From: Martin Lau <kafai@fb.com>
To: Jakub Sitnicki <jakub@cloudflare.com>
Cc: John Fastabend <john.fastabend@gmail.com>,
Alexei Starovoitov <alexei.starovoitov@gmail.com>,
"bpf@vger.kernel.org" <bpf@vger.kernel.org>,
"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
"kernel-team@cloudflare.com" <kernel-team@cloudflare.com>
Subject: Re: [PATCH bpf-next 5/8] bpf: Allow selecting reuseport socket from a SOCKMAP
Date: Tue, 26 Nov 2019 19:03:04 +0000 [thread overview]
Message-ID: <20191126190301.quwvjihpdzfjhdbe@kafai-mbp.dhcp.thefacebook.com> (raw)
In-Reply-To: <87imn6ogke.fsf@cloudflare.com>
On Tue, Nov 26, 2019 at 03:30:57PM +0100, Jakub Sitnicki wrote:
> On Mon, Nov 25, 2019 at 11:07 PM CET, Martin Lau wrote:
> > On Mon, Nov 25, 2019 at 11:40:41AM +0100, Jakub Sitnicki wrote:
> >> On Mon, Nov 25, 2019 at 05:17 AM CET, John Fastabend wrote:
> >> > Alexei Starovoitov wrote:
> >> >> On Sat, Nov 23, 2019 at 12:07:48PM +0100, Jakub Sitnicki wrote:
> >> >> > SOCKMAP now supports storing references to listening sockets. Nothing keeps
> >> >> > us from using it as an array of sockets to select from in SK_REUSEPORT
> >> >> > programs.
> >> >> >
> >> >> > Whitelist the map type with the BPF helper for selecting socket. However,
> >> >> > impose a restriction that the selected socket needs to be a listening TCP
> >> >> > socket or a bound UDP socket (connected or not).
> >> >> >
> >> >> > The only other map type that works with the BPF reuseport helper,
> >> >> > REUSEPORT_SOCKARRAY, has a corresponding check in its update operation
> >> >> > handler.
> >> >> >
> >> >> > Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
> >> >> > ---
> >> >
> >> > [...]
> >> >
> >> >> > diff --git a/net/core/filter.c b/net/core/filter.c
> >> >> > index 49ded4a7588a..e3fb77353248 100644
> >> >> > --- a/net/core/filter.c
> >> >> > +++ b/net/core/filter.c
> >> >> > @@ -8723,6 +8723,8 @@ BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
> >> >> > selected_sk = map->ops->map_lookup_elem(map, key);
> >> >> > if (!selected_sk)
> >> >> > return -ENOENT;
> >> >> > + if (!sock_flag(selected_sk, SOCK_RCU_FREE))
> >> >> > + return -EINVAL;
> > If I read it correctly,
> > this is to avoid the following "if (!reuse)" to return -ENOENT,
> > and instead returns -EINVAL for non TCP_LISTEN tcp_sock.
> > It should at least only be done under the "if (!reuse)" then.
>
> Yes, exactly. For an established TCP socket in SOCKMAP we would get
> -ENOENT because sk_reuseport_cb is not set. Which is a bit confusing
> since the map entry exists.
>
> Returning -EINVAL matches the REUSEPORT_SOCKARRAY update operation
> semantics for established TCP sockets.
>
> But this is just about returning an informative error so you're
> completely right that this should be done under "if (!reuse)" branch to
> avoid the extra cost on the happy path.
>
> > Checking SOCK_RCU_FREE to imply TCP_LISTEN is not ideal.
> > It is not immediately obvious. Why not directly check
> > TCP_LISTEN?
>
> I agree, it's not obvious. When I first saw this check in
> reuseport_array_update_check it got me puzzled too. I should have added
> an explanatory comment there.
>
> Thing is we're not matching on just TCP_LISTEN. REUSEPORT_SOCKARRAY
> allows selecting a connected UDP socket as a target as well. It takes
> some effort to set up but it's possible even if obscure.
How about this instead:
if (!reuse)
/* reuseport_array only has sk that has non NULL sk_reuseport_cb.
* The only (!reuse) case here is, the sk has already been removed from
* reuseport_array, so treat it as -ENOENT.
*
* Other maps (e.g. sock_map) do not provide this guarantee and the sk may
* never be in the reuseport to begin with.
