Workaround for webservers that lie about Content-Encoding: Tor now tries to... (69fa5be7) · Commits · jarl / tor

src/common/torgzip.c

+16 −0

Original line number	Diff line number	Diff line
		@@ -134,6 +134,7 @@ tor_gzip_compress(char *out, size_t out_len,
		return -1;
		}

		/* DOCDOC -- sets out to NULL on failure. /
		int
		tor_gzip_uncompress(char *out, size_t out_len,
		const char *in, size_t in_len,
		@@ -224,3 +225,18 @@ tor_gzip_uncompress(char *out, size_t out_len,
		return -1;
		}

		/** Try to tell whether the <b>in_len</b>-byte string in <b>in</b> is likely
		* to be compressed or not. If it is, return the likeliest compression method.
		* Otherwise, return 0.
		*/
		int detect_compression_method(const char *in, size_t in_len)
		{
		if (in_len > 2 && in[0] == 0x1f && in[1] == 0x8b) {
		return GZIP_METHOD;
		} else if (in_len > 2 && (in[0] & 0x0f) == 8 &&
		get_uint16(in) % 31 == 0) {
		return ZLIB_METHOD;
		} else {
		return 0;
		}
		}

src/common/torgzip.h

+5 −1

Original line number	Diff line number	Diff line
		@@ -11,7 +11,9 @@
		#define __TORGZIP_H
		#define TORGZIP_H_ID "$Id$"

		typedef enum { GZIP_METHOD=1, ZLIB_METHOD=2 } compress_method_t;
		typedef enum {
		GZIP_METHOD=1, ZLIB_METHOD=2, UNKNOWN_METHOD=3
		} compress_method_t;

		int
		tor_gzip_compress(char *out, size_t out_len,
		@@ -24,4 +26,6 @@ tor_gzip_uncompress(char *out, size_t out_len,

		int is_gzip_supported(void);

		int detect_compression_method(const char *in, size_t in_len);

		#endif

src/or/directory.c

+74 −10

Original line number	Diff line number	Diff line
		@@ -48,6 +48,7 @@ directory_send_command(connection_t conn, const char platform,
		int purpose, const char *resource,
		const char *payload, size_t payload_len);
		static int directory_handle_command(connection_t *conn);
		static int body_is_plausible(const char *body, size_t body_len);

		/******* START VARIABLES ********/

		@@ -483,7 +484,10 @@ parse_http_url(char headers, char *url)
		* "HTTP/1.\%d \%d\%s\r\n...".
		* If it's well-formed, assign *<b>code</b>, point and return 0.
		* If <b>date</b> is provided, set *date to the Date header in the
		* http headers, or 0 if no such header is found.
		* http headers, or 0 if no such header is found. If <b>compression</b>
		* is provided, set *<b>compression</b> to the compression method given
		* in the Content-Encoding header, or 0 if no such header is found, or -1
		* if the value of the header is not recognized.
		* Otherwise, return -1.
		*/
		static int
		@@ -534,8 +538,8 @@ parse_http_response(const char headers, int code, time_t *date,
		} else if (!strcmp(enc, "gzip") \|\| !strcmp(enc, "x-gzip")) {
		*compression = GZIP_METHOD;
		} else {
		log_fn(LOG_WARN, "Unrecognized content encoding: '%s'", enc);
		*compression = 0;
		log_fn(LOG_INFO, "Unrecognized content encoding: '%s'", enc);
		*compression = -1;
		}
		}
		SMARTLIST_FOREACH(parsed_headers, char *, s, tor_free(s));
		@@ -544,6 +548,27 @@ parse_http_response(const char headers, int code, time_t *date,
		return 0;
		}

		/** Return true iff <b>body</b> doesn't start with a plausible router or
		* running-list or directory opening. This is a sign of possible compression.
		**/
		static int
		body_is_plausible(const char *body, size_t len)
		{
		int i;
		if (len < 32)
		return 0;
		if (!strcmpstart(body,"router") \|\|
		!strcmpstart(body,"signed-directory") \|\|
		!strcmpstart(body,"network-status") \|\|
		!strcmpstart(body,"running-routers"))
		return 1;
		for (i=0;i<32;++i) {
		if (!isprint(body[i]) && !isspace(body[i]))
		return 0;
		}
		return 1;
		}

		/** We are a client, and we've finished reading the server's
		* response. Parse and it and act appropriately.
		*
		@@ -560,6 +585,7 @@ connection_dir_client_reached_eof(connection_t *conn)
		time_t now, date_header=0;
		int delta;
		int compression;
		int plausible;

		switch (fetch_from_buf_http(conn->inbuf,
		&headers, MAX_HEADERS_SIZE,
		@@ -592,18 +618,56 @@ connection_dir_client_reached_eof(connection_t *conn)
		}
		}

		if (compression != 0) {
		char *new_body;
		size_t new_len;
		if (tor_gzip_uncompress(&new_body, &new_len, body, body_len, compression)) {
		plausible = body_is_plausible(body, body_len);
		if (compression \|\| !plausible) {
		char *new_body = NULL;
		size_t new_len = 0;
		int guessed = detect_compression_method(body, body_len);
		if (compression <= 0 \|\| guessed != compression) {
		/* Tell the user if we don't believe what we're told about compression.*/
		const char description1, description2;
		if (compression == ZLIB_METHOD)
		description1 = "as deflated";
		else if (compression = GZIP_METHOD)
		description1 = "as gzipped";
		else if (compression == 0)
		description1 = "as uncompressed";
		else
		description1 = "with an unknown Content-Encoding";
		if (guessed == ZLIB_METHOD)
		description2 = "deflated";
		else if (guessed == GZIP_METHOD)
		description2 = "gzipped";
		else if (!plausible)
		description2 = "confusing binary junk";
		else
		description2 = "uncompressed";

		log_fn(LOG_INFO, "HTTP body from server '%s' was labeled %s,"
		"but it seems to be %s.%s",
		conn->address, description1, description2,
		(compression>0 && guessed>0)?" Trying both.":"");
		}
		/* Try declared compression first if we can. */
		if (compression > 0)
		tor_gzip_uncompress(&new_body, &new_len, body, body_len, compression);
		/* Okay, if that didn't work, and we think that it was compressed
		* differently, try that. */
		if (!new_body && guessed > 0 && compression != guessed)
		tor_gzip_uncompress(&new_body, &new_len, body, body_len, guessed);
		/* If we're pretty sure that we have a compressed directory, and
		* we didn't manage to uncompress it, then warn and bail. */
		if (!plausible && !new_body) {
		log_fn(LOG_WARN, "Unable to decompress HTTP body (server '%s').", conn->address);
		tor_free(body); tor_free(headers);
		return -1;
		}
		if (new_body) {
		tor_free(body);
		body = new_body;
		body_len = new_len;
		}
		}

		if (conn->purpose == DIR_PURPOSE_FETCH_DIR) {
		/* fetch/process the directory to learn about new routers. */