1 // SPDX-License-Identifier: GPL-2.0
6 #include "disk_groups.h"
10 #include "journal_sb.h"
11 #include "journal_seq_blacklist.h"
16 #include "sb-errors.h"
17 #include "sb-members.h"
23 #include <linux/backing-dev.h>
24 #include <linux/sort.h>
26 static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
29 struct bch2_metadata_version {
35 static const struct bch2_metadata_version bch2_metadata_versions[] = {
36 #define x(n, v, _recovery_passes) { \
39 .recovery_passes = _recovery_passes, \
41 BCH_METADATA_VERSIONS()
45 void bch2_version_to_text(struct printbuf *out, unsigned v)
47 const char *str = "(unknown version)";
49 for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
50 if (bch2_metadata_versions[i].version == v) {
51 str = bch2_metadata_versions[i].name;
55 prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
58 unsigned bch2_latest_compatible_version(unsigned v)
60 if (!BCH_VERSION_MAJOR(v))
63 for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
64 if (bch2_metadata_versions[i].version > v &&
65 BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
67 v = bch2_metadata_versions[i].version;
72 u64 bch2_upgrade_recovery_passes(struct bch_fs *c,
78 for (const struct bch2_metadata_version *i = bch2_metadata_versions;
79 i < bch2_metadata_versions + ARRAY_SIZE(bch2_metadata_versions);
81 if (i->version > old_version && i->version <= new_version) {
82 if (i->recovery_passes & RECOVERY_PASS_ALL_FSCK)
83 ret |= bch2_fsck_recovery_passes();
84 ret |= i->recovery_passes;
87 return ret &= ~RECOVERY_PASS_ALL_FSCK;
90 const char * const bch2_sb_fields[] = {
91 #define x(name, nr) #name,
97 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
100 struct bch_sb_field *bch2_sb_field_get_id(struct bch_sb *sb,
101 enum bch_sb_field_type type)
103 struct bch_sb_field *f;
105 /* XXX: need locking around superblock to access optional fields */
107 vstruct_for_each(sb, f)
108 if (le32_to_cpu(f->type) == type)
113 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
114 struct bch_sb_field *f,
117 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
118 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
120 BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
125 f = vstruct_last(sb->sb);
126 memset(f, 0, sizeof(u64) * u64s);
127 f->u64s = cpu_to_le32(u64s);
132 src = vstruct_end(f);
135 f->u64s = cpu_to_le32(u64s);
136 dst = vstruct_end(f);
141 memmove(dst, src, vstruct_end(sb->sb) - src);
144 memset(src, 0, dst - src);
147 sb->sb->u64s = cpu_to_le32(sb_u64s);
149 return u64s ? f : NULL;
152 void bch2_sb_field_delete(struct bch_sb_handle *sb,
153 enum bch_sb_field_type type)
155 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
158 __bch2_sb_field_resize(sb, f, 0);
161 /* Superblock realloc/free: */
163 void bch2_free_super(struct bch_sb_handle *sb)
166 if (!IS_ERR_OR_NULL(sb->bdev))
167 blkdev_put(sb->bdev, sb->holder);
172 memset(sb, 0, sizeof(*sb));
175 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
177 size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
178 size_t new_buffer_size;
179 struct bch_sb *new_sb;
183 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
185 new_buffer_size = roundup_pow_of_two(new_bytes);
187 if (sb->sb && sb->buffer_size >= new_buffer_size)
190 if (sb->sb && sb->have_layout) {
191 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
193 if (new_bytes > max_bytes) {
194 pr_err("%pg: superblock too big: want %zu but have %llu",
195 sb->bdev, new_bytes, max_bytes);
196 return -BCH_ERR_ENOSPC_sb;
200 if (sb->buffer_size >= new_buffer_size && sb->sb)
203 if (dynamic_fault("bcachefs:add:super_realloc"))
204 return -BCH_ERR_ENOMEM_sb_realloc_injected;
206 new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
208 return -BCH_ERR_ENOMEM_sb_buf_realloc;
213 unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);
215 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
217 return -BCH_ERR_ENOMEM_sb_bio_realloc;
219 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
225 sb->buffer_size = new_buffer_size;
230 struct bch_sb_field *bch2_sb_field_resize_id(struct bch_sb_handle *sb,
