kernel - Automatically downscasle NPROC resource limit

[dragonfly.git] / sys / kern / kern_fork.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_fork.c 8.6 (Berkeley) 4/8/94
 * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $
 */

#include "opt_ktrace.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
#include <sys/filedesc.h>
#include <sys/kernel.h>
#include <sys/sysctl.h>
#include <sys/malloc.h>
#include <sys/proc.h>
#include <sys/resourcevar.h>
#include <sys/vnode.h>
#include <sys/acct.h>
#include <sys/ktrace.h>
#include <sys/unistd.h>
#include <sys/jail.h>
#include <sys/lwp.h>

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

#include <sys/vmmeter.h>
#include <sys/refcount.h>
#include <sys/thread2.h>
#include <sys/signal2.h>
#include <sys/spinlock2.h>

#include <sys/dsched.h>

static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
static MALLOC_DEFINE(M_REAPER, "reaper", "process reapers");

/*
 * These are the stuctures used to create a callout list for things to do
 * when forking a process
 */
struct forklist {
        forklist_fn function;
        TAILQ_ENTRY(forklist) next;
};

TAILQ_HEAD(forklist_head, forklist);
static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);

static struct lwp       *lwp_fork(struct lwp *, struct proc *, int flags,
                            const cpumask_t *mask);
static int              lwp_create1(struct lwp_params *params,
                            const cpumask_t *mask);

int forksleep; /* Place for fork1() to sleep on. */

/*
 * Red-Black tree support for LWPs
 */

static int
rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
{
        if (lp1->lwp_tid < lp2->lwp_tid)
                return(-1);
        if (lp1->lwp_tid > lp2->lwp_tid)
                return(1);
        return(0);
}

RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);

/*
 * fork() system call
 */
int
sys_fork(struct fork_args *uap)
{
        struct lwp *lp = curthread->td_lwp;
        struct proc *p2;
        int error;

        error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
        if (error == 0) {
                PHOLD(p2);
                start_forked_proc(lp, p2);
                uap->sysmsg_fds[0] = p2->p_pid;
                uap->sysmsg_fds[1] = 0;
                PRELE(p2);
        }
        return error;
}

/*
 * vfork() system call
 */
int
sys_vfork(struct vfork_args *uap)
{
        struct lwp *lp = curthread->td_lwp;
        struct proc *p2;
        int error;

        error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
        if (error == 0) {
                PHOLD(p2);
                start_forked_proc(lp, p2);
                uap->sysmsg_fds[0] = p2->p_pid;
                uap->sysmsg_fds[1] = 0;
                PRELE(p2);
        }
        return error;
}

/*
 * Handle rforks.  An rfork may (1) operate on the current process without
 * creating a new, (2) create a new process that shared the current process's
 * vmspace, signals, and/or descriptors, or (3) create a new process that does
 * not share these things (normal fork).
 *
 * Note that we only call start_forked_proc() if a new process is actually
 * created.
 *
 * rfork { int flags }
 */
int
sys_rfork(struct rfork_args *uap)
{
        struct lwp *lp = curthread->td_lwp;
        struct proc *p2;
        int error;

        if ((uap->flags & RFKERNELONLY) != 0)
                return (EINVAL);

        error = fork1(lp, uap->flags | RFPGLOCK, &p2);
        if (error == 0) {
                if (p2) {
                        PHOLD(p2);
                        start_forked_proc(lp, p2);
                        uap->sysmsg_fds[0] = p2->p_pid;
                        uap->sysmsg_fds[1] = 0;
                        PRELE(p2);
                } else {
                        uap->sysmsg_fds[0] = 0;
                        uap->sysmsg_fds[1] = 0;
                }
        }
        return error;
}

static int
lwp_create1(struct lwp_params *uprm, const cpumask_t *umask)
{
        struct proc *p = curproc;
        struct lwp *lp;
        struct lwp_params params;
        cpumask_t *mask = NULL, mask0;
        int error;

