gnupg/g10/getkey.c

2325 lines
65 KiB
C
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/* getkey.c - Get a key from the database
* Copyright (C) 1998, 1999, 2000 Free Software Foundation, Inc.
*
* This file is part of GnuPG.
*
* GnuPG is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* GnuPG 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 General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA
*/
#include <config.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <ctype.h>
#include "util.h"
#include "packet.h"
#include <gcrypt.h>
#include "iobuf.h"
#include "keydb.h"
#include "options.h"
#include "main.h"
#include "trustdb.h"
#include "i18n.h"
#if 0
#define MAX_UNK_CACHE_ENTRIES 1000 /* we use a linked list - so I guess
* this is a reasonable limit */
#define MAX_PK_CACHE_ENTRIES 50
#endif
#define MAX_UID_CACHE_ENTRIES 50
/* A map of the all characters valid used for word_match()
* Valid characters are in in this table converted to uppercase.
* because the upper 128 bytes have special meaning, we assume
* that they are all valid.
* Note: We must use numerical values here in case that this program
* will be converted to those little blue HAL9000s with their strange
* EBCDIC character set (user ids are UTF-8).
* wk 2000-04-13: Hmmm, does this really make sense, given the fact that
* we can run gpg now on a S/390 running GNU/Linux, where the code
* translation is done by the device drivers?
*/
static const byte word_match_chars[256] = {
/* 00 */ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 08 */ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 10 */ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 18 */ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 20 */ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 28 */ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 30 */ 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37,
/* 38 */ 0x38, 0x39, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 40 */ 0x00, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47,
/* 48 */ 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f,
/* 50 */ 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57,
/* 58 */ 0x58, 0x59, 0x5a, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 60 */ 0x00, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47,
/* 68 */ 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f,
/* 70 */ 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57,
/* 78 */ 0x58, 0x59, 0x5a, 0x00, 0x00, 0x00, 0x00, 0x00,
/* 80 */ 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87,
/* 88 */ 0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f,
/* 90 */ 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97,
/* 98 */ 0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f,
/* a0 */ 0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7,
/* a8 */ 0xa8, 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, 0xaf,
/* b0 */ 0xb0, 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7,
/* b8 */ 0xb8, 0xb9, 0xba, 0xbb, 0xbc, 0xbd, 0xbe, 0xbf,
/* c0 */ 0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7,
/* c8 */ 0xc8, 0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, 0xcf,
/* d0 */ 0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7,
/* d8 */ 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf,
/* e0 */ 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7,
/* e8 */ 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef,
/* f0 */ 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
/* f8 */ 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff
};
typedef struct {
int mode;
u32 keyid[2];
byte fprint[20];
char *namebuf;
const char *name;
} getkey_item_t;
struct getkey_ctx_s {
/* make an array or a linked list from dome fields */
int primary;
KBNODE keyblock;
KBPOS kbpos;
KBNODE found_key; /* pointer into some keyblock */
int last_rc;
int req_usage;
int req_algo;
ulong count;
int not_allocated;
int nitems;
getkey_item_t items[1];
};
#if 0
static struct {
int any;
int okay_count;
int nokey_count;
int error_count;
} lkup_stats[21];
#endif
typedef struct keyid_list {
struct keyid_list *next;
u32 keyid[2];
} *keyid_list_t;
#if MAX_UNK_CACHE_ENTRIES
static keyid_list_t unknown_keyids;
static int unk_cache_entries; /* number of entries in unknown keys cache */
static int unk_cache_disabled;
#endif
#if MAX_PK_CACHE_ENTRIES
typedef struct pk_cache_entry {
struct pk_cache_entry *next;
u32 keyid[2];
PKT_public_key *pk;
} *pk_cache_entry_t;
static pk_cache_entry_t pk_cache;
static int pk_cache_entries; /* number of entries in pk cache */
static int pk_cache_disabled;
#endif
#if MAX_UID_CACHE_ENTRIES < 5
#error we really need the userid cache
#endif
typedef struct user_id_db {
struct user_id_db *next;
keyid_list_t keyids;
int len;
char name[1];
} *user_id_db_t;
static user_id_db_t user_id_db;
static int uid_cache_entries; /* number of entries in uid cache */
static char* prepare_word_match( const byte *name );
static int lookup( GETKEY_CTX ctx, KBNODE *ret_kb, int secmode );
#if 0
static void
print_stats()
{
int i;
for(i=0; i < DIM(lkup_stats); i++ ) {
if( lkup_stats[i].any )
fprintf(stderr,
"lookup stats: mode=%-2d ok=%-6d nokey=%-6d err=%-6d\n",
i,
lkup_stats[i].okay_count,
lkup_stats[i].nokey_count,
lkup_stats[i].error_count );
}
}
#endif
static void
cache_public_key( PKT_public_key *pk )
{
#if MAX_PK_CACHE_ENTRIES
pk_cache_entry_t ce;
u32 keyid[2];
if( pk_cache_disabled )
return;
if( is_ELGAMAL(pk->pubkey_algo)
|| pk->pubkey_algo == GCRY_PK_DSA
|| is_RSA(pk->pubkey_algo) ) {
keyid_from_pk( pk, keyid );
}
else
return; /* don't know how to get the keyid */
for( ce = pk_cache; ce; ce = ce->next )
if( ce->keyid[0] == keyid[0] && ce->keyid[1] == keyid[1] ) {
if( DBG_CACHE )
log_debug("cache_public_key: already in cache\n");
return;
}
if( pk_cache_entries >= MAX_PK_CACHE_ENTRIES ) {
/* fixme: use another algorithm to free some cache slots */
pk_cache_disabled=1;
if( opt.verbose > 1 )
log_info(_("too many entries in pk cache - disabled\n"));
return;
}
pk_cache_entries++;
ce = gcry_xmalloc( sizeof *ce );
ce->next = pk_cache;
pk_cache = ce;
ce->pk = copy_public_key( NULL, pk );
ce->keyid[0] = keyid[0];
ce->keyid[1] = keyid[1];
#endif
}
/*
* Return the user ID from the given keyblock.
* We use the primary uid flag which has been set by the merge_selfsigs
* function. The returned value is only valid as long as then given
* keyblock is not changed
*/
static const char *
get_primary_uid ( KBNODE keyblock, size_t *uidlen )
{
KBNODE k;
for (k=keyblock; k; k=k->next ) {
if ( k->pkt->pkttype == PKT_USER_ID
&& k->pkt->pkt.user_id->is_primary ) {
*uidlen = k->pkt->pkt.user_id->len;
return k->pkt->pkt.user_id->name;
}
}
*uidlen = 12;
return "[No user ID]";
}
static void
release_keyid_list ( keyid_list_t k )
{
while ( k ) {
keyid_list_t k2 = k->next;
gcry_free (k);
k = k2;
}
}
/****************
* Store the association of keyid and userid
* Feed only public keys to this function.
