xref: /freebsd/sys/kern/kern_fork.c (revision cfd6422a5217410fbd66f7a7a8a64d9d85e61229)
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/ptrace.h>
63 #include <sys/racct.h>
64 #include <sys/resourcevar.h>
65 #include <sys/sched.h>
66 #include <sys/syscall.h>
67 #include <sys/vmmeter.h>
68 #include <sys/vnode.h>
69 #include <sys/acct.h>
70 #include <sys/ktr.h>
71 #include <sys/ktrace.h>
72 #include <sys/unistd.h>
73 #include <sys/sdt.h>
74 #include <sys/sx.h>
75 #include <sys/sysent.h>
76 #include <sys/signalvar.h>
77 
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80 
81 #include <vm/vm.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_extern.h>
85 #include <vm/uma.h>
86 
87 #ifdef KDTRACE_HOOKS
88 #include <sys/dtrace_bsd.h>
89 dtrace_fork_func_t	dtrace_fasttrap_fork;
90 #endif
91 
92 SDT_PROVIDER_DECLARE(proc);
93 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
94 
95 #ifndef _SYS_SYSPROTO_H_
96 struct fork_args {
97 	int     dummy;
98 };
99 #endif
100 
101 /* ARGSUSED */
102 int
103 sys_fork(struct thread *td, struct fork_args *uap)
104 {
105 	struct fork_req fr;
106 	int error, pid;
107 
108 	bzero(&fr, sizeof(fr));
109 	fr.fr_flags = RFFDG | RFPROC;
110 	fr.fr_pidp = &pid;
111 	error = fork1(td, &fr);
112 	if (error == 0) {
113 		td->td_retval[0] = pid;
114 		td->td_retval[1] = 0;
115 	}
116 	return (error);
117 }
118 
119 /* ARGUSED */
120 int
121 sys_pdfork(struct thread *td, struct pdfork_args *uap)
122 {
123 	struct fork_req fr;
124 	int error, fd, pid;
125 
126 	bzero(&fr, sizeof(fr));
127 	fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
128 	fr.fr_pidp = &pid;
129 	fr.fr_pd_fd = &fd;
130 	fr.fr_pd_flags = uap->flags;
131 	AUDIT_ARG_FFLAGS(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 	/* RFSPAWN must not appear with others */
174 	if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
175 		return (EINVAL);
176 
177 	AUDIT_ARG_FFLAGS(uap->flags);
178 	bzero(&fr, sizeof(fr));
179 	if ((uap->flags & RFSPAWN) != 0) {
180 		fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
181 		fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
182 	} else {
183 		fr.fr_flags = uap->flags;
184 	}
185 	fr.fr_pidp = &pid;
186 	error = fork1(td, &fr);
187 	if (error == 0) {
188 		td->td_retval[0] = pid;
189 		td->td_retval[1] = 0;
190 	}
191 	return (error);
192 }
193 
194 int __exclusive_cache_line	nprocs = 1;		/* process 0 */
195 int	lastpid = 0;
196 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
197     "Last used PID");
198 
199 /*
200  * Random component to lastpid generation.  We mix in a random factor to make
201  * it a little harder to predict.  We sanity check the modulus value to avoid
202  * doing it in critical paths.  Don't let it be too small or we pointlessly
203  * waste randomness entropy, and don't let it be impossibly large.  Using a
204  * modulus that is too big causes a LOT more process table scans and slows
205  * down fork processing as the pidchecked caching is defeated.
206  */
207 static int randompid = 0;
208 
209 static int
210 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
211 {
212 	int error, pid;
213 
214 	error = sysctl_wire_old_buffer(req, sizeof(int));
215 	if (error != 0)
216 		return(error);
217 	sx_xlock(&allproc_lock);
218 	pid = randompid;
219 	error = sysctl_handle_int(oidp, &pid, 0, req);
220 	if (error == 0 && req->newptr != NULL) {
221 		if (pid == 0)
222 			randompid = 0;
223 		else if (pid == 1)
224 			/* generate a random PID modulus between 100 and 1123 */
225 			randompid = 100 + arc4random() % 1024;
226 		else if (pid < 0 || pid > pid_max - 100)
227 			/* out of range */
228 			randompid = pid_max - 100;
229 		else if (pid < 100)
230 			/* Make it reasonable */
231 			randompid = 100;
232 		else
233 			randompid = pid;
234 	}
235 	sx_xunlock(&allproc_lock);
236 	return (error);
237 }
238 
239 SYSCTL_PROC(_kern, OID_AUTO, randompid,
240     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
241     sysctl_kern_randompid, "I",
242     "Random PID modulus. Special values: 0: disable, 1: choose random value");
243 
244 extern bitstr_t proc_id_pidmap;
245 extern bitstr_t proc_id_grpidmap;
246 extern bitstr_t proc_id_sessidmap;
247 extern bitstr_t proc_id_reapmap;
248 
249 /*
250  * Find an unused process ID
251  *
252  * If RFHIGHPID is set (used during system boot), do not allocate
253  * low-numbered pids.
