xref: /freebsd/sys/kern/kern_fork.c (revision dce6e6518b85561495cff38a3074a69d29d58a55)
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  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ktrace.h"
45 #include "opt_mac.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/sysproto.h>
50 #include <sys/eventhandler.h>
51 #include <sys/filedesc.h>
52 #include <sys/kernel.h>
53 #include <sys/sysctl.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/mutex.h>
57 #include <sys/proc.h>
58 #include <sys/pioctl.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
61 #include <sys/syscall.h>
62 #include <sys/vnode.h>
63 #include <sys/acct.h>
64 #include <sys/mac.h>
65 #include <sys/ktr.h>
66 #include <sys/ktrace.h>
67 #include <sys/kthread.h>
68 #include <sys/unistd.h>
69 #include <sys/jail.h>
70 #include <sys/sx.h>
71 
72 #include <vm/vm.h>
73 #include <vm/pmap.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_extern.h>
76 #include <vm/uma.h>
77 
78 #include <sys/vmmeter.h>
79 #include <sys/user.h>
80 #include <machine/critical.h>
81 
82 #ifndef _SYS_SYSPROTO_H_
83 struct fork_args {
84 	int     dummy;
85 };
86 #endif
87 
88 static int forksleep; /* Place for fork1() to sleep on. */
89 
90 /*
91  * MPSAFE
92  */
93 /* ARGSUSED */
94 int
95 fork(td, uap)
96 	struct thread *td;
97 	struct fork_args *uap;
98 {
99 	int error;
100 	struct proc *p2;
101 
102 	error = fork1(td, RFFDG | RFPROC, 0, &p2);
103 	if (error == 0) {
104 		td->td_retval[0] = p2->p_pid;
105 		td->td_retval[1] = 0;
106 	}
107 	return error;
108 }
109 
110 /*
111  * MPSAFE
112  */
113 /* ARGSUSED */
114 int
115 vfork(td, uap)
116 	struct thread *td;
117 	struct vfork_args *uap;
118 {
119 	int error;
120 	struct proc *p2;
121 
122 	error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
123 	if (error == 0) {
124 		td->td_retval[0] = p2->p_pid;
125 		td->td_retval[1] = 0;
126 	}
127 	return error;
128 }
129 
130 /*
131  * MPSAFE
132  */
133 int
134 rfork(td, uap)
135 	struct thread *td;
136 	struct rfork_args *uap;
137 {
138 	int error;
139 	struct proc *p2;
140 
141 	/* Don't allow kernel only flags. */
142 	if ((uap->flags & RFKERNELONLY) != 0)
143 		return (EINVAL);
144 	error = fork1(td, uap->flags, 0, &p2);
145 	if (error == 0) {
146 		td->td_retval[0] = p2 ? p2->p_pid : 0;
147 		td->td_retval[1] = 0;
148 	}
149 	return error;
150 }
151 
152 
153 int	nprocs = 1;				/* process 0 */
154 int	lastpid = 0;
155 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
156     "Last used PID");
157 
158 /*
159  * Random component to lastpid generation.  We mix in a random factor to make
160  * it a little harder to predict.  We sanity check the modulus value to avoid
161  * doing it in critical paths.  Don't let it be too small or we pointlessly
162  * waste randomness entropy, and don't let it be impossibly large.  Using a
163  * modulus that is too big causes a LOT more process table scans and slows
164  * down fork processing as the pidchecked caching is defeated.
165  */
166 static int randompid = 0;
167 
168 static int
169 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
170 {
171 	int error, pid;
172 
173 	sysctl_wire_old_buffer(req, sizeof(int));
174 	sx_xlock(&allproc_lock);
175 	pid = randompid;
176 	error = sysctl_handle_int(oidp, &pid, 0, req);
177 	if (error == 0 && req->newptr != NULL) {
178 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
179 			pid = PID_MAX - 100;
180 		else if (pid < 2)			/* NOP */
181 			pid = 0;
182 		else if (pid < 100)			/* Make it reasonable */
183 			pid = 100;
184 		randompid = pid;
185 	}
186 	sx_xunlock(&allproc_lock);
187 	return (error);
188 }
189 
190 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
191     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
192 
193 int
194 fork1(td, flags, pages, procp)
195 	struct thread *td;			/* parent proc */
196 	int flags;
197 	int pages;
198 	struct proc **procp;			/* child proc */
199 {
200 	struct proc *p2, *pptr;
201 	uid_t uid;
202 	struct proc *newproc;
203 	int trypid;
204 	int ok;
205 	static int curfail, pidchecked = 0;
206 	static struct timeval lastfail;
207 	struct filedesc *fd;
208 	struct filedesc_to_leader *fdtol;
209 	struct proc *p1 = td->td_proc;
210 	struct thread *td2;
211 	struct kse *ke2;
212 	struct ksegrp *kg2;
213 	struct sigacts *newsigacts;
214 	int error;
215 
216 	/* Can't copy and clear */
217 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
218 		return (EINVAL);
219 
220 	mtx_lock(&Giant);
221 	/*
222 	 * Here we don't create a new process, but we divorce
223 	 * certain parts of a process from itself.
