You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by Ruediger Pluem <rp...@apache.org> on 2009/10/26 00:44:17 UTC
Re: svn commit: r829619 - in /httpd/httpd/trunk: ./ modules/ssl/
On 10/25/2009 06:21 PM, jorton@apache.org wrote:
> Author: jorton
> Date: Sun Oct 25 17:21:10 2009
> New Revision: 829619
>
> URL: http://svn.apache.org/viewvc?rev=829619&view=rev
> Log:
> Add support for OCSP "stapling":
>
> * modules/ssl/ssl_util_stapling.c: New file.
>
> * modules/ssl/config.m4, modules/ssl/mod_ssl.dsp: Build it.
>
> * modules/ssl/ssl_toolkit_compat.h: Define HAVE_OCSP_STAPLING if
> OpenSSL is of suitable version (>= 0.9.8g) and capability (TLS
> extension support enabled).
>
> * modules/ssl/mod_ssl.c: Add config directives.
>
> * modules/ssl/ssl_private.h: Add prototypes for new functions.
> (SSLModConfigRec): Add fields for stapling socache instance and
> associated mutex.
> (modssl_ctx_t): Add config fields for stapling.
>
> * modules/ssl/ssl_engine_init.c (ssl_init_Module, ssl_init_Child):
> Call the stapling initialization functions.
>
> * modules/ssl/ssl_engine_config.c: Add config hooks.
>
> * modules/ssl/ssl_scache.c: Create, initialize and destroy the socache
> instance for OCSP responses.
>
> Submitted by: Dr Stephen Henson <shenson oss-institute.org>
Will there be documentation patches for the new directives?
>
> Added:
> httpd/httpd/trunk/modules/ssl/ssl_util_stapling.c
> Modified:
> httpd/httpd/trunk/CHANGES
> httpd/httpd/trunk/modules/ssl/config.m4
> httpd/httpd/trunk/modules/ssl/mod_ssl.c
> httpd/httpd/trunk/modules/ssl/mod_ssl.dsp
> httpd/httpd/trunk/modules/ssl/ssl_engine_config.c
> httpd/httpd/trunk/modules/ssl/ssl_engine_init.c
> httpd/httpd/trunk/modules/ssl/ssl_private.h
> httpd/httpd/trunk/modules/ssl/ssl_scache.c
> httpd/httpd/trunk/modules/ssl/ssl_toolkit_compat.h
>
> Added: httpd/httpd/trunk/modules/ssl/ssl_util_stapling.c
> URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/modules/ssl/ssl_util_stapling.c?rev=829619&view=auto
> ==============================================================================
> --- httpd/httpd/trunk/modules/ssl/ssl_util_stapling.c (added)
> +++ httpd/httpd/trunk/modules/ssl/ssl_util_stapling.c Sun Oct 25 17:21:10 2009
> +/*
> + * OCSP response caching code. The response is preceded by a flag value
> + * which indicates whether the response was invalid when it was stored.
> + * the purpose of this flag is to avoid repeated queries to a server
> + * which has given an invalid response while allowing a response which
> + * has subsequently become invalid to be retried immediately.
> + *
> + * The key for the cache is the hash of the certificate the response
> + * is for.
> + */
> +static BOOL stapling_cache_response(server_rec *s, modssl_ctx_t *mctx,
> + OCSP_RESPONSE *rsp, certinfo *cinf,
> + BOOL ok, apr_pool_t *pool)
> +{
> + SSLModConfigRec *mc = myModConfig(s);
> + unsigned char resp_der[MAX_STAPLING_DER];
> + unsigned char *p;
> + int resp_derlen;
> + BOOL rv;
> + time_t timeout;
> +
> + resp_derlen = i2d_OCSP_RESPONSE(rsp, NULL) + 1;
> +
> + if (resp_derlen <= 0) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "OCSP stapling response encode error??");
> + return FALSE;
> + }
> +
> + if (resp_derlen > sizeof resp_der) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "OCSP stapling response too big (%u bytes)", resp_derlen);
> + return FALSE;
> + }
> +
> +
> + p = resp_der;
> +
> + if (ok == TRUE) {
> + *p++ = 1;
> + timeout = mctx->stapling_cache_timeout;
> + } else {
> + *p++ = 0;
> + timeout = mctx->stapling_errcache_timeout;
> + }
> +
> + timeout += time(NULL);
Shouldn't we use apr_time_now here?
