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"
21 #include <linux/backing-dev.h>
22 #include <linux/sort.h>
24 #include <trace/events/bcachefs.h>
26 const char * const bch2_sb_fields[] = {
27 #define x(name, nr) #name,
33 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
36 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
37 enum bch_sb_field_type type)
39 struct bch_sb_field *f;
41 /* XXX: need locking around superblock to access optional fields */
43 vstruct_for_each(sb, f)
44 if (le32_to_cpu(f->type) == type)
49 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
50 struct bch_sb_field *f,
53 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
54 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
56 BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
61 f = vstruct_last(sb->sb);
62 memset(f, 0, sizeof(u64) * u64s);
63 f->u64s = cpu_to_le32(u64s);
71 f->u64s = cpu_to_le32(u64s);
77 memmove(dst, src, vstruct_end(sb->sb) - src);
80 memset(src, 0, dst - src);
83 sb->sb->u64s = cpu_to_le32(sb_u64s);
85 return u64s ? f : NULL;
88 void bch2_sb_field_delete(struct bch_sb_handle *sb,
89 enum bch_sb_field_type type)
91 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
94 __bch2_sb_field_resize(sb, f, 0);
97 /* Superblock realloc/free: */
99 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 char buf[BDEVNAME_SIZE];
131 pr_err("%s: superblock too big: want %zu but have %llu",
132 bdevname(sb->bdev, buf), new_bytes, max_bytes);
137 if (sb->buffer_size >= new_buffer_size && sb->sb)
140 if (dynamic_fault("bcachefs:add:super_realloc"))
144 bio = bio_kmalloc(GFP_KERNEL,
145 DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
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 pr_buf(out, "Not a bcachefs superblock layout");
218 if (layout->layout_type != 0) {
219 pr_buf(out, "Invalid superblock layout type %u",
220 layout->layout_type);
224 if (!layout->nr_superblocks) {
225 pr_buf(out, "Invalid superblock layout: no superblocks");
229 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
230 pr_buf(out, "Invalid superblock layout: too many superblocks");
234 max_sectors = 1 << layout->sb_max_size_bits;
236 prev_offset = le64_to_cpu(layout->sb_offset[0]);
238 for (i = 1; i < layout->nr_superblocks; i++) {
239 offset = le64_to_cpu(layout->sb_offset[i]);
241 if (offset < prev_offset + max_sectors) {
242 pr_buf(out, "Invalid superblock layout: superblocks overlap\n"
243 " (sb %u ends at %llu next starts at %llu",
244 i - 1, prev_offset + max_sectors, offset);
247 prev_offset = offset;
253 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
256 struct bch_sb *sb = disk_sb->sb;
257 struct bch_sb_field *f;
258 struct bch_sb_field_members *mi;
259 enum bch_opt_id opt_id;
260 u32 version, version_min;
264 version = le16_to_cpu(sb->version);
265 version_min = version >= bcachefs_metadata_version_bkey_renumber
266 ? le16_to_cpu(sb->version_min)
269 if (version >= bcachefs_metadata_version_max) {
270 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
271 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
275 if (version_min < bcachefs_metadata_version_min) {
276 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
277 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
281 if (version_min > version) {
282 pr_buf(out, "Bad minimum version %u, greater than version field %u",
283 version_min, version);
287 if (sb->features[1] ||
288 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
289 pr_buf(out, "Filesystem has incompatible features");
293 block_size = le16_to_cpu(sb->block_size);
295 if (block_size > PAGE_SECTORS) {
296 pr_buf(out, "Block size too big (got %u, max %u)",
297 block_size, PAGE_SECTORS);
301 if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
302 pr_buf(out, "Bad user UUID (got zeroes)");
306 if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
307 pr_buf(out, "Bad intenal UUID (got zeroes)");
311 if (!sb->nr_devices ||
312 sb->nr_devices > BCH_SB_MEMBERS_MAX) {
313 pr_buf(out, "Bad number of member devices %u (max %u)",
