+ char *new_url = NULL;
+ int hits;
+
+ assert(old_url);
+ assert(pcrs_command);
+
+ new_url = pcrs_execute_single_command(old_url, pcrs_command, &hits);
+
+ if (hits == 0)
+ {
+ log_error(LOG_LEVEL_REDIRECTS,
+ "pcrs command \"%s\" didn't change \"%s\".",
+ pcrs_command, old_url);
+ freez(new_url);
+ }
+ else if (hits < 0)
+ {
+ log_error(LOG_LEVEL_REDIRECTS,
+ "executing pcrs command \"%s\" to rewrite %s failed: %s",
+ pcrs_command, old_url, pcrs_strerror(hits));
+ freez(new_url);
+ }
+ else if (strncmpic(new_url, "http://", 7) && strncmpic(new_url, "https://", 8))
+ {
+ log_error(LOG_LEVEL_ERROR,
+ "pcrs command \"%s\" changed \"%s\" to \"%s\" (%u hi%s), "
+ "but the result doesn't look like a valid URL and will be ignored.",
+ pcrs_command, old_url, new_url, hits, (hits == 1) ? "t" : "ts");
+ freez(new_url);
+ }
+ else
+ {
+ log_error(LOG_LEVEL_REDIRECTS,
+ "pcrs command \"%s\" changed \"%s\" to \"%s\" (%u hi%s).",
+ pcrs_command, old_url, new_url, hits, (hits == 1) ? "t" : "ts");
+ }
+
+ return new_url;
+
+}
+
+
+#ifdef FEATURE_FAST_REDIRECTS
+/*********************************************************************
+ *
+ * Function : get_last_url
+ *
+ * Description : Search for the last URL inside a string.
+ * If the string already is a URL, it will
+ * be the first URL found.
+ *
+ * Parameters :
+ * 1 : subject = the string to check
+ * 2 : redirect_mode = +fast-redirect{} mode
+ *
+ * Returns : NULL if no URL was found, or
+ * the last URL found.
+ *
+ *********************************************************************/
+char *get_last_url(char *subject, const char *redirect_mode)
+{
+ char *new_url = NULL;
+ char *tmp;
+
+ assert(subject);
+ assert(redirect_mode);
+
+ subject = strdup(subject);
+ if (subject == NULL)
+ {
+ log_error(LOG_LEVEL_ERROR, "Out of memory while searching for redirects.");
+ return NULL;
+ }
+
+ if (0 == strcmpic(redirect_mode, "check-decoded-url"))
+ {
+ log_error(LOG_LEVEL_REDIRECTS, "Decoding \"%s\" if necessary.", subject);
+ new_url = url_decode(subject);
+ if (new_url != NULL)
+ {
+ freez(subject);
+ subject = new_url;
+ }
+ else
+ {
+ log_error(LOG_LEVEL_ERROR, "Unable to decode \"%s\".", subject);
+ }
+ }
+
+ log_error(LOG_LEVEL_REDIRECTS, "Checking \"%s\" for redirects.", subject);
+
+ /*
+ * Find the last URL encoded in the request
+ */
+ tmp = subject;
+ while ((tmp = strstr(tmp, "http://")) != NULL)
+ {
+ new_url = tmp++;
+ }
+ tmp = (new_url != NULL) ? new_url : subject;
+ while ((tmp = strstr(tmp, "https://")) != NULL)
+ {
+ new_url = tmp++;
+ }
+
+ if ((new_url != NULL)
+ && ( (new_url != subject)
+ || (0 == strncmpic(subject, "http://", 7))
+ || (0 == strncmpic(subject, "https://", 8))
+ ))
+ {
+ /*
+ * Return new URL if we found a redirect
+ * or if the subject already was a URL.
+ *
+ * The second case makes sure that we can
+ * chain get_last_url after another redirection check
+ * (like rewrite_url) without losing earlier redirects.
