Update bcachefs sources to 75e8a078b8 bcachefs: improved flush_held_btree_writes()

1 files changed, 232 insertions, 214 deletions

diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c
index 8ff8cfa8..f108a282 100644
--- a/libbcachefs/journal.c
+++ b/libbcachefs/journal.c
@@ -17,23 +17,14 @@
 
 #include <trace/events/bcachefs.h>
 
-static bool journal_entry_is_open(struct journal *j)
+static bool __journal_entry_is_open(union journal_res_state state)
 {
-	return j->reservations.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
+	return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
 }
 
-void bch2_journal_buf_put_slowpath(struct journal *j, bool need_write_just_set)
+static bool journal_entry_is_open(struct journal *j)
 {
-	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);
-
-	closure_call(&j->io, bch2_journal_write, system_highpri_wq, NULL);
+	return __journal_entry_is_open(j->reservations);
 }
 
 static void journal_pin_new_entry(struct journal *j, int count)
@@ -77,39 +68,71 @@ static inline bool journal_entry_empty(struct jset *j)
 	return true;
 }
 
-static enum {
-	JOURNAL_ENTRY_ERROR,
-	JOURNAL_ENTRY_INUSE,
-	JOURNAL_ENTRY_CLOSED,
-	JOURNAL_UNLOCKED,
-} journal_buf_switch(struct journal *j, bool need_write_just_set)
+void bch2_journal_halt(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;
+
+		new.cur_entry_offset = JOURNAL_ENTRY_ERROR_VAL;
+	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
+				       old.v, new.v)) != old.v);
+
+	journal_wake(j);
+	closure_wake_up(&journal_cur_buf(j)->wait);
+}
+
+/* journal entry close/open: */
+
+void __bch2_journal_buf_put(struct journal *j, bool need_write_just_set)
+{
+	if (!need_write_just_set &&
+	    test_bit(JOURNAL_NEED_WRITE, &j->flags))
+		bch2_time_stats_update(j->delay_time,
+				       j->need_write_time);
+
+	clear_bit(JOURNAL_NEED_WRITE, &j->flags);
+
+	closure_call(&j->io, bch2_journal_write, system_highpri_wq, NULL);
+}
+
+/*
+ * Returns true if journal entry is now closed:
+ */
+static bool __journal_entry_close(struct journal *j)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_buf *buf = journal_cur_buf(j);
 	union journal_res_state old, new;
 	u64 v = atomic64_read(&j->reservations.counter);
+	bool set_need_write = false;
+	unsigned sectors;
 
 	lockdep_assert_held(&j->lock);
 
 	do {
 		old.v = new.v = v;
 		if (old.cur_entry_offset == JOURNAL_ENTRY_CLOSED_VAL)
-			return JOURNAL_ENTRY_CLOSED;
+			return true;
 
 		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL) {
 			/* this entry will never be written: */
 			closure_wake_up(&buf->wait);
-			return JOURNAL_ENTRY_ERROR;
+			return true;
 		}
 
-		if (new.prev_buf_unwritten)
-			return JOURNAL_ENTRY_INUSE;
+		if (!test_bit(JOURNAL_NEED_WRITE, &j->flags)) {
+			set_bit(JOURNAL_NEED_WRITE, &j->flags);
+			j->need_write_time = local_clock();
+			set_need_write = true;
+		}
 
-		/*
-		 * avoid race between setting buf->data->u64s and
-		 * journal_res_put starting write:
-		 */
-		journal_state_inc(&new);
+		if (new.prev_buf_unwritten)
+			return false;
 
 		new.cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL;
 		new.idx++;
@@ -119,15 +142,12 @@ static enum {
 	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
 				       old.v, new.v)) != old.v);
 
-	clear_bit(JOURNAL_NEED_WRITE, &j->flags);
-
 	buf->data->u64s		= cpu_to_le32(old.cur_entry_offset);
 
-	j->prev_buf_sectors =
-		vstruct_blocks_plus(buf->data, c->block_bits,
-				    buf->u64s_reserved) *
-		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;
 
 	bkey_extent_init(&buf->key);
 
@@ -150,7 +170,6 @@ static enum {
 	 * Hence, we want update/set last_seq on the current journal entry right
 	 * before we open a new one:
 	 */
-	bch2_journal_reclaim_fast(j);
 	buf->data->last_seq	= cpu_to_le64(journal_last_seq(j));
 
 	if (journal_entry_empty(buf->data))
@@ -163,32 +182,22 @@ static enum {
 	bch2_journal_buf_init(j);
 
 	cancel_delayed_work(&j->write_work);
-	spin_unlock(&j->lock);
 
-	/* 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);
+	bch2_journal_space_available(j);
 
-	return JOURNAL_UNLOCKED;
+	bch2_journal_buf_put(j, old.idx, set_need_write);
+	return true;
 }
 
-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_close(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;
 }
 
 /*
@@ -196,46 +205,39 @@ void bch2_journal_halt(struct journal *j)
  * 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
+ * 0:		success
+ * -ENOSPC:	journal currently full, must invoke reclaim
+ * -EAGAIN:	journal blocked, must wait
+ * -EROFS:	insufficient rw devices or journal error
  */
 static int journal_entry_open(struct journal *j)
 {
 	struct journal_buf *buf = journal_cur_buf(j);
 	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));
 
-	if (!fifo_free(&j->pin))
-		return 0;
+	if (j->blocked)
+		return -EAGAIN;
 
-	sectors = bch2_journal_entry_sectors(j);
-	if (sectors <= 0)
-		return sectors;
+	if (j->cur_entry_error)
+		return j->cur_entry_error;
 
-	buf->disk_sectors	= sectors;
-	buf->u64s_reserved	= j->entry_u64s_reserved;
+	BUG_ON(!j->cur_entry_sectors);
 
-	sectors = min_t(unsigned, sectors, buf->size >> 9);
-	j->cur_buf_sectors	= sectors;
-
-	u64s = (sectors << 9) / sizeof(u64);
-
-	/* Subtract the journal header */
-	u64s -= sizeof(struct jset) / sizeof(u64);
-	u64s -= buf->u64s_reserved;
-	u64s  = max_t(ssize_t, 0L, u64s);
+	buf->u64s_reserved	= j->entry_u64s_reserved;
+	buf->disk_sectors	= j->cur_entry_sectors;
+	buf->sectors		= min(buf->disk_sectors, buf->buf_size >> 9);
 
-	BUG_ON(u64s >= JOURNAL_ENTRY_CLOSED_VAL);
+	u64s = (int) (buf->sectors << 9) / sizeof(u64) -
+		journal_entry_overhead(j);
+	u64s  = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 
 	if (u64s <= le32_to_cpu(buf->data->u64s))
-		return 0;
+		return -ENOSPC;
 
 	/*
 	 * Must be set before marking the journal entry as open:
@@ -246,11 +248,14 @@ static int journal_entry_open(struct journal *j)
 	do {
 		old.v = new.v = v;
 
+		EBUG_ON(journal_state_count(new, new.idx));
+
 		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL)
-			return -EIO;
+			return -EROFS;
 
 		/* Handle any already added entries */
 		new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
+		journal_state_inc(&new);
 	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
 				       old.v, new.v)) != old.v);
 
@@ -263,37 +268,22 @@ static int journal_entry_open(struct journal *j)
 			 &j->write_work,
 			 msecs_to_jiffies(j->write_delay_ms));
 	journal_wake(j);
-	return 1;
+	return 0;
 }
 
-static bool __journal_entry_close(struct journal *j)
+static bool journal_quiesced(struct journal *j)
 {
-	bool set_need_write;
-
-	if (!journal_entry_is_open(j)) {
-		spin_unlock(&j->lock);
-		return true;
-	}
-
-	set_need_write = !test_and_set_bit(JOURNAL_NEED_WRITE, &j->flags);
-	if (set_need_write)
-		j->need_write_time = local_clock();
+	union journal_res_state state = READ_ONCE(j->reservations);
+	bool ret = !state.prev_buf_unwritten && !__journal_entry_is_open(state);
 
-	switch (journal_buf_switch(j, set_need_write)) {
-	case JOURNAL_ENTRY_INUSE:
-		spin_unlock(&j->lock);
-		return false;
-	default:
-		spin_unlock(&j->lock);
-	case JOURNAL_UNLOCKED:
-		return true;
-	}
+	if (!ret)
+		journal_entry_close(j);
+	return ret;
 }
 
-static bool journal_entry_close(struct journal *j)
+static void journal_quiesce(struct journal *j)
 {
-	spin_lock(&j->lock);
-	return __journal_entry_close(j);
+	wait_event(j->wait, journal_quiesced(j));
 }
 
 static void journal_write_work(struct work_struct *work)
@@ -337,7 +327,11 @@ retry:
 	if (journal_res_get_fast(j, res, flags))
 		return 0;
 
