netdev - Re: [PATCH net-next] bpf: expose sk_priority through struct bpf_sock

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <E92D5306-8776-4377-96F8-35644EBC57FD@fb.com>
Date:   Mon, 13 Nov 2017 20:09:51 +0000
From:   Lawrence Brakmo <brakmo@...com>
To:     Vlad Dumitrescu <vlad@...itrescu.ro>,
        Alexei Starovoitov <ast@...com>,
        Daniel Borkmann <daniel@...earbox.net>,
        "davem@...emloft.net" <davem@...emloft.net>
CC:     "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        Craig Gallek <kraigatgoog@...il.com>
Subject: Re: [PATCH net-next] bpf: expose sk_priority through struct
 bpf_sock_ops

On 11/13/17, 11:01 AM, "Vlad Dumitrescu" <vlad@...itrescu.ro> wrote:

    On Sat, Nov 11, 2017 at 2:38 PM, Alexei Starovoitov <ast@...com> wrote:
    >
    > On 11/12/17 4:46 AM, Daniel Borkmann wrote:
    >>
    >> On 11/11/2017 05:06 AM, Alexei Starovoitov wrote:
    >>>
    >>> On 11/11/17 6:07 AM, Daniel Borkmann wrote:
    >>>>
    >>>> On 11/10/2017 08:17 PM, Vlad Dumitrescu wrote:
    >>>>>
    >>>>> From: Vlad Dumitrescu <vladum@...gle.com>
    >>>>>
    >>>>> Allows BPF_PROG_TYPE_SOCK_OPS programs to read sk_priority.
    >>>>>
    >>>>> Signed-off-by: Vlad Dumitrescu <vladum@...gle.com>
    >>>>> ---
    >>>>>   include/uapi/linux/bpf.h       |  1 +
    >>>>>   net/core/filter.c              | 11 +++++++++++
    >>>>>   tools/include/uapi/linux/bpf.h |  1 +
    >>>>>   3 files changed, 13 insertions(+)
    >>>>>
    >>>>> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
    >>>>> index e880ae6434ee..9757a2002513 100644
    >>>>> --- a/include/uapi/linux/bpf.h
    >>>>> +++ b/include/uapi/linux/bpf.h
    >>>>> @@ -947,6 +947,7 @@ struct bpf_sock_ops {
    >>>>>       __u32 local_ip6[4];    /* Stored in network byte order */
    >>>>>       __u32 remote_port;    /* Stored in network byte order */
    >>>>>       __u32 local_port;    /* stored in host byte order */
    >>>>> +    __u32 priority;
    >>>>>   };
    >>>>>     /* List of known BPF sock_ops operators.
    >>>>> diff --git a/net/core/filter.c b/net/core/filter.c
    >>>>> index 61c791f9f628..a6329642d047 100644
    >>>>> --- a/net/core/filter.c
    >>>>> +++ b/net/core/filter.c
    >>>>> @@ -4449,6 +4449,17 @@ static u32 sock_ops_convert_ctx_access(enum
    >>>>> bpf_access_type type,
    >>>>>           *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
    >>>>>                         offsetof(struct sock_common, skc_num));
    >>>>>           break;
    >>>>> +
    >>>>> +    case offsetof(struct bpf_sock_ops, priority):
    >>>>> +        BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
    >>>>> +
    >>>>> +        *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
    >>>>> +                        struct bpf_sock_ops_kern, sk),
    >>>>> +                      si->dst_reg, si->src_reg,
    >>>>> +                      offsetof(struct bpf_sock_ops_kern, sk));
    >>>>> +        *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
    >>>>> +                      offsetof(struct sock, sk_priority));
    >>>>> +        break;
    >>>>
    >>>>
    >>>> Hm, I don't think this would work, I actually think your initial patch
    >>>> was ok.
    >>>> bpf_setsockopt() as well as bpf_getsockopt() check for sk_fullsock(sk)
    >>>> right
    >>>> before accessing options on either socket or TCP level, and bail out
    >>>> with error
    >>>> otherwise; in such cases we'd read something else here and assume it's
    >>>> sk_priority.
    >>>
    >>>
    >>> even if it's not fullsock, it will just read zero, no? what's a problem
    >>> with that?
    >>> In non-fullsock hooks like BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB
    >>> the program author will know that it's meaningless to read sk_priority,
    >>> so returning zero with minimal checks is fine.
    >>> While adding extra runtime if (sk_fullsock(sk)) is unnecessary,
    >>> since the safety is not compromised.
    >>
    >>
    >> Hm, on my kernel, struct sock has the 4 bytes sk_priority at offset 440,
    >> struct request_sock itself is only 232 byte long in total, and the struct
    >> inet_timewait_sock is 208 byte long, so you'd be accessing out of bounds
    >> that way, so it cannot be ignored and assumed zero.
    >
    >
    > I thought we always pass fully allocated sock but technically not fullsock yet. My mistake. We do: tcp_timeout_init((struct sock *)req))
    > so yeah ctx rewrite approach won't work.
    > Let's go back to access via helper.
    >
    
    TIL. Thanks!
    
    Is there anything else needed from me to get the helper approach accepted?

I plan to add access to TCP state variables (cwnd, rtt, etc.) and I have been thinking
about this issue. I think it is possible to access it directly as long as we use a value
like 0xffffffff to represent an invalid value (e.g. not fullsock). The ctx rewrite just
needs to add a conditional to determine what to return. I would probably add a
field into the internal kernel struct to indicate if it is fullsock or not (we should
know when we call tcp_call_bpf whether it is a fullsock or not based on context).

Let me do a sample patch that I can send for review and get feedback from
Alexi and Daniel.