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 const char *bch2_sb_field_validate(struct bch_sb *,
31 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 const char *validate_sb_layout(struct bch_sb_layout *layout)
207 u64 offset, prev_offset, max_sectors;
210 if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
211 return "Not a bcachefs superblock layout";
213 if (layout->layout_type != 0)
214 return "Invalid superblock layout type";
216 if (!layout->nr_superblocks)
217 return "Invalid superblock layout: no superblocks";
219 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
220 return "Invalid superblock layout: too many superblocks";
222 max_sectors = 1 << layout->sb_max_size_bits;
224 prev_offset = le64_to_cpu(layout->sb_offset[0]);
226 for (i = 1; i < layout->nr_superblocks; i++) {
227 offset = le64_to_cpu(layout->sb_offset[i]);
229 if (offset < prev_offset + max_sectors)
230 return "Invalid superblock layout: superblocks overlap";
231 prev_offset = offset;
237 const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
239 struct bch_sb *sb = disk_sb->sb;
240 struct bch_sb_field *f;
241 struct bch_sb_field_members *mi;
243 u32 version, version_min;
246 version = le16_to_cpu(sb->version);
247 version_min = version >= bcachefs_metadata_version_new_versioning
248 ? le16_to_cpu(sb->version_min)
251 if (version >= bcachefs_metadata_version_max ||
252 version_min < bcachefs_metadata_version_min)
253 return "Unsupported superblock version";
255 if (version_min > version)
256 return "Bad minimum version";
258 if (sb->features[1] ||
259 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
260 return "Filesystem has incompatible features";
262 block_size = le16_to_cpu(sb->block_size);
264 if (!is_power_of_2(block_size) ||
265 block_size > PAGE_SECTORS)
266 return "Bad block size";
268 if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
269 return "Bad user UUID";
271 if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
272 return "Bad internal UUID";
274 if (!sb->nr_devices ||
275 sb->nr_devices <= sb->dev_idx ||
276 sb->nr_devices > BCH_SB_MEMBERS_MAX)
277 return "Bad number of member devices";
279 if (!BCH_SB_META_REPLICAS_WANT(sb) ||
280 BCH_SB_META_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
281 return "Invalid number of metadata replicas";
283 if (!BCH_SB_META_REPLICAS_REQ(sb) ||
284 BCH_SB_META_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
285 return "Invalid number of metadata replicas";
287 if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
288 BCH_SB_DATA_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
289 return "Invalid number of data replicas";
291 if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
292 BCH_SB_DATA_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
293 return "Invalid number of data replicas";
295 if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
296 return "Invalid metadata checksum type";
298 if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
299 return "Invalid metadata checksum type";
301 if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
302 return "Invalid compression type";
304 if (!BCH_SB_BTREE_NODE_SIZE(sb))
305 return "Btree node size not set";
307 if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
308 return "Btree node size not a power of two";
310 if (BCH_SB_GC_RESERVE(sb) < 5)
311 return "gc reserve percentage too small";
313 if (!sb->time_precision ||
314 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
315 return "invalid time precision";
317 /* validate layout */
318 err = validate_sb_layout(&sb->layout);
322 vstruct_for_each(sb, f) {
324 return "Invalid superblock: invalid optional field";
326 if (vstruct_next(f) > vstruct_last(sb))
327 return "Invalid superblock: invalid optional field";
330 /* members must be validated first: */
331 mi = bch2_sb_get_members(sb);
333 return "Invalid superblock: member info area missing";
335 err = bch2_sb_field_validate(sb, &mi->field);
339 vstruct_for_each(sb, f) {
340 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
343 err = bch2_sb_field_validate(sb, f);
353 static void bch2_sb_update(struct bch_fs *c)
355 struct bch_sb *src = c->disk_sb.sb;
356 struct bch_sb_field_members *mi = bch2_sb_get_members(src);
360 lockdep_assert_held(&c->sb_lock);
362 c->sb.uuid = src->uuid;
363 c->sb.user_uuid = src->user_uuid;
364 c->sb.version = le16_to_cpu(src->version);
365 c->sb.version_min = le16_to_cpu(src->version_min);
366 c->sb.nr_devices = src->nr_devices;
367 c->sb.clean = BCH_SB_CLEAN(src);
368 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
369 c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
371 c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
372 c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
