Content-Type: multipart/mixed; boundary="===============6280769739976966801==" MIME-Version: 1.0 From: Mat Martineau To: mptcp at lists.01.org Subject: [MPTCP] Re: [PATCH mptcp-next v3] mptcp: track window announced to peer Date: Thu, 22 Oct 2020 17:43:57 -0700 Message-ID: In-Reply-To: 20201022212738.24258-1-fw@strlen.de X-Status: X-Keywords: X-UID: 6413 --===============6280769739976966801== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable On Thu, 22 Oct 2020, Florian Westphal wrote: > OoO handling attemtps to detect when packet is out-of-window by testing > current ack sequence and remaining space vs. sequence number. > > This doesn't work reliably. Store the highest allowed sequence number > that we've announced and use it to detect oow packets. > > Do this when mptcp options get written to the packet (wire format). > For this to work we need to move the write_options call until after stack > selected a new tcp window. > > Signed-off-by: Florian Westphal > --- > Changes in v3: > - add a few line breaks > - update rcvwnd even for non-dss packets > > include/net/mptcp.h | 3 ++- > net/ipv4/tcp_output.c | 11 +++++++---- > net/mptcp/options.c | 22 +++++++++++++++++++++- > net/mptcp/protocol.c | 12 +++++++----- > net/mptcp/protocol.h | 1 + > 5 files changed, 38 insertions(+), 11 deletions(-) > > diff --git a/include/net/mptcp.h b/include/net/mptcp.h > index 6e706d838e4e..b6cf07143a8a 100644 > --- a/include/net/mptcp.h > +++ b/include/net/mptcp.h > @@ -88,7 +88,8 @@ bool mptcp_established_options(struct sock *sk, struct = sk_buff *skb, > struct mptcp_out_options *opts); > void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb); > > -void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts); > +void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp, > + struct mptcp_out_options *opts); > > /* move the skb extension owership, with the assumption that 'to' is > * newly allocated > diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c > index 9abf5a0358d5..a127856ab7df 100644 > --- a/net/ipv4/tcp_output.c > +++ b/net/ipv4/tcp_output.c > @@ -445,11 +445,12 @@ struct tcp_out_options { > struct mptcp_out_options mptcp; > }; > > -static void mptcp_options_write(__be32 *ptr, struct tcp_out_options *opt= s) > +static void mptcp_options_write(__be32 *ptr, const struct tcp_sock *tp, > + struct tcp_out_options *opts) > { > #if IS_ENABLED(CONFIG_MPTCP) > if (unlikely(OPTION_MPTCP & opts->options)) > - mptcp_write_options(ptr, &opts->mptcp); > + mptcp_write_options(ptr, tp, &opts->mptcp); > #endif > } > > @@ -701,7 +702,7 @@ static void tcp_options_write(__be32 *ptr, struct tcp= _sock *tp, > > smc_options_write(ptr, &options); > > - mptcp_options_write(ptr, opts); > + mptcp_options_write(ptr, tp, opts); > } > > static void smc_set_option(const struct tcp_sock *tp, > @@ -1348,7 +1349,6 @@ static int __tcp_transmit_skb(struct sock *sk, stru= ct sk_buff *skb, > } > } > > - tcp_options_write((__be32 *)(th + 1), tp, &opts); > skb_shinfo(skb)->gso_type =3D sk->sk_gso_type; > if (likely(!(tcb->tcp_flags & TCPHDR_SYN))) { > th->window =3D htons(tcp_select_window(sk)); These lines of code between the old tcp_options_write() location and the = new location were last touched by Eric, doing some cache miss = optimizations. It doesn't seem like populating the TCP options just after = th->window instead of just before should behave similarly with the data = cache - are you concerned about this at all? > @@ -1359,6 +1359,9 @@ static int __tcp_transmit_skb(struct sock *sk, stru= ct sk_buff *skb, > */ > th->window =3D htons(min(tp->rcv_wnd, 65535U)); > } > + > + tcp_options_write((__be32 *)(th + 1), tp, &opts); > + > #ifdef CONFIG_TCP_MD5SIG > /* Calculate the