xref: /freebsd/sys/kern/kern_fork.c (revision d7d962ead0b6e5e8a39202d0590022082bf5bfb6)
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/msan.h>
59 #include <sys/mutex.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/procdesc.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 	AUDIT_ARG_FFLAGS(uap->flags);
133 	/*
134 	 * It is necessary to return fd by reference because 0 is a valid file
135 	 * descriptor number, and the child needs to be able to distinguish
136 	 * itself from the parent using the return value.
137 	 */
138 	error = fork1(td, &fr);
139 	if (error == 0) {
140 		td->td_retval[0] = pid;
141 		td->td_retval[1] = 0;
142 		error = copyout(&fd, uap->fdp, sizeof(fd));
143 	}
144 	return (error);
145 }
146 
147 /* ARGSUSED */
148 int
149 sys_vfork(struct thread *td, struct vfork_args *uap)
150 {
151 	struct fork_req fr;
152 	int error, pid;
153 
154 	bzero(&fr, sizeof(fr));
155 	fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
156 	fr.fr_pidp = &pid;
157 	error = fork1(td, &fr);
158 	if (error == 0) {
159 		td->td_retval[0] = pid;
160 		td->td_retval[1] = 0;
161 	}
162 	return (error);
163 }
164 
165 int
166 sys_rfork(struct thread *td, struct rfork_args *uap)
167 {
168 	struct fork_req fr;
169 	int error, pid;
170 
171 	/* Don't allow kernel-only flags. */
172 	if ((uap->flags & RFKERNELONLY) != 0)
173 		return (EINVAL);
174 	/* RFSPAWN must not appear with others */
175 	if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
176 		return (EINVAL);
177 
178 	AUDIT_ARG_FFLAGS(uap->flags);
179 	bzero(&fr, sizeof(fr));
180 	if ((uap->flags & RFSPAWN) != 0) {
181 		fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
182 		fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
183 	} else {
184 		fr.fr_flags = uap->flags;
185 	}
186 	fr.fr_pidp = &pid;
187 	error = fork1(td, &fr);
188 	if (error == 0) {
189 		td->td_retval[0] = pid;
190 		td->td_retval[1] = 0;
191 	}
192 	return (error);
193 }
194 
195 int __exclusive_cache_line	nprocs = 1;		/* process 0 */
196 int	lastpid = 0;
197 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
198     "Last used PID");
199 
200 /*
201  * Random component to lastpid generation.  We mix in a random factor to make
202  * it a little harder to predict.  We sanity check the modulus value to avoid
203  * doing it in critical paths.  Don't let it be too small or we pointlessly
204  * waste randomness entropy, and don't let it be impossibly large.  Using a
205  * modulus that is too big causes a LOT more process table scans and slows
206  * down fork processing as the pidchecked caching is defeated.
207  */
208 static int randompid = 0;
209 
210 static int
211 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
212 {
213 	int error, pid;
214 
215 	error = sysctl_wire_old_buffer(req, sizeof(int));
216 	if (error != 0)
217 		return(error);
218 	sx_xlock(&allproc_lock);
219 	pid = randompid;
220 	error = sysctl_handle_int(oidp, &pid, 0, req);
221 	if (error == 0 && req->newptr != NULL) {
222 		if (pid == 0)
223 			randompid = 0;
224 		else if (pid == 1)
225 			/* generate a random PID modulus between 100 and 1123 */
226 			randompid = 100 + arc4random() % 1024;
227 		else if (pid < 0 || pid > pid_max - 100)
228 			/* out of range */
229 			randompid = pid_max - 100;
230 		else if (pid < 100)
231 			/* Make it reasonable */
232 			randompid = 100;
233 		else
234 			randompid = pid;
235 	}
236 	sx_xunlock(&allproc_lock);
237 	return (error);
238 }
239 
240 SYSCTL_PROC(_kern, OID_AUTO, randompid,
241     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
242     sysctl_kern_randompid, "I",
243     "Random PID modulus. Special values: 0: disable, 1: choose random value");
244 
245 extern bitstr_t proc_id_pidmap;
246 extern bitstr_t proc_id_grpidmap;
247 extern bitstr_t proc_id_sessidmap;
248 extern bitstr_t proc_id_reapmap;
249 
250 /*
251  * Find an unused process ID
252  *
253  * If RFHIGHPID is set (used during system boot), do not allocate
254  * low-numbered pids.
