1
0
mirror of git://git.gnupg.org/gnupg.git synced 2024-11-04 20:38:50 +01:00
gnupg/tools/gpg-pair-tool.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1966 lines
55 KiB
C
Raw Normal View History

tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
/* gpg-pair-tool.c - The tool to run the pairing protocol.
* Copyright (C) 2018 g10 Code GmbH
*
* This file is part of GnuPG.
*
* This file is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as
* published by the Free Software Foundation; either version 2.1 of
* the License, or (at your option) any later version.
*
* This file is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with this program; if not, see <https://www.gnu.org/licenses/>.
* SPDX-License-Identifier: LGPL-2.1-or-later
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
*/
/* Protocol:
*
* Initiator Responder
* | |
* | COMMIT |
* |-------------------->|
* | |
* | DHPART1 |
* |<--------------------|
* | |
* | DHPART2 |
* |-------------------->|
* | |
* | CONFIRM |
* |<--------------------|
* | |
*
* The initiator creates a keypair (PKi,SKi) and sends this COMMIT
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
* message to the responder:
*
* 7 byte Magic, value: "GPG-pa1"
* 1 byte MessageType, value 1 (COMMIT)
* 8 byte SessionId, value: 8 random bytes
* 1 byte Realm, value 1
* 2 byte reserved, value 0
* 5 byte ExpireTime, value: seconds since Epoch as an unsigned int.
* 32 byte Hash(PKi)
*
* The initiator also needs to locally store the sessionid, the realm,
* the expiration time, the keypair and a hash of the entire message
* sent.
*
* The responder checks that the received message has not expired and
* stores sessionid, realm, expiretime and the Hash(PKi). The
* Responder then creates and locally stores its own keypair (PKr,SKr)
* and sends the DHPART1 message back:
*
* 7 byte Magic, value: "GPG-pa1"
* 1 byte MessageType, value 2 (DHPART1)
* 8 byte SessionId from COMMIT message
* 32 byte PKr
* 32 byte Hash(Hash(COMMIT) || DHPART1[0..47])
*
* Note that Hash(COMMIT) is the hash over the entire received COMMIT
* message. DHPART1[0..47] are the first 48 bytes of the created
* DHPART1 message.
*
* The Initiator receives the DHPART1 message and checks that the hash
* matches. Although this hash is easily malleable it is later in the
* protocol used to assert the integrity of all messages. The
* Initiator then computes the shared master secret from its SKi and
* the received PKr. Using this master secret several keys are
* derived:
*
* - HMACi-key using the label "GPG-pa1-HMACi-key".
* - SYMx-key using the label "GPG-pa1-SYMx-key"
*
* For details on the KDF see the implementation of the function kdf.
* The master secret is stored securely in the local state. The
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
* DHPART2 message is then created and send to the Responder:
*
* 7 byte Magic, value: "GPG-pa1"
* 1 byte MessageType, value 3 (DHPART2)
* 8 byte SessionId from COMMIT message
* 32 byte PKi
* 32 byte MAC(HMACi-key, Hash(DHPART1) || DHPART2[0..47] || SYMx-key)
*
* The Responder receives the DHPART2 message and checks that the hash
* of the received PKi matches the Hash(PKi) value as received earlier
* with the COMMIT message. The Responder now also computes the
* shared master secret from its SKr and the received PKi and derives
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
* the keys:
*
* - HMACi-key using the label "GPG-pa1-HMACi-key".
* - HMACr-key using the label "GPG-pa1-HMACr-key".
* - SYMx-key using the label "GPG-pa1-SYMx-key"
* - SAS using the label "GPG-pa1-SAS"
*
* With these keys the MAC from the received DHPART2 message is
* checked. On success a SAS is displayed to the user and a CONFIRM
* message send back:
*
* 7 byte Magic, value: "GPG-pa1"
* 1 byte MessageType, value 4 (CONFIRM)
* 8 byte SessionId from COMMIT message
* 32 byte MAC(HMACr-key, Hash(DHPART2) || CONFIRM[0..15] || SYMx-key)
*
* The Initiator receives this CONFIRM message, gets the master shared
* secret from its local state and derives the keys. It checks the
* MAC in the received CONFIRM message and ask the user to enter
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
* the SAS as displayed by the responder. Iff the SAS matches the
* master key is flagged as confirmed and the Initiator may now use a
* derived key to send encrypted data to the Responder.
*
* In case the Responder also needs to send encrypted data we need to
* introduce another final message to tell the responder that the
* Initiator validated the SAS.
*
* TODO: Encrypt the state files using a key stored in gpg-agent's cache.
*
*/
#include <config.h>
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <dirent.h>
#include <stdarg.h>
#include "../common/util.h"
#include "../common/status.h"
#include "../common/i18n.h"
#include "../common/sysutils.h"
#include "../common/init.h"
#include "../common/name-value.h"
/* Constants to identify the commands and options. */
enum cmd_and_opt_values
{
aNull = 0,
oQuiet = 'q',
oVerbose = 'v',
oOutput = 'o',
oArmor = 'a',
aInitiate = 400,
aRespond = 401,
aGet = 402,
aCleanup = 403,
oDebug = 500,
oStatusFD,
oHomedir,
oSAS,
oDummy
};
/* The list of commands and options. */
static gpgrt_opt_t opts[] = {
ARGPARSE_group (300, ("@Commands:\n ")),
ARGPARSE_c (aInitiate, "initiate", N_("initiate a pairing request")),
ARGPARSE_c (aRespond, "respond", N_("respond to a pairing request")),
ARGPARSE_c (aGet, "get", N_("return the keys")),
ARGPARSE_c (aCleanup, "cleanup", N_("remove expired states etc.")),
ARGPARSE_group (301, ("@\nOptions:\n ")),
ARGPARSE_s_n (oVerbose, "verbose", N_("verbose")),
ARGPARSE_s_n (oQuiet, "quiet", N_("be somewhat more quiet")),
ARGPARSE_s_n (oArmor, "armor", N_("create ascii armored output")),
ARGPARSE_s_s (oSAS, "sas", N_("|SAS|the SAS as shown by the peer")),
ARGPARSE_s_s (oDebug, "debug", "@"),
ARGPARSE_s_s (oOutput, "output", N_("|FILE|write the request to FILE")),
ARGPARSE_s_i (oStatusFD, "status-fd", N_("|FD|write status info to this FD")),
ARGPARSE_s_s (oHomedir, "homedir", "@"),
ARGPARSE_end ()
};
/* We keep all global options in the structure OPT. */
static struct
{
int verbose;
unsigned int debug;
int quiet;
int armor;
const char *output;
estream_t statusfp;
unsigned int ttl;
const char *sas;
} opt;
/* Debug values and macros. */
#define DBG_MESSAGE_VALUE 2 /* Debug the messages. */
#define DBG_CRYPTO_VALUE 4 /* Debug low level crypto. */
#define DBG_MEMORY_VALUE 32 /* Debug memory allocation stuff. */
#define DBG_MESSAGE (opt.debug & DBG_MESSAGE_VALUE)
#define DBG_CRYPTO (opt.debug & DBG_CRYPTO_VALUE)
/* The list of supported debug flags. */
static struct debug_flags_s debug_flags [] =
{
{ DBG_MESSAGE_VALUE, "message" },
{ DBG_CRYPTO_VALUE , "crypto" },
{ DBG_MEMORY_VALUE , "memory" },
{ 0, NULL }
};
