net_design.nr

早期freebsd实现

allbox, tab(+);
l l l l.
Option+Send+Forward+Receive
=
Padding+may be set+-+-
Security+reject+ignore+discard
Source Route+\fIclnp_srcroute()\fR+\fIclnp_srcroute()\fR+-
Record Route+-+\fIclnp_dooptions()\fR+-
QOS+added+congestion bit set+tpclnp_ctlinput()
Priority+reject+ignore+-
.TE
.)b
.sh 2 "DT NPDU Segmentation"
.pp
Segmentation is the process by which initial NPDUs are segmented into 
smaller derived NPDUs when the initial NPDU is too large for transmission
on a network interface.
Segmentation is accomplished by \fIclnp_fragment()\fR. 
This function chops the NPDU into pieces and individually places the pieces
in the appropriate network interface's output queue. 
Each piece is made as large as possible. 
Note: The phrase "fragmentation" is used synonymously with "segmentation"
throughout this prose and the CLNP fragmentation code. This is due to 
this author's familiarity with the DoD Internet Protocol which uses
the term "fragment."
.sh 2 "DT NPDU Reassembly"
.pp
Derived NPDUs are put back together by the process called 
reassembly. 
Reassembly is performed only at the destination end system.
When a derived NPDU arrives, it is passed to \fIclnp_reass()\fR. 
This function scans a linked list of NPDUs awaiting reassembly. 
Each packet in the list is represented by a fragment list
descriptor, which is stored in an \fImbuf\fR:
.(b
\fC
.TS
tab(+);
l s s s.
struct clnp_fragl {
.T&
l l l l.
+struct iso_addr+cfl_src;+/* source */
+struct iso_addr+cfl_dst;+/* destination */
+u_short+cfl_id;+/* id of the pkt */
+u_char+cfl_ttl;+/* time to live */
+u_short+cfl_last;+/* offset of last 
+++byte of packet */
+struct mbuf +*cfl_orighdr;+/* ptr to 
+++original header */
+struct clnp_frag+*cfl_frags;+/* linked list 
+++of fragments */
+struct clnp_fragl+*cfl_next;+/* next pkt be-
+++ing reassembled */
};
.TE
\fR
.)b
The fields \fIcfl_src\fR, \fIcfl_dst\fR, and \fIcfl_id\fR are used to
match an incoming derived NPDU with a fragment list. 
\fICfl_orighdr\fR contains a copy of the NPDU header of the first fragment received. 
The linked list of fragments pertaining to the packet is stored in the
\fIcfl_frags\fR field. 
Each NPDU fragment represented by a \fIclnp_frag\fR structure, 
stored in an \fImbuf\fR:
.(b
\fC
.TS
tab(+);
l s s s.
struct clnp_frag {
.T&
l l l l.
+u_int+cfr_first;+/* offset of 
+++first byte of this frag */
+u_int+cfr_last;+/* offset of last 
+++byte of this frag */
+u_int+cfr_bytes;+/* bytes to shave */
+struct mbuf+*cfr_data;+/* ptr to data */
+struct clnp_frag+*cfr_next;+/* next frag */
};
.TE
\fR
.)b
The fields \fIcfr_first\fR and \fIcfr_last\fR indicate the first and
last octet of the fragment. 
\fICfr_data\fR points to an mbuf chain
which contains the data for the fragment.
.pp
If \fIclnp_reass()\fR finds a \fIclnp_fragl\fR structure matching the
incoming derived NPDU, \fIclnp_insert_frag()\fR is called to create
a \fIclnp_frag\fR structure and insert it in the linked list of
packet fragments. 
If no \fIclnp_fragl\fR structure is found, 
\fIclnp_newpkt()\fR is invoked to create a new fragment list structure.
.pp
The last task \fIclnp_reass()\fR performs is to check if the fragment
that just arrived completes the reassembly of the initial NPDU. 
If it does, the reassembled NPDU is rearranged to 
look like it just arrived intact.
It accomplishes this by linking the \fImbuf\fRs holding
the fragments into one \fImbuf\fR chain that represents the initial
NPDU.
