xref: /freebsd/sys/kern/kern_fork.c (revision c193de568b428d4159b2722b4962034cdcd3d36b)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_kstack_pages.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/eventhandler.h>
48 #include <sys/filedesc.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/sysctl.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mutex.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/pioctl.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
61 #include <sys/syscall.h>
62 #include <sys/vmmeter.h>
63 #include <sys/vnode.h>
64 #include <sys/acct.h>
65 #include <sys/ktr.h>
66 #include <sys/ktrace.h>
67 #include <sys/unistd.h>
68 #include <sys/sdt.h>
69 #include <sys/sx.h>
70 #include <sys/signalvar.h>
71 
72 #include <security/audit/audit.h>
73 #include <security/mac/mac_framework.h>
74 
75 #include <vm/vm.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_extern.h>
79 #include <vm/uma.h>
80 
81 #ifdef KDTRACE_HOOKS
82 #include <sys/dtrace_bsd.h>
83 dtrace_fork_func_t	dtrace_fasttrap_fork;
84 #endif
85 
86 SDT_PROVIDER_DECLARE(proc);
87 SDT_PROBE_DEFINE(proc, kernel, , create);
88 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
90 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
91 
92 #ifndef _SYS_SYSPROTO_H_
93 struct fork_args {
94 	int     dummy;
95 };
96 #endif
97 
98 /* ARGSUSED */
99 int
100 fork(td, uap)
101 	struct thread *td;
102 	struct fork_args *uap;
103 {
104 	int error;
105 	struct proc *p2;
106 
107 	error = fork1(td, RFFDG | RFPROC, 0, &p2);
108 	if (error == 0) {
109 		td->td_retval[0] = p2->p_pid;
110 		td->td_retval[1] = 0;
111 	}
112 	return (error);
113 }
114 
115 /* ARGSUSED */
116 int
117 vfork(td, uap)
118 	struct thread *td;
119 	struct vfork_args *uap;
120 {
121 	int error, flags;
122 	struct proc *p2;
123 
124 #ifdef XEN
125 	flags = RFFDG | RFPROC; /* validate that this is still an issue */
126 #else
127 	flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
128 #endif
129 	error = fork1(td, flags, 0, &p2);
130 	if (error == 0) {
131 		td->td_retval[0] = p2->p_pid;
132 		td->td_retval[1] = 0;
133 	}
134 	return (error);
135 }
136 
137 int
138 rfork(td, uap)
139 	struct thread *td;
140 	struct rfork_args *uap;
141 {
142 	struct proc *p2;
143 	int error;
144 
145 	/* Don't allow kernel-only flags. */
146 	if ((uap->flags & RFKERNELONLY) != 0)
147 		return (EINVAL);
148 
149 	AUDIT_ARG_FFLAGS(uap->flags);
150 	error = fork1(td, uap->flags, 0, &p2);
151 	if (error == 0) {
152 		td->td_retval[0] = p2 ? p2->p_pid : 0;
153 		td->td_retval[1] = 0;
154 	}
155 	return (error);
156 }
157 
158 int	nprocs = 1;		/* process 0 */
159 int	lastpid = 0;
160 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
161     "Last used PID");
162 
163 /*
164  * Random component to lastpid generation.  We mix in a random factor to make
165  * it a little harder to predict.  We sanity check the modulus value to avoid
166  * doing it in critical paths.  Don't let it be too small or we pointlessly
167  * waste randomness entropy, and don't let it be impossibly large.  Using a
168  * modulus that is too big causes a LOT more process table scans and slows
169  * down fork processing as the pidchecked caching is defeated.
