Add a client-body-tagger action
[privoxy.git] / jcc.c
1 /*********************************************************************
2  *
3  * File        :  $Source: /cvsroot/ijbswa/current/jcc.c,v $
4  *
5  * Purpose     :  Main file.  Contains main() method, main loop, and
6  *                the main connection-handling function.
7  *
8  * Copyright   :  Written by and Copyright (C) 2001-2022 the
9  *                Privoxy team. https://www.privoxy.org/
10  *
11  *                Based on the Internet Junkbuster originally written
12  *                by and Copyright (C) 1997 Anonymous Coders and
13  *                Junkbusters Corporation.  http://www.junkbusters.com
14  *
15  *                This program is free software; you can redistribute it
16  *                and/or modify it under the terms of the GNU General
17  *                Public License as published by the Free Software
18  *                Foundation; either version 2 of the License, or (at
19  *                your option) any later version.
20  *
21  *                This program is distributed in the hope that it will
22  *                be useful, but WITHOUT ANY WARRANTY; without even the
23  *                implied warranty of MERCHANTABILITY or FITNESS FOR A
24  *                PARTICULAR PURPOSE.  See the GNU General Public
25  *                License for more details.
26  *
27  *                The GNU General Public License should be included with
28  *                this file.  If not, you can view it at
29  *                http://www.gnu.org/copyleft/gpl.html
30  *                or write to the Free Software Foundation, Inc., 59
31  *                Temple Place - Suite 330, Boston, MA  02111-1307, USA.
32  *
33  *********************************************************************/
34
35
36 #include "config.h"
37
38 #include <stdio.h>
39 #include <sys/types.h>
40 #include <stdlib.h>
41 #include <string.h>
42 #include <signal.h>
43 #include <fcntl.h>
44 #include <errno.h>
45 #include <assert.h>
46
47 #ifdef _WIN32
48 # ifndef FEATURE_PTHREAD
49 #  ifndef STRICT
50 #   define STRICT
51 #  endif
52 #  include <winsock2.h>
53 #  include <windows.h>
54 #  include <process.h>
55 # endif /* ndef FEATURE_PTHREAD */
56
57 # include "win32.h"
58 # ifndef _WIN_CONSOLE
59 #  include "w32log.h"
60 # endif /* ndef _WIN_CONSOLE */
61 # include "w32svrapi.h"
62
63 #else /* ifndef _WIN32 */
64
65 # include <unistd.h>
66 # include <sys/wait.h>
67 # include <sys/time.h>
68 # include <sys/stat.h>
69 # include <sys/ioctl.h>
70
71 #ifdef sun
72 #include <sys/termios.h>
73 #endif /* sun */
74
75 #ifdef unix
76 #include <pwd.h>
77 #include <grp.h>
78 #endif
79
80 # include <signal.h>
81
82 # ifdef __BEOS__
83 #  include <socket.h>  /* BeOS has select() for sockets only. */
84 #  include <OS.h>      /* declarations for threads and stuff. */
85 # endif
86
87 #ifdef HAVE_POLL
88 #ifdef __GLIBC__
89 #include <sys/poll.h>
90 #else
91 #include <poll.h>
92 #endif /* def __GLIBC__ */
93 #else
94 # ifndef FD_ZERO
95 #  include <select.h>
96 # endif
97 #warning poll() appears to be unavailable. Your platform will become unsupported in the future.
98 #endif /* HAVE_POLL */
99
100 #endif
101
102 #include "project.h"
103 #include "list.h"
104 #include "jcc.h"
105 #ifdef FEATURE_HTTPS_INSPECTION
106 #include "ssl.h"
107 #endif
108 #include "filters.h"
109 #include "loaders.h"
110 #include "parsers.h"
111 #include "miscutil.h"
112 #include "errlog.h"
113 #include "jbsockets.h"
114 #include "gateway.h"
115 #include "actions.h"
116 #include "cgi.h"
117 #include "loadcfg.h"
118 #include "urlmatch.h"
119 #ifdef FEATURE_CLIENT_TAGS
120 #include "client-tags.h"
121 #endif
122
123 int daemon_mode = 1;
124 struct client_states clients[1];
125 struct file_list     files[1];
126
127 #ifdef FEATURE_STATISTICS
128 int urls_read     = 0;     /* total nr of urls read inc rejected */
129 int urls_rejected = 0;     /* total nr of urls rejected */
130 #ifdef MUTEX_LOCKS_AVAILABLE
131 unsigned long long number_of_requests_received = 0;
132 unsigned long long number_of_requests_blocked = 0;
133 #endif
134 #endif /* def FEATURE_STATISTICS */
135
136 #ifdef FEATURE_GRACEFUL_TERMINATION
137 int g_terminate = 0;
138 #endif
139
140 #if !defined(_WIN32)
141 static void sig_handler(int the_signal);
142 #endif
143 static int client_protocol_is_unsupported(struct client_state *csp, char *req);
144 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
145 static jb_err get_server_headers(struct client_state *csp);
146 static const char *crunch_reason(const struct http_response *rsp);
147 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
148 static char *get_request_line(struct client_state *csp);
149 static jb_err receive_client_request(struct client_state *csp);
150 static jb_err parse_client_request(struct client_state *csp);
151 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
152 static jb_err change_request_destination(struct client_state *csp);
153 static void handle_established_connection(struct client_state *csp);
154 static void chat(struct client_state *csp);
155 static void serve(struct client_state *csp);
156 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
157 static void usage(const char *myname);
158 #endif
159 static void initialize_mutexes(void);
160 static jb_socket bind_port_helper(const char *haddr, int hport, int backlog);
161 static void bind_ports_helper(struct configuration_spec *config, jb_socket sockets[]);
162 static void close_ports_helper(jb_socket sockets[]);
163 static void listen_loop(void);
164 static void serve(struct client_state *csp);
165
166 #ifdef __BEOS__
167 static int32 server_thread(void *data);
168 #endif /* def __BEOS__ */
169
170 #ifdef _WIN32
171 #define sleep(N)  Sleep(((N) * 1000))
172 #endif
173
174 #ifdef FUZZ
175 int process_fuzzed_input(char *fuzz_input_type, char *fuzz_input_file);
176 void show_fuzz_usage(const char *name);
177 #endif
178
179 #ifdef MUTEX_LOCKS_AVAILABLE
180 /*
181  * XXX: Does the locking stuff really belong in this file?
182  */
183 privoxy_mutex_t log_mutex;
184 privoxy_mutex_t log_init_mutex;
185 privoxy_mutex_t connection_reuse_mutex;
186
187 #ifdef FEATURE_HTTPS_INSPECTION
188 privoxy_mutex_t certificate_mutex;
189 privoxy_mutex_t ssl_init_mutex;
190 #endif
191
192 #ifdef FEATURE_EXTERNAL_FILTERS
193 privoxy_mutex_t external_filter_mutex;
194 #endif
195 #ifdef FEATURE_CLIENT_TAGS
196 privoxy_mutex_t client_tags_mutex;
197 #endif
198 #ifdef FEATURE_STATISTICS
199 privoxy_mutex_t block_statistics_mutex;
200 #endif
201 #ifdef FEATURE_EXTENDED_STATISTICS
202 privoxy_mutex_t filter_statistics_mutex;
203 privoxy_mutex_t block_reason_statistics_mutex;
204 #endif
205
206 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
207 privoxy_mutex_t resolver_mutex;
208 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
209
210 #ifndef HAVE_GMTIME_R
211 privoxy_mutex_t gmtime_mutex;
212 #endif /* ndef HAVE_GMTIME_R */
213
214 #ifndef HAVE_LOCALTIME_R
215 privoxy_mutex_t localtime_mutex;
216 #endif /* ndef HAVE_GMTIME_R */
217
218 #if !defined(HAVE_ARC4RANDOM) && !defined(HAVE_RANDOM)
219 privoxy_mutex_t rand_mutex;
220 #endif /* !defined(HAVE_ARC4RANDOM) && !defined(HAVE_RANDOM) */
221
222 #endif /* def MUTEX_LOCKS_AVAILABLE */
223
224 #if defined(unix)
225 const char *basedir = NULL;
226 const char *pidfile = NULL;
227 static int received_hup_signal = 0;
228 #endif /* defined unix */
229
230 /* HTTP snipplets. */
231 static const char CSUCCEED[] =
232    "HTTP/1.1 200 Connection established\r\n\r\n";
233
234 static const char CHEADER[] =
235    "HTTP/1.1 400 Invalid header received from client\r\n"
236    "Content-Type: text/plain\r\n"
237    "Connection: close\r\n\r\n"
238    "Invalid header received from client.\n";
239
240 static const char FTP_RESPONSE[] =
241    "HTTP/1.1 400 Invalid request received from client\r\n"
242    "Content-Type: text/plain\r\n"
243    "Connection: close\r\n\r\n"
244    "Invalid request. Privoxy doesn't support FTP.\n";
245
246 static const char GOPHER_RESPONSE[] =
247    "HTTP/1.1 400 Invalid request received from client\r\n"
248    "Content-Type: text/plain\r\n"
249    "Connection: close\r\n\r\n"
250    "Invalid request. Privoxy doesn't support gopher.\n";
251
252 /* XXX: should be a template */
253 static const char MISSING_DESTINATION_RESPONSE[] =
254    "HTTP/1.1 400 Bad request received from client\r\n"
255    "Content-Type: text/plain\r\n"
256    "Connection: close\r\n\r\n"
257    "Bad request. Privoxy was unable to extract the destination.\n";
258
259 /* XXX: should be a template */
260 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
261    "HTTP/1.1 502 Server or forwarder response invalid\r\n"
262    "Content-Type: text/plain\r\n"
263    "Connection: close\r\n\r\n"
264    "Bad response. The server or forwarder response doesn't look like HTTP.\n";
265
266 /* XXX: should be a template */
267 static const char MESSED_UP_REQUEST_RESPONSE[] =
268    "HTTP/1.1 400 Malformed request after rewriting\r\n"
269    "Content-Type: text/plain\r\n"
270    "Connection: close\r\n\r\n"
271    "Bad request. Messed up with header filters.\n";
272
273 static const char TOO_MANY_CONNECTIONS_RESPONSE[] =
274    "HTTP/1.1 503 Too many open connections\r\n"
275    "Content-Type: text/plain\r\n"
276    "Connection: close\r\n\r\n"
277    "Maximum number of open connections reached.\n";
278
279 static const char CLIENT_CONNECTION_TIMEOUT_RESPONSE[] =
280    "HTTP/1.1 504 Connection timeout\r\n"
281    "Content-Type: text/plain\r\n"
282    "Connection: close\r\n\r\n"
283    "The connection timed out because the client request didn't arrive in time.\n";
284
285 static const char CLIENT_BODY_PARSE_ERROR_RESPONSE[] =
286    "HTTP/1.1 400 Failed reading client body\r\n"
287    "Content-Type: text/plain\r\n"
288    "Connection: close\r\n\r\n"
289    "Failed parsing or buffering the chunk-encoded client body.\n";
290
291 static const char CLIENT_BODY_BUFFER_ERROR_RESPONSE[] =
292    "HTTP/1.1 400 Failed reading client body\r\n"
293    "Content-Type: text/plain\r\n"
294    "Connection: close\r\n\r\n"
295    "Failed to buffer the client body to apply content filters.\n"
296    "Could be caused by a socket timeout\n";
297
298 static const char UNSUPPORTED_CLIENT_EXPECTATION_ERROR_RESPONSE[] =
299    "HTTP/1.1 417 Expecting too much\r\n"
300    "Content-Type: text/plain\r\n"
301    "Connection: close\r\n\r\n"
302    "Privoxy detected an unsupported Expect header value.\n";
303
304 /* A function to crunch a response */
305 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
306
307 /* Crunch function flags */
308 #define CF_NO_FLAGS        0
309 /* Cruncher applies to forced requests as well */
310 #define CF_IGNORE_FORCE    1
311 /* Crunched requests are counted for the block statistics */
312 #define CF_COUNT_AS_REJECT 2
313
314 /* A crunch function and its flags */
315 struct cruncher
316 {
317    const crunch_func_ptr cruncher;
318    const int flags;
319 };
320
321 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
322
323 /* Complete list of cruncher functions */
324 static const struct cruncher crunchers_all[] = {
325    { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
326    { block_url,       CF_COUNT_AS_REJECT },
327 #ifdef FEATURE_TRUST
328    { trust_url,       CF_COUNT_AS_REJECT },
329 #endif /* def FEATURE_TRUST */
330    { redirect_url,    CF_NO_FLAGS  },
331    { dispatch_cgi,    CF_IGNORE_FORCE},
332    { NULL,            0 }
333 };
334
335 /* Light version, used after tags are applied */
336 static const struct cruncher crunchers_light[] = {
337    { block_url,       CF_COUNT_AS_REJECT },
338    { redirect_url,    CF_NO_FLAGS },
339    { NULL,            0 }
340 };
341
342
343 /*
344  * XXX: Don't we really mean
345  *
346  * #if defined(unix)
347  *
348  * here?
349  */
350 #if !defined(_WIN32)
351 /*********************************************************************
352  *
353  * Function    :  sig_handler
354  *
355  * Description :  Signal handler for different signals.
356  *                Exit gracefully on TERM and INT
357  *                or set a flag that will cause the errlog
358  *                to be reopened by the main thread on HUP.
359  *
360  * Parameters  :
361  *          1  :  the_signal = the signal cause this function to call
362  *
363  * Returns     :  -
364  *
365  *********************************************************************/
366 static void sig_handler(int the_signal)
367 {
368    switch(the_signal)
369    {
370       case SIGTERM:
371       case SIGINT:
372          log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
373 #if defined(unix)
374          if (pidfile)
375          {
376             unlink(pidfile);
377          }
378 #endif /* unix */
379          exit(the_signal);
380          break;
381
382       case SIGHUP:
383 #if defined(unix)
384          received_hup_signal = 1;
385 #endif
386          break;
387
388       default:
389          /*
390           * We shouldn't be here, unless we catch signals
391           * in main() that we can't handle here!
392           */
393          log_error(LOG_LEVEL_FATAL,
394             "sig_handler: exiting on unexpected signal %d", the_signal);
395    }
396    return;
397
398 }
399 #endif
400
401
402 /*********************************************************************
403  *
404  * Function    :  get_write_delay
405  *
406  * Description :  Parse the delay-response parameter.
407  *
408  * Parameters  :
409  *          1  :  csp = Current client state (buffers, headers, etc...)
410  *
411  * Returns     :  Number of milliseconds to delay writes.
412  *
413  *********************************************************************/
414 static unsigned int get_write_delay(const struct client_state *csp)
415 {
416    unsigned int delay;
417    char *endptr;
418    char *newval;
419
420    if ((csp->action->flags & ACTION_DELAY_RESPONSE) == 0)
421    {
422       return 0;
423    }
424    newval = csp->action->string[ACTION_STRING_DELAY_RESPONSE];
425
426    delay = (unsigned)strtol(newval, &endptr, 0);
427    if (*endptr != '\0')
428    {
429       log_error(LOG_LEVEL_FATAL,
430          "Invalid delay-response{} parameter: '%s'", newval);
431    }
432
433    return delay;
434
435 }
436
437
438 /*********************************************************************
439  *
440  * Function    :  client_protocol_is_unsupported
441  *
442  * Description :  Checks if the client used a known unsupported
443  *                protocol and deals with it by sending an error
444  *                response.
445  *
446  * Parameters  :
447  *          1  :  csp = Current client state (buffers, headers, etc...)
448  *          2  :  req = the first request line send by the client
449  *
450  * Returns     :  TRUE if an error response has been generated, or
451  *                FALSE if the request doesn't look invalid.
452  *
453  *********************************************************************/
454 static int client_protocol_is_unsupported(struct client_state *csp, char *req)
455 {
456    /*
457     * If it's a FTP or gopher request, we don't support it.
458     *
459     * These checks are better than nothing, but they might
460     * not work in all configurations and some clients might
461     * have problems digesting the answer.
462     *
463     * They should, however, never cause more problems than
464     * Privoxy's old behaviour (returning the misleading HTML
465     * error message:
466     *
467     * "Could not resolve http://(ftp|gopher)://example.org").