        error = copyin(uprm, &params, sizeof(params));
        if (error)
                goto fail2;

        if (umask != NULL) {
                error = copyin(umask, &mask0, sizeof(mask0));
                if (error)
                        goto fail2;
                CPUMASK_ANDMASK(mask0, smp_active_mask);
                if (CPUMASK_TESTNZERO(mask0))
                        mask = &mask0;
        }

        lwkt_gettoken(&p->p_token);
        plimit_lwp_fork(p);     /* force exclusive access */
        lp = lwp_fork(curthread->td_lwp, p, RFPROC | RFMEM, mask);
        error = cpu_prepare_lwp(lp, &params);
        if (error)
                goto fail;
        if (params.lwp_tid1 != NULL &&
            (error = copyout(&lp->lwp_tid, params.lwp_tid1, sizeof(lp->lwp_tid))))
                goto fail;
        if (params.lwp_tid2 != NULL &&
            (error = copyout(&lp->lwp_tid, params.lwp_tid2, sizeof(lp->lwp_tid))))
                goto fail;

        /*
         * Now schedule the new lwp. 
         */
        p->p_usched->resetpriority(lp);
        crit_enter();
        lp->lwp_stat = LSRUN;
        p->p_usched->setrunqueue(lp);
        crit_exit();
        lwkt_reltoken(&p->p_token);

        return (0);

fail:
        /*
         * 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, "lwpfail", 1);
        lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
        --p->p_nthreads;
        /* lwp_dispose expects an exited lwp, and a held proc */
        atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
        lp->lwp_thread->td_flags |= TDF_EXITING;
        lwkt_remove_tdallq(lp->lwp_thread);
        PHOLD(p);
        biosched_done(lp->lwp_thread);
        dsched_exit_thread(lp->lwp_thread);
        lwp_dispose(lp);
        lwkt_reltoken(&p->p_token);
fail2:
        return (error);
}

/*
 * Low level thread create used by pthreads.
 */
int
sys_lwp_create(struct lwp_create_args *uap)
{

        return (lwp_create1(uap->params, NULL));
}

int
sys_lwp_create2(struct lwp_create2_args *uap)
{

        return (lwp_create1(uap->params, uap->mask));
}

int     nprocs = 1;             /* process 0 */

int
fork1(struct lwp *lp1, int flags, struct proc **procp)
{
        struct proc *p1 = lp1->lwp_proc;
        struct proc *p2;
        struct proc *pptr;
        struct pgrp *p1grp;
        struct pgrp *plkgrp;
        struct sysreaper *reap;
        uid_t uid;
        int ok, error;
        static int curfail = 0;
        static struct timeval lastfail;
        struct forklist *ep;
        struct filedesc_to_leader *fdtol;

        if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
                return (EINVAL);

        lwkt_gettoken(&p1->p_token);
        plkgrp = NULL;
        p2 = NULL;

        /*
         * Here we don't create a new process, but we divorce
         * certain parts of a process from itself.
         */
        if ((flags & RFPROC) == 0) {
                /*
                 * This kind of stunt does not work anymore if
                 * there are native threads (lwps) running
                 */
                if (p1->p_nthreads != 1) {
                        error = EINVAL;
                        goto done;
                }

                vm_fork(p1, 0, flags);

                /*
                 * Close all file descriptors.
                 */
                if (flags & RFCFDG) {
                        struct filedesc *fdtmp;
                        fdtmp = fdinit(p1);
                        fdfree(p1, fdtmp);
                }

                /*
                 * Unshare file descriptors (from parent.)
                 */
                if (flags & RFFDG) {
                        if (p1->p_fd->fd_refcnt > 1) {
                                struct filedesc *newfd;
                                error = fdcopy(p1, &newfd);
                                if (error != 0) {
                                        error = ENOMEM;
                                        goto done;
                                }
                                fdfree(p1, newfd);
                        }
                }
                *procp = NULL;
                error = 0;
                goto done;
        }

