Re: [PATCH bpf-next] bpf: add tcp_notsent_lowat bpf setsockopt

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



чт, 8 окт. 2020 г. в 23:27, Yonghong Song <yhs@xxxxxx>:
>
>
>
> On 10/8/20 10:08 PM, Nikita V. Shirokov wrote:
> > Adding support for TCP_NOTSENT_LOWAT sockoption
> > (https://lwn.net/Articles/560082/ ) in tcpbpf
> >
> > Signed-off-by: Nikita V. Shirokov <tehnerd@xxxxxxxxxxx>
> > ---
> >   include/uapi/linux/bpf.h                          |  2 +-
> >   net/core/filter.c                                 |  4 ++++
> >   tools/testing/selftests/bpf/progs/connect4_prog.c | 15 +++++++++++++++
> >   3 files changed, 20 insertions(+), 1 deletion(-)
> >
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index d83561e8cd2c..42d2df799397 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -1698,7 +1698,7 @@ union bpf_attr {
> >    *            **TCP_CONGESTION**, **TCP_BPF_IW**,
> >    *            **TCP_BPF_SNDCWND_CLAMP**, **TCP_SAVE_SYN**,
> >    *            **TCP_KEEPIDLE**, **TCP_KEEPINTVL**, **TCP_KEEPCNT**,
> > - *             **TCP_SYNCNT**, **TCP_USER_TIMEOUT**.
> > + *             **TCP_SYNCNT**, **TCP_USER_TIMEOUT**, **TCP_NOTSENT_LOWAT**.
> >    *          * **IPPROTO_IP**, which supports *optname* **IP_TOS**.
> >    *          * **IPPROTO_IPV6**, which supports *optname* **IPV6_TCLASS**.
> >    *  Return
> > diff --git a/net/core/filter.c b/net/core/filter.c
> > index 05df73780dd3..5da44b11e1ec 100644
> > --- a/net/core/filter.c
> > +++ b/net/core/filter.c
> > @@ -4827,6 +4827,10 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname,
> >                               else
> >                                       icsk->icsk_user_timeout = val;
> >                               break;
> > +                     case TCP_NOTSENT_LOWAT:
> > +                             tp->notsent_lowat = val;
> > +                             sk->sk_write_space(sk);
> > +                             break;
>
> This looks good to me. It is the same as in do_tcp_setsockopt().
>
> >                       default:
> >                               ret = -EINVAL;
> >                       }
> > diff --git a/tools/testing/selftests/bpf/progs/connect4_prog.c b/tools/testing/selftests/bpf/progs/connect4_prog.c
> > index b1b2773c0b9d..b10e7fbace7b 100644
> > --- a/tools/testing/selftests/bpf/progs/connect4_prog.c
> > +++ b/tools/testing/selftests/bpf/progs/connect4_prog.c
> > @@ -128,6 +128,18 @@ static __inline int set_keepalive(struct bpf_sock_addr *ctx)
> >       return 0;
> >   }
> >
> > +static __inline int set_notsent_lowat(struct bpf_sock_addr *ctx)
> > +{
> > +     int lowat = 65535;
> > +
> > +     if (ctx->type == SOCK_STREAM) {
> > +             if (bpf_setsockopt(ctx, SOL_TCP, TCP_NOTSENT_LOWAT, &lowat, sizeof(lowat)))
>
> In my build system, I hit a compilation error.
>
> progs/connect4_prog.c:137:36: error: use of undeclared identifier
> 'TCP_NOTSENT_LOWAT'
>                  if (bpf_setsockopt(ctx, SOL_TCP, TCP_NOTSENT_LOWAT,
> &lowat, sizeof(lowat)))
>
> TCP_NOTSENT_LOWAT is included in /usr/include/linux/tcp.h. But this file
> includes netinet/tcp.h and it contains some same symbol definitions as
> linux/tcp.h so I can include both.
>
> Adding the following can fix the issue
>
> #ifndef TCP_NOTSENT_LOWAT
> #define TCP_NOTSENT_LOWAT       25
> #endif
>
> Not sure where TCP_NOTSENT_LOWAT is defined in your system.

Hey, thanks for checking. will send v2 in a few. as for my system it
is defined here:

/usr/include/netinet/tcp.h
64:#define TCP_NOTSENT_LOWAT     25 /* Limit number of unsent bytes in


>
> > +                     return 1;
> > +     }
> > +
> > +     return 0;
> > +}
> > +
> >   SEC("cgroup/connect4")
> >   int connect_v4_prog(struct bpf_sock_addr *ctx)
> >   {
> > @@ -148,6 +160,9 @@ int connect_v4_prog(struct bpf_sock_addr *ctx)
> >       if (set_keepalive(ctx))
> >               return 0;
> >
> > +     if (set_notsent_lowat(ctx))
> > +             return 0;
> > +
> >       if (ctx->type != SOCK_STREAM && ctx->type != SOCK_DGRAM)
> >               return 0;
> >       else if (ctx->type == SOCK_STREAM)
> >

--
Nikita




[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux