From mboxrd@z Thu Jan 1 00:00:00 1970 Authentication-Results: passt.top; dmarc=none (p=none dis=none) header.from=gibson.dropbear.id.au Authentication-Results: passt.top; dkim=pass (2048-bit key; secure) header.d=gibson.dropbear.id.au header.i=@gibson.dropbear.id.au header.a=rsa-sha256 header.s=202510 header.b=JfExeDty; dkim-atps=neutral Received: from mail.ozlabs.org (mail.ozlabs.org [IPv6:2404:9400:2221:ea00::3]) by passt.top (Postfix) with ESMTPS id 42E3F5A061E for ; Fri, 14 Nov 2025 01:47:03 +0100 (CET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gibson.dropbear.id.au; s=202510; t=1763081221; bh=yJiKpSUtcLyL+1t9p0paL34/bRea6Q0gu8NX95uVkFA=; h=Date:From:To:Cc:Subject:References:In-Reply-To:From; b=JfExeDtyz9/flm3/HPrv4WTorpCQxIPl5aQ8GLxAuXumK9CU54IKiSc+qYZnkCJJS DHa7UUCqfPSdjDtrOL3Ji7RweqntKcGxQdJDVtntCPInqsK9CLGJKhK1hrPZdBtATf Swdr6YwEk/QmQxxpfWlhY8zIfEieJgob68p04ICYdjmWKKekstDnFLnsxVIHVN8JMn 3zmgE8jOSbFaZP2dcmRTVyUY2/K2dvbRnFvZACw5SOxR+CmvNw3HqRW05msgN3n7nh z2EibMhQ7JykFb3wl4h/1bnTm06HFT57M28mLzUJ9oGZuYwMYgtyW6BHOjpJ/kElN/ 0JoRbokRqnclw== Received: by gandalf.ozlabs.org (Postfix, from userid 1007) id 4d6z4500fWz4wCv; Fri, 14 Nov 2025 11:47:00 +1100 (AEDT) Date: Fri, 14 Nov 2025 11:31:14 +1100 From: David Gibson To: Stefano Brivio Subject: Re: [PATCH v8 4/6] tcp: Resend SYN for inbound connections Message-ID: References: <20251110093137.87705-1-yuhuang@redhat.com> <20251110093137.87705-5-yuhuang@redhat.com> <20251114010053.62389abd@elisabeth> MIME-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha512; protocol="application/pgp-signature"; boundary="m2Fc3KycxwkJg2ds" Content-Disposition: inline In-Reply-To: <20251114010053.62389abd@elisabeth> Message-ID-Hash: TYPC4HDQEUNUEP4N3MD4NKQW3ERTX5I3 X-Message-ID-Hash: TYPC4HDQEUNUEP4N3MD4NKQW3ERTX5I3 X-MailFrom: dgibson@gandalf.ozlabs.org X-Mailman-Rule-Misses: dmarc-mitigation; no-senders; approved; emergency; loop; banned-address; member-moderation; nonmember-moderation; administrivia; implicit-dest; max-recipients; max-size; news-moderation; no-subject; digests; suspicious-header CC: Yumei Huang , passt-dev@passt.top X-Mailman-Version: 3.3.8 Precedence: list List-Id: Development discussion and patches for passt Archived-At: Archived-At: List-Archive: List-Archive: List-Help: List-Owner: List-Post: List-Subscribe: List-Unsubscribe: --m2Fc3KycxwkJg2ds Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Fri, Nov 14, 2025 at 01:00:53AM +0100, Stefano Brivio wrote: > On Mon, 10 Nov 2025 21:46:55 +1100 > David Gibson wrote: >=20 > > On Mon, Nov 10, 2025 at 05:31:35PM +0800, Yumei Huang wrote: > > > If a client connects while guest is not connected or ready yet, > > > resend SYN instead of just resetting connection after 10 seconds. > > >=20 > > > Use the same backoff calculation for the timeout as Linux kernel. > > >=20 > > > Link: https://bugs.passt.top/show_bug.cgi?id=3D153 > > > Signed-off-by: Yumei Huang =20 > >=20 > > Reviewed-by: David Gibson > >=20 > > Though I have one small concern remaining > >=20 > > > --- > > > tcp.c | 62 ++++++++++++++++++++++++++++++++++++++++++++++++++-------= -- > > > tcp.h | 4 ++++ > > > 2 files changed, 57 insertions(+), 9 deletions(-) > > >=20 > > > diff --git a/tcp.c b/tcp.c > > > index 2f49327..da40a99 100644 > > > --- a/tcp.c > > > +++ b/tcp.c > > > @@ -179,9 +179,11 @@ > > > * > > > * Timeouts are implemented by means of timerfd timers, set based on= flags: > > > * > > > - * - SYN_TIMEOUT: if no ACK is received from tap/guest during handsh= ake (flag > > > - * ACK_FROM_TAP_DUE without ESTABLISHED event) within this time, r= eset the > > > - * connection > > > + * - SYN_TIMEOUT_INIT: if no SYN,ACK is received from tap/guest duri= ng > > > + * handshake (flag ACK_FROM_TAP_DUE without ESTABLISHED event) wit= hin > > > + * this time, resend SYN. It's the starting timeout for the first = SYN > > > + * retry. Retry for TCP_MAX_RETRIES or (syn_retries + syn_linear_t= imeouts) > > > + * times, reset the connection > > > * > > > * - ACK_TIMEOUT: if no ACK segment was received from tap/guest, aft= er sending > > > * data (flag ACK_FROM_TAP_DUE with ESTABLISHED event), re-send da= ta from the > > > @@ -340,7 +342,7 @@ enum { > > > #define WINDOW_DEFAULT 14600 /* RFC 6928 */ > > > =20 > > > #define ACK_INTERVAL 10 /* ms */ > > > -#define SYN_TIMEOUT 10 /* s */ > > > +#define SYN_TIMEOUT_INIT 1 /* s, RFC 6928 */ > > > #define ACK_TIMEOUT 2 > > > #define FIN_TIMEOUT 60 > > > #define ACT_TIMEOUT 7200 > > > @@ -365,6 +367,13 @@ uint8_t tcp_migrate_rcv_queue [TCP_MIGRATE_RCV_= QUEUE_MAX]; > > > =20 > > > #define TCP_MIGRATE_RESTORE_CHUNK_MIN 1024 /* Try smaller when above= this */ > > > =20 > > > +#define SYN_RETRIES "/proc/sys/net/ipv4/tcp_syn_retries" > > > +#define SYN_LINEAR_TIMEOUTS "/proc/sys/net/ipv4/tcp_syn_linear_timeo= uts" > > > + > > > +#define SYN_RETRIES_DEFAULT 6 > > > +#define SYN_LINEAR_TIMEOUTS_DEFAULT 4 > > > +#define MAX_SYNCNT 127 /* derived from kernel's limit */ > > > + > > > /* "Extended" data (not stored in the flow table) for TCP flow migra= tion */ > > > static struct tcp_tap_transfer_ext migrate_ext[FLOW_MAX]; > > > =20 > > > @@ -585,10 +594,13 @@ static void tcp_timer_ctl(const struct ctx *c, = struct tcp_tap_conn *conn) > > > if (conn->flags & ACK_TO_TAP_DUE) { > > > it.it_value.tv_nsec =3D (long)ACK_INTERVAL * 1000 * 1000; > > > } else if (conn->flags & ACK_FROM_TAP_DUE) { > > > - if (!(conn->events & ESTABLISHED)) > > > - it.it_value.tv_sec =3D SYN_TIMEOUT; > > > - else > > > + if (!