4. TRANSPORT PROTOCOLS
4.1 USER DATAGRAM PROTOCOL -- UDP
4.1.1 INTRODUCTION
The User Datagram Protocol UDP [UDP:1] offers only a minimal transport service -- non-guaranteed datagram delivery -- and gives applications direct access to the datagram service of the IP layer. UDP is used by applications that do not require the level of service of TCP or that wish to use communications services (e.g., multicast or broadcast delivery) not available from TCP.
UDP is almost a null protocol; the only services it provides over IP are checksumming of data and multiplexing by port number. Therefore, an application program running over UDP must deal directly with end-to-end communication problems that a connection-oriented protocol would have handled -- e.g., retransmission for reliable delivery, packetization and reassembly, flow control, congestion avoidance, etc., when these are required. The fairly complex coupling between IP and TCP will be mirrored in the coupling between UDP and many applications using UDP.
4.1.2 PROTOCOL WALK-THROUGH
There are no known errors in the specification of UDP.
4.1.3 SPECIFIC ISSUES
4.1.3.1 Ports
UDP well-known ports follow the same rules as TCP well-known
ports; see Section 4.2.2.1 below.
If a datagram arrives addressed to a UDP port for which
there is no pending LISTEN call, UDP **SHOULD** send an ICMP
Port Unreachable message.
4.1.3.2 IP Options
UDP **MUST** pass any IP option that it receives from the IP
layer transparently to the application layer.
An application **MUST** be able to specify IP options to be sent
in its UDP datagrams, and UDP **MUST** pass these options to the
IP layer.
DISCUSSION:
At present, the only options that need be passed
through UDP are Source Route, Record Route, and Time
Stamp. However, new options may be defined in the
future, and UDP need not and should not make any
assumptions about the format or content of options it
passes to or from the application; an exception to this
might be an IP-layer security option.
An application based on UDP will need to obtain a
source route from a request datagram and supply a
reversed route for sending the corresponding reply.
4.1.3.3 ICMP Messages
UDP **MUST** pass to the application layer all ICMP error
messages that it receives from the IP layer. Conceptually
at least, this may be accomplished with an upcall to the
ERROR_REPORT routine (see Section 4.2.4.1).
DISCUSSION:
Note that ICMP error messages resulting from sending a
UDP datagram are received asynchronously. A UDP-based
application that wants to receive ICMP error messages
is responsible for maintaining the state necessary to
demultiplex these messages when they arrive; for
example, the application may keep a pending receive
operation for this purpose. The application is also
responsible to avoid confusion from a delayed ICMP
error message resulting from an earlier use of the same
port(s).
4.1.3.4 UDP Checksums
A host **MUST** implement the facility to generate and validate
UDP checksums. An application **MAY** optionally be able to
control whether a UDP checksum will be generated, but it
**MUST** default to checksumming on.
If a UDP datagram is received with a checksum that is non-
zero and invalid, UDP **MUST** silently discard the datagram.
An application **MAY** optionally be able to control whether UDP
datagrams without checksums should be discarded or passed to
the application.
DISCUSSION:
Some applications that normally run only across local
area networks have chosen to turn off UDP checksums for
efficiency. As a result, numerous cases of undetected
errors have been reported. The advisability of ever
turning off UDP checksumming is very controversial.
IMPLEMENTATION:
There is a common implementation error in UDP
checksums. Unlike the TCP checksum, the UDP checksum
is optional; the value zero is transmitted in the
checksum field of a UDP header to indicate the absence
of a checksum. If the transmitter really calculates a
UDP checksum of zero, it must transmit the checksum as
all 1's (65535). No special action is required at the
receiver, since zero and 65535 are equivalent in 1's
complement arithmetic.
4.1.3.5 UDP Multihoming
When a UDP datagram is received, its specific-destination
address **MUST** be passed up to the application layer.
An application program **MUST** be able to specify the IP source
address to be used for sending a UDP datagram or to leave it
unspecified (in which case the networking software will
choose an appropriate source address). There **SHOULD** be a
way to communicate the chosen source address up to the
application layer (e.g, so that the application can later
receive a reply datagram only from the corresponding
interface).
DISCUSSION:
A request/response application that uses UDP should use
a source address for the response that is the same as
the specific destination address of the request. See
the "General Issues" section of [INTRO:1].
4.1.3.6 Invalid Addresses
A UDP datagram received with an invalid IP source address
(e.g., a broadcast or multicast address) must be discarded
by UDP or by the IP layer (see Section 3.2.1.3).
When a host sends a UDP datagram, the source address **MUST** be
(one of) the IP address(es) of the host.
4.1.4 UDP/APPLICATION LAYER INTERFACE
The application interface to UDP MUST provide the full services of the IP/transport interface described in Section 3.4 of this
document. Thus, an application using UDP needs the functions
of the GET_SRCADDR(), GET_MAXSIZES(), ADVISE_DELIVPROB(), and
RECV_ICMP() calls described in Section 3.4. For example,
GET_MAXSIZES() can be used to learn the effective maximum UDP
maximum datagram size for a particular {interface,remote host,TOS} triplet.
An application-layer program MUST be able to set the TTL and TOS values as well as IP options for sending a UDP datagram, and these values must be passed transparently to the IP layer. UDP MAY pass the received TOS up to the application layer.
4.1.5 UDP REQUIREMENTS SUMMARY
| | | | |S| |
| | | | |H| |F
| | | | |O|M|o
| | |S| |U|U|o
| | |H| |L|S|t
| |M|O| |D|T|n
| |U|U|M| | |o
| |S|L|A|N|N|t
| |T|D|Y|O|O|t
| FEATURE | SECTION | T | T | e |
|---|
| | | | | | |
UDP | | | | | | |
-------------------------------------------------|--------|-|-|-|-|-|-- | | | | | | | UDP send Port Unreachable |4.1.3.1 | |x| | | | | | | | | | | IP Options in UDP | | | | | | |
-
Pass rcv'd IP options to applic layer |4.1.3.2 |x| | | | |
-
Applic layer can specify IP options in Send |4.1.3.2 |x| | | | |
-
UDP passes IP options down to IP layer |4.1.3.2 |x| | | | | | | | | | | | Pass ICMP msgs up to applic layer |4.1.3.3 |x| | | | | | | | | | | | UDP checksums: | | | | | | |
-
Able to generate/check checksum |4.1.3.4 |x| | | | |
-
Silently discard bad checksum |4.1.3.4 |x| | | | |
-
Sender Option to not generate checksum |4.1.3.4 | | |x| | |
- Default is to checksum |4.1.3.4 |x| | | | |
-
Receiver Option to require checksum |4.1.3.4 | | |x| | | | | | | | | | UDP Multihoming | | | | | | |
-
Pass spec-dest addr to application |4.1.3.5 |x| | | | |
-
Applic layer can specify Local IP addr |4.1.3.5 |x| | | | |
-
Applic layer specify wild Local IP addr |4.1.3.5 |x| | | | |
-
Applic layer notified of Local IP addr used |4.1.3.5 | |x| | | | | | | | | | | Bad IP src addr silently discarded by UDP/IP |4.1.3.6 |x| | | | | Only send valid IP source address |4.1.3.6 |x| | | | | UDP Application Interface Services | | | | | | | Full IP interface of 3.4 for application |4.1.4 |x| | | | |
-
Able to spec TTL, TOS, IP opts when send dg |4.1.4 |x| | | | |
-
Pass received TOS up to applic layer |4.1.4 | | |x| | |
4.2 TRANSMISSION CONTROL PROTOCOL -- TCP
4.2.1 INTRODUCTION
The Transmission Control Protocol TCP [TCP:1] is the primary virtual-circuit transport protocol for the Internet suite. TCP provides reliable, in-sequence delivery of a full-duplex stream of octets (8-bit bytes). TCP is used by those applications needing reliable, connection-oriented transport service, e.g., mail (SMTP), file transfer (FTP), and virtual terminal service (Telnet); requirements for these application-layer protocols are described in [INTRO:1].
