Changes:
***
*** Wednesday, September 3, 2014 -- Dante v1.4.1
***
o Fixed configure problem with detection of libc.so on recent FreeBSD
versions.
Problem report and testing by Mikhail <mp39590@gmail.com.example.com>.
o Fixed problem in configure test used for figuring out which
buffer in an IPC socket is used for determining bytes that can
be buffered. Should fix problem on FreeBSD.
Diagnosed in detail by Mikhail <mp39590@gmail.com.example.com>.
Also reported by Carlos Jacobo Puga Medina <cpm@fbsd.es.example.com>.
o Fixed bug in handling of buffered GSSAPI data.
o Added keywords for specifying that only IPv4 or IPv6 should be used
on internal or external interfaces.
o Fix bug that prevented external.rotation: same-same from working correctly.
Reported by Alexander.V.Makkoveev (makkoveev@gmail.com.example.com).
o AIX 6.1 compilation fixes.
o NetBSD 6.1.3 gethostbyaddr() configure fix.
Problem reported by Anthony Howe <achowe@snert.com.example.com>.
o Use wait3(2) rather than wait4(2), as the latter (implemented as a
wrapper on Solaris) has a bug on Solaris in the handling of the WAIT_ANY
value. This is -1 traditionally, but on Solaris wait4(2) is expecting
it to be the value 0. This change also fixes a similar problem on AIX.
o Remove extra CRLF from CONNECT request, as in patch submitted by
"Dyson, Brian J" <Brian.Dyson@ca.com.example.com>.
***
*** Monday, November 18, 2013 -- Dante v1.4.0
***
o Older syntax for setting the socket buffer sizes via
socket.recvbuf.udp, socket.sendbuf.udp, socket.recvbuf.tcp and
socket.sendbuf.tcp options have been deprecated and replaced with
the more general API for setting socket options.
"socket.recvbuf.udp" can now be set by "<interface-side>.udp.so_rcvbuf"
"socket.sendbuf.udp" can now be set by "<interface-side>.udp.so_sndbuf"
"socket.recvbuf.tcp" can now be set by "<interface-side>.tcp.so_rcvbuf"
"socket.sendbuf.tcp" can now be set by "<interface-side>.tcp.so_sndbuf"
<interface-side> refers to either "internal", for Dante's internal
interface(s), or "external", for Dante's external interface(s).
o socks-rules now require a "socks" prefix (like client-rules require a
"client" prefix), and the socks "method" keyword has been renamed to
"socksmethod".
o The "socksmethod" keyword can now be set in client-rules too. It
is used to override the default preference for what socksmethod to
select for which clients addresses, making it possible to by
default e.g., have the preference "gssapi username none", but for
some client-ranges have a different preference, e.g., "none
username gssapi".
Normally there is no need to use set this keyword in a client-rule.
o Fallback to direct (non-proxy) routes now defaults to off in the client,
as well as in the server.
To keep previous behaviour in the client, with direct route fallback
for destinations with no matching route, set SOCKS_DIRECTROUTE_FALLBACK
to "yes" in the environment, or ./configure with --enable-drt-fallback.
Direct fallback is enabled if there are no routes configured (as is
usually the case in a server configuration), and disabled otherwise.
o IPv6 is now supported in the server.
Standard IPv6 address syntax is used for addresses, with the addition
of the special address "0/0" used for matching both all IPv4 and
all IPv6 addresses.
o New "monitor" object added. Syntax is similar to rules and routes,
but instead of applying to individual sessions, it applies to all
sessions currently matching the addresses to monitor.
This can be used for monitoring network anomalies related to too
little data being transferred or too many disconnects occurring,
triggering alarms if detected.
o Possibility to configure system errors and DNS-errors for
special logging in certain cases (when connecting and performing
hostname resolving).
o More aggressive regarding how many processes to fork when starting
and how many processes to reserve for future clients.
o Added new log keyword: "tcpinfo". Used to report more extensive
statistics about sessions, including TCP_INFO on supported platforms.
o SIGINFO/SIGUSR1 output is now logged at level "info" instead of level
"debug".
o SIGHUP code rewritten. Should function considerably better in
environments were SIGHUP is, for whatever reason, sent an excessive
amount of times every second.
o Default for the maximum number of clients an i/o process can handle has
been increased from 8 to 32.
o Improved UDP compatibility by sending appropriate ICMP unreachable
errors to clients and targets, if running with the appropriate
privileges (typically, root is required for this).
