BUILD INSTRUCTIONS A makefile was supplied with this which should have built the program. If it fails please let us know, and here are some hints for building on different platforms. Tested OK on Linux/x86 with gcc3.2. cc -O3 -pedantic -Wuninitialized -Wall -pipe -mcpu=pentium -march=pentium -fomit-frame-pointer -ffast-math -finline-functions -funroll-loops clamav-milter.c -pthread -lmilter ../libclamav/.libs/libclamav.a ../clamd/cfgfile.o ../clamd/others.o Compiles OK on Linux/x86 with tcc 0.9.16, but fails to link errors with 'atexit' tcc -g -b -lmilter -lpthread clamav-milter.c... Fails to compile on Linux/x86 with icc6.0 (complains about stdio.h...) icc -O3 -tpp7 -xiMKW -ipo -parallel -i_dynamic -w2 clamav-milter.c... Fails to build on Linux/x86 with icc7.1 with -ipo (fails on libclamav.a - keeps saying run ranlib). Otherwise it builds and runs OK. icc -O2 -tpp7 -xiMKW -parallel -i_dynamic -w2 -march=pentium4 -mcpu=pentium4 clamav-milter.c... Tested with Electric Fence 2.2.2 Compiles OK on Linux/ppc (YDL2.3) with gcc2.95.4. Needs -lsmutil to link. cc -O3 -pedantic -Wuninitialized -Wall -pipe -fomit-frame-pointer -ffast-math -finline-functions -funroll-loop -pthread -lmilter ../libclamav/.libs/libclamav.a ../clamd/cfgfile.o ../clamd/others.o -lsmutil I haven't tested it further on this platform yet. YDL3.0 should compile out of the box Sendmail on MacOS/X (10.1) is provided without a development package so this can't be run "out of the box" Solaris 8 doesn't have milter support so clamav-milter won't work unless you rebuild sendmail from source. FreeBSD4.7 use /usr/local/bin/gcc30. GCC3.0 is an optional extra on FreeBSD. It comes with getopt.h which is handy. To link you need -lgnugetopt gcc30 -O3 -DCONFDIR=\"/usr/local/etc\" -I. -I.. -I../clamd -I../libclamav -pedantic -Wuninitialized -Wall -pipe -mcpu=pentium -march=pentium -fomit-frame-pointer -ffast-math -finline-functions -funroll-loops clamav-milter.c -pthread -lmilter ../libclamav/.libs/libclamav.a ../clamd/cfgfile.o ../clamd/others.o -lgnugetopt FreeBSD4.8: compiles out of the box with either gcc2.95 or gcc3 OpenBSD3.4: the supplied sendmail does not come with Milter support. Do this *before* running configure (thanks for Per-Olov Sjöhol <peo_s@incedo.org> for these instructions). echo WANT_LIBMILTER=1 > /etc/mk.conf cd /usr/src/gnu/usr.sbin/sendmail make depend make make install kill -HUP `sed q /var/run/sendmail.pid` Then do this to make the milter headers available to clamav... (the libmilter.a file is already in the right place after the sendmail recompiles above) cd /usr/include ln -s ../src/gnu/usr.sbin/sendmail/include/libmilter libmilter Solaris 9 and FreeBSD5 have milter support in the supplied sendmail, but doesn't include libmilter so you can't develop milter applications on it. Go to sendmail.org, download the lastest sendmail, cd to libmilter and "make install" there. Needs -lresolv on Solaris INSTALLATION Install into /usr/local/sbin/clamav-milter. See http://www.nmt.edu/~wcolburn/sendmail-8.12.5/libmilter/docs/sample.html Installations for RedHat Linux and it's derivatives such as YellowDog: Ensure that you have the sendmail-devel RPM installed Add to /etc/mail/sendmail.mc: INPUT_MAIL_FILTER(`clamav', `S=local:/var/run/clamav/clmilter.sock, F=, T=S:4m;R:4m')dnl define(`confINPUT_MAIL_FILTERS', `clamav') Check entry in /usr/local/etc/clamav.conf of the form: LocalSocket /var/run/clamav/clamd.sock StreamSaveToDisk If you already have a filter (such as spamassassin-milter from http://savannah.nongnu.org/projects/spamass-milt) add it thus: INPUT_MAIL_FILTER(`clamav', `S=local:/var/run/clamav/clmilter.sock, F=, T=S:4m;R:4m')dnl