bcachefs: Redo replicas gc mechanism

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

fs/bcachefs/migrate.c

@@ -48,10 +48,6 @@ static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 	iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 				   POS_MIN, BTREE_ITER_PREFETCH);
 
-	mutex_lock(&c->replicas_gc_lock);
-	bch2_replicas_gc_start(c, (1 << BCH_DATA_USER)|(1 << BCH_DATA_CACHED));
-
-
 	while ((k = bch2_btree_iter_peek(iter)).k &&
 	       !(ret = bkey_err(k))) {
 		if (!bkey_extent_is_data(k.k) ||
@@ -97,12 +93,9 @@ static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 			break;
 	}
 
-	BUG_ON(ret == -EINTR);
-
-	bch2_trans_exit(&trans);
+	ret = bch2_trans_exit(&trans) ?: ret;
 
-	bch2_replicas_gc_end(c, ret);
-	mutex_unlock(&c->replicas_gc_lock);
+	BUG_ON(ret == -EINTR);
 
 	return ret;
 }
@@ -123,9 +116,6 @@ static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 	bch2_trans_init(&trans, c);
 	closure_init_stack(&cl);
 
-	mutex_lock(&c->replicas_gc_lock);
-	bch2_replicas_gc_start(c, 1 << BCH_DATA_BTREE);
-
 	for (id = 0; id < BTREE_ID_NR; id++) {
 		for_each_btree_node(&trans, iter, id, POS_MIN,
 				    BTREE_ITER_PREFETCH, b) {
@@ -178,10 +168,9 @@ static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 
 	ret = 0;
 err:
-	bch2_trans_exit(&trans);
+	ret = bch2_trans_exit(&trans) ?: ret;
 
-	ret = bch2_replicas_gc_end(c, ret);
-	mutex_unlock(&c->replicas_gc_lock);
+	BUG_ON(ret == -EINTR);
 
 	return ret;
 }
@@ -189,5 +178,6 @@ static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 {
 	return bch2_dev_usrdata_drop(c, dev_idx, flags) ?:
-		bch2_dev_metadata_drop(c, dev_idx, flags);
+		bch2_dev_metadata_drop(c, dev_idx, flags) ?:
+		bch2_replicas_gc2(c);
 }

fs/bcachefs/move.c

@@ -619,64 +619,6 @@ int bch2_move_data(struct bch_fs *c,
 	return ret;
 }
 
-static int bch2_gc_data_replicas(struct bch_fs *c)
-{
-	struct btree_trans trans;
-	struct btree_iter *iter;
-	struct bkey_s_c k;
-	int ret;
-
-	bch2_trans_init(&trans, c);
-
-	mutex_lock(&c->replicas_gc_lock);
-	bch2_replicas_gc_start(c, (1 << BCH_DATA_USER)|(1 << BCH_DATA_CACHED));
-
-	for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, POS_MIN,
-			   BTREE_ITER_PREFETCH, k, ret) {
-		ret = bch2_mark_bkey_replicas(c, k);
-		if (ret)
-			break;
-	}
-	ret = bch2_trans_exit(&trans) ?: ret;
-
-	bch2_replicas_gc_end(c, ret);
-	mutex_unlock(&c->replicas_gc_lock);
-
-	return ret;
-}
-
-static int bch2_gc_btree_replicas(struct bch_fs *c)
-{
-	struct btree_trans trans;
-	struct btree_iter *iter;
-	struct btree *b;
-	unsigned id;
-	int ret = 0;
-
-	bch2_trans_init(&trans, c);
-
-	mutex_lock(&c->replicas_gc_lock);
-	bch2_replicas_gc_start(c, 1 << BCH_DATA_BTREE);
-
-	for (id = 0; id < BTREE_ID_NR; id++) {
-		for_each_btree_node(&trans, iter, id, POS_MIN,
-				    BTREE_ITER_PREFETCH, b) {
-			ret = bch2_mark_bkey_replicas(c, bkey_i_to_s_c(&b->key));
-
-			bch2_trans_cond_resched(&trans);
-		}
-
-		ret = bch2_trans_iter_free(&trans, iter) ?: ret;
-	}
-
-	bch2_trans_exit(&trans);
-
-	bch2_replicas_gc_end(c, ret);
-	mutex_unlock(&c->replicas_gc_lock);
-
-	return ret;
-}
-
 static int bch2_move_btree(struct bch_fs *c,
 			   move_pred_fn pred,
 			   void *arg,
@@ -803,14 +745,14 @@ int bch2_data_job(struct bch_fs *c,
 			bch2_journal_meta(&c->journal);
 		}
 
