Basic SOAP client, reworked transport streams.
[shibboleth/cpp-xmltooling.git] / xmltooling / soap / impl / CURLSOAPTransport.cpp
1 /*
2  *  Copyright 2001-2006 Internet2
3  * 
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *     http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16
17 /**
18  * CURLSOAPTransport.cpp
19  * 
20  * libcurl-based SOAPTransport implementation
21  */
22
23 #include "internal.h"
24 #include "exceptions.h"
25 #include "security/OpenSSLTrustEngine.h"
26 #include "signature/OpenSSLCredentialResolver.h"
27 #include "soap/HTTPSOAPTransport.h"
28 #include "soap/OpenSSLSOAPTransport.h"
29 #include "util/NDC.h"
30 #include "util/Threads.h"
31
32 #include <list>
33 #include <curl/curl.h>
34 #include <log4cpp/Category.hh>
35 #include <openssl/x509_vfy.h>
36
37 using namespace xmlsignature;
38 using namespace xmltooling;
39 using namespace log4cpp;
40 using namespace std;
41
42 namespace xmltooling {
43
44     // Manages cache of socket connections via CURL handles.
45     class XMLTOOL_DLLLOCAL CURLPool
46     {
47     public:
48         CURLPool() : m_size(256), m_lock(Mutex::create()),
49             m_log(Category::getInstance(XMLTOOLING_LOGCAT".SOAPTransport.CURLPool")) {}
50         ~CURLPool();
51         
52         CURL* get(const string& to, const char* endpoint);
53         void put(const string& to, const char* endpoint, CURL* handle);
54     
55     private:    
56         typedef map<string,vector<CURL*> > poolmap_t;
57         poolmap_t m_bindingMap;
58         list< vector<CURL*>* > m_pools;
59         long m_size;
60         Mutex* m_lock;
61         Category& m_log;
62     };
63     
64     static XMLTOOL_DLLLOCAL CURLPool* g_CURLPool = NULL;
65     
66     class XMLTOOL_DLLLOCAL CURLSOAPTransport : public HTTPSOAPTransport, public OpenSSLSOAPTransport
67     {
68     public:
69         CURLSOAPTransport(const KeyInfoSource& peer, const char* endpoint)
70                 : m_peer(peer), m_endpoint(endpoint), m_handle(NULL), m_headers(NULL),
71                     m_credResolver(NULL), m_trustEngine(NULL), m_keyResolver(NULL),
72                     m_ssl_callback(NULL), m_ssl_userptr(NULL) {
73             m_handle = g_CURLPool->get(peer.getName(), endpoint);
74             curl_easy_setopt(m_handle,CURLOPT_URL,endpoint);
75             curl_easy_setopt(m_handle,CURLOPT_CONNECTTIMEOUT,15);
76             curl_easy_setopt(m_handle,CURLOPT_TIMEOUT,30);
77             curl_easy_setopt(m_handle,CURLOPT_HTTPAUTH,0);
78             curl_easy_setopt(m_handle,CURLOPT_USERPWD,NULL);
79             curl_easy_setopt(m_handle,CURLOPT_HEADERDATA,this);
80             m_headers=curl_slist_append(m_headers,"Content-Type: text/xml");
81         }
82         
83         virtual ~CURLSOAPTransport() {
84             curl_slist_free_all(m_headers);
85             curl_easy_setopt(m_handle,CURLOPT_ERRORBUFFER,NULL);
86             g_CURLPool->put(m_peer.getName(), m_endpoint.c_str(), m_handle);
87         }
88
89         bool setConnectTimeout(long timeout) const {
90             return (curl_easy_setopt(m_handle,CURLOPT_CONNECTTIMEOUT,timeout)==CURLE_OK);
91         }
92         
93         bool setTimeout(long timeout) const {
94             return (curl_easy_setopt(m_handle,CURLOPT_TIMEOUT,timeout)==CURLE_OK);
95         }
96         
97         bool setAuth(transport_auth_t authType, const char* username=NULL, const char* password=NULL) const;
98         
99         bool setCredentialResolver(const CredentialResolver* credResolver) const {
100             const OpenSSLCredentialResolver* down = dynamic_cast<const OpenSSLCredentialResolver*>(credResolver);
101             if (!down) {
102                 m_credResolver = NULL;
103                 return (credResolver==NULL);
104             }
105             m_credResolver = down;
106             return true;
107         }
108         
109         bool setTrustEngine(const X509TrustEngine* trustEngine, const KeyResolver* keyResolver=NULL) const {
110             const OpenSSLTrustEngine* down = dynamic_cast<const OpenSSLTrustEngine*>(trustEngine);
111             if (!down) {
112                 m_trustEngine = NULL;
113                 m_keyResolver = NULL;
114                 return (trustEngine==NULL);
115             }
116             m_trustEngine = down;
117             m_keyResolver = keyResolver;
118             return true;
119         }
120         
121         void send(istream& in);
122         
123         istream& receive() {
124             return m_stream;
125         }
126         
127         string getContentType() const;
128         
129         bool setRequestHeader(const char* name, const char* val) const {
130             string temp(name);
131             temp=temp + ": " + val;
132             m_headers=curl_slist_append(m_headers,temp.c_str());
133             return true;
134         }
135         
136         const vector<string>& getResponseHeader(const char* val) const;
137         
138         bool setSSLCallback(ssl_ctx_callback_fn fn, void* userptr=NULL) const {
139             m_ssl_callback=fn;
140             m_ssl_userptr=userptr;
141             return true;
142         }
143
144     private:        
145         // per-call state
146         const KeyInfoSource& m_peer;
147         string m_endpoint;
148         CURL* m_handle;
149         stringstream m_stream;
150         mutable struct curl_slist* m_headers;
151         map<string,vector<string> > m_response_headers;
152         mutable const OpenSSLCredentialResolver* m_credResolver;
153         mutable const OpenSSLTrustEngine* m_trustEngine;
154         mutable const KeyResolver* m_keyResolver;
155         mutable ssl_ctx_callback_fn m_ssl_callback;
156         mutable void* m_ssl_userptr;
157         
158         friend size_t XMLTOOL_DLLLOCAL curl_header_hook(void* ptr, size_t size, size_t nmemb, void* stream);
159         friend CURLcode XMLTOOL_DLLLOCAL xml_ssl_ctx_callback(CURL* curl, SSL_CTX* ssl_ctx, void* userptr);
160         friend int XMLTOOL_DLLLOCAL verify_callback(X509_STORE_CTX* x509_ctx, void* arg);
161     };
162
163     // libcurl callback functions
164     size_t XMLTOOL_DLLLOCAL curl_header_hook(void* ptr, size_t size, size_t nmemb, void* stream);
165     size_t XMLTOOL_DLLLOCAL curl_write_hook(void* ptr, size_t size, size_t nmemb, void* stream);
166     size_t XMLTOOL_DLLLOCAL curl_read_hook( void *ptr, size_t size, size_t nmemb, void *stream);
167     int XMLTOOL_DLLLOCAL curl_debug_hook(CURL* handle, curl_infotype type, char* data, size_t len, void* ptr);
168     CURLcode XMLTOOL_DLLLOCAL xml_ssl_ctx_callback(CURL* curl, SSL_CTX* ssl_ctx, void* userptr);
169     int XMLTOOL_DLLLOCAL verify_callback(X509_STORE_CTX* x509_ctx, void* arg);
170
171     SOAPTransport* CURLSOAPTransportFactory(const pair<const KeyInfoSource*,const char*>& dest)
172     {
173         return new CURLSOAPTransport(*dest.first, dest.second);
174     }
175 };
176
177 void xmltooling::registerSOAPTransports()
178 {
179     XMLToolingConfig& conf=XMLToolingConfig::getConfig();
180     conf.SOAPTransportManager.registerFactory("http", CURLSOAPTransportFactory);
181     conf.SOAPTransportManager.registerFactory("https", CURLSOAPTransportFactory);
182 }
183
184 void xmltooling::initSOAPTransports()
185 {
186     g_CURLPool=new CURLPool();
187 }
188
189 void xmltooling::termSOAPTransports()
190 {
191     delete g_CURLPool;
192     g_CURLPool = NULL;
193 }
194
195 CURLPool::~CURLPool()
196 {
197     for (poolmap_t::iterator i=m_bindingMap.begin(); i!=m_bindingMap.end(); i++) {
198         for (vector<CURL*>::iterator j=i->second.begin(); j!=i->second.end(); j++)
199             curl_easy_cleanup(*j);
200     }
201     delete m_lock;
202 }
203
204 CURL* CURLPool::get(const string& to, const char* endpoint)
205 {
206 #ifdef _DEBUG
207     xmltooling::NDC("get");
208 #endif
209     m_log.debug("getting connection handle to %s", endpoint);
210     m_lock->lock();
211     poolmap_t::iterator i=m_bindingMap.find(to + "|" + endpoint);
212     
213     if (i!=m_bindingMap.end()) {
214         // Move this pool to the front of the list.
