xref: /dragonfly/sys/kern/kern_fork.c (revision 31524921)
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. 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  * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $
36  */
37 
38 #include "opt_ktrace.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/filedesc.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/malloc.h>
47 #include <sys/proc.h>
48 #include <sys/resourcevar.h>
49 #include <sys/vnode.h>
50 #include <sys/acct.h>
51 #include <sys/ktrace.h>
52 #include <sys/unistd.h>
53 #include <sys/jail.h>
54 
55 #include <vm/vm.h>
56 #include <sys/lock.h>
57 #include <vm/pmap.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_extern.h>
60 
61 #include <sys/vmmeter.h>
62 #include <sys/refcount.h>
63 #include <sys/thread2.h>
64 #include <sys/signal2.h>
65 #include <sys/spinlock2.h>
66 
67 #include <sys/dsched.h>
68 
69 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
70 static MALLOC_DEFINE(M_REAPER, "reaper", "process reapers");
71 
72 /*
73  * These are the stuctures used to create a callout list for things to do
74  * when forking a process
75  */
76 struct forklist {
77 	forklist_fn function;
78 	TAILQ_ENTRY(forklist) next;
79 };
80 
81 TAILQ_HEAD(forklist_head, forklist);
82 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
83 
84 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
85 
86 int forksleep; /* Place for fork1() to sleep on. */
87 
88 /*
89  * Red-Black tree support for LWPs
90  */
91 
92 static int
93 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
94 {
95 	if (lp1->lwp_tid < lp2->lwp_tid)
96 		return(-1);
97 	if (lp1->lwp_tid > lp2->lwp_tid)
98 		return(1);
99 	return(0);
100 }
101 
102 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
103 
104 /*
105  * fork() system call
106  */
107 int
108 sys_fork(struct fork_args *uap)
109 {
110 	struct lwp *lp = curthread->td_lwp;
111 	struct proc *p2;
112 	int error;
113 
114 	error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
115 	if (error == 0) {
116 		PHOLD(p2);
117 		start_forked_proc(lp, p2);
118 		uap->sysmsg_fds[0] = p2->p_pid;
119 		uap->sysmsg_fds[1] = 0;
120 		PRELE(p2);
121 	}
122 	return error;
123 }
124 
125 /*
126  * vfork() system call
127  */
128 int
129 sys_vfork(struct vfork_args *uap)
130 {
131 	struct lwp *lp = curthread->td_lwp;
132 	struct proc *p2;
133 	int error;
134 
135 	error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
136 	if (error == 0) {
137 		PHOLD(p2);
138 		start_forked_proc(lp, p2);
139 		uap->sysmsg_fds[0] = p2->p_pid;
140 		uap->sysmsg_fds[1] = 0;
141 		PRELE(p2);
142 	}
143 	return error;
144 }
145 
146 /*
147  * Handle rforks.  An rfork may (1) operate on the current process without
148  * creating a new, (2) create a new process that shared the current process's
149  * vmspace, signals, and/or descriptors, or (3) create a new process that does
150  * not share these things (normal fork).
151  *
152  * Note that we only call start_forked_proc() if a new process is actually
153  * created.
154  *
155  * rfork { int flags }
156  */
157 int
158 sys_rfork(struct rfork_args *uap)
159 {
160 	struct lwp *lp = curthread->td_lwp;
161 	struct proc *p2;
162 	int error;
163 
164 	if ((uap->flags & RFKERNELONLY) != 0)
165 		return (EINVAL);
166 
167 	error = fork1(lp, uap->flags | RFPGLOCK, &p2);
168 	if (error == 0) {
169 		if (p2) {
170 			PHOLD(p2);
171 			start_forked_proc(lp, p2);
172 			uap->sysmsg_fds[0] = p2->p_pid;
173 			uap->sysmsg_fds[1] = 0;
174 			PRELE(p2);
175 		} else {
176 			uap->sysmsg_fds[0] = 0;
177 			uap->sysmsg_fds[1] = 0;
178 		}
179 	}
180 	return error;
181 }
182 
183 /*
184  * Low level thread create used by pthreads.
