xref: /openbsd-src/sys/kern/kern_fork.c (revision e88d58b26ba3c1eb3d963ca7eef30960d3ac48c2) 
1 /*	$OpenBSD: kern_fork.c,v 1.117 2010/06/30 00:40:28 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 int	nprocs = 1;		/* process 0 */
69 int	randompid;		/* when set to 1, pid's go random */
70 pid_t	lastpid;
71 struct	forkstat forkstat;
72 
73 void fork_return(void *);
74 int pidtaken(pid_t);
75 
76 void process_new(struct proc *, struct proc *);
77 
78 void
79 fork_return(void *arg)
80 {
81 	struct proc *p = (struct proc *)arg;
82 
83 	if (p->p_flag & P_TRACED)
84 		psignal(p, SIGTRAP);
85 
86 	child_return(p);
87 }
88 
89 /*ARGSUSED*/
90 int
91 sys_fork(struct proc *p, void *v, register_t *retval)
92 {
93 	int flags;
94 
95 	flags = FORK_FORK;
96 	if (p->p_ptmask & PTRACE_FORK)
97 		flags |= FORK_PTRACE;
98 	return (fork1(p, SIGCHLD, flags, NULL, 0,
99 	    fork_return, NULL, retval, NULL));
100 }
101 
102 /*ARGSUSED*/
103 int
104 sys_vfork(struct proc *p, void *v, register_t *retval)
105 {
106 	return (fork1(p, SIGCHLD, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
107 	    NULL, retval, NULL));
108 }
109 
110 int
111 sys_rfork(struct proc *p, void *v, register_t *retval)
112 {
113 	struct sys_rfork_args /* {
114 		syscallarg(int) flags;
115 	} */ *uap = v;
116 
117 	int rforkflags;
118 	int flags;
119 
120 	flags = FORK_RFORK;
121 	rforkflags = SCARG(uap, flags);
122 
123 	if ((rforkflags & RFPROC) == 0)
124 		return (EINVAL);
125 
126 	switch(rforkflags & (RFFDG|RFCFDG)) {
127 	case (RFFDG|RFCFDG):
128 		return EINVAL;
129 	case RFCFDG:
130 		flags |= FORK_CLEANFILES;
131 		break;
132 	case RFFDG:
133 		break;
134 	default:
135 		flags |= FORK_SHAREFILES;
136 		break;
137 	}
138 
139 	if (rforkflags & RFNOWAIT)
140 		flags |= FORK_NOZOMBIE;
141 
142 	if (rforkflags & RFMEM)
143 		flags |= FORK_SHAREVM;
144 
145 	if (rforkflags & RFTHREAD)
146 		flags |= FORK_THREAD | FORK_SIGHAND | FORK_NOZOMBIE;
147 
148 	return (fork1(p, SIGCHLD, flags, NULL, 0, NULL, NULL, retval, NULL));
149 }
150 
151 /*
152  * Allocate and initialize a new process.
153  */
154 void
155 process_new(struct proc *newproc, struct proc *parentproc)
156 {
157 	struct process *pr, *parent;
158 
159 	pr = pool_get(&process_pool, PR_WAITOK);
160 	pr->ps_mainproc = newproc;
161 	TAILQ_INIT(&pr->ps_threads);
162 	TAILQ_INSERT_TAIL(&pr->ps_threads, newproc, p_thr_link);
163 	pr->ps_refcnt = 1;
164 
165 	parent = parentproc->p_p;
166 
167 	/*
168 	 * Make a process structure for the new process.
169 	 * Start by zeroing the section of proc that is zero-initialized,
170 	 * then copy the section that is copied directly from the parent.
