xref: /dflybsd-src/sys/kern/kern_fork.c (revision 4bfff613235390e418af6fe340d465e53afba83a)
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: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $
40  * $DragonFly: src/sys/kern/kern_fork.c,v 1.77 2008/05/18 20:02:02 nth Exp $
41  */
42 
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/proc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/vnode.h>
55 #include <sys/acct.h>
56 #include <sys/ktrace.h>
57 #include <sys/unistd.h>
58 #include <sys/jail.h>
59 #include <sys/caps.h>
60 
61 #include <vm/vm.h>
62 #include <sys/lock.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_extern.h>
66 
67 #include <sys/vmmeter.h>
68 #include <sys/refcount.h>
69 #include <sys/thread2.h>
70 #include <sys/signal2.h>
71 #include <sys/spinlock2.h>
72 
73 #include <sys/dsched.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 TAILQ_HEAD(forklist_head, forklist);
87 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
88 
89 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
90 
91 int forksleep; /* Place for fork1() to sleep on. */
92 
93 /*
94  * Red-Black tree support for LWPs
95  */
96 
97 static int
98 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
99 {
100 	if (lp1->lwp_tid < lp2->lwp_tid)
101 		return(-1);
102 	if (lp1->lwp_tid > lp2->lwp_tid)
103 		return(1);
104 	return(0);
105 }
106 
107 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
108 
109 /*
110  * Fork system call
111  *
112  * MPALMOSTSAFE
113  */
114 int
115 sys_fork(struct fork_args *uap)
116 {
117 	struct lwp *lp = curthread->td_lwp;
118 	struct proc *p2;
119 	int error;
120 
121 	error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
122 	if (error == 0) {
123 		start_forked_proc(lp, p2);
124 		uap->sysmsg_fds[0] = p2->p_pid;
125 		uap->sysmsg_fds[1] = 0;
126 	}
127 	return error;
128 }
129 
130 /*
131  * MPALMOSTSAFE
132  */
133 int
134 sys_vfork(struct vfork_args *uap)
135 {
136 	struct lwp *lp = curthread->td_lwp;
137 	struct proc *p2;
138 	int error;
139 
140 	error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
141 	if (error == 0) {
142 		start_forked_proc(lp, p2);
143 		uap->sysmsg_fds[0] = p2->p_pid;
144 		uap->sysmsg_fds[1] = 0;
145 	}
146 	return error;
147 }
148 
149 /*
150  * Handle rforks.  An rfork may (1) operate on the current process without
151  * creating a new, (2) create a new process that shared the current process's
152  * vmspace, signals, and/or descriptors, or (3) create a new process that does
153  * not share these things (normal fork).
154  *
155  * Note that we only call start_forked_proc() if a new process is actually
156  * created.
157  *
158  * rfork { int flags }
159  *
160  * MPALMOSTSAFE
161  */
162 int
163 sys_rfork(struct rfork_args *uap)
164 {
165 	struct lwp *lp = curthread->td_lwp;
166 	struct proc *p2;
167 	int error;
168 
169 	if ((uap->flags & RFKERNELONLY) != 0)
170 		return (EINVAL);
171 
172 	error = fork1(lp, uap->flags | RFPGLOCK, &p2);
173 	if (error == 0) {
174 		if (p2)
175 			start_forked_proc(lp, p2);
176 		uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
177 		uap->sysmsg_fds[1] = 0;
178 	}
179 	return error;
180 }
181 
182 /*
183  * MPALMOSTSAFE
184  */
185 int
186 sys_lwp_create(struct lwp_create_args *uap)
187 {
188 	struct proc *p = curproc;
189 	struct lwp *lp;
190 	struct lwp_params params;
191 	int error;
192 
193 	error = copyin(uap->params, &params, sizeof(params));
194 	if (error)
195 		goto fail2;
196 
197 	lwkt_gettoken(&p->p_token);
198 	plimit_lwp_fork(p);	/* force exclusive access */
199 	lp = lwp_fork(curthread->td_lwp, p, RFPROC);
200 	error = cpu_prepare_lwp(lp, &params);
201 	if (params.tid1 != NULL &&
202 	    (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
203 		goto fail;
204 	if (params.tid2 != NULL &&
205 	    (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
206 		goto fail;
207 
208 	/*
209 	 * Now schedule the new lwp.