*/
return map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY ? -ENOENT : -EINVAL;
>
> > Note that the SOCK_RCU_FREE check at the 'slow-path'
> > reuseport_array_update_check() is because reuseport_array does depend on
> > call_rcu(&sk->sk_rcu,...) to work, e.g. the reuseport_array
> > does not hold the sk_refcnt.
>
> Oh, so it's not only about socket state like I thought.
>
> This raises the question - does REUSEPORT_SOCKARRAY allow storing
> connected UDP sockets by design or is it a happy accident? It doesn't
> seem particularly useful.
Not by design/accident on the REUSEPORT_SOCKARRAY side ;)
The intention of REUSEPORT_SOCKARRAY is to allow sk that can be added to
reuse->socks[].
next prev parent reply other threads:[~2019-11-26 19:03 UTC|newest]
Thread overview: 39+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-11-23 11:07 [PATCH bpf-next 0/8] Extend SOCKMAP to store listening sockets Jakub Sitnicki
2019-11-23 11:07 ` [PATCH bpf-next 1/8] bpf, sockmap: Return socket cookie on lookup from syscall Jakub Sitnicki
2019-11-24 5:32 ` John Fastabend
2019-11-23 11:07 ` [PATCH bpf-next 2/8] bpf, sockmap: Let all kernel-land lookup values in SOCKMAP Jakub Sitnicki
2019-11-24 5:35 ` John Fastabend
2019-11-23 11:07 ` [PATCH bpf-next 3/8] bpf, sockmap: Allow inserting listening TCP sockets into SOCKMAP Jakub Sitnicki
2019-11-24 5:38 ` John Fastabend
2019-11-23 11:07 ` [PATCH bpf-next 4/8] bpf, sockmap: Don't let child socket inherit psock or its ops on copy Jakub Sitnicki
2019-11-24 5:56 ` John Fastabend
2019-11-25 22:38 ` Martin Lau
2019-11-26 15:54 ` Jakub Sitnicki
2019-11-26 17:16 ` Martin Lau
2019-11-26 18:36 ` Jakub Sitnicki
[not found] ` <87sglsfdda.fsf@cloudflare.com>
2019-12-11 17:20 ` Martin Lau
2019-12-12 11:27 ` Jakub Sitnicki
2019-12-12 19:23 ` Martin Lau
2019-12-17 15:06 ` Jakub Sitnicki
2019-11-26 18:43 ` John Fastabend
2019-11-27 22:18 ` Jakub Sitnicki
2019-11-23 11:07 ` [PATCH bpf-next 5/8] bpf: Allow selecting reuseport socket from a SOCKMAP Jakub Sitnicki
2019-11-24 5:57 ` John Fastabend
2019-11-25 1:24 ` Alexei Starovoitov
2019-11-25 4:17 ` John Fastabend
2019-11-25 10:40 ` Jakub Sitnicki
2019-11-25 22:07 ` Martin Lau
2019-11-26 14:30 ` Jakub Sitnicki
2019-11-26 19:03 ` Martin Lau [this message]
2019-11-27 21:34 ` Jakub Sitnicki
2019-11-23 11:07 ` [PATCH bpf-next 6/8] libbpf: Recognize SK_REUSEPORT programs from section name Jakub Sitnicki
2019-11-24 5:57 ` John Fastabend
2019-11-23 11:07 ` [PATCH bpf-next 7/8] selftests/bpf: Extend SK_REUSEPORT tests to cover SOCKMAP Jakub Sitnicki
2019-11-24 6:00 ` John Fastabend
2019-11-25 22:30 ` Martin Lau
2019-11-26 14:32 ` Jakub Sitnicki
2019-12-12 10:30 ` Jakub Sitnicki
2019-11-23 11:07 ` [PATCH bpf-next 8/8] selftests/bpf: Tests for SOCKMAP holding listening sockets Jakub Sitnicki
2019-11-24 6:04 ` John Fastabend
2019-11-24 6:10 ` [PATCH bpf-next 0/8] Extend SOCKMAP to store " John Fastabend
2019-11-25 9:22 ` Jakub Sitnicki
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=20191126190301.quwvjihpdzfjhdbe@kafai-mbp.dhcp.thefacebook.com \
--to=kafai@fb.com \
--cc=alexei.starovoitov@gmail.com \
--cc=bpf@vger.kernel.org \
--cc=jakub@cloudflare.com \
--cc=john.fastabend@gmail.com \
--cc=kernel-team@cloudflare.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).