xref: /freebsd/sys/kern/kern_fork.c (revision 1dbf9dcc20ca9dcb5fde8e70f32760c3f165f3d0)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_kstack_pages.h"
43 #include "opt_procdesc.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/fcntl.h>
50 #include <sys/filedesc.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/kthread.h>
54 #include <sys/sysctl.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mutex.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/procdesc.h>
61 #include <sys/pioctl.h>
62 #include <sys/racct.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sched.h>
65 #include <sys/syscall.h>
66 #include <sys/vmmeter.h>
67 #include <sys/vnode.h>
68 #include <sys/acct.h>
69 #include <sys/ktr.h>
70 #include <sys/ktrace.h>
71 #include <sys/unistd.h>
72 #include <sys/sdt.h>
73 #include <sys/sx.h>
74 #include <sys/sysent.h>
75 #include <sys/signalvar.h>
76 
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/pmap.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_extern.h>
84 #include <vm/uma.h>
85 
86 #ifdef KDTRACE_HOOKS
87 #include <sys/dtrace_bsd.h>
88 dtrace_fork_func_t	dtrace_fasttrap_fork;
89 #endif
90 
91 SDT_PROVIDER_DECLARE(proc);
92 SDT_PROBE_DEFINE(proc, kernel, , create, create);
93 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
94 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
95 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
96 
97 #ifndef _SYS_SYSPROTO_H_
98 struct fork_args {
99 	int     dummy;
100 };
101 #endif
102 
103 /* ARGSUSED */
104 int
105 sys_fork(struct thread *td, struct fork_args *uap)
106 {
107 	int error;
108 	struct proc *p2;
109 
110 	error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0);
111 	if (error == 0) {
112 		td->td_retval[0] = p2->p_pid;
113 		td->td_retval[1] = 0;
114 	}
115 	return (error);
116 }
117 
118 /* ARGUSED */
119 int
120 sys_pdfork(td, uap)
121 	struct thread *td;
122 	struct pdfork_args *uap;
123 {
124 #ifdef PROCDESC
125 	int error, fd;
126 	struct proc *p2;
127 
128 	/*
129 	 * It is necessary to return fd by reference because 0 is a valid file
130 	 * descriptor number, and the child needs to be able to distinguish
131 	 * itself from the parent using the return value.
132 	 */
133 	error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2,
134 	    &fd, uap->flags);
135 	if (error == 0) {
136 		td->td_retval[0] = p2->p_pid;
137 		td->td_retval[1] = 0;
138 		error = copyout(&fd, uap->fdp, sizeof(fd));
139 	}
140 	return (error);
141 #else
142 	return (ENOSYS);
143 #endif
144 }
145 
146 /* ARGSUSED */
147 int
148 sys_vfork(struct thread *td, struct vfork_args *uap)
149 {
150 	int error, flags;
151 	struct proc *p2;
152 
153 #ifdef XEN
154 	flags = RFFDG | RFPROC; /* validate that this is still an issue */
155 #else
156 	flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
157 #endif
158 	error = fork1(td, flags, 0, &p2, NULL, 0);
159 	if (error == 0) {
160 		td->td_retval[0] = p2->p_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 proc *p2;
170 	int error;
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 	error = fork1(td, uap->flags, 0, &p2, NULL, 0);
178 	if (error == 0) {
179 		td->td_retval[0] = p2 ? p2->p_pid : 0;
180 		td->td_retval[1] = 0;
181 	}
182 	return (error);
183 }
184 
185 int	nprocs = 1;		/* process 0 */
186 int	lastpid = 0;
187 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
188     "Last used PID");
189 
190 /*
191  * Random component to lastpid generation.  We mix in a random factor to make
192  * it a little harder to predict.  We sanity check the modulus value to avoid
193  * doing it in critical paths.  Don't let it be too small or we pointlessly
194  * waste randomness entropy, and don't let it be impossibly large.  Using a
195  * modulus that is too big causes a LOT more process table scans and slows
196  * down fork processing as the pidchecked caching is defeated.
197  */
198 static int randompid = 0;
199 
200 static int
201 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
202 {
203 	int error, pid;
204 
205 	error = sysctl_wire_old_buffer(req, sizeof(int));
206 	if (error != 0)
207 		return(error);
208 	sx_xlock(&allproc_lock);
209 	pid = randompid;
210 	error = sysctl_handle_int(oidp, &pid, 0, req);
211 	if (error == 0 && req->newptr != NULL) {
212 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
213 			pid = PID_MAX - 100;
214 		else if (pid < 2)			/* NOP */
215 			pid = 0;
216 		else if (pid < 100)			/* Make it reasonable */
217 			pid = 100;
218 		randompid = pid;
219 	}
220 	sx_xunlock(&allproc_lock);
221 	return (error);
222 }
223 
224 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
225     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
226 
227 static int
228 fork_findpid(int flags)
229 {
230 	struct proc *p;
231 	int trypid;
232 	static int pidchecked = 0;
233 
234 	/*
235 	 * Requires allproc_lock in order to iterate over the list
236 	 * of processes, and proctree_lock to access p_pgrp.
237 	 */
238 	sx_assert(&allproc_lock, SX_LOCKED);
239 	sx_assert(&proctree_lock, SX_LOCKED);
240 
241 	/*
242 	 * Find an unused process ID.  We remember a range of unused IDs
243 	 * ready to use (from lastpid+1 through pidchecked-1).
244 	 *
245 	 * If RFHIGHPID is set (used during system boot), do not allocate
246 	 * low-numbered pids.
247 	 */
248 	trypid = lastpid + 1;
249 	if (flags & RFHIGHPID) {
250 		if (trypid < 10)
251 			trypid = 10;
252 	} else {
253 		if (randompid)
254 			trypid += arc4random() % randompid;
255 	}
256 retry:
257 	/*
258 	 * If the process ID prototype has wrapped around,
259 	 * restart somewhat above 0, as the low-numbered procs
260 	 * tend to include daemons that don't exit.
261 	 */
262 	if (trypid >= PID_MAX) {
263 		trypid = trypid % PID_MAX;
264 		if (trypid < 100)
265 			trypid += 100;
266 		pidchecked = 0;
267 	}
268 	if (trypid >= pidchecked) {
269 		int doingzomb = 0;
270 
271 		pidchecked = PID_MAX;
272 		/*
273 		 * Scan the active and zombie procs to check whether this pid
274 		 * is in use.  Remember the lowest pid that's greater
275 		 * than trypid, so we can avoid checking for a while.
276 		 */
277 		p = LIST_FIRST(&allproc);
278 again:
279 		for (; p != NULL; p = LIST_NEXT(p, p_list)) {
280 			while (p->p_pid == trypid ||
281 			    (p->p_pgrp != NULL &&
282 			    (p->p_pgrp->pg_id == trypid ||
283 			    (p->p_session != NULL &&
284 			    p->p_session->s_sid == trypid)))) {
285 				trypid++;
286 				if (trypid >= pidchecked)
287 					goto retry;
288 			}
289 			if (p->p_pid > trypid && pidchecked > p->p_pid)
290 				pidchecked = p->p_pid;
291 			if (p->p_pgrp != NULL) {
292 				if (p->p_pgrp->pg_id > trypid &&
293 				    pidchecked > p->p_pgrp->pg_id)
294 					pidchecked = p->p_pgrp->pg_id;
295 				if (p->p_session != NULL &&
296 				    p->p_session->s_sid > trypid &&
297 				    pidchecked > p->p_session->s_sid)
298 					pidchecked = p->p_session->s_sid;
299 			}
300 		}
301 		if (!doingzomb) {
302 			doingzomb = 1;
303 			p = LIST_FIRST(&zombproc);
304 			goto again;
305 		}
306 	}
307 
308 	/*
309 	 * RFHIGHPID does not mess with the lastpid counter during boot.
310 	 */
311 	if (flags & RFHIGHPID)
312 		pidchecked = 0;
313 	else
314 		lastpid = trypid;
315 
316 	return (trypid);
317 }
318 
319 static int
320 fork_norfproc(struct thread *td, int flags)
321 {
322 	int error;
323 	struct proc *p1;
324 
325 	KASSERT((flags & RFPROC) == 0,
326 	    ("fork_norfproc called with RFPROC set"));
327 	p1 = td->td_proc;
328 
329 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
330 	    (flags & (RFCFDG | RFFDG))) {
331 		PROC_LOCK(p1);
332 		if (thread_single(SINGLE_BOUNDARY)) {
333 			PROC_UNLOCK(p1);
334 			return (ERESTART);
335 		}
336 		PROC_UNLOCK(p1);
337 	}
338 
339 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
340 	if (error)
341 		goto fail;
342 
343 	/*
344 	 * Close all file descriptors.
345 	 */
346 	if (flags & RFCFDG) {
347 		struct filedesc *fdtmp;
348 		fdtmp = fdinit(td->td_proc->p_fd);
349 		fdfree(td);
350 		p1->p_fd = fdtmp;
351 	}
352 
353 	/*
354 	 * Unshare file descriptors (from parent).
355 	 */
356 	if (flags & RFFDG)
357 		fdunshare(p1, td);
358 
359 fail:
360 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
361 	    (flags & (RFCFDG | RFFDG))) {
362 		PROC_LOCK(p1);
363 		thread_single_end();
364 		PROC_UNLOCK(p1);
365 	}
366 	return (error);
367 }
368 
369 static void
370 do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
371     struct vmspace *vm2, int pdflags)
372 {
373 	struct proc *p1, *pptr;
374 	int p2_held, trypid;
375 	struct filedesc *fd;
376 	struct filedesc_to_leader *fdtol;
377 	struct sigacts *newsigacts;
378 
379 	sx_assert(&proctree_lock, SX_SLOCKED);
380 	sx_assert(&allproc_lock, SX_XLOCKED);
381 
382 	p2_held = 0;
383 	p1 = td->td_proc;
384 
385 	/*
386 	 * Increment the nprocs resource before blocking can occur.  There
387 	 * are hard-limits as to the number of processes that can run.
388 	 */
389 	nprocs++;
390 
391 	trypid = fork_findpid(flags);
392 
393 	sx_sunlock(&proctree_lock);
394 
395 	p2->p_state = PRS_NEW;		/* protect against others */
396 	p2->p_pid = trypid;
397 	AUDIT_ARG_PID(p2->p_pid);
398 	LIST_INSERT_HEAD(&allproc, p2, p_list);
399 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
400 	tidhash_add(td2);
401 	PROC_LOCK(p2);
402 	PROC_LOCK(p1);
403 
404 	sx_xunlock(&allproc_lock);
405 
406 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
407 	    __rangeof(struct proc, p_startcopy, p_endcopy));
408 	pargs_hold(p2->p_args);
409 	PROC_UNLOCK(p1);
410 
411 	bzero(&p2->p_startzero,
412 	    __rangeof(struct proc, p_startzero, p_endzero));
413 
414 	p2->p_ucred = crhold(td->td_ucred);
415 
416 	/* Tell the prison that we exist. */
417 	prison_proc_hold(p2->p_ucred->cr_prison);
418 
419 	PROC_UNLOCK(p2);
420 
421 	/*
422 	 * Malloc things while we don't hold any locks.
423 	 */
424 	if (flags & RFSIGSHARE)
425 		newsigacts = NULL;
426 	else
427 		newsigacts = sigacts_alloc();
428 
429 	/*
430 	 * Copy filedesc.
431 	 */
432 	if (flags & RFCFDG) {
433 		fd = fdinit(p1->p_fd);
434 		fdtol = NULL;
435 	} else if (flags & RFFDG) {
436 		fd = fdcopy(p1->p_fd);
437 		fdtol = NULL;
438 	} else {
439 		fd = fdshare(p1->p_fd);
440 		if (p1->p_fdtol == NULL)
441 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
442 			    p1->p_leader);
443 		if ((flags & RFTHREAD) != 0) {
444 			/*
445 			 * Shared file descriptor table, and shared
446 			 * process leaders.
447 			 */
448 			fdtol = p1->p_fdtol;
449 			FILEDESC_XLOCK(p1->p_fd);
450 			fdtol->fdl_refcount++;
451 			FILEDESC_XUNLOCK(p1->p_fd);
452 		} else {
453 			/*
454 			 * Shared file descriptor table, and different
455 			 * process leaders.
456 			 */
457 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
458 			    p1->p_fd, p2);
459 		}
460 	}
461 	/*
462 	 * Make a proc table entry for the new process.
463 	 * Start by zeroing the section of proc that is zero-initialized,
464 	 * then copy the section that is copied directly from the parent.
465 	 */
466 
467 	PROC_LOCK(p2);
468 	PROC_LOCK(p1);
469 
470 	bzero(&td2->td_startzero,
471 	    __rangeof(struct thread, td_startzero, td_endzero));
472 
473 	bcopy(&td->td_startcopy, &td2->td_startcopy,
474 	    __rangeof(struct thread, td_startcopy, td_endcopy));
475 
476 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
477 	td2->td_sigstk = td->td_sigstk;
478 	td2->td_sigmask = td->td_sigmask;
479 	td2->td_flags = TDF_INMEM;
480 	td2->td_lend_user_pri = PRI_MAX;
481 
482 #ifdef VIMAGE
483 	td2->td_vnet = NULL;
484 	td2->td_vnet_lpush = NULL;
485 #endif
486 
487 	/*
488 	 * Allow the scheduler to initialize the child.
489 	 */
490 	thread_lock(td);
491 	sched_fork(td, td2);
492 	thread_unlock(td);
493 
494 	/*
495 	 * Duplicate sub-structures as needed.
496 	 * Increase reference counts on shared objects.
497 	 */
498 	p2->p_flag = P_INMEM;
499 	p2->p_swtick = ticks;
500 	if (p1->p_flag & P_PROFIL)
501 		startprofclock(p2);
502 	td2->td_ucred = crhold(p2->p_ucred);
503 
504 	if (flags & RFSIGSHARE) {
505 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
506 	} else {
507 		sigacts_copy(newsigacts, p1->p_sigacts);
508 		p2->p_sigacts = newsigacts;
509 	}
510 
511 	if (flags & RFTSIGZMB)
512 	        p2->p_sigparent = RFTSIGNUM(flags);
513 	else if (flags & RFLINUXTHPN)
514 	        p2->p_sigparent = SIGUSR1;
515 	else
516 	        p2->p_sigparent = SIGCHLD;
517 
518 	p2->p_textvp = p1->p_textvp;
519 	p2->p_fd = fd;
520 	p2->p_fdtol = fdtol;
521 
522 	/*
523 	 * p_limit is copy-on-write.  Bump its refcount.
524 	 */
525 	lim_fork(p1, p2);
526 
527 	pstats_fork(p1->p_stats, p2->p_stats);
528 
529 	PROC_UNLOCK(p1);
530 	PROC_UNLOCK(p2);
531 
532 	/* Bump references to the text vnode (for procfs). */
533 	if (p2->p_textvp)
534 		vref(p2->p_textvp);
535 
536 	/*
537 	 * Set up linkage for kernel based threading.
538 	 */
539 	if ((flags & RFTHREAD) != 0) {
540 		mtx_lock(&ppeers_lock);
541 		p2->p_peers = p1->p_peers;
542 		p1->p_peers = p2;
543 		p2->p_leader = p1->p_leader;
544 		mtx_unlock(&ppeers_lock);
545 		PROC_LOCK(p1->p_leader);
546 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
547 			PROC_UNLOCK(p1->p_leader);
548 			/*
549 			 * The task leader is exiting, so process p1 is
550 			 * going to be killed shortly.  Since p1 obviously
551 			 * isn't dead yet, we know that the leader is either
552 			 * sending SIGKILL's to all the processes in this
553 			 * task or is sleeping waiting for all the peers to
554 			 * exit.  We let p1 complete the fork, but we need
555 			 * to go ahead and kill the new process p2 since
556 			 * the task leader may not get a chance to send
557 			 * SIGKILL to it.  We leave it on the list so that
558 			 * the task leader will wait for this new process
559 			 * to commit suicide.
560 			 */
561 			PROC_LOCK(p2);
562 			kern_psignal(p2, SIGKILL);
563 			PROC_UNLOCK(p2);
564 		} else
565 			PROC_UNLOCK(p1->p_leader);
566 	} else {
567 		p2->p_peers = NULL;
568 		p2->p_leader = p2;
569 	}
570 
571 	sx_xlock(&proctree_lock);
572 	PGRP_LOCK(p1->p_pgrp);
573 	PROC_LOCK(p2);
574 	PROC_LOCK(p1);
575 
576 	/*
577 	 * Preserve some more flags in subprocess.  P_PROFIL has already
578 	 * been preserved.
579 	 */
580 	p2->p_flag |= p1->p_flag & P_SUGID;
581 	td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
582 	SESS_LOCK(p1->p_session);
583 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
584 		p2->p_flag |= P_CONTROLT;
585 	SESS_UNLOCK(p1->p_session);
586 	if (flags & RFPPWAIT)
587 		p2->p_flag |= P_PPWAIT;
588 
589 	p2->p_pgrp = p1->p_pgrp;
590 	LIST_INSERT_AFTER(p1, p2, p_pglist);
591 	PGRP_UNLOCK(p1->p_pgrp);
592 	LIST_INIT(&p2->p_children);
593 
594 	callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
595 
596 	/*
597 	 * If PF_FORK is set, the child process inherits the
598 	 * procfs ioctl flags from its parent.
599 	 */
600 	if (p1->p_pfsflags & PF_FORK) {
601 		p2->p_stops = p1->p_stops;
602 		p2->p_pfsflags = p1->p_pfsflags;
603 	}
604 
605 	/*
606 	 * This begins the section where we must prevent the parent
607 	 * from being swapped.
608 	 */
609 	_PHOLD(p1);
610 	PROC_UNLOCK(p1);
611 
612 	/*
613 	 * Attach the new process to its parent.
614 	 *
615 	 * If RFNOWAIT is set, the newly created process becomes a child
616 	 * of init.  This effectively disassociates the child from the
617 	 * parent.
618 	 */
619 	if (flags & RFNOWAIT)
620 		pptr = initproc;
621 	else
622 		pptr = p1;
623 	p2->p_pptr = pptr;
624 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
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, flags);
640 
641 	if (flags == (RFFDG | RFPROC)) {
642 		PCPU_INC(cnt.v_forks);
643 		PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
644 		    p2->p_vmspace->vm_ssize);
645 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
646 		PCPU_INC(cnt.v_vforks);
647 		PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
648 		    p2->p_vmspace->vm_ssize);
649 	} else if (p1 == &proc0) {
650 		PCPU_INC(cnt.v_kthreads);
651 		PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
652 		    p2->p_vmspace->vm_ssize);
653 	} else {
654 		PCPU_INC(cnt.v_rforks);
655 		PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
656 		    p2->p_vmspace->vm_ssize);
657 	}
658 
659 #ifdef PROCDESC
660 	/*
661 	 * Associate the process descriptor with the process before anything
662 	 * can happen that might cause that process to need the descriptor.
663 	 * However, don't do this until after fork(2) can no longer fail.
664 	 */
665 	if (flags & RFPROCDESC)
666 		procdesc_new(p2, pdflags);
667 #endif
668 
669 	/*
670 	 * Both processes are set up, now check if any loadable modules want
671 	 * to adjust anything.
672 	 */
673 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
674 
675 	/*
676 	 * Set the child start time and mark the process as being complete.
677 	 */
678 	PROC_LOCK(p2);
679 	PROC_LOCK(p1);
680 	microuptime(&p2->p_stats->p_start);
681 	PROC_SLOCK(p2);
682 	p2->p_state = PRS_NORMAL;
683 	PROC_SUNLOCK(p2);
684 
685 #ifdef KDTRACE_HOOKS
686 	/*
687 	 * Tell the DTrace fasttrap provider about the new process
688 	 * if it has registered an interest. We have to do this only after
689 	 * p_state is PRS_NORMAL since the fasttrap module will use pfind()
690 	 * later on.
691 	 */
692 	if (dtrace_fasttrap_fork)
693 		dtrace_fasttrap_fork(p1, p2);
694 #endif
695 	if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED |
696 	    P_FOLLOWFORK)) {
697 		/*
698 		 * Arrange for debugger to receive the fork event.
699 		 *
700 		 * We can report PL_FLAG_FORKED regardless of
701 		 * P_FOLLOWFORK settings, but it does not make a sense
702 		 * for runaway child.
703 		 */
704 		td->td_dbgflags |= TDB_FORK;
705 		td->td_dbg_forked = p2->p_pid;
706 		td2->td_dbgflags |= TDB_STOPATFORK;
707 		_PHOLD(p2);
708 		p2_held = 1;
709 	}
710 	PROC_UNLOCK(p2);
711 	if ((flags & RFSTOPPED) == 0) {
712 		/*
713 		 * If RFSTOPPED not requested, make child runnable and
714 		 * add to run queue.
715 		 */
716 		thread_lock(td2);
717 		TD_SET_CAN_RUN(td2);
718 		sched_add(td2, SRQ_BORING);
719 		thread_unlock(td2);
720 	}
721 
722 	/*
723 	 * Now can be swapped.
724 	 */
725 	_PRELE(p1);
726 	PROC_UNLOCK(p1);
727 
728 	/*
729 	 * Tell any interested parties about the new process.
730 	 */
731 	knote_fork(&p1->p_klist, p2->p_pid);
732 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
733 
734 	/*
735 	 * Wait until debugger is attached to child.
736 	 */
737 	PROC_LOCK(p2);
738 	while ((td2->td_dbgflags & TDB_STOPATFORK) != 0)
739 		cv_wait(&p2->p_dbgwait, &p2->p_mtx);
740 	if (p2_held)
741 		_PRELE(p2);
742 
743 	/*
744 	 * Preserve synchronization semantics of vfork.  If waiting for
745 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
746 	 * proc (in case of exit).
747 	 */
748 	while (p2->p_flag & P_PPWAIT)
749 		cv_wait(&p2->p_pwait, &p2->p_mtx);
750 	PROC_UNLOCK(p2);
751 }
752 
753 int
754 fork1(struct thread *td, int flags, int pages, struct proc **procp,
755     int *procdescp, int pdflags)
756 {
757 	struct proc *p1;
758 	struct proc *newproc;
759 	int ok;
760 	struct thread *td2;
761 	struct vmspace *vm2;
762 	vm_ooffset_t mem_charged;
763 	int error;
764 	static int curfail;
765 	static struct timeval lastfail;
766 #ifdef PROCDESC
767 	struct file *fp_procdesc = NULL;
768 #endif
769 
770 	/* Check for the undefined or unimplemented flags. */
771 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
772 		return (EINVAL);
773 
774 	/* Signal value requires RFTSIGZMB. */
775 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
776 		return (EINVAL);
777 
778 	/* Can't copy and clear. */
779 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
780 		return (EINVAL);
781 
782 	/* Check the validity of the signal number. */
783 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
784 		return (EINVAL);
785 
786 #ifdef PROCDESC
787 	if ((flags & RFPROCDESC) != 0) {
788 		/* Can't not create a process yet get a process descriptor. */
789 		if ((flags & RFPROC) == 0)
790 			return (EINVAL);
791 
792 		/* Must provide a place to put a procdesc if creating one. */
793 		if (procdescp == NULL)
794 			return (EINVAL);
795 	}
796 #endif
797 
798 	p1 = td->td_proc;
799 
800 	/*
801 	 * Here we don't create a new process, but we divorce
802 	 * certain parts of a process from itself.
803 	 */
804 	if ((flags & RFPROC) == 0) {
805 		*procp = NULL;
806 		return (fork_norfproc(td, flags));
807 	}
808 
809 #ifdef PROCDESC
810 	/*
811 	 * If required, create a process descriptor in the parent first; we
812 	 * will abandon it if something goes wrong. We don't finit() until
813 	 * later.
814 	 */
815 	if (flags & RFPROCDESC) {
816 		error = falloc(td, &fp_procdesc, procdescp, 0);
817 		if (error != 0)
818 			return (error);
819 	}
820 #endif
821 
822 	mem_charged = 0;
823 	vm2 = NULL;
824 	if (pages == 0)
825 		pages = KSTACK_PAGES;
826 	/* Allocate new proc. */
827 	newproc = uma_zalloc(proc_zone, M_WAITOK);
828 	td2 = FIRST_THREAD_IN_PROC(newproc);
829 	if (td2 == NULL) {
830 		td2 = thread_alloc(pages);
831 		if (td2 == NULL) {
832 			error = ENOMEM;
833 			goto fail1;
834 		}
835 		proc_linkup(newproc, td2);
836 	} else {
837 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
838 			if (td2->td_kstack != 0)
839 				vm_thread_dispose(td2);
840 			if (!thread_alloc_stack(td2, pages)) {
841 				error = ENOMEM;
842 				goto fail1;
843 			}
844 		}
845 	}
846 
847 	if ((flags & RFMEM) == 0) {
848 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
849 		if (vm2 == NULL) {
850 			error = ENOMEM;
851 			goto fail1;
852 		}
853 		if (!swap_reserve(mem_charged)) {
854 			/*
855 			 * The swap reservation failed. The accounting
856 			 * from the entries of the copied vm2 will be
857 			 * substracted in vmspace_free(), so force the
858 			 * reservation there.
859 			 */
860 			swap_reserve_force(mem_charged);
861 			error = ENOMEM;
862 			goto fail1;
863 		}
864 	} else
865 		vm2 = NULL;
866 
867 	/*
868 	 * XXX: This is ugly; when we copy resource usage, we need to bump
869 	 *      per-cred resource counters.
870 	 */
871 	newproc->p_ucred = p1->p_ucred;
872 
873 	/*
874 	 * Initialize resource accounting for the child process.
875 	 */
876 	error = racct_proc_fork(p1, newproc);
877 	if (error != 0) {
878 		error = EAGAIN;
879 		goto fail1;
880 	}
881 
882 #ifdef RACCT
883 	PROC_LOCK(newproc);
884 	error = racct_add(newproc, RACCT_NPROC, 1);
885 	error += racct_add(newproc, RACCT_NTHR, 1);
886 	PROC_UNLOCK(newproc);
887 	if (error != 0) {
888 		error = EAGAIN;
889 		goto fail1;
890 	}
891 #endif
892 
893 #ifdef MAC
894 	mac_proc_init(newproc);
895 #endif
896 	knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
897 	STAILQ_INIT(&newproc->p_ktr);
898 
899 	/* We have to lock the process tree while we look for a pid. */
900 	sx_slock(&proctree_lock);
901 
902 	/*
903 	 * Although process entries are dynamically created, we still keep
904 	 * a global limit on the maximum number we will create.  Don't allow
905 	 * a nonprivileged user to use the last ten processes; don't let root
906 	 * exceed the limit. The variable nprocs is the current number of
907 	 * processes, maxproc is the limit.
908 	 */
909 	sx_xlock(&allproc_lock);
910 	if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
911 	    PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
912 		error = EAGAIN;
913 		goto fail;
914 	}
915 
916 	/*
917 	 * Increment the count of procs running with this uid. Don't allow
918 	 * a nonprivileged user to exceed their current limit.
919 	 *
920 	 * XXXRW: Can we avoid privilege here if it's not needed?
921 	 */
922 	error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
923 	if (error == 0)
924 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
925 	else {
926 		PROC_LOCK(p1);
927 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
928 		    lim_cur(p1, RLIMIT_NPROC));
929 		PROC_UNLOCK(p1);
930 	}
931 	if (ok) {
932 		do_fork(td, flags, newproc, td2, vm2, pdflags);
933 
934 		/*
935 		 * Return child proc pointer to parent.
936 		 */
937 		*procp = newproc;
938 #ifdef PROCDESC
939 		if (flags & RFPROCDESC)
940 			procdesc_finit(newproc->p_procdesc, fp_procdesc);
941 #endif
942 		return (0);
943 	}
944 
945 	error = EAGAIN;
946 fail:
947 	sx_sunlock(&proctree_lock);
948 	if (ppsratecheck(&lastfail, &curfail, 1))
949 		printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
950 		    td->td_ucred->cr_ruid);
951 	sx_xunlock(&allproc_lock);
952 #ifdef MAC
953 	mac_proc_destroy(newproc);
954 #endif
955 fail1:
956 	racct_proc_exit(newproc);
957 	if (vm2 != NULL)
958 		vmspace_free(vm2);
959 	uma_zfree(proc_zone, newproc);
960 #ifdef PROCDESC
961 	if (((flags & RFPROCDESC) != 0) && (fp_procdesc != NULL))
962 		fdrop(fp_procdesc, td);
963 #endif
964 	pause("fork", hz / 2);
965 	return (error);
966 }
967 
968 /*
969  * Handle the return of a child process from fork1().  This function
970  * is called from the MD fork_trampoline() entry point.
971  */
972 void
973 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
974     struct trapframe *frame)
975 {
976 	struct proc *p;
977 	struct thread *td;
978 	struct thread *dtd;
979 
980 	td = curthread;
981 	p = td->td_proc;
982 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
983 
984 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
985 		td, td->td_sched, p->p_pid, td->td_name);
986 
987 	sched_fork_exit(td);
988 	/*
989 	* Processes normally resume in mi_switch() after being
990 	* cpu_switch()'ed to, but when children start up they arrive here
991 	* instead, so we must do much the same things as mi_switch() would.
992 	*/
993 	if ((dtd = PCPU_GET(deadthread))) {
994 		PCPU_SET(deadthread, NULL);
995 		thread_stash(dtd);
996 	}
997 	thread_unlock(td);
998 
999 	/*
1000 	 * cpu_set_fork_handler intercepts this function call to
1001 	 * have this call a non-return function to stay in kernel mode.
1002 	 * initproc has its own fork handler, but it does return.
1003 	 */
1004 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1005 	callout(arg, frame);
1006 
1007 	/*
1008 	 * Check if a kernel thread misbehaved and returned from its main
1009 	 * function.
1010 	 */
1011 	if (p->p_flag & P_KTHREAD) {
1012 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1013 		    td->td_name, p->p_pid);
1014 		kproc_exit(0);
1015 	}
1016 	mtx_assert(&Giant, MA_NOTOWNED);
1017 
1018 	if (p->p_sysent->sv_schedtail != NULL)
1019 		(p->p_sysent->sv_schedtail)(td);
1020 }
1021 
1022 /*
1023  * Simplified back end of syscall(), used when returning from fork()
1024  * directly into user mode.  Giant is not held on entry, and must not
1025  * be held on return.  This function is passed in to fork_exit() as the
1026  * first parameter and is called when returning to a new userland process.
1027  */
1028 void
1029 fork_return(struct thread *td, struct trapframe *frame)
1030 {
1031 	struct proc *p, *dbg;
1032 
1033 	if (td->td_dbgflags & TDB_STOPATFORK) {
1034 		p = td->td_proc;
1035 		sx_xlock(&proctree_lock);
1036 		PROC_LOCK(p);
1037 		if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) ==
1038 		    (P_TRACED | P_FOLLOWFORK)) {
1039 			/*
1040 			 * If debugger still wants auto-attach for the
1041 			 * parent's children, do it now.
1042 			 */
1043 			dbg = p->p_pptr->p_pptr;
1044 			p->p_flag |= P_TRACED;
1045 			p->p_oppid = p->p_pptr->p_pid;
1046 			proc_reparent(p, dbg);
1047 			sx_xunlock(&proctree_lock);
1048 			ptracestop(td, SIGSTOP);
1049 		} else {
1050 			/*
1051 			 * ... otherwise clear the request.
1052 			 */
1053 			sx_xunlock(&proctree_lock);
1054 			td->td_dbgflags &= ~TDB_STOPATFORK;
1055 			cv_broadcast(&p->p_dbgwait);
1056 		}
1057 		PROC_UNLOCK(p);
1058 	}
1059 
1060 	userret(td, frame);
1061 
1062 #ifdef KTRACE
1063 	if (KTRPOINT(td, KTR_SYSRET))
1064 		ktrsysret(SYS_fork, 0, 0);
1065 #endif
1066 	mtx_assert(&Giant, MA_NOTOWNED);
1067 }
1068