+	if (bch2_journal_error(j))
+		return -EROFS;
+
 	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()
@@ -355,56 +349,43 @@ 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:
+	if (journal_entry_is_open(j) &&
+	    !__journal_entry_close(j)) {
 		/*
-		 * The current journal entry is still open, but we failed to get
-		 * a journal reservation because there's not enough space in it,
-		 * and we can't close it and start another because we haven't
-		 * finished writing out the previous entry:
+		 * We failed to get a reservation on the current open journal
+		 * entry because it's full, and we can't close it because
+		 * there's still a previous one in flight:
 		 */
-		spin_unlock(&j->lock);
 		trace_journal_entry_full(c);
-		goto blocked;
-	case JOURNAL_ENTRY_CLOSED:
-		break;
-	case JOURNAL_UNLOCKED:
-		goto retry;
+		ret = -EAGAIN;
+	} else {
+		ret = journal_entry_open(j);
 	}
 
-	/* We now have a new, closed journal buf - see if we can open it: */
-	ret = journal_entry_open(j);
+	if ((ret == -EAGAIN || ret == -ENOSPC) &&
+	    !j->res_get_blocked_start)
+		j->res_get_blocked_start = local_clock() ?: 1;
+
 	spin_unlock(&j->lock);
 
-	if (ret < 0)
-		return ret;
-	if (ret)
+	if (!ret)
 		goto retry;
+	if (ret == -ENOSPC) {
+		/*
+		 * Journal is full - can't rely on reclaim from work item due to
+		 * freezing:
+		 */
+		trace_journal_full(c);
+		if (!(flags & JOURNAL_RES_GET_NONBLOCK))
+			bch2_journal_reclaim_work(&j->reclaim_work.work);
+		ret = -EAGAIN;
+	}
 
-	/* Journal's full, we have to wait */
-
-	/*
-	 * Direct reclaim - can't rely on reclaim from work item
-	 * due to freezing..
-	 */
-	bch2_journal_reclaim_work(&j->reclaim_work.work);
-
-	trace_journal_full(c);
-blocked:
-	if (!j->res_get_blocked_start)
-		j->res_get_blocked_start = local_clock() ?: 1;
-	return -EAGAIN;
+	return ret;
 }
 
 /*
@@ -422,7 +403,7 @@ int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 {
 	int ret;
 
-	wait_event(j->wait,
+	closure_wait_event(&j->async_wait,
 		   (ret = __journal_res_get(j, res, flags)) != -EAGAIN ||
 		   (flags & JOURNAL_RES_GET_NONBLOCK));
 	return ret;
@@ -441,9 +422,9 @@ void bch2_journal_entry_res_resize(struct journal *j,
 
 	j->entry_u64s_reserved += d;
 	if (d <= 0)
-		goto out_unlock;
+		goto out;
 
-	j->cur_entry_u64s -= d;
+	j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
 	smp_mb();
 	state = READ_ONCE(j->reservations);
 
@@ -454,15 +435,12 @@ void bch2_journal_entry_res_resize(struct journal *j,
 		 * Not enough room in current journal entry, have to flush it:
 		 */
 		__journal_entry_close(j);
-		goto out;
+	} else {
+		journal_cur_buf(j)->u64s_reserved += d;
 	}
-
-	journal_cur_buf(j)->u64s_reserved += d;
-out_unlock:
-	spin_unlock(&j->lock);
 out:
+	spin_unlock(&j->lock);
 	res->u64s += d;
-	return;
 }
 
 /* journal flushing: */
@@ -492,47 +470,47 @@ int bch2_journal_open_seq_async(struct journal *j, u64 seq, struct closure *cl)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	int ret;
-retry:
+
 	spin_lock(&j->lock);
 
-	if (seq < journal_cur_seq(j) ||
+	/*
+	 * Can't try to open more than one sequence number ahead:
+	 */
+	BUG_ON(journal_cur_seq(j) < seq && !journal_entry_is_open(j));
+
+	if (journal_cur_seq(j) > seq ||
 	    journal_entry_is_open(j)) {
 		spin_unlock(&j->lock);
 		return 0;
 	}
 
