X-Git-Url: http://www.privoxy.org/gitweb/?p=privoxy.git;a=blobdiff_plain;f=parsers.c;h=14ce3b6bb14c4fcf08753818ae3fc921b91e8c3f;hp=f4e67ade510c84ca78021a747f635c548da7a29f;hb=c5c165e95d33f316bc3d9334101b81aefb54d30b;hpb=3f442dfbb7e2d279820abb6913adefafb125cc75 diff --git a/parsers.c b/parsers.c index f4e67ade..14ce3b6b 100644 --- a/parsers.c +++ b/parsers.c @@ -1,4 +1,4 @@ -const char parsers_rcs[] = "$Id: parsers.c,v 1.73 2006/09/23 13:26:38 roro Exp $"; +const char parsers_rcs[] = "$Id: parsers.c,v 1.96 2007/04/12 12:53:58 fabiankeil Exp $"; /********************************************************************* * * File : $Source: /cvsroot/ijbswa/current/parsers.c,v $ @@ -10,10 +10,14 @@ const char parsers_rcs[] = "$Id: parsers.c,v 1.73 2006/09/23 13:26:38 roro Exp $ * `client_uagent', `client_x_forwarded', * `client_x_forwarded_adder', `client_xtra_adder', * `content_type', `crumble', `destroy_list', `enlist', - * `flush_socket', ``get_header', `sed', - * and `server_set_cookie'. - * - * Copyright : Written by and Copyright (C) 2001 the SourceForge + * `flush_socket', ``get_header', `sed', `filter_header' + * `server_content_encoding', `server_content_disposition', + * `server_last_modified', `client_accept_language', + * `crunch_client_header', `client_if_modified_since', + * `client_if_none_match', `get_destination_from_headers', + * `parse_header_time', `decompress_iob' and `server_set_cookie'. + * + * Copyright : Written by and Copyright (C) 2001-2007 the SourceForge * Privoxy team. http://www.privoxy.org/ * * Based on the Internet Junkbuster originally written @@ -40,6 +44,137 @@ const char parsers_rcs[] = "$Id: parsers.c,v 1.73 2006/09/23 13:26:38 roro Exp $ * * Revisions : * $Log: parsers.c,v $ + * Revision 1.96 2007/04/12 12:53:58 fabiankeil + * Log a warning if the content is compressed, filtering is + * enabled and Privoxy was compiled without zlib support. + * Closes FR#1673938. + * + * Revision 1.95 2007/03/25 14:26:40 fabiankeil + * - Fix warnings when compiled with glibc. + * - Don't use crumble() for cookie crunching. + * - Move cookie time parsing into parse_header_time(). + * - Let parse_header_time() return a jb_err code + * instead of a pointer that can only be used to + * check for NULL anyway. + * + * Revision 1.94 2007/03/21 12:23:53 fabiankeil + * - Add better protection against malicious gzip headers. + * - Stop logging the first hundred bytes of decompressed content. + * It looks like it's working and there is always debug 16. + * - Log the content size after decompression in decompress_iob() + * instead of pcrs_filter_response(). + * + * Revision 1.93 2007/03/20 15:21:44 fabiankeil + * - Use dedicated header filter actions instead of abusing "filter". + * Replace "filter-client-headers" and "filter-client-headers" + * with "server-header-filter" and "client-header-filter". + * - Remove filter_client_header() and filter_client_header(), + * filter_header() now checks the shiny new + * CSP_FLAG_CLIENT_HEADER_PARSING_DONE flag instead. + * + * Revision 1.92 2007/03/05 13:25:32 fabiankeil + * - Cosmetical changes for LOG_LEVEL_RE_FILTER messages. + * - Handle "Cookie:" and "Connection:" headers a bit smarter + * (don't crunch them just to recreate them later on). + * - Add another non-standard time format for the cookie + * expiration date detection. + * - Fix a valgrind warning. + * + * Revision 1.91 2007/02/24 12:27:32 fabiankeil + * Improve cookie expiration date detection. + * + * Revision 1.90 2007/02/08 19:12:35 fabiankeil + * Don't run server_content_length() the first time + * sed() parses server headers; only adjust the + * Content-Length header if the page was modified. + * + * Revision 1.89 2007/02/07 16:52:11 fabiankeil + * Fix log messages regarding the cookie time format + * (cookie and request URL were mixed up). + * + * Revision 1.88 2007/02/07 11:27:12 fabiankeil + * - Let decompress_iob() + * - not corrupt the content if decompression fails + * early. (the first byte(s) were lost). + * - use pointer arithmetics with defined outcome for + * a change. + * - Use a different kludge to remember a failed decompression. + * + * Revision 1.87 2007/01/31 16:21:38 fabiankeil + * Search for Max-Forwards headers case-insensitive, + * don't generate the "501 unsupported" message for invalid + * Max-Forwards values and don't increase negative ones. + * + * Revision 1.86 2007/01/30 13:05:26 fabiankeil + * - Let server_set_cookie() check the expiration date + * of cookies and don't touch the ones that are already + * expired. Fixes problems with low quality web applications + * as described in BR 932612. + * + * - Adjust comment in client_max_forwards to reality; + * remove invalid Max-Forwards headers. + * + * Revision 1.85 2007/01/26 15:33:46 fabiankeil + * Stop filter_header() from unintentionally removing + * empty header lines that were enlisted by the continue + * hack. + * + * Revision 1.84 2007/01/24 12:56:52 fabiankeil + * - Repeat the request URL before logging any headers. + * Makes reading the log easier in case of simultaneous requests. + * - If there are more than one Content-Type headers in one request, + * use the first one and remove the others. + * - Remove "newval" variable in server_content_type(). + * It's only used once. + * + * Revision 1.83 2007/01/12 15:03:02 fabiankeil + * Correct a cast, check inflateEnd() exit code + * to see if we have to, replace sprintf calls + * with snprintf. + * + * Revision 1.82 2007/01/01 19:36:37 fabiankeil + * Integrate a modified version of Wil Mahan's + * zlib patch (PR #895531). + * + * Revision 1.81 2006/12/31 22:21:33 fabiankeil + * Skip empty filter files in filter_header() + * but don't ignore the ones that come afterwards. + * Fixes BR 1619208, this time for real. + * + * Revision 1.80 2006/12/29 19:08:22 fabiankeil + * Reverted parts of my last commit + * to keep error handling working. + * + * Revision 1.79 2006/12/29 18:04:40 fabiankeil + * Fixed gcc43 conversion warnings. + * + * Revision 1.78 2006/12/26 17:19:20 fabiankeil + * Bringing back the "useless" localtime() call + * I removed in revision 1.67. On some platforms + * it's necessary to prevent time zone offsets. + * + * Revision 1.77 2006/12/07 18:44:26 fabiankeil + * Rebuild request URL in get_destination_from_headers() + * to make sure redirect{pcrs command} works as expected + * for intercepted requests. + * + * Revision 1.76 2006/12/06 19:52:25 fabiankeil + * Added get_destination_from_headers(). + * + * Revision 1.75 2006/11/13 19:05:51 fabiankeil + * Make pthread mutex locking more generic. Instead of + * checking for OSX and OpenBSD, check for FEATURE_PTHREAD + * and use mutex locking unless there is an _r function + * available. Better safe than sorry. + * + * Fixes "./configure --disable-pthread" and should result + * in less threading-related problems on pthread-using platforms, + * but it still doesn't fix BR#1122404. + * + * Revision 1.74 2006/10/02 16:59:12 fabiankeil + * The special header "X-Filter: No" now disables + * header filtering as well. + * * Revision 1.73 2006/09/23 13:26:38 roro * Replace TABs by spaces in source code. * @@ -527,19 +662,29 @@ const char parsers_rcs[] = "$Id: parsers.c,v 1.73 2006/09/23 13:26:38 roro Exp $ #include #include #include + +#ifdef __GLIBC__ +/* + * Convince GNU's libc to provide a strptime prototype. + */ +#define __USE_XOPEN +#endif /*__GLIBC__ */ #include +#ifdef FEATURE_ZLIB +#include +#endif + #if !defined(_WIN32) && !defined(__OS2__) #include #endif #include "project.h" -#ifdef OSX_DARWIN -#include +#ifdef FEATURE_PTHREAD #include "jcc.h" /* jcc.h is