xref: /freebsd/sys/kern/kern_fork.c (revision 35a472461a9878d41c8588aed731080efdc2ad24)
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  * $FreeBSD$
40  */
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/syscall.h>
55 #include <sys/vnode.h>
56 #include <sys/acct.h>
57 #include <sys/ktr.h>
58 #include <sys/ktrace.h>
59 #include <sys/kthread.h>
60 #include <sys/unistd.h>
61 #include <sys/jail.h>
62 #include <sys/sx.h>
63 
64 #include <vm/vm.h>
65 #include <sys/lock.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_map.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_zone.h>
70 
71 #include <sys/vmmeter.h>
72 #include <sys/user.h>
73 
74 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
75 
76 static int	fast_vfork = 1;
77 SYSCTL_INT(_kern, OID_AUTO, fast_vfork, CTLFLAG_RW, &fast_vfork, 0,
78     "flag to indicate whether we have a fast vfork()");
79 
80 /*
81  * These are the stuctures used to create a callout list for things to do
82  * when forking a process
83  */
84 struct forklist {
85 	forklist_fn function;
86 	TAILQ_ENTRY(forklist) next;
87 };
88 
89 static struct sx fork_list_lock;
90 
91 TAILQ_HEAD(forklist_head, forklist);
92 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
93 
94 #ifndef _SYS_SYSPROTO_H_
95 struct fork_args {
96 	int     dummy;
97 };
98 #endif
99 
100 static void
101 init_fork_list(void *data __unused)
102 {
103 
104 	sx_init(&fork_list_lock, "fork list");
105 }
106 SYSINIT(fork_list, SI_SUB_INTRINSIC, SI_ORDER_ANY, init_fork_list, NULL);
107 
108 /* ARGSUSED */
109 int
110 fork(p, uap)
111 	struct proc *p;
112 	struct fork_args *uap;
113 {
114 	int error;
115 	struct proc *p2;
116 
117 	error = fork1(p, RFFDG | RFPROC, &p2);
118 	if (error == 0) {
119 		p->p_retval[0] = p2->p_pid;
120 		p->p_retval[1] = 0;
121 	}
122 	return error;
123 }
124 
125 /* ARGSUSED */
126 int
127 vfork(p, uap)
128 	struct proc *p;
129 	struct vfork_args *uap;
130 {
131 	int error;
132 	struct proc *p2;
133 
134 	error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
135 	if (error == 0) {
136 		p->p_retval[0] = p2->p_pid;
137 		p->p_retval[1] = 0;
138 	}
139 	return error;
140 }
141 
142 int
143 rfork(p, uap)
144 	struct proc *p;
145 	struct rfork_args *uap;
146 {
147 	int error;
148 	struct proc *p2;
149 
150 	/* mask kernel only flags out of the user flags */
151 	error = fork1(p, uap->flags & ~RFKERNELONLY, &p2);
152 	if (error == 0) {
153 		p->p_retval[0] = p2 ? p2->p_pid : 0;
154 		p->p_retval[1] = 0;
155 	}
156 	return error;
157 }
158 
159 
160 int	nprocs = 1;				/* process 0 */
161 static int nextpid = 0;
162 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &nextpid, 0,
163     "Last used PID");
164 
165 /*
166  * Random component to nextpid generation.  We mix in a random factor to make
167  * it a little harder to predict.  We sanity check the modulus value to avoid
168  * doing it in critical paths.  Don't let it be too small or we pointlessly
169  * waste randomness entropy, and don't let it be impossibly large.  Using a
170  * modulus that is too big causes a LOT more process table scans and slows
171  * down fork processing as the pidchecked caching is defeated.
