September 2011 - stunnel-users

netword.c & fdprintf access access violation fix
by Alexey Drozdov 12 Sep '11

12 Sep '11

Hi, all One more bug was found out at work stunnel through a proxy. There is a reference to already released memory in fdprintf function. Here a small fix of this situation: *** network.c 2011-09-07 20:46:19.000000000 +0400 --- network.c_patch 2011-09-12 23:52:46.015618574 +0400 *************** *** 766,771 **** --- 766,772 ---- int fdprintf(CLI *c, int fd, const char *format, ...) { va_list ap; char *line; + int outsize; va_start(ap, format); line=str_vprintf(format, ap); *************** *** 775,782 **** longjmp(c->err, 1); } fdputline(c, fd, line); str_free(line); ! return strlen(line)+2; } #define INET_SOCKET_PAIR --- 776,784 ---- longjmp(c->err, 1); } fdputline(c, fd, line); + outsize = strlen(line)+2; str_free(line); ! return outsize; } #define INET_SOCKET_PAIR / Alexey Drozdov

1 0

SNI and some troubles solving
by Alexey Drozdov 10 Sep '11

10 Sep '11

Hi All, At first I would like to thank Marzena Trojnara for the useful program! Occurrence of support SNI has especially interest. I have decided to use the given possibility, but have detect pair of bugs which has independently successfully solved. To begin with I dump part of my config: [virtual_server] accept = 443 connect = localhost:1081 cert = /usr/local/etc/stunnel/alexCA.crt key = /usr/local/etc/stunnel/alexCA.key ; need peer verification for default access verify = 2 CAfile = /usr/local/etc/stunnel/alexCA.crt ; use restricted access form some target services [socks5] sni = virtual_server:localhost sni = virtual_server:89.250.5.185 sni = virtual_server:127.0.0.1 sni = virtual_server:server cert = /usr/local/etc/stunnel/alexCA.crt key = /usr/local/etc/stunnel/alexCA.key verify = 2 CAfile = /usr/local/etc/stunnel/alexCA.crt connect = localhost:1081 ; public service [dreamhub_ru] sni = virtual_server:dreamhub.ru sni = virtual_server:pad.dreamhub.ru cert = /usr/local/etc/stunnel/dreamhub.cer key = /usr/local/etc/stunnel/dreamhub.pem connect = localhost:80 I want to use stunnel on 443rd port for ssl-socks5 service with peer certificates verification and use public https web-service (https://dreamhub.ru) And so it has appeared that value for "verify" are always used from global section (in my case [virtual_server]), and values of "verify" parameters from virtual services sections aren't used. Problem it has appeared that in the SNI check callback static int servername_cb(SSL *ssl, int *ad, void *arg) ... for(list=section->servername_list_head; list; list=list->next) if(!strcasecmp(servername, list->servername)) { c=SSL_get_ex_data(ssl, cli_index); c->opt=list->opt; SSL_set_SSL_CTX(ssl, c->opt->ctx); s_log(LOG_NOTICE, "SNI: switched to section %s", c->opt->servname); SSL_set_SSL_CTX(ssl, c->opt->ctx); does not copy ctx->verify_mode (SSL*)s->verify_mode=ctx->verify_mode; It has been solved by explicit calling SSL_set_verify *** ctx.c 2011-09-07 14:43:50.000000000 +0400 --- ctx.c_patched 2011-09-10 01:12:48.000000000 +0400 *************** *** 156,161 **** --- 156,162 ---- c=SSL_get_ex_data(ssl, cli_index); c->opt=list->opt; SSL_set_SSL_CTX(ssl, c->opt->ctx); + SSL_set_verify(ssl, SSL_CTX_get_verify_mode(c->opt->ctx), +NULL); s_log(LOG_NOTICE, "SNI: switched to section %s", c->opt->servname); #ifdef USE_LIBWRAP The second moment concerns uses stunnel as client through HTTPS a proxy (protocol=connect). All works perfectly, only it has appeared that value for SNI used from "connect" parameter instead of value of "protocolHost". On a code all is quite obvious: if(c->opt->host_name) { s_log(LOG_DEBUG, "SNI: host name: %s", c->opt->host_name); if(!SSL_set_tlsext_host_name(c->ssl, c->opt->host_name)) { sslerror("SSL_set_tlsext_host_name"); longjmp(c->err, 1); } } I solved it such: *** options.c 2011-09-07 20:38:56.000000000 +0400 --- options.c_patch 2011-09-10 16:34:56.882619049 +0400 *************** *** 695,700 **** --- 695,705 ---- s_log(LOG_INFO, "Cannot resolve '%s' - delaying DNS lookup", arg); section->option.delayed_lookup=1; } + + /* check - is host_name already filled */ + if (section->host_name != NULL) + return NULL; /* OK */ + tmpstr=strrchr(arg, ':'); if(tmpstr) { *tmpstr='\0'; *************** *** 1077,1082 **** --- 1082,1095 ---- if(strcasecmp(opt, "protocolHost")) break; section->protocol_host=str_dup_err(arg); + + /* Overwrite section->host_nam because of + protocolHost should be used for client SNI */ + tmpstr=strrchr(arg, ':'); + if(tmpstr) { + *tmpstr='\0'; + section->host_name=str_dup_err(arg); + } return NULL; /* OK */ case CMD_DEFAULT: break; I think that the given changes can be useful for other :) / Alexey Drozdov

2 1

SNI and some troubles solving
by Alexey Drozdov 10 Sep '11

10 Sep '11

Hi All, At first I would like to thank Marzena Trojnara for the useful program! Occurrence of support SNI has especially interest. I have decided to use the given possibility, but have detect pair of bugs which has independently successfully solved. To begin with I dump part of my config: [virtual_server] accept = 443 connect = localhost:1081 cert = /usr/local/etc/stunnel/alexCA.crt key = /usr/local/etc/stunnel/alexCA.key ; need peer verification for default access verify = 2 CAfile = /usr/local/etc/stunnel/alexCA.crt ; use restricted access form some target services [socks5] sni = virtual_server:localhost sni = virtual_server:89.250.5.185 sni = virtual_server:127.0.0.1 sni = virtual_server:server cert = /usr/local/etc/stunnel/alexCA.crt key = /usr/local/etc/stunnel/alexCA.key verify = 2 CAfile = /usr/local/etc/stunnel/alexCA.crt connect = localhost:1081 ; public service [dreamhub_ru] sni = virtual_server:dreamhub.ru sni = virtual_server:pad.dreamhub.ru cert = /usr/local/etc/stunnel/dreamhub.cer key = /usr/local/etc/stunnel/dreamhub.pem connect = localhost:80 I want to use stunnel on 443rd port for ssl-socks5 service with peer certificates verification and use public https web-service (https://dreamhub.ru) And so it has appeared that value for "verify" are always used from global section (in my case [virtual_server]), and values of "verify" parameters from virtual services sections aren't used. Problem it has appeared that in the SNI check callback static int servername_cb(SSL *ssl, int *ad, void *arg) ... for(list=section->servername_list_head; list; list=list->next) if(!strcasecmp(servername, list->servername)) { c=SSL_get_ex_data(ssl, cli_index); c->opt=list->opt; SSL_set_SSL_CTX(ssl, c->opt->ctx); s_log(LOG_NOTICE, "SNI: switched to section %s", c->opt->servname); SSL_set_SSL_CTX(ssl, c->opt->ctx); does not copy ctx->verify_mode (SSL*)s->verify_mode=ctx->verify_mode; It has been solved by explicit calling SSL_set_verify *** ctx.c 2011-09-07 14:43:50.000000000 +0400 --- ctx.c_patched 2011-09-10 01:12:48.000000000 +0400 *************** *** 156,161 **** --- 156,162 ---- c=SSL_get_ex_data(ssl, cli_index); c->opt=list->opt; SSL_set_SSL_CTX(ssl, c->opt->ctx); + SSL_set_verify(ssl, SSL_CTX_get_verify_mode(c->opt->ctx), +NULL); s_log(LOG_NOTICE, "SNI: switched to section %s", c->opt->servname); #ifdef USE_LIBWRAP The second moment concerns uses stunnel as client through HTTPS a proxy (protocol=connect). All works perfectly, only it has appeared that value for SNI used from "connect" parameter instead of value of "protocolHost". On a code all is quite obvious: if(c->opt->host_name) { s_log(LOG_DEBUG, "SNI: host name: %s", c->opt->host_name); if(!SSL_set_tlsext_host_name(c->ssl, c->opt->host_name)) { sslerror("SSL_set_tlsext_host_name"); longjmp(c->err, 1); } } I solved it such: *** options.c 2011-09-07 20:38:56.000000000 +0400 --- options.c_patch 2011-09-10 16:34:56.882619049 +0400 *************** *** 695,700 **** --- 695,705 ---- s_log(LOG_INFO, "Cannot resolve '%s' - delaying DNS lookup", arg); section->option.delayed_lookup=1; } + + /* check - is host_name already filled */ + if (section->host_name != NULL) + return NULL; /* OK */ + tmpstr=strrchr(arg, ':'); if(tmpstr) { *tmpstr='\0'; *************** *** 1077,1082 **** --- 1082,1095 ---- if(strcasecmp(opt, "protocolHost")) break; section->protocol_host=str_dup_err(arg); + + /* Overwrite section->host_nam because of + protocolHost should be used for client SNI */ + tmpstr=strrchr(arg, ':'); + if(tmpstr) { + *tmpstr='\0'; + section->host_name=str_dup_err(arg); + } return NULL; /* OK */ case CMD_DEFAULT: break; I think that the given changes can be useful for other :) / Alexey Drozdov

