X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fjournal.c;h=5699a9d8d9f957aa747765fc38e28f6dc37a0352;hb=fa358537725c8065b058b558125cf15359936f94;hp=b4fe27f8f5ca3fb780d9ffea3c095c67a810841e;hpb=17e2f2775be6e10b966cd958bc0461aab662571a;p=bcachefs-tools-debian

diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c
index b4fe27f..5699a9d 100644
--- a/libbcachefs/journal.c
+++ b/libbcachefs/journal.c
@@ -1,3 +1,4 @@
+// SPDX-License-Identifier: GPL-2.0
 /*
  * bcachefs journalling code, for btree insertions
  *
@@ -5,236 +6,275 @@
  */
 
 #include "bcachefs.h"
-#include "alloc.h"
+#include "alloc_foreground.h"
 #include "bkey_methods.h"
 #include "btree_gc.h"
+#include "btree_update.h"
 #include "buckets.h"
+#include "error.h"
 #include "journal.h"
 #include "journal_io.h"
 #include "journal_reclaim.h"
+#include "journal_sb.h"
 #include "journal_seq_blacklist.h"
-#include "super-io.h"
 
 #include <trace/events/bcachefs.h>
 
-static bool journal_entry_is_open(struct journal *j)
+#define x(n)	#n,
+static const char * const bch2_journal_watermarks[] = {
+	JOURNAL_WATERMARKS()
+	NULL
+};
+
+static const char * const bch2_journal_errors[] = {
+	JOURNAL_ERRORS()
+	NULL
+};
+#undef x
+
+static inline bool journal_seq_unwritten(struct journal *j, u64 seq)
 {
-	return j->reservations.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
+	return seq > j->seq_ondisk;
 }
 
-void bch2_journal_buf_put_slowpath(struct journal *j, bool need_write_just_set)
+static bool __journal_entry_is_open(union journal_res_state state)
 {
-	struct journal_buf *w = journal_prev_buf(j);
-
-	atomic_dec_bug(&journal_seq_pin(j, le64_to_cpu(w->data->seq))->count);
-
-	if (!need_write_just_set &&
-	    test_bit(JOURNAL_NEED_WRITE, &j->flags))
-		bch2_time_stats_update(j->delay_time,
-				       j->need_write_time);
-#if 0
-	closure_call(&j->io, bch2_journal_write, NULL, NULL);
-#else
-	/* Shut sparse up: */
-	closure_init(&j->io, NULL);
-	set_closure_fn(&j->io, bch2_journal_write, NULL);
-	bch2_journal_write(&j->io);
-#endif
+	return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
 }
 
-static void journal_pin_new_entry(struct journal *j, int count)
+static inline unsigned nr_unwritten_journal_entries(struct journal *j)
 {
-	struct journal_entry_pin_list *p;
-
-	/*
-	 * The fifo_push() needs to happen at the same time as j->seq is
-	 * incremented for journal_last_seq() to be calculated correctly
-	 */
-	atomic64_inc(&j->seq);
-	p = fifo_push_ref(&j->pin);
+	return atomic64_read(&j->seq) - j->seq_ondisk;
+}
 
-	INIT_LIST_HEAD(&p->list);
-	INIT_LIST_HEAD(&p->flushed);
-	atomic_set(&p->count, count);
-	p->devs.nr = 0;
+static bool journal_entry_is_open(struct journal *j)
+{
+	return __journal_entry_is_open(j->reservations);
 }
 
-static void bch2_journal_buf_init(struct journal *j)
+static inline struct journal_buf *
+journal_seq_to_buf(struct journal *j, u64 seq)
 {
-	struct journal_buf *buf = journal_cur_buf(j);
+	struct journal_buf *buf = NULL;
 
-	memset(buf->has_inode, 0, sizeof(buf->has_inode));
+	EBUG_ON(seq > journal_cur_seq(j));
 
-	memset(buf->data, 0, sizeof(*buf->data));
-	buf->data->seq	= cpu_to_le64(journal_cur_seq(j));
-	buf->data->u64s	= 0;
+	if (journal_seq_unwritten(j, seq)) {
+		buf = j->buf + (seq & JOURNAL_BUF_MASK);
+		EBUG_ON(le64_to_cpu(buf->data->seq) != seq);
+	}
+	return buf;
 }
 
-static inline size_t journal_entry_u64s_reserve(struct journal_buf *buf)
+static void journal_pin_list_init(struct journal_entry_pin_list *p, int count)
 {
-	return BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_EXTENT_U64s_MAX);
+	unsigned i;
+	for (i = 0; i < ARRAY_SIZE(p->list); i++)
+		INIT_LIST_HEAD(&p->list[i]);
+	INIT_LIST_HEAD(&p->flushed);
+	atomic_set(&p->count, count);
+	p->devs.nr = 0;
 }
 
-static inline bool journal_entry_empty(struct jset *j)
-{
-	struct jset_entry *i;
+/* journal entry close/open: */
 
-	if (j->seq != j->last_seq)
-		return false;
+void __bch2_journal_buf_put(struct journal *j)
+{
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 
-	vstruct_for_each(j, i)
-		if (i->type || i->u64s)
-			return false;
-	return true;
+	closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
 }
 
-static enum {
-	JOURNAL_ENTRY_ERROR,
-	JOURNAL_ENTRY_INUSE,
-	JOURNAL_ENTRY_CLOSED,
-	JOURNAL_UNLOCKED,
-} journal_buf_switch(struct journal *j, bool need_write_just_set)
+/*
+ * Returns true if journal entry is now closed:
+ *
+ * We don't close a journal_buf until the next journal_buf is finished writing,
+ * and can be opened again - this also initializes the next journal_buf:
+ */
+static void __journal_entry_close(struct journal *j, unsigned closed_val)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
-	struct journal_buf *buf;
+	struct journal_buf *buf = journal_cur_buf(j);
 	union journal_res_state old, new;
 	u64 v = atomic64_read(&j->reservations.counter);
+	unsigned sectors;
+
+	BUG_ON(closed_val != JOURNAL_ENTRY_CLOSED_VAL &&
+	       closed_val != JOURNAL_ENTRY_ERROR_VAL);
 
 	lockdep_assert_held(&j->lock);
 
 	do {
 		old.v = new.v = v;
-		if (old.cur_entry_offset == JOURNAL_ENTRY_CLOSED_VAL)
-			return JOURNAL_ENTRY_CLOSED;
-
-		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL)
-			return JOURNAL_ENTRY_ERROR;
+		new.cur_entry_offset = closed_val;
 
-		if (new.prev_buf_unwritten)
-			return JOURNAL_ENTRY_INUSE;
-
-		/*
-		 * avoid race between setting buf->data->u64s and
-		 * journal_res_put starting write:
-		 */
-		journal_state_inc(&new);
-
-		new.cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL;
-		new.idx++;
-		new.prev_buf_unwritten = 1;
-
-		BUG_ON(journal_state_count(new, new.idx));
+		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL ||
+		    old.cur_entry_offset == new.cur_entry_offset)
+			return;
 	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
 				       old.v, new.v)) != old.v);
 
-	clear_bit(JOURNAL_NEED_WRITE, &j->flags);
+	if (!__journal_entry_is_open(old))
+		return;
 
-	buf = &j->buf[old.idx];
+	/* Close out old buffer: */
 	buf->data->u64s		= cpu_to_le32(old.cur_entry_offset);
 
-	j->prev_buf_sectors =
-		vstruct_blocks_plus(buf->data, c->block_bits,
-				    journal_entry_u64s_reserve(buf)) *
-		c->opts.block_size;
-	BUG_ON(j->prev_buf_sectors > j->cur_buf_sectors);
+	sectors = vstruct_blocks_plus(buf->data, c->block_bits,
+				      buf->u64s_reserved) << c->block_bits;
+	BUG_ON(sectors > buf->sectors);
+	buf->sectors = sectors;
 
-	bch2_journal_reclaim_fast(j);
-	/* XXX: why set this here, and not in bch2_journal_write()? */
-	buf->data->last_seq	= cpu_to_le64(journal_last_seq(j));
+	/*
+	 * We have to set last_seq here, _before_ opening a new journal entry:
+	 *
+	 * A threads may replace an old pin with a new pin on their current
+	 * journal reservation - the expectation being that the journal will
+	 * contain either what the old pin protected or what the new pin
+	 * protects.
+	 *
+	 * After the old pin is dropped journal_last_seq() won't include the old
+	 * pin, so we can only write the updated last_seq on the entry that
+	 * contains whatever the new pin protects.
+	 *
+	 * Restated, we can _not_ update last_seq for a given entry if there
+	 * could be a newer entry open with reservations/pins that have been
+	 * taken against it.
+	 *
+	 * Hence, we want update/set last_seq on the current journal entry right
+	 * before we open a new one:
+	 */
+	buf->last_seq		= journal_last_seq(j);
+	buf->data->last_seq	= cpu_to_le64(buf->last_seq);
+	BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq));
 
