xref: /freebsd/sys/kern/kern_fork.c (revision 59e50df3cd1493537cfa916daf3c51a01b7ff06e)
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  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ktrace.h"
43 #include "opt_kstack_pages.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/bitstring.h>
48 #include <sys/sysproto.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
52 #include <sys/jail.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/sysctl.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mutex.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/procdesc.h>
62 #include <sys/pioctl.h>
63 #include <sys/ptrace.h>
64 #include <sys/racct.h>
65 #include <sys/resourcevar.h>
66 #include <sys/sched.h>
67 #include <sys/syscall.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
70 #include <sys/acct.h>
71 #include <sys/ktr.h>
72 #include <sys/ktrace.h>
73 #include <sys/unistd.h>
74 #include <sys/sdt.h>
75 #include <sys/sx.h>
76 #include <sys/sysent.h>
77 #include <sys/signalvar.h>
78 
79 #include <security/audit/audit.h>
80 #include <security/mac/mac_framework.h>
81 
82 #include <vm/vm.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_extern.h>
86 #include <vm/uma.h>
87 
88 #ifdef KDTRACE_HOOKS
89 #include <sys/dtrace_bsd.h>
90 dtrace_fork_func_t	dtrace_fasttrap_fork;
91 #endif
92 
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
95 
96 #ifndef _SYS_SYSPROTO_H_
97 struct fork_args {
98 	int     dummy;
99 };
100 #endif
101 
102 /* ARGSUSED */
103 int
104 sys_fork(struct thread *td, struct fork_args *uap)
105 {
106 	struct fork_req fr;
107 	int error, pid;
108 
109 	bzero(&fr, sizeof(fr));
110 	fr.fr_flags = RFFDG | RFPROC;
111 	fr.fr_pidp = &pid;
112 	error = fork1(td, &fr);
113 	if (error == 0) {
114 		td->td_retval[0] = pid;
115 		td->td_retval[1] = 0;
116 	}
117 	return (error);
118 }
119 
120 /* ARGUSED */
121 int
122 sys_pdfork(struct thread *td, struct pdfork_args *uap)
123 {
124 	struct fork_req fr;
125 	int error, fd, pid;
126 
127 	bzero(&fr, sizeof(fr));
128 	fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
129 	fr.fr_pidp = &pid;
130 	fr.fr_pd_fd = &fd;
131 	fr.fr_pd_flags = uap->flags;
132 	/*
133 	 * It is necessary to return fd by reference because 0 is a valid file
134 	 * descriptor number, and the child needs to be able to distinguish
135 	 * itself from the parent using the return value.
136 	 */
137 	error = fork1(td, &fr);
138 	if (error == 0) {
139 		td->td_retval[0] = pid;
140 		td->td_retval[1] = 0;
141 		error = copyout(&fd, uap->fdp, sizeof(fd));
142 	}
143 	return (error);
144 }
145 
146 /* ARGSUSED */
147 int
148 sys_vfork(struct thread *td, struct vfork_args *uap)
149 {
150 	struct fork_req fr;
151 	int error, pid;
152 
153 	bzero(&fr, sizeof(fr));
154 	fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
155 	fr.fr_pidp = &pid;
156 	error = fork1(td, &fr);
157 	if (error == 0) {
158 		td->td_retval[0] = pid;
159 		td->td_retval[1] = 0;
160 	}
161 	return (error);
162 }
163 
164 int
165 sys_rfork(struct thread *td, struct rfork_args *uap)
166 {
167 	struct fork_req fr;
168 	int error, pid;
169 
170 	/* Don't allow kernel-only flags. */
171 	if ((uap->flags & RFKERNELONLY) != 0)
172 		return (EINVAL);
173 
174 	AUDIT_ARG_FFLAGS(uap->flags);
175 	bzero(&fr, sizeof(fr));
176 	fr.fr_flags = uap->flags;
177 	fr.fr_pidp = &pid;
178 	error = fork1(td, &fr);
179 	if (error == 0) {
180 		td->td_retval[0] = pid;
181 		td->td_retval[1] = 0;
182 	}
183 	return (error);
184 }
185 
186 int __exclusive_cache_line	nprocs = 1;		/* process 0 */
187 int	lastpid = 0;
188 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
189     "Last used PID");
190 
191 /*
192  * Random component to lastpid generation.  We mix in a random factor to make
193  * it a little harder to predict.  We sanity check the modulus value to avoid
194  * doing it in critical paths.  Don't let it be too small or we pointlessly
195  * waste randomness entropy, and don't let it be impossibly large.  Using a
196  * modulus that is too big causes a LOT more process table scans and slows
197  * down fork processing as the pidchecked caching is defeated.
