xref: /freebsd/sys/kern/kern_fork.c (revision 17d6c636720d00f77e5d098daf4c278f89d84f7b)
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 			if (p1->p_fd->fd_refcnt > 1) {
273 				struct filedesc *newfd;
274 				newfd = fdcopy(td);
275 				PROC_LOCK(p1);
276 				fdfree(td);
277 				p1->p_fd = newfd;
278 				PROC_UNLOCK(p1);
279 			}
280 		}
281 		*procp = NULL;
282 		return (0);
283 	}
284 
285 	/*
286 	 * Although process entries are dynamically created, we still keep
287 	 * a global limit on the maximum number we will create.  Don't allow
288 	 * a nonprivileged user to use the last process; don't let root
289 	 * exceed the limit. The variable nprocs is the current number of
290 	 * processes, maxproc is the limit.
291 	 */
292 	uid = p1->p_ucred->cr_ruid;
293 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
294 		tablefull("proc");
295 		return (EAGAIN);
296 	}
297 	/*
298 	 * Increment the nprocs resource before blocking can occur.  There
299 	 * are hard-limits as to the number of processes that can run.
300 	 */
301 	nprocs++;
302 
303 	/*
304 	 * Increment the count of procs running with this uid. Don't allow
305 	 * a nonprivileged user to exceed their current limit.
306 	 */
307 	ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
308 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
309 	if (!ok) {
310 		/*
311 		 * Back out the process count
312 		 */
313 		nprocs--;
314 		return (EAGAIN);
315 	}
316 
317 	/* Allocate new proc. */
318 	newproc = zalloc(proc_zone);
319 
320 	/*
321 	 * Setup linkage for kernel based threading
322 	 */
323 	if((flags & RFTHREAD) != 0) {
324 		newproc->p_peers = p1->p_peers;
325 		p1->p_peers = newproc;
326 		newproc->p_leader = p1->p_leader;
327 	} else {
328 		newproc->p_peers = NULL;
329 		newproc->p_leader = newproc;
330 	}
331 
332 	newproc->p_vmspace = NULL;
333 
334 	/*
335 	 * Find an unused process ID.  We remember a range of unused IDs
336 	 * ready to use (from lastpid+1 through pidchecked-1).
337 	 *
338 	 * If RFHIGHPID is set (used during system boot), do not allocate
339 	 * low-numbered pids.
340 	 */
341 	sx_xlock(&allproc_lock);
342 	trypid = lastpid + 1;
343 	if (flags & RFHIGHPID) {
344 		if (trypid < 10) {
345 			trypid = 10;
346 		}
347 	} else {
348 		if (randompid)
349 			trypid += arc4random() % randompid;
350 	}
351 retry:
352 	/*
353 	 * If the process ID prototype has wrapped around,
354 	 * restart somewhat above 0, as the low-numbered procs
355 	 * tend to include daemons that don't exit.
356 	 */
357 	if (trypid >= PID_MAX) {
358 		trypid = trypid % PID_MAX;
359 		if (trypid < 100)
360 			trypid += 100;
361 		pidchecked = 0;
362 	}
363 	if (trypid >= pidchecked) {
364 		int doingzomb = 0;
365 
366 		pidchecked = PID_MAX;
367 		/*
368 		 * Scan the active and zombie procs to check whether this pid
369 		 * is in use.  Remember the lowest pid that's greater
370 		 * than trypid, so we can avoid checking for a while.
371 		 */
372 		p2 = LIST_FIRST(&allproc);
373 again:
374 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
375 			while (p2->p_pid == trypid ||
376 			    p2->p_pgrp->pg_id == trypid ||
377 			    p2->p_session->s_sid == trypid) {
378 				trypid++;
379 				if (trypid >= pidchecked)
380 					goto retry;
381 			}
382 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
383 				pidchecked = p2->p_pid;
384 			if (p2->p_pgrp->pg_id > trypid &&
385 			    pidchecked > p2->p_pgrp->pg_id)
386 				pidchecked = p2->p_pgrp->pg_id;
387 			if (p2->p_session->s_sid > trypid &&
388 			    pidchecked > p2->p_session->s_sid)
389 				pidchecked = p2->p_session->s_sid;
390 		}
391 		if (!doingzomb) {
392 			doingzomb = 1;
393 			p2 = LIST_FIRST(&zombproc);
394 			goto again;
395 		}
396 	}
397 
398 	/*
399 	 * RFHIGHPID does not mess with the lastpid counter during boot.
