On Thu, Aug 28, 2014 at 05:30:09AM +0200, Mateusz Guzik wrote: > @@ -791,6 +791,8 @@ fill_kinfo_proc_only(struct proc *p, struct kinfo_proc > *kp) > struct ucred *cred; > struct sigacts *ps; > > + /* For proc_realparent. */ > + sx_assert(&proctree_lock, SX_LOCKED); > PROC_LOCK_ASSERT(p, MA_OWNED); > bzero(kp, sizeof(*kp)); > > @@ -920,7 +922,9 @@ fill_kinfo_proc_only(struct proc *p, struct kinfo_proc > *kp) > kp->ki_acflag = p->p_acflag; > kp->ki_lock = p->p_lock; > if (p->p_pptr) > - kp->ki_ppid = p->p_pptr->p_pid; > + kp->ki_ppid = proc_realparent(p)->p_pid; Is the check for p_pptr != NULL still needed for the call to proc_realparent() ? If yes, I think it indicates a bug in proc_realparent(), which should incorporate the check, instead of enforcing it on the callers. It seems to be there for the kernel process (pid 0).
If the test can be removed, and proc_realparent() called unconditionally, I suggest to remove assert about proctree_lock at the start of fill_kinfo_proc_only(), since the check is done in proc_realparent(). Whatever decision is made there, it can be implemented after your change is landed. The patch looks fine for me.
pgpVrUv1Dl5f7.pgp
Description: PGP signature