198  */
199 static int randompid = 0;
200 
201 static int
202 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
203 {
204 	int error, pid;
205 
206 	error = sysctl_wire_old_buffer(req, sizeof(int));
207 	if (error != 0)
208 		return(error);
209 	sx_xlock(&allproc_lock);
210 	pid = randompid;
211 	error = sysctl_handle_int(oidp, &pid, 0, req);
212 	if (error == 0 && req->newptr != NULL) {
213 		if (pid == 0)
214 			randompid = 0;
215 		else if (pid == 1)
216 			/* generate a random PID modulus between 100 and 1123 */
217 			randompid = 100 + arc4random() % 1024;
218 		else if (pid < 0 || pid > pid_max - 100)
219 			/* out of range */
220 			randompid = pid_max - 100;
221 		else if (pid < 100)
222 			/* Make it reasonable */
223 			randompid = 100;
224 		else
225 			randompid = pid;
226 	}
227 	sx_xunlock(&allproc_lock);
228 	return (error);
229 }
230 
231 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
232     0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value");
233 
234 extern bitstr_t proc_id_pidmap;
235 extern bitstr_t proc_id_grpidmap;
236 extern bitstr_t proc_id_sessidmap;
237 extern bitstr_t proc_id_reapmap;
238 
239 /*
240  * Find an unused process ID
241  *
242  * If RFHIGHPID is set (used during system boot), do not allocate
243  * low-numbered pids.
244  */
245 static int
246 fork_findpid(int flags)
247 {
248 	pid_t result;
249 	int trypid, random;
250 
251 	/*
252 	 * Avoid calling arc4random with procid_lock held.
253 	 */
254 	random = 0;
255 	if (__predict_false(randompid))
256 		random = arc4random() % randompid;
257 
258 	mtx_lock(&procid_lock);
259 
260 	trypid = lastpid + 1;
261 	if (flags & RFHIGHPID) {
262 		if (trypid < 10)
263 			trypid = 10;
264 	} else {
265 		trypid += random;
266 	}
267 retry:
268 	if (trypid >= pid_max)
269 		trypid = 2;
270 
271 	bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
272 	if (result == -1) {
273 		KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
274 		trypid = 2;
275 		goto retry;
276 	}
277 	if (bit_test(&proc_id_grpidmap, result) ||
278 	    bit_test(&proc_id_sessidmap, result) ||
279 	    bit_test(&proc_id_reapmap, result)) {
280 		trypid = result + 1;
281 		goto retry;
282 	}
283 
284 	/*
285 	 * RFHIGHPID does not mess with the lastpid counter during boot.
286 	 */
287 	if ((flags & RFHIGHPID) == 0)
288 		lastpid = result;
289 
290 	bit_set(&proc_id_pidmap, result);
291 	mtx_unlock(&procid_lock);
292 
293 	return (result);
294 }
295 
296 static int
297 fork_norfproc(struct thread *td, int flags)
298 {
299 	int error;
300 	struct proc *p1;
301 
302 	KASSERT((flags & RFPROC) == 0,
303 	    ("fork_norfproc called with RFPROC set"));
304 	p1 = td->td_proc;
305 
306 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
307 	    (flags & (RFCFDG | RFFDG))) {
308 		PROC_LOCK(p1);
309 		if (thread_single(p1, SINGLE_BOUNDARY)) {
310 			PROC_UNLOCK(p1);
311 			return (ERESTART);
312 		}
313 		PROC_UNLOCK(p1);
314 	}
315 
316 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
317 	if (error)
318 		goto fail;
319 
320 	/*
321 	 * Close all file descriptors.
322 	 */
323 	if (flags & RFCFDG) {
324 		struct filedesc *fdtmp;
325 		fdtmp = fdinit(td->td_proc->p_fd, false);
326 		fdescfree(td);
327 		p1->p_fd = fdtmp;
328 	}
329 
330 	/*
331 	 * Unshare file descriptors (from parent).