254  */
255 static int
256 fork_findpid(int flags)
257 {
258 	pid_t result;
259 	int trypid, random;
260 
261 	/*
262 	 * Avoid calling arc4random with procid_lock held.
263 	 */
264 	random = 0;
265 	if (__predict_false(randompid))
266 		random = arc4random() % randompid;
267 
268 	mtx_lock(&procid_lock);
269 
270 	trypid = lastpid + 1;
271 	if (flags & RFHIGHPID) {
272 		if (trypid < 10)
273 			trypid = 10;
274 	} else {
275 		trypid += random;
276 	}
277 retry:
278 	if (trypid >= pid_max)
279 		trypid = 2;
280 
281 	bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
282 	if (result == -1) {
283 		KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
284 		trypid = 2;
285 		goto retry;
286 	}
287 	if (bit_test(&proc_id_grpidmap, result) ||
288 	    bit_test(&proc_id_sessidmap, result) ||
289 	    bit_test(&proc_id_reapmap, result)) {
290 		trypid = result + 1;
291 		goto retry;
292 	}
293 
294 	/*
295 	 * RFHIGHPID does not mess with the lastpid counter during boot.
296 	 */
297 	if ((flags & RFHIGHPID) == 0)
298 		lastpid = result;
299 
300 	bit_set(&proc_id_pidmap, result);
301 	mtx_unlock(&procid_lock);
302 
303 	return (result);
304 }
305 
306 static int
307 fork_norfproc(struct thread *td, int flags)
308 {
309 	int error;
310 	struct proc *p1;
311 
312 	KASSERT((flags & RFPROC) == 0,
313 	    ("fork_norfproc called with RFPROC set"));
314 	p1 = td->td_proc;
315 
316 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
317 	    (flags & (RFCFDG | RFFDG))) {
318 		PROC_LOCK(p1);
319 		if (thread_single(p1, SINGLE_BOUNDARY)) {
320 			PROC_UNLOCK(p1);
321 			return (ERESTART);
322 		}
323 		PROC_UNLOCK(p1);
324 	}
325 
326 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
327 	if (error)
328 		goto fail;
329 
330 	/*
331 	 * Close all file descriptors.
332 	 */
333 	if (flags & RFCFDG) {
334 		struct filedesc *fdtmp;
335 		struct pwddesc *pdtmp;
336 		pdtmp = pdinit(td->td_proc->p_pd, false);
337 		fdtmp = fdinit(td->td_proc->p_fd, false, NULL);
338 		pdescfree(td);
339 		fdescfree(td);
340 		p1->p_fd = fdtmp;
341 		p1->p_pd = pdtmp;
342 	}
343 
344 	/*
345 	 * Unshare file descriptors (from parent).
346 	 */
347 	if (flags & RFFDG) {
348 		fdunshare(td);
349 		pdunshare(td);
350 	}
351 
352 fail:
353 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
354 	    (flags & (RFCFDG | RFFDG))) {
355 		PROC_LOCK(p1);
356 		thread_single_end(p1, SINGLE_BOUNDARY);
357 		PROC_UNLOCK(p1);
358 	}
359 	return (error);
360 }
361 
362 static void
363 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
364     struct vmspace *vm2, struct file *fp_procdesc)
365 {
366 	struct proc *p1, *pptr;
367 	struct filedesc *fd;
368 	struct filedesc_to_leader *fdtol;
369 	struct pwddesc *pd;
370 	struct sigacts *newsigacts;
371 
372 	p1 = td->td_proc;
373 
374 	PROC_LOCK(p1);
375 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
376 	    __rangeof(struct proc, p_startcopy, p_endcopy));
377 	pargs_hold(p2->p_args);
378 	PROC_UNLOCK(p1);
379 
380 	bzero(&p2->p_startzero,
381 	    __rangeof(struct proc, p_startzero, p_endzero));
382 
383 	/* Tell the prison that we exist. */
384 	prison_proc_hold(p2->p_ucred->cr_prison);
385 
386 	p2->p_state = PRS_NEW;		/* protect against others */
387 	p2->p_pid = fork_findpid(fr->fr_flags);
388 	AUDIT_ARG_PID(p2->p_pid);
389 
390 	sx_xlock(&allproc_lock);
391 	LIST_INSERT_HEAD(&allproc, p2, p_list);
392 	allproc_gen++;
393 	sx_xunlock(&allproc_lock);
394 
395 	sx_xlock(PIDHASHLOCK(p2->p_pid));
396 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
397 	sx_xunlock(PIDHASHLOCK(p2->p_pid));
398 
399 	tidhash_add(td2);
400 
401 	/*
402 	 * Malloc things while we don't hold any locks.
