Context Navigation

source: lib/http_client.c @ 8bd866f

Last change on this file since 8bd866f was 8bd866f, checked in by Wilmer van der Gaast <wilmer@…>, at 2012-11-10T22:25:58Z
Reworked http_client a little bit to support streaming besides just buffering a complete response before giving it back to the caller.
Property mode set to `100644`
File size: 14.5 KB

Line
1	/********************************************************************\
2	* BitlBee -- An IRC to other IM-networks gateway *
3	* *
4	* Copyright 2002-2011 Wilmer van der Gaast and others *
5	\********************************************************************/
6
7	/* HTTP(S) module */
8
9	/*
10	This program is free software; you can redistribute it and/or modify
11	it under the terms of the GNU General Public License as published by
12	the Free Software Foundation; either version 2 of the License, or
13	(at your option) any later version.
14
15	This program is distributed in the hope that it will be useful,
16	but WITHOUT ANY WARRANTY; without even the implied warranty of
17	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18	GNU General Public License for more details.
19
20	You should have received a copy of the GNU General Public License with
21	the Debian GNU/Linux distribution in /usr/share/common-licenses/GPL;
22	if not, write to the Free Software Foundation, Inc., 59 Temple Place,
23	Suite 330, Boston, MA 02111-1307 USA
24	*/
25
26	#include <string.h>
27	#include <stdio.h>
28
29	#include "http_client.h"
30	#include "url.h"
31	#include "sock.h"
32
33
34	static gboolean http_connected( gpointer data, int source, b_input_condition cond );
35	static gboolean http_ssl_connected( gpointer data, int returncode, void *source, b_input_condition cond );
36	static gboolean http_incoming_data( gpointer data, int source, b_input_condition cond );
37	static void http_free( struct http_request *req );
38
39
40	struct http_request http_dorequest( char host, int port, int ssl, char *request, http_input_function func, gpointer data )
41	{
42	struct http_request *req;
43	int error = 0;
44
45	req = g_new0( struct http_request, 1 );
46
47	if( ssl )
48	{
49	req->ssl = ssl_connect( host, port, TRUE, http_ssl_connected, req );
50	if( req->ssl == NULL )
51	error = 1;
52	}
53	else
54	{
55	req->fd = proxy_connect( host, port, http_connected, req );
56	if( req->fd < 0 )
57	error = 1;
58	}
59
60	if( error )
61	{
62	http_free( req );
63	return NULL;
64	}
65
66	req->func = func;
67	req->data = data;
68	req->request = g_strdup( request );
69	req->request_length = strlen( request );
70	req->redir_ttl = 3;
71
72	if( getenv( "BITLBEE_DEBUG" ) )
73	printf( "About to send HTTP request:\n%s\n", req->request );
74
75	return req;
76	}
77
78	struct http_request http_dorequest_url( char url_string, http_input_function func, gpointer data )
79	{
80	url_t *url = g_new0( url_t, 1 );
81	char *request;
82	void *ret;
83
84	if( !url_set( url, url_string ) )
85	{
86	g_free( url );
87	return NULL;
88	}
89
90	if( url->proto != PROTO_HTTP && url->proto != PROTO_HTTPS )
91	{
92	g_free( url );
93	return NULL;
94	}
95
96	request = g_strdup_printf( "GET %s HTTP/1.0\r\n"
97	"Host: %s\r\n"
98	"Connection: close\r\n"
99	"User-Agent: BitlBee " BITLBEE_VERSION " " ARCH "/" CPU "\r\n"
100	"\r\n", url->file, url->host );
101
102	ret = http_dorequest( url->host, url->port,
103	url->proto == PROTO_HTTPS, request, func, data );
104
105	g_free( url );
106	g_free( request );
107	return ret;
108	}
109
110	/* This one is actually pretty simple... Might get more calls if we can't write
111	the whole request at once. */
112	static gboolean http_connected( gpointer data, int source, b_input_condition cond )
113	{
114	struct http_request *req = data;
115	int st;
116
117	if( source < 0 )
118	goto error;
119
120	if( req->inpa > 0 )
121	b_event_remove( req->inpa );
122
123	sock_make_nonblocking( req->fd );
124
125	if( req->ssl )
126	{
127	st = ssl_write( req->ssl, req->request + req->bytes_written,
128	req->request_length - req->bytes_written );
129	if( st < 0 )
130	{
131	if( ssl_errno != SSL_AGAIN )
132	{
133	ssl_disconnect( req->ssl );
134	goto error;
135	}
136	}
137	}
138	else
139	{
