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_seq_blacklist.h"
20 #include <linux/backing-dev.h>
21 #include <linux/sort.h>
23 const char * const bch2_sb_fields[] = {
24 #define x(name, nr) #name,
30 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
33 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
34 enum bch_sb_field_type type)
36 struct bch_sb_field *f;
38 /* XXX: need locking around superblock to access optional fields */
40 vstruct_for_each(sb, f)
41 if (le32_to_cpu(f->type) == type)
46 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
47 struct bch_sb_field *f,
50 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
51 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
53 BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
58 f = vstruct_last(sb->sb);
59 memset(f, 0, sizeof(u64) * u64s);
60 f->u64s = cpu_to_le32(u64s);
68 f->u64s = cpu_to_le32(u64s);
74 memmove(dst, src, vstruct_end(sb->sb) - src);
77 memset(src, 0, dst - src);
80 sb->sb->u64s = cpu_to_le32(sb_u64s);
82 return u64s ? f : NULL;
85 void bch2_sb_field_delete(struct bch_sb_handle *sb,
86 enum bch_sb_field_type type)
88 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
91 __bch2_sb_field_resize(sb, f, 0);
94 /* Superblock realloc/free: */
96 void bch2_free_super(struct bch_sb_handle *sb)
100 if (!IS_ERR_OR_NULL(sb->bdev))
101 blkdev_put(sb->bdev, sb->mode);
104 memset(sb, 0, sizeof(*sb));
107 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
109 size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
110 size_t new_buffer_size;
111 struct bch_sb *new_sb;
115 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
117 new_buffer_size = roundup_pow_of_two(new_bytes);
119 if (sb->sb && sb->buffer_size >= new_buffer_size)
122 if (sb->have_layout) {
123 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
125 if (new_bytes > max_bytes) {
126 char buf[BDEVNAME_SIZE];
128 pr_err("%s: superblock too big: want %zu but have %llu",
129 bdevname(sb->bdev, buf), new_bytes, max_bytes);
134 if (sb->buffer_size >= new_buffer_size && sb->sb)
137 if (dynamic_fault("bcachefs:add:super_realloc"))
141 bio = bio_kmalloc(GFP_KERNEL,
142 DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
151 new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
156 sb->buffer_size = new_buffer_size;
161 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
162 enum bch_sb_field_type type,
165 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
166 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
167 ssize_t d = -old_u64s + u64s;
169 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
173 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
177 lockdep_assert_held(&c->sb_lock);
179 /* XXX: we're not checking that offline device have enough space */
181 for_each_online_member(ca, c, i) {
182 struct bch_sb_handle *sb = &ca->disk_sb;
184 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
185 percpu_ref_put(&ca->ref);
191 f = bch2_sb_field_get(sb->sb, type);
192 f = __bch2_sb_field_resize(sb, f, u64s);
194 f->type = cpu_to_le32(type);
198 /* Superblock validate: */
200 static inline void __bch2_sb_layout_size_assert(void)
202 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
205 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
207 u64 offset, prev_offset, max_sectors;
210 if (uuid_le_cmp(layout->magic, BCACHE_MAGIC)) {
211 pr_buf(out, "Not a bcachefs superblock layout");
215 if (layout->layout_type != 0) {
216 pr_buf(out, "Invalid superblock layout type %u",
217 layout->layout_type);
221 if (!layout->nr_superblocks) {
222 pr_buf(out, "Invalid superblock layout: no superblocks");
226 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
227 pr_buf(out, "Invalid superblock layout: too many superblocks");
231 max_sectors = 1 << layout->sb_max_size_bits;
233 prev_offset = le64_to_cpu(layout->sb_offset[0]);
235 for (i = 1; i < layout->nr_superblocks; i++) {
236 offset = le64_to_cpu(layout->sb_offset[i]);
238 if (offset < prev_offset + max_sectors) {
239 pr_buf(out, "Invalid superblock layout: superblocks overlap\n"
240 " (sb %u ends at %llu next starts at %llu",
241 i - 1, prev_offset + max_sectors, offset);
244 prev_offset = offset;
250 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out)
252 struct bch_sb *sb = disk_sb->sb;
253 struct bch_sb_field *f;
