xref: /netbsd-src/sys/kern/kern_fork.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: kern_fork.c,v 1.157 2008/01/28 20:09:06 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2001, 2004, 2006, 2007 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1989, 1991, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  * (c) UNIX System Laboratories, Inc.
44  * All or some portions of this file are derived from material licensed
45  * to the University of California by American Telephone and Telegraph
46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47  * the permission of UNIX System Laboratories, Inc.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  * 3. Neither the name of the University nor the names of its contributors
58  *    may be used to endorse or promote products derived from this software
59  *    without specific prior written permission.
60  *
61  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
62  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
65  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71  * SUCH DAMAGE.
72  *
73  *	@(#)kern_fork.c	8.8 (Berkeley) 2/14/95
74  */
75 
76 #include <sys/cdefs.h>
77 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.157 2008/01/28 20:09:06 ad Exp $");
78 
79 #include "opt_ktrace.h"
80 #include "opt_multiprocessor.h"
81 
82 #include <sys/param.h>
83 #include <sys/systm.h>
84 #include <sys/filedesc.h>
85 #include <sys/kernel.h>
86 #include <sys/malloc.h>
87 #include <sys/pool.h>
88 #include <sys/mount.h>
89 #include <sys/proc.h>
90 #include <sys/ras.h>
91 #include <sys/resourcevar.h>
92 #include <sys/vnode.h>
93 #include <sys/file.h>
94 #include <sys/acct.h>
95 #include <sys/ktrace.h>
96 #include <sys/vmmeter.h>
97 #include <sys/sched.h>
98 #include <sys/signalvar.h>
99 #include <sys/kauth.h>
100 #include <sys/atomic.h>
101 #include <sys/syscallargs.h>
102 
103 #include <uvm/uvm_extern.h>
104 
105 u_int	nprocs = 1;		/* process 0 */
106 
107 /*
108  * Number of ticks to sleep if fork() would fail due to process hitting
109  * limits. Exported in miliseconds to userland via sysctl.
110  */
111 int	forkfsleep = 0;
112 
113 /*ARGSUSED*/
114 int
115 sys_fork(struct lwp *l, const void *v, register_t *retval)
116 {
117 
118 	return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
119 }
120 
121 /*
122  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
123  * Address space is not shared, but parent is blocked until child exit.
124  */
125 /*ARGSUSED*/
126 int
127 sys_vfork(struct lwp *l, const void *v, register_t *retval)
128 {
129 
130 	return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
131 	    retval, NULL));
132 }
133 
134 /*
135  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
136  * semantics.  Address space is shared, and parent is blocked until child exit.
137  */
138 /*ARGSUSED*/
139 int
140 sys___vfork14(struct lwp *l, const void *v, register_t *retval)
141 {
142 
143 	return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
144 	    NULL, NULL, retval, NULL));
145 }
146 
147 /*
148  * Linux-compatible __clone(2) system call.
149  */
150 int
151 sys___clone(struct lwp *l, const struct sys___clone_args *uap, register_t *retval)
152 {
153 	/* {
154 		syscallarg(int) flags;
155 		syscallarg(void *) stack;
156 	} */
157 	int flags, sig;
158 
159 	/*
160 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
161 	 */
162 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
163 		return (EINVAL);
164 
165 	/*
166 	 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
167 	 */
168 	if (SCARG(uap, flags) & CLONE_SIGHAND
169 	    && (SCARG(uap, flags) & CLONE_VM) == 0)
170 		return (EINVAL);
171 
172 	flags = 0;
173 
174 	if (SCARG(uap, flags) & CLONE_VM)
175 		flags |= FORK_SHAREVM;
176 	if (SCARG(uap, flags) & CLONE_FS)
177 		flags |= FORK_SHARECWD;
178 	if (SCARG(uap, flags) & CLONE_FILES)
179 		flags |= FORK_SHAREFILES;
180 	if (SCARG(uap, flags) & CLONE_SIGHAND)
181 		flags |= FORK_SHARESIGS;
182 	if (SCARG(uap, flags) & CLONE_VFORK)
183 		flags |= FORK_PPWAIT;
184 
185 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
186 	if (sig < 0 || sig >= _NSIG)
187 		return (EINVAL);
188 
189 	/*
190 	 * Note that the Linux API does not provide a portable way of
191 	 * specifying the stack area; the caller must know if the stack
192 	 * grows up or down.  So, we pass a stack size of 0, so that the
193 	 * code that makes this adjustment is a noop.
