1 // SPDX-License-Identifier: GPL-2.0
5 #include "disk_groups.h"
9 #include "journal_sb.h"
10 #include "journal_seq_blacklist.h"
15 #include "sb-counters.h"
16 #include "sb-downgrade.h"
17 #include "sb-errors.h"
18 #include "sb-members.h"
24 #include <linux/backing-dev.h>
25 #include <linux/sort.h>
27 static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
30 struct bch2_metadata_version {
35 static const struct bch2_metadata_version bch2_metadata_versions[] = {
40 BCH_METADATA_VERSIONS()
44 void bch2_version_to_text(struct printbuf *out, unsigned v)
46 const char *str = "(unknown version)";
48 for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
49 if (bch2_metadata_versions[i].version == v) {
50 str = bch2_metadata_versions[i].name;
54 prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
57 unsigned bch2_latest_compatible_version(unsigned v)
59 if (!BCH_VERSION_MAJOR(v))
62 for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
63 if (bch2_metadata_versions[i].version > v &&
64 BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
66 v = bch2_metadata_versions[i].version;
71 const char * const bch2_sb_fields[] = {
72 #define x(name, nr) #name,
78 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
81 struct bch_sb_field *bch2_sb_field_get_id(struct bch_sb *sb,
82 enum bch_sb_field_type type)
84 /* XXX: need locking around superblock to access optional fields */
86 vstruct_for_each(sb, f)
87 if (le32_to_cpu(f->type) == type)
92 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
93 struct bch_sb_field *f,
96 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
97 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
99 BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
104 f = vstruct_last(sb->sb);
105 memset(f, 0, sizeof(u64) * u64s);
106 f->u64s = cpu_to_le32(u64s);
111 src = vstruct_end(f);
114 f->u64s = cpu_to_le32(u64s);
115 dst = vstruct_end(f);
120 memmove(dst, src, vstruct_end(sb->sb) - src);
123 memset(src, 0, dst - src);
126 sb->sb->u64s = cpu_to_le32(sb_u64s);
128 return u64s ? f : NULL;
131 void bch2_sb_field_delete(struct bch_sb_handle *sb,
132 enum bch_sb_field_type type)
134 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
137 __bch2_sb_field_resize(sb, f, 0);
140 /* Superblock realloc/free: */
142 void bch2_free_super(struct bch_sb_handle *sb)
145 if (!IS_ERR_OR_NULL(sb->bdev_handle))
146 bdev_release(sb->bdev_handle);
151 memset(sb, 0, sizeof(*sb));
154 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
156 size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
157 size_t new_buffer_size;
158 struct bch_sb *new_sb;
162 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
164 new_buffer_size = roundup_pow_of_two(new_bytes);
166 if (sb->sb && sb->buffer_size >= new_buffer_size)
169 if (sb->sb && sb->have_layout) {
170 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
172 if (new_bytes > max_bytes) {
173 struct printbuf buf = PRINTBUF;
175 prt_bdevname(&buf, sb->bdev);
176 prt_printf(&buf, ": superblock too big: want %zu but have %llu", new_bytes, max_bytes);
177 pr_err("%s", buf.buf);
179 return -BCH_ERR_ENOSPC_sb;
183 if (sb->buffer_size >= new_buffer_size && sb->sb)
186 if (dynamic_fault("bcachefs:add:super_realloc"))
187 return -BCH_ERR_ENOMEM_sb_realloc_injected;
189 new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
191 return -BCH_ERR_ENOMEM_sb_buf_realloc;
196 unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);
198 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
200 return -BCH_ERR_ENOMEM_sb_bio_realloc;
202 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
208 sb->buffer_size = new_buffer_size;
213 struct bch_sb_field *bch2_sb_field_resize_id(struct bch_sb_handle *sb,
214 enum bch_sb_field_type type,
217 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
218 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
219 ssize_t d = -old_u64s + u64s;
221 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
225 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
