xref: /openbsd-src/sys/kern/kern_fork.c (revision e6fd2ff84860d099ea5ae17d34f29629f598ebc7) 
1 /*	$OpenBSD: kern_fork.c,v 1.136 2012/04/10 15:50:52 guenther Exp $	*/
2 /*	$NetBSD: kern_fork.c,v 1.29 1996/02/09 18:59:34 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1989, 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/filedesc.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mount.h>
46 #include <sys/proc.h>
47 #include <sys/exec.h>
48 #include <sys/resourcevar.h>
49 #include <sys/signalvar.h>
50 #include <sys/vnode.h>
51 #include <sys/file.h>
52 #include <sys/acct.h>
53 #include <sys/ktrace.h>
54 #include <sys/sched.h>
55 #include <dev/rndvar.h>
56 #include <sys/pool.h>
57 #include <sys/mman.h>
58 #include <sys/ptrace.h>
59 
60 #include <sys/syscallargs.h>
61 
62 #include "systrace.h"
63 #include <dev/systrace.h>
64 
65 #include <uvm/uvm_extern.h>
66 #include <uvm/uvm_map.h>
67 
68 #ifdef __HAVE_MD_TCB
69 # include <machine/tcb.h>
70 #endif
71 
72 int	nprocesses = 1;		/* process 0 */
73 int	nthreads = 1;		/* proc 0 */
74 int	randompid;		/* when set to 1, pid's go random */
75 pid_t	lastpid;
76 struct	forkstat forkstat;
77 
78 void fork_return(void *);
79 void tfork_child_return(void *);
80 int pidtaken(pid_t);
81 
82 void process_new(struct proc *, struct process *);
83 
84 void
85 fork_return(void *arg)
86 {
87 	struct proc *p = (struct proc *)arg;
88 
89 	if (p->p_p->ps_flags & PS_TRACED)
90 		psignal(p, SIGTRAP);
91 
92 	child_return(p);
93 }
94 
95 /*ARGSUSED*/
96 int
97 sys_fork(struct proc *p, void *v, register_t *retval)
98 {
99 	int flags;
100 
101 	flags = FORK_FORK;
102 	if (p->p_p->ps_ptmask & PTRACE_FORK)
103 		flags |= FORK_PTRACE;
104 	return (fork1(p, SIGCHLD, flags, NULL, 0,
105 	    fork_return, NULL, retval, NULL));
106 }
107 
108 /*ARGSUSED*/
109 int
110 sys_vfork(struct proc *p, void *v, register_t *retval)
111 {
112 	return (fork1(p, SIGCHLD, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
113 	    NULL, retval, NULL));
114 }
115 
116 int
117 sys_rfork(struct proc *p, void *v, register_t *retval)
118 {
119 	struct sys_rfork_args /* {
120 		syscallarg(int) flags;
121 	} */ *uap = v;
122 
123 	int rforkflags;
124 	int flags;
125 
126 	flags = FORK_RFORK;
127 	rforkflags = SCARG(uap, flags);
128 
129 	if ((rforkflags & RFPROC) == 0)
130 		return (EINVAL);
131 
132 	switch(rforkflags & (RFFDG|RFCFDG)) {
133 	case (RFFDG|RFCFDG):
134 		return EINVAL;
135 	case RFCFDG:
136 		flags |= FORK_CLEANFILES;
137 		break;
138 	case RFFDG:
139 		break;
140 	default:
141 		flags |= FORK_SHAREFILES;
142 		break;
143 	}
144 
145 	if (rforkflags & RFNOWAIT)
146 		flags |= FORK_NOZOMBIE;
147 
148 	if (rforkflags & RFMEM)
149 		flags |= FORK_SHAREVM;
150 
151 	if (rforkflags & RFTHREAD)
152 		flags |= FORK_THREAD | FORK_SIGHAND | FORK_NOZOMBIE;
153 
154 	return (fork1(p, SIGCHLD, flags, NULL, 0, NULL, NULL, retval, NULL));
155 }
156 
157 int
158 sys___tfork(struct proc *p, void *v, register_t *retval)
159 {
160 	struct sys___tfork_args /* {
161 		syscallarg(struct __tfork) *param;
162 	} */ *uap = v;
163 	struct __tfork param;
164 	int flags;
165 	int error;
166 
167 	if ((error = copyin(SCARG(uap, param), &param, sizeof(param))))
168 		return (error);
169 
170 	/* XXX will supersede rfork at some point... */
171 	if (param.tf_flags != 0)
172 		return (EINVAL);
173 
174 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
175 	    | FORK_NOZOMBIE | FORK_SHAREFILES;
176 
177 	return (fork1(p, 0, flags, NULL, param.tf_tid, tfork_child_return,
178 	    param.tf_tcb, retval, NULL));
179 }
180 
181 void
182 tfork_child_return(void *arg)
183 {
184 	struct proc *p = curproc;
185 
186 	TCB_SET(p, arg);
187 	child_return(p);
188 }
189 
190 /*
191  * Allocate and initialize a new process.