255  */
256 static int
257 fork_findpid(int flags)
258 {
259 	pid_t result;
260 	int trypid, random;
261 
262 	/*
263 	 * Avoid calling arc4random with procid_lock held.
264 	 */
265 	random = 0;
266 	if (__predict_false(randompid))
267 		random = arc4random() % randompid;
268 
269 	mtx_lock(&procid_lock);
270 
271 	trypid = lastpid + 1;
272 	if (flags & RFHIGHPID) {
273 		if (trypid < 10)
274 			trypid = 10;
275 	} else {
276 		trypid += random;
277 	}
278 retry:
279 	if (trypid >= pid_max)
280 		trypid = 2;
281 
282 	bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
283 	if (result == -1) {
284 		KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
285 		trypid = 2;
286 		goto retry;
287 	}
288 	if (bit_test(&proc_id_grpidmap, result) ||
289 	    bit_test(&proc_id_sessidmap, result) ||
290 	    bit_test(&proc_id_reapmap, result)) {
291 		trypid = result + 1;
292 		goto retry;
293 	}
294 
295 	/*
296 	 * RFHIGHPID does not mess with the lastpid counter during boot.
297 	 */
298 	if ((flags & RFHIGHPID) == 0)
299 		lastpid = result;
300 
301 	bit_set(&proc_id_pidmap, result);
302 	mtx_unlock(&procid_lock);
303 
304 	return (result);
305 }
306 
307 static int
308 fork_norfproc(struct thread *td, int flags)
309 {
310 	int error;
311 	struct proc *p1;
312 
313 	KASSERT((flags & RFPROC) == 0,
314 	    ("fork_norfproc called with RFPROC set"));
315 	p1 = td->td_proc;
316 
317 	/*
318 	 * Quiesce other threads if necessary.  If RFMEM is not specified we
319 	 * must ensure that other threads do not concurrently create a second
320 	 * process sharing the vmspace, see vmspace_unshare().
321 	 */
322 	if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
323 	    ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
324 		PROC_LOCK(p1);
325 		if (thread_single(p1, SINGLE_BOUNDARY)) {
326 			PROC_UNLOCK(p1);
327 			return (ERESTART);
328 		}
329 		PROC_UNLOCK(p1);
330 	}
331 
332 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
333 	if (error)
334 		goto fail;
335 
336 	/*
337 	 * Close all file descriptors.
338 	 */
339 	if (flags & RFCFDG) {
340 		struct filedesc *fdtmp;
341 		struct pwddesc *pdtmp;
342 		pdtmp = pdinit(td->td_proc->p_pd, false);
343 		fdtmp = fdinit(td->td_proc->p_fd, false, NULL);
344 		pdescfree(td);
345 		fdescfree(td);
346 		p1->p_fd = fdtmp;
347 		p1->p_pd = pdtmp;
348 	}
349 
350 	/*
351 	 * Unshare file descriptors (from parent).
352 	 */
353 	if (flags & RFFDG) {
354 		fdunshare(td);
355 		pdunshare(td);
356 	}
357 
358 fail:
359 	if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
360 	    ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
361 		PROC_LOCK(p1);
362 		thread_single_end(p1, SINGLE_BOUNDARY);
363 		PROC_UNLOCK(p1);
364 	}
365 	return (error);
366 }
367 
368 static void
369 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
370     struct vmspace *vm2, struct file *fp_procdesc)
371 {
372 	struct proc *p1, *pptr;
373 	struct filedesc *fd;
374 	struct filedesc_to_leader *fdtol;
375 	struct pwddesc *pd;
376 	struct sigacts *newsigacts;
377 
378 	p1 = td->td_proc;
379 
380 	PROC_LOCK(p1);
381 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
382 	    __rangeof(struct proc, p_startcopy, p_endcopy));
383 	pargs_hold(p2->p_args);
384 	PROC_UNLOCK(p1);
385 
386 	bzero(&p2->p_startzero,
387 	    __rangeof(struct proc, p_startzero, p_endzero));
388 
389 	/* Tell the prison that we exist. */
390 	prison_proc_hold(p2->p_ucred->cr_prison);
391 
392 	p2->p_state = PRS_NEW;		/* protect against others */
393 	p2->p_pid = fork_findpid(fr->fr_flags);
394 	AUDIT_ARG_PID(p2->p_pid);
395 
396 	sx_xlock(&allproc_lock);
397 	LIST_INSERT_HEAD(&allproc, p2, p_list);
398 	allproc_gen++;
399 	sx_xunlock(&allproc_lock);
400 
401 	sx_xlock(PIDHASHLOCK(p2->p_pid));
402 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
403 	sx_xunlock(PIDHASHLOCK(p2->p_pid));
404 
405 	tidhash_add(td2);
406 
407 	/*
408 	 * Malloc things while we don't hold any locks.
