xref: /freebsd/sys/kern/kern_fork.c (revision 19fe57fdb4fd2c18a37f2a972617c8769609cdb8)
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 /* ARGSUSED */
103 int
104 sys_fork(struct thread *td, struct fork_args *uap)
105 {
106 	struct fork_req fr;
107 	int error, pid;
108 
109 	bzero(&fr, sizeof(fr));
110 	fr.fr_flags = RFFDG | RFPROC;
111 	fr.fr_pidp = &pid;
112 	error = fork1(td, &fr);
113 	if (error == 0) {
114 		td->td_retval[0] = pid;
115 		td->td_retval[1] = 0;
116 	}
117 	return (error);
118 }
119 
120 /* ARGUSED */
121 int
122 sys_pdfork(struct thread *td, struct pdfork_args *uap)
123 {
124 	struct fork_req fr;
125 	int error, fd, pid;
126 
127 	bzero(&fr, sizeof(fr));
128 	fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
129 	fr.fr_pidp = &pid;
130 	fr.fr_pd_fd = &fd;
131 	fr.fr_pd_flags = uap->flags;
132 	/*
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 		fdtmp = fdinit(td->td_proc->p_fd, false);
336 		fdescfree(td);
337 		p1->p_fd = fdtmp;
338 	}
339 
340 	/*
341 	 * Unshare file descriptors (from parent).
342 	 */
343 	if (flags & RFFDG)
344 		fdunshare(td);
345 
346 fail:
347 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
348 	    (flags & (RFCFDG | RFFDG))) {
349 		PROC_LOCK(p1);
350 		thread_single_end(p1, SINGLE_BOUNDARY);
351 		PROC_UNLOCK(p1);
352 	}
353 	return (error);
354 }
355 
356 static void
357 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
358     struct vmspace *vm2, struct file *fp_procdesc)
359 {
360 	struct proc *p1, *pptr;
361 	struct filedesc *fd;
362 	struct filedesc_to_leader *fdtol;
363 	struct sigacts *newsigacts;
364 
365 	p1 = td->td_proc;
366 
367 	PROC_LOCK(p1);
368 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
369 	    __rangeof(struct proc, p_startcopy, p_endcopy));
370 	pargs_hold(p2->p_args);
371 	PROC_UNLOCK(p1);
372 
373 	bzero(&p2->p_startzero,
374 	    __rangeof(struct proc, p_startzero, p_endzero));
375 
376 	/* Tell the prison that we exist. */
377 	prison_proc_hold(p2->p_ucred->cr_prison);
378 
379 	p2->p_state = PRS_NEW;		/* protect against others */
380 	p2->p_pid = fork_findpid(fr->fr_flags);
381 	AUDIT_ARG_PID(p2->p_pid);
382 
383 	sx_xlock(&allproc_lock);
384 	LIST_INSERT_HEAD(&allproc, p2, p_list);
385 	allproc_gen++;
386 	sx_xunlock(&allproc_lock);
387 
388 	sx_xlock(PIDHASHLOCK(p2->p_pid));
389 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
390 	sx_xunlock(PIDHASHLOCK(p2->p_pid));
391 
392 	tidhash_add(td2);
393 
394 	/*
395 	 * Malloc things while we don't hold any locks.
396 	 */
397 	if (fr->fr_flags & RFSIGSHARE)
398 		newsigacts = NULL;
399 	else
400 		newsigacts = sigacts_alloc();
401 
402 	/*
403 	 * Copy filedesc.
404 	 */
405 	if (fr->fr_flags & RFCFDG) {
406 		fd = fdinit(p1->p_fd, false);
407 		fdtol = NULL;
408 	} else if (fr->fr_flags & RFFDG) {
409 		fd = fdcopy(p1->p_fd);
410 		fdtol = NULL;
411 	} else {
412 		fd = fdshare(p1->p_fd);
413 		if (p1->p_fdtol == NULL)
414 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
415 			    p1->p_leader);
416 		if ((fr->fr_flags & RFTHREAD) != 0) {
417 			/*
418 			 * Shared file descriptor table, and shared
419 			 * process leaders.
420 			 */
421 			fdtol = p1->p_fdtol;
422 			FILEDESC_XLOCK(p1->p_fd);
423 			fdtol->fdl_refcount++;
424 			FILEDESC_XUNLOCK(p1->p_fd);
425 		} else {
426 			/*
427 			 * Shared file descriptor table, and different
428 			 * process leaders.
429 			 */
430 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
431 			    p1->p_fd, p2);
432 		}
433 	}
434 	/*
435 	 * Make a proc table entry for the new process.
436 	 * Start by zeroing the section of proc that is zero-initialized,
437 	 * then copy the section that is copied directly from the parent.
438 	 */
439 
440 	PROC_LOCK(p2);
441 	PROC_LOCK(p1);
442 
443 	bzero(&td2->td_startzero,
444 	    __rangeof(struct thread, td_startzero, td_endzero));
445 
446 	bcopy(&td->td_startcopy, &td2->td_startcopy,
447 	    __rangeof(struct thread, td_startcopy, td_endcopy));
448 
449 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
450 	td2->td_sigstk = td->td_sigstk;
451 	td2->td_flags = TDF_INMEM;
452 	td2->td_lend_user_pri = PRI_MAX;
453 
454 #ifdef VIMAGE
455 	td2->td_vnet = NULL;
456 	td2->td_vnet_lpush = NULL;
457 #endif
458 
459 	/*
460 	 * Allow the scheduler to initialize the child.
461 	 */
462 	thread_lock(td);
463 	sched_fork(td, td2);
464 	thread_unlock(td);
465 
466 	/*
467 	 * Duplicate sub-structures as needed.
468 	 * Increase reference counts on shared objects.
469 	 */
470 	p2->p_flag = P_INMEM;
471 	p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
472 	    P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
473 	    P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
474 	    P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC);
475 	p2->p_swtick = ticks;
476 	if (p1->p_flag & P_PROFIL)
477 		startprofclock(p2);
478 
479 	if (fr->fr_flags & RFSIGSHARE) {
480 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
481 	} else {
482 		sigacts_copy(newsigacts, p1->p_sigacts);
483 		p2->p_sigacts = newsigacts;
484 		if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) {
485 			mtx_lock(&p2->p_sigacts->ps_mtx);
486 			sig_drop_caught(p2);
487 			mtx_unlock(&p2->p_sigacts->ps_mtx);
488 		}
489 	}
490 
491 	if (fr->fr_flags & RFTSIGZMB)
492 	        p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
493 	else if (fr->fr_flags & RFLINUXTHPN)
494 	        p2->p_sigparent = SIGUSR1;
495 	else
496 	        p2->p_sigparent = SIGCHLD;
497 
498 	p2->p_textvp = p1->p_textvp;
499 	p2->p_fd = fd;
500 	p2->p_fdtol = fdtol;
501 
502 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
503 		p2->p_flag |= P_PROTECTED;
504 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
505 	}
506 
507 	/*
508 	 * p_limit is copy-on-write.  Bump its refcount.
509 	 */
510 	lim_fork(p1, p2);
511 
512 	thread_cow_get_proc(td2, p2);
513 
514 	pstats_fork(p1->p_stats, p2->p_stats);
515 
516 	PROC_UNLOCK(p1);
517 	PROC_UNLOCK(p2);
518 
519 	/* Bump references to the text vnode (for procfs). */
520 	if (p2->p_textvp)
521 		vrefact(p2->p_textvp);
522 
523 	/*
524 	 * Set up linkage for kernel based threading.
525 	 */
526 	if ((fr->fr_flags & RFTHREAD) != 0) {
527 		mtx_lock(&ppeers_lock);
528 		p2->p_peers = p1->p_peers;
529 		p1->p_peers = p2;
530 		p2->p_leader = p1->p_leader;
531 		mtx_unlock(&ppeers_lock);
532 		PROC_LOCK(p1->p_leader);
533 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
534 			PROC_UNLOCK(p1->p_leader);
535 			/*
536 			 * The task leader is exiting, so process p1 is
537 			 * going to be killed shortly.  Since p1 obviously
538 			 * isn't dead yet, we know that the leader is either
539 			 * sending SIGKILL's to all the processes in this
540 			 * task or is sleeping waiting for all the peers to
541 			 * exit.  We let p1 complete the fork, but we need
542 			 * to go ahead and kill the new process p2 since
543 			 * the task leader may not get a chance to send
544 			 * SIGKILL to it.  We leave it on the list so that
545 			 * the task leader will wait for this new process
546 			 * to commit suicide.
547 			 */
548 			PROC_LOCK(p2);
549 			kern_psignal(p2, SIGKILL);
550 			PROC_UNLOCK(p2);
551 		} else
552 			PROC_UNLOCK(p1->p_leader);
553 	} else {
554 		p2->p_peers = NULL;
555 		p2->p_leader = p2;
556 	}
557 
558 	sx_xlock(&proctree_lock);
559 	PGRP_LOCK(p1->p_pgrp);
560 	PROC_LOCK(p2);
561 	PROC_LOCK(p1);
562 
563 	/*
564 	 * Preserve some more flags in subprocess.  P_PROFIL has already
565 	 * been preserved.
566 	 */
567 	p2->p_flag |= p1->p_flag & P_SUGID;
568 	td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK |
569 	    TDP_SIGFASTBLOCK)) | TDP_FORKING;
570 	SESS_LOCK(p1->p_session);
571 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
572 		p2->p_flag |= P_CONTROLT;
573 	SESS_UNLOCK(p1->p_session);
574 	if (fr->fr_flags & RFPPWAIT)
575 		p2->p_flag |= P_PPWAIT;
576 
577 	p2->p_pgrp = p1->p_pgrp;
578 	LIST_INSERT_AFTER(p1, p2, p_pglist);
579 	PGRP_UNLOCK(p1->p_pgrp);
580 	LIST_INIT(&p2->p_children);
581 	LIST_INIT(&p2->p_orphans);
582 
583 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
584 
585 	/*
586 	 * If PF_FORK is set, the child process inherits the
587 	 * procfs ioctl flags from its parent.
588 	 */
589 	if (p1->p_pfsflags & PF_FORK) {
590 		p2->p_stops = p1->p_stops;
591 		p2->p_pfsflags = p1->p_pfsflags;
592 	}
593 
594 	/*
595 	 * This begins the section where we must prevent the parent
596 	 * from being swapped.
597 	 */
598 	_PHOLD(p1);
599 	PROC_UNLOCK(p1);
600 
601 	/*
602 	 * Attach the new process to its parent.
603 	 *
604 	 * If RFNOWAIT is set, the newly created process becomes a child
605 	 * of init.  This effectively disassociates the child from the
606 	 * parent.
607 	 */
608 	if ((fr->fr_flags & RFNOWAIT) != 0) {
609 		pptr = p1->p_reaper;
610 		p2->p_reaper = pptr;
611 	} else {
612 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
613 		    p1 : p1->p_reaper;
614 		pptr = p1;
615 	}
616 	p2->p_pptr = pptr;
617 	p2->p_oppid = pptr->p_pid;
618 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
619 	LIST_INIT(&p2->p_reaplist);
620 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
621 	if (p2->p_reaper == p1 && p1 != initproc) {
622 		p2->p_reapsubtree = p2->p_pid;
623 		proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
624 	}
625 	sx_xunlock(&proctree_lock);
626 
627 	/* Inform accounting that we have forked. */
628 	p2->p_acflag = AFORK;
629 	PROC_UNLOCK(p2);
630 
631 #ifdef KTRACE
632 	ktrprocfork(p1, p2);
633 #endif
634 
635 	/*
636 	 * Finish creating the child process.  It will return via a different
637 	 * execution path later.  (ie: directly into user mode)
638 	 */
639 	vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
640 
641 	if (fr->fr_flags == (RFFDG | RFPROC)) {
642 		VM_CNT_INC(v_forks);
643 		VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
644 		    p2->p_vmspace->vm_ssize);
645 	} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
646 		VM_CNT_INC(v_vforks);
647 		VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
648 		    p2->p_vmspace->vm_ssize);
649 	} else if (p1 == &proc0) {
650 		VM_CNT_INC(v_kthreads);
651 		VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
652 		    p2->p_vmspace->vm_ssize);
653 	} else {
654 		VM_CNT_INC(v_rforks);
655 		VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
656 		    p2->p_vmspace->vm_ssize);
657 	}
658 
659 	/*
660 	 * Associate the process descriptor with the process before anything
661 	 * can happen that might cause that process to need the descriptor.
662 	 * However, don't do this until after fork(2) can no longer fail.
663 	 */
664 	if (fr->fr_flags & RFPROCDESC)
665 		procdesc_new(p2, fr->fr_pd_flags);
666 
667 	/*
668 	 * Both processes are set up, now check if any loadable modules want
669 	 * to adjust anything.
670 	 */
671 	EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
672 
673 	/*
674 	 * Set the child start time and mark the process as being complete.
675 	 */
676 	PROC_LOCK(p2);
677 	PROC_LOCK(p1);
678 	microuptime(&p2->p_stats->p_start);
679 	PROC_SLOCK(p2);
680 	p2->p_state = PRS_NORMAL;
681 	PROC_SUNLOCK(p2);
682 
683 #ifdef KDTRACE_HOOKS
684 	/*
685 	 * Tell the DTrace fasttrap provider about the new process so that any
686 	 * tracepoints inherited from the parent can be removed. We have to do
687 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
688 	 * use pfind() later on.
689 	 */
690 	if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
691 		dtrace_fasttrap_fork(p1, p2);
692 #endif
693 	if (fr->fr_flags & RFPPWAIT) {
694 		td->td_pflags |= TDP_RFPPWAIT;
695 		td->td_rfppwait_p = p2;
696 		td->td_dbgflags |= TDB_VFORK;
697 	}
698 	PROC_UNLOCK(p2);
699 
700 	/*
701 	 * Tell any interested parties about the new process.
702 	 */
703 	knote_fork(p1->p_klist, p2->p_pid);
704 
705 	/*
706 	 * Now can be swapped.
707 	 */
708 	_PRELE(p1);
709 	PROC_UNLOCK(p1);
710 	SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
711 
712 	if (fr->fr_flags & RFPROCDESC) {
713 		procdesc_finit(p2->p_procdesc, fp_procdesc);
714 		fdrop(fp_procdesc, td);
715 	}
716 
717 	/*
718 	 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
719 	 * synced with forks in progress so it is OK if we miss it
720 	 * if being set atm.
721 	 */
722 	if ((p1->p_ptevents & PTRACE_FORK) != 0) {
723 		sx_xlock(&proctree_lock);
724 		PROC_LOCK(p2);
725 
726 		/*
727 		 * p1->p_ptevents & p1->p_pptr are protected by both
728 		 * process and proctree locks for modifications,
729 		 * so owning proctree_lock allows the race-free read.
730 		 */
731 		if ((p1->p_ptevents & PTRACE_FORK) != 0) {
732 			/*
733 			 * Arrange for debugger to receive the fork event.
734 			 *
735 			 * We can report PL_FLAG_FORKED regardless of
736 			 * P_FOLLOWFORK settings, but it does not make a sense
737 			 * for runaway child.
738 			 */
739 			td->td_dbgflags |= TDB_FORK;
740 			td->td_dbg_forked = p2->p_pid;
741 			td2->td_dbgflags |= TDB_STOPATFORK;
742 			proc_set_traced(p2, true);
743 			CTR2(KTR_PTRACE,
744 			    "do_fork: attaching to new child pid %d: oppid %d",
745 			    p2->p_pid, p2->p_oppid);
746 			proc_reparent(p2, p1->p_pptr, false);
747 		}
748 		PROC_UNLOCK(p2);
749 		sx_xunlock(&proctree_lock);
750 	}
751 
752 	racct_proc_fork_done(p2);
753 
754 	if ((fr->fr_flags & RFSTOPPED) == 0) {
755 		if (fr->fr_pidp != NULL)
756 			*fr->fr_pidp = p2->p_pid;
757 		/*
758 		 * If RFSTOPPED not requested, make child runnable and
759 		 * add to run queue.
760 		 */
761 		thread_lock(td2);
762 		TD_SET_CAN_RUN(td2);
763 		sched_add(td2, SRQ_BORING);
764 	} else {
765 		*fr->fr_procp = p2;
766 	}
767 }
768 
769 void
770 fork_rfppwait(struct thread *td)
771 {
772 	struct proc *p, *p2;
773 
774 	MPASS(td->td_pflags & TDP_RFPPWAIT);
775 
776 	p = td->td_proc;
777 	/*
778 	 * Preserve synchronization semantics of vfork.  If
779 	 * waiting for child to exec or exit, fork set
780 	 * P_PPWAIT on child, and there we sleep on our proc
781 	 * (in case of exit).
782 	 *
783 	 * Do it after the ptracestop() above is finished, to
784 	 * not block our debugger until child execs or exits
785 	 * to finish vfork wait.
786 	 */
787 	td->td_pflags &= ~TDP_RFPPWAIT;
788 	p2 = td->td_rfppwait_p;
789 again:
790 	PROC_LOCK(p2);
791 	while (p2->p_flag & P_PPWAIT) {
792 		PROC_LOCK(p);
793 		if (thread_suspend_check_needed()) {
794 			PROC_UNLOCK(p2);
795 			thread_suspend_check(0);
796 			PROC_UNLOCK(p);
797 			goto again;
798 		} else {
799 			PROC_UNLOCK(p);
800 		}
801 		cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
802 	}
803 	PROC_UNLOCK(p2);
804 
805 	if (td->td_dbgflags & TDB_VFORK) {
806 		PROC_LOCK(p);
807 		if (p->p_ptevents & PTRACE_VFORK)
808 			ptracestop(td, SIGTRAP, NULL);
809 		td->td_dbgflags &= ~TDB_VFORK;
810 		PROC_UNLOCK(p);
811 	}
812 }
813 
814 int
815 fork1(struct thread *td, struct fork_req *fr)
816 {
817 	struct proc *p1, *newproc;
818 	struct thread *td2;
819 	struct vmspace *vm2;
820 	struct ucred *cred;
821 	struct file *fp_procdesc;
822 	vm_ooffset_t mem_charged;
823 	int error, nprocs_new;
824 	static int curfail;
825 	static struct timeval lastfail;
826 	int flags, pages;
827 
828 	flags = fr->fr_flags;
829 	pages = fr->fr_pages;
830 
831 	if ((flags & RFSTOPPED) != 0)
832 		MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
833 	else
834 		MPASS(fr->fr_procp == NULL);
835 
836 	/* Check for the undefined or unimplemented flags. */
837 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
838 		return (EINVAL);
839 
840 	/* Signal value requires RFTSIGZMB. */
841 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
842 		return (EINVAL);
843 
844 	/* Can't copy and clear. */
845 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
846 		return (EINVAL);
847 
848 	/* Check the validity of the signal number. */
849 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
850 		return (EINVAL);
851 
852 	if ((flags & RFPROCDESC) != 0) {
853 		/* Can't not create a process yet get a process descriptor. */
854 		if ((flags & RFPROC) == 0)
855 			return (EINVAL);
856 
857 		/* Must provide a place to put a procdesc if creating one. */
858 		if (fr->fr_pd_fd == NULL)
859 			return (EINVAL);
860 
861 		/* Check if we are using supported flags. */
862 		if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
863 			return (EINVAL);
864 	}
865 
866 	p1 = td->td_proc;
867 
868 	/*
869 	 * Here we don't create a new process, but we divorce
870 	 * certain parts of a process from itself.
871 	 */
872 	if ((flags & RFPROC) == 0) {
873 		if (fr->fr_procp != NULL)
874 			*fr->fr_procp = NULL;
875 		else if (fr->fr_pidp != NULL)
876 			*fr->fr_pidp = 0;
877 		return (fork_norfproc(td, flags));
878 	}
879 
880 	fp_procdesc = NULL;
881 	newproc = NULL;
882 	vm2 = NULL;
883 
884 	/*
885 	 * Increment the nprocs resource before allocations occur.
886 	 * Although process entries are dynamically created, we still
887 	 * keep a global limit on the maximum number we will
888 	 * create. There are hard-limits as to the number of processes
889 	 * that can run, established by the KVA and memory usage for
890 	 * the process data.
891 	 *
892 	 * Don't allow a nonprivileged user to use the last ten
893 	 * processes; don't let root exceed the limit.
894 	 */
895 	nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
896 	if (nprocs_new >= maxproc - 10) {
897 		if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
898 		    nprocs_new >= maxproc) {
899 			error = EAGAIN;
900 			sx_xlock(&allproc_lock);
901 			if (ppsratecheck(&lastfail, &curfail, 1)) {
902 				printf("maxproc limit exceeded by uid %u "
903 				    "(pid %d); see tuning(7) and "
904 				    "login.conf(5)\n",
905 				    td->td_ucred->cr_ruid, p1->p_pid);
906 			}
907 			sx_xunlock(&allproc_lock);
908 			goto fail2;
909 		}
910 	}
911 
912 	/*
913 	 * If required, create a process descriptor in the parent first; we
914 	 * will abandon it if something goes wrong. We don't finit() until
915 	 * later.
916 	 */
917 	if (flags & RFPROCDESC) {
918 		error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
919 		    fr->fr_pd_flags, fr->fr_pd_fcaps);
920 		if (error != 0)
921 			goto fail2;
922 	}
923 
924 	mem_charged = 0;
925 	if (pages == 0)
926 		pages = kstack_pages;
927 	/* Allocate new proc. */
928 	newproc = uma_zalloc(proc_zone, M_WAITOK);
929 	td2 = FIRST_THREAD_IN_PROC(newproc);
930 	if (td2 == NULL) {
931 		td2 = thread_alloc(pages);
932 		if (td2 == NULL) {
933 			error = ENOMEM;
934 			goto fail2;
935 		}
936 		proc_linkup(newproc, td2);
937 	} else {
938 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
939 			if (td2->td_kstack != 0)
940 				vm_thread_dispose(td2);
941 			if (!thread_alloc_stack(td2, pages)) {
942 				error = ENOMEM;
943 				goto fail2;
944 			}
945 		}
946 	}
947 
948 	if ((flags & RFMEM) == 0) {
949 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
950 		if (vm2 == NULL) {
951 			error = ENOMEM;
952 			goto fail2;
953 		}
954 		if (!swap_reserve(mem_charged)) {
955 			/*
956 			 * The swap reservation failed. The accounting
957 			 * from the entries of the copied vm2 will be
958 			 * subtracted in vmspace_free(), so force the
959 			 * reservation there.
960 			 */
961 			swap_reserve_force(mem_charged);
962 			error = ENOMEM;
963 			goto fail2;
964 		}
965 	} else
966 		vm2 = NULL;
967 
968 	/*
969 	 * XXX: This is ugly; when we copy resource usage, we need to bump
970 	 *      per-cred resource counters.
971 	 */
972 	proc_set_cred_init(newproc, crhold(td->td_ucred));
973 
974 	/*
975 	 * Initialize resource accounting for the child process.
976 	 */
977 	error = racct_proc_fork(p1, newproc);
978 	if (error != 0) {
979 		error = EAGAIN;
980 		goto fail1;
981 	}
982 
983 #ifdef MAC
984 	mac_proc_init(newproc);
985 #endif
986 	newproc->p_klist = knlist_alloc(&newproc->p_mtx);
987 	STAILQ_INIT(&newproc->p_ktr);
988 
989 	/*
990 	 * Increment the count of procs running with this uid. Don't allow
991 	 * a nonprivileged user to exceed their current limit.
