&mi->field)) {
943 prt_printf(err, "too many devices for section size");
944 return -BCH_ERR_invalid_sb_members;
947 for (i = 0; i < sb->nr_devices; i++) {
948 struct bch_member *m = mi->members + i;
950 if (!bch2_member_exists(m))
953 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
954 prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
955 i, le64_to_cpu(m->nbuckets), LONG_MAX);
956 return -BCH_ERR_invalid_sb_members;
959 if (le64_to_cpu(m->nbuckets) -
960 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
961 prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
962 i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
963 return -BCH_ERR_invalid_sb_members;
966 if (le16_to_cpu(m->bucket_size) <
967 le16_to_cpu(sb->block_size)) {
968 prt_printf(err, "device %u: bucket size %u smaller than block size %u",
969 i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
970 return -BCH_ERR_invalid_sb_members;
973 if (le16_to_cpu(m->bucket_size) <
974 BCH_SB_BTREE_NODE_SIZE(sb)) {
975 prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
976 i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
977 return -BCH_ERR_invalid_sb_members;
984 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
985 struct bch_sb_field *f)
987 struct bch_sb_field_members *mi = field_to_type(f, members);
988 struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
991 for (i = 0; i < sb->nr_devices; i++) {
992 struct bch_member *m = mi->members + i;
993 unsigned data_have = bch2_sb_dev_has_data(sb, i);
994 u64 bucket_size = le16_to_cpu(m->bucket_size);
995 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
997 if (!bch2_member_exists(m))
1000 prt_printf(out, "Device:");
1002 prt_printf(out, "%u", i);
1005 printbuf_indent_add(out, 2);
1007 prt_printf(out, "UUID:");
1009 pr_uuid(out, m->uuid.b);
1012 prt_printf(out, "Size:");
1014 prt_units_u64(out, device_size << 9);
1017 prt_printf(out, "Bucket size:");
1019 prt_units_u64(out, bucket_size << 9);
1022 prt_printf(out, "First bucket:");
1024 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1027 prt_printf(out, "Buckets:");
1029 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1032 prt_printf(out, "Last mount:");
1035 pr_time(out, le64_to_cpu(m->last_mount));
1037 prt_printf(out, "(never)");
1040 prt_printf(out, "State:");
1042 prt_printf(out, "%s",
1043 BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1044 ? bch2_member_states[BCH_MEMBER_STATE(m)]
1048 prt_printf(out, "Label:");
1050 if (BCH_MEMBER_GROUP(m)) {
1051 unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1053 if (idx < disk_groups_nr(gi))
1054 prt_printf(out, "%s (%u)",
1055 gi->entries[idx].label, idx);
1057 prt_printf(out, "(bad disk labels section)");
1059 prt_printf(out, "(none)");
1063 prt_printf(out, "Data allowed:");
1065 if (BCH_MEMBER_DATA_ALLOWED(m))
1066 prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1068 prt_printf(out, "(none)");
1071 prt_printf(out, "Has data:");
1074 prt_bitflags(out, bch2_data_types, data_have);
1076 prt_printf(out, "(none)");
1079 prt_printf(out, "Discard:");
1081 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1084 prt_printf(out, "Freespace initialized:");
1086 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1089 printbuf_indent_sub(out, 2);
1093 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1094 .validate = bch2_sb_members_validate,
1095 .to_text = bch2_sb_members_to_text,
1098 /* BCH_SB_FIELD_crypt: */
1100 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1101 struct bch_sb_field *f,
1102 struct printbuf *err)
1104 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1106 if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1107 prt_printf(err, "wrong size (got %zu should be %zu)",
1108 vstruct_bytes(&crypt->field), sizeof(*crypt));
1109 return -BCH_ERR_invalid_sb_crypt;
1112 if (BCH_CRYPT_KDF_TYPE(crypt)) {
1113 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1114 return -BCH_ERR_invalid_sb_crypt;
1120 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1121 struct bch_sb_field *f)
1123 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1125 prt_printf(out, "KFD: %llu", BCH_CRYPT_KDF_TYPE(crypt));
1127 prt_printf(out, "scrypt n: %llu", BCH_KDF_SCRYPT_N(crypt));
1129 prt_printf(out, "scrypt r: %llu", BCH_KDF_SCRYPT_R(crypt));
1131 prt_printf(out, "scrypt p: %llu", BCH_KDF_SCRYPT_P(crypt));
1135 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1136 .validate = bch2_sb_crypt_validate,
1137 .to_text = bch2_sb_crypt_to_text,
1140 /* BCH_SB_FIELD_clean: */
1142 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1144 struct jset_entry *entry;
1147 for (entry = clean->start;
1148 entry < (struct jset_entry *) vstruct_end(&clean->field);
1149 entry = vstruct_next(entry)) {
1150 ret = bch2_journal_entry_validate(c, NULL, entry,
1151 le16_to_cpu(c->disk_sb.sb->version),
1152 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1161 int bch2_fs_mark_dirty(struct bch_fs *c)
