xref: /freebsd/sys/kern/kern_fork.c (revision 9517e866259191fcd39434a97ad849a9b59b9b9f)
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 
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/jail.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 	vm_ooffset_t mem_charged;
217 	int error;
218 
219 	/* Can't copy and clear. */
220 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
221 		return (EINVAL);
222 
223 	p1 = td->td_proc;
224 
225 	/*
226 	 * Here we don't create a new process, but we divorce
227 	 * certain parts of a process from itself.
228 	 */
229 	if ((flags & RFPROC) == 0) {
230 		if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
231 		    (flags & (RFCFDG | RFFDG))) {
232 			PROC_LOCK(p1);
233 			if (thread_single(SINGLE_BOUNDARY)) {
234 				PROC_UNLOCK(p1);
235 				return (ERESTART);
236 			}
237 			PROC_UNLOCK(p1);
238 		}
239 
240 		error = vm_forkproc(td, NULL, NULL, NULL, flags);
241 		if (error)
242 			goto norfproc_fail;
243 
244 		/*
245 		 * Close all file descriptors.
246 		 */
247 		if (flags & RFCFDG) {
248 			struct filedesc *fdtmp;
249 			fdtmp = fdinit(td->td_proc->p_fd);
250 			fdfree(td);
251 			p1->p_fd = fdtmp;
252 		}
253 
254 		/*
255 		 * Unshare file descriptors (from parent).
256 		 */
257 		if (flags & RFFDG)
258 			fdunshare(p1, td);
259 
260 norfproc_fail:
261 		if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
262 		    (flags & (RFCFDG | RFFDG))) {
263 			PROC_LOCK(p1);
264 			thread_single_end();
265 			PROC_UNLOCK(p1);
266 		}
267 		*procp = NULL;
268 		return (error);
269 	}
270 
271 	/*
272 	 * XXX
273 	 * We did have single-threading code here
274 	 * however it proved un-needed and caused problems
275 	 */
276 
277 	mem_charged = 0;
278 	vm2 = NULL;
279 	/* Allocate new proc. */
280 	newproc = uma_zalloc(proc_zone, M_WAITOK);
281 	if (TAILQ_EMPTY(&newproc->p_threads)) {
282 		td2 = thread_alloc();
283 		if (td2 == NULL) {
284 			error = ENOMEM;
285 			goto fail1;
286 		}
287 		proc_linkup(newproc, td2);
288 	} else
289 		td2 = FIRST_THREAD_IN_PROC(newproc);
290 
291 	/* Allocate and switch to an alternate kstack if specified. */
292 	if (pages != 0) {
293 		if (!vm_thread_new_altkstack(td2, pages)) {
294 			error = ENOMEM;
295 			goto fail1;
296 		}
297 	}
298 	if ((flags & RFMEM) == 0) {
299 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
300 		if (vm2 == NULL) {
301 			error = ENOMEM;
302 			goto fail1;
303 		}
304 		if (!swap_reserve(mem_charged)) {
305 			/*
306 			 * The swap reservation failed. The accounting
307 			 * from the entries of the copied vm2 will be
308 			 * substracted in vmspace_free(), so force the
309 			 * reservation there.
310 			 */
311 			swap_reserve_force(mem_charged);
312 			error = ENOMEM;
313 			goto fail1;
314 		}
315 	} else
316 		vm2 = NULL;
317 #ifdef MAC
318 	mac_proc_init(newproc);
319 #endif
320 	knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
321 	STAILQ_INIT(&newproc->p_ktr);
322 
323 	/* We have to lock the process tree while we look for a pid. */
324 	sx_slock(&proctree_lock);
325 
326 	/*
327 	 * Although process entries are dynamically created, we still keep
328 	 * a global limit on the maximum number we will create.  Don't allow
329 	 * a nonprivileged user to use the last ten processes; don't let root
330 	 * exceed the limit. The variable nprocs is the current number of
331 	 * processes, maxproc is the limit.
332 	 */
333 	sx_xlock(&allproc_lock);
334 	if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
335 	    PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
336 		error = EAGAIN;
337 		goto fail;
338 	}
339 
340 	/*
341 	 * Increment the count of procs running with this uid. Don't allow
342 	 * a nonprivileged user to exceed their current limit.
343 	 *
344 	 * XXXRW: Can we avoid privilege here if it's not needed?
