xref: /freebsd/sys/kern/kern_fork.c (revision 2d223ccdaed67ad336474f023a661c300bd0ef38)
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 | FR2_KPROC)) != 0) {
501 			mtx_lock(&p2->p_sigacts->ps_mtx);
502 			if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
503 				sig_drop_caught(p2);
504 			if ((fr->fr_flags2 & FR2_KPROC) != 0)
505 				p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
506 			mtx_unlock(&p2->p_sigacts->ps_mtx);
507 		}
508 	}
509 
510 	if (fr->fr_flags & RFTSIGZMB)
511 	        p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
512 	else if (fr->fr_flags & RFLINUXTHPN)
513 	        p2->p_sigparent = SIGUSR1;
514 	else
515 	        p2->p_sigparent = SIGCHLD;
516 
517 	if ((fr->fr_flags2 & FR2_KPROC) != 0) {
518 		p2->p_flag |= P_SYSTEM | P_KPROC;
519 		td2->td_pflags |= TDP_KTHREAD;
520 	}
521 
522 	p2->p_textvp = p1->p_textvp;
523 	p2->p_fd = fd;
524 	p2->p_fdtol = fdtol;
525 	p2->p_pd = pd;
526 
527 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
528 		p2->p_flag |= P_PROTECTED;
529 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
530 	}
531 
532 	/*
533 	 * p_limit is copy-on-write.  Bump its refcount.
534 	 */
535 	lim_fork(p1, p2);
536 
537 	thread_cow_get_proc(td2, p2);
538 
539 	pstats_fork(p1->p_stats, p2->p_stats);
540 
541 	PROC_UNLOCK(p1);
542 	PROC_UNLOCK(p2);
543 
544 	/* Bump references to the text vnode (for procfs). */
545 	if (p2->p_textvp)
546 		vrefact(p2->p_textvp);
547 
548 	/*
549 	 * Set up linkage for kernel based threading.
550 	 */
551 	if ((fr->fr_flags & RFTHREAD) != 0) {
552 		mtx_lock(&ppeers_lock);
553 		p2->p_peers = p1->p_peers;
554 		p1->p_peers = p2;
555 		p2->p_leader = p1->p_leader;
556 		mtx_unlock(&ppeers_lock);
557 		PROC_LOCK(p1->p_leader);
558 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
559 			PROC_UNLOCK(p1->p_leader);
560 			/*
561 			 * The task leader is exiting, so process p1 is
562 			 * going to be killed shortly.  Since p1 obviously
563 			 * isn't dead yet, we know that the leader is either
564 			 * sending SIGKILL's to all the processes in this
565 			 * task or is sleeping waiting for all the peers to
566 			 * exit.  We let p1 complete the fork, but we need
567 			 * to go ahead and kill the new process p2 since
568 			 * the task leader may not get a chance to send
569 			 * SIGKILL to it.  We leave it on the list so that
570 			 * the task leader will wait for this new process
571 			 * to commit suicide.
572 			 */
573 			PROC_LOCK(p2);
574 			kern_psignal(p2, SIGKILL);
575 			PROC_UNLOCK(p2);
576 		} else
577 			PROC_UNLOCK(p1->p_leader);
578 	} else {
579 		p2->p_peers = NULL;
580 		p2->p_leader = p2;
581 	}
582 
583 	sx_xlock(&proctree_lock);
584 	PGRP_LOCK(p1->p_pgrp);
585 	PROC_LOCK(p2);
586 	PROC_LOCK(p1);
587 
588 	/*
589 	 * Preserve some more flags in subprocess.  P_PROFIL has already
590 	 * been preserved.
591 	 */
592 	p2->p_flag |= p1->p_flag & P_SUGID;
593 	td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK |
594 	    TDP_SIGFASTBLOCK)) | TDP_FORKING;
595 	SESS_LOCK(p1->p_session);
596 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
597 		p2->p_flag |= P_CONTROLT;
598 	SESS_UNLOCK(p1->p_session);
599 	if (fr->fr_flags & RFPPWAIT)
600 		p2->p_flag |= P_PPWAIT;
601 
602 	p2->p_pgrp = p1->p_pgrp;
603 	LIST_INSERT_AFTER(p1, p2, p_pglist);
604 	PGRP_UNLOCK(p1->p_pgrp);
605 	LIST_INIT(&p2->p_children);
606 	LIST_INIT(&p2->p_orphans);
607 
608 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
609 
610 	/*
611 	 * This begins the section where we must prevent the parent
612 	 * from being swapped.