140	st = write( source, req->request + req->bytes_written,
141	req->request_length - req->bytes_written );
142	if( st < 0 )
143	{
144	if( !sockerr_again() )
145	{
146	closesocket( req->fd );
147	goto error;
148	}
149	}
150	}
151
152	if( st > 0 )
153	req->bytes_written += st;
154
155	if( req->bytes_written < req->request_length )
156	req->inpa = b_input_add( source,
157	req->ssl ? ssl_getdirection( req->ssl ) : B_EV_IO_WRITE,
158	http_connected, req );
159	else
160	req->inpa = b_input_add( source, B_EV_IO_READ, http_incoming_data, req );
161
162	return FALSE;
163
164	error:
165	if( req->status_string == NULL )
166	req->status_string = g_strdup( "Error while writing HTTP request" );
167
168	req->func( req );
169	http_free( req );
170	return FALSE;
171	}
172
173	static gboolean http_ssl_connected( gpointer data, int returncode, void *source, b_input_condition cond )
174	{
175	struct http_request *req = data;
176
177	if( source == NULL )
178	{
179	if( returncode != 0 )
180	{
181	char *err = ssl_verify_strerror( returncode );
182	req->status_string = g_strdup_printf(
183	"Certificate verification problem 0x%x: %s",
184	returncode, err ? err : "Unknown" );
185	g_free( err );
186	}
187	return http_connected( data, -1, cond );
188	}
189
190	req->fd = ssl_getfd( source );
191
192	return http_connected( data, req->fd, cond );
193	}
194
195	static gboolean http_handle_headers( struct http_request *req );
196
197	static gboolean http_incoming_data( gpointer data, int source, b_input_condition cond )
198	{
199	struct http_request *req = data;
200	char buffer[2048];
201	char *s;
202	size_t content_length;
203	int st;
204
205	if( req->inpa > 0 )
206	b_event_remove( req->inpa );
207
208	if( req->ssl )
209	{
210	st = ssl_read( req->ssl, buffer, sizeof( buffer ) );
211	if( st < 0 )
212	{
213	if( ssl_errno != SSL_AGAIN )
214	{
215	/* goto cleanup; */
216
217	/* YAY! We have to deal with crappy Microsoft
218	servers that LOVE to send invalid TLS
219	packets that abort connections! \o/ */
220
221	goto eof;
222	}
223	}
224	else if( st == 0 )
225	{
226	goto eof;
227	}
228	}
229	else
230	{
231	st = read( req->fd, buffer, sizeof( buffer ) );
232	if( st < 0 )
233	{
234	if( !sockerr_again() )
235	{
236	req->status_string = g_strdup( strerror( errno ) );
237	goto cleanup;
238	}
239	}
240	else if( st == 0 )
241	{
242	goto eof;
243	}
244	}
245
246	if( st > 0 && !req->sbuf )
247	{
248	req->reply_headers = g_realloc( req->reply_headers, req->bytes_read + st + 1 );
249	memcpy( req->reply_headers + req->bytes_read, buffer, st );
250	req->bytes_read += st;
251
252	st = 0;
253	}
254
255	if( st >= 0 && ( req->flags & HTTPC_STREAMING ) )
256	{
257	if( !req->reply_body &&
258	( strstr( req->reply_headers, "\r\n\r\n" ) \|\|
259	strstr( req->reply_headers, "\n\n" ) ) )
260	{
261	size_t hlen;
262
263	/* We've now received all headers, so process them once
264	before we start feeding back data. */
265	if( !http_handle_headers( req ) )
266	return FALSE;
267
268	hlen = req->reply_body - req->reply_headers;
269
270	req->sblen = req->bytes_read - hlen;
271	req->sbuf = g_memdup( req->reply_body, req->sblen + 1 );
272	req->reply_headers = g_realloc( req->reply_headers, hlen + 1 );
273
274	req->reply_body = req->sbuf;
275	}
276
277	if( st > 0 )
278	{
279	int pos = req->reply_body - req->sbuf;
280	req->sbuf = g_realloc( req->sbuf, req->sblen + st + 1 );
281	memcpy( req->sbuf + req->sblen, buffer, st );
282	req->bytes_read += st;
283	req->sblen += st;
284	req->reply_body = req->sbuf + pos;
285	req->body_size = req->sblen - pos;
286	}
287
288	if( req->reply_body )
289	req->func( req );
290	}
291
292	if( ssl_pending( req->ssl ) )
293	return http_incoming_data( data, source, cond );
294
295	/* There will be more! */
296	req->inpa = b_input_add( req->fd,
297	req->ssl ? ssl_getdirection( req->ssl ) : B_EV_IO_READ,
298	http_incoming_data, req );
299
300	return FALSE;
301
302	eof:
303	/* Maybe if the webserver is overloaded, or when there's bad SSL
304	support... */
305	if( req->bytes_read == 0 )
306	{
307	req->status_string = g_strdup( "Empty HTTP reply" );
308	goto cleanup;
309	}
310
311	if( !( req->flags & HTTPC_STREAMING ) )
312	{
313	/* Returns FALSE if we were redirected, in which case we should abort