172  */
173 static int randompid = 0;
174 
175 static int
176 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
177 {
178 	int error, pid;
179 
180 	pid = randompid;
181 	error = sysctl_handle_int(oidp, &pid, 0, req);
182 	if (error || !req->newptr)
183 		return (error);
184 	if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
185 		pid = PID_MAX - 100;
186 	else if (pid < 2)			/* NOP */
187 		pid = 0;
188 	else if (pid < 100)			/* Make it reasonable */
189 		pid = 100;
190 	randompid = pid;
191 	return (error);
192 }
193 
194 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
195     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
196 
197 int
198 fork1(p1, flags, procp)
199 	struct proc *p1;			/* parent proc */
200 	int flags;
201 	struct proc **procp;			/* child proc */
202 {
203 	struct proc *p2, *pptr;
204 	uid_t uid;
205 	struct proc *newproc;
206 	int trypid;
207 	int ok;
208 	static int pidchecked = 0;
209 	struct forklist *ep;
210 	struct filedesc *fd;
211 
212 	/* Can't copy and clear */
213 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
214 		return (EINVAL);
215 
216 	/*
217 	 * Here we don't create a new process, but we divorce
218 	 * certain parts of a process from itself.
219 	 */
220 	if ((flags & RFPROC) == 0) {
221 
222 		vm_fork(p1, 0, flags);
223 
224 		/*
225 		 * Close all file descriptors.
226 		 */
227 		if (flags & RFCFDG) {
228 			struct filedesc *fdtmp;
229 			fdtmp = fdinit(p1);
230 			PROC_LOCK(p1);
231 			fdfree(p1);
232 			p1->p_fd = fdtmp;
233 			PROC_UNLOCK(p1);
234 		}
235 
236 		/*
237 		 * Unshare file descriptors (from parent.)
238 		 */
239 		if (flags & RFFDG) {
240 			if (p1->p_fd->fd_refcnt > 1) {
241 				struct filedesc *newfd;
242 				newfd = fdcopy(p1);
243 				PROC_LOCK(p1);
244 				fdfree(p1);
245 				p1->p_fd = newfd;
246 				PROC_UNLOCK(p1);
247 			}
248 		}
249 		*procp = NULL;
250 		return (0);
251 	}
252 
253 	/*
254 	 * Although process entries are dynamically created, we still keep
255 	 * a global limit on the maximum number we will create.  Don't allow
256 	 * a nonprivileged user to use the last process; don't let root
257 	 * exceed the limit. The variable nprocs is the current number of
258 	 * processes, maxproc is the limit.
259 	 */
260 	uid = p1->p_cred->p_ruid;
261 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
262 		tablefull("proc");
263 		return (EAGAIN);
264 	}
265 	/*
266 	 * Increment the nprocs resource before blocking can occur.  There
267 	 * are hard-limits as to the number of processes that can run.
268 	 */
269 	nprocs++;
270 
271 	/*
272 	 * Increment the count of procs running with this uid. Don't allow
273 	 * a nonprivileged user to exceed their current limit.
274 	 */
275 	ok = chgproccnt(p1->p_cred->p_uidinfo, 1,
276 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
277 	if (!ok) {
278 		/*
279 		 * Back out the process count
280 		 */
281 		nprocs--;
282 		return (EAGAIN);
283 	}
284 
285 	/* Allocate new proc. */
286 	newproc = zalloc(proc_zone);
287 
288 	/*
289 	 * Setup linkage for kernel based threading
290 	 */
291 	if((flags & RFTHREAD) != 0) {
292 		newproc->p_peers = p1->p_peers;
293 		p1->p_peers = newproc;
294 		newproc->p_leader = p1->p_leader;
295 	} else {
296 		newproc->p_peers = NULL;
297 		newproc->p_leader = newproc;
298 	}
299 
300 	newproc->p_vmspace = NULL;
301 
302 	/*
303 	 * Find an unused process ID.  We remember a range of unused IDs
304 	 * ready to use (from nextpid+1 through pidchecked-1).
305 	 *
306 	 * If RFHIGHPID is set (used during system boot), do not allocate
307 	 * low-numbered pids.
