1 const char filters_rcs[] = "$Id: filters.c,v 1.97 2007/11/30 15:37:03 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 1.97 2007/11/30 15:37:03 fabiankeil
44 * Use freez instead of free.
46 * Revision 1.96 2007/10/19 16:53:28 fabiankeil
47 * Add helper function to check if any content filters are enabled.
49 * Revision 1.95 2007/10/17 19:31:20 fabiankeil
50 * Omitting the zero chunk that ends the chunk transfer encoding seems
51 * to be the new black. Log the problem and continue filtering anyway.
53 * Revision 1.94 2007/09/29 13:20:20 fabiankeil
54 * Remove two redundant and one useless log messages.
56 * Revision 1.93 2007/09/29 10:21:16 fabiankeil
57 * - Move get_filter_function() from jcc.c to filters.c
58 * so the filter functions can be static.
59 * - Don't bother filtering body-less responses.
61 * Revision 1.92 2007/09/28 16:38:55 fabiankeil
62 * - Execute content filters through execute_content_filter().
63 * - Add prepare_for_filtering() so filter functions don't have to
64 * care about de-chunking and decompression. As a side effect this enables
65 * decompression for gif_deanimate_response() and jpeg_inspect_response().
66 * - Change remove_chunked_transfer_coding()'s return type to jb_err.
67 * Some clowns feel like chunking empty responses in which case
68 * (size == 0) is valid but previously would be interpreted as error.
70 * Revision 1.91 2007/09/02 15:31:20 fabiankeil
71 * Move match_portlist() from filter.c to urlmatch.c.
72 * It's used for url matching, not for filtering.
74 * Revision 1.90 2007/09/02 12:44:17 fabiankeil
75 * Remove newline at the end of a log_error() message.
77 * Revision 1.89 2007/08/05 13:42:23 fabiankeil
78 * #1763173 from Stefan Huehner: declare some more functions static.
80 * Revision 1.88 2007/06/01 16:41:11 fabiankeil
81 * Add forward-override{} to change the forwarding settings through
82 * action sections. This is mainly interesting to forward different
83 * clients differently (for example based on User-Agent or request
86 * Revision 1.87 2007/04/30 15:53:10 fabiankeil
87 * Make sure filters with dynamic jobs actually use them.
89 * Revision 1.86 2007/04/30 15:03:28 fabiankeil
90 * - Introduce dynamic pcrs jobs that can resolve variables.
91 * - Don't run redirect functions more than once,
92 * unless they are activated more than once.
94 * Revision 1.85 2007/03/21 12:24:47 fabiankeil
95 * - Log the content size after decompression in decompress_iob()
96 * instead of pcrs_filter_response().
98 * Revision 1.84 2007/03/20 15:16:34 fabiankeil
99 * Use dedicated header filter actions instead of abusing "filter".
100 * Replace "filter-client-headers" and "filter-client-headers"
101 * with "server-header-filter" and "client-header-filter".
103 * Revision 1.83 2007/03/17 15:20:05 fabiankeil
104 * New config option: enforce-blocks.
106 * Revision 1.82 2007/03/13 11:28:43 fabiankeil
107 * - Fix port handling in acl_addr() and use a temporary acl spec
108 * copy so error messages don't contain a truncated version.
109 * - Log size of iob before and after decompression.
111 * Revision 1.81 2007/03/05 14:40:53 fabiankeil
112 * - Cosmetical changes for LOG_LEVEL_RE_FILTER messages.
113 * - Hide the "Go there anyway" link for blocked CONNECT
114 * requests where going there anyway doesn't work anyway.
116 * Revision 1.80 2007/02/07 10:55:20 fabiankeil
117 * - Save the reason for generating http_responses.
118 * - Block (+block) with status code 403 instead of 404.
119 * - Use a different kludge to remember a failed decompression.
121 * Revision 1.79 2007/01/31 16:21:38 fabiankeil
122 * Search for Max-Forwards headers case-insensitive,
123 * don't generate the "501 unsupported" message for invalid
124 * Max-Forwards values and don't increase negative ones.
126 * Revision 1.78 2007/01/28 13:41:18 fabiankeil
127 * - Add HEAD support to finish_http_response.
128 * - Add error favicon to internal HTML error messages.
130 * Revision 1.77 2007/01/12 15:36:44 fabiankeil
131 * Mark *csp as immutable for is_untrusted_url()
132 * and is_imageurl(). Closes FR 1237736.
134 * Revision 1.76 2007/01/01 19:36:37 fabiankeil
135 * Integrate a modified version of Wil Mahan's
136 * zlib patch (PR #895531).
138 * Revision 1.75 2006/12/29 18:30:46 fabiankeil
139 * Fixed gcc43 conversion warnings,
140 * changed sprintf calls to snprintf.
142 * Revision 1.74 2006/12/24 17:37:38 fabiankeil
143 * Adjust comment in pcrs_filter_response()
144 * to recent pcrs changes. Hohoho.
146 * Revision 1.73 2006/12/23 16:01:02 fabiankeil
147 * Don't crash if pcre returns an error code
148 * that pcrs didn't expect. Fixes BR 1621173.
150 * Revision 1.72 2006/12/22 18:52:53 fabiankeil
151 * Modified is_untrusted_url to complain in case of
152 * write errors and to give a reason when adding new
153 * entries to the trustfile. Closes FR 1097611.
155 * Revision 1.71 2006/12/22 14:24:52 fabiankeil
156 * Skip empty filter files in pcrs_filter_response,
157 * but don't ignore the ones that come afterwards.
158 * Fixes parts of BR 1619208.
160 * Revision 1.70 2006/12/09 13:33:15 fabiankeil
161 * Added some sanity checks for get_last_url().
162 * Fixed possible segfault caused by my last commit.
164 * Revision 1.69 2006/12/08 12:39:13 fabiankeil
165 * Let get_last_url() catch https URLs as well.
167 * Revision 1.68 2006/12/05 14:45:48 fabiankeil
168 * Make sure get_last_url() behaves like advertised
169 * and fast-redirects{} can be combined with redirect{}.
171 * Revision 1.67 2006/11/28 15:19:43 fabiankeil
172 * Implemented +redirect{s@foo@bar@} to generate
173 * a redirect based on a rewritten version of the
176 * Revision 1.66 2006/09/23 13:26:38 roro
177 * Replace TABs by spaces in source code.
179 * Revision 1.65 2006/09/21 12:54:43 fabiankeil
180 * Fix +redirect{}. Didn't work with -fast-redirects.
182 * Revision 1.64 2006/08/31 10:55:49 fabiankeil
183 * Block requests for untrusted URLs with status
184 * code 403 instead of 200.
186 * Revision 1.63 2006/08/31 10:11:28 fabiankeil
187 * Don't free p which is still in use and will be later
188 * freed by free_map(). Don't claim the referrer is unknown
189 * when the client didn't set one.
191 * Revision 1.62 2006/08/14 00:27:47 david__schmidt
192 * Feature request 595948: Re-Filter logging in single line
194 * Revision 1.61 2006/08/03 02:46:41 david__schmidt
195 * Incorporate Fabian Keil's patch work:
\rhttp://www.fabiankeil.de/sourcecode/privoxy/
197 * Revision 1.60 2006/07/18 14:48:46 david__schmidt
198 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
199 * with what was really the latest development (the v_3_0_branch branch)
201 * Revision 1.58.2.9 2006/01/29 23:10:56 david__schmidt
202 * Multiple filter file support
204 * Revision 1.58.2.8 2005/05/07 21:50:55 david__schmidt
205 * A few memory leaks plugged (mostly on error paths)
207 * Revision 1.58.2.7 2004/10/03 12:53:32 david__schmidt
208 * Add the ability to check jpeg images for invalid
209 * lengths of comment blocks. Defensive strategy
210 * against the exploit:
211 * Microsoft Security Bulletin MS04-028
212 * Buffer Overrun in JPEG Processing (GDI+) Could
213 * Allow Code Execution (833987)
214 * Enabled with +inspect-jpegs in actions files.
216 * Revision 1.58.2.6 2003/12/06 22:18:27 gliptak
217 * Correcting compile problem with FEATURE_IMAGE_BLOCKING
219 * Revision 1.58.2.5 2003/11/11 13:10:31 oes
220 * Fixed bug #839859: "See why" link URL now gets url-encoded.
222 * Revision 1.58.2.4 2003/02/28 12:52:45 oes
225 * Revision 1.58.2.3 2002/09/25 14:51:51 oes
226 * Added basic support for OPTIONS and TRACE HTTP methods:
227 * New function direct_response which handles OPTIONS and
228 * TRACE requests whose Max-Forwards header field is zero.
230 * Revision 1.58.2.2 2002/08/01 17:18:28 oes
231 * Fixed BR 537651 / SR 579724 (MSIE image detect improper for IE/Mac)
233 * Revision 1.58.2.1 2002/07/26 15:18:53 oes
234 * - Bugfix: Executing a filters without jobs no longer results in
235 * turing off *all* filters.