468     */
469    if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
470    {
471       const char *response = NULL;
472       const char *protocol = NULL;
473
474       if (!strncmpic(req, "GET ftp://", 10))
475       {
476          response = FTP_RESPONSE;
477          protocol = "FTP";
478       }
479       else
480       {
481          response = GOPHER_RESPONSE;
482          protocol = "GOPHER";
483       }
484       log_error(LOG_LEVEL_ERROR,
485          "%s tried to use Privoxy as %s proxy: %s",
486          csp->ip_addr_str, protocol, req);
487       log_error(LOG_LEVEL_CLF,
488          "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
489       freez(req);
490
491 #ifdef FEATURE_HTTPS_INSPECTION
492       if (client_use_ssl(csp))
493       {
494          ssl_send_data_delayed(&(csp->ssl_client_attr),
495             (const unsigned char *)response, strlen(response),
496             get_write_delay(csp));
497       }
498       else
499 #endif
500       {
501          write_socket_delayed(csp->cfd, response, strlen(response),
502             get_write_delay(csp));
503       }
504
505       return TRUE;
506    }
507
508    return FALSE;
509 }
510
511
512 /*********************************************************************
513  *
514  * Function    :  client_has_unsupported_expectations
515  *
516  * Description :  Checks if the client used an unsupported expectation
517  *                in which case an error message is delivered.
518  *
519  * Parameters  :
520  *          1  :  csp = Current client state (buffers, headers, etc...)
521  *
522  * Returns     :  TRUE if an error response has been generated, or
523  *                FALSE if the request doesn't look invalid.
524  *
525  *********************************************************************/
526 static int client_has_unsupported_expectations(const struct client_state *csp)
527 {
528    if ((csp->flags & CSP_FLAG_UNSUPPORTED_CLIENT_EXPECTATION))
529    {
530       log_error(LOG_LEVEL_ERROR,
531          "Rejecting request from client %s with unsupported Expect header value",
532          csp->ip_addr_str);
533       log_error(LOG_LEVEL_CLF,
534          "%s - - [%T] \"%s\" 417 0", csp->ip_addr_str, csp->http->cmd);
535       write_socket_delayed(csp->cfd,
536          UNSUPPORTED_CLIENT_EXPECTATION_ERROR_RESPONSE,
537          strlen(UNSUPPORTED_CLIENT_EXPECTATION_ERROR_RESPONSE),
538          get_write_delay(csp));
539
540       return TRUE;
541    }
542
543    return FALSE;
544
545 }
546
547
548 /*********************************************************************
549  *
550  * Function    :  get_request_destination_elsewhere
551  *
552  * Description :  If the client's request was redirected into
553  *                Privoxy without the client's knowledge,
554  *                the request line lacks the destination host.
555  *
556  *                This function tries to get it elsewhere,
557  *                provided accept-intercepted-requests is enabled.
558  *
559  *                "Elsewhere" currently only means "Host: header",
560  *                but in the future we may ask the redirecting
561  *                packet filter to look the destination up.
562  *
563  *                If the destination stays unknown, an error
564  *                response is send to the client and headers
565  *                are freed so that chat() can return directly.
566  *
567  * Parameters  :
568  *          1  :  csp = Current client state (buffers, headers, etc...)
569  *          2  :  headers = a header list
570  *
571  * Returns     :  JB_ERR_OK if the destination is now known, or
572  *                JB_ERR_PARSE if it isn't.
573  *
574  *********************************************************************/
575 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
576 {
577    if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
578    {
579       log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
580          " Privoxy isn't configured to accept intercepted requests.",
581          csp->ip_addr_str, csp->http->cmd);
582       /* XXX: Use correct size */
583       log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
584          csp->ip_addr_str, csp->http->cmd);
585
586       write_socket_delayed(csp->cfd, CHEADER, strlen(CHEADER),
587          get_write_delay(csp));
588       destroy_list(headers);
589
590       return JB_ERR_PARSE;
591    }
592    else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
593    {
594       /* Split the domain we just got for pattern matching */
595       init_domain_components(csp->http);
596
597       return JB_ERR_OK;
598    }
599    else
600    {
601       /* We can't work without destination. Go spread the news.*/
602
603       /* XXX: Use correct size */
604       log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
605          csp->ip_addr_str, csp->http->cmd);
606       log_error(LOG_LEVEL_ERROR,
607          "Privoxy was unable to get the destination for %s's request: %s",
608          csp->ip_addr_str, csp->http->cmd);
609
610       write_socket_delayed(csp->cfd, MISSING_DESTINATION_RESPONSE,
611          strlen(MISSING_DESTINATION_RESPONSE), get_write_delay(csp));
612       destroy_list(headers);
613
614       return JB_ERR_PARSE;
615    }
616    /*
617     * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
618     * to get the destination IP address, use it as host directly
619     * or do a reverse DNS lookup first.
620     */
621 }
622
623
624 /*********************************************************************
625  *
626  * Function    :  get_server_headers
627  *
628  * Description :  Parses server headers in iob and fills them
629  *                into csp->headers so that they can later be
630  *                handled by sed().
631  *
632  * Parameters  :
633  *          1  :  csp = Current client state (buffers, headers, etc...)
634  *
635  * Returns     :  JB_ERR_OK if everything went fine, or
636  *                JB_ERR_PARSE if the headers were incomplete.
637  *
638  *********************************************************************/
639 static jb_err get_server_headers(struct client_state *csp)
640 {
641    int continue_hack_in_da_house = 0;
642    char * header;
643
644    while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
645    {
646       if (header == NULL)
647       {
648          /*
649           * continue hack in da house. Ignore the ending of
650           * this head and continue enlisting header lines.
651           * The reason is described below.
652           */
653          enlist(csp->headers, "");
654          continue_hack_in_da_house = 0;
655          continue;
656       }
657       else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
658       {
659          /*
660           * It's a bodyless continue response, don't
661           * stop header parsing after reaching its end.
662           *
663           * As a result Privoxy will concatenate the
664           * next response's head and parse and deliver
665           * the headers as if they belonged to one request.
666           *
667           * The client will separate them because of the
668           * empty line between them.
669           *
670           * XXX: What we're doing here is clearly against
671           * the intended purpose of the continue header,
672           * and under some conditions (HTTP/1.0 client request)
673           * it's a standard violation.
674           *
675           * Anyway, "sort of against the spec" is preferable
676           * to "always getting confused by Continue responses"
677           * (Privoxy's behaviour before this hack was added)
678           */
679          log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
680          continue_hack_in_da_house = 1;
681       }
682       else if (*header == '\0')
683       {
684          /*
685           * If the header is empty, but the Continue hack
686           * isn't active, we can assume that we reached the
687           * end of the buffer before we hit the end of the
688           * head.
689           *
690           * Inform the caller an let it decide how to handle it.
691           */
692          return JB_ERR_PARSE;
693       }
694
695       if (JB_ERR_MEMORY == enlist(csp->headers, header))
696       {
697          /*
698           * XXX: Should we quit the request and return a
699           * out of memory error page instead?
700           */
701          log_error(LOG_LEVEL_ERROR,
702             "Out of memory while enlisting server headers. %s lost.",
703             header);
704       }
705       freez(header);
706    }
707
708    return JB_ERR_OK;
709 }
710
711
712 /*********************************************************************
713  *
714  * Function    :  crunch_reason
715  *
716  * Description :  Translates the crunch reason code into a string.
717  *
718  * Parameters  :
719  *          1  :  rsp = a http_response
720  *
721  * Returns     :  A string with the crunch reason or an error description.
722  *
723  *********************************************************************/
724 static const char *crunch_reason(const struct http_response *rsp)
725 {
726    char * reason = NULL;
727
728    assert(rsp != NULL);
729    if (rsp == NULL)
730    {
731       return "Internal error while searching for crunch reason";
732    }
733
734    switch (rsp->crunch_reason)
735    {
736       case UNSUPPORTED:
737          reason = "Unsupported HTTP feature";
738          break;
739       case BLOCKED:
740          reason = "Blocked";
741          break;
742       case UNTRUSTED:
743          reason = "Untrusted";
744          break;
745       case REDIRECTED:
746          reason = "Redirected";
747          break;
748       case CGI_CALL:
749          reason = "CGI Call";
750          break;
751       case NO_SUCH_DOMAIN:
752          reason = "DNS failure";
753          break;
754       case FORWARDING_FAILED:
755          reason = "Forwarding failed";
756          break;
757       case CONNECT_FAILED:
758          reason = "Connection failure";
759          break;
760       case OUT_OF_MEMORY:
761          reason = "Out of memory (may mask other reasons)";
762          break;
763       case CONNECTION_TIMEOUT:
764          reason = "Connection timeout";
765          break;
766       case NO_SERVER_DATA:
767          reason = "No server data received";
768          break;
769       default:
770          reason = "No reason recorded";
771          break;
772    }
773
774    return reason;
775 }
776
777
778 /*********************************************************************
779  *
780  * Function    :  log_applied_actions
781  *
782  * Description :  Logs the applied actions if LOG_LEVEL_ACTIONS is
783  *                enabled.
784  *
785  * Parameters  :
786  *          1  :  actions = Current action spec to log
787  *
788  * Returns     :  Nothing.
789  *
790  *********************************************************************/
791 static void log_applied_actions(const struct current_action_spec *actions)
792 {
793    /*
794     * The conversion to text requires lots of memory allocations so
795     * we only do the conversion if the user is actually interested.
796     */
797    if (debug_level_is_enabled(LOG_LEVEL_ACTIONS))
798    {
799       char *actions_as_text = actions_to_line_of_text(actions);
800       log_error(LOG_LEVEL_ACTIONS, "%s", actions_as_text);
801       freez(actions_as_text);
802    }
803 }
804
805
806 /*********************************************************************
807  *
808  * Function    :  send_crunch_response
809  *
810  * Description :  Delivers already prepared response for
811  *                intercepted requests, logs the interception
812  *                and frees the response.
813  *
814  * Parameters  :
815  *          1  :  csp = Current client state (buffers, headers, etc...)
816  *          2  :  rsp = Fully prepared response. Will be freed on exit.
817  *
818  * Returns     :  Nothing.
819  *
820  *********************************************************************/
821 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
822 {
823       const struct http_request *http = csp->http;
824       char status_code[4];
825
826       assert(rsp != NULL);
827       assert(rsp->head != NULL);
828
829       if (rsp == NULL)
830       {
831          log_error(LOG_LEVEL_FATAL, "NULL response in send_crunch_response.");
832       }
833
834       /*
835        * Extract the status code from the actual head
836        * that will be send to the client. It is the only
837        * way to get it right for all requests, including
838        * the fixed ones for out-of-memory problems.
839        *
840        * A head starts like this: 'HTTP/1.1 200...'
841        *                           0123456789|11
842        *                                     10
843        */
844       status_code[0] = rsp->head[9];
845       status_code[1] = rsp->head[10];
846       status_code[2] = rsp->head[11];
847       status_code[3] = '\0';
848
849       /* Log that the request was crunched and why. */
850       log_applied_actions(csp->action);
851 #ifdef FEATURE_HTTPS_INSPECTION
852       if (client_use_ssl(csp))
853       {
854          log_error(LOG_LEVEL_CRUNCH, "%s: https://%s%s", crunch_reason(rsp),
855             http->hostport, http->path);
856          log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s https://%s%s %s\" %s %lu",
857             csp->ip_addr_str, http->gpc, http->hostport, http->path,
858             http->version, status_code, rsp->content_length);
859       }
860       else
861 #endif
862       {
863          log_error(LOG_LEVEL_CRUNCH, "%s: %s", crunch_reason(rsp), http->url);
864          log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %lu",
865             csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
866       }
867       /* Write the answer to the client */
868 #ifdef FEATURE_HTTPS_INSPECTION
869       if (client_use_ssl(csp))
870       {
871          if ((ssl_send_data_delayed(&(csp->ssl_client_attr),
872                 (const unsigned char *)rsp->head, rsp->head_length,
873                 get_write_delay(csp)) < 0)
874           || (ssl_send_data_delayed(&(csp->ssl_client_attr),
875                 (const unsigned char *)rsp->body, rsp->content_length,
876                 get_write_delay(csp)) < 0))
877          {
878             /* There is nothing we can do about it. */
879             log_error(LOG_LEVEL_CONNECT, "Couldn't deliver the error message "
880                "for https://%s%s through client socket %d using TLS/SSL",
881                http->hostport, http->url, csp->cfd);
882          }
883       }
884       else
885 #endif
886       {
887          if (write_socket_delayed(csp->cfd, rsp->head, rsp->head_length,
888                 get_write_delay(csp))
889           || write_socket_delayed(csp->cfd, rsp->body, rsp->content_length,
890                 get_write_delay(csp)))
891          {
892             /* There is nothing we can do about it. */
893             log_error(LOG_LEVEL_CONNECT,
894                "Couldn't deliver the error message for %s through client socket %d: %E",
895                http->url, csp->cfd);
896          }
897       }
898
899       /* Clean up and return */
900       if (cgi_error_memory() != rsp)
901       {
902          free_http_response(rsp);
903       }
904       return;
905 }
906
907
908 /*********************************************************************
909  *
910  * Function    :  crunch_response_triggered
911  *
912  * Description :  Checks if the request has to be crunched,
913  *                and delivers the crunch response if necessary.
914  *
915  * Parameters  :
916  *          1  :  csp = Current client state (buffers, headers, etc...)
917  *          2  :  crunchers = list of cruncher functions to run
918  *
919  * Returns     :  TRUE if the request was answered with a crunch response
920  *                FALSE otherwise.
921  *
922  *********************************************************************/
923 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
924 {
925    struct http_response *rsp = NULL;
926    const struct cruncher *c;
927
928    /*
929     * If CGI request crunching is disabled,
930     * check the CGI dispatcher out of order to
931     * prevent unintentional blocks or redirects.
932     */
933    if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
934        && (NULL != (rsp = dispatch_cgi(csp))))
935    {
936       /* Deliver, log and free the interception response. */
937       send_crunch_response(csp, rsp);
938       csp->flags |= CSP_FLAG_CRUNCHED;
939       return TRUE;
940    }
941
942    for (c = crunchers; c->cruncher != NULL; c++)
943    {
944       /*
945        * Check the cruncher if either Privoxy is toggled
946        * on and the request isn't forced, or if the cruncher
947        * applies to forced requests as well.
948        */
949       if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
950           !(csp->flags & CSP_FLAG_FORCED)) ||
951           (c->flags & CF_IGNORE_FORCE))
952       {
953          rsp = c->cruncher(csp);
954          if (NULL != rsp)
955          {
956             /* Deliver, log and free the interception response. */
957             send_crunch_response(csp, rsp);
958             csp->flags |= CSP_FLAG_CRUNCHED;
959 #ifdef FEATURE_STATISTICS
960             if (c->flags & CF_COUNT_AS_REJECT)
961             {
962 #ifdef MUTEX_LOCKS_AVAILABLE
963                privoxy_mutex_lock(&block_statistics_mutex);
964                number_of_requests_blocked++;
965                privoxy_mutex_unlock(&block_statistics_mutex);
966 #endif
967                csp->flags |= CSP_FLAG_REJECTED;
968             }
969 #endif /* def FEATURE_STATISTICS */
970
971             return TRUE;
972          }
973       }
974    }
975
976    return FALSE;
977 }
978
979
980 /*********************************************************************
981  *
982  * Function    :  build_request_line
983  *
984  * Description :  Builds the HTTP request line.
985  *
986  *                If a HTTP forwarder is used it expects the whole URL,
987  *                web servers only get the path.
988  *
989  * Parameters  :
990  *          1  :  csp = Current client state (buffers, headers, etc...)
991  *          2  :  fwd = The forwarding spec used for the request.
992  *                      Can be NULL.
993  *          3  :  request_line = The old request line which will be replaced.
994  *
995  * Returns     :  Nothing. Terminates in case of memory problems.
996  *
997  *********************************************************************/
998 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
999 {
1000    struct http_request *http = csp->http;
1001
1002    /*
1003     * Downgrade http version from 1.1 to 1.0
1004     * if +downgrade action applies.
