http-backend: Protect GIT_PROJECT_ROOT from /../ requests

Eons ago HPA taught git-daemon how to protect itself from /../
attacks, which Junio brought back into service in d79374c
("daemon.c and path.enter_repo(): revamp path validation").

I did not carry this into git-http-backend as originally we relied
only upon PATH_TRANSLATED, and assumed the HTTP server had done
its access control checks to validate the resolved path was within
a directory permitting access from the remote client.  This would
usually be sufficient to protect a server from requests for its
/etc/passwd file by http://host/smart/../etc/passwd sorts of URLs.

However in 917adc0 Mark Lodato added GIT_PROJECT_ROOT as an
additional method of configuring the CGI.  When this environment
variable is used the web server does not generate the final access
path and therefore may blindly pass through "/../etc/passwd"
in PATH_INFO under the assumption that "/../" might have special
meaning to the invoked CGI.

Instead of permitting these sorts of malformed path requests, we
now reject them back at the client, with an error message for the
server log.  This matches git-daemon behavior.

Signed-off-by: Shawn O. Pearce <[email protected]>
Signed-off-by: Junio C Hamano <[email protected]>

cache.h

@@ -656,6 +656,7 @@ const char *make_relative_path(const char *abs, const char *base);
 int normalize_path_copy(char *dst, const char *src);
 int longest_ancestor_length(const char *path, const char *prefix_list);
 char *strip_path_suffix(const char *path, const char *suffix);
+int daemon_avoid_alias(const char *path);
 
 /* Read and unpack a sha1 file into memory, write memory to a sha1 file */
 extern int sha1_object_info(const unsigned char *, unsigned long *);

daemon.c

@@ -101,53 +101,6 @@ static void NORETURN daemon_die(const char *err, va_list params)
 	exit(1);
 }
 
-static int avoid_alias(char *p)
-{
-	int sl, ndot;
-
-	/*
-	 * This resurrects the belts and suspenders paranoia check by HPA
-	 * done in <[email protected]> thread, now enter_repo()
-	 * does not do getcwd() based path canonicalizations.
-	 *
-	 * sl becomes true immediately after seeing '/' and continues to
-	 * be true as long as dots continue after that without intervening
-	 * non-dot character.
-	 */
-	if (!p || (*p != '/' && *p != '~'))
-		return -1;
-	sl = 1; ndot = 0;
-	p++;
-
-	while (1) {
-		char ch = *p++;
-		if (sl) {
-			if (ch == '.')
-				ndot++;
-			else if (ch == '/') {
-				if (ndot < 3)
-					/* reject //, /./ and /../ */
-					return -1;
-				ndot = 0;
-			}
-			else if (ch == 0) {
-				if (0 < ndot && ndot < 3)
-					/* reject /.$ and /..$ */
-					return -1;
-				return 0;
-			}
-			else
-				sl = ndot = 0;
-		}
-		else if (ch == 0)
-			return 0;
-		else if (ch == '/') {
-			sl = 1;
-			ndot = 0;
-		}
-	}
-}
-
 static char *path_ok(char *directory)
 {
 	static char rpath[PATH_MAX];
@@ -157,7 +110,7 @@ static char *path_ok(char *directory)
 
 	dir = directory;
 
-	if (avoid_alias(dir)) {
+	if (daemon_avoid_alias(dir)) {
 		logerror("'%s': aliased", dir);
 		return NULL;
 	}

http-backend.c

@@ -559,7 +559,13 @@ static char* getdir(void)
 	if (root && *root) {
 		if (!pathinfo || !*pathinfo)
 			die("GIT_PROJECT_ROOT is set but PATH_INFO is not");
+		if (daemon_avoid_alias(pathinfo))
+			die("'%s': aliased", pathinfo);
 		strbuf_addstr(&buf, root);
+		if (buf.buf[buf.len - 1] != '/')
+			strbuf_addch(&buf, '/');
+		if (pathinfo[0] == '/')
+			pathinfo++;
 		strbuf_addstr(&buf, pathinfo);
 		return strbuf_detach(&buf, NULL);
 	} else if (path && *path) {

path.c

@@ -564,3 +564,50 @@ char *strip_path_suffix(const char *path, const char *suffix)
 		return NULL;
 	return xstrndup(path, chomp_trailing_dir_sep(path, path_len));
 }
+
+int daemon_avoid_alias(const char *p)
+{
+	int sl, ndot;
+
+	/*
+	 * This resurrects the belts and suspenders paranoia check by HPA
+	 * done in <[email protected]> thread, now enter_repo()
+	 * does not do getcwd() based path canonicalizations.
+	 *
+	 * sl becomes true immediately after seeing '/' and continues to
+	 * be true as long as dots continue after that without intervening
+	 * non-dot character.
+	 */
+	if (!p || (*p != '/' && *p != '~'))
+		return -1;
+	sl = 1; ndot = 0;
+	p++;
+
+	while (1) {
+		char ch = *p++;
+		if (sl) {
+			if (ch == '.')
+				ndot++;
+			else if (ch == '/') {
+				if (ndot < 3)
+					/* reject //, /./ and /../ */
+					return -1;
+				ndot = 0;
+			}
+			else if (ch == 0) {
+				if (0 < ndot && ndot < 3)
+					/* reject /.$ and /..$ */
+					return -1;
+				return 0;
+			}
+			else
+				sl = ndot = 0;
+		}
+		else if (ch == 0)
+			return 0;
+		else if (ch == '/') {
+			sl = 1;
+			ndot = 0;
+		}
+	}
+}

t/t5560-http-backend.sh

@@ -146,6 +146,37 @@ test_expect_success 'http.receivepack false' '
 	POST git-receive-pack 0000 "403 Forbidden"
 '
 
+run_backend() {
+	REQUEST_METHOD=GET \
+	GIT_PROJECT_ROOT="$HTTPD_DOCUMENT_ROOT_PATH" \
+	PATH_INFO="$2" \
+	git http-backend >act.out 2>act.err
+}
+
+path_info() {
+	if test $1 = 0; then
+		run_backend "$2"
+	else
+		test_must_fail run_backend "$2" &&
+		echo "fatal: '$2': aliased" >exp.err &&
+		test_cmp exp.err act.err
+	fi
+}
+
+test_expect_success 'http-backend blocks bad PATH_INFO' '
+	config http.getanyfile true &&
+
+	run_backend 0 /repo.git/HEAD &&
+
+	run_backend 1 /repo.git/../HEAD &&
+	run_backend 1 /../etc/passwd &&
+	run_backend 1 ../etc/passwd &&
+	run_backend 1 /etc//passwd &&
+	run_backend 1 /etc/./passwd &&
+	run_backend 1 /etc/.../passwd &&
+	run_backend 1 //domain/data.txt
+'
+
 cat >exp <<EOF
 
 ###  refs/heads/master