xref: /openbsd-src/sys/kern/kern_fork.c (revision e54c1b3dc44fe36eaefd1841275d07a8ac09c0c8)
1 /*	$OpenBSD: kern_fork.c,v 1.126 2011/06/06 17:05:46 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/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 	parent = parentproc->p_p;
162 
163 	TAILQ_INIT(&pr->ps_threads);
164 	TAILQ_INSERT_TAIL(&pr->ps_threads, newproc, p_thr_link);
165 	LIST_INSERT_AFTER(parent, pr, ps_pglist);
166 	pr->ps_pptr = parent;
167 	LIST_INSERT_HEAD(&parent->ps_children, pr, ps_sibling);
168 	LIST_INIT(&pr->ps_children);
169 	pr->ps_refcnt = 1;
170 
171 	/*
172 	 * Make a process structure for the new process.
173 	 * Start by zeroing the section of proc that is zero-initialized,
174 	 * then copy the section that is copied directly from the parent.
175 	 */
176 	bzero(&pr->ps_startzero,
177 	    (unsigned) ((caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero));
178 	bcopy(&parent->ps_startcopy, &pr->ps_startcopy,
179 	    (unsigned) ((caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy));
180 
181 	/* post-copy fixups */
182 	pr->ps_cred = pool_get(&pcred_pool, PR_WAITOK);
183 	bcopy(parent->ps_cred, pr->ps_cred, sizeof(*pr->ps_cred));
184 	crhold(parent->ps_cred->pc_ucred);
185 	pr->ps_limit->p_refcnt++;
186 
187 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
188 	if (parent->ps_session->s_ttyvp != NULL &&
189 	    parent->ps_flags & PS_CONTROLT)
190 		atomic_setbits_int(&pr->ps_flags, PS_CONTROLT);
191 
192 	newproc->p_p = pr;
193 }
194 
195 /* print the 'table full' message once per 10 seconds */
196 struct timeval fork_tfmrate = { 10, 0 };
197 
198 int
199 fork1(struct proc *p1, int exitsig, int flags, void *stack, size_t stacksize,
200     void (*func)(void *), void *arg, register_t *retval,
201     struct proc **rnewprocp)
202 {
203 	struct proc *p2;
204 	uid_t uid;
205 	struct vmspace *vm;
206 	int count;
207 	vaddr_t uaddr;
208 	int s;
209 	extern void endtsleep(void *);
210 	extern void realitexpire(void *);
211 	struct  ptrace_state *newptstat = NULL;
212 #if NSYSTRACE > 0
213 	void *newstrp = NULL;
214 #endif
215 
216 	/* sanity check some flag combinations */
217 	if (flags & FORK_THREAD) {
218 		if (!rthreads_enabled)
219 			return (ENOTSUP);
220 		if ((flags & (FORK_SIGHAND | FORK_NOZOMBIE)) !=
221 		    (FORK_SIGHAND | FORK_NOZOMBIE))
222 			return (EINVAL);
223 	}
224 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
225 		return (EINVAL);
226 
227 	/*
228 	 * Although process entries are dynamically created, we still keep
229 	 * a global limit on the maximum number we will create. We reserve
230 	 * the last 5 processes to root. The variable nprocs is the current
231 	 * number of processes, maxproc is the limit.
232 	 */
233 	uid = p1->p_cred->p_ruid;
234 	if ((nprocs >= maxproc - 5 && uid != 0) || nprocs >= maxproc) {
235 		static struct timeval lasttfm;
236 
237 		if (ratecheck(&lasttfm, &fork_tfmrate))
238 			tablefull("proc");
239 		return (EAGAIN);
240 	}
241 	nprocs++;
242 
243 	/*
244 	 * Increment the count of procs running with this uid. Don't allow
245 	 * a nonprivileged user to exceed their current limit.
246 	 */
247 	count = chgproccnt(uid, 1);
248 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
249 		(void)chgproccnt(uid, -1);
250 		nprocs--;
251 		return (EAGAIN);
252 	}
253 
254 	uaddr = uvm_km_kmemalloc_pla(kernel_map, uvm.kernel_object, USPACE,
255 	    USPACE_ALIGN, UVM_KMF_ZERO,
256 	    no_constraint.ucr_low, no_constraint.ucr_high,
257 	    0, 0, USPACE/PAGE_SIZE);
258 	if (uaddr == 0) {
259 		chgproccnt(uid, -1);
260 		nprocs--;
261 		return (ENOMEM);
262 	}
263 
264 	/*
265 	 * From now on, we're committed to the fork and cannot fail.
