xref: /freebsd/sys/kern/kern_fork.c (revision 28f4385e45a2681c14bd04b83fe1796eaefe8265)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ktrace.h"
43 #include "opt_kstack_pages.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/bitstring.h>
48 #include <sys/sysproto.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
52 #include <sys/jail.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/sysctl.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mutex.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/procdesc.h>
62 #include <sys/pioctl.h>
63 #include <sys/ptrace.h>
64 #include <sys/racct.h>
65 #include <sys/resourcevar.h>
66 #include <sys/sched.h>
67 #include <sys/syscall.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
70 #include <sys/acct.h>
71 #include <sys/ktr.h>
72 #include <sys/ktrace.h>
73 #include <sys/unistd.h>
74 #include <sys/sdt.h>
75 #include <sys/sx.h>
76 #include <sys/sysent.h>
77 #include <sys/signalvar.h>
78 
79 #include <security/audit/audit.h>
80 #include <security/mac/mac_framework.h>
81 
82 #include <vm/vm.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_extern.h>
86 #include <vm/uma.h>
87 
88 #ifdef KDTRACE_HOOKS
89 #include <sys/dtrace_bsd.h>
90 dtrace_fork_func_t	dtrace_fasttrap_fork;
91 #endif
92 
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
95 
96 #ifndef _SYS_SYSPROTO_H_
97 struct fork_args {
98 	int     dummy;
99 };
100 #endif
101 
102 EVENTHANDLER_LIST_DECLARE(process_fork);
103 
104 /* ARGSUSED */
105 int
106 sys_fork(struct thread *td, struct fork_args *uap)
107 {
108 	struct fork_req fr;
109 	int error, pid;
110 
111 	bzero(&fr, sizeof(fr));
112 	fr.fr_flags = RFFDG | RFPROC;
113 	fr.fr_pidp = &pid;
114 	error = fork1(td, &fr);
115 	if (error == 0) {
116 		td->td_retval[0] = pid;
117 		td->td_retval[1] = 0;
118 	}
119 	return (error);
120 }
121 
122 /* ARGUSED */
123 int
124 sys_pdfork(struct thread *td, struct pdfork_args *uap)
125 {
126 	struct fork_req fr;
127 	int error, fd, pid;
128 
129 	bzero(&fr, sizeof(fr));
130 	fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
131 	fr.fr_pidp = &pid;
132 	fr.fr_pd_fd = &fd;
133 	fr.fr_pd_flags = uap->flags;
134 	/*
135 	 * It is necessary to return fd by reference because 0 is a valid file
136 	 * descriptor number, and the child needs to be able to distinguish
137 	 * itself from the parent using the return value.
138 	 */
139 	error = fork1(td, &fr);
140 	if (error == 0) {
141 		td->td_retval[0] = pid;
142 		td->td_retval[1] = 0;
143 		error = copyout(&fd, uap->fdp, sizeof(fd));
144 	}
145 	return (error);
146 }
147 
148 /* ARGSUSED */
149 int
150 sys_vfork(struct thread *td, struct vfork_args *uap)
151 {
152 	struct fork_req fr;
153 	int error, pid;
154 
155 	bzero(&fr, sizeof(fr));
156 	fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
157 	fr.fr_pidp = &pid;
158 	error = fork1(td, &fr);
159 	if (error == 0) {
160 		td->td_retval[0] = pid;
161 		td->td_retval[1] = 0;
162 	}
163 	return (error);
164 }
165 
166 int
167 sys_rfork(struct thread *td, struct rfork_args *uap)
168 {
169 	struct fork_req fr;
170 	int error, pid;
171 
172 	/* Don't allow kernel-only flags. */
173 	if ((uap->flags & RFKERNELONLY) != 0)
174 		return (EINVAL);
175 
176 	AUDIT_ARG_FFLAGS(uap->flags);
177 	bzero(&fr, sizeof(fr));
178 	fr.fr_flags = uap->flags;
179 	fr.fr_pidp = &pid;
180 	error = fork1(td, &fr);
181 	if (error == 0) {
182 		td->td_retval[0] = pid;
183 		td->td_retval[1] = 0;
184 	}
185 	return (error);
186 }
187 
188 int	nprocs = 1;		/* process 0 */
189 int	lastpid = 0;
190 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
191     "Last used PID");
192 
193 /*
194  * Random component to lastpid generation.  We mix in a random factor to make
195  * it a little harder to predict.  We sanity check the modulus value to avoid
196  * doing it in critical paths.  Don't let it be too small or we pointlessly
197  * waste randomness entropy, and don't let it be impossibly large.  Using a
198  * modulus that is too big causes a LOT more process table scans and slows
199  * down fork processing as the pidchecked caching is defeated.