409 	 */
410 	if (fr->fr_flags & RFSIGSHARE)
411 		newsigacts = NULL;
412 	else
413 		newsigacts = sigacts_alloc();
414 
415 	/*
416 	 * Copy filedesc.
417 	 */
418 	if (fr->fr_flags & RFCFDG) {
419 		pd = pdinit(p1->p_pd, false);
420 		fd = fdinit(p1->p_fd, false, NULL);
421 		fdtol = NULL;
422 	} else if (fr->fr_flags & RFFDG) {
423 		if (fr->fr_flags2 & FR2_SHARE_PATHS)
424 			pd = pdshare(p1->p_pd);
425 		else
426 			pd = pdcopy(p1->p_pd);
427 		fd = fdcopy(p1->p_fd);
428 		fdtol = NULL;
429 	} else {
430 		if (fr->fr_flags2 & FR2_SHARE_PATHS)
431 			pd = pdcopy(p1->p_pd);
432 		else
433 			pd = pdshare(p1->p_pd);
434 		fd = fdshare(p1->p_fd);
435 		if (p1->p_fdtol == NULL)
436 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
437 			    p1->p_leader);
438 		if ((fr->fr_flags & RFTHREAD) != 0) {
439 			/*
440 			 * Shared file descriptor table, and shared
441 			 * process leaders.
442 			 */
443 			fdtol = p1->p_fdtol;
444 			FILEDESC_XLOCK(p1->p_fd);
445 			fdtol->fdl_refcount++;
446 			FILEDESC_XUNLOCK(p1->p_fd);
447 		} else {
448 			/*
449 			 * Shared file descriptor table, and different
450 			 * process leaders.
451 			 */
452 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
453 			    p1->p_fd, p2);
454 		}
455 	}
456 	/*
457 	 * Make a proc table entry for the new process.
458 	 * Start by zeroing the section of proc that is zero-initialized,
459 	 * then copy the section that is copied directly from the parent.
460 	 */
461 
462 	PROC_LOCK(p2);
463 	PROC_LOCK(p1);
464 
465 	bzero(&td2->td_startzero,
466 	    __rangeof(struct thread, td_startzero, td_endzero));
467 
468 	bcopy(&td->td_startcopy, &td2->td_startcopy,
469 	    __rangeof(struct thread, td_startcopy, td_endcopy));
470 
471 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
472 	td2->td_sigstk = td->td_sigstk;
473 	td2->td_flags = TDF_INMEM;
474 	td2->td_lend_user_pri = PRI_MAX;
475 
476 #ifdef VIMAGE
477 	td2->td_vnet = NULL;
478 	td2->td_vnet_lpush = NULL;
479 #endif
480 
481 	/*
482 	 * Allow the scheduler to initialize the child.
483 	 */
484 	thread_lock(td);
485 	sched_fork(td, td2);
486 	thread_unlock(td);
487 
488 	/*
489 	 * Duplicate sub-structures as needed.
490 	 * Increase reference counts on shared objects.
491 	 */
492 	p2->p_flag = P_INMEM;
493 	p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
494 	    P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
495 	    P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
496 	    P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS);
497 	p2->p_swtick = ticks;
498 	if (p1->p_flag & P_PROFIL)
499 		startprofclock(p2);
500 
501 	if (fr->fr_flags & RFSIGSHARE) {
502 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
503 	} else {
504 		sigacts_copy(newsigacts, p1->p_sigacts);
505 		p2->p_sigacts = newsigacts;
506 		if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) {
507 			mtx_lock(&p2->p_sigacts->ps_mtx);
508 			if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
509 				sig_drop_caught(p2);
510 			if ((fr->fr_flags2 & FR2_KPROC) != 0)
511 				p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
512 			mtx_unlock(&p2->p_sigacts->ps_mtx);
513 		}
514 	}
515 
516 	if (fr->fr_flags & RFTSIGZMB)
517 	        p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
518 	else if (fr->fr_flags & RFLINUXTHPN)
519 	        p2->p_sigparent = SIGUSR1;
520 	else
521 	        p2->p_sigparent = SIGCHLD;
522 
523 	if ((fr->fr_flags2 & FR2_KPROC) != 0) {
524 		p2->p_flag |= P_SYSTEM | P_KPROC;
525 		td2->td_pflags |= TDP_KTHREAD;
526 	}
527 
528 	p2->p_textvp = p1->p_textvp;
529 	p2->p_fd = fd;
530 	p2->p_fdtol = fdtol;
531 	p2->p_pd = pd;
532 
533 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
534 		p2->p_flag |= P_PROTECTED;
535 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
536 	}
537 
538 	/*
539 	 * p_limit is copy-on-write.  Bump its refcount.