for mutex semapores only */ -#endif /* def OSX_DARWIN */ +#endif /* def FEATURE_PTHREAD */ #include "list.h" #include "parsers.h" #include "encode.h" @@ -548,6 +693,7 @@ const char parsers_rcs[] = "$Id: parsers.c,v 1.73 2006/09/23 13:26:38 roro Exp $ #include "jbsockets.h" #include "miscutil.h" #include "list.h" +#include "actions.h" #ifndef HAVE_STRPTIME #include "strptime.h" @@ -557,7 +703,7 @@ const char parsers_h_rcs[] = PARSERS_H_VERSION; /* Fix a problem with Solaris. There should be no effect on other * platforms. - * Solaris's isspace() is a macro which uses it's argument directly + * Solaris's isspace() is a macro which uses its argument directly * as an array index. Therefore we need to make sure that high-bit * characters generate +ve values, and ideally we also want to make * the argument match the declared parameter type of "int". @@ -568,6 +714,8 @@ const char parsers_h_rcs[] = PARSERS_H_VERSION; #define ijb_isupper(__X) isupper((int)(unsigned char)(__X)) #define ijb_tolower(__X) tolower((int)(unsigned char)(__X)) +jb_err header_tagger(struct client_state *csp, char *header); +jb_err scan_headers(struct client_state *csp); const struct parsers client_patterns[] = { { "referer:", 8, client_referrer }, @@ -581,23 +729,22 @@ const struct parsers client_patterns[] = { { "Host:", 5, client_host }, { "if-modified-since:", 18, client_if_modified_since }, { "Keep-Alive:", 11, crumble }, - { "connection:", 11, crumble }, + { "connection:", 11, connection }, { "proxy-connection:", 17, crumble }, { "max-forwards:", 13, client_max_forwards }, { "Accept-Language:", 16, client_accept_language }, { "if-none-match:", 14, client_if_none_match }, { "X-Filter:", 9, client_x_filter }, { "*", 0, crunch_client_header }, - { "*", 0, filter_client_header }, + { "*", 0, filter_header }, { NULL, 0, NULL } }; const struct parsers server_patterns[] = { { "HTTP", 4, server_http }, { "set-cookie:", 11, server_set_cookie }, - { "connection:", 11, crumble }, + { "connection:", 11, connection }, { "Content-Type:", 13, server_content_type }, - { "Content-Length:", 15, server_content_length }, { "Content-MD5:", 12, server_content_md5 }, { "Content-Encoding:", 17, server_content_encoding }, { "Transfer-Encoding:", 18, server_transfer_coding }, @@ -605,13 +752,16 @@ const struct parsers server_patterns[] = { { "content-disposition:", 20, server_content_disposition }, { "Last-Modified:", 14, server_last_modified }, { "*", 0, crunch_server_header }, - { "*", 0, filter_server_header }, + { "*", 0, filter_header }, { NULL, 0, NULL } }; const struct parsers server_patterns_light[] = { { "Content-Length:", 15, server_content_length }, { "Transfer-Encoding:", 18, server_transfer_coding }, +#ifdef FEATURE_ZLIB + { "Content-Encoding:", 17, server_content_encoding }, +#endif /* def FEATURE_ZLIB */ { NULL, 0, NULL } }; @@ -693,9 +843,9 @@ jb_err add_to_iob(struct client_state *csp, char *buf, int n) if (n <= 0) return JB_ERR_OK; - used = iob->eod - iob->buf; - offset = iob->cur - iob->buf; - need = used + n + 1; + used = (size_t)(iob->eod - iob->buf); + offset = (size_t)(iob->cur - iob->buf); + need = used + (size_t)n + 1; /* * If the buffer can't hold the new data, extend it first. @@ -745,6 +895,380 @@ jb_err add_to_iob(struct client_state *csp, char *buf, int n) } +#ifdef FEATURE_ZLIB +/********************************************************************* + * + * Function : decompress_iob + * + * Description : Decompress buffered page, expanding the + * buffer as necessary. csp->iob->cur + * should point to the the beginning of the + * compressed data block. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * + * Returns : JB_ERR_OK on success, + * JB_ERR_MEMORY if out-of-memory limit reached, and + * JB_ERR_COMPRESS if error decompressing buffer. + * + *********************************************************************/ +jb_err decompress_iob(struct client_state *csp) +{ + char *buf; /* new, uncompressed buffer */ + char *cur; /* Current iob position (to keep the original + * iob->cur unmodified if we return early) */ + size_t bufsize; /* allocated size of the new buffer */ + size_t old_size; /* Content size before decompression */ + size_t skip_size; /* Number of bytes at the beginning of the iob + that we should NOT decompress. */ + int status; /* return status of the inflate() call */ + z_stream zstr; /* used by calls to zlib */ + + assert(csp->iob->cur - csp->iob->buf > 0); + assert(csp->iob->eod - csp->iob->cur > 0); + + bufsize = csp->iob->size; + skip_size = (size_t)(csp->iob->cur - csp->iob->buf); + old_size = (size_t)(csp->iob->eod - csp->iob->cur); + + cur = csp->iob->cur; + + if (bufsize < 10) + { + /* + * This is to protect the parsing of gzipped data, + * but it should(?) be valid for deflated data also. + */ + log_error (LOG_LEVEL_ERROR, "Buffer too small decompressing iob"); + return JB_ERR_COMPRESS; + } + + if (csp->content_type & CT_GZIP) + { + /* + * Our task is slightly complicated by the facts that data + * compressed by gzip does not include a zlib header, and + * that there is no easily accessible interface in zlib to + * handle a gzip header. We strip off the gzip header by + * hand, and later inform zlib not to expect a header. + */ + + /* + * Strip off the gzip header. Please see RFC 1952 for more + * explanation of the appropriate fields. + */ + if ((*cur++ != (char)0x1f) + || (*cur++ != (char)0x8b) + || (*cur++ != Z_DEFLATED)) + { + log_error (LOG_LEVEL_ERROR, "Invalid gzip header when decompressing"); + return JB_ERR_COMPRESS; + } + else + { + int flags = *cur++; + /* + * XXX: These magic numbers should be replaced + * with macros to give a better idea what they do. + */ + if (flags & 0xe0) + { + /* The gzip header has reserved bits set; bail out. */ + log_error (LOG_LEVEL_ERROR, "Invalid gzip header flags when decompressing"); + return JB_ERR_COMPRESS; + } + cur += 6; + + /* Skip extra fields if necessary. */ + if (flags & 0x04) + { + /* + * Skip a given number of bytes, specified + * as a 16-bit little-endian value. + */ + /* + * XXX: This code used to be: + * + * csp->iob->cur += *csp->iob->cur++ + (*csp->iob->cur++ << 8); + * + * which I had to change into: + * + * cur += *cur++ + (*cur++ << 8); + * + * at which point gcc43 finally noticed that the value + * of cur is undefined (it depends on which of the + * summands is evaluated first). + * + * I haven't come across a site where this + * code is actually executed yet, but I hope + * it works anyway. + */ + int skip_bytes; + skip_bytes = *cur++; + skip_bytes = *cur++ << 8; + + assert(skip_bytes == *csp->iob->cur - 2 + ((*csp->iob->cur - 1) << 8)); + + /* + * The number of bytes to skip should be positive + * and we'd like to stay in the buffer. + */ + if((skip_bytes < 0) || (skip_bytes >= (csp->iob->eod - cur))) + { + log_error (LOG_LEVEL_ERROR, + "Unreasonable amount of bytes to skip (%d). Stopping decompression", + skip_bytes); + return JB_ERR_COMPRESS; + } + log_error (LOG_LEVEL_INFO, + "Skipping %d bytes for gzip compression. Does this sound right?", + skip_bytes); + cur += skip_bytes; + } + + /* Skip the filename if necessary. */ + if (flags & 0x08) + { + /* A null-terminated string is supposed to follow. */ + while (*cur++ && (cur < csp->iob->eod)); + + } + + /* Skip the comment if necessary. */ + if (flags & 0x10) + { + /* A null-terminated string is supposed to follow. */ + while (*cur++ && (cur < csp->iob->eod)); + } + + /* Skip the CRC if necessary. */ + if (flags & 0x02) + { + cur += 2; + } + + if (cur >= csp->iob->eod) + { + /* + * If the current position pointer reached or passed + * the