xref: /freebsd/sys/kern/kern_fork.c (revision 06a31d6a6779b74405b41c8ad4579b5d81db4d4c)
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 #include "opt_mac.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/filedesc.h>
50 #include <sys/kernel.h>
51 #include <sys/sysctl.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/pioctl.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/syscall.h>
60 #include <sys/vnode.h>
61 #include <sys/acct.h>
62 #include <sys/mac.h>
63 #include <sys/ktr.h>
64 #include <sys/ktrace.h>
65 #include <sys/kthread.h>
66 #include <sys/unistd.h>
67 #include <sys/jail.h>
68 #include <sys/sx.h>
69 
70 #include <vm/vm.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_extern.h>
74 #include <vm/uma.h>
75 
76 #include <sys/vmmeter.h>
77 #include <sys/user.h>
78 #include <machine/critical.h>
79 
80 #ifndef _SYS_SYSPROTO_H_
81 struct fork_args {
82 	int     dummy;
83 };
84 #endif
85 
86 static int forksleep; /* Place for fork1() to sleep on. */
87 
88 /*
89  * MPSAFE
90  */
91 /* ARGSUSED */
92 int
93 fork(td, uap)
94 	struct thread *td;
95 	struct fork_args *uap;
96 {
97 	int error;
98 	struct proc *p2;
99 
100 	error = fork1(td, RFFDG | RFPROC, 0, &p2);
101 	if (error == 0) {
102 		td->td_retval[0] = p2->p_pid;
103 		td->td_retval[1] = 0;
104 	}
105 	return error;
106 }
107 
108 /*
109  * MPSAFE
110  */
111 /* ARGSUSED */
112 int
113 vfork(td, uap)
114 	struct thread *td;
115 	struct vfork_args *uap;
116 {
117 	int error;
118 	struct proc *p2;
119 
120 	error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
121 	if (error == 0) {
122 		td->td_retval[0] = p2->p_pid;
123 		td->td_retval[1] = 0;
124 	}
125 	return error;
126 }
127 
128 /*
129  * MPSAFE
130  */
131 int
132 rfork(td, uap)
133 	struct thread *td;
134 	struct rfork_args *uap;
135 {
136 	int error;
137 	struct proc *p2;
138 
139 	/* Don't allow kernel only flags. */
140 	if ((uap->flags & RFKERNELONLY) != 0)
141 		return (EINVAL);
142 	error = fork1(td, uap->flags, 0, &p2);
143 	if (error == 0) {
144 		td->td_retval[0] = p2 ? p2->p_pid : 0;
145 		td->td_retval[1] = 0;
146 	}
147 	return error;
148 }
149 
150 
151 int	nprocs = 1;				/* process 0 */
152 int	lastpid = 0;
153 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
154     "Last used PID");
155 
156 /*
157  * Random component to lastpid generation.  We mix in a random factor to make
158  * it a little harder to predict.  We sanity check the modulus value to avoid
159  * doing it in critical paths.  Don't let it be too small or we pointlessly
160  * waste randomness entropy, and don't let it be impossibly large.  Using a
161  * modulus that is too big causes a LOT more process table scans and slows
162  * down fork processing as the pidchecked caching is defeated.
163  */
164 static int randompid = 0;
165 
166 static int
167 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
168 {
169 	int error, pid;
170 
171 	sysctl_wire_old_buffer(req, sizeof(int));
172 	sx_xlock(&allproc_lock);
173 	pid = randompid;
174 	error = sysctl_handle_int(oidp, &pid, 0, req);
175 	if (error == 0 && req->newptr != NULL) {
176 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
177 			pid = PID_MAX - 100;
178 		else if (pid < 2)			/* NOP */
179 			pid = 0;
180 		else if (pid < 100)			/* Make it reasonable */
181 			pid = 100;
182 		randompid = pid;
183 	}
184 	sx_xunlock(&allproc_lock);
185 	return (error);
186 }
187 
188 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
189     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
190 
191 int
192 fork1(td, flags, pages, procp)
193 	struct thread *td;			/* parent proc */
194 	int flags;
195 	int pages;
196 	struct proc **procp;			/* child proc */
197 {
198 	struct proc *p2, *pptr;
199 	uid_t uid;
200 	struct proc *newproc;
201 	int trypid;
202 	int ok;
203 	static int pidchecked = 0;
204 	struct filedesc *fd;
205 	struct filedesc_to_leader *fdtol;
206 	struct proc *p1 = td->td_proc;
207 	struct thread *td2;
208 	struct kse *ke2;
209 	struct ksegrp *kg2;
210 	struct sigacts *newsigacts;
211 	int error;
212 
213 	/* Can't copy and clear */
214 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
215 		return (EINVAL);
216 
217 	mtx_lock(&Giant);
218 	/*
219 	 * Here we don't create a new process, but we divorce
220 	 * certain parts of a process from itself.
