[PATCH bpf-next 01/11] bpf: Disable zero-extension for BPF_MEMSX

[Ilya Leoshkevich <iii@linux.ibm.com>]

On the architectures that use bpf_jit_needs_zext(), e.g., s390x, the
verifier incorrectly inserts a zero-extension after BPF_MEMSX, leading
to miscompilations like the one below:

      24:       89 1a ff fe 00 00 00 00 "r1 = *(s16 *)(r10 - 2);"       # zext_dst set
   0x3ff7fdb910e:       lgh     %r2,-2(%r13,%r0)                        # load halfword
   0x3ff7fdb9114:       llgfr   %r2,%r2                                 # wrong!
      25:       65 10 00 03 00 00 7f ff if r1 s> 32767 goto +3 <l0_1>   # check_cond_jmp_op()

Disable such zero-extensions. The JITs need to insert sign-extension
themselves, if necessary.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
---
 kernel/bpf/verifier.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index bb78212fa5b2..097985a46edc 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -3110,7 +3110,9 @@ static void mark_insn_zext(struct bpf_verifier_env *env,
 {
 	s32 def_idx = reg->subreg_def;
 
-	if (def_idx == DEF_NOT_SUBREG)
+	if (def_idx == DEF_NOT_SUBREG ||
+	    (BPF_CLASS(env->prog->insnsi[def_idx - 1].code) == BPF_LDX &&
+	     BPF_MODE(env->prog->insnsi[def_idx - 1].code) == BPF_MEMSX))
 		return;
 
 	env->insn_aux_data[def_idx - 1].zext_dst = true;
-- 
2.41.0