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 if (BCH_SB_INITIALIZED(c->disk_sb.sb))
459 set_bit(BCH_FS_INITIALIZED, &c->flags);
461 ret = bch2_sb_replicas_to_cpu_replicas(c);
465 ret = bch2_sb_disk_groups_to_cpu(c);
473 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
475 struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
476 struct bch_sb_field_journal *journal_buckets =
477 bch2_sb_get_journal(dst);
478 unsigned journal_u64s = journal_buckets
479 ? le32_to_cpu(journal_buckets->field.u64s)
481 unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
484 ret = bch2_sb_realloc(&ca->disk_sb, u64s);
488 __copy_super(&ca->disk_sb, src);
492 /* read superblock: */
494 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
496 struct bch_csum csum;
497 u32 version, version_min;
502 bio_set_dev(sb->bio, sb->bdev);
503 sb->bio->bi_iter.bi_sector = offset;
504 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
505 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
507 ret = submit_bio_wait(sb->bio);
509 pr_buf(err, "IO error: %i", ret);
513 if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
514 pr_buf(err, "Not a bcachefs superblock");
518 version = le16_to_cpu(sb->sb->version);
519 version_min = version >= bcachefs_metadata_version_new_versioning
520 ? le16_to_cpu(sb->sb->version_min)
523 if (version >= bcachefs_metadata_version_max) {
524 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
525 version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
529 if (version_min < bcachefs_metadata_version_min) {
530 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
531 version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
535 bytes = vstruct_bytes(sb->sb);
537 if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
538 pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
539 bytes, 512UL << sb->sb->layout.sb_max_size_bits);
543 if (bytes > sb->buffer_size) {
544 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
549 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
550 pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
554 /* XXX: verify MACs */
555 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
556 null_nonce(), sb->sb);
558 if (bch2_crc_cmp(csum, sb->sb->csum)) {
559 pr_buf(err, "bad checksum");
563 sb->seq = le64_to_cpu(sb->sb->seq);
568 int bch2_read_super(const char *path, struct bch_opts *opts,
569 struct bch_sb_handle *sb)
571 u64 offset = opt_get(*opts, sb);
572 struct bch_sb_layout layout;
578 _err = kmalloc(4096, GFP_KERNEL);
581 err = _PBUF(_err, 4096);
583 pr_verbose_init(*opts, "");
585 memset(sb, 0, sizeof(*sb));
586 sb->mode = FMODE_READ;
589 if (!opt_get(*opts, noexcl))
590 sb->mode |= FMODE_EXCL;
592 if (!opt_get(*opts, nochanges))
593 sb->mode |= FMODE_WRITE;
595 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
596 if (IS_ERR(sb->bdev) &&
597 PTR_ERR(sb->bdev) == -EACCES &&
598 opt_get(*opts, read_only)) {
599 sb->mode &= ~FMODE_WRITE;
601 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
602 if (!IS_ERR(sb->bdev))
603 opt_set(*opts, nochanges, true);
606 if (IS_ERR(sb->bdev)) {
607 ret = PTR_ERR(sb->bdev);
611 ret = bch2_sb_realloc(sb, 0);
613 pr_buf(&err, "error allocating memory for superblock");
617 if (bch2_fs_init_fault("read_super")) {
618 pr_buf(&err, "dynamic fault");
623 ret = read_one_super(sb, offset, &err);
627 if (opt_defined(*opts, sb))
630 printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
632 err = _PBUF(_err, 4096);
635 * Error reading primary superblock - read location of backup
639 bio_set_dev(sb->bio, sb->bdev);
640 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
641 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
643 * use sb buffer to read layout, since sb buffer is page aligned but
646 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
648 ret = submit_bio_wait(sb->bio);
650 pr_buf(&err, "IO error: %i", ret);
654 memcpy(&layout, sb->sb, sizeof(layout));
655 ret = validate_sb_layout(&layout, &err);
659 for (i = layout.sb_offset;
660 i < layout.sb_offset + layout.nr_superblocks; i++) {
661 offset = le64_to_cpu(*i);
663 if (offset == opt_get(*opts, sb))
666 ret = read_one_super(sb, offset, &err);
674 if (le16_to_cpu(sb->sb->block_size) << 9 <
675 bdev_logical_block_size(sb->bdev)) {
676 pr_buf(&err, "block size (%u) smaller than device block size (%u)",
677 le16_to_cpu(sb->sb->block_size) << 9,
678 bdev_logical_block_size(sb->bdev));
684 sb->have_layout = true;
686 ret = bch2_sb_validate(sb, &err);
688 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
693 pr_verbose_init(*opts, "ret %i", ret);
697 printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
704 /* write superblock: */
706 static void write_super_endio(struct bio *bio)
708 struct bch_dev *ca = bio->bi_private;
710 /* XXX: return errors directly */
712 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
