pptpd/TODO

This is the pptpd project TODO list.

(Note: tasks that are probably easy are marked with a [JJ], which
means Junior Jobs, following the KDE tradition suggested by Adriaan de
Groot in May 2004.)

20060714-1, 1.3.2, bcrelay is built even if --enable-bcrelay is not used.

20050105-1, 1.2.1, denial of service, for ten seconds (or --stimeout)
the pptpd will not process further connections while it waits for a
client to send data.  The second select() will prevent any further
accept() calls.  To be tested.

20041108-0, 1.2.1, vpnstats does not parse RHEL3 log output
successfully.  It also creates a cron.weekly entry that is not removed
when the package is removed.  Yet to test if vpnstats.pl works okay.
[JJ]

20040811-0, 1.2.1, RFC2637 timers IDLE_WAIT and MAX_ECHO_WAIT, in
pptpdefs.h, make these configurable, requested by Tobias Brox.  [JJ]

20040623-0, 1.2.1, "Plugin /usr/lib/pptpd/pptpd-logwtmp.so is for pppd
version 2.4.2, this is 2.4.3" happens if pppd 2.4.3 is installed, or
vice versa if 2.4.2 is installed.  [JJ] Workaround on common
distributions is to use makepackage instead of "make install".
pptpd.init is also affected, as it has a hardcoded path to pptpd.
Configure script should detect or accept version of pppd to build
against.

20040521-4, 1.2.0-b3, running on Fedora Core 1, kernel ? , when a
connection arrives a warning is issued by the kernel "application bug:
pptpctrl($PID) has SIGCHLD set to SIG_IGN but calls wait().  (see the
NOTES section of 'man 2 wait').  Workaround activated."

20040521-0, 1.2.0-b3, write bcrelay man page.  (Debian lintian for
pptpd 1.1.4-0b4-1)  [JJ]

feature: nuke prior login by user, by writing a new pppd plugin.

feature: prevent login by user if prior login still in progress, using
pppd plugin that captures old hook and re-uses it, during chap auth.

20040428-7, CVS HEAD, minor, pptpctrl.c refers to hard coded path
/usr/lib/pptpd/pptpd-logwtmp.so instead of path derived from prefix
given to configure script.  also, the rpm build fails if not run in a
fakeroot.

review debian bug reports for pptpd package
http://bugs.debian.org/pptpd

20040426-0 bug: terminating immediately on CALL_CLR_RQST from client
(1.2.0-b3) may violate the RFC, but was done to fix zombie pppd, Bruno
will try older version to gather system state information to see why
it happened, but there is indeed a design defect that doesn't reap
(wait() on the PID of) the pppd it had previously forked.  it probably
isn't right to simply kill() the pppd until it has had a chance to
finish up, e.g. ip-down scripts.  context is pptpctrl.c.  on the other
hand, if a stop control connection request arrives, are the active
connections supposed to be closed; there are reports of pppd living
on.

minor bug: pptpctrl.c:302 (1.11) there is a test for absolute file
descriptor numbers, which is probably not portable.

bug: 1.2.0-b3 configfile.c opens, reads and closes /etc/pptpd.conf as
many times as there are possible option values, and only returns the
first match for a keyword, not the last.

20030214, ignore EPERM on GRE write for a short time in case netfilter
is being adjusted by ip-up scripts, Bug #181005.

20030801, 1.1.3, XP seems to need mtu negotiation 4 lower than
serverside.  MS PPTP-server handles this correctly.

20030620, 1.1.3, PPTP daemon waiting for infinite time if the
pptpd.conf file does not have a new line character at the EOF. So we
need a config-check routine. It is allready there, but it seems to
output to stdout.

20030611, 1.1.3, pptpd incorrectly hides all pppd error output.  This
is a defect and should be fixed. It apparently doesn't even wait for
and report the pppd exit status.

20030526, 1.1.3, Make choice between slirp, pppd and BSD userspace
PPPD a runtime one instead of compile time.

20030218, 1.1.3, ask IP-info from a real DHCP-server. It is possible
to get a Radius-server to do this.

20030128, 1.1.3, Make the logentries more verbose, so that they
actually mean something to the admins.

20021220, 1.1.3, Add a way to define pools for different usergroups.

19990802, 1.1.3, We don't actually implement any kind of window
protocol, congestion control, etc, etc.  Very simple algorithm which
works well on LANs and good performance links but could be improved
for slow distant dialups/bad routing and links.

19990705, 1.1.3, PNS mode and all things which go with it aren't
supported properly.  We're only an IP tunnelling PAC, and our clients
act as a PNS, as per the draft.  We may implement PNS mode and related
things like multiple calls from one client IP in the future
(post-1.0).

19990413, 1.1.3, SET LINK INFO (CTRL Message 15) is not fully
implemented. This doesn't cause any problem with Win95/98/NT or Linux
clients.  NT clients send a default set of ACCMs which we validate and
ignore, the others don't send anything.  Probably fine for all IP
tunneling clients (post-1.0).