use CreateProcess instead of system to run scripts.

That prevents the pop up of terminal windows.

[dwmw2: Use -ETIMEDOUT instead of -ETIME which doesn't seem to be present in
        my Fedora 20 MinGW setup. Do not prepend 'cscript' to vpnc_script
        string now that we invoke it that way unconditionally.]

Signed-off-by: Nikos Mavrogiannopoulos <nmav@redhat.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

use recv() and closesocket() on the cmd_fd on windows

[dwmw2: Reduce #ifdefs by using recv() everywhere, not just on Windows]

Signed-off-by: Nikos Mavrogiannopoulos <nmav@redhat.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

use windows types and return codes for socket errors

That is, use SOCKET instead of int for cmd_pipe, and return INVALID
socket for cmd_pipe.

Signed-off-by: Nikos Mavrogiannopoulos <nmav@redhat.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Import translations from GNOME

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Check to avoid calling iconv_close(-1) in openconnect_vpninfo_free

Signed-off-by: Chad Catlett <chad@catlett.info>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Use inet_pton() to provide inet_aton() functionality

This is cleaner and simpler than the original compat code, and avoids
the strange issues that Nikos was reporting on his Windows 7
installation with inet_addr() not working correctly.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix DeviceIoControl() calls to make NDIS6 TAP-Windows driver happy

All calls appear to need both input and output buffers.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Merge branch 'hotp-java-v1' of git://github.com/cernekee/openconnect

Clean up OpenSSL key loading

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Remove cert_type field from struct openconnect_info

We stopped actually using this a long time ago (4.00, June 2012).

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Stop using obsolete vpninfo->cert_type in OpenSSL code

This wants cleaning up a little but that can come in separate commits.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Stop using obsolete vpninfo->cert_type in GnuTLS code

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Use BIO_CLOSE when importing PEM private key with OpenSSL

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

www: Add notes on SecurID PIN usage

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

jni: Fix onTokenLock() invocation from native code

Fix the type signature to reflect 0 arguments, and make sure the int
return value propagates back to the library.

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

java: Add constant for OC_TOKEN_MODE_HOTP

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

Resync translations with sources

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Reread token file in lock_token()

If someone else has used a HOTP token between the initial invocation of
openconnect and the time the token gets used, we want to use the current
value of the counter not the original value.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Disable soft token on failure

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add software token documentation

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Move stoken code to stoken.c

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Move HOTP/TOTP code to oath.c

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Shift TOTP/HOTP functions in library.c under a single #ifdef HAVE_LIBOATH

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Shift TOTP/HOTP functions in auth.c under a single #ifdef HAVE_LIBOATH

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add PSKC support

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add token lock/unlock callbacks to Java bindings

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Update changelog

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Actually lock and unlock/write back HOTP tokens when counter updates

This was kind of the point...

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Support writing back to token files from openconnect(8) itself

This is quite basic; we probably ought to re-read the file and call
openconnect_set_token_mode() again if it's changed, and we ought to
have locking on the file to prevent concurrent updates. But this is a
good start.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add openconnect_set_token_callbacks() API

This allows the application to lock the file (or other) storage when
a token is about to be generated, then to unlock it and write the new
counter back to the file when we're done.

This gives us a clean way to use HOTP tokens safely.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Remember what form we got the HOTP secret in

We'll care about this shortly...

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Allow HOTP/TOTP secrets to be specified in hex form

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Allow whitespace at end of HOTP and TOTP tokens

When we're reading them from a file, in particular, we want to tolerate
having a newline at the end.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

For HOTP tokens increment vpninfo->token_time in do_gen_hotp_code()

This allows us to do a lock/increment/unlock cycle in one place.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Update changelog

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Be consistent about spelling of cancel{l,}ed

The source uses 'cancelled', and the en_US file can be 'canceled' if that's
the way they prefer it.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Convert all args using GetCommandLineW() on Windows

Checking for ? isn't enough — some characters are 'loosely' translated on
the 8-bit command line judging by https://bugzilla.redhat.com/1124436#c2

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Update translations from GNOME

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Minor SSPI fixes

When using NTLM, the payload sometimes does get encrypted, so preserve it.
And after DecryptMessage() don't free the buffer. It'll have been decrypted
in-place.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix ntlm_auth handling to accept AF in auth reply

This is what it should have been giving all along, if it weren't for
Samba bug #10691. Also report the error better if we do get one.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix endless loop when automatic NTLM auth fails

We never *do* actually fall back from automatic to manual NTLM auth since
our criteria for using each are mutually exclusive. So arguably we could
simplify things somewhat and just set the state to AUTH_FAILED if the SSO
version fails. But let's preserve the option to fall back later, if we do
start prompting for a password on demand.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix SSPI SOCKS auth crash on double-cleanup