210 	 */
211 	p->p_usched->resetpriority(lp);
212 	crit_enter();
213 	lp->lwp_stat = LSRUN;
214 	p->p_usched->setrunqueue(lp);
215 	crit_exit();
216 	lwkt_reltoken(&p->p_token);
217 
218 	return (0);
219 
220 fail:
221 	lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
222 	--p->p_nthreads;
223 	/* lwp_dispose expects an exited lwp, and a held proc */
224 	lp->lwp_flag |= LWP_WEXIT;
225 	lp->lwp_thread->td_flags |= TDF_EXITING;
226 	PHOLD(p);
227 	lwp_dispose(lp);
228 	lwkt_reltoken(&p->p_token);
229 fail2:
230 	return (error);
231 }
232 
233 int	nprocs = 1;		/* process 0 */
234 
235 int
236 fork1(struct lwp *lp1, int flags, struct proc **procp)
237 {
238 	struct proc *p1 = lp1->lwp_proc;
239 	struct proc *p2, *pptr;
240 	struct pgrp *p1grp;
241 	struct pgrp *plkgrp;
242 	uid_t uid;
243 	int ok, error;
244 	static int curfail = 0;
245 	static struct timeval lastfail;
246 	struct forklist *ep;
247 	struct filedesc_to_leader *fdtol;
248 
249 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
250 		return (EINVAL);
251 
252 	lwkt_gettoken(&p1->p_token);
253 	plkgrp = NULL;
254 
255 	/*
256 	 * Here we don't create a new process, but we divorce
257 	 * certain parts of a process from itself.
258 	 */
259 	if ((flags & RFPROC) == 0) {
260 		/*
261 		 * This kind of stunt does not work anymore if
262 		 * there are native threads (lwps) running
263 		 */
264 		if (p1->p_nthreads != 1) {
265 			error = EINVAL;
266 			goto done;
267 		}
268 
269 		vm_fork(p1, 0, flags);
270 
271 		/*
272 		 * Close all file descriptors.
273 		 */
274 		if (flags & RFCFDG) {
275 			struct filedesc *fdtmp;
276 			fdtmp = fdinit(p1);
277 			fdfree(p1, fdtmp);
278 		}
279 
280 		/*
281 		 * Unshare file descriptors (from parent.)
282 		 */
283 		if (flags & RFFDG) {
284 			if (p1->p_fd->fd_refcnt > 1) {
285 				struct filedesc *newfd;
286 				newfd = fdcopy(p1);
287 				fdfree(p1, newfd);
288 			}
289 		}
290 		*procp = NULL;
291 		error = 0;
292 		goto done;
293 	}
294 
295 	/*
296 	 * Interlock against process group signal delivery.  If signals
297 	 * are pending after the interlock is obtained we have to restart
298 	 * the system call to process the signals.  If we don't the child
299 	 * can miss a pgsignal (such as ^C) sent during the fork.
300 	 *
301 	 * We can't use CURSIG() here because it will process any STOPs
302 	 * and cause the process group lock to be held indefinitely.  If
303 	 * a STOP occurs, the fork will be restarted after the CONT.
304 	 */
305 	p1grp = p1->p_pgrp;
306 	if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
307 		pgref(plkgrp);
308 		lockmgr(&plkgrp->pg_lock, LK_SHARED);
309 		if (CURSIG_NOBLOCK(lp1)) {
310 			error = ERESTART;
311 			goto done;
312 		}
313 	}
314 
315 	/*
316 	 * Although process entries are dynamically created, we still keep
317 	 * a global limit on the maximum number we will create.  Don't allow
318 	 * a nonprivileged user to use the last ten processes; don't let root
319 	 * exceed the limit. The variable nprocs is the current number of
320 	 * processes, maxproc is the limit.
321 	 */
322 	uid = lp1->lwp_thread->td_ucred->cr_ruid;
323 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
324 		if (ppsratecheck(&lastfail, &curfail, 1))
325 			kprintf("maxproc limit exceeded by uid %d, please "
326 			       "see tuning(7) and login.conf(5).\n", uid);
327 		tsleep(&forksleep, 0, "fork", hz / 2);
328 		error = EAGAIN;
329 		goto done;
330 	}
331 	/*
332 	 * Increment the nprocs resource before blocking can occur.  There
333 	 * are hard-limits as to the number of processes that can run.
334 	 */
335 	nprocs++;
336 
337 	/*
338 	 * Increment the count of procs running with this uid. Don't allow
339 	 * a nonprivileged user to exceed their current limit.