221 	 */
222 	if ((flags & RFPROC) == 0) {
223 		vm_forkproc(td, NULL, NULL, flags);
224 
225 		/*
226 		 * Close all file descriptors.
227 		 */
228 		if (flags & RFCFDG) {
229 			struct filedesc *fdtmp;
230 			fdtmp = fdinit(td->td_proc->p_fd);
231 			fdfree(td);
232 			p1->p_fd = fdtmp;
233 		}
234 
235 		/*
236 		 * Unshare file descriptors (from parent.)
237 		 */
238 		if (flags & RFFDG) {
239 			FILEDESC_LOCK(p1->p_fd);
240 			if (p1->p_fd->fd_refcnt > 1) {
241 				struct filedesc *newfd;
242 
243 				newfd = fdcopy(td->td_proc->p_fd);
244 				FILEDESC_UNLOCK(p1->p_fd);
245 				fdfree(td);
246 				p1->p_fd = newfd;
247 			} else
248 				FILEDESC_UNLOCK(p1->p_fd);
249 		}
250 		mtx_unlock(&Giant);
251 		*procp = NULL;
252 		return (0);
253 	}
254 
255 	/*
256 	 * Note 1:1 allows for forking with one thread coming out on the
257 	 * other side with the expectation that the process is about to
258 	 * exec.
259 	 */
260 	if (p1->p_flag & P_THREADED) {
261 		/*
262 		 * Idle the other threads for a second.
263 		 * Since the user space is copied, it must remain stable.
264 		 * In addition, all threads (from the user perspective)
265 		 * need to either be suspended or in the kernel,
266 		 * where they will try restart in the parent and will
267 		 * be aborted in the child.
268 		 */
269 		PROC_LOCK(p1);
270 		if (thread_single(SINGLE_NO_EXIT)) {
271 			/* Abort.. someone else is single threading before us */
272 			PROC_UNLOCK(p1);
273 			mtx_unlock(&Giant);
274 			return (ERESTART);
275 		}
276 		PROC_UNLOCK(p1);
277 		/*
278 		 * All other activity in this process
279 		 * is now suspended at the user boundary,
280 		 * (or other safe places if we think of any).
281 		 */
282 	}
283 
284 	/* Allocate new proc. */
285 	newproc = uma_zalloc(proc_zone, M_WAITOK);
286 #ifdef MAC
287 	mac_init_proc(newproc);
288 #endif
289 
290 	/*
291 	 * Although process entries are dynamically created, we still keep
292 	 * a global limit on the maximum number we will create.  Don't allow
293 	 * a nonprivileged user to use the last ten processes; don't let root
294 	 * exceed the limit. The variable nprocs is the current number of
295 	 * processes, maxproc is the limit.
296 	 */
297 	sx_xlock(&allproc_lock);
298 	uid = td->td_ucred->cr_ruid;
299 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
300 		error = EAGAIN;
301 		goto fail;
302 	}
303 
304 	/*
305 	 * Increment the count of procs running with this uid. Don't allow
306 	 * a nonprivileged user to exceed their current limit.
307 	 */
308 	PROC_LOCK(p1);
309 	ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
310 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
311 	PROC_UNLOCK(p1);
312 	if (!ok) {
313 		error = EAGAIN;
314 		goto fail;
315 	}
316 
317 	/*
318 	 * Increment the nprocs resource before blocking can occur.  There
319 	 * are hard-limits as to the number of processes that can run.
320 	 */
321 	nprocs++;
322 
323 	/*
324 	 * Find an unused process ID.  We remember a range of unused IDs
325 	 * ready to use (from lastpid+1 through pidchecked-1).
326 	 *
327 	 * If RFHIGHPID is set (used during system boot), do not allocate
328 	 * low-numbered pids.