A pointer to this \fImbuf\fR chain is returned by \fIclnp_reass()\fR.
.pp
If the newly arrived fragment does not complete an initial NPDU, 
\fIclnp_reass()\fR returns NULL.
.sh 3 "Reassembly Lifetime Control"
.pp
One function of the CLNP is to prevent
a proliferation of fragments awaiting reassembly from
consuming buffers in an end system for indefinite periods of time.
This function is called reassembly lifetime control.
It is accomplished by 
periodic traversal of
the list of \fIclnp_fragl\fR structures, decrementing the 
\fIcfl_ttl\fR field. 
This field is a copy of the NPDU time-to-live
field. If \fIcfl_ttl\fR reaches zero, all resources associated with the
fragment are released.
The procedure
\fIclnp_slowtimo()\fR, which is called by the system
clock every 500 milliseconds (every half-second),
performs the CLNP reassembly lifetime control.
.sh 2 "ER NPDU"
.pp
An ER NPDU is sent to the originator of a packet when a DT NPDU is
discarded and the error report function is not suppressed. Suppression
of the error report function is accomplished by setting the "no ER"
bit in the CLNP header.
A packet is discarded by \fIclnp_discard()\fR. 
Before it
returns the \fImbufs\fR used to store the 
the discarded packet to the \fImbuf\fR free list,
\fIclnp_discard()\fR 
determines if the error report function is suppressed. 
If not, 
an ER NPDU will be sent to the originator of the discarded packet by
calling \fIclnp_emit_er()\fR.
.pp
\fIClnp_emit_er()\fR will create an ER NPDU, address it to the 
originator of the discarded packet, route the NPDU, 
and transmit it, sending the header of the discarded NPDU as data. 
ER NPDUs may not be segmented. 
If the ER NPDU is too large for the outgoing network interface, 
the packet is truncated.
.sh 2 "Raw CLNP"
.pp
In order to test CLNP in isolation from higher layer
protocols, ARGO provides a \*(lqraw\*(rq interface to CLNP.
This raw interface is selected with the \fISOCK_RAW\fR parameter to
the
\fIsocket()\fR
system call.
When a \*(rqraw\*(rq socket is open,
and CLNP receives an NPDU,
CLNP must determine whether the incoming NPDU is destined for
the 
\*(rqraw\*(rq interface or for the interface to the 
OSI transport protocol entity.
ARGO addresses this problem by using non-standard NPDU types
for packets sent on \*(rqraw\*(rq sockets.
The type field in the CLNP NPDU header
is set to \fICLNP_RAW\fR (hex 1d) rather than \fICLNP_DT\fR
in NPDUs that originate from 
\*(rqraw\*(rq sockets.
This non-standard type value is used by \fIclnp_input()\fR
to decide which upper layer protocol should receive the packet.
See \fIclnptest(8)\fR for more information about the.
\*(rqraw\*(rq CLNP interface.
.sh 2 "CLNP Echo"
.pp
In the DoD world, ICMP supports an \fIecho\fR service. 
This allows one to \*(lqping\*(rq a distant gateway and 
to receive an echo response (a packet in return) if the gateway is working.
There is no counterpart to \*(lqecho\*(rq in ISO 8473 (CLNP). 
ARGO provides this non-standard feature in its connectionless
network layer.
.pp
Like raw CLNP, implementing an echo function requires a non-standard
NPDU type value to allow
\fIclnp_input()\fR to differentiate between a DT NPDU to be forwarded
or passed to a higher layer protocol, and an NPDU that is to be echoed.
When requesting an echo, 
the CLNP type field is set to \fICLNP_EC\fR (hex 1E) rather
than CLNP_DT. 
When \fIclnp_input()\fR receives a packet with type
\fICLNP_EC\fR, 
it swaps the source and destination addresses, sets the
type field to \fICLNP_ECR\fR (hex 1F) and forwards
the packet back to the sender. 
See also \fIclnpping(8)\fR.
.sh 2 "Timers"
.pp
The only timer used by CLNP is the 
500 millisecond timer, which is 
user for reassembly lifetime control.