236 * - Security fix: Malicious web servers can't cause a seg fault
237 * through bogus chunk sizes anymore
239 * Revision 1.58 2002/04/24 02:11:17 oes
240 * Jon's multiple AF patch: url_actions now evaluates rules
243 * Revision 1.57 2002/04/08 20:38:34 swa
246 * Revision 1.56 2002/04/05 15:51:24 oes
247 * - bugfix: error-pages now get correct request protocol
248 * - fix for invalid HTML in trust info
250 * Revision 1.55 2002/04/02 16:13:51 oes
251 * Fix: No "Go there anyway" for SSL
253 * Revision 1.54 2002/04/02 14:55:56 oes
254 * Bugfix: is_untrusted_url() now depends on FEATURE_TRUST, not FEATURE_COOKIE_JAR
256 * Revision 1.53 2002/03/26 22:29:54 swa
257 * we have a new homepage!
259 * Revision 1.52 2002/03/24 16:35:57 jongfoster
262 * Revision 1.51 2002/03/24 15:23:33 jongfoster
265 * Revision 1.50 2002/03/24 13:25:43 swa
266 * name change related issues
268 * Revision 1.49 2002/03/16 20:29:14 oes
271 * Revision 1.48 2002/03/13 20:25:34 oes
272 * Better logging for content filters
274 * Revision 1.47 2002/03/13 00:30:52 jongfoster
276 * Added option of always sending redirect for imageblock,
277 * currently disabled with #if 0.
279 * Revision 1.46 2002/03/12 01:42:49 oes
280 * Introduced modular filters
282 * Revision 1.45 2002/03/08 16:47:50 oes
283 * Added choice beween GIF and PNG built-in images
285 * Revision 1.44 2002/03/07 03:49:31 oes
286 * - Fixed compiler warnings etc
287 * - Changed built-in images from GIF to PNG
288 * (with regard to Unisys patent issue)
289 * - Added a 4x4 pattern PNG which is less intrusive
290 * than the logo but also clearly marks the deleted banners
292 * Revision 1.43 2002/01/22 23:51:59 jongfoster
293 * Replacing strsav() with the safer string_append().
295 * Adding missing html_encode() to error message generators. Where encoded
296 * and unencoded versions of a string were provided, removing the unencoded
299 * Revision 1.42 2002/01/17 21:00:32 jongfoster
300 * Moving all our URL and URL pattern parsing code to urlmatch.c.
302 * Using a single, simple url_match(pattern,url) function - rather than
303 * the 3-line match routine which was repeated all over the place.
305 * Renaming free_url to free_url_spec, since it frees a struct url_spec.
307 * Using parse_http_url() to parse URLs without faking a HTTP
308 * request line for parse_http_request().
310 * Revision 1.41 2001/11/13 00:14:07 jongfoster
311 * Fixing stupid bug now I've figured out what || means.
312 * (It always returns 0 or 1, not one of it's paramaters.)
314 * Revision 1.40 2001/10/26 17:37:55 oes
315 * - Re-enabled Netscape 200/404 bug workaround in block_url():
316 * - Removed OS/2 special case
317 * - Made block_url() independant from sed() having been run
318 * - Made trust_url independant from sed() having been run
319 * - Made is_imageurl independant from sed() having been run.
320 * It now checks User-Agent: and Accept: by itself.
323 * Revision 1.39 2001/10/25 03:40:48 david__schmidt
324 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
325 * threads to call select() simultaneously. So, it's time to do a real, live,
326 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
327 * (native). Both versions will work, but using __OS2__ offers multi-threading.
329 * Revision 1.38 2001/10/23 21:32:33 jongfoster
330 * Adding error-checking to selected functions
332 * Revision 1.37 2001/10/22 15:33:56 david__schmidt
333 * Special-cased OS/2 out of the Netscape-abort-on-404-in-js problem in
334 * filters.c. Added a FIXME in front of the offending code. I'll gladly
335 * put in a better/more robust fix for all parties if one is presented...
336 * It seems that just returning 200 instead of 404 would pretty much fix
337 * it for everyone, but I don't know all the history of the problem.
339 * Revision 1.36 2001/10/10 16:44:16 oes
340 * Added match_portlist function
342 * Revision 1.35 2001/10/07 15:41:23 oes
343 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
345 * New function remove_chunked_transfer_coding that strips chunked
346 * transfer coding to plain and is called by pcrs_filter_response
347 * and gif_deanimate_response if neccessary
349 * Improved handling of zero-change re_filter runs
351 * pcrs_filter_response and gif_deanimate_response now remove
352 * chunked transfer codeing before processing the body.
354 * Revision 1.34 2001/09/20 15:49:36 steudten
356 * Fix BUG: Change int size to size_t size in pcrs_filter_response().
357 * See cgi.c fill_template().
359 * Revision 1.33 2001/09/16 17:05:14 jongfoster
360 * Removing unused #include showarg.h
362 * Revision 1.32 2001/09/16 13:21:27 jongfoster
363 * Changes to use new list functions.
365 * Revision 1.31 2001/09/16 11:38:02 jongfoster
366 * Splitting fill_template() into 2 functions:
367 * template_load() loads the file
368 * template_fill() performs the PCRS regexps.
369 * This is because the CGI edit interface has a "table row"
370 * template which is used many times in the page - this
371 * change means it's only loaded from disk once.
373 * Revision 1.30 2001/09/16 11:00:10 jongfoster
374 * New function alloc_http_response, for symmetry with free_http_response
376 * Revision 1.29 2001/09/13 23:32:40 jongfoster
377 * Moving image data to cgi.c rather than cgi.h
378 * Fixing a GPF under Win32 (and any other OS that protects global
379 * constants from being written to).
381 * Revision 1.28 2001/09/10 10:18:51 oes
382 * Silenced compiler warnings
384 * Revision 1.27 2001/08/05 16:06:20 jongfoster
385 * Modifiying "struct map" so that there are now separate header and
386 * "map_entry" structures. This means that functions which modify a
387 * map no longer need to return a pointer to the modified map.
388 * Also, it no longer reverses the order of the entries (which may be
389 * important with some advanced template substitutions).
391 * Revision 1.26 2001/07/30 22:08:36 jongfoster
392 * Tidying up #defines:
393 * - All feature #defines are now of the form FEATURE_xxx
394 * - Permanently turned off WIN_GUI_EDIT
395 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
397 * Revision 1.25 2001/07/26 10:09:46 oes
398 * Made browser detection a little less naive
400 * Revision 1.24 2001/07/25 17:22:51 oes
401 * Added workaround for Netscape bug that prevents display of page when loading a component fails.
403 * Revision 1.23 2001/07/23 13:40:12 oes
404 * Fixed bug that caused document body to be dropped when pcrs joblist was empty.
406 * Revision 1.22 2001/07/18 12:29:34 oes
407 * - Made gif_deanimate_response respect
408 * csp->action->string[ACTION_STRING_DEANIMATE]
409 * - Logging cosmetics
411 * Revision 1.21 2001/07/13 13:59:53 oes
412 * - Introduced gif_deanimate_response which shares the
413 * generic content modification interface of pcrs_filter_response
414 * and acts as a wrapper to deanimate.c:gif_deanimate()
415 * - Renamed re_process_buffer to pcrs_filter_response
416 * - pcrs_filter_response now returns NULL on failiure
417 * - Removed all #ifdef PCRS
419 * Revision 1.20 2001/07/01 17:01:04 oes
420 * Added comments and missing return statement in is_untrusted_url()
422 * Revision 1.19 2001/06/29 21:45:41 oes
423 * Indentation, CRLF->LF, Tab-> Space
425 * Revision 1.18 2001/06/29 13:27:38 oes
426 * - Cleaned up, renamed and reorderd functions
427 * and improved comments
430 * - Ported to CGI platform. Now delivers
431 * http_response or NULL
432 * - Unified HTML and GIF generation (moved image detection
433 * and GIF generation here from jcc.c:chat())
434 * - Fixed HTTP status to:
435 * - 403 (Forbidden) for the "blocked" HTML message
436 * - 200 (OK) for GIF answers
437 * - 302 (Redirect) for redirect to GIF
440 * - Ported to CGI platform. Now delivers
441 * http_response or NULL
442 * - Separated detection of untrusted URL into
443 * (bool)is_untrusted_url
444 * - Added enforcement of untrusted requests
446 * - Moved redirect_url() from cgi.c to here
447 * and ported it to the CGI platform
449 * - Removed logentry from cancelled commit
451 * Revision 1.17 2001/06/09 10:55:28 jongfoster
452 * Changing BUFSIZ ==> BUFFER_SIZE
454 * Revision 1.16 2001/06/07 23:10:26 jongfoster
455 * Allowing unanchored domain patterns to back off and retry
456 * if they partially match. Optimized right-anchored patterns.
457 * Moving ACL and forward files into config file.
