X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fsuper-io.c;h=b6021b734bf02c1cc3e770b8717315274778874a;hb=54b6beabf05b6cf62092f98f0c06395e4242b064;hp=54de9fac6e2283e63487a691f18523ba89522f24;hpb=17e2f2775be6e10b966cd958bc0461aab662571a;p=bcachefs-tools-debian

diff --git a/libbcachefs/super-io.c b/libbcachefs/super-io.c
index 54de9fa..b6021b7 100644
--- a/libbcachefs/super-io.c
+++ b/libbcachefs/super-io.c
@@ -1,19 +1,89 @@
+// SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
 #include "checksum.h"
+#include "counters.h"
 #include "disk_groups.h"
+#include "ec.h"
 #include "error.h"
 #include "io.h"
 #include "journal.h"
+#include "journal_sb.h"
+#include "journal_seq_blacklist.h"
+#include "recovery.h"
 #include "replicas.h"
 #include "quota.h"
+#include "sb-clean.h"
+#include "sb-members.h"
 #include "super-io.h"
 #include "super.h"
+#include "trace.h"
 #include "vstructs.h"
 
 #include <linux/backing-dev.h>
 #include <linux/sort.h>
 
+struct bch2_metadata_version {
+	u16		version;
+	const char	*name;
+	u64		recovery_passes;
+};
+
+static const struct bch2_metadata_version bch2_metadata_versions[] = {
+#define x(n, v, _recovery_passes) {		\
+	.version = v,				\
+	.name = #n,				\
+	.recovery_passes = _recovery_passes,	\
+},
+	BCH_METADATA_VERSIONS()
+#undef x
+};
+
+void bch2_version_to_text(struct printbuf *out, unsigned v)
+{
+	const char *str = "(unknown version)";
+
+	for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
+		if (bch2_metadata_versions[i].version == v) {
+			str = bch2_metadata_versions[i].name;
+			break;
+		}
+
+	prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
+}
+
+unsigned bch2_latest_compatible_version(unsigned v)
+{
+	if (!BCH_VERSION_MAJOR(v))
+		return v;
+
+	for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
+		if (bch2_metadata_versions[i].version > v &&
+		    BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
+		    BCH_VERSION_MAJOR(v))
+			v = bch2_metadata_versions[i].version;
+
+	return v;
+}
+
+u64 bch2_upgrade_recovery_passes(struct bch_fs *c,
+				 unsigned old_version,
+				 unsigned new_version)
+{
+	u64 ret = 0;
+
+	for (const struct bch2_metadata_version *i = bch2_metadata_versions;
+	     i < bch2_metadata_versions + ARRAY_SIZE(bch2_metadata_versions);
+	     i++)
+		if (i->version > old_version && i->version <= new_version) {
+			if (i->recovery_passes & RECOVERY_PASS_ALL_FSCK)
+				ret |= bch2_fsck_recovery_passes();
+			ret |= i->recovery_passes;
+		}
+
+	return ret &= ~RECOVERY_PASS_ALL_FSCK;
+}
+
 const char * const bch2_sb_fields[] = {
 #define x(name, nr)	#name,
 	BCH_SB_FIELDS()
@@ -21,8 +91,8 @@ const char * const bch2_sb_fields[] = {
 	NULL
 };
 
-static const char *bch2_sb_field_validate(struct bch_sb *,
-					  struct bch_sb_field *);
+static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
+				  struct printbuf *);
 
 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
 				      enum bch_sb_field_type type)
@@ -44,10 +114,11 @@ static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
 	unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 	unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
 
-	BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
-	       sb->page_order);
+	BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
 
-	if (!f) {
+	if (!f && !u64s) {
+		/* nothing to do: */
+	} else if (!f) {
 		f = vstruct_last(sb->sb);
 		memset(f, 0, sizeof(u64) * u64s);
 		f->u64s = cpu_to_le32(u64s);
@@ -56,8 +127,13 @@ static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
 		void *src, *dst;
 
 		src = vstruct_end(f);
-		f->u64s = cpu_to_le32(u64s);
-		dst = vstruct_end(f);
+
+		if (u64s) {
+			f->u64s = cpu_to_le32(u64s);
+			dst = vstruct_end(f);
+		} else {
+			dst = f;
+		}
 
 		memmove(dst, src, vstruct_end(sb->sb) - src);
 
@@ -67,71 +143,81 @@ static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
 
 	sb->sb->u64s = cpu_to_le32(sb_u64s);
 
-	return f;
+	return u64s ? f : NULL;
+}
+
+void bch2_sb_field_delete(struct bch_sb_handle *sb,
+			  enum bch_sb_field_type type)
+{
+	struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
+
+	if (f)
+		__bch2_sb_field_resize(sb, f, 0);
 }
 
 /* Superblock realloc/free: */
 
 void bch2_free_super(struct bch_sb_handle *sb)
 {
-	if (sb->bio)
-		bio_put(sb->bio);
+	kfree(sb->bio);
 	if (!IS_ERR_OR_NULL(sb->bdev))
 		blkdev_put(sb->bdev, sb->mode);
 
-	free_pages((unsigned long) sb->sb, sb->page_order);
+	kfree(sb->sb);
 	memset(sb, 0, sizeof(*sb));
 }
 
 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 {
 	size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
-	unsigned order = get_order(new_bytes);
+	size_t new_buffer_size;
 	struct bch_sb *new_sb;
 	struct bio *bio;
 
-	if (sb->sb && sb->page_order >= order)
+	if (sb->bdev)
+		new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
+
+	new_buffer_size = roundup_pow_of_two(new_bytes);
+
+	if (sb->sb && sb->buffer_size >= new_buffer_size)
 		return 0;
 
 	if (sb->have_layout) {
 		u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 
 		if (new_bytes > max_bytes) {
-			char buf[BDEVNAME_SIZE];
-
-			pr_err("%s: superblock too big: want %zu but have %llu",
-			       bdevname(sb->bdev, buf), new_bytes, max_bytes);
-			return -ENOSPC;
+			pr_err("%pg: superblock too big: want %zu but have %llu",
+			       sb->bdev, new_bytes, max_bytes);
+			return -BCH_ERR_ENOSPC_sb;
 		}
 	}
 
