bcachefs: Normal update/commit path now works before going RW

This improves __bch2_trans_commit - early in the recovery process, when
we're running btree_gc and before we want to go RW, it now uses
bch2_journal_key_insert() to add the update to the list of updates for
journal replay to do, instead of btree_gc having to use separate
interfaces depending on whether we're running at bringup or, later,
runtime.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

fs/bcachefs/bcachefs.h

@@ -515,6 +515,7 @@ enum {
 	BCH_FS_TOPOLOGY_REPAIR_DONE,
 	BCH_FS_FSCK_DONE,
 	BCH_FS_STARTED,
+	BCH_FS_MAY_GO_RW,
 	BCH_FS_RW,
 	BCH_FS_WAS_RW,
 

fs/bcachefs/btree_gc.c

@@ -753,7 +753,8 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 			atomic64_set(&c->key_version, k->k->version.lo);
 	}
 
-	ret = bch2_mark_key(trans, old, *k, flags);
+	ret = __bch2_trans_do(trans, NULL, NULL, 0,
+			bch2_mark_key(trans, old, *k, flags));
 fsck_err:
 err:
 	if (ret)
@@ -1259,7 +1260,7 @@ static int bch2_gc_start(struct bch_fs *c,
 
 static int bch2_alloc_write_key(struct btree_trans *trans,
 				struct btree_iter *iter,
-				bool initial, bool metadata_only)
+				bool metadata_only)
 {
 	struct bch_fs *c = trans->c;
 	struct bch_dev *ca = bch_dev_bkey_exists(c, iter->pos.inode);
@@ -1327,14 +1328,12 @@ static int bch2_alloc_write_key(struct btree_trans *trans,
 	if (IS_ERR(a))
 		return PTR_ERR(a);
 
-	ret = initial
-		? bch2_journal_key_insert(c, BTREE_ID_alloc, 0, &a->k)
-		: bch2_trans_update(trans, iter, &a->k, BTREE_TRIGGER_NORUN);
+	ret = bch2_trans_update(trans, iter, &a->k, BTREE_TRIGGER_NORUN);
 fsck_err:
 	return ret;
 }
 
-static int bch2_gc_alloc_done(struct bch_fs *c, bool initial, bool metadata_only)
+static int bch2_gc_alloc_done(struct bch_fs *c, bool metadata_only)
 {
 	struct btree_trans trans;
 	struct btree_iter iter;
@@ -1356,7 +1355,7 @@ static int bch2_gc_alloc_done(struct bch_fs *c, bool initial, bool metadata_only
 			ret = __bch2_trans_do(&trans, NULL, NULL,
 					      BTREE_INSERT_LAZY_RW,
 					bch2_alloc_write_key(&trans, &iter,
-							     initial, metadata_only));
+							     metadata_only));
 			if (ret)
 				break;
 		}
@@ -1373,7 +1372,7 @@ static int bch2_gc_alloc_done(struct bch_fs *c, bool initial, bool metadata_only
 	return ret;
 }
 
-static int bch2_gc_alloc_start(struct bch_fs *c, bool initial, bool metadata_only)
+static int bch2_gc_alloc_start(struct bch_fs *c, bool metadata_only)
 {
 	struct bch_dev *ca;
 	unsigned i;
@@ -1397,7 +1396,7 @@ static int bch2_gc_alloc_start(struct bch_fs *c, bool initial, bool metadata_onl
 	return bch2_alloc_read(c, true, metadata_only);
 }
 
-static void bch2_gc_alloc_reset(struct bch_fs *c, bool initial, bool metadata_only)
+static void bch2_gc_alloc_reset(struct bch_fs *c, bool metadata_only)
 {
 	struct bch_dev *ca;
 	unsigned i;
@@ -1418,8 +1417,7 @@ static void bch2_gc_alloc_reset(struct bch_fs *c, bool initial, bool metadata_on
 	};
 }
 
-static int bch2_gc_reflink_done(struct bch_fs *c, bool initial,
-				bool metadata_only)
+static int bch2_gc_reflink_done(struct bch_fs *c, bool metadata_only)
 {
 	struct btree_trans trans;
 	struct btree_iter iter;
@@ -1466,22 +1464,12 @@ static int bch2_gc_reflink_done(struct bch_fs *c, bool initial,
 
 			bkey_reassemble(new, k);
 
