xref: /freebsd/sys/kern/kern_fork.c (revision 119b75925c562202145d7bac7b676b98029c6cb9)
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 #include <vm/vm_domain.h>
84 
85 #ifdef KDTRACE_HOOKS
86 #include <sys/dtrace_bsd.h>
87 dtrace_fork_func_t	dtrace_fasttrap_fork;
88 #endif
89 
90 SDT_PROVIDER_DECLARE(proc);
91 SDT_PROBE_DEFINE3(proc, kernel, , create, "struct proc *",
92     "struct proc *", "int");
93 
94 #ifndef _SYS_SYSPROTO_H_
95 struct fork_args {
96 	int     dummy;
97 };
98 #endif
99 
100 /* ARGSUSED */
101 int
102 sys_fork(struct thread *td, struct fork_args *uap)
103 {
104 	int error;
105 	struct proc *p2;
106 
107 	error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0);
108 	if (error == 0) {
109 		td->td_retval[0] = p2->p_pid;
110 		td->td_retval[1] = 0;
111 	}
112 	return (error);
113 }
114 
115 /* ARGUSED */
116 int
117 sys_pdfork(td, uap)
118 	struct thread *td;
119 	struct pdfork_args *uap;
120 {
121 	int error, fd;
122 	struct proc *p2;
123 
124 	/*
125 	 * It is necessary to return fd by reference because 0 is a valid file
126 	 * descriptor number, and the child needs to be able to distinguish
127 	 * itself from the parent using the return value.
128 	 */
129 	error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2,
130 	    &fd, uap->flags);
131 	if (error == 0) {
132 		td->td_retval[0] = p2->p_pid;
133 		td->td_retval[1] = 0;
134 		error = copyout(&fd, uap->fdp, sizeof(fd));
135 	}
136 	return (error);
137 }
138 
139 /* ARGSUSED */
140 int
141 sys_vfork(struct thread *td, struct vfork_args *uap)
142 {
143 	int error, flags;
144 	struct proc *p2;
145 
146 	flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
147 	error = fork1(td, flags, 0, &p2, NULL, 0);
148 	if (error == 0) {
149 		td->td_retval[0] = p2->p_pid;
150 		td->td_retval[1] = 0;
151 	}
152 	return (error);
153 }
154 
155 int
156 sys_rfork(struct thread *td, struct rfork_args *uap)
157 {
158 	struct proc *p2;
159 	int error;
160 
161 	/* Don't allow kernel-only flags. */
162 	if ((uap->flags & RFKERNELONLY) != 0)
163 		return (EINVAL);
164 
165 	AUDIT_ARG_FFLAGS(uap->flags);
166 	error = fork1(td, uap->flags, 0, &p2, NULL, 0);
167 	if (error == 0) {
168 		td->td_retval[0] = p2 ? p2->p_pid : 0;
169 		td->td_retval[1] = 0;
170 	}
171 	return (error);
172 }
173 
174 int	nprocs = 1;		/* process 0 */
175 int	lastpid = 0;
176 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
177     "Last used PID");
178 
179 /*
180  * Random component to lastpid generation.  We mix in a random factor to make
181  * it a little harder to predict.  We sanity check the modulus value to avoid
182  * doing it in critical paths.  Don't let it be too small or we pointlessly
183  * waste randomness entropy, and don't let it be impossibly large.  Using a
184  * modulus that is too big causes a LOT more process table scans and slows
185  * down fork processing as the pidchecked caching is defeated.
186  */
187 static int randompid = 0;
188 
189 static int
190 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
191 {
192 	int error, pid;
193 
194 	error = sysctl_wire_old_buffer(req, sizeof(int));
195 	if (error != 0)
196 		return(error);
197 	sx_xlock(&allproc_lock);
198 	pid = randompid;
199 	error = sysctl_handle_int(oidp, &pid, 0, req);
200 	if (error == 0 && req->newptr != NULL) {
201 		if (pid < 0 || pid > pid_max - 100)	/* out of range */
202 			pid = pid_max - 100;
203 		else if (pid < 2)			/* NOP */
204 			pid = 0;
205 		else if (pid < 100)			/* Make it reasonable */
206 			pid = 100;
207 		randompid = pid;
208 	}
209 	sx_xunlock(&allproc_lock);
210 	return (error);
211 }
212 
213 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
214     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
215 
216 static int
217 fork_findpid(int flags)
218 {
219 	struct proc *p;
220 	int trypid;
221 	static int pidchecked = 0;
222 
223 	/*
224 	 * Requires allproc_lock in order to iterate over the list
225 	 * of processes, and proctree_lock to access p_pgrp.
226 	 */
227 	sx_assert(&allproc_lock, SX_LOCKED);
228 	sx_assert(&proctree_lock, SX_LOCKED);
229 
230 	/*
231 	 * Find an unused process ID.  We remember a range of unused IDs
232 	 * ready to use (from lastpid+1 through pidchecked-1).
233 	 *
234 	 * If RFHIGHPID is set (used during system boot), do not allocate
235 	 * low-numbered pids.
236 	 */
237 	trypid = lastpid + 1;
238 	if (flags & RFHIGHPID) {
239 		if (trypid < 10)
240 			trypid = 10;
241 	} else {
242 		if (randompid)
243 			trypid += arc4random() % randompid;
244 	}
245 retry:
246 	/*
247 	 * If the process ID prototype has wrapped around,
248 	 * restart somewhat above 0, as the low-numbered procs
249 	 * tend to include daemons that don't exit.
250 	 */
251 	if (trypid >= pid_max) {
252 		trypid = trypid % pid_max;
253 		if (trypid < 100)
254 			trypid += 100;
255 		pidchecked = 0;
256 	}
257 	if (trypid >= pidchecked) {
258 		int doingzomb = 0;
259 
260 		pidchecked = PID_MAX;
261 		/*
262 		 * Scan the active and zombie procs to check whether this pid
263 		 * is in use.  Remember the lowest pid that's greater
264 		 * than trypid, so we can avoid checking for a while.
265 		 *
266 		 * Avoid reuse of the process group id, session id or
267 		 * the reaper subtree id.  Note that for process group
268 		 * and sessions, the amount of reserved pids is
269 		 * limited by process limit.  For the subtree ids, the
270 		 * id is kept reserved only while there is a
271 		 * non-reaped process in the subtree, so amount of
272 		 * reserved pids is limited by process limit times
273 		 * two.
274 		 */
275 		p = LIST_FIRST(&allproc);
276 again:
277 		for (; p != NULL; p = LIST_NEXT(p, p_list)) {
278 			while (p->p_pid == trypid ||
279 			    p->p_reapsubtree == trypid ||
280 			    (p->p_pgrp != NULL &&
281 			    (p->p_pgrp->pg_id == trypid ||
282 			    (p->p_session != NULL &&
283 			    p->p_session->s_sid == trypid)))) {
284 				trypid++;
285 				if (trypid >= pidchecked)
286 					goto retry;
287 			}
288 			if (p->p_pid > trypid && pidchecked > p->p_pid)
289 				pidchecked = p->p_pid;
290 			if (p->p_pgrp != NULL) {
291 				if (p->p_pgrp->pg_id > trypid &&
292 				    pidchecked > p->p_pgrp->pg_id)
293 					pidchecked = p->p_pgrp->pg_id;
294 				if (p->p_session != NULL &&
295 				    p->p_session->s_sid > trypid &&
296 				    pidchecked > p->p_session->s_sid)
297 					pidchecked = p->p_session->s_sid;
298 			}
299 		}
300 		if (!doingzomb) {
301 			doingzomb = 1;
302 			p = LIST_FIRST(&zombproc);
303 			goto again;
304 		}
305 	}
306 
307 	/*
308 	 * RFHIGHPID does not mess with the lastpid counter during boot.