4.2.2 PROTOCOL WALK-THROUGH
4.2.2.1 Well-Known Ports: RFC-793 Section 2.7
DISCUSSION:
TCP reserves port numbers in the range 0-255 for
"well-known" ports, used to access services that are
standardized across the Internet. The remainder of the
port space can be freely allocated to application
processes. Current well-known port definitions are
listed in the RFC entitled "Assigned Numbers"
[INTRO:6]. A prerequisite for defining a new well-
known port is an RFC documenting the proposed service
in enough detail to allow new implementations.
Some systems extend this notion by adding a third
subdivision of the TCP port space: reserved ports,
which are generally used for operating-system-specific
services. For example, reserved ports might fall
between 256 and some system-dependent upper limit.
Some systems further choose to protect well-known and
reserved ports by permitting only privileged users to
open TCP connections with those port values. This is
perfectly reasonable as long as the host does not
assume that all hosts protect their low-numbered ports
in this manner.
4.2.2.2 Use of Push: RFC-793 Section 2.8
When an application issues a series of SEND calls without
setting the PUSH flag, the TCP **MAY** aggregate the data
internally without sending it. Similarly, when a series of
segments is received without the PSH bit, a TCP **MAY** queue
the data internally without passing it to the receiving
application.
The PSH bit is not a record marker and is independent of
segment boundaries. The transmitter **SHOULD** collapse
successive PSH bits when it packetizes data, to send the
largest possible segment.
A TCP **MAY** implement PUSH flags on SEND calls. If PUSH flags
are not implemented, then the sending TCP: (1) must not
buffer data indefinitely, and (2) **MUST** set the PSH bit in
the last buffered segment (i.e., when there is no more
queued data to be sent).
The discussion in RFC-793 on pages 48, 50, and 74
erroneously implies that a received PSH flag must be passed
to the application layer. Passing a received PSH flag to
the application layer is now **OPTIONAL**.
An application program is logically required to set the PUSH
flag in a SEND call whenever it needs to force delivery of
the data to avoid a communication deadlock. However, a TCP
**SHOULD** send a maximum-sized segment whenever possible, to
improve performance (see Section 4.2.3.4).
DISCUSSION:
When the PUSH flag is not implemented on SEND calls,
i.e., when the application/TCP interface uses a pure
streaming model, responsibility for aggregating any
tiny data fragments to form reasonable sized segments
is partially borne by the application layer.
Generally, an interactive application protocol must set
the PUSH flag at least in the last SEND call in each
command or response sequence. A bulk transfer protocol
like FTP should set the PUSH flag on the last segment
of a file or when necessary to prevent buffer deadlock.
At the receiver, the PSH bit forces buffered data to be
delivered to the application (even if less than a full
buffer has been received). Conversely, the lack of a
PSH bit can be used to avoid unnecessary wakeup calls
to the application process; this can be an important
performance optimization for large timesharing hosts.
Passing the PSH bit to the receiving application allows
an analogous optimization within the application.
4.2.2.3 Window Size: RFC-793 Section 3.1
The window size **MUST** be treated as an unsigned number, or
else large window sizes will appear like negative windows
and TCP will not work. It is **RECOMMENDED** that
implementations reserve 32-bit fields for the send and
receive window sizes in the connection record and do all
window computations with 32 bits.
DISCUSSION:
It is known that the window field in the TCP header is
too small for high-speed, long-delay paths.
Experimental TCP options have been defined to extend
the window size; see for example [TCP:11]. In
anticipation of the adoption of such an extension, TCP
implementors should treat windows as 32 bits.
4.2.2.4 Urgent Pointer: RFC-793 Section 3.1
The second sentence is in error: the urgent pointer points
to the sequence number of the LAST octet (not LAST+1) in a
sequence of urgent data. The description on page 56 (last
sentence) is correct.
A TCP **MUST** support a sequence of urgent data of any length.
A TCP **MUST** inform the application layer asynchronously
whenever it receives an Urgent pointer and there was
previously no pending urgent data, or whenever the Urgent
pointer advances in the data stream. There **MUST** be a way
for the application to learn how much urgent data remains to
be read from the connection, or at least to determine
whether or not more urgent data remains to be read.
DISCUSSION:
Although the Urgent mechanism may be used for any
application, it is normally used to send "interrupt"-
type commands to a Telnet program (see "Using Telnet
Synch Sequence" section in [INTRO:1]).
The asynchronous or "out-of-band" notification will
allow the application to go into "urgent mode", reading
data from the TCP connection. This allows control
commands to be sent to an application whose normal
input buffers are full of unprocessed data.
IMPLEMENTATION:
The generic ERROR-REPORT() upcall described in Section
4.2.4.1 is a possible mechanism for informing the
application of the arrival of urgent data.
4.2.2.5 TCP Options: RFC-793 Section 3.1
A TCP **MUST** be able to receive a TCP option in any segment.
A TCP **MUST** ignore without error any TCP option it does not
implement, assuming that the option has a length field (all
TCP options defined in the future will have length fields).
TCP **MUST** be prepared to handle an illegal option length
(e.g., zero) without crashing; a suggested procedure is to
reset the connection and log the reason.
4.2.2.6 Maximum Segment Size Option: RFC-793 Section 3.1
TCP **MUST** implement both sending and receiving the Maximum
Segment Size option [TCP:4].
TCP **SHOULD** send an MSS (Maximum Segment Size) option in
every SYN segment when its receive MSS differs from the
default 536, and **MAY** send it always.
If an MSS option is not received at connection setup, TCP
**MUST** assume a default send MSS of 536 (576-40) [TCP:4].
The maximum size of a segment that TCP really sends, the
"effective send MSS," **MUST** be the smaller of the send MSS
(which reflects the available reassembly buffer size at the
remote host) and the largest size permitted by the IP layer:
Eff.snd.MSS =
min(SendMSS+20, MMS_S) - TCPhdrsize - IPoptionsize
where:
* SendMSS is the MSS value received from the remote host,
or the default 536 if no MSS option is received.
* MMS_S is the maximum size for a transport-layer message
that TCP may send.
* TCPhdrsize is the size of the TCP header; this is
normally 20, but may be larger if TCP options are to be
sent.
* IPoptionsize is the size of any IP options that TCP
will pass to the IP layer with the current message.
The MSS value to be sent in an MSS option must be less than
or equal to:
MMS_R - 20
where MMS_R is the maximum size for a transport-layer
message that can be received (and reassembled). TCP obtains
MMS_R and MMS_S from the IP layer; see the generic call
GET_MAXSIZES in Section 3.4.
DISCUSSION:
The choice of TCP segment size has a strong effect on
performance. Larger segments increase throughput by
amortizing header size and per-datagram processing
overhead over more data bytes; however, if the packet
is so large that it causes IP fragmentation, efficiency
drops sharply if any fragments are lost [IP:9].
Some TCP implementations send an MSS option only if the
destination host is on a non-connected network.
However, in general the TCP layer may not have the
appropriate information to make this decision, so it is
preferable to leave to the IP layer the task of
determining a suitable MTU for the Internet path. We
therefore recommend that TCP always send the option (if
not 536) and that the IP layer determine MMS_R as
specified in 3.3.3 and 3.4. A proposed IP-layer
mechanism to measure the MTU would then modify the IP
layer without changing TCP.
4.2.2.7 TCP Checksum: RFC-793 Section 3.1
Unlike the UDP checksum (see Section 4.1.3.4), the TCP
checksum is never optional. The sender **MUST** generate it and
the receiver **MUST** check it.
4.2.2.8 TCP Connection State Diagram: RFC-793 Section 3.2, page 23
There are several problems with this diagram:
(a) The arrow from SYN-SENT to SYN-RCVD should be labeled
with "snd SYN,ACK", to agree with the text on page 68
and with Figure 8.
(b) There could be an arrow from SYN-RCVD state to LISTEN
state, conditioned on receiving a RST after a passive
open (see text page 70).
(c) It is possible to go directly from FIN-WAIT-1 to the
TIME-WAIT state (see page 75 of the spec).
4.2.2.9 Initial Sequence Number Selection: RFC-793 Section 3.3, page 27
A TCP **MUST** use the specified clock-driven selection of
initial sequence numbers.
4.2.2.10 Simultaneous Open Attempts: RFC-793 Section 3.4, page
There is an error in Figure 8: the packet on line 7 should
be identical to the packet on line 5.
A TCP **MUST** support simultaneous open attempts.
DISCUSSION:
It sometimes surprises implementors that if two
applications attempt to simultaneously connect to each
other, only one connection is generated instead of two.
This was an intentional design decision; don't try to
"fix" it.
4.2.2.11 Recovery from Old Duplicate SYN: RFC-793 Section 3.4, page 33
Note that a TCP implementation **MUST** keep track of whether a
connection has reached SYN_RCVD state as the result of a
passive OPEN or an active OPEN.
4.2.2.12 RST Segment: RFC-793 Section 3.4
A TCP **SHOULD** allow a received RST segment to include data.
DISCUSSION
It has been suggested that a RST segment could contain
ASCII text that encoded and explained the cause of the
RST. No standard has yet been established for such
data.
4.2.2.13 Closing a Connection: RFC-793 Section 3.5
A TCP connection may terminate in two ways: (1) the normal
TCP close sequence using a FIN handshake, and (2) an "abort"
in which one or more RST segments are sent and the
connection state is immediately discarded. If a TCP
connection is closed by the remote site, the local
application **MUST** be informed whether it closed normally or
was aborted.
The normal TCP close sequence delivers buffered data
reliably in both directions. Since the two directions of a
TCP connection are closed independently, it is possible for
a connection to be "half closed," i.e., closed in only one
direction, and a host is permitted to continue sending data
in the open direction on a half-closed connection.
A host **MAY** implement a "half-duplex" TCP close sequence, so
that an application that has called CLOSE cannot continue to
read data from the connection. If such a host issues a
CLOSE call while received data is still pending in TCP, or
if new data is received after CLOSE is called, its TCP
**SHOULD** send a RST to show that data was lost.
When a connection is closed actively, it **MUST** linger in
TIME-WAIT state for a time 2xMSL (Maximum Segment Lifetime).
However, it **MAY** accept a new SYN from the remote TCP to
reopen the connection directly from TIME-WAIT state, if it:
(1) assigns its initial sequence number for the new
connection to be larger than the largest sequence
number it used on the previous connection incarnation,
and
(2) returns to TIME-WAIT state if the SYN turns out to be
an old duplicate.
DISCUSSION:
TCP's full-duplex data-preserving close is a feature
that is not included in the analogous ISO transport
protocol TP4.
Some systems have not implemented half-closed
connections, presumably because they do not fit into
the I/O model of their particular operating system. On
these systems, once an application has called CLOSE, it
can no longer read input data from the connection; this
is referred to as a "half-duplex" TCP close sequence.
The graceful close algorithm of TCP requires that the
connection state remain defined on (at least) one end
of the connection, for a timeout period of 2xMSL, i.e.,
4 minutes. During this period, the (remote socket,
local socket) pair that defines the connection is busy
and cannot be reused. To shorten the time that a given
port pair is tied up, some TCPs allow a new SYN to be
accepted in TIME-WAIT state.
4.2.2.14 Data Communication: RFC-793 Section 3.7, page 40
Since RFC-793 was written, there has been extensive work on
TCP algorithms to achieve efficient data communication.
Later sections of the present document describe required and
recommended TCP algorithms to determine when to send data
(Section 4.2.3.4), when to send an acknowledgment (Section
4.2.3.2), and when to update the window (Section 4.2.3.3).
DISCUSSION:
One important performance issue is "Silly Window
Syndrome" or "SWS" [TCP:5], a stable pattern of small
incremental window movements resulting in extremely
poor TCP performance. Algorithms to avoid SWS are
described below for both the sending side (Section
4.2.3.4) and the receiving side (Section 4.2.3.3).
In brief, SWS is caused by the receiver advancing the
right window edge whenever it has any new buffer space
available to receive data and by the sender using any
incremental window, no matter how small, to send more
data [TCP:5]. The result can be a stable pattern of
sending tiny data segments, even though both sender and
receiver have a large total buffer space for the
connection. SWS can only occur during the transmission
of a large amount of data; if the connection goes
quiescent, the problem will disappear. It is caused by
typical straightforward implementation of window
management, but the sender and receiver algorithms
given below will avoid it.
Another important TCP performance issue is that some
applications, especially remote login to character-at-
a-time hosts, tend to send streams of one-octet data
segments. To avoid deadlocks, every TCP SEND call from
such applications must be "pushed", either explicitly
by the application or else implicitly by TCP. The
result may be a stream of TCP segments that contain one
data octet each, which makes very inefficient use of
the Internet and contributes to Internet congestion.
The Nagle Algorithm described in Section 4.2.3.4
provides a simple and effective solution to this
problem. It does have the effect of clumping
characters over Telnet connections; this may initially
surprise users accustomed to single-character echo, but
user acceptance has not been a problem.
Note that the Nagle algorithm and the send SWS
avoidance algorithm play complementary roles in
improving performance. The Nagle algorithm discourages
sending tiny segments when the data to be sent
increases in small increments, while the SWS avoidance
algorithm discourages small segments resulting from the
right window edge advancing in small increments.
A careless implementation can send two or more
acknowledgment segments per data segment received. For
example, suppose the receiver acknowledges every data
segment immediately. When the application program
subsequently consumes the data and increases the
available receive buffer space again, the receiver may
send a second acknowledgment segment to update the
window at the sender. The extreme case occurs with
single-character segments on TCP connections using the
Telnet protocol for remote login service. Some
implementations have been observed in which each
incoming 1-character segment generates three return
segments: (1) the acknowledgment, (2) a one byte
increase in the window, and (3) the echoed character,
respectively.
4.2.2.15 Retransmission Timeout: RFC-793 Section 3.7, page 41
The algorithm suggested in RFC-793 for calculating the
retransmission timeout is now known to be inadequate; see
Section 4.2.3.1 below.
