-const char parsers_rcs[] = "$Id: parsers.c,v 1.81 2006/12/31 22:21:33 fabiankeil Exp $";
+const char parsers_rcs[] = "$Id: parsers.c,v 1.87 2007/01/31 16:21:38 fabiankeil Exp $";
/*********************************************************************
*
* File : $Source: /cvsroot/ijbswa/current/parsers.c,v $
* `client_if_none_match', `get_destination_from_headers',
* `parse_header_time' and `server_set_cookie'.
*
- * Copyright : Written by and Copyright (C) 2001-2006 the SourceForge
+ * Copyright : Written by and Copyright (C) 2001-2007 the SourceForge
* Privoxy team. http://www.privoxy.org/
*
* Based on the Internet Junkbuster originally written
*
* Revisions :
* $Log: parsers.c,v $
+ * 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.
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 skip_size; /* Number of bytes at the beginning of the iob
that we should NOT decompress. */
bufsize = csp->iob->size;
skip_size = (size_t)(csp->iob->cur - csp->iob->buf);
+ cur = csp->iob->cur;
+
if (bufsize < 10)
{
/*
* Strip off the gzip header. Please see RFC 1952 for more
* explanation of the appropriate fields.
*/
- if ((*csp->iob->cur++ != (char)0x1f)
- || (*csp->iob->cur++ != (char)0x8b)
- || (*csp->iob->cur++ != Z_DEFLATED))
+ 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 = *csp->iob->cur++;
+ 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 when decompressing");
+ log_error (LOG_LEVEL_ERROR, "Invalid gzip header flags when decompressing");
return JB_ERR_COMPRESS;
}
- csp->iob->cur += 6;
+ cur += 6;
/* Skip extra fields if necessary. */
if (flags & 0x04)
* Skip a given number of bytes, specified
* as a 16-bit little-endian value.
*/
- csp->iob->cur += *csp->iob->cur++ + (*csp->iob->cur++ << 8);
+ /*
+ * 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 follows. */
- while (*csp->iob->cur++);
+ while (*cur++);
}
/* Skip the comment if necessary. */
if (flags & 0x10)
{
- while (*csp->iob->cur++);
+ while (*cur++);
}
/* Skip the CRC if necessary. */
if (flags & 0x02)
{
- csp->iob->cur += 2;
+ cur += 2;
}
}
}
else if (csp->content_type & CT_DEFLATE)
{
- log_error (LOG_LEVEL_INFO, "Decompressing deflated iob: %d", *csp->iob->cur);
+ /*
+ * 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
}
/* Set up the fields required by zlib. */
- zstr.next_in = (Bytef *)csp->iob->cur;
- zstr.avail_in = (unsigned long)(csp->iob->eod - csp->iob->cur);
+ 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;
}
}
- inflateEnd(&zstr);
+ 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. */
if (first_run) /* Parse and print */
{
+ log_error(LOG_LEVEL_HEADER, "scanning headers for: %s", csp->http->url);
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)
}
}
- 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);
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
*********************************************************************/
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);
- assert(!csp->content_type || (csp->content_type == CT_TABOO));
+ return JB_ERR_OK;
+ }
if (!(csp->content_type & CT_TABOO))
{
{
freez(*header);
*header = strdup("Content-Type: ");
- string_append(header, newval);
+ string_append(header, csp->action->string[ACTION_STRING_CONTENT_TYPE]);
if (header == NULL)
{
- log_error(LOG_LEVEL_HEADER,
- "Insufficient memory. Content-Type crunched without replacement!");
+ log_error(LOG_LEVEL_HEADER, "Insufficient memory to replace Content-Type!");
return JB_ERR_MEMORY;
}
log_error(LOG_LEVEL_HEADER, "Modified: %s!", *header);
jb_err server_content_encoding(struct client_state *csp, char **header)
{
#ifdef FEATURE_ZLIB
- /* XXX: Why would we modify the content if it was taboo? */
- if ((csp->flags & CSP_FLAG_MODIFIED) && !(csp->content_type & CT_TABOO))
+ if ((csp->flags & CSP_FLAG_MODIFIED)
+ && (csp->content_type & (CT_GZIP | CT_DEFLATE)))
{
/*
* We successfully decompressed the content,
*********************************************************************/
jb_err server_content_length(struct client_state *csp, char **header)
{
+ const size_t max_header_length = 80;
if (csp->content_length != 0) /* Content length could have been modified */
{
/*
* 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;
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;
*********************************************************************/
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)
{
- sprintf(*header, "Max-Forwards: %u", max_forwards);
- log_error(LOG_LEVEL_HEADER, "Max forwards of %s request now %d", csp->http->gpc, max_forwards);
+ 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)
+ {
+ 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;
{
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
{
* 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...)
*********************************************************************/
jb_err server_set_cookie(struct client_state *csp, char **header)
{
+ time_t now;
+ time_t cookie_time;
+ struct tm tm_now;
+ struct tm tm_cookie;
+ time(&now);
+
#ifdef FEATURE_COOKIE_JAR
if (csp->config->jar)
{
* 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 FEATURE_PTHREAD
next_tag = cur_tag + strlen(cur_tag);
}
- /* Is this the "Expires" tag? */
+ /*
+ * 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)
{
- /* 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.)
+ char *match;
+ /*
+ * Try the valid time formats we know about.
+ *
+ * XXX: Maybe the log messages should be removed
+ * for the next stable release. They just exist to
+ * see which time format gets the most hits and
+ * should be checked for first.
*/
- memmove(cur_tag, next_tag, strlen(next_tag) + 1);
+ if (NULL != (match = strptime(cur_tag, "expires=%a, %e-%b-%y %H:%M:%S ", &tm_cookie)))
+ {
+ log_error(LOG_LEVEL_HEADER,
+ "cookie \'%s\' send by %s appears to be using time format 1.",
+ csp->http->url, *header);
+ }
+ else if (NULL != (match = strptime(cur_tag, "expires=%A, %e-%b-%Y %H:%M:%S ", &tm_cookie)))
+ {
+ log_error(LOG_LEVEL_HEADER,
+ "cookie \'%s\' send by %s appears to be using time format 2.",
+ csp->http->url, *header);
- /* That changed the header, need to issue a log message */
- changed = 1;
+ }
+ else if (NULL != (match = strptime(cur_tag, "expires=%a, %e-%b-%Y %H:%M:%S ", &tm_cookie)))
+ {
+ log_error(LOG_LEVEL_HEADER,
+ "cookie \'%s\' send by %s appears to be using time format 3.",
+ csp->http->url, *header);
+ }
+
+ /* Did any of them match? */
+ if (NULL == match)
+ {
+ /*
+ * 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. Unsupported time format?", cur_tag);
+ 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.
+ */
+ cookie_time = timegm(&tm_cookie);
+ 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
+ {
+ log_error(LOG_LEVEL_HEADER,
+ "Cookie \'%s\' is still valid and has to be rewritten.", *header);
+
+ /*
+ * 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);
+
+ /* 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
{
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);
}
}