-	if (journal_cur_seq(j) < seq) {
-		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: */
-			trace_journal_entry_full(c);
-			goto blocked;
-		case JOURNAL_ENTRY_CLOSED:
-			break;
-		case JOURNAL_UNLOCKED:
-			goto retry;
-		}
-	}
-
-	BUG_ON(journal_cur_seq(j) < seq);
+	if (journal_cur_seq(j) < seq &&
+	    !__journal_entry_close(j)) {
+		/* haven't finished writing out the previous one: */
+		trace_journal_entry_full(c);
+		ret = -EAGAIN;
+	} else {
+		BUG_ON(journal_cur_seq(j) != seq);
 
-	ret = journal_entry_open(j);
-	if (ret) {
-		spin_unlock(&j->lock);
-		return ret < 0 ? ret : 0;
+		ret = journal_entry_open(j);
 	}
-blocked:
-	if (!j->res_get_blocked_start)
+
+	if ((ret == -EAGAIN || ret == -ENOSPC) &&
+	    !j->res_get_blocked_start)
 		j->res_get_blocked_start = local_clock() ?: 1;
 
-	closure_wait(&j->async_wait, cl);
+	if (ret == -EAGAIN || ret == -ENOSPC)
+		closure_wait(&j->async_wait, cl);
+
 	spin_unlock(&j->lock);
 
-	bch2_journal_reclaim_work(&j->reclaim_work.work);
-	return -EAGAIN;
+	if (ret == -ENOSPC) {
+		trace_journal_full(c);
+		bch2_journal_reclaim_work(&j->reclaim_work.work);
+		ret = -EAGAIN;
+	}
+
+	return ret;
 }
 
 static int journal_seq_error(struct journal *j, u64 seq)
@@ -615,8 +593,7 @@ void bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 
 	if (seq == journal_cur_seq(j))
 		__journal_entry_close(j);
-	else
-		spin_unlock(&j->lock);
+	spin_unlock(&j->lock);
 }
 
 static int journal_seq_flushed(struct journal *j, u64 seq)
@@ -628,8 +605,7 @@ static int journal_seq_flushed(struct journal *j, u64 seq)
 
 	if (seq == journal_cur_seq(j))
 		__journal_entry_close(j);
-	else
-		spin_unlock(&j->lock);
+	spin_unlock(&j->lock);
 
 	return ret;
 }
@@ -721,6 +697,26 @@ int bch2_journal_flush(struct journal *j)
 	return bch2_journal_flush_seq(j, seq);
 }
 
+/* block/unlock the journal: */
+
+void bch2_journal_unblock(struct journal *j)
+{
+	spin_lock(&j->lock);
+	j->blocked--;
+	spin_unlock(&j->lock);
+
+	journal_wake(j);
+}
+
+void bch2_journal_block(struct journal *j)
+{
+	spin_lock(&j->lock);
+	j->blocked++;
+	spin_unlock(&j->lock);
+
+	journal_quiesce(j);
+}
+
 /* allocate journal on a device: */
 
 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
@@ -743,7 +739,7 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 		goto err;
 
 	journal_buckets = bch2_sb_resize_journal(&ca->disk_sb,
-				nr + sizeof(*journal_buckets) / sizeof(u64));
+						 nr + sizeof(*journal_buckets) / sizeof(u64));
 	if (!journal_buckets)
 		goto err;
 
@@ -806,9 +802,9 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 		ja->nr++;
 
 		bch2_mark_metadata_bucket(c, ca, bucket, BCH_DATA_JOURNAL,
-				ca->mi.bucket_size,
-				gc_phase(GC_PHASE_SB),
-				0);
+					  ca->mi.bucket_size,
+					  gc_phase(GC_PHASE_SB),
+					  0);
 
 		if (c) {
 			spin_unlock(&c->journal.lock);
@@ -859,7 +855,7 @@ int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 		 */
 
 		if (bch2_disk_reservation_get(c, &disk_res,
-				bucket_to_sector(ca, nr - ja->nr), 1, 0)) {
+					      bucket_to_sector(ca, nr - ja->nr), 1, 0)) {
 			mutex_unlock(&c->sb_lock);
 			return -ENOSPC;
 		}
@@ -930,8 +926,7 @@ void bch2_fs_journal_stop(struct journal *j)
 	    c->btree_roots_dirty)
 		bch2_journal_meta(j);
 
-	BUG_ON(journal_entry_is_open(j) ||
-	       j->reservations.prev_buf_unwritten);
+	journal_quiesce(j);
 