254 struct bch_sb_field_members *mi;
255 u32 version, version_min;
259 version = le16_to_cpu(sb->version);
260 version_min = version >= bcachefs_metadata_version_new_versioning
261 ? le16_to_cpu(sb->version_min)
264 if (version >= bcachefs_metadata_version_max) {
265 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
266 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
270 if (version_min < bcachefs_metadata_version_min) {
271 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
272 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
276 if (version_min > version) {
277 pr_buf(out, "Bad minimum version %u, greater than version field %u",
278 version_min, version);
282 if (sb->features[1] ||
283 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
284 pr_buf(out, "Filesystem has incompatible features");
288 block_size = le16_to_cpu(sb->block_size);
290 if (block_size > PAGE_SECTORS) {
291 pr_buf(out, "Block size too big (got %u, max %u)",
292 block_size, PAGE_SECTORS);
296 if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
297 pr_buf(out, "Bad user UUID (got zeroes)");
301 if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
302 pr_buf(out, "Bad intenal UUID (got zeroes)");
306 if (!sb->nr_devices ||
307 sb->nr_devices > BCH_SB_MEMBERS_MAX) {
308 pr_buf(out, "Bad number of member devices %u (max %u)",
309 sb->nr_devices, BCH_SB_MEMBERS_MAX);
313 if (sb->dev_idx >= sb->nr_devices) {
314 pr_buf(out, "Bad dev_idx (got %u, nr_devices %u)",
315 sb->dev_idx, sb->nr_devices);
319 if (!sb->time_precision ||
320 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
321 pr_buf(out, "Invalid time precision: %u (min 1, max %lu)",
322 le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
326 /* validate layout */
327 ret = validate_sb_layout(&sb->layout, out);
331 vstruct_for_each(sb, f) {
333 pr_buf(out, "Invalid superblock: optional with size 0 (type %u)",
334 le32_to_cpu(f->type));
338 if (vstruct_next(f) > vstruct_last(sb)) {
339 pr_buf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
340 le32_to_cpu(f->type));
345 /* members must be validated first: */
346 mi = bch2_sb_get_members(sb);
348 pr_buf(out, "Invalid superblock: member info area missing");
352 ret = bch2_sb_field_validate(sb, &mi->field, out);
356 vstruct_for_each(sb, f) {
357 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
360 ret = bch2_sb_field_validate(sb, f, out);
370 static void bch2_sb_update(struct bch_fs *c)
372 struct bch_sb *src = c->disk_sb.sb;
373 struct bch_sb_field_members *mi = bch2_sb_get_members(src);
377 lockdep_assert_held(&c->sb_lock);
379 c->sb.uuid = src->uuid;
380 c->sb.user_uuid = src->user_uuid;
381 c->sb.version = le16_to_cpu(src->version);
382 c->sb.version_min = le16_to_cpu(src->version_min);
383 c->sb.nr_devices = src->nr_devices;
384 c->sb.clean = BCH_SB_CLEAN(src);
385 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
387 c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
388 c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
390 /* XXX this is wrong, we need a 96 or 128 bit integer type */
391 c->sb.time_base_lo = div_u64(le64_to_cpu(src->time_base_lo),
392 c->sb.nsec_per_time_unit);
393 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
395 c->sb.features = le64_to_cpu(src->features[0]);
396 c->sb.compat = le64_to_cpu(src->compat[0]);
398 for_each_member_device(ca, c, i)
399 ca->mi = bch2_mi_to_cpu(mi->members + i);
402 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
404 struct bch_sb_field *src_f, *dst_f;
405 struct bch_sb *dst = dst_handle->sb;
408 dst->version = src->version;
409 dst->version_min = src->version_min;
411 dst->uuid = src->uuid;
412 dst->user_uuid = src->user_uuid;
413 memcpy(dst->label, src->label, sizeof(dst->label));
415 dst->block_size = src->block_size;
416 dst->nr_devices = src->nr_devices;
418 dst->time_base_lo = src->time_base_lo;
419 dst->time_base_hi = src->time_base_hi;
420 dst->time_precision = src->time_precision;
422 memcpy(dst->flags, src->flags, sizeof(dst->flags));
423 memcpy(dst->features, src->features, sizeof(dst->features));
424 memcpy(dst->compat, src->compat, sizeof(dst->compat));
426 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
427 if (i == BCH_SB_FIELD_journal)
430 src_f = bch2_sb_field_get(src, i);
431 dst_f = bch2_sb_field_get(dst, i);
432 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