1005     */
1006    if ((csp->action->flags & ACTION_DOWNGRADE)
1007      && (!strcmpic(http->version, "HTTP/1.1")))
1008    {
1009       freez(http->version);
1010       http->version = strdup_or_die("HTTP/1.0");
1011    }
1012
1013    /*
1014     * Rebuild the request line.
1015     */
1016    freez(*request_line);
1017 #ifdef FEATURE_HTTPS_INSPECTION
1018    if (fwd != NULL && fwd->forward_host &&
1019        fwd->type != FORWARD_WEBSERVER && client_use_ssl(csp))
1020    {
1021       *request_line = strdup("CONNECT ");
1022    }
1023    else
1024 #endif
1025    {
1026       *request_line = strdup(http->gpc);
1027       string_append(request_line, " ");
1028    }
1029
1030    if (fwd != NULL && fwd->forward_host && fwd->type != FORWARD_WEBSERVER)
1031    {
1032 #ifdef FEATURE_HTTPS_INSPECTION
1033       if (client_use_ssl(csp))
1034       {
1035          char port_string[10];
1036
1037          string_append(request_line, http->host);
1038          snprintf(port_string, sizeof(port_string), ":%d", http->port);
1039          string_append(request_line, port_string);
1040       }
1041       else
1042 #endif
1043       {
1044          string_append(request_line, http->url);
1045       }
1046    }
1047    else
1048    {
1049       string_append(request_line, http->path);
1050    }
1051    string_append(request_line, " ");
1052    string_append(request_line, http->version);
1053
1054    if (*request_line == NULL)
1055    {
1056       log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1057    }
1058    log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1059 }
1060
1061
1062 /*********************************************************************
1063  *
1064  * Function    :  change_request_destination
1065  *
1066  * Description :  Parse a (rewritten) request line and regenerate
1067  *                the http request data.
1068  *
1069  * Parameters  :
1070  *          1  :  csp = Current client state (buffers, headers, etc...)
1071  *
1072  * Returns     :  Forwards the parse_http_request() return code.
1073  *                Terminates in case of memory problems.
1074  *
1075  *********************************************************************/
1076 static jb_err change_request_destination(struct client_state *csp)
1077 {
1078    struct http_request *http = csp->http;
1079    jb_err err;
1080
1081    log_error(LOG_LEVEL_REDIRECTS, "Rewrite detected: %s",
1082       csp->headers->first->str);
1083    free_http_request(http);
1084    err = parse_http_request(csp->headers->first->str, http);
1085    if (JB_ERR_OK != err)
1086    {
1087       log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1088          jb_err_to_string(err));
1089    }
1090    if (http->ssl && strcmpic(csp->http->gpc, "CONNECT"))
1091    {
1092       /*
1093        * A client header filter changed the request URL from
1094        * http:// to https:// which we currently don't support.
1095        */
1096       log_error(LOG_LEVEL_ERROR, "Changing the request destination from http "
1097          "to https behind the client's back currently isn't supported.");
1098       return JB_ERR_PARSE;
1099    }
1100
1101    return err;
1102 }
1103
1104
1105 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
1106 /*********************************************************************
1107  *
1108  * Function    :  server_response_is_complete
1109  *
1110  * Description :  Determines whether we should stop reading
1111  *                from the server socket.
1112  *
1113  * Parameters  :
1114  *          1  :  csp = Current client state (buffers, headers, etc...)
1115  *          2  :  content_length = Length of content received so far.
1116  *
1117  * Returns     :  TRUE if the response is complete,
1118  *                FALSE otherwise.
1119  *
1120  *********************************************************************/
1121 static int server_response_is_complete(struct client_state *csp,
1122    unsigned long long content_length)
1123 {
1124    int content_length_known = !!(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
1125
1126    if (!strcmpic(csp->http->gpc, "HEAD"))
1127    {
1128       /*
1129        * "HEAD" implies no body, we are thus expecting
1130        * no content. XXX: incomplete "list" of methods?
1131        */
1132       csp->expected_content_length = 0;
1133       content_length_known = TRUE;
1134       csp->flags |= CSP_FLAG_SERVER_CONTENT_LENGTH_SET;
1135    }
1136
1137    if (csp->http->status == 204 || csp->http->status == 304)
1138    {
1139       /*
1140        * Expect no body. XXX: incomplete "list" of status codes?
1141        */
1142       csp->expected_content_length = 0;
1143       content_length_known = TRUE;
1144       csp->flags |= CSP_FLAG_SERVER_CONTENT_LENGTH_SET;
1145    }
1146
1147    return (content_length_known && ((0 == csp->expected_content_length)
1148             || (csp->expected_content_length <= content_length)));
1149 }
1150
1151
1152 #ifdef FEATURE_CONNECTION_SHARING
1153 /*********************************************************************
1154  *
1155  * Function    :  wait_for_alive_connections
1156  *
1157  * Description :  Waits for alive connections to timeout.
1158  *
1159  * Parameters  :  N/A
1160  *
1161  * Returns     :  N/A
1162  *
1163  *********************************************************************/
1164 static void wait_for_alive_connections(void)
1165 {
1166    int connections_alive = close_unusable_connections();
1167
1168    while (0 < connections_alive)
1169    {
1170       log_error(LOG_LEVEL_CONNECT,
1171          "Waiting for %d connections to timeout.",
1172          connections_alive);
1173       sleep(60);
1174       connections_alive = close_unusable_connections();
1175    }
1176
1177    log_error(LOG_LEVEL_CONNECT, "No connections to wait for left.");
1178
1179 }
1180 #endif /* def FEATURE_CONNECTION_SHARING */
1181
1182
1183 /*********************************************************************
1184  *
1185  * Function    :  save_connection_destination
1186  *
1187  * Description :  Remembers a connection for reuse later on.
1188  *
1189  * Parameters  :
1190  *          1  :  sfd  = Open socket to remember.
1191  *          2  :  http = The destination for the connection.
1192  *          3  :  fwd  = The forwarder settings used.
1193  *          4  :  server_connection  = storage.
1194  *
1195  * Returns     : void
1196  *
1197  *********************************************************************/
1198 static void save_connection_destination(jb_socket sfd,
1199                                         const struct http_request *http,
1200                                         const struct forward_spec *fwd,
1201                                         struct reusable_connection *server_connection)
1202 {
1203    assert(sfd != JB_INVALID_SOCKET);
1204    assert(NULL != http->host);
1205
1206    server_connection->sfd = sfd;
1207    server_connection->host = strdup_or_die(http->host);
1208    server_connection->port = http->port;
1209
1210    assert(NULL != fwd);
1211    assert(server_connection->gateway_host == NULL);
1212    assert(server_connection->gateway_port == 0);
1213    assert(server_connection->forwarder_type == 0);
1214    assert(server_connection->forward_host == NULL);
1215    assert(server_connection->forward_port == 0);
1216
1217    server_connection->forwarder_type = fwd->type;
1218    if (NULL != fwd->gateway_host)
1219    {
1220       server_connection->gateway_host = strdup_or_die(fwd->gateway_host);
1221    }
1222    else
1223    {
1224       server_connection->gateway_host = NULL;
1225    }
1226    server_connection->gateway_port = fwd->gateway_port;
1227    if (NULL != fwd->auth_username)
1228    {
1229       server_connection->auth_username = strdup_or_die(fwd->auth_username);
1230    }
1231    else
1232    {
1233       server_connection->auth_username = NULL;
1234    }
1235    if (NULL != fwd->auth_password)
1236    {
1237       server_connection->auth_password = strdup_or_die(fwd->auth_password);
1238    }
1239    else
1240    {
1241       server_connection->auth_password = NULL;
1242    }
1243
1244    if (NULL != fwd->forward_host)
1245    {
1246       server_connection->forward_host = strdup_or_die(fwd->forward_host);
1247    }
1248    else
1249    {
1250       server_connection->forward_host = NULL;
1251    }
1252    server_connection->forward_port = fwd->forward_port;
1253 }
1254 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
1255
1256
1257 /*********************************************************************
1258  *
1259  * Function    : verify_request_length
1260  *
1261  * Description : Checks if we already got the whole client requests
1262  *               and sets CSP_FLAG_CLIENT_REQUEST_COMPLETELY_READ if
1263  *               we do.
1264  *
1265  *               Data that doesn't belong to the current request is
1266  *               either thrown away to let the client retry on a clean
1267  *               socket, or stashed to be dealt with after the current
1268  *               request is served.
1269  *
1270  * Parameters  :
1271  *          1  :  csp = Current client state (buffers, headers, etc...)
1272  *
1273  * Returns     :  void
1274  *
1275  *********************************************************************/
1276 static void verify_request_length(struct client_state *csp)
1277 {
1278    unsigned long long buffered_request_bytes =
1279       (unsigned long long)(csp->client_iob->eod - csp->client_iob->cur);
1280
1281    if ((csp->expected_client_content_length != 0)
1282       && (buffered_request_bytes != 0))
1283    {
1284       if (csp->expected_client_content_length >= buffered_request_bytes)
1285       {
1286          csp->expected_client_content_length -= buffered_request_bytes;
1287          log_error(LOG_LEVEL_CONNECT, "Reduced expected bytes to %llu "
1288             "to account for the %llu ones we already got.",
1289             csp->expected_client_content_length, buffered_request_bytes);
1290       }
1291       else
1292       {
1293          assert(csp->client_iob->eod > csp->client_iob->cur + csp->expected_client_content_length);
1294          csp->client_iob->eod = csp->client_iob->cur + csp->expected_client_content_length;
1295          log_error(LOG_LEVEL_CONNECT, "Reducing expected bytes to 0. "
1296             "Marking the server socket tainted after throwing %llu bytes away.",
1297             buffered_request_bytes - csp->expected_client_content_length);
1298          csp->expected_client_content_length = 0;
1299          csp->flags |= CSP_FLAG_SERVER_SOCKET_TAINTED;
1300       }
1301
1302       if (csp->expected_client_content_length == 0)
1303       {
1304          csp->flags |= CSP_FLAG_CLIENT_REQUEST_COMPLETELY_READ;
1305       }
1306    }
1307
1308    if (!(csp->flags & CSP_FLAG_CLIENT_REQUEST_COMPLETELY_READ)
1309       && ((csp->client_iob->cur < csp->client_iob->eod)
1310          || (csp->expected_client_content_length != 0)))
1311    {
1312       if (strcmpic(csp->http->gpc, "GET")
1313          && strcmpic(csp->http->gpc, "HEAD")
1314          && strcmpic(csp->http->gpc, "TRACE")
1315          && strcmpic(csp->http->gpc, "OPTIONS")
1316          && strcmpic(csp->http->gpc, "DELETE"))
1317       {
1318          /* XXX: this is an incomplete hack */
1319          csp->flags &= ~CSP_FLAG_CLIENT_REQUEST_COMPLETELY_READ;
1320          log_error(LOG_LEVEL_CONNECT, "There better be a request body.");
1321       }
1322       else
1323       {
1324          csp->flags |= CSP_FLAG_CLIENT_REQUEST_COMPLETELY_READ;
1325
1326          if ((csp->config->feature_flags & RUNTIME_FEATURE_TOLERATE_PIPELINING) == 0)
1327          {
1328             csp->flags |= CSP_FLAG_SERVER_SOCKET_TAINTED;
1329             log_error(LOG_LEVEL_CONNECT,
1330                "Possible pipeline attempt detected. The connection will not "
1331                "be kept alive and we will only serve the first request.");
1332             /* Nuke the pipelined requests from orbit, just to be sure. */
1333             clear_iob(csp->client_iob);
1334          }
1335          else
1336          {
1337             /*
1338              * Keep the pipelined data around for now, we'll deal with
1339              * it once we're done serving the current request.
1340              */
1341             csp->flags |= CSP_FLAG_PIPELINED_REQUEST_WAITING;
1342             assert(csp->client_iob->eod >= csp->client_iob->cur);
1343             log_error(LOG_LEVEL_CONNECT, "Complete client request followed by "
1344                "%d bytes of pipelined data received.",
1345                (int)(csp->client_iob->eod - csp->client_iob->cur));
1346          }
1347       }
1348    }
1349    else
1350    {
1351       csp->flags |= CSP_FLAG_CLIENT_REQUEST_COMPLETELY_READ;
1352       log_error(LOG_LEVEL_CONNECT, "Complete client request received.");
1353    }
1354 }
1355
1356
1357 /*********************************************************************
1358  *
1359  * Function    :  mark_server_socket_tainted
1360  *
1361  * Description :  Makes sure we don't reuse a server socket
1362  *                (if we didn't read everything the server sent
1363  *                us reusing the socket would lead to garbage).
1364  *
1365  * Parameters  :
1366  *          1  :  csp = Current client state (buffers, headers, etc...)
1367  *
1368  * Returns     :  void.
1369  *
1370  *********************************************************************/
1371 static void mark_server_socket_tainted(struct client_state *csp)
1372 {
1373    /*
1374     * For consistency we always mark the server socket
1375     * tainted, however, to reduce the log noise we only
1376     * emit a log message if the server socket could have
1377     * actually been reused.
1378     */
1379    if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE)
1380       && !(csp->flags & CSP_FLAG_SERVER_SOCKET_TAINTED))
1381    {
1382       log_error(LOG_LEVEL_CONNECT,
1383          "Marking the server socket %d tainted.",
1384          csp->server_connection.sfd);
1385    }
1386    csp->flags |= CSP_FLAG_SERVER_SOCKET_TAINTED;
1387 }
1388
1389 /*********************************************************************
1390  *
1391  * Function    :  get_request_line
1392  *
1393  * Description : Read the client request line.
1394  *
1395  * Parameters  :
1396  *          1  :  csp = Current client state (buffers, headers, etc...)
1397  *
1398  * Returns     :  Pointer to request line or NULL in case of errors.
1399  *
1400  *********************************************************************/
1401 static char *get_request_line(struct client_state *csp)
1402 {
1403    char buf[BUFFER_SIZE];
1404    char *request_line = NULL;
1405    int len;
1406
1407    memset(buf, 0, sizeof(buf));
1408
1409    if ((csp->flags & CSP_FLAG_PIPELINED_REQUEST_WAITING) != 0)
1410    {
1411       /*
1412        * If there are multiple pipelined requests waiting,
1413        * the flag will be set again once the next request
1414        * has been parsed.
1415        */
1416       csp->flags &= ~CSP_FLAG_PIPELINED_REQUEST_WAITING;
1417
1418       request_line = get_header(csp->client_iob);
1419       if ((NULL != request_line) && ('\0' != *request_line))
1420       {
1421          return request_line;
1422       }
1423       else
1424       {
1425          log_error(LOG_LEVEL_CONNECT, "No complete request line "
1426             "received yet. Continuing reading from %d.", csp->cfd);
1427       }
1428    }
1429
1430    do
1431    {
1432       if (
1433 #ifdef FUZZ
1434           0 == (csp->flags & CSP_FLAG_FUZZED_INPUT) &&
1435 #endif
1436           !data_is_available(csp->cfd, csp->config->socket_timeout)
1437           )
1438       {
1439          if (socket_is_still_alive(csp->cfd))
1440          {
1441             log_error(LOG_LEVEL_CONNECT,
1442                "No request line on socket %d received in time. Timeout: %d.",
1443                csp->cfd, csp->config->socket_timeout);
1444             write_socket_delayed(csp->cfd, CLIENT_CONNECTION_TIMEOUT_RESPONSE,
1445                strlen(CLIENT_CONNECTION_TIMEOUT_RESPONSE),
1446                get_write_delay(csp));
1447          }
1448          else
1449          {
1450             log_error(LOG_LEVEL_CONNECT,
1451                "The client side of the connection on socket %d got "
1452                "closed without sending a complete request line.", csp->cfd);
1453          }
1454          return NULL;
1455       }
1456
1457       len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1458
1459       if (len <= 0) return NULL;
1460
1461       /*
1462        * If there is no memory left for buffering the
1463        * request, there is nothing we can do but hang up
1464        */
1465       if (add_to_iob(csp->client_iob, csp->config->buffer_limit, buf, len))
1466       {
1467          return NULL;
1468       }
1469
1470       request_line = get_header(csp->client_iob);
1471
1472    } while ((NULL != request_line) && ('\0' == *request_line));
1473
1474    return request_line;
1475
1476 }
1477
1478 enum chunk_status
1479 {
1480    CHUNK_STATUS_MISSING_DATA,
1481    CHUNK_STATUS_BODY_COMPLETE,
1482    CHUNK_STATUS_PARSE_ERROR
1483 };
1484
1485
1486 /*********************************************************************
1487  *
1488  * Function    :  chunked_body_is_complete
1489  *
1490  * Description :  Figures out whether or not a chunked body is complete.
