1 const char filters_rcs[] = "$Id: filters.c,v 1.98 2008/01/04 17:43:45 fabiankeil Exp $";
2 /*********************************************************************
4 * File : $Source: /cvsroot/ijbswa/current/filters.c,v $
6 * Purpose : Declares functions to parse/crunch headers and pages.
7 * Functions declared include:
8 * `acl_addr', `add_stats', `block_acl', `block_imageurl',
9 * `block_url', `url_actions', `domain_split',
10 * `filter_popups', `forward_url', 'redirect_url',
11 * `ij_untrusted_url', `intercept_url', `pcrs_filter_respose',
12 * `ijb_send_banner', `trust_url', `gif_deanimate_response',
13 * `jpeg_inspect_response', `execute_single_pcrs_command',
14 * `rewrite_url', `get_last_url'
16 * Copyright : Written by and Copyright (C) 2001, 2004-2007 the SourceForge
17 * Privoxy team. http://www.privoxy.org/
19 * Based on the Internet Junkbuster originally written
20 * by and Copyright (C) 1997 Anonymous Coders and
21 * Junkbusters Corporation. http://www.junkbusters.com
23 * This program is free software; you can redistribute it
24 * and/or modify it under the terms of the GNU General
25 * Public License as published by the Free Software
26 * Foundation; either version 2 of the License, or (at
27 * your option) any later version.
29 * This program is distributed in the hope that it will
30 * be useful, but WITHOUT ANY WARRANTY; without even the
31 * implied warranty of MERCHANTABILITY or FITNESS FOR A
32 * PARTICULAR PURPOSE. See the GNU General Public
33 * License for more details.
35 * The GNU General Public License should be included with
36 * this file. If not, you can view it at
37 * http://www.gnu.org/copyleft/gpl.html
38 * or write to the Free Software Foundation, Inc., 59
39 * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
43 * Revision 1.98 2008/01/04 17:43:45 fabiankeil
44 * Improve the warning messages that get logged if the action files
45 * "enable" filters but no filters of that type have been loaded.
47 * Revision 1.97 2007/11/30 15:37:03 fabiankeil
48 * Use freez instead of free.
50 * Revision 1.96 2007/10/19 16:53:28 fabiankeil
51 * Add helper function to check if any content filters are enabled.
53 * Revision 1.95 2007/10/17 19:31:20 fabiankeil
54 * Omitting the zero chunk that ends the chunk transfer encoding seems
55 * to be the new black. Log the problem and continue filtering anyway.
57 * Revision 1.94 2007/09/29 13:20:20 fabiankeil
58 * Remove two redundant and one useless log messages.
60 * Revision 1.93 2007/09/29 10:21:16 fabiankeil
61 * - Move get_filter_function() from jcc.c to filters.c
62 * so the filter functions can be static.
63 * - Don't bother filtering body-less responses.
65 * Revision 1.92 2007/09/28 16:38:55 fabiankeil
66 * - Execute content filters through execute_content_filter().
67 * - Add prepare_for_filtering() so filter functions don't have to
68 * care about de-chunking and decompression. As a side effect this enables
69 * decompression for gif_deanimate_response() and jpeg_inspect_response().
70 * - Change remove_chunked_transfer_coding()'s return type to jb_err.
71 * Some clowns feel like chunking empty responses in which case
72 * (size == 0) is valid but previously would be interpreted as error.
74 * Revision 1.91 2007/09/02 15:31:20 fabiankeil
75 * Move match_portlist() from filter.c to urlmatch.c.
76 * It's used for url matching, not for filtering.
78 * Revision 1.90 2007/09/02 12:44:17 fabiankeil
79 * Remove newline at the end of a log_error() message.
81 * Revision 1.89 2007/08/05 13:42:23 fabiankeil
82 * #1763173 from Stefan Huehner: declare some more functions static.
84 * Revision 1.88 2007/06/01 16:41:11 fabiankeil
85 * Add forward-override{} to change the forwarding settings through
86 * action sections. This is mainly interesting to forward different
87 * clients differently (for example based on User-Agent or request
90 * Revision 1.87 2007/04/30 15:53:10 fabiankeil
91 * Make sure filters with dynamic jobs actually use them.
93 * Revision 1.86 2007/04/30 15:03:28 fabiankeil
94 * - Introduce dynamic pcrs jobs that can resolve variables.
95 * - Don't run redirect functions more than once,
96 * unless they are activated more than once.
98 * Revision 1.85 2007/03/21 12:24:47 fabiankeil
99 * - Log the content size after decompression in decompress_iob()
100 * instead of pcrs_filter_response().
102 * Revision 1.84 2007/03/20 15:16:34 fabiankeil
103 * Use dedicated header filter actions instead of abusing "filter".
104 * Replace "filter-client-headers" and "filter-client-headers"
105 * with "server-header-filter" and "client-header-filter".
107 * Revision 1.83 2007/03/17 15:20:05 fabiankeil
108 * New config option: enforce-blocks.
110 * Revision 1.82 2007/03/13 11:28:43 fabiankeil
111 * - Fix port handling in acl_addr() and use a temporary acl spec
112 * copy so error messages don't contain a truncated version.
113 * - Log size of iob before and after decompression.
115 * Revision 1.81 2007/03/05 14:40:53 fabiankeil
116 * - Cosmetical changes for LOG_LEVEL_RE_FILTER messages.
117 * - Hide the "Go there anyway" link for blocked CONNECT
118 * requests where going there anyway doesn't work anyway.
120 * Revision 1.80 2007/02/07 10:55:20 fabiankeil
121 * - Save the reason for generating http_responses.
122 * - Block (+block) with status code 403 instead of 404.
123 * - Use a different kludge to remember a failed decompression.
125 * Revision 1.79 2007/01/31 16:21:38 fabiankeil
126 * Search for Max-Forwards headers case-insensitive,
127 * don't generate the "501 unsupported" message for invalid
128 * Max-Forwards values and don't increase negative ones.
130 * Revision 1.78 2007/01/28 13:41:18 fabiankeil
131 * - Add HEAD support to finish_http_response.
132 * - Add error favicon to internal HTML error messages.
134 * Revision 1.77 2007/01/12 15:36:44 fabiankeil
135 * Mark *csp as immutable for is_untrusted_url()
136 * and is_imageurl(). Closes FR 1237736.
138 * Revision 1.76 2007/01/01 19:36:37 fabiankeil
139 * Integrate a modified version of Wil Mahan's
140 * zlib patch (PR #895531).
142 * Revision 1.75 2006/12/29 18:30:46 fabiankeil
143 * Fixed gcc43 conversion warnings,
144 * changed sprintf calls to snprintf.
146 * Revision 1.74 2006/12/24 17:37:38 fabiankeil
147 * Adjust comment in pcrs_filter_response()
148 * to recent pcrs changes. Hohoho.
150 * Revision 1.73 2006/12/23 16:01:02 fabiankeil
151 * Don't crash if pcre returns an error code
152 * that pcrs didn't expect. Fixes BR 1621173.
154 * Revision 1.72 2006/12/22 18:52:53 fabiankeil
155 * Modified is_untrusted_url to complain in case of
156 * write errors and to give a reason when adding new
157 * entries to the trustfile. Closes FR 1097611.
159 * Revision 1.71 2006/12/22 14:24:52 fabiankeil
160 * Skip empty filter files in pcrs_filter_response,
161 * but don't ignore the ones that come afterwards.
162 * Fixes parts of BR 1619208.
164 * Revision 1.70 2006/12/09 13:33:15 fabiankeil
165 * Added some sanity checks for get_last_url().
166 * Fixed possible segfault caused by my last commit.
168 * Revision 1.69 2006/12/08 12:39:13 fabiankeil
169 * Let get_last_url() catch https URLs as well.
171 * Revision 1.68 2006/12/05 14:45:48 fabiankeil
172 * Make sure get_last_url() behaves like advertised
173 * and fast-redirects{} can be combined with redirect{}.
175 * Revision 1.67 2006/11/28 15:19:43 fabiankeil
176 * Implemented +redirect{s@foo@bar@} to generate
177 * a redirect based on a rewritten version of the
180 * Revision 1.66 2006/09/23 13:26:38 roro
181 * Replace TABs by spaces in source code.
183 * Revision 1.65 2006/09/21 12:54:43 fabiankeil
184 * Fix +redirect{}. Didn't work with -fast-redirects.
186 * Revision 1.64 2006/08/31 10:55:49 fabiankeil
187 * Block requests for untrusted URLs with status
188 * code 403 instead of 200.
190 * Revision 1.63 2006/08/31 10:11:28 fabiankeil
191 * Don't free p which is still in use and will be later
192 * freed by free_map(). Don't claim the referrer is unknown
193 * when the client didn't set one.
