Simtel MSDOS - Coast to Coast/simteldosarchivecoasttocoast2.iso/ncsatlnt/dev_note.txt

Text File | 1994-03-07 | 56KB | 1,327 lines

NCSA Telnet
Developers Guide

This is an ASCII-printable version of the real documentation.
It was crudely transformed from the original Macintosh MSWord version
which is also included on this disk.  The Mac version is much easier
to read and contains the same words.



Version 2.2 release
July 15, 1988
National Center for Supercomputing Applications
The code and documentation are in the public domain.


Note to developers:

Thanks for your interest and support of this source code.  Please give us your
comments on where the libraries and/or source need more documentation.

We keep a running list of bugs and modify the source code ourselves,  so
please submit diffs for the changes you make to avoid duplicated effort.

Those of you who care will note that there are ifdefs in "whatami.h" which
control the difference between compiling on a PC or Macintosh with AppleTalk
or Ethernet.  Most of the code runs on both machines and we will keep it that
way.

The enclosed "enhancements" list is a list of projects that we ARE NOT
currently working on, though we may pick up one or two.  What we plan to do is
maintain this list with names of people who are working on each project.  If
you do pick up one of the larger projects, let us know and we will funnel
questions, help and ideas in your direction.  The mailing list may help here
also.

For questions, send us email first, think about it for a while, read the
developers documentation again, then give us a call.

Chris Wilson
[e-mail]
[phone]

Quincey Koziol
[e-mail]
[phone]



Compiler Notes

Source changes from 2.1 to 2.2

To help the port to MPW C, internal ints in the protocol code are
declared with a length, like int16 or int32.
Some additional calls and events were added to support better
functionality in the FTP server.
The drivers for Ethernet were combined on the PC, function
pointers are used to reference the board calls.
The AppleTalk and EtherTalk low level routines are switchable in
the Mac version.
On the Mac, the stored IP number and configuration information
were moved to an "NCSA Telnet Settings" file in the System Folder.
RARP support was added, the special IP number "R","A","R","P" is
used to indicate that the value has not been looked up yet.

PC

The PC version compiles in Lattice C version 3.0 or 3.1.  The Macro Assembler
code must be modified to work with Microsoft C 4.0 - I don't know what MSC 5.0
does.  Modifying each driver means taking each one and converting the public
labels and also the return values for all LONGS and POINTERS.  MSC uses a
different calling convention than Lattice C.  Also, static pointers in the
rgX.c programs must be changed.  If I get a real good DIFF for the MSC code, I
will post it.

The assembler code works for Lattice C in IBM Macro Assembler 2.0 (MS 3.0),
but the Microsoft 4.0 assembler support requires MS Assember 4.0.  There are
different assembler files for supporting each compiler.  They are not hard to
translate, NCSAIO even supports both.

If you have the Lattice C compiler and MASM installed in your PATH, just run
the batch file: COMPILE.BAT.

Mac

Execute the Unix Aliases script in the Examples directory.  The source code as
posted will make and run under MPW/MPW C 2.02.

We compiled the release 2.2 under Aztec C, but future releases will not use
Aztec C.  Because we are converting as of this release date, Aztec C
compatible source is available ONLY upon request.



Suggestions for source code enhancements

User FTP
The Macinstosh needs a user FTP.  The source to user FTP for the
PC and the source to the background FTP server could be used as an
example for how to get the file transferred.  The user interface
should obviously be in the Macintosh style and allow file
transfers to be specified by dragging file names or icons.
Other UNIX utilities
They won't be direct ports, but it should be relatively easy to
create a user rcp, rlogin, finger, rwho, and other BSD utilities.
PING and finger have been common requests.
lpr
lpr is not a difficult protocol, and an lpr application for the PC
and/or the Mac would be very handy.  The problem would be that all
of the PCs and Macs would have to be trusted hosts.  One
suggestion there is to modify lpd to accept members of a
particular netgroup in addition to/instead of trusted hosts.
New drivers
There are many more Ethernet and Token ring boards for the PC.  To
write a new driver, look at the files in the enet directory for
examples.  For a new Ethernet board, you should only have to
replace this one file and hook it into pctools.c, no other mods
will be required.
New video drivers for PC
More RG routines can be written for different PC video options.
Driver for smart board
The code is divided into the TCP/IP layer and the session layer.
Certain routines in the TCP/IP layer, all beginning with "net"
could be replaced with routines which reside inside a "smart"
board.  netopen, netclose, netlisten, netread, netwrite, netsetip,
for starters, and all of the status routines have to be integrated
with the board (netqlen, netest).  You also have to make sure that
the events for netevent get posted when anything on the board
changes.
New terminal emulator (ADM, DG, etc)
There is additional documentation on the insertion of new
emulators.  These character oriented terminal emulators should be
easy to incorporate.
User mail for PC and Mac
We need to find a POP server to standardize on.  With that, a POP
client with local disk storage would be easy to do.
UDP
The UDP support needs to be filled out by adding support for
allocation of several ports, each of which buffer data waiting to
be read by the applications.  A buffering scheme for each port
would keep the last few packets for each port which is being
listened to.  These buffers should be allocated when the UDP port
is opened (like TCP).
Tek 4105
Filter all 4105 commands, find a subset of features to emulate so
that we can take advantage of color.
tn3270
Some interest in different models of the 3270 series emulated,
color, 44 lines, etc.
TSR telnet for PC
A Sidekick-like program which pops up on a certain key sequence
Windows/DesqView version for PC
Multiple session output goes to different windows on PC
Network "ping" monitor
Ping a bunch of hosts to see which ones are up
SLIP
Serial Line IP is now a part of 4.3BSD UNIX, so there is a lot of
interest in providing all of the NCSA Telnet capabilities over
phone lines.
FTP server improvements
A DIR command (LIST) could show file sizes and other information.
The names of subdirectories should show up in the list.



