Re: [PATCH bpf v2] bpf: Update bpf_{g,s}etsockopt() documentation

[Ji Rongfeng <SikoJobs@xxxxxxxxxxx>]

On 2022/11/27 11:27, Ji Rongfeng wrote:
> On 2022/11/24 8:40, Martin KaFai Lau wrote:
> > On 11/23/22 4:18 PM, Daniel Borkmann wrote:
> > > On 11/18/22 9:18 AM, Ji Rongfeng wrote:
> > > > * append missing optnames to the end
> > > > * simplify bpf_getsockopt()'s doc
> > > >
> > > > Signed-off-by: Ji Rongfeng <SikoJobs@xxxxxxxxxxx>
> > > > ---
> > > >   include/uapi/linux/bpf.h       | 20 ++++++++++++--------
> > > >   tools/include/uapi/linux/bpf.h | 20 ++++++++++++--------
> > > >   2 files changed, 24 insertions(+), 16 deletions(-)
> > > >
> > > > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > > > index 51b9aa640ad2..14f29d95ea71 100644
> > > > --- a/include/uapi/linux/bpf.h
> > > > +++ b/include/uapi/linux/bpf.h
> > > > @@ -2576,14 +2576,19 @@ union bpf_attr {
> > > >    *         * **SOL_SOCKET**, which supports the following 
> > > > *optname*\ s:
> > > >    *           **SO_RCVBUF**, **SO_SNDBUF**, **SO_MAX_PACING_RATE**,
> > > >    *           **SO_PRIORITY**, **SO_RCVLOWAT**, **SO_MARK**,
> > > > - *           **SO_BINDTODEVICE**, **SO_KEEPALIVE**.
> > > > + *           **SO_BINDTODEVICE**, **SO_KEEPALIVE**, **SO_REUSEADDR**,
> > > > + *           **SO_REUSEPORT**, **SO_BINDTOIFINDEX**, **SO_TXREHASH**.
> > > >    *         * **IPPROTO_TCP**, which supports the following 
> > > > *optname*\ s:
> > > >    *           **TCP_CONGESTION**, **TCP_BPF_IW**,
> > > >    *           **TCP_BPF_SNDCWND_CLAMP**, **TCP_SAVE_SYN**,
> > > >    *           **TCP_KEEPIDLE**, **TCP_KEEPINTVL**, **TCP_KEEPCNT**,
> > > > - *          **TCP_SYNCNT**, **TCP_USER_TIMEOUT**, 
> > > > **TCP_NOTSENT_LOWAT**.
> > > > + *           **TCP_SYNCNT**, **TCP_USER_TIMEOUT**, 
> > > > **TCP_NOTSENT_LOWAT**,
> > > > + *           **TCP_NODELAY**, **TCP_MAXSEG**, **TCP_WINDOW_CLAMP**,
> > > > + *           **TCP_THIN_LINEAR_TIMEOUTS**, **TCP_BPF_DELACK_MAX**,
> > > > + *           **TCP_BPF_RTO_MIN**.
> > > >    *         * **IPPROTO_IP**, which supports *optname* **IP_TOS**.
> > > > - *         * **IPPROTO_IPV6**, which supports *optname* 
> > > > **IPV6_TCLASS**.
> > > > + *         * **IPPROTO_IPV6**, which supports the following 
> > > > *optname*\ s:
> > > > + *           **IPV6_TCLASS**, **IPV6_AUTOFLOWLABEL**.
> > > >    *     Return
> > > >    *         0 on success, or a negative error in case of failure.
> > > >    *
> > > > @@ -2800,12 +2805,11 @@ union bpf_attr {
> > > >    *           and **BPF_CGROUP_INET6_CONNECT**.
> > > >    *
> > > >    *         This helper actually implements a subset of 
> > > > **getsockopt()**.
> > > > - *         It supports the following *level*\ s:
> > > > + *         It supports the same set of *optname*\ s that supported by
> > >
> > > nit: that is supported by
> > >
> > > > + *         **bpf_setsockopt**\ () helper with a few exceptions:
> > > >    *
> > > > - *         * **IPPROTO_TCP**, which supports *optname*
> > > > - *           **TCP_CONGESTION**.
> > > > - *         * **IPPROTO_IP**, which supports *optname* **IP_TOS**.
> > > > - *         * **IPPROTO_IPV6**, which supports *optname* 
> > > > **IPV6_TCLASS**.
> > > > + *         * **bpf_setsockopt**\ () helper only: **TCP_BPF_***.
> > > > + *         * **bpf_getsockopt**\ () helper only: **TCP_SAVED_SYNC**.
> > >
> > > I think from a user PoV the above is a bit hard to follow, maybe take 
> > > Martin's
> > > earlier feedback into account and add a proper sentence; it will be 
> > > much easier
> > > to understand.
> >
> > +1  Made the change and also fixed TCP_SAVED_SYNC with s/SYNC/SYN/ 
> > while applying.  Thanks!
>
> Thanks for the helpful reviews. I chose the form of lists was because we 
> could
> append more optnames easily in the future. But I believe it's not late 
> to apply
> that form when we really need it : )
>
> In my opinion, this patch doesn't contain any new feature, but just a 

new features

> kind of fix
> to the documentation, according to the corresponding code in the bpf tree,
> which hasn't been modified yet in the bpf-next tree. So I targeted the 
> former,
> as this patch could be useful there. Please let me know if there's any 

if there're

> customary
> rules outside bpf_devel_QA. Thanks!

I just found that SO_BINDTODEVICE is bpf_setsockopt() only. I checked 
sock_getbindtodevice() and there's nothing special comparing with 
sock_setbindtodevice(), except "down_read(&devnet_rename_sem);" and 
"up_read(&devnet_rename_sem);". Martin once wrote:

> The only exception is SO_BINDTODEVICE because it needs to acquire a
> blocking lock.  Thus, SO_BINDTODEVICE is not supported.

Were you referring to "down_read(&devnet_rename_sem);"? Seems it's not 
acquiring a blocking lock. Maybe "devnet_rename_sem" has been locked for 
writing somewhere in bpf before? Please let me know. Thanks!