xref: /freebsd/sys/kern/kern_fork.c (revision 77a0943ded95b9e6438f7db70c4a28e4d93946d4)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
39  * $FreeBSD$
40  */
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/vnode.h>
55 #include <sys/acct.h>
56 #include <sys/ktr.h>
57 #include <sys/ktrace.h>
58 #include <sys/unistd.h>
59 #include <sys/jail.h>
60 
61 #include <vm/vm.h>
62 #include <sys/lock.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_zone.h>
67 
68 #include <sys/user.h>
69 
70 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
71 
72 static int	fast_vfork = 1;
73 SYSCTL_INT(_kern, OID_AUTO, fast_vfork, CTLFLAG_RW, &fast_vfork, 0,
74     "flag to indicate whether we have a fast vfork()");
75 
76 /*
77  * These are the stuctures used to create a callout list for things to do
78  * when forking a process
79  */
80 struct forklist {
81 	forklist_fn function;
82 	TAILQ_ENTRY(forklist) next;
83 };
84 
85 TAILQ_HEAD(forklist_head, forklist);
86 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
87 
88 #ifndef _SYS_SYSPROTO_H_
89 struct fork_args {
90 	int     dummy;
91 };
92 #endif
93 
94 /* ARGSUSED */
95 int
96 fork(p, uap)
97 	struct proc *p;
98 	struct fork_args *uap;
99 {
100 	int error;
101 	struct proc *p2;
102 
103 	error = fork1(p, RFFDG | RFPROC, &p2);
104 	if (error == 0) {
105 		p->p_retval[0] = p2->p_pid;
106 		p->p_retval[1] = 0;
107 	}
108 	return error;
109 }
110 
111 /* ARGSUSED */
112 int
113 vfork(p, uap)
114 	struct proc *p;
115 	struct vfork_args *uap;
116 {
117 	int error;
118 	struct proc *p2;
119 
120 	error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
121 	if (error == 0) {
122 		p->p_retval[0] = p2->p_pid;
123 		p->p_retval[1] = 0;
124 	}
125 	return error;
126 }
127 
128 int
129 rfork(p, uap)
130 	struct proc *p;
131 	struct rfork_args *uap;
132 {
133 	int error;
134 	struct proc *p2;
135 
136 	/* mask kernel only flags out of the user flags */
137 	error = fork1(p, uap->flags & ~RFKERNELONLY, &p2);
138 	if (error == 0) {
139 		p->p_retval[0] = p2 ? p2->p_pid : 0;
140 		p->p_retval[1] = 0;
141 	}
142 	return error;
143 }
144 
145 
146 int	nprocs = 1;		/* process 0 */
147 static int nextpid = 0;
148 
149 /*
150  * Random component to nextpid generation.  We mix in a random factor to make
151  * it a little harder to predict.  We sanity check the modulus value to avoid
152  * doing it in critical paths.  Don't let it be too small or we pointlessly
153  * waste randomness entropy, and don't let it be impossibly large.  Using a
154  * modulus that is too big causes a LOT more process table scans and slows
155  * down fork processing as the pidchecked caching is defeated.
156  */
157 static int randompid = 0;
158 
159 static int
160 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
161 {
162 		int error, pid;
163 
164 		pid = randompid;
165 		error = sysctl_handle_int(oidp, &pid, 0, req);
166 		if (error || !req->newptr)
167 			return (error);
168 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
169 			pid = PID_MAX - 100;
170 		else if (pid < 2)			/* NOP */
171 			pid = 0;
172 		else if (pid < 100)			/* Make it reasonable */
173 			pid = 100;
174 		randompid = pid;
175 		return (error);
176 }
177 
178 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
179     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
180 
181 int
182 fork1(p1, flags, procp)
183 	struct proc *p1;				    /* parent proc */
184 	int flags;
185 	struct proc **procp;				    /* child proc */
186 {
187 	struct proc *p2, *pptr;
188 	uid_t uid;
189 	struct proc *newproc;
190 	int trypid;
191 	int ok;
192 	static int pidchecked = 0;
193 	struct forklist *ep;
194 
195 	/* Can't copy and clear */
196 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
197 		return (EINVAL);
198 
199 	/*
200 	 * Here we don't create a new process, but we divorce
201 	 * certain parts of a process from itself.
202 	 */
203 	if ((flags & RFPROC) == 0) {
204 
205 		vm_fork(p1, 0, flags);
206 
207 		/*
208 		 * Close all file descriptors.
209 		 */
210 		if (flags & RFCFDG) {
211 			struct filedesc *fdtmp;
212 			fdtmp = fdinit(p1);
213 			fdfree(p1);
214 			p1->p_fd = fdtmp;
215 		}
216 
217 		/*
218 		 * Unshare file descriptors (from parent.)
219 		 */
220 		if (flags & RFFDG) {
221 			if (p1->p_fd->fd_refcnt > 1) {
222 				struct filedesc *newfd;
223 				newfd = fdcopy(p1);
224 				fdfree(p1);
225 				p1->p_fd = newfd;
226 			}
227 		}
228 		*procp = NULL;
229 		return (0);
230 	}
231 
232 	/*
233 	 * Although process entries are dynamically created, we still keep
234 	 * a global limit on the maximum number we will create.  Don't allow
235 	 * a nonprivileged user to use the last process; don't let root
236 	 * exceed the limit. The variable nprocs is the current number of
237 	 * processes, maxproc is the limit.
238 	 */
239 	uid = p1->p_cred->p_ruid;
240 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
241 		tablefull("proc");
242 		return (EAGAIN);
243 	}
244 	/*
245 	 * Increment the nprocs resource before blocking can occur.  There
246 	 * are hard-limits as to the number of processes that can run.
247 	 */
248 	nprocs++;
249 
250 	/*
251 	 * Increment the count of procs running with this uid. Don't allow
252 	 * a nonprivileged user to exceed their current limit.
253 	 */
254 	ok = chgproccnt(p1->p_cred->p_uidinfo, 1,
255 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
256 	if (!ok) {
257 		/*
258 		 * Back out the process count
259 		 */
260 		nprocs--;
261 		return (EAGAIN);
262 	}
263 
264 	/* Allocate new proc. */
265 	newproc = zalloc(proc_zone);
266 
267 	/*
268 	 * Setup linkage for kernel based threading
269 	 */
270 	if((flags & RFTHREAD) != 0) {
271 		newproc->p_peers = p1->p_peers;
272 		p1->p_peers = newproc;
273 		newproc->p_leader = p1->p_leader;
274 	} else {
275 		newproc->p_peers = 0;
276 		newproc->p_leader = newproc;
277 	}
278 
279 	newproc->p_wakeup = 0;
280 
281 	newproc->p_vmspace = NULL;
282 
283 	/*
284 	 * Find an unused process ID.  We remember a range of unused IDs
285 	 * ready to use (from nextpid+1 through pidchecked-1).
286 	 *
287 	 * If RFHIGHPID is set (used during system boot), do not allocate
288 	 * low-numbered pids.
289 	 */
290 	lockmgr(&allproc_lock, LK_EXCLUSIVE, NULL, CURPROC);
291 	trypid = nextpid + 1;
292 	if (flags & RFHIGHPID) {
293 		if (trypid < 10) {
294 			trypid = 10;
295 		}
296 	} else {
297 	if (randompid)
298 			trypid += arc4random() % randompid;
299 	}
300 retry:
301 	/*
302 	 * If the process ID prototype has wrapped around,
303 	 * restart somewhat above 0, as the low-numbered procs
304 	 * tend to include daemons that don't exit.
305 	 */
306 	if (trypid >= PID_MAX) {
307 		trypid = trypid % PID_MAX;
308 		if (trypid < 100)
309 			trypid += 100;
310 		pidchecked = 0;
311 	}
312 	if (trypid >= pidchecked) {
313 		int doingzomb = 0;
314 
315 		pidchecked = PID_MAX;
316 		/*
317 		 * Scan the active and zombie procs to check whether this pid
318 		 * is in use.  Remember the lowest pid that's greater
319 		 * than trypid, so we can avoid checking for a while.
320 		 */
321 		p2 = LIST_FIRST(&allproc);
322 again:
323 		for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) {
324 			while (p2->p_pid == trypid ||
325 			    p2->p_pgrp->pg_id == trypid ||
326 			    p2->p_session->s_sid == trypid) {
327 				trypid++;
328 				if (trypid >= pidchecked)
329 					goto retry;
330 			}
331 			if (p2->p_pid > trypid && pidchecked > p2->p_pid)
332 				pidchecked = p2->p_pid;
333 			if (p2->p_pgrp->pg_id > trypid &&
334 			    pidchecked > p2->p_pgrp->pg_id)
335 				pidchecked = p2->p_pgrp->pg_id;
336 			if (p2->p_session->s_sid > trypid &&
337 			    pidchecked > p2->p_session->s_sid)
338 				pidchecked = p2->p_session->s_sid;
339 		}
340 		if (!doingzomb) {
341 			doingzomb = 1;
342 			p2 = LIST_FIRST(&zombproc);
343 			goto again;
344 		}
345 	}
346 
347 	/*
348 	 * RFHIGHPID does not mess with the nextpid counter during boot.