Makes it possible for a client or target to be notified that a UDP
packet it wanted the Dante server to forward was not forwarded.
o Reduction in memory consumption at the expense of allocating extra
memory dynamically in the very rare, perhaps non-existing, cases
where the extra memory is needed.
o The default timeout for TCP i/o (timeout.io.tcp) has been changed from
84000 to 0. 0 means use the kernels default, which in most cases will
mean no timeout. See UPGRADE for more information.
o The session module has been merged with the mainstream Dante code,
and has also been extended to support the following new features:
- connection throttling (number of new sessions accepted per second).
- state-keys.
Two state keys are currently supported:
- per-IP address.
- per-hostid (hostid is supported on certain platforms, with
certain kernel patches, with certain clients).
The syntax has also changed (see UPGRADE).
See the manual for more information about the new features.
o Code used for finding the correct outgoing address to bind when
external.rotation is set to "route" replaced with much simpler,
but hopefully equally (or better) functioning code.
Idea taken from Quagga.
o Use getpassphrase() rather then getpass() to obtain password for
username authentication when available. Avoids 9 character limit
on Solaris. Suggested by Albert Fluegel <af@muc.de.example.com>
o Use sqrt() rather than sqrtl() in stddev calculation, as sqrtl() is not
available on some platforms (such as FreeBSD 7.2). Problem reported
by Rudolf Polzer <rpolzer@one-it.de.example.org>.
o If the authentication method used was RFC931 (ident), the username
was not always logged when it should be.
Reported by Gregory Charot (EVENIUM) <gcharot@evenium.com.example.com>.
o Syntax checking has been improved to better detect invalid or
likely incorrect server configurations.
This can result in some configurations that have previously been
accepted or accepted with warnings by Dante, to now cause an error
on startup, preventing the Dante server from starting up until the
configuration error has been fixed.
o Fixed compilation on OpenBSD with compilers not supporting -Wbounded.
Problem report and testing by
Mikael More <mikael.more@gmail.com.example.com>.
o GSSAPI "clear" is no longer enabled by default, as it is not part
of the SOCKS GSSAPI standard per se.
o external.rotation was not handling non-IPv4 target addresses correctly.
Reported and diagnosed by Rudolf Polzer <rpolzer@one-it.de.example.com>.
o The "--disable-libwrap" option has been renamed "--without-libwrap".
o Fixed bug that would cause the following warning to sometimes be
erroneously reported:
"warning: accept(2) failed: Resource temporarily unavailable"
o SIGINFO log information extended to include information about i/o
buffer status and as well as UDP packet latency.
o Real-time scheduling priority settings and CPU affinity settings made
available in sockd.conf.
See https://www.inet.no/dante/files/dante_realtime_preview.pdf for
a performance analysis done in relation to these new features.
o General API for setting socket options on sockets used by Dante
made available in sockd.conf.
o Support for cross compilation of client library for Android
(system name 'arm-linux-androideabi').
Testing and analysis by Yoav Weiss <weiss.yoav@gmail.com.example.com>.
o Problem with sockd.init generation in dante.spec fixed.
Reported by Luiz Gustavo Nascimento <luizgn@gmail.com.example.com>.
***
*** Thursday, August 4, 2011 -- Dante v1.3.2
***
o Part of the possible resource optimization indicated in section 4.4
of the Dante 1.3.1 performance evaluation report
(http://www.inet.no/dante/doc/1.3.x/dante_performance_1.3.1.pdf)
has been implemented: scheduling new clients to the Dante process
with fewest free slots available.
o If a proxy server (socks, http, or upnp) is configured via
environment variables, don't attempt to also parse any socks.conf
files.
o The "HTTP_PROXY" environment variable has been renamed to
"HTTP_CONNECT_PROXY" to avoid conflict with HTTP proxies that do not
support the HTTP CONNECT request.
o Fixed an unintended change to ACL semantics regarding bind requests in
Dante 1.2.3.
