xref: /freebsd/sys/kern/kern_fork.c (revision 0b87f79976047c8f4332bbf7dc03146f6b0de79f)
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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
39  * $FreeBSD$
40  */
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/resourcevar.h>
55 #include <sys/syscall.h>
56 #include <sys/vnode.h>
57 #include <sys/acct.h>
58 #include <sys/ktr.h>
59 #include <sys/ktrace.h>
60 #include <sys/kthread.h>
61 #include <sys/unistd.h>
62 #include <sys/jail.h>
63 #include <sys/sx.h>
64 
65 #include <vm/vm.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_map.h>
68 #include <vm/vm_extern.h>
69 #include <vm/uma.h>
70 
71 #include <sys/vmmeter.h>
72 #include <sys/user.h>
73 #include <machine/critical.h>
74 
75 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
76 
77 /*
78  * These are the stuctures used to create a callout list for things to do
79  * when forking a process
80  */
81 struct forklist {
82 	forklist_fn function;
83 	TAILQ_ENTRY(forklist) next;
84 };
85 
86 static struct sx fork_list_lock;
87 
88 TAILQ_HEAD(forklist_head, forklist);
89 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
90 
91 #ifndef _SYS_SYSPROTO_H_
92 struct fork_args {
93 	int     dummy;
94 };
95 #endif
96 
97 int forksleep; /* Place for fork1() to sleep on. */
98 
99 static void
100 init_fork_list(void *data __unused)
101 {
102 
103 	sx_init(&fork_list_lock, "fork list");
104 }
105 SYSINIT(fork_list, SI_SUB_INTRINSIC, SI_ORDER_ANY, init_fork_list, NULL);
106 
107 /*
108  * MPSAFE
109  */
110 /* ARGSUSED */
111 int
112 fork(td, uap)
113 	struct thread *td;
114 	struct fork_args *uap;
115 {
116 	int error;
117 	struct proc *p2;
118 
119 	mtx_lock(&Giant);
120 	error = fork1(td, RFFDG | RFPROC, &p2);
121 	if (error == 0) {
122 		td->td_retval[0] = p2->p_pid;
123 		td->td_retval[1] = 0;
124 	}
125 	mtx_unlock(&Giant);
126 	return error;
127 }
128 
129 /*
130  * MPSAFE
131  */
132 /* ARGSUSED */
133 int
134 vfork(td, uap)
135 	struct thread *td;
136 	struct vfork_args *uap;
137 {
138 	int error;
139 	struct proc *p2;
140 
141 	mtx_lock(&Giant);
142 	error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
143 	if (error == 0) {
144 		td->td_retval[0] = p2->p_pid;
145 		td->td_retval[1] = 0;
146 	}
147 	mtx_unlock(&Giant);
148 	return error;
149 }
150 
151 /*
152  * MPSAFE
153  */
154 int
155 rfork(td, uap)
156 	struct thread *td;
157 	struct rfork_args *uap;
158 {
159 	int error;
160 	struct proc *p2;
161 
162 	/* Don't allow kernel only flags. */
163 	if ((uap->flags & RFKERNELONLY) != 0)
164 		return (EINVAL);
165 	mtx_lock(&Giant);
166 	error = fork1(td, uap->flags, &p2);
167 	if (error == 0) {
168 		td->td_retval[0] = p2 ? p2->p_pid : 0;
169 		td->td_retval[1] = 0;
170 	}
171 	mtx_unlock(&Giant);
172 	return error;
173 }
174 
175 
176 int	nprocs = 1;				/* process 0 */
177 int	lastpid = 0;
178 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
179     "Last used PID");
180 
181 /*
182  * Random component to lastpid generation.  We mix in a random factor to make
183  * it a little harder to predict.  We sanity check the modulus value to avoid
184  * doing it in critical paths.  Don't let it be too small or we pointlessly
185  * waste randomness entropy, and don't let it be impossibly large.  Using a
186  * modulus that is too big causes a LOT more process table scans and slows
187  * down fork processing as the pidchecked caching is defeated.
188  */
189 static int randompid = 0;
190 
191 static int
192 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
193 {
194 	int error, pid;
195 
196 	sx_xlock(&allproc_lock);
197 	pid = randompid;
198 	error = sysctl_handle_int(oidp, &pid, 0, req);
199 	if (error == 0 && req->newptr != NULL) {
200 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
201 			pid = PID_MAX - 100;
202 		else if (pid < 2)			/* NOP */
203 			pid = 0;
204 		else if (pid < 100)			/* Make it reasonable */
205 			pid = 100;
206 		randompid = pid;
207 	}
208 	sx_xunlock(&allproc_lock);
209 	return (error);
210 }
211 
212 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
213     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
214 
215 int
216 fork1(td, flags, procp)
217 	struct thread *td;			/* parent proc */
218 	int flags;
219 	struct proc **procp;			/* child proc */
220 {
221 	struct proc *p2, *pptr;
222 	uid_t uid;
223 	struct proc *newproc;
224 	int trypid;
225 	int ok;
226 	static int pidchecked = 0;
227 	struct forklist *ep;
228 	struct filedesc *fd;
229 	struct proc *p1 = td->td_proc;
230 	struct thread *td2;
231 	struct kse *ke2;
232 	struct ksegrp *kg2;
233 	struct sigacts *newsigacts;
234 	struct procsig *newprocsig;
235 
236 	GIANT_REQUIRED;
237 
238 	/* Can't copy and clear */
239 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
240 		return (EINVAL);
241 
242 	/*
243 	 * Here we don't create a new process, but we divorce
244 	 * certain parts of a process from itself.
245 	 */
246 	if ((flags & RFPROC) == 0) {
247 		vm_forkproc(td, NULL, NULL, flags);
248 
249 		/*
250 		 * Close all file descriptors.
251 		 */
252 		if (flags & RFCFDG) {
253 			struct filedesc *fdtmp;
254 			fdtmp = fdinit(td);	/* XXXKSE */
255 			PROC_LOCK(p1);
256 			fdfree(td);		/* XXXKSE */
257 			p1->p_fd = fdtmp;
258 			PROC_UNLOCK(p1);
259 		}
260 
261 		/*
262 		 * Unshare file descriptors (from parent.)
263 		 */
264 		if (flags & RFFDG) {
265 			FILEDESC_LOCK(p1->p_fd);
266 			if (p1->p_fd->fd_refcnt > 1) {
267 				struct filedesc *newfd;
268 
269 				newfd = fdcopy(td);
270 				FILEDESC_UNLOCK(p1->p_fd);
271 				PROC_LOCK(p1);
272 				fdfree(td);
273 				p1->p_fd = newfd;
274 				PROC_UNLOCK(p1);
275 			} else
276 				FILEDESC_UNLOCK(p1->p_fd);
277 		}
278 		*procp = NULL;
279 		return (0);
280 	}
281 
282 	if (p1->p_flag & P_KSES) {
283 		/*
284 		 * Idle the other threads for a second.
285 		 * Since the user space is copied, it must remain stable.
286 		 * In addition, all threads (from the user perspective)
287 		 * need to either be suspended or in the kernel,
288 		 * where they will try restart in the parent and will
289 		 * be aborted in the child.
290 		 */
291 		PROC_LOCK(p1);
292 		if (thread_single(SNGLE_NO_EXIT)) {
293 			/* Abort.. someone else is single threading before us */
294 			PROC_UNLOCK(p1);
295 			return (ERESTART);
296 		}
297 		PROC_UNLOCK(p1);
298 		/*
299 		 * All other activity in this process
300 		 * is now suspended at the user boundary,
301 		 * (or other safe places if we think of any).
302 		 */
303 	}
304 
305 	/* Allocate new proc. */
306 	newproc = uma_zalloc(proc_zone, M_WAITOK);
307 
308 	/*
309 	 * Although process entries are dynamically created, we still keep
310 	 * a global limit on the maximum number we will create.  Don't allow
311 	 * a nonprivileged user to use the last process; don't let root
312 	 * exceed the limit. The variable nprocs is the current number of
313 	 * processes, maxproc is the limit.
314 	 */
315 	sx_xlock(&allproc_lock);
316 	uid = td->td_ucred->cr_ruid;
317 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
318 		sx_xunlock(&allproc_lock);
319 		uma_zfree(proc_zone, newproc);
320 		if (p1->p_flag & P_KSES) {
321 			PROC_LOCK(p1);
322 			thread_single_end();
323 			PROC_UNLOCK(p1);
324 		}
325 		tsleep(&forksleep, PUSER, "fork", hz / 2);
326 		return (EAGAIN);
327 	}
328 	/*
329 	 * Increment the count of procs running with this uid. Don't allow
330 	 * a nonprivileged user to exceed their current limit.
331 	 */
332 	PROC_LOCK(p1);
333 	ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
334 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
335 	PROC_UNLOCK(p1);
336 	if (!ok) {
337 		sx_xunlock(&allproc_lock);
338 		uma_zfree(proc_zone, newproc);
339 		if (p1->p_flag & P_KSES) {
340 			PROC_LOCK(p1);
341 			thread_single_end();
342 			PROC_UNLOCK(p1);
343 		}
344 		tsleep(&forksleep, PUSER, "fork", hz / 2);
345 		return (EAGAIN);
346 	}
347 
348 	/*
349 	 * Increment the nprocs resource before blocking can occur.  There
350 	 * are hard-limits as to the number of processes that can run.
351 	 */
352 	nprocs++;
353 
354 	/*
355 	 * Find an unused process ID.  We remember a range of unused IDs
356 	 * ready to use (from lastpid+1 through pidchecked-1).
357 	 *
358 	 * If RFHIGHPID is set (used during system boot), do not allocate
359 	 * low-numbered pids.
360 	 */
361 	trypid = lastpid + 1;
362 	if (flags & RFHIGHPID) {
363 		if (trypid < 10) {
364 			trypid = 10;
365 		}
366 	} else {
367 		if (randompid)
368 			trypid += arc4random() % randompid;
369 	}
370 retry:
371 	/*
372 	 * If the process ID prototype has wrapped around,
373 	 * restart somewhat above 0, as the low-numbered procs
374 	 * tend to include daemons that don't exit.
375 	 */
376 	if (trypid >= PID_MAX) {
377 		trypid = trypid % PID_MAX;
378 		if (trypid < 100)
379 			trypid += 100;
380 		pidchecked = 0;
381 	}
382 	if (trypid >= pidchecked) {
383 		int doingzomb = 0;
384 
385 		pidchecked = PID_MAX;
386 		/*
387 		 * Scan the active and zombie procs to check whether this pid
388 		 * is in use.  Remember the lowest pid that's greater
389 		 * than trypid, so we can avoid checking for a while.
390 		 */
391 		p2 = LIST_FIRST(&allproc);
392 again:
393 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
394 			PROC_LOCK(p2);
395 			while (p2->p_pid == trypid ||
396 			    p2->p_pgrp->pg_id == trypid ||
397 			    p2->p_session->s_sid == trypid) {
398 				trypid++;
399 				if (trypid >= pidchecked) {
400 					PROC_UNLOCK(p2);
401 					goto retry;
402 				}
403 			}
404 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
405 				pidchecked = p2->p_pid;
406 			if (p2->p_pgrp->pg_id > trypid &&
407 			    pidchecked > p2->p_pgrp->pg_id)
408 				pidchecked = p2->p_pgrp->pg_id;
409 			if (p2->p_session->s_sid > trypid &&
410 			    pidchecked > p2->p_session->s_sid)
411 				pidchecked = p2->p_session->s_sid;
412 			PROC_UNLOCK(p2);
413 		}
414 		if (!doingzomb) {
415 			doingzomb = 1;
416 			p2 = LIST_FIRST(&zombproc);
417 			goto again;
418 		}
419 	}
420 
421 	/*
422 	 * RFHIGHPID does not mess with the lastpid counter during boot.
423 	 */
424 	if (flags & RFHIGHPID)
425 		pidchecked = 0;
426 	else
427 		lastpid = trypid;
428 
429 	p2 = newproc;
430 	p2->p_state = PRS_NEW;		/* protect against others */
431 	p2->p_pid = trypid;
432 	LIST_INSERT_HEAD(&allproc, p2, p_list);
433 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
434 	sx_xunlock(&allproc_lock);
435 
436 	/*
437 	 * Malloc things while we don't hold any locks.
438 	 */
439 	if (flags & RFSIGSHARE) {
440 		MALLOC(newsigacts, struct sigacts *,
441 		    sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
442 		newprocsig = NULL;
443 	} else {
444 		newsigacts = NULL;
445 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
446 		    M_SUBPROC, M_WAITOK);
447 	}
448 
449 	/*
450 	 * Copy filedesc.
451 	 * XXX: This is busted.  fd*() need to not take proc
452 	 * arguments or something.
453 	 */
454 	if (flags & RFCFDG)
455 		fd = fdinit(td);
456 	else if (flags & RFFDG) {
457 		FILEDESC_LOCK(p1->p_fd);
458 		fd = fdcopy(td);
459 		FILEDESC_UNLOCK(p1->p_fd);
460 	} else
461 		fd = fdshare(p1);
462 
463 	/*
464 	 * Make a proc table entry for the new process.
465 	 * Start by zeroing the section of proc that is zero-initialized,
466 	 * then copy the section that is copied directly from the parent.
467 	 */
468 	td2 = thread_alloc();
469 	ke2 = &p2->p_kse;
470 	kg2 = &p2->p_ksegrp;
471 
472 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start))
473 
474 	bzero(&p2->p_startzero,
475 	    (unsigned) RANGEOF(struct proc, p_startzero, p_endzero));
476 	bzero(&ke2->ke_startzero,
477 	    (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero));
478 #if 0 /* bzero'd by the thread allocator */
479 	bzero(&td2->td_startzero,
480 	    (unsigned) RANGEOF(struct thread, td_startzero, td_endzero));
481 #endif
482 	bzero(&kg2->kg_startzero,
483 	    (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero));
484 
485 	mtx_init(&p2->p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK);
486 	PROC_LOCK(p2);
487 	PROC_LOCK(p1);
488 
489 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
490 	    (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy));
491 	bcopy(&td->td_kse->ke_startcopy, &ke2->ke_startcopy,
492 	    (unsigned) RANGEOF(struct kse, ke_startcopy, ke_endcopy));
493 	bcopy(&td->td_startcopy, &td2->td_startcopy,
494 	    (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy));
495 	bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
496 	    (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy));
497 #undef RANGEOF
498 
499 	/*
500 	 * XXXKSE Theoretically only the running thread would get copied
501 	 * Others in the kernel would be 'aborted' in the child.
502 	 * i.e return E*something*
503 	 * On SMP we would have to stop them running on
504 	 * other CPUs! (set a flag in the proc that stops
505 	 * all returns to userland until completed)
506 	 * This is wrong but ok for 1:1.
507 	 */
508 	proc_linkup(p2, kg2, ke2, td2);
509 
510 	/* Set up the thread as an active thread (as if runnable). */
511 	TAILQ_REMOVE(&kg2->kg_iq, ke2, ke_kgrlist);
512 	kg2->kg_idle_kses--;
513 	ke2->ke_state = KES_THREAD;
514 	ke2->ke_thread = td2;
515 	td2->td_kse = ke2;
516 	td2->td_flags &= ~TDF_UNBOUND; /* For the rest of this syscall. */
517 
518 	/* note.. XXXKSE no pcb or u-area yet */
519 
520 	/*
521 	 * Duplicate sub-structures as needed.
522 	 * Increase reference counts on shared objects.
523 	 * The p_stats and p_sigacts substructs are set in vm_forkproc.
524 	 */
525 	p2->p_flag = 0;
526 	mtx_lock_spin(&sched_lock);
527 	p2->p_sflag = PS_INMEM;
528 	if (p1->p_sflag & PS_PROFIL)
529 		startprofclock(p2);
530 	mtx_unlock_spin(&sched_lock);
531 	p2->p_ucred = crhold(td->td_ucred);
532 	td2->td_ucred = crhold(p2->p_ucred);	/* XXXKSE */
533 
534 	/*
535 	 * Setup linkage for kernel based threading
536 	 */
537 	if((flags & RFTHREAD) != 0) {
538 		/*
539 		 * XXX: This assumes a leader is a parent or grandparent of
540 		 * all processes in a task.
541 		 */
542 		if (p1->p_leader != p1)
543 			PROC_LOCK(p1->p_leader);
544 		p2->p_peers = p1->p_peers;
545 		p1->p_peers = p2;
546 		p2->p_leader = p1->p_leader;
547 		if (p1->p_leader != p1)
548 			PROC_UNLOCK(p1->p_leader);
549 	} else {
550 		p2->p_peers = NULL;
551 		p2->p_leader = p2;
552 	}
553 
554 	pargs_hold(p2->p_args);
555 
556 	if (flags & RFSIGSHARE) {
557 		p2->p_procsig = p1->p_procsig;
558 		p2->p_procsig->ps_refcnt++;
559 		if (p1->p_sigacts == &p1->p_uarea->u_sigacts) {
560 			/*
561 			 * Set p_sigacts to the new shared structure.
562 			 * Note that this is updating p1->p_sigacts at the
563 			 * same time, since p_sigacts is just a pointer to
564 			 * the shared p_procsig->ps_sigacts.
565 			 */
566 			p2->p_sigacts  = newsigacts;
567 			newsigacts = NULL;
568 			*p2->p_sigacts = p1->p_uarea->u_sigacts;
569 		}
570 	} else {
571 		p2->p_procsig = newprocsig;
572 		newprocsig = NULL;
573 		bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
574 		p2->p_procsig->ps_refcnt = 1;
575 		p2->p_sigacts = NULL;	/* finished in vm_forkproc() */
576 	}
577 	if (flags & RFLINUXTHPN)
578 	        p2->p_sigparent = SIGUSR1;
579 	else
580 	        p2->p_sigparent = SIGCHLD;
581 
582 	/* Bump references to the text vnode (for procfs) */
583 	p2->p_textvp = p1->p_textvp;
584 	if (p2->p_textvp)
585 		VREF(p2->p_textvp);
586 	p2->p_fd = fd;
587 	PROC_UNLOCK(p1);
588 	PROC_UNLOCK(p2);
589 
590 	/*
591 	 * If p_limit is still copy-on-write, bump refcnt,
592 	 * otherwise get a copy that won't be modified.
593 	 * (If PL_SHAREMOD is clear, the structure is shared
594 	 * copy-on-write.)
595 	 */
596 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
597 		p2->p_limit = limcopy(p1->p_limit);
598 	else {
599 		p2->p_limit = p1->p_limit;
600 		p2->p_limit->p_refcnt++;
601 	}
602 
603 	sx_xlock(&proctree_lock);
604 	PGRP_LOCK(p1->p_pgrp);
605 	PROC_LOCK(p2);
606 	PROC_LOCK(p1);
607 
608 	/*
609 	 * Preserve some more flags in subprocess.  PS_PROFIL has already
610 	 * been preserved.
611 	 */
612 	p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
613 	SESS_LOCK(p1->p_session);
614 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
615 		p2->p_flag |= P_CONTROLT;
616 	SESS_UNLOCK(p1->p_session);
617 	if (flags & RFPPWAIT)
618 		p2->p_flag |= P_PPWAIT;
619 
620 	LIST_INSERT_AFTER(p1, p2, p_pglist);
621 	PGRP_UNLOCK(p1->p_pgrp);
622 	LIST_INIT(&p2->p_children);
623 	LIST_INIT(&td2->td_contested); /* XXXKSE only 1 thread? */
624 
625 	callout_init(&p2->p_itcallout, 0);
626 	callout_init(&td2->td_slpcallout, 1); /* XXXKSE */
627 
628 #ifdef KTRACE
629 	/*
630 	 * Copy traceflag and tracefile if enabled.
631 	 */
632 	mtx_lock(&ktrace_mtx);
633 	KASSERT(p2->p_tracep == NULL, ("new process has a ktrace vnode"));
634 	if (p1->p_traceflag & KTRFAC_INHERIT) {
635 		p2->p_traceflag = p1->p_traceflag;
636 		if ((p2->p_tracep = p1->p_tracep) != NULL)
637 			VREF(p2->p_tracep);
638 	}
639 	mtx_unlock(&ktrace_mtx);
640 #endif
641 
642 	/*
643 	 * set priority of child to be that of parent
644 	 * XXXKSE hey! copying the estcpu seems dodgy.. should split it..
645 	 */
646 	mtx_lock_spin(&sched_lock);
647 	p2->p_ksegrp.kg_estcpu = p1->p_ksegrp.kg_estcpu;
648 	mtx_unlock_spin(&sched_lock);
649 
650 	/*
651 	 * This begins the section where we must prevent the parent
652 	 * from being swapped.
653 	 */
654 	_PHOLD(p1);
655 	PROC_UNLOCK(p1);
656 
657 	/*
658 	 * Attach the new process to its parent.
659 	 *
660 	 * If RFNOWAIT is set, the newly created process becomes a child
661 	 * of init.  This effectively disassociates the child from the
662 	 * parent.
663 	 */
664 	if (flags & RFNOWAIT)
665 		pptr = initproc;
666 	else
667 		pptr = p1;
668 	p2->p_pptr = pptr;
669 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
670 	PROC_UNLOCK(p2);
671 	sx_xunlock(&proctree_lock);
672 
673 	/*
674 	 * XXXKSE: In KSE, there would be a race here if one thread was
675 	 * dieing due to a signal (or calling exit1() for that matter) while
676 	 * another thread was calling fork1().  Not sure how KSE wants to work
677 	 * around that.  The problem is that up until the point above, if p1
678 	 * gets killed, it won't find p2 in its list in order for it to be
679 	 * reparented.  Alternatively, we could add a new p_flag that gets set
680 	 * before we reparent all the children that we check above and just
681 	 * use init as our parent if that if that flag is set.  (Either that
682 	 * or abort the fork if the flag is set since our parent died trying
683 	 * to fork us (which is evil)).
684 	 */
685 
686 	KASSERT(newprocsig == NULL, ("unused newprocsig"));
687 	if (newsigacts != NULL)
688 		FREE(newsigacts, M_SUBPROC);
689 	/*
690 	 * Finish creating the child process.  It will return via a different
691 	 * execution path later.  (ie: directly into user mode)
692 	 */
693 	vm_forkproc(td, p2, td2, flags);
694 
695 	if (flags == (RFFDG | RFPROC)) {
696 		cnt.v_forks++;
697 		cnt.v_forkpages += p2->p_vmspace->vm_dsize +
698 		    p2->p_vmspace->vm_ssize;
699 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
700 		cnt.v_vforks++;
701 		cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
702 		    p2->p_vmspace->vm_ssize;
703 	} else if (p1 == &proc0) {
704 		cnt.v_kthreads++;
705 		cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
706 		    p2->p_vmspace->vm_ssize;
707 	} else {
708 		cnt.v_rforks++;
709 		cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
710 		    p2->p_vmspace->vm_ssize;
711 	}
712 
713 	/*
714 	 * Both processes are set up, now check if any loadable modules want
715 	 * to adjust anything.
716 	 *   What if they have an error? XXX
717 	 */
718 	sx_slock(&fork_list_lock);
719 	TAILQ_FOREACH(ep, &fork_list, next) {
720 		(*ep->function)(p1, p2, flags);
721 	}
722 	sx_sunlock(&fork_list_lock);
723 
724 	/*
725 	 * If RFSTOPPED not requested, make child runnable and add to
726 	 * run queue.
727 	 */
728 	microtime(&(p2->p_stats->p_start));
729 	p2->p_acflag = AFORK;
730 	if ((flags & RFSTOPPED) == 0) {
731 		mtx_lock_spin(&sched_lock);
732 		p2->p_state = PRS_NORMAL;
733 		setrunqueue(td2);
734 		mtx_unlock_spin(&sched_lock);
735 	}
736 
737 	/*
738 	 * Now can be swapped.
739 	 */
740 	PROC_LOCK(p1);
741 	_PRELE(p1);
742 
743 	/*
744 	 * tell any interested parties about the new process
745 	 */
746 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
747 	PROC_UNLOCK(p1);
748 
749 	/*
750 	 * Preserve synchronization semantics of vfork.  If waiting for
751 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
752 	 * proc (in case of exit).
753 	 */
754 	PROC_LOCK(p2);
755 	while (p2->p_flag & P_PPWAIT)
756 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
757 	PROC_UNLOCK(p2);
758 
759 	/*
760 	 * Return child proc pointer to parent.
761 	 */
762 	*procp = p2;
763 	return (0);
764 }
765 
766 /*
767  * The next two functionms are general routines to handle adding/deleting
768  * items on the fork callout list.
769  *
770  * at_fork():
771  * Take the arguments given and put them onto the fork callout list,
772  * However first make sure that it's not already there.
773  * Returns 0 on success or a standard error number.
774  */
775 
776 int
777 at_fork(function)
778 	forklist_fn function;
779 {
780 	struct forklist *ep;
781 
782 #ifdef INVARIANTS
783 	/* let the programmer know if he's been stupid */
784 	if (rm_at_fork(function))
785 		printf("WARNING: fork callout entry (%p) already present\n",
786 		    function);
787 #endif
788 	ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
789 	if (ep == NULL)
790 		return (ENOMEM);
791 	ep->function = function;
792 	sx_xlock(&fork_list_lock);
793 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
794 	sx_xunlock(&fork_list_lock);
795 	return (0);
796 }
797 
798 /*
799  * Scan the exit callout list for the given item and remove it..
800  * Returns the number of items removed (0 or 1)
801  */
802 
803 int
804 rm_at_fork(function)
805 	forklist_fn function;
806 {
807 	struct forklist *ep;
808 
809 	sx_xlock(&fork_list_lock);
810 	TAILQ_FOREACH(ep, &fork_list, next) {
811 		if (ep->function == function) {
812 			TAILQ_REMOVE(&fork_list, ep, next);
813 			sx_xunlock(&fork_list_lock);
814 			free(ep, M_ATFORK);
815 			return(1);
816 		}
817 	}
818 	sx_xunlock(&fork_list_lock);
819 	return (0);
820 }
821 
822 /*
823  * Handle the return of a child process from fork1().  This function
824  * is called from the MD fork_trampoline() entry point.
825  */
826 void
827 fork_exit(callout, arg, frame)
828 	void (*callout)(void *, struct trapframe *);
829 	void *arg;
830 	struct trapframe *frame;
831 {
832 	struct thread *td = curthread;
833 	struct proc *p = td->td_proc;
834 
835 	td->td_kse->ke_oncpu = PCPU_GET(cpuid);
836 	p->p_state = PRS_NORMAL;
837 	/*
838 	 * Finish setting up thread glue.  We need to initialize
839 	 * the thread into a td_critnest=1 state.  Some platforms
840 	 * may have already partially or fully initialized td_critnest
841 	 * and/or td_md.md_savecrit (when applciable).
842 	 *
843 	 * see <arch>/<arch>/critical.c
844 	 */
845 	sched_lock.mtx_lock = (uintptr_t)td;
846 	sched_lock.mtx_recurse = 0;
847 	cpu_critical_fork_exit();
848 	CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid,
849 	    p->p_comm);
850 	if (PCPU_GET(switchtime.sec) == 0)
851 		binuptime(PCPU_PTR(switchtime));
852 	PCPU_SET(switchticks, ticks);
853 	mtx_unlock_spin(&sched_lock);
854 
855 	/*
856 	 * cpu_set_fork_handler intercepts this function call to
857          * have this call a non-return function to stay in kernel mode.
858          * initproc has its own fork handler, but it does return.
859          */
860 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
861 	callout(arg, frame);
862 
863 	/*
864 	 * Check if a kernel thread misbehaved and returned from its main
865 	 * function.
866 	 */
867 	PROC_LOCK(p);
868 	if (p->p_flag & P_KTHREAD) {
869 		PROC_UNLOCK(p);
870 		mtx_lock(&Giant);
871 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
872 		    p->p_comm, p->p_pid);
873 		kthread_exit(0);
874 	}
875 	PROC_UNLOCK(p);
876 #ifdef DIAGNOSTIC
877 	cred_free_thread(td);
878 #endif
879 	mtx_assert(&Giant, MA_NOTOWNED);
880 }
881 
882 /*
883  * Simplified back end of syscall(), used when returning from fork()
884  * directly into user mode.  Giant is not held on entry, and must not
885  * be held on return.  This function is passed in to fork_exit() as the
886  * first parameter and is called when returning to a new userland process.
887  */
888 void
889 fork_return(td, frame)
890 	struct thread *td;
891 	struct trapframe *frame;
892 {
893 
894 	userret(td, frame, 0);
895 #ifdef KTRACE
896 	if (KTRPOINT(td, KTR_SYSRET))
897 		ktrsysret(SYS_fork, 0, 0);
898 #endif
899 	mtx_assert(&Giant, MA_NOTOWNED);
900 }
901