374 /* XXX this is wrong, we need a 96 or 128 bit integer type */
375 c->sb.time_base_lo = div_u64(le64_to_cpu(src->time_base_lo),
376 c->sb.nsec_per_time_unit);
377 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
379 c->sb.features = le64_to_cpu(src->features[0]);
380 c->sb.compat = le64_to_cpu(src->compat[0]);
382 for_each_member_device(ca, c, i)
383 ca->mi = bch2_mi_to_cpu(mi->members + i);
386 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
388 struct bch_sb_field *src_f, *dst_f;
389 struct bch_sb *dst = dst_handle->sb;
392 dst->version = src->version;
393 dst->version_min = src->version_min;
395 dst->uuid = src->uuid;
396 dst->user_uuid = src->user_uuid;
397 memcpy(dst->label, src->label, sizeof(dst->label));
399 dst->block_size = src->block_size;
400 dst->nr_devices = src->nr_devices;
402 dst->time_base_lo = src->time_base_lo;
403 dst->time_base_hi = src->time_base_hi;
404 dst->time_precision = src->time_precision;
406 memcpy(dst->flags, src->flags, sizeof(dst->flags));
407 memcpy(dst->features, src->features, sizeof(dst->features));
408 memcpy(dst->compat, src->compat, sizeof(dst->compat));
410 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
411 if (i == BCH_SB_FIELD_journal)
414 src_f = bch2_sb_field_get(src, i);
415 dst_f = bch2_sb_field_get(dst, i);
416 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
417 src_f ? le32_to_cpu(src_f->u64s) : 0);
420 memcpy(dst_f, src_f, vstruct_bytes(src_f));
424 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
426 struct bch_sb_field_journal *journal_buckets =
427 bch2_sb_get_journal(src);
428 unsigned journal_u64s = journal_buckets
429 ? le32_to_cpu(journal_buckets->field.u64s)
433 lockdep_assert_held(&c->sb_lock);
435 ret = bch2_sb_realloc(&c->disk_sb,
436 le32_to_cpu(src->u64s) - journal_u64s);
440 __copy_super(&c->disk_sb, src);
442 if (BCH_SB_HAS_ERRORS(c->disk_sb.sb))
443 set_bit(BCH_FS_ERROR, &c->flags);
444 if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb))
445 set_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags);
447 ret = bch2_sb_replicas_to_cpu_replicas(c);
451 ret = bch2_sb_disk_groups_to_cpu(c);
459 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
461 struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
462 struct bch_sb_field_journal *journal_buckets =
463 bch2_sb_get_journal(dst);
464 unsigned journal_u64s = journal_buckets
465 ? le32_to_cpu(journal_buckets->field.u64s)
467 unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
470 ret = bch2_sb_realloc(&ca->disk_sb, u64s);
474 __copy_super(&ca->disk_sb, src);
478 /* read superblock: */
480 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
482 struct bch_csum csum;
486 bio_set_dev(sb->bio, sb->bdev);
487 sb->bio->bi_iter.bi_sector = offset;
488 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
489 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
491 if (submit_bio_wait(sb->bio))
494 if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
495 return "Not a bcachefs superblock";
497 if (le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_min ||
498 le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
499 return "Unsupported superblock version";
501 bytes = vstruct_bytes(sb->sb);
503 if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
504 return "Bad superblock: too big";
506 if (bytes > sb->buffer_size) {
507 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
508 return "cannot allocate memory";
512 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
513 return "unknown csum type";
515 /* XXX: verify MACs */
516 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
517 null_nonce(), sb->sb);
519 if (bch2_crc_cmp(csum, sb->sb->csum))
520 return "bad checksum reading superblock";
522 sb->seq = le64_to_cpu(sb->sb->seq);
527 int bch2_read_super(const char *path, struct bch_opts *opts,
528 struct bch_sb_handle *sb)
530 u64 offset = opt_get(*opts, sb);
531 struct bch_sb_layout layout;
536 pr_verbose_init(*opts, "");
538 memset(sb, 0, sizeof(*sb));
539 sb->mode = FMODE_READ;
542 if (!opt_get(*opts, noexcl))
543 sb->mode |= FMODE_EXCL;
545 if (!opt_get(*opts, nochanges))
546 sb->mode |= FMODE_WRITE;
548 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
549 if (IS_ERR(sb->bdev) &&
550 PTR_ERR(sb->bdev) == -EACCES &&
551 opt_get(*opts, read_only)) {
552 sb->mode &= ~FMODE_WRITE;
554 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
555 if (!IS_ERR(sb->bdev))
556 opt_set(*opts, nochanges, true);
559 if (IS_ERR(sb->bdev)) {
560 ret = PTR_ERR(sb->bdev);
564 err = "cannot allocate memory";
565 ret = bch2_sb_realloc(sb, 0);
570 err = "dynamic fault";
571 if (bch2_fs_init_fault("read_super"))
575 err = read_one_super(sb, offset);
579 if (opt_defined(*opts, sb))
582 pr_err("error reading default superblock: %s", err);
