xref: /dragonfly/sys/kern/kern_fork.c (revision 6fb88001)
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 1.44 2005/11/14 18:50:05 dillon 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 #include <vm/vm_zone.h>
67 
68 #include <sys/vmmeter.h>
69 #include <sys/user.h>
70 #include <sys/thread2.h>
71 
72 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
73 
74 /*
75  * These are the stuctures used to create a callout list for things to do
76  * when forking a process
77  */
78 struct forklist {
79 	forklist_fn function;
80 	TAILQ_ENTRY(forklist) next;
81 };
82 
83 TAILQ_HEAD(forklist_head, forklist);
84 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
85 
86 int forksleep; /* Place for fork1() to sleep on. */
87 
88 /* ARGSUSED */
89 int
90 fork(struct fork_args *uap)
91 {
92 	struct lwp *lp = curthread->td_lwp;
93 	struct proc *p2;
94 	int error;
95 
96 	error = fork1(lp, RFFDG | RFPROC, &p2);
97 	if (error == 0) {
98 		start_forked_proc(lp, p2);
99 		uap->sysmsg_fds[0] = p2->p_pid;
100 		uap->sysmsg_fds[1] = 0;
101 	}
102 	return error;
103 }
104 
105 /* ARGSUSED */
106 int
107 vfork(struct vfork_args *uap)
108 {
109 	struct lwp *lp = curthread->td_lwp;
110 	struct proc *p2;
111 	int error;
112 
113 	error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
114 	if (error == 0) {
115 		start_forked_proc(lp, p2);
116 		uap->sysmsg_fds[0] = p2->p_pid;
117 		uap->sysmsg_fds[1] = 0;
118 	}
119 	return error;
120 }
121 
122 /*
123  * Handle rforks.  An rfork may (1) operate on the current process without
124  * creating a new, (2) create a new process that shared the current process's
125  * vmspace, signals, and/or descriptors, or (3) create a new process that does
126  * not share these things (normal fork).
127  *
128  * Note that we only call start_forked_proc() if a new process is actually
129  * created.
130  *
131  * rfork { int flags }
132  */
133 int
134 rfork(struct rfork_args *uap)
135 {
136 	struct lwp *lp = curthread->td_lwp;
137 	struct proc *p2;
138 	int error;
139 
140 	if ((uap->flags & RFKERNELONLY) != 0)
141 		return (EINVAL);
142 
143 	error = fork1(lp, uap->flags, &p2);
144 	if (error == 0) {
145 		if (p2)
146 			start_forked_proc(lp, p2);
147 		uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
148 		uap->sysmsg_fds[1] = 0;
149 	}
150 	return error;
151 }
152 
153 
154 int	nprocs = 1;		/* process 0 */
155 static int nextpid = 0;
156 
157 /*
158  * Random component to nextpid generation.  We mix in a random factor to make
159  * it a little harder to predict.  We sanity check the modulus value to avoid
160  * doing it in critical paths.  Don't let it be too small or we pointlessly
161  * waste randomness entropy, and don't let it be impossibly large.  Using a
162  * modulus that is too big causes a LOT more process table scans and slows
163  * down fork processing as the pidchecked caching is defeated.
164  */
165 static int randompid = 0;
166 
167 static int
168 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
169 {
170 		int error, pid;
171 
172 		pid = randompid;
173 		error = sysctl_handle_int(oidp, &pid, 0, req);
174 		if (error || !req->newptr)
175 			return (error);
176 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
177 			pid = PID_MAX - 100;
178 		else if (pid < 2)			/* NOP */
179 			pid = 0;
180 		else if (pid < 100)			/* Make it reasonable */
181 			pid = 100;
182 		randompid = pid;
183 		return (error);
184 }
185 
186 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
187     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
188 
189 int
190 fork1(struct lwp *lp1, int flags, struct proc **procp)
191 {
192 	struct proc *p1 = lp1->lwp_proc;
193 	struct proc *p2, *pptr;
194 	struct lwp *lp2;
195 	uid_t uid;
196 	struct proc *newproc;
197 	int ok;
198 	static int curfail = 0, pidchecked = 0;
199 	static struct timeval lastfail;
200 	struct forklist *ep;
201 	struct filedesc_to_leader *fdtol;
202 
203 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
204 		return (EINVAL);
205 
206 	/*
207 	 * Here we don't create a new process, but we divorce
208 	 * certain parts of a process from itself.
