xref: /openbsd-src/sys/kern/kern_fork.c (revision 512dfaa60ec59661b8e5334d886db1901c2f9faa) 
1 /*	$OpenBSD: kern_fork.c,v 1.166 2014/05/06 11:50:14 mpi 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 <dev/rndvar.h>
57 #include <sys/pool.h>
58 #include <sys/mman.h>
59 #include <sys/ptrace.h>
60 
61 #include <sys/syscallargs.h>
62 
63 #include "systrace.h"
64 #include <dev/systrace.h>
65 
66 #include <uvm/uvm.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 struct	forkstat forkstat;
76 
77 void fork_return(void *);
78 void tfork_child_return(void *);
79 int pidtaken(pid_t);
80 
81 void process_new(struct proc *, struct process *, int);
82 
83 void
84 fork_return(void *arg)
85 {
86 	struct proc *p = (struct proc *)arg;
87 
88 	if (p->p_p->ps_flags & PS_TRACED)
89 		psignal(p, SIGTRAP);
90 
91 	child_return(p);
92 }
93 
94 /*ARGSUSED*/
95 int
96 sys_fork(struct proc *p, void *v, register_t *retval)
97 {
98 	int flags;
99 
100 	flags = FORK_FORK;
101 	if (p->p_p->ps_ptmask & PTRACE_FORK)
102 		flags |= FORK_PTRACE;
103 	return (fork1(p, flags, NULL, 0, fork_return, NULL, retval, NULL));
104 }
105 
106 /*ARGSUSED*/
107 int
108 sys_vfork(struct proc *p, void *v, register_t *retval)
109 {
110 	return (fork1(p, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
111 	    NULL, retval, NULL));
112 }
113 
114 int
115 sys___tfork(struct proc *p, void *v, register_t *retval)
116 {
117 	struct sys___tfork_args /* {
118 		syscallarg(const struct __tfork) *param;
119 		syscallarg(size_t) psize;
120 	} */ *uap = v;
121 	size_t psize = SCARG(uap, psize);
122 	struct __tfork param = { 0 };
123 	int flags;
124 	int error;
125 
126 	if (psize == 0 || psize > sizeof(param))
127 		return (EINVAL);
128 	if ((error = copyin(SCARG(uap, param), &param, psize)))
129 		return (error);
130 #ifdef KTRACE
131 	if (KTRPOINT(p, KTR_STRUCT))
132 		ktrstruct(p, "tfork", &param, sizeof(param));
133 #endif
134 
135 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
136 	    | FORK_SHAREFILES;
137 
138 	return (fork1(p, flags, param.tf_stack, param.tf_tid,
139 	    tfork_child_return, param.tf_tcb, retval, NULL));
140 }
141 
142 void
143 tfork_child_return(void *arg)
144 {
145 	struct proc *p = curproc;
146 
147 	TCB_SET(p, arg);
148 	child_return(p);
149 }
150 
151 /*
152  * Allocate and initialize a new process.
153  */
154 void
155 process_new(struct proc *p, struct process *parent, int flags)
156 {
157 	struct process *pr;
158 
159 	pr = pool_get(&process_pool, PR_WAITOK);
160 	pr->ps_mainproc = p;
161 
162 	TAILQ_INIT(&pr->ps_threads);
163 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
164 	pr->ps_pptr = parent;
165 	LIST_INIT(&pr->ps_children);
166 	pr->ps_refcnt = 1;
167 
168 	/*
169 	 * Make a process structure for the new process.
170 	 * Start by zeroing the section of proc that is zero-initialized,
171 	 * then copy the section that is copied directly from the parent.
172 	 */
173 	memset(&pr->ps_startzero, 0,
174 	    (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero);
175 	memcpy(&pr->ps_startcopy, &parent->ps_startcopy,
176 	    (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy);
177 
178 	/* post-copy fixups */
179 	pr->ps_ucred = p->p_ucred;
180 	crhold(pr->ps_ucred);
181 	KASSERT(p->p_ucred->cr_ref >= 3); /* fork thr, new thr, new process */
182 	pr->ps_limit->p_refcnt++;
183 
184 	/* bump references to the text vnode (for procfs) */
185 	pr->ps_textvp = parent->ps_textvp;
186 	if (pr->ps_textvp)
187 		vref(pr->ps_textvp);
188 
189 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
190 
191 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
192 	if (parent->ps_session->s_ttyvp != NULL &&
193 	    parent->ps_flags & PS_CONTROLT)
194 		atomic_setbits_int(&pr->ps_flags, PS_CONTROLT);
195 
196 	p->p_p = pr;
197 
198 	/*
199 	 * Create signal actions for the child process.
