xref: /dflybsd-src/sys/kern/kern_fork.c (revision 32832096b2b814ac219c4c4dc7fece32162b9ca4)
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: src/sys/kern/kern_fork.c,v 1.72.2.13 2003/06/06 20:21:32 tegge Exp $
40  * $DragonFly: src/sys/kern/kern_fork.c,v 1.16 2003/11/05 23:26:20 dillon Exp $
41  */
42 
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/proc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/vnode.h>
55 #include <sys/acct.h>
56 #include <sys/ktrace.h>
57 #include <sys/unistd.h>
58 #include <sys/jail.h>
59 
60 #include <vm/vm.h>
61 #include <sys/lock.h>
62 #include <vm/pmap.h>
63 #include <vm/vm_map.h>
64 #include <vm/vm_extern.h>
65 #include <vm/vm_zone.h>
66 
67 #include <sys/vmmeter.h>
68 #include <sys/user.h>
69 
70 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
71 
72 /*
73  * These are the stuctures used to create a callout list for things to do
74  * when forking a process
75  */
76 struct forklist {
77 	forklist_fn function;
78 	TAILQ_ENTRY(forklist) next;
79 };
80 
81 TAILQ_HEAD(forklist_head, forklist);
82 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
83 
84 int forksleep; /* Place for fork1() to sleep on. */
85 
86 /* ARGSUSED */
87 int
88 fork(struct fork_args *uap)
89 {
90 	struct proc *p = curproc;
91 	struct proc *p2;
92 	int error;
93 
94 	error = fork1(p, RFFDG | RFPROC, &p2);
95 	if (error == 0) {
96 		start_forked_proc(p, p2);
97 		uap->sysmsg_fds[0] = p2->p_pid;
98 		uap->sysmsg_fds[1] = 0;
99 	}
100 	return error;
101 }
102 
103 /* ARGSUSED */
104 int
105 vfork(struct vfork_args *uap)
106 {
107 	struct proc *p = curproc;
108 	struct proc *p2;
109 	int error;
110 
111 	error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
112 	if (error == 0) {
113 		start_forked_proc(p, p2);
114 		uap->sysmsg_fds[0] = p2->p_pid;
115 		uap->sysmsg_fds[1] = 0;
116 	}
117 	return error;
118 }
119 
120 int
121 rfork(struct rfork_args *uap)
122 {
123 	struct proc *p = curproc;
124 	struct proc *p2;
125 	int error;
126 
127 	error = fork1(p, uap->flags, &p2);
128 	if (error == 0) {
129 		start_forked_proc(p, p2);
130 		uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
131 		uap->sysmsg_fds[1] = 0;
132 	}
133 	return error;
134 }
135 
136 
137 int	nprocs = 1;		/* process 0 */
138 static int nextpid = 0;
139 
140 /*
141  * Random component to nextpid generation.  We mix in a random factor to make
142  * it a little harder to predict.  We sanity check the modulus value to avoid
143  * doing it in critical paths.  Don't let it be too small or we pointlessly
144  * waste randomness entropy, and don't let it be impossibly large.  Using a
145  * modulus that is too big causes a LOT more process table scans and slows
146  * down fork processing as the pidchecked caching is defeated.
147  */
148 static int randompid = 0;
149 
150 static int
151 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
152 {
153 		int error, pid;
154 
155 		pid = randompid;
156 		error = sysctl_handle_int(oidp, &pid, 0, req);
157 		if (error || !req->newptr)
158 			return (error);
159 		if (pid < 0 || pid > PID_MAX - 100)	/* out of range */
160 			pid = PID_MAX - 100;
161 		else if (pid < 2)			/* NOP */
162 			pid = 0;
163 		else if (pid < 100)			/* Make it reasonable */
164 			pid = 100;
165 		randompid = pid;
166 		return (error);
167 }
168 
169 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
170     0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
171 
172 int
173 fork1(p1, flags, procp)
174 	struct proc *p1;
175 	int flags;
176 	struct proc **procp;
177 {
178 	struct proc *p2, *pptr;
179 	uid_t uid;
180 	struct proc *newproc;
181 	int ok;
182 	static int pidchecked = 0;
183 	struct forklist *ep;
184 	struct filedesc_to_leader *fdtol;
185 
186 	if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
187 		return (EINVAL);
188 
189 	/*
190 	 * Here we don't create a new process, but we divorce
191 	 * certain parts of a process from itself.
192 	 */
193 	if ((flags & RFPROC) == 0) {
194 
195 		vm_fork(p1, 0, flags);
196 
197 		/*
198 		 * Close all file descriptors.
199 		 */
200 		if (flags & RFCFDG) {
201 			struct filedesc *fdtmp;
202 			fdtmp = fdinit(p1);
203 			fdfree(p1);
204 			p1->p_fd = fdtmp;
205 		}
206 
207 		/*
208 		 * Unshare file descriptors (from parent.)