/* The directory name below the cache dir to store paring states. */
#define PAIRING_STATE_DIR "state"
/* Message types. */
#define MSG_TYPE_COMMIT 1
#define MSG_TYPE_DHPART1 2
#define MSG_TYPE_DHPART2 3
#define MSG_TYPE_CONFIRM 4
/* Realm values. */
#define REALM_STANDARD 1
/* Local prototypes. */
static void wrong_args (const char *text) GPGRT_ATTR_NORETURN;
static void xnvc_set_printf (nvc_t nvc, const char *name, const char *format,
...) GPGRT_ATTR_PRINTF(3,4);
static void *hash_data (void *result, size_t resultsize,
...) GPGRT_ATTR_SENTINEL(0);
static void *hmac_data (void *result, size_t resultsize,
const unsigned char *key, size_t keylen,
...) GPGRT_ATTR_SENTINEL(0);
static gpg_error_t command_initiate (void);
static gpg_error_t command_respond (void);
static gpg_error_t command_cleanup (void);
static gpg_error_t command_get (const char *sessionidstr);
/* Print usage information and provide strings for help. */
static const char *
my_strusage( int level )
{
const char *p;
switch (level)
{
case 9: p = "LGPL-2.1-or-later"; break;
case 11: p = "gpg-pair-tool"; break;
case 12: p = "@GNUPG@"; break;
case 13: p = VERSION; break;
case 14: p = GNUPG_DEF_COPYRIGHT_LINE; break;
case 17: p = PRINTABLE_OS_NAME; break;
case 19: p = _("Please report bugs to <@EMAIL@>.\n"); break;
case 1:
case 40:
p = ("Usage: gpg-pair-tool [command] [options] [args] (-h for help)");
break;
case 41:
p = ("Syntax: gpg-pair-tool [command] [options] [args]\n"
"Client to run the pairing protocol\n");
break;
default: p = NULL; break;
}
return p;
}
static void
wrong_args (const char *text)
{
es_fprintf (es_stderr, _("usage: %s [options] %s\n"),
gpgrt_strusage (11), text);
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
exit (2);
}
/* Set the status FD. */
static void
set_status_fd (int fd)
{
static int last_fd = -1;
if (fd != -1 && last_fd == fd)
return;
if (opt.statusfp && opt.statusfp != es_stdout && opt.statusfp != es_stderr)
es_fclose (opt.statusfp);
opt.statusfp = NULL;
if (fd == -1)
return;
if (fd == 1)
opt.statusfp = es_stdout;
else if (fd == 2)
opt.statusfp = es_stderr;
else
opt.statusfp = es_fdopen (fd, "w");
if (!opt.statusfp)
{
log_fatal ("can't open fd %d for status output: %s\n",
fd, gpg_strerror (gpg_error_from_syserror ()));
}
last_fd = fd;
}
/* Write a status line with code NO followed by the output of the
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
* printf style FORMAT. The caller needs to make sure that LFs and
* CRs are not printed. */
static void
write_status (int no, const char *format, ...)
{
va_list arg_ptr;
if (!opt.statusfp)
return; /* Not enabled. */
es_fputs ("[GNUPG:] ", opt.statusfp);
es_fputs (get_status_string (no), opt.statusfp);
if (format)
{
es_putc (' ', opt.statusfp);
va_start (arg_ptr, format);
es_vfprintf (opt.statusfp, format, arg_ptr);
va_end (arg_ptr);
}
es_putc ('\n', opt.statusfp);
}
/* gpg-pair-tool main. */
int
main (int argc, char **argv)
{
gpg_error_t err;
gpgrt_argparse_t pargs = { &argc, &argv };
enum cmd_and_opt_values cmd = 0;
opt.ttl = 8*3600; /* Default to 8 hours. */
gnupg_reopen_std ("gpg-pair-tool");
gpgrt_set_strusage (my_strusage);
log_set_prefix ("gpg-pair-tool", GPGRT_LOG_WITH_PREFIX);
/* Make sure that our subsystems are ready. */
i18n_init();
init_common_subsystems (&argc, &argv);
/* Parse the command line. */
while (gpgrt_argparse (NULL, &pargs, opts))
{
switch (pargs.r_opt)
{
case oQuiet: opt.quiet = 1; break;
case oVerbose: opt.verbose++; break;
case oArmor: opt.armor = 1; break;
case oDebug:
if (parse_debug_flag (pargs.r.ret_str, &opt.debug, debug_flags))
{
pargs.r_opt = ARGPARSE_INVALID_ARG;
pargs.err = ARGPARSE_PRINT_ERROR;
}
break;
case oOutput:
opt.output = pargs.r.ret_str;
break;
case oStatusFD:
set_status_fd (translate_sys2libc_fd_int (pargs.r.ret_int, 1));
break;
case oHomedir:
gnupg_set_homedir (pargs.r.ret_str);
break;
case oSAS:
opt.sas = pargs.r.ret_str;
break;
case aInitiate:
case aRespond:
case aGet:
case aCleanup:
if (cmd && cmd != pargs.r_opt)
log_error (_("conflicting commands\n"));
else
cmd = pargs.r_opt;
break;
default: pargs.err = ARGPARSE_PRINT_WARNING; break;
}
}
gpgrt_argparse (NULL, &pargs, NULL); /* Release internal state. */
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
/* Print a warning if an argument looks like an option. */
if (!opt.quiet && !(pargs.flags & ARGPARSE_FLAG_STOP_SEEN))
{
int i;
for (i=0; i < argc; i++)
if (argv[i][0] == '-' && argv[i][1] == '-')
log_info (("NOTE: '%s' is not considered an option\n"), argv[i]);
}
gpgrt_argparse (NULL, &pargs, NULL); /* Free internal memory. */
if (opt.sas)
{
if (strlen (opt.sas) != 11
|| !digitp (opt.sas+0) || !digitp (opt.sas+1) || !digitp (opt.sas+2)
|| opt.sas[3] != '-'
|| !digitp (opt.sas+4) || !digitp (opt.sas+5) || !digitp (opt.sas+6)
|| opt.sas[7] != '-'
|| !digitp (opt.sas+8) || !digitp (opt.sas+9) || !digitp (opt.sas+10))
log_error ("invalid formatted SAS\n");
}
/* Stop if any error, including ARGPARSE_PRINT_WARNING, occurred. */
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
if (log_get_errorcount (0))
exit (2);
if (DBG_CRYPTO)
gcry_control (GCRYCTL_SET_DEBUG_FLAGS, 1|2);
/* Now run the requested command. */
switch (cmd)
{
case aInitiate:
if (argc)
wrong_args ("--initiate");
err = command_initiate ();
break;
case aRespond:
if (argc)
wrong_args ("--respond");
err = command_respond ();
break;
case aGet:
if (argc > 1)
wrong_args ("--respond [sessionid]");
err = command_get (argc? *argv:NULL);
break;
case aCleanup:
if (argc)
wrong_args ("--cleanup");
err = command_cleanup ();
break;
default:
gpgrt_usage (1);
err = 0;
break;
}
if (err)
write_status (STATUS_FAILURE, "- %u", err);
else if (log_get_errorcount (0))
write_status (STATUS_FAILURE, "- %u", GPG_ERR_GENERAL);
else
write_status (STATUS_SUCCESS, NULL);
return log_get_errorcount (0)? 1:0;
}
/* Wrapper around nvc_new which terminates in the error case. */
static nvc_t
xnvc_new (void)
{
nvc_t c = nvc_new ();
if (!c)
log_fatal ("error creating NVC object: %s\n",
gpg_strerror (gpg_error_from_syserror ()));
return c;
}
/* Wrapper around nvc_set which terminates in the error case. */
static void
xnvc_set (nvc_t nvc, const char *name, const char *value)
{
gpg_error_t err = nvc_set (nvc, name, value);
if (err)
log_fatal ("error updating NVC object: %s\n", gpg_strerror (err));
}
/* Call vnc_set with (BUFFER, BUFLEN) converted to a hex string as
* value. Terminates in the error case. */
static void
xnvc_set_hex (nvc_t nvc, const char *name, const void *buffer, size_t buflen)
{
char *hex;
hex = bin2hex (buffer, buflen, NULL);
if (!hex)
xoutofcore ();
strlwr (hex);
xnvc_set (nvc, name, hex);
xfree (hex);
}
/* Call nvc_set with a value created from the string generated using
* the printf style FORMAT. Terminates in the error case. */
static void
xnvc_set_printf (nvc_t nvc, const char *name, const char *format, ...)