171 	 */
172 	bzero(&pr->ps_startzero,
173 	    (unsigned) ((caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero));
174 	bcopy(&parent->ps_startcopy, &pr->ps_startcopy,
175 	    (unsigned) ((caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy));
176 
177 	/* post-copy fixups */
178 	pr->ps_cred = pool_get(&pcred_pool, PR_WAITOK);
179 	bcopy(parent->ps_cred, pr->ps_cred, sizeof(*pr->ps_cred));
180 	crhold(parent->ps_cred->pc_ucred);
181 	pr->ps_limit->p_refcnt++;
182 
183 	newproc->p_p = pr;
184 }
185 
186 /* print the 'table full' message once per 10 seconds */
187 struct timeval fork_tfmrate = { 10, 0 };
188 
189 int
190 fork1(struct proc *p1, int exitsig, int flags, void *stack, size_t stacksize,
191     void (*func)(void *), void *arg, register_t *retval,
192     struct proc **rnewprocp)
193 {
194 	struct proc *p2;
195 	uid_t uid;
196 	struct vmspace *vm;
197 	int count;
198 	vaddr_t uaddr;
199 	int s;
200 	extern void endtsleep(void *);
201 	extern void realitexpire(void *);
202 	struct  ptrace_state *newptstat = NULL;
203 #if NSYSTRACE > 0
204 	void *newstrp = NULL;
205 #endif
206 
207 	/* sanity check some flag combinations */
208 	if (flags & FORK_THREAD)
209 	{
210 		if (!rthreads_enabled)
211 			return (ENOTSUP);
212 		if ((flags & (FORK_SIGHAND | FORK_NOZOMBIE)) !=
213 		    (FORK_SIGHAND | FORK_NOZOMBIE))
214 			return (EINVAL);
215 	}
216 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
217 		return (EINVAL);
218 
219 	/*
220 	 * Although process entries are dynamically created, we still keep
221 	 * a global limit on the maximum number we will create. We reserve
222 	 * the last 5 processes to root. The variable nprocs is the current
223 	 * number of processes, maxproc is the limit.
224 	 */
225 	uid = p1->p_cred->p_ruid;
226 	if ((nprocs >= maxproc - 5 && uid != 0) || nprocs >= maxproc) {
227 		static struct timeval lasttfm;
228 
229 		if (ratecheck(&lasttfm, &fork_tfmrate))
230 			tablefull("proc");
231 		return (EAGAIN);
232 	}
233 	nprocs++;
234 
235 	/*
236 	 * Increment the count of procs running with this uid. Don't allow
237 	 * a nonprivileged user to exceed their current limit.
238 	 */
239 	count = chgproccnt(uid, 1);
240 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
241 		(void)chgproccnt(uid, -1);
242 		nprocs--;
243 		return (EAGAIN);
244 	}
245 
246 	uaddr = uvm_km_alloc1(kernel_map, USPACE, USPACE_ALIGN, 1);
247 	if (uaddr == 0) {
248 		chgproccnt(uid, -1);
249 		nprocs--;
250 		return (ENOMEM);
251 	}
252 
253 	/*
254 	 * From now on, we're committed to the fork and cannot fail.
255 	 */
256 
257 	/* Allocate new proc. */
258 	p2 = pool_get(&proc_pool, PR_WAITOK);
259 
260 	p2->p_stat = SIDL;			/* protect against others */
261 	p2->p_exitsig = exitsig;
262 	p2->p_flag = 0;
263 
264 	if (flags & FORK_THREAD) {
265 		atomic_setbits_int(&p2->p_flag, P_THREAD);
266 		p2->p_p = p1->p_p;
267 		TAILQ_INSERT_TAIL(&p2->p_p->ps_threads, p2, p_thr_link);
268 		p2->p_p->ps_refcnt++;
269 	} else {
270 		process_new(p2, p1);
271 	}
272 
273 	/*
274 	 * Make a proc table entry for the new process.
275 	 * Start by zeroing the section of proc that is zero-initialized,
276 	 * then copy the section that is copied directly from the parent.
277 	 */
278 	bzero(&p2->p_startzero,
279 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
280 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
281 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
282 
283 	/*
284 	 * Initialize the timeouts.
285 	 */
286 	timeout_set(&p2->p_sleep_to, endtsleep, p2);
287 	timeout_set(&p2->p_realit_to, realitexpire, p2);
288 
289 	/*
290 	 * Duplicate sub-structures as needed.
291 	 * Increase reference counts on shared objects.
292 	 * The p_stats and p_sigacts substructs are set in vm_fork.
