xref: /freebsd/sys/kern/kern_fork.c (revision a812392203d7c4c3f0db9d8a0f3391374c49c71f)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_ktrace.h"
41 #include "opt_kstack_pages.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
48 #include <sys/filedesc.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/sysctl.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mutex.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/procdesc.h>
59 #include <sys/pioctl.h>
60 #include <sys/racct.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sched.h>
63 #include <sys/syscall.h>
64 #include <sys/vmmeter.h>
65 #include <sys/vnode.h>
66 #include <sys/acct.h>
67 #include <sys/ktr.h>
68 #include <sys/ktrace.h>
69 #include <sys/unistd.h>
70 #include <sys/sdt.h>
71 #include <sys/sx.h>
72 #include <sys/sysent.h>
73 #include <sys/signalvar.h>
74 
75 #include <security/audit/audit.h>
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_extern.h>
82 #include <vm/uma.h>
83 
84 #ifdef KDTRACE_HOOKS
85 #include <sys/dtrace_bsd.h>
86 dtrace_fork_func_t	dtrace_fasttrap_fork;
87 #endif
88 
89 SDT_PROVIDER_DECLARE(proc);
90 SDT_PROBE_DEFINE3(proc, kernel, , create, "struct proc *",
91     "struct proc *", "int");
92 
93 #ifndef _SYS_SYSPROTO_H_
94 struct fork_args {
95 	int     dummy;
96 };
97 #endif
98 
99 /* ARGSUSED */
100 int
101 sys_fork(struct thread *td, struct fork_args *uap)
102 {
103 	int error;
104 	struct proc *p2;
105 
106 	error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0);
107 	if (error == 0) {
108 		td->td_retval[0] = p2->p_pid;
109 		td->td_retval[1] = 0;
110 	}
111 	return (error);
112 }
113 
114 /* ARGUSED */
115 int
116 sys_pdfork(td, uap)
117 	struct thread *td;
118 	struct pdfork_args *uap;
119 {
120 	int error, fd;
121 	struct proc *p2;
122 
123 	/*
124 	 * It is necessary to return fd by reference because 0 is a valid file
125 	 * descriptor number, and the child needs to be able to distinguish
126 	 * itself from the parent using the return value.
127 	 */
128 	error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2,
129 	    &fd, uap->flags);
130 	if (error == 0) {
131 		td->td_retval[0] = p2->p_pid;
132 		td->td_retval[1] = 0;
133 		error = copyout(&fd, uap->fdp, sizeof(fd));
134 	}
135 	return (error);
136 }
137 
138 /* ARGSUSED */
139 int
140 sys_vfork(struct thread *td, struct vfork_args *uap)
141 {
142 	int error, flags;
143 	struct proc *p2;
144 
145 	flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
146 	error = fork1(td, flags, 0, &p2, NULL, 0);
147 	if (error == 0) {
148 		td->td_retval[0] = p2->p_pid;
149 		td->td_retval[1] = 0;
150 	}
151 	return (error);
152 }
153 
154 int
155 sys_rfork(struct thread *td, struct rfork_args *uap)
156 {
157 	struct proc *p2;
158 	int error;
159 
160 	/* Don't allow kernel-only flags. */
161 	if ((uap->flags & RFKERNELONLY) != 0)
162 		return (EINVAL);
163 
164 	AUDIT_ARG_FFLAGS(uap->flags);
165 	error = fork1(td, uap->flags, 0, &p2, NULL, 0);
166 	if (error == 0) {
167 		td->td_retval[0] = p2 ? p2->p_pid : 0;
168 		td->td_retval[1] = 0;
169 	}
170 	return (error);
171 }
172 
173 int	nprocs = 1;		/* process 0 */
174 int	lastpid = 0;
175 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
176     "Last used PID");
177 
178 /*
179  * Random component to lastpid generation.  We mix in a random factor to make
180  * it a little harder to predict.  We sanity check the modulus value to avoid
181  * doing it in critical paths.  Don't let it be too small or we pointlessly
182  * waste randomness entropy, and don't let it be impossibly large.  Using a
183  * modulus that is too big causes a LOT more process table scans and slows
184  * down fork processing as the pidchecked caching is defeated.