195 * Revision 1.62 2006/08/14 00:27:47 david__schmidt
196 * Feature request 595948: Re-Filter logging in single line
198 * Revision 1.61 2006/08/03 02:46:41 david__schmidt
199 * Incorporate Fabian Keil's patch work:
\rhttp://www.fabiankeil.de/sourcecode/privoxy/
201 * Revision 1.60 2006/07/18 14:48:46 david__schmidt
202 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
203 * with what was really the latest development (the v_3_0_branch branch)
205 * Revision 1.58.2.9 2006/01/29 23:10:56 david__schmidt
206 * Multiple filter file support
208 * Revision 1.58.2.8 2005/05/07 21:50:55 david__schmidt
209 * A few memory leaks plugged (mostly on error paths)
211 * Revision 1.58.2.7 2004/10/03 12:53:32 david__schmidt
212 * Add the ability to check jpeg images for invalid
213 * lengths of comment blocks. Defensive strategy
214 * against the exploit:
215 * Microsoft Security Bulletin MS04-028
216 * Buffer Overrun in JPEG Processing (GDI+) Could
217 * Allow Code Execution (833987)
218 * Enabled with +inspect-jpegs in actions files.
220 * Revision 1.58.2.6 2003/12/06 22:18:27 gliptak
221 * Correcting compile problem with FEATURE_IMAGE_BLOCKING
223 * Revision 1.58.2.5 2003/11/11 13:10:31 oes
224 * Fixed bug #839859: "See why" link URL now gets url-encoded.
226 * Revision 1.58.2.4 2003/02/28 12:52:45 oes
229 * Revision 1.58.2.3 2002/09/25 14:51:51 oes
230 * Added basic support for OPTIONS and TRACE HTTP methods:
231 * New function direct_response which handles OPTIONS and
232 * TRACE requests whose Max-Forwards header field is zero.
234 * Revision 1.58.2.2 2002/08/01 17:18:28 oes
235 * Fixed BR 537651 / SR 579724 (MSIE image detect improper for IE/Mac)
237 * Revision 1.58.2.1 2002/07/26 15:18:53 oes
238 * - Bugfix: Executing a filters without jobs no longer results in
239 * turing off *all* filters.
240 * - Security fix: Malicious web servers can't cause a seg fault
241 * through bogus chunk sizes anymore
243 * Revision 1.58 2002/04/24 02:11:17 oes
244 * Jon's multiple AF patch: url_actions now evaluates rules
247 * Revision 1.57 2002/04/08 20:38:34 swa
250 * Revision 1.56 2002/04/05 15:51:24 oes
251 * - bugfix: error-pages now get correct request protocol
252 * - fix for invalid HTML in trust info
254 * Revision 1.55 2002/04/02 16:13:51 oes
255 * Fix: No "Go there anyway" for SSL
257 * Revision 1.54 2002/04/02 14:55:56 oes
258 * Bugfix: is_untrusted_url() now depends on FEATURE_TRUST, not FEATURE_COOKIE_JAR
260 * Revision 1.53 2002/03/26 22:29:54 swa
261 * we have a new homepage!
263 * Revision 1.52 2002/03/24 16:35:57 jongfoster
266 * Revision 1.51 2002/03/24 15:23:33 jongfoster
269 * Revision 1.50 2002/03/24 13:25:43 swa
270 * name change related issues
272 * Revision 1.49 2002/03/16 20:29:14 oes
275 * Revision 1.48 2002/03/13 20:25:34 oes
276 * Better logging for content filters
278 * Revision 1.47 2002/03/13 00:30:52 jongfoster
280 * Added option of always sending redirect for imageblock,
281 * currently disabled with #if 0.
283 * Revision 1.46 2002/03/12 01:42:49 oes
284 * Introduced modular filters
286 * Revision 1.45 2002/03/08 16:47:50 oes
287 * Added choice beween GIF and PNG built-in images
289 * Revision 1.44 2002/03/07 03:49:31 oes
290 * - Fixed compiler warnings etc
291 * - Changed built-in images from GIF to PNG
292 * (with regard to Unisys patent issue)
293 * - Added a 4x4 pattern PNG which is less intrusive
294 * than the logo but also clearly marks the deleted banners
296 * Revision 1.43 2002/01/22 23:51:59 jongfoster
297 * Replacing strsav() with the safer string_append().
299 * Adding missing html_encode() to error message generators. Where encoded
300 * and unencoded versions of a string were provided, removing the unencoded
303 * Revision 1.42 2002/01/17 21:00:32 jongfoster
304 * Moving all our URL and URL pattern parsing code to urlmatch.c.
306 * Using a single, simple url_match(pattern,url) function - rather than
307 * the 3-line match routine which was repeated all over the place.
309 * Renaming free_url to free_url_spec, since it frees a struct url_spec.
311 * Using parse_http_url() to parse URLs without faking a HTTP
312 * request line for parse_http_request().
314 * Revision 1.41 2001/11/13 00:14:07 jongfoster
315 * Fixing stupid bug now I've figured out what || means.
316 * (It always returns 0 or 1, not one of it's paramaters.)
318 * Revision 1.40 2001/10/26 17:37:55 oes
319 * - Re-enabled Netscape 200/404 bug workaround in block_url():
320 * - Removed OS/2 special case
321 * - Made block_url() independant from sed() having been run
322 * - Made trust_url independant from sed() having been run
323 * - Made is_imageurl independant from sed() having been run.
324 * It now checks User-Agent: and Accept: by itself.
327 * Revision 1.39 2001/10/25 03:40:48 david__schmidt
328 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
329 * threads to call select() simultaneously. So, it's time to do a real, live,
330 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
331 * (native). Both versions will work, but using __OS2__ offers multi-threading.
333 * Revision 1.38 2001/10/23 21:32:33 jongfoster
334 * Adding error-checking to selected functions
336 * Revision 1.37 2001/10/22 15:33:56 david__schmidt
337 * Special-cased OS/2 out of the Netscape-abort-on-404-in-js problem in
338 * filters.c. Added a FIXME in front of the offending code. I'll gladly
339 * put in a better/more robust fix for all parties if one is presented...
340 * It seems that just returning 200 instead of 404 would pretty much fix
341 * it for everyone, but I don't know all the history of the problem.
343 * Revision 1.36 2001/10/10 16:44:16 oes
344 * Added match_portlist function
346 * Revision 1.35 2001/10/07 15:41:23 oes
347 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
349 * New function remove_chunked_transfer_coding that strips chunked
350 * transfer coding to plain and is called by pcrs_filter_response
351 * and gif_deanimate_response if neccessary
353 * Improved handling of zero-change re_filter runs
355 * pcrs_filter_response and gif_deanimate_response now remove
356 * chunked transfer codeing before processing the body.
358 * Revision 1.34 2001/09/20 15:49:36 steudten
360 * Fix BUG: Change int size to size_t size in pcrs_filter_response().
361 * See cgi.c fill_template().
363 * Revision 1.33 2001/09/16 17:05:14 jongfoster
364 * Removing unused #include showarg.h
366 * Revision 1.32 2001/09/16 13:21:27 jongfoster
367 * Changes to use new list functions.
369 * Revision 1.31 2001/09/16 11:38:02 jongfoster
370 * Splitting fill_template() into 2 functions:
371 * template_load() loads the file
372 * template_fill() performs the PCRS regexps.
373 * This is because the CGI edit interface has a "table row"
374 * template which is used many times in the page - this
375 * change means it's only loaded from disk once.
377 * Revision 1.30 2001/09/16 11:00:10 jongfoster
378 * New function alloc_http_response, for symmetry with free_http_response
380 * Revision 1.29 2001/09/13 23:32:40 jongfoster
381 * Moving image data to cgi.c rather than cgi.h
382 * Fixing a GPF under Win32 (and any other OS that protects global
383 * constants from being written to).
385 * Revision 1.28 2001/09/10 10:18:51 oes
386 * Silenced compiler warnings
388 * Revision 1.27 2001/08/05 16:06:20 jongfoster
389 * Modifiying "struct map" so that there are now separate header and
390 * "map_entry" structures. This means that functions which modify a
391 * map no longer need to return a pointer to the modified map.
392 * Also, it no longer reverses the order of the entries (which may be
393 * important with some advanced template substitutions).
395 * Revision 1.26 2001/07/30 22:08:36 jongfoster
396 * Tidying up #defines:
397 * - All feature #defines are now of the form FEATURE_xxx
398 * - Permanently turned off WIN_GUI_EDIT
399 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
401 * Revision 1.25 2001/07/26 10:09:46 oes
402 * Made browser detection a little less naive
404 * Revision 1.24 2001/07/25 17:22:51 oes
405 * Added workaround for Netscape bug that prevents display of page when loading a component fails.
407 * Revision 1.23 2001/07/23 13:40:12 oes
408 * Fixed bug that caused document body to be dropped when pcrs joblist was empty.
