xref: /freebsd/sys/kern/kern_fork.c (revision a259b98fa211ed87bfee58c575de4e2de94ee0fa)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include "opt_ktrace.h"
38 #include "opt_kstack_pages.h"
39 
40 #define EXTERR_CATEGORY	EXTERR_CAT_FORK
41 #include <sys/systm.h>
42 #include <sys/acct.h>
43 #include <sys/bitstring.h>
44 #include <sys/capsicum.h>
45 #include <sys/eventhandler.h>
46 #include <sys/exterrvar.h>
47 #include <sys/fcntl.h>
48 #include <sys/filedesc.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/ktr.h>
53 #include <sys/ktrace.h>
54 #include <sys/sysctl.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/msan.h>
58 #include <sys/mutex.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/procdesc.h>
62 #include <sys/ptrace.h>
63 #include <sys/racct.h>
64 #include <sys/resourcevar.h>
65 #include <sys/sched.h>
66 #include <sys/sdt.h>
67 #include <sys/signalvar.h>
68 #include <sys/sx.h>
69 #include <sys/syscall.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/sysent.h>
72 #include <sys/sysproto.h>
73 #include <sys/vmmeter.h>
74 #include <sys/vnode.h>
75 #include <sys/unistd.h>
76 
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/pmap.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_extern.h>
84 #include <vm/uma.h>
85 
86 #ifdef KDTRACE_HOOKS
87 #include <sys/dtrace_bsd.h>
88 dtrace_fork_func_t	dtrace_fasttrap_fork;
89 #endif
90 
91 SDT_PROVIDER_DECLARE(proc);
92 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
93 
94 static bool pdfork_implicit_nowaitpid;
95 SYSCTL_BOOL(_kern, OID_AUTO, pdfork_implicit_nowaitpid, CTLFLAG_RWTUN,
96     &pdfork_implicit_nowaitpid, 0,
97     "PD_NOWAITPID is assumed to be always set");
98 
99 #ifndef _SYS_SYSPROTO_H_
100 struct fork_args {
101 	int     dummy;
102 };
103 #endif
104 
105 /* ARGSUSED */
106 int
107 sys_fork(struct thread *td, struct fork_args *uap)
108 {
109 	struct fork_req fr;
110 	int error, pid;
111 
112 	bzero(&fr, sizeof(fr));
113 	fr.fr_flags = RFFDG | RFPROC;
114 	fr.fr_pidp = &pid;
115 	error = fork1(td, &fr);
116 	if (error == 0) {
117 		td->td_retval[0] = pid;
118 		td->td_retval[1] = 0;
119 	}
120 	return (error);
121 }
122 
123 /* ARGUSED */
124 int
125 sys_pdfork(struct thread *td, struct pdfork_args *uap)
126 {
127 	struct fork_req fr;
128 	struct filecaps fcaps;
129 	int error, fd, pid;
130 
131 	bzero(&fr, sizeof(fr));
132 	fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
133 	fr.fr_pidp = &pid;
134 	fr.fr_pd_fd = &fd;
135 	fr.fr_pd_flags = uap->flags;
136 	filecaps_fill(&fcaps);
137 	if ((uap->flags & PD_PTRACE_CAP) == 0)
138 		cap_rights_clear(&fcaps.fc_rights, CAP_PTRACE);
139 	fr.fr_pd_fcaps = &fcaps;
140 	AUDIT_ARG_FFLAGS(uap->flags);
141 	/*
142 	 * It is necessary to return fd by reference because 0 is a valid file
143 	 * descriptor number, and the child needs to be able to distinguish
144 	 * itself from the parent using the return value.
145 	 */
146 	error = fork1(td, &fr);
147 	if (error == 0) {
148 		td->td_retval[0] = pid;
149 		td->td_retval[1] = 0;
150 		error = copyout(&fd, uap->fdp, sizeof(fd));
151 	}
152 	return (error);
153 }
154 
155 /* ARGSUSED */
156 int
157 sys_vfork(struct thread *td, struct vfork_args *uap)
158 {
159 	struct fork_req fr;
160 	int error, pid;
161 
162 	bzero(&fr, sizeof(fr));
163 	fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
164 	fr.fr_pidp = &pid;
165 	error = fork1(td, &fr);
166 	if (error == 0) {
167 		td->td_retval[0] = pid;
168 		td->td_retval[1] = 0;
169 	}
170 	return (error);
171 }
172 
173 int
174 sys_rfork(struct thread *td, struct rfork_args *uap)
175 {
176 	struct fork_req fr;
177 	int error, pid;
178 
179 	/* Don't allow kernel-only flags. */
180 	if ((uap->flags & RFKERNELONLY) != 0)
181 		return (EXTERROR(EINVAL, "Kernel-only flags %#jx", uap->flags));
182 	/* RFSPAWN must not appear with others */
183 	if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
184 		return (EXTERROR(EINVAL, "RFSPAWN must be the only flag %#jx",
185 		    uap->flags));
186 
187 	AUDIT_ARG_FFLAGS(uap->flags);
188 	bzero(&fr, sizeof(fr));
189 	if ((uap->flags & RFSPAWN) != 0) {
190 		fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
191 		fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
192 	} else {
193 		fr.fr_flags = uap->flags;
194 	}
195 	fr.fr_pidp = &pid;
196 	error = fork1(td, &fr);
197 	if (error == 0) {
198 		td->td_retval[0] = pid;
199 		td->td_retval[1] = 0;
200 	}
201 	return (error);
202 }
203 
204 int
205 sys_pdrfork(struct thread *td, struct pdrfork_args *uap)
206 {
207 	struct fork_req fr;
208 	struct filecaps fcaps;
209 	int error, fd, pid;
210 
211 	bzero(&fr, sizeof(fr));
212 	fd = -1;
213 
214 	AUDIT_ARG_FFLAGS(uap->pdflags);
215 	AUDIT_ARG_CMD(uap->rfflags);
216 
217 	if ((uap->rfflags & (RFSTOPPED | RFHIGHPID)) != 0)
218 		return (EXTERROR(EINVAL,
219 		    "Kernel-only flags %#jx", uap->rfflags));