224 	 */
225 	if ((flags & RFPROC) == 0) {
226 		vm_forkproc(td, NULL, NULL, flags);
227 
228 		/*
229 		 * Close all file descriptors.
230 		 */
231 		if (flags & RFCFDG) {
232 			struct filedesc *fdtmp;
233 			fdtmp = fdinit(td->td_proc->p_fd);
234 			fdfree(td);
235 			p1->p_fd = fdtmp;
236 		}
237 
238 		/*
239 		 * Unshare file descriptors (from parent.)
240 		 */
241 		if (flags & RFFDG) {
242 			FILEDESC_LOCK(p1->p_fd);
243 			if (p1->p_fd->fd_refcnt > 1) {
244 				struct filedesc *newfd;
245 
246 				newfd = fdcopy(td->td_proc->p_fd);
247 				FILEDESC_UNLOCK(p1->p_fd);
248 				fdfree(td);
249 				p1->p_fd = newfd;
250 			} else
251 				FILEDESC_UNLOCK(p1->p_fd);
252 		}
253 		mtx_unlock(&Giant);
254 		*procp = NULL;
255 		return (0);
256 	}
257 
258 	/*
259 	 * Note 1:1 allows for forking with one thread coming out on the
260 	 * other side with the expectation that the process is about to
261 	 * exec.
262 	 */
263 	if (p1->p_flag & P_SA) {
264 		/*
265 		 * Idle the other threads for a second.
266 		 * Since the user space is copied, it must remain stable.
267 		 * In addition, all threads (from the user perspective)
268 		 * need to either be suspended or in the kernel,
269 		 * where they will try restart in the parent and will
270 		 * be aborted in the child.
271 		 */
272 		PROC_LOCK(p1);
273 		if (thread_single(SINGLE_NO_EXIT)) {
274 			/* Abort.. someone else is single threading before us */
275 			PROC_UNLOCK(p1);
276 			mtx_unlock(&Giant);
277 			return (ERESTART);
278 		}
279 		PROC_UNLOCK(p1);
280 		/*
281 		 * All other activity in this process
282 		 * is now suspended at the user boundary,
283 		 * (or other safe places if we think of any).
284 		 */
285 	}
286 
287 	/* Allocate new proc. */
288 	newproc = uma_zalloc(proc_zone, M_WAITOK);
289 #ifdef MAC
290 	mac_init_proc(newproc);
291 #endif
292 
293 	/*
294 	 * Although process entries are dynamically created, we still keep
295 	 * a global limit on the maximum number we will create.  Don't allow
296 	 * a nonprivileged user to use the last ten processes; don't let root
297 	 * exceed the limit. The variable nprocs is the current number of
298 	 * processes, maxproc is the limit.
299 	 */
300 	sx_xlock(&allproc_lock);
301 	uid = td->td_ucred->cr_ruid;
302 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
303 		error = EAGAIN;
304 		goto fail;
305 	}
306 
307 	/*
308 	 * Increment the count of procs running with this uid. Don't allow
309 	 * a nonprivileged user to exceed their current limit.
310 	 */
311 	PROC_LOCK(p1);
312 	ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
313 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
314 	PROC_UNLOCK(p1);
315 	if (!ok) {
316 		error = EAGAIN;
317 		goto fail;
318 	}
319 
320 	/*
321 	 * Increment the nprocs resource before blocking can occur.  There
322 	 * are hard-limits as to the number of processes that can run.
323 	 */
324 	nprocs++;
325 
326 	/*
327 	 * Find an unused process ID.  We remember a range of unused IDs
328 	 * ready to use (from lastpid+1 through pidchecked-1).
329 	 *
330 	 * If RFHIGHPID is set (used during system boot), do not allocate
331 	 * low-numbered pids.