TCP Kernel Calls

Tim Krauskopf                        July 11, 1988

netsetip(ipnum)
char ipnum[4];        IP address to use.
Set my IP address.  This routine must be called before
calling netinit, in order for netinit to know what return
address to use during initialization.  Usually called by a
session layer routine.

netconfig(hw)
char *hw;        Hardware type - string of <10 chars.
This call tells the kernel what type of hardware to use,
"AppleTalk" or "Ether" on the Mac, for example.  When
multiple hardware types are supported, the lower layer must
be informed of which type to use.  Called by Snetinit()
after the value is read from the configuration file.

netparms(irq,address,ioaddr)
int irq;            Hardware parameter,  irq is the interrupt request
level
int address;        Shared memory segment address to use for starting
Ethernet driver
int ioaddr;        Hardware I/O address parameter
Set hardware dependent network configuration.  The address
and I/O address parameters only apply if the hardware needs
them.  These values are filled in by the Session layer in
most cases.

neteventinit()
Initializes network event queue so that errors and
connection information can be posted.  VERY IMPORTANT.
Called by Snetinit().

netarptime(secs)
int secs;        Amount of time in seconds to try to find local hosts or
local gateways.
A particularly slow machine may take five seconds to respond
to an ARP request.  Most return an answer within 1/2 second.
This timeout produces the "local host not responding" error.

netsetmask(mask)
char mask[4];        Subnet mask to use for future gateway calculations
Set the subnetting mask.  If this is not called before
netinit(),  netinit() will install the default addressing
mask for the class A,B or C network according to the address
in netsetip.

netgetmask(mask)
char mask[4];        location to put the copy of the subnet mask
Copies the subnetting mask.  Not very useful if called
before the network initialization.  Session layer routines
are typically used to set the value before you try to call
netgetmask().

netgetip(ipnum)
char ipnum[4];        Location to place IP address
Get my IP address.  Used when one session layer routine sets
the IP address but another one needs to know the value.  Any
time after netsetip, netgetip can obtain a copy of the
address.

char *neterrstring(errno)
int errno;        error number of look up
Get a pointer to the error string associated with "errno".
Your custom error handler can use this to look up a "canned"
error message.  Copy the message, print it, whatever.  Don't
worry if the errno is -1, that is a special case that works
OK.  Neterrstring will always return a valid string.

netinit()
Sets up the hardware and initializes vars.  Must be called
before netopen.  Use of Snetinit() is recommended instead of
netinit().  It calls Sreadhosts(), netinit() and Ssetgates()
for you.

netsetgate(ipnum)
char ipnum[4];        IP address of gateway
Install this IP number as a gateway.  Does an ARP request
immediately, because we assume we will need its Ethernet
address.

dl = netdlayer(ipnum)
char *dl;        48-bit Ethernet address or psuedo-address from ARP cache
char  ipnum[4];        IP number to look up with ARP.
Queries the network with a broadcast ARP request every
second and returns the Ethernet address for the requested IP
address.  Returns NULL on failure.  Blocks for arptime
seconds or until the response arrives.  One of the very few
blocking calls, only done this way because it is easier and
ARP is so quick.

netgetrarp()
Pulls the local Ethernet address stored during the netinit()
and sends a RARP packet every second.  Waits for the
response and returns 0 for success or negative for failure.
If a valid response comes in, then netgetip() will return
the new IP number.  Unfortunately, to get everything to
work, the combination netgetip(); netsetip() must be
executed to set all of the internal variables.

netsegsize(size)
int size;        new segment size to use for future connection attempts
Set the maximum segment size (for incoming packets).  This
value will take affect on all future opening connections,
whether from netopen() or netlisten().  Can be changed any
time, but cannot affect current connections, only future
ones.  Netlisten takes this value when called, not when the
connection eventually opens.

netquench(limit)
int limit;        largest TCP window that can be advertised.
Set the maximum window (for incoming packets) that future
connections should allow.   Will have no effect when set to
a value over the program's maximum.

netfromport(port)
int port;        TCP protocol number to use for next netopen request.
When netopen starts a connection, the destination port
number is requested by the application, i.e. 23 for the
well-known telnet port.  The return (local) port number is
usually randomly generated to be unique from all local ports
in use.  If the application needs a specific local port
number (as the default port for FTP data transfers
requires), this call can force a particular number.  The
setting goes away after the next netopen call.

pn = netopen(ipnum,service)
char ipnum[4];        Address of machine to connect to, in binary
int service;        TCP port to attempt to connect to, often the WKA for a
service
int pn;            port number for use in later calls to just about every
routine available
Open a connection to another machine (up to 30 times).
Returns a file descriptor to use in later calls.  Netopen is
often used only in a session layer routine which does name
to IP address translation before calling netopen.  See
Snetopen() for open which includes name translation.

cnt = netread(pn,buf,len)
int pn;            port number from netopen()
char *buf;        pointer to data space to read into
int len;            maximum length of data to read
int cnt;            returns # of bytes read, -1 on closed connection, 0
for waiting
Similar to UNIX read, but using a file descriptor from
netopen().  Does not block EVER.

cnt = netwrite(pn,buf,len)
int pn;            port number from netopen()
char *buf;        pointer to data space to write from
int len;            length of data to try to write
int cnt;            returns # of bytes written, -1 on closed connection, 0
for waiting

Similar to UNIX write, but using a file descriptor from
netopen().  Does not block EVER.

netclose(pn)
int pn;            port number from netopen()
Close a connection, like closing a file.  You must make sure
(with netpush()) that all data which went through netwrite()
to this file descriptor has been sent before closing.
Before closing, you will also want to read all of the
available data from the connection if you don't want to lose
it.

netshut()
Shut down communication, disable interrupts.  It is not wise
to call any network routines at all after calling netshut().

netsleep(secs)
int secs;        number of seconds to wait before returning
This is what keeps the network drivers alive.  It makes sure
that any required ACKs are sent.  Often called with a
parameter of zero, but can be made to wait some # of secs.
Has a resolution of 1/18th of a second on PC, 1/60th on Mac.