215         m_pools.remove(&(i->second));
216         m_pools.push_front(&(i->second));
217         
218         // If a free connection exists, return it.
219         if (!(i->second.empty())) {
220             CURL* handle=i->second.back();
221             i->second.pop_back();
222             m_size--;
223             m_lock->unlock();
224             m_log.debug("returning existing connection handle from pool");
225             return handle;
226         }
227     }
228     
229     m_lock->unlock();
230     m_log.debug("nothing free in pool, returning new connection handle");
231     
232     // Create a new connection and set non-varying options.
233     CURL* handle=curl_easy_init();
234     if (!handle)
235         return NULL;
236     curl_easy_setopt(handle,CURLOPT_NOPROGRESS,1);
237     curl_easy_setopt(handle,CURLOPT_NOSIGNAL,1);
238     curl_easy_setopt(handle,CURLOPT_FAILONERROR,1);
239     curl_easy_setopt(handle,CURLOPT_SSLVERSION,3);
240     curl_easy_setopt(handle,CURLOPT_SSL_VERIFYHOST,2);
241     curl_easy_setopt(handle,CURLOPT_HEADERFUNCTION,&curl_header_hook);
242     curl_easy_setopt(handle,CURLOPT_READFUNCTION,&curl_read_hook);
243     curl_easy_setopt(handle,CURLOPT_WRITEFUNCTION,&curl_write_hook);
244     curl_easy_setopt(handle,CURLOPT_DEBUGFUNCTION,&curl_debug_hook);
245
246     return handle;
247 }
248
249 void CURLPool::put(const string& to, const char* endpoint, CURL* handle)
250 {
251     string key = to + "|" + endpoint;
252     m_lock->lock();
253     poolmap_t::iterator i=m_bindingMap.find(key);
254     if (i==m_bindingMap.end())
255         m_pools.push_front(&(m_bindingMap.insert(poolmap_t::value_type(key,vector<CURL*>(1,handle))).first->second));
256     else
257         i->second.push_back(handle);
258     
259     CURL* killit=NULL;
260     if (++m_size > 256) {
261         // Kick a handle out from the back of the bus.
262         while (true) {
263             vector<CURL*>* corpse=m_pools.back();
264             if (!corpse->empty()) {
265                 killit=corpse->back();
266                 corpse->pop_back();
267                 m_size--;
268                 break;
269             }
270             
271             // Move an empty pool up to the front so we don't keep hitting it.
272             m_pools.pop_back();
273             m_pools.push_front(corpse);
274         }
275     }
276     m_lock->unlock();
277     if (killit) {
278         curl_easy_cleanup(killit);
279 #ifdef _DEBUG
280         xmltooling::NDC("put");
281 #endif
282         m_log.info("conn_pool_max limit reached, dropping an old connection");
283     }
284 }
285
286 bool CURLSOAPTransport::setAuth(transport_auth_t authType, const char* username, const char* password) const
287 {
288     if (authType==transport_auth_none) {
289         if (curl_easy_setopt(m_handle,CURLOPT_HTTPAUTH,0)!=CURLE_OK)
290             return false;
291         return (curl_easy_setopt(m_handle,CURLOPT_USERPWD,NULL)==CURLE_OK);
292     }
293     long flag=0;
294     switch (authType) {
295         case transport_auth_basic:    flag = CURLAUTH_BASIC; break;
296         case transport_auth_digest:   flag = CURLAUTH_DIGEST; break;
297         case transport_auth_ntlm:     flag = CURLAUTH_NTLM; break;
298         case transport_auth_gss:      flag = CURLAUTH_GSSNEGOTIATE; break;
299         default:            return false;
300     }
301     if (curl_easy_setopt(m_handle,CURLOPT_HTTPAUTH,flag)!=CURLE_OK)
302         return false;
303     string creds = string(username ? username : "") + ':' + (password ? password : "");
304     return (curl_easy_setopt(m_handle,CURLOPT_USERPWD,creds.c_str())==CURLE_OK);
305 }
306
307 const vector<string>& CURLSOAPTransport::getResponseHeader(const char* name) const
308 {
