xref: /netbsd-src/sys/kern/kern_fork.c (revision deb6f0161a9109e7de9b519dc8dfb9478668dcdd)
1 /*	$NetBSD: kern_fork.c,v 1.205 2018/05/01 16:37:23 kamil Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2001, 2004, 2006, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1989, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  * (c) UNIX System Laboratories, Inc.
37  * All or some portions of this file are derived from material licensed
38  * to the University of California by American Telephone and Telegraph
39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40  * the permission of UNIX System Laboratories, Inc.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)kern_fork.c	8.8 (Berkeley) 2/14/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.205 2018/05/01 16:37:23 kamil Exp $");
71 
72 #include "opt_ktrace.h"
73 #include "opt_dtrace.h"
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/filedesc.h>
78 #include <sys/kernel.h>
79 #include <sys/pool.h>
80 #include <sys/mount.h>
81 #include <sys/proc.h>
82 #include <sys/ras.h>
83 #include <sys/resourcevar.h>
84 #include <sys/vnode.h>
85 #include <sys/file.h>
86 #include <sys/acct.h>
87 #include <sys/ktrace.h>
88 #include <sys/sched.h>
89 #include <sys/signalvar.h>
90 #include <sys/kauth.h>
91 #include <sys/atomic.h>
92 #include <sys/syscallargs.h>
93 #include <sys/uidinfo.h>
94 #include <sys/sdt.h>
95 #include <sys/ptrace.h>
96 
97 #include <uvm/uvm_extern.h>
98 
99 /*
100  * DTrace SDT provider definitions
101  */
102 SDT_PROVIDER_DECLARE(proc);
103 SDT_PROBE_DEFINE3(proc, kernel, , create,
104     "struct proc *", /* new process */
105     "struct proc *", /* parent process */
106     "int" /* flags */);
107 
108 u_int	nprocs __cacheline_aligned = 1;		/* process 0 */
109 
110 /*
111  * Number of ticks to sleep if fork() would fail due to process hitting
112  * limits. Exported in miliseconds to userland via sysctl.
113  */
114 int	forkfsleep = 0;
115 
116 int
117 sys_fork(struct lwp *l, const void *v, register_t *retval)
118 {
119 
120 	return fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval);
121 }
122 
123 /*
124  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
125  * Address space is not shared, but parent is blocked until child exit.
126  */
127 int
128 sys_vfork(struct lwp *l, const void *v, register_t *retval)
129 {
130 
131 	return fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
132 	    retval);
133 }
134 
135 /*
136  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
137  * semantics.  Address space is shared, and parent is blocked until child exit.
138  */
139 int
140 sys___vfork14(struct lwp *l, const void *v, register_t *retval)
141 {
142 
143 	return fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
144 	    NULL, NULL, retval);
145 }
146 
147 /*
148  * Linux-compatible __clone(2) system call.
149  */
150 int
151 sys___clone(struct lwp *l, const struct sys___clone_args *uap,
152     register_t *retval)
153 {
154 	/* {
155 		syscallarg(int) flags;
156 		syscallarg(void *) stack;
157 	} */
158 	int flags, sig;
159 
160 	/*
161 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
162 	 */
163 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
164 		return EINVAL;
165 
166 	/*
167 	 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
168 	 */
169 	if (SCARG(uap, flags) & CLONE_SIGHAND
170 	    && (SCARG(uap, flags) & CLONE_VM) == 0)
171 		return EINVAL;
172 
173 	flags = 0;
174 
175 	if (SCARG(uap, flags) & CLONE_VM)
176 		flags |= FORK_SHAREVM;
177 	if (SCARG(uap, flags) & CLONE_FS)
178 		flags |= FORK_SHARECWD;
179 	if (SCARG(uap, flags) & CLONE_FILES)
180 		flags |= FORK_SHAREFILES;
181 	if (SCARG(uap, flags) & CLONE_SIGHAND)
182 		flags |= FORK_SHARESIGS;
183 	if (SCARG(uap, flags) & CLONE_VFORK)
184 		flags |= FORK_PPWAIT;
185 
186 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
187 	if (sig < 0 || sig >= _NSIG)
188 		return EINVAL;
189 
190 	/*
191 	 * Note that the Linux API does not provide a portable way of
192 	 * specifying the stack area; the caller must know if the stack
193 	 * grows up or down.  So, we pass a stack size of 0, so that the
194 	 * code that makes this adjustment is a noop.