332 	 */
333 	trypid = lastpid + 1;
334 	if (flags & RFHIGHPID) {
335 		if (trypid < 10) {
336 			trypid = 10;
337 		}
338 	} else {
339 		if (randompid)
340 			trypid += arc4random() % randompid;
341 	}
342 retry:
343 	/*
344 	 * If the process ID prototype has wrapped around,
345 	 * restart somewhat above 0, as the low-numbered procs
346 	 * tend to include daemons that don't exit.
347 	 */
348 	if (trypid >= PID_MAX) {
349 		trypid = trypid % PID_MAX;
350 		if (trypid < 100)
351 			trypid += 100;
352 		pidchecked = 0;
353 	}
354 	if (trypid >= pidchecked) {
355 		int doingzomb = 0;
356 
357 		pidchecked = PID_MAX;
358 		/*
359 		 * Scan the active and zombie procs to check whether this pid
360 		 * is in use.  Remember the lowest pid that's greater
361 		 * than trypid, so we can avoid checking for a while.
362 		 */
363 		p2 = LIST_FIRST(&allproc);
364 again:
365 		for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
366 			PROC_LOCK(p2);
367 			while (p2->p_pid == trypid ||
368 			    p2->p_pgrp->pg_id == trypid ||
369 			    p2->p_session->s_sid == trypid) {
370 				trypid++;
371 				if (trypid >= pidchecked) {
372 					PROC_UNLOCK(p2);
373 					goto retry;
374 				}
375 			}
376 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
377 				pidchecked = p2->p_pid;
378 			if (p2->p_pgrp->pg_id > trypid &&
379 			    pidchecked > p2->p_pgrp->pg_id)
380 				pidchecked = p2->p_pgrp->pg_id;
381 			if (p2->p_session->s_sid > trypid &&
382 			    pidchecked > p2->p_session->s_sid)
383 				pidchecked = p2->p_session->s_sid;
384 			PROC_UNLOCK(p2);
385 		}
386 		if (!doingzomb) {
387 			doingzomb = 1;
388 			p2 = LIST_FIRST(&zombproc);
389 			goto again;
390 		}
391 	}
392 
393 	/*
394 	 * RFHIGHPID does not mess with the lastpid counter during boot.
395 	 */
396 	if (flags & RFHIGHPID)
397 		pidchecked = 0;
398 	else
399 		lastpid = trypid;
400 
401 	p2 = newproc;
402 	p2->p_state = PRS_NEW;		/* protect against others */
403 	p2->p_pid = trypid;
404 	LIST_INSERT_HEAD(&allproc, p2, p_list);
405 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
406 	sx_xunlock(&allproc_lock);
407 
408 	/*
409 	 * Malloc things while we don't hold any locks.
410 	 */
411 	if (flags & RFSIGSHARE)
412 		newsigacts = NULL;
413 	else
414 		newsigacts = sigacts_alloc();
415 
416 	/*
417 	 * Copy filedesc.
418 	 */
419 	if (flags & RFCFDG) {
420 		fd = fdinit(td->td_proc->p_fd);
421 		fdtol = NULL;
422 	} else if (flags & RFFDG) {
423 		FILEDESC_LOCK(p1->p_fd);
424 		fd = fdcopy(td->td_proc->p_fd);
425 		FILEDESC_UNLOCK(p1->p_fd);
426 		fdtol = NULL;
427 	} else {
428 		fd = fdshare(p1->p_fd);
429 		if (p1->p_fdtol == NULL)
430 			p1->p_fdtol =
431 				filedesc_to_leader_alloc(NULL,
432 							 NULL,
433 							 p1->p_leader);
434 		if ((flags & RFTHREAD) != 0) {
435 			/*
436 			 * Shared file descriptor table and
437 			 * shared process leaders.
438 			 */
439 			fdtol = p1->p_fdtol;
440 			FILEDESC_LOCK(p1->p_fd);
441 			fdtol->fdl_refcount++;
442 			FILEDESC_UNLOCK(p1->p_fd);
443 		} else {
444 			/*
445 			 * Shared file descriptor table, and
446 			 * different process leaders
447 			 */
448 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
449 							 p1->p_fd,
450 							 p2);
451 		}
452 	}
453 	/*
454 	 * Make a proc table entry for the new process.
455 	 * Start by zeroing the section of proc that is zero-initialized,
456 	 * then copy the section that is copied directly from the parent.