329 	 */
330 	trypid = lastpid + 1;
331 	if (flags & RFHIGHPID) {
332 		if (trypid < 10) {
333 			trypid = 10;
334 		}
335 	} else {
336 		if (randompid)
337 			trypid += arc4random() % randompid;
338 	}
339 retry:
340 	/*
341 	 * If the process ID prototype has wrapped around,
342 	 * restart somewhat above 0, as the low-numbered procs
343 	 * tend to include daemons that don't exit.
344 	 */
345 	if (trypid >= PID_MAX) {
346 		trypid = trypid % PID_MAX;
347 		if (trypid < 100)
348 			trypid += 100;
349 		pidchecked = 0;
350 	}
351 	if (trypid >= pidchecked) {
352 		int doingzomb = 0;
353 
354 		pidchecked = PID_MAX;
355 		/*
356 		 * Scan the active and zombie procs to check whether this pid
357 		 * is in use.  Remember the lowest pid that's greater
358 		 * than trypid, so we can avoid checking for a while.
359 		 */
360 		p2 = LIST_FIRST(&allproc);
361 again:
362 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
363 			PROC_LOCK(p2);
364 			while (p2->p_pid == trypid ||
365 			    p2->p_pgrp->pg_id == trypid ||
366 			    p2->p_session->s_sid == trypid) {
367 				trypid++;
368 				if (trypid >= pidchecked) {
369 					PROC_UNLOCK(p2);
370 					goto retry;
371 				}
372 			}
373 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
374 				pidchecked = p2->p_pid;
375 			if (p2->p_pgrp->pg_id > trypid &&
376 			    pidchecked > p2->p_pgrp->pg_id)
377 				pidchecked = p2->p_pgrp->pg_id;
378 			if (p2->p_session->s_sid > trypid &&
379 			    pidchecked > p2->p_session->s_sid)
380 				pidchecked = p2->p_session->s_sid;
381 			PROC_UNLOCK(p2);
382 		}
383 		if (!doingzomb) {
384 			doingzomb = 1;
385 			p2 = LIST_FIRST(&zombproc);
386 			goto again;
387 		}
388 	}
389 
390 	/*
391 	 * RFHIGHPID does not mess with the lastpid counter during boot.
392 	 */
393 	if (flags & RFHIGHPID)
394 		pidchecked = 0;
395 	else
396 		lastpid = trypid;
397 
398 	p2 = newproc;
399 	p2->p_state = PRS_NEW;		/* protect against others */
400 	p2->p_pid = trypid;
401 	LIST_INSERT_HEAD(&allproc, p2, p_list);
402 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
403 	sx_xunlock(&allproc_lock);
404 
405 	/*
406 	 * Malloc things while we don't hold any locks.
407 	 */
408 	if (flags & RFSIGSHARE)
409 		newsigacts = NULL;
410 	else
411 		newsigacts = sigacts_alloc();
412 
413 	/*
414 	 * Copy filedesc.
415 	 */
416 	if (flags & RFCFDG) {
417 		fd = fdinit(td->td_proc->p_fd);
418 		fdtol = NULL;
419 	} else if (flags & RFFDG) {
420 		FILEDESC_LOCK(p1->p_fd);
421 		fd = fdcopy(td->td_proc->p_fd);
422 		FILEDESC_UNLOCK(p1->p_fd);
423 		fdtol = NULL;
424 	} else {
425 		fd = fdshare(p1->p_fd);
426 		if (p1->p_fdtol == NULL)
427 			p1->p_fdtol =
428 				filedesc_to_leader_alloc(NULL,
429 							 NULL,
430 							 p1->p_leader);
431 		if ((flags & RFTHREAD) != 0) {
432 			/*
433 			 * Shared file descriptor table and
434 			 * shared process leaders.
435 			 */
436 			fdtol = p1->p_fdtol;
437 			FILEDESC_LOCK(p1->p_fd);
438 			fdtol->fdl_refcount++;
439 			FILEDESC_UNLOCK(p1->p_fd);
440 		} else {
441 			/*
442 			 * Shared file descriptor table, and
443 			 * different process leaders
444 			 */
445 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
446 							 p1->p_fd,
447 							 p2);
448 		}
449 	}
450 	/*
451 	 * Make a proc table entry for the new process.
452 	 * Start by zeroing the section of proc that is zero-initialized,
453 	 * then copy the section that is copied directly from the parent.
