xref: /freebsd/sys/kern/kern_fork.c (revision 70fe064ad7cab6c0444b91622f60ec6a462f308a)
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/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/resourcevar.h>
55 #include <sys/syscall.h>
56 #include <sys/vnode.h>
57 #include <sys/acct.h>
58 #include <sys/ktr.h>
59 #include <sys/ktrace.h>
60 #include <sys/kthread.h>
61 #include <sys/unistd.h>
62 #include <sys/jail.h>
63 #include <sys/sx.h>
64 
65 #include <vm/vm.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 /*
77  * These are the stuctures used to create a callout list for things to do
78  * when forking a process
79  */
80 struct forklist {
81 	forklist_fn function;
82 	TAILQ_ENTRY(forklist) next;
83 };
84 
85 static struct sx fork_list_lock;
86 
87 TAILQ_HEAD(forklist_head, forklist);
88 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
89 
90 #ifndef _SYS_SYSPROTO_H_
91 struct fork_args {
92 	int     dummy;
93 };
94 #endif
95 
96 static void
97 init_fork_list(void *data __unused)
98 {
99 
100 	sx_init(&fork_list_lock, "fork list");
101 }
102 SYSINIT(fork_list, SI_SUB_INTRINSIC, SI_ORDER_ANY, init_fork_list, NULL);
103 
104 /*
105  * MPSAFE
106  */
107 /* ARGSUSED */
108 int
109 fork(td, uap)
110 	struct thread *td;
111 	struct fork_args *uap;
112 {
113 	int error;
114 	struct proc *p2;
115 
116 	mtx_lock(&Giant);
117 	error = fork1(td, RFFDG | RFPROC, &p2);
118 	if (error == 0) {
119 		td->td_retval[0] = p2->p_pid;
120 		td->td_retval[1] = 0;
121 	}
122 	mtx_unlock(&Giant);
123 	return error;
124 }
125 
126 /*
127  * MPSAFE
128  */
129 /* ARGSUSED */
130 int
131 vfork(td, uap)
132 	struct thread *td;
133 	struct vfork_args *uap;
134 {
135 	int error;
136 	struct proc *p2;
137 
138 	mtx_lock(&Giant);
139 	error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
140 	if (error == 0) {
141 		td->td_retval[0] = p2->p_pid;
142 		td->td_retval[1] = 0;
143 	}
144 	mtx_unlock(&Giant);
145 	return error;
146 }
147 
148 /*
149  * MPSAFE
150  */
151 int
152 rfork(td, uap)
153 	struct thread *td;
154 	struct rfork_args *uap;
155 {
156 	int error;
157 	struct proc *p2;
158 
159 	/* Don't allow kernel only flags. */
160 	if ((uap->flags & RFKERNELONLY) != 0)
161 		return (EINVAL);
162 	mtx_lock(&Giant);
163 	error = fork1(td, uap->flags, &p2);
164 	if (error == 0) {
165 		td->td_retval[0] = p2 ? p2->p_pid : 0;
166 		td->td_retval[1] = 0;
167 	}
168 	mtx_unlock(&Giant);
169 	return error;
170 }
171 
172 
173 int	nprocs = 1;				/* process 0 */
174 int	lastpid = 0;
175 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
176     "Last used PID");
177 
178 /*
179  * Random component to lastpid generation.  We mix in a random factor to make
180  * it a little harder to predict.  We sanity check the modulus value to avoid
181  * doing it in critical paths.  Don't let it be too small or we pointlessly
182  * waste randomness entropy, and don't let it be impossibly large.  Using a
183  * modulus that is too big causes a LOT more process table scans and slows
184  * down fork processing as the pidchecked caching is defeated.