*/
void
cache_user_id( KBNODE keyblock )
{
user_id_db_t r;
const char *uid;
size_t uidlen;
keyid_list_t keyids = NULL;
KBNODE k;
for (k=keyblock; k; k = k->next ) {
if ( k->pkt->pkttype == PKT_PUBLIC_KEY
|| k->pkt->pkttype == PKT_PUBLIC_SUBKEY ) {
keyid_list_t a = gcry_xcalloc ( 1, sizeof *a );
/* Hmmm: For a long list of keyids it might be an advantage
* to append the keys */
keyid_from_pk( k->pkt->pkt.public_key, a->keyid );
/* first check for duplicates */
for(r=user_id_db; r; r = r->next ) {
keyid_list_t b = r->keyids;
for ( b = r->keyids; b; b = b->next ) {
if( b->keyid[0] == a->keyid[0]
&& b->keyid[1] == a->keyid[1] ) {
if( DBG_CACHE )
log_debug("cache_user_id: already in cache\n");
release_keyid_list ( keyids );
gcry_free ( a );
return;
}
}
}
/* now put it into the cache */
a->next = keyids;
keyids = a;
}
}
if ( !keyids )
BUG (); /* No key no fun */
uid = get_primary_uid ( keyblock, &uidlen );
if( uid_cache_entries >= MAX_UID_CACHE_ENTRIES ) {
/* fixme: use another algorithm to free some cache slots */
r = user_id_db;
user_id_db = r->next;
release_keyid_list ( r->keyids );
gcry_free(r);
uid_cache_entries--;
}
r = gcry_xmalloc( sizeof *r + uidlen-1 );
r->keyids = keyids;
r->len = uidlen;
memcpy(r->name, uid, r->len);
r->next = user_id_db;
user_id_db = r;
uid_cache_entries++;
}
void
getkey_disable_caches()
{
#if MAX_UNK_CACHE_ENTRIES
{
keyid_list_t kl, kl2;
for( kl = unknown_keyids; kl; kl = kl2 ) {
kl2 = kl->next;
gcry_free(kl);
}
unknown_keyids = NULL;
unk_cache_disabled = 1;
}
#endif
#if MAX_PK_CACHE_ENTRIES
{
pk_cache_entry_t ce, ce2;
for( ce = pk_cache; ce; ce = ce2 ) {
ce2 = ce->next;
free_public_key( ce->pk );
gcry_free( ce );
}
pk_cache_disabled=1;
pk_cache_entries = 0;
pk_cache = NULL;
}
#endif
/* fixme: disable user id cache ? */
}
static void
pk_from_block ( GETKEY_CTX ctx,
PKT_public_key *pk, KBNODE keyblock, const char *namehash )
{
KBNODE a = ctx->found_key ? ctx->found_key : keyblock;
assert ( a->pkt->pkttype == PKT_PUBLIC_KEY
|| a->pkt->pkttype == PKT_PUBLIC_SUBKEY );
copy_public_key_new_namehash( pk, a->pkt->pkt.public_key, namehash);
}
static void
sk_from_block ( GETKEY_CTX ctx,
PKT_secret_key *sk, KBNODE keyblock )
{
KBNODE a = ctx->found_key ? ctx->found_key : keyblock;
assert ( a->pkt->pkttype == PKT_SECRET_KEY
|| a->pkt->pkttype == PKT_SECRET_SUBKEY );
copy_secret_key( sk, a->pkt->pkt.secret_key);
}
/****************
* Get a public key and store it into the allocated pk
* can be called with PK set to NULL to just read it into some
* internal structures.
*/
int
get_pubkey( PKT_public_key *pk, u32 *keyid )
{
int internal = 0;
int rc = 0;
#if MAX_UNK_CACHE_ENTRIES
{ /* let's see whether we checked the keyid already */
keyid_list_t kl;
for( kl = unknown_keyids; kl; kl = kl->next )
if( kl->keyid[0] == keyid[0] && kl->keyid[1] == keyid[1] )
return GPGERR_NO_PUBKEY; /* already checked and not found */
}
#endif
#if MAX_PK_CACHE_ENTRIES
{ /* Try to get it from the cache */
pk_cache_entry_t ce;
for( ce = pk_cache; ce; ce = ce->next ) {
if( ce->keyid[0] == keyid[0] && ce->keyid[1] == keyid[1] ) {
if( pk )
copy_public_key( pk, ce->pk );
return 0;
}
}
}
#endif
/* more init stuff */
if( !pk ) {
pk = gcry_xcalloc( 1, sizeof *pk );
internal++;
}
/* do a lookup */
{ struct getkey_ctx_s ctx;
KBNODE kb = NULL;
memset( &ctx, 0, sizeof ctx );
ctx.not_allocated = 1;
ctx.nitems = 1;
ctx.items[0].mode = 11;
ctx.items[0].keyid[0] = keyid[0];
ctx.items[0].keyid[1] = keyid[1];
ctx.req_algo = pk->req_algo;
ctx.req_usage = pk->req_usage;
rc = lookup( &ctx, &kb, 0 );
if ( !rc ) {
pk_from_block ( &ctx, pk, kb, NULL );
}
get_pubkey_end( &ctx );
release_kbnode ( kb );
}
if( !rc )
goto leave;
#if MAX_UNK_CACHE_ENTRIES
/* not found: store it for future reference */
if( unk_cache_disabled )
;
else if( ++unk_cache_entries > MAX_UNK_CACHE_ENTRIES ) {
unk_cache_disabled = 1;
if( opt.verbose > 1 )
log_info(_("too many entries in unk cache - disabled\n"));
}
else {
keyid_list_t kl;
kl = gcry_xmalloc( sizeof *kl );
kl->keyid[0] = keyid[0];
kl->keyid[1] = keyid[1];
kl->next = unknown_keyids;
unknown_keyids = kl;
}
#endif
rc = GPGERR_NO_PUBKEY;
leave:
if( !rc )
cache_public_key( pk );
if( internal )
free_public_key(pk);
return rc;
}
KBNODE
get_pubkeyblock( u32 *keyid )
{
struct getkey_ctx_s ctx;
int rc = 0;
KBNODE keyblock = NULL;
memset( &ctx, 0, sizeof ctx );
ctx.not_allocated = 1;
ctx.nitems = 1;
ctx.items[0].mode = 11;
ctx.items[0].keyid[0] = keyid[0];
ctx.items[0].keyid[1] = keyid[1];
rc = lookup( &ctx, &keyblock, 0 );
get_pubkey_end( &ctx );
return rc ? NULL : keyblock;
}
/****************
* Get a secret key and store it into sk
*/
int
get_seckey( PKT_secret_key *sk, u32 *keyid )
{
int rc;
struct getkey_ctx_s ctx;
KBNODE kb = NULL;
memset( &ctx, 0, sizeof ctx );
ctx.not_allocated = 1;
ctx.nitems = 1;
ctx.items[0].mode = 11;
ctx.items[0].keyid[0] = keyid[0];
ctx.items[0].keyid[1] = keyid[1];
ctx.req_algo = sk->req_algo;
ctx.req_usage = sk->req_usage;
rc = lookup( &ctx, &kb, 1 );
if ( !rc ) {
sk_from_block ( &ctx, sk, kb );
}
get_seckey_end( &ctx );
release_kbnode ( kb );
if( !rc ) {
/* check the secret key (this may prompt for a passprase to
* unlock the secret key
*/
rc = check_secret_key( sk, 0 );
}
return rc;
}
/****************
* Check whether the secret key is available
* Returns: 0 := key is available
* GPGERR_NO_SECKEY := not availabe
*/
int
seckey_available( u32 *keyid )
{
int rc;
struct getkey_ctx_s ctx;
KBNODE kb = NULL;
memset( &ctx, 0, sizeof ctx );
ctx.not_allocated = 1;
ctx.nitems = 1;
ctx.items[0].mode = 11;
ctx.items[0].keyid[0] = keyid[0];
ctx.items[0].keyid[1] = keyid[1];
rc = lookup( &ctx, &kb, 1 );
get_seckey_end( &ctx );
release_kbnode ( kb );
return rc;
}
static int
hextobyte( const byte *s )
{
int c;
if( *s >= '0' && *s <= '9' )
c = 16 * (*s - '0');
else if( *s >= 'A' && *s <= 'F' )
c = 16 * (10 + *s - 'A');
else if( *s >= 'a' && *s <= 'f' )
c = 16 * (10 + *s - 'a');
else
return -1;
s++;
if( *s >= '0' && *s <= '9' )
c += *s - '0';
else if( *s >= 'A' && *s <= 'F' )
c += 10 + *s - 'A';
else if( *s >= 'a' && *s <= 'f' )
c += 10 + *s - 'a';
else
return -1;
return c;
}
/****************
* Return the type of the user id:
*
* 0 = Invalid user ID
* 1 = exact match
* 2 = match a substring
* 3 = match an email address
* 4 = match a substring of an email address
* 5 = match an email address, but compare from end
* 6 = word match mode
* 10 = it is a short KEYID (don't care about keyid[0])
* 11 = it is a long KEYID
* 12 = it is a trustdb index (keyid is looked up)
* 16 = it is a 16 byte fingerprint
* 20 = it is a 20 byte fingerprint
*
* if fprint is not NULL, it should be an array of at least 20 bytes.
*
* Rules used:
* - If the username starts with 8,9,16 or 17 hex-digits (the first one
* must be in the range 0..9), this is considered a keyid; depending
* on the length a short or complete one.
* - If the username starts with 32,33,40 or 41 hex-digits (the first one
* must be in the range 0..9), this is considered a fingerprint.
* - If the username starts with a left angle, we assume it is a complete
* email address and look only at this part.
* - If the username starts with a '.', we assume it is the ending
* part of an email address
* - If the username starts with an '@', we assume it is a part of an
* email address
* - If the userid start with an '=' an exact compare is done.
* - If the userid starts with a '*' a case insensitive substring search is
* done (This is the default).
* - If the userid starts with a '+' we will compare individual words
* and a match requires that all the words are in the userid.
* Words are delimited by white space or "()<>[]{}.@-+_,;/&!"
* (note that you can't search for these characters). Compare
* is not case sensitive.
*/
int
classify_user_id( const char *name, u32 *keyid, byte *fprint,
const char **retstr, size_t *retlen )
{
const char * s;
int mode = 0;
int hexprefix = 0;
int hexlength;
/* skip leading spaces. FIXME: what is with leading spaces? */
for(s = name; *s && isspace(*s); s++ )
;
switch (*s) {
case 0: /* empty string is an error */
return 0;
case '.': /* an email address, compare from end */
mode = 5;
s++;
break;
case '<': /* an email address */
mode = 3;
break;
case '@': /* part of an email address */
mode = 4;
s++;
break;
case '=': /* exact compare */
mode = 1;
s++;
break;
case '*': /* case insensitive substring search */
mode = 2;
s++;
break;
case '+': /* compare individual words */
mode = 6;
s++;
break;
case '#': /* local user id */
mode = 12;
s++;
if (keyid) {
if (keyid_from_lid(strtoul(s, NULL, 10), keyid))
keyid[0] = keyid[1] = 0;
}
break;
default:
if (s[0] == '0' && s[1] == 'x') {
hexprefix = 1;
s += 2;
}
hexlength = strspn(s, "0123456789abcdefABCDEF");
/* check if a hexadecimal number is terminated by EOS or blank */
if (hexlength && s[hexlength] && !isspace(s[hexlength])) {
if (hexprefix) /* a "0x" prefix without correct */
return 0; /* termination is an error */
else /* The first chars looked like */
hexlength = 0; /* a hex number, but really were not. */
}
if (hexlength == 8
|| (!hexprefix && hexlength == 9 && *s == '0')){
/* short keyid */
if (hexlength == 9)
s++;
if (keyid) {
keyid[0] = 0;
keyid[1] = strtoul( s, NULL, 16 );
}
mode = 10;
}
else if (hexlength == 16
|| (!hexprefix && hexlength == 17 && *s == '0')) {
/* complete keyid */
char buf[9];
if (hexlength == 17)
s++;
mem2str(buf, s, 9 );
keyid[0] = strtoul( buf, NULL, 16 );
keyid[1] = strtoul( s+8, NULL, 16 );
mode = 11;
}
else if (hexlength == 32 || (!hexprefix && hexlength == 33
&& *s == '0')) {
/* md5 fingerprint */
int i;
if (hexlength == 33)
s++;
if (fprint) {
memset(fprint+16, 4, 0);
for (i=0; i < 16; i++, s+=2) {
int c = hextobyte(s);
if (c == -1)
return 0;
fprint[i] = c;
}
}
mode = 16;
}
else if (hexlength == 40 || (!hexprefix && hexlength == 41
&& *s == '0')) {
/* sha1/rmd160 fingerprint */
int i;
if (hexlength == 41)
s++;
if (fprint) {
for (i=0; i < 20; i++, s+=2) {
int c = hextobyte(s);
if (c == -1)
return 0;
fprint[i] = c;
}
}
mode = 20;
}
else {
if (hexprefix) /* This was a hex number with a prefix */
return 0; /* and a wrong length */
mode = 2; /* Default is case insensitive substring search */
}
}
if( retstr )
*retstr = s;
if( retlen )
*retlen = strlen(s);
return mode;
}
/****************
* Try to get the pubkey by the userid. This function looks for the
* first pubkey certificate which has the given name in a user_id.
* if pk/sk has the pubkey algo set, the function will only return
* a pubkey with that algo.
* The caller must provide provide storage for either the pk or the sk.
* If ret_kb is not NULL the funtion will return the keyblock there.
*/
static int
key_byname( GETKEY_CTX *retctx, STRLIST namelist,
PKT_public_key *pk, PKT_secret_key *sk, KBNODE *ret_kb )
{
int rc = 0;
int n;
STRLIST r;
GETKEY_CTX ctx;
KBNODE help_kb = NULL;
if( retctx ) /* reset the returned context in case of error */
*retctx = NULL;
/* build the search context */
/* Performance hint: Use a static buffer if there is only one name */
/* and we don't have mode 6 */
for(n=0, r=namelist; r; r = r->next )
n++;
ctx = gcry_xcalloc( 1, sizeof *ctx + (n-1)*sizeof ctx->items );
ctx->nitems = n;
for(n=0, r=namelist; r; r = r->next, n++ ) {
ctx->items[n].mode = classify_user_id( r->d,
ctx->items[n].keyid,
ctx->items[n].fprint,
&ctx->items[n].name,
NULL );
if( !ctx->items[n].mode ) {
gcry_free( ctx );
return GPGERR_INV_USER_ID;
}
if( ctx->items[n].mode == 6 ) {
ctx->items[n].namebuf = prepare_word_match(ctx->items[n].name);
ctx->items[n].name = ctx->items[n].namebuf;
}
}
/* and call the lookup function */
ctx->primary = 1; /* we want to look for the primary key only */
if ( !ret_kb )
ret_kb = &help_kb;
if( sk ) {
rc = lookup( ctx, ret_kb, 1 );
if ( !rc && sk ) {
sk_from_block ( ctx, sk, *ret_kb );
}
}
else {
rc = lookup( ctx, ret_kb, 0 );
if ( !rc && pk ) {
pk_from_block ( ctx, pk, *ret_kb, NULL /* FIXME need to get the namehash*/ );
}
}
release_kbnode ( help_kb );
if( retctx ) /* caller wants the context */
*retctx = ctx;
else {
/* Hmmm, why not get_pubkey-end here?? */
enum_keyblocks( 2, &ctx->kbpos, NULL ); /* close */
for(n=0; n < ctx->nitems; n++ )
gcry_free( ctx->items[n].namebuf );
gcry_free( ctx );
}
return rc;
}
int
get_pubkey_byname( GETKEY_CTX *retctx, PKT_public_key *pk,
const char *name, KBNODE *ret_keyblock )
{
int rc;
STRLIST namelist = NULL;
add_to_strlist( &namelist, name );
rc = key_byname( retctx, namelist, pk, NULL, ret_keyblock );
free_strlist( namelist );
return rc;
}
int
get_pubkey_bynames( GETKEY_CTX *retctx, PKT_public_key *pk,
STRLIST names, KBNODE *ret_keyblock )
{
return key_byname( retctx, names, pk, NULL, ret_keyblock );
}
int
get_pubkey_next( GETKEY_CTX ctx, PKT_public_key *pk, KBNODE *ret_keyblock )
{
int rc;
rc = lookup( ctx, ret_keyblock, 0 );
if ( !rc && pk && ret_keyblock )
pk_from_block ( ctx, pk, *ret_keyblock, NULL );
return rc;
}
void
get_pubkey_end( GETKEY_CTX ctx )
{
if( ctx ) {
int n;
enum_keyblocks( 2, &ctx->kbpos, NULL ); /* close */
for(n=0; n < ctx->nitems; n++ )
gcry_free( ctx->items[n].namebuf );
if( !ctx->not_allocated )
gcry_free( ctx );
}
}
/****************
* Search for a key with the given fingerprint.
*/
int
get_pubkey_byfprint( PKT_public_key *pk,
const byte *fprint, size_t fprint_len)
{
int rc;
if( fprint_len == 20 || fprint_len == 16 ) {
struct getkey_ctx_s ctx;
KBNODE kb = NULL;
memset( &ctx, 0, sizeof ctx );
ctx.not_allocated = 1;
ctx.nitems = 1;
ctx.items[0].mode = fprint_len;
memcpy( ctx.items[0].fprint, fprint, fprint_len );
rc = lookup( &ctx, &kb, 0 );
if (!rc && pk )
pk_from_block ( &ctx, pk, kb, NULL );
release_kbnode ( kb );
get_pubkey_end( &ctx );
}
else
rc = GPGERR_GENERAL; /* Oops */
return rc;
}
/****************
* Search for a key with the given fingerprint and return the
* complete keyblock which may have more than only this key.
*/
int
get_keyblock_byfprint( KBNODE *ret_keyblock, const byte *fprint,
size_t fprint_len )
{
int rc;
if( fprint_len == 20 || fprint_len == 16 ) {
struct getkey_ctx_s ctx;
memset( &ctx, 0, sizeof ctx );
ctx.not_allocated = 1;
ctx.nitems = 1;
ctx.items[0].mode = fprint_len;
memcpy( ctx.items[0].fprint, fprint, fprint_len );
rc = lookup( &ctx, ret_keyblock, 0 );
get_pubkey_end( &ctx );
}
else
rc = GPGERR_GENERAL; /* Oops */
return rc;
}
/****************
* Search for a key with the given lid and return the entire keyblock
*/
int
get_keyblock_bylid( KBNODE *ret_keyblock, ulong lid )
{
int rc;
struct getkey_ctx_s ctx;
u32 kid[2];
if( keyid_from_lid( lid, kid ) )
kid[0] = kid[1] = 0;
memset( &ctx, 0, sizeof ctx );
ctx.not_allocated = 1;
ctx.nitems = 1;
ctx.items[0].mode = 12;
ctx.items[0].keyid[0] = kid[0];
ctx.items[0].keyid[1] = kid[1];
rc = lookup( &ctx, ret_keyblock, 0 );
get_pubkey_end( &ctx );
return rc;
}
/****************
* Get a secret key by name and store it into sk
* If NAME is NULL use the default key
*/
int
get_seckey_byname( PKT_secret_key *sk, const char *name, int unprotect )
{
STRLIST namelist = NULL;
int rc;
if( !name && opt.def_secret_key && *opt.def_secret_key ) {
add_to_strlist( &namelist, opt.def_secret_key );
rc = key_byname( NULL, namelist, NULL, sk, NULL );
}
else if( !name ) { /* use the first one as default key */
struct getkey_ctx_s ctx;
KBNODE kb = NULL;
memset( &ctx, 0, sizeof ctx );
ctx.not_allocated = 1;
ctx.primary = 1;
ctx.nitems = 1;
ctx.items[0].mode = 15;
rc = lookup( &ctx, &kb, 1 );
if (!rc && sk )
sk_from_block ( &ctx, sk, kb );
release_kbnode ( kb );
get_seckey_end( &ctx );
}
else {
add_to_strlist( &namelist, name );
rc = key_byname( NULL, namelist, NULL, sk, NULL );
}
free_strlist( namelist );
if( !rc && unprotect )
rc = check_secret_key( sk, 0 );
return rc;
}
int
get_seckey_bynames( GETKEY_CTX *retctx, PKT_secret_key *sk,
STRLIST names, KBNODE *ret_keyblock )
{
return key_byname( retctx, names, NULL, sk, ret_keyblock );
}
int
get_seckey_next( GETKEY_CTX ctx, PKT_secret_key *sk, KBNODE *ret_keyblock )
{
int rc;
rc = lookup( ctx, ret_keyblock, 1 );
if ( !rc && sk && ret_keyblock )
sk_from_block ( ctx, sk, *ret_keyblock );
return rc;
}
void
get_seckey_end( GETKEY_CTX ctx )
{
get_pubkey_end( ctx );
}
/*******************************************************
************** compare functions **********************
*******************************************************/
/****************
* Do a word match (original user id starts with a '+').
* The pattern is already tokenized to a more suitable format:
* There are only the real words in it delimited by one space
* and all converted to uppercase.
*
* Returns: 0 if all words match.
*
* Note: This algorithm is a straightforward one and not very
* fast. It works for UTF-8 strings. The uidlen should
* be removed but due to the fact that old versions of
* pgp don't use UTF-8 we still use the length; this should
* be fixed in parse-packet (and replace \0 by some special
* UTF-8 encoding)
*/
static int
word_match( const byte *uid, size_t uidlen, const byte *pattern )
{
size_t wlen, n;
const byte *p;
const byte *s;
for( s=pattern; *s; ) {
do {
/* skip leading delimiters */
while( uidlen && !word_match_chars[*uid] )
uid++, uidlen--;
/* get length of the word */
n = uidlen; p = uid;
while( n && word_match_chars[*p] )
p++, n--;
wlen = p - uid;
/* and compare against the current word from pattern */
for(n=0, p=uid; n < wlen && s[n] != ' ' && s[n] ; n++, p++ ) {
if( word_match_chars[*p] != s[n] )
break;
}
if( n == wlen && (s[n] == ' ' || !s[n]) )
break; /* found */
uid += wlen;
uidlen -= wlen;
} while( uidlen );
if( !uidlen )
return -1; /* not found */
/* advance to next word in pattern */
for(; *s != ' ' && *s ; s++ )
;
if( *s )
s++ ;
}
return 0; /* found */
}
/****************
* prepare word word_match; that is parse the name and
* build the pattern.
* caller has to free the returned pattern
*/
static char*
prepare_word_match( const byte *name )
{
byte *pattern, *p;
int c;
/* the original length is always enough for the pattern */
p = pattern = gcry_xmalloc(strlen(name)+1);
do {
/* skip leading delimiters */
while( *name && !word_match_chars[*name] )
name++;
/* copy as long as we don't have a delimiter and convert
* to uppercase.
* fixme: how can we handle utf8 uppercasing */
for( ; *name && (c=word_match_chars[*name]); name++ )
*p++ = c;
*p++ = ' '; /* append pattern delimiter */
} while( *name );
p[-1] = 0; /* replace last pattern delimiter by EOS */
return pattern;
}
static int
compare_name( const char *uid, size_t uidlen, const char *name, int mode )
{
int i;
const char *s, *se;
if( mode == 1 ) { /* exact match */
for(i=0; name[i] && uidlen; i++, uidlen-- )
if( uid[i] != name[i] )
break;
if( !uidlen && !name[i] )
return 0; /* found */
}
else if( mode == 2 ) { /* case insensitive substring */
if( memistr( uid, uidlen, name ) )
return 0;
}
else if( mode >= 3 && mode <= 5 ) { /* look at the email address */
for( i=0, s= uid; i < uidlen && *s != '<'; s++, i++ )
;
if( i < uidlen ) {
/* skip opening delim and one char and look for the closing one*/
s++; i++;
for( se=s+1, i++; i < uidlen && *se != '>'; se++, i++ )
;
if( i < uidlen ) {
i = se - s;
if( mode == 3 ) { /* exact email address */
if( strlen(name)-2 == i && !memicmp( s, name+1, i) )
return 0;
}
else if( mode == 4 ) { /* email substring */
if( memistr( s, i, name ) )
return 0;
}
else { /* email from end */
/* nyi */
}
}
}
}
else if( mode == 6 )
return word_match( uid, uidlen, name );
else
BUG();
return -1; /* not found */
}
/************************************************
************* Merging stuff ********************
************************************************/
/****************
* merge all selfsignatures with the keys.
* FIXME: replace this at least for the public key parts
* by merge_selfsigs
*/
void
merge_keys_and_selfsig( KBNODE keyblock )
{
PKT_public_key *pk = NULL;
PKT_secret_key *sk = NULL;
PKT_signature *sig;
KBNODE k;
u32 kid[2] = { 0, 0 };
u32 sigdate = 0;
for(k=keyblock; k; k = k->next ) {
if( k->pkt->pkttype == PKT_PUBLIC_KEY
|| k->pkt->pkttype == PKT_PUBLIC_SUBKEY ) {
pk = k->pkt->pkt.public_key; sk = NULL;
if( pk->version < 4 )
pk = NULL; /* not needed for old keys */
else if( k->pkt->pkttype == PKT_PUBLIC_KEY )
keyid_from_pk( pk, kid );
else if( !pk->expiredate ) { /* and subkey */
/* insert the expiration date here */
/*FIXME!!! pk->expiredate = subkeys_expiretime( k, kid );*/
}
sigdate = 0;
}
else if( k->pkt->pkttype == PKT_SECRET_KEY
|| k->pkt->pkttype == PKT_SECRET_SUBKEY ) {
pk = NULL; sk = k->pkt->pkt.secret_key;
if( sk->version < 4 )
sk = NULL;
else if( k->pkt->pkttype == PKT_SECRET_KEY )
keyid_from_sk( sk, kid );
sigdate = 0;
}
else if( (pk || sk ) && k->pkt->pkttype == PKT_SIGNATURE
&& (sig=k->pkt->pkt.signature)->sig_class >= 0x10
&& sig->sig_class <= 0x30 && sig->version > 3
&& !(sig->sig_class == 0x18 || sig->sig_class == 0x28)
&& sig->keyid[0] == kid[0] && sig->keyid[1] == kid[1] ) {
/* okay this is a self-signature which can be used.
* This is not used for subkey binding signature, becuase this
* is done above.
* FIXME: We should only use this if the signature is valid
* but this is time consuming - we must provide another
* way to handle this
*/
const byte *p;
u32 ed;
p = parse_sig_subpkt( sig->hashed_data, SIGSUBPKT_KEY_EXPIRE, NULL );
if( pk ) {
ed = p? pk->timestamp + buffer_to_u32(p):0;
if( sig->timestamp > sigdate ) {
pk->expiredate = ed;
sigdate = sig->timestamp;
}
}
else {
ed = p? sk->timestamp + buffer_to_u32(p):0;
if( sig->timestamp > sigdate ) {
sk->expiredate = ed;
sigdate = sig->timestamp;
}
}
}
}
}
static void
fixup_uidnode ( KBNODE uidnode, KBNODE signode )
{
PKT_user_id *uid = uidnode->pkt->pkt.user_id;
PKT_signature *sig = signode->pkt->pkt.signature;
const byte *p;
size_t n;
uid->created = 0; /* not created == invalid */
if ( !signode )
return; /* no self-signature */
if ( IS_UID_REV ( sig ) )
return; /* has been revoked */
uid->created = sig->timestamp; /* this one is okay */
/* store the key flags in the helper variable for later processing */
uid->help_key_usage = 0;
p = parse_sig_subpkt ( sig->hashed_data, SIGSUBPKT_KEY_FLAGS, &n );
if ( p && n ) {
/* first octet of the keyflags */
if ( (*p & 3) )
uid->help_key_usage |= GCRY_PK_USAGE_SIGN;
if ( (*p & 12) )
uid->help_key_usage |= GCRY_PK_USAGE_ENCR;
}
/* ditto or the key expiration */
uid->help_key_expire = 0;
p = parse_sig_subpkt ( sig->hashed_data, SIGSUBPKT_KEY_EXPIRE, NULL);
if ( p ) {
uid->help_key_expire = sig->timestamp + buffer_to_u32(p);
}
/* Set the primary user ID flag - we will later wipe out some
* of them to only have one in out keyblock */
uid->is_primary = 0;
p = parse_sig_subpkt ( sig->hashed_data, SIGSUBPKT_PRIMARY_UID, NULL );
if ( p && *p )
uid->is_primary = 1;
/* We could also query this from the unhashed area if it is not in
* the hased area and then later try to decide which is the better
* there should be no security problem with this.
* For now we only look at the hashed one.
*/
}
static void
merge_selfsigs_main( KBNODE keyblock, int *r_revoked )
{
PKT_public_key *pk = NULL;
KBNODE k;
u32 kid[2];
u32 sigdate = 0, uiddate=0, uiddate2;
KBNODE signode, uidnode, uidnode2;
u32 curtime = make_timestamp ();
unsigned int key_usage = 0;
u32 key_expire = 0;
int key_expire_seen = 0;
*r_revoked = 0;
if ( keyblock->pkt->pkttype != PKT_PUBLIC_KEY )
BUG ();
pk = keyblock->pkt->pkt.public_key;
pk->created = 0;
keyid_from_pk( pk, kid );
pk->main_keyid[0] = kid[0];
pk->main_keyid[1] = kid[1];
if ( pk->version < 4 )
return; /* nothing to do for old keys FIXME: This is wrong!!!!*/
/* first pass: find the latest direct key self-signature.
* We assume that the newest one overrides all others
*/
signode = NULL;
sigdate = 0; /* helper to find the latest signature */
for(k=keyblock; k && k->pkt->pkttype != PKT_PUBLIC_SUBKEY; k = k->next ) {
if ( k->pkt->pkttype == PKT_SIGNATURE ) {
PKT_signature *sig = k->pkt->pkt.signature;
if ( sig->keyid[0] == kid[0] && sig->keyid[1]==kid[1] ) {
if ( check_key_signature( keyblock, k, NULL ) )
; /* signature did not verify */
else if ( IS_KEY_REV (sig) ){
/* key has been revoked - there is no way to override
* such a revocation, so we can stop now.
* we can't cope with expiration times for revocations
* here because we have to assumethat an attacker can
* generate all kinds of signatures.
*/
*r_revoked = 1;
return;
}
else if ( IS_KEY_SIG (sig) && sig->timestamp >= sigdate ) {
const byte *p;
p = parse_sig_subpkt( sig->hashed_data,
SIGSUBPKT_SIG_EXPIRE, NULL );
if ( p && (sig->timestamp + buffer_to_u32(p)) >= curtime )
; /* signature has expired - ignore it */
else {
sigdate = sig->timestamp;
signode = k;
}
}
}
}
}
if ( signode ) {
/* some information from a direct key signature take precedence
* over the same information given in UID sigs.
*/
PKT_signature *sig = signode->pkt->pkt.signature;
const byte *p;
size_t n;
p = parse_sig_subpkt ( sig->hashed_data, SIGSUBPKT_KEY_FLAGS, &n );
if ( p && n ) {
/* first octet of the keyflags */
if ( (*p & 3) )
key_usage |= GCRY_PK_USAGE_SIGN;
if ( (*p & 12) )
key_usage |= GCRY_PK_USAGE_ENCR;
}
p = parse_sig_subpkt ( sig->hashed_data, SIGSUBPKT_KEY_EXPIRE, NULL);
if ( p ) {
key_expire = sig->timestamp + buffer_to_u32(p);
key_expire_seen = 1;
}
/* and set the created field */
pk->created = sigdate;
/* and mark that key as valid: one direct key signature should
* render a key as valid */
pk->is_valid = 1;
}
/* second pass: look at the self-signature of all user IDs */
signode = uidnode = NULL;
sigdate = 0; /* helper to find the latest signature in one user ID */
uiddate = 0; /* and over of all user IDs */
for(k=keyblock; k && k->pkt->pkttype != PKT_PUBLIC_SUBKEY; k = k->next ) {
if ( k->pkt->pkttype == PKT_USER_ID
|| k->pkt->pkttype == PKT_PHOTO_ID ) {
if ( uidnode )
fixup_uidnode ( uidnode, signode );
uidnode = k;
signode = NULL;
if ( sigdate > uiddate )
uiddate = sigdate;
sigdate = 0;
}
else if ( k->pkt->pkttype == PKT_SIGNATURE && uidnode ) {
PKT_signature *sig = k->pkt->pkt.signature;
if ( sig->keyid[0] == kid[0] && sig->keyid[1]==kid[1] ) {
if ( check_key_signature( keyblock, k, NULL ) )
; /* signature did not verify */
else if ( IS_UID_SIG (sig) || IS_UID_REV (sig)) {
/* Note: we allow to invalidated cert revocations
* by a newer signature. An attacker can't use this
* because a key should be revoced with a key revocation.
* The reason why we have to allow for that is that at
* one time an email address may become invalid but later
* the same email address may become valid again (hired,
* fired, hired again).
*/
const byte *p;
p = parse_sig_subpkt( sig->hashed_data,
SIGSUBPKT_SIG_EXPIRE, NULL );
if ( p && (sig->timestamp + buffer_to_u32(p)) >= curtime )
; /* signature/revocation has expired - ignore it */
else {
sigdate = sig->timestamp;
signode = k;
}
}
}
}
}
if ( uidnode ) {
fixup_uidnode ( uidnode, signode );
pk->is_valid = 1;
}
if ( sigdate > uiddate )
uiddate = sigdate;
/* if we do not have a direct key signature, take the key creation date
* from the latest user ID. Hmmm, another possibilty would be to take
* it from the latest primary user ID - but we don't implement it for
* now */
if ( !pk->created )
pk->created = uiddate;
if ( !pk->created ) {
/* oops, still no creation date: use the timestamp */
if (DBG_CACHE)
log_debug( "merge_selfsigs_main: "
"using timestamp as creation date\n");
pk->created = pk->timestamp;
}
/* Now that we had a look at all user IDs we can now get some information
* from those user IDs.