-	if (journal_entry_empty(buf->data))
-		clear_bit(JOURNAL_NOT_EMPTY, &j->flags);
-	else
-		set_bit(JOURNAL_NOT_EMPTY, &j->flags);
+	__bch2_journal_pin_put(j, le64_to_cpu(buf->data->seq));
 
-	journal_pin_new_entry(j, 1);
+	cancel_delayed_work(&j->write_work);
 
-	bch2_journal_buf_init(j);
+	bch2_journal_space_available(j);
 
-	cancel_delayed_work(&j->write_work);
+	bch2_journal_buf_put(j, old.idx);
+}
+
+void bch2_journal_halt(struct journal *j)
+{
+	spin_lock(&j->lock);
+	__journal_entry_close(j, JOURNAL_ENTRY_ERROR_VAL);
+	if (!j->err_seq)
+		j->err_seq = journal_cur_seq(j);
 	spin_unlock(&j->lock);
+}
 
-	if (c->bucket_journal_seq > 1 << 14) {
-		c->bucket_journal_seq = 0;
-		bch2_bucket_seq_cleanup(c);
-	}
+static bool journal_entry_want_write(struct journal *j)
+{
+	bool ret = !journal_entry_is_open(j) ||
+		journal_cur_seq(j) == journal_last_unwritten_seq(j);
 
-	c->bucket_journal_seq++;
+	/* Don't close it yet if we already have a write in flight: */
+	if (ret)
+		__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
+	else if (nr_unwritten_journal_entries(j)) {
+		struct journal_buf *buf = journal_cur_buf(j);
 
-	/* ugh - might be called from __journal_res_get() under wait_event() */
-	__set_current_state(TASK_RUNNING);
-	bch2_journal_buf_put(j, old.idx, need_write_just_set);
+		if (!buf->flush_time) {
+			buf->flush_time	= local_clock() ?: 1;
+			buf->expires = jiffies;
+		}
+	}
 
-	return JOURNAL_UNLOCKED;
+	return ret;
 }
 
-void bch2_journal_halt(struct journal *j)
+static bool journal_entry_close(struct journal *j)
 {
-	union journal_res_state old, new;
-	u64 v = atomic64_read(&j->reservations.counter);
-
-	do {
-		old.v = new.v = v;
-		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL)
-			return;
+	bool ret;
 
-		new.cur_entry_offset = JOURNAL_ENTRY_ERROR_VAL;
-	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
-				       old.v, new.v)) != old.v);
+	spin_lock(&j->lock);
+	ret = journal_entry_want_write(j);
+	spin_unlock(&j->lock);
 
-	journal_wake(j);
-	closure_wake_up(&journal_cur_buf(j)->wait);
-	closure_wake_up(&journal_prev_buf(j)->wait);
+	return ret;
 }
 
 /*
  * should _only_ called from journal_res_get() - when we actually want a
  * journal reservation - journal entry is open means journal is dirty:
- *
- * returns:
- * 1:		success
- * 0:		journal currently full (must wait)
- * -EROFS:	insufficient rw devices
- * -EIO:	journal error
  */
 static int journal_entry_open(struct journal *j)
 {
-	struct journal_buf *buf = journal_cur_buf(j);
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	struct journal_buf *buf = j->buf +
+		((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK);
 	union journal_res_state old, new;
-	ssize_t u64s;
-	int sectors;
+	int u64s;
 	u64 v;
 
 	lockdep_assert_held(&j->lock);
 	BUG_ON(journal_entry_is_open(j));
+	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
+
+	if (j->blocked)
+		return JOURNAL_ERR_blocked;
+
+	if (j->cur_entry_error)
+		return j->cur_entry_error;
+
+	if (bch2_journal_error(j))
+		return JOURNAL_ERR_insufficient_devices; /* -EROFS */
 
 	if (!fifo_free(&j->pin))
-		return 0;
+		return JOURNAL_ERR_journal_pin_full;
+
+	if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
+		return JOURNAL_ERR_max_in_flight;
+
+	BUG_ON(!j->cur_entry_sectors);
+
+	buf->expires		=
+		(journal_cur_seq(j) == j->flushed_seq_ondisk
+		 ? jiffies
+		 : j->last_flush_write) +
+		msecs_to_jiffies(c->opts.journal_flush_delay);
 
-	sectors = bch2_journal_entry_sectors(j);
-	if (sectors <= 0)
-		return sectors;
+	buf->u64s_reserved	= j->entry_u64s_reserved;
+	buf->disk_sectors	= j->cur_entry_sectors;
+	buf->sectors		= min(buf->disk_sectors, buf->buf_size >> 9);
 
-	buf->disk_sectors	= sectors;
+	u64s = (int) (buf->sectors << 9) / sizeof(u64) -
+		journal_entry_overhead(j);
+	u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 
-	sectors = min_t(unsigned, sectors, buf->size >> 9);
-	j->cur_buf_sectors	= sectors;
+	if (u64s <= (ssize_t) j->early_journal_entries.nr)
+		return JOURNAL_ERR_journal_full;
 
-	u64s = (sectors << 9) / sizeof(u64);
+	if (fifo_empty(&j->pin) && j->reclaim_thread)
+		wake_up_process(j->reclaim_thread);
 
-	/* Subtract the journal header */
-	u64s -= sizeof(struct jset) / sizeof(u64);
 	/*
-	 * Btree roots, prio pointers don't get added until right before we do
-	 * the write:
+	 * The fifo_push() needs to happen at the same time as j->seq is
+	 * incremented for journal_last_seq() to be calculated correctly
 	 */
-	u64s -= journal_entry_u64s_reserve(buf);
-	u64s  = max_t(ssize_t, 0L, u64s);
+	atomic64_inc(&j->seq);
+	journal_pin_list_init(fifo_push_ref(&j->pin), 1);
 
-	BUG_ON(u64s >= JOURNAL_ENTRY_CLOSED_VAL);
+	BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf);
 
-	if (u64s <= le32_to_cpu(buf->data->u64s))
-		return 0;
+	bkey_extent_init(&buf->key);
+	buf->noflush	= false;
+	buf->must_flush	= false;
+	buf->separate_flush = false;
+	buf->flush_time	= 0;
+
+	memset(buf->data, 0, sizeof(*buf->data));
+	buf->data->seq	= cpu_to_le64(journal_cur_seq(j));
+	buf->data->u64s	= 0;
+
+	if (j->early_journal_entries.nr) {
+		memcpy(buf->data->_data, j->early_journal_entries.data,
+		       j->early_journal_entries.nr * sizeof(u64));
+		le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr);
+	}
 
 	/*
 	 * Must be set before marking the journal entry as open:
@@ -245,8 +285,13 @@ static int journal_entry_open(struct journal *j)
 	do {
 		old.v = new.v = v;
 
-		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL)
-			return -EIO;
+		BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
+
+		new.idx++;
+		BUG_ON(journal_state_count(new, new.idx));
+		BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
+
+		journal_state_inc(&new);
 
 		/* Handle any already added entries */
 		new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
@@ -258,88 +303,83 @@ static int journal_entry_open(struct journal *j)
 				       j->res_get_blocked_start);
 	j->res_get_blocked_start = 0;
 
-	mod_delayed_work(system_freezable_wq,
+	mod_delayed_work(c->io_complete_wq,
 			 &j->write_work,
-			 msecs_to_jiffies(j->write_delay_ms));
+			 msecs_to_jiffies(c->opts.journal_flush_delay));
 	journal_wake(j);
-	return 1;
+
+	if (j->early_journal_entries.nr)
+		darray_exit(&j->early_journal_entries);
+	return 0;
 }
 
-/*
- * returns true if there's nothing to flush and no journal write still in flight
- */
-static bool journal_flush_write(struct journal *j)
+static bool journal_quiesced(struct journal *j)
 {
-	bool ret;
-
-	spin_lock(&j->lock);
-	ret = !j->reservations.prev_buf_unwritten;
-
-	if (!journal_entry_is_open(j)) {
-		spin_unlock(&j->lock);
-		return ret;
-	}
+	bool ret = atomic64_read(&j->seq) == j->seq_ondisk;
 
-	set_bit(JOURNAL_NEED_WRITE, &j->flags);
-	if (journal_buf_switch(j, false) == JOURNAL_UNLOCKED)
-		ret = false;
-	else
-		spin_unlock(&j->lock);
+	if (!ret)
+		journal_entry_close(j);
 	return ret;
 }
 
