xref: /dflybsd-src/sys/kern/kern_fork.c (revision 90ea502b8c5d21f908cedff6680ee2bc9e74ce74)
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  * Fork system call
108  *
109  * MPALMOSTSAFE
110  */
111 int
112 sys_fork(struct fork_args *uap)
113 {
114 	struct lwp *lp = curthread->td_lwp;
115 	struct proc *p2;
116 	int error;
117 
118 	get_mplock();
119 	error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
120 	if (error == 0) {
121 		start_forked_proc(lp, p2);
122 		uap->sysmsg_fds[0] = p2->p_pid;
123 		uap->sysmsg_fds[1] = 0;
124 	}
125 	rel_mplock();
126 	return error;
127 }
128 
129 /*
130  * MPALMOSTSAFE
131  */
132 int
133 sys_vfork(struct vfork_args *uap)
134 {
135 	struct lwp *lp = curthread->td_lwp;
136 	struct proc *p2;
137 	int error;
138 
139 	get_mplock();
140 	error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
141 	if (error == 0) {
142 		start_forked_proc(lp, p2);
143 		uap->sysmsg_fds[0] = p2->p_pid;
144 		uap->sysmsg_fds[1] = 0;
145 	}
146 	rel_mplock();
147 	return error;
148 }
149 
150 /*
151  * Handle rforks.  An rfork may (1) operate on the current process without
152  * creating a new, (2) create a new process that shared the current process's
153  * vmspace, signals, and/or descriptors, or (3) create a new process that does
154  * not share these things (normal fork).
155  *
156  * Note that we only call start_forked_proc() if a new process is actually
157  * created.
158  *
159  * rfork { int flags }
160  *
161  * MPALMOSTSAFE
162  */
163 int
164 sys_rfork(struct rfork_args *uap)
165 {
166 	struct lwp *lp = curthread->td_lwp;
167 	struct proc *p2;
168 	int error;
169 
170 	if ((uap->flags & RFKERNELONLY) != 0)
171 		return (EINVAL);
172 
173 	get_mplock();
174 	error = fork1(lp, uap->flags | RFPGLOCK, &p2);
175 	if (error == 0) {
176 		if (p2)
177 			start_forked_proc(lp, p2);
178 		uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
179 		uap->sysmsg_fds[1] = 0;
180 	}
181 	rel_mplock();
182 	return error;
183 }
184 
185 /*
186  * MPALMOSTSAFE
187  */
188 int
189 sys_lwp_create(struct lwp_create_args *uap)
190 {
191 	struct proc *p = curproc;
192 	struct lwp *lp;
193 	struct lwp_params params;
194 	int error;
195 
196 	error = copyin(uap->params, &params, sizeof(params));
197 	if (error)
198 		goto fail2;
199 
200 	get_mplock();
201 	plimit_lwp_fork(p);	/* force exclusive access */
202 	lp = lwp_fork(curthread->td_lwp, p, RFPROC);
203 	error = cpu_prepare_lwp(lp, &params);
204 	if (params.tid1 != NULL &&
205 	    (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
206 		goto fail;
207 	if (params.tid2 != NULL &&
208 	    (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
209 		goto fail;
210 
211 	/*
212 	 * Now schedule the new lwp.
213 	 */
214 	p->p_usched->resetpriority(lp);
215 	crit_enter();
216 	lp->lwp_stat = LSRUN;
217 	p->p_usched->setrunqueue(lp);
218 	crit_exit();
219 	rel_mplock();
220 
221 	return (0);
222 
223 fail:
224 	lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
225 	--p->p_nthreads;
226 	/* lwp_dispose expects an exited lwp, and a held proc */
227 	lp->lwp_flag |= LWP_WEXIT;
228 	lp->lwp_thread->td_flags |= TDF_EXITING;
229 	PHOLD(p);
230 	lwp_dispose(lp);
231 	rel_mplock();
232 fail2:
233 	return (error);
234 }
235 
236 int	nprocs = 1;		/* process 0 */
237 
238 int
239 fork1(struct lwp *lp1, int flags, struct proc **procp)
240 {
241 	struct proc *p1 = lp1->lwp_proc;
242 	struct proc *p2, *pptr;
243 	struct pgrp *pgrp;
244 	uid_t uid;
245 	int ok, error;
246 	static int curfail = 0;
247 	static struct timeval lastfail;
248 	struct forklist *ep;
249 	struct filedesc_to_leader *fdtol;
250 
251 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
252 		return (EINVAL);
253 
254 	/*
255 	 * Here we don't create a new process, but we divorce
256 	 * certain parts of a process from itself.