410 * Revision 1.22 2001/07/18 12:29:34 oes
411 * - Made gif_deanimate_response respect
412 * csp->action->string[ACTION_STRING_DEANIMATE]
413 * - Logging cosmetics
415 * Revision 1.21 2001/07/13 13:59:53 oes
416 * - Introduced gif_deanimate_response which shares the
417 * generic content modification interface of pcrs_filter_response
418 * and acts as a wrapper to deanimate.c:gif_deanimate()
419 * - Renamed re_process_buffer to pcrs_filter_response
420 * - pcrs_filter_response now returns NULL on failiure
421 * - Removed all #ifdef PCRS
423 * Revision 1.20 2001/07/01 17:01:04 oes
424 * Added comments and missing return statement in is_untrusted_url()
426 * Revision 1.19 2001/06/29 21:45:41 oes
427 * Indentation, CRLF->LF, Tab-> Space
429 * Revision 1.18 2001/06/29 13:27:38 oes
430 * - Cleaned up, renamed and reorderd functions
431 * and improved comments
434 * - Ported to CGI platform. Now delivers
435 * http_response or NULL
436 * - Unified HTML and GIF generation (moved image detection
437 * and GIF generation here from jcc.c:chat())
438 * - Fixed HTTP status to:
439 * - 403 (Forbidden) for the "blocked" HTML message
440 * - 200 (OK) for GIF answers
441 * - 302 (Redirect) for redirect to GIF
444 * - Ported to CGI platform. Now delivers
445 * http_response or NULL
446 * - Separated detection of untrusted URL into
447 * (bool)is_untrusted_url
448 * - Added enforcement of untrusted requests
450 * - Moved redirect_url() from cgi.c to here
451 * and ported it to the CGI platform
453 * - Removed logentry from cancelled commit
455 * Revision 1.17 2001/06/09 10:55:28 jongfoster
456 * Changing BUFSIZ ==> BUFFER_SIZE
458 * Revision 1.16 2001/06/07 23:10:26 jongfoster
459 * Allowing unanchored domain patterns to back off and retry
460 * if they partially match. Optimized right-anchored patterns.
461 * Moving ACL and forward files into config file.
462 * Replacing struct gateway with struct forward_spec
464 * Revision 1.15 2001/06/03 19:12:00 oes
465 * extracted-CGI relevant stuff
467 * Revision 1.14 2001/06/01 10:30:55 oes
468 * Added optional left-anchoring to domaincmp
470 * Revision 1.13 2001/05/31 21:21:30 jongfoster
471 * Permissionsfile / actions file changes:
472 * - Changed "permission" to "action" throughout
473 * - changes to file format to allow string parameters
474 * - Moved helper functions to actions.c
476 * Revision 1.12 2001/05/31 17:35:20 oes
478 * - Enhanced domain part globbing with infix and prefix asterisk
479 * matching and optional unanchored operation
481 * Revision 1.11 2001/05/29 11:53:23 oes
482 * "See why" link added to "blocked" page
484 * Revision 1.10 2001/05/29 09:50:24 jongfoster
485 * Unified blocklist/imagelist/permissionslist.
486 * File format is still under discussion, but the internal changes
489 * Also modified interceptor behaviour:
490 * - We now intercept all URLs beginning with one of the following
491 * prefixes (and *only* these prefixes):
493 * * http://ijbswa.sf.net/config/
494 * * http://ijbswa.sourceforge.net/config/
495 * - New interceptors "home page" - go to http://i.j.b/ to see it.
496 * - Internal changes so that intercepted and fast redirect pages
497 * are not replaced with an image.
498 * - Interceptors now have the option to send a binary page direct
499 * to the client. (i.e. ijb-send-banner uses this)
500 * - Implemented show-url-info interceptor. (Which is why I needed
501 * the above interceptors changes - a typical URL is
502 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
503 * The previous mechanism would not have intercepted that, and
504 * if it had been intercepted then it then it would have replaced
507 * Revision 1.9 2001/05/27 22:17:04 oes
509 * - re_process_buffer no longer writes the modified buffer
510 * to the client, which was very ugly. It now returns the
511 * buffer, which it is then written by chat.
513 * - content_length now adjusts the Content-Length: header
514 * for modified documents rather than crunch()ing it.
515 * (Length info in csp->content_length, which is 0 for
516 * unmodified documents)
518 * - For this to work, sed() is called twice when filtering.
520 * Revision 1.8 2001/05/26 17:13:28 jongfoster
521 * Filled in a function comment.
523 * Revision 1.7 2001/05/26 15:26:15 jongfoster
524 * ACL feature now provides more security by immediately dropping
525 * connections from untrusted hosts.
527 * Revision 1.6 2001/05/26 00:28:36 jongfoster
528 * Automatic reloading of config file.
529 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
530 * Most of the global variables have been moved to a new
531 * struct configuration_spec, accessed through csp->config->globalname
532 * Most of the globals remaining are used by the Win32 GUI.
534 * Revision 1.5 2001/05/25 22:34:30 jongfoster
537 * Revision 1.4 2001/05/22 18:46:04 oes
539 * - Enabled filtering banners by size rather than URL
540 * by adding patterns that replace all standard banner
541 * sizes with the "Junkbuster" gif to the re_filterfile
543 * - Enabled filtering WebBugs by providing a pattern
544 * which kills all 1x1 images
546 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
547 * which is selected by the (nonstandard and therefore
548 * capital) letter 'U' in the option string.
549 * It causes the quantifiers to be ungreedy by default.
550 * Appending a ? turns back to greedy (!).
552 * - Added a new interceptor ijb-send-banner, which
553 * sends back the "Junkbuster" gif. Without imagelist or
554 * MSIE detection support, or if tinygif = 1, or the
555 * URL isn't recognized as an imageurl, a lame HTML
556 * explanation is sent instead.
558 * - Added new feature, which permits blocking remote
559 * script redirects and firing back a local redirect
561 * The feature is conditionally compiled, i.e. it
562 * can be disabled with --disable-fast-redirects,
563 * plus it must be activated by a "fast-redirects"
564 * line in the config file, has its own log level
565 * and of course wants to be displayed by show-proxy-args
566 * Note: Boy, all the #ifdefs in 1001 locations and
567 * all the fumbling with configure.in and acconfig.h
568 * were *way* more work than the feature itself :-(
570 * - Because a generic redirect template was needed for
571 * this, tinygif = 3 now uses the same.
573 * - Moved GIFs, and other static HTTP response templates
578 * - Removed some >400 CRs again (Jon, you really worked
581 * Revision 1.3 2001/05/20 16:44:47 jongfoster
582 * Removing last hardcoded Junkbusters.com URLs.
584 * Revision 1.2 2001/05/20 01:21:20 jongfoster
585 * Version 2.9.4 checkin.
586 * - Merged popupfile and cookiefile, and added control over PCRS
587 * filtering, in new "permissionsfile".
588 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
589 * file error you now get a message box (in the Win32 GUI) rather
590 * than the program exiting with no explanation.
591 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
593 * - Removed tabs from "config"
594 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
595 * - Bumped up version number.
597 * Revision 1.1.1.1 2001/05/15 13:58:52 oes
598 * Initial import of version 2.9.3 source tree
601 *********************************************************************/
607 #include <sys/types.h>
616 #endif /* ndef __OS2__ */
617 #include <netinet/in.h>
619 #include <winsock2.h>
620 #endif /* ndef _WIN32 */
624 #endif /* def __OS2__ */
632 #include "jbsockets.h"
633 #include "miscutil.h"
637 #include "deanimate.h"
638 #include "urlmatch.h"
645 const char filters_h_rcs[] = FILTERS_H_VERSION;
647 /* Fix a problem with Solaris. There should be no effect on other
649 * Solaris's isspace() is a macro which uses it's argument directly
650 * as an array index. Therefore we need to make sure that high-bit
651 * characters generate +ve values, and ideally we also want to make
652 * the argument match the declared parameter type of "int".
654 #define ijb_isdigit(__X) isdigit((int)(unsigned char)(__X))
656 static jb_err remove_chunked_transfer_coding(char *buffer, size_t *size);
657 static jb_err prepare_for_filtering(struct client_state *csp);
660 /*********************************************************************
662 * Function : block_acl
664 * Description : Block this request?
665 * Decide yes or no based on ACL file.
668 * 1 : dst = The proxy or gateway address this is going to.
669 * Or NULL to check all possible targets.
670 * 2 : csp = Current client state (buffers, headers, etc...)
671 * Also includes the client IP address.
673 * Returns : 0 = FALSE (don't block) and 1 = TRUE (do block)
675 *********************************************************************/
676 int block_acl(struct access_control_addr *dst, struct client_state *csp)
678 struct access_control_list *acl = csp->config->acl;
680 /* if not using an access control list, then permit the connection */
686 /* search the list */
689 if ((csp->ip_addr_long & acl->src->mask) == acl->src->addr)
693 /* Just want to check if they have any access */
694 if (acl->action == ACL_PERMIT)
699 else if ( ((dst->addr & acl->dst->mask) == acl->dst->addr)
700 && ((dst->port == acl->dst->port) || (acl->dst->port == 0)))
702 if (acl->action == ACL_PERMIT)
720 /*********************************************************************
722 * Function : acl_addr
724 * Description : Called from `load_config' to parse an ACL address.
727 * 1 : aspec = String specifying ACL address.
728 * 2 : aca = struct access_control_addr to fill in.
730 * Returns : 0 => Ok, everything else is an error.
732 *********************************************************************/
733 int acl_addr(const char *aspec, struct access_control_addr *aca)
738 char *acl_spec = NULL;
744 * Use a temporary acl spec copy so we can log
745 * the unmodified original in case of parse errors.
747 acl_spec = strdup(aspec);
748 if (acl_spec == NULL)
750 /* XXX: This will be logged as parse error. */
754 if ((p = strchr(acl_spec, '/')) != NULL)
757 if (ijb_isdigit(*p) == 0)
762 masklength = atoi(p);
765 if ((masklength < 0) || (masklength > 32))
771 if ((p = strchr(acl_spec, ':')) != NULL)
776 port = strtol(p, &endptr, 10);
778 if (port <= 0 || port > 65535 || *endptr != '\0')
785 aca->port = (unsigned long)port;
787 aca->addr = ntohl(resolve_hostname_to_ip(acl_spec));
790 if (aca->addr == INADDR_NONE)
792 /* XXX: This will be logged as parse error. */
796 /* build the netmask */
798 for (i=1; i <= masklength ; i++)
800 aca->mask |= (1 << (32 - i));
803 /* now mask off the host portion of the ip address
804 * (i.e. save on the network portion of the address).