314	and not run any callback yet. */
315	if( !http_handle_headers( req ) )
316	return FALSE;
317	}
318
319	cleanup:
320	if( req->ssl )
321	ssl_disconnect( req->ssl );
322	else
323	closesocket( req->fd );
324
325	if( ( s = get_rfc822_header( req->reply_headers, "Content-Length", 0 ) ) &&
326	sscanf( s, "%zd", &content_length ) == 1 )
327	{
328	if( content_length < req->body_size )
329	{
330	req->status_code = -1;
331	g_free( req->status_string );
332	req->status_string = g_strdup( "Response truncated" );
333	}
334	}
335	g_free( s );
336
337	if( getenv( "BITLBEE_DEBUG" ) && req )
338	printf( "Finishing HTTP request with status: %s\n",
339	req->status_string ? req->status_string : "NULL" );
340
341	req->func( req );
342	http_free( req );
343	return FALSE;
344	}
345
346	/* Splits headers and body. Checks result code, in case of 300s it'll handle
347	redirects. If this returns FALSE, don't call any callbacks! */
348	static gboolean http_handle_headers( struct http_request *req )
349	{
350	char end1, end2;
351	int evil_server = 0;
352
353	/* Zero termination is very convenient. */
354	req->reply_headers[req->bytes_read] = '\0';
355
356	/* Find the separation between headers and body, and keep stupid
357	webservers in mind. */
358	end1 = strstr( req->reply_headers, "\r\n\r\n" );
359	end2 = strstr( req->reply_headers, "\n\n" );
360
361	if( end2 && end2 < end1 )
362	{
363	end1 = end2 + 1;
364	evil_server = 1;
365	}
366	else if( end1 )
367	{
368	end1 += 2;
369	}
370	else
371	{
372	req->status_string = g_strdup( "Malformed HTTP reply" );
373	return TRUE;
374	}
375
376	*end1 = 0;
377
378	if( getenv( "BITLBEE_DEBUG" ) )
379	printf( "HTTP response headers:\n%s\n", req->reply_headers );
380
381	if( evil_server )
382	req->reply_body = end1 + 1;
383	else
384	req->reply_body = end1 + 2;
385
386	req->body_size = req->reply_headers + req->bytes_read - req->reply_body;
387
388	if( ( end1 = strchr( req->reply_headers, ' ' ) ) != NULL )
389	{
390	if( sscanf( end1 + 1, "%hd", &req->status_code ) != 1 )
391	{
392	req->status_string = g_strdup( "Can't parse status code" );
393	req->status_code = -1;
394	}
395	else
396	{
397	char *eol;
398
399	if( evil_server )
400	eol = strchr( end1, '\n' );
401	else
402	eol = strchr( end1, '\r' );
403
404	req->status_string = g_strndup( end1 + 1, eol - end1 - 1 );
405
406	/* Just to be sure... */
407	if( ( eol = strchr( req->status_string, '\r' ) ) )
408	*eol = 0;
409	if( ( eol = strchr( req->status_string, '\n' ) ) )
410	*eol = 0;
411	}
412	}
413	else
414	{
415	req->status_string = g_strdup( "Can't locate status code" );
416	req->status_code = -1;
417	}
418
419	if( ( ( req->status_code >= 301 && req->status_code <= 303 ) \|\|
420	req->status_code == 307 ) && req->redir_ttl-- > 0 )
421	{
422	char loc, new_request, *new_host;
423	int error = 0, new_port, new_proto;
424
425	/* We might fill it again, so let's not leak any memory. */
426	g_free( req->status_string );
427	req->status_string = NULL;
428
429	loc = strstr( req->reply_headers, "\nLocation: " );
430	if( loc == NULL ) /* We can't handle this redirect... */
431	{
432	req->status_string = g_strdup( "Can't locate Location: header" );
433	return TRUE;
434	}
435
436	loc += 11;
437	while( *loc == ' ' )
438	loc ++;
439
440	/* TODO/FIXME: Possibly have to handle relative redirections,
441	and rewrite Host: headers. Not necessary for now, it's
442	enough for passport authentication like this. */
443
444	if( *loc == '/' )
445	{
446	/* Just a different pathname... */
447
448	/* Since we don't cache the servername, and since we
449	don't need this yet anyway, I won't implement it. */
450
451	req->status_string = g_strdup( "Can't handle recursive redirects" );
452
453	return TRUE;
454	}
455	else
456	{
457	/* A whole URL */
458	url_t *url;
459	char *s;
460	const char *new_method;
461
462	s = strstr( loc, "\r\n" );
463	if( s == NULL )
464	return TRUE;
465
466	url = g_new0( url_t, 1 );
467	*s = 0;
468
469	if( !url_set( url, loc ) )
470	{
471	req->status_string = g_strdup( "Malformed redirect URL" );
472	g_free( url );
473	return TRUE;
474	}
475
476	/* Find all headers and, if necessary, the POST request contents.