netest(pn)
int pn;            port number from netopen()
Returns 0 if the connection is in established mode, <0 if
one side or the other has closed the connection.

cnt = netpush(pn)
int pn;            port number from netopen()
int cnt;            number of bytes still waiting to be delivered
(outgoing)
Sets push bit on transmit, returns size of output queue
(unacknowledged data).  Check to see that this value is zero
before calling netclose().  Unless you don't care.

netqlen(pn)
int pn;            portnumber from netopen()
Returns the number of bytes waiting to be read (incoming)
which have been acked, but have not been taken from the
queue.

netroom(pn)
int pn;            port number from netopen()
Get information on buffer space from a port.  netroom
returns the number of bytes available for a netwrite()
command to write into.  If the TCP is caught up, this will
be WINDOWSIZE, if we are loaded, this could be zero.

netgetftp(a,pn)
int a[8];        return values
int pn;            port number from netopen()
Get information about an active connection.  Returns the IP
address of the other machine which is connected to you and
the incoming and outgoing  TCP port numbers in use.  a[0] to
a[3] is the IP number of the other host.  Note that these
are integers, not characters!  a[4]  and a[5] are the high
and low bytes (yet stored in integers) of your local port
number.  a[6] and a[7] are the high and low bytes of the
other side's port number for this connection.

pn = netlisten(service)
int service;        TCP port number to listen to.
int pn;            Returns a port number equivalent to a netopen() port
number
Listen to a TCP port number.  The TCP will establish the
connection for you automatically.   You will be notified
with an event when the connection occurs.  Try not to get
the TCP port number confused with the netopen() port number.
The TCP port number defines the network connection between
machines while the netopen() port number is a local
descriptor for a given connection.

netusend(ipnum,port,retport,buf,len)
char *ipnum[4];    IP number of other machine
int port,retport;        to and from ports for the other machine's UDP socket
char *buf;        data to be sent in the packet
int len;            number of bytes of data to send, must fit in 512 byte
packet
Send a UDP packet to another machine, to a certain port,
with the data in buf, length n.

netulisten(port)
int port;        port to be sensitive about
Listen for a UDP packet on a certain incoming port.  Often
port is the same as retport on a netusend().  Any packets
for this port will be kept for reading later.  Another
packet for this port before a neturead will overwrite the
last one.  This should be expanded later to give similar
socket addressing as TCP uses.

theevent = netgetevent(class,theclass,dat)
int class;        Classes to search for (OR combination)
int *theclass;        Actual class of the event returned (return value)
int *dat;        Data which tags the event
int theevent;        Which event has occurred
Receive the next event in the queue.  Classes are in
whatami.h.  Not usually called by the user because the user
will want the session layer handling of Sgetevent().
Returns 0 on no event available.

netputevent(class,event,dat)
int class;        Class of the event to post
int event;        Which event within that class to post
int dat;            The data to carry in the event queue
Place an event into the event queue to be picked up later.
The user may want to define classes to be posted this way.
See Sgetevent().

netputuev(class,event,dat)
int class;        Class of the event to post
int event;        Which event within that class to post
int dat;            The data to carry in the event queue
Place an event into the event queue.  First checks to see if
there already is an identical such event.  If so, another
copy will not be posted.

netposterr(enum)
int enum;        Error number in errorclass.
Post a user error message.  The error message is one from
the list in TOOLS.C.  This message will wait in the queue
until an ERRORCLASS event is requested.

cnt = neturead(buf)
char *buf;        data area to drop the UDP packet (<512 bytes)
int cnt;            actual length of the packet which is ready
Reads the data from a UDP packet which was received.
Returns the number of bytes read into buf, or -1 if there is
no packet available to read.  Automatically clears the
incoming UDP buffer for the next read.

Session Layer routines

hostform.h
Include file which contains the structure of the machine
information record and the configuration information record.
Use these structures with the following calls to read or
manipulate machine specific information.  Includes struct
machinfo and struct config.
whatami.h
Currently contains the definition of PC vs. Mac in terms of
program and compiler settings to get the code to compile on
both the Mac and PC.  Also includes netevent.h.
netevent.h
Contains the list of event types, both low-level event types
used internally, and the events which the application
program will pay attention to.   See documentation section
on rules for event handling that explain what you can do
with events.


Sgetconfig(conp)
struct config *conp;    Pointer to a config structure which will receive a
copy of current data
Copies the config structure to user memory.  The hosts file
contains a lot of configuration information that the
application may need to know.  The structure of the data
area is in hostform.h.

mp = Shostlook(name)
char *name;        name of machine to get information about
struct machinfo *mp;    Pointer to machine information record
Takes a machine name (standard ASCII string) and returns a
pointer to the machine information record associated with
that name.  Searches the sname field first, then the hname
field.  Returns NULL if not found.  The structure definition
is in hostform.h.  Take care when writing to fields in this
structure (i.e. don't).  If you mess up the data, strange
things may happen.  Shostlook does  a simple lookup on the
name field, so the "default" record can be looked up if
necessary.

mp = Sgethost(name)
char *name;        name or IP number of machine to get information about
struct machinfo *mp;    Pointer to machine information record
Takes a machine name or IP number in ASCII format
(192.17.20.10) or a special shorthand form (#10) and returns
a pointer to the machine information record associated with
that name.   Uses Shostlook() to look up name if number is
not used.  Returns NULL if not found.  The structure
definition is in hostform.h.  Sgethost guarantees that the
record returned has a valid IP number which Snetopen() will
accept.  Sgethost should be the primary lookup call to
determine whether the domain name server is required or not.
If the IP number is given, a pointer to the "default"
machine is returned with a temporary IP number installed.
"default" cannot be returned as a real machine because
Sgethost guarantees a real IP number in the machine record.

mp = Slooknum(mnum)
int mnum;        Machine number to look for
struct machinfo *mp;    Pointer to machine information record
Used primarily for domain lookups, this number is unique for
all entries in the memory host list.

mp = Slookip(ipnum)
char *ipnum;        IP number of the host to look for
struct machinfo *mp;    Pointer to machine information record
Look up host information when you know the host's IP number.
Used by background ftp server.

mp = Smadd(name)
char *name;        ASCII name of the host to add to the memory host list
struct machinfo *mp;    Pointer to machine information record
Adds a machine to the host list, generating a unique host
number for it.  Copies the configuration information from
"default".  Returns a pointer to the machine record created.
Returns NULL if no memory could be allocated.  Be careful
updating the information in the machine record.  If the name
already exists somewhere as a session or host name, it just
returns that record pointer.