458 * Replacing struct gateway with struct forward_spec
460 * Revision 1.15 2001/06/03 19:12:00 oes
461 * extracted-CGI relevant stuff
463 * Revision 1.14 2001/06/01 10:30:55 oes
464 * Added optional left-anchoring to domaincmp
466 * Revision 1.13 2001/05/31 21:21:30 jongfoster
467 * Permissionsfile / actions file changes:
468 * - Changed "permission" to "action" throughout
469 * - changes to file format to allow string parameters
470 * - Moved helper functions to actions.c
472 * Revision 1.12 2001/05/31 17:35:20 oes
474 * - Enhanced domain part globbing with infix and prefix asterisk
475 * matching and optional unanchored operation
477 * Revision 1.11 2001/05/29 11:53:23 oes
478 * "See why" link added to "blocked" page
480 * Revision 1.10 2001/05/29 09:50:24 jongfoster
481 * Unified blocklist/imagelist/permissionslist.
482 * File format is still under discussion, but the internal changes
485 * Also modified interceptor behaviour:
486 * - We now intercept all URLs beginning with one of the following
487 * prefixes (and *only* these prefixes):
489 * * http://ijbswa.sf.net/config/
490 * * http://ijbswa.sourceforge.net/config/
491 * - New interceptors "home page" - go to http://i.j.b/ to see it.
492 * - Internal changes so that intercepted and fast redirect pages
493 * are not replaced with an image.
494 * - Interceptors now have the option to send a binary page direct
495 * to the client. (i.e. ijb-send-banner uses this)
496 * - Implemented show-url-info interceptor. (Which is why I needed
497 * the above interceptors changes - a typical URL is
498 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
499 * The previous mechanism would not have intercepted that, and
500 * if it had been intercepted then it then it would have replaced
503 * Revision 1.9 2001/05/27 22:17:04 oes
505 * - re_process_buffer no longer writes the modified buffer
506 * to the client, which was very ugly. It now returns the
507 * buffer, which it is then written by chat.
509 * - content_length now adjusts the Content-Length: header
510 * for modified documents rather than crunch()ing it.
511 * (Length info in csp->content_length, which is 0 for
512 * unmodified documents)
514 * - For this to work, sed() is called twice when filtering.
516 * Revision 1.8 2001/05/26 17:13:28 jongfoster
517 * Filled in a function comment.
519 * Revision 1.7 2001/05/26 15:26:15 jongfoster
520 * ACL feature now provides more security by immediately dropping
521 * connections from untrusted hosts.
523 * Revision 1.6 2001/05/26 00:28:36 jongfoster
524 * Automatic reloading of config file.
525 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
526 * Most of the global variables have been moved to a new
527 * struct configuration_spec, accessed through csp->config->globalname
528 * Most of the globals remaining are used by the Win32 GUI.
530 * Revision 1.5 2001/05/25 22:34:30 jongfoster
533 * Revision 1.4 2001/05/22 18:46:04 oes
535 * - Enabled filtering banners by size rather than URL
536 * by adding patterns that replace all standard banner
537 * sizes with the "Junkbuster" gif to the re_filterfile
539 * - Enabled filtering WebBugs by providing a pattern
540 * which kills all 1x1 images
542 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
543 * which is selected by the (nonstandard and therefore
544 * capital) letter 'U' in the option string.
545 * It causes the quantifiers to be ungreedy by default.
546 * Appending a ? turns back to greedy (!).
548 * - Added a new interceptor ijb-send-banner, which
549 * sends back the "Junkbuster" gif. Without imagelist or
550 * MSIE detection support, or if tinygif = 1, or the
551 * URL isn't recognized as an imageurl, a lame HTML
552 * explanation is sent instead.
554 * - Added new feature, which permits blocking remote
555 * script redirects and firing back a local redirect
557 * The feature is conditionally compiled, i.e. it
558 * can be disabled with --disable-fast-redirects,
559 * plus it must be activated by a "fast-redirects"
560 * line in the config file, has its own log level
561 * and of course wants to be displayed by show-proxy-args
562 * Note: Boy, all the #ifdefs in 1001 locations and
563 * all the fumbling with configure.in and acconfig.h
564 * were *way* more work than the feature itself :-(
566 * - Because a generic redirect template was needed for
567 * this, tinygif = 3 now uses the same.
569 * - Moved GIFs, and other static HTTP response templates
574 * - Removed some >400 CRs again (Jon, you really worked
577 * Revision 1.3 2001/05/20 16:44:47 jongfoster
578 * Removing last hardcoded Junkbusters.com URLs.
580 * Revision 1.2 2001/05/20 01:21:20 jongfoster
581 * Version 2.9.4 checkin.
582 * - Merged popupfile and cookiefile, and added control over PCRS
583 * filtering, in new "permissionsfile".
584 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
585 * file error you now get a message box (in the Win32 GUI) rather
586 * than the program exiting with no explanation.
587 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
589 * - Removed tabs from "config"
590 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
591 * - Bumped up version number.
593 * Revision 1.1.1.1 2001/05/15 13:58:52 oes
594 * Initial import of version 2.9.3 source tree
597 *********************************************************************/
603 #include <sys/types.h>
612 #endif /* ndef __OS2__ */
613 #include <netinet/in.h>
615 #include <winsock2.h>
616 #endif /* ndef _WIN32 */
620 #endif /* def __OS2__ */
628 #include "jbsockets.h"
629 #include "miscutil.h"
633 #include "deanimate.h"
634 #include "urlmatch.h"
641 const char filters_h_rcs[] = FILTERS_H_VERSION;
643 /* Fix a problem with Solaris. There should be no effect on other
645 * Solaris's isspace() is a macro which uses it's argument directly
646 * as an array index. Therefore we need to make sure that high-bit
647 * characters generate +ve values, and ideally we also want to make
648 * the argument match the declared parameter type of "int".
650 #define ijb_isdigit(__X) isdigit((int)(unsigned char)(__X))
652 static jb_err remove_chunked_transfer_coding(char *buffer, size_t *size);
653 static jb_err prepare_for_filtering(struct client_state *csp);
656 /*********************************************************************
658 * Function : block_acl
660 * Description : Block this request?
661 * Decide yes or no based on ACL file.
664 * 1 : dst = The proxy or gateway address this is going to.
665 * Or NULL to check all possible targets.
666 * 2 : csp = Current client state (buffers, headers, etc...)
667 * Also includes the client IP address.
669 * Returns : 0 = FALSE (don't block) and 1 = TRUE (do block)
671 *********************************************************************/
672 int block_acl(struct access_control_addr *dst, struct client_state *csp)
674 struct access_control_list *acl = csp->config->acl;
676 /* if not using an access control list, then permit the connection */
682 /* search the list */
685 if ((csp->ip_addr_long & acl->src->mask) == acl->src->addr)
689 /* Just want to check if they have any access */
690 if (acl->action == ACL_PERMIT)
695 else if ( ((dst->addr & acl->dst->mask) == acl->dst->addr)
696 && ((dst->port == acl->dst->port) || (acl->dst->port == 0)))
698 if (acl->action == ACL_PERMIT)
716 /*********************************************************************
718 * Function : acl_addr
720 * Description : Called from `load_config' to parse an ACL address.
723 * 1 : aspec = String specifying ACL address.
724 * 2 : aca = struct access_control_addr to fill in.
726 * Returns : 0 => Ok, everything else is an error.
728 *********************************************************************/
729 int acl_addr(const char *aspec, struct access_control_addr *aca)
734 char *acl_spec = NULL;
740 * Use a temporary acl spec copy so we can log
741 * the unmodified original in case of parse errors.
743 acl_spec = strdup(aspec);
744 if (acl_spec == NULL)
746 /* XXX: This will be logged as parse error. */
750 if ((p = strchr(acl_spec, '/')) != NULL)
753 if (ijb_isdigit(*p) == 0)
758 masklength = atoi(p);
761 if ((masklength < 0) || (masklength > 32))
767 if ((p = strchr(acl_spec, ':')) != NULL)
772 port = strtol(p, &endptr, 10);
774 if (port <= 0 || port > 65535 || *endptr != '\0')
781 aca->port = (unsigned long)port;
783 aca->addr = ntohl(resolve_hostname_to_ip(acl_spec));
786 if (aca->addr == INADDR_NONE)
788 /* XXX: This will be logged as parse error. */
792 /* build the netmask */
794 for (i=1; i <= masklength ; i++)
796 aca->mask |= (1 << (32 - i));
799 /* now mask off the host portion of the ip address
800 * (i.e. save on the network portion of the address).
802 aca->addr = aca->addr & aca->mask;
807 #endif /* def FEATURE_ACL */
810 /*********************************************************************
812 * Function : block_url
814 * Description : Called from `chat'. Check to see if we need to block this.
817 * 1 : csp = Current client state (buffers, headers, etc...)