185  */
186 static int randompid = 0;
187 
188 static int
189 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
190 {
191 	int error, pid;
192 
193 	error = sysctl_wire_old_buffer(req, sizeof(int));
194 	if (error != 0)
195 		return(error);
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 static int
216 fork_findpid(int flags)
217 {
218 	struct proc *p;
219 	int trypid;
220 	static int pidchecked = 0;
221 
222 	/*
223 	 * Requires allproc_lock in order to iterate over the list
224 	 * of processes, and proctree_lock to access p_pgrp.
225 	 */
226 	sx_assert(&allproc_lock, SX_LOCKED);
227 	sx_assert(&proctree_lock, SX_LOCKED);
228 
229 	/*
230 	 * Find an unused process ID.  We remember a range of unused IDs
231 	 * ready to use (from lastpid+1 through pidchecked-1).
232 	 *
233 	 * If RFHIGHPID is set (used during system boot), do not allocate
234 	 * low-numbered pids.
235 	 */
236 	trypid = lastpid + 1;
237 	if (flags & RFHIGHPID) {
238 		if (trypid < 10)
239 			trypid = 10;
240 	} else {
241 		if (randompid)
242 			trypid += arc4random() % randompid;
243 	}
244 retry:
245 	/*
246 	 * If the process ID prototype has wrapped around,
247 	 * restart somewhat above 0, as the low-numbered procs
248 	 * tend to include daemons that don't exit.
249 	 */
250 	if (trypid >= pid_max) {
251 		trypid = trypid % pid_max;
252 		if (trypid < 100)
253 			trypid += 100;
254 		pidchecked = 0;
255 	}
256 	if (trypid >= pidchecked) {
257 		int doingzomb = 0;
258 
259 		pidchecked = PID_MAX;
260 		/*
261 		 * Scan the active and zombie procs to check whether this pid
262 		 * is in use.  Remember the lowest pid that's greater
263 		 * than trypid, so we can avoid checking for a while.
264 		 *
265 		 * Avoid reuse of the process group id, session id or
266 		 * the reaper subtree id.  Note that for process group
267 		 * and sessions, the amount of reserved pids is
268 		 * limited by process limit.  For the subtree ids, the
269 		 * id is kept reserved only while there is a
270 		 * non-reaped process in the subtree, so amount of
271 		 * reserved pids is limited by process limit times
272 		 * two.
273 		 */
274 		p = LIST_FIRST(&allproc);
275 again:
276 		for (; p != NULL; p = LIST_NEXT(p, p_list)) {
277 			while (p->p_pid == trypid ||
278 			    p->p_reapsubtree == trypid ||
279 			    (p->p_pgrp != NULL &&
280 			    (p->p_pgrp->pg_id == trypid ||
281 			    (p->p_session != NULL &&
282 			    p->p_session->s_sid == trypid)))) {
283 				trypid++;
284 				if (trypid >= pidchecked)
285 					goto retry;
286 			}
287 			if (p->p_pid > trypid && pidchecked > p->p_pid)
288 				pidchecked = p->p_pid;
289 			if (p->p_pgrp != NULL) {
290 				if (p->p_pgrp->pg_id > trypid &&
291 				    pidchecked > p->p_pgrp->pg_id)
292 					pidchecked = p->p_pgrp->pg_id;
293 				if (p->p_session != NULL &&
294 				    p->p_session->s_sid > trypid &&
295 				    pidchecked > p->p_session->s_sid)
296 					pidchecked = p->p_session->s_sid;
297 			}
298 		}
299 		if (!doingzomb) {
300 			doingzomb = 1;
301 			p = LIST_FIRST(&zombproc);
302 			goto again;
303 		}
304 	}
305 
306 	/*
307 	 * RFHIGHPID does not mess with the lastpid counter during boot.