433 src_f ? le32_to_cpu(src_f->u64s) : 0);
436 memcpy(dst_f, src_f, vstruct_bytes(src_f));
440 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
442 struct bch_sb_field_journal *journal_buckets =
443 bch2_sb_get_journal(src);
444 unsigned journal_u64s = journal_buckets
445 ? le32_to_cpu(journal_buckets->field.u64s)
449 lockdep_assert_held(&c->sb_lock);
451 ret = bch2_sb_realloc(&c->disk_sb,
452 le32_to_cpu(src->u64s) - journal_u64s);
456 __copy_super(&c->disk_sb, src);
458 ret = bch2_sb_replicas_to_cpu_replicas(c);
462 ret = bch2_sb_disk_groups_to_cpu(c);
470 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
472 struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
473 struct bch_sb_field_journal *journal_buckets =
474 bch2_sb_get_journal(dst);
475 unsigned journal_u64s = journal_buckets
476 ? le32_to_cpu(journal_buckets->field.u64s)
478 unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
481 ret = bch2_sb_realloc(&ca->disk_sb, u64s);
485 __copy_super(&ca->disk_sb, src);
489 /* read superblock: */
491 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
493 struct bch_csum csum;
494 u32 version, version_min;
499 bio_set_dev(sb->bio, sb->bdev);
500 sb->bio->bi_iter.bi_sector = offset;
501 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
502 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
504 ret = submit_bio_wait(sb->bio);
506 pr_buf(err, "IO error: %i", ret);
510 if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
511 pr_buf(err, "Not a bcachefs superblock");
515 version = le16_to_cpu(sb->sb->version);
516 version_min = version >= bcachefs_metadata_version_new_versioning
517 ? le16_to_cpu(sb->sb->version_min)
520 if (version >= bcachefs_metadata_version_max) {
521 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
522 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
526 if (version_min < bcachefs_metadata_version_min) {
527 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
528 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
532 bytes = vstruct_bytes(sb->sb);
534 if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
535 pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
536 bytes, 512UL << sb->sb->layout.sb_max_size_bits);
540 if (bytes > sb->buffer_size) {
541 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
546 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
547 pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
551 /* XXX: verify MACs */
552 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
553 null_nonce(), sb->sb);
555 if (bch2_crc_cmp(csum, sb->sb->csum)) {
556 pr_buf(err, "bad checksum");
560 sb->seq = le64_to_cpu(sb->sb->seq);
565 int bch2_read_super(const char *path, struct bch_opts *opts,
566 struct bch_sb_handle *sb)
568 u64 offset = opt_get(*opts, sb);
569 struct bch_sb_layout layout;
570 struct printbuf err = PRINTBUF;
574 pr_verbose_init(*opts, "");
576 memset(sb, 0, sizeof(*sb));
577 sb->mode = FMODE_READ;
580 if (!opt_get(*opts, noexcl))
581 sb->mode |= FMODE_EXCL;
583 if (!opt_get(*opts, nochanges))
584 sb->mode |= FMODE_WRITE;
586 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
587 if (IS_ERR(sb->bdev) &&
588 PTR_ERR(sb->bdev) == -EACCES &&
589 opt_get(*opts, read_only)) {
590 sb->mode &= ~FMODE_WRITE;
592 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
593 if (!IS_ERR(sb->bdev))
594 opt_set(*opts, nochanges, true);
597 if (IS_ERR(sb->bdev)) {
598 ret = PTR_ERR(sb->bdev);
602 ret = bch2_sb_realloc(sb, 0);
604 pr_buf(&err, "error allocating memory for superblock");
608 if (bch2_fs_init_fault("read_super")) {
609 pr_buf(&err, "dynamic fault");
614 ret = read_one_super(sb, offset, &err);
618 if (opt_defined(*opts, sb))
621 printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
623 printbuf_reset(&err);
626 * Error reading primary superblock - read location of backup
630 bio_set_dev(sb->bio, sb->bdev);
631 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
632 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
634 * use sb buffer to read layout, since sb buffer is page aligned but
637 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
639 ret = submit_bio_wait(sb->bio);
641 pr_buf(&err, "IO error: %i", ret);
645 memcpy(&layout, sb->sb, sizeof(layout));
646 ret = validate_sb_layout(&layout, &err);
650 for (i = layout.sb_offset;