200  */
201 static int randompid = 0;
202 
203 static int
204 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
205 {
206 	int error, pid;
207 
208 	error = sysctl_wire_old_buffer(req, sizeof(int));
209 	if (error != 0)
210 		return(error);
211 	sx_xlock(&allproc_lock);
212 	pid = randompid;
213 	error = sysctl_handle_int(oidp, &pid, 0, req);
214 	if (error == 0 && req->newptr != NULL) {
215 		if (pid == 0)
216 			randompid = 0;
217 		else if (pid == 1)
218 			/* generate a random PID modulus between 100 and 1123 */
219 			randompid = 100 + arc4random() % 1024;
220 		else if (pid < 0 || pid > pid_max - 100)
221 			/* out of range */
222 			randompid = pid_max - 100;
223 		else if (pid < 100)
224 			/* Make it reasonable */
225 			randompid = 100;
226 		else
227 			randompid = pid;
228 	}
229 	sx_xunlock(&allproc_lock);
230 	return (error);
231 }
232 
233 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
234     0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value");
235 
236 extern bitstr_t proc_id_pidmap;
237 extern bitstr_t proc_id_grpidmap;
238 extern bitstr_t proc_id_sessidmap;
239 extern bitstr_t proc_id_reapmap;
240 
241 /*
242  * Find an unused process ID
243  *
244  * If RFHIGHPID is set (used during system boot), do not allocate
245  * low-numbered pids.
246  */
247 static int
248 fork_findpid(int flags)
249 {
250 	pid_t result;
251 	int trypid;
252 
253 	trypid = lastpid + 1;
254 	if (flags & RFHIGHPID) {
255 		if (trypid < 10)
256 			trypid = 10;
257 	} else {
258 		if (randompid)
259 			trypid += arc4random() % randompid;
260 	}
261 	mtx_lock(&procid_lock);
262 retry:
263 	/*
264 	 * If the process ID prototype has wrapped around,
265 	 * restart somewhat above 0, as the low-numbered procs
266 	 * tend to include daemons that don't exit.
267 	 */
268 	if (trypid >= pid_max) {
269 		trypid = trypid % pid_max;
270 		if (trypid < 100)
271 			trypid += 100;
272 	}
273 
274 	bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
275 	if (result == -1) {
276 		trypid = 100;
277 		goto retry;
278 	}
279 	if (bit_test(&proc_id_grpidmap, result) ||
280 	    bit_test(&proc_id_sessidmap, result) ||
281 	    bit_test(&proc_id_reapmap, result)) {
282 		trypid = result + 1;
283 		goto retry;
284 	}
285 
286 	/*
287 	 * RFHIGHPID does not mess with the lastpid counter during boot.
288 	 */
289 	if ((flags & RFHIGHPID) == 0)
290 		lastpid = result;
291 
292 	bit_set(&proc_id_pidmap, result);
293 	mtx_unlock(&procid_lock);
294 
295 	return (result);
296 }
297 
298 static int
299 fork_norfproc(struct thread *td, int flags)
300 {
301 	int error;
302 	struct proc *p1;
303 
304 	KASSERT((flags & RFPROC) == 0,
305 	    ("fork_norfproc called with RFPROC set"));
306 	p1 = td->td_proc;
307 
308 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
309 	    (flags & (RFCFDG | RFFDG))) {
310 		PROC_LOCK(p1);
311 		if (thread_single(p1, SINGLE_BOUNDARY)) {
312 			PROC_UNLOCK(p1);
313 			return (ERESTART);
314 		}
315 		PROC_UNLOCK(p1);
316 	}
317 
318 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
319 	if (error)
320 		goto fail;
321 
322 	/*
323 	 * Close all file descriptors.
324 	 */
325 	if (flags & RFCFDG) {
326 		struct filedesc *fdtmp;
327 		fdtmp = fdinit(td->td_proc->p_fd, false);
328 		fdescfree(td);
329 		p1->p_fd = fdtmp;
330 	}
331 
332 	/*
333 	 * Unshare file descriptors (from parent).