170  */
171 static int randompid = 0;
172 
173 static int
174 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
175 {
176 	int error, pid;
177 
178 	error = sysctl_wire_old_buffer(req, sizeof(int));
179 	if (error != 0)
180 		return(error);
181 	sx_xlock(&allproc_lock);
182 	pid = randompid;
183 	error = sysctl_handle_int(oidp, &pid, 0, req);
184 	if (error == 0 && req->newptr != NULL) {
185 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
186 			pid = PID_MAX - 100;
187 		else if (pid < 2)			/* NOP */
188 			pid = 0;
189 		else if (pid < 100)			/* Make it reasonable */
190 			pid = 100;
191 		randompid = pid;
192 	}
193 	sx_xunlock(&allproc_lock);
194 	return (error);
195 }
196 
197 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
198     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
199 
200 int
201 fork1(td, flags, pages, procp)
202 	struct thread *td;
203 	int flags;
204 	int pages;
205 	struct proc **procp;
206 {
207 	struct proc *p1, *p2, *pptr;
208 	struct proc *newproc;
209 	int ok, trypid;
210 	static int curfail, pidchecked = 0;
211 	static struct timeval lastfail;
212 	struct filedesc *fd;
213 	struct filedesc_to_leader *fdtol;
214 	struct thread *td2;
215 	struct sigacts *newsigacts;
216 	struct vmspace *vm2;
217 	vm_ooffset_t mem_charged;
218 	int error;
219 
220 	/* Can't copy and clear. */
221 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
222 		return (EINVAL);
223 
224 	p1 = td->td_proc;
225 
226 	/*
227 	 * Here we don't create a new process, but we divorce
228 	 * certain parts of a process from itself.
229 	 */
230 	if ((flags & RFPROC) == 0) {
231 		if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
232 		    (flags & (RFCFDG | RFFDG))) {
233 			PROC_LOCK(p1);
234 			if (thread_single(SINGLE_BOUNDARY)) {
235 				PROC_UNLOCK(p1);
236 				return (ERESTART);
237 			}
238 			PROC_UNLOCK(p1);
239 		}
240 
241 		error = vm_forkproc(td, NULL, NULL, NULL, flags);
242 		if (error)
243 			goto norfproc_fail;
244 
245 		/*
246 		 * Close all file descriptors.
247 		 */
248 		if (flags & RFCFDG) {
249 			struct filedesc *fdtmp;
250 			fdtmp = fdinit(td->td_proc->p_fd);
251 			fdfree(td);
252 			p1->p_fd = fdtmp;
253 		}
254 
255 		/*
256 		 * Unshare file descriptors (from parent).
257 		 */
258 		if (flags & RFFDG)
259 			fdunshare(p1, td);
260 
261 norfproc_fail:
262 		if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
263 		    (flags & (RFCFDG | RFFDG))) {
264 			PROC_LOCK(p1);
265 			thread_single_end();
266 			PROC_UNLOCK(p1);
267 		}
268 		*procp = NULL;
269 		return (error);
270 	}
271 
272 	/*
273 	 * XXX
274 	 * We did have single-threading code here
275 	 * however it proved un-needed and caused problems
276 	 */
277 
278 	mem_charged = 0;
279 	vm2 = NULL;
280 	if (pages == 0)
281 		pages = KSTACK_PAGES;
282 	/* Allocate new proc. */
283 	newproc = uma_zalloc(proc_zone, M_WAITOK);
284 	td2 = FIRST_THREAD_IN_PROC(newproc);
285 	if (td2 == NULL) {
286 		td2 = thread_alloc(pages);
287 		if (td2 == NULL) {
288 			error = ENOMEM;
289 			goto fail1;
290 		}
291 		proc_linkup(newproc, td2);
292 	} else {
293 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
294 			if (td2->td_kstack != 0)
295 				vm_thread_dispose(td2);
296 			if (!thread_alloc_stack(td2, pages)) {
297 				error = ENOMEM;
298 				goto fail1;
299 			}
300 		}
301 	}
302 
303 	if ((flags & RFMEM) == 0) {
304 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
305 		if (vm2 == NULL) {
306 			error = ENOMEM;
307 			goto fail1;
308 		}
309 		if (!swap_reserve(mem_charged)) {
310 			/*
311 			 * The swap reservation failed. The accounting
312 			 * from the entries of the copied vm2 will be
313 			 * substracted in vmspace_free(), so force the
314 			 * reservation there.
