From: Daniel Borkmann <dan...@iogearbox.net>
Date: Sun, 14 May 2017 16:31:10 +0200
> On 05/13/2017 04:28 AM, David Miller wrote:
>> @@ -823,10 +825,27 @@ static int check_pkt_ptr_alignment(const struct
>> bpf_reg_state *reg,
>> }
>>
>> static int check_val_ptr_alignment(const struct bpf_reg_state *reg,
>> - int size, bool strict)
>> + int off, int size, bool strict)
>> {
>> - if (strict && size != 1) {
>> - verbose("Unknown alignment. Only byte-sized access allowed in
>> value
>> - access.\n");
>> + int reg_off;
>> +
>> + /* Byte size accesses are always allowed. */
>> + if (!strict || size == 1)
>> + return 0;
>> +
>> + reg_off = reg->off;
>> + if (reg->id) {
>> + if (reg->aux_off_align % size) {
>> + verbose("Value access is only %u byte aligned, %d byte access not
>> allowed\n",
>> + reg->aux_off_align, size);
>> + return -EACCES;
>> + }
>> + reg_off += reg->aux_off;
>> + }
>
> What are the semantics of using id here? In ptr_to_pkt, we have it,
> so that eventually, in find_good_pkt_pointers() we can match on id
> and update the range for all such regs with the same id. I'm just
> wondering as the side effect of this is that this makes state
> pruning worse.
Ok. I was advancing reg->id so that it can be used here as the signal
that there is "auxiliary" components to the pointer, and thus we need
to take reg->aux_off_align and reg->aux_off into account.
I did not realize the state pruning component of reg->id so I'll need
to look more deeply into this.
We could use something other than reg->id to decide if there are
variable components to the pointer if necessary.
> Also, reg->off is currently only used in ptr_to_pkt types and
> checked as well in check_packet_access(). Now as semantics change,
> do we need to check for it as well in check_map_access_adj() which
> we currently don't do?
It should not be necessary. Both before and after my changes we
validate the access range using the reg->min_value and reg->max_value.
>> - /* a constant was added to pkt_ptr.
>> + /* a constant was added to the pointer.
>> * Remember it while keeping the same 'id'
>> */
>> dst_reg->off += imm;
>
> Can this now overflow for map type? Also in the UNKNOWN_VALUE case
> below since overflow checks are then only enforced in ptr_to_pkt case?
Indeed, we will have to do "something". The reg->off used to be u16
and is now a u32 with my changes. So it can handle something larger
than MAX_PACKET_OFF.
But we still have to handle overflow. I am not so sure what range of
offsets is reasonable for these MAP pointers, can you make a
suggestion?
>
>> } else {
>> - bool had_id;
>> -
>> - if (src_reg->type == PTR_TO_PACKET) {
>> + if (is_packet && src_reg->type == PTR_TO_PACKET) {
>> /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
>> tmp_reg = *dst_reg; /* save r7 state */
>> *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
>
> I believe clang could probably generate something similar also for
> map value pointers.
Ok, it should be easy to make that part work both with packet pointers
and MAPs.
Thanks for your feedback, I'll try to refine this some more.