Using the state for this is horrid; we should make it idempotent like
the GSSAPI one is.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix SOCKS SSPI auth to send final token client->server if needed

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix SOCKS GSSAPI auth to send final token client->server if needed

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Update translations from GNOME

Silence iconv() compiler warnings on systems which expect (const char **)

Some systems expect the src argument to be const, so just cast it to (void *)
to shut them up.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Use O_BINARY for xmlconfig file

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix UTF-8 handling of config file name

If we're going to convert the filename *from* UTF-8 when opening it, then
we really ought to be convert it *to* UTF-8 first. This would have shown
up if invoking openconnect in a non-UTF-8 locale, with a non-ASCII
filename.

On POSIX systems this whole thing is a no-op, since converting from any
arbitrary 8-bit charset to UTF-8 and back again effectively does nothing.

However, on Windows the convert_arg_to_utf8() function doesn't just work
on the input string; it looks aside with GetCommandLineW() to find the
*original* text on the command line, and uses that instead. So we really
do have to do the double "conversion" — even when it's simply converting to
UTF-8, *immediately* converting back again (in fopen_utf8()), and *never*
using it again.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix NetBSD compile warnings with isspace()

As mentioned in the comment in openssl.c, we need to cast the arguments
for ctype functions first to (unsigned char) and then to (int) to make
NedBSD happy.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Use openconnect_[f]open_utf8() for xmlconfig, pidfile, config_file

Each option was sanity-tested under Linux.

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

ssl: Support more read/write modes in openconnect_fopen_utf8()

This adds ASCII "r"/"w" and binary "wb".

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

ssl: Allow openconnect_open_utf8() to be used with O_CREAT

Pass in the optional mode/pmode argument; per the API docs this will be
ignored if O_CREAT is not used.

The _WIN32 version is based on [1] but untested.

[1] http://msdn.microsoft.com/en-us/library/z0kc8e3z.aspx

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

Rename and export open_utf8(), fopen_utf8()

Make these functions available to main.c and xml.c.

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

auth: Refactor stoken form handling

Split the devid/pass form from the PIN form, to allow the use of new "v3"
tokens from the command line.  On newer libstoken versions, query the
PIN status and token interval time; use this information to figure out
whether to request a PIN for concatenation (ala hard tokens) and whether
to use a 30- or 60-second delta in response to the next tokencode prompt.

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

auth: Rearrange stoken support code

Put everything under one giant #ifdef to make it easier to add helper
functions later.  No change to the program logic.

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

main: Allow reading --token-secret from a file

If the token string starts with '@' or '/', assume it refers to a
filename.  None of the current token backends would recognize a string
starting with '@' or '/'.

The main use cases are:

 - Keeping token secrets from showing up in "ps" output
 - Allowing the use of SecurID SDTID XML files without import/conversion

This accepts "raw" HOTP/TOTP/SecurID tokens, not .stokenrc files.
Therefore it cannot use stoken_import_rcfile().

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

main: Restore tty state if password prompt is aborted

On Linux, hitting ^C on any of the password prompts now leaves ECHO
disabled, so the user needs to run "stty sane".  Restore the correct
settings prior to exiting.

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

xml: Call openconnect_parse_url() on <HostAddress> entries

Per the Cisco docs[1] these should just contain a FQDN, not a full URL.
But calling openconnect_parse_url() populates other internal fields, like
protocol and port; the latter is misreported as 0 right now:

    Host "MIT" has address "vpn.mit.edu"
    POST https://vpn.mit.edu:0/
    Attempting to connect to server 18.9.37.5:443

[1] http://www.cisco.com/c/en/us/td/docs/security/vpn_client/anyconnect/anyconnect30/administration/guide/anyconnectadmin30/ac13vpnxmlref.html

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

xml: Trim whitespace from xmlconfig entries

The Cisco implementations seem to be more tolerant of extraneous
whitespace in these files; for us, it can break hostname resolution:

    Host "MIT" has address "vpn.mit.edu "
    POST https://vpn.mit.edu :0/
    getaddrinfo failed for host 'vpn.mit.edu ': Name or service not known
    Failed to open HTTPS connection to vpn.mit.edu
    Failed to obtain WebVPN cookie

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

xml: Make a generic function to read a file into a string

This replaces load_xml_conf_file(); it can be called by main.c for other
purposes.