buffer end, we were obviously tricked to skip + * too much. + */ + log_error (LOG_LEVEL_ERROR, + "Malformed gzip header detected. Aborting decompression."); + return JB_ERR_COMPRESS; + } + } + } + else if (csp->content_type & CT_DEFLATE) + { + /* + * XXX: The debug level should be lowered + * before the next stable release. + */ + log_error (LOG_LEVEL_INFO, "Decompressing deflated iob: %d", *cur); + /* + * In theory (that is, according to RFC 1950), deflate-compressed + * data should begin with a two-byte zlib header and have an + * adler32 checksum at the end. It seems that in practice only + * the raw compressed data is sent. Note that this means that + * we are not RFC 1950-compliant here, but the advantage is that + * this actually works. :) + * + * We add a dummy null byte to tell zlib where the data ends, + * and later inform it not to expect a header. + * + * Fortunately, add_to_iob() has thoughtfully null-terminated + * the buffer; we can just increment the end pointer to include + * the dummy byte. + */ + csp->iob->eod++; + } + else + { + log_error (LOG_LEVEL_ERROR, + "Unable to determine compression format for decompression"); + return JB_ERR_COMPRESS; + } + + /* Set up the fields required by zlib. */ + zstr.next_in = (Bytef *)cur; + zstr.avail_in = (unsigned int)(csp->iob->eod - cur); + zstr.zalloc = Z_NULL; + zstr.zfree = Z_NULL; + zstr.opaque = Z_NULL; + + /* + * Passing -MAX_WBITS to inflateInit2 tells the library + * that there is no zlib header. + */ + if (inflateInit2 (&zstr, -MAX_WBITS) != Z_OK) + { + log_error (LOG_LEVEL_ERROR, "Error initializing decompression"); + return JB_ERR_COMPRESS; + } + + /* + * Next, we allocate new storage for the inflated data. + * We don't modify the existing iob yet, so in case there + * is error in decompression we can recover gracefully. + */ + buf = zalloc (bufsize); + if (NULL == buf) + { + log_error (LOG_LEVEL_ERROR, "Out of memory decompressing iob"); + return JB_ERR_MEMORY; + } + + assert(bufsize >= skip_size); + memcpy(buf, csp->iob->buf, skip_size); + zstr.avail_out = bufsize - skip_size; + zstr.next_out = (Bytef *)buf + skip_size; + + /* Try to decompress the whole stream in one shot. */ + while (Z_BUF_ERROR == (status = inflate(&zstr, Z_FINISH))) + { + /* We need to allocate more memory for the output buffer. */ + + char *tmpbuf; /* used for realloc'ing the buffer */ + size_t oldbufsize = bufsize; /* keep track of the old bufsize */ + + /* + * If zlib wants more data then there's a problem, because + * the complete compressed file should have been buffered. + */ + if (0 == zstr.avail_in) + { + log_error(LOG_LEVEL_ERROR, "Unexpected end of compressed iob"); + return JB_ERR_COMPRESS; + } + + /* + * If we tried the limit and still didn't have enough + * memory, just give up. + */ + if (bufsize == csp->config->buffer_limit) + { + log_error(LOG_LEVEL_ERROR, "Buffer limit reached while decompressing iob"); + return JB_ERR_MEMORY; + } + + /* Try doubling the buffer size each time. */ + bufsize *= 2; + + /* Don't exceed the buffer limit. */ + if (bufsize > csp->config->buffer_limit) + { + bufsize = csp->config->buffer_limit; + } + + /* Try to allocate the new buffer. */ + tmpbuf = realloc(buf, bufsize); + if (NULL == tmpbuf) + { + log_error(LOG_LEVEL_ERROR, "Out of memory decompressing iob"); + freez(buf); + return JB_ERR_MEMORY; + } + else + { + char *oldnext_out = (char *)zstr.next_out; + + /* + * Update the fields for inflate() to use the new + * buffer, which may be in a location different from + * the old one. + */ + zstr.avail_out += bufsize - oldbufsize; + zstr.next_out = (Bytef *)tmpbuf + bufsize - zstr.avail_out; + + /* + * Compare with an uglier method of calculating these values + * that doesn't require the extra oldbufsize variable. + */ + assert(zstr.avail_out == tmpbuf + bufsize - (char *)zstr.next_out); + assert((char *)zstr.next_out == tmpbuf + ((char *)oldnext_out - buf)); + assert(zstr.avail_out > 0); + + buf = tmpbuf; + } + } + + if (Z_STREAM_ERROR == inflateEnd(&zstr)) + { + log_error(LOG_LEVEL_ERROR, + "Inconsistent stream state after decompression: %s", zstr.msg); + /* + * XXX: Intentionally no return. + * + * According to zlib.h, Z_STREAM_ERROR is returned + * "if the stream state was inconsistent". + * + * I assume in this case inflate()'s status + * would also be something different than Z_STREAM_END + * so this check should be redundant, but lets see. + */ + } + + if (status != Z_STREAM_END) + { + /* We failed to decompress the stream. */ + log_error(LOG_LEVEL_ERROR, + "Error in decompressing to the buffer (iob): %s", zstr.msg); + return JB_ERR_COMPRESS; + } + + /* + * Finally, we can actually update the iob, since the + * decompression was successful. First, free the old + * buffer. + */ + freez(csp->iob->buf); + + /* Now, update the iob to use the new buffer. */ + csp->iob->buf = buf; + csp->iob->cur = csp->iob->buf + skip_size; + csp->iob->eod = (char *)zstr.next_out; + csp->iob->size = bufsize; + + /* + * Make sure the new uncompressed iob obeys some minimal + * consistency conditions. + */ + if ((csp->iob->buf < csp->iob->cur) + && (csp->iob->cur <= csp->iob->eod) + && (csp->iob->eod <= csp->iob->buf + csp->iob->size)) + { + const size_t new_size = (size_t)(csp->iob->eod - csp->iob->cur); + if (new_size > 0) + { + log_error(LOG_LEVEL_RE_FILTER, + "Decompression successful. Old size: %d, new size: %d.", + old_size, new_size); + } + else + { + /* zlib thinks this is OK, so lets do the same. */ + log_error(LOG_LEVEL_INFO, "Decompression didn't result in any content."); + } + } + else + { + /* It seems that zlib did something weird. */ + log_error(LOG_LEVEL_ERROR, + "Unexpected error decompressing the buffer (iob): %d==%d, %d>%d, %d<%d", + csp->iob->cur, csp->iob->buf + skip_size, csp->iob->eod, csp->iob->buf, + csp->iob->eod, csp->iob->buf + csp->iob->size); + return JB_ERR_COMPRESS; + } + + return JB_ERR_OK; + +} +#endif /* defined(FEATURE_ZLIB) */ + + /********************************************************************* * * Function : get_header @@ -849,6 +1373,40 @@ char *get_header_value(const struct list *header_list, const char *header_name) } + +/********************************************************************* + * + * Function : scan_headers + * + * Description : Scans headers, applies tags and updates action bits. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * + * Returns : JB_ERR_OK + * + *********************************************************************/ +jb_err scan_headers(struct client_state *csp) +{ + struct list_entry *h; /* Header */ + jb_err err = JB_ERR_OK; + + log_error(LOG_LEVEL_HEADER, "scanning headers for: %s", csp->http->url); + + for (h = csp->headers->first; (err == JB_ERR_OK) && (h != NULL) ; h = h->next) + { + /* Header crunch()ed in previous run? -> ignore */ + if (h->str == NULL) continue; + log_error(LOG_LEVEL_HEADER, "scan: %s", h->str); + err = header_tagger(csp, h->str); + } + + update_action_bits(csp); + + return err; +} + + /********************************************************************* * * Function : sed @@ -889,6 +1447,8 @@ char *sed(const struct parsers pats[], if (first_run) /* Parse and print */ { + scan_headers(csp); + for (v = pats; (err == JB_ERR_OK) && (v->str != NULL) ; v++) { for (p = csp->headers->first; (err == JB_ERR_OK) && (p != NULL) ; p = p->next) @@ -896,8 +1456,6 @@ char *sed(const struct parsers pats[], /* Header crunch()ed in previous run? -> ignore */ if (p->str == NULL) continue; - if (v == pats) log_error(LOG_LEVEL_HEADER, "scan: %s", p->str); - /* Does the current parser handle this header? */ if ((strncmpic(p->str, v->str, v->len) == 0) || (v->len == CHECK_EVERY_HEADER_REMAINING)) { @@ -919,7 +1477,7 @@ char *sed(const struct parsers pats[], */ if (strncmpic(csp->http->cmd, "HEAD", 4)) { - /*XXX: Code duplication*/ + /*XXX: Code duplication */ for (v = pats; (err == JB_ERR_OK) && (v->str != NULL) ; v++) { for (p = csp->headers->first; (err == JB_ERR_OK) && (p != NULL) ; p = p->next) @@ -946,66 +1504,179 @@ char *sed(const struct parsers pats[], } -/* here begins the family of parser functions that reformat header lines */ /********************************************************************* * - * Function : filter_server_header + * Function : header_tagger * - * Description : Checks if server header filtering is enabled. - * If it is, filter_header is called to do the work. + * Description : Executes all text substitutions from applying + * tag actions and saves the result as tag. + * + * XXX: Shares enough code with filter_header() and + * pcrs_filter_response() to warrant some helper functions. * * Parameters : * 1 : csp = Current client state (buffers, headers, etc...) - * 2 : header = On input, pointer to header to modify. - * On output, pointer to the modified header, or NULL - * to remove the header. This function frees the - * original string if necessary. + * 2 : header = Header that is used as tagger input * * Returns : JB_ERR_OK on success and always succeeds * *********************************************************************/ -jb_err filter_server_header(struct client_state *csp, char **header) +jb_err header_tagger(struct client_state *csp, char *header) { - if (csp->action->flags & ACTION_FILTER_SERVER_HEADERS) + int wanted_filter_type; + int multi_action_index; + int i; + pcrs_job *job; + + struct file_list *fl; + struct re_filterfile_spec *b; + struct list_entry *tag_name; + + int found_filters = 0; + const size_t header_length = strlen(header); + + if (csp->flags & CSP_FLAG_CLIENT_HEADER_PARSING_DONE) { - filter_header(csp, header); + wanted_filter_type = FT_SERVER_HEADER_TAGGER; + multi_action_index = ACTION_MULTI_SERVER_HEADER_TAGGER; + } + else + { + wanted_filter_type = FT_CLIENT_HEADER_TAGGER; + multi_action_index = ACTION_MULTI_CLIENT_HEADER_TAGGER; } - return(JB_ERR_OK); -} -/********************************************************************* - * - * Function : filter_client_header - * - * Description : Checks if client header filtering is enabled. - * If it is, filter_header is called to do the work. - * - * Parameters : - * 1 : csp = Current client state (buffers, headers, etc...) - * 2 : header = On input, pointer to header to modify. - * On output, pointer to the modified header, or NULL - * to remove the header. This function frees the - * original string if necessary. - * - * Returns : JB_ERR_OK on success and always succeeds - * - *********************************************************************/ -jb_err filter_client_header(struct client_state *csp, char **header) -{ - if (csp->action->flags & ACTION_FILTER_CLIENT_HEADERS) + /* Check if there are any filters */ + for (i = 0; i < MAX_AF_FILES; i++) { - filter_header(csp, header); + fl = csp->rlist[i]; + if (NULL != fl) + { + if (NULL != fl->f) + { + found_filters = 1; + break; + } + } } - return(JB_ERR_OK); + + if (0 == found_filters) + { + log_error(LOG_LEVEL_ERROR, "Unable to get current state of regex tagging."); + return(JB_ERR_OK); + } + + 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 filters, */ + for (b = fl->f; b; b = b->next) + { + if (b->type != wanted_filter_type) + { + /* skip the ones we don't care about, */ + continue; + } + /* leaving only taggers that could apply, of which we use the ones, */ + for (tag_name = csp->action->multi[multi_action_index]->first; + NULL != tag_name; tag_name = tag_name->next) + { + /* that do apply, and */ + if (strcmp(b->name, tag_name->str) == 0) + { + char *modified_tag = NULL; + char *tag = header; + size_t size = header_length; + + if (NULL == b->joblist) + { + log_error(LOG_LEVEL_RE_FILTER, + "Tagger %s has empty joblist. Nothing to do.", b->name); + continue; + } + + /* execute their pcrs_joblist on the header. */ + for (job = b->joblist; NULL != job; job = job->next) + { + const int hits = pcrs_execute(job, tag, size, &modified_tag, &size); + + if (0 < hits) + { + /* Success, continue with the modified version. */ + if (tag != header) + { + freez(tag); + } + tag = modified_tag; + } + else + { + /* Tagger doesn't match */ + if (0 > hits) + { + /* Regex failure, log it but continue anyway. */ + log_error(LOG_LEVEL_ERROR, + "Problems with tagger \'%s\' and header \'%s\': %s", + b->name, *header, pcrs_strerror(hits)); + } + freez(modified_tag); + } + } + + /* If this tagger matched */ + if (tag != header) + { + /* and there is something left to save, */ + if (0 < size) + { + /* enlist a unique version of it as tag. */ + if (JB_ERR_OK != enlist_unique(csp->tags, tag, 0)) + { + log_error(LOG_LEVEL_ERROR, + "Insufficient memory to add tag \'%s\', " + "based on tagger \'%s\' and header \'%s\'", + tag, b->name, *header); + } + else + { + log_error(LOG_LEVEL_HEADER, + "Adding tag \'%s\' created by header tagger \'%s\'", + tag, b->name); + } + } + freez(tag); + } + } /* if the tagger applies */ + } /* for every tagger that could apply */ + } /* for all filters */ + } /* for all filter files */ + + return JB_ERR_OK; } +/* here begins the family of parser functions that reformat header lines */ + /********************************************************************* * * Function : filter_header * * Description : Executes all text substitutions from all applying - * +filter actions on the header. + * +(server|client)-header-filter actions on the header. * Most of the code was copied from pcrs_filter_response, * including the rather short variable names * @@ -1033,6 +1704,19 @@ jb_err filter_header(struct client_state *csp, char **header) struct list_entry *filtername; int i, found_filters = 0; + int wanted_filter_type; + int multi_action_index; + + if (csp->flags & CSP_FLAG_CLIENT_HEADER_PARSING_DONE) + { + wanted_filter_type = FT_SERVER_HEADER_FILTER; + multi_action_index = ACTION_MULTI_SERVER_HEADER_FILTER; + } + else + { + wanted_filter_type = FT_CLIENT_HEADER_FILTER; + multi_action_index = ACTION_MULTI_CLIENT_HEADER_FILTER; + } /* * Need to check the set of re_filterfiles... @@ -1053,14 +1737,25 @@ jb_err filter_header(struct client_state *csp, char **header) if (0 == found_filters) { log_error(LOG_LEVEL_ERROR, "Unable to get current state of regexp filtering."); - return(JB_ERR_OK); + return(JB_ERR_OK); } for (i = 0; i < MAX_AF_FILES; i++) { fl = csp->rlist[i]; if ((NULL == fl) || (NULL == fl->f)) - break; + { + /* + * 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 its pcrs_joblist on the @@ -1068,7 +1763,13 @@ jb_err filter_header(struct client_state *csp, char **header) */ for (b = fl->f; b; b = b->next) { - for (filtername = csp->action->multi[ACTION_MULTI_FILTER]->first; + if (b->type != wanted_filter_type) + { + /* Skip other filter types */ + continue; + } + + for (filtername = csp->action->multi[multi_action_index]->first; filtername ; filtername = filtername->next) { if (strcmp(b->name, filtername->str) == 0) @@ -1081,7 +1782,7 @@ jb_err filter_header(struct client_state *csp, char **header) continue; } - log_error(LOG_LEVEL_RE_FILTER, "re_filtering %s (size %d) with filter %s...", + log_error(LOG_LEVEL_RE_FILTER, "filtering \'%s\' (size %d) with \'%s\' ...", *header, size, b->name); /* Apply all jobs from the joblist */ @@ -1112,14 +1813,19 @@ jb_err filter_header(struct client_state *csp, char **header) } } } - log_error(LOG_LEVEL_RE_FILTER, " ...produced %d hits (new size %d).", current_hits, size); + log_error(LOG_LEVEL_RE_FILTER, "... produced %d hits (new size %d).", current_hits, size); hits += current_hits; } } } } - if ( 0 == size ) + /* + * Additionally checking for hits is important because if + * the continue hack is triggered, server headers can + * arrive empty to separate multiple heads from each other. + */ + if ((0 == size) && hits) { log_error(LOG_LEVEL_HEADER, "Removing empty header %s", *header); freez(*header); @@ -1129,6 +1835,56 @@ jb_err filter_header(struct client_state *csp, char **header) } +/********************************************************************* + * + * Function : connection + * + * Description : Makes sure that the value of the Connection: header + * is "close" and signals connection_close_adder + * to do nothing. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 2 : header = On input, pointer to header to modify. + * On output, pointer to the modified header, or NULL + * to remove the header. This function frees the + * original string if necessary. + * + * Returns : JB_ERR_OK on success, or + * JB_ERR_MEMORY on out-of-memory error. + * + *********************************************************************/ +jb_err connection(struct client_state *csp, char **header) +{ + char *old_header = *header; + + /* Do we have a 'Connection: close' header? */ + if (strcmpic(*header, "Connection: close")) + { + /* No, create one */ + *header = strdup("Connection: close"); + if (header == NULL) + { + return JB_ERR_MEMORY; + } + log_error(LOG_LEVEL_HEADER, "Replaced: \'%s\' with \'%s\'", old_header, *header); + freez(old_header); + } + + /* Signal connection_close_adder() to return early. */ + if (csp->flags & CSP_FLAG_CLIENT_HEADER_PARSING_DONE) + { + csp->flags |= CSP_FLAG_SERVER_CONNECTION_CLOSE_SET; + } + else + { + csp->flags |= CSP_FLAG_CLIENT_CONNECTION_CLOSE_SET; + } + + return JB_ERR_OK; +} + + /********************************************************************* * * Function : crumble @@ -1173,13 +1929,13 @@ jb_err crumble(struct client_state *csp, char **header) jb_err crunch_server_header(struct client_state *csp, char **header) { const char *crunch_pattern; - /*Is there a header to crunch*/ + /* Do we feel like crunching? */ if ((csp->action->flags & ACTION_CRUNCH_SERVER_HEADER)) { crunch_pattern = csp->action->string[ACTION_STRING_SERVER_HEADER]; - /*Is the current header the lucky one?*/ + /* Is the current header the lucky one? */ if (strstr(*header, crunch_pattern)) { log_error(LOG_LEVEL_HEADER, "Crunching server header: %s (contains: %s)", *header, crunch_pattern); @@ -1189,12 +1945,14 @@ jb_err crunch_server_header(struct client_state *csp, char **header) return JB_ERR_OK; } + + /********************************************************************* * * Function : server_content_type * * Description : Set the content-type for filterable types (text/.*, - * javascript and image/gif) unless filtering has been + * .*xml.*, javascript and image/gif) unless filtering has been * forbidden (CT_TABOO) while parsing earlier headers. * NOTE: Since text/plain is commonly used by web servers * for files whose correct type is unknown, we don't @@ -1213,38 +1971,59 @@ jb_err crunch_server_header(struct client_state *csp, char **header) *********************************************************************/ jb_err server_content_type(struct client_state *csp, char **header) { - const char *newval; - - newval = csp->action->string[ACTION_STRING_CONTENT_TYPE]; + /* Remove header if it isn't the first Content-Type header */ + if(csp->content_type && (csp->content_type != CT_TABOO)) + { + /* + * Another, slightly slower, way to see if + * we already parsed another Content-Type header. + */ + assert(NULL != get_header_value(csp->headers, "Content-Type:")); + + log_error(LOG_LEVEL_ERROR, + "Multiple Content-Type headers. Removing and ignoring: \'%s\'", + *header); + freez(*header); + + return JB_ERR_OK; + } - if (csp->content_type != CT_TABOO) + if (!(csp->content_type & CT_TABOO)) { if ((strstr(*header, " text/") && !strstr(*header, "plain")) - || strstr(*header, "xml") - || strstr(*header, "application/x-javascript")) - csp->content_type = CT_TEXT; + || strstr(*header, "xml") + || strstr(*header, "application/x-javascript")) + { + csp->content_type |= CT_TEXT; + } else if (strstr(*header, " image/gif")) - csp->content_type = CT_GIF; + { + csp->content_type |= CT_GIF; + } else if (strstr(*header, " image/jpeg")) - csp->content_type = CT_JPEG; + { + csp->content_type |= CT_JPEG; + } else + { csp->content_type = 0; + } } /* * 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) + /* + * 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; + csp->content_type |= CT_TEXT; log_error(LOG_LEVEL_HEADER, "Text mode enabled by force. Take cover!"); } } @@ -1253,28 +2032,28 @@ jb_err server_content_type(struct client_state *csp, char **header) */ if (csp->action->flags & ACTION_CONTENT_TYPE_OVERWRITE) { - /* - * Make sure the user doesn't accidently - * change the content type of binary documents. - */ - if (csp->content_type == CT_TEXT) - { - freez(*header); - *header = strdup("Content-Type: "); - string_append(header, newval); - - if (header == NULL) - { - log_error(LOG_LEVEL_HEADER, "Insufficient memory. Conten-Type crunched without replacement!"); - return JB_ERR_MEMORY; - } - log_error(LOG_LEVEL_HEADER, "Modified: %s!", *header); - } - else - { - log_error(LOG_LEVEL_HEADER, "%s not replaced. It doesn't look like text. " - "Enable force-text-mode if you know what you're doing.", *header); - } + /* + * Make sure the user doesn't accidently + * change the content type of binary documents. + */ + if (csp->content_type & CT_TEXT) + { + freez(*header); + *header = strdup("Content-Type: "); + string_append(header, csp->action->string[ACTION_STRING_CONTENT_TYPE]); + + if (header == NULL) + { + log_error(LOG_LEVEL_HEADER, "Insufficient memory to replace Content-Type!"); + return JB_ERR_MEMORY; + } + log_error(LOG_LEVEL_HEADER, "Modified: %s!", *header); + } + else + { + log_error(LOG_LEVEL_HEADER, "%s not replaced. It doesn't look like text. " + "Enable force-text-mode if you know what you're doing.", *header); + } } return JB_ERR_OK; } @@ -1307,6 +2086,13 @@ jb_err server_transfer_coding(struct client_state *csp, char **header) */ if (strstr(*header, "gzip") || strstr(*header, "compress") || strstr(*header, "deflate")) { +#ifdef FEATURE_ZLIB + /* + * XXX: Added to test if we could use CT_GZIP and CT_DEFLATE here. + */ + log_error(LOG_LEVEL_INFO, "Marking content type for %s as CT_TABOO because of %s.", + csp->http->cmd, *header); +#endif /* def FEATURE_ZLIB */ csp->content_type = CT_TABOO; } @@ -1319,7 +2105,10 @@ jb_err server_transfer_coding(struct client_state *csp, char **header) /* * If the body was modified, it has been de-chunked first - * and the header must be removed. + * and the header must be removed. + * + * FIXME: If there is more than one transfer encoding, + * only the "chunked" part should be removed here. */ if (csp->flags & CSP_FLAG_MODIFIED) { @@ -1336,7 +2125,16 @@ jb_err server_transfer_coding(struct client_state *csp, char **header) * * Function : server_content_encoding * - * Description : Prohibit filtering (CT_TABOO) if content encoding compresses + * Description : This function is run twice for each request, + * unless FEATURE_ZLIB and filtering are disabled. + * + * The first run is used to check if the content + * is compressed, if FEATURE_ZLIB is disabled + * filtering is then disabled as well, if