185  */
186 static int randompid = 0;
187 
188 static int
189 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
190 {
191 	int error, pid;
192 
193 	pid = randompid;
194 	error = sysctl_handle_int(oidp, &pid, 0, req);
195 	if (error || !req->newptr)
196 		return (error);
197 	if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
198 		pid = PID_MAX - 100;
199 	else if (pid < 2)			/* NOP */
200 		pid = 0;
201 	else if (pid < 100)			/* Make it reasonable */
202 		pid = 100;
203 	randompid = pid;
204 	return (error);
205 }
206 
207 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
208     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
209 
210 #if 0
211 void
212 kse_init(struct kse *kse1, struct kse *kse2)
213 {
214 }
215 
216 void
217 thread_init(struct thread *thread1, struct thread *thread2)
218 {
219 }
220 
221 void
222 ksegrp_init(struct ksegrp *ksegrp1, struct ksegrp *ksegrp2)
223 {
224 }
225 #endif
226 
227 int
228 fork1(td, flags, procp)
229 	struct thread *td;			/* parent proc */
230 	int flags;
231 	struct proc **procp;			/* child proc */
232 {
233 	struct proc *p2, *pptr;
234 	uid_t uid;
235 	struct proc *newproc;
236 	int trypid;
237 	int ok;
238 	static int pidchecked = 0;
239 	struct forklist *ep;
240 	struct filedesc *fd;
241 	struct proc *p1 = td->td_proc;
242 
243 	GIANT_REQUIRED;
244 
245 	/* Can't copy and clear */
246 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
247 		return (EINVAL);
248 
249 	/*
250 	 * Here we don't create a new process, but we divorce
251 	 * certain parts of a process from itself.
252 	 */
253 	if ((flags & RFPROC) == 0) {
254 		vm_forkproc(td, 0, flags);
255 
256 		/*
257 		 * Close all file descriptors.
258 		 */
259 		if (flags & RFCFDG) {
260 			struct filedesc *fdtmp;
261 			fdtmp = fdinit(td);	/* XXXKSE */
262 			PROC_LOCK(p1);
263 			fdfree(td);		/* XXXKSE */
264 			p1->p_fd = fdtmp;
265 			PROC_UNLOCK(p1);
266 		}
267 
268 		/*
269 		 * Unshare file descriptors (from parent.)
270 		 */
271 		if (flags & RFFDG) {
272 			FILEDESC_LOCK(p1->p_fd);
273 			if (p1->p_fd->fd_refcnt > 1) {
274 				struct filedesc *newfd;
275 
276 				newfd = fdcopy(td);
277 				FILEDESC_UNLOCK(p1->p_fd);
278 				PROC_LOCK(p1);
279 				fdfree(td);
280 				p1->p_fd = newfd;
281 				PROC_UNLOCK(p1);
282 			} else
283 				FILEDESC_UNLOCK(p1->p_fd);
284 		}
285 		*procp = NULL;
286 		return (0);
287 	}
288 
289 	/*
290 	 * Although process entries are dynamically created, we still keep
291 	 * a global limit on the maximum number we will create.  Don't allow
292 	 * a nonprivileged user to use the last process; don't let root
293 	 * exceed the limit. The variable nprocs is the current number of
294 	 * processes, maxproc is the limit.
295 	 */
296 	uid = p1->p_ucred->cr_ruid;
297 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
298 		tablefull("proc");
299 		return (EAGAIN);
300 	}
301 	/*
302 	 * Increment the nprocs resource before blocking can occur.  There
303 	 * are hard-limits as to the number of processes that can run.
304 	 */
305 	nprocs++;
306 
307 	/*
308 	 * Increment the count of procs running with this uid. Don't allow
309 	 * a nonprivileged user to exceed their current limit.
310 	 */
311 	ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
312 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
313 	if (!ok) {
314 		/*
315 		 * Back out the process count
316 		 */
317 		nprocs--;
318 		return (EAGAIN);
319 	}
320 
321 	/* Allocate new proc. */
322 	newproc = zalloc(proc_zone);
323 
324 	/*
325 	 * Setup linkage for kernel based threading
326 	 */
327 	if((flags & RFTHREAD) != 0) {
328 		newproc->p_peers = p1->p_peers;
329 		p1->p_peers = newproc;
330 		newproc->p_leader = p1->p_leader;
331 	} else {
332 		newproc->p_peers = NULL;
333 		newproc->p_leader = newproc;
334 	}
335 
336 	newproc->p_vmspace = NULL;
337 
338 	/*
339 	 * Find an unused process ID.  We remember a range of unused IDs
340 	 * ready to use (from lastpid+1 through pidchecked-1).
341 	 *
342 	 * If RFHIGHPID is set (used during system boot), do not allocate
343 	 * low-numbered pids.
344 	 */
345 	sx_xlock(&allproc_lock);
346 	trypid = lastpid + 1;
347 	if (flags & RFHIGHPID) {
348 		if (trypid < 10) {
349 			trypid = 10;
350 		}
351 	} else {
352 		if (randompid)
353 			trypid += arc4random() % randompid;
354 	}
355 retry:
356 	/*
357 	 * If the process ID prototype has wrapped around,
358 	 * restart somewhat above 0, as the low-numbered procs
359 	 * tend to include daemons that don't exit.