227 lockdep_assert_held(&c->sb_lock);
229 /* XXX: we're not checking that offline device have enough space */
231 for_each_online_member(c, ca) {
232 struct bch_sb_handle *dev_sb = &ca->disk_sb;
234 if (bch2_sb_realloc(dev_sb, le32_to_cpu(dev_sb->sb->u64s) + d)) {
235 percpu_ref_put(&ca->ref);
241 f = bch2_sb_field_get_id(sb->sb, type);
242 f = __bch2_sb_field_resize(sb, f, u64s);
244 f->type = cpu_to_le32(type);
248 struct bch_sb_field *bch2_sb_field_get_minsize_id(struct bch_sb_handle *sb,
249 enum bch_sb_field_type type,
252 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
254 if (!f || le32_to_cpu(f->u64s) < u64s)
255 f = bch2_sb_field_resize_id(sb, type, u64s);
259 /* Superblock validate: */
261 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
263 u64 offset, prev_offset, max_sectors;
266 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
268 if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
269 !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
270 prt_printf(out, "Not a bcachefs superblock layout");
271 return -BCH_ERR_invalid_sb_layout;
274 if (layout->layout_type != 0) {
275 prt_printf(out, "Invalid superblock layout type %u",
276 layout->layout_type);
277 return -BCH_ERR_invalid_sb_layout_type;
280 if (!layout->nr_superblocks) {
281 prt_printf(out, "Invalid superblock layout: no superblocks");
282 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
285 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
286 prt_printf(out, "Invalid superblock layout: too many superblocks");
287 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
290 max_sectors = 1 << layout->sb_max_size_bits;
292 prev_offset = le64_to_cpu(layout->sb_offset[0]);
294 for (i = 1; i < layout->nr_superblocks; i++) {
295 offset = le64_to_cpu(layout->sb_offset[i]);
297 if (offset < prev_offset + max_sectors) {
298 prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
299 " (sb %u ends at %llu next starts at %llu",
300 i - 1, prev_offset + max_sectors, offset);
301 return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
303 prev_offset = offset;
309 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
311 u16 version = le16_to_cpu(sb->version);
312 u16 version_min = le16_to_cpu(sb->version_min);
314 if (!bch2_version_compatible(version)) {
315 prt_str(out, "Unsupported superblock version ");
316 bch2_version_to_text(out, version);
317 prt_str(out, " (min ");
318 bch2_version_to_text(out, bcachefs_metadata_version_min);
319 prt_str(out, ", max ");
320 bch2_version_to_text(out, bcachefs_metadata_version_current);
322 return -BCH_ERR_invalid_sb_version;
325 if (!bch2_version_compatible(version_min)) {
326 prt_str(out, "Unsupported superblock version_min ");
327 bch2_version_to_text(out, version_min);
328 prt_str(out, " (min ");
329 bch2_version_to_text(out, bcachefs_metadata_version_min);
330 prt_str(out, ", max ");
331 bch2_version_to_text(out, bcachefs_metadata_version_current);
333 return -BCH_ERR_invalid_sb_version;
336 if (version_min > version) {
337 prt_str(out, "Bad minimum version ");
338 bch2_version_to_text(out, version_min);
339 prt_str(out, ", greater than version field ");
340 bch2_version_to_text(out, version);
341 return -BCH_ERR_invalid_sb_version;
347 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
350 struct bch_sb *sb = disk_sb->sb;
351 struct bch_sb_field_members_v1 *mi;
352 enum bch_opt_id opt_id;
356 ret = bch2_sb_compatible(sb, out);
360 if (sb->features[1] ||
361 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
362 prt_printf(out, "Filesystem has incompatible features");
363 return -BCH_ERR_invalid_sb_features;
366 block_size = le16_to_cpu(sb->block_size);
368 if (block_size > PAGE_SECTORS) {
369 prt_printf(out, "Block size too big (got %u, max %u)",
370 block_size, PAGE_SECTORS);
371 return -BCH_ERR_invalid_sb_block_size;
374 if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
375 prt_printf(out, "Bad user UUID (got zeroes)");
376 return -BCH_ERR_invalid_sb_uuid;