195 	 */
196 	return fork1(l, flags, sig, SCARG(uap, stack), 0,
197 	    NULL, NULL, retval);
198 }
199 
200 /*
201  * Print the 'table full' message once per 10 seconds.
202  */
203 static struct timeval fork_tfmrate = { 10, 0 };
204 
205 /*
206  * General fork call.  Note that another LWP in the process may call exec()
207  * or exit() while we are forking.  It's safe to continue here, because
208  * neither operation will complete until all LWPs have exited the process.
209  */
210 int
211 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
212     void (*func)(void *), void *arg, register_t *retval)
213 {
214 	struct proc	*p1, *p2, *parent;
215 	struct plimit   *p1_lim;
216 	uid_t		uid;
217 	struct lwp	*l2;
218 	int		count;
219 	vaddr_t		uaddr;
220 	int		tnprocs;
221 	int		tracefork, tracevfork, tracevforkdone;
222 	int		error = 0;
223 
224 	p1 = l1->l_proc;
225 	uid = kauth_cred_getuid(l1->l_cred);
226 	tnprocs = atomic_inc_uint_nv(&nprocs);
227 
228 	/*
229 	 * Although process entries are dynamically created, we still keep
230 	 * a global limit on the maximum number we will create.
231 	 */
232 	if (__predict_false(tnprocs >= maxproc))
233 		error = -1;
234 	else
235 		error = kauth_authorize_process(l1->l_cred,
236 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
237 
238 	if (error) {
239 		static struct timeval lasttfm;
240 		atomic_dec_uint(&nprocs);
241 		if (ratecheck(&lasttfm, &fork_tfmrate))
242 			tablefull("proc", "increase kern.maxproc or NPROC");
243 		if (forkfsleep)
244 			kpause("forkmx", false, forkfsleep, NULL);
245 		return EAGAIN;
246 	}
247 
248 	/*
249 	 * Enforce limits.
250 	 */
251 	count = chgproccnt(uid, 1);
252 	if (__predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
253 		if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
254 		    p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
255 		    &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0) {
256 			(void)chgproccnt(uid, -1);
257 			atomic_dec_uint(&nprocs);
258 			if (forkfsleep)
259 				kpause("forkulim", false, forkfsleep, NULL);
260 			return EAGAIN;
261 		}
262 	}
263 
264 	/*
265 	 * Allocate virtual address space for the U-area now, while it
266 	 * is still easy to abort the fork operation if we're out of
267 	 * kernel virtual address space.
268 	 */
269 	uaddr = uvm_uarea_alloc();
270 	if (__predict_false(uaddr == 0)) {
271 		(void)chgproccnt(uid, -1);
272 		atomic_dec_uint(&nprocs);
273 		return ENOMEM;
274 	}
275 
276 	/*
277 	 * We are now committed to the fork.  From here on, we may
278 	 * block on resources, but resource allocation may NOT fail.
279 	 */
280 
281 	/* Allocate new proc. */
282 	p2 = proc_alloc();
283 
284 	/*
285 	 * Make a proc table entry for the new process.
286 	 * Start by zeroing the section of proc that is zero-initialized,
287 	 * then copy the section that is copied directly from the parent.
288 	 */
289 	memset(&p2->p_startzero, 0,
290 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
291 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
292 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
293 
294 	TAILQ_INIT(&p2->p_sigpend.sp_info);
295 
296 	LIST_INIT(&p2->p_lwps);
297 	LIST_INIT(&p2->p_sigwaiters);
298 
299 	/*
300 	 * Duplicate sub-structures as needed.
301 	 * Increase reference counts on shared objects.