360 	 */
361 	if (trypid >= PID_MAX) {
362 		trypid = trypid % PID_MAX;
363 		if (trypid < 100)
364 			trypid += 100;
365 		pidchecked = 0;
366 	}
367 	if (trypid >= pidchecked) {
368 		int doingzomb = 0;
369 
370 		pidchecked = PID_MAX;
371 		/*
372 		 * Scan the active and zombie procs to check whether this pid
373 		 * is in use.  Remember the lowest pid that's greater
374 		 * than trypid, so we can avoid checking for a while.
375 		 */
376 		p2 = LIST_FIRST(&allproc);
377 again:
378 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
379 			while (p2->p_pid == trypid ||
380 			    p2->p_pgrp->pg_id == trypid ||
381 			    p2->p_session->s_sid == trypid) {
382 				trypid++;
383 				if (trypid >= pidchecked)
384 					goto retry;
385 			}
386 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
387 				pidchecked = p2->p_pid;
388 			if (p2->p_pgrp->pg_id > trypid &&
389 			    pidchecked > p2->p_pgrp->pg_id)
390 				pidchecked = p2->p_pgrp->pg_id;
391 			if (p2->p_session->s_sid > trypid &&
392 			    pidchecked > p2->p_session->s_sid)
393 				pidchecked = p2->p_session->s_sid;
394 		}
395 		if (!doingzomb) {
396 			doingzomb = 1;
397 			p2 = LIST_FIRST(&zombproc);
398 			goto again;
399 		}
400 	}
401 
402 	/*
403 	 * RFHIGHPID does not mess with the lastpid counter during boot.
404 	 */
405 	if (flags & RFHIGHPID)
406 		pidchecked = 0;
407 	else
408 		lastpid = trypid;
409 
410 	p2 = newproc;
411 	p2->p_stat = SIDL;			/* protect against others */
412 	p2->p_pid = trypid;
413 	LIST_INSERT_HEAD(&allproc, p2, p_list);
414 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
415 	sx_xunlock(&allproc_lock);
416 
417 	/*
418 	 * Make a proc table entry for the new process.
419 	 * Start by zeroing the section of proc that is zero-initialized,
420 	 * then copy the section that is copied directly from the parent.
421 	 */
422 	bzero(&p2->p_startzero,
423 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
424 	bzero(&p2->p_kse.ke_startzero,
425 	    (unsigned) ((caddr_t)&p2->p_kse.ke_endzero
426 			- (caddr_t)&p2->p_kse.ke_startzero));
427 	bzero(&p2->p_thread.td_startzero,
428 	    (unsigned) ((caddr_t)&p2->p_thread.td_endzero
429 			- (caddr_t)&p2->p_thread.td_startzero));
430 	bzero(&p2->p_ksegrp.kg_startzero,
431 	    (unsigned) ((caddr_t)&p2->p_ksegrp.kg_endzero
432 			- (caddr_t)&p2->p_ksegrp.kg_startzero));
433 	PROC_LOCK(p1);
434 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
435 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
436 
437 	bcopy(&p1->p_kse.ke_startcopy, &p2->p_kse.ke_startcopy,
438 	    (unsigned) ((caddr_t)&p2->p_kse.ke_endcopy
439 			- (caddr_t)&p2->p_kse.ke_startcopy));
440 
441 	bcopy(&p1->p_thread.td_startcopy, &p2->p_thread.td_startcopy,
442 	    (unsigned) ((caddr_t)&p2->p_thread.td_endcopy
443 			- (caddr_t)&p2->p_thread.td_startcopy));
444 
445 	bcopy(&p1->p_ksegrp.kg_startcopy, &p2->p_ksegrp.kg_startcopy,
446 	    (unsigned) ((caddr_t)&p2->p_ksegrp.kg_endcopy
447 			- (caddr_t)&p2->p_ksegrp.kg_startcopy));
448 	PROC_UNLOCK(p1);
449 
450 	/*
451 	 * XXXKSE Theoretically only the running thread would get copied
452 	 * Others in the kernel would be 'aborted' in the child.
453 	 * i.e return E*something*
454 	 */
455 	proc_linkup(p2);
456 
457 	mtx_init(&p2->p_mtx, "process lock", MTX_DEF);
458 	PROC_LOCK(p2);
459 	/* note.. XXXKSE no pcb or u-area yet */
460 
461 	/*
462 	 * Duplicate sub-structures as needed.