1166 * Unconditionally write superblock, to verify it hasn't changed before
1170 mutex_lock(&c->sb_lock);
1171 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1172 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1173 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1174 ret = bch2_write_super(c);
1175 mutex_unlock(&c->sb_lock);
1180 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1182 struct jset_entry *entry = *end;
1183 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1185 memset(entry, 0, u64s * sizeof(u64));
1187 * The u64s field counts from the start of data, ignoring the shared
1190 entry->u64s = cpu_to_le16(u64s - 1);
1192 *end = vstruct_next(*end);
1196 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1197 struct jset_entry **end,
1203 percpu_down_read(&c->mark_lock);
1206 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1207 bch2_fs_usage_acc_to_base(c, i);
1209 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1213 struct jset_entry_usage *u =
1214 container_of(jset_entry_init(end, sizeof(*u)),
1215 struct jset_entry_usage, entry);
1217 u->entry.type = BCH_JSET_ENTRY_usage;
1218 u->entry.btree_id = BCH_FS_USAGE_inodes;
1219 u->v = cpu_to_le64(c->usage_base->nr_inodes);
1223 struct jset_entry_usage *u =
1224 container_of(jset_entry_init(end, sizeof(*u)),
1225 struct jset_entry_usage, entry);
1227 u->entry.type = BCH_JSET_ENTRY_usage;
1228 u->entry.btree_id = BCH_FS_USAGE_key_version;
1229 u->v = cpu_to_le64(atomic64_read(&c->key_version));
1232 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1233 struct jset_entry_usage *u =
1234 container_of(jset_entry_init(end, sizeof(*u)),
1235 struct jset_entry_usage, entry);
1237 u->entry.type = BCH_JSET_ENTRY_usage;
1238 u->entry.btree_id = BCH_FS_USAGE_reserved;
1240 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1243 for (i = 0; i < c->replicas.nr; i++) {
1244 struct bch_replicas_entry *e =
1245 cpu_replicas_entry(&c->replicas, i);
1246 struct jset_entry_data_usage *u =
1247 container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1248 struct jset_entry_data_usage, entry);
1250 u->entry.type = BCH_JSET_ENTRY_data_usage;
1251 u->v = cpu_to_le64(c->usage_base->replicas[i]);
1252 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
1253 "embedded variable length struct");
1256 for_each_member_device(ca, c, dev) {
1257 unsigned b = sizeof(struct jset_entry_dev_usage) +
1258 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1259 struct jset_entry_dev_usage *u =
1260 container_of(jset_entry_init(end, b),
1261 struct jset_entry_dev_usage, entry);
1263 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1264 u->dev = cpu_to_le32(dev);
1265 u->buckets_ec = cpu_to_le64(ca->usage_base->buckets_ec);
1267 for (i = 0; i < BCH_DATA_NR; i++) {
1268 u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1269 u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1270 u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1274 percpu_up_read(&c->mark_lock);
1276 for (i = 0; i < 2; i++) {
1277 struct jset_entry_clock *clock =
1278 container_of(jset_entry_init(end, sizeof(*clock)),
1279 struct jset_entry_clock, entry);
1281 clock->entry.type = BCH_JSET_ENTRY_clock;
1283 clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1287 void bch2_fs_mark_clean(struct bch_fs *c)
1289 struct bch_sb_field_clean *sb_clean;
1290 struct jset_entry *entry;
1294 mutex_lock(&c->sb_lock);
1295 if (BCH_SB_CLEAN(c->disk_sb.sb))
1298 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1300 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1301 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1302 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1303 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1305 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1307 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1309 bch_err(c, "error resizing superblock while setting filesystem clean");
1313 sb_clean->flags = 0;
1314 sb_clean->journal_seq = cpu_to_le64(atomic64_read(&c->journal.seq));
1316 /* Trying to catch outstanding bug: */
1317 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1319 entry = sb_clean->start;
1320 bch2_journal_super_entries_add_common(c, &entry, 0);
1321 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1322 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1325 vstruct_end(&sb_clean->field) - (void *) entry);
1328 * this should be in the write path, and we should be validating every
1329 * superblock section:
1331 ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1333 bch_err(c, "error writing marking filesystem clean: validate error");
1337 bch2_write_super(c);
1339 mutex_unlock(&c->sb_lock);
1342 static int bch2_sb_clean_validate(struct bch_sb *sb,
1343 struct bch_sb_field *f,
1344 struct printbuf *err)
1346 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1348 if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1349 prt_printf(err, "wrong size (got %zu should be %zu)",
