[PATCH] fix httpd lockup in cgi POSTs

[Matthias Reichl]

While debugging some lockup problems with the openwrt webif^2 firmware
upload page I discovered a bug in httpd.c:

In line 1216 it must really be safe_read(), not full_read().
Otherwise httpd may lock up in the case where the cgi-script
has sent some data before fully receiving all POSTed data.

httpd.c multiplexes fine using select() and only calls safe_read()
if data is available, whereas full_read() loops over safe_read()
without checking if data is available. In this case read() will
block...

so long,

Hias

--- busybox-1.4.1.orig/networking/httpd.c	2007-01-24 22:34:34.000000000 +0100
+++ busybox-1.4.1/networking/httpd.c	2007-02-11 20:37:01.000000000 +0100
@@ -1211,9 +1211,10 @@
 #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);
+			/* reverted back to safe_read, otherwise httpd may */
+			/* block if the cgi-script outputs page date before */
+			/* it has fully received all (POST) data */
+			count = safe_read(inFd, rbuf, PIPESIZE);
 			if (count == 0)
 				break;  /* closed */
 			if (count < 0)