MD5 hash, as we have all we need now */ > if (md5) { > diff --git a/net/mptcp/options.c b/net/mptcp/options.c > index f2d1e27a2bc1..2e9f1c4ea008 100644 > --- a/net/mptcp/options.c > +++ b/net/mptcp/options.c > @@ -991,7 +991,24 @@ void mptcp_incoming_options(struct sock *sk, struct = sk_buff *skb) > } > } > > -void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts) > +static void mptcp_set_rwin(const struct tcp_sock *tp) > +{ > + const struct sock *ssk =3D (const struct sock *)tp; > + const struct mptcp_subflow_context *subflow; > + struct mptcp_sock *msk; > + u64 ack_seq; > + > + subflow =3D mptcp_subflow_ctx(ssk); > + msk =3D mptcp_sk(subflow->conn); > + > + ack_seq =3D READ_ONCE(msk->ack_seq) + tp->rcv_wnd; > + > + if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent))) > + WRITE_ONCE(msk->rcv_wnd_sent, ack_seq); This looks racy if packet headers on multiple subflows are created = concurrently: msk->rcv_wnd_sent could end up with an earlier sequence = number than it should have, leading to possible extra drops. It looks like = a hard race to trigger, but do you think rcv_wnd_sent should use cmpxchg = like wnd_end and snd_una? Mat > +} > + > +void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp, > + struct mptcp_out_options *opts) > { > if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK | > OPTION_MPTCP_MPC_ACK) & opts->suboptions) { > @@ -1148,4 +1165,7 @@ void mptcp_write_options(__be32 *ptr, struct mptcp_= out_options *opts) > TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); > } > } > + > + if (tp) > + mptcp_set_rwin(tp); > } > diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c > index 6c131759f3e5..cd747696e7af 100644 > --- a/net/mptcp/protocol.c > +++ b/net/mptcp/protocol.c > @@ -168,19 +168,19 @@ static void mptcp_data_queue_ofo(struct mptcp_sock = *msk, struct sk_buff *skb) > struct rb_node **p, *parent; > u64 seq, end_seq, max_seq; > struct sk_buff *skb1; > - int space; > > seq =3D MPTCP_SKB_CB(skb)->map_seq; > end_seq =3D MPTCP_SKB_CB(skb)->end_seq; > - space =3D tcp_space(sk); > - max_seq =3D space > 0 ? space + msk->ack_seq : msk->ack_seq; > + max_seq =3D READ_ONCE(msk->rcv_wnd_sent); > > pr_debug("msk=3D%p seq=3D%llx limit=3D%llx empty=3D%d", msk, seq, max_se= q, > RB_EMPTY_ROOT(&msk->out_of_order_queue)); > - if (after64(seq, max_seq)) { > + if (after64(end_seq, max_seq)) { > /* out of window */ > mptcp_drop(sk, skb); > - pr_debug("oow by %ld", (unsigned long)seq - (unsigned long)max_seq); > + pr_debug("oow by %lld, rcv_wnd_sent %llu\n", > + (unsigned long long)end_seq - (unsigned long)max_seq, > + (unsigned long long)msk->rcv_wnd_sent); > MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW); > return; > } > @@ -2258,6 +2258,7 @@ struct sock *mptcp_sk_clone(const struct sock *sk, > mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq); > ack_seq++; > WRITE_ONCE(msk->ack_seq, ack_seq); > + WRITE_ONCE(msk->rcv_wnd_sent, ack_seq); > } > > #if !IS_ENABLED(CONFIG_KASAN) > @@ -2567,6 +2568,7 @@ void mptcp_finish_connect(struct sock *ssk) > WRITE_ONCE(msk->write_seq, subflow->idsn + 1); > WRITE_ONCE(msk->snd_nxt, msk->write_seq); > WRITE_ONCE(msk->ack_seq, ack_seq); > + WRITE_ONCE(msk->rcv_wnd_sent, ack_seq); > WRITE_ONCE(msk->can_ack, 1); > atomic64_set(&msk->snd_una, msk->write_seq); > > diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h > index b4c8dbe9236b..bf15c2fc0ba1 100644 > --- a/net/mptcp/protocol.h > +++ b/net/mptcp/protocol.h > @@ -211,6 +211,7 @@ struct mptcp_sock { > u64 write_seq; > u64 snd_nxt; > u64 ack_seq; > + u64 rcv_wnd_sent; > u64 rcv_data_fin_seq; > struct sock *last_snd; > int snd_burst; > -- = > 2.26.2 -- Mat Martineau Intel --===============6280769739976966801==--