345 	 */
346 	error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
347 	if (error == 0)
348 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
349 	else {
350 		PROC_LOCK(p1);
351 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
352 		    lim_cur(p1, RLIMIT_NPROC));
353 		PROC_UNLOCK(p1);
354 	}
355 	if (!ok) {
356 		error = EAGAIN;
357 		goto fail;
358 	}
359 
360 	/*
361 	 * Increment the nprocs resource before blocking can occur.  There
362 	 * are hard-limits as to the number of processes that can run.
363 	 */
364 	nprocs++;
365 
366 	/*
367 	 * Find an unused process ID.  We remember a range of unused IDs
368 	 * ready to use (from lastpid+1 through pidchecked-1).
369 	 *
370 	 * If RFHIGHPID is set (used during system boot), do not allocate
371 	 * low-numbered pids.
372 	 */
373 	trypid = lastpid + 1;
374 	if (flags & RFHIGHPID) {
375 		if (trypid < 10)
376 			trypid = 10;
377 	} else {
378 		if (randompid)
379 			trypid += arc4random() % randompid;
380 	}
381 retry:
382 	/*
383 	 * If the process ID prototype has wrapped around,
384 	 * restart somewhat above 0, as the low-numbered procs
385 	 * tend to include daemons that don't exit.
386 	 */
387 	if (trypid >= PID_MAX) {
388 		trypid = trypid % PID_MAX;
389 		if (trypid < 100)
390 			trypid += 100;
391 		pidchecked = 0;
392 	}
393 	if (trypid >= pidchecked) {
394 		int doingzomb = 0;
395 
396 		pidchecked = PID_MAX;
397 		/*
398 		 * Scan the active and zombie procs to check whether this pid
399 		 * is in use.  Remember the lowest pid that's greater
400 		 * than trypid, so we can avoid checking for a while.
401 		 */
402 		p2 = LIST_FIRST(&allproc);
403 again:
404 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
405 			while (p2->p_pid == trypid ||
406 			    (p2->p_pgrp != NULL &&
407 			    (p2->p_pgrp->pg_id == trypid ||
408 			    (p2->p_session != NULL &&
409 			    p2->p_session->s_sid == trypid)))) {
410 				trypid++;
411 				if (trypid >= pidchecked)
412 					goto retry;
413 			}
414 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
415 				pidchecked = p2->p_pid;
416 			if (p2->p_pgrp != NULL) {
417 				if (p2->p_pgrp->pg_id > trypid &&
418 				    pidchecked > p2->p_pgrp->pg_id)
419 					pidchecked = p2->p_pgrp->pg_id;
420 				if (p2->p_session != NULL &&
421 				    p2->p_session->s_sid > trypid &&
422 				    pidchecked > p2->p_session->s_sid)
423 					pidchecked = p2->p_session->s_sid;
424 			}
425 		}
426 		if (!doingzomb) {
427 			doingzomb = 1;
428 			p2 = LIST_FIRST(&zombproc);
429 			goto again;
430 		}
431 	}
432 	sx_sunlock(&proctree_lock);
433 
434 	/*
435 	 * RFHIGHPID does not mess with the lastpid counter during boot.
436 	 */
437 	if (flags & RFHIGHPID)
438 		pidchecked = 0;
439 	else
440 		lastpid = trypid;
441 
442 	p2 = newproc;
443 	p2->p_state = PRS_NEW;		/* protect against others */
444 	p2->p_pid = trypid;
445 	/*
446 	 * Allow the scheduler to initialize the child.
447 	 */
448 	thread_lock(td);
449 	sched_fork(td, td2);
450 	thread_unlock(td);
451 	AUDIT_ARG_PID(p2->p_pid);
452 	LIST_INSERT_HEAD(&allproc, p2, p_list);
453 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
454 
455 	PROC_LOCK(p2);
456 	PROC_LOCK(p1);
457 
458 	sx_xunlock(&allproc_lock);
459 
460 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
461 	    __rangeof(struct proc, p_startcopy, p_endcopy));
462 	pargs_hold(p2->p_args);
463 	PROC_UNLOCK(p1);
464 
465 	bzero(&p2->p_startzero,
466 	    __rangeof(struct proc, p_startzero, p_endzero));
467 
468 	p2->p_ucred = crhold(td->td_ucred);
469 
470 	/* Tell the prison that we exist. */
471 	prison_proc_hold(p2->p_ucred->cr_prison);
472 
473 	PROC_UNLOCK(p2);
474 
475 	/*
476 	 * Malloc things while we don't hold any locks.
477 	 */
478 	if (flags & RFSIGSHARE)
479 		newsigacts = NULL;
480 	else
481 		newsigacts = sigacts_alloc();
482 
483 	/*
484 	 * Copy filedesc.