334 	 */
335 	if (flags & RFFDG)
336 		fdunshare(td);
337 
338 fail:
339 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
340 	    (flags & (RFCFDG | RFFDG))) {
341 		PROC_LOCK(p1);
342 		thread_single_end(p1, SINGLE_BOUNDARY);
343 		PROC_UNLOCK(p1);
344 	}
345 	return (error);
346 }
347 
348 static void
349 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
350     struct vmspace *vm2, struct file *fp_procdesc)
351 {
352 	struct proc *p1, *pptr;
353 	int trypid;
354 	struct filedesc *fd;
355 	struct filedesc_to_leader *fdtol;
356 	struct sigacts *newsigacts;
357 
358 	sx_assert(&allproc_lock, SX_XLOCKED);
359 
360 	p1 = td->td_proc;
361 
362 	trypid = fork_findpid(fr->fr_flags);
363 	p2->p_state = PRS_NEW;		/* protect against others */
364 	p2->p_pid = trypid;
365 	AUDIT_ARG_PID(p2->p_pid);
366 	LIST_INSERT_HEAD(&allproc, p2, p_list);
367 	allproc_gen++;
368 	sx_xlock(PIDHASHLOCK(p2->p_pid));
369 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
370 	sx_xunlock(PIDHASHLOCK(p2->p_pid));
371 	PROC_LOCK(p2);
372 	PROC_LOCK(p1);
373 
374 	sx_xunlock(&allproc_lock);
375 
376 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
377 	    __rangeof(struct proc, p_startcopy, p_endcopy));
378 	pargs_hold(p2->p_args);
379 
380 	PROC_UNLOCK(p1);
381 
382 	bzero(&p2->p_startzero,
383 	    __rangeof(struct proc, p_startzero, p_endzero));
384 
385 	/* Tell the prison that we exist. */
386 	prison_proc_hold(p2->p_ucred->cr_prison);
387 
388 	PROC_UNLOCK(p2);
389 
390 	tidhash_add(td2);
391 
392 	/*
393 	 * Malloc things while we don't hold any locks.
394 	 */
395 	if (fr->fr_flags & RFSIGSHARE)
396 		newsigacts = NULL;
397 	else
398 		newsigacts = sigacts_alloc();
399 
400 	/*
401 	 * Copy filedesc.
402 	 */
403 	if (fr->fr_flags & RFCFDG) {
404 		fd = fdinit(p1->p_fd, false);
405 		fdtol = NULL;
406 	} else if (fr->fr_flags & RFFDG) {
407 		fd = fdcopy(p1->p_fd);
408 		fdtol = NULL;
409 	} else {
410 		fd = fdshare(p1->p_fd);
411 		if (p1->p_fdtol == NULL)
412 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
413 			    p1->p_leader);
414 		if ((fr->fr_flags & RFTHREAD) != 0) {
415 			/*
416 			 * Shared file descriptor table, and shared
417 			 * process leaders.
418 			 */
419 			fdtol = p1->p_fdtol;
420 			FILEDESC_XLOCK(p1->p_fd);
421 			fdtol->fdl_refcount++;
422 			FILEDESC_XUNLOCK(p1->p_fd);
423 		} else {
424 			/*
425 			 * Shared file descriptor table, and different
426 			 * process leaders.
427 			 */
428 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
429 			    p1->p_fd, p2);
430 		}
431 	}
432 	/*
433 	 * Make a proc table entry for the new process.
434 	 * Start by zeroing the section of proc that is zero-initialized,
435 	 * then copy the section that is copied directly from the parent.
436 	 */
437 
438 	PROC_LOCK(p2);
439 	PROC_LOCK(p1);
440 
441 	bzero(&td2->td_startzero,
442 	    __rangeof(struct thread, td_startzero, td_endzero));
443 
444 	bcopy(&td->td_startcopy, &td2->td_startcopy,
445 	    __rangeof(struct thread, td_startcopy, td_endcopy));
446 
447 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
448 	td2->td_sigstk = td->td_sigstk;
449 	td2->td_flags = TDF_INMEM;
450 	td2->td_lend_user_pri = PRI_MAX;
451 
452 #ifdef VIMAGE
453 	td2->td_vnet = NULL;
454 	td2->td_vnet_lpush = NULL;
455 #endif
456 
457 	/*
458 	 * Allow the scheduler to initialize the child.
459 	 */
460 	thread_lock(td);
461 	sched_fork(td, td2);
462 	thread_unlock(td);
463 
464 	/*
465 	 * Duplicate sub-structures as needed.
466 	 * Increase reference counts on shared objects.