819 * Returns : NULL => unblocked, else HTTP block response
821 *********************************************************************/
822 struct http_response *block_url(struct client_state *csp)
824 struct http_response *rsp;
825 const char *new_content_type = NULL;
828 * If it's not blocked, don't block it ;-)
830 if ((csp->action->flags & ACTION_BLOCK) == 0)
834 if (csp->action->flags & ACTION_REDIRECT)
836 log_error(LOG_LEVEL_ERROR, "redirect{} overruled by block.");
839 * Else, prepare a response
841 if (NULL == (rsp = alloc_http_response()))
843 return cgi_error_memory();
847 * If it's an image-url, send back an image or redirect
848 * as specified by the relevant +image action
850 #ifdef FEATURE_IMAGE_BLOCKING
851 if (((csp->action->flags & ACTION_IMAGE_BLOCKER) != 0)
855 /* determine HOW images should be blocked */
856 p = csp->action->string[ACTION_STRING_IMAGE_BLOCKER];
858 if(csp->action->flags & ACTION_HANDLE_AS_EMPTY_DOCUMENT)
860 log_error(LOG_LEVEL_ERROR, "handle-as-empty-document overruled by handle-as-image.");
862 #if 1 /* Two alternative strategies, use this one for now: */
864 /* and handle accordingly: */
865 if ((p == NULL) || (0 == strcmpic(p, "pattern")))
867 rsp->status = strdup("403 Request blocked by Privoxy");
868 if (rsp->status == NULL)
870 free_http_response(rsp);
871 return cgi_error_memory();
873 rsp->body = bindup(image_pattern_data, image_pattern_length);
874 if (rsp->body == NULL)
876 free_http_response(rsp);
877 return cgi_error_memory();
879 rsp->content_length = image_pattern_length;
881 if (enlist_unique_header(rsp->headers, "Content-Type", BUILTIN_IMAGE_MIMETYPE))
883 free_http_response(rsp);
884 return cgi_error_memory();
888 else if (0 == strcmpic(p, "blank"))
890 rsp->status = strdup("403 Request blocked by Privoxy");
891 if (rsp->status == NULL)
893 free_http_response(rsp);
894 return cgi_error_memory();
896 rsp->body = bindup(image_blank_data, image_blank_length);
897 if (rsp->body == NULL)
899 free_http_response(rsp);
900 return cgi_error_memory();
902 rsp->content_length = image_blank_length;
904 if (enlist_unique_header(rsp->headers, "Content-Type", BUILTIN_IMAGE_MIMETYPE))
906 free_http_response(rsp);
907 return cgi_error_memory();
913 rsp->status = strdup("302 Local Redirect from Privoxy");
914 if (rsp->status == NULL)
916 free_http_response(rsp);
917 return cgi_error_memory();
920 if (enlist_unique_header(rsp->headers, "Location", p))
922 free_http_response(rsp);
923 return cgi_error_memory();
927 #else /* Following code is disabled for now */
929 /* and handle accordingly: */
930 if ((p == NULL) || (0 == strcmpic(p, "pattern")))
932 p = CGI_PREFIX "send-banner?type=pattern";
934 else if (0 == strcmpic(p, "blank"))
936 p = CGI_PREFIX "send-banner?type=blank";
938 rsp->status = strdup("302 Local Redirect from Privoxy");
939 if (rsp->status == NULL)
941 free_http_response(rsp);
942 return cgi_error_memory();
945 if (enlist_unique_header(rsp->headers, "Location", p))
947 free_http_response(rsp);
948 return cgi_error_memory();
950 #endif /* Preceeding code is disabled for now */
952 else if(csp->action->flags & ACTION_HANDLE_AS_EMPTY_DOCUMENT)
955 * Send empty document.
957 new_content_type = csp->action->string[ACTION_STRING_CONTENT_TYPE];
960 rsp->body = strdup(" ");
961 rsp->content_length = 1;
963 rsp->status = strdup("403 Request blocked by Privoxy");
964 if (rsp->status == NULL)
966 free_http_response(rsp);
967 return cgi_error_memory();
969 if (new_content_type != 0)
971 log_error(LOG_LEVEL_HEADER, "Overwriting Content-Type with %s", new_content_type);
972 if (enlist_unique_header(rsp->headers, "Content-Type", new_content_type))
974 free_http_response(rsp);
975 return cgi_error_memory();
980 #endif /* def FEATURE_IMAGE_BLOCKING */
983 * Else, generate an HTML "blocked" message:
987 struct map * exports;
991 * Workaround for stupid Netscape bug which prevents
992 * pages from being displayed if loading a referenced
993 * JavaScript or style sheet fails. So make it appear
994 * as if it succeeded.
996 if ( NULL != (p = get_header_value(csp->headers, "User-Agent:"))
997 && !strncmpic(p, "mozilla", 7) /* Catch Netscape but */
998 && !strstr(p, "Gecko") /* save Mozilla, */
999 && !strstr(p, "compatible") /* MSIE */
1000 && !strstr(p, "Opera")) /* and Opera. */
1002 rsp->status = strdup("200 Request for blocked URL");
1006 rsp->status = strdup("403 Request for blocked URL");
1009 if (rsp->status == NULL)
1011 free_http_response(rsp);
1012 return cgi_error_memory();
1015 exports = default_exports(csp, NULL);
1016 if (exports == NULL)
1018 free_http_response(rsp);
1019 return cgi_error_memory();
1022 #ifdef FEATURE_FORCE_LOAD
1023 err = map(exports, "force-prefix", 1, FORCE_PREFIX, 1);
1025 * Export the force conditional block killer if
1027 * - Privoxy was compiled without FEATURE_FORCE_LOAD, or
1028 * - Privoxy is configured to enforce blocks, or
1029 * - it's a CONNECT request and enforcing wouldn't work anyway.
1031 if ((csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1032 || (0 == strcmpic(csp->http->gpc, "connect")))
1033 #endif /* ndef FEATURE_FORCE_LOAD */
1035 err = map_block_killer(exports, "force-support");
1038 if (!err) err = map(exports, "protocol", 1, csp->http->ssl ? "https://" : "http://", 1);
1039 if (!err) err = map(exports, "hostport", 1, html_encode(csp->http->hostport), 0);
1040 if (!err) err = map(exports, "path", 1, html_encode(csp->http->path), 0);
1041 if (!err) err = map(exports, "path-ue", 1, url_encode(csp->http->path), 0);
1046 free_http_response(rsp);
1047 return cgi_error_memory();
1050 err = template_fill_for_cgi(csp, "blocked", exports, rsp);
1053 free_http_response(rsp);
1054 return cgi_error_memory();
1057 rsp->reason = RSP_REASON_BLOCKED;
1059 return finish_http_response(csp, rsp);
1064 #ifdef FEATURE_TRUST
1065 /*********************************************************************
1067 * Function : trust_url FIXME: I should be called distrust_url
1069 * Description : Calls is_untrusted_url to determine if the URL is trusted
1070 * and if not, returns a HTTP 403 response with a reject message.
1073 * 1 : csp = Current client state (buffers, headers, etc...)
1075 * Returns : NULL => trusted, else http_response.
1077 *********************************************************************/
1078 struct http_response *trust_url(struct client_state *csp)
1080 struct http_response *rsp;
1081 struct map * exports;
1082 char buf[BUFFER_SIZE];
1084 struct url_spec **tl;
1089 * Don't bother to work on trusted URLs
1091 if (!is_untrusted_url(csp))
1097 * Else, prepare a response:
1099 if (NULL == (rsp = alloc_http_response()))
1101 return cgi_error_memory();
1104 rsp->status = strdup("403 Request blocked by Privoxy");
1105 exports = default_exports(csp, NULL);
1106 if (exports == NULL || rsp->status == NULL)
1108 free_http_response(rsp);
1109 return cgi_error_memory();
1113 * Export the protocol, host, port, and referrer information
1115 err = map(exports, "hostport", 1, csp->http->hostport, 1);
1116 if (!err) err = map(exports, "protocol", 1, csp->http->ssl ? "https://" : "http://", 1);
1117 if (!err) err = map(exports, "path", 1, csp->http->path, 1);
1119 if (NULL != (p = get_header_value(csp->headers, "Referer:")))
1121 if (!err) err = map(exports, "referrer", 1, html_encode(p), 0);
1125 if (!err) err = map(exports, "referrer", 1, "none set", 1);
1131 free_http_response(rsp);
1132 return cgi_error_memory();
1136 * Export the trust list
1139 for (tl = csp->config->trust_list; (t = *tl) != NULL ; tl++)
1141 snprintf(buf, sizeof(buf), "<li>%s</li>\n", t->spec);
1142 string_append(&p, buf);
1144 err = map(exports, "trusted-referrers", 1, p, 0);
1149 free_http_response(rsp);
1150 return cgi_error_memory();
1154 * Export the trust info, if available
1156 if (csp->config->trust_info->first)
1158 struct list_entry *l;
1161 for (l = csp->config->trust_info->first; l ; l = l->next)
1163 snprintf(buf, sizeof(buf), "<li> <a href=\"%s\">%s</a><br>\n", l->str, l->str);
1164 string_append(&p, buf);
1166 err = map(exports, "trust-info", 1, p, 0);
1170 err = map_block_killer(exports, "have-trust-info");
1176 free_http_response(rsp);
1177 return cgi_error_memory();
1181 * Export the force conditional block killer if
1183 * - Privoxy was compiled without FEATURE_FORCE_LOAD, or
1184 * - Privoxy is configured to enforce blocks, or
1185 * - it's a CONNECT request and enforcing wouldn't work anyway.