See the section \*(lqReassembly Lifetime Control.\*(rq
.sh 1 "End System to Intermediate System Routing Protocol (ES-IS)"
.\" ROB
.sh 2 "Overview"
.pp
This section describes the implementation of the ES-IS routing protocol.
This protocol is used primarily to resolve NSAP address to SNPA address 
translations. It is also used to identify end systems
and intermediate systems on
the local subnetwork. 
All of this work is accomplished by transmitting
packets of the type End System Hello (ESH), Intermediate System Hello (ISH)
and Request Redirect (RD).
.pp
For the purpose of this section, the following definitions of end system (ES)
and intermediate system (IS) apply.
.ip \(bu 5
An \fIend system\fR is an open system that
is an OSI end system in the standard OSI sense
(that it supports a full OSI protocol suite in addition to the network layer)
and that
implements the functions of the
the ES-IS protocol that are mandatory for end systems,
such as the Query Configuration function and the Record Redirect
function,
but that does not implement
the functions of the ES-IS protocol that are for intermediate systems.
.ip \(bu 5
An \fIintermediate system\fR is an open system that
is an OSI intermediate system in the standard OSI sense
(that it performs packet routing in the network layer)
and that
implements the functions of the
the ES-IS protocol that are mandatory for intermediate systems,
such as the Request Redirect function, 
but not the functions of the ES-IS protocol that are for end systems.
.pp
While system may be an ES or an IS or both according to the
standard OSI definitions, this is not the case in the context of
the ES-IS protocol.
.pp
An ARGO system is by default an end system, by the definitions given above.
An ARGO system can be made to function as an intermediate system
instead of an end system with the \fIclnlutil\fR program. 
See \fIclnlutil(8)\fR for more information.
.sh 2 "Report Configuration Function"
.pp
The report configuration function is used by end systems and intermediate
systems to inform each other of their reachability and current subnetwork
addresses. 
This function is invoked whenever the configuration timer
expires. 
This timer fires at a frequency of once every
\fIesis_config_time\fR seconds. 
By default, this value is 60 (seconds), 
but it may be changed with the \fIclnlutil\fR program.
.pp
The report configuration function is contained in the C function 
\fIesis_config()\fR. Called every \fIesis_config_time\fR seconds, 
\fIesis_config()\fR searches the list of active network interfaces
calling \fIesis_shoutput\fR for each interface that is up, has
broadcast ability and has an ISO address configured.
.pp
The function \fIesis_shoutput()\fR has the responsibility of building and 
transmitting ESH and ISH packets.
It takes several arguments, including  a pointer to a network interface
and
a packet type (ESH or ISH).
If the packet type is ESH, then
each NSAP address configured on the specified interface is added to
the ESH NPDU. ISH NPDUs may only contain a single NSAP address\**.
.(f
\** Actually, ISH packets contain Network Entity Titles (NETs). ARGO
does not make a distinction between NETs and NSAPs.
.)f
After the packet is built, it is transmitted on the subnetwork. ESH packets
are sent to the multicast address \fIall intermediate systems\fR, whereas
ISH packets are sent to the multicast address \fIall end systems\fR.
.pp
Each ISH and ESH NPDU contains 
a holding timer setting. This setting (specified 
in seconds) is used by the receiver of the NPDU to set its
holding timer. When its holding timer expires, the information from
the NPDU is erased. The holding timer value sent on each ISH and ESH NPDU
is contained in the variable \fIesis_holding_time\fR. By default, this
timer setting is 120 seconds. This value may be changed with the 
\fIclnlutil\fR utility program.
.sh 2 "Record Configuration Function"
.pp
The Record Configuration function receives ESH or ISH NPDUs, extracts the
configuration information, and updates kernel-resident tables. 
The two functions \fIesis_eshinput()\fR and \fIesis_ishinput()\fR 
process incoming ESH and ISH NPDUs, respectively.
.pp
The ES-IS entity maintains a table that
associates a SNPA-addresses with NSAP-addresses.
This table is called the \fISNPA cache\fR.
.pp
Whenever an ESH or ISH NPDU is received, 
an entry is made in the SNPA cache
via the \fIsnpac_add()\fR function. 