651 i < layout.sb_offset + layout.nr_superblocks; i++) {
652 offset = le64_to_cpu(*i);
654 if (offset == opt_get(*opts, sb))
657 ret = read_one_super(sb, offset, &err);
665 if (le16_to_cpu(sb->sb->block_size) << 9 <
666 bdev_logical_block_size(sb->bdev)) {
667 pr_buf(&err, "block size (%u) smaller than device block size (%u)",
668 le16_to_cpu(sb->sb->block_size) << 9,
669 bdev_logical_block_size(sb->bdev));
675 sb->have_layout = true;
677 ret = bch2_sb_validate(sb, &err);
679 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
684 pr_verbose_init(*opts, "ret %i", ret);
688 printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
695 /* write superblock: */
697 static void write_super_endio(struct bio *bio)
699 struct bch_dev *ca = bio->bi_private;
701 /* XXX: return errors directly */
703 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
704 bch2_blk_status_to_str(bio->bi_status)))
705 ca->sb_write_error = 1;
707 closure_put(&ca->fs->sb_write);
708 percpu_ref_put(&ca->io_ref);
711 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
713 struct bch_sb *sb = ca->disk_sb.sb;
714 struct bio *bio = ca->disk_sb.bio;
717 bio_set_dev(bio, ca->disk_sb.bdev);
718 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
719 bio->bi_end_io = write_super_endio;
720 bio->bi_private = ca;
721 bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
722 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
724 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
727 percpu_ref_get(&ca->io_ref);
728 closure_bio_submit(bio, &c->sb_write);
731 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
733 struct bch_sb *sb = ca->disk_sb.sb;
734 struct bio *bio = ca->disk_sb.bio;
736 sb->offset = sb->layout.sb_offset[idx];
738 SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
739 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
743 bio_set_dev(bio, ca->disk_sb.bdev);
744 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
745 bio->bi_end_io = write_super_endio;
746 bio->bi_private = ca;
747 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
748 bch2_bio_map(bio, sb,
749 roundup((size_t) vstruct_bytes(sb),
750 bdev_logical_block_size(ca->disk_sb.bdev)));
752 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
755 percpu_ref_get(&ca->io_ref);
756 closure_bio_submit(bio, &c->sb_write);
759 int bch2_write_super(struct bch_fs *c)
761 struct closure *cl = &c->sb_write;
763 struct printbuf err = PRINTBUF;
764 unsigned i, sb = 0, nr_wrote;
765 struct bch_devs_mask sb_written;
766 bool wrote, can_mount_without_written, can_mount_with_written;
767 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
770 if (c->opts.very_degraded)
771 degraded_flags |= BCH_FORCE_IF_LOST;
773 lockdep_assert_held(&c->sb_lock);
775 closure_init_stack(cl);
776 memset(&sb_written, 0, sizeof(sb_written));
778 le64_add_cpu(&c->disk_sb.sb->seq, 1);
780 if (test_bit(BCH_FS_ERROR, &c->flags))
781 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
782 if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
783 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
785 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
787 for_each_online_member(ca, c, i)
788 bch2_sb_from_fs(c, ca);
790 for_each_online_member(ca, c, i) {
791 printbuf_reset(&err);
793 ret = bch2_sb_validate(&ca->disk_sb, &err);
795 bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
796 percpu_ref_put(&ca->io_ref);
801 if (c->opts.nochanges)
804 for_each_online_member(ca, c, i) {
805 __set_bit(ca->dev_idx, sb_written.d);
806 ca->sb_write_error = 0;
809 for_each_online_member(ca, c, i)
810 read_back_super(c, ca);
813 for_each_online_member(ca, c, i) {
814 if (ca->sb_write_error)
817 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
818 bch2_fs_fatal_error(c,
819 "Superblock write was silently dropped! (seq %llu expected %llu)",
820 le64_to_cpu(ca->sb_read_scratch->seq),
822 percpu_ref_put(&ca->io_ref);
827 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
828 bch2_fs_fatal_error(c,
829 "Superblock modified by another process (seq %llu expected %llu)",
830 le64_to_cpu(ca->sb_read_scratch->seq),
832 percpu_ref_put(&ca->io_ref);
840 for_each_online_member(ca, c, i)
841 if (!ca->sb_write_error &&
842 sb < ca->disk_sb.sb->layout.nr_superblocks) {
843 write_one_super(c, ca, sb);
850 for_each_online_member(ca, c, i) {
