X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fjournal_io.c;h=3c3a6c4dc6372f8f8ebc50168e97083237426bfa;hb=b6740e5392a0e0a5fddc5ad4ffac0567078e114a;hp=45b1b839783de66f712aaac1b337cf522e365d84;hpb=ec28e6bfa269daf953af95610c0d1af3e7df1c0a;p=bcachefs-tools-debian

diff --git a/libbcachefs/journal_io.c b/libbcachefs/journal_io.c
index 45b1b83..3c3a6c4 100644
--- a/libbcachefs/journal_io.c
+++ b/libbcachefs/journal_io.c
@@ -4,18 +4,18 @@
 #include "alloc_foreground.h"
 #include "btree_io.h"
 #include "btree_update_interior.h"
+#include "btree_write_buffer.h"
 #include "buckets.h"
 #include "checksum.h"
 #include "disk_groups.h"
 #include "error.h"
-#include "io.h"
 #include "journal.h"
 #include "journal_io.h"
 #include "journal_reclaim.h"
 #include "journal_seq_blacklist.h"
 #include "replicas.h"
-
-#include <trace/events/bcachefs.h>
+#include "sb-clean.h"
+#include "trace.h"
 
 static struct nonce journal_nonce(const struct jset *jset)
 {
@@ -141,7 +141,8 @@ static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
 		if (!dup->csum_good)
 			goto replace;
 
-		fsck_err(c, "found duplicate but non identical journal entries (seq %llu)",
+		fsck_err(c, journal_entry_replicas_data_mismatch,
+			 "found duplicate but non identical journal entries (seq %llu)",
 			 le64_to_cpu(j->seq));
 		i = dup;
 		goto found;
@@ -209,38 +210,47 @@ static void journal_entry_null_range(void *start, void *end)
 #define JOURNAL_ENTRY_BAD	7
 
 static void journal_entry_err_msg(struct printbuf *out,
+				  u32 version,
 				  struct jset *jset,
 				  struct jset_entry *entry)
 {
-	prt_str(out, "invalid journal entry ");
-	if (entry)
-		prt_printf(out, "%s ", bch2_jset_entry_types[entry->type]);
-
-	if (!jset)
-		prt_printf(out, "in superblock");
-	else if (!entry)
-		prt_printf(out, "at seq %llu", le64_to_cpu(jset->seq));
-	else
-		prt_printf(out, "at offset %zi/%u seq %llu",
-			   (u64 *) entry - jset->_data,
-			   le32_to_cpu(jset->u64s),
-			   le64_to_cpu(jset->seq));
+	prt_str(out, "invalid journal entry, version=");
+	bch2_version_to_text(out, version);
+
+	if (entry) {
+		prt_str(out, " type=");
+		prt_str(out, bch2_jset_entry_types[entry->type]);
+	}
+
+	if (!jset) {
+		prt_printf(out, " in superblock");
+	} else {
+
+		prt_printf(out, " seq=%llu", le64_to_cpu(jset->seq));
+
+		if (entry)
+			prt_printf(out, " offset=%zi/%u",
+				   (u64 *) entry - jset->_data,
+				   le32_to_cpu(jset->u64s));
+	}
+
 	prt_str(out, ": ");
 }
 
-#define journal_entry_err(c, jset, entry, msg, ...)			\
+#define journal_entry_err(c, version, jset, entry, _err, msg, ...)	\
 ({									\
-	struct printbuf buf = PRINTBUF;					\
+	struct printbuf _buf = PRINTBUF;				\
 									\
-	journal_entry_err_msg(&buf, jset, entry);			\
-	prt_printf(&buf, msg, ##__VA_ARGS__);				\
+	journal_entry_err_msg(&_buf, version, jset, entry);		\
+	prt_printf(&_buf, msg, ##__VA_ARGS__);				\
 									\
-	switch (write) {						\
+	switch (flags & BKEY_INVALID_WRITE) {				\
 	case READ:							\
-		mustfix_fsck_err(c, "%s", buf.buf);			\
+		mustfix_fsck_err(c, _err, "%s", _buf.buf);		\
 		break;							\
 	case WRITE:							\
-		bch_err(c, "corrupt metadata before write: %s\n", buf.buf);\
+		bch2_sb_error_count(c, BCH_FSCK_ERR_##_err);		\
+		bch_err(c, "corrupt metadata before write: %s\n", _buf.buf);\
 		if (bch2_fs_inconsistent(c)) {				\
 			ret = -BCH_ERR_fsck_errors_not_fixed;		\
 			goto fsck_err;					\
@@ -248,12 +258,12 @@ static void journal_entry_err_msg(struct printbuf *out,
 		break;							\
 	}								\
 									\
-	printbuf_exit(&buf);						\
+	printbuf_exit(&_buf);						\
 	true;								\
 })
 
-#define journal_entry_err_on(cond, c, jset, entry, msg, ...)		\
-	((cond) ? journal_entry_err(c, jset, entry, msg, ##__VA_ARGS__) : false)
+#define journal_entry_err_on(cond, ...)					\
+	((cond) ? journal_entry_err(__VA_ARGS__) : false)
 