1187 #ifdef FEATURE_FORCE_LOAD
1188 if ((csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
1189 || (0 == strcmpic(csp->http->gpc, "connect")))
1191 err = map_block_killer(exports, "force-support");
1195 err = map(exports, "force-prefix", 1, FORCE_PREFIX, 1);
1197 #else /* ifndef FEATURE_FORCE_LOAD */
1198 err = map_block_killer(exports, "force-support");
1199 #endif /* ndef FEATURE_FORCE_LOAD */
1204 free_http_response(rsp);
1205 return cgi_error_memory();
1209 * Build the response
1211 err = template_fill_for_cgi(csp, "untrusted", exports, rsp);
1214 free_http_response(rsp);
1215 return cgi_error_memory();
1217 rsp->reason = RSP_REASON_UNTRUSTED;
1219 return finish_http_response(csp, rsp);
1221 #endif /* def FEATURE_TRUST */
1224 /*********************************************************************
1226 * Function : compile_dynamic_pcrs_job_list
1228 * Description : Compiles a dynamic pcrs job list (one with variables
1229 * resolved at request time)
1232 * 1 : csp = Current client state (buffers, headers, etc...)
1233 * 2 : b = The filter list to compile
1235 * Returns : NULL in case of errors, otherwise the
1238 *********************************************************************/
1239 pcrs_job *compile_dynamic_pcrs_job_list(const struct client_state *csp, const struct re_filterfile_spec *b)
1241 struct list_entry *pattern;
1242 pcrs_job *job_list = NULL;
1243 pcrs_job *dummy = NULL;
1244 pcrs_job *lastjob = NULL;
1247 const struct pcrs_variable variables[] =
1249 {"url", csp->http->url, 1},
1250 {"path", csp->http->path, 1},
1251 {"host", csp->http->host, 1},
1252 {"origin", csp->ip_addr_str, 1},
1256 for (pattern = b->patterns->first; pattern != NULL; pattern = pattern->next)
1258 assert(pattern->str != NULL);
1260 dummy = pcrs_compile_dynamic_command(pattern->str, variables, &error);
1264 log_error(LOG_LEVEL_ERROR,
1265 "Adding filter job \'%s\' to dynamic filter %s failed: %s",
1266 pattern->str, b->name, pcrs_strerror(error));
1271 if (error == PCRS_WARN_TRUNCATION)
1273 log_error(LOG_LEVEL_ERROR,
1274 "At least one of the variables in \'%s\' had to "
1275 "be truncated before compilation", pattern->str);
1277 if (job_list == NULL)
1283 lastjob->next = dummy;
1293 /*********************************************************************
1295 * Function : rewrite_url
1297 * Description : Rewrites a URL with a single pcrs command
1298 * and returns the result if it differs from the
1299 * original and isn't obviously invalid.
1302 * 1 : old_url = URL to rewrite.
1303 * 2 : pcrs_command = pcrs command formatted as string (s@foo@bar@)
1306 * Returns : NULL if the pcrs_command didn't change the url, or
1307 * the result of the modification.
1309 *********************************************************************/
1310 char *rewrite_url(char *old_url, const char *pcrs_command)
1312 char *new_url = NULL;
1316 assert(pcrs_command);
1318 new_url = pcrs_execute_single_command(old_url, pcrs_command, &hits);
1322 log_error(LOG_LEVEL_REDIRECTS,
1323 "pcrs command \"%s\" didn't change \"%s\".",
1324 pcrs_command, old_url);
1329 log_error(LOG_LEVEL_REDIRECTS,
1330 "executing pcrs command \"%s\" to rewrite %s failed: %s",
1331 pcrs_command, old_url, pcrs_strerror(hits));
1334 else if (strncmpic(new_url, "http://", 7) && strncmpic(new_url, "https://", 8))
1336 log_error(LOG_LEVEL_ERROR,
1337 "pcrs command \"%s\" changed \"%s\" to \"%s\" (%u hi%s), "
1338 "but the result doesn't look like a valid URL and will be ignored.",
1339 pcrs_command, old_url, new_url, hits, (hits == 1) ? "t" : "ts");
1344 log_error(LOG_LEVEL_REDIRECTS,
1345 "pcrs command \"%s\" changed \"%s\" to \"%s\" (%u hi%s).",
1346 pcrs_command, old_url, new_url, hits, (hits == 1) ? "t" : "ts");
1354 #ifdef FEATURE_FAST_REDIRECTS
1355 /*********************************************************************
1357 * Function : get_last_url
1359 * Description : Search for the last URL inside a string.
1360 * If the string already is a URL, it will
1361 * be the first URL found.
1364 * 1 : subject = the string to check
1365 * 2 : redirect_mode = +fast-redirect{} mode
1367 * Returns : NULL if no URL was found, or
1368 * the last URL found.
1370 *********************************************************************/
1371 char *get_last_url(char *subject, const char *redirect_mode)
1373 char *new_url = NULL;
1377 assert(redirect_mode);
1379 subject = strdup(subject);
1380 if (subject == NULL)
1382 log_error(LOG_LEVEL_ERROR, "Out of memory while searching for redirects.");
1386 if (0 == strcmpic(redirect_mode, "check-decoded-url"))
1388 log_error(LOG_LEVEL_REDIRECTS, "Decoding \"%s\" if necessary.", subject);
1389 new_url = url_decode(subject);
1390 if (new_url != NULL)
1397 log_error(LOG_LEVEL_ERROR, "Unable to decode \"%s\".", subject);
1401 log_error(LOG_LEVEL_REDIRECTS, "Checking \"%s\" for redirects.", subject);
1404 * Find the last URL encoded in the request
1407 while ((tmp = strstr(tmp, "http://")) != NULL)
1411 tmp = (new_url != NULL) ? new_url : subject;
1412 while ((tmp = strstr(tmp, "https://")) != NULL)
1417 if ((new_url != NULL)
1418 && ( (new_url != subject)
1419 || (0 == strncmpic(subject, "http://", 7))
1420 || (0 == strncmpic(subject, "https://", 8))
1424 * Return new URL if we found a redirect
1425 * or if the subject already was a URL.
1427 * The second case makes sure that we can
1428 * chain get_last_url after another redirection check
1429 * (like rewrite_url) without losing earlier redirects.
1431 new_url = strdup(new_url);
1440 #endif /* def FEATURE_FAST_REDIRECTS */
1443 /*********************************************************************
1445 * Function : redirect_url
1447 * Description : Checks if Privoxy should answer the request with
1448 * a HTTP redirect and generates the redirect if
1452 * 1 : csp = Current client state (buffers, headers, etc...)
1454 * Returns : NULL if the request can pass, HTTP redirect otherwise.
1456 *********************************************************************/
1457 struct http_response *redirect_url(struct client_state *csp)
1459 struct http_response *rsp;
1460 #ifdef FEATURE_FAST_REDIRECTS
1462 * XXX: Do we still need FEATURE_FAST_REDIRECTS
1463 * as compile-time option? The user can easily disable
1464 * it in his action file.
1466 char * redirect_mode;
1467 #endif /* def FEATURE_FAST_REDIRECTS */
1468 char *old_url = NULL;
1469 char *new_url = NULL;
1470 char *redirection_string;
1472 if ((csp->action->flags & ACTION_REDIRECT))
1474 redirection_string = csp->action->string[ACTION_STRING_REDIRECT];
1477 * If the redirection string begins with 's',
1478 * assume it's a pcrs command, otherwise treat it as
1479 * properly formatted URL and use it for the redirection
1482 * According to RFC 2616 section 14.30 the URL
1483 * has to be absolute and if the user tries:
1484 * +redirect{shit/this/will/be/parsed/as/pcrs_command.html}
1485 * she would get undefined results anyway.
1489 if (*redirection_string == 's')
1491 old_url = csp->http->url;
1492 new_url = rewrite_url(old_url, redirection_string);
1496 log_error(LOG_LEVEL_REDIRECTS,
1497 "No pcrs command recognized, assuming that \"%s\" is already properly formatted.",
1498 redirection_string);
1499 new_url = strdup(redirection_string);
1503 #ifdef FEATURE_FAST_REDIRECTS
1504 if ((csp->action->flags & ACTION_FAST_REDIRECTS))
1506 redirect_mode = csp->action->string[ACTION_STRING_FAST_REDIRECTS];
1509 * If it exists, use the previously rewritten URL as input
1510 * otherwise just use the old path.
1512 old_url = (new_url != NULL) ? new_url : strdup(csp->http->path);
1513 new_url = get_last_url(old_url, redirect_mode);
1518 * Disable redirect checkers, so that they
1519 * will be only run more than once if the user
1520 * also enables them through tags.
1522 * From a performance point of view
1523 * it doesn't matter, but the duplicated
1524 * log messages are annoying.