1 0

stunnel 4.43 released
by Michal Trojnara 07 Sep '11

07 Sep '11

Dear Users, I have released version 4.43 of stunnel. The ChangeLog entry: Version 4.43, 2011.09.07, urgency: MEDIUM: * New features - Updated Win32 DLLs for OpenSSL 1.0.0e. - Major optimization of the logging subsystem. Benchmarks indicate up to 15% performance improvement. * Bugfixes - Fixed WIN32 configuration file reload. - Fixed FORK and UCONTEXT threading models. - Corrected INSTALL.W32 file. Home page: http://www.stunnel.org/ Download: ftp://ftp.stunnel.org/stunnel/ SHA-256 hash for stunnel-4.43.tar.gz: 93a002d9e1652d7684756af75b44b00f99aa93574e8a5a2e69f88656221d5ce2 Best regards, Mike

2 1

SSL3_READ_BYTES:tlsv1 alert internal error
by Andrew Heuneman 07 Sep '11

07 Sep '11

Hello, We recently implemented Stunnel on Centos 5.6 for ssl offloading for our java application. The application has applets to communicate java objects over https to a tomcat server on the server side. We have it setup in front of our Alteon/Radware load balancer. This hardware load balancer is capable of ssl load balancing, but has produced a very specific packet reset that only presents itself in ssl processing. We decided to implement Stunnel in front of this load balancer to fix this problem. Ssl offloading was working great with Stunnel until we ran into Java 7. if I run any version of our applets on java 6 they work. If i run java 7 they do not work. I have tried googling and looking for this error but I have only found some references to SNI... is this correct? Is there anything I can do. Please forgive me if i have omitted any details I will be more than happy to include a packet capture or other details if needed. I compiled stunnel with the following options ./configure --disable-libwrap --bindir=/usr/sbin --sbindir=/usr/sbin --sysconfdir=/etc --with-ssl=/usr/local/ssl also i compiled OpenSSL 1.0.0d with the following ./Configure threads shared linux-generic64 stunnel -version No limit detected for the number of clients signal_pipe: FD=3 allocated (non-blocking mode) signal_pipe: FD=4 allocated (non-blocking mode) stunnel 4.42 on x86_64-unknown-linux-gnu platform Compiled/running with OpenSSL 1.0.0d 8 Feb 2011 Threading:PTHREAD SSL:ENGINE Auth:none Sockets:POLL,IPv6 stunnel 4.42 on x86_64-unknown-linux-gnu platform Compiled/running with OpenSSL 1.0.0d 8 Feb 2011 Threading:PTHREAD SSL:ENGINE Auth:none Sockets:POLL,IPv6 Global option defaults debug = daemon.notice pid = /usr/local/var/run/stunnel/stunnel.pid RNDbytes = 64 RNDfile = /dev/urandom RNDoverwrite = yes Service-level option defaults ciphers = ALL:!SSLv2:!aNULL:!EXP:!LOW:-MEDIUM:RC4:+HIGH curve = prime256v1 session = 300 seconds sslVersion = TLSv1 for client, all for server stack = 65536 bytes TIMEOUTbusy = 300 