 #define FSCK_DELETED_KEY	5
 
@@ -262,13 +272,18 @@ static int journal_validate_key(struct bch_fs *c,
 				struct jset_entry *entry,
 				unsigned level, enum btree_id btree_id,
 				struct bkey_i *k,
-				unsigned version, int big_endian, int write)
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
+	int write = flags & BKEY_INVALID_WRITE;
 	void *next = vstruct_next(entry);
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
-	if (journal_entry_err_on(!k->k.u64s, c, jset, entry, "k->u64s 0")) {
+	if (journal_entry_err_on(!k->k.u64s,
+				 c, version, jset, entry,
+				 journal_entry_bkey_u64s_0,
+				 "k->u64s 0")) {
 		entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
 		journal_entry_null_range(vstruct_next(entry), next);
 		return FSCK_DELETED_KEY;
@@ -276,7 +291,8 @@ static int journal_validate_key(struct bch_fs *c,
 
 	if (journal_entry_err_on((void *) bkey_next(k) >
 				 (void *) vstruct_next(entry),
-				 c, jset, entry,
+				 c, version, jset, entry,
+				 journal_entry_bkey_past_end,
 				 "extends past end of journal entry")) {
 		entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
 		journal_entry_null_range(vstruct_next(entry), next);
@@ -284,7 +300,8 @@ static int journal_validate_key(struct bch_fs *c,
 	}
 
 	if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT,
-				 c, jset, entry,
+				 c, version, jset, entry,
+				 journal_entry_bkey_bad_format,
 				 "bad format %u", k->k.format)) {
 		le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
 		memmove(k, bkey_next(k), next - (void *) bkey_next(k));
@@ -299,11 +316,7 @@ static int journal_validate_key(struct bch_fs *c,
 	if (bch2_bkey_invalid(c, bkey_i_to_s_c(k),
 			      __btree_node_type(level, btree_id), write, &buf)) {
 		printbuf_reset(&buf);
-		prt_printf(&buf, "invalid journal entry %s at offset %zi/%u seq %llu:",
-			   bch2_jset_entry_types[entry->type],
-			   (u64 *) entry - jset->_data,
-			   le32_to_cpu(jset->u64s),
-			   le64_to_cpu(jset->seq));
+		journal_entry_err_msg(&buf, version, jset, entry);
 		prt_newline(&buf);
 		printbuf_indent_add(&buf, 2);
 
@@ -312,7 +325,8 @@ static int journal_validate_key(struct bch_fs *c,
 		bch2_bkey_invalid(c, bkey_i_to_s_c(k),
 				  __btree_node_type(level, btree_id), write, &buf);
 
-		mustfix_fsck_err(c, "%s", buf.buf);
+		mustfix_fsck_err(c, journal_entry_bkey_invalid,
+				 "%s", buf.buf);
 
 		le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
 		memmove(k, bkey_next(k), next - (void *) bkey_next(k));
@@ -331,9 +345,10 @@ fsck_err:
 }
 
 static int journal_entry_btree_keys_validate(struct bch_fs *c,
-					     struct jset *jset,
-					     struct jset_entry *entry,
-					     unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	struct bkey_i *k = entry->start;
 
@@ -341,7 +356,8 @@ static int journal_entry_btree_keys_validate(struct bch_fs *c,
 		int ret = journal_validate_key(c, jset, entry,
 					       entry->level,
 					       entry->btree_id,
-					       k, version, big_endian, write|BKEY_INVALID_FROM_JOURNAL);
+					       k, version, big_endian,
+					       flags|BKEY_INVALID_JOURNAL);
 		if (ret == FSCK_DELETED_KEY)
 			continue;
 
@@ -362,23 +378,25 @@ static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs
 			prt_newline(out);
 			prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
 		}
-		prt_printf(out, "btree=%s l=%u ", bch2_btree_ids[entry->btree_id], entry->level);
+		prt_printf(out, "btree=%s l=%u ", bch2_btree_id_str(entry->btree_id), entry->level);
 		bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
 		first = false;
 	}
 }
 
 static int journal_entry_btree_root_validate(struct bch_fs *c,
-					     struct jset *jset,
-					     struct jset_entry *entry,
-					     unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	struct bkey_i *k = entry->start;
 	int ret = 0;
 
 	if (journal_entry_err_on(!entry->u64s ||
 				 le16_to_cpu(entry->u64s) != k->k.u64s,
-				 c, jset, entry,
+				 c, version, jset, entry,
+				 journal_entry_btree_root_bad_size,
 				 "invalid btree root journal entry: wrong number of keys")) {
 		void *next = vstruct_next(entry);
 		/*
@@ -392,7 +410,7 @@ static int journal_entry_btree_root_validate(struct bch_fs *c,
 	}
 
 	return journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
-				    version, big_endian, write);
+				    version, big_endian, flags);
 fsck_err:
 	return ret;
 }
@@ -404,9 +422,10 @@ static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs
 }
 
 static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
-					    struct jset *jset,
-					    struct jset_entry *entry,
-					    unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	/* obsolete, don't care: */
 	return 0;
@@ -418,14 +437,16 @@ static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs
 }
 
 static int journal_entry_blacklist_validate(struct bch_fs *c,
-					    struct jset *jset,
-					    struct jset_entry *entry,
-					    unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	int ret = 0;
 
 	if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1,
-				 c, jset, entry,
+				 c, version, jset, entry,
+				 journal_entry_blacklist_bad_size,
 		"invalid journal seq blacklist entry: bad size")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 	}
@@ -443,15 +464,17 @@ static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs
 }
 
 static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
-					       struct jset *jset,
-					       struct jset_entry *entry,
-					       unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	struct jset_entry_blacklist_v2 *bl_entry;
 	int ret = 0;
 
 	if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2,
-				 c, jset, entry,
+				 c, version, jset, entry,
+				 journal_entry_blacklist_v2_bad_size,
 		"invalid journal seq blacklist entry: bad size")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 		goto out;
@@ -461,7 +484,8 @@ static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
 
 	if (journal_entry_err_on(le64_to_cpu(bl_entry->start) >
 				 le64_to_cpu(bl_entry->end),
-				 c, jset, entry,
+				 c, version, jset, entry,
+				 journal_entry_blacklist_v2_start_past_end,
 		"invalid journal seq blacklist entry: start > end")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 	}
@@ -482,9 +506,10 @@ static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_
 }
 