185  */
186 int
187 sys_lwp_create(struct lwp_create_args *uap)
188 {
189 	struct proc *p = curproc;
190 	struct lwp *lp;
191 	struct lwp_params params;
192 	int error;
193 
194 	error = copyin(uap->params, &params, sizeof(params));
195 	if (error)
196 		goto fail2;
197 
198 	lwkt_gettoken(&p->p_token);
199 	plimit_lwp_fork(p);	/* force exclusive access */
200 	lp = lwp_fork(curthread->td_lwp, p, RFPROC);
201 	error = cpu_prepare_lwp(lp, &params);
202 	if (error)
203 		goto fail;
204 	if (params.lwp_tid1 != NULL &&
205 	    (error = copyout(&lp->lwp_tid, params.lwp_tid1, sizeof(lp->lwp_tid))))
206 		goto fail;
207 	if (params.lwp_tid2 != NULL &&
208 	    (error = copyout(&lp->lwp_tid, params.lwp_tid2, sizeof(lp->lwp_tid))))
209 		goto fail;
210 
211 	/*
212 	 * Now schedule the new lwp.
213 	 */
214 	p->p_usched->resetpriority(lp);
215 	crit_enter();
216 	lp->lwp_stat = LSRUN;
217 	p->p_usched->setrunqueue(lp);
218 	crit_exit();
219 	lwkt_reltoken(&p->p_token);
220 
221 	return (0);
222 
223 fail:
224 	lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
225 	--p->p_nthreads;
226 	/* lwp_dispose expects an exited lwp, and a held proc */
227 	atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
228 	lp->lwp_thread->td_flags |= TDF_EXITING;
229 	lwkt_remove_tdallq(lp->lwp_thread);
230 	PHOLD(p);
231 	biosched_done(lp->lwp_thread);
232 	dsched_exit_thread(lp->lwp_thread);
233 	lwp_dispose(lp);
234 	lwkt_reltoken(&p->p_token);
235 fail2:
236 	return (error);
237 }
238 
239 int	nprocs = 1;		/* process 0 */
240 
241 int
242 fork1(struct lwp *lp1, int flags, struct proc **procp)
243 {
244 	struct proc *p1 = lp1->lwp_proc;
245 	struct proc *p2;
246 	struct proc *pptr;
247 	struct pgrp *p1grp;
248 	struct pgrp *plkgrp;
249 	struct sysreaper *reap;
250 	uid_t uid;
251 	int ok, error;
252 	static int curfail = 0;
253 	static struct timeval lastfail;
254 	struct forklist *ep;
255 	struct filedesc_to_leader *fdtol;
256 
257 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
258 		return (EINVAL);
259 
260 	lwkt_gettoken(&p1->p_token);
261 	plkgrp = NULL;
262 	p2 = NULL;
263 
264 	/*
265 	 * Here we don't create a new process, but we divorce
266 	 * certain parts of a process from itself.
267 	 */
268 	if ((flags & RFPROC) == 0) {
269 		/*
270 		 * This kind of stunt does not work anymore if
271 		 * there are native threads (lwps) running
272 		 */
273 		if (p1->p_nthreads != 1) {
274 			error = EINVAL;
275 			goto done;
276 		}
277 
278 		vm_fork(p1, 0, flags);
279 
280 		/*
281 		 * Close all file descriptors.
282 		 */
283 		if (flags & RFCFDG) {
284 			struct filedesc *fdtmp;
285 			fdtmp = fdinit(p1);
286 			fdfree(p1, fdtmp);
287 		}
288 
289 		/*
290 		 * Unshare file descriptors (from parent.)
291 		 */
292 		if (flags & RFFDG) {
293 			if (p1->p_fd->fd_refcnt > 1) {
294 				struct filedesc *newfd;
295 				error = fdcopy(p1, &newfd);
296 				if (error != 0) {
297 					error = ENOMEM;
298 					goto done;
299 				}
300 				fdfree(p1, newfd);
301 			}
302 		}
303 		*procp = NULL;
304 		error = 0;
305 		goto done;
306 	}
307 
308 	/*
309 	 * Interlock against process group signal delivery.  If signals
310 	 * are pending after the interlock is obtained we have to restart
311 	 * the system call to process the signals.  If we don't the child
312 	 * can miss a pgsignal (such as ^C) sent during the fork.
313 	 *
314 	 * We can't use CURSIG() here because it will process any STOPs
315 	 * and cause the process group lock to be held indefinitely.  If
316 	 * a STOP occurs, the fork will be restarted after the CONT.
