kernel - Remove SMP bottlenecks on uidinfo, descriptors, and lockf

[dragonfly.git] / sys / kern / kern_exit.c

/*
 * Copyright (c) 1982, 1986, 1989, 1991, 1993
 *      The Regents of the University of California.  All rights reserved.
 * (c) UNIX System Laboratories, Inc.
 * All or some portions of this file are derived from material licensed
 * to the University of California by American Telephone and Telegraph
 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
 * the permission of UNIX System Laboratories, Inc.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *      @(#)kern_exit.c 8.7 (Berkeley) 2/12/94
 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $
 */

#include "opt_ktrace.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/proc.h>
#include <sys/ktrace.h>
#include <sys/pioctl.h>
#include <sys/tty.h>
#include <sys/wait.h>
#include <sys/vnode.h>
#include <sys/resourcevar.h>
#include <sys/signalvar.h>
#include <sys/taskqueue.h>
#include <sys/ptrace.h>
#include <sys/acct.h>           /* for acct_process() function prototype */
#include <sys/filedesc.h>
#include <sys/shm.h>
#include <sys/sem.h>
#include <sys/jail.h>
#include <sys/kern_syscall.h>
#include <sys/unistd.h>
#include <sys/eventhandler.h>
#include <sys/dsched.h>

#include <vm/vm.h>
#include <vm/vm_param.h>
#include <sys/lock.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_extern.h>
#include <sys/user.h>

#include <sys/refcount.h>
#include <sys/thread2.h>
#include <sys/spinlock2.h>
#include <sys/mplock2.h>

#include <machine/vmm.h>

static void reaplwps(void *context, int dummy);
static void reaplwp(struct lwp *lp);
static void killlwps(struct lwp *lp);

static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");

/*
 * callout list for things to do at exit time
 */
struct exitlist {
        exitlist_fn function;
        TAILQ_ENTRY(exitlist) next;
};

TAILQ_HEAD(exit_list_head, exitlist);
static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);

/*
 * LWP reaper data
 */
static struct task *deadlwp_task[MAXCPU];
static struct lwplist deadlwp_list[MAXCPU];
static struct lwkt_token deadlwp_token[MAXCPU];

/*
 * exit --
 *      Death of process.
 *
 * SYS_EXIT_ARGS(int rval)
 */
int
sys_exit(struct exit_args *uap)
{
        exit1(W_EXITCODE(uap->rval, 0));
        /* NOTREACHED */
}

/*
 * Extended exit --
 *      Death of a lwp or process with optional bells and whistles.
 */
int
sys_extexit(struct extexit_args *uap)
{
        struct proc *p = curproc;
        int action, who;
        int error;

        action = EXTEXIT_ACTION(uap->how);
        who = EXTEXIT_WHO(uap->how);

        /* Check parameters before we might perform some action */
        switch (who) {
        case EXTEXIT_PROC:
        case EXTEXIT_LWP:
                break;
        default:
                return (EINVAL);
        }

        switch (action) {
        case EXTEXIT_SIMPLE:
                break;
        case EXTEXIT_SETINT:
                error = copyout(&uap->status, uap->addr, sizeof(uap->status));
                if (error)
                        return (error);
                break;
        default:
                return (EINVAL);
        }

        lwkt_gettoken(&p->p_token);

        switch (who) {
        case EXTEXIT_LWP:
                /*
                 * Be sure only to perform a simple lwp exit if there is at
                 * least one more lwp in the proc, which will call exit1()
                 * later, otherwise the proc will be an UNDEAD and not even a
                 * SZOMB!
                 */
                if (p->p_nthreads > 1) {
                        lwp_exit(0, NULL);      /* called w/ p_token held */
                        /* NOT REACHED */
                }
                /* else last lwp in proc:  do the real thing */
                /* FALLTHROUGH */
        default:        /* to help gcc */
        case EXTEXIT_PROC:
                lwkt_reltoken(&p->p_token);
                exit1(W_EXITCODE(uap->status, 0));
                /* NOTREACHED */
        }

        /* NOTREACHED */
        lwkt_reltoken(&p->p_token);     /* safety */
}

/*
 * Kill all lwps associated with the current process except the
 * current lwp.   Return an error if we race another thread trying to
 * do the same thing and lose the race.
 *
 * If forexec is non-zero the current thread and process flags are
 * cleaned up so they can be reused.
 *
 * Caller must hold curproc->p_token
 */
int
killalllwps(int forexec)
{
        struct lwp *lp = curthread->td_lwp;
        struct proc *p = lp->lwp_proc;
        int fakestop;

        /*
         * Interlock against P_WEXIT.  Only one of the process's thread
         * is allowed to do the master exit.
         */
        if (p->p_flags & P_WEXIT)
                return (EALREADY);
        p->p_flags |= P_WEXIT;

