xref: /freebsd/sys/kern/kern_fork.c (revision 2e3f49888ec8851bafb22011533217487764fdb0)
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 
37 #include <sys/cdefs.h>
38 #include "opt_ktrace.h"
39 #include "opt_kstack_pages.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/bitstring.h>
44 #include <sys/sysproto.h>
45 #include <sys/eventhandler.h>
46 #include <sys/fcntl.h>
47 #include <sys/filedesc.h>
48 #include <sys/jail.h>
49 #include <sys/kernel.h>
50 #include <sys/kthread.h>
51 #include <sys/sysctl.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/msan.h>
55 #include <sys/mutex.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/procdesc.h>
59 #include <sys/ptrace.h>
60 #include <sys/racct.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sched.h>
63 #include <sys/syscall.h>
64 #include <sys/vmmeter.h>
65 #include <sys/vnode.h>
66 #include <sys/acct.h>
67 #include <sys/ktr.h>
68 #include <sys/ktrace.h>
69 #include <sys/unistd.h>
70 #include <sys/sdt.h>
71 #include <sys/sx.h>
72 #include <sys/sysent.h>
73 #include <sys/signalvar.h>
74 
75 #include <security/audit/audit.h>
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_extern.h>
82 #include <vm/uma.h>
83 
84 #ifdef KDTRACE_HOOKS
85 #include <sys/dtrace_bsd.h>
86 dtrace_fork_func_t	dtrace_fasttrap_fork;
87 #endif
88 
89 SDT_PROVIDER_DECLARE(proc);
90 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
91 
92 #ifndef _SYS_SYSPROTO_H_
93 struct fork_args {
94 	int     dummy;
95 };
96 #endif
97 
98 /* ARGSUSED */
99 int
100 sys_fork(struct thread *td, struct fork_args *uap)
101 {
102 	struct fork_req fr;
103 	int error, pid;
104 
105 	bzero(&fr, sizeof(fr));
106 	fr.fr_flags = RFFDG | RFPROC;
107 	fr.fr_pidp = &pid;
108 	error = fork1(td, &fr);
109 	if (error == 0) {
110 		td->td_retval[0] = pid;
111 		td->td_retval[1] = 0;
112 	}
113 	return (error);
114 }
115 
116 /* ARGUSED */
117 int
118 sys_pdfork(struct thread *td, struct pdfork_args *uap)
119 {
120 	struct fork_req fr;
121 	int error, fd, pid;
122 
123 	bzero(&fr, sizeof(fr));
124 	fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
125 	fr.fr_pidp = &pid;
126 	fr.fr_pd_fd = &fd;
127 	fr.fr_pd_flags = uap->flags;
128 	AUDIT_ARG_FFLAGS(uap->flags);
129 	/*
130 	 * It is necessary to return fd by reference because 0 is a valid file
131 	 * descriptor number, and the child needs to be able to distinguish
132 	 * itself from the parent using the return value.
133 	 */
134 	error = fork1(td, &fr);
135 	if (error == 0) {
136 		td->td_retval[0] = pid;
137 		td->td_retval[1] = 0;
138 		error = copyout(&fd, uap->fdp, sizeof(fd));
139 	}
140 	return (error);
141 }
142 
143 /* ARGSUSED */
144 int
145 sys_vfork(struct thread *td, struct vfork_args *uap)
146 {
147 	struct fork_req fr;
148 	int error, pid;
149 
150 	bzero(&fr, sizeof(fr));
151 	fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
152 	fr.fr_pidp = &pid;
153 	error = fork1(td, &fr);
154 	if (error == 0) {
155 		td->td_retval[0] = pid;
156 		td->td_retval[1] = 0;
157 	}
158 	return (error);
159 }
160 
161 int
162 sys_rfork(struct thread *td, struct rfork_args *uap)
163 {
164 	struct fork_req fr;
165 	int error, pid;
166 
167 	/* Don't allow kernel-only flags. */
168 	if ((uap->flags & RFKERNELONLY) != 0)
169 		return (EINVAL);
170 	/* RFSPAWN must not appear with others */
171 	if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
172 		return (EINVAL);
173 
174 	AUDIT_ARG_FFLAGS(uap->flags);
175 	bzero(&fr, sizeof(fr));
176 	if ((uap->flags & RFSPAWN) != 0) {
177 		fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
178 		fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
179 	} else {
180 		fr.fr_flags = uap->flags;
181 	}
182 	fr.fr_pidp = &pid;
183 	error = fork1(td, &fr);
184 	if (error == 0) {
185 		td->td_retval[0] = pid;
186 		td->td_retval[1] = 0;
187 	}
188 	return (error);
189 }
190 
191 int __exclusive_cache_line	nprocs = 1;		/* process 0 */
192 int	lastpid = 0;
193 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
194     "Last used PID");
195 
196 /*
197  * Random component to lastpid generation.  We mix in a random factor to make
198  * it a little harder to predict.  We sanity check the modulus value to avoid
199  * doing it in critical paths.  Don't let it be too small or we pointlessly
200  * waste randomness entropy, and don't let it be impossibly large.  Using a
201  * modulus that is too big causes a LOT more process table scans and slows
202  * down fork processing as the pidchecked caching is defeated.