266 	 */
267 
268 	/* Allocate new proc. */
269 	p2 = pool_get(&proc_pool, PR_WAITOK);
270 
271 	p2->p_stat = SIDL;			/* protect against others */
272 	p2->p_exitsig = exitsig;
273 	p2->p_flag = 0;
274 
275 	if (flags & FORK_THREAD) {
276 		atomic_setbits_int(&p2->p_flag, P_THREAD);
277 		p2->p_p = p1->p_p;
278 		TAILQ_INSERT_TAIL(&p2->p_p->ps_threads, p2, p_thr_link);
279 		p2->p_p->ps_refcnt++;
280 	} else {
281 		process_new(p2, p1);
282 	}
283 
284 	/*
285 	 * Make a proc table entry for the new process.
286 	 * Start by zeroing the section of proc that is zero-initialized,
287 	 * then copy the section that is copied directly from the parent.
288 	 */
289 	bzero(&p2->p_startzero,
290 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
291 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
292 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
293 
294 	/*
295 	 * Initialize the timeouts.
296 	 */
297 	timeout_set(&p2->p_sleep_to, endtsleep, p2);
298 	timeout_set(&p2->p_realit_to, realitexpire, p2);
299 
300 	/*
301 	 * Duplicate sub-structures as needed.
302 	 * Increase reference counts on shared objects.
303 	 * The p_stats and p_sigacts substructs are set in vm_fork.
304 	 */
305 	if (p1->p_flag & P_PROFIL)
306 		startprofclock(p2);
307 	if (flags & FORK_PTRACE)
308 		atomic_setbits_int(&p2->p_flag, p1->p_flag & P_TRACED);
309 
310 	/* bump references to the text vnode (for procfs) */
311 	p2->p_textvp = p1->p_textvp;
312 	if (p2->p_textvp)
313 		vref(p2->p_textvp);
314 
315 	if (flags & FORK_CLEANFILES)
316 		p2->p_fd = fdinit(p1);
317 	else if (flags & FORK_SHAREFILES)
318 		p2->p_fd = fdshare(p1);
319 	else
320 		p2->p_fd = fdcopy(p1);
321 
322 	if (flags & FORK_PPWAIT) {
323 		atomic_setbits_int(&p2->p_p->ps_flags, PS_PPWAIT);
324 		atomic_setbits_int(&p1->p_p->ps_flags, PS_ISPWAIT);
325 	}
326 	if (flags & FORK_NOZOMBIE)
327 		atomic_setbits_int(&p2->p_flag, P_NOZOMBIE);
328 
329 #ifdef KTRACE
330 	/*
331 	 * Copy traceflag and tracefile if enabled.
332 	 * If not inherited, these were zeroed above.
333 	 */
334 	if (p1->p_traceflag & KTRFAC_INHERIT) {
335 		p2->p_traceflag = p1->p_traceflag;
336 		if ((p2->p_tracep = p1->p_tracep) != NULL)
337 			vref(p2->p_tracep);
338 	}
339 #endif
340 
341 	/*
342 	 * set priority of child to be that of parent
343 	 * XXX should move p_estcpu into the region of struct proc which gets
344 	 * copied.
345 	 */
346 	scheduler_fork_hook(p1, p2);
347 
348 	/*
349 	 * Create signal actions for the child process.
350 	 */
351 	if (flags & FORK_SIGHAND)
352 		sigactsshare(p1, p2);
353 	else
354 		p2->p_sigacts = sigactsinit(p1);
355 
356 	/*
357 	 * If emulation has process fork hook, call it now.
358 	 */
359 	if (p2->p_emul->e_proc_fork)
360 		(*p2->p_emul->e_proc_fork)(p2, p1);
361 
362 	p2->p_addr = (struct user *)uaddr;
363 
364 	/*
365 	 * Finish creating the child process.  It will return through a
366 	 * different path later.
367 	 */
368 	uvm_fork(p1, p2, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
369 	    stacksize, func ? func : child_return, arg ? arg : p2);
370 
371 	timeout_set(&p2->p_stats->p_virt_to, virttimer_trampoline, p2);
372 	timeout_set(&p2->p_stats->p_prof_to, proftimer_trampoline, p2);
373 
374 	vm = p2->p_vmspace;
375 
376 	if (flags & FORK_FORK) {
377 		forkstat.cntfork++;
378 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
379 	} else if (flags & FORK_VFORK) {
380 		forkstat.cntvfork++;
381 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
382 	} else if (flags & FORK_RFORK) {
383 		forkstat.cntrfork++;
384 		forkstat.sizrfork += vm->vm_dsize + vm->vm_ssize;
385 	} else {
386 		forkstat.cntkthread++;
387 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
388 	}
389 
390 	if (p2->p_flag & P_TRACED && flags & FORK_FORK)
391 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
392 #if NSYSTRACE > 0
393 	if (ISSET(p1->p_flag, P_SYSTRACE))
394 		newstrp = systrace_getproc();
395 #endif
396 
397 	/* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
398 	do {
399 		lastpid = 1 + (randompid ? arc4random() : lastpid) % PID_MAX;
400 	} while (pidtaken(lastpid));
401 	p2->p_pid = lastpid;
402 
403 	LIST_INSERT_HEAD(&allproc, p2, p_list);
404 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
405 
406 	if (p2->p_flag & P_TRACED) {
407 		p2->p_oppid = p1->p_pid;
408 		if ((flags & FORK_THREAD) == 0 &&
409 		    p2->p_p->ps_pptr != p1->p_p->ps_pptr)
410 			proc_reparent(p2->p_p, p1->p_p->ps_pptr);
411 
412 		/*
413 		 * Set ptrace status.