713 bch2_blk_status_to_str(bio->bi_status)))
714 ca->sb_write_error = 1;
716 closure_put(&ca->fs->sb_write);
717 percpu_ref_put(&ca->io_ref);
720 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
722 struct bch_sb *sb = ca->disk_sb.sb;
723 struct bio *bio = ca->disk_sb.bio;
726 bio_set_dev(bio, ca->disk_sb.bdev);
727 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
728 bio->bi_end_io = write_super_endio;
729 bio->bi_private = ca;
730 bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
731 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
733 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
736 percpu_ref_get(&ca->io_ref);
737 closure_bio_submit(bio, &c->sb_write);
740 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
742 struct bch_sb *sb = ca->disk_sb.sb;
743 struct bio *bio = ca->disk_sb.bio;
745 sb->offset = sb->layout.sb_offset[idx];
747 SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
748 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
752 bio_set_dev(bio, ca->disk_sb.bdev);
753 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
754 bio->bi_end_io = write_super_endio;
755 bio->bi_private = ca;
756 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
757 bch2_bio_map(bio, sb,
758 roundup((size_t) vstruct_bytes(sb),
759 bdev_logical_block_size(ca->disk_sb.bdev)));
761 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
764 percpu_ref_get(&ca->io_ref);
765 closure_bio_submit(bio, &c->sb_write);
768 int bch2_write_super(struct bch_fs *c)
770 struct closure *cl = &c->sb_write;
772 unsigned i, sb = 0, nr_wrote;
773 struct bch_devs_mask sb_written;
774 bool wrote, can_mount_without_written, can_mount_with_written;
775 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
778 if (c->opts.very_degraded)
779 degraded_flags |= BCH_FORCE_IF_LOST;
781 lockdep_assert_held(&c->sb_lock);
783 closure_init_stack(cl);
784 memset(&sb_written, 0, sizeof(sb_written));
786 le64_add_cpu(&c->disk_sb.sb->seq, 1);
788 if (test_bit(BCH_FS_ERROR, &c->flags))
789 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
790 if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
791 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
793 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
795 for_each_online_member(ca, c, i)
796 bch2_sb_from_fs(c, ca);
798 for_each_online_member(ca, c, i) {
799 struct printbuf buf = { NULL, NULL };
801 ret = bch2_sb_validate(&ca->disk_sb, &buf);
803 char *_buf = kmalloc(4096, GFP_NOFS);
805 buf = _PBUF(_buf, 4096);
806 bch2_sb_validate(&ca->disk_sb, &buf);
809 bch2_fs_inconsistent(c, "sb invalid before write: %s", _buf);
811 percpu_ref_put(&ca->io_ref);
816 if (c->opts.nochanges)
819 for_each_online_member(ca, c, i) {
820 __set_bit(ca->dev_idx, sb_written.d);
821 ca->sb_write_error = 0;
824 for_each_online_member(ca, c, i)
825 read_back_super(c, ca);
828 for_each_online_member(ca, c, i) {
829 if (ca->sb_write_error)
832 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
833 bch2_fs_fatal_error(c,
834 "Superblock write was silently dropped! (seq %llu expected %llu)",
835 le64_to_cpu(ca->sb_read_scratch->seq),
837 percpu_ref_put(&ca->io_ref);
842 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
843 bch2_fs_fatal_error(c,
844 "Superblock modified by another process (seq %llu expected %llu)",
845 le64_to_cpu(ca->sb_read_scratch->seq),
847 percpu_ref_put(&ca->io_ref);
855 for_each_online_member(ca, c, i)
856 if (!ca->sb_write_error &&
857 sb < ca->disk_sb.sb->layout.nr_superblocks) {
858 write_one_super(c, ca, sb);
865 for_each_online_member(ca, c, i) {
866 if (ca->sb_write_error)
867 __clear_bit(ca->dev_idx, sb_written.d);
869 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
872 nr_wrote = dev_mask_nr(&sb_written);
874 can_mount_with_written =
875 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
877 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
878 sb_written.d[i] = ~sb_written.d[i];
880 can_mount_without_written =
881 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
884 * If we would be able to mount _without_ the devices we successfully
885 * wrote superblocks to, we weren't able to write to enough devices:
887 * Exception: if we can mount without the successes because we haven't
888 * written anything (new filesystem), we continue if we'd be able to
889 * mount with the devices we did successfully write to:
891 if (bch2_fs_fatal_err_on(!nr_wrote ||
892 !can_mount_with_written ||
893 (can_mount_without_written &&
894 !can_mount_with_written), c,
895 "Unable to write superblock to sufficient devices (from %ps)",
899 /* Make new options visible after they're persistent: */
904 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
906 mutex_lock(&c->sb_lock);
907 if (!(c->sb.features & (1ULL << feat))) {
908 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
912 mutex_unlock(&c->sb_lock);