379 if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
380 prt_printf(out, "Bad internal UUID (got zeroes)");
381 return -BCH_ERR_invalid_sb_uuid;
384 if (!sb->nr_devices ||
385 sb->nr_devices > BCH_SB_MEMBERS_MAX) {
386 prt_printf(out, "Bad number of member devices %u (max %u)",
387 sb->nr_devices, BCH_SB_MEMBERS_MAX);
388 return -BCH_ERR_invalid_sb_too_many_members;
391 if (sb->dev_idx >= sb->nr_devices) {
392 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
393 sb->dev_idx, sb->nr_devices);
394 return -BCH_ERR_invalid_sb_dev_idx;
397 if (!sb->time_precision ||
398 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
399 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
400 le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
401 return -BCH_ERR_invalid_sb_time_precision;
406 * Been seeing a bug where these are getting inexplicably
407 * zeroed, so we're now validating them, but we have to be
408 * careful not to preven people's filesystems from mounting:
410 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
411 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
412 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
413 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
415 if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
416 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
419 for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
420 const struct bch_option *opt = bch2_opt_table + opt_id;
422 if (opt->get_sb != BCH2_NO_SB_OPT) {
423 u64 v = bch2_opt_from_sb(sb, opt_id);
425 prt_printf(out, "Invalid option ");
426 ret = bch2_opt_validate(opt, v, out);
434 /* validate layout */
435 ret = validate_sb_layout(&sb->layout, out);
439 vstruct_for_each(sb, f) {
441 prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
442 le32_to_cpu(f->type));
443 return -BCH_ERR_invalid_sb_field_size;
446 if (vstruct_next(f) > vstruct_last(sb)) {
447 prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
448 le32_to_cpu(f->type));
449 return -BCH_ERR_invalid_sb_field_size;
453 /* members must be validated first: */
454 mi = bch2_sb_field_get(sb, members_v1);
456 prt_printf(out, "Invalid superblock: member info area missing");
457 return -BCH_ERR_invalid_sb_members_missing;
460 ret = bch2_sb_field_validate(sb, &mi->field, out);
464 vstruct_for_each(sb, f) {
465 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
468 ret = bch2_sb_field_validate(sb, f, out);
478 static unsigned long le_ulong_to_cpu(unsigned long v)
480 return sizeof(unsigned long) == 8
485 static void le_bitvector_to_cpu(unsigned long *dst, unsigned long *src, unsigned nr)
487 BUG_ON(nr & (BITS_PER_TYPE(long) - 1));
489 for (unsigned i = 0; i < BITS_TO_LONGS(nr); i++)
490 dst[i] = le_ulong_to_cpu(src[i]);
493 static void bch2_sb_update(struct bch_fs *c)
495 struct bch_sb *src = c->disk_sb.sb;
497 lockdep_assert_held(&c->sb_lock);
499 c->sb.uuid = src->uuid;
500 c->sb.user_uuid = src->user_uuid;
501 c->sb.version = le16_to_cpu(src->version);
502 c->sb.version_min = le16_to_cpu(src->version_min);
503 c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
504 c->sb.nr_devices = src->nr_devices;
505 c->sb.clean = BCH_SB_CLEAN(src);
506 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
508 c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
509 c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
511 /* XXX this is wrong, we need a 96 or 128 bit integer type */
512 c->sb.time_base_lo = div_u64(le64_to_cpu(src->time_base_lo),
513 c->sb.nsec_per_time_unit);
514 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
516 c->sb.features = le64_to_cpu(src->features[0]);
517 c->sb.compat = le64_to_cpu(src->compat[0]);
519 memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));
521 struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
523 le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
524 sizeof(c->sb.errors_silent) * 8);
526 for_each_member_device(c, ca) {
527 struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
