X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fsuper-io.c;h=a2b789b4ac68a344dde31b82ffe1687373f0eecc;hb=6f5afc0c12bbf56ffdabe5b2c5297aef255c4baa;hp=9568cb46755c6c912a2a0966b046e5e0d86d3d24;hpb=ddb58076ef4fe4572ab8537785fc67052f47bf5b;p=bcachefs-tools-debian

diff --git a/libbcachefs/super-io.c b/libbcachefs/super-io.c
index 9568cb4..a2b789b 100644
--- a/libbcachefs/super-io.c
+++ b/libbcachefs/super-io.c
@@ -1,20 +1,29 @@
+// SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
+#include "btree_update_interior.h"
+#include "buckets.h"
 #include "checksum.h"
 #include "disk_groups.h"
 #include "ec.h"
 #include "error.h"
 #include "io.h"
 #include "journal.h"
+#include "journal_io.h"
+#include "journal_sb.h"
+#include "journal_seq_blacklist.h"
 #include "replicas.h"
 #include "quota.h"
 #include "super-io.h"
 #include "super.h"
 #include "vstructs.h"
+#include "counters.h"
 
 #include <linux/backing-dev.h>
 #include <linux/sort.h>
 
+#include <trace/events/bcachefs.h>
+
 const char * const bch2_sb_fields[] = {
 #define x(name, nr)	#name,
 	BCH_SB_FIELDS()
@@ -22,8 +31,8 @@ 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)
@@ -45,10 +54,11 @@ 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);
@@ -94,18 +104,23 @@ void bch2_free_super(struct bch_sb_handle *sb)
 	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) {
@@ -120,14 +135,15 @@ int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 		}
 	}
 
-	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;
 
 	if (sb->have_bio) {
-		bio = bio_kmalloc(GFP_KERNEL, 1 << order);
+		bio = bio_kmalloc(GFP_KERNEL,
+			DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
 		if (!bio)
 			return -ENOMEM;
 
@@ -136,17 +152,12 @@ int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 		sb->bio = bio;
 	}
 
-	new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
+	new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 	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;
 }
@@ -195,22 +206,31 @@ 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";
+	if (uuid_le_cmp(layout->magic, BCACHE_MAGIC)) {
+		pr_buf(out, "Not a bcachefs superblock layout");
+		return -EINVAL;
+	}
 
-	if (layout->layout_type != 0)
-		return "Invalid superblock layout type";
+	if (layout->layout_type != 0) {
+		pr_buf(out, "Invalid superblock layout type %u",
+		       layout->layout_type);
+		return -EINVAL;
+	}
 
-	if (!layout->nr_superblocks)
-		return "Invalid superblock layout: no superblocks";
+	if (!layout->nr_superblocks) {
+		pr_buf(out, "Invalid superblock layout: no superblocks");
+		return -EINVAL;
+	}
 
-	if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
-		return "Invalid superblock layout: too many superblocks";
+	if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
+		pr_buf(out, "Invalid superblock layout: too many superblocks");
+		return -EINVAL;
+	}
 
 	max_sectors = 1 << layout->sb_max_size_bits;
 
@@ -219,126 +239,163 @@ 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) {
+			pr_buf(out, "Invalid superblock layout: superblocks overlap\n"
+			       "  (sb %u ends at %llu next starts at %llu",
+			       i - 1, prev_offset + max_sectors, offset);
+			return -EINVAL;
+		}
 		prev_offset = offset;
 	}
 
-	return NULL;
+	return 0;
 }
 
-const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
+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;
 	u32 version, version_min;
 	u16 block_size;
+	int ret;
 
 	version		= le16_to_cpu(sb->version);
-	version_min	= version >= bcachefs_metadata_version_new_versioning
+	version_min	= version >= bcachefs_metadata_version_bkey_renumber
 		? le16_to_cpu(sb->version_min)
 		: version;
 
-	if (version    >= bcachefs_metadata_version_max ||
-	    version_min < bcachefs_metadata_version_min)
-		return "Unsupported superblock version";
+	if (version    >= bcachefs_metadata_version_max) {
+		pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
+		       version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
+		return -EINVAL;
+	}
 
-	if (version_min > version)
-		return "Bad minimum version";
+	if (version_min < bcachefs_metadata_version_min) {
+		pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
+		       version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
+		return -EINVAL;
+	}
+
+	if (version_min > version) {
+		pr_buf(out, "Bad minimum version %u, greater than version field %u",
+		       version_min, version);
+		return -EINVAL;
+	}
 
 	if (sb->features[1] ||
-	    (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
-		return "Filesystem has incompatible features";
+	    (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
+		pr_buf(out, "Filesystem has incompatible features");
+		return -EINVAL;
+	}
 