 	BUG_ON(!bch2_journal_error(j) &&
 	       test_bit(JOURNAL_NOT_EMPTY, &j->flags));
@@ -957,7 +952,7 @@ void bch2_fs_journal_start(struct journal *j)
 		journal_pin_new_entry(j, 0);
 
 	/*
-	 * journal_buf_switch() only inits the next journal entry when it
+	 * __journal_entry_close() only inits the next journal entry when it
 	 * closes an open journal entry - the very first journal entry gets
 	 * initialized here:
 	 */
@@ -966,6 +961,7 @@ void bch2_fs_journal_start(struct journal *j)
 
 	c->last_bucket_seq_cleanup = journal_cur_seq(j);
 
+	bch2_journal_space_available(j);
 	spin_unlock(&j->lock);
 
 	/*
@@ -975,7 +971,7 @@ void bch2_fs_journal_start(struct journal *j)
 	 */
 	bch2_journal_seq_blacklist_write(j);
 
-	queue_delayed_work(system_freezable_wq, &j->reclaim_work, 0);
+	queue_delayed_work(c->journal_reclaim_wq, &j->reclaim_work, 0);
 }
 
 /* init/exit: */
@@ -1021,8 +1017,8 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 
 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);
+	kvpfree(j->buf[1].data, j->buf[1].buf_size);
+	kvpfree(j->buf[0].data, j->buf[0].buf_size);
 	free_fifo(&j->pin);
 }
 
@@ -1046,8 +1042,8 @@ int bch2_fs_journal_init(struct journal *j)
 
 	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->buf[0].buf_size	= JOURNAL_ENTRY_SIZE_MIN;
+	j->buf[1].buf_size	= JOURNAL_ENTRY_SIZE_MIN;
 	j->write_delay_ms	= 1000;
 	j->reclaim_delay_ms	= 100;
 
@@ -1060,8 +1056,8 @@ int bch2_fs_journal_init(struct journal *j)
 		 { .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))) {
+	    !(j->buf[0].data = kvpmalloc(j->buf[0].buf_size, GFP_KERNEL)) ||
+	    !(j->buf[1].data = kvpmalloc(j->buf[1].buf_size, GFP_KERNEL))) {
 		ret = -ENOMEM;
 		goto out;
 	}
@@ -1078,35 +1074,54 @@ ssize_t bch2_journal_print_debug(struct journal *j, char *buf)
 {
 	struct printbuf out = _PBUF(buf, PAGE_SIZE);
 	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;
 
 	rcu_read_lock();
 	spin_lock(&j->lock);
+	s = READ_ONCE(j->reservations);
 
 	pr_buf(&out,
 	       "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",
+	       "current entry:\t\t",
 	       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,
+	       j->last_seq_ondisk);
+
+	switch (s.cur_entry_offset) {
+	case JOURNAL_ENTRY_ERROR_VAL:
+		pr_buf(&out, "error\n");
+		break;
+	case JOURNAL_ENTRY_CLOSED_VAL:
+		pr_buf(&out, "closed\n");
+		break;
+	default:
+		pr_buf(&out, "%u/%u\n",
+		       s.cur_entry_offset,
+		       j->cur_entry_u64s);
+		break;
+	}
+
+	pr_buf(&out,
+	       "current entry refs:\t%u\n"
+	       "prev entry unwritten:\t",
+	       journal_state_count(s, s.idx));
+
+	if (s.prev_buf_unwritten)
+		pr_buf(&out, "yes, ref %u\n",
+		       journal_state_count(s, !s.idx));
+	else
+		pr_buf(&out, "no\n");
+
+	pr_buf(&out,
+	       "need write:\t\t%i\n"
+	       "replay done:\t\t%i\n",
 	       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,
@@ -1119,9 +1134,12 @@ ssize_t bch2_journal_print_debug(struct journal *j, char *buf)
 		pr_buf(&out,
 		       "dev %u:\n"
 		       "\tnr\t\t%u\n"
+		       "\tavailable\t%u:%u\n"
 		       "\tcur_idx\t\t%u (seq %llu)\n"
 		       "\tlast_idx\t%u (seq %llu)\n",
 		       iter, ja->nr,
+		       bch2_journal_dev_buckets_available(j, ja),
+		       ja->sectors_free,
 		       ja->cur_idx,	ja->bucket_seq[ja->cur_idx],
 		       ja->last_idx,	ja->bucket_seq[ja->last_idx]);
 	}