xref: /freebsd/sys/kern/kern_fork.c (revision a9148abd9da5db2f1c682fb17bed791845fc41c9)
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_mac.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/eventhandler.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/kthread.h>
51 #include <sys/sysctl.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/pioctl.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sched.h>
60 #include <sys/syscall.h>
61 #include <sys/vmmeter.h>
62 #include <sys/vnode.h>
63 #include <sys/acct.h>
64 #include <sys/ktr.h>
65 #include <sys/ktrace.h>
66 #include <sys/unistd.h>
67 #include <sys/sdt.h>
68 #include <sys/sx.h>
69 #include <sys/signalvar.h>
70 
71 #include <security/audit/audit.h>
72 #include <security/mac/mac_framework.h>
73 
74 #include <vm/vm.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_extern.h>
78 #include <vm/uma.h>
79 
80 #ifdef KDTRACE_HOOKS
81 #include <sys/dtrace_bsd.h>
82 dtrace_fork_func_t	dtrace_fasttrap_fork;
83 #endif
84 
85 SDT_PROVIDER_DECLARE(proc);
86 SDT_PROBE_DEFINE(proc, kernel, , create);
87 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
88 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
90 
91 #ifndef _SYS_SYSPROTO_H_
92 struct fork_args {
93 	int     dummy;
94 };
95 #endif
96 
97 /* ARGSUSED */
98 int
99 fork(td, uap)
100 	struct thread *td;
101 	struct fork_args *uap;
102 {
103 	int error;
104 	struct proc *p2;
105 
106 	error = fork1(td, RFFDG | RFPROC, 0, &p2);
107 	if (error == 0) {
108 		td->td_retval[0] = p2->p_pid;
109 		td->td_retval[1] = 0;
110 	}
111 	return (error);
112 }
113 
114 /* ARGSUSED */
115 int
116 vfork(td, uap)
117 	struct thread *td;
118 	struct vfork_args *uap;
119 {
120 	int error, flags;
121 	struct proc *p2;
122 
123 #ifdef XEN
124 	flags = RFFDG | RFPROC; /* validate that this is still an issue */
125 #else
126 	flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
127 #endif
128 	error = fork1(td, flags, 0, &p2);
129 	if (error == 0) {
130 		td->td_retval[0] = p2->p_pid;
131 		td->td_retval[1] = 0;
132 	}
133 	return (error);
134 }
135 
136 int
137 rfork(td, uap)
138 	struct thread *td;
139 	struct rfork_args *uap;
140 {
141 	struct proc *p2;
142 	int error;
143 
144 	/* Don't allow kernel-only flags. */
145 	if ((uap->flags & RFKERNELONLY) != 0)
146 		return (EINVAL);
147 
148 	AUDIT_ARG(fflags, uap->flags);
149 	error = fork1(td, uap->flags, 0, &p2);
150 	if (error == 0) {
151 		td->td_retval[0] = p2 ? p2->p_pid : 0;
152 		td->td_retval[1] = 0;
153 	}
154 	return (error);
155 }
156 
157 int	nprocs = 1;		/* process 0 */
158 int	lastpid = 0;
159 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
160     "Last used PID");
161 
162 /*
163  * Random component to lastpid generation.  We mix in a random factor to make
164  * it a little harder to predict.  We sanity check the modulus value to avoid
165  * doing it in critical paths.  Don't let it be too small or we pointlessly
166  * waste randomness entropy, and don't let it be impossibly large.  Using a
167  * modulus that is too big causes a LOT more process table scans and slows
168  * down fork processing as the pidchecked caching is defeated.
169  */
170 static int randompid = 0;
171 
172 static int
173 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
174 {
175 	int error, pid;
176 
177 	error = sysctl_wire_old_buffer(req, sizeof(int));
178 	if (error != 0)
179 		return(error);
180 	sx_xlock(&allproc_lock);
181 	pid = randompid;
182 	error = sysctl_handle_int(oidp, &pid, 0, req);
183 	if (error == 0 && req->newptr != NULL) {
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 	}
192 	sx_xunlock(&allproc_lock);
193 	return (error);
194 }
195 
196 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
197     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
198 
199 int
200 fork1(td, flags, pages, procp)
201 	struct thread *td;
202 	int flags;
203 	int pages;
204 	struct proc **procp;
205 {
206 	struct proc *p1, *p2, *pptr;
207 	struct proc *newproc;
208 	int ok, trypid;
209 	static int curfail, pidchecked = 0;
210 	static struct timeval lastfail;
211 	struct filedesc *fd;
212 	struct filedesc_to_leader *fdtol;
213 	struct thread *td2;
214 	struct sigacts *newsigacts;
215 	struct vmspace *vm2;
216 	int error;
217 
218 	/* Can't copy and clear. */
219 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
220 		return (EINVAL);
221 
222 	p1 = td->td_proc;
223 
224 	/*
225 	 * Here we don't create a new process, but we divorce
226 	 * certain parts of a process from itself.
