xref: /dflybsd-src/sys/kern/kern_fork.c (revision 3df4d8b028117b78bad9315141e641a022a7a967)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
39  * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $
40  * $DragonFly: src/sys/kern/kern_fork.c,v 1.77 2008/05/18 20:02:02 nth Exp $
41  */
42 
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/proc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/vnode.h>
55 #include <sys/acct.h>
56 #include <sys/ktrace.h>
57 #include <sys/unistd.h>
58 #include <sys/jail.h>
59 #include <sys/caps.h>
60 
61 #include <vm/vm.h>
62 #include <sys/lock.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_extern.h>
66 
67 #include <sys/vmmeter.h>
68 #include <sys/thread2.h>
69 #include <sys/signal2.h>
70 #include <sys/spinlock2.h>
71 #include <sys/mplock2.h>
72 
73 #include <sys/dsched.h>
74 
75 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
76 
77 /*
78  * These are the stuctures used to create a callout list for things to do
79  * when forking a process
80  */
81 struct forklist {
82 	forklist_fn function;
83 	TAILQ_ENTRY(forklist) next;
84 };
85 
86 TAILQ_HEAD(forklist_head, forklist);
87 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
88 
89 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
90 
91 int forksleep; /* Place for fork1() to sleep on. */
92 
93 /*
94  * Red-Black tree support for LWPs
95  */
96 
97 static int
98 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
99 {
100 	if (lp1->lwp_tid < lp2->lwp_tid)
101 		return(-1);
102 	if (lp1->lwp_tid > lp2->lwp_tid)
103 		return(1);
104 	return(0);
105 }
106 
107 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
108 
109 /*
110  * Fork system call
111  *
112  * MPALMOSTSAFE
113  */
114 int
115 sys_fork(struct fork_args *uap)
116 {
117 	struct lwp *lp = curthread->td_lwp;
118 	struct proc *p2;
119 	int error;
120 
121 	get_mplock();
122 	error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
123 	if (error == 0) {
124 		start_forked_proc(lp, p2);
125 		uap->sysmsg_fds[0] = p2->p_pid;
126 		uap->sysmsg_fds[1] = 0;
127 	}
128 	rel_mplock();
129 	return error;
130 }
131 
132 /*
133  * MPALMOSTSAFE
134  */
135 int
136 sys_vfork(struct vfork_args *uap)
137 {
138 	struct lwp *lp = curthread->td_lwp;
139 	struct proc *p2;
140 	int error;
141 
142 	get_mplock();
143 	error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
144 	if (error == 0) {
145 		start_forked_proc(lp, p2);
146 		uap->sysmsg_fds[0] = p2->p_pid;
147 		uap->sysmsg_fds[1] = 0;
148 	}
149 	rel_mplock();
150 	return error;
151 }
152 
153 /*
154  * Handle rforks.  An rfork may (1) operate on the current process without
155  * creating a new, (2) create a new process that shared the current process's
156  * vmspace, signals, and/or descriptors, or (3) create a new process that does
157  * not share these things (normal fork).
158  *
159  * Note that we only call start_forked_proc() if a new process is actually
160  * created.
161  *
162  * rfork { int flags }
163  *
164  * MPALMOSTSAFE
165  */
166 int
167 sys_rfork(struct rfork_args *uap)
168 {
169 	struct lwp *lp = curthread->td_lwp;
170 	struct proc *p2;
171 	int error;
172 
173 	if ((uap->flags & RFKERNELONLY) != 0)
174 		return (EINVAL);
175 
176 	get_mplock();
177 	error = fork1(lp, uap->flags | RFPGLOCK, &p2);
178 	if (error == 0) {
179 		if (p2)
180 			start_forked_proc(lp, p2);
181 		uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
182 		uap->sysmsg_fds[1] = 0;
183 	}
184 	rel_mplock();
185 	return error;
186 }
187 
188 /*
189  * MPALMOSTSAFE
190  */
191 int
192 sys_lwp_create(struct lwp_create_args *uap)
193 {
194 	struct proc *p = curproc;
195 	struct lwp *lp;
196 	struct lwp_params params;
197 	int error;
198 
199 	error = copyin(uap->params, &params, sizeof(params));
200 	if (error)
201 		goto fail2;
202 
203 	get_mplock();
204 	plimit_lwp_fork(p);	/* force exclusive access */
205 	lp = lwp_fork(curthread->td_lwp, p, RFPROC);
206 	error = cpu_prepare_lwp(lp, &params);
207 	if (params.tid1 != NULL &&
208 	    (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
209 		goto fail;
210 	if (params.tid2 != NULL &&
211 	    (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
212 		goto fail;
213 
214 	/*
215 	 * Now schedule the new lwp.