-			if (!r->refcount) {
+			if (!r->refcount)
 				new->k.type = KEY_TYPE_deleted;
-				/*
-				 * XXX ugly: bch2_journal_key_insert() queues up
-				 * the key for the journal replay code, which
-				 * doesn't run the extent overwrite pass
-				 */
-				if (initial)
-					new->k.size = 0;
-			} else {
+			else
 				*bkey_refcount(new) = cpu_to_le64(r->refcount);
-			}
 
-			ret = initial
-			       ? bch2_journal_key_insert(c, BTREE_ID_stripes, 0, new)
-			       : __bch2_trans_do(&trans, NULL, NULL, 0,
+			ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 				__bch2_btree_insert(&trans, BTREE_ID_reflink, new));
 			kfree(new);
 
@@ -1496,7 +1484,7 @@ static int bch2_gc_reflink_done(struct bch_fs *c, bool initial,
 	return ret;
 }
 
-static int bch2_gc_reflink_start(struct bch_fs *c, bool initial,
+static int bch2_gc_reflink_start(struct bch_fs *c,
 				 bool metadata_only)
 {
 	struct btree_trans trans;
@@ -1535,8 +1523,7 @@ static int bch2_gc_reflink_start(struct bch_fs *c, bool initial,
 	return ret;
 }
 
-static void bch2_gc_reflink_reset(struct bch_fs *c, bool initial,
-				  bool metadata_only)
+static void bch2_gc_reflink_reset(struct bch_fs *c, bool metadata_only)
 {
 	struct genradix_iter iter;
 	struct reflink_gc *r;
@@ -1545,8 +1532,7 @@ static void bch2_gc_reflink_reset(struct bch_fs *c, bool initial,
 		r->refcount = 0;
 }
 
-static int bch2_gc_stripes_done(struct bch_fs *c, bool initial,
-				bool metadata_only)
+static int bch2_gc_stripes_done(struct bch_fs *c, bool metadata_only)
 {
 	struct btree_trans trans;
 	struct btree_iter iter;
@@ -1594,9 +1580,7 @@ static int bch2_gc_stripes_done(struct bch_fs *c, bool initial,
 			for (i = 0; i < new->v.nr_blocks; i++)
 				stripe_blockcount_set(&new->v, i, m ? m->block_sectors[i] : 0);
 
-			ret = initial
-				? bch2_journal_key_insert(c, BTREE_ID_stripes, 0, &new->k_i)
-				: __bch2_trans_do(&trans, NULL, NULL, 0,
+			ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 				__bch2_btree_insert(&trans, BTREE_ID_reflink, &new->k_i));
 			kfree(new);
 		}
@@ -1608,8 +1592,7 @@ static int bch2_gc_stripes_done(struct bch_fs *c, bool initial,
 	return ret;
 }
 
-static void bch2_gc_stripes_reset(struct bch_fs *c, bool initial,
-				bool metadata_only)
+static void bch2_gc_stripes_reset(struct bch_fs *c, bool metadata_only)
 {
 	genradix_free(&c->gc_stripes);
 }
@@ -1649,8 +1632,8 @@ int bch2_gc(struct bch_fs *c, bool initial, bool metadata_only)
 			   !bch2_btree_interior_updates_nr_pending(c));
 
 	ret   = bch2_gc_start(c, metadata_only) ?:
-		bch2_gc_alloc_start(c, initial, metadata_only) ?:
-		bch2_gc_reflink_start(c, initial, metadata_only);
+		bch2_gc_alloc_start(c, metadata_only) ?:
+		bch2_gc_reflink_start(c, metadata_only);
 	if (ret)
 		goto out;
 again:
@@ -1705,9 +1688,9 @@ int bch2_gc(struct bch_fs *c, bool initial, bool metadata_only)
 		clear_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags);
 		__gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
 
-		bch2_gc_stripes_reset(c, initial, metadata_only);
-		bch2_gc_alloc_reset(c, initial, metadata_only);
-		bch2_gc_reflink_reset(c, initial, metadata_only);
+		bch2_gc_stripes_reset(c, metadata_only);
+		bch2_gc_alloc_reset(c, metadata_only);
+		bch2_gc_reflink_reset(c, metadata_only);
 