403 	 */
404 	if (fr->fr_flags & RFSIGSHARE)
405 		newsigacts = NULL;
406 	else
407 		newsigacts = sigacts_alloc();
408 
409 	/*
410 	 * Copy filedesc.
411 	 */
412 	if (fr->fr_flags & RFCFDG) {
413 		pd = pdinit(p1->p_pd, false);
414 		fd = fdinit(p1->p_fd, false, NULL);
415 		fdtol = NULL;
416 	} else if (fr->fr_flags & RFFDG) {
417 		if (fr->fr_flags2 & FR2_SHARE_PATHS)
418 			pd = pdshare(p1->p_pd);
419 		else
420 			pd = pdcopy(p1->p_pd);
421 		fd = fdcopy(p1->p_fd);
422 		fdtol = NULL;
423 	} else {
424 		if (fr->fr_flags2 & FR2_SHARE_PATHS)
425 			pd = pdcopy(p1->p_pd);
426 		else
427 			pd = pdshare(p1->p_pd);
428 		fd = fdshare(p1->p_fd);
429 		if (p1->p_fdtol == NULL)
430 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
431 			    p1->p_leader);
432 		if ((fr->fr_flags & RFTHREAD) != 0) {
433 			/*
434 			 * Shared file descriptor table, and shared
435 			 * process leaders.
436 			 */
437 			fdtol = p1->p_fdtol;
438 			FILEDESC_XLOCK(p1->p_fd);
439 			fdtol->fdl_refcount++;
440 			FILEDESC_XUNLOCK(p1->p_fd);
441 		} else {
442 			/*
443 			 * Shared file descriptor table, and different
444 			 * process leaders.
445 			 */
446 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
447 			    p1->p_fd, p2);
448 		}
449 	}
450 	/*
451 	 * Make a proc table entry for the new process.
452 	 * Start by zeroing the section of proc that is zero-initialized,
453 	 * then copy the section that is copied directly from the parent.
454 	 */
455 
456 	PROC_LOCK(p2);
457 	PROC_LOCK(p1);
458 
459 	bzero(&td2->td_startzero,
460 	    __rangeof(struct thread, td_startzero, td_endzero));
461 
462 	bcopy(&td->td_startcopy, &td2->td_startcopy,
463 	    __rangeof(struct thread, td_startcopy, td_endcopy));
464 
465 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
466 	td2->td_sigstk = td->td_sigstk;
467 	td2->td_flags = TDF_INMEM;
468 	td2->td_lend_user_pri = PRI_MAX;
469 
470 #ifdef VIMAGE
471 	td2->td_vnet = NULL;
472 	td2->td_vnet_lpush = NULL;
473 #endif
474 
475 	/*
476 	 * Allow the scheduler to initialize the child.
477 	 */
478 	thread_lock(td);
479 	sched_fork(td, td2);
480 	thread_unlock(td);
481 
482 	/*
483 	 * Duplicate sub-structures as needed.
484 	 * Increase reference counts on shared objects.
485 	 */
486 	p2->p_flag = P_INMEM;
487 	p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
488 	    P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
489 	    P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
490 	    P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC);
491 	p2->p_swtick = ticks;
492 	if (p1->p_flag & P_PROFIL)
493 		startprofclock(p2);
494 
495 	if (fr->fr_flags & RFSIGSHARE) {
496 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
497 	} else {
498 		sigacts_copy(newsigacts, p1->p_sigacts);
499 		p2->p_sigacts = newsigacts;
500 		if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) {
501 			mtx_lock(&p2->p_sigacts->ps_mtx);
502 			sig_drop_caught(p2);
503 			mtx_unlock(&p2->p_sigacts->ps_mtx);
504 		}
505 	}
506 
507 	if (fr->fr_flags & RFTSIGZMB)
508 	        p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
509 	else if (fr->fr_flags & RFLINUXTHPN)
510 	        p2->p_sigparent = SIGUSR1;
511 	else
512 	        p2->p_sigparent = SIGCHLD;
513 
514 	p2->p_textvp = p1->p_textvp;
515 	p2->p_fd = fd;
516 	p2->p_fdtol = fdtol;
517 	p2->p_pd = pd;
518 
519 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
520 		p2->p_flag |= P_PROTECTED;
521 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
522 	}
523 
524 	/*
525 	 * p_limit is copy-on-write.  Bump its refcount.
526 	 */
527 	lim_fork(p1, p2);
528 
529 	thread_cow_get_proc(td2, p2);
530 
531 	pstats_fork(p1->p_stats, p2->p_stats);
532 
533 	PROC_UNLOCK(p1);
534 	PROC_UNLOCK(p2);
535 
536 	/* Bump references to the text vnode (for procfs). */
537 	if (p2->p_textvp)
538 		vrefact(p2->p_textvp);
539 
540 	/*
541 	 * Set up linkage for kernel based threading.
542 	 */
543 	if ((fr->fr_flags & RFTHREAD) != 0) {
544 		mtx_lock(&ppeers_lock);
545 		p2->p_peers = p1->p_peers;
546 		p1->p_peers = p2;
547 		p2->p_leader = p1->p_leader;
548 		mtx_unlock(&ppeers_lock);
549 		PROC_LOCK(p1->p_leader);
550 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
551 			PROC_UNLOCK(p1->p_leader);
552 			/*
553 			 * The task leader is exiting, so process p1 is
554 			 * going to be killed shortly.  Since p1 obviously
555 			 * isn't dead yet, we know that the leader is either
556 			 * sending SIGKILL's to all the processes in this
557 			 * task or is sleeping waiting for all the peers to
558 			 * exit.  We let p1 complete the fork, but we need
559 			 * to go ahead and kill the new process p2 since
560 			 * the task leader may not get a chance to send
561 			 * SIGKILL to it.  We leave it on the list so that
562 			 * the task leader will wait for this new process
563 			 * to commit suicide.
564 			 */
565 			PROC_LOCK(p2);
566 			kern_psignal(p2, SIGKILL);
567 			PROC_UNLOCK(p2);
568 		} else
569 			PROC_UNLOCK(p1->p_leader);
570 	} else {
571 		p2->p_peers = NULL;
572 		p2->p_leader = p2;
573 	}
574 
575 	sx_xlock(&proctree_lock);
576 	PGRP_LOCK(p1->p_pgrp);
577 	PROC_LOCK(p2);
578 	PROC_LOCK(p1);
579 
580 	/*
581 	 * Preserve some more flags in subprocess.  P_PROFIL has already
582 	 * been preserved.
583 	 */
584 	p2->p_flag |= p1->p_flag & P_SUGID;
585 	td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK |
586 	    TDP_SIGFASTBLOCK)) | TDP_FORKING;
587 	SESS_LOCK(p1->p_session);
588 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
589 		p2->p_flag |= P_CONTROLT;
590 	SESS_UNLOCK(p1->p_session);
591 	if (fr->fr_flags & RFPPWAIT)
592 		p2->p_flag |= P_PPWAIT;
593 
594 	p2->p_pgrp = p1->p_pgrp;
595 	LIST_INSERT_AFTER(p1, p2, p_pglist);
596 	PGRP_UNLOCK(p1->p_pgrp);
597 	LIST_INIT(&p2->p_children);
598 	LIST_INIT(&p2->p_orphans);
599 
600 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
601 
602 	/*
603 	 * This begins the section where we must prevent the parent
604 	 * from being swapped.
605 	 */
606 	_PHOLD(p1);
607 	PROC_UNLOCK(p1);
608 
609 	/*
610 	 * Attach the new process to its parent.
611 	 *
612 	 * If RFNOWAIT is set, the newly created process becomes a child
613 	 * of init.  This effectively disassociates the child from the
614 	 * parent.