203  */
204 static int randompid = 0;
205 
206 static int
207 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
208 {
209 	int error, pid;
210 
211 	error = sysctl_wire_old_buffer(req, sizeof(int));
212 	if (error != 0)
213 		return(error);
214 	sx_xlock(&allproc_lock);
215 	pid = randompid;
216 	error = sysctl_handle_int(oidp, &pid, 0, req);
217 	if (error == 0 && req->newptr != NULL) {
218 		if (pid == 0)
219 			randompid = 0;
220 		else if (pid == 1)
221 			/* generate a random PID modulus between 100 and 1123 */
222 			randompid = 100 + arc4random() % 1024;
223 		else if (pid < 0 || pid > pid_max - 100)
224 			/* out of range */
225 			randompid = pid_max - 100;
226 		else if (pid < 100)
227 			/* Make it reasonable */
228 			randompid = 100;
229 		else
230 			randompid = pid;
231 	}
232 	sx_xunlock(&allproc_lock);
233 	return (error);
234 }
235 
236 SYSCTL_PROC(_kern, OID_AUTO, randompid,
237     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
238     sysctl_kern_randompid, "I",
239     "Random PID modulus. Special values: 0: disable, 1: choose random value");
240 
241 extern bitstr_t proc_id_pidmap;
242 extern bitstr_t proc_id_grpidmap;
243 extern bitstr_t proc_id_sessidmap;
244 extern bitstr_t proc_id_reapmap;
245 
246 /*
247  * Find an unused process ID
248  *
249  * If RFHIGHPID is set (used during system boot), do not allocate
250  * low-numbered pids.
251  */
252 static int
253 fork_findpid(int flags)
254 {
255 	pid_t result;
256 	int trypid, random;
257 
258 	/*
259 	 * Avoid calling arc4random with procid_lock held.
260 	 */
261 	random = 0;
262 	if (__predict_false(randompid))
263 		random = arc4random() % randompid;
264 
265 	mtx_lock(&procid_lock);
266 
267 	trypid = lastpid + 1;
268 	if (flags & RFHIGHPID) {
269 		if (trypid < 10)
270 			trypid = 10;
271 	} else {
272 		trypid += random;
273 	}
274 retry:
275 	if (trypid >= pid_max)
276 		trypid = 2;
277 
278 	bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
279 	if (result == -1) {
280 		KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
281 		trypid = 2;
282 		goto retry;
283 	}
284 	if (bit_test(&proc_id_grpidmap, result) ||
285 	    bit_test(&proc_id_sessidmap, result) ||
286 	    bit_test(&proc_id_reapmap, result)) {
287 		trypid = result + 1;
288 		goto retry;
289 	}
290 
291 	/*
292 	 * RFHIGHPID does not mess with the lastpid counter during boot.
293 	 */
294 	if ((flags & RFHIGHPID) == 0)
295 		lastpid = result;
296 
297 	bit_set(&proc_id_pidmap, result);
298 	mtx_unlock(&procid_lock);
299 
300 	return (result);
301 }
302 
303 static int
304 fork_norfproc(struct thread *td, int flags)
305 {
306 	struct proc *p1;
307 	int error;
308 
309 	KASSERT((flags & RFPROC) == 0,
310 	    ("fork_norfproc called with RFPROC set"));
311 	p1 = td->td_proc;
312 
313 	/*
314 	 * Quiesce other threads if necessary.  If RFMEM is not specified we
315 	 * must ensure that other threads do not concurrently create a second
316 	 * process sharing the vmspace, see vmspace_unshare().
317 	 */
318 	if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
319 	    ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
320 		PROC_LOCK(p1);
321 		if (thread_single(p1, SINGLE_BOUNDARY)) {
322 			PROC_UNLOCK(p1);
323 			return (ERESTART);
324 		}
325 		PROC_UNLOCK(p1);
326 	}
327 
328 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
329 	if (error != 0)
330 		goto fail;
331 
332 	/*
333 	 * Close all file descriptors.
334 	 */
335 	if ((flags & RFCFDG) != 0) {
336 		struct filedesc *fdtmp;
337 		struct pwddesc *pdtmp;
338 
339 		pdtmp = pdinit(td->td_proc->p_pd, false);
340 		fdtmp = fdinit();
341 		pdescfree(td);
342 		fdescfree(td);
343 		p1->p_fd = fdtmp;
344 		p1->p_pd = pdtmp;
345 	}
346 
347 	/*
348 	 * Unshare file descriptors (from parent).