308 	 */
309 	if (flags & RFHIGHPID)
310 		pidchecked = 0;
311 	else
312 		lastpid = trypid;
313 
314 	return (trypid);
315 }
316 
317 static int
318 fork_norfproc(struct thread *td, int flags)
319 {
320 	int error;
321 	struct proc *p1;
322 
323 	KASSERT((flags & RFPROC) == 0,
324 	    ("fork_norfproc called with RFPROC set"));
325 	p1 = td->td_proc;
326 
327 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
328 	    (flags & (RFCFDG | RFFDG))) {
329 		PROC_LOCK(p1);
330 		if (thread_single(p1, SINGLE_BOUNDARY)) {
331 			PROC_UNLOCK(p1);
332 			return (ERESTART);
333 		}
334 		PROC_UNLOCK(p1);
335 	}
336 
337 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
338 	if (error)
339 		goto fail;
340 
341 	/*
342 	 * Close all file descriptors.
343 	 */
344 	if (flags & RFCFDG) {
345 		struct filedesc *fdtmp;
346 		fdtmp = fdinit(td->td_proc->p_fd, false);
347 		fdescfree(td);
348 		p1->p_fd = fdtmp;
349 	}
350 
351 	/*
352 	 * Unshare file descriptors (from parent).
353 	 */
354 	if (flags & RFFDG)
355 		fdunshare(td);
356 
357 fail:
358 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
359 	    (flags & (RFCFDG | RFFDG))) {
360 		PROC_LOCK(p1);
361 		thread_single_end(p1, SINGLE_BOUNDARY);
362 		PROC_UNLOCK(p1);
363 	}
364 	return (error);
365 }
366 
367 static void
368 do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
369     struct vmspace *vm2, int pdflags)
370 {
371 	struct proc *p1, *pptr;
372 	int p2_held, trypid;
373 	struct filedesc *fd;
374 	struct filedesc_to_leader *fdtol;
375 	struct sigacts *newsigacts;
376 
377 	sx_assert(&proctree_lock, SX_SLOCKED);
378 	sx_assert(&allproc_lock, SX_XLOCKED);
379 
380 	p2_held = 0;
381 	p1 = td->td_proc;
382 
383 	/*
384 	 * Increment the nprocs resource before blocking can occur.  There
385 	 * are hard-limits as to the number of processes that can run.
386 	 */
387 	nprocs++;
388 
389 	trypid = fork_findpid(flags);
390 
391 	sx_sunlock(&proctree_lock);
392 
393 	p2->p_state = PRS_NEW;		/* protect against others */
394 	p2->p_pid = trypid;
395 	AUDIT_ARG_PID(p2->p_pid);
396 	LIST_INSERT_HEAD(&allproc, p2, p_list);
397 	allproc_gen++;
398 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
399 	tidhash_add(td2);
400 	PROC_LOCK(p2);
401 	PROC_LOCK(p1);
402 
403 	sx_xunlock(&allproc_lock);
404 
405 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
406 	    __rangeof(struct proc, p_startcopy, p_endcopy));
407 	pargs_hold(p2->p_args);
408 	PROC_UNLOCK(p1);
409 
410 	bzero(&p2->p_startzero,
411 	    __rangeof(struct proc, p_startzero, p_endzero));
412 
413 	p2->p_ucred = crhold(td->td_ucred);
414 
415 	/* Tell the prison that we exist. */
416 	prison_proc_hold(p2->p_ucred->cr_prison);
417 
418 	PROC_UNLOCK(p2);
419 
420 	/*
421 	 * Malloc things while we don't hold any locks.
422 	 */
423 	if (flags & RFSIGSHARE)
424 		newsigacts = NULL;
425 	else
426 		newsigacts = sigacts_alloc();
427 
428 	/*
429 	 * Copy filedesc.
430 	 */
431 	if (flags & RFCFDG) {
432 		fd = fdinit(p1->p_fd, false);
433 		fdtol = NULL;
434 	} else if (flags & RFFDG) {
435 		fd = fdcopy(p1->p_fd);
436 		fdtol = NULL;
437 	} else {
438 		fd = fdshare(p1->p_fd);
439 		if (p1->p_fdtol == NULL)
440 			p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
441 			    p1->p_leader);
442 		if ((flags & RFTHREAD) != 0) {
443 			/*
444 			 * Shared file descriptor table, and shared
445 			 * process leaders.
446 			 */
447 			fdtol = p1->p_fdtol;
448 			FILEDESC_XLOCK(p1->p_fd);
449 			fdtol->fdl_refcount++;
450 			FILEDESC_XUNLOCK(p1->p_fd);
451 		} else {
452 			/*
453 			 * Shared file descriptor table, and different
454 			 * process leaders.