457 	 */
458 	td2 = FIRST_THREAD_IN_PROC(p2);
459 	kg2 = FIRST_KSEGRP_IN_PROC(p2);
460 	ke2 = FIRST_KSE_IN_KSEGRP(kg2);
461 
462 	/* Allocate and switch to an alternate kstack if specified */
463 	if (pages != 0)
464 		vm_thread_new_altkstack(td2, pages);
465 
466 	PROC_LOCK(p2);
467 	PROC_LOCK(p1);
468 
469 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start))
470 
471 	bzero(&p2->p_startzero,
472 	    (unsigned) RANGEOF(struct proc, p_startzero, p_endzero));
473 	bzero(&ke2->ke_startzero,
474 	    (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero));
475 	bzero(&td2->td_startzero,
476 	    (unsigned) RANGEOF(struct thread, td_startzero, td_endzero));
477 	bzero(&kg2->kg_startzero,
478 	    (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero));
479 
480 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
481 	    (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy));
482 	bcopy(&td->td_startcopy, &td2->td_startcopy,
483 	    (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy));
484 	bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
485 	    (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy));
486 #undef RANGEOF
487 
488 	/* Set up the thread as an active thread (as if runnable). */
489 	ke2->ke_state = KES_THREAD;
490 	ke2->ke_thread = td2;
491 	td2->td_kse = ke2;
492 
493 	/*
494 	 * Duplicate sub-structures as needed.
495 	 * Increase reference counts on shared objects.
496 	 * The p_stats substruct is set in vm_forkproc.
497 	 */
498 	p2->p_flag = 0;
499 	if (p1->p_flag & P_PROFIL)
500 		startprofclock(p2);
501 	mtx_lock_spin(&sched_lock);
502 	p2->p_sflag = PS_INMEM;
503 	/*
504 	 * Allow the scheduler to adjust the priority of the child and
505 	 * parent while we hold the sched_lock.
506 	 */
507 	sched_fork(p1, p2);
508 
509 	mtx_unlock_spin(&sched_lock);
510 	p2->p_ucred = crhold(td->td_ucred);
511 	td2->td_ucred = crhold(p2->p_ucred);	/* XXXKSE */
512 
513 	pargs_hold(p2->p_args);
514 
515 	if (flags & RFSIGSHARE) {
516 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
517 	} else {
518 		sigacts_copy(newsigacts, p1->p_sigacts);
519 		p2->p_sigacts = newsigacts;
520 	}
521 	if (flags & RFLINUXTHPN)
522 	        p2->p_sigparent = SIGUSR1;
523 	else
524 	        p2->p_sigparent = SIGCHLD;
525 
526 	/* Bump references to the text vnode (for procfs) */
527 	p2->p_textvp = p1->p_textvp;
528 	if (p2->p_textvp)
529 		VREF(p2->p_textvp);
530 	p2->p_fd = fd;
531 	p2->p_fdtol = fdtol;
532 	PROC_UNLOCK(p1);
533 	PROC_UNLOCK(p2);
534 
535 	/*
536 	 * p_limit is copy-on-write, bump refcnt,
537 	 */
538 	p2->p_limit = p1->p_limit;
539 	p2->p_limit->p_refcnt++;
540 
541 	/*
542 	 * Setup linkage for kernel based threading
543 	 */
544 	if((flags & RFTHREAD) != 0) {
545 		mtx_lock(&ppeers_lock);
546 		p2->p_peers = p1->p_peers;
547 		p1->p_peers = p2;
548 		p2->p_leader = p1->p_leader;
549 		mtx_unlock(&ppeers_lock);
550 		PROC_LOCK(p1->p_leader);
551 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
552 			PROC_UNLOCK(p1->p_leader);
553 			/*
554 			 * The task leader is exiting, so process p1 is
555 			 * going to be killed shortly.  Since p1 obviously
556 			 * isn't dead yet, we know that the leader is either
557 			 * sending SIGKILL's to all the processes in this
558 			 * task or is sleeping waiting for all the peers to
559 			 * exit.  We let p1 complete the fork, but we need
560 			 * to go ahead and kill the new process p2 since
561 			 * the task leader may not get a chance to send
562 			 * SIGKILL to it.  We leave it on the list so that
563 			 * the task leader will wait for this new process
564 			 * to commit suicide.
