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)
104 if (!IS_ERR_OR_NULL(sb->bdev))
105 blkdev_put(sb->bdev, sb->mode);
108 memset(sb, 0, sizeof(*sb));
111 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
113 size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
114 size_t new_buffer_size;
115 struct bch_sb *new_sb;
119 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
121 new_buffer_size = roundup_pow_of_two(new_bytes);
123 if (sb->sb && sb->buffer_size >= new_buffer_size)
126 if (sb->have_layout) {
127 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
129 if (new_bytes > max_bytes) {
130 char buf[BDEVNAME_SIZE];
132 pr_err("%s: superblock too big: want %zu but have %llu",
133 bdevname(sb->bdev, buf), new_bytes, max_bytes);
138 if (sb->buffer_size >= new_buffer_size && sb->sb)
141 if (dynamic_fault("bcachefs:add:super_realloc"))
145 bio = bio_kmalloc(GFP_KERNEL,
146 DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
155 new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
160 sb->buffer_size = new_buffer_size;
165 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
166 enum bch_sb_field_type type,
169 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
170 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
171 ssize_t d = -old_u64s + u64s;
173 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
177 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
181 lockdep_assert_held(&c->sb_lock);
183 /* XXX: we're not checking that offline device have enough space */
185 for_each_online_member(ca, c, i) {
186 struct bch_sb_handle *sb = &ca->disk_sb;
188 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
189 percpu_ref_put(&ca->ref);
195 f = bch2_sb_field_get(sb->sb, type);
196 f = __bch2_sb_field_resize(sb, f, u64s);
198 f->type = cpu_to_le32(type);
202 /* Superblock validate: */
204 static inline void __bch2_sb_layout_size_assert(void)
206 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
209 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
211 u64 offset, prev_offset, max_sectors;
214 if (uuid_le_cmp(layout->magic, BCACHE_MAGIC)) {
215 pr_buf(out, "Not a bcachefs superblock layout");
219 if (layout->layout_type != 0) {
220 pr_buf(out, "Invalid superblock layout type %u",
221 layout->layout_type);
225 if (!layout->nr_superblocks) {
226 pr_buf(out, "Invalid superblock layout: no superblocks");
230 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
231 pr_buf(out, "Invalid superblock layout: too many 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 pr_buf(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);
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 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
272 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
276 if (version_min < bcachefs_metadata_version_min) {
277 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
278 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
282 if (version_min > version) {
283 pr_buf(out, "Bad minimum version %u, greater than version field %u",
284 version_min, version);
288 if (sb->features[1] ||
289 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
290 pr_buf(out, "Filesystem has incompatible features");
294 block_size = le16_to_cpu(sb->block_size);
296 if (block_size > PAGE_SECTORS) {
297 pr_buf(out, "Block size too big (got %u, max %u)",
298 block_size, PAGE_SECTORS);
302 if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
303 pr_buf(out, "Bad user UUID (got zeroes)");
307 if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
308 pr_buf(out, "Bad intenal UUID (got zeroes)");
312 if (!sb->nr_devices ||
313 sb->nr_devices > BCH_SB_MEMBERS_MAX) {
314 pr_buf(out, "Bad number of member devices %u (max %u)",
315 sb->nr_devices, BCH_SB_MEMBERS_MAX);
319 if (sb->dev_idx >= sb->nr_devices) {
320 pr_buf(out, "Bad dev_idx (got %u, nr_devices %u)",
321 sb->dev_idx, sb->nr_devices);
325 if (!sb->time_precision ||
326 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
327 pr_buf(out, "Invalid time precision: %u (min 1, max %lu)",
328 le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
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 pr_buf(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 pr_buf(out, "Invalid superblock: optional with size 0 (type %u)",
