path: root/proc/wait.c

/* Implementation of wait
   Copyright (C) 1994, 1995, 1996, 2001 Free Software Foundation

   This program is free software; you can redistribute it and/or
   modify it under the terms of the GNU General Public License as
   published by the Free Software Foundation; either version 2, or (at
   your option) any later version.

   This program is distributed in the hope that it will be useful, but
   WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */

#include <mach.h>
#include <sys/types.h>
#include <hurd/hurd_types.h>
#include <sys/resource.h>
#include <sys/time.h>
#include <mach/task_info.h>

#include "proc.h"

#include <signal.h>
#include <string.h>
#include <sys/wait.h>
#include <errno.h>
#include <stdlib.h>
#include <assert.h>

#include "process_S.h"
#include <mach/mig_errors.h>

static inline void
rusage_add (struct rusage *acc, const struct rusage *b)
{
  timeradd (&acc->ru_utime, &b->ru_utime, &acc->ru_utime);
  timeradd (&acc->ru_stime, &b->ru_stime, &acc->ru_stime);

  /* Check <bits/resource.h> definition of `struct rusage'
     to make sure this gets all the fields.  */
  acc->ru_maxrss += b->ru_maxrss;
  acc->ru_ixrss += b->ru_ixrss;
  acc->ru_idrss += b->ru_idrss;
  acc->ru_isrss += b->ru_isrss;
  acc->ru_minflt += b->ru_minflt;
  acc->ru_majflt += b->ru_majflt;
  acc->ru_nswap += b->ru_nswap;
  acc->ru_inblock += b->ru_inblock;
  acc->ru_oublock += b->ru_oublock;
  acc->ru_msgsnd += b->ru_msgsnd;
  acc->ru_msgrcv += b->ru_msgrcv;
  acc->ru_nsignals += b->ru_nsignals;
  acc->ru_nvcsw += b->ru_nvcsw;
  acc->ru_nivcsw += b->ru_nivcsw;
}

/* XXX This is real half-assed.

   We want to collect usage statistics from dead processes to return to its
   parent for its proc_wait calls and its aggregate child statistics.

   The microkernel provides no access to this information once the task is
   terminated.  So the best we can do is take a sample at some time while
   the task is still alive but not too long before it dies.  Our results
   are always inaccurate, because they don't account for the final part of
   the task's lifetime.  But perhaps it's better than nothing at all.

   The obvious place to take this sample is in proc_mark_exit, which in
   normal circumstances a task is calling immediately before terminating
   itself.  So in the best of cases, our data omits only the interval in
   which our RPC returns to the task and it calls task_terminate.  We could
   take samples in other places just to have something rather than nothing
   if the task dies unexpectedly (e.g. SIGKILL); but it may not be worthwhile
   since the end result is never going to be accurate anyway.

   The only way to get correct results is by adding some microkernel
   feature to report the task statistics data post-mortem.  */

void
sample_rusage (struct proc *p)
{
  struct task_basic_info bi;
  struct task_events_info ei;
  struct task_thread_times_info tti;
  mach_msg_type_number_t count;
  error_t err;

  count = TASK_BASIC_INFO_COUNT;
  err = task_info (p->p_task, TASK_BASIC_INFO,
		   (task_info_t) &bi, &count);
  if (err)
    memset (&bi, 0, sizeof bi);

  count = TASK_EVENTS_INFO_COUNT;
  err = task_info (p->p_task, TASK_EVENTS_INFO,
		   (task_info_t) &ei, &count);
  if (err)
    memset (&ei, 0, sizeof ei);

  count = TASK_THREAD_TIMES_INFO_COUNT;
  err = task_info (p->p_task, TASK_THREAD_TIMES_INFO,
		   (task_info_t) &tti, &count);
  if (err)
    memset (&tti, 0, sizeof tti);

  time_value_add (&bi.user_time, &tti.user_time);
  time_value_add (&bi.system_time, &tti.system_time);

  memset (&p->p_rusage, 0, sizeof (struct rusage));

  p->p_rusage.ru_utime.tv_sec = bi.user_time.seconds;
  p->p_rusage.ru_utime.tv_usec = bi.user_time.microseconds;
  p->p_rusage.ru_stime.tv_sec = bi.system_time.seconds;
  p->p_rusage.ru_stime.tv_usec = bi.system_time.microseconds;

  /* These statistics map only approximately.  */
  p->p_rusage.ru_majflt = ei.pageins;
  p->p_rusage.ru_minflt = ei.faults - ei.pageins;
  p->p_rusage.ru_msgsnd = ei.messages_sent; /* Mach IPC, not SysV IPC */
  p->p_rusage.ru_msgrcv = ei.messages_received; /* ditto */
}

/* Return nonzero if a `waitpid' on WAIT_PID by a process
   in MYPGRP cares about the death of PID/PGRP.  */
static inline int
waiter_cares (pid_t wait_pid, pid_t mypgrp,
	      pid_t pid, pid_t pgrp)
{
  return (wait_pid == pid ||
	  wait_pid == -pgrp ||
	  wait_pid == WAIT_ANY ||
	  (wait_pid == WAIT_MYPGRP && pgrp == mypgrp));
}