455 			 */
456 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
457 			    p1->p_fd, p2);
458 		}
459 	}
460 	/*
461 	 * Make a proc table entry for the new process.
462 	 * Start by zeroing the section of proc that is zero-initialized,
463 	 * then copy the section that is copied directly from the parent.
464 	 */
465 
466 	PROC_LOCK(p2);
467 	PROC_LOCK(p1);
468 
469 	bzero(&td2->td_startzero,
470 	    __rangeof(struct thread, td_startzero, td_endzero));
471 
472 	bcopy(&td->td_startcopy, &td2->td_startcopy,
473 	    __rangeof(struct thread, td_startcopy, td_endcopy));
474 
475 	bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
476 	td2->td_sigstk = td->td_sigstk;
477 	td2->td_flags = TDF_INMEM;
478 	td2->td_lend_user_pri = PRI_MAX;
479 
480 #ifdef VIMAGE
481 	td2->td_vnet = NULL;
482 	td2->td_vnet_lpush = NULL;
483 #endif
484 
485 	/*
486 	 * Allow the scheduler to initialize the child.
487 	 */
488 	thread_lock(td);
489 	sched_fork(td, td2);
490 	thread_unlock(td);
491 
492 	/*
493 	 * Duplicate sub-structures as needed.
494 	 * Increase reference counts on shared objects.
495 	 */
496 	p2->p_flag = P_INMEM;
497 	p2->p_flag2 = 0;
498 	p2->p_swtick = ticks;
499 	if (p1->p_flag & P_PROFIL)
500 		startprofclock(p2);
501 	td2->td_ucred = crhold(p2->p_ucred);
502 
503 	if (flags & RFSIGSHARE) {
504 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
505 	} else {
506 		sigacts_copy(newsigacts, p1->p_sigacts);
507 		p2->p_sigacts = newsigacts;
508 	}
509 
510 	if (flags & RFTSIGZMB)
511 	        p2->p_sigparent = RFTSIGNUM(flags);
512 	else if (flags & RFLINUXTHPN)
513 	        p2->p_sigparent = SIGUSR1;
514 	else
515 	        p2->p_sigparent = SIGCHLD;
516 
517 	p2->p_textvp = p1->p_textvp;
518 	p2->p_fd = fd;
519 	p2->p_fdtol = fdtol;
520 
521 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
522 		p2->p_flag |= P_PROTECTED;
523 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
524 	}
525 
526 	/*
527 	 * p_limit is copy-on-write.  Bump its refcount.
528 	 */
529 	lim_fork(p1, p2);
530 
531 	pstats_fork(p1->p_stats, p2->p_stats);
532 
533 	PROC_UNLOCK(p1);
534 	PROC_UNLOCK(p2);
535 
536 	/* Bump references to the text vnode (for procfs). */
537 	if (p2->p_textvp)
538 		vref(p2->p_textvp);
539 
540 	/*
541 	 * Set up linkage for kernel based threading.
542 	 */
543 	if ((flags & RFTHREAD) != 0) {
544 		mtx_lock(&ppeers_lock);
545 		p2->p_peers = p1->p_peers;
546 		p1->p_peers = p2;
547 		p2->p_leader = p1->p_leader;
548 		mtx_unlock(&ppeers_lock);
549 		PROC_LOCK(p1->p_leader);
550 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
551 			PROC_UNLOCK(p1->p_leader);
552 			/*
553 			 * The task leader is exiting, so process p1 is
554 			 * going to be killed shortly.  Since p1 obviously
555 			 * isn't dead yet, we know that the leader is either
556 			 * sending SIGKILL's to all the processes in this
557 			 * task or is sleeping waiting for all the peers to
558 			 * exit.  We let p1 complete the fork, but we need
559 			 * to go ahead and kill the new process p2 since
560 			 * the task leader may not get a chance to send
561 			 * SIGKILL to it.  We leave it on the list so that
562 			 * the task leader will wait for this new process
563 			 * to commit suicide.