340 	 */
341 	ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
342 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
343 	if (!ok) {
344 		/*
345 		 * Back out the process count
346 		 */
347 		nprocs--;
348 		if (ppsratecheck(&lastfail, &curfail, 1))
349 			kprintf("maxproc limit exceeded by uid %d, please "
350 			       "see tuning(7) and login.conf(5).\n", uid);
351 		tsleep(&forksleep, 0, "fork", hz / 2);
352 		error = EAGAIN;
353 		goto done;
354 	}
355 
356 	/* Allocate new proc. */
357 	p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
358 
359 	/*
360 	 * Setup linkage for kernel based threading XXX lwp
361 	 */
362 	if (flags & RFTHREAD) {
363 		p2->p_peers = p1->p_peers;
364 		p1->p_peers = p2;
365 		p2->p_leader = p1->p_leader;
366 	} else {
367 		p2->p_leader = p2;
368 	}
369 
370 	RB_INIT(&p2->p_lwp_tree);
371 	spin_init(&p2->p_spin);
372 	lwkt_token_init(&p2->p_token, "iproc");
373 	p2->p_lasttid = -1;	/* first tid will be 0 */
374 
375 	/*
376 	 * Setting the state to SIDL protects the partially initialized
377 	 * process once it starts getting hooked into the rest of the system.
378 	 */
379 	p2->p_stat = SIDL;
380 	proc_add_allproc(p2);
381 
382 	/*
383 	 * Make a proc table entry for the new process.
384 	 * The whole structure was zeroed above, so copy the section that is
385 	 * copied directly from the parent.
386 	 */
387 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
388 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
389 
390 	/*
391 	 * Duplicate sub-structures as needed.  Increase reference counts
392 	 * on shared objects.
393 	 *
394 	 * NOTE: because we are now on the allproc list it is possible for
395 	 *	 other consumers to gain temporary references to p2
396 	 *	 (p2->p_lock can change).
397 	 */
398 	if (p1->p_flag & P_PROFIL)
399 		startprofclock(p2);
400 	p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
401 
402 	if (jailed(p2->p_ucred))
403 		p2->p_flag |= P_JAILED;
404 
405 	if (p2->p_args)
406 		refcount_acquire(&p2->p_args->ar_ref);
407 
408 	p2->p_usched = p1->p_usched;
409 	/* XXX: verify copy of the secondary iosched stuff */
410 	dsched_new_proc(p2);
411 
412 	if (flags & RFSIGSHARE) {
413 		p2->p_sigacts = p1->p_sigacts;
414 		refcount_acquire(&p2->p_sigacts->ps_refcnt);
415 	} else {
416 		p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
417 					M_SUBPROC, M_WAITOK);
418 		bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
419 		refcount_init(&p2->p_sigacts->ps_refcnt, 1);
420 	}
421 	if (flags & RFLINUXTHPN)
422 	        p2->p_sigparent = SIGUSR1;
423 	else
424 	        p2->p_sigparent = SIGCHLD;
425 
426 	/* bump references to the text vnode (for procfs) */
427 	p2->p_textvp = p1->p_textvp;
428 	if (p2->p_textvp)
429 		vref(p2->p_textvp);
430 
431 	/* copy namecache handle to the text file */
432 	if (p1->p_textnch.mount)
433 		cache_copy(&p1->p_textnch, &p2->p_textnch);
434 
435 	/*
436 	 * Handle file descriptors
437 	 */
438 	if (flags & RFCFDG) {
439 		p2->p_fd = fdinit(p1);
440 		fdtol = NULL;
441 	} else if (flags & RFFDG) {
442 		p2->p_fd = fdcopy(p1);
443 		fdtol = NULL;
444 	} else {
445 		p2->p_fd = fdshare(p1);
446 		if (p1->p_fdtol == NULL) {
447 			lwkt_gettoken(&p1->p_token);
448 			p1->p_fdtol =
449 				filedesc_to_leader_alloc(NULL,
450 							 p1->p_leader);
451 			lwkt_reltoken(&p1->p_token);
452 		}
453 		if ((flags & RFTHREAD) != 0) {
454 			/*
455 			 * Shared file descriptor table and
456 			 * shared process leaders.
457 			 */
458 			fdtol = p1->p_fdtol;
459 			fdtol->fdl_refcount++;
460 		} else {
461 			/*
462 			 * Shared file descriptor table, and
463 			 * different process leaders
464 			 */
465 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
466 		}
467 	}
468 	p2->p_fdtol = fdtol;
469 	p2->p_limit = plimit_fork(p1);
470 
471 	/*
472 	 * Preserve some more flags in subprocess.  P_PROFIL has already
473 	 * been preserved.