209 		 */
210 		if (flags & RFFDG) {
211 			if (p1->p_fd->fd_refcnt > 1) {
212 				struct filedesc *newfd;
213 				newfd = fdcopy(p1);
214 				fdfree(p1);
215 				p1->p_fd = newfd;
216 			}
217 		}
218 		*procp = NULL;
219 		return (0);
220 	}
221 
222 	/*
223 	 * Although process entries are dynamically created, we still keep
224 	 * a global limit on the maximum number we will create.  Don't allow
225 	 * a nonprivileged user to use the last ten processes; don't let root
226 	 * exceed the limit. The variable nprocs is the current number of
227 	 * processes, maxproc is the limit.
228 	 */
229 	uid = p1->p_ucred->cr_ruid;
230 	if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
231 		tsleep(&forksleep, 0, "fork", hz / 2);
232 		return (EAGAIN);
233 	}
234 	/*
235 	 * Increment the nprocs resource before blocking can occur.  There
236 	 * are hard-limits as to the number of processes that can run.
237 	 */
238 	nprocs++;
239 
240 	/*
241 	 * Increment the count of procs running with this uid. Don't allow
242 	 * a nonprivileged user to exceed their current limit.
243 	 */
244 	ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
245 		(uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
246 	if (!ok) {
247 		/*
248 		 * Back out the process count
249 		 */
250 		nprocs--;
251 		tsleep(&forksleep, 0, "fork", hz / 2);
252 		return (EAGAIN);
253 	}
254 
255 	/* Allocate new proc. */
256 	newproc = zalloc(proc_zone);
257 
258 	/*
259 	 * Setup linkage for kernel based threading
260 	 */
261 	if((flags & RFTHREAD) != 0) {
262 		newproc->p_peers = p1->p_peers;
263 		p1->p_peers = newproc;
264 		newproc->p_leader = p1->p_leader;
265 	} else {
266 		newproc->p_peers = 0;
267 		newproc->p_leader = newproc;
268 	}
269 
270 	newproc->p_wakeup = 0;
271 	newproc->p_vmspace = NULL;
272 
273 	/*
274 	 * Find an unused process ID.  We remember a range of unused IDs
275 	 * ready to use (from nextpid+1 through pidchecked-1).
276 	 */
277 	nextpid++;
278 	if (randompid)
279 		nextpid += arc4random() % randompid;
280 retry:
281 	/*
282 	 * If the process ID prototype has wrapped around,
283 	 * restart somewhat above 0, as the low-numbered procs
284 	 * tend to include daemons that don't exit.
285 	 */
286 	if (nextpid >= PID_MAX) {
287 		nextpid = nextpid % PID_MAX;
288 		if (nextpid < 100)
289 			nextpid += 100;
290 		pidchecked = 0;
291 	}
292 	if (nextpid >= pidchecked) {
293 		int doingzomb = 0;
294 
295 		pidchecked = PID_MAX;
296 		/*
297 		 * Scan the active and zombie procs to check whether this pid
298 		 * is in use.  Remember the lowest pid that's greater
299 		 * than nextpid, so we can avoid checking for a while.
300 		 */
301 		p2 = LIST_FIRST(&allproc);
302 again:
303 		for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) {
304 			while (p2->p_pid == nextpid ||
305 			    p2->p_pgrp->pg_id == nextpid ||
306 			    p2->p_session->s_sid == nextpid) {
307 				nextpid++;
308 				if (nextpid >= pidchecked)
309 					goto retry;
310 			}
311 			if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
312 				pidchecked = p2->p_pid;
313 			if (p2->p_pgrp->pg_id > nextpid &&
314 			    pidchecked > p2->p_pgrp->pg_id)
315 				pidchecked = p2->p_pgrp->pg_id;
316 			if (p2->p_session->s_sid > nextpid &&
317 			    pidchecked > p2->p_session->s_sid)
318 				pidchecked = p2->p_session->s_sid;
319 		}
320 		if (!doingzomb) {
321 			doingzomb = 1;
322 			p2 = LIST_FIRST(&zombproc);
323 			goto again;
324 		}
325 	}
326 
327 	p2 = newproc;
328 	p2->p_stat = SIDL;			/* protect against others */
329 	p2->p_pid = nextpid;
330 	LIST_INSERT_HEAD(&allproc, p2, p_list);
331 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
332 
333 	/*
334 	 * Make a proc table entry for the new process.
335 	 * Start by zeroing the section of proc that is zero-initialized,
336 	 * then copy the section that is copied directly from the parent.