 	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) {
+		pr_buf(out, "Block size too big (got %u, max %u)",
+		       block_size, PAGE_SECTORS);
+		return -EINVAL;
+	}
 
-	if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
-		return "Bad user UUID";
+	if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
+		pr_buf(out, "Bad user UUID (got zeroes)");
+		return -EINVAL;
+	}
 
-	if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
-		return "Bad internal UUID";
+	if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
+		pr_buf(out, "Bad intenal UUID (got zeroes)");
+		return -EINVAL;
+	}
 
 	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) {
+		pr_buf(out, "Bad number of member devices %u (max %u)",
+		       sb->nr_devices, BCH_SB_MEMBERS_MAX);
+		return -EINVAL;
+	}
 
-	if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
-		return "Invalid metadata checksum type";
+	if (sb->dev_idx >= sb->nr_devices) {
+		pr_buf(out, "Bad dev_idx (got %u, nr_devices %u)",
+		       sb->dev_idx, sb->nr_devices);
+		return -EINVAL;
+	}
 
-	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) {
+		pr_buf(out, "Invalid time precision: %u (min 1, max %lu)",
+		       le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
+		return -EINVAL;
+	}
 
-	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'r 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_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";
+			pr_buf(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) {
+			pr_buf(out, "Invalid superblock: optional with size 0 (type %u)",
+			       le32_to_cpu(f->type));
+			return -EINVAL;
+		}
 
-		if (vstruct_next(f) > vstruct_last(sb))
-			return "Invalid superblock: invalid optional field";
+		if (vstruct_next(f) > vstruct_last(sb)) {
+			pr_buf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
+			       le32_to_cpu(f->type));
+			return -EINVAL;
+		}
 	}
 
 	/* members must be validated first: */
 	mi = bch2_sb_get_members(sb);
-	if (!mi)
-		return "Invalid superblock: member info area missing";
+	if (!mi) {
+		pr_buf(out, "Invalid superblock: member info area missing");
+		return -EINVAL;
+	}
 
-	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;
 	}
 
-	return NULL;
+	return 0;
 }
 
 /* device open: */
@@ -355,13 +412,19 @@ 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.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]);
 
@@ -369,7 +432,6 @@ static void bch2_sb_update(struct bch_fs *c)
 		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)
 {
 	struct bch_sb_field *src_f, *dst_f;
@@ -395,7 +457,7 @@ static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 	memcpy(dst->compat,	src->compat,	sizeof(dst->compat));
 
 	for (i = 0; i < BCH_SB_FIELD_NR; i++) {
-		if (i == BCH_SB_FIELD_journal)
+		if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 			continue;
 
 		src_f = bch2_sb_field_get(src, i);
@@ -459,52 +521,78 @@ int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 
 /* 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;
+	u32 version, version_min;
 	size_t bytes;
+	int ret;
 reread:
 	bio_reset(sb->bio);
 	bio_set_dev(sb->bio, sb->bdev);
 	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);
+
+	ret = submit_bio_wait(sb->bio);
+	if (ret) {
+		pr_buf(err, "IO error: %i", ret);
+		return ret;
+	}
+
+	if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
+		pr_buf(err, "Not a bcachefs superblock");
+		return -EINVAL;
+	}
 
-	if (submit_bio_wait(sb->bio))
-		return "IO error";
+	version		= le16_to_cpu(sb->sb->version);
+	version_min	= version >= bcachefs_metadata_version_bkey_renumber
+		? le16_to_cpu(sb->sb->version_min)
+		: version;
 
-	if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
-		return "Not a bcachefs superblock";
+	if (version    >= bcachefs_metadata_version_max) {
+		pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
+		       version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
+		return -EINVAL;
+	}
 
-	if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
-	    le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
-		return "Unsupported superblock version";
+	if (version_min < bcachefs_metadata_version_min) {
+		pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
+		       version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
+		return -EINVAL;
+	}
 
 	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) {
+		pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
+		       bytes, 512UL << sb->sb->layout.sb_max_size_bits);
+		return -EINVAL;
+	}
 
-	if (get_order(bytes) > sb->page_order) {
+	if (bytes > sb->buffer_size) {
 		if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
-			return "cannot allocate memory";
+			return -ENOMEM;
 		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) {
+		pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
+		return -EINVAL;
+	}
 
 	/* 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)) {
+		pr_buf(err, "bad checksum");
+		return -EINVAL;
+	}
 
 	sb->seq = le64_to_cpu(sb->sb->seq);
 