332 	 */
333 	if (flags & RFFDG)
334 		fdunshare(td);
335 
336 fail:
337 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
338 	    (flags & (RFCFDG | RFFDG))) {
339 		PROC_LOCK(p1);
340 		thread_single_end(p1, SINGLE_BOUNDARY);
341 		PROC_UNLOCK(p1);
342 	}
343 	return (error);
344 }
345 
346 static void
347 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
348     struct vmspace *vm2, struct file *fp_procdesc)
349 {
350 	struct proc *p1, *pptr;
351 	struct filedesc *fd;
352 	struct filedesc_to_leader *fdtol;
353 	struct sigacts *newsigacts;
354 
355 	p1 = td->td_proc;
356 
357 	PROC_LOCK(p1);
358 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
359 	    __rangeof(struct proc, p_startcopy, p_endcopy));
360 	pargs_hold(p2->p_args);
361 	PROC_UNLOCK(p1);
362 
363 	bzero(&p2->p_startzero,
364 	    __rangeof(struct proc, p_startzero, p_endzero));
365 
366 	/* Tell the prison that we exist. */
367 	prison_proc_hold(p2->p_ucred->cr_prison);
368 
369 	p2->p_state = PRS_NEW;		/* protect against others */
370 	p2->p_pid = fork_findpid(fr->fr_flags);
371 	AUDIT_ARG_PID(p2->p_pid);
372 
373 	sx_xlock(&allproc_lock);
374 	LIST_INSERT_HEAD(&allproc, p2, p_list);
375 	allproc_gen++;
376 	sx_xunlock(&allproc_lock);
377 
378 	sx_xlock(PIDHASHLOCK(p2->p_pid));
379 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
380 	sx_xunlock(PIDHASHLOCK(p2->p_pid));
381 
382 	tidhash_add(td2);
383 
384 	/*
385 	 * Malloc things while we don't hold any locks.
386 	 */
387 	if (fr->fr_flags & RFSIGSHARE)
388 		newsigacts = NULL;
389 	else
390 		newsigacts = sigacts_alloc();
391 
392 	/*
393 	 * Copy filedesc.
394 	 */
395 	if (fr->fr_flags & RFCFDG) {
396 		fd = fdinit(p1->p_fd, false);
397 		fdtol = NULL;
398 	} else if (fr->fr_flags & RFFDG) {
399 		fd = fdcopy(p1->p_fd);
400 		fdtol = NULL;
401 	} else {
402 		fd = fdshare(p1->p_fd);
403 		if (p1->p_fdtol == NULL)
404 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
405 			    p1->p_leader);
406 		if ((fr->fr_flags & RFTHREAD) != 0) {
407 			/*
408 			 * Shared file descriptor table, and shared
409 			 * process leaders.
410 			 */
411 			fdtol = p1->p_fdtol;
412 			FILEDESC_XLOCK(p1->p_fd);
413 			fdtol->fdl_refcount++;
414 			FILEDESC_XUNLOCK(p1->p_fd);
415 		} else {
416 			/*
417 			 * Shared file descriptor table, and different
418 			 * process leaders.
419 			 */
420 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
421 			    p1->p_fd, p2);
422 		}
423 	}
424 	/*
425 	 * Make a proc table entry for the new process.
426 	 * Start by zeroing the section of proc that is zero-initialized,
427 	 * then copy the section that is copied directly from the parent.
428 	 */
429 
430 	PROC_LOCK(p2);
431 	PROC_LOCK(p1);
432 
433 	bzero(&td2->td_startzero,
434 	    __rangeof(struct thread, td_startzero, td_endzero));
435 
436 	bcopy(&td->td_startcopy, &td2->td_startcopy,
437 	    __rangeof(struct thread, td_startcopy, td_endcopy));
438 
439 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
440 	td2->td_sigstk = td->td_sigstk;
441 	td2->td_flags = TDF_INMEM;
442 	td2->td_lend_user_pri = PRI_MAX;
443 
444 #ifdef VIMAGE
445 	td2->td_vnet = NULL;
446 	td2->td_vnet_lpush = NULL;
447 #endif
448 
449 	/*
450 	 * Allow the scheduler to initialize the child.
451 	 */
452 	thread_lock(td);
453 	sched_fork(td, td2);
454 	thread_unlock(td);
455 
456 	/*
457 	 * Duplicate sub-structures as needed.
458 	 * Increase reference counts on shared objects.