        /*
         * Interlock against process group signal delivery.  If signals
         * are pending after the interlock is obtained we have to restart
         * the system call to process the signals.  If we don't the child
         * can miss a pgsignal (such as ^C) sent during the fork.
         *
         * We can't use CURSIG() here because it will process any STOPs
         * and cause the process group lock to be held indefinitely.  If
         * a STOP occurs, the fork will be restarted after the CONT.
         */
        p1grp = p1->p_pgrp;
        if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
                pgref(plkgrp);
                lockmgr(&plkgrp->pg_lock, LK_SHARED);
                if (CURSIG_NOBLOCK(lp1)) {
                        error = ERESTART;
                        goto done;
                }
        }

        /*
         * Although process entries are dynamically created, we still keep
         * a global limit on the maximum number we will create.  Don't allow
         * a nonprivileged user to use the last ten processes; don't let root
         * exceed the limit. The variable nprocs is the current number of
         * processes, maxproc is the limit.
         */
        uid = lp1->lwp_thread->td_ucred->cr_ruid;
        if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
                if (ppsratecheck(&lastfail, &curfail, 1))
                        kprintf("maxproc limit exceeded by uid %d, please "
                               "see tuning(7) and login.conf(5).\n", uid);
                tsleep(&forksleep, 0, "fork", hz / 2);
                error = EAGAIN;
                goto done;
        }

        /*
         * Increment the nprocs resource before blocking can occur.  There
         * are hard-limits as to the number of processes that can run.
         */
        atomic_add_int(&nprocs, 1);

        /*
         * Increment the count of procs running with this uid.  This also
         * applies to root.
         */
        ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
                        plimit_getadjvalue(RLIMIT_NPROC));
        if (!ok) {
                /*
                 * Back out the process count
                 */
                atomic_add_int(&nprocs, -1);
                if (ppsratecheck(&lastfail, &curfail, 1)) {
                        kprintf("maxproc limit of %jd "
                                "exceeded by \"%s\" uid %d, "
                                "please see tuning(7) and login.conf(5).\n",
                                plimit_getadjvalue(RLIMIT_NPROC),
                                p1->p_comm,
                                uid);
                }
                tsleep(&forksleep, 0, "fork", hz / 2);
                error = EAGAIN;
                goto done;
        }

        /*
         * Allocate a new process, don't get fancy: zero the structure.
         */
        p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);

        /*
         * Core initialization.  SIDL is a safety state that protects the
         * partially initialized process once it starts getting hooked
         * into system structures and becomes addressable.
         *
         * We must be sure to acquire p2->p_token as well, we must hold it
         * once the process is on the allproc list to avoid things such
         * as competing modifications to p_flags.
         */
        mycpu->gd_forkid += ncpus;
        p2->p_forkid = mycpu->gd_forkid + mycpu->gd_cpuid;
        p2->p_lasttid = -1;     /* first tid will be 0 */
        p2->p_stat = SIDL;

        /*
         * NOTE: Process 0 will not have a reaper, but process 1 (init) and
         *       all other processes always will.
         */
        if ((reap = p1->p_reaper) != NULL) {
                reaper_hold(reap);
                p2->p_reaper = reap;
        } else {
                p2->p_reaper = NULL;
        }

        RB_INIT(&p2->p_lwp_tree);
        spin_init(&p2->p_spin, "procfork1");
        lwkt_token_init(&p2->p_token, "proc");
        lwkt_gettoken(&p2->p_token);

        /*
         * Setup linkage for kernel based threading XXX lwp.  Also add the
         * process to the allproclist.
         *
         * The process structure is addressable after this point.
         */
        if (flags & RFTHREAD) {
                p2->p_peers = p1->p_peers;
                p1->p_peers = p2;
                p2->p_leader = p1->p_leader;
        } else {
                p2->p_leader = p2;
        }
        proc_add_allproc(p2);

        /*
         * Initialize the section which is copied verbatim from the parent.
         */
        bcopy(&p1->p_startcopy, &p2->p_startcopy,
              ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));