> +
> + i2d_OCSP_RESPONSE(rsp, &p);
Is this correct? p is already char *. So &p would be char **.
If p gets changed by i2d_OCSP_RESPONSE our flag set above would get lost ???
> +
> + rv = mc->stapling_cache->store(mc->stapling_cache_context, s,
> + cinf->idx, sizeof(cinf->idx),
> + timeout, resp_der, resp_derlen, pool);
> + if (rv != APR_SUCCESS) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_cache_response: OCSP response session store error!");
> + return FALSE;
> + }
> +
> + return TRUE;
> +}
> +
> +static BOOL stapling_get_cached_response(server_rec *s, OCSP_RESPONSE **prsp,
> + BOOL *pok, certinfo *cinf,
> + apr_pool_t *pool)
> +{
> + SSLModConfigRec *mc = myModConfig(s);
> + apr_status_t rv;
> + OCSP_RESPONSE *rsp;
> + unsigned char resp_der[MAX_STAPLING_DER];
> + const unsigned char *p;
> + unsigned int resp_derlen = MAX_STAPLING_DER;
> +
> + rv = mc->stapling_cache->retrieve(mc->stapling_cache_context, s,
> + cinf->idx, sizeof(cinf->idx),
> + resp_der, &resp_derlen, pool);
> + if (rv != APR_SUCCESS) {
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_get_cached_response: cache miss");
> + return TRUE;
> + }
> + if (resp_derlen <= 1) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_get_cached_response: response length invalid??");
> + return TRUE;
> + }
> + p = resp_der;
> + if (pok) {
> + if (*p)
> + *pok = TRUE;
> + else
> + *pok = FALSE;
> + }
> + p++;
> + resp_derlen--;
> + rsp = d2i_OCSP_RESPONSE(NULL, &p, resp_derlen);
Why &p and not p?
> + if (!rsp) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_get_cached_response: response parse error??");
> + return TRUE;
> + }
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_get_cached_response: cache hit");
> +
> + *prsp = rsp;
> +
> + return TRUE;
> +}
> +
> +static BOOL stapling_renew_response(server_rec *s, modssl_ctx_t *mctx, SSL *ssl,
> + certinfo *cinf, OCSP_RESPONSE **prsp,
> + apr_pool_t *pool)
> +{
> + conn_rec *conn = (conn_rec *)SSL_get_app_data(ssl);
> + apr_pool_t *vpool;
> + OCSP_REQUEST *req = NULL;
> + OCSP_CERTID *id = NULL;
> + STACK_OF(X509_EXTENSION) *exts;
> + int i;
> + BOOL ok = FALSE;
> + BOOL rv = TRUE;
> + const char *ocspuri;
> + apr_uri_t uri;
> +
> + *prsp = NULL;
> + /* Build up OCSP query from server certificate info */
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_renew_response: querying responder");
> +
> + req = OCSP_REQUEST_new();
> + if (!req)
> + goto err;
> + id = OCSP_CERTID_dup(cinf->cid);
> + if (!id)
> + goto err;
> + if (!OCSP_request_add0_id(req, id))
> + goto err;
> + id = NULL;
> + /* Add any extensions to the request */
> + SSL_get_tlsext_status_exts(ssl, &exts);
> + for (i = 0; i < sk_X509_EXTENSION_num(exts); i++) {
> + X509_EXTENSION *ext = sk_X509_EXTENSION_value(exts, i);
> + if (!OCSP_REQUEST_add_ext(req, ext, -1))
> + goto err;
> + }
> +
> + if (mctx->stapling_force_url)
> + ocspuri = mctx->stapling_force_url;
> + else
> + ocspuri = cinf->uri;
> +
> + /* Create a temporary pool to constrain memory use */
> + apr_pool_create(&vpool, conn->pool);
> +
> + ok = apr_uri_parse(vpool, ocspuri, &uri);
> + if (ok != APR_SUCCESS) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_renew_response: Error parsing uri %s",
> + ocspuri);
> + rv = FALSE;
> + goto done;
I guess we are missing some indention here.