528 ca->mi = bch2_mi_to_cpu(&m);
532 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
534 struct bch_sb_field *src_f, *dst_f;
535 struct bch_sb *dst = dst_handle->sb;
538 dst->version = src->version;
539 dst->version_min = src->version_min;
541 dst->uuid = src->uuid;
542 dst->user_uuid = src->user_uuid;
543 memcpy(dst->label, src->label, sizeof(dst->label));
545 dst->block_size = src->block_size;
546 dst->nr_devices = src->nr_devices;
548 dst->time_base_lo = src->time_base_lo;
549 dst->time_base_hi = src->time_base_hi;
550 dst->time_precision = src->time_precision;
551 dst->write_time = src->write_time;
553 memcpy(dst->flags, src->flags, sizeof(dst->flags));
554 memcpy(dst->features, src->features, sizeof(dst->features));
555 memcpy(dst->compat, src->compat, sizeof(dst->compat));
557 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
560 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
563 src_f = bch2_sb_field_get_id(src, i);
564 dst_f = bch2_sb_field_get_id(dst, i);
566 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
567 (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
569 int ret = bch2_sb_realloc(dst_handle,
570 le32_to_cpu(dst_handle->sb->u64s) + d);
575 dst = dst_handle->sb;
576 dst_f = bch2_sb_field_get_id(dst, i);
579 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
580 src_f ? le32_to_cpu(src_f->u64s) : 0);
583 memcpy(dst_f, src_f, vstruct_bytes(src_f));
589 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
593 lockdep_assert_held(&c->sb_lock);
595 ret = bch2_sb_realloc(&c->disk_sb, 0) ?:
596 __copy_super(&c->disk_sb, src) ?:
597 bch2_sb_replicas_to_cpu_replicas(c) ?:
598 bch2_sb_disk_groups_to_cpu(c);
606 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
608 return __copy_super(&ca->disk_sb, c->disk_sb.sb);
611 /* read superblock: */
613 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
618 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
619 sb->bio->bi_iter.bi_sector = offset;
620 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
622 ret = submit_bio_wait(sb->bio);
624 prt_printf(err, "IO error: %i", ret);
628 if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
629 !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
630 prt_str(err, "Not a bcachefs superblock (got magic ");
631 pr_uuid(err, sb->sb->magic.b);
633 return -BCH_ERR_invalid_sb_magic;
636 ret = bch2_sb_compatible(sb->sb, err);
640 bytes = vstruct_bytes(sb->sb);
642 if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
643 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
644 bytes, 512UL << sb->sb->layout.sb_max_size_bits);
645 return -BCH_ERR_invalid_sb_too_big;
648 if (bytes > sb->buffer_size) {
649 ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
655 enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
656 if (csum_type >= BCH_CSUM_NR) {
657 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
658 return -BCH_ERR_invalid_sb_csum_type;
661 /* XXX: verify MACs */
662 struct bch_csum csum = csum_vstruct(NULL, csum_type, null_nonce(), sb->sb);
663 if (bch2_crc_cmp(csum, sb->sb->csum)) {
664 bch2_csum_err_msg(err, csum_type, sb->sb->csum, csum);
665 return -BCH_ERR_invalid_sb_csum;
668 sb->seq = le64_to_cpu(sb->sb->seq);
673 static int __bch2_read_super(const char *path, struct bch_opts *opts,
674 struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
676 u64 offset = opt_get(*opts, sb);
677 struct bch_sb_layout layout;
678 struct printbuf err = PRINTBUF;
679 struct printbuf err2 = PRINTBUF;
685 memset(sb, 0, sizeof(*sb));
686 sb->mode = BLK_OPEN_READ;
688 sb->holder = kmalloc(1, GFP_KERNEL);
692 sb->sb_name = kstrdup(path, GFP_KERNEL);
697 if (opt_get(*opts, direct_io) == false)
698 sb->mode |= BLK_OPEN_BUFFERED;
701 if (!opt_get(*opts, noexcl))
702 sb->mode |= BLK_OPEN_EXCL;
704 if (!opt_get(*opts, nochanges))
705 sb->mode |= BLK_OPEN_WRITE;
707 sb->bdev_handle = bdev_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
708 if (IS_ERR(sb->bdev_handle) &&