308 	 */
309 	ALLPROC_LOCK(AP_EXCLUSIVE);
310 	trypid = nextpid + 1;
311 	if (flags & RFHIGHPID) {
312 		if (trypid < 10) {
313 			trypid = 10;
314 		}
315 	} else {
316 		if (randompid)
317 			trypid += arc4random() % randompid;
318 	}
319 retry:
320 	/*
321 	 * If the process ID prototype has wrapped around,
322 	 * restart somewhat above 0, as the low-numbered procs
323 	 * tend to include daemons that don't exit.
324 	 */
325 	if (trypid >= PID_MAX) {
326 		trypid = trypid % PID_MAX;
327 		if (trypid < 100)
328 			trypid += 100;
329 		pidchecked = 0;
330 	}
331 	if (trypid >= pidchecked) {
332 		int doingzomb = 0;
333 
334 		pidchecked = PID_MAX;
335 		/*
336 		 * Scan the active and zombie procs to check whether this pid
337 		 * is in use.  Remember the lowest pid that's greater
338 		 * than trypid, so we can avoid checking for a while.
339 		 */
340 		p2 = LIST_FIRST(&allproc);
341 again:
342 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
343 			while (p2->p_pid == trypid ||
344 			    p2->p_pgrp->pg_id == trypid ||
345 			    p2->p_session->s_sid == trypid) {
346 				trypid++;
347 				if (trypid >= pidchecked)
348 					goto retry;
349 			}
350 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
351 				pidchecked = p2->p_pid;
352 			if (p2->p_pgrp->pg_id > trypid &&
353 			    pidchecked > p2->p_pgrp->pg_id)
354 				pidchecked = p2->p_pgrp->pg_id;
355 			if (p2->p_session->s_sid > trypid &&
356 			    pidchecked > p2->p_session->s_sid)
357 				pidchecked = p2->p_session->s_sid;
358 		}
359 		if (!doingzomb) {
360 			doingzomb = 1;
361 			p2 = LIST_FIRST(&zombproc);
362 			goto again;
363 		}
364 	}
365 
366 	/*
367 	 * RFHIGHPID does not mess with the nextpid counter during boot.
368 	 */
369 	if (flags & RFHIGHPID)
370 		pidchecked = 0;
371 	else
372 		nextpid = trypid;
373 
374 	p2 = newproc;
375 	p2->p_intr_nesting_level = 0;
376 	p2->p_stat = SIDL;			/* protect against others */
377 	p2->p_pid = trypid;
378 	LIST_INSERT_HEAD(&allproc, p2, p_list);
379 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
380 	ALLPROC_LOCK(AP_RELEASE);
381 
382 	/*
383 	 * Make a proc table entry for the new process.
384 	 * Start by zeroing the section of proc that is zero-initialized,
385 	 * then copy the section that is copied directly from the parent.
386 	 */
387 	bzero(&p2->p_startzero,
388 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
389 	PROC_LOCK(p1);
390 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
391 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
392 	PROC_UNLOCK(p1);
393 
394 	mtx_init(&p2->p_mtx, "process lock", MTX_DEF);
395 	PROC_LOCK(p2);
396 	p2->p_aioinfo = NULL;
397 
398 	/*
399 	 * Duplicate sub-structures as needed.
400 	 * Increase reference counts on shared objects.
401 	 * The p_stats and p_sigacts substructs are set in vm_fork.
402 	 */
403 	p2->p_flag = 0;
404 	mtx_lock_spin(&sched_lock);
405 	p2->p_sflag = PS_INMEM;
406 	if (p1->p_sflag & PS_PROFIL)
407 		startprofclock(p2);
408 	mtx_unlock_spin(&sched_lock);
409 	/*
410 	 * We start off holding one spinlock after fork: sched_lock.