851 if (ca->sb_write_error)
852 __clear_bit(ca->dev_idx, sb_written.d);
854 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
857 nr_wrote = dev_mask_nr(&sb_written);
859 can_mount_with_written =
860 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
862 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
863 sb_written.d[i] = ~sb_written.d[i];
865 can_mount_without_written =
866 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
869 * If we would be able to mount _without_ the devices we successfully
870 * wrote superblocks to, we weren't able to write to enough devices:
872 * Exception: if we can mount without the successes because we haven't
873 * written anything (new filesystem), we continue if we'd be able to
874 * mount with the devices we did successfully write to:
876 if (bch2_fs_fatal_err_on(!nr_wrote ||
877 !can_mount_with_written ||
878 (can_mount_without_written &&
879 !can_mount_with_written), c,
880 "Unable to write superblock to sufficient devices (from %ps)",
884 /* Make new options visible after they're persistent: */
890 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
892 mutex_lock(&c->sb_lock);
893 if (!(c->sb.features & (1ULL << feat))) {
894 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
898 mutex_unlock(&c->sb_lock);
901 /* BCH_SB_FIELD_journal: */
903 static int u64_cmp(const void *_l, const void *_r)
905 u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
907 return l < r ? -1 : l > r ? 1 : 0;
910 static int bch2_sb_journal_validate(struct bch_sb *sb,
911 struct bch_sb_field *f,
912 struct printbuf *err)
914 struct bch_sb_field_journal *journal = field_to_type(f, journal);
915 struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
921 nr = bch2_nr_journal_buckets(journal);
925 b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
929 for (i = 0; i < nr; i++)
930 b[i] = le64_to_cpu(journal->buckets[i]);
932 sort(b, nr, sizeof(u64), u64_cmp, NULL);
935 pr_buf(err, "journal bucket at sector 0");
939 if (b[0] < le16_to_cpu(m->first_bucket)) {
940 pr_buf(err, "journal bucket %llu before first bucket %u",
941 b[0], le16_to_cpu(m->first_bucket));
945 if (b[nr - 1] >= le64_to_cpu(m->nbuckets)) {
946 pr_buf(err, "journal bucket %llu past end of device (nbuckets %llu)",
947 b[nr - 1], le64_to_cpu(m->nbuckets));
951 for (i = 0; i + 1 < nr; i++)
952 if (b[i] == b[i + 1]) {
953 pr_buf(err, "duplicate journal buckets %llu", b[i]);
963 static void bch2_sb_journal_to_text(struct printbuf *out, struct bch_sb *sb,
964 struct bch_sb_field *f)
966 struct bch_sb_field_journal *journal = field_to_type(f, journal);
967 unsigned i, nr = bch2_nr_journal_buckets(journal);
969 pr_buf(out, "Buckets: ");
970 for (i = 0; i < nr; i++)
971 pr_buf(out, " %llu", le64_to_cpu(journal->buckets[i]));
975 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
976 .validate = bch2_sb_journal_validate,
977 .to_text = bch2_sb_journal_to_text,
980 /* BCH_SB_FIELD_members: */
982 static int bch2_sb_members_validate(struct bch_sb *sb,
983 struct bch_sb_field *f,
984 struct printbuf *err)
986 struct bch_sb_field_members *mi = field_to_type(f, members);
989 if ((void *) (mi->members + sb->nr_devices) >
990 vstruct_end(&mi->field)) {
991 pr_buf(err, "too many devices for section size");
995 for (i = 0; i < sb->nr_devices; i++) {
996 struct bch_member *m = mi->members + i;
998 if (!bch2_member_exists(m))
1001 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
1002 pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
1003 i, le64_to_cpu(m->nbuckets), LONG_MAX);
1007 if (le64_to_cpu(m->nbuckets) -
1008 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
1009 pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
1010 i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
1014 if (le16_to_cpu(m->bucket_size) <
1015 le16_to_cpu(sb->block_size)) {
1016 pr_buf(err, "device %u: bucket size %u smaller than block size %u",
1017 i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
1021 if (le16_to_cpu(m->bucket_size) <
1022 BCH_SB_BTREE_NODE_SIZE(sb)) {
1023 pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
1024 i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
1032 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
1033 struct bch_sb_field *f)
1035 struct bch_sb_field_members *mi = field_to_type(f, members);