314 sb->nr_devices, BCH_SB_MEMBERS_MAX);
318 if (sb->dev_idx >= sb->nr_devices) {
319 pr_buf(out, "Bad dev_idx (got %u, nr_devices %u)",
320 sb->dev_idx, sb->nr_devices);
324 if (!sb->time_precision ||
325 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
326 pr_buf(out, "Invalid time precision: %u (min 1, max %lu)",
327 le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
333 * Been seeing a bug where these are getting inexplicably
334 * zeroed, so we'r now validating them, but we have to be
335 * careful not to preven people's filesystems from mounting:
337 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
338 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
339 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
340 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
343 for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
344 const struct bch_option *opt = bch2_opt_table + opt_id;
346 if (opt->get_sb != BCH2_NO_SB_OPT) {
347 u64 v = bch2_opt_from_sb(sb, opt_id);
349 pr_buf(out, "Invalid option ");
350 ret = bch2_opt_validate(opt, v, out);
358 /* validate layout */
359 ret = validate_sb_layout(&sb->layout, out);
363 vstruct_for_each(sb, f) {
365 pr_buf(out, "Invalid superblock: optional with size 0 (type %u)",
366 le32_to_cpu(f->type));
370 if (vstruct_next(f) > vstruct_last(sb)) {
371 pr_buf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
372 le32_to_cpu(f->type));
377 /* members must be validated first: */
378 mi = bch2_sb_get_members(sb);
380 pr_buf(out, "Invalid superblock: member info area missing");
384 ret = bch2_sb_field_validate(sb, &mi->field, out);
388 vstruct_for_each(sb, f) {
389 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
392 ret = bch2_sb_field_validate(sb, f, out);
402 static void bch2_sb_update(struct bch_fs *c)
404 struct bch_sb *src = c->disk_sb.sb;
405 struct bch_sb_field_members *mi = bch2_sb_get_members(src);
409 lockdep_assert_held(&c->sb_lock);
411 c->sb.uuid = src->uuid;
412 c->sb.user_uuid = src->user_uuid;
413 c->sb.version = le16_to_cpu(src->version);
414 c->sb.version_min = le16_to_cpu(src->version_min);
415 c->sb.nr_devices = src->nr_devices;
416 c->sb.clean = BCH_SB_CLEAN(src);
417 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
419 c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
420 c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
422 /* XXX this is wrong, we need a 96 or 128 bit integer type */
423 c->sb.time_base_lo = div_u64(le64_to_cpu(src->time_base_lo),
424 c->sb.nsec_per_time_unit);
425 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
427 c->sb.features = le64_to_cpu(src->features[0]);
428 c->sb.compat = le64_to_cpu(src->compat[0]);
430 for_each_member_device(ca, c, i)
431 ca->mi = bch2_mi_to_cpu(mi->members + i);
434 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
436 struct bch_sb_field *src_f, *dst_f;
437 struct bch_sb *dst = dst_handle->sb;
440 dst->version = src->version;
441 dst->version_min = src->version_min;
443 dst->uuid = src->uuid;
444 dst->user_uuid = src->user_uuid;
445 memcpy(dst->label, src->label, sizeof(dst->label));
447 dst->block_size = src->block_size;
448 dst->nr_devices = src->nr_devices;
450 dst->time_base_lo = src->time_base_lo;
451 dst->time_base_hi = src->time_base_hi;
452 dst->time_precision = src->time_precision;
454 memcpy(dst->flags, src->flags, sizeof(dst->flags));
455 memcpy(dst->features, src->features, sizeof(dst->features));
456 memcpy(dst->compat, src->compat, sizeof(dst->compat));
458 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
459 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
462 src_f = bch2_sb_field_get(src, i);
463 dst_f = bch2_sb_field_get(dst, i);
464 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
465 src_f ? le32_to_cpu(src_f->u64s) : 0);
468 memcpy(dst_f, src_f, vstruct_bytes(src_f));
472 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
474 struct bch_sb_field_journal *journal_buckets =
475 bch2_sb_get_journal(src);
476 unsigned journal_u64s = journal_buckets
477 ? le32_to_cpu(journal_buckets->field.u64s)