317 	 */
318 	p1grp = p1->p_pgrp;
319 	if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
320 		pgref(plkgrp);
321 		lockmgr(&plkgrp->pg_lock, LK_SHARED);
322 		if (CURSIG_NOBLOCK(lp1)) {
323 			error = ERESTART;
324 			goto done;
325 		}
326 	}
327 
328 	/*
329 	 * Although process entries are dynamically created, we still keep
330 	 * a global limit on the maximum number we will create.  Don't allow
331 	 * a nonprivileged user to use the last ten processes; don't let root
332 	 * exceed the limit. The variable nprocs is the current number of
333 	 * processes, maxproc is the limit.
334 	 */
335 	uid = lp1->lwp_thread->td_ucred->cr_ruid;
336 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
337 		if (ppsratecheck(&lastfail, &curfail, 1))
338 			kprintf("maxproc limit exceeded by uid %d, please "
339 			       "see tuning(7) and login.conf(5).\n", uid);
340 		tsleep(&forksleep, 0, "fork", hz / 2);
341 		error = EAGAIN;
342 		goto done;
343 	}
344 
345 	/*
346 	 * Increment the nprocs resource before blocking can occur.  There
347 	 * are hard-limits as to the number of processes that can run.
348 	 */
349 	atomic_add_int(&nprocs, 1);
350 
351 	/*
352 	 * Increment the count of procs running with this uid. Don't allow
353 	 * a nonprivileged user to exceed their current limit.
354 	 */
355 	ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
356 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
357 	if (!ok) {
358 		/*
359 		 * Back out the process count
360 		 */
361 		atomic_add_int(&nprocs, -1);
362 		if (ppsratecheck(&lastfail, &curfail, 1))
363 			kprintf("maxproc limit exceeded by uid %d, please "
364 			       "see tuning(7) and login.conf(5).\n", uid);
365 		tsleep(&forksleep, 0, "fork", hz / 2);
366 		error = EAGAIN;
367 		goto done;
368 	}
369 
370 	/*
371 	 * Allocate a new process, don't get fancy: zero the structure.
372 	 */
373 	p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
374 
375 	/*
376 	 * Core initialization.  SIDL is a safety state that protects the
377 	 * partially initialized process once it starts getting hooked
378 	 * into system structures and becomes addressable.
379 	 *
380 	 * We must be sure to acquire p2->p_token as well, we must hold it
381 	 * once the process is on the allproc list to avoid things such
382 	 * as competing modifications to p_flags.
383 	 */
384 	mycpu->gd_forkid += ncpus;
385 	p2->p_forkid = mycpu->gd_forkid + mycpu->gd_cpuid;
386 	p2->p_lasttid = -1;	/* first tid will be 0 */
387 	p2->p_stat = SIDL;
388 
389 	/*
390 	 * NOTE: Process 0 will not have a reaper, but process 1 (init) and
391 	 *	 all other processes always will.
392 	 */
393 	if ((reap = p1->p_reaper) != NULL) {
394 		reaper_hold(reap);
395 		p2->p_reaper = reap;
396 	} else {
397 		p2->p_reaper = NULL;
398 	}
399 
400 	RB_INIT(&p2->p_lwp_tree);
401 	spin_init(&p2->p_spin, "procfork1");
402 	lwkt_token_init(&p2->p_token, "proc");
403 	lwkt_gettoken(&p2->p_token);
404 
405 	/*
406 	 * Setup linkage for kernel based threading XXX lwp.  Also add the
407 	 * process to the allproclist.
408 	 *
409 	 * The process structure is addressable after this point.
410 	 */
411 	if (flags & RFTHREAD) {
412 		p2->p_peers = p1->p_peers;
413 		p1->p_peers = p2;
414 		p2->p_leader = p1->p_leader;
415 	} else {
416 		p2->p_leader = p2;
417 	}
418 	proc_add_allproc(p2);
419 
420 	/*
421 	 * Initialize the section which is copied verbatim from the parent.
422 	 */
423 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
424 	      ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
425 
426 	/*
427 	 * Duplicate sub-structures as needed.  Increase reference counts
428 	 * on shared objects.
429 	 *
430 	 * NOTE: because we are now on the allproc list it is possible for
431 	 *	 other consumers to gain temporary references to p2
432 	 *	 (p2->p_lock can change).