216 	 */
217 	p->p_usched->resetpriority(lp);
218 	crit_enter();
219 	lp->lwp_stat = LSRUN;
220 	p->p_usched->setrunqueue(lp);
221 	crit_exit();
222 	rel_mplock();
223 
224 	return (0);
225 
226 fail:
227 	lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
228 	--p->p_nthreads;
229 	/* lwp_dispose expects an exited lwp, and a held proc */
230 	lp->lwp_flag |= LWP_WEXIT;
231 	lp->lwp_thread->td_flags |= TDF_EXITING;
232 	PHOLD(p);
233 	lwp_dispose(lp);
234 	rel_mplock();
235 fail2:
236 	return (error);
237 }
238 
239 int	nprocs = 1;		/* process 0 */
240 
241 int
242 fork1(struct lwp *lp1, int flags, struct proc **procp)
243 {
244 	struct proc *p1 = lp1->lwp_proc;
245 	struct proc *p2, *pptr;
246 	struct pgrp *pgrp;
247 	uid_t uid;
248 	int ok, error;
249 	static int curfail = 0;
250 	static struct timeval lastfail;
251 	struct forklist *ep;
252 	struct filedesc_to_leader *fdtol;
253 
254 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
255 		return (EINVAL);
256 
257 	/*
258 	 * Here we don't create a new process, but we divorce
259 	 * certain parts of a process from itself.
260 	 */
261 	if ((flags & RFPROC) == 0) {
262 		/*
263 		 * This kind of stunt does not work anymore if
264 		 * there are native threads (lwps) running
265 		 */
266 		if (p1->p_nthreads != 1)
267 			return (EINVAL);
268 
269 		vm_fork(p1, 0, flags);
270 
271 		/*
272 		 * Close all file descriptors.
273 		 */
274 		if (flags & RFCFDG) {
275 			struct filedesc *fdtmp;
276 			fdtmp = fdinit(p1);
277 			fdfree(p1, fdtmp);
278 		}
279 
280 		/*
281 		 * Unshare file descriptors (from parent.)
282 		 */
283 		if (flags & RFFDG) {
284 			if (p1->p_fd->fd_refcnt > 1) {
285 				struct filedesc *newfd;
286 				newfd = fdcopy(p1);
287 				fdfree(p1, newfd);
288 			}
289 		}
290 		*procp = NULL;
291 		return (0);
292 	}
293 
294 	/*
295 	 * Interlock against process group signal delivery.  If signals
296 	 * are pending after the interlock is obtained we have to restart
297 	 * the system call to process the signals.  If we don't the child
298 	 * can miss a pgsignal (such as ^C) sent during the fork.
299 	 *
300 	 * We can't use CURSIG() here because it will process any STOPs
301 	 * and cause the process group lock to be held indefinitely.  If
302 	 * a STOP occurs, the fork will be restarted after the CONT.
303 	 */
304 	error = 0;
305 	pgrp = NULL;
306 	if ((flags & RFPGLOCK) && (pgrp = p1->p_pgrp) != NULL) {
307 		lockmgr(&pgrp->pg_lock, LK_SHARED);
308 		if (CURSIG_NOBLOCK(lp1)) {
309 			error = ERESTART;
310 			goto done;
311 		}
312 	}
313 
314 	/*
315 	 * Although process entries are dynamically created, we still keep
316 	 * a global limit on the maximum number we will create.  Don't allow
317 	 * a nonprivileged user to use the last ten processes; don't let root
318 	 * exceed the limit. The variable nprocs is the current number of
319 	 * processes, maxproc is the limit.
320 	 */
321 	uid = lp1->lwp_thread->td_ucred->cr_ruid;
322 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
323 		if (ppsratecheck(&lastfail, &curfail, 1))
324 			kprintf("maxproc limit exceeded by uid %d, please "
325 			       "see tuning(7) and login.conf(5).\n", uid);
326 		tsleep(&forksleep, 0, "fork", hz / 2);
327 		error = EAGAIN;
328 		goto done;
329 	}
330 	/*
331 	 * Increment the nprocs resource before blocking can occur.  There
332 	 * are hard-limits as to the number of processes that can run.