-	return NULL;
+	return 0;
 }
 
 int bch2_read_super(const char *path, struct bch_opts *opts,
@@ -512,7 +600,7 @@ 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;
 
@@ -544,25 +632,28 @@ 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) {
+		pr_buf(&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")) {
+		pr_buf(&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
@@ -571,21 +662,22 @@ int bch2_read_super(const char *path, struct bch_opts *opts,
 	bio_reset(sb->bio);
 	bio_set_dev(sb->bio, sb->bdev);
 	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) {
+		pr_buf(&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;
@@ -595,32 +687,41 @@ 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)) {
+		pr_buf(&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 = -EINVAL;
 		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;
 }
 
@@ -632,7 +733,8 @@ 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);
@@ -647,13 +749,12 @@ static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 	bio_reset(bio);
 	bio_set_dev(bio, ca->disk_sb.bdev);
 	bio->bi_iter.bi_sector	= le64_to_cpu(sb->layout.sb_offset[0]);
-	bio->bi_iter.bi_size	= 4096;
 	bio->bi_end_io		= write_super_endio;
 	bio->bi_private		= ca;
 	bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
-	bch2_bio_map(bio, ca->sb_read_scratch);
+	bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 
-	this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_SB],
+	this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 		     bio_sectors(bio));
 
 	percpu_ref_get(&ca->io_ref);
@@ -667,22 +768,21 @@ 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->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);
@@ -693,12 +793,18 @@ 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_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);
@@ -706,20 +812,37 @@ int bch2_write_super(struct bch_fs *c)
 
 	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);
-			ret = -1;
+		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) {
@@ -732,11 +855,24 @@ int bch2_write_super(struct bch_fs *c)
 	closure_sync(cl);
 
 	for_each_online_member(ca, c, i) {
-		if (!ca->sb_write_error &&
-		    ca->disk_sb.seq !=
-		    le64_to_cpu(ca->sb_read_scratch->seq)) {
+		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 modified by another process");
+				"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 = -EROFS;
+			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 = -EROFS;
 			goto out;
@@ -765,15 +901,13 @@ int bch2_write_super(struct bch_fs *c)
 	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
@@ -784,150 +918,258 @@ int bch2_write_super(struct bch_fs *c)
 	 * mount with the devices we did successfully write to:
 	 */
 	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"))
+		"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)
+/* BCH_SB_FIELD_members: */
+
+static int bch2_sb_members_validate(struct bch_sb *sb,
+				    struct bch_sb_field *f,
+				    struct printbuf *err)
 {
-	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;
+	struct bch_sb_field_members *mi = field_to_type(f, members);
 	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";
+	if ((void *) (mi->members + sb->nr_devices) >
+	    vstruct_end(&mi->field)) {
+		pr_buf(err, "too many devices for section size");
+		return -EINVAL;
+	}
 
-	for (i = 0; i < nr; i++)
-		b[i] = le64_to_cpu(journal->buckets[i]);
+	for (i = 0; i < sb->nr_devices; i++) {
+		struct bch_member *m = mi->members + i;
 
-	sort(b, nr, sizeof(u64), u64_cmp, NULL);
+		if (!bch2_member_exists(m))
+			continue;
 
-	err = "journal bucket at sector 0";
-	if (!b[0])
-		goto err;
+		if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
+			pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
+			       i, le64_to_cpu(m->nbuckets), LONG_MAX);
+			return -EINVAL;
+		}
 
-	err = "journal bucket before first bucket";
-	if (m && b[0] < le16_to_cpu(m->first_bucket))
-		goto err;
+		if (le64_to_cpu(m->nbuckets) -
+		    le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
+			pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
+			       i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
+			return -EINVAL;
+		}
 
-	err = "journal bucket past end of device";
-	if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
-		goto err;
+		if (le16_to_cpu(m->bucket_size) <
+		    le16_to_cpu(sb->block_size)) {
+			pr_buf(err, "device %u: bucket size %u smaller than block size %u",
+			       i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
+			return -EINVAL;
+		}
 
-	err = "duplicate journal buckets";
-	for (i = 0; i + 1 < nr; i++)
-		if (b[i] == b[i + 1])
-			goto err;
+		if (le16_to_cpu(m->bucket_size) <
+		    BCH_SB_BTREE_NODE_SIZE(sb)) {
+			pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
+			       i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
+			return -EINVAL;
+		}
+	}
 