411 	 */
412 	p2->p_spinlocks = 1;
413 	PROC_UNLOCK(p2);
414 	MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred),
415 	    M_SUBPROC, M_WAITOK);
416 	PROC_LOCK(p2);
417 	PROC_LOCK(p1);
418 	bcopy(p1->p_cred, p2->p_cred, sizeof(*p2->p_cred));
419 	p2->p_cred->p_refcnt = 1;
420 	crhold(p1->p_ucred);
421 	uihold(p1->p_cred->p_uidinfo);
422 
423 	if (p2->p_args)
424 		p2->p_args->ar_ref++;
425 
426 	if (flags & RFSIGSHARE) {
427 		p2->p_procsig = p1->p_procsig;
428 		p2->p_procsig->ps_refcnt++;
429 		if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
430 			struct sigacts *newsigacts;
431 
432 			PROC_UNLOCK(p1);
433 			PROC_UNLOCK(p2);
434 			/* Create the shared sigacts structure */
435 			MALLOC(newsigacts, struct sigacts *,
436 			    sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
437 			PROC_LOCK(p2);
438 			PROC_LOCK(p1);
439 			/*
440 			 * Set p_sigacts to the new shared structure.
441 			 * Note that this is updating p1->p_sigacts at the
442 			 * same time, since p_sigacts is just a pointer to
443 			 * the shared p_procsig->ps_sigacts.
444 			 */
445 			p2->p_sigacts  = newsigacts;
446 			bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
447 			    sizeof(*p2->p_sigacts));
448 			*p2->p_sigacts = p1->p_addr->u_sigacts;
449 		}
450 	} else {
451 		PROC_UNLOCK(p1);
452 		PROC_UNLOCK(p2);
453 		MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
454 		    M_SUBPROC, M_WAITOK);
455 		PROC_LOCK(p2);
456 		PROC_LOCK(p1);
457 		bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
458 		p2->p_procsig->ps_refcnt = 1;
459 		p2->p_sigacts = NULL;	/* finished in vm_fork() */
460 	}
461 	if (flags & RFLINUXTHPN)
462 	        p2->p_sigparent = SIGUSR1;
463 	else
464 	        p2->p_sigparent = SIGCHLD;
465 
466 	/* bump references to the text vnode (for procfs) */
467 	p2->p_textvp = p1->p_textvp;
468 	PROC_UNLOCK(p1);
469 	PROC_UNLOCK(p2);
470 	if (p2->p_textvp)
471 		VREF(p2->p_textvp);
472 
473 	if (flags & RFCFDG)
474 		fd = fdinit(p1);
475 	else if (flags & RFFDG)
476 		fd = fdcopy(p1);
477 	else
478 		fd = fdshare(p1);
479 	PROC_LOCK(p2);
480 	p2->p_fd = fd;
481 
482 	/*
483 	 * If p_limit is still copy-on-write, bump refcnt,
484 	 * otherwise get a copy that won't be modified.
485 	 * (If PL_SHAREMOD is clear, the structure is shared
486 	 * copy-on-write.)
487 	 */
488 	PROC_LOCK(p1);
489 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
490 		p2->p_limit = limcopy(p1->p_limit);
491 	else {
492 		p2->p_limit = p1->p_limit;
493 		p2->p_limit->p_refcnt++;
494 	}
495 
496 	/*
497 	 * Preserve some more flags in subprocess.  PS_PROFIL has already
498 	 * been preserved.
499 	 */
500 	p2->p_flag |= p1->p_flag & P_SUGID;
501 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
502 		p2->p_flag |= P_CONTROLT;
503 	if (flags & RFPPWAIT)
504 		p2->p_flag |= P_PPWAIT;
505 
506 	LIST_INSERT_AFTER(p1, p2, p_pglist);
507 	PROC_UNLOCK(p1);
508 	PROC_UNLOCK(p2);
509 
510 	/*
511 	 * Attach the new process to its parent.
512 	 *
513 	 * If RFNOWAIT is set, the newly created process becomes a child
514 	 * of init.  This effectively disassociates the child from the
515 	 * parent.