+ */
+ new_url = strdup(new_url);
+ freez(subject);
+ return new_url;
+ }
+
+ freez(subject);
+ return NULL;
+
+}
+#endif /* def FEATURE_FAST_REDIRECTS */
+
+
+/*********************************************************************
+ *
+ * Function : redirect_url
+ *
+ * Description : Checks if Privoxy should answer the request with
+ * a HTTP redirect and generates the redirect if
+ * necessary.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : NULL if the request can pass, HTTP redirect otherwise.
+ *
+ *********************************************************************/
+struct http_response *redirect_url(struct client_state *csp)
+{
+ struct http_response *rsp;
+#ifdef FEATURE_FAST_REDIRECTS
+ /*
+ * XXX: Do we still need FEATURE_FAST_REDIRECTS
+ * as compile-time option? The user can easily disable
+ * it in his action file.
+ */
+ char * redirect_mode;
+#endif /* def FEATURE_FAST_REDIRECTS */
+ char *old_url = NULL;
+ char *new_url = NULL;
+ char *redirection_string;
+
+ if ((csp->action->flags & ACTION_REDIRECT))
+ {
+ redirection_string = csp->action->string[ACTION_STRING_REDIRECT];
+
+ /*
+ * If the redirection string begins with 's',
+ * assume it's a pcrs command, otherwise treat it as
+ * properly formatted URL and use it for the redirection
+ * directly.
+ *
+ * According to RFC 2616 section 14.30 the URL
+ * has to be absolute and if the user tries:
+ * +redirect{shit/this/will/be/parsed/as/pcrs_command.html}
+ * she would get undefined results anyway.
+ *
+ */
+
+ if (*redirection_string == 's')
+ {
+ old_url = csp->http->url;
+ new_url = rewrite_url(old_url, redirection_string);
+ }
+ else
+ {
+ log_error(LOG_LEVEL_REDIRECTS,
+ "No pcrs command recognized, assuming that \"%s\" is already properly formatted.",
+ redirection_string);
+ new_url = strdup(redirection_string);
+ }
+ }
+
+#ifdef FEATURE_FAST_REDIRECTS
+ if ((csp->action->flags & ACTION_FAST_REDIRECTS))
+ {
+ redirect_mode = csp->action->string[ACTION_STRING_FAST_REDIRECTS];
+
+ /*
+ * If it exists, use the previously rewritten URL as input
+ * otherwise just use the old path.
+ */
+ old_url = (new_url != NULL) ? new_url : strdup(csp->http->path);
+ new_url = get_last_url(old_url, redirect_mode);
+ freez(old_url);
+ }
+
+ /*
+ * Disable redirect checkers, so that they
+ * will be only run more than once if the user
+ * also enables them through tags.
+ *
+ * From a performance point of view
+ * it doesn't matter, but the duplicated
+ * log messages are annoying.
+ */
+ csp->action->flags &= ~ACTION_FAST_REDIRECTS;
+#endif /* def FEATURE_FAST_REDIRECTS */
+ csp->action->flags &= ~ACTION_REDIRECT;
+
+ /* Did any redirect action trigger? */
+ if (new_url)
+ {
+ if (0 == strcmpic(new_url, csp->http->url))
+ {
+ log_error(LOG_LEVEL_ERROR,
+ "New URL \"%s\" and old URL \"%s\" are the same. Redirection loop prevented.",
+ csp->http->url, new_url);
+ freez(new_url);
+ }
+ else
+ {
+ log_error(LOG_LEVEL_REDIRECTS, "New URL is: %s", new_url);
+
+ if (NULL == (rsp = alloc_http_response()))
+ {
+ freez(new_url);
+ return cgi_error_memory();
+ }
+
+ if ( enlist_unique_header(rsp->headers, "Location", new_url)
+ || (NULL == (rsp->status = strdup("302 Local Redirect from Privoxy"))) )
+ {
+ freez(new_url);
+ free_http_response(rsp);
+ return cgi_error_memory();
+ }
+ rsp->reason = RSP_REASON_REDIRECTED;
+ freez(new_url);
+
+ return finish_http_response(csp, rsp);
+ }
+ }
+
+ /* Only reached if no redirect is required */
+ return NULL;
+
+}
+
+
+#ifdef FEATURE_IMAGE_BLOCKING
+/*********************************************************************
+ *
+ * Function : is_imageurl
+ *
+ * Description : Given a URL, decide whether it is an image or not,
+ * using either the info from a previous +image action
+ * or, #ifdef FEATURE_IMAGE_DETECT_MSIE, and the browser
+ * is MSIE and not on a Mac, tell from the browser's accept
+ * header.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : True (nonzero) if URL is an image, false (0)
+ * otherwise
+ *
+ *********************************************************************/
+int is_imageurl(const struct client_state *csp)
+{
+#ifdef FEATURE_IMAGE_DETECT_MSIE
+ char *tmp;
+
+ tmp = get_header_value(csp->headers, "User-Agent:");
+ if (tmp && strstr(tmp, "MSIE") && !strstr(tmp, "Mac_"))
+ {
+ tmp = get_header_value(csp->headers, "Accept:");
+ if (tmp && strstr(tmp, "image/gif"))
+ {
+ /* Client will accept HTML. If this seems counterintuitive,
+ * blame Microsoft.