1526 csp->action->flags &= ~ACTION_FAST_REDIRECTS;
1527 #endif /* def FEATURE_FAST_REDIRECTS */
1528 csp->action->flags &= ~ACTION_REDIRECT;
1530 /* Did any redirect action trigger? */
1533 if (0 == strcmpic(new_url, csp->http->url))
1535 log_error(LOG_LEVEL_ERROR,
1536 "New URL \"%s\" and old URL \"%s\" are the same. Redirection loop prevented.",
1537 csp->http->url, new_url);
1542 log_error(LOG_LEVEL_REDIRECTS, "New URL is: %s", new_url);
1544 if (NULL == (rsp = alloc_http_response()))
1547 return cgi_error_memory();
1550 if ( enlist_unique_header(rsp->headers, "Location", new_url)
1551 || (NULL == (rsp->status = strdup("302 Local Redirect from Privoxy"))) )
1554 free_http_response(rsp);
1555 return cgi_error_memory();
1557 rsp->reason = RSP_REASON_REDIRECTED;
1560 return finish_http_response(csp, rsp);
1564 /* Only reached if no redirect is required */
1570 #ifdef FEATURE_IMAGE_BLOCKING
1571 /*********************************************************************
1573 * Function : is_imageurl
1575 * Description : Given a URL, decide whether it is an image or not,
1576 * using either the info from a previous +image action
1577 * or, #ifdef FEATURE_IMAGE_DETECT_MSIE, and the browser
1578 * is MSIE and not on a Mac, tell from the browser's accept
1582 * 1 : csp = Current client state (buffers, headers, etc...)
1584 * Returns : True (nonzero) if URL is an image, false (0)
1587 *********************************************************************/
1588 int is_imageurl(const struct client_state *csp)
1590 #ifdef FEATURE_IMAGE_DETECT_MSIE
1593 tmp = get_header_value(csp->headers, "User-Agent:");
1594 if (tmp && strstr(tmp, "MSIE") && !strstr(tmp, "Mac_"))
1596 tmp = get_header_value(csp->headers, "Accept:");
1597 if (tmp && strstr(tmp, "image/gif"))
1599 /* Client will accept HTML. If this seems counterintuitive,
1609 #endif /* def FEATURE_IMAGE_DETECT_MSIE */
1611 return ((csp->action->flags & ACTION_IMAGE) != 0);
1614 #endif /* def FEATURE_IMAGE_BLOCKING */
1617 #ifdef FEATURE_TRUST
1618 /*********************************************************************
1620 * Function : is_untrusted_url
1622 * Description : Should we "distrust" this URL (and block it)?
1624 * Yes if it matches a line in the trustfile, or if the
1625 * referrer matches a line starting with "+" in the
1630 * 1 : csp = Current client state (buffers, headers, etc...)
1632 * Returns : 0 => trusted, 1 => untrusted
1634 *********************************************************************/
1635 int is_untrusted_url(const struct client_state *csp)
1637 struct file_list *fl;
1638 struct block_spec *b;
1639 struct url_spec **trusted_url;
1640 struct http_request rhttp[1];
1641 const char * referer;
1645 * If we don't have a trustlist, we trust everybody
1647 if (((fl = csp->tlist) == NULL) || ((b = fl->f) == NULL))
1652 memset(rhttp, '\0', sizeof(*rhttp));
1655 * Do we trust the request URL itself?
1657 for (b = b->next; b ; b = b->next)
1659 if (url_match(b->url, csp->http))
1665 if (NULL == (referer = get_header_value(csp->headers, "Referer:")))
1667 /* no referrer was supplied */
1673 * If not, do we maybe trust its referrer?
1675 err = parse_http_url(referer, rhttp, csp);
1681 for (trusted_url = csp->config->trust_list; *trusted_url != NULL; trusted_url++)
1683 if (url_match(*trusted_url, rhttp))
1685 /* if the URL's referrer is from a trusted referrer, then
1686 * add the target spec to the trustfile as an unblocked
1687 * domain and return 0 (which means it's OK).
1692 if (NULL != (fp = fopen(csp->config->trustfile, "a")))
1696 char * new_entry = strdup("~");
1698 string_append(&new_entry, csp->http->hostport);
1700 path = csp->http->path;
1701 if ( (path[0] == '/')
1703 && ((path_end = strchr(path + 2, '/')) != NULL))
1705 /* since this path points into a user's home space
1706 * be sure to include this spec in the trustfile.
1708 int path_len = path_end - path; /* save offset */
1709 path = strdup(path); /* Copy string */
1712 path_end = path + path_len; /* regenerate ptr to new buffer */
1713 *(path_end + 1) = '\0'; /* Truncate path after '/' */
1715 string_join(&new_entry, path);
1719 * Give a reason for generating this entry.
1721 string_append(&new_entry, " # Trusted referrer was: ");
1722 string_append(&new_entry, referer);
1724 if (new_entry != NULL)
1726 if (-1 == fprintf(fp, "%s\n", new_entry))
1728 log_error(LOG_LEVEL_ERROR, "Failed to append \'%s\' to trustfile \'%s\': %E",
1729 new_entry, csp->config->trustfile);
1735 /* FIXME: No way to handle out-of memory, so mostly ignoring it */
1736 log_error(LOG_LEVEL_ERROR, "Out of memory adding pattern to trust file");
1743 log_error(LOG_LEVEL_ERROR, "Failed to append new entry for \'%s\' to trustfile \'%s\': %E",
1744 csp->http->hostport, csp->config->trustfile);
1752 #endif /* def FEATURE_TRUST */
1755 /*********************************************************************
1757 * Function : pcrs_filter_response
1759 * Description : Execute all text substitutions from all applying
1760 * +filter actions on the text buffer that's been
1761 * accumulated in csp->iob->buf.
1764 * 1 : csp = Current client state (buffers, headers, etc...)
1766 * Returns : a pointer to the (newly allocated) modified buffer.
1767 * or NULL if there were no hits or something went wrong
1769 *********************************************************************/
1770 static char *pcrs_filter_response(struct client_state *csp)
1773 size_t size, prev_size;
1779 struct file_list *fl;
1780 struct re_filterfile_spec *b;
1781 struct list_entry *filtername;
1783 int i, found_filters = 0;
1788 if (csp->iob->cur >= csp->iob->eod)
1794 * Need to check the set of re_filterfiles...
1796 for (i = 0; i < MAX_AF_FILES; i++)
1809 if (0 == found_filters)
1811 log_error(LOG_LEVEL_ERROR, "Inconsistent configuration: "
1812 "content filtering enabled, but no content filters available.");
1816 size = (size_t)(csp->iob->eod - csp->iob->cur);
1817 old = csp->iob->cur;
1819 for (i = 0; i < MAX_AF_FILES; i++)
1822 if ((NULL == fl) || (NULL == fl->f))
1825 * Either there are no filter files
1826 * left, or this filter file just
1827 * contains no valid filters.
1829 * Continue to be sure we don't miss
1830 * valid filter files that are chained
1831 * after empty or invalid ones.
1836 * For all applying +filter actions, look if a filter by that
1837 * name exists and if yes, execute it's pcrs_joblist on the
1840 for (b = fl->f; b; b = b->next)
1842 if (b->type != FT_CONTENT_FILTER)
1844 /* Skip header filters */
1848 for (filtername = csp->action->multi[ACTION_MULTI_FILTER]->first;
1849 filtername ; filtername = filtername->next)
1851 if (strcmp(b->name, filtername->str) == 0)
1853 int current_hits = 0; /* Number of hits caused by this filter */
1854 int job_number = 0; /* Which job we're currently executing */
1855 int job_hits = 0; /* How many hits the current job caused */
1856 pcrs_job *joblist = b->joblist;
1858 if (b->dynamic) joblist = compile_dynamic_pcrs_job_list(csp, b);
1860 if (NULL == joblist)
1862 log_error(LOG_LEVEL_RE_FILTER, "Filter %s has empty joblist. Nothing to do.", b->name);
1867 /* Apply all jobs from the joblist */
1868 for (job = joblist; NULL != job; job = job->next)
1871 job_hits = pcrs_execute(job, old, size, &new, &size);
1876 * That went well. Continue filtering
1877 * and use the result of this job as
1878 * input for the next one.
1880 current_hits += job_hits;
1881 if (old != csp->iob->cur)
1890 * This job caused an unexpected error. Inform the user
1891 * and skip the rest of the jobs in this filter. We could
1892 * continue with the next job, but usually the jobs
1893 * depend on each other or are similar enough to
1894 * fail for the same reason.
1896 * At the moment our pcrs expects the error codes of pcre 3.4,
1897 * but newer pcre versions can return additional error codes.
1898 * As a result pcrs_strerror()'s error message might be
1899 * "Unknown error ...", therefore we print the numerical value
1902 * XXX: Is this important enough for LOG_LEVEL_ERROR or
1903 * should we use LOG_LEVEL_RE_FILTER instead?
