From: Kuniyuki Iwashima <kuniyu@amazon.com>
To: <sinquersw@gmail.com>
Cc: <andrii@kernel.org>, <ast@kernel.org>, <bpf@vger.kernel.org>,
<daniel@iogearbox.net>, <davem@davemloft.net>,
<dsahern@kernel.org>, <edumazet@google.com>, <haoluo@google.com>,
<john.fastabend@gmail.com>, <jolsa@kernel.org>,
<kpsingh@kernel.org>, <kuba@kernel.org>, <kuni1840@gmail.com>,
<kuniyu@amazon.com>, <martin.lau@linux.dev>, <mykolal@fb.com>,
<netdev@vger.kernel.org>, <pabeni@redhat.com>, <sdf@google.com>,
<song@kernel.org>, <yonghong.song@linux.dev>
Subject: Re: [PATCH v1 bpf-next 00/11] bpf: tcp: Add SYN Cookie generation/validation SOCK_OPS hooks.
Date: Mon, 23 Oct 2023 18:22:08 -0700 [thread overview]
Message-ID: <20231024012208.82653-1-kuniyu@amazon.com> (raw)
In-Reply-To: <9aebb3e9-70c5-428c-bc31-7b38a04e4848@gmail.com>
> On 10/23/23 14:35, Martin KaFai Lau wrote:
> > On 10/20/23 11:48 PM, Kuniyuki Iwashima wrote:
> > > I think this was doable. With the diff below, I was able to skip
> > > validation in cookie_v[46]_check() when if skb->sk is not NULL.
> > >
> > > The kfunc allocates req and set req->syncookie to 1, which is usually
> > > set in TX path, so if it's 1 in RX (inet_steal_sock()), we can see
> > > that req is allocated by kfunc (at least, req->syncookie &&
> > > req->rsk_listener never be true in the current TCP stack).
> > >
> > > The difference here is that req allocated by kfunc holds refcnt of
> > > rsk_listener (passing true to inet_reqsk_alloc()) to prevent freeing
> > > the listener until req reaches cookie_v[46]_check().
> >
> > The cookie_v[46]_check() holds the listener sk refcnt now?
>
> The caller of cookie_v[46]_check() should hold a refcnt of the listener.
No, it need not.
When we handle the default syn cookie, cookie_tcp_reqsk_alloc() passes
false to inet_reqsk_alloc(), then reqsk does not hold refcnt of the
listener.
If inet_csk_reqsk_queue_add() in tcp_get_cookie_sock() succeeds, we know
the listener is still alive.
> If the listener is destroyed, the callers of cookie_v[46]_check() should
> fail to lookup a sock for the skb. However, in this case, the kfunc sets
> a sock to skb->sk, and the lookup function
> (__inet_lookup_skb()) steals sock from skb. So, there is no guarantee
> ensuring the listener is still alive.
>
> One solution is let the stealing function to lookup the listener if
> inet_reqsk(skb->sk)->syncookie is true.
kfunc at least guarantees that the listener is not freed until req
is freed. There's two cases where the listener could be close()d
after kfunc:
1. close()d before lookup
-> kfree_skb(skb) calls reqsk_put() and releases the last
refcnt of the listener
2. close()d between lookup and inet_csk_reqsk_queue_add()
-> inet_csk_reqsk_queue_add() fails and __reqsk_free()
releases the last refcnt of the listener.
So, we need not look up the listener again in inet_steal_sock().
> >
> > >
> > > The cookie generation at least should be done at tc/xdp. The
> > > valdation can be done earlier as well on tc/xdp, but it could
> > > add another complexity, listener's life cycle if we allocate
> > > req there.
> >
> > I think your code below looks pretty close already.
> >
> > It seems the only concern/complexity is the extra rsk_listener refcnt (btw the
> > concern is on performance for the extra refcnt? or there is correctness issue?).
Yes, that's the only concern and I think it's all ok now.