257 	 */
258 	if ((flags & RFPROC) == 0) {
259 		/*
260 		 * This kind of stunt does not work anymore if
261 		 * there are native threads (lwps) running
262 		 */
263 		if (p1->p_nthreads != 1)
264 			return (EINVAL);
265 
266 		vm_fork(p1, 0, flags);
267 
268 		/*
269 		 * Close all file descriptors.
270 		 */
271 		if (flags & RFCFDG) {
272 			struct filedesc *fdtmp;
273 			fdtmp = fdinit(p1);
274 			fdfree(p1, fdtmp);
275 		}
276 
277 		/*
278 		 * Unshare file descriptors (from parent.)
279 		 */
280 		if (flags & RFFDG) {
281 			if (p1->p_fd->fd_refcnt > 1) {
282 				struct filedesc *newfd;
283 				newfd = fdcopy(p1);
284 				fdfree(p1, newfd);
285 			}
286 		}
287 		*procp = NULL;
288 		return (0);
289 	}
290 
291 	/*
292 	 * Interlock against process group signal delivery.  If signals
293 	 * are pending after the interlock is obtained we have to restart
294 	 * the system call to process the signals.  If we don't the child
295 	 * can miss a pgsignal (such as ^C) sent during the fork.
296 	 *
297 	 * We can't use CURSIG() here because it will process any STOPs
298 	 * and cause the process group lock to be held indefinitely.  If
299 	 * a STOP occurs, the fork will be restarted after the CONT.
300 	 */
301 	error = 0;
302 	pgrp = NULL;
303 	if ((flags & RFPGLOCK) && (pgrp = p1->p_pgrp) != NULL) {
304 		lockmgr(&pgrp->pg_lock, LK_SHARED);
305 		if (CURSIG_NOBLOCK(lp1)) {
306 			error = ERESTART;
307 			goto done;
308 		}
309 	}
310 
311 	/*
312 	 * Although process entries are dynamically created, we still keep
313 	 * a global limit on the maximum number we will create.  Don't allow
314 	 * a nonprivileged user to use the last ten processes; don't let root
315 	 * exceed the limit. The variable nprocs is the current number of
316 	 * processes, maxproc is the limit.
317 	 */
318 	uid = lp1->lwp_thread->td_ucred->cr_ruid;
319 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
320 		if (ppsratecheck(&lastfail, &curfail, 1))
321 			kprintf("maxproc limit exceeded by uid %d, please "
322 			       "see tuning(7) and login.conf(5).\n", uid);
323 		tsleep(&forksleep, 0, "fork", hz / 2);
324 		error = EAGAIN;
325 		goto done;
326 	}
327 	/*
328 	 * Increment the nprocs resource before blocking can occur.  There
329 	 * are hard-limits as to the number of processes that can run.
330 	 */
331 	nprocs++;
332 
333 	/*
334 	 * Increment the count of procs running with this uid. Don't allow
335 	 * a nonprivileged user to exceed their current limit.