485 	 */
486 	if (flags & RFCFDG) {
487 		fd = fdinit(p1->p_fd);
488 		fdtol = NULL;
489 	} else if (flags & RFFDG) {
490 		fd = fdcopy(p1->p_fd);
491 		fdtol = NULL;
492 	} else {
493 		fd = fdshare(p1->p_fd);
494 		if (p1->p_fdtol == NULL)
495 			p1->p_fdtol =
496 				filedesc_to_leader_alloc(NULL,
497 							 NULL,
498 							 p1->p_leader);
499 		if ((flags & RFTHREAD) != 0) {
500 			/*
501 			 * Shared file descriptor table and
502 			 * shared process leaders.
503 			 */
504 			fdtol = p1->p_fdtol;
505 			FILEDESC_XLOCK(p1->p_fd);
506 			fdtol->fdl_refcount++;
507 			FILEDESC_XUNLOCK(p1->p_fd);
508 		} else {
509 			/*
510 			 * Shared file descriptor table, and
511 			 * different process leaders
512 			 */
513 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
514 							 p1->p_fd,
515 							 p2);
516 		}
517 	}
518 	/*
519 	 * Make a proc table entry for the new process.
520 	 * Start by zeroing the section of proc that is zero-initialized,
521 	 * then copy the section that is copied directly from the parent.
522 	 */
523 
524 	PROC_LOCK(p2);
525 	PROC_LOCK(p1);
526 
527 	bzero(&td2->td_startzero,
528 	    __rangeof(struct thread, td_startzero, td_endzero));
529 
530 	bcopy(&td->td_startcopy, &td2->td_startcopy,
531 	    __rangeof(struct thread, td_startcopy, td_endcopy));
532 
533 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
534 	td2->td_sigstk = td->td_sigstk;
535 	td2->td_sigmask = td->td_sigmask;
536 	td2->td_flags = TDF_INMEM;
537 
538 #ifdef VIMAGE
539 	td2->td_vnet = NULL;
540 	td2->td_vnet_lpush = NULL;
541 #endif
542 
543 	/*
544 	 * Duplicate sub-structures as needed.
545 	 * Increase reference counts on shared objects.
546 	 */
547 	p2->p_flag = P_INMEM;
548 	p2->p_swtick = ticks;
549 	if (p1->p_flag & P_PROFIL)
550 		startprofclock(p2);
551 	td2->td_ucred = crhold(p2->p_ucred);
552 
553 	if (flags & RFSIGSHARE) {
554 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
555 	} else {
556 		sigacts_copy(newsigacts, p1->p_sigacts);
557 		p2->p_sigacts = newsigacts;
558 	}
559 	if (flags & RFLINUXTHPN)
560 	        p2->p_sigparent = SIGUSR1;
561 	else
562 	        p2->p_sigparent = SIGCHLD;
563 
564 	p2->p_textvp = p1->p_textvp;
565 	p2->p_fd = fd;
566 	p2->p_fdtol = fdtol;
567 
568 	/*
569 	 * p_limit is copy-on-write.  Bump its refcount.
570 	 */
571 	lim_fork(p1, p2);
572 
573 	pstats_fork(p1->p_stats, p2->p_stats);
574 
575 	PROC_UNLOCK(p1);
576 	PROC_UNLOCK(p2);
577 
578 	/* Bump references to the text vnode (for procfs) */
579 	if (p2->p_textvp)
580 		vref(p2->p_textvp);
581 
582 	/*
583 	 * Set up linkage for kernel based threading.
584 	 */
585 	if ((flags & RFTHREAD) != 0) {
586 		mtx_lock(&ppeers_lock);
587 		p2->p_peers = p1->p_peers;
588 		p1->p_peers = p2;
589 		p2->p_leader = p1->p_leader;
590 		mtx_unlock(&ppeers_lock);
591 		PROC_LOCK(p1->p_leader);
592 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
593 			PROC_UNLOCK(p1->p_leader);
594 			/*
595 			 * The task leader is exiting, so process p1 is
596 			 * going to be killed shortly.  Since p1 obviously
597 			 * isn't dead yet, we know that the leader is either
598 			 * sending SIGKILL's to all the processes in this
599 			 * task or is sleeping waiting for all the peers to
600 			 * exit.  We let p1 complete the fork, but we need
601 			 * to go ahead and kill the new process p2 since
602 			 * the task leader may not get a chance to send
603 			 * SIGKILL to it.  We leave it on the list so that
604 			 * the task leader will wait for this new process
605 			 * to commit suicide.