349 	 */
350 	if ((flags & RFFDG) != 0) {
351 		fdunshare(td);
352 		pdunshare(td);
353 	}
354 
355 fail:
356 	if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
357 	    ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
358 		PROC_LOCK(p1);
359 		thread_single_end(p1, SINGLE_BOUNDARY);
360 		PROC_UNLOCK(p1);
361 	}
362 	return (error);
363 }
364 
365 static void
366 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
367     struct vmspace *vm2, struct file *fp_procdesc)
368 {
369 	struct proc *p1, *pptr;
370 	struct filedesc *fd;
371 	struct filedesc_to_leader *fdtol;
372 	struct pwddesc *pd;
373 	struct sigacts *newsigacts;
374 
375 	p1 = td->td_proc;
376 
377 	PROC_LOCK(p1);
378 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
379 	    __rangeof(struct proc, p_startcopy, p_endcopy));
380 	pargs_hold(p2->p_args);
381 	PROC_UNLOCK(p1);
382 
383 	bzero(&p2->p_startzero,
384 	    __rangeof(struct proc, p_startzero, p_endzero));
385 
386 	/* Tell the prison that we exist. */
387 	prison_proc_hold(p2->p_ucred->cr_prison);
388 
389 	p2->p_state = PRS_NEW;		/* protect against others */
390 	p2->p_pid = fork_findpid(fr->fr_flags);
391 	AUDIT_ARG_PID(p2->p_pid);
392 	TSFORK(p2->p_pid, p1->p_pid);
393 
394 	sx_xlock(&allproc_lock);
395 	LIST_INSERT_HEAD(&allproc, p2, p_list);
396 	allproc_gen++;
397 	prison_proc_link(p2->p_ucred->cr_prison, p2);
398 	sx_xunlock(&allproc_lock);
399 
400 	sx_xlock(PIDHASHLOCK(p2->p_pid));
401 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
402 	sx_xunlock(PIDHASHLOCK(p2->p_pid));
403 
404 	tidhash_add(td2);
405 
406 	/*
407 	 * Malloc things while we don't hold any locks.
408 	 */
409 	if (fr->fr_flags & RFSIGSHARE)
410 		newsigacts = NULL;
411 	else
412 		newsigacts = sigacts_alloc();
413 
414 	/*
415 	 * Copy filedesc.
416 	 */
417 	if (fr->fr_flags & RFCFDG) {
418 		pd = pdinit(p1->p_pd, false);
419 		fd = fdinit();
420 		fdtol = NULL;
421 	} else if (fr->fr_flags & RFFDG) {
422 		if (fr->fr_flags2 & FR2_SHARE_PATHS)
423 			pd = pdshare(p1->p_pd);
424 		else
425 			pd = pdcopy(p1->p_pd);
426 		fd = fdcopy(p1->p_fd);
427 		fdtol = NULL;
428 	} else {
429 		if (fr->fr_flags2 & FR2_SHARE_PATHS)
430 			pd = pdcopy(p1->p_pd);
431 		else
432 			pd = pdshare(p1->p_pd);
433 		fd = fdshare(p1->p_fd);
434 		if (p1->p_fdtol == NULL)
435 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
436 			    p1->p_leader);
437 		if ((fr->fr_flags & RFTHREAD) != 0) {
438 			/*
439 			 * Shared file descriptor table, and shared
440 			 * process leaders.
441 			 */
442 			fdtol = filedesc_to_leader_share(p1->p_fdtol, p1->p_fd);
443 		} else {
444 			/*
445 			 * Shared file descriptor table, and different
446 			 * process leaders.
447 			 */
448 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
449 			    p1->p_fd, p2);
450 		}
451 	}
452 	/*
453 	 * Make a proc table entry for the new process.
454 	 * Start by zeroing the section of proc that is zero-initialized,
455 	 * then copy the section that is copied directly from the parent.
456 	 */
457 
458 	PROC_LOCK(p2);
459 	PROC_LOCK(p1);
460 
461 	bzero(&td2->td_startzero,
462 	    __rangeof(struct thread, td_startzero, td_endzero));
463 
464 	bcopy(&td->td_startcopy, &td2->td_startcopy,
465 	    __rangeof(struct thread, td_startcopy, td_endcopy));
466 
467 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
468 	td2->td_sigstk = td->td_sigstk;
469 	td2->td_flags = TDF_INMEM;
470 	td2->td_lend_user_pri = PRI_MAX;
471 
472 #ifdef VIMAGE
473 	td2->td_vnet = NULL;
474 	td2->td_vnet_lpush = NULL;
475 #endif
476 
477 	/*
478 	 * Allow the scheduler to initialize the child.
479 	 */
480 	thread_lock(td);
481 	sched_fork(td, td2);
482 	/*
483 	 * Request AST to check for TDP_RFPPWAIT.  Do it here
484 	 * to avoid calling thread_lock() again.
485 	 */
486 	if ((fr->fr_flags & RFPPWAIT) != 0)
487 		ast_sched_locked(td, TDA_VFORK);
488 	thread_unlock(td);
489 
490 	/*
491 	 * Duplicate sub-structures as needed.
492 	 * Increase reference counts on shared objects.