-static void journal_write_work(struct work_struct *work)
+static void journal_quiesce(struct journal *j)
 {
-	struct journal *j = container_of(work, struct journal, write_work.work);
-
-	journal_flush_write(j);
+	wait_event(j->wait, journal_quiesced(j));
 }
 
-/*
- * Given an inode number, if that inode number has data in the journal that
- * hasn't yet been flushed, return the journal sequence number that needs to be
- * flushed:
- */
-u64 bch2_inode_journal_seq(struct journal *j, u64 inode)
+static void journal_write_work(struct work_struct *work)
 {
-	size_t h = hash_64(inode, ilog2(sizeof(j->buf[0].has_inode) * 8));
-	u64 seq = 0;
-
-	if (!test_bit(h, j->buf[0].has_inode) &&
-	    !test_bit(h, j->buf[1].has_inode))
-		return 0;
+	struct journal *j = container_of(work, struct journal, write_work.work);
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	long delta;
 
 	spin_lock(&j->lock);
-	if (test_bit(h, journal_cur_buf(j)->has_inode))
-		seq = journal_cur_seq(j);
-	else if (test_bit(h, journal_prev_buf(j)->has_inode))
-		seq = journal_cur_seq(j) - 1;
-	spin_unlock(&j->lock);
+	if (!__journal_entry_is_open(j->reservations))
+		goto unlock;
+
+	delta = journal_cur_buf(j)->expires - jiffies;
 
-	return seq;
+	if (delta > 0)
+		mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
+	else
+		__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
+unlock:
+	spin_unlock(&j->lock);
 }
 
 static int __journal_res_get(struct journal *j, struct journal_res *res,
-			      unsigned u64s_min, unsigned u64s_max)
+			     unsigned flags)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_buf *buf;
+	bool can_discard;
 	int ret;
 retry:
-	ret = journal_res_get_fast(j, res, u64s_min, u64s_max);
-	if (ret)
-		return ret;
+	if (journal_res_get_fast(j, res, flags))
+		return 0;
+
+	if (bch2_journal_error(j))
+		return -BCH_ERR_erofs_journal_err;
 
 	spin_lock(&j->lock);
+
 	/*
 	 * Recheck after taking the lock, so we don't race with another thread
 	 * that just did journal_entry_open() and call journal_entry_close()
 	 * unnecessarily
 	 */
-	ret = journal_res_get_fast(j, res, u64s_min, u64s_max);
-	if (ret) {
+	if (journal_res_get_fast(j, res, flags)) {
 		spin_unlock(&j->lock);
-		return 1;
+		return 0;
+	}
+
+	if ((flags & JOURNAL_WATERMARK_MASK) < j->watermark) {
+		/*
+		 * Don't want to close current journal entry, just need to
+		 * invoke reclaim:
+		 */
+		ret = JOURNAL_ERR_journal_full;
+		goto unlock;
 	}
 
 	/*
@@ -349,51 +389,71 @@ retry:
 	 */
 	buf = journal_cur_buf(j);
 	if (journal_entry_is_open(j) &&
-	    buf->size >> 9 < buf->disk_sectors &&
-	    buf->size < JOURNAL_ENTRY_SIZE_MAX)
-		j->buf_size_want = max(j->buf_size_want, buf->size << 1);
+	    buf->buf_size >> 9 < buf->disk_sectors &&
+	    buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
+		j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 
-	/*
-	 * Close the current journal entry if necessary, then try to start a new
-	 * one:
-	 */
-	switch (journal_buf_switch(j, false)) {
-	case JOURNAL_ENTRY_ERROR:
-		spin_unlock(&j->lock);
-		return -EROFS;
-	case JOURNAL_ENTRY_INUSE:
-		/* haven't finished writing out the previous one: */
-		spin_unlock(&j->lock);
-		trace_journal_entry_full(c);
-		goto blocked;
-	case JOURNAL_ENTRY_CLOSED:
-		break;
-	case JOURNAL_UNLOCKED:
-		goto retry;
+	__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
+	ret = journal_entry_open(j);
+
+	if (ret == JOURNAL_ERR_max_in_flight)
+		trace_and_count(c, journal_entry_full, c);
+unlock:
+	if ((ret && ret != JOURNAL_ERR_insufficient_devices) &&
+	    !j->res_get_blocked_start) {
+		j->res_get_blocked_start = local_clock() ?: 1;
+		trace_and_count(c, journal_full, c);
 	}
 
-	/* We now have a new, closed journal buf - see if we can open it: */
-	ret = journal_entry_open(j);
+	can_discard = j->can_discard;
 	spin_unlock(&j->lock);
 
-	if (ret < 0)
-		return ret;
-	if (ret)
+	if (!ret)
 		goto retry;
 
-	/* Journal's full, we have to wait */
+	if ((ret == JOURNAL_ERR_journal_full ||
+	     ret == JOURNAL_ERR_journal_pin_full) &&
+	    !can_discard &&
+	    !nr_unwritten_journal_entries(j) &&
+	    (flags & JOURNAL_WATERMARK_MASK) == JOURNAL_WATERMARK_reserved) {
+		struct printbuf buf = PRINTBUF;
+
+		bch_err(c, "Journal stuck! Hava a pre-reservation but journal full (ret %s)",
+			bch2_journal_errors[ret]);
+
+		bch2_journal_debug_to_text(&buf, j);
+		bch_err(c, "%s", buf.buf);
+
+		printbuf_reset(&buf);
+		bch2_journal_pins_to_text(&buf, j);
+		bch_err(c, "Journal pins:\n%s", buf.buf);
+
+		printbuf_exit(&buf);
+		bch2_fatal_error(c);
+		dump_stack();
+	}
 
 	/*
-	 * Direct reclaim - can't rely on reclaim from work item
-	 * due to freezing..
+	 * Journal is full - can't rely on reclaim from work item due to
+	 * freezing:
 	 */
-	bch2_journal_reclaim_work(&j->reclaim_work.work);
+	if ((ret == JOURNAL_ERR_journal_full ||
+	     ret == JOURNAL_ERR_journal_pin_full) &&
+	    !(flags & JOURNAL_RES_GET_NONBLOCK)) {
+		if (can_discard) {
+			bch2_journal_do_discards(j);
+			goto retry;
+		}
 
-	trace_journal_full(c);
-blocked:
-	if (!j->res_get_blocked_start)
-		j->res_get_blocked_start = local_clock() ?: 1;
-	return 0;
+		if (mutex_trylock(&j->reclaim_lock)) {
+			bch2_journal_reclaim(j);
+			mutex_unlock(&j->reclaim_lock);
+		}
+	}
+
+	return ret == JOURNAL_ERR_insufficient_devices
+		? -EROFS
+		: -BCH_ERR_journal_res_get_blocked;
 }
 
 /*
@@ -407,208 +467,169 @@ blocked:
  * btree node write locks.
  */
 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
-				 unsigned u64s_min, unsigned u64s_max)
+				  unsigned flags)
 {
 	int ret;
 
-	wait_event(j->wait,
-		   (ret = __journal_res_get(j, res, u64s_min,
-					    u64s_max)));
-	return ret < 0 ? ret : 0;
+	closure_wait_event(&j->async_wait,
+		   (ret = __journal_res_get(j, res, flags)) !=
+		   -BCH_ERR_journal_res_get_blocked||
+		   (flags & JOURNAL_RES_GET_NONBLOCK));
+	return ret;
 }
 
-u64 bch2_journal_last_unwritten_seq(struct journal *j)
+/* journal_preres: */
+
+static bool journal_preres_available(struct journal *j,
+				     struct journal_preres *res,
+				     unsigned new_u64s,
+				     unsigned flags)
 {
-	u64 seq;
+	bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags, true);
 
-	spin_lock(&j->lock);
-	seq = journal_cur_seq(j);
-	if (j->reservations.prev_buf_unwritten)
-		seq--;
-	spin_unlock(&j->lock);
+	if (!ret && mutex_trylock(&j->reclaim_lock)) {
+		bch2_journal_reclaim(j);
+		mutex_unlock(&j->reclaim_lock);
+	}
 
-	return seq;
+	return ret;
 }
 
-/**
- * bch2_journal_open_seq_async - try to open a new journal entry if @seq isn't
- * open yet, or wait if we cannot
- *
- * used by the btree interior update machinery, when it needs to write a new
- * btree root - every journal entry contains the roots of all the btrees, so it
- * doesn't need to bother with getting a journal reservation
- */
-int bch2_journal_open_seq_async(struct journal *j, u64 seq, struct closure *parent)
+int __bch2_journal_preres_get(struct journal *j,
+			      struct journal_preres *res,
+			      unsigned new_u64s,
+			      unsigned flags)
 {
 	int ret;
 