1036 struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1039 for (i = 0; i < sb->nr_devices; i++) {
1040 struct bch_member *m = mi->members + i;
1041 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1042 u64 bucket_size = le16_to_cpu(m->bucket_size);
1043 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1045 if (!bch2_member_exists(m))
1048 pr_buf(out, "Device:");
1050 pr_buf(out, "%u", i);
1053 pr_indent_push(out, 2);
1055 pr_buf(out, "UUID:");
1057 pr_uuid(out, m->uuid.b);
1060 pr_buf(out, "Size:");
1062 pr_units(out, device_size, device_size << 9);
1065 pr_buf(out, "Bucket size:");
1067 pr_units(out, bucket_size, bucket_size << 9);
1070 pr_buf(out, "First bucket:");
1072 pr_buf(out, "%u", le16_to_cpu(m->first_bucket));
1075 pr_buf(out, "Buckets:");
1077 pr_buf(out, "%llu", le64_to_cpu(m->nbuckets));
1080 pr_buf(out, "Last mount:");
1083 pr_time(out, le64_to_cpu(m->last_mount));
1085 pr_buf(out, "(never)");
1088 pr_buf(out, "State:");
1091 BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1092 ? bch2_member_states[BCH_MEMBER_STATE(m)]
1096 pr_buf(out, "Group:");
1098 if (BCH_MEMBER_GROUP(m)) {
1099 unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1101 if (idx < disk_groups_nr(gi))
1102 pr_buf(out, "%s (%u)",
1103 gi->entries[idx].label, idx);
1105 pr_buf(out, "(bad disk labels section)");
1107 pr_buf(out, "(none)");
1111 pr_buf(out, "Data allowed:");
1113 if (BCH_MEMBER_DATA_ALLOWED(m))
1114 bch2_flags_to_text(out, bch2_data_types,
1115 BCH_MEMBER_DATA_ALLOWED(m));
1117 pr_buf(out, "(none)");
1120 pr_buf(out, "Has data:");
1123 bch2_flags_to_text(out, bch2_data_types, data_have);
1125 pr_buf(out, "(none)");
1128 pr_buf(out, "Discard:");
1130 pr_buf(out, "%llu", BCH_MEMBER_DISCARD(m));
1133 pr_indent_pop(out, 2);
1137 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1138 .validate = bch2_sb_members_validate,
1139 .to_text = bch2_sb_members_to_text,
1142 /* BCH_SB_FIELD_crypt: */
1144 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1145 struct bch_sb_field *f,
1146 struct printbuf *err)
1148 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1150 if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1151 pr_buf(err, "wrong size (got %zu should be %zu)",
1152 vstruct_bytes(&crypt->field), sizeof(*crypt));
1156 if (BCH_CRYPT_KDF_TYPE(crypt)) {
1157 pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1164 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1165 struct bch_sb_field *f)
1167 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1169 pr_buf(out, "KFD: %llu", BCH_CRYPT_KDF_TYPE(crypt));
1171 pr_buf(out, "scrypt n: %llu", BCH_KDF_SCRYPT_N(crypt));
1173 pr_buf(out, "scrypt r: %llu", BCH_KDF_SCRYPT_R(crypt));
1175 pr_buf(out, "scrypt p: %llu", BCH_KDF_SCRYPT_P(crypt));
1179 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1180 .validate = bch2_sb_crypt_validate,
1181 .to_text = bch2_sb_crypt_to_text,
1184 /* BCH_SB_FIELD_clean: */
1186 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1188 struct jset_entry *entry;
1191 for (entry = clean->start;
1192 entry < (struct jset_entry *) vstruct_end(&clean->field);
1193 entry = vstruct_next(entry)) {
1194 ret = bch2_journal_entry_validate(c, "superblock", entry,
1195 le16_to_cpu(c->disk_sb.sb->version),
1196 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1205 int bch2_fs_mark_dirty(struct bch_fs *c)
1210 * Unconditionally write superblock, to verify it hasn't changed before
1214 mutex_lock(&c->sb_lock);
1215 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1216 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1217 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1218 ret = bch2_write_super(c);
1219 mutex_unlock(&c->sb_lock);
1224 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1226 struct jset_entry *entry = *end;
1227 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1229 memset(entry, 0, u64s * sizeof(u64));
1231 * The u64s field counts from the start of data, ignoring the shared
1234 entry->u64s = cpu_to_le16(u64s - 1);
1236 *end = vstruct_next(*end);
1240 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1241 struct jset_entry **end,
1247 percpu_down_read(&c->mark_lock);
1250 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1251 bch2_fs_usage_acc_to_base(c, i);
1253 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1257 struct jset_entry_usage *u =