227 	 */
228 	if ((flags & RFPROC) == 0) {
229 		if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
230 		    (flags & (RFCFDG | RFFDG))) {
231 			PROC_LOCK(p1);
232 			if (thread_single(SINGLE_BOUNDARY)) {
233 				PROC_UNLOCK(p1);
234 				return (ERESTART);
235 			}
236 			PROC_UNLOCK(p1);
237 		}
238 
239 		error = vm_forkproc(td, NULL, NULL, NULL, flags);
240 		if (error)
241 			goto norfproc_fail;
242 
243 		/*
244 		 * Close all file descriptors.
245 		 */
246 		if (flags & RFCFDG) {
247 			struct filedesc *fdtmp;
248 			fdtmp = fdinit(td->td_proc->p_fd);
249 			fdfree(td);
250 			p1->p_fd = fdtmp;
251 		}
252 
253 		/*
254 		 * Unshare file descriptors (from parent).
255 		 */
256 		if (flags & RFFDG)
257 			fdunshare(p1, td);
258 
259 norfproc_fail:
260 		if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
261 		    (flags & (RFCFDG | RFFDG))) {
262 			PROC_LOCK(p1);
263 			thread_single_end();
264 			PROC_UNLOCK(p1);
265 		}
266 		*procp = NULL;
267 		return (error);
268 	}
269 
270 	/*
271 	 * XXX
272 	 * We did have single-threading code here
273 	 * however it proved un-needed and caused problems
274 	 */
275 
276 	vm2 = NULL;
277 	/* Allocate new proc. */
278 	newproc = uma_zalloc(proc_zone, M_WAITOK);
279 	if (TAILQ_EMPTY(&newproc->p_threads)) {
280 		td2 = thread_alloc();
281 		if (td2 == NULL) {
282 			error = ENOMEM;
283 			goto fail1;
284 		}
285 		proc_linkup(newproc, td2);
286 	} else
287 		td2 = FIRST_THREAD_IN_PROC(newproc);
288 
289 	/* Allocate and switch to an alternate kstack if specified. */
290 	if (pages != 0) {
291 		if (!vm_thread_new_altkstack(td2, pages)) {
292 			error = ENOMEM;
293 			goto fail1;
294 		}
295 	}
296 	if ((flags & RFMEM) == 0) {
297 		vm2 = vmspace_fork(p1->p_vmspace);
298 		if (vm2 == NULL) {
299 			error = ENOMEM;
300 			goto fail1;
301 		}
302 	}
303 #ifdef MAC
304 	mac_proc_init(newproc);
305 #endif
306 	knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
307 	STAILQ_INIT(&newproc->p_ktr);
308 
309 	/* We have to lock the process tree while we look for a pid. */
310 	sx_slock(&proctree_lock);
311 
312 	/*
313 	 * Although process entries are dynamically created, we still keep
314 	 * a global limit on the maximum number we will create.  Don't allow
315 	 * a nonprivileged user to use the last ten processes; don't let root
316 	 * exceed the limit. The variable nprocs is the current number of
317 	 * processes, maxproc is the limit.
318 	 */
319 	sx_xlock(&allproc_lock);
320 	if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
321 	    PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
322 		error = EAGAIN;
323 		goto fail;
324 	}
325 
326 	/*
327 	 * Increment the count of procs running with this uid. Don't allow
328 	 * a nonprivileged user to exceed their current limit.
329 	 *
330 	 * XXXRW: Can we avoid privilege here if it's not needed?
331 	 */
332 	error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
333 	if (error == 0)
334 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
335 	else {
336 		PROC_LOCK(p1);
337 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
338 		    lim_cur(p1, RLIMIT_NPROC));
339 		PROC_UNLOCK(p1);
340 	}
341 	if (!ok) {
342 		error = EAGAIN;
343 		goto fail;
344 	}
345 
346 	/*
347 	 * Increment the nprocs resource before blocking can occur.  There
348 	 * are hard-limits as to the number of processes that can run.