481 lockdep_assert_held(&c->sb_lock);
483 ret = bch2_sb_realloc(&c->disk_sb,
484 le32_to_cpu(src->u64s) - journal_u64s);
488 __copy_super(&c->disk_sb, src);
490 ret = bch2_sb_replicas_to_cpu_replicas(c);
494 ret = bch2_sb_disk_groups_to_cpu(c);
502 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
504 struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
505 struct bch_sb_field_journal *journal_buckets =
506 bch2_sb_get_journal(dst);
507 unsigned journal_u64s = journal_buckets
508 ? le32_to_cpu(journal_buckets->field.u64s)
510 unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
513 ret = bch2_sb_realloc(&ca->disk_sb, u64s);
517 __copy_super(&ca->disk_sb, src);
521 /* read superblock: */
523 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
525 struct bch_csum csum;
526 u32 version, version_min;
531 bio_set_dev(sb->bio, sb->bdev);
532 sb->bio->bi_iter.bi_sector = offset;
533 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
534 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
536 ret = submit_bio_wait(sb->bio);
538 pr_buf(err, "IO error: %i", ret);
542 if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
543 pr_buf(err, "Not a bcachefs superblock");
547 version = le16_to_cpu(sb->sb->version);
548 version_min = version >= bcachefs_metadata_version_bkey_renumber
549 ? le16_to_cpu(sb->sb->version_min)
552 if (version >= bcachefs_metadata_version_max) {
553 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
554 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
558 if (version_min < bcachefs_metadata_version_min) {
559 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
560 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
564 bytes = vstruct_bytes(sb->sb);
566 if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
567 pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
568 bytes, 512UL << sb->sb->layout.sb_max_size_bits);
572 if (bytes > sb->buffer_size) {
573 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
578 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
579 pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
583 /* XXX: verify MACs */
584 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
585 null_nonce(), sb->sb);
587 if (bch2_crc_cmp(csum, sb->sb->csum)) {
588 pr_buf(err, "bad checksum");
592 sb->seq = le64_to_cpu(sb->sb->seq);
597 int bch2_read_super(const char *path, struct bch_opts *opts,
598 struct bch_sb_handle *sb)
600 u64 offset = opt_get(*opts, sb);
601 struct bch_sb_layout layout;
602 struct printbuf err = PRINTBUF;
606 pr_verbose_init(*opts, "");
608 memset(sb, 0, sizeof(*sb));
609 sb->mode = FMODE_READ;
612 if (!opt_get(*opts, noexcl))
613 sb->mode |= FMODE_EXCL;
615 if (!opt_get(*opts, nochanges))
616 sb->mode |= FMODE_WRITE;
618 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
619 if (IS_ERR(sb->bdev) &&
620 PTR_ERR(sb->bdev) == -EACCES &&
621 opt_get(*opts, read_only)) {
622 sb->mode &= ~FMODE_WRITE;
624 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
625 if (!IS_ERR(sb->bdev))
626 opt_set(*opts, nochanges, true);
629 if (IS_ERR(sb->bdev)) {
630 ret = PTR_ERR(sb->bdev);
634 ret = bch2_sb_realloc(sb, 0);
636 pr_buf(&err, "error allocating memory for superblock");
640 if (bch2_fs_init_fault("read_super")) {
641 pr_buf(&err, "dynamic fault");
646 ret = read_one_super(sb, offset, &err);
650 if (opt_defined(*opts, sb))
653 printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
655 printbuf_reset(&err);
658 * Error reading primary superblock - read location of backup
662 bio_set_dev(sb->bio, sb->bdev);
663 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
664 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
666 * use sb buffer to read layout, since sb buffer is page aligned but
669 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
671 ret = submit_bio_wait(sb->bio);
673 pr_buf(&err, "IO error: %i", ret);
677 memcpy(&layout, sb->sb, sizeof(layout));
678 ret = validate_sb_layout(&layout, &err);
682 for (i = layout.sb_offset;
683 i < layout.sb_offset + layout.nr_superblocks; i++) {
