lsm: make security_socket_getpeersec_stream() sockptr_t safe
authorPaul Moore <paul@paul-moore.com>
Mon, 10 Oct 2022 16:31:21 +0000 (12:31 -0400)
committerPaul Moore <paul@paul-moore.com>
Sat, 5 Nov 2022 03:25:30 +0000 (23:25 -0400)
commitb10b9c342f7571f287fd422be5d5c0beb26ba974
treee2afb492a670ba89d761a4f94d875cf0a171b17e
parent610b17b05c5c682fbb8fefedae1aacaab412eac3
lsm: make security_socket_getpeersec_stream() sockptr_t safe

Commit 4ff09db1b79b ("bpf: net: Change sk_getsockopt() to take the
sockptr_t argument") made it possible to call sk_getsockopt()
with both user and kernel address space buffers through the use of
the sockptr_t type.  Unfortunately at the time of conversion the
security_socket_getpeersec_stream() LSM hook was written to only
accept userspace buffers, and in a desire to avoid having to change
the LSM hook the commit author simply passed the sockptr_t's
userspace buffer pointer.  Since the only sk_getsockopt() callers
at the time of conversion which used kernel sockptr_t buffers did
not allow SO_PEERSEC, and hence the
security_socket_getpeersec_stream() hook, this was acceptable but
also very fragile as future changes presented the possibility of
silently passing kernel space pointers to the LSM hook.

There are several ways to protect against this, including careful
code review of future commits, but since relying on code review to
catch bugs is a recipe for disaster and the upstream eBPF maintainer
is "strongly against defensive programming", this patch updates the
LSM hook, and all of the implementations to support sockptr_t and
safely handle both user and kernel space buffers.

Acked-by: Casey Schaufler <casey@schaufler-ca.com>
Acked-by: John Johansen <john.johansen@canonical.com>
Signed-off-by: Paul Moore <paul@paul-moore.com>
include/linux/lsm_hook_defs.h
include/linux/lsm_hooks.h
include/linux/security.h
net/core/sock.c
security/apparmor/lsm.c
security/security.c
security/selinux/hooks.c
security/smack/smack_lsm.c