On 6/13/19 3:25 PM, Andrii Nakryiko wrote: > On Thu, Jun 13, 2019 at 9:50 AM Alexei Starovoitov <ast@xxxxxxxxxx> wrote: >> >> This patch extends is_branch_taken() logic from JMP+K instructions >> to JMP+X instructions. >> Conditional branches are often done when src and dst registers >> contain known scalars. In such case the verifier can follow >> the branch that is going to be taken when program executes on CPU. >> That speeds up the verification and essential feature to support > > typo: and *is* essential > >> bounded loops. >> >> Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx> >> --- >> kernel/bpf/verifier.c | 16 +++++++++++++--- >> 1 file changed, 13 insertions(+), 3 deletions(-) >> >> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c >> index a21bafd7d931..c79c09586a9e 100644 >> --- a/kernel/bpf/verifier.c >> +++ b/kernel/bpf/verifier.c >> @@ -5263,10 +5263,11 @@ static int check_cond_jmp_op(struct bpf_verifier_env *env, >> struct bpf_verifier_state *this_branch = env->cur_state; >> struct bpf_verifier_state *other_branch; >> struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; >> - struct bpf_reg_state *dst_reg, *other_branch_regs; >> + struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL; >> u8 opcode = BPF_OP(insn->code); >> bool is_jmp32; >> int err; >> + u64 cond_val; > > reverse Christmas tree > >> >> /* Only conditional jumps are expected to reach here. */ >> if (opcode == BPF_JA || opcode > BPF_JSLE) { >> @@ -5290,6 +5291,7 @@ static int check_cond_jmp_op(struct bpf_verifier_env *env, >> insn->src_reg); >> return -EACCES; >> } >> + src_reg = ®s[insn->src_reg]; >> } else { >> if (insn->src_reg != BPF_REG_0) { >> verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); >> @@ -5306,8 +5308,11 @@ static int check_cond_jmp_op(struct bpf_verifier_env *env, >> is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32; >> >> if (BPF_SRC(insn->code) == BPF_K) { >> - int pred = is_branch_taken(dst_reg, insn->imm, opcode, >> - is_jmp32); >> + int pred; >> + >> + cond_val = insn->imm; >> +check_taken: >> + pred = is_branch_taken(dst_reg, cond_val, opcode, is_jmp32); >> >> if (pred == 1) { >> /* only follow the goto, ignore fall-through */ >> @@ -5319,6 +5324,11 @@ static int check_cond_jmp_op(struct bpf_verifier_env *env, >> */ >> return 0; >> } >> + } else if (BPF_SRC(insn->code) == BPF_X && >> + src_reg->type == SCALAR_VALUE && >> + tnum_is_const(src_reg->var_off)) { >> + cond_val = src_reg->var_off.value; >> + goto check_taken; >> } > > To eliminate goto, how about this; > > int pred = -1; > > if (BPF_SRC(insn->code) == BPF_K) > pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32); > else if (BPF_SRC(insn->code) == BPF_X && > src_reg->type == SCALAR_VALUE && > tnum_is_const(src_reg->var_off) > pred = is_branch_taken(dst_reg, src_reg->var_off.value, > opcode, is_jmp32); > > /* here do pred == 1 and pred == 0 special handling, otherwise fall-through */ > > Again, more linear and no unnecessary gotos. pred == -1 has already a > meaning of "don't know, have to try both". sure. should work.