615 	 */
616 	if ((fr->fr_flags & RFNOWAIT) != 0) {
617 		pptr = p1->p_reaper;
618 		p2->p_reaper = pptr;
619 	} else {
620 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
621 		    p1 : p1->p_reaper;
622 		pptr = p1;
623 	}
624 	p2->p_pptr = pptr;
625 	p2->p_oppid = pptr->p_pid;
626 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
627 	LIST_INIT(&p2->p_reaplist);
628 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
629 	if (p2->p_reaper == p1 && p1 != initproc) {
630 		p2->p_reapsubtree = p2->p_pid;
631 		proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
632 	}
633 	sx_xunlock(&proctree_lock);
634 
635 	/* Inform accounting that we have forked. */
636 	p2->p_acflag = AFORK;
637 	PROC_UNLOCK(p2);
638 
639 #ifdef KTRACE
640 	ktrprocfork(p1, p2);
641 #endif
642 
643 	/*
644 	 * Finish creating the child process.  It will return via a different
645 	 * execution path later.  (ie: directly into user mode)
646 	 */
647 	vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
648 
649 	if (fr->fr_flags == (RFFDG | RFPROC)) {
650 		VM_CNT_INC(v_forks);
651 		VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
652 		    p2->p_vmspace->vm_ssize);
653 	} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
654 		VM_CNT_INC(v_vforks);
655 		VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
656 		    p2->p_vmspace->vm_ssize);
657 	} else if (p1 == &proc0) {
658 		VM_CNT_INC(v_kthreads);
659 		VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
660 		    p2->p_vmspace->vm_ssize);
661 	} else {
662 		VM_CNT_INC(v_rforks);
663 		VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
664 		    p2->p_vmspace->vm_ssize);
665 	}
666 
667 	/*
668 	 * Associate the process descriptor with the process before anything
669 	 * can happen that might cause that process to need the descriptor.
670 	 * However, don't do this until after fork(2) can no longer fail.
671 	 */
672 	if (fr->fr_flags & RFPROCDESC)
673 		procdesc_new(p2, fr->fr_pd_flags);
674 
675 	/*
676 	 * Both processes are set up, now check if any loadable modules want
677 	 * to adjust anything.
678 	 */
679 	EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
680 
681 	/*
682 	 * Set the child start time and mark the process as being complete.
683 	 */
684 	PROC_LOCK(p2);
685 	PROC_LOCK(p1);
686 	microuptime(&p2->p_stats->p_start);
687 	PROC_SLOCK(p2);
688 	p2->p_state = PRS_NORMAL;
689 	PROC_SUNLOCK(p2);
690 
691 #ifdef KDTRACE_HOOKS
692 	/*
693 	 * Tell the DTrace fasttrap provider about the new process so that any
694 	 * tracepoints inherited from the parent can be removed. We have to do
695 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
696 	 * use pfind() later on.
697 	 */
698 	if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
699 		dtrace_fasttrap_fork(p1, p2);
700 #endif
701 	if (fr->fr_flags & RFPPWAIT) {
702 		td->td_pflags |= TDP_RFPPWAIT;
703 		td->td_rfppwait_p = p2;
704 		td->td_dbgflags |= TDB_VFORK;
705 	}
706 	PROC_UNLOCK(p2);
707 
708 	/*
709 	 * Tell any interested parties about the new process.
710 	 */
711 	knote_fork(p1->p_klist, p2->p_pid);
712 
713 	/*
714 	 * Now can be swapped.
715 	 */
716 	_PRELE(p1);
717 	PROC_UNLOCK(p1);
718 	SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
719 
720 	if (fr->fr_flags & RFPROCDESC) {
721 		procdesc_finit(p2->p_procdesc, fp_procdesc);
722 		fdrop(fp_procdesc, td);
723 	}
724 
725 	/*
726 	 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
727 	 * synced with forks in progress so it is OK if we miss it
728 	 * if being set atm.
729 	 */
730 	if ((p1->p_ptevents & PTRACE_FORK) != 0) {
731 		sx_xlock(&proctree_lock);
732 		PROC_LOCK(p2);
733 
734 		/*
735 		 * p1->p_ptevents & p1->p_pptr are protected by both
736 		 * process and proctree locks for modifications,
737 		 * so owning proctree_lock allows the race-free read.
738 		 */
739 		if ((p1->p_ptevents & PTRACE_FORK) != 0) {
740 			/*
741 			 * Arrange for debugger to receive the fork event.
742 			 *
743 			 * We can report PL_FLAG_FORKED regardless of
744 			 * P_FOLLOWFORK settings, but it does not make a sense
745 			 * for runaway child.