        /*
         * Duplicate sub-structures as needed.  Increase reference counts
         * on shared objects.
         *
         * NOTE: because we are now on the allproc list it is possible for
         *       other consumers to gain temporary references to p2
         *       (p2->p_lock can change).
         */
        if (p1->p_flags & P_PROFIL)
                startprofclock(p2);
        p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);

        if (jailed(p2->p_ucred))
                p2->p_flags |= P_JAILED;

        if (p2->p_args)
                refcount_acquire(&p2->p_args->ar_ref);

        p2->p_usched = p1->p_usched;
        /* XXX: verify copy of the secondary iosched stuff */
        dsched_enter_proc(p2);

        if (flags & RFSIGSHARE) {
                p2->p_sigacts = p1->p_sigacts;
                refcount_acquire(&p2->p_sigacts->ps_refcnt);
        } else {
                p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
                                        M_SUBPROC, M_WAITOK);
                bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
                refcount_init(&p2->p_sigacts->ps_refcnt, 1);
        }
        if (flags & RFLINUXTHPN) 
                p2->p_sigparent = SIGUSR1;
        else
                p2->p_sigparent = SIGCHLD;

        /* bump references to the text vnode (for procfs) */
        p2->p_textvp = p1->p_textvp;
        if (p2->p_textvp)
                vref(p2->p_textvp);

        /* copy namecache handle to the text file */
        if (p1->p_textnch.mount)
                cache_copy(&p1->p_textnch, &p2->p_textnch);

        /*
         * Handle file descriptors
         */
        if (flags & RFCFDG) {
                p2->p_fd = fdinit(p1);
                fdtol = NULL;
        } else if (flags & RFFDG) {
                error = fdcopy(p1, &p2->p_fd);
                if (error != 0) {
                        error = ENOMEM;
                        goto done;
                }
                fdtol = NULL;
        } else {
                p2->p_fd = fdshare(p1);
                if (p1->p_fdtol == NULL) {
                        p1->p_fdtol = filedesc_to_leader_alloc(NULL,
                                                               p1->p_leader);
                }
                if ((flags & RFTHREAD) != 0) {
                        /*
                         * Shared file descriptor table and
                         * shared process leaders.
                         */
                        fdtol = p1->p_fdtol;
                        fdtol->fdl_refcount++;
                } else {
                        /* 
                         * Shared file descriptor table, and
                         * different process leaders 
                         */
                        fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
                }
        }
        p2->p_fdtol = fdtol;
        p2->p_limit = plimit_fork(p1);

        /*
         * Adjust depth for resource downscaling
         */
        if ((p2->p_depth & 31) != 31)
                ++p2->p_depth;

        /*
         * Preserve some more flags in subprocess.  P_PROFIL has already
         * been preserved.
         */
        p2->p_flags |= p1->p_flags & P_SUGID;
        if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT))
                p2->p_flags |= P_CONTROLT;
        if (flags & RFPPWAIT) {
                p2->p_flags |= P_PPWAIT;
                if (p1->p_upmap)
                        atomic_add_int(&p1->p_upmap->invfork, 1);
        }

        /*
         * Inherit the virtual kernel structure (allows a virtual kernel
         * to fork to simulate multiple cpus).
         */
        if (p1->p_vkernel)
                vkernel_inherit(p1, p2);

        /*
         * Once we are on a pglist we may receive signals.  XXX we might
         * race a ^C being sent to the process group by not receiving it
         * at all prior to this line.
         */
        pgref(p1grp);
        lwkt_gettoken(&p1grp->pg_token);
        LIST_INSERT_AFTER(p1, p2, p_pglist);
        lwkt_reltoken(&p1grp->pg_token);