467 	 */
468 	p2->p_flag = P_INMEM;
469 	p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP);
470 	p2->p_swtick = ticks;
471 	if (p1->p_flag & P_PROFIL)
472 		startprofclock(p2);
473 
474 	if (fr->fr_flags & RFSIGSHARE) {
475 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
476 	} else {
477 		sigacts_copy(newsigacts, p1->p_sigacts);
478 		p2->p_sigacts = newsigacts;
479 	}
480 
481 	if (fr->fr_flags & RFTSIGZMB)
482 	        p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
483 	else if (fr->fr_flags & RFLINUXTHPN)
484 	        p2->p_sigparent = SIGUSR1;
485 	else
486 	        p2->p_sigparent = SIGCHLD;
487 
488 	p2->p_textvp = p1->p_textvp;
489 	p2->p_fd = fd;
490 	p2->p_fdtol = fdtol;
491 
492 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
493 		p2->p_flag |= P_PROTECTED;
494 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
495 	}
496 
497 	/*
498 	 * p_limit is copy-on-write.  Bump its refcount.
499 	 */
500 	lim_fork(p1, p2);
501 
502 	thread_cow_get_proc(td2, p2);
503 
504 	pstats_fork(p1->p_stats, p2->p_stats);
505 
506 	PROC_UNLOCK(p1);
507 	PROC_UNLOCK(p2);
508 
509 	/* Bump references to the text vnode (for procfs). */
510 	if (p2->p_textvp)
511 		vrefact(p2->p_textvp);
512 
513 	/*
514 	 * Set up linkage for kernel based threading.
515 	 */
516 	if ((fr->fr_flags & RFTHREAD) != 0) {
517 		mtx_lock(&ppeers_lock);
518 		p2->p_peers = p1->p_peers;
519 		p1->p_peers = p2;
520 		p2->p_leader = p1->p_leader;
521 		mtx_unlock(&ppeers_lock);
522 		PROC_LOCK(p1->p_leader);
523 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
524 			PROC_UNLOCK(p1->p_leader);
525 			/*
526 			 * The task leader is exiting, so process p1 is
527 			 * going to be killed shortly.  Since p1 obviously
528 			 * isn't dead yet, we know that the leader is either
529 			 * sending SIGKILL's to all the processes in this
530 			 * task or is sleeping waiting for all the peers to
531 			 * exit.  We let p1 complete the fork, but we need
532 			 * to go ahead and kill the new process p2 since
533 			 * the task leader may not get a chance to send
534 			 * SIGKILL to it.  We leave it on the list so that
535 			 * the task leader will wait for this new process
536 			 * to commit suicide.
537 			 */
538 			PROC_LOCK(p2);
539 			kern_psignal(p2, SIGKILL);
540 			PROC_UNLOCK(p2);
541 		} else
542 			PROC_UNLOCK(p1->p_leader);
543 	} else {
544 		p2->p_peers = NULL;
545 		p2->p_leader = p2;
546 	}
547 
548 	sx_xlock(&proctree_lock);
549 	PGRP_LOCK(p1->p_pgrp);
550 	PROC_LOCK(p2);
551 	PROC_LOCK(p1);
552 
553 	/*
554 	 * Preserve some more flags in subprocess.  P_PROFIL has already
555 	 * been preserved.
556 	 */
557 	p2->p_flag |= p1->p_flag & P_SUGID;
558 	td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
559 	SESS_LOCK(p1->p_session);
560 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
561 		p2->p_flag |= P_CONTROLT;
562 	SESS_UNLOCK(p1->p_session);
563 	if (fr->fr_flags & RFPPWAIT)
564 		p2->p_flag |= P_PPWAIT;
565 
566 	p2->p_pgrp = p1->p_pgrp;
567 	LIST_INSERT_AFTER(p1, p2, p_pglist);
568 	PGRP_UNLOCK(p1->p_pgrp);
569 	LIST_INIT(&p2->p_children);
570 	LIST_INIT(&p2->p_orphans);
571 
572 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
573 
574 	/*
575 	 * If PF_FORK is set, the child process inherits the
576 	 * procfs ioctl flags from its parent.
577 	 */
578 	if (p1->p_pfsflags & PF_FORK) {
579 		p2->p_stops = p1->p_stops;
580 		p2->p_pfsflags = p1->p_pfsflags;
581 	}
582 
583 	/*
584 	 * This begins the section where we must prevent the parent
585 	 * from being swapped.
586 	 */
587 	_PHOLD(p1);
588 	PROC_UNLOCK(p1);
589 
590 	/*
591 	 * Attach the new process to its parent.
592 	 *
593 	 * If RFNOWAIT is set, the newly created process becomes a child
594 	 * of init.  This effectively disassociates the child from the
595 	 * parent.