        /*
         * Set temporary stopped state in case we are racing a coredump.
         * Otherwise the coredump may hang forever.
         */
        if (lp->lwp_mpflags & LWP_MP_WSTOP) {
                fakestop = 0;
        } else {
                atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
                ++p->p_nstopped;
                fakestop = 1;
                wakeup(&p->p_nstopped);
        }

        /*
         * Interlock with LWP_MP_WEXIT and kill any remaining LWPs
         */
        atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
        if (p->p_nthreads > 1)
                killlwps(lp);

        /*
         * Undo temporary stopped state
         */
        if (fakestop) {
                atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
                --p->p_nstopped;
        }

        /*
         * If doing this for an exec, clean up the remaining thread
         * (us) for continuing operation after all the other threads
         * have been killed.
         */
        if (forexec) {
                atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
                p->p_flags &= ~P_WEXIT;
        }
        return(0);
}

/*
 * Kill all LWPs except the current one.  Do not try to signal
 * LWPs which have exited on their own or have already been
 * signaled.
 */
static void
killlwps(struct lwp *lp)
{
        struct proc *p = lp->lwp_proc;
        struct lwp *tlp;

        /*
         * Kill the remaining LWPs.  We must send the signal before setting
         * LWP_MP_WEXIT.  The setting of WEXIT is optional but helps reduce
         * races.  tlp must be held across the call as it might block and
         * allow the target lwp to rip itself out from under our loop.
         */
        FOREACH_LWP_IN_PROC(tlp, p) {
                LWPHOLD(tlp);
                lwkt_gettoken(&tlp->lwp_token);
                if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) {
                        atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT);
                        lwpsignal(p, tlp, SIGKILL);
                }
                lwkt_reltoken(&tlp->lwp_token);
                LWPRELE(tlp);
        }

        /*
         * Wait for everything to clear out.  Also make sure any tstop()s
         * are signalled (we are holding p_token for the interlock).
         */
        wakeup(p);
        while (p->p_nthreads > 1)
                tsleep(&p->p_nthreads, 0, "killlwps", 0);
}

/*
 * Exit: deallocate address space and other resources, change proc state
 * to zombie, and unlink proc from allproc and parent's lists.  Save exit
 * status and rusage for wait().  Check for child processes and orphan them.
 */
void
exit1(int rv)
{
        struct thread *td = curthread;
        struct proc *p = td->td_proc;
        struct lwp *lp = td->td_lwp;
        struct proc *q;
        struct proc *pp;
        struct proc *reproc;
        struct sysreaper *reap;
        struct vmspace *vm;
        struct vnode *vtmp;
        struct exitlist *ep;
        int error;

        lwkt_gettoken(&p->p_token);

        if (p->p_pid == 1) {
                kprintf("init died (signal %d, exit %d)\n",
                    WTERMSIG(rv), WEXITSTATUS(rv));
                panic("Going nowhere without my init!");
        }
        varsymset_clean(&p->p_varsymset);
        lockuninit(&p->p_varsymset.vx_lock);

        /*
         * Kill all lwps associated with the current process, return an
         * error if we race another thread trying to do the same thing
         * and lose the race.
         */
        error = killalllwps(0);
        if (error) {
                lwp_exit(0, NULL);
                /* NOT REACHED */
        }

        /* are we a task leader? */
        if (p == p->p_leader) {
                struct kill_args killArgs;
                killArgs.signum = SIGKILL;
                q = p->p_peers;
                while(q) {
                        killArgs.pid = q->p_pid;
                        /*
                         * The interface for kill is better
                         * than the internal signal
                         */
                        sys_kill(&killArgs);
                        q = q->p_peers;
                }
                while (p->p_peers) 
                        tsleep((caddr_t)p, 0, "exit1", 0);
        }

#ifdef PGINPROF
        vmsizmon();
#endif
        STOPEVENT(p, S_EXIT, rv);
        p->p_flags |= P_POSTEXIT;       /* stop procfs stepping */

        /* 
         * Check if any loadable modules need anything done at process exit.
         * e.g. SYSV IPC stuff
         * XXX what if one of these generates an error?
         */
        p->p_xstat = rv;

        /*
         * XXX: imho, the eventhandler stuff is much cleaner than this.
         *      Maybe we should move everything to use eventhandler.
         */
        TAILQ_FOREACH(ep, &exit_list, next) 
                (*ep->function)(td);

        if (p->p_flags & P_PROFIL)
                stopprofclock(p);

        SIGEMPTYSET(p->p_siglist);
        SIGEMPTYSET(lp->lwp_siglist);
        if (timevalisset(&p->p_realtimer.it_value))
                callout_stop_sync(&p->p_ithandle);

        /*
         * Reset any sigio structures pointing to us as a result of
         * F_SETOWN with our pid.
         */
        funsetownlst(&p->p_sigiolst);

