1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_update_interior.h"
7 #include "disk_groups.h"
12 #include "journal_seq_blacklist.h"
19 #include <linux/backing-dev.h>
20 #include <linux/sort.h>
22 const char * const bch2_sb_fields[] = {
23 #define x(name, nr) #name,
29 static const char *bch2_sb_field_validate(struct bch_sb *,
30 struct bch_sb_field *);
32 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
33 enum bch_sb_field_type type)
35 struct bch_sb_field *f;
37 /* XXX: need locking around superblock to access optional fields */
39 vstruct_for_each(sb, f)
40 if (le32_to_cpu(f->type) == type)
45 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
46 struct bch_sb_field *f,
49 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
50 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
52 BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
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);
103 free_pages((unsigned long) sb->sb, sb->page_order);
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 unsigned order = get_order(new_bytes);
111 struct bch_sb *new_sb;
114 if (sb->sb && sb->page_order >= order)
117 if (sb->have_layout) {
118 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
120 if (new_bytes > max_bytes) {
121 char buf[BDEVNAME_SIZE];
123 pr_err("%s: superblock too big: want %zu but have %llu",
124 bdevname(sb->bdev, buf), new_bytes, max_bytes);
129 if (sb->page_order >= order && sb->sb)
132 if (dynamic_fault("bcachefs:add:super_realloc"))
136 bio = bio_kmalloc(GFP_KERNEL, 1 << order);
145 new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
150 memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
152 free_pages((unsigned long) sb->sb, sb->page_order);
155 sb->page_order = order;
160 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
161 enum bch_sb_field_type type,
164 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
165 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
166 ssize_t d = -old_u64s + u64s;
168 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
172 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
176 lockdep_assert_held(&c->sb_lock);
178 /* XXX: we're not checking that offline device have enough space */
180 for_each_online_member(ca, c, i) {
181 struct bch_sb_handle *sb = &ca->disk_sb;
183 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
184 percpu_ref_put(&ca->ref);
190 f = bch2_sb_field_get(sb->sb, type);
191 f = __bch2_sb_field_resize(sb, f, u64s);
193 f->type = cpu_to_le32(type);
197 /* Superblock validate: */
199 static inline void __bch2_sb_layout_size_assert(void)
201 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
204 static const char *validate_sb_layout(struct bch_sb_layout *layout)
206 u64 offset, prev_offset, max_sectors;
209 if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
210 return "Not a bcachefs superblock layout";
212 if (layout->layout_type != 0)
213 return "Invalid superblock layout type";
215 if (!layout->nr_superblocks)
216 return "Invalid superblock layout: no superblocks";
218 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
219 return "Invalid superblock layout: too many superblocks";
221 max_sectors = 1 << layout->sb_max_size_bits;
223 prev_offset = le64_to_cpu(layout->sb_offset[0]);
225 for (i = 1; i < layout->nr_superblocks; i++) {
226 offset = le64_to_cpu(layout->sb_offset[i]);
228 if (offset < prev_offset + max_sectors)
229 return "Invalid superblock layout: superblocks overlap";
230 prev_offset = offset;
236 const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
238 struct bch_sb *sb = disk_sb->sb;
239 struct bch_sb_field *f;
240 struct bch_sb_field_members *mi;
242 u32 version, version_min;
245 version = le16_to_cpu(sb->version);
246 version_min = version >= bcachefs_metadata_version_new_versioning
247 ? le16_to_cpu(sb->version_min)
250 if (version >= bcachefs_metadata_version_max ||
251 version_min < bcachefs_metadata_version_min)
252 return "Unsupported superblock version";
254 if (version_min > version)
255 return "Bad minimum version";
257 if (sb->features[1] ||
258 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
259 return "Filesystem has incompatible features";