433 	 */
434 	if (p1->p_flags & P_PROFIL)
435 		startprofclock(p2);
436 	p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
437 
438 	if (jailed(p2->p_ucred))
439 		p2->p_flags |= P_JAILED;
440 
441 	if (p2->p_args)
442 		refcount_acquire(&p2->p_args->ar_ref);
443 
444 	p2->p_usched = p1->p_usched;
445 	/* XXX: verify copy of the secondary iosched stuff */
446 	dsched_new_proc(p2);
447 
448 	if (flags & RFSIGSHARE) {
449 		p2->p_sigacts = p1->p_sigacts;
450 		refcount_acquire(&p2->p_sigacts->ps_refcnt);
451 	} else {
452 		p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
453 					M_SUBPROC, M_WAITOK);
454 		bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
455 		refcount_init(&p2->p_sigacts->ps_refcnt, 1);
456 	}
457 	if (flags & RFLINUXTHPN)
458 	        p2->p_sigparent = SIGUSR1;
459 	else
460 	        p2->p_sigparent = SIGCHLD;
461 
462 	/* bump references to the text vnode (for procfs) */
463 	p2->p_textvp = p1->p_textvp;
464 	if (p2->p_textvp)
465 		vref(p2->p_textvp);
466 
467 	/* copy namecache handle to the text file */
468 	if (p1->p_textnch.mount)
469 		cache_copy(&p1->p_textnch, &p2->p_textnch);
470 
471 	/*
472 	 * Handle file descriptors
473 	 */
474 	if (flags & RFCFDG) {
475 		p2->p_fd = fdinit(p1);
476 		fdtol = NULL;
477 	} else if (flags & RFFDG) {
478 		error = fdcopy(p1, &p2->p_fd);
479 		if (error != 0) {
480 			error = ENOMEM;
481 			goto done;
482 		}
483 		fdtol = NULL;
484 	} else {
485 		p2->p_fd = fdshare(p1);
486 		if (p1->p_fdtol == NULL) {
487 			p1->p_fdtol = filedesc_to_leader_alloc(NULL,
488 							       p1->p_leader);
489 		}
490 		if ((flags & RFTHREAD) != 0) {
491 			/*
492 			 * Shared file descriptor table and
493 			 * shared process leaders.
494 			 */
495 			fdtol = p1->p_fdtol;
496 			fdtol->fdl_refcount++;
497 		} else {
498 			/*
499 			 * Shared file descriptor table, and
500 			 * different process leaders
501 			 */
502 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
503 		}
504 	}
505 	p2->p_fdtol = fdtol;
506 	p2->p_limit = plimit_fork(p1);
507 
508 	/*
509 	 * Preserve some more flags in subprocess.  P_PROFIL has already
510 	 * been preserved.
511 	 */
512 	p2->p_flags |= p1->p_flags & P_SUGID;
513 	if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT))
514 		p2->p_flags |= P_CONTROLT;
515 	if (flags & RFPPWAIT) {
516 		p2->p_flags |= P_PPWAIT;
517 		if (p1->p_upmap)
518 			p1->p_upmap->invfork = 1;
519 	}
520 
521 
522 	/*
523 	 * Inherit the virtual kernel structure (allows a virtual kernel
524 	 * to fork to simulate multiple cpus).
525 	 */
526 	if (p1->p_vkernel)
527 		vkernel_inherit(p1, p2);
528 
529 	/*
530 	 * Once we are on a pglist we may receive signals.  XXX we might
531 	 * race a ^C being sent to the process group by not receiving it
532 	 * at all prior to this line.
533 	 */
534 	pgref(p1grp);
535 	lwkt_gettoken(&p1grp->pg_token);
536 	LIST_INSERT_AFTER(p1, p2, p_pglist);
537 	lwkt_reltoken(&p1grp->pg_token);
538 
539 	/*
540 	 * Attach the new process to its parent.
541 	 *
542 	 * If RFNOWAIT is set, the newly created process becomes a child
543 	 * of the reaper (typically init).  This effectively disassociates
544 	 * the child from the parent.
545 	 *
546 	 * Temporarily hold pptr for the RFNOWAIT case to avoid ripouts.