684 offset = le64_to_cpu(*i);
686 if (offset == opt_get(*opts, sb))
689 ret = read_one_super(sb, offset, &err);
697 if (le16_to_cpu(sb->sb->block_size) << 9 <
698 bdev_logical_block_size(sb->bdev)) {
699 pr_buf(&err, "block size (%u) smaller than device block size (%u)",
700 le16_to_cpu(sb->sb->block_size) << 9,
701 bdev_logical_block_size(sb->bdev));
707 sb->have_layout = true;
709 ret = bch2_sb_validate(sb, &err, READ);
711 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
716 pr_verbose_init(*opts, "ret %i", ret);
720 printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
727 /* write superblock: */
729 static void write_super_endio(struct bio *bio)
731 struct bch_dev *ca = bio->bi_private;
733 /* XXX: return errors directly */
735 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
736 bch2_blk_status_to_str(bio->bi_status)))
737 ca->sb_write_error = 1;
739 closure_put(&ca->fs->sb_write);
740 percpu_ref_put(&ca->io_ref);
743 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
745 struct bch_sb *sb = ca->disk_sb.sb;
746 struct bio *bio = ca->disk_sb.bio;
749 bio_set_dev(bio, ca->disk_sb.bdev);
750 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
751 bio->bi_end_io = write_super_endio;
752 bio->bi_private = ca;
753 bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
754 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
756 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
759 percpu_ref_get(&ca->io_ref);
760 closure_bio_submit(bio, &c->sb_write);
763 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
765 struct bch_sb *sb = ca->disk_sb.sb;
766 struct bio *bio = ca->disk_sb.bio;
768 sb->offset = sb->layout.sb_offset[idx];
770 SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
771 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
775 bio_set_dev(bio, ca->disk_sb.bdev);
776 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
777 bio->bi_end_io = write_super_endio;
778 bio->bi_private = ca;
779 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
780 bch2_bio_map(bio, sb,
781 roundup((size_t) vstruct_bytes(sb),
782 bdev_logical_block_size(ca->disk_sb.bdev)));
784 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
787 percpu_ref_get(&ca->io_ref);
788 closure_bio_submit(bio, &c->sb_write);
791 int bch2_write_super(struct bch_fs *c)
793 struct closure *cl = &c->sb_write;
795 struct printbuf err = PRINTBUF;
796 unsigned i, sb = 0, nr_wrote;
797 struct bch_devs_mask sb_written;
798 bool wrote, can_mount_without_written, can_mount_with_written;
799 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
802 trace_write_super(c, _RET_IP_);
804 if (c->opts.very_degraded)
805 degraded_flags |= BCH_FORCE_IF_LOST;
807 lockdep_assert_held(&c->sb_lock);
809 closure_init_stack(cl);
810 memset(&sb_written, 0, sizeof(sb_written));
812 le64_add_cpu(&c->disk_sb.sb->seq, 1);
814 if (test_bit(BCH_FS_ERROR, &c->flags))
815 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
816 if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
817 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
819 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
821 for_each_online_member(ca, c, i)
822 bch2_sb_from_fs(c, ca);
824 for_each_online_member(ca, c, i) {
825 printbuf_reset(&err);
827 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
829 bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
830 percpu_ref_put(&ca->io_ref);
835 if (c->opts.nochanges)
839 * Defer writing the superblock until filesystem initialization is
840 * complete - don't write out a partly initialized superblock:
842 if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
845 for_each_online_member(ca, c, i) {
846 __set_bit(ca->dev_idx, sb_written.d);
847 ca->sb_write_error = 0;
850 for_each_online_member(ca, c, i)
851 read_back_super(c, ca);
854 for_each_online_member(ca, c, i) {
855 if (ca->sb_write_error)
858 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
859 bch2_fs_fatal_error(c,
860 "Superblock write was silently dropped! (seq %llu expected %llu)",
861 le64_to_cpu(ca->sb_read_scratch->seq),