709 PTR_ERR(sb->bdev_handle) == -EACCES &&
710 opt_get(*opts, read_only)) {
711 sb->mode &= ~BLK_OPEN_WRITE;
713 sb->bdev_handle = bdev_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
714 if (!IS_ERR(sb->bdev_handle))
715 opt_set(*opts, nochanges, true);
718 if (IS_ERR(sb->bdev_handle)) {
719 ret = PTR_ERR(sb->bdev_handle);
722 sb->bdev = sb->bdev_handle->bdev;
724 ret = bch2_sb_realloc(sb, 0);
726 prt_printf(&err, "error allocating memory for superblock");
730 if (bch2_fs_init_fault("read_super")) {
731 prt_printf(&err, "dynamic fault");
736 ret = read_one_super(sb, offset, &err);
740 if (opt_defined(*opts, sb))
743 prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
745 if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
746 printk(KERN_INFO "%s", err2.buf);
748 printk(KERN_ERR "%s", err2.buf);
750 printbuf_exit(&err2);
751 printbuf_reset(&err);
754 * Error reading primary superblock - read location of backup
757 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
758 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
760 * use sb buffer to read layout, since sb buffer is page aligned but
763 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
765 ret = submit_bio_wait(sb->bio);
767 prt_printf(&err, "IO error: %i", ret);
771 memcpy(&layout, sb->sb, sizeof(layout));
772 ret = validate_sb_layout(&layout, &err);
776 for (i = layout.sb_offset;
777 i < layout.sb_offset + layout.nr_superblocks; i++) {
778 offset = le64_to_cpu(*i);
780 if (offset == opt_get(*opts, sb))
783 ret = read_one_super(sb, offset, &err);
791 if (le16_to_cpu(sb->sb->block_size) << 9 <
792 bdev_logical_block_size(sb->bdev) &&
793 opt_get(*opts, direct_io)) {
795 opt_set(*opts, direct_io, false);
799 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
800 le16_to_cpu(sb->sb->block_size) << 9,
801 bdev_logical_block_size(sb->bdev));
802 ret = -BCH_ERR_block_size_too_small;
807 sb->have_layout = true;
809 ret = bch2_sb_validate(sb, &err, READ);
811 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
819 printk(KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
826 int bch2_read_super(const char *path, struct bch_opts *opts,
827 struct bch_sb_handle *sb)
829 return __bch2_read_super(path, opts, sb, false);
832 /* provide a silenced version for mount.bcachefs */
834 int bch2_read_super_silent(const char *path, struct bch_opts *opts,
835 struct bch_sb_handle *sb)
837 return __bch2_read_super(path, opts, sb, true);
840 /* write superblock: */
842 static void write_super_endio(struct bio *bio)
844 struct bch_dev *ca = bio->bi_private;
846 /* XXX: return errors directly */
848 if (bch2_dev_io_err_on(bio->bi_status, ca,
850 ? BCH_MEMBER_ERROR_write
851 : BCH_MEMBER_ERROR_read,
852 "superblock %s error: %s",
853 bio_data_dir(bio) ? "write" : "read",
854 bch2_blk_status_to_str(bio->bi_status)))
855 ca->sb_write_error = 1;
857 closure_put(&ca->fs->sb_write);
858 percpu_ref_put(&ca->io_ref);
861 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
863 struct bch_sb *sb = ca->disk_sb.sb;
864 struct bio *bio = ca->disk_sb.bio;
866 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
867 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
868 bio->bi_end_io = write_super_endio;
869 bio->bi_private = ca;
870 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
872 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
875 percpu_ref_get(&ca->io_ref);
876 closure_bio_submit(bio, &c->sb_write);
879 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
881 struct bch_sb *sb = ca->disk_sb.sb;
882 struct bio *bio = ca->disk_sb.bio;
884 sb->offset = sb->layout.sb_offset[idx];
886 SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
887 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
890 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
891 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
892 bio->bi_end_io = write_super_endio;
893 bio->bi_private = ca;
894 bch2_bio_map(bio, sb,
895 roundup((size_t) vstruct_bytes(sb),