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>

Update GUI page

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Remove stray perror() in convert_str()

This was for debugging only.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Wrap ifreq_set_ifname() in #ifndef __sun__ to shut the compiler up there

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix csd_wrapper handling for non-UTF8 locales

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Check inputs for invalid UTF-8

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Make buf_append_utf16le() suitable for validation-only

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix buf_append_utf16le() error handling harder

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Remove static out_pkt

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix asprintf() handling in openconnect_passphrase_from_fsid()

It returns -1 on error. Not non-zero. We were always returning -ENOMEM but
nobody cared; if this fails they'll just ignore it and let us ask the user
for the passphrase anyway.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Handle UTF-8 sslkey in Windows openconnect_passphrase_from_fsid

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Handle UTF-8 sslkey in Linux openconnect_passphrase_from_fsid

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Handle UTF-8 sslkey in BSD openconnect_passphrase_from_fsid

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Update changelog

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add web page about charset handling

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Convert last remaining OpenSSL code to open_utf8()

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Convert cafile to legacy charset for OpenSSL

On Windows this is OK, but legacy UNIX needs the filename to be converted
back to the legacy charset.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Avoid SSL_CTX_use_RSAPrivateKey_file() because it doesn't handle UTF-8.

Open the file for ourselves with fopen_utf8() then do it manually.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Extend reload_pem_cert() to become load_cert_chain_file()

Because SSL_CTX_use_certificate_chain_file() doesn't cope with UTF-8
filenames, and we have a couple of other minor reasons for wanting to
reimplement it anyway (which is why we had reload_pem_cert() in the first
place).

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Factor out OpenSSL install_extra_certs() from load_pkcs12_certificate()

We'll want to use this again in a moment...

Also fix up the translations, since nobody had translated 'PKCS#12' anyway.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Convert vpnc_script to legacy charset (or UTF-16 for Windows)

The script_config_tun() function ends up being *just* different enough
between the two implementations that it's not quite worth trying to
reduce the duplication.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Convert $CISCO_BANNER environment variable to legacy charset

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Convert vpninfo->ifname to legacy charset for $TUNDEV and configuration

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Use open_utf8() in GnuTLS load_datum()

This should cover all cert-related files including vpninfo->cafile.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add open_utf8() and fopen_utf8() functions for opening files

These will convert to legacy charset as required (or UTF-16 for Windows).

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add openconnect_utf8_to_legacy() helper function for charset conversion

In some cases, the library itself really is going to have to know about
legacy non-UTF8 locales — when opening files or setting environment
variables, for example.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix some arguments missing dup_config_arg()

These weren't getting convert to UTF-8.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Take a copy of the string that nl_langinfo() returns

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Use setlocale() unconditionally, complain if non-UTF8 locale and no iconv

Hopefully setlocale() is ubiquitous. Certainly nl_langinfo() seems to be
more often available than iconv(), so we can certainly complain if we are
run in a legacy non-UTF8 locale and we *don't* have conversion support.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add missing 'Proxy CONNECT request failed: %d' translation back

We had a '%s' version of this, so just put that one back with the
appropriate substitution.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Add conversion to UTF-8 of command-line arguments

Hm, this isn't ideal in all cases; we end up converting filenames into
UTF-8 too, and we don't (yet) have facilities to convert *back* to the
local charset when trying to access them. And it's not entirely clear
we really want to have to.

But it does fix other things like '--interface=TAP♥', and is consistent
with the methodology of converting *everything* on the way in and out,
and being entirely consistent with UTF-8 internally.

We *might* want to add special-case hacks for keeping filenames in the
legacy charset internally, but then again we'd only have to convert *to*
UTF-8 for printing them if we did that.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Introduce dup_config_arg() macro

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Move Windows UTF-8 stdio functions up to live with POSIX ones

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fix tun device matching with --interface on Windows

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Use _wputenv() for Windows environment

This fixes the handling of non-ASCII (and out-of-codepage) tundev names.
It looks like cscript/vpnc-script-win.js/netsh all Just Work™ at least
here on Windows 7 with my 'TAP♥' device.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Fixes for buf_append_utf16le()

Make it set buf_error() when it returns -EINVAL, to ensure callers will
catch the problem. Previously, ntlm_set_string_binary() and  sspi_setup()
would not have noticed.

Also, NUL-terminate the buffer. Not accounting for it, but just putting
two zero bytes after the string so that it can be used as a
NUL-terminated (wchar_t *). Not entirely sure how sspi_setup() was
working before this, except perhaps pure luck.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Handle UTF-8 output on Windows too

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Handle UTF-8 input on Windows

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Separate read_stdin() implementations for Windows vs. POSIX

They're about to get even *more* different...

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Convert to UTF-8 when reading from stdin

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Attempt to convert output to locale charset from UTF-8

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>