565 			 */
566 			PROC_LOCK(p2);
567 			psignal(p2, SIGKILL);
568 			PROC_UNLOCK(p2);
569 		} else
570 			PROC_UNLOCK(p1->p_leader);
571 	} else {
572 		p2->p_peers = NULL;
573 		p2->p_leader = p2;
574 	}
575 
576 	sx_xlock(&proctree_lock);
577 	PGRP_LOCK(p1->p_pgrp);
578 	PROC_LOCK(p2);
579 	PROC_LOCK(p1);
580 
581 	/*
582 	 * Preserve some more flags in subprocess.  P_PROFIL has already
583 	 * been preserved.
584 	 */
585 	p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
586 	SESS_LOCK(p1->p_session);
587 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
588 		p2->p_flag |= P_CONTROLT;
589 	SESS_UNLOCK(p1->p_session);
590 	if (flags & RFPPWAIT)
591 		p2->p_flag |= P_PPWAIT;
592 
593 	LIST_INSERT_AFTER(p1, p2, p_pglist);
594 	PGRP_UNLOCK(p1->p_pgrp);
595 	LIST_INIT(&p2->p_children);
596 
597 	callout_init(&p2->p_itcallout, 1);
598 
599 #ifdef KTRACE
600 	/*
601 	 * Copy traceflag and tracefile if enabled.
602 	 */
603 	mtx_lock(&ktrace_mtx);
604 	KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
605 	if (p1->p_traceflag & KTRFAC_INHERIT) {
606 		p2->p_traceflag = p1->p_traceflag;
607 		if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
608 			VREF(p2->p_tracevp);
609 			KASSERT(p1->p_tracecred != NULL,
610 			    ("ktrace vnode with no cred"));
611 			p2->p_tracecred = crhold(p1->p_tracecred);
612 		}
613 	}
614 	mtx_unlock(&ktrace_mtx);
615 #endif
616 
617 	/*
618 	 * If PF_FORK is set, the child process inherits the
619 	 * procfs ioctl flags from its parent.
620 	 */
621 	if (p1->p_pfsflags & PF_FORK) {
622 		p2->p_stops = p1->p_stops;
623 		p2->p_pfsflags = p1->p_pfsflags;
624 	}
625 
626 	/*
627 	 * This begins the section where we must prevent the parent
628 	 * from being swapped.
629 	 */
630 	_PHOLD(p1);
631 	PROC_UNLOCK(p1);
632 
633 	/*
634 	 * Attach the new process to its parent.
635 	 *
636 	 * If RFNOWAIT is set, the newly created process becomes a child
637 	 * of init.  This effectively disassociates the child from the
638 	 * parent.
639 	 */
640 	if (flags & RFNOWAIT)
641 		pptr = initproc;
642 	else
643 		pptr = p1;
644 	p2->p_pptr = pptr;
645 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
646 	sx_xunlock(&proctree_lock);
647 
648 	/* Inform accounting that we have forked. */
649 	p2->p_acflag = AFORK;
650 	PROC_UNLOCK(p2);
651 
652 	/*
653 	 * Finish creating the child process.  It will return via a different
654 	 * execution path later.  (ie: directly into user mode)
655 	 */
656 	vm_forkproc(td, p2, td2, flags);
657 
658 	if (flags == (RFFDG | RFPROC)) {
659 		cnt.v_forks++;
660 		cnt.v_forkpages += p2->p_vmspace->vm_dsize +
661 		    p2->p_vmspace->vm_ssize;
662 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
663 		cnt.v_vforks++;
664 		cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
665 		    p2->p_vmspace->vm_ssize;
666 	} else if (p1 == &proc0) {
667 		cnt.v_kthreads++;
668 		cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
669 		    p2->p_vmspace->vm_ssize;
670 	} else {
671 		cnt.v_rforks++;
672 		cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
673 		    p2->p_vmspace->vm_ssize;
674 	}
675 
676 	/*
677 	 * Both processes are set up, now check if any loadable modules want
678 	 * to adjust anything.
679 	 *   What if they have an error? XXX
680 	 */
681 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
682 
683 	/*
684 	 * If RFSTOPPED not requested, make child runnable and add to
685 	 * run queue.
686 	 */
687 	microuptime(&p2->p_stats->p_start);
688 	if ((flags & RFSTOPPED) == 0) {
689 		mtx_lock_spin(&sched_lock);
690 		p2->p_state = PRS_NORMAL;
691 		TD_SET_CAN_RUN(td2);
692 		setrunqueue(td2);
693 		mtx_unlock_spin(&sched_lock);
694 	}
695 
696 	/*
697 	 * Now can be swapped.