493 	 */
494 	p2->p_flag = P_INMEM;
495 	p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
496 	    P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
497 	    P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
498 	    P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS |
499 	    P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
500 	p2->p_swtick = ticks;
501 	if (p1->p_flag & P_PROFIL)
502 		startprofclock(p2);
503 
504 	if (fr->fr_flags & RFSIGSHARE) {
505 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
506 	} else {
507 		sigacts_copy(newsigacts, p1->p_sigacts);
508 		p2->p_sigacts = newsigacts;
509 		if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) {
510 			mtx_lock(&p2->p_sigacts->ps_mtx);
511 			if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
512 				sig_drop_caught(p2);
513 			if ((fr->fr_flags2 & FR2_KPROC) != 0)
514 				p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
515 			mtx_unlock(&p2->p_sigacts->ps_mtx);
516 		}
517 	}
518 
519 	if (fr->fr_flags & RFTSIGZMB)
520 	        p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
521 	else if (fr->fr_flags & RFLINUXTHPN)
522 	        p2->p_sigparent = SIGUSR1;
523 	else
524 	        p2->p_sigparent = SIGCHLD;
525 
526 	if ((fr->fr_flags2 & FR2_KPROC) != 0) {
527 		p2->p_flag |= P_SYSTEM | P_KPROC;
528 		td2->td_pflags |= TDP_KTHREAD;
529 	}
530 
531 	p2->p_textvp = p1->p_textvp;
532 	p2->p_textdvp = p1->p_textdvp;
533 	p2->p_fd = fd;
534 	p2->p_fdtol = fdtol;
535 	p2->p_pd = pd;
536 
537 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
538 		p2->p_flag |= P_PROTECTED;
539 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
540 	}
541 
542 	/*
543 	 * p_limit is copy-on-write.  Bump its refcount.
544 	 */
545 	lim_fork(p1, p2);
546 
547 	thread_cow_get_proc(td2, p2);
548 
549 	pstats_fork(p1->p_stats, p2->p_stats);
550 
551 	PROC_UNLOCK(p1);
552 	PROC_UNLOCK(p2);
553 
554 	/*
555 	 * Bump references to the text vnode and directory, and copy
556 	 * the hardlink name.
557 	 */
558 	if (p2->p_textvp != NULL)
559 		vrefact(p2->p_textvp);
560 	if (p2->p_textdvp != NULL)
561 		vrefact(p2->p_textdvp);
562 	p2->p_binname = p1->p_binname == NULL ? NULL :
563 	    strdup(p1->p_binname, M_PARGS);
564 
565 	/*
566 	 * Set up linkage for kernel based threading.
567 	 */
568 	if ((fr->fr_flags & RFTHREAD) != 0) {
569 		mtx_lock(&ppeers_lock);
570 		p2->p_peers = p1->p_peers;
571 		p1->p_peers = p2;
572 		p2->p_leader = p1->p_leader;
573 		mtx_unlock(&ppeers_lock);
574 		PROC_LOCK(p1->p_leader);
575 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
576 			PROC_UNLOCK(p1->p_leader);
577 			/*
578 			 * The task leader is exiting, so process p1 is
579 			 * going to be killed shortly.  Since p1 obviously
580 			 * isn't dead yet, we know that the leader is either
581 			 * sending SIGKILL's to all the processes in this
582 			 * task or is sleeping waiting for all the peers to
583 			 * exit.  We let p1 complete the fork, but we need
584 			 * to go ahead and kill the new process p2 since
585 			 * the task leader may not get a chance to send
586 			 * SIGKILL to it.  We leave it on the list so that
587 			 * the task leader will wait for this new process
588 			 * to commit suicide.
589 			 */
590 			PROC_LOCK(p2);
591 			kern_psignal(p2, SIGKILL);
592 			PROC_UNLOCK(p2);
593 		} else
594 			PROC_UNLOCK(p1->p_leader);
595 	} else {
596 		p2->p_peers = NULL;
597 		p2->p_leader = p2;
598 	}
599 
600 	sx_xlock(&proctree_lock);
601 	PGRP_LOCK(p1->p_pgrp);
602 	PROC_LOCK(p2);
603 	PROC_LOCK(p1);
604 
605 	/*
606 	 * Preserve some more flags in subprocess.  P_PROFIL has already
607 	 * been preserved.
608 	 */
609 	p2->p_flag |= p1->p_flag & P_SUGID;
610 	td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK | TDP_SIGFASTBLOCK));
611 	SESS_LOCK(p1->p_session);
612 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
613 		p2->p_flag |= P_CONTROLT;
614 	SESS_UNLOCK(p1->p_session);
615 	if (fr->fr_flags & RFPPWAIT)
616 		p2->p_flag |= P_PPWAIT;
617 
618 	p2->p_pgrp = p1->p_pgrp;
619 	LIST_INSERT_AFTER(p1, p2, p_pglist);
620 	PGRP_UNLOCK(p1->p_pgrp);
621 	LIST_INIT(&p2->p_children);
622 	LIST_INIT(&p2->p_orphans);
623 
624 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
625 
626 	/*
627 	 * This begins the section where we must prevent the parent
628 	 * from being swapped.
629 	 */
630 	_PHOLD(p1);
631 	PROC_UNLOCK(p1);
632 
633 	/*
634 	 * Attach the new process to its parent.
635 	 *
636 	 * If RFNOWAIT is set, the newly created process becomes a child
637 	 * of init.  This effectively disassociates the child from the
638 	 * parent.