        /*
         * Close open files and release open-file table.
         * This may block!
         */
        fdfree(p, NULL);

        if (p->p_leader->p_peers) {
                q = p->p_leader;
                while(q->p_peers != p)
                        q = q->p_peers;
                q->p_peers = p->p_peers;
                wakeup((caddr_t)p->p_leader);
        }

        /*
         * XXX Shutdown SYSV semaphores
         */
        semexit(p);

        /* The next two chunks should probably be moved to vmspace_exit. */
        vm = p->p_vmspace;

        /*
         * Clean up data related to virtual kernel operation.  Clean up
         * any vkernel context related to the current lwp now so we can
         * destroy p_vkernel.
         */
        if (p->p_vkernel) {
                vkernel_lwp_exit(lp);
                vkernel_exit(p);
        }

        /*
         * Release the user portion of address space.  The exitbump prevents
         * the vmspace from being completely eradicated (using holdcnt).
         * This releases references to vnodes, which could cause I/O if the
         * file has been unlinked.  We need to do this early enough that
         * we can still sleep.
         *
         * We can't free the entire vmspace as the kernel stack may be mapped
         * within that space also.
         *
         * Processes sharing the same vmspace may exit in one order, and
         * get cleaned up by vmspace_exit() in a different order.  The
         * last exiting process to reach this point releases as much of
         * the environment as it can, and the last process cleaned up
         * by vmspace_exit() (which decrements exitingcnt) cleans up the
         * remainder.
         *
         * NOTE: Releasing p_token around this call is helpful if the
         *       vmspace had a huge RSS.  Otherwise some other process
         *       trying to do an allproc or other scan (like 'ps') may
         *       stall for a long time.
         */
        lwkt_reltoken(&p->p_token);
        vmspace_relexit(vm);
        lwkt_gettoken(&p->p_token);

        if (SESS_LEADER(p)) {
                struct session *sp = p->p_session;

                if (sp->s_ttyvp) {
                        /*
                         * We are the controlling process.  Signal the 
                         * foreground process group, drain the controlling
                         * terminal, and revoke access to the controlling
                         * terminal.
                         *
                         * NOTE: while waiting for the process group to exit
                         * it is possible that one of the processes in the
                         * group will revoke the tty, so the ttyclosesession()
                         * function will re-check sp->s_ttyvp.
                         */
                        if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
                                if (sp->s_ttyp->t_pgrp)
                                        pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
                                ttywait(sp->s_ttyp);
                                ttyclosesession(sp, 1); /* also revoke */
                        }
                        /*
                         * Release the tty.  If someone has it open via
                         * /dev/tty then close it (since they no longer can
                         * once we've NULL'd it out).
                         */
                        ttyclosesession(sp, 0);

                        /*
                         * s_ttyp is not zero'd; we use this to indicate
                         * that the session once had a controlling terminal.
                         * (for logging and informational purposes)
                         */
                }
                sp->s_leader = NULL;
        }
        fixjobc(p, p->p_pgrp, 0);
        (void)acct_process(p);
#ifdef KTRACE
        /*
         * release trace file
         */
        if (p->p_tracenode)
                ktrdestroy(&p->p_tracenode);
        p->p_traceflag = 0;
#endif
        /*
         * Release reference to text vnode
         */
        if ((vtmp = p->p_textvp) != NULL) {
                p->p_textvp = NULL;
                vrele(vtmp);
        }

        /* Release namecache handle to text file */
        if (p->p_textnch.ncp)
                cache_drop(&p->p_textnch);

        /*
         * We have to handle PPWAIT here or proc_move_allproc_zombie()
         * will block on the PHOLD() the parent is doing.
         *
         * We are using the flag as an interlock so an atomic op is
         * necessary to synchronize with the parent's cpu.
         */
        if (p->p_flags & P_PPWAIT) {
                if (p->p_pptr && p->p_pptr->p_upmap)
                        atomic_add_int(&p->p_pptr->p_upmap->invfork, -1);
                atomic_clear_int(&p->p_flags, P_PPWAIT);
                wakeup(p->p_pptr);
        }

        /*
         * Move the process to the zombie list.  This will block
         * until the process p_lock count reaches 0.  The process will
         * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
         * which is called from cpu_proc_exit().
         *
         * Interlock against waiters using p_waitgen.  We increment
         * p_waitgen after completing the move of our process to the
         * zombie list.
         *
         * WARNING: pp becomes stale when we block, clear it now as a
         *          reminder.
         */
        proc_move_allproc_zombie(p);
        pp = p->p_pptr;
        atomic_add_long(&pp->p_waitgen, 1);
        pp = NULL;

        /*
         * release controlled reaper for exit if we own it and return the
         * remaining reaper (the one for us), which we will drop after we
         * are done.
         */
        reap = reaper_exit(p);