547 	 */
548 	if (flags & RFNOWAIT) {
549 		pptr = reaper_get(reap);
550 		if (pptr == NULL) {
551 			pptr = initproc;
552 			PHOLD(pptr);
553 		}
554 	} else {
555 		pptr = p1;
556 	}
557 	p2->p_pptr = pptr;
558 	LIST_INIT(&p2->p_children);
559 
560 	lwkt_gettoken(&pptr->p_token);
561 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
562 	lwkt_reltoken(&pptr->p_token);
563 
564 	if (flags & RFNOWAIT)
565 		PRELE(pptr);
566 
567 	varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
568 	callout_init_mp(&p2->p_ithandle);
569 
570 #ifdef KTRACE
571 	/*
572 	 * Copy traceflag and tracefile if enabled.  If not inherited,
573 	 * these were zeroed above but we still could have a trace race
574 	 * so make sure p2's p_tracenode is NULL.
575 	 */
576 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
577 		p2->p_traceflag = p1->p_traceflag;
578 		p2->p_tracenode = ktrinherit(p1->p_tracenode);
579 	}
580 #endif
581 
582 	/*
583 	 * This begins the section where we must prevent the parent
584 	 * from being swapped.
585 	 *
586 	 * Gets PRELE'd in the caller in start_forked_proc().
587 	 */
588 	PHOLD(p1);
589 
590 	vm_fork(p1, p2, flags);
591 
592 	/*
593 	 * Create the first lwp associated with the new proc.
594 	 * It will return via a different execution path later, directly
595 	 * into userland, after it was put on the runq by
596 	 * start_forked_proc().
597 	 */
598 	lwp_fork(lp1, p2, flags);
599 
600 	if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
601 		mycpu->gd_cnt.v_forks++;
602 		mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize +
603 					     p2->p_vmspace->vm_ssize;
604 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
605 		mycpu->gd_cnt.v_vforks++;
606 		mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
607 					      p2->p_vmspace->vm_ssize;
608 	} else if (p1 == &proc0) {
609 		mycpu->gd_cnt.v_kthreads++;
610 		mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
611 						p2->p_vmspace->vm_ssize;
612 	} else {
613 		mycpu->gd_cnt.v_rforks++;
614 		mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
615 					      p2->p_vmspace->vm_ssize;
616 	}
617 
618 	/*
619 	 * Both processes are set up, now check if any loadable modules want
620 	 * to adjust anything.
621 	 *   What if they have an error? XXX
622 	 */
623 	TAILQ_FOREACH(ep, &fork_list, next) {
624 		(*ep->function)(p1, p2, flags);
625 	}
626 
627 	/*
628 	 * Set the start time.  Note that the process is not runnable.  The
629 	 * caller is responsible for making it runnable.
630 	 */
631 	microtime(&p2->p_start);
632 	p2->p_acflag = AFORK;
633 
634 	/*
635 	 * tell any interested parties about the new process
636 	 */
637 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
638 
639 	/*
640 	 * Return child proc pointer to parent.
641 	 */
642 	*procp = p2;
643 	error = 0;
644 done:
645 	if (p2)
646 		lwkt_reltoken(&p2->p_token);
647 	lwkt_reltoken(&p1->p_token);
648 	if (plkgrp) {
649 		lockmgr(&plkgrp->pg_lock, LK_RELEASE);
650 		pgrel(plkgrp);
651 	}
652 	return (error);
653 }
654 
655 static struct lwp *
656 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
657 {
658 	globaldata_t gd = mycpu;
659 	struct lwp *lp;
660 	struct thread *td;
661 
662 	lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
663 
664 	lp->lwp_proc = destproc;
665 	lp->lwp_vmspace = destproc->p_vmspace;
666 	lp->lwp_stat = LSRUN;
667 	bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
668 	    (unsigned) ((caddr_t)&lp->lwp_endcopy -
669 			(caddr_t)&lp->lwp_startcopy));
670 	lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK;
671 	/*
672 	 * Set cpbase to the last timeout that occured (not the upcoming
673 	 * timeout).
674 	 *
675 	 * A critical section is required since a timer IPI can update
676 	 * scheduler specific data.
677 	 */
678 	crit_enter();
679 	lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic;
680 	destproc->p_usched->heuristic_forking(origlp, lp);
681 	crit_exit();
682 	CPUMASK_ANDMASK(lp->lwp_cpumask, usched_mastermask);
683 	lwkt_token_init(&lp->lwp_token, "lwp_token");
684 	spin_init(&lp->lwp_spin, "lwptoken");
685 
686 	/*
687 	 * Assign the thread to the current cpu to begin with so we
688 	 * can manipulate it.