[ I was seeing a weird refcnt warning, but I missed *refcounted was true
in inet_steal_sock() for reqsk and forgot to flipping it to false :S ]
> >
> > Asking because bpf_sk_assign() can already assign a listener to skb->sk and it
> > also does not take a refcnt on the listener. The same no refcnt needed on
> > req->rsk_listener should be doable also. sock_pfree may need to be smarter to
> > check req->syncookie. What else may need to change?
I was wondering if we are in the same RCU period between tc and
cookie_v[46]_check(), but yeah, probably sock_pfree() can check
req->syncookie and set NULL to rsk_listener so that reqsk_put()
will not touch the listener.
> >
> > >
> > > I'm wondering which place to add the validation capability, and
> > > I think SOCK_OPS is simpler than tc.
> > >
> > > #1 validate cookie and allocate req at tc, and skip validation
> > >
> > > #2 validate cookie (and update bpf map at xdp/tc, and look up bpf
> > > map) and allocate req at SOCK_OPS hook
> > >
> > > Given SYN proxy is usually on the other node and incoming cookie
> > > is almost always valid, we might need not validate it in the early
> > > stage in the stack.
> > >
> > > What do you think ?
> >
> > Yeah, supporting validation in sock_ops is an open option if the tc side is too
> > hard but I feel you are pretty close on the tc side.
Now I think I can go v2 with tc.
Thanks for your guide!
> >
> > >
> > > ---8<---
> > > diff --git a/include/net/inet_hashtables.h b/include/net/inet_hashtables.h
> > > index 3ecfeadbfa06..e5e4627bf270 100644
> > > --- a/include/net/inet_hashtables.h
> > > +++ b/include/net/inet_hashtables.h
> > > @@ -462,9 +462,19 @@ struct sock *inet_steal_sock(struct net *net, struct sk_buff *skb, int doff,
> > > if (!sk)
> > > return NULL;
> > >
> > > - if (!prefetched || !sk_fullsock(sk))
> > > + if (!prefetched)
> > > return sk;
> > >
> > > + if (!sk_fullsock(sk)) {
> > > + if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
> > > + skb->sk = sk;
> > > + skb->destructor = sock_pfree;
> > > + sk = inet_reqsk(sk)->rsk_listener;
> > > + }
> > > +
> > > + return sk;
> > > + }
> > > +
> > > if (sk->sk_protocol == IPPROTO_TCP) {
> > > if (sk->sk_state != TCP_LISTEN)
> > > return sk;
> > > diff --git a/net/core/filter.c b/net/core/filter.c
> > > index cc2e4babc85f..bca491ddf42c 100644
> > > --- a/net/core/filter.c
> > > +++ b/net/core/filter.c
> > > @@ -11800,6 +11800,71 @@ __bpf_kfunc int bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern *sa_kern,
> > >
> > > return 0;
> > > }
> > > +
> > > +__bpf_kfunc int bpf_sk_assign_tcp_reqsk(struct sk_buff *skb, struct sock *sk,
> > > + struct tcp_options_received *tcp_opt,
> > > + int tcp_opt__sz, u16 mss)
> > > +{
> > > + const struct tcp_request_sock_ops *af_ops;
> > > + const struct request_sock_ops *ops;
> > > + struct inet_request_sock *ireq;
> > > + struct tcp_request_sock *treq;
> > > + struct request_sock *req;
> > > +
> > > + if (!sk)
> > > + return -EINVAL;
> > > +
> > > + if (!skb_at_tc_ingress(skb))
> > > + return -EINVAL;
> > > +
> > > + if (dev_net(skb->dev) != sock_net(sk))
> > > + return -ENETUNREACH;
> > > +