367 le32_to_cpu(f->type));
371 if (vstruct_next(f) > vstruct_last(sb)) {
372 pr_buf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
373 le32_to_cpu(f->type));
378 /* members must be validated first: */
379 mi = bch2_sb_get_members(sb);
381 pr_buf(out, "Invalid superblock: member info area 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 void __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++) {
460 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
463 src_f = bch2_sb_field_get(src, i);
464 dst_f = bch2_sb_field_get(dst, i);
465 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
466 src_f ? le32_to_cpu(src_f->u64s) : 0);
469 memcpy(dst_f, src_f, vstruct_bytes(src_f));
473 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
475 struct bch_sb_field_journal *journal_buckets =
476 bch2_sb_get_journal(src);
477 unsigned journal_u64s = journal_buckets
478 ? le32_to_cpu(journal_buckets->field.u64s)
482 lockdep_assert_held(&c->sb_lock);
484 ret = bch2_sb_realloc(&c->disk_sb,
485 le32_to_cpu(src->u64s) - journal_u64s);
489 __copy_super(&c->disk_sb, src);
491 ret = bch2_sb_replicas_to_cpu_replicas(c);
495 ret = bch2_sb_disk_groups_to_cpu(c);
503 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
505 struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
506 struct bch_sb_field_journal *journal_buckets =
507 bch2_sb_get_journal(dst);
508 unsigned journal_u64s = journal_buckets
509 ? le32_to_cpu(journal_buckets->field.u64s)
511 unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
514 ret = bch2_sb_realloc(&ca->disk_sb, u64s);
518 __copy_super(&ca->disk_sb, src);
522 /* read superblock: */
524 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
526 struct bch_csum csum;
527 u32 version, version_min;
532 bio_set_dev(sb->bio, sb->bdev);
533 sb->bio->bi_iter.bi_sector = offset;
534 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
535 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
537 ret = submit_bio_wait(sb->bio);
539 pr_buf(err, "IO error: %i", ret);
543 if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
544 pr_buf(err, "Not a bcachefs superblock");
548 version = le16_to_cpu(sb->sb->version);
549 version_min = version >= bcachefs_metadata_version_bkey_renumber
550 ? le16_to_cpu(sb->sb->version_min)
553 if (version >= bcachefs_metadata_version_max) {
554 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
555 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
559 if (version_min < bcachefs_metadata_version_min) {
560 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
561 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
565 bytes = vstruct_bytes(sb->sb);
567 if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
568 pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
569 bytes, 512UL << sb->sb->layout.sb_max_size_bits);
573 if (bytes > sb->buffer_size) {
574 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
579 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
580 pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
584 /* XXX: verify MACs */
585 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
586 null_nonce(), sb->sb);
588 if (bch2_crc_cmp(csum, sb->sb->csum)) {
589 pr_buf(err, "bad checksum");
593 sb->seq = le64_to_cpu(sb->sb->seq);
598 int bch2_read_super(const char *path, struct bch_opts *opts,
599 struct bch_sb_handle *sb)
601 u64 offset = opt_get(*opts, sb);
602 struct bch_sb_layout layout;
603 struct printbuf err = PRINTBUF;
607 pr_verbose_init(*opts, "");
609 memset(sb, 0, sizeof(*sb));
610 sb->mode = FMODE_READ;
613 if (!opt_get(*opts, noexcl))
614 sb->mode |= FMODE_EXCL;
616 if (!opt_get(*opts, nochanges))
617 sb->mode |= FMODE_WRITE;
619 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
620 if (IS_ERR(sb->bdev) &&
621 PTR_ERR(sb->bdev) == -EACCES &&
622 opt_get(*opts, read_only)) {
623 sb->mode &= ~FMODE_WRITE;
625 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
626 if (!IS_ERR(sb->bdev))
627 opt_set(*opts, nochanges, true);
630 if (IS_ERR(sb->bdev)) {
631 ret = PTR_ERR(sb->bdev);
635 ret = bch2_sb_realloc(sb, 0);
637 pr_buf(&err, "error allocating memory for superblock");