        /*
         * Reparent all of this process's children to the init process or
         * to the designated reaper.  We must hold the reaper's p_token in
         * order to safely mess with p_children.
         *
         * We already hold p->p_token (to remove the children from our list).
         */
        reproc = NULL;
        q = LIST_FIRST(&p->p_children);
        if (q) {
                reproc = reaper_get(reap);
                lwkt_gettoken(&reproc->p_token);
                while ((q = LIST_FIRST(&p->p_children)) != NULL) {
                        PHOLD(q);
                        lwkt_gettoken(&q->p_token);
                        if (q != LIST_FIRST(&p->p_children)) {
                                lwkt_reltoken(&q->p_token);
                                PRELE(q);
                                continue;
                        }
                        LIST_REMOVE(q, p_sibling);
                        LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling);
                        q->p_pptr = reproc;
                        q->p_ppid = reproc->p_pid;
                        q->p_sigparent = SIGCHLD;

                        /*
                         * Traced processes are killed
                         * since their existence means someone is screwing up.
                         */
                        if (q->p_flags & P_TRACED) {
                                q->p_flags &= ~P_TRACED;
                                ksignal(q, SIGKILL);
                        }
                        lwkt_reltoken(&q->p_token);
                        PRELE(q);
                }
                lwkt_reltoken(&reproc->p_token);
                wakeup(reproc);
        }

        /*
         * Save exit status and final rusage info, adding in child rusage
         * info and self times.
         */
        calcru_proc(p, &p->p_ru);
        ruadd(&p->p_ru, &p->p_cru);

        /*
         * notify interested parties of our demise.
         */
        KNOTE(&p->p_klist, NOTE_EXIT);

        /*
         * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
         * flag set, or if the handler is set to SIG_IGN, notify the reaper
         * instead (it will handle this situation).
         *
         * NOTE: The reaper can still be the parent process.
         *
         * (must reload pp)
         */
        if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
                if (reproc == NULL)
                        reproc = reaper_get(reap);
                proc_reparent(p, reproc);
        }
        if (reproc)
                PRELE(reproc);
        if (reap)
                reaper_drop(reap);

        /*
         * Signal (possibly new) parent.
         */
        pp = p->p_pptr;
        PHOLD(pp);
        if (p->p_sigparent && pp != initproc) {
                int sig = p->p_sigparent;

                if (sig != SIGUSR1 && sig != SIGCHLD)
                        sig = SIGCHLD;
                ksignal(pp, sig);
        } else {
                ksignal(pp, SIGCHLD);
        }
        p->p_flags &= ~P_TRACED;
        PRELE(pp);

        /*
         * cpu_exit is responsible for clearing curproc, since
         * it is heavily integrated with the thread/switching sequence.
         *
         * Other substructures are freed from wait().
         */
        if (p->p_limit) {
                struct plimit *rlimit;

                rlimit = p->p_limit;
                p->p_limit = NULL;
                plimit_free(rlimit);
        }

        /*
         * Finally, call machine-dependent code to release as many of the
         * lwp's resources as we can and halt execution of this thread.
         *
         * pp is a wild pointer now but still the correct wakeup() target.
         * lwp_exit() only uses it to send the wakeup() signal to the likely
         * parent.  Any reparenting race that occurs will get a signal
         * automatically and not be an issue.
         */
        lwp_exit(1, pp);
}

/*
 * Eventually called by every exiting LWP
 *
 * p->p_token must be held.  mplock may be held and will be released.
 */
void
lwp_exit(int masterexit, void *waddr)
{
        struct thread *td = curthread;
        struct lwp *lp = td->td_lwp;
        struct proc *p = lp->lwp_proc;
        int dowake = 0;

        /*
         * Release the current user process designation on the process so
         * the userland scheduler can work in someone else.
         */
        p->p_usched->release_curproc(lp);

        /*
         * lwp_exit() may be called without setting LWP_MP_WEXIT, so
         * make sure it is set here.
         */
        ASSERT_LWKT_TOKEN_HELD(&p->p_token);
        atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);

        /*
         * Clean up any virtualization
         */
        if (lp->lwp_vkernel)
                vkernel_lwp_exit(lp);

        if (td->td_vmm)
                vmm_vmdestroy();

        /*
         * Clean up select/poll support
         */
        kqueue_terminate(&lp->lwp_kqueue);

        /*
         * Clean up any syscall-cached ucred or rlimit.
         */
        if (td->td_ucred) {
                crfree(td->td_ucred);
                td->td_ucred = NULL;
        }
        if (td->td_limit) {
                struct plimit *rlimit;

                rlimit = td->td_limit;
                td->td_limit = NULL;
                plimit_free(rlimit);
        }