698 	 */
699 	PROC_LOCK(p1);
700 	_PRELE(p1);
701 
702 	/*
703 	 * tell any interested parties about the new process
704 	 */
705 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
706 	PROC_UNLOCK(p1);
707 
708 	/*
709 	 * Preserve synchronization semantics of vfork.  If waiting for
710 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
711 	 * proc (in case of exit).
712 	 */
713 	PROC_LOCK(p2);
714 	while (p2->p_flag & P_PPWAIT)
715 		msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
716 	PROC_UNLOCK(p2);
717 
718 	/*
719 	 * If other threads are waiting, let them continue now
720 	 */
721 	if (p1->p_flag & P_SA) {
722 		PROC_LOCK(p1);
723 		thread_single_end();
724 		PROC_UNLOCK(p1);
725 	}
726 
727 	/*
728 	 * Return child proc pointer to parent.
729 	 */
730 	mtx_unlock(&Giant);
731 	*procp = p2;
732 	return (0);
733 fail:
734 	if (ppsratecheck(&lastfail, &curfail, 1))
735 		printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
736 			uid);
737 	sx_xunlock(&allproc_lock);
738 	uma_zfree(proc_zone, newproc);
739 	if (p1->p_flag & P_SA) {
740 		PROC_LOCK(p1);
741 		thread_single_end();
742 		PROC_UNLOCK(p1);
743 	}
744 	tsleep(&forksleep, PUSER, "fork", hz / 2);
745 	mtx_unlock(&Giant);
746 	return (error);
747 }
748 
749 /*
750  * Handle the return of a child process from fork1().  This function
751  * is called from the MD fork_trampoline() entry point.
752  */
753 void
754 fork_exit(callout, arg, frame)
755 	void (*callout)(void *, struct trapframe *);
756 	void *arg;
757 	struct trapframe *frame;
758 {
759 	struct thread *td;
760 	struct proc *p;
761 
762 	if ((td = PCPU_GET(deadthread))) {
763 		PCPU_SET(deadthread, NULL);
764 		thread_stash(td);
765 	}
766 	td = curthread;
767 	p = td->td_proc;
768 	td->td_oncpu = PCPU_GET(cpuid);
769 	p->p_state = PRS_NORMAL;
770 	/*
771 	 * Finish setting up thread glue.  We need to initialize
772 	 * the thread into a td_critnest=1 state.  Some platforms
773 	 * may have already partially or fully initialized td_critnest
774 	 * and/or td_md.md_savecrit (when applciable).
775 	 *
776 	 * see <arch>/<arch>/critical.c
777 	 */
778 	sched_lock.mtx_lock = (uintptr_t)td;
779 	sched_lock.mtx_recurse = 0;
780 	cpu_critical_fork_exit();
781 	CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid,
782 	    p->p_comm);
783 	if (PCPU_GET(switchtime.sec) == 0)
784 		binuptime(PCPU_PTR(switchtime));
785 	PCPU_SET(switchticks, ticks);
786 	mtx_unlock_spin(&sched_lock);
787 
788 	/*
789 	 * cpu_set_fork_handler intercepts this function call to
790          * have this call a non-return function to stay in kernel mode.
791          * initproc has its own fork handler, but it does return.
792          */
793 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
794 	callout(arg, frame);
795 
796 	/*
797 	 * Check if a kernel thread misbehaved and returned from its main
798 	 * function.
799 	 */
800 	PROC_LOCK(p);
801 	if (p->p_flag & P_KTHREAD) {
802 		PROC_UNLOCK(p);
803 		mtx_lock(&Giant);
804 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
805 		    p->p_comm, p->p_pid);
806 		kthread_exit(0);
807 	}
808 	PROC_UNLOCK(p);
809 #ifdef DIAGNOSTIC
810 	cred_free_thread(td);
811 #endif
812 	mtx_assert(&Giant, MA_NOTOWNED);
813 }
814 
815 /*
816  * Simplified back end of syscall(), used when returning from fork()
817  * directly into user mode.  Giant is not held on entry, and must not
818  * be held on return.  This function is passed in to fork_exit() as the
819  * first parameter and is called when returning to a new userland process.
820  */
821 void
822 fork_return(td, frame)
823 	struct thread *td;
824 	struct trapframe *frame;
825 {
826 
827 	userret(td, frame, 0);
828 #ifdef KTRACE
829 	if (KTRPOINT(td, KTR_SYSRET))
830 		ktrsysret(SYS_fork, 0, 0);
831 #endif
832 	mtx_assert(&Giant, MA_NOTOWNED);
833 }
834