516 	 */
517 	if (flags & RFNOWAIT)
518 		pptr = initproc;
519 	else
520 		pptr = p1;
521 	PROCTREE_LOCK(PT_EXCLUSIVE);
522 	PROC_LOCK(p2);
523 	p2->p_pptr = pptr;
524 	PROC_UNLOCK(p2);
525 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
526 	PROCTREE_LOCK(PT_RELEASE);
527 	PROC_LOCK(p2);
528 	LIST_INIT(&p2->p_children);
529 	LIST_INIT(&p2->p_heldmtx);
530 	LIST_INIT(&p2->p_contested);
531 
532 	callout_init(&p2->p_itcallout, 0);
533 	callout_init(&p2->p_slpcallout, 1);
534 
535 	PROC_LOCK(p1);
536 #ifdef KTRACE
537 	/*
538 	 * Copy traceflag and tracefile if enabled.
539 	 * If not inherited, these were zeroed above.
540 	 */
541 	if (p1->p_traceflag & KTRFAC_INHERIT) {
542 		p2->p_traceflag = p1->p_traceflag;
543 		if ((p2->p_tracep = p1->p_tracep) != NULL) {
544 			PROC_UNLOCK(p1);
545 			PROC_UNLOCK(p2);
546 			VREF(p2->p_tracep);
547 			PROC_LOCK(p2);
548 			PROC_LOCK(p1);
549 		}
550 	}
551 #endif
552 
553 	/*
554 	 * set priority of child to be that of parent
555 	 */
556 	mtx_lock_spin(&sched_lock);
557 	p2->p_estcpu = p1->p_estcpu;
558 	mtx_unlock_spin(&sched_lock);
559 
560 	/*
561 	 * This begins the section where we must prevent the parent
562 	 * from being swapped.
563 	 */
564 	_PHOLD(p1);
565 	PROC_UNLOCK(p1);
566 	PROC_UNLOCK(p2);
567 
568 	/*
569 	 * Finish creating the child process.  It will return via a different
570 	 * execution path later.  (ie: directly into user mode)
571 	 */
572 	vm_fork(p1, p2, flags);
573 
574 	if (flags == (RFFDG | RFPROC)) {
575 		cnt.v_forks++;
576 		cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
577 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
578 		cnt.v_vforks++;
579 		cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
580 	} else if (p1 == &proc0) {
581 		cnt.v_kthreads++;
582 		cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
583 	} else {
584 		cnt.v_rforks++;
585 		cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
586 	}
587 
588 	/*
589 	 * Both processes are set up, now check if any loadable modules want
590 	 * to adjust anything.
591 	 *   What if they have an error? XXX
592 	 */
593 	sx_slock(&fork_list_lock);
594 	TAILQ_FOREACH(ep, &fork_list, next) {
595 		(*ep->function)(p1, p2, flags);
596 	}
597 	sx_sunlock(&fork_list_lock);
598 
599 	/*
600 	 * If RFSTOPPED not requested, make child runnable and add to
601 	 * run queue.
602 	 */
603 	microtime(&(p2->p_stats->p_start));
604 	p2->p_acflag = AFORK;
605 	if ((flags & RFSTOPPED) == 0) {
606 		mtx_lock_spin(&sched_lock);
607 		p2->p_stat = SRUN;
608 		setrunqueue(p2);
609 		mtx_unlock_spin(&sched_lock);
610 	}
611 
612 	/*
613 	 * Now can be swapped.
614 	 */
615 	PROC_LOCK(p1);
616 	_PRELE(p1);
617 
618 	/*
619 	 * tell any interested parties about the new process
620 	 */
621 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
622 	PROC_UNLOCK(p1);
623 
624 	/*
625 	 * Preserve synchronization semantics of vfork.  If waiting for
626 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
627 	 * proc (in case of exit).
628 	 */
629 	PROC_LOCK(p2);
630 	while (p2->p_flag & P_PPWAIT)
631 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
632 	PROC_UNLOCK(p2);
633 
634 	/*
635 	 * Return child proc pointer to parent.
636 	 */
637 	*procp = p2;
638 	return (0);
639 }
640 
641 /*
642  * The next two functionms are general routines to handle adding/deleting
643  * items on the fork callout list.
644  *
645  * at_fork():
646  * Take the arguments given and put them onto the fork callout list,
647  * However first make sure that it's not already there.
648  * Returns 0 on success or a standard error number.
