bcachefs: Bring back metadata only gc

This is useful for the filesystem dump debugging tool - when we're
hitting bugs we want to skip as much of the recovery process as
possible, and the dump tool only needs to know where metadata lives.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/btree_gc.c

@@ -378,12 +378,13 @@ static int btree_gc_mark_node(struct bch_fs *c, struct btree *b, u8 *max_stale,
 }
 
 static int bch2_gc_btree(struct bch_fs *c, enum btree_id btree_id,
-			 bool initial)
+			 bool initial, bool metadata_only)
 {
 	struct btree_trans trans;
 	struct btree_iter *iter;
 	struct btree *b;
-	unsigned depth = bch2_expensive_debug_checks	? 0
+	unsigned depth = metadata_only			? 1
+		: bch2_expensive_debug_checks		? 0
 		: !btree_node_type_needs_gc(btree_id)	? 1
 		: 0;
 	u8 max_stale = 0;
@@ -526,11 +527,13 @@ static int bch2_gc_btree_init_recurse(struct bch_fs *c, struct btree *b,
 }
 
 static int bch2_gc_btree_init(struct bch_fs *c,
-			      enum btree_id btree_id)
+			      enum btree_id btree_id,
+			      bool metadata_only)
 {
 	struct btree *b;
-	unsigned target_depth = bch2_expensive_debug_checks	? 0
-		: !btree_node_type_needs_gc(btree_id)		? 1
+	unsigned target_depth = metadata_only		? 1
+		: bch2_expensive_debug_checks		? 0
+		: !btree_node_type_needs_gc(btree_id)	? 1
 		: 0;
 	u8 max_stale = 0;
 	int ret = 0;
@@ -576,7 +579,7 @@ static inline int btree_id_gc_phase_cmp(enum btree_id l, enum btree_id r)
 		(int) btree_id_to_gc_phase(r);
 }
 
-static int bch2_gc_btrees(struct bch_fs *c, bool initial)
+static int bch2_gc_btrees(struct bch_fs *c, bool initial, bool metadata_only)
 {
 	enum btree_id ids[BTREE_ID_NR];
 	unsigned i;
@@ -588,8 +591,8 @@ static int bch2_gc_btrees(struct bch_fs *c, bool initial)
 	for (i = 0; i < BTREE_ID_NR; i++) {
 		enum btree_id id = ids[i];
 		int ret = initial
-			? bch2_gc_btree_init(c, id)
-			: bch2_gc_btree(c, id, initial);
+			? bch2_gc_btree_init(c, id, metadata_only)
+			: bch2_gc_btree(c, id, initial, metadata_only);
 		if (ret) {
 			bch_err(c, "%s: ret %i", __func__, ret);
 			return ret;
@@ -758,10 +761,10 @@ static void bch2_gc_free(struct bch_fs *c)
 }
 
 static int bch2_gc_done(struct bch_fs *c,
-			bool initial)
+			bool initial, bool metadata_only)
 {
 	struct bch_dev *ca;
-	bool verify = (!initial ||
+	bool verify = !metadata_only && (!initial ||
 		       (c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)));
 	unsigned i, dev;
 	int ret = 0;
@@ -800,7 +803,7 @@ static int bch2_gc_done(struct bch_fs *c,
 #define copy_fs_field(_f, _msg, ...)					\
 	copy_field(_f, "fs has wrong " _msg, ##__VA_ARGS__)
 
-	{
+	if (!metadata_only) {
 		struct genradix_iter iter = genradix_iter_init(&c->stripes[1], 0);
 		struct stripe *dst, *src;
 
@@ -873,20 +876,28 @@ static int bch2_gc_done(struct bch_fs *c,
 
 		copy_fs_field(hidden,		"hidden");
 		copy_fs_field(btree,		"btree");
-		copy_fs_field(data,	"data");
-		copy_fs_field(cached,	"cached");
-		copy_fs_field(reserved,	"reserved");
-		copy_fs_field(nr_inodes,"nr_inodes");
 