596 	 */
597 	if ((fr->fr_flags & RFNOWAIT) != 0) {
598 		pptr = p1->p_reaper;
599 		p2->p_reaper = pptr;
600 	} else {
601 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
602 		    p1 : p1->p_reaper;
603 		pptr = p1;
604 	}
605 	p2->p_pptr = pptr;
606 	p2->p_oppid = pptr->p_pid;
607 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
608 	LIST_INIT(&p2->p_reaplist);
609 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
610 	if (p2->p_reaper == p1 && p1 != initproc) {
611 		p2->p_reapsubtree = p2->p_pid;
612 		proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
613 	}
614 	sx_xunlock(&proctree_lock);
615 
616 	/* Inform accounting that we have forked. */
617 	p2->p_acflag = AFORK;
618 	PROC_UNLOCK(p2);
619 
620 #ifdef KTRACE
621 	ktrprocfork(p1, p2);
622 #endif
623 
624 	/*
625 	 * Finish creating the child process.  It will return via a different
626 	 * execution path later.  (ie: directly into user mode)
627 	 */
628 	vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
629 
630 	if (fr->fr_flags == (RFFDG | RFPROC)) {
631 		VM_CNT_INC(v_forks);
632 		VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
633 		    p2->p_vmspace->vm_ssize);
634 	} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
635 		VM_CNT_INC(v_vforks);
636 		VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
637 		    p2->p_vmspace->vm_ssize);
638 	} else if (p1 == &proc0) {
639 		VM_CNT_INC(v_kthreads);
640 		VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
641 		    p2->p_vmspace->vm_ssize);
642 	} else {
643 		VM_CNT_INC(v_rforks);
644 		VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
645 		    p2->p_vmspace->vm_ssize);
646 	}
647 
648 	/*
649 	 * Associate the process descriptor with the process before anything
650 	 * can happen that might cause that process to need the descriptor.
651 	 * However, don't do this until after fork(2) can no longer fail.
652 	 */
653 	if (fr->fr_flags & RFPROCDESC)
654 		procdesc_new(p2, fr->fr_pd_flags);
655 
656 	/*
657 	 * Both processes are set up, now check if any loadable modules want
658 	 * to adjust anything.
659 	 */
660 	EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
661 
662 	/*
663 	 * Set the child start time and mark the process as being complete.
664 	 */
665 	PROC_LOCK(p2);
666 	PROC_LOCK(p1);
667 	microuptime(&p2->p_stats->p_start);
668 	PROC_SLOCK(p2);
669 	p2->p_state = PRS_NORMAL;
670 	PROC_SUNLOCK(p2);
671 
672 #ifdef KDTRACE_HOOKS
673 	/*
674 	 * Tell the DTrace fasttrap provider about the new process so that any
675 	 * tracepoints inherited from the parent can be removed. We have to do
676 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
677 	 * use pfind() later on.
678 	 */
679 	if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
680 		dtrace_fasttrap_fork(p1, p2);
681 #endif
682 	if (fr->fr_flags & RFPPWAIT) {
683 		td->td_pflags |= TDP_RFPPWAIT;
684 		td->td_rfppwait_p = p2;
685 		td->td_dbgflags |= TDB_VFORK;
686 	}
687 	PROC_UNLOCK(p2);
688 
689 	/*
690 	 * Tell any interested parties about the new process.
691 	 */
692 	knote_fork(p1->p_klist, p2->p_pid);
693 
694 	/*
695 	 * Now can be swapped.
696 	 */
697 	_PRELE(p1);
698 	PROC_UNLOCK(p1);
699 	SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
700 
701 	if (fr->fr_flags & RFPROCDESC) {
702 		procdesc_finit(p2->p_procdesc, fp_procdesc);
703 		fdrop(fp_procdesc, td);
704 	}
705 
706 	/*
707 	 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
708 	 * synced with forks in progress so it is OK if we miss it
709 	 * if being set atm.
710 	 */
711 	if ((p1->p_ptevents & PTRACE_FORK) != 0) {
712 		sx_xlock(&proctree_lock);
713 		PROC_LOCK(p2);
714 
715 		/*
716 		 * p1->p_ptevents & p1->p_pptr are protected by both
717 		 * process and proctree locks for modifications,
718 		 * so owning proctree_lock allows the race-free read.
719 		 */
720 		if ((p1->p_ptevents & PTRACE_FORK) != 0) {
721 			/*
722 			 * Arrange for debugger to receive the fork event.