806 aca->addr = aca->addr & aca->mask;
811 #endif /* def FEATURE_ACL */
814 /*********************************************************************
816 * Function : block_url
818 * Description : Called from `chat'. Check to see if we need to block this.
821 * 1 : csp = Current client state (buffers, headers, etc...)
823 * Returns : NULL => unblocked, else HTTP block response
825 *********************************************************************/
826 struct http_response *block_url(struct client_state *csp)
828 struct http_response *rsp;
829 const char *new_content_type = NULL;
832 * If it's not blocked, don't block it ;-)
834 if ((csp->action->flags & ACTION_BLOCK) == 0)
838 if (csp->action->flags & ACTION_REDIRECT)
840 log_error(LOG_LEVEL_ERROR, "redirect{} overruled by block.");
843 * Else, prepare a response
845 if (NULL == (rsp = alloc_http_response()))
847 return cgi_error_memory();
851 * If it's an image-url, send back an image or redirect
852 * as specified by the relevant +image action
854 #ifdef FEATURE_IMAGE_BLOCKING
855 if (((csp->action->flags & ACTION_IMAGE_BLOCKER) != 0)
859 /* determine HOW images should be blocked */
860 p = csp->action->string[ACTION_STRING_IMAGE_BLOCKER];
862 if(csp->action->flags & ACTION_HANDLE_AS_EMPTY_DOCUMENT)
864 log_error(LOG_LEVEL_ERROR, "handle-as-empty-document overruled by handle-as-image.");
866 #if 1 /* Two alternative strategies, use this one for now: */
868 /* and handle accordingly: */
869 if ((p == NULL) || (0 == strcmpic(p, "pattern")))
871 rsp->status = strdup("403 Request blocked by Privoxy");
872 if (rsp->status == NULL)
874 free_http_response(rsp);
875 return cgi_error_memory();
877 rsp->body = bindup(image_pattern_data, image_pattern_length);
878 if (rsp->body == NULL)
880 free_http_response(rsp);
881 return cgi_error_memory();
883 rsp->content_length = image_pattern_length;
885 if (enlist_unique_header(rsp->headers, "Content-Type", BUILTIN_IMAGE_MIMETYPE))
887 free_http_response(rsp);
888 return cgi_error_memory();
892 else if (0 == strcmpic(p, "blank"))
894 rsp->status = strdup("403 Request blocked by Privoxy");
895 if (rsp->status == NULL)
897 free_http_response(rsp);
898 return cgi_error_memory();
900 rsp->body = bindup(image_blank_data, image_blank_length);
901 if (rsp->body == NULL)
903 free_http_response(rsp);
904 return cgi_error_memory();
906 rsp->content_length = image_blank_length;
908 if (enlist_unique_header(rsp->headers, "Content-Type", BUILTIN_IMAGE_MIMETYPE))
910 free_http_response(rsp);
911 return cgi_error_memory();
917 rsp->status = strdup("302 Local Redirect from Privoxy");
918 if (rsp->status == NULL)
920 free_http_response(rsp);
921 return cgi_error_memory();
924 if (enlist_unique_header(rsp->headers, "Location", p))
926 free_http_response(rsp);
927 return cgi_error_memory();
931 #else /* Following code is disabled for now */
933 /* and handle accordingly: */
934 if ((p == NULL) || (0 == strcmpic(p, "pattern")))
936 p = CGI_PREFIX "send-banner?type=pattern";
938 else if (0 == strcmpic(p, "blank"))
940 p = CGI_PREFIX "send-banner?type=blank";
942 rsp->status = strdup("302 Local Redirect from Privoxy");
943 if (rsp->status == NULL)
945 free_http_response(rsp);
946 return cgi_error_memory();
949 if (enlist_unique_header(rsp->headers, "Location", p))
951 free_http_response(rsp);
952 return cgi_error_memory();
954 #endif /* Preceeding code is disabled for now */
956 else if(csp->action->flags & ACTION_HANDLE_AS_EMPTY_DOCUMENT)
959 * Send empty document.
961 new_content_type = csp->action->string[ACTION_STRING_CONTENT_TYPE];
964 rsp->body = strdup(" ");
965 rsp->content_length = 1;
967 rsp->status = strdup("403 Request blocked by Privoxy");
968 if (rsp->status == NULL)
970 free_http_response(rsp);
971 return cgi_error_memory();
973 if (new_content_type != 0)
975 log_error(LOG_LEVEL_HEADER, "Overwriting Content-Type with %s", new_content_type);
976 if (enlist_unique_header(rsp->headers, "Content-Type", new_content_type))
978 free_http_response(rsp);
979 return cgi_error_memory();
984 #endif /* def FEATURE_IMAGE_BLOCKING */
987 * Else, generate an HTML "blocked" message:
991 struct map * exports;
995 * Workaround for stupid Netscape bug which prevents
996 * pages from being displayed if loading a referenced
997 * JavaScript or style sheet fails. So make it appear
998 * as if it succeeded.
1000 if ( NULL != (p = get_header_value(csp->headers, "User-Agent:"))
1001 && !strncmpic(p, "mozilla", 7) /* Catch Netscape but */
1002 && !strstr(p, "Gecko") /* save Mozilla, */
1003 && !strstr(p, "compatible") /* MSIE */
1004 && !strstr(p, "Opera")) /* and Opera. */
1006 rsp->status = strdup("200 Request for blocked URL");
1010 rsp->status = strdup("403 Request for blocked URL");
1013 if (rsp->status == NULL)
1015 free_http_response(rsp);
1016 return cgi_error_memory();
1019 exports = default_exports(csp, NULL);
1020 if (exports == NULL)
1022 free_http_response(rsp);
1023 return cgi_error_memory();
1026 #ifdef FEATURE_FORCE_LOAD
1027 err = map(exports, "force-prefix", 1, FORCE_PREFIX, 1);
1029 * Export the force conditional block killer if
1031 * - Privoxy was compiled without FEATURE_FORCE_LOAD, or
1032 * - Privoxy is configured to enforce blocks, or
1033 * - it's a CONNECT request and enforcing wouldn't work anyway.
1035 if ((csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1036 || (0 == strcmpic(csp->http->gpc, "connect")))
1037 #endif /* ndef FEATURE_FORCE_LOAD */
1039 err = map_block_killer(exports, "force-support");
1042 if (!err) err = map(exports, "protocol", 1, csp->http->ssl ? "https://" : "http://", 1);
1043 if (!err) err = map(exports, "hostport", 1, html_encode(csp->http->hostport), 0);
1044 if (!err) err = map(exports, "path", 1, html_encode(csp->http->path), 0);
1045 if (!err) err = map(exports, "path-ue", 1, url_encode(csp->http->path), 0);
1050 free_http_response(rsp);
1051 return cgi_error_memory();
1054 err = template_fill_for_cgi(csp, "blocked", exports, rsp);
1057 free_http_response(rsp);
1058 return cgi_error_memory();
1061 rsp->reason = RSP_REASON_BLOCKED;
1063 return finish_http_response(csp, rsp);
1068 #ifdef FEATURE_TRUST
1069 /*********************************************************************
1071 * Function : trust_url FIXME: I should be called distrust_url
1073 * Description : Calls is_untrusted_url to determine if the URL is trusted
1074 * and if not, returns a HTTP 403 response with a reject message.
1077 * 1 : csp = Current client state (buffers, headers, etc...)
1079 * Returns : NULL => trusted, else http_response.
1081 *********************************************************************/
1082 struct http_response *trust_url(struct client_state *csp)
1084 struct http_response *rsp;
1085 struct map * exports;
1086 char buf[BUFFER_SIZE];
1088 struct url_spec **tl;
1093 * Don't bother to work on trusted URLs
1095 if (!is_untrusted_url(csp))
1101 * Else, prepare a response:
1103 if (NULL == (rsp = alloc_http_response()))
1105 return cgi_error_memory();
1108 rsp->status = strdup("403 Request blocked by Privoxy");
1109 exports = default_exports(csp, NULL);
1110 if (exports == NULL || rsp->status == NULL)
1112 free_http_response(rsp);
1113 return cgi_error_memory();
1117 * Export the protocol, host, port, and referrer information
1119 err = map(exports, "hostport", 1, csp->http->hostport, 1);
1120 if (!err) err = map(exports, "protocol", 1, csp->http->ssl ? "https://" : "http://", 1);
1121 if (!err) err = map(exports, "path", 1, csp->http->path, 1);
1123 if (NULL != (p = get_header_value(csp->headers, "Referer:")))
1125 if (!err) err = map(exports, "referrer", 1, html_encode(p), 0);
1129 if (!err) err = map(exports, "referrer", 1, "none set", 1);
1135 free_http_response(rsp);
1136 return cgi_error_memory();
1140 * Export the trust list
1143 for (tl = csp->config->trust_list; (t = *tl) != NULL ; tl++)
1145 snprintf(buf, sizeof(buf), "<li>%s</li>\n", t->spec);
1146 string_append(&p, buf);
1148 err = map(exports, "trusted-referrers", 1, p, 0);
1153 free_http_response(rsp);
1154 return cgi_error_memory();
1158 * Export the trust info, if available
1160 if (csp->config->trust_info->first)
1162 struct list_entry *l;
1165 for (l = csp->config->trust_info->first; l ; l = l->next)
1167 snprintf(buf, sizeof(buf), "<li> <a href=\"%s\">%s</a><br>\n", l->str, l->str);
1168 string_append(&p, buf);
1170 err = map(exports, "trust-info", 1, p, 0);
1174 err = map_block_killer(exports, "have-trust-info");
1180 free_http_response(rsp);
1181 return cgi_error_memory();
1185 * Export the force conditional block killer if
1187 * - Privoxy was compiled without FEATURE_FORCE_LOAD, or
1188 * - Privoxy is configured to enforce blocks, or
1189 * - it's a CONNECT request and enforcing wouldn't work anyway.