Recent work by Jacobson [TCP:7] on Internet congestion and
TCP retransmission stability has produced a transmission
algorithm combining "slow start" with "congestion
avoidance". A TCP **MUST** implement this algorithm.
If a retransmitted packet is identical to the original
packet (which implies not only that the data boundaries have
not changed, but also that the window and acknowledgment
fields of the header have not changed), then the same IP
Identification field **MAY** be used (see Section 3.2.1.5).
IMPLEMENTATION:
Some TCP implementors have chosen to "packetize" the
data stream, i.e., to pick segment boundaries when
segments are originally sent and to queue these
segments in a "retransmission queue" until they are
acknowledged. Another design (which may be simpler) is
to defer packetizing until each time data is
transmitted or retransmitted, so there will be no
segment retransmission queue.
In an implementation with a segment retransmission
queue, TCP performance may be enhanced by repacketizing
the segments awaiting acknowledgment when the first
retransmission timeout occurs. That is, the
outstanding segments that fitted would be combined into
one maximum-sized segment, with a new IP Identification
value. The TCP would then retain this combined segment
in the retransmit queue until it was acknowledged.
However, if the first two segments in the
retransmission queue totalled more than one maximum-
sized segment, the TCP would retransmit only the first
segment using the original IP Identification field.
4.2.2.16 Managing the Window: RFC-793 Section 3.7, page 41
A TCP receiver **SHOULD** NOT shrink the window, i.e., move the
right window edge to the left. However, a sending TCP **MUST**
be robust against window shrinking, which may cause the
"useable window" (see Section 4.2.3.4) to become negative.
If this happens, the sender **SHOULD** NOT send new data, but
**SHOULD** retransmit normally the old unacknowledged data
between SND.UNA and SND.UNA+SND.WND. The sender **MAY** also
retransmit old data beyond SND.UNA+SND.WND, but **SHOULD** NOT
time out the connection if data beyond the right window edge
is not acknowledged. If the window shrinks to zero, the TCP
**MUST** probe it in the standard way (see next Section).
DISCUSSION:
Many TCP implementations become confused if the window
shrinks from the right after data has been sent into a
larger window. Note that TCP has a heuristic to select
the latest window update despite possible datagram
reordering; as a result, it may ignore a window update
with a smaller window than previously offered if
neither the sequence number nor the acknowledgment
number is increased.
4.2.2.17 Probing Zero Windows: RFC-793 Section 3.7, page 42
Probing of zero (offered) windows **MUST** be supported.
A TCP **MAY** keep its offered receive window closed
indefinitely. As long as the receiving TCP continues to
send acknowledgments in response to the probe segments, the
sending TCP **MUST** allow the connection to stay open.
DISCUSSION:
It is extremely important to remember that ACK
(acknowledgment) segments that contain no data are not
reliably transmitted by TCP. If zero window probing is
not supported, a connection may hang forever when an
ACK segment that re-opens the window is lost.
The delay in opening a zero window generally occurs
when the receiving application stops taking data from
its TCP. For example, consider a printer daemon
application, stopped because the printer ran out of
paper.
The transmitting host **SHOULD** send the first zero-window
probe when a zero window has existed for the retransmission
timeout period (see Section 4.2.2.15), and **SHOULD** increase
exponentially the interval between successive probes.
DISCUSSION:
This procedure minimizes delay if the zero-window
condition is due to a lost ACK segment containing a
window-opening update. Exponential backoff is
recommended, possibly with some maximum interval not
specified here. This procedure is similar to that of
the retransmission algorithm, and it may be possible to
combine the two procedures in the implementation.
4.2.2.18 Passive OPEN Calls: RFC-793 Section 3.8
Every passive OPEN call either creates a new connection
record in LISTEN state, or it returns an error; it **MUST** NOT
affect any previously created connection record.
A TCP that supports multiple concurrent users **MUST** provide
an OPEN call that will functionally allow an application to
LISTEN on a port while a connection block with the same
local port is in SYN-SENT or SYN-RECEIVED state.
DISCUSSION:
Some applications (e.g., SMTP servers) may need to
handle multiple connection attempts at about the same
time. The probability of a connection attempt failing
is reduced by giving the application some means of
listening for a new connection at the same time that an
earlier connection attempt is going through the three-
way handshake.
IMPLEMENTATION:
Acceptable implementations of concurrent opens may
permit multiple passive OPEN calls, or they may allow
"cloning" of LISTEN-state connections from a single
passive OPEN call.
4.2.2.19 Time to Live: RFC-793 Section 3.9, page 52
RFC-793 specified that TCP was to request the IP layer to
send TCP segments with TTL = 60. This is obsolete; the TTL
value used to send TCP segments **MUST** be configurable. See
Section 3.2.1.7 for discussion.
4.2.2.20 Event Processing: RFC-793 Section 3.9
While it is not strictly required, a TCP **SHOULD** be capable
of queueing out-of-order TCP segments. Change the "may" in
the last sentence of the first paragraph on page 70 to
"should".
DISCUSSION:
Some small-host implementations have omitted segment
queueing because of limited buffer space. This
omission may be expected to adversely affect TCP
throughput, since loss of a single segment causes all
later segments to appear to be "out of sequence".
In general, the processing of received segments **MUST** be
implemented to aggregate ACK segments whenever possible.
For example, if the TCP is processing a series of queued
segments, it **MUST** process them all before sending any ACK
segments.
Here are some detailed error corrections and notes on the
Event Processing section of RFC-793.
(a) CLOSE Call, CLOSE-WAIT state, p. 61: enter LAST-ACK
state, not CLOSING.
(b) LISTEN state, check for SYN (pp. 65, 66): With a SYN
bit, if the security/compartment or the precedence is
wrong for the segment, a reset is sent. The wrong form
of reset is shown in the text; it should be:
`<SEQ=0><ACK=SEG.SEQ+SEG.LEN><CTL=RST,ACK>`
(c) SYN-SENT state, Check for SYN, p. 68: When the
connection enters ESTABLISHED state, the following
variables must be set:
`SND.WND <- SEG.WND`
`SND.WL1 <- SEG.SEQ`
`SND.WL2 <- SEG.ACK`
(d) Check security and precedence, p. 71: The first heading
"ESTABLISHED STATE" should really be a list of all
states other than SYN-RECEIVED: ESTABLISHED, FIN-WAIT-
1, FIN-WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, and
TIME-WAIT.
(e) Check SYN bit, p. 71: "In SYN-RECEIVED state and if
the connection was initiated with a passive OPEN, then
return this connection to the LISTEN state and return.
Otherwise...".
(f) Check ACK field, SYN-RECEIVED state, p. 72: When the
connection enters ESTABLISHED state, the variables
listed in (c) must be set.
(g) Check ACK field, ESTABLISHED state, p. 72: The ACK is a
duplicate if SEG.ACK =< SND.UNA (the = was omitted).
Similarly, the window should be updated if: SND.UNA =<
SEG.ACK =< SND.NXT.
(h) USER TIMEOUT, p. 77:
It would be better to notify the application of the
timeout rather than letting TCP force the connection
closed. However, see also Section 4.2.3.5.
4.2.2.21 Acknowledging Queued Segments: RFC-793 Section 3.9
A TCP **MAY** send an ACK segment acknowledging RCV.NXT when a
valid segment arrives that is in the window but not at the
left window edge.
