Home | History | Annotate | Line # | Download | only in kern
kern_fork.c revision 1.209
      1 /*	$NetBSD: kern_fork.c,v 1.209 2019/04/07 14:50:41 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 1.209 2019/04/07 14:50:41 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  * General fork call.  Note that another LWP in the process may call exec()
    208  * or exit() while we are forking.  It's safe to continue here, because
    209  * neither operation will complete until all LWPs have exited the process.
    210  */
    211 int
    212 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
    213     void (*func)(void *), void *arg, register_t *retval)
    214 {
    215 	struct proc	*p1, *p2, *parent;
    216 	struct plimit   *p1_lim;
    217 	uid_t		uid;
    218 	struct lwp	*l2;
    219 	int		count;
    220 	vaddr_t		uaddr;
    221 	int		tnprocs;
    222 	int		tracefork, tracevfork, tracevforkdone;
    223 	int		error = 0;
    224 
    225 	p1 = l1->l_proc;
    226 	uid = kauth_cred_getuid(l1->l_cred);
    227 	tnprocs = atomic_inc_uint_nv(&nprocs);
    228 
    229 	/*
    230 	 * Although process entries are dynamically created, we still keep
    231 	 * a global limit on the maximum number we will create.
    232 	 */
    233 	if (__predict_false(tnprocs >= maxproc))
    234 		error = -1;
    235 	else
    236 		error = kauth_authorize_process(l1->l_cred,
    237 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
    238 
    239 	if (error) {
    240 		static struct timeval lasttfm;
    241 		atomic_dec_uint(&nprocs);
    242 		if (ratecheck(&lasttfm, &fork_tfmrate))
    243 			tablefull("proc", "increase kern.maxproc or NPROC");
    244 		if (forkfsleep)
    245 			kpause("forkmx", false, forkfsleep, NULL);
    246 		return EAGAIN;
    247 	}
    248 
    249 	/*
    250 	 * Enforce limits.
    251 	 */
    252 	count = chgproccnt(uid, 1);
    253 	if (__predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
    254 		if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
    255 		    p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
    256 		    &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0) {
    257 			(void)chgproccnt(uid, -1);
    258 			atomic_dec_uint(&nprocs);
    259 			if (forkfsleep)
    260 				kpause("forkulim", false, forkfsleep, NULL);
    261 			return EAGAIN;
    262 		}
    263 	}
    264 
    265 	/*
    266 	 * Allocate virtual address space for the U-area now, while it
    267 	 * is still easy to abort the fork operation if we're out of
    268 	 * kernel virtual address space.
    269 	 */
    270 	uaddr = uvm_uarea_alloc();
    271 	if (__predict_false(uaddr == 0)) {
    272 		(void)chgproccnt(uid, -1);
    273 		atomic_dec_uint(&nprocs);
    274 		return ENOMEM;
    275 	}
    276 
    277 	/*
    278 	 * We are now committed to the fork.  From here on, we may
    279 	 * block on resources, but resource allocation may NOT fail.
    280 	 */
    281 
    282 	/* Allocate new proc. */
    283 	p2 = proc_alloc();
    284 
    285 	/*
    286 	 * Make a proc table entry for the new process.
    287 	 * Start by zeroing the section of proc that is zero-initialized,
    288 	 * then copy the section that is copied directly from the parent.
    289 	 */
    290 	memset(&p2->p_startzero, 0,
    291 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
    292 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
    293 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
    294 
    295 	TAILQ_INIT(&p2->p_sigpend.sp_info);
    296 
    297 	LIST_INIT(&p2->p_lwps);
    298 	LIST_INIT(&p2->p_sigwaiters);
    299 
    300 	/*
    301 	 * Duplicate sub-structures as needed.
    302 	 * Increase reference counts on shared objects.
    303 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
    304 	 * handling are important in order to keep a consistent behaviour
    305 	 * for the child after the fork.  If we are a 32-bit process, the
    306 	 * child will be too.
    307 	 */
    308 	p2->p_flag =
    309 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
    310 	p2->p_emul = p1->p_emul;
    311 	p2->p_execsw = p1->p_execsw;
    312 
    313 	if (flags & FORK_SYSTEM) {
    314 		/*
    315 		 * Mark it as a system process.  Set P_NOCLDWAIT so that
    316 		 * children are reparented to init(8) when they exit.
    317 		 * init(8) can easily wait them out for us.