-		for (i = 0; i < BCH_REPLICAS_MAX; i++)
-			copy_fs_field(persistent_reserved[i],
-				      "persistent_reserved[%i]", i);
+		if (!metadata_only) {
+			copy_fs_field(data,	"data");
+			copy_fs_field(cached,	"cached");
+			copy_fs_field(reserved,	"reserved");
+			copy_fs_field(nr_inodes,"nr_inodes");
+
+			for (i = 0; i < BCH_REPLICAS_MAX; i++)
+				copy_fs_field(persistent_reserved[i],
+					      "persistent_reserved[%i]", i);
+		}
 
 		for (i = 0; i < c->replicas.nr; i++) {
 			struct bch_replicas_entry *e =
 				cpu_replicas_entry(&c->replicas, i);
 			char buf[80];
 
+			if (metadata_only &&
+			    (e->data_type == BCH_DATA_user ||
+			     e->data_type == BCH_DATA_cached))
+				continue;
+
 			bch2_replicas_entry_to_text(&PBUF(buf), e);
 
 			copy_fs_field(replicas[i], "%s", buf);
@@ -904,7 +915,8 @@ static int bch2_gc_done(struct bch_fs *c,
 	return ret;
 }
 
-static int bch2_gc_start(struct bch_fs *c)
+static int bch2_gc_start(struct bch_fs *c,
+			 bool metadata_only)
 {
 	struct bch_dev *ca;
 	unsigned i;
@@ -968,6 +980,13 @@ static int bch2_gc_start(struct bch_fs *c)
 
 			d->_mark.gen = dst->b[b].oldest_gen = s->mark.gen;
 			d->gen_valid = s->gen_valid;
+
+			if (metadata_only &&
+			    (s->mark.data_type == BCH_DATA_user ||
+			     s->mark.data_type == BCH_DATA_cached)) {
+				d->_mark = s->mark;
+				d->_mark.owned_by_allocator = 0;
+			}
 		}
 	};
 
@@ -994,7 +1013,7 @@ static int bch2_gc_start(struct bch_fs *c)
  *    move around - if references move backwards in the ordering GC
  *    uses, GC could skip past them
  */
-int bch2_gc(struct bch_fs *c, bool initial)
+int bch2_gc(struct bch_fs *c, bool initial, bool metadata_only)
 {
 	struct bch_dev *ca;
 	u64 start_time = local_clock();
@@ -1010,13 +1029,13 @@ int bch2_gc(struct bch_fs *c, bool initial)
 	closure_wait_event(&c->btree_interior_update_wait,
 			   !bch2_btree_interior_updates_nr_pending(c));
 again:
-	ret = bch2_gc_start(c);
+	ret = bch2_gc_start(c, metadata_only);
 	if (ret)
 		goto out;
 
 	bch2_mark_superblocks(c);
 
-	ret = bch2_gc_btrees(c, initial);
+	ret = bch2_gc_btrees(c, initial, metadata_only);
 	if (ret)
 		goto out;
 
@@ -1054,7 +1073,7 @@ int bch2_gc(struct bch_fs *c, bool initial)
 		bch2_journal_block(&c->journal);
 
 		percpu_down_write(&c->mark_lock);
-		ret = bch2_gc_done(c, initial);
+		ret = bch2_gc_done(c, initial, metadata_only);
 
 		bch2_journal_unblock(&c->journal);
 	} else {

fs/bcachefs/btree_gc.h

@@ -6,7 +6,7 @@
 
 void bch2_coalesce(struct bch_fs *);
 
-int bch2_gc(struct bch_fs *, bool);
+int bch2_gc(struct bch_fs *, bool, bool);
 int bch2_gc_gens(struct bch_fs *);
 void bch2_gc_thread_stop(struct bch_fs *);
 int bch2_gc_thread_start(struct bch_fs *);

fs/bcachefs/recovery.c

@@ -1117,9 +1117,11 @@ int bch2_fs_recovery(struct bch_fs *c)
 	    !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
 	    !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
 	    test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
+		bool metadata_only = c->opts.norecovery;
+
 		bch_info(c, "starting mark and sweep");
 		err = "error in mark and sweep";
-		ret = bch2_gc(c, true);
+		ret = bch2_gc(c, true, metadata_only);
 		if (ret)
 			goto err;
 		bch_verbose(c, "mark and sweep done");