Re: [PATCH bpf-next v2] bpf: check attach_func_proto more carefully in check_return_code

[Martin KaFai Lau <kafai@xxxxxx>]

On Thu, Jul 07, 2022 at 12:18:33PM -0700, sdf@xxxxxxxxxx wrote:
> On 07/07, Martin KaFai Lau wrote:
> > On Thu, Jul 07, 2022 at 09:02:33AM -0700, Stanislav Fomichev wrote:
> > > Syzkaller reports the following crash:
> > > RIP: 0010:check_return_code kernel/bpf/verifier.c:10575 [inline]
> > > RIP: 0010:do_check kernel/bpf/verifier.c:12346 [inline]
> > > RIP: 0010:do_check_common+0xb3d2/0xd250 kernel/bpf/verifier.c:14610
> > >
> > > With the following reproducer:
> > > bpf$PROG_LOAD_XDP(0x5, &(0x7f00000004c0)={0xd, 0x3,
> > &(0x7f0000000000)=ANY=[@ANYBLOB="1800000000000019000000000000000095"],
> > &(0x7f0000000300)='GPL\x00', 0x0, 0x0, 0x0, 0x0, 0x0, '\x00', 0x0, 0x2b,
> > 0xffffffffffffffff, 0x8, 0x0, 0x0, 0x10, 0x0}, 0x80)
> > >
> > > Because we don't enforce expected_attach_type for XDP programs,
> > > we end up in hitting 'if (prog->expected_attach_type == BPF_LSM_CGROUP'
> > > part in check_return_code and follow up with testing
> > > `prog->aux->attach_func_proto->type`, but `prog->aux->attach_func_proto`
> > > is NULL.
> > >
> > > Add explicit prog_type check for the "Note, BPF_LSM_CGROUP that
> > > attach ..." condition. Also, don't skip return code check for
> > > LSM/STRUCT_OPS.
> > >
> > > The above actually brings an issue with existing selftest which
> > > tries to return EPERM from void inet_csk_clone. Fix the
> > > test (and move called_socket_clone to make sure it's not
> > > incremented in case of an error) and add a new one to explicitly
> > > verify this condition.
> > >
> > > v2:
> > > - Martin: don't add new helper, check prog_type instead
> > > - Martin: check expected_attach_type as well at the function entry
> > > - Update selftest to verify this condition
> > >
> > > Fixes: 69fd337a975c ("bpf: per-cgroup lsm flavor")
> > > Reported-by: syzbot+5cc0730bd4b4d2c5f152@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > Signed-off-by: Stanislav Fomichev <sdf@xxxxxxxxxx>
> > > ---
> > >  kernel/bpf/verifier.c                              |  2 ++
> > >  .../testing/selftests/bpf/prog_tests/lsm_cgroup.c  | 12 ++++++++++++
> > >  tools/testing/selftests/bpf/progs/lsm_cgroup.c     | 12 ++++++------
> > >  .../selftests/bpf/progs/lsm_cgroup_nonvoid.c       | 14 ++++++++++++++
> > >  4 files changed, 34 insertions(+), 6 deletions(-)
> > >  create mode 100644
> > tools/testing/selftests/bpf/progs/lsm_cgroup_nonvoid.c
> > >
> > > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > > index df3ec6b05f05..2bc1e7252778 100644
> > > --- a/kernel/bpf/verifier.c
> > > +++ b/kernel/bpf/verifier.c
> > > @@ -10445,6 +10445,7 @@ static int check_return_code(struct
> > bpf_verifier_env *env)
> > >
> > >  	/* LSM and struct_ops func-ptr's return type could be "void" */
> > >  	if (!is_subprog &&
> > > +	    prog->expected_attach_type != BPF_LSM_CGROUP &&
> > BPF_PROG_TYPE_STRUCT_OPS also uses the expected_attach_type,
> > so the expected_attach_type check should only be done for LSM prog alone.
> > Others lgtm.
> 
> In this case, something like the following should be sufficient?
> 
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 2bc1e7252778..6702a5fc12e6 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -10445,11 +10445,13 @@ static int check_return_code(struct
> bpf_verifier_env *env)
> 
>  	/* LSM and struct_ops func-ptr's return type could be "void" */
>  	if (!is_subprog &&
> -	    prog->expected_attach_type != BPF_LSM_CGROUP &&
> -	    (prog_type == BPF_PROG_TYPE_STRUCT_OPS ||
> -	     prog_type == BPF_PROG_TYPE_LSM) &&
> -	    !prog->aux->attach_func_proto->type)
> -		return 0;
> +	    !prog->aux->attach_func_proto->type) {
prog_type check has to be done first since prog->aux->attach_func_proto
depends on the prog_type.

How about a small tweak on top of yours ?

	/* LSM and struct_ops func-ptr's return type could be "void" */
	if (!is_subprog) {
		switch (prog_type) {
		case BPF_PROG_TYPE_LSM:
			if (prog->expected_attach_type == BPF_LSM_CGROUP)
				/* cgroup prog needs to return 0 or 1 */
				break;
			fallthrough;
		case BPF_PROG_TYPE_STRUCT_OPS:
			if (!prog->aux->attach_func_proto->type)
				return 0;
			break;
		default:
			break;
		}
	}
	    
> +		if (prog_type == BPF_PROG_TYPE_STRUCT_OPS)
> +			return 0;
> +		if (prog_type == BPF_PROG_TYPE_LSM &&
> +		    prog->expected_attach_type != BPF_LSM_CGROUP)
> +			return 0;
> +	}
> 
>  	/* eBPF calling convention is such that R0 is used
>  	 * to return the value from eBPF program.
> 
> > >  	    (prog_type == BPF_PROG_TYPE_STRUCT_OPS ||
> > >  	     prog_type == BPF_PROG_TYPE_LSM) &&
> > >  	    !prog->aux->attach_func_proto->type)
> > > @@ -10572,6 +10573,7 @@ static int check_return_code(struct
> > bpf_verifier_env *env)
> > >  	if (!tnum_in(range, reg->var_off)) {
> > >  		verbose_invalid_scalar(env, reg, &range, "program exit", "R0");
> > >  		if (prog->expected_attach_type == BPF_LSM_CGROUP &&
> > > +		    prog_type == BPF_PROG_TYPE_LSM &&
> > >  		    !prog->aux->attach_func_proto->type)
> > >  			verbose(env, "Note, BPF_LSM_CGROUP that attach to void LSM hooks
> > can't modify return value!\n");
> > >  		return -EINVAL;