1350 vstruct_bytes(&clean->field), sizeof(*clean));
1351 return -BCH_ERR_invalid_sb_clean;
1357 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1358 struct bch_sb_field *f)
1360 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1361 struct jset_entry *entry;
1363 prt_printf(out, "flags: %x", le32_to_cpu(clean->flags));
1365 prt_printf(out, "journal_seq: %llu", le64_to_cpu(clean->journal_seq));
1368 for (entry = clean->start;
1369 entry != vstruct_end(&clean->field);
1370 entry = vstruct_next(entry)) {
1371 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1375 bch2_journal_entry_to_text(out, NULL, entry);
1380 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1381 .validate = bch2_sb_clean_validate,
1382 .to_text = bch2_sb_clean_to_text,
1385 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1387 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1392 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1393 struct printbuf *err)
1395 unsigned type = le32_to_cpu(f->type);
1396 struct printbuf field_err = PRINTBUF;
1399 if (type >= BCH_SB_FIELD_NR)
1402 ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1404 prt_printf(err, "Invalid superblock section %s: %s",
1405 bch2_sb_fields[type],
1408 bch2_sb_field_to_text(err, sb, f);
1411 printbuf_exit(&field_err);
1415 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1416 struct bch_sb_field *f)
1418 unsigned type = le32_to_cpu(f->type);
1419 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1420 ? bch2_sb_field_ops[type] : NULL;
1422 if (!out->nr_tabstops)
1423 printbuf_tabstop_push(out, 32);
1426 prt_printf(out, "%s", bch2_sb_fields[type]);
1428 prt_printf(out, "(unknown field %u)", type);
1430 prt_printf(out, " (size %zu):", vstruct_bytes(f));
1433 if (ops && ops->to_text) {
1434 printbuf_indent_add(out, 2);
1435 bch2_sb_field_ops[type]->to_text(out, sb, f);
1436 printbuf_indent_sub(out, 2);
1440 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1444 prt_printf(out, "Type: %u", l->layout_type);
1447 prt_str(out, "Superblock max size: ");
1448 prt_units_u64(out, 512 << l->sb_max_size_bits);
1451 prt_printf(out, "Nr superblocks: %u", l->nr_superblocks);
1454 prt_str(out, "Offsets: ");
1455 for (i = 0; i < l->nr_superblocks; i++) {
1458 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1463 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1464 bool print_layout, unsigned fields)
1466 struct bch_sb_field_members *mi;
1467 struct bch_sb_field *f;
1468 u64 fields_have = 0;
1469 unsigned nr_devices = 0;
1471 if (!out->nr_tabstops)
1472 printbuf_tabstop_push(out, 44);
1474 mi = bch2_sb_get_members(sb);
1476 struct bch_member *m;
1478 for (m = mi->members;
1479 m < mi->members + sb->nr_devices;
1481 nr_devices += bch2_member_exists(m);
1484 prt_printf(out, "External UUID:");
1486 pr_uuid(out, sb->user_uuid.b);
1489 prt_printf(out, "Internal UUID:");
1491 pr_uuid(out, sb->uuid.b);
1494 prt_str(out, "Device index:");
1496 prt_printf(out, "%u", sb->dev_idx);
1499 prt_str(out, "Label:");
1501 prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1504 prt_str(out, "Version:");
1506 prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1509 prt_printf(out, "Oldest version on disk:");
1511 prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1514 prt_printf(out, "Created:");
1516 if (sb->time_base_lo)
1517 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1519 prt_printf(out, "(not set)");
1522 prt_printf(out, "Sequence number:");
1524 prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1527 prt_printf(out, "Superblock size:");
1529 prt_printf(out, "%zu", vstruct_bytes(sb));
1532 prt_printf(out, "Clean:");
1534 prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1537 prt_printf(out, "Devices:");
1539 prt_printf(out, "%u", nr_devices);
1542 prt_printf(out, "Sections:");
1543 vstruct_for_each(sb, f)
1544 fields_have |= 1 << le32_to_cpu(f->type);
1546 prt_bitflags(out, bch2_sb_fields, fields_have);
1549 prt_printf(out, "Features:");
1551 prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1554 prt_printf(out, "Compat features:");
1556 prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1560 prt_printf(out, "Options:");
1562 printbuf_indent_add(out, 2);
1566 for (id = 0; id < bch2_opts_nr; id++) {
1567 const struct bch_option *opt = bch2_opt_table + id;
1569 if (opt->get_sb != BCH2_NO_SB_OPT) {
1570 u64 v = bch2_opt_from_sb(sb, id);
1572 prt_printf(out, "%s:", opt->attr.name);
1574 bch2_opt_to_text(out, NULL, sb, opt, v,
1575 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1581 printbuf_indent_sub(out, 2);
1585 prt_printf(out, "layout:");
1587 printbuf_indent_add(out, 2);
1588 bch2_sb_layout_to_text(out, &sb->layout);
1589 printbuf_indent_sub(out, 2);
1592 vstruct_for_each(sb, f)
1593 if (fields & (1 << le32_to_cpu(f->type))) {
1595 bch2_sb_field_to_text(out, sb, f);