220 
221 	/* RFSPAWN must not appear with others */
222 	if ((uap->rfflags & RFSPAWN) != 0) {
223 		if (uap->rfflags != RFSPAWN)
224 			return (EXTERROR(EINVAL,
225 			    "RFSPAWN must be the only flag %#jx",
226 			    uap->rfflags));
227 		fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPROCDESC;
228 		fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
229 	} else {
230 		if ((uap->rfflags & (RFPROC | RFPROCDESC)) !=
231 		    (RFPROC | RFPROCDESC)) {
232 			return (EXTERROR(EINVAL,
233 			    "RFPROC|RFPROCDESC required %#jx", uap->rfflags));
234 		}
235 		fr.fr_flags = uap->rfflags;
236 	}
237 
238 	fr.fr_pidp = &pid;
239 	fr.fr_pd_fd = &fd;
240 	fr.fr_pd_flags = uap->pdflags;
241 	filecaps_fill(&fcaps);
242 	if ((uap->pdflags & PD_PTRACE_CAP) == 0)
243 		cap_rights_clear(&fcaps.fc_rights, CAP_PTRACE);
244 	fr.fr_pd_fcaps = &fcaps;
245 	error = fork1(td, &fr);
246 	if (error == 0) {
247 		td->td_retval[0] = pid;
248 		td->td_retval[1] = 0;
249 		if ((fr.fr_flags & (RFPROC | RFPROCDESC)) ==
250 		    (RFPROC | RFPROCDESC) || uap->rfflags == RFSPAWN)
251 			error = copyout(&fd, uap->fdp, sizeof(fd));
252 	}
253 	return (error);
254 }
255 
256 int __exclusive_cache_line	nprocs = 1;		/* process 0 */
257 int	lastpid = 0;
258 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
259     "Last used PID");
260 
261 /*
262  * Random component to lastpid generation.  We mix in a random factor to make
263  * it a little harder to predict.  We sanity check the modulus value to avoid
264  * doing it in critical paths.  Don't let it be too small or we pointlessly
265  * waste randomness entropy, and don't let it be impossibly large.  Using a
266  * modulus that is too big causes a LOT more process table scans and slows
267  * down fork processing as the pidchecked caching is defeated.
268  */
269 static int randompid = 0;
270 
271 static int
272 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
273 {
274 	int error, pid;
275 
276 	error = sysctl_wire_old_buffer(req, sizeof(int));
277 	if (error != 0)
278 		return(error);
279 	sx_xlock(&allproc_lock);
280 	pid = randompid;
281 	error = sysctl_handle_int(oidp, &pid, 0, req);
282 	if (error == 0 && req->newptr != NULL) {
283 		if (pid == 0)
284 			randompid = 0;
285 		else if (pid == 1)
286 			/* generate a random PID modulus between 100 and 1123 */
287 			randompid = 100 + arc4random() % 1024;
288 		else if (pid < 0 || pid > pid_max - 100)
289 			/* out of range */
290 			randompid = pid_max - 100;
291 		else if (pid < 100)
292 			/* Make it reasonable */
293 			randompid = 100;
294 		else
295 			randompid = pid;
296 	}
297 	sx_xunlock(&allproc_lock);
298 	return (error);
299 }
300 
301 SYSCTL_PROC(_kern, OID_AUTO, randompid,
302     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
303     sysctl_kern_randompid, "I",
304     "Random PID modulus. Special values: 0: disable, 1: choose random value");
305 
306 extern bitstr_t proc_id_pidmap;
307 extern bitstr_t proc_id_grpidmap;
308 extern bitstr_t proc_id_sessidmap;
309 extern bitstr_t proc_id_reapmap;
310 
311 /*
312  * Find an unused process ID
313  *
314  * If RFHIGHPID is set (used during system boot), do not allocate
315  * low-numbered pids.
316  */
317 static int
318 fork_findpid(int flags)
319 {
320 	pid_t result;
321 	int trypid, random;
322 
323 	/*
324 	 * Avoid calling arc4random with procid_lock held.
325 	 */
326 	random = 0;
327 	if (__predict_false(randompid))
328 		random = arc4random() % randompid;
329 
330 	mtx_lock(&procid_lock);
331 
332 	trypid = lastpid + 1;
333 	if (flags & RFHIGHPID) {
334 		if (trypid < 10)
335 			trypid = 10;
336 	} else {
337 		trypid += random;
338 	}
339 retry:
340 	if (trypid >= pid_max)
341 		trypid = 2;
342 
343 	bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
344 	if (result == -1) {
345 		KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
346 		trypid = 2;
347 		goto retry;
348 	}
349 	if (bit_test(&proc_id_grpidmap, result) ||
350 	    bit_test(&proc_id_sessidmap, result) ||
351 	    bit_test(&proc_id_reapmap, result)) {
352 		trypid = result + 1;
353 		goto retry;
354 	}
355 
356 	/*
357 	 * RFHIGHPID does not mess with the lastpid counter during boot.
358 	 */
359 	if ((flags & RFHIGHPID) == 0)
360 		lastpid = result;
361 
362 	bit_set(&proc_id_pidmap, result);
363 	mtx_unlock(&procid_lock);
364 
365 	return (result);
366 }
367 
368 static int
369 fork_norfproc(struct thread *td, int flags)
370 {
371 	struct proc *p1;
372 	int error;
373 
374 	KASSERT((flags & RFPROC) == 0,
375 	    ("fork_norfproc called with RFPROC set"));
376 	p1 = td->td_proc;
377 
378 	/*
379 	 * Quiesce other threads if necessary.  If RFMEM is not specified we
380 	 * must ensure that other threads do not concurrently create a second
381 	 * process sharing the vmspace, see vmspace_unshare().
