fusefs: handle evil servers that return illegal inode numbers

* If during FUSE_CREATE, FUSE_MKDIR, etc the server returns the same
  inode number for the new file as for its parent directory, reject it.
  Previously this would triggers a recurse-on-non-recursive lock panic.

* If during FUSE_LINK the server returns a different inode number for
  the new name as for the old one, reject it.  Obviously, that can't be
  a hard link.

* If during FUSE_LOOKUP the server returns the same inode number for the
  new file as for its parent directory, reject it.  Nothing good can
  come of this.

PR:		263662
Reported by:	Robert Morris <rtm@lcs.mit.edu>
MFC after:	2 weeks
Reviewed by:	pfg
Differential Revision: https://reviews.freebsd.org/D35128

6 files changed, 226 insertions, 0 deletions

diff --git a/tests/sys/fs/fusefs/create.cc b/tests/sys/fs/fusefs/create.cc
index df3225ed1837..9f5820a00b3a 100644
--- a/tests/sys/fs/fusefs/create.cc
+++ b/tests/sys/fs/fusefs/create.cc
@@ -371,6 +371,47 @@ TEST_F(Create, ok)
 }
 
 /*
+ * Nothing bad should happen if the server returns the parent's inode number
+ * for the newly created file.  Regression test for bug 263662
+ * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=263662
+ */
+TEST_F(Create, parent_inode)
+{
+	const char FULLPATH[] = "mountpoint/some_dir/some_file.txt";
+	const char RELDIRPATH[] = "some_dir";
+	const char RELPATH[] = "some_file.txt";
+	mode_t mode = 0755;
+	uint64_t ino = 42;
+	int fd;
+
+	expect_lookup(RELDIRPATH, ino, S_IFDIR | mode, 0, 1);
+	EXPECT_LOOKUP(ino, RELPATH)
+		.WillOnce(Invoke(ReturnErrno(ENOENT)));
+	expect_create(RELPATH, S_IFREG | mode,
+		ReturnImmediate([=](auto in __unused, auto& out) {
+		SET_OUT_HEADER_LEN(out, create);
+		out.body.create.entry.attr.mode = S_IFREG | mode;
+		/* Return the same inode as the parent dir */
+		out.body.create.entry.nodeid = ino;
+		out.body.create.entry.entry_valid = UINT64_MAX;
+		out.body.create.entry.attr_valid = UINT64_MAX;
+	}));
+	// FUSE_RELEASE happens asynchronously, so it may or may not arrive
+	// before the test completes.
+	EXPECT_CALL(*m_mock, process(
+		ResultOf([=](auto in) {
+			return (in.header.opcode == FUSE_RELEASE);
+		}, Eq(true)),
+		_)
+	).Times(AtMost(1))
+	.WillOnce(Invoke([=](auto in __unused, auto &out __unused) { }));
+
+	fd = open(FULLPATH, O_CREAT | O_EXCL, mode);
+	ASSERT_EQ(-1, fd);
+	EXPECT_EQ(EIO, errno);
+}
+
+/*
  * A regression test for a bug that affected old FUSE implementations:
  * open(..., O_WRONLY | O_CREAT, 0444) should work despite the seeming
  * contradiction between O_WRONLY and 0444
diff --git a/tests/sys/fs/fusefs/link.cc b/tests/sys/fs/fusefs/link.cc
index 3d9a5a4e0e8f..789d3dcc3494 100644
--- a/tests/sys/fs/fusefs/link.cc
+++ b/tests/sys/fs/fusefs/link.cc
@@ -176,6 +176,53 @@ TEST_F(Link, emlink)
 	EXPECT_EQ(EMLINK, errno);
 }
 