231 enum bch_sb_field_type type,
234 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
235 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
236 ssize_t d = -old_u64s + u64s;
238 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
242 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
246 lockdep_assert_held(&c->sb_lock);
248 /* XXX: we're not checking that offline device have enough space */
250 for_each_online_member(ca, c, i) {
251 struct bch_sb_handle *dev_sb = &ca->disk_sb;
253 if (bch2_sb_realloc(dev_sb, le32_to_cpu(dev_sb->sb->u64s) + d)) {
254 percpu_ref_put(&ca->ref);
260 f = bch2_sb_field_get_id(sb->sb, type);
261 f = __bch2_sb_field_resize(sb, f, u64s);
263 f->type = cpu_to_le32(type);
267 /* Superblock validate: */
269 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
271 u64 offset, prev_offset, max_sectors;
274 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
276 if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
277 !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
278 prt_printf(out, "Not a bcachefs superblock layout");
279 return -BCH_ERR_invalid_sb_layout;
282 if (layout->layout_type != 0) {
283 prt_printf(out, "Invalid superblock layout type %u",
284 layout->layout_type);
285 return -BCH_ERR_invalid_sb_layout_type;
288 if (!layout->nr_superblocks) {
289 prt_printf(out, "Invalid superblock layout: no superblocks");
290 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
293 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
294 prt_printf(out, "Invalid superblock layout: too many superblocks");
295 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
298 max_sectors = 1 << layout->sb_max_size_bits;
300 prev_offset = le64_to_cpu(layout->sb_offset[0]);
302 for (i = 1; i < layout->nr_superblocks; i++) {
303 offset = le64_to_cpu(layout->sb_offset[i]);
305 if (offset < prev_offset + max_sectors) {
306 prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
307 " (sb %u ends at %llu next starts at %llu",
308 i - 1, prev_offset + max_sectors, offset);
309 return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
311 prev_offset = offset;
317 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
319 u16 version = le16_to_cpu(sb->version);
320 u16 version_min = le16_to_cpu(sb->version_min);
322 if (!bch2_version_compatible(version)) {
323 prt_str(out, "Unsupported superblock version ");
324 bch2_version_to_text(out, version);
325 prt_str(out, " (min ");
326 bch2_version_to_text(out, bcachefs_metadata_version_min);
327 prt_str(out, ", max ");
328 bch2_version_to_text(out, bcachefs_metadata_version_current);
330 return -BCH_ERR_invalid_sb_version;
333 if (!bch2_version_compatible(version_min)) {
334 prt_str(out, "Unsupported superblock version_min ");
335 bch2_version_to_text(out, version_min);
336 prt_str(out, " (min ");
337 bch2_version_to_text(out, bcachefs_metadata_version_min);
338 prt_str(out, ", max ");
339 bch2_version_to_text(out, bcachefs_metadata_version_current);
341 return -BCH_ERR_invalid_sb_version;
344 if (version_min > version) {
345 prt_str(out, "Bad minimum version ");
346 bch2_version_to_text(out, version_min);
347 prt_str(out, ", greater than version field ");
348 bch2_version_to_text(out, version);
349 return -BCH_ERR_invalid_sb_version;
355 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
358 struct bch_sb *sb = disk_sb->sb;
359 struct bch_sb_field *f;
360 struct bch_sb_field_members_v1 *mi;
361 enum bch_opt_id opt_id;
365 ret = bch2_sb_compatible(sb, out);
369 if (sb->features[1] ||
370 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
371 prt_printf(out, "Filesystem has incompatible features");
372 return -BCH_ERR_invalid_sb_features;
375 block_size = le16_to_cpu(sb->block_size);
377 if (block_size > PAGE_SECTORS) {
378 prt_printf(out, "Block size too big (got %u, max %u)",
379 block_size, PAGE_SECTORS);
380 return -BCH_ERR_invalid_sb_block_size;
383 if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
384 prt_printf(out, "Bad user UUID (got zeroes)");
385 return -BCH_ERR_invalid_sb_uuid;
388 if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