 static int journal_entry_usage_validate(struct bch_fs *c,
-					struct jset *jset,
-					struct jset_entry *entry,
-					unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	struct jset_entry_usage *u =
 		container_of(entry, struct jset_entry_usage, entry);
@@ -492,7 +517,8 @@ static int journal_entry_usage_validate(struct bch_fs *c,
 	int ret = 0;
 
 	if (journal_entry_err_on(bytes < sizeof(*u),
-				 c, jset, entry,
+				 c, version, jset, entry,
+				 journal_entry_usage_bad_size,
 				 "invalid journal entry usage: bad size")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 		return ret;
@@ -514,24 +540,36 @@ static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
 }
 
 static int journal_entry_data_usage_validate(struct bch_fs *c,
-					struct jset *jset,
-					struct jset_entry *entry,
-					unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	struct jset_entry_data_usage *u =
 		container_of(entry, struct jset_entry_data_usage, entry);
 	unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
+	struct printbuf err = PRINTBUF;
 	int ret = 0;
 
 	if (journal_entry_err_on(bytes < sizeof(*u) ||
 				 bytes < sizeof(*u) + u->r.nr_devs,
-				 c, jset, entry,
+				 c, version, jset, entry,
+				 journal_entry_data_usage_bad_size,
 				 "invalid journal entry usage: bad size")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
-		return ret;
+		goto out;
 	}
 
+	if (journal_entry_err_on(bch2_replicas_entry_validate(&u->r, c->disk_sb.sb, &err),
+				 c, version, jset, entry,
+				 journal_entry_data_usage_bad_size,
+				 "invalid journal entry usage: %s", err.buf)) {
+		journal_entry_null_range(entry, vstruct_next(entry));
+		goto out;
+	}
+out:
 fsck_err:
+	printbuf_exit(&err);
 	return ret;
 }
 
@@ -546,9 +584,10 @@ static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs
 }
 
 static int journal_entry_clock_validate(struct bch_fs *c,
-					struct jset *jset,
-					struct jset_entry *entry,
-					unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	struct jset_entry_clock *clock =
 		container_of(entry, struct jset_entry_clock, entry);
@@ -556,13 +595,17 @@ static int journal_entry_clock_validate(struct bch_fs *c,
 	int ret = 0;
 
 	if (journal_entry_err_on(bytes != sizeof(*clock),
-				 c, jset, entry, "bad size")) {
+				 c, version, jset, entry,
+				 journal_entry_clock_bad_size,
+				 "bad size")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 		return ret;
 	}
 
 	if (journal_entry_err_on(clock->rw > 1,
-				 c, jset, entry, "bad rw")) {
+				 c, version, jset, entry,
+				 journal_entry_clock_bad_rw,
+				 "bad rw")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 		return ret;
 	}
@@ -581,9 +624,10 @@ static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
 }
 
 static int journal_entry_dev_usage_validate(struct bch_fs *c,
-					    struct jset *jset,
-					    struct jset_entry *entry,
-					    unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	struct jset_entry_dev_usage *u =
 		container_of(entry, struct jset_entry_dev_usage, entry);
@@ -593,7 +637,9 @@ static int journal_entry_dev_usage_validate(struct bch_fs *c,
 	int ret = 0;
 
 	if (journal_entry_err_on(bytes < expected,
-				 c, jset, entry, "bad size (%u < %u)",
+				 c, version, jset, entry,
+				 journal_entry_dev_usage_bad_size,
+				 "bad size (%u < %u)",
 				 bytes, expected)) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 		return ret;
@@ -602,13 +648,17 @@ static int journal_entry_dev_usage_validate(struct bch_fs *c,
 	dev = le32_to_cpu(u->dev);
 
 	if (journal_entry_err_on(!bch2_dev_exists2(c, dev),
-				 c, jset, entry, "bad dev")) {
+				 c, version, jset, entry,
+				 journal_entry_dev_usage_bad_dev,
+				 "bad dev")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 		return ret;
 	}
 
 	if (journal_entry_err_on(u->pad,
-				 c, jset, entry, "bad pad")) {
+				 c, version, jset, entry,
+				 journal_entry_dev_usage_bad_pad,
+				 "bad pad")) {
 		journal_entry_null_range(entry, vstruct_next(entry));
 		return ret;
 	}
@@ -636,14 +686,13 @@ static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs
 		       le64_to_cpu(u->d[i].sectors),
 		       le64_to_cpu(u->d[i].fragmented));
 	}
-
-	prt_printf(out, " buckets_ec: %llu", le64_to_cpu(u->buckets_ec));
 }
 
 static int journal_entry_log_validate(struct bch_fs *c,
-				      struct jset *jset,
-				      struct jset_entry *entry,
-				      unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	return 0;
 }
@@ -658,9 +707,10 @@ static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
 }
 
 static int journal_entry_overwrite_validate(struct bch_fs *c,
-				      struct jset *jset,
-				      struct jset_entry *entry,
-				      unsigned version, int big_endian, int write)
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	return journal_entry_btree_keys_validate(c, jset, entry,
 				version, big_endian, READ);
@@ -672,9 +722,26 @@ static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs
 	journal_entry_btree_keys_to_text(out, c, entry);
 }
 
+static int journal_entry_write_buffer_keys_validate(struct bch_fs *c,
+				struct jset *jset,
+				struct jset_entry *entry,
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
+{
+	return journal_entry_btree_keys_validate(c, jset, entry,
+				version, big_endian, READ);
+}
+
+static void journal_entry_write_buffer_keys_to_text(struct printbuf *out, struct bch_fs *c,
+					    struct jset_entry *entry)
+{
+	journal_entry_btree_keys_to_text(out, c, entry);
+}
+
 struct jset_entry_ops {
 	int (*validate)(struct bch_fs *, struct jset *,
-			struct jset_entry *, unsigned, int, int);
+			struct jset_entry *, unsigned, int,
+			enum bkey_invalid_flags);
 	void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
 };
 