seconds TIMEOUTclose = 60 seconds TIMEOUTconnect = 10 seconds TIMEOUTidle = 43200 seconds verify = none str_stats: 112 block(s), 4046 byte(s) stunnel.conf cert=/etc/stunnel/stunnel.pem debug=7 output=/var/log/stunnel.log socket=l:TCP_NODELAY=1 socket=r:TCP_NODELAY=1 [https] accept=0.0.0.0:443 connect=172.16.18.100:80 session = 300 TIMEOUTbusy = 300 TIMEOUTconnect = 10 TIMEOUTidle = 43200 client = no stunnel.log 2011.08.22 16:58:49 LOG7[438154:47689394220768]: Service https accepted FD=2 from 10.0.11.27:46830 2011.08.22 16:58:49 LOG7[438154:1104877888]: Service https started 2011.08.22 16:58:49 LOG7[438154:1104877888]: Option TCP_NODELAY set on local socket 2011.08.22 16:58:49 LOG5[438154:1104877888]: Service https accepted connection from 10.0.11.27:46830 2011.08.22 16:58:49 LOG7[438154:1104877888]: SSL state (accept): before/accept initialization 2011.08.22 16:58:49 LOG7[438154:1104877888]: SSL state (accept): SSLv3 read client hello A 2011.08.22 16:58:49 LOG7[438154:1104877888]: SSL state (accept): SSLv3 write server hello A 2011.08.22 16:58:49 LOG7[438154:1104877888]: SSL state (accept): SSLv3 write certificate A 2011.08.22 16:58:49 LOG7[438154:1104877888]: SSL state (accept): SSLv3 write key exchange A 2011.08.22 16:58:49 LOG7[438154:1104877888]: SSL state (accept): SSLv3 write certificate request A 2011.08.22 16:58:49 LOG7[438154:1104877888]: SSL state (accept): SSLv3 flush data 2011.08.22 16:58:49 LOG7[438154:1104877888]: SSL alert (read): fatal: internal error 2011.08.22 16:58:49 LOG3[438154:1104877888]: SSL_accept: 14094438: error:14094438:SSL routines:SSL3_READ_BYTES:tlsv1 alert internal error 2011.08.22 16:58:49 LOG5[438154:1104877888]: Connection reset: 0 bytes sent to SSL, 0 bytes sent to socket 2011.08.22 16:58:49 LOG7[438154:1104877888]: Service https finished (1 left) 2011.08.22 16:58:49 LOG7[438154:1104877888]: str_stats: 0 block(s), 0 byte(s) 2011.08.22 16:59:01 LOG7[438154:1104947520]: Socket closed on read 2011.08.22 16:59:01 LOG7[438154:1104947520]: Sending SSL write shutdown 2011.08.22 16:59:01 LOG7[438154:1104947520]: SSL alert (write): warning: close notify 2011.08.22 16:59:01 LOG6[438154:1104947520]: SSL_shutdown successfully sent close_notify 2011.08.22 16:59:01 LOG7[438154:1104947520]: SSL alert (read): warning: close notify 2011.08.22 16:59:01 LOG7[438154:1104947520]: SSL closed on SSL_read 2011.08.22 16:59:01 LOG7[438154:1104947520]: Sending socket write shutdown 2011.08.22 16:59:01 LOG5[438154:1104947520]: Connection closed: 49445 bytes sent to SSL, 8175 bytes sent to socket -- Thank You, Andrew Heuneman Senior Systems Administrator Reading Plus®/Taylor Associates Helping students become proficient silent readers. <http://twitter.com/readingplus> <http://www.facebook.com/pages/Reading-Plus/165970877038>