463 	 * Increase reference counts on shared objects.
464 	 * The p_stats and p_sigacts substructs are set in vm_forkproc.
465 	 */
466 	p2->p_flag = 0;
467 	mtx_lock_spin(&sched_lock);
468 	p2->p_sflag = PS_INMEM;
469 	if (p1->p_sflag & PS_PROFIL)
470 		startprofclock(p2);
471 	mtx_unlock_spin(&sched_lock);
472 	PROC_LOCK(p1);
473 	p2->p_ucred = crhold(p1->p_ucred);
474 	p2->p_thread.td_ucred = crhold(p2->p_ucred);	/* XXXKSE */
475 
476 	if (p2->p_args)
477 		p2->p_args->ar_ref++;
478 
479 	if (flags & RFSIGSHARE) {
480 		p2->p_procsig = p1->p_procsig;
481 		p2->p_procsig->ps_refcnt++;
482 		if (p1->p_sigacts == &p1->p_uarea->u_sigacts) {
483 			struct sigacts *newsigacts;
484 
485 			PROC_UNLOCK(p1);
486 			PROC_UNLOCK(p2);
487 			/* Create the shared sigacts structure */
488 			MALLOC(newsigacts, struct sigacts *,
489 			    sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
490 			PROC_LOCK(p2);
491 			PROC_LOCK(p1);
492 			/*
493 			 * Set p_sigacts to the new shared structure.
494 			 * Note that this is updating p1->p_sigacts at the
495 			 * same time, since p_sigacts is just a pointer to
496 			 * the shared p_procsig->ps_sigacts.
497 			 */
498 			p2->p_sigacts  = newsigacts;
499 			*p2->p_sigacts = p1->p_uarea->u_sigacts;
500 		}
501 	} else {
502 		PROC_UNLOCK(p1);
503 		PROC_UNLOCK(p2);
504 		MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
505 		    M_SUBPROC, M_WAITOK);
506 		PROC_LOCK(p2);
507 		PROC_LOCK(p1);
508 		bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
509 		p2->p_procsig->ps_refcnt = 1;
510 		p2->p_sigacts = NULL;	/* finished in vm_forkproc() */
511 	}
512 	if (flags & RFLINUXTHPN)
513 	        p2->p_sigparent = SIGUSR1;
514 	else
515 	        p2->p_sigparent = SIGCHLD;
516 
517 	/* bump references to the text vnode (for procfs) */
518 	p2->p_textvp = p1->p_textvp;
519 	PROC_UNLOCK(p1);
520 	PROC_UNLOCK(p2);
521 	if (p2->p_textvp)
522 		VREF(p2->p_textvp);
523 
524 	if (flags & RFCFDG)
525 		fd = fdinit(td);
526 	else if (flags & RFFDG) {
527 		FILEDESC_LOCK(p1->p_fd);
528 		fd = fdcopy(td);
529 		FILEDESC_UNLOCK(p1->p_fd);
530 	} else
531 		fd = fdshare(p1);
532 	PROC_LOCK(p2);
533 	p2->p_fd = fd;
534 
535 	/*
536 	 * If p_limit is still copy-on-write, bump refcnt,
537 	 * otherwise get a copy that won't be modified.
538 	 * (If PL_SHAREMOD is clear, the structure is shared
539 	 * copy-on-write.)
540 	 */
541 	PROC_LOCK(p1);
542 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
543 		p2->p_limit = limcopy(p1->p_limit);
544 	else {
545 		p2->p_limit = p1->p_limit;
546 		p2->p_limit->p_refcnt++;
547 	}
548 
549 	/*
550 	 * Preserve some more flags in subprocess.  PS_PROFIL has already
551 	 * been preserved.
552 	 */
553 	p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
554 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
555 		p2->p_flag |= P_CONTROLT;
556 	if (flags & RFPPWAIT)
557 		p2->p_flag |= P_PPWAIT;
558 
559 	LIST_INSERT_AFTER(p1, p2, p_pglist);
560 	PROC_UNLOCK(p1);
561 	PROC_UNLOCK(p2);
562 
563 	/*
564 	 * Attach the new process to its parent.
565 	 *
566 	 * If RFNOWAIT is set, the newly created process becomes a child
567 	 * of init.  This effectively disassociates the child from the
568 	 * parent.