293 	 */
294 	if (p1->p_flag & P_PROFIL)
295 		startprofclock(p2);
296 	atomic_setbits_int(&p2->p_flag, p1->p_flag & (P_SUGID | P_SUGIDEXEC));
297 	if (flags & FORK_PTRACE)
298 		atomic_setbits_int(&p2->p_flag, p1->p_flag & P_TRACED);
299 
300 	/* bump references to the text vnode (for procfs) */
301 	p2->p_textvp = p1->p_textvp;
302 	if (p2->p_textvp)
303 		vref(p2->p_textvp);
304 
305 	if (flags & FORK_CLEANFILES)
306 		p2->p_fd = fdinit(p1);
307 	else if (flags & FORK_SHAREFILES)
308 		p2->p_fd = fdshare(p1);
309 	else
310 		p2->p_fd = fdcopy(p1);
311 
312 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
313 		atomic_setbits_int(&p2->p_flag, P_CONTROLT);
314 	if (flags & FORK_PPWAIT)
315 		atomic_setbits_int(&p2->p_flag, P_PPWAIT);
316 	p2->p_pptr = p1;
317 	if (flags & FORK_NOZOMBIE)
318 		atomic_setbits_int(&p2->p_flag, P_NOZOMBIE);
319 	LIST_INIT(&p2->p_children);
320 
321 #ifdef KTRACE
322 	/*
323 	 * Copy traceflag and tracefile if enabled.
324 	 * If not inherited, these were zeroed above.
325 	 */
326 	if (p1->p_traceflag & KTRFAC_INHERIT) {
327 		p2->p_traceflag = p1->p_traceflag;
328 		if ((p2->p_tracep = p1->p_tracep) != NULL)
329 			vref(p2->p_tracep);
330 	}
331 #endif
332 
333 	/*
334 	 * set priority of child to be that of parent
335 	 * XXX should move p_estcpu into the region of struct proc which gets
336 	 * copied.
337 	 */
338 	scheduler_fork_hook(p1, p2);
339 
340 	/*
341 	 * Create signal actions for the child process.
342 	 */
343 	if (flags & FORK_SIGHAND)
344 		sigactsshare(p1, p2);
345 	else
346 		p2->p_sigacts = sigactsinit(p1);
347 
348 	/*
349 	 * If emulation has process fork hook, call it now.
350 	 */
351 	if (p2->p_emul->e_proc_fork)
352 		(*p2->p_emul->e_proc_fork)(p2, p1);
353 
354 	p2->p_addr = (struct user *)uaddr;
355 
356 	/*
357 	 * Finish creating the child process.  It will return through a
358 	 * different path later.
359 	 */
360 	uvm_fork(p1, p2, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
361 	    stacksize, func ? func : child_return, arg ? arg : p2);
362 
363 	timeout_set(&p2->p_stats->p_virt_to, virttimer_trampoline, p2);
364 	timeout_set(&p2->p_stats->p_prof_to, proftimer_trampoline, p2);
365 
366 	vm = p2->p_vmspace;
367 
368 	if (flags & FORK_FORK) {
369 		forkstat.cntfork++;
370 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
371 	} else if (flags & FORK_VFORK) {
372 		forkstat.cntvfork++;
373 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
374 	} else if (flags & FORK_RFORK) {
375 		forkstat.cntrfork++;
376 		forkstat.sizrfork += vm->vm_dsize + vm->vm_ssize;
377 	} else {
378 		forkstat.cntkthread++;
379 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
380 	}
381 
382 	if (p2->p_flag & P_TRACED && flags & FORK_FORK)
383 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
384 #if NSYSTRACE > 0
385 	if (ISSET(p1->p_flag, P_SYSTRACE))
386 		newstrp = systrace_getproc();
387 #endif
388 
389 	/* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
390 	rw_enter_write(&allproclk);
391 	do {
392 		lastpid = 1 + (randompid ? arc4random() : lastpid) % PID_MAX;
393 	} while (pidtaken(lastpid));
394 	p2->p_pid = lastpid;
395 
396 	LIST_INSERT_HEAD(&allproc, p2, p_list);
397 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
398 	rw_exit_write(&allproclk);
399 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
400 	LIST_INSERT_AFTER(p1, p2, p_pglist);
401 	if (p2->p_flag & P_TRACED) {
402 		p2->p_oppid = p1->p_pid;
403 		if (p2->p_pptr != p1->p_pptr)
404 			proc_reparent(p2, p1->p_pptr);
405 
406 		/*
407 		 * Set ptrace status.