261 block_size = le16_to_cpu(sb->block_size);
263 if (!is_power_of_2(block_size) ||
264 block_size > PAGE_SECTORS)
265 return "Bad block size";
267 if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
268 return "Bad user UUID";
270 if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
271 return "Bad internal UUID";
273 if (!sb->nr_devices ||
274 sb->nr_devices <= sb->dev_idx ||
275 sb->nr_devices > BCH_SB_MEMBERS_MAX)
276 return "Bad number of member devices";
278 if (!BCH_SB_META_REPLICAS_WANT(sb) ||
279 BCH_SB_META_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
280 return "Invalid number of metadata replicas";
282 if (!BCH_SB_META_REPLICAS_REQ(sb) ||
283 BCH_SB_META_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
284 return "Invalid number of metadata replicas";
286 if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
287 BCH_SB_DATA_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
288 return "Invalid number of data replicas";
290 if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
291 BCH_SB_DATA_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
292 return "Invalid number of data replicas";
294 if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
295 return "Invalid metadata checksum type";
297 if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
298 return "Invalid metadata checksum type";
300 if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
301 return "Invalid compression type";
303 if (!BCH_SB_BTREE_NODE_SIZE(sb))
304 return "Btree node size not set";
306 if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
307 return "Btree node size not a power of two";
309 if (BCH_SB_GC_RESERVE(sb) < 5)
310 return "gc reserve percentage too small";
312 if (!sb->time_precision ||
313 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
314 return "invalid time precision";
316 /* validate layout */
317 err = validate_sb_layout(&sb->layout);
321 vstruct_for_each(sb, f) {
323 return "Invalid superblock: invalid optional field";
325 if (vstruct_next(f) > vstruct_last(sb))
326 return "Invalid superblock: invalid optional field";
329 /* members must be validated first: */
330 mi = bch2_sb_get_members(sb);
332 return "Invalid superblock: member info area missing";
334 err = bch2_sb_field_validate(sb, &mi->field);
338 vstruct_for_each(sb, f) {
339 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
342 err = bch2_sb_field_validate(sb, f);
352 static void bch2_sb_update(struct bch_fs *c)
354 struct bch_sb *src = c->disk_sb.sb;
355 struct bch_sb_field_members *mi = bch2_sb_get_members(src);
359 lockdep_assert_held(&c->sb_lock);
361 c->sb.uuid = src->uuid;
362 c->sb.user_uuid = src->user_uuid;
363 c->sb.version = le16_to_cpu(src->version);
364 c->sb.nr_devices = src->nr_devices;
365 c->sb.clean = BCH_SB_CLEAN(src);
366 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
367 c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
368 c->sb.time_base_lo = le64_to_cpu(src->time_base_lo);
369 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
370 c->sb.time_precision = le32_to_cpu(src->time_precision);
371 c->sb.features = le64_to_cpu(src->features[0]);
372 c->sb.compat = le64_to_cpu(src->compat[0]);
374 for_each_member_device(ca, c, i)
375 ca->mi = bch2_mi_to_cpu(mi->members + i);
378 /* doesn't copy member info */
379 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
381 struct bch_sb_field *src_f, *dst_f;
382 struct bch_sb *dst = dst_handle->sb;
385 dst->version = src->version;
386 dst->version_min = src->version_min;
388 dst->uuid = src->uuid;
389 dst->user_uuid = src->user_uuid;
390 memcpy(dst->label, src->label, sizeof(dst->label));
392 dst->block_size = src->block_size;
393 dst->nr_devices = src->nr_devices;
395 dst->time_base_lo = src->time_base_lo;
396 dst->time_base_hi = src->time_base_hi;
397 dst->time_precision = src->time_precision;
399 memcpy(dst->flags, src->flags, sizeof(dst->flags));
400 memcpy(dst->features, src->features, sizeof(dst->features));
401 memcpy(dst->compat, src->compat, sizeof(dst->compat));
403 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
404 if (i == BCH_SB_FIELD_journal)
407 src_f = bch2_sb_field_get(src, i);