INPUT_MAIL_FILTER(`spamassassin', `S=local:/var/run/spamass.sock, F=, T=C:15m;S:4m;R:4m;E:10m') define(`confINPUT_MAIL_FILTERS', `spamassassin,clamav')dnl mkdir /var/run/clamav chown clamav /var/run/clamav (if you use User clamav in clamav.conf) chmod 700 /var/run/clamav You may find INPUT_MAIL_FILTERS is not needed on your machine, however it is recommended by the Sendmail documentation and I recommend going along with that. If you see an unsafe socket error from sendmail, it means that the permissions of the /var/run/clamav directory are too open. Check you have correctly run chown and chmod, it may also mean that clamav-milter hasn't started, run ps and check your logs. The above example shows clamav-milter, clamd and sendmail all on the same machine, however using TCP they may reside on different machines, indeed clamav-milter is capable of talking to multiple clamds for redundancy and load balancing. I suggest putting SpamAssassin first since you're more likely to get spam than a virus/worm sent to you. Add to /etc/sysconfig/clamav-milter CLAMAV_FLAGS="--max-children=2 local:/var/run/clamav/clmilter.sock" or if clamd is on a different machine CLAMAV_FLAGS="--max-children=2 --server=192.168.1.9 local:/var/run/clamav/clmilter.sock" If you want clamav-milter to listen on TCP for communication with sendmail, for example if they are on different machines use inet:<port>. On machine A (running sendmail) you would have in sendmail.mc: INPUT_MAIL_FILTER(`clamav', `S=inet:3311@machineb, F=, T=S:4m;R:4m')dnl On machine B (running clamav-milter) you would start up clamav-milter thus: clamav-milter inet:3311 You should have received a script to put into /etc/init.d with this software. You should always start clamd before clamav-milter. As with all software it is wise to ensure that clamav-milter has the least privileges it needs to run. So don't run it as root and don't store the sockets in a directory that can be written by everyone. For example ensure that /var/run is owned and writable only by root and add entries for 'User' and 'FixStaleSocket' in clamav.conf. When using UNIX domain sockets via the LocalSocket option of clamav.conf, we recommend that you use the --quarantine-dir option since that may improve performance. CHANGE HISTORY Changes 0.2: 4/3/03 clamfi_abort() now always calls pthread_mutex_unlock 5/3/03 Only send a bounce if -b is set Version now uses -v not -V --config-file couldn't be set by -c 0.3 7/3/03 Enhanced the Solaris compile time comment No need to save the return result of LogSyslog Use LogVerbose 0.4 9/3/03 Initialise dataSocket/cmdSocket correctly 10/3/03 Say why we don't connect() to clamd Enhanced '-l' usage message 0.5 18/3/03 Ported to FreeBSD 4.7 Source no longer in support, so remove one .. from the build instructions Corrected the use of strerror_r 0.51 20/3/03 Mention StreamSaveToDisk in the installation Added -s option which allows clamd to run on a different machine from the milter 0.52 20/3/03 -b flag now only stops the bounce, sends warning to recipient and postmaster 0.53 24/3/03 %d->%u in syslog call 27/3/03 tcpSocket is now of type in_port_t 27/3/03 Use PING/PONG 0.54 23/5/03 Allow a range of IP addresses as outgoing ones that need not be checked 0.55 24/5/03 Use inet_ntop() instead of inet_ntoa() Thanks to Krzysztof Olędzki <ole@ans.pl> 0.60 11/7/03 Some TODOs done by Nigel Kukard <nkukard@lbsd.net> Should stop a couple of remote chances of crashes 0.60a 22/7/03 Tidied up message when sender is unknown 0.60b 17/8/03 Optionally set postmaster address. Usually one uses /etc/aliases, but not everyone want's to... 0.60c 22/8/03 Another go at Solaris support 0.60d 26/8/03 Removed superfluous buffer and unneeded