408 		 */
409 		if (flags & FORK_FORK) {
410 			p2->p_ptstat = newptstat;
411 			newptstat = NULL;
412 			p1->p_ptstat->pe_report_event = PTRACE_FORK;
413 			p2->p_ptstat->pe_report_event = PTRACE_FORK;
414 			p1->p_ptstat->pe_other_pid = p2->p_pid;
415 			p2->p_ptstat->pe_other_pid = p1->p_pid;
416 		}
417 	}
418 
419 #if NSYSTRACE > 0
420 	if (newstrp)
421 		systrace_fork(p1, p2, newstrp);
422 #endif
423 
424 	/*
425 	 * Make child runnable, set start time, and add to run queue.
426 	 */
427 	SCHED_LOCK(s);
428  	getmicrotime(&p2->p_stats->p_start);
429 	p2->p_acflag = AFORK;
430 	p2->p_stat = SRUN;
431 	p2->p_cpu = sched_choosecpu_fork(p1, flags);
432 	setrunqueue(p2);
433 	SCHED_UNLOCK(s);
434 
435 	if (newptstat)
436 		free(newptstat, M_SUBPROC);
437 
438 	/*
439 	 * Notify any interested parties about the new process.
440 	 */
441 	if ((flags & FORK_THREAD) == 0)
442 		KNOTE(&p1->p_p->ps_klist, NOTE_FORK | p2->p_pid);
443 
444 	/*
445 	 * Update stats now that we know the fork was successful.
446 	 */
447 	uvmexp.forks++;
448 	if (flags & FORK_PPWAIT)
449 		uvmexp.forks_ppwait++;
450 	if (flags & FORK_SHAREVM)
451 		uvmexp.forks_sharevm++;
452 
453 	/*
454 	 * Pass a pointer to the new process to the caller.
455 	 */
456 	if (rnewprocp != NULL)
457 		*rnewprocp = p2;
458 
459 	/*
460 	 * Preserve synchronization semantics of vfork.  If waiting for
461 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
462 	 * proc (in case of exit).
463 	 */
464 	if (flags & FORK_PPWAIT)
465 		while (p2->p_flag & P_PPWAIT)
466 			tsleep(p1, PWAIT, "ppwait", 0);
467 
468 	/*
469 	 * If we're tracing the child, alert the parent too.
470 	 */
471 	if ((flags & FORK_PTRACE) && (p1->p_flag & P_TRACED))
472 		psignal(p1, SIGTRAP);
473 
474 	/*
475 	 * Return child pid to parent process,
476 	 * marking us as parent via retval[1].
477 	 */
478 	if (retval != NULL) {
479 		retval[0] = p2->p_pid;
480 		retval[1] = 0;
481 	}
482 	return (0);
483 }
484 
485 /*
486  * Checks for current use of a pid, either as a pid or pgid.
487  */
488 int
489 pidtaken(pid_t pid)
490 {
491 	struct proc *p;
492 
493 	if (pfind(pid) != NULL)
494 		return (1);
495 	if (pgfind(pid) != NULL)
496 		return (1);
497 	LIST_FOREACH(p, &zombproc, p_list)
498 		if (p->p_pid == pid || (p->p_pgrp && p->p_pgrp->pg_id == pid))
499 			return (1);
500 	return (0);
501 }
502 
503 #if defined(MULTIPROCESSOR)
504 /*
505  * XXX This is a slight hack to get newly-formed processes to
506  * XXX acquire the kernel lock as soon as they run.
507  */
508 void
509 proc_trampoline_mp(void)
510 {
511 	struct proc *p;
512 
513 	p = curproc;
514 
515 	SCHED_ASSERT_LOCKED();
516 	__mp_unlock(&sched_lock);
517 	spl0();
518 	SCHED_ASSERT_UNLOCKED();
519 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
520 
521 	KERNEL_PROC_LOCK(p);
522 }
523 #endif
524