382 	 */
383 	if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
384 	    ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
385 		PROC_LOCK(p1);
386 		if (thread_single(p1, SINGLE_BOUNDARY)) {
387 			PROC_UNLOCK(p1);
388 			return (ERESTART);
389 		}
390 		PROC_UNLOCK(p1);
391 	}
392 
393 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
394 	if (error != 0)
395 		goto fail;
396 
397 	/*
398 	 * Close all file descriptors.
399 	 */
400 	if ((flags & RFCFDG) != 0) {
401 		struct filedesc *fdtmp;
402 		struct pwddesc *pdtmp;
403 
404 		pdtmp = pdinit(td->td_proc->p_pd, false);
405 		fdtmp = fdinit();
406 		pdescfree(td);
407 		fdescfree(td);
408 		p1->p_fd = fdtmp;
409 		p1->p_pd = pdtmp;
410 	}
411 
412 	/*
413 	 * Unshare file descriptors (from parent).
414 	 */
415 	if ((flags & RFFDG) != 0) {
416 		fdunshare(td);
417 		pdunshare(td);
418 	}
419 
420 fail:
421 	if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
422 	    ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
423 		PROC_LOCK(p1);
424 		thread_single_end(p1, SINGLE_BOUNDARY);
425 		PROC_UNLOCK(p1);
426 	}
427 	return (error);
428 }
429 
430 static void
431 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
432     struct vmspace *vm2, struct file *fp_procdesc)
433 {
434 	struct proc *p1, *pptr;
435 	struct filedesc *fd;
436 	struct filedesc_to_leader *fdtol;
437 	struct pwddesc *pd;
438 	struct sigacts *newsigacts;
439 
440 	p1 = td->td_proc;
441 
442 	PROC_LOCK(p1);
443 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
444 	    __rangeof(struct proc, p_startcopy, p_endcopy));
445 	pargs_hold(p2->p_args);
446 	PROC_UNLOCK(p1);
447 
448 	bzero(&p2->p_startzero,
449 	    __rangeof(struct proc, p_startzero, p_endzero));
450 
451 	/* Tell the prison that we exist. */
452 	prison_proc_hold(p2->p_ucred->cr_prison);
453 
454 	p2->p_state = PRS_NEW;		/* protect against others */
455 	p2->p_pid = fork_findpid(fr->fr_flags);
456 	AUDIT_ARG_PID(p2->p_pid);
457 	TSFORK(p2->p_pid, p1->p_pid);
458 
459 	sx_xlock(&allproc_lock);
460 	LIST_INSERT_HEAD(&allproc, p2, p_list);
461 	allproc_gen++;
462 	prison_proc_link(p2->p_ucred->cr_prison, p2);
463 	sx_xunlock(&allproc_lock);
464 
465 	sx_xlock(PIDHASHLOCK(p2->p_pid));
466 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
467 	sx_xunlock(PIDHASHLOCK(p2->p_pid));
468 
469 	tidhash_add(td2);
470 
471 	/*
472 	 * Malloc things while we don't hold any locks.
473 	 */
474 	if (fr->fr_flags & RFSIGSHARE)
475 		newsigacts = NULL;
476 	else
477 		newsigacts = sigacts_alloc();
478 
479 	/*
480 	 * Copy filedesc.
481 	 */
482 	if (fr->fr_flags & RFCFDG) {
483 		pd = pdinit(p1->p_pd, false);
484 		fd = fdinit();
485 		fdtol = NULL;
486 	} else if (fr->fr_flags & RFFDG) {
487 		if (fr->fr_flags2 & FR2_SHARE_PATHS)
488 			pd = pdshare(p1->p_pd);
489 		else
490 			pd = pdcopy(p1->p_pd);
491 		fd = fdcopy(p1->p_fd, p2);
492 		fdtol = NULL;
493 	} else {
494 		if (fr->fr_flags2 & FR2_SHARE_PATHS)
495 			pd = pdcopy(p1->p_pd);
496 		else
497 			pd = pdshare(p1->p_pd);
498 		fd = fdshare(p1->p_fd);
499 		if (p1->p_fdtol == NULL)
500 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
501 			    p1->p_leader);
502 		if ((fr->fr_flags & RFTHREAD) != 0) {
503 			/*
504 			 * Shared file descriptor table, and shared
505 			 * process leaders.
506 			 */
507 			fdtol = filedesc_to_leader_share(p1->p_fdtol, p1->p_fd);
508 		} else {
509 			/*
510 			 * Shared file descriptor table, and different
511 			 * process leaders.
512 			 */
513 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
514 			    p1->p_fd, p2);
515 		}
516 	}
517 	/*
518 	 * Make a proc table entry for the new process.
519 	 * Start by zeroing the section of proc that is zero-initialized,
520 	 * then copy the section that is copied directly from the parent.
521 	 */
522 
523 	PROC_LOCK(p2);
524 	PROC_LOCK(p1);
525 
526 	bzero(&td2->td_startzero,
527 	    __rangeof(struct thread, td_startzero, td_endzero));
528 
529 	bcopy(&td->td_startcopy, &td2->td_startcopy,
530 	    __rangeof(struct thread, td_startcopy, td_endcopy));
531 
532 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
533 	td2->td_sigstk = td->td_sigstk;
534 	td2->td_flags = TDF_INMEM;
535 	td2->td_lend_user_pri = PRI_MAX;
536 
537 #ifdef VIMAGE
538 	td2->td_vnet = NULL;
539 	td2->td_vnet_lpush = NULL;
540 #endif
541 
542 	/*
543 	 * Allow the scheduler to initialize the child.