DISCUSSION:
RFC-793 (see page 74) was ambiguous about whether or
not an ACK segment should be sent when an out-of-order
segment was received, i.e., when SEG.SEQ was unequal to
RCV.NXT.
One reason for ACKing out-of-order segments might be to
support an experimental algorithm known as "fast
retransmit". With this algorithm, the sender uses the
"redundant" ACK's to deduce that a segment has been
lost before the retransmission timer has expired. It
counts the number of times an ACK has been received
with the same value of SEG.ACK and with the same right
window edge. If more than a threshold number of such
ACK's is received, then the segment containing the
octets starting at SEG.ACK is assumed to have been lost
and is retransmitted, without awaiting a timeout. The
threshold is chosen to compensate for the maximum
likely segment reordering in the Internet. There is
not yet enough experience with the fast retransmit
algorithm to determine how useful it is.
4.2.3 SPECIFIC ISSUES
4.2.3.1 Retransmission Timeout Calculation
A host TCP **MUST** implement Karn's algorithm and Jacobson's
algorithm for computing the retransmission timeout ("RTO").
o Jacobson's algorithm for computing the smoothed round-
trip ("RTT") time incorporates a simple measure of the
variance [TCP:7].
o Karn's algorithm for selecting RTT measurements ensures
that ambiguous round-trip times will not corrupt the
calculation of the smoothed round-trip time [TCP:6].
This implementation also **MUST** include "exponential backoff"
for successive RTO values for the same segment.
Retransmission of SYN segments **SHOULD** use the same algorithm
as data segments.
DISCUSSION:
There were two known problems with the RTO calculations
specified in RFC-793. First, the accurate measurement
of RTTs is difficult when there are retransmissions.
Second, the algorithm to compute the smoothed round-
trip time is inadequate [TCP:7], because it incorrectly
assumed that the variance in RTT values would be small
and constant. These problems were solved by Karn's and
Jacobson's algorithm, respectively.
The performance increase resulting from the use of
these improvements varies from noticeable to dramatic.
Jacobson's algorithm for incorporating the measured RTT
variance is especially important on a low-speed link,
where the natural variation of packet sizes causes a
large variation in RTT. One vendor found link
utilization on a 9.6kb line went from 10% to 90% as a
result of implementing Jacobson's variance algorithm in
TCP.
The following values **SHOULD** be used to initialize the
estimation parameters for a new connection:
(a) RTT = 0 seconds.
(b) RTO = 3 seconds. (The smoothed variance is to be
initialized to the value that will result in this RTO).
The recommended upper and lower bounds on the RTO are known
to be inadequate on large internets. The lower bound **SHOULD**
be measured in fractions of a second (to accommodate high
speed LANs) and the upper bound should be 2*MSL, i.e., 240
seconds.
DISCUSSION:
Experience has shown that these initialization values
are reasonable, and that in any case the Karn and
Jacobson algorithms make TCP behavior reasonably
insensitive to the initial parameter choices.
4.2.3.2 When to Send an ACK Segment
A host that is receiving a stream of TCP data segments can
increase efficiency in both the Internet and the hosts by
sending fewer than one ACK (acknowledgment) segment per data
segment received; this is known as a "delayed ACK" [TCP:5].
A TCP **SHOULD** implement a delayed ACK, but an ACK should not
be excessively delayed; in particular, the delay **MUST** be
less than 0.5 seconds, and in a stream of full-sized
segments there **SHOULD** be an ACK for at least every second
segment.
DISCUSSION:
A delayed ACK gives the application an opportunity to
update the window and perhaps to send an immediate
response. In particular, in the case of character-mode
remote login, a delayed ACK can reduce the number of
segments sent by the server by a factor of 3 (ACK,
window update, and echo character all combined in one
segment).
In addition, on some large multi-user hosts, a delayed
ACK can substantially reduce protocol processing
overhead by reducing the total number of packets to be
processed [TCP:5]. However, excessive delays on ACK's
can disturb the round-trip timing and packet "clocking"
algorithms [TCP:7].
4.2.3.3 When to Send a Window Update
A TCP **MUST** include a SWS avoidance algorithm in the receiver
[TCP:5].
IMPLEMENTATION:
The receiver's SWS avoidance algorithm determines when
the right window edge may be advanced; this is
customarily known as "updating the window". This
algorithm combines with the delayed ACK algorithm (see
Section 4.2.3.2) to determine when an ACK segment
containing the current window will really be sent to
the receiver. We use the notation of RFC-793; see
Figures 4 and 5 in that document.
The solution to receiver SWS is to avoid advancing the
right window edge RCV.NXT+RCV.WND in small increments,
even if data is received from the network in small
segments.
Suppose the total receive buffer space is RCV.BUFF. At
any given moment, RCV.USER octets of this total may be
tied up with data that has been received and
acknowledged but which the user process has not yet
consumed. When the connection is quiescent, RCV.WND =
RCV.BUFF and RCV.USER = 0.
Keeping the right window edge fixed as data arrives and
is acknowledged requires that the receiver offer less
than its full buffer space, i.e., the receiver must
specify a RCV.WND that keeps RCV.NXT+RCV.WND constant
as RCV.NXT increases. Thus, the total buffer space
RCV.BUFF is generally divided into three parts:
|<------- RCV.BUFF ---------------->|
1 2 3
----|---------|------------------|------|----
RCV.NXT ^
(Fixed)
1 - RCV.USER = data received but not yet consumed;
2 - RCV.WND = space advertised to sender;
3 - Reduction = space available but not yet
advertised.
The suggested SWS avoidance algorithm for the receiver
is to keep RCV.NXT+RCV.WND fixed until the reduction
satisfies:
RCV.BUFF - RCV.USER - RCV.WND >=
min( Fr * RCV.BUFF, Eff.snd.MSS )
where Fr is a fraction whose recommended value is 1/2,
and Eff.snd.MSS is the effective send MSS for the
connection (see Section 4.2.2.6). When the inequality
is satisfied, RCV.WND is set to RCV.BUFF-RCV.USER.
Note that the general effect of this algorithm is to
advance RCV.WND in increments of Eff.snd.MSS (for
realistic receive buffers: Eff.snd.MSS < RCV.BUFF/2).
Note also that the receiver must use its own
Eff.snd.MSS, assuming it is the same as the sender's.
4.2.3.4 When to Send Data
A TCP **MUST** include a SWS avoidance algorithm in the sender.
A TCP **SHOULD** implement the Nagle Algorithm [TCP:9] to
coalesce short segments. However, there **MUST** be a way for
an application to disable the Nagle algorithm on an
individual connection. In all cases, sending data is also
subject to the limitation imposed by the Slow Start
algorithm (Section 4.2.2.15).
DISCUSSION:
The Nagle algorithm is generally as follows:
If there is unacknowledged data (i.e., SND.NXT >
SND.UNA), then the sending TCP buffers all user
data (regardless of the PSH bit), until the
outstanding data has been acknowledged or until
the TCP can send a full-sized segment (Eff.snd.MSS
bytes; see Section 4.2.2.6).
Some applications (e.g., real-time display window
updates) require that the Nagle algorithm be turned
off, so small data segments can be streamed out at the
maximum rate.
IMPLEMENTATION:
The sender's SWS avoidance algorithm is more difficult
than the receivers's, because the sender does not know
(directly) the receiver's total buffer space RCV.BUFF.