613 	 */
614 	_PHOLD(p1);
615 	PROC_UNLOCK(p1);
616 
617 	/*
618 	 * Attach the new process to its parent.
619 	 *
620 	 * If RFNOWAIT is set, the newly created process becomes a child
621 	 * of init.  This effectively disassociates the child from the
622 	 * parent.
623 	 */
624 	if ((fr->fr_flags & RFNOWAIT) != 0) {
625 		pptr = p1->p_reaper;
626 		p2->p_reaper = pptr;
627 	} else {
628 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
629 		    p1 : p1->p_reaper;
630 		pptr = p1;
631 	}
632 	p2->p_pptr = pptr;
633 	p2->p_oppid = pptr->p_pid;
634 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
635 	LIST_INIT(&p2->p_reaplist);
636 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
637 	if (p2->p_reaper == p1 && p1 != initproc) {
638 		p2->p_reapsubtree = p2->p_pid;
639 		proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
640 	}
641 	sx_xunlock(&proctree_lock);
642 
643 	/* Inform accounting that we have forked. */
644 	p2->p_acflag = AFORK;
645 	PROC_UNLOCK(p2);
646 
647 #ifdef KTRACE
648 	ktrprocfork(p1, p2);
649 #endif
650 
651 	/*
652 	 * Finish creating the child process.  It will return via a different
653 	 * execution path later.  (ie: directly into user mode)
654 	 */
655 	vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
656 
657 	if (fr->fr_flags == (RFFDG | RFPROC)) {
658 		VM_CNT_INC(v_forks);
659 		VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
660 		    p2->p_vmspace->vm_ssize);
661 	} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
662 		VM_CNT_INC(v_vforks);
663 		VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
664 		    p2->p_vmspace->vm_ssize);
665 	} else if (p1 == &proc0) {
666 		VM_CNT_INC(v_kthreads);
667 		VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
668 		    p2->p_vmspace->vm_ssize);
669 	} else {
670 		VM_CNT_INC(v_rforks);
671 		VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
672 		    p2->p_vmspace->vm_ssize);
673 	}
674 
675 	/*
676 	 * Associate the process descriptor with the process before anything
677 	 * can happen that might cause that process to need the descriptor.
678 	 * However, don't do this until after fork(2) can no longer fail.
679 	 */
680 	if (fr->fr_flags & RFPROCDESC)
681 		procdesc_new(p2, fr->fr_pd_flags);
682 
683 	/*
684 	 * Both processes are set up, now check if any loadable modules want
685 	 * to adjust anything.
686 	 */
687 	EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
688 
689 	/*
690 	 * Set the child start time and mark the process as being complete.
691 	 */
692 	PROC_LOCK(p2);
693 	PROC_LOCK(p1);
694 	microuptime(&p2->p_stats->p_start);
695 	PROC_SLOCK(p2);
696 	p2->p_state = PRS_NORMAL;
697 	PROC_SUNLOCK(p2);
698 
699 #ifdef KDTRACE_HOOKS
700 	/*
701 	 * Tell the DTrace fasttrap provider about the new process so that any
702 	 * tracepoints inherited from the parent can be removed. We have to do
703 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
704 	 * use pfind() later on.
705 	 */
706 	if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
707 		dtrace_fasttrap_fork(p1, p2);
708 #endif
709 	if (fr->fr_flags & RFPPWAIT) {
710 		td->td_pflags |= TDP_RFPPWAIT;
711 		td->td_rfppwait_p = p2;
712 		td->td_dbgflags |= TDB_VFORK;
713 	}
714 	PROC_UNLOCK(p2);
715 
716 	/*
717 	 * Tell any interested parties about the new process.
718 	 */
719 	knote_fork(p1->p_klist, p2->p_pid);
720 
721 	/*
722 	 * Now can be swapped.
723 	 */
724 	_PRELE(p1);
725 	PROC_UNLOCK(p1);
726 	SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
727 
728 	if (fr->fr_flags & RFPROCDESC) {
729 		procdesc_finit(p2->p_procdesc, fp_procdesc);
730 		fdrop(fp_procdesc, td);
731 	}
732 
733 	/*
734 	 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
735 	 * synced with forks in progress so it is OK if we miss it
736 	 * if being set atm.
737 	 */
738 	if ((p1->p_ptevents & PTRACE_FORK) != 0) {
739 		sx_xlock(&proctree_lock);
740 		PROC_LOCK(p2);
741 
742 		/*
743 		 * p1->p_ptevents & p1->p_pptr are protected by both
744 		 * process and proctree locks for modifications,
745 		 * so owning proctree_lock allows the race-free read.