389 prt_printf(out, "Bad internal UUID (got zeroes)");
390 return -BCH_ERR_invalid_sb_uuid;
393 if (!sb->nr_devices ||
394 sb->nr_devices > BCH_SB_MEMBERS_MAX) {
395 prt_printf(out, "Bad number of member devices %u (max %u)",
396 sb->nr_devices, BCH_SB_MEMBERS_MAX);
397 return -BCH_ERR_invalid_sb_too_many_members;
400 if (sb->dev_idx >= sb->nr_devices) {
401 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
402 sb->dev_idx, sb->nr_devices);
403 return -BCH_ERR_invalid_sb_dev_idx;
406 if (!sb->time_precision ||
407 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
408 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
409 le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
410 return -BCH_ERR_invalid_sb_time_precision;
415 * Been seeing a bug where these are getting inexplicably
416 * zeroed, so we're now validating them, but we have to be
417 * careful not to preven people's filesystems from mounting:
419 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
420 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
421 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
422 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
424 if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
425 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
428 for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
429 const struct bch_option *opt = bch2_opt_table + opt_id;
431 if (opt->get_sb != BCH2_NO_SB_OPT) {
432 u64 v = bch2_opt_from_sb(sb, opt_id);
434 prt_printf(out, "Invalid option ");
435 ret = bch2_opt_validate(opt, v, out);
443 /* validate layout */
444 ret = validate_sb_layout(&sb->layout, out);
448 vstruct_for_each(sb, f) {
450 prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
451 le32_to_cpu(f->type));
452 return -BCH_ERR_invalid_sb_field_size;
455 if (vstruct_next(f) > vstruct_last(sb)) {
456 prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
457 le32_to_cpu(f->type));
458 return -BCH_ERR_invalid_sb_field_size;
462 /* members must be validated first: */
463 mi = bch2_sb_field_get(sb, members_v1);
465 prt_printf(out, "Invalid superblock: member info area missing");
466 return -BCH_ERR_invalid_sb_members_missing;
469 ret = bch2_sb_field_validate(sb, &mi->field, out);
473 vstruct_for_each(sb, f) {
474 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
477 ret = bch2_sb_field_validate(sb, f, out);
487 static void bch2_sb_update(struct bch_fs *c)
489 struct bch_sb *src = c->disk_sb.sb;
493 lockdep_assert_held(&c->sb_lock);
495 c->sb.uuid = src->uuid;
496 c->sb.user_uuid = src->user_uuid;
497 c->sb.version = le16_to_cpu(src->version);
498 c->sb.version_min = le16_to_cpu(src->version_min);
499 c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
500 c->sb.nr_devices = src->nr_devices;
501 c->sb.clean = BCH_SB_CLEAN(src);
502 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
504 c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
505 c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
507 /* XXX this is wrong, we need a 96 or 128 bit integer type */
508 c->sb.time_base_lo = div_u64(le64_to_cpu(src->time_base_lo),
509 c->sb.nsec_per_time_unit);
510 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
512 c->sb.features = le64_to_cpu(src->features[0]);
513 c->sb.compat = le64_to_cpu(src->compat[0]);
515 for_each_member_device(ca, c, i) {
516 struct bch_member m = bch2_sb_member_get(src, i);
517 ca->mi = bch2_mi_to_cpu(&m);
521 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
523 struct bch_sb_field *src_f, *dst_f;
524 struct bch_sb *dst = dst_handle->sb;
527 dst->version = src->version;
528 dst->version_min = src->version_min;
530 dst->uuid = src->uuid;
531 dst->user_uuid = src->user_uuid;
532 memcpy(dst->label, src->label, sizeof(dst->label));
534 dst->block_size = src->block_size;
535 dst->nr_devices = src->nr_devices;
537 dst->time_base_lo = src->time_base_lo;
538 dst->time_base_hi = src->time_base_hi;
539 dst->time_precision = src->time_precision;
541 memcpy(dst->flags, src->flags, sizeof(dst->flags));
542 memcpy(dst->features, src->features, sizeof(dst->features));