1191 #ifdef FEATURE_FORCE_LOAD
1192 if ((csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1193 || (0 == strcmpic(csp->http->gpc, "connect")))
1195 err = map_block_killer(exports, "force-support");
1199 err = map(exports, "force-prefix", 1, FORCE_PREFIX, 1);
1201 #else /* ifndef FEATURE_FORCE_LOAD */
1202 err = map_block_killer(exports, "force-support");
1203 #endif /* ndef FEATURE_FORCE_LOAD */
1208 free_http_response(rsp);
1209 return cgi_error_memory();
1213 * Build the response
1215 err = template_fill_for_cgi(csp, "untrusted", exports, rsp);
1218 free_http_response(rsp);
1219 return cgi_error_memory();
1221 rsp->reason = RSP_REASON_UNTRUSTED;
1223 return finish_http_response(csp, rsp);
1225 #endif /* def FEATURE_TRUST */
1228 /*********************************************************************
1230 * Function : compile_dynamic_pcrs_job_list
1232 * Description : Compiles a dynamic pcrs job list (one with variables
1233 * resolved at request time)
1236 * 1 : csp = Current client state (buffers, headers, etc...)
1237 * 2 : b = The filter list to compile
1239 * Returns : NULL in case of errors, otherwise the
1242 *********************************************************************/
1243 pcrs_job *compile_dynamic_pcrs_job_list(const struct client_state *csp, const struct re_filterfile_spec *b)
1245 struct list_entry *pattern;
1246 pcrs_job *job_list = NULL;
1247 pcrs_job *dummy = NULL;
1248 pcrs_job *lastjob = NULL;
1251 const struct pcrs_variable variables[] =
1253 {"url", csp->http->url, 1},
1254 {"path", csp->http->path, 1},
1255 {"host", csp->http->host, 1},
1256 {"origin", csp->ip_addr_str, 1},
1260 for (pattern = b->patterns->first; pattern != NULL; pattern = pattern->next)
1262 assert(pattern->str != NULL);
1264 dummy = pcrs_compile_dynamic_command(pattern->str, variables, &error);
1268 log_error(LOG_LEVEL_ERROR,
1269 "Adding filter job \'%s\' to dynamic filter %s failed: %s",
1270 pattern->str, b->name, pcrs_strerror(error));
1275 if (error == PCRS_WARN_TRUNCATION)
1277 log_error(LOG_LEVEL_ERROR,
1278 "At least one of the variables in \'%s\' had to "
1279 "be truncated before compilation", pattern->str);
1281 if (job_list == NULL)
1287 lastjob->next = dummy;
1297 /*********************************************************************
1299 * Function : rewrite_url
1301 * Description : Rewrites a URL with a single pcrs command
1302 * and returns the result if it differs from the
1303 * original and isn't obviously invalid.
1306 * 1 : old_url = URL to rewrite.
1307 * 2 : pcrs_command = pcrs command formatted as string (s@foo@bar@)
1310 * Returns : NULL if the pcrs_command didn't change the url, or
1311 * the result of the modification.
1313 *********************************************************************/
1314 char *rewrite_url(char *old_url, const char *pcrs_command)
1316 char *new_url = NULL;
1320 assert(pcrs_command);
1322 new_url = pcrs_execute_single_command(old_url, pcrs_command, &hits);
1326 log_error(LOG_LEVEL_REDIRECTS,
1327 "pcrs command \"%s\" didn't change \"%s\".",
1328 pcrs_command, old_url);
1333 log_error(LOG_LEVEL_REDIRECTS,
1334 "executing pcrs command \"%s\" to rewrite %s failed: %s",
1335 pcrs_command, old_url, pcrs_strerror(hits));
1338 else if (strncmpic(new_url, "http://", 7) && strncmpic(new_url, "https://", 8))
1340 log_error(LOG_LEVEL_ERROR,
1341 "pcrs command \"%s\" changed \"%s\" to \"%s\" (%u hi%s), "
1342 "but the result doesn't look like a valid URL and will be ignored.",
1343 pcrs_command, old_url, new_url, hits, (hits == 1) ? "t" : "ts");
1348 log_error(LOG_LEVEL_REDIRECTS,
1349 "pcrs command \"%s\" changed \"%s\" to \"%s\" (%u hi%s).",
1350 pcrs_command, old_url, new_url, hits, (hits == 1) ? "t" : "ts");
1358 #ifdef FEATURE_FAST_REDIRECTS
1359 /*********************************************************************
1361 * Function : get_last_url
1363 * Description : Search for the last URL inside a string.
1364 * If the string already is a URL, it will
1365 * be the first URL found.
1368 * 1 : subject = the string to check
1369 * 2 : redirect_mode = +fast-redirect{} mode
1371 * Returns : NULL if no URL was found, or
1372 * the last URL found.
1374 *********************************************************************/
1375 char *get_last_url(char *subject, const char *redirect_mode)
1377 char *new_url = NULL;
1381 assert(redirect_mode);
1383 subject = strdup(subject);
1384 if (subject == NULL)
1386 log_error(LOG_LEVEL_ERROR, "Out of memory while searching for redirects.");
1390 if (0 == strcmpic(redirect_mode, "check-decoded-url"))
1392 log_error(LOG_LEVEL_REDIRECTS, "Decoding \"%s\" if necessary.", subject);
1393 new_url = url_decode(subject);
1394 if (new_url != NULL)
1401 log_error(LOG_LEVEL_ERROR, "Unable to decode \"%s\".", subject);
1405 log_error(LOG_LEVEL_REDIRECTS, "Checking \"%s\" for redirects.", subject);
1408 * Find the last URL encoded in the request
1411 while ((tmp = strstr(tmp, "http://")) != NULL)
1415 tmp = (new_url != NULL) ? new_url : subject;
1416 while ((tmp = strstr(tmp, "https://")) != NULL)
1421 if ((new_url != NULL)
1422 && ( (new_url != subject)
1423 || (0 == strncmpic(subject, "http://", 7))
1424 || (0 == strncmpic(subject, "https://", 8))
1428 * Return new URL if we found a redirect
1429 * or if the subject already was a URL.
1431 * The second case makes sure that we can
1432 * chain get_last_url after another redirection check
1433 * (like rewrite_url) without losing earlier redirects.
1435 new_url = strdup(new_url);
1444 #endif /* def FEATURE_FAST_REDIRECTS */
1447 /*********************************************************************
1449 * Function : redirect_url
1451 * Description : Checks if Privoxy should answer the request with
1452 * a HTTP redirect and generates the redirect if
1456 * 1 : csp = Current client state (buffers, headers, etc...)
1458 * Returns : NULL if the request can pass, HTTP redirect otherwise.
1460 *********************************************************************/
1461 struct http_response *redirect_url(struct client_state *csp)
1463 struct http_response *rsp;
1464 #ifdef FEATURE_FAST_REDIRECTS
1466 * XXX: Do we still need FEATURE_FAST_REDIRECTS
1467 * as compile-time option? The user can easily disable
1468 * it in his action file.
1470 char * redirect_mode;
1471 #endif /* def FEATURE_FAST_REDIRECTS */
1472 char *old_url = NULL;
1473 char *new_url = NULL;
1474 char *redirection_string;
1476 if ((csp->action->flags & ACTION_REDIRECT))
1478 redirection_string = csp->action->string[ACTION_STRING_REDIRECT];
1481 * If the redirection string begins with 's',
1482 * assume it's a pcrs command, otherwise treat it as
1483 * properly formatted URL and use it for the redirection
1486 * According to RFC 2616 section 14.30 the URL
1487 * has to be absolute and if the user tries:
1488 * +redirect{shit/this/will/be/parsed/as/pcrs_command.html}
1489 * she would get undefined results anyway.
1493 if (*redirection_string == 's')
1495 old_url = csp->http->url;
1496 new_url = rewrite_url(old_url, redirection_string);
1500 log_error(LOG_LEVEL_REDIRECTS,
1501 "No pcrs command recognized, assuming that \"%s\" is already properly formatted.",
1502 redirection_string);
1503 new_url = strdup(redirection_string);
1507 #ifdef FEATURE_FAST_REDIRECTS
1508 if ((csp->action->flags & ACTION_FAST_REDIRECTS))
1510 redirect_mode = csp->action->string[ACTION_STRING_FAST_REDIRECTS];
1513 * If it exists, use the previously rewritten URL as input
1514 * otherwise just use the old path.