-	err = NULL;
-err:
-	kfree(b);
-	return err;
+	return 0;
 }
 
-static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
-	.validate	= bch2_sb_validate_journal,
-};
-
-/* BCH_SB_FIELD_members: */
-
-static const char *bch2_sb_validate_members(struct bch_sb *sb,
-					    struct bch_sb_field *f)
+static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
+				    struct bch_sb_field *f)
 {
 	struct bch_sb_field_members *mi = field_to_type(f, members);
-	struct bch_member *m;
+	struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
+	unsigned i;
 
-	if ((void *) (mi->members + sb->nr_devices) >
-	    vstruct_end(&mi->field))
-		return "Invalid superblock: bad member info";
+	for (i = 0; i < sb->nr_devices; i++) {
+		struct bch_member *m = mi->members + i;
+		unsigned data_have = bch2_sb_dev_has_data(sb, i);
+		u64 bucket_size = le16_to_cpu(m->bucket_size);
+		u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
 
-	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";
+		pr_buf(out, "Device:");
+		pr_tab(out);
+		pr_buf(out, "%u", i);
+		pr_newline(out);
+
+		pr_indent_push(out, 2);
+
+		pr_buf(out, "UUID:");
+		pr_tab(out);
+		pr_uuid(out, m->uuid.b);
+		pr_newline(out);
+
+		pr_buf(out, "Size:");
+		pr_tab(out);
+		pr_units(out, device_size, device_size << 9);
+		pr_newline(out);
+
+		pr_buf(out, "Bucket size:");
+		pr_tab(out);
+		pr_units(out, bucket_size, bucket_size << 9);
+		pr_newline(out);
+
+		pr_buf(out, "First bucket:");
+		pr_tab(out);
+		pr_buf(out, "%u", le16_to_cpu(m->first_bucket));
+		pr_newline(out);
+
+		pr_buf(out, "Buckets:");
+		pr_tab(out);
+		pr_buf(out, "%llu", le64_to_cpu(m->nbuckets));
+		pr_newline(out);
+
+		pr_buf(out, "Last mount:");
+		pr_tab(out);
+		if (m->last_mount)
+			pr_time(out, le64_to_cpu(m->last_mount));
+		else
+			pr_buf(out, "(never)");
+		pr_newline(out);
+
+		pr_buf(out, "State:");
+		pr_tab(out);
+		pr_buf(out, "%s",
+		       BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
+		       ? bch2_member_states[BCH_MEMBER_STATE(m)]
+		       : "unknown");
+		pr_newline(out);
+
+		pr_buf(out, "Group:");
+		pr_tab(out);
+		if (BCH_MEMBER_GROUP(m)) {
+			unsigned idx = BCH_MEMBER_GROUP(m) - 1;
+
+			if (idx < disk_groups_nr(gi))
+				pr_buf(out, "%s (%u)",
+				       gi->entries[idx].label, idx);
+			else
+				pr_buf(out, "(bad disk labels section)");
+		} else {
+			pr_buf(out, "(none)");
+		}
+		pr_newline(out);
 
-		if (le64_to_cpu(m->nbuckets) -
-		    le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
-			return "Not enough buckets";
+		pr_buf(out, "Data allowed:");
+		pr_tab(out);
+		if (BCH_MEMBER_DATA_ALLOWED(m))
+			bch2_flags_to_text(out, bch2_data_types,
+					   BCH_MEMBER_DATA_ALLOWED(m));
+		else
+			pr_buf(out, "(none)");
+		pr_newline(out);
 
-		if (le16_to_cpu(m->bucket_size) <
-		    le16_to_cpu(sb->block_size))
-			return "bucket size smaller than block size";
+		pr_buf(out, "Has data:");
+		pr_tab(out);
+		if (data_have)
+			bch2_flags_to_text(out, bch2_data_types, data_have);
+		else
+			pr_buf(out, "(none)");
+		pr_newline(out);
 
-		if (le16_to_cpu(m->bucket_size) <
-		    BCH_SB_BTREE_NODE_SIZE(sb))
-			return "bucket size smaller than btree node size";
-	}
+		pr_buf(out, "Discard:");
+		pr_tab(out);
+		pr_buf(out, "%llu", BCH_MEMBER_DISCARD(m));
+		pr_newline(out);
 
-	return NULL;
+		pr_buf(out, "Freespace initialized:");
+		pr_tab(out);
+		pr_buf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
+		pr_newline(out);
+
+		pr_indent_pop(out, 2);
+	}
 }
 
 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
-	.validate	= bch2_sb_validate_members,
+	.validate	= bch2_sb_members_validate,
+	.to_text	= bch2_sb_members_to_text,
 };
 
