xref: /freebsd/sys/kern/kern_fork.c (revision e4e9813eb92cd7c4d4b819a8fbed5cbd3d92f5d8)
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 <security/audit/audit.h>
70 
71 #include <vm/vm.h>
72 #include <vm/pmap.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_extern.h>
75 #include <vm/uma.h>
76 
77 
78 #ifndef _SYS_SYSPROTO_H_
79 struct fork_args {
80 	int     dummy;
81 };
82 #endif
83 
84 static int forksleep; /* Place for fork1() to sleep on. */
85 
86 /*
87  * MPSAFE
88  */
89 /* ARGSUSED */
90 int
91 fork(td, uap)
92 	struct thread *td;
93 	struct fork_args *uap;
94 {
95 	int error;
96 	struct proc *p2;
97 
98 	error = fork1(td, RFFDG | RFPROC, 0, &p2);
99 	if (error == 0) {
100 		td->td_retval[0] = p2->p_pid;
101 		td->td_retval[1] = 0;
102 	}
103 	return (error);
104 }
105 
106 /*
107  * MPSAFE
108  */
109 /* ARGSUSED */
110 int
111 vfork(td, uap)
112 	struct thread *td;
113 	struct vfork_args *uap;
114 {
115 	int error;
116 	struct proc *p2;
117 
118 	error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
119 	if (error == 0) {
120 		td->td_retval[0] = p2->p_pid;
121 		td->td_retval[1] = 0;
122 	}
123 	return (error);
124 }
125 
126 /*
127  * MPSAFE
128  */
129 int
130 rfork(td, uap)
131 	struct thread *td;
132 	struct rfork_args *uap;
133 {
134 	struct proc *p2;
135 	int error;
136 
137 	/* Don't allow kernel-only flags. */
138 	if ((uap->flags & RFKERNELONLY) != 0)
139 		return (EINVAL);
140 
141 	AUDIT_ARG(fflags, uap->flags);
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 int	nprocs = 1;		/* process 0 */
151 int	lastpid = 0;
152 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
153     "Last used PID");
154 
155 /*
156  * Random component to lastpid generation.  We mix in a random factor to make
157  * it a little harder to predict.  We sanity check the modulus value to avoid
158  * doing it in critical paths.  Don't let it be too small or we pointlessly
159  * waste randomness entropy, and don't let it be impossibly large.  Using a
160  * modulus that is too big causes a LOT more process table scans and slows
161  * down fork processing as the pidchecked caching is defeated.
162  */
163 static int randompid = 0;
164 
165 static int
166 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
167 {
168 	int error, pid;
169 
170 	error = sysctl_wire_old_buffer(req, sizeof(int));
171 	if (error != 0)
172 		return(error);
173 	sx_xlock(&allproc_lock);
174 	pid = randompid;
175 	error = sysctl_handle_int(oidp, &pid, 0, req);
176 	if (error == 0 && req->newptr != NULL) {
177 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
178 			pid = PID_MAX - 100;
179 		else if (pid < 2)			/* NOP */
180 			pid = 0;
181 		else if (pid < 100)			/* Make it reasonable */
182 			pid = 100;
183 		randompid = pid;
184 	}
185 	sx_xunlock(&allproc_lock);
186 	return (error);
187 }
188 
189 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
190     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
191 
192 int
193 fork1(td, flags, pages, procp)
194 	struct thread *td;
195 	int flags;
196 	int pages;
197 	struct proc **procp;
198 {
199 	struct proc *p1, *p2, *pptr;
200 	struct proc *newproc;
201 	int ok, trypid;
202 	static int curfail, pidchecked = 0;
203 	static struct timeval lastfail;
204 	struct filedesc *fd;
205 	struct filedesc_to_leader *fdtol;
206 	struct thread *td2;
207 	struct ksegrp *kg2;
208 	struct sigacts *newsigacts;
209 	int error;
210 
211 	/* Can't copy and clear. */
212 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
213 		return (EINVAL);
214 
215 	p1 = td->td_proc;
216 
217 	/*
218 	 * Here we don't create a new process, but we divorce
219 	 * certain parts of a process from itself.
220 	 */
221 	if ((flags & RFPROC) == 0) {
222 		if ((p1->p_flag & P_HADTHREADS) &&
223 		    (flags & (RFCFDG | RFFDG))) {
224 			PROC_LOCK(p1);
225 			if (thread_single(SINGLE_BOUNDARY)) {
226 				PROC_UNLOCK(p1);
227 				return (ERESTART);
228 			}
229 			PROC_UNLOCK(p1);
230 		}
231 
232 		vm_forkproc(td, NULL, NULL, flags);
233 
234 		/*
235 		 * Close all file descriptors.
236 		 */
237 		if (flags & RFCFDG) {
238 			struct filedesc *fdtmp;
239 			fdtmp = fdinit(td->td_proc->p_fd);
240 			fdfree(td);
241 			p1->p_fd = fdtmp;
242 		}
243 
244 		/*
245 		 * Unshare file descriptors (from parent).
246 		 */
247 		if (flags & RFFDG)
248 			fdunshare(p1, td);
249 
250 		if ((p1->p_flag & P_HADTHREADS) &&
251 		    (flags & (RFCFDG | RFFDG))) {
252 			PROC_LOCK(p1);
253 			thread_single_end();
254 			PROC_UNLOCK(p1);
255 		}
256 		*procp = NULL;
257 		return (0);
258 	}
259 
260 	/*
261 	 * Note 1:1 allows for forking with one thread coming out on the
262 	 * other side with the expectation that the process is about to
263 	 * exec.
264 	 */
265 	if (p1->p_flag & P_HADTHREADS) {
266 		/*
267 		 * Idle the other threads for a second.
268 		 * Since the user space is copied, it must remain stable.
269 		 * In addition, all threads (from the user perspective)
270 		 * need to either be suspended or in the kernel,
271 		 * where they will try restart in the parent and will
272 		 * be aborted in the child.
273 		 */
274 		PROC_LOCK(p1);
275 		if (thread_single(SINGLE_NO_EXIT)) {
276 			/* Abort. Someone else is single threading before us. */
277 			PROC_UNLOCK(p1);
278 			return (ERESTART);
279 		}
280 		PROC_UNLOCK(p1);
281 		/*
282 		 * All other activity in this process
283 		 * is now suspended at the user boundary,
284 		 * (or other safe places if we think of any).
285 		 */
286 	}
287 
288 	/* Allocate new proc. */
289 	newproc = uma_zalloc(proc_zone, M_WAITOK);
290 #ifdef MAC
291 	mac_init_proc(newproc);
292 #endif
293 #ifdef AUDIT
294 	audit_proc_alloc(newproc);
295 #endif
296 	knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
297 	STAILQ_INIT(&newproc->p_ktr);
298 
299 	/* We have to lock the process tree while we look for a pid. */
300 	sx_slock(&proctree_lock);
301 
302 	/*
303 	 * Although process entries are dynamically created, we still keep
304 	 * a global limit on the maximum number we will create.  Don't allow
305 	 * a nonprivileged user to use the last ten processes; don't let root
306 	 * exceed the limit. The variable nprocs is the current number of
307 	 * processes, maxproc is the limit.
308 	 */
309 	sx_xlock(&allproc_lock);
310 	if ((nprocs >= maxproc - 10 &&
311 	    suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
312 	    nprocs >= maxproc) {
313 		error = EAGAIN;
314 		goto fail;
315 	}
316 
317 	/*
318 	 * Increment the count of procs running with this uid. Don't allow
319 	 * a nonprivileged user to exceed their current limit.
320 	 */
321 	error = suser_cred(td->td_ucred, SUSER_RUID | SUSER_ALLOWJAIL);
322 	if (error == 0)
323 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
324 	else {
325 		PROC_LOCK(p1);
326 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
327 		    lim_cur(p1, RLIMIT_NPROC));
328 		PROC_UNLOCK(p1);
329 	}
330 	if (!ok) {
331 		error = EAGAIN;
332 		goto fail;
333 	}
334 
335 	/*
336 	 * Increment the nprocs resource before blocking can occur.  There
337 	 * are hard-limits as to the number of processes that can run.
338 	 */
339 	nprocs++;
340 
341 	/*
342 	 * Find an unused process ID.  We remember a range of unused IDs
343 	 * ready to use (from lastpid+1 through pidchecked-1).
344 	 *
345 	 * If RFHIGHPID is set (used during system boot), do not allocate
346 	 * low-numbered pids.
347 	 */
348 	trypid = lastpid + 1;
349 	if (flags & RFHIGHPID) {
350 		if (trypid < 10)
351 			trypid = 10;
352 	} else {
353 		if (randompid)
354 			trypid += arc4random() % randompid;
355 	}
356 retry:
357 	/*
358 	 * If the process ID prototype has wrapped around,
359 	 * restart somewhat above 0, as the low-numbered procs
360 	 * tend to include daemons that don't exit.
361 	 */
362 	if (trypid >= PID_MAX) {
363 		trypid = trypid % PID_MAX;
364 		if (trypid < 100)
365 			trypid += 100;
366 		pidchecked = 0;
367 	}
368 	if (trypid >= pidchecked) {
369 		int doingzomb = 0;
370 
371 		pidchecked = PID_MAX;
372 		/*
373 		 * Scan the active and zombie procs to check whether this pid
374 		 * is in use.  Remember the lowest pid that's greater
375 		 * than trypid, so we can avoid checking for a while.
376 		 */
377 		p2 = LIST_FIRST(&allproc);
378 again:
379 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
380 			while (p2->p_pid == trypid ||
381 			    (p2->p_pgrp != NULL &&
382 			    (p2->p_pgrp->pg_id == trypid ||
383 			    (p2->p_session != NULL &&
384 			    p2->p_session->s_sid == trypid)))) {
385 				trypid++;
386 				if (trypid >= pidchecked)
387 					goto retry;
388 			}
389 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
390 				pidchecked = p2->p_pid;
391 			if (p2->p_pgrp != NULL) {
392 				if (p2->p_pgrp->pg_id > trypid &&
393 				    pidchecked > p2->p_pgrp->pg_id)
394 					pidchecked = p2->p_pgrp->pg_id;
395 				if (p2->p_session != NULL &&
396 				    p2->p_session->s_sid > trypid &&
397 				    pidchecked > p2->p_session->s_sid)
398 					pidchecked = p2->p_session->s_sid;
399 			}
400 		}
401 		if (!doingzomb) {
402 			doingzomb = 1;
403 			p2 = LIST_FIRST(&zombproc);
404 			goto again;
405 		}
406 	}
407 	sx_sunlock(&proctree_lock);
408 
409 	/*
410 	 * RFHIGHPID does not mess with the lastpid counter during boot.
411 	 */
412 	if (flags & RFHIGHPID)
413 		pidchecked = 0;
414 	else
415 		lastpid = trypid;
416 
417 	p2 = newproc;
418 	p2->p_state = PRS_NEW;		/* protect against others */
419 	p2->p_pid = trypid;
420 	AUDIT_ARG(pid, p2->p_pid);
421 	LIST_INSERT_HEAD(&allproc, p2, p_list);
422 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
423 	sx_xunlock(&allproc_lock);
424 
425 	/*
426 	 * Malloc things while we don't hold any locks.
427 	 */
428 	if (flags & RFSIGSHARE)
429 		newsigacts = NULL;
430 	else
431 		newsigacts = sigacts_alloc();
432 
433 	/*
434 	 * Copy filedesc.
435 	 */
436 	if (flags & RFCFDG) {
437 		fd = fdinit(p1->p_fd);
438 		fdtol = NULL;
439 	} else if (flags & RFFDG) {
440 		fd = fdcopy(p1->p_fd);
441 		fdtol = NULL;
442 	} else {
443 		fd = fdshare(p1->p_fd);
444 		if (p1->p_fdtol == NULL)
445 			p1->p_fdtol =
446 				filedesc_to_leader_alloc(NULL,
447 							 NULL,
448 							 p1->p_leader);
449 		if ((flags & RFTHREAD) != 0) {
450 			/*
451 			 * Shared file descriptor table and
452 			 * shared process leaders.
453 			 */
454 			fdtol = p1->p_fdtol;
455 			FILEDESC_LOCK_FAST(p1->p_fd);
456 			fdtol->fdl_refcount++;
457 			FILEDESC_UNLOCK_FAST(p1->p_fd);
458 		} else {
459 			/*
460 			 * Shared file descriptor table, and
461 			 * different process leaders
462 			 */
463 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
464 							 p1->p_fd,
465 							 p2);
466 		}
467 	}
468 	/*
469 	 * Make a proc table entry for the new process.
470 	 * Start by zeroing the section of proc that is zero-initialized,
471 	 * then copy the section that is copied directly from the parent.
472 	 */
473 	td2 = FIRST_THREAD_IN_PROC(p2);
474 	kg2 = FIRST_KSEGRP_IN_PROC(p2);
475 
476 	/* Allocate and switch to an alternate kstack if specified. */
477 	if (pages != 0)
478 		vm_thread_new_altkstack(td2, pages);
479 
480 	PROC_LOCK(p2);
481 	PROC_LOCK(p1);
482 
483 	bzero(&p2->p_startzero,
484 	    __rangeof(struct proc, p_startzero, p_endzero));
485 	bzero(&td2->td_startzero,
486 	    __rangeof(struct thread, td_startzero, td_endzero));
487 	bzero(&kg2->kg_startzero,
488 	    __rangeof(struct ksegrp, kg_startzero, kg_endzero));
489 
490 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
491 	    __rangeof(struct proc, p_startcopy, p_endcopy));
492 	bcopy(&td->td_startcopy, &td2->td_startcopy,
493 	    __rangeof(struct thread, td_startcopy, td_endcopy));
494 	bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
495 	    __rangeof(struct ksegrp, kg_startcopy, kg_endcopy));
496 
497 	td2->td_sigstk = td->td_sigstk;
498 	td2->td_sigmask = td->td_sigmask;
499 
500 	/*
501 	 * Duplicate sub-structures as needed.
502 	 * Increase reference counts on shared objects.
503 	 */
504 	p2->p_flag = 0;
505 	if (p1->p_flag & P_PROFIL)
506 		startprofclock(p2);
507 	mtx_lock_spin(&sched_lock);
508 	p2->p_sflag = PS_INMEM;
509 	/*
510 	 * Allow the scheduler to adjust the priority of the child and
511 	 * parent while we hold the sched_lock.
512 	 */
513 	sched_fork(td, td2);
514 
515 	mtx_unlock_spin(&sched_lock);
516 	p2->p_ucred = crhold(td->td_ucred);
517 	td2->td_ucred = crhold(p2->p_ucred);	/* XXXKSE */
518 #ifdef AUDIT
519 	audit_proc_fork(p1, p2);
520 #endif
521 	pargs_hold(p2->p_args);
522 
523 	if (flags & RFSIGSHARE) {
524 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
525 	} else {
526 		sigacts_copy(newsigacts, p1->p_sigacts);
527 		p2->p_sigacts = newsigacts;
528 	}
529 	if (flags & RFLINUXTHPN)
530 	        p2->p_sigparent = SIGUSR1;
531 	else
532 	        p2->p_sigparent = SIGCHLD;
533 
534 	p2->p_textvp = p1->p_textvp;
535 	p2->p_fd = fd;
536 	p2->p_fdtol = fdtol;
537 
538 	/*
539 	 * p_limit is copy-on-write.  Bump its refcount.
540 	 */
541 	p2->p_limit = lim_hold(p1->p_limit);
542 
543 	pstats_fork(p1->p_stats, p2->p_stats);
544 
545 	PROC_UNLOCK(p1);
546 	PROC_UNLOCK(p2);
547 
548 	/* Bump references to the text vnode (for procfs) */
549 	if (p2->p_textvp)
550 		vref(p2->p_textvp);
551 
552 	/*
553 	 * Set up linkage for kernel based threading.
554 	 */
555 	if ((flags & RFTHREAD) != 0) {
556 		mtx_lock(&ppeers_lock);
557 		p2->p_peers = p1->p_peers;
558 		p1->p_peers = p2;
559 		p2->p_leader = p1->p_leader;
560 		mtx_unlock(&ppeers_lock);
561 		PROC_LOCK(p1->p_leader);
562 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
563 			PROC_UNLOCK(p1->p_leader);
564 			/*
565 			 * The task leader is exiting, so process p1 is
566 			 * going to be killed shortly.  Since p1 obviously
567 			 * isn't dead yet, we know that the leader is either
568 			 * sending SIGKILL's to all the processes in this
569 			 * task or is sleeping waiting for all the peers to
570 			 * exit.  We let p1 complete the fork, but we need
571 			 * to go ahead and kill the new process p2 since
572 			 * the task leader may not get a chance to send
573 			 * SIGKILL to it.  We leave it on the list so that
574 			 * the task leader will wait for this new process
575 			 * to commit suicide.
576 			 */
577 			PROC_LOCK(p2);
578 			psignal(p2, SIGKILL);
579 			PROC_UNLOCK(p2);
580 		} else
581 			PROC_UNLOCK(p1->p_leader);
582 	} else {
583 		p2->p_peers = NULL;
584 		p2->p_leader = p2;
585 	}
586 
587 	sx_xlock(&proctree_lock);
588 	PGRP_LOCK(p1->p_pgrp);
589 	PROC_LOCK(p2);
590 	PROC_LOCK(p1);
591 
592 	/*
593 	 * Preserve some more flags in subprocess.  P_PROFIL has already
594 	 * been preserved.
595 	 */
596 	p2->p_flag |= p1->p_flag & P_SUGID;
597 	td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
598 	SESS_LOCK(p1->p_session);
599 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
600 		p2->p_flag |= P_CONTROLT;
601 	SESS_UNLOCK(p1->p_session);
602 	if (flags & RFPPWAIT)
603 		p2->p_flag |= P_PPWAIT;
604 
605 	p2->p_pgrp = p1->p_pgrp;
606 	LIST_INSERT_AFTER(p1, p2, p_pglist);
607 	PGRP_UNLOCK(p1->p_pgrp);
608 	LIST_INIT(&p2->p_children);
609 
610 	callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
611 
612 #ifdef KTRACE
613 	/*
614 	 * Copy traceflag and tracefile if enabled.
615 	 */
616 	mtx_lock(&ktrace_mtx);
617 	KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
618 	if (p1->p_traceflag & KTRFAC_INHERIT) {
619 		p2->p_traceflag = p1->p_traceflag;
620 		if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
621 			VREF(p2->p_tracevp);
622 			KASSERT(p1->p_tracecred != NULL,
623 			    ("ktrace vnode with no cred"));
624 			p2->p_tracecred = crhold(p1->p_tracecred);
625 		}
626 	}
627 	mtx_unlock(&ktrace_mtx);
628 #endif
629 
630 	/*
631 	 * If PF_FORK is set, the child process inherits the
632 	 * procfs ioctl flags from its parent.
633 	 */
634 	if (p1->p_pfsflags & PF_FORK) {
635 		p2->p_stops = p1->p_stops;
636 		p2->p_pfsflags = p1->p_pfsflags;
637 	}
638 
639 	/*
640 	 * This begins the section where we must prevent the parent
641 	 * from being swapped.
642 	 */
643 	_PHOLD(p1);
644 	PROC_UNLOCK(p1);
645 
646 	/*
647 	 * Attach the new process to its parent.
648 	 *
649 	 * If RFNOWAIT is set, the newly created process becomes a child
650 	 * of init.  This effectively disassociates the child from the
651 	 * parent.
652 	 */
653 	if (flags & RFNOWAIT)
654 		pptr = initproc;
655 	else
656 		pptr = p1;
657 	p2->p_pptr = pptr;
658 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
659 	sx_xunlock(&proctree_lock);
660 
661 	/* Inform accounting that we have forked. */
662 	p2->p_acflag = AFORK;
663 	PROC_UNLOCK(p2);
664 
665 	/*
666 	 * Finish creating the child process.  It will return via a different
667 	 * execution path later.  (ie: directly into user mode)
668 	 */
669 	vm_forkproc(td, p2, td2, flags);
670 
671 	if (flags == (RFFDG | RFPROC)) {
672 		atomic_add_int(&cnt.v_forks, 1);
673 		atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize +
674 		    p2->p_vmspace->vm_ssize);
675 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
676 		atomic_add_int(&cnt.v_vforks, 1);
677 		atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
678 		    p2->p_vmspace->vm_ssize);
679 	} else if (p1 == &proc0) {
680 		atomic_add_int(&cnt.v_kthreads, 1);
681 		atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
682 		    p2->p_vmspace->vm_ssize);
683 	} else {
684 		atomic_add_int(&cnt.v_rforks, 1);
685 		atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
686 		    p2->p_vmspace->vm_ssize);
687 	}
688 
689 	/*
690 	 * Both processes are set up, now check if any loadable modules want
691 	 * to adjust anything.
692 	 *   What if they have an error? XXX
693 	 */
694 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
695 
696 	/*
697 	 * Set the child start time and mark the process as being complete.
698 	 */
699 	microuptime(&p2->p_stats->p_start);
700 	mtx_lock_spin(&sched_lock);
701 	p2->p_state = PRS_NORMAL;
702 
703 	/*
704 	 * If RFSTOPPED not requested, make child runnable and add to
705 	 * run queue.
706 	 */
707 	if ((flags & RFSTOPPED) == 0) {
708 		TD_SET_CAN_RUN(td2);
709 		setrunqueue(td2, SRQ_BORING);
710 	}
711 	mtx_unlock_spin(&sched_lock);
712 
713 	/*
714 	 * Now can be swapped.
715 	 */
716 	PROC_LOCK(p1);
717 	_PRELE(p1);
718 
719 	/*
720 	 * Tell any interested parties about the new process.
721 	 */
722 	KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
723 
724 	PROC_UNLOCK(p1);
725 
726 	/*
727 	 * Preserve synchronization semantics of vfork.  If waiting for
728 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
729 	 * proc (in case of exit).
730 	 */
731 	PROC_LOCK(p2);
732 	while (p2->p_flag & P_PPWAIT)
733 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
734 	PROC_UNLOCK(p2);
735 
736 	/*
737 	 * If other threads are waiting, let them continue now.
738 	 */
739 	if (p1->p_flag & P_HADTHREADS) {
740 		PROC_LOCK(p1);
741 		thread_single_end();
742 		PROC_UNLOCK(p1);
743 	}
744 
745 	/*
746 	 * Return child proc pointer to parent.
747 	 */
748 	*procp = p2;
749 	return (0);
750 fail:
751 	sx_sunlock(&proctree_lock);
752 	if (ppsratecheck(&lastfail, &curfail, 1))
753 		printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
754 		    td->td_ucred->cr_ruid);
755 	sx_xunlock(&allproc_lock);
756 #ifdef MAC
757 	mac_destroy_proc(newproc);
758 #endif
759 #ifdef AUDIT
760 	audit_proc_free(newproc);
761 #endif
762 	uma_zfree(proc_zone, newproc);
763 	if (p1->p_flag & P_HADTHREADS) {
764 		PROC_LOCK(p1);
765 		thread_single_end();
766 		PROC_UNLOCK(p1);
767 	}
768 	tsleep(&forksleep, PUSER, "fork", hz / 2);
769 	return (error);
770 }
771 
772 /*
773  * Handle the return of a child process from fork1().  This function
774  * is called from the MD fork_trampoline() entry point.
775  */
776 void
777 fork_exit(callout, arg, frame)
778 	void (*callout)(void *, struct trapframe *);
779 	void *arg;
780 	struct trapframe *frame;
781 {
782 	struct proc *p;
783 	struct thread *td;
784 
785 	/*
786 	 * Finish setting up thread glue so that it begins execution in a
787 	 * non-nested critical section with sched_lock held but not recursed.
788 	 */
789 	td = curthread;
790 	p = td->td_proc;
791 	td->td_oncpu = PCPU_GET(cpuid);
792 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
793 
794 	sched_lock.mtx_lock = (uintptr_t)td;
795 	mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
796 	CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
797 		td, td->td_sched, p->p_pid, p->p_comm);
798 
799 	/*
800 	 * Processes normally resume in mi_switch() after being
801 	 * cpu_switch()'ed to, but when children start up they arrive here
802 	 * instead, so we must do much the same things as mi_switch() would.
803 	 */
804 
805 	if ((td = PCPU_GET(deadthread))) {
806 		PCPU_SET(deadthread, NULL);
807 		thread_stash(td);
808 	}
809 	td = curthread;
810 	mtx_unlock_spin(&sched_lock);
811 
812 	/*
813 	 * cpu_set_fork_handler intercepts this function call to
814 	 * have this call a non-return function to stay in kernel mode.
815 	 * initproc has its own fork handler, but it does return.
816 	 */
817 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
818 	callout(arg, frame);
819 
820 	/*
821 	 * Check if a kernel thread misbehaved and returned from its main
822 	 * function.
823 	 */
824 	if (p->p_flag & P_KTHREAD) {
825 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
826 		    p->p_comm, p->p_pid);
827 		kthread_exit(0);
828 	}
829 	mtx_assert(&Giant, MA_NOTOWNED);
830 }
831 
832 /*
833  * Simplified back end of syscall(), used when returning from fork()
834  * directly into user mode.  Giant is not held on entry, and must not
835  * be held on return.  This function is passed in to fork_exit() as the
836  * first parameter and is called when returning to a new userland process.
837  */
838 void
839 fork_return(td, frame)
840 	struct thread *td;
841 	struct trapframe *frame;
842 {
843 
844 	userret(td, frame);
845 #ifdef KTRACE
846 	if (KTRPOINT(td, KTR_SYSRET))
847 		ktrsysret(SYS_fork, 0, 0);
848 #endif
849 	mtx_assert(&Giant, MA_NOTOWNED);
850 }
851