639 	 */
640 	if ((fr->fr_flags & RFNOWAIT) != 0) {
641 		pptr = p1->p_reaper;
642 		p2->p_reaper = pptr;
643 	} else {
644 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
645 		    p1 : p1->p_reaper;
646 		pptr = p1;
647 	}
648 	p2->p_pptr = pptr;
649 	p2->p_oppid = pptr->p_pid;
650 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
651 	LIST_INIT(&p2->p_reaplist);
652 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
653 	if (p2->p_reaper == p1 && p1 != initproc) {
654 		p2->p_reapsubtree = p2->p_pid;
655 		proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
656 	}
657 	sx_xunlock(&proctree_lock);
658 
659 	/* Inform accounting that we have forked. */
660 	p2->p_acflag = AFORK;
661 	PROC_UNLOCK(p2);
662 
663 #ifdef KTRACE
664 	ktrprocfork(p1, p2);
665 #endif
666 
667 	/*
668 	 * Finish creating the child process.  It will return via a different
669 	 * execution path later.  (ie: directly into user mode)
670 	 */
671 	vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
672 
673 	if (fr->fr_flags == (RFFDG | RFPROC)) {
674 		VM_CNT_INC(v_forks);
675 		VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
676 		    p2->p_vmspace->vm_ssize);
677 	} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
678 		VM_CNT_INC(v_vforks);
679 		VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
680 		    p2->p_vmspace->vm_ssize);
681 	} else if (p1 == &proc0) {
682 		VM_CNT_INC(v_kthreads);
683 		VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
684 		    p2->p_vmspace->vm_ssize);
685 	} else {
686 		VM_CNT_INC(v_rforks);
687 		VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
688 		    p2->p_vmspace->vm_ssize);
689 	}
690 
691 	/*
692 	 * Associate the process descriptor with the process before anything
693 	 * can happen that might cause that process to need the descriptor.
694 	 * However, don't do this until after fork(2) can no longer fail.
695 	 */
696 	if (fr->fr_flags & RFPROCDESC)
697 		procdesc_new(p2, fr->fr_pd_flags);
698 
699 	/*
700 	 * Both processes are set up, now check if any loadable modules want
701 	 * to adjust anything.
702 	 */
703 	EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
704 
705 	/*
706 	 * Set the child start time and mark the process as being complete.
707 	 */
708 	PROC_LOCK(p2);
709 	PROC_LOCK(p1);
710 	microuptime(&p2->p_stats->p_start);
711 	PROC_SLOCK(p2);
712 	p2->p_state = PRS_NORMAL;
713 	PROC_SUNLOCK(p2);
714 
715 #ifdef KDTRACE_HOOKS
716 	/*
717 	 * Tell the DTrace fasttrap provider about the new process so that any
718 	 * tracepoints inherited from the parent can be removed. We have to do
719 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
720 	 * use pfind() later on.
721 	 */
722 	if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
723 		dtrace_fasttrap_fork(p1, p2);
724 #endif
725 	if (fr->fr_flags & RFPPWAIT) {
726 		td->td_pflags |= TDP_RFPPWAIT;
727 		td->td_rfppwait_p = p2;
728 		td->td_dbgflags |= TDB_VFORK;
729 	}
730 	PROC_UNLOCK(p2);
731 
732 	/*
733 	 * Tell any interested parties about the new process.
734 	 */
735 	knote_fork(p1->p_klist, p2->p_pid);
736 
737 	/*
738 	 * Now can be swapped.
739 	 */
740 	_PRELE(p1);
741 	PROC_UNLOCK(p1);
742 	SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
743 
744 	if (fr->fr_flags & RFPROCDESC) {
745 		procdesc_finit(p2->p_procdesc, fp_procdesc);
746 		fdrop(fp_procdesc, td);
747 	}
748 
749 	/*
750 	 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
751 	 * synced with forks in progress so it is OK if we miss it
752 	 * if being set atm.
753 	 */
754 	if ((p1->p_ptevents & PTRACE_FORK) != 0) {
755 		sx_xlock(&proctree_lock);
756 		PROC_LOCK(p2);
757 
758 		/*
759 		 * p1->p_ptevents & p1->p_pptr are protected by both
760 		 * process and proctree locks for modifications,
761 		 * so owning proctree_lock allows the race-free read.
762 		 */
763 		if ((p1->p_ptevents & PTRACE_FORK) != 0) {
764 			/*
765 			 * Arrange for debugger to receive the fork event.
766 			 *
767 			 * We can report PL_FLAG_FORKED regardless of
768 			 * P_FOLLOWFORK settings, but it does not make a sense
769 			 * for runaway child.
770 			 */
771 			td->td_dbgflags |= TDB_FORK;
772 			td->td_dbg_forked = p2->p_pid;
773 			td2->td_dbgflags |= TDB_STOPATFORK;
774 			proc_set_traced(p2, true);
775 			CTR2(KTR_PTRACE,
776 			    "do_fork: attaching to new child pid %d: oppid %d",
777 			    p2->p_pid, p2->p_oppid);
778 			proc_reparent(p2, p1->p_pptr, false);
779 		}
780 		PROC_UNLOCK(p2);
781 		sx_xunlock(&proctree_lock);
782 	}
783 
784 	racct_proc_fork_done(p2);
785 
786 	if ((fr->fr_flags & RFSTOPPED) == 0) {
787 		if (fr->fr_pidp != NULL)
788 			*fr->fr_pidp = p2->p_pid;
789 		/*
790 		 * If RFSTOPPED not requested, make child runnable and
791 		 * add to run queue.