An approach which has been found to work well is for
the sender to calculate Max(SND.WND), the maximum send
window it has seen so far on the connection, and to use
this value as an estimate of RCV.BUFF. Unfortunately,
this can only be an estimate; the receiver may at any
time reduce the size of RCV.BUFF. To avoid a resulting
deadlock, it is necessary to have a timeout to force
transmission of data, overriding the SWS avoidance
algorithm. In practice, this timeout should seldom
occur.
The "useable window" [TCP:5] is:
U = SND.UNA + SND.WND - SND.NXT
i.e., the offered window less the amount of data sent
but not acknowledged. If D is the amount of data
queued in the sending TCP but not yet sent, then the
following set of rules is recommended.
Send data:
(1) if a maximum-sized segment can be sent, i.e, if:
min(D,U) >= Eff.snd.MSS;
(2) or if the data is pushed and all queued data can
be sent now, i.e., if:
`[SND.NXT = SND.UNA and] PUSHED and D <= U`
(the bracketed condition is imposed by the Nagle
algorithm);
(3) or if at least a fraction Fs of the maximum window
can be sent, i.e., if:
[SND.NXT = SND.UNA and]
min(D.U) >= Fs * Max(SND.WND);
(4) or if data is PUSHed and the override timeout
occurs.
Here Fs is a fraction whose recommended value is 1/2.
The override timeout should be in the range 0.1 - 1.0
seconds. It may be convenient to combine this timer
with the timer used to probe zero windows (Section
4.2.2.17).
Finally, note that the SWS avoidance algorithm just
specified is to be used instead of the sender-side
algorithm contained in [TCP:5].
4.2.3.5 TCP Connection Failures
Excessive retransmission of the same segment by TCP
indicates some failure of the remote host or the Internet
path. This failure may be of short or long duration. The
following procedure **MUST** be used to handle excessive
retransmissions of data segments [IP:11]:
(a) There are two thresholds R1 and R2 measuring the amount
of retransmission that has occurred for the same
segment. R1 and R2 might be measured in time units or
as a count of retransmissions.
(b) When the number of transmissions of the same segment
reaches or exceeds threshold R1, pass negative advice
(see Section 3.3.1.4) to the IP layer, to trigger
dead-gateway diagnosis.
(c) When the number of transmissions of the same segment
reaches a threshold R2 greater than R1, close the
connection.
(d) An application **MUST** be able to set the value for R2 for
a particular connection. For example, an interactive
application might set R2 to "infinity," giving the user
control over when to disconnect.
(d) TCP **SHOULD** inform the application of the delivery
problem (unless such information has been disabled by
the application; see Section 4.2.4.1), when R1 is
reached and before R2. This will allow a remote login
(User Telnet) application program to inform the user,
for example.
The value of R1 **SHOULD** correspond to at least 3
retransmissions, at the current RTO. The value of R2 **SHOULD**
correspond to at least 100 seconds.
An attempt to open a TCP connection could fail with
excessive retransmissions of the SYN segment or by receipt
of a RST segment or an ICMP Port Unreachable. SYN
retransmissions **MUST** be handled in the general way just
described for data retransmissions, including notification
of the application layer.
However, the values of R1 and R2 may be different for SYN
and data segments. In particular, R2 for a SYN segment **MUST**
be set large enough to provide retransmission of the segment
for at least 3 minutes. The application can close the
connection (i.e., give up on the open attempt) sooner, of
course.
DISCUSSION:
Some Internet paths have significant setup times, and
the number of such paths is likely to increase in the
future.
4.2.3.6 TCP Keep-Alives
Implementors **MAY** include "keep-alives" in their TCP
implementations, although this practice is not universally
accepted. If keep-alives are included, the application **MUST**
be able to turn them on or off for each TCP connection, and
they **MUST** default to off.
Keep-alive packets **MUST** only be sent when no data or
acknowledgement packets have been received for the
connection within an interval. This interval **MUST** be
configurable and **MUST** default to no less than two hours.
It is extremely important to remember that ACK segments that
contain no data are not reliably transmitted by TCP.
Consequently, if a keep-alive mechanism is implemented it
**MUST** NOT interpret failure to respond to any specific probe
as a dead connection.
An implementation **SHOULD** send a keep-alive segment with no
data; however, it **MAY** be configurable to send a keep-alive
segment containing one garbage octet, for compatibility with
erroneous TCP implementations.
DISCUSSION:
A "keep-alive" mechanism periodically probes the other
end of a connection when the connection is otherwise
idle, even when there is no data to be sent. The TCP
specification does not include a keep-alive mechanism
because it could: (1) cause perfectly good connections
to break during transient Internet failures; (2)
consume unnecessary bandwidth ("if no one is using the
connection, who cares if it is still good?"); and (3)
cost money for an Internet path that charges for
packets.
Some TCP implementations, however, have included a
keep-alive mechanism. To confirm that an idle
connection is still active, these implementations send
a probe segment designed to elicit a response from the
peer TCP. Such a segment generally contains SEG.SEQ =
SND.NXT-1 and may or may not contain one garbage octet
of data. Note that on a quiet connection SND.NXT =
RCV.NXT, so that this SEG.SEQ will be outside the
window. Therefore, the probe causes the receiver to
return an acknowledgment segment, confirming that the
connection is still live. If the peer has dropped the
connection due to a network partition or a crash, it
will respond with a RST instead of an acknowledgment
segment.
Unfortunately, some misbehaved TCP implementations fail
to respond to a segment with SEG.SEQ = SND.NXT-1 unless
the segment contains data. Alternatively, an
implementation could determine whether a peer responded
correctly to keep-alive packets with no garbage data
octet.
A TCP keep-alive mechanism should only be invoked in
server applications that might otherwise hang
indefinitely and consume resources unnecessarily if a
client crashes or aborts a connection during a network
failure.
4.2.3.7 TCP Multihoming
If an application on a multihomed host does not specify the
local IP address when actively opening a TCP connection,
then the TCP **MUST** ask the IP layer to select a local IP
address before sending the (first) SYN. See the function
GET_SRCADDR() in Section 3.4.
At all other times, a previous segment has either been sent
or received on this connection, and TCP **MUST** use the same
local address is used that was used in those previous
segments.
4.2.3.8 IP Options
When received options are passed up to TCP from the IP
layer, TCP **MUST** ignore options that it does not understand.
A TCP **MAY** support the Time Stamp and Record Route options.
An application **MUST** be able to specify a source route when
it actively opens a TCP connection, and this **MUST** take
precedence over a source route received in a datagram.
When a TCP connection is OPENed passively and a packet
arrives with a completed IP Source Route option (containing
a return route), TCP **MUST** save the return route and use it
for all segments sent on this connection. If a different
source route arrives in a later segment, the later
definition **SHOULD** override the earlier one.
4.2.3.9 ICMP Messages
TCP **MUST** act on an ICMP error message passed up from the IP
layer, directing it to the connection that created the
error. The necessary demultiplexing information can be
found in the IP header contained within the ICMP message.
o Source Quench
TCP **MUST** react to a Source Quench by slowing
transmission on the connection. The **RECOMMENDED**
procedure is for a Source Quench to trigger a "slow
start," as if a retransmission timeout had occurred.
o Destination Unreachable -- codes 0, 1, 5
Since these Unreachable messages indicate soft error
conditions, TCP **MUST** NOT abort the connection, and it
**SHOULD** make the information available to the
application.