*/
if ( !key_usage ) {
/* find the latest user ID with key flags set */
uiddate = 0; /* helper to find the latest user ID */
for(k=keyblock; k && k->pkt->pkttype != PKT_PUBLIC_SUBKEY;
k = k->next ) {
if ( k->pkt->pkttype == PKT_USER_ID
|| k->pkt->pkttype == PKT_PHOTO_ID ) {
PKT_user_id *uid = k->pkt->pkt.user_id;
if ( uid->help_key_usage && uid->created > uiddate ) {
key_usage = uid->help_key_usage;
uiddate = uid->created;
}
}
}
}
if ( !key_usage ) { /* no key flags at all: get it from the algo */
key_usage = openpgp_pk_algo_usage ( pk->pubkey_algo );
}
else { /* check that the usage matches the usage as given by the algo */
int x = openpgp_pk_algo_usage ( pk->pubkey_algo );
if ( x ) /* mask it down to the actual allowed usage */
key_usage &= x;
}
pk->pubkey_usage = key_usage;
if ( !key_expire_seen ) {
/* find the latest valid user ID with a key expiration set
* Note, that this may be a diferent one from the above because
* some user IDs may have no expiration date set */
uiddate = 0;
for(k=keyblock; k && k->pkt->pkttype != PKT_PUBLIC_SUBKEY;
k = k->next ) {
if ( k->pkt->pkttype == PKT_USER_ID
|| k->pkt->pkttype == PKT_PHOTO_ID ) {
PKT_user_id *uid = k->pkt->pkt.user_id;
if ( uid->help_key_expire && uid->created > uiddate ) {
key_expire = uid->help_key_expire;
uiddate = uid->created;
}
}
}
}
if ( key_expire >= curtime )
pk->has_expired = key_expire;
/* FIXME: we should see how to get rid of the expiretime fields */
/* and now find the real primary user ID and delete all others */
uiddate = uiddate2 = 0;
uidnode = uidnode2 = NULL;
for(k=keyblock; k && k->pkt->pkttype != PKT_PUBLIC_SUBKEY; k = k->next ) {
if ( k->pkt->pkttype == PKT_USER_ID
|| k->pkt->pkttype == PKT_PHOTO_ID ) {
PKT_user_id *uid = k->pkt->pkt.user_id;
if ( uid->is_primary && uid->created > uiddate ) {
uiddate = uid->created;
uidnode = k;
}
if ( !uid->is_primary && uid->created > uiddate2 ) {
uiddate2 = uid->created;
uidnode2 = k;
}
}
}
if ( uidnode ) {
for(k=keyblock; k && k->pkt->pkttype != PKT_PUBLIC_SUBKEY;
k = k->next ) {
if ( k->pkt->pkttype == PKT_USER_ID
|| k->pkt->pkttype == PKT_PHOTO_ID ) {
PKT_user_id *uid = k->pkt->pkt.user_id;
if ( k != uidnode )
uid->is_primary = 0;
}
}
}
else if( uidnode2 ) {
/* none is flagged primary - use the latest user ID we have */
uidnode2->pkt->pkt.user_id->is_primary = 1;
}
}
static void
merge_selfsigs_subkey( KBNODE keyblock, KBNODE subnode )
{
PKT_public_key *mainpk = NULL, *subpk = NULL;
PKT_signature *sig;
KBNODE k;
u32 mainkid[2];
u32 sigdate = 0;
KBNODE signode;
u32 curtime = make_timestamp ();
unsigned int key_usage = 0;
u32 key_expire = 0;
const byte *p;
size_t n;
if ( subnode->pkt->pkttype != PKT_PUBLIC_SUBKEY )
BUG ();
mainpk = keyblock->pkt->pkt.public_key;
if ( mainpk->version < 4 )
return; /* (actually this should never happen) */
keyid_from_pk( mainpk, mainkid );
subpk = subnode->pkt->pkt.public_key;
subpk->is_valid = 0;
subpk->main_keyid[0] = mainpk->main_keyid[0];
subpk->main_keyid[1] = mainpk->main_keyid[1];
if ( subpk->version < 4 )
return; /* there are no v3 subkeys */
/* find the latest key binding self-signature. */
signode = NULL;
sigdate = 0; /* helper to find the latest signature */
for(k=subnode->next; k && k->pkt->pkttype != PKT_PUBLIC_SUBKEY;
k = k->next ) {
if ( k->pkt->pkttype == PKT_SIGNATURE ) {
sig = k->pkt->pkt.signature;
if ( sig->keyid[0] == mainkid[0] && sig->keyid[1]==mainkid[1] ) {
if ( check_key_signature( keyblock, k, NULL ) )
; /* signature did not verify */
else if ( IS_SUBKEY_REV (sig) ) {
/* key has been revoked - given the fact that it is easy
* to create a new subkey, it does not make sense to
* revive a revoked key. So we can stop here.
*/
subpk->is_revoked = 1;
return;
}
else if ( IS_SUBKEY_SIG (sig) && sig->timestamp >= sigdate ) {
p = parse_sig_subpkt( sig->hashed_data,
SIGSUBPKT_SIG_EXPIRE, NULL );
if ( p && (sig->timestamp + buffer_to_u32(p)) >= curtime )
; /* signature has expired - ignore it */
else {
sigdate = sig->timestamp;
signode = k;
}
}
}
}
}
if ( !signode ) {
subpk->created = subpk->timestamp;
return; /* no valid key binding */
}
subpk->is_valid = 1;
subpk->created = sigdate;
sig = signode->pkt->pkt.signature;
p = parse_sig_subpkt ( sig->hashed_data, SIGSUBPKT_KEY_FLAGS, &n );
if ( p && n ) {
/* first octet of the keyflags */
if ( (*p & 3) )
key_usage |= GCRY_PK_USAGE_SIGN;
if ( (*p & 12) )
key_usage |= GCRY_PK_USAGE_ENCR;
}
if ( !key_usage ) { /* no key flags at all: get it from the algo */
key_usage = openpgp_pk_algo_usage ( subpk->pubkey_algo );
}
else { /* check that the usage matches the usage as given by the algo */
int x = openpgp_pk_algo_usage ( subpk->pubkey_algo );
if ( x ) /* mask it down to the actual allowed usage */
key_usage &= x;
}
subpk->pubkey_usage = key_usage;
p = parse_sig_subpkt ( sig->hashed_data, SIGSUBPKT_KEY_EXPIRE, NULL);
if ( p )
key_expire = sig->timestamp + buffer_to_u32(p);
else
key_expire = 0;
subpk->has_expired = key_expire >= curtime? key_expire : 0;
}
/*
* Merge information from the self-signatures with the key, so that
* we can later use them more easy.
* The function works by first applying the self signatures to the
* primary key and the to each subkey.
* Here are the rules we use to decide which inormation from which
* self-signature is used:
* We check all self signatures or validity and ignore all invalid signatures.
* All signatures are then ordered by their creation date ....
* For the primary key:
* FIXME the docs
*/
static void
merge_selfsigs( KBNODE keyblock )
{
KBNODE k;
int revoked;
PKT_public_key *main_pk;
if ( keyblock->pkt->pkttype != PKT_PUBLIC_KEY )
BUG ();
merge_selfsigs_main ( keyblock, &revoked );
main_pk = keyblock->pkt->pkt.public_key;
if ( revoked ) {
/* if the primary key has been revoked we better set the revoke
* flag on that key and all subkeys */
for(k=keyblock; k; k = k->next ) {
if ( k->pkt->pkttype == PKT_PUBLIC_KEY
|| k->pkt->pkttype == PKT_PUBLIC_SUBKEY ) {
PKT_public_key *pk = k->pkt->pkt.public_key;
pk->is_revoked = 1;
pk->main_keyid[0] = main_pk->main_keyid[0];
pk->main_keyid[1] = main_pk->main_keyid[1];
}
}
return;
}
/* now merge in the data from each of the subkeys */
for(k=keyblock; k; k = k->next ) {
if ( k->pkt->pkttype == PKT_PUBLIC_SUBKEY ) {
merge_selfsigs_subkey ( keyblock, k );
}
}
}
/*
* Merge the secret keys from secblock into the pubblock thereby
* replacing the public (sub)keys with their secret counterparts Hmmm:
* It might be better to get away from the concept of entire secret
* keys at all and have a way to store just the real secret parts
* from the key.