 /* BCH_SB_FIELD_crypt: */
 
-static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
-					  struct bch_sb_field *f)
+static int bch2_sb_crypt_validate(struct bch_sb *sb,
+				  struct bch_sb_field *f,
+				  struct printbuf *err)
 {
 	struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 
-	if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
-		return "invalid field crypt: wrong size";
+	if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
+		pr_buf(err, "wrong size (got %zu should be %zu)",
+		       vstruct_bytes(&crypt->field), sizeof(*crypt));
+		return -EINVAL;
+	}
+
+	if (BCH_CRYPT_KDF_TYPE(crypt)) {
+		pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
+		return -EINVAL;
+	}
 
-	if (BCH_CRYPT_KDF_TYPE(crypt))
-		return "invalid field crypt: bad kdf type";
+	return 0;
+}
 
-	return NULL;
+static void bch2_sb_crypt_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);
+
+	pr_buf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
+	pr_newline(out);
+	pr_buf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
+	pr_newline(out);
+	pr_buf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
+	pr_newline(out);
+	pr_buf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
+	pr_newline(out);
 }
 
 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
-	.validate	= bch2_sb_validate_crypt,
+	.validate	= bch2_sb_crypt_validate,
+	.to_text	= bch2_sb_crypt_to_text,
 };
 
 /* BCH_SB_FIELD_clean: */
 
-void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
+int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
 {
 	struct jset_entry *entry;
+	int ret;
 
 	for (entry = clean->start;
 	     entry < (struct jset_entry *) vstruct_end(&clean->field);
-	     entry = vstruct_next(entry))
-		bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
+	     entry = vstruct_next(entry)) {
+		ret = bch2_journal_entry_validate(c, "superblock", entry,
+						  le16_to_cpu(c->disk_sb.sb->version),
+						  BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
+						  write);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
 }
 
 int bch2_fs_mark_dirty(struct bch_fs *c)
@@ -941,105 +1183,118 @@ int bch2_fs_mark_dirty(struct bch_fs *c)
 
 	mutex_lock(&c->sb_lock);
 	SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
-	c->disk_sb.sb->compat[0] &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
+	c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
+	c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
 	ret = bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
 	return ret;
 }
 
-struct jset_entry *
-bch2_journal_super_entries_add_common(struct bch_fs *c,
-				      struct jset_entry *entry)
+static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
 {
-	struct btree_root *r;
-	unsigned i;
+	struct jset_entry *entry = *end;
+	unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
 
-	mutex_lock(&c->btree_root_lock);
+	memset(entry, 0, u64s * sizeof(u64));
+	/*
+	 * The u64s field counts from the start of data, ignoring the shared
+	 * fields.
+	 */
+	entry->u64s = cpu_to_le16(u64s - 1);
 
-	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);
+	*end = vstruct_next(*end);
+	return entry;
+}
 
-			entry = vstruct_next(entry);
-		}
-	c->btree_roots_dirty = false;
+void bch2_journal_super_entries_add_common(struct bch_fs *c,
+					   struct jset_entry **end,
+					   u64 journal_seq)
+{
+	struct bch_dev *ca;
+	unsigned i, dev;
 
-	mutex_unlock(&c->btree_root_lock);
+	percpu_down_read(&c->mark_lock);
 
-	percpu_down_read_preempt_disable(&c->mark_lock);
+	if (!journal_seq) {
+		for (i = 0; i < ARRAY_SIZE(c->usage); i++)
+			bch2_fs_usage_acc_to_base(c, i);
+	} else {
+		bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
+	}
 
 	{
-		u64 nr_inodes = percpu_u64_get(&c->usage[0]->nr_inodes);
 		struct jset_entry_usage *u =
-			container_of(entry, struct jset_entry_usage, entry);
+			container_of(jset_entry_init(end, sizeof(*u)),
+				     struct jset_entry_usage, entry);
 
-		memset(u, 0, sizeof(*u));
-		u->entry.u64s	= DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 		u->entry.type	= BCH_JSET_ENTRY_usage;
-		u->entry.btree_id = FS_USAGE_INODES;
-		u->v		= cpu_to_le64(nr_inodes);
-
-		entry = vstruct_next(entry);
+		u->entry.btree_id = BCH_FS_USAGE_inodes;
+		u->v		= cpu_to_le64(c->usage_base->nr_inodes);
 	}
 
 	{
 		struct jset_entry_usage *u =
-			container_of(entry, struct jset_entry_usage, entry);
+			container_of(jset_entry_init(end, sizeof(*u)),
+				     struct jset_entry_usage, entry);
 