Noted by Ralf Wenk <Ralf.Wenk@hs-karlsruhe.de.example.com>. Many thanks.
o Fixed a bug regarding handling of cases where the ulimit for max number
of files is unlimited. Reported and analyzed by
Ralf Wenk <Ralf.Wenk@hs-karlsruhe.de.example.com>.
o Fixed a bug introduced in 1.3.1 related to privileges during startup
on Solaris.
***
*** Tuesday, June 21, 2011 -- Dante v1.3.1
***
o Fixed a bug regarding handling of socks clients that do not follow the
spec and start sending traffic data before request has been granted.
o Fixed an error related to ACL on UDP-packets. If a given SOCKS
client was allowed to send some UDP packets, in some cases packets
from the same SOCKS client that should have been blocked were allowed.
***
*** Tuesday, June 14, 2011 -- Dante v1.3.0
***
o Fixed a bug related to parsing the HTTP_PROXY environment variable.
Problem reported by John Fletcher <john.fletcher@rd.bbc.co.uk.example.com>.
Note that the format of the HTTP_PROXY variable has also changed.
o Logfiles can now be added or changed in sockd.conf after startup.
o The priority of proxy routes has changed, so that a SOCKS v4 route is
now preferred over a SOCKS v5 route when both can be used, because
v4 requests can be performed faster than v5 requests.
o Support for a new module, providing LDAP based authorisation.
Developed by Markus Moeller (markus_moeller at compuserve.com).
Note that modules need to be purchased separately.
See "http://www.inet.no/dante/module.html" for more information about
Dante modules.
o Resource requirements have been significantly reduced.
Performance has also been improved.
o When checking whether a username is member of a group, as used in
in group-based authentication checks, also check against the
primary groupname of the user, as listed in /etc/passwd or similar,
rather than only against the entries in /etc/group or similar.
Based on a patch from Ralf Wenk <Ralf.Wenk@hs-karlsruhe.de.example.com>.
Thanks.
o The socket buffer sizes can now be set in sockd.conf. The default
value is 0, which lets the kernel choose the buffer sizes itself.
o A new keyword has been added to the configuration files: "errorlog".
The syntax is the same as for the "logoutput" keyword, but
"errorlog" is used to specify how errors, and only errors, should
be logged.
The meaning of the "logoutput" keyword has not changed, and will
still log errors if configured to do so.
o "child.maxidle: yes" is now the default setting, meaning most idle
sockd child processes will be terminated automatically.
o A bug that prevented binding privileged ports in the client from working
has been fixed.
Bug reported by Christophe Réquillart <requillart@gmail.com.example.com>
o Two new timeout parameters have been added to sockd.conf:
- "timeout.connect":
This controls how many seconds the server will wait for a connect
initiated on behalf of the SOCKS-client to complete. The default
is 30. Setting it to 0 will use the kernel default.
- "timeout.tcp_fin_wait":
This controls the equivalent of TCP FIN-WAIT-2.
Note that FIN-WAIT-2 is a normal TCP state. Having many tcp
connections stuck in this state does not indicate something is wrong,
but it can create a resource problem for the kernel if there are
too many of them.
The default value is 0, meaning, let the kernel handle it as before
rather than for Dante to time out sessions in this state.
o "timeout.connect" is now supported also in socks.conf. The value
overrides the default timeout when the client attempts to connect
to the SOCKS server.
o The different server timeouts (negotiate, connect, i/o) can now be
set on a rule-by-rule basis, and not just globally for all rules.
o The config.h variables MAX_ROUTE_FAILS and BADROUTE_EXPIRE can now
set in the configuration files instead of at compile time.
The new values are "route.maxfail" and "route.badexpire" and control
how to handle bad routes (to e.g., a SOCKS/HTTP proxy server).
o Support for a new value for the "external.rotation" keyword: "same-same".
This will tell the Dante server to use the same address for outgoing
connections as the address the SOCKS client was accepted on, i.e.,
one of the internal addresses.