459 	 */
460 	p2->p_flag = P_INMEM;
461 	p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
462 	    P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
463 	    P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
464 	    P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC);
465 	p2->p_swtick = ticks;
466 	if (p1->p_flag & P_PROFIL)
467 		startprofclock(p2);
468 
469 	if (fr->fr_flags & RFSIGSHARE) {
470 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
471 	} else {
472 		sigacts_copy(newsigacts, p1->p_sigacts);
473 		p2->p_sigacts = newsigacts;
474 	}
475 
476 	if (fr->fr_flags & RFTSIGZMB)
477 	        p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
478 	else if (fr->fr_flags & RFLINUXTHPN)
479 	        p2->p_sigparent = SIGUSR1;
480 	else
481 	        p2->p_sigparent = SIGCHLD;
482 
483 	p2->p_textvp = p1->p_textvp;
484 	p2->p_fd = fd;
485 	p2->p_fdtol = fdtol;
486 
487 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
488 		p2->p_flag |= P_PROTECTED;
489 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
490 	}
491 
492 	/*
493 	 * p_limit is copy-on-write.  Bump its refcount.
494 	 */
495 	lim_fork(p1, p2);
496 
497 	thread_cow_get_proc(td2, p2);
498 
499 	pstats_fork(p1->p_stats, p2->p_stats);
500 
501 	PROC_UNLOCK(p1);
502 	PROC_UNLOCK(p2);
503 
504 	/* Bump references to the text vnode (for procfs). */
505 	if (p2->p_textvp)
506 		vrefact(p2->p_textvp);
507 
508 	/*
509 	 * Set up linkage for kernel based threading.
510 	 */
511 	if ((fr->fr_flags & RFTHREAD) != 0) {
512 		mtx_lock(&ppeers_lock);
513 		p2->p_peers = p1->p_peers;
514 		p1->p_peers = p2;
515 		p2->p_leader = p1->p_leader;
516 		mtx_unlock(&ppeers_lock);
517 		PROC_LOCK(p1->p_leader);
518 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
519 			PROC_UNLOCK(p1->p_leader);
520 			/*
521 			 * The task leader is exiting, so process p1 is
522 			 * going to be killed shortly.  Since p1 obviously
523 			 * isn't dead yet, we know that the leader is either
524 			 * sending SIGKILL's to all the processes in this
525 			 * task or is sleeping waiting for all the peers to
526 			 * exit.  We let p1 complete the fork, but we need
527 			 * to go ahead and kill the new process p2 since
528 			 * the task leader may not get a chance to send
529 			 * SIGKILL to it.  We leave it on the list so that
530 			 * the task leader will wait for this new process
531 			 * to commit suicide.
532 			 */
533 			PROC_LOCK(p2);
534 			kern_psignal(p2, SIGKILL);
535 			PROC_UNLOCK(p2);
536 		} else
537 			PROC_UNLOCK(p1->p_leader);
538 	} else {
539 		p2->p_peers = NULL;
540 		p2->p_leader = p2;
541 	}
542 
543 	sx_xlock(&proctree_lock);
544 	PGRP_LOCK(p1->p_pgrp);
545 	PROC_LOCK(p2);
546 	PROC_LOCK(p1);
547 
548 	/*
549 	 * Preserve some more flags in subprocess.  P_PROFIL has already
550 	 * been preserved.
551 	 */
552 	p2->p_flag |= p1->p_flag & P_SUGID;
553 	td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
554 	SESS_LOCK(p1->p_session);
555 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
556 		p2->p_flag |= P_CONTROLT;
557 	SESS_UNLOCK(p1->p_session);
558 	if (fr->fr_flags & RFPPWAIT)
559 		p2->p_flag |= P_PPWAIT;
560 
561 	p2->p_pgrp = p1->p_pgrp;
562 	LIST_INSERT_AFTER(p1, p2, p_pglist);
563 	PGRP_UNLOCK(p1->p_pgrp);
564 	LIST_INIT(&p2->p_children);
565 	LIST_INIT(&p2->p_orphans);
566 
567 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
568 
569 	/*
570 	 * If PF_FORK is set, the child process inherits the
571 	 * procfs ioctl flags from its parent.
572 	 */
573 	if (p1->p_pfsflags & PF_FORK) {
574 		p2->p_stops = p1->p_stops;
575 		p2->p_pfsflags = p1->p_pfsflags;
576 	}
577 
578 	/*
579 	 * This begins the section where we must prevent the parent
580 	 * from being swapped.
581 	 */
582 	_PHOLD(p1);
583 	PROC_UNLOCK(p1);
584 
585 	/*
586 	 * Attach the new process to its parent.
587 	 *
588 	 * If RFNOWAIT is set, the newly created process becomes a child
589 	 * of init.  This effectively disassociates the child from the
590 	 * parent.