*/
static void
merge_public_with_secret ( KBNODE pubblock, KBNODE secblock )
{
KBNODE pub;
int deleting = 0;
int any_deleted = 0;
assert ( pubblock->pkt->pkttype == PKT_PUBLIC_KEY );
assert ( secblock->pkt->pkttype == PKT_SECRET_KEY );
for (pub=pubblock; pub; pub = pub->next ) {
if ( pub->pkt->pkttype == PKT_PUBLIC_KEY ) {
PKT_public_key *pk = pub->pkt->pkt.public_key;
PKT_secret_key *sk = secblock->pkt->pkt.secret_key;
assert ( pub == pubblock ); /* only in the first node */
/* there is nothing to compare in this case, so just replace
* some information */
copy_public_parts_to_secret_key ( pk, sk );
free_public_key ( pk );
pub->pkt->pkttype = PKT_SECRET_KEY;
pub->pkt->pkt.secret_key = copy_secret_key (NULL, sk);
}
else if ( pub->pkt->pkttype == PKT_PUBLIC_SUBKEY ) {
KBNODE sec;
PKT_public_key *pk = pub->pkt->pkt.public_key;
deleting = 0;
/* this is more complicated: it may happen that the sequence
* of the subkeys dosn't match, so we have to find the
* appropriate secret key */
for (sec=secblock->next; sec; sec = sec->next ) {
if ( sec->pkt->pkttype == PKT_SECRET_SUBKEY ) {
PKT_secret_key *sk = sec->pkt->pkt.secret_key;
if ( !cmp_public_secret_key ( pk, sk ) ) {
copy_public_parts_to_secret_key ( pk, sk );
free_public_key ( pk );
pub->pkt->pkttype = PKT_SECRET_KEY;
pub->pkt->pkt.secret_key = copy_secret_key (NULL, sk);
break;
}
}
}
if ( !sec ) {
log_error ( "no corresponding secret subkey "
"for public subkey - removing\n" );
/* better remove the public subkey in this case */
delete_kbnode ( pub );
deleting = 1;
any_deleted = 1;
}
}
else if ( deleting ) {
delete_kbnode (pub);
}
}
if ( any_deleted ) {
/* because we have not deleted the root node, we don't need to
* update the pubblock */
pub = pubblock;
commit_kbnode ( &pubblock );
assert ( pub == pubblock );
}
}
/************************************************
************* Find stuff ***********************
************************************************/
static int
find_by_name( KBNODE keyblock, const char *name,
int mode, byte *namehash )
{
KBNODE k;
for(k=keyblock; k; k = k->next ) {
if( k->pkt->pkttype == PKT_USER_ID
&& !compare_name( k->pkt->pkt.user_id->name,
k->pkt->pkt.user_id->len, name, mode)) {
/* we found a matching name, look for the key */
if( k->pkt->pkt.user_id->photo ) {
/* oops: this can never happen */
gcry_md_hash_buffer( GCRY_MD_RMD160, namehash,
k->pkt->pkt.user_id->photo,
k->pkt->pkt.user_id->photolen );
}
else {
gcry_md_hash_buffer( GCRY_MD_RMD160, namehash,
k->pkt->pkt.user_id->name,
k->pkt->pkt.user_id->len );
}
return 1;
}
}
return 0;
}
static KBNODE
find_by_keyid( KBNODE keyblock, u32 *keyid, int mode )
{
KBNODE k;
for(k=keyblock; k; k = k->next ) {
if( k->pkt->pkttype == PKT_PUBLIC_KEY
|| k->pkt->pkttype == PKT_PUBLIC_SUBKEY ) {
u32 aki[2];
keyid_from_pk( k->pkt->pkt.public_key, aki );
if( aki[1] == keyid[1] && ( mode == 10 || aki[0] == keyid[0] ) ) {
return k; /* found */
}
}
}
return NULL;
}
static KBNODE
find_by_fpr( KBNODE keyblock, const char *name, int mode )
{
KBNODE k;
for(k=keyblock; k; k = k->next ) {
if( k->pkt->pkttype == PKT_PUBLIC_KEY
|| k->pkt->pkttype == PKT_PUBLIC_SUBKEY ) {
byte afp[MAX_FINGERPRINT_LEN];
size_t an;
fingerprint_from_pk(k->pkt->pkt.public_key, afp, &an );
if( an == mode && !memcmp( afp, name, an) ) {
return k;
}
}
}
return NULL;
}
/* See see whether the key fits
* our requirements and in case we do not
* request a the primary key, we should select
* a suitable subkey.
* FIXME: Check against PGP 7 whether we still need a kludge
* to favor type 16 keys over type 20 keys when type 20
* has not been explitely requested.
* Returns: True when a suitable key has been found.
*
* We have to distinguish four cases:
* 1. No usage and no primary key requested
* Examples for this case are that we have a keyID to be used
* for decrytion or verification.
* 2. No usage but primary key requested
* This is the case for all functions which work on an
* entire keyblock, e.g. for editing or listing
* 3. Usage and primary key requested
* FXME
* 4. Usage but no primary key requested
* FIXME
* FIXME: Tell what is going to happen here and something about the rationale
*
*/
static int
finish_lookup( GETKEY_CTX ctx, KBNODE foundk )
{
KBNODE keyblock = ctx->keyblock;
KBNODE k;
#define USAGE_MASK (GCRY_PK_USAGE_SIGN|GCRY_PK_USAGE_ENCR)
unsigned int req_usage = ( ctx->req_usage & USAGE_MASK );
u32 latest_date;
KBNODE latest_key;
assert( !foundk || foundk->pkt->pkttype == PKT_PUBLIC_KEY
|| foundk->pkt->pkttype == PKT_PUBLIC_SUBKEY );
assert( keyblock->pkt->pkttype == PKT_PUBLIC_KEY );
ctx->found_key = NULL;
if ( DBG_CACHE )
log_debug( "finish_lookup: checking %s (req_usage=%x)\n",
foundk? "one key":"all keys", req_usage);
latest_date = 0;
latest_key = NULL;
/* We do check the subkeys only if we either have requested a specific
* usage or have not requested to get the primary key. */
if ( (req_usage || !ctx->primary)
&& (!foundk || foundk->pkt->pkttype == PKT_PUBLIC_SUBKEY) ) {
KBNODE nextk;
/* either start a loop or check just this one subkey */
for (k=foundk?foundk:keyblock; k; k = nextk ) {
PKT_public_key *pk;
nextk = k->next;
if ( k->pkt->pkttype != PKT_PUBLIC_SUBKEY )
continue;
if ( foundk )
nextk = NULL; /* what a hack */
pk = k->pkt->pkt.public_key;
if ( !pk->is_valid ) {
if (DBG_CACHE)
log_debug( "\tsubkey not valid\n");
continue;
}
if ( pk->is_revoked ) {
if (DBG_CACHE)
log_debug( "\tsubkey has been revoked\n");
continue;
}
if ( pk->has_expired ) {
if (DBG_CACHE)
log_debug( "\tsubkey has expired\n");
continue;
}
if ( req_usage &&
!((pk->pubkey_usage&USAGE_MASK) & req_usage) ) {
if (DBG_CACHE)
log_debug( "\tusage does not match: want=%x have=%x\n",
req_usage, pk->pubkey_usage );
continue;
}
if (DBG_CACHE)
log_debug( "\tconsidering key created %lu\n",
(ulong)pk->created);
if ( pk->created > latest_date ) {
latest_date = pk->created;
latest_key = k;
}
}
}
if ( !latest_key ) {
PKT_public_key *pk;
if (DBG_CACHE && !foundk )
log_debug( "\tno suitable subkeys found - trying primary\n");
pk = keyblock->pkt->pkt.public_key;
if ( !pk->is_valid ) {
if (DBG_CACHE)
log_debug( "\tprimary key not valid\n");
}
else if ( pk->is_revoked ) {
if (DBG_CACHE)
log_debug( "\tprimary key has been revoked\n");
}
else if ( pk->has_expired ) {
if (DBG_CACHE)
log_debug( "\tprimary key has expired\n");
}
else if ( req_usage
&& !((pk->pubkey_usage&USAGE_MASK) & req_usage) ) {
if (DBG_CACHE)
log_debug( "\tusage does not match: want=%x have=%x\n",
req_usage, pk->pubkey_usage );
}
else { /* okay */
if (DBG_CACHE)
log_debug( "\tprimary key may be used\n");
latest_key = keyblock;
latest_date = pk->created;
}
}
if ( !latest_key ) {
if (DBG_CACHE)
log_debug("\tno suitable key found - giving up\n");
return 0;
}
if (DBG_CACHE)
log_debug( "\tusing key created %lu\n", (ulong)latest_date );
ctx->found_key = latest_key;
if ( latest_key != keyblock ) {
log_info(_("using secondary key %08lX "
