Re: [libvirt] [PATCH 3/4] lxc: validate container process during load config

On Fri, May 30, 2008 at 11:40:00AM -0700, Dave Leskovec wrote: > Daniel P. Berrange wrote: >> On Fri, May 30, 2008 at 02:28:46AM -0400, Daniel Veillard wrote: >>> On Thu, May 29, 2008 at 03:20:15PM -0700, Dave Leskovec wrote: >>>> + if (ESRCH == errno) { >>>> + rc = 0; >>>> + DEBUG("pid %d no longer exists", def->id); >>>> + goto done; >>>> + } >>>> + >>>> + lxcError(NULL, NULL, VIR_ERR_INTERNAL_ERROR, >>>> + _("error checking container process: %d %s"), >>>> + def->id, strerror(errno)); >>>> + goto done; >>>> + } >>> The problem though is that by doing just a passive test for the PID >>> it feels like there is a possible race if the process counter rolled over and >>> another process with the same PID got create in the meantime. >>> i have the feeling that a test based on the state of the file descriptors >>> used to communicate with the container would be more reliable. Basically if the >>> container disapear, then the pipe should get in a half-closed state, >>> detecting the change at that level sounds like it would be more reliable, >>> don't you think so ? >> Yes, after checking the PID still exists, it needs to validate /proc/$PID/exe >> to verify it points to the binary we expect it to. > Hmmm.... Worked with this a bit and I don't think we can reliably know what to > expect /proc/$PID/exe to point to. For scripts, /proc/$PID/exe seems to point > the shell. Also, if the container does an exec, /proc/$PID/exe points to > whatever it exec'd rather than the init program. Well the path won't change once launched, so why not store the original value of /proc/$PID/exe in the /var/lib/libivrt/lxc/NAME.pid file too, so you can read it back out later & validate.