544 	 */
545 	thread_lock(td);
546 	sched_fork(td, td2);
547 	/*
548 	 * Request AST to check for TDP_RFPPWAIT.  Do it here
549 	 * to avoid calling thread_lock() again.
550 	 */
551 	if ((fr->fr_flags & RFPPWAIT) != 0)
552 		ast_sched_locked(td, TDA_VFORK);
553 	thread_unlock(td);
554 
555 	/*
556 	 * Duplicate sub-structures as needed.
557 	 * Increase reference counts on shared objects.
558 	 */
559 	p2->p_flag = P_INMEM;
560 	p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
561 	    P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
562 	    P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
563 	    P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS |
564 	    P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC | P2_LOGSIGEXIT_CTL |
565 	    P2_LOGSIGEXIT_ENABLE);
566 	if ((fr->fr_flags & RFPROCDESC) != 0) {
567 		p2->p_zombieref = PZOMBIEREF_PROCDESC;
568 		if (((fr->fr_pd_flags & PD_NOWAITPID) == 0 &&
569 		    !pdfork_implicit_nowaitpid) &&
570 		    (fr->fr_flags & RFNOWAIT) == 0)
571 			p2->p_zombieref |= (PZOMBIEREF_PARENT |
572 			    PZOMBIEREF_NEEDPARENT);
573 	} else {
574 		p2->p_zombieref = PZOMBIEREF_PARENT | PZOMBIEREF_NEEDPARENT;
575 	}
576 	p2->p_swtick = ticks;
577 	if (p1->p_flag & P_PROFIL)
578 		startprofclock(p2);
579 
580 	if (fr->fr_flags & RFSIGSHARE) {
581 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
582 	} else {
583 		sigacts_copy(newsigacts, p1->p_sigacts);
584 		p2->p_sigacts = newsigacts;
585 		if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) {
586 			mtx_lock(&p2->p_sigacts->ps_mtx);
587 			if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
588 				sig_drop_caught(p2);
589 			if ((fr->fr_flags2 & FR2_KPROC) != 0)
590 				p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
591 			mtx_unlock(&p2->p_sigacts->ps_mtx);
592 		}
593 	}
594 
595 	if (fr->fr_flags & RFTSIGZMB)
596 	        p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
597 	else if (fr->fr_flags & RFLINUXTHPN)
598 	        p2->p_sigparent = SIGUSR1;
599 	else
600 	        p2->p_sigparent = SIGCHLD;
601 
602 	if ((fr->fr_flags2 & FR2_KPROC) != 0) {
603 		p2->p_flag |= P_SYSTEM | P_KPROC;
604 		td2->td_pflags |= TDP_KTHREAD;
605 	}
606 
607 	p2->p_textvp = p1->p_textvp;
608 	p2->p_textdvp = p1->p_textdvp;
609 	p2->p_fd = fd;
610 	p2->p_fdtol = fdtol;
611 	p2->p_pd = pd;
612 
613 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
614 		p2->p_flag |= P_PROTECTED;
615 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
616 	}
617 
618 	/*
619 	 * p_limit is copy-on-write.  Bump its refcount.
620 	 */
621 	lim_fork(p1, p2);
622 
623 	thread_cow_get_proc(td2, p2);
624 
625 	pstats_fork(p1->p_stats, p2->p_stats);
626 
627 	PROC_UNLOCK(p1);
628 	PROC_UNLOCK(p2);
629 
630 	/*
631 	 * Bump references to the text vnode and directory, and copy
632 	 * the hardlink name.
633 	 */
634 	if (p2->p_textvp != NULL)
635 		vrefact(p2->p_textvp);
636 	if (p2->p_textdvp != NULL)
637 		vrefact(p2->p_textdvp);
638 	p2->p_binname = p1->p_binname == NULL ? NULL :
639 	    strdup(p1->p_binname, M_PARGS);
640 
641 	/*
642 	 * Set up linkage for kernel based threading.
643 	 */
644 	if ((fr->fr_flags & RFTHREAD) != 0) {
645 		mtx_lock(&ppeers_lock);
646 		p2->p_peers = p1->p_peers;
647 		p1->p_peers = p2;
648 		p2->p_leader = p1->p_leader;
649 		mtx_unlock(&ppeers_lock);
650 		PROC_LOCK(p1->p_leader);
651 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
652 			PROC_UNLOCK(p1->p_leader);
653 			/*
654 			 * The task leader is exiting, so process p1 is
655 			 * going to be killed shortly.  Since p1 obviously
656 			 * isn't dead yet, we know that the leader is either
657 			 * sending SIGKILL's to all the processes in this
658 			 * task or is sleeping waiting for all the peers to
659 			 * exit.  We let p1 complete the fork, but we need
660 			 * to go ahead and kill the new process p2 since
661 			 * the task leader may not get a chance to send
662 			 * SIGKILL to it.  We leave it on the list so that
663 			 * the task leader will wait for this new process
664 			 * to commit suicide.
665 			 */
666 			PROC_LOCK(p2);
667 			kern_psignal(p2, SIGKILL);
668 			PROC_UNLOCK(p2);
669 		} else
670 			PROC_UNLOCK(p1->p_leader);
671 	} else {
672 		p2->p_peers = NULL;
673 		p2->p_leader = p2;
674 	}
675 
676 	sx_xlock(&proctree_lock);
677 	PGRP_LOCK(p1->p_pgrp);
678 	PROC_LOCK(p2);
679 	PROC_LOCK(p1);
680 
681 	/*
682 	 * Preserve some more flags in subprocess.  P_PROFIL has already
683 	 * been preserved.