    318 		 */
    319 		p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT);
    320 	}
    321 
    322 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
    323 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
    324 	rw_init(&p2->p_reflock);
    325 	cv_init(&p2->p_waitcv, "wait");
    326 	cv_init(&p2->p_lwpcv, "lwpwait");
    327 
    328 	/*
    329 	 * Share a lock between the processes if they are to share signal
    330 	 * state: we must synchronize access to it.
    331 	 */
    332 	if (flags & FORK_SHARESIGS) {
    333 		p2->p_lock = p1->p_lock;
    334 		mutex_obj_hold(p1->p_lock);
    335 	} else
    336 		p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    337 
    338 	kauth_proc_fork(p1, p2);
    339 
    340 	p2->p_raslist = NULL;
    341 #if defined(__HAVE_RAS)
    342 	ras_fork(p1, p2);
    343 #endif
    344 
    345 	/* bump references to the text vnode (for procfs) */
    346 	p2->p_textvp = p1->p_textvp;
    347 	if (p2->p_textvp)
    348 		vref(p2->p_textvp);
    349 	if (p1->p_path)
    350 		p2->p_path = kmem_strdupsize(p1->p_path, NULL, KM_SLEEP);
    351 	else
    352 		p2->p_path = NULL;
    353 
    354 	if (flags & FORK_SHAREFILES)
    355 		fd_share(p2);
    356 	else if (flags & FORK_CLEANFILES)
    357 		p2->p_fd = fd_init(NULL);
    358 	else
    359 		p2->p_fd = fd_copy();
    360 
    361 	/* XXX racy */
    362 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
    363 
    364 	if (flags & FORK_SHARECWD)
    365 		cwdshare(p2);
    366 	else
    367 		p2->p_cwdi = cwdinit();
    368 
    369 	/*
    370 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
    371 	 * we just need increase pl_refcnt.
    372 	 */
    373 	p1_lim = p1->p_limit;
    374 	if (!p1_lim->pl_writeable) {
    375 		lim_addref(p1_lim);
    376 		p2->p_limit = p1_lim;
    377 	} else {
    378 		p2->p_limit = lim_copy(p1_lim);
    379 	}
    380 
    381 	if (flags & FORK_PPWAIT) {
    382 		/* Mark ourselves as waiting for a child. */
    383 		l1->l_pflag |= LP_VFORKWAIT;
    384 		p2->p_lflag = PL_PPWAIT;
    385 		p2->p_vforklwp = l1;
    386 	} else {
    387 		p2->p_lflag = 0;
    388 	}
    389 	p2->p_sflag = 0;
    390 	p2->p_slflag = 0;
    391 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
    392 	p2->p_pptr = parent;
    393 	p2->p_ppid = parent->p_pid;
    394 	LIST_INIT(&p2->p_children);
    395 
    396 	p2->p_aio = NULL;
    397 
    398 #ifdef KTRACE
    399 	/*
    400 	 * Copy traceflag and tracefile if enabled.
    401 	 * If not inherited, these were zeroed above.
    402 	 */
    403 	if (p1->p_traceflag & KTRFAC_INHERIT) {
    404 		mutex_enter(&ktrace_lock);
    405 		p2->p_traceflag = p1->p_traceflag;
    406 		if ((p2->p_tracep = p1->p_tracep) != NULL)
    407 			ktradref(p2);
    408 		mutex_exit(&ktrace_lock);
    409 	}
    410 #endif
    411 
    412 	/*
    413 	 * Create signal actions for the child process.
    414 	 */
    415 	p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS);
    416 	mutex_enter(p1->p_lock);
    417 	p2->p_sflag |=
    418 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
    419 	sched_proc_fork(p1, p2);
    420 	mutex_exit(p1->p_lock);
    421 
    422 	p2->p_stflag = p1->p_stflag;
    423 
    424 	/*
    425 	 * p_stats.
    426 	 * Copy parts of p_stats, and zero out the rest.
    427 	 */
    428 	p2->p_stats = pstatscopy(p1->p_stats);
    429 
    430 	/*
    431 	 * Set up the new process address space.
    432 	 */
    433 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false);
    434 
    435 	/*
    436 	 * Finish creating the child process.
    437 	 * It will return through a different path later.
    438 	 */
    439 	lwp_create(l1, p2, uaddr, (flags & FORK_PPWAIT) ? LWP_VFORK : 0,
    440 	    stack, stacksize, (func != NULL) ? func : child_return, arg, &l2,
    441 	    l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
    442 
    443 	/*
    444 	 * Inherit l_private from the parent.