315 			 */
316 			swap_reserve_force(mem_charged);
317 			error = ENOMEM;
318 			goto fail1;
319 		}
320 	} else
321 		vm2 = NULL;
322 #ifdef MAC
323 	mac_proc_init(newproc);
324 #endif
325 	knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
326 	STAILQ_INIT(&newproc->p_ktr);
327 
328 	/* We have to lock the process tree while we look for a pid. */
329 	sx_slock(&proctree_lock);
330 
331 	/*
332 	 * Although process entries are dynamically created, we still keep
333 	 * a global limit on the maximum number we will create.  Don't allow
334 	 * a nonprivileged user to use the last ten processes; don't let root
335 	 * exceed the limit. The variable nprocs is the current number of
336 	 * processes, maxproc is the limit.
337 	 */
338 	sx_xlock(&allproc_lock);
339 	if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
340 	    PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
341 		error = EAGAIN;
342 		goto fail;
343 	}
344 
345 	/*
346 	 * Increment the count of procs running with this uid. Don't allow
347 	 * a nonprivileged user to exceed their current limit.
348 	 *
349 	 * XXXRW: Can we avoid privilege here if it's not needed?
350 	 */
351 	error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
352 	if (error == 0)
353 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
354 	else {
355 		PROC_LOCK(p1);
356 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
357 		    lim_cur(p1, RLIMIT_NPROC));
358 		PROC_UNLOCK(p1);
359 	}
360 	if (!ok) {
361 		error = EAGAIN;
362 		goto fail;
363 	}
364 
365 	/*
366 	 * Increment the nprocs resource before blocking can occur.  There
367 	 * are hard-limits as to the number of processes that can run.
368 	 */
369 	nprocs++;
370 
371 	/*
372 	 * Find an unused process ID.  We remember a range of unused IDs
373 	 * ready to use (from lastpid+1 through pidchecked-1).
374 	 *
375 	 * If RFHIGHPID is set (used during system boot), do not allocate
376 	 * low-numbered pids.
377 	 */
378 	trypid = lastpid + 1;
379 	if (flags & RFHIGHPID) {
380 		if (trypid < 10)
381 			trypid = 10;
382 	} else {
383 		if (randompid)
384 			trypid += arc4random() % randompid;
385 	}
386 retry:
387 	/*
388 	 * If the process ID prototype has wrapped around,
389 	 * restart somewhat above 0, as the low-numbered procs
390 	 * tend to include daemons that don't exit.
391 	 */
392 	if (trypid >= PID_MAX) {
393 		trypid = trypid % PID_MAX;
394 		if (trypid < 100)
395 			trypid += 100;
396 		pidchecked = 0;
397 	}
398 	if (trypid >= pidchecked) {
399 		int doingzomb = 0;
400 
401 		pidchecked = PID_MAX;
402 		/*
403 		 * Scan the active and zombie procs to check whether this pid
404 		 * is in use.  Remember the lowest pid that's greater
405 		 * than trypid, so we can avoid checking for a while.
406 		 */
407 		p2 = LIST_FIRST(&allproc);
408 again:
409 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
410 			while (p2->p_pid == trypid ||
411 			    (p2->p_pgrp != NULL &&
412 			    (p2->p_pgrp->pg_id == trypid ||
413 			    (p2->p_session != NULL &&
414 			    p2->p_session->s_sid == trypid)))) {
415 				trypid++;
416 				if (trypid >= pidchecked)
417 					goto retry;
418 			}
419 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
420 				pidchecked = p2->p_pid;
421 			if (p2->p_pgrp != NULL) {
422 				if (p2->p_pgrp->pg_id > trypid &&
423 				    pidchecked > p2->p_pgrp->pg_id)
424 					pidchecked = p2->p_pgrp->pg_id;
425 				if (p2->p_session != NULL &&
426 				    p2->p_session->s_sid > trypid &&
427 				    pidchecked > p2->p_session->s_sid)
428 					pidchecked = p2->p_session->s_sid;
429 			}
430 		}
431 		if (!doingzomb) {
432 			doingzomb = 1;
433 			p2 = LIST_FIRST(&zombproc);
434 			goto again;
435 		}
436 	}
437 	sx_sunlock(&proctree_lock);
438 
439 	/*
440 	 * RFHIGHPID does not mess with the lastpid counter during boot.