1491  *
1492  *                Currently it always starts at the beginning of the
1493  *                buffer which is somewhat wasteful and prevents Privoxy
1494  *                from starting to forward the correctly parsed chunks
1495  *                as soon as theoretically possible.
1496  *
1497  *                Should be modified to work with a common buffer,
1498  *                and allow the caller to skip already parsed chunks.
1499  *
1500  *                This would allow the function to be used for unbuffered
1501  *                response bodies as well.
1502  *
1503  * Parameters  :
1504  *          1  :  iob = Buffer with the body to check.
1505  *          2  :  length = Length of complete body
1506  *
1507  * Returns     :  Enum with the result of the check.
1508  *
1509  *********************************************************************/
1510 static enum chunk_status chunked_body_is_complete(struct iob *iob, size_t *length)
1511 {
1512    unsigned int chunksize;
1513    char *p = iob->cur;
1514
1515    do
1516    {
1517       /*
1518        * We need at least a single digit, followed by "\r\n",
1519        * followed by an unknown amount of data, followed by "\r\n".
1520        */
1521       if (p + 5 > iob->eod)
1522       {
1523          return CHUNK_STATUS_MISSING_DATA;
1524       }
1525       if (sscanf(p, "%x", &chunksize) != 1)
1526       {
1527          return CHUNK_STATUS_PARSE_ERROR;
1528       }
1529
1530       /*
1531        * We want at least a single digit, followed by "\r\n",
1532        * followed by the specified amount of data, followed by "\r\n".
1533        */
1534       if (p + chunksize + 5 > iob->eod)
1535       {
1536          return CHUNK_STATUS_MISSING_DATA;
1537       }
1538
1539       /* Skip chunk-size. */
1540       p = strstr(p, "\r\n");
1541       if (NULL == p)
1542       {
1543          return CHUNK_STATUS_PARSE_ERROR;
1544       }
1545       /* Move beyond the chunkdata. */
1546       p += 2 + chunksize;
1547
1548       /* Make sure we're still within the buffer and have two bytes left */
1549       if (p + 2 > iob->eod)
1550       {
1551          return CHUNK_STATUS_MISSING_DATA;
1552       }
1553
1554       /* There should be another "\r\n" to skip */
1555       if (memcmp(p, "\r\n", 2))
1556       {
1557          return CHUNK_STATUS_PARSE_ERROR;
1558       }
1559       p += 2;
1560    } while (chunksize > 0U);
1561
1562    *length = (size_t)(p - iob->cur);
1563    assert(*length <= (size_t)(iob->eod - iob->cur));
1564    assert(p <= iob->eod);
1565
1566    return CHUNK_STATUS_BODY_COMPLETE;
1567
1568 }
1569
1570
1571 /*********************************************************************
1572  *
1573  * Function    : receive_chunked_client_request_body
1574  *
1575  * Description : Read the chunk-encoded client request body.
1576  *               Failures are dealt with.
1577  *
1578  * Parameters  :
1579  *          1  :  csp = Current client state (buffers, headers, etc...)
1580  *
1581  * Returns     :  JB_ERR_OK or JB_ERR_PARSE
1582  *
1583  *********************************************************************/
1584 static jb_err receive_chunked_client_request_body(struct client_state *csp)
1585 {
1586    size_t body_length;
1587    enum chunk_status status;
1588
1589    while (CHUNK_STATUS_MISSING_DATA ==
1590       (status = chunked_body_is_complete(csp->client_iob, &body_length)))
1591    {
1592       char buf[BUFFER_SIZE];
1593       int len;
1594
1595       if (!data_is_available(csp->cfd, csp->config->socket_timeout))
1596       {
1597          log_error(LOG_LEVEL_ERROR,
1598             "Timeout while waiting for the client body.");
1599          break;
1600       }
1601       len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1602       if (len <= 0)
1603       {
1604          log_error(LOG_LEVEL_ERROR,
1605             "Reading the client body failed: %E");
1606          break;
1607       }
1608       if (add_to_iob(csp->client_iob, csp->config->buffer_limit, buf, len))
1609       {
1610          break;
1611       }
1612    }
1613    if (status != CHUNK_STATUS_BODY_COMPLETE)
1614    {
1615       write_socket_delayed(csp->cfd, CLIENT_BODY_PARSE_ERROR_RESPONSE,
1616          strlen(CLIENT_BODY_PARSE_ERROR_RESPONSE), get_write_delay(csp));
1617       log_error(LOG_LEVEL_CLF,
1618          "%s - - [%T] \"Failed reading chunked client body\" 400 0", csp->ip_addr_str);
1619       return JB_ERR_PARSE;
1620    }
1621    log_error(LOG_LEVEL_CONNECT,
1622       "Chunked client body completely read. Length: %lu", body_length);
1623    csp->expected_client_content_length = body_length;
1624
1625    return JB_ERR_OK;
1626
1627 }
1628
1629
1630 #ifdef FUZZ
1631 /*********************************************************************
1632  *
1633  * Function    :  fuzz_chunked_transfer_encoding
1634  *
1635  * Description :  Treat the fuzzed input as chunked transfer encoding
1636  *                to check and dechunk.
1637  *
1638  * Parameters  :
1639  *          1  :  csp      = Used to store the data.
1640  *          2  :  fuzz_input_file = File to read the input from.
1641  *
1642  * Returns     : Result of dechunking
1643  *
1644  *********************************************************************/
1645 extern int fuzz_chunked_transfer_encoding(struct client_state *csp, char *fuzz_input_file)
1646 {
1647    size_t length;
1648    size_t size = (size_t)(csp->iob->eod - csp->iob->cur);
1649    enum chunk_status status;
1650
1651    status = chunked_body_is_complete(csp->iob, &length);
1652    if (CHUNK_STATUS_BODY_COMPLETE != status)
1653    {
1654       log_error(LOG_LEVEL_INFO, "Chunked body is incomplete or invalid");
1655    }
1656
1657    return (JB_ERR_OK == remove_chunked_transfer_coding(csp->iob->cur, &size));
1658
1659 }
1660
1661
1662 /*********************************************************************
1663  *
1664  * Function    : fuzz_client_request
1665  *
1666  * Description : Try to get a client request from the fuzzed input.
1667  *
1668  * Parameters  :
1669  *          1  :  csp = Current client state (buffers, headers, etc...)
1670  *          2  :  fuzz_input_file = File to read the input from.
1671  *
1672  * Returns     :  Result of fuzzing.
1673  *
1674  *********************************************************************/
1675 extern int fuzz_client_request(struct client_state *csp, char *fuzz_input_file)
1676 {
1677    jb_err err;
1678
1679    csp->cfd = 0;
1680    csp->ip_addr_str = "fuzzer";
1681
1682    if (strcmp(fuzz_input_file, "-") != 0)
1683    {
1684       log_error(LOG_LEVEL_FATAL,
1685          "Fuzzed client requests can currently only be read from stdin (-).");
1686    }
1687    err = receive_client_request(csp);
1688    if (err != JB_ERR_OK)
1689    {
1690       return 1;
1691    }
1692    err = parse_client_request(csp);
1693    if (err != JB_ERR_OK)
1694    {
1695       return 1;
1696    }
1697
1698    return 0;
1699
1700 }
1701 #endif  /* def FUZZ */
1702
1703
1704 #ifdef FEATURE_FORCE_LOAD
1705 /*********************************************************************
1706  *
1707  * Function    :  force_required
1708  *
1709  * Description : Checks a request line to see if it contains
1710  *               the FORCE_PREFIX. If it does, it is removed
1711  *               unless enforcing requests has beend disabled.
1712  *
1713  * Parameters  :
1714  *          1  :  request_line = HTTP request line
1715  *
1716  * Returns     :  TRUE if force is required, FALSE otherwise.
1717  *
1718  *********************************************************************/
1719 static int force_required(const struct client_state *csp, char *request_line)
1720 {
1721    char *p;
1722
1723    p = strstr(request_line, "http://");
1724    if (p != NULL)
1725    {
1726       /* Skip protocol */
1727       p += strlen("http://");
1728    }
1729    else
1730    {
1731       /* Intercepted request usually don't specify the protocol. */
1732       p = request_line;
1733    }
1734
1735    /* Go to the beginning of the path */
1736    p = strstr(p, "/");
1737    if (p == NULL)
1738    {
1739       /*
1740        * If the path is missing the request line is invalid and we
1741        * are done here. The client-visible rejection happens later on.
1742        */
1743       return 0;
1744    }
1745
1746    if (0 == strncmpic(p, FORCE_PREFIX, strlen(FORCE_PREFIX) - 1))
1747    {
1748       if (!(csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS))
1749       {
1750          /* XXX: Should clean more carefully */
1751          strclean(request_line, FORCE_PREFIX);
1752          log_error(LOG_LEVEL_FORCE,
1753             "Enforcing request: \"%s\".", request_line);
1754
1755          return 1;
1756       }
1757       log_error(LOG_LEVEL_FORCE,
1758          "Ignored force prefix in request: \"%s\".", request_line);
1759    }
1760
1761    return 0;
1762
1763 }
1764 #endif /* def FEATURE_FORCE_LOAD */
1765
1766
1767 /*********************************************************************
1768  *
1769  * Function    :  receive_client_request
1770  *
1771  * Description : Read the client's request (more precisely the
1772  *               client headers) and answer it if necessary.
1773  *
1774  * Parameters  :
1775  *          1  :  csp = Current client state (buffers, headers, etc...)
1776  *
1777  * Returns     :  JB_ERR_OK, JB_ERR_PARSE or JB_ERR_MEMORY
1778  *
1779  *********************************************************************/
1780 static jb_err receive_client_request(struct client_state *csp)
1781 {
1782    char buf[BUFFER_SIZE];
1783    char *p;
1784    char *req = NULL;
1785    struct http_request *http;
1786    int len;
1787    jb_err err;
1788
1789    /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1790    struct list header_list;
1791    struct list *headers = &header_list;
1792
1793    /* We don't care if the arriving data is a valid HTTP request or not. */
1794    csp->requests_received_total++;
1795
1796    http = csp->http;
1797
1798    memset(buf, 0, sizeof(buf));
1799
1800    req = get_request_line(csp);
1801    if (req == NULL)
1802    {
1803       mark_server_socket_tainted(csp);
1804       return JB_ERR_PARSE;
1805    }
1806    assert(*req != '\0');
1807
1808    if (client_protocol_is_unsupported(csp, req))
1809    {
1810       return JB_ERR_PARSE;
1811    }
1812
1813 #ifdef FEATURE_FORCE_LOAD
1814    if (force_required(csp, req))
1815    {
1816       csp->flags |= CSP_FLAG_FORCED;
1817    }
1818 #endif /* def FEATURE_FORCE_LOAD */
1819
1820    err = parse_http_request(req, http);
1821    freez(req);
1822    if (JB_ERR_OK != err)
1823    {
1824       write_socket_delayed(csp->cfd, CHEADER, strlen(CHEADER),
1825          get_write_delay(csp));
1826       /* XXX: Use correct size */
1827       log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1828       log_error(LOG_LEVEL_ERROR,
1829          "Couldn't parse request line received from %s: %s",
1830          csp->ip_addr_str, jb_err_to_string(err));
1831
1832       free_http_request(http);
1833       return JB_ERR_PARSE;
1834    }
1835    if (http->ssl && strcmpic(http->gpc, "CONNECT"))
1836    {
1837       write_socket_delayed(csp->cfd, CHEADER, strlen(CHEADER),
1838          get_write_delay(csp));
1839       /* XXX: Use correct size */
1840       log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0",
1841          csp->ip_addr_str);
1842       log_error(LOG_LEVEL_ERROR, "Client %s tried to send a https "
1843          "URL without sending a CONNECT request first",
1844          csp->ip_addr_str);
1845       free_http_request(http);
1846       return JB_ERR_PARSE;
1847    }
1848
1849    /* grab the rest of the client's headers */
1850    init_list(headers);
1851    for (;;)
1852    {
1853       p = get_header(csp->client_iob);
1854
1855       if (p == NULL)
1856       {
1857          /* There are no additional headers to read. */
1858          break;
1859       }
1860
1861       if (*p == '\0')
1862       {
1863          /*
1864           * We didn't receive a complete header
1865           * line yet, get the rest of it.
1866           */
1867          if (!data_is_available(csp->cfd, csp->config->socket_timeout))
1868          {
1869             log_error(LOG_LEVEL_ERROR,
1870                "Client headers did not arrive in time. Timeout: %d",
1871                csp->config->socket_timeout);
1872             destroy_list(headers);
1873             return JB_ERR_PARSE;
1874          }
1875
1876          len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1877          if (len <= 0)
1878          {
1879             log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
1880             destroy_list(headers);
1881             return JB_ERR_PARSE;
1882          }
1883
1884          if (add_to_iob(csp->client_iob, csp->config->buffer_limit, buf, len))
1885          {
1886             /*
1887              * If there is no memory left for buffering the
1888              * request, there is nothing we can do but hang up
1889              */
1890             destroy_list(headers);
1891             return JB_ERR_MEMORY;
1892          }
1893       }
1894       else
1895       {
1896          if (!strncmpic(p, "Transfer-Encoding:", 18))
1897          {
1898             /*
1899              * XXX: should be called through sed()
1900              *      but currently can't.
1901              */
1902             client_transfer_encoding(csp, &p);
1903          }
1904          /*
1905           * We were able to read a complete
1906           * header and can finally enlist it.
1907           */
1908          enlist(headers, p);
1909          freez(p);
1910       }
1911    }
1912
1913    if (http->host == NULL)
1914    {
1915       /*
1916        * If we still don't know the request destination,
1917        * the request is invalid or the client uses
1918        * Privoxy without its knowledge.
1919        */
1920       if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
1921       {
1922          /*
1923           * Our attempts to get the request destination
1924           * elsewhere failed or Privoxy is configured
1925           * to only accept proxy requests.
1926           *
1927           * An error response has already been sent
1928           * and we're done here.
1929           */
1930          return JB_ERR_PARSE;
1931       }
1932    }
1933
1934 #ifdef FEATURE_CLIENT_TAGS
1935    /* XXX: If the headers were enlisted sooner, passing csp would do. */
1936    set_client_address(csp, headers);
1937    get_tag_list_for_client(csp->client_tags, csp->client_address);
1938 #endif
1939
1940    /*
1941     * Determine the actions for this URL
1942     */
1943 #ifdef FEATURE_TOGGLE
1944    if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
1945    {
1946       /* Most compatible set of actions (i.e. none) */
1947       init_current_action(csp->action);
1948    }
1949    else
1950 #endif /* ndef FEATURE_TOGGLE */
1951    {
1952       get_url_actions(csp, http);
1953    }
1954
1955    enlist(csp->headers, http->cmd);
1956
1957    /* Append the previously read headers */
1958    err = list_append_list_unique(csp->headers, headers);
1959    destroy_list(headers);
1960
1961    return err;
1962
1963 }
1964
1965
1966 /*********************************************************************
1967  *
1968  * Function    : parse_client_request
1969  *
1970  * Description : Parses the client's request and decides what to do
1971  *               with it.
1972  *
1973  *               Note that since we're not using select() we could get
1974  *               blocked here if a client connected, then didn't say
1975  *               anything!
1976  *
1977  * Parameters  :
1978  *          1  :  csp = Current client state (buffers, headers, etc...)
1979  *
1980  * Returns     :  JB_ERR_OK or JB_ERR_PARSE
1981  *
1982  *********************************************************************/
1983 static jb_err parse_client_request(struct client_state *csp)
1984 {
1985    struct http_request *http = csp->http;
1986    jb_err err;
1987
1988 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
1989    if ((csp->config->feature_flags & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE)
1990     && (!strcmpic(csp->http->version, "HTTP/1.1"))
1991     && (csp->http->ssl == 0))
1992    {
1993       /* Assume persistence until further notice */
1994       csp->flags |= CSP_FLAG_CLIENT_CONNECTION_KEEP_ALIVE;
1995    }
1996 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
1997
1998    if (csp->http->ssl == 0)
1999    {
2000       /*
2001        * This whole block belongs to chat() but currently
2002        * has to be executed before sed().