349 	 */
350 	if (flags & RFHIGHPID)
351 		pidchecked = 0;
352 	else
353 		nextpid = trypid;
354 
355 	p2 = newproc;
356 	p2->p_stat = SIDL;			/* protect against others */
357 	p2->p_pid = trypid;
358 	LIST_INSERT_HEAD(&allproc, p2, p_list);
359 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
360 	lockmgr(&allproc_lock, LK_RELEASE, NULL, CURPROC);
361 
362 	/*
363 	 * Make a proc table entry for the new process.
364 	 * Start by zeroing the section of proc that is zero-initialized,
365 	 * then copy the section that is copied directly from the parent.
366 	 */
367 	bzero(&p2->p_startzero,
368 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
369 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
370 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
371 
372 	p2->p_aioinfo = NULL;
373 
374 	/*
375 	 * Duplicate sub-structures as needed.
376 	 * Increase reference counts on shared objects.
377 	 * The p_stats and p_sigacts substructs are set in vm_fork.
378 	 */
379 	p2->p_flag = P_INMEM;
380 	if (p1->p_flag & P_PROFIL)
381 		startprofclock(p2);
382 	MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred),
383 	    M_SUBPROC, M_WAITOK);
384 	bcopy(p1->p_cred, p2->p_cred, sizeof(*p2->p_cred));
385 	p2->p_cred->p_refcnt = 1;
386 	crhold(p1->p_ucred);
387 	uihold(p1->p_cred->p_uidinfo);
388 
389 	if (p2->p_prison) {
390 		p2->p_prison->pr_ref++;
391 		p2->p_flag |= P_JAILED;
392 	}
393 
394 	if (p2->p_args)
395 		p2->p_args->ar_ref++;
396 
397 	if (flags & RFSIGSHARE) {
398 		p2->p_procsig = p1->p_procsig;
399 		p2->p_procsig->ps_refcnt++;
400 		if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
401 			struct sigacts *newsigacts;
402 			int s;
403 
404 			/* Create the shared sigacts structure */
405 			MALLOC(newsigacts, struct sigacts *,
406 			    sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
407 			s = splhigh();
408 			/*
409 			 * Set p_sigacts to the new shared structure.
410 			 * Note that this is updating p1->p_sigacts at the
411 			 * same time, since p_sigacts is just a pointer to
412 			 * the shared p_procsig->ps_sigacts.
413 			 */
414 			p2->p_sigacts  = newsigacts;
415 			bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
416 			    sizeof(*p2->p_sigacts));
417 			*p2->p_sigacts = p1->p_addr->u_sigacts;
418 			splx(s);
419 		}
420 	} else {
421 		MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
422 		    M_SUBPROC, M_WAITOK);
423 		bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
424 		p2->p_procsig->ps_refcnt = 1;
425 		p2->p_sigacts = NULL;	/* finished in vm_fork() */
426 	}
427 	if (flags & RFLINUXTHPN)
428 	        p2->p_sigparent = SIGUSR1;
429 	else
430 	        p2->p_sigparent = SIGCHLD;
431 
432 	/* bump references to the text vnode (for procfs) */
433 	p2->p_textvp = p1->p_textvp;
434 	if (p2->p_textvp)
435 		VREF(p2->p_textvp);
436 
437 	if (flags & RFCFDG)
438 		p2->p_fd = fdinit(p1);
439 	else if (flags & RFFDG)
440 		p2->p_fd = fdcopy(p1);
441 	else
442 		p2->p_fd = fdshare(p1);
443 
444 	/*
445 	 * If p_limit is still copy-on-write, bump refcnt,
446 	 * otherwise get a copy that won't be modified.
447 	 * (If PL_SHAREMOD is clear, the structure is shared
448 	 * copy-on-write.)
449 	 */
450 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
451 		p2->p_limit = limcopy(p1->p_limit);
452 	else {
453 		p2->p_limit = p1->p_limit;
454 		p2->p_limit->p_refcnt++;
455 	}
456 
457 	/*
458 	 * Preserve some more flags in subprocess.  P_PROFIL has already
459 	 * been preserved.