Shostfile(newname)
char *newname;    pointer to new configuration file name
Takes a pointer to static storage of the host file name.
The storage associated with the name must remain allocated.
Default is "config.tel".  Typically called with an argv[]
parm.

Snetinit()
Calls Sreadhosts(), netinit(), neteventinit(), and
Ssetgates() for you and initializes the timer queue.
Returns non-zero on error.  Sreadhosts() sets up the config
structure, so Sgetconfig() can be called after Snetinit().
RARP handling is done in Snetinit() - if RARP fails,
Snetinit returns -2 -- this condition requires a netshut()
to clear any interrupt drivers.  A -1 return means that the
netinit() never succeeded so the interrupt drivers were
never installed.

Sreadhosts()
Reads the hosts file, using the filename provided by
nethostfile().  Uses the new type of config file.  Sets
Smachlist to NULL before reading the file and creating the
list.  Processes and logs all information to the config
structure or into the machinfo list.  Calls some of the
setup options for ftp, rcp, screen modes, etc.  Called by
Snetinit, user does not need to call this routine.

Ssetgates()
This routine is called by Snetinit() automatically.  It sets
the netmask with the field read from the hosts file, turns
on ftp and rcp if they are specified in the hosts file, and
traverses the machinfo list to set up any gateways which are
marked.

pn = Snetopen(mp,tport)
struct machinfo *mp;    Machine information record obtained from Sgethost()
int tport;        TCP port to attempt connection with
int pn;            Port number passed through from netopen()
Tries to connect to the specified port on that machine and
returns the port descriptor for netread and netwrite.
Should be used instead of netopen().  Calls netopen() and
uses any special configuration information found in the
machinfo record.  Sets a timer for the connection timeout
which will be posted in an event.  You cannot access this
connection until you are notified with an event that shows
the connection opened or failed.

Sdomain(name)
char *name;        name of the machine to look up, s/b in DOMAIN format
Tries to use the DOMAIN name server to look up the IP number
of the named host.   Sdomain() posts appropriate events to
notify you of the results.  If the name does not have any
periods, the default suffix from the config structure is
added to the lookup.  All domain events return a machine
number which can be looked up with Slooknum() to find out
which machine is referred to.   Remember, the local host
file lookup is faster, so look up the machine name with
Sgethost() first.  Returns 0 on success, -1 if there is no
name server to query.

Snewns()
Rotate name servers.  If there is more than one nameserver
in the machine list, the one with the next highest number
will become the current nameserver.  Wraps to nameserver 1
when there are no more nameservers.

Ssetns(ipn)
char ipn[4];        IP number of nameserver to become primary nameserver.
For situations when your primary nameserver is not indicated
in the config file, but is obtained over the network
somehow, this sets that host up as nameserver #1.  The ns #1
set by Snewns() will rotate to ns #2 as necessary.

Stask()
Should be called instead of netsleep() to support the timer
queue and network updates.   Sgetevent() calls Stask() for
you.  The FTP server can't function without repeated Stask()
calls.

Stimerset(class,event,dat,howlong)
int class;        Class of event which should be posted
int event;        Event number which should be posted
int dat;            Associated data which should be posted
int howlong;        Seconds to wait before posting this event
Stimerset() adds events to the timer queue.  Stask()
contains the checks to post any events which have come due.

Stimerunset(class,event,dat)
int class;        Class of event which should be dequeued
int event;        Event number which should be dequeued
int dat;            Associated data which should be dequeued
Searches the timer queue for a matching entry and removes
the entry from the queue without generating an event.  The
timer may have just gone off with the event in the event
queue.  This procedure will not prevent that event from
occurring, only those which have unexpired timers are
removed.

theevent = Sgetevent(class,theclass,dat)
int class;        Classes to search for (OR combination)
int *theclass;        Actual class of the event returned (return value)
int *dat;        Data which tags the event
int theevent;        Which event has occurred
Sgetevent() does the background processing of ftp, rcp and
domain name lookup.  Events are posted with netputevent()
and are usually posted by lower layer routines.  Calling
Snetopen() will guarantee that you will get either a CONOPEN
or CONFAIL event depending on how things went.  The
associated data for CONXXXX events is always the port number
of the connection involved.  The routine will return 0 if
there are no events available.  Class types are in whatami.h
and the user may define new events and classes as described
under "events".
Currently, it is quite common to receive undesired extra
events which are for ports which you don't have defined.
Check the data tag for validity.

FILE *Sopencap()
Opens the capture file with the previously stored name.
Returns a file pointer or NULL if there is an error.  Use
Snewcap() to change the name.  Always opens for append.
Snewcap(name)
char *name;        Name to use for a capture file
Copies the name into private storage for use by Sopencap().
Limit of 80 chars on name length (no warning).

Stekmode(m)
Sftpmode(m)
Srcpmode(m)
Scwritemode(m)
int m;            Value of boolean flag to set
Sets modes and turns switches for the session layer.  Tek is
tektronix mode.  FTP and rcp turn the listeners on and off.
When writemode is 1 (true), direct writes to the screen
should be enabled.  Tek and writemode are simply services to
the upper layers, they only store the boolean flag.

Stmode()
Sfmode()
Srmode()
Scmode()
Returns the value of the boolean flag set above.