@@ -691,11 +758,12 @@ static const struct jset_entry_ops bch2_jset_entry_ops[] = {
 int bch2_journal_entry_validate(struct bch_fs *c,
 				struct jset *jset,
 				struct jset_entry *entry,
-				unsigned version, int big_endian, int write)
+				unsigned version, int big_endian,
+				enum bkey_invalid_flags flags)
 {
 	return entry->type < BCH_JSET_ENTRY_NR
 		? bch2_jset_entry_ops[entry->type].validate(c, jset, entry,
-				version, big_endian, write)
+				version, big_endian, flags)
 		: 0;
 }
 
@@ -711,22 +779,23 @@ void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
 }
 
 static int jset_validate_entries(struct bch_fs *c, struct jset *jset,
-				 int write)
+				 enum bkey_invalid_flags flags)
 {
 	struct jset_entry *entry;
+	unsigned version = le32_to_cpu(jset->version);
 	int ret = 0;
 
 	vstruct_for_each(jset, entry) {
-		if (journal_entry_err_on(vstruct_next(entry) >
-					 vstruct_last(jset), c, jset, entry,
+		if (journal_entry_err_on(vstruct_next(entry) > vstruct_last(jset),
+				c, version, jset, entry,
+				journal_entry_past_jset_end,
 				"journal entry extends past end of jset")) {
 			jset->u64s = cpu_to_le32((u64 *) entry - jset->_data);
 			break;
 		}
 
 		ret = bch2_journal_entry_validate(c, jset, entry,
-					le32_to_cpu(jset->version),
-					JSET_BIG_ENDIAN(jset), write);
+					version, JSET_BIG_ENDIAN(jset), flags);
 		if (ret)
 			break;
 	}
@@ -737,7 +806,7 @@ fsck_err:
 static int jset_validate(struct bch_fs *c,
 			 struct bch_dev *ca,
 			 struct jset *jset, u64 sector,
-			 int write)
+			 enum bkey_invalid_flags flags)
 {
 	unsigned version;
 	int ret = 0;
@@ -746,20 +815,21 @@ static int jset_validate(struct bch_fs *c,
 		return JOURNAL_ENTRY_NONE;
 
 	version = le32_to_cpu(jset->version);
-	if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD &&
-				  version < bcachefs_metadata_version_min) ||
-				 version >= bcachefs_metadata_version_max,
-				 c, jset, NULL,
-			"%s sector %llu seq %llu: unknown journal entry version %u",
+	if (journal_entry_err_on(!bch2_version_compatible(version),
+			c, version, jset, NULL,
+			jset_unsupported_version,
+			"%s sector %llu seq %llu: incompatible journal entry version %u.%u",
 			ca ? ca->name : c->name,
 			sector, le64_to_cpu(jset->seq),
-			version)) {
+			BCH_VERSION_MAJOR(version),
+			BCH_VERSION_MINOR(version))) {
 		/* don't try to continue: */
 		return -EINVAL;
 	}
 
 	if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)),
-				 c, jset, NULL,
+			c, version, jset, NULL,
+			jset_unknown_csum,
 			"%s sector %llu seq %llu: journal entry with unknown csum type %llu",
 			ca ? ca->name : c->name,
 			sector, le64_to_cpu(jset->seq),
@@ -769,7 +839,8 @@ static int jset_validate(struct bch_fs *c,
 	/* last_seq is ignored when JSET_NO_FLUSH is true */
 	if (journal_entry_err_on(!JSET_NO_FLUSH(jset) &&
 				 le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq),
-				 c, jset, NULL,
+				 c, version, jset, NULL,
+				 jset_last_seq_newer_than_seq,
 				 "invalid journal entry: last_seq > seq (%llu > %llu)",
 				 le64_to_cpu(jset->last_seq),
 				 le64_to_cpu(jset->seq))) {
@@ -777,7 +848,7 @@ static int jset_validate(struct bch_fs *c,
 		return JOURNAL_ENTRY_BAD;
 	}
 
-	ret = jset_validate_entries(c, jset, write);
+	ret = jset_validate_entries(c, jset, flags);
 fsck_err:
 	return ret;
 }
@@ -790,21 +861,21 @@ static int jset_validate_early(struct bch_fs *c,
 {
 	size_t bytes = vstruct_bytes(jset);
 	unsigned version;
-	int write = READ;
+	enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
 	int ret = 0;
 
 	if (le64_to_cpu(jset->magic) != jset_magic(c))
 		return JOURNAL_ENTRY_NONE;
 
 	version = le32_to_cpu(jset->version);
-	if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD &&
-				  version < bcachefs_metadata_version_min) ||
-				 version >= bcachefs_metadata_version_max,
-				 c, jset, NULL,
-			"%s sector %llu seq %llu: unknown journal entry version %u",
+	if (journal_entry_err_on(!bch2_version_compatible(version),
+			c, version, jset, NULL,
+			jset_unsupported_version,
+			"%s sector %llu seq %llu: unknown journal entry version %u.%u",
 			ca ? ca->name : c->name,
 			sector, le64_to_cpu(jset->seq),
-			version)) {
+			BCH_VERSION_MAJOR(version),
+			BCH_VERSION_MINOR(version))) {
 		/* don't try to continue: */
 		return -EINVAL;
 	}
@@ -814,7 +885,8 @@ static int jset_validate_early(struct bch_fs *c,
 		return JOURNAL_ENTRY_REREAD;
 
 	if (journal_entry_err_on(bytes > bucket_sectors_left << 9,
-				 c, jset, NULL,
+			c, version, jset, NULL,
+			jset_past_bucket_end,
 			"%s sector %llu seq %llu: journal entry too big (%zu bytes)",
 			ca ? ca->name : c->name,
 			sector, le64_to_cpu(jset->seq), bytes))
@@ -883,7 +955,7 @@ reread:
 			ret = submit_bio_wait(bio);
 			kfree(bio);
 