{
va_list arg_ptr;
char *buffer;
va_start (arg_ptr, format);
if (gpgrt_vasprintf (&buffer, format, arg_ptr) < 0)
log_fatal ("estream_asprintf failed: %s\n",
gpg_strerror (gpg_error_from_syserror ()));
va_end (arg_ptr);
xnvc_set (nvc, name, buffer);
xfree (buffer);
}
/* Return the string for the first entry in NVC with NAME. If NAME is
* missing, an empty string is returned. The returned string is a
* pointer into NVC. */
static const char *
xnvc_get_string (nvc_t nvc, const char *name)
{
nve_t item;
if (!nvc)
return "";
item = nvc_lookup (nvc, name);
if (!item)
return "";
return nve_value (item);
}
/* Return a string for MSGTYPE. */
const char *
msgtypestr (int msgtype)
{
switch (msgtype)
{
case MSG_TYPE_COMMIT: return "Commit";
case MSG_TYPE_DHPART1: return "DHPart1";
case MSG_TYPE_DHPART2: return "DHPart2";
case MSG_TYPE_CONFIRM: return "Confirm";
}
return "?";
}
/* Private to {get,set}_session_id(). */
static struct {
int initialized;
unsigned char sessid[8];
} session_id;
/* Return the 8 octet session. */
static unsigned char *
get_session_id (void)
{
if (!session_id.initialized)
{
session_id.initialized = 1;
gcry_create_nonce (session_id.sessid, sizeof session_id.sessid);
}
return session_id.sessid;
}
static void
set_session_id (const void *sessid, size_t len)
{
log_assert (!session_id.initialized);
if (len > sizeof session_id.sessid)
len = sizeof session_id.sessid;
memcpy (session_id.sessid, sessid, len);
if (len < sizeof session_id.sessid)
memset (session_id.sessid+len, 0, sizeof session_id.sessid - len);
session_id.initialized = 1;
}
/* Return a string with the hexified session id. */
static const char *
get_session_id_hex (void)
{
static char hexstr[16+1];
bin2hex (get_session_id (), 8, hexstr);
strlwr (hexstr);
return hexstr;
}
/* Return a fixed string with the directory used to store the state of
* pairings. On error a diagnostic is printed but the file name is
* returned anyway. It is expected that the expected failure of the
* following open is responsible for error handling. */
static const char *
get_pairing_statedir (void)
{
static char *fname;
gpg_error_t err = 0;
char *tmpstr;
struct stat statbuf;
if (fname)
return fname;
fname = make_filename (gnupg_homedir (), GNUPG_CACHE_DIR, NULL);
if (gnupg_stat (fname, &statbuf) && errno == ENOENT)
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
{
if (gnupg_mkdir (fname, "-rwx"))
{
err = gpg_error_from_syserror ();
log_error (_("can't create directory '%s': %s\n"),
fname, gpg_strerror (err) );
}
else if (!opt.quiet)
log_info (_("directory '%s' created\n"), fname);
}
tmpstr = make_filename (fname, PAIRING_STATE_DIR, NULL);
xfree (fname);
fname = tmpstr;
if (gnupg_stat (fname, &statbuf) && errno == ENOENT)
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
{
if (gnupg_mkdir (fname, "-rwx"))
{
if (!err)
{
err = gpg_error_from_syserror ();
log_error (_("can't create directory '%s': %s\n"),
fname, gpg_strerror (err) );
}
}
else if (!opt.quiet)
log_info (_("directory '%s' created\n"), fname);
}
return fname;
}
/* Open the pairing state file. SESSIONID is a 8 byte buffer with the
* session-id. If CREATE_FLAG is set the file is created and will
* always return a valid stream. If CREATE_FLAG is not set the file
* is opened for reading and writing. If the file does not exist NULL
* is return; in all other error cases the process is terminated. If
* R_FNAME is not NULL the name of the file is stored there and the
* caller needs to free it. */
static estream_t
open_pairing_state (const unsigned char *sessionid, int create_flag,
char **r_fname)
{
gpg_error_t err;
char *fname, *tmpstr;
estream_t fp;
/* The filename is the session id with a "pa1" suffix. Note that
* the state dir may eventually be used for other purposes as well
* and thus the suffix identifies that the file belongs to this
* tool. We use lowercase file names for no real reason. */
tmpstr = bin2hex (sessionid, 8, NULL);
if (!tmpstr)
xoutofcore ();
strlwr (tmpstr);
fname = xstrconcat (tmpstr, ".pa1", NULL);
xfree (tmpstr);
tmpstr = make_filename (get_pairing_statedir (), fname, NULL);
xfree (fname);
fname = tmpstr;
fp = es_fopen (fname, create_flag? "wbx,mode=-rw": "rb+,mode=-rw");
if (!fp)
{
err = gpg_error_from_syserror ();
if (create_flag)
{
/* We should always be able to create a file. Also we use a
* 64 bit session id, it is theoretically possible that such
* a session already exists. However, that is rare enough
* and thus the fatal error message should still be okay. */
log_fatal ("can't create '%s': %s\n", fname, gpg_strerror (err));
}
else if (gpg_err_code (err) == GPG_ERR_ENOENT)
{
/* That is an expected error; return NULL. */
}
else
{
log_fatal ("can't open '%s': %s\n", fname, gpg_strerror (err));
}
}
if (r_fname)
*r_fname = fname;
else
xfree (fname);
return fp;
}
/* Write the state to a possible new state file. */
static void
write_state (nvc_t state, int create_flag)
{
gpg_error_t err;
char *fname = NULL;
estream_t fp;
fp = open_pairing_state (get_session_id (), create_flag, &fname);
log_assert (fp);
err = nvc_write (state, fp);
if (err)
{
es_fclose (fp);
gnupg_remove (fname);
log_fatal ("error writing '%s': %s\n", fname, gpg_strerror (err));
}
/* If we did not create the file, we need to truncate the file. */
if (!create_flag && ftruncate (es_fileno (fp), es_ftello (fp)))
{
err = gpg_error_from_syserror ();
log_fatal ("error truncating '%s': %s\n", fname, gpg_strerror (err));
}
if (es_ferror (fp) || es_fclose (fp))
{
err = gpg_error_from_syserror ();
es_fclose (fp);
gnupg_remove (fname);
log_fatal ("error writing '%s': %s\n", fname, gpg_strerror (err));
}
}
/* Read the state into a newly allocated state object and store that
* at R_STATE. If no state is available GPG_ERR_NOT_FOUND is returned
* and as with all errors NULL is stored at R_STATE. SESSIONID is an
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
* input with the 8 session id. */
static gpg_error_t
read_state (nvc_t *r_state)
{
gpg_error_t err;
char *fname = NULL;
estream_t fp;
nvc_t state = NULL;
nve_t item;
const char *value;
unsigned long expire;
*r_state = NULL;
fp = open_pairing_state (get_session_id (), 0, &fname);
if (!fp)
return gpg_error (GPG_ERR_NOT_FOUND);
err = nvc_parse (&state, NULL, fp);
if (err)
{
log_info ("failed to parse state file '%s': %s\n",
fname, gpg_strerror (err));
goto leave;
}
/* Check whether the state already expired. */
item = nvc_lookup (state, "Expires:");
if (!item)
{
log_info ("invalid state file '%s': %s\n",
fname, "field 'expire' not found");
goto leave;
}
value = nve_value (item);
if (!value || !(expire = strtoul (value, NULL, 10)))
{
log_info ("invalid state file '%s': %s\n",
fname, "field 'expire' has an invalid value");
goto leave;
}