400 	 */
401 	if (flags & RFHIGHPID)
402 		pidchecked = 0;
403 	else
404 		lastpid = trypid;
405 
406 	p2 = newproc;
407 	p2->p_stat = SIDL;			/* protect against others */
408 	p2->p_pid = trypid;
409 	LIST_INSERT_HEAD(&allproc, p2, p_list);
410 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
411 	sx_xunlock(&allproc_lock);
412 
413 	/*
414 	 * Make a proc table entry for the new process.
415 	 * Start by zeroing the section of proc that is zero-initialized,
416 	 * then copy the section that is copied directly from the parent.
417 	 */
418 	bzero(&p2->p_startzero,
419 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
420 	bzero(&p2->p_kse.ke_startzero,
421 	    (unsigned) ((caddr_t)&p2->p_kse.ke_endzero
422 			- (caddr_t)&p2->p_kse.ke_startzero));
423 	bzero(&p2->p_thread.td_startzero,
424 	    (unsigned) ((caddr_t)&p2->p_thread.td_endzero
425 			- (caddr_t)&p2->p_thread.td_startzero));
426 	bzero(&p2->p_ksegrp.kg_startzero,
427 	    (unsigned) ((caddr_t)&p2->p_ksegrp.kg_endzero
428 			- (caddr_t)&p2->p_ksegrp.kg_startzero));
429 	PROC_LOCK(p1);
430 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
431 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
432 
433 	bcopy(&p1->p_kse.ke_startcopy, &p2->p_kse.ke_startcopy,
434 	    (unsigned) ((caddr_t)&p2->p_kse.ke_endcopy
435 			- (caddr_t)&p2->p_kse.ke_startcopy));
436 
437 	bcopy(&p1->p_thread.td_startcopy, &p2->p_thread.td_startcopy,
438 	    (unsigned) ((caddr_t)&p2->p_thread.td_endcopy
439 			- (caddr_t)&p2->p_thread.td_startcopy));
440 
441 	bcopy(&p1->p_ksegrp.kg_startcopy, &p2->p_ksegrp.kg_startcopy,
442 	    (unsigned) ((caddr_t)&p2->p_ksegrp.kg_endcopy
443 			- (caddr_t)&p2->p_ksegrp.kg_startcopy));
444 	PROC_UNLOCK(p1);
445 
446 	/*
447 	 * XXXKSE Theoretically only the running thread would get copied
448 	 * Others in the kernel would be 'aborted' in the child.
449 	 * i.e return E*something*
450 	 */
451 	proc_linkup(p2);
452 
453 	mtx_init(&p2->p_mtx, "process lock", MTX_DEF);
454 	PROC_LOCK(p2);
455 	/* note.. XXXKSE no pcb or u-area yet */
456 
457 	/*
458 	 * Duplicate sub-structures as needed.
459 	 * Increase reference counts on shared objects.
460 	 * The p_stats and p_sigacts substructs are set in vm_forkproc.
461 	 */
462 	p2->p_flag = 0;
463 	mtx_lock_spin(&sched_lock);
464 	p2->p_sflag = PS_INMEM;
465 	if (p1->p_sflag & PS_PROFIL)
466 		startprofclock(p2);
467 	mtx_unlock_spin(&sched_lock);
468 	PROC_LOCK(p1);
469 	p2->p_ucred = crhold(p1->p_ucred);
470 	p2->p_thread.td_ucred = crhold(p2->p_ucred);	/* XXXKSE */
471 
472 	if (p2->p_args)
473 		p2->p_args->ar_ref++;
474 
475 	if (flags & RFSIGSHARE) {
476 		p2->p_procsig = p1->p_procsig;
477 		p2->p_procsig->ps_refcnt++;
478 		if (p1->p_sigacts == &p1->p_uarea->u_sigacts) {
479 			struct sigacts *newsigacts;
480 
481 			PROC_UNLOCK(p1);
482 			PROC_UNLOCK(p2);
483 			/* Create the shared sigacts structure */
484 			MALLOC(newsigacts, struct sigacts *,
485 			    sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
486 			PROC_LOCK(p2);
487 			PROC_LOCK(p1);
488 			/*
489 			 * Set p_sigacts to the new shared structure.