746 			 */
747 			td->td_dbgflags |= TDB_FORK;
748 			td->td_dbg_forked = p2->p_pid;
749 			td2->td_dbgflags |= TDB_STOPATFORK;
750 			proc_set_traced(p2, true);
751 			CTR2(KTR_PTRACE,
752 			    "do_fork: attaching to new child pid %d: oppid %d",
753 			    p2->p_pid, p2->p_oppid);
754 			proc_reparent(p2, p1->p_pptr, false);
755 		}
756 		PROC_UNLOCK(p2);
757 		sx_xunlock(&proctree_lock);
758 	}
759 
760 	racct_proc_fork_done(p2);
761 
762 	if ((fr->fr_flags & RFSTOPPED) == 0) {
763 		if (fr->fr_pidp != NULL)
764 			*fr->fr_pidp = p2->p_pid;
765 		/*
766 		 * If RFSTOPPED not requested, make child runnable and
767 		 * add to run queue.
768 		 */
769 		thread_lock(td2);
770 		TD_SET_CAN_RUN(td2);
771 		sched_add(td2, SRQ_BORING);
772 	} else {
773 		*fr->fr_procp = p2;
774 	}
775 }
776 
777 void
778 fork_rfppwait(struct thread *td)
779 {
780 	struct proc *p, *p2;
781 
782 	MPASS(td->td_pflags & TDP_RFPPWAIT);
783 
784 	p = td->td_proc;
785 	/*
786 	 * Preserve synchronization semantics of vfork.  If
787 	 * waiting for child to exec or exit, fork set
788 	 * P_PPWAIT on child, and there we sleep on our proc
789 	 * (in case of exit).
790 	 *
791 	 * Do it after the ptracestop() above is finished, to
792 	 * not block our debugger until child execs or exits
793 	 * to finish vfork wait.
794 	 */
795 	td->td_pflags &= ~TDP_RFPPWAIT;
796 	p2 = td->td_rfppwait_p;
797 again:
798 	PROC_LOCK(p2);
799 	while (p2->p_flag & P_PPWAIT) {
800 		PROC_LOCK(p);
801 		if (thread_suspend_check_needed()) {
802 			PROC_UNLOCK(p2);
803 			thread_suspend_check(0);
804 			PROC_UNLOCK(p);
805 			goto again;
806 		} else {
807 			PROC_UNLOCK(p);
808 		}
809 		cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
810 	}
811 	PROC_UNLOCK(p2);
812 
813 	if (td->td_dbgflags & TDB_VFORK) {
814 		PROC_LOCK(p);
815 		if (p->p_ptevents & PTRACE_VFORK)
816 			ptracestop(td, SIGTRAP, NULL);
817 		td->td_dbgflags &= ~TDB_VFORK;
818 		PROC_UNLOCK(p);
819 	}
820 }
821 
822 int
823 fork1(struct thread *td, struct fork_req *fr)
824 {
825 	struct proc *p1, *newproc;
826 	struct thread *td2;
827 	struct vmspace *vm2;
828 	struct ucred *cred;
829 	struct file *fp_procdesc;
830 	vm_ooffset_t mem_charged;
831 	int error, nprocs_new;
832 	static int curfail;
833 	static struct timeval lastfail;
834 	int flags, pages;
835 
836 	flags = fr->fr_flags;
837 	pages = fr->fr_pages;
838 
839 	if ((flags & RFSTOPPED) != 0)
840 		MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
841 	else
842 		MPASS(fr->fr_procp == NULL);
843 
844 	/* Check for the undefined or unimplemented flags. */
845 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
846 		return (EINVAL);
847 
848 	/* Signal value requires RFTSIGZMB. */
849 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
850 		return (EINVAL);
851 
852 	/* Can't copy and clear. */
853 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
854 		return (EINVAL);
855 
856 	/* Check the validity of the signal number. */
857 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
858 		return (EINVAL);
859 
860 	if ((flags & RFPROCDESC) != 0) {
861 		/* Can't not create a process yet get a process descriptor. */
862 		if ((flags & RFPROC) == 0)
863 			return (EINVAL);
864 
865 		/* Must provide a place to put a procdesc if creating one. */
866 		if (fr->fr_pd_fd == NULL)
867 			return (EINVAL);
868 
869 		/* Check if we are using supported flags. */
870 		if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
871 			return (EINVAL);
872 	}
873 
874 	p1 = td->td_proc;
875 
876 	/*
877 	 * Here we don't create a new process, but we divorce
878 	 * certain parts of a process from itself.