564 			 */
565 			PROC_LOCK(p2);
566 			kern_psignal(p2, SIGKILL);
567 			PROC_UNLOCK(p2);
568 		} else
569 			PROC_UNLOCK(p1->p_leader);
570 	} else {
571 		p2->p_peers = NULL;
572 		p2->p_leader = p2;
573 	}
574 
575 	sx_xlock(&proctree_lock);
576 	PGRP_LOCK(p1->p_pgrp);
577 	PROC_LOCK(p2);
578 	PROC_LOCK(p1);
579 
580 	/*
581 	 * Preserve some more flags in subprocess.  P_PROFIL has already
582 	 * been preserved.
583 	 */
584 	p2->p_flag |= p1->p_flag & P_SUGID;
585 	td2->td_pflags |= td->td_pflags & TDP_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 	p2->p_pgrp = p1->p_pgrp;
594 	LIST_INSERT_AFTER(p1, p2, p_pglist);
595 	PGRP_UNLOCK(p1->p_pgrp);
596 	LIST_INIT(&p2->p_children);
597 	LIST_INIT(&p2->p_orphans);
598 
599 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
600 
601 	/*
602 	 * If PF_FORK is set, the child process inherits the
603 	 * procfs ioctl flags from its parent.
604 	 */
605 	if (p1->p_pfsflags & PF_FORK) {
606 		p2->p_stops = p1->p_stops;
607 		p2->p_pfsflags = p1->p_pfsflags;
608 	}
609 
610 	/*
611 	 * This begins the section where we must prevent the parent
612 	 * from being swapped.
613 	 */
614 	_PHOLD(p1);
615 	PROC_UNLOCK(p1);
616 
617 	/*
618 	 * Attach the new process to its parent.
619 	 *
620 	 * If RFNOWAIT is set, the newly created process becomes a child
621 	 * of init.  This effectively disassociates the child from the
622 	 * parent.
623 	 */
624 	if ((flags & RFNOWAIT) != 0) {
625 		pptr = p1->p_reaper;
626 		p2->p_reaper = pptr;
627 	} else {
628 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
629 		    p1 : p1->p_reaper;
630 		pptr = p1;
631 	}
632 	p2->p_pptr = pptr;
633 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
634 	LIST_INIT(&p2->p_reaplist);
635 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
636 	if (p2->p_reaper == p1)
637 		p2->p_reapsubtree = p2->p_pid;
638 	sx_xunlock(&proctree_lock);
639 
640 	/* Inform accounting that we have forked. */
641 	p2->p_acflag = AFORK;
642 	PROC_UNLOCK(p2);
643 
644 #ifdef KTRACE
645 	ktrprocfork(p1, p2);
646 #endif
647 
648 	/*
649 	 * Finish creating the child process.  It will return via a different
650 	 * execution path later.  (ie: directly into user mode)
651 	 */
652 	vm_forkproc(td, p2, td2, vm2, flags);
653 
654 	if (flags == (RFFDG | RFPROC)) {
655 		PCPU_INC(cnt.v_forks);
656 		PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
657 		    p2->p_vmspace->vm_ssize);
658 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
659 		PCPU_INC(cnt.v_vforks);
660 		PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
661 		    p2->p_vmspace->vm_ssize);
662 	} else if (p1 == &proc0) {
663 		PCPU_INC(cnt.v_kthreads);
664 		PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
665 		    p2->p_vmspace->vm_ssize);
666 	} else {
667 		PCPU_INC(cnt.v_rforks);
668 		PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
669 		    p2->p_vmspace->vm_ssize);
670 	}
671 
672 	/*
673 	 * Associate the process descriptor with the process before anything
674 	 * can happen that might cause that process to need the descriptor.
675 	 * However, don't do this until after fork(2) can no longer fail.
676 	 */
677 	if (flags & RFPROCDESC)
678 		procdesc_new(p2, pdflags);
679 
680 	/*
681 	 * Both processes are set up, now check if any loadable modules want
682 	 * to adjust anything.
683 	 */
684 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
685 
686 	/*
687 	 * Set the child start time and mark the process as being complete.