490 			 * Note that this is updating p1->p_sigacts at the
491 			 * same time, since p_sigacts is just a pointer to
492 			 * the shared p_procsig->ps_sigacts.
493 			 */
494 			p2->p_sigacts  = newsigacts;
495 			*p2->p_sigacts = p1->p_uarea->u_sigacts;
496 		}
497 	} else {
498 		PROC_UNLOCK(p1);
499 		PROC_UNLOCK(p2);
500 		MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
501 		    M_SUBPROC, M_WAITOK);
502 		PROC_LOCK(p2);
503 		PROC_LOCK(p1);
504 		bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
505 		p2->p_procsig->ps_refcnt = 1;
506 		p2->p_sigacts = NULL;	/* finished in vm_forkproc() */
507 	}
508 	if (flags & RFLINUXTHPN)
509 	        p2->p_sigparent = SIGUSR1;
510 	else
511 	        p2->p_sigparent = SIGCHLD;
512 
513 	/* bump references to the text vnode (for procfs) */
514 	p2->p_textvp = p1->p_textvp;
515 	PROC_UNLOCK(p1);
516 	PROC_UNLOCK(p2);
517 	if (p2->p_textvp)
518 		VREF(p2->p_textvp);
519 
520 	if (flags & RFCFDG)
521 		fd = fdinit(td);
522 	else if (flags & RFFDG)
523 		fd = fdcopy(td);
524 	else
525 		fd = fdshare(p1);
526 	PROC_LOCK(p2);
527 	p2->p_fd = fd;
528 
529 	/*
530 	 * If p_limit is still copy-on-write, bump refcnt,
531 	 * otherwise get a copy that won't be modified.
532 	 * (If PL_SHAREMOD is clear, the structure is shared
533 	 * copy-on-write.)
534 	 */
535 	PROC_LOCK(p1);
536 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
537 		p2->p_limit = limcopy(p1->p_limit);
538 	else {
539 		p2->p_limit = p1->p_limit;
540 		p2->p_limit->p_refcnt++;
541 	}
542 
543 	/*
544 	 * Preserve some more flags in subprocess.  PS_PROFIL has already
545 	 * been preserved.
546 	 */
547 	p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
548 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
549 		p2->p_flag |= P_CONTROLT;
550 	if (flags & RFPPWAIT)
551 		p2->p_flag |= P_PPWAIT;
552 
553 	LIST_INSERT_AFTER(p1, p2, p_pglist);
554 	PROC_UNLOCK(p1);
555 	PROC_UNLOCK(p2);
556 
557 	/*
558 	 * Attach the new process to its parent.
559 	 *
560 	 * If RFNOWAIT is set, the newly created process becomes a child
561 	 * of init.  This effectively disassociates the child from the
562 	 * parent.
563 	 */
564 	if (flags & RFNOWAIT)
565 		pptr = initproc;
566 	else
567 		pptr = p1;
568 	sx_xlock(&proctree_lock);
569 	PROC_LOCK(p2);
570 	p2->p_pptr = pptr;
571 	PROC_UNLOCK(p2);
572 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
573 	sx_xunlock(&proctree_lock);
574 	PROC_LOCK(p2);
575 	LIST_INIT(&p2->p_children);
576 	LIST_INIT(&p2->p_thread.td_contested); /* XXXKSE only 1 thread? */
577 
578 	callout_init(&p2->p_itcallout, 0);
579 	callout_init(&p2->p_thread.td_slpcallout, 1); /* XXXKSE */
580 
581 	PROC_LOCK(p1);
582 #ifdef KTRACE
583 	/*
584 	 * Copy traceflag and tracefile if enabled.  If not inherited,
585 	 * these were zeroed above but we still could have a trace race
586 	 * so make sure p2's p_tracep is NULL.