-	if (sb->page_order >= order && sb->sb)
+	if (sb->buffer_size >= new_buffer_size && sb->sb)
 		return 0;
 
 	if (dynamic_fault("bcachefs:add:super_realloc"))
-		return -ENOMEM;
+		return -BCH_ERR_ENOMEM_sb_realloc_injected;
+
+	new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
+	if (!new_sb)
+		return -BCH_ERR_ENOMEM_sb_buf_realloc;
+
+	sb->sb = new_sb;
 
 	if (sb->have_bio) {
-		bio = bio_kmalloc(GFP_KERNEL, 1 << order);
+		unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);
+
+		bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 		if (!bio)
-			return -ENOMEM;
+			return -BCH_ERR_ENOMEM_sb_bio_realloc;
 
-		if (sb->bio)
-			bio_put(sb->bio);
+		bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
+
+		kfree(sb->bio);
 		sb->bio = bio;
 	}
 
-	new_sb = (void *) __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
-	if (!new_sb)
-		return -ENOMEM;
-
-	if (sb->sb)
-		memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
-
-	free_pages((unsigned long) sb->sb, sb->page_order);
-	sb->sb = new_sb;
-
-	sb->page_order = order;
+	sb->buffer_size = new_buffer_size;
 
 	return 0;
 }
@@ -166,34 +252,43 @@ struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 		}
 	}
 
+	f = bch2_sb_field_get(sb->sb, type);
 	f = __bch2_sb_field_resize(sb, f, u64s);
-	f->type = cpu_to_le32(type);
+	if (f)
+		f->type = cpu_to_le32(type);
 	return f;
 }
 
 /* Superblock validate: */
 
-static inline void __bch2_sb_layout_size_assert(void)
-{
-	BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
-}
-
-static const char *validate_sb_layout(struct bch_sb_layout *layout)
+static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 {
 	u64 offset, prev_offset, max_sectors;
 	unsigned i;
 
-	if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
-		return "Not a bcachefs superblock layout";
+	BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 
-	if (layout->layout_type != 0)
-		return "Invalid superblock layout type";
+	if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
+	    !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
+		prt_printf(out, "Not a bcachefs superblock layout");
+		return -BCH_ERR_invalid_sb_layout;
+	}
 
-	if (!layout->nr_superblocks)
-		return "Invalid superblock layout: no superblocks";
+	if (layout->layout_type != 0) {
+		prt_printf(out, "Invalid superblock layout type %u",
+		       layout->layout_type);
+		return -BCH_ERR_invalid_sb_layout_type;
+	}
 
-	if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
-		return "Invalid superblock layout: too many superblocks";
+	if (!layout->nr_superblocks) {
+		prt_printf(out, "Invalid superblock layout: no superblocks");
+		return -BCH_ERR_invalid_sb_layout_nr_superblocks;
+	}
+
+	if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
+		prt_printf(out, "Invalid superblock layout: too many superblocks");
+		return -BCH_ERR_invalid_sb_layout_nr_superblocks;
+	}
 
 	max_sectors = 1 << layout->sb_max_size_bits;
 
@@ -202,125 +297,184 @@ static const char *validate_sb_layout(struct bch_sb_layout *layout)
 	for (i = 1; i < layout->nr_superblocks; i++) {
 		offset = le64_to_cpu(layout->sb_offset[i]);
 
-		if (offset < prev_offset + max_sectors)
-			return "Invalid superblock layout: superblocks overlap";
+		if (offset < prev_offset + max_sectors) {
+			prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
+			       "  (sb %u ends at %llu next starts at %llu",
+			       i - 1, prev_offset + max_sectors, offset);
+			return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
+		}
 		prev_offset = offset;
 	}
 
-	return NULL;
+	return 0;
 }
 
-const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
+static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
+{
+	u16 version		= le16_to_cpu(sb->version);
+	u16 version_min		= le16_to_cpu(sb->version_min);
+
+	if (!bch2_version_compatible(version)) {
+		prt_str(out, "Unsupported superblock version ");
+		bch2_version_to_text(out, version);
+		prt_str(out, " (min ");
+		bch2_version_to_text(out, bcachefs_metadata_version_min);
+		prt_str(out, ", max ");
+		bch2_version_to_text(out, bcachefs_metadata_version_current);
+		prt_str(out, ")");
+		return -BCH_ERR_invalid_sb_version;
+	}
+
+	if (!bch2_version_compatible(version_min)) {
+		prt_str(out, "Unsupported superblock version_min ");
+		bch2_version_to_text(out, version_min);
+		prt_str(out, " (min ");
+		bch2_version_to_text(out, bcachefs_metadata_version_min);
+		prt_str(out, ", max ");
+		bch2_version_to_text(out, bcachefs_metadata_version_current);
+		prt_str(out, ")");
+		return -BCH_ERR_invalid_sb_version;
+	}
+
+	if (version_min > version) {
+		prt_str(out, "Bad minimum version ");
+		bch2_version_to_text(out, version_min);
+		prt_str(out, ", greater than version field ");
+		bch2_version_to_text(out, version);
+		return -BCH_ERR_invalid_sb_version;
+	}
+
+	return 0;
+}
+
+static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
+			    int rw)
 {
 	struct bch_sb *sb = disk_sb->sb;
 	struct bch_sb_field *f;
 	struct bch_sb_field_members *mi;
-	const char *err;
+	enum bch_opt_id opt_id;
 	u16 block_size;
+	int ret;
 
-	if (le64_to_cpu(sb->version) < BCH_SB_VERSION_MIN ||
-	    le64_to_cpu(sb->version) > BCH_SB_VERSION_MAX)
-		return"Unsupported superblock version";
+	ret = bch2_sb_compatible(sb, out);
+	if (ret)
+		return ret;
 
-	if (le64_to_cpu(sb->version) < BCH_SB_VERSION_EXTENT_MAX) {
-		SET_BCH_SB_ENCODED_EXTENT_MAX_BITS(sb, 7);
-		SET_BCH_SB_POSIX_ACL(sb, 1);
+	if (sb->features[1] ||
+	    (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
+		prt_printf(out, "Filesystem has incompatible features");
+		return -BCH_ERR_invalid_sb_features;
 	}
 
 	block_size = le16_to_cpu(sb->block_size);
 
-	if (!is_power_of_2(block_size) ||
-	    block_size > PAGE_SECTORS)
-		return "Bad block size";
+	if (block_size > PAGE_SECTORS) {
+		prt_printf(out, "Block size too big (got %u, max %u)",
+		       block_size, PAGE_SECTORS);
+		return -BCH_ERR_invalid_sb_block_size;
+	}
 
-	if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
-		return "Bad user UUID";
+	if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
+		prt_printf(out, "Bad user UUID (got zeroes)");
+		return -BCH_ERR_invalid_sb_uuid;
+	}
 