+ */
+ return(0);
+ }
+ else
+ {
+ return(1);
+ }
+ }
+#endif /* def FEATURE_IMAGE_DETECT_MSIE */
+
+ return ((csp->action->flags & ACTION_IMAGE) != 0);
+
+}
+#endif /* def FEATURE_IMAGE_BLOCKING */
+
+
+#ifdef FEATURE_TRUST
+/*********************************************************************
+ *
+ * Function : is_untrusted_url
+ *
+ * Description : Should we "distrust" this URL (and block it)?
+ *
+ * Yes if it matches a line in the trustfile, or if the
+ * referrer matches a line starting with "+" in the
+ * trustfile.
+ * No otherwise.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : 0 => trusted, 1 => untrusted
+ *
+ *********************************************************************/
+int is_untrusted_url(const struct client_state *csp)
+{
+ struct file_list *fl;
+ struct block_spec *b;
+ struct url_spec **trusted_url;
+ struct http_request rhttp[1];
+ const char * referer;
+ jb_err err;
+
+ /*
+ * If we don't have a trustlist, we trust everybody
+ */
+ if (((fl = csp->tlist) == NULL) || ((b = fl->f) == NULL))
+ {
+ return 0;
+ }
+
+ memset(rhttp, '\0', sizeof(*rhttp));
+
+ /*
+ * Do we trust the request URL itself?
+ */
+ for (b = b->next; b ; b = b->next)
+ {
+ if (url_match(b->url, csp->http))
+ {
+ return b->reject;
+ }
+ }
+
+ if (NULL == (referer = get_header_value(csp->headers, "Referer:")))
+ {
+ /* no referrer was supplied */
+ return 1;
+ }
+
+
+ /*
+ * If not, do we maybe trust its referrer?
+ */
+ err = parse_http_url(referer, rhttp, csp);
+ if (err)
+ {
+ return 1;
+ }
+
+ for (trusted_url = csp->config->trust_list; *trusted_url != NULL; trusted_url++)
+ {
+ if (url_match(*trusted_url, rhttp))
+ {
+ /* if the URL's referrer is from a trusted referrer, then
+ * add the target spec to the trustfile as an unblocked
+ * domain and return 0 (which means it's OK).
+ */
+
+ FILE *fp;
+
+ if (NULL != (fp = fopen(csp->config->trustfile, "a")))
+ {
+ char * path;
+ char * path_end;
+ char * new_entry = strdup("~");
+
+ string_append(&new_entry, csp->http->hostport);
+
+ path = csp->http->path;
+ if ( (path[0] == '/')
+ && (path[1] == '~')
+ && ((path_end = strchr(path + 2, '/')) != NULL))
+ {
+ /* since this path points into a user's home space
+ * be sure to include this spec in the trustfile.