-		ret = bch2_gc_btree_replicas(c) ?: ret;
+		ret = bch2_replicas_gc2(c) ?: ret;
 
 		ret = bch2_move_data(c, NULL,
 				     writepoint_hashed((unsigned long) current),
 				     op.start,
 				     op.end,
 				     rereplicate_pred, c, stats) ?: ret;
-		ret = bch2_gc_data_replicas(c) ?: ret;
+		ret = bch2_replicas_gc2(c) ?: ret;
 		break;
 	case BCH_DATA_OP_MIGRATE:
 		if (op.migrate.dev >= c->sb.nr_devices)
@@ -820,14 +762,14 @@ int bch2_data_job(struct bch_fs *c,
 		ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 
 		ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret;
-		ret = bch2_gc_btree_replicas(c) ?: ret;
+		ret = bch2_replicas_gc2(c) ?: ret;
 
 		ret = bch2_move_data(c, NULL,
 				     writepoint_hashed((unsigned long) current),
 				     op.start,
 				     op.end,
 				     migrate_pred, &op, stats) ?: ret;
-		ret = bch2_gc_data_replicas(c) ?: ret;
+		ret = bch2_replicas_gc2(c) ?: ret;
 		break;
 	default:
 		ret = -EINVAL;

fs/bcachefs/replicas.c

@@ -571,6 +571,64 @@ int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 	return 0;
 }
 
+int bch2_replicas_gc2(struct bch_fs *c)
+{
+	struct bch_replicas_cpu new = { 0 };
+	unsigned i, nr;
+	int ret = 0;
+
+	bch2_journal_meta(&c->journal);
+retry:
+	nr		= READ_ONCE(c->replicas.nr);
+	new.entry_size	= READ_ONCE(c->replicas.entry_size);
+	new.entries	= kcalloc(nr, new.entry_size, GFP_KERNEL);
+	if (!new.entries)
+		return -ENOMEM;
+
+	mutex_lock(&c->sb_lock);
+	percpu_down_write(&c->mark_lock);
+
+	if (nr			!= c->replicas.nr ||
+	    new.entry_size	!= c->replicas.entry_size) {
+		percpu_up_write(&c->mark_lock);
+		mutex_unlock(&c->sb_lock);
+		kfree(new.entries);
+		goto retry;
+	}
+
+	for (i = 0; i < c->replicas.nr; i++) {
+		struct bch_replicas_entry *e =
+			cpu_replicas_entry(&c->replicas, i);
+
+		if (e->data_type == BCH_DATA_JOURNAL ||
+		    c->usage_base->replicas[i] ||
+		    percpu_u64_get(&c->usage[0]->replicas[i]) ||
+		    percpu_u64_get(&c->usage[1]->replicas[i]))
+			memcpy(cpu_replicas_entry(&new, new.nr++),
+			       e, new.entry_size);
+	}
+
+	bch2_cpu_replicas_sort(&new);
+
+	if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
+		ret = -ENOSPC;
+		goto err;
+	}
+
+	ret = replicas_table_update(c, &new);
+err:
+	kfree(new.entries);
+
+	percpu_up_write(&c->mark_lock);
+
+	if (!ret)
+		bch2_write_super(c);
+
+	mutex_unlock(&c->sb_lock);
+
+	return ret;
+}
+
 int bch2_replicas_set_usage(struct bch_fs *c,
 			    struct bch_replicas_entry *r,
 			    u64 sectors)

fs/bcachefs/replicas.h

@@ -59,6 +59,7 @@ unsigned bch2_dev_has_data(struct bch_fs *, struct bch_dev *);
 
 int bch2_replicas_gc_end(struct bch_fs *, int);
 int bch2_replicas_gc_start(struct bch_fs *, unsigned);
+int bch2_replicas_gc2(struct bch_fs *);
 
 int bch2_replicas_set_usage(struct bch_fs *,
 			    struct bch_replicas_entry *,