-		memset(u, 0, sizeof(*u));
-		u->entry.u64s	= DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 		u->entry.type	= BCH_JSET_ENTRY_usage;
-		u->entry.btree_id = FS_USAGE_KEY_VERSION;
+		u->entry.btree_id = BCH_FS_USAGE_key_version;
 		u->v		= cpu_to_le64(atomic64_read(&c->key_version));
-
-		entry = vstruct_next(entry);
 	}
 
 	for (i = 0; i < BCH_REPLICAS_MAX; i++) {
 		struct jset_entry_usage *u =
-			container_of(entry, struct jset_entry_usage, entry);
-		u64 sectors = percpu_u64_get(&c->usage[0]->persistent_reserved[i]);
-
-		if (!sectors)
-			continue;
+			container_of(jset_entry_init(end, sizeof(*u)),
+				     struct jset_entry_usage, entry);
 
-		memset(u, 0, sizeof(*u));
-		u->entry.u64s	= DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 		u->entry.type	= BCH_JSET_ENTRY_usage;
-		u->entry.btree_id = FS_USAGE_RESERVED;
+		u->entry.btree_id = BCH_FS_USAGE_reserved;
 		u->entry.level	= i;
-		u->v		= sectors;
-
-		entry = vstruct_next(entry);
+		u->v		= cpu_to_le64(c->usage_base->persistent_reserved[i]);
 	}
 
 	for (i = 0; i < c->replicas.nr; i++) {
 		struct bch_replicas_entry *e =
 			cpu_replicas_entry(&c->replicas, i);
-		u64 sectors = percpu_u64_get(&c->usage[0]->replicas[i]);
 		struct jset_entry_data_usage *u =
-			container_of(entry, struct jset_entry_data_usage, entry);
+			container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
+				     struct jset_entry_data_usage, entry);
 
-		memset(u, 0, sizeof(*u));
-		u->entry.u64s	= DIV_ROUND_UP(sizeof(*u) + e->nr_devs,
-					       sizeof(u64)) - 1;
 		u->entry.type	= BCH_JSET_ENTRY_data_usage;
-		u->v		= cpu_to_le64(sectors);
+		u->v		= cpu_to_le64(c->usage_base->replicas[i]);
 		memcpy(&u->r, e, replicas_entry_bytes(e));
+	}
 
-		entry = vstruct_next(entry);
+	for_each_member_device(ca, c, dev) {
+		unsigned b = sizeof(struct jset_entry_dev_usage) +
+			sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
+		struct jset_entry_dev_usage *u =
+			container_of(jset_entry_init(end, b),
+				     struct jset_entry_dev_usage, entry);
+
+		u->entry.type = BCH_JSET_ENTRY_dev_usage;
+		u->dev = cpu_to_le32(dev);
+		u->buckets_ec		= cpu_to_le64(ca->usage_base->buckets_ec);
+
+		for (i = 0; i < BCH_DATA_NR; i++) {
+			u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
+			u->d[i].sectors	= cpu_to_le64(ca->usage_base->d[i].sectors);
+			u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
+		}
 	}
 
-	percpu_up_read_preempt_enable(&c->mark_lock);
+	percpu_up_read(&c->mark_lock);
 
-	return entry;
+	for (i = 0; i < 2; i++) {
+		struct jset_entry_clock *clock =
+			container_of(jset_entry_init(end, sizeof(*clock)),
+				     struct jset_entry_clock, entry);
+
+		clock->entry.type = BCH_JSET_ENTRY_clock;
+		clock->rw	= i;
+		clock->time	= cpu_to_le64(atomic64_read(&c->io_clock[i].now));
+	}
 }
 
 void bch2_fs_mark_clean(struct bch_fs *c)
@@ -1047,6 +1302,7 @@ void bch2_fs_mark_clean(struct bch_fs *c)
 	struct bch_sb_field_clean *sb_clean;
 	struct jset_entry *entry;
 	unsigned u64s;
+	int ret;
 
 	mutex_lock(&c->sb_lock);
 	if (BCH_SB_CLEAN(c->disk_sb.sb))
@@ -1054,7 +1310,10 @@ void bch2_fs_mark_clean(struct bch_fs *c)
 
 	SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
 
-	c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
+	c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
+	c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
+	c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
+	c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
 
 	u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
 
@@ -1065,42 +1324,75 @@ void bch2_fs_mark_clean(struct bch_fs *c)
 	}
 
 	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	= cpu_to_le64(journal_cur_seq(&c->journal) - 1);
+	sb_clean->journal_seq	= cpu_to_le64(atomic64_read(&c->journal.seq));
 
