xref: /freebsd/sys/kern/kern_fork.c (revision db612abe8df3355d1eb23bb3b50fdd97bc21e979)
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/priv.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/vmmeter.h>
61 #include <sys/vnode.h>
62 #include <sys/acct.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 #include <security/mac/mac_framework.h>
71 
72 #include <vm/vm.h>
73 #include <vm/pmap.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_extern.h>
76 #include <vm/uma.h>
77 
78 
79 #ifndef _SYS_SYSPROTO_H_
80 struct fork_args {
81 	int     dummy;
82 };
83 #endif
84 
85 /* ARGSUSED */
86 int
87 fork(td, uap)
88 	struct thread *td;
89 	struct fork_args *uap;
90 {
91 	int error;
92 	struct proc *p2;
93 
94 	error = fork1(td, RFFDG | RFPROC, 0, &p2);
95 	if (error == 0) {
96 		td->td_retval[0] = p2->p_pid;
97 		td->td_retval[1] = 0;
98 	}
99 	return (error);
100 }
101 
102 /* ARGSUSED */
103 int
104 vfork(td, uap)
105 	struct thread *td;
106 	struct vfork_args *uap;
107 {
108 	int error;
109 	struct proc *p2;
110 
111 	error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
112 	if (error == 0) {
113 		td->td_retval[0] = p2->p_pid;
114 		td->td_retval[1] = 0;
115 	}
116 	return (error);
117 }
118 
119 int
120 rfork(td, uap)
121 	struct thread *td;
122 	struct rfork_args *uap;
123 {
124 	struct proc *p2;
125 	int error;
126 
127 	/* Don't allow kernel-only flags. */
128 	if ((uap->flags & RFKERNELONLY) != 0)
129 		return (EINVAL);
130 
131 	AUDIT_ARG(fflags, uap->flags);
132 	error = fork1(td, uap->flags, 0, &p2);
133 	if (error == 0) {
134 		td->td_retval[0] = p2 ? p2->p_pid : 0;
135 		td->td_retval[1] = 0;
136 	}
137 	return (error);
138 }
139 
140 int	nprocs = 1;		/* process 0 */
141 int	lastpid = 0;
142 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
143     "Last used PID");
144 
145 /*
146  * Random component to lastpid generation.  We mix in a random factor to make
147  * it a little harder to predict.  We sanity check the modulus value to avoid
148  * doing it in critical paths.  Don't let it be too small or we pointlessly
149  * waste randomness entropy, and don't let it be impossibly large.  Using a
150  * modulus that is too big causes a LOT more process table scans and slows
151  * down fork processing as the pidchecked caching is defeated.
152  */
153 static int randompid = 0;
154 
155 static int
156 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
157 {
158 	int error, pid;
159 
160 	error = sysctl_wire_old_buffer(req, sizeof(int));
161 	if (error != 0)
162 		return(error);
163 	sx_xlock(&allproc_lock);
164 	pid = randompid;
165 	error = sysctl_handle_int(oidp, &pid, 0, req);
166 	if (error == 0 && req->newptr != NULL) {
167 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
168 			pid = PID_MAX - 100;
169 		else if (pid < 2)			/* NOP */
170 			pid = 0;
171 		else if (pid < 100)			/* Make it reasonable */
172 			pid = 100;
173 		randompid = pid;
174 	}
175 	sx_xunlock(&allproc_lock);
176 	return (error);
177 }
178 
179 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
180     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
181 
182 int
183 fork1(td, flags, pages, procp)
184 	struct thread *td;
185 	int flags;
186 	int pages;
187 	struct proc **procp;
188 {
189 	struct proc *p1, *p2, *pptr;
190 	struct proc *newproc;
191 	int ok, trypid;
192 	static int curfail, pidchecked = 0;
193 	static struct timeval lastfail;
194 	struct filedesc *fd;
195 	struct filedesc_to_leader *fdtol;
196 	struct thread *td2;
197 	struct sigacts *newsigacts;
198 	struct vmspace *vm2;
199 	int error;
200 
201 	/* Can't copy and clear. */
202 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
203 		return (EINVAL);
204 
205 	p1 = td->td_proc;
206 
207 	/*
208 	 * Here we don't create a new process, but we divorce
209 	 * certain parts of a process from itself.