        /*
         * Attach the new process to its parent.
         *
         * If RFNOWAIT is set, the newly created process becomes a child
         * of the reaper (typically init).  This effectively disassociates
         * the child from the parent.
         *
         * Temporarily hold pptr for the RFNOWAIT case to avoid ripouts.
         */
        if (flags & RFNOWAIT) {
                pptr = reaper_get(reap);
                if (pptr == NULL) {
                        pptr = initproc;
                        PHOLD(pptr);
                }
        } else {
                pptr = p1;
        }
        p2->p_pptr = pptr;
        LIST_INIT(&p2->p_children);

        lwkt_gettoken(&pptr->p_token);
        LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
        lwkt_reltoken(&pptr->p_token);

        if (flags & RFNOWAIT)
                PRELE(pptr);

        varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
        callout_init_mp(&p2->p_ithandle);

#ifdef KTRACE
        /*
         * Copy traceflag and tracefile if enabled.  If not inherited,
         * these were zeroed above but we still could have a trace race
         * so make sure p2's p_tracenode is NULL.
         */
        if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
                p2->p_traceflag = p1->p_traceflag;
                p2->p_tracenode = ktrinherit(p1->p_tracenode);
        }
#endif

        /*
         * This begins the section where we must prevent the parent
         * from being swapped.
         *
         * Gets PRELE'd in the caller in start_forked_proc().
         */
        PHOLD(p1);

        vm_fork(p1, p2, flags);

        /*
         * Create the first lwp associated with the new proc.
         * It will return via a different execution path later, directly
         * into userland, after it was put on the runq by
         * start_forked_proc().
         */
        lwp_fork(lp1, p2, flags, NULL);

        if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
                mycpu->gd_cnt.v_forks++;
                mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize +
                                             p2->p_vmspace->vm_ssize;
        } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
                mycpu->gd_cnt.v_vforks++;
                mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
                                              p2->p_vmspace->vm_ssize;
        } else if (p1 == &proc0) {
                mycpu->gd_cnt.v_kthreads++;
                mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
                                                p2->p_vmspace->vm_ssize;
        } else {
                mycpu->gd_cnt.v_rforks++;
                mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
                                              p2->p_vmspace->vm_ssize;
        }

        /*
         * Both processes are set up, now check if any loadable modules want
         * to adjust anything.
         *   What if they have an error? XXX
         */
        TAILQ_FOREACH(ep, &fork_list, next) {
                (*ep->function)(p1, p2, flags);
        }

        /*
         * Set the start time.  Note that the process is not runnable.  The
         * caller is responsible for making it runnable.
         */
        microtime(&p2->p_start);
        p2->p_acflag = AFORK;

        /*
         * tell any interested parties about the new process
         */
        KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);

        /*
         * Return child proc pointer to parent.
         */
        *procp = p2;
        error = 0;
done:
        if (p2)
                lwkt_reltoken(&p2->p_token);
        lwkt_reltoken(&p1->p_token);
        if (plkgrp) {
                lockmgr(&plkgrp->pg_lock, LK_RELEASE);
                pgrel(plkgrp);
        }
        return (error);
}

static struct lwp *
lwp_fork(struct lwp *origlp, struct proc *destproc, int flags,
    const cpumask_t *mask)
{
        globaldata_t gd = mycpu;
        struct lwp *lp;
        struct thread *td;

        lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);

        lp->lwp_proc = destproc;
        lp->lwp_vmspace = destproc->p_vmspace;
        lp->lwp_stat = LSRUN;
        bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
            (unsigned) ((caddr_t)&lp->lwp_endcopy -
                        (caddr_t)&lp->lwp_startcopy));
        if (mask != NULL)
                lp->lwp_cpumask = *mask;

        /*
         * Reset the sigaltstack if memory is shared, otherwise inherit
         * it.
         */
        if (flags & RFMEM) {
                lp->lwp_sigstk.ss_flags = SS_DISABLE;
                lp->lwp_sigstk.ss_size = 0;
                lp->lwp_sigstk.ss_sp = NULL;
                lp->lwp_flags &= ~LWP_ALTSTACK;
        } else {
                lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK;
        }