> + } else if (strcmp(uri.scheme, "http")) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_renew_response: Unsupported uri %s", ocspuri);
> + rv = FALSE;
> + goto done;
I guess we are missing some indention here.
> + }
> +
> + *prsp = modssl_dispatch_ocsp_request(&uri, mctx->stapling_responder_timeout,
> + req, conn, vpool);
> +
> + apr_pool_destroy(vpool);
> +
> + if (!*prsp) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_renew_response: responder error");
> + if (mctx->stapling_fake_trylater) {
> + *prsp = OCSP_response_create(OCSP_RESPONSE_STATUS_TRYLATER, NULL);
> + }
> + else {
> + goto done;
> + }
> + } else {
> + int response_status = OCSP_response_status(*prsp);
> +
> + if (response_status == OCSP_RESPONSE_STATUS_SUCCESSFUL) {
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_renew_response: query response received");
> + stapling_check_response(s, mctx, cinf, *prsp, &ok);
> + if (ok == FALSE) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_renew_response: error in retreived response!");
> + }
> + } else {
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_renew_response: responder error %s",
> + OCSP_response_status_str(response_status));
> + }
> + }
> + if (stapling_cache_response(s, mctx, *prsp, cinf, ok, pool) == FALSE) {
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_renew_response: error caching response!");
> + }
> +
> +done:
> + if (id)
> + OCSP_CERTID_free(id);
> + if (req)
> + OCSP_REQUEST_free(req);
> + return rv;
> +err:
> + rv = FALSE;
> + goto done;
Looks a little bit like spaghetti code :-).
> +}
> +
> +/* Certificate Status callback. This is called when a client includes a
> + * certificate status request extension.
> + *
> + * Check for cached responses in session cache. If valid send back to
> + * client. If absent or no longer valid query responder and update
> + * cache. */
> +static int stapling_cb(SSL *ssl, void *arg)
> +{
> + conn_rec *conn = (conn_rec *)SSL_get_app_data(ssl);
> + server_rec *s = conn->base_server;
Shouldn't we use mySrvFromConn(conn) here instead of conn->base_server?
> +
> + SSLSrvConfigRec *sc = mySrvConfig(s);
> + SSLConnRec *sslconn = myConnConfig(conn);
> + modssl_ctx_t *mctx = myCtxConfig(sslconn, sc);
> + certinfo *cinf = NULL;
> + OCSP_RESPONSE *rsp = NULL;
> + int rv;
> + BOOL ok;
> +
> + if (sc->server->stapling_enabled != TRUE) {
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_cb: OCSP Stapling disabled");
> + return SSL_TLSEXT_ERR_NOACK;
> + }
> +
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_cb: OCSP Stapling callback called");
> +
> + cinf = stapling_get_cert_info(s, mctx, ssl);
> + if (cinf == NULL) {
> + return SSL_TLSEXT_ERR_NOACK;
> + }
> +
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_cb: retrieved cached certificate data");
> +
> + /* Check to see if we already have a response for this certificate */
> + stapling_mutex_on(s);
> +
> + rv = stapling_get_cached_response(s, &rsp, &ok, cinf, conn->pool);
> + if (rv == FALSE) {
> + stapling_mutex_off(s);
> + return SSL_TLSEXT_ERR_ALERT_FATAL;
> + }
> +
> + if (rsp) {
> + /* see if response is acceptable */
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_cb: retrieved cached response");
> + rv = stapling_check_response(s, mctx, cinf, rsp, NULL);
> + if (rv == SSL_TLSEXT_ERR_ALERT_FATAL) {
> + OCSP_RESPONSE_free(rsp);
> + stapling_mutex_off(s);
> + return SSL_TLSEXT_ERR_ALERT_FATAL;
> + }
> + else if (rv == SSL_TLSEXT_ERR_NOACK) {
> + /* Error in response. If this error was not present when it was
> + * stored (i.e. response no longer valid) then it can be
> + * renewed straight away.
> + *
> + * If the error *was* present at the time it was stored then we
> + * don't renew the response straight away we just wait for the
> + * cached response to expire.