585 * Error reading primary superblock - read location of backup
589 bio_set_dev(sb->bio, sb->bdev);
590 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
591 bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
593 * use sb buffer to read layout, since sb buffer is page aligned but
596 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
599 if (submit_bio_wait(sb->bio))
602 memcpy(&layout, sb->sb, sizeof(layout));
603 err = validate_sb_layout(&layout);
607 for (i = layout.sb_offset;
608 i < layout.sb_offset + layout.nr_superblocks; i++) {
609 offset = le64_to_cpu(*i);
611 if (offset == opt_get(*opts, sb))
614 err = read_one_super(sb, offset);
623 err = "Superblock block size smaller than device block size";
625 if (le16_to_cpu(sb->sb->block_size) << 9 <
626 bdev_logical_block_size(sb->bdev))
630 sb->have_layout = true;
632 pr_verbose_init(*opts, "ret %i", ret);
636 pr_err("error reading superblock: %s", err);
640 /* write superblock: */
642 static void write_super_endio(struct bio *bio)
644 struct bch_dev *ca = bio->bi_private;
646 /* XXX: return errors directly */
648 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
649 bch2_blk_status_to_str(bio->bi_status)))
650 ca->sb_write_error = 1;
652 closure_put(&ca->fs->sb_write);
653 percpu_ref_put(&ca->io_ref);
656 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
658 struct bch_sb *sb = ca->disk_sb.sb;
659 struct bio *bio = ca->disk_sb.bio;
662 bio_set_dev(bio, ca->disk_sb.bdev);
663 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
664 bio->bi_end_io = write_super_endio;
665 bio->bi_private = ca;
666 bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
667 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
669 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
672 percpu_ref_get(&ca->io_ref);
673 closure_bio_submit(bio, &c->sb_write);
676 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
678 struct bch_sb *sb = ca->disk_sb.sb;
679 struct bio *bio = ca->disk_sb.bio;
681 sb->offset = sb->layout.sb_offset[idx];
683 SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
684 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
688 bio_set_dev(bio, ca->disk_sb.bdev);
689 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
690 bio->bi_end_io = write_super_endio;
691 bio->bi_private = ca;
692 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
693 bch2_bio_map(bio, sb,
694 roundup((size_t) vstruct_bytes(sb),
695 bdev_logical_block_size(ca->disk_sb.bdev)));
697 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
700 percpu_ref_get(&ca->io_ref);
701 closure_bio_submit(bio, &c->sb_write);
704 int bch2_write_super(struct bch_fs *c)
706 struct closure *cl = &c->sb_write;
708 unsigned i, sb = 0, nr_wrote;
710 struct bch_devs_mask sb_written;
711 bool wrote, can_mount_without_written, can_mount_with_written;
712 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
715 if (c->opts.very_degraded)
716 degraded_flags |= BCH_FORCE_IF_LOST;
718 lockdep_assert_held(&c->sb_lock);
720 closure_init_stack(cl);
721 memset(&sb_written, 0, sizeof(sb_written));
723 le64_add_cpu(&c->disk_sb.sb->seq, 1);
725 if (test_bit(BCH_FS_ERROR, &c->flags))
726 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
727 if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
728 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
730 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
732 for_each_online_member(ca, c, i)
733 bch2_sb_from_fs(c, ca);
735 for_each_online_member(ca, c, i) {
736 err = bch2_sb_validate(&ca->disk_sb);
738 bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
744 if (c->opts.nochanges)
747 for_each_online_member(ca, c, i) {
748 __set_bit(ca->dev_idx, sb_written.d);
749 ca->sb_write_error = 0;
752 for_each_online_member(ca, c, i)
753 read_back_super(c, ca);
756 for_each_online_member(ca, c, i) {
757 if (!ca->sb_write_error &&
759 le64_to_cpu(ca->sb_read_scratch->seq)) {
760 bch2_fs_fatal_error(c,
761 "Superblock modified by another process");
762 percpu_ref_put(&ca->io_ref);
770 for_each_online_member(ca, c, i)
771 if (!ca->sb_write_error &&
772 sb < ca->disk_sb.sb->layout.nr_superblocks) {
773 write_one_super(c, ca, sb);
780 for_each_online_member(ca, c, i) {
781 if (ca->sb_write_error)
782 __clear_bit(ca->dev_idx, sb_written.d);
784 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
787 nr_wrote = dev_mask_nr(&sb_written);
789 can_mount_with_written =
790 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
792 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
793 sb_written.d[i] = ~sb_written.d[i];
795 can_mount_without_written =
796 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
799 * If we would be able to mount _without_ the devices we successfully
800 * wrote superblocks to, we weren't able to write to enough devices:
802 * Exception: if we can mount without the successes because we haven't
803 * written anything (new filesystem), we continue if we'd be able to
804 * mount with the devices we did successfully write to:
806 if (bch2_fs_fatal_err_on(!nr_wrote ||
807 !can_mount_with_written ||
808 (can_mount_without_written &&
809 !can_mount_with_written), c,
810 "Unable to write superblock to sufficient devices"))
813 /* Make new options visible after they're persistent: */
818 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
820 mutex_lock(&c->sb_lock);
821 if (!(c->sb.features & (1ULL << feat))) {
822 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
826 mutex_unlock(&c->sb_lock);
829 /* BCH_SB_FIELD_journal: */
831 static int u64_cmp(const void *_l, const void *_r)
833 u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
835 return l < r ? -1 : l > r ? 1 : 0;
838 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
839 struct bch_sb_field *f)
841 struct bch_sb_field_journal *journal = field_to_type(f, journal);
842 struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
848 journal = bch2_sb_get_journal(sb);
852 nr = bch2_nr_journal_buckets(journal);
856 b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
858 return "cannot allocate memory";
860 for (i = 0; i < nr; i++)
861 b[i] = le64_to_cpu(journal->buckets[i]);
863 sort(b, nr, sizeof(u64), u64_cmp, NULL);
865 err = "journal bucket at sector 0";
869 err = "journal bucket before first bucket";
870 if (m && b[0] < le16_to_cpu(m->first_bucket))
873 err = "journal bucket past end of device";
874 if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
877 err = "duplicate journal buckets";
878 for (i = 0; i + 1 < nr; i++)
879 if (b[i] == b[i + 1])
888 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
889 .validate = bch2_sb_validate_journal,
892 /* BCH_SB_FIELD_members: */
894 static const char *bch2_sb_validate_members(struct bch_sb *sb,
895 struct bch_sb_field *f)
897 struct bch_sb_field_members *mi = field_to_type(f, members);
898 struct bch_member *m;
900 if ((void *) (mi->members + sb->nr_devices) >
901 vstruct_end(&mi->field))
902 return "Invalid superblock: bad member info";
904 for (m = mi->members;
905 m < mi->members + sb->nr_devices;
907 if (!bch2_member_exists(m))
910 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
911 return "Too many buckets";
913 if (le64_to_cpu(m->nbuckets) -
914 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
915 return "Not enough buckets";
917 if (le16_to_cpu(m->bucket_size) <
918 le16_to_cpu(sb->block_size))
919 return "bucket size smaller than block size";
921 if (le16_to_cpu(m->bucket_size) <
922 BCH_SB_BTREE_NODE_SIZE(sb))
923 return "bucket size smaller than btree node size";
929 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
930 .validate = bch2_sb_validate_members,
933 /* BCH_SB_FIELD_crypt: */
935 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
936 struct bch_sb_field *f)
938 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
940 if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
941 return "invalid field crypt: wrong size";
943 if (BCH_CRYPT_KDF_TYPE(crypt))
944 return "invalid field crypt: bad kdf type";
949 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
950 .validate = bch2_sb_validate_crypt,
953 /* BCH_SB_FIELD_clean: */
955 int bch2_sb_clean_validate(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
957 struct jset_entry *entry;
960 for (entry = clean->start;
961 entry < (struct jset_entry *) vstruct_end(&clean->field);
962 entry = vstruct_next(entry)) {
963 ret = bch2_journal_entry_validate(c, "superblock", entry,
964 le16_to_cpu(c->disk_sb.sb->version),
965 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
974 int bch2_fs_mark_dirty(struct bch_fs *c)
979 * Unconditionally write superblock, to verify it hasn't changed before
983 mutex_lock(&c->sb_lock);
984 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
985 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALWAYS;
986 ret = bch2_write_super(c);
987 mutex_unlock(&c->sb_lock);
992 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
994 struct jset_entry *entry = *end;
995 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
997 memset(entry, 0, u64s * sizeof(u64));
999 * The u64s field counts from the start of data, ignoring the shared
1002 entry->u64s = u64s - 1;
1004 *end = vstruct_next(*end);
1008 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1009 struct jset_entry **end,
1015 percpu_down_read(&c->mark_lock);
1018 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1019 bch2_fs_usage_acc_to_base(c, i);
1021 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1025 struct jset_entry_usage *u =
1026 container_of(jset_entry_init(end, sizeof(*u)),
1027 struct jset_entry_usage, entry);