2003        */
2004       if (csp->flags & CSP_FLAG_CHUNKED_CLIENT_BODY)
2005       {
2006          if (receive_chunked_client_request_body(csp) != JB_ERR_OK)
2007          {
2008             return JB_ERR_PARSE;
2009          }
2010       }
2011       else
2012       {
2013          csp->expected_client_content_length = get_expected_content_length(csp->headers);
2014       }
2015       verify_request_length(csp);
2016    }
2017 #ifndef FEATURE_HTTPS_INSPECTION
2018    else
2019    {
2020       csp->flags |= CSP_FLAG_SERVER_SOCKET_TAINTED;
2021    }
2022 #endif
2023
2024    err = sed(csp, FILTER_CLIENT_HEADERS);
2025    if (JB_ERR_OK != err)
2026    {
2027       log_error(LOG_LEVEL_ERROR, "Failed to parse client request from %s.",
2028          csp->ip_addr_str);
2029       log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
2030          csp->ip_addr_str, csp->http->cmd);
2031       write_socket_delayed(csp->cfd, CHEADER, strlen(CHEADER), get_write_delay(csp));
2032       return JB_ERR_PARSE;
2033    }
2034    csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2035
2036    /* Check request line for rewrites. */
2037    if ((NULL == csp->headers->first->str)
2038       || (strcmp(http->cmd, csp->headers->first->str) &&
2039          (JB_ERR_OK != change_request_destination(csp))))
2040    {
2041       /*
2042        * A header filter broke the request line - bail out.
2043        */
2044       write_socket_delayed(csp->cfd, MESSED_UP_REQUEST_RESPONSE,
2045          strlen(MESSED_UP_REQUEST_RESPONSE), get_write_delay(csp));
2046       /* XXX: Use correct size */
2047       log_error(LOG_LEVEL_CLF,
2048          "%s - - [%T] \"Invalid request generated\" 400 0", csp->ip_addr_str);
2049       log_error(LOG_LEVEL_ERROR,
2050          "Invalid request line after applying header filters.");
2051       free_http_request(http);
2052
2053       return JB_ERR_PARSE;
2054    }
2055
2056    if (client_has_unsupported_expectations(csp))
2057    {
2058       return JB_ERR_PARSE;
2059    }
2060
2061    return JB_ERR_OK;
2062
2063 }
2064
2065
2066 /*********************************************************************
2067  *
2068  * Function    : read_http_request_body
2069  *
2070  * Description : Reads remaining request body from the client.
2071  *
2072  * Parameters  :
2073  *          1  :  csp = Current client state (buffers, headers, etc...)
2074  *
2075  * Returns     :  0 on success, anything else is an error.
2076  *
2077  *********************************************************************/
2078 static int read_http_request_body(struct client_state *csp)
2079 {
2080    size_t to_read = csp->expected_client_content_length;
2081    int len;
2082
2083    assert(to_read != 0);
2084
2085    /* check if all data has been already read */
2086    if (to_read <= (csp->client_iob->eod - csp->client_iob->cur))
2087    {
2088       return 0;
2089    }
2090
2091    for (to_read -= (size_t)(csp->client_iob->eod - csp->client_iob->cur);
2092         to_read > 0 && data_is_available(csp->cfd, csp->config->socket_timeout);
2093         to_read -= (unsigned)len)
2094    {
2095       char buf[BUFFER_SIZE];
2096       size_t max_bytes_to_read = to_read < sizeof(buf) ? to_read : sizeof(buf);
2097
2098       log_error(LOG_LEVEL_CONNECT,
2099          "Waiting for up to %lu bytes of request body from the client.",
2100          max_bytes_to_read);
2101       len = read_socket(csp->cfd, buf, (int)max_bytes_to_read);
2102       if (len <= -1)
2103       {
2104          log_error(LOG_LEVEL_CONNECT, "Failed receiving request body from %s: %E", csp->ip_addr_str);
2105          return 1;
2106       }
2107       if (add_to_iob(csp->client_iob, csp->config->buffer_limit, (char *)buf, len))
2108       {
2109          return 1;
2110       }
2111       assert(to_read >= len);
2112    }
2113
2114    if (to_read != 0)
2115    {
2116       log_error(LOG_LEVEL_CONNECT,
2117          "Not enough request body has been read: expected %lu more bytes",
2118          to_read);
2119       return 1;
2120    }
2121    log_error(LOG_LEVEL_CONNECT,
2122       "The last %d bytes of the request body have been read", len);
2123    return 0;
2124 }
2125
2126
2127 /*********************************************************************
2128  *
2129  * Function    : update_client_headers
2130  *
2131  * Description : Updates the HTTP headers from the client request.
2132  *
2133  * Parameters  :
2134  *          1  :  csp = Current client state (buffers, headers, etc...)
2135  *          2  :  new_content_length = new content length value to set
2136  *
2137  * Returns     :  0 on success, anything else is an error.
2138  *
2139  *********************************************************************/
2140 static int update_client_headers(struct client_state *csp, size_t new_content_length)
2141 {
2142    static const char content_length[] = "Content-Length:";
2143    int updated = 0;
2144    struct list_entry *p;
2145
2146 #ifndef FEATURE_HTTPS_INSPECTION
2147    for (p = csp->headers->first;
2148 #else
2149    for (p = csp->http->client_ssl ? csp->https_headers->first : csp->headers->first;
2150 #endif
2151         !updated  && (p != NULL); p = p->next)
2152    {
2153       /* Header crunch()ed in previous run? -> ignore */
2154       if (p->str == NULL)
2155       {
2156          continue;
2157       }
2158
2159       /* Does the current parser handle this header? */
2160       if (0 == strncmpic(p->str, content_length, sizeof(content_length) - 1))
2161       {
2162          updated = (JB_ERR_OK == header_adjust_content_length((char **)&(p->str), new_content_length));
2163          if (!updated)
2164          {
2165             return 1;
2166          }
2167       }
2168    }
2169
2170    return !updated;
2171 }
2172
2173
2174 /*********************************************************************
2175  *
2176  * Function    : can_buffer_request_body
2177  *
2178  * Description : Checks if the current request body can be stored in
2179  *               the client_iob without hitting buffer limit.
2180  *
2181  * Parameters  :
2182  *          1  : csp = Current client state (buffers, headers, etc...)
2183  *
2184  * Returns     : TRUE if the current request size do not exceed buffer limit
2185  *               FALSE otherwise.
2186  *
2187  *********************************************************************/
2188 static int can_buffer_request_body(const struct client_state *csp)
2189 {
2190    if (!can_add_to_iob(csp->client_iob, csp->config->buffer_limit,
2191                        csp->expected_client_content_length))
2192    {
2193       log_error(LOG_LEVEL_INFO,
2194          "Not filtering request body from %s: buffer limit %lu will be exceeded "
2195          "(content length %llu)", csp->ip_addr_str, csp->config->buffer_limit,
2196          csp->expected_client_content_length);
2197       return FALSE;
2198    }
2199    return TRUE;
2200 }
2201
2202
2203 /*********************************************************************
2204  *
2205  * Function    : send_http_request
2206  *
2207  * Description : Sends the HTTP headers from the client request
2208  *               and all the body data that has already been received.
2209  *
2210  * Parameters  :
2211  *          1  :  csp = Current client state (buffers, headers, etc...)
2212  *
2213  * Returns     :  0 on success, 1 on error, 2 if the request got crunched.
2214  *
2215  *********************************************************************/
2216 static int send_http_request(struct client_state *csp)
2217 {
2218    char *hdr;
2219    int write_failure;
2220
2221    hdr = list_to_text(csp->headers);
2222    if (hdr == NULL)
2223    {
2224       /* FIXME Should handle error properly */
2225       log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2226    }
2227    list_remove_all(csp->headers);
2228
2229    /*
2230     * Write the client's (modified) header to the server
2231     * (along with anything else that may be in the buffer)
2232     */
2233    write_failure = 0 != write_socket(csp->server_connection.sfd, hdr, strlen(hdr));
2234    freez(hdr);
2235
2236    if (write_failure)
2237    {
2238       log_error(LOG_LEVEL_CONNECT, "Failed sending request headers to: %s: %E",
2239          csp->http->hostport);
2240       return 1;
2241    }
2242
2243    /* XXX: Filtered data is not sent if there's a pipelined request? */
2244    if (((csp->flags & CSP_FLAG_PIPELINED_REQUEST_WAITING) == 0)
2245       && (flush_iob(csp->server_connection.sfd, csp->client_iob, 0) < 0))
2246    {
2247       log_error(LOG_LEVEL_CONNECT, "Failed sending request body to: %s: %E",
2248          csp->http->hostport);
2249       return 1;
2250    }
2251    return 0;
2252 }
2253
2254
2255 #ifdef FEATURE_HTTPS_INSPECTION
2256 /*********************************************************************
2257  *
2258  * Function    : read_https_request_body
2259  *
2260  * Description : Reads remaining request body from the client.
2261  *
2262  * Parameters  :
2263  *          1  :  csp = Current client state (buffers, headers, etc...)
2264  *
2265  * Returns     :  0 on success, anything else is an error.
2266  *
2267  *********************************************************************/
2268 static int read_https_request_body(struct client_state *csp)
2269 {
2270    size_t to_read = csp->expected_client_content_length;
2271    int len;
2272
2273    assert(to_read != 0);
2274
2275    /* check if all data has been already read */
2276    if (to_read <= (csp->client_iob->eod - csp->client_iob->cur))
2277    {
2278       return 0;
2279    }
2280
2281    for (to_read -= (size_t)(csp->client_iob->eod - csp->client_iob->cur);
2282         to_read > 0 && (is_ssl_pending(&(csp->ssl_client_attr)) ||
2283           data_is_available(csp->cfd, csp->config->socket_timeout));
2284         to_read -= (unsigned)len)
2285    {
2286       unsigned char buf[BUFFER_SIZE];
2287       size_t max_bytes_to_read = to_read < sizeof(buf) ? to_read : sizeof(buf);
2288
2289       log_error(LOG_LEVEL_CONNECT,
2290          "Buffering encrypted client body. Prepared to read up to %lu bytes.",
2291          max_bytes_to_read);
2292       len = ssl_recv_data(&(csp->ssl_client_attr), buf,
2293          (unsigned)max_bytes_to_read);
2294       if (len <= 0)
2295       {
2296          log_error(LOG_LEVEL_CONNECT,
2297             "Did not receive the whole encrypted request body from %s",
2298             csp->ip_addr_str);
2299          return 1;
2300       }
2301       if (add_to_iob(csp->client_iob, csp->config->buffer_limit, (char *)buf, len))
2302       {
2303          return 1;
2304       }
2305       assert(to_read >= len);
2306    }
2307
2308    if (to_read != 0)
2309    {
2310       log_error(LOG_LEVEL_CONNECT,
2311          "Not enough encrypted request body has been read: expected %lu more bytes",
2312          to_read);
2313       return 1;
2314    }
2315
2316    log_error(LOG_LEVEL_CONNECT,
2317       "The last %llu bytes of the encrypted request body have been read",
2318       csp->expected_client_content_length);
2319    return 0;
2320 }
2321
2322
2323 /*********************************************************************
2324  *
2325  * Function    : receive_and_send_encrypted_post_data
2326  *
2327  * Description : Reads remaining request body from the client and sends
2328  *               it to the server.
2329  *
2330  * Parameters  :
2331  *          1  :  csp = Current client state (buffers, headers, etc...)
2332  *
2333  * Returns     :  0 on success, anything else is an error.
2334  *
2335  *********************************************************************/
2336 static int receive_and_send_encrypted_post_data(struct client_state *csp)
2337 {
2338    int content_length_known = csp->expected_client_content_length != 0;
2339
2340    while ((content_length_known && csp->expected_client_content_length != 0) ||
2341       (is_ssl_pending(&(csp->ssl_client_attr)) ||
2342             data_is_available(csp->cfd, csp->config->socket_timeout)))
2343    {
2344       unsigned char buf[BUFFER_SIZE];
2345       int len;
2346       int max_bytes_to_read = sizeof(buf);
2347
2348       if (content_length_known && csp->expected_client_content_length < sizeof(buf))
2349       {
2350          max_bytes_to_read = (int)csp->expected_client_content_length;
2351       }
2352       log_error(LOG_LEVEL_CONNECT,
2353          "Prepared to read up to %d bytes of encrypted request body from the client.",
2354          max_bytes_to_read);
2355       len = ssl_recv_data(&(csp->ssl_client_attr), buf,
2356          (unsigned)max_bytes_to_read);
2357       if (len == -1)
2358       {
2359          return 1;
2360       }
2361       if (len == 0)
2362       {
2363          /* XXX: Does this actually happen? */
2364          break;
2365       }
2366       log_error(LOG_LEVEL_CONNECT, "Forwarding %d bytes of encrypted request body",
2367          len);
2368       len = ssl_send_data(&(csp->ssl_server_attr), buf, (size_t)len);
2369       if (len == -1)
2370       {
2371          return 1;
2372       }
2373       if (csp->expected_client_content_length != 0)
2374       {
2375          if (csp->expected_client_content_length >= len)
2376          {
2377             csp->expected_client_content_length -= (unsigned)len;
2378          }
2379          if (csp->expected_client_content_length == 0)
2380          {
2381             log_error(LOG_LEVEL_CONNECT, "Forwarded the last %d bytes", len);
2382             break;
2383          }
2384       }
2385    }
2386
2387    log_error(LOG_LEVEL_CONNECT, "Done forwarding encrypted request body");
2388
2389    return 0;
2390
2391 }
2392
2393
2394 /*********************************************************************
2395  *
2396  * Function    : send_https_request
2397  *
2398  * Description : Sends the HTTP headers from the client request
2399  *               and all the body data that has already been received.
2400  *
2401  * Parameters  :
2402  *          1  :  csp = Current client state (buffers, headers, etc...)
2403  *
2404  * Returns     :  0 on success, anything else is an error.
2405  *
2406  *********************************************************************/
2407 static int send_https_request(struct client_state *csp)
2408 {
2409    char *hdr;
2410    int ret;
2411    long flushed = 0;
2412
2413    hdr = list_to_text(csp->https_headers);
2414    if (hdr == NULL)
2415    {
2416       /* FIXME Should handle error properly */
2417       log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2418    }
2419    list_remove_all(csp->https_headers);
2420
2421    /*
2422     * Write the client's (modified) header to the server
2423     * (along with anything else that may be in the buffer)
2424     */
2425    ret = ssl_send_data(&(csp->ssl_server_attr),
2426       (const unsigned char *)hdr, strlen(hdr));
2427    freez(hdr);
2428
2429    if (ret < 0)
2430    {
2431       log_error(LOG_LEVEL_CONNECT,
2432          "Failed sending encrypted request headers to: %s: %E",
2433          csp->http->hostport);
2434       mark_server_socket_tainted(csp);
2435       return 1;
2436    }
2437
2438    /* XXX: Client body isn't sent if there's pipelined data? */
2439    if (((csp->flags & CSP_FLAG_PIPELINED_REQUEST_WAITING) == 0)
2440       && ((flushed = ssl_flush_socket(&(csp->ssl_server_attr),
2441             csp->client_iob)) < 0))
2442    {
2443       log_error(LOG_LEVEL_CONNECT, "Failed sending request body to: %s: %E",
2444          csp->http->hostport);
2445       return 1;
2446    }
2447    if (flushed != 0 || csp->expected_client_content_length != 0)
2448    {
2449       if (csp->expected_client_content_length != 0)
2450       {
2451          if (csp->expected_client_content_length < flushed)
2452          {
2453             log_error(LOG_LEVEL_ERROR,
2454                "Flushed %ld bytes of request body while only expecting %llu",
2455                flushed, csp->expected_client_content_length);
2456             csp->expected_client_content_length = 0;
2457          }
2458          else
2459          {
2460             log_error(LOG_LEVEL_CONNECT,
2461                "Flushed %ld bytes of request body while expecting %llu",
2462                flushed, csp->expected_client_content_length);
2463             csp->expected_client_content_length -= (unsigned)flushed;
2464             if (receive_and_send_encrypted_post_data(csp))
2465             {
2466                return 1;
2467             }
2468          }
2469       }
2470       else
2471       {
2472          log_error(LOG_LEVEL_CONNECT,
2473             "Flushed %ld bytes of request body", flushed);
2474       }
2475    }
2476
2477    log_error(LOG_LEVEL_CONNECT, "Encrypted request sent");
2478
2479    return 0;
2480
2481 }
2482
2483
2484 /*********************************************************************
2485  *
2486  * Function    :  receive_encrypted_request_headers
2487  *
2488  * Description :  Receives the encrypted request headers when
2489  *                https-inspecting.