540 	 */
541 	lim_fork(p1, p2);
542 
543 	thread_cow_get_proc(td2, p2);
544 
545 	pstats_fork(p1->p_stats, p2->p_stats);
546 
547 	PROC_UNLOCK(p1);
548 	PROC_UNLOCK(p2);
549 
550 	/* Bump references to the text vnode (for procfs). */
551 	if (p2->p_textvp)
552 		vrefact(p2->p_textvp);
553 
554 	/*
555 	 * Set up linkage for kernel based threading.
556 	 */
557 	if ((fr->fr_flags & RFTHREAD) != 0) {
558 		mtx_lock(&ppeers_lock);
559 		p2->p_peers = p1->p_peers;
560 		p1->p_peers = p2;
561 		p2->p_leader = p1->p_leader;
562 		mtx_unlock(&ppeers_lock);
563 		PROC_LOCK(p1->p_leader);
564 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
565 			PROC_UNLOCK(p1->p_leader);
566 			/*
567 			 * The task leader is exiting, so process p1 is
568 			 * going to be killed shortly.  Since p1 obviously
569 			 * isn't dead yet, we know that the leader is either
570 			 * sending SIGKILL's to all the processes in this
571 			 * task or is sleeping waiting for all the peers to
572 			 * exit.  We let p1 complete the fork, but we need
573 			 * to go ahead and kill the new process p2 since
574 			 * the task leader may not get a chance to send
575 			 * SIGKILL to it.  We leave it on the list so that
576 			 * the task leader will wait for this new process
577 			 * to commit suicide.
578 			 */
579 			PROC_LOCK(p2);
580 			kern_psignal(p2, SIGKILL);
581 			PROC_UNLOCK(p2);
582 		} else
583 			PROC_UNLOCK(p1->p_leader);
584 	} else {
585 		p2->p_peers = NULL;
586 		p2->p_leader = p2;
587 	}
588 
589 	sx_xlock(&proctree_lock);
590 	PGRP_LOCK(p1->p_pgrp);
591 	PROC_LOCK(p2);
592 	PROC_LOCK(p1);
593 
594 	/*
595 	 * Preserve some more flags in subprocess.  P_PROFIL has already
596 	 * been preserved.
597 	 */
598 	p2->p_flag |= p1->p_flag & P_SUGID;
599 	td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK |
600 	    TDP_SIGFASTBLOCK)) | TDP_FORKING;
601 	SESS_LOCK(p1->p_session);
602 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
603 		p2->p_flag |= P_CONTROLT;
604 	SESS_UNLOCK(p1->p_session);
605 	if (fr->fr_flags & RFPPWAIT)
606 		p2->p_flag |= P_PPWAIT;
607 
608 	p2->p_pgrp = p1->p_pgrp;
609 	LIST_INSERT_AFTER(p1, p2, p_pglist);
610 	PGRP_UNLOCK(p1->p_pgrp);
611 	LIST_INIT(&p2->p_children);
612 	LIST_INIT(&p2->p_orphans);
613 
614 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
615 	TAILQ_INIT(&p2->p_kqtim_stop);
616 
617 	/*
618 	 * This begins the section where we must prevent the parent
619 	 * from being swapped.
620 	 */
621 	_PHOLD(p1);
622 	PROC_UNLOCK(p1);
623 
624 	/*
625 	 * Attach the new process to its parent.
626 	 *
627 	 * If RFNOWAIT is set, the newly created process becomes a child
628 	 * of init.  This effectively disassociates the child from the
629 	 * parent.