> > > + switch (sk->sk_family) {
> > > + case AF_INET: /* TODO: MPTCP */
> > > + ops = &tcp_request_sock_ops;
> > > + af_ops = &tcp_request_sock_ipv4_ops;
> > > + break;
> > > +#if IS_ENABLED(CONFIG_IPV6)
> > > + case AF_INET6:
> > > + ops = &tcp6_request_sock_ops;
> > > + af_ops = &tcp_request_sock_ipv6_ops;
> > > + break;
> > > +#endif
> > > + default:
> > > + return -EINVAL;
> > > + }
> > > +
> > > + if (sk->sk_type != SOCK_STREAM || sk->sk_state != TCP_LISTEN)
> > > + return -EINVAL;
> > > +
> > > + req = inet_reqsk_alloc(ops, sk, true);
> > > + if (!req)
> > > + return -ENOMEM;
> > > +
> > > + ireq = inet_rsk(req);
> > > + treq = tcp_rsk(req);
> > > +
> > > + refcount_set(&req->rsk_refcnt, 1);
> > > + req->syncookie = 1;
> > > + req->mss = mss;
> > > + req->ts_recent = tcp_opt->saw_tstamp ? tcp_opt->rcv_tsval : 0;
> > > +
> > > + ireq->snd_wscale = tcp_opt->snd_wscale;
> > > + ireq->sack_ok = tcp_opt->sack_ok;
> > > + ireq->wscale_ok = tcp_opt->wscale_ok;
> > > + ireq->tstamp_ok = tcp_opt->saw_tstamp;
> > > +
> > > + tcp_rsk(req)->af_specific = af_ops;
> > > + tcp_rsk(req)->ts_off = tcp_opt->rcv_tsecr - tcp_ns_to_ts(tcp_clock_ns());
> > > +
> > > + skb_orphan(skb);
> > > + skb->sk = req_to_sk(req);
> > > + skb->destructor = sock_pfree;
> > > +
> > > + return 0;
> > > +}
> > > +
> > > __diag_pop();
> > >
> > > int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
> > > @@ -11828,6 +11893,10 @@ BTF_SET8_START(bpf_kfunc_check_set_sock_addr)
> > > BTF_ID_FLAGS(func, bpf_sock_addr_set_sun_path)
> > > BTF_SET8_END(bpf_kfunc_check_set_sock_addr)
> > >
> > > +BTF_SET8_START(bpf_kfunc_check_set_tcp_reqsk)
> > > +BTF_ID_FLAGS(func, bpf_sk_assign_tcp_reqsk)
> > > +BTF_SET8_END(bpf_kfunc_check_set_tcp_reqsk)
> > > +
> > > static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
> > > .owner = THIS_MODULE,
> > > .set = &bpf_kfunc_check_set_skb,
> > > @@ -11843,6 +11912,11 @@ static const struct btf_kfunc_id_set bpf_kfunc_set_sock_addr = {
> > > .set = &bpf_kfunc_check_set_sock_addr,
> > > };
> > >
> > > +static const struct btf_kfunc_id_set bpf_kfunc_set_tcp_reqsk = {
> > > + .owner = THIS_MODULE,
> > > + .set = &bpf_kfunc_check_set_tcp_reqsk,
> > > +};
> > > +
> > > static int __init bpf_kfunc_init(void)
> > > {
> > > int ret;
> > > @@ -11858,8 +11932,10 @@ static int __init bpf_kfunc_init(void)
> > > ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
> > > ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
> > > ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
> > > - return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
> > > - &bpf_kfunc_set_sock_addr);
> > > + ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
> > > + &bpf_kfunc_set_sock_addr);
> > > + ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_tcp_reqsk);
> > > + return ret;
> > > }
> > > late_initcall(bpf_kfunc_init);
> > >
> > > ---8<---
next prev parent reply other threads:[~2023-10-24 1:22 UTC|newest]