200 	 */
201 	if (flags & FORK_SIGHAND)
202 		pr->ps_sigacts = sigactsshare(parent);
203 	else
204 		pr->ps_sigacts = sigactsinit(parent);
205 
206 	if (parent->ps_flags & PS_PROFIL)
207 		startprofclock(pr);
208 	if ((flags & FORK_PTRACE) && (parent->ps_flags & PS_TRACED))
209 		atomic_setbits_int(&pr->ps_flags, PS_TRACED);
210 	if (flags & FORK_NOZOMBIE)
211 		atomic_setbits_int(&pr->ps_flags, PS_NOZOMBIE);
212 	if (flags & FORK_SYSTEM)
213 		atomic_setbits_int(&pr->ps_flags, PS_SYSTEM);
214 
215 	/* it's sufficiently inited to be globally visible */
216 	LIST_INSERT_HEAD(&allprocess, pr, ps_list);
217 }
218 
219 /* print the 'table full' message once per 10 seconds */
220 struct timeval fork_tfmrate = { 10, 0 };
221 
222 int
223 fork1(struct proc *curp, int flags, void *stack, pid_t *tidptr,
224     void (*func)(void *), void *arg, register_t *retval,
225     struct proc **rnewprocp)
226 {
227 	struct process *curpr = curp->p_p;
228 	struct process *pr;
229 	struct proc *p;
230 	uid_t uid;
231 	struct vmspace *vm;
232 	int count;
233 	vaddr_t uaddr;
234 	int s;
235 	struct  ptrace_state *newptstat = NULL;
236 #if NSYSTRACE > 0
237 	void *newstrp = NULL;
238 #endif
239 
240 	/* sanity check some flag combinations */
241 	if (flags & FORK_THREAD) {
242 		if ((flags & FORK_SIGHAND) == 0 || (flags & FORK_SYSTEM) != 0)
243 			return (EINVAL);
244 	}
245 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
246 		return (EINVAL);
247 
248 	/*
249 	 * Although process entries are dynamically created, we still keep
250 	 * a global limit on the maximum number we will create. We reserve
251 	 * the last 5 processes to root. The variable nprocesses is the
252 	 * current number of processes, maxprocess is the limit.  Similar
253 	 * rules for threads (struct proc): we reserve the last 5 to root;
254 	 * the variable nthreads is the current number of procs, maxthread is
255 	 * the limit.
256 	 */
257 	uid = curp->p_ucred->cr_ruid;
258 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
259 		static struct timeval lasttfm;
260 
261 		if (ratecheck(&lasttfm, &fork_tfmrate))
262 			tablefull("proc");
263 		return (EAGAIN);
264 	}
265 	nthreads++;
266 
267 	if ((flags & FORK_THREAD) == 0) {
268 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
269 		    nprocesses >= maxprocess) {
270 			static struct timeval lasttfm;
271 
272 			if (ratecheck(&lasttfm, &fork_tfmrate))
273 				tablefull("process");
274 			nthreads--;
275 			return (EAGAIN);
276 		}
277 		nprocesses++;
278 
279 		/*
280 		 * Increment the count of processes running with
281 		 * this uid.  Don't allow a nonprivileged user to
282 		 * exceed their current limit.
283 		 */
284 		count = chgproccnt(uid, 1);
285 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
286 			(void)chgproccnt(uid, -1);
287 			nprocesses--;
288 			nthreads--;
289 			return (EAGAIN);
290 		}
291 	}
292 
293 	uaddr = uvm_uarea_alloc();
294 	if (uaddr == 0) {
295 		if ((flags & FORK_THREAD) == 0) {
296 			(void)chgproccnt(uid, -1);
297 			nprocesses--;
298 		}
299 		nthreads--;
300 		return (ENOMEM);
301 	}
302 
303 	/*
304 	 * From now on, we're committed to the fork and cannot fail.
305 	 */
306 
307 	/* Allocate new proc. */
308 	p = pool_get(&proc_pool, PR_WAITOK);
309 
310 	p->p_stat = SIDL;			/* protect against others */
311 	p->p_flag = 0;
312 
313 	/*
314 	 * Make a proc table entry for the new process.