630 	 */
631 	if ((fr->fr_flags & RFNOWAIT) != 0) {
632 		pptr = p1->p_reaper;
633 		p2->p_reaper = pptr;
634 	} else {
635 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
636 		    p1 : p1->p_reaper;
637 		pptr = p1;
638 	}
639 	p2->p_pptr = pptr;
640 	p2->p_oppid = pptr->p_pid;
641 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
642 	LIST_INIT(&p2->p_reaplist);
643 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
644 	if (p2->p_reaper == p1 && p1 != initproc) {
645 		p2->p_reapsubtree = p2->p_pid;
646 		proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
647 	}
648 	sx_xunlock(&proctree_lock);
649 
650 	/* Inform accounting that we have forked. */
651 	p2->p_acflag = AFORK;
652 	PROC_UNLOCK(p2);
653 
654 #ifdef KTRACE
655 	ktrprocfork(p1, p2);
656 #endif
657 
658 	/*
659 	 * Finish creating the child process.  It will return via a different
660 	 * execution path later.  (ie: directly into user mode)
661 	 */
662 	vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
663 
664 	if (fr->fr_flags == (RFFDG | RFPROC)) {
665 		VM_CNT_INC(v_forks);
666 		VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
667 		    p2->p_vmspace->vm_ssize);
668 	} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
669 		VM_CNT_INC(v_vforks);
670 		VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
671 		    p2->p_vmspace->vm_ssize);
672 	} else if (p1 == &proc0) {
673 		VM_CNT_INC(v_kthreads);
674 		VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
675 		    p2->p_vmspace->vm_ssize);
676 	} else {
677 		VM_CNT_INC(v_rforks);
678 		VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
679 		    p2->p_vmspace->vm_ssize);
680 	}
681 
682 	/*
683 	 * Associate the process descriptor with the process before anything
684 	 * can happen that might cause that process to need the descriptor.
685 	 * However, don't do this until after fork(2) can no longer fail.
686 	 */
687 	if (fr->fr_flags & RFPROCDESC)
688 		procdesc_new(p2, fr->fr_pd_flags);
689 
690 	/*
691 	 * Both processes are set up, now check if any loadable modules want
692 	 * to adjust anything.
693 	 */
694 	EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
695 
696 	/*
697 	 * Set the child start time and mark the process as being complete.
698 	 */
699 	PROC_LOCK(p2);
700 	PROC_LOCK(p1);
701 	microuptime(&p2->p_stats->p_start);
702 	PROC_SLOCK(p2);
703 	p2->p_state = PRS_NORMAL;
704 	PROC_SUNLOCK(p2);
705 
706 #ifdef KDTRACE_HOOKS
707 	/*
708 	 * Tell the DTrace fasttrap provider about the new process so that any
709 	 * tracepoints inherited from the parent can be removed. We have to do
710 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
711 	 * use pfind() later on.
712 	 */
713 	if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
714 		dtrace_fasttrap_fork(p1, p2);
715 #endif
716 	if (fr->fr_flags & RFPPWAIT) {
717 		td->td_pflags |= TDP_RFPPWAIT;
718 		td->td_rfppwait_p = p2;
719 		td->td_dbgflags |= TDB_VFORK;
720 	}
721 	PROC_UNLOCK(p2);
722 
723 	/*
724 	 * Tell any interested parties about the new process.
725 	 */
726 	knote_fork(p1->p_klist, p2->p_pid);
727 
728 	/*
729 	 * Now can be swapped.
730 	 */
731 	_PRELE(p1);
732 	PROC_UNLOCK(p1);
733 	SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
734 
735 	if (fr->fr_flags & RFPROCDESC) {
736 		procdesc_finit(p2->p_procdesc, fp_procdesc);
737 		fdrop(fp_procdesc, td);
738 	}
739 
740 	/*
741 	 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
742 	 * synced with forks in progress so it is OK if we miss it
743 	 * if being set atm.
744 	 */
745 	if ((p1->p_ptevents & PTRACE_FORK) != 0) {
746 		sx_xlock(&proctree_lock);
747 		PROC_LOCK(p2);
748 
749 		/*
750 		 * p1->p_ptevents & p1->p_pptr are protected by both
751 		 * process and proctree locks for modifications,
752 		 * so owning proctree_lock allows the race-free read.
753 		 */
754 		if ((p1->p_ptevents & PTRACE_FORK) != 0) {
755 			/*
756 			 * Arrange for debugger to receive the fork event.
757 			 *
758 			 * We can report PL_FLAG_FORKED regardless of
759 			 * P_FOLLOWFORK settings, but it does not make a sense
760 			 * for runaway child.