641 if (bch2_fs_init_fault("read_super")) {
642 pr_buf(&err, "dynamic fault");
647 ret = read_one_super(sb, offset, &err);
651 if (opt_defined(*opts, sb))
654 printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
656 printbuf_reset(&err);
659 * Error reading primary superblock - read location of backup
663 bio_set_dev(sb->bio, sb->bdev);
664 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
665 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
667 * use sb buffer to read layout, since sb buffer is page aligned but
670 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
672 ret = submit_bio_wait(sb->bio);
674 pr_buf(&err, "IO error: %i", ret);
678 memcpy(&layout, sb->sb, sizeof(layout));
679 ret = validate_sb_layout(&layout, &err);
683 for (i = layout.sb_offset;
684 i < layout.sb_offset + layout.nr_superblocks; i++) {
685 offset = le64_to_cpu(*i);
687 if (offset == opt_get(*opts, sb))
690 ret = read_one_super(sb, offset, &err);
698 if (le16_to_cpu(sb->sb->block_size) << 9 <
699 bdev_logical_block_size(sb->bdev)) {
700 pr_buf(&err, "block size (%u) smaller than device block size (%u)",
701 le16_to_cpu(sb->sb->block_size) << 9,
702 bdev_logical_block_size(sb->bdev));
708 sb->have_layout = true;
710 ret = bch2_sb_validate(sb, &err, READ);
712 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
717 pr_verbose_init(*opts, "ret %i", ret);
721 printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
728 /* write superblock: */
730 static void write_super_endio(struct bio *bio)
732 struct bch_dev *ca = bio->bi_private;
734 /* XXX: return errors directly */
736 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
737 bch2_blk_status_to_str(bio->bi_status)))
738 ca->sb_write_error = 1;
740 closure_put(&ca->fs->sb_write);
741 percpu_ref_put(&ca->io_ref);
744 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
746 struct bch_sb *sb = ca->disk_sb.sb;
747 struct bio *bio = ca->disk_sb.bio;
750 bio_set_dev(bio, ca->disk_sb.bdev);
751 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
752 bio->bi_end_io = write_super_endio;
753 bio->bi_private = ca;
754 bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
755 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
757 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
760 percpu_ref_get(&ca->io_ref);
761 closure_bio_submit(bio, &c->sb_write);
764 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
766 struct bch_sb *sb = ca->disk_sb.sb;
767 struct bio *bio = ca->disk_sb.bio;
769 sb->offset = sb->layout.sb_offset[idx];
771 SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
772 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
776 bio_set_dev(bio, ca->disk_sb.bdev);
777 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
778 bio->bi_end_io = write_super_endio;
779 bio->bi_private = ca;
780 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
781 bch2_bio_map(bio, sb,
782 roundup((size_t) vstruct_bytes(sb),
783 bdev_logical_block_size(ca->disk_sb.bdev)));
785 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
788 percpu_ref_get(&ca->io_ref);
789 closure_bio_submit(bio, &c->sb_write);
792 int bch2_write_super(struct bch_fs *c)
794 struct closure *cl = &c->sb_write;
796 struct printbuf err = PRINTBUF;
797 unsigned i, sb = 0, nr_wrote;
798 struct bch_devs_mask sb_written;
799 bool wrote, can_mount_without_written, can_mount_with_written;
800 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
803 trace_write_super(c, _RET_IP_);
805 if (c->opts.very_degraded)
806 degraded_flags |= BCH_FORCE_IF_LOST;
808 lockdep_assert_held(&c->sb_lock);
810 closure_init_stack(cl);
811 memset(&sb_written, 0, sizeof(sb_written));
813 le64_add_cpu(&c->disk_sb.sb->seq, 1);
815 if (test_bit(BCH_FS_ERROR, &c->flags))
816 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
817 if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
818 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
820 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
822 bch2_sb_counters_from_cpu(c);
824 for_each_online_member(ca, c, i)
825 bch2_sb_from_fs(c, ca);
827 for_each_online_member(ca, c, i) {
828 printbuf_reset(&err);
830 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
832 bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