688 	 */
689 	PROC_LOCK(p2);
690 	PROC_LOCK(p1);
691 	microuptime(&p2->p_stats->p_start);
692 	PROC_SLOCK(p2);
693 	p2->p_state = PRS_NORMAL;
694 	PROC_SUNLOCK(p2);
695 
696 #ifdef KDTRACE_HOOKS
697 	/*
698 	 * Tell the DTrace fasttrap provider about the new process so that any
699 	 * tracepoints inherited from the parent can be removed. We have to do
700 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
701 	 * use pfind() later on.
702 	 */
703 	if ((flags & RFMEM) == 0 && dtrace_fasttrap_fork)
704 		dtrace_fasttrap_fork(p1, p2);
705 #endif
706 	if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED |
707 	    P_FOLLOWFORK)) {
708 		/*
709 		 * Arrange for debugger to receive the fork event.
710 		 *
711 		 * We can report PL_FLAG_FORKED regardless of
712 		 * P_FOLLOWFORK settings, but it does not make a sense
713 		 * for runaway child.
714 		 */
715 		td->td_dbgflags |= TDB_FORK;
716 		td->td_dbg_forked = p2->p_pid;
717 		td2->td_dbgflags |= TDB_STOPATFORK;
718 		_PHOLD(p2);
719 		p2_held = 1;
720 	}
721 	if (flags & RFPPWAIT) {
722 		td->td_pflags |= TDP_RFPPWAIT;
723 		td->td_rfppwait_p = p2;
724 	}
725 	PROC_UNLOCK(p2);
726 	if ((flags & RFSTOPPED) == 0) {
727 		/*
728 		 * If RFSTOPPED not requested, make child runnable and
729 		 * add to run queue.
730 		 */
731 		thread_lock(td2);
732 		TD_SET_CAN_RUN(td2);
733 		sched_add(td2, SRQ_BORING);
734 		thread_unlock(td2);
735 	}
736 
737 	/*
738 	 * Now can be swapped.
739 	 */
740 	_PRELE(p1);
741 	PROC_UNLOCK(p1);
742 
743 	/*
744 	 * Tell any interested parties about the new process.
745 	 */
746 	knote_fork(&p1->p_klist, p2->p_pid);
747 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
748 
749 	/*
750 	 * Wait until debugger is attached to child.
751 	 */
752 	PROC_LOCK(p2);
753 	while ((td2->td_dbgflags & TDB_STOPATFORK) != 0)
754 		cv_wait(&p2->p_dbgwait, &p2->p_mtx);
755 	if (p2_held)
756 		_PRELE(p2);
757 	PROC_UNLOCK(p2);
758 }
759 
760 int
761 fork1(struct thread *td, int flags, int pages, struct proc **procp,
762     int *procdescp, int pdflags)
763 {
764 	struct proc *p1;
765 	struct proc *newproc;
766 	int ok;
767 	struct thread *td2;
768 	struct vmspace *vm2;
769 	vm_ooffset_t mem_charged;
770 	int error;
771 	static int curfail;
772 	static struct timeval lastfail;
773 	struct file *fp_procdesc = NULL;
774 
775 	/* Check for the undefined or unimplemented flags. */
776 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
777 		return (EINVAL);
778 
779 	/* Signal value requires RFTSIGZMB. */
780 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
781 		return (EINVAL);
782 
783 	/* Can't copy and clear. */
784 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
785 		return (EINVAL);
786 
787 	/* Check the validity of the signal number. */
788 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
789 		return (EINVAL);
790 
791 	if ((flags & RFPROCDESC) != 0) {
792 		/* Can't not create a process yet get a process descriptor. */
793 		if ((flags & RFPROC) == 0)
794 			return (EINVAL);
795 
796 		/* Must provide a place to put a procdesc if creating one. */
797 		if (procdescp == NULL)
798 			return (EINVAL);
799 	}
800 
801 	p1 = td->td_proc;
802 
803 	/*
804 	 * Here we don't create a new process, but we divorce
805 	 * certain parts of a process from itself.
806 	 */
807 	if ((flags & RFPROC) == 0) {
808 		*procp = NULL;
809 		return (fork_norfproc(td, flags));
810 	}
811 
812 	/*
813 	 * If required, create a process descriptor in the parent first; we
814 	 * will abandon it if something goes wrong. We don't finit() until
815 	 * later.
