AF_UNIX crash in bind() triggered by strace test

[Sven Schnelle]

Hi,

the following crash was reported in our CI on s390x with
debug_defconfig, which enables FORTIFY_SOURCE:

detected buffer overflow in __fortify_strlen
------------[ cut here ]------------
kernel BUG at lib/string_helpers.c:1031!
monitor event: 0040 ilc:2 [#1] PREEMPT SMP 
Modules linked in: [..]
CPU: 0 PID: 243755 Comm: net-accept-conn Not tainted 6.5.0-20230719.rc2.git1.f1311c9e1695.300.fc38.s390x+debug #1
Call Trace:
 [<000000003465b7a2>] fortify_panic+0x2a/0x30 
([<000000003465b79e>] fortify_panic+0x26/0x30)
 [<0000000034a3a77e>] unix_bind_bsd+0x86/0x390 
 [<00000000348839d0>] __sys_bind+0xe0/0xe8 
 [<0000000034926270>] __do_compat_sys_socketcall+0x260/0x4d0 
 [<0000000034be3b66>] __do_syscall+0x1de/0x208 
 [<0000000034bfaf38>] system_call+0x70/0x98 
INFO: lockdep is turned off.
Last Breaking-Event-Address:
 [<0000000034774880>] __s390_indirect_jump_r14+0x0/0x10
Kernel panic - not syncing: Fatal exception: panic_on_oops

This is caused by a test case who sends an unterminated sun_path
to the kernel in a bind() system call from the strace test suite.
As a test i made the following quick fix, which "fixed" the issue:

diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 123b35ddfd71..c505edd74d8c 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -1206,11 +1206,13 @@ static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
 	struct unix_address *addr;
 	struct dentry *dentry;
 	struct path parent;
-	int err;
+	int pathlen, err;
 
 	unix_mkname_bsd(sunaddr, addr_len);
-	addr_len = strlen(sunaddr->sun_path) +
-		offsetof(struct sockaddr_un, sun_path) + 1;
+	pathlen = strnlen(sunaddr->sun_path, UNIX_PATH_MAX);
+	if (pathlen == UNIX_PATH_MAX)
+		return -EINVAL;
+	addr_len = pathlen + offsetof(struct sockaddr_un, sun_path) + 1;
 
 	addr = unix_create_addr(sunaddr, addr_len);
 	if (!addr)

However, unix(7) says "The pathname in sun_path *should* be
null-terminated." So this change might break userspace. I'm not sure
whether we should return -EINVAL, or just truncate the name to
UNIX_PATH_MAX. From a quick read, it looks like connect() would trigger
the same problem. Any thoughts?