337 	 */
338 	bzero(&p2->p_startzero,
339 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
340 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
341 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
342 
343 	p2->p_aioinfo = NULL;
344 
345 	/*
346 	 * Duplicate sub-structures as needed.
347 	 * Increase reference counts on shared objects.
348 	 * The p_stats and p_sigacts substructs are set in vm_fork.
349 	 *
350 	 * P_CP_RELEASED indicates that the process is starting out in
351 	 * the kernel (in the fork trampoline).  The flag will be converted
352 	 * to P_CURPROC when the new process calls userret() and attempts
353 	 * to return to userland
354 	 */
355 	p2->p_flag = P_INMEM | P_CP_RELEASED;
356 	if (p1->p_flag & P_PROFIL)
357 		startprofclock(p2);
358 	p2->p_ucred = crhold(p1->p_ucred);
359 
360 	if (p2->p_ucred->cr_prison) {
361 		p2->p_ucred->cr_prison->pr_ref++;
362 		p2->p_flag |= P_JAILED;
363 	}
364 
365 	if (p2->p_args)
366 		p2->p_args->ar_ref++;
367 
368 	if (flags & RFSIGSHARE) {
369 		p2->p_procsig = p1->p_procsig;
370 		p2->p_procsig->ps_refcnt++;
371 		if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
372 			struct sigacts *newsigacts;
373 			int s;
374 
375 			/* Create the shared sigacts structure */
376 			MALLOC(newsigacts, struct sigacts *,
377 			    sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
378 			s = splhigh();
379 			/*
380 			 * Set p_sigacts to the new shared structure.
381 			 * Note that this is updating p1->p_sigacts at the
382 			 * same time, since p_sigacts is just a pointer to
383 			 * the shared p_procsig->ps_sigacts.
384 			 */
385 			p2->p_sigacts  = newsigacts;
386 			bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
387 			    sizeof(*p2->p_sigacts));
388 			*p2->p_sigacts = p1->p_addr->u_sigacts;
389 			splx(s);
390 		}
391 	} else {
392 		MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
393 		    M_SUBPROC, M_WAITOK);
394 		bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
395 		p2->p_procsig->ps_refcnt = 1;
396 		p2->p_sigacts = NULL;	/* finished in vm_fork() */
397 	}
398 	if (flags & RFLINUXTHPN)
399 	        p2->p_sigparent = SIGUSR1;
400 	else
401 	        p2->p_sigparent = SIGCHLD;
402 
403 	/* bump references to the text vnode (for procfs) */
404 	p2->p_textvp = p1->p_textvp;
405 	if (p2->p_textvp)
406 		VREF(p2->p_textvp);
407 
408 	if (flags & RFCFDG) {
409 		p2->p_fd = fdinit(p1);
410 		fdtol = NULL;
411 	} else if (flags & RFFDG) {
412 		p2->p_fd = fdcopy(p1);
413 		fdtol = NULL;
414 	} else {
415 		p2->p_fd = fdshare(p1);
416 		if (p1->p_fdtol == NULL)
417 			p1->p_fdtol =
418 				filedesc_to_leader_alloc(NULL,
419 							 p1->p_leader);
420 		if ((flags & RFTHREAD) != 0) {
421 			/*
422 			 * Shared file descriptor table and
423 			 * shared process leaders.
424 			 */
425 			fdtol = p1->p_fdtol;
426 			fdtol->fdl_refcount++;
427 		} else {
428 			/*
429 			 * Shared file descriptor table, and
430 			 * different process leaders
431 			 */
432 			fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
433 		}
434 	}
435 	p2->p_fdtol = fdtol;
436 
437 	/*
438 	 * If p_limit is still copy-on-write, bump refcnt,
439 	 * otherwise get a copy that won't be modified.
440 	 * (If PL_SHAREMOD is clear, the structure is shared
441 	 * copy-on-write.)
442 	 */
443 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
444 		p2->p_limit = limcopy(p1->p_limit);
445 	else {
446 		p2->p_limit = p1->p_limit;
447 		p2->p_limit->p_refcnt++;
448 	}
449 
450 	/*
451 	 * Preserve some more flags in subprocess.  P_PROFIL has already
452 	 * been preserved.
453 	 */
454 	p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
455 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
456 		p2->p_flag |= P_CONTROLT;
457 	if (flags & RFPPWAIT)
458 		p2->p_flag |= P_PPWAIT;
459 
460 	LIST_INSERT_AFTER(p1, p2, p_pglist);
461 
462 	/*
463 	 * Attach the new process to its parent.
464 	 *
465 	 * If RFNOWAIT is set, the newly created process becomes a child
466 	 * of init.  This effectively disassociates the child from the
467 	 * parent.