315 	 * Start by zeroing the section of proc that is zero-initialized,
316 	 * then copy the section that is copied directly from the parent.
317 	 */
318 	memset(&p->p_startzero, 0,
319 	    (caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero);
320 	memcpy(&p->p_startcopy, &curp->p_startcopy,
321 	    (caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy);
322 	crhold(p->p_ucred);
323 
324 	/*
325 	 * Initialize the timeouts.
326 	 */
327 	timeout_set(&p->p_sleep_to, endtsleep, p);
328 
329 	if (flags & FORK_THREAD) {
330 		atomic_setbits_int(&p->p_flag, P_THREAD);
331 		p->p_p = pr = curpr;
332 		pr->ps_refcnt++;
333 	} else {
334 		process_new(p, curpr, flags);
335 		pr = p->p_p;
336 	}
337 	if (pr->ps_flags & PS_SYSTEM)
338 		atomic_setbits_int(&p->p_flag, P_SYSTEM);
339 
340 	/*
341 	 * Duplicate sub-structures as needed.
342 	 * Increase reference counts on shared objects.
343 	 */
344 	if (flags & FORK_SHAREFILES)
345 		p->p_fd = fdshare(curp);
346 	else
347 		p->p_fd = fdcopy(curp);
348 
349 	if (flags & FORK_PPWAIT) {
350 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
351 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
352 	}
353 
354 #ifdef KTRACE
355 	/*
356 	 * Copy traceflag and tracefile if enabled.
357 	 * If not inherited, these were zeroed above.
358 	 */
359 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
360 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
361 		    curpr->ps_tracecred);
362 #endif
363 
364 	/*
365 	 * set priority of child to be that of parent
366 	 * XXX should move p_estcpu into the region of struct proc which gets
367 	 * copied.
368 	 */
369 	scheduler_fork_hook(curp, p);
370 
371 	if (flags & FORK_THREAD)
372 		sigstkinit(&p->p_sigstk);
373 
374 	/*
375 	 * If emulation has thread fork hook, call it now.
376 	 */
377 	if (pr->ps_emul->e_proc_fork)
378 		(*pr->ps_emul->e_proc_fork)(p, curp);
379 
380 	p->p_addr = (struct user *)uaddr;
381 
382 	/*
383 	 * Finish creating the child process.  It will return through a
384 	 * different path later.
385 	 */
386 	uvm_fork(curp, p, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
387 	    0, func ? func : child_return, arg ? arg : p);
388 
389 	vm = p->p_vmspace;
390 
391 	if (flags & FORK_FORK) {
392 		forkstat.cntfork++;
393 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
394 	} else if (flags & FORK_VFORK) {
395 		forkstat.cntvfork++;
396 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
397 	} else if (flags & FORK_TFORK) {
398 		forkstat.cnttfork++;
399 	} else {
400 		forkstat.cntkthread++;
401 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
402 	}
403 
404 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
405 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
406 #if NSYSTRACE > 0
407 	if (ISSET(curp->p_flag, P_SYSTRACE))
408 		newstrp = systrace_getproc();
409 #endif
410 
411 	p->p_pid = allocpid();
412 
413 	LIST_INSERT_HEAD(&allproc, p, p_list);
414 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
415 	if ((flags & FORK_THREAD) == 0) {
416 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
417 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
418 
419 		if (pr->ps_flags & PS_TRACED) {
420 			pr->ps_oppid = curpr->ps_pid;
421 			if (pr->ps_pptr != curpr->ps_pptr)
422 				proc_reparent(pr, curpr->ps_pptr);
423 
424 			/*
425 			 * Set ptrace status.
426 			 */
427 			if (flags & FORK_FORK) {
428 				pr->ps_ptstat = newptstat;
429 				newptstat = NULL;
430 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
431 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
432 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
433 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
434 			}
435 		}
436 	} else {
437 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
438 		/*
439 		 * if somebody else wants to take us to single threaded mode,
440 		 * count ourselves in.
441 		 */
442 		if (pr->ps_single) {
443 			curpr->ps_singlecount++;
444 			atomic_setbits_int(&p->p_flag, P_SUSPSINGLE);
445 		}
446 	}
447 
448 #if NSYSTRACE > 0
449 	if (newstrp)
450 		systrace_fork(curp, p, newstrp);
451 #endif
452 
453 	if (tidptr != NULL) {
454 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
455 
456 		if (copyout(&pid, tidptr, sizeof(pid)))
457 			psignal(curp, SIGSEGV);
458 	}
459 
460 	/*
461 	 * For new processes, set accounting bits
462 	 */
463 	if ((flags & FORK_THREAD) == 0) {
464 		getnanotime(&pr->ps_start);
465 		pr->ps_acflag = AFORK;
466 	}
467 
468 	/*
469 	 * Make child runnable and add to run queue.