454 	 */
455 	td2 = FIRST_THREAD_IN_PROC(p2);
456 	kg2 = FIRST_KSEGRP_IN_PROC(p2);
457 	ke2 = FIRST_KSE_IN_KSEGRP(kg2);
458 
459 	/* Allocate and switch to an alternate kstack if specified */
460 	if (pages != 0)
461 		pmap_new_altkstack(td2, pages);
462 
463 	PROC_LOCK(p2);
464 	PROC_LOCK(p1);
465 
466 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start))
467 
468 	bzero(&p2->p_startzero,
469 	    (unsigned) RANGEOF(struct proc, p_startzero, p_endzero));
470 	bzero(&ke2->ke_startzero,
471 	    (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero));
472 	bzero(&td2->td_startzero,
473 	    (unsigned) RANGEOF(struct thread, td_startzero, td_endzero));
474 	bzero(&kg2->kg_startzero,
475 	    (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero));
476 
477 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
478 	    (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy));
479 	bcopy(&td->td_startcopy, &td2->td_startcopy,
480 	    (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy));
481 	bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
482 	    (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy));
483 #undef RANGEOF
484 
485 	/* Set up the thread as an active thread (as if runnable). */
486 	ke2->ke_state = KES_THREAD;
487 	ke2->ke_thread = td2;
488 	td2->td_kse = ke2;
489 
490 	/*
491 	 * Duplicate sub-structures as needed.
492 	 * Increase reference counts on shared objects.
493 	 * The p_stats substruct is set in vm_forkproc.
494 	 */
495 	p2->p_flag = 0;
496 	if (p1->p_flag & P_PROFIL)
497 		startprofclock(p2);
498 	mtx_lock_spin(&sched_lock);
499 	p2->p_sflag = PS_INMEM;
500 	/*
501 	 * Allow the scheduler to adjust the priority of the child and
502 	 * parent while we hold the sched_lock.
503 	 */
504 	sched_fork(p1, p2);
505 
506 	mtx_unlock_spin(&sched_lock);
507 	p2->p_ucred = crhold(td->td_ucred);
508 	td2->td_ucred = crhold(p2->p_ucred);	/* XXXKSE */
509 
510 	pargs_hold(p2->p_args);
511 
512 	if (flags & RFSIGSHARE) {
513 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
514 	} else {
515 		sigacts_copy(newsigacts, p1->p_sigacts);
516 		p2->p_sigacts = newsigacts;
517 	}
518 	if (flags & RFLINUXTHPN)
519 	        p2->p_sigparent = SIGUSR1;
520 	else
521 	        p2->p_sigparent = SIGCHLD;
522 
523 	/* Bump references to the text vnode (for procfs) */
524 	p2->p_textvp = p1->p_textvp;
525 	if (p2->p_textvp)
526 		VREF(p2->p_textvp);
527 	p2->p_fd = fd;
528 	p2->p_fdtol = fdtol;
529 	PROC_UNLOCK(p1);
530 	PROC_UNLOCK(p2);
531 
532 	/*
533 	 * p_limit is copy-on-write, bump refcnt,
534 	 */
535 	p2->p_limit = p1->p_limit;
536 	p2->p_limit->p_refcnt++;
537 
538 	/*
539 	 * Setup linkage for kernel based threading
540 	 */
541 	if((flags & RFTHREAD) != 0) {
542 		mtx_lock(&ppeers_lock);
543 		p2->p_peers = p1->p_peers;
544 		p1->p_peers = p2;
545 		p2->p_leader = p1->p_leader;
546 		mtx_unlock(&ppeers_lock);
547 		PROC_LOCK(p1->p_leader);
548 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
549 			PROC_UNLOCK(p1->p_leader);
550 			/*
551 			 * The task leader is exiting, so process p1 is
552 			 * going to be killed shortly.  Since p1 obviously
553 			 * isn't dead yet, we know that the leader is either
554 			 * sending SIGKILL's to all the processes in this
555 			 * task or is sleeping waiting for all the peers to
556 			 * exit.  We let p1 complete the fork, but we need
557 			 * to go ahead and kill the new process p2 since
558 			 * the task leader may not get a chance to send
559 			 * SIGKILL to it.  We leave it on the list so that
560 			 * the task leader will wait for this new process
561 			 * to commit suicide.