414 		 */
415 		if (flags & FORK_FORK) {
416 			p2->p_ptstat = newptstat;
417 			newptstat = NULL;
418 			p1->p_ptstat->pe_report_event = PTRACE_FORK;
419 			p2->p_ptstat->pe_report_event = PTRACE_FORK;
420 			p1->p_ptstat->pe_other_pid = p2->p_pid;
421 			p2->p_ptstat->pe_other_pid = p1->p_pid;
422 		}
423 	}
424 
425 #if NSYSTRACE > 0
426 	if (newstrp)
427 		systrace_fork(p1, p2, newstrp);
428 #endif
429 
430 	/*
431 	 * Make child runnable, set start time, and add to run queue.
432 	 */
433 	SCHED_LOCK(s);
434  	getmicrotime(&p2->p_stats->p_start);
435 	p2->p_acflag = AFORK;
436 	p2->p_stat = SRUN;
437 	p2->p_cpu = sched_choosecpu_fork(p1, flags);
438 	setrunqueue(p2);
439 	SCHED_UNLOCK(s);
440 
441 	if (newptstat)
442 		free(newptstat, M_SUBPROC);
443 
444 	/*
445 	 * Notify any interested parties about the new process.
446 	 */
447 	if ((flags & FORK_THREAD) == 0)
448 		KNOTE(&p1->p_p->ps_klist, NOTE_FORK | p2->p_pid);
449 
450 	/*
451 	 * Update stats now that we know the fork was successful.
452 	 */
453 	uvmexp.forks++;
454 	if (flags & FORK_PPWAIT)
455 		uvmexp.forks_ppwait++;
456 	if (flags & FORK_SHAREVM)
457 		uvmexp.forks_sharevm++;
458 
459 	/*
460 	 * Pass a pointer to the new process to the caller.
461 	 */
462 	if (rnewprocp != NULL)
463 		*rnewprocp = p2;
464 
465 	/*
466 	 * Preserve synchronization semantics of vfork.  If waiting for
467 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
468 	 * on ourselves, and sleep on our process for the latter flag
469 	 * to go away.
470 	 * XXX Need to stop other rthreads in the parent
471 	 */
472 	if (flags & FORK_PPWAIT)
473 		while (p1->p_p->ps_flags & PS_ISPWAIT)
474 			tsleep(p1->p_p, PWAIT, "ppwait", 0);
475 
476 	/*
477 	 * If we're tracing the child, alert the parent too.
478 	 */
479 	if ((flags & FORK_PTRACE) && (p1->p_flag & P_TRACED))
480 		psignal(p1, SIGTRAP);
481 
482 	/*
483 	 * Return child pid to parent process,
484 	 * marking us as parent via retval[1].
485 	 */
486 	if (retval != NULL) {
487 		retval[0] = p2->p_pid +
488 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
489 		retval[1] = 0;
490 	}
491 	return (0);
492 }
493 
494 /*
495  * Checks for current use of a pid, either as a pid or pgid.
496  */
497 int
498 pidtaken(pid_t pid)
499 {
500 	struct proc *p;
501 
502 	if (pfind(pid) != NULL)
503 		return (1);
504 	if (pgfind(pid) != NULL)
505 		return (1);
506 	LIST_FOREACH(p, &zombproc, p_list) {
507 		if (p->p_pid == pid || (p->p_p->ps_pgrp && p->p_p->ps_pgrp->pg_id == pid))
508 			return (1);
509 	}
510 	return (0);
511 }
512 
513 #if defined(MULTIPROCESSOR)
514 /*
515  * XXX This is a slight hack to get newly-formed processes to
516  * XXX acquire the kernel lock as soon as they run.
517  */
518 void
519 proc_trampoline_mp(void)
520 {
521 	struct proc *p;
522 
523 	p = curproc;
524 
525 	SCHED_ASSERT_LOCKED();
526 	__mp_unlock(&sched_lock);
527 	spl0();
528 	SCHED_ASSERT_UNLOCKED();
529 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
530 
531 	KERNEL_PROC_LOCK(p);
532 }
533 #endif
534