xref: /netbsd-src/sys/kern/kern_fork.c (revision 76c7fc5f6b13ed0b1508e6b313e88e59977ed78e)
1 /*	$NetBSD: kern_fork.c,v 1.215 2019/10/12 10:55:23 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 1.215 2019/10/12 10:55:23 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/syscall.h>
91 #include <sys/kauth.h>
92 #include <sys/atomic.h>
93 #include <sys/syscallargs.h>
94 #include <sys/uidinfo.h>
95 #include <sys/sdt.h>
96 #include <sys/ptrace.h>
97 
98 #include <uvm/uvm_extern.h>
99 
100 /*
101  * DTrace SDT provider definitions
102  */
103 SDT_PROVIDER_DECLARE(proc);
104 SDT_PROBE_DEFINE3(proc, kernel, , create,
105     "struct proc *", /* new process */
106     "struct proc *", /* parent process */
107     "int" /* flags */);
108 
109 u_int	nprocs __cacheline_aligned = 1;		/* process 0 */
110 
111 /*
112  * Number of ticks to sleep if fork() would fail due to process hitting
113  * limits. Exported in miliseconds to userland via sysctl.
114  */
115 int	forkfsleep = 0;
116 
117 int
118 sys_fork(struct lwp *l, const void *v, register_t *retval)
119 {
120 
121 	return fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval);
122 }
123 
124 /*
125  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
126  * Address space is not shared, but parent is blocked until child exit.
127  */
128 int
129 sys_vfork(struct lwp *l, const void *v, register_t *retval)
130 {
131 
132 	return fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
133 	    retval);
134 }
135 
136 /*
137  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
138  * semantics.  Address space is shared, and parent is blocked until child exit.
139  */
140 int
141 sys___vfork14(struct lwp *l, const void *v, register_t *retval)
142 {
143 
144 	return fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
145 	    NULL, NULL, retval);
146 }
147 
148 /*
149  * Linux-compatible __clone(2) system call.
150  */
151 int
152 sys___clone(struct lwp *l, const struct sys___clone_args *uap,
153     register_t *retval)
154 {
155 	/* {
156 		syscallarg(int) flags;
157 		syscallarg(void *) stack;
158 	} */
159 	int flags, sig;
160 
161 	/*
162 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
163 	 */
164 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
165 		return EINVAL;
166 
167 	/*
168 	 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
169 	 */
170 	if (SCARG(uap, flags) & CLONE_SIGHAND
171 	    && (SCARG(uap, flags) & CLONE_VM) == 0)
172 		return EINVAL;
173 
174 	flags = 0;
175 
176 	if (SCARG(uap, flags) & CLONE_VM)
177 		flags |= FORK_SHAREVM;
178 	if (SCARG(uap, flags) & CLONE_FS)
179 		flags |= FORK_SHARECWD;
180 	if (SCARG(uap, flags) & CLONE_FILES)
181 		flags |= FORK_SHAREFILES;
182 	if (SCARG(uap, flags) & CLONE_SIGHAND)
183 		flags |= FORK_SHARESIGS;
184 	if (SCARG(uap, flags) & CLONE_VFORK)
185 		flags |= FORK_PPWAIT;
186 
187 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
188 	if (sig < 0 || sig >= _NSIG)
189 		return EINVAL;
190 
191 	/*
192 	 * Note that the Linux API does not provide a portable way of
193 	 * specifying the stack area; the caller must know if the stack
194 	 * grows up or down.  So, we pass a stack size of 0, so that the
195 	 * code that makes this adjustment is a noop.
196 	 */
197 	return fork1(l, flags, sig, SCARG(uap, stack), 0,
198 	    NULL, NULL, retval);
199 }
200 
201 /*
202  * Print the 'table full' message once per 10 seconds.
203  */
204 static struct timeval fork_tfmrate = { 10, 0 };
205 
206 /*
207  * Check if a process is traced and shall inform about FORK events.