761 			 */
762 			td->td_dbgflags |= TDB_FORK;
763 			td->td_dbg_forked = p2->p_pid;
764 			td2->td_dbgflags |= TDB_STOPATFORK;
765 			proc_set_traced(p2, true);
766 			CTR2(KTR_PTRACE,
767 			    "do_fork: attaching to new child pid %d: oppid %d",
768 			    p2->p_pid, p2->p_oppid);
769 			proc_reparent(p2, p1->p_pptr, false);
770 		}
771 		PROC_UNLOCK(p2);
772 		sx_xunlock(&proctree_lock);
773 	}
774 
775 	racct_proc_fork_done(p2);
776 
777 	if ((fr->fr_flags & RFSTOPPED) == 0) {
778 		if (fr->fr_pidp != NULL)
779 			*fr->fr_pidp = p2->p_pid;
780 		/*
781 		 * If RFSTOPPED not requested, make child runnable and
782 		 * add to run queue.
783 		 */
784 		thread_lock(td2);
785 		TD_SET_CAN_RUN(td2);
786 		sched_add(td2, SRQ_BORING);
787 	} else {
788 		*fr->fr_procp = p2;
789 	}
790 }
791 
792 void
793 fork_rfppwait(struct thread *td)
794 {
795 	struct proc *p, *p2;
796 
797 	MPASS(td->td_pflags & TDP_RFPPWAIT);
798 
799 	p = td->td_proc;
800 	/*
801 	 * Preserve synchronization semantics of vfork.  If
802 	 * waiting for child to exec or exit, fork set
803 	 * P_PPWAIT on child, and there we sleep on our proc
804 	 * (in case of exit).
805 	 *
806 	 * Do it after the ptracestop() above is finished, to
807 	 * not block our debugger until child execs or exits
808 	 * to finish vfork wait.
809 	 */
810 	td->td_pflags &= ~TDP_RFPPWAIT;
811 	p2 = td->td_rfppwait_p;
812 again:
813 	PROC_LOCK(p2);
814 	while (p2->p_flag & P_PPWAIT) {
815 		PROC_LOCK(p);
816 		if (thread_suspend_check_needed()) {
817 			PROC_UNLOCK(p2);
818 			thread_suspend_check(0);
819 			PROC_UNLOCK(p);
820 			goto again;
821 		} else {
822 			PROC_UNLOCK(p);
823 		}
824 		cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
825 	}
826 	PROC_UNLOCK(p2);
827 
828 	if (td->td_dbgflags & TDB_VFORK) {
829 		PROC_LOCK(p);
830 		if (p->p_ptevents & PTRACE_VFORK)
831 			ptracestop(td, SIGTRAP, NULL);
832 		td->td_dbgflags &= ~TDB_VFORK;
833 		PROC_UNLOCK(p);
834 	}
835 }
836 
837 int
838 fork1(struct thread *td, struct fork_req *fr)
839 {
840 	struct proc *p1, *newproc;
841 	struct thread *td2;
842 	struct vmspace *vm2;
843 	struct ucred *cred;
844 	struct file *fp_procdesc;
845 	vm_ooffset_t mem_charged;
846 	int error, nprocs_new;
847 	static int curfail;
848 	static struct timeval lastfail;
849 	int flags, pages;
850 
851 	flags = fr->fr_flags;
852 	pages = fr->fr_pages;
853 
854 	if ((flags & RFSTOPPED) != 0)
855 		MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
856 	else
857 		MPASS(fr->fr_procp == NULL);
858 
859 	/* Check for the undefined or unimplemented flags. */
860 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
861 		return (EINVAL);
862 
863 	/* Signal value requires RFTSIGZMB. */
864 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
865 		return (EINVAL);
866 
867 	/* Can't copy and clear. */
868 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
869 		return (EINVAL);
870 
871 	/* Check the validity of the signal number. */
872 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
873 		return (EINVAL);
874 
875 	if ((flags & RFPROCDESC) != 0) {
876 		/* Can't not create a process yet get a process descriptor. */
877 		if ((flags & RFPROC) == 0)
878 			return (EINVAL);
879 
880 		/* Must provide a place to put a procdesc if creating one. */
881 		if (fr->fr_pd_fd == NULL)
882 			return (EINVAL);
883 
884 		/* Check if we are using supported flags. */
885 		if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
886 			return (EINVAL);
887 	}
888 
889 	p1 = td->td_proc;
890 
891 	/*
892 	 * Here we don't create a new process, but we divorce
893 	 * certain parts of a process from itself.