684 	 */
685 	p2->p_flag |= p1->p_flag & P_SUGID;
686 	td2->td_pflags |= td->td_pflags & (TDP_ALTSTACK | TDP_SIGFASTBLOCK);
687 	td2->td_pflags2 |= td->td_pflags2 & TDP2_UEXTERR;
688 	if (p1->p_flag & P_CONTROLT) {
689 		SESS_LOCK(p1->p_session);
690 		if (p1->p_session->s_ttyvp != NULL)
691 			p2->p_flag |= P_CONTROLT;
692 		SESS_UNLOCK(p1->p_session);
693 	}
694 	if (fr->fr_flags & RFPPWAIT)
695 		p2->p_flag |= P_PPWAIT;
696 
697 	p2->p_pgrp = p1->p_pgrp;
698 	LIST_INSERT_AFTER(p1, p2, p_pglist);
699 	PGRP_UNLOCK(p1->p_pgrp);
700 	LIST_INIT(&p2->p_children);
701 	LIST_INIT(&p2->p_orphans);
702 
703 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
704 
705 	PROC_UNLOCK(p1);
706 
707 	/*
708 	 * Attach the new process to its parent.
709 	 *
710 	 * If RFNOWAIT is set, the newly created process becomes a child
711 	 * of init.  This effectively disassociates the child from the
712 	 * parent.
713 	 */
714 	if ((fr->fr_flags & RFNOWAIT) != 0) {
715 		pptr = p1->p_reaper;
716 		p2->p_reaper = pptr;
717 	} else {
718 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
719 		    p1 : p1->p_reaper;
720 		pptr = p1;
721 	}
722 	p2->p_pptr = pptr;
723 	p2->p_oppid = pptr->p_pid;
724 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
725 	LIST_INIT(&p2->p_reaplist);
726 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
727 	if (p2->p_reaper == p1 && p1 != initproc) {
728 		p2->p_reapsubtree = p2->p_pid;
729 		proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
730 	} else {
731 		/*
732 		 * Explicitly copy this field under the proctree lock, as it
733 		 * might have changed since the bulk copying of the parent's
734 		 * fields.
735 		 */
736 		p2->p_reapsubtree = p1->p_reapsubtree;
737 	}
738 	sx_xunlock(&proctree_lock);
739 
740 	/* Inform accounting that we have forked. */
741 	p2->p_acflag = AFORK;
742 	PROC_UNLOCK(p2);
743 
744 #ifdef KTRACE
745 	ktrprocfork(p1, p2);
746 #endif
747 
748 	/*
749 	 * Finish creating the child process.  It will return via a different
750 	 * execution path later.  (ie: directly into user mode)
751 	 */
752 	vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
753 
754 	if (fr->fr_flags == (RFFDG | RFPROC)) {
755 		VM_CNT_INC(v_forks);
756 		VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
757 		    p2->p_vmspace->vm_ssize);
758 	} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
759 		VM_CNT_INC(v_vforks);
760 		VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
761 		    p2->p_vmspace->vm_ssize);
762 	} else if (p1 == &proc0) {
763 		VM_CNT_INC(v_kthreads);
764 		VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
765 		    p2->p_vmspace->vm_ssize);
766 	} else {
767 		VM_CNT_INC(v_rforks);
768 		VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
769 		    p2->p_vmspace->vm_ssize);
770 	}
771 
772 	/*
773 	 * Associate the process descriptor with the process before anything
774 	 * can happen that might cause that process to need the descriptor.
775 	 * However, don't do this until after fork(2) can no longer fail.
776 	 */
777 	if (fr->fr_flags & RFPROCDESC)
778 		procdesc_new(p2, fr->fr_pd_flags);
779 
780 	/*
781 	 * Both processes are set up, now check if any loadable modules want
782 	 * to adjust anything.
783 	 */
784 	EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
785 
786 	/*
787 	 * Set the child start time and mark the process as being complete.
788 	 */
789 	PROC_LOCK(p2);
790 	PROC_LOCK(p1);
791 	microuptime(&p2->p_stats->p_start);
792 	PROC_SLOCK(p2);
793 	p2->p_state = PRS_NORMAL;
794 	PROC_SUNLOCK(p2);
795 
796 #ifdef KDTRACE_HOOKS
797 	/*
798 	 * Tell the DTrace fasttrap provider about the new process so that any
799 	 * tracepoints inherited from the parent can be removed. We have to do
800 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
801 	 * use pfind() later on.
802 	 */
803 	if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
804 		dtrace_fasttrap_fork(p1, p2);
805 #endif
806 	if (fr->fr_flags & RFPPWAIT) {
807 		td->td_pflags |= TDP_RFPPWAIT;
808 		td->td_rfppwait_p = p2;
809 		td->td_dbgflags |= TDB_VFORK;
810 	}
811 	PROC_UNLOCK(p2);
812 
813 	/*
814 	 * Tell any interested parties about the new process.
815 	 */
816 	knote_fork(p1->p_klist, p2->p_pid);
817 
818 	PROC_UNLOCK(p1);
819 	SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
820 
821 	if (fr->fr_flags & RFPROCDESC) {
822 		procdesc_finit(p2->p_procdesc, fp_procdesc);
823 		fdrop(fp_procdesc, td);
824 	}
825 
826 	/*
827 	 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
828 	 * synced with forks in progress so it is OK if we miss it
829 	 * if being set atm.
830 	 */
831 	if ((p1->p_ptevents & PTRACE_FORK) != 0) {
832 		sx_xlock(&proctree_lock);
833 		PROC_LOCK(p2);
834 
835 		/*
836 		 * p1->p_ptevents & p1->p_pptr are protected by both
837 		 * process and proctree locks for modifications,
838 		 * so owning proctree_lock allows the race-free read.
839 		 */
840 		if ((p1->p_ptevents & PTRACE_FORK) != 0) {
841 			/*
842 			 * Arrange for debugger to receive the fork event.
