This is the mail archive of the cygwin@sources.redhat.com mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: <defunct> processes under Win9x...


Hi, Ken

I had a similar experience with <defunct> processes in
Apache port for cygwin but, in my case, it was with
the version 1.3.9 distribuited in the Web site of
cygwin porting project

http://www.student.uni-koeln.de/cygwin/Distribution/Source/apache/

I have gcc 2.95.2 and cygwin 1.1.4 (2000-08-03) and my
system of work is Windows 98; the visualization of Web
pages is good, but when I tried run the CGI scripts,
the child processes gone to <defunct> state as well as
in your same situation. I had beed depuring the first
chield proccess whit the purpose of find the
instruction or the function in which the error ocurrs,
and I believe that I already discovered the source of
the problem.

In appearance, Apache fail whit CGI scripts in Windows
98 when it come to function ap_select (in file
http_protocol.c).  At this point, the program stay
stopped until TIMEOUT variable expires (this variable
is defined in httpd.conf, with default to 300 sec),
but this last function isn't the real cause of the
problem. The true error ocurrs in the primitive
function read (located at buff.c, inside of function
ap_read), because this primitive don't wait, but that
it returns inmediatly whitout reading any data; by
this reason, the read function returns a value of -1,
and it don't breaks the later processes.

The cause of why the read function don't wait for any
input, is because the O_NONBLOCK flag is set for the
file fb->fd_in. Setting the next patch for avoid the
non-blocking of this file, the problem is resolved, at
least on Win98:

diff -urN apache_1.3.9-cygwin/src/main/http_protocol.c
apache_1.3.9-cygwin1/src/main/http_protocol.c
--- apache_1.3.9-cygwin/src/main/http_protocol.c      
 Sat Aug 14 09:21:20 1999
+++ apache_1.3.9-cygwin1/src/main/http_protocol.c     
 Tue Nov 14 00:03:40 2000
@@ -2065,7 +2065,7 @@
     /* Make fb unbuffered and non-blocking */
     ap_bsetflag(fb, B_RD, 0);
-#ifndef TPF
+#if !defined(TPF) && !defined(__CYGWIN__) &&
!defined(__CYGWIN32__)
     ap_bnonblock(fb, B_RD);
 #endif
     fd = ap_bfileno(fb, B_RD);

I don't had tested this patch on WinNT system (because
I have'n any system on my work whit this OS); I'll
give thanks to all people that can test this on WinNT
and say me the results obtained.

Some others changes to the source that I applied to
this port of Apache were:

diff -urN apache_1.3.9-cygwin/src/main/http_main.c
apache_1.3.9-cygwin1/src/main/http_main.c
--- apache_1.3.9-cygwin/src/main/http_main.c    Sat
Oct  2 17:15:52 1999
+++ apache_1.3.9-cygwin1/src/main/http_main.c   Tue
Nov 14 00:05:54 2000
@@ -2911,8 +2911,8 @@
     int x;

     chdir("/");
-#if !defined(MPE) && !defined(OS2) && !defined(TPF)
&& \
-    !defined(__CYGWIN__) && !defined(__CYGWIN32__)
+#if !defined(MPE) && !defined(OS2) && !defined(TPF)
/* && \
+    !defined(__CYGWIN__) && !defined(__CYGWIN32__) */
 /* Don't detach for MPE because child processes can't
survive the death of
    the parent. */
     if ((x = fork()) > 0)
diff -urN apache_1.3.9-cygwin/configure
apache_1.3.9-cygwin1/configure
--- apache_1.3.9-cygwin/configure       Sat Aug 14
08:29:06 1999
+++ apache_1.3.9-cygwin1/configure      Tue Nov 14
00:07:08 2000
@@ -241,6 +241,9 @@
         iflags_program="$iflags_program -e .exe"
         iflags_core="$iflags_core -e .exe"
         ;;
+    *-cygwin32* )
+        iflags_program="$iflags_program -e .exe"
+        ;;
 esac

 #   various other flags
diff -urN apache_1.3.9-cygwin/src/support/logresolve.c
apache_1.3.9-cygwin1/src/support/logresolve.c
--- apache_1.3.9-cygwin/src/support/logresolve.c      
 Mon Jul 13 11:33:00 1998
+++ apache_1.3.9-cygwin1/src/support/logresolve.c     
 Tue Nov 14 00:10:06 2000
@@ -95,7 +95,7 @@
  * statistics - obvious
  */

-#ifndef h_errno
+#if !defined(h_errno) && !defined(__CYGWIN__) &&
!defined(__CYGWIN32__)
 extern int h_errno; /* some machines don't have this
in their headers */
 #endif

Regards,

Edgar Ruiz


On Mon, Jun 05, 2000 at 11:00:47AM -0700, Ken Coleman
wrote:
>I've debugged a problem I reported a few weeks ago a
little bit more,
>and I can tell you fairly well what's happening.
>
>I'm trying to get Apache 1.3.12 running, and when I
run CGI scripts,
>the subprocesses that run the script become <defunct>
and stay that way
>forever.  It fork()s Apache, then calls one of the
exec() functions to
>run the CGI script synchronously.  In one of the
processes, it's
>spinning at the top of wait_subproc in sigproc.cc. 
It's appears that
>it will be in the loop forever, although I haven't
let the processes
>run for more than a few minutes.
>
>In that function, WaitForMultipleObjects() is
consistently returning
>WAIT_TIMEOUT, but proc_loop_wait is still 500.  It
appears that you set
>it to 0 when it's time for the process to terminate,
but somehow that
>has not occurred.  The process has already received
several
>PROC_CHILDTERMINATED signals into proc_subproc, which
is why it's set
>to <defunct> in the process table (zombie flag is
set).
>
>Any help would be greatly appreciated.  There are
several strace logs
>that I sent in my earlier message that may be useful
- it's dated
>5/23/2000.  It works under WinNT (I ran it on my work
system), but
>fails under Win9x.  There are two strace logs - one
for each OS.  I
>know you hate dealing with Win9x, but I just can't
really upgrade this
>system to Win2K.

_________________________________________________________
Do You Yahoo!?
Obtenga su dirección de correo-e gratis @yahoo.com
en http://correo.espanol.yahoo.com

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]