xref: /openbsd-src/sys/kern/kern_fork.c (revision e6c6495d62998a12dd778aed8d40dc250167a61c) 
1 /*	$OpenBSD: kern_fork.c,v 1.170 2014/07/08 17:19:25 deraadt 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/vmmeter.h>
52 #include <sys/file.h>
53 #include <sys/acct.h>
54 #include <sys/ktrace.h>
55 #include <sys/sched.h>
56 #include <sys/sysctl.h>
57 #include <dev/rndvar.h>
58 #include <sys/pool.h>
59 #include <sys/mman.h>
60 #include <sys/ptrace.h>
61 
62 #include <sys/syscallargs.h>
63 
64 #include "systrace.h"
65 #include <dev/systrace.h>
66 
67 #include <uvm/uvm.h>
68 
69 #ifdef __HAVE_MD_TCB
70 # include <machine/tcb.h>
71 #endif
72 
73 int	nprocesses = 1;		/* process 0 */
74 int	nthreads = 1;		/* proc 0 */
75 int	randompid;		/* when set to 1, pid's go random */
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 *, int);
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, flags, NULL, 0, fork_return, NULL, retval, NULL));
105 }
106 
107 /*ARGSUSED*/
108 int
109 sys_vfork(struct proc *p, void *v, register_t *retval)
110 {
111 	return (fork1(p, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
112 	    NULL, retval, NULL));
113 }
114 
115 int
116 sys___tfork(struct proc *p, void *v, register_t *retval)
117 {
118 	struct sys___tfork_args /* {
119 		syscallarg(const struct __tfork) *param;
120 		syscallarg(size_t) psize;
121 	} */ *uap = v;
122 	size_t psize = SCARG(uap, psize);
123 	struct __tfork param = { 0 };
124 	int flags;
125 	int error;
126 
127 	if (psize == 0 || psize > sizeof(param))
128 		return (EINVAL);
129 	if ((error = copyin(SCARG(uap, param), &param, psize)))
130 		return (error);
131 #ifdef KTRACE
132 	if (KTRPOINT(p, KTR_STRUCT))
133 		ktrstruct(p, "tfork", &param, sizeof(param));
134 #endif
135 
136 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
137 	    | FORK_SHAREFILES;
138 
139 	return (fork1(p, flags, param.tf_stack, param.tf_tid,
140 	    tfork_child_return, param.tf_tcb, retval, NULL));
141 }
142 
143 void
144 tfork_child_return(void *arg)
145 {
146 	struct proc *p = curproc;
147 
148 	TCB_SET(p, arg);
149 	child_return(p);
150 }
151 
152 /*
153  * Allocate and initialize a new process.
154  */
155 void
156 process_new(struct proc *p, struct process *parent, int flags)
157 {
158 	struct process *pr;
159 
160 	pr = pool_get(&process_pool, PR_WAITOK);
161 	pr->ps_mainproc = p;
162 
163 	TAILQ_INIT(&pr->ps_threads);
164 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
165 	pr->ps_pptr = parent;
166 	LIST_INIT(&pr->ps_children);
167 	pr->ps_refcnt = 1;
168 
169 	/*
170 	 * Make a process structure for the new process.
171 	 * Start by zeroing the section of proc that is zero-initialized,
172 	 * then copy the section that is copied directly from the parent.
173 	 */
174 	memset(&pr->ps_startzero, 0,
175 	    (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero);
176 	memcpy(&pr->ps_startcopy, &parent->ps_startcopy,
177 	    (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy);
178 
179 	/* post-copy fixups */
180 	pr->ps_ucred = p->p_ucred;
181 	crhold(pr->ps_ucred);
182 	KASSERT(p->p_ucred->cr_ref >= 3); /* fork thr, new thr, new process */
183 	pr->ps_limit->p_refcnt++;
184 
185 	/* bump references to the text vnode (for procfs) */
186 	pr->ps_textvp = parent->ps_textvp;
187 	if (pr->ps_textvp)
188 		vref(pr->ps_textvp);
189 
190 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
191 
192 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
193 	if (parent->ps_session->s_ttyvp != NULL)
194 		pr->ps_flags |= parent->ps_flags & PS_CONTROLT;
195 
196 	p->p_p = pr;
197 
198 	/*
199 	 * Duplicate sub-structures as needed.
200 	 * Increase reference counts on shared objects.