543 memcpy(dst->compat, src->compat, sizeof(dst->compat));
545 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
548 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
551 src_f = bch2_sb_field_get_id(src, i);
552 dst_f = bch2_sb_field_get_id(dst, i);
554 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
555 (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
557 int ret = bch2_sb_realloc(dst_handle,
558 le32_to_cpu(dst_handle->sb->u64s) + d);
563 dst = dst_handle->sb;
564 dst_f = bch2_sb_field_get_id(dst, i);
567 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
568 src_f ? le32_to_cpu(src_f->u64s) : 0);
571 memcpy(dst_f, src_f, vstruct_bytes(src_f));
577 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
581 lockdep_assert_held(&c->sb_lock);
583 ret = bch2_sb_realloc(&c->disk_sb, 0) ?:
584 __copy_super(&c->disk_sb, src) ?:
585 bch2_sb_replicas_to_cpu_replicas(c) ?:
586 bch2_sb_disk_groups_to_cpu(c);
594 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
596 return __copy_super(&ca->disk_sb, c->disk_sb.sb);
599 /* read superblock: */
601 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
603 struct bch_csum csum;
607 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
608 sb->bio->bi_iter.bi_sector = offset;
609 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
611 ret = submit_bio_wait(sb->bio);
613 prt_printf(err, "IO error: %i", ret);
617 if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
618 !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
619 prt_printf(err, "Not a bcachefs superblock");
620 return -BCH_ERR_invalid_sb_magic;
623 ret = bch2_sb_compatible(sb->sb, err);
627 bytes = vstruct_bytes(sb->sb);
629 if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
630 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
631 bytes, 512UL << sb->sb->layout.sb_max_size_bits);
632 return -BCH_ERR_invalid_sb_too_big;
635 if (bytes > sb->buffer_size) {
636 ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
642 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
643 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
644 return -BCH_ERR_invalid_sb_csum_type;
647 /* XXX: verify MACs */
648 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
649 null_nonce(), sb->sb);
651 if (bch2_crc_cmp(csum, sb->sb->csum)) {
652 prt_printf(err, "bad checksum");
653 return -BCH_ERR_invalid_sb_csum;
656 sb->seq = le64_to_cpu(sb->sb->seq);
661 int __bch2_read_super(const char *path, struct bch_opts *opts,
662 struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
664 u64 offset = opt_get(*opts, sb);
665 struct bch_sb_layout layout;
666 struct printbuf err = PRINTBUF;
667 struct printbuf err2 = PRINTBUF;
673 memset(sb, 0, sizeof(*sb));
674 sb->mode = BLK_OPEN_READ;
676 sb->holder = kmalloc(1, GFP_KERNEL);
680 sb->sb_name = kstrdup(path, GFP_KERNEL);
685 if (opt_get(*opts, direct_io) == false)
686 sb->mode |= BLK_OPEN_BUFFERED;
689 if (!opt_get(*opts, noexcl))
690 sb->mode |= BLK_OPEN_EXCL;
692 if (!opt_get(*opts, nochanges))
693 sb->mode |= BLK_OPEN_WRITE;
695 sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
696 if (IS_ERR(sb->bdev) &&
697 PTR_ERR(sb->bdev) == -EACCES &&
698 opt_get(*opts, read_only)) {
699 sb->mode &= ~BLK_OPEN_WRITE;
701 sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
702 if (!IS_ERR(sb->bdev))
703 opt_set(*opts, nochanges, true);
706 if (IS_ERR(sb->bdev)) {
707 ret = PTR_ERR(sb->bdev);
711 ret = bch2_sb_realloc(sb, 0);
713 prt_printf(&err, "error allocating memory for superblock");
717 if (bch2_fs_init_fault("read_super")) {
718 prt_printf(&err, "dynamic fault");
723 ret = read_one_super(sb, offset, &err);
727 if (opt_defined(*opts, sb))
730 prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
732 if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
733 printk(KERN_INFO "%s", err2.buf);
735 printk(KERN_ERR "%s", err2.buf);
737 printbuf_exit(&err2);
738 printbuf_reset(&err);
741 * Error reading primary superblock - read location of backup
744 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
745 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