if (expire <= gnupg_get_time ())
{
es_fclose (fp);
fp = NULL;
if (gnupg_remove (fname))
{
err = gpg_error_from_syserror ();
log_info ("failed to delete state file '%s': %s\n",
fname, gpg_strerror (err));
}
else if (opt.verbose)
log_info ("state file '%s' deleted\n", fname);
err = gpg_error (GPG_ERR_NOT_FOUND);
goto leave;
}
*r_state = state;
state = NULL;
leave:
nvc_release (state);
es_fclose (fp);
return err;
}
/* Send (MSG,MSGLEN) to the output device. */
static void
send_message (const unsigned char *msg, size_t msglen)
{
gpg_error_t err;
if (opt.verbose)
log_info ("session %s: sending %s message\n",
get_session_id_hex (), msgtypestr (msg[7]));
if (DBG_MESSAGE)
log_printhex (msg, msglen, "send msg(%s):", msgtypestr (msg[7]));
/* FIXME: For now only stdout. */
if (opt.armor)
{
gpgrt_b64state_t state;
state = gpgrt_b64enc_start (es_stdout, "");
if (!state)
log_fatal ("error setting up base64 encoder: %s\n",
gpg_strerror (gpg_error_from_syserror ()));
err = gpgrt_b64enc_write (state, msg, msglen);
if (!err)
err = gpgrt_b64enc_finish (state);
if (err)
log_fatal ("error writing base64 to stdout: %s\n", gpg_strerror (err));
}
else
{
if (es_fwrite (msg, msglen, 1, es_stdout) != 1)
log_fatal ("error writing to stdout: %s\n",
gpg_strerror (gpg_error_from_syserror ()));
}
es_fputc ('\n', es_stdout);
}
/* Read a message from stdin and store it at the address (R_MSG,
* R_MSGLEN). This function detects armoring and removes it. On
* error NULL is stored at R_MSG, a diagnostic printed and an error
* code returned. The returned message has a proper message type and
* an appropriate length. The message type is stored at R_MSGTYPE and
* if a state is available it is stored at R_STATE. */
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
static gpg_error_t
read_message (unsigned char **r_msg, size_t *r_msglen, int *r_msgtype,
nvc_t *r_state)
{
gpg_error_t err;
unsigned char msg[128]; /* max msg size is 80 but 107 with base64. */
size_t msglen;
size_t reqlen;
*r_msg = NULL;
*r_state = NULL;
es_setvbuf (es_stdin, NULL, _IONBF, 0);
es_set_binary (es_stdin);
if (es_read (es_stdin, msg, sizeof msg, &msglen))
{
err = gpg_error_from_syserror ();
log_error ("error reading from message: %s\n", gpg_strerror (err));
return err;
}
if (msglen > 4 && !memcmp (msg, "R1BH", 4))
{
/* This is base64 of the first 3 bytes. */
gpgrt_b64state_t state = gpgrt_b64dec_start (NULL);
if (!state)
log_fatal ("error setting up base64 decoder: %s\n",
gpg_strerror (gpg_error_from_syserror ()));
err = gpgrt_b64dec_proc (state, msg, msglen, &msglen);
gpgrt_b64dec_finish (state);
if (err)
{
log_error ("error decoding message: %s\n", gpg_strerror (err));
return err;
}
}
if (msglen < 16 || memcmp (msg, "GPG-pa1", 7))
{
log_error ("error parsing message: %s\n",
msglen? "invalid header":"empty message");
return gpg_error (GPG_ERR_INV_RESPONSE);
}
switch (msg[7])
{
case MSG_TYPE_COMMIT: reqlen = 56; break;
case MSG_TYPE_DHPART1: reqlen = 80; break;
case MSG_TYPE_DHPART2: reqlen = 80; break;
case MSG_TYPE_CONFIRM: reqlen = 48; break;
default:
log_error ("error parsing message: %s\n", "invalid message type");
return gpg_error (GPG_ERR_INV_RESPONSE);
}
if (msglen < reqlen)
{
log_error ("error parsing message: %s\n", "message too short");
return gpg_error (GPG_ERR_INV_RESPONSE);
}
if (DBG_MESSAGE)
log_printhex (msg, msglen, "recv msg(%s):", msgtypestr (msg[7]));
/* Note that we ignore any garbage at the end of a message. */
msglen = reqlen;
set_session_id (msg+8, 8);
if (opt.verbose)
log_info ("session %s: received %s message\n",
get_session_id_hex (), msgtypestr (msg[7]));
/* Read the state. */
err = read_state (r_state);
if (err && gpg_err_code (err) != GPG_ERR_NOT_FOUND)
return err;
*r_msg = xmalloc (msglen);
memcpy (*r_msg, msg, msglen);
*r_msglen = msglen;
*r_msgtype = msg[7];
return err;
}
/* Display the Short Authentication String (SAS). If WAIT is true the
* function waits until the user has entered the SAS as seen at the
* peer.
*
* To construct the SAS we take the 4 most significant octets of HASH,
* interpret them as a 32 bit big endian unsigned integer, divide that
* integer by 10^9 and take the remainder. The remainder is displayed
* as 3 groups of 3 decimal digits delimited by a hyphens. This gives
* a search space of close to 2^30 and is still easy to compare.
*/
static gpg_error_t
display_sas (const unsigned char *hash, size_t hashlen, int wait)
{
gpg_error_t err = 0;
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
unsigned long sas = 0;
char sasbuf[12];
log_assert (hashlen >= 4);
sas |= (unsigned long)hash[20] << 24;
sas |= (unsigned long)hash[21] << 16;
sas |= (unsigned long)hash[22] << 8;
sas |= (unsigned long)hash[23];
sas %= 1000000000ul;
snprintf (sasbuf, sizeof sasbuf, "%09lu", sas);
memmove (sasbuf+8, sasbuf+6, 3);
memmove (sasbuf+4, sasbuf+3, 3);
sasbuf[3] = sasbuf[7] = '-';
sasbuf[11] = 0;
if (wait)
log_info ("Please check the SAS:\n");
else
log_info ("Please note the SAS:\n");
log_info ("\n");
log_info (" %s\n", sasbuf);
log_info ("\n");
if (wait)
{
if (!opt.sas || strcmp (sasbuf, opt.sas))
err = gpg_error (GPG_ERR_NOT_CONFIRMED);
else
log_info ("SAS confirmed\n");
}
if (err)
log_info ("checking SAS failed: %s\n", gpg_strerror (err));
return err;
}
static gpg_error_t
create_dh_keypair (unsigned char *dh_secret, size_t dh_secret_len,
unsigned char *dh_public, size_t dh_public_len)
{
gpg_error_t err;
unsigned char *p;
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
/* We need a temporary buffer for the public key. Check the length
* for the later memcpy. */
if (dh_public_len < 32 || dh_secret_len < 32)
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
return gpg_error (GPG_ERR_BUFFER_TOO_SHORT);
if (gcry_ecc_get_algo_keylen (GCRY_ECC_CURVE25519) > dh_public_len)
return gpg_error (GPG_ERR_BUFFER_TOO_SHORT);
p = gcry_random_bytes (32, GCRY_VERY_STRONG_RANDOM);
if (!p)
return gpg_error_from_syserror ();
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
memcpy (dh_secret, p, 32);
xfree (p);
err = gcry_ecc_mul_point (GCRY_ECC_CURVE25519, dh_public, dh_secret, NULL);
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
if (err)
return err;
if (DBG_CRYPTO)
{
log_printhex (dh_secret, 32, "DH secret:");
log_printhex (dh_public, 32, "DH public:");
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
}
return 0;
}
/* SHA256 the data given as varargs tuples of (const void*, size_t)
* and store the result in RESULT. The end of the list is indicated
* by a NULL element in a tuple. RESULTLEN gives the length of the
* RESULT buffer which must be at least 32. Note that the second item
* of the tuple is the length and it is a size_t. */
static void *
hash_data (void *result, size_t resultsize, ...)