-			if (bch2_dev_io_err_on(ret, ca,
+			if (bch2_dev_io_err_on(ret, ca, BCH_MEMBER_ERROR_read,
 					       "journal read error: sector %llu",
 					       offset) ||
 			    bch2_meta_read_fault("journal")) {
@@ -939,7 +1011,8 @@ reread:
 		ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
 
 		csum_good = jset_csum_good(c, j);
-		if (!csum_good)
+		if (bch2_dev_io_err_on(!csum_good, ca, BCH_MEMBER_ERROR_checksum,
+				       "journal checksum error"))
 			saw_bad = true;
 
 		ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j),
@@ -977,10 +1050,9 @@ next_block:
 	return 0;
 }
 
-static void bch2_journal_read_device(struct closure *cl)
+static CLOSURE_CALLBACK(bch2_journal_read_device)
 {
-	struct journal_device *ja =
-		container_of(cl, struct journal_device, read);
+	closure_type(ja, struct journal_device, read);
 	struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
 	struct bch_fs *c = ca->fs;
 	struct journal_list *jlist =
@@ -1031,12 +1103,20 @@ found:
 
 	if (ja->bucket_seq[ja->cur_idx] &&
 	    ja->sectors_free == ca->mi.bucket_size) {
+#if 0
+		/*
+		 * Debug code for ZNS support, where we (probably) want to be
+		 * correlated where we stopped in the journal to the zone write
+		 * points:
+		 */
 		bch_err(c, "ja->sectors_free == ca->mi.bucket_size");
 		bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr);
 		for (i = 0; i < 3; i++) {
 			unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr;
+
 			bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]);
 		}
+#endif
 		ja->sectors_free = 0;
 	}
 
@@ -1131,7 +1211,7 @@ int bch2_journal_read(struct bch_fs *c,
 	 * those entries will be blacklisted:
 	 */
 	genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
-		int write = READ;
+		enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
 
 		i = *_i;
 
@@ -1153,7 +1233,8 @@ int bch2_journal_read(struct bch_fs *c,
 		}
 
 		if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq),
-					 c, &i->j, NULL,
+					 c, le32_to_cpu(i->j.version), &i->j, NULL,
+					 jset_last_seq_newer_than_seq,
 					 "invalid journal entry: last_seq > seq (%llu > %llu)",
 					 le64_to_cpu(i->j.last_seq),
 					 le64_to_cpu(i->j.seq)))
@@ -1170,7 +1251,8 @@ int bch2_journal_read(struct bch_fs *c,
 	}
 
 	if (!*last_seq) {
-		fsck_err(c, "journal read done, but no entries found after dropping non-flushes");
+		fsck_err(c, dirty_but_no_journal_entries_post_drop_nonflushes,
+			 "journal read done, but no entries found after dropping non-flushes");
 		return 0;
 	}
 
@@ -1196,6 +1278,7 @@ int bch2_journal_read(struct bch_fs *c,
 
 		if (bch2_journal_seq_is_blacklisted(c, seq, true)) {
 			fsck_err_on(!JSET_NO_FLUSH(&i->j), c,
+				    jset_seq_blacklisted,
 				    "found blacklisted journal entry %llu", seq);
 			i->ignore = true;
 		}
@@ -1236,7 +1319,8 @@ int bch2_journal_read(struct bch_fs *c,
 			bch2_journal_ptrs_to_text(&buf2, c, i);
 
 			missing_end = seq - 1;
-			fsck_err(c, "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
+			fsck_err(c, journal_entries_missing,
+				 "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
 				 "  prev at %s\n"
 				 "  next at %s",
 				 missing_start, missing_end,
@@ -1263,7 +1347,7 @@ int bch2_journal_read(struct bch_fs *c,
 			continue;
 
 		for (ptr = 0; ptr < i->nr_ptrs; ptr++) {
-			struct bch_dev *ca = bch_dev_bkey_exists(c, i->ptrs[ptr].dev);
+			ca = bch_dev_bkey_exists(c, i->ptrs[ptr].dev);
 
 			if (!i->ptrs[ptr].csum_good)
 				bch_err_dev_offset(ca, i->ptrs[ptr].sector,
@@ -1285,18 +1369,15 @@ int bch2_journal_read(struct bch_fs *c,
 
 		bch2_replicas_entry_sort(&replicas.e);
 
-		/*
-		 * If we're mounting in degraded mode - if we didn't read all
-		 * the devices - this is wrong:
-		 */
-
 		printbuf_reset(&buf);
 		bch2_replicas_entry_to_text(&buf, &replicas.e);
 
 		if (!degraded &&
-		    fsck_err_on(!bch2_replicas_marked(c, &replicas.e), c,
-				"superblock not marked as containing replicas %s",
-				buf.buf)) {
+		    !bch2_replicas_marked(c, &replicas.e) &&
+		    (le64_to_cpu(i->j.seq) == *last_seq ||
+		     fsck_err(c, journal_entry_replicas_not_marked,
+			      "superblock not marked as containing replicas for journal entry %llu\n  %s",
+			      le64_to_cpu(i->j.seq), buf.buf))) {
 			ret = bch2_mark_replicas(c, &replicas.e);
 			if (ret)
 				goto err;
@@ -1365,16 +1446,21 @@ static void __journal_write_alloc(struct journal *j,
 }
 