591 	 */
592 	if ((fr->fr_flags & RFNOWAIT) != 0) {
593 		pptr = p1->p_reaper;
594 		p2->p_reaper = pptr;
595 	} else {
596 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
597 		    p1 : p1->p_reaper;
598 		pptr = p1;
599 	}
600 	p2->p_pptr = pptr;
601 	p2->p_oppid = pptr->p_pid;
602 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
603 	LIST_INIT(&p2->p_reaplist);
604 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
605 	if (p2->p_reaper == p1 && p1 != initproc) {
606 		p2->p_reapsubtree = p2->p_pid;
607 		proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
608 	}
609 	sx_xunlock(&proctree_lock);
610 
611 	/* Inform accounting that we have forked. */
612 	p2->p_acflag = AFORK;
613 	PROC_UNLOCK(p2);
614 
615 #ifdef KTRACE
616 	ktrprocfork(p1, p2);
617 #endif
618 
619 	/*
620 	 * Finish creating the child process.  It will return via a different
621 	 * execution path later.  (ie: directly into user mode)
622 	 */
623 	vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
624 
625 	if (fr->fr_flags == (RFFDG | RFPROC)) {
626 		VM_CNT_INC(v_forks);
627 		VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
628 		    p2->p_vmspace->vm_ssize);
629 	} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
630 		VM_CNT_INC(v_vforks);
631 		VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
632 		    p2->p_vmspace->vm_ssize);
633 	} else if (p1 == &proc0) {
634 		VM_CNT_INC(v_kthreads);
635 		VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
636 		    p2->p_vmspace->vm_ssize);
637 	} else {
638 		VM_CNT_INC(v_rforks);
639 		VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
640 		    p2->p_vmspace->vm_ssize);
641 	}
642 
643 	/*
644 	 * Associate the process descriptor with the process before anything
645 	 * can happen that might cause that process to need the descriptor.
646 	 * However, don't do this until after fork(2) can no longer fail.
647 	 */
648 	if (fr->fr_flags & RFPROCDESC)
649 		procdesc_new(p2, fr->fr_pd_flags);
650 
651 	/*
652 	 * Both processes are set up, now check if any loadable modules want
653 	 * to adjust anything.
654 	 */
655 	EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
656 
657 	/*
658 	 * Set the child start time and mark the process as being complete.
659 	 */
660 	PROC_LOCK(p2);
661 	PROC_LOCK(p1);
662 	microuptime(&p2->p_stats->p_start);
663 	PROC_SLOCK(p2);
664 	p2->p_state = PRS_NORMAL;
665 	PROC_SUNLOCK(p2);
666 
667 #ifdef KDTRACE_HOOKS
668 	/*
669 	 * Tell the DTrace fasttrap provider about the new process so that any
670 	 * tracepoints inherited from the parent can be removed. We have to do
671 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
672 	 * use pfind() later on.
673 	 */
674 	if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
675 		dtrace_fasttrap_fork(p1, p2);
676 #endif
677 	if (fr->fr_flags & RFPPWAIT) {
678 		td->td_pflags |= TDP_RFPPWAIT;
679 		td->td_rfppwait_p = p2;
680 		td->td_dbgflags |= TDB_VFORK;
681 	}
682 	PROC_UNLOCK(p2);
683 
684 	/*
685 	 * Tell any interested parties about the new process.
686 	 */
687 	knote_fork(p1->p_klist, p2->p_pid);
688 
689 	/*
690 	 * Now can be swapped.
691 	 */
692 	_PRELE(p1);
693 	PROC_UNLOCK(p1);
694 	SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
695 
696 	if (fr->fr_flags & RFPROCDESC) {
697 		procdesc_finit(p2->p_procdesc, fp_procdesc);
698 		fdrop(fp_procdesc, td);
699 	}
700 
701 	/*
702 	 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
703 	 * synced with forks in progress so it is OK if we miss it
704 	 * if being set atm.
705 	 */
706 	if ((p1->p_ptevents & PTRACE_FORK) != 0) {
707 		sx_xlock(&proctree_lock);
708 		PROC_LOCK(p2);
709 
710 		/*
711 		 * p1->p_ptevents & p1->p_pptr are protected by both
712 		 * process and proctree locks for modifications,
713 		 * so owning proctree_lock allows the race-free read.
714 		 */
715 		if ((p1->p_ptevents & PTRACE_FORK) != 0) {
716 			/*
717 			 * Arrange for debugger to receive the fork event.
