xref: /dragonfly/sys/kern/kern_fork.c (revision e293de53)
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 
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 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
87 
88 int forksleep; /* Place for fork1() to sleep on. */
89 
90 /*
91  * Red-Black tree support for LWPs
92  */
93 
94 static int
95 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
96 {
97 	if (lp1->lwp_tid < lp2->lwp_tid)
98 		return(-1);
99 	if (lp1->lwp_tid > lp2->lwp_tid)
100 		return(1);
101 	return(0);
102 }
103 
104 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
105 
106 
107 /* ARGSUSED */
108 int
109 sys_fork(struct fork_args *uap)
110 {
111 	struct lwp *lp = curthread->td_lwp;
112 	struct proc *p2;
113 	int error;
114 
115 	error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
116 	if (error == 0) {
117 		start_forked_proc(lp, p2);
118 		uap->sysmsg_fds[0] = p2->p_pid;
119 		uap->sysmsg_fds[1] = 0;
120 	}
121 	return error;
122 }
123 
124 /* ARGSUSED */
125 int
126 sys_vfork(struct vfork_args *uap)
127 {
128 	struct lwp *lp = curthread->td_lwp;
129 	struct proc *p2;
130 	int error;
131 
132 	error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
133 	if (error == 0) {
134 		start_forked_proc(lp, p2);
135 		uap->sysmsg_fds[0] = p2->p_pid;
136 		uap->sysmsg_fds[1] = 0;
137 	}
138 	return error;
139 }
140 
141 /*
142  * Handle rforks.  An rfork may (1) operate on the current process without
143  * creating a new, (2) create a new process that shared the current process's
144  * vmspace, signals, and/or descriptors, or (3) create a new process that does
145  * not share these things (normal fork).
146  *
147  * Note that we only call start_forked_proc() if a new process is actually
148  * created.
149  *
150  * rfork { int flags }
151  */
152 int
153 sys_rfork(struct rfork_args *uap)
154 {
155 	struct lwp *lp = curthread->td_lwp;
156 	struct proc *p2;
157 	int error;
158 
159 	if ((uap->flags & RFKERNELONLY) != 0)
160 		return (EINVAL);
161 
162 	error = fork1(lp, uap->flags | RFPGLOCK, &p2);
163 	if (error == 0) {
164 		if (p2)
165 			start_forked_proc(lp, p2);
166 		uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
167 		uap->sysmsg_fds[1] = 0;
168 	}
169 	return error;
170 }
171 
172 int
173 sys_lwp_create(struct lwp_create_args *uap)
174 {
175 	struct proc *p = curproc;
176 	struct lwp *lp;
177 	struct lwp_params params;
178 	int error;
179 
180 	error = copyin(uap->params, &params, sizeof(params));
181 	if (error)
182 		goto fail2;
183 
184 	plimit_lwp_fork(p);	/* force exclusive access */
185 	lp = lwp_fork(curthread->td_lwp, p, RFPROC);
186 	error = cpu_prepare_lwp(lp, &params);
187 	if (params.tid1 != NULL &&
188 	    (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
189 		goto fail;
190 	if (params.tid2 != NULL &&
191 	    (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
192 		goto fail;
193 
194 	/*
195 	 * Now schedule the new lwp.
196 	 */
197 	p->p_usched->resetpriority(lp);
198 	crit_enter();
199 	lp->lwp_stat = LSRUN;
200 	p->p_usched->setrunqueue(lp);
201 	crit_exit();
202 
203 	return (0);
204 
205 fail:
206 	lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
207 	--p->p_nthreads;
208 	/* lwp_dispose expects an exited lwp, and a held proc */
209 	lp->lwp_flag |= LWP_WEXIT;
210 	lp->lwp_thread->td_flags |= TDF_EXITING;
211 	PHOLD(p);
212 	lwp_dispose(lp);
213 fail2:
214 	return (error);
215 }
216 
217 int	nprocs = 1;		/* process 0 */
218 
219 int
220 fork1(struct lwp *lp1, int flags, struct proc **procp)
221 {
222 	struct proc *p1 = lp1->lwp_proc;
223 	struct proc *p2, *pptr;
224 	struct pgrp *pgrp;
225 	uid_t uid;
226 	int ok, error;
227 	static int curfail = 0;
228 	static struct timeval lastfail;
229 	struct forklist *ep;
230 	struct filedesc_to_leader *fdtol;
231 
232 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
233 		return (EINVAL);
234 
235 	/*
236 	 * Here we don't create a new process, but we divorce
237 	 * certain parts of a process from itself.
238 	 */
239 	if ((flags & RFPROC) == 0) {
240 		/*
241 		 * This kind of stunt does not work anymore if
242 		 * there are native threads (lwps) running
243 		 */
244 		if (p1->p_nthreads != 1)
245 			return (EINVAL);
246 
247 		vm_fork(p1, 0, flags);
248 
249 		/*
250 		 * Close all file descriptors.
251 		 */
252 		if (flags & RFCFDG) {
253 			struct filedesc *fdtmp;
254 			fdtmp = fdinit(p1);
255 			fdfree(p1);
256 			p1->p_fd = fdtmp;
257 		}
258 
259 		/*
260 		 * Unshare file descriptors (from parent.)
261 		 */
262 		if (flags & RFFDG) {
263 			if (p1->p_fd->fd_refcnt > 1) {
264 				struct filedesc *newfd;
265 				newfd = fdcopy(p1);
266 				fdfree(p1);
267 				p1->p_fd = newfd;
268 			}
269 		}
270 		*procp = NULL;
271 		return (0);
272 	}
273 
274 	/*
275 	 * Interlock against process group signal delivery.  If signals
276 	 * are pending after the interlock is obtained we have to restart
277 	 * the system call to process the signals.  If we don't the child
278 	 * can miss a pgsignal (such as ^C) sent during the fork.
279 	 *
280 	 * We can't use CURSIG() here because it will process any STOPs
281 	 * and cause the process group lock to be held indefinitely.  If
282 	 * a STOP occurs, the fork will be restarted after the CONT.
283 	 */
284 	error = 0;
285 	pgrp = NULL;
286 	if ((flags & RFPGLOCK) && (pgrp = p1->p_pgrp) != NULL) {
287 		lockmgr(&pgrp->pg_lock, LK_SHARED);
288 		if (CURSIG_NOBLOCK(lp1)) {
289 			error = ERESTART;
290 			goto done;
291 		}
292 	}
293 
294 	/*
295 	 * Although process entries are dynamically created, we still keep
296 	 * a global limit on the maximum number we will create.  Don't allow
297 	 * a nonprivileged user to use the last ten processes; don't let root
298 	 * exceed the limit. The variable nprocs is the current number of
299 	 * processes, maxproc is the limit.
300 	 */
301 	uid = p1->p_ucred->cr_ruid;
302 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
303 		if (ppsratecheck(&lastfail, &curfail, 1))
304 			kprintf("maxproc limit exceeded by uid %d, please "
305 			       "see tuning(7) and login.conf(5).\n", uid);
306 		tsleep(&forksleep, 0, "fork", hz / 2);
307 		error = EAGAIN;
308 		goto done;
309 	}
310 	/*
311 	 * Increment the nprocs resource before blocking can occur.  There
312 	 * are hard-limits as to the number of processes that can run.
313 	 */
314 	nprocs++;
315 
316 	/*
317 	 * Increment the count of procs running with this uid. Don't allow
318 	 * a nonprivileged user to exceed their current limit.
319 	 */
320 	ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
321 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
322 	if (!ok) {
323 		/*
324 		 * Back out the process count
325 		 */
326 		nprocs--;
327 		if (ppsratecheck(&lastfail, &curfail, 1))
328 			kprintf("maxproc limit exceeded by uid %d, please "
329 			       "see tuning(7) and login.conf(5).\n", uid);
330 		tsleep(&forksleep, 0, "fork", hz / 2);
331 		error = EAGAIN;
332 		goto done;
333 	}
334 
335 	/* Allocate new proc. */
336 	p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
337 
338 	/*
339 	 * Setup linkage for kernel based threading XXX lwp
340 	 */
341 	if (flags & RFTHREAD) {
342 		p2->p_peers = p1->p_peers;
343 		p1->p_peers = p2;
344 		p2->p_leader = p1->p_leader;
345 	} else {
346 		p2->p_leader = p2;
347 	}
348 
349 	RB_INIT(&p2->p_lwp_tree);
350 	spin_init(&p2->p_spin);
351 	p2->p_lasttid = -1;	/* first tid will be 0 */
352 
353 	/*
354 	 * Setting the state to SIDL protects the partially initialized
355 	 * process once it starts getting hooked into the rest of the system.
356 	 */
357 	p2->p_stat = SIDL;
358 	proc_add_allproc(p2);
359 
360 	/*
361 	 * Make a proc table entry for the new process.
362 	 * The whole structure was zeroed above, so copy the section that is
363 	 * copied directly from the parent.
364 	 */
365 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
366 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
367 
368 	/*
369 	 * Duplicate sub-structures as needed.
370 	 * Increase reference counts on shared objects.
371 	 */
372 	if (p1->p_flag & P_PROFIL)
373 		startprofclock(p2);
374 	p2->p_ucred = crhold(p1->p_ucred);
375 	if (p2->p_lock)
376 		kprintf("Debug: p_lock race averted\n");
377 	p2->p_lock = 0;
378 
379 	if (jailed(p2->p_ucred))
380 		p2->p_flag |= P_JAILED;
381 
382 	if (p2->p_args)
383 		p2->p_args->ar_ref++;
384 
385 	p2->p_usched = p1->p_usched;
386 
387 	if (flags & RFSIGSHARE) {
388 		p2->p_sigacts = p1->p_sigacts;
389 		p2->p_sigacts->ps_refcnt++;
390 	} else {
391 		p2->p_sigacts = (struct sigacts *)kmalloc(sizeof(*p2->p_sigacts),
392 		    M_SUBPROC, M_WAITOK);
393 		bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
394 		p2->p_sigacts->ps_refcnt = 1;
395 	}
396 	if (flags & RFLINUXTHPN)
397 	        p2->p_sigparent = SIGUSR1;
398 	else
399 	        p2->p_sigparent = SIGCHLD;
400 
401 	/* bump references to the text vnode (for procfs) */
402 	p2->p_textvp = p1->p_textvp;
403 	if (p2->p_textvp)
404 		vref(p2->p_textvp);
405 
406 	/*
407 	 * Handle file descriptors
408 	 */
409 	if (flags & RFCFDG) {
410 		p2->p_fd = fdinit(p1);
411 		fdtol = NULL;
412 	} else if (flags & RFFDG) {
413 		p2->p_fd = fdcopy(p1);
414 		fdtol = NULL;
415 	} else {
416 		p2->p_fd = fdshare(p1);
417 		if (p1->p_fdtol == NULL)
418 			p1->p_fdtol =
419 				filedesc_to_leader_alloc(NULL,
420 							 p1->p_leader);
421 		if ((flags & RFTHREAD) != 0) {
422 			/*
423 			 * Shared file descriptor table and
424 			 * shared process leaders.
425 			 */
426 			fdtol = p1->p_fdtol;
427 			fdtol->fdl_refcount++;
428 		} else {
429 			/*
430 			 * Shared file descriptor table, and
431 			 * different process leaders
432 			 */
433 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
434 		}
435 	}
436 	p2->p_fdtol = fdtol;
437 	p2->p_limit = plimit_fork(p1);
438 
439 	/*
440 	 * Preserve some more flags in subprocess.  P_PROFIL has already
441 	 * been preserved.
442 	 */
443 	p2->p_flag |= p1->p_flag & P_SUGID;
444 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
445 		p2->p_flag |= P_CONTROLT;
446 	if (flags & RFPPWAIT)
447 		p2->p_flag |= P_PPWAIT;
448 
449 	/*
450 	 * Inherit the virtual kernel structure (allows a virtual kernel
451 	 * to fork to simulate multiple cpus).
452 	 */
453 	if (p1->p_vkernel)
454 		vkernel_inherit(p1, p2);
455 
456 	/*
457 	 * Once we are on a pglist we may receive signals.  XXX we might
458 	 * race a ^C being sent to the process group by not receiving it
459 	 * at all prior to this line.
460 	 */
461 	LIST_INSERT_AFTER(p1, p2, p_pglist);
462 
463 	/*
464 	 * Attach the new process to its parent.
465 	 *
466 	 * If RFNOWAIT is set, the newly created process becomes a child
467 	 * of init.  This effectively disassociates the child from the
468 	 * parent.
469 	 */
470 	if (flags & RFNOWAIT)
471 		pptr = initproc;
472 	else
473 		pptr = p1;
474 	p2->p_pptr = pptr;
475 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
476 	LIST_INIT(&p2->p_children);
477 	varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
478 	callout_init(&p2->p_ithandle);
479 
480 #ifdef KTRACE
481 	/*
482 	 * Copy traceflag and tracefile if enabled.  If not inherited,
483 	 * these were zeroed above but we still could have a trace race
484 	 * so make sure p2's p_tracenode is NULL.
485 	 */
486 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
487 		p2->p_traceflag = p1->p_traceflag;
488 		p2->p_tracenode = ktrinherit(p1->p_tracenode);
489 	}
490 #endif
491 
492 	/*
493 	 * This begins the section where we must prevent the parent
494 	 * from being swapped.
495 	 *
496 	 * Gets PRELE'd in the caller in start_forked_proc().
497 	 */
498 	PHOLD(p1);
499 
500 	vm_fork(p1, p2, flags);
501 
502 	/*
503 	 * Create the first lwp associated with the new proc.
504 	 * It will return via a different execution path later, directly
505 	 * into userland, after it was put on the runq by
506 	 * start_forked_proc().
507 	 */
508 	lwp_fork(lp1, p2, flags);
509 
510 	if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
511 		mycpu->gd_cnt.v_forks++;
512 		mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
513 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
514 		mycpu->gd_cnt.v_vforks++;
515 		mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
516 	} else if (p1 == &proc0) {
517 		mycpu->gd_cnt.v_kthreads++;
518 		mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
519 	} else {
520 		mycpu->gd_cnt.v_rforks++;
521 		mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
522 	}
523 
524 	/*
525 	 * Both processes are set up, now check if any loadable modules want
526 	 * to adjust anything.
527 	 *   What if they have an error? XXX
528 	 */
529 	TAILQ_FOREACH(ep, &fork_list, next) {
530 		(*ep->function)(p1, p2, flags);
531 	}
532 
533 	/*
534 	 * Set the start time.  Note that the process is not runnable.  The
535 	 * caller is responsible for making it runnable.
536 	 */
537 	microtime(&p2->p_start);
538 	p2->p_acflag = AFORK;
539 
540 	/*
541 	 * tell any interested parties about the new process
542 	 */
543 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
544 
545 	/*
546 	 * Return child proc pointer to parent.
547 	 */
548 	*procp = p2;
549 done:
550 	if (pgrp)
551 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
552 	return (error);
553 }
554 
555 static struct lwp *
556 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
557 {
558 	struct lwp *lp;
559 	struct thread *td;
560 
561 	lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
562 
563 	lp->lwp_proc = destproc;
564 	lp->lwp_vmspace = destproc->p_vmspace;
565 	lp->lwp_stat = LSRUN;
566 	bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
567 	    (unsigned) ((caddr_t)&lp->lwp_endcopy -
568 			(caddr_t)&lp->lwp_startcopy));
569 	lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
570 	/*
571 	 * Set cpbase to the last timeout that occured (not the upcoming
572 	 * timeout).
573 	 *
574 	 * A critical section is required since a timer IPI can update
575 	 * scheduler specific data.
576 	 */
577 	crit_enter();
578 	lp->lwp_cpbase = mycpu->gd_schedclock.time -
579 			mycpu->gd_schedclock.periodic;
580 	destproc->p_usched->heuristic_forking(origlp, lp);
581 	crit_exit();
582 	lp->lwp_cpumask &= usched_mastermask;
583 
584 	/*
585 	 * Assign a TID to the lp.  Loop until the insert succeeds (returns
586 	 * NULL).
587 	 */
588 	lp->lwp_tid = destproc->p_lasttid;
589 	do {
590 		if (++lp->lwp_tid < 0)
591 			lp->lwp_tid = 1;
592 	} while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
593 	destproc->p_lasttid = lp->lwp_tid;
594 	destproc->p_nthreads++;
595 
596 	td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
597 	lp->lwp_thread = td;
598 	td->td_proc = destproc;
599 	td->td_lwp = lp;
600 	td->td_switch = cpu_heavy_switch;
601 #ifdef SMP
602 	KKASSERT(td->td_mpcount == 1);
603 #endif
604 	lwkt_setpri(td, TDPRI_KERN_USER);
605 	lwkt_set_comm(td, "%s", destproc->p_comm);
606 
607 	/*
608 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
609 	 * and make the child ready to run.
610 	 */
611 	cpu_fork(origlp, lp, flags);
612 	caps_fork(origlp->lwp_thread, lp->lwp_thread);
613 
614 	return (lp);
615 }
616 
617 /*
618  * The next two functionms are general routines to handle adding/deleting
619  * items on the fork callout list.
620  *
621  * at_fork():
622  * Take the arguments given and put them onto the fork callout list,
623  * However first make sure that it's not already there.
624  * Returns 0 on success or a standard error number.
625  */
626 int
627 at_fork(forklist_fn function)
628 {
629 	struct forklist *ep;
630 
631 #ifdef INVARIANTS
632 	/* let the programmer know if he's been stupid */
633 	if (rm_at_fork(function)) {
634 		kprintf("WARNING: fork callout entry (%p) already present\n",
635 		    function);
636 	}
637 #endif
638 	ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
639 	ep->function = function;
640 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
641 	return (0);
642 }
643 
644 /*
645  * Scan the exit callout list for the given item and remove it..
646  * Returns the number of items removed (0 or 1)
647  */
648 int
649 rm_at_fork(forklist_fn function)
650 {
651 	struct forklist *ep;
652 
653 	TAILQ_FOREACH(ep, &fork_list, next) {
654 		if (ep->function == function) {
655 			TAILQ_REMOVE(&fork_list, ep, next);
656 			kfree(ep, M_ATFORK);
657 			return(1);
658 		}
659 	}
660 	return (0);
661 }
662 
663 /*
664  * Add a forked process to the run queue after any remaining setup, such
665  * as setting the fork handler, has been completed.
666  */
667 void
668 start_forked_proc(struct lwp *lp1, struct proc *p2)
669 {
670 	struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
671 
672 	/*
673 	 * Move from SIDL to RUN queue, and activate the process's thread.
674 	 * Activation of the thread effectively makes the process "a"
675 	 * current process, so we do not setrunqueue().
676 	 *
677 	 * YYY setrunqueue works here but we should clean up the trampoline
678 	 * code so we just schedule the LWKT thread and let the trampoline
679 	 * deal with the userland scheduler on return to userland.
680 	 */
681 	KASSERT(p2->p_stat == SIDL,
682 	    ("cannot start forked process, bad status: %p", p2));
683 	p2->p_usched->resetpriority(lp2);
684 	crit_enter();
685 	p2->p_stat = SACTIVE;
686 	lp2->lwp_stat = LSRUN;
687 	p2->p_usched->setrunqueue(lp2);
688 	crit_exit();
689 
690 	/*
691 	 * Now can be swapped.
692 	 */
693 	PRELE(lp1->lwp_proc);
694 
695 	/*
696 	 * Preserve synchronization semantics of vfork.  If waiting for
697 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
698 	 * proc (in case of exit).
699 	 */
700 	while (p2->p_flag & P_PPWAIT)
701 		tsleep(lp1->lwp_proc, 0, "ppwait", 0);
702 }
703