863 percpu_ref_put(&ca->io_ref);
868 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
869 bch2_fs_fatal_error(c,
870 "Superblock modified by another process (seq %llu expected %llu)",
871 le64_to_cpu(ca->sb_read_scratch->seq),
873 percpu_ref_put(&ca->io_ref);
881 for_each_online_member(ca, c, i)
882 if (!ca->sb_write_error &&
883 sb < ca->disk_sb.sb->layout.nr_superblocks) {
884 write_one_super(c, ca, sb);
891 for_each_online_member(ca, c, i) {
892 if (ca->sb_write_error)
893 __clear_bit(ca->dev_idx, sb_written.d);
895 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
898 nr_wrote = dev_mask_nr(&sb_written);
900 can_mount_with_written =
901 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
903 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
904 sb_written.d[i] = ~sb_written.d[i];
906 can_mount_without_written =
907 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
910 * If we would be able to mount _without_ the devices we successfully
911 * wrote superblocks to, we weren't able to write to enough devices:
913 * Exception: if we can mount without the successes because we haven't
914 * written anything (new filesystem), we continue if we'd be able to
915 * mount with the devices we did successfully write to:
917 if (bch2_fs_fatal_err_on(!nr_wrote ||
918 !can_mount_with_written ||
919 (can_mount_without_written &&
920 !can_mount_with_written), c,
921 "Unable to write superblock to sufficient devices (from %ps)",
925 /* Make new options visible after they're persistent: */
931 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
933 mutex_lock(&c->sb_lock);
934 if (!(c->sb.features & (1ULL << feat))) {
935 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
939 mutex_unlock(&c->sb_lock);
942 /* BCH_SB_FIELD_members: */
944 static int bch2_sb_members_validate(struct bch_sb *sb,
945 struct bch_sb_field *f,
946 struct printbuf *err)
948 struct bch_sb_field_members *mi = field_to_type(f, members);
951 if ((void *) (mi->members + sb->nr_devices) >
952 vstruct_end(&mi->field)) {
953 pr_buf(err, "too many devices for section size");
957 for (i = 0; i < sb->nr_devices; i++) {
958 struct bch_member *m = mi->members + i;
960 if (!bch2_member_exists(m))
963 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
964 pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
965 i, le64_to_cpu(m->nbuckets), LONG_MAX);
969 if (le64_to_cpu(m->nbuckets) -
970 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
971 pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
972 i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
976 if (le16_to_cpu(m->bucket_size) <
977 le16_to_cpu(sb->block_size)) {
978 pr_buf(err, "device %u: bucket size %u smaller than block size %u",
979 i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
983 if (le16_to_cpu(m->bucket_size) <
984 BCH_SB_BTREE_NODE_SIZE(sb)) {
985 pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
986 i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
994 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
995 struct bch_sb_field *f)
997 struct bch_sb_field_members *mi = field_to_type(f, members);
998 struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1001 for (i = 0; i < sb->nr_devices; i++) {
1002 struct bch_member *m = mi->members + i;
1003 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1004 u64 bucket_size = le16_to_cpu(m->bucket_size);
1005 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1007 if (!bch2_member_exists(m))
1010 pr_buf(out, "Device:");
1012 pr_buf(out, "%u", i);
1015 pr_indent_push(out, 2);
1017 pr_buf(out, "UUID:");
1019 pr_uuid(out, m->uuid.b);
1022 pr_buf(out, "Size:");
1024 pr_units(out, device_size, device_size << 9);
1027 pr_buf(out, "Bucket size:");
1029 pr_units(out, bucket_size, bucket_size << 9);
1032 pr_buf(out, "First bucket:");
1034 pr_buf(out, "%u", le16_to_cpu(m->first_bucket));
1037 pr_buf(out, "Buckets:");
1039 pr_buf(out, "%llu", le64_to_cpu(m->nbuckets));
1042 pr_buf(out, "Last mount:");
1045 pr_time(out, le64_to_cpu(m->last_mount));