2490  *
2491  * Parameters  :
2492  *          1  :  csp = Current client state (buffers, headers, etc...)
2493  *
2494  * Returns     :  JB_ERR_OK on success,
2495  *                JB_ERR_PARSE or JB_ERR_MEMORY otherwise
2496  *
2497  *********************************************************************/
2498 static jb_err receive_encrypted_request_headers(struct client_state *csp)
2499 {
2500    char buf[BUFFER_SIZE];
2501    int len;
2502    char *p;
2503
2504    do
2505    {
2506       log_error(LOG_LEVEL_HEADER, "Waiting for encrypted client headers");
2507       if (!is_ssl_pending(&(csp->ssl_client_attr)) &&
2508           !data_is_available(csp->cfd, csp->config->socket_timeout))
2509       {
2510          log_error(LOG_LEVEL_CONNECT,
2511             "Socket %d timed out while waiting for client headers", csp->cfd);
2512          return JB_ERR_PARSE;
2513       }
2514       len = ssl_recv_data(&(csp->ssl_client_attr),
2515          (unsigned char *)buf, sizeof(buf));
2516       if (len == 0)
2517       {
2518          log_error(LOG_LEVEL_CONNECT,
2519             "Socket %d closed while waiting for client headers", csp->cfd);
2520          return JB_ERR_PARSE;
2521       }
2522       if (len == -1)
2523       {
2524          return JB_ERR_PARSE;
2525       }
2526       if (add_to_iob(csp->client_iob, csp->config->buffer_limit, buf, len))
2527       {
2528          return JB_ERR_MEMORY;
2529       }
2530       p = strstr(csp->client_iob->cur, "\r\n\r\n");
2531    } while (p == NULL);
2532
2533    log_error(LOG_LEVEL_HEADER, "Encrypted headers received completely");
2534
2535    return JB_ERR_OK;
2536 }
2537
2538
2539 /*********************************************************************
2540  *
2541  * Function    :  change_encrypted_request_destination
2542  *
2543  * Description :  Parse a (rewritten) request line from an encrypted
2544  *                request and regenerate the http request data.
2545  *
2546  * Parameters  :
2547  *          1  :  csp = Current client state (buffers, headers, etc...)
2548  *
2549  * Returns     :  Forwards the parse_http_request() return code.
2550  *                Terminates in case of memory problems.
2551  *
2552  *********************************************************************/
2553 static jb_err change_encrypted_request_destination(struct client_state *csp)
2554 {
2555    jb_err err;
2556    char *original_host = csp->http->host;
2557    int original_port = csp->http->port;
2558
2559    log_error(LOG_LEVEL_REDIRECTS, "Rewrite detected: %s",
2560       csp->https_headers->first->str);
2561    csp->http->host = NULL;
2562    free_http_request(csp->http);
2563    err = parse_http_request(csp->https_headers->first->str, csp->http);
2564    if (JB_ERR_OK != err)
2565    {
2566       log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
2567          jb_err_to_string(err));
2568       freez(original_host);
2569       return err;
2570    }
2571
2572    if (csp->http->host == NULL)
2573    {
2574       char port_string[10];
2575       /*
2576        * The rewritten request line did not specify a host
2577        * which means we can use the original host specified
2578        * by the client.
2579        */
2580       csp->http->host = original_host;
2581       csp->http->port = original_port;
2582       log_error(LOG_LEVEL_REDIRECTS, "Keeping the original host: %s",
2583          csp->http->host);
2584       /*
2585        * If the rewritten request line didn't contain a host
2586        * it also didn't contain a port so we can reuse the host
2587        * port.
2588        */
2589       freez(csp->http->hostport);
2590       csp->http->hostport = strdup_or_die(csp->http->host);
2591       snprintf(port_string, sizeof(port_string), ":%d", original_port);
2592       err = string_append(&csp->http->hostport, port_string);
2593       if (err != JB_ERR_OK)
2594       {
2595          log_error(LOG_LEVEL_ERROR, "Failed to rebuild hostport: %s.",
2596             jb_err_to_string(err));
2597          return err;
2598       }
2599
2600       /*
2601        * While the request line didn't mention it,
2602        * we're https-inspecting and want to speak TLS
2603        * with the server.
2604        */
2605       csp->http->server_ssl = 1;
2606       csp->http->ssl = 1;
2607    }
2608    else
2609    {
2610       /* The rewrite filter added a host so we can ditch the original */
2611       freez(original_host);
2612       csp->http->server_ssl = csp->http->ssl;
2613    }
2614
2615    csp->http->client_ssl = 1;
2616
2617    freez(csp->https_headers->first->str);
2618    build_request_line(csp, NULL, &csp->https_headers->first->str);
2619
2620    if (!server_use_ssl(csp))
2621    {
2622       log_error(LOG_LEVEL_REDIRECTS,
2623          "Rewritten request line results in downgrade to http");
2624       /*
2625        * Replace the unencryptd headers received with the
2626        * CONNECT request with the ones we received securely.
2627        */
2628       destroy_list(csp->headers);
2629       csp->headers->first = csp->https_headers->first;
2630       csp->headers->last  = csp->https_headers->last;
2631       csp->https_headers->first = NULL;
2632       csp->https_headers->last = NULL;
2633    }
2634
2635    return JB_ERR_OK;
2636
2637 }
2638
2639
2640 /*********************************************************************
2641  *
2642  * Function    :  process_encrypted_request_headers
2643  *
2644  * Description :  Receives and parses the encrypted headers send
2645  *                by the client when https-inspecting.
2646  *
2647  * Parameters  :
2648  *          1  :  csp = Current client state (buffers, headers, etc...)
2649  *
2650  * Returns     :  JB_ERR_OK on success,
2651  *                JB_ERR_PARSE or JB_ERR_MEMORY otherwise
2652  *
2653  *********************************************************************/
2654 static jb_err process_encrypted_request_headers(struct client_state *csp)
2655 {
2656    char *p;
2657    char *request_line;
2658    jb_err err;
2659    /* Temporary copy of the client's headers before they get enlisted in csp->https_headers */
2660    struct list header_list;
2661    struct list *headers = &header_list;
2662
2663    assert(csp->ssl_with_client_is_opened);
2664
2665 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2666    if (csp->config->feature_flags & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE)
2667    {
2668       csp->flags |= CSP_FLAG_CLIENT_CONNECTION_KEEP_ALIVE;
2669    }
2670 #endif
2671    err = receive_encrypted_request_headers(csp);
2672    if (err != JB_ERR_OK)
2673    {
2674       if (csp->client_iob->cur == NULL ||
2675           csp->client_iob->cur == csp->client_iob->eod)
2676       {
2677          /*
2678           * We did not receive any data, most likely because the
2679           * client is done. Don't log this as a parse failure.
2680           */
2681          return JB_ERR_PARSE;
2682       }
2683       /* XXX: Also used for JB_ERR_MEMORY */
2684       log_error(LOG_LEVEL_ERROR, "Failed to receive encrypted request: %s",
2685          jb_err_to_string(err));
2686       ssl_send_data_delayed(&(csp->ssl_client_attr),
2687          (const unsigned char *)CHEADER, strlen(CHEADER), get_write_delay(csp));
2688       return err;
2689    }
2690
2691    /* We don't need get_request_line() because the whole HTTP head is buffered. */
2692    request_line = get_header(csp->client_iob);
2693    if (request_line == NULL)
2694    {
2695       log_error(LOG_LEVEL_ERROR, "Failed to get the encrypted request line");
2696       ssl_send_data_delayed(&(csp->ssl_client_attr),
2697          (const unsigned char *)CHEADER, strlen(CHEADER), get_write_delay(csp));
2698       return JB_ERR_PARSE;
2699    }
2700    assert(*request_line != '\0');
2701
2702    if (client_protocol_is_unsupported(csp, request_line))
2703    {
2704       /*
2705        * If the protocol is unsupported we're done here.
2706        * client_protocol_is_unsupported() took care of sending
2707        * the error response and logging the error message.
2708        */
2709       return JB_ERR_PARSE;
2710    }
2711
2712 #ifdef FEATURE_FORCE_LOAD
2713    if (force_required(csp, request_line))
2714    {
2715       csp->flags |= CSP_FLAG_FORCED;
2716    }
2717 #endif /* def FEATURE_FORCE_LOAD */
2718
2719    free_http_request(csp->http);
2720
2721    err = parse_http_request(request_line, csp->http);
2722    /* XXX: Restore ssl setting. This is ugly */
2723    csp->http->client_ssl = 1;
2724    csp->http->server_ssl = 1;
2725
2726    freez(request_line);
2727    if (JB_ERR_OK != err)
2728    {
2729       ssl_send_data_delayed(&(csp->ssl_client_attr),
2730          (const unsigned char *)CHEADER, strlen(CHEADER), get_write_delay(csp));
2731       /* XXX: Use correct size */
2732       log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2733       log_error(LOG_LEVEL_ERROR,
2734          "Couldn't parse request line received from %s: %s",
2735          csp->ip_addr_str, jb_err_to_string(err));
2736
2737       free_http_request(csp->http);
2738       return JB_ERR_PARSE;
2739    }
2740
2741    /* Parse the rest of the client's headers. */
2742    init_list(headers);
2743    for (;;)
2744    {
2745       p = get_header(csp->client_iob);
2746
2747       if (p == NULL)
2748       {
2749          /* There are no additional headers to read. */
2750          break;
2751       }
2752       enlist(headers, p);
2753       freez(p);
2754    }
2755
2756    if (JB_ERR_OK != get_destination_from_https_headers(headers, csp->http))
2757    {
2758       /*
2759        * Our attempts to get the request destination
2760        * elsewhere failed.
2761        */
2762       log_error(LOG_LEVEL_ERROR,
2763          "Failed to get the encrypted request destination");
2764       ssl_send_data_delayed(&(csp->ssl_client_attr),
2765          (const unsigned char *)CHEADER, strlen(CHEADER), get_write_delay(csp));
2766       destroy_list(headers);
2767
2768       return JB_ERR_PARSE;
2769    }
2770
2771    /* Split the domain we just got for pattern matching */
2772    init_domain_components(csp->http);
2773
2774 #ifdef FEATURE_CLIENT_TAGS
2775    /* XXX: If the headers were enlisted sooner, passing csp would do. */
2776    if (csp->client_address == NULL)
2777    {
2778       set_client_address(csp, headers);
2779       get_tag_list_for_client(csp->client_tags, csp->client_address);
2780    }
2781 #endif
2782
2783 #ifdef FEATURE_TOGGLE
2784    if ((csp->flags & CSP_FLAG_TOGGLED_ON) != 0)
2785 #endif
2786    {
2787       /*
2788        * Determine the actions for this request after
2789        * clearing the ones from the previous one.
2790        */
2791       free_current_action(csp->action);
2792       get_url_actions(csp, csp->http);
2793    }
2794
2795    enlist(csp->https_headers, csp->http->cmd);
2796
2797    /* Append the previously read headers */
2798    err = list_append_list_unique(csp->https_headers, headers);
2799    destroy_list(headers);
2800    if (JB_ERR_OK != err)
2801    {
2802       /* XXX: Send error message */
2803       return err;
2804    }
2805
2806    /* XXX: Work around crash */
2807    csp->error_message = NULL;
2808
2809    /* XXX: Why do this here? */
2810    csp->http->ssl = 1;
2811
2812    err = sed_https(csp);
2813    if (JB_ERR_OK != err)
2814    {
2815       ssl_send_data_delayed(&(csp->ssl_client_attr),
2816          (const unsigned char *)CHEADER, strlen(CHEADER), get_write_delay(csp));
2817       log_error(LOG_LEVEL_ERROR, "Failed to parse client request from %s.",
2818          csp->ip_addr_str);
2819       log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
2820          csp->ip_addr_str, csp->http->cmd);
2821       return JB_ERR_PARSE;
2822    }
2823
2824    if ((NULL == csp->https_headers->first->str)
2825       || (strcmp(csp->http->cmd, csp->https_headers->first->str) &&
2826          (JB_ERR_OK != change_encrypted_request_destination(csp))))
2827    {
2828       ssl_send_data_delayed(&(csp->ssl_client_attr),
2829          (const unsigned char *)MESSED_UP_REQUEST_RESPONSE,
2830          strlen(MESSED_UP_REQUEST_RESPONSE), get_write_delay(csp));
2831       log_error(LOG_LEVEL_ERROR,
2832          "Invalid request line after applying header filters.");
2833       /* XXX: Use correct size */
2834       log_error(LOG_LEVEL_CLF,
2835          "%s - - [%T] \"Invalid request generated\" 400 0", csp->ip_addr_str);
2836
2837       return JB_ERR_PARSE;
2838    }
2839
2840    log_error(LOG_LEVEL_HEADER, "Encrypted request headers processed");
2841    log_error(LOG_LEVEL_REQUEST, "https://%s%s", csp->http->hostport,
2842       csp->http->path);
2843
2844    return err;
2845
2846 }
2847
2848 /*********************************************************************
2849  *
2850  * Function    :  cgi_page_requested
2851  *
2852  * Description :  Checks if a request is for an internal CGI page.
2853  *
2854  * Parameters  :
2855  *          1  :  host = The host requested by the client.
2856  *
2857  * Returns     :  1 if a CGI page has been requested, 0 otherwise
2858  *
2859  *********************************************************************/
2860 static int cgi_page_requested(const char *host)
2861 {
2862    if ((0 == strcmpic(host, CGI_SITE_1_HOST))
2863     || (0 == strcmpic(host, CGI_SITE_1_HOST "."))
2864     || (0 == strcmpic(host, CGI_SITE_2_HOST))
2865     || (0 == strcmpic(host, CGI_SITE_2_HOST ".")))
2866    {
2867       return 1;
2868    }
2869
2870    return 0;
2871
2872 }
2873
2874
2875 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2876 /*********************************************************************
2877  *
2878  * Function    :  continue_https_chat
2879  *
2880  * Description :  Behaves similar to chat() but only deals with
2881  *                https-inspected requests that arrive on an already
2882  *                established connection. The first request is always
2883  *                served by chat() which is a lot more complex as it
2884  *                has to deal with forwarding settings and connection
2885  *                failures etc.
2886  *
2887  *                If a connection to the server has already been
2888  *                opened it is reused unless the request is blocked
2889  *                or the forwarder changed.
2890  *
2891  *                If a connection to the server has not yet been
2892  *                opened (because the previous request was crunched),
2893  *                or the forwarder changed, the connection is dropped
2894  *                so that the client retries on a fresh one.
2895  *
2896  * Parameters  :
2897  *          1  :  csp = Current client state (buffers, headers, etc...)
2898  *
2899  * Returns     :  Nothing.
2900  *
2901  *********************************************************************/
2902 static void continue_https_chat(struct client_state *csp)
2903 {
2904    const struct forward_spec *fwd;
2905
2906    if (JB_ERR_OK != process_encrypted_request_headers(csp))
2907    {
2908       csp->flags &= ~CSP_FLAG_CLIENT_CONNECTION_KEEP_ALIVE;
2909       return;
2910    }
2911
2912 #if defined(FEATURE_STATISTICS) && defined(MUTEX_LOCKS_AVAILABLE)
2913    privoxy_mutex_lock(&block_statistics_mutex);
2914    number_of_requests_received++;
2915    privoxy_mutex_unlock(&block_statistics_mutex);
2916 #endif
2917
2918    csp->requests_received_total++;
2919
2920    /*
2921     * We have an encrypted request. Check if one of the crunchers wants it.