208  */
209 static inline bool
210 tracefork(struct proc *p, int flags)
211 {
212 
213 	return (p->p_slflag & (PSL_TRACEFORK|PSL_TRACED)) ==
214 	    (PSL_TRACEFORK|PSL_TRACED) && (flags & FORK_PPWAIT) == 0;
215 }
216 
217 /*
218  * Check if a process is traced and shall inform about VFORK events.
219  */
220 static inline bool
221 tracevfork(struct proc *p, int flags)
222 {
223 
224 	return (p->p_slflag & (PSL_TRACEVFORK|PSL_TRACED)) ==
225 	    (PSL_TRACEVFORK|PSL_TRACED) && (flags & FORK_PPWAIT) != 0;
226 }
227 
228 /*
229  * Check if a process is traced and shall inform about VFORK_DONE events.
230  */
231 static inline bool
232 tracevforkdone(struct proc *p, int flags)
233 {
234 
235 	return (p->p_slflag & (PSL_TRACEVFORK_DONE|PSL_TRACED)) ==
236 	    (PSL_TRACEVFORK_DONE|PSL_TRACED) && (flags & FORK_PPWAIT);
237 }
238 
239 /*
240  * General fork call.  Note that another LWP in the process may call exec()
241  * or exit() while we are forking.  It's safe to continue here, because
242  * neither operation will complete until all LWPs have exited the process.
243  */
244 int
245 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
246     void (*func)(void *), void *arg, register_t *retval)
247 {
248 	struct proc	*p1, *p2, *parent;
249 	struct plimit   *p1_lim;
250 	uid_t		uid;
251 	struct lwp	*l2;
252 	int		count;
253 	vaddr_t		uaddr;
254 	int		tnprocs;
255 	int		error = 0;
256 
257 	p1 = l1->l_proc;
258 	uid = kauth_cred_getuid(l1->l_cred);
259 	tnprocs = atomic_inc_uint_nv(&nprocs);
260 
261 	/*
262 	 * Although process entries are dynamically created, we still keep
263 	 * a global limit on the maximum number we will create.
264 	 */
265 	if (__predict_false(tnprocs >= maxproc))
266 		error = -1;
267 	else
268 		error = kauth_authorize_process(l1->l_cred,
269 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
270 
271 	if (error) {
272 		static struct timeval lasttfm;
273 		atomic_dec_uint(&nprocs);
274 		if (ratecheck(&lasttfm, &fork_tfmrate))
275 			tablefull("proc", "increase kern.maxproc or NPROC");
276 		if (forkfsleep)
277 			kpause("forkmx", false, forkfsleep, NULL);
278 		return EAGAIN;
279 	}
280 
281 	/*
282 	 * Enforce limits.
283 	 */
284 	count = chgproccnt(uid, 1);
285 	if (__predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
286 		if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
287 		    p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
288 		    &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0) {
289 			(void)chgproccnt(uid, -1);
290 			atomic_dec_uint(&nprocs);
291 			if (forkfsleep)
292 				kpause("forkulim", false, forkfsleep, NULL);
293 			return EAGAIN;
294 		}
295 	}
296 
297 	/*
298 	 * Allocate virtual address space for the U-area now, while it
299 	 * is still easy to abort the fork operation if we're out of
300 	 * kernel virtual address space.
301 	 */
302 	uaddr = uvm_uarea_alloc();
303 	if (__predict_false(uaddr == 0)) {
304 		(void)chgproccnt(uid, -1);
305 		atomic_dec_uint(&nprocs);
306 		return ENOMEM;
307 	}
308 
309 	/*
310 	 * We are now committed to the fork.  From here on, we may
311 	 * block on resources, but resource allocation may NOT fail.
312 	 */
313 
314 	/* Allocate new proc. */
315 	p2 = proc_alloc();
316 
317 	/*
318 	 * Make a proc table entry for the new process.