333 	 */
334 	nprocs++;
335 
336 	/*
337 	 * Increment the count of procs running with this uid. Don't allow
338 	 * a nonprivileged user to exceed their current limit.
339 	 */
340 	ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
341 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
342 	if (!ok) {
343 		/*
344 		 * Back out the process count
345 		 */
346 		nprocs--;
347 		if (ppsratecheck(&lastfail, &curfail, 1))
348 			kprintf("maxproc limit exceeded by uid %d, please "
349 			       "see tuning(7) and login.conf(5).\n", uid);
350 		tsleep(&forksleep, 0, "fork", hz / 2);
351 		error = EAGAIN;
352 		goto done;
353 	}
354 
355 	/* Allocate new proc. */
356 	p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
357 
358 	/*
359 	 * Setup linkage for kernel based threading XXX lwp
360 	 */
361 	if (flags & RFTHREAD) {
362 		p2->p_peers = p1->p_peers;
363 		p1->p_peers = p2;
364 		p2->p_leader = p1->p_leader;
365 	} else {
366 		p2->p_leader = p2;
367 	}
368 
369 	RB_INIT(&p2->p_lwp_tree);
370 	spin_init(&p2->p_spin);
371 	lwkt_token_init(&p2->p_token, "iproc");
372 	p2->p_lasttid = -1;	/* first tid will be 0 */
373 
374 	/*
375 	 * Setting the state to SIDL protects the partially initialized
376 	 * process once it starts getting hooked into the rest of the system.
377 	 */
378 	p2->p_stat = SIDL;
379 	proc_add_allproc(p2);
380 
381 	/*
382 	 * Make a proc table entry for the new process.
383 	 * The whole structure was zeroed above, so copy the section that is
384 	 * copied directly from the parent.
385 	 */
386 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
387 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
388 
389 	/*
390 	 * Duplicate sub-structures as needed.
391 	 * Increase reference counts on shared objects.
392 	 */
393 	if (p1->p_flag & P_PROFIL)
394 		startprofclock(p2);
395 	p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
396 	KKASSERT(p2->p_lock == 0);
397 
398 	if (jailed(p2->p_ucred))
399 		p2->p_flag |= P_JAILED;
400 
401 	if (p2->p_args)
402 		p2->p_args->ar_ref++;
403 
404 	p2->p_usched = p1->p_usched;
405 	/* XXX: verify copy of the secondary iosched stuff */
406 	dsched_new_proc(p2);
407 
408 	if (flags & RFSIGSHARE) {
409 		p2->p_sigacts = p1->p_sigacts;
410 		p2->p_sigacts->ps_refcnt++;
411 	} else {
412 		p2->p_sigacts = (struct sigacts *)kmalloc(sizeof(*p2->p_sigacts),
413 		    M_SUBPROC, M_WAITOK);
414 		bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
415 		p2->p_sigacts->ps_refcnt = 1;
416 	}
417 	if (flags & RFLINUXTHPN)
418 	        p2->p_sigparent = SIGUSR1;
419 	else
420 	        p2->p_sigparent = SIGCHLD;
421 
422 	/* bump references to the text vnode (for procfs) */
423 	p2->p_textvp = p1->p_textvp;
424 	if (p2->p_textvp)
425 		vref(p2->p_textvp);
426 
427 	/* copy namecache handle to the text file */
428 	if (p1->p_textnch.mount)
429 		cache_copy(&p1->p_textnch, &p2->p_textnch);
430 
431 	/*
432 	 * Handle file descriptors
433 	 */
434 	if (flags & RFCFDG) {
435 		p2->p_fd = fdinit(p1);
436 		fdtol = NULL;
437 	} else if (flags & RFFDG) {
438 		p2->p_fd = fdcopy(p1);
439 		fdtol = NULL;
440 	} else {
441 		p2->p_fd = fdshare(p1);
442 		if (p1->p_fdtol == NULL)
443 			p1->p_fdtol =
444 				filedesc_to_leader_alloc(NULL,
445 							 p1->p_leader);
446 		if ((flags & RFTHREAD) != 0) {
447 			/*
448 			 * Shared file descriptor table and
449 			 * shared process leaders.