+/*
+ * A hard link should always have the same inode as its source.  If it doesn't,
+ * then it's not a hard link.
+ */
+TEST_F(Link, bad_inode)
+{
+	const char FULLPATH[] = "mountpoint/src";
+	const char RELPATH[] = "src";
+	const char FULLDST[] = "mountpoint/dst";
+	const char RELDST[] = "dst";
+	const uint64_t src_ino = 42;
+	const uint64_t dst_ino = 43;
+	mode_t mode = S_IFREG | 0644;
+
+	EXPECT_LOOKUP(FUSE_ROOT_ID, RELPATH)
+	.WillOnce(Invoke(ReturnErrno(ENOENT)));
+	EXPECT_LOOKUP(FUSE_ROOT_ID, RELDST)
+	.WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto& out) {
+		SET_OUT_HEADER_LEN(out, entry);
+		out.body.entry.attr.mode = mode;
+		out.body.entry.nodeid = dst_ino;
+		out.body.entry.attr.nlink = 1;
+		out.body.entry.attr_valid = UINT64_MAX;
+		out.body.entry.entry_valid = UINT64_MAX;
+	})));
+	EXPECT_CALL(*m_mock, process(
+		ResultOf([=](auto in) {
+			const char *name = (const char*)in.body.bytes
+				+ sizeof(struct fuse_link_in);
+			return (in.header.opcode == FUSE_LINK &&
+				in.body.link.oldnodeid == dst_ino &&
+				(0 == strcmp(name, RELPATH)));
+		}, Eq(true)),
+		_)
+	).WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto& out) {
+		SET_OUT_HEADER_LEN(out, entry);
+		out.body.entry.nodeid = src_ino;
+		out.body.entry.attr.mode = mode;
+		out.body.entry.attr.nlink = 2;
+		out.body.entry.attr_valid = UINT64_MAX;
+		out.body.entry.entry_valid = UINT64_MAX;
+	})));
+
+	ASSERT_EQ(-1, link(FULLDST, FULLPATH));
+	ASSERT_EQ(EIO, errno);
+}
+
 TEST_F(Link, ok)
 {
 	const char FULLPATH[] = "mountpoint/src";
diff --git a/tests/sys/fs/fusefs/lookup.cc b/tests/sys/fs/fusefs/lookup.cc
index 0ec02913f66a..c654dd46bae5 100644
--- a/tests/sys/fs/fusefs/lookup.cc
+++ b/tests/sys/fs/fusefs/lookup.cc
@@ -430,6 +430,37 @@ TEST_F(Lookup, ok)
 	ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno);
 }
 
+/*
+ * Lookup in a subdirectory of the fuse mount.  The naughty server returns the
+ * same inode for the child as for the parent.
+ */
+TEST_F(Lookup, parent_inode)
+{
+	const char FULLPATH[] = "mountpoint/some_dir/some_file.txt";
+	const char DIRPATH[] = "some_dir";
+	const char RELPATH[] = "some_file.txt";
+	uint64_t dir_ino = 2;
+
+	EXPECT_LOOKUP(FUSE_ROOT_ID, DIRPATH)
+	.WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto& out) {
+		SET_OUT_HEADER_LEN(out, entry);
+		out.body.entry.attr.mode = S_IFDIR | 0755;
+		out.body.entry.nodeid = dir_ino;
+	})));
+	EXPECT_LOOKUP(dir_ino, RELPATH)
+	.WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto& out) {
+		SET_OUT_HEADER_LEN(out, entry);
+		out.body.entry.attr.mode = S_IFREG | 0644;
+		out.body.entry.nodeid = dir_ino;
+	})));
+	/*
+	 * access(2) is one of the few syscalls that will not (always) follow
+	 * up a successful VOP_LOOKUP with another VOP.
+	 */
+	ASSERT_EQ(-1, access(FULLPATH, F_OK));
+	ASSERT_EQ(EIO, errno);
+}
+
 // Lookup in a subdirectory of the fuse mount
 TEST_F(Lookup, subdir)
 {
diff --git a/tests/sys/fs/fusefs/mkdir.cc b/tests/sys/fs/fusefs/mkdir.cc
index 45efd08cfc80..f47189d9bf53 100644
--- a/tests/sys/fs/fusefs/mkdir.cc
+++ b/tests/sys/fs/fusefs/mkdir.cc
@@ -193,6 +193,59 @@ TEST_F(Mkdir, ok)
 	ASSERT_EQ(0, mkdir(FULLPATH, mode)) << strerror(errno);
 }
 
