From: Yonghong Song <yhs@fb.com>
To: Jiri Olsa <jolsa@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>, Alexei Starovoitov <ast@kernel.org>,
"Daniel Borkmann" <daniel@iogearbox.net>,
"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
"bpf@vger.kernel.org" <bpf@vger.kernel.org>,
"Andrii Nakryiko" <andriin@fb.com>, Martin Lau <kafai@fb.com>,
Jakub Kicinski <jakub.kicinski@netronome.com>,
David Miller <davem@redhat.com>
Subject: Re: Re: [PATCH 1/5] bpf: Allow non struct type for btf ctx access
Date: Tue, 7 Jan 2020 18:28:11 +0000 [thread overview]
Message-ID: <962209d7-cdfa-9ee3-8a12-6375091843cf@fb.com> (raw)
In-Reply-To: <c8ed83dc-3f3b-30d2-69fa-3a5c59152034@fb.com>
On 1/7/20 9:55 AM, Yonghong Song wrote:
>
>
> On 1/7/20 7:50 AM, Jiri Olsa wrote:
>> On Tue, Jan 07, 2020 at 01:13:23PM +0100, Jiri Olsa wrote:
>>> On Mon, Jan 06, 2020 at 09:36:17PM +0000, Yonghong Song wrote:
>>>>
>>>>
>>>> On 12/29/19 6:37 AM, Jiri Olsa wrote:
>>>>> I'm not sure why the restriction was added,
>>>>> but I can't access pointers to POD types like
>>>>> const char * when probing vfs_read function.
>>>>>
>>>>> Removing the check and allow non struct type
>>>>> access in context.
>>>>>
>>>>> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
>>>>> ---
>>>>> kernel/bpf/btf.c | 6 ------
>>>>> 1 file changed, 6 deletions(-)
>>>>>
>>>>> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
>>>>> index ed2075884724..ae90f60ac1b8 100644
>>>>> --- a/kernel/bpf/btf.c
>>>>> +++ b/kernel/bpf/btf.c
>>>>> @@ -3712,12 +3712,6 @@ bool btf_ctx_access(int off, int size, enum bpf_access_type type,
>>>>> /* skip modifiers */
>>>>> while (btf_type_is_modifier(t))
>>>>> t = btf_type_by_id(btf, t->type);
>>>>> - if (!btf_type_is_struct(t)) {
>>>>> - bpf_log(log,
>>>>> - "func '%s' arg%d type %s is not a struct\n",
>>>>> - tname, arg, btf_kind_str[BTF_INFO_KIND(t->info)]);
>>>>> - return false;
>>>>> - }
>>>>
>>>> Hi, Jiri, the RFC looks great! Especially, you also referenced this will
>>>> give great performance boost for bcc scripts.
>>>>
>>>> Could you provide more context on why the above change is needed?
>>>> The function btf_ctx_access is used to check validity of accessing
>>>> function parameters which are wrapped inside a structure, I am wondering
>>>> what kinds of accesses you tried to address here.
>>>
>>> when I was transforming opensnoop.py to use this I got fail in
>>> there when I tried to access filename arg in do_sys_open
>>>
>>> but actualy it seems this should get recognized earlier by:
>>>
>>> if (btf_type_is_int(t))
>>> /* accessing a scalar */
>>> return true;
>>>
>>> I'm not sure why it did not pass for const char*, I'll check
>>
>> it seems we don't check for pointer to scalar (just void),
>> which is the case in my example 'const char *filename'
>
> Thanks for clarification. See some comments below.
>
>>
>> I'll post this in v2 with other changes
>>
>> jirka
>>
>>
>> ---
>> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
>> index ed2075884724..650df4ed346e 100644
>> --- a/kernel/bpf/btf.c
>> +++ b/kernel/bpf/btf.c
>> @@ -3633,7 +3633,7 @@ bool btf_ctx_access(int off, int size, enum bpf_access_type type,
>> const struct bpf_prog *prog,
>> struct bpf_insn_access_aux *info)
>> {
>> - const struct btf_type *t = prog->aux->attach_func_proto;
>> + const struct btf_type *tp, *t = prog->aux->attach_func_proto;
>> struct bpf_prog *tgt_prog = prog->aux->linked_prog;
>> struct btf *btf = bpf_prog_get_target_btf(prog);
>> const char *tname = prog->aux->attach_func_name;
>> @@ -3695,6 +3695,17 @@ bool btf_ctx_access(int off, int size, enum bpf_access_type type,
>> */
>> return true;
>>
>> + tp = btf_type_by_id(btf, t->type);
>> + /* skip modifiers */
>> + while (btf_type_is_modifier(tp))
>> + tp = btf_type_by_id(btf, tp->type);
>> +
>> + if (btf_type_is_int(tp))
>> + /* This is a pointer scalar.