319 	 * Start by zeroing the section of proc that is zero-initialized,
320 	 * then copy the section that is copied directly from the parent.
321 	 */
322 	memset(&p2->p_startzero, 0,
323 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
324 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
325 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
326 
327 	TAILQ_INIT(&p2->p_sigpend.sp_info);
328 
329 	LIST_INIT(&p2->p_lwps);
330 	LIST_INIT(&p2->p_sigwaiters);
331 
332 	/*
333 	 * Duplicate sub-structures as needed.
334 	 * Increase reference counts on shared objects.
335 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
336 	 * handling are important in order to keep a consistent behaviour
337 	 * for the child after the fork.  If we are a 32-bit process, the
338 	 * child will be too.
339 	 */
340 	p2->p_flag =
341 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
342 	p2->p_emul = p1->p_emul;
343 	p2->p_execsw = p1->p_execsw;
344 
345 	if (flags & FORK_SYSTEM) {
346 		/*
347 		 * Mark it as a system process.  Set P_NOCLDWAIT so that
348 		 * children are reparented to init(8) when they exit.
349 		 * init(8) can easily wait them out for us.
350 		 */
351 		p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT);
352 	}
353 
354 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
355 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
356 	rw_init(&p2->p_reflock);
357 	cv_init(&p2->p_waitcv, "wait");
358 	cv_init(&p2->p_lwpcv, "lwpwait");
359 
360 	/*
361 	 * Share a lock between the processes if they are to share signal
362 	 * state: we must synchronize access to it.
363 	 */
364 	if (flags & FORK_SHARESIGS) {
365 		p2->p_lock = p1->p_lock;
366 		mutex_obj_hold(p1->p_lock);
367 	} else
368 		p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
369 
370 	kauth_proc_fork(p1, p2);
371 
372 	p2->p_raslist = NULL;
373 #if defined(__HAVE_RAS)
374 	ras_fork(p1, p2);
375 #endif
376 
377 	/* bump references to the text vnode (for procfs) */
378 	p2->p_textvp = p1->p_textvp;
379 	if (p2->p_textvp)
380 		vref(p2->p_textvp);
381 	if (p1->p_path)
382 		p2->p_path = kmem_strdupsize(p1->p_path, NULL, KM_SLEEP);
383 	else
384 		p2->p_path = NULL;
385 
386 	if (flags & FORK_SHAREFILES)
387 		fd_share(p2);
388 	else if (flags & FORK_CLEANFILES)
389 		p2->p_fd = fd_init(NULL);
390 	else
391 		p2->p_fd = fd_copy();
392 
393 	/* XXX racy */
394 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
395 
396 	if (flags & FORK_SHARECWD)
397 		cwdshare(p2);
398 	else
399 		p2->p_cwdi = cwdinit();
400 
401 	/*
402 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
403 	 * we just need increase pl_refcnt.
404 	 */
405 	p1_lim = p1->p_limit;
406 	if (!p1_lim->pl_writeable) {
407 		lim_addref(p1_lim);
408 		p2->p_limit = p1_lim;
409 	} else {
410 		p2->p_limit = lim_copy(p1_lim);
411 	}
412 
413 	if (flags & FORK_PPWAIT) {
414 		/* Mark ourselves as waiting for a child. */
415 		p2->p_lflag = PL_PPWAIT;
416 		l1->l_vforkwaiting = true;
417 		p2->p_vforklwp = l1;
418 	} else {
419 		p2->p_lflag = 0;
420 		l1->l_vforkwaiting = false;
421 	}
422 	p2->p_sflag = 0;
423 	p2->p_slflag = 0;
424 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
425 	p2->p_pptr = parent;
426 	p2->p_ppid = parent->p_pid;
427 	LIST_INIT(&p2->p_children);
428 
429 	p2->p_aio = NULL;
430 
431 #ifdef KTRACE
432 	/*
433 	 * Copy traceflag and tracefile if enabled.