606 			 */
607 			PROC_LOCK(p2);
608 			psignal(p2, SIGKILL);
609 			PROC_UNLOCK(p2);
610 		} else
611 			PROC_UNLOCK(p1->p_leader);
612 	} else {
613 		p2->p_peers = NULL;
614 		p2->p_leader = p2;
615 	}
616 
617 	sx_xlock(&proctree_lock);
618 	PGRP_LOCK(p1->p_pgrp);
619 	PROC_LOCK(p2);
620 	PROC_LOCK(p1);
621 
622 	/*
623 	 * Preserve some more flags in subprocess.  P_PROFIL has already
624 	 * been preserved.
625 	 */
626 	p2->p_flag |= p1->p_flag & P_SUGID;
627 	td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
628 	SESS_LOCK(p1->p_session);
629 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
630 		p2->p_flag |= P_CONTROLT;
631 	SESS_UNLOCK(p1->p_session);
632 	if (flags & RFPPWAIT)
633 		p2->p_flag |= P_PPWAIT;
634 
635 	p2->p_pgrp = p1->p_pgrp;
636 	LIST_INSERT_AFTER(p1, p2, p_pglist);
637 	PGRP_UNLOCK(p1->p_pgrp);
638 	LIST_INIT(&p2->p_children);
639 
640 	callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
641 
642 #ifdef KTRACE
643 	/*
644 	 * Copy traceflag and tracefile if enabled.
645 	 */
646 	mtx_lock(&ktrace_mtx);
647 	KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
648 	if (p1->p_traceflag & KTRFAC_INHERIT) {
649 		p2->p_traceflag = p1->p_traceflag;
650 		if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
651 			VREF(p2->p_tracevp);
652 			KASSERT(p1->p_tracecred != NULL,
653 			    ("ktrace vnode with no cred"));
654 			p2->p_tracecred = crhold(p1->p_tracecred);
655 		}
656 	}
657 	mtx_unlock(&ktrace_mtx);
658 #endif
659 
660 	/*
661 	 * If PF_FORK is set, the child process inherits the
662 	 * procfs ioctl flags from its parent.
663 	 */
664 	if (p1->p_pfsflags & PF_FORK) {
665 		p2->p_stops = p1->p_stops;
666 		p2->p_pfsflags = p1->p_pfsflags;
667 	}
668 
669 #ifdef KDTRACE_HOOKS
670 	/*
671 	 * Tell the DTrace fasttrap provider about the new process
672 	 * if it has registered an interest.
673 	 */
674 	if (dtrace_fasttrap_fork)
675 		dtrace_fasttrap_fork(p1, p2);
676 #endif
677 
678 	/*
679 	 * This begins the section where we must prevent the parent
680 	 * from being swapped.
681 	 */
682 	_PHOLD(p1);
683 	PROC_UNLOCK(p1);
684 
685 	/*
686 	 * Attach the new process to its parent.
687 	 *
688 	 * If RFNOWAIT is set, the newly created process becomes a child
689 	 * of init.  This effectively disassociates the child from the
690 	 * parent.
691 	 */
692 	if (flags & RFNOWAIT)
693 		pptr = initproc;
694 	else
695 		pptr = p1;
696 	p2->p_pptr = pptr;
697 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
698 	sx_xunlock(&proctree_lock);
699 
700 	/* Inform accounting that we have forked. */
701 	p2->p_acflag = AFORK;
702 	PROC_UNLOCK(p2);
703 
704 	/*
705 	 * Finish creating the child process.  It will return via a different
706 	 * execution path later.  (ie: directly into user mode)
707 	 */
708 	vm_forkproc(td, p2, td2, vm2, flags);
709 
710 	if (flags == (RFFDG | RFPROC)) {
711 		PCPU_INC(cnt.v_forks);
712 		PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
713 		    p2->p_vmspace->vm_ssize);
714 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
715 		PCPU_INC(cnt.v_vforks);
716 		PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
717 		    p2->p_vmspace->vm_ssize);
718 	} else if (p1 == &proc0) {
719 		PCPU_INC(cnt.v_kthreads);
720 		PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
721 		    p2->p_vmspace->vm_ssize);
722 	} else {
723 		PCPU_INC(cnt.v_rforks);
724 		PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
725 		    p2->p_vmspace->vm_ssize);
726 	}
727 
728 	/*
729 	 * Both processes are set up, now check if any loadable modules want
730 	 * to adjust anything.
