linux-stable/Documentation/networking/dccp.txt

208 lines
9.4 KiB
Text
Raw Normal View History

DCCP protocol
=============
Contents
========
- Introduction
- Missing features
- Socket options
- Sysctl variables
- IOCTLs
- Other tunables
- Notes
Introduction
============
Datagram Congestion Control Protocol (DCCP) is an unreliable, connection
[DCCP]: Promote CCID2 as default CCID This patch addresses the following problems: 1. DCCP relies for its proper functioning on having at least one CCID module enabled (as in TCP plugable congestion control). Currently it is possible to disable both CCIDs and thus leave the DCCP module in a compiled, but entirely non-functional state: no sockets can be created when no CCID is available. Furthermore, the protocol is (again like TCP) not intended to be used without CCIDs. Last, a non-empty CCID list is needed for doing CCID feature negotiation. 2. Internally the default CCID that is advertised by the Linux host is set to CCID2 (DCCPF_INITIAL_CCID in include/linux/dccp.h). Disabling CCID2 in the Kconfig menu without changing the defaults leads to a failure `module not found' when trying to load the dccp module (which internally tries to load the default CCID). 3. The specification (RFC 4340, sec. 10) treats CCID2 somewhat like a `minimum common denominator'; the specification says that: * "New connections start with CCID 2 for both endpoints" * "A DCCP implementation intended for general use, such as an implementation in a general-purpose operating system kernel, SHOULD implement at least CCID 2. The intent is to make CCID 2 broadly available for interoperability [...]" Providing CCID2 as minimum-required CCID (like Reno/Cubic in TCP) thus seems reasonable. Hence this patch automatically selects CCID2 when DCCP is enabled. Documentation also added. Discussions with Ian McDonald on this subject are gratefully acknowledged. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk> Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-11-21 12:09:56 +00:00
oriented protocol designed to solve issues present in UDP and TCP, particularly
for real-time and multimedia (streaming) traffic.
It divides into a base protocol (RFC 4340) and pluggable congestion control
modules called CCIDs. Like pluggable TCP congestion control, at least one CCID
[DCCP]: Promote CCID2 as default CCID This patch addresses the following problems: 1. DCCP relies for its proper functioning on having at least one CCID module enabled (as in TCP plugable congestion control). Currently it is possible to disable both CCIDs and thus leave the DCCP module in a compiled, but entirely non-functional state: no sockets can be created when no CCID is available. Furthermore, the protocol is (again like TCP) not intended to be used without CCIDs. Last, a non-empty CCID list is needed for doing CCID feature negotiation. 2. Internally the default CCID that is advertised by the Linux host is set to CCID2 (DCCPF_INITIAL_CCID in include/linux/dccp.h). Disabling CCID2 in the Kconfig menu without changing the defaults leads to a failure `module not found' when trying to load the dccp module (which internally tries to load the default CCID). 3. The specification (RFC 4340, sec. 10) treats CCID2 somewhat like a `minimum common denominator'; the specification says that: * "New connections start with CCID 2 for both endpoints" * "A DCCP implementation intended for general use, such as an implementation in a general-purpose operating system kernel, SHOULD implement at least CCID 2. The intent is to make CCID 2 broadly available for interoperability [...]" Providing CCID2 as minimum-required CCID (like Reno/Cubic in TCP) thus seems reasonable. Hence this patch automatically selects CCID2 when DCCP is enabled. Documentation also added. Discussions with Ian McDonald on this subject are gratefully acknowledged. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk> Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-11-21 12:09:56 +00:00
needs to be enabled in order for the protocol to function properly. In the Linux
implementation, this is the TCP-like CCID2 (RFC 4341). Additional CCIDs, such as
the TCP-friendly CCID3 (RFC 4342), are optional.
For a brief introduction to CCIDs and suggestions for choosing a CCID to match
given applications, see section 10 of RFC 4340.
It has a base protocol and pluggable congestion control IDs (CCIDs).
DCCP is a Proposed Standard (RFC 2026), and the homepage for DCCP as a protocol
is at http://www.ietf.org/html.charters/dccp-charter.html
Missing features
================
The Linux DCCP implementation does not currently support all the features that are
specified in RFCs 4340...42.
The known bugs are at:
Docs/Kconfig: Update: osdl.org -> linuxfoundation.org Some of the documentation refers to web pages under the domain `osdl.org'. However, `osdl.org' now redirects to `linuxfoundation.org'. Rather than rely on redirections, this patch updates the addresses appropriately; for the most part, only documentation that is meant to be current has been updated. The patch should be pretty quick to scan and check; each new web-page url was gotten by trying out the original URL in a browser and then simply copying the the redirected URL (formatting as necessary). There is some conflict as to which one of these domain names is preferred: linuxfoundation.org linux-foundation.org So, I wrote: info@linuxfoundation.org and got this reply: Message-ID: <4CE17EE6.9040807@linuxfoundation.org> Date: Mon, 15 Nov 2010 10:41:42 -0800 From: David Ames <david@linuxfoundation.org> ... linuxfoundation.org is preferred. The canonical name for our web site is www.linuxfoundation.org. Our list site is actually lists.linux-foundation.org. Regarding email linuxfoundation.org is preferred there are a few people who choose to use linux-foundation.org for their own reasons. Consequently, I used `linuxfoundation.org' for web pages and `lists.linux-foundation.org' for mailing-list web pages and email addresses; the only personal email address I updated from `@osdl.org' was that of Andrew Morton, who prefers `linux-foundation.org' according `git log'. Signed-off-by: Michael Witten <mfwitten@gmail.com> Signed-off-by: Jiri Kosina <jkosina@suse.cz>
2010-11-15 19:55:34 +00:00
http://www.linuxfoundation.org/collaborate/workgroups/networking/todo#DCCP
For more up-to-date versions of the DCCP implementation, please consider using
the experimental DCCP test tree; instructions for checking this out are on:
Docs/Kconfig: Update: osdl.org -> linuxfoundation.org Some of the documentation refers to web pages under the domain `osdl.org'. However, `osdl.org' now redirects to `linuxfoundation.org'. Rather than rely on redirections, this patch updates the addresses appropriately; for the most part, only documentation that is meant to be current has been updated. The patch should be pretty quick to scan and check; each new web-page url was gotten by trying out the original URL in a browser and then simply copying the the redirected URL (formatting as necessary). There is some conflict as to which one of these domain names is preferred: linuxfoundation.org linux-foundation.org So, I wrote: info@linuxfoundation.org and got this reply: Message-ID: <4CE17EE6.9040807@linuxfoundation.org> Date: Mon, 15 Nov 2010 10:41:42 -0800 From: David Ames <david@linuxfoundation.org> ... linuxfoundation.org is preferred. The canonical name for our web site is www.linuxfoundation.org. Our list site is actually lists.linux-foundation.org. Regarding email linuxfoundation.org is preferred there are a few people who choose to use linux-foundation.org for their own reasons. Consequently, I used `linuxfoundation.org' for web pages and `lists.linux-foundation.org' for mailing-list web pages and email addresses; the only personal email address I updated from `@osdl.org' was that of Andrew Morton, who prefers `linux-foundation.org' according `git log'. Signed-off-by: Michael Witten <mfwitten@gmail.com> Signed-off-by: Jiri Kosina <jkosina@suse.cz>
2010-11-15 19:55:34 +00:00
http://www.linuxfoundation.org/collaborate/workgroups/networking/dccp_testing#Experimental_DCCP_source_tree
Socket options
==============
DCCP_SOCKOPT_QPOLICY_ID sets the dequeuing policy for outgoing packets. It takes
a policy ID as argument and can only be set before the connection (i.e. changes
during an established connection are not supported). Currently, two policies are
defined: the "simple" policy (DCCPQ_POLICY_SIMPLE), which does nothing special,
and a priority-based variant (DCCPQ_POLICY_PRIO). The latter allows to pass an
u32 priority value as ancillary data to sendmsg(), where higher numbers indicate
a higher packet priority (similar to SO_PRIORITY). This ancillary data needs to
be formatted using a cmsg(3) message header filled in as follows:
cmsg->cmsg_level = SOL_DCCP;
cmsg->cmsg_type = DCCP_SCM_PRIORITY;
cmsg->cmsg_len = CMSG_LEN(sizeof(uint32_t)); /* or CMSG_LEN(4) */
DCCP_SOCKOPT_QPOLICY_TXQLEN sets the maximum length of the output queue. A zero
value is always interpreted as unbounded queue length. If different from zero,
the interpretation of this parameter depends on the current dequeuing policy
(see above): the "simple" policy will enforce a fixed queue size by returning
EAGAIN, whereas the "prio" policy enforces a fixed queue length by dropping the
lowest-priority packet first. The default value for this parameter is
initialised from /proc/sys/net/dccp/default/tx_qlen.
DCCP_SOCKOPT_SERVICE sets the service. The specification mandates use of
service codes (RFC 4340, sec. 8.1.2); if this socket option is not set,
the socket will fall back to 0 (which means that no meaningful service code
is present). On active sockets this is set before connect(); specifying more
than one code has no effect (all subsequent service codes are ignored). The
case is different for passive sockets, where multiple service codes (up to 32)
can be set before calling bind().
DCCP_SOCKOPT_GET_CUR_MPS is read-only and retrieves the current maximum packet
size (application payload size) in bytes, see RFC 4340, section 14.
DCCP_SOCKOPT_AVAILABLE_CCIDS is also read-only and returns the list of CCIDs
supported by the endpoint. The option value is an array of type uint8_t whose
size is passed as option length. The minimum array size is 4 elements, the
value returned in the optlen argument always reflects the true number of
built-in CCIDs.
DCCP_SOCKOPT_CCID is write-only and sets both the TX and RX CCIDs at the same
time, combining the operation of the next two socket options. This option is
preferable over the latter two, since often applications will use the same
type of CCID for both directions; and mixed use of CCIDs is not currently well
understood. This socket option takes as argument at least one uint8_t value, or
an array of uint8_t values, which must match available CCIDS (see above). CCIDs
must be registered on the socket before calling connect() or listen().
DCCP_SOCKOPT_TX_CCID is read/write. It returns the current CCID (if set) or sets
the preference list for the TX CCID, using the same format as DCCP_SOCKOPT_CCID.
Please note that the getsockopt argument type here is `int', not uint8_t.
DCCP_SOCKOPT_RX_CCID is analogous to DCCP_SOCKOPT_TX_CCID, but for the RX CCID.
DCCP_SOCKOPT_SERVER_TIMEWAIT enables the server (listening socket) to hold
timewait state when closing the connection (RFC 4340, 8.3). The usual case is
that the closing server sends a CloseReq, whereupon the client holds timewait
state. When this boolean socket option is on, the server sends a Close instead
and will enter TIMEWAIT. This option must be set after accept() returns.
DCCP_SOCKOPT_SEND_CSCOV and DCCP_SOCKOPT_RECV_CSCOV are used for setting the
partial checksum coverage (RFC 4340, sec. 9.2). The default is that checksums
always cover the entire packet and that only fully covered application data is
accepted by the receiver. Hence, when using this feature on the sender, it must
be enabled at the receiver, too with suitable choice of CsCov.
DCCP_SOCKOPT_SEND_CSCOV sets the sender checksum coverage. Values in the
range 0..15 are acceptable. The default setting is 0 (full coverage),
values between 1..15 indicate partial coverage.
DCCP_SOCKOPT_RECV_CSCOV is for the receiver and has a different meaning: it
sets a threshold, where again values 0..15 are acceptable. The default
of 0 means that all packets with a partial coverage will be discarded.
Values in the range 1..15 indicate that packets with minimally such a
coverage value are also acceptable. The higher the number, the more
restrictive this setting (see [RFC 4340, sec. 9.2.1]). Partial coverage
settings are inherited to the child socket after accept().
The following two options apply to CCID 3 exclusively and are getsockopt()-only.
In either case, a TFRC info struct (defined in <linux/tfrc.h>) is returned.
DCCP_SOCKOPT_CCID_RX_INFO
Returns a `struct tfrc_rx_info' in optval; the buffer for optval and
optlen must be set to at least sizeof(struct tfrc_rx_info).
DCCP_SOCKOPT_CCID_TX_INFO
Returns a `struct tfrc_tx_info' in optval; the buffer for optval and
optlen must be set to at least sizeof(struct tfrc_tx_info).
[DCCP]: Honour and make use of shutdown option set by user This extends the DCCP socket API by honouring any shutdown(2) option set by the user. The behaviour is, as much as possible, made consistent with the API for TCP's shutdown. This patch exploits the information provided by the user via the socket API to reduce processing costs: * if the read end is closed (SHUT_RD), it is not necessary to deliver to input CCID; * if the write end is closed (SHUT_WR), the same idea applies, but with a difference - as long as the TX queue has not been drained, we need to receive feedback to keep congestion-control rates up to date. Hence SHUT_WR is honoured only after the last packet (under congestion control) has been sent; * although SHUT_RDWR seems nonsensical, it is nevertheless supported in the same manner as for TCP (and agrees with test for SHUTDOWN_MASK in dccp_poll() in net/dccp/proto.c). Furthermore, most of the code already honours the sk_shutdown flags (dccp_recvmsg() for instance sets the read length to 0 if SHUT_RD had been called); CCID handling is now added to this by the present patch. There will also no longer be any delivery when the socket is in the final stages, i.e. when one of dccp_close(), dccp_fin(), or dccp_done() has been called - which is fine since at that stage the connection is its final stages. Motivation and background are on http://www.erg.abdn.ac.uk/users/gerrit/dccp/notes/shutdown A FIXME has been added to notify the other end if SHUT_RD has been set (RFC 4340, 11.7). Note: There is a comment in inet_shutdown() in net/ipv4/af_inet.c which asks to "make sure the socket is a TCP socket". This should probably be extended to mean `TCP or DCCP socket' (the code is also used by UDP and raw sockets). Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk> Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2007-11-21 11:56:48 +00:00
On unidirectional connections it is useful to close the unused half-connection
via shutdown (SHUT_WR or SHUT_RD): this will reduce per-packet processing costs.
Sysctl variables
================
Several DCCP default parameters can be managed by the following sysctls
(sysctl net.dccp.default or /proc/sys/net/dccp/default):
request_retries
The number of active connection initiation retries (the number of
Requests minus one) before timing out. In addition, it also governs
the behaviour of the other, passive side: this variable also sets
the number of times DCCP repeats sending a Response when the initial
handshake does not progress from RESPOND to OPEN (i.e. when no Ack
is received after the initial Request). This value should be greater
than 0, suggested is less than 10. Analogue of tcp_syn_retries.
retries1
How often a DCCP Response is retransmitted until the listening DCCP
side considers its connecting peer dead. Analogue of tcp_retries1.
retries2
The number of times a general DCCP packet is retransmitted. This has
importance for retransmitted acknowledgments and feature negotiation,
data packets are never retransmitted. Analogue of tcp_retries2.
tx_ccid = 2
Default CCID for the sender-receiver half-connection. Depending on the
choice of CCID, the Send Ack Vector feature is enabled automatically.
rx_ccid = 2
Default CCID for the receiver-sender half-connection; see tx_ccid.
seq_window = 100
The initial sequence window (sec. 7.5.2) of the sender. This influences
the local ackno validity and the remote seqno validity windows (7.5.1).
Values in the range Wmin = 32 (RFC 4340, 7.5.2) up to 2^32-1 can be set.
tx_qlen = 5
The size of the transmit buffer in packets. A value of 0 corresponds
to an unbounded transmit buffer.
sync_ratelimit = 125 ms
The timeout between subsequent DCCP-Sync packets sent in response to
sequence-invalid packets on the same socket (RFC 4340, 7.5.4). The unit
of this parameter is milliseconds; a value of 0 disables rate-limiting.
IOCTLS
======
FIONREAD
Works as in udp(7): returns in the `int' argument pointer the size of
the next pending datagram in bytes, or 0 when no datagram is pending.
Other tunables
==============
Per-route rto_min support
CCID-2 supports the RTAX_RTO_MIN per-route setting for the minimum value
of the RTO timer. This setting can be modified via the 'rto_min' option
of iproute2; for example:
> ip route change 10.0.0.0/24 rto_min 250j dev wlan0
> ip route add 10.0.0.254/32 rto_min 800j dev wlan0
> ip route show dev wlan0
CCID-3 also supports the rto_min setting: it is used to define the lower
bound for the expiry of the nofeedback timer. This can be useful on LANs
with very low RTTs (e.g., loopback, Gbit ethernet).
Notes
=====
DCCP does not travel through NAT successfully at present on many boxes. This is
because the checksum covers the pseudo-header as per TCP and UDP. Linux NAT
support for DCCP has been added.