        /*
         * Set cpbase to the last timeout that occured (not the upcoming
         * timeout).
         *
         * A critical section is required since a timer IPI can update
         * scheduler specific data.
         */
        crit_enter();
        lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic;
        destproc->p_usched->heuristic_forking(origlp, lp);
        crit_exit();
        CPUMASK_ANDMASK(lp->lwp_cpumask, usched_mastermask);
        lwkt_token_init(&lp->lwp_token, "lwp_token");
        spin_init(&lp->lwp_spin, "lwptoken");

        /*
         * Assign the thread to the current cpu to begin with so we
         * can manipulate it.
         */
        td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0);
        lp->lwp_thread = td;
        td->td_ucred = crhold(destproc->p_ucred);
        td->td_proc = destproc;
        td->td_lwp = lp;
        td->td_switch = cpu_heavy_switch;
#ifdef NO_LWKT_SPLIT_USERPRI
        lwkt_setpri(td, TDPRI_USER_NORM);
#else
        lwkt_setpri(td, TDPRI_KERN_USER);
#endif
        lwkt_set_comm(td, "%s", destproc->p_comm);

        /*
         * cpu_fork will copy and update the pcb, set up the kernel stack,
         * and make the child ready to run.
         */
        cpu_fork(origlp, lp, flags);
        kqueue_init(&lp->lwp_kqueue, destproc->p_fd);

        /*
         * Assign a TID to the lp.  Loop until the insert succeeds (returns
         * NULL).
         *
         * If we are in a vfork assign the same TID as the lwp that did the
         * vfork().  This way if the user program messes around with
         * pthread calls inside the vfork(), it will operate like an
         * extension of the (blocked) parent.  Also note that since the
         * address space is being shared, insofar as pthreads is concerned,
         * the code running in the vfork() is part of the original process.
         */
        if (flags & RFPPWAIT) {
                lp->lwp_tid = origlp->lwp_tid - 1;
        } else {
                lp->lwp_tid = destproc->p_lasttid;
        }

        do {
                if (++lp->lwp_tid < 0)
                        lp->lwp_tid = 1;
        } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);

        destproc->p_lasttid = lp->lwp_tid;
        destproc->p_nthreads++;

        /*
         * This flag is set and never cleared.  It means that the process
         * was threaded at some point.  Used to improve exit performance.
         */
        destproc->p_flags |= P_MAYBETHREADED;

        return (lp);
}

/*
 * The next two functionms are general routines to handle adding/deleting
 * items on the fork callout list.
 *
 * at_fork():
 * Take the arguments given and put them onto the fork callout list,
 * However first make sure that it's not already there.
 * Returns 0 on success or a standard error number.
 */
int
at_fork(forklist_fn function)
{
        struct forklist *ep;

#ifdef INVARIANTS
        /* let the programmer know if he's been stupid */
        if (rm_at_fork(function)) {
                kprintf("WARNING: fork callout entry (%p) already present\n",
                    function);
        }
#endif
        ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
        ep->function = function;
        TAILQ_INSERT_TAIL(&fork_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_fork(forklist_fn function)
{
        struct forklist *ep;

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

/*
 * Add a forked process to the run queue after any remaining setup, such
 * as setting the fork handler, has been completed.
 *
 * p2 is held by the caller.
 */
void
start_forked_proc(struct lwp *lp1, struct proc *p2)
{
        struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
        int pflags;

        /*
         * Move from SIDL to RUN queue, and activate the process's thread.
         * Activation of the thread effectively makes the process "a"
         * current process, so we do not setrunqueue().
         *
         * YYY setrunqueue works here but we should clean up the trampoline
         * code so we just schedule the LWKT thread and let the trampoline
         * deal with the userland scheduler on return to userland.
         */
        KASSERT(p2->p_stat == SIDL,
            ("cannot start forked process, bad status: %p", p2));
        p2->p_usched->resetpriority(lp2);
        crit_enter();
        p2->p_stat = SACTIVE;
        lp2->lwp_stat = LSRUN;
        p2->p_usched->setrunqueue(lp2);
        crit_exit();