-	spin_lock(&j->lock);
-	BUG_ON(seq > journal_cur_seq(j));
-
-	if (seq < journal_cur_seq(j) ||
-	    journal_entry_is_open(j)) {
-		spin_unlock(&j->lock);
-		return 1;
-	}
-
-	ret = journal_entry_open(j);
-	if (!ret)
-		closure_wait(&j->async_wait, parent);
-	spin_unlock(&j->lock);
-
-	if (!ret)
-		bch2_journal_reclaim_work(&j->reclaim_work.work);
-
+	closure_wait_event(&j->preres_wait,
+		   (ret = bch2_journal_error(j)) ||
+		   journal_preres_available(j, res, new_u64s, flags));
 	return ret;
 }
 
-/**
- * bch2_journal_wait_on_seq - wait for a journal entry to be written
- *
- * does _not_ cause @seq to be written immediately - if there is no other
- * activity to cause the relevant journal entry to be filled up or flushed it
- * can wait for an arbitrary amount of time (up to @j->write_delay_ms, which is
- * configurable).
- */
-void bch2_journal_wait_on_seq(struct journal *j, u64 seq, struct closure *parent)
-{
-	spin_lock(&j->lock);
+/* journal_entry_res: */
 
-	BUG_ON(seq > journal_cur_seq(j));
+void bch2_journal_entry_res_resize(struct journal *j,
+				   struct journal_entry_res *res,
+				   unsigned new_u64s)
+{
+	union journal_res_state state;
+	int d = new_u64s - res->u64s;
 
-	if (bch2_journal_error(j)) {
-		spin_unlock(&j->lock);
-		return;
-	}
+	spin_lock(&j->lock);
 
-	if (seq == journal_cur_seq(j)) {
-		if (!closure_wait(&journal_cur_buf(j)->wait, parent))
-			BUG();
-	} else if (seq + 1 == journal_cur_seq(j) &&
-		   j->reservations.prev_buf_unwritten) {
-		if (!closure_wait(&journal_prev_buf(j)->wait, parent))
-			BUG();
+	j->entry_u64s_reserved += d;
+	if (d <= 0)
+		goto out;
 
-		smp_mb();
+	j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
+	smp_mb();
+	state = READ_ONCE(j->reservations);
 
-		/* check if raced with write completion (or failure) */
-		if (!j->reservations.prev_buf_unwritten ||
-		    bch2_journal_error(j))
-			closure_wake_up(&journal_prev_buf(j)->wait);
+	if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
+	    state.cur_entry_offset > j->cur_entry_u64s) {
+		j->cur_entry_u64s += d;
+		/*
+		 * Not enough room in current journal entry, have to flush it:
+		 */
+		__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
+	} else {
+		journal_cur_buf(j)->u64s_reserved += d;
 	}
-
+out:
 	spin_unlock(&j->lock);
+	res->u64s += d;
 }
 
+/* journal flushing: */
+
 /**
  * bch2_journal_flush_seq_async - wait for a journal entry to be written
  *
  * like bch2_journal_wait_on_seq, except that it triggers a write immediately if
  * necessary
  */
-void bch2_journal_flush_seq_async(struct journal *j, u64 seq, struct closure *parent)
+int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
+				 struct closure *parent)
 {
 	struct journal_buf *buf;
+	int ret = 0;
+
+	if (seq <= j->flushed_seq_ondisk)
+		return 1;
 
 	spin_lock(&j->lock);
 
-	BUG_ON(seq > journal_cur_seq(j));
+	if (WARN_ONCE(seq > journal_cur_seq(j),
+		      "requested to flush journal seq %llu, but currently at %llu",
+		      seq, journal_cur_seq(j)))
+		goto out;
 
-	if (bch2_journal_error(j)) {
-		spin_unlock(&j->lock);
-		return;
+	/* Recheck under lock: */
+	if (j->err_seq && seq >= j->err_seq) {
+		ret = -EIO;
+		goto out;
 	}
 
-	if (seq == journal_cur_seq(j)) {
-		bool set_need_write = false;
-
-		buf = journal_cur_buf(j);
-
-		if (parent && !closure_wait(&buf->wait, parent))
-			BUG();
-
-		if (!test_and_set_bit(JOURNAL_NEED_WRITE, &j->flags)) {
-			j->need_write_time = local_clock();
-			set_need_write = true;
-		}
-
-		switch (journal_buf_switch(j, set_need_write)) {
-		case JOURNAL_ENTRY_ERROR:
-			if (parent)
-				closure_wake_up(&buf->wait);
-			break;
-		case JOURNAL_ENTRY_CLOSED:
-			/*
-			 * Journal entry hasn't been opened yet, but caller
-			 * claims it has something
-			 */
-			BUG();
-		case JOURNAL_ENTRY_INUSE:
-			break;
-		case JOURNAL_UNLOCKED:
-			return;
-		}
-	} else if (parent &&
-		   seq + 1 == journal_cur_seq(j) &&
-		   j->reservations.prev_buf_unwritten) {
-		buf = journal_prev_buf(j);
-
-		if (!closure_wait(&buf->wait, parent))
-			BUG();
-
-		smp_mb();
-
-		/* check if raced with write completion (or failure) */
-		if (!j->reservations.prev_buf_unwritten ||
-		    bch2_journal_error(j))
-			closure_wake_up(&buf->wait);
+	if (seq <= j->flushed_seq_ondisk) {
+		ret = 1;
+		goto out;
 	}
 
-	spin_unlock(&j->lock);
-}
+	/* if seq was written, but not flushed - flush a newer one instead */
+	seq = max(seq, journal_last_unwritten_seq(j));
 
-static int journal_seq_flushed(struct journal *j, u64 seq)
-{
-	struct journal_buf *buf;
-	int ret = 1;
+recheck_need_open:
+	if (seq > journal_cur_seq(j)) {
+		struct journal_res res = { 0 };
 
-	spin_lock(&j->lock);
-	BUG_ON(seq > journal_cur_seq(j));
+		if (journal_entry_is_open(j))
+			__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 
-	if (seq == journal_cur_seq(j)) {
-		bool set_need_write = false;
+		spin_unlock(&j->lock);
 
-		ret = 0;
+		ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
+		if (ret)
+			return ret;
 
-		buf = journal_cur_buf(j);
+		seq = res.seq;
+		buf = j->buf + (seq & JOURNAL_BUF_MASK);
+		buf->must_flush = true;
 
-		if (!test_and_set_bit(JOURNAL_NEED_WRITE, &j->flags)) {
-			j->need_write_time = local_clock();
-			set_need_write = true;
+		if (!buf->flush_time) {
+			buf->flush_time	= local_clock() ?: 1;
+			buf->expires = jiffies;
 		}
 
-		switch (journal_buf_switch(j, set_need_write)) {
-		case JOURNAL_ENTRY_ERROR:
-			ret = -EIO;
-			break;
-		case JOURNAL_ENTRY_CLOSED:
-			/*
-			 * Journal entry hasn't been opened yet, but caller
-			 * claims it has something
-			 */
+		if (parent && !closure_wait(&buf->wait, parent))
 			BUG();
-		case JOURNAL_ENTRY_INUSE:
-			break;
-		case JOURNAL_UNLOCKED:
-			return 0;
-		}
-	} else if (seq + 1 == journal_cur_seq(j) &&
-		   j->reservations.prev_buf_unwritten) {
-		ret = bch2_journal_error(j);
+
+		bch2_journal_res_put(j, &res);
+
+		spin_lock(&j->lock);
+		goto want_write;
 	}
 
-	spin_unlock(&j->lock);
+	/*
+	 * if write was kicked off without a flush, flush the next sequence
+	 * number instead
+	 */
+	buf = journal_seq_to_buf(j, seq);
+	if (buf->noflush) {
+		seq++;
+		goto recheck_need_open;
+	}
+
+	buf->must_flush = true;
 
+	if (parent && !closure_wait(&buf->wait, parent))
+		BUG();
+want_write:
+	if (seq == journal_cur_seq(j))
+		journal_entry_want_write(j);
+out:
+	spin_unlock(&j->lock);
 	return ret;
 }
 
@@ -617,88 +638,115 @@ int bch2_journal_flush_seq(struct journal *j, u64 seq)
 	u64 start_time = local_clock();
 	int ret, ret2;
 