562 			 */
563 			PROC_LOCK(p2);
564 			psignal(p2, SIGKILL);
565 			PROC_UNLOCK(p2);
566 		} else
567 			PROC_UNLOCK(p1->p_leader);
568 	} else {
569 		p2->p_peers = NULL;
570 		p2->p_leader = p2;
571 	}
572 
573 	sx_xlock(&proctree_lock);
574 	PGRP_LOCK(p1->p_pgrp);
575 	PROC_LOCK(p2);
576 	PROC_LOCK(p1);
577 
578 	/*
579 	 * Preserve some more flags in subprocess.  P_PROFIL has already
580 	 * been preserved.
581 	 */
582 	p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
583 	SESS_LOCK(p1->p_session);
584 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
585 		p2->p_flag |= P_CONTROLT;
586 	SESS_UNLOCK(p1->p_session);
587 	if (flags & RFPPWAIT)
588 		p2->p_flag |= P_PPWAIT;
589 
590 	LIST_INSERT_AFTER(p1, p2, p_pglist);
591 	PGRP_UNLOCK(p1->p_pgrp);
592 	LIST_INIT(&p2->p_children);
593 
594 	callout_init(&p2->p_itcallout, 1);
595 
596 #ifdef KTRACE
597 	/*
598 	 * Copy traceflag and tracefile if enabled.
599 	 */
600 	mtx_lock(&ktrace_mtx);
601 	KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
602 	if (p1->p_traceflag & KTRFAC_INHERIT) {
603 		p2->p_traceflag = p1->p_traceflag;
604 		if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
605 			VREF(p2->p_tracevp);
606 			KASSERT(p1->p_tracecred != NULL,
607 			    ("ktrace vnode with no cred"));
608 			p2->p_tracecred = crhold(p1->p_tracecred);
609 		}
610 	}
611 	mtx_unlock(&ktrace_mtx);
612 #endif
613 
614 	/*
615 	 * If PF_FORK is set, the child process inherits the
616 	 * procfs ioctl flags from its parent.
617 	 */
618 	if (p1->p_pfsflags & PF_FORK) {
619 		p2->p_stops = p1->p_stops;
620 		p2->p_pfsflags = p1->p_pfsflags;
621 	}
622 
623 	/*
624 	 * This begins the section where we must prevent the parent
625 	 * from being swapped.
626 	 */
627 	_PHOLD(p1);
628 	PROC_UNLOCK(p1);
629 
630 	/*
631 	 * Attach the new process to its parent.
632 	 *
633 	 * If RFNOWAIT is set, the newly created process becomes a child
634 	 * of init.  This effectively disassociates the child from the
635 	 * parent.
636 	 */
637 	if (flags & RFNOWAIT)
638 		pptr = initproc;
639 	else
640 		pptr = p1;
641 	p2->p_pptr = pptr;
642 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
643 	sx_xunlock(&proctree_lock);
644 
645 	/* Inform accounting that we have forked. */
646 	p2->p_acflag = AFORK;
647 	PROC_UNLOCK(p2);
648 
649 	/*
650 	 * Finish creating the child process.  It will return via a different
651 	 * execution path later.  (ie: directly into user mode)
652 	 */
653 	vm_forkproc(td, p2, td2, flags);
654 
655 	if (flags == (RFFDG | RFPROC)) {
656 		cnt.v_forks++;
657 		cnt.v_forkpages += p2->p_vmspace->vm_dsize +
658 		    p2->p_vmspace->vm_ssize;
659 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
660 		cnt.v_vforks++;
661 		cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
662 		    p2->p_vmspace->vm_ssize;
663 	} else if (p1 == &proc0) {
664 		cnt.v_kthreads++;
665 		cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
666 		    p2->p_vmspace->vm_ssize;
667 	} else {
668 		cnt.v_rforks++;
669 		cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
670 		    p2->p_vmspace->vm_ssize;
671 	}
672 
673 	/*
674 	 * Both processes are set up, now check if any loadable modules want
675 	 * to adjust anything.
676 	 *   What if they have an error? XXX
677 	 */
678 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
679 
680 	/*
681 	 * If RFSTOPPED not requested, make child runnable and add to
682 	 * run queue.
683 	 */
684 	microuptime(&p2->p_stats->p_start);
685 	if ((flags & RFSTOPPED) == 0) {
686 		mtx_lock_spin(&sched_lock);
687 		p2->p_state = PRS_NORMAL;
688 		TD_SET_CAN_RUN(td2);
689 		setrunqueue(td2);
690 		mtx_unlock_spin(&sched_lock);
691 	}
692 
693 	/*
694 	 * Now can be swapped.