2922     */
2923    if (crunch_response_triggered(csp, crunchers_all))
2924    {
2925       /*
2926        * Yes. The client got the crunch response and we're done here.
2927        */
2928       return;
2929    }
2930    if (csp->ssl_with_server_is_opened == 0)
2931    {
2932       log_error(LOG_LEVEL_CONNECT,
2933          "Dropping the client connection on socket %d. "
2934          "The server connection has not been established yet.",
2935          csp->cfd);
2936       csp->flags &= ~CSP_FLAG_CLIENT_CONNECTION_KEEP_ALIVE;
2937       return;
2938    }
2939    assert(csp->server_connection.sfd != JB_INVALID_SOCKET);
2940
2941    if (csp->expected_client_content_length != 0 &&
2942       (client_body_filters_enabled(csp->action) ||
2943        client_body_taggers_enabled(csp->action)) &&
2944       can_buffer_request_body(csp))
2945    {
2946       int content_modified;
2947
2948       if (read_https_request_body(csp))
2949       {
2950          /* XXX: handle */
2951          return;
2952       }
2953       if (client_body_taggers_enabled(csp->action))
2954       {
2955          execute_client_body_taggers(csp, csp->expected_client_content_length);
2956          if (crunch_response_triggered(csp, crunchers_all))
2957          {
2958             /*
2959              * Yes. The client got the crunch response and we're done here.
2960              */
2961             return;
2962          }
2963       }
2964       if (client_body_filters_enabled(csp->action))
2965       {
2966          size_t modified_content_length = csp->expected_client_content_length;
2967          content_modified = execute_client_body_filters(csp,
2968             &modified_content_length);
2969          if ((content_modified == 1) &&
2970             (modified_content_length != csp->expected_client_content_length) &&
2971             update_client_headers(csp, modified_content_length))
2972          {
2973             /* XXX: Send error response */
2974             log_error(LOG_LEVEL_HEADER, "Error updating client headers");
2975             return;
2976          }
2977       }
2978       csp->expected_client_content_length = 0;
2979    }
2980
2981    fwd = forward_url(csp, csp->http);
2982    if (!connection_destination_matches(&csp->server_connection, csp->http, fwd))
2983    {
2984       log_error(LOG_LEVEL_CONNECT,
2985          "Dropping the client connection on socket %d with "
2986          "server socket %d connected to %s. The forwarder has changed.",
2987          csp->cfd, csp->server_connection.sfd, csp->server_connection.host);
2988       csp->flags &= ~CSP_FLAG_CLIENT_CONNECTION_KEEP_ALIVE;
2989       return;
2990    }
2991
2992    log_applied_actions(csp->action);
2993
2994    log_error(LOG_LEVEL_CONNECT,
2995       "Reusing server socket %d connected to %s. Requests already sent: %u.",
2996       csp->server_connection.sfd, csp->server_connection.host,
2997       csp->server_connection.requests_sent_total);
2998
2999    if (send_https_request(csp))
3000    {
3001       /*
3002        * Most likely the server connection timed out. We can't easily
3003        * create a new one so simply drop the client connection without a
3004        * error response to let the client retry.
3005        */
3006       log_error(LOG_LEVEL_CONNECT,
3007          "Dropping client connection on socket %d. "
3008          "Forwarding the encrypted client request failed.",
3009          csp->cfd);
3010       return;
3011    }
3012    csp->server_connection.request_sent = time(NULL);
3013    csp->server_connection.requests_sent_total++;
3014    handle_established_connection(csp);
3015    freez(csp->receive_buffer);
3016 }
3017 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3018 #endif
3019
3020
3021 /*********************************************************************
3022  *
3023  * Function    :  handle_established_connection
3024  *
3025  * Description :  Shuffle data between client and server once the
3026  *                connection has been established and the request
3027  *                has been sent.
3028  *
3029  * Parameters  :
3030  *          1  :  csp = Current client state (buffers, headers, etc...)
3031  *
3032  * Returns     :  Nothing.
3033  *
3034  *********************************************************************/
3035 static void handle_established_connection(struct client_state *csp)
3036 {
3037    char *hdr;
3038    char *p;
3039    int n;
3040 #ifdef HAVE_POLL
3041    struct pollfd poll_fds[2];
3042 #else
3043    fd_set rfds;
3044    jb_socket maxfd;
3045    struct timeval timeout;
3046 #endif
3047    int server_body;
3048    int ms_iis5_hack = 0;
3049    unsigned long long byte_count = 0;
3050    struct http_request *http;
3051    long len = 0; /* for buffer sizes (and negative error codes) */
3052    int buffer_and_filter_content = 0;
3053    unsigned int write_delay;
3054 #ifdef FEATURE_HTTPS_INSPECTION
3055    int ret = 0;
3056    int use_ssl_tunnel = 0;
3057    csp->dont_verify_certificate = 0;
3058
3059    if (csp->http->ssl && !(csp->action->flags & ACTION_HTTPS_INSPECTION))
3060    {
3061       /* Pass encrypted content without filtering. */
3062       use_ssl_tunnel = 1;
3063    }
3064 #endif
3065
3066    /* Skeleton for HTTP response, if we should intercept the request */
3067    struct http_response *rsp;
3068 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3069    int watch_client_socket;
3070 #endif
3071
3072    csp->receive_buffer_size = csp->config->receive_buffer_size;
3073    csp->receive_buffer = zalloc(csp->receive_buffer_size + 1);
3074    if (csp->receive_buffer == NULL)
3075    {
3076       log_error(LOG_LEVEL_ERROR,
3077          "Out of memory. Failed to allocate the receive buffer.");
3078       rsp = cgi_error_memory();
3079       send_crunch_response(csp, rsp);
3080       return;
3081    }
3082
3083    http = csp->http;
3084
3085 #ifndef HAVE_POLL
3086    maxfd = (csp->cfd > csp->server_connection.sfd) ?
3087       csp->cfd : csp->server_connection.sfd;
3088 #endif
3089
3090    /* pass data between the client and server
3091     * until one or the other shuts down the connection.
3092     */
3093
3094    server_body = 0;
3095
3096 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3097    watch_client_socket = 0 == (csp->flags & CSP_FLAG_PIPELINED_REQUEST_WAITING);
3098 #endif
3099    write_delay = get_write_delay(csp);
3100
3101    for (;;)
3102    {
3103 #ifdef FEATURE_HTTPS_INSPECTION
3104       if (server_use_ssl(csp) && is_ssl_pending(&(csp->ssl_server_attr)))
3105       {
3106          /*
3107           * It's possible that the TLS library already consumed all the
3108           * data the server intends to send. If that happens poll() and
3109           * select() will no longer see the data as available so we have
3110           * to skip the calls.
3111           */
3112          goto server_wants_to_talk;
3113       }
3114       if (watch_client_socket && client_use_ssl(csp) &&
3115          is_ssl_pending(&(csp->ssl_client_attr)))
3116       {
3117          /*
3118           * The TLS libray may also consume all of the remaining data
3119           * from the client when we're shuffling the data from an
3120           * unbuffered request body to the server.
3121           */
3122          goto client_wants_to_talk;
3123       }
3124 #endif
3125 #ifndef HAVE_POLL
3126       FD_ZERO(&rfds);
3127 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3128       if (!watch_client_socket)
3129       {
3130          maxfd = csp->server_connection.sfd;
3131       }
3132       else
3133 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3134       {
3135          FD_SET(csp->cfd, &rfds);
3136       }
3137
3138       FD_SET(csp->server_connection.sfd, &rfds);
3139 #endif /* ndef HAVE_POLL */
3140
3141 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3142       if ((csp->flags & CSP_FLAG_CHUNKED)
3143          && !(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET)
3144          && ((csp->iob->eod - csp->iob->cur) >= 5)
3145          && !memcmp(csp->iob->eod-5, "0\r\n\r\n", 5))
3146       {
3147          /*
3148           * XXX: This check should be obsolete now,
3149           *      but let's wait a while to be sure.
3150           */
3151          log_error(LOG_LEVEL_CONNECT,
3152             "Looks like we got the last chunk together with "
3153             "the server headers but didn't detect it earlier. "
3154             "We better stop reading.");
3155          byte_count = (unsigned long long)(csp->iob->eod - csp->iob->cur);
3156          csp->expected_content_length = byte_count;
3157          csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
3158       }
3159       if (server_body && server_response_is_complete(csp, byte_count))
3160       {
3161          if (csp->expected_content_length == byte_count)
3162          {
3163             log_error(LOG_LEVEL_CONNECT,
3164                "Done reading from server. Content length: %llu as expected. "
3165                "Bytes most recently read: %ld.",
3166                byte_count, len);
3167          }
3168          else
3169          {
3170             log_error(LOG_LEVEL_CONNECT,
3171                "Done reading from server. Expected content length: %llu. "
3172                "Actual content length: %llu. Bytes most recently read: %ld.",
3173                csp->expected_content_length, byte_count, len);
3174          }
3175          len = 0;
3176          /*
3177           * XXX: Should not jump around, handle_established_connection()
3178           * is complicated enough already.
3179           */
3180          goto reading_done;
3181       }
3182 #endif  /* FEATURE_CONNECTION_KEEP_ALIVE */
3183
3184 #ifdef HAVE_POLL
3185       poll_fds[0].fd = csp->cfd;
3186 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3187       if (!watch_client_socket)
3188       {
3189          /*
3190           * Ignore incoming data, but still watch out
3191           * for disconnects etc. These flags are always
3192           * implied anyway but explicitly setting them
3193           * doesn't hurt.
3194           */
3195          poll_fds[0].events = POLLERR|POLLHUP;
3196       }
3197       else
3198 #endif
3199       {
3200          poll_fds[0].events = POLLIN;
3201       }
3202       poll_fds[1].fd = csp->server_connection.sfd;
3203       poll_fds[1].events = POLLIN;
3204       n = poll(poll_fds, 2, csp->config->socket_timeout * 1000);
3205 #else
3206       timeout.tv_sec = csp->config->socket_timeout;
3207       timeout.tv_usec = 0;
3208       n = select((int)maxfd + 1, &rfds, NULL, NULL, &timeout);
3209 #endif /* def HAVE_POLL */
3210
3211       /* Server or client not responding in timeout */
3212       if (n == 0)
3213       {
3214          log_error(LOG_LEVEL_CONNECT, "Socket timeout %d reached: %s",
3215             csp->config->socket_timeout, http->url);
3216          if ((byte_count == 0) && (http->ssl == 0))
3217          {
3218             send_crunch_response(csp, error_response(csp, "connection-timeout"));
3219          }
3220          mark_server_socket_tainted(csp);
3221 #ifdef FEATURE_HTTPS_INSPECTION
3222          close_client_and_server_ssl_connections(csp);
3223 #endif
3224          return;
3225       }
3226       else if (n < 0)
3227       {
3228 #ifdef HAVE_POLL
3229          log_error(LOG_LEVEL_ERROR, "poll() failed!: %E");
3230 #else
3231          log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
3232 #endif
3233          mark_server_socket_tainted(csp);
3234 #ifdef FEATURE_HTTPS_INSPECTION
3235          close_client_and_server_ssl_connections(csp);
3236 #endif
3237          return;
3238       }
3239
3240       /*
3241        * This is the body of the browser's request,
3242        * just read and write it.
3243        *
3244        * Receives data from browser and sends it to server
3245        *
3246        * XXX: Make sure the client doesn't use pipelining
3247        * behind Privoxy's back.
3248        */
3249 #ifdef HAVE_POLL
3250       if ((poll_fds[0].revents & (POLLERR|POLLHUP|POLLNVAL)) != 0)
3251       {
3252          log_error(LOG_LEVEL_CONNECT,
3253             "The client socket %d has become unusable while "
3254             "the server socket %d is still open.",
3255             csp->cfd, csp->server_connection.sfd);
3256          mark_server_socket_tainted(csp);
3257          break;
3258       }
3259
3260       if (poll_fds[0].revents != 0)
3261 #else
3262       if (FD_ISSET(csp->cfd, &rfds))
3263 #endif /* def HAVE_POLL*/
3264       {
3265          int max_bytes_to_read;
3266
3267 #ifdef FEATURE_HTTPS_INSPECTION
3268          client_wants_to_talk:
3269 #endif
3270
3271          max_bytes_to_read = (int)csp->receive_buffer_size;
3272
3273 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3274          if ((csp->flags & CSP_FLAG_CLIENT_REQUEST_COMPLETELY_READ))
3275          {
3276             if (data_is_available(csp->cfd, 0))
3277             {
3278                /*
3279                 * If the next request is already waiting, we have
3280                 * to stop select()ing the client socket. Otherwise
3281                 * we would always return right away and get nothing
3282                 * else done.
3283                 */
3284                watch_client_socket = 0;
3285                log_error(LOG_LEVEL_CONNECT,
3286                   "Stop watching client socket %d. "
3287                   "There's already another request waiting.",
3288                   csp->cfd);
3289                continue;
3290             }
3291             /*
3292              * If the client socket is set, but there's no data
3293              * available on the socket, the client went fishing
3294              * and continuing talking to the server makes no sense.
3295              */
3296             log_error(LOG_LEVEL_CONNECT,
3297                "The client closed socket %d while "
3298                "the server socket %d is still open.",
3299                csp->cfd, csp->server_connection.sfd);
3300             mark_server_socket_tainted(csp);
3301             break;
3302          }
3303          if (csp->expected_client_content_length != 0)
3304          {
3305             if (csp->expected_client_content_length < csp->receive_buffer_size)
3306             {
3307                max_bytes_to_read = (int)csp->expected_client_content_length;
3308             }
3309             log_error(LOG_LEVEL_CONNECT,
3310                "Waiting for up to %d bytes from the client.",
3311                max_bytes_to_read);
3312          }
3313          assert(max_bytes_to_read <= csp->receive_buffer_size);
3314 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3315
3316 #ifdef FEATURE_HTTPS_INSPECTION
3317          if (client_use_ssl(csp))
3318          {
3319             if (csp->http->status == 101)
3320             {
3321                len = ssl_recv_data(&(csp->ssl_client_attr),
3322                   (unsigned char *)csp->receive_buffer,
3323                   (size_t)max_bytes_to_read);
3324                if (len == -1)
3325                {
3326                   log_error(LOG_LEVEL_ERROR, "Failed to receive data "
3327                      "on client socket %d for an upgraded connection",
3328                      csp->cfd);
3329                   break;
3330                }
3331                if (len == 0)
3332                {
3333                   log_error(LOG_LEVEL_CONNECT, "Done receiving data "
3334                      "on client socket %d for an upgraded connection",
3335                      csp->cfd);
3336                   break;
3337                }
3338                byte_count += (unsigned long long)len;
3339                len = ssl_send_data(&(csp->ssl_server_attr),
3340                   (unsigned char *)csp->receive_buffer, (size_t)len);
3341                if (len == -1)
3342                {
3343                   log_error(LOG_LEVEL_ERROR, "Failed to send data "
3344                      "on server socket %d for an upgraded connection",
3345                      csp->server_connection.sfd);
3346                   break;
3347                }
3348                continue;
3349             }
3350             log_error(LOG_LEVEL_CONNECT, "Breaking with TLS/SSL.");
3351             break;
3352          }
3353          else
3354 #endif /* def FEATURE_HTTPS_INSPECTION */
3355          {
3356             len = read_socket(csp->cfd, csp->receive_buffer, max_bytes_to_read);
3357
3358             if (len <= 0)
3359             {
3360                /* XXX: not sure if this is necessary. */
3361                mark_server_socket_tainted(csp);
3362                break; /* "game over, man" */
3363             }
3364
3365 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3366             if (csp->expected_client_content_length != 0)
3367             {
3368                assert(len <= max_bytes_to_read);
3369                csp->expected_client_content_length -= (unsigned)len;
3370                log_error(LOG_LEVEL_CONNECT,
3371                   "Expected client content length set to %llu "
3372                   "after reading %ld bytes.",
3373                   csp->expected_client_content_length, len);
3374                if (csp->expected_client_content_length == 0)
3375                {
3376                   log_error(LOG_LEVEL_CONNECT,
3377                      "Done reading from the client.");
3378                   csp->flags |= CSP_FLAG_CLIENT_REQUEST_COMPLETELY_READ;
3379                }
3380             }
3381 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3382
3383             if (write_socket(csp->server_connection.sfd, csp->receive_buffer, (size_t)len))
3384             {
3385                log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
3386                mark_server_socket_tainted(csp);
3387                return;
3388             }
3389          }
3390          continue;
3391       }
3392
3393       /*
3394        * The server wants to talk. It could be the header or the body.
