xref: /openbsd-src/sys/kern/kern_fork.c (revision 351c20c593075aab5a3342f59ad7ce1e96b7b1ad) 
1 /*	$OpenBSD: kern_fork.c,v 1.113 2010/06/29 00:28:14 tedu 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 | PR_ZERO);
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 	pr->ps_rdomain = parent->ps_rdomain;
167 
168 	newproc->p_p = pr;
169 }
170 
171 /* print the 'table full' message once per 10 seconds */
172 struct timeval fork_tfmrate = { 10, 0 };
173 
174 int
175 fork1(struct proc *p1, int exitsig, int flags, void *stack, size_t stacksize,
176     void (*func)(void *), void *arg, register_t *retval,
177     struct proc **rnewprocp)
178 {
179 	struct proc *p2;
180 	uid_t uid;
181 	struct vmspace *vm;
182 	int count;
183 	vaddr_t uaddr;
184 	int s;
185 	extern void endtsleep(void *);
186 	extern void realitexpire(void *);
187 	struct  ptrace_state *newptstat = NULL;
188 #if NSYSTRACE > 0
189 	void *newstrp = NULL;
190 #endif
191 
192 	/* sanity check some flag combinations */
193 	if (flags & FORK_THREAD)
194 	{
195 		if (!rthreads_enabled)
196 			return (ENOTSUP);
197 		if ((flags & (FORK_SIGHAND | FORK_NOZOMBIE)) !=
198 		    (FORK_SIGHAND | FORK_NOZOMBIE))
199 			return (EINVAL);
200 	}
201 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
202 		return (EINVAL);
203 
204 	/*
205 	 * Although process entries are dynamically created, we still keep
206 	 * a global limit on the maximum number we will create. We reserve
207 	 * the last 5 processes to root. The variable nprocs is the current
208 	 * number of processes, maxproc is the limit.
209 	 */
210 	uid = p1->p_cred->p_ruid;
211 	if ((nprocs >= maxproc - 5 && uid != 0) || nprocs >= maxproc) {
212 		static struct timeval lasttfm;
213 
214 		if (ratecheck(&lasttfm, &fork_tfmrate))
215 			tablefull("proc");
216 		return (EAGAIN);
217 	}
218 	nprocs++;
219 
220 	/*
221 	 * Increment the count of procs running with this uid. Don't allow
222 	 * a nonprivileged user to exceed their current limit.
223 	 */
224 	count = chgproccnt(uid, 1);
225 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
226 		(void)chgproccnt(uid, -1);
227 		nprocs--;
228 		return (EAGAIN);
229 	}
230 
231 	uaddr = uvm_km_alloc1(kernel_map, USPACE, USPACE_ALIGN, 1);
232 	if (uaddr == 0) {
233 		chgproccnt(uid, -1);
234 		nprocs--;
235 		return (ENOMEM);
236 	}
237 
238 	/*
239 	 * From now on, we're committed to the fork and cannot fail.
240 	 */
241 
242 	/* Allocate new proc. */
243 	p2 = pool_get(&proc_pool, PR_WAITOK);
244 
245 	p2->p_stat = SIDL;			/* protect against others */
246 	p2->p_exitsig = exitsig;
247 	p2->p_flag = 0;
248 
249 	if (flags & FORK_THREAD) {
250 		atomic_setbits_int(&p2->p_flag, P_THREAD);
251 		p2->p_p = p1->p_p;
252 		TAILQ_INSERT_TAIL(&p2->p_p->ps_threads, p2, p_thr_link);
253 		p2->p_p->ps_refcnt++;
254 	} else {
255 		process_new(p2, p1);
256 	}
257 
258 	/*
259 	 * Make a proc table entry for the new process.
260 	 * Start by zeroing the section of proc that is zero-initialized,
261 	 * then copy the section that is copied directly from the parent.
262 	 */
263 	bzero(&p2->p_startzero,
264 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
265 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
266 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
267 
268 	/*
269 	 * Initialize the timeouts.
270 	 */
271 	timeout_set(&p2->p_sleep_to, endtsleep, p2);
272 	timeout_set(&p2->p_realit_to, realitexpire, p2);
273 
274 	/*
275 	 * Duplicate sub-structures as needed.
276 	 * Increase reference counts on shared objects.
277 	 * The p_stats and p_sigacts substructs are set in vm_fork.