{
va_list arg_ptr;
gpg_error_t err;
gcry_md_hd_t hd;
const void *data;
size_t datalen;
log_assert (resultsize >= 32);
err = gcry_md_open (&hd, GCRY_MD_SHA256, 0);
if (err)
log_fatal ("error creating a Hash handle: %s\n", gpg_strerror (err));
/* log_printhex ("", 0, "Hash-256:"); */
va_start (arg_ptr, resultsize);
while ((data = va_arg (arg_ptr, const void *)))
{
datalen = va_arg (arg_ptr, size_t);
/* log_printhex (data, datalen, " data:"); */
gcry_md_write (hd, data, datalen);
}
va_end (arg_ptr);
memcpy (result, gcry_md_read (hd, 0), 32);
/* log_printhex (result, 32, " result:"); */
gcry_md_close (hd);
return result;
}
/* HMAC-SHA256 the data given as varargs tuples of (const void*,
* size_t) using (KEYLEN,KEY) and store the result in RESULT. The end
* of the list is indicated by a NULL element in a tuple. RESULTLEN
* gives the length of the RESULT buffer which must be at least 32.
* Note that the second item of the tuple is the length and it is a
* size_t. */
static void *
hmac_data (void *result, size_t resultsize,
const unsigned char *key, size_t keylen, ...)
{
va_list arg_ptr;
gpg_error_t err;
gcry_mac_hd_t hd;
const void *data;
size_t datalen;
log_assert (resultsize >= 32);
err = gcry_mac_open (&hd, GCRY_MAC_HMAC_SHA256, 0, NULL);
if (err)
log_fatal ("error creating a MAC handle: %s\n", gpg_strerror (err));
err = gcry_mac_setkey (hd, key, keylen);
if (err)
log_fatal ("error setting the MAC key: %s\n", gpg_strerror (err));
/* log_printhex (key, keylen, "HMAC-key:"); */
va_start (arg_ptr, keylen);
while ((data = va_arg (arg_ptr, const void *)))
{
datalen = va_arg (arg_ptr, size_t);
/* log_printhex (data, datalen, " data:"); */
err = gcry_mac_write (hd, data, datalen);
if (err)
log_fatal ("error writing to the MAC handle: %s\n", gpg_strerror (err));
}
va_end (arg_ptr);
err = gcry_mac_read (hd, result, &resultsize);
if (err || resultsize != 32)
log_fatal ("error reading MAC value: %s\n", gpg_strerror (err));
/* log_printhex (result, resultsize, " result:"); */
gcry_mac_close (hd);
return result;
}
/* Key derivation function:
*
* FIXME(doc)
*/
static void
kdf (unsigned char *result, size_t resultlen,
const unsigned char *master, size_t masterlen,
const unsigned char *sessionid, size_t sessionidlen,
const unsigned char *expire, size_t expirelen,
const char *label)
{
log_assert (masterlen == 32 && sessionidlen == 8 && expirelen == 5);
log_assert (*label);
log_assert (resultlen == 32);
hmac_data (result, resultlen, master, masterlen,
"\x00\x00\x00\x01", (size_t)4, /* Counter=1*/
label, strlen (label) + 1, /* Label, 0x00 */
sessionid, sessionidlen, /* Context */
expire, expirelen, /* Context */
"\x00\x00\x01\x00", (size_t)4, /* L=256 */
NULL);
}
static gpg_error_t
compute_master_secret (unsigned char *master, size_t masterlen,
const unsigned char *sk_a, size_t sk_a_len,
const unsigned char *pk_b, size_t pk_b_len)
{
gpg_error_t err;
log_assert (masterlen == 32);
log_assert (sk_a_len == 32);
log_assert (pk_b_len == 32);
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
err = gcry_ecc_mul_point (GCRY_ECC_CURVE25519, master, sk_a, pk_b);
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
if (err)
log_error ("error computing DH: %s\n", gpg_strerror (err));
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
return err;
}
/* We are the Initiator: Create the commit message. This function
* sends the COMMIT message and writes STATE. */
static gpg_error_t
make_msg_commit (nvc_t state)
{
gpg_error_t err;
uint64_t now, expire;
unsigned char secret[32];
unsigned char public[32];
unsigned char *newmsg;
size_t newmsglen;
unsigned char tmphash[32];
err = create_dh_keypair (secret, sizeof secret, public, sizeof public );
if (err)
log_error ("creating DH keypair failed: %s\n", gpg_strerror (err));
now = gnupg_get_time ();
expire = now + opt.ttl;
newmsglen = 7+1+8+1+2+5+32;
newmsg = xmalloc (newmsglen);
memcpy (newmsg+0, "GPG-pa1", 7);
newmsg[7] = MSG_TYPE_COMMIT;
memcpy (newmsg+8, get_session_id (), 8);
newmsg[16] = REALM_STANDARD;
newmsg[17] = 0;
newmsg[18] = 0;
newmsg[19] = expire >> 32;
newmsg[20] = expire >> 24;
newmsg[21] = expire >> 16;
newmsg[22] = expire >> 8;
newmsg[23] = expire;
gcry_md_hash_buffer (GCRY_MD_SHA256, newmsg+24, public, 32);
/* Create the state file. */
xnvc_set (state, "State:", "Commit-sent");
xnvc_set_printf (state, "Created:", "%llu", (unsigned long long)now);
xnvc_set_printf (state, "Expires:", "%llu", (unsigned long long)expire);
xnvc_set_hex (state, "DH-PKi:", public, 32);
xnvc_set_hex (state, "DH-SKi:", secret, 32);
gcry_md_hash_buffer (GCRY_MD_SHA256, tmphash, newmsg, newmsglen);
xnvc_set_hex (state, "Hash-Commit:", tmphash, 32);
/* Write the state. Note that we need to create it. The state
* updating should in theory be done atomically with send_message.