723 			 *
724 			 * We can report PL_FLAG_FORKED regardless of
725 			 * P_FOLLOWFORK settings, but it does not make a sense
726 			 * for runaway child.
727 			 */
728 			td->td_dbgflags |= TDB_FORK;
729 			td->td_dbg_forked = p2->p_pid;
730 			td2->td_dbgflags |= TDB_STOPATFORK;
731 			proc_set_traced(p2, true);
732 			CTR2(KTR_PTRACE,
733 			    "do_fork: attaching to new child pid %d: oppid %d",
734 			    p2->p_pid, p2->p_oppid);
735 			proc_reparent(p2, p1->p_pptr, false);
736 		}
737 		PROC_UNLOCK(p2);
738 		sx_xunlock(&proctree_lock);
739 	}
740 
741 	racct_proc_fork_done(p2);
742 
743 	if ((fr->fr_flags & RFSTOPPED) == 0) {
744 		if (fr->fr_pidp != NULL)
745 			*fr->fr_pidp = p2->p_pid;
746 		/*
747 		 * If RFSTOPPED not requested, make child runnable and
748 		 * add to run queue.
749 		 */
750 		thread_lock(td2);
751 		TD_SET_CAN_RUN(td2);
752 		sched_add(td2, SRQ_BORING);
753 		thread_unlock(td2);
754 	} else {
755 		*fr->fr_procp = p2;
756 	}
757 }
758 
759 void
760 fork_rfppwait(struct thread *td)
761 {
762 	struct proc *p, *p2;
763 
764 	MPASS(td->td_pflags & TDP_RFPPWAIT);
765 
766 	p = td->td_proc;
767 	/*
768 	 * Preserve synchronization semantics of vfork.  If
769 	 * waiting for child to exec or exit, fork set
770 	 * P_PPWAIT on child, and there we sleep on our proc
771 	 * (in case of exit).
772 	 *
773 	 * Do it after the ptracestop() above is finished, to
774 	 * not block our debugger until child execs or exits
775 	 * to finish vfork wait.
776 	 */
777 	td->td_pflags &= ~TDP_RFPPWAIT;
778 	p2 = td->td_rfppwait_p;
779 again:
780 	PROC_LOCK(p2);
781 	while (p2->p_flag & P_PPWAIT) {
782 		PROC_LOCK(p);
783 		if (thread_suspend_check_needed()) {
784 			PROC_UNLOCK(p2);
785 			thread_suspend_check(0);
786 			PROC_UNLOCK(p);
787 			goto again;
788 		} else {
789 			PROC_UNLOCK(p);
790 		}
791 		cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
792 	}
793 	PROC_UNLOCK(p2);
794 
795 	if (td->td_dbgflags & TDB_VFORK) {
796 		PROC_LOCK(p);
797 		if (p->p_ptevents & PTRACE_VFORK)
798 			ptracestop(td, SIGTRAP, NULL);
799 		td->td_dbgflags &= ~TDB_VFORK;
800 		PROC_UNLOCK(p);
801 	}
802 }
803 
804 int
805 fork1(struct thread *td, struct fork_req *fr)
806 {
807 	struct proc *p1, *newproc;
808 	struct thread *td2;
809 	struct vmspace *vm2;
810 	struct file *fp_procdesc;
811 	vm_ooffset_t mem_charged;
812 	int error, nprocs_new, ok;
813 	static int curfail;
814 	static struct timeval lastfail;
815 	int flags, pages;
816 
817 	flags = fr->fr_flags;
818 	pages = fr->fr_pages;
819 
820 	if ((flags & RFSTOPPED) != 0)
821 		MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
822 	else
823 		MPASS(fr->fr_procp == NULL);
824 
825 	/* Check for the undefined or unimplemented flags. */
826 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
827 		return (EINVAL);
828 
829 	/* Signal value requires RFTSIGZMB. */
830 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
831 		return (EINVAL);
832 
833 	/* Can't copy and clear. */
834 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
835 		return (EINVAL);
836 
837 	/* Check the validity of the signal number. */
838 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
839 		return (EINVAL);
840 
841 	if ((flags & RFPROCDESC) != 0) {
842 		/* Can't not create a process yet get a process descriptor. */
843 		if ((flags & RFPROC) == 0)
844 			return (EINVAL);
845 
846 		/* Must provide a place to put a procdesc if creating one. */
847 		if (fr->fr_pd_fd == NULL)
848 			return (EINVAL);
849 
850 		/* Check if we are using supported flags. */
851 		if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
852 			return (EINVAL);
853 	}
854 
855 	p1 = td->td_proc;
856 
857 	/*
858 	 * Here we don't create a new process, but we divorce
859 	 * certain parts of a process from itself.