192  */
193 void
194 process_new(struct proc *p, struct process *parent)
195 {
196 	struct process *pr;
197 
198 	pr = pool_get(&process_pool, PR_WAITOK);
199 	pr->ps_mainproc = p;
200 
201 	TAILQ_INIT(&pr->ps_threads);
202 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
203 	pr->ps_pptr = parent;
204 	LIST_INIT(&pr->ps_children);
205 	pr->ps_refcnt = 1;
206 
207 	/*
208 	 * Make a process structure for the new process.
209 	 * Start by zeroing the section of proc that is zero-initialized,
210 	 * then copy the section that is copied directly from the parent.
211 	 */
212 	bzero(&pr->ps_startzero,
213 	    (unsigned) ((caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero));
214 	bcopy(&parent->ps_startcopy, &pr->ps_startcopy,
215 	    (unsigned) ((caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy));
216 
217 	/* post-copy fixups */
218 	pr->ps_cred = pool_get(&pcred_pool, PR_WAITOK);
219 	bcopy(parent->ps_cred, pr->ps_cred, sizeof(*pr->ps_cred));
220 	crhold(parent->ps_cred->pc_ucred);
221 	pr->ps_limit->p_refcnt++;
222 
223 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
224 	timeout_set(&pr->ps_virt_to, virttimer_trampoline, pr);
225 	timeout_set(&pr->ps_prof_to, proftimer_trampoline, pr);
226 
227 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
228 	if (parent->ps_session->s_ttyvp != NULL &&
229 	    parent->ps_flags & PS_CONTROLT)
230 		atomic_setbits_int(&pr->ps_flags, PS_CONTROLT);
231 
232 	p->p_p = pr;
233 }
234 
235 /* print the 'table full' message once per 10 seconds */
236 struct timeval fork_tfmrate = { 10, 0 };
237 
238 int
239 fork1(struct proc *curp, int exitsig, int flags, void *stack, pid_t *tidptr,
240     void (*func)(void *), void *arg, register_t *retval,
241     struct proc **rnewprocp)
242 {
243 	struct process *curpr = curp->p_p;
244 	struct process *pr;
245 	struct proc *p;
246 	uid_t uid;
247 	struct vmspace *vm;
248 	int count;
249 	vaddr_t uaddr;
250 	int s;
251 	struct  ptrace_state *newptstat = NULL;
252 #if NSYSTRACE > 0
253 	void *newstrp = NULL;
254 #endif
255 
256 	/* sanity check some flag combinations */
257 	if (flags & FORK_THREAD) {
258 		if (!rthreads_enabled)
259 			return (ENOTSUP);
260 		if ((flags & (FORK_SIGHAND | FORK_NOZOMBIE)) !=
261 		    (FORK_SIGHAND | FORK_NOZOMBIE))
262 			return (EINVAL);
263 	}
264 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
265 		return (EINVAL);
266 
267 	/*
268 	 * Although process entries are dynamically created, we still keep
269 	 * a global limit on the maximum number we will create. We reserve
270 	 * the last 5 processes to root. The variable nprocesses is the
271 	 * current number of processes, maxprocess is the limit.  Similar
272 	 * rules for threads (struct proc): we reserve the last 5 to root;
273 	 * the variable nthreads is the current number of procs, maxthread is
274 	 * the limit.
275 	 */
276 	uid = curp->p_cred->p_ruid;
277 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
278 		static struct timeval lasttfm;
279 
280 		if (ratecheck(&lasttfm, &fork_tfmrate))
281 			tablefull("proc");
282 		return (EAGAIN);
283 	}
284 	nthreads++;
285 
286 	if ((flags & FORK_THREAD) == 0) {
287 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
288 		    nprocesses >= maxprocess) {
289 			static struct timeval lasttfm;
290 
291 			if (ratecheck(&lasttfm, &fork_tfmrate))
292 				tablefull("process");
293 			nthreads--;
294 			return (EAGAIN);
295 		}
296 		nprocesses++;
297 
298 		/*
299 		 * Increment the count of processes running with
300 		 * this uid.  Don't allow a nonprivileged user to
301 		 * exceed their current limit.