        /*
         * Now can be swapped.
         */
        PRELE(lp1->lwp_proc);

        /*
         * Preserve synchronization semantics of vfork.  P_PPWAIT is set in
         * the child until it has retired the parent's resources.  The parent
         * must wait for the flag to be cleared by the child.
         *
         * Interlock the flag/tsleep with atomic ops to avoid unnecessary
         * p_token conflicts.
         *
         * XXX Is this use of an atomic op on a field that is not normally
         *     manipulated with atomic ops ok?
         */
        while ((pflags = p2->p_flags) & P_PPWAIT) {
                cpu_ccfence();
                tsleep_interlock(lp1->lwp_proc, 0);
                if (atomic_cmpset_int(&p2->p_flags, pflags, pflags))
                        tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0);
        }
}

/*
 * procctl (idtype_t idtype, id_t id, int cmd, void *arg)
 */
int
sys_procctl(struct procctl_args *uap)
{
        struct proc *p = curproc;
        struct proc *p2;
        struct sysreaper *reap;
        union reaper_info udata;
        int error;

        if (uap->idtype != P_PID || uap->id != (id_t)p->p_pid)
                return EINVAL;

        switch(uap->cmd) {
        case PROC_REAP_ACQUIRE:
                lwkt_gettoken(&p->p_token);
                reap = kmalloc(sizeof(*reap), M_REAPER, M_WAITOK|M_ZERO);
                if (p->p_reaper == NULL || p->p_reaper->p != p) {
                        reaper_init(p, reap);
                        error = 0;
                } else {
                        kfree(reap, M_REAPER);
                        error = EALREADY;
                }
                lwkt_reltoken(&p->p_token);
                break;
        case PROC_REAP_RELEASE:
                lwkt_gettoken(&p->p_token);
release_again:
                reap = p->p_reaper;
                KKASSERT(reap != NULL);
                if (reap->p == p) {
                        reaper_hold(reap);      /* in case of thread race */
                        lockmgr(&reap->lock, LK_EXCLUSIVE);
                        if (reap->p != p) {
                                lockmgr(&reap->lock, LK_RELEASE);
                                reaper_drop(reap);
                                goto release_again;
                        }
                        reap->p = NULL;
                        p->p_reaper = reap->parent;
                        if (p->p_reaper)
                                reaper_hold(p->p_reaper);
                        lockmgr(&reap->lock, LK_RELEASE);
                        reaper_drop(reap);      /* our ref */
                        reaper_drop(reap);      /* old p_reaper ref */
                        error = 0;
                } else {
                        error = ENOTCONN;
                }
                lwkt_reltoken(&p->p_token);
                break;
        case PROC_REAP_STATUS:
                bzero(&udata, sizeof(udata));
                lwkt_gettoken_shared(&p->p_token);
                if ((reap = p->p_reaper) != NULL && reap->p == p) {
                        udata.status.flags = reap->flags;
                        udata.status.refs = reap->refs - 1; /* minus ours */
                }
                p2 = LIST_FIRST(&p->p_children);
                udata.status.pid_head = p2 ? p2->p_pid : -1;
                lwkt_reltoken(&p->p_token);

                if (uap->data) {
                        error = copyout(&udata, uap->data,
                                        sizeof(udata.status));
                } else {
                        error = 0;
                }
                break;
        default:
                error = EINVAL;
                break;
        }
        return error;
}

/*
 * Bump ref on reaper, preventing destruction
 */
void
reaper_hold(struct sysreaper *reap)
{
        KKASSERT(reap->refs > 0);
        refcount_acquire(&reap->refs);
}

/*
 * Drop ref on reaper, destroy the structure on the 1->0
 * transition and loop on the parent.
 */
void
reaper_drop(struct sysreaper *next)
{
        struct sysreaper *reap;

        while ((reap = next) != NULL) {
                if (refcount_release(&reap->refs)) {
                        next = reap->parent;
                        KKASSERT(reap->p == NULL);
                        reap->parent = NULL;
                        kfree(reap, M_REAPER);
                } else {
                        next = NULL;
                }
        }
}