FEATURE_ZLIB + * is enabled the content is marked for decompression. + * + * The second run is used to remove the Content-Encoding + * header if the decompression was successful. * * Parameters : * 1 : csp = Current client state (buffers, headers, etc...) @@ -1351,13 +2149,60 @@ jb_err server_transfer_coding(struct client_state *csp, char **header) *********************************************************************/ jb_err server_content_encoding(struct client_state *csp, char **header) { - /* - * Turn off pcrs and gif filtering if body compressed - */ +#ifdef FEATURE_ZLIB + if ((csp->flags & CSP_FLAG_MODIFIED) + && (csp->content_type & (CT_GZIP | CT_DEFLATE))) + { + /* + * We successfully decompressed the content, + * and have to clean the header now, so the + * client no longer expects compressed data.. + * + * XXX: There is a difference between cleaning + * and removing it completely. + */ + log_error(LOG_LEVEL_HEADER, "Crunching: %s", *header); + freez(*header); + } + else if (strstr(*header, "gzip")) + { + /* Mark for gzip decompression */ + csp->content_type |= CT_GZIP; + } + else if (strstr(*header, "deflate")) + { + /* Mark for zlib decompression */ + csp->content_type |= CT_DEFLATE; + } + else if (strstr(*header, "compress")) + { + /* + * We can't decompress this; therefore we can't filter + * it either. + */ + csp->content_type |= CT_TABOO; + } +#else /* !defined(FEATURE_ZLIB) */ if (strstr(*header, "gzip") || strstr(*header, "compress") || strstr(*header, "deflate")) { - csp->content_type = CT_TABOO; + /* + * Body is compressed, turn off pcrs and gif filtering. + */ + csp->content_type |= CT_TABOO; + + /* + * Log a warning if the user expects the content to be filtered. + */ + if ((csp->rlist != NULL) && + (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER]))) + { + log_error(LOG_LEVEL_INFO, + "Compressed content detected, content filtering disabled. " + "Consider recompiling Privoxy with zlib support or " + "enable the prevent-compression action."); + } } +#endif /* defined(FEATURE_ZLIB) */ return JB_ERR_OK; @@ -1384,22 +2229,22 @@ jb_err server_content_encoding(struct client_state *csp, char **header) *********************************************************************/ jb_err server_content_length(struct client_state *csp, char **header) { - if (csp->content_length != 0) /* Content length could have been modified */ + const size_t max_header_length = 80; + + /* Regenerate header if the content was modified. */ + if (csp->flags & CSP_FLAG_MODIFIED) { - /* - * XXX: Shouldn't we check if csp->content_length - * is different than the original value? - */ freez(*header); - *header = (char *) zalloc(100); + *header = (char *) zalloc(max_header_length); if (*header == NULL) { return JB_ERR_MEMORY; } - sprintf(*header, "Content-Length: %d", (int) csp->content_length); - - log_error(LOG_LEVEL_HEADER, "Adjust Content-Length to %d", (int) csp->content_length); + snprintf(*header, max_header_length, "Content-Length: %d", + (int)csp->content_length); + log_error(LOG_LEVEL_HEADER, "Adjusted Content-Length to %d", + (int)csp->content_length); } return JB_ERR_OK; @@ -1572,32 +2417,34 @@ jb_err server_last_modified(struct client_state *csp, char **header) } else if (0 == strcmpic(newval, "randomize")) { + const char *header_time = *header + sizeof("Last-Modified:"); + log_error(LOG_LEVEL_HEADER, "Randomizing: %s", *header); now = time(NULL); #ifdef HAVE_GMTIME_R timeptr = gmtime_r(&now, &gmt); -#elif OSX_DARWIN +#elif FEATURE_PTHREAD pthread_mutex_lock(&gmtime_mutex); timeptr = gmtime(&now); pthread_mutex_unlock(&gmtime_mutex); #else timeptr = gmtime(&now); #endif - if ((timeptr = parse_header_time(*header, &last_modified)) == NULL) + if (JB_ERR_OK != parse_header_time(header_time, &last_modified)) { - log_error(LOG_LEVEL_HEADER, "Couldn't parse: %s (crunching!)", *header); + log_error(LOG_LEVEL_HEADER, "Couldn't parse: %s in %s (crunching!)", header_time, *header); freez(*header); } else { - rtime = difftime(now, last_modified); + rtime = (long int)difftime(now, last_modified); if (rtime) { rtime = pick_from_range(rtime); last_modified += rtime; #ifdef HAVE_GMTIME_R timeptr = gmtime_r(&last_modified, &gmt); -#elif OSX_DARWIN +#elif FEATURE_PTHREAD pthread_mutex_lock(&gmtime_mutex); timeptr = gmtime(&last_modified); pthread_mutex_unlock(&gmtime_mutex); @@ -1637,6 +2484,7 @@ jb_err server_last_modified(struct client_state *csp, char **header) return JB_ERR_OK; } + /********************************************************************* * * Function : client_accept_encoding @@ -1892,11 +2740,13 @@ jb_err client_accept_language(struct client_state *csp, char **header) if (*header == NULL) { - log_error(LOG_LEVEL_ERROR, " Insufficent memory. Accept-Language header crunched without replacement."); + log_error(LOG_LEVEL_ERROR, + "Insufficent memory. Accept-Language header crunched without replacement."); } else { - log_error(LOG_LEVEL_HEADER, "Accept-Language header crunched and replaced with: %s", *header); + log_error(LOG_LEVEL_HEADER, + "Accept-Language header crunched and replaced with: %s", *header); } } return (*header == NULL) ? JB_ERR_MEMORY : JB_ERR_OK; @@ -1922,13 +2772,13 @@ jb_err client_accept_language(struct client_state *csp, char **header) jb_err crunch_client_header(struct client_state *csp, char **header) { const char *crunch_pattern; - /*Is there a header to crunch*/ - + + /* Do we feel like crunching? */ if ((csp->action->flags & ACTION_CRUNCH_CLIENT_HEADER)) { crunch_pattern = csp->action->string[ACTION_STRING_CLIENT_HEADER]; - /*Is the current header the lucky one?*/ + /* Is the current header the lucky one? */ if (strstr(*header, crunch_pattern)) { log_error(LOG_LEVEL_HEADER, "Crunching client header: %s (contains: %s)", *header, crunch_pattern); @@ -2064,9 +2914,10 @@ jb_err client_from(struct client_state *csp, char **header) * * Function : client_send_cookie * - * Description : Handle the "cookie" header properly. Called from `sed'. - * If cookie is accepted, add it to the cookie_list, - * else we crunch it. Mmmmmmmmmmm ... cookie ...... + * Description : Crunches the "cookie" header if necessary. + * Called from `sed'. + * + * XXX: Stupid name, doesn't send squat. * * Parameters : * 1 : csp = Current client state (buffers, headers, etc...) @@ -2081,25 +2932,13 @@ jb_err client_from(struct client_state *csp, char **header) *********************************************************************/ jb_err client_send_cookie(struct client_state *csp, char **header) { - jb_err result = JB_ERR_OK; - - if ((csp->action->flags & ACTION_NO_COOKIE_READ) == 0) - { - /* strlen("cookie: ") == 8 */ - result = enlist(csp->cookie_list, *header + 8); - } - else + if (csp->action->flags & ACTION_NO_COOKIE_READ) { - log_error(LOG_LEVEL_HEADER, "Crunched outgoing cookie -- yum!"); + log_error(LOG_LEVEL_HEADER, "Crunched outgoing cookie: %s", *header); + freez(*header); } - /* - * Always remove the cookie here. The cookie header - * will be sent at the end of the header. - */ - freez(*header); - - return result; + return JB_ERR_OK; } @@ -2165,23 +3004,42 @@ jb_err client_x_forwarded(struct client_state *csp, char **header) *********************************************************************/ jb_err client_max_forwards(struct client_state *csp, char **header) { - unsigned int max_forwards; + int max_forwards; - if ((0 == strcmpic(csp->http->gpc, "trace")) - || (0 == strcmpic(csp->http->gpc, "options"))) + if ((0 == strcmpic(csp->http->gpc, "trace")) || + (0 == strcmpic(csp->http->gpc, "options"))) { - if (1 == sscanf(*header, "Max-Forwards: %u", &max_forwards)) + assert(*(*header+12) == ':'); + if (1 == sscanf(*header+12, ": %u", &max_forwards)) { - if (max_forwards-- >= 1) + if (max_forwards > 0) + { + snprintf(*header, strlen(*header)+1, "Max-Forwards: %u", --max_forwards); + log_error(LOG_LEVEL_HEADER, "Max-Forwards value for %s request reduced to %u.", + csp->http->gpc, max_forwards); + } + else if (max_forwards < 0) { - sprintf(*header, "Max-Forwards: %u", max_forwards); - log_error(LOG_LEVEL_HEADER, "Max forwards of %s request now %d", csp->http->gpc, max_forwards); + log_error(LOG_LEVEL_ERROR, "Crunching invalid header: %s", *header); + freez(*header); } else { - log_error(LOG_LEVEL_ERROR, "Non-intercepted %s request with Max-Forwards zero!", csp->http->gpc); + /* + * Not supposed to be reached. direct_response() which + * was already called earlier in chat() should have + * intercepted the request. + */ + log_error(LOG_LEVEL_ERROR, + "Non-intercepted %s request with Max-Forwards zero!", csp->http->gpc); + assert(max_forwards != 0); } } + else + { + log_error(LOG_LEVEL_ERROR, "Crunching invalid header: %s", *header); + freez(*header); + } } return JB_ERR_OK; @@ -2261,6 +3119,9 @@ jb_err client_host(struct client_state *csp, char **header) csp->http->hostport, csp->http->host, csp->http->port); } + /* Signal client_host_adder() to return right away */ + csp->flags |= CSP_FLAG_HOST_HEADER_IS_SET; + return JB_ERR_OK; } @@ -2318,9 +3179,11 @@ jb_err client_if_modified_since(struct client_state *csp, char **header) } else /* add random value */ { - if ((timeptr = parse_header_time(*header, &tm)) == NULL) + const char *header_time = *header + sizeof("If-Modified-Since:"); + + if (JB_ERR_OK != parse_header_time(header_time, &tm)) { - log_error(LOG_LEVEL_HEADER, "Couldn't parse: %s (crunching!)", *header); + log_error(LOG_LEVEL_HEADER, "Couldn't parse: %s in %s (crunching!)", header_time, *header); freez(*header); } else @@ -2346,7 +3209,7 @@ jb_err client_if_modified_since(struct client_state *csp, char **header) tm += rtime * (negative ? -1 : 1); #ifdef HAVE_GMTIME_R timeptr = gmtime_r(&tm, &gmt); -#elif OSX_DARWIN +#elif FEATURE_PTHREAD pthread_mutex_lock(&gmtime_mutex); timeptr = gmtime(&tm); pthread_mutex_unlock(&gmtime_mutex); @@ -2439,7 +3302,8 @@ jb_err client_x_filter(struct client_state *csp, char **header) { if (csp->action->flags & ACTION_FORCE_TEXT_MODE) { - log_error(LOG_LEVEL_HEADER, "force-text-mode overruled the client's request to fetch without filtering!"); + log_error(LOG_LEVEL_HEADER, + "force-text-mode overruled the client's request to fetch without filtering!"); } else { @@ -2476,8 +3340,16 @@ jb_err client_host_adder(struct client_state *csp) char *p; jb_err err; + if (csp->flags & CSP_FLAG_HOST_HEADER_IS_SET) + { + /* Header already set by the client, nothing to do. */ + return JB_ERR_OK; + } + if ( !csp->http->hostport || !*(csp->http->hostport)) { + /* XXX: When does this happen and why is it OK? */ + log_error(LOG_LEVEL_INFO, "Weirdness in client_host_adder detected and ignored."); return JB_ERR_OK; } @@ -2493,6 +3365,7 @@ jb_err client_host_adder(struct client_state *csp) p = csp->http->hostport; } + /* XXX: Just add it, we already made sure that it will be unique */ log_error(LOG_LEVEL_HEADER, "addh-unique: Host: %s", p); err = enlist_unique_header(csp->headers, "Host", p); return err; @@ -2506,6 +3379,8 @@ jb_err client_host_adder(struct client_state *csp) * * Description : Used in the add_client_headers list. Called from `sed'. * + * XXX: Remove csp->cookie_list which is no longer used. + * * Parameters : * 1 : csp = Current client state (buffers, headers, etc...) * @@ -2586,6 +3461,8 @@ jb_err client_cookie_adder(struct client_state *csp) *********************************************************************/ jb_err client_accept_encoding_adder(struct client_state *csp) { + assert(0); /* Not in use */ + if ( ((csp->action->flags & ACTION_NO_COMPRESSION) != 0) && (!strcmpic(csp->http->ver, "HTTP/1.1")) ) { @@ -2681,9 +3558,10 @@ jb_err client_x_forwarded_adder(struct client_state *csp) * * Function : connection_close_adder * - * Description : Adds a "Connection: close" header to csp->headers - * as a temporary fix for the needed but missing HTTP/1.1 - * support. Called from `sed'. + * Description : "Temporary" fix for the needed but missing HTTP/1.1 + * support. Adds a "Connection: close" header to csp->headers + * unless the header was already present. Called from `sed'. + * * FIXME: This whole function shouldn't be neccessary! * * Parameters : @@ -2695,7 +3573,27 @@ jb_err client_x_forwarded_adder(struct client_state *csp) *********************************************************************/ jb_err connection_close_adder(struct client_state *csp) { + const unsigned int flags = csp->flags; + + /* + * Return right away if + * + * - we're parsing server headers and the server header + * "Connection: close" is already set, or if + * + * - we're parsing client headers and the client header + * "Connection: close" is already set. + */ + if ((flags & CSP_FLAG_CLIENT_HEADER_PARSING_DONE + && flags & CSP_FLAG_SERVER_CONNECTION_CLOSE_SET) + ||(!(flags & CSP_FLAG_CLIENT_HEADER_PARSING_DONE) + && flags & CSP_FLAG_CLIENT_CONNECTION_CLOSE_SET)) + { + return JB_ERR_OK; + } + log_error(LOG_LEVEL_HEADER, "Adding: Connection: close"); + return enlist(csp->headers, "Connection: close"); } @@ -2744,8 +3642,13 @@ jb_err server_http(struct client_state *csp, char **header) * Function : server_set_cookie * * Description : Handle the server "cookie" header properly. - * Log cookie to the jar file. Then "crunch" it, - * or accept it. Called from `sed'. + * Log cookie to the jar file. Then "crunch", + * accept or rewrite it to a session cookie. + * Called from `sed'. + * + * TODO: Allow the user to specify a new expiration + * time to cause the cookie to expire even before the + * browser is closed. * * Parameters : * 1 : csp = Current client state (buffers, headers, etc...) @@ -2760,6 +3663,11 @@ jb_err server_http(struct client_state *csp, char **header) *********************************************************************/ jb_err server_set_cookie(struct client_state *csp, char **header) { + time_t now; + time_t cookie_time; + struct tm tm_now; + time(&now); + #ifdef FEATURE_COOKIE_JAR if (csp->config->jar) { @@ -2770,17 +3678,15 @@ jb_err server_set_cookie(struct client_state *csp, char **header) * the %z field in strftime() */ char tempbuf[ BUFFER_SIZE ]; - time_t now; - struct tm tm_now; - time (&now); + #ifdef HAVE_LOCALTIME_R tm_now = *localtime_r(&now, &tm_now); -#elif OSX_DARWIN +#elif FEATURE_PTHREAD pthread_mutex_lock(&localtime_mutex); - tm_now = *localtime (&now); + tm_now = *localtime (&now); pthread_mutex_unlock(&localtime_mutex); #else - tm_now = *localtime (&now); + tm_now = *localtime (&now); #endif strftime(tempbuf, BUFFER_SIZE-6, "%b %d %H:%M:%S ", &tm_now); @@ -2791,8 +3697,8 @@ jb_err server_set_cookie(struct client_state *csp, char **header) if ((csp->action->flags & ACTION_NO_COOKIE_SET) != 0) { - log_error(LOG_LEVEL_HEADER, "Crunched incoming cookie -- yum!"); - return crumble(csp, header); + log_error(LOG_LEVEL_HEADER, "Crunching incoming cookie: %s", *header); + freez(*header); } else if ((csp->action->flags & ACTION_NO_COOKIE_KEEP) != 0) { @@ -2833,22 +3739,93 @@ jb_err server_set_cookie(struct client_state *csp, char **header) next_tag = cur_tag + strlen(cur_tag); } - /* Is this the "Expires" tag? */ - if (strncmpic(cur_tag, "expires=", 8) == 0) + /* + * Check the expiration date to see + * if the cookie is still valid, if yes, + * rewrite it to a session cookie. + */ + if ((strncmpic(cur_tag, "expires=", 8) == 0) && *(cur_tag + 8)) { - /* Delete the tag by copying the rest of the string over it. - * (Note that we cannot just use "strcpy(cur_tag, next_tag)", - * since the behaviour of strcpy is undefined for overlapping - * strings.) - */ - memmove(cur_tag, next_tag, strlen(next_tag) + 1); + char *expiration_date = cur_tag + 8; /* Skip "[Ee]xpires=" */ - /* That changed the header, need to issue a log message */ - changed = 1; + /* Did we detect the date properly? */ + if (JB_ERR_OK != parse_header_time(expiration_date, &cookie_time)) + { + /* + * Nope, treat it as if it was still valid. + * + * XXX: Should we remove the whole cookie instead? + */ + log_error(LOG_LEVEL_ERROR, + "Can't parse \'%s\', send by %s. Unsupported time format?", cur_tag, csp->http->url); + memmove(cur_tag, next_tag, strlen(next_tag) + 1); + changed = 1; + } + else + { + /* + * Yes. Check if the cookie is still valid. + * + * If the cookie is already expired it's probably + * a delete cookie and even if it isn't, the browser + * will discard it anyway. + */ + + /* + * XXX: timegm() isn't available on some AmigaOS + * versions and our replacement doesn't work. + * + * Our options are to either: + * + * - disable session-cookies-only completely if timegm + * is missing, + * + * - to simply remove all expired tags, like it has + * been done until Privoxy 3.0.6 and to live with + * the consequence that it can cause login/logout + * problems on servers that don't validate their + * input properly, or + * + * - to replace it with mktime in which + * case there is a slight chance of valid cookies + * passing as already expired. + * + * This is the way it's currently done and it's not + * as bad as it sounds. If the missing GMT offset is + * enough to change the result of the expiration check + * the cookie will be only valid for a few hours + * anyway, which in many cases will be shorter + * than a browser session. + */ + if (cookie_time - now < 0) + { + log_error(LOG_LEVEL_HEADER, + "Cookie \'%s\' is already expired and can pass unmodified.", *header); + /* Just in case some clown sets more then one expiration date */ + cur_tag = next_tag; + } + else + { + /* + * Still valid, delete expiration date by copying + * the rest of the string over it. + * + * (Note that we cannot just use "strcpy(cur_tag, next_tag)", + * since the behaviour of strcpy is undefined for overlapping + * strings.) + */ + memmove(cur_tag, next_tag, strlen(next_tag) + 1); + + /* That changed the header, need to issue a log message */ + changed = 1; + + /* + * Note that the next tag has now been moved to *cur_tag, + * so we do not need to update the cur_tag pointer. + */ + } + } - /* Note that the next tag has now been moved to *cur_tag, - * so we do not need to update the cur_tag pointer. - */ } else { @@ -2859,7 +3836,9 @@ jb_err server_set_cookie(struct client_state *csp, char **header) if (changed) { - log_error(LOG_LEVEL_HEADER, "Changed cookie to a temporary one."); + assert(NULL != *header); + log_error(LOG_LEVEL_HEADER, "Cookie rewritten to a temporary one: %s", + *header); } } @@ -2884,9 +3863,12 @@ jb_err server_set_cookie(struct client_state *csp, char **header) *********************************************************************/ int strclean(const char *string, const char *substring) { - int hits = 0, len = strlen(substring); + int hits = 0; + size_t len; char *pos, *p; + len = strlen(substring); + while((pos = strstr(string, substring)) != NULL) { p = pos + len; @@ -2907,35 +3889,130 @@ int strclean(const char *string, const char *substring) * * Function : parse_header_time * - * Description : Transforms time inside a HTTP header into - * the usual time format. + * Description : Parses time formats used in HTTP header strings + * to get the numerical respresentation. * * Parameters : - * 1 : header = header to parse - * 2 : tm = storage for the resulting time in seconds + * 1 : header_time = HTTP header time as string. + * 2 : result = storage for header_time in seconds * - * Returns : Time struct containing the header time, or - * NULL in case of a parsing problem. + * Returns : JB_ERR_OK if the time format was recognized, or + * JB_ERR_PARSE otherwise. * *********************************************************************/ -struct tm *parse_header_time(char *header, time_t *tm) { - - char * timestring; +jb_err parse_header_time(const char *header_time, time_t *result) +{ struct tm gmt; - struct tm * timeptr; - /* Skipping header name */ - timestring = strstr(header, ": "); - if (strptime(timestring, ": %a, %d %b %Y %H:%M:%S", &gmt) == NULL) + /* + * Zero out gmt to prevent time zone offsets. + * + * While this is only necessary on some platforms + * (mingw32 for example), I don't know how to + * detect these automatically and doing it everywhere + * shouldn't hurt. + */ + memset(&gmt, 0, sizeof(gmt)); + + /* Tue, 02 Jun 2037 20:00:00 */ + if ((NULL == strptime(header_time, "%a, %d %b %Y %H:%M:%S", &gmt)) + /* Tue, 02-Jun-2037 20:00:00 */ + && (NULL == strptime(header_time, "%a, %d-%b-%Y %H:%M:%S", &gmt)) + /* Tue, 02-Jun-37 20:00:00 */ + && (NULL == strptime(header_time, "%a, %d-%b-%y %H:%M:%S", &gmt)) + /* Tuesday, 02-Jun-2037 20:00:00 */ + && (NULL == strptime(header_time, "%A, %d-%b-%Y %H:%M:%S", &gmt)) + /* Tuesday Jun 02 20:00:00 2037 */ + && (NULL == strptime(header_time, "%A %b %d %H:%M:%S %Y", &gmt))) { - timeptr = NULL; + return JB_ERR_PARSE; + } + + *result = timegm(&gmt); + + return JB_ERR_OK; + +} + +/********************************************************************* + * + * Function : get_destination_from_headers + * + * Description : Parse the "Host:" header to get the request's destination. + * Only needed if the client's request was forcefully + * redirected into Privoxy. + * + * Code mainly copied from client_host() which is currently + * run too late for this purpose. + * + * Parameters : + * 1 : headers = List of headers (one of them hopefully being + * the "Host:" header) + * 2 : http = storage for the result (host, port and hostport). + * + * Returns : JB_ERR_MEMORY in case of memory problems, + * JB_ERR_PARSE if the host header couldn't be found, + * JB_ERR_OK otherwise. + * + *********************************************************************/ +jb_err get_destination_from_headers(const struct list *headers, struct http_request *http) +{ + char *q; + char *p; + char *host; + + host = get_header_value(headers, "Host:"); + + if (NULL == host) + { + log_error(LOG_LEVEL_ERROR, "No \"Host:\" header found."); + return JB_ERR_PARSE; + } + + if (NULL == (p = strdup((host)))) + { + log_error(LOG_LEVEL_ERROR, "Out of memory while parsing \"Host:\" header"); + return JB_ERR_MEMORY; + } + chomp(p); + if (NULL == (q = strdup(p))) + { + freez(p); + log_error(LOG_LEVEL_ERROR, "Out of memory while parsing \"Host:\" header"); + return JB_ERR_MEMORY; + } + + freez(http->hostport); + http->hostport = p; + freez(http->host); + http->host = q; + q = strchr(http->host, ':'); + if (q != NULL) + { + /* Terminate hostname and evaluate port string */ + *q++ = '\0'; + http->port = atoi(q); } else { - *tm = timegm(&gmt); - timeptr=&gmt; + http->port = http->ssl ? 443 : 80; } - return(timeptr); + + /* Rebuild request URL */ + freez(http->url); + http->url = strdup(http->ssl ? "https://" : "http://"); + string_append(&http->url, http->hostport); + string_append(&http->url, http->path); + if (http->url == NULL) + { + return JB_ERR_MEMORY; + } + + log_error(LOG_LEVEL_HEADER, "Destination extracted from \"Host:\" header. New request URL: %s", + http->url); + + return JB_ERR_OK; + }