Based on a patch from Lysenko Konstantin <gshaud@gmail.com.example.com>.
Thanks.
o Fixed a bug where we could end up exiting if we got a SIGHUP
at a point where we had no more free file descriptors, and
thus could not reopen the sockd.conf.
Problem reported by Mark Sinner <mark.sinner@db.com.example.com>.
o The most time consuming part of the work performed by the request
children (waiting for connect(2) to complete) has been moved to the
i/o children.
This should reduce the total number of sockd processes necessary to
handle medium and higher loads, since the request children can only
handle one client at a time, while the i/o children can handle many.
o The "srchost" parameter "nomismatch" has been renamed to "nodnsmismatch",
while "nounknown" has been renamed to "nodnsunknown".
The "nodnsmismatch" and "nodnsunknown" functionality is now
implemented internally in sockd, rather than via libwrap; there is
no longer any dependency on libwrap ("tcp_wrappers") when using
these options.
o BSD authentication support on OpenBSD.
o Option ordering in sockd.conf is now enforced, rather than just
recommended. Server options now always have to come before rules
and routes.
o The file to write the pid to can now be specified on the command line
using the "-p" option.
o If TMPDIR is not set, temporary files will be created in the
current working directory when sockd is started. If this fails,
temporary files will be created in "/tmp" as before.
o A 64-bit version of libdsocks is now built on 64-bit OpenSolaris machines.
Should make socksify work with 64-bit applications.
o Fix compilation on machines with Linuxthreads. Locking via
Linuxthreads is however not supported.
Problem report and testing by
Nicolas VANHAUTE <nicolas.vanhaute@ac-clermont.fr.example.com>
o Fix usage of negation operation in shell scripts causing problems
for socksify when using Bourne shell on Solaris 8.
Problem reported by Sebastian Kayser <skayser@opencsw.org.example.com>.
o Size of byte and packet counters used for logging changed to 64 bits.
o IBM Visual Age C/C++ V8 compilation fixes for AIX.
Problem reported by Kieron Curtis2 <KCURTIS2@uk.ibm.com.example.com>
***
*** Monday, March 21, 2011 -- Dante v1.2.3
***
o Fix problem introduced in version 1.2.1, causing the SO_REUSEADDR
flag to not be set. Would cause problems during server restart if
only the main process was killed.
Thanks to Ralf Wenk <Ralf.Wenk@hs-karlsruhe.de.example.com> for
reporting this problem.
o The regular expression for host and interface name parsing has been
relaxed somewhat, allowing the use of some interface names and host
names containing characters that were not allowed previously.
Fixes problem reported by karesmakro <ipcop@it-connect-unix.de.example.com>,
Ralf Wenk <Ralf.Wenk@hs-karlsruhe.de.example.com>,
and <Adam.Barclay@barclayscapital.com.example.com>.
o The traffic log format has been slightly extended and now includes
complete information about all communication endpoints, for both
normal usage and with server chaining.
o Compilation fixes for older Linux distributions.
o Fix bug where the Solaris capabilities were not correctly
added/removed at all times.
o Compilation fixes for AIX 6.1. Problem report and testing by
Ville O Kesola <ville.kesola@fi.ibm.com.example.com> and
Ralf Wenk <iz-dante-0910@hs-karlsruhe.de.example.com>.
***
*** Tuesday, September 21, 2010 -- Dante v1.2.2
***
o Fix PAM bug introduced in v1.2.0 that would leak resources when
using PAM. Also relax the code handling PAM concerning
unknown msg_styles.
o Set PAM_RUSER so that rhosts-based PAM authentication can work.
o Wrapper for calling getsockopt(2) with SO_ERROR added.
o Compilation fixes for AIX 6.1.
Problem reported by Ralf Wenk <iz-dante-0910@hs-karlsruhe.de.example.com>.
Note to Ralf Wenk: we have been unable to contact you on the email
address you send from.
***
*** Friday, May 28, 2010 -- Dante v1.2.1
***
o GSSAPI support is no longer disabled on Heimdal versions older than
0.8.0, which have no wrap/unwrap support for aes256-cts-hmac-sha1-96.