408 dst_f = bch2_sb_field_get(dst, i);
409 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
410 src_f ? le32_to_cpu(src_f->u64s) : 0);
413 memcpy(dst_f, src_f, vstruct_bytes(src_f));
417 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
419 struct bch_sb_field_journal *journal_buckets =
420 bch2_sb_get_journal(src);
421 unsigned journal_u64s = journal_buckets
422 ? le32_to_cpu(journal_buckets->field.u64s)
426 lockdep_assert_held(&c->sb_lock);
428 ret = bch2_sb_realloc(&c->disk_sb,
429 le32_to_cpu(src->u64s) - journal_u64s);
433 __copy_super(&c->disk_sb, src);
435 ret = bch2_sb_replicas_to_cpu_replicas(c);
439 ret = bch2_sb_disk_groups_to_cpu(c);
447 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
449 struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
450 struct bch_sb_field_journal *journal_buckets =
451 bch2_sb_get_journal(dst);
452 unsigned journal_u64s = journal_buckets
453 ? le32_to_cpu(journal_buckets->field.u64s)
455 unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
458 ret = bch2_sb_realloc(&ca->disk_sb, u64s);
462 __copy_super(&ca->disk_sb, src);
466 /* read superblock: */
468 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
470 struct bch_csum csum;
474 bio_set_dev(sb->bio, sb->bdev);
475 sb->bio->bi_iter.bi_sector = offset;
476 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
477 bch2_bio_map(sb->bio, sb->sb, PAGE_SIZE << sb->page_order);
479 if (submit_bio_wait(sb->bio))
482 if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
483 return "Not a bcachefs superblock";
485 if (le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_min ||
486 le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
487 return "Unsupported superblock version";
489 bytes = vstruct_bytes(sb->sb);
491 if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
492 return "Bad superblock: too big";
494 if (get_order(bytes) > sb->page_order) {
495 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
496 return "cannot allocate memory";
500 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
501 return "unknown csum type";
503 /* XXX: verify MACs */
504 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
505 null_nonce(), sb->sb);
507 if (bch2_crc_cmp(csum, sb->sb->csum))
508 return "bad checksum reading superblock";
510 sb->seq = le64_to_cpu(sb->sb->seq);
515 int bch2_read_super(const char *path, struct bch_opts *opts,
516 struct bch_sb_handle *sb)
518 u64 offset = opt_get(*opts, sb);
519 struct bch_sb_layout layout;
524 pr_verbose_init(*opts, "");
526 memset(sb, 0, sizeof(*sb));
527 sb->mode = FMODE_READ;
530 if (!opt_get(*opts, noexcl))
531 sb->mode |= FMODE_EXCL;
533 if (!opt_get(*opts, nochanges))
534 sb->mode |= FMODE_WRITE;
536 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
537 if (IS_ERR(sb->bdev) &&
538 PTR_ERR(sb->bdev) == -EACCES &&
539 opt_get(*opts, read_only)) {
540 sb->mode &= ~FMODE_WRITE;
542 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
543 if (!IS_ERR(sb->bdev))
544 opt_set(*opts, nochanges, true);
547 if (IS_ERR(sb->bdev)) {
548 ret = PTR_ERR(sb->bdev);
552 err = "cannot allocate memory";
553 ret = bch2_sb_realloc(sb, 0);
558 err = "dynamic fault";
559 if (bch2_fs_init_fault("read_super"))
563 err = read_one_super(sb, offset);
567 if (opt_defined(*opts, sb))
570 pr_err("error reading default superblock: %s", err);
573 * Error reading primary superblock - read location of backup
577 bio_set_dev(sb->bio, sb->bdev);
578 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
579 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
581 * use sb buffer to read layout, since sb buffer is page aligned but
584 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
587 if (submit_bio_wait(sb->bio))
590 memcpy(&layout, sb->sb, sizeof(layout));
591 err = validate_sb_layout(&layout);
595 for (i = layout.sb_offset;
596 i < layout.sb_offset + layout.nr_superblocks; i++) {
597 offset = le64_to_cpu(*i);
599 if (offset == opt_get(*opts, sb))
602 err = read_one_super(sb, offset);
611 err = "Superblock block size smaller than device block size";
613 if (le16_to_cpu(sb->sb->block_size) << 9 <
614 bdev_logical_block_size(sb->bdev))
618 sb->have_layout = true;