+ */
+ int path_len = path_end - path; /* save offset */
+ path = strdup(path); /* Copy string */
+ if (path != NULL)
+ {
+ path_end = path + path_len; /* regenerate ptr to new buffer */
+ *(path_end + 1) = '\0'; /* Truncate path after '/' */
+ }
+ string_join(&new_entry, path);
+ }
+
+ /*
+ * Give a reason for generating this entry.
+ */
+ string_append(&new_entry, " # Trusted referrer was: ");
+ string_append(&new_entry, referer);
+
+ if (new_entry != NULL)
+ {
+ if (-1 == fprintf(fp, "%s\n", new_entry))
+ {
+ log_error(LOG_LEVEL_ERROR, "Failed to append \'%s\' to trustfile \'%s\': %E",
+ new_entry, csp->config->trustfile);
+ }
+ freez(new_entry);
+ }
+ else
+ {
+ /* FIXME: No way to handle out-of memory, so mostly ignoring it */
+ log_error(LOG_LEVEL_ERROR, "Out of memory adding pattern to trust file");
+ }
+
+ fclose(fp);
+ }
+ else
+ {
+ log_error(LOG_LEVEL_ERROR, "Failed to append new entry for \'%s\' to trustfile \'%s\': %E",
+ csp->http->hostport, csp->config->trustfile);
+ }
+ return 0;
+ }
+ }
+
+ return 1;
+}
+#endif /* def FEATURE_TRUST */
+
+
+/*********************************************************************
+ *
+ * Function : pcrs_filter_response
+ *
+ * Description : Execute all text substitutions from all applying
+ * +filter actions on the text buffer that's been
+ * accumulated in csp->iob->buf.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : a pointer to the (newly allocated) modified buffer.
+ * or NULL if there were no hits or something went wrong
+ *
+ *********************************************************************/
+static char *pcrs_filter_response(struct client_state *csp)
+{
+ int hits=0;
+ size_t size, prev_size;
+
+ char *old = NULL;
+ char *new = NULL;
+ pcrs_job *job;
+
+ struct file_list *fl;
+ struct re_filterfile_spec *b;
+ struct list_entry *filtername;
+
+ int i, found_filters = 0;
+
+ /*
+ * Sanity first
+ */
+ if (csp->iob->cur >= csp->iob->eod)
+ {
+ return(NULL);
+ }
+
+ /*
+ * Need to check the set of re_filterfiles...
+ */
+ for (i = 0; i < MAX_AF_FILES; i++)
+ {
+ fl = csp->rlist[i];
+ if (NULL != fl)
+ {
+ if (NULL != fl->f)
+ {
+ found_filters = 1;
+ break;
+ }
+ }
+ }
+
+ if (0 == found_filters)
+ {
+ log_error(LOG_LEVEL_ERROR, "Inconsistent configuration: "
+ "content filtering enabled, but no content filters available.");
+ return(NULL);
+ }
+
+ size = (size_t)(csp->iob->eod - csp->iob->cur);
+ old = csp->iob->cur;
+
+ for (i = 0; i < MAX_AF_FILES; i++)
+ {
+ fl = csp->rlist[i];
+ if ((NULL == fl) || (NULL == fl->f))
+ {
+ /*
+ * Either there are no filter files
+ * left, or this filter file just
+ * contains no valid filters.
+ *
+ * Continue to be sure we don't miss
+ * valid filter files that are chained
+ * after empty or invalid ones.
+ */
+ continue;
+ }
+ /*
+ * For all applying +filter actions, look if a filter by that
+ * name exists and if yes, execute it's pcrs_joblist on the
+ * buffer.