441 	 */
442 	if (flags & RFHIGHPID)
443 		pidchecked = 0;
444 	else
445 		lastpid = trypid;
446 
447 	p2 = newproc;
448 	p2->p_state = PRS_NEW;		/* protect against others */
449 	p2->p_pid = trypid;
450 	/*
451 	 * Allow the scheduler to initialize the child.
452 	 */
453 	thread_lock(td);
454 	sched_fork(td, td2);
455 	thread_unlock(td);
456 	AUDIT_ARG_PID(p2->p_pid);
457 	LIST_INSERT_HEAD(&allproc, p2, p_list);
458 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
459 
460 	PROC_LOCK(p2);
461 	PROC_LOCK(p1);
462 
463 	sx_xunlock(&allproc_lock);
464 
465 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
466 	    __rangeof(struct proc, p_startcopy, p_endcopy));
467 	pargs_hold(p2->p_args);
468 	PROC_UNLOCK(p1);
469 
470 	bzero(&p2->p_startzero,
471 	    __rangeof(struct proc, p_startzero, p_endzero));
472 
473 	p2->p_ucred = crhold(td->td_ucred);
474 
475 	/* Tell the prison that we exist. */
476 	prison_proc_hold(p2->p_ucred->cr_prison);
477 
478 	PROC_UNLOCK(p2);
479 
480 	/*
481 	 * Malloc things while we don't hold any locks.
482 	 */
483 	if (flags & RFSIGSHARE)
484 		newsigacts = NULL;
485 	else
486 		newsigacts = sigacts_alloc();
487 
488 	/*
489 	 * Copy filedesc.
490 	 */
491 	if (flags & RFCFDG) {
492 		fd = fdinit(p1->p_fd);
493 		fdtol = NULL;
494 	} else if (flags & RFFDG) {
495 		fd = fdcopy(p1->p_fd);
496 		fdtol = NULL;
497 	} else {
498 		fd = fdshare(p1->p_fd);
499 		if (p1->p_fdtol == NULL)
500 			p1->p_fdtol =
501 				filedesc_to_leader_alloc(NULL,
502 							 NULL,
503 							 p1->p_leader);
504 		if ((flags & RFTHREAD) != 0) {
505 			/*
506 			 * Shared file descriptor table and
507 			 * shared process leaders.
508 			 */
509 			fdtol = p1->p_fdtol;
510 			FILEDESC_XLOCK(p1->p_fd);
511 			fdtol->fdl_refcount++;
512 			FILEDESC_XUNLOCK(p1->p_fd);
513 		} else {
514 			/*
515 			 * Shared file descriptor table, and
516 			 * different process leaders
517 			 */
518 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
519 							 p1->p_fd,
520 							 p2);
521 		}
522 	}
523 	/*
524 	 * Make a proc table entry for the new process.
525 	 * Start by zeroing the section of proc that is zero-initialized,
526 	 * then copy the section that is copied directly from the parent.
527 	 */
528 
529 	PROC_LOCK(p2);
530 	PROC_LOCK(p1);
531 
532 	bzero(&td2->td_startzero,
533 	    __rangeof(struct thread, td_startzero, td_endzero));
534 	bzero(&td2->td_rux, sizeof(td2->td_rux));
535 
536 	bcopy(&td->td_startcopy, &td2->td_startcopy,
537 	    __rangeof(struct thread, td_startcopy, td_endcopy));
538 
539 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
540 	td2->td_sigstk = td->td_sigstk;
541 	td2->td_sigmask = td->td_sigmask;
542 	td2->td_flags = TDF_INMEM;
543 
544 #ifdef VIMAGE
545 	td2->td_vnet = NULL;
546 	td2->td_vnet_lpush = NULL;
547 #endif
548 
549 	/*
550 	 * Duplicate sub-structures as needed.