    445 	 * Note that we cannot use lwp_setprivate() here since that
    446 	 * also sets the CPU TLS register, which is incorrect if the
    447 	 * process has changed that without letting the kernel know.
    448 	 */
    449 	l2->l_private = l1->l_private;
    450 
    451 	/*
    452 	 * If emulation has a process fork hook, call it now.
    453 	 */
    454 	if (p2->p_emul->e_proc_fork)
    455 		(*p2->p_emul->e_proc_fork)(p2, l1, flags);
    456 
    457 	/*
    458 	 * ...and finally, any other random fork hooks that subsystems
    459 	 * might have registered.
    460 	 */
    461 	doforkhooks(p2, p1);
    462 
    463 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
    464 
    465 	/*
    466 	 * It's now safe for the scheduler and other processes to see the
    467 	 * child process.
    468 	 */
    469 	mutex_enter(proc_lock);
    470 
    471 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
    472 		p2->p_lflag |= PL_CONTROLT;
    473 
    474 	LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
    475 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
    476 
    477 	/*
    478 	 * Trace fork(2) and vfork(2)-like events on demand in a debugger.
    479 	 */
    480 	tracefork = (p1->p_slflag & (PSL_TRACEFORK|PSL_TRACED)) ==
    481 	    (PSL_TRACEFORK|PSL_TRACED) && (flags & FORK_PPWAIT) == 0;
    482 	tracevfork = (p1->p_slflag & (PSL_TRACEVFORK|PSL_TRACED)) ==
    483 	    (PSL_TRACEVFORK|PSL_TRACED) && (flags & FORK_PPWAIT) != 0;
    484 	tracevforkdone = (p1->p_slflag & (PSL_TRACEVFORK_DONE|PSL_TRACED)) ==
    485 	    (PSL_TRACEVFORK_DONE|PSL_TRACED) && (flags & FORK_PPWAIT);
    486 	if (tracefork || tracevfork)
    487 		proc_changeparent(p2, p1->p_pptr);
    488 	if (tracefork) {
    489 		p1->p_fpid = p2->p_pid;
    490 		p2->p_fpid = p1->p_pid;
    491 	}
    492 	if (tracevfork) {
    493 		p1->p_vfpid = p2->p_pid;
    494 		p2->p_vfpid = p1->p_pid;
    495 	}
    496 
    497 	LIST_INSERT_AFTER(p1, p2, p_pglist);
    498 	LIST_INSERT_HEAD(&allproc, p2, p_list);
    499 
    500 	p2->p_trace_enabled = trace_is_enabled(p2);
    501 #ifdef __HAVE_SYSCALL_INTERN
    502 	(*p2->p_emul->e_syscall_intern)(p2);
    503 #endif
    504 
    505 	/*
    506 	 * Update stats now that we know the fork was successful.
    507 	 */
    508 	uvmexp.forks++;
    509 	if (flags & FORK_PPWAIT)
    510 		uvmexp.forks_ppwait++;
    511 	if (flags & FORK_SHAREVM)
    512 		uvmexp.forks_sharevm++;
    513 
    514 	if (ktrpoint(KTR_EMUL))
    515 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
    516 
    517 	/*
    518 	 * Notify any interested parties about the new process.
    519 	 */
    520 	if (!SLIST_EMPTY(&p1->p_klist)) {
    521 		mutex_exit(proc_lock);
    522 		KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
    523 		mutex_enter(proc_lock);
    524 	}
    525 
    526 	/*
    527 	 * Make child runnable, set start time, and add to run queue except
    528 	 * if the parent requested the child to start in SSTOP state.
    529 	 */
    530 	mutex_enter(p2->p_lock);
    531 
    532 	/*
    533 	 * Start profiling.