        /*
         * Cleanup any cached descriptors for this thread
         */
        if (p->p_fd)
                fexitcache(td);

        /*
         * Nobody actually wakes us when the lock
         * count reaches zero, so just wait one tick.
         */
        while (lp->lwp_lock > 0)
                tsleep(lp, 0, "lwpexit", 1);

        /* Hand down resource usage to our proc */
        ruadd(&p->p_ru, &lp->lwp_ru);

        /*
         * If we don't hold the process until the LWP is reaped wait*()
         * may try to dispose of its vmspace before all the LWPs have
         * actually terminated.
         */
        PHOLD(p);

        /*
         * Do any remaining work that might block on us.  We should be
         * coded such that further blocking is ok after decrementing
         * p_nthreads but don't take the chance.
         */
        dsched_exit_thread(td);
        biosched_done(curthread);

        /*
         * We have to use the reaper for all the LWPs except the one doing
         * the master exit.  The LWP doing the master exit can just be
         * left on p_lwps and the process reaper will deal with it
         * synchronously, which is much faster.
         *
         * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
         *
         * The process is left held until the reaper calls lwp_dispose() on
         * the lp (after calling lwp_wait()).
         */
        if (masterexit == 0) {
                int cpu = mycpuid;

                lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
                --p->p_nthreads;
                if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
                        dowake = 1;
                lwkt_gettoken(&deadlwp_token[cpu]);
                LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry);
                taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]);
                lwkt_reltoken(&deadlwp_token[cpu]);
        } else {
                --p->p_nthreads;
                if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
                        dowake = 1;
        }

        /*
         * We no longer need p_token.
         *
         * Tell the userland scheduler that we are going away
         */
        lwkt_reltoken(&p->p_token);
        p->p_usched->heuristic_exiting(lp, p);

        /*
         * Issue late wakeups after releasing our token to give us a chance
         * to deschedule and switch away before another cpu in a wait*()
         * reaps us.  This is done as late as possible to reduce contention.
         */
        if (dowake)
                wakeup(&p->p_nthreads);
        if (waddr)
                wakeup(waddr);

        cpu_lwp_exit();
}

/*
 * Wait until a lwp is completely dead.  The final interlock in this drama
 * is when TDF_EXITING is set in cpu_thread_exit() just before the final
 * switchout.
 *
 * At the point TDF_EXITING is set a complete exit is accomplished when
 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear.  td_mpflags has two
 * post-switch interlock flags that can be used to wait for the TDF_
 * flags to clear.
 *
 * Returns non-zero on success, and zero if the caller needs to retry
 * the lwp_wait().
 */
static int
lwp_wait(struct lwp *lp)
{
        struct thread *td = lp->lwp_thread;
        u_int mpflags;

        KKASSERT(lwkt_preempted_proc() != lp);

        /*
         * This bit of code uses the thread destruction interlock
         * managed by lwkt_switch_return() to wait for the lwp's
         * thread to completely disengage.
         *
         * It is possible for us to race another cpu core so we
         * have to do this correctly.
         */
        for (;;) {
                mpflags = td->td_mpflags;
                cpu_ccfence();
                if (mpflags & TDF_MP_EXITSIG)
                        break;
                tsleep_interlock(td, 0);
                if (atomic_cmpset_int(&td->td_mpflags, mpflags,
                                      mpflags | TDF_MP_EXITWAIT)) {
                        tsleep(td, PINTERLOCKED, "lwpxt", 0);
                }
        }

        /*
         * We've already waited for the core exit but there can still
         * be other refs from e.g. process scans and such.
         */
        if (lp->lwp_lock > 0) {
                tsleep(lp, 0, "lwpwait1", 1);
                return(0);
        }
        if (td->td_refs) {
                tsleep(td, 0, "lwpwait2", 1);
                return(0);
        }

        /*
         * Now that we have the thread destruction interlock these flags
         * really should already be cleaned up, keep a check for safety.
         *
         * We can't rip its stack out from under it until TDF_EXITING is
         * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
         * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
         * will be cleared temporarily if a thread gets preempted.
         */
        while ((td->td_flags & (TDF_RUNNING |
                                TDF_RUNQ |
                                TDF_PREEMPT_LOCK |
                                TDF_EXITING)) != TDF_EXITING) {
                tsleep(lp, 0, "lwpwait3", 1);
                return (0);
        }

        KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0,
                ("lwp_wait: td %p (%s) still on run or sleep queue",
                td, td->td_comm));
        return (1);
}

/*
 * Release the resources associated with a lwp.
 * The lwp must be completely dead.
 */
void
lwp_dispose(struct lwp *lp)
{
        struct thread *td = lp->lwp_thread;