731 	 *   What if they have an error? XXX
732 	 */
733 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
734 
735 	/*
736 	 * Set the child start time and mark the process as being complete.
737 	 */
738 	microuptime(&p2->p_stats->p_start);
739 	PROC_SLOCK(p2);
740 	p2->p_state = PRS_NORMAL;
741 	PROC_SUNLOCK(p2);
742 
743 	/*
744 	 * If RFSTOPPED not requested, make child runnable and add to
745 	 * run queue.
746 	 */
747 	if ((flags & RFSTOPPED) == 0) {
748 		thread_lock(td2);
749 		TD_SET_CAN_RUN(td2);
750 		sched_add(td2, SRQ_BORING);
751 		thread_unlock(td2);
752 	}
753 
754 	/*
755 	 * Now can be swapped.
756 	 */
757 	PROC_LOCK(p1);
758 	_PRELE(p1);
759 	PROC_UNLOCK(p1);
760 
761 	/*
762 	 * Tell any interested parties about the new process.
763 	 */
764 	knote_fork(&p1->p_klist, p2->p_pid);
765 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
766 
767 	/*
768 	 * Preserve synchronization semantics of vfork.  If waiting for
769 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
770 	 * proc (in case of exit).
771 	 */
772 	PROC_LOCK(p2);
773 	while (p2->p_flag & P_PPWAIT)
774 		cv_wait(&p2->p_pwait, &p2->p_mtx);
775 	PROC_UNLOCK(p2);
776 
777 	/*
778 	 * Return child proc pointer to parent.
779 	 */
780 	*procp = p2;
781 	return (0);
782 fail:
783 	sx_sunlock(&proctree_lock);
784 	if (ppsratecheck(&lastfail, &curfail, 1))
785 		printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
786 		    td->td_ucred->cr_ruid);
787 	sx_xunlock(&allproc_lock);
788 #ifdef MAC
789 	mac_proc_destroy(newproc);
790 #endif
791 fail1:
792 	if (vm2 != NULL)
793 		vmspace_free(vm2);
794 	uma_zfree(proc_zone, newproc);
795 	pause("fork", hz / 2);
796 	return (error);
797 }
798 
799 /*
800  * Handle the return of a child process from fork1().  This function
801  * is called from the MD fork_trampoline() entry point.
802  */
803 void
804 fork_exit(callout, arg, frame)
805 	void (*callout)(void *, struct trapframe *);
806 	void *arg;
807 	struct trapframe *frame;
808 {
809 	struct proc *p;
810 	struct thread *td;
811 	struct thread *dtd;
812 
813 	td = curthread;
814 	p = td->td_proc;
815 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
816 
817 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
818 		td, td->td_sched, p->p_pid, td->td_name);
819 
820 	sched_fork_exit(td);
821 	/*
822 	* Processes normally resume in mi_switch() after being
823 	* cpu_switch()'ed to, but when children start up they arrive here
824 	* instead, so we must do much the same things as mi_switch() would.
825 	*/
826 	if ((dtd = PCPU_GET(deadthread))) {
827 		PCPU_SET(deadthread, NULL);
828 		thread_stash(dtd);
829 	}
830 	thread_unlock(td);
831 
832 	/*
833 	 * cpu_set_fork_handler intercepts this function call to
834 	 * have this call a non-return function to stay in kernel mode.
835 	 * initproc has its own fork handler, but it does return.
836 	 */
837 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
838 	callout(arg, frame);
839 
840 	/*
841 	 * Check if a kernel thread misbehaved and returned from its main
842 	 * function.
843 	 */
844 	if (p->p_flag & P_KTHREAD) {
845 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
846 		    td->td_name, p->p_pid);
847 		kproc_exit(0);
848 	}
849 	mtx_assert(&Giant, MA_NOTOWNED);
850 
851 	EVENTHANDLER_INVOKE(schedtail, p);
852 }
853 
854 /*
855  * Simplified back end of syscall(), used when returning from fork()
856  * directly into user mode.  Giant is not held on entry, and must not
857  * be held on return.  This function is passed in to fork_exit() as the
858  * first parameter and is called when returning to a new userland process.
859  */
860 void
861 fork_return(td, frame)
862 	struct thread *td;
863 	struct trapframe *frame;
864 {
865 
866 	userret(td, frame);
867 #ifdef KTRACE
868 	if (KTRPOINT(td, KTR_SYSRET))
869 		ktrsysret(SYS_fork, 0, 0);
870 #endif
871 	mtx_assert(&Giant, MA_NOTOWNED);
872 }
873