* However, we can't assure that the message will actually be
* delivered and thus it doesn't matter whether we have an already
* update state when we later fail in send_message. */
write_state (state, 1);
/* Write the message. */
send_message (newmsg, newmsglen);
xfree (newmsg);
return err;
}
/* We are the Responder: Process a commit message in (MSG,MSGLEN)
* which has already been validated to have a correct header and
* message type. Sends the DHPart1 message and writes STATE. */
static gpg_error_t
proc_msg_commit (nvc_t state, const unsigned char *msg, size_t msglen)
{
gpg_error_t err;
uint64_t now, expire;
unsigned char tmphash[32];
unsigned char secret[32];
unsigned char public[32];
unsigned char *newmsg = NULL;
size_t newmsglen;
log_assert (msglen >= 56);
now = gnupg_get_time ();
/* Check that the message has not expired. */
expire = (uint64_t)msg[19] << 32;
expire |= (uint64_t)msg[20] << 24;
expire |= (uint64_t)msg[21] << 16;
expire |= (uint64_t)msg[22] << 8;
expire |= (uint64_t)msg[23];
if (expire < now)
{
log_error ("received %s message is too old\n",
msgtypestr (MSG_TYPE_COMMIT));
err = gpg_error (GPG_ERR_TOO_OLD);
goto leave;
}
/* Create the response. */
err = create_dh_keypair (secret, sizeof secret, public, sizeof public );
if (err)
{
log_error ("creating DH keypair failed: %s\n", gpg_strerror (err));
goto leave;
}
newmsglen = 7+1+8+32+32;
newmsg = xmalloc (newmsglen);
memcpy (newmsg+0, "GPG-pa1", 7);
newmsg[7] = MSG_TYPE_DHPART1;
memcpy (newmsg+8, msg + 8, 8); /* SessionID. */
memcpy (newmsg+16, public, 32); /* PKr */
/* Hash(Hash(Commit) || DHPart1[0..47]) */
gcry_md_hash_buffer (GCRY_MD_SHA256, tmphash, msg, msglen);
hash_data (newmsg+48, 32,
tmphash, sizeof tmphash,
newmsg, (size_t)48,
NULL);
/* Update the state. */
xnvc_set (state, "State:", "DHPart1-sent");
xnvc_set_printf (state, "Created:", "%llu", (unsigned long long)now);
xnvc_set_printf (state, "Expires:", "%llu", (unsigned long long)expire);
xnvc_set_hex (state, "Hash-PKi:", msg+24, 32);
xnvc_set_hex (state, "DH-PKr:", public, 32);
xnvc_set_hex (state, "DH-SKr:", secret, 32);
gcry_md_hash_buffer (GCRY_MD_SHA256, tmphash, newmsg, newmsglen);
xnvc_set_hex (state, "Hash-DHPart1:", tmphash, 32);
/* Write the state. Note that we need to create it. */
write_state (state, 1);
/* Write the message. */
send_message (newmsg, newmsglen);
leave:
xfree (newmsg);
return err;
}
/* We are the Initiator: Process a DHPART1 message in (MSG,MSGLEN)
* which has already been validated to have a correct header and
* message type. Sends the DHPart2 message and writes STATE. */
static gpg_error_t
proc_msg_dhpart1 (nvc_t state, const unsigned char *msg, size_t msglen)
{
gpg_error_t err;
unsigned char hash[32];
unsigned char tmphash[32];
unsigned char pki[32];
unsigned char pkr[32];
unsigned char ski[32];
unsigned char master[32];
uint64_t expire;
unsigned char expirebuf[5];
unsigned char hmacikey[32];
unsigned char symxkey[32];
unsigned char *newmsg = NULL;
size_t newmsglen;
log_assert (msglen >= 80);
/* Check that the message includes the Hash(Commit). */
if (hex2bin (xnvc_get_string (state, "Hash-Commit:"), hash, sizeof hash) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'Hash-Commit' in our state file\n");
goto leave;
}
hash_data (tmphash, 32,
hash, sizeof hash,
msg, (size_t)48,
NULL);
if (memcmp (msg+48, tmphash, 32))
{
err = gpg_error (GPG_ERR_BAD_DATA);
log_error ("manipulation of received %s message detected: %s\n",
msgtypestr (MSG_TYPE_DHPART1), "Bad Hash");
goto leave;
}
/* Check that the received PKr is different from our PKi and copy
* PKr into PKR. */
if (hex2bin (xnvc_get_string (state, "DH-PKi:"), pki, sizeof pki) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'DH-PKi' in our state file\n");
goto leave;
}
if (!memcmp (msg+16, pki, 32))
{
/* This can only happen if the state file leaked to the
* responder. */
err = gpg_error (GPG_ERR_BAD_DATA);
log_error ("received our own public key PKi instead of PKr\n");
goto leave;
}
memcpy (pkr, msg+16, 32);
/* Put the expire value into a buffer. */
expire = string_to_u64 (xnvc_get_string (state, "Expires:"));
if (!expire)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no 'Expire' in our state file\n");
goto leave;
}
expirebuf[0] = expire >> 32;
expirebuf[1] = expire >> 24;
expirebuf[2] = expire >> 16;
expirebuf[3] = expire >> 8;
expirebuf[4] = expire;
/* Get our secret from the state. */
if (hex2bin (xnvc_get_string (state, "DH-SKi:"), ski, sizeof ski) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'DH-SKi' in our state file\n");
goto leave;
}
/* Compute the shared secrets. */
err = compute_master_secret (master, sizeof master,
ski, sizeof ski, pkr, sizeof pkr);
if (err)
{
log_error ("creating DH keypair failed: %s\n", gpg_strerror (err));
goto leave;
}
kdf (hmacikey, sizeof hmacikey,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-HMACi-key");
kdf (symxkey, sizeof symxkey,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-SYMx-key");
/* Create the response. */
newmsglen = 7+1+8+32+32;
newmsg = xmalloc (newmsglen);
memcpy (newmsg+0, "GPG-pa1", 7);
newmsg[7] = MSG_TYPE_DHPART2;
memcpy (newmsg+8, msg + 8, 8); /* SessionID. */
memcpy (newmsg+16, pki, 32); /* PKi */
/* MAC(HMACi-key, Hash(DHPART1) || DHPART2[0..47] || SYMx-key) */
gcry_md_hash_buffer (GCRY_MD_SHA256, tmphash, msg, msglen);
hmac_data (newmsg+48, 32, hmacikey, sizeof hmacikey,
tmphash, sizeof tmphash,
newmsg, (size_t)48,
symxkey, sizeof symxkey,
NULL);
/* Update the state. */
xnvc_set (state, "State:", "DHPart2-sent");
xnvc_set_hex (state, "DH-Master:", master, sizeof master);
gcry_md_hash_buffer (GCRY_MD_SHA256, tmphash, newmsg, newmsglen);
xnvc_set_hex (state, "Hash-DHPart2:", tmphash, 32);
/* Write the state. */
write_state (state, 0);
/* Write the message. */
send_message (newmsg, newmsglen);
leave:
xfree (newmsg);
return err;
}
/* We are the Responder: Process a DHPART2 message in (MSG,MSGLEN)
* which has already been validated to have a correct header and
* message type. Sends the CONFIRM message and writes STATE. */
static gpg_error_t
proc_msg_dhpart2 (nvc_t state, const unsigned char *msg, size_t msglen)
{
gpg_error_t err;
unsigned char hash[32];
unsigned char tmphash[32];
uint64_t expire;
unsigned char expirebuf[5];
unsigned char pki[32];
unsigned char pkr[32];
unsigned char skr[32];
unsigned char master[32];
unsigned char hmacikey[32];
unsigned char hmacrkey[32];
unsigned char symxkey[32];
unsigned char sas[32];
unsigned char *newmsg = NULL;
size_t newmsglen;
log_assert (msglen >= 80);
/* Check that the PKi in the message matches the Hash(Pki) received
* with the Commit message. */
memcpy (pki, msg + 16, 32);
gcry_md_hash_buffer (GCRY_MD_SHA256, hash, pki, 32);
if (hex2bin (xnvc_get_string (state, "Hash-PKi:"),
tmphash, sizeof tmphash) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'Hash-PKi' in our state file\n");
goto leave;
}
if (memcmp (hash, tmphash, 32))
{
err = gpg_error (GPG_ERR_BAD_DATA);
log_error ("Initiator sent a different key in %s than announced in %s\n",
msgtypestr (MSG_TYPE_DHPART2),
msgtypestr (MSG_TYPE_COMMIT));
goto leave;
}
/* Check that the received PKi is different from our PKr. */
if (hex2bin (xnvc_get_string (state, "DH-PKr:"), pkr, sizeof pkr) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'DH-PKr' in our state file\n");
goto leave;
}
if (!memcmp (pkr, pki, 32))
{
err = gpg_error (GPG_ERR_BAD_DATA);
log_error ("Initiator sent our own PKr back\n");
goto leave;
}
/* Put the expire value into a buffer. */
expire = string_to_u64 (xnvc_get_string (state, "Expires:"));
if (!expire)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no 'Expire' in our state file\n");
goto leave;
}
expirebuf[0] = expire >> 32;
expirebuf[1] = expire >> 24;
expirebuf[2] = expire >> 16;
expirebuf[3] = expire >> 8;
expirebuf[4] = expire;
/* Get our secret from the state. */
if (hex2bin (xnvc_get_string (state, "DH-SKr:"), skr, sizeof skr) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'DH-SKr' in our state file\n");
goto leave;
}
/* Compute the shared secrets. */
err = compute_master_secret (master, sizeof master,
skr, sizeof skr, pki, sizeof pki);
if (err)
{
log_error ("creating DH keypair failed: %s\n", gpg_strerror (err));
goto leave;
}
kdf (hmacikey, sizeof hmacikey,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-HMACi-key");
kdf (hmacrkey, sizeof hmacrkey,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-HMACr-key");
kdf (symxkey, sizeof symxkey,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-SYMx-key");
kdf (sas, sizeof sas,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-SAS");
/* Check the MAC from the message which is
* MAC(HMACi-key, Hash(DHPART1) || DHPART2[0..47] || SYMx-key).