569 	 */
570 	if (flags & RFNOWAIT)
571 		pptr = initproc;
572 	else
573 		pptr = p1;
574 	sx_xlock(&proctree_lock);
575 	PROC_LOCK(p2);
576 	p2->p_pptr = pptr;
577 	PROC_UNLOCK(p2);
578 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
579 	sx_xunlock(&proctree_lock);
580 	PROC_LOCK(p2);
581 	LIST_INIT(&p2->p_children);
582 	LIST_INIT(&p2->p_thread.td_contested); /* XXXKSE only 1 thread? */
583 
584 	callout_init(&p2->p_itcallout, 0);
585 	callout_init(&p2->p_thread.td_slpcallout, 1); /* XXXKSE */
586 
587 	PROC_LOCK(p1);
588 #ifdef KTRACE
589 	/*
590 	 * Copy traceflag and tracefile if enabled.  If not inherited,
591 	 * these were zeroed above but we still could have a trace race
592 	 * so make sure p2's p_tracep is NULL.
593 	 */
594 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
595 		p2->p_traceflag = p1->p_traceflag;
596 		if ((p2->p_tracep = p1->p_tracep) != NULL) {
597 			PROC_UNLOCK(p1);
598 			PROC_UNLOCK(p2);
599 			VREF(p2->p_tracep);
600 			PROC_LOCK(p2);
601 			PROC_LOCK(p1);
602 		}
603 	}
604 #endif
605 
606 	/*
607 	 * set priority of child to be that of parent
608 	 * XXXKSE hey! copying the estcpu seems dodgy.. should split it..
609 	 */
610 	mtx_lock_spin(&sched_lock);
611 	p2->p_ksegrp.kg_estcpu = p1->p_ksegrp.kg_estcpu;
612 	mtx_unlock_spin(&sched_lock);
613 
614 	/*
615 	 * This begins the section where we must prevent the parent
616 	 * from being swapped.
617 	 */
618 	_PHOLD(p1);
619 	PROC_UNLOCK(p1);
620 	PROC_UNLOCK(p2);
621 
622 	/*
623 	 * Finish creating the child process.  It will return via a different
624 	 * execution path later.  (ie: directly into user mode)
625 	 */
626 	vm_forkproc(td, p2, flags);
627 
628 	if (flags == (RFFDG | RFPROC)) {
629 		cnt.v_forks++;
630 		cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
631 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
632 		cnt.v_vforks++;
633 		cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
634 	} else if (p1 == &proc0) {
635 		cnt.v_kthreads++;
636 		cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
637 	} else {
638 		cnt.v_rforks++;
639 		cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
640 	}
641 
642 	/*
643 	 * Both processes are set up, now check if any loadable modules want
644 	 * to adjust anything.
645 	 *   What if they have an error? XXX
646 	 */
647 	sx_slock(&fork_list_lock);
648 	TAILQ_FOREACH(ep, &fork_list, next) {
649 		(*ep->function)(p1, p2, flags);
650 	}
651 	sx_sunlock(&fork_list_lock);
652 
653 	/*
654 	 * If RFSTOPPED not requested, make child runnable and add to
655 	 * run queue.
656 	 */
657 	microtime(&(p2->p_stats->p_start));
658 	p2->p_acflag = AFORK;
659 	if ((flags & RFSTOPPED) == 0) {
660 		mtx_lock_spin(&sched_lock);
661 		p2->p_stat = SRUN;
662 		setrunqueue(&p2->p_thread);
663 		mtx_unlock_spin(&sched_lock);
664 	}
665 
666 	/*
667 	 * Now can be swapped.
668 	 */
669 	PROC_LOCK(p1);
670 	_PRELE(p1);
671 
672 	/*
673 	 * tell any interested parties about the new process
674 	 */
675 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
676 	PROC_UNLOCK(p1);
677 
678 	/*
679 	 * Preserve synchronization semantics of vfork.  If waiting for
680 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
681 	 * proc (in case of exit).
682 	 */
683 	PROC_LOCK(p2);
684 	while (p2->p_flag & P_PPWAIT)
685 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
686 	PROC_UNLOCK(p2);
687 
688 	/*
689 	 * Return child proc pointer to parent.
690 	 */
691 	*procp = p2;
692 	return (0);
693 }
694 
695 /*
696  * The next two functionms are general routines to handle adding/deleting
697  * items on the fork callout list.