468 	 */
469 	if (flags & RFNOWAIT)
470 		pptr = initproc;
471 	else
472 		pptr = p1;
473 	p2->p_pptr = pptr;
474 	LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
475 	LIST_INIT(&p2->p_children);
476 	varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
477 
478 #ifdef KTRACE
479 	/*
480 	 * Copy traceflag and tracefile if enabled.  If not inherited,
481 	 * these were zeroed above but we still could have a trace race
482 	 * so make sure p2's p_tracep is NULL.
483 	 */
484 	if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
485 		p2->p_traceflag = p1->p_traceflag;
486 		if ((p2->p_tracep = p1->p_tracep) != NULL)
487 			VREF(p2->p_tracep);
488 	}
489 #endif
490 
491 	/*
492 	 * set priority of child to be that of parent
493 	 */
494 	p2->p_estcpu = p1->p_estcpu;
495 
496 	/*
497 	 * This begins the section where we must prevent the parent
498 	 * from being swapped.
499 	 */
500 	PHOLD(p1);
501 
502 	/*
503 	 * Finish creating the child process.  It will return via a different
504 	 * execution path later.  (ie: directly into user mode)
505 	 */
506 	vm_fork(p1, p2, flags);
507 
508 	if (flags == (RFFDG | RFPROC)) {
509 		mycpu->gd_cnt.v_forks++;
510 		mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
511 	} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
512 		mycpu->gd_cnt.v_vforks++;
513 		mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
514 	} else if (p1 == &proc0) {
515 		mycpu->gd_cnt.v_kthreads++;
516 		mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
517 	} else {
518 		mycpu->gd_cnt.v_rforks++;
519 		mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
520 	}
521 
522 	/*
523 	 * Both processes are set up, now check if any loadable modules want
524 	 * to adjust anything.
525 	 *   What if they have an error? XXX
526 	 */
527 	TAILQ_FOREACH(ep, &fork_list, next) {
528 		(*ep->function)(p1, p2, flags);
529 	}
530 
531 	/*
532 	 * Make child runnable and add to run queue.
533 	 */
534 	microtime(&(p2->p_stats->p_start));
535 	p2->p_acflag = AFORK;
536 
537 	/*
538 	 * tell any interested parties about the new process
539 	 */
540 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
541 
542 	/*
543 	 * Return child proc pointer to parent.
544 	 */
545 	*procp = p2;
546 	return (0);
547 }
548 
549 /*
550  * The next two functionms are general routines to handle adding/deleting
551  * items on the fork callout list.
552  *
553  * at_fork():
554  * Take the arguments given and put them onto the fork callout list,
555  * However first make sure that it's not already there.
556  * Returns 0 on success or a standard error number.
557  */
558 
559 int
560 at_fork(function)
561 	forklist_fn function;
562 {
563 	struct forklist *ep;
564 
565 #ifdef INVARIANTS
566 	/* let the programmer know if he's been stupid */
567 	if (rm_at_fork(function))
568 		printf("WARNING: fork callout entry (%p) already present\n",
569 		    function);
570 #endif
571 	ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
572 	if (ep == NULL)
573 		return (ENOMEM);
574 	ep->function = function;
575 	TAILQ_INSERT_TAIL(&fork_list, ep, next);
576 	return (0);
577 }
578 
579 /*
580  * Scan the exit callout list for the given item and remove it..
581  * Returns the number of items removed (0 or 1)
582  */
583 
584 int
585 rm_at_fork(function)
586 	forklist_fn function;
587 {
588 	struct forklist *ep;
589 
590 	TAILQ_FOREACH(ep, &fork_list, next) {
591 		if (ep->function == function) {
592 			TAILQ_REMOVE(&fork_list, ep, next);
593 			free(ep, M_ATFORK);
594 			return(1);
595 		}
596 	}
597 	return (0);
598 }
599 
600 /*
601  * Add a forked process to the run queue after any remaining setup, such
602  * as setting the fork handler, has been completed.
603  */
604 
605 void
606 start_forked_proc(struct proc *p1, struct proc *p2)
607 {
608 	/*
609 	 * Move from SIDL to RUN queue, and activate the process's thread.
610 	 * Activation of the thread effectively makes the process "a"
611 	 * current process, so we do not setrunqueue().
612 	 */
613 	KASSERT(p2->p_stat == SIDL,
614 	    ("cannot start forked process, bad status: %p", p2));
615 	(void) splhigh();
616 	p2->p_stat = SRUN;
617 	setrunqueue(p2);
618 	(void) spl0();
619 
620 	/*
621 	 * Now can be swapped.
622 	 */
623 	PRELE(p1);
624 
625 	/*
626 	 * Preserve synchronization semantics of vfork.  If waiting for
627 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
628 	 * proc (in case of exit).
629 	 */
630 	while (p2->p_flag & P_PPWAIT)
631 		tsleep(p1, 0, "ppwait", 0);
632 }
633 
634