792 		 */
793 		thread_lock(td2);
794 		TD_SET_CAN_RUN(td2);
795 		sched_add(td2, SRQ_BORING);
796 	} else {
797 		*fr->fr_procp = p2;
798 	}
799 }
800 
801 static void
802 ast_vfork(struct thread *td, int tda __unused)
803 {
804 	struct proc *p, *p2;
805 
806 	MPASS(td->td_pflags & TDP_RFPPWAIT);
807 
808 	p = td->td_proc;
809 	/*
810 	 * Preserve synchronization semantics of vfork.  If
811 	 * waiting for child to exec or exit, fork set
812 	 * P_PPWAIT on child, and there we sleep on our proc
813 	 * (in case of exit).
814 	 *
815 	 * Do it after the ptracestop() above is finished, to
816 	 * not block our debugger until child execs or exits
817 	 * to finish vfork wait.
818 	 */
819 	td->td_pflags &= ~TDP_RFPPWAIT;
820 	p2 = td->td_rfppwait_p;
821 again:
822 	PROC_LOCK(p2);
823 	while (p2->p_flag & P_PPWAIT) {
824 		PROC_LOCK(p);
825 		if (thread_suspend_check_needed()) {
826 			PROC_UNLOCK(p2);
827 			thread_suspend_check(0);
828 			PROC_UNLOCK(p);
829 			goto again;
830 		} else {
831 			PROC_UNLOCK(p);
832 		}
833 		cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
834 	}
835 	PROC_UNLOCK(p2);
836 
837 	if (td->td_dbgflags & TDB_VFORK) {
838 		PROC_LOCK(p);
839 		if (p->p_ptevents & PTRACE_VFORK)
840 			ptracestop(td, SIGTRAP, NULL);
841 		td->td_dbgflags &= ~TDB_VFORK;
842 		PROC_UNLOCK(p);
843 	}
844 }
845 
846 int
847 fork1(struct thread *td, struct fork_req *fr)
848 {
849 	struct proc *p1, *newproc;
850 	struct thread *td2;
851 	struct vmspace *vm2;
852 	struct ucred *cred;
853 	struct file *fp_procdesc;
854 	struct pgrp *pg;
855 	vm_ooffset_t mem_charged;
856 	int error, nprocs_new;
857 	static int curfail;
858 	static struct timeval lastfail;
859 	int flags, pages;
860 	bool killsx_locked, singlethreaded;
861 
862 	flags = fr->fr_flags;
863 	pages = fr->fr_pages;
864 
865 	if ((flags & RFSTOPPED) != 0)
866 		MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
867 	else
868 		MPASS(fr->fr_procp == NULL);
869 
870 	/* Check for the undefined or unimplemented flags. */
871 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
872 		return (EINVAL);
873 
874 	/* Signal value requires RFTSIGZMB. */
875 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
876 		return (EINVAL);
877 
878 	/* Can't copy and clear. */
879 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
880 		return (EINVAL);
881 
882 	/* Check the validity of the signal number. */
883 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
884 		return (EINVAL);
885 
886 	if ((flags & RFPROCDESC) != 0) {
887 		/* Can't not create a process yet get a process descriptor. */
888 		if ((flags & RFPROC) == 0)
889 			return (EINVAL);
890 
891 		/* Must provide a place to put a procdesc if creating one. */
892 		if (fr->fr_pd_fd == NULL)
893 			return (EINVAL);
894 
895 		/* Check if we are using supported flags. */
896 		if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
897 			return (EINVAL);
898 	}
899 
900 	p1 = td->td_proc;
901 
902 	/*
903 	 * Here we don't create a new process, but we divorce
904 	 * certain parts of a process from itself.
905 	 */
906 	if ((flags & RFPROC) == 0) {
907 		if (fr->fr_procp != NULL)
908 			*fr->fr_procp = NULL;
909 		else if (fr->fr_pidp != NULL)
910 			*fr->fr_pidp = 0;
911 		return (fork_norfproc(td, flags));
912 	}
913 
914 	fp_procdesc = NULL;
915 	newproc = NULL;
916 	vm2 = NULL;
917 	killsx_locked = false;
918 	singlethreaded = false;
919 
920 	/*
921 	 * Increment the nprocs resource before allocations occur.
922 	 * Although process entries are dynamically created, we still
923 	 * keep a global limit on the maximum number we will
924 	 * create. There are hard-limits as to the number of processes
925 	 * that can run, established by the KVA and memory usage for
926 	 * the process data.
927 	 *
928 	 * Don't allow a nonprivileged user to use the last ten
929 	 * processes; don't let root exceed the limit.
930 	 */
931 	nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
932 	if (nprocs_new >= maxproc - 10) {
933 		if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
934 		    nprocs_new >= maxproc) {
935 			error = EAGAIN;
936 			sx_xlock(&allproc_lock);
937 			if (ppsratecheck(&lastfail, &curfail, 1)) {
938 				printf("maxproc limit exceeded by uid %u "
939 				    "(pid %d); see tuning(7) and "
940 				    "login.conf(5)\n",
941 				    td->td_ucred->cr_ruid, p1->p_pid);
942 			}
943 			sx_xunlock(&allproc_lock);
944 			goto fail2;
945 		}
946 	}
947 
948 	/*
949 	 * If we are possibly multi-threaded, and there is a process
950 	 * sending a signal to our group right now, ensure that our
951 	 * other threads cannot be chosen for the signal queueing.