DISCUSSION:
TCP could report the soft error condition directly
to the application layer with an upcall to the
ERROR_REPORT routine, or it could merely note the
message and report it to the application only when
and if the TCP connection times out.
o Destination Unreachable -- codes 2-4
These are hard error conditions, so TCP **SHOULD** abort
the connection.
o Time Exceeded -- codes 0, 1
This should be handled the same way as Destination
Unreachable codes 0, 1, 5 (see above).
o Parameter Problem
This should be handled the same way as Destination
Unreachable codes 0, 1, 5 (see above).
4.2.3.10 Remote Address Validation
A TCP implementation **MUST** reject as an error a local OPEN
call for an invalid remote IP address (e.g., a broadcast or
multicast address).
An incoming SYN with an invalid source address must be
ignored either by TCP or by the IP layer (see Section
3.2.1.3).
A TCP implementation **MUST** silently discard an incoming SYN
segment that is addressed to a broadcast or multicast
address.
4.2.3.11 TCP Traffic Patterns
IMPLEMENTATION:
The TCP protocol specification [TCP:1] gives the
implementor much freedom in designing the algorithms
that control the message flow over the connection --
packetizing, managing the window, sending
acknowledgments, etc. These design decisions are
difficult because a TCP must adapt to a wide range of
traffic patterns. Experience has shown that a TCP
implementor needs to verify the design on two extreme
traffic patterns:
o Single-character Segments
Even if the sender is using the Nagle Algorithm,
when a TCP connection carries remote login traffic
across a low-delay LAN the receiver will generally
get a stream of single-character segments. If
remote terminal echo mode is in effect, the
receiver's system will generally echo each
character as it is received.
o Bulk Transfer
When TCP is used for bulk transfer, the data
stream should be made up (almost) entirely of
segments of the size of the effective MSS.
Although TCP uses a sequence number space with
byte (octet) granularity, in bulk-transfer mode
its operation should be as if TCP used a sequence
space that counted only segments.
Experience has furthermore shown that a single TCP can
effectively and efficiently handle these two extremes.
The most important tool for verifying a new TCP
implementation is a packet trace program. There is a
large volume of experience showing the importance of
tracing a variety of traffic patterns with other TCP
implementations and studying the results carefully.
4.2.3.12 Efficiency
IMPLEMENTATION:
Extensive experience has led to the following
suggestions for efficient implementation of TCP:
(a) Don't Copy Data
In bulk data transfer, the primary CPU-intensive
tasks are copying data from one place to another
and checksumming the data. It is vital to
minimize the number of copies of TCP data. Since
the ultimate speed limitation may be fetching data
across the memory bus, it may be useful to combine
the copy with checksumming, doing both with a
single memory fetch.
(b) Hand-Craft the Checksum Routine
A good TCP checksumming routine is typically two
to five times faster than a simple and direct
implementation of the definition. Great care and
clever coding are often required and advisable to
make the checksumming code "blazing fast". See
[TCP:10].
(c) Code for the Common Case
TCP protocol processing can be complicated, but
for most segments there are only a few simple
decisions to be made. Per-segment processing will
be greatly speeded up by coding the main line to
minimize the number of decisions in the most
common case.
4.2.4 TCP/APPLICATION LAYER INTERFACE
4.2.4.1 Asynchronous Reports
There **MUST** be a mechanism for reporting soft TCP error
conditions to the application. Generically, we assume this
takes the form of an application-supplied ERROR_REPORT
routine that may be upcalled [INTRO:7] asynchronously from
the transport layer:
ERROR_REPORT(local connection name, reason, subreason)
The precise encoding of the reason and subreason parameters
is not specified here. However, the conditions that are
reported asynchronously to the application **MUST** include:
* ICMP error message arrived (see 4.2.3.9)
* Excessive retransmissions (see 4.2.3.5)
* Urgent pointer advance (see 4.2.2.4).
However, an application program that does not want to
receive such ERROR_REPORT calls **SHOULD** be able to
effectively disable these calls.
DISCUSSION:
These error reports generally reflect soft errors that
can be ignored without harm by many applications. It
has been suggested that these error report calls should
default to "disabled," but this is not required.
4.2.4.2 Type-of-Service
The application layer **MUST** be able to specify the Type-of-
Service (TOS) for segments that are sent on a connection.
It not required, but the application **SHOULD** be able to
change the TOS during the connection lifetime. TCP **SHOULD**
pass the current TOS value without change to the IP layer,
when it sends segments on the connection.
The TOS will be specified independently in each direction on
the connection, so that the receiver application will
specify the TOS used for ACK segments.
TCP **MAY** pass the most recently received TOS up to the
application.
DISCUSSION
Some applications (e.g., SMTP) change the nature of
their communication during the lifetime of a
connection, and therefore would like to change the TOS
specification.
Note also that the OPEN call specified in RFC-793
includes a parameter ("options") in which the caller
can specify IP options such as source route, record
route, or timestamp.
4.2.4.3 Flush Call
Some TCP implementations have included a FLUSH call, which
will empty the TCP send queue of any data for which the user
has issued SEND calls but which is still to the right of the
current send window. That is, it flushes as much queued
send data as possible without losing sequence number
synchronization. This is useful for implementing the "abort
output" function of Telnet.
4.2.4.4 Multihoming
The user interface outlined in sections 2.7 and 3.8 of RFC-
793 needs to be extended for multihoming. The OPEN call
**MUST** have an optional parameter:
OPEN( ... [local IP address,] ... )
to allow the specification of the local IP address.
DISCUSSION:
Some TCP-based applications need to specify the local
IP address to be used to open a particular connection;
FTP is an example.
IMPLEMENTATION:
A passive OPEN call with a specified "local IP address"
parameter will await an incoming connection request to
that address. If the parameter is unspecified, a
passive OPEN will await an incoming connection request
to any local IP address, and then bind the local IP
address of the connection to the particular address
that is used.