587 	 */
588 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
589 		p2->p_traceflag = p1->p_traceflag;
590 		if ((p2->p_tracep = p1->p_tracep) != NULL) {
591 			PROC_UNLOCK(p1);
592 			PROC_UNLOCK(p2);
593 			VREF(p2->p_tracep);
594 			PROC_LOCK(p2);
595 			PROC_LOCK(p1);
596 		}
597 	}
598 #endif
599 
600 	/*
601 	 * set priority of child to be that of parent
602 	 * XXXKSE hey! copying the estcpu seems dodgy.. should split it..
603 	 */
604 	mtx_lock_spin(&sched_lock);
605 	p2->p_ksegrp.kg_estcpu = p1->p_ksegrp.kg_estcpu;
606 	mtx_unlock_spin(&sched_lock);
607 
608 	/*
609 	 * This begins the section where we must prevent the parent
610 	 * from being swapped.
611 	 */
612 	_PHOLD(p1);
613 	PROC_UNLOCK(p1);
614 	PROC_UNLOCK(p2);
615 
616 	/*
617 	 * Finish creating the child process.  It will return via a different
618 	 * execution path later.  (ie: directly into user mode)
619 	 */
620 	vm_forkproc(td, p2, flags);
621 
622 	if (flags == (RFFDG | RFPROC)) {
623 		cnt.v_forks++;
624 		cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
625 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
626 		cnt.v_vforks++;
627 		cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
628 	} else if (p1 == &proc0) {
629 		cnt.v_kthreads++;
630 		cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
631 	} else {
632 		cnt.v_rforks++;
633 		cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
634 	}
635 
636 	/*
637 	 * Both processes are set up, now check if any loadable modules want
638 	 * to adjust anything.
639 	 *   What if they have an error? XXX
640 	 */
641 	sx_slock(&fork_list_lock);
642 	TAILQ_FOREACH(ep, &fork_list, next) {
643 		(*ep->function)(p1, p2, flags);
644 	}
645 	sx_sunlock(&fork_list_lock);
646 
647 	/*
648 	 * If RFSTOPPED not requested, make child runnable and add to
649 	 * run queue.
650 	 */
651 	microtime(&(p2->p_stats->p_start));
652 	p2->p_acflag = AFORK;
653 	if ((flags & RFSTOPPED) == 0) {
654 		mtx_lock_spin(&sched_lock);
655 		p2->p_stat = SRUN;
656 		setrunqueue(&p2->p_thread);
657 		mtx_unlock_spin(&sched_lock);
658 	}
659 
660 	/*
661 	 * Now can be swapped.
662 	 */
663 	PROC_LOCK(p1);
664 	_PRELE(p1);
665 
666 	/*
667 	 * tell any interested parties about the new process
668 	 */
669 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
670 	PROC_UNLOCK(p1);
671 
672 	/*
673 	 * Preserve synchronization semantics of vfork.  If waiting for
674 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
675 	 * proc (in case of exit).
676 	 */
677 	PROC_LOCK(p2);
678 	while (p2->p_flag & P_PPWAIT)
679 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
680 	PROC_UNLOCK(p2);
681 
682 	/*
683 	 * Return child proc pointer to parent.
684 	 */
685 	*procp = p2;
686 	return (0);
687 }
688 
689 /*
690  * The next two functionms are general routines to handle adding/deleting
691  * items on the fork callout list.
692  *
693  * at_fork():
694  * Take the arguments given and put them onto the fork callout list,
695  * However first make sure that it's not already there.
696  * Returns 0 on success or a standard error number.
