bpf, s390x: do not reload skb pointers in non-skb context
authorDaniel Borkmann <daniel@iogearbox.net>
Fri, 22 Dec 2017 15:23:01 +0000 (16:23 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 25 Dec 2017 13:26:31 +0000 (14:26 +0100)
[ Upstream commit 6d59b7dbf72ed20d0138e2f9b75ca3d4a9d4faca ]

The assumption of unconditionally reloading skb pointers on
BPF helper calls where bpf_helper_changes_pkt_data() holds
true is wrong. There can be different contexts where the
BPF helper would enforce a reload such as in case of XDP.
Here, we do have a struct xdp_buff instead of struct sk_buff
as context, thus this will access garbage.

JITs only ever need to deal with cached skb pointer reload
when ld_abs/ind was seen, therefore guard the reload behind
SEEN_SKB only. Tested on s390x.

Fixes: 9db7f2b81880 ("s390/bpf: recache skb->data/hlen for skb_vlan_push/pop")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/s390/net/bpf_jit_comp.c

index b15cd2f..33e2785 100644 (file)
@@ -55,8 +55,7 @@ struct bpf_jit {
 #define SEEN_LITERAL   8       /* code uses literals */
 #define SEEN_FUNC      16      /* calls C functions */
 #define SEEN_TAIL_CALL 32      /* code uses tail calls */
-#define SEEN_SKB_CHANGE        64      /* code changes skb data */
-#define SEEN_REG_AX    128     /* code uses constant blinding */
+#define SEEN_REG_AX    64      /* code uses constant blinding */
 #define SEEN_STACK     (SEEN_FUNC | SEEN_MEM | SEEN_SKB)
 
 /*
@@ -448,12 +447,12 @@ static void bpf_jit_prologue(struct bpf_jit *jit)
                        EMIT6_DISP_LH(0xe3000000, 0x0024, REG_W1, REG_0,
                                      REG_15, 152);
        }
-       if (jit->seen & SEEN_SKB)
+       if (jit->seen & SEEN_SKB) {
                emit_load_skb_data_hlen(jit);
-       if (jit->seen & SEEN_SKB_CHANGE)
                /* stg %b1,ST_OFF_SKBP(%r0,%r15) */
                EMIT6_DISP_LH(0xe3000000, 0x0024, BPF_REG_1, REG_0, REG_15,
                              STK_OFF_SKBP);
+       }
 }
 
 /*
@@ -983,8 +982,8 @@ static noinline int bpf_jit_insn(struct bpf_jit *jit, struct bpf_prog *fp, int i
                EMIT2(0x0d00, REG_14, REG_W1);
                /* lgr %b0,%r2: load return value into %b0 */
                EMIT4(0xb9040000, BPF_REG_0, REG_2);
-               if (bpf_helper_changes_pkt_data((void *)func)) {
-                       jit->seen |= SEEN_SKB_CHANGE;
+               if ((jit->seen & SEEN_SKB) &&
+                   bpf_helper_changes_pkt_data((void *)func)) {
                        /* lg %b1,ST_OFF_SKBP(%r15) */
                        EMIT6_DISP_LH(0xe3000000, 0x0004, BPF_REG_1, REG_0,
                                      REG_15, STK_OFF_SKBP);