strerror call ETIMEDOUT isn't an error, but should give a warning 0.60e 09/9/03 Added -P and -q flags by "Nicholas M. Kirsch" <nick@kirsch.org> 0.60f 17/9/03 Changed fprintf to fputs where possible Redirect stdin from /dev/null, stdout&stderr to /dev/console 0.60g 26/9/03 Handle sendmail calling abort after calling cleanup (Should never happen - but it does) Added -noxheader patch from dirk.meyer@dinoex.sub.org 0.60h 28/9/03 Support MaxThreads option in config file, overriden by --max-children. Patch from "Richard G. Roberto" <rgr@dedlegend.com> 0.60i 30/9/03 clamfi_envfrom() now correctly returns SMFIS_TEMPFAIL, in a few circumstances it used to return EX_TEMPFAIL Patch from Matt Sullivan <matt@sullivan.gen.nz> 0.60j 1/10/03 strerror_r doesn't work on Linux, attempting workaround Added support for hard-coded list of email addresses who's e-mail is not scanned 0.60k 5/10/03 Only remove old UNIX domain socket if FixStaleSocket is set 0.60l 11/10/03 port is now unsigned Removed remote possibility of crash if the target e-mail address is very long No longer calls clamdscan to get the version 0.60m 12/10/03 Now does sanity check if using localSocket Gets version info from clamd Only reset fd's 0/1/2 if !ForeGround 0.60n 22/10/03 Call pthread_cont_broadcast more often 0.60o 31/10/03 Optionally accept all mails if scanning procedure fails (Joe Talbott <josepht@cstone.net>) 0.60p 5/11/03 Only call mutex_unlock when max_children is set Tidy up the call to pthread_cond_timedwait 0.60q 11/11/03 Fixed handling of % characters in e-mail addresses pointed out by dotslash@snosoft.com 0.65 15/11/03 Upissue of clamav 0.65a 19/11/03 Close cmdSocket earlier Added setpgrp() 0.65b 22/11/03 Ensure milter is not run as root if requested Added quarantine support 0.65c 24/11/03 Support AllowSupplementaryGroups Fix warning about root usage 0.65d 25/11/03 Handle empty hostname or hostaddr Fix based on a submission by Michael Dankov <misha@btrc.ru> 0.65e 29/11/03 Fix problem of possible confused pointers if large number of recipients given. Fix by Michael Dankov <misha@btrc.ru>. 0.65f 29/11/03 Added --quarantine-dir Thanks to Michael Dankov <misha@btrc.ru>. 0.65g 2/12/03 Use setsid if setpgrp is not present. Thanks to Eugene Crosser <crosser@rol.ru> 0.65h 4/12/03 Added call to umask to ensure that the local socket is not publically writeable. If it is sendmail will (correctly!) refuse to start this program Thanks for Nicklaus Wicker <n.wicker@cnk-networks.de> Don't sent From as the first line since that means clamd will think it is an mbox and not handle unescaped From at the start of lines properly Thanks to Michael Dankov <misha@btrc.ru> 0.65i 9/12/03 Use the location of sendmail discovered by configure 0.65j 10/12/03 Timeout on waiting for data from clamd 0.65k 12/12/03 A couple of calls to clamfi_cleanup were missing before return cl_error 0.66 13/12/03 Upissue 0.66a 22/12/03 Added --sign 0.66b 27/12/03 --sign moved to privdata 0.66c 31/12/03 Included the sendmail queue ID in the log, from an idea by Andy Fiddaman <af@jeamland.org> 0.66d 10/1/04 Added OpenBSD instructions Added --signature-file option 0.66e 12/1/04 FixStaleSocket: no longer complain if asked to remove an old socket when there was none to remove 0.66f 24/1/04 -s: Allow clamd server name as well as IPaddress 0.66g 25/1/04 Corrected usage message Started to honour --debug Dump core on LINUX if CL_DEBUG set Support multiple servers separated by colons 0.66h 26/1/04 Corrected endian problem (ntohs instead of htons) 0.66i 28/1/04 Fixed compilation error with --enable-debug 0.66j 29/1/03 Added --noreject flag, based on a patch by "Vijay Sarvepalli" <vssarvep@office.uncg.edu> 0.66k 2/2/04 