* For that we need to fetch the stored hash from the state. */
if (hex2bin (xnvc_get_string (state, "Hash-DHPart1:"),
tmphash, sizeof tmphash) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'Hash-DHPart1' in our state file\n");
goto leave;
}
hmac_data (hash, 32, hmacikey, sizeof hmacikey,
tmphash, sizeof tmphash,
msg, 48,
symxkey, sizeof symxkey,
NULL);
if (memcmp (msg+48, hash, 32))
{
err = gpg_error (GPG_ERR_BAD_DATA);
log_error ("manipulation of received %s message detected: %s\n",
msgtypestr (MSG_TYPE_DHPART2), "Bad MAC");
goto leave;
}
/* Create the response. */
newmsglen = 7+1+8+32;
newmsg = xmalloc (newmsglen);
memcpy (newmsg+0, "GPG-pa1", 7);
newmsg[7] = MSG_TYPE_CONFIRM;
memcpy (newmsg+8, msg + 8, 8); /* SessionID. */
/* MAC(HMACr-key, Hash(DHPART2) || CONFIRM[0..15] || SYMx-key) */
gcry_md_hash_buffer (GCRY_MD_SHA256, tmphash, msg, msglen);
hmac_data (newmsg+16, 32, hmacrkey, sizeof hmacrkey,
tmphash, sizeof tmphash,
newmsg, (size_t)16,
symxkey, sizeof symxkey,
NULL);
/* Update the state. */
xnvc_set (state, "State:", "Confirm-sent");
xnvc_set_hex (state, "DH-Master:", master, sizeof master);
/* Write the state. */
write_state (state, 0);
/* Write the message. */
send_message (newmsg, newmsglen);
display_sas (sas, sizeof sas, 0);
leave:
xfree (newmsg);
return err;
}
/* We are the Initiator: Process a CONFIRM message in (MSG,MSGLEN)
* which has already been validated to have a correct header and
* message type. Does not send anything back. */
static gpg_error_t
proc_msg_confirm (nvc_t state, const unsigned char *msg, size_t msglen)
{
gpg_error_t err;
unsigned char hash[32];
unsigned char tmphash[32];
unsigned char master[32];
uint64_t expire;
unsigned char expirebuf[5];
unsigned char hmacrkey[32];
unsigned char symxkey[32];
unsigned char sas[32];
log_assert (msglen >= 48);
/* Put the expire value into a buffer. */
expire = string_to_u64 (xnvc_get_string (state, "Expires:"));
if (!expire)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no 'Expire' in our state file\n");
goto leave;
}
expirebuf[0] = expire >> 32;
expirebuf[1] = expire >> 24;
expirebuf[2] = expire >> 16;
expirebuf[3] = expire >> 8;
expirebuf[4] = expire;
/* Get the master secret. */
if (hex2bin (xnvc_get_string (state, "DH-Master:"),master,sizeof master) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'DH-Master' in our state file\n");
goto leave;
}
kdf (hmacrkey, sizeof hmacrkey,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-HMACr-key");
kdf (symxkey, sizeof symxkey,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-SYMx-key");
kdf (sas, sizeof sas,
master, sizeof master, msg+8, 8, expirebuf, sizeof expirebuf,
"GPG-pa1-SAS");
/* Check the MAC from the message which is */
/* MAC(HMACr-key, Hash(DHPART2) || CONFIRM[0..15] || SYMx-key). */
if (hex2bin (xnvc_get_string (state, "Hash-DHPart2:"),
tmphash, sizeof tmphash) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("no or garbled 'Hash-DHPart2' in our state file\n");
goto leave;
}
hmac_data (hash, 32, hmacrkey, sizeof hmacrkey,
tmphash, sizeof tmphash,
msg, (size_t)16,
symxkey, sizeof symxkey,
NULL);
if (!memcmp (msg+48, hash, 32))
{
err = gpg_error (GPG_ERR_BAD_DATA);
log_error ("manipulation of received %s message detected: %s\n",
msgtypestr (MSG_TYPE_CONFIRM), "Bad MAC");
goto leave;
}
err = display_sas (sas, sizeof sas, 1);
if (err)
goto leave;
/* Update the state. */
xnvc_set (state, "State:", "Confirmed");
/* Write the state. */
write_state (state, 0);
leave:
return err;
}
/* Expire old state files. This loops over all state files and remove
* those which are expired. */
static void
expire_old_states (void)
{
gpg_error_t err = 0;
const char *dirname;
gnupg_dir_t dir = NULL;
gnupg_dirent_t dir_entry;
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
char *fname = NULL;
estream_t fp = NULL;
nvc_t nvc = NULL;
nve_t item;
const char *value;
unsigned long expire;
unsigned long now = gnupg_get_time ();
dirname = get_pairing_statedir ();
dir = gnupg_opendir (dirname);
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
if (!dir)
{
err = gpg_error_from_syserror ();
goto leave;
}
while ((dir_entry = gnupg_readdir (dir)))
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
{
if (strlen (dir_entry->d_name) != 16+4
|| strcmp (dir_entry->d_name + 16, ".pa1"))
continue;
xfree (fname);
fname = make_filename (dirname, dir_entry->d_name, NULL);
es_fclose (fp);
fp = es_fopen (fname, "rb");
if (!fp)
{
err = gpg_error_from_syserror ();
if (gpg_err_code (err) != GPG_ERR_ENOENT)
log_info ("failed to open state file '%s': %s\n",
fname, gpg_strerror (err));
continue;
}
nvc_release (nvc);
/* NB.: The following is similar to code in read_state. */
err = nvc_parse (&nvc, NULL, fp);
if (err)
{
log_info ("failed to parse state file '%s': %s\n",
fname, gpg_strerror (err));
continue; /* Skip */
}
item = nvc_lookup (nvc, "Expires:");
if (!item)
{
log_info ("invalid state file '%s': %s\n",
fname, "field 'expire' not found");
continue; /* Skip */
}
value = nve_value (item);
if (!value || !(expire = strtoul (value, NULL, 10)))
{
log_info ("invalid state file '%s': %s\n",
fname, "field 'expire' has an invalid value");
continue; /* Skip */
}
if (expire <= now)
{
es_fclose (fp);
fp = NULL;
if (gnupg_remove (fname))
{
err = gpg_error_from_syserror ();
log_info ("failed to delete state file '%s': %s\n",
fname, gpg_strerror (err));
}
else if (opt.verbose)
log_info ("state file '%s' deleted\n", fname);
}
}
leave:
if (err)
log_error ("expiring old states in '%s' failed: %s\n",
dirname, gpg_strerror (err));
gnupg_closedir (dir);
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
es_fclose (fp);
xfree (fname);
}
/* Initiate a pairing. The output needs to be conveyed to the