434 	 * If not inherited, these were zeroed above.
435 	 */
436 	if (p1->p_traceflag & KTRFAC_INHERIT) {
437 		mutex_enter(&ktrace_lock);
438 		p2->p_traceflag = p1->p_traceflag;
439 		if ((p2->p_tracep = p1->p_tracep) != NULL)
440 			ktradref(p2);
441 		mutex_exit(&ktrace_lock);
442 	}
443 #endif
444 
445 	/*
446 	 * Create signal actions for the child process.
447 	 */
448 	p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS);
449 	mutex_enter(p1->p_lock);
450 	p2->p_sflag |=
451 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
452 	sched_proc_fork(p1, p2);
453 	mutex_exit(p1->p_lock);
454 
455 	p2->p_stflag = p1->p_stflag;
456 
457 	/*
458 	 * p_stats.
459 	 * Copy parts of p_stats, and zero out the rest.
460 	 */
461 	p2->p_stats = pstatscopy(p1->p_stats);
462 
463 	/*
464 	 * Set up the new process address space.
465 	 */
466 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false);
467 
468 	/*
469 	 * Finish creating the child process.
470 	 * It will return through a different path later.
471 	 */
472 	lwp_create(l1, p2, uaddr, (flags & FORK_PPWAIT) ? LWP_VFORK : 0,
473 	    stack, stacksize, (func != NULL) ? func : child_return, arg, &l2,
474 	    l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
475 
476 	/*
477 	 * Inherit l_private from the parent.
478 	 * Note that we cannot use lwp_setprivate() here since that
479 	 * also sets the CPU TLS register, which is incorrect if the
480 	 * process has changed that without letting the kernel know.
481 	 */
482 	l2->l_private = l1->l_private;
483 
484 	/*
485 	 * If emulation has a process fork hook, call it now.
486 	 */
487 	if (p2->p_emul->e_proc_fork)
488 		(*p2->p_emul->e_proc_fork)(p2, l1, flags);
489 
490 	/*
491 	 * ...and finally, any other random fork hooks that subsystems
492 	 * might have registered.
493 	 */
494 	doforkhooks(p2, p1);
495 
496 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
497 
498 	/*
499 	 * It's now safe for the scheduler and other processes to see the
500 	 * child process.
501 	 */
502 	mutex_enter(proc_lock);
503 
504 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
505 		p2->p_lflag |= PL_CONTROLT;
506 
507 	LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
508 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
509 
510 	/*
511 	 * Trace fork(2) and vfork(2)-like events on demand in a debugger.
512 	 */
513 	if (tracefork(p1, flags) || tracevfork(p1, flags))
514 		proc_changeparent(p2, p1->p_pptr);
515 
516 	LIST_INSERT_AFTER(p1, p2, p_pglist);
517 	LIST_INSERT_HEAD(&allproc, p2, p_list);
518 
519 	p2->p_trace_enabled = trace_is_enabled(p2);
520 #ifdef __HAVE_SYSCALL_INTERN
521 	(*p2->p_emul->e_syscall_intern)(p2);
522 #endif
523 
524 	/*
525 	 * Update stats now that we know the fork was successful.
526 	 */
527 	uvmexp.forks++;
528 	if (flags & FORK_PPWAIT)
529 		uvmexp.forks_ppwait++;
530 	if (flags & FORK_SHAREVM)
531 		uvmexp.forks_sharevm++;
532 
533 	if (ktrpoint(KTR_EMUL))
534 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
535 
536 	/*
537 	 * Notify any interested parties about the new process.
538 	 */
539 	if (!SLIST_EMPTY(&p1->p_klist)) {
540 		mutex_exit(proc_lock);
541 		KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
542 		mutex_enter(proc_lock);
543 	}
544 
545 	/*
546 	 * Make child runnable, set start time, and add to run queue except
547 	 * if the parent requested the child to start in SSTOP state.