1047 pr_buf(out, "(never)");
1050 pr_buf(out, "State:");
1053 BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1054 ? bch2_member_states[BCH_MEMBER_STATE(m)]
1058 pr_buf(out, "Group:");
1060 if (BCH_MEMBER_GROUP(m)) {
1061 unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1063 if (idx < disk_groups_nr(gi))
1064 pr_buf(out, "%s (%u)",
1065 gi->entries[idx].label, idx);
1067 pr_buf(out, "(bad disk labels section)");
1069 pr_buf(out, "(none)");
1073 pr_buf(out, "Data allowed:");
1075 if (BCH_MEMBER_DATA_ALLOWED(m))
1076 bch2_flags_to_text(out, bch2_data_types,
1077 BCH_MEMBER_DATA_ALLOWED(m));
1079 pr_buf(out, "(none)");
1082 pr_buf(out, "Has data:");
1085 bch2_flags_to_text(out, bch2_data_types, data_have);
1087 pr_buf(out, "(none)");
1090 pr_buf(out, "Discard:");
1092 pr_buf(out, "%llu", BCH_MEMBER_DISCARD(m));
1095 pr_buf(out, "Freespace initialized:");
1097 pr_buf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1100 pr_indent_pop(out, 2);
1104 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1105 .validate = bch2_sb_members_validate,
1106 .to_text = bch2_sb_members_to_text,
1109 /* BCH_SB_FIELD_crypt: */
1111 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1112 struct bch_sb_field *f,
1113 struct printbuf *err)
1115 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1117 if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1118 pr_buf(err, "wrong size (got %zu should be %zu)",
1119 vstruct_bytes(&crypt->field), sizeof(*crypt));
1123 if (BCH_CRYPT_KDF_TYPE(crypt)) {
1124 pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1131 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1132 struct bch_sb_field *f)
1134 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1136 pr_buf(out, "KFD: %llu", BCH_CRYPT_KDF_TYPE(crypt));
1138 pr_buf(out, "scrypt n: %llu", BCH_KDF_SCRYPT_N(crypt));
1140 pr_buf(out, "scrypt r: %llu", BCH_KDF_SCRYPT_R(crypt));
1142 pr_buf(out, "scrypt p: %llu", BCH_KDF_SCRYPT_P(crypt));
1146 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1147 .validate = bch2_sb_crypt_validate,
1148 .to_text = bch2_sb_crypt_to_text,
1151 /* BCH_SB_FIELD_clean: */
1153 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1155 struct jset_entry *entry;
1158 for (entry = clean->start;
1159 entry < (struct jset_entry *) vstruct_end(&clean->field);
1160 entry = vstruct_next(entry)) {
1161 ret = bch2_journal_entry_validate(c, "superblock", entry,
1162 le16_to_cpu(c->disk_sb.sb->version),
1163 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1172 int bch2_fs_mark_dirty(struct bch_fs *c)
1177 * Unconditionally write superblock, to verify it hasn't changed before
1181 mutex_lock(&c->sb_lock);
1182 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1183 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1184 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1185 ret = bch2_write_super(c);
1186 mutex_unlock(&c->sb_lock);
1191 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1193 struct jset_entry *entry = *end;
1194 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1196 memset(entry, 0, u64s * sizeof(u64));
1198 * The u64s field counts from the start of data, ignoring the shared
1201 entry->u64s = cpu_to_le16(u64s - 1);
1203 *end = vstruct_next(*end);
1207 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1208 struct jset_entry **end,
1214 percpu_down_read(&c->mark_lock);
1217 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1218 bch2_fs_usage_acc_to_base(c, i);
1220 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1224 struct jset_entry_usage *u =
1225 container_of(jset_entry_init(end, sizeof(*u)),
1226 struct jset_entry_usage, entry);
1228 u->entry.type = BCH_JSET_ENTRY_usage;
1229 u->entry.btree_id = BCH_FS_USAGE_inodes;
1230 u->v = cpu_to_le64(c->usage_base->nr_inodes);
1234 struct jset_entry_usage *u =
1235 container_of(jset_entry_init(end, sizeof(*u)),
1236 struct jset_entry_usage, entry);
1238 u->entry.type = BCH_JSET_ENTRY_usage;
1239 u->entry.btree_id = BCH_FS_USAGE_key_version;
1240 u->v = cpu_to_le64(atomic64_read(&c->key_version));