698  *
699  * at_fork():
700  * Take the arguments given and put them onto the fork callout list,
701  * However first make sure that it's not already there.
702  * Returns 0 on success or a standard error number.
703  */
704 
705 int
706 at_fork(function)
707 	forklist_fn function;
708 {
709 	struct forklist *ep;
710 
711 #ifdef INVARIANTS
712 	/* let the programmer know if he's been stupid */
713 	if (rm_at_fork(function))
714 		printf("WARNING: fork callout entry (%p) already present\n",
715 		    function);
716 #endif
717 	ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
718 	if (ep == NULL)
719 		return (ENOMEM);
720 	ep->function = function;
721 	sx_xlock(&fork_list_lock);
722 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
723 	sx_xunlock(&fork_list_lock);
724 	return (0);
725 }
726 
727 /*
728  * Scan the exit callout list for the given item and remove it..
729  * Returns the number of items removed (0 or 1)
730  */
731 
732 int
733 rm_at_fork(function)
734 	forklist_fn function;
735 {
736 	struct forklist *ep;
737 
738 	sx_xlock(&fork_list_lock);
739 	TAILQ_FOREACH(ep, &fork_list, next) {
740 		if (ep->function == function) {
741 			TAILQ_REMOVE(&fork_list, ep, next);
742 			sx_xunlock(&fork_list_lock);
743 			free(ep, M_ATFORK);
744 			return(1);
745 		}
746 	}
747 	sx_xunlock(&fork_list_lock);
748 	return (0);
749 }
750 
751 /*
752  * Handle the return of a child process from fork1().  This function
753  * is called from the MD fork_trampoline() entry point.
754  */
755 void
756 fork_exit(callout, arg, frame)
757 	void (*callout)(void *, struct trapframe *);
758 	void *arg;
759 	struct trapframe *frame;
760 {
761 	struct thread *td = curthread;
762 	struct proc *p = td->td_proc;
763 
764 	td->td_kse->ke_oncpu = PCPU_GET(cpuid);
765 	/*
766 	 * Setup the sched_lock state so that we can release it.
767 	 */
768 	sched_lock.mtx_lock = (uintptr_t)td;
769 	sched_lock.mtx_recurse = 0;
770 	td->td_critnest = 1;
771 	td->td_savecrit = CRITICAL_FORK;
772 	CTR3(KTR_PROC, "fork_exit: new proc %p (pid %d, %s)", p, p->p_pid,
773 	    p->p_comm);
774 	if (PCPU_GET(switchtime.tv_sec) == 0)
775 		microuptime(PCPU_PTR(switchtime));
776 	PCPU_SET(switchticks, ticks);
777 	mtx_unlock_spin(&sched_lock);
778 
779 	/*
780 	 * cpu_set_fork_handler intercepts this function call to
781          * have this call a non-return function to stay in kernel mode.
782          * initproc has its own fork handler, but it does return.
783          */
784 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
785 	callout(arg, frame);
786 
787 	/*
788 	 * Check if a kernel thread misbehaved and returned from its main
789 	 * function.
790 	 */
791 	PROC_LOCK(p);
792 	if (p->p_flag & P_KTHREAD) {
793 		PROC_UNLOCK(p);
794 		mtx_lock(&Giant);
795 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
796 		    p->p_comm, p->p_pid);
797 		kthread_exit(0);
798 	}
799 	PROC_UNLOCK(p);
800 	mtx_lock(&Giant);
801 	crfree(td->td_ucred);
802 	mtx_unlock(&Giant);
803 	td->td_ucred = NULL;
804 	mtx_assert(&Giant, MA_NOTOWNED);
805 }
806 
807 /*
808  * Simplified back end of syscall(), used when returning from fork()
809  * directly into user mode.  Giant is not held on entry, and must not
810  * be held on return.  This function is passed in to fork_exit() as the
811  * first parameter and is called when returning to a new userland process.
812  */
813 void
814 fork_return(td, frame)
815 	struct thread *td;
816 	struct trapframe *frame;
817 {
818 
819 	userret(td, frame, 0);
820 #ifdef KTRACE
821 	if (KTRPOINT(td->td_proc, KTR_SYSRET)) {
822 		ktrsysret(td->td_proc->p_tracep, SYS_fork, 0, 0);
823 	}
824 #endif
825 	mtx_assert(&Giant, MA_NOTOWNED);
826 }
827