-	ret = wait_event_killable(j->wait, (ret2 = journal_seq_flushed(j, seq)));
+	/*
+	 * Don't update time_stats when @seq is already flushed:
+	 */
+	if (seq <= j->flushed_seq_ondisk)
+		return 0;
+
+	ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 
-	bch2_time_stats_update(j->flush_seq_time, start_time);
+	if (!ret)
+		bch2_time_stats_update(j->flush_seq_time, start_time);
 
 	return ret ?: ret2 < 0 ? ret2 : 0;
 }
 
-/**
- * bch2_journal_meta_async - force a journal entry to be written
+/*
+ * bch2_journal_flush_async - if there is an open journal entry, or a journal
+ * still being written, write it and wait for the write to complete
  */
-void bch2_journal_meta_async(struct journal *j, struct closure *parent)
+void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 {
-	struct journal_res res;
-	unsigned u64s = jset_u64s(0);
+	bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent);
+}
 
-	memset(&res, 0, sizeof(res));
+int bch2_journal_flush(struct journal *j)
+{
+	return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
+}
 
-	bch2_journal_res_get(j, &res, u64s, u64s);
-	bch2_journal_res_put(j, &res);
+/*
+ * bch2_journal_noflush_seq - tell the journal not to issue any flushes before
+ * @seq
+ */
+bool bch2_journal_noflush_seq(struct journal *j, u64 seq)
+{
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	u64 unwritten_seq;
+	bool ret = false;
+
+	if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush)))
+		return false;
+
+	if (seq <= c->journal.flushed_seq_ondisk)
+		return false;
+
+	spin_lock(&j->lock);
+	if (seq <= c->journal.flushed_seq_ondisk)
+		goto out;
+
+	for (unwritten_seq = journal_last_unwritten_seq(j);
+	     unwritten_seq < seq;
+	     unwritten_seq++) {
+		struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq);
+
+		/* journal write is already in flight, and was a flush write: */
+		if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush)
+			goto out;
+
+		buf->noflush = true;
+	}
 
-	bch2_journal_flush_seq_async(j, res.seq, parent);
+	ret = true;
+out:
+	spin_unlock(&j->lock);
+	return ret;
 }
 
 int bch2_journal_meta(struct journal *j)
 {
+	struct journal_buf *buf;
 	struct journal_res res;
-	unsigned u64s = jset_u64s(0);
 	int ret;
 
 	memset(&res, 0, sizeof(res));
 
-	ret = bch2_journal_res_get(j, &res, u64s, u64s);
+	ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 	if (ret)
 		return ret;
 
+	buf = j->buf + (res.seq & JOURNAL_BUF_MASK);
+	buf->must_flush = true;
+
+	if (!buf->flush_time) {
+		buf->flush_time	= local_clock() ?: 1;
+		buf->expires = jiffies;
+	}
+
 	bch2_journal_res_put(j, &res);
 
 	return bch2_journal_flush_seq(j, res.seq);
 }
 
-/*
- * bch2_journal_flush_async - if there is an open journal entry, or a journal
- * still being written, write it and wait for the write to complete
- */
-void bch2_journal_flush_async(struct journal *j, struct closure *parent)
-{
-	u64 seq, journal_seq;
+/* block/unlock the journal: */
 
+void bch2_journal_unblock(struct journal *j)
+{
 	spin_lock(&j->lock);
-	journal_seq = journal_cur_seq(j);
-
-	if (journal_entry_is_open(j)) {
-		seq = journal_seq;
-	} else if (journal_seq) {
-		seq = journal_seq - 1;
-	} else {
-		spin_unlock(&j->lock);
-		return;
-	}
+	j->blocked--;
 	spin_unlock(&j->lock);
 
-	bch2_journal_flush_seq_async(j, seq, parent);
+	journal_wake(j);
 }
 
-int bch2_journal_flush(struct journal *j)
+void bch2_journal_block(struct journal *j)
 {
-	u64 seq, journal_seq;
-
 	spin_lock(&j->lock);
-	journal_seq = journal_cur_seq(j);
-
-	if (journal_entry_is_open(j)) {
-		seq = journal_seq;
-	} else if (journal_seq) {
-		seq = journal_seq - 1;
-	} else {
-		spin_unlock(&j->lock);
-		return 0;
-	}
+	j->blocked++;
 	spin_unlock(&j->lock);
 
-	return bch2_journal_flush_seq(j, seq);
+	journal_quiesce(j);
 }
 
 /* allocate journal on a device: */
@@ -708,101 +756,129 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 {
 	struct bch_fs *c = ca->fs;
 	struct journal_device *ja = &ca->journal;
-	struct bch_sb_field_journal *journal_buckets;
 	u64 *new_bucket_seq = NULL, *new_buckets = NULL;
+	struct open_bucket **ob = NULL;
+	long *bu = NULL;
+	unsigned i, pos, nr_got = 0, nr_want = nr - ja->nr;
 	int ret = 0;
 
-	/* don't handle reducing nr of buckets yet: */
-	if (nr <= ja->nr)
-		return 0;
+	BUG_ON(nr <= ja->nr);
 
-	ret = -ENOMEM;
-	new_buckets	= kzalloc(nr * sizeof(u64), GFP_KERNEL);
-	new_bucket_seq	= kzalloc(nr * sizeof(u64), GFP_KERNEL);
-	if (!new_buckets || !new_bucket_seq)
-		goto err;
+	bu		= kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
+	ob		= kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
+	new_buckets	= kcalloc(nr, sizeof(u64), GFP_KERNEL);
+	new_bucket_seq	= kcalloc(nr, sizeof(u64), GFP_KERNEL);
+	if (!bu || !ob || !new_buckets || !new_bucket_seq) {
+		ret = -ENOMEM;
+		goto err_free;
+	}
 
-	journal_buckets = bch2_sb_resize_journal(&ca->disk_sb,
-				nr + sizeof(*journal_buckets) / sizeof(u64));
-	if (!journal_buckets)
-		goto err;
+	for (nr_got = 0; nr_got < nr_want; nr_got++) {
+		if (new_fs) {
+			bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
+			if (bu[nr_got] < 0) {
+				ret = -BCH_ERR_ENOSPC_bucket_alloc;
+				break;
+			}
+		} else {
+			ob[nr_got] = bch2_bucket_alloc(c, ca, RESERVE_none, cl);
+			ret = PTR_ERR_OR_ZERO(ob[nr_got]);
+			if (ret)
+				break;
+
+			ret = bch2_trans_run(c,
+				bch2_trans_mark_metadata_bucket(&trans, ca,
+						ob[nr_got]->bucket, BCH_DATA_journal,
+						ca->mi.bucket_size));
+			if (ret) {
+				bch2_open_bucket_put(c, ob[nr_got]);
+				bch_err(c, "error marking new journal buckets: %s", bch2_err_str(ret));
+				break;
+			}
 
-	if (c)
-		spin_lock(&c->journal.lock);
+			bu[nr_got] = ob[nr_got]->bucket;
+		}
+	}
+
+	if (!nr_got)
+		goto err_free;
+
+	/* Don't return an error if we successfully allocated some buckets: */
+	ret = 0;
+
+	if (c) {
+		bch2_journal_flush_all_pins(&c->journal);
+		bch2_journal_block(&c->journal);
+		mutex_lock(&c->sb_lock);
+	}
 
 	memcpy(new_buckets,	ja->buckets,	ja->nr * sizeof(u64));
 	memcpy(new_bucket_seq,	ja->bucket_seq,	ja->nr * sizeof(u64));
-	swap(new_buckets,	ja->buckets);
-	swap(new_bucket_seq,	ja->bucket_seq);
 
-	if (c)
-		spin_unlock(&c->journal.lock);
+	BUG_ON(ja->discard_idx > ja->nr);
 
-	while (ja->nr < nr) {
-		struct open_bucket *ob = NULL;
-		long bucket;
+	pos = ja->discard_idx ?: ja->nr;
 
-		if (new_fs) {
-			percpu_down_read_preempt_disable(&c->usage_lock);
-			bucket = bch2_bucket_alloc_new_fs(ca);
-			percpu_up_read_preempt_enable(&c->usage_lock);
+	memmove(new_buckets + pos + nr_got,
+		new_buckets + pos,
+		sizeof(new_buckets[0]) * (ja->nr - pos));
+	memmove(new_bucket_seq + pos + nr_got,
+		new_bucket_seq + pos,
+		sizeof(new_bucket_seq[0]) * (ja->nr - pos));
 
-			if (bucket < 0) {
-				ret = -ENOSPC;
-				goto err;
-			}
-		} else {
-			int ob_idx = bch2_bucket_alloc(c, ca, RESERVE_ALLOC, false, cl);
-			if (ob_idx < 0) {
-				ret = cl ? -EAGAIN : -ENOSPC;
-				goto err;
-			}
+	for (i = 0; i < nr_got; i++) {
+		new_buckets[pos + i] = bu[i];
+		new_bucket_seq[pos + i] = 0;
+	}
 