833 percpu_ref_put(&ca->io_ref);
838 if (c->opts.nochanges)
842 * Defer writing the superblock until filesystem initialization is
843 * complete - don't write out a partly initialized superblock:
845 if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
848 for_each_online_member(ca, c, i) {
849 __set_bit(ca->dev_idx, sb_written.d);
850 ca->sb_write_error = 0;
853 for_each_online_member(ca, c, i)
854 read_back_super(c, ca);
857 for_each_online_member(ca, c, i) {
858 if (ca->sb_write_error)
861 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
862 bch2_fs_fatal_error(c,
863 "Superblock write was silently dropped! (seq %llu expected %llu)",
864 le64_to_cpu(ca->sb_read_scratch->seq),
866 percpu_ref_put(&ca->io_ref);
871 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
872 bch2_fs_fatal_error(c,
873 "Superblock modified by another process (seq %llu expected %llu)",
874 le64_to_cpu(ca->sb_read_scratch->seq),
876 percpu_ref_put(&ca->io_ref);
884 for_each_online_member(ca, c, i)
885 if (!ca->sb_write_error &&
886 sb < ca->disk_sb.sb->layout.nr_superblocks) {
887 write_one_super(c, ca, sb);
894 for_each_online_member(ca, c, i) {
895 if (ca->sb_write_error)
896 __clear_bit(ca->dev_idx, sb_written.d);
898 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
901 nr_wrote = dev_mask_nr(&sb_written);
903 can_mount_with_written =
904 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
906 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
907 sb_written.d[i] = ~sb_written.d[i];
909 can_mount_without_written =
910 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
913 * If we would be able to mount _without_ the devices we successfully
914 * wrote superblocks to, we weren't able to write to enough devices:
916 * Exception: if we can mount without the successes because we haven't
917 * written anything (new filesystem), we continue if we'd be able to
918 * mount with the devices we did successfully write to:
920 if (bch2_fs_fatal_err_on(!nr_wrote ||
921 !can_mount_with_written ||
922 (can_mount_without_written &&
923 !can_mount_with_written), c,
924 "Unable to write superblock to sufficient devices (from %ps)",
928 /* Make new options visible after they're persistent: */
934 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
936 mutex_lock(&c->sb_lock);
937 if (!(c->sb.features & (1ULL << feat))) {
938 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
942 mutex_unlock(&c->sb_lock);
945 /* BCH_SB_FIELD_members: */
947 static int bch2_sb_members_validate(struct bch_sb *sb,
948 struct bch_sb_field *f,
949 struct printbuf *err)
951 struct bch_sb_field_members *mi = field_to_type(f, members);
954 if ((void *) (mi->members + sb->nr_devices) >
955 vstruct_end(&mi->field)) {
956 pr_buf(err, "too many devices for section size");
960 for (i = 0; i < sb->nr_devices; i++) {
961 struct bch_member *m = mi->members + i;
963 if (!bch2_member_exists(m))
966 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
967 pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
968 i, le64_to_cpu(m->nbuckets), LONG_MAX);
972 if (le64_to_cpu(m->nbuckets) -
973 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
974 pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
975 i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
979 if (le16_to_cpu(m->bucket_size) <
980 le16_to_cpu(sb->block_size)) {
981 pr_buf(err, "device %u: bucket size %u smaller than block size %u",
982 i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
986 if (le16_to_cpu(m->bucket_size) <
987 BCH_SB_BTREE_NODE_SIZE(sb)) {
988 pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
989 i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
997 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
998 struct bch_sb_field *f)
1000 struct bch_sb_field_members *mi = field_to_type(f, members);
1001 struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1004 for (i = 0; i < sb->nr_devices; i++) {
1005 struct bch_member *m = mi->members + i;
1006 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1007 u64 bucket_size = le16_to_cpu(m->bucket_size);
1008 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1010 if (!bch2_member_exists(m))
1013 pr_buf(out, "Device:");
1015 pr_buf(out, "%u", i);