1905 log_error(LOG_LEVEL_ERROR, "Skipped filter \'%s\' after job number %u: %s (%d)",
1906 b->name, job_number, pcrs_strerror(job_hits), job_hits);
1911 if (b->dynamic) pcrs_free_joblist(joblist);
1913 log_error(LOG_LEVEL_RE_FILTER,
1914 "filtering %s%s (size %d) with \'%s\' produced %d hits (new size %d).",
1915 csp->http->hostport, csp->http->path, prev_size, b->name, current_hits, size);
1917 hits += current_hits;
1924 * If there were no hits, destroy our copy and let
1925 * chat() use the original in csp->iob
1933 csp->flags |= CSP_FLAG_MODIFIED;
1934 csp->content_length = size;
1942 /*********************************************************************
1944 * Function : gif_deanimate_response
1946 * Description : Deanimate the GIF image that has been accumulated in
1947 * csp->iob->buf, set csp->content_length to the modified
1948 * size and raise the CSP_FLAG_MODIFIED flag.
1951 * 1 : csp = Current client state (buffers, headers, etc...)
1953 * Returns : a pointer to the (newly allocated) modified buffer.
1954 * or NULL in case something went wrong.
1956 *********************************************************************/
1957 static char *gif_deanimate_response(struct client_state *csp)
1959 struct binbuffer *in, *out;
1963 size = (size_t)(csp->iob->eod - csp->iob->cur);
1965 if ( (NULL == (in = (struct binbuffer *)zalloc(sizeof *in )))
1966 || (NULL == (out = (struct binbuffer *)zalloc(sizeof *out))) )
1968 log_error(LOG_LEVEL_DEANIMATE, "failed! (no mem)");
1972 in->buffer = csp->iob->cur;
1975 if (gif_deanimate(in, out, strncmp("last", csp->action->string[ACTION_STRING_DEANIMATE], 4)))
1977 log_error(LOG_LEVEL_DEANIMATE, "failed! (gif parsing)");
1984 if ((int)size == out->offset)
1986 log_error(LOG_LEVEL_DEANIMATE, "GIF not changed.");
1990 log_error(LOG_LEVEL_DEANIMATE, "Success! GIF shrunk from %d bytes to %d.", size, out->offset);
1992 csp->content_length = out->offset;
1993 csp->flags |= CSP_FLAG_MODIFIED;
2003 /*********************************************************************
2005 * Function : jpeg_inspect_response
2010 * 1 : csp = Current client state (buffers, headers, etc...)
2012 * Returns : a pointer to the (newly allocated) modified buffer
2013 * or NULL in case something went wrong.
2015 *********************************************************************/
2016 static char *jpeg_inspect_response(struct client_state *csp)
2018 struct binbuffer *in = NULL;
2019 struct binbuffer *out = NULL;
2023 size = (size_t)(csp->iob->eod - csp->iob->cur);
2025 if (NULL == (in = (struct binbuffer *)zalloc(sizeof *in )))
2027 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg no mem 1)");
2031 if (NULL == (out = (struct binbuffer *)zalloc(sizeof *out)))
2033 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg no mem 2)");
2037 in->buffer = csp->iob->cur;
2041 * Calling jpeg_inspect has the side-effect of creating and
2042 * modifying the image buffer of "out" directly.
2044 if (jpeg_inspect(in, out))
2046 log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg parsing)");
2054 csp->content_length = out->offset;
2055 csp->flags |= CSP_FLAG_MODIFIED;
2065 /*********************************************************************
2067 * Function : get_filter_function
2069 * Description : Decides which content filter function has
2070 * to be applied (if any).
2072 * XXX: Doesn't handle filter_popups()
2073 * because of the different prototype. Probably
2074 * we should ditch filter_popups() anyway, it's
2075 * even less reliable than popup blocking based
2079 * 1 : csp = Current client state (buffers, headers, etc...)
2081 * Returns : The content filter function to run, or
2082 * NULL if no content filter is active
2084 *********************************************************************/
2085 filter_function_ptr get_filter_function(struct client_state *csp)
2087 filter_function_ptr filter_function = NULL;
2090 * Are we enabling text mode by force?
2092 if (csp->action->flags & ACTION_FORCE_TEXT_MODE)
2095 * Do we really have to?
2097 if (csp->content_type & CT_TEXT)
2099 log_error(LOG_LEVEL_HEADER, "Text mode is already enabled.");
2103 csp->content_type |= CT_TEXT;
2104 log_error(LOG_LEVEL_HEADER, "Text mode enabled by force. Take cover!");
2108 if (!(csp->content_type & CT_DECLARED))
2111 * The server didn't bother to declare a MIME-Type.
2112 * Assume it's text that can be filtered.
2114 * This also regulary happens with 304 responses,
2115 * therefore logging anything here would cause
2118 csp->content_type |= CT_TEXT;
2122 * Choose the applying filter function based on
2123 * the content type and action settings.
2125 if ((csp->content_type & CT_TEXT) &&
2126 (csp->rlist != NULL) &&
2127 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER])))
2129 filter_function = pcrs_filter_response;
2131 else if ((csp->content_type & CT_GIF) &&
2132 (csp->action->flags & ACTION_DEANIMATE))
2134 filter_function = gif_deanimate_response;
2136 else if ((csp->content_type & CT_JPEG) &&
2137 (csp->action->flags & ACTION_JPEG_INSPECT))
2139 filter_function = jpeg_inspect_response;
2142 return filter_function;
2146 /*********************************************************************
2148 * Function : remove_chunked_transfer_coding
2150 * Description : In-situ remove the "chunked" transfer coding as defined
2151 * in rfc2616 from a buffer.
2154 * 1 : buffer = Pointer to the text buffer
2155 * 2 : size = In: Number of bytes to be processed,
2156 * Out: Number of bytes after de-chunking.
2157 * (undefined in case of errors)
2159 * Returns : JB_ERR_OK for success,
2160 * JB_ERR_PARSE otherwise
2162 *********************************************************************/
2163 static jb_err remove_chunked_transfer_coding(char *buffer, size_t *size)
2166 unsigned int chunksize = 0;
2167 char *from_p, *to_p;
2170 from_p = to_p = buffer;
2172 if (sscanf(buffer, "%x", &chunksize) != 1)
2174 log_error(LOG_LEVEL_ERROR, "Invalid first chunksize while stripping \"chunked\" transfer coding");
2175 return JB_ERR_PARSE;
2178 while (chunksize > 0)
2180 if (NULL == (from_p = strstr(from_p, "\r\n")))
2182 log_error(LOG_LEVEL_ERROR, "Parse error while stripping \"chunked\" transfer coding");
2183 return JB_ERR_PARSE;
2186 if ((newsize += chunksize) >= *size)
2188 log_error(LOG_LEVEL_ERROR, "Chunksize exceeds buffer in \"chunked\" transfer coding");
2189 return JB_ERR_PARSE;
2193 memmove(to_p, from_p, (size_t) chunksize);
2194 to_p = buffer + newsize;
2195 from_p += chunksize + 2;
2197 if (sscanf(from_p, "%x", &chunksize) != 1)
2199 log_error(LOG_LEVEL_INFO, "Invalid \"chunked\" transfer encoding detected and ignored.");
2204 /* XXX: Should get its own loglevel. */
2205 log_error(LOG_LEVEL_RE_FILTER, "De-chunking successful. Shrunk from %d to %d", *size, newsize);
2214 /*********************************************************************
2216 * Function : prepare_for_filtering
2218 * Description : If necessary, de-chunks and decompresses
2219 * the content so it can get filterd.
2222 * 1 : csp = Current client state (buffers, headers, etc...)
2224 * Returns : JB_ERR_OK for success,
2225 * JB_ERR_PARSE otherwise
2227 *********************************************************************/
2228 static jb_err prepare_for_filtering(struct client_state *csp)
2230 jb_err err = JB_ERR_OK;
2233 * If the body has a "chunked" transfer-encoding,
2234 * get rid of it, adjusting size and iob->eod
2236 if (csp->flags & CSP_FLAG_CHUNKED)
2238 size_t size = (size_t)(csp->iob->eod - csp->iob->cur);
2240 log_error(LOG_LEVEL_RE_FILTER, "Need to de-chunk first");
2241 err = remove_chunked_transfer_coding(csp->iob->cur, &size);
2242 if (JB_ERR_OK == err)
2244 csp->iob->eod = csp->iob->cur + size;
2245 csp->flags |= CSP_FLAG_MODIFIED;
2249 return JB_ERR_PARSE;
2255 * If the body has a supported transfer-encoding,
2256 * decompress it, adjusting size and iob->eod.
2258 if (csp->content_type & (CT_GZIP|CT_DEFLATE))
2260 if (0 == csp->iob->eod - csp->iob->cur)
2262 /* Nothing left after de-chunking. */
2266 err = decompress_iob(csp);
2268 if (JB_ERR_OK == err)
2270 csp->flags |= CSP_FLAG_MODIFIED;
2271 csp->content_type &= ~CT_TABOO;
2276 * Unset CT_GZIP and CT_DEFLATE to remember not
2277 * to modify the Content-Encoding header later.
2279 csp->content_type &= ~CT_GZIP;
2280 csp->content_type &= ~CT_DEFLATE;
2289 /*********************************************************************
2291 * Function : execute_content_filter
2293 * Description : Executes a given content filter.
2296 * 1 : csp = Current client state (buffers, headers, etc...)