-	if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
-		return "Bad internal UUID";
+	if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
+		prt_printf(out, "Bad intenal UUID (got zeroes)");
+		return -BCH_ERR_invalid_sb_uuid;
+	}
 
 	if (!sb->nr_devices ||
-	    sb->nr_devices <= sb->dev_idx ||
-	    sb->nr_devices > BCH_SB_MEMBERS_MAX)
-		return "Bad number of member devices";
-
-	if (!BCH_SB_META_REPLICAS_WANT(sb) ||
-	    BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
-		return "Invalid number of metadata replicas";
-
-	if (!BCH_SB_META_REPLICAS_REQ(sb) ||
-	    BCH_SB_META_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
-		return "Invalid number of metadata replicas";
-
-	if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
-	    BCH_SB_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
-		return "Invalid number of data replicas";
-
-	if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
-	    BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
-		return "Invalid number of data replicas";
+	    sb->nr_devices > BCH_SB_MEMBERS_MAX) {
+		prt_printf(out, "Bad number of member devices %u (max %u)",
+		       sb->nr_devices, BCH_SB_MEMBERS_MAX);
+		return -BCH_ERR_invalid_sb_too_many_members;
+	}
 
-	if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
-		return "Invalid metadata checksum type";
+	if (sb->dev_idx >= sb->nr_devices) {
+		prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
+		       sb->dev_idx, sb->nr_devices);
+		return -BCH_ERR_invalid_sb_dev_idx;
+	}
 
-	if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
-		return "Invalid metadata checksum type";
+	if (!sb->time_precision ||
+	    le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
+		prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
+		       le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
+		return -BCH_ERR_invalid_sb_time_precision;
+	}
 
-	if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
-		return "Invalid compression type";
+	if (rw == READ) {
+		/*
+		 * Been seeing a bug where these are getting inexplicably
+		 * zeroed, so we're now validating them, but we have to be
+		 * careful not to preven people's filesystems from mounting:
+		 */
+		if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
+			SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
+		if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
+			SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
+
+		if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
+			SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
+	}
 
-	if (!BCH_SB_BTREE_NODE_SIZE(sb))
-		return "Btree node size not set";
+	for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
+		const struct bch_option *opt = bch2_opt_table + opt_id;
 
-	if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
-		return "Btree node size not a power of two";
+		if (opt->get_sb != BCH2_NO_SB_OPT) {
+			u64 v = bch2_opt_from_sb(sb, opt_id);
 
-	if (BCH_SB_GC_RESERVE(sb) < 5)
-		return "gc reserve percentage too small";
+			prt_printf(out, "Invalid option ");
+			ret = bch2_opt_validate(opt, v, out);
+			if (ret)
+				return ret;
 
-	if (!sb->time_precision ||
-	    le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
-		return "invalid time precision";
+			printbuf_reset(out);
+		}
+	}
 
 	/* validate layout */
-	err = validate_sb_layout(&sb->layout);
-	if (err)
-		return err;
+	ret = validate_sb_layout(&sb->layout, out);
+	if (ret)
+		return ret;
 
 	vstruct_for_each(sb, f) {
-		if (!f->u64s)
-			return "Invalid superblock: invalid optional field";
+		if (!f->u64s) {
+			prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
+			       le32_to_cpu(f->type));
+			return -BCH_ERR_invalid_sb_field_size;
+		}
 
-		if (vstruct_next(f) > vstruct_last(sb))
-			return "Invalid superblock: invalid optional field";
+		if (vstruct_next(f) > vstruct_last(sb)) {
+			prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
+			       le32_to_cpu(f->type));
+			return -BCH_ERR_invalid_sb_field_size;
+		}
 	}
 
 	/* members must be validated first: */
 	mi = bch2_sb_get_members(sb);
-	if (!mi)
-		return "Invalid superblock: member info area missing";
+	if (!mi) {
+		prt_printf(out, "Invalid superblock: member info area missing");
+		return -BCH_ERR_invalid_sb_members_missing;
+	}
 
-	err = bch2_sb_field_validate(sb, &mi->field);
-	if (err)
-		return err;
+	ret = bch2_sb_field_validate(sb, &mi->field, out);
+	if (ret)
+		return ret;
 
 	vstruct_for_each(sb, f) {
 		if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 			continue;
 
-		err = bch2_sb_field_validate(sb, f);
-		if (err)
-			return err;
+		ret = bch2_sb_field_validate(sb, f, out);
+		if (ret)
+			return ret;
 	}
 
-	if (le64_to_cpu(sb->version) < BCH_SB_VERSION_EXTENT_NONCE_V1 &&
-	    bch2_sb_get_crypt(sb) &&
-	    BCH_SB_INITIALIZED(sb))
-		return "Incompatible extent nonces";
-
-	sb->version = cpu_to_le64(BCH_SB_VERSION_MAX);
-
-	return NULL;
+	return 0;
 }
 
 /* device open: */
@@ -336,26 +490,36 @@ static void bch2_sb_update(struct bch_fs *c)
 
 	c->sb.uuid		= src->uuid;
 	c->sb.user_uuid		= src->user_uuid;
+	c->sb.version		= le16_to_cpu(src->version);
+	c->sb.version_min	= le16_to_cpu(src->version_min);
+	c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
 	c->sb.nr_devices	= src->nr_devices;
 	c->sb.clean		= BCH_SB_CLEAN(src);
 	c->sb.encryption_type	= BCH_SB_ENCRYPTION_TYPE(src);
-	c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
-	c->sb.time_base_lo	= le64_to_cpu(src->time_base_lo);
+
+	c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
+	c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
+
+	/* XXX this is wrong, we need a 96 or 128 bit integer type */
+	c->sb.time_base_lo	= div_u64(le64_to_cpu(src->time_base_lo),
+					  c->sb.nsec_per_time_unit);
 	c->sb.time_base_hi	= le32_to_cpu(src->time_base_hi);
-	c->sb.time_precision	= le32_to_cpu(src->time_precision);
+
 	c->sb.features		= le64_to_cpu(src->features[0]);
+	c->sb.compat		= le64_to_cpu(src->compat[0]);
 
 	for_each_member_device(ca, c, i)
 		ca->mi = bch2_mi_to_cpu(mi->members + i);
 }
 
-/* doesn't copy member info */
-static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
+static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 {
 	struct bch_sb_field *src_f, *dst_f;
 	struct bch_sb *dst = dst_handle->sb;
+	unsigned i;
 
 	dst->version		= src->version;
+	dst->version_min	= src->version_min;
 	dst->seq		= src->seq;
 	dst->uuid		= src->uuid;
 	dst->user_uuid		= src->user_uuid;
@@ -372,41 +536,48 @@ static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 	memcpy(dst->features,	src->features,	sizeof(dst->features));
 	memcpy(dst->compat,	src->compat,	sizeof(dst->compat));
 