992 	 */
993 	cred = td->td_ucred;
994 	if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
995 		if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
996 			goto fail0;
997 		chgproccnt(cred->cr_ruidinfo, 1, 0);
998 	}
999 
1000 	do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1001 	return (0);
1002 fail0:
1003 	error = EAGAIN;
1004 #ifdef MAC
1005 	mac_proc_destroy(newproc);
1006 #endif
1007 	racct_proc_exit(newproc);
1008 fail1:
1009 	crfree(newproc->p_ucred);
1010 	newproc->p_ucred = NULL;
1011 fail2:
1012 	if (vm2 != NULL)
1013 		vmspace_free(vm2);
1014 	uma_zfree(proc_zone, newproc);
1015 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1016 		fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1017 		fdrop(fp_procdesc, td);
1018 	}
1019 	atomic_add_int(&nprocs, -1);
1020 	pause("fork", hz / 2);
1021 	return (error);
1022 }
1023 
1024 /*
1025  * Handle the return of a child process from fork1().  This function
1026  * is called from the MD fork_trampoline() entry point.
1027  */
1028 void
1029 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1030     struct trapframe *frame)
1031 {
1032 	struct proc *p;
1033 	struct thread *td;
1034 	struct thread *dtd;
1035 
1036 	td = curthread;
1037 	p = td->td_proc;
1038 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1039 
1040 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1041 	    td, td_get_sched(td), p->p_pid, td->td_name);
1042 
1043 	sched_fork_exit(td);
1044 	/*
1045 	* Processes normally resume in mi_switch() after being
1046 	* cpu_switch()'ed to, but when children start up they arrive here
1047 	* instead, so we must do much the same things as mi_switch() would.
1048 	*/
1049 	if ((dtd = PCPU_GET(deadthread))) {
1050 		PCPU_SET(deadthread, NULL);
1051 		thread_stash(dtd);
1052 	}
1053 	thread_unlock(td);
1054 
1055 	/*
1056 	 * cpu_fork_kthread_handler intercepts this function call to
1057 	 * have this call a non-return function to stay in kernel mode.
1058 	 * initproc has its own fork handler, but it does return.
1059 	 */
1060 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1061 	callout(arg, frame);
1062 
1063 	/*
1064 	 * Check if a kernel thread misbehaved and returned from its main
1065 	 * function.
1066 	 */
1067 	if (p->p_flag & P_KPROC) {
1068 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1069 		    td->td_name, p->p_pid);
1070 		kthread_exit();
1071 	}
1072 	mtx_assert(&Giant, MA_NOTOWNED);
1073 
1074 	if (p->p_sysent->sv_schedtail != NULL)
1075 		(p->p_sysent->sv_schedtail)(td);
1076 	td->td_pflags &= ~TDP_FORKING;
1077 }
1078 
1079 /*
1080  * Simplified back end of syscall(), used when returning from fork()
1081  * directly into user mode.  This function is passed in to fork_exit()
1082  * as the first parameter and is called when returning to a new
1083  * userland process.
1084  */
1085 void
1086 fork_return(struct thread *td, struct trapframe *frame)
1087 {
1088 	struct proc *p;
1089 
1090 	p = td->td_proc;
1091 	if (td->td_dbgflags & TDB_STOPATFORK) {
1092 		PROC_LOCK(p);
1093 		if ((p->p_flag & P_TRACED) != 0) {
1094 			/*
1095 			 * Inform the debugger if one is still present.
1096 			 */
1097 			td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1098 			ptracestop(td, SIGSTOP, NULL);
1099 			td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1100 		} else {
1101 			/*
1102 			 * ... otherwise clear the request.
1103 			 */
1104 			td->td_dbgflags &= ~TDB_STOPATFORK;
1105 		}
1106 		PROC_UNLOCK(p);
1107 	} else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1108  		/*
1109 		 * This is the start of a new thread in a traced
1110 		 * process.  Report a system call exit event.
1111 		 */
1112 		PROC_LOCK(p);
1113 		td->td_dbgflags |= TDB_SCX;
1114 		_STOPEVENT(p, S_SCX, td->td_sa.code);
1115 		if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1116 		    (td->td_dbgflags & TDB_BORN) != 0)
1117 			ptracestop(td, SIGTRAP, NULL);
1118 		td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1119 		PROC_UNLOCK(p);
1120 	}
1121 
1122 	userret(td, frame);
1123 
1124 #ifdef KTRACE
1125 	if (KTRPOINT(td, KTR_SYSRET))
1126 		ktrsysret(SYS_fork, 0, 0);
1127 #endif
1128 }
1129