201 	 */
202 	if (flags & FORK_SHAREFILES)
203 		pr->ps_fd = fdshare(parent);
204 	else
205 		pr->ps_fd = fdcopy(parent);
206 	if (flags & FORK_SIGHAND)
207 		pr->ps_sigacts = sigactsshare(parent);
208 	else
209 		pr->ps_sigacts = sigactsinit(parent);
210 	if (flags & FORK_SHAREVM)
211 		pr->ps_vmspace = uvmspace_share(parent);
212 	else
213 		pr->ps_vmspace = uvmspace_fork(parent);
214 
215 	if (parent->ps_flags & PS_PROFIL)
216 		startprofclock(pr);
217 	if (flags & FORK_PTRACE)
218 		pr->ps_flags |= parent->ps_flags & PS_TRACED;
219 	if (flags & FORK_NOZOMBIE)
220 		pr->ps_flags |= PS_NOZOMBIE;
221 	if (flags & FORK_SYSTEM)
222 		pr->ps_flags |= PS_SYSTEM;
223 
224 	/*
225 	 * Mark as embryo to protect against others.
226 	 * Done with atomic_* to force visibility of all of the above flags
227 	 */
228 	atomic_setbits_int(&pr->ps_flags, PS_EMBRYO);
229 
230 	/* it's sufficiently inited to be globally visible */
231 	LIST_INSERT_HEAD(&allprocess, pr, ps_list);
232 }
233 
234 /* print the 'table full' message once per 10 seconds */
235 struct timeval fork_tfmrate = { 10, 0 };
236 
237 int
238 fork1(struct proc *curp, int flags, void *stack, pid_t *tidptr,
239     void (*func)(void *), void *arg, register_t *retval,
240     struct proc **rnewprocp)
241 {
242 	struct process *curpr = curp->p_p;
243 	struct process *pr;
244 	struct proc *p;
245 	uid_t uid;
246 	struct vmspace *vm;
247 	int count;
248 	vaddr_t uaddr;
249 	int s;
250 	struct  ptrace_state *newptstat = NULL;
251 #if NSYSTRACE > 0
252 	void *newstrp = NULL;
253 #endif
254 
255 	/* sanity check some flag combinations */
256 	if (flags & FORK_THREAD) {
257 		if ((flags & FORK_SHAREFILES) == 0 ||
258 		    (flags & FORK_SIGHAND) == 0 ||
259 		    (flags & FORK_SYSTEM) != 0)
260 			return (EINVAL);
261 	}
262 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
263 		return (EINVAL);
264 
265 	/*
266 	 * Although process entries are dynamically created, we still keep
267 	 * a global limit on the maximum number we will create. We reserve
268 	 * the last 5 processes to root. The variable nprocesses is the
269 	 * current number of processes, maxprocess is the limit.  Similar
270 	 * rules for threads (struct proc): we reserve the last 5 to root;
271 	 * the variable nthreads is the current number of procs, maxthread is
272 	 * the limit.
273 	 */
274 	uid = curp->p_ucred->cr_ruid;
275 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
276 		static struct timeval lasttfm;
277 
278 		if (ratecheck(&lasttfm, &fork_tfmrate))
279 			tablefull("proc");
280 		return (EAGAIN);
281 	}
282 	nthreads++;
283 
284 	if ((flags & FORK_THREAD) == 0) {
285 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
286 		    nprocesses >= maxprocess) {
287 			static struct timeval lasttfm;
288 
289 			if (ratecheck(&lasttfm, &fork_tfmrate))
290 				tablefull("process");
291 			nthreads--;
292 			return (EAGAIN);
293 		}
294 		nprocesses++;
295 
296 		/*
297 		 * Increment the count of processes running with
298 		 * this uid.  Don't allow a nonprivileged user to
299 		 * exceed their current limit.
300 		 */
301 		count = chgproccnt(uid, 1);
302 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
303 			(void)chgproccnt(uid, -1);
304 			nprocesses--;
305 			nthreads--;
306 			return (EAGAIN);
307 		}
308 	}
309 
310 	uaddr = uvm_uarea_alloc();
311 	if (uaddr == 0) {
312 		if ((flags & FORK_THREAD) == 0) {
313 			(void)chgproccnt(uid, -1);
314 			nprocesses--;
315 		}
316 		nthreads--;
317 		return (ENOMEM);
318 	}
319 
320 	/*
321 	 * From now on, we're committed to the fork and cannot fail.
322 	 */
323 
324 	/* Allocate new proc. */
325 	p = pool_get(&proc_pool, PR_WAITOK);
326 
327 	p->p_stat = SIDL;			/* protect against others */
328 	p->p_flag = 0;
329 
330 	/*
331 	 * Make a proc table entry for the new process.