551 	 * Increase reference counts on shared objects.
552 	 */
553 	p2->p_flag = P_INMEM;
554 	p2->p_swtick = ticks;
555 	if (p1->p_flag & P_PROFIL)
556 		startprofclock(p2);
557 	td2->td_ucred = crhold(p2->p_ucred);
558 
559 	if (flags & RFSIGSHARE) {
560 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
561 	} else {
562 		sigacts_copy(newsigacts, p1->p_sigacts);
563 		p2->p_sigacts = newsigacts;
564 	}
565 	if (flags & RFLINUXTHPN)
566 	        p2->p_sigparent = SIGUSR1;
567 	else
568 	        p2->p_sigparent = SIGCHLD;
569 
570 	p2->p_textvp = p1->p_textvp;
571 	p2->p_fd = fd;
572 	p2->p_fdtol = fdtol;
573 
574 	/*
575 	 * p_limit is copy-on-write.  Bump its refcount.
576 	 */
577 	lim_fork(p1, p2);
578 
579 	pstats_fork(p1->p_stats, p2->p_stats);
580 
581 	PROC_UNLOCK(p1);
582 	PROC_UNLOCK(p2);
583 
584 	/* Bump references to the text vnode (for procfs) */
585 	if (p2->p_textvp)
586 		vref(p2->p_textvp);
587 
588 	/*
589 	 * Set up linkage for kernel based threading.
590 	 */
591 	if ((flags & RFTHREAD) != 0) {
592 		mtx_lock(&ppeers_lock);
593 		p2->p_peers = p1->p_peers;
594 		p1->p_peers = p2;
595 		p2->p_leader = p1->p_leader;
596 		mtx_unlock(&ppeers_lock);
597 		PROC_LOCK(p1->p_leader);
598 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
599 			PROC_UNLOCK(p1->p_leader);
600 			/*
601 			 * The task leader is exiting, so process p1 is
602 			 * going to be killed shortly.  Since p1 obviously
603 			 * isn't dead yet, we know that the leader is either
604 			 * sending SIGKILL's to all the processes in this
605 			 * task or is sleeping waiting for all the peers to
606 			 * exit.  We let p1 complete the fork, but we need
607 			 * to go ahead and kill the new process p2 since
608 			 * the task leader may not get a chance to send
609 			 * SIGKILL to it.  We leave it on the list so that
610 			 * the task leader will wait for this new process
611 			 * to commit suicide.
612 			 */
613 			PROC_LOCK(p2);
614 			psignal(p2, SIGKILL);
615 			PROC_UNLOCK(p2);
616 		} else
617 			PROC_UNLOCK(p1->p_leader);
618 	} else {
619 		p2->p_peers = NULL;
620 		p2->p_leader = p2;
621 	}
622 
623 	sx_xlock(&proctree_lock);
624 	PGRP_LOCK(p1->p_pgrp);
625 	PROC_LOCK(p2);
626 	PROC_LOCK(p1);
627 
628 	/*
629 	 * Preserve some more flags in subprocess.  P_PROFIL has already
630 	 * been preserved.
631 	 */
632 	p2->p_flag |= p1->p_flag & P_SUGID;
633 	td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
634 	SESS_LOCK(p1->p_session);
635 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
636 		p2->p_flag |= P_CONTROLT;
637 	SESS_UNLOCK(p1->p_session);
638 	if (flags & RFPPWAIT)
639 		p2->p_flag |= P_PPWAIT;
640 
641 	p2->p_pgrp = p1->p_pgrp;
642 	LIST_INSERT_AFTER(p1, p2, p_pglist);
643 	PGRP_UNLOCK(p1->p_pgrp);
644 	LIST_INIT(&p2->p_children);
645 
646 	callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
647 
648 #ifdef KTRACE
649 	/*
650 	 * Copy traceflag and tracefile if enabled.