Snewpsfile(s)
Snewhpfile(s)
Snewtekfile(s)
char *s            new file name to use
For each of the graphic options, Postscript, tek and HPGL,
the configuration structure maintains a filename as a
service to higher layers.  These routines change those
filenames.  Pointers to the filenames are obtained through
the structure copy which is retrieved from Sgetconfig().

Scheckpass(user,passwd)
char *user;        User name to check the password of
char *passwd;        The password to check the validity of
Returns true if the passwd associated with the user field
matches the value in the password file associated with
telnet.  If there is a password file and the username is not
present, returns false.  If no password file is specified in
the hosts file, returns true.

Sneedpass()
Returns a boolean flag indicating whether we need to check
FTP passwords.  The determination is made depending upon
whether there has been a filename specified with the
passfile option in the config file.

Scompass(password,encrypt)
char *password;    The password to compare
char *encrypt;        The encryption to check against
Checks to see if the encrypted string has been encrypted
from a particular password string.   Returns true or false.

Sftpname(s)
char *s;        Space to copy the name of the file being transferred
The name of the current file being transferred by FTP is
stored internally.  Use Sftpname() to get a copy of that
internal name.

Sftphost(host)
char *host;        Space to copy the name of the host involved.
The IP number of the FTP client for FTP transfers is stored
internally.  Use Sftphost() to get a copy of that internal
number.  Then you can look up the host's name with
Slookip().

Sftpuser(user)
char *user;        Space to copy the username.
The name field from the FTP USER command is stored
internally.  Use Sftpuser() to get a copy of that name.

Sftpstat(bytes)
long *bytes;        Pointer to long, number of bytes
The value copied into this long int returns number of bytes
transferred or left to transfer in the currently active ftp
transfer.  Used to update status displays.



Events -- how to use them

Non-blocking code
NCSA Telnet is centered around an event structure similar to
the Macintosh toolbox event manager.  There are no blocking
routines in the lower layers of NCSA Telnet.  We don't
intend to ever add any blocking routines and we continue to
eliminate higher layer routines which block for user input.
A re-write of the user input routines which uses events to
map keystrokes is not out of the question.  Simply put,
anywhere that a programmer might want to put in code which
blocks to wait for something to happen, you will find an
event defined.
netgetevent() and Sgetevent()
The routines netgetevent() (at a low level) and
Sgetevent()(at the session level) look for events which are
stored in NCSA Telnet's event queue.  Sgetevent() calls
netgetevent(), so if you are using the session layer
library, you will never call netgetevent().  Sgetevent()
sifts through the events and captures events related to the
background name serving and ftp file transfers.  All other
events are passed through to the application.
Classes
Events are divided into classes to help different portions
of the application look for only certain kinds of events.
When you call one of the getevent routines, the class which
you pass to the routine is a mask of the OR combination of
the classes that you want to retrieve from the event queue.
When the routine returns with a greater than zero event
number, that mask will have been transformed into the exact
class to which the returned event belongs.  The classes are
defined below.
Events
For each class, several events are defined.  They are
documented individually, along with the use of the data
value which is returned from the getevent call.
User-defined events
Applications programmers are welcome to use class 0x80 (128)
for their own events.  Events in each of the predefined
classes in the range 128-255 are also available for use.
Specifically, use events 128-255 of the USERCLASS because
they will be more convenient to use.  Use of any events or
classes not in these ranges may conflict with future NCSA
software.  If you want certain events to be defined by NCSA,
let us know.  The official "defined events" list will be
updated periodically.
Timer events
Stimerset() and Stimerunset() control a separate queue from
the event queue.  The important thing to know is that when a
timer goes off, it places its specific event into the event
queue.  You may use user-defined event types to set timers
for yourself in this queue.  Some events like DOMFAIL and
CONFAIL are caused by timers if something doesn't occur to
unset the timers for those events.
minitel.c
Look at minitel.c for the simplest possible telnet.  This
one doesn't even have a telnet command parser and it only
talks to BSD hosts.  It does show you how to handle events
for the simple cases.

The following events are defined in netevent.h:


#define USERCLASS    1
#define ICMPCLASS    2
#define ERRCLASS        4
#define SCLASS    8
#define CONCLASS    0x10

#define ERR1        1    /* an error message is waiting, ERRCLASS */

#define IREDIR    1    /* ICMP redirect, ICMPCLASS */

#define CONOPEN     1    /* connection has opened, CONCLASS */
#define CONDATA     2      /* there is data available on this connection */
#define CONCLOSE     3    /* the other side has closed its side of the connection */
#define CONFAIL     4    /* connection open attempt has failed */

#define UDPDATA     1    /* UDP data has arrived on listening port, USERCLASS */
#define DOMOK        2    /* domain name ready */
#define DOMFAIL     3    /* domain name lookup failed */
#define FTPCOPEN     20     /* FTP command connection has opened */
#define FTPCLOSE     21     /* FTP command connection has closed */
#define FTPBEGIN    22     /* FTP transfer beginning, dat =1 for get, 0 for put */
#define FTPEND       23     /* FTP transfer ending */
#define FTPLIST     24     /* FTP file listing taking place */
#define FTPUSER      25     /* FTP user name has been entered */
#define FTPPWOK     26     /* FTP password verified */
#define FTPPWNO     27     /* FTP password failed */
#define RCPBEGIN     30    /* RCP beginning */
#define RCPEND     31    /* RCP ending */

#define UDPTO     1    /* UDP request from DOMAIN timed out, SCLASS */
#define FTPACT    2    /* FTP transfer is active, keep sending */
#define TCPTO      3    /* TCP for DOMAIN timed out */
#define RCPACT     4    /* rcp is active, needs CPU time */
#define RETRYCON     5    /* retry connection packet, might be lost */