332 	 * Start by zeroing the section of proc that is zero-initialized,
333 	 * then copy the section that is copied directly from the parent.
334 	 */
335 	memset(&p->p_startzero, 0,
336 	    (caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero);
337 	memcpy(&p->p_startcopy, &curp->p_startcopy,
338 	    (caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy);
339 	crhold(p->p_ucred);
340 
341 	/*
342 	 * Initialize the timeouts.
343 	 */
344 	timeout_set(&p->p_sleep_to, endtsleep, p);
345 
346 	if (flags & FORK_THREAD) {
347 		atomic_setbits_int(&p->p_flag, P_THREAD);
348 		p->p_p = pr = curpr;
349 		pr->ps_refcnt++;
350 	} else {
351 		process_new(p, curpr, flags);
352 		pr = p->p_p;
353 	}
354 	p->p_fd		= pr->ps_fd;
355 	p->p_vmspace	= pr->ps_vmspace;
356 	if (pr->ps_flags & PS_SYSTEM)
357 		atomic_setbits_int(&p->p_flag, P_SYSTEM);
358 
359 	if (flags & FORK_PPWAIT) {
360 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
361 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
362 	}
363 
364 #ifdef KTRACE
365 	/*
366 	 * Copy traceflag and tracefile if enabled.
367 	 * If not inherited, these were zeroed above.
368 	 */
369 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
370 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
371 		    curpr->ps_tracecred);
372 #endif
373 
374 	/*
375 	 * set priority of child to be that of parent
376 	 * XXX should move p_estcpu into the region of struct proc which gets
377 	 * copied.
378 	 */
379 	scheduler_fork_hook(curp, p);
380 
381 	if (flags & FORK_THREAD)
382 		sigstkinit(&p->p_sigstk);
383 
384 	/*
385 	 * If emulation has thread fork hook, call it now.
386 	 */
387 	if (pr->ps_emul->e_proc_fork)
388 		(*pr->ps_emul->e_proc_fork)(p, curp);
389 
390 	p->p_addr = (struct user *)uaddr;
391 
392 	/*
393 	 * Finish creating the child thread.  cpu_fork() will copy
394 	 * and update the pcb and make the child ready to run.  If
395 	 * this is a normal user fork, the child will exit directly
396 	 * to user mode via child_return() on its first time slice
397 	 * and will not return here.  If this is a kernel thread,
398 	 * the specified entry point will be executed.
399 	 */
400 	cpu_fork(curp, p, stack, 0, func ? func : child_return, arg ? arg : p);
401 
402 	vm = pr->ps_vmspace;
403 
404 	if (flags & FORK_FORK) {
405 		forkstat.cntfork++;
406 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
407 	} else if (flags & FORK_VFORK) {
408 		forkstat.cntvfork++;
409 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
410 	} else if (flags & FORK_TFORK) {
411 		forkstat.cnttfork++;
412 	} else {
413 		forkstat.cntkthread++;
414 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
415 	}
416 
417 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
418 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
419 #if NSYSTRACE > 0
420 	if (ISSET(curp->p_flag, P_SYSTRACE))
421 		newstrp = systrace_getproc();
422 #endif
423 
424 	p->p_pid = allocpid();
425 
426 	LIST_INSERT_HEAD(&allproc, p, p_list);
427 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
428 	if ((flags & FORK_THREAD) == 0) {
429 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
430 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
431 
432 		if (pr->ps_flags & PS_TRACED) {
433 			pr->ps_oppid = curpr->ps_pid;
434 			if (pr->ps_pptr != curpr->ps_pptr)
435 				proc_reparent(pr, curpr->ps_pptr);
436 
437 			/*
438 			 * Set ptrace status.
439 			 */
440 			if (flags & FORK_FORK) {
441 				pr->ps_ptstat = newptstat;
442 				newptstat = NULL;
443 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
444 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
445 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
446 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
447 			}
448 		}
449 	} else {
450 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
451 		/*
452 		 * if somebody else wants to take us to single threaded mode,
453 		 * count ourselves in.
454 		 */
455 		if (pr->ps_single) {
456 			curpr->ps_singlecount++;
457 			atomic_setbits_int(&p->p_flag, P_SUSPSINGLE);
458 		}
459 	}
460 
461 #if NSYSTRACE > 0
462 	if (newstrp)
463 		systrace_fork(curp, p, newstrp);
464 #endif
465 
466 	if (tidptr != NULL) {
467 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
468 
469 		if (copyout(&pid, tidptr, sizeof(pid)))
470 			psignal(curp, SIGSEGV);
471 	}
472 
473 	/*
474 	 * For new processes, set accounting bits and mark as complete.