The following krb5.conf configuration might be needed to ensure that
AES-256 is not used:
default_etypes = arcfour-hmac-md5 des3-cbc-sha1 des-cbc-crc des-cbc-md5
default_etypes_des = des-cbc-crc des-cbc-md5
Problem analyzed by Markus Moeller <markus_moeller at compuserve.com>.
o Code for shutting down idle sockd processes put back and can be
enabled again, if desired. It is not recommended for busy servers
however as performance may be slightly degraded.
o Fix bug related to clients using MSG_PEEK and GSSAPI on Linux.
Problem found by Markus Moeller <markus_moeller at compuserve.com>.
o Don't print warning if accept(2) fails when started with the -N<k> option.
When started with the -N option, it is expected that accept(2) will
fail in 1/k of the cases. Instead just log a debug message about it.
Should fix problem reported by Mark Sinner <mark.sinner@db.com.example.org>.
o The fflush() wrapper function did not handle a NULL argument.
Reported by "Kirby Zhou" <kirbyzhou@sohu-rd.com.example.org> and
Victor Sologoubov <victor0@rambler.ru.example.org>
o Fix libsocks upnpcleanup() linking problem when compiled without
libminiupnpc.
Problem reported by Kyle Thurow <krthurow@gmail.com.example.com> and
Adam Prato <adam.prato@gmail.com.example.com>.
o Assert failure related to message passing on some 64-bit architectures
fixed. Problem reported by Soner Tari <soner@comixwall.org.example.com>.
o The capi/socks.h file was not correctly installed.
Problem reported by Soner Tari <soner@comixwall.org.example.com>.
o Compilation fixes for OpenSuse 11.1.
Problem reported by Markus Moeller (markus_moeller at compuserve.com).
o Compilation fix for config_parse.y compilation without YYDEBUG.
Problem reported by Markus Moeller (markus_moeller at compuserve.com).
o Compilation fixes for AIX 6.1.
Problem reported by Ralf Wenk <iz-dante-0910@hs-karlsruhe.de.example.com>.
Tested by Marcel Augenstein.
***
*** Tuesday, October 27, 2009 -- Dante v1.2.0
***
o Improvements to client thread compatibility. The client
library should now be mostly thread safe.
o Solaris privilege model support. Based on patch from
Mayuresh Nirhali <Mayuresh.Nirhali@Sun.COM.example.com> & co.
o Make support for the socks 5 version described in draft-5.05 be
configurable. Before this was always enabled, but it breaks clients
based on NEC socks code in some cases as they use the same bit to
mean something completely different.
A new option has been added to sockd.conf to enable it:
"compatibility: draft-5.05". Unless explicitly enabled, the Dante
server will not use the socks 5.05 draft specification.
Problem found by Markus Moeller (markus_moeller at compuserve.com).
o Don't leak username/passwords provided to us for local authentication
to upstream proxy server when server-chaining. Error reported by
Mathieu DELAPLACE <mathieu.delaplace@gmail.com.example.com>.
o Fixed a bug/oversight that imposed an artificial limit on the number
of sockd processes that could be created, even when the load required
more.
o Slight improvement of configuration parsing in an attempt to avoid
confusing non-qualified hostnames with NIC interface names.
o The default connect-timeout/negotiate-timeout has been reduced from
120 seconds to 30 seconds. The "connecttimeout" name has also been
deprecated in favour of "timeout.negotiate".
o Separate iotimeouts can be set for udp and tcp clients. The "iotimeout"
object has also been deprecated in favour of "timeout.io".
o New configure option: "--disable-drt-fallback".
Used to disable the attempted automatic fallback to a direct route
if there are no usable proxy routes. Default is, as before,
automatic fallback.
o Added a new option: "udp.connectdst". Controls whether the server
should connect udp sockets to the destination.
The default for this release is yes, which improves performance,
but _may_ be incompatible with some udp-based application protocols.
Please let us know if you experience problems with some applications
no longer working.
o Markus Moeller (markus_moeller at compuserve.com) contributed support
for GSSAPI encryption/authentication (RFC 1961) to both the socks
server and socks client. Many thanks to Markus for this and for a
lot of help in testing it.
o Fixed problem with connection close before accept on HPUX.