336 	 */
337 	ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
338 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
339 	if (!ok) {
340 		/*
341 		 * Back out the process count
342 		 */
343 		nprocs--;
344 		if (ppsratecheck(&lastfail, &curfail, 1))
345 			kprintf("maxproc limit exceeded by uid %d, please "
346 			       "see tuning(7) and login.conf(5).\n", uid);
347 		tsleep(&forksleep, 0, "fork", hz / 2);
348 		error = EAGAIN;
349 		goto done;
350 	}
351 
352 	/* Allocate new proc. */
353 	p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
354 
355 	/*
356 	 * Setup linkage for kernel based threading XXX lwp
357 	 */
358 	if (flags & RFTHREAD) {
359 		p2->p_peers = p1->p_peers;
360 		p1->p_peers = p2;
361 		p2->p_leader = p1->p_leader;
362 	} else {
363 		p2->p_leader = p2;
364 	}
365 
366 	RB_INIT(&p2->p_lwp_tree);
367 	spin_init(&p2->p_spin);
368 	p2->p_lasttid = -1;	/* first tid will be 0 */
369 
370 	/*
371 	 * Setting the state to SIDL protects the partially initialized
372 	 * process once it starts getting hooked into the rest of the system.
373 	 */
374 	p2->p_stat = SIDL;
375 	proc_add_allproc(p2);
376 
377 	/*
378 	 * Make a proc table entry for the new process.
379 	 * The whole structure was zeroed above, so copy the section that is
380 	 * copied directly from the parent.
381 	 */
382 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
383 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
384 
385 	/*
386 	 * Duplicate sub-structures as needed.
387 	 * Increase reference counts on shared objects.
388 	 */
389 	if (p1->p_flag & P_PROFIL)
390 		startprofclock(p2);
391 	p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
392 	KKASSERT(p2->p_lock == 0);
393 
394 	if (jailed(p2->p_ucred))
395 		p2->p_flag |= P_JAILED;
396 
397 	if (p2->p_args)
398 		p2->p_args->ar_ref++;
399 
400 	p2->p_usched = p1->p_usched;
401 
402 	if (flags & RFSIGSHARE) {
403 		p2->p_sigacts = p1->p_sigacts;
404 		p2->p_sigacts->ps_refcnt++;
405 	} else {
406 		p2->p_sigacts = (struct sigacts *)kmalloc(sizeof(*p2->p_sigacts),
407 		    M_SUBPROC, M_WAITOK);
408 		bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
409 		p2->p_sigacts->ps_refcnt = 1;
410 	}
411 	if (flags & RFLINUXTHPN)
412 	        p2->p_sigparent = SIGUSR1;
413 	else
414 	        p2->p_sigparent = SIGCHLD;
415 
416 	/* bump references to the text vnode (for procfs) */
417 	p2->p_textvp = p1->p_textvp;
418 	if (p2->p_textvp)
419 		vref(p2->p_textvp);
420 
421 	/*
422 	 * Handle file descriptors
423 	 */
424 	if (flags & RFCFDG) {
425 		p2->p_fd = fdinit(p1);
426 		fdtol = NULL;
427 	} else if (flags & RFFDG) {
428 		p2->p_fd = fdcopy(p1);
429 		fdtol = NULL;
430 	} else {
431 		p2->p_fd = fdshare(p1);
432 		if (p1->p_fdtol == NULL)
433 			p1->p_fdtol =
434 				filedesc_to_leader_alloc(NULL,
435 							 p1->p_leader);
436 		if ((flags & RFTHREAD) != 0) {
437 			/*
438 			 * Shared file descriptor table and
439 			 * shared process leaders.
440 			 */
441 			fdtol = p1->p_fdtol;
442 			fdtol->fdl_refcount++;
443 		} else {
444 			/*
445 			 * Shared file descriptor table, and
446 			 * different process leaders
447 			 */
448 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
449 		}
450 	}
451 	p2->p_fdtol = fdtol;
452 	p2->p_limit = plimit_fork(p1);
453 
454 	/*
455 	 * Preserve some more flags in subprocess.  P_PROFIL has already
456 	 * been preserved.
457 	 */
458 	p2->p_flag |= p1->p_flag & P_SUGID;
459 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
460 		p2->p_flag |= P_CONTROLT;
461 	if (flags & RFPPWAIT)
462 		p2->p_flag |= P_PPWAIT;
463 
464 	/*
465 	 * Inherit the virtual kernel structure (allows a virtual kernel
466 	 * to fork to simulate multiple cpus).
