bpf: handle 32-bit zext during constant blinding

Since BPF constant blinding is performed after the verifier pass, the
ALU32 instructions inserted for doubleword immediate loads don't have a
corresponding zext instruction. This is causing a kernel oops on powerpc
and can be reproduced by running 'test_cgroup_storage' with
bpf_jit_harden=2.

Fix this by emitting BPF_ZEXT during constant blinding if
prog->aux->verifier_zext is set.

Fixes: a4b1d3c1ddf6cb ("bpf: verifier: insert zero extension according to analysis result")
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
---
Changes since RFC:
- Removed changes to ALU32 and JMP32 ops since those don't alter program 
  execution, and the verifier would have already accounted for them.  


 kernel/bpf/core.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

Naveen N. Rao writes:

> Since BPF constant blinding is performed after the verifier pass, the
> ALU32 instructions inserted for doubleword immediate loads don't have a
> corresponding zext instruction. This is causing a kernel oops on powerpc
> and can be reproduced by running 'test_cgroup_storage' with
> bpf_jit_harden=2.
>
> Fix this by emitting BPF_ZEXT during constant blinding if
> prog->aux->verifier_zext is set.
>
> Fixes: a4b1d3c1ddf6cb ("bpf: verifier: insert zero extension according to analysis result")
> Reported-by: Michael Ellerman <mpe@ellerman.id.au>
> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>

Thanks for the fix.

Reviewed-by: Jiong Wang <jiong.wang@netronome.com>

Just two other comments during review in case I am wrong on somewhere.

  - Use verifier_zext instead of bpf_jit_needs_zext() seems better, even
    though the latter could avoid extending function argument.

    Because JIT back-ends look at verifier_zext, true means zext inserted
    by verifier so JITs won't do the code-gen.

    Use verifier_zext is sort of keeping JIT blinding the same behaviour
    has verifier even though blinding doesn't belong to verifier, but for
    such insn patching, it could be seen as a extension of verifier,
    therefore use verifier_zext seems better than bpf_jit_needs_zext() to
    me.
   
  - JIT blinding is also escaping the HI32 randomization which happens
    inside verifier, otherwise x86-64 regression should have caught this issue.

Regards,
Jiong

> ---
> Changes since RFC:
> - Removed changes to ALU32 and JMP32 ops since those don't alter program 
>   execution, and the verifier would have already accounted for them.  
>
>
>  kernel/bpf/core.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> index 8191a7db2777..66088a9e9b9e 100644
> --- a/kernel/bpf/core.c
> +++ b/kernel/bpf/core.c
> @@ -890,7 +890,8 @@ int bpf_jit_get_func_addr(const struct bpf_prog *prog,
>  
>  static int bpf_jit_blind_insn(const struct bpf_insn *from,
>  			      const struct bpf_insn *aux,
> -			      struct bpf_insn *to_buff)
> +			      struct bpf_insn *to_buff,
> +			      bool emit_zext)
>  {
>  	struct bpf_insn *to = to_buff;
>  	u32 imm_rnd = get_random_int();
> @@ -1005,6 +1006,8 @@ static int bpf_jit_blind_insn(const struct bpf_insn *from,
>  	case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
>  		*to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
>  		*to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
> +		if (emit_zext)
> +			*to++ = BPF_ZEXT_REG(BPF_REG_AX);
>  		*to++ = BPF_ALU64_REG(BPF_OR,  aux[0].dst_reg, BPF_REG_AX);
>  		break;
>  
> @@ -1088,7 +1091,8 @@ struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
>  		    insn[1].code == 0)
>  			memcpy(aux, insn, sizeof(aux));
>  
> -		rewritten = bpf_jit_blind_insn(insn, aux, insn_buff);
> +		rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
> +						clone->aux->verifier_zext);
>  		if (!rewritten)
>  			continue;

Jiong Wang wrote:
> 
> Naveen N. Rao writes:
> 
>> Since BPF constant blinding is performed after the verifier pass, the
>> ALU32 instructions inserted for doubleword immediate loads don't have a
>> corresponding zext instruction. This is causing a kernel oops on powerpc
>> and can be reproduced by running 'test_cgroup_storage' with
>> bpf_jit_harden=2.
>>
>> Fix this by emitting BPF_ZEXT during constant blinding if
>> prog->aux->verifier_zext is set.
>>
>> Fixes: a4b1d3c1ddf6cb ("bpf: verifier: insert zero extension according to analysis result")
>> Reported-by: Michael Ellerman <mpe@ellerman.id.au>
>> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
> 
> Thanks for the fix.
> 
> Reviewed-by: Jiong Wang <jiong.wang@netronome.com>
> 
> Just two other comments during review in case I am wrong on somewhere.
> 
>   - Use verifier_zext instead of bpf_jit_needs_zext() seems better, even
>     though the latter could avoid extending function argument.
> 
>     Because JIT back-ends look at verifier_zext, true means zext inserted
>     by verifier so JITs won't do the code-gen.
> 
>     Use verifier_zext is sort of keeping JIT blinding the same behaviour
>     has verifier even though blinding doesn't belong to verifier, but for
>     such insn patching, it could be seen as a extension of verifier,
>     therefore use verifier_zext seems better than bpf_jit_needs_zext() to
>     me.
>    
>   - JIT blinding is also escaping the HI32 randomization which happens
>     inside verifier, otherwise x86-64 regression should have caught this issue.

Jiong,
Thanks for the review.

Alexei, Daniel,
Can you please pick this up for v5.3. This is a regression and is 
causing a crash on powerpc.


- Naveen

On 8/21/19 9:23 PM, Naveen N. Rao wrote:
> Since BPF constant blinding is performed after the verifier pass, the
> ALU32 instructions inserted for doubleword immediate loads don't have a
> corresponding zext instruction. This is causing a kernel oops on powerpc
> and can be reproduced by running 'test_cgroup_storage' with
> bpf_jit_harden=2.
> 
> Fix this by emitting BPF_ZEXT during constant blinding if
> prog->aux->verifier_zext is set.
> 
> Fixes: a4b1d3c1ddf6cb ("bpf: verifier: insert zero extension according to analysis result")
> Reported-by: Michael Ellerman <mpe@ellerman.id.au>
> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>

LGTM, applied to bpf, thanks!