Hopefully the comment I added in the last commit in combination with
the simple code change below will be understandable and perhaps even
correct.
It works for me with both mod_cgid and another module which gets
apr_proc_create() to do more of the pipe setup.
Index: srclib/apr/threadproc/unix/proc.c
===================================================================
RCS file: /home/cvspublic/apr/threadproc/unix/proc.c,v
retrieving revision 1.52
diff -u -r1.52 proc.c
--- srclib/apr/threadproc/unix/proc.c 2001/10/27 00:48:19 1.52
+++ srclib/apr/threadproc/unix/proc.c 2001/10/27 02:32:02
@@ -290,8 +290,7 @@
int status;
/* child process */
- /* XXX major SNAFU
- *
+ /*
* If we do exec cleanup before the dup2() calls to set up pipes
* on 0-2, we accidentally close the pipes used by programs like
* mod_cgid.
@@ -299,9 +298,27 @@
* If we do exec cleanup after the dup2() calls, cleanup can
accidentally
* close our pipes which replaced any files which previously had
* descriptors 0-2.
+ *
+ * The solution is to kill the cleanup for the pipes, then do
+ * exec cleanup, then do the dup2() calls.
*/
if (attr->child_in) {
+ apr_pool_cleanup_kill(apr_file_pool_get(attr->child_in),
+ attr->child_in, apr_unix_file_cleanup);
+ }
+ if (attr->child_out) {
+ apr_pool_cleanup_kill(apr_file_pool_get(attr->child_out),
+ attr->child_out, apr_unix_file_cleanup);
+ }
+ if (attr->child_err) {
+ apr_pool_cleanup_kill(apr_file_pool_get(attr->child_err),
+ attr->child_err, apr_unix_file_cleanup);
+ }
+
+ apr_pool_cleanup_for_exec();
+
+ if (attr->child_in) {
apr_file_close(attr->parent_in);
dup2(attr->child_in->filedes, STDIN_FILENO);
apr_file_close(attr->child_in);
@@ -324,8 +341,6 @@
exit(-1); /* We have big problems, the child should exit. */
}
}
-
- apr_pool_cleanup_for_exec();
if ((status = limit_proc(attr)) != APR_SUCCESS) {
return status;
--
Jeff Trawick | [EMAIL PROTECTED] | PGP public key at web site:
http://www.geocities.com/SiliconValley/Park/9289/
Born in Roswell... married an alien...