1516 old_url = (new_url != NULL) ? new_url : strdup(csp->http->path);
1517 new_url = get_last_url(old_url, redirect_mode);
1522 * Disable redirect checkers, so that they
1523 * will be only run more than once if the user
1524 * also enables them through tags.
1526 * From a performance point of view
1527 * it doesn't matter, but the duplicated
1528 * log messages are annoying.
1530 csp->action->flags &= ~ACTION_FAST_REDIRECTS;
1531 #endif /* def FEATURE_FAST_REDIRECTS */
1532 csp->action->flags &= ~ACTION_REDIRECT;
1534 /* Did any redirect action trigger? */
1537 if (0 == strcmpic(new_url, csp->http->url))
1539 log_error(LOG_LEVEL_ERROR,
1540 "New URL \"%s\" and old URL \"%s\" are the same. Redirection loop prevented.",
1541 csp->http->url, new_url);
1546 log_error(LOG_LEVEL_REDIRECTS, "New URL is: %s", new_url);
1548 if (NULL == (rsp = alloc_http_response()))
1551 return cgi_error_memory();
1554 if ( enlist_unique_header(rsp->headers, "Location", new_url)
1555 || (NULL == (rsp->status = strdup("302 Local Redirect from Privoxy"))) )
1558 free_http_response(rsp);
1559 return cgi_error_memory();
1561 rsp->reason = RSP_REASON_REDIRECTED;
1564 return finish_http_response(csp, rsp);
1568 /* Only reached if no redirect is required */
1574 #ifdef FEATURE_IMAGE_BLOCKING
1575 /*********************************************************************
1577 * Function : is_imageurl
1579 * Description : Given a URL, decide whether it is an image or not,
1580 * using either the info from a previous +image action
1581 * or, #ifdef FEATURE_IMAGE_DETECT_MSIE, and the browser
1582 * is MSIE and not on a Mac, tell from the browser's accept
1586 * 1 : csp = Current client state (buffers, headers, etc...)
1588 * Returns : True (nonzero) if URL is an image, false (0)
1591 *********************************************************************/
1592 int is_imageurl(const struct client_state *csp)
1594 #ifdef FEATURE_IMAGE_DETECT_MSIE
1597 tmp = get_header_value(csp->headers, "User-Agent:");
1598 if (tmp && strstr(tmp, "MSIE") && !strstr(tmp, "Mac_"))
1600 tmp = get_header_value(csp->headers, "Accept:");
1601 if (tmp && strstr(tmp, "image/gif"))
1603 /* Client will accept HTML. If this seems counterintuitive,
1613 #endif /* def FEATURE_IMAGE_DETECT_MSIE */
1615 return ((csp->action->flags & ACTION_IMAGE) != 0);
1618 #endif /* def FEATURE_IMAGE_BLOCKING */
1621 #ifdef FEATURE_TRUST
1622 /*********************************************************************
1624 * Function : is_untrusted_url
1626 * Description : Should we "distrust" this URL (and block it)?
1628 * Yes if it matches a line in the trustfile, or if the
1629 * referrer matches a line starting with "+" in the
1634 * 1 : csp = Current client state (buffers, headers, etc...)
1636 * Returns : 0 => trusted, 1 => untrusted
1638 *********************************************************************/
1639 int is_untrusted_url(const struct client_state *csp)
1641 struct file_list *fl;
1642 struct block_spec *b;
1643 struct url_spec **trusted_url;
1644 struct http_request rhttp[1];
1645 const char * referer;
1649 * If we don't have a trustlist, we trust everybody
1651 if (((fl = csp->tlist) == NULL) || ((b = fl->f) == NULL))
1656 memset(rhttp, '\0', sizeof(*rhttp));
1659 * Do we trust the request URL itself?
1661 for (b = b->next; b ; b = b->next)
1663 if (url_match(b->url, csp->http))
1669 if (NULL == (referer = get_header_value(csp->headers, "Referer:")))
1671 /* no referrer was supplied */
1677 * If not, do we maybe trust its referrer?
1679 err = parse_http_url(referer, rhttp, csp);
1685 for (trusted_url = csp->config->trust_list; *trusted_url != NULL; trusted_url++)
1687 if (url_match(*trusted_url, rhttp))
1689 /* if the URL's referrer is from a trusted referrer, then
1690 * add the target spec to the trustfile as an unblocked
1691 * domain and return 0 (which means it's OK).
1696 if (NULL != (fp = fopen(csp->config->trustfile, "a")))
1700 char * new_entry = strdup("~");
1702 string_append(&new_entry, csp->http->hostport);
1704 path = csp->http->path;
1705 if ( (path[0] == '/')
1707 && ((path_end = strchr(path + 2, '/')) != NULL))
1709 /* since this path points into a user's home space
1710 * be sure to include this spec in the trustfile.
1712 int path_len = path_end - path; /* save offset */
1713 path = strdup(path); /* Copy string */
1716 path_end = path + path_len; /* regenerate ptr to new buffer */
1717 *(path_end + 1) = '\0'; /* Truncate path after '/' */
1719 string_join(&new_entry, path);
1723 * Give a reason for generating this entry.
1725 string_append(&new_entry, " # Trusted referrer was: ");
1726 string_append(&new_entry, referer);
1728 if (new_entry != NULL)
1730 if (-1 == fprintf(fp, "%s\n", new_entry))
1732 log_error(LOG_LEVEL_ERROR, "Failed to append \'%s\' to trustfile \'%s\': %E",
1733 new_entry, csp->config->trustfile);
1739 /* FIXME: No way to handle out-of memory, so mostly ignoring it */
1740 log_error(LOG_LEVEL_ERROR, "Out of memory adding pattern to trust file");
1747 log_error(LOG_LEVEL_ERROR, "Failed to append new entry for \'%s\' to trustfile \'%s\': %E",
1748 csp->http->hostport, csp->config->trustfile);
1756 #endif /* def FEATURE_TRUST */
1759 /*********************************************************************
1761 * Function : pcrs_filter_response
1763 * Description : Execute all text substitutions from all applying
1764 * +filter actions on the text buffer that's been
1765 * accumulated in csp->iob->buf.
1768 * 1 : csp = Current client state (buffers, headers, etc...)
1770 * Returns : a pointer to the (newly allocated) modified buffer.
1771 * or NULL if there were no hits or something went wrong
1773 *********************************************************************/
1774 static char *pcrs_filter_response(struct client_state *csp)
1777 size_t size, prev_size;
1783 struct file_list *fl;
1784 struct re_filterfile_spec *b;
1785 struct list_entry *filtername;
1787 int i, found_filters = 0;
1792 if (csp->iob->cur >= csp->iob->eod)
1798 * Need to check the set of re_filterfiles...
1800 for (i = 0; i < MAX_AF_FILES; i++)
1813 if (0 == found_filters)
1815 log_error(LOG_LEVEL_ERROR, "Inconsistent configuration: "
1816 "content filtering enabled, but no content filters available.");
1820 size = (size_t)(csp->iob->eod - csp->iob->cur);
1821 old = csp->iob->cur;
1823 for (i = 0; i < MAX_AF_FILES; i++)
1826 if ((NULL == fl) || (NULL == fl->f))
1829 * Either there are no filter files
1830 * left, or this filter file just
1831 * contains no valid filters.
1833 * Continue to be sure we don't miss
1834 * valid filter files that are chained
1835 * after empty or invalid ones.
1840 * For all applying +filter actions, look if a filter by that
1841 * name exists and if yes, execute it's pcrs_joblist on the
1844 for (b = fl->f; b; b = b->next)
1846 if (b->type != FT_CONTENT_FILTER)
1848 /* Skip header filters */
1852 for (filtername = csp->action->multi[ACTION_MULTI_FILTER]->first;
1853 filtername ; filtername = filtername->next)
1855 if (strcmp(b->name, filtername->str) == 0)
1857 int current_hits = 0; /* Number of hits caused by this filter */
1858 int job_number = 0; /* Which job we're currently executing */
1859 int job_hits = 0; /* How many hits the current job caused */
1860 pcrs_job *joblist = b->joblist;
1862 if (b->dynamic) joblist = compile_dynamic_pcrs_job_list(csp, b);
1864 if (NULL == joblist)
1866 log_error(LOG_LEVEL_RE_FILTER, "Filter %s has empty joblist. Nothing to do.", b->name);
1871 /* Apply all jobs from the joblist */
1872 for (job = joblist; NULL != job; job = job->next)
1875 job_hits = pcrs_execute(job, old, size, &new, &size);
1880 * That went well. Continue filtering
1881 * and use the result of this job as
1882 * input for the next one.
1884 current_hits += job_hits;
1885 if (old != csp->iob->cur)
1894 * This job caused an unexpected error. Inform the user
1895 * and skip the rest of the jobs in this filter. We could
1896 * continue with the next job, but usually the jobs
1897 * depend on each other or are similar enough to
1898 * fail for the same reason.
1900 * At the moment our pcrs expects the error codes of pcre 3.4,
1901 * but newer pcre versions can return additional error codes.
1902 * As a result pcrs_strerror()'s error message might be
1903 * "Unknown error ...", therefore we print the numerical value
1906 * XXX: Is this important enough for LOG_LEVEL_ERROR or
1907 * should we use LOG_LEVEL_RE_FILTER instead?