/*
 * Initialize a static or newly allocated reaper structure
 */
void
reaper_init(struct proc *p, struct sysreaper *reap)
{
        reap->parent = p->p_reaper;
        reap->p = p;
        if (p == initproc) {
                reap->flags = REAPER_STAT_OWNED | REAPER_STAT_REALINIT;
                reap->refs = 2;
        } else {
                reap->flags = REAPER_STAT_OWNED;
                reap->refs = 1;
        }
        lockinit(&reap->lock, "subrp", 0, 0);
        cpu_sfence();
        p->p_reaper = reap;
}

/*
 * Called with p->p_token held during exit.
 *
 * This is a bit simpler than RELEASE because there are no threads remaining
 * to race.  We only release if we own the reaper, the exit code will handle
 * the final p_reaper release.
 */
struct sysreaper *
reaper_exit(struct proc *p)
{
        struct sysreaper *reap;

        /*
         * Release acquired reaper
         */
        if ((reap = p->p_reaper) != NULL && reap->p == p) {
                lockmgr(&reap->lock, LK_EXCLUSIVE);
                p->p_reaper = reap->parent;
                if (p->p_reaper)
                        reaper_hold(p->p_reaper);
                reap->p = NULL;
                lockmgr(&reap->lock, LK_RELEASE);
                reaper_drop(reap);
        }

        /*
         * Return and clear reaper (caller is holding p_token for us)
         * (reap->p does not equal p).  Caller must drop it.
         */
        if ((reap = p->p_reaper) != NULL) {
                p->p_reaper = NULL;
        }
        return reap;
}

/*
 * Return a held (PHOLD) process representing the reaper for process (p).
 * NULL should not normally be returned.  Caller should PRELE() the returned
 * reaper process when finished.
 *
 * Remove dead internal nodes while we are at it.
 *
 * Process (p)'s token must be held on call.
 * The returned process's token is NOT acquired by this routine.
 */
struct proc *
reaper_get(struct sysreaper *reap)
{
        struct sysreaper *next;
        struct proc *reproc;

        if (reap == NULL)
                return NULL;

        /*
         * Extra hold for loop
         */
        reaper_hold(reap);

        while (reap) {
                lockmgr(&reap->lock, LK_SHARED);
                if (reap->p) {
                        /*
                         * Probable reaper
                         */
                        if (reap->p) {
                                reproc = reap->p;
                                PHOLD(reproc);
                                lockmgr(&reap->lock, LK_RELEASE);
                                reaper_drop(reap);
                                return reproc;
                        }

                        /*
                         * Raced, try again
                         */
                        lockmgr(&reap->lock, LK_RELEASE);
                        continue;
                }

                /*
                 * Traverse upwards in the reaper topology, destroy
                 * dead internal nodes when possible.
                 *
                 * NOTE: Our ref on next means that a dead node should
                 *       have 2 (ours and reap->parent's).
                 */
                next = reap->parent;
                while (next) {
                        reaper_hold(next);
                        if (next->refs == 2 && next->p == NULL) {
                                lockmgr(&reap->lock, LK_RELEASE);
                                lockmgr(&reap->lock, LK_EXCLUSIVE);
                                if (next->refs == 2 &&
                                    reap->parent == next &&
                                    next->p == NULL) {
                                        /*
                                         * reap->parent inherits ref from next.
                                         */
                                        reap->parent = next->parent;
                                        next->parent = NULL;
                                        reaper_drop(next);      /* ours */
                                        reaper_drop(next);      /* old parent */
                                        next = reap->parent;
                                        continue;       /* possible chain */
                                }
                        }
                        break;
                }
                lockmgr(&reap->lock, LK_RELEASE);
                reaper_drop(reap);
                reap = next;
        }
        return NULL;
}