548 	 */
549 	mutex_enter(p2->p_lock);
550 
551 	/*
552 	 * Start profiling.
553 	 */
554 	if ((p2->p_stflag & PST_PROFIL) != 0) {
555 		mutex_spin_enter(&p2->p_stmutex);
556 		startprofclock(p2);
557 		mutex_spin_exit(&p2->p_stmutex);
558 	}
559 
560 	getmicrotime(&p2->p_stats->p_start);
561 	p2->p_acflag = AFORK;
562 	lwp_lock(l2);
563 	KASSERT(p2->p_nrlwps == 1);
564 	if (p2->p_sflag & PS_STOPFORK) {
565 		struct schedstate_percpu *spc = &l2->l_cpu->ci_schedstate;
566 		p2->p_nrlwps = 0;
567 		p2->p_stat = SSTOP;
568 		p2->p_waited = 0;
569 		p1->p_nstopchild++;
570 		l2->l_stat = LSSTOP;
571 		KASSERT(l2->l_wchan == NULL);
572 		lwp_unlock_to(l2, spc->spc_lwplock);
573 	} else {
574 		p2->p_nrlwps = 1;
575 		p2->p_stat = SACTIVE;
576 		l2->l_stat = LSRUN;
577 		sched_enqueue(l2, false);
578 		lwp_unlock(l2);
579 	}
580 
581 	/*
582 	 * Return child pid to parent process,
583 	 * marking us as parent via retval[1].
584 	 */
585 	if (retval != NULL) {
586 		retval[0] = p2->p_pid;
587 		retval[1] = 0;
588 	}
589 
590 	mutex_exit(p2->p_lock);
591 
592 	/*
593 	 * Let the parent know that we are tracing its child.
594 	 */
595 	if (tracefork(p1, flags) || tracevfork(p1, flags)) {
596 		mutex_enter(p1->p_lock);
597 		eventswitch(TRAP_CHLD,
598 		    tracefork(p1, flags) ? PTRACE_FORK : PTRACE_VFORK,
599 		    retval[0]);
600 		mutex_enter(proc_lock);
601 	}
602 
603 	/*
604 	 * Preserve synchronization semantics of vfork.  If waiting for
605 	 * child to exec or exit, sleep until it clears p_vforkwaiting.
606 	 */
607 	while (l1->l_vforkwaiting)
608 		cv_wait(&l1->l_waitcv, proc_lock);
609 
610 	/*
611 	 * Let the parent know that we are tracing its child.
612 	 */
613 	if (tracevforkdone(p1, flags)) {
614 		mutex_enter(p1->p_lock);
615 		eventswitch(TRAP_CHLD, PTRACE_VFORK_DONE, retval[0]);
616 	} else
617 		mutex_exit(proc_lock);
618 
619 	return 0;
620 }
621 
622 /*
623  * MI code executed in each newly spawned process before returning to userland.
624  */
625 void
626 child_return(void *arg)
627 {
628 	struct lwp *l = arg;
629 	struct proc *p = l->l_proc;
630 
631 	if (p->p_slflag & PSL_TRACED) {
632 		/* Paranoid check */
633 		mutex_enter(proc_lock);
634 		if (!(p->p_slflag & PSL_TRACED)) {
635 			mutex_exit(proc_lock);
636 			goto my_tracer_is_gone;
637 		}
638 		mutex_enter(p->p_lock);
639 		eventswitch(TRAP_CHLD,
640 		    ISSET(p->p_lflag, PL_PPWAIT) ? PTRACE_VFORK : PTRACE_FORK,
641 		    p->p_opptr->p_pid);
642 	}
643 
644 my_tracer_is_gone:
645 	md_child_return(l);
646 
647 	/*
648 	 * Return SYS_fork for all fork types, including vfork(2) and clone(2).
649 	 *
650 	 * This approach simplifies the code and avoids extra locking.
651 	 */
652 	ktrsysret(SYS_fork, 0, 0);
653 }
654