        KKASSERT(lwkt_preempted_proc() != lp);
        KKASSERT(lp->lwp_lock == 0);
        KKASSERT(td->td_refs == 0);
        KKASSERT((td->td_flags & (TDF_RUNNING |
                                  TDF_RUNQ |
                                  TDF_PREEMPT_LOCK |
                                  TDF_EXITING)) == TDF_EXITING);

        PRELE(lp->lwp_proc);
        lp->lwp_proc = NULL;
        if (td != NULL) {
                td->td_proc = NULL;
                td->td_lwp = NULL;
                lp->lwp_thread = NULL;
                lwkt_free_thread(td);
        }
        kfree(lp, M_LWP);
}

int
sys_wait4(struct wait_args *uap)
{
        struct rusage rusage;
        int error, status;

        error = kern_wait(uap->pid, (uap->status ? &status : NULL),
                          uap->options, (uap->rusage ? &rusage : NULL),
                          &uap->sysmsg_result);

        if (error == 0 && uap->status)
                error = copyout(&status, uap->status, sizeof(*uap->status));
        if (error == 0 && uap->rusage)
                error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
        return (error);
}

/*
 * wait1()
 *
 * wait_args(int pid, int *status, int options, struct rusage *rusage)
 */
int
kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
{
        struct thread *td = curthread;
        struct lwp *lp;
        struct proc *q = td->td_proc;
        struct proc *p, *t;
        struct ucred *cr;
        struct pargs *pa;
        struct sigacts *ps;
        int nfound, error;
        long waitgen;

        if (pid == 0)
                pid = -q->p_pgid;
        if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
                return (EINVAL);

        /*
         * Protect the q->p_children list
         */
        lwkt_gettoken(&q->p_token);
loop:
        /*
         * All sorts of things can change due to blocking so we have to loop
         * all the way back up here.
         *
         * The problem is that if a process group is stopped and the parent
         * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
         * of the child and then stop itself when it tries to return from the
         * system call.  When the process group is resumed the parent will
         * then get the STOP status even though the child has now resumed
         * (a followup wait*() will get the CONT status).
         *
         * Previously the CONT would overwrite the STOP because the tstop
         * was handled within tsleep(), and the parent would only see
         * the CONT when both are stopped and continued together.  This little
         * two-line hack restores this effect.
         */
        if (STOPLWP(q, td->td_lwp))
            tstop();

        nfound = 0;

        /*
         * Loop on children.
         *
         * NOTE: We don't want to break q's p_token in the loop for the
         *       case where no children are found or we risk breaking the
         *       interlock between child and parent.
         */
        waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
        LIST_FOREACH(p, &q->p_children, p_sibling) {
                if (pid != WAIT_ANY &&
                    p->p_pid != pid && p->p_pgid != -pid) {
                        continue;
                }

                /*
                 * This special case handles a kthread spawned by linux_clone
                 * (see linux_misc.c).  The linux_wait4 and linux_waitpid 
                 * functions need to be able to distinguish between waiting
                 * on a process and waiting on a thread.  It is a thread if
                 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
                 * signifies we want to wait for threads and not processes.
                 */
                if ((p->p_sigparent != SIGCHLD) ^ 
                    ((options & WLINUXCLONE) != 0)) {
                        continue;
                }

                nfound++;
                if (p->p_stat == SZOMB) {
                        /*
                         * We may go into SZOMB with threads still present.
                         * We must wait for them to exit before we can reap
                         * the master thread, otherwise we may race reaping
                         * non-master threads.
                         *
                         * Only this routine can remove a process from
                         * the zombie list and destroy it, use PACQUIREZOMB()
                         * to serialize us and loop if it blocks (interlocked
                         * by the parent's q->p_token).
                         *
                         * WARNING!  (p) can be invalid when PHOLDZOMB(p)
                         *           returns non-zero.  Be sure not to
                         *           mess with it.
                         */
                        if (PHOLDZOMB(p))
                                goto loop;
                        lwkt_gettoken(&p->p_token);
                        if (p->p_pptr != q) {
                                lwkt_reltoken(&p->p_token);
                                PRELEZOMB(p);
                                goto loop;
                        }
                        while (p->p_nthreads > 0) {
                                tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
                        }

                        /*
                         * Reap any LWPs left in p->p_lwps.  This is usually
                         * just the last LWP.  This must be done before
                         * we loop on p_lock since the lwps hold a ref on
                         * it as a vmspace interlock.
                         *
                         * Once that is accomplished p_nthreads had better
                         * be zero.
                         */
                        while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
                                /*
                                 * Make sure no one is using this lwp, before
                                 * it is removed from the tree.  If we didn't
                                 * wait it here, lwp tree iteration with
                                 * blocking operation would be broken.
                                 */
                                while (lp->lwp_lock > 0)
                                        tsleep(lp, 0, "zomblwp", 1);
                                lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
                                reaplwp(lp);
                        }
                        KKASSERT(p->p_nthreads == 0);