879 	 */
880 	if ((flags & RFPROC) == 0) {
881 		if (fr->fr_procp != NULL)
882 			*fr->fr_procp = NULL;
883 		else if (fr->fr_pidp != NULL)
884 			*fr->fr_pidp = 0;
885 		return (fork_norfproc(td, flags));
886 	}
887 
888 	fp_procdesc = NULL;
889 	newproc = NULL;
890 	vm2 = NULL;
891 
892 	/*
893 	 * Increment the nprocs resource before allocations occur.
894 	 * Although process entries are dynamically created, we still
895 	 * keep a global limit on the maximum number we will
896 	 * create. There are hard-limits as to the number of processes
897 	 * that can run, established by the KVA and memory usage for
898 	 * the process data.
899 	 *
900 	 * Don't allow a nonprivileged user to use the last ten
901 	 * processes; don't let root exceed the limit.
902 	 */
903 	nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
904 	if (nprocs_new >= maxproc - 10) {
905 		if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
906 		    nprocs_new >= maxproc) {
907 			error = EAGAIN;
908 			sx_xlock(&allproc_lock);
909 			if (ppsratecheck(&lastfail, &curfail, 1)) {
910 				printf("maxproc limit exceeded by uid %u "
911 				    "(pid %d); see tuning(7) and "
912 				    "login.conf(5)\n",
913 				    td->td_ucred->cr_ruid, p1->p_pid);
914 			}
915 			sx_xunlock(&allproc_lock);
916 			goto fail2;
917 		}
918 	}
919 
920 	/*
921 	 * If required, create a process descriptor in the parent first; we
922 	 * will abandon it if something goes wrong. We don't finit() until
923 	 * later.
924 	 */
925 	if (flags & RFPROCDESC) {
926 		error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
927 		    fr->fr_pd_flags, fr->fr_pd_fcaps);
928 		if (error != 0)
929 			goto fail2;
930 		AUDIT_ARG_FD(*fr->fr_pd_fd);
931 	}
932 
933 	mem_charged = 0;
934 	if (pages == 0)
935 		pages = kstack_pages;
936 	/* Allocate new proc. */
937 	newproc = uma_zalloc(proc_zone, M_WAITOK);
938 	td2 = FIRST_THREAD_IN_PROC(newproc);
939 	if (td2 == NULL) {
940 		td2 = thread_alloc(pages);
941 		if (td2 == NULL) {
942 			error = ENOMEM;
943 			goto fail2;
944 		}
945 		proc_linkup(newproc, td2);
946 	} else {
947 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
948 			if (td2->td_kstack != 0)
949 				vm_thread_dispose(td2);
950 			if (!thread_alloc_stack(td2, pages)) {
951 				error = ENOMEM;
952 				goto fail2;
953 			}
954 		}
955 	}
956 
957 	if ((flags & RFMEM) == 0) {
958 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
959 		if (vm2 == NULL) {
960 			error = ENOMEM;
961 			goto fail2;
962 		}
963 		if (!swap_reserve(mem_charged)) {
964 			/*
965 			 * The swap reservation failed. The accounting
966 			 * from the entries of the copied vm2 will be
967 			 * subtracted in vmspace_free(), so force the
968 			 * reservation there.
969 			 */
970 			swap_reserve_force(mem_charged);
971 			error = ENOMEM;
972 			goto fail2;
973 		}
974 	} else
975 		vm2 = NULL;
976 
977 	/*
978 	 * XXX: This is ugly; when we copy resource usage, we need to bump
979 	 *      per-cred resource counters.
980 	 */
981 	proc_set_cred_init(newproc, td->td_ucred);
982 
983 	/*
984 	 * Initialize resource accounting for the child process.
985 	 */
986 	error = racct_proc_fork(p1, newproc);
987 	if (error != 0) {
988 		error = EAGAIN;
989 		goto fail1;
990 	}
991 
992 #ifdef MAC
993 	mac_proc_init(newproc);
994 #endif
995 	newproc->p_klist = knlist_alloc(&newproc->p_mtx);
996 	STAILQ_INIT(&newproc->p_ktr);
997 
998 	/*
999 	 * Increment the count of procs running with this uid. Don't allow
1000 	 * a nonprivileged user to exceed their current limit.