-	vstruct_for_each(src, src_f) {
-		if (src_f->type == BCH_SB_FIELD_journal)
+	for (i = 0; i < BCH_SB_FIELD_NR; i++) {
+		int d;
+
+		if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 			continue;
 
-		dst_f = bch2_sb_field_get(dst, le32_to_cpu(src_f->type));
+		src_f = bch2_sb_field_get(src, i);
+		dst_f = bch2_sb_field_get(dst, i);
+
+		d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
+		    (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
+		if (d > 0) {
+			int ret = bch2_sb_realloc(dst_handle,
+					le32_to_cpu(dst_handle->sb->u64s) + d);
+
+			if (ret)
+				return ret;
+
+			dst = dst_handle->sb;
+			dst_f = bch2_sb_field_get(dst, i);
+		}
+
 		dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
-					       le32_to_cpu(src_f->u64s));
+				src_f ? le32_to_cpu(src_f->u64s) : 0);
 
-		memcpy(dst_f, src_f, vstruct_bytes(src_f));
+		if (src_f)
+			memcpy(dst_f, src_f, vstruct_bytes(src_f));
 	}
+
+	return 0;
 }
 
 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 {
-	struct bch_sb_field_journal *journal_buckets =
-		bch2_sb_get_journal(src);
-	unsigned journal_u64s = journal_buckets
-		? le32_to_cpu(journal_buckets->field.u64s)
-		: 0;
 	int ret;
 
 	lockdep_assert_held(&c->sb_lock);
 
-	ret = bch2_sb_realloc(&c->disk_sb,
-			      le32_to_cpu(src->u64s) - journal_u64s);
-	if (ret)
-		return ret;
-
-	__copy_super(&c->disk_sb, src);
-
-	ret = bch2_sb_replicas_to_cpu_replicas(c);
-	if (ret)
-		return ret;
-
-	ret = bch2_sb_disk_groups_to_cpu(c);
+	ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
+		__copy_super(&c->disk_sb, src) ?:
+		bch2_sb_replicas_to_cpu_replicas(c) ?:
+		bch2_sb_disk_groups_to_cpu(c);
 	if (ret)
 		return ret;
 
@@ -416,69 +587,69 @@ int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 
 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 {
-	struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
-	struct bch_sb_field_journal *journal_buckets =
-		bch2_sb_get_journal(dst);
-	unsigned journal_u64s = journal_buckets
-		? le32_to_cpu(journal_buckets->field.u64s)
-		: 0;
-	unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
-	int ret;
-
-	ret = bch2_sb_realloc(&ca->disk_sb, u64s);
-	if (ret)
-		return ret;
-
-	__copy_super(&ca->disk_sb, src);
-	return 0;
+	return __copy_super(&ca->disk_sb, c->disk_sb.sb);
 }
 
 /* read superblock: */
 
-static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
+static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 {
 	struct bch_csum csum;
 	size_t bytes;
+	int ret;
 reread:
-	bio_reset(sb->bio);
-	bio_set_dev(sb->bio, sb->bdev);
+	bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 	sb->bio->bi_iter.bi_sector = offset;
-	sb->bio->bi_iter.bi_size = PAGE_SIZE << sb->page_order;
-	bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
-	bch2_bio_map(sb->bio, sb->sb);
+	bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 
-	if (submit_bio_wait(sb->bio))
-		return "IO error";
+	ret = submit_bio_wait(sb->bio);
+	if (ret) {
+		prt_printf(err, "IO error: %i", ret);
+		return ret;
+	}
 
-	if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
-		return "Not a bcachefs superblock";
+	if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
+	    !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
+		prt_printf(err, "Not a bcachefs superblock");
+		return -BCH_ERR_invalid_sb_magic;
+	}
 
-	if (le64_to_cpu(sb->sb->version) < BCH_SB_VERSION_MIN ||
-	    le64_to_cpu(sb->sb->version) > BCH_SB_VERSION_MAX)
-		return"Unsupported superblock version";
+	ret = bch2_sb_compatible(sb->sb, err);
+	if (ret)
+		return ret;
 
 	bytes = vstruct_bytes(sb->sb);
 
-	if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
-		return "Bad superblock: too big";
+	if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
+		prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
+		       bytes, 512UL << sb->sb->layout.sb_max_size_bits);
+		return -BCH_ERR_invalid_sb_too_big;
+	}
 
-	if (get_order(bytes) > sb->page_order) {
-		if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
-			return "cannot allocate memory";
+	if (bytes > sb->buffer_size) {
+		ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
+		if (ret)
+			return ret;
 		goto reread;
 	}
 
-	if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
-		return "unknown csum type";
+	if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
+		prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
+		return -BCH_ERR_invalid_sb_csum_type;
+	}
 
 	/* XXX: verify MACs */
 	csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 			    null_nonce(), sb->sb);
 
-	if (bch2_crc_cmp(csum, sb->sb->csum))
-		return "bad checksum reading superblock";
+	if (bch2_crc_cmp(csum, sb->sb->csum)) {
+		prt_printf(err, "bad checksum");
+		return -BCH_ERR_invalid_sb_csum;
+	}
+
+	sb->seq = le64_to_cpu(sb->sb->seq);
 
-	return NULL;
+	return 0;
 }
 
 int bch2_read_super(const char *path, struct bch_opts *opts,
@@ -486,16 +657,21 @@ int bch2_read_super(const char *path, struct bch_opts *opts,
 {
 	u64 offset = opt_get(*opts, sb);
 	struct bch_sb_layout layout;
-	const char *err;
+	struct printbuf err = PRINTBUF;
 	__le64 *i;
 	int ret;
-
-	pr_verbose_init(*opts, "");
-
+#ifndef __KERNEL__
+retry:
+#endif
 	memset(sb, 0, sizeof(*sb));
 	sb->mode	= FMODE_READ;
 	sb->have_bio	= true;
 
+#ifndef __KERNEL__
+	if (opt_get(*opts, direct_io) == false)
+		sb->mode |= FMODE_BUFFERED;
+#endif
+
 	if (!opt_get(*opts, noexcl))
 		sb->mode |= FMODE_EXCL;
 