-			ob = c->open_buckets + ob_idx;
-			bucket = sector_to_bucket(ca, ob->ptr.offset);
-		}
+	nr = ja->nr + nr_got;
 
-		if (c) {
-			percpu_down_read_preempt_disable(&c->usage_lock);
-			spin_lock(&c->journal.lock);
-		}
+	ret = bch2_journal_buckets_to_sb(c, ca, new_buckets, nr);
+	if (ret)
+		goto err_unblock;
 
-		__array_insert_item(ja->buckets,		ja->nr, ja->last_idx);
-		__array_insert_item(ja->bucket_seq,		ja->nr, ja->last_idx);
-		__array_insert_item(journal_buckets->buckets,	ja->nr, ja->last_idx);
+	if (!new_fs)
+		bch2_write_super(c);
 
-		ja->buckets[ja->last_idx] = bucket;
-		ja->bucket_seq[ja->last_idx] = 0;
-		journal_buckets->buckets[ja->last_idx] = cpu_to_le64(bucket);
+	/* Commit: */
+	if (c)
+		spin_lock(&c->journal.lock);
 
-		if (ja->last_idx < ja->nr) {
-			if (ja->cur_idx >= ja->last_idx)
-				ja->cur_idx++;
-			ja->last_idx++;
-		}
-		ja->nr++;
-
-		bch2_mark_metadata_bucket(c, ca, bucket, BCH_DATA_JOURNAL,
-				ca->mi.bucket_size,
-				gc_phase(GC_PHASE_SB),
-				new_fs
-				? BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE
-				: 0);
-
-		if (c) {
-			spin_unlock(&c->journal.lock);
-			percpu_up_read_preempt_enable(&c->usage_lock);
-		}
+	swap(new_buckets,	ja->buckets);
+	swap(new_bucket_seq,	ja->bucket_seq);
+	ja->nr = nr;
+
+	if (pos <= ja->discard_idx)
+		ja->discard_idx = (ja->discard_idx + nr_got) % ja->nr;
+	if (pos <= ja->dirty_idx_ondisk)
+		ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + nr_got) % ja->nr;
+	if (pos <= ja->dirty_idx)
+		ja->dirty_idx = (ja->dirty_idx + nr_got) % ja->nr;
+	if (pos <= ja->cur_idx)
+		ja->cur_idx = (ja->cur_idx + nr_got) % ja->nr;
 
-		if (!new_fs)
-			bch2_open_bucket_put(c, ob);
+	if (c)
+		spin_unlock(&c->journal.lock);
+err_unblock:
+	if (c) {
+		bch2_journal_unblock(&c->journal);
+		mutex_unlock(&c->sb_lock);
 	}
 
-	ret = 0;
-err:
+	if (ret && !new_fs)
+		for (i = 0; i < nr_got; i++)
+			bch2_trans_run(c,
+				bch2_trans_mark_metadata_bucket(&trans, ca,
+						bu[i], BCH_DATA_free, 0));
+err_free:
+	if (!new_fs)
+		for (i = 0; i < nr_got; i++)
+			bch2_open_bucket_put(c, ob[i]);
+
 	kfree(new_bucket_seq);
 	kfree(new_buckets);
-
+	kfree(ob);
+	kfree(bu);
 	return ret;
 }
 
@@ -815,41 +891,49 @@ int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 {
 	struct journal_device *ja = &ca->journal;
 	struct closure cl;
-	unsigned current_nr;
-	int ret;
+	int ret = 0;
 
 	closure_init_stack(&cl);
 
-	do {
-		struct disk_reservation disk_res = { 0, 0 };
+	down_write(&c->state_lock);
 
-		closure_sync(&cl);
+	/* don't handle reducing nr of buckets yet: */
+	if (nr < ja->nr)
+		goto unlock;
 
-		mutex_lock(&c->sb_lock);
-		current_nr = ja->nr;
+	while (ja->nr < nr) {
+		struct disk_reservation disk_res = { 0, 0 };
 
 		/*
 		 * note: journal buckets aren't really counted as _sectors_ used yet, so
 		 * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
 		 * when space used goes up without a reservation - but we do need the
 		 * reservation to ensure we'll actually be able to allocate:
+		 *
+		 * XXX: that's not right, disk reservations only ensure a
+		 * filesystem-wide allocation will succeed, this is a device
+		 * specific allocation - we can hang here:
 		 */
 
-		if (bch2_disk_reservation_get(c, &disk_res,
-				bucket_to_sector(ca, nr - ja->nr), 1, 0)) {
-			mutex_unlock(&c->sb_lock);
-			return -ENOSPC;
-		}
+		ret = bch2_disk_reservation_get(c, &disk_res,
+						bucket_to_sector(ca, nr - ja->nr), 1, 0);
+		if (ret)
+			break;
 
 		ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 
 		bch2_disk_reservation_put(c, &disk_res);
 
-		if (ja->nr != current_nr)
-			bch2_write_super(c);
-		mutex_unlock(&c->sb_lock);
-	} while (ret == -EAGAIN);
+		closure_sync(&cl);
 
+		if (ret && ret != -BCH_ERR_bucket_alloc_blocked)
+			break;
+	}
+
+	if (ret)
+		bch_err(c, "%s: err %s", __func__, bch2_err_str(ret));
+unlock:
+	up_write(&c->state_lock);
 	return ret;
 }
 
@@ -860,14 +944,17 @@ int bch2_dev_journal_alloc(struct bch_dev *ca)
 	if (dynamic_fault("bcachefs:add:journal_alloc"))
 		return -ENOMEM;
 
+	/* 1/128th of the device by default: */
+	nr = ca->mi.nbuckets >> 7;
+
 	/*
-	 * clamp journal size to 1024 buckets or 512MB (in sectors), whichever
+	 * clamp journal size to 8192 buckets or 8GB (in sectors), whichever
 	 * is smaller:
 	 */
-	nr = clamp_t(unsigned, ca->mi.nbuckets >> 8,
+	nr = clamp_t(unsigned, nr,
 		     BCH_JOURNAL_BUCKETS_MIN,
-		     min(1 << 10,
-			 (1 << 20) / ca->mi.bucket_size));
+		     min(1 << 13,
+			 (1 << 24) / ca->mi.bucket_size));
 
 	return __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
 }
@@ -876,16 +963,18 @@ int bch2_dev_journal_alloc(struct bch_dev *ca)
 
 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
 {
-	union journal_res_state state;
-	struct journal_buf *w;
-	bool ret;
+	bool ret = false;
+	u64 seq;
 
 	spin_lock(&j->lock);
-	state = READ_ONCE(j->reservations);
-	w = j->buf + !state.idx;
+	for (seq = journal_last_unwritten_seq(j);
+	     seq <= journal_cur_seq(j) && !ret;
+	     seq++) {
+		struct journal_buf *buf = journal_seq_to_buf(j, seq);
 
-	ret = state.prev_buf_unwritten &&
-		bch2_extent_has_device(bkey_i_to_s_c_extent(&w->key), dev_idx);
+		if (bch2_bkey_has_device_c(bkey_i_to_s_c(&buf->key), dev_idx))
+			ret = true;
+	}
 	spin_unlock(&j->lock);
 
 	return ret;
@@ -893,64 +982,115 @@ static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
 
 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
 {
-	spin_lock(&j->lock);
-	bch2_extent_drop_device(bkey_i_to_s_extent(&j->key), ca->dev_idx);
-	spin_unlock(&j->lock);
-
 	wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
 }
 
 void bch2_fs_journal_stop(struct journal *j)
 {
-	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	bch2_journal_reclaim_stop(j);
+	bch2_journal_flush_all_pins(j);
 
-	wait_event(j->wait, journal_flush_write(j));
+	wait_event(j->wait, journal_entry_close(j));
 
-	/* do we need to write another journal entry? */
-	if (test_bit(JOURNAL_NOT_EMPTY, &j->flags) ||
-	    c->btree_roots_dirty)
-		bch2_journal_meta(j);
+	/*
+	 * Always write a new journal entry, to make sure the clock hands are up
+	 * to date (and match the superblock)
+	 */
+	bch2_journal_meta(j);
+
+	journal_quiesce(j);
 
 	BUG_ON(!bch2_journal_error(j) &&
-	       test_bit(JOURNAL_NOT_EMPTY, &j->flags));
+	       test_bit(JOURNAL_REPLAY_DONE, &j->flags) &&
+	       j->last_empty_seq != journal_cur_seq(j));
 
 	cancel_delayed_work_sync(&j->write_work);
-	cancel_delayed_work_sync(&j->reclaim_work);
 }
 