747 * use sb buffer to read layout, since sb buffer is page aligned but
750 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
752 ret = submit_bio_wait(sb->bio);
754 prt_printf(&err, "IO error: %i", ret);
758 memcpy(&layout, sb->sb, sizeof(layout));
759 ret = validate_sb_layout(&layout, &err);
763 for (i = layout.sb_offset;
764 i < layout.sb_offset + layout.nr_superblocks; i++) {
765 offset = le64_to_cpu(*i);
767 if (offset == opt_get(*opts, sb))
770 ret = read_one_super(sb, offset, &err);
778 if (le16_to_cpu(sb->sb->block_size) << 9 <
779 bdev_logical_block_size(sb->bdev) &&
780 opt_get(*opts, direct_io)) {
782 opt_set(*opts, direct_io, false);
786 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
787 le16_to_cpu(sb->sb->block_size) << 9,
788 bdev_logical_block_size(sb->bdev));
789 ret = -BCH_ERR_block_size_too_small;
794 sb->have_layout = true;
796 ret = bch2_sb_validate(sb, &err, READ);
798 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
806 printk(KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
813 int bch2_read_super(const char *path, struct bch_opts *opts,
814 struct bch_sb_handle *sb)
816 return __bch2_read_super(path, opts, sb, false);
819 /* provide a silenced version for mount.bcachefs */
821 int bch2_read_super_silent(const char *path, struct bch_opts *opts,
822 struct bch_sb_handle *sb)
824 return __bch2_read_super(path, opts, sb, true);
827 /* write superblock: */
829 static void write_super_endio(struct bio *bio)
831 struct bch_dev *ca = bio->bi_private;
833 /* XXX: return errors directly */
835 if (bch2_dev_io_err_on(bio->bi_status, ca,
837 ? BCH_MEMBER_ERROR_write
838 : BCH_MEMBER_ERROR_read,
839 "superblock %s error: %s",
840 bio_data_dir(bio) ? "write" : "read",
841 bch2_blk_status_to_str(bio->bi_status)))
842 ca->sb_write_error = 1;
844 closure_put(&ca->fs->sb_write);
845 percpu_ref_put(&ca->io_ref);
848 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
850 struct bch_sb *sb = ca->disk_sb.sb;
851 struct bio *bio = ca->disk_sb.bio;
853 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
854 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
855 bio->bi_end_io = write_super_endio;
856 bio->bi_private = ca;
857 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
859 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
862 percpu_ref_get(&ca->io_ref);
863 closure_bio_submit(bio, &c->sb_write);
866 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
868 struct bch_sb *sb = ca->disk_sb.sb;
869 struct bio *bio = ca->disk_sb.bio;
871 sb->offset = sb->layout.sb_offset[idx];
873 SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
874 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
877 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
878 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
879 bio->bi_end_io = write_super_endio;
880 bio->bi_private = ca;
881 bch2_bio_map(bio, sb,
882 roundup((size_t) vstruct_bytes(sb),
883 bdev_logical_block_size(ca->disk_sb.bdev)));
885 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
888 percpu_ref_get(&ca->io_ref);
889 closure_bio_submit(bio, &c->sb_write);
892 int bch2_write_super(struct bch_fs *c)
894 struct closure *cl = &c->sb_write;
896 struct printbuf err = PRINTBUF;
897 unsigned i, sb = 0, nr_wrote;
898 struct bch_devs_mask sb_written;
899 bool wrote, can_mount_without_written, can_mount_with_written;
900 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
903 trace_and_count(c, write_super, c, _RET_IP_);
905 if (c->opts.very_degraded)
906 degraded_flags |= BCH_FORCE_IF_LOST;
908 lockdep_assert_held(&c->sb_lock);
910 closure_init_stack(cl);
911 memset(&sb_written, 0, sizeof(sb_written));
913 /* Make sure we're using the new magic numbers: */
914 c->disk_sb.sb->magic = BCHFS_MAGIC;
915 c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
917 le64_add_cpu(&c->disk_sb.sb->seq, 1);
919 if (test_bit(BCH_FS_ERROR, &c->flags))
920 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
921 if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
922 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