952 	 * Otherwise, this might delay signal action, and make the new
953 	 * child escape the signaling.
954 	 */
955 	pg = p1->p_pgrp;
956 	if (p1->p_numthreads > 1) {
957 		if (sx_try_slock(&pg->pg_killsx) != 0) {
958 			killsx_locked = true;
959 		} else {
960 			PROC_LOCK(p1);
961 			if (thread_single(p1, SINGLE_BOUNDARY)) {
962 				PROC_UNLOCK(p1);
963 				error = ERESTART;
964 				goto fail2;
965 			}
966 			PROC_UNLOCK(p1);
967 			singlethreaded = true;
968 		}
969 	}
970 
971 	/*
972 	 * Atomically check for signals and block processes from sending
973 	 * a signal to our process group until the child is visible.
974 	 */
975 	if (!killsx_locked && sx_slock_sig(&pg->pg_killsx) != 0) {
976 		error = ERESTART;
977 		goto fail2;
978 	}
979 	if (__predict_false(p1->p_pgrp != pg || sig_intr() != 0)) {
980 		/*
981 		 * Either the process was moved to other process
982 		 * group, or there is pending signal.  sx_slock_sig()
983 		 * does not check for signals if not sleeping for the
984 		 * lock.
985 		 */
986 		sx_sunlock(&pg->pg_killsx);
987 		killsx_locked = false;
988 		error = ERESTART;
989 		goto fail2;
990 	} else {
991 		killsx_locked = true;
992 	}
993 
994 	/*
995 	 * If required, create a process descriptor in the parent first; we
996 	 * will abandon it if something goes wrong. We don't finit() until
997 	 * later.
998 	 */
999 	if (flags & RFPROCDESC) {
1000 		error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
1001 		    fr->fr_pd_flags, fr->fr_pd_fcaps);
1002 		if (error != 0)
1003 			goto fail2;
1004 		AUDIT_ARG_FD(*fr->fr_pd_fd);
1005 	}
1006 
1007 	mem_charged = 0;
1008 	if (pages == 0)
1009 		pages = kstack_pages;
1010 	/* Allocate new proc. */
1011 	newproc = uma_zalloc(proc_zone, M_WAITOK);
1012 	td2 = FIRST_THREAD_IN_PROC(newproc);
1013 	if (td2 == NULL) {
1014 		td2 = thread_alloc(pages);
1015 		if (td2 == NULL) {
1016 			error = ENOMEM;
1017 			goto fail2;
1018 		}
1019 		proc_linkup(newproc, td2);
1020 	} else {
1021 		kmsan_thread_alloc(td2);
1022 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
1023 			if (td2->td_kstack != 0)
1024 				vm_thread_dispose(td2);
1025 			if (!thread_alloc_stack(td2, pages)) {
1026 				error = ENOMEM;
1027 				goto fail2;
1028 			}
1029 		}
1030 	}
1031 
1032 	if ((flags & RFMEM) == 0) {
1033 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
1034 		if (vm2 == NULL) {
1035 			error = ENOMEM;
1036 			goto fail2;
1037 		}
1038 		if (!swap_reserve(mem_charged)) {
1039 			/*
1040 			 * The swap reservation failed. The accounting
1041 			 * from the entries of the copied vm2 will be
1042 			 * subtracted in vmspace_free(), so force the
1043 			 * reservation there.
1044 			 */
1045 			swap_reserve_force(mem_charged);
1046 			error = ENOMEM;
1047 			goto fail2;
1048 		}
1049 	} else
1050 		vm2 = NULL;
1051 
1052 	/*
1053 	 * XXX: This is ugly; when we copy resource usage, we need to bump
1054 	 *      per-cred resource counters.
1055 	 */
1056 	newproc->p_ucred = crcowget(td->td_ucred);
1057 
1058 	/*
1059 	 * Initialize resource accounting for the child process.
1060 	 */
1061 	error = racct_proc_fork(p1, newproc);
1062 	if (error != 0) {
1063 		error = EAGAIN;
1064 		goto fail1;
1065 	}
1066 
1067 #ifdef MAC
1068 	mac_proc_init(newproc);
1069 #endif
1070 	newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1071 	STAILQ_INIT(&newproc->p_ktr);
1072 
1073 	/*
1074 	 * Increment the count of procs running with this uid. Don't allow
1075 	 * a nonprivileged user to exceed their current limit.