689 	 */
690 	td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0);
691 	lp->lwp_thread = td;
692 	td->td_ucred = crhold(destproc->p_ucred);
693 	td->td_proc = destproc;
694 	td->td_lwp = lp;
695 	td->td_switch = cpu_heavy_switch;
696 #ifdef NO_LWKT_SPLIT_USERPRI
697 	lwkt_setpri(td, TDPRI_USER_NORM);
698 #else
699 	lwkt_setpri(td, TDPRI_KERN_USER);
700 #endif
701 	lwkt_set_comm(td, "%s", destproc->p_comm);
702 
703 	/*
704 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
705 	 * and make the child ready to run.
706 	 */
707 	cpu_fork(origlp, lp, flags);
708 	kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
709 
710 	/*
711 	 * Assign a TID to the lp.  Loop until the insert succeeds (returns
712 	 * NULL).
713 	 */
714 	lp->lwp_tid = destproc->p_lasttid;
715 	do {
716 		if (++lp->lwp_tid < 0)
717 			lp->lwp_tid = 1;
718 	} while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
719 	destproc->p_lasttid = lp->lwp_tid;
720 	destproc->p_nthreads++;
721 
722 	/*
723 	 * This flag is set and never cleared.  It means that the process
724 	 * was threaded at some point.  Used to improve exit performance.
725 	 */
726 	destproc->p_flags |= P_MAYBETHREADED;
727 
728 	return (lp);
729 }
730 
731 /*
732  * The next two functionms are general routines to handle adding/deleting
733  * items on the fork callout list.
734  *
735  * at_fork():
736  * Take the arguments given and put them onto the fork callout list,
737  * However first make sure that it's not already there.
738  * Returns 0 on success or a standard error number.
739  */
740 int
741 at_fork(forklist_fn function)
742 {
743 	struct forklist *ep;
744 
745 #ifdef INVARIANTS
746 	/* let the programmer know if he's been stupid */
747 	if (rm_at_fork(function)) {
748 		kprintf("WARNING: fork callout entry (%p) already present\n",
749 		    function);
750 	}
751 #endif
752 	ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
753 	ep->function = function;
754 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
755 	return (0);
756 }
757 
758 /*
759  * Scan the exit callout list for the given item and remove it..
760  * Returns the number of items removed (0 or 1)
761  */
762 int
763 rm_at_fork(forklist_fn function)
764 {
765 	struct forklist *ep;
766 
767 	TAILQ_FOREACH(ep, &fork_list, next) {
768 		if (ep->function == function) {
769 			TAILQ_REMOVE(&fork_list, ep, next);
770 			kfree(ep, M_ATFORK);
771 			return(1);
772 		}
773 	}
774 	return (0);
775 }
776 
777 /*
778  * Add a forked process to the run queue after any remaining setup, such
779  * as setting the fork handler, has been completed.
780  *
781  * p2 is held by the caller.
782  */
783 void
784 start_forked_proc(struct lwp *lp1, struct proc *p2)
785 {
786 	struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
787 	int pflags;
788 
789 	/*
790 	 * Move from SIDL to RUN queue, and activate the process's thread.
791 	 * Activation of the thread effectively makes the process "a"
792 	 * current process, so we do not setrunqueue().
793 	 *
794 	 * YYY setrunqueue works here but we should clean up the trampoline
795 	 * code so we just schedule the LWKT thread and let the trampoline
796 	 * deal with the userland scheduler on return to userland.
797 	 */
798 	KASSERT(p2->p_stat == SIDL,
799 	    ("cannot start forked process, bad status: %p", p2));
800 	p2->p_usched->resetpriority(lp2);
801 	crit_enter();
802 	p2->p_stat = SACTIVE;
803 	lp2->lwp_stat = LSRUN;
804 	p2->p_usched->setrunqueue(lp2);
805 	crit_exit();
806 
807 	/*
808 	 * Now can be swapped.
809 	 */
810 	PRELE(lp1->lwp_proc);
811 
812 	/*
813 	 * Preserve synchronization semantics of vfork.  P_PPWAIT is set in
814 	 * the child until it has retired the parent's resources.  The parent
815 	 * must wait for the flag to be cleared by the child.
816 	 *
817 	 * Interlock the flag/tsleep with atomic ops to avoid unnecessary
818 	 * p_token conflicts.