302 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
303 	 * handling are important in order to keep a consistent behaviour
304 	 * for the child after the fork.  If we are a 32-bit process, the
305 	 * child will be too.
306 	 */
307 	p2->p_flag =
308 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
309 	p2->p_emul = p1->p_emul;
310 	p2->p_execsw = p1->p_execsw;
311 
312 	if (flags & FORK_SYSTEM) {
313 		/*
314 		 * Mark it as a system process.  Set P_NOCLDWAIT so that
315 		 * children are reparented to init(8) when they exit.
316 		 * init(8) can easily wait them out for us.
317 		 */
318 		p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT);
319 	}
320 
321 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
322 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
323 	rw_init(&p2->p_reflock);
324 	cv_init(&p2->p_waitcv, "wait");
325 	cv_init(&p2->p_lwpcv, "lwpwait");
326 
327 	/*
328 	 * Share a lock between the processes if they are to share signal
329 	 * state: we must synchronize access to it.
330 	 */
331 	if (flags & FORK_SHARESIGS) {
332 		p2->p_lock = p1->p_lock;
333 		mutex_obj_hold(p1->p_lock);
334 	} else
335 		p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
336 
337 	kauth_proc_fork(p1, p2);
338 
339 	p2->p_raslist = NULL;
340 #if defined(__HAVE_RAS)
341 	ras_fork(p1, p2);
342 #endif
343 
344 	/* bump references to the text vnode (for procfs) */
345 	p2->p_textvp = p1->p_textvp;
346 	if (p2->p_textvp)
347 		vref(p2->p_textvp);
348 	if (p1->p_path)
349 		p2->p_path = kmem_strdupsize(p1->p_path, NULL, KM_SLEEP);
350 	else
351 		p2->p_path = NULL;
352 
353 	if (flags & FORK_SHAREFILES)
354 		fd_share(p2);
355 	else if (flags & FORK_CLEANFILES)
356 		p2->p_fd = fd_init(NULL);
357 	else
358 		p2->p_fd = fd_copy();
359 
360 	/* XXX racy */
361 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
362 
363 	if (flags & FORK_SHARECWD)
364 		cwdshare(p2);
365 	else
366 		p2->p_cwdi = cwdinit();
367 
368 	/*
369 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
370 	 * we just need increase pl_refcnt.
371 	 */
372 	p1_lim = p1->p_limit;
373 	if (!p1_lim->pl_writeable) {
374 		lim_addref(p1_lim);
375 		p2->p_limit = p1_lim;
376 	} else {
377 		p2->p_limit = lim_copy(p1_lim);
378 	}
379 
380 	if (flags & FORK_PPWAIT) {
381 		/* Mark ourselves as waiting for a child. */
382 		l1->l_pflag |= LP_VFORKWAIT;
383 		p2->p_lflag = PL_PPWAIT;
384 		p2->p_vforklwp = l1;
385 	} else {
386 		p2->p_lflag = 0;
387 	}
388 	p2->p_sflag = 0;
389 	p2->p_slflag = 0;
390 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
391 	p2->p_pptr = parent;
392 	p2->p_ppid = parent->p_pid;
393 	LIST_INIT(&p2->p_children);
394 
395 	p2->p_aio = NULL;
396 
397 #ifdef KTRACE
398 	/*
399 	 * Copy traceflag and tracefile if enabled.
400 	 * If not inherited, these were zeroed above.
401 	 */
402 	if (p1->p_traceflag & KTRFAC_INHERIT) {
403 		mutex_enter(&ktrace_lock);
404 		p2->p_traceflag = p1->p_traceflag;
405 		if ((p2->p_tracep = p1->p_tracep) != NULL)
406 			ktradref(p2);
407 		mutex_exit(&ktrace_lock);
408 	}
409 #endif
410 
411 	/*
412 	 * Create signal actions for the child process.