467 	 */
468 	if (p1->p_vkernel)
469 		vkernel_inherit(p1, p2);
470 
471 	/*
472 	 * Once we are on a pglist we may receive signals.  XXX we might
473 	 * race a ^C being sent to the process group by not receiving it
474 	 * at all prior to this line.
475 	 */
476 	LIST_INSERT_AFTER(p1, p2, p_pglist);
477 
478 	/*
479 	 * Attach the new process to its parent.
480 	 *
481 	 * If RFNOWAIT is set, the newly created process becomes a child
482 	 * of init.  This effectively disassociates the child from the
483 	 * parent.
484 	 */
485 	if (flags & RFNOWAIT)
486 		pptr = initproc;
487 	else
488 		pptr = p1;
489 	p2->p_pptr = pptr;
490 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
491 	LIST_INIT(&p2->p_children);
492 	varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
493 	callout_init(&p2->p_ithandle);
494 
495 #ifdef KTRACE
496 	/*
497 	 * Copy traceflag and tracefile if enabled.  If not inherited,
498 	 * these were zeroed above but we still could have a trace race
499 	 * so make sure p2's p_tracenode is NULL.
500 	 */
501 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
502 		p2->p_traceflag = p1->p_traceflag;
503 		p2->p_tracenode = ktrinherit(p1->p_tracenode);
504 	}
505 #endif
506 
507 	/*
508 	 * This begins the section where we must prevent the parent
509 	 * from being swapped.
510 	 *
511 	 * Gets PRELE'd in the caller in start_forked_proc().
512 	 */
513 	PHOLD(p1);
514 
515 	vm_fork(p1, p2, flags);
516 
517 	/*
518 	 * Create the first lwp associated with the new proc.
519 	 * It will return via a different execution path later, directly
520 	 * into userland, after it was put on the runq by
521 	 * start_forked_proc().
522 	 */
523 	lwp_fork(lp1, p2, flags);
524 
525 	if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
526 		mycpu->gd_cnt.v_forks++;
527 		mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
528 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
529 		mycpu->gd_cnt.v_vforks++;
530 		mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
531 	} else if (p1 == &proc0) {
532 		mycpu->gd_cnt.v_kthreads++;
533 		mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
534 	} else {
535 		mycpu->gd_cnt.v_rforks++;
536 		mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
537 	}
538 
539 	/*
540 	 * Both processes are set up, now check if any loadable modules want
541 	 * to adjust anything.
542 	 *   What if they have an error? XXX
543 	 */
544 	TAILQ_FOREACH(ep, &fork_list, next) {
545 		(*ep->function)(p1, p2, flags);
546 	}
547 
548 	/*
549 	 * Set the start time.  Note that the process is not runnable.  The
550 	 * caller is responsible for making it runnable.
551 	 */
552 	microtime(&p2->p_start);
553 	p2->p_acflag = AFORK;
554 
555 	/*
556 	 * tell any interested parties about the new process
557 	 */
558 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
559 
560 	/*
561 	 * Return child proc pointer to parent.
562 	 */
563 	*procp = p2;
564 done:
565 	if (pgrp)
566 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
567 	return (error);
568 }
569 
570 static struct lwp *
571 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
572 {
573 	struct lwp *lp;
574 	struct thread *td;
575 
576 	lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
577 
578 	lp->lwp_proc = destproc;
579 	lp->lwp_vmspace = destproc->p_vmspace;
580 	lp->lwp_stat = LSRUN;
581 	bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
582 	    (unsigned) ((caddr_t)&lp->lwp_endcopy -
583 			(caddr_t)&lp->lwp_startcopy));
584 	lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
585 	/*
586 	 * Set cpbase to the last timeout that occured (not the upcoming
587 	 * timeout).
588 	 *
589 	 * A critical section is required since a timer IPI can update
590 	 * scheduler specific data.