278 	 */
279 	if (p1->p_flag & P_PROFIL)
280 		startprofclock(p2);
281 	atomic_setbits_int(&p2->p_flag, p1->p_flag & (P_SUGID | P_SUGIDEXEC));
282 	if (flags & FORK_PTRACE)
283 		atomic_setbits_int(&p2->p_flag, p1->p_flag & P_TRACED);
284 	if (flags & FORK_THREAD) {
285 		/* nothing */
286 	} else {
287 		p2->p_p->ps_cred = pool_get(&pcred_pool, PR_WAITOK);
288 		bcopy(p1->p_p->ps_cred, p2->p_p->ps_cred, sizeof(*p2->p_p->ps_cred));
289 		p2->p_p->ps_cred->p_refcnt = 1;
290 		crhold(p1->p_ucred);
291 	}
292 
293 	/* bump references to the text vnode (for procfs) */
294 	p2->p_textvp = p1->p_textvp;
295 	if (p2->p_textvp)
296 		vref(p2->p_textvp);
297 
298 	if (flags & FORK_CLEANFILES)
299 		p2->p_fd = fdinit(p1);
300 	else if (flags & FORK_SHAREFILES)
301 		p2->p_fd = fdshare(p1);
302 	else
303 		p2->p_fd = fdcopy(p1);
304 
305 	/*
306 	 * If ps_limit is still copy-on-write, bump refcnt,
307 	 * otherwise get a copy that won't be modified.
308 	 * (If PL_SHAREMOD is clear, the structure is shared
309 	 * copy-on-write.)
310 	 */
311 	if (flags & FORK_THREAD) {
312 		/* nothing */
313 	} else {
314 		if (p1->p_p->ps_limit->p_lflags & PL_SHAREMOD)
315 			p2->p_p->ps_limit = limcopy(p1->p_p->ps_limit);
316 		else {
317 			p2->p_p->ps_limit = p1->p_p->ps_limit;
318 			p2->p_p->ps_limit->p_refcnt++;
319 		}
320 	}
321 
322 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
323 		atomic_setbits_int(&p2->p_flag, P_CONTROLT);
324 	if (flags & FORK_PPWAIT)
325 		atomic_setbits_int(&p2->p_flag, P_PPWAIT);
326 	p2->p_pptr = p1;
327 	if (flags & FORK_NOZOMBIE)
328 		atomic_setbits_int(&p2->p_flag, P_NOZOMBIE);
329 	LIST_INIT(&p2->p_children);
330 
331 #ifdef KTRACE
332 	/*
333 	 * Copy traceflag and tracefile if enabled.
334 	 * If not inherited, these were zeroed above.
335 	 */
336 	if (p1->p_traceflag & KTRFAC_INHERIT) {
337 		p2->p_traceflag = p1->p_traceflag;
338 		if ((p2->p_tracep = p1->p_tracep) != NULL)
339 			vref(p2->p_tracep);
340 	}
341 #endif
342 
343 	/*
344 	 * set priority of child to be that of parent
345 	 * XXX should move p_estcpu into the region of struct proc which gets
346 	 * copied.
347 	 */
348 	scheduler_fork_hook(p1, p2);
349 
350 	/*
351 	 * Create signal actions for the child process.
352 	 */
353 	if (flags & FORK_SIGHAND)
354 		sigactsshare(p1, p2);
355 	else
356 		p2->p_sigacts = sigactsinit(p1);
357 
358 	/*
359 	 * If emulation has process fork hook, call it now.
360 	 */
361 	if (p2->p_emul->e_proc_fork)
362 		(*p2->p_emul->e_proc_fork)(p2, p1);
363 
364 	p2->p_addr = (struct user *)uaddr;
365 
366 	/*
367 	 * Finish creating the child process.  It will return through a
368 	 * different path later.
