bcachefs: Fix bch2_delete_dead_inodes()

 - the fsck_err() check for the filesystem being clean was incorrect,
   causing us to always fail to delete unlinked inodes
 - if a snapshot had been taken, the unlinked inode needs to be
   propagated to snapshot leaves so the unlink can happen there - fixed.

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

2 files changed, 69 insertions, 22 deletions

diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index 925d1b7f2887..789f8958f685 100644
--- a/fs/bcachefs/inode.c
+++ b/fs/bcachefs/inode.c
@@ -362,9 +362,10 @@ int bch2_inode_peek(struct btree_trans *trans,
 	return ret;
 }
 
-int bch2_inode_write(struct btree_trans *trans,
+int bch2_inode_write_flags(struct btree_trans *trans,
 		     struct btree_iter *iter,
-		     struct bch_inode_unpacked *inode)
+		     struct bch_inode_unpacked *inode,
+		     enum btree_update_flags flags)
 {
 	struct bkey_inode_buf *inode_p;
 
@@ -374,7 +375,7 @@ int bch2_inode_write(struct btree_trans *trans,
 
 	bch2_inode_pack_inlined(inode_p, inode);
 	inode_p->inode.k.p.snapshot = iter->snapshot;
-	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
+	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags);
 }
 
 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
@@ -1055,24 +1056,18 @@ err:
 	return ret ?: -BCH_ERR_transaction_restart_nested;
 }
 
-static int may_delete_deleted_inode(struct btree_trans *trans, struct bpos pos)
+static int may_delete_deleted_inode(struct btree_trans *trans,
+				    struct btree_iter *iter,
+				    struct bpos pos,
+				    bool *need_another_pass)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter iter;
+	struct btree_iter inode_iter;
 	struct bkey_s_c k;
 	struct bch_inode_unpacked inode;
 	int ret;
 
-	if (bch2_snapshot_is_internal_node(c, pos.snapshot))
-		return 0;
-
-	if (!fsck_err_on(c->sb.clean, c,
-			 deleted_inode_but_clean,
-			 "filesystem marked as clean but have deleted inode %llu:%u",
-			 pos.offset, pos.snapshot))
-		return 0;
-
-	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
+	k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
 	ret = bkey_err(k);
 	if (ret)
 		return ret;
@@ -1086,7 +1081,7 @@ static int may_delete_deleted_inode(struct btree_trans *trans, struct bpos pos)
 
 	ret = bch2_inode_unpack(k, &inode);
 	if (ret)
-		goto err;
+		goto out;
 
 	if (fsck_err_on(S_ISDIR(inode.bi_mode), c,
 			deleted_inode_is_dir,
@@ -1100,12 +1095,46 @@ static int may_delete_deleted_inode(struct btree_trans *trans, struct bpos pos)
 			pos.offset, pos.snapshot))
 		goto delete;
 
-	return 1;
-err:
+	if (c->sb.clean &&
+	    !fsck_err(c,
+		      deleted_inode_but_clean,
+		      "filesystem marked as clean but have deleted inode %llu:%u",
+		      pos.offset, pos.snapshot)) {
+		ret = 0;
+		goto out;
+	}
+
+	if (bch2_snapshot_is_internal_node(c, pos.snapshot)) {
+		struct bpos new_min_pos;
+
+		ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos);
+		if (ret)
+			goto out;
+
+		inode.bi_flags &= ~BCH_INODE_UNLINKED;
+
+		ret = bch2_inode_write_flags(trans, &inode_iter, &inode,
+					     BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+		bch_err_msg(c, ret, "clearing inode unlinked flag");
+		if (ret)
+			goto out;
+
+		/*
+		 * We'll need another write buffer flush to pick up the new
+		 * unlinked inodes in the snapshot leaves:
+		 */
+		*need_another_pass = true;
+		return 0;
+	}
+
+	ret = 1;
+out:
 fsck_err:
+	bch2_trans_iter_exit(trans, &inode_iter);
 	return ret;
 delete:
-	return bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
+	ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
+	goto out;
 }
 
 int bch2_delete_dead_inodes(struct bch_fs *c)
@@ -1113,7 +1142,10 @@ int bch2_delete_dead_inodes(struct bch_fs *c)
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct btree_iter iter;
 	struct bkey_s_c k;
+	bool need_another_pass;
 	int ret;
+again:
+	need_another_pass = false;
 
 	ret = bch2_btree_write_buffer_flush_sync(trans);
 	if (ret)
@@ -1127,7 +1159,8 @@ int bch2_delete_dead_inodes(struct bch_fs *c)
 	 */
 	for_each_btree_key(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
 			   BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-		ret = lockrestart_do(trans, may_delete_deleted_inode(trans, k.k->p));
+		ret = lockrestart_do(trans, may_delete_deleted_inode(trans, &iter, k.k->p,
+								     &need_another_pass));
 		if (ret < 0)
 			break;
 
@@ -1137,12 +1170,17 @@ int bch2_delete_dead_inodes(struct bch_fs *c)
 				bch2_fs_lazy_rw(c);
 			}
 
+			bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot);
+
 			ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
 			if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 				break;
 		}
 	}
 	bch2_trans_iter_exit(trans, &iter);
+
+	if (!ret && need_another_pass)
+		goto again;
 err:
 	bch2_trans_put(trans);
 
diff --git a/fs/bcachefs/inode.h b/fs/bcachefs/inode.h
index 74c62e6c16cc..b09fda452d6f 100644
--- a/fs/bcachefs/inode.h
+++ b/fs/bcachefs/inode.h
@@ -3,6 +3,7 @@
 #define _BCACHEFS_INODE_H
 
 #include "bkey.h"
+#include "bkey_methods.h"
 #include "opts.h"
 
 enum bkey_invalid_flags;
@@ -101,8 +102,16 @@ void bch2_inode_unpacked_to_text(struct printbuf *, struct bch_inode_unpacked *)
 
 int bch2_inode_peek(struct btree_trans *, struct btree_iter *,
 		    struct bch_inode_unpacked *, subvol_inum, unsigned);
-int bch2_inode_write(struct btree_trans *, struct btree_iter *,
-		     struct bch_inode_unpacked *);
+
+int bch2_inode_write_flags(struct btree_trans *, struct btree_iter *,
+		     struct bch_inode_unpacked *, enum btree_update_flags);
+
+static inline int bch2_inode_write(struct btree_trans *trans,
+		     struct btree_iter *iter,
+		     struct bch_inode_unpacked *inode)
+{
+	return bch2_inode_write_flags(trans, iter, inode, 0);
+}
 
 void bch2_inode_init_early(struct bch_fs *,
 			   struct bch_inode_unpacked *);