1243 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1244 struct jset_entry_usage *u =
1245 container_of(jset_entry_init(end, sizeof(*u)),
1246 struct jset_entry_usage, entry);
1248 u->entry.type = BCH_JSET_ENTRY_usage;
1249 u->entry.btree_id = BCH_FS_USAGE_reserved;
1251 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1254 for (i = 0; i < c->replicas.nr; i++) {
1255 struct bch_replicas_entry *e =
1256 cpu_replicas_entry(&c->replicas, i);
1257 struct jset_entry_data_usage *u =
1258 container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1259 struct jset_entry_data_usage, entry);
1261 u->entry.type = BCH_JSET_ENTRY_data_usage;
1262 u->v = cpu_to_le64(c->usage_base->replicas[i]);
1263 memcpy(&u->r, e, replicas_entry_bytes(e));
1266 for_each_member_device(ca, c, dev) {
1267 unsigned b = sizeof(struct jset_entry_dev_usage) +
1268 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1269 struct jset_entry_dev_usage *u =
1270 container_of(jset_entry_init(end, b),
1271 struct jset_entry_dev_usage, entry);
1273 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1274 u->dev = cpu_to_le32(dev);
1275 u->buckets_ec = cpu_to_le64(ca->usage_base->buckets_ec);
1277 for (i = 0; i < BCH_DATA_NR; i++) {
1278 u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1279 u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1280 u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1284 percpu_up_read(&c->mark_lock);
1286 for (i = 0; i < 2; i++) {
1287 struct jset_entry_clock *clock =
1288 container_of(jset_entry_init(end, sizeof(*clock)),
1289 struct jset_entry_clock, entry);
1291 clock->entry.type = BCH_JSET_ENTRY_clock;
1293 clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1297 void bch2_fs_mark_clean(struct bch_fs *c)
1299 struct bch_sb_field_clean *sb_clean;
1300 struct jset_entry *entry;
1304 mutex_lock(&c->sb_lock);
1305 if (BCH_SB_CLEAN(c->disk_sb.sb))
1308 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1310 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1311 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1312 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1313 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1315 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1317 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1319 bch_err(c, "error resizing superblock while setting filesystem clean");
1323 sb_clean->flags = 0;
1324 sb_clean->journal_seq = cpu_to_le64(atomic64_read(&c->journal.seq));
1326 /* Trying to catch outstanding bug: */
1327 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1329 entry = sb_clean->start;
1330 bch2_journal_super_entries_add_common(c, &entry, 0);
1331 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1332 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1335 vstruct_end(&sb_clean->field) - (void *) entry);
1338 * this should be in the write path, and we should be validating every
1339 * superblock section:
1341 ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1343 bch_err(c, "error writing marking filesystem clean: validate error");
1347 bch2_write_super(c);
1349 mutex_unlock(&c->sb_lock);
1352 static int bch2_sb_clean_validate(struct bch_sb *sb,
1353 struct bch_sb_field *f,
1354 struct printbuf *err)
1356 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1358 if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1359 pr_buf(err, "wrong size (got %zu should be %zu)",
1360 vstruct_bytes(&clean->field), sizeof(*clean));
1367 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1368 struct bch_sb_field *f)
1370 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1371 struct jset_entry *entry;
1373 pr_buf(out, "flags: %x", le32_to_cpu(clean->flags));
1375 pr_buf(out, "journal_seq: %llu", le64_to_cpu(clean->journal_seq));
1378 for (entry = clean->start;
1379 entry != vstruct_end(&clean->field);
1380 entry = vstruct_next(entry)) {
1381 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1385 bch2_journal_entry_to_text(out, NULL, entry);
1390 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1391 .validate = bch2_sb_clean_validate,