This entry is kept in the cache until the holding timer expires. 
In addition to adding an entry to the SNPA cache, \fIsnpac_add()\fR creates
a default ISO route toward the sender of the ISH.
One such route is kept so that the ES-IS entity has at most one
route to an IS at any time.
Note that ISHs from different sources will 
cause the route to the source of the earlier ISH to be 
overwritten.
The default route
will be removed when the ISH holding timer expires.
.pp
If, at the time an ESH or ISH NPDU is received, the SNPA cache
contains no entry for the NSAP address in the NPDU just received, 
an ESH or ISH (depending on the system type) NPDU is
transmitted to the sender of the NPDU just received.
.sh 2 "Resolving NSAP addresses to SNPA addresses: Query Configuration Function"
.pp
Whenever a device driver needs to resolve an NSAP address to 
an SNPA address, it calls \fIiso_snparesolve()\fR. This function first looks
up the NSAP address in the SNPA cache. If a match is found, the
corresponding SNPA address is returned. If a match is not found and the
system is an end system, and there is a known intermediate system, then
the SNPA address of the intermediate system is returned. It is assumed that
the intermediate system will forward the packet and transmit a redirect back
(see "Redirection Generation", below).
If a match is not found and the system is an end system, but there is no
known intermediate system, then \fIiso_snparesolve()\fR will return 
the multicast address \fIall end systems\fR. 
In all other cases, \fIiso_snparesolve()\fR will return an error.
This is known as the query configuration function. 
.sh 3 "Configuration Response Function"
.pp
In order for the query configuration function to be effective, the network
entity that receives a CLNP DT sent to the \fIall end system\fR
multicast address must transmit an ESH back to the sender of the DT.
This is called the configuration response function and is accomplished by
calling \fIsh_output()\fR from within \fIclnp_input()\fR.
.sh 2 "Redirection Generation"
.pp
When an intermediate system forwards a packet onto the same interface 
upon which 
the packet arrived, a redirect (RD) NPDU is generated. This NPDU is
transmitted by calling \fIesis_rdoutput()\fR from within \fIclnp_forward()\fR.
Note that end systems may forward packets but they do not generate RD PDUs.
.sh 2 "Redirection Receipt"
.pp
RD NPDUs direct an end system to create an SNPA cache entry 
for an NSAP address, or, if such an entry exists, to change
the SNPA address associated with the NSAP address.
The receipt of RD NPDUs is handled by \fIesis_rdinput()\fR. 
This function
parses the RD NPDU and adds an entry to the SNPA cache for the corresponding
destination NSAP address.
If the redirect is toward an intermediate system,
meaning that the RD NPDU contains an SNPA address
of an intermediate system (gateway),
a route is created for the destination NSAP with the intermediate system as
the first hop, or gateway, in the route.
.sh 2 "Multicast Addresses"
.pp
As specified by the December 1987 NBS agreements, the address
\fIall end systems\fR is {0x09, 0x00, 0x2B, 0x00, 0x00, x04} and the address
\fIall intermediate systems\fR is {0x09, 0x00, x02B, 0x00, 0x00, 0x05}. 
These multicast addresses are only used on the 802.3 subnetwork (baseband).
Broadcast addresses are used on the 802.5 subnetwork (token ring). See
the comment in \fC/sys/netargo/iso_snpac.c\fR for more information on 
multicast addresses.
.sh 1 "Connection Oriented Network Service and Subnetwork Service"
.pp
The following sections describe the design of the Connection Oriented 
Network Service (CONS) and the Connection Oriented Subnetwork Service
(COSNS).
The CONS and COSNS are provided by two functionally separate but related
modules, a connection manager and the ISO 8208 (X.25) protocols.
The connection manager is also known in OSI terminology as a 
subnetwork dependent convergence function, or SNDCF.
In ARGO it is used for more than an SNDCF, and it is a sort of 
"glue" that binds a transport service, a network service, a
subnetwork service, and a device driver together, so 
hereinafter it is called "the glue".
This code performs the some of the functions of ISO 8878,
which specifies how ISO 8208 (X.25) can be used to provide the OSI