From: Alexei Starovoitov Subject: [PATCH net] sparc: bpf_jit: fix loads from negative offsets Date: Tue, 23 Sep 2014 13:50:10 -0700 Message-ID: <1411505410-25215-1-git-send-email-ast@plumgrid.com> Cc: netdev@vger.kernel.org To: "David S. Miller" Return-path: Received: from mail-pd0-f181.google.com ([209.85.192.181]:41607 "EHLO mail-pd0-f181.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753947AbaIWUuV (ORCPT ); Tue, 23 Sep 2014 16:50:21 -0400 Received: by mail-pd0-f181.google.com with SMTP id r10so7205910pdi.40 for ; Tue, 23 Sep 2014 13:50:21 -0700 (PDT) Sender: netdev-owner@vger.kernel.org List-ID: - fix BPF_LD|ABS|IND from negative offsets: make sure to sign extend lower 32 bits in 64-bit register before calling C helpers from JITed code, otherwise 'int k' argument of bpf_internal_load_pointer_neg_helper() function will be added as large unsigned integer, causing packet size check to trigger and abort the program. It's worth noting that JITed code for 'A = A op K' will affect upper 32 bits differently depending whether K is simm13 or not. Since small constants are sign extended, whereas large constants are stored in temp register and zero extended. That is ok and we don't have to pay a penalty of sign extension for every sethi, since all classic BPF instructions have 32-bit semantics and we only need to set correct upper bits when transitioning from JITed code into C. - though instructions 'A &= 0' and 'A *= 0' are odd, JIT compiler should not optimize them out Signed-off-by: Alexei Starovoitov --- More detailed explanation of 1st bug: BPF_STMT(BPF_LD | BPF_B | BPF_ABS, SKF_LL_OFF) will be JITed as: sethi %hi(0xffe00000), %g3 or %g3, 0, %g3 call bpf_jit_load_byte_negative_offset which will do: mov %g3, %o1 call bpf_internal_load_pointer_neg_helper mov 1, %o2 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size) { ... ptr = skb_mac_header(skb) + k - SKF_LL_OFF; here 'k' will be added as 0xffe00000 instead of SKF_LL_OFF == -0x200000 After these fixes, test_bpf is now clean on sparc64. Compile tested only on sparc32 arch/sparc/net/bpf_jit_asm.S | 3 +++ arch/sparc/net/bpf_jit_comp.c | 2 +- 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/arch/sparc/net/bpf_jit_asm.S b/arch/sparc/net/bpf_jit_asm.S index 9d016c7..8c83f4b 100644 --- a/arch/sparc/net/bpf_jit_asm.S +++ b/arch/sparc/net/bpf_jit_asm.S @@ -6,10 +6,12 @@ #define SAVE_SZ 176 #define SCRATCH_OFF STACK_BIAS + 128 #define BE_PTR(label) be,pn %xcc, label +#define SIGN_EXTEND(reg) sra reg, 0, reg #else #define SAVE_SZ 96 #define SCRATCH_OFF 72 #define BE_PTR(label) be label +#define SIGN_EXTEND(reg) #endif #define SKF_MAX_NEG_OFF (-0x200000) /* SKF_LL_OFF from filter.h */ @@ -135,6 +137,7 @@ bpf_slow_path_byte_msh: save %sp, -SAVE_SZ, %sp; \ mov %i0, %o0; \ mov r_OFF, %o1; \ + SIGN_EXTEND(%o1); \ call bpf_internal_load_pointer_neg_helper; \ mov (LEN), %o2; \ mov %o0, r_TMP; \ diff --git a/arch/sparc/net/bpf_jit_comp.c b/arch/sparc/net/bpf_jit_comp.c index 51ae87b..ece4af0 100644 --- a/arch/sparc/net/bpf_jit_comp.c +++ b/arch/sparc/net/bpf_jit_comp.c @@ -184,7 +184,7 @@ do { \ */ #define emit_alu_K(OPCODE, K) \ do { \ - if (K) { \ + if (K || OPCODE == AND || OPCODE == MUL) { \ unsigned int _insn = OPCODE; \ _insn |= RS1(r_A) | RD(r_A); \ if (is_simm13(K)) { \ -- 1.7.9.5