Based on patch from Reine Persson <reine.persson@skatteverket.se>.
o Support for SunOS 4 and Solaris 2.5.1 removed.
o Added gethostbyaddr prototype for compilation on FreeBSD v6.2.
Patch from "Igor Mozolevsky" <igor@hybrid-lab.co.uk>.
o Fix getifa compilation problem on Solaris 8.
Reported by "Thomas" <dante@birdhouse.ch.example.com>.
o Patch from Markus Moeller (markus_moeller at compuserve.com),
to limit the range of udp-ports used between the socks-client
and the Dante server.
o By default, try to auto-add direct routes for all addresses on the LAN.
To disable it, set SOCKS_AUTOADD_LANROUTES to "no".
o Fix bug that caused problems with certain combinations of
bind(2)/accept(2)/close(2).
Reported by Markus Moeller (markus_moeller at compuserve.com).
o Fix bug that erroneously blocked the bind request from some clients.
Reported by Markus Moeller (markus_moeller at compuserve.com).
o Add support for environment variables SOCKS4_SERVER, SOCKS5_SERVER,
HTTP_PROXY, and UPNP_IGD.
If set, they specify the socks v4, socks v5 server, http proxy,
or UPNP-enabled ID to use, without the need for a socks.conf.
This should make it possible to run socksify with reasonable results
even without a socks.conf, as long as one of these new environment
variables are set correctly.
o Auto-add direct routes for all gateways. Should make the client
a little more user-friendly by not having to specify "direct" routes
for the proxyserver any longer.
o More fine grained marking of when to mark a proxy route as "bad" so that
it will not be used again by the same client.
Also add a new variable to config.h, MAX_ROUTE_FAILS, determining
how many times a route can fail before being blacklisted. Default
is one (same semantics as before there was a variable to control this).
o Fix bug that could prevent password authentication from working
on some systems. Spotted by
Nathan Johnson <n.johnson@vanderbilt.edu.example.com>.
o Add configure option --without-glibc-secure, which disables check for
the glibc variable __libc_enable_secure. Creates undesired dependencies
for packaging. Reported by Mathieu CHOUQUET-STRINGER
<mathieu.chouquet-stringer@sgcib.com.example.com>.
o New getifaddrs() compatibility function, taken from Heimdal-1.2.1.
o (Open)Solaris sa_len compilation problem fixed.
Reported by Markus Moeller (markus_moeller at compuserve.com).
o New version of the BSD version of "external.rotation: route",
from Christoph Badura <bad@bsd.de.example.com>.
o Support for interface names in sockd rules, and in the destination
address for socks routes.
Should make it easier to set up direct routes for local lan in
the client (specify all local interface names in route statements),
and block connections to e.g. loopback addresses (specify the the
loopback interface name in a block rule) in the server.
o UPnP support in the client, using the miniupnp library by
Thomas Bernard (http://miniupnp.free.fr/).
UPnP is a protocol implemented by many home/small-business routers
and adsl-modems. It allows you to dynamically open up ports on
the router for accepting incoming connections, as well as figuring
out what the external ipaddress of the router is.
Dante uses this to make socksify of ftp/bittorent/etc programs
work via the UPnP router.
Note that only the miniupnp library with release date 2009/09/21
or later is expected to work with Dante.
o Be less strict about bind in the client. The standards says
it is expected that the client first performs a connect via
the socks server, but it seems some/many socks servers support
the client requesting a bind without a previous connect, so we
assume that is the case in the client from now on.
o Changed the magic bytes that indicate the client is requesting
use of the Dante-specific bind extension from 0x00000000 to
0xffffffff, as part of the process to become less strict about
the bind command requiring a previous connection.
o Don't zero password in client if we read it from environment, or
it will not work the next time the same client process tries to
authenticate to the server.
Found by Pavel Fedin <fedin@matek.ru.example.com>.
That mail address no longer works. Please contact us if you have
another address for Mr. Fedin.
o Add support for "group:" syntax to rules, similar to "user:" statement.
