Re: [patch v2] tcp.7: Add description for TCP_FASTOPEN and TCP_FASTOPEN_CONNECT options

On Mon, Sep 20, 2021 at 12:50 PM Alejandro Colomar (man-pages)
[off-list ref] wrote:

Hello Wei,

On 9/17/21 6:17 AM, Wei Wang wrote:

quoted

TCP_FASTOPEN socket option was added by:
commit 8336886f786fdacbc19b719c1f7ea91eb70706d4
TCP_FASTOPEN_CONNECT socket option was added by the following patch
series:
commit 065263f40f0972d5f1cd294bb0242bd5aa5f06b2
commit 25776aa943401662617437841b3d3ea4693ee98a
commit 19f6d3f3c8422d65b5e3d2162e30ef07c6e21ea2
commit 3979ad7e82dfe3fb94a51c3915e64ec64afa45c3
Add detailed description for these 2 options.
Also add descriptions for /proc entry tcp_fastopen and tcp_fastopen_key.

Signed-off-by: Wei Wang <redacted>
Reviewed-by: Yuchung Cheng <redacted>

Thanks for the patch (and the review, Yuchung)!

Please see some comments below.

Cheers,

Alex

quoted

---
Change in v2: corrected some format issues

  man7/tcp.7 | 110 +++++++++++++++++++++++++++++++++++++++++++++++++++++
  1 file changed, 110 insertions(+)

diff --git a/man7/tcp.7 b/man7/tcp.7
index 0a7c61a37..5a6fa7f50 100644
--- a/man7/tcp.7
+++ b/man7/tcp.7

@@ -423,6 +423,28 @@ option.
  .\" Since 2.4.0-test7
  Enable RFC\ 2883 TCP Duplicate SACK support.
  .TP
+.IR tcp_fastopen  " (Bitmask; default: 0x1; since Linux 3.7)"
+Enables RFC\ 7413 Fast Open support.
+The flag is used as a bitmap with the following values:
+.RS
+.IP 0x1
+Enables client side Fast Open support
+.IP 0x2
+Enables server side Fast Open support
+.IP 0x4
+Allows client side to transmit data in SYN without Fast Open option
+.IP 0x200
+Allows server side to accept SYN data without Fast Open option
+.IP 0x400
+Enables Fast Open on all listeners without
+.B TCP_FASTOPEN
+socket option
+.RE
+.TP
+.IR tcp_fastopen_key " (since Linux 3.7)"
+Set server side RFC\ 7413 Fast Open key to generate Fast Open cookie
+when server side Fast Open support is enabled.
+.TP
  .IR tcp_ecn " (Integer; default: see below; since Linux 2.4)"
  .\" Since 2.4.0-test7
  Enable RFC\ 3168 Explicit Congestion Notification.

@@ -1202,6 +1224,94 @@ Bound the size of the advertised window to this value.
  The kernel imposes a minimum size of SOCK_MIN_RCVBUF/2.
  This option should not be used in code intended to be
  portable.
+.TP
+.BR TCP_FASTOPEN " (since Linux 3.6)"
+This option enables Fast Open (RFC\ 7413) on the listener socket.
+The value specifies the maximum length of pending SYNs
+(similar to the backlog argument in
+.BR listen (2)).
+Once enabled,
+the listener socket grants the TCP Fast Open cookie on incoming
+SYN with TCP Fast Open option.
+.IP
+More importantly it accepts the data in SYN with a valid Fast Open cookie
+and responds SYN-ACK acknowledging both the data and the SYN sequence.
+.BR accept (2)
+returns a socket that is available for read and write when the handshake
+has not completed yet.
+Thus the data exchange can commence before the handshake completes.
+This option requires enabling the server-side support on sysctl
+.IR net.ipv4.tcp_fastopen
+(see above).
+For TCP Fast Open client-side support,
+see
+.BR send (2)
+.B MSG_FASTOPEN
+or
+.B TCP_FASTOPEN_CONNECT
+below.
+.TP
+.BR TCP_FASTOPEN_CONNECT " (since Linux 4.11)"
+This option enables an alternative way to perform Fast Open on the active
+side (client).
+When this option is enabled,
+.BR connect (2)
+would behave differently depending if a Fast Open cookie is available for
+the destination.
+.IP
+If a cookie is not available (i.e. first contact to the destination),
+.BR connect (2)
+behaves as usual by sending a SYN immediately,
+except the SYN would include an empty Fast Open cookie option to solicit a
+cookie.
+.IP
+If a cookie is available,
+.BR connect (2)
+would return 0 immediately but the SYN transmission is defered.
+A subsequent
+.BR write (2)
+or
+.BR sendmsg (2)
+would trigger a SYN with data plus cookie in the Fast Open option.
+In other words,
+the actual connect operation is deferred until data is supplied.
+.IP
+.B Note:
+While this option is designed for convenience,
+enabling it does change the behaviors and might set new
+.I errnos

typo?

errno values?

quoted

+of socket calls.

The above is not very clear to me.

Will update.

quoted

+With cookie present,
+.BR write (2)
+/

Does this mean an "or"?  If so, prefer the "or".

Yes. Ack.

quoted

+.BR sendmsg (2)
+must be called right after
+.BR connect (2)
+in order to send out SYN+data to complete 3WHS and establish connection.
+Calling
+.BR read (2)
+right after
+.BR connect (2)
+without
+.BR write (2)
+will cause the blocking socket to be blocked forever.

quoted

+The application should use either
+.B TCP_FASTOPEN_CONNECT
+or
+.BR send (2)

This is not clear to me.  So TCP_FASTOPEN_CONNECT can use write(2) and
sendmsg(2) (mentioned above), and TCP_FASTOPEN can only use send(2)?  Or
what did you mean?

The application should either set TCP_FASTOPEN_CONNECT socket option
before calling write() or sendmsg(), or call write() or sendmsg() with
MSG_FASTOPEN flag directly, but not both at the same time.

quoted

+with
+.B MSG_FASTOPEN ,
+instead of both on the same connection.

 From "The application ...":
Does this have relation with the text just above it?  It appears to me
to be a more generic statement that both options shouldn't be mixed, so
maybe a new paragraph is more appropriate.

I think a new line does make sense, since this is a general statement
for TCP_FASTOPEN_CONNECT option.

quoted

+.IP
+Here is the typical call flow with this new option:
+  s = socket();
+  setsockopt(s, IPPROTO_TCP, TCP_FASTOPEN_CONNECT, 1, ...);
+  connect(s);
+  write(s); // write() should always follow connect() in order to
+            // trigger SYN to go out
+  read(s)/write(s);
+  ... > +  close(s);

See man-pages(7):

    Indentation of structure definitions, shell session  logs,  and
        so on
        When  structure  definitions, shell session logs, and so on
        are included in running  text,  indent  them  by  4  spaces
        (i.e.,  a  block  enclosed by .in +4n and .in), format them
        using the .EX and EE macros, and surround them  with  suit‐
        able paragraph markers (either .PP or .IP).  For example:

                .PP
                .in +4n
                .EX
                int
                main(int argc, char *argv[])
                {
                    return 0;
                }
                .EE
                .in
                .PP

Ack. Thanks.
Will send out a new version with the above addressed.

quoted

  .SS Sockets API
  TCP provides limited support for out-of-band data,
  in the form of (a single byte of) urgent data.

--
Alejandro Colomar
Linux man-pages comaintainer; https://www.kernel.org/doc/man-pages/
http://www.alejandro-colomar.es/