                        /*
                         * Don't do anything really bad until all references
                         * to the process go away.  This may include other
                         * LWPs which are still in the process of being
                         * reaped.  We can't just pull the rug out from under
                         * them because they may still be using the VM space.
                         *
                         * Certain kernel facilities such as /proc will also
                         * put a hold on the process for short periods of
                         * time.
                         */
                        PRELE(p);
                        PSTALL(p, "reap3", 0);

                        /* Take care of our return values. */
                        *res = p->p_pid;

                        if (status)
                                *status = p->p_xstat;
                        if (rusage)
                                *rusage = p->p_ru;

                        /*
                         * If we got the child via a ptrace 'attach',
                         * we need to give it back to the old parent.
                         */
                        if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
                                PHOLD(p);
                                p->p_oppid = 0;
                                proc_reparent(p, t);
                                ksignal(t, SIGCHLD);
                                wakeup((caddr_t)t);
                                error = 0;
                                PRELE(t);
                                lwkt_reltoken(&p->p_token);
                                PRELEZOMB(p);
                                goto done;
                        }

                        /*
                         * Unlink the proc from its process group so that
                         * the following operations won't lead to an
                         * inconsistent state for processes running down
                         * the zombie list.
                         */
                        proc_remove_zombie(p);
                        proc_userunmap(p);
                        lwkt_reltoken(&p->p_token);
                        leavepgrp(p);

                        p->p_xstat = 0;
                        ruadd(&q->p_cru, &p->p_ru);

                        /*
                         * Decrement the count of procs running with this uid.
                         */
                        chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);

                        /*
                         * Free up credentials.  p_spin is required to
                         * avoid races against allproc scans.
                         */
                        spin_lock(&p->p_spin);
                        cr = p->p_ucred;
                        p->p_ucred = NULL;
                        spin_unlock(&p->p_spin);
                        crfree(cr);

                        /*
                         * Remove unused arguments
                         */
                        pa = p->p_args;
                        p->p_args = NULL;
                        if (pa && refcount_release(&pa->ar_ref)) {
                                kfree(pa, M_PARGS);
                                pa = NULL;
                        }

                        ps = p->p_sigacts;
                        p->p_sigacts = NULL;
                        if (ps && refcount_release(&ps->ps_refcnt)) {
                                kfree(ps, M_SUBPROC);
                                ps = NULL;
                        }

                        /*
                         * Our exitingcount was incremented when the process
                         * became a zombie, now that the process has been
                         * removed from (almost) all lists we should be able
                         * to safely destroy its vmspace.  Wait for any current
                         * holders to go away (so the vmspace remains stable),
                         * then scrap it.
                         *
                         * NOTE: Releasing the parent process (q) p_token
                         *       across the vmspace_exitfree() call is
                         *       important here to reduce stalls on
                         *       interactions with (q) (such as
                         *       fork/exec/wait or 'ps').
                         */
                        PSTALL(p, "reap4", 0);
                        lwkt_reltoken(&q->p_token);
                        vmspace_exitfree(p);
                        lwkt_gettoken(&q->p_token);
                        PSTALL(p, "reap5", 0);

                        /*
                         * NOTE: We have to officially release ZOMB in order
                         *       to ensure that a racing thread in kern_wait()
                         *       which blocked on ZOMB is woken up.
                         */
                        PHOLD(p);
                        PRELEZOMB(p);
                        kfree(p->p_uidpcpu, M_SUBPROC);
                        kfree(p, M_PROC);
                        atomic_add_int(&nprocs, -1);
                        error = 0;
                        goto done;
                }
                if ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
                    (p->p_flags & P_WAITED) == 0 &&
                    ((p->p_flags & P_TRACED) || (options & WUNTRACED))) {
                        PHOLD(p);
                        lwkt_gettoken(&p->p_token);
                        if (p->p_pptr != q) {
                                lwkt_reltoken(&p->p_token);
                                PRELE(p);
                                goto loop;
                        }
                        if ((p->p_stat != SSTOP && p->p_stat != SCORE) ||
                            (p->p_flags & P_WAITED) != 0 ||
                            ((p->p_flags & P_TRACED) == 0 &&
                             (options & WUNTRACED) == 0)) {
                                lwkt_reltoken(&p->p_token);
                                PRELE(p);
                                goto loop;
                        }

                        p->p_flags |= P_WAITED;

                        *res = p->p_pid;
                        if (status)
                                *status = W_STOPCODE(p->p_xstat);
                        /* Zero rusage so we get something consistent. */
                        if (rusage)
                                bzero(rusage, sizeof(*rusage));
                        error = 0;
                        lwkt_reltoken(&p->p_token);
                        PRELE(p);
                        goto done;
                }
                if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) {
                        PHOLD(p);
                        lwkt_gettoken(&p->p_token);
                        if (p->p_pptr != q) {
                                lwkt_reltoken(&p->p_token);
                                PRELE(p);
                                goto loop;
                        }
                        if ((p->p_flags & P_CONTINUED) == 0) {
                                lwkt_reltoken(&p->p_token);
                                PRELE(p);
                                goto loop;
                        }