620 pr_verbose_init(*opts, "ret %i", ret);
624 pr_err("error reading superblock: %s", err);
628 /* write superblock: */
630 static void write_super_endio(struct bio *bio)
632 struct bch_dev *ca = bio->bi_private;
634 /* XXX: return errors directly */
636 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
637 bch2_blk_status_to_str(bio->bi_status)))
638 ca->sb_write_error = 1;
640 closure_put(&ca->fs->sb_write);
641 percpu_ref_put(&ca->io_ref);
644 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
646 struct bch_sb *sb = ca->disk_sb.sb;
647 struct bio *bio = ca->disk_sb.bio;
650 bio_set_dev(bio, ca->disk_sb.bdev);
651 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
652 bio->bi_end_io = write_super_endio;
653 bio->bi_private = ca;
654 bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
655 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
657 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
660 percpu_ref_get(&ca->io_ref);
661 closure_bio_submit(bio, &c->sb_write);
664 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
666 struct bch_sb *sb = ca->disk_sb.sb;
667 struct bio *bio = ca->disk_sb.bio;
669 sb->offset = sb->layout.sb_offset[idx];
671 SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
672 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
676 bio_set_dev(bio, ca->disk_sb.bdev);
677 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
678 bio->bi_end_io = write_super_endio;
679 bio->bi_private = ca;
680 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
681 bch2_bio_map(bio, sb,
682 roundup((size_t) vstruct_bytes(sb),
683 bdev_logical_block_size(ca->disk_sb.bdev)));
685 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
688 percpu_ref_get(&ca->io_ref);
689 closure_bio_submit(bio, &c->sb_write);
692 int bch2_write_super(struct bch_fs *c)
694 struct closure *cl = &c->sb_write;
696 unsigned i, sb = 0, nr_wrote;
698 struct bch_devs_mask sb_written;
699 bool wrote, can_mount_without_written, can_mount_with_written;
702 lockdep_assert_held(&c->sb_lock);
704 closure_init_stack(cl);
705 memset(&sb_written, 0, sizeof(sb_written));
707 le64_add_cpu(&c->disk_sb.sb->seq, 1);
709 if (test_bit(BCH_FS_ERROR, &c->flags))
710 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
712 for_each_online_member(ca, c, i)
713 bch2_sb_from_fs(c, ca);
715 for_each_online_member(ca, c, i) {
716 err = bch2_sb_validate(&ca->disk_sb);
718 bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
724 if (c->opts.nochanges)
727 for_each_online_member(ca, c, i) {
728 __set_bit(ca->dev_idx, sb_written.d);
729 ca->sb_write_error = 0;
732 for_each_online_member(ca, c, i)
733 read_back_super(c, ca);
736 for_each_online_member(ca, c, i) {
737 if (!ca->sb_write_error &&
739 le64_to_cpu(ca->sb_read_scratch->seq)) {
740 bch2_fs_fatal_error(c,
741 "Superblock modified by another process");
742 percpu_ref_put(&ca->io_ref);
750 for_each_online_member(ca, c, i)
751 if (!ca->sb_write_error &&
752 sb < ca->disk_sb.sb->layout.nr_superblocks) {
753 write_one_super(c, ca, sb);
760 for_each_online_member(ca, c, i) {
761 if (ca->sb_write_error)
762 __clear_bit(ca->dev_idx, sb_written.d);
764 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
767 nr_wrote = dev_mask_nr(&sb_written);
769 can_mount_with_written =
770 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
771 BCH_FORCE_IF_DEGRADED);
773 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
774 sb_written.d[i] = ~sb_written.d[i];
776 can_mount_without_written =
777 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
778 BCH_FORCE_IF_DEGRADED);
781 * If we would be able to mount _without_ the devices we successfully
782 * wrote superblocks to, we weren't able to write to enough devices:
784 * Exception: if we can mount without the successes because we haven't
785 * written anything (new filesystem), we continue if we'd be able to
786 * mount with the devices we did successfully write to:
788 if (bch2_fs_fatal_err_on(!nr_wrote ||
789 (can_mount_without_written &&
790 !can_mount_with_written), c,
791 "Unable to write superblock to sufficient devices"))
794 /* Make new options visible after they're persistent: */
799 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