1029 u->entry.type = BCH_JSET_ENTRY_usage;
1030 u->entry.btree_id = FS_USAGE_INODES;
1031 u->v = cpu_to_le64(c->usage_base->nr_inodes);
1035 struct jset_entry_usage *u =
1036 container_of(jset_entry_init(end, sizeof(*u)),
1037 struct jset_entry_usage, entry);
1039 u->entry.type = BCH_JSET_ENTRY_usage;
1040 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1041 u->v = cpu_to_le64(atomic64_read(&c->key_version));
1044 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1045 struct jset_entry_usage *u =
1046 container_of(jset_entry_init(end, sizeof(*u)),
1047 struct jset_entry_usage, entry);
1049 u->entry.type = BCH_JSET_ENTRY_usage;
1050 u->entry.btree_id = FS_USAGE_RESERVED;
1052 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1055 for (i = 0; i < c->replicas.nr; i++) {
1056 struct bch_replicas_entry *e =
1057 cpu_replicas_entry(&c->replicas, i);
1058 struct jset_entry_data_usage *u =
1059 container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1060 struct jset_entry_data_usage, entry);
1062 u->entry.type = BCH_JSET_ENTRY_data_usage;
1063 u->v = cpu_to_le64(c->usage_base->replicas[i]);
1064 memcpy(&u->r, e, replicas_entry_bytes(e));
1067 for_each_member_device(ca, c, dev) {
1068 unsigned b = sizeof(struct jset_entry_dev_usage) +
1069 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1070 struct jset_entry_dev_usage *u =
1071 container_of(jset_entry_init(end, b),
1072 struct jset_entry_dev_usage, entry);
1074 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1075 u->dev = cpu_to_le32(dev);
1076 u->buckets_ec = cpu_to_le64(ca->usage_base->buckets_ec);
1077 u->buckets_unavailable = cpu_to_le64(ca->usage_base->buckets_unavailable);
1079 for (i = 0; i < BCH_DATA_NR; i++) {
1080 u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1081 u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1082 u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1086 percpu_up_read(&c->mark_lock);
1088 for (i = 0; i < 2; i++) {
1089 struct jset_entry_clock *clock =
1090 container_of(jset_entry_init(end, sizeof(*clock)),
1091 struct jset_entry_clock, entry);
1093 clock->entry.type = BCH_JSET_ENTRY_clock;
1095 clock->time = atomic64_read(&c->io_clock[i].now);
1099 void bch2_fs_mark_clean(struct bch_fs *c)
1101 struct bch_sb_field_clean *sb_clean;
1102 struct jset_entry *entry;
1106 mutex_lock(&c->sb_lock);
1107 if (BCH_SB_CLEAN(c->disk_sb.sb))
1110 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1112 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_alloc_info;
1113 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_alloc_metadata;
1114 c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
1115 c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
1117 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1119 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1121 bch_err(c, "error resizing superblock while setting filesystem clean");
1125 sb_clean->flags = 0;
1126 sb_clean->journal_seq = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1128 /* Trying to catch outstanding bug: */
1129 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1131 entry = sb_clean->start;
1132 bch2_journal_super_entries_add_common(c, &entry, 0);
1133 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1134 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1137 vstruct_end(&sb_clean->field) - (void *) entry);
1140 * this should be in the write path, and we should be validating every
1141 * superblock section:
1143 ret = bch2_sb_clean_validate(c, sb_clean, WRITE);
1145 bch_err(c, "error writing marking filesystem clean: validate error");
1149 bch2_write_super(c);
1151 mutex_unlock(&c->sb_lock);
1154 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1155 struct bch_sb_field *f)
1157 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1159 if (vstruct_bytes(&clean->field) < sizeof(*clean))
1160 return "invalid field crypt: wrong size";
1165 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1166 .validate = bch2_sb_validate_clean,
1169 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1171 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1176 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1177 struct bch_sb_field *f)
1179 unsigned type = le32_to_cpu(f->type);
1181 return type < BCH_SB_FIELD_NR
1182 ? bch2_sb_field_ops[type]->validate(sb, f)
1186 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1187 struct bch_sb_field *f)
1189 unsigned type = le32_to_cpu(f->type);
1190 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1191 ? bch2_sb_field_ops[type] : NULL;
1194 pr_buf(out, "%s", bch2_sb_fields[type]);
1196 pr_buf(out, "(unknown field %u)", type);
1198 pr_buf(out, " (size %llu):", vstruct_bytes(f));
1200 if (ops && ops->to_text)
1201 bch2_sb_field_ops[type]->to_text(out, sb, f);