>> + * It is the same as scalar from the verifier safety pov.
>> + */
>> + return true;
>
> This should work since:
> - the int pointer will be treated as a scalar later on
> - bpf_probe_read() will be used to read the contents
>
> I am wondering whether we should add proper verifier support
> to allow pointer to int ctx access. There, users do not need
> to use bpf_probe_read() to dereference the pointer.
>
> Discussed with Martin, maybe somewhere in check_ptr_to_btf_access(),
> before btf_struct_access(), checking if it is a pointer to int/enum,
> it should just allow and return SCALAR_VALUE.
double checked check_ptr_to_btf_access() and btf_struct_access().
btf_struct_access() already returns SCALAR_VALUE for pointer to
int/enum. So verifier change is probably not needed.
In your above code, could you do
btf_type_is_int(t) || btf_type_is_enum(t)
which will cover pointer to enum as well?
>
> If you do verifier changes, please ensure bpf_probe_read() is not
> needed any more. In bcc, you need to hack to prevent rewriter to
> re-introduce bpf_probe_read() :-).
>
>> +
>> /* this is a pointer to another type */
>> info->reg_type = PTR_TO_BTF_ID;
>> info->btf_id = t->type;
>>
next prev parent reply other threads:[~2020-01-07 18:28 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-12-29 14:37 [RFC 0/5] bpf: Add trampoline helpers Jiri Olsa
2019-12-29 14:37 ` [PATCH 1/5] bpf: Allow non struct type for btf ctx access Jiri Olsa
2020-01-06 21:36 ` Yonghong Song
2020-01-07 12:13 ` Jiri Olsa
2020-01-07 15:50 ` Jiri Olsa
2020-01-07 17:55 ` Yonghong Song
2020-01-07 18:28 ` Yonghong Song [this message]
2020-01-08 14:38 ` Jiri Olsa
2019-12-29 14:37 ` [PATCH 2/5] bpf: Add bpf_perf_event_output_kfunc Jiri Olsa
2020-01-06 23:27 ` Alexei Starovoitov
2020-01-07 12:25 ` Jiri Olsa
2020-01-07 22:13 ` Alexei Starovoitov
2020-01-08 10:24 ` Jiri Olsa
2019-12-29 14:37 ` [PATCH 3/5] bpf: Add bpf_get_stackid_kfunc Jiri Olsa
2019-12-29 14:37 ` [PATCH 4/5] bpf: Add bpf_get_stack_kfunc Jiri Olsa
2019-12-29 14:37 ` [PATCH 5/5] bpf: Allow to resolve bpf trampoline in unwind Jiri Olsa
2020-01-06 23:46 ` Alexei Starovoitov
2020-01-07 8:30 ` Daniel Borkmann
2020-01-07 13:15 ` Jiri Olsa
2020-01-07 19:30 ` Daniel Borkmann
2020-01-07 13:05 ` Jiri Olsa
2020-01-07 13:30 ` Björn Töpel
2020-01-13 9:43 ` Jiri Olsa
2020-01-13 12:21 ` Björn Töpel
2020-01-13 12:31 ` Björn Töpel
2020-01-13 12:37 ` Jiri Olsa
2020-02-03 19:58 ` Jiri Olsa
2020-02-03 20:27 ` Björn Töpel
2020-02-03 20:45 ` Toke Høiland-Jørgensen
2020-02-04 8:18 ` Jiri Olsa
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=962209d7-cdfa-9ee3-8a12-6375091843cf@fb.com \
--to=yhs@fb.com \
--cc=andriin@fb.com \
--cc=ast@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=daniel@iogearbox.net \
--cc=davem@redhat.com \
--cc=jakub.kicinski@netronome.com \
--cc=jolsa@kernel.org \
--cc=jolsa@redhat.com \
--cc=kafai@fb.com \
--cc=netdev@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).