@@ -518,48 +694,50 @@ int bch2_read_super(const char *path, struct bch_opts *opts,
 		goto out;
 	}
 
-	err = "cannot allocate memory";
 	ret = bch2_sb_realloc(sb, 0);
-	if (ret)
+	if (ret) {
+		prt_printf(&err, "error allocating memory for superblock");
 		goto err;
+	}
 
-	ret = -EFAULT;
-	err = "dynamic fault";
-	if (bch2_fs_init_fault("read_super"))
+	if (bch2_fs_init_fault("read_super")) {
+		prt_printf(&err, "dynamic fault");
+		ret = -EFAULT;
 		goto err;
+	}
 
-	ret = -EINVAL;
-	err = read_one_super(sb, offset);
-	if (!err)
+	ret = read_one_super(sb, offset, &err);
+	if (!ret)
 		goto got_super;
 
 	if (opt_defined(*opts, sb))
 		goto err;
 
-	pr_err("error reading default superblock: %s", err);
+	printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
+	       path, err.buf);
+	printbuf_reset(&err);
 
 	/*
 	 * Error reading primary superblock - read location of backup
 	 * superblocks:
 	 */
-	bio_reset(sb->bio);
-	bio_set_dev(sb->bio, sb->bdev);
+	bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 	sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
-	sb->bio->bi_iter.bi_size = sizeof(struct bch_sb_layout);
-	bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 	/*
 	 * use sb buffer to read layout, since sb buffer is page aligned but
 	 * layout won't be:
 	 */
-	bch2_bio_map(sb->bio, sb->sb);
+	bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 
-	err = "IO error";
-	if (submit_bio_wait(sb->bio))
+	ret = submit_bio_wait(sb->bio);
+	if (ret) {
+		prt_printf(&err, "IO error: %i", ret);
 		goto err;
+	}
 
 	memcpy(&layout, sb->sb, sizeof(layout));
-	err = validate_sb_layout(&layout);
-	if (err)
+	ret = validate_sb_layout(&layout, &err);
+	if (ret)
 		goto err;
 
 	for (i = layout.sb_offset;
@@ -569,32 +747,46 @@ int bch2_read_super(const char *path, struct bch_opts *opts,
 		if (offset == opt_get(*opts, sb))
 			continue;
 
-		err = read_one_super(sb, offset);
-		if (!err)
+		ret = read_one_super(sb, offset, &err);
+		if (!ret)
 			goto got_super;
 	}
 
-	ret = -EINVAL;
 	goto err;
 
 got_super:
-	err = "Superblock block size smaller than device block size";
-	ret = -EINVAL;
 	if (le16_to_cpu(sb->sb->block_size) << 9 <
-	    bdev_logical_block_size(sb->bdev))
+	    bdev_logical_block_size(sb->bdev) &&
+	    opt_get(*opts, direct_io)) {
+#ifndef __KERNEL__
+		opt_set(*opts, direct_io, false);
+		bch2_free_super(sb);
+		goto retry;
+#endif
+		prt_printf(&err, "block size (%u) smaller than device block size (%u)",
+		       le16_to_cpu(sb->sb->block_size) << 9,
+		       bdev_logical_block_size(sb->bdev));
+		ret = -BCH_ERR_block_size_too_small;
 		goto err;
+	}
 
-	if (sb->mode & FMODE_WRITE)
-		bdev_get_queue(sb->bdev)->backing_dev_info->capabilities
-			|= BDI_CAP_STABLE_WRITES;
 	ret = 0;
 	sb->have_layout = true;
+
+	ret = bch2_sb_validate(sb, &err, READ);
+	if (ret) {
+		printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
+		       path, err.buf);
+		goto err_no_print;
+	}
 out:
-	pr_verbose_init(*opts, "ret %i", ret);
+	printbuf_exit(&err);
 	return ret;
 err:
+	printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
+	       path, err.buf);
+err_no_print:
 	bch2_free_super(sb);
-	pr_err("error reading superblock: %s", err);
 	goto out;
 }
 
@@ -606,13 +798,32 @@ static void write_super_endio(struct bio *bio)
 
 	/* XXX: return errors directly */
 
-	if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write"))
+	if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
+			       bch2_blk_status_to_str(bio->bi_status)))
 		ca->sb_write_error = 1;
 
 	closure_put(&ca->fs->sb_write);
 	percpu_ref_put(&ca->io_ref);
 }
 
+static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
+{
+	struct bch_sb *sb = ca->disk_sb.sb;
+	struct bio *bio = ca->disk_sb.bio;
+
+	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
+	bio->bi_iter.bi_sector	= le64_to_cpu(sb->layout.sb_offset[0]);
+	bio->bi_end_io		= write_super_endio;
+	bio->bi_private		= ca;
+	bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
+
+	this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
+		     bio_sectors(bio));
+
+	percpu_ref_get(&ca->io_ref);
+	closure_bio_submit(bio, &c->sb_write);
+}
+
 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 {
 	struct bch_sb *sb = ca->disk_sb.sb;
@@ -620,57 +831,83 @@ static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 
 	sb->offset = sb->layout.sb_offset[idx];
 
-	SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
+	SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 	sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 				null_nonce(), sb);
 
-	bio_reset(bio);
-	bio_set_dev(bio, ca->disk_sb.bdev);
+	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 	bio->bi_iter.bi_sector	= le64_to_cpu(sb->offset);
-	bio->bi_iter.bi_size	=
-		roundup((size_t) vstruct_bytes(sb),
-			bdev_logical_block_size(ca->disk_sb.bdev));
 	bio->bi_end_io		= write_super_endio;
 	bio->bi_private		= ca;
-	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
-	bch2_bio_map(bio, sb);
+	bch2_bio_map(bio, sb,
+		     roundup((size_t) vstruct_bytes(sb),
+			     bdev_logical_block_size(ca->disk_sb.bdev)));
 
-	this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_SB],
+	this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 		     bio_sectors(bio));
 
 	percpu_ref_get(&ca->io_ref);
 	closure_bio_submit(bio, &c->sb_write);
 }
 
-void bch2_write_super(struct bch_fs *c)
+int bch2_write_super(struct bch_fs *c)
 {
 	struct closure *cl = &c->sb_write;
 	struct bch_dev *ca;
+	struct printbuf err = PRINTBUF;
 	unsigned i, sb = 0, nr_wrote;
-	const char *err;
 	struct bch_devs_mask sb_written;
 	bool wrote, can_mount_without_written, can_mount_with_written;
+	unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
+	int ret = 0;
+
+	trace_and_count(c, write_super, c, _RET_IP_);
+
+	if (c->opts.very_degraded)
+		degraded_flags |= BCH_FORCE_IF_LOST;
 
 	lockdep_assert_held(&c->sb_lock);
 
 	closure_init_stack(cl);
 	memset(&sb_written, 0, sizeof(sb_written));
 
