bcachefs: New tracepoint for bch2_trans_get_iter()

Trying to debug an issue where after traverse_all() we shouldn't have to
traverse any iterators... yet we are

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

2 files changed, 79 insertions, 9 deletions

diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index 757e5650f33c..525afe3509ec 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -2020,6 +2020,13 @@ struct btree_iter *__bch2_trans_get_iter(struct btree_trans *trans,
 					 unsigned flags)
 {
 	struct btree_iter *iter, *best = NULL;
+	struct bpos real_pos, pos_min = POS_MIN;
+
+	if ((flags & BTREE_ITER_TYPE) != BTREE_ITER_NODES &&
+	    btree_node_type_is_extents(btree_id) &&
+	    !(flags & BTREE_ITER_NOT_EXTENTS) &&
+	    !(flags & BTREE_ITER_ALL_SNAPSHOTS))
+		flags |= BTREE_ITER_IS_EXTENTS;
 
 	if ((flags & BTREE_ITER_TYPE) != BTREE_ITER_NODES &&
 	    !btree_type_has_snapshots(btree_id))
@@ -2029,6 +2036,12 @@ struct btree_iter *__bch2_trans_get_iter(struct btree_trans *trans,
 		pos.snapshot = btree_type_has_snapshots(btree_id)
 			? U32_MAX : 0;
 
+	real_pos = pos;
+
+	if ((flags & BTREE_ITER_IS_EXTENTS) &&
+	    bkey_cmp(pos, POS_MAX))
+		real_pos = bpos_nosnap_successor(pos);
+
 	trans_for_each_iter(trans, iter) {
 		if (btree_iter_type(iter) != (flags & BTREE_ITER_TYPE))
 			continue;
@@ -2037,8 +2050,8 @@ struct btree_iter *__bch2_trans_get_iter(struct btree_trans *trans,
 			continue;
 
 		if (best) {
-			int cmp = bkey_cmp(bpos_diff(best->real_pos, pos),
-					   bpos_diff(iter->real_pos, pos));
+			int cmp = bkey_cmp(bpos_diff(best->real_pos, real_pos),
+					   bpos_diff(iter->real_pos, real_pos));
 
 			if (cmp < 0 ||
 			    ((cmp == 0 && btree_iter_keep(trans, iter))))
@@ -2048,6 +2061,13 @@ struct btree_iter *__bch2_trans_get_iter(struct btree_trans *trans,
 		best = iter;
 	}
 
+	trace_trans_get_iter(_RET_IP_, trans->ip,
+			     btree_id,
+			     &real_pos, locks_want,
+			     best ? &best->real_pos : &pos_min,
+			     best ? best->locks_want : 0,
+			     best ? best->uptodate : BTREE_ITER_NEED_TRAVERSE);
+
 	if (!best) {
 		iter = btree_trans_iter_alloc(trans);
 		bch2_btree_iter_init(trans, iter, btree_id);
@@ -2061,12 +2081,6 @@ struct btree_iter *__bch2_trans_get_iter(struct btree_trans *trans,
 	trans->iters_live	|= 1ULL << iter->idx;
 	trans->iters_touched	|= 1ULL << iter->idx;
 
-	if ((flags & BTREE_ITER_TYPE) != BTREE_ITER_NODES &&
-	    btree_node_type_is_extents(btree_id) &&
-	    !(flags & BTREE_ITER_NOT_EXTENTS) &&
-	    !(flags & BTREE_ITER_ALL_SNAPSHOTS))
-		flags |= BTREE_ITER_IS_EXTENTS;
-
 	iter->flags = flags;
 
 	iter->snapshot = pos.snapshot;
@@ -2098,7 +2112,7 @@ struct btree_iter *__bch2_trans_get_iter(struct btree_trans *trans,
 	iter->min_depth	= depth;
 
 	bch2_btree_iter_set_pos(iter, pos);
-	btree_iter_set_search_pos(iter, btree_iter_search_key(iter));
+	btree_iter_set_search_pos(iter, real_pos);
 
 	return iter;
 }
diff --git a/fs/bcachefs/trace.h b/fs/bcachefs/trace.h
index c6d98f4c50e7..7e518f7618d4 100644
--- a/fs/bcachefs/trace.h
+++ b/fs/bcachefs/trace.h
@@ -528,6 +528,62 @@ TRACE_EVENT(copygc,
 		__entry->buckets_moved, __entry->buckets_not_moved)
 );
 
+TRACE_EVENT(trans_get_iter,
+	TP_PROTO(unsigned long caller, unsigned long ip,
+		 enum btree_id btree_id,
+		 struct bpos *pos_want,
+		 unsigned locks_want,
+		 struct bpos *pos_found,
+		 unsigned locks_found,
+		 unsigned uptodate),
+	TP_ARGS(caller, ip, btree_id,
+		pos_want, locks_want,
+		pos_found, locks_found,
+		uptodate),
+
+	TP_STRUCT__entry(
+		__field(unsigned long,	caller			)
+		__field(unsigned long,	ip			)
+		__field(u8,		btree_id		)
+		__field(u8,		uptodate		)
+		__field(u8,		locks_want		)
+		__field(u8,		locks_found		)
+		__field(u64,		pos_want_inode		)
+		__field(u64,		pos_want_offset		)
+		__field(u32,		pos_want_snapshot	)
+		__field(u64,		pos_found_inode		)
+		__field(u64,		pos_found_offset	)
+		__field(u32,		pos_found_snapshot	)
+	),
+
+	TP_fast_assign(
+		__entry->caller			= caller;
+		__entry->ip			= ip;
+		__entry->btree_id		= btree_id;
+		__entry->uptodate		= uptodate;
+		__entry->pos_want_inode		= pos_want->inode;
+		__entry->pos_want_offset	= pos_want->offset;
+		__entry->pos_want_snapshot	= pos_want->snapshot;
+		__entry->pos_found_inode	= pos_found->inode;
+		__entry->pos_found_offset	= pos_found->offset;
+		__entry->pos_found_snapshot	= pos_found->snapshot;
+	),
+
+	TP_printk("%ps %pS btree %u uptodate %u want %llu:%llu:%u locks %u found %llu:%llu:%u locks %u",
+		  (void *) __entry->caller,
+		  (void *) __entry->ip,
+		  __entry->btree_id,
+		  __entry->uptodate,
+		  __entry->pos_want_inode,
+		  __entry->pos_want_offset,
+		  __entry->pos_want_snapshot,
+		  __entry->locks_want,
+		  __entry->pos_found_inode,
+		  __entry->pos_found_offset,
+		  __entry->pos_found_snapshot,
+		  __entry->locks_found)
+);
+
 TRACE_EVENT(transaction_restart_ip,
 	TP_PROTO(unsigned long caller, unsigned long ip),
 	TP_ARGS(caller, ip),