915 /* BCH_SB_FIELD_journal: */
917 static int u64_cmp(const void *_l, const void *_r)
919 u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
921 return l < r ? -1 : l > r ? 1 : 0;
924 static int bch2_sb_validate_journal(struct bch_sb *sb,
925 struct bch_sb_field *f,
926 struct printbuf *err)
928 struct bch_sb_field_journal *journal = field_to_type(f, journal);
929 struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
935 nr = bch2_nr_journal_buckets(journal);
939 b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
943 for (i = 0; i < nr; i++)
944 b[i] = le64_to_cpu(journal->buckets[i]);
946 sort(b, nr, sizeof(u64), u64_cmp, NULL);
949 pr_buf(err, "journal bucket at sector 0");
953 if (b[0] < le16_to_cpu(m->first_bucket)) {
954 pr_buf(err, "journal bucket %llu before first bucket %u",
955 b[0], le16_to_cpu(m->first_bucket));
959 if (b[nr - 1] >= le64_to_cpu(m->nbuckets)) {
960 pr_buf(err, "journal bucket %llu past end of device (nbuckets %llu)",
961 b[nr - 1], le64_to_cpu(m->nbuckets));
965 for (i = 0; i + 1 < nr; i++)
966 if (b[i] == b[i + 1]) {
967 pr_buf(err, "duplicate journal buckets %llu", b[i]);
977 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
978 .validate = bch2_sb_validate_journal,
981 /* BCH_SB_FIELD_members: */
983 static int bch2_sb_validate_members(struct bch_sb *sb,
984 struct bch_sb_field *f,
985 struct printbuf *err)
987 struct bch_sb_field_members *mi = field_to_type(f, members);
990 if ((void *) (mi->members + sb->nr_devices) >
991 vstruct_end(&mi->field)) {
992 pr_buf(err, "too many devices for section size");
996 for (i = 0; i < sb->nr_devices; i++) {
997 struct bch_member *m = mi->members + i;
999 if (!bch2_member_exists(m))
1002 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
1003 pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
1004 i, le64_to_cpu(m->nbuckets), LONG_MAX);
1008 if (le64_to_cpu(m->nbuckets) -
1009 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
1010 pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
1011 i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
1015 if (le16_to_cpu(m->bucket_size) <
1016 le16_to_cpu(sb->block_size)) {
1017 pr_buf(err, "device %u: bucket size %u smaller than block size %u",
1018 i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
1022 if (le16_to_cpu(m->bucket_size) <
1023 BCH_SB_BTREE_NODE_SIZE(sb)) {
1024 pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
1025 i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
1033 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1034 .validate = bch2_sb_validate_members,
1037 /* BCH_SB_FIELD_crypt: */
1039 static int bch2_sb_validate_crypt(struct bch_sb *sb,
1040 struct bch_sb_field *f,
1041 struct printbuf *err)
1043 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1045 if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1046 pr_buf(err, "wrong size (got %llu should be %zu)",
1047 vstruct_bytes(&crypt->field), sizeof(*crypt));
1051 if (BCH_CRYPT_KDF_TYPE(crypt)) {
1052 pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1059 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1060 .validate = bch2_sb_validate_crypt,
1063 /* BCH_SB_FIELD_clean: */
1065 int bch2_sb_clean_validate(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1067 struct jset_entry *entry;
1070 for (entry = clean->start;
1071 entry < (struct jset_entry *) vstruct_end(&clean->field);
1072 entry = vstruct_next(entry)) {
1073 ret = bch2_journal_entry_validate(c, "superblock", entry,
1074 le16_to_cpu(c->disk_sb.sb->version),
1075 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1084 int bch2_fs_mark_dirty(struct bch_fs *c)
1089 * Unconditionally write superblock, to verify it hasn't changed before
1093 mutex_lock(&c->sb_lock);
1094 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1095 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1096 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1097 ret = bch2_write_super(c);
1098 mutex_unlock(&c->sb_lock);
1103 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1105 struct jset_entry *entry = *end;
1106 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1108 memset(entry, 0, u64s * sizeof(u64));
1110 * The u64s field counts from the start of data, ignoring the shared
1113 entry->u64s = cpu_to_le16(u64s - 1);
1115 *end = vstruct_next(*end);
1119 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1120 struct jset_entry **end,
1126 percpu_down_read(&c->mark_lock);
1129 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1130 bch2_fs_usage_acc_to_base(c, i);
1132 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1136 struct jset_entry_usage *u =
1137 container_of(jset_entry_init(end, sizeof(*u)),
1138 struct jset_entry_usage, entry);
1140 u->entry.type = BCH_JSET_ENTRY_usage;
1141 u->entry.btree_id = BCH_FS_USAGE_inodes;
1142 u->v = cpu_to_le64(c->usage_base->nr_inodes);