843 			 *
844 			 * We can report PL_FLAG_FORKED regardless of
845 			 * P_FOLLOWFORK settings, but it does not make a sense
846 			 * for runaway child.
847 			 */
848 			td->td_dbgflags |= TDB_FORK;
849 			td->td_dbg_forked = p2->p_pid;
850 			td2->td_dbgflags |= TDB_STOPATFORK;
851 			proc_set_traced(p2, true);
852 			CTR2(KTR_PTRACE,
853 			    "do_fork: attaching to new child pid %d: oppid %d",
854 			    p2->p_pid, p2->p_oppid);
855 			proc_reparent(p2, p1->p_pptr, false);
856 		}
857 		PROC_UNLOCK(p2);
858 		sx_xunlock(&proctree_lock);
859 	}
860 
861 	/*
862 	 * Activate procdesc NOTE_FORK after we attached the debugger
863 	 * to the child.  This guarantees that a debugger which does
864 	 * kevent() on the process descriptor to get notifications of
865 	 * fork events, can properly observe the child right after the
866 	 * notification fired.
867 	 */
868 	procdesc_fork(p1, p2->p_pid);
869 
870 	racct_proc_fork_done(p2);
871 
872 	if ((fr->fr_flags & RFSTOPPED) == 0) {
873 		if (fr->fr_pidp != NULL)
874 			*fr->fr_pidp = p2->p_pid;
875 		/*
876 		 * If RFSTOPPED not requested, make child runnable and
877 		 * add to run queue.
878 		 */
879 		thread_lock(td2);
880 		TD_SET_CAN_RUN(td2);
881 		sched_add(td2, SRQ_BORING);
882 	} else {
883 		*fr->fr_procp = p2;
884 	}
885 }
886 
887 static void
888 ast_vfork(struct thread *td, int tda __unused)
889 {
890 	struct proc *p, *p2;
891 
892 	MPASS(td->td_pflags & TDP_RFPPWAIT);
893 
894 	p = td->td_proc;
895 	/*
896 	 * Preserve synchronization semantics of vfork.  If
897 	 * waiting for child to exec or exit, fork set
898 	 * P_PPWAIT on child, and there we sleep on our proc
899 	 * (in case of exit).
900 	 *
901 	 * Do it after the ptracestop() above is finished, to
902 	 * not block our debugger until child execs or exits
903 	 * to finish vfork wait.
904 	 */
905 	td->td_pflags &= ~TDP_RFPPWAIT;
906 	p2 = td->td_rfppwait_p;
907 again:
908 	PROC_LOCK(p2);
909 	while (p2->p_flag & P_PPWAIT) {
910 		PROC_LOCK(p);
911 		if (thread_suspend_check_needed()) {
912 			PROC_UNLOCK(p2);
913 			thread_suspend_check(0);
914 			PROC_UNLOCK(p);
915 			goto again;
916 		} else {
917 			PROC_UNLOCK(p);
918 		}
919 		cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
920 	}
921 	PROC_UNLOCK(p2);
922 
923 	if (td->td_dbgflags & TDB_VFORK) {
924 		PROC_LOCK(p);
925 		if (p->p_ptevents & PTRACE_VFORK)
926 			ptracestop(td, SIGTRAP, NULL);
927 		td->td_dbgflags &= ~TDB_VFORK;
928 		PROC_UNLOCK(p);
929 	}
930 }
931 
932 int
933 fork1(struct thread *td, struct fork_req *fr)
934 {
935 	struct proc *p1, *newproc;
936 	struct thread *td2;
937 	struct vmspace *vm2;
938 	struct ucred *cred;
939 	struct file *fp_procdesc;
940 	struct pgrp *pg;
941 	vm_ooffset_t mem_charged;
942 	int error, nprocs_new;
943 	static int curfail;
944 	static struct timeval lastfail;
945 	int flags, pages;
946 	bool killsx_locked, singlethreaded;
947 
948 	flags = fr->fr_flags;
949 	pages = fr->fr_pages;
950 
951 	if ((flags & RFSTOPPED) != 0)
952 		MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
953 	else
954 		MPASS(fr->fr_procp == NULL);
955 
956 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
957 		return (EXTERROR(EINVAL,
958 		    "Undef or unimplemented flags %#jx", flags));
959 
960 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
961 		return (EXTERROR(EINVAL,
962 		    "Signal value requires RFTSIGZMB", flags));
963 
964 	if ((flags & (RFFDG | RFCFDG)) == (RFFDG | RFCFDG))
965 		return (EXTERROR(EINVAL, "Can not copy and clear"));
966 
967 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
968 		return (EXTERROR(EINVAL, "Invalid signal", RFTSIGNUM(flags)));
969 
970 	if ((flags & RFPROCDESC) != 0) {
971 		if ((flags & RFPROC) == 0)
972 			return (EXTERROR(EINVAL,
973 	    "Can not not create a process yet get a process descriptor"));
974 
975 		if (fr->fr_pd_fd == NULL)
976 			return (EXTERROR(EINVAL,
977 		    "Must provide a place to put a procdesc if creating one"));
978 
979 		if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
980 			return (EXTERROR(EINVAL,
981 			    "Invalid pdflags at fork %#jx", fr->fr_pd_flags));
982 	}
983 
984 	p1 = td->td_proc;
985 
986 	/*
987 	 * Here we don't create a new process, but we divorce
988 	 * certain parts of a process from itself.
989 	 */
990 	if ((flags & RFPROC) == 0) {
991 		if (fr->fr_procp != NULL)
992 			*fr->fr_procp = NULL;
993 		else if (fr->fr_pidp != NULL)
994 			*fr->fr_pidp = 0;
995 		return (fork_norfproc(td, flags));
996 	}
997 
998 	fp_procdesc = NULL;
999 	newproc = NULL;
1000 	vm2 = NULL;
1001 	killsx_locked = false;
1002 	singlethreaded = false;
1003 
1004 	/*
1005 	 * Increment the nprocs resource before allocations occur.