924 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
926 bch2_sb_counters_from_cpu(c);
927 bch2_sb_members_from_cpu(c);
928 bch2_sb_members_cpy_v2_v1(&c->disk_sb);
929 bch2_sb_errors_from_cpu(c);
931 for_each_online_member(ca, c, i)
932 bch2_sb_from_fs(c, ca);
934 for_each_online_member(ca, c, i) {
935 printbuf_reset(&err);
937 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
939 bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
940 percpu_ref_put(&ca->io_ref);
945 if (c->opts.nochanges)
949 * Defer writing the superblock until filesystem initialization is
950 * complete - don't write out a partly initialized superblock:
952 if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
955 for_each_online_member(ca, c, i) {
956 __set_bit(ca->dev_idx, sb_written.d);
957 ca->sb_write_error = 0;
960 for_each_online_member(ca, c, i)
961 read_back_super(c, ca);
964 for_each_online_member(ca, c, i) {
965 if (ca->sb_write_error)
968 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
969 bch2_fs_fatal_error(c,
970 "Superblock write was silently dropped! (seq %llu expected %llu)",
971 le64_to_cpu(ca->sb_read_scratch->seq),
973 percpu_ref_put(&ca->io_ref);
974 ret = -BCH_ERR_erofs_sb_err;
978 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
979 bch2_fs_fatal_error(c,
980 "Superblock modified by another process (seq %llu expected %llu)",
981 le64_to_cpu(ca->sb_read_scratch->seq),
983 percpu_ref_put(&ca->io_ref);
984 ret = -BCH_ERR_erofs_sb_err;
991 for_each_online_member(ca, c, i)
992 if (!ca->sb_write_error &&
993 sb < ca->disk_sb.sb->layout.nr_superblocks) {
994 write_one_super(c, ca, sb);
1001 for_each_online_member(ca, c, i) {
1002 if (ca->sb_write_error)
1003 __clear_bit(ca->dev_idx, sb_written.d);
1005 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
1008 nr_wrote = dev_mask_nr(&sb_written);
1010 can_mount_with_written =
1011 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1013 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
1014 sb_written.d[i] = ~sb_written.d[i];
1016 can_mount_without_written =
1017 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1020 * If we would be able to mount _without_ the devices we successfully
1021 * wrote superblocks to, we weren't able to write to enough devices:
1023 * Exception: if we can mount without the successes because we haven't
1024 * written anything (new filesystem), we continue if we'd be able to
1025 * mount with the devices we did successfully write to:
1027 if (bch2_fs_fatal_err_on(!nr_wrote ||
1028 !can_mount_with_written ||
1029 (can_mount_without_written &&
1030 !can_mount_with_written), c,
1031 "Unable to write superblock to sufficient devices (from %ps)",
1035 /* Make new options visible after they're persistent: */
1037 printbuf_exit(&err);
1041 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
1043 mutex_lock(&c->sb_lock);
1044 if (!(c->sb.features & (1ULL << feat))) {
1045 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1047 bch2_write_super(c);
1049 mutex_unlock(&c->sb_lock);
1052 /* Downgrade if superblock is at a higher version than currently supported: */
1053 void bch2_sb_maybe_downgrade(struct bch_fs *c)
1055 lockdep_assert_held(&c->sb_lock);
1058 * Downgrade, if superblock is at a higher version than currently
1061 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
1062 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1063 if (c->sb.version > bcachefs_metadata_version_current)
1064 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1065 if (c->sb.version_min > bcachefs_metadata_version_current)
1066 c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1067 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1070 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
1072 lockdep_assert_held(&c->sb_lock);
1074 c->disk_sb.sb->version = cpu_to_le16(new_version);
1075 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1078 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1080 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1085 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1087 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1089 return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1090 ? bch2_sb_field_ops[type]