896 bdev_logical_block_size(ca->disk_sb.bdev)));
898 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
901 percpu_ref_get(&ca->io_ref);
902 closure_bio_submit(bio, &c->sb_write);
905 int bch2_write_super(struct bch_fs *c)
907 struct closure *cl = &c->sb_write;
908 struct printbuf err = PRINTBUF;
909 unsigned sb = 0, nr_wrote;
910 struct bch_devs_mask sb_written;
911 bool wrote, can_mount_without_written, can_mount_with_written;
912 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
915 trace_and_count(c, write_super, c, _RET_IP_);
917 if (c->opts.very_degraded)
918 degraded_flags |= BCH_FORCE_IF_LOST;
920 lockdep_assert_held(&c->sb_lock);
922 closure_init_stack(cl);
923 memset(&sb_written, 0, sizeof(sb_written));
925 /* Make sure we're using the new magic numbers: */
926 c->disk_sb.sb->magic = BCHFS_MAGIC;
927 c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
929 le64_add_cpu(&c->disk_sb.sb->seq, 1);
931 struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
932 for_each_online_member(c, ca)
933 __bch2_members_v2_get_mut(mi, ca->dev_idx)->seq = c->disk_sb.sb->seq;
934 c->disk_sb.sb->write_time = cpu_to_le64(ktime_get_real_seconds());
936 if (test_bit(BCH_FS_error, &c->flags))
937 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
938 if (test_bit(BCH_FS_topology_error, &c->flags))
939 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
941 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
943 bch2_sb_counters_from_cpu(c);
944 bch2_sb_members_from_cpu(c);
945 bch2_sb_members_cpy_v2_v1(&c->disk_sb);
946 bch2_sb_errors_from_cpu(c);
947 bch2_sb_downgrade_update(c);
949 for_each_online_member(c, ca)
950 bch2_sb_from_fs(c, ca);
952 for_each_online_member(c, ca) {
953 printbuf_reset(&err);
955 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
957 bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
958 percpu_ref_put(&ca->io_ref);
963 if (c->opts.nochanges)
967 * Defer writing the superblock until filesystem initialization is
968 * complete - don't write out a partly initialized superblock:
970 if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
973 if (le16_to_cpu(c->disk_sb.sb->version) > bcachefs_metadata_version_current) {
974 struct printbuf buf = PRINTBUF;
975 prt_printf(&buf, "attempting to write superblock that wasn't version downgraded (");
976 bch2_version_to_text(&buf, le16_to_cpu(c->disk_sb.sb->version));
977 prt_str(&buf, " > ");
978 bch2_version_to_text(&buf, bcachefs_metadata_version_current);
980 bch2_fs_fatal_error(c, "%s", buf.buf);
982 return -BCH_ERR_sb_not_downgraded;
985 for_each_online_member(c, ca) {
986 __set_bit(ca->dev_idx, sb_written.d);
987 ca->sb_write_error = 0;
990 for_each_online_member(c, ca)
991 read_back_super(c, ca);
994 for_each_online_member(c, ca) {
995 if (ca->sb_write_error)
998 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
999 bch2_fs_fatal_error(c,
1000 "Superblock write was silently dropped! (seq %llu expected %llu)",
1001 le64_to_cpu(ca->sb_read_scratch->seq),
1003 percpu_ref_put(&ca->io_ref);
1004 ret = -BCH_ERR_erofs_sb_err;
1008 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
1009 bch2_fs_fatal_error(c,
1010 "Superblock modified by another process (seq %llu expected %llu)",
1011 le64_to_cpu(ca->sb_read_scratch->seq),
1013 percpu_ref_put(&ca->io_ref);
1014 ret = -BCH_ERR_erofs_sb_err;
1021 for_each_online_member(c, ca)
1022 if (!ca->sb_write_error &&
1023 sb < ca->disk_sb.sb->layout.nr_superblocks) {
1024 write_one_super(c, ca, sb);
1031 for_each_online_member(c, ca) {
1032 if (ca->sb_write_error)
1033 __clear_bit(ca->dev_idx, sb_written.d);
1035 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
1038 nr_wrote = dev_mask_nr(&sb_written);
1040 can_mount_with_written =
1041 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1043 for (unsigned i = 0; i < ARRAY_SIZE(sb_written.d); i++)
1044 sb_written.d[i] = ~sb_written.d[i];
1046 can_mount_without_written =
1047 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1050 * If we would be able to mount _without_ the devices we successfully