309     static vector<string> emptyVector;
310
311     map<string,vector<string> >::const_iterator i=m_response_headers.find(name);
312     if (i!=m_response_headers.end())
313         return i->second;
314     
315     for (map<string,vector<string> >::const_iterator j=m_response_headers.begin(); j!=m_response_headers.end(); j++) {
316 #ifdef HAVE_STRCASECMP
317         if (!strcasecmp(j->first.c_str(), name))
318 #else
319         if (!stricmp(j->first.c_str(), name))
320 #endif
321             return j->second;
322     }
323     
324     return emptyVector;
325 }
326
327 string CURLSOAPTransport::getContentType() const
328 {
329     char* content_type=NULL;
330     curl_easy_getinfo(m_handle,CURLINFO_CONTENT_TYPE,&content_type);
331     return content_type ? content_type : "";
332 }
333
334 void CURLSOAPTransport::send(istream& in)
335 {
336 #ifdef _DEBUG
337     xmltooling::NDC ndc("send");
338 #endif
339     Category& log=Category::getInstance(XMLTOOLING_LOGCAT".SOAPTransport");
340     Category& log_curl=Category::getInstance(XMLTOOLING_LOGCAT".libcurl");
341
342     // By this time, the handle has been prepared with the URL to use and the
343     // caller should have executed any set functions to manipulate it.
344
345     // Setup standard per-call curl properties.
346     curl_easy_setopt(m_handle,CURLOPT_POST,1);
347     curl_easy_setopt(m_handle,CURLOPT_READDATA,&in);
348     curl_easy_setopt(m_handle,CURLOPT_FILE,&m_stream);
349     curl_easy_setopt(m_handle,CURLOPT_DEBUGDATA,&log_curl);
350
351     char curl_errorbuf[CURL_ERROR_SIZE];
352     curl_errorbuf[0]=0;
353     curl_easy_setopt(m_handle,CURLOPT_ERRORBUFFER,curl_errorbuf);
354     if (log_curl.isDebugEnabled())
355         curl_easy_setopt(m_handle,CURLOPT_VERBOSE,1);
356
357     // Set request headers (possibly appended by hooks).
358     curl_easy_setopt(m_handle,CURLOPT_HTTPHEADER,m_headers);
359
360     if (m_ssl_callback || m_credResolver || m_trustEngine) {
361         curl_easy_setopt(m_handle,CURLOPT_SSL_CTX_FUNCTION,xml_ssl_ctx_callback);
362         curl_easy_setopt(m_handle,CURLOPT_SSL_CTX_DATA,this);
363     }
364     else {
365         curl_easy_setopt(m_handle,CURLOPT_SSL_CTX_FUNCTION,NULL);
366         curl_easy_setopt(m_handle,CURLOPT_SSL_CTX_DATA,NULL);
367     }
368     
369     // Verification of the peer is via TrustEngine only.
370     curl_easy_setopt(m_handle,CURLOPT_SSL_VERIFYPEER,0);
371
372     // Make the call.
373     log.info("sending SOAP message to %s", m_endpoint.c_str());
374     if (curl_easy_perform(m_handle) != CURLE_OK) {
375         log.error("failed communicating with SOAP endpoint: %s",
376             (curl_errorbuf[0] ? curl_errorbuf : "no further information available"));
377         throw IOException(
378             string("CURLSOAPTransport::send() failed while contacting SOAP responder: ") +
379                 (curl_errorbuf[0] ? curl_errorbuf : "no further information available"));
380     }
381 }
382
383 // callback to buffer headers from server
384 size_t xmltooling::curl_header_hook(void* ptr, size_t size, size_t nmemb, void* stream)
385 {
386     // only handle single-byte data
387     if (size!=1)
388         return 0;
389     CURLSOAPTransport* ctx = reinterpret_cast<CURLSOAPTransport*>(stream);
390     char* buf = (char*)malloc(nmemb + 1);
391     if (buf) {
392         memset(buf,0,nmemb + 1);
393         memcpy(buf,ptr,nmemb);
394         char* sep=(char*)strchr(buf,':');
395         if (sep) {
396             *(sep++)=0;
397             while (*sep==' ')
398                 *(sep++)=0;
399             char* white=buf+nmemb-1;
400             while (isspace(*white))
401                 *(white--)=0;
402             ctx->m_response_headers[buf].push_back(sep);
403         }
404         free(buf);