3395        * If `hdr' is null, then it's the header otherwise it's the body.
3396        * FIXME: Does `hdr' really mean `host'? No.
3397        */
3398 #ifdef HAVE_POLL
3399       if (poll_fds[1].revents != 0)
3400 #else
3401       if (FD_ISSET(csp->server_connection.sfd, &rfds))
3402 #endif /* HAVE_POLL */
3403       {
3404 #ifdef FEATURE_HTTPS_INSPECTION
3405          server_wants_to_talk:
3406 #endif
3407 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3408          /*
3409           * If we are buffering content, we don't want to eat up to
3410           * buffer-limit bytes if the client no longer cares about them.
3411           * If we aren't buffering, however, a dead client socket will be
3412           * noticed pretty much right away anyway, so we can reduce the
3413           * overhead by skipping the check.
3414           */
3415          if (buffer_and_filter_content && !socket_is_still_alive(csp->cfd))
3416          {
3417 #ifdef _WIN32
3418             log_error(LOG_LEVEL_CONNECT,
3419                "The server still wants to talk, but the client may already have hung up on us.");
3420 #else
3421             log_error(LOG_LEVEL_CONNECT,
3422                "The server still wants to talk, but the client hung up on us.");
3423             mark_server_socket_tainted(csp);
3424 #ifdef FEATURE_HTTPS_INSPECTION
3425             close_client_and_server_ssl_connections(csp);
3426 #endif
3427             return;
3428 #endif /* def _WIN32 */
3429          }
3430 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3431
3432 #ifdef FEATURE_HTTPS_INSPECTION
3433          /*
3434           * Reading data from standard or secured connection (HTTP/HTTPS)
3435           */
3436          if (server_use_ssl(csp))
3437          {
3438             len = ssl_recv_data(&(csp->ssl_server_attr),
3439                (unsigned char *)csp->receive_buffer, csp->receive_buffer_size);
3440          }
3441          else
3442 #endif
3443          {
3444             len = read_socket(csp->server_connection.sfd, csp->receive_buffer,
3445                (int)csp->receive_buffer_size);
3446          }
3447
3448          if (len < 0)
3449          {
3450             log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
3451
3452             if ((http->ssl && (csp->fwd == NULL))
3453 #ifdef FEATURE_HTTPS_INSPECTION
3454                && use_ssl_tunnel
3455 #endif
3456                 )
3457             {
3458                /*
3459                 * Just hang up. We already confirmed the client's CONNECT
3460                 * request with status code 200 and unencrypted content is
3461                 * no longer welcome.
3462                 */
3463                log_error(LOG_LEVEL_ERROR,
3464                   "CONNECT already confirmed. Unable to tell the client about the problem.");
3465                return;
3466             }
3467             else if (byte_count)
3468             {
3469                /*
3470                 * Just hang up. We already transmitted the original headers
3471                 * and parts of the original content and therefore missed the
3472                 * chance to send an error message (without risking data corruption).
3473                 *
3474                 * XXX: we could retry with a fancy range request here.
3475                 */
3476                log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
3477                   "Unable to tell the client about the problem.");
3478                mark_server_socket_tainted(csp);
3479 #ifdef FEATURE_HTTPS_INSPECTION
3480                close_client_and_server_ssl_connections(csp);
3481 #endif
3482                return;
3483             }
3484             /*
3485              * XXX: Consider handling the cases above the same.
3486              */
3487             mark_server_socket_tainted(csp);
3488             len = 0;
3489          }
3490
3491 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3492          if (csp->flags & CSP_FLAG_CHUNKED)
3493          {
3494             if ((len >= 5) && !memcmp(csp->receive_buffer+len-5, "0\r\n\r\n", 5))
3495             {
3496                /* XXX: this is a temporary hack */
3497                log_error(LOG_LEVEL_CONNECT,
3498                   "Looks like we reached the end of the last chunk. "
3499                   "We better stop reading.");
3500                csp->expected_content_length = byte_count + (unsigned long long)len;
3501                csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
3502             }
3503          }
3504          reading_done:
3505 #endif  /* FEATURE_CONNECTION_KEEP_ALIVE */
3506
3507          /*
3508           * This is guaranteed by allocating with zalloc_or_die()
3509           * and never (intentionally) writing to the last byte.
3510           *
3511           * csp->receive_buffer_size is the size of the part of the
3512           * buffer we intentionally write to, but we actually
3513           * allocated csp->receive_buffer_size+1 bytes so the assertion
3514           * stays within the allocated range.
3515           */
3516          assert(csp->receive_buffer[csp->receive_buffer_size] == '\0');
3517
3518          /*
3519           * Add a trailing zero to let be able to use string operations.
3520           * XXX: do we still need this with filter_popups gone?
3521           */
3522          assert(len <= csp->receive_buffer_size);
3523          csp->receive_buffer[len] = '\0';
3524
3525          /*
3526           * Normally, this would indicate that we've read
3527           * as much as the server has sent us and we can
3528           * close the client connection.  However, Microsoft
3529           * in its wisdom has released IIS/5 with a bug that
3530           * prevents it from sending the trailing \r\n in
3531           * a 302 redirect header (and possibly other headers).
3532           * To work around this if we've haven't parsed
3533           * a full header we'll append a trailing \r\n
3534           * and see if this now generates a valid one.
3535           *
3536           * This hack shouldn't have any impacts.  If we've
3537           * already transmitted the header or if this is a
3538           * SSL connection, then we won't bother with this
3539           * hack.  So we only work on partially received
3540           * headers.  If we append a \r\n and this still
3541           * doesn't generate a valid header, then we won't
3542           * transmit anything to the client.
3543           */
3544          if (len == 0)
3545          {
3546
3547             if (server_body || (http->ssl
3548 #ifdef FEATURE_HTTPS_INSPECTION
3549                   && use_ssl_tunnel
3550 #endif
3551                ))
3552             {
3553                /*
3554                 * If we have been buffering up the document,
3555                 * now is the time to apply content modification
3556                 * and send the result to the client.
3557                 */
3558                if (buffer_and_filter_content)
3559                {
3560                   p = execute_content_filters(csp);
3561                   /*
3562                    * If content filtering fails, use the original
3563                    * buffer and length.
3564                    * (see p != NULL ? p : csp->iob->cur below)
3565                    */
3566                   if (NULL == p)
3567                   {
3568                      csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
3569                   }
3570 #ifdef FEATURE_COMPRESSION
3571                   else if ((csp->flags & CSP_FLAG_CLIENT_SUPPORTS_DEFLATE)
3572                      && (csp->content_length > LOWER_LENGTH_LIMIT_FOR_COMPRESSION))
3573                   {
3574                      char *compressed_content = compress_buffer(p,
3575                         (size_t *)&csp->content_length, csp->config->compression_level);
3576                      if (compressed_content != NULL)
3577                      {
3578                         freez(p);
3579                         p = compressed_content;
3580                         csp->flags |= CSP_FLAG_BUFFERED_CONTENT_DEFLATED;
3581                      }
3582                   }
3583 #endif
3584
3585                   if (JB_ERR_OK != update_server_headers(csp))
3586                   {
3587                      log_error(LOG_LEVEL_FATAL,
3588                         "Failed to update server headers. after filtering.");
3589                   }
3590
3591                   hdr = list_to_text(csp->headers);
3592                   if (hdr == NULL)
3593                   {
3594                      /* FIXME Should handle error properly */
3595                      log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
3596                   }
3597
3598 #ifdef FEATURE_HTTPS_INSPECTION
3599                   /*
3600                    * Sending data with standard or secured connection (HTTP/HTTPS)
3601                    */
3602                   if (client_use_ssl(csp))
3603                   {
3604                      if ((ssl_send_data_delayed(&(csp->ssl_client_attr),
3605                               (const unsigned char *)hdr, strlen(hdr),
3606                               get_write_delay(csp)) < 0)
3607                         || (ssl_send_data_delayed(&(csp->ssl_client_attr),
3608                               (const unsigned char *) ((p != NULL) ? p : csp->iob->cur),
3609                               csp->content_length, get_write_delay(csp)) < 0))
3610                      {
3611                         log_error(LOG_LEVEL_ERROR,
3612                            "Failed to send the modified content to the client over TLS");
3613                         freez(hdr);
3614                         freez(p);
3615                         mark_server_socket_tainted(csp);
3616                         close_client_and_server_ssl_connections(csp);
3617                         return;
3618                      }
3619                   }
3620                   else
3621 #endif /* def FEATURE_HTTPS_INSPECTION */
3622                   {
3623                      if (write_socket_delayed(csp->cfd, hdr, strlen(hdr), write_delay)
3624                       || write_socket_delayed(csp->cfd, ((p != NULL) ? p : csp->iob->cur),
3625                          (size_t)csp->content_length, write_delay))
3626                      {
3627                         log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
3628                         freez(hdr);
3629                         freez(p);
3630                         mark_server_socket_tainted(csp);
3631                         return;
3632                      }
3633                   }
3634
3635                   freez(hdr);
3636                   freez(p);
3637                }
3638
3639                break; /* "game over, man" */
3640             }
3641
3642             /*
3643              * This is not the body, so let's pretend the server just sent
3644              * us a blank line.
3645              */
3646             snprintf(csp->receive_buffer, csp->receive_buffer_size, "\r\n");
3647             len = (int)strlen(csp->receive_buffer);
3648
3649             /*
3650              * Now, let the normal header parsing algorithm below do its
3651              * job.  If it fails, we'll exit instead of continuing.
3652              */
3653
3654             ms_iis5_hack = 1;
3655          }
3656
3657          /*
3658           * If we're in the body of the server document, just write it to
3659           * the client, unless we need to buffer the body for later
3660           * content-filtering.
3661           */
3662          if (server_body || (http->ssl
3663 #ifdef FEATURE_HTTPS_INSPECTION
3664                && use_ssl_tunnel
3665 #endif
3666             ))
3667          {
3668             if (buffer_and_filter_content)
3669             {
3670                /*
3671                 * If there is no memory left for buffering the content, or the buffer limit
3672                 * has been reached, switch to non-filtering mode, i.e. make & write the
3673                 * header, flush the iob and buf, and get out of the way.
3674                 */
3675                if (add_to_iob(csp->iob, csp->config->buffer_limit, csp->receive_buffer, len))
3676                {
3677                   size_t hdrlen;
3678                   long flushed;
3679
3680                   log_error(LOG_LEVEL_INFO,
3681                      "Flushing header and buffers. Stepping back from filtering.");
3682
3683                   hdr = list_to_text(csp->headers);
3684                   if (hdr == NULL)
3685                   {
3686                      /*
3687                       * Memory is too tight to even generate the header.
3688                       * Send our static "Out-of-memory" page.
3689                       */
3690                      log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
3691                      rsp = cgi_error_memory();
3692                      send_crunch_response(csp, rsp);
3693                      mark_server_socket_tainted(csp);
3694 #ifdef FEATURE_HTTPS_INSPECTION
3695                      close_client_and_server_ssl_connections(csp);
3696 #endif
3697                      return;
3698                   }
3699                   hdrlen = strlen(hdr);
3700
3701 #ifdef FEATURE_HTTPS_INSPECTION
3702                   /*
3703                    * Sending data with standard or secured connection (HTTP/HTTPS)
3704                    */
3705                   if (client_use_ssl(csp))
3706                   {
3707                      if ((ssl_send_data_delayed(&(csp->ssl_client_attr),
3708                              (const unsigned char *)hdr, hdrlen, get_write_delay(csp)) < 0)
3709                         || ((flushed = ssl_flush_socket(&(csp->ssl_client_attr),
3710                                 csp->iob)) < 0)
3711                         || (ssl_send_data_delayed(&(csp->ssl_client_attr),
3712                               (const unsigned char *)csp->receive_buffer, (size_t)len,
3713                               get_write_delay(csp)) < 0))
3714                      {
3715                         log_error(LOG_LEVEL_CONNECT,
3716                            "Flush header and buffers to client failed");
3717                         freez(hdr);
3718                         mark_server_socket_tainted(csp);
3719                         close_client_and_server_ssl_connections(csp);
3720                         return;
3721                      }
3722                   }
3723                   else
3724 #endif /* def FEATURE_HTTPS_INSPECTION */
3725                   {
3726                      if (write_socket_delayed(csp->cfd, hdr, hdrlen, write_delay)
3727                       || ((flushed = flush_iob(csp->cfd, csp->iob, write_delay)) < 0)
3728                       || write_socket_delayed(csp->cfd, csp->receive_buffer, (size_t)len,
3729                             write_delay))
3730                      {
3731                         log_error(LOG_LEVEL_CONNECT,
3732                            "Flush header and buffers to client failed: %E");
3733                         freez(hdr);
3734                         mark_server_socket_tainted(csp);
3735                         return;
3736                      }
3737                   }
3738
3739                   /*
3740                    * Reset the byte_count to the amount of bytes
3741                    * we just flushed. len will be added a few lines below,
3742                    * hdrlen doesn't matter for LOG_LEVEL_CLF.
3743                    */
3744                   byte_count = (unsigned long long)flushed;
3745                   freez(hdr);
3746                   buffer_and_filter_content = 0;
3747                   server_body = 1;
3748                }
3749             }
3750             else
3751             {
3752 #ifdef FEATURE_HTTPS_INSPECTION
3753                /*
3754                 * Sending data with standard or secured connection (HTTP/HTTPS)
3755                 */
3756                if (client_use_ssl(csp))
3757                {
3758                   ret = ssl_send_data_delayed(&(csp->ssl_client_attr),
3759                      (const unsigned char *)csp->receive_buffer, (size_t)len,
3760                      get_write_delay(csp));
3761                   if (ret < 0)
3762                   {
3763                      log_error(LOG_LEVEL_ERROR,
3764                         "Sending data to client failed");
3765                      mark_server_socket_tainted(csp);
3766                      close_client_and_server_ssl_connections(csp);
3767                      return;
3768                   }
3769                }
3770                else
3771 #endif /* def FEATURE_HTTPS_INSPECTION */
3772                {
3773                   if (write_socket_delayed(csp->cfd, csp->receive_buffer,
3774                         (size_t)len, write_delay))
3775                   {
3776                      log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
3777                      mark_server_socket_tainted(csp);
3778                      return;
3779                   }
3780                }
3781             }
3782             byte_count += (unsigned long long)len;
3783             continue;
3784          }
3785          else
3786          {
3787             /*
3788              * We're still looking for the end of the server's header.
3789              * Buffer up the data we just read.  If that fails, there's
3790              * little we can do but send our static out-of-memory page.
3791              */
3792             if (add_to_iob(csp->iob, csp->config->buffer_limit, csp->receive_buffer, len))
3793             {
3794                log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
3795                rsp = cgi_error_memory();
3796                send_crunch_response(csp, rsp);
3797                mark_server_socket_tainted(csp);
3798 #ifdef FEATURE_HTTPS_INSPECTION
3799                close_client_and_server_ssl_connections(csp);
3800 #endif
3801                return;
3802             }
3803
3804             /* Convert iob into something sed() can digest */
3805             if (JB_ERR_PARSE == get_server_headers(csp))
3806             {
3807                if (ms_iis5_hack)
3808                {
3809                   /*
3810                    * Well, we tried our MS IIS/5 hack and it didn't work.
3811                    * The header is incomplete and there isn't anything
3812                    * we can do about it.
3813                    */
3814                   log_error(LOG_LEVEL_ERROR, "Invalid server headers. "
3815                      "Applying the MS IIS5 hack didn't help.");
3816                   log_error(LOG_LEVEL_CLF,
3817                      "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
3818 #ifdef FEATURE_HTTPS_INSPECTION
3819                   /*
3820                    * Sending data with standard or secured connection (HTTP/HTTPS)
3821                    */
3822                   if (client_use_ssl(csp))
3823                   {
3824                      ssl_send_data_delayed(&(csp->ssl_client_attr),
3825                         (const unsigned char *)INVALID_SERVER_HEADERS_RESPONSE,