gecko-dev/security/nss/lib/certhigh/certhtml.c

322 строки
9.0 KiB
C

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
/*
* certhtml.c --- convert a cert to html
*/
#include "seccomon.h"
#include "secitem.h"
#include "sechash.h"
#include "cert.h"
#include "keyhi.h"
#include "secder.h"
#include "prprf.h"
#include "secport.h"
#include "secasn1.h"
#include "pk11func.h"
static char *hex = "0123456789ABCDEF";
/*
** Convert a der-encoded integer to a hex printable string form
*/
char *
CERT_Hexify(SECItem *i, int do_colon)
{
unsigned char *cp, *end;
char *rv, *o;
if (!i->len) {
return PORT_Strdup("00");
}
rv = o = (char *)PORT_Alloc(i->len * 3);
if (!rv)
return rv;
cp = i->data;
end = cp + i->len;
while (cp < end) {
unsigned char ch = *cp++;
*o++ = hex[(ch >> 4) & 0xf];
*o++ = hex[ch & 0xf];
if (cp != end) {
if (do_colon) {
*o++ = ':';
}
}
}
*o = 0; /* Null terminate the string */
return rv;
}
#define BREAK "<br>"
#define BREAKLEN 4
#define COMMA ", "
#define COMMALEN 2
#define MAX_OUS 20
#define MAX_DC MAX_OUS
char *
CERT_FormatName(CERTName *name)
{
CERTRDN **rdns;
CERTRDN *rdn;
CERTAVA **avas;
CERTAVA *ava;
char *buf = 0;
char *tmpbuf = 0;
SECItem *cn = 0;
SECItem *email = 0;
SECItem *org = 0;
SECItem *loc = 0;
SECItem *state = 0;
SECItem *country = 0;
SECItem *dq = 0;
unsigned len = 0;
int tag;
int i;
int ou_count = 0;
int dc_count = 0;
PRBool first;
SECItem *orgunit[MAX_OUS];
SECItem *dc[MAX_DC];
/* Loop over name components and gather the interesting ones */
rdns = name->rdns;
while ((rdn = *rdns++) != 0) {
avas = rdn->avas;
while ((ava = *avas++) != 0) {
tag = CERT_GetAVATag(ava);
switch (tag) {
case SEC_OID_AVA_COMMON_NAME:
if (cn) {
break;
}
cn = CERT_DecodeAVAValue(&ava->value);
if (!cn) {
goto loser;
}
len += cn->len;
// cn will always have BREAK after it
len += BREAKLEN;
break;
case SEC_OID_AVA_COUNTRY_NAME:
if (country) {
break;
}
country = CERT_DecodeAVAValue(&ava->value);
if (!country) {
goto loser;
}
len += country->len;
// country may have COMMA after it (if we over-count len,
// that's fine - we'll just allocate a buffer larger than we
// need)
len += COMMALEN;
break;
case SEC_OID_AVA_LOCALITY:
if (loc) {
break;
}
loc = CERT_DecodeAVAValue(&ava->value);
if (!loc) {
goto loser;
}
len += loc->len;
// loc may have COMMA after it
len += COMMALEN;
break;
case SEC_OID_AVA_STATE_OR_PROVINCE:
if (state) {
break;
}
state = CERT_DecodeAVAValue(&ava->value);
if (!state) {
goto loser;
}
len += state->len;
// state currently won't have COMMA after it, but this is a
// (probably vain) attempt to future-proof this code
len += COMMALEN;
break;
case SEC_OID_AVA_ORGANIZATION_NAME:
if (org) {
break;
}
org = CERT_DecodeAVAValue(&ava->value);
if (!org) {
goto loser;
}
len += org->len;
// org will have BREAK after it
len += BREAKLEN;
break;
case SEC_OID_AVA_DN_QUALIFIER:
if (dq) {
break;
}
dq = CERT_DecodeAVAValue(&ava->value);
if (!dq) {
goto loser;
}
len += dq->len;