210 	 */
211 	if ((flags & RFPROC) == 0) {
212 		if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
213 		    (flags & (RFCFDG | RFFDG))) {
214 			PROC_LOCK(p1);
215 			if (thread_single(SINGLE_BOUNDARY)) {
216 				PROC_UNLOCK(p1);
217 				return (ERESTART);
218 			}
219 			PROC_UNLOCK(p1);
220 		}
221 
222 		error = vm_forkproc(td, NULL, NULL, NULL, flags);
223 		if (error)
224 			goto norfproc_fail;
225 
226 		/*
227 		 * Close all file descriptors.
228 		 */
229 		if (flags & RFCFDG) {
230 			struct filedesc *fdtmp;
231 			fdtmp = fdinit(td->td_proc->p_fd);
232 			fdfree(td);
233 			p1->p_fd = fdtmp;
234 		}
235 
236 		/*
237 		 * Unshare file descriptors (from parent).
238 		 */
239 		if (flags & RFFDG)
240 			fdunshare(p1, td);
241 
242 norfproc_fail:
243 		if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
244 		    (flags & (RFCFDG | RFFDG))) {
245 			PROC_LOCK(p1);
246 			thread_single_end();
247 			PROC_UNLOCK(p1);
248 		}
249 		*procp = NULL;
250 		return (error);
251 	}
252 
253 	/*
254 	 * XXX
255 	 * We did have single-threading code here
256 	 * however it proved un-needed and caused problems
257 	 */
258 
259 	vm2 = NULL;
260 	/* Allocate new proc. */
261 	newproc = uma_zalloc(proc_zone, M_WAITOK);
262 	if (TAILQ_EMPTY(&newproc->p_threads)) {
263 		td2 = thread_alloc();
264 		if (td2 == NULL) {
265 			error = ENOMEM;
266 			goto fail1;
267 		}
268 		proc_linkup(newproc, td2);
269 	} else
270 		td2 = FIRST_THREAD_IN_PROC(newproc);
271 
272 	/* Allocate and switch to an alternate kstack if specified. */
273 	if (pages != 0) {
274 		if (!vm_thread_new_altkstack(td2, pages)) {
275 			error = ENOMEM;
276 			goto fail1;
277 		}
278 	}
279 	if ((flags & RFMEM) == 0) {
280 		vm2 = vmspace_fork(p1->p_vmspace);
281 		if (vm2 == NULL) {
282 			error = ENOMEM;
283 			goto fail1;
284 		}
285 	}
286 #ifdef MAC
287 	mac_proc_init(newproc);
288 #endif
289 	knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
290 	STAILQ_INIT(&newproc->p_ktr);
291 
292 	/* We have to lock the process tree while we look for a pid. */
293 	sx_slock(&proctree_lock);
294 
295 	/*
296 	 * Although process entries are dynamically created, we still keep
297 	 * a global limit on the maximum number we will create.  Don't allow
298 	 * a nonprivileged user to use the last ten processes; don't let root
299 	 * exceed the limit. The variable nprocs is the current number of
300 	 * processes, maxproc is the limit.
301 	 */
302 	sx_xlock(&allproc_lock);
303 	if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
304 	    PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
305 		error = EAGAIN;
306 		goto fail;
307 	}
308 
309 	/*
310 	 * Increment the count of procs running with this uid. Don't allow
311 	 * a nonprivileged user to exceed their current limit.
312 	 *
313 	 * XXXRW: Can we avoid privilege here if it's not needed?
314 	 */
315 	error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
316 	if (error == 0)
317 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
318 	else {
319 		PROC_LOCK(p1);
320 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
321 		    lim_cur(p1, RLIMIT_NPROC));
322 		PROC_UNLOCK(p1);
323 	}
324 	if (!ok) {
325 		error = EAGAIN;
326 		goto fail;
327 	}
328 
329 	/*
330 	 * Increment the nprocs resource before blocking can occur.  There
331 	 * are hard-limits as to the number of processes that can run.
332 	 */
333 	nprocs++;
334 
335 	/*
336 	 * Find an unused process ID.  We remember a range of unused IDs
337 	 * ready to use (from lastpid+1 through pidchecked-1).