413 	 */
414 	p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS);
415 	mutex_enter(p1->p_lock);
416 	p2->p_sflag |=
417 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
418 	sched_proc_fork(p1, p2);
419 	mutex_exit(p1->p_lock);
420 
421 	p2->p_stflag = p1->p_stflag;
422 
423 	/*
424 	 * p_stats.
425 	 * Copy parts of p_stats, and zero out the rest.
426 	 */
427 	p2->p_stats = pstatscopy(p1->p_stats);
428 
429 	/*
430 	 * Set up the new process address space.
431 	 */
432 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false);
433 
434 	/*
435 	 * Finish creating the child process.
436 	 * It will return through a different path later.
437 	 */
438 	lwp_create(l1, p2, uaddr, (flags & FORK_PPWAIT) ? LWP_VFORK : 0,
439 	    stack, stacksize, (func != NULL) ? func : child_return, arg, &l2,
440 	    l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
441 
442 	/*
443 	 * Inherit l_private from the parent.
444 	 * Note that we cannot use lwp_setprivate() here since that
445 	 * also sets the CPU TLS register, which is incorrect if the
446 	 * process has changed that without letting the kernel know.
447 	 */
448 	l2->l_private = l1->l_private;
449 
450 	/*
451 	 * If emulation has a process fork hook, call it now.
452 	 */
453 	if (p2->p_emul->e_proc_fork)
454 		(*p2->p_emul->e_proc_fork)(p2, l1, flags);
455 
456 	/*
457 	 * ...and finally, any other random fork hooks that subsystems
458 	 * might have registered.
459 	 */
460 	doforkhooks(p2, p1);
461 
462 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
463 
464 	/*
465 	 * It's now safe for the scheduler and other processes to see the
466 	 * child process.
467 	 */
468 	mutex_enter(proc_lock);
469 
470 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
471 		p2->p_lflag |= PL_CONTROLT;
472 
473 	LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
474 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
475 
476 	/*
477 	 * Trace fork(2) and vfork(2)-like events on demand in a debugger.
478 	 */
479 	tracefork = (p1->p_slflag & (PSL_TRACEFORK|PSL_TRACED)) ==
480 	    (PSL_TRACEFORK|PSL_TRACED) && (flags && FORK_PPWAIT) == 0;
481 	tracevfork = (p1->p_slflag & (PSL_TRACEVFORK|PSL_TRACED)) ==
482 	    (PSL_TRACEVFORK|PSL_TRACED) && (flags && FORK_PPWAIT) != 0;
483 	tracevforkdone = (p1->p_slflag & (PSL_TRACEVFORK_DONE|PSL_TRACED)) ==
484 	    (PSL_TRACEVFORK_DONE|PSL_TRACED) && (flags && FORK_PPWAIT);
485 	if (tracefork || tracevfork)
486 		proc_changeparent(p2, p1->p_pptr);
487 	if (tracefork) {
488 		p1->p_fpid = p2->p_pid;
489 		p2->p_fpid = p1->p_pid;
490 	}
491 	if (tracevfork) {
492 		p1->p_vfpid = p2->p_pid;
493 		p2->p_vfpid = p1->p_pid;
494 	}
495 
496 	LIST_INSERT_AFTER(p1, p2, p_pglist);
497 	LIST_INSERT_HEAD(&allproc, p2, p_list);
498 
499 	p2->p_trace_enabled = trace_is_enabled(p2);
500 #ifdef __HAVE_SYSCALL_INTERN
501 	(*p2->p_emul->e_syscall_intern)(p2);
502 #endif
503 
504 	/* if we are being traced, give the owner a chance to interfere */
505 	if (p2->p_slflag & PSL_TRACED) {
506 		ksiginfo_t ksi;
507 
508 		KSI_INIT_EMPTY(&ksi);
509 		ksi.ksi_signo = SIGTRAP;
510 		ksi.ksi_code = TRAP_CHLD;
511 		ksi.ksi_lid = l2->l_lid;
512 		kpsignal(p2, &ksi, NULL);
513 	}
514 
515 	/*
516 	 * Update stats now that we know the fork was successful.