718 			 *
719 			 * We can report PL_FLAG_FORKED regardless of
720 			 * P_FOLLOWFORK settings, but it does not make a sense
721 			 * for runaway child.
722 			 */
723 			td->td_dbgflags |= TDB_FORK;
724 			td->td_dbg_forked = p2->p_pid;
725 			td2->td_dbgflags |= TDB_STOPATFORK;
726 			proc_set_traced(p2, true);
727 			CTR2(KTR_PTRACE,
728 			    "do_fork: attaching to new child pid %d: oppid %d",
729 			    p2->p_pid, p2->p_oppid);
730 			proc_reparent(p2, p1->p_pptr, false);
731 		}
732 		PROC_UNLOCK(p2);
733 		sx_xunlock(&proctree_lock);
734 	}
735 
736 	racct_proc_fork_done(p2);
737 
738 	if ((fr->fr_flags & RFSTOPPED) == 0) {
739 		if (fr->fr_pidp != NULL)
740 			*fr->fr_pidp = p2->p_pid;
741 		/*
742 		 * If RFSTOPPED not requested, make child runnable and
743 		 * add to run queue.
744 		 */
745 		thread_lock(td2);
746 		TD_SET_CAN_RUN(td2);
747 		sched_add(td2, SRQ_BORING);
748 		thread_unlock(td2);
749 	} else {
750 		*fr->fr_procp = p2;
751 	}
752 }
753 
754 void
755 fork_rfppwait(struct thread *td)
756 {
757 	struct proc *p, *p2;
758 
759 	MPASS(td->td_pflags & TDP_RFPPWAIT);
760 
761 	p = td->td_proc;
762 	/*
763 	 * Preserve synchronization semantics of vfork.  If
764 	 * waiting for child to exec or exit, fork set
765 	 * P_PPWAIT on child, and there we sleep on our proc
766 	 * (in case of exit).
767 	 *
768 	 * Do it after the ptracestop() above is finished, to
769 	 * not block our debugger until child execs or exits
770 	 * to finish vfork wait.
771 	 */
772 	td->td_pflags &= ~TDP_RFPPWAIT;
773 	p2 = td->td_rfppwait_p;
774 again:
775 	PROC_LOCK(p2);
776 	while (p2->p_flag & P_PPWAIT) {
777 		PROC_LOCK(p);
778 		if (thread_suspend_check_needed()) {
779 			PROC_UNLOCK(p2);
780 			thread_suspend_check(0);
781 			PROC_UNLOCK(p);
782 			goto again;
783 		} else {
784 			PROC_UNLOCK(p);
785 		}
786 		cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
787 	}
788 	PROC_UNLOCK(p2);
789 
790 	if (td->td_dbgflags & TDB_VFORK) {
791 		PROC_LOCK(p);
792 		if (p->p_ptevents & PTRACE_VFORK)
793 			ptracestop(td, SIGTRAP, NULL);
794 		td->td_dbgflags &= ~TDB_VFORK;
795 		PROC_UNLOCK(p);
796 	}
797 }
798 
799 int
800 fork1(struct thread *td, struct fork_req *fr)
801 {
802 	struct proc *p1, *newproc;
803 	struct thread *td2;
804 	struct vmspace *vm2;
805 	struct ucred *cred;
806 	struct file *fp_procdesc;
807 	vm_ooffset_t mem_charged;
808 	int error, nprocs_new;
809 	static int curfail;
810 	static struct timeval lastfail;
811 	int flags, pages;
812 
813 	flags = fr->fr_flags;
814 	pages = fr->fr_pages;
815 
816 	if ((flags & RFSTOPPED) != 0)
817 		MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
818 	else
819 		MPASS(fr->fr_procp == NULL);
820 
821 	/* Check for the undefined or unimplemented flags. */
822 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
823 		return (EINVAL);
824 
825 	/* Signal value requires RFTSIGZMB. */
826 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
827 		return (EINVAL);
828 
829 	/* Can't copy and clear. */
830 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
831 		return (EINVAL);
832 
833 	/* Check the validity of the signal number. */
834 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
835 		return (EINVAL);
836 
837 	if ((flags & RFPROCDESC) != 0) {
838 		/* Can't not create a process yet get a process descriptor. */
839 		if ((flags & RFPROC) == 0)
840 			return (EINVAL);
841 
842 		/* Must provide a place to put a procdesc if creating one. */
843 		if (fr->fr_pd_fd == NULL)
844 			return (EINVAL);
845 
846 		/* Check if we are using supported flags. */
847 		if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
848 			return (EINVAL);
849 	}
850 
851 	p1 = td->td_proc;
852 
853 	/*
854 	 * Here we don't create a new process, but we divorce
855 	 * certain parts of a process from itself.