338 	 *
339 	 * If RFHIGHPID is set (used during system boot), do not allocate
340 	 * low-numbered pids.
341 	 */
342 	trypid = lastpid + 1;
343 	if (flags & RFHIGHPID) {
344 		if (trypid < 10)
345 			trypid = 10;
346 	} else {
347 		if (randompid)
348 			trypid += arc4random() % randompid;
349 	}
350 retry:
351 	/*
352 	 * If the process ID prototype has wrapped around,
353 	 * restart somewhat above 0, as the low-numbered procs
354 	 * tend to include daemons that don't exit.
355 	 */
356 	if (trypid >= PID_MAX) {
357 		trypid = trypid % PID_MAX;
358 		if (trypid < 100)
359 			trypid += 100;
360 		pidchecked = 0;
361 	}
362 	if (trypid >= pidchecked) {
363 		int doingzomb = 0;
364 
365 		pidchecked = PID_MAX;
366 		/*
367 		 * Scan the active and zombie procs to check whether this pid
368 		 * is in use.  Remember the lowest pid that's greater
369 		 * than trypid, so we can avoid checking for a while.
370 		 */
371 		p2 = LIST_FIRST(&allproc);
372 again:
373 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
374 			while (p2->p_pid == trypid ||
375 			    (p2->p_pgrp != NULL &&
376 			    (p2->p_pgrp->pg_id == trypid ||
377 			    (p2->p_session != NULL &&
378 			    p2->p_session->s_sid == trypid)))) {
379 				trypid++;
380 				if (trypid >= pidchecked)
381 					goto retry;
382 			}
383 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
384 				pidchecked = p2->p_pid;
385 			if (p2->p_pgrp != NULL) {
386 				if (p2->p_pgrp->pg_id > trypid &&
387 				    pidchecked > p2->p_pgrp->pg_id)
388 					pidchecked = p2->p_pgrp->pg_id;
389 				if (p2->p_session != NULL &&
390 				    p2->p_session->s_sid > trypid &&
391 				    pidchecked > p2->p_session->s_sid)
392 					pidchecked = p2->p_session->s_sid;
393 			}
394 		}
395 		if (!doingzomb) {
396 			doingzomb = 1;
397 			p2 = LIST_FIRST(&zombproc);
398 			goto again;
399 		}
400 	}
401 	sx_sunlock(&proctree_lock);
402 
403 	/*
404 	 * RFHIGHPID does not mess with the lastpid counter during boot.
405 	 */
406 	if (flags & RFHIGHPID)
407 		pidchecked = 0;
408 	else
409 		lastpid = trypid;
410 
411 	p2 = newproc;
412 	p2->p_state = PRS_NEW;		/* protect against others */
413 	p2->p_pid = trypid;
414 	/*
415 	 * Allow the scheduler to initialize the child.
416 	 */
417 	thread_lock(td);
418 	sched_fork(td, td2);
419 	thread_unlock(td);
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 
424 	PROC_LOCK(p2);
425 	PROC_LOCK(p1);
426 
427 	sx_xunlock(&allproc_lock);
428 
429 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
430 	    __rangeof(struct proc, p_startcopy, p_endcopy));
431 	PROC_UNLOCK(p1);
432 
433 	bzero(&p2->p_startzero,
434 	    __rangeof(struct proc, p_startzero, p_endzero));
435 
436 	p2->p_ucred = crhold(td->td_ucred);
437 	PROC_UNLOCK(p2);
438 
439 	/*
440 	 * Malloc things while we don't hold any locks.
441 	 */
442 	if (flags & RFSIGSHARE)
443 		newsigacts = NULL;
444 	else
445 		newsigacts = sigacts_alloc();
446 
447 	/*
448 	 * Copy filedesc.
449 	 */
450 	if (flags & RFCFDG) {
451 		fd = fdinit(p1->p_fd);
452 		fdtol = NULL;
453 	} else if (flags & RFFDG) {
454 		fd = fdcopy(p1->p_fd);
455 		fdtol = NULL;
456 	} else {
457 		fd = fdshare(p1->p_fd);
458 		if (p1->p_fdtol == NULL)
459 			p1->p_fdtol =
460 				filedesc_to_leader_alloc(NULL,
461 							 NULL,
462 							 p1->p_leader);
463 		if ((flags & RFTHREAD) != 0) {
464 			/*
465 			 * Shared file descriptor table and
466 			 * shared process leaders.
467 			 */
468 			fdtol = p1->p_fdtol;
469 			FILEDESC_XLOCK(p1->p_fd);
470 			fdtol->fdl_refcount++;
471 			FILEDESC_XUNLOCK(p1->p_fd);
472 		} else {
473 			/*
474 			 * Shared file descriptor table, and
475 			 * different process leaders
476 			 */
477 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
478 							 p1->p_fd,
479 							 p2);
480 		}
481 	}
482 	/*
483 	 * Make a proc table entry for the new process.
484 	 * Start by zeroing the section of proc that is zero-initialized,
485 	 * then copy the section that is copied directly from the parent.
486 	 */
487 
488 	PROC_LOCK(p2);
489 	PROC_LOCK(p1);
490 
491 	bzero(&td2->td_startzero,
492 	    __rangeof(struct thread, td_startzero, td_endzero));
493 
494 	bcopy(&td->td_startcopy, &td2->td_startcopy,
495 	    __rangeof(struct thread, td_startcopy, td_endcopy));
496 
497 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
498 	td2->td_sigstk = td->td_sigstk;
499 	td2->td_sigmask = td->td_sigmask;
500 	td2->td_flags = TDF_INMEM;
501 
502 	/*
503 	 * Duplicate sub-structures as needed.
504 	 * Increase reference counts on shared objects.
505 	 */
506 	p2->p_flag = P_INMEM;
507 	p2->p_swtick = ticks;
508 	if (p1->p_flag & P_PROFIL)
509 		startprofclock(p2);
510 	td2->td_ucred = crhold(p2->p_ucred);
511 	pargs_hold(p2->p_args);
512 
513 	if (flags & RFSIGSHARE) {
514 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
515 	} else {
516 		sigacts_copy(newsigacts, p1->p_sigacts);
517 		p2->p_sigacts = newsigacts;
518 	}
519 	if (flags & RFLINUXTHPN)
520 	        p2->p_sigparent = SIGUSR1;
521 	else
522 	        p2->p_sigparent = SIGCHLD;
523 
524 	p2->p_textvp = p1->p_textvp;
525 	p2->p_fd = fd;
526 	p2->p_fdtol = fdtol;
527 
528 	/*
529 	 * p_limit is copy-on-write.  Bump its refcount.
530 	 */
531 	lim_fork(p1, p2);
532 
533 	pstats_fork(p1->p_stats, p2->p_stats);
534 
535 	PROC_UNLOCK(p1);
536 	PROC_UNLOCK(p2);
537 
538 	/* Bump references to the text vnode (for procfs) */
539 	if (p2->p_textvp)
540 		vref(p2->p_textvp);
541 
542 	/*
543 	 * Set up linkage for kernel based threading.
544 	 */
545 	if ((flags & RFTHREAD) != 0) {
546 		mtx_lock(&ppeers_lock);
547 		p2->p_peers = p1->p_peers;
548 		p1->p_peers = p2;
549 		p2->p_leader = p1->p_leader;
550 		mtx_unlock(&ppeers_lock);
551 		PROC_LOCK(p1->p_leader);
552 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
553 			PROC_UNLOCK(p1->p_leader);
554 			/*
555 			 * The task leader is exiting, so process p1 is
556 			 * going to be killed shortly.  Since p1 obviously
557 			 * isn't dead yet, we know that the leader is either
558 			 * sending SIGKILL's to all the processes in this
559 			 * task or is sleeping waiting for all the peers to
560 			 * exit.  We let p1 complete the fork, but we need
561 			 * to go ahead and kill the new process p2 since
562 			 * the task leader may not get a chance to send
563 			 * SIGKILL to it.  We leave it on the list so that
564 			 * the task leader will wait for this new process
565 			 * to commit suicide.
566 			 */
567 			PROC_LOCK(p2);
568 			psignal(p2, SIGKILL);
569 			PROC_UNLOCK(p2);
570 		} else
571 			PROC_UNLOCK(p1->p_leader);
572 	} else {
573 		p2->p_peers = NULL;
574 		p2->p_leader = p2;
575 	}
576 
577 	sx_xlock(&proctree_lock);
578 	PGRP_LOCK(p1->p_pgrp);
579 	PROC_LOCK(p2);
580 	PROC_LOCK(p1);
581 
582 	/*
583 	 * Preserve some more flags in subprocess.  P_PROFIL has already
584 	 * been preserved.
585 	 */
586 	p2->p_flag |= p1->p_flag & P_SUGID;
587 	td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
588 	SESS_LOCK(p1->p_session);
589 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
590 		p2->p_flag |= P_CONTROLT;
591 	SESS_UNLOCK(p1->p_session);
592 	if (flags & RFPPWAIT)
593 		p2->p_flag |= P_PPWAIT;
594 
595 	p2->p_pgrp = p1->p_pgrp;
596 	LIST_INSERT_AFTER(p1, p2, p_pglist);
597 	PGRP_UNLOCK(p1->p_pgrp);
598 	LIST_INIT(&p2->p_children);
599 
600 	callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
601 
602 #ifdef KTRACE
603 	/*
604 	 * Copy traceflag and tracefile if enabled.
605 	 */
606 	mtx_lock(&ktrace_mtx);
607 	KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
608 	if (p1->p_traceflag & KTRFAC_INHERIT) {
609 		p2->p_traceflag = p1->p_traceflag;
610 		if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
611 			VREF(p2->p_tracevp);
612 			KASSERT(p1->p_tracecred != NULL,
613 			    ("ktrace vnode with no cred"));
614 			p2->p_tracecred = crhold(p1->p_tracecred);
615 		}
616 	}
617 	mtx_unlock(&ktrace_mtx);
618 #endif
619 
620 	/*
621 	 * If PF_FORK is set, the child process inherits the
622 	 * procfs ioctl flags from its parent.
623 	 */
624 	if (p1->p_pfsflags & PF_FORK) {
625 		p2->p_stops = p1->p_stops;
626 		p2->p_pfsflags = p1->p_pfsflags;
627 	}
628 
629 	/*
630 	 * This begins the section where we must prevent the parent
631 	 * from being swapped.
632 	 */
633 	_PHOLD(p1);
634 	PROC_UNLOCK(p1);
635 
636 	/*
637 	 * Attach the new process to its parent.
638 	 *
639 	 * If RFNOWAIT is set, the newly created process becomes a child
640 	 * of init.  This effectively disassociates the child from the
641 	 * parent.
642 	 */
643 	if (flags & RFNOWAIT)
644 		pptr = initproc;
645 	else
646 		pptr = p1;
647 	p2->p_pptr = pptr;
648 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
649 	sx_xunlock(&proctree_lock);
650 
651 	/* Inform accounting that we have forked. */
652 	p2->p_acflag = AFORK;
653 	PROC_UNLOCK(p2);
654 
655 	/*
656 	 * Finish creating the child process.  It will return via a different
657 	 * execution path later.  (ie: directly into user mode)
658 	 */
659 	vm_forkproc(td, p2, td2, vm2, flags);
660 
661 	if (flags == (RFFDG | RFPROC)) {
662 		PCPU_INC(cnt.v_forks);
663 		PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
664 		    p2->p_vmspace->vm_ssize);
665 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
666 		PCPU_INC(cnt.v_vforks);
667 		PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
668 		    p2->p_vmspace->vm_ssize);
669 	} else if (p1 == &proc0) {
670 		PCPU_INC(cnt.v_kthreads);
671 		PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
672 		    p2->p_vmspace->vm_ssize);
673 	} else {
674 		PCPU_INC(cnt.v_rforks);
675 		PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
676 		    p2->p_vmspace->vm_ssize);
677 	}
678 
679 	/*
680 	 * Both processes are set up, now check if any loadable modules want
681 	 * to adjust anything.
682 	 *   What if they have an error? XXX
683 	 */
684 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
685 
686 	/*
687 	 * Set the child start time and mark the process as being complete.
688 	 */
689 	microuptime(&p2->p_stats->p_start);
690 	PROC_SLOCK(p2);
691 	p2->p_state = PRS_NORMAL;
692 	PROC_SUNLOCK(p2);
693 
694 	/*
695 	 * If RFSTOPPED not requested, make child runnable and add to
696 	 * run queue.
697 	 */
698 	if ((flags & RFSTOPPED) == 0) {
699 		thread_lock(td2);
700 		TD_SET_CAN_RUN(td2);
701 		sched_add(td2, SRQ_BORING);
702 		thread_unlock(td2);
703 	}
704 
705 	/*
706 	 * Now can be swapped.
707 	 */
708 	PROC_LOCK(p1);
709 	_PRELE(p1);
710 
711 	/*
712 	 * Tell any interested parties about the new process.
713 	 */
714 	KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
715 
716 	PROC_UNLOCK(p1);
717 
718 	/*
719 	 * Preserve synchronization semantics of vfork.  If waiting for
720 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
721 	 * proc (in case of exit).
722 	 */
723 	PROC_LOCK(p2);
724 	while (p2->p_flag & P_PPWAIT)
725 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
726 	PROC_UNLOCK(p2);
727 
728 	/*
729 	 * Return child proc pointer to parent.
730 	 */
731 	*procp = p2;
732 	return (0);
733 fail:
734 	sx_sunlock(&proctree_lock);
735 	if (ppsratecheck(&lastfail, &curfail, 1))
736 		printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
737 		    td->td_ucred->cr_ruid);
738 	sx_xunlock(&allproc_lock);
739 #ifdef MAC
740 	mac_proc_destroy(newproc);
741 #endif
742 fail1:
743 	if (vm2 != NULL)
744 		vmspace_free(vm2);
745 	uma_zfree(proc_zone, newproc);
746 	pause("fork", hz / 2);
747 	return (error);
748 }
749 
750 /*
751  * Handle the return of a child process from fork1().  This function
752  * is called from the MD fork_trampoline() entry point.
753  */
754 void
755 fork_exit(callout, arg, frame)
756 	void (*callout)(void *, struct trapframe *);
757 	void *arg;
758 	struct trapframe *frame;
759 {
760 	struct proc *p;
761 	struct thread *td;
762 	struct thread *dtd;
763 
764 	td = curthread;
765 	p = td->td_proc;
766 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
767 
768 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
769 		td, td->td_sched, p->p_pid, td->td_name);
770 
771 	sched_fork_exit(td);
772 	/*
773 	* Processes normally resume in mi_switch() after being
774 	* cpu_switch()'ed to, but when children start up they arrive here
775 	* instead, so we must do much the same things as mi_switch() would.
776 	*/
777 	if ((dtd = PCPU_GET(deadthread))) {
778 		PCPU_SET(deadthread, NULL);
779 		thread_stash(dtd);
780 	}
781 	thread_unlock(td);
782 
783 	/*
784 	 * cpu_set_fork_handler intercepts this function call to
785 	 * have this call a non-return function to stay in kernel mode.
786 	 * initproc has its own fork handler, but it does return.
787 	 */
788 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
789 	callout(arg, frame);
790 
791 	/*
792 	 * Check if a kernel thread misbehaved and returned from its main
793 	 * function.
794 	 */
795 	if (p->p_flag & P_KTHREAD) {
796 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
797 		    td->td_name, p->p_pid);
798 		kproc_exit(0);
799 	}
800 	mtx_assert(&Giant, MA_NOTOWNED);
801 
802 	EVENTHANDLER_INVOKE(schedtail, p);
803 }
804 
805 /*
806  * Simplified back end of syscall(), used when returning from fork()
807  * directly into user mode.  Giant is not held on entry, and must not
808  * be held on return.  This function is passed in to fork_exit() as the
809  * first parameter and is called when returning to a new userland process.
810  */
811 void
812 fork_return(td, frame)
813 	struct thread *td;
814 	struct trapframe *frame;
815 {
816 
817 	userret(td, frame);
818 #ifdef KTRACE
819 	if (KTRPOINT(td, KTR_SYSRET))
820 		ktrsysret(SYS_fork, 0, 0);
821 #endif
822 	mtx_assert(&Giant, MA_NOTOWNED);
823 }
824