209 	 */
210 	if ((flags & RFPROC) == 0) {
211 
212 		vm_fork(p1, 0, flags);
213 
214 		/*
215 		 * Close all file descriptors.
216 		 */
217 		if (flags & RFCFDG) {
218 			struct filedesc *fdtmp;
219 			fdtmp = fdinit(p1);
220 			fdfree(p1);
221 			p1->p_fd = fdtmp;
222 		}
223 
224 		/*
225 		 * Unshare file descriptors (from parent.)
226 		 */
227 		if (flags & RFFDG) {
228 			if (p1->p_fd->fd_refcnt > 1) {
229 				struct filedesc *newfd;
230 				newfd = fdcopy(p1);
231 				fdfree(p1);
232 				p1->p_fd = newfd;
233 			}
234 		}
235 		*procp = NULL;
236 		return (0);
237 	}
238 
239 	/*
240 	 * Although process entries are dynamically created, we still keep
241 	 * a global limit on the maximum number we will create.  Don't allow
242 	 * a nonprivileged user to use the last ten processes; don't let root
243 	 * exceed the limit. The variable nprocs is the current number of
244 	 * processes, maxproc is the limit.
245 	 */
246 	uid = p1->p_ucred->cr_ruid;
247 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
248 		if (ppsratecheck(&lastfail, &curfail, 1))
249 			printf("maxproc limit exceeded by uid %d, please "
250 			       "see tuning(7) and login.conf(5).\n", uid);
251 		tsleep(&forksleep, 0, "fork", hz / 2);
252 		return (EAGAIN);
253 	}
254 	/*
255 	 * Increment the nprocs resource before blocking can occur.  There
256 	 * are hard-limits as to the number of processes that can run.
257 	 */
258 	nprocs++;
259 
260 	/*
261 	 * Increment the count of procs running with this uid. Don't allow
262 	 * a nonprivileged user to exceed their current limit.
263 	 */
264 	ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
265 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
266 	if (!ok) {
267 		/*
268 		 * Back out the process count
269 		 */
270 		nprocs--;
271 		if (ppsratecheck(&lastfail, &curfail, 1))
272 			printf("maxproc limit exceeded by uid %d, please "
273 			       "see tuning(7) and login.conf(5).\n", uid);
274 		tsleep(&forksleep, 0, "fork", hz / 2);
275 		return (EAGAIN);
276 	}
277 
278 	/* Allocate new proc. */
279 	newproc = zalloc(proc_zone);
280 
281 	/*
282 	 * Setup linkage for kernel based threading XXX lwp
283 	 */
284 	if ((flags & RFTHREAD) != 0) {
285 		newproc->p_peers = p1->p_peers;
286 		p1->p_peers = newproc;
287 		newproc->p_leader = p1->p_leader;
288 	} else {
289 		newproc->p_peers = 0;
290 		newproc->p_leader = newproc;
291 	}
292 
293 	newproc->p_wakeup = 0;
294 	newproc->p_vmspace = NULL;
295 	TAILQ_INIT(&newproc->p_lwp.lwp_sysmsgq);
296 	LIST_INIT(&newproc->p_lwps);
297 
298 	/* XXX lwp */
299 	lp2 = &newproc->p_lwp;
300 	lp2->lwp_proc = newproc;
301 	lp2->lwp_tid = 0;
302 	LIST_INSERT_HEAD(&newproc->p_lwps, lp2, lwp_list);
303 	newproc->p_nthreads = 1;
304 
305 	/*
306 	 * Find an unused process ID.  We remember a range of unused IDs
307 	 * ready to use (from nextpid+1 through pidchecked-1).
308 	 */
309 	nextpid++;
310 	if (randompid)
311 		nextpid += arc4random() % randompid;
312 retry:
313 	/*
314 	 * If the process ID prototype has wrapped around,
315 	 * restart somewhat above 0, as the low-numbered procs
316 	 * tend to include daemons that don't exit.
317 	 */
318 	if (nextpid >= PID_MAX) {
319 		nextpid = nextpid % PID_MAX;
320 		if (nextpid < 100)
321 			nextpid += 100;
322 		pidchecked = 0;
323 	}
324 	if (nextpid >= pidchecked) {
325 		int doingzomb = 0;
326 
327 		pidchecked = PID_MAX;
328 		/*
329 		 * Scan the active and zombie procs to check whether this pid
330 		 * is in use.  Remember the lowest pid that's greater
331 		 * than nextpid, so we can avoid checking for a while.
332 		 */
333 		p2 = LIST_FIRST(&allproc);
334 again:
335 		for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) {
336 			while (p2->p_pid == nextpid ||
337 			    p2->p_pgrp->pg_id == nextpid ||
338 			    p2->p_session->s_sid == nextpid) {
339 				nextpid++;
340 				if (nextpid >= pidchecked)
341 					goto retry;
342 			}
343 			if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
344 				pidchecked = p2->p_pid;
345 			if (p2->p_pgrp->pg_id > nextpid &&
346 			    pidchecked > p2->p_pgrp->pg_id)
347 				pidchecked = p2->p_pgrp->pg_id;
348 			if (p2->p_session->s_sid > nextpid &&
349 			    pidchecked > p2->p_session->s_sid)
350 				pidchecked = p2->p_session->s_sid;
351 		}
352 		if (!doingzomb) {
353 			doingzomb = 1;
354 			p2 = LIST_FIRST(&zombproc);
355 			goto again;
356 		}
357 	}
358 
359 	p2 = newproc;
360 	p2->p_stat = SIDL;			/* protect against others */
361 	p2->p_pid = nextpid;
362 	LIST_INSERT_HEAD(&allproc, p2, p_list);
363 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
364 
365 	/*
366 	 * Make a proc table entry for the new process.
367 	 * Start by zeroing the section of proc that is zero-initialized,
368 	 * then copy the section that is copied directly from the parent.
369 	 */
370 	bzero(&p2->p_startzero,
371 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
372 	bzero(&lp2->lwp_startzero,
373 	    (unsigned) ((caddr_t)&lp2->lwp_endzero -
374 			(caddr_t)&lp2->lwp_startzero));
375 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
376 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
377 	bcopy(&p1->p_lwp.lwp_startcopy, &lp2->lwp_startcopy,
378 	    (unsigned) ((caddr_t)&lp2->lwp_endcopy -
379 			(caddr_t)&lp2->lwp_startcopy));
380 
381 	p2->p_aioinfo = NULL;
382 
383 	/*
384 	 * Duplicate sub-structures as needed.
385 	 * Increase reference counts on shared objects.
386 	 * The p_stats and p_sigacts substructs are set in vm_fork.
387 	 */
388 	p2->p_flag = 0;
389 	if (p1->p_flag & P_PROFIL)
390 		startprofclock(p2);
391 	p2->p_ucred = crhold(p1->p_ucred);
392 
393 	if (jailed(p2->p_ucred))
394 		p2->p_flag |= P_JAILED;
395 
396 	if (p2->p_args)
397 		p2->p_args->ar_ref++;
398 
399 	if (flags & RFSIGSHARE) {
400 		p2->p_procsig = p1->p_procsig;
401 		p2->p_procsig->ps_refcnt++;
402 		if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
403 			struct sigacts *newsigacts;
404 
405 			/* Create the shared sigacts structure */
406 			MALLOC(newsigacts, struct sigacts *,
407 			    sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
408 			crit_enter();
409 			/*
410 			 * Set p_sigacts to the new shared structure.
411 			 * Note that this is updating p1->p_sigacts at the
412 			 * same time, since p_sigacts is just a pointer to
413 			 * the shared p_procsig->ps_sigacts.
414 			 */
415 			p2->p_sigacts  = newsigacts;
416 			bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
417 			    sizeof(*p2->p_sigacts));
418 			*p2->p_sigacts = p1->p_addr->u_sigacts;
419 			crit_exit();
420 		}
421 	} else {
422 		MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
423 		    M_SUBPROC, M_WAITOK);
424 		bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
425 		p2->p_procsig->ps_refcnt = 1;
426 		p2->p_sigacts = NULL;	/* finished in vm_fork() */
427 	}
428 	if (flags & RFLINUXTHPN)
429 	        p2->p_sigparent = SIGUSR1;
430 	else
431 	        p2->p_sigparent = SIGCHLD;
432 
433 	/* bump references to the text vnode (for procfs) */
434 	p2->p_textvp = p1->p_textvp;
435 	if (p2->p_textvp)
436 		vref(p2->p_textvp);
437 
438 	if (flags & RFCFDG) {
439 		p2->p_fd = fdinit(p1);
440 		fdtol = NULL;
441 	} else if (flags & RFFDG) {
442 		p2->p_fd = fdcopy(p1);
443 		fdtol = NULL;
444 	} else {
445 		p2->p_fd = fdshare(p1);
446 		if (p1->p_fdtol == NULL)
447 			p1->p_fdtol =
448 				filedesc_to_leader_alloc(NULL,
449 							 p1->p_leader);
450 		if ((flags & RFTHREAD) != 0) {
451 			/*
452 			 * Shared file descriptor table and
453 			 * shared process leaders.
454 			 */
455 			fdtol = p1->p_fdtol;
456 			fdtol->fdl_refcount++;
457 		} else {
458 			/*
459 			 * Shared file descriptor table, and
460 			 * different process leaders
461 			 */
462 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
463 		}
464 	}
465 	p2->p_fdtol = fdtol;
466 
467 	/*
468 	 * If p_limit is still copy-on-write, bump refcnt,
469 	 * otherwise get a copy that won't be modified.
470 	 * (If PL_SHAREMOD is clear, the structure is shared
471 	 * copy-on-write.)
472 	 */
473 	if (p1->p_limit->p_lflags & PL_SHAREMOD) {
474 		p2->p_limit = limcopy(p1->p_limit);
475 	} else {
476 		p2->p_limit = p1->p_limit;
477 		p2->p_limit->p_refcnt++;
478 	}
479 
480 	/*
481 	 * Preserve some more flags in subprocess.  P_PROFIL has already
482 	 * been preserved.
483 	 */
484 	p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
485 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
486 		p2->p_flag |= P_CONTROLT;
487 	if (flags & RFPPWAIT)
488 		p2->p_flag |= P_PPWAIT;
489 
490 	/*
491 	 * Once we are on a pglist we may receive signals.  XXX we might
492 	 * race a ^C being sent to the process group by not receiving it
493 	 * at all prior to this line.
494 	 */
495 	LIST_INSERT_AFTER(p1, p2, p_pglist);
496 
497 	/*
498 	 * Attach the new process to its parent.
499 	 *
500 	 * If RFNOWAIT is set, the newly created process becomes a child
501 	 * of init.  This effectively disassociates the child from the
502 	 * parent.
503 	 */
504 	if (flags & RFNOWAIT)
505 		pptr = initproc;
506 	else
507 		pptr = p1;
508 	p2->p_pptr = pptr;
509 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
510 	LIST_INIT(&p2->p_children);
511 	varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
512 	callout_init(&p2->p_ithandle);
513 
514 #ifdef KTRACE
515 	/*
516 	 * Copy traceflag and tracefile if enabled.  If not inherited,
517 	 * these were zeroed above but we still could have a trace race
518 	 * so make sure p2's p_tracep is NULL.
519 	 */
520 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
521 		p2->p_traceflag = p1->p_traceflag;
522 		if ((p2->p_tracep = p1->p_tracep) != NULL)
523 			vref(p2->p_tracep);
524 	}
525 #endif
526 
527 	/*
528 	 * Inherit the scheduler and initialize scheduler-related fields.
529 	 * Set cpbase to the last timeout that occured (not the upcoming
530 	 * timeout).
531 	 */
532 	p2->p_usched = p1->p_usched;
533 	lp2->lwp_cpbase = mycpu->gd_schedclock.time -
534 			mycpu->gd_schedclock.periodic;
535 	p2->p_usched->heuristic_forking(&p1->p_lwp, lp2);
536 
537 	/*
538 	 * This begins the section where we must prevent the parent
539 	 * from being swapped.