1018 pr_indent_push(out, 2);
1020 pr_buf(out, "UUID:");
1022 pr_uuid(out, m->uuid.b);
1025 pr_buf(out, "Size:");
1027 pr_units(out, device_size, device_size << 9);
1030 pr_buf(out, "Bucket size:");
1032 pr_units(out, bucket_size, bucket_size << 9);
1035 pr_buf(out, "First bucket:");
1037 pr_buf(out, "%u", le16_to_cpu(m->first_bucket));
1040 pr_buf(out, "Buckets:");
1042 pr_buf(out, "%llu", le64_to_cpu(m->nbuckets));
1045 pr_buf(out, "Last mount:");
1048 pr_time(out, le64_to_cpu(m->last_mount));
1050 pr_buf(out, "(never)");
1053 pr_buf(out, "State:");
1056 BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1057 ? bch2_member_states[BCH_MEMBER_STATE(m)]
1061 pr_buf(out, "Group:");
1063 if (BCH_MEMBER_GROUP(m)) {
1064 unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1066 if (idx < disk_groups_nr(gi))
1067 pr_buf(out, "%s (%u)",
1068 gi->entries[idx].label, idx);
1070 pr_buf(out, "(bad disk labels section)");
1072 pr_buf(out, "(none)");
1076 pr_buf(out, "Data allowed:");
1078 if (BCH_MEMBER_DATA_ALLOWED(m))
1079 bch2_flags_to_text(out, bch2_data_types,
1080 BCH_MEMBER_DATA_ALLOWED(m));
1082 pr_buf(out, "(none)");
1085 pr_buf(out, "Has data:");
1088 bch2_flags_to_text(out, bch2_data_types, data_have);
1090 pr_buf(out, "(none)");
1093 pr_buf(out, "Discard:");
1095 pr_buf(out, "%llu", BCH_MEMBER_DISCARD(m));
1098 pr_buf(out, "Freespace initialized:");
1100 pr_buf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1103 pr_indent_pop(out, 2);
1107 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1108 .validate = bch2_sb_members_validate,
1109 .to_text = bch2_sb_members_to_text,
1112 /* BCH_SB_FIELD_crypt: */
1114 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1115 struct bch_sb_field *f,
1116 struct printbuf *err)
1118 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1120 if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1121 pr_buf(err, "wrong size (got %zu should be %zu)",
1122 vstruct_bytes(&crypt->field), sizeof(*crypt));
1126 if (BCH_CRYPT_KDF_TYPE(crypt)) {
1127 pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1134 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1135 struct bch_sb_field *f)
1137 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1139 pr_buf(out, "KFD: %llu", BCH_CRYPT_KDF_TYPE(crypt));
1141 pr_buf(out, "scrypt n: %llu", BCH_KDF_SCRYPT_N(crypt));
1143 pr_buf(out, "scrypt r: %llu", BCH_KDF_SCRYPT_R(crypt));
1145 pr_buf(out, "scrypt p: %llu", BCH_KDF_SCRYPT_P(crypt));
1149 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1150 .validate = bch2_sb_crypt_validate,
1151 .to_text = bch2_sb_crypt_to_text,
1154 /* BCH_SB_FIELD_clean: */
1156 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1158 struct jset_entry *entry;
1161 for (entry = clean->start;
1162 entry < (struct jset_entry *) vstruct_end(&clean->field);
1163 entry = vstruct_next(entry)) {
1164 ret = bch2_journal_entry_validate(c, "superblock", entry,
1165 le16_to_cpu(c->disk_sb.sb->version),
1166 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1175 int bch2_fs_mark_dirty(struct bch_fs *c)
1180 * Unconditionally write superblock, to verify it hasn't changed before
1184 mutex_lock(&c->sb_lock);
1185 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1186 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1187 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1188 ret = bch2_write_super(c);
1189 mutex_unlock(&c->sb_lock);
1194 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1196 struct jset_entry *entry = *end;
1197 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1199 memset(entry, 0, u64s * sizeof(u64));
1201 * The u64s field counts from the start of data, ignoring the shared
1204 entry->u64s = cpu_to_le16(u64s - 1);
1206 *end = vstruct_next(*end);
1210 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1211 struct jset_entry **end,
1217 percpu_down_read(&c->mark_lock);
1220 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1221 bch2_fs_usage_acc_to_base(c, i);
1223 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1227 struct jset_entry_usage *u =
1228 container_of(jset_entry_init(end, sizeof(*u)),
1229 struct jset_entry_usage, entry);
1231 u->entry.type = BCH_JSET_ENTRY_usage;