819 	 *
820 	 * XXX Is this use of an atomic op on a field that is not normally
821 	 *     manipulated with atomic ops ok?
822 	 */
823 	while ((pflags = p2->p_flags) & P_PPWAIT) {
824 		cpu_ccfence();
825 		tsleep_interlock(lp1->lwp_proc, 0);
826 		if (atomic_cmpset_int(&p2->p_flags, pflags, pflags))
827 			tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0);
828 	}
829 }
830 
831 /*
832  * procctl (idtype_t idtype, id_t id, int cmd, void *arg)
833  */
834 int
835 sys_procctl(struct procctl_args *uap)
836 {
837 	struct proc *p = curproc;
838 	struct proc *p2;
839 	struct sysreaper *reap;
840 	union reaper_info udata;
841 	int error;
842 
843 	if (uap->idtype != P_PID || uap->id != (id_t)p->p_pid)
844 		return EINVAL;
845 
846 	switch(uap->cmd) {
847 	case PROC_REAP_ACQUIRE:
848 		lwkt_gettoken(&p->p_token);
849 		reap = kmalloc(sizeof(*reap), M_REAPER, M_WAITOK|M_ZERO);
850 		if (p->p_reaper == NULL || p->p_reaper->p != p) {
851 			reaper_init(p, reap);
852 			error = 0;
853 		} else {
854 			kfree(reap, M_REAPER);
855 			error = EALREADY;
856 		}
857 		lwkt_reltoken(&p->p_token);
858 		break;
859 	case PROC_REAP_RELEASE:
860 		lwkt_gettoken(&p->p_token);
861 release_again:
862 		reap = p->p_reaper;
863 		KKASSERT(reap != NULL);
864 		if (reap->p == p) {
865 			reaper_hold(reap);	/* in case of thread race */
866 			lockmgr(&reap->lock, LK_EXCLUSIVE);
867 			if (reap->p != p) {
868 				lockmgr(&reap->lock, LK_RELEASE);
869 				reaper_drop(reap);
870 				goto release_again;
871 			}
872 			reap->p = NULL;
873 			p->p_reaper = reap->parent;
874 			if (p->p_reaper)
875 				reaper_hold(p->p_reaper);
876 			lockmgr(&reap->lock, LK_RELEASE);
877 			reaper_drop(reap);	/* our ref */
878 			reaper_drop(reap);	/* old p_reaper ref */
879 			error = 0;
880 		} else {
881 			error = ENOTCONN;
882 		}
883 		lwkt_reltoken(&p->p_token);
884 		break;
885 	case PROC_REAP_STATUS:
886 		bzero(&udata, sizeof(udata));
887 		lwkt_gettoken_shared(&p->p_token);
888 		if ((reap = p->p_reaper) != NULL && reap->p == p) {
889 			udata.status.flags = reap->flags;
890 			udata.status.refs = reap->refs - 1; /* minus ours */
891 		}
892 		p2 = LIST_FIRST(&p->p_children);
893 		udata.status.pid_head = p2 ? p2->p_pid : -1;
894 		lwkt_reltoken(&p->p_token);
895 
896 		if (uap->data) {
897 			error = copyout(&udata, uap->data,
898 					sizeof(udata.status));
899 		} else {
900 			error = 0;
901 		}
902 		break;
903 	default:
904 		error = EINVAL;
905 		break;
906 	}
907 	return error;
908 }
909 
910 /*
911  * Bump ref on reaper, preventing destruction
912  */
913 void
914 reaper_hold(struct sysreaper *reap)
915 {
916 	KKASSERT(reap->refs > 0);
917 	refcount_acquire(&reap->refs);
918 }
919 
920 /*
921  * Drop ref on reaper, destroy the structure on the 1->0
922  * transition and loop on the parent.
923  */
924 void
925 reaper_drop(struct sysreaper *next)
926 {
927 	struct sysreaper *reap;
928 
929 	while ((reap = next) != NULL) {
930 		if (refcount_release(&reap->refs)) {
931 			next = reap->parent;
932 			KKASSERT(reap->p == NULL);
933 			reap->parent = NULL;
934 			kfree(reap, M_REAPER);
935 		} else {
936 			next = NULL;
937 		}
938 	}
939 }
940 
941 /*
942  * Initialize a static or newly allocated reaper structure
943  */
944 void
945 reaper_init(struct proc *p, struct sysreaper *reap)
946 {
947 	reap->parent = p->p_reaper;
948 	reap->p = p;
949 	if (p == initproc) {
950 		reap->flags = REAPER_STAT_OWNED | REAPER_STAT_REALINIT;
951 		reap->refs = 2;
952 	} else {
953 		reap->flags = REAPER_STAT_OWNED;
954 		reap->refs = 1;
955 	}
956 	lockinit(&reap->lock, "subrp", 0, 0);
957 	cpu_sfence();
958 	p->p_reaper = reap;
959 }
960 
961 /*
962  * Called with p->p_token held during exit.