+/*
+ * Nothing bad should happen if the server returns the parent's inode number
+ * for the newly created directory.  Regression test for bug 263662.
+ * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=263662
+ */
+TEST_F(Mkdir, parent_inode)
+{
+	const char FULLPATH[] = "mountpoint/parent/some_dir";
+	const char PPATH[] = "parent";
+	const char RELPATH[] = "some_dir";
+	mode_t mode = 0755;
+	uint64_t ino = 42;
+	mode_t mask;
+
+	mask = umask(0);
+	(void)umask(mask);
+
+	expect_lookup(PPATH, ino, S_IFDIR | 0755, 0, 1);
+	EXPECT_LOOKUP(ino, RELPATH)
+	.WillOnce(Invoke(ReturnErrno(ENOENT)));
+
+	EXPECT_CALL(*m_mock, process(
+		ResultOf([=](auto in) {
+			const char *name = (const char*)in.body.bytes +
+				sizeof(fuse_mkdir_in);
+			return (in.header.opcode == FUSE_MKDIR &&
+				in.body.mkdir.mode == (S_IFDIR | mode) &&
+				in.body.mkdir.umask == mask &&
+				(0 == strcmp(RELPATH, name)));
+		}, Eq(true)),
+		_)
+	).WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto& out) {
+		SET_OUT_HEADER_LEN(out, entry);
+		out.body.create.entry.attr.mode = S_IFDIR | mode;
+		out.body.create.entry.nodeid = ino;
+		out.body.create.entry.entry_valid = UINT64_MAX;
+		out.body.create.entry.attr_valid = UINT64_MAX;
+	})));
+	// FUSE_FORGET happens asynchronously, so it may or may not arrive
+	// before the test completes.
+	EXPECT_CALL(*m_mock, process(
+		ResultOf([=](auto in) {
+			return (in.header.opcode == FUSE_FORGET);
+		}, Eq(true)),
+		_)
+	).Times(AtMost(1))
+	.WillOnce(Invoke([=](auto in __unused, auto &out __unused) { }));
+
+	ASSERT_EQ(-1, mkdir(FULLPATH, mode));
+	ASSERT_EQ(EIO, errno);
+	usleep(100000);
+}
+
 TEST_F(Mkdir_7_8, ok)
 {
 	const char FULLPATH[] = "mountpoint/some_dir";
diff --git a/tests/sys/fs/fusefs/mknod.cc b/tests/sys/fs/fusefs/mknod.cc
index 7bb8e116c3f8..75f7c2c46adc 100644
--- a/tests/sys/fs/fusefs/mknod.cc
+++ b/tests/sys/fs/fusefs/mknod.cc
@@ -245,6 +245,38 @@ TEST_F(Mknod, socket)
 	leak(fd);
 }
 
+/*
+ * Nothing bad should happen if the server returns the parent's inode number
+ * for the newly created file.  Regression test for bug 263662.
+ * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=263662
+ */
+TEST_F(Mknod, parent_inode)
+{
+	const char FULLPATH[] = "mountpoint/parent/some_node";
+	const char PPATH[] = "parent";
+	const char RELPATH[] = "some_node";
+	mode_t mode = S_IFSOCK | 0755;
+	struct sockaddr_un sa;
+	int fd;
+	dev_t rdev = -1;	/* Really it's a don't care */
+	uint64_t ino = 42;
+
+	expect_lookup(PPATH, ino, S_IFDIR | 0755, 0, 1);
+	EXPECT_LOOKUP(ino, RELPATH)
+	.WillOnce(Invoke(ReturnErrno(ENOENT)));
+	expect_mknod(ino, RELPATH, ino, mode, rdev);
+
+	fd = socket(AF_UNIX, SOCK_STREAM, 0);
+	ASSERT_LE(0, fd) << strerror(errno);
+	sa.sun_family = AF_UNIX;
+	strlcpy(sa.sun_path, FULLPATH, sizeof(sa.sun_path));
+	sa.sun_len = sizeof(FULLPATH);
+	ASSERT_EQ(-1, bind(fd, (struct sockaddr*)&sa, sizeof(sa)));
+	ASSERT_EQ(EIO, errno);
+
+	leak(fd);
+}
+
 /* 
  * fusefs(5) lacks VOP_WHITEOUT support.  No bugzilla entry, because that's a
  * feature, not a bug
diff --git a/tests/sys/fs/fusefs/symlink.cc b/tests/sys/fs/fusefs/symlink.cc
index 1ee5f79f91fb..bef06c90c3db 100644
--- a/tests/sys/fs/fusefs/symlink.cc
+++ b/tests/sys/fs/fusefs/symlink.cc
@@ -165,6 +165,28 @@ TEST_F(Symlink, ok)
 	EXPECT_EQ(0, symlink(dst, FULLPATH)) << strerror(errno);
 }
 
+/*
+ * Nothing bad should happen if the server returns the parent's inode number
+ * for the newly created symlink.  Regression test for bug 263662.
+ * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=263662
+ */
+TEST_F(Symlink, parent_ino)
+{
+	const char FULLPATH[] = "mountpoint/parent/src";
+	const char PPATH[] = "parent";
+	const char RELPATH[] = "src";
+	const char dst[] = "dst";
+	const uint64_t ino = 42;
+
+	expect_lookup(PPATH, ino, S_IFDIR | 0755, 0, 1);
+	EXPECT_LOOKUP(ino, RELPATH)
+	.WillOnce(Invoke(ReturnErrno(ENOENT)));
+	expect_symlink(ino, dst, RELPATH);
+
+	EXPECT_EQ(-1, symlink(dst, FULLPATH));
+	EXPECT_EQ(EIO, errno);
+}
+
 TEST_F(Symlink_7_8, ok)
 {
 	const char FULLPATH[] = "mountpoint/src";