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