1006 	 * Although process entries are dynamically created, we still
1007 	 * keep a global limit on the maximum number we will
1008 	 * create. There are hard-limits as to the number of processes
1009 	 * that can run, established by the KVA and memory usage for
1010 	 * the process data.
1011 	 *
1012 	 * Don't allow a nonprivileged user to use the last ten
1013 	 * processes; don't let root exceed the limit.
1014 	 */
1015 	nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
1016 	if (nprocs_new >= maxproc - 10) {
1017 		if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
1018 		    nprocs_new >= maxproc) {
1019 			error = EAGAIN;
1020 			sx_xlock(&allproc_lock);
1021 			if (ppsratecheck(&lastfail, &curfail, 1)) {
1022 				printf("maxproc limit exceeded by uid %u "
1023 				    "(pid %d); see tuning(7) and "
1024 				    "login.conf(5)\n",
1025 				    td->td_ucred->cr_ruid, p1->p_pid);
1026 			}
1027 			sx_xunlock(&allproc_lock);
1028 			goto fail2;
1029 		}
1030 	}
1031 
1032 	/*
1033 	 * If we are possibly multi-threaded, and there is a process
1034 	 * sending a signal to our group right now, ensure that our
1035 	 * other threads cannot be chosen for the signal queueing.
1036 	 * Otherwise, this might delay signal action, and make the new
1037 	 * child escape the signaling.
1038 	 */
1039 	pg = p1->p_pgrp;
1040 	if (p1->p_numthreads > 1) {
1041 		if (sx_try_slock(&pg->pg_killsx) != 0) {
1042 			killsx_locked = true;
1043 		} else {
1044 			PROC_LOCK(p1);
1045 			if (thread_single(p1, SINGLE_BOUNDARY)) {
1046 				PROC_UNLOCK(p1);
1047 				error = ERESTART;
1048 				goto fail2;
1049 			}
1050 			PROC_UNLOCK(p1);
1051 			singlethreaded = true;
1052 		}
1053 	}
1054 
1055 	/*
1056 	 * Atomically check for signals and block processes from sending
1057 	 * a signal to our process group until the child is visible.
1058 	 */
1059 	if (!killsx_locked && sx_slock_sig(&pg->pg_killsx) != 0) {
1060 		error = ERESTART;
1061 		goto fail2;
1062 	}
1063 	if (__predict_false(p1->p_pgrp != pg || sig_intr() != 0)) {
1064 		/*
1065 		 * Either the process was moved to other process
1066 		 * group, or there is pending signal.  sx_slock_sig()
1067 		 * does not check for signals if not sleeping for the
1068 		 * lock.
1069 		 */
1070 		sx_sunlock(&pg->pg_killsx);
1071 		killsx_locked = false;
1072 		error = ERESTART;
1073 		goto fail2;
1074 	} else {
1075 		killsx_locked = true;
1076 	}
1077 
1078 	/*
1079 	 * If required, create a process descriptor in the parent first; we
1080 	 * will abandon it if something goes wrong. We don't finit() until
1081 	 * later.
1082 	 */
1083 	if (flags & RFPROCDESC) {
1084 		error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
1085 		    fr->fr_pd_flags, fr->fr_pd_fcaps);
1086 		if (error != 0)
1087 			goto fail2;
1088 		fr->fr_pd_fcaps = NULL;
1089 		AUDIT_ARG_FD(*fr->fr_pd_fd);
1090 	}
1091 
1092 	mem_charged = 0;
1093 	if (pages == 0)
1094 		pages = kstack_pages;
1095 	/* Allocate new proc. */
1096 	newproc = uma_zalloc(proc_zone, M_WAITOK);
1097 	PROC_TREE_REF(newproc);
1098 	td2 = FIRST_THREAD_IN_PROC(newproc);
1099 	if (td2 == NULL) {
1100 		td2 = thread_alloc(pages);
1101 		if (td2 == NULL) {
1102 			error = ENOMEM;
1103 			goto fail2;
1104 		}
1105 		proc_linkup(newproc, td2);
1106 	} else {
1107 		error = thread_recycle(td2, pages);
1108 		if (error != 0)
1109 			goto fail2;
1110 	}
1111 
1112 	if ((flags & RFMEM) == 0) {
1113 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
1114 		if (vm2 == NULL) {
1115 			error = ENOMEM;
1116 			goto fail2;
1117 		}
1118 		if (!swap_reserve(mem_charged)) {
1119 			/*
1120 			 * The swap reservation failed. The accounting
1121 			 * from the entries of the copied vm2 will be
1122 			 * subtracted in vmspace_free(), so force the
1123 			 * reservation there.
1124 			 */
1125 			swap_reserve_force(mem_charged);
1126 			error = ENOMEM;
1127 			goto fail2;
1128 		}
1129 	} else
1130 		vm2 = NULL;
1131 
1132 	/*
1133 	 * XXX: This is ugly; when we copy resource usage, we need to bump
1134 	 *      per-cred resource counters.
1135 	 */
1136 	newproc->p_ucred = crcowget(td->td_ucred);
1137 
1138 	/*
1139 	 * Initialize resource accounting for the child process.
1140 	 */
1141 	error = racct_proc_fork(p1, newproc);
1142 	if (error != 0) {
1143 		error = EAGAIN;
1144 		goto fail1;
1145 	}
1146 
1147 #ifdef MAC
1148 	mac_proc_init(newproc);
1149 #endif
1150 
1151 	/*
1152 	 * Increment the count of procs running with this uid. Don't allow
1153 	 * a nonprivileged user to exceed their current limit.