> + */
> + if (ok) {
> + OCSP_RESPONSE_free(rsp);
> + rsp = NULL;
> + }
> + else if (!mctx->stapling_return_errors) {
> + OCSP_RESPONSE_free(rsp);
> + stapling_mutex_off(s);
> + return SSL_TLSEXT_ERR_NOACK;
> + }
> + }
> + }
> +
> + if (rsp == NULL) {
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_cb: renewing cached response");
> + rv = stapling_renew_response(s, mctx, ssl, cinf, &rsp, conn->pool);
> +
> + if (rv == FALSE) {
> + stapling_mutex_off(s);
> + ap_log_error(APLOG_MARK, APLOG_ERR, 0, s,
> + "stapling_cb: fatal error");
> + return SSL_TLSEXT_ERR_ALERT_FATAL;
> + }
> + }
> + stapling_mutex_off(s);
> +
> + if (rsp) {
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_cb: setting response");
> + if (!stapling_set_response(ssl, rsp))
> + return SSL_TLSEXT_ERR_ALERT_FATAL;
> + return SSL_TLSEXT_ERR_OK;
> + }
> + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s,
> + "stapling_cb: no response available");
> +
> + return SSL_TLSEXT_ERR_NOACK;
> +
> +}
> +
Regards
RĂ¼diger
Re: svn commit: r829619 - in /httpd/httpd/trunk: ./ modules/ssl/
Posted by Joe Orton <jo...@redhat.com>.
On Mon, Oct 26, 2009 at 12:44:17AM +0100, Ruediger Pluem wrote:
> On 10/25/2009 06:21 PM, jorton@apache.org wrote:
> > Author: jorton
> > Date: Sun Oct 25 17:21:10 2009
> > New Revision: 829619
> >
> > URL: http://svn.apache.org/viewvc?rev=829619&view=rev
> > Log:
> > Add support for OCSP "stapling":
...
> Will there be documentation patches for the new directives?
Yes, I will try to get that done, I might not have time until after
ApacheCon though.
Thanks greatly for the review!
> > + timeout += time(NULL);
>
> Shouldn't we use apr_time_now here?
Yup, fixed in r830551.
> > +
> > + i2d_OCSP_RESPONSE(rsp, &p);
>
> Is this correct? p is already char *. So &p would be char **.
> If p gets changed by i2d_OCSP_RESPONSE our flag set above would get lost ???
This is correct. The preceding lines:
p = resp_der;
if (ok == TRUE) {
*p++ = 1;
timeout = mctx->stapling_cache_timeout;
}
else {
*p++ = 0;
timeout = mctx->stapling_errcache_timeout;
}
set the flag value in the first byte of the data block. The i2d_*
macros all take a char ** pointer, and move the pointer to the next byte
after the serialized data. (see e.g. the man page for i2d_X509)
> > + p++;
> > + resp_derlen--;
> > + rsp = d2i_OCSP_RESPONSE(NULL, &p, resp_derlen);
>
> Why &p and not p?
As above, the d2i_ macros also take a char ** pointer and move it to the
first byte after the encoded OCSP_RESPONSE.
...
> I guess we are missing some indention here.
Sorry about that - Guenter has fixed it all, thanks Guenter!
> > +done:
> > + if (id)
> > + OCSP_CERTID_free(id);
> > + if (req)
> > + OCSP_REQUEST_free(req);
> > + return rv;
> > +err:
> > + rv = FALSE;
> > + goto done;
>
> Looks a little bit like spaghetti code :-).
It's a bit annoying; it could be done using pool cleanups but that adds
significant complexity, or by in-lining the _free calls everywhere,
which is a bit error-pront. Not really sure either alternative is
better. C sucks ;)
> > + * Check for cached responses in session cache. If valid send back to
> > + * client. If absent or no longer valid query responder and update
> > + * cache. */
> > +static int stapling_cb(SSL *ssl, void *arg)
> > +{
> > + conn_rec *conn = (conn_rec *)SSL_get_app_data(ssl);
> > + server_rec *s = conn->base_server;
>
> Shouldn't we use mySrvFromConn(conn) here instead of conn->base_server?
Yup, fixed in r830546.
Thanks again!
Regards, Joe