651 	 */
652 	mtx_lock(&ktrace_mtx);
653 	KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
654 	if (p1->p_traceflag & KTRFAC_INHERIT) {
655 		p2->p_traceflag = p1->p_traceflag;
656 		if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
657 			VREF(p2->p_tracevp);
658 			KASSERT(p1->p_tracecred != NULL,
659 			    ("ktrace vnode with no cred"));
660 			p2->p_tracecred = crhold(p1->p_tracecred);
661 		}
662 	}
663 	mtx_unlock(&ktrace_mtx);
664 #endif
665 
666 	/*
667 	 * If PF_FORK is set, the child process inherits the
668 	 * procfs ioctl flags from its parent.
669 	 */
670 	if (p1->p_pfsflags & PF_FORK) {
671 		p2->p_stops = p1->p_stops;
672 		p2->p_pfsflags = p1->p_pfsflags;
673 	}
674 
675 #ifdef KDTRACE_HOOKS
676 	/*
677 	 * Tell the DTrace fasttrap provider about the new process
678 	 * if it has registered an interest.
679 	 */
680 	if (dtrace_fasttrap_fork)
681 		dtrace_fasttrap_fork(p1, p2);
682 #endif
683 
684 	/*
685 	 * This begins the section where we must prevent the parent
686 	 * from being swapped.
687 	 */
688 	_PHOLD(p1);
689 	PROC_UNLOCK(p1);
690 
691 	/*
692 	 * Attach the new process to its parent.
693 	 *
694 	 * If RFNOWAIT is set, the newly created process becomes a child
695 	 * of init.  This effectively disassociates the child from the
696 	 * parent.
697 	 */
698 	if (flags & RFNOWAIT)
699 		pptr = initproc;
700 	else
701 		pptr = p1;
702 	p2->p_pptr = pptr;
703 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
704 	sx_xunlock(&proctree_lock);
705 
706 	/* Inform accounting that we have forked. */
707 	p2->p_acflag = AFORK;
708 	PROC_UNLOCK(p2);
709 
710 	/*
711 	 * Finish creating the child process.  It will return via a different
712 	 * execution path later.  (ie: directly into user mode)
713 	 */
714 	vm_forkproc(td, p2, td2, vm2, flags);
715 
716 	if (flags == (RFFDG | RFPROC)) {
717 		PCPU_INC(cnt.v_forks);
718 		PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
719 		    p2->p_vmspace->vm_ssize);
720 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
721 		PCPU_INC(cnt.v_vforks);
722 		PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
723 		    p2->p_vmspace->vm_ssize);
724 	} else if (p1 == &proc0) {
725 		PCPU_INC(cnt.v_kthreads);
726 		PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
727 		    p2->p_vmspace->vm_ssize);
728 	} else {
729 		PCPU_INC(cnt.v_rforks);
730 		PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
731 		    p2->p_vmspace->vm_ssize);
732 	}
733 
734 	/*
735 	 * Both processes are set up, now check if any loadable modules want
736 	 * to adjust anything.
737 	 *   What if they have an error? XXX
738 	 */
739 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
740 
741 	/*
742 	 * Set the child start time and mark the process as being complete.
743 	 */
744 	microuptime(&p2->p_stats->p_start);
745 	PROC_SLOCK(p2);
746 	p2->p_state = PRS_NORMAL;
747 	PROC_SUNLOCK(p2);
748 
749 	/*
750 	 * If RFSTOPPED not requested, make child runnable and add to
751 	 * run queue.