349 	 */
350 	nprocs++;
351 
352 	/*
353 	 * Find an unused process ID.  We remember a range of unused IDs
354 	 * ready to use (from lastpid+1 through pidchecked-1).
355 	 *
356 	 * If RFHIGHPID is set (used during system boot), do not allocate
357 	 * low-numbered pids.
358 	 */
359 	trypid = lastpid + 1;
360 	if (flags & RFHIGHPID) {
361 		if (trypid < 10)
362 			trypid = 10;
363 	} else {
364 		if (randompid)
365 			trypid += arc4random() % randompid;
366 	}
367 retry:
368 	/*
369 	 * If the process ID prototype has wrapped around,
370 	 * restart somewhat above 0, as the low-numbered procs
371 	 * tend to include daemons that don't exit.
372 	 */
373 	if (trypid >= PID_MAX) {
374 		trypid = trypid % PID_MAX;
375 		if (trypid < 100)
376 			trypid += 100;
377 		pidchecked = 0;
378 	}
379 	if (trypid >= pidchecked) {
380 		int doingzomb = 0;
381 
382 		pidchecked = PID_MAX;
383 		/*
384 		 * Scan the active and zombie procs to check whether this pid
385 		 * is in use.  Remember the lowest pid that's greater
386 		 * than trypid, so we can avoid checking for a while.
387 		 */
388 		p2 = LIST_FIRST(&allproc);
389 again:
390 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
391 			while (p2->p_pid == trypid ||
392 			    (p2->p_pgrp != NULL &&
393 			    (p2->p_pgrp->pg_id == trypid ||
394 			    (p2->p_session != NULL &&
395 			    p2->p_session->s_sid == trypid)))) {
396 				trypid++;
397 				if (trypid >= pidchecked)
398 					goto retry;
399 			}
400 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
401 				pidchecked = p2->p_pid;
402 			if (p2->p_pgrp != NULL) {
403 				if (p2->p_pgrp->pg_id > trypid &&
404 				    pidchecked > p2->p_pgrp->pg_id)
405 					pidchecked = p2->p_pgrp->pg_id;
406 				if (p2->p_session != NULL &&
407 				    p2->p_session->s_sid > trypid &&
408 				    pidchecked > p2->p_session->s_sid)
409 					pidchecked = p2->p_session->s_sid;
410 			}
411 		}
412 		if (!doingzomb) {
413 			doingzomb = 1;
414 			p2 = LIST_FIRST(&zombproc);
415 			goto again;
416 		}
417 	}
418 	sx_sunlock(&proctree_lock);
419 
420 	/*
421 	 * RFHIGHPID does not mess with the lastpid counter during boot.
422 	 */
423 	if (flags & RFHIGHPID)
424 		pidchecked = 0;
425 	else
426 		lastpid = trypid;
427 
428 	p2 = newproc;
429 	p2->p_state = PRS_NEW;		/* protect against others */
430 	p2->p_pid = trypid;
431 	/*
432 	 * Allow the scheduler to initialize the child.
433 	 */
434 	thread_lock(td);
435 	sched_fork(td, td2);
436 	thread_unlock(td);
437 	AUDIT_ARG(pid, p2->p_pid);
438 	LIST_INSERT_HEAD(&allproc, p2, p_list);
439 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
440 
441 	PROC_LOCK(p2);
442 	PROC_LOCK(p1);
443 
444 	sx_xunlock(&allproc_lock);
445 
446 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
447 	    __rangeof(struct proc, p_startcopy, p_endcopy));
448 	pargs_hold(p2->p_args);
449 	PROC_UNLOCK(p1);
450 
451 	bzero(&p2->p_startzero,
452 	    __rangeof(struct proc, p_startzero, p_endzero));
453 
454 	p2->p_ucred = crhold(td->td_ucred);
455 	PROC_UNLOCK(p2);
456 
457 	/*
458 	 * Malloc things while we don't hold any locks.
459 	 */
460 	if (flags & RFSIGSHARE)
461 		newsigacts = NULL;
462 	else
463 		newsigacts = sigacts_alloc();
464 
465 	/*
466 	 * Copy filedesc.
467 	 */
468 	if (flags & RFCFDG) {
469 		fd = fdinit(p1->p_fd);
470 		fdtol = NULL;
471 	} else if (flags & RFFDG) {
472 		fd = fdcopy(p1->p_fd);
473 		fdtol = NULL;
474 	} else {
475 		fd = fdshare(p1->p_fd);
476 		if (p1->p_fdtol == NULL)
477 			p1->p_fdtol =
478 				filedesc_to_leader_alloc(NULL,
479 							 NULL,
480 							 p1->p_leader);
481 		if ((flags & RFTHREAD) != 0) {
482 			/*
483 			 * Shared file descriptor table and
484 			 * shared process leaders.
485 			 */
486 			fdtol = p1->p_fdtol;
487 			FILEDESC_XLOCK(p1->p_fd);
488 			fdtol->fdl_refcount++;
489 			FILEDESC_XUNLOCK(p1->p_fd);
490 		} else {
491 			/*
492 			 * Shared file descriptor table, and
493 			 * different process leaders
494 			 */
495 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
496 							 p1->p_fd,
497 							 p2);
498 		}
499 	}
500 	/*
501 	 * Make a proc table entry for the new process.
502 	 * Start by zeroing the section of proc that is zero-initialized,
503 	 * then copy the section that is copied directly from the parent.
504 	 */
505 
506 	PROC_LOCK(p2);
507 	PROC_LOCK(p1);
508 
509 	bzero(&td2->td_startzero,
510 	    __rangeof(struct thread, td_startzero, td_endzero));
511 
512 	bcopy(&td->td_startcopy, &td2->td_startcopy,
513 	    __rangeof(struct thread, td_startcopy, td_endcopy));
514 
515 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
516 	td2->td_sigstk = td->td_sigstk;
517 	td2->td_sigmask = td->td_sigmask;
518 	td2->td_flags = TDF_INMEM;
519 
520 	/*
521 	 * Duplicate sub-structures as needed.
522 	 * Increase reference counts on shared objects.
523 	 */
524 	p2->p_flag = P_INMEM;
525 	p2->p_swtick = ticks;
526 	if (p1->p_flag & P_PROFIL)
527 		startprofclock(p2);
528 	td2->td_ucred = crhold(p2->p_ucred);
529 
530 	if (flags & RFSIGSHARE) {
531 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
532 	} else {
533 		sigacts_copy(newsigacts, p1->p_sigacts);
534 		p2->p_sigacts = newsigacts;
535 	}
536 	if (flags & RFLINUXTHPN)
537 	        p2->p_sigparent = SIGUSR1;
538 	else
539 	        p2->p_sigparent = SIGCHLD;
540 
541 	p2->p_textvp = p1->p_textvp;
542 	p2->p_fd = fd;
543 	p2->p_fdtol = fdtol;
544 
545 	/*
546 	 * p_limit is copy-on-write.  Bump its refcount.
547 	 */
548 	lim_fork(p1, p2);
549 
550 	pstats_fork(p1->p_stats, p2->p_stats);
551 
552 	PROC_UNLOCK(p1);
553 	PROC_UNLOCK(p2);
554 
555 	/* Bump references to the text vnode (for procfs) */
556 	if (p2->p_textvp)
557 		vref(p2->p_textvp);
558 
559 	/*
560 	 * Set up linkage for kernel based threading.
561 	 */
562 	if ((flags & RFTHREAD) != 0) {
563 		mtx_lock(&ppeers_lock);
564 		p2->p_peers = p1->p_peers;
565 		p1->p_peers = p2;
566 		p2->p_leader = p1->p_leader;
567 		mtx_unlock(&ppeers_lock);
568 		PROC_LOCK(p1->p_leader);
569 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
570 			PROC_UNLOCK(p1->p_leader);
571 			/*
572 			 * The task leader is exiting, so process p1 is
573 			 * going to be killed shortly.  Since p1 obviously
574 			 * isn't dead yet, we know that the leader is either
575 			 * sending SIGKILL's to all the processes in this
576 			 * task or is sleeping waiting for all the peers to
577 			 * exit.  We let p1 complete the fork, but we need
578 			 * to go ahead and kill the new process p2 since
579 			 * the task leader may not get a chance to send
580 			 * SIGKILL to it.  We leave it on the list so that
581 			 * the task leader will wait for this new process
582 			 * to commit suicide.
583 			 */
584 			PROC_LOCK(p2);
585 			psignal(p2, SIGKILL);
586 			PROC_UNLOCK(p2);
587 		} else
588 			PROC_UNLOCK(p1->p_leader);
589 	} else {
590 		p2->p_peers = NULL;
591 		p2->p_leader = p2;
592 	}
593 
594 	sx_xlock(&proctree_lock);
595 	PGRP_LOCK(p1->p_pgrp);
596 	PROC_LOCK(p2);
597 	PROC_LOCK(p1);
598 
599 	/*
600 	 * Preserve some more flags in subprocess.  P_PROFIL has already
601 	 * been preserved.
602 	 */
603 	p2->p_flag |= p1->p_flag & P_SUGID;
604 	td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
605 	SESS_LOCK(p1->p_session);
606 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
607 		p2->p_flag |= P_CONTROLT;
608 	SESS_UNLOCK(p1->p_session);
609 	if (flags & RFPPWAIT)
610 		p2->p_flag |= P_PPWAIT;
611 
612 	p2->p_pgrp = p1->p_pgrp;
613 	LIST_INSERT_AFTER(p1, p2, p_pglist);
614 	PGRP_UNLOCK(p1->p_pgrp);
615 	LIST_INIT(&p2->p_children);
616 
617 	callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
618 
619 #ifdef KTRACE
620 	/*
621 	 * Copy traceflag and tracefile if enabled.
622 	 */
623 	mtx_lock(&ktrace_mtx);
624 	KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
625 	if (p1->p_traceflag & KTRFAC_INHERIT) {
626 		p2->p_traceflag = p1->p_traceflag;
627 		if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
628 			VREF(p2->p_tracevp);
629 			KASSERT(p1->p_tracecred != NULL,
630 			    ("ktrace vnode with no cred"));
631 			p2->p_tracecred = crhold(p1->p_tracecred);
632 		}
633 	}
634 	mtx_unlock(&ktrace_mtx);
635 #endif
636 
637 	/*
638 	 * If PF_FORK is set, the child process inherits the
639 	 * procfs ioctl flags from its parent.
640 	 */
641 	if (p1->p_pfsflags & PF_FORK) {
642 		p2->p_stops = p1->p_stops;
643 		p2->p_pfsflags = p1->p_pfsflags;
644 	}
645 
646 #ifdef KDTRACE_HOOKS
647 	/*
648 	 * Tell the DTrace fasttrap provider about the new process
649 	 * if it has registered an interest.
650 	 */
651 	if (dtrace_fasttrap_fork)
652 		dtrace_fasttrap_fork(p1, p2);
653 #endif
654 
655 	/*
656 	 * This begins the section where we must prevent the parent
657 	 * from being swapped.
658 	 */
659 	_PHOLD(p1);
660 	PROC_UNLOCK(p1);
661 
662 	/*
663 	 * Attach the new process to its parent.
664 	 *
665 	 * If RFNOWAIT is set, the newly created process becomes a child
666 	 * of init.  This effectively disassociates the child from the
667 	 * parent.
668 	 */
669 	if (flags & RFNOWAIT)
670 		pptr = initproc;
671 	else
672 		pptr = p1;
673 	p2->p_pptr = pptr;
674 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
675 	sx_xunlock(&proctree_lock);
676 
677 	/* Inform accounting that we have forked. */
678 	p2->p_acflag = AFORK;
679 	PROC_UNLOCK(p2);
680 
681 	/*
682 	 * Finish creating the child process.  It will return via a different
683 	 * execution path later.  (ie: directly into user mode)
684 	 */
685 	vm_forkproc(td, p2, td2, vm2, flags);
686 
687 	if (flags == (RFFDG | RFPROC)) {
688 		PCPU_INC(cnt.v_forks);
689 		PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
690 		    p2->p_vmspace->vm_ssize);
691 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
692 		PCPU_INC(cnt.v_vforks);
693 		PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
694 		    p2->p_vmspace->vm_ssize);
695 	} else if (p1 == &proc0) {
696 		PCPU_INC(cnt.v_kthreads);
697 		PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
698 		    p2->p_vmspace->vm_ssize);
699 	} else {
700 		PCPU_INC(cnt.v_rforks);
701 		PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
702 		    p2->p_vmspace->vm_ssize);
703 	}
704 
705 	/*
706 	 * Both processes are set up, now check if any loadable modules want
707 	 * to adjust anything.
708 	 *   What if they have an error? XXX
709 	 */
710 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
711 
712 	/*
713 	 * Set the child start time and mark the process as being complete.