695 	 */
696 	PROC_LOCK(p1);
697 	_PRELE(p1);
698 
699 	/*
700 	 * tell any interested parties about the new process
701 	 */
702 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
703 	PROC_UNLOCK(p1);
704 
705 	/*
706 	 * Preserve synchronization semantics of vfork.  If waiting for
707 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
708 	 * proc (in case of exit).
709 	 */
710 	PROC_LOCK(p2);
711 	while (p2->p_flag & P_PPWAIT)
712 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
713 	PROC_UNLOCK(p2);
714 
715 	/*
716 	 * If other threads are waiting, let them continue now
717 	 */
718 	if (p1->p_flag & P_THREADED) {
719 		PROC_LOCK(p1);
720 		thread_single_end();
721 		PROC_UNLOCK(p1);
722 	}
723 
724 	/*
725 	 * Return child proc pointer to parent.
726 	 */
727 	mtx_unlock(&Giant);
728 	*procp = p2;
729 	return (0);
730 fail:
731 	sx_xunlock(&allproc_lock);
732 	uma_zfree(proc_zone, newproc);
733 	if (p1->p_flag & P_THREADED) {
734 		PROC_LOCK(p1);
735 		thread_single_end();
736 		PROC_UNLOCK(p1);
737 	}
738 	tsleep(&forksleep, PUSER, "fork", hz / 2);
739 	mtx_unlock(&Giant);
740 	return (error);
741 }
742 
743 /*
744  * Handle the return of a child process from fork1().  This function
745  * is called from the MD fork_trampoline() entry point.
746  */
747 void
748 fork_exit(callout, arg, frame)
749 	void (*callout)(void *, struct trapframe *);
750 	void *arg;
751 	struct trapframe *frame;
752 {
753 	struct thread *td;
754 	struct proc *p;
755 
756 	if ((td = PCPU_GET(deadthread))) {
757 		PCPU_SET(deadthread, NULL);
758 		thread_stash(td);
759 	}
760 	td = curthread;
761 	p = td->td_proc;
762 	td->td_oncpu = PCPU_GET(cpuid);
763 	p->p_state = PRS_NORMAL;
764 	/*
765 	 * Finish setting up thread glue.  We need to initialize
766 	 * the thread into a td_critnest=1 state.  Some platforms
767 	 * may have already partially or fully initialized td_critnest
768 	 * and/or td_md.md_savecrit (when applciable).
769 	 *
770 	 * see <arch>/<arch>/critical.c
771 	 */
772 	sched_lock.mtx_lock = (uintptr_t)td;
773 	sched_lock.mtx_recurse = 0;
774 	cpu_critical_fork_exit();
775 	CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid,
776 	    p->p_comm);
777 	if (PCPU_GET(switchtime.sec) == 0)
778 		binuptime(PCPU_PTR(switchtime));
779 	PCPU_SET(switchticks, ticks);
780 	mtx_unlock_spin(&sched_lock);
781 
782 	/*
783 	 * cpu_set_fork_handler intercepts this function call to
784          * have this call a non-return function to stay in kernel mode.
785          * initproc has its own fork handler, but it does return.
786          */
787 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
788 	callout(arg, frame);
789 
790 	/*
791 	 * Check if a kernel thread misbehaved and returned from its main
792 	 * function.
793 	 */
794 	PROC_LOCK(p);
795 	if (p->p_flag & P_KTHREAD) {
796 		PROC_UNLOCK(p);
797 		mtx_lock(&Giant);
798 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
799 		    p->p_comm, p->p_pid);
800 		kthread_exit(0);
801 	}
802 	PROC_UNLOCK(p);
803 #ifdef DIAGNOSTIC
804 	cred_free_thread(td);
805 #endif
806 	mtx_assert(&Giant, MA_NOTOWNED);
807 }
808 
809 /*
810  * Simplified back end of syscall(), used when returning from fork()
811  * directly into user mode.  Giant is not held on entry, and must not
812  * be held on return.  This function is passed in to fork_exit() as the
813  * first parameter and is called when returning to a new userland process.
814  */
815 void
816 fork_return(td, frame)
817 	struct thread *td;
818 	struct trapframe *frame;
819 {
820 
821 	userret(td, frame, 0);
822 #ifdef KTRACE
823 	if (KTRPOINT(td, KTR_SYSRET))
824 		ktrsysret(SYS_fork, 0, 0);
825 #endif
826 	mtx_assert(&Giant, MA_NOTOWNED);
827 }
828