1258 container_of(jset_entry_init(end, sizeof(*u)),
1259 struct jset_entry_usage, entry);
1261 u->entry.type = BCH_JSET_ENTRY_usage;
1262 u->entry.btree_id = BCH_FS_USAGE_inodes;
1263 u->v = cpu_to_le64(c->usage_base->nr_inodes);
1267 struct jset_entry_usage *u =
1268 container_of(jset_entry_init(end, sizeof(*u)),
1269 struct jset_entry_usage, entry);
1271 u->entry.type = BCH_JSET_ENTRY_usage;
1272 u->entry.btree_id = BCH_FS_USAGE_key_version;
1273 u->v = cpu_to_le64(atomic64_read(&c->key_version));
1276 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1277 struct jset_entry_usage *u =
1278 container_of(jset_entry_init(end, sizeof(*u)),
1279 struct jset_entry_usage, entry);
1281 u->entry.type = BCH_JSET_ENTRY_usage;
1282 u->entry.btree_id = BCH_FS_USAGE_reserved;
1284 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1287 for (i = 0; i < c->replicas.nr; i++) {
1288 struct bch_replicas_entry *e =
1289 cpu_replicas_entry(&c->replicas, i);
1290 struct jset_entry_data_usage *u =
1291 container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1292 struct jset_entry_data_usage, entry);
1294 u->entry.type = BCH_JSET_ENTRY_data_usage;
1295 u->v = cpu_to_le64(c->usage_base->replicas[i]);
1296 memcpy(&u->r, e, replicas_entry_bytes(e));
1299 for_each_member_device(ca, c, dev) {
1300 unsigned b = sizeof(struct jset_entry_dev_usage) +
1301 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1302 struct jset_entry_dev_usage *u =
1303 container_of(jset_entry_init(end, b),
1304 struct jset_entry_dev_usage, entry);
1306 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1307 u->dev = cpu_to_le32(dev);
1308 u->buckets_ec = cpu_to_le64(ca->usage_base->buckets_ec);
1309 u->buckets_unavailable = cpu_to_le64(ca->usage_base->buckets_unavailable);
1311 for (i = 0; i < BCH_DATA_NR; i++) {
1312 u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1313 u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1314 u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1318 percpu_up_read(&c->mark_lock);
1320 for (i = 0; i < 2; i++) {
1321 struct jset_entry_clock *clock =
1322 container_of(jset_entry_init(end, sizeof(*clock)),
1323 struct jset_entry_clock, entry);
1325 clock->entry.type = BCH_JSET_ENTRY_clock;
1327 clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1331 void bch2_fs_mark_clean(struct bch_fs *c)
1333 struct bch_sb_field_clean *sb_clean;
1334 struct jset_entry *entry;
1338 mutex_lock(&c->sb_lock);
1339 if (BCH_SB_CLEAN(c->disk_sb.sb))
1342 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1344 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1345 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1346 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1347 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1349 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1351 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1353 bch_err(c, "error resizing superblock while setting filesystem clean");
1357 sb_clean->flags = 0;
1358 sb_clean->journal_seq = cpu_to_le64(atomic64_read(&c->journal.seq));
1360 /* Trying to catch outstanding bug: */
1361 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1363 entry = sb_clean->start;
1364 bch2_journal_super_entries_add_common(c, &entry, 0);
1365 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1366 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1369 vstruct_end(&sb_clean->field) - (void *) entry);
1372 * this should be in the write path, and we should be validating every
1373 * superblock section:
1375 ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1377 bch_err(c, "error writing marking filesystem clean: validate error");
1381 bch2_write_super(c);
1383 mutex_unlock(&c->sb_lock);
1386 static int bch2_sb_clean_validate(struct bch_sb *sb,
1387 struct bch_sb_field *f,
1388 struct printbuf *err)
1390 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1392 if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1393 pr_buf(err, "wrong size (got %zu should be %zu)",
1394 vstruct_bytes(&clean->field), sizeof(*clean));
1401 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1402 struct bch_sb_field *f)
1404 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1405 struct jset_entry *entry;
1407 pr_buf(out, "flags: %x", le32_to_cpu(clean->flags));
1409 pr_buf(out, "journal_seq: %llu", le64_to_cpu(clean->journal_seq));