2297 * 2 : content_filter = The filter function to execute
2299 * Returns : Pointer to the modified buffer, or
2300 * NULL if filtering failed or wasn't necessary.
2302 *********************************************************************/
2303 char *execute_content_filter(struct client_state *csp, filter_function_ptr content_filter)
2305 if (0 == csp->iob->eod - csp->iob->cur)
2308 * No content (probably status code 301, 302 ...),
2309 * no filtering necessary.
2314 if (JB_ERR_OK != prepare_for_filtering(csp))
2317 * failed to de-chunk or decompress.
2322 if (0 == csp->iob->eod - csp->iob->cur)
2325 * Clown alarm: chunked and/or compressed nothing delivered.
2330 return ((*content_filter)(csp));
2334 /*********************************************************************
2336 * Function : url_actions
2338 * Description : Gets the actions for this URL.
2341 * 1 : http = http_request request for blocked URLs
2342 * 2 : csp = Current client state (buffers, headers, etc...)
2346 *********************************************************************/
2347 void url_actions(struct http_request *http,
2348 struct client_state *csp)
2350 struct file_list *fl;
2351 struct url_actions *b;
2354 init_current_action(csp->action);
2356 for (i = 0; i < MAX_AF_FILES; i++)
2358 if (((fl = csp->actions_list[i]) == NULL) || ((b = fl->f) == NULL))
2363 apply_url_actions(csp->action, http, b);
2370 /*********************************************************************
2372 * Function : apply_url_actions
2374 * Description : Applies a list of URL actions.
2377 * 1 : action = Destination.
2378 * 2 : http = Current URL
2379 * 3 : b = list of URL actions to apply
2383 *********************************************************************/
2384 void apply_url_actions(struct current_action_spec *action,
2385 struct http_request *http,
2386 struct url_actions *b)
2390 /* Should never happen */
2394 for (b = b->next; NULL != b; b = b->next)
2396 if (url_match(b->url, http))
2398 merge_current_action(action, b->action);
2404 /*********************************************************************
2406 * Function : get_forward_override_settings
2408 * Description : Returns forward settings as specified with the
2409 * forward-override{} action. forward-override accepts
2410 * forward lines similar to the one used in the
2411 * configuration file, but without the URL pattern.
2417 * in the configuration file can be replaced with
2418 * the action section:
2420 * {+forward-override{forward .}}
2424 * 1 : csp = Current client state (buffers, headers, etc...)
2426 * Returns : Pointer to forwarding structure in case of success.
2427 * Invalid syntax is fatal.
2429 *********************************************************************/
2430 static const struct forward_spec *get_forward_override_settings(struct client_state *csp)
2432 const char *forward_override_line = csp->action->string[ACTION_STRING_FORWARD_OVERRIDE];
2433 char forward_settings[BUFFER_SIZE];
2434 char *http_parent = NULL;
2435 /* variable names were chosen for consistency reasons. */
2436 struct forward_spec *fwd = NULL;
2440 assert(csp->action->flags & ACTION_FORWARD_OVERRIDE);
2441 /* Should be enforced by load_one_actions_file() */
2442 assert(strlen(forward_override_line) < sizeof(forward_settings) - 1);
2444 /* Create a copy ssplit can modify */
2445 strlcpy(forward_settings, forward_override_line, sizeof(forward_settings));
2447 if (NULL != csp->fwd)
2450 * XXX: Currently necessary to prevent memory
2451 * leaks when the show-url-info cgi page is visited.
2453 unload_forward_spec(csp->fwd);
2457 * allocate a new forward node, valid only for
2458 * the lifetime of this request. Save its location
2459 * in csp as well, so sweep() can free it later on.
2461 fwd = csp->fwd = zalloc(sizeof(*fwd));
2464 log_error(LOG_LEVEL_FATAL,
2465 "can't allocate memory for forward-override{%s}", forward_override_line);
2466 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2469 vec_count = ssplit(forward_settings, " \t", vec, SZ(vec), 1, 1);
2470 if ((vec_count == 2) && !strcasecmp(vec[0], "forward"))
2472 fwd->type = SOCKS_NONE;
2474 /* Parse the parent HTTP proxy host:port */
2475 http_parent = vec[1];
2478 else if (vec_count == 3)
2480 char *socks_proxy = NULL;
2482 if (!strcasecmp(vec[0], "forward-socks4"))
2484 fwd->type = SOCKS_4;
2485 socks_proxy = vec[1];
2487 else if (!strcasecmp(vec[0], "forward-socks4a"))
2489 fwd->type = SOCKS_4A;
2490 socks_proxy = vec[1];
2493 if (NULL != socks_proxy)
2495 /* Parse the SOCKS proxy host[:port] */
2496 fwd->gateway_host = strdup(socks_proxy);
2498 if (NULL != (socks_proxy = strchr(fwd->gateway_host, ':')))
2500 *socks_proxy++ = '\0';
2501 fwd->gateway_port = strtol(socks_proxy, NULL, 0);
2504 if (fwd->gateway_port <= 0)
2506 fwd->gateway_port = 1080;
2509 http_parent = vec[2];
2513 if (NULL == http_parent)
2515 log_error(LOG_LEVEL_FATAL,
2516 "Invalid forward-override syntax in: %s", forward_override_line);
2517 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2520 /* Parse http forwarding settings */
2521 if (strcmp(http_parent, ".") != 0)
2523 fwd->forward_host = strdup(http_parent);
2525 if (NULL != (http_parent = strchr(fwd->forward_host, ':')))
2527 *http_parent++ = '\0';
2528 fwd->forward_port = strtol(http_parent, NULL, 0);
2531 if (fwd->forward_port <= 0)
2533 fwd->forward_port = 8000;
2537 assert (NULL != fwd);
2539 log_error(LOG_LEVEL_CONNECT,
2540 "Overriding forwarding settings based on \'%s\'", forward_override_line);
2546 /*********************************************************************
2548 * Function : forward_url
2550 * Description : Should we forward this to another proxy?
2552 * XXX: Should be changed to make use of csp->fwd.
2555 * 1 : http = http_request request for current URL
2556 * 2 : csp = Current client state (buffers, headers, etc...)
2558 * Returns : Pointer to forwarding information.
2560 *********************************************************************/
2561 const struct forward_spec * forward_url(struct http_request *http,
2562 struct client_state *csp)
2564 static const struct forward_spec fwd_default[1] = { FORWARD_SPEC_INITIALIZER };
2565 struct forward_spec *fwd = csp->config->forward;
2567 if (csp->action->flags & ACTION_FORWARD_OVERRIDE)
2569 return get_forward_override_settings(csp);
2579 if (url_match(fwd->url, http))
2590 /*********************************************************************
2592 * Function : direct_response
2594 * Description : Check if Max-Forwards == 0 for an OPTIONS or TRACE
2595 * request and if so, return a HTTP 501 to the client.
2597 * FIXME: I have a stupid name and I should handle the
2598 * requests properly. Still, what we do here is rfc-
2599 * compliant, whereas ignoring or forwarding are not.
2602 * 1 : csp = Current client state (buffers, headers, etc...)
2604 * Returns : http_response if , NULL if nonmatch or handler fail
2606 *********************************************************************/
2607 struct http_response *direct_response(struct client_state *csp)
2609 struct http_response *rsp;
2610 struct list_entry *p;
2612 if ((0 == strcmpic(csp->http->gpc, "trace"))
2613 || (0 == strcmpic(csp->http->gpc, "options")))
2615 for (p = csp->headers->first; (p != NULL) ; p = p->next)
2617 if (!strncmpic("Max-Forwards:", p->str, 13))
2619 unsigned int max_forwards;
2622 * If it's a Max-Forwards value of zero,
2623 * we have to intercept the request.
2625 if (1 == sscanf(p->str+12, ": %u", &max_forwards) && max_forwards == 0)
2628 * FIXME: We could handle at least TRACE here,
2629 * but that would require a verbatim copy of
2630 * the request which we don't have anymore
2632 log_error(LOG_LEVEL_HEADER,
2633 "Detected header \'%s\' in OPTIONS or TRACE request. Returning 501.",
2636 /* Get mem for response or fail*/
2637 if (NULL == (rsp = alloc_http_response()))
2639 return cgi_error_memory();
2642 if (NULL == (rsp->status = strdup("501 Not Implemented")))
2644 free_http_response(rsp);
2645 return cgi_error_memory();
2649 rsp->reason = RSP_REASON_UNSUPPORTED;
2651 return(finish_http_response(csp, rsp));
2660 /*********************************************************************
2662 * Function : content_filters_enabled
2664 * Description : Checks whether there are any content filters
2665 * enabled for the current request.
2668 * 1 : csp = Current client state (buffers, headers, etc...)
2670 * Returns : TRUE for yes, FALSE otherwise
2672 *********************************************************************/
2673 inline int content_filters_enabled(const struct client_state *csp)
2675 return (((csp->rlist != NULL) &&
2676 (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER]))) ||
2677 (csp->action->flags & (ACTION_DEANIMATE|ACTION_JPEG_INSPECT|ACTION_NO_POPUPS)));