 		/* flush fsck errors, reset counters */
 		bch2_flush_fsck_errs(c);
@@ -1717,9 +1700,9 @@ int bch2_gc(struct bch_fs *c, bool initial, bool metadata_only)
 	if (!ret) {
 		bch2_journal_block(&c->journal);
 
-		ret   = bch2_gc_stripes_done(c, initial, metadata_only) ?:
-			bch2_gc_reflink_done(c, initial, metadata_only) ?:
-			bch2_gc_alloc_done(c, initial, metadata_only) ?:
+		ret   = bch2_gc_stripes_done(c, metadata_only) ?:
+			bch2_gc_reflink_done(c, metadata_only) ?:
+			bch2_gc_alloc_done(c, metadata_only) ?:
 			bch2_gc_done(c, initial, metadata_only);
 
 		bch2_journal_unblock(&c->journal);

fs/bcachefs/btree_update_leaf.c

@@ -968,6 +968,27 @@ static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
 	return 0;
 }
 
+/*
+ * This is for updates done in the early part of fsck - btree_gc - before we've
+ * gone RW. we only add the new key to the list of keys for journal replay to
+ * do.
+ */
+static noinline int
+do_bch2_trans_commit_to_journal_replay(struct btree_trans *trans)
+{
+	struct bch_fs *c = trans->c;
+	struct btree_insert_entry *i;
+	int ret = 0;
+
+	trans_for_each_update(trans, i) {
+		ret = bch2_journal_key_insert(c, i->btree_id, i->level, i->k);
+		if (ret)
+			break;
+	}
+
+	return ret;
+}
+
 int __bch2_trans_commit(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
@@ -986,6 +1007,11 @@ int __bch2_trans_commit(struct btree_trans *trans)
 	if (ret)
 		goto out_reset;
 
+	if (unlikely(!test_bit(BCH_FS_MAY_GO_RW, &c->flags))) {
+		ret = do_bch2_trans_commit_to_journal_replay(trans);
+		goto out_reset;
+	}
+
 	if (!(trans->flags & BTREE_INSERT_NOCHECK_RW) &&
 	    unlikely(!percpu_ref_tryget(&c->writes))) {
 		ret = bch2_trans_commit_get_rw_cold(trans);

fs/bcachefs/buckets.c

@@ -1164,9 +1164,11 @@ static int bch2_mark_reservation(struct btree_trans *trans,
 	return 0;
 }
 
-static s64 __bch2_mark_reflink_p(struct bch_fs *c, struct bkey_s_c_reflink_p p,
+static s64 __bch2_mark_reflink_p(struct btree_trans *trans,
+				 struct bkey_s_c_reflink_p p,
 				 u64 *idx, unsigned flags, size_t r_idx)
 {
+	struct bch_fs *c = trans->c;
 	struct reflink_gc *r;
 	int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
 	s64 ret = 0;
@@ -1199,7 +1201,7 @@ static s64 __bch2_mark_reflink_p(struct bch_fs *c, struct bkey_s_c_reflink_p p,
 		new.k.type	= KEY_TYPE_error;
 		new.k.p		= p.k->p;
 		new.k.size	= p.k->size;
-		ret = bch2_journal_key_insert(c, BTREE_ID_extents, 0, &new.k_i);
+		ret = __bch2_btree_insert(trans, BTREE_ID_extents, &new.k_i);
 	}
 fsck_err:
 	return ret;
@@ -1238,7 +1240,7 @@ static int bch2_mark_reflink_p(struct btree_trans *trans,
 	}
 
 	while (idx < end && !ret)
-		ret = __bch2_mark_reflink_p(c, p, &idx, flags, l++);
+		ret = __bch2_mark_reflink_p(trans, p, &idx, flags, l++);
 
 	return ret;
 }

fs/bcachefs/recovery.c

@@ -1157,6 +1157,7 @@ int bch2_fs_recovery(struct bch_fs *c)
 
 	clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 	set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
+	set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 
 	/*
 	 * Skip past versions that might have possibly been used (as nonces),
@@ -1317,6 +1318,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	mutex_unlock(&c->sb_lock);
 
 	set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
+	set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 	set_bit(BCH_FS_FSCK_DONE, &c->flags);
 
 	for (i = 0; i < BTREE_ID_NR; i++)