309 	 */
310 	if (flags & RFHIGHPID)
311 		pidchecked = 0;
312 	else
313 		lastpid = trypid;
314 
315 	return (trypid);
316 }
317 
318 static int
319 fork_norfproc(struct thread *td, int flags)
320 {
321 	int error;
322 	struct proc *p1;
323 
324 	KASSERT((flags & RFPROC) == 0,
325 	    ("fork_norfproc called with RFPROC set"));
326 	p1 = td->td_proc;
327 
328 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
329 	    (flags & (RFCFDG | RFFDG))) {
330 		PROC_LOCK(p1);
331 		if (thread_single(p1, SINGLE_BOUNDARY)) {
332 			PROC_UNLOCK(p1);
333 			return (ERESTART);
334 		}
335 		PROC_UNLOCK(p1);
336 	}
337 
338 	error = vm_forkproc(td, NULL, NULL, NULL, flags);
339 	if (error)
340 		goto fail;
341 
342 	/*
343 	 * Close all file descriptors.
344 	 */
345 	if (flags & RFCFDG) {
346 		struct filedesc *fdtmp;
347 		fdtmp = fdinit(td->td_proc->p_fd, false);
348 		fdescfree(td);
349 		p1->p_fd = fdtmp;
350 	}
351 
352 	/*
353 	 * Unshare file descriptors (from parent).
354 	 */
355 	if (flags & RFFDG)
356 		fdunshare(td);
357 
358 fail:
359 	if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
360 	    (flags & (RFCFDG | RFFDG))) {
361 		PROC_LOCK(p1);
362 		thread_single_end(p1, SINGLE_BOUNDARY);
363 		PROC_UNLOCK(p1);
364 	}
365 	return (error);
366 }
367 
368 static void
369 do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
370     struct vmspace *vm2, int pdflags)
371 {
372 	struct proc *p1, *pptr;
373 	int p2_held, trypid;
374 	struct filedesc *fd;
375 	struct filedesc_to_leader *fdtol;
376 	struct sigacts *newsigacts;
377 
378 	sx_assert(&proctree_lock, SX_SLOCKED);
379 	sx_assert(&allproc_lock, SX_XLOCKED);
380 
381 	p2_held = 0;
382 	p1 = td->td_proc;
383 
384 	/*
385 	 * Increment the nprocs resource before blocking can occur.  There
386 	 * are hard-limits as to the number of processes that can run.
387 	 */
388 	nprocs++;
389 
390 	trypid = fork_findpid(flags);
391 
392 	sx_sunlock(&proctree_lock);
393 
394 	p2->p_state = PRS_NEW;		/* protect against others */
395 	p2->p_pid = trypid;
396 	AUDIT_ARG_PID(p2->p_pid);
397 	LIST_INSERT_HEAD(&allproc, p2, p_list);
398 	allproc_gen++;
399 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
400 	tidhash_add(td2);
401 	PROC_LOCK(p2);
402 	PROC_LOCK(p1);
403 
404 	sx_xunlock(&allproc_lock);
405 
406 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
407 	    __rangeof(struct proc, p_startcopy, p_endcopy));
408 	pargs_hold(p2->p_args);
409 
410 	PROC_UNLOCK(p1);
411 
412 	bzero(&p2->p_startzero,
413 	    __rangeof(struct proc, p_startzero, p_endzero));
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 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC);
498 	p2->p_swtick = ticks;
499 	if (p1->p_flag & P_PROFIL)
500 		startprofclock(p2);
501 
502 	/*
503 	 * Whilst the proc lock is held, copy the VM domain data out
504 	 * using the VM domain method.
505 	 */
506 	vm_domain_policy_init(&p2->p_vm_dom_policy);
507 	vm_domain_policy_localcopy(&p2->p_vm_dom_policy,
508 	    &p1->p_vm_dom_policy);
509 
510 	if (flags & RFSIGSHARE) {
511 		p2->p_sigacts = sigacts_hold(p1->p_sigacts);
512 	} else {
513 		sigacts_copy(newsigacts, p1->p_sigacts);
514 		p2->p_sigacts = newsigacts;
515 	}
516 
517 	if (flags & RFTSIGZMB)
518 	        p2->p_sigparent = RFTSIGNUM(flags);
519 	else if (flags & RFLINUXTHPN)
520 	        p2->p_sigparent = SIGUSR1;
521 	else
522 	        p2->p_sigparent = SIGCHLD;
523 
524 	p2->p_textvp = p1->p_textvp;
525 	p2->p_fd = fd;
526 	p2->p_fdtol = fdtol;
527 
528 	if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
529 		p2->p_flag |= P_PROTECTED;
530 		p2->p_flag2 |= P2_INHERIT_PROTECTED;
531 	}
532 
533 	/*
534 	 * p_limit is copy-on-write.  Bump its refcount.
535 	 */
536 	lim_fork(p1, p2);
537 
538 	thread_cow_get_proc(td2, p2);
539 
540 	pstats_fork(p1->p_stats, p2->p_stats);
541 
542 	PROC_UNLOCK(p1);
543 	PROC_UNLOCK(p2);
544 
545 	/* Bump references to the text vnode (for procfs). */
546 	if (p2->p_textvp)
547 		vref(p2->p_textvp);
548 
549 	/*
550 	 * Set up linkage for kernel based threading.
551 	 */
552 	if ((flags & RFTHREAD) != 0) {
553 		mtx_lock(&ppeers_lock);
554 		p2->p_peers = p1->p_peers;
555 		p1->p_peers = p2;
556 		p2->p_leader = p1->p_leader;
557 		mtx_unlock(&ppeers_lock);
558 		PROC_LOCK(p1->p_leader);
559 		if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
560 			PROC_UNLOCK(p1->p_leader);
561 			/*
562 			 * The task leader is exiting, so process p1 is
563 			 * going to be killed shortly.  Since p1 obviously
564 			 * isn't dead yet, we know that the leader is either
565 			 * sending SIGKILL's to all the processes in this
566 			 * task or is sleeping waiting for all the peers to
567 			 * exit.  We let p1 complete the fork, but we need
568 			 * to go ahead and kill the new process p2 since
569 			 * the task leader may not get a chance to send
570 			 * SIGKILL to it.  We leave it on the list so that
571 			 * the task leader will wait for this new process
572 			 * to commit suicide.
573 			 */
574 			PROC_LOCK(p2);
575 			kern_psignal(p2, SIGKILL);
576 			PROC_UNLOCK(p2);
577 		} else
578 			PROC_UNLOCK(p1->p_leader);
579 	} else {
580 		p2->p_peers = NULL;
581 		p2->p_leader = p2;
582 	}
583 
584 	sx_xlock(&proctree_lock);
585 	PGRP_LOCK(p1->p_pgrp);
586 	PROC_LOCK(p2);
587 	PROC_LOCK(p1);
588 
589 	/*
590 	 * Preserve some more flags in subprocess.  P_PROFIL has already
591 	 * been preserved.
592 	 */
593 	p2->p_flag |= p1->p_flag & P_SUGID;
594 	td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
595 	SESS_LOCK(p1->p_session);
596 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
597 		p2->p_flag |= P_CONTROLT;
598 	SESS_UNLOCK(p1->p_session);
599 	if (flags & RFPPWAIT)
600 		p2->p_flag |= P_PPWAIT;
601 
602 	p2->p_pgrp = p1->p_pgrp;
603 	LIST_INSERT_AFTER(p1, p2, p_pglist);
604 	PGRP_UNLOCK(p1->p_pgrp);
605 	LIST_INIT(&p2->p_children);
606 	LIST_INIT(&p2->p_orphans);
607 
608 	callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
609 
610 	/*
611 	 * If PF_FORK is set, the child process inherits the
612 	 * procfs ioctl flags from its parent.
613 	 */
614 	if (p1->p_pfsflags & PF_FORK) {
615 		p2->p_stops = p1->p_stops;
616 		p2->p_pfsflags = p1->p_pfsflags;
617 	}
618 
619 	/*
620 	 * This begins the section where we must prevent the parent
621 	 * from being swapped.
622 	 */
623 	_PHOLD(p1);
624 	PROC_UNLOCK(p1);
625 
626 	/*
627 	 * Attach the new process to its parent.
628 	 *
629 	 * If RFNOWAIT is set, the newly created process becomes a child
630 	 * of init.  This effectively disassociates the child from the
631 	 * parent.
632 	 */
633 	if ((flags & RFNOWAIT) != 0) {
634 		pptr = p1->p_reaper;
635 		p2->p_reaper = pptr;
636 	} else {
637 		p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
638 		    p1 : p1->p_reaper;
639 		pptr = p1;
640 	}
641 	p2->p_pptr = pptr;
642 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
643 	LIST_INIT(&p2->p_reaplist);
644 	LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
645 	if (p2->p_reaper == p1)
646 		p2->p_reapsubtree = p2->p_pid;
647 	sx_xunlock(&proctree_lock);
648 
649 	/* Inform accounting that we have forked. */
650 	p2->p_acflag = AFORK;
651 	PROC_UNLOCK(p2);
652 
653 #ifdef KTRACE
654 	ktrprocfork(p1, p2);
655 #endif
656 
657 	/*
658 	 * Finish creating the child process.  It will return via a different
659 	 * execution path later.  (ie: directly into user mode)
660 	 */
661 	vm_forkproc(td, p2, td2, vm2, flags);
662 
663 	if (flags == (RFFDG | RFPROC)) {
664 		PCPU_INC(cnt.v_forks);
665 		PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
666 		    p2->p_vmspace->vm_ssize);
667 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
668 		PCPU_INC(cnt.v_vforks);
669 		PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
670 		    p2->p_vmspace->vm_ssize);
671 	} else if (p1 == &proc0) {
672 		PCPU_INC(cnt.v_kthreads);
673 		PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
674 		    p2->p_vmspace->vm_ssize);
675 	} else {
676 		PCPU_INC(cnt.v_rforks);
677 		PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
678 		    p2->p_vmspace->vm_ssize);
679 	}
680 
681 	/*
682 	 * Associate the process descriptor with the process before anything
683 	 * can happen that might cause that process to need the descriptor.
684 	 * However, don't do this until after fork(2) can no longer fail.
685 	 */
686 	if (flags & RFPROCDESC)
687 		procdesc_new(p2, pdflags);
688 
689 	/*
690 	 * Both processes are set up, now check if any loadable modules want
691 	 * to adjust anything.
692 	 */
693 	EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
694 
695 	/*
696 	 * Set the child start time and mark the process as being complete.
697 	 */
698 	PROC_LOCK(p2);
699 	PROC_LOCK(p1);
700 	microuptime(&p2->p_stats->p_start);
701 	PROC_SLOCK(p2);
702 	p2->p_state = PRS_NORMAL;
703 	PROC_SUNLOCK(p2);
704 
705 #ifdef KDTRACE_HOOKS
706 	/*
707 	 * Tell the DTrace fasttrap provider about the new process so that any
708 	 * tracepoints inherited from the parent can be removed. We have to do
709 	 * this only after p_state is PRS_NORMAL since the fasttrap module will
710 	 * use pfind() later on.
711 	 */
712 	if ((flags & RFMEM) == 0 && dtrace_fasttrap_fork)
713 		dtrace_fasttrap_fork(p1, p2);
714 #endif
715 	if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED |
716 	    P_FOLLOWFORK)) {
717 		/*
718 		 * Arrange for debugger to receive the fork event.
719 		 *
720 		 * We can report PL_FLAG_FORKED regardless of
721 		 * P_FOLLOWFORK settings, but it does not make a sense
722 		 * for runaway child.
723 		 */
724 		td->td_dbgflags |= TDB_FORK;
725 		td->td_dbg_forked = p2->p_pid;
726 		td2->td_dbgflags |= TDB_STOPATFORK;
727 		_PHOLD(p2);
728 		p2_held = 1;
729 	}
730 	if (flags & RFPPWAIT) {
731 		td->td_pflags |= TDP_RFPPWAIT;
732 		td->td_rfppwait_p = p2;
733 	}
734 	PROC_UNLOCK(p2);
735 	if ((flags & RFSTOPPED) == 0) {
736 		/*
737 		 * If RFSTOPPED not requested, make child runnable and
738 		 * add to run queue.
739 		 */
740 		thread_lock(td2);
741 		TD_SET_CAN_RUN(td2);
742 		sched_add(td2, SRQ_BORING);
743 		thread_unlock(td2);
744 	}
745 
746 	/*
747 	 * Now can be swapped.
748 	 */
749 	_PRELE(p1);
750 	PROC_UNLOCK(p1);
751 
752 	/*
753 	 * Tell any interested parties about the new process.
754 	 */
755 	knote_fork(&p1->p_klist, p2->p_pid);
756 	SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
757 
758 	/*
759 	 * Wait until debugger is attached to child.
760 	 */
761 	PROC_LOCK(p2);
762 	while ((td2->td_dbgflags & TDB_STOPATFORK) != 0)
763 		cv_wait(&p2->p_dbgwait, &p2->p_mtx);
764 	if (p2_held)
765 		_PRELE(p2);
766 	PROC_UNLOCK(p2);
767 }
768 
769 int
770 fork1(struct thread *td, int flags, int pages, struct proc **procp,
771     int *procdescp, int pdflags)
772 {
773 	struct proc *p1;
774 	struct proc *newproc;
775 	int ok;
776 	struct thread *td2;
777 	struct vmspace *vm2;
778 	vm_ooffset_t mem_charged;
779 	int error;
780 	static int curfail;
781 	static struct timeval lastfail;
782 	struct file *fp_procdesc = NULL;
783 
784 	/* Check for the undefined or unimplemented flags. */
785 	if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
786 		return (EINVAL);
787 
788 	/* Signal value requires RFTSIGZMB. */
789 	if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
790 		return (EINVAL);
791 
792 	/* Can't copy and clear. */
793 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
794 		return (EINVAL);
795 
796 	/* Check the validity of the signal number. */
797 	if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
798 		return (EINVAL);
799 
800 	if ((flags & RFPROCDESC) != 0) {
801 		/* Can't not create a process yet get a process descriptor. */
802 		if ((flags & RFPROC) == 0)
803 			return (EINVAL);
804 
805 		/* Must provide a place to put a procdesc if creating one. */
806 		if (procdescp == NULL)
807 			return (EINVAL);
808 	}
809 
810 	p1 = td->td_proc;
811 
812 	/*
813 	 * Here we don't create a new process, but we divorce
814 	 * certain parts of a process from itself.
815 	 */
816 	if ((flags & RFPROC) == 0) {
817 		*procp = NULL;
818 		return (fork_norfproc(td, flags));
819 	}
820 
821 	/*
822 	 * If required, create a process descriptor in the parent first; we
823 	 * will abandon it if something goes wrong. We don't finit() until
824 	 * later.