                        *res = p->p_pid;
                        p->p_flags &= ~P_CONTINUED;

                        if (status)
                                *status = SIGCONT;
                        error = 0;
                        lwkt_reltoken(&p->p_token);
                        PRELE(p);
                        goto done;
                }
        }
        if (nfound == 0) {
                error = ECHILD;
                goto done;
        }
        if (options & WNOHANG) {
                *res = 0;
                error = 0;
                goto done;
        }

        /*
         * Wait for signal - interlocked using q->p_waitgen.
         */
        error = 0;
        while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
                tsleep_interlock(q, PCATCH);
                waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
                if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
                        error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0);
                        break;
                }
        }
        if (error) {
done:
                lwkt_reltoken(&q->p_token);
                return (error);
        }
        goto loop;
}

/*
 * Change child's parent process to parent.
 *
 * p_children/p_sibling requires the parent's token, and
 * changing pptr requires the child's token, so we have to
 * get three tokens to do this operation.  We also need to
 * hold pointers that might get ripped out from under us to
 * preserve structural integrity.
 *
 * It is possible to race another reparent or disconnect or other
 * similar operation.  We must retry when this situation occurs.
 * Once we successfully reparent the process we no longer care
 * about any races.
 */
void
proc_reparent(struct proc *child, struct proc *parent)
{
        struct proc *opp;

        PHOLD(parent);
        while ((opp = child->p_pptr) != parent) {
                PHOLD(opp);
                lwkt_gettoken(&opp->p_token);
                lwkt_gettoken(&child->p_token);
                lwkt_gettoken(&parent->p_token);
                if (child->p_pptr != opp) {
                        lwkt_reltoken(&parent->p_token);
                        lwkt_reltoken(&child->p_token);
                        lwkt_reltoken(&opp->p_token);
                        PRELE(opp);
                        continue;
                }
                LIST_REMOVE(child, p_sibling);
                LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
                child->p_pptr = parent;
                child->p_ppid = parent->p_pid;
                lwkt_reltoken(&parent->p_token);
                lwkt_reltoken(&child->p_token);
                lwkt_reltoken(&opp->p_token);
                if (LIST_EMPTY(&opp->p_children))
                        wakeup(opp);
                PRELE(opp);
                break;
        }
        PRELE(parent);
}

/*
 * The next two functions are to handle adding/deleting items on the
 * exit callout list
 * 
 * at_exit():
 * Take the arguments given and put them onto the exit callout list,
 * However first make sure that it's not already there.
 * returns 0 on success.
 */

int
at_exit(exitlist_fn function)
{
        struct exitlist *ep;

#ifdef INVARIANTS
        /* Be noisy if the programmer has lost track of things */
        if (rm_at_exit(function)) 
                kprintf("WARNING: exit callout entry (%p) already present\n",
                    function);
#endif
        ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
        if (ep == NULL)
                return (ENOMEM);
        ep->function = function;
        TAILQ_INSERT_TAIL(&exit_list, ep, next);
        return (0);
}

/*
 * Scan the exit callout list for the given item and remove it.
 * Returns the number of items removed (0 or 1)
 */
int
rm_at_exit(exitlist_fn function)
{
        struct exitlist *ep;

        TAILQ_FOREACH(ep, &exit_list, next) {
                if (ep->function == function) {
                        TAILQ_REMOVE(&exit_list, ep, next);
                        kfree(ep, M_ATEXIT);
                        return(1);
                }
        }       
        return (0);
}

/*
 * LWP reaper related code.
 */
static void
reaplwps(void *context, int dummy)
{
        struct lwplist *lwplist = context;
        struct lwp *lp;
        int cpu = mycpuid;

        lwkt_gettoken(&deadlwp_token[cpu]);
        while ((lp = LIST_FIRST(lwplist))) {
                LIST_REMOVE(lp, u.lwp_reap_entry);
                reaplwp(lp);
        }
        lwkt_reltoken(&deadlwp_token[cpu]);
}

static void
reaplwp(struct lwp *lp)
{
        while (lwp_wait(lp) == 0)
                ;
        lwp_dispose(lp);
}

static void
deadlwp_init(void)
{
        int cpu;

        for (cpu = 0; cpu < ncpus; cpu++) {
                lwkt_token_init(&deadlwp_token[cpu], "deadlwpl");
                LIST_INIT(&deadlwp_list[cpu]);
                deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]),
                                            M_DEVBUF, M_WAITOK);
                TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
        }
}

SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);