894 	 */
895 	if ((flags & RFPROC) == 0) {
896 		if (fr->fr_procp != NULL)
897 			*fr->fr_procp = NULL;
898 		else if (fr->fr_pidp != NULL)
899 			*fr->fr_pidp = 0;
900 		return (fork_norfproc(td, flags));
901 	}
902 
903 	fp_procdesc = NULL;
904 	newproc = NULL;
905 	vm2 = NULL;
906 
907 	/*
908 	 * Increment the nprocs resource before allocations occur.
909 	 * Although process entries are dynamically created, we still
910 	 * keep a global limit on the maximum number we will
911 	 * create. There are hard-limits as to the number of processes
912 	 * that can run, established by the KVA and memory usage for
913 	 * the process data.
914 	 *
915 	 * Don't allow a nonprivileged user to use the last ten
916 	 * processes; don't let root exceed the limit.
917 	 */
918 	nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
919 	if (nprocs_new >= maxproc - 10) {
920 		if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
921 		    nprocs_new >= maxproc) {
922 			error = EAGAIN;
923 			sx_xlock(&allproc_lock);
924 			if (ppsratecheck(&lastfail, &curfail, 1)) {
925 				printf("maxproc limit exceeded by uid %u "
926 				    "(pid %d); see tuning(7) and "
927 				    "login.conf(5)\n",
928 				    td->td_ucred->cr_ruid, p1->p_pid);
929 			}
930 			sx_xunlock(&allproc_lock);
931 			goto fail2;
932 		}
933 	}
934 
935 	/*
936 	 * If required, create a process descriptor in the parent first; we
937 	 * will abandon it if something goes wrong. We don't finit() until
938 	 * later.
939 	 */
940 	if (flags & RFPROCDESC) {
941 		error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
942 		    fr->fr_pd_flags, fr->fr_pd_fcaps);
943 		if (error != 0)
944 			goto fail2;
945 		AUDIT_ARG_FD(*fr->fr_pd_fd);
946 	}
947 
948 	mem_charged = 0;
949 	if (pages == 0)
950 		pages = kstack_pages;
951 	/* Allocate new proc. */
952 	newproc = uma_zalloc(proc_zone, M_WAITOK);
953 	td2 = FIRST_THREAD_IN_PROC(newproc);
954 	if (td2 == NULL) {
955 		td2 = thread_alloc(pages);
956 		if (td2 == NULL) {
957 			error = ENOMEM;
958 			goto fail2;
959 		}
960 		proc_linkup(newproc, td2);
961 	} else {
962 		kmsan_thread_alloc(td2);
963 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
964 			if (td2->td_kstack != 0)
965 				vm_thread_dispose(td2);
966 			if (!thread_alloc_stack(td2, pages)) {
967 				error = ENOMEM;
968 				goto fail2;
969 			}
970 		}
971 	}
972 
973 	if ((flags & RFMEM) == 0) {
974 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
975 		if (vm2 == NULL) {
976 			error = ENOMEM;
977 			goto fail2;
978 		}
979 		if (!swap_reserve(mem_charged)) {
980 			/*
981 			 * The swap reservation failed. The accounting
982 			 * from the entries of the copied vm2 will be
983 			 * subtracted in vmspace_free(), so force the
984 			 * reservation there.
985 			 */
986 			swap_reserve_force(mem_charged);
987 			error = ENOMEM;
988 			goto fail2;
989 		}
990 	} else
991 		vm2 = NULL;
992 
993 	/*
994 	 * XXX: This is ugly; when we copy resource usage, we need to bump
995 	 *      per-cred resource counters.
996 	 */
997 	proc_set_cred_init(newproc, td->td_ucred);
998 
999 	/*
1000 	 * Initialize resource accounting for the child process.
1001 	 */
1002 	error = racct_proc_fork(p1, newproc);
1003 	if (error != 0) {
1004 		error = EAGAIN;
1005 		goto fail1;
1006 	}
1007 
1008 #ifdef MAC
1009 	mac_proc_init(newproc);
1010 #endif
1011 	newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1012 	STAILQ_INIT(&newproc->p_ktr);
1013 
1014 	/*
1015 	 * Increment the count of procs running with this uid. Don't allow
1016 	 * a nonprivileged user to exceed their current limit.