474 	 */
475 	p2->p_flag |= p1->p_flag & P_SUGID;
476 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
477 		p2->p_flag |= P_CONTROLT;
478 	if (flags & RFPPWAIT)
479 		p2->p_flag |= P_PPWAIT;
480 
481 	/*
482 	 * Inherit the virtual kernel structure (allows a virtual kernel
483 	 * to fork to simulate multiple cpus).
484 	 */
485 	if (p1->p_vkernel)
486 		vkernel_inherit(p1, p2);
487 
488 	/*
489 	 * Once we are on a pglist we may receive signals.  XXX we might
490 	 * race a ^C being sent to the process group by not receiving it
491 	 * at all prior to this line.
492 	 */
493 	pgref(p1grp);
494 	lwkt_gettoken(&p1grp->pg_token);
495 	LIST_INSERT_AFTER(p1, p2, p_pglist);
496 	lwkt_reltoken(&p1grp->pg_token);
497 
498 	/*
499 	 * Attach the new process to its parent.
500 	 *
501 	 * If RFNOWAIT is set, the newly created process becomes a child
502 	 * of init.  This effectively disassociates the child from the
503 	 * parent.
504 	 */
505 	if (flags & RFNOWAIT)
506 		pptr = initproc;
507 	else
508 		pptr = p1;
509 	p2->p_pptr = pptr;
510 	LIST_INIT(&p2->p_children);
511 
512 	lwkt_gettoken(&pptr->p_token);
513 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
514 	lwkt_reltoken(&pptr->p_token);
515 
516 	varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
517 	callout_init(&p2->p_ithandle);
518 
519 #ifdef KTRACE
520 	/*
521 	 * Copy traceflag and tracefile if enabled.  If not inherited,
522 	 * these were zeroed above but we still could have a trace race
523 	 * so make sure p2's p_tracenode is NULL.
524 	 */
525 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
526 		p2->p_traceflag = p1->p_traceflag;
527 		p2->p_tracenode = ktrinherit(p1->p_tracenode);
528 	}
529 #endif
530 
531 	/*
532 	 * This begins the section where we must prevent the parent
533 	 * from being swapped.
534 	 *
535 	 * Gets PRELE'd in the caller in start_forked_proc().
536 	 */
537 	PHOLD(p1);
538 
539 	vm_fork(p1, p2, flags);
540 
541 	/*
542 	 * Create the first lwp associated with the new proc.
543 	 * It will return via a different execution path later, directly
544 	 * into userland, after it was put on the runq by
545 	 * start_forked_proc().
546 	 */
547 	lwp_fork(lp1, p2, flags);
548 
549 	if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
550 		mycpu->gd_cnt.v_forks++;
551 		mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
552 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
553 		mycpu->gd_cnt.v_vforks++;
554 		mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
555 	} else if (p1 == &proc0) {
556 		mycpu->gd_cnt.v_kthreads++;
557 		mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
558 	} else {
559 		mycpu->gd_cnt.v_rforks++;
560 		mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
561 	}
562 
563 	/*
564 	 * Both processes are set up, now check if any loadable modules want
565 	 * to adjust anything.
566 	 *   What if they have an error? XXX
567 	 */
568 	TAILQ_FOREACH(ep, &fork_list, next) {
569 		(*ep->function)(p1, p2, flags);
570 	}
571 
572 	/*
573 	 * Set the start time.  Note that the process is not runnable.  The
574 	 * caller is responsible for making it runnable.
575 	 */
576 	microtime(&p2->p_start);
577 	p2->p_acflag = AFORK;
578 
579 	/*
580 	 * tell any interested parties about the new process
581 	 */
582 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
583 
584 	/*
585 	 * Return child proc pointer to parent.
586 	 */
587 	*procp = p2;
588 	error = 0;
589 done:
590 	lwkt_reltoken(&p1->p_token);
591 	if (plkgrp) {
592 		lockmgr(&plkgrp->pg_lock, LK_RELEASE);
593 		pgrel(plkgrp);
594 	}
595 	return (error);
596 }
597 
598 static struct lwp *
599 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
600 {
601 	struct lwp *lp;
602 	struct thread *td;
603 
604 	lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
605 
606 	lp->lwp_proc = destproc;
607 	lp->lwp_vmspace = destproc->p_vmspace;
608 	lp->lwp_stat = LSRUN;
609 	bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
610 	    (unsigned) ((caddr_t)&lp->lwp_endcopy -
611 			(caddr_t)&lp->lwp_startcopy));
612 	lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
613 	/*
614 	 * Set cpbase to the last timeout that occured (not the upcoming
615 	 * timeout).