-void bch2_fs_journal_start(struct journal *j)
+int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
 {
-	struct journal_seq_blacklist *bl;
-	u64 blacklist = 0;
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	struct journal_entry_pin_list *p;
+	struct journal_replay *i, **_i;
+	struct genradix_iter iter;
+	bool had_entries = false;
+	unsigned ptr;
+	u64 last_seq = cur_seq, nr, seq;
 
-	list_for_each_entry(bl, &j->seq_blacklist, list)
-		blacklist = max(blacklist, bl->end);
+	genradix_for_each_reverse(&c->journal_entries, iter, _i) {
+		i = *_i;
+
+		if (!i || i->ignore)
+			continue;
+
+		last_seq = le64_to_cpu(i->j.last_seq);
+		break;
+	}
+
+	nr = cur_seq - last_seq;
+
+	if (nr + 1 > j->pin.size) {
+		free_fifo(&j->pin);
+		init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
+		if (!j->pin.data) {
+			bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
+			return -ENOMEM;
+		}
+	}
+
+	j->replay_journal_seq	= last_seq;
+	j->replay_journal_seq_end = cur_seq;
+	j->last_seq_ondisk	= last_seq;
+	j->flushed_seq_ondisk	= cur_seq - 1;
+	j->seq_ondisk		= cur_seq - 1;
+	j->pin.front		= last_seq;
+	j->pin.back		= cur_seq;
+	atomic64_set(&j->seq, cur_seq - 1);
+
+	fifo_for_each_entry_ptr(p, &j->pin, seq)
+		journal_pin_list_init(p, 1);
+
+	genradix_for_each(&c->journal_entries, iter, _i) {
+		i = *_i;
+
+		if (!i || i->ignore)
+			continue;
+
+		seq = le64_to_cpu(i->j.seq);
+		BUG_ON(seq >= cur_seq);
+
+		if (seq < last_seq)
+			continue;
+
+		if (journal_entry_empty(&i->j))
+			j->last_empty_seq = le64_to_cpu(i->j.seq);
+
+		p = journal_seq_pin(j, seq);
+
+		p->devs.nr = 0;
+		for (ptr = 0; ptr < i->nr_ptrs; ptr++)
+			bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
+
+		had_entries = true;
+	}
+
+	if (!had_entries)
+		j->last_empty_seq = cur_seq;
 
 	spin_lock(&j->lock);
 
 	set_bit(JOURNAL_STARTED, &j->flags);
+	j->last_flush_write = jiffies;
 
-	while (journal_cur_seq(j) < blacklist)
-		journal_pin_new_entry(j, 0);
+	j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j);
+	j->reservations.unwritten_idx++;
 
-	/*
-	 * journal_buf_switch() only inits the next journal entry when it
-	 * closes an open journal entry - the very first journal entry gets
-	 * initialized here:
-	 */
-	journal_pin_new_entry(j, 1);
-	bch2_journal_buf_init(j);
+	c->last_bucket_seq_cleanup = journal_cur_seq(j);
 
+	bch2_journal_space_available(j);
 	spin_unlock(&j->lock);
 
-	/*
-	 * Adding entries to the next journal entry before allocating space on
-	 * disk for the next journal entry - this is ok, because these entries
-	 * only have to go down with the next journal entry we write:
-	 */
-	bch2_journal_seq_blacklist_write(j);
-
-	queue_delayed_work(system_freezable_wq, &j->reclaim_work, 0);
+	return bch2_journal_reclaim_start(j);
 }
 
 /* init/exit: */
@@ -971,33 +1111,61 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 	struct journal_device *ja = &ca->journal;
 	struct bch_sb_field_journal *journal_buckets =
 		bch2_sb_get_journal(sb);
-	unsigned i;
+	struct bch_sb_field_journal_v2 *journal_buckets_v2 =
+		bch2_sb_get_journal_v2(sb);
+	unsigned i, nr_bvecs;
+
+	ja->nr = 0;
 
-	ja->nr = bch2_nr_journal_buckets(journal_buckets);
+	if (journal_buckets_v2) {
+		unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
+
+		for (i = 0; i < nr; i++)
+			ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
+	} else if (journal_buckets) {
+		ja->nr = bch2_nr_journal_buckets(journal_buckets);
+	}
 
 	ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
 	if (!ja->bucket_seq)
 		return -ENOMEM;
 
-	ca->journal.bio = bio_kmalloc(GFP_KERNEL,
-			DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE));
+	nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
+
+	ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 	if (!ca->journal.bio)
 		return -ENOMEM;
 
+	bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
+
 	ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
 	if (!ja->buckets)
 		return -ENOMEM;
 
-	for (i = 0; i < ja->nr; i++)
-		ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
+	if (journal_buckets_v2) {
+		unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
+		unsigned j, dst = 0;
+
+		for (i = 0; i < nr; i++)
+			for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
+				ja->buckets[dst++] =
+					le64_to_cpu(journal_buckets_v2->d[i].start) + j;
+	} else if (journal_buckets) {
+		for (i = 0; i < ja->nr; i++)
+			ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
+	}
 
 	return 0;
 }
 
 void bch2_fs_journal_exit(struct journal *j)
 {
-	kvpfree(j->buf[1].data, j->buf[1].size);
-	kvpfree(j->buf[0].data, j->buf[0].size);
+	unsigned i;
+
+	darray_exit(&j->early_journal_entries);
+
+	for (i = 0; i < ARRAY_SIZE(j->buf); i++)
+		kvpfree(j->buf[i].data, j->buf[i].buf_size);
 	free_fifo(&j->pin);
 }
 
@@ -1005,6 +1173,7 @@ int bch2_fs_journal_init(struct journal *j)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	static struct lock_class_key res_key;
+	unsigned i;
 	int ret = 0;
 
 	pr_verbose_init(c->opts, "");
@@ -1013,31 +1182,31 @@ int bch2_fs_journal_init(struct journal *j)
 	spin_lock_init(&j->err_lock);
 	init_waitqueue_head(&j->wait);
 	INIT_DELAYED_WORK(&j->write_work, journal_write_work);
-	INIT_DELAYED_WORK(&j->reclaim_work, bch2_journal_reclaim_work);
-	mutex_init(&j->blacklist_lock);
-	INIT_LIST_HEAD(&j->seq_blacklist);
+	init_waitqueue_head(&j->reclaim_wait);
+	init_waitqueue_head(&j->pin_flush_wait);
 	mutex_init(&j->reclaim_lock);
+	mutex_init(&j->discard_lock);
 
 	lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
 
-	j->buf[0].size		= JOURNAL_ENTRY_SIZE_MIN;
-	j->buf[1].size		= JOURNAL_ENTRY_SIZE_MIN;
-	j->write_delay_ms	= 1000;
-	j->reclaim_delay_ms	= 100;
-
-	bkey_extent_init(&j->key);
-
 	atomic64_set(&j->reservations.counter,
 		((union journal_res_state)
 		 { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);
 
-	if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)) ||
-	    !(j->buf[0].data = kvpmalloc(j->buf[0].size, GFP_KERNEL)) ||
-	    !(j->buf[1].data = kvpmalloc(j->buf[1].size, GFP_KERNEL))) {
+	if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL))) {
 		ret = -ENOMEM;
 		goto out;
 	}
 
+	for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
+		j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
+		j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
+		if (!j->buf[i].data) {
+			ret = -ENOMEM;
+			goto out;
+		}
+	}
+
 	j->pin.front = j->pin.back = 1;
 out:
 	pr_verbose_init(c->opts, "ret %i", ret);
@@ -1046,92 +1215,185 @@ out:
 
 /* debug: */
 
-ssize_t bch2_journal_print_debug(struct journal *j, char *buf)
+void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
-	union journal_res_state *s = &j->reservations;
+	union journal_res_state s;
 	struct bch_dev *ca;
-	unsigned iter;
-	ssize_t ret = 0;
+	unsigned long now = jiffies;
+	u64 seq;
+	unsigned i;
+
+	if (!out->nr_tabstops)
+		printbuf_tabstop_push(out, 24);
+	out->atomic++;
 