405         return nmemb;
406     }
407     return 0;
408 }
409
410 // callback to send data to server
411 size_t xmltooling::curl_read_hook(void* ptr, size_t size, size_t nmemb, void* stream)
412 {
413     // *stream is actually an istream object
414     istream& buf=*(reinterpret_cast<istream*>(stream));
415     buf.read(reinterpret_cast<char*>(ptr),size*nmemb);
416     return buf.gcount();
417 }
418
419 // callback to buffer data from server
420 size_t xmltooling::curl_write_hook(void* ptr, size_t size, size_t nmemb, void* stream)
421 {
422     size_t len = size*nmemb;
423     reinterpret_cast<stringstream*>(stream)->write(reinterpret_cast<const char*>(ptr),len);
424     return len;
425 }
426
427 // callback for curl debug data
428 int xmltooling::curl_debug_hook(CURL* handle, curl_infotype type, char* data, size_t len, void* ptr)
429 {
430     // *ptr is actually a logging object
431     if (!ptr) return 0;
432     CategoryStream log=reinterpret_cast<Category*>(ptr)->debugStream();
433     for (char* ch=data; len && (isprint(*ch) || isspace(*ch)); len--)
434         log << *ch++;
435     log << CategoryStream::ENDLINE;
436     return 0;
437 }
438
439 int xmltooling::verify_callback(X509_STORE_CTX* x509_ctx, void* arg)
440 {
441     Category::getInstance("OpenSSL").debug("invoking X509 verify callback");
442 #if (OPENSSL_VERSION_NUMBER >= 0x00907000L)
443     CURLSOAPTransport* ctx = reinterpret_cast<CURLSOAPTransport*>(arg);
444 #else
445     // Yes, this sucks. I'd use TLS, but there's no really obvious spot to put the thread key
446     // and global variables suck too. We can't access the X509_STORE_CTX depth directly because
447     // OpenSSL only copies it into the context if it's >=0, and the unsigned pointer may be
448     // negative in the SSL structure's int member.
449     CURLSOAPTransport* ctx = reinterpret_cast<CURLSOAPTransport*>(
450         SSL_get_verify_depth(
451             reinterpret_cast<SSL*>(X509_STORE_CTX_get_ex_data(x509_ctx,SSL_get_ex_data_X509_STORE_CTX_idx()))
452             )
453         );
454 #endif
455
456      // Bypass name check (handled for us by curl).
457     if (!ctx->m_trustEngine->validate(x509_ctx->cert,x509_ctx->untrusted,ctx->m_peer,false,ctx->m_keyResolver)) {
458         x509_ctx->error=X509_V_ERR_APPLICATION_VERIFICATION;     // generic error, check log for plugin specifics
459         return 0;
460     }
461     
462     // Signal success. Hopefully it doesn't matter what's actually in the structure now.
463     return 1;
464 }
465
466 // callback to invoke a caller-defined SSL callback
467 CURLcode xmltooling::xml_ssl_ctx_callback(CURL* curl, SSL_CTX* ssl_ctx, void* userptr)
468 {
469     CURLSOAPTransport* conf = reinterpret_cast<CURLSOAPTransport*>(userptr);
470     if (conf->m_credResolver)
471         conf->m_credResolver->attach(ssl_ctx);
472
473     if (conf->m_trustEngine) {
474         SSL_CTX_set_verify(ssl_ctx,SSL_VERIFY_PEER,NULL);
475 #if (OPENSSL_VERSION_NUMBER >= 0x00907000L)
476         // With 0.9.7, we can pass a callback argument directly.
477         SSL_CTX_set_cert_verify_callback(ssl_ctx,verify_callback,userptr);
478 #else
479         // With 0.9.6, there's no argument, so we're going to use a really embarrassing hack and
480         // stuff the argument in the depth property where it will get copied to the context object
481         // that's handed to the callback.
482         SSL_CTX_set_cert_verify_callback(ssl_ctx,reinterpret_cast<int (*)()>(verify_callback),NULL);
483         SSL_CTX_set_verify_depth(ssl_ctx,reinterpret_cast<int>(userptr));
484 #endif
485     }
486         
487     if (!conf->m_ssl_callback(ssl_ctx,conf->m_ssl_userptr))
488         return CURLE_SSL_CERTPROBLEM;
489         
490     return CURLE_OK;
491 }