 /**
- * journal_next_bucket - move on to the next journal bucket if possible
+ * journal_write_alloc - decide where to write next journal entry
+ *
+ * @j:		journal object
+ * @w:		journal buf (entry to be written)
+ *
+ * Returns: 0 on success, or -EROFS on failure
  */
-static int journal_write_alloc(struct journal *j, struct journal_buf *w,
-			       unsigned sectors)
+static int journal_write_alloc(struct journal *j, struct journal_buf *w)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct bch_devs_mask devs;
 	struct journal_device *ja;
 	struct bch_dev *ca;
 	struct dev_alloc_list devs_sorted;
+	unsigned sectors = vstruct_sectors(w->data, c->block_bits);
 	unsigned target = c->opts.metadata_target ?:
 		c->opts.foreground_target;
 	unsigned i, replicas = 0, replicas_want =
@@ -1432,6 +1518,8 @@ done:
 
 static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+
 	/* we aren't holding j->lock: */
 	unsigned new_size = READ_ONCE(j->buf_size_want);
 	void *new_buf;
@@ -1439,7 +1527,12 @@ static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
 	if (buf->buf_size >= new_size)
 		return;
 
-	new_buf = kvpmalloc(new_size, GFP_NOIO|__GFP_NOWARN);
+	size_t btree_write_buffer_size = new_size / 64;
+
+	if (bch2_btree_write_buffer_resize(c, btree_write_buffer_size))
+		return;
+
+	new_buf = kvpmalloc(new_size, GFP_NOFS|__GFP_NOWARN);
 	if (!new_buf)
 		return;
 
@@ -1458,9 +1551,9 @@ static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
 	return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
 }
 
-static void journal_write_done(struct closure *cl)
+static CLOSURE_CALLBACK(journal_write_done)
 {
-	struct journal *j = container_of(cl, struct journal, io);
+	closure_type(j, struct journal, io);
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_buf *w = journal_last_unwritten_buf(j);
 	struct bch_replicas_padded replicas;
@@ -1513,7 +1606,7 @@ static void journal_write_done(struct closure *cl)
 	 * Must come before signaling write completion, for
 	 * bch2_fs_journal_stop():
 	 */
-	if (j->watermark)
+	if (j->watermark != BCH_WATERMARK_stripe)
 		journal_reclaim_kick(&c->journal);
 
 	/* also must come before signalling write completion: */
@@ -1528,13 +1621,18 @@ static void journal_write_done(struct closure *cl)
 	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
 				       old.v, new.v)) != old.v);
 
+	bch2_journal_reclaim_fast(j);
 	bch2_journal_space_available(j);
 
+	track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight],
+			   &j->max_in_flight_start, false);
+
 	closure_wake_up(&w->wait);
 	journal_wake(j);
 
 	if (!journal_state_count(new, new.unwritten_idx) &&
 	    journal_last_unwritten_seq(j) <= journal_cur_seq(j)) {
+		spin_unlock(&j->lock);
 		closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
 	} else if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
 		   new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
@@ -1547,10 +1645,11 @@ static void journal_write_done(struct closure *cl)
 		 * might want to be written now:
 		 */
 
+		spin_unlock(&j->lock);
 		mod_delayed_work(c->io_complete_wq, &j->write_work, max(0L, delta));
+	} else {
+		spin_unlock(&j->lock);
 	}
-
-	spin_unlock(&j->lock);
 }
 
 static void journal_write_endio(struct bio *bio)
@@ -1560,7 +1659,8 @@ static void journal_write_endio(struct bio *bio)
 	struct journal_buf *w = journal_last_unwritten_buf(j);
 	unsigned long flags;
 
-	if (bch2_dev_io_err_on(bio->bi_status, ca, "error writing journal entry %llu: %s",
+	if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write,
+			       "error writing journal entry %llu: %s",
 			       le64_to_cpu(w->data->seq),
 			       bch2_blk_status_to_str(bio->bi_status)) ||
 	    bch2_meta_write_fault("journal")) {
@@ -1573,9 +1673,9 @@ static void journal_write_endio(struct bio *bio)
 	percpu_ref_put(&ca->io_ref);
 }
 
-static void do_journal_write(struct closure *cl)
+static CLOSURE_CALLBACK(do_journal_write)
 {
-	struct journal *j = container_of(cl, struct journal, io);
+	closure_type(j, struct journal, io);
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct bch_dev *ca;
 	struct journal_buf *w = journal_last_unwritten_buf(j);
@@ -1618,12 +1718,19 @@ static void do_journal_write(struct closure *cl)
 	}
 
 	continue_at(cl, journal_write_done, c->io_complete_wq);
-	return;
 }
 
-static void bch2_journal_entries_postprocess(struct bch_fs *c, struct jset *jset)
+static int bch2_journal_write_prep(struct journal *j, struct journal_buf *w)
 {
-	struct jset_entry *i, *next, *prev = NULL;
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	struct jset_entry *start, *end, *i;
+	struct jset *jset = w->data;
+	struct journal_keys_to_wb wb = { NULL };
+	unsigned sectors, bytes, u64s;
+	unsigned long btree_roots_have = 0;
+	bool validate_before_checksum = false;
+	u64 seq = le64_to_cpu(jset->seq);
+	int ret;
 
 	/*
 	 * Simple compaction, dropping empty jset_entries (from journal
@@ -1633,62 +1740,113 @@ static void bch2_journal_entries_postprocess(struct bch_fs *c, struct jset *jset
 	 * If we wanted to be really fancy here, we could sort all the keys in
 	 * the jset and drop keys that were overwritten - probably not worth it:
 	 */
-	vstruct_for_each_safe(jset, i, next) {
+	vstruct_for_each(jset, i) {
 		unsigned u64s = le16_to_cpu(i->u64s);
 
 		/* Empty entry: */
 		if (!u64s)
 			continue;
 