816 	 */
817 	if (flags & RFPROCDESC) {
818 		error = falloc(td, &fp_procdesc, procdescp, 0);
819 		if (error != 0)
820 			return (error);
821 	}
822 
823 	mem_charged = 0;
824 	vm2 = NULL;
825 	if (pages == 0)
826 		pages = KSTACK_PAGES;
827 	/* Allocate new proc. */
828 	newproc = uma_zalloc(proc_zone, M_WAITOK);
829 	td2 = FIRST_THREAD_IN_PROC(newproc);
830 	if (td2 == NULL) {
831 		td2 = thread_alloc(pages);
832 		if (td2 == NULL) {
833 			error = ENOMEM;
834 			goto fail1;
835 		}
836 		proc_linkup(newproc, td2);
837 	} else {
838 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
839 			if (td2->td_kstack != 0)
840 				vm_thread_dispose(td2);
841 			if (!thread_alloc_stack(td2, pages)) {
842 				error = ENOMEM;
843 				goto fail1;
844 			}
845 		}
846 	}
847 
848 	if ((flags & RFMEM) == 0) {
849 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
850 		if (vm2 == NULL) {
851 			error = ENOMEM;
852 			goto fail1;
853 		}
854 		if (!swap_reserve(mem_charged)) {
855 			/*
856 			 * The swap reservation failed. The accounting
857 			 * from the entries of the copied vm2 will be
858 			 * substracted in vmspace_free(), so force the
859 			 * reservation there.
860 			 */
861 			swap_reserve_force(mem_charged);
862 			error = ENOMEM;
863 			goto fail1;
864 		}
865 	} else
866 		vm2 = NULL;
867 
868 	/*
869 	 * XXX: This is ugly; when we copy resource usage, we need to bump
870 	 *      per-cred resource counters.
871 	 */
872 	newproc->p_ucred = p1->p_ucred;
873 
874 	/*
875 	 * Initialize resource accounting for the child process.
876 	 */
877 	error = racct_proc_fork(p1, newproc);
878 	if (error != 0) {
879 		error = EAGAIN;
880 		goto fail1;
881 	}
882 
883 #ifdef MAC
884 	mac_proc_init(newproc);
885 #endif
886 	knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
887 	STAILQ_INIT(&newproc->p_ktr);
888 
889 	/* We have to lock the process tree while we look for a pid. */
890 	sx_slock(&proctree_lock);
891 
892 	/*
893 	 * Although process entries are dynamically created, we still keep
894 	 * a global limit on the maximum number we will create.  Don't allow
895 	 * a nonprivileged user to use the last ten processes; don't let root
896 	 * exceed the limit. The variable nprocs is the current number of
897 	 * processes, maxproc is the limit.
898 	 */
899 	sx_xlock(&allproc_lock);
900 	if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
901 	    PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
902 		error = EAGAIN;
903 		goto fail;
904 	}
905 
906 	/*
907 	 * Increment the count of procs running with this uid. Don't allow
908 	 * a nonprivileged user to exceed their current limit.
909 	 *
910 	 * XXXRW: Can we avoid privilege here if it's not needed?
911 	 */
912 	error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
913 	if (error == 0)
914 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
915 	else {
916 		PROC_LOCK(p1);
917 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
918 		    lim_cur(p1, RLIMIT_NPROC));
919 		PROC_UNLOCK(p1);
920 	}
921 	if (ok) {
922 		do_fork(td, flags, newproc, td2, vm2, pdflags);
923 
924 		/*
925 		 * Return child proc pointer to parent.