856 	 */
857 	if ((flags & RFPROC) == 0) {
858 		if (fr->fr_procp != NULL)
859 			*fr->fr_procp = NULL;
860 		else if (fr->fr_pidp != NULL)
861 			*fr->fr_pidp = 0;
862 		return (fork_norfproc(td, flags));
863 	}
864 
865 	fp_procdesc = NULL;
866 	newproc = NULL;
867 	vm2 = NULL;
868 
869 	/*
870 	 * Increment the nprocs resource before allocations occur.
871 	 * Although process entries are dynamically created, we still
872 	 * keep a global limit on the maximum number we will
873 	 * create. There are hard-limits as to the number of processes
874 	 * that can run, established by the KVA and memory usage for
875 	 * the process data.
876 	 *
877 	 * Don't allow a nonprivileged user to use the last ten
878 	 * processes; don't let root exceed the limit.
879 	 */
880 	nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
881 	if (nprocs_new >= maxproc - 10) {
882 		if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
883 		    nprocs_new >= maxproc) {
884 			error = EAGAIN;
885 			sx_xlock(&allproc_lock);
886 			if (ppsratecheck(&lastfail, &curfail, 1)) {
887 				printf("maxproc limit exceeded by uid %u "
888 				    "(pid %d); see tuning(7) and "
889 				    "login.conf(5)\n",
890 				    td->td_ucred->cr_ruid, p1->p_pid);
891 			}
892 			sx_xunlock(&allproc_lock);
893 			goto fail2;
894 		}
895 	}
896 
897 	/*
898 	 * If required, create a process descriptor in the parent first; we
899 	 * will abandon it if something goes wrong. We don't finit() until
900 	 * later.
901 	 */
902 	if (flags & RFPROCDESC) {
903 		error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
904 		    fr->fr_pd_flags, fr->fr_pd_fcaps);
905 		if (error != 0)
906 			goto fail2;
907 	}
908 
909 	mem_charged = 0;
910 	if (pages == 0)
911 		pages = kstack_pages;
912 	/* Allocate new proc. */
913 	newproc = uma_zalloc(proc_zone, M_WAITOK);
914 	td2 = FIRST_THREAD_IN_PROC(newproc);
915 	if (td2 == NULL) {
916 		td2 = thread_alloc(pages);
917 		if (td2 == NULL) {
918 			error = ENOMEM;
919 			goto fail2;
920 		}
921 		proc_linkup(newproc, td2);
922 	} else {
923 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
924 			if (td2->td_kstack != 0)
925 				vm_thread_dispose(td2);
926 			if (!thread_alloc_stack(td2, pages)) {
927 				error = ENOMEM;
928 				goto fail2;
929 			}
930 		}
931 	}
932 
933 	if ((flags & RFMEM) == 0) {
934 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
935 		if (vm2 == NULL) {
936 			error = ENOMEM;
937 			goto fail2;
938 		}
939 		if (!swap_reserve(mem_charged)) {
940 			/*
941 			 * The swap reservation failed. The accounting
942 			 * from the entries of the copied vm2 will be
943 			 * subtracted in vmspace_free(), so force the
944 			 * reservation there.
945 			 */
946 			swap_reserve_force(mem_charged);
947 			error = ENOMEM;
948 			goto fail2;
949 		}
950 	} else
951 		vm2 = NULL;
952 
953 	/*
954 	 * XXX: This is ugly; when we copy resource usage, we need to bump
955 	 *      per-cred resource counters.
956 	 */
957 	proc_set_cred_init(newproc, crhold(td->td_ucred));
958 
959 	/*
960 	 * Initialize resource accounting for the child process.
961 	 */
962 	error = racct_proc_fork(p1, newproc);
963 	if (error != 0) {
964 		error = EAGAIN;
965 		goto fail1;
966 	}
967 
968 #ifdef MAC
969 	mac_proc_init(newproc);
970 #endif
971 	newproc->p_klist = knlist_alloc(&newproc->p_mtx);
972 	STAILQ_INIT(&newproc->p_ktr);
973 
974 	/*
975 	 * Increment the count of procs running with this uid. Don't allow
976 	 * a nonprivileged user to exceed their current limit.
