bcachefs: Start copygc when first going read-write

In the distant past, it wasn't possible to start copygc until after
journal replay had finished. Now, the btree iterator code overlays keys
from the journal, so there's no reason not to start it earlier - and it
solves a rare deadlock.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/recovery.c

@@ -1267,6 +1267,20 @@ int bch2_fs_recovery(struct bch_fs *c)
 			goto err;
 		bch_verbose(c, "done checking need_discard and freespace btrees");
 
+		if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
+			err = "error creating root snapshot node";
+			ret = bch2_fs_initialize_subvolumes(c);
+			if (ret)
+				goto err;
+		}
+
+		bch_verbose(c, "reading snapshots table");
+		err = "error reading snapshots table";
+		ret = bch2_fs_snapshots_start(c);
+		if (ret)
+			goto err;
+		bch_verbose(c, "reading snapshots done");
+
 		set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 
 		bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
@@ -1293,7 +1307,6 @@ int bch2_fs_recovery(struct bch_fs *c)
 		bch_verbose(c, "done checking alloc to lru refs");
 		set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
 	} else {
-		set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 		set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 		set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
 		set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
@@ -1302,35 +1315,35 @@ int bch2_fs_recovery(struct bch_fs *c)
 		if (c->opts.norecovery)
 			goto out;
 
-		bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
-		err = "journal replay failed";
-		ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
+		if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
+			err = "error creating root snapshot node";
+			ret = bch2_fs_initialize_subvolumes(c);
 			if (ret)
 				goto err;
-		if (c->opts.verbose || !c->sb.clean)
-			bch_info(c, "journal replay done");
 		}
 
-	err = "error initializing freespace";
-	ret = bch2_fs_freespace_init(c);
+		bch_verbose(c, "reading snapshots table");
+		err = "error reading snapshots table";
+		ret = bch2_fs_snapshots_start(c);
 		if (ret)
 			goto err;
+		bch_verbose(c, "reading snapshots done");
 
-	if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
-		bch2_fs_lazy_rw(c);
+		set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 
-		err = "error creating root snapshot node";
-		ret = bch2_fs_initialize_subvolumes(c);
+		bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
+		err = "journal replay failed";
+		ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
 		if (ret)
 			goto err;
+		if (c->opts.verbose || !c->sb.clean)
+			bch_info(c, "journal replay done");
 	}
 
-	bch_verbose(c, "reading snapshots table");
-	err = "error reading snapshots table";
-	ret = bch2_fs_snapshots_start(c);
+	err = "error initializing freespace";
+	ret = bch2_fs_freespace_init(c);
 	if (ret)
 		goto err;
-	bch_verbose(c, "reading snapshots done");
 
 	if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
 		/* set bi_subvol on root inode */

fs/bcachefs/super.c

@@ -332,26 +332,12 @@ static int bch2_fs_read_write_late(struct bch_fs *c)
 {
 	int ret;
 
-	ret = bch2_gc_thread_start(c);
-	if (ret) {
-		bch_err(c, "error starting gc thread");
-		return ret;
-	}
-
-	ret = bch2_copygc_start(c);
-	if (ret) {
-		bch_err(c, "error starting copygc thread");
-		return ret;
-	}
-
 	ret = bch2_rebalance_start(c);
 	if (ret) {
 		bch_err(c, "error starting rebalance thread");
 		return ret;
 	}
 
-	schedule_work(&c->ec_stripe_delete_work);
-
 	return 0;
 }
 
@@ -398,6 +384,18 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 		bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);
 
+	ret = bch2_gc_thread_start(c);
+	if (ret) {
+		bch_err(c, "error starting gc thread");
+		return ret;
+	}
+
+	ret = bch2_copygc_start(c);
+	if (ret) {
+		bch_err(c, "error starting copygc thread");
+		return ret;
+	}
+
 	if (!early) {
 		ret = bch2_fs_read_write_late(c);
 		if (ret)