// dq will have BREAK after it
len += BREAKLEN;
break;
case SEC_OID_AVA_ORGANIZATIONAL_UNIT_NAME:
if (ou_count < MAX_OUS) {
orgunit[ou_count] = CERT_DecodeAVAValue(&ava->value);
if (!orgunit[ou_count]) {
goto loser;
}
len += orgunit[ou_count++]->len;
// each ou will have BREAK after it
len += BREAKLEN;
}
break;
case SEC_OID_AVA_DC:
if (dc_count < MAX_DC) {
dc[dc_count] = CERT_DecodeAVAValue(&ava->value);
if (!dc[dc_count]) {
goto loser;
}
len += dc[dc_count++]->len;
// each dc will have BREAK after it
len += BREAKLEN;
}
break;
case SEC_OID_PKCS9_EMAIL_ADDRESS:
case SEC_OID_RFC1274_MAIL:
if (email) {
break;
}
email = CERT_DecodeAVAValue(&ava->value);
if (!email) {
goto loser;
}
len += email->len;
// email will have BREAK after it
len += BREAKLEN;
break;
default:
break;
}
}
}
// there may be a final BREAK
len += BREAKLEN;
/* allocate buffer */
buf = (char *)PORT_Alloc(len);
if (!buf) {
goto loser;
}
tmpbuf = buf;
if (cn) {
PORT_Memcpy(tmpbuf, cn->data, cn->len);
tmpbuf += cn->len;
PORT_Memcpy(tmpbuf, BREAK, BREAKLEN);
tmpbuf += BREAKLEN;
}
if (email) {
PORT_Memcpy(tmpbuf, email->data, email->len);
tmpbuf += (email->len);
PORT_Memcpy(tmpbuf, BREAK, BREAKLEN);
tmpbuf += BREAKLEN;
}
for (i = ou_count - 1; i >= 0; i--) {
PORT_Memcpy(tmpbuf, orgunit[i]->data, orgunit[i]->len);
tmpbuf += (orgunit[i]->len);
PORT_Memcpy(tmpbuf, BREAK, BREAKLEN);
tmpbuf += BREAKLEN;
}
if (dq) {
PORT_Memcpy(tmpbuf, dq->data, dq->len);
tmpbuf += (dq->len);
PORT_Memcpy(tmpbuf, BREAK, BREAKLEN);
tmpbuf += BREAKLEN;
}
if (org) {
PORT_Memcpy(tmpbuf, org->data, org->len);
tmpbuf += (org->len);
PORT_Memcpy(tmpbuf, BREAK, BREAKLEN);
tmpbuf += BREAKLEN;
}
for (i = dc_count - 1; i >= 0; i--) {
PORT_Memcpy(tmpbuf, dc[i]->data, dc[i]->len);
tmpbuf += (dc[i]->len);
PORT_Memcpy(tmpbuf, BREAK, BREAKLEN);
tmpbuf += BREAKLEN;
}
first = PR_TRUE;
if (loc) {
PORT_Memcpy(tmpbuf, loc->data, loc->len);
tmpbuf += (loc->len);
first = PR_FALSE;
}
if (state) {
if (!first) {
PORT_Memcpy(tmpbuf, COMMA, COMMALEN);
tmpbuf += COMMALEN;
}
PORT_Memcpy(tmpbuf, state->data, state->len);
tmpbuf += (state->len);
first = PR_FALSE;
}
if (country) {
if (!first) {
PORT_Memcpy(tmpbuf, COMMA, COMMALEN);
tmpbuf += COMMALEN;
}
PORT_Memcpy(tmpbuf, country->data, country->len);
tmpbuf += (country->len);
first = PR_FALSE;
}
if (!first) {
PORT_Memcpy(tmpbuf, BREAK, BREAKLEN);
tmpbuf += BREAKLEN;
}
*tmpbuf = 0;
/* fall through and clean */
loser:
if (cn) {
SECITEM_FreeItem(cn, PR_TRUE);
}
if (email) {
SECITEM_FreeItem(email, PR_TRUE);
}
for (i = ou_count - 1; i >= 0; i--) {
SECITEM_FreeItem(orgunit[i], PR_TRUE);
}
if (dq) {
SECITEM_FreeItem(dq, PR_TRUE);
}
if (org) {
SECITEM_FreeItem(org, PR_TRUE);
}
for (i = dc_count - 1; i >= 0; i--) {
SECITEM_FreeItem(dc[i], PR_TRUE);
}
if (loc) {
SECITEM_FreeItem(loc, PR_TRUE);
}
if (state) {
SECITEM_FreeItem(state, PR_TRUE);
}
if (country) {
SECITEM_FreeItem(country, PR_TRUE);
}
return (buf);
}