977 	 */
978 	cred = td->td_ucred;
979 	if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
980 		if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
981 			goto fail0;
982 		chgproccnt(cred->cr_ruidinfo, 1, 0);
983 	}
984 
985 	do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
986 	return (0);
987 fail0:
988 	error = EAGAIN;
989 #ifdef MAC
990 	mac_proc_destroy(newproc);
991 #endif
992 	racct_proc_exit(newproc);
993 fail1:
994 	crfree(newproc->p_ucred);
995 	newproc->p_ucred = NULL;
996 fail2:
997 	if (vm2 != NULL)
998 		vmspace_free(vm2);
999 	uma_zfree(proc_zone, newproc);
1000 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1001 		fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1002 		fdrop(fp_procdesc, td);
1003 	}
1004 	atomic_add_int(&nprocs, -1);
1005 	pause("fork", hz / 2);
1006 	return (error);
1007 }
1008 
1009 /*
1010  * Handle the return of a child process from fork1().  This function
1011  * is called from the MD fork_trampoline() entry point.
1012  */
1013 void
1014 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1015     struct trapframe *frame)
1016 {
1017 	struct proc *p;
1018 	struct thread *td;
1019 	struct thread *dtd;
1020 
1021 	td = curthread;
1022 	p = td->td_proc;
1023 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1024 
1025 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1026 	    td, td_get_sched(td), p->p_pid, td->td_name);
1027 
1028 	sched_fork_exit(td);
1029 	/*
1030 	* Processes normally resume in mi_switch() after being
1031 	* cpu_switch()'ed to, but when children start up they arrive here
1032 	* instead, so we must do much the same things as mi_switch() would.
1033 	*/
1034 	if ((dtd = PCPU_GET(deadthread))) {
1035 		PCPU_SET(deadthread, NULL);
1036 		thread_stash(dtd);
1037 	}
1038 	thread_unlock(td);
1039 
1040 	/*
1041 	 * cpu_fork_kthread_handler intercepts this function call to
1042 	 * have this call a non-return function to stay in kernel mode.
1043 	 * initproc has its own fork handler, but it does return.
1044 	 */
1045 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1046 	callout(arg, frame);
1047 
1048 	/*
1049 	 * Check if a kernel thread misbehaved and returned from its main
1050 	 * function.
1051 	 */
1052 	if (p->p_flag & P_KPROC) {
1053 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1054 		    td->td_name, p->p_pid);
1055 		kthread_exit();
1056 	}
1057 	mtx_assert(&Giant, MA_NOTOWNED);
1058 
1059 	if (p->p_sysent->sv_schedtail != NULL)
1060 		(p->p_sysent->sv_schedtail)(td);
1061 	td->td_pflags &= ~TDP_FORKING;
1062 }
1063 
1064 /*
1065  * Simplified back end of syscall(), used when returning from fork()
1066  * directly into user mode.  This function is passed in to fork_exit()
1067  * as the first parameter and is called when returning to a new
1068  * userland process.
1069  */
1070 void
1071 fork_return(struct thread *td, struct trapframe *frame)
1072 {
1073 	struct proc *p;
1074 
1075 	p = td->td_proc;
1076 	if (td->td_dbgflags & TDB_STOPATFORK) {
1077 		PROC_LOCK(p);
1078 		if ((p->p_flag & P_TRACED) != 0) {
1079 			/*
1080 			 * Inform the debugger if one is still present.
1081 			 */
1082 			td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1083 			ptracestop(td, SIGSTOP, NULL);
1084 			td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1085 		} else {
1086 			/*
1087 			 * ... otherwise clear the request.
1088 			 */
1089 			td->td_dbgflags &= ~TDB_STOPATFORK;
1090 		}
1091 		PROC_UNLOCK(p);
1092 	} else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1093  		/*
1094 		 * This is the start of a new thread in a traced
1095 		 * process.  Report a system call exit event.
1096 		 */
1097 		PROC_LOCK(p);
1098 		td->td_dbgflags |= TDB_SCX;
1099 		_STOPEVENT(p, S_SCX, td->td_sa.code);
1100 		if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1101 		    (td->td_dbgflags & TDB_BORN) != 0)
1102 			ptracestop(td, SIGTRAP, NULL);
1103 		td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1104 		PROC_UNLOCK(p);
1105 	}
1106 
1107 	userret(td, frame);
1108 
1109 #ifdef KTRACE
1110 	if (KTRPOINT(td, KTR_SYSRET))
1111 		ktrsysret(SYS_fork, 0, 0);
1112 #endif
1113 }
1114