Re: [PATCH v3 bpf-next 06/11] bpf: implement accurate raw_tp context access via BTF

[Daniel Borkmann <daniel@iogearbox.net>]

On 10/16/19 11:28 PM, Alexei Starovoitov wrote:
> On Wed, Oct 16, 2019 at 2:22 PM Daniel Borkmann <daniel@iogearbox.net> wrote:
>> On 10/16/19 5:25 AM, Alexei Starovoitov wrote:
>>> libbpf analyzes bpf C program, searches in-kernel BTF for given type name
>>> and stores it into expected_attach_type.
>>> The kernel verifier expects this btf_id to point to something like:
>>> typedef void (*btf_trace_kfree_skb)(void *, struct sk_buff *skb, void *loc);
>>> which represents signature of raw_tracepoint "kfree_skb".
>>>
>>> Then btf_ctx_access() matches ctx+0 access in bpf program with 'skb'
>>> and 'ctx+8' access with 'loc' arguments of "kfree_skb" tracepoint.
>>> In first case it passes btf_id of 'struct sk_buff *' back to the verifier core
>>> and 'void *' in second case.
>>>
>>> Then the verifier tracks PTR_TO_BTF_ID as any other pointer type.
>>> Like PTR_TO_SOCKET points to 'struct bpf_sock',
>>> PTR_TO_TCP_SOCK points to 'struct bpf_tcp_sock', and so on.
>>> PTR_TO_BTF_ID points to in-kernel structs.
>>> If 1234 is btf_id of 'struct sk_buff' in vmlinux's BTF
>>> then PTR_TO_BTF_ID#1234 points to one of in kernel skbs.
>>>
>>> When PTR_TO_BTF_ID#1234 is dereferenced (like r2 = *(u64 *)r1 + 32)
>>> the btf_struct_access() checks which field of 'struct sk_buff' is
>>> at offset 32. Checks that size of access matches type definition
>>> of the field and continues to track the dereferenced type.
>>> If that field was a pointer to 'struct net_device' the r2's type
>>> will be PTR_TO_BTF_ID#456. Where 456 is btf_id of 'struct net_device'
>>> in vmlinux's BTF.
>>>
>>> Such verifier analysis prevents "cheating" in BPF C program.
>>> The program cannot cast arbitrary pointer to 'struct sk_buff *'
>>> and access it. C compiler would allow type cast, of course,
>>> but the verifier will notice type mismatch based on BPF assembly
>>> and in-kernel BTF.
>>>
>>> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
>>
>> Overall set looks great!
>>
>> [...]
>>> +int btf_struct_access(struct bpf_verifier_log *log,
>>> +                   const struct btf_type *t, int off, int size,
>>> +                   enum bpf_access_type atype,
>>> +                   u32 *next_btf_id)
>>> +{
>>> +     const struct btf_member *member;
>>> +     const struct btf_type *mtype;
>>> +     const char *tname, *mname;
>>> +     int i, moff = 0, msize;
>>> +
>>> +again:
>>> +     tname = __btf_name_by_offset(btf_vmlinux, t->name_off);
>>
>> More of a high-level question wrt btf_ctx_access(), is there a reason the ctx
>> access is only done for raw_tp? I presume kprobes is still on todo (?), what
>> about uprobes which also have pt_regs and could benefit from this work, but is
>> not fixed to btf_vmlinux to search its ctx type.
> 
> Optimized kprobes via ftrace entry point are on immediate todo list
> to follow up. I'm still debating on the best way to handle it.
> uprobes - I haven't though about. Likely necessary as well.
> Not sure what types to give to pt_regs yet.
> 
>> I presume BPF_LDX | BPF_PROBE_MEM | BPF_* would need no additional encoding,
>> but JIT emission would have to differ depending on the prog type.
> 
> you mean for kprobes/uprobes? Why would it need to be different?
> The idea was to keep LDX|PROBE_MEM as normal LDX|MEM load as much as possible.

Agree, makes sense.

> The only difference vs normal load is to populate extable which is
> arch dependent.

Wouldn't you also need to switch to USER_DS similarly to what probe_kernel_read()
vs probe_user_read() differentiates?

Thanks,
Daniel