1412 for (entry = clean->start;
1413 entry != vstruct_end(&clean->field);
1414 entry = vstruct_next(entry)) {
1415 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1419 bch2_journal_entry_to_text(out, NULL, entry);
1424 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1425 .validate = bch2_sb_clean_validate,
1426 .to_text = bch2_sb_clean_to_text,
1429 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1431 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1436 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1437 struct printbuf *err)
1439 unsigned type = le32_to_cpu(f->type);
1440 struct printbuf field_err = PRINTBUF;
1443 if (type >= BCH_SB_FIELD_NR)
1446 ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1448 pr_buf(err, "Invalid superblock section %s: %s",
1449 bch2_sb_fields[type],
1452 bch2_sb_field_to_text(err, sb, f);
1455 printbuf_exit(&field_err);
1459 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1460 struct bch_sb_field *f)
1462 unsigned type = le32_to_cpu(f->type);
1463 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1464 ? bch2_sb_field_ops[type] : NULL;
1466 if (!out->tabstops[0])
1467 out->tabstops[0] = 32;
1470 pr_buf(out, "%s", bch2_sb_fields[type]);
1472 pr_buf(out, "(unknown field %u)", type);
1474 pr_buf(out, " (size %zu):", vstruct_bytes(f));
1477 if (ops && ops->to_text) {
1478 pr_indent_push(out, 2);
1479 bch2_sb_field_ops[type]->to_text(out, sb, f);
1480 pr_indent_pop(out, 2);
1484 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1488 pr_buf(out, "Type: %u", l->layout_type);
1491 pr_buf(out, "Superblock max size: ");
1493 1 << l->sb_max_size_bits,
1494 512 << l->sb_max_size_bits);
1497 pr_buf(out, "Nr superblocks: %u", l->nr_superblocks);
1500 pr_buf(out, "Offsets: ");
1501 for (i = 0; i < l->nr_superblocks; i++) {
1504 pr_buf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1509 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1510 bool print_layout, unsigned fields)
1512 struct bch_sb_field_members *mi;
1513 struct bch_sb_field *f;
1514 u64 fields_have = 0;
1515 unsigned nr_devices = 0;
1517 if (!out->tabstops[0])
1518 out->tabstops[0] = 32;
1520 mi = bch2_sb_get_members(sb);
1522 struct bch_member *m;
1524 for (m = mi->members;
1525 m < mi->members + sb->nr_devices;
1527 nr_devices += bch2_member_exists(m);
1530 pr_buf(out, "External UUID:");
1532 pr_uuid(out, sb->user_uuid.b);
1535 pr_buf(out, "Internal UUID:");
1537 pr_uuid(out, sb->uuid.b);
1540 pr_buf(out, "Device index:");
1542 pr_buf(out, "%u", sb->dev_idx);
1545 pr_buf(out, "Label:");
1547 pr_buf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1550 pr_buf(out, "Version:");
1552 pr_buf(out, "%u", le16_to_cpu(sb->version));
1555 pr_buf(out, "Oldest version on disk:");
1557 pr_buf(out, "%u", le16_to_cpu(sb->version_min));
1560 pr_buf(out, "Created:");
1562 if (sb->time_base_lo)
1563 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1565 pr_buf(out, "(not set)");
1568 pr_buf(out, "Sequence number:");
1570 pr_buf(out, "%llu", le64_to_cpu(sb->seq));
1573 pr_buf(out, "Superblock size:");
1575 pr_buf(out, "%zu", vstruct_bytes(sb));
1578 pr_buf(out, "Clean:");
1580 pr_buf(out, "%llu", BCH_SB_CLEAN(sb));
1583 pr_buf(out, "Devices:");
1585 pr_buf(out, "%u", nr_devices);
1588 pr_buf(out, "Sections:");
1589 vstruct_for_each(sb, f)
1590 fields_have |= 1 << le32_to_cpu(f->type);
1592 bch2_flags_to_text(out, bch2_sb_fields, fields_have);
1595 pr_buf(out, "Features:");
1597 bch2_flags_to_text(out, bch2_sb_features,
1598 le64_to_cpu(sb->features[0]));
1601 pr_buf(out, "Compat features:");
1603 bch2_flags_to_text(out, bch2_sb_compat,
1604 le64_to_cpu(sb->compat[0]));
1608 pr_buf(out, "Options:");
1610 pr_indent_push(out, 2);
1614 for (id = 0; id < bch2_opts_nr; id++) {
1615 const struct bch_option *opt = bch2_opt_table + id;
1617 if (opt->get_sb != BCH2_NO_SB_OPT) {
1618 u64 v = bch2_opt_from_sb(sb, id);
1620 pr_buf(out, "%s:", opt->attr.name);
1622 bch2_opt_to_text(out, NULL, opt, v, OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1628 pr_indent_pop(out, 2);
1632 pr_buf(out, "layout:");
1634 pr_indent_push(out, 2);
1635 bch2_sb_layout_to_text(out, &sb->layout);
1636 pr_indent_pop(out, 2);
1639 vstruct_for_each(sb, f)
1640 if (fields & (1 << le32_to_cpu(f->type))) {
1642 bch2_sb_field_to_text(out, sb, f);