1017 	 */
1018 	cred = td->td_ucred;
1019 	if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1020 		if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1021 			goto fail0;
1022 		chgproccnt(cred->cr_ruidinfo, 1, 0);
1023 	}
1024 
1025 	do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1026 	return (0);
1027 fail0:
1028 	error = EAGAIN;
1029 #ifdef MAC
1030 	mac_proc_destroy(newproc);
1031 #endif
1032 	racct_proc_exit(newproc);
1033 fail1:
1034 	proc_unset_cred(newproc);
1035 fail2:
1036 	if (vm2 != NULL)
1037 		vmspace_free(vm2);
1038 	uma_zfree(proc_zone, newproc);
1039 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1040 		fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1041 		fdrop(fp_procdesc, td);
1042 	}
1043 	atomic_add_int(&nprocs, -1);
1044 	pause("fork", hz / 2);
1045 	return (error);
1046 }
1047 
1048 /*
1049  * Handle the return of a child process from fork1().  This function
1050  * is called from the MD fork_trampoline() entry point.
1051  */
1052 void
1053 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1054     struct trapframe *frame)
1055 {
1056 	struct proc *p;
1057 	struct thread *td;
1058 	struct thread *dtd;
1059 
1060 	kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED);
1061 
1062 	td = curthread;
1063 	p = td->td_proc;
1064 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1065 
1066 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1067 	    td, td_get_sched(td), p->p_pid, td->td_name);
1068 
1069 	sched_fork_exit(td);
1070 	/*
1071 	* Processes normally resume in mi_switch() after being
1072 	* cpu_switch()'ed to, but when children start up they arrive here
1073 	* instead, so we must do much the same things as mi_switch() would.
1074 	*/
1075 	if ((dtd = PCPU_GET(deadthread))) {
1076 		PCPU_SET(deadthread, NULL);
1077 		thread_stash(dtd);
1078 	}
1079 	thread_unlock(td);
1080 
1081 	/*
1082 	 * cpu_fork_kthread_handler intercepts this function call to
1083 	 * have this call a non-return function to stay in kernel mode.
1084 	 * initproc has its own fork handler, but it does return.
1085 	 */
1086 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1087 	callout(arg, frame);
1088 
1089 	/*
1090 	 * Check if a kernel thread misbehaved and returned from its main
1091 	 * function.
1092 	 */
1093 	if (p->p_flag & P_KPROC) {
1094 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1095 		    td->td_name, p->p_pid);
1096 		kthread_exit();
1097 	}
1098 	mtx_assert(&Giant, MA_NOTOWNED);
1099 
1100 	if (p->p_sysent->sv_schedtail != NULL)
1101 		(p->p_sysent->sv_schedtail)(td);
1102 	td->td_pflags &= ~TDP_FORKING;
1103 }
1104 
1105 /*
1106  * Simplified back end of syscall(), used when returning from fork()
1107  * directly into user mode.  This function is passed in to fork_exit()
1108  * as the first parameter and is called when returning to a new
1109  * userland process.
1110  */
1111 void
1112 fork_return(struct thread *td, struct trapframe *frame)
1113 {
1114 	struct proc *p;
1115 
1116 	p = td->td_proc;
1117 	if (td->td_dbgflags & TDB_STOPATFORK) {
1118 		PROC_LOCK(p);
1119 		if ((p->p_flag & P_TRACED) != 0) {
1120 			/*
1121 			 * Inform the debugger if one is still present.
1122 			 */
1123 			td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1124 			ptracestop(td, SIGSTOP, NULL);
1125 			td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1126 		} else {
1127 			/*
1128 			 * ... otherwise clear the request.
1129 			 */
1130 			td->td_dbgflags &= ~TDB_STOPATFORK;
1131 		}
1132 		PROC_UNLOCK(p);
1133 	} else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1134  		/*
1135 		 * This is the start of a new thread in a traced
1136 		 * process.  Report a system call exit event.
1137 		 */
1138 		PROC_LOCK(p);
1139 		td->td_dbgflags |= TDB_SCX;
1140 		if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1141 		    (td->td_dbgflags & TDB_BORN) != 0)
1142 			ptracestop(td, SIGTRAP, NULL);
1143 		td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1144 		PROC_UNLOCK(p);
1145 	}
1146 
1147 	/*
1148 	 * If the prison was killed mid-fork, die along with it.
1149 	 */
1150 	if (!prison_isalive(td->td_ucred->cr_prison))
1151 		exit1(td, 0, SIGKILL);
1152 
1153 	userret(td, frame);
1154 
1155 #ifdef KTRACE
1156 	if (KTRPOINT(td, KTR_SYSRET))
1157 		ktrsysret(SYS_fork, 0, 0);
1158 #endif
1159 }
1160