1909 log_error(LOG_LEVEL_ERROR, "Skipped filter \'%s\' after job number %u: %s (%d)",
1910 b->name, job_number, pcrs_strerror(job_hits), job_hits);
1915 if (b->dynamic) pcrs_free_joblist(joblist);
1917 log_error(LOG_LEVEL_RE_FILTER,
1918 "filtering %s%s (size %d) with \'%s\' produced %d hits (new size %d).",
1919 csp->http->hostport, csp->http->path, prev_size, b->name, current_hits, size);
1921 hits += current_hits;
1928 * If there were no hits, destroy our copy and let
1929 * chat() use the original in csp->iob
1937 csp->flags |= CSP_FLAG_MODIFIED;
1938 csp->content_length = size;
1946 /*********************************************************************
1948 * Function : gif_deanimate_response
1950 * Description : Deanimate the GIF image that has been accumulated in
1951 * csp->iob->buf, set csp->content_length to the modified
1952 * size and raise the CSP_FLAG_MODIFIED flag.
1955 * 1 : csp = Current client state (buffers, headers, etc...)
1957 * Returns : a pointer to the (newly allocated) modified buffer.
1958 * or NULL in case something went wrong.
1960 *********************************************************************/
1961 static char *gif_deanimate_response(struct client_state *csp)
1963 struct binbuffer *in, *out;
1967 size = (size_t)(csp->iob->eod - csp->iob->cur);
1969 if ( (NULL == (in = (struct binbuffer *)zalloc(sizeof *in )))
1970 || (NULL == (out = (struct binbuffer *)zalloc(sizeof *out))) )
1972 log_error(LOG_LEVEL_DEANIMATE, "failed! (no mem)");
1976 in->buffer = csp->iob->cur;
1979 if (gif_deanimate(in, out, strncmp("last", csp->action->string[ACTION_STRING_DEANIMATE], 4)))
1981 log_error(LOG_LEVEL_DEANIMATE, "failed! (gif parsing)");
1988 if ((int)size == out->offset)
1990 log_error(LOG_LEVEL_DEANIMATE, "GIF not changed.");
1994 log_error(LOG_LEVEL_DEANIMATE, "Success! GIF shrunk from %d bytes to %d.", size, out->offset);
1996 csp->content_length = out->offset;
1997 csp->flags |= CSP_FLAG_MODIFIED;
2007 /*********************************************************************
2009 * Function : jpeg_inspect_response
2014 * 1 : csp = Current client state (buffers, headers, etc...)
2016 * Returns : a pointer to the (newly allocated) modified buffer
2017 * or NULL in case something went wrong.
2019 *********************************************************************/
2020 static char *jpeg_inspect_response(struct client_state *csp)
2022 struct binbuffer *in = NULL;
2023 struct binbuffer *out = NULL;
2027 size = (size_t)(csp->iob->eod - csp->iob->cur);
2029 if (NULL == (in = (struct binbuffer *)zalloc(sizeof *in )))
2031 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg no mem 1)");
2035 if (NULL == (out = (struct binbuffer *)zalloc(sizeof *out)))
2037 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg no mem 2)");
2041 in->buffer = csp->iob->cur;
2045 * Calling jpeg_inspect has the side-effect of creating and
2046 * modifying the image buffer of "out" directly.
2048 if (jpeg_inspect(in, out))
2050 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg parsing)");
2058 csp->content_length = out->offset;
2059 csp->flags |= CSP_FLAG_MODIFIED;
2069 /*********************************************************************
2071 * Function : get_filter_function
2073 * Description : Decides which content filter function has
2074 * to be applied (if any).
2076 * XXX: Doesn't handle filter_popups()
2077 * because of the different prototype. Probably
2078 * we should ditch filter_popups() anyway, it's
2079 * even less reliable than popup blocking based
2083 * 1 : csp = Current client state (buffers, headers, etc...)
2085 * Returns : The content filter function to run, or
2086 * NULL if no content filter is active
2088 *********************************************************************/
2089 filter_function_ptr get_filter_function(struct client_state *csp)
2091 filter_function_ptr filter_function = NULL;
2094 * Are we enabling text mode by force?
2096 if (csp->action->flags & ACTION_FORCE_TEXT_MODE)
2099 * Do we really have to?
2101 if (csp->content_type & CT_TEXT)
2103 log_error(LOG_LEVEL_HEADER, "Text mode is already enabled.");
2107 csp->content_type |= CT_TEXT;
2108 log_error(LOG_LEVEL_HEADER, "Text mode enabled by force. Take cover!");
2112 if (!(csp->content_type & CT_DECLARED))
2115 * The server didn't bother to declare a MIME-Type.
2116 * Assume it's text that can be filtered.
2118 * This also regulary happens with 304 responses,
2119 * therefore logging anything here would cause
2122 csp->content_type |= CT_TEXT;
2126 * Choose the applying filter function based on
2127 * the content type and action settings.
2129 if ((csp->content_type & CT_TEXT) &&
2130 (csp->rlist != NULL) &&
2131 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER])))
2133 filter_function = pcrs_filter_response;
2135 else if ((csp->content_type & CT_GIF) &&
2136 (csp->action->flags & ACTION_DEANIMATE))
2138 filter_function = gif_deanimate_response;
2140 else if ((csp->content_type & CT_JPEG) &&
2141 (csp->action->flags & ACTION_JPEG_INSPECT))
2143 filter_function = jpeg_inspect_response;
2146 return filter_function;
2150 /*********************************************************************
2152 * Function : remove_chunked_transfer_coding
2154 * Description : In-situ remove the "chunked" transfer coding as defined
2155 * in rfc2616 from a buffer.
2158 * 1 : buffer = Pointer to the text buffer
2159 * 2 : size = In: Number of bytes to be processed,
2160 * Out: Number of bytes after de-chunking.
2161 * (undefined in case of errors)
2163 * Returns : JB_ERR_OK for success,
2164 * JB_ERR_PARSE otherwise
2166 *********************************************************************/
2167 static jb_err remove_chunked_transfer_coding(char *buffer, size_t *size)
2170 unsigned int chunksize = 0;
2171 char *from_p, *to_p;
2174 from_p = to_p = buffer;
2176 if (sscanf(buffer, "%x", &chunksize) != 1)
2178 log_error(LOG_LEVEL_ERROR, "Invalid first chunksize while stripping \"chunked\" transfer coding");
2179 return JB_ERR_PARSE;
2182 while (chunksize > 0)
2184 if (NULL == (from_p = strstr(from_p, "\r\n")))
2186 log_error(LOG_LEVEL_ERROR, "Parse error while stripping \"chunked\" transfer coding");
2187 return JB_ERR_PARSE;
2190 if ((newsize += chunksize) >= *size)
2192 log_error(LOG_LEVEL_ERROR, "Chunksize exceeds buffer in \"chunked\" transfer coding");
2193 return JB_ERR_PARSE;
2197 memmove(to_p, from_p, (size_t) chunksize);
2198 to_p = buffer + newsize;
2199 from_p += chunksize + 2;
2201 if (sscanf(from_p, "%x", &chunksize) != 1)
2203 log_error(LOG_LEVEL_INFO, "Invalid \"chunked\" transfer encoding detected and ignored.");
2208 /* XXX: Should get its own loglevel. */
2209 log_error(LOG_LEVEL_RE_FILTER, "De-chunking successful. Shrunk from %d to %d", *size, newsize);
2218 /*********************************************************************
2220 * Function : prepare_for_filtering
2222 * Description : If necessary, de-chunks and decompresses
2223 * the content so it can get filterd.
2226 * 1 : csp = Current client state (buffers, headers, etc...)
2228 * Returns : JB_ERR_OK for success,
2229 * JB_ERR_PARSE otherwise
2231 *********************************************************************/
2232 static jb_err prepare_for_filtering(struct client_state *csp)
2234 jb_err err = JB_ERR_OK;
2237 * If the body has a "chunked" transfer-encoding,
2238 * get rid of it, adjusting size and iob->eod
2240 if (csp->flags & CSP_FLAG_CHUNKED)
2242 size_t size = (size_t)(csp->iob->eod - csp->iob->cur);
2244 log_error(LOG_LEVEL_RE_FILTER, "Need to de-chunk first");
2245 err = remove_chunked_transfer_coding(csp->iob->cur, &size);
2246 if (JB_ERR_OK == err)
2248 csp->iob->eod = csp->iob->cur + size;
2249 csp->flags |= CSP_FLAG_MODIFIED;
2253 return JB_ERR_PARSE;
2259 * If the body has a supported transfer-encoding,
2260 * decompress it, adjusting size and iob->eod.
2262 if (csp->content_type & (CT_GZIP|CT_DEFLATE))
2264 if (0 == csp->iob->eod - csp->iob->cur)
2266 /* Nothing left after de-chunking. */
2270 err = decompress_iob(csp);
2272 if (JB_ERR_OK == err)
2274 csp->flags |= CSP_FLAG_MODIFIED;
2275 csp->content_type &= ~CT_TABOO;
2280 * Unset CT_GZIP and CT_DEFLATE to remember not
2281 * to modify the Content-Encoding header later.
2283 csp->content_type &= ~CT_GZIP;
2284 csp->content_type &= ~CT_DEFLATE;
2293 /*********************************************************************
2295 * Function : execute_content_filter
2297 * Description : Executes a given content filter.