450 			 */
451 			fdtol = p1->p_fdtol;
452 			fdtol->fdl_refcount++;
453 		} else {
454 			/*
455 			 * Shared file descriptor table, and
456 			 * different process leaders
457 			 */
458 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
459 		}
460 	}
461 	p2->p_fdtol = fdtol;
462 	p2->p_limit = plimit_fork(p1);
463 
464 	/*
465 	 * Preserve some more flags in subprocess.  P_PROFIL has already
466 	 * been preserved.
467 	 */
468 	p2->p_flag |= p1->p_flag & P_SUGID;
469 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
470 		p2->p_flag |= P_CONTROLT;
471 	if (flags & RFPPWAIT)
472 		p2->p_flag |= P_PPWAIT;
473 
474 	/*
475 	 * Inherit the virtual kernel structure (allows a virtual kernel
476 	 * to fork to simulate multiple cpus).
477 	 */
478 	if (p1->p_vkernel)
479 		vkernel_inherit(p1, p2);
480 
481 	/*
482 	 * Once we are on a pglist we may receive signals.  XXX we might
483 	 * race a ^C being sent to the process group by not receiving it
484 	 * at all prior to this line.
485 	 */
486 	LIST_INSERT_AFTER(p1, p2, p_pglist);
487 
488 	/*
489 	 * Attach the new process to its parent.
490 	 *
491 	 * If RFNOWAIT is set, the newly created process becomes a child
492 	 * of init.  This effectively disassociates the child from the
493 	 * parent.
494 	 */
495 	if (flags & RFNOWAIT)
496 		pptr = initproc;
497 	else
498 		pptr = p1;
499 	p2->p_pptr = pptr;
500 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
501 	LIST_INIT(&p2->p_children);
502 	varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
503 	callout_init(&p2->p_ithandle);
504 
505 #ifdef KTRACE
506 	/*
507 	 * Copy traceflag and tracefile if enabled.  If not inherited,
508 	 * these were zeroed above but we still could have a trace race
509 	 * so make sure p2's p_tracenode is NULL.
510 	 */
511 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
512 		p2->p_traceflag = p1->p_traceflag;
513 		p2->p_tracenode = ktrinherit(p1->p_tracenode);
514 	}
515 #endif
516 
517 	/*
518 	 * This begins the section where we must prevent the parent
519 	 * from being swapped.
520 	 *
521 	 * Gets PRELE'd in the caller in start_forked_proc().
522 	 */
523 	PHOLD(p1);
524 
525 	vm_fork(p1, p2, flags);
526 
527 	/*
528 	 * Create the first lwp associated with the new proc.
529 	 * It will return via a different execution path later, directly
530 	 * into userland, after it was put on the runq by
531 	 * start_forked_proc().
532 	 */
533 	lwp_fork(lp1, p2, flags);
534 
535 	if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
536 		mycpu->gd_cnt.v_forks++;
537 		mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
538 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
539 		mycpu->gd_cnt.v_vforks++;
540 		mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
541 	} else if (p1 == &proc0) {
542 		mycpu->gd_cnt.v_kthreads++;
543 		mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
544 	} else {
545 		mycpu->gd_cnt.v_rforks++;
546 		mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
547 	}
548 
549 	/*
550 	 * Both processes are set up, now check if any loadable modules want
551 	 * to adjust anything.
552 	 *   What if they have an error? XXX
553 	 */
554 	TAILQ_FOREACH(ep, &fork_list, next) {
555 		(*ep->function)(p1, p2, flags);
556 	}
557 
558 	/*
559 	 * Set the start time.  Note that the process is not runnable.  The
560 	 * caller is responsible for making it runnable.
561 	 */
562 	microtime(&p2->p_start);
563 	p2->p_acflag = AFORK;
564 
565 	/*
566 	 * tell any interested parties about the new process
567 	 */
568 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
569 
570 	/*
571 	 * Return child proc pointer to parent.
572 	 */
573 	*procp = p2;
574 done:
575 	if (pgrp)
576 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
577 	return (error);
578 }
579 
580 static struct lwp *
581 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
582 {
583 	struct lwp *lp;
584 	struct thread *td;
585 
586 	lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
587 
588 	lp->lwp_proc = destproc;
589 	lp->lwp_vmspace = destproc->p_vmspace;
590 	lp->lwp_stat = LSRUN;
591 	bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
592 	    (unsigned) ((caddr_t)&lp->lwp_endcopy -
593 			(caddr_t)&lp->lwp_startcopy));
594 	lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
595 	/*
596 	 * Set cpbase to the last timeout that occured (not the upcoming
597 	 * timeout).