369 	 */
370 	uvm_fork(p1, p2, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
371 	    stacksize, func ? func : child_return, arg ? arg : p2);
372 
373 	timeout_set(&p2->p_stats->p_virt_to, virttimer_trampoline, p2);
374 	timeout_set(&p2->p_stats->p_prof_to, proftimer_trampoline, p2);
375 
376 	vm = p2->p_vmspace;
377 
378 	if (flags & FORK_FORK) {
379 		forkstat.cntfork++;
380 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
381 	} else if (flags & FORK_VFORK) {
382 		forkstat.cntvfork++;
383 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
384 	} else if (flags & FORK_RFORK) {
385 		forkstat.cntrfork++;
386 		forkstat.sizrfork += vm->vm_dsize + vm->vm_ssize;
387 	} else {
388 		forkstat.cntkthread++;
389 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
390 	}
391 
392 	if (p2->p_flag & P_TRACED && flags & FORK_FORK)
393 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
394 #if NSYSTRACE > 0
395 	if (ISSET(p1->p_flag, P_SYSTRACE))
396 		newstrp = systrace_getproc();
397 #endif
398 
399 	/* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
400 	rw_enter_write(&allproclk);
401 	do {
402 		lastpid = 1 + (randompid ? arc4random() : lastpid) % PID_MAX;
403 	} while (pidtaken(lastpid));
404 	p2->p_pid = lastpid;
405 
406 	LIST_INSERT_HEAD(&allproc, p2, p_list);
407 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
408 	rw_exit_write(&allproclk);
409 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
410 	LIST_INSERT_AFTER(p1, p2, p_pglist);
411 	if (p2->p_flag & P_TRACED) {
412 		p2->p_oppid = p1->p_pid;
413 		if (p2->p_pptr != p1->p_pptr)
414 			proc_reparent(p2, p1->p_pptr);
415 
416 		/*
417 		 * Set ptrace status.
418 		 */
419 		if (flags & FORK_FORK) {
420 			p2->p_ptstat = newptstat;
421 			newptstat = NULL;
422 			p1->p_ptstat->pe_report_event = PTRACE_FORK;
423 			p2->p_ptstat->pe_report_event = PTRACE_FORK;
424 			p1->p_ptstat->pe_other_pid = p2->p_pid;
425 			p2->p_ptstat->pe_other_pid = p1->p_pid;
426 		}
427 	}
428 
429 #if NSYSTRACE > 0
430 	if (newstrp)
431 		systrace_fork(p1, p2, newstrp);
432 #endif
433 
434 	/*
435 	 * Make child runnable, set start time, and add to run queue.
436 	 */
437 	SCHED_LOCK(s);
438  	getmicrotime(&p2->p_stats->p_start);
439 	p2->p_acflag = AFORK;
440 	p2->p_stat = SRUN;
441 	p2->p_cpu = sched_choosecpu_fork(p1, flags);
442 	setrunqueue(p2);
443 	SCHED_UNLOCK(s);
444 
445 	if (newptstat)
446 		free(newptstat, M_SUBPROC);
447 
448 	/*
449 	 * Notify any interested parties about the new process.
450 	 */
451 	if ((flags & FORK_THREAD) == 0)
452 		KNOTE(&p1->p_p->ps_klist, NOTE_FORK | p2->p_pid);
453 
454 	/*
455 	 * Update stats now that we know the fork was successful.
456 	 */
457 	uvmexp.forks++;
458 	if (flags & FORK_PPWAIT)
459 		uvmexp.forks_ppwait++;
460 	if (flags & FORK_SHAREVM)
461 		uvmexp.forks_sharevm++;
462 
463 	/*
464 	 * Pass a pointer to the new process to the caller.
465 	 */
466 	if (rnewprocp != NULL)
467 		*rnewprocp = p2;
468 
469 	/*
470 	 * Preserve synchronization semantics of vfork.  If waiting for
471 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
472 	 * proc (in case of exit).
473 	 */
474 	if (flags & FORK_PPWAIT)
475 		while (p2->p_flag & P_PPWAIT)
476 			tsleep(p1, PWAIT, "ppwait", 0);
477 
478 	/*
479 	 * If we're tracing the child, alert the parent too.
480 	 */
481 	if ((flags & FORK_PTRACE) && (p1->p_flag & P_TRACED))
482 		psignal(p1, SIGTRAP);
483 
484 	/*
485 	 * Return child pid to parent process,
486 	 * marking us as parent via retval[1].
487 	 */
488 	if (retval != NULL) {
489 		retval[0] = p2->p_pid;
490 		retval[1] = 0;
491 	}
492 	return (0);
493 }
494 
495 /*
496  * Checks for current use of a pid, either as a pid or pgid.
497  */
498 int
499 pidtaken(pid_t pid)
500 {
501 	struct proc *p;
502 
503 	if (pfind(pid) != NULL)
504 		return (1);
505 	if (pgfind(pid) != NULL)
506 		return (1);
507 	LIST_FOREACH(p, &zombproc, p_list)
508 		if (p->p_pid == pid || (p->p_pgrp && p->p_pgrp->pg_id == pid))
509 			return (1);
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