1051 * wrote superblocks to, we weren't able to write to enough devices:
1053 * Exception: if we can mount without the successes because we haven't
1054 * written anything (new filesystem), we continue if we'd be able to
1055 * mount with the devices we did successfully write to:
1057 if (bch2_fs_fatal_err_on(!nr_wrote ||
1058 !can_mount_with_written ||
1059 (can_mount_without_written &&
1060 !can_mount_with_written), c,
1061 "Unable to write superblock to sufficient devices (from %ps)",
1065 /* Make new options visible after they're persistent: */
1067 printbuf_exit(&err);
1071 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
1073 mutex_lock(&c->sb_lock);
1074 if (!(c->sb.features & (1ULL << feat))) {
1075 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1077 bch2_write_super(c);
1079 mutex_unlock(&c->sb_lock);
1082 /* Downgrade if superblock is at a higher version than currently supported: */
1083 bool bch2_check_version_downgrade(struct bch_fs *c)
1085 bool ret = bcachefs_metadata_version_current < c->sb.version;
1087 lockdep_assert_held(&c->sb_lock);
1090 * Downgrade, if superblock is at a higher version than currently
1093 * c->sb will be checked before we write the superblock, so update it as
1096 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current) {
1097 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1098 c->sb.version_upgrade_complete = bcachefs_metadata_version_current;
1100 if (c->sb.version > bcachefs_metadata_version_current) {
1101 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1102 c->sb.version = bcachefs_metadata_version_current;
1104 if (c->sb.version_min > bcachefs_metadata_version_current) {
1105 c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1106 c->sb.version_min = bcachefs_metadata_version_current;
1108 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1112 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
1114 lockdep_assert_held(&c->sb_lock);
1116 if (BCH_VERSION_MAJOR(new_version) >
1117 BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
1118 bch2_sb_field_resize(&c->disk_sb, downgrade, 0);
1120 c->disk_sb.sb->version = cpu_to_le16(new_version);
1121 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1124 static int bch2_sb_ext_validate(struct bch_sb *sb, struct bch_sb_field *f,
1125 struct printbuf *err)
1127 if (vstruct_bytes(f) < 88) {
1128 prt_printf(err, "field too small (%zu < %u)", vstruct_bytes(f), 88);
1129 return -BCH_ERR_invalid_sb_ext;
1135 static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
1136 struct bch_sb_field *f)
1138 struct bch_sb_field_ext *e = field_to_type(f, ext);
1140 prt_printf(out, "Recovery passes required:");
1142 prt_bitflags(out, bch2_recovery_passes,
1143 bch2_recovery_passes_from_stable(le64_to_cpu(e->recovery_passes_required[0])));
1146 unsigned long *errors_silent = kmalloc(sizeof(e->errors_silent), GFP_KERNEL);
1147 if (errors_silent) {
1148 le_bitvector_to_cpu(errors_silent, (void *) e->errors_silent, sizeof(e->errors_silent) * 8);
1150 prt_printf(out, "Errors to silently fix:");
1152 prt_bitflags_vector(out, bch2_sb_error_strs, errors_silent, sizeof(e->errors_silent) * 8);
1155 kfree(errors_silent);
1159 static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
1160 .validate = bch2_sb_ext_validate,
1161 .to_text = bch2_sb_ext_to_text,
1164 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1166 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1171 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1173 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1175 return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1176 ? bch2_sb_field_ops[type]
1177 : &bch2_sb_field_null_ops;
1180 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1181 struct printbuf *err)
1183 unsigned type = le32_to_cpu(f->type);
1184 struct printbuf field_err = PRINTBUF;
1185 const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1188 ret = ops->validate ? ops->validate(sb, f, &field_err) : 0;