616 	 *
617 	 * A critical section is required since a timer IPI can update
618 	 * scheduler specific data.
619 	 */
620 	crit_enter();
621 	lp->lwp_cpbase = mycpu->gd_schedclock.time -
622 			mycpu->gd_schedclock.periodic;
623 	destproc->p_usched->heuristic_forking(origlp, lp);
624 	crit_exit();
625 	lp->lwp_cpumask &= usched_mastermask;
626 
627 	/*
628 	 * Assign a TID to the lp.  Loop until the insert succeeds (returns
629 	 * NULL).
630 	 */
631 	lp->lwp_tid = destproc->p_lasttid;
632 	do {
633 		if (++lp->lwp_tid < 0)
634 			lp->lwp_tid = 1;
635 	} while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
636 	destproc->p_lasttid = lp->lwp_tid;
637 	destproc->p_nthreads++;
638 
639 	td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
640 	lp->lwp_thread = td;
641 	td->td_proc = destproc;
642 	td->td_lwp = lp;
643 	td->td_switch = cpu_heavy_switch;
644 	lwkt_setpri(td, TDPRI_KERN_USER);
645 	lwkt_set_comm(td, "%s", destproc->p_comm);
646 
647 	/*
648 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
649 	 * and make the child ready to run.
650 	 */
651 	cpu_fork(origlp, lp, flags);
652 	caps_fork(origlp->lwp_thread, lp->lwp_thread);
653 	kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
654 
655 	return (lp);
656 }
657 
658 /*
659  * The next two functionms are general routines to handle adding/deleting
660  * items on the fork callout list.
661  *
662  * at_fork():
663  * Take the arguments given and put them onto the fork callout list,
664  * However first make sure that it's not already there.
665  * Returns 0 on success or a standard error number.
666  */
667 int
668 at_fork(forklist_fn function)
669 {
670 	struct forklist *ep;
671 
672 #ifdef INVARIANTS
673 	/* let the programmer know if he's been stupid */
674 	if (rm_at_fork(function)) {
675 		kprintf("WARNING: fork callout entry (%p) already present\n",
676 		    function);
677 	}
678 #endif
679 	ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
680 	ep->function = function;
681 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
682 	return (0);
683 }
684 
685 /*
686  * Scan the exit callout list for the given item and remove it..
687  * Returns the number of items removed (0 or 1)
688  */
689 int
690 rm_at_fork(forklist_fn function)
691 {
692 	struct forklist *ep;
693 
694 	TAILQ_FOREACH(ep, &fork_list, next) {
695 		if (ep->function == function) {
696 			TAILQ_REMOVE(&fork_list, ep, next);
697 			kfree(ep, M_ATFORK);
698 			return(1);
699 		}
700 	}
701 	return (0);
702 }
703 
704 /*
705  * Add a forked process to the run queue after any remaining setup, such
706  * as setting the fork handler, has been completed.
707  */
708 void
709 start_forked_proc(struct lwp *lp1, struct proc *p2)
710 {
711 	struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
712 
713 	/*
714 	 * Move from SIDL to RUN queue, and activate the process's thread.
715 	 * Activation of the thread effectively makes the process "a"
716 	 * current process, so we do not setrunqueue().
717 	 *
718 	 * YYY setrunqueue works here but we should clean up the trampoline
719 	 * code so we just schedule the LWKT thread and let the trampoline
720 	 * deal with the userland scheduler on return to userland.
721 	 */
722 	KASSERT(p2->p_stat == SIDL,
723 	    ("cannot start forked process, bad status: %p", p2));
724 	p2->p_usched->resetpriority(lp2);
725 	crit_enter();
726 	p2->p_stat = SACTIVE;
727 	lp2->lwp_stat = LSRUN;
728 	p2->p_usched->setrunqueue(lp2);
729 	crit_exit();
730 
731 	/*
732 	 * Now can be swapped.
733 	 */
734 	PRELE(lp1->lwp_proc);
735 
736 	/*
737 	 * Preserve synchronization semantics of vfork.  If waiting for
738 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
739 	 * proc (in case of exit).
740 	 */
741 	while (p2->p_flag & P_PPWAIT)
742 		tsleep(lp1->lwp_proc, 0, "ppwait", 0);
743 }
744