When --postmaster-only is given, include the system ID of the message in the warning e-mail, since that will help the administrator when sifting through the mail logs. Based on an idea by Jim Allen, <Jim.Allen@Heartsine.co.uk> 0.66l 7/2/04 Updated URL reference Added new config.h mechanism 0.66m 9/2/04 Added Hflag from "Leonid Zeitlin" <lz@europe.com> 0.66n 13/2/04 Added TCPwrappers support Removed duplication in version string Handle machines that don't have in_port_t 0.67 16/2/04 Upissued to 0.67 0.67a 16/2/04 Added clamfi_free 0.67b 17/2/04 Removed compilation warning - now compiles on FreeBSD5.2 Don't allow --force to overwride TCPwrappers 0.67c 18/2/04 Added dont-log-clean flag 0.67d 19/2/04 Reworked TCPwrappers code Thanks to "Hector M. Rulot Segovia" <Hector.Rulot@uv.es> Changed some printf/puts to cli_dbgmsg 0.67e 20/2/04 Moved the definition of the sendmail pipe The recent changes to the configure script changed the order of includes so some prototypes weren't getting in 0.67f 20/2/04 Added checkClamd() - if possible attempts to see if clamd has died 0.67g 21/2/04 Don't run if the quarantine-dir is publically accessable 0.67h 22/2/04 Change the log level TCPwrapper denying Handle ERROR message from clamd Moved smfi_setconn to avoid race condictions when an e-mail is received just as the milter is starting but isn't ready to handle it causing the milter to go to an error state Hardend umask 0.67i 27/2/04 Dropping priv message now same as clamd Only use TCPwrappers when using TCP/IP to establish communications with the milter 0.67j 27/2/04 Call checkClamd() before attempting to connect, it's a way of warning the user if they've started the milter before clamd checkClamd() now stashes pid in syslog Ensure installation instructions tally with man page and put sockets into subdirectory for security clamfi_close debug, change assert to debug message Better way to force TCPwrappers only with TCP/IP 0.67k 7/3/04 Ensure cli_dbgmsg's end with \n Fixed some warning messages with icc Use cli_[cm]alloc Included extra information if --headers is given (based on an idea from "Leonid Zeitlin" <lz@europe.com> 0.67l 10/3/04 Use new HAVE_STRERROR_R rather than TARGET_OS_SOLARIS to determine if strerror_r exists 0.70 17/3/04 Up-issued to 0.70 0.70a 20/3/04 strerror_r is a bit confused on Fedora Linux. The man page says it returns an int, but the prototype in string.h says it returns a char * Say how many bytes can't be written to clamd - it may give a clue what's wrong 0.70b 26/3/04 Display errno information on write failure to clamd Ensure errno is passed to strerror Print fd in clamfi_send debug 0.70c 27/3/04 Timestamp clamfi_send messages Call cli_warnmsg if ERROR received Minor code tidy Delay connection to clamd to handle clamd's appetite for timing out when the remote end (the end talking to sendmail) is slow Prefer cli_dbgmsg/cli_warnmsg over printf 0.70d 29/3/04 Print the sendmail ID with the virus note in syslog config file location has changed 0.70e 1/4/04 Fix a remote possibility of a file descriptor leak in PingServer() if clamd has died Fix by Andrey J. Melnikoff (TEMHOTA) <temnota@kmv.ru> Corrected some debug messages reported by Sergey Y. Afonin <asy@kraft-s.ru> 0.70f 1/4/04 Added auto-submitted header to messages generated here Suggested by "Andrey J. Melnikoff (TEMHOTA)" <temnota@kmv.ru> Add advice that --quarantine-dir may improve performance when LocalSocket is used ThreadTimeout seems to have been changed to ReadTimeout 0.70g 3/4/04 Error if ReadTimeout is -ve Honour StreamMaxLength 0.70h 8/4/04 Cleanup StreamMaxLength code 0.70i 9/4/04 Handle clamd giving up on StreamMaxLength before clamav-milter 0.70j 15/4/04 