470 	 */
471 	if ((flags & FORK_IDLE) == 0) {
472 		SCHED_LOCK(s);
473 		p->p_stat = SRUN;
474 		p->p_cpu = sched_choosecpu_fork(curp, flags);
475 		setrunqueue(p);
476 		SCHED_UNLOCK(s);
477 	} else
478 		p->p_cpu = arg;
479 
480 	if (newptstat)
481 		free(newptstat, M_SUBPROC);
482 
483 	/*
484 	 * Notify any interested parties about the new process.
485 	 */
486 	if ((flags & FORK_THREAD) == 0)
487 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
488 
489 	/*
490 	 * Update stats now that we know the fork was successful.
491 	 */
492 	uvmexp.forks++;
493 	if (flags & FORK_PPWAIT)
494 		uvmexp.forks_ppwait++;
495 	if (flags & FORK_SHAREVM)
496 		uvmexp.forks_sharevm++;
497 
498 	/*
499 	 * Pass a pointer to the new process to the caller.
500 	 */
501 	if (rnewprocp != NULL)
502 		*rnewprocp = p;
503 
504 	/*
505 	 * Preserve synchronization semantics of vfork.  If waiting for
506 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
507 	 * on ourselves, and sleep on our process for the latter flag
508 	 * to go away.
509 	 * XXX Need to stop other rthreads in the parent
510 	 */
511 	if (flags & FORK_PPWAIT)
512 		while (curpr->ps_flags & PS_ISPWAIT)
513 			tsleep(curpr, PWAIT, "ppwait", 0);
514 
515 	/*
516 	 * If we're tracing the child, alert the parent too.
517 	 */
518 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
519 		psignal(curp, SIGTRAP);
520 
521 	/*
522 	 * Return child pid to parent process,
523 	 * marking us as parent via retval[1].
524 	 */
525 	if (retval != NULL) {
526 		retval[0] = p->p_pid +
527 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
528 		retval[1] = 0;
529 	}
530 	return (0);
531 }
532 
533 /*
534  * Checks for current use of a pid, either as a pid or pgid.
535  */
536 pid_t oldpids[100];
537 int
538 ispidtaken(pid_t pid)
539 {
540 	uint32_t i;
541 	struct process *pr;
542 
543 	for (i = 0; i < nitems(oldpids); i++)
544 		if (pid == oldpids[i])
545 			return (1);
546 
547 	if (pfind(pid) != NULL)
548 		return (1);
549 	if (pgfind(pid) != NULL)
550 		return (1);
551 	LIST_FOREACH(pr, &zombprocess, ps_list) {
552 		if (pr->ps_pid == pid ||
553 		    (pr->ps_pgrp && pr->ps_pgrp->pg_id == pid))
554 			return (1);
555 	}
556 	return (0);
557 }
558 
559 /* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
560 pid_t
561 allocpid(void)
562 {
563 	static pid_t lastpid;
564 	pid_t pid;
565 
566 	if (!randompid) {
567 		/* only used early on for system processes */
568 		pid = ++lastpid;
569 	} else {
570 		do {
571 			pid = 1 + arc4random_uniform(PID_MAX);
572 		} while (ispidtaken(pid));
573 	}
574 
575 	return pid;
576 }
577 
578 void
579 freepid(pid_t pid)
580 {
581 	static uint32_t idx;
582 
583 	oldpids[idx++ % nitems(oldpids)] = pid;
584 }
585 
586 #if defined(MULTIPROCESSOR)
587 /*
588  * XXX This is a slight hack to get newly-formed processes to
589  * XXX acquire the kernel lock as soon as they run.
590  */
591 void
592 proc_trampoline_mp(void)
593 {
594 	struct proc *p;
595 
596 	p = curproc;
597 
598 	SCHED_ASSERT_LOCKED();
599 	__mp_unlock(&sched_lock);
600 	spl0();
601 	SCHED_ASSERT_UNLOCKED();
602 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
603 
604 	KERNEL_LOCK();
605 }
606 #endif
607