1076 	 */
1077 	cred = td->td_ucred;
1078 	if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1079 		if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1080 			goto fail0;
1081 		chgproccnt(cred->cr_ruidinfo, 1, 0);
1082 	}
1083 
1084 	do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1085 	error = 0;
1086 	goto cleanup;
1087 fail0:
1088 	error = EAGAIN;
1089 #ifdef MAC
1090 	mac_proc_destroy(newproc);
1091 #endif
1092 	racct_proc_exit(newproc);
1093 fail1:
1094 	proc_unset_cred(newproc);
1095 fail2:
1096 	if (vm2 != NULL)
1097 		vmspace_free(vm2);
1098 	uma_zfree(proc_zone, newproc);
1099 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1100 		fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1101 		fdrop(fp_procdesc, td);
1102 	}
1103 	atomic_add_int(&nprocs, -1);
1104 cleanup:
1105 	if (killsx_locked)
1106 		sx_sunlock(&pg->pg_killsx);
1107 	if (singlethreaded) {
1108 		PROC_LOCK(p1);
1109 		thread_single_end(p1, SINGLE_BOUNDARY);
1110 		PROC_UNLOCK(p1);
1111 	}
1112 	if (error != 0)
1113 		pause("fork", hz / 2);
1114 	return (error);
1115 }
1116 
1117 /*
1118  * Handle the return of a child process from fork1().  This function
1119  * is called from the MD fork_trampoline() entry point.
1120  */
1121 void
1122 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1123     struct trapframe *frame)
1124 {
1125 	struct proc *p;
1126 	struct thread *td;
1127 	struct thread *dtd;
1128 
1129 	kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED);
1130 
1131 	td = curthread;
1132 	p = td->td_proc;
1133 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1134 
1135 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1136 	    td, td_get_sched(td), p->p_pid, td->td_name);
1137 
1138 	sched_fork_exit(td);
1139 
1140 	/*
1141 	 * Processes normally resume in mi_switch() after being
1142 	 * cpu_switch()'ed to, but when children start up they arrive here
1143 	 * instead, so we must do much the same things as mi_switch() would.
1144 	 */
1145 	if ((dtd = PCPU_GET(deadthread))) {
1146 		PCPU_SET(deadthread, NULL);
1147 		thread_stash(dtd);
1148 	}
1149 	thread_unlock(td);
1150 
1151 	/*
1152 	 * cpu_fork_kthread_handler intercepts this function call to
1153 	 * have this call a non-return function to stay in kernel mode.
1154 	 * initproc has its own fork handler, but it does return.
1155 	 */
1156 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1157 	callout(arg, frame);
1158 
1159 	/*
1160 	 * Check if a kernel thread misbehaved and returned from its main
1161 	 * function.
1162 	 */
1163 	if (p->p_flag & P_KPROC) {
1164 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1165 		    td->td_name, p->p_pid);
1166 		kthread_exit();
1167 	}
1168 	mtx_assert(&Giant, MA_NOTOWNED);
1169 
1170 	/*
1171 	 * Now going to return to userland.
1172 	 */
1173 
1174 	if (p->p_sysent->sv_schedtail != NULL)
1175 		(p->p_sysent->sv_schedtail)(td);
1176 
1177 	userret(td, frame);
1178 }
1179 
1180 /*
1181  * Simplified back end of syscall(), used when returning from fork()
1182  * directly into user mode.  This function is passed in to fork_exit()
1183  * as the first parameter and is called when returning to a new
1184  * userland process.
1185  */
1186 void
1187 fork_return(struct thread *td, struct trapframe *frame)
1188 {
1189 	struct proc *p;
1190 
1191 	p = td->td_proc;
1192 	if (td->td_dbgflags & TDB_STOPATFORK) {
1193 		PROC_LOCK(p);
1194 		if ((p->p_flag & P_TRACED) != 0) {
1195 			/*
1196 			 * Inform the debugger if one is still present.
1197 			 */
1198 			td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1199 			ptracestop(td, SIGSTOP, NULL);
1200 			td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1201 		} else {
1202 			/*
1203 			 * ... otherwise clear the request.
1204 			 */
1205 			td->td_dbgflags &= ~TDB_STOPATFORK;
1206 		}
1207 		PROC_UNLOCK(p);
1208 	} else if (p->p_flag & P_TRACED) {
1209  		/*
1210 		 * This is the start of a new thread in a traced
1211 		 * process.  Report a system call exit event.
1212 		 */
1213 		PROC_LOCK(p);
1214 		td->td_dbgflags |= TDB_SCX;
1215 		if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1216 		    (td->td_dbgflags & TDB_BORN) != 0)
1217 			ptracestop(td, SIGTRAP, NULL);
1218 		td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1219 		PROC_UNLOCK(p);
1220 	}
1221 
1222 	/*
1223 	 * If the prison was killed mid-fork, die along with it.
1224 	 */
1225 	if (!prison_isalive(td->td_ucred->cr_prison))
1226 		exit1(td, 0, SIGKILL);
1227 
1228 #ifdef KTRACE
1229 	if (KTRPOINT(td, KTR_SYSRET))
1230 		ktrsysret(td->td_sa.code, 0, 0);
1231 #endif
1232 }
1233 
1234 static void
1235 fork_init(void *arg __unused)
1236 {
1237 	ast_register(TDA_VFORK, ASTR_ASTF_REQUIRED | ASTR_TDP, TDP_RFPPWAIT,
1238 	    ast_vfork);
1239 }
1240 SYSINIT(fork, SI_SUB_INTRINSIC, SI_ORDER_ANY, fork_init, NULL);
1241