BUG: File Descriptor leak

while working with onmpeg parameter, I have found that some file descriptors were leak to sub-process.

It can have several effects:

• a good subprocess can wait forever that a file is not used anymore (I am in this case)
• netcam may not restart after a crash (device is busy)
• an evil or buggy subprocess can play with netcam or opened jpg, mmpeg...

The boggus functions in event.c are exec_command, send_mail, send_sms (functions that are using fork and an exec* function) The three functions need to call

• ffmpeg_close(cnt->ffmpeg_new)
• ffmpeg_close(cnt->ffmpeg_motion)
• ffmpeg_close(cnt->ffmpeg_timelapse)
• netcam_close(netcam)

send_sms and send_mail also forgot to call vid_close().

Test case

Add in motion.conf onmpeg /usr/sbin/lsof -n >> /tmp/lsof.out

Example

motion  6691 kmaster    5w   REG   8,24 722190 4514570 /mnt/data/webcam/2005/0321/20050321-081240-03.avi
sh      6716 kmaster    5w   REG   8,24 722190 4514570 /mnt/data/webcam/2005/0321/20050321-081240-03.avi

Environment

Motion version:	3.1.19
ffmpeg version:	0.4.9-pre1
Shared libraries:	curl, xmlrpc, ffmpeg, mysql, postgresql
Server OS:	Fedora Core 3, kernel 2.6.10

-- ChristopheGRENIER - 21 Mar 2005

Follow up

I have been playing with this for several hours.

I tried this code in exec_command():

#ifdef HAVE_FFMPEG
    if (cnt->ffmpeg_new) ffmpeg_close(cnt->ffmpeg_new);
    if (cnt->ffmpeg_motion) ffmpeg_close(cnt->ffmpeg_motion);
    if (cnt->ffmpeg_timelapse) ffmpeg_close(cnt->ffmpeg_timelapse);
#endif /* HAVE_FFMPEG */

The is not enough. In this example I have timelapse enabled on 9 cameras. And it seems all open exec forks carry the same file descriptors which then needs to be closed in each. (output from your testcase)