926 		 */
927 		*procp = newproc;
928 		if (flags & RFPROCDESC) {
929 			procdesc_finit(newproc->p_procdesc, fp_procdesc);
930 			fdrop(fp_procdesc, td);
931 		}
932 		racct_proc_fork_done(newproc);
933 		return (0);
934 	}
935 
936 	error = EAGAIN;
937 fail:
938 	sx_sunlock(&proctree_lock);
939 	if (ppsratecheck(&lastfail, &curfail, 1))
940 		printf("maxproc limit exceeded by uid %u (pid %d); see tuning(7) and login.conf(5)\n",
941 		    td->td_ucred->cr_ruid, p1->p_pid);
942 	sx_xunlock(&allproc_lock);
943 #ifdef MAC
944 	mac_proc_destroy(newproc);
945 #endif
946 	racct_proc_exit(newproc);
947 fail1:
948 	if (vm2 != NULL)
949 		vmspace_free(vm2);
950 	uma_zfree(proc_zone, newproc);
951 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
952 		fdclose(td->td_proc->p_fd, fp_procdesc, *procdescp, td);
953 		fdrop(fp_procdesc, td);
954 	}
955 	pause("fork", hz / 2);
956 	return (error);
957 }
958 
959 /*
960  * Handle the return of a child process from fork1().  This function
961  * is called from the MD fork_trampoline() entry point.
962  */
963 void
964 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
965     struct trapframe *frame)
966 {
967 	struct proc *p;
968 	struct thread *td;
969 	struct thread *dtd;
970 
971 	td = curthread;
972 	p = td->td_proc;
973 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
974 
975 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
976 		td, td->td_sched, p->p_pid, td->td_name);
977 
978 	sched_fork_exit(td);
979 	/*
980 	* Processes normally resume in mi_switch() after being
981 	* cpu_switch()'ed to, but when children start up they arrive here
982 	* instead, so we must do much the same things as mi_switch() would.
983 	*/
984 	if ((dtd = PCPU_GET(deadthread))) {
985 		PCPU_SET(deadthread, NULL);
986 		thread_stash(dtd);
987 	}
988 	thread_unlock(td);
989 
990 	/*
991 	 * cpu_set_fork_handler intercepts this function call to
992 	 * have this call a non-return function to stay in kernel mode.
993 	 * initproc has its own fork handler, but it does return.
994 	 */
995 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
996 	callout(arg, frame);
997 
998 	/*
999 	 * Check if a kernel thread misbehaved and returned from its main
1000 	 * function.
1001 	 */
1002 	if (p->p_flag & P_KTHREAD) {
1003 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1004 		    td->td_name, p->p_pid);
1005 		kproc_exit(0);
1006 	}
1007 	mtx_assert(&Giant, MA_NOTOWNED);
1008 
1009 	if (p->p_sysent->sv_schedtail != NULL)
1010 		(p->p_sysent->sv_schedtail)(td);
1011 }
1012 
1013 /*
1014  * Simplified back end of syscall(), used when returning from fork()
1015  * directly into user mode.  Giant is not held on entry, and must not
1016  * be held on return.  This function is passed in to fork_exit() as the
1017  * first parameter and is called when returning to a new userland process.
1018  */
1019 void
1020 fork_return(struct thread *td, struct trapframe *frame)
1021 {
1022 	struct proc *p, *dbg;
1023 
1024 	if (td->td_dbgflags & TDB_STOPATFORK) {
1025 		p = td->td_proc;
1026 		sx_xlock(&proctree_lock);
1027 		PROC_LOCK(p);
1028 		if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) ==
1029 		    (P_TRACED | P_FOLLOWFORK)) {
1030 			/*
1031 			 * If debugger still wants auto-attach for the
1032 			 * parent's children, do it now.
1033 			 */
1034 			dbg = p->p_pptr->p_pptr;
1035 			p->p_flag |= P_TRACED;
1036 			p->p_oppid = p->p_pptr->p_pid;
1037 			proc_reparent(p, dbg);
1038 			sx_xunlock(&proctree_lock);
1039 			td->td_dbgflags |= TDB_CHILD;
1040 			ptracestop(td, SIGSTOP);
1041 			td->td_dbgflags &= ~TDB_CHILD;
1042 		} else {
1043 			/*
1044 			 * ... otherwise clear the request.
1045 			 */
1046 			sx_xunlock(&proctree_lock);
1047 			td->td_dbgflags &= ~TDB_STOPATFORK;
1048 			cv_broadcast(&p->p_dbgwait);
1049 		}
1050 		PROC_UNLOCK(p);
1051 	}
1052 
1053 	userret(td, frame);
1054 
1055 #ifdef KTRACE
1056 	if (KTRPOINT(td, KTR_SYSRET))
1057 		ktrsysret(SYS_fork, 0, 0);
1058 #endif
1059 }
1060