477	Skip the old Host: header though. This crappy code here means
478	anything using this http_client MUST put the Host: header at
479	the top. */
480	if( !( ( s = strstr( req->request, "\r\nHost: " ) ) &&
481	( s = strstr( s + strlen( "\r\nHost: " ), "\r\n" ) ) ) )
482	{
483	req->status_string = g_strdup( "Error while rebuilding request string" );
484	g_free( url );
485	return TRUE;
486	}
487
488	/* More or less HTTP/1.0 compliant, from my reading of RFC 2616.
489	Always perform a GET request unless we received a 301. 303 was
490	meant for this but it's HTTP/1.1-only and we're specifically
491	speaking HTTP/1.0. ...
492
493	Well except someone at identi.ca's didn't bother reading any
494	RFCs and just return HTTP/1.1-specific status codes to HTTP/1.0
495	requests. Fuckers. So here we are, handle 301..303,307. */
496	if( strncmp( req->request, "GET", 3 ) == 0 )
497	/* GETs never become POSTs. */
498	new_method = "GET";
499	else if( req->status_code == 302 \|\| req->status_code == 303 )
500	/* 302 de-facto becomes GET, 303 as specified by RFC 2616#10.3.3 */
501	new_method = "GET";
502	else
503	/* 301 de-facto should stay POST, 307 specifally RFC 2616#10.3.8 */
504	new_method = "POST";
505
506	/* Okay, this isn't fun! We have to rebuild the request... :-( */
507	new_request = g_strdup_printf( "%s %s HTTP/1.0\r\nHost: %s%s",
508	new_method, url->file, url->host, s );
509
510	new_host = g_strdup( url->host );
511	new_port = url->port;
512	new_proto = url->proto;
513
514	/* If we went from POST to GET, truncate the request content. */
515	if( new_request[0] != req->request[0] && new_request[0] == 'G' &&
516	( s = strstr( new_request, "\r\n\r\n" ) ) )
517	s[4] = '\0';
518
519	g_free( url );
520	}
521
522	if( req->ssl )
523	ssl_disconnect( req->ssl );
524	else
525	closesocket( req->fd );
526
527	req->fd = -1;
528	req->ssl = NULL;
529
530	if( getenv( "BITLBEE_DEBUG" ) )
531	printf( "New headers for redirected HTTP request:\n%s\n", new_request );
532
533	if( new_proto == PROTO_HTTPS )
534	{
535	req->ssl = ssl_connect( new_host, new_port, TRUE, http_ssl_connected, req );
536	if( req->ssl == NULL )
537	error = 1;
538	}
539	else
540	{
541	req->fd = proxy_connect( new_host, new_port, http_connected, req );
542	if( req->fd < 0 )
543	error = 1;
544	}
545	g_free( new_host );
546
547	if( error )
548	{
549	req->status_string = g_strdup( "Connection problem during redirect" );
550	g_free( new_request );
551	return TRUE;
552	}
553
554	g_free( req->request );
555	g_free( req->reply_headers );
556	req->request = new_request;
557	req->request_length = strlen( new_request );
558	req->bytes_read = req->bytes_written = req->inpa = 0;
559	req->reply_headers = req->reply_body = NULL;
560
561	return FALSE;
562	}
563
564	return TRUE;
565	}
566
567	void http_flush_bytes( struct http_request *req, size_t len )
568	{
569	if( len > 0 && len <= req->body_size )
570	{
571	req->reply_body += len;
572	req->body_size -= len;
573	}
574	}
575
576	static void http_free( struct http_request *req )
577	{
578	g_free( req->request );
579	g_free( req->reply_headers );
580	g_free( req->status_string );
581	g_free( req->sbuf );
582	g_free( req );
583	}

Note: See TracBrowser for help on using the repository browser.

Download in other formats: