xref: /dflybsd-src/sys/kern/kern_fork.c (revision 2381583524bd7679b9e475f7ce722874d1eea618)
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 1.76 2008/05/09 06:38:19 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/thread2.h>
70 #include <sys/signal2.h>
71 #include <sys/spinlock2.h>
72 
73 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
74 
75 /*
76  * These are the stuctures used to create a callout list for things to do
77  * when forking a process
78  */
79 struct forklist {
80 	forklist_fn function;
81 	TAILQ_ENTRY(forklist) next;
82 };
83 
84 TAILQ_HEAD(forklist_head, forklist);
85 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
86 
87 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
88 
89 int forksleep; /* Place for fork1() to sleep on. */
90 
91 /*
92  * Red-Black tree support for LWPs
93  */
94 
95 static int
96 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
97 {
98 	if (lp1->lwp_tid < lp2->lwp_tid)
99 		return(-1);
100 	if (lp1->lwp_tid > lp2->lwp_tid)
101 		return(1);
102 	return(0);
103 }
104 
105 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
106 
107 
108 /* ARGSUSED */
109 int
110 sys_fork(struct fork_args *uap)
111 {
112 	struct lwp *lp = curthread->td_lwp;
113 	struct proc *p2;
114 	int error;
115 
116 	error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
117 	if (error == 0) {
118 		start_forked_proc(lp, p2);
119 		uap->sysmsg_fds[0] = p2->p_pid;
120 		uap->sysmsg_fds[1] = 0;
121 	}
122 	return error;
123 }
124 
125 /* ARGSUSED */
126 int
127 sys_vfork(struct vfork_args *uap)
128 {
129 	struct lwp *lp = curthread->td_lwp;
130 	struct proc *p2;
131 	int error;
132 
133 	error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
134 	if (error == 0) {
135 		start_forked_proc(lp, p2);
136 		uap->sysmsg_fds[0] = p2->p_pid;
137 		uap->sysmsg_fds[1] = 0;
138 	}
139 	return error;
140 }
141 
142 /*
143  * Handle rforks.  An rfork may (1) operate on the current process without
144  * creating a new, (2) create a new process that shared the current process's
145  * vmspace, signals, and/or descriptors, or (3) create a new process that does
146  * not share these things (normal fork).
147  *
148  * Note that we only call start_forked_proc() if a new process is actually
149  * created.
150  *
151  * rfork { int flags }
152  */
153 int
154 sys_rfork(struct rfork_args *uap)
155 {
156 	struct lwp *lp = curthread->td_lwp;
157 	struct proc *p2;
158 	int error;
159 
160 	if ((uap->flags & RFKERNELONLY) != 0)
161 		return (EINVAL);
162 
163 	error = fork1(lp, uap->flags | RFPGLOCK, &p2);
164 	if (error == 0) {
165 		if (p2)
166 			start_forked_proc(lp, p2);
167 		uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
168 		uap->sysmsg_fds[1] = 0;
169 	}
170 	return error;
171 }
172 
173 int
174 sys_lwp_create(struct lwp_create_args *uap)
175 {
176 	struct proc *p = curproc;
177 	struct lwp *lp;
178 	struct lwp_params params;
179 	int error;
180 
181 	error = copyin(uap->params, &params, sizeof(params));
182 	if (error)
183 		goto fail2;
184 
185 	plimit_lwp_fork(p);	/* force exclusive access */
186 	lp = lwp_fork(curthread->td_lwp, p, RFPROC);
187 	error = cpu_prepare_lwp(lp, &params);
188 	if (params.tid1 != NULL &&
189 	    (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
190 		goto fail;
191 	if (params.tid2 != NULL &&
192 	    (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
193 		goto fail;
194 
195 	/*
196 	 * Now schedule the new lwp.
197 	 */
198 	p->p_usched->resetpriority(lp);
199 	crit_enter();
200 	lp->lwp_stat = LSRUN;
201 	p->p_usched->setrunqueue(lp);
202 	crit_exit();
203 
204 	return (0);
205 
206 fail:
207 	lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
208 	--p->p_nthreads;
209 	/* lwp_dispose expects an exited lwp, and a held proc */
210 	lp->lwp_flag |= LWP_WEXIT;
211 	lp->lwp_thread->td_flags |= TDF_EXITING;
212 	PHOLD(p);
213 	lwp_dispose(lp);
214 fail2:
215 	return (error);
216 }
217 
218 int	nprocs = 1;		/* process 0 */
219 
220 int
221 fork1(struct lwp *lp1, int flags, struct proc **procp)
222 {
223 	struct proc *p1 = lp1->lwp_proc;
224 	struct proc *p2, *pptr;
225 	struct pgrp *pgrp;
226 	uid_t uid;
227 	int ok, error;
228 	static int curfail = 0;
229 	static struct timeval lastfail;
230 	struct forklist *ep;
231 	struct filedesc_to_leader *fdtol;
232 
233 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
234 		return (EINVAL);
235 
236 	/*
237 	 * Here we don't create a new process, but we divorce
238 	 * certain parts of a process from itself.
239 	 */
240 	if ((flags & RFPROC) == 0) {
241 		/*
242 		 * This kind of stunt does not work anymore if
243 		 * there are native threads (lwps) running
244 		 */
245 		if (p1->p_nthreads != 1)
246 			return (EINVAL);
247 
248 		vm_fork(p1, 0, flags);
249 
250 		/*
251 		 * Close all file descriptors.
252 		 */
253 		if (flags & RFCFDG) {
254 			struct filedesc *fdtmp;
255 			fdtmp = fdinit(p1);
256 			fdfree(p1);
257 			p1->p_fd = fdtmp;
258 		}
259 
260 		/*
261 		 * Unshare file descriptors (from parent.)
262 		 */
263 		if (flags & RFFDG) {
264 			if (p1->p_fd->fd_refcnt > 1) {
265 				struct filedesc *newfd;
266 				newfd = fdcopy(p1);
267 				fdfree(p1);
268 				p1->p_fd = newfd;
269 			}
270 		}
271 		*procp = NULL;
272 		return (0);
273 	}
274 
275 	/*
276 	 * Interlock against process group signal delivery.  If signals
277 	 * are pending after the interlock is obtained we have to restart
278 	 * the system call to process the signals.  If we don't the child
279 	 * can miss a pgsignal (such as ^C) sent during the fork.
280 	 *
281 	 * We can't use CURSIG() here because it will process any STOPs
282 	 * and cause the process group lock to be held indefinitely.  If
283 	 * a STOP occurs, the fork will be restarted after the CONT.
284 	 */
285 	error = 0;
286 	pgrp = NULL;
287 	if ((flags & RFPGLOCK) && (pgrp = p1->p_pgrp) != NULL) {
288 		lockmgr(&pgrp->pg_lock, LK_SHARED);
289 		if (CURSIGNB(lp1)) {
290 			error = ERESTART;
291 			goto done;
292 		}
293 	}
294 
295 	/*
296 	 * Although process entries are dynamically created, we still keep
297 	 * a global limit on the maximum number we will create.  Don't allow
298 	 * a nonprivileged user to use the last ten processes; don't let root
299 	 * exceed the limit. The variable nprocs is the current number of
300 	 * processes, maxproc is the limit.
301 	 */
302 	uid = p1->p_ucred->cr_ruid;
303 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
304 		if (ppsratecheck(&lastfail, &curfail, 1))
305 			kprintf("maxproc limit exceeded by uid %d, please "
306 			       "see tuning(7) and login.conf(5).\n", uid);
307 		tsleep(&forksleep, 0, "fork", hz / 2);
308 		error = EAGAIN;
309 		goto done;
310 	}
311 	/*
312 	 * Increment the nprocs resource before blocking can occur.  There
313 	 * are hard-limits as to the number of processes that can run.
314 	 */
315 	nprocs++;
316 
317 	/*
318 	 * Increment the count of procs running with this uid. Don't allow
319 	 * a nonprivileged user to exceed their current limit.
320 	 */
321 	ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
322 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
323 	if (!ok) {
324 		/*
325 		 * Back out the process count
326 		 */
327 		nprocs--;
328 		if (ppsratecheck(&lastfail, &curfail, 1))
329 			kprintf("maxproc limit exceeded by uid %d, please "
330 			       "see tuning(7) and login.conf(5).\n", uid);
331 		tsleep(&forksleep, 0, "fork", hz / 2);
332 		error = EAGAIN;
333 		goto done;
334 	}
335 
336 	/* Allocate new proc. */
337 	p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
338 
339 	/*
340 	 * Setup linkage for kernel based threading XXX lwp
341 	 */
342 	if (flags & RFTHREAD) {
343 		p2->p_peers = p1->p_peers;
344 		p1->p_peers = p2;
345 		p2->p_leader = p1->p_leader;
346 	} else {
347 		p2->p_leader = p2;
348 	}
349 
350 	RB_INIT(&p2->p_lwp_tree);
351 	spin_init(&p2->p_spin);
352 	p2->p_lasttid = -1;	/* first tid will be 0 */
353 
354 	/*
355 	 * Setting the state to SIDL protects the partially initialized
356 	 * process once it starts getting hooked into the rest of the system.
357 	 */
358 	p2->p_stat = SIDL;
359 	proc_add_allproc(p2);
360 
361 	/*
362 	 * Make a proc table entry for the new process.
363 	 * The whole structure was zeroed above, so copy the section that is
364 	 * copied directly from the parent.
365 	 */
366 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
367 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
368 
369 	/*
370 	 * Duplicate sub-structures as needed.
371 	 * Increase reference counts on shared objects.
372 	 */
373 	if (p1->p_flag & P_PROFIL)
374 		startprofclock(p2);
375 	p2->p_ucred = crhold(p1->p_ucred);
376 
377 	if (jailed(p2->p_ucred))
378 		p2->p_flag |= P_JAILED;
379 
380 	if (p2->p_args)
381 		p2->p_args->ar_ref++;
382 
383 	p2->p_usched = p1->p_usched;
384 
385 	if (flags & RFSIGSHARE) {
386 		p2->p_sigacts = p1->p_sigacts;
387 		p2->p_sigacts->ps_refcnt++;
388 	} else {
389 		p2->p_sigacts = (struct sigacts *)kmalloc(sizeof(*p2->p_sigacts),
390 		    M_SUBPROC, M_WAITOK);
391 		bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
392 		p2->p_sigacts->ps_refcnt = 1;
393 	}
394 	if (flags & RFLINUXTHPN)
395 	        p2->p_sigparent = SIGUSR1;
396 	else
397 	        p2->p_sigparent = SIGCHLD;
398 
399 	/* bump references to the text vnode (for procfs) */
400 	p2->p_textvp = p1->p_textvp;
401 	if (p2->p_textvp)
402 		vref(p2->p_textvp);
403 
404 	/*
405 	 * Handle file descriptors
406 	 */
407 	if (flags & RFCFDG) {
408 		p2->p_fd = fdinit(p1);
409 		fdtol = NULL;
410 	} else if (flags & RFFDG) {
411 		p2->p_fd = fdcopy(p1);
412 		fdtol = NULL;
413 	} else {
414 		p2->p_fd = fdshare(p1);
415 		if (p1->p_fdtol == NULL)
416 			p1->p_fdtol =
417 				filedesc_to_leader_alloc(NULL,
418 							 p1->p_leader);
419 		if ((flags & RFTHREAD) != 0) {
420 			/*
421 			 * Shared file descriptor table and
422 			 * shared process leaders.
423 			 */
424 			fdtol = p1->p_fdtol;
425 			fdtol->fdl_refcount++;
426 		} else {
427 			/*
428 			 * Shared file descriptor table, and
429 			 * different process leaders
430 			 */
431 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
432 		}
433 	}
434 	p2->p_fdtol = fdtol;
435 	p2->p_limit = plimit_fork(p1);
436 
437 	/*
438 	 * Preserve some more flags in subprocess.  P_PROFIL has already
439 	 * been preserved.
440 	 */
441 	p2->p_flag |= p1->p_flag & P_SUGID;
442 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
443 		p2->p_flag |= P_CONTROLT;
444 	if (flags & RFPPWAIT)
445 		p2->p_flag |= P_PPWAIT;
446 
447 	/*
448 	 * Inherit the virtual kernel structure (allows a virtual kernel
449 	 * to fork to simulate multiple cpus).
450 	 */
451 	if (p1->p_vkernel)
452 		vkernel_inherit(p1, p2);
453 
454 	/*
455 	 * Once we are on a pglist we may receive signals.  XXX we might
456 	 * race a ^C being sent to the process group by not receiving it
457 	 * at all prior to this line.
458 	 */
459 	LIST_INSERT_AFTER(p1, p2, p_pglist);
460 
461 	/*
462 	 * Attach the new process to its parent.
463 	 *
464 	 * If RFNOWAIT is set, the newly created process becomes a child
465 	 * of init.  This effectively disassociates the child from the
466 	 * parent.
467 	 */
468 	if (flags & RFNOWAIT)
469 		pptr = initproc;
470 	else
471 		pptr = p1;
472 	p2->p_pptr = pptr;
473 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
474 	LIST_INIT(&p2->p_children);
475 	varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
476 	callout_init(&p2->p_ithandle);
477 
478 #ifdef KTRACE
479 	/*
480 	 * Copy traceflag and tracefile if enabled.  If not inherited,
481 	 * these were zeroed above but we still could have a trace race
482 	 * so make sure p2's p_tracenode is NULL.
483 	 */
484 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
485 		p2->p_traceflag = p1->p_traceflag;
486 		p2->p_tracenode = ktrinherit(p1->p_tracenode);
487 	}
488 #endif
489 
490 	/*
491 	 * This begins the section where we must prevent the parent
492 	 * from being swapped.
493 	 *
494 	 * Gets PRELE'd in the caller in start_forked_proc().
495 	 */
496 	PHOLD(p1);
497 
498 	vm_fork(p1, p2, flags);
499 
500 	/*
501 	 * Create the first lwp associated with the new proc.
502 	 * It will return via a different execution path later, directly
503 	 * into userland, after it was put on the runq by
504 	 * start_forked_proc().
505 	 */
506 	lwp_fork(lp1, p2, flags);
507 
508 	if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
509 		mycpu->gd_cnt.v_forks++;
510 		mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
511 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
512 		mycpu->gd_cnt.v_vforks++;
513 		mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
514 	} else if (p1 == &proc0) {
515 		mycpu->gd_cnt.v_kthreads++;
516 		mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
517 	} else {
518 		mycpu->gd_cnt.v_rforks++;
519 		mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
520 	}
521 
522 	/*
523 	 * Both processes are set up, now check if any loadable modules want
524 	 * to adjust anything.
525 	 *   What if they have an error? XXX
526 	 */
527 	TAILQ_FOREACH(ep, &fork_list, next) {
528 		(*ep->function)(p1, p2, flags);
529 	}
530 
531 	/*
532 	 * Set the start time.  Note that the process is not runnable.  The
533 	 * caller is responsible for making it runnable.
534 	 */
535 	microtime(&p2->p_start);
536 	p2->p_acflag = AFORK;
537 
538 	/*
539 	 * tell any interested parties about the new process
540 	 */
541 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
542 
543 	/*
544 	 * Return child proc pointer to parent.
545 	 */
546 	*procp = p2;
547 done:
548 	if (pgrp)
549 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
550 	return (error);
551 }
552 
553 static struct lwp *
554 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
555 {
556 	struct lwp *lp;
557 	struct thread *td;
558 
559 	lp = zalloc(lwp_zone);
560 	bzero(lp, sizeof(*lp));
561 
562 	lp->lwp_proc = destproc;
563 	lp->lwp_vmspace = destproc->p_vmspace;
564 	lp->lwp_stat = LSRUN;
565 	bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
566 	    (unsigned) ((caddr_t)&lp->lwp_endcopy -
567 			(caddr_t)&lp->lwp_startcopy));
568 	lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
569 	/*
570 	 * Set cpbase to the last timeout that occured (not the upcoming
571 	 * timeout).
572 	 *
573 	 * A critical section is required since a timer IPI can update
574 	 * scheduler specific data.
575 	 */
576 	crit_enter();
577 	lp->lwp_cpbase = mycpu->gd_schedclock.time -
578 			mycpu->gd_schedclock.periodic;
579 	destproc->p_usched->heuristic_forking(origlp, lp);
580 	crit_exit();
581 	lp->lwp_cpumask &= usched_mastermask;
582 
583 	/*
584 	 * Assign a TID to the lp.  Loop until the insert succeeds (returns
585 	 * NULL).
586 	 */
587 	lp->lwp_tid = destproc->p_lasttid;
588 	do {
589 		if (++lp->lwp_tid < 0)
590 			lp->lwp_tid = 1;
591 	} while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
592 	destproc->p_lasttid = lp->lwp_tid;
593 	destproc->p_nthreads++;
594 
595 	td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
596 	lp->lwp_thread = td;
597 	td->td_proc = destproc;
598 	td->td_lwp = lp;
599 	td->td_switch = cpu_heavy_switch;
600 #ifdef SMP
601 	KKASSERT(td->td_mpcount == 1);
602 #endif
603 	lwkt_setpri(td, TDPRI_KERN_USER);
604 	lwkt_set_comm(td, "%s", destproc->p_comm);
605 
606 	/*
607 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
608 	 * and make the child ready to run.
609 	 */
610 	cpu_fork(origlp, lp, flags);
611 	caps_fork(origlp->lwp_thread, lp->lwp_thread);
612 
613 	return (lp);
614 }
615 
616 /*
617  * The next two functionms are general routines to handle adding/deleting
618  * items on the fork callout list.
619  *
620  * at_fork():
621  * Take the arguments given and put them onto the fork callout list,
622  * However first make sure that it's not already there.
623  * Returns 0 on success or a standard error number.
624  */
625 int
626 at_fork(forklist_fn function)
627 {
628 	struct forklist *ep;
629 
630 #ifdef INVARIANTS
631 	/* let the programmer know if he's been stupid */
632 	if (rm_at_fork(function)) {
633 		kprintf("WARNING: fork callout entry (%p) already present\n",
634 		    function);
635 	}
636 #endif
637 	ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
638 	ep->function = function;
639 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
640 	return (0);
641 }
642 
643 /*
644  * Scan the exit callout list for the given item and remove it..
645  * Returns the number of items removed (0 or 1)
646  */
647 int
648 rm_at_fork(forklist_fn function)
649 {
650 	struct forklist *ep;
651 
652 	TAILQ_FOREACH(ep, &fork_list, next) {
653 		if (ep->function == function) {
654 			TAILQ_REMOVE(&fork_list, ep, next);
655 			kfree(ep, M_ATFORK);
656 			return(1);
657 		}
658 	}
659 	return (0);
660 }
661 
662 /*
663  * Add a forked process to the run queue after any remaining setup, such
664  * as setting the fork handler, has been completed.
665  */
666 void
667 start_forked_proc(struct lwp *lp1, struct proc *p2)
668 {
669 	struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
670 
671 	/*
672 	 * Move from SIDL to RUN queue, and activate the process's thread.
673 	 * Activation of the thread effectively makes the process "a"
674 	 * current process, so we do not setrunqueue().
675 	 *
676 	 * YYY setrunqueue works here but we should clean up the trampoline
677 	 * code so we just schedule the LWKT thread and let the trampoline
678 	 * deal with the userland scheduler on return to userland.
679 	 */
680 	KASSERT(p2->p_stat == SIDL,
681 	    ("cannot start forked process, bad status: %p", p2));
682 	p2->p_usched->resetpriority(lp2);
683 	crit_enter();
684 	p2->p_stat = SACTIVE;
685 	lp2->lwp_stat = LSRUN;
686 	p2->p_usched->setrunqueue(lp2);
687 	crit_exit();
688 
689 	/*
690 	 * Now can be swapped.
691 	 */
692 	PRELE(lp1->lwp_proc);
693 
694 	/*
695 	 * Preserve synchronization semantics of vfork.  If waiting for
696 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
697 	 * proc (in case of exit).
698 	 */
699 	while (p2->p_flag & P_PPWAIT)
700 		tsleep(lp1->lwp_proc, 0, "ppwait", 0);
701 }
702