649  */
650 
651 int
652 at_fork(function)
653 	forklist_fn function;
654 {
655 	struct forklist *ep;
656 
657 #ifdef INVARIANTS
658 	/* let the programmer know if he's been stupid */
659 	if (rm_at_fork(function))
660 		printf("WARNING: fork callout entry (%p) already present\n",
661 		    function);
662 #endif
663 	ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
664 	if (ep == NULL)
665 		return (ENOMEM);
666 	ep->function = function;
667 	sx_xlock(&fork_list_lock);
668 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
669 	sx_xunlock(&fork_list_lock);
670 	return (0);
671 }
672 
673 /*
674  * Scan the exit callout list for the given item and remove it..
675  * Returns the number of items removed (0 or 1)
676  */
677 
678 int
679 rm_at_fork(function)
680 	forklist_fn function;
681 {
682 	struct forklist *ep;
683 
684 	sx_xlock(&fork_list_lock);
685 	TAILQ_FOREACH(ep, &fork_list, next) {
686 		if (ep->function == function) {
687 			TAILQ_REMOVE(&fork_list, ep, next);
688 			sx_xunlock(&fork_list_lock);
689 			free(ep, M_ATFORK);
690 			return(1);
691 		}
692 	}
693 	sx_xunlock(&fork_list_lock);
694 	return (0);
695 }
696 
697 /*
698  * Handle the return of a child process from fork1().  This function
699  * is called from the MD fork_trampoline() entry point.
700  */
701 void
702 fork_exit(callout, arg, frame)
703 	void (*callout)(void *, struct trapframe *);
704 	void *arg;
705 	struct trapframe *frame;
706 {
707 	struct proc *p;
708 
709 	p = curproc;
710 
711 	/*
712 	 * Setup the sched_lock state so that we can release it.
713 	 */
714 	sched_lock.mtx_lock = (uintptr_t)p;
715 	sched_lock.mtx_recurse = 0;
716 	/*
717 	 * XXX: We really shouldn't have to do this.
718 	 */
719 	mtx_intr_enable(&sched_lock);
720 	mtx_unlock_spin(&sched_lock);
721 
722 #ifdef SMP
723 	if (PCPU_GET(switchtime.tv_sec) == 0)
724 		microuptime(PCPU_PTR(switchtime));
725 	PCPU_SET(switchticks, ticks);
726 #endif
727 
728 	/*
729 	 * cpu_set_fork_handler intercepts this function call to
730          * have this call a non-return function to stay in kernel mode.
731          * initproc has its own fork handler, but it does return.
732          */
733 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
734 	callout(arg, frame);
735 
736 	/*
737 	 * Check if a kernel thread misbehaved and returned from its main
738 	 * function.
739 	 */
740 	PROC_LOCK(p);
741 	if (p->p_flag & P_KTHREAD) {
742 		PROC_UNLOCK(p);
743 		mtx_lock(&Giant);
744 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
745 		    p->p_comm, p->p_pid);
746 		kthread_exit(0);
747 	}
748 	PROC_UNLOCK(p);
749 	mtx_assert(&Giant, MA_NOTOWNED);
750 }
751 
752 /*
753  * Simplified back end of syscall(), used when returning from fork()
754  * directly into user mode.  Giant is not held on entry, and must not
755  * be held on return.  This function is passed in to fork_exit() as the
756  * first parameter and is called when returning to a new userland process.
757  */
758 void
759 fork_return(p, frame)
760 	struct proc *p;
761 	struct trapframe *frame;
762 {
763 
764 	userret(p, frame, 0);
765 #ifdef KTRACE
766 	if (KTRPOINT(p, KTR_SYSRET)) {
767 		if (!mtx_owned(&Giant))
768 			mtx_lock(&Giant);
769 		ktrsysret(p->p_tracep, SYS_fork, 0, 0);
770 	}
771 #endif
772 	if (mtx_owned(&Giant))
773 		mtx_unlock(&Giant);
774 	mtx_assert(&Giant, MA_NOTOWNED);
775 }
776