2300 * 1 : csp = Current client state (buffers, headers, etc...)
2301 * 2 : content_filter = The filter function to execute
2303 * Returns : Pointer to the modified buffer, or
2304 * NULL if filtering failed or wasn't necessary.
2306 *********************************************************************/
2307 char *execute_content_filter(struct client_state *csp, filter_function_ptr content_filter)
2309 if (0 == csp->iob->eod - csp->iob->cur)
2312 * No content (probably status code 301, 302 ...),
2313 * no filtering necessary.
2318 if (JB_ERR_OK != prepare_for_filtering(csp))
2321 * failed to de-chunk or decompress.
2326 if (0 == csp->iob->eod - csp->iob->cur)
2329 * Clown alarm: chunked and/or compressed nothing delivered.
2334 return ((*content_filter)(csp));
2338 /*********************************************************************
2340 * Function : url_actions
2342 * Description : Gets the actions for this URL.
2345 * 1 : http = http_request request for blocked URLs
2346 * 2 : csp = Current client state (buffers, headers, etc...)
2350 *********************************************************************/
2351 void url_actions(struct http_request *http,
2352 struct client_state *csp)
2354 struct file_list *fl;
2355 struct url_actions *b;
2358 init_current_action(csp->action);
2360 for (i = 0; i < MAX_AF_FILES; i++)
2362 if (((fl = csp->actions_list[i]) == NULL) || ((b = fl->f) == NULL))
2367 apply_url_actions(csp->action, http, b);
2374 /*********************************************************************
2376 * Function : apply_url_actions
2378 * Description : Applies a list of URL actions.
2381 * 1 : action = Destination.
2382 * 2 : http = Current URL
2383 * 3 : b = list of URL actions to apply
2387 *********************************************************************/
2388 void apply_url_actions(struct current_action_spec *action,
2389 struct http_request *http,
2390 struct url_actions *b)
2394 /* Should never happen */
2398 for (b = b->next; NULL != b; b = b->next)
2400 if (url_match(b->url, http))
2402 merge_current_action(action, b->action);
2408 /*********************************************************************
2410 * Function : get_forward_override_settings
2412 * Description : Returns forward settings as specified with the
2413 * forward-override{} action. forward-override accepts
2414 * forward lines similar to the one used in the
2415 * configuration file, but without the URL pattern.
2421 * in the configuration file can be replaced with
2422 * the action section:
2424 * {+forward-override{forward .}}
2428 * 1 : csp = Current client state (buffers, headers, etc...)
2430 * Returns : Pointer to forwarding structure in case of success.
2431 * Invalid syntax is fatal.
2433 *********************************************************************/
2434 static const struct forward_spec *get_forward_override_settings(struct client_state *csp)
2436 const char *forward_override_line = csp->action->string[ACTION_STRING_FORWARD_OVERRIDE];
2437 char forward_settings[BUFFER_SIZE];
2438 char *http_parent = NULL;
2439 /* variable names were chosen for consistency reasons. */
2440 struct forward_spec *fwd = NULL;
2444 assert(csp->action->flags & ACTION_FORWARD_OVERRIDE);
2445 /* Should be enforced by load_one_actions_file() */
2446 assert(strlen(forward_override_line) < sizeof(forward_settings) - 1);
2448 /* Create a copy ssplit can modify */
2449 strlcpy(forward_settings, forward_override_line, sizeof(forward_settings));
2451 if (NULL != csp->fwd)
2454 * XXX: Currently necessary to prevent memory
2455 * leaks when the show-url-info cgi page is visited.
2457 unload_forward_spec(csp->fwd);
2461 * allocate a new forward node, valid only for
2462 * the lifetime of this request. Save its location
2463 * in csp as well, so sweep() can free it later on.
2465 fwd = csp->fwd = zalloc(sizeof(*fwd));
2468 log_error(LOG_LEVEL_FATAL,
2469 "can't allocate memory for forward-override{%s}", forward_override_line);
2470 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2473 vec_count = ssplit(forward_settings, " \t", vec, SZ(vec), 1, 1);
2474 if ((vec_count == 2) && !strcasecmp(vec[0], "forward"))
2476 fwd->type = SOCKS_NONE;
2478 /* Parse the parent HTTP proxy host:port */
2479 http_parent = vec[1];
2482 else if (vec_count == 3)
2484 char *socks_proxy = NULL;
2486 if (!strcasecmp(vec[0], "forward-socks4"))
2488 fwd->type = SOCKS_4;
2489 socks_proxy = vec[1];
2491 else if (!strcasecmp(vec[0], "forward-socks4a"))
2493 fwd->type = SOCKS_4A;
2494 socks_proxy = vec[1];
2496 else if (!strcasecmp(vec[0], "forward-socks5"))
2498 fwd->type = SOCKS_5;
2499 socks_proxy = vec[1];
2502 if (NULL != socks_proxy)
2504 /* Parse the SOCKS proxy host[:port] */
2505 fwd->gateway_host = strdup(socks_proxy);
2507 if (NULL != (socks_proxy = strchr(fwd->gateway_host, ':')))
2509 *socks_proxy++ = '\0';
2510 fwd->gateway_port = strtol(socks_proxy, NULL, 0);
2513 if (fwd->gateway_port <= 0)
2515 fwd->gateway_port = 1080;
2518 http_parent = vec[2];
2522 if (NULL == http_parent)
2524 log_error(LOG_LEVEL_FATAL,
2525 "Invalid forward-override syntax in: %s", forward_override_line);
2526 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2529 /* Parse http forwarding settings */
2530 if (strcmp(http_parent, ".") != 0)
2532 fwd->forward_host = strdup(http_parent);
2534 if (NULL != (http_parent = strchr(fwd->forward_host, ':')))
2536 *http_parent++ = '\0';
2537 fwd->forward_port = strtol(http_parent, NULL, 0);
2540 if (fwd->forward_port <= 0)
2542 fwd->forward_port = 8000;
2546 assert (NULL != fwd);
2548 log_error(LOG_LEVEL_CONNECT,
2549 "Overriding forwarding settings based on \'%s\'", forward_override_line);
2555 /*********************************************************************
2557 * Function : forward_url
2559 * Description : Should we forward this to another proxy?
2561 * XXX: Should be changed to make use of csp->fwd.
2564 * 1 : http = http_request request for current URL
2565 * 2 : csp = Current client state (buffers, headers, etc...)
2567 * Returns : Pointer to forwarding information.
2569 *********************************************************************/
2570 const struct forward_spec * forward_url(struct http_request *http,
2571 struct client_state *csp)
2573 static const struct forward_spec fwd_default[1] = { FORWARD_SPEC_INITIALIZER };
2574 struct forward_spec *fwd = csp->config->forward;
2576 if (csp->action->flags & ACTION_FORWARD_OVERRIDE)
2578 return get_forward_override_settings(csp);
2588 if (url_match(fwd->url, http))
2599 /*********************************************************************
2601 * Function : direct_response
2603 * Description : Check if Max-Forwards == 0 for an OPTIONS or TRACE
2604 * request and if so, return a HTTP 501 to the client.
2606 * FIXME: I have a stupid name and I should handle the
2607 * requests properly. Still, what we do here is rfc-
2608 * compliant, whereas ignoring or forwarding are not.
2611 * 1 : csp = Current client state (buffers, headers, etc...)
2613 * Returns : http_response if , NULL if nonmatch or handler fail
2615 *********************************************************************/
2616 struct http_response *direct_response(struct client_state *csp)
2618 struct http_response *rsp;
2619 struct list_entry *p;
2621 if ((0 == strcmpic(csp->http->gpc, "trace"))
2622 || (0 == strcmpic(csp->http->gpc, "options")))
2624 for (p = csp->headers->first; (p != NULL) ; p = p->next)
2626 if (!strncmpic("Max-Forwards:", p->str, 13))
2628 unsigned int max_forwards;
2631 * If it's a Max-Forwards value of zero,
2632 * we have to intercept the request.
2634 if (1 == sscanf(p->str+12, ": %u", &max_forwards) && max_forwards == 0)
2637 * FIXME: We could handle at least TRACE here,
2638 * but that would require a verbatim copy of
2639 * the request which we don't have anymore
2641 log_error(LOG_LEVEL_HEADER,
2642 "Detected header \'%s\' in OPTIONS or TRACE request. Returning 501.",
2645 /* Get mem for response or fail*/
2646 if (NULL == (rsp = alloc_http_response()))
2648 return cgi_error_memory();
2651 if (NULL == (rsp->status = strdup("501 Not Implemented")))
2653 free_http_response(rsp);
2654 return cgi_error_memory();
2658 rsp->reason = RSP_REASON_UNSUPPORTED;
2660 return(finish_http_response(csp, rsp));
2669 /*********************************************************************
2671 * Function : content_filters_enabled
2673 * Description : Checks whether there are any content filters
2674 * enabled for the current request.
2677 * 1 : csp = Current client state (buffers, headers, etc...)
2679 * Returns : TRUE for yes, FALSE otherwise
2681 *********************************************************************/
2682 inline int content_filters_enabled(const struct client_state *csp)
2684 return (((csp->rlist != NULL) &&
2685 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER]))) ||
2686 (csp->action->flags & (ACTION_DEANIMATE|ACTION_JPEG_INSPECT|ACTION_NO_POPUPS)));