+	/* Make sure we're using the new magic numbers: */
+	c->disk_sb.sb->magic = BCHFS_MAGIC;
+	c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
+
 	le64_add_cpu(&c->disk_sb.sb->seq, 1);
 
+	if (test_bit(BCH_FS_ERROR, &c->flags))
+		SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
+	if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
+		SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
+
+	SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
+
+	bch2_sb_counters_from_cpu(c);
+
 	for_each_online_member(ca, c, i)
 		bch2_sb_from_fs(c, ca);
 
 	for_each_online_member(ca, c, i) {
-		err = bch2_sb_validate(&ca->disk_sb);
-		if (err) {
-			bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
+		printbuf_reset(&err);
+
+		ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
+		if (ret) {
+			bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
+			percpu_ref_put(&ca->io_ref);
 			goto out;
 		}
 	}
 
-	if (c->opts.nochanges ||
-	    test_bit(BCH_FS_ERROR, &c->flags))
+	if (c->opts.nochanges)
+		goto out;
+
+	/*
+	 * Defer writing the superblock until filesystem initialization is
+	 * complete - don't write out a partly initialized superblock:
+	 */
+	if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 		goto out;
 
 	for_each_online_member(ca, c, i) {
@@ -678,10 +915,40 @@ void bch2_write_super(struct bch_fs *c)
 		ca->sb_write_error = 0;
 	}
 
+	for_each_online_member(ca, c, i)
+		read_back_super(c, ca);
+	closure_sync(cl);
+
+	for_each_online_member(ca, c, i) {
+		if (ca->sb_write_error)
+			continue;
+
+		if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
+			bch2_fs_fatal_error(c,
+				"Superblock write was silently dropped! (seq %llu expected %llu)",
+				le64_to_cpu(ca->sb_read_scratch->seq),
+				ca->disk_sb.seq);
+			percpu_ref_put(&ca->io_ref);
+			ret = -BCH_ERR_erofs_sb_err;
+			goto out;
+		}
+
+		if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
+			bch2_fs_fatal_error(c,
+				"Superblock modified by another process (seq %llu expected %llu)",
+				le64_to_cpu(ca->sb_read_scratch->seq),
+				ca->disk_sb.seq);
+			percpu_ref_put(&ca->io_ref);
+			ret = -BCH_ERR_erofs_sb_err;
+			goto out;
+		}
+	}
+
 	do {
 		wrote = false;
 		for_each_online_member(ca, c, i)
-			if (sb < ca->disk_sb.sb->layout.nr_superblocks) {
+			if (!ca->sb_write_error &&
+			    sb < ca->disk_sb.sb->layout.nr_superblocks) {
 				write_one_super(c, ca, sb);
 				wrote = true;
 			}
@@ -689,22 +956,23 @@ void bch2_write_super(struct bch_fs *c)
 		sb++;
 	} while (wrote);
 
-	for_each_online_member(ca, c, i)
+	for_each_online_member(ca, c, i) {
 		if (ca->sb_write_error)
 			__clear_bit(ca->dev_idx, sb_written.d);
+		else
+			ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
+	}
 
 	nr_wrote = dev_mask_nr(&sb_written);
 
 	can_mount_with_written =
-		bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
-				      BCH_FORCE_IF_DEGRADED);
+		bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 
 	for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 		sb_written.d[i] = ~sb_written.d[i];
 
 	can_mount_without_written =
-		bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
-				      BCH_FORCE_IF_DEGRADED);
+		bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 
 	/*
 	 * If we would be able to mount _without_ the devices we successfully
@@ -714,255 +982,277 @@ void bch2_write_super(struct bch_fs *c)
 	 * written anything (new filesystem), we continue if we'd be able to
 	 * mount with the devices we did successfully write to:
 	 */
-	bch2_fs_fatal_err_on(!nr_wrote ||
-			     (can_mount_without_written &&
-			      !can_mount_with_written), c,
-		"Unable to write superblock to sufficient devices");
+	if (bch2_fs_fatal_err_on(!nr_wrote ||
+				 !can_mount_with_written ||
+				 (can_mount_without_written &&
+				  !can_mount_with_written), c,
+		"Unable to write superblock to sufficient devices (from %ps)",
+		(void *) _RET_IP_))
+		ret = -1;
 out:
 	/* Make new options visible after they're persistent: */
 	bch2_sb_update(c);
+	printbuf_exit(&err);
+	return ret;
 }
 
-/* BCH_SB_FIELD_journal: */
-
-static int u64_cmp(const void *_l, const void *_r)
+void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 {
-	u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
+	mutex_lock(&c->sb_lock);
+	if (!(c->sb.features & (1ULL << feat))) {
+		c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 
-	return l < r ? -1 : l > r ? 1 : 0;
+		bch2_write_super(c);
+	}
+	mutex_unlock(&c->sb_lock);
 }
 
-static const char *bch2_sb_validate_journal(struct bch_sb *sb,
-					    struct bch_sb_field *f)
+/* Downgrade if superblock is at a higher version than currently supported: */
+void bch2_sb_maybe_downgrade(struct bch_fs *c)
 {
-	struct bch_sb_field_journal *journal = field_to_type(f, journal);
-	struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
-	const char *err;
-	unsigned nr;
-	unsigned i;
-	u64 *b;
-
-	journal = bch2_sb_get_journal(sb);
-	if (!journal)
-		return NULL;
-
-	nr = bch2_nr_journal_buckets(journal);
-	if (!nr)
-		return NULL;
-
-	b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
-	if (!b)
-		return "cannot allocate memory";
-
-	for (i = 0; i < nr; i++)
-		b[i] = le64_to_cpu(journal->buckets[i]);
-
-	sort(b, nr, sizeof(u64), u64_cmp, NULL);
-
-	err = "journal bucket at sector 0";
-	if (!b[0])
-		goto err;
-
-	err = "journal bucket before first bucket";
-	if (m && b[0] < le16_to_cpu(m->first_bucket))
-		goto err;
+	lockdep_assert_held(&c->sb_lock);
 