+ */
+ for (b = fl->f; b; b = b->next)
+ {
+ if (b->type != FT_CONTENT_FILTER)
+ {
+ /* Skip header filters */
+ continue;
+ }
+
+ for (filtername = csp->action->multi[ACTION_MULTI_FILTER]->first;
+ filtername ; filtername = filtername->next)
+ {
+ if (strcmp(b->name, filtername->str) == 0)
+ {
+ int current_hits = 0; /* Number of hits caused by this filter */
+ int job_number = 0; /* Which job we're currently executing */
+ int job_hits = 0; /* How many hits the current job caused */
+ pcrs_job *joblist = b->joblist;
+
+ if (b->dynamic) joblist = compile_dynamic_pcrs_job_list(csp, b);
+
+ if (NULL == joblist)
+ {
+ log_error(LOG_LEVEL_RE_FILTER, "Filter %s has empty joblist. Nothing to do.", b->name);
+ continue;
+ }
+
+ prev_size = size;
+ /* Apply all jobs from the joblist */
+ for (job = joblist; NULL != job; job = job->next)
+ {
+ job_number++;
+ job_hits = pcrs_execute(job, old, size, &new, &size);
+
+ if (job_hits >= 0)
+ {
+ /*
+ * That went well. Continue filtering
+ * and use the result of this job as
+ * input for the next one.
+ */
+ current_hits += job_hits;
+ if (old != csp->iob->cur)
+ {
+ freez(old);
+ }
+ old = new;
+ }
+ else
+ {
+ /*
+ * This job caused an unexpected error. Inform the user
+ * and skip the rest of the jobs in this filter. We could
+ * continue with the next job, but usually the jobs
+ * depend on each other or are similar enough to
+ * fail for the same reason.
+ *
+ * At the moment our pcrs expects the error codes of pcre 3.4,
+ * but newer pcre versions can return additional error codes.
+ * As a result pcrs_strerror()'s error message might be
+ * "Unknown error ...", therefore we print the numerical value
+ * as well.
+ *
+ * XXX: Is this important enough for LOG_LEVEL_ERROR or
+ * should we use LOG_LEVEL_RE_FILTER instead?
+ */
+ log_error(LOG_LEVEL_ERROR, "Skipped filter \'%s\' after job number %u: %s (%d)",
+ b->name, job_number, pcrs_strerror(job_hits), job_hits);
+ break;
+ }
+ }
+
+ if (b->dynamic) pcrs_free_joblist(joblist);
+
+ log_error(LOG_LEVEL_RE_FILTER,
+ "filtering %s%s (size %d) with \'%s\' produced %d hits (new size %d).",
+ csp->http->hostport, csp->http->path, prev_size, b->name, current_hits, size);
+
+ hits += current_hits;
+ }
+ }
+ }
+ }
+
+ /*
+ * If there were no hits, destroy our copy and let
+ * chat() use the original in csp->iob
+ */
+ if (!hits)
+ {
+ freez(new);
+ return(NULL);
+ }
+
+ csp->flags |= CSP_FLAG_MODIFIED;
+ csp->content_length = size;
+ IOB_RESET(csp);
+
+ return(new);
+
+}
+
+
+/*********************************************************************
+ *
+ * Function : gif_deanimate_response
+ *
+ * Description : Deanimate the GIF image that has been accumulated in
+ * csp->iob->buf, set csp->content_length to the modified
+ * size and raise the CSP_FLAG_MODIFIED flag.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : a pointer to the (newly allocated) modified buffer.
+ * or NULL in case something went wrong.
+ *
+ *********************************************************************/
+static char *gif_deanimate_response(struct client_state *csp)
+{
+ struct binbuffer *in, *out;
+ char *p;
+ size_t size;
+
+ size = (size_t)(csp->iob->eod - csp->iob->cur);
+
+ if ( (NULL == (in = (struct binbuffer *)zalloc(sizeof *in )))
+ || (NULL == (out = (struct binbuffer *)zalloc(sizeof *out))) )
+ {
+ log_error(LOG_LEVEL_DEANIMATE, "failed! (no mem)");
+ return NULL;
+ }
+
+ in->buffer = csp->iob->cur;
+ in->size = size;
+
+ if (gif_deanimate(in, out, strncmp("last", csp->action->string[ACTION_STRING_DEANIMATE], 4)))
+ {
+ log_error(LOG_LEVEL_DEANIMATE, "failed! (gif parsing)");
+ freez(in);
+ buf_free(out);
+ return(NULL);
+ }
+ else
+ {
+ if ((int)size == out->offset)
+ {
+ log_error(LOG_LEVEL_DEANIMATE, "GIF not changed.");
+ }
+ else
+ {
+ log_error(LOG_LEVEL_DEANIMATE, "Success! GIF shrunk from %d bytes to %d.", size, out->offset);
+ }
+ csp->content_length = out->offset;
+ csp->flags |= CSP_FLAG_MODIFIED;
+ p = out->buffer;
+ freez(in);
+ freez(out);
+ return(p);
+ }
+
+}
+
+
+/*********************************************************************
+ *
+ * Function : jpeg_inspect_response
+ *
+ * Description :
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : a pointer to the (newly allocated) modified buffer
+ * or NULL in case something went wrong.
