Re: [PATCH v2 2/3] tcp: Resend SYN for inbound connections

[Yumei Huang <yuhuang@redhat.com>]

On Mon, Oct 13, 2025 at 8:21 AM David Gibson
<david@gibson.dropbear.id.au> wrote:
>
> On Fri, Oct 10, 2025 at 03:46:59PM +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.
> >
> > Use the same backoff calculation for the timeout as linux kernel.
> >
> > Signed-off-by: Yumei Huang <yuhuang@redhat.com>
> > ---
> >  tcp.c | 95 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
> >  tcp.h |  2 ++
> >  2 files changed, 89 insertions(+), 8 deletions(-)
> >
> > diff --git a/tcp.c b/tcp.c
> > index 2ec4b0c..85bbdac 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 handshake (flag
> > - *   ACK_FROM_TAP_DUE without ESTABLISHED event) within this time, reset the
> > - *   connection
> > + * - SYN_TIMEOUT_INIT: if no ACK is received from tap/guest during handshake
> > + *   (flag ACK_FROM_TAP_DUE without ESTABLISHED event) within this time, resend
> > + *   SYN. It's the starting timeout for the first SYN retry. If this persists
> > + *   for more than TCP_MAX_RETRIES or (tcp_syn_retries +
> > + *   tcp_syn_linear_timeouts) times in a row, reset the connection
> >   *
> >   * - ACK_TIMEOUT: if no ACK segment was received from tap/guest, after sending
> >   *   data (flag ACK_FROM_TAP_DUE with ESTABLISHED event), re-send data from the
> > @@ -309,6 +311,7 @@
> >  #include "tcp_internal.h"
> >  #include "tcp_buf.h"
> >  #include "tcp_vu.h"
> > +#include "lineread.h"
> >
> >  /*
> >   * The size of TCP header (including options) is given by doff (Data Offset)
> > @@ -340,7 +343,7 @@ enum {
> >  #define WINDOW_DEFAULT                       14600           /* RFC 6928 */
> >
> >  #define ACK_INTERVAL                 10              /* ms */
> > -#define SYN_TIMEOUT                  10              /* s */
> > +#define SYN_TIMEOUT_INIT             1               /* s */
> >  #define ACK_TIMEOUT                  2
>
> As we've discussed in some places, the RFCs largely treat SYN retries
> the same as data retransmits.  ACK_TIMEOUT controls the latter.  2s is
> a little odd, IIRC the RFC suggests 3, I'm not sure what Linux does by
> default.  In any case, I'm wondering if we should use a single define
> for the initial value of both timeouts.

Well, as Stefano pointed out, according to Appendix A in RFC 6298, the
initial value is suggested to be one second.

BTW, as this is for data retransmits which is addressed by the third
patch, I kept it not changed in this second patch.  I agree we could
use one single define for both timeouts. Just not sure how we are
supposed to split the patches. Maybe we could remove the
SYN_TIMEOUT_INIT in the third patch? What do you think?

>
> >  #define FIN_TIMEOUT                  60
> >  #define ACT_TIMEOUT                  7200
> > @@ -365,6 +368,10 @@ uint8_t tcp_migrate_rcv_queue            [TCP_MIGRATE_RCV_QUEUE_MAX];
> >
> >  #define TCP_MIGRATE_RESTORE_CHUNK_MIN        1024 /* Try smaller when above this */
> >
> > +#define TCP_SYN_RETRIES_SYSCTL               "/proc/sys/net/ipv4/tcp_syn_retries"
> > +#define TCP_SYN_LINEAR_TIMEOUTS_SYSCTL                                               \
> > +     "/proc/sys/net/ipv4/tcp_syn_linear_timeouts"
> > +
> >  /* "Extended" data (not stored in the flow table) for TCP flow migration */
> >  static struct tcp_tap_transfer_ext migrate_ext[FLOW_MAX];
> >
> > @@ -581,8 +588,13 @@ 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 = (long)ACK_INTERVAL * 1000 * 1000;
> >       } else if (conn->flags & ACK_FROM_TAP_DUE) {
> > -             if (!(conn->events & ESTABLISHED))
> > -                     it.it_value.tv_sec = SYN_TIMEOUT;
> > +             if (!(conn->events & ESTABLISHED)) {
> > +                     if (conn->retries < c->tcp.syn_linear_timeouts)
> > +                             it.it_value.tv_sec = SYN_TIMEOUT_INIT;
> > +                     else
> > +                             it.it_value.tv_sec = SYN_TIMEOUT_INIT <<
> > +                                     (conn->retries - c->tcp.syn_linear_timeouts);
> > +             }
> >               else
> >                       it.it_value.tv_sec = ACK_TIMEOUT;
> >       } else if (CONN_HAS(conn, SOCK_FIN_SENT | TAP_FIN_ACKED)) {
> > @@ -1961,6 +1973,7 @@ static void tcp_conn_from_sock_finish(const struct ctx *c,
> >       }
> >
> >       tcp_send_flag(c, conn, ACK);
> > +     conn->retries = 0;
>
> I know Stefano said you needed to add this, but on a closer
> examination I don't think you do.  tcp_tap_handler() calls
> tcp_update_seqack_from_tap() if !ESTABLISHED.  That will clear
> retrans/retries if the ack number from the guest has advanced.  That
> will occur when the guest SYN-ACKs our SYN, which is exactly the point
> at which we should clear the retries count.