801 mutex_lock(&c->sb_lock);
802 if (!(c->sb.features & (1ULL << feat))) {
803 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
807 mutex_unlock(&c->sb_lock);
810 /* BCH_SB_FIELD_journal: */
812 static int u64_cmp(const void *_l, const void *_r)
814 u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
816 return l < r ? -1 : l > r ? 1 : 0;
819 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
820 struct bch_sb_field *f)
822 struct bch_sb_field_journal *journal = field_to_type(f, journal);
823 struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
829 journal = bch2_sb_get_journal(sb);
833 nr = bch2_nr_journal_buckets(journal);
837 b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
839 return "cannot allocate memory";
841 for (i = 0; i < nr; i++)
842 b[i] = le64_to_cpu(journal->buckets[i]);
844 sort(b, nr, sizeof(u64), u64_cmp, NULL);
846 err = "journal bucket at sector 0";
850 err = "journal bucket before first bucket";
851 if (m && b[0] < le16_to_cpu(m->first_bucket))
854 err = "journal bucket past end of device";
855 if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
858 err = "duplicate journal buckets";
859 for (i = 0; i + 1 < nr; i++)
860 if (b[i] == b[i + 1])
869 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
870 .validate = bch2_sb_validate_journal,
873 /* BCH_SB_FIELD_members: */
875 static const char *bch2_sb_validate_members(struct bch_sb *sb,
876 struct bch_sb_field *f)
878 struct bch_sb_field_members *mi = field_to_type(f, members);
879 struct bch_member *m;
881 if ((void *) (mi->members + sb->nr_devices) >
882 vstruct_end(&mi->field))
883 return "Invalid superblock: bad member info";
885 for (m = mi->members;
886 m < mi->members + sb->nr_devices;
888 if (!bch2_member_exists(m))
891 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
892 return "Too many buckets";
894 if (le64_to_cpu(m->nbuckets) -
895 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
896 return "Not enough buckets";
898 if (le16_to_cpu(m->bucket_size) <
899 le16_to_cpu(sb->block_size))
900 return "bucket size smaller than block size";
902 if (le16_to_cpu(m->bucket_size) <
903 BCH_SB_BTREE_NODE_SIZE(sb))
904 return "bucket size smaller than btree node size";
910 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
911 .validate = bch2_sb_validate_members,
914 /* BCH_SB_FIELD_crypt: */
916 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
917 struct bch_sb_field *f)
919 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
921 if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
922 return "invalid field crypt: wrong size";
924 if (BCH_CRYPT_KDF_TYPE(crypt))
925 return "invalid field crypt: bad kdf type";
930 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
931 .validate = bch2_sb_validate_crypt,
934 /* BCH_SB_FIELD_clean: */
936 void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
938 struct jset_entry *entry;
940 for (entry = clean->start;
941 entry < (struct jset_entry *) vstruct_end(&clean->field);
942 entry = vstruct_next(entry))
943 bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
946 int bch2_fs_mark_dirty(struct bch_fs *c)
951 * Unconditionally write superblock, to verify it hasn't changed before
955 mutex_lock(&c->sb_lock);
956 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
957 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
958 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_extents_above_btree_updates;
959 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_btree_updates_journalled;
960 ret = bch2_write_super(c);
961 mutex_unlock(&c->sb_lock);
967 entry_init_u64s(struct jset_entry *entry, unsigned u64s)
969 memset(entry, 0, u64s * sizeof(u64));
972 * The u64s field counts from the start of data, ignoring the shared
975 entry->u64s = u64s - 1;
979 entry_init_size(struct jset_entry *entry, size_t size)
981 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
982 entry_init_u64s(entry, u64s);
986 bch2_journal_super_entries_add_common(struct bch_fs *c,
987 struct jset_entry *entry,
992 percpu_down_write(&c->mark_lock);
995 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
996 bch2_fs_usage_acc_to_base(c, i);