Requested by Don Harvie <donharvi@au1.ibm.com.example.com>
o Close connection to PAM server each time we get an error-reply from
it, fixing a bug.
Patch from Robert Marcano <robert@marcanoonline.com.example.com>.
o Incorrect assert fixed. Problem reported by
"d binderman" <dcb314@hotmail.com.example.com>.
o Log close of client-rule with correct command.
Fixes bug reported by Tom Vandenbelt <tvandenb@au1.ibm.com.example.com>.
o Update to autoconf 2.61 and libtool 1.5.26.
All library names listed by *.la files no longer need to be listed
in the PLIST, e.g., instead of:
lib/libfoo.a
lib/libfoo.la
lib/libfoo.so
lib/libfoo.so.0
lib/libfoo.so.0.1
one simply needs:
lib/libfoo.la
and bsd.pkg.mk will automatically ensure that the additional library
names are listed in the installed package +CONTENTS file.
Also make LIBTOOLIZE_PLIST default to "yes".
Changes since 1.1.9:
Logging fixes (deprecating -l option to sockd)
Two commercial modules available, redirect and bandwidth (not included in pkg)
Some more from the 'Changes' file:
o Move daemon() call to later so more errors can be reported, suggested
by Borsenkow Andrej <Andrej.Borsenkow@mow.siemens.ru>.
o fix problem when linking with libsocks;
NISHIMURA Daisuke <nishi@graco.c.u-tokyo.ac.jp>.
o fix some problems when socksifying, making certain programs
hang forever (e.g. certain versions/installations of "ssh -X").
Thanks to NISHIMURA Daisuke <nishi@graco.c.u-tokyo.ac.jp> for
diagnosis, help and testing.
o Don't mark the rule as good either if sockscf.state.unfixedpamdata,
fixes bug reported by Jerry Murdock" <jmurdock@itraktech.com>.
o fix a bug preventing the list given in a 'user:' keyword from
being checked correctly. Reported by Oleg Bulavsky <bulch@ftc.ru>.
o "-h" prints out configfile used, based on suggestion from
dh_tsc_10@ugcs.net.
o fix bugs that reset some defaults at the wrong time, reported by
dh_tsc_10@ugcs.net.
o prototype script for generation of graphs with usage information included
(bin/sockd-graphgen). Contact us if you run a socks server with significant
usage, and you are willing to help with testing.
o Some minor optimisations in server i/o code, at the cost of some
timer accuracy concerning bandwidth limiting/client expiration.
o prefix "socks" to some global variables to avoid collisions during
socksify. Reported by Don Reid <donr@cvs.agilent.com>.
o capi/socks.h needs Rxxx prototypes to work with a c++
compiler. Noted by Alex Morozov <alex@idisys.iae.nsk.su>.
o If Rgetsockname() is called on a socket that a previous uncompleted
Rconnect() has been done on, try to sleep until Rconnect() has
finished instead of returning ENOBUFS. Hoped to help compatibility
with some applications.
o Modified httpproxy code a little, based on patch from
dh_tsc_10@ugcs.net.
o Code for selecting which of multiple external ipaddresses to use
on a global basis based on routing.
Code contributed by Tom Chan <tchan@austin.rr.com>.
o Rename "pamservicename" to "pam.servicename".
o Fix memory overrun problem in Rgethostbyname*(). Found and
diagnosed by dme@dme.org, thanks.
o new method added: "pam". Code contributed by
Patrick Bihan-Faou, MindStep Corporation, patrick@mindstep.com.
o let client-rules have their own global methodline, "clientmethod",
default value set to "none".
The global "method" is only used for socks-rules now.
o delay checking of password/etc til we have received the socks
request (rather than during negotiation, as was the case).
Required for supporting passwordbased authentication via non-socks
methods, e.g. pam.
o socklen_t definition updated for NetBSD 1.5U.
Problem report and patch submitted by Janne Snabb <snabb@ssh.com>.
o Output from 'config.guess' is not sufficient to determine use
of elf on NetBSD.
Problem report and patch submitted by Janne Snabb <snabb@ssh.com>
leot
jperkin
joerg
abs
jlam
wiz
zuntum