Thread overview: 44+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-10-13 22:04 [PATCH v1 bpf-next 00/11] bpf: tcp: Add SYN Cookie generation/validation SOCK_OPS hooks Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 01/11] tcp: Clean up reverse xmas tree in cookie_v[46]_check() Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 02/11] tcp: Cache sock_net(sk) " Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 03/11] tcp: Clean up goto labels " Kuniyuki Iwashima
2023-10-17 0:00 ` Kui-Feng Lee
2023-10-17 0:30 ` Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 04/11] tcp: Don't initialise tp->tsoffset in tcp_get_cookie_sock() Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 05/11] bpf: tcp: Add SYN Cookie generation SOCK_OPS hook Kuniyuki Iwashima
2023-10-18 0:54 ` Martin KaFai Lau
2023-10-18 17:00 ` Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 06/11] bpf: tcp: Add SYN Cookie validation " Kuniyuki Iwashima
2023-10-16 20:38 ` Stanislav Fomichev
2023-10-16 22:02 ` Kuniyuki Iwashima
2023-10-17 16:52 ` Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 07/11] bpf: Make bpf_sock_ops.replylong[1] writable Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 08/11] bpf: tcp: Make TS available for SYN Cookie storage Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 09/11] tcp: Split cookie_ecn_ok() Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 10/11] bpf: tcp: Make WS, SACK, ECN configurable from BPF SYN Cookie Kuniyuki Iwashima
2023-10-18 1:08 ` Martin KaFai Lau
2023-10-18 17:02 ` Kuniyuki Iwashima
2023-10-13 22:04 ` [PATCH v1 bpf-next 11/11] selftest: bpf: Test BPF_SOCK_OPS_(GEN|CHECK)_SYNCOOKIE_CB Kuniyuki Iwashima
2023-10-17 5:50 ` Martin KaFai Lau
2023-10-17 16:29 ` Kuniyuki Iwashima
2023-10-16 13:05 ` [PATCH v1 bpf-next 00/11] bpf: tcp: Add SYN Cookie generation/validation SOCK_OPS hooks Daniel Borkmann
2023-10-16 16:11 ` Kuniyuki Iwashima
2023-10-16 14:19 ` Willem de Bruijn
2023-10-16 16:46 ` Kuniyuki Iwashima
2023-10-16 18:41 ` Willem de Bruijn
2023-10-17 5:53 ` Martin KaFai Lau
2023-10-17 16:48 ` Kuniyuki Iwashima
2023-10-18 6:19 ` Martin KaFai Lau
2023-10-18 8:02 ` Eric Dumazet
2023-10-18 17:20 ` Kuniyuki Iwashima
2023-10-18 21:47 ` Kui-Feng Lee
2023-10-18 22:31 ` Kuniyuki Iwashima
2023-10-19 7:25 ` Martin KaFai Lau
2023-10-19 18:01 ` Kuniyuki Iwashima
2023-10-20 19:59 ` Martin KaFai Lau
2023-10-20 23:10 ` Kuniyuki Iwashima
2023-10-21 6:48 ` Kuniyuki Iwashima
2023-10-23 21:35 ` Martin KaFai Lau
2023-10-24 0:37 ` Kui-Feng Lee
2023-10-24 1:22 ` Kuniyuki Iwashima [this message]
2023-10-24 17:55 ` Kui-Feng Lee
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=20231024012208.82653-1-kuniyu@amazon.com \
--to=kuniyu@amazon.com \
--cc=andrii@kernel.org \
--cc=ast@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=daniel@iogearbox.net \
--cc=davem@davemloft.net \
--cc=dsahern@kernel.org \
--cc=edumazet@google.com \
--cc=haoluo@google.com \
--cc=john.fastabend@gmail.com \
--cc=jolsa@kernel.org \
--cc=kpsingh@kernel.org \
--cc=kuba@kernel.org \
--cc=kuni1840@gmail.com \
--cc=martin.lau@linux.dev \
--cc=mykolal@fb.com \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=sdf@google.com \
--cc=sinquersw@gmail.com \
--cc=song@kernel.org \
--cc=yonghong.song@linux.dev \
/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).