xref: /freebsd/sys/kern/kern_fork.c (revision 63f9a4cb2684a303e3eb2ffed39c03a2e2b28ae0)
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_ktrace.h"
41 #include "opt_mac.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/eventhandler.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/kthread.h>
50 #include <sys/sysctl.h>
51 #include <sys/lock.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.h>
55 #include <sys/pioctl.h>
56 #include <sys/resourcevar.h>
57 #include <sys/sched.h>
58 #include <sys/syscall.h>
59 #include <sys/vmmeter.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/unistd.h>
66 #include <sys/sx.h>
67 #include <sys/signalvar.h>
68 
69 #include <vm/vm.h>
70 #include <vm/pmap.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_extern.h>
73 #include <vm/uma.h>
74 
75 #include <machine/critical.h>
76 
77 #ifndef _SYS_SYSPROTO_H_
78 struct fork_args {
79 	int     dummy;
80 };
81 #endif
82 
83 static int forksleep; /* Place for fork1() to sleep on. */
84 
85 /*
86  * MPSAFE
87  */
88 /* ARGSUSED */
89 int
90 fork(td, uap)
91 	struct thread *td;
92 	struct fork_args *uap;
93 {
94 	int error;
95 	struct proc *p2;
96 
97 	error = fork1(td, RFFDG | RFPROC, 0, &p2);
98 	if (error == 0) {
99 		td->td_retval[0] = p2->p_pid;
100 		td->td_retval[1] = 0;
101 	}
102 	return (error);
103 }
104 
105 /*
106  * MPSAFE
107  */
108 /* ARGSUSED */
109 int
110 vfork(td, uap)
111 	struct thread *td;
112 	struct vfork_args *uap;
113 {
114 	int error;
115 	struct proc *p2;
116 
117 	error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
118 	if (error == 0) {
119 		td->td_retval[0] = p2->p_pid;
120 		td->td_retval[1] = 0;
121 	}
122 	return (error);
123 }
124 
125 /*
126  * MPSAFE
127  */
128 int
129 rfork(td, uap)
130 	struct thread *td;
131 	struct rfork_args *uap;
132 {
133 	struct proc *p2;
134 	int error;
135 
136 	/* Don't allow kernel-only flags. */
137 	if ((uap->flags & RFKERNELONLY) != 0)
138 		return (EINVAL);
139 
140 	error = fork1(td, uap->flags, 0, &p2);
141 	if (error == 0) {
142 		td->td_retval[0] = p2 ? p2->p_pid : 0;
143 		td->td_retval[1] = 0;
144 	}
145 	return (error);
146 }
147 
148 int	nprocs = 1;		/* process 0 */
149 int	lastpid = 0;
150 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
151     "Last used PID");
152 
153 /*
154  * Random component to lastpid generation.  We mix in a random factor to make
155  * it a little harder to predict.  We sanity check the modulus value to avoid
156  * doing it in critical paths.  Don't let it be too small or we pointlessly
157  * waste randomness entropy, and don't let it be impossibly large.  Using a
158  * modulus that is too big causes a LOT more process table scans and slows
159  * down fork processing as the pidchecked caching is defeated.
160  */
161 static int randompid = 0;
162 
163 static int
164 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
165 {
166 	int error, pid;
167 
168 	error = sysctl_wire_old_buffer(req, sizeof(int));
169 	if (error != 0)
170 		return(error);
171 	sx_xlock(&allproc_lock);
172 	pid = randompid;
173 	error = sysctl_handle_int(oidp, &pid, 0, req);
174 	if (error == 0 && req->newptr != NULL) {
175 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
176 			pid = PID_MAX - 100;
177 		else if (pid < 2)			/* NOP */
178 			pid = 0;
179 		else if (pid < 100)			/* Make it reasonable */
180 			pid = 100;
181 		randompid = pid;
182 	}
183 	sx_xunlock(&allproc_lock);
184 	return (error);
185 }
186 
187 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
188     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
189 
190 int
191 fork1(td, flags, pages, procp)
192 	struct thread *td;
193 	int flags;
194 	int pages;
195 	struct proc **procp;
196 {
197 	struct proc *p1, *p2, *pptr;
198 	uid_t uid;
199 	struct proc *newproc;
200 	int ok, trypid;
201 	static int curfail, pidchecked = 0;
202 	static struct timeval lastfail;
203 	struct filedesc *fd;
204 	struct filedesc_to_leader *fdtol;
205 	struct thread *td2;
206 	struct ksegrp *kg2;
207 	struct sigacts *newsigacts;
208 	int error;
209 
210 	/* Can't copy and clear. */
211 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
212 		return (EINVAL);
213 
214 	p1 = td->td_proc;
215 
216 	/*
217 	 * Here we don't create a new process, but we divorce
218 	 * certain parts of a process from itself.
219 	 */
220 	if ((flags & RFPROC) == 0) {
221 		vm_forkproc(td, NULL, NULL, flags);
222 
223 		/*
224 		 * Close all file descriptors.
225 		 */
226 		if (flags & RFCFDG) {
227 			struct filedesc *fdtmp;
228 			fdtmp = fdinit(td->td_proc->p_fd);
229 			fdfree(td);
230 			p1->p_fd = fdtmp;
231 		}
232 
233 		/*
234 		 * Unshare file descriptors (from parent).
235 		 */
236 		if (flags & RFFDG) {
237 			FILEDESC_LOCK_FAST(p1->p_fd);
238 			if (p1->p_fd->fd_refcnt > 1) {
239 				struct filedesc *newfd;
240 
241 				FILEDESC_UNLOCK_FAST(p1->p_fd);
242 				newfd = fdcopy(p1->p_fd);
243 				fdfree(td);
244 				p1->p_fd = newfd;
245 			} else
246 				FILEDESC_UNLOCK_FAST(p1->p_fd);
247 		}
248 		*procp = NULL;
249 		return (0);
250 	}
251 
252 	/*
253 	 * Note 1:1 allows for forking with one thread coming out on the
254 	 * other side with the expectation that the process is about to
255 	 * exec.
256 	 */
257 	if (p1->p_flag & P_HADTHREADS) {
258 		/*
259 		 * Idle the other threads for a second.
260 		 * Since the user space is copied, it must remain stable.
261 		 * In addition, all threads (from the user perspective)
262 		 * need to either be suspended or in the kernel,
263 		 * where they will try restart in the parent and will
264 		 * be aborted in the child.
265 		 */
266 		PROC_LOCK(p1);
267 		if (thread_single(SINGLE_NO_EXIT)) {
268 			/* Abort. Someone else is single threading before us. */
269 			PROC_UNLOCK(p1);
270 			return (ERESTART);
271 		}
272 		PROC_UNLOCK(p1);
273 		/*
274 		 * All other activity in this process
275 		 * is now suspended at the user boundary,
276 		 * (or other safe places if we think of any).
277 		 */
278 	}
279 
280 	/* Allocate new proc. */
281 	newproc = uma_zalloc(proc_zone, M_WAITOK);
282 #ifdef MAC
283 	mac_init_proc(newproc);
284 #endif
285 	knlist_init(&newproc->p_klist, &newproc->p_mtx);
286 
287 	/* We have to lock the process tree while we look for a pid. */
288 	sx_slock(&proctree_lock);
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 &&
300 	    suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
301 	    nprocs >= maxproc) {
302 		error = EAGAIN;
303 		goto fail;
304 	}
305 
306 	/*
307 	 * Increment the count of procs running with this uid. Don't allow
308 	 * a nonprivileged user to exceed their current limit.
309 	 */
310 	PROC_LOCK(p1);
311 	ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
312 		(uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0);
313 	PROC_UNLOCK(p1);
314 	if (!ok) {
315 		error = EAGAIN;
316 		goto fail;
317 	}
318 
319 	/*
320 	 * Increment the nprocs resource before blocking can occur.  There
321 	 * are hard-limits as to the number of processes that can run.
322 	 */
323 	nprocs++;
324 
325 	/*
326 	 * Find an unused process ID.  We remember a range of unused IDs
327 	 * ready to use (from lastpid+1 through pidchecked-1).
328 	 *
329 	 * If RFHIGHPID is set (used during system boot), do not allocate
330 	 * low-numbered pids.
331 	 */
332 	trypid = lastpid + 1;
333 	if (flags & RFHIGHPID) {
334 		if (trypid < 10)
335 			trypid = 10;
336 	} else {
337 		if (randompid)
338 			trypid += arc4random() % randompid;
339 	}
340 retry:
341 	/*
342 	 * If the process ID prototype has wrapped around,
343 	 * restart somewhat above 0, as the low-numbered procs
344 	 * tend to include daemons that don't exit.
345 	 */
346 	if (trypid >= PID_MAX) {
347 		trypid = trypid % PID_MAX;
348 		if (trypid < 100)
349 			trypid += 100;
350 		pidchecked = 0;
351 	}
352 	if (trypid >= pidchecked) {
353 		int doingzomb = 0;
354 
355 		pidchecked = PID_MAX;
356 		/*
357 		 * Scan the active and zombie procs to check whether this pid
358 		 * is in use.  Remember the lowest pid that's greater
359 		 * than trypid, so we can avoid checking for a while.
360 		 */
361 		p2 = LIST_FIRST(&allproc);
362 again:
363 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
364 			PROC_LOCK(p2);
365 			while (p2->p_pid == trypid ||
366 			    (p2->p_pgrp != NULL &&
367 			    (p2->p_pgrp->pg_id == trypid ||
368 			    (p2->p_session != NULL &&
369 			    p2->p_session->s_sid == trypid)))) {
370 				trypid++;
371 				if (trypid >= pidchecked) {
372 					PROC_UNLOCK(p2);
373 					goto retry;
374 				}
375 			}
376 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
377 				pidchecked = p2->p_pid;
378 			if (p2->p_pgrp != NULL) {
379 				if (p2->p_pgrp->pg_id > trypid &&
380 				    pidchecked > p2->p_pgrp->pg_id)
381 					pidchecked = p2->p_pgrp->pg_id;
382 				if (p2->p_session != NULL &&
383 				    p2->p_session->s_sid > trypid &&
384 				    pidchecked > p2->p_session->s_sid)
385 					pidchecked = p2->p_session->s_sid;
386 			}
387 			PROC_UNLOCK(p2);
388 		}
389 		if (!doingzomb) {
390 			doingzomb = 1;
391 			p2 = LIST_FIRST(&zombproc);
392 			goto again;
393 		}
394 	}
395 	sx_sunlock(&proctree_lock);
396 
397 	/*
398 	 * RFHIGHPID does not mess with the lastpid counter during boot.
399 	 */
400 	if (flags & RFHIGHPID)
401 		pidchecked = 0;
402 	else
403 		lastpid = trypid;
404 
405 	p2 = newproc;
406 	p2->p_state = PRS_NEW;		/* protect against others */
407 	p2->p_pid = trypid;
408 	LIST_INSERT_HEAD(&allproc, p2, p_list);
409 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
410 	sx_xunlock(&allproc_lock);
411 
412 	/*
413 	 * Malloc things while we don't hold any locks.
414 	 */
415 	if (flags & RFSIGSHARE)
416 		newsigacts = NULL;
417 	else
418 		newsigacts = sigacts_alloc();
419 
420 	/*
421 	 * Copy filedesc.
422 	 */
423 	if (flags & RFCFDG) {
424 		fd = fdinit(p1->p_fd);
425 		fdtol = NULL;
426 	} else if (flags & RFFDG) {
427 		fd = fdcopy(p1->p_fd);
428 		fdtol = NULL;
429 	} else {
430 		fd = fdshare(p1->p_fd);
431 		if (p1->p_fdtol == NULL)
432 			p1->p_fdtol =
433 				filedesc_to_leader_alloc(NULL,
434 							 NULL,
435 							 p1->p_leader);
436 		if ((flags & RFTHREAD) != 0) {
437 			/*
438 			 * Shared file descriptor table and
439 			 * shared process leaders.
440 			 */
441 			fdtol = p1->p_fdtol;
442 			FILEDESC_LOCK_FAST(p1->p_fd);
443 			fdtol->fdl_refcount++;
444 			FILEDESC_UNLOCK_FAST(p1->p_fd);
445 		} else {
446 			/*
447 			 * Shared file descriptor table, and
448 			 * different process leaders
449 			 */
450 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
451 							 p1->p_fd,
452 							 p2);
453 		}
454 	}
455 	/*
456 	 * Make a proc table entry for the new process.
457 	 * Start by zeroing the section of proc that is zero-initialized,
458 	 * then copy the section that is copied directly from the parent.
459 	 */
460 	td2 = FIRST_THREAD_IN_PROC(p2);
461 	kg2 = FIRST_KSEGRP_IN_PROC(p2);
462 
463 	/* Allocate and switch to an alternate kstack if specified. */
464 	if (pages != 0)
465 		vm_thread_new_altkstack(td2, pages);
466 
467 	PROC_LOCK(p2);
468 	PROC_LOCK(p1);
469 
470 	bzero(&p2->p_startzero,
471 	    __rangeof(struct proc, p_startzero, p_endzero));
472 	bzero(&td2->td_startzero,
473 	    __rangeof(struct thread, td_startzero, td_endzero));
474 	bzero(&kg2->kg_startzero,
475 	    __rangeof(struct ksegrp, kg_startzero, kg_endzero));
476 
477 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
478 	    __rangeof(struct proc, p_startcopy, p_endcopy));
479 	bcopy(&td->td_startcopy, &td2->td_startcopy,
480 	    __rangeof(struct thread, td_startcopy, td_endcopy));
481 	bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
482 	    __rangeof(struct ksegrp, kg_startcopy, kg_endcopy));
483 
484 	td2->td_sigstk = td->td_sigstk;
485 
486 	/*
487 	 * Duplicate sub-structures as needed.
488 	 * Increase reference counts on shared objects.
489 	 */
490 	p2->p_flag = 0;
491 	if (p1->p_flag & P_PROFIL)
492 		startprofclock(p2);
493 	mtx_lock_spin(&sched_lock);
494 	p2->p_sflag = PS_INMEM;
495 	/*
496 	 * Allow the scheduler to adjust the priority of the child and
497 	 * parent while we hold the sched_lock.
498 	 */
499 	sched_fork(td, td2);
500 
501 	mtx_unlock_spin(&sched_lock);
502 	p2->p_ucred = crhold(td->td_ucred);
503 	td2->td_ucred = crhold(p2->p_ucred);	/* XXXKSE */
504 
505 	pargs_hold(p2->p_args);
506 
507 	if (flags & RFSIGSHARE) {
508 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
509 	} else {
510 		sigacts_copy(newsigacts, p1->p_sigacts);
511 		p2->p_sigacts = newsigacts;
512 	}
513 	if (flags & RFLINUXTHPN)
514 	        p2->p_sigparent = SIGUSR1;
515 	else
516 	        p2->p_sigparent = SIGCHLD;
517 
518 	p2->p_textvp = p1->p_textvp;
519 	p2->p_fd = fd;
520 	p2->p_fdtol = fdtol;
521 
522 	/*
523 	 * p_limit is copy-on-write.  Bump its refcount.
524 	 */
525 	p2->p_limit = lim_hold(p1->p_limit);
526 
527 	pstats_fork(p1->p_stats, p2->p_stats);
528 
529 	PROC_UNLOCK(p1);
530 	PROC_UNLOCK(p2);
531 
532 	/* Bump references to the text vnode (for procfs) */
533 	if (p2->p_textvp)
534 		vref(p2->p_textvp);
535 
536 	/*
537 	 * Set up linkage for kernel based threading.
538 	 */
539 	if ((flags & RFTHREAD) != 0) {
540 		mtx_lock(&ppeers_lock);
541 		p2->p_peers = p1->p_peers;
542 		p1->p_peers = p2;
543 		p2->p_leader = p1->p_leader;
544 		mtx_unlock(&ppeers_lock);
545 		PROC_LOCK(p1->p_leader);
546 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
547 			PROC_UNLOCK(p1->p_leader);
548 			/*
549 			 * The task leader is exiting, so process p1 is
550 			 * going to be killed shortly.  Since p1 obviously
551 			 * isn't dead yet, we know that the leader is either
552 			 * sending SIGKILL's to all the processes in this
553 			 * task or is sleeping waiting for all the peers to
554 			 * exit.  We let p1 complete the fork, but we need
555 			 * to go ahead and kill the new process p2 since
556 			 * the task leader may not get a chance to send
557 			 * SIGKILL to it.  We leave it on the list so that
558 			 * the task leader will wait for this new process
559 			 * to commit suicide.
560 			 */
561 			PROC_LOCK(p2);
562 			psignal(p2, SIGKILL);
563 			PROC_UNLOCK(p2);
564 		} else
565 			PROC_UNLOCK(p1->p_leader);
566 	} else {
567 		p2->p_peers = NULL;
568 		p2->p_leader = p2;
569 	}
570 
571 	sx_xlock(&proctree_lock);
572 	PGRP_LOCK(p1->p_pgrp);
573 	PROC_LOCK(p2);
574 	PROC_LOCK(p1);
575 
576 	/*
577 	 * Preserve some more flags in subprocess.  P_PROFIL has already
578 	 * been preserved.
579 	 */
580 	p2->p_flag |= p1->p_flag & P_SUGID;
581 	td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
582 	SESS_LOCK(p1->p_session);
583 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
584 		p2->p_flag |= P_CONTROLT;
585 	SESS_UNLOCK(p1->p_session);
586 	if (flags & RFPPWAIT)
587 		p2->p_flag |= P_PPWAIT;
588 
589 	p2->p_pgrp = p1->p_pgrp;
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, CALLOUT_MPSAFE);
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 		atomic_add_int(&cnt.v_forks, 1);
657 		atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize +
658 		    p2->p_vmspace->vm_ssize);
659 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
660 		atomic_add_int(&cnt.v_vforks, 1);
661 		atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
662 		    p2->p_vmspace->vm_ssize);
663 	} else if (p1 == &proc0) {
664 		atomic_add_int(&cnt.v_kthreads, 1);
665 		atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
666 		    p2->p_vmspace->vm_ssize);
667 	} else {
668 		atomic_add_int(&cnt.v_rforks, 1);
669 		atomic_add_int(&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 	 * Set the child start time and mark the process as being complete.
682 	 */
683 	microuptime(&p2->p_stats->p_start);
684 	mtx_lock_spin(&sched_lock);
685 	p2->p_state = PRS_NORMAL;
686 
687 	/*
688 	 * If RFSTOPPED not requested, make child runnable and add to
689 	 * run queue.
690 	 */
691 	if ((flags & RFSTOPPED) == 0) {
692 		TD_SET_CAN_RUN(td2);
693 		setrunqueue(td2, SRQ_BORING);
694 	}
695 	mtx_unlock_spin(&sched_lock);
696 
697 	/*
698 	 * Now can be swapped.
699 	 */
700 	PROC_LOCK(p1);
701 	_PRELE(p1);
702 
703 	/*
704 	 * Tell any interested parties about the new process.
705 	 */
706 	KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
707 
708 	PROC_UNLOCK(p1);
709 
710 	/*
711 	 * Preserve synchronization semantics of vfork.  If waiting for
712 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
713 	 * proc (in case of exit).
714 	 */
715 	PROC_LOCK(p2);
716 	while (p2->p_flag & P_PPWAIT)
717 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
718 	PROC_UNLOCK(p2);
719 
720 	/*
721 	 * If other threads are waiting, let them continue now.
722 	 */
723 	if (p1->p_flag & P_HADTHREADS) {
724 		PROC_LOCK(p1);
725 		thread_single_end();
726 		PROC_UNLOCK(p1);
727 	}
728 
729 	/*
730 	 * Return child proc pointer to parent.
731 	 */
732 	*procp = p2;
733 	return (0);
734 fail:
735 	sx_sunlock(&proctree_lock);
736 	if (ppsratecheck(&lastfail, &curfail, 1))
737 		printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
738 			uid);
739 	sx_xunlock(&allproc_lock);
740 #ifdef MAC
741 	mac_destroy_proc(newproc);
742 #endif
743 	uma_zfree(proc_zone, newproc);
744 	if (p1->p_flag & P_HADTHREADS) {
745 		PROC_LOCK(p1);
746 		thread_single_end();
747 		PROC_UNLOCK(p1);
748 	}
749 	tsleep(&forksleep, PUSER, "fork", hz / 2);
750 	return (error);
751 }
752 
753 /*
754  * Handle the return of a child process from fork1().  This function
755  * is called from the MD fork_trampoline() entry point.
756  */
757 void
758 fork_exit(callout, arg, frame)
759 	void (*callout)(void *, struct trapframe *);
760 	void *arg;
761 	struct trapframe *frame;
762 {
763 	struct proc *p;
764 	struct thread *td;
765 
766 	/*
767 	 * Finish setting up thread glue so that it begins execution in a
768 	 * non-nested critical section with sched_lock held but not recursed.
769 	 */
770 	td = curthread;
771 	p = td->td_proc;
772 	td->td_oncpu = PCPU_GET(cpuid);
773 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
774 
775 	sched_lock.mtx_lock = (uintptr_t)td;
776 	mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
777 	cpu_critical_fork_exit();
778 	CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
779 		td, td->td_sched, p->p_pid, p->p_comm);
780 
781 	/*
782 	 * Processes normally resume in mi_switch() after being
783 	 * cpu_switch()'ed to, but when children start up they arrive here
784 	 * instead, so we must do much the same things as mi_switch() would.
785 	 */
786 
787 	if ((td = PCPU_GET(deadthread))) {
788 		PCPU_SET(deadthread, NULL);
789 		thread_stash(td);
790 	}
791 	td = curthread;
792 	mtx_unlock_spin(&sched_lock);
793 
794 	/*
795 	 * cpu_set_fork_handler intercepts this function call to
796 	 * have this call a non-return function to stay in kernel mode.
797 	 * initproc has its own fork handler, but it does return.
798 	 */
799 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
800 	callout(arg, frame);
801 
802 	/*
803 	 * Check if a kernel thread misbehaved and returned from its main
804 	 * function.
805 	 */
806 	PROC_LOCK(p);
807 	if (p->p_flag & P_KTHREAD) {
808 		PROC_UNLOCK(p);
809 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
810 		    p->p_comm, p->p_pid);
811 		kthread_exit(0);
812 	}
813 	PROC_UNLOCK(p);
814 	mtx_assert(&Giant, MA_NOTOWNED);
815 }
816 
817 /*
818  * Simplified back end of syscall(), used when returning from fork()
819  * directly into user mode.  Giant is not held on entry, and must not
820  * be held on return.  This function is passed in to fork_exit() as the
821  * first parameter and is called when returning to a new userland process.
822  */
823 void
824 fork_return(td, frame)
825 	struct thread *td;
826 	struct trapframe *frame;
827 {
828 
829 	userret(td, frame, 0);
830 #ifdef KTRACE
831 	if (KTRPOINT(td, KTR_SYSRET))
832 		ktrsysret(SYS_fork, 0, 0);
833 #endif
834 	mtx_assert(&Giant, MA_NOTOWNED);
835 }
836