475 	 */
476 	if ((flags & FORK_THREAD) == 0) {
477 		getnanotime(&pr->ps_start);
478 		pr->ps_acflag = AFORK;
479 		atomic_clearbits_int(&pr->ps_flags, PS_EMBRYO);
480 	}
481 
482 	/*
483 	 * Make child runnable and add to run queue.
484 	 */
485 	if ((flags & FORK_IDLE) == 0) {
486 		SCHED_LOCK(s);
487 		p->p_stat = SRUN;
488 		p->p_cpu = sched_choosecpu_fork(curp, flags);
489 		setrunqueue(p);
490 		SCHED_UNLOCK(s);
491 	} else
492 		p->p_cpu = arg;
493 
494 	if (newptstat)
495 		free(newptstat, M_SUBPROC);
496 
497 	/*
498 	 * Notify any interested parties about the new process.
499 	 */
500 	if ((flags & FORK_THREAD) == 0)
501 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
502 
503 	/*
504 	 * Update stats now that we know the fork was successful.
505 	 */
506 	uvmexp.forks++;
507 	if (flags & FORK_PPWAIT)
508 		uvmexp.forks_ppwait++;
509 	if (flags & FORK_SHAREVM)
510 		uvmexp.forks_sharevm++;
511 
512 	/*
513 	 * Pass a pointer to the new process to the caller.
514 	 */
515 	if (rnewprocp != NULL)
516 		*rnewprocp = p;
517 
518 	/*
519 	 * Preserve synchronization semantics of vfork.  If waiting for
520 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
521 	 * on ourselves, and sleep on our process for the latter flag
522 	 * to go away.
523 	 * XXX Need to stop other rthreads in the parent
524 	 */
525 	if (flags & FORK_PPWAIT)
526 		while (curpr->ps_flags & PS_ISPWAIT)
527 			tsleep(curpr, PWAIT, "ppwait", 0);
528 
529 	/*
530 	 * If we're tracing the child, alert the parent too.
531 	 */
532 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
533 		psignal(curp, SIGTRAP);
534 
535 	/*
536 	 * Return child pid to parent process,
537 	 * marking us as parent via retval[1].
538 	 */
539 	if (retval != NULL) {
540 		retval[0] = p->p_pid +
541 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
542 		retval[1] = 0;
543 	}
544 	return (0);
545 }
546 
547 /*
548  * Checks for current use of a pid, either as a pid or pgid.
549  */
550 pid_t oldpids[100];
551 int
552 ispidtaken(pid_t pid)
553 {
554 	uint32_t i;
555 	struct process *pr;
556 
557 	for (i = 0; i < nitems(oldpids); i++)
558 		if (pid == oldpids[i])
559 			return (1);
560 
561 	if (pfind(pid) != NULL)
562 		return (1);
563 	if (pgfind(pid) != NULL)
564 		return (1);
565 	LIST_FOREACH(pr, &zombprocess, ps_list) {
566 		if (pr->ps_pid == pid ||
567 		    (pr->ps_pgrp && pr->ps_pgrp->pg_id == pid))
568 			return (1);
569 	}
570 	return (0);
571 }
572 
573 /* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
574 pid_t
575 allocpid(void)
576 {
577 	static pid_t lastpid;
578 	pid_t pid;
579 
580 	if (!randompid) {
581 		/* only used early on for system processes */
582 		pid = ++lastpid;
583 	} else {
584 		do {
585 			pid = 1 + arc4random_uniform(PID_MAX);
586 		} while (ispidtaken(pid));
587 	}
588 
589 	return pid;
590 }
591 
592 void
593 freepid(pid_t pid)
594 {
595 	static uint32_t idx;
596 
597 	oldpids[idx++ % nitems(oldpids)] = pid;
598 }
599 
600 #if defined(MULTIPROCESSOR)
601 /*
602  * XXX This is a slight hack to get newly-formed processes to
603  * XXX acquire the kernel lock as soon as they run.
604  */
605 void
606 proc_trampoline_mp(void)
607 {
608 	struct proc *p;
609 
610 	p = curproc;
611 
612 	SCHED_ASSERT_LOCKED();
613 	__mp_unlock(&sched_lock);
614 	spl0();
615 	SCHED_ASSERT_UNLOCKED();
616 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
617 
618 	KERNEL_LOCK();
619 }
620 #endif
621