746 		 */
747 		if ((p1->p_ptevents & PTRACE_FORK) != 0) {
748 			/*
749 			 * Arrange for debugger to receive the fork event.
750 			 *
751 			 * We can report PL_FLAG_FORKED regardless of
752 			 * P_FOLLOWFORK settings, but it does not make a sense
753 			 * for runaway child.
754 			 */
755 			td->td_dbgflags |= TDB_FORK;
756 			td->td_dbg_forked = p2->p_pid;
757 			td2->td_dbgflags |= TDB_STOPATFORK;
758 			proc_set_traced(p2, true);
759 			CTR2(KTR_PTRACE,
760 			    "do_fork: attaching to new child pid %d: oppid %d",
761 			    p2->p_pid, p2->p_oppid);
762 			proc_reparent(p2, p1->p_pptr, false);
763 		}
764 		PROC_UNLOCK(p2);
765 		sx_xunlock(&proctree_lock);
766 	}
767 
768 	racct_proc_fork_done(p2);
769 
770 	if ((fr->fr_flags & RFSTOPPED) == 0) {
771 		if (fr->fr_pidp != NULL)
772 			*fr->fr_pidp = p2->p_pid;
773 		/*
774 		 * If RFSTOPPED not requested, make child runnable and
775 		 * add to run queue.
776 		 */
777 		thread_lock(td2);
778 		TD_SET_CAN_RUN(td2);
779 		sched_add(td2, SRQ_BORING);
780 	} else {
781 		*fr->fr_procp = p2;
782 	}
783 }
784 
785 void
786 fork_rfppwait(struct thread *td)
787 {
788 	struct proc *p, *p2;
789 
790 	MPASS(td->td_pflags & TDP_RFPPWAIT);
791 
792 	p = td->td_proc;
793 	/*
794 	 * Preserve synchronization semantics of vfork.  If
795 	 * waiting for child to exec or exit, fork set
796 	 * P_PPWAIT on child, and there we sleep on our proc
797 	 * (in case of exit).
798 	 *
799 	 * Do it after the ptracestop() above is finished, to
800 	 * not block our debugger until child execs or exits
801 	 * to finish vfork wait.
802 	 */
803 	td->td_pflags &= ~TDP_RFPPWAIT;
804 	p2 = td->td_rfppwait_p;
805 again:
806 	PROC_LOCK(p2);
807 	while (p2->p_flag & P_PPWAIT) {
808 		PROC_LOCK(p);
809 		if (thread_suspend_check_needed()) {
810 			PROC_UNLOCK(p2);
811 			thread_suspend_check(0);
812 			PROC_UNLOCK(p);
813 			goto again;
814 		} else {
815 			PROC_UNLOCK(p);
816 		}
817 		cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
818 	}
819 	PROC_UNLOCK(p2);
820 
821 	if (td->td_dbgflags & TDB_VFORK) {
822 		PROC_LOCK(p);
823 		if (p->p_ptevents & PTRACE_VFORK)
824 			ptracestop(td, SIGTRAP, NULL);
825 		td->td_dbgflags &= ~TDB_VFORK;
826 		PROC_UNLOCK(p);
827 	}
828 }
829 
830 int
831 fork1(struct thread *td, struct fork_req *fr)
832 {
833 	struct proc *p1, *newproc;
834 	struct thread *td2;
835 	struct vmspace *vm2;
836 	struct ucred *cred;
837 	struct file *fp_procdesc;
838 	vm_ooffset_t mem_charged;
839 	int error, nprocs_new;
840 	static int curfail;
841 	static struct timeval lastfail;
842 	int flags, pages;
843 
844 	flags = fr->fr_flags;
845 	pages = fr->fr_pages;
846 
847 	if ((flags & RFSTOPPED) != 0)
848 		MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
849 	else
850 		MPASS(fr->fr_procp == NULL);
851 
852 	/* Check for the undefined or unimplemented flags. */
853 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
854 		return (EINVAL);
855 
856 	/* Signal value requires RFTSIGZMB. */
857 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
858 		return (EINVAL);
859 
860 	/* Can't copy and clear. */
861 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
862 		return (EINVAL);
863 
864 	/* Check the validity of the signal number. */
865 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
866 		return (EINVAL);
867 
868 	if ((flags & RFPROCDESC) != 0) {
869 		/* Can't not create a process yet get a process descriptor. */
870 		if ((flags & RFPROC) == 0)
871 			return (EINVAL);
872 
873 		/* Must provide a place to put a procdesc if creating one. */
874 		if (fr->fr_pd_fd == NULL)
875 			return (EINVAL);
876 
877 		/* Check if we are using supported flags. */
878 		if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
879 			return (EINVAL);
880 	}
881 
882 	p1 = td->td_proc;
883 
884 	/*
885 	 * Here we don't create a new process, but we divorce
886 	 * certain parts of a process from itself.
887 	 */
888 	if ((flags & RFPROC) == 0) {
889 		if (fr->fr_procp != NULL)
890 			*fr->fr_procp = NULL;
891 		else if (fr->fr_pidp != NULL)
892 			*fr->fr_pidp = 0;
893 		return (fork_norfproc(td, flags));
894 	}
895 
896 	fp_procdesc = NULL;
897 	newproc = NULL;
898 	vm2 = NULL;
899 
900 	/*
901 	 * Increment the nprocs resource before allocations occur.
902 	 * Although process entries are dynamically created, we still
903 	 * keep a global limit on the maximum number we will
904 	 * create. There are hard-limits as to the number of processes
905 	 * that can run, established by the KVA and memory usage for
906 	 * the process data.
907 	 *
908 	 * Don't allow a nonprivileged user to use the last ten
909 	 * processes; don't let root exceed the limit.
910 	 */
911 	nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
912 	if (nprocs_new >= maxproc - 10) {
913 		if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
914 		    nprocs_new >= maxproc) {
915 			error = EAGAIN;
916 			sx_xlock(&allproc_lock);
917 			if (ppsratecheck(&lastfail, &curfail, 1)) {
918 				printf("maxproc limit exceeded by uid %u "
919 				    "(pid %d); see tuning(7) and "
920 				    "login.conf(5)\n",
921 				    td->td_ucred->cr_ruid, p1->p_pid);
922 			}
923 			sx_xunlock(&allproc_lock);
924 			goto fail2;
925 		}
926 	}
927 
928 	/*
929 	 * If required, create a process descriptor in the parent first; we
930 	 * will abandon it if something goes wrong. We don't finit() until
931 	 * later.
932 	 */
933 	if (flags & RFPROCDESC) {
934 		error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
935 		    fr->fr_pd_flags, fr->fr_pd_fcaps);
936 		if (error != 0)
937 			goto fail2;
938 		AUDIT_ARG_FD(*fr->fr_pd_fd);
939 	}
940 
941 	mem_charged = 0;
942 	if (pages == 0)
943 		pages = kstack_pages;
944 	/* Allocate new proc. */
945 	newproc = uma_zalloc(proc_zone, M_WAITOK);
946 	td2 = FIRST_THREAD_IN_PROC(newproc);
947 	if (td2 == NULL) {
948 		td2 = thread_alloc(pages);
949 		if (td2 == NULL) {
950 			error = ENOMEM;
951 			goto fail2;
952 		}
953 		proc_linkup(newproc, td2);
954 	} else {
955 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
956 			if (td2->td_kstack != 0)
957 				vm_thread_dispose(td2);
958 			if (!thread_alloc_stack(td2, pages)) {
959 				error = ENOMEM;
960 				goto fail2;
961 			}
962 		}
963 	}
964 
965 	if ((flags & RFMEM) == 0) {
966 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
967 		if (vm2 == NULL) {
968 			error = ENOMEM;
969 			goto fail2;
970 		}
971 		if (!swap_reserve(mem_charged)) {
972 			/*
973 			 * The swap reservation failed. The accounting
974 			 * from the entries of the copied vm2 will be
975 			 * subtracted in vmspace_free(), so force the
976 			 * reservation there.
977 			 */
978 			swap_reserve_force(mem_charged);
979 			error = ENOMEM;
980 			goto fail2;
981 		}
982 	} else
983 		vm2 = NULL;
984 
985 	/*
986 	 * XXX: This is ugly; when we copy resource usage, we need to bump
987 	 *      per-cred resource counters.
988 	 */
989 	proc_set_cred_init(newproc, td->td_ucred);
990 
991 	/*
992 	 * Initialize resource accounting for the child process.
993 	 */
994 	error = racct_proc_fork(p1, newproc);
995 	if (error != 0) {
996 		error = EAGAIN;
997 		goto fail1;
998 	}
999 
1000 #ifdef MAC
1001 	mac_proc_init(newproc);
1002 #endif
1003 	newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1004 	STAILQ_INIT(&newproc->p_ktr);
1005 
1006 	/*
1007 	 * Increment the count of procs running with this uid. Don't allow
1008 	 * a nonprivileged user to exceed their current limit.