1392 .to_text = bch2_sb_clean_to_text,
1395 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1397 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1402 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1403 struct printbuf *err)
1405 unsigned type = le32_to_cpu(f->type);
1406 struct printbuf field_err = PRINTBUF;
1409 if (type >= BCH_SB_FIELD_NR)
1412 ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1414 pr_buf(err, "Invalid superblock section %s: %s",
1415 bch2_sb_fields[type],
1418 bch2_sb_field_to_text(err, sb, f);
1421 printbuf_exit(&field_err);
1425 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1426 struct bch_sb_field *f)
1428 unsigned type = le32_to_cpu(f->type);
1429 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1430 ? bch2_sb_field_ops[type] : NULL;
1432 if (!out->tabstops[0])
1433 out->tabstops[0] = 32;
1436 pr_buf(out, "%s", bch2_sb_fields[type]);
1438 pr_buf(out, "(unknown field %u)", type);
1440 pr_buf(out, " (size %zu):", vstruct_bytes(f));
1443 if (ops && ops->to_text) {
1444 pr_indent_push(out, 2);
1445 bch2_sb_field_ops[type]->to_text(out, sb, f);
1446 pr_indent_pop(out, 2);
1450 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1454 pr_buf(out, "Type: %u", l->layout_type);
1457 pr_buf(out, "Superblock max size: ");
1459 1 << l->sb_max_size_bits,
1460 512 << l->sb_max_size_bits);
1463 pr_buf(out, "Nr superblocks: %u", l->nr_superblocks);
1466 pr_buf(out, "Offsets: ");
1467 for (i = 0; i < l->nr_superblocks; i++) {
1470 pr_buf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1475 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1476 bool print_layout, unsigned fields)
1478 struct bch_sb_field_members *mi;
1479 struct bch_sb_field *f;
1480 u64 fields_have = 0;
1481 unsigned nr_devices = 0;
1483 if (!out->tabstops[0])
1484 out->tabstops[0] = 32;
1486 mi = bch2_sb_get_members(sb);
1488 struct bch_member *m;
1490 for (m = mi->members;
1491 m < mi->members + sb->nr_devices;
1493 nr_devices += bch2_member_exists(m);
1496 pr_buf(out, "External UUID:");
1498 pr_uuid(out, sb->user_uuid.b);
1501 pr_buf(out, "Internal UUID:");
1503 pr_uuid(out, sb->uuid.b);
1506 pr_buf(out, "Device index:");
1508 pr_buf(out, "%u", sb->dev_idx);
1511 pr_buf(out, "Label:");
1513 pr_buf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1516 pr_buf(out, "Version:");
1518 pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1521 pr_buf(out, "Oldest version on disk:");
1523 pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1526 pr_buf(out, "Created:");
1528 if (sb->time_base_lo)
1529 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1531 pr_buf(out, "(not set)");
1534 pr_buf(out, "Sequence number:");
1536 pr_buf(out, "%llu", le64_to_cpu(sb->seq));
1539 pr_buf(out, "Superblock size:");
1541 pr_buf(out, "%zu", vstruct_bytes(sb));
1544 pr_buf(out, "Clean:");
1546 pr_buf(out, "%llu", BCH_SB_CLEAN(sb));
1549 pr_buf(out, "Devices:");
1551 pr_buf(out, "%u", nr_devices);
1554 pr_buf(out, "Sections:");
1555 vstruct_for_each(sb, f)
1556 fields_have |= 1 << le32_to_cpu(f->type);
1558 bch2_flags_to_text(out, bch2_sb_fields, fields_have);
1561 pr_buf(out, "Features:");
1563 bch2_flags_to_text(out, bch2_sb_features,
1564 le64_to_cpu(sb->features[0]));
1567 pr_buf(out, "Compat features:");
1569 bch2_flags_to_text(out, bch2_sb_compat,
1570 le64_to_cpu(sb->compat[0]));
1574 pr_buf(out, "Options:");
1576 pr_indent_push(out, 2);
1580 for (id = 0; id < bch2_opts_nr; id++) {
1581 const struct bch_option *opt = bch2_opt_table + id;
1583 if (opt->get_sb != BCH2_NO_SB_OPT) {
1584 u64 v = bch2_opt_from_sb(sb, id);
1586 pr_buf(out, "%s:", opt->attr.name);
1588 bch2_opt_to_text(out, NULL, sb, opt, v,
1589 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1595 pr_indent_pop(out, 2);
1599 pr_buf(out, "layout:");
1601 pr_indent_push(out, 2);
1602 bch2_sb_layout_to_text(out, &sb->layout);
1603 pr_indent_pop(out, 2);
1606 vstruct_for_each(sb, f)
1607 if (fields & (1 << le32_to_cpu(f->type))) {
1609 bch2_sb_field_to_text(out, sb, f);