 	/* Trying to catch outstanding bug: */
 	BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
 
 	entry = sb_clean->start;
-	entry = bch2_journal_super_entries_add_common(c, entry);
+	bch2_journal_super_entries_add_common(c, &entry, 0);
+	entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
 	BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
 
 	memset(entry, 0,
 	       vstruct_end(&sb_clean->field) - (void *) entry);
 
-	if (le16_to_cpu(c->disk_sb.sb->version) <
-	    bcachefs_metadata_version_bkey_renumber)
-		bch2_sb_clean_renumber(sb_clean, WRITE);
+	/*
+	 * this should be in the write path, and we should be validating every
+	 * superblock section:
+	 */
+	ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
+	if (ret) {
+		bch_err(c, "error writing marking filesystem clean: validate error");
+		goto out;
+	}
 
 	bch2_write_super(c);
 out:
 	mutex_unlock(&c->sb_lock);
 }
 
-static const char *bch2_sb_validate_clean(struct bch_sb *sb,
-					  struct bch_sb_field *f)
+static int bch2_sb_clean_validate(struct bch_sb *sb,
+				  struct bch_sb_field *f,
+				  struct printbuf *err)
 {
 	struct bch_sb_field_clean *clean = field_to_type(f, clean);
 
-	if (vstruct_bytes(&clean->field) < sizeof(*clean))
-		return "invalid field crypt: wrong size";
+	if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
+		pr_buf(err, "wrong size (got %zu should be %zu)",
+		       vstruct_bytes(&clean->field), sizeof(*clean));
+		return -EINVAL;
+	}
 
-	return NULL;
+	return 0;
+}
+
+static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
+				  struct bch_sb_field *f)
+{
+	struct bch_sb_field_clean *clean = field_to_type(f, clean);
+	struct jset_entry *entry;
+
+	pr_buf(out, "flags:          %x",	le32_to_cpu(clean->flags));
+	pr_newline(out);
+	pr_buf(out, "journal_seq:    %llu",	le64_to_cpu(clean->journal_seq));
+	pr_newline(out);
+
+	for (entry = clean->start;
+	     entry != vstruct_end(&clean->field);
+	     entry = vstruct_next(entry)) {
+		if (entry->type == BCH_JSET_ENTRY_btree_keys &&
+		    !entry->u64s)
+			continue;
+
+		bch2_journal_entry_to_text(out, NULL, entry);
+		pr_newline(out);
+	}
 }
 
 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
-	.validate	= bch2_sb_validate_clean,
+	.validate	= bch2_sb_clean_validate,
+	.to_text	= bch2_sb_clean_to_text,
 };
 
 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
@@ -1110,14 +1402,27 @@ static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
 #undef x
 };
 
-static const char *bch2_sb_field_validate(struct bch_sb *sb,
-					  struct bch_sb_field *f)
+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;
+	int ret;
 
-	return type < BCH_SB_FIELD_NR
-		? bch2_sb_field_ops[type]->validate(sb, f)
-		: NULL;
+	if (type >= BCH_SB_FIELD_NR)
+		return 0;
+
+	ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
+	if (ret) {
+		pr_buf(err, "Invalid superblock section %s: %s",
+		       bch2_sb_fields[type],
+		       field_err.buf);
+		pr_newline(err);
+		bch2_sb_field_to_text(err, sb, f);
+	}
+
+	printbuf_exit(&field_err);
+	return ret;
 }
 
 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
@@ -1127,13 +1432,183 @@ void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
 	const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
 		? bch2_sb_field_ops[type] : NULL;
 
+	if (!out->tabstops[0])
+		out->tabstops[0] = 32;
+
 	if (ops)
 		pr_buf(out, "%s", bch2_sb_fields[type]);
 	else
 		pr_buf(out, "(unknown field %u)", type);
 
-	pr_buf(out, " (size %llu):", vstruct_bytes(f));
+	pr_buf(out, " (size %zu):", vstruct_bytes(f));
+	pr_newline(out);
 