860 	 */
861 	if ((flags & RFPROC) == 0) {
862 		if (fr->fr_procp != NULL)
863 			*fr->fr_procp = NULL;
864 		else if (fr->fr_pidp != NULL)
865 			*fr->fr_pidp = 0;
866 		return (fork_norfproc(td, flags));
867 	}
868 
869 	fp_procdesc = NULL;
870 	newproc = NULL;
871 	vm2 = NULL;
872 
873 	/*
874 	 * Increment the nprocs resource before allocations occur.
875 	 * Although process entries are dynamically created, we still
876 	 * keep a global limit on the maximum number we will
877 	 * create. There are hard-limits as to the number of processes
878 	 * that can run, established by the KVA and memory usage for
879 	 * the process data.
880 	 *
881 	 * Don't allow a nonprivileged user to use the last ten
882 	 * processes; don't let root exceed the limit.
883 	 */
884 	nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
885 	if ((nprocs_new >= maxproc - 10 &&
886 	    priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0) ||
887 	    nprocs_new >= maxproc) {
888 		error = EAGAIN;
889 		sx_xlock(&allproc_lock);
890 		if (ppsratecheck(&lastfail, &curfail, 1)) {
891 			printf("maxproc limit exceeded by uid %u (pid %d); "
892 			    "see tuning(7) and login.conf(5)\n",
893 			    td->td_ucred->cr_ruid, p1->p_pid);
894 		}
895 		sx_xunlock(&allproc_lock);
896 		goto fail2;
897 	}
898 
899 	/*
900 	 * If required, create a process descriptor in the parent first; we
901 	 * will abandon it if something goes wrong. We don't finit() until
902 	 * later.
903 	 */
904 	if (flags & RFPROCDESC) {
905 		error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
906 		    fr->fr_pd_flags, fr->fr_pd_fcaps);
907 		if (error != 0)
908 			goto fail2;
909 	}
910 
911 	mem_charged = 0;
912 	if (pages == 0)
913 		pages = kstack_pages;
914 	/* Allocate new proc. */
915 	newproc = uma_zalloc(proc_zone, M_WAITOK);
916 	td2 = FIRST_THREAD_IN_PROC(newproc);
917 	if (td2 == NULL) {
918 		td2 = thread_alloc(pages);
919 		if (td2 == NULL) {
920 			error = ENOMEM;
921 			goto fail2;
922 		}
923 		proc_linkup(newproc, td2);
924 	} else {
925 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
926 			if (td2->td_kstack != 0)
927 				vm_thread_dispose(td2);
928 			if (!thread_alloc_stack(td2, pages)) {
929 				error = ENOMEM;
930 				goto fail2;
931 			}
932 		}
933 	}
934 
935 	if ((flags & RFMEM) == 0) {
936 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
937 		if (vm2 == NULL) {
938 			error = ENOMEM;
939 			goto fail2;
940 		}
941 		if (!swap_reserve(mem_charged)) {
942 			/*
943 			 * The swap reservation failed. The accounting
944 			 * from the entries of the copied vm2 will be
945 			 * subtracted in vmspace_free(), so force the
946 			 * reservation there.
947 			 */
948 			swap_reserve_force(mem_charged);
949 			error = ENOMEM;
950 			goto fail2;
951 		}
952 	} else
953 		vm2 = NULL;
954 
955 	/*
956 	 * XXX: This is ugly; when we copy resource usage, we need to bump
957 	 *      per-cred resource counters.
958 	 */
959 	proc_set_cred_init(newproc, crhold(td->td_ucred));
960 
961 	/*
962 	 * Initialize resource accounting for the child process.
963 	 */
964 	error = racct_proc_fork(p1, newproc);
965 	if (error != 0) {
966 		error = EAGAIN;
967 		goto fail1;
968 	}
969 
970 #ifdef MAC
971 	mac_proc_init(newproc);
972 #endif
973 	newproc->p_klist = knlist_alloc(&newproc->p_mtx);
974 	STAILQ_INIT(&newproc->p_ktr);
975 
976 	sx_xlock(&allproc_lock);
977 
978 	/*
979 	 * Increment the count of procs running with this uid. Don't allow
980 	 * a nonprivileged user to exceed their current limit.
981 	 *
982 	 * XXXRW: Can we avoid privilege here if it's not needed?