    534 	 */
    535 	if ((p2->p_stflag & PST_PROFIL) != 0) {
    536 		mutex_spin_enter(&p2->p_stmutex);
    537 		startprofclock(p2);
    538 		mutex_spin_exit(&p2->p_stmutex);
    539 	}
    540 
    541 	getmicrotime(&p2->p_stats->p_start);
    542 	p2->p_acflag = AFORK;
    543 	lwp_lock(l2);
    544 	KASSERT(p2->p_nrlwps == 1);
    545 	if (p2->p_sflag & PS_STOPFORK) {
    546 		struct schedstate_percpu *spc = &l2->l_cpu->ci_schedstate;
    547 		p2->p_nrlwps = 0;
    548 		p2->p_stat = SSTOP;
    549 		p2->p_waited = 0;
    550 		p1->p_nstopchild++;
    551 		l2->l_stat = LSSTOP;
    552 		KASSERT(l2->l_wchan == NULL);
    553 		lwp_unlock_to(l2, spc->spc_lwplock);
    554 	} else {
    555 		p2->p_nrlwps = 1;
    556 		p2->p_stat = SACTIVE;
    557 		l2->l_stat = LSRUN;
    558 		sched_enqueue(l2, false);
    559 		lwp_unlock(l2);
    560 	}
    561 
    562 	/*
    563 	 * Return child pid to parent process,
    564 	 * marking us as parent via retval[1].
    565 	 */
    566 	if (retval != NULL) {
    567 		retval[0] = p2->p_pid;
    568 		retval[1] = 0;
    569 	}
    570 
    571 	mutex_exit(p2->p_lock);
    572 
    573 	/*
    574 	 * Let the parent know that we are tracing its child.
    575 	 */
    576 	if (tracefork || tracevfork) {
    577 		mutex_enter(p1->p_lock);
    578 		p1->p_xsig = SIGTRAP;
    579 		p1->p_sigctx.ps_faked = true; // XXX
    580 		p1->p_sigctx.ps_info._signo = p1->p_xsig;
    581 		p1->p_sigctx.ps_info._code = TRAP_CHLD;
    582 		sigswitch(0, SIGTRAP, false);
    583 		// XXX ktrpoint(KTR_PSIG)
    584 		mutex_exit(p1->p_lock);
    585 		mutex_enter(proc_lock);
    586 	}
    587 
    588 	/*
    589 	 * Preserve synchronization semantics of vfork.  If waiting for
    590 	 * child to exec or exit, sleep until it clears LP_VFORKWAIT.
    591 	 */
    592 	while (p2->p_lflag & PL_PPWAIT) // XXX: p2 can go invalid
    593 		cv_wait(&p1->p_waitcv, proc_lock);
    594 
    595 	/*
    596 	 * Let the parent know that we are tracing its child.
    597 	 */
    598 	if (tracevforkdone) {
    599 		mutex_enter(p1->p_lock);
    600 		p1->p_xsig = SIGTRAP;
    601 		p1->p_sigctx.ps_faked = true; // XXX
    602 		p1->p_sigctx.ps_info._signo = p1->p_xsig;
    603 		p1->p_sigctx.ps_info._code = TRAP_CHLD;
    604 		p1->p_vfpid_done = retval[0];
    605 		sigswitch(0, SIGTRAP, false);
    606 		// XXX ktrpoint(KTR_PSIG)
    607 		mutex_exit(p1->p_lock);
    608 		// proc_lock unlocked
    609 	} else
    610 		mutex_exit(proc_lock);
    611 
    612 	return 0;
    613 }
    614 
    615 void
    616 child_return(void *arg)
    617 {
    618 	struct lwp *l = arg;
    619 	struct proc *p = l->l_proc;
    620 
    621 	if (p->p_slflag & PSL_TRACED) {
    622 		/* Paranoid check */
    623 		mutex_enter(proc_lock);
    624 		if (!(p->p_slflag & PSL_TRACED)) {
    625 			mutex_exit(proc_lock);
    626 			goto my_tracer_is_gone;
    627 		}
    628 
    629 		mutex_enter(p->p_lock);
    630 		p->p_xsig = SIGTRAP;
    631 		p->p_sigctx.ps_faked = true; // XXX
    632 		p->p_sigctx.ps_info._signo = p->p_xsig;
    633 		p->p_sigctx.ps_info._code = TRAP_CHLD;
    634 		sigswitch(0, SIGTRAP, false);
    635 		// XXX ktrpoint(KTR_PSIG)
    636 		mutex_exit(p->p_lock);
    637 	}
    638 
    639 my_tracer_is_gone:
    640 	md_child_return(l);
    641 
    642 	/*
    643 	 * Return SYS_fork for all fork types, including vfork(2) and clone(2).
    644 	 *
    645 	 * This approach simplifies the code and avoids extra locking.
    646 	 */
    647 	ktrsysret(SYS_fork, 0, 0);
    648 }
    649