1190 prt_printf(err, "Invalid superblock section %s: %s",
1191 bch2_sb_fields[type], field_err.buf);
1193 bch2_sb_field_to_text(err, sb, f);
1196 printbuf_exit(&field_err);
1200 void __bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1201 struct bch_sb_field *f)
1203 unsigned type = le32_to_cpu(f->type);
1204 const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1206 if (!out->nr_tabstops)
1207 printbuf_tabstop_push(out, 32);
1210 ops->to_text(out, sb, f);
1213 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1214 struct bch_sb_field *f)
1216 unsigned type = le32_to_cpu(f->type);
1218 if (type < BCH_SB_FIELD_NR)
1219 prt_printf(out, "%s", bch2_sb_fields[type]);
1221 prt_printf(out, "(unknown field %u)", type);
1223 prt_printf(out, " (size %zu):", vstruct_bytes(f));
1226 __bch2_sb_field_to_text(out, sb, f);
1229 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1233 prt_printf(out, "Type: %u", l->layout_type);
1236 prt_str(out, "Superblock max size: ");
1237 prt_units_u64(out, 512 << l->sb_max_size_bits);
1240 prt_printf(out, "Nr superblocks: %u", l->nr_superblocks);
1243 prt_str(out, "Offsets: ");
1244 for (i = 0; i < l->nr_superblocks; i++) {
1247 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1252 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1253 bool print_layout, unsigned fields)
1255 u64 fields_have = 0;
1256 unsigned nr_devices = 0;
1258 if (!out->nr_tabstops)
1259 printbuf_tabstop_push(out, 44);
1261 for (int i = 0; i < sb->nr_devices; i++)
1262 nr_devices += bch2_dev_exists(sb, i);
1264 prt_printf(out, "External UUID:");
1266 pr_uuid(out, sb->user_uuid.b);
1269 prt_printf(out, "Internal UUID:");
1271 pr_uuid(out, sb->uuid.b);
1274 prt_printf(out, "Magic number:");
1276 pr_uuid(out, sb->magic.b);
1279 prt_str(out, "Device index:");
1281 prt_printf(out, "%u", sb->dev_idx);
1284 prt_str(out, "Label:");
1286 prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1289 prt_str(out, "Version:");
1291 bch2_version_to_text(out, le16_to_cpu(sb->version));
1294 prt_str(out, "Version upgrade complete:");
1296 bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1299 prt_printf(out, "Oldest version on disk:");
1301 bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1304 prt_printf(out, "Created:");
1306 if (sb->time_base_lo)
1307 bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1309 prt_printf(out, "(not set)");
1312 prt_printf(out, "Sequence number:");
1314 prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1317 prt_printf(out, "Time of last write:");
1319 bch2_prt_datetime(out, le64_to_cpu(sb->write_time));
1322 prt_printf(out, "Superblock size:");
1324 prt_units_u64(out, vstruct_bytes(sb));
1326 prt_units_u64(out, 512ULL << sb->layout.sb_max_size_bits);
1329 prt_printf(out, "Clean:");
1331 prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1334 prt_printf(out, "Devices:");
1336 prt_printf(out, "%u", nr_devices);
1339 prt_printf(out, "Sections:");
1340 vstruct_for_each(sb, f)
1341 fields_have |= 1 << le32_to_cpu(f->type);
1343 prt_bitflags(out, bch2_sb_fields, fields_have);
1346 prt_printf(out, "Features:");
1348 prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1351 prt_printf(out, "Compat features:");
1353 prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1357 prt_printf(out, "Options:");
1359 printbuf_indent_add(out, 2);
1363 for (id = 0; id < bch2_opts_nr; id++) {
1364 const struct bch_option *opt = bch2_opt_table + id;
1366 if (opt->get_sb != BCH2_NO_SB_OPT) {
1367 u64 v = bch2_opt_from_sb(sb, id);
1369 prt_printf(out, "%s:", opt->attr.name);
1371 bch2_opt_to_text(out, NULL, sb, opt, v,
1372 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1378 printbuf_indent_sub(out, 2);
1382 prt_printf(out, "layout:");
1384 printbuf_indent_add(out, 2);
1385 bch2_sb_layout_to_text(out, &sb->layout);
1386 printbuf_indent_sub(out, 2);
1389 vstruct_for_each(sb, f)
1390 if (fields & (1 << le32_to_cpu(f->type))) {
1392 bch2_sb_field_to_text(out, sb, f);