1009 	 */
1010 	cred = td->td_ucred;
1011 	if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1012 		if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1013 			goto fail0;
1014 		chgproccnt(cred->cr_ruidinfo, 1, 0);
1015 	}
1016 
1017 	do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1018 	return (0);
1019 fail0:
1020 	error = EAGAIN;
1021 #ifdef MAC
1022 	mac_proc_destroy(newproc);
1023 #endif
1024 	racct_proc_exit(newproc);
1025 fail1:
1026 	proc_unset_cred(newproc);
1027 fail2:
1028 	if (vm2 != NULL)
1029 		vmspace_free(vm2);
1030 	uma_zfree(proc_zone, newproc);
1031 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1032 		fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1033 		fdrop(fp_procdesc, td);
1034 	}
1035 	atomic_add_int(&nprocs, -1);
1036 	pause("fork", hz / 2);
1037 	return (error);
1038 }
1039 
1040 /*
1041  * Handle the return of a child process from fork1().  This function
1042  * is called from the MD fork_trampoline() entry point.
1043  */
1044 void
1045 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1046     struct trapframe *frame)
1047 {
1048 	struct proc *p;
1049 	struct thread *td;
1050 	struct thread *dtd;
1051 
1052 	td = curthread;
1053 	p = td->td_proc;
1054 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1055 
1056 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1057 	    td, td_get_sched(td), p->p_pid, td->td_name);
1058 
1059 	sched_fork_exit(td);
1060 	/*
1061 	* Processes normally resume in mi_switch() after being
1062 	* cpu_switch()'ed to, but when children start up they arrive here
1063 	* instead, so we must do much the same things as mi_switch() would.
1064 	*/
1065 	if ((dtd = PCPU_GET(deadthread))) {
1066 		PCPU_SET(deadthread, NULL);
1067 		thread_stash(dtd);
1068 	}
1069 	thread_unlock(td);
1070 
1071 	/*
1072 	 * cpu_fork_kthread_handler intercepts this function call to
1073 	 * have this call a non-return function to stay in kernel mode.
1074 	 * initproc has its own fork handler, but it does return.
1075 	 */
1076 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1077 	callout(arg, frame);
1078 
1079 	/*
1080 	 * Check if a kernel thread misbehaved and returned from its main
1081 	 * function.
1082 	 */
1083 	if (p->p_flag & P_KPROC) {
1084 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1085 		    td->td_name, p->p_pid);
1086 		kthread_exit();
1087 	}
1088 	mtx_assert(&Giant, MA_NOTOWNED);
1089 
1090 	if (p->p_sysent->sv_schedtail != NULL)
1091 		(p->p_sysent->sv_schedtail)(td);
1092 	td->td_pflags &= ~TDP_FORKING;
1093 }
1094 
1095 /*
1096  * Simplified back end of syscall(), used when returning from fork()
1097  * directly into user mode.  This function is passed in to fork_exit()
1098  * as the first parameter and is called when returning to a new
1099  * userland process.
1100  */
1101 void
1102 fork_return(struct thread *td, struct trapframe *frame)
1103 {
1104 	struct proc *p;
1105 
1106 	p = td->td_proc;
1107 	if (td->td_dbgflags & TDB_STOPATFORK) {
1108 		PROC_LOCK(p);
1109 		if ((p->p_flag & P_TRACED) != 0) {
1110 			/*
1111 			 * Inform the debugger if one is still present.
1112 			 */
1113 			td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1114 			ptracestop(td, SIGSTOP, NULL);
1115 			td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1116 		} else {
1117 			/*
1118 			 * ... otherwise clear the request.
1119 			 */
1120 			td->td_dbgflags &= ~TDB_STOPATFORK;
1121 		}
1122 		PROC_UNLOCK(p);
1123 	} else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1124  		/*
1125 		 * This is the start of a new thread in a traced
1126 		 * process.  Report a system call exit event.
1127 		 */
1128 		PROC_LOCK(p);
1129 		td->td_dbgflags |= TDB_SCX;
1130 		if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1131 		    (td->td_dbgflags & TDB_BORN) != 0)
1132 			ptracestop(td, SIGTRAP, NULL);
1133 		td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1134 		PROC_UNLOCK(p);
1135 	}
1136 
1137 	/*
1138 	 * If the prison was killed mid-fork, die along with it.
1139 	 */
1140 	if (!prison_isalive(td->td_ucred->cr_prison))
1141 		exit1(td, 0, SIGKILL);
1142 
1143 	userret(td, frame);
1144 
1145 #ifdef KTRACE
1146 	if (KTRPOINT(td, KTR_SYSRET))
1147 		ktrsysret(SYS_fork, 0, 0);
1148 #endif
1149 }
1150