Class USERCLASS
Event UDPDATA
To get this event you must have already called netulisten
with a port number.  A UDP packet with some data has arrived
for you on that port.  The data field for the event is the
UDP port number which you were listening to.  In the future,
you will be able to listen to multiple ports and the data
field will become important.  Port 998 is arbitrarily used
by our domain name server.
Event DOMOK
To get this event, you must have already placed a domain
name request.  You may get more than one DOMOK for one
request.  The data field contains the machine number of the
name you looked up.  You can look up the machine record with
Slooknum() now.  The IP number has been placed into the
record.  A Snetopen() with that machine record should
succeed now.
Event DOMFAIL
To get this event, you must have already placed a domain
name request.  You may get more than one DOMFAIL for one
request.  The data field contains the machine number of the
name you looked up.  That machine name could not be
resolved.   You may also have an error event waiting for you
which explains why the resolve failed.
Event FTPCOPEN
A command connection has been established to your background
FTP server.  NCSA Telnet uses this to post informative
messages.
Event FTPCLOSE
The command connection for your server FTP has closed.
Event FTPBEGIN
A file transfer to or from your machine has begun.  The data
field=1 for get (transfer from) and 0 for put (transfer to)
transfers.
Event FTPEND
A file transfer or a LIST command has ended.
Event FTPLIST
A LIST command has begun for your background FTP.
Event FTPUSER
A USER command has been entered for the background FTP.
Event FTPPWOK
An FTP PASS command has been entered and verified by
comparing to the entries in the password file.
Event FTPPWNO
An FTP PASS command has been entered and rejected by failing
to match anything in the password file.
Event RCPBEGIN
Someone has started an rcp transfer to or from your
computer.
Event RCPEND
That rcp has ended.
Class ICMPCLASS
Event IREDIR
The netsleep() routine automatically does a netgetevent()
for ICMP redirect messages and performs the addressing
change required by the ICMP redirect.  When an ICMP redirect
packet is received, this event is posted by the ICMP
routine.  Bugfix in 2.2 makes sure ALL redirects are
registered.
Class ERRCLASS
Event ERR1
There is currently only one error event which pertains to
all of the possible errors.  These events should be
reclassified as ERRWARNING, ERRINFO, ERRFATAL, etc., so the
rest of this class is reserved.  The data field contains the
error number which can be looked up with neterrstring().
Class SCLASS (session class)
The entire SCLASS is for special session layer events.  You
will not receive these events from Sgetevent() because they
are always removed.
Event UDPTO
The current UDP domain request timed out.
Event FTPACT
When FTP is active, it uses as much CPU as it can get.  To
get more CPU it posts events back to itself.
Event TCPTO
not used, we don't need TCP lookups.
Event RCPACT
rcp posts this event back to itself when it needs CPU.
Event RETRYCON
Snetopen() posts this in a timer to send multiple SYN
packets if the first one is lost.
Class CONCLASS
Event CONOPEN
A connection has just opened.  This may be a connection
which you are listening to, or one which you are trying to
open with Snetopen().  Remember Snetopen() does not block
until the connection is open.  The data field returns the
port number (descriptor) of the connection.
Event CONDATA
Data has arrived for a certain port number.  The data field
contains that port number.  Because this event always comes
up when there is data in the incoming TCP buffer, you never
have to poll netread() to find out if there is data to be
read.  If you get CONDATA events, there is data to be read,
if you are not getting CONDATA events, then there won't be
data to be read.  Unless you drop an event, of course.
CONDATA events which arrive after a CONCLOSE event or a
netclose() call may get a -1 return from netread(),
indicating that the connection has no more data or does not
exist.
Event CONCLOSE
If the host you are talking to decides to drop the
connection, or reset it, then you may get a CONCLOSE event.
This can be interpreted as the last CONDATA event that you
will get.  There may be a large amount of data waiting for
you to pick it up with netread() even after receiving a
CONCLOSE event.  The correct procedure is to take data from
netread() until you receive a 0 or -1 return code and then
call netclose() to finalize your side of the connection.
Event CONFAIL
The timer ran out on a connection which you tried to open.
This does not mean that you are done.  You must netclose()
to allow the memory for that port to be re-used.


Virtual Screen Kernel Calls

Gaige B. Paulsen                            October 31, 1987
                                    updated July 14,
1988

int VSinit(max)
int max;        Maximum number of Virtual screens to allow.
Initializes the Virtual Screen Kernel for use with up to max
screens.
Returns 0 if successful.

VSscrn *VSwhereis(i)
int i;            Virtual Screen to examine.
Returns the current address of the virtual screen control
record.  This is used for debugging.

int VSnewscreen( maxlines, screensave, maxwid, IDC)
int maxwid;        How wide can this window be.  Only tested for 80 and 132.
int maxlines;        Maximum number of lines to save.
int screensave;    Is scrollback enabled at start?
int IDC;        Do we have support for insert and delete of characters.

Creates a new screen if possible.   VSnewscreen returns a
number >=0 if it succeeds, which is to be used as the window
parameter(w) in all subsequent calls to vs routines.  If the
return value is <0, there was an error, and no screen was
created.

VSdestroy(w)
int w;            Window number to destroy.
Destroy the window corresponding to w.  Returns an error if
applicable.

VSdetatch(w)
int w;            Window number to detach.
Detach the window corresponding to w.  Returns an error if
applicable.  Note that this function currently just calls
VSdestroy(w).  Previously and hopefully sometime in the
future as well, it did/will allow for the destruction of a
screen without freeing memory.

VSredraw(w,x1,y1,x2,y2)
int w;            Window number to redraw.
int x1, y1,x2,y2;    Bounds of the rectangle to be drawn, in local window
coordinates.
Redraw the portion of window w which lies within (x1,y1) -
(x2,y2).  The coordinates are local to the window and should
NOT be adjusted for scrollback, as most of scrollback is
handled by the VS routines and not the RS.

VSwrite(w,ptr,len)
int w;            Window number to write to.
int  len;        How many characters to write.
char *ptr;        Where to get the characters.
The big routine.  This one is called to write a character
stream to the virtual screen w.  VSwrite calls VSem to parse
the emulation.

