fusefs: fix error handling when reading a directory's sticky bit

When trying to delete or rename a file, fuse_vnop_lookup must check
whether its parent directory's sticky bit is set.  Realistically, the
parent directory's attributes will almost always be cached.  But it's
possible that they won't be, and in that case we must send a new
FUSE_GETATTR request to the server.  If that request fails for some
reason, then we must fail the lookup.  Prior to this change fusefs would
ignore failure of that request.

Reported by:	Yuxiang Yang, Yizhou Zhao, Ao Wang, Xuewei Feng, Qi Li,
                and Ke Xu of Tsinghua University
MFC after:	2 weeks
Reviewed by:	markj
Differential Revision: https://reviews.freebsd.org/D57588

2 files changed, 59 insertions, 4 deletions

diff --git a/sys/fs/fuse/fuse_vnops.c b/sys/fs/fuse/fuse_vnops.c
index 242e023a38dd..f5c832cbbd13 100644
--- a/sys/fs/fuse/fuse_vnops.c
+++ b/sys/fs/fuse/fuse_vnops.c
@@ -1743,14 +1743,16 @@ fuse_vnop_lookup(struct vop_lookup_args *ap)
 				 * Need to figure out the vnode locking to make
 				 * this work.
 				 */
-				fuse_internal_getattr(dvp, &dvattr, cred, td);
-				if ((dvattr.va_mode & S_ISTXT) &&
+				err = fuse_internal_getattr(dvp, &dvattr, cred,
+						td);
+				if (err == 0 &&
+					(dvattr.va_mode & S_ISTXT) &&
 					fuse_internal_access(dvp, VADMIN, td,
 						cred) &&
 					fuse_internal_access(*vpp, VADMIN, td,
-						cred)) {
+						cred))
+				{
 					err = EPERM;
-					goto out;
 				}
 			}
 		}
diff --git a/tests/sys/fs/fusefs/default_permissions.cc b/tests/sys/fs/fusefs/default_permissions.cc
index 4b04297d97ac..f9a37f904d30 100644
--- a/tests/sys/fs/fusefs/default_permissions.cc
+++ b/tests/sys/fs/fusefs/default_permissions.cc
@@ -1562,6 +1562,59 @@ TEST_F(Unlink, sticky_directory)
 	ASSERT_EQ(EPERM, errno);
 }
 
+/*
+ * When trying to delete or rename a file, we must check its parent directory's
+ * sticky bit.  That may entail a FUSE_GETATTR.  If that operation fails, then
+ * we must fail the rename or delete.
+ */
+TEST_F(Unlink, sticky_directory_io_during_getattr)
+{
+	const char FULLPATH[] = "mountpoint/some_file.txt";
+	const char RELPATH[] = "some_file.txt";
+	Sequence seq;
+	uint64_t ino = 42;
+
+	EXPECT_CALL(*m_mock, process(
+		ResultOf([=](auto in) {
+			return (in.header.opcode == FUSE_GETATTR &&
+				in.header.nodeid == FUSE_ROOT_ID);
+		}, Eq(true)),
+		_)
+	).Times(2)
+	.InSequence(seq)
+	.WillRepeatedly(Invoke(ReturnImmediate([=](auto i __unused, auto& out) {
+		SET_OUT_HEADER_LEN(out, attr);
+		out.body.attr.attr.ino = FUSE_ROOT_ID;
+		out.body.attr.attr.mode = S_IFDIR | 0177;
+		/*
+		 * Realistically, the parent directory's attributes will almost
+		 * always be cached when we try to lookup the sticky bit.  For
+		 * this test case, set attr_valid to 0 so that won't be the
+		 * case.
+		 */
+		out.body.attr.attr_valid = 0;
+	})));
+	EXPECT_CALL(*m_mock, process(
+		ResultOf([=](auto in) {
+			return (in.header.opcode == FUSE_GETATTR &&
+				in.header.nodeid == FUSE_ROOT_ID);
+		}, Eq(true)),
+		_)
+	).Times(1)
+	.InSequence(seq)
+	.WillOnce(Invoke(ReturnErrno(EIO)));
+	expect_lookup(RELPATH, ino, S_IFREG | 0644, UINT64_MAX, 0);
+	EXPECT_CALL(*m_mock, process(
+		ResultOf([=](auto in) {
+			return (in.header.opcode == FUSE_UNLINK);
+		}, Eq(true)),
+		_)
+	).Times(0);
+
+	ASSERT_EQ(-1, unlink(FULLPATH));
+	ASSERT_EQ(EIO, errno);
+}
+
 /* A write by a non-owner should clear a file's SUID bit */
 TEST_F(Write, clear_suid)
 {