-	err = "journal bucket past end of device";
-	if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
-		goto err;
+	/*
+	 * Downgrade, if superblock is at a higher version than currently
+	 * supported:
+	 */
+	if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
+		SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
+	if (c->sb.version > bcachefs_metadata_version_current)
+		c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
+	if (c->sb.version_min > bcachefs_metadata_version_current)
+		c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
+	c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
+}
 
-	err = "duplicate journal buckets";
-	for (i = 0; i + 1 < nr; i++)
-		if (b[i] == b[i + 1])
-			goto err;
+void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
+{
+	lockdep_assert_held(&c->sb_lock);
 
-	err = NULL;
-err:
-	kfree(b);
-	return err;
+	c->disk_sb.sb->version = cpu_to_le16(new_version);
+	c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
 }
 
-static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
-	.validate	= bch2_sb_validate_journal,
+static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
+#define x(f, nr)					\
+	[BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
+	BCH_SB_FIELDS()
+#undef x
 };
 
-/* BCH_SB_FIELD_members: */
+static const struct bch_sb_field_ops bch2_sb_field_null_ops;
 
-static const char *bch2_sb_validate_members(struct bch_sb *sb,
-					    struct bch_sb_field *f)
+static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
 {
-	struct bch_sb_field_members *mi = field_to_type(f, members);
-	struct bch_member *m;
-
-	if ((void *) (mi->members + sb->nr_devices) >
-	    vstruct_end(&mi->field))
-		return "Invalid superblock: bad member info";
-
-	for (m = mi->members;
-	     m < mi->members + sb->nr_devices;
-	     m++) {
-		if (!bch2_member_exists(m))
-			continue;
-
-		if (le64_to_cpu(m->nbuckets) > LONG_MAX)
-			return "Too many buckets";
-
-		if (le64_to_cpu(m->nbuckets) -
-		    le16_to_cpu(m->first_bucket) < 1 << 10)
-			return "Not enough buckets";
+	return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
+		? bch2_sb_field_ops[type]
+		: &bch2_sb_field_null_ops;
+}
 
-		if (le16_to_cpu(m->bucket_size) <
-		    le16_to_cpu(sb->block_size))
-			return "bucket size smaller than block size";
+static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
+				  struct printbuf *err)
+{
+	unsigned type = le32_to_cpu(f->type);
+	struct printbuf field_err = PRINTBUF;
+	const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
+	int ret;
 
-		if (le16_to_cpu(m->bucket_size) <
-		    BCH_SB_BTREE_NODE_SIZE(sb))
-			return "bucket size smaller than btree node size";
+	ret = ops->validate ? ops->validate(sb, f, &field_err) : 0;
+	if (ret) {
+		prt_printf(err, "Invalid superblock section %s: %s",
+			   bch2_sb_fields[type], field_err.buf);
+		prt_newline(err);
+		bch2_sb_field_to_text(err, sb, f);
 	}
 
-	if (le64_to_cpu(sb->version) < BCH_SB_VERSION_EXTENT_MAX)
-		for (m = mi->members;
-		     m < mi->members + sb->nr_devices;
-		     m++)
-			SET_BCH_MEMBER_DATA_ALLOWED(m, ~0);
-
-	return NULL;
+	printbuf_exit(&field_err);
+	return ret;
 }
 
-static const struct bch_sb_field_ops bch_sb_field_ops_members = {
-	.validate	= bch2_sb_validate_members,
-};
-
-/* BCH_SB_FIELD_crypt: */
-
-static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
-					  struct bch_sb_field *f)
+void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
+			   struct bch_sb_field *f)
 {
-	struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
-
-	if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
-		return "invalid field crypt: wrong size";
+	unsigned type = le32_to_cpu(f->type);
+	const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
 
-	if (BCH_CRYPT_KDF_TYPE(crypt))
-		return "invalid field crypt: bad kdf type";
+	if (!out->nr_tabstops)
+		printbuf_tabstop_push(out, 32);
 
-	return NULL;
-}
+	if (type < BCH_SB_FIELD_NR)
+		prt_printf(out, "%s", bch2_sb_fields[type]);
+	else
+		prt_printf(out, "(unknown field %u)", type);
 
-static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
-	.validate	= bch2_sb_validate_crypt,
-};
+	prt_printf(out, " (size %zu):", vstruct_bytes(f));
+	prt_newline(out);
 
-/* BCH_SB_FIELD_clean: */
+	if (ops->to_text) {
+		printbuf_indent_add(out, 2);
+		ops->to_text(out, sb, f);
+		printbuf_indent_sub(out, 2);
+	}
+}
 
-void bch2_fs_mark_clean(struct bch_fs *c, bool clean)
+void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
 {
-	struct bch_sb_field_clean *sb_clean;
-	unsigned u64s = sizeof(*sb_clean) / sizeof(u64);
-	struct jset_entry *entry;
-	struct btree_root *r;
-
-	mutex_lock(&c->sb_lock);
-	if (clean == BCH_SB_CLEAN(c->disk_sb.sb))
-		goto out;
-
-	SET_BCH_SB_CLEAN(c->disk_sb.sb, clean);
+	unsigned i;
 
-	if (!clean)
-		goto write_super;
+	prt_printf(out, "Type:                    %u", l->layout_type);
+	prt_newline(out);
 
-	mutex_lock(&c->btree_root_lock);
+	prt_str(out, "Superblock max size:     ");
+	prt_units_u64(out, 512 << l->sb_max_size_bits);
+	prt_newline(out);
 
-	for (r = c->btree_roots;
-	     r < c->btree_roots + BTREE_ID_NR;
-	     r++)
-		if (r->alive)
-			u64s += jset_u64s(r->key.u64s);
+	prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
+	prt_newline(out);
 