VSclear(w)
int w;            Window number to clear.
Clear the virtual screen of window w.

char VSkbsend(w,k,echo)
int w;            Window number to send the characters from.
int echo;        Boolean - whether to local echo the key.
unsigned char k;    VT-100 key code to send
Sends the VT-102 representation of the key represented by k.
This routine uses somewhat of a hack by calling RSsendstring
which is supposed to know how to send characters to
whichever port corresponds to window w.  The echo flag only
applies to special keys (like arrows) and will cause them to
echo on the local screen as they are transmitted.

VSclearall(w)
int w;            Window number to clear all of.
This routine currently does nothing.  It was originally
intended clear the current screen and all of the scrollback
as well as setting the top of the scrollback to the top of
the available storage space, but this was deemed
unnecessary.....at least for now.

VSreset(w)
int w;            Window number to reset.
Resets the virtual screen (w) with respect to VT-100 modes.
Changes the wrap mode to off and clears the screen as well
as setting all other modes to their ORIGINAL POWER-ON
defaults as specified by the VT-100 programmers manual.

char *VSgetline(w,y)
int w;            Window number to get line from.
int  y;            Line from which to get the pointer
This routine in now out of date and SHOULD NOT BE USED
unless you really know what you are doing.  It retrives a
pointer to the data string associated with line (y) of the
screen (w).  This ONLY works with lines that are on the
active vt100 screen (lines 0-23).

VSsetrgn(w,x1,y1,x2,y2)
int w;            Window number to set the display region for.
int x1,y1,x2,y2;    The vs's bounding rectangle in GLOBAL VS coordinates .
Set the current display region for window w to (x1,y1)-
(x2,y2).  This will perform
scrollback/scrollforward/scrollleft/scrollright/etc. as
necessary to make the current screen reflect the values of
the passed display region.  It restricts you from doing
anything really stupid (I hope).

VSgetrgn(w,x1,y1,x2,y2)
int w;            Window number to scroll forward in.
int *x1,*y1,*x2,*y2;    Where to retrieve the display region into.

Get the current value of the display region (as set by
setrgn and/or modified by the scroll routines as well as
auto--scroll).  Pass pointers to the integers you would like
to have the values put into.

VSscrolback(w,in)
int w;            Window number to scroll back in.
int in;            Number of lines to scroll back.
Scroll back in lines in window w if possible.  Note that
this routine will prevent you from scrolling back beyond the
top of the buffer, so it is safe to call it whenever you get
a request to scroll back further, even if you don't know if
there is data left to scroll back into.

VSscrolforward(w,n)
int w;            Window number to scroll forward in.
int n;            Number of lines to scroll forward.
Scrolls forward n lines in window w.  See VSscrolback for
details.

VSscrolright(w,n)
int w;            Window number to scroll right in.
int n;            Number of columns to scroll right.
Scrolls right n columns in window w.  See VSscrolback for
details.

VSscrolleft(w,n)
int w;            Window number to scroll forward in.
int n;            Number of columns to scroll left.
Scrolls left n columns in window w.  See VSscrolback for
details.

VSscrolcontrol(w,scrolon,savescroll)
int w;            Window number to set scroll variables for.
int scrolon;        Do we scroll or don't we?
int savescroll;        Save cleared lines into scrollback.
Changes the value of the scrolon variable in window w,
therefore disabling and enabling scrollback, not really
necessary.  If savescroll is on, all clearscreen commands
will load the text into the scrollback buffer.

VSsnapshot(w)
int w;            Window number to take snapshot of.
Currently does nothing.  Will/should eventually make a copy
of the current screen into the scrollback buffer.  This is
most easily done by sending 24 line feeds and then copying
the values of lines (-24) - (-1) into 0-24.  But, I hope to
do this before  2.1 and the source release.

int VSmaxwidth(w)
int w;            Window number to retrieve the maximum width of.
Retrieves the current maximum width of window w.  Used
primarily to get the maximum possible size of a selection
for copying. NOTE: a 132 column window in 80 column mode
returns 80.

long VSgettext(w, x1, y1, x2, y2, charp, max, EOLS)
int w;            Window number to get text from.
int x1,y1,x2,y2;    The starting and ending points of the text in global VS
coordinates .
char *charp;        Pointer to the destination of the gotten text.
char  *EOLS;        Pointer to the string to be used for end of line.
long max;        Maximum number of characters charp can hold.
Copies a section of text from window w, which corresponds to
an area starting from the x1th character of line y1 to the
x2th character of line y2, inclusive of all of the
characters on lines y1+1 to y2-1  into a buffer pointed at
by charp to exceed no more than max and inserting the string
pointed at by EOLS whenever there is a line transition.
(...whew...) This is mainly used for copy and print of VS
data.  Extraneous blanks are ommitted at the end of lines by
searching backward to the beginning of the line.

Global  VS coordinates vs. local coordinates

Global
Global coordinates are used to represent a range of
information that can span the entire VS space as it
currently exists.   To accurately portray the area, we must
have some standard method of referring to scrollback.  For
this, we have chosen to refer to the top of the currently
active VT-102 screen as line 0, the bottom of the active
screen as line 23, and the lines of scrollback as negative
offsets from line 0.  Hence, a line that has been scrolled
off by a CR-LF would be seen as line -1.  To perform actions
such as scrolling back to the top of the virtual screen, you
would ask for lines from (top of buffer) to (top of buffer
+length of display window).  Similarly, if you wished to
scroll to make the right hand side of a 132 column window
visible, you would ask for the columns from (maxwidth -20)
to (maxwidth).
Local
Local coordinates are used to represent things that happen
relative to the area being displayed by the Virtual Screen.
An example would be a portion (or all) of the virtual screen
being made visible.  This action would cause whichever lines
are in the visible area to be redrawn, therefore you will
need to inform the VS routines to do this.  Since the VS
routines attempt to maintain as much of the information
about scrollback by themselves so as not to burden you with
extraneous information, you pass these coordinates as local
to the currently active screen.  Therefore, a full redraw
would be line 0 character 0 through line 23 character 79,
which would then be translated by VSredraw to whatever
global coordinates are appropriate.