(conn->events & ESTABLISHED)) { > > > + int exp =3D conn->retries - c->tcp.syn_linear_timeouts; =20 > >=20 > > As discussed in the thread on the last version, the subtraction will > > be done unsigned, so the final result depends on behaviour of casting > > a "negative" (that is, close to max value) unsigned value into a > > signed variable. I'm pretty sure that will do the right thing in > > practice, but I don't believe that behaviour is guaranteed by the C > > standard. >=20 > Surprisingly to me, C99 6.2.6.2 "Integer types" admits three > interpretations of the sign bit for signed types, quoting: >=20 > =E2=80=94 the corresponding value with sign bit 0 is negated (sign and > magnitude); > =E2=80=94 the sign bit has the value =E2=88=92(2N ) (two=E2=80=99s comp= lement); > =E2=80=94 the sign bit has the value =E2=88=92(2N =E2=88=92 1) (ones=E2= =80=99 complement). >=20 > and this, together with the conversion rule from 6.3.1.3 "Signed > and unsigned integers": >=20 > Otherwise, the new type is signed and the value cannot be represented > in it; either the result is implementation-defined or an > implementation-defined signal is raised. >=20 > seems to confirm your interpretation, even though I'm really having > a hard time convincing myself that we should actually do stuff like: >=20 > int exp; >=20 > exp =3D (int)conn->retries - c->tcp.syn_linear_timeouts; >=20 > and I still have the suspicion we're missing something. Is this syntax > what you're suggesting, by the way? Yes, that's what I'm suggesting. > > This probably isn't worth a respin, though. >=20 > Maybe not but: >=20 > > > + it.it_value.tv_sec =3D SYN_TIMEOUT_INIT << MAX(exp, 0); > > > + } > > > + else { >=20 > this is another bit I would need to change. The coding style dictates: >=20 > if (x) { > ... > } else { > ... > } >=20 > *not*: >=20 > if (x) { > ... > } > else { > ... > } >=20 > ...on the other hand, it goes away in 5/6, so I don't care > particularly. Right, that was my reasoning too. Mind you might as well fix it if there's going to be a respin anyway. > It still matters a very tiny bit because if we need to revert the > commit resulting from 5/6 for any reason we'll end up with broken > coding style. But that's not a realistic scenario and it's very minor > anyway. >=20 > > > it.it_value.tv_sec =3D ACK_TIMEOUT; > > > + } > > > } else if (CONN_HAS(conn, SOCK_FIN_SENT | TAP_FIN_ACKED)) { > > > it.it_value.tv_sec =3D FIN_TIMEOUT; > > > } else { > > > @@ -2425,8 +2437,18 @@ void tcp_timer_handler(const struct ctx *c, un= ion epoll_ref ref) > > > tcp_timer_ctl(c, conn); > > > } else if (conn->flags & ACK_FROM_TAP_DUE) { > > > if (!(conn->events & ESTABLISHED)) { > > > - flow_dbg(conn, "handshake timeout"); > > > - tcp_rst(c, conn); > > > + int max; > > > + max =3D c->tcp.syn_retries + c->tcp.syn_linear_timeouts; > > > + max =3D MIN(TCP_MAX_RETRIES, max); > > > + if (conn->retries >=3D max) { > > > + flow_dbg(conn, "handshake timeout"); > > > + tcp_rst(c, conn); > > > + } else { > > > + flow_trace(conn, "SYN timeout, retry"); > > > + tcp_send_flag(c, conn, SYN); > > > + conn->retries++; > > > + tcp_timer_ctl(c, conn); > > > + } > > > } else if (CONN_HAS(conn, SOCK_FIN_SENT | TAP_FIN_ACKED)) { > > > flow_dbg(conn, "FIN timeout"); > > > tcp_rst(c, conn); > > > @@ -2782,6 +2804,26 @@ static socklen_t tcp_probe_tcp_info(void) > > > return sl; > > > } > > > =20 > > > +/** > > > + * tcp_get_rto_params() - Get host kernel RTO parameters > > > + * @c: Execution context > > > + */ > > > +void tcp_get_rto_params(struct ctx *c) > > > +{ > > > + intmax_t v; > > > + > > > + v =3D read_file_integer(SYN_RETRIES, SYN_RETRIES_DEFAULT); > > > + c->tcp.syn_retries =3D MIN(v, MAX_SYNCNT); > > > + > > > + v =3D read_file_integer(SYN_LINEAR_TIMEOUTS, SYN_LINEAR_TIMEOUTS_DE= FAULT); > > > + c->tcp.syn_linear_timeouts =3D MIN(v, MAX_SYNCNT); > > > + > > > + debug("Read sysctl values syn_retries: %"PRIu8 > > > + ", syn_linear_timeouts: %"PRIu8, > > > + c->tcp.syn_retries, > > > + c->tcp.syn_linear_timeouts); > > > +} > > > + > > > /** > > > * tcp_init() - Get initial sequence, hash secret, initialise per-so= cket data > > > * @c: Execution context > > > @@ -2792,6 +2834,8 @@ int tcp_init(struct ctx *c) > > > { > > > ASSERT(!c->no_tcp); > > > =20 > > > + tcp_get_rto_params(c); > > > + > > > tcp_sock_iov_init(c); > > > =20 > > > memset(init_sock_pool4, 0xff, sizeof(init_sock_pool4)); > > > diff --git a/tcp.h b/tcp.h > > > index 0082386..37d7758 100644 > > > --- a/tcp.h > > > +++ b/tcp.h > > > @@ -60,12 +60,16 @@ union tcp_listen_epoll_ref { > > > * @fwd_out: Port forwarding configuration for outbound packets > > > * @timer_run: Timestamp of most recent timer run > > > * @pipe_size: Size of pipes for spliced connections > > > + * @syn_retries: SYN retries using exponential backoff timeout > > > + * @syn_linear_timeouts: SYN retries before using exponential backof= f timeout > > > */ > > > struct tcp_ctx { > > > struct fwd_ports fwd_in; > > > struct fwd_ports fwd_out; > > > struct timespec timer_run; > > > size_t pipe_size; > > > + uint8_t syn_retries; > > > + uint8_t syn_linear_timeouts; > > > }; > > > =20 > > > #endif /* TCP_H */ > > > --=20 > > > 2.51.0 >=20 > --=20 > Stefano >=20 --=20 David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson --m2Fc3KycxwkJg2ds Content-Type: application/pgp-signature; name=signature.asc -----BEGIN PGP SIGNATURE----- iQIzBAEBCgAdFiEEO+dNsU4E3yXUXRK2zQJF27ox2GcFAmkWeFEACgkQzQJF27ox 2GcDNg//aBM26vWwR54ttn9zfq4NYBq+MpYswP3gcbXW8ADnCwVzKBY1JpxRRYdm HCGjgmzTSGnfMo5hDvvs6cqdxaCKIGFamAOmaSQnsb9ymuHfUVgNagETCrRDckY+ 0y+E4QG5bN7ZBG9RlZVengBTWutXsVkGRMgOQOyRLCCnYY9vbdjnXl5f0ncn0eQJ DknPGTx6zt0JVwqvcfFBGDoWVz+vxQN+tmmsMpym8BZ5UePRHA+rUFQ5Ppxq7Max +Zhs0L9Nj5Hs5WcDySYxH9CjxQlMFjaCcZ6c5gmfqmggNDzVct5W2/4tpoMsaTTj gC3fGX6QlmLuMC5x/lhb69RalPvuodhclpUk/v/jNBOHm0uzEZTCEc5KgTSit6mX XcwDmQSuN6spQdvQX71wTVCfYU8n3gim/1cNb4xU1nfwSazCGfxDWeJHAN+L5ehh ywNrCy/ReuGPx23glB8+fd+oshPSTIEw7wj+i7S8ytJKWCdmGk1o9NZC5tPCXy2x nLcbWaLIO2rBv0HuBPLRrhKF5L92q8E7TdOAoPoaRlITVWPqzUnbq4G3LgNyjzG6 Fwbmbv7Qg4wqiqrc5c+GKIPwGG/l1z4Hf38uXJ39102k/+x1m08/uImBBreZnlqG kGPZaS4OAVWFEL76OWF0pfRY2owERtZL/dcUy3863a2kO2Bc7C8= =QzZn -----END PGP SIGNATURE----- --m2Fc3KycxwkJg2ds--