* peer */
static gpg_error_t
command_initiate (void)
{
gpg_error_t err;
nvc_t state;
state = xnvc_new ();
xnvc_set (state, "Version:", "GPG-pa1");
xnvc_set_hex (state, "Session:", get_session_id (), 8);
xnvc_set (state, "Role:", "Initiator");
err = make_msg_commit (state);
nvc_release (state);
return err;
}
/* Helper for command_respond(). */
static gpg_error_t
expect_state (int msgtype, const char *statestr, const char *expected)
{
if (strcmp (statestr, expected))
{
log_error ("received %s message in %s state (should be %s)\n",
msgtypestr (msgtype), statestr, expected);
return gpg_error (GPG_ERR_INV_RESPONSE);
}
return 0;
}
/* Respond to a pairing initiation. This is used by the peer and later
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
* by the original responder. Depending on the state the output needs
* to be conveyed to the peer. */
static gpg_error_t
command_respond (void)
{
gpg_error_t err;
unsigned char *msg;
size_t msglen = 0; /* In case that read_message returns an error. */
int msgtype = 0; /* ditto. */
tools: Add experimental code for a pairing protocol * configure.ac (GNUPG_CACHE_DIR): New const. * tools/Makefile.am (libexec_PROGRAMS): Add gpg-pair-tool. (gpg_pair_tool_SOURCES, gpg_pair_tool_CFLAGS) (gpg_pair_tool_LDADD): New. * tools/gpg-pair-tool.c: New. -- This is a first try on a protocol to pair two devices so that they can agree on a shared secret to exchange secret keys. The idea is that if you want to sync your secret keys to another machine (e.g. from desktop to mobile) you have physical access to both devices and thus a pairing protocol allows to authenitcate the connection using a short string. See the source for a protocol description. How to test: $ gpg-pair-tool -va --homedir . --initiate >msg.commit $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.commit >msg.dhpart1 $ gpg-pair-tool -va --homedir . --respond \ <msg.dhpart1 >msg.dhpart2 $ gpg-pair-tool -va --homedir 2ndhome --respond \ <msg.dhpart2 >msg.confirm Now set the SAS as printed by the responder into SAS and run $ gpg-pair-tool -va --homedir . --respond --sas $SAS <msg.confirm Storing the secret on disk is obviously not the right thing to do. With the new PUT_SECRET and GET_SECRET commands of gpg-agent we can change this to store it all in gpg-agent instead. This will make it also easier for gpg to access the secret and we won't need an option to return it from gpg-pair-tool. Thus gpg-pair-tool can be dedicated to run the protocol and maybe to popup info dialogs. Adding a second expiration time for running the protocol in addition to the expiration of the secret is probably a better idea than just that simple catch-all TTL. Signed-off-by: Werner Koch <wk@gnupg.org>
2018-07-05 09:40:35 +02:00
nvc_t state;
const char *rolestr;
const char *statestr;
err = read_message (&msg, &msglen, &msgtype, &state);
if (err && gpg_err_code (err) != GPG_ERR_NOT_FOUND)
goto leave;
rolestr = xnvc_get_string (state, "Role:");
statestr = xnvc_get_string (state, "State:");
if (DBG_MESSAGE)
{
if (!state)
log_debug ("no state available\n");
else
log_debug ("we are %s, our current state is %s\n", rolestr, statestr);
log_debug ("got message of type %s (%d)\n",
msgtypestr (msgtype), msgtype);
}
if (!state)
{
if (msgtype == MSG_TYPE_COMMIT)
{
state = xnvc_new ();
xnvc_set (state, "Version:", "GPG-pa1");
xnvc_set_hex (state, "Session:", get_session_id (), 8);
xnvc_set (state, "Role:", "Responder");
err = proc_msg_commit (state, msg, msglen);
}
else
{
log_error ("%s message expected but got %s\n",
msgtypestr (MSG_TYPE_COMMIT), msgtypestr (msgtype));
if (msgtype == MSG_TYPE_DHPART1)
log_info ("the pairing probably took too long and timed out\n");
err = gpg_error (GPG_ERR_INV_RESPONSE);
goto leave;
}
}
else if (!strcmp (rolestr, "Initiator"))
{
if (msgtype == MSG_TYPE_DHPART1)
{
if (!(err = expect_state (msgtype, statestr, "Commit-sent")))
err = proc_msg_dhpart1 (state, msg, msglen);
}
else if (msgtype == MSG_TYPE_CONFIRM)
{
if (!(err = expect_state (msgtype, statestr, "DHPart2-sent")))
err = proc_msg_confirm (state, msg, msglen);
}
else
{
log_error ("%s message not expected by Initiator\n",
msgtypestr (msgtype));
err = gpg_error (GPG_ERR_INV_RESPONSE);
goto leave;
}
}
else if (!strcmp (rolestr, "Responder"))
{
if (msgtype == MSG_TYPE_DHPART2)
{
if (!(err = expect_state (msgtype, statestr, "DHPart1-sent")))
err = proc_msg_dhpart2 (state, msg, msglen);
}
else
{
log_error ("%s message not expected by Responder\n",
msgtypestr (msgtype));
err = gpg_error (GPG_ERR_INV_RESPONSE);
goto leave;
}
}
else
log_fatal ("invalid role '%s' in state file\n", rolestr);
leave:
xfree (msg);
nvc_release (state);
return err;
}
/* Return the keys for SESSIONIDSTR or the last one if it is NULL.
* Two keys are returned: The first is the one for sending encrypted
* data and the second one for decrypting received data. The keys are
* always returned hex encoded and both are terminated by a LF. */
static gpg_error_t
command_get (const char *sessionidstr)
{
gpg_error_t err;
unsigned char sessid[8];
nvc_t state;
if (!sessionidstr)
{
log_error ("calling without session-id is not yet implemented\n");
err = gpg_error (GPG_ERR_NOT_IMPLEMENTED);
goto leave;
}
if (hex2bin (sessionidstr, sessid, sizeof sessid) < 0)
{
err = gpg_error (GPG_ERR_INV_VALUE);
log_error ("invalid session id given\n");
goto leave;
}
set_session_id (sessid, sizeof sessid);
err = read_state (&state);
if (err)
{
log_error ("reading state of session %s failed: %s\n",
sessionidstr, gpg_strerror (err));
goto leave;
}
leave:
return err;
}
/* Cleanup command. */
static gpg_error_t
command_cleanup (void)
{
expire_old_states ();
return 0;
}