+ *
+ *********************************************************************/
+static char *jpeg_inspect_response(struct client_state *csp)
+{
+ struct binbuffer *in = NULL;
+ struct binbuffer *out = NULL;
+ char *p = NULL;
+ size_t size;
+
+ size = (size_t)(csp->iob->eod - csp->iob->cur);
+
+ if (NULL == (in = (struct binbuffer *)zalloc(sizeof *in )))
+ {
+ log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg no mem 1)");
+ return NULL;
+ }
+
+ if (NULL == (out = (struct binbuffer *)zalloc(sizeof *out)))
+ {
+ log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg no mem 2)");
+ return NULL;
+ }
+
+ in->buffer = csp->iob->cur;
+ in->size = size;
+
+ /*
+ * Calling jpeg_inspect has the side-effect of creating and
+ * modifying the image buffer of "out" directly.
+ */
+ if (jpeg_inspect(in, out))
+ {
+ log_error(LOG_LEVEL_DEANIMATE, "failed! (jpeg parsing)");
+ freez(in);
+ buf_free(out);
+ return(NULL);
+
+ }
+ else
+ {
+ csp->content_length = out->offset;
+ csp->flags |= CSP_FLAG_MODIFIED;
+ p = out->buffer;
+ freez(in);
+ freez(out);
+ return(p);
+ }
+
+}
+
+
+/*********************************************************************
+ *
+ * Function : get_filter_function
+ *
+ * Description : Decides which content filter function has
+ * to be applied (if any).
+ *
+ * XXX: Doesn't handle filter_popups()
+ * because of the different prototype. Probably
+ * we should ditch filter_popups() anyway, it's
+ * even less reliable than popup blocking based
+ * on pcrs filters.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : The content filter function to run, or
+ * NULL if no content filter is active
+ *
+ *********************************************************************/
+filter_function_ptr get_filter_function(struct client_state *csp)
+{
+ filter_function_ptr filter_function = NULL;
+
+ /*
+ * Are we enabling text mode by force?
+ */
+ if (csp->action->flags & ACTION_FORCE_TEXT_MODE)
+ {
+ /*
+ * Do we really have to?
+ */
+ if (csp->content_type & CT_TEXT)
+ {
+ log_error(LOG_LEVEL_HEADER, "Text mode is already enabled.");
+ }
+ else
+ {
+ csp->content_type |= CT_TEXT;
+ log_error(LOG_LEVEL_HEADER, "Text mode enabled by force. Take cover!");
+ }
+ }
+
+ if (!(csp->content_type & CT_DECLARED))
+ {
+ /*
+ * The server didn't bother to declare a MIME-Type.
+ * Assume it's text that can be filtered.
+ *
+ * This also regulary happens with 304 responses,
+ * therefore logging anything here would cause
+ * too much noise.
+ */
+ csp->content_type |= CT_TEXT;
+ }
+
+ /*
+ * Choose the applying filter function based on
+ * the content type and action settings.
+ */
+ if ((csp->content_type & CT_TEXT) &&
+ (csp->rlist != NULL) &&
+ (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER])))
+ {
+ filter_function = pcrs_filter_response;
+ }
+ else if ((csp->content_type & CT_GIF) &&
+ (csp->action->flags & ACTION_DEANIMATE))
+ {
+ filter_function = gif_deanimate_response;