COMMAND   PID USER   FD   TYPE DEVICE  SIZE    NODE NAME
motion  16806 root   34w   REG  253,0 18034 9930483 /usr/local/apache2/htdocs/cam5/2005032523-timelapse.mpg
sh      16833 root   34w   REG  253,0 18034 9930483 /usr/local/apache2/htdocs/cam5/2005032523-timelapse.mpg
sh      16836 root   34w   REG  253,0 18034 9930483 /usr/local/apache2/htdocs/cam5/2005032523-timelapse.mpg
sh      16837 root   34w   REG  253,0 18034 9930483 /usr/local/apache2/htdocs/cam5/2005032523-timelapse.mpg
sh      16842 root   34w   REG  253,0 18034 9930483 /usr/local/apache2/htdocs/cam5/2005032523-timelapse.mpg
sh      16843 root   34w   REG  253,0 18034 9930483 /usr/local/apache2/htdocs/cam5/2005032523-timelapse.mpg
sh      16848 root   34w   REG  253,0 18034 9930483 /usr/local/apache2/htdocs/cam5/2005032523-timelapse.mpg
sh      16849 root   34w   REG  253,0 18034 9930483 /usr/local/apache2/htdocs/cam5/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE  SIZE    NODE NAME
motion  16806 root   41w   REG  253,0 25898 9228223 /usr/local/apache2/htdocs/cam4/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE SIZE     NODE NAME
motion  16806 root   38w   REG  253,0 9515 10716748 /usr/local/apache2/htdocs/cam2/2005032523-timelapse.mpg
sh      16843 root   38w   REG  253,0 9515 10716748 /usr/local/apache2/htdocs/cam2/2005032523-timelapse.mpg
sh      16848 root   38w   REG  253,0 9515 10716748 /usr/local/apache2/htdocs/cam2/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE SIZE    NODE NAME
motion  16806 root   35w   REG  253,0 4409 8735556 /usr/local/apache2/htdocs/cam6/2005032523-timelapse.mpg
sh      16836 root   35w   REG  253,0 4409 8735556 /usr/local/apache2/htdocs/cam6/2005032523-timelapse.mpg
sh      16837 root   35w   REG  253,0 4409 8735556 /usr/local/apache2/htdocs/cam6/2005032523-timelapse.mpg
sh      16843 root   35w   REG  253,0 4409 8735556 /usr/local/apache2/htdocs/cam6/2005032523-timelapse.mpg
sh      16848 root   35w   REG  253,0 4409 8735556 /usr/local/apache2/htdocs/cam6/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE  SIZE    NODE NAME
motion  16806 root   33w   REG  253,0 38604 8507904 /usr/local/apache2/htdocs/cam9/2005032523-timelapse.mpg
sh      16833 root   33w   REG  253,0 38604 8507904 /usr/local/apache2/htdocs/cam9/2005032523-timelapse.mpg
sh      16836 root   33w   REG  253,0 38604 8507904 /usr/local/apache2/htdocs/cam9/2005032523-timelapse.mpg
sh      16837 root   33w   REG  253,0 38604 8507904 /usr/local/apache2/htdocs/cam9/2005032523-timelapse.mpg
sh      16843 root   33w   REG  253,0 38604 8507904 /usr/local/apache2/htdocs/cam9/2005032523-timelapse.mpg
sh      16848 root   33w   REG  253,0 38604 8507904 /usr/local/apache2/htdocs/cam9/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE  SIZE     NODE NAME
motion  16806 root   37w   REG  253,0 12931 10094976 /usr/local/apache2/htdocs/cam1/2005032523-timelapse.mpg
sh      16843 root   37w   REG  253,0 12931 10094976 /usr/local/apache2/htdocs/cam1/2005032523-timelapse.mpg
sh      16848 root   37w   REG  253,0 12931 10094976 /usr/local/apache2/htdocs/cam1/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE  SIZE    NODE NAME
motion  16806 root   36w   REG  253,0 16806 9522034 /usr/local/apache2/htdocs/cam7/2005032523-timelapse.mpg
sh      16843 root   36w   REG  253,0 16806 9522034 /usr/local/apache2/htdocs/cam7/2005032523-timelapse.mpg
sh      16848 root   36w   REG  253,0 16806 9522034 /usr/local/apache2/htdocs/cam7/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE  SIZE     NODE NAME
motion  16806 root   39w   REG  253,0 13586 10814151 /usr/local/apache2/htdocs/cam3/2005032523-timelapse.mpg
sh      16848 root   39w   REG  253,0 13586 10814151 /usr/local/apache2/htdocs/cam3/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE  SIZE     NODE NAME
motion  16806 root   40w   REG  253,0 24541 10749052 /usr/local/apache2/htdocs/cam8/2005032523-timelapse.mpg
COMMAND   PID USER   FD   TYPE DEVICE  SIZE     NODE NAME
motion  16806 root   42w   REG  253,0 17268 10094991 /usr/local/apache2/htdocs/cam1/01-20050325233012.avi
COMMAND   PID USER   FD   TYPE DEVICE  SIZE     NODE NAME
motion  16806 root   43w   REG  253,0 15802 10094994 /usr/local/apache2/htdocs/cam1/01-20050325233012m.avi

-- KennethLavrsen - 25 Mar 2005

Fix record

This fix will be known as the BFG9000 fix (ever played Quake?).

In send_sms, send_mail and exec_command I added this

if (!fork()) {
    int i;
  
    /* Detach from parent */
    setsid();

    /* Close any file descripter except console because we will like to see error messages */
    for (i=getdtablesize(); i>2; --i)
        close(i);
      
    exec....

That cleans up any mess effectively and without too much cpu overhead I believe.

Implemented in motion-3.2.1_snap10 and motion-3.1.20_snap3.