"instead of primary key %08lX\n"),
(ulong)keyid_from_pk( latest_key->pkt->pkt.public_key, NULL),
(ulong)keyid_from_pk( keyblock->pkt->pkt.public_key, NULL) );
}
cache_user_id( keyblock );
return 1; /* found */
}
static int
lookup( GETKEY_CTX ctx, KBNODE *ret_keyblock, int secmode )
{
int rc;
int oldmode = set_packet_list_mode(0);
byte namehash[20];
int use_namehash=0;
KBNODE secblock = NULL; /* helper */
if( !ctx->count ) /* first time */
rc = enum_keyblocks( secmode?5:0, &ctx->kbpos, &ctx->keyblock );
else
rc = 0;
if( !rc ) {
while( !(rc = enum_keyblocks( 1, &ctx->kbpos, &ctx->keyblock )) ) {
int n;
getkey_item_t *item;
if ( secmode ) {
/* find the correspondig public key and use this
* this one for the selection process */
u32 aki[2];
KBNODE k = ctx->keyblock;
if ( k->pkt->pkttype != PKT_SECRET_KEY )
BUG();
keyid_from_sk( k->pkt->pkt.secret_key, aki );
k = get_pubkeyblock( aki );
if( !k ) {
log_info(_("key %08lX: secret key without public key "
"- skipped\n"), (ulong)aki[1] );
goto skip;
}
secblock = ctx->keyblock;
ctx->keyblock = k;
}
/* loop over all the user ids we want to look for */
item = ctx->items;
for(n=0; n < ctx->nitems; n++, item++ ) {
KBNODE k = NULL;
int found = 0;
if( item->mode < 10 ) {
found = find_by_name( ctx->keyblock,
item->name, item->mode,
namehash );
use_namehash = found;
}
else if( item->mode >= 10 && item->mode <= 12 ) {
k = find_by_keyid( ctx->keyblock,
item->keyid, item->mode );
found = !!k;
}
else if( item->mode == 15 ) {
found = 1;
}
else if( item->mode == 16 || item->mode == 20 ) {
k = find_by_fpr( ctx->keyblock,
item->fprint, item->mode );
found = !!k;
}
else
BUG();
if( found ) {
/* this keyblock looks fine - do further investigation */
merge_selfsigs ( ctx->keyblock );
if ( finish_lookup( ctx, k ) ) {
if ( secmode ) {
merge_public_with_secret ( ctx->keyblock,
secblock);
release_kbnode (secblock);
secblock = NULL;
}
goto found;
}
}
}
skip:
/* release resources and try the next keyblock */
if ( secmode ) {
release_kbnode( secblock );
secblock = NULL;
}
release_kbnode( ctx->keyblock );
ctx->keyblock = NULL;
}
found:
;
}
if( rc && rc != -1 )
log_error("enum_keyblocks failed: %s\n", gpg_errstr(rc));
if( !rc ) {
*ret_keyblock = ctx->keyblock; /* return the keyblock */
ctx->keyblock = NULL;
}
else if( rc == -1 )
rc = secmode ? GPGERR_NO_SECKEY : GPGERR_NO_PUBKEY;
if ( secmode ) {
release_kbnode( secblock );
secblock = NULL;
}
release_kbnode( ctx->keyblock );
ctx->keyblock = NULL;
set_packet_list_mode(oldmode);
#if 0
if( opt.debug & DBG_MEMSTAT_VALUE ) {
static int initialized;
if( !initialized ) {
initialized = 1;
atexit( print_stats );
}
assert( ctx->mode < DIM(lkup_stats) );
lkup_stats[ctx->mode].any = 1;
if( !rc )
lkup_stats[ctx->mode].okay_count++;
else if ( rc == GPGERR_NO_PUBKEY || rc == GPGERR_NO_SECKEY )
lkup_stats[ctx->mode].nokey_count++;
else
lkup_stats[ctx->mode].error_count++;
}
#endif
ctx->last_rc = rc;
ctx->count++;
return rc;
}
/****************
* FIXME: Replace by the generic function
* It does not work as it is right now - it is used at
* 2 places: a) to get the key for an anonyous recipient
* b) to get the ultimately trusted keys.
* The a) usage might have some problems.
*
* Enumerate all primary secret keys. Caller must use these procedure:
* 1) create a void pointer and initialize it to NULL
* 2) pass this void pointer by reference to this function
* and provide space for the secret key (pass a buffer for sk)
* 3) call this function as long as it does not return -1
* to indicate EOF.
* 4) Always call this function a last time with SK set to NULL,
* so that can free it's context.
*
*
*/
int
enum_secret_keys( void **context, PKT_secret_key *sk, int with_subkeys )
{
int rc=0;
PACKET pkt;
int save_mode;
struct {
int eof;
int sequence;
const char *name;
IOBUF iobuf;
} *c = *context;
if( !c ) { /* make a new context */
c = gcry_xcalloc( 1, sizeof *c );
*context = c;
c->sequence = 0;
c->name = enum_keyblock_resources( &c->sequence, 1 );
}
if( !sk ) { /* free the context */
if( c->iobuf )
iobuf_close(c->iobuf);
gcry_free( c );
*context = NULL;
return 0;
}
if( c->eof )
return -1;
/* FIXME: This assumes a plain keyring file */
for( ; c->name; c->name = enum_keyblock_resources( &c->sequence, 1 ) ) {
if( !c->iobuf ) {
if( !(c->iobuf = iobuf_open( c->name ) ) ) {
log_error("enum_secret_keys: can't open `%s'\n", c->name );
continue; /* try next file */
}
}
save_mode = set_packet_list_mode(0);
init_packet(&pkt);
while( (rc=parse_packet(c->iobuf, &pkt)) != -1 ) {
if( rc )
; /* e.g. unknown packet */
else if( pkt.pkttype == PKT_SECRET_KEY
|| ( with_subkeys && pkt.pkttype == PKT_SECRET_SUBKEY ) ) {
copy_secret_key( sk, pkt.pkt.secret_key );
set_packet_list_mode(save_mode);
return 0; /* found */
}
free_packet(&pkt);
}
set_packet_list_mode(save_mode);
iobuf_close(c->iobuf); c->iobuf = NULL;
}
c->eof = 1;
return -1;
}
/*********************************************
*********** user ID printing helpers *******
*********************************************/
/****************
* Return a string with a printable representation of the user_id.
* this string must be freed by m_free.
*/
char*
get_user_id_string( u32 *keyid )
{
user_id_db_t r;
char *p;
int pass=0;
/* try it two times; second pass reads from key resources */
do {
for(r=user_id_db; r; r = r->next ) {
keyid_list_t a;
for (a=r->keyids; a; a= a->next ) {
if( a->keyid[0] == keyid[0] && a->keyid[1] == keyid[1] ) {
p = gcry_xmalloc( r->len + 10 );
sprintf(p, "%08lX %.*s",
(ulong)keyid[1], r->len, r->name );
return p;
}
}
}
} while( ++pass < 2 && !get_pubkey( NULL, keyid ) );
p = gcry_xmalloc( 15 );
sprintf(p, "%08lX [?]", (ulong)keyid[1] );
return p;
}
char*
get_user_id_string_native( u32 *keyid )
{
char *p = get_user_id_string( keyid );
char *p2 = utf8_to_native( p, strlen(p) );
gcry_free(p);
return p2;
}
char*
get_long_user_id_string( u32 *keyid )
{
user_id_db_t r;
char *p;
int pass=0;
/* try it two times; second pass reads from key resources */
do {
for(r=user_id_db; r; r = r->next ) {
keyid_list_t a;
for (a=r->keyids; a; a= a->next ) {
if( a->keyid[0] == keyid[0] && a->keyid[1] == keyid[1] ) {
p = gcry_xmalloc( r->len + 20 );
sprintf(p, "%08lX%08lX %.*s",
(ulong)keyid[0], (ulong)keyid[1],
r->len, r->name );
return p;
}
}
}
} while( ++pass < 2 && !get_pubkey( NULL, keyid ) );
p = gcry_xmalloc( 25 );
sprintf(p, "%08lX%08lX [?]", (ulong)keyid[0], (ulong)keyid[1] );
return p;
}
char*
get_user_id( u32 *keyid, size_t *rn )
{
user_id_db_t r;
char *p;
int pass=0;
/* try it two times; second pass reads from key resources */
do {
for(r=user_id_db; r; r = r->next ) {
keyid_list_t a;
for (a=r->keyids; a; a= a->next ) {
if( a->keyid[0] == keyid[0] && a->keyid[1] == keyid[1] ) {
p = gcry_xmalloc( r->len );
memcpy(p, r->name, r->len );
*rn = r->len;
return p;
}
}
}
} while( ++pass < 2 && !get_pubkey( NULL, keyid ) );
p = gcry_xstrdup( _("[User id not found]") );
*rn = strlen(p);
return p;
}