1001 	 */
1002 	cred = td->td_ucred;
1003 	if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1004 		if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1005 			goto fail0;
1006 		chgproccnt(cred->cr_ruidinfo, 1, 0);
1007 	}
1008 
1009 	do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1010 	return (0);
1011 fail0:
1012 	error = EAGAIN;
1013 #ifdef MAC
1014 	mac_proc_destroy(newproc);
1015 #endif
1016 	racct_proc_exit(newproc);
1017 fail1:
1018 	proc_unset_cred(newproc);
1019 fail2:
1020 	if (vm2 != NULL)
1021 		vmspace_free(vm2);
1022 	uma_zfree(proc_zone, newproc);
1023 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1024 		fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1025 		fdrop(fp_procdesc, td);
1026 	}
1027 	atomic_add_int(&nprocs, -1);
1028 	pause("fork", hz / 2);
1029 	return (error);
1030 }
1031 
1032 /*
1033  * Handle the return of a child process from fork1().  This function
1034  * is called from the MD fork_trampoline() entry point.
1035  */
1036 void
1037 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1038     struct trapframe *frame)
1039 {
1040 	struct proc *p;
1041 	struct thread *td;
1042 	struct thread *dtd;
1043 
1044 	td = curthread;
1045 	p = td->td_proc;
1046 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1047 
1048 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1049 	    td, td_get_sched(td), p->p_pid, td->td_name);
1050 
1051 	sched_fork_exit(td);
1052 	/*
1053 	* Processes normally resume in mi_switch() after being
1054 	* cpu_switch()'ed to, but when children start up they arrive here
1055 	* instead, so we must do much the same things as mi_switch() would.
1056 	*/
1057 	if ((dtd = PCPU_GET(deadthread))) {
1058 		PCPU_SET(deadthread, NULL);
1059 		thread_stash(dtd);
1060 	}
1061 	thread_unlock(td);
1062 
1063 	/*
1064 	 * cpu_fork_kthread_handler intercepts this function call to
1065 	 * have this call a non-return function to stay in kernel mode.
1066 	 * initproc has its own fork handler, but it does return.
1067 	 */
1068 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1069 	callout(arg, frame);
1070 
1071 	/*
1072 	 * Check if a kernel thread misbehaved and returned from its main
1073 	 * function.
1074 	 */
1075 	if (p->p_flag & P_KPROC) {
1076 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1077 		    td->td_name, p->p_pid);
1078 		kthread_exit();
1079 	}
1080 	mtx_assert(&Giant, MA_NOTOWNED);
1081 
1082 	if (p->p_sysent->sv_schedtail != NULL)
1083 		(p->p_sysent->sv_schedtail)(td);
1084 	td->td_pflags &= ~TDP_FORKING;
1085 }
1086 
1087 /*
1088  * Simplified back end of syscall(), used when returning from fork()
1089  * directly into user mode.  This function is passed in to fork_exit()
1090  * as the first parameter and is called when returning to a new
1091  * userland process.
1092  */
1093 void
1094 fork_return(struct thread *td, struct trapframe *frame)
1095 {
1096 	struct proc *p;
1097 
1098 	p = td->td_proc;
1099 	if (td->td_dbgflags & TDB_STOPATFORK) {
1100 		PROC_LOCK(p);
1101 		if ((p->p_flag & P_TRACED) != 0) {
1102 			/*
1103 			 * Inform the debugger if one is still present.
1104 			 */
1105 			td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1106 			ptracestop(td, SIGSTOP, NULL);
1107 			td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1108 		} else {
1109 			/*
1110 			 * ... otherwise clear the request.
1111 			 */
1112 			td->td_dbgflags &= ~TDB_STOPATFORK;
1113 		}
1114 		PROC_UNLOCK(p);
1115 	} else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1116  		/*
1117 		 * This is the start of a new thread in a traced
1118 		 * process.  Report a system call exit event.
1119 		 */
1120 		PROC_LOCK(p);
1121 		td->td_dbgflags |= TDB_SCX;
1122 		if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1123 		    (td->td_dbgflags & TDB_BORN) != 0)
1124 			ptracestop(td, SIGTRAP, NULL);
1125 		td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1126 		PROC_UNLOCK(p);
1127 	}
1128 
1129 	userret(td, frame);
1130 
1131 #ifdef KTRACE
1132 	if (KTRPOINT(td, KTR_SYSRET))
1133 		ktrsysret(SYS_fork, 0, 0);
1134 #endif
1135 }
1136