825 	 */
826 	if (flags & RFPROCDESC) {
827 		error = falloc(td, &fp_procdesc, procdescp, 0);
828 		if (error != 0)
829 			return (error);
830 	}
831 
832 	mem_charged = 0;
833 	vm2 = NULL;
834 	if (pages == 0)
835 		pages = KSTACK_PAGES;
836 	/* Allocate new proc. */
837 	newproc = uma_zalloc(proc_zone, M_WAITOK);
838 	td2 = FIRST_THREAD_IN_PROC(newproc);
839 	if (td2 == NULL) {
840 		td2 = thread_alloc(pages);
841 		if (td2 == NULL) {
842 			error = ENOMEM;
843 			goto fail2;
844 		}
845 		proc_linkup(newproc, td2);
846 	} else {
847 		if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
848 			if (td2->td_kstack != 0)
849 				vm_thread_dispose(td2);
850 			if (!thread_alloc_stack(td2, pages)) {
851 				error = ENOMEM;
852 				goto fail2;
853 			}
854 		}
855 	}
856 
857 	if ((flags & RFMEM) == 0) {
858 		vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
859 		if (vm2 == NULL) {
860 			error = ENOMEM;
861 			goto fail2;
862 		}
863 		if (!swap_reserve(mem_charged)) {
864 			/*
865 			 * The swap reservation failed. The accounting
866 			 * from the entries of the copied vm2 will be
867 			 * substracted in vmspace_free(), so force the
868 			 * reservation there.
869 			 */
870 			swap_reserve_force(mem_charged);
871 			error = ENOMEM;
872 			goto fail2;
873 		}
874 	} else
875 		vm2 = NULL;
876 
877 	/*
878 	 * XXX: This is ugly; when we copy resource usage, we need to bump
879 	 *      per-cred resource counters.
880 	 */
881 	proc_set_cred_init(newproc, crhold(td->td_ucred));
882 
883 	/*
884 	 * Initialize resource accounting for the child process.
885 	 */
886 	error = racct_proc_fork(p1, newproc);
887 	if (error != 0) {
888 		error = EAGAIN;
889 		goto fail1;
890 	}
891 
892 #ifdef MAC
893 	mac_proc_init(newproc);
894 #endif
895 	knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
896 	STAILQ_INIT(&newproc->p_ktr);
897 
898 	/* We have to lock the process tree while we look for a pid. */
899 	sx_slock(&proctree_lock);
900 
901 	/*
902 	 * Although process entries are dynamically created, we still keep
903 	 * a global limit on the maximum number we will create.  Don't allow
904 	 * a nonprivileged user to use the last ten processes; don't let root
905 	 * exceed the limit. The variable nprocs is the current number of
906 	 * processes, maxproc is the limit.
907 	 */
908 	sx_xlock(&allproc_lock);
909 	if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
910 	    PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
911 		error = EAGAIN;
912 		goto fail;
913 	}
914 
915 	/*
916 	 * Increment the count of procs running with this uid. Don't allow
917 	 * a nonprivileged user to exceed their current limit.
918 	 *
919 	 * XXXRW: Can we avoid privilege here if it's not needed?
920 	 */
921 	error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
922 	if (error == 0)
923 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
924 	else {
925 		ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
926 		    lim_cur(td, RLIMIT_NPROC));
927 	}
928 	if (ok) {
929 		do_fork(td, flags, newproc, td2, vm2, pdflags);
930 
931 		/*
932 		 * Return child proc pointer to parent.
933 		 */
934 		*procp = newproc;
935 		if (flags & RFPROCDESC) {
936 			procdesc_finit(newproc->p_procdesc, fp_procdesc);
937 			fdrop(fp_procdesc, td);
938 		}
939 		racct_proc_fork_done(newproc);
940 		return (0);
941 	}
942 
943 	error = EAGAIN;
944 fail:
945 	sx_sunlock(&proctree_lock);
946 	if (ppsratecheck(&lastfail, &curfail, 1))
947 		printf("maxproc limit exceeded by uid %u (pid %d); see tuning(7) and login.conf(5)\n",
948 		    td->td_ucred->cr_ruid, p1->p_pid);
949 	sx_xunlock(&allproc_lock);
950 #ifdef MAC
951 	mac_proc_destroy(newproc);
952 #endif
953 	racct_proc_exit(newproc);
954 fail1:
955 	crfree(newproc->p_ucred);
956 	newproc->p_ucred = NULL;
957 fail2:
958 	if (vm2 != NULL)
959 		vmspace_free(vm2);
960 	uma_zfree(proc_zone, newproc);
961 	if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
962 		fdclose(td, fp_procdesc, *procdescp);
963 		fdrop(fp_procdesc, td);
964 	}
965 	pause("fork", hz / 2);
966 	return (error);
967 }
968 
969 /*
970  * Handle the return of a child process from fork1().  This function
971  * is called from the MD fork_trampoline() entry point.