1091 : &bch2_sb_field_null_ops;
1094 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1095 struct printbuf *err)
1097 unsigned type = le32_to_cpu(f->type);
1098 struct printbuf field_err = PRINTBUF;
1099 const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1102 ret = ops->validate ? ops->validate(sb, f, &field_err) : 0;
1104 prt_printf(err, "Invalid superblock section %s: %s",
1105 bch2_sb_fields[type], field_err.buf);
1107 bch2_sb_field_to_text(err, sb, f);
1110 printbuf_exit(&field_err);
1114 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1115 struct bch_sb_field *f)
1117 unsigned type = le32_to_cpu(f->type);
1118 const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1120 if (!out->nr_tabstops)
1121 printbuf_tabstop_push(out, 32);
1123 if (type < BCH_SB_FIELD_NR)
1124 prt_printf(out, "%s", bch2_sb_fields[type]);
1126 prt_printf(out, "(unknown field %u)", type);
1128 prt_printf(out, " (size %zu):", vstruct_bytes(f));
1132 printbuf_indent_add(out, 2);
1133 ops->to_text(out, sb, f);
1134 printbuf_indent_sub(out, 2);
1138 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1142 prt_printf(out, "Type: %u", l->layout_type);
1145 prt_str(out, "Superblock max size: ");
1146 prt_units_u64(out, 512 << l->sb_max_size_bits);
1149 prt_printf(out, "Nr superblocks: %u", l->nr_superblocks);
1152 prt_str(out, "Offsets: ");
1153 for (i = 0; i < l->nr_superblocks; i++) {
1156 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1161 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1162 bool print_layout, unsigned fields)
1164 struct bch_sb_field *f;
1165 u64 fields_have = 0;
1166 unsigned nr_devices = 0;
1168 if (!out->nr_tabstops)
1169 printbuf_tabstop_push(out, 44);
1171 for (int i = 0; i < sb->nr_devices; i++)
1172 nr_devices += bch2_dev_exists(sb, i);
1174 prt_printf(out, "External UUID:");
1176 pr_uuid(out, sb->user_uuid.b);
1179 prt_printf(out, "Internal UUID:");
1181 pr_uuid(out, sb->uuid.b);
1184 prt_str(out, "Device index:");
1186 prt_printf(out, "%u", sb->dev_idx);
1189 prt_str(out, "Label:");
1191 prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1194 prt_str(out, "Version:");
1196 bch2_version_to_text(out, le16_to_cpu(sb->version));
1199 prt_str(out, "Version upgrade complete:");
1201 bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1204 prt_printf(out, "Oldest version on disk:");
1206 bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1209 prt_printf(out, "Created:");
1211 if (sb->time_base_lo)
1212 bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1214 prt_printf(out, "(not set)");
1217 prt_printf(out, "Sequence number:");
1219 prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1222 prt_printf(out, "Superblock size:");
1224 prt_printf(out, "%zu", vstruct_bytes(sb));
1227 prt_printf(out, "Clean:");
1229 prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1232 prt_printf(out, "Devices:");
1234 prt_printf(out, "%u", nr_devices);
1237 prt_printf(out, "Sections:");
1238 vstruct_for_each(sb, f)
1239 fields_have |= 1 << le32_to_cpu(f->type);
1241 prt_bitflags(out, bch2_sb_fields, fields_have);
1244 prt_printf(out, "Features:");
1246 prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1249 prt_printf(out, "Compat features:");
1251 prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1255 prt_printf(out, "Options:");
1257 printbuf_indent_add(out, 2);
1261 for (id = 0; id < bch2_opts_nr; id++) {
1262 const struct bch_option *opt = bch2_opt_table + id;
1264 if (opt->get_sb != BCH2_NO_SB_OPT) {
1265 u64 v = bch2_opt_from_sb(sb, id);
1267 prt_printf(out, "%s:", opt->attr.name);
1269 bch2_opt_to_text(out, NULL, sb, opt, v,
1270 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1276 printbuf_indent_sub(out, 2);
1280 prt_printf(out, "layout:");
1282 printbuf_indent_add(out, 2);
1283 bch2_sb_layout_to_text(out, &sb->layout);
1284 printbuf_indent_sub(out, 2);
1287 vstruct_for_each(sb, f)
1288 if (fields & (1 << le32_to_cpu(f->type))) {
1290 bch2_sb_field_to_text(out, sb, f);