714 	 */
715 	microuptime(&p2->p_stats->p_start);
716 	PROC_SLOCK(p2);
717 	p2->p_state = PRS_NORMAL;
718 	PROC_SUNLOCK(p2);
719 
720 	/*
721 	 * If RFSTOPPED not requested, make child runnable and add to
722 	 * run queue.
723 	 */
724 	if ((flags & RFSTOPPED) == 0) {
725 		thread_lock(td2);
726 		TD_SET_CAN_RUN(td2);
727 		sched_add(td2, SRQ_BORING);
728 		thread_unlock(td2);
729 	}
730 
731 	/*
732 	 * Now can be swapped.
733 	 */
734 	PROC_LOCK(p1);
735 	_PRELE(p1);
736 	PROC_UNLOCK(p1);
737 
738 	/*
739 	 * Tell any interested parties about the new process.
740 	 */
741 	knote_fork(&p1->p_klist, p2->p_pid);
742 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
743 
744 	/*
745 	 * Preserve synchronization semantics of vfork.  If waiting for
746 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
747 	 * proc (in case of exit).
748 	 */
749 	PROC_LOCK(p2);
750 	while (p2->p_flag & P_PPWAIT)
751 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
752 	PROC_UNLOCK(p2);
753 
754 	/*
755 	 * Return child proc pointer to parent.
756 	 */
757 	*procp = p2;
758 	return (0);
759 fail:
760 	sx_sunlock(&proctree_lock);
761 	if (ppsratecheck(&lastfail, &curfail, 1))
762 		printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
763 		    td->td_ucred->cr_ruid);
764 	sx_xunlock(&allproc_lock);
765 #ifdef MAC
766 	mac_proc_destroy(newproc);
767 #endif
768 fail1:
769 	if (vm2 != NULL)
770 		vmspace_free(vm2);
771 	uma_zfree(proc_zone, newproc);
772 	pause("fork", hz / 2);
773 	return (error);
774 }
775 
776 /*
777  * Handle the return of a child process from fork1().  This function
778  * is called from the MD fork_trampoline() entry point.
779  */
780 void
781 fork_exit(callout, arg, frame)
782 	void (*callout)(void *, struct trapframe *);
783 	void *arg;
784 	struct trapframe *frame;
785 {
786 	struct proc *p;
787 	struct thread *td;
788 	struct thread *dtd;
789 
790 	td = curthread;
791 	p = td->td_proc;
792 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
793 
794 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
795 		td, td->td_sched, p->p_pid, td->td_name);
796 
797 	sched_fork_exit(td);
798 	/*
799 	* Processes normally resume in mi_switch() after being
800 	* cpu_switch()'ed to, but when children start up they arrive here
801 	* instead, so we must do much the same things as mi_switch() would.
802 	*/
803 	if ((dtd = PCPU_GET(deadthread))) {
804 		PCPU_SET(deadthread, NULL);
805 		thread_stash(dtd);
806 	}
807 	thread_unlock(td);
808 
809 	/*
810 	 * cpu_set_fork_handler intercepts this function call to
811 	 * have this call a non-return function to stay in kernel mode.
812 	 * initproc has its own fork handler, but it does return.
813 	 */
814 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
815 	callout(arg, frame);
816 
817 	/*
818 	 * Check if a kernel thread misbehaved and returned from its main
819 	 * function.
820 	 */
821 	if (p->p_flag & P_KTHREAD) {
822 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
823 		    td->td_name, p->p_pid);
824 		kproc_exit(0);
825 	}
826 	mtx_assert(&Giant, MA_NOTOWNED);
827 
828 	EVENTHANDLER_INVOKE(schedtail, p);
829 }
830 
831 /*
832  * Simplified back end of syscall(), used when returning from fork()
833  * directly into user mode.  Giant is not held on entry, and must not
834  * be held on return.  This function is passed in to fork_exit() as the
835  * first parameter and is called when returning to a new userland process.
836  */
837 void
838 fork_return(td, frame)
839 	struct thread *td;
840 	struct trapframe *frame;
841 {
842 
843 	userret(td, frame);
844 #ifdef KTRACE
845 	if (KTRPOINT(td, KTR_SYSRET))
846 		ktrsysret(SYS_fork, 0, 0);
847 #endif
848 	mtx_assert(&Giant, MA_NOTOWNED);
849 }
850