972  */
973 void
974 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
975     struct trapframe *frame)
976 {
977 	struct proc *p;
978 	struct thread *td;
979 	struct thread *dtd;
980 
981 	td = curthread;
982 	p = td->td_proc;
983 	KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
984 
985 	CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
986 		td, td->td_sched, p->p_pid, td->td_name);
987 
988 	sched_fork_exit(td);
989 	/*
990 	* Processes normally resume in mi_switch() after being
991 	* cpu_switch()'ed to, but when children start up they arrive here
992 	* instead, so we must do much the same things as mi_switch() would.
993 	*/
994 	if ((dtd = PCPU_GET(deadthread))) {
995 		PCPU_SET(deadthread, NULL);
996 		thread_stash(dtd);
997 	}
998 	thread_unlock(td);
999 
1000 	/*
1001 	 * cpu_set_fork_handler intercepts this function call to
1002 	 * have this call a non-return function to stay in kernel mode.
1003 	 * initproc has its own fork handler, but it does return.
1004 	 */
1005 	KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1006 	callout(arg, frame);
1007 
1008 	/*
1009 	 * Check if a kernel thread misbehaved and returned from its main
1010 	 * function.
1011 	 */
1012 	if (p->p_flag & P_KTHREAD) {
1013 		printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1014 		    td->td_name, p->p_pid);
1015 		kproc_exit(0);
1016 	}
1017 	mtx_assert(&Giant, MA_NOTOWNED);
1018 
1019 	if (p->p_sysent->sv_schedtail != NULL)
1020 		(p->p_sysent->sv_schedtail)(td);
1021 }
1022 
1023 /*
1024  * Simplified back end of syscall(), used when returning from fork()
1025  * directly into user mode.  Giant is not held on entry, and must not
1026  * be held on return.  This function is passed in to fork_exit() as the
1027  * first parameter and is called when returning to a new userland process.
1028  */
1029 void
1030 fork_return(struct thread *td, struct trapframe *frame)
1031 {
1032 	struct proc *p, *dbg;
1033 
1034 	if (td->td_dbgflags & TDB_STOPATFORK) {
1035 		p = td->td_proc;
1036 		sx_xlock(&proctree_lock);
1037 		PROC_LOCK(p);
1038 		if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) ==
1039 		    (P_TRACED | P_FOLLOWFORK)) {
1040 			/*
1041 			 * If debugger still wants auto-attach for the
1042 			 * parent's children, do it now.
1043 			 */
1044 			dbg = p->p_pptr->p_pptr;
1045 			p->p_flag |= P_TRACED;
1046 			p->p_oppid = p->p_pptr->p_pid;
1047 			CTR2(KTR_PTRACE,
1048 		    "fork_return: attaching to new child pid %d: oppid %d",
1049 			    p->p_pid, p->p_oppid);
1050 			proc_reparent(p, dbg);
1051 			sx_xunlock(&proctree_lock);
1052 			td->td_dbgflags |= TDB_CHILD;
1053 			ptracestop(td, SIGSTOP);
1054 			td->td_dbgflags &= ~TDB_CHILD;
1055 		} else {
1056 			/*
1057 			 * ... otherwise clear the request.
1058 			 */
1059 			sx_xunlock(&proctree_lock);
1060 			td->td_dbgflags &= ~TDB_STOPATFORK;
1061 			cv_broadcast(&p->p_dbgwait);
1062 		}
1063 		PROC_UNLOCK(p);
1064 	}
1065 
1066 	userret(td, frame);
1067 
1068 #ifdef KTRACE
1069 	if (KTRPOINT(td, KTR_SYSRET))
1070 		ktrsysret(SYS_fork, 0, 0);
1071 #endif
1072 }
1073