302 		 */
303 		count = chgproccnt(uid, 1);
304 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
305 			(void)chgproccnt(uid, -1);
306 			nprocesses--;
307 			nthreads--;
308 			return (EAGAIN);
309 		}
310 	}
311 
312 	uaddr = uvm_km_kmemalloc_pla(kernel_map, uvm.kernel_object, USPACE,
313 	    USPACE_ALIGN, UVM_KMF_ZERO,
314 	    no_constraint.ucr_low, no_constraint.ucr_high,
315 	    0, 0, USPACE/PAGE_SIZE);
316 	if (uaddr == 0) {
317 		chgproccnt(uid, -1);
318 		nprocesses--;
319 		nthreads--;
320 		return (ENOMEM);
321 	}
322 
323 	/*
324 	 * From now on, we're committed to the fork and cannot fail.
325 	 */
326 
327 	/* Allocate new proc. */
328 	p = pool_get(&proc_pool, PR_WAITOK);
329 
330 	p->p_stat = SIDL;			/* protect against others */
331 	p->p_exitsig = exitsig;
332 	p->p_flag = 0;
333 
334 	if (flags & FORK_THREAD) {
335 		atomic_setbits_int(&p->p_flag, P_THREAD);
336 		p->p_p = pr = curpr;
337 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
338 		pr->ps_refcnt++;
339 	} else {
340 		process_new(p, curpr);
341 		pr = p->p_p;
342 	}
343 
344 	/*
345 	 * Make a proc table entry for the new process.
346 	 * Start by zeroing the section of proc that is zero-initialized,
347 	 * then copy the section that is copied directly from the parent.
348 	 */
349 	bzero(&p->p_startzero,
350 	    (unsigned) ((caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero));
351 	bcopy(&curp->p_startcopy, &p->p_startcopy,
352 	    (unsigned) ((caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy));
353 
354 	/*
355 	 * Initialize the timeouts.
356 	 */
357 	timeout_set(&p->p_sleep_to, endtsleep, p);
358 
359 	/*
360 	 * Duplicate sub-structures as needed.
361 	 * Increase reference counts on shared objects.
362 	 */
363 	if (curp->p_flag & P_PROFIL)
364 		startprofclock(p);
365 	if (flags & FORK_PTRACE)
366 		atomic_setbits_int(&pr->ps_flags, curpr->ps_flags & PS_TRACED);
367 
368 	/* bump references to the text vnode (for procfs) */
369 	p->p_textvp = curp->p_textvp;
370 	if (p->p_textvp)
371 		vref(p->p_textvp);
372 
373 	if (flags & FORK_CLEANFILES)
374 		p->p_fd = fdinit(curp);
375 	else if (flags & FORK_SHAREFILES)
376 		p->p_fd = fdshare(curp);
377 	else
378 		p->p_fd = fdcopy(curp);
379 
380 	if (flags & FORK_PPWAIT) {
381 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
382 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
383 	}
384 	if (flags & FORK_NOZOMBIE)
385 		atomic_setbits_int(&p->p_flag, P_NOZOMBIE);
386 
387 #ifdef KTRACE
388 	/*
389 	 * Copy traceflag and tracefile if enabled.
390 	 * If not inherited, these were zeroed above.
391 	 */
392 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
393 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
394 		    curpr->ps_tracecred);
395 #endif
396 
397 	/*
398 	 * set priority of child to be that of parent
399 	 * XXX should move p_estcpu into the region of struct proc which gets
400 	 * copied.
401 	 */
402 	scheduler_fork_hook(curp, p);
403 
404 	/*
405 	 * Create signal actions for the child process.
406 	 */
407 	if (flags & FORK_SIGHAND)
408 		p->p_sigacts = sigactsshare(curp);
409 	else
410 		p->p_sigacts = sigactsinit(curp);
411 	if (flags & FORK_THREAD)
412 		sigstkinit(&p->p_sigstk);
413 
414 	/*
415 	 * If emulation has process fork hook, call it now.
416 	 */
417 	if (p->p_emul->e_proc_fork)
418 		(*p->p_emul->e_proc_fork)(p, curp);
419 
420 	p->p_addr = (struct user *)uaddr;
421 
422 	/*
423 	 * Finish creating the child process.  It will return through a
424 	 * different path later.