-	if (ops && ops->to_text)
+	if (ops && ops->to_text) {
+		pr_indent_push(out, 2);
 		bch2_sb_field_ops[type]->to_text(out, sb, f);
+		pr_indent_pop(out, 2);
+	}
+}
+
+void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
+{
+	unsigned i;
+
+	pr_buf(out, "Type:                    %u", l->layout_type);
+	pr_newline(out);
+
+	pr_buf(out, "Superblock max size:     ");
+	pr_units(out,
+		 1 << l->sb_max_size_bits,
+		 512 << l->sb_max_size_bits);
+	pr_newline(out);
+
+	pr_buf(out, "Nr superblocks:          %u", l->nr_superblocks);
+	pr_newline(out);
+
+	pr_buf(out, "Offsets:                 ");
+	for (i = 0; i < l->nr_superblocks; i++) {
+		if (i)
+			pr_buf(out, ", ");
+		pr_buf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
+	}
+	pr_newline(out);
+}
+
+void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
+		     bool print_layout, unsigned fields)
+{
+	struct bch_sb_field_members *mi;
+	struct bch_sb_field *f;
+	u64 fields_have = 0;
+	unsigned nr_devices = 0;
+
+	if (!out->tabstops[0])
+		out->tabstops[0] = 32;
+
+	mi = bch2_sb_get_members(sb);
+	if (mi) {
+		struct bch_member *m;
+
+		for (m = mi->members;
+		     m < mi->members + sb->nr_devices;
+		     m++)
+			nr_devices += bch2_member_exists(m);
+	}
+
+	pr_buf(out, "External UUID:");
+	pr_tab(out);
+	pr_uuid(out, sb->user_uuid.b);
+	pr_newline(out);
+
+	pr_buf(out, "Internal UUID:");
+	pr_tab(out);
+	pr_uuid(out, sb->uuid.b);
+	pr_newline(out);
+
+	pr_buf(out, "Device index:");
+	pr_tab(out);
+	pr_buf(out, "%u", sb->dev_idx);
+	pr_newline(out);
+
+	pr_buf(out, "Label:");
+	pr_tab(out);
+	pr_buf(out, "%.*s", (int) sizeof(sb->label), sb->label);
+	pr_newline(out);
+
+	pr_buf(out, "Version:");
+	pr_tab(out);
+	pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
+	pr_newline(out);
+
+	pr_buf(out, "Oldest version on disk:");
+	pr_tab(out);
+	pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
+	pr_newline(out);
+
+	pr_buf(out, "Created:");
+	pr_tab(out);
+	if (sb->time_base_lo)
+		pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
+	else
+		pr_buf(out, "(not set)");
+	pr_newline(out);
+
+	pr_buf(out, "Sequence number:");
+	pr_tab(out);
+	pr_buf(out, "%llu", le64_to_cpu(sb->seq));
+	pr_newline(out);
+
+	pr_buf(out, "Superblock size:");
+	pr_tab(out);
+	pr_buf(out, "%zu", vstruct_bytes(sb));
+	pr_newline(out);
+
+	pr_buf(out, "Clean:");
+	pr_tab(out);
+	pr_buf(out, "%llu", BCH_SB_CLEAN(sb));
+	pr_newline(out);
+
+	pr_buf(out, "Devices:");
+	pr_tab(out);
+	pr_buf(out, "%u", nr_devices);
+	pr_newline(out);
+
+	pr_buf(out, "Sections:");
+	vstruct_for_each(sb, f)
+		fields_have |= 1 << le32_to_cpu(f->type);
+	pr_tab(out);
+	bch2_flags_to_text(out, bch2_sb_fields, fields_have);
+	pr_newline(out);
+
+	pr_buf(out, "Features:");
+	pr_tab(out);
+	bch2_flags_to_text(out, bch2_sb_features,
+			   le64_to_cpu(sb->features[0]));
+	pr_newline(out);
+
+	pr_buf(out, "Compat features:");
+	pr_tab(out);
+	bch2_flags_to_text(out, bch2_sb_compat,
+			   le64_to_cpu(sb->compat[0]));
+	pr_newline(out);
+
+	pr_newline(out);
+	pr_buf(out, "Options:");
+	pr_newline(out);
+	pr_indent_push(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);
+
+				pr_buf(out, "%s:", opt->attr.name);
+				pr_tab(out);
+				bch2_opt_to_text(out, NULL, sb, opt, v,
+						 OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
+				pr_newline(out);
+			}
+		}
+	}
+
+	pr_indent_pop(out, 2);
+
+	if (print_layout) {
+		pr_newline(out);
+		pr_buf(out, "layout:");
+		pr_newline(out);
+		pr_indent_push(out, 2);
+		bch2_sb_layout_to_text(out, &sb->layout);
+		pr_indent_pop(out, 2);
+	}
+
+	vstruct_for_each(sb, f)
+		if (fields & (1 << le32_to_cpu(f->type))) {
+			pr_newline(out);
+			bch2_sb_field_to_text(out, sb, f);
+		}
 }