2 1

Internal error after reload modified server mode section
by Javier 07 Sep '11

07 Sep '11

Hi, I didn't found this question, so I hope isn't duplicate. I'm using the latest stable 4.42 version and I'm running Windows 2000 and I want to ask is this is a normal behavior. What is happening is that, after change the server section in the config file (the rest are all client mode), if it has been any connection to that server (and only if it has been a connection), when I reload the modified conf, it gives...: INTERNAL ERROR: get_alloc_list_ptr: Bad magic str_stats: 380 block(s), 96684 byte(s) This happens after no errors in the load (except for DH being hardcoded), SSL context initialized Configuration successful after open every service (clients and server) and then, after the error, closes every service. Trying to reload the conf, doesn't nothing, just stunnel hangs (is not responding to reloads). The only way to restart all is to close stunnel and reopen. Then, the configuration is accepted without problems and works perfectly. This happens, as I said, after modify the conf file and if it was a connection. The section is related to a local test server, where I modify the IP and/or ports where to accept and where to connect. As general options I have: debug = debug delay = yes And the server section: [mySSL] client = no cert = mytestpem.pem accept = 127.0.0.1:443 connect = 127.0.0.1:80 What I'm doing wrong? Is the normal behavior? Is it a new bug? Regards, Javier.

3 5

Compiling Stunnel For Windows
by Albert 05 Sep '11

05 Sep '11

Hi, I want to compile stunnel for windows, to include the XMPP patch (that I managed to apply cleanly to the current stunnel code) that I saw on the mailing list earlier as I need to get it to connect to a SSL only server in client mode. I followed the instructions in the Install.w32 file 1) apt-get install gcc-mingw32 - Worked fine with no issues 2) Download the recent zlib from http://www.zlib.net/ , Update the following definitions in win32/Makefile.gcc file: etc *snip* Worked fine as far as I can tell 3) Download the recent OpenSSL in unpack it to /usr/src/ directory., cd /usr/src && tar zvxf ~/openssl-(version).tar.gz I got OpenSSL 1.0.0d 4) Build OpenSSL with cross_mingw32.sh script. I ran instead ./Configure --cross-compile-prefix=i586-mingw32msvc- mingw shared zlib-dynamic && make it seems to have worked fine 5) Download and unpack stunnel-(version).tar.gz. -got stunnel 4.42 6) Configure stunnel. cd stunnel-(version) && ./configure --with-ssl=/path/to/openssl-(version) This step worked fine 7) Build windows executable. cd src && make stunnel.exe The first few compile steps work fine, but I get stuck on: ...i586-mingw32msvc/bin/ld: cannot find -lcrypto.dll I did a few hours of troubleshooting and trying to move a few files around but couldn't find crypto.dll anywhere I'm running a fresh install of Ubuntu 11.04 in a VM if that makes any difference. Help would be greatly appreciated :) Thanks, Albert

2 1

Changing the secure tunnel 'encryption' method ....
by VANDOOREN, Pascal 02 Sep '11

02 Sep '11

Hello Everyone, I have a question, Why use you a certificate and not a hash AES, or other crypting algorithm to make a secure tunnel ? Because i use not a validate cert and i can only generate a cert valid for maximum 1year ... In the config file you can let the choice to use the cert or hash .. It's only a idea Kind regards Pascal -- This e-mail is confidential. If you are not the addressee or an authorized recipient of this message, any distribution, copying, publication or use of this information for any purpose is prohibited. Please notify the sender immediately by e-mail and then delete this message. Ce message est confidentiel. Si vous n'etes pas le destinataire designe de ce message ou une personne autorisee a l'utiliser, toute distribution, copie, publication ou usage a quelques fins que ce soit des informations contenues dans ce message sont interdits. Merci d'informer immediatement l'expediteur par messagerie electronique et d'ensuite detruire ce message. --

1 0