Handle systems without inet_ntop 0.70k 17/4/04 Put the virus message in the 550 rejection 0.70l 19/4/04 Started coding e-mail template support 0.70m 19/4/04 Started code to parse header to find the real infected machine Added the --from flag Return SMFIS_TEMPFAIL when out of memory idea by Joe Maimon <jmaimon@ttec.com> Some still to be done Based on an idea by Christian Pelissier <Christian.Pelissier@onera.fr>. Store different day's quarantines in different directories to make them easier to manage 0.70n 20/4/04 Allow for "i" macro not defined in sendmail.cf clamfi_connect: print better message if hostaddr is null 0.70o 20/4/04 Added X-Virus-Status Always add X-Virus-Scanned If hostaddr is NULL assume it's a local connection. This is probably a safe assumption but it should be verified 0.70p 20/4/04 If /dev/console fails to open, open /dev/null instead on fds 1 and 2 TCP_WRAPPERS code now uses inet_ntop() Simplify virus string Sort out tabs in the hard coded e-mail message 0.70q 22/4/04 No need to parse the received line if --headers is given If -outgoing is given put generated emails in the deferred queue to avoid the milter being called twice at the same time (one on the incoming one on the outgoing) header_list_print, ensure From lines are escaped, may not be needed but it is better to be on the safe side When loadbalancing, fail to start only if no servers can be reached (used to fail if any one server could not be reached) Not all servers were load balanced 0.70r 23/4/04 Ensure only From lines are escaped Also defer generated emails if --force-scan is given Better subject for quarantine e-mails 0.70s 25/4/04 Added --pidfile support 0.70t 28/4/04 Better quarantine message error report when failing to create the temporary file Send 554 after DATA received, not 550 Don't send rejection notices to rejection notices, we just end up playing ping-pong (patch by "Andrey J.Melnikoff (TEMHOTA)" <temnota@kmv.ru> If CL_DEBUG is defined, don't redirect stdout/stderr Don't attempt to return an old signature if no filename has been given. There has never been one to return 0.70u 29/4/04 When changing from realloc to cli_realloc I forgot to keep the assignment of signature 0.70v 6/5/04 clamfi_close now always checks privdata is NULL, not only when debugging Allow transfers of exactly streamMaxLength Warn if a clean file can't be removed from the quarantine When streamMaxLength is exceeded add a header where possible, unless --noxheader is given 0.70x 7/5/04 Only report that we've dropped privilege if the setuid succeeded, fix by Jens Elkner <elkner@linofee.org> If logVerbose is set state both starting and started messages (based on an idea by "Sergey Y. Afonin" <asy@kraft-s.ru> Also added X-Infected-Received-From: header by Sergey Fix from Damian Menscher <menscher@uiuc.edu> ensures that when a child dies we continue when max children is hit Report an error if inet_ntop fails in tcp_wrappers 0.71 16/5/04 Up issue 0.71a 21/5/04 --from wasn't always a recognised option Write failure to quarantine file now logs the name of the file Commented out TKs advice about using quarantine when using localSocket, sys admins were confused by it 0.71b 24/5/04 Add which host did the virus scanning 0.71c 25/5/04 X-Virus-Status: Not Scanned - StreamMaxLength exceeded was not always being added Now says host running clamd rather than host running clamav-milter, useful for checking load balancing etc. 0.72 3/6/04 Up-issued BUG REPORTS Please send bug reports and/or comments to Nigel Horne <njh@despammed.com> or bugs@clamav.net. TODO There are several ideas marked as TODO in the source code. If anyone has any other suggestions please feel free to contact me.