460 	 */
461 	p2->p_flag |= p1->p_flag & P_SUGID;
462 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
463 		p2->p_flag |= P_CONTROLT;
464 	if (flags & RFPPWAIT)
465 		p2->p_flag |= P_PPWAIT;
466 
467 	LIST_INSERT_AFTER(p1, p2, p_pglist);
468 
469 	/*
470 	 * Attach the new process to its parent.
471 	 *
472 	 * If RFNOWAIT is set, the newly created process becomes a child
473 	 * of init.  This effectively disassociates the child from the
474 	 * parent.
475 	 */
476 	if (flags & RFNOWAIT)
477 		pptr = initproc;
478 	else
479 		pptr = p1;
480 	p2->p_pptr = pptr;
481 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
482 	LIST_INIT(&p2->p_children);
483 	LIST_INIT(&p2->p_heldmtx);
484 	LIST_INIT(&p2->p_contested);
485 
486 #ifdef KTRACE
487 	/*
488 	 * Copy traceflag and tracefile if enabled.
489 	 * If not inherited, these were zeroed above.
490 	 */
491 	if (p1->p_traceflag&KTRFAC_INHERIT) {
492 		p2->p_traceflag = p1->p_traceflag;
493 		if ((p2->p_tracep = p1->p_tracep) != NULL)
494 			VREF(p2->p_tracep);
495 	}
496 #endif
497 
498 	/*
499 	 * set priority of child to be that of parent
500 	 */
501 	p2->p_estcpu = p1->p_estcpu;
502 
503 	/*
504 	 * This begins the section where we must prevent the parent
505 	 * from being swapped.
506 	 */
507 	PHOLD(p1);
508 
509 	/*
510 	 * Finish creating the child process.  It will return via a different
511 	 * execution path later.  (ie: directly into user mode)
512 	 */
513 	vm_fork(p1, p2, flags);
514 
515 	/*
516 	 * Both processes are set up, now check if any loadable modules want
517 	 * to adjust anything.
518 	 *   What if they have an error? XXX
519 	 */
520 	TAILQ_FOREACH(ep, &fork_list, next) {
521 		(*ep->function)(p1, p2, flags);
522 	}
523 
524 	/*
525 	 * If RFSTOPPED not requested, make child runnable and add to
526 	 * run queue.
527 	 */
528 	microtime(&(p2->p_stats->p_start));
529 	p2->p_acflag = AFORK;
530 	if ((flags & RFSTOPPED) == 0) {
531 		splhigh();
532 		mtx_enter(&sched_lock, MTX_SPIN);
533 		p2->p_stat = SRUN;
534 		setrunqueue(p2);
535 		mtx_exit(&sched_lock, MTX_SPIN);
536 		spl0();
537 	}
538 
539 	/*
540 	 * Now can be swapped.
541 	 */
542 	PRELE(p1);
543 
544 	/*
545 	 * tell any interested parties about the new process
546 	 */
547 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
548 
549 	/*
550 	 * Preserve synchronization semantics of vfork.  If waiting for
551 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
552 	 * proc (in case of exit).
553 	 */
554 	while (p2->p_flag & P_PPWAIT)
555 		tsleep(p1, PWAIT, "ppwait", 0);
556 
557 	/*
558 	 * Return child proc pointer to parent.
559 	 */
560 	*procp = p2;
561 	return (0);
562 }
563 
564 /*
565  * The next two functionms are general routines to handle adding/deleting
566  * items on the fork callout list.
567  *
568  * at_fork():
569  * Take the arguments given and put them onto the fork callout list,
570  * However first make sure that it's not already there.
571  * Returns 0 on success or a standard error number.
572  */
573 
574 int
575 at_fork(function)
576 	forklist_fn function;
577 {
578 	struct forklist *ep;
579 
580 #ifdef INVARIANTS
581 	/* let the programmer know if he's been stupid */
582 	if (rm_at_fork(function))
583 		printf("WARNING: fork callout entry (%p) already present\n",
584 		    function);
585 #endif
586 	ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
587 	if (ep == NULL)
588 		return (ENOMEM);
589 	ep->function = function;
590 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
591 	return (0);
592 }
593 
594 /*
595  * Scan the exit callout list for the given item and remove it..
596  * Returns the number of items removed (0 or 1)
597  */
598 
599 int
600 rm_at_fork(function)
601 	forklist_fn function;
602 {
603 	struct forklist *ep;
604 
605 	TAILQ_FOREACH(ep, &fork_list, next) {
606 		if (ep->function == function) {
607 			TAILQ_REMOVE(&fork_list, ep, next);
608 			free(ep, M_ATFORK);
609 			return(1);
610 		}
611 	}
612 	return (0);
613 }
614