752 	 */
753 	if ((flags & RFSTOPPED) == 0) {
754 		thread_lock(td2);
755 		TD_SET_CAN_RUN(td2);
756 		sched_add(td2, SRQ_BORING);
757 		thread_unlock(td2);
758 	}
759 
760 	/*
761 	 * Now can be swapped.
762 	 */
763 	PROC_LOCK(p1);
764 	_PRELE(p1);
765 	PROC_UNLOCK(p1);
766 
767 	/*
768 	 * Tell any interested parties about the new process.
769 	 */
770 	knote_fork(&p1->p_klist, p2->p_pid);
771 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
772 
773 	/*
774 	 * Preserve synchronization semantics of vfork.  If waiting for
775 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
776 	 * proc (in case of exit).
777 	 */
778 	PROC_LOCK(p2);
779 	while (p2->p_flag & P_PPWAIT)
780 		cv_wait(&p2->p_pwait, &p2->p_mtx);
781 	PROC_UNLOCK(p2);
782 
783 	/*
784 	 * Return child proc pointer to parent.
785 	 */
786 	*procp = p2;
787 	return (0);
788 fail:
789 	sx_sunlock(&proctree_lock);
790 	if (ppsratecheck(&lastfail, &curfail, 1))
791 		printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
792 		    td->td_ucred->cr_ruid);
793 	sx_xunlock(&allproc_lock);
794 #ifdef MAC
795 	mac_proc_destroy(newproc);
796 #endif
797 fail1:
798 	if (vm2 != NULL)
799 		vmspace_free(vm2);
800 	uma_zfree(proc_zone, newproc);
801 	pause("fork", hz / 2);
802 	return (error);
803 }
804 
805 /*
806  * Handle the return of a child process from fork1().  This function
807  * is called from the MD fork_trampoline() entry point.
808  */
809 void
810 fork_exit(callout, arg, frame)
811 	void (*callout)(void *, struct trapframe *);
812 	void *arg;
813 	struct trapframe *frame;
814 {
815 	struct proc *p;
816 	struct thread *td;
817 	struct thread *dtd;
818 
819 	td = curthread;
820 	p = td->td_proc;
821 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
822 
823 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
824 		td, td->td_sched, p->p_pid, td->td_name);
825 
826 	sched_fork_exit(td);
827 	/*
828 	* Processes normally resume in mi_switch() after being
829 	* cpu_switch()'ed to, but when children start up they arrive here
830 	* instead, so we must do much the same things as mi_switch() would.
831 	*/
832 	if ((dtd = PCPU_GET(deadthread))) {
833 		PCPU_SET(deadthread, NULL);
834 		thread_stash(dtd);
835 	}
836 	thread_unlock(td);
837 
838 	/*
839 	 * cpu_set_fork_handler intercepts this function call to
840 	 * have this call a non-return function to stay in kernel mode.
841 	 * initproc has its own fork handler, but it does return.
842 	 */
843 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
844 	callout(arg, frame);
845 
846 	/*
847 	 * Check if a kernel thread misbehaved and returned from its main
848 	 * function.
849 	 */
850 	if (p->p_flag & P_KTHREAD) {
851 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
852 		    td->td_name, p->p_pid);
853 		kproc_exit(0);
854 	}
855 	mtx_assert(&Giant, MA_NOTOWNED);
856 
857 	EVENTHANDLER_INVOKE(schedtail, p);
858 }
859 
860 /*
861  * Simplified back end of syscall(), used when returning from fork()
862  * directly into user mode.  Giant is not held on entry, and must not
863  * be held on return.  This function is passed in to fork_exit() as the
864  * first parameter and is called when returning to a new userland process.
865  */
866 void
867 fork_return(td, frame)
868 	struct thread *td;
869 	struct trapframe *frame;
870 {
871 
872 	userret(td, frame);
873 #ifdef KTRACE
874 	if (KTRPOINT(td, KTR_SYSRET))
875 		ktrsysret(SYS_fork, 0, 0);
876 #endif
877 	mtx_assert(&Giant, MA_NOTOWNED);
878 }
879