 	rcu_read_lock();
-	spin_lock(&j->lock);
+	s = READ_ONCE(j->reservations);
+
+	prt_printf(out, "dirty journal entries:\t%llu/%llu\n",	fifo_used(&j->pin), j->pin.size);
+	prt_printf(out, "seq:\t\t\t%llu\n",			journal_cur_seq(j));
+	prt_printf(out, "seq_ondisk:\t\t%llu\n",		j->seq_ondisk);
+	prt_printf(out, "last_seq:\t\t%llu\n",		journal_last_seq(j));
+	prt_printf(out, "last_seq_ondisk:\t%llu\n",		j->last_seq_ondisk);
+	prt_printf(out, "flushed_seq_ondisk:\t%llu\n",	j->flushed_seq_ondisk);
+	prt_printf(out, "prereserved:\t\t%u/%u\n",		j->prereserved.reserved, j->prereserved.remaining);
+	prt_printf(out, "watermark:\t\t%s\n",		bch2_journal_watermarks[j->watermark]);
+	prt_printf(out, "each entry reserved:\t%u\n",	j->entry_u64s_reserved);
+	prt_printf(out, "nr flush writes:\t%llu\n",		j->nr_flush_writes);
+	prt_printf(out, "nr noflush writes:\t%llu\n",	j->nr_noflush_writes);
+	prt_printf(out, "nr direct reclaim:\t%llu\n",	j->nr_direct_reclaim);
+	prt_printf(out, "nr background reclaim:\t%llu\n",	j->nr_background_reclaim);
+	prt_printf(out, "reclaim kicked:\t\t%u\n",		j->reclaim_kicked);
+	prt_printf(out, "reclaim runs in:\t%u ms\n",	time_after(j->next_reclaim, now)
+	       ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
+	prt_printf(out, "current entry sectors:\t%u\n",	j->cur_entry_sectors);
+	prt_printf(out, "current entry error:\t%s\n",	bch2_journal_errors[j->cur_entry_error]);
+	prt_printf(out, "current entry:\t\t");
+
+	switch (s.cur_entry_offset) {
+	case JOURNAL_ENTRY_ERROR_VAL:
+		prt_printf(out, "error");
+		break;
+	case JOURNAL_ENTRY_CLOSED_VAL:
+		prt_printf(out, "closed");
+		break;
+	default:
+		prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s);
+		break;
+	}
+
+	prt_newline(out);
+
+	for (seq = journal_cur_seq(j);
+	     seq >= journal_last_unwritten_seq(j);
+	     --seq) {
+		i = seq & JOURNAL_BUF_MASK;
+
+		prt_printf(out, "unwritten entry:");
+		prt_tab(out);
+		prt_printf(out, "%llu", seq);
+		prt_newline(out);
+		printbuf_indent_add(out, 2);
+
+		prt_printf(out, "refcount:");
+		prt_tab(out);
+		prt_printf(out, "%u", journal_state_count(s, i));
+		prt_newline(out);
+
+		prt_printf(out, "sectors:");
+		prt_tab(out);
+		prt_printf(out, "%u", j->buf[i].sectors);
+		prt_newline(out);
+
+		prt_printf(out, "expires");
+		prt_tab(out);
+		prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies);
+		prt_newline(out);
+
+		printbuf_indent_sub(out, 2);
+	}
 
-	ret += scnprintf(buf + ret, PAGE_SIZE - ret,
-			 "active journal entries:\t%llu\n"
-			 "seq:\t\t\t%llu\n"
-			 "last_seq:\t\t%llu\n"
-			 "last_seq_ondisk:\t%llu\n"
-			 "reservation count:\t%u\n"
-			 "reservation offset:\t%u\n"
-			 "current entry u64s:\t%u\n"
-			 "io in flight:\t\t%i\n"
-			 "need write:\t\t%i\n"
-			 "dirty:\t\t\t%i\n"
-			 "replay done:\t\t%i\n",
-			 fifo_used(&j->pin),
-			 journal_cur_seq(j),
-			 journal_last_seq(j),
-			 j->last_seq_ondisk,
-			 journal_state_count(*s, s->idx),
-			 s->cur_entry_offset,
-			 j->cur_entry_u64s,
-			 s->prev_buf_unwritten,
-			 test_bit(JOURNAL_NEED_WRITE,	&j->flags),
-			 journal_entry_is_open(j),
-			 test_bit(JOURNAL_REPLAY_DONE,	&j->flags));
-
-	for_each_member_device_rcu(ca, c, iter,
-				   &c->rw_devs[BCH_DATA_JOURNAL]) {
+	prt_printf(out,
+	       "replay done:\t\t%i\n",
+	       test_bit(JOURNAL_REPLAY_DONE,	&j->flags));
+
+	prt_printf(out, "space:\n");
+	prt_printf(out, "\tdiscarded\t%u:%u\n",
+	       j->space[journal_space_discarded].next_entry,
+	       j->space[journal_space_discarded].total);
+	prt_printf(out, "\tclean ondisk\t%u:%u\n",
+	       j->space[journal_space_clean_ondisk].next_entry,
+	       j->space[journal_space_clean_ondisk].total);
+	prt_printf(out, "\tclean\t\t%u:%u\n",
+	       j->space[journal_space_clean].next_entry,
+	       j->space[journal_space_clean].total);
+	prt_printf(out, "\ttotal\t\t%u:%u\n",
+	       j->space[journal_space_total].next_entry,
+	       j->space[journal_space_total].total);
+
+	for_each_member_device_rcu(ca, c, i,
+				   &c->rw_devs[BCH_DATA_journal]) {
 		struct journal_device *ja = &ca->journal;
 
+		if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d))
+			continue;
+
 		if (!ja->nr)
 			continue;
 
-		ret += scnprintf(buf + ret, PAGE_SIZE - ret,
-				 "dev %u:\n"
-				 "\tnr\t\t%u\n"
-				 "\tcur_idx\t\t%u (seq %llu)\n"
-				 "\tlast_idx\t%u (seq %llu)\n",
-				 iter, ja->nr,
-				 ja->cur_idx,	ja->bucket_seq[ja->cur_idx],
-				 ja->last_idx,	ja->bucket_seq[ja->last_idx]);
+		prt_printf(out, "dev %u:\n",		i);
+		prt_printf(out, "\tnr\t\t%u\n",		ja->nr);
+		prt_printf(out, "\tbucket size\t%u\n",	ca->mi.bucket_size);
+		prt_printf(out, "\tavailable\t%u:%u\n",	bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free);
+		prt_printf(out, "\tdiscard_idx\t%u\n",	ja->discard_idx);
+		prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk,	ja->bucket_seq[ja->dirty_idx_ondisk]);
+		prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx,		ja->bucket_seq[ja->dirty_idx]);
+		prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx,		ja->bucket_seq[ja->cur_idx]);
 	}
 
-	spin_unlock(&j->lock);
 	rcu_read_unlock();
 
-	return ret;
+	--out->atomic;
 }
 
-ssize_t bch2_journal_print_pins(struct journal *j, char *buf)
+void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
+{
+	spin_lock(&j->lock);
+	__bch2_journal_debug_to_text(out, j);
+	spin_unlock(&j->lock);
+}
+
+bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
 {
 	struct journal_entry_pin_list *pin_list;
 	struct journal_entry_pin *pin;
-	ssize_t ret = 0;
-	u64 i;
+	unsigned i;
 
 	spin_lock(&j->lock);
-	fifo_for_each_entry_ptr(pin_list, &j->pin, i) {
-		ret += scnprintf(buf + ret, PAGE_SIZE - ret,
-				 "%llu: count %u\n",
-				 i, atomic_read(&pin_list->count));
-
-		list_for_each_entry(pin, &pin_list->list, list)
-			ret += scnprintf(buf + ret, PAGE_SIZE - ret,
-					 "\t%p %pf\n",
-					 pin, pin->flush);
-
-		if (!list_empty(&pin_list->flushed))
-			ret += scnprintf(buf + ret, PAGE_SIZE - ret,
-					 "flushed:\n");
-
-		list_for_each_entry(pin, &pin_list->flushed, list)
-			ret += scnprintf(buf + ret, PAGE_SIZE - ret,
-					 "\t%p %pf\n",
-					 pin, pin->flush);
+	*seq = max(*seq, j->pin.front);
+
+	if (*seq >= j->pin.back) {
+		spin_unlock(&j->lock);
+		return true;
 	}
+
+	out->atomic++;
+
+	pin_list = journal_seq_pin(j, *seq);
+
+	prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count));
+	prt_newline(out);
+	printbuf_indent_add(out, 2);
+
+	for (i = 0; i < ARRAY_SIZE(pin_list->list); i++)
+		list_for_each_entry(pin, &pin_list->list[i], list) {
+			prt_printf(out, "\t%px %ps", pin, pin->flush);
+			prt_newline(out);
+		}
+
+	if (!list_empty(&pin_list->flushed)) {
+		prt_printf(out, "flushed:");
+		prt_newline(out);
+	}
+
+	list_for_each_entry(pin, &pin_list->flushed, list) {
+		prt_printf(out, "\t%px %ps", pin, pin->flush);
+		prt_newline(out);
+	}
+
+	printbuf_indent_sub(out, 2);
+
+	--out->atomic;
 	spin_unlock(&j->lock);
 
-	return ret;
+	return false;
+}
+
+void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
+{
+	u64 seq = 0;
+
+	while (!bch2_journal_seq_pins_to_text(out, j, &seq))
+		seq++;
 }