1146 struct jset_entry_usage *u =
1147 container_of(jset_entry_init(end, sizeof(*u)),
1148 struct jset_entry_usage, entry);
1150 u->entry.type = BCH_JSET_ENTRY_usage;
1151 u->entry.btree_id = BCH_FS_USAGE_key_version;
1152 u->v = cpu_to_le64(atomic64_read(&c->key_version));
1155 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1156 struct jset_entry_usage *u =
1157 container_of(jset_entry_init(end, sizeof(*u)),
1158 struct jset_entry_usage, entry);
1160 u->entry.type = BCH_JSET_ENTRY_usage;
1161 u->entry.btree_id = BCH_FS_USAGE_reserved;
1163 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1166 for (i = 0; i < c->replicas.nr; i++) {
1167 struct bch_replicas_entry *e =
1168 cpu_replicas_entry(&c->replicas, i);
1169 struct jset_entry_data_usage *u =
1170 container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1171 struct jset_entry_data_usage, entry);
1173 u->entry.type = BCH_JSET_ENTRY_data_usage;
1174 u->v = cpu_to_le64(c->usage_base->replicas[i]);
1175 memcpy(&u->r, e, replicas_entry_bytes(e));
1178 for_each_member_device(ca, c, dev) {
1179 unsigned b = sizeof(struct jset_entry_dev_usage) +
1180 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1181 struct jset_entry_dev_usage *u =
1182 container_of(jset_entry_init(end, b),
1183 struct jset_entry_dev_usage, entry);
1185 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1186 u->dev = cpu_to_le32(dev);
1187 u->buckets_ec = cpu_to_le64(ca->usage_base->buckets_ec);
1188 u->buckets_unavailable = cpu_to_le64(ca->usage_base->buckets_unavailable);
1190 for (i = 0; i < BCH_DATA_NR; i++) {
1191 u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1192 u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1193 u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1197 percpu_up_read(&c->mark_lock);
1199 for (i = 0; i < 2; i++) {
1200 struct jset_entry_clock *clock =
1201 container_of(jset_entry_init(end, sizeof(*clock)),
1202 struct jset_entry_clock, entry);
1204 clock->entry.type = BCH_JSET_ENTRY_clock;
1206 clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1210 void bch2_fs_mark_clean(struct bch_fs *c)
1212 struct bch_sb_field_clean *sb_clean;
1213 struct jset_entry *entry;
1217 mutex_lock(&c->sb_lock);
1218 if (BCH_SB_CLEAN(c->disk_sb.sb))
1221 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1223 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1224 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1225 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1226 c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1228 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1230 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1232 bch_err(c, "error resizing superblock while setting filesystem clean");
1236 sb_clean->flags = 0;
1237 sb_clean->journal_seq = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1239 /* Trying to catch outstanding bug: */
1240 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1242 entry = sb_clean->start;
1243 bch2_journal_super_entries_add_common(c, &entry, 0);
1244 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1245 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1248 vstruct_end(&sb_clean->field) - (void *) entry);
1251 * this should be in the write path, and we should be validating every
1252 * superblock section:
1254 ret = bch2_sb_clean_validate(c, sb_clean, WRITE);
1256 bch_err(c, "error writing marking filesystem clean: validate error");
1260 bch2_write_super(c);
1262 mutex_unlock(&c->sb_lock);
1265 static int bch2_sb_validate_clean(struct bch_sb *sb,
1266 struct bch_sb_field *f,
1267 struct printbuf *err)
1269 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1271 if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1272 pr_buf(err, "wrong size (got %llu should be %zu)",
1273 vstruct_bytes(&clean->field), sizeof(*clean));
1280 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1281 .validate = bch2_sb_validate_clean,
1284 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1286 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1291 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1292 struct printbuf *orig_err)
1294 unsigned type = le32_to_cpu(f->type);
1295 struct printbuf err = *orig_err;
1298 if (type >= BCH_SB_FIELD_NR)
1301 pr_buf(&err, "Invalid superblock section %s: ", bch2_sb_fields[type]);
1303 ret = bch2_sb_field_ops[type]->validate(sb, f, &err);
1306 bch2_sb_field_to_text(&err, sb, f);
1313 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1314 struct bch_sb_field *f)
1316 unsigned type = le32_to_cpu(f->type);
1317 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1318 ? bch2_sb_field_ops[type] : NULL;
1321 pr_buf(out, "%s", bch2_sb_fields[type]);
1323 pr_buf(out, "(unknown field %u)", type);
1325 pr_buf(out, " (size %llu):", vstruct_bytes(f));
1327 if (ops && ops->to_text)
1328 bch2_sb_field_ops[type]->to_text(out, sb, f);