1154 	 */
1155 	cred = td->td_ucred;
1156 	if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1157 		if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1158 			goto fail0;
1159 		chgproccnt(cred->cr_ruidinfo, 1, 0);
1160 	}
1161 
1162 	newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1163 
1164 	do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1165 	error = 0;
1166 	goto cleanup;
1167 fail0:
1168 	error = EAGAIN;
1169 #ifdef MAC
1170 	mac_proc_destroy(newproc);
1171 #endif
1172 	racct_proc_exit(newproc);
1173 fail1:
1174 	proc_unset_cred(newproc, false);
1175 fail2:
1176 	if (vm2 != NULL)
1177 		vmspace_free(vm2);
1178 	if (newproc != NULL)
1179 		PROC_TREE_UNREF(newproc);
1180 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1181 		fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1182 		fdrop(fp_procdesc, td);
1183 	}
1184 	if (fr->fr_pd_fcaps != NULL)
1185 		filecaps_free(fr->fr_pd_fcaps);
1186 	atomic_add_int(&nprocs, -1);
1187 cleanup:
1188 	if (killsx_locked)
1189 		sx_sunlock(&pg->pg_killsx);
1190 	if (singlethreaded) {
1191 		PROC_LOCK(p1);
1192 		thread_single_end(p1, SINGLE_BOUNDARY);
1193 		PROC_UNLOCK(p1);
1194 	}
1195 	if (error != 0)
1196 		pause("fork", hz / 2);
1197 	return (error);
1198 }
1199 
1200 /*
1201  * Handle the return of a child process from fork1().  This function
1202  * is called from the MD fork_trampoline() entry point.
1203  */
1204 void
1205 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1206     struct trapframe *frame)
1207 {
1208 	struct proc *p;
1209 	struct thread *td;
1210 	struct thread *dtd;
1211 
1212 	kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED);
1213 
1214 	td = curthread;
1215 	p = td->td_proc;
1216 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1217 
1218 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1219 	    td, td_get_sched(td), p->p_pid, td->td_name);
1220 
1221 	sched_fork_exit(td);
1222 
1223 	/*
1224 	 * Processes normally resume in mi_switch() after being
1225 	 * cpu_switch()'ed to, but when children start up they arrive here
1226 	 * instead, so we must do much the same things as mi_switch() would.
1227 	 */
1228 	if ((dtd = PCPU_GET(deadthread))) {
1229 		PCPU_SET(deadthread, NULL);
1230 		thread_stash(dtd);
1231 	}
1232 	thread_unlock(td);
1233 
1234 	/*
1235 	 * cpu_fork_kthread_handler intercepts this function call to
1236 	 * have this call a non-return function to stay in kernel mode.
1237 	 * initproc has its own fork handler, but it does return.
1238 	 */
1239 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1240 	callout(arg, frame);
1241 
1242 	/*
1243 	 * Check if a kernel thread misbehaved and returned from its main
1244 	 * function.
1245 	 */
1246 	if (p->p_flag & P_KPROC) {
1247 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1248 		    td->td_name, p->p_pid);
1249 		kthread_exit();
1250 	}
1251 	mtx_assert(&Giant, MA_NOTOWNED);
1252 
1253 	/*
1254 	 * Now going to return to userland.
1255 	 */
1256 
1257 	if (p->p_sysent->sv_schedtail != NULL)
1258 		(p->p_sysent->sv_schedtail)(td);
1259 
1260 	userret(td, frame);
1261 }
1262 
1263 /*
1264  * Simplified back end of syscall(), used when returning from fork()
1265  * directly into user mode.  This function is passed in to fork_exit()
1266  * as the first parameter and is called when returning to a new
1267  * userland process.
1268  */
1269 void
1270 fork_return(struct thread *td, struct trapframe *frame)
1271 {
1272 	struct proc *p;
1273 
1274 	p = td->td_proc;
1275 	if (td->td_dbgflags & TDB_STOPATFORK) {
1276 		PROC_LOCK(p);
1277 		if ((p->p_flag & P_TRACED) != 0) {
1278 			/*
1279 			 * Inform the debugger if one is still present.
1280 			 */
1281 			td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1282 			ptracestop(td, SIGSTOP, NULL);
1283 			td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1284 		} else {
1285 			/*
1286 			 * ... otherwise clear the request.
1287 			 */
1288 			td->td_dbgflags &= ~TDB_STOPATFORK;
1289 		}
1290 		PROC_UNLOCK(p);
1291 	} else if (p->p_flag & P_TRACED) {
1292  		/*
1293 		 * This is the start of a new thread in a traced
1294 		 * process.  Report a system call exit event.
1295 		 */
1296 		PROC_LOCK(p);
1297 		td->td_dbgflags |= TDB_SCX;
1298 		if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1299 		    (td->td_dbgflags & TDB_BORN) != 0)
1300 			ptracestop(td, SIGTRAP, NULL);
1301 		td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1302 		PROC_UNLOCK(p);
1303 	}
1304 
1305 	/*
1306 	 * If the prison was killed mid-fork, die along with it.
1307 	 */
1308 	if (!prison_isalive(td->td_ucred->cr_prison))
1309 		kern_exit(td, 0, SIGKILL);
1310 
1311 #ifdef KTRACE
1312 	if (KTRPOINT(td, KTR_SYSRET))
1313 		ktrsysret(td->td_sa.code, 0, 0);
1314 #endif
1315 }
1316 
1317 static void
1318 fork_init(void *arg __unused)
1319 {
1320 	ast_register(TDA_VFORK, ASTR_ASTF_REQUIRED | ASTR_TDP, TDP_RFPPWAIT,
1321 	    ast_vfork);
1322 }
1323 SYSINIT(fork, SI_SUB_INTRINSIC, SI_ORDER_ANY, fork_init, NULL);
1324