963  *
964  * This is a bit simpler than RELEASE because there are no threads remaining
965  * to race.  We only release if we own the reaper, the exit code will handle
966  * the final p_reaper release.
967  */
968 struct sysreaper *
969 reaper_exit(struct proc *p)
970 {
971 	struct sysreaper *reap;
972 
973 	/*
974 	 * Release acquired reaper
975 	 */
976 	if ((reap = p->p_reaper) != NULL && reap->p == p) {
977 		lockmgr(&reap->lock, LK_EXCLUSIVE);
978 		p->p_reaper = reap->parent;
979 		if (p->p_reaper)
980 			reaper_hold(p->p_reaper);
981 		reap->p = NULL;
982 		lockmgr(&reap->lock, LK_RELEASE);
983 		reaper_drop(reap);
984 	}
985 
986 	/*
987 	 * Return and clear reaper (caller is holding p_token for us)
988 	 * (reap->p does not equal p).  Caller must drop it.
989 	 */
990 	if ((reap = p->p_reaper) != NULL) {
991 		p->p_reaper = NULL;
992 	}
993 	return reap;
994 }
995 
996 /*
997  * Return a held (PHOLD) process representing the reaper for process (p).
998  * NULL should not normally be returned.  Caller should PRELE() the returned
999  * reaper process when finished.
1000  *
1001  * Remove dead internal nodes while we are at it.
1002  *
1003  * Process (p)'s token must be held on call.
1004  * The returned process's token is NOT acquired by this routine.
1005  */
1006 struct proc *
1007 reaper_get(struct sysreaper *reap)
1008 {
1009 	struct sysreaper *next;
1010 	struct proc *reproc;
1011 
1012 	if (reap == NULL)
1013 		return NULL;
1014 
1015 	/*
1016 	 * Extra hold for loop
1017 	 */
1018 	reaper_hold(reap);
1019 
1020 	while (reap) {
1021 		lockmgr(&reap->lock, LK_SHARED);
1022 		if (reap->p) {
1023 			/*
1024 			 * Probable reaper
1025 			 */
1026 			if (reap->p) {
1027 				reproc = reap->p;
1028 				PHOLD(reproc);
1029 				lockmgr(&reap->lock, LK_RELEASE);
1030 				reaper_drop(reap);
1031 				return reproc;
1032 			}
1033 
1034 			/*
1035 			 * Raced, try again
1036 			 */
1037 			lockmgr(&reap->lock, LK_RELEASE);
1038 			continue;
1039 		}
1040 
1041 		/*
1042 		 * Traverse upwards in the reaper topology, destroy
1043 		 * dead internal nodes when possible.
1044 		 *
1045 		 * NOTE: Our ref on next means that a dead node should
1046 		 *	 have 2 (ours and reap->parent's).
1047 		 */
1048 		next = reap->parent;
1049 		while (next) {
1050 			reaper_hold(next);
1051 			if (next->refs == 2 && next->p == NULL) {
1052 				lockmgr(&reap->lock, LK_RELEASE);
1053 				lockmgr(&reap->lock, LK_EXCLUSIVE);
1054 				if (next->refs == 2 &&
1055 				    reap->parent == next &&
1056 				    next->p == NULL) {
1057 					/*
1058 					 * reap->parent inherits ref from next.
1059 					 */
1060 					reap->parent = next->parent;
1061 					next->parent = NULL;
1062 					reaper_drop(next);	/* ours */
1063 					reaper_drop(next);	/* old parent */
1064 					next = reap->parent;
1065 					continue;	/* possible chain */
1066 				}
1067 			}
1068 			break;
1069 		}
1070 		lockmgr(&reap->lock, LK_RELEASE);
1071 		reaper_drop(reap);
1072 		reap = next;
1073 	}
1074 	return NULL;
1075 }
1076