598 	 *
599 	 * A critical section is required since a timer IPI can update
600 	 * scheduler specific data.
601 	 */
602 	crit_enter();
603 	lp->lwp_cpbase = mycpu->gd_schedclock.time -
604 			mycpu->gd_schedclock.periodic;
605 	destproc->p_usched->heuristic_forking(origlp, lp);
606 	crit_exit();
607 	lp->lwp_cpumask &= usched_mastermask;
608 
609 	/*
610 	 * Assign a TID to the lp.  Loop until the insert succeeds (returns
611 	 * NULL).
612 	 */
613 	lp->lwp_tid = destproc->p_lasttid;
614 	do {
615 		if (++lp->lwp_tid < 0)
616 			lp->lwp_tid = 1;
617 	} while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
618 	destproc->p_lasttid = lp->lwp_tid;
619 	destproc->p_nthreads++;
620 
621 	td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
622 	lp->lwp_thread = td;
623 	td->td_proc = destproc;
624 	td->td_lwp = lp;
625 	td->td_switch = cpu_heavy_switch;
626 	lwkt_setpri(td, TDPRI_KERN_USER);
627 	lwkt_set_comm(td, "%s", destproc->p_comm);
628 
629 	/*
630 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
631 	 * and make the child ready to run.
632 	 */
633 	cpu_fork(origlp, lp, flags);
634 	caps_fork(origlp->lwp_thread, lp->lwp_thread);
635 	kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
636 
637 	return (lp);
638 }
639 
640 /*
641  * The next two functionms are general routines to handle adding/deleting
642  * items on the fork callout list.
643  *
644  * at_fork():
645  * Take the arguments given and put them onto the fork callout list,
646  * However first make sure that it's not already there.
647  * Returns 0 on success or a standard error number.
648  */
649 int
650 at_fork(forklist_fn function)
651 {
652 	struct forklist *ep;
653 
654 #ifdef INVARIANTS
655 	/* let the programmer know if he's been stupid */
656 	if (rm_at_fork(function)) {
657 		kprintf("WARNING: fork callout entry (%p) already present\n",
658 		    function);
659 	}
660 #endif
661 	ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
662 	ep->function = function;
663 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
664 	return (0);
665 }
666 
667 /*
668  * Scan the exit callout list for the given item and remove it..
669  * Returns the number of items removed (0 or 1)
670  */
671 int
672 rm_at_fork(forklist_fn function)
673 {
674 	struct forklist *ep;
675 
676 	TAILQ_FOREACH(ep, &fork_list, next) {
677 		if (ep->function == function) {
678 			TAILQ_REMOVE(&fork_list, ep, next);
679 			kfree(ep, M_ATFORK);
680 			return(1);
681 		}
682 	}
683 	return (0);
684 }
685 
686 /*
687  * Add a forked process to the run queue after any remaining setup, such
688  * as setting the fork handler, has been completed.
689  */
690 void
691 start_forked_proc(struct lwp *lp1, struct proc *p2)
692 {
693 	struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
694 
695 	/*
696 	 * Move from SIDL to RUN queue, and activate the process's thread.
697 	 * Activation of the thread effectively makes the process "a"
698 	 * current process, so we do not setrunqueue().
699 	 *
700 	 * YYY setrunqueue works here but we should clean up the trampoline
701 	 * code so we just schedule the LWKT thread and let the trampoline
702 	 * deal with the userland scheduler on return to userland.
703 	 */
704 	KASSERT(p2->p_stat == SIDL,
705 	    ("cannot start forked process, bad status: %p", p2));
706 	p2->p_usched->resetpriority(lp2);
707 	crit_enter();
708 	p2->p_stat = SACTIVE;
709 	lp2->lwp_stat = LSRUN;
710 	p2->p_usched->setrunqueue(lp2);
711 	crit_exit();
712 
713 	/*
714 	 * Now can be swapped.
715 	 */
716 	PRELE(lp1->lwp_proc);
717 
718 	/*
719 	 * Preserve synchronization semantics of vfork.  If waiting for
720 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
721 	 * proc (in case of exit).
722 	 */
723 	while (p2->p_flag & P_PPWAIT)
724 		tsleep(lp1->lwp_proc, 0, "ppwait", 0);
725 }
726