httpd: fix for POSTDATA handling bugs:

erroneous close(0)
full_read -> safe_read (with explanation)

1 files changed, 24 insertions, 24 deletions

diff --git a/networking/httpd.c b/networking/httpd.c
index fb363c56f..6f4ca05a7 100644
--- a/networking/httpd.c
+++ b/networking/httpd.c
@@ -950,7 +950,6 @@ static int getLine(void)
  *      (int bodyLen)  . . . . . . . . Length of the post body.
  *      (const char *cookie) . . . . . For set HTTP_COOKIE.
  *      (const char *content_type) . . For set CONTENT_TYPE.
-
  *
  * $Return: (char *)  . . . . A pointer to the decoded string (same as input).
  *
@@ -984,31 +983,32 @@ static int sendCgi(const char *url,
 	if (!pid) {
 		/* child process */
 		char *script;
-		char *purl = xstrdup(url);
+		char *purl;
 		char realpath_buff[MAXPATHLEN];
 
-		if (purl == NULL)
-			_exit(242);
-
-		inFd = toCgi[0];
-		outFd = fromCgi[1];
+		if (config->accepted_socket > 1)
+			close(config->accepted_socket);
+		if (config->server_socket > 1)
+			close(config->server_socket);
 
-		dup2(inFd, 0);  // replace stdin with the pipe
-		dup2(outFd, 1);  // replace stdout with the pipe
+		dup2(toCgi[0], 0);  // replace stdin with the pipe
+		dup2(fromCgi[1], 1);  // replace stdout with the pipe
+		/* Huh? User seeing stderr can be a security problem...
+		 * and if cgi really wants that, it can always dup2(1,2)...
 		if (!DEBUG)
-			dup2(outFd, 2);  // replace stderr with the pipe
-
+			dup2(fromCgi[1], 2);  // replace stderr with the pipe
+		*/
+		/* I think we cannot inadvertently close 0, 1 here... */
 		close(toCgi[0]);
 		close(toCgi[1]);
 		close(fromCgi[0]);
 		close(fromCgi[1]);
 
-		close(config->accepted_socket);
-		close(config->server_socket);
-
 		/*
 		 * Find PATH_INFO.
 		 */
+		xfunc_error_retval = 242;
+		purl = xstrdup(url);
 		script = purl;
 		while ((script = strchr(script + 1, '/')) != NULL) {
 			/* have script.cgi/PATH_INFO or dirs/script.cgi[/PATH_INFO] */
@@ -1210,33 +1210,33 @@ static int sendCgi(const char *url,
 #if PIPESIZE >= MAX_MEMORY_BUFF
 # error "PIPESIZE >= MAX_MEMORY_BUFF"
 #endif
-			/* NB: was safe_read. If it *has to be* safe_read, */
-			/* please explain why in this comment... */
-			count = full_read(inFd, rbuf, PIPESIZE);
+			/* Must use safe_read, not full_read, because
+			 * cgi may output a few first lines and then wait
+			 * for POSTDATA without closing stdout.
+			 * With full_read we may wait here forever. */
+			count = safe_read(inFd, rbuf, PIPESIZE);
 			if (count == 0)
 				break;  /* closed */
 			if (count < 0)
 				continue; /* huh, error, why continue?? */
 
 			if (firstLine) {
-				/* full_read (above) avoids
-				 * "chopped up into small chunks" syndrome here */
+				static const char HTTP_200[] = "HTTP/1.0 200 OK\r\n\r\n";
 				rbuf[count] = '\0';
 				/* check to see if the user script added headers */
-#define HTTP_200 "HTTP/1.0 200 OK\r\n\r\n"
+				/* (NB: buggy wrt cgi splitting "HTTP OK") */
 				if (memcmp(rbuf, HTTP_200, 4) != 0) {
 					/* there is no "HTTP", do it ourself */
 					full_write(s, HTTP_200, sizeof(HTTP_200)-1);
 				}
-#undef HTTP_200
 				/* Example of valid GCI without "Content-type:"
 				 * echo -en "HTTP/1.0 302 Found\r\n"
 				 * echo -en "Location: http://www.busybox.net\r\n"
 				 * echo -en "\r\n"
+				if (!strstr(rbuf, "ontent-")) {
+					full_write(s, "Content-type: text/plain\r\n\r\n", 28);
+				}
 				 */
-				//if (!strstr(rbuf, "ontent-")) {
-				//	full_write(s, "Content-type: text/plain\r\n\r\n", 28);
-				//}
 				firstLine = 0;
 			}
 			if (full_write(s, rbuf, count) != count)