425 	 */
426 	uvm_fork(curp, p, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
427 	    0, func ? func : child_return, arg ? arg : p);
428 
429 	vm = p->p_vmspace;
430 
431 	if (flags & FORK_FORK) {
432 		forkstat.cntfork++;
433 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
434 	} else if (flags & FORK_VFORK) {
435 		forkstat.cntvfork++;
436 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
437 #if 0
438 	} else if (flags & FORK_TFORK) {
439 		forkstat.cnttfork++;
440 #endif
441 	} else if (flags & FORK_RFORK) {
442 		forkstat.cntrfork++;
443 		forkstat.sizrfork += vm->vm_dsize + vm->vm_ssize;
444 	} else {
445 		forkstat.cntkthread++;
446 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
447 	}
448 
449 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
450 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
451 #if NSYSTRACE > 0
452 	if (ISSET(curp->p_flag, P_SYSTRACE))
453 		newstrp = systrace_getproc();
454 #endif
455 
456 	/* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
457 	do {
458 		lastpid = 1 + (randompid ? arc4random() : lastpid) % PID_MAX;
459 	} while (pidtaken(lastpid));
460 	p->p_pid = lastpid;
461 
462 	LIST_INSERT_HEAD(&allproc, p, p_list);
463 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
464 	if ((flags & FORK_THREAD) == 0) {
465 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
466 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
467 
468 		if (pr->ps_flags & PS_TRACED) {
469 			pr->ps_oppid = curpr->ps_pid;
470 			if (pr->ps_pptr != curpr->ps_pptr)
471 				proc_reparent(pr, curpr->ps_pptr);
472 
473 			/*
474 			 * Set ptrace status.
475 			 */
476 			if (flags & FORK_FORK) {
477 				pr->ps_ptstat = newptstat;
478 				newptstat = NULL;
479 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
480 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
481 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
482 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
483 			}
484 		}
485 	}
486 
487 #if NSYSTRACE > 0
488 	if (newstrp)
489 		systrace_fork(curp, p, newstrp);
490 #endif
491 
492 	if (tidptr != NULL) {
493 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
494 
495 		if (copyout(&pid, tidptr, sizeof(pid)))
496 			psignal(curp, SIGSEGV);
497 	}
498 
499 	/*
500 	 * Make child runnable, set start time, and add to run queue.
501 	 */
502 	SCHED_LOCK(s);
503  	getmicrotime(&pr->ps_start);
504 	p->p_acflag = AFORK;
505 	p->p_stat = SRUN;
506 	p->p_cpu = sched_choosecpu_fork(curp, flags);
507 	setrunqueue(p);
508 	SCHED_UNLOCK(s);
509 
510 	if (newptstat)
511 		free(newptstat, M_SUBPROC);
512 
513 	/*
514 	 * Notify any interested parties about the new process.
515 	 */
516 	if ((flags & FORK_THREAD) == 0)
517 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
518 
519 	/*
520 	 * Update stats now that we know the fork was successful.
521 	 */
522 	uvmexp.forks++;
523 	if (flags & FORK_PPWAIT)
524 		uvmexp.forks_ppwait++;
525 	if (flags & FORK_SHAREVM)
526 		uvmexp.forks_sharevm++;
527 
528 	/*
529 	 * Pass a pointer to the new process to the caller.
530 	 */
531 	if (rnewprocp != NULL)
532 		*rnewprocp = p;
533 
534 	/*
535 	 * Preserve synchronization semantics of vfork.  If waiting for
536 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
537 	 * on ourselves, and sleep on our process for the latter flag
538 	 * to go away.
539 	 * XXX Need to stop other rthreads in the parent
540 	 */
541 	if (flags & FORK_PPWAIT)
542 		while (curpr->ps_flags & PS_ISPWAIT)
543 			tsleep(curpr, PWAIT, "ppwait", 0);
544 
545 	/*
546 	 * If we're tracing the child, alert the parent too.
547 	 */
548 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
549 		psignal(curp, SIGTRAP);
550 
551 	/*
552 	 * Return child pid to parent process,
553 	 * marking us as parent via retval[1].
554 	 */
555 	if (retval != NULL) {
556 		retval[0] = p->p_pid +
557 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
558 		retval[1] = 0;
559 	}
560 	return (0);
561 }
562 
563 /*
564  * Checks for current use of a pid, either as a pid or pgid.
565  */
566 int
567 pidtaken(pid_t pid)
568 {
569 	struct proc *p;
570 
571 	if (pfind(pid) != NULL)
572 		return (1);
573 	if (pgfind(pid) != NULL)
574 		return (1);
575 	LIST_FOREACH(p, &zombproc, p_list) {
576 		if (p->p_pid == pid || (p->p_p->ps_pgrp && p->p_p->ps_pgrp->pg_id == pid))
577 			return (1);
578 	}
579 	return (0);
580 }
581 
582 #if defined(MULTIPROCESSOR)
583 /*
584  * XXX This is a slight hack to get newly-formed processes to
585  * XXX acquire the kernel lock as soon as they run.
586  */
587 void
588 proc_trampoline_mp(void)
589 {
590 	struct proc *p;
591 
592 	p = curproc;
593 
594 	SCHED_ASSERT_LOCKED();
595 	__mp_unlock(&sched_lock);
596 	spl0();
597 	SCHED_ASSERT_UNLOCKED();
598 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
599 
600 	KERNEL_LOCK();
601 }
602 #endif
603