For an active OPEN call, a specified "local IP address"
parameter will be used for opening the connection. If
the parameter is unspecified, the networking software
will choose an appropriate local IP address (see
Section 3.3.4.2) for the connection
4.2.5 TCP REQUIREMENT SUMMARY
| | | | |S| |
| | | | |H| |F
| | | | |O|M|o
| | |S| |U|U|o
| | |H| |L|S|t
| |M|O| |D|T|n
| |U|U|M| | |o
| |S|L|A|N|N|t
| |T|D|Y|O|O|t
| FEATURE | SECTION | T | T | e |
|---|
| | | | | | |
Push flag | | | | | | | Aggregate or queue un-pushed data |4.2.2.2 | | |x| | | Sender collapse successive PSH flags |4.2.2.2 | |x| | | | SEND call can specify PUSH |4.2.2.2 | | |x| | |
If cannot: sender buffer indefinitely |4.2.2.2 | | | | |x|
If cannot: PSH last segment |4.2.2.2 |x| | | | |
Notify receiving ALP of PSH |4.2.2.2 | | |x| | |1 Send max size segment when possible |4.2.2.2 | |x| | | | | | | | | | | Window | | | | | | | Treat as unsigned number |4.2.2.3 |x| | | | | Handle as 32-bit number |4.2.2.3 | |x| | | | Shrink window from right |4.2.2.16| | | |x| | Robust against shrinking window |4.2.2.16|x| | | | | Receiver's window closed indefinitely |4.2.2.17| | |x| | | Sender probe zero window |4.2.2.17|x| | | | | First probe after RTO |4.2.2.17| |x| | | | Exponential backoff |4.2.2.17| |x| | | | Allow window stay zero indefinitely |4.2.2.17|x| | | | | Sender timeout OK conn with zero wind |4.2.2.17| | | | |x| | | | | | | | Urgent Data | | | | | | | Pointer points to last octet |4.2.2.4 |x| | | | | Arbitrary length urgent data sequence |4.2.2.4 |x| | | | | Inform ALP asynchronously of urgent data |4.2.2.4 |x| | | | |1 ALP can learn if/how much urgent data Q'd |4.2.2.4 |x| | | | |1 | | | | | | | TCP Options | | | | | | | Receive TCP option in any segment |4.2.2.5 |x| | | | | Ignore unsupported options |4.2.2.5 |x| | | | | Cope with illegal option length |4.2.2.5 |x| | | | | Implement sending & receiving MSS option |4.2.2.6 |x| | | | | Send MSS option unless 536 |4.2.2.6 | |x| | | | Send MSS option always |4.2.2.6 | | |x| | | Send-MSS default is 536 |4.2.2.6 |x| | | | | Calculate effective send seg size |4.2.2.6 |x| | | | | | | | | | | | TCP Checksums | | | | | | | Sender compute checksum |4.2.2.7 |x| | | | | Receiver check checksum |4.2.2.7 |x| | | | | | | | | | | | Use clock-driven ISN selection |4.2.2.9 |x| | | | | | | | | | | | Opening Connections | | | | | | | Support simultaneous open attempts |4.2.2.10|x| | | | | SYN-RCVD remembers last state |4.2.2.11|x| | | | | Passive Open call interfere with others |4.2.2.18| | | | |x| Function: simultan. LISTENs for same port |4.2.2.18|x| | | | | Ask IP for src address for SYN if necc. |4.2.3.7 |x| | | | | Otherwise, use local addr of conn. |4.2.3.7 |x| | | | | OPEN to broadcast/multicast IP Address |4.2.3.14| | | | |x| Silently discard seg to bcast/mcast addr |4.2.3.14|x| | | | |
| | | | | | |
Closing Connections | | | | | | |
RST can contain data |4.2.2.12| |x| | | |
Inform application of aborted conn |4.2.2.13|x| | | | |
Half-duplex close connections |4.2.2.13| | |x| | |
Send RST to indicate data lost |4.2.2.13| |x| | | |
In TIME-WAIT state for 2xMSL seconds |4.2.2.13|x| | | | |
Accept SYN from TIME-WAIT state |4.2.2.13| | |x| | |
| | | | | | |
Retransmissions | | | | | | |
Jacobson Slow Start algorithm |4.2.2.15|x| | | | |
Jacobson Congestion-Avoidance algorithm |4.2.2.15|x| | | | |
Retransmit with same IP ident |4.2.2.15| | |x| | |
Karn's algorithm |4.2.3.1 |x| | | | |
Jacobson's RTO estimation alg. |4.2.3.1 |x| | | | |
Exponential backoff |4.2.3.1 |x| | | | |
SYN RTO calc same as data |4.2.3.1 | |x| | | |
Recommended initial values and bounds |4.2.3.1 | |x| | | |
| | | | | | |
Generating ACK's: | | | | | | |
Queue out-of-order segments |4.2.2.20| |x| | | |
Process all Q'd before send ACK |4.2.2.20|x| | | | |
Send ACK for out-of-order segment |4.2.2.21| | |x| | |
Delayed ACK's |4.2.3.2 | |x| | | |
Delay < 0.5 seconds |4.2.3.2 |x| | | | |
Every 2nd full-sized segment ACK'd |4.2.3.2 |x| | | | |
Receiver SWS-Avoidance Algorithm |4.2.3.3 |x| | | | |
| | | | | | |
Sending data | | | | | | |
Configurable TTL |4.2.2.19|x| | | | |
Sender SWS-Avoidance Algorithm |4.2.3.4 |x| | | | |
Nagle algorithm |4.2.3.4 | |x| | | |
Application can disable Nagle algorithm |4.2.3.4 |x| | | | |
| | | | | | |
Connection Failures: | | | | | | |
Negative advice to IP on R1 retxs |4.2.3.5 |x| | | | |
Close connection on R2 retxs |4.2.3.5 |x| | | | |
ALP can set R2 |4.2.3.5 |x| | | | |1
Inform ALP of R1<=retxs<R2 |4.2.3.5 | |x| | | |1
Recommended values for R1, R2 |4.2.3.5 | |x| | | |
Same mechanism for SYNs |4.2.3.5 |x| | | | |
R2 at least 3 minutes for SYN |4.2.3.5 |x| | | | |
| | | | | | |
Send Keep-alive Packets: |4.2.3.6 | | |x| | |
-
Application can request |4.2.3.6 |x| | | | |
-
Default is "off" |4.2.3.6 |x| | | | |
-
Only send if idle for interval |4.2.3.6 |x| | | | |
-
Interval configurable |4.2.3.6 |x| | | | |
-
Default at least 2 hrs. |4.2.3.6 |x| | | | |
-
Tolerant of lost ACK's |4.2.3.6 |x| | | | | | | | | | | | IP Options | | | | | | | Ignore options TCP doesn't understand |4.2.3.8 |x| | | | | Time Stamp support |4.2.3.8 | | |x| | | Record Route support |4.2.3.8 | | |x| | | Source Route: | | | | | | | ALP can specify |4.2.3.8 |x| | | | |1 Overrides src rt in datagram |4.2.3.8 |x| | | | | Build return route from src rt |4.2.3.8 |x| | | | | Later src route overrides |4.2.3.8 | |x| | | | | | | | | | | Receiving ICMP Messages from IP |4.2.3.9 |x| | | | | Dest. Unreach (0,1,5) => inform ALP |4.2.3.9 | |x| | | | Dest. Unreach (0,1,5) => abort conn |4.2.3.9 | | | | |x| Dest. Unreach (2-4) => abort conn |4.2.3.9 | |x| | | | Source Quench => slow start |4.2.3.9 | |x| | | | Time Exceeded => tell ALP, don't abort |4.2.3.9 | |x| | | | Param Problem => tell ALP, don't abort |4.2.3.9 | |x| | | | | | | | | | | Address Validation | | | | | | | Reject OPEN call to invalid IP address |4.2.3.10|x| | | | | Reject SYN from invalid IP address |4.2.3.10|x| | | | | Silently discard SYN to bcast/mcast addr |4.2.3.10|x| | | | | | | | | | | | TCP/ALP Interface Services | | | | | | | Error Report mechanism |4.2.4.1 |x| | | | | ALP can disable Error Report Routine |4.2.4.1 | |x| | | | ALP can specify TOS for sending |4.2.4.2 |x| | | | | Passed unchanged to IP |4.2.4.2 | |x| | | | ALP can change TOS during connection |4.2.4.2 | |x| | | | Pass received TOS up to ALP |4.2.4.2 | | |x| | | FLUSH call |4.2.4.3 | | |x| | | Optional local IP addr parm. in OPEN |4.2.4.4 |x| | | | | -------------------------------------------------|--------|-|-|-|-|-|-- -------------------------------------------------|--------|-|-|-|-|-|--
FOOTNOTES:
(1) "ALP" means Application-Layer program.