1232 u->entry.btree_id = BCH_FS_USAGE_inodes;
1233 u->v = cpu_to_le64(c->usage_base->nr_inodes);
1237 struct jset_entry_usage *u =
1238 container_of(jset_entry_init(end, sizeof(*u)),
1239 struct jset_entry_usage, entry);
1241 u->entry.type = BCH_JSET_ENTRY_usage;
1242 u->entry.btree_id = BCH_FS_USAGE_key_version;
1243 u->v = cpu_to_le64(atomic64_read(&c->key_version));
1246 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1247 struct jset_entry_usage *u =
1248 container_of(jset_entry_init(end, sizeof(*u)),
1249 struct jset_entry_usage, entry);
1251 u->entry.type = BCH_JSET_ENTRY_usage;
1252 u->entry.btree_id = BCH_FS_USAGE_reserved;
1254 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1257 for (i = 0; i < c->replicas.nr; i++) {
1258 struct bch_replicas_entry *e =
1259 cpu_replicas_entry(&c->replicas, i);
1260 struct jset_entry_data_usage *u =
1261 container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1262 struct jset_entry_data_usage, entry);
1264 u->entry.type = BCH_JSET_ENTRY_data_usage;
1265 u->v = cpu_to_le64(c->usage_base->replicas[i]);
1266 memcpy(&u->r, e, replicas_entry_bytes(e));
1269 for_each_member_device(ca, c, dev) {
1270 unsigned b = sizeof(struct jset_entry_dev_usage) +
1271 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1272 struct jset_entry_dev_usage *u =
1273 container_of(jset_entry_init(end, b),
1274 struct jset_entry_dev_usage, entry);
1276 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1277 u->dev = cpu_to_le32(dev);
1278 u->buckets_ec = cpu_to_le64(ca->usage_base->buckets_ec);
1280 for (i = 0; i < BCH_DATA_NR; i++) {
1281 u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1282 u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1283 u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1287 percpu_up_read(&c->mark_lock);
1289 for (i = 0; i < 2; i++) {
1290 struct jset_entry_clock *clock =
1291 container_of(jset_entry_init(end, sizeof(*clock)),
1292 struct jset_entry_clock, entry);
1294 clock->entry.type = BCH_JSET_ENTRY_clock;
1296 clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1300 void bch2_fs_mark_clean(struct bch_fs *c)
1302 struct bch_sb_field_clean *sb_clean;
1303 struct jset_entry *entry;
1307 mutex_lock(&c->sb_lock);
1308 if (BCH_SB_CLEAN(c->disk_sb.sb))
1311 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1313 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1314 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1315 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1316 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1318 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1320 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1322 bch_err(c, "error resizing superblock while setting filesystem clean");
1326 sb_clean->flags = 0;
1327 sb_clean->journal_seq = cpu_to_le64(atomic64_read(&c->journal.seq));
1329 /* Trying to catch outstanding bug: */
1330 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1332 entry = sb_clean->start;
1333 bch2_journal_super_entries_add_common(c, &entry, 0);
1334 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1335 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1338 vstruct_end(&sb_clean->field) - (void *) entry);
1341 * this should be in the write path, and we should be validating every
1342 * superblock section:
1344 ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1346 bch_err(c, "error writing marking filesystem clean: validate error");
1350 bch2_write_super(c);
1352 mutex_unlock(&c->sb_lock);
1355 static int bch2_sb_clean_validate(struct bch_sb *sb,
1356 struct bch_sb_field *f,
1357 struct printbuf *err)
1359 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1361 if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1362 pr_buf(err, "wrong size (got %zu should be %zu)",
1363 vstruct_bytes(&clean->field), sizeof(*clean));
1370 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1371 struct bch_sb_field *f)
1373 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1374 struct jset_entry *entry;
1376 pr_buf(out, "flags: %x", le32_to_cpu(clean->flags));
1378 pr_buf(out, "journal_seq: %llu", le64_to_cpu(clean->journal_seq));
1381 for (entry = clean->start;
1382 entry != vstruct_end(&clean->field);
1383 entry = vstruct_next(entry)) {