998 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1002 struct jset_entry_usage *u =
1003 container_of(entry, struct jset_entry_usage, entry);
1005 entry_init_size(entry, sizeof(*u));
1006 u->entry.type = BCH_JSET_ENTRY_usage;
1007 u->entry.btree_id = FS_USAGE_INODES;
1008 u->v = cpu_to_le64(c->usage_base->nr_inodes);
1010 entry = vstruct_next(entry);
1014 struct jset_entry_usage *u =
1015 container_of(entry, struct jset_entry_usage, entry);
1017 entry_init_size(entry, sizeof(*u));
1018 u->entry.type = BCH_JSET_ENTRY_usage;
1019 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1020 u->v = cpu_to_le64(atomic64_read(&c->key_version));
1022 entry = vstruct_next(entry);
1025 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1026 struct jset_entry_usage *u =
1027 container_of(entry, struct jset_entry_usage, entry);
1029 entry_init_size(entry, sizeof(*u));
1030 u->entry.type = BCH_JSET_ENTRY_usage;
1031 u->entry.btree_id = FS_USAGE_RESERVED;
1033 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1035 entry = vstruct_next(entry);
1038 for (i = 0; i < c->replicas.nr; i++) {
1039 struct bch_replicas_entry *e =
1040 cpu_replicas_entry(&c->replicas, i);
1041 struct jset_entry_data_usage *u =
1042 container_of(entry, struct jset_entry_data_usage, entry);
1044 entry_init_size(entry, sizeof(*u) + e->nr_devs);
1045 u->entry.type = BCH_JSET_ENTRY_data_usage;
1046 u->v = cpu_to_le64(c->usage_base->replicas[i]);
1047 memcpy(&u->r, e, replicas_entry_bytes(e));
1049 entry = vstruct_next(entry);
1052 percpu_up_write(&c->mark_lock);
1057 void bch2_fs_mark_clean(struct bch_fs *c)
1059 struct bch_sb_field_clean *sb_clean;
1060 struct jset_entry *entry;
1063 mutex_lock(&c->sb_lock);
1064 if (BCH_SB_CLEAN(c->disk_sb.sb))
1067 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1069 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1070 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA;
1071 c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
1072 c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
1074 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1076 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1078 bch_err(c, "error resizing superblock while setting filesystem clean");
1082 sb_clean->flags = 0;
1083 sb_clean->read_clock = cpu_to_le16(c->bucket_clock[READ].hand);
1084 sb_clean->write_clock = cpu_to_le16(c->bucket_clock[WRITE].hand);
1085 sb_clean->journal_seq = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1087 /* Trying to catch outstanding bug: */
1088 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1090 entry = sb_clean->start;
1091 entry = bch2_journal_super_entries_add_common(c, entry, 0);
1092 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1093 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1096 vstruct_end(&sb_clean->field) - (void *) entry);
1098 if (le16_to_cpu(c->disk_sb.sb->version) <
1099 bcachefs_metadata_version_bkey_renumber)
1100 bch2_sb_clean_renumber(sb_clean, WRITE);
1102 bch2_write_super(c);
1104 mutex_unlock(&c->sb_lock);
1107 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1108 struct bch_sb_field *f)
1110 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1112 if (vstruct_bytes(&clean->field) < sizeof(*clean))
1113 return "invalid field crypt: wrong size";
1118 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1119 .validate = bch2_sb_validate_clean,
1122 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1124 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1129 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1130 struct bch_sb_field *f)
1132 unsigned type = le32_to_cpu(f->type);
1134 return type < BCH_SB_FIELD_NR
1135 ? bch2_sb_field_ops[type]->validate(sb, f)
1139 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1140 struct bch_sb_field *f)
1142 unsigned type = le32_to_cpu(f->type);
1143 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1144 ? bch2_sb_field_ops[type] : NULL;
1147 pr_buf(out, "%s", bch2_sb_fields[type]);
1149 pr_buf(out, "(unknown field %u)", type);
1151 pr_buf(out, " (size %llu):", vstruct_bytes(f));
1153 if (ops && ops->to_text)
1154 bch2_sb_field_ops[type]->to_text(out, sb, f);