697  */
698 
699 int
700 at_fork(function)
701 	forklist_fn function;
702 {
703 	struct forklist *ep;
704 
705 #ifdef INVARIANTS
706 	/* let the programmer know if he's been stupid */
707 	if (rm_at_fork(function))
708 		printf("WARNING: fork callout entry (%p) already present\n",
709 		    function);
710 #endif
711 	ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
712 	if (ep == NULL)
713 		return (ENOMEM);
714 	ep->function = function;
715 	sx_xlock(&fork_list_lock);
716 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
717 	sx_xunlock(&fork_list_lock);
718 	return (0);
719 }
720 
721 /*
722  * Scan the exit callout list for the given item and remove it..
723  * Returns the number of items removed (0 or 1)
724  */
725 
726 int
727 rm_at_fork(function)
728 	forklist_fn function;
729 {
730 	struct forklist *ep;
731 
732 	sx_xlock(&fork_list_lock);
733 	TAILQ_FOREACH(ep, &fork_list, next) {
734 		if (ep->function == function) {
735 			TAILQ_REMOVE(&fork_list, ep, next);
736 			sx_xunlock(&fork_list_lock);
737 			free(ep, M_ATFORK);
738 			return(1);
739 		}
740 	}
741 	sx_xunlock(&fork_list_lock);
742 	return (0);
743 }
744 
745 /*
746  * Handle the return of a child process from fork1().  This function
747  * is called from the MD fork_trampoline() entry point.
748  */
749 void
750 fork_exit(callout, arg, frame)
751 	void (*callout)(void *, struct trapframe *);
752 	void *arg;
753 	struct trapframe *frame;
754 {
755 	struct thread *td = curthread;
756 	struct proc *p = td->td_proc;
757 
758 	td->td_kse->ke_oncpu = PCPU_GET(cpuid);
759 	/*
760 	 * Setup the sched_lock state so that we can release it.
761 	 */
762 	sched_lock.mtx_lock = (uintptr_t)td;
763 	sched_lock.mtx_recurse = 0;
764 	td->td_critnest = 1;
765 	td->td_savecrit = CRITICAL_FORK;
766 	CTR3(KTR_PROC, "fork_exit: new proc %p (pid %d, %s)", p, p->p_pid,
767 	    p->p_comm);
768 	if (PCPU_GET(switchtime.tv_sec) == 0)
769 		microuptime(PCPU_PTR(switchtime));
770 	PCPU_SET(switchticks, ticks);
771 	mtx_unlock_spin(&sched_lock);
772 
773 	/*
774 	 * cpu_set_fork_handler intercepts this function call to
775          * have this call a non-return function to stay in kernel mode.
776          * initproc has its own fork handler, but it does return.
777          */
778 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
779 	callout(arg, frame);
780 
781 	/*
782 	 * Check if a kernel thread misbehaved and returned from its main
783 	 * function.
784 	 */
785 	PROC_LOCK(p);
786 	if (p->p_flag & P_KTHREAD) {
787 		PROC_UNLOCK(p);
788 		mtx_lock(&Giant);
789 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
790 		    p->p_comm, p->p_pid);
791 		kthread_exit(0);
792 	}
793 	PROC_UNLOCK(p);
794 	mtx_lock(&Giant);
795 	crfree(td->td_ucred);
796 	mtx_unlock(&Giant);
797 	td->td_ucred = NULL;
798 	mtx_assert(&Giant, MA_NOTOWNED);
799 }
800 
801 /*
802  * Simplified back end of syscall(), used when returning from fork()
803  * directly into user mode.  Giant is not held on entry, and must not
804  * be held on return.  This function is passed in to fork_exit() as the
805  * first parameter and is called when returning to a new userland process.
806  */
807 void
808 fork_return(td, frame)
809 	struct thread *td;
810 	struct trapframe *frame;
811 {
812 
813 	userret(td, frame, 0);
814 #ifdef KTRACE
815 	if (KTRPOINT(td->td_proc, KTR_SYSRET)) {
816 		ktrsysret(td->td_proc->p_tracep, SYS_fork, 0, 0);
817 	}
818 #endif
819 	mtx_assert(&Giant, MA_NOTOWNED);
820 }
821