1384 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1388 bch2_journal_entry_to_text(out, NULL, entry);
1393 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1394 .validate = bch2_sb_clean_validate,
1395 .to_text = bch2_sb_clean_to_text,
1398 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1400 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1405 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1406 struct printbuf *err)
1408 unsigned type = le32_to_cpu(f->type);
1409 struct printbuf field_err = PRINTBUF;
1412 if (type >= BCH_SB_FIELD_NR)
1415 ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1417 pr_buf(err, "Invalid superblock section %s: %s",
1418 bch2_sb_fields[type],
1421 bch2_sb_field_to_text(err, sb, f);
1424 printbuf_exit(&field_err);
1428 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1429 struct bch_sb_field *f)
1431 unsigned type = le32_to_cpu(f->type);
1432 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1433 ? bch2_sb_field_ops[type] : NULL;
1435 if (!out->tabstops[0])
1436 out->tabstops[0] = 32;
1439 pr_buf(out, "%s", bch2_sb_fields[type]);
1441 pr_buf(out, "(unknown field %u)", type);
1443 pr_buf(out, " (size %zu):", vstruct_bytes(f));
1446 if (ops && ops->to_text) {
1447 pr_indent_push(out, 2);
1448 bch2_sb_field_ops[type]->to_text(out, sb, f);
1449 pr_indent_pop(out, 2);
1453 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1457 pr_buf(out, "Type: %u", l->layout_type);
1460 pr_buf(out, "Superblock max size: ");
1462 1 << l->sb_max_size_bits,
1463 512 << l->sb_max_size_bits);
1466 pr_buf(out, "Nr superblocks: %u", l->nr_superblocks);
1469 pr_buf(out, "Offsets: ");
1470 for (i = 0; i < l->nr_superblocks; i++) {
1473 pr_buf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1478 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1479 bool print_layout, unsigned fields)
1481 struct bch_sb_field_members *mi;
1482 struct bch_sb_field *f;
1483 u64 fields_have = 0;
1484 unsigned nr_devices = 0;
1486 if (!out->tabstops[0])
1487 out->tabstops[0] = 32;
1489 mi = bch2_sb_get_members(sb);
1491 struct bch_member *m;
1493 for (m = mi->members;
1494 m < mi->members + sb->nr_devices;
1496 nr_devices += bch2_member_exists(m);
1499 pr_buf(out, "External UUID:");
1501 pr_uuid(out, sb->user_uuid.b);
1504 pr_buf(out, "Internal UUID:");
1506 pr_uuid(out, sb->uuid.b);
1509 pr_buf(out, "Device index:");
1511 pr_buf(out, "%u", sb->dev_idx);
1514 pr_buf(out, "Label:");
1516 pr_buf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1519 pr_buf(out, "Version:");
1521 pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1524 pr_buf(out, "Oldest version on disk:");
1526 pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1529 pr_buf(out, "Created:");
1531 if (sb->time_base_lo)
1532 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1534 pr_buf(out, "(not set)");
1537 pr_buf(out, "Sequence number:");
1539 pr_buf(out, "%llu", le64_to_cpu(sb->seq));
1542 pr_buf(out, "Superblock size:");
1544 pr_buf(out, "%zu", vstruct_bytes(sb));
1547 pr_buf(out, "Clean:");
1549 pr_buf(out, "%llu", BCH_SB_CLEAN(sb));
1552 pr_buf(out, "Devices:");
1554 pr_buf(out, "%u", nr_devices);
1557 pr_buf(out, "Sections:");
1558 vstruct_for_each(sb, f)
1559 fields_have |= 1 << le32_to_cpu(f->type);
1561 bch2_flags_to_text(out, bch2_sb_fields, fields_have);
1564 pr_buf(out, "Features:");
1566 bch2_flags_to_text(out, bch2_sb_features,
1567 le64_to_cpu(sb->features[0]));
1570 pr_buf(out, "Compat features:");
1572 bch2_flags_to_text(out, bch2_sb_compat,
1573 le64_to_cpu(sb->compat[0]));
1577 pr_buf(out, "Options:");
1579 pr_indent_push(out, 2);
1583 for (id = 0; id < bch2_opts_nr; id++) {
1584 const struct bch_option *opt = bch2_opt_table + id;
1586 if (opt->get_sb != BCH2_NO_SB_OPT) {
1587 u64 v = bch2_opt_from_sb(sb, id);
1589 pr_buf(out, "%s:", opt->attr.name);
1591 bch2_opt_to_text(out, NULL, sb, opt, v,
1592 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1598 pr_indent_pop(out, 2);
1602 pr_buf(out, "layout:");
1604 pr_indent_push(out, 2);
1605 bch2_sb_layout_to_text(out, &sb->layout);
1606 pr_indent_pop(out, 2);
1609 vstruct_for_each(sb, f)
1610 if (fields & (1 << le32_to_cpu(f->type))) {
1612 bch2_sb_field_to_text(out, sb, f);