517 	 */
518 	uvmexp.forks++;
519 	if (flags & FORK_PPWAIT)
520 		uvmexp.forks_ppwait++;
521 	if (flags & FORK_SHAREVM)
522 		uvmexp.forks_sharevm++;
523 
524 	if (ktrpoint(KTR_EMUL))
525 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
526 
527 	/*
528 	 * Notify any interested parties about the new process.
529 	 */
530 	if (!SLIST_EMPTY(&p1->p_klist)) {
531 		mutex_exit(proc_lock);
532 		KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
533 		mutex_enter(proc_lock);
534 	}
535 
536 	/*
537 	 * Make child runnable, set start time, and add to run queue except
538 	 * if the parent requested the child to start in SSTOP state.
539 	 */
540 	mutex_enter(p2->p_lock);
541 
542 	/*
543 	 * Start profiling.
544 	 */
545 	if ((p2->p_stflag & PST_PROFIL) != 0) {
546 		mutex_spin_enter(&p2->p_stmutex);
547 		startprofclock(p2);
548 		mutex_spin_exit(&p2->p_stmutex);
549 	}
550 
551 	getmicrotime(&p2->p_stats->p_start);
552 	p2->p_acflag = AFORK;
553 	lwp_lock(l2);
554 	KASSERT(p2->p_nrlwps == 1);
555 	if (p2->p_sflag & PS_STOPFORK) {
556 		struct schedstate_percpu *spc = &l2->l_cpu->ci_schedstate;
557 		p2->p_nrlwps = 0;
558 		p2->p_stat = SSTOP;
559 		p2->p_waited = 0;
560 		p1->p_nstopchild++;
561 		l2->l_stat = LSSTOP;
562 		KASSERT(l2->l_wchan == NULL);
563 		lwp_unlock_to(l2, spc->spc_lwplock);
564 	} else {
565 		p2->p_nrlwps = 1;
566 		p2->p_stat = SACTIVE;
567 		l2->l_stat = LSRUN;
568 		sched_enqueue(l2, false);
569 		lwp_unlock(l2);
570 	}
571 
572 	/*
573 	 * Return child pid to parent process,
574 	 * marking us as parent via retval[1].
575 	 */
576 	if (retval != NULL) {
577 		retval[0] = p2->p_pid;
578 		retval[1] = 0;
579 	}
580 
581 	mutex_exit(p2->p_lock);
582 
583 	/*
584 	 * Let the parent know that we are tracing its child.
585 	 */
586 	if (tracefork || tracevfork) {
587 		mutex_enter(p1->p_lock);
588 		p1->p_xsig = SIGTRAP;
589 		p1->p_sigctx.ps_faked = true; // XXX
590 		p1->p_sigctx.ps_info._signo = p1->p_xsig;
591 		p1->p_sigctx.ps_info._code = TRAP_CHLD;
592 		sigswitch(0, SIGTRAP, false);
593 		// XXX ktrpoint(KTR_PSIG)
594 		mutex_exit(p1->p_lock);
595 		mutex_enter(proc_lock);
596 	}
597 
598 	/*
599 	 * Preserve synchronization semantics of vfork.  If waiting for
600 	 * child to exec or exit, sleep until it clears LP_VFORKWAIT.
601 	 */
602 	while (p2->p_lflag & PL_PPWAIT) // XXX: p2 can go invalid
603 		cv_wait(&p1->p_waitcv, proc_lock);
604 
605 	/*
606 	 * Let the parent know that we are tracing its child.
607 	 */
608 	if (tracevforkdone) {
609 		ksiginfo_t ksi;
610 
611 		KSI_INIT_EMPTY(&ksi);
612 		ksi.ksi_signo = SIGTRAP;
613 		ksi.ksi_code = TRAP_CHLD;
614 		ksi.ksi_lid = l1->l_lid;
615 		kpsignal(p1, &ksi, NULL);
616 
617 		p1->p_vfpid_done = retval[0];
618 	}
619 	mutex_exit(proc_lock);
620 
621 	return 0;
622 }
623