/* A process is dying.  Send SIGCHLD to the parent.
   Wake the parent if it is waiting for us to exit. */
void
alert_parent (struct proc *p)
{
  /* We accumulate the aggregate usage stats of all our dead children.  */
  rusage_add (&p->p_parent->p_child_rusage, &p->p_rusage);

  send_signal (p->p_parent->p_msgport, SIGCHLD, p->p_parent->p_task);

  if (!p->p_exiting)
    {
      p->p_status = W_EXITCODE (0, SIGKILL);
      p->p_sigcode = -1;
    }

  if (p->p_parent->p_waiting)
    {
      condition_broadcast (&p->p_parent->p_wakeup);
      p->p_parent->p_waiting = 0;
    }
}

kern_return_t
S_proc_wait (struct proc *p,
	     mach_port_t reply_port,
	     mach_msg_type_name_t reply_port_type,
	     pid_t pid,
	     int options,
	     int *status,
	     int *sigcode,
	     struct rusage *ru,
	     pid_t *pid_status)
{
  int cancel;

  int reap (struct proc *child)
    {
      if (child->p_waited
	  || (!child->p_dead
	      && (!child->p_stopped
		  || !(child->p_traced || (options & WUNTRACED)))))
	return 0;
      child->p_waited = 1;
      *status = child->p_status;
      *sigcode = child->p_sigcode;
      *ru = child->p_rusage; /* all zeros if !p_dead */
      *pid_status = child->p_pid;
      if (child->p_dead)
	complete_exit (child);
      return 1;
    }

  if (!p)
    return EOPNOTSUPP;

 start_over:
  /* See if we can satisfy the request with a stopped
     child; also check for invalid arguments here. */
  if (!p->p_ochild)
    return ECHILD;

  if (pid > 0)
    {
      struct proc *child = pid_find_allow_zombie (pid);
      if (!child || child->p_parent != p)
	return ECHILD;
      if (reap (child))
	return 0;
    }
  else
    {
      struct proc *child;
      int had_a_match = pid == 0;

      for (child = p->p_ochild; child; child = child->p_sib)
	if (waiter_cares (pid, p->p_pgrp->pg_pgid,
			  child->p_pid, child->p_pgrp->pg_pgid))
	  {
	    if (reap (child))
	      return 0;
	    had_a_match = 1;
	  }

      if (!had_a_match)
	return ECHILD;
    }

  if (options & WNOHANG)
    return EWOULDBLOCK;

  p->p_waiting = 1;
  cancel = hurd_condition_wait (&p->p_wakeup, &global_lock);
  if (p->p_dead)
    return EOPNOTSUPP;
  if (cancel)
    return EINTR;
  goto start_over;
}

/* Implement proc_mark_stop as described in <hurd/process.defs>. */
kern_return_t
S_proc_mark_stop (struct proc *p,
		  int signo,
		  int sigcode)
{
  if (!p)
    return EOPNOTSUPP;

  p->p_stopped = 1;
  p->p_status = W_STOPCODE (signo);
  p->p_sigcode = sigcode;
  p->p_waited = 0;

  if (p->p_parent->p_waiting)
    {
      condition_broadcast (&p->p_parent->p_wakeup);
      p->p_parent->p_waiting = 0;
    }

  if (!p->p_parent->p_nostopcld)
    send_signal (p->p_parent->p_msgport, SIGCHLD, p->p_parent->p_task);

  return 0;
}

/* Implement proc_mark_exit as described in <hurd/process.defs>. */
kern_return_t
S_proc_mark_exit (struct proc *p,
		  int status,
		  int sigcode)
{
  if (!p)
    return EOPNOTSUPP;

  if (WIFSTOPPED (status))
    return EINVAL;

  sample_rusage (p);		/* See comments above sample_rusage.  */

  if (p->p_exiting)
    return EBUSY;

  p->p_exiting = 1;
  p->p_status = status;
  p->p_sigcode = sigcode;
  return 0;
}

/* Implement proc_mark_cont as described in <hurd/process.defs>. */
kern_return_t
S_proc_mark_cont (struct proc *p)
{
  if (!p)
    return EOPNOTSUPP;
  p->p_stopped = 0;
  return 0;
}

/* Implement proc_mark_traced as described in <hurd/process.defs>. */
kern_return_t
S_proc_mark_traced (struct proc *p)
{
  if (!p)
    return EOPNOTSUPP;
  p->p_traced = 1;
  return 0;
}

/* Implement proc_mark_nostopchild as described in <hurd/process.defs>. */
kern_return_t
S_proc_mod_stopchild (struct proc *p,
		      int value)
{
  if (!p)
    return EOPNOTSUPP;
  /* VALUE is nonzero if we should send SIGCHLD.  */
  p->p_nostopcld = ! value;
  return 0;
}