-		if (i->type == BCH_JSET_ENTRY_btree_root)
+		/*
+		 * New btree roots are set by journalling them; when the journal
+		 * entry gets written we have to propagate them to
+		 * c->btree_roots
+		 *
+		 * But, every journal entry we write has to contain all the
+		 * btree roots (at least for now); so after we copy btree roots
+		 * to c->btree_roots we have to get any missing btree roots and
+		 * add them to this journal entry:
+		 */
+		switch (i->type) {
+		case BCH_JSET_ENTRY_btree_root:
 			bch2_journal_entry_to_btree_root(c, i);
-
-		/* Can we merge with previous entry? */
-		if (prev &&
-		    i->btree_id == prev->btree_id &&
-		    i->level	== prev->level &&
-		    i->type	== prev->type &&
-		    i->type	== BCH_JSET_ENTRY_btree_keys &&
-		    le16_to_cpu(prev->u64s) + u64s <= U16_MAX) {
-			memmove_u64s_down(vstruct_next(prev),
-					  i->_data,
-					  u64s);
-			le16_add_cpu(&prev->u64s, u64s);
-			continue;
+			__set_bit(i->btree_id, &btree_roots_have);
+			break;
+		case BCH_JSET_ENTRY_write_buffer_keys:
+			EBUG_ON(!w->need_flush_to_write_buffer);
+
+			if (!wb.wb)
+				bch2_journal_keys_to_write_buffer_start(c, &wb, seq);
+
+			struct bkey_i *k;
+			jset_entry_for_each_key(i, k) {
+				ret = bch2_journal_key_to_wb(c, &wb, i->btree_id, k);
+				if (ret) {
+					bch2_fs_fatal_error(c, "-ENOMEM flushing journal keys to btree write buffer");
+					bch2_journal_keys_to_write_buffer_end(c, &wb);
+					return ret;
+				}
+			}
+			i->type = BCH_JSET_ENTRY_btree_keys;
+			break;
 		}
+	}
 
-		/* Couldn't merge, move i into new position (after prev): */
-		prev = prev ? vstruct_next(prev) : jset->start;
-		if (i != prev)
-			memmove_u64s_down(prev, i, jset_u64s(u64s));
+	if (wb.wb)
+		bch2_journal_keys_to_write_buffer_end(c, &wb);
+	w->need_flush_to_write_buffer = false;
+
+	start = end = vstruct_last(jset);
+
+	end	= bch2_btree_roots_to_journal_entries(c, end, btree_roots_have);
+
+	bch2_journal_super_entries_add_common(c, &end, seq);
+	u64s	= (u64 *) end - (u64 *) start;
+	BUG_ON(u64s > j->entry_u64s_reserved);
+
+	le32_add_cpu(&jset->u64s, u64s);
+
+	sectors = vstruct_sectors(jset, c->block_bits);
+	bytes	= vstruct_bytes(jset);
+
+	if (sectors > w->sectors) {
+		bch2_fs_fatal_error(c, "aieeee! journal write overran available space, %zu > %u (extra %u reserved %u/%u)",
+				    vstruct_bytes(jset), w->sectors << 9,
+				    u64s, w->u64s_reserved, j->entry_u64s_reserved);
+		return -EINVAL;
 	}
 
-	prev = prev ? vstruct_next(prev) : jset->start;
-	jset->u64s = cpu_to_le32((u64 *) prev - jset->_data);
-}
+	jset->magic		= cpu_to_le64(jset_magic(c));
+	jset->version		= cpu_to_le32(c->sb.version);
 
-void bch2_journal_write(struct closure *cl)
-{
-	struct journal *j = container_of(cl, struct journal, io);
-	struct bch_fs *c = container_of(j, struct bch_fs, journal);
-	struct bch_dev *ca;
-	struct journal_buf *w = journal_last_unwritten_buf(j);
-	struct jset_entry *start, *end;
-	struct jset *jset;
-	struct bio *bio;
-	struct printbuf journal_debug_buf = PRINTBUF;
-	bool validate_before_checksum = false;
-	unsigned i, sectors, bytes, u64s, nr_rw_members = 0;
-	int ret;
+	SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
+	SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
 
-	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
+	if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
+		j->last_empty_seq = seq;
 
-	journal_buf_realloc(j, w);
-	jset = w->data;
+	if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
+		validate_before_checksum = true;
 
-	j->write_start_time = local_clock();
+	if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
+		validate_before_checksum = true;
 
-	spin_lock(&j->lock);
+	if (validate_before_checksum &&
+	    (ret = jset_validate(c, NULL, jset, 0, WRITE)))
+		return ret;
+
+	ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
+		    jset->encrypted_start,
+		    vstruct_end(jset) - (void *) jset->encrypted_start);
+	if (bch2_fs_fatal_err_on(ret, c,
+			"error decrypting journal entry: %i", ret))
+		return ret;
+
+	jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
+				  journal_nonce(jset), jset);
+
+	if (!validate_before_checksum &&
+	    (ret = jset_validate(c, NULL, jset, 0, WRITE)))
+		return ret;
+
+	memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
+	return 0;
+}
+
+static int bch2_journal_write_pick_flush(struct journal *j, struct journal_buf *w)
+{
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	int error = bch2_journal_error(j);
 