983 	 */
984 	error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT);
985 	if (error == 0)
986 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
987 	else {
988 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
989 		    lim_cur(td, RLIMIT_NPROC));
990 	}
991 	if (ok) {
992 		do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
993 		return (0);
994 	}
995 
996 	error = EAGAIN;
997 	sx_xunlock(&allproc_lock);
998 #ifdef MAC
999 	mac_proc_destroy(newproc);
1000 #endif
1001 	racct_proc_exit(newproc);
1002 fail1:
1003 	crfree(newproc->p_ucred);
1004 	newproc->p_ucred = NULL;
1005 fail2:
1006 	if (vm2 != NULL)
1007 		vmspace_free(vm2);
1008 	uma_zfree(proc_zone, newproc);
1009 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1010 		fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1011 		fdrop(fp_procdesc, td);
1012 	}
1013 	atomic_add_int(&nprocs, -1);
1014 	pause("fork", hz / 2);
1015 	return (error);
1016 }
1017 
1018 /*
1019  * Handle the return of a child process from fork1().  This function
1020  * is called from the MD fork_trampoline() entry point.
1021  */
1022 void
1023 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1024     struct trapframe *frame)
1025 {
1026 	struct proc *p;
1027 	struct thread *td;
1028 	struct thread *dtd;
1029 
1030 	td = curthread;
1031 	p = td->td_proc;
1032 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1033 
1034 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1035 	    td, td_get_sched(td), p->p_pid, td->td_name);
1036 
1037 	sched_fork_exit(td);
1038 	/*
1039 	* Processes normally resume in mi_switch() after being
1040 	* cpu_switch()'ed to, but when children start up they arrive here
1041 	* instead, so we must do much the same things as mi_switch() would.
1042 	*/
1043 	if ((dtd = PCPU_GET(deadthread))) {
1044 		PCPU_SET(deadthread, NULL);
1045 		thread_stash(dtd);
1046 	}
1047 	thread_unlock(td);
1048 
1049 	/*
1050 	 * cpu_fork_kthread_handler intercepts this function call to
1051 	 * have this call a non-return function to stay in kernel mode.
1052 	 * initproc has its own fork handler, but it does return.
1053 	 */
1054 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1055 	callout(arg, frame);
1056 
1057 	/*
1058 	 * Check if a kernel thread misbehaved and returned from its main
1059 	 * function.
1060 	 */
1061 	if (p->p_flag & P_KPROC) {
1062 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1063 		    td->td_name, p->p_pid);
1064 		kthread_exit();
1065 	}
1066 	mtx_assert(&Giant, MA_NOTOWNED);
1067 
1068 	if (p->p_sysent->sv_schedtail != NULL)
1069 		(p->p_sysent->sv_schedtail)(td);
1070 	td->td_pflags &= ~TDP_FORKING;
1071 }
1072 
1073 /*
1074  * Simplified back end of syscall(), used when returning from fork()
1075  * directly into user mode.  This function is passed in to fork_exit()
1076  * as the first parameter and is called when returning to a new
1077  * userland process.
1078  */
1079 void
1080 fork_return(struct thread *td, struct trapframe *frame)
1081 {
1082 	struct proc *p;
1083 
1084 	p = td->td_proc;
1085 	if (td->td_dbgflags & TDB_STOPATFORK) {
1086 		PROC_LOCK(p);
1087 		if ((p->p_flag & P_TRACED) != 0) {
1088 			/*
1089 			 * Inform the debugger if one is still present.
1090 			 */
1091 			td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1092 			ptracestop(td, SIGSTOP, NULL);
1093 			td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1094 		} else {
1095 			/*
1096 			 * ... otherwise clear the request.
1097 			 */
1098 			td->td_dbgflags &= ~TDB_STOPATFORK;
1099 		}
1100 		PROC_UNLOCK(p);
1101 	} else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1102  		/*
1103 		 * This is the start of a new thread in a traced
1104 		 * process.  Report a system call exit event.
1105 		 */
1106 		PROC_LOCK(p);
1107 		td->td_dbgflags |= TDB_SCX;
1108 		_STOPEVENT(p, S_SCX, td->td_sa.code);
1109 		if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1110 		    (td->td_dbgflags & TDB_BORN) != 0)
1111 			ptracestop(td, SIGTRAP, NULL);
1112 		td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1113 		PROC_UNLOCK(p);
1114 	}
1115 
1116 	userret(td, frame);
1117 
1118 #ifdef KTRACE
1119 	if (KTRPOINT(td, KTR_SYSRET))
1120 		ktrsysret(SYS_fork, 0, 0);
1121 #endif
1122 }
1123