I see, thanks. Will remove it in v3.
>
> >
> >       /* The client might have sent data already, which we didn't
> >        * dequeue waiting for SYN,ACK from tap -- check now.
> > @@ -2409,8 +2422,16 @@ void tcp_timer_handler(const struct ctx *c, union 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);
> > +                     if (conn->retries == MIN(TCP_MAX_RETRIES,
> > +                             (c->tcp.tcp_syn_retries + c->tcp.tcp_syn_retries))) {
>
> Should this be tcp_syn_retries + linear_timeouts?

Yes, sorry, there must be a copy error. I should have double checked
before sending out.
>
> Also, probably safer to use a >= rather than == - if we somehow send
> an extra retry, we don't want to then retry forever.
>
> > +                             flow_dbg(conn, "handshake timeout");
> > +                             tcp_rst(c, conn);
> > +                     } else {
> > +                             flow_dbg(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);
> > @@ -2766,6 +2787,62 @@ static socklen_t tcp_probe_tcp_info(void)
> >       return sl;
> >  }
> >
> > +/**
> > + * get_tcp_syn_param() - Read SYN parameters from /proc/sys
> > + * @path: Path to the sysctl file
> > + * @fallback: Default value if file can't be read
> > + *
> > + * Return: Parameter value, fallback on failure
> > +*/
> > +int get_tcp_syn_param(const char *path, int fallback)
>
> I wonder if it might be worth making a new function in util.c to read
> a file containing a single number.

I can do that. So it could be reused in the future.
>
> > +{
> > +     char *line, *end;
> > +     struct lineread lr;
> > +     long value;
> > +     ssize_t len;
> > +     int fd;
> > +
> > +     fd = open(path, O_RDONLY | O_CLOEXEC);
> > +     if (fd < 0) {
> > +             debug("Unable to open %s", path);
> > +             return fallback;
> > +     }
> > +
> > +     lineread_init(&lr, fd);
> > +     len = lineread_get(&lr, &line);
> > +     close(fd);
> > +
> > +     if (len < 0) {
> > +             debug("Unable to read %s", path);
> > +             return fallback;
> > +     }
> > +
> > +     errno = 0;
> > +     value = strtol(line, &end, 10);
> > +     if (*end && *end != '\n') {
> > +             debug("Invalid format in %s", path);
> > +             return fallback;
> > +     }
> > +     if (errno || value < 0 || value > INT_MAX) {
> > +             debug("Invalid value in %s: %ld", path, value);
> > +             return fallback;
> > +     }
> > +     return (int)value;
>
> You return an (int) here, but store the value into a uint8_t in the
> caller.  That will cause unexpected results if the value you read is
> greater than 255.  Somewhere you need to check for this and clamp the
> value.  I'd suggest the function actually reading /proc return a long
> long (for reuseability), then clamp it in the caller.

Yeah, I noticed that. Seems tcp_syn_retries should not be higher than
127 according to Documentation/networking/ip-sysctl.rst, but
tcp_syn_linear_timeouts is not limited. Anyway, I should check them.
Will update in v3.


>
> > +}
> > +
> > +/**
> > + * tcp_syn_params_init() - Get initial syn params for inbound connection
> > + * @c:               Execution context
> > +*/
> > +void tcp_syn_params_init(struct ctx *c)
> > +{
> > +     c->tcp.tcp_syn_retries = get_tcp_syn_param(TCP_SYN_RETRIES_SYSCTL, 8);
> > +     c->tcp.syn_linear_timeouts =
> > +             get_tcp_syn_param(TCP_SYN_LINEAR_TIMEOUTS_SYSCTL, 1);
> > +     debug("TCP SYN parameters: retries=%d, linear_timeouts=%d",
> > +               c->tcp.tcp_syn_retries, c->tcp.syn_linear_timeouts);
> > +}
> > +
> >  /**
> >   * tcp_init() - Get initial sequence, hash secret, initialise per-socket data
> >   * @c:               Execution context
> > @@ -2776,6 +2853,8 @@ int tcp_init(struct ctx *c)
> >  {
> >       ASSERT(!c->no_tcp);
> >
> > +     tcp_syn_params_init(c);
> > +
> >       tcp_sock_iov_init(c);
> >
> >       memset(init_sock_pool4,         0xff,   sizeof(init_sock_pool4));
> > diff --git a/tcp.h b/tcp.h
> > index 234a803..df699a4 100644
> > --- a/tcp.h
> > +++ b/tcp.h
> > @@ -65,6 +65,8 @@ struct tcp_ctx {
> >       struct fwd_ports fwd_out;
> >       struct timespec timer_run;
> >       size_t pipe_size;
> > +     uint8_t tcp_syn_retries;
> > +     uint8_t syn_linear_timeouts;
> >  };
> >
> >  #endif /* TCP_H */
> > --
> > 2.47.0
> >
>
> --
> 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



-- 
Thanks,

Yumei Huang