 	/*
 	 * If the journal is in an error state - we did an emergency shutdown -
@@ -1706,105 +1864,79 @@ void bch2_journal_write(struct closure *cl)
 	 * So if we're in an error state, and we're still starting up, we don't
 	 * write anything at all.
 	 */
-	if (!test_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags) &&
-	    (bch2_journal_error(j) ||
-	     w->noflush ||
-	     (!w->must_flush &&
-	      (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
-	      test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags)))) {
+	if (error && test_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags))
+		return -EIO;
+
+	if (error ||
+	    w->noflush ||
+	    (!w->must_flush &&
+	     (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
+	     test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags))) {
 		w->noflush = true;
-		SET_JSET_NO_FLUSH(jset, true);
-		jset->last_seq	= 0;
-		w->last_seq	= 0;
+		SET_JSET_NO_FLUSH(w->data, true);
+		w->data->last_seq	= 0;
+		w->last_seq		= 0;
 
 		j->nr_noflush_writes++;
-	} else if (!bch2_journal_error(j)) {
+	} else {
 		j->last_flush_write = jiffies;
 		j->nr_flush_writes++;
 		clear_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags);
-	} else {
-		spin_unlock(&j->lock);
-		goto err;
 	}
-	spin_unlock(&j->lock);
-
-	/*
-	 * New btree roots are set by journalling them; when the journal entry
-	 * gets written we have to propagate them to c->btree_roots
-	 *
-	 * But, every journal entry we write has to contain all the btree roots
-	 * (at least for now); so after we copy btree roots to c->btree_roots we
-	 * have to get any missing btree roots and add them to this journal
-	 * entry:
-	 */
 
-	bch2_journal_entries_postprocess(c, jset);
-
-	start = end = vstruct_last(jset);
-
-	end	= bch2_btree_roots_to_journal_entries(c, jset->start, end);
-
-	bch2_journal_super_entries_add_common(c, &end,
-				le64_to_cpu(jset->seq));
-	u64s	= (u64 *) end - (u64 *) start;
-	BUG_ON(u64s > j->entry_u64s_reserved);
-
-	le32_add_cpu(&jset->u64s, u64s);
-	BUG_ON(vstruct_sectors(jset, c->block_bits) > w->sectors);
-
-	jset->magic		= cpu_to_le64(jset_magic(c));
-	jset->version		= c->sb.version < bcachefs_metadata_version_bkey_renumber
-		? cpu_to_le32(BCH_JSET_VERSION_OLD)
-		: cpu_to_le32(c->sb.version);
-
-	SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
-	SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
-
-	if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
-		j->last_empty_seq = le64_to_cpu(jset->seq);
+	return 0;
+}
 
-	if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
-		validate_before_checksum = true;
+CLOSURE_CALLBACK(bch2_journal_write)
+{
+	closure_type(j, struct journal, io);
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	struct bch_dev *ca;
+	struct journal_buf *w = journal_last_unwritten_buf(j);
+	struct bch_replicas_padded replicas;
+	struct bio *bio;
+	struct printbuf journal_debug_buf = PRINTBUF;
+	unsigned i, nr_rw_members = 0;
+	int ret;
 
-	if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
-		validate_before_checksum = true;
+	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 
-	if (validate_before_checksum &&
-	    jset_validate(c, NULL, jset, 0, WRITE))
-		goto err;
+	j->write_start_time = local_clock();
 
-	ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
-		    jset->encrypted_start,
-		    vstruct_end(jset) - (void *) jset->encrypted_start);
-	if (bch2_fs_fatal_err_on(ret, c,
-			"error decrypting journal entry: %i", ret))
+	spin_lock(&j->lock);
+	ret = bch2_journal_write_pick_flush(j, w);
+	spin_unlock(&j->lock);
+	if (ret)
 		goto err;
 
-	jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
-				  journal_nonce(jset), jset);
+	mutex_lock(&j->buf_lock);
+	journal_buf_realloc(j, w);
 
-	if (!validate_before_checksum &&
-	    jset_validate(c, NULL, jset, 0, WRITE))
+	ret = bch2_journal_write_prep(j, w);
+	mutex_unlock(&j->buf_lock);
+	if (ret)
 		goto err;
 
-	sectors = vstruct_sectors(jset, c->block_bits);
-	BUG_ON(sectors > w->sectors);
-
-	bytes = vstruct_bytes(jset);
-	memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
+	j->entry_bytes_written += vstruct_bytes(w->data);
 
-retry_alloc:
-	spin_lock(&j->lock);
-	ret = journal_write_alloc(j, w, sectors);
+	while (1) {
+		spin_lock(&j->lock);
+		ret = journal_write_alloc(j, w);
+		if (!ret || !j->can_discard)
+			break;
 
-	if (ret && j->can_discard) {
 		spin_unlock(&j->lock);
 		bch2_journal_do_discards(j);
-		goto retry_alloc;
 	}
 
-	if (ret)
+	if (ret) {
 		__bch2_journal_debug_to_text(&journal_debug_buf, j);
+		spin_unlock(&j->lock);
+		bch_err(c, "Unable to allocate journal write:\n%s",
+			journal_debug_buf.buf);
+		printbuf_exit(&journal_debug_buf);
+		goto err;
+	}
 
 	/*
 	 * write is allocated, no longer need to account for it in
@@ -1819,15 +1951,6 @@ retry_alloc:
 	bch2_journal_space_available(j);
 	spin_unlock(&j->lock);
 
-	if (ret) {
-		bch_err(c, "Unable to allocate journal write:\n%s",
-			journal_debug_buf.buf);
-		printbuf_exit(&journal_debug_buf);
-		bch2_fatal_error(c);
-		continue_at(cl, journal_write_done, c->io_complete_wq);
-		return;
-	}
-
 	w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
 
 	if (c->opts.nochanges)
@@ -1839,7 +1962,17 @@ retry_alloc:
 	if (nr_rw_members > 1)
 		w->separate_flush = true;
 
-	if (!JSET_NO_FLUSH(jset) && w->separate_flush) {
+	/*
+	 * Mark journal replicas before we submit the write to guarantee
+	 * recovery will find the journal entries after a crash.
+	 */
+	bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
+				 w->devs_written);
+	ret = bch2_mark_replicas(c, &replicas.e);
+	if (ret)
+		goto err;
+
+	if (!JSET_NO_FLUSH(w->data) && w->separate_flush) {
 		for_each_rw_member(ca, c, i) {
 			percpu_ref_get(&ca->io_ref);