591 	 */
592 	crit_enter();
593 	lp->lwp_cpbase = mycpu->gd_schedclock.time -
594 			mycpu->gd_schedclock.periodic;
595 	destproc->p_usched->heuristic_forking(origlp, lp);
596 	crit_exit();
597 	lp->lwp_cpumask &= usched_mastermask;
598 
599 	/*
600 	 * Assign a TID to the lp.  Loop until the insert succeeds (returns
601 	 * NULL).
602 	 */
603 	lp->lwp_tid = destproc->p_lasttid;
604 	do {
605 		if (++lp->lwp_tid < 0)
606 			lp->lwp_tid = 1;
607 	} while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
608 	destproc->p_lasttid = lp->lwp_tid;
609 	destproc->p_nthreads++;
610 
611 	td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
612 	lp->lwp_thread = td;
613 	td->td_proc = destproc;
614 	td->td_lwp = lp;
615 	td->td_switch = cpu_heavy_switch;
616 #ifdef SMP
617 	KKASSERT(td->td_mpcount == 1);
618 #endif
619 	lwkt_setpri(td, TDPRI_KERN_USER);
620 	lwkt_set_comm(td, "%s", destproc->p_comm);
621 
622 	/*
623 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
624 	 * and make the child ready to run.
625 	 */
626 	cpu_fork(origlp, lp, flags);
627 	caps_fork(origlp->lwp_thread, lp->lwp_thread);
628 
629 	return (lp);
630 }
631 
632 /*
633  * The next two functionms are general routines to handle adding/deleting
634  * items on the fork callout list.
635  *
636  * at_fork():
637  * Take the arguments given and put them onto the fork callout list,
638  * However first make sure that it's not already there.
639  * Returns 0 on success or a standard error number.
640  */
641 int
642 at_fork(forklist_fn function)
643 {
644 	struct forklist *ep;
645 
646 #ifdef INVARIANTS
647 	/* let the programmer know if he's been stupid */
648 	if (rm_at_fork(function)) {
649 		kprintf("WARNING: fork callout entry (%p) already present\n",
650 		    function);
651 	}
652 #endif
653 	ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
654 	ep->function = function;
655 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
656 	return (0);
657 }
658 
659 /*
660  * Scan the exit callout list for the given item and remove it..
661  * Returns the number of items removed (0 or 1)
662  */
663 int
664 rm_at_fork(forklist_fn function)
665 {
666 	struct forklist *ep;
667 
668 	TAILQ_FOREACH(ep, &fork_list, next) {
669 		if (ep->function == function) {
670 			TAILQ_REMOVE(&fork_list, ep, next);
671 			kfree(ep, M_ATFORK);
672 			return(1);
673 		}
674 	}
675 	return (0);
676 }
677 
678 /*
679  * Add a forked process to the run queue after any remaining setup, such
680  * as setting the fork handler, has been completed.
681  */
682 void
683 start_forked_proc(struct lwp *lp1, struct proc *p2)
684 {
685 	struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
686 
687 	/*
688 	 * Move from SIDL to RUN queue, and activate the process's thread.
689 	 * Activation of the thread effectively makes the process "a"
690 	 * current process, so we do not setrunqueue().
691 	 *
692 	 * YYY setrunqueue works here but we should clean up the trampoline
693 	 * code so we just schedule the LWKT thread and let the trampoline
694 	 * deal with the userland scheduler on return to userland.
695 	 */
696 	KASSERT(p2->p_stat == SIDL,
697 	    ("cannot start forked process, bad status: %p", p2));
698 	p2->p_usched->resetpriority(lp2);
699 	crit_enter();
700 	p2->p_stat = SACTIVE;
701 	lp2->lwp_stat = LSRUN;
702 	p2->p_usched->setrunqueue(lp2);
703 	crit_exit();
704 
705 	/*
706 	 * Now can be swapped.
707 	 */
708 	PRELE(lp1->lwp_proc);
709 
710 	/*
711 	 * Preserve synchronization semantics of vfork.  If waiting for
712 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
713 	 * proc (in case of exit).
714 	 */
715 	while (p2->p_flag & P_PPWAIT)
716 		tsleep(lp1->lwp_proc, 0, "ppwait", 0);
717 }
718