540 	 */
541 	PHOLD(p1);
542 
543 	/*
544 	 * Finish creating the child process.  It will return via a different
545 	 * execution path later.  (ie: directly into user mode)
546 	 */
547 	vm_fork(p1, p2, flags);
548 	caps_fork(p1, p2, flags);
549 
550 	if (flags == (RFFDG | RFPROC)) {
551 		mycpu->gd_cnt.v_forks++;
552 		mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
553 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
554 		mycpu->gd_cnt.v_vforks++;
555 		mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
556 	} else if (p1 == &proc0) {
557 		mycpu->gd_cnt.v_kthreads++;
558 		mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
559 	} else {
560 		mycpu->gd_cnt.v_rforks++;
561 		mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
562 	}
563 
564 	/*
565 	 * Both processes are set up, now check if any loadable modules want
566 	 * to adjust anything.
567 	 *   What if they have an error? XXX
568 	 */
569 	TAILQ_FOREACH(ep, &fork_list, next) {
570 		(*ep->function)(p1, p2, flags);
571 	}
572 
573 	/*
574 	 * Set the start time.  Note that the process is not runnable.  The
575 	 * caller is responsible for making it runnable.
576 	 */
577 	microtime(&p2->p_start);
578 	p2->p_acflag = AFORK;
579 
580 	/*
581 	 * tell any interested parties about the new process
582 	 */
583 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
584 
585 	/*
586 	 * Return child proc pointer to parent.
587 	 */
588 	*procp = p2;
589 	return (0);
590 }
591 
592 /*
593  * The next two functionms are general routines to handle adding/deleting
594  * items on the fork callout list.
595  *
596  * at_fork():
597  * Take the arguments given and put them onto the fork callout list,
598  * However first make sure that it's not already there.
599  * Returns 0 on success or a standard error number.
600  */
601 int
602 at_fork(forklist_fn function)
603 {
604 	struct forklist *ep;
605 
606 #ifdef INVARIANTS
607 	/* let the programmer know if he's been stupid */
608 	if (rm_at_fork(function)) {
609 		printf("WARNING: fork callout entry (%p) already present\n",
610 		    function);
611 	}
612 #endif
613 	ep = malloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
614 	ep->function = function;
615 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
616 	return (0);
617 }
618 
619 /*
620  * Scan the exit callout list for the given item and remove it..
621  * Returns the number of items removed (0 or 1)
622  */
623 int
624 rm_at_fork(forklist_fn function)
625 {
626 	struct forklist *ep;
627 
628 	TAILQ_FOREACH(ep, &fork_list, next) {
629 		if (ep->function == function) {
630 			TAILQ_REMOVE(&fork_list, ep, next);
631 			free(ep, M_ATFORK);
632 			return(1);
633 		}
634 	}
635 	return (0);
636 }
637 
638 /*
639  * Add a forked process to the run queue after any remaining setup, such
640  * as setting the fork handler, has been completed.
641  */
642 void
643 start_forked_proc(struct lwp *lp1, struct proc *p2)
644 {
645 	struct lwp *lp2;
646 
647 	KKASSERT(p2 != NULL && p2->p_nthreads == 1);
648 
649 	lp2 = LIST_FIRST(&p2->p_lwps);
650 
651 	/*
652 	 * Move from SIDL to RUN queue, and activate the process's thread.
653 	 * Activation of the thread effectively makes the process "a"
654 	 * current process, so we do not setrunqueue().
655 	 *
656 	 * YYY setrunqueue works here but we should clean up the trampoline
657 	 * code so we just schedule the LWKT thread and let the trampoline
658 	 * deal with the userland scheduler on return to userland.
659 	 */
660 	KASSERT(p2->p_stat == SIDL,
661 	    ("cannot start forked process, bad status: %p", p2));
662 	p2->p_usched->resetpriority(lp2);
663 	crit_enter();
664 	p2->p_stat = SRUN;
665 	p2->p_usched->setrunqueue(lp2);
666 	crit_exit();
667 
668 	/*
669 	 * Now can be swapped.
670 	 */
671 	PRELE(lp1->lwp_proc);
672 
673 	/*
674 	 * Preserve synchronization semantics of vfork.  If waiting for
675 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
676 	 * proc (in case of exit).
677 	 */
678 	while (p2->p_flag & P_PPWAIT)
679 		tsleep(lp1->lwp_proc, 0, "ppwait", 0);
680 }
681