Replacing the Terminal Emulation package in the VS
routines.

The VS routines can be used as a basis for just about any 24 line screen
terminal emulator.  This document shows how to use the internal routines in
vsintern.c to construct a  vsem.c file as a basis for a new terminal emulator.

VSem(c,ctr) -
char *c     pointer to the buffer to use as input to the emulator
int ctr        how many characters to read from the input buffer

This is the main emulation routine.  It is the entry point from VSwrite and
should be the only thing that needs to be replaced when changing the terminal
emulation.

    escflg: current escape state
        0 - normal character wait
        1 - have escape, waiting for next character
        2 - escape '[' sequence, waiting for parameters
        3 - escape '#' sequence
        4 - escape '(' sequence
        5 - escape ')' sequence
    c:     pointer to character currently being parsed
    VSIw: points to the window structure for the current window.
    VSIwn: is the current window number

VSI routines from vsintern.c for use in emulation

VSIreset()            - Cause a terminal reset
VSItabinit()        - Initialize tabs to default stat (every 8
        characters)
VSItabclear()        - Clear the tab at the current position

VSIdellines(n,s)        - delete n lines at s (current line if s<0)
VSIinslines(n,s)        - insert n lines at s (current line if s<0)

VSIeeol()            - erase to end of line
VSIebol()            - erase to beginning of line
VSIel(s)            - erase the entire line s (current line if s<0)
VSIeeos()            - erase to end of screen
VSIebos()            - erase to beginning of screen
VSIes()            - erase the whole screen

VSIrange()            - check and resolve range errors for x and y

VSIdraw( VSIwn, x, y, a, len, c)
                - draw len characters at (x,y) in attribute a
                    with the chars at c
VSIinsstring(len,start)    - insert len characters at the current position
            with the chars at start

VSIdelchars(x)        - delete x characters at the current position
VSIinschar(x)        - insert x blank characters at the current position
VSItab()            - move to next tab stop
VSIsave()            - save current x, y, a
VSIrestore()            - restore previous x,y,a

VSIindex()            - index one line
VSIrindex()            - reverse index one line

VTsendpos()            - Send the VT-100 cursor report
VTsendstat()            - Send the VT-100 status report
VTsendident()        - Send the VT-100 identity report
VSIsetoption(toggle)    - Set the VT-100 option depending on toggle



How to write a new PC driver

There are six routines which must be provided in a new
driver.  All of these routines are currently provided in
assembly language in one file.   When you write a new
driver, you will probably take a current driver and replace
each of the six routines in the file.  The current examples
are in the ENET subdirectory and the three best examples
are:

NET501.ASM
3COM 3C501 driver.  Uses interrupts,  and will be a
reasonable template for other interrupt-driven drivers.  I
highly discourage trying to improve this driver for the
3C501.  I've tried.
NETUB.ASM
Ungermann-Bass (IBM) NIC board driver.  Does not use
interrupts.  This driver may be a good template for simple
shared-memory boards which divide the memory into pages.
NET5210.ASM
MICOM NI5210 driver.  The NI5210 uses the Intel 82586 chip.
You will need the Intel databook for this chip in order to
understand the driver.  This driver should be directly
applicable to all boards which are built around the 82586.
Works with 8K or 16K boards.  Only uses 8K of a 16K board,
so save your money.

With dual-ported RAM and intelligent chipsets, interrupts are entirely
unnecessary.  The board will automatically buffer a certain number of packets
which are picked up when NCSA Telnet has time to process incoming packets.
TCP takes care of preventing overruns.

The six routines are described below.  They generally return -1 on error.
getaddr() is usually called first and etopen must be called next for any of
the other routines to make sense.  Function pointers for the routines are
declared and installed in PCTOOLS.C.  A new driver should install new pointers
right alongside the existing ones.

etopen(eaddr,irq,addr,ioaddr)
char  eaddr[6];        Hardware ethernet address to use on board.
int irq;            Interrupt request level from configuration file.
int addr;        Shared memory base segment address from configuration file.
int ioaddr;        I/O address from configuration file.
This routine must do all of the initializations for the
Ethernet board.  Any interrupt drivers must be placed,
interrupts turned on, base register variables set up,
packets initialized and the address set for the board.  Do
not return until you are receiving packets.  Most drivers
will ignore some of the fields provided.  Map them to
whatever values are appropriate for your hardware.

getaddr(eaddr,addr,ioaddr)
char eaddr[6];        Buffer to place address.
int addr;        Base segment address for shared memory
int ioaddr;        Base I/O address for board
Some of these parameters are also ignored, depending upon
the hardware.  The hardware address from the board's EPROMs
should be copied into the eaddr buffer.  This address is
generally given back to etopen, so getaddr cannot depend on
etopen's initialization.

etclose()
Shut down Ethernet board.  Primarily to turn off interrupts.
This routine can be just a RET for non-interrupt boards.

recv()
Receive a packet.  Used only for boards that do not use
interrupts.  The interrupt handler generally replaces the
need for recv().  The interrupt handler or recv() must use
my undocumented buffering scheme.  Generally, keep the same
code and replace the board handling side.

xmit(packet,size)
char *packet;        Address of packet to transmit.
int size;        Size of packet to transmit, including all headers.
Send out one packet.  This is generally not interrupt
driven.  The packet size may be smaller than the minimum
Ethernet packet size.  If so, this routine must take care of
the details.  The frame to transmit always includes the
entire Ethernet frame header, so non-802.3 drivers may need
to strip the header to modify higher layer code.

etupdate()
Remove one packet from the buffer.  This routine is
generally unmodified from driver to driver.  The incoming
packet is not copied from the buffer until fully processed,
so the driver must wait until etupdate() is called before
freeing the space back to the buffer.