-	sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
-	if (!sb_clean) {
-		bch_err(c, "error resizing superblock while setting filesystem clean");
-		goto out;
+	prt_str(out, "Offsets:                 ");
+	for (i = 0; i < l->nr_superblocks; i++) {
+		if (i)
+			prt_str(out, ", ");
+		prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
 	}
-
-	sb_clean->flags		= 0;
-	sb_clean->read_clock	= cpu_to_le16(c->bucket_clock[READ].hand);
-	sb_clean->write_clock	= cpu_to_le16(c->bucket_clock[WRITE].hand);
-	sb_clean->journal_seq	= journal_cur_seq(&c->journal) - 1;
-
-	entry = sb_clean->start;
-	memset(entry, 0,
-	       vstruct_end(&sb_clean->field) - (void *) entry);
-
-	for (r = c->btree_roots;
-	     r < c->btree_roots + BTREE_ID_NR;
-	     r++)
-		if (r->alive) {
-			entry->u64s	= r->key.u64s;
-			entry->btree_id	= r - c->btree_roots;
-			entry->level	= r->level;
-			entry->type	= BCH_JSET_ENTRY_btree_root;
-			bkey_copy(&entry->start[0], &r->key);
-			entry = vstruct_next(entry);
-			BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
-		}
-
-	BUG_ON(entry != vstruct_end(&sb_clean->field));
-
-	mutex_unlock(&c->btree_root_lock);
-write_super:
-	bch2_write_super(c);
-out:
-	mutex_unlock(&c->sb_lock);
+	prt_newline(out);
 }
 
-static const char *bch2_sb_validate_clean(struct bch_sb *sb,
-					  struct bch_sb_field *f)
+void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
+		     bool print_layout, unsigned fields)
 {
-	struct bch_sb_field_clean *clean = field_to_type(f, clean);
-
-	if (vstruct_bytes(&clean->field) < sizeof(*clean))
-		return "invalid field crypt: wrong size";
+	struct bch_sb_field_members *mi;
+	struct bch_sb_field *f;
+	u64 fields_have = 0;
+	unsigned nr_devices = 0;
 
-	return NULL;
-}
+	if (!out->nr_tabstops)
+		printbuf_tabstop_push(out, 44);
 
-static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
-	.validate	= bch2_sb_validate_clean,
-};
+	mi = bch2_sb_get_members(sb);
+	if (mi) {
+		struct bch_member *m;
 
-static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
-#define x(f, nr)					\
-	[BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
-	BCH_SB_FIELDS()
-#undef x
-};
+		for (m = mi->members;
+		     m < mi->members + sb->nr_devices;
+		     m++)
+			nr_devices += bch2_member_exists(m);
+	}
 
-static const char *bch2_sb_field_validate(struct bch_sb *sb,
-					  struct bch_sb_field *f)
-{
-	unsigned type = le32_to_cpu(f->type);
+	prt_printf(out, "External UUID:");
+	prt_tab(out);
+	pr_uuid(out, sb->user_uuid.b);
+	prt_newline(out);
+
+	prt_printf(out, "Internal UUID:");
+	prt_tab(out);
+	pr_uuid(out, sb->uuid.b);
+	prt_newline(out);
+
+	prt_str(out, "Device index:");
+	prt_tab(out);
+	prt_printf(out, "%u", sb->dev_idx);
+	prt_newline(out);
+
+	prt_str(out, "Label:");
+	prt_tab(out);
+	prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
+	prt_newline(out);
+
+	prt_str(out, "Version:");
+	prt_tab(out);
+	bch2_version_to_text(out, le16_to_cpu(sb->version));
+	prt_newline(out);
+
+	prt_str(out, "Version upgrade complete:");
+	prt_tab(out);
+	bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
+	prt_newline(out);
+
+	prt_printf(out, "Oldest version on disk:");
+	prt_tab(out);
+	bch2_version_to_text(out, le16_to_cpu(sb->version_min));
+	prt_newline(out);
+
+	prt_printf(out, "Created:");
+	prt_tab(out);
+	if (sb->time_base_lo)
+		pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
+	else
+		prt_printf(out, "(not set)");
+	prt_newline(out);
+
+	prt_printf(out, "Sequence number:");
+	prt_tab(out);
+	prt_printf(out, "%llu", le64_to_cpu(sb->seq));
+	prt_newline(out);
+
+	prt_printf(out, "Superblock size:");
+	prt_tab(out);
+	prt_printf(out, "%zu", vstruct_bytes(sb));
+	prt_newline(out);
+
+	prt_printf(out, "Clean:");
+	prt_tab(out);
+	prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
+	prt_newline(out);
+
+	prt_printf(out, "Devices:");
+	prt_tab(out);
+	prt_printf(out, "%u", nr_devices);
+	prt_newline(out);
+
+	prt_printf(out, "Sections:");
+	vstruct_for_each(sb, f)
+		fields_have |= 1 << le32_to_cpu(f->type);
+	prt_tab(out);
+	prt_bitflags(out, bch2_sb_fields, fields_have);
+	prt_newline(out);
+
+	prt_printf(out, "Features:");
+	prt_tab(out);
+	prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
+	prt_newline(out);
+
+	prt_printf(out, "Compat features:");
+	prt_tab(out);
+	prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
+	prt_newline(out);
+
+	prt_newline(out);
+	prt_printf(out, "Options:");
+	prt_newline(out);
+	printbuf_indent_add(out, 2);
+	{
+		enum bch_opt_id id;
+
+		for (id = 0; id < bch2_opts_nr; id++) {
+			const struct bch_option *opt = bch2_opt_table + id;
+
+			if (opt->get_sb != BCH2_NO_SB_OPT) {
+				u64 v = bch2_opt_from_sb(sb, id);
+
+				prt_printf(out, "%s:", opt->attr.name);
+				prt_tab(out);
+				bch2_opt_to_text(out, NULL, sb, opt, v,
+						 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
+				prt_newline(out);
+			}
+		}
+	}
 
-	return type < BCH_SB_FIELD_NR
-		? bch2_sb_field_ops[type]->validate(sb, f)
-		: NULL;
-}
+	printbuf_indent_sub(out, 2);
 
-size_t bch2_sb_field_to_text(char *buf, size_t size,
-			     struct bch_sb *sb, struct bch_sb_field *f)
-{
-	unsigned type = le32_to_cpu(f->type);
-	size_t (*to_text)(char *, size_t, struct bch_sb *,
-				   struct bch_sb_field *) =
-		type < BCH_SB_FIELD_NR
-		? bch2_sb_field_ops[type]->to_text
-		: NULL;
-
-	if (!to_text) {
-		if (size)
-			buf[0] = '\0';
-		return 0;
+	if (print_layout) {
+		prt_newline(out);
+		prt_printf(out, "layout:");
+		prt_newline(out);
+		printbuf_indent_add(out, 2);
+		bch2_sb_layout_to_text(out, &sb->layout);
+		printbuf_indent_sub(out, 2);
 	}
 
-	return to_text(buf, size, sb, f);
+	vstruct_for_each(sb, f)
+		if (fields & (1 << le32_to_cpu(f->type))) {
+			prt_newline(out);
+			bch2_sb_field_to_text(out, sb, f);
+		}
 }