194 	 */
195 	return (fork1(l, flags, sig, SCARG(uap, stack), 0,
196 	    NULL, NULL, retval, NULL));
197 }
198 
199 /* print the 'table full' message once per 10 seconds */
200 struct timeval fork_tfmrate = { 10, 0 };
201 
202 /*
203  * General fork call.  Note that another LWP in the process may call exec()
204  * or exit() while we are forking.  It's safe to continue here, because
205  * neither operation will complete until all LWPs have exited the process.
206  */
207 int
208 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
209     void (*func)(void *), void *arg, register_t *retval,
210     struct proc **rnewprocp)
211 {
212 	struct proc	*p1, *p2, *parent;
213 	struct plimit   *p1_lim;
214 	uid_t		uid;
215 	struct lwp	*l2;
216 	int		count;
217 	vaddr_t		uaddr;
218 	bool		inmem;
219 	int		tmp;
220 	int		tnprocs;
221 	int		error = 0;
222 
223 	p1 = l1->l_proc;
224 	mutex_enter(&p1->p_mutex);
225 	uid = kauth_cred_getuid(p1->p_cred);
226 	mutex_exit(&p1->p_mutex);
227 	tnprocs = atomic_inc_uint_nv(&nprocs);
228 
229 	/*
230 	 * Although process entries are dynamically created, we still keep
231 	 * a global limit on the maximum number we will create.
232 	 */
233 	if (__predict_false(tnprocs >= maxproc))
234 		error = -1;
235 	else
236 		error = kauth_authorize_process(l1->l_cred,
237 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
238 
239 	if (error) {
240 		static struct timeval lasttfm;
241 		atomic_dec_uint(&nprocs);
242 		if (ratecheck(&lasttfm, &fork_tfmrate))
243 			tablefull("proc", "increase kern.maxproc or NPROC");
244 		if (forkfsleep)
245 			(void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep);
246 		return (EAGAIN);
247 	}
248 
249 	/*
250 	 * Enforce limits.
251 	 */
252 	count = chgproccnt(uid, 1);
253 	if (__predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
254 		(void)chgproccnt(uid, -1);
255 		atomic_dec_uint(&nprocs);
256 		if (forkfsleep)
257 			(void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep);
258 		return (EAGAIN);
259 	}
260 
261 	/*
262 	 * Allocate virtual address space for the U-area now, while it
263 	 * is still easy to abort the fork operation if we're out of
264 	 * kernel virtual address space.  The actual U-area pages will
265 	 * be allocated and wired in uvm_fork() if needed.
266 	 */
267 
268 	inmem = uvm_uarea_alloc(&uaddr);
269 	if (__predict_false(uaddr == 0)) {
270 		(void)chgproccnt(uid, -1);
271 		atomic_dec_uint(&nprocs);
272 		return (ENOMEM);
273 	}
274 
275 	/*
276 	 * We are now committed to the fork.  From here on, we may
277 	 * block on resources, but resource allocation may NOT fail.
278 	 */
279 
280 	/* Allocate new proc. */
281 	p2 = proc_alloc();
282 
283 	/*
284 	 * Make a proc table entry for the new process.
285 	 * Start by zeroing the section of proc that is zero-initialized,
286 	 * then copy the section that is copied directly from the parent.
287 	 */
288 	memset(&p2->p_startzero, 0,
289 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
290 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
291 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
292 
293 	CIRCLEQ_INIT(&p2->p_sigpend.sp_info);
294 
295 	LIST_INIT(&p2->p_lwps);
296 	LIST_INIT(&p2->p_sigwaiters);
297 
298 	/*
299 	 * Duplicate sub-structures as needed.
300 	 * Increase reference counts on shared objects.
301 	 * The p_stats and p_sigacts substructs are set in uvm_fork().
302 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
303 	 * handling are important in order to keep a consistent behaviour
304 	 * for the child after the fork.
305 	 */
306 	p2->p_flag = p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN);
307 	p2->p_emul = p1->p_emul;
308 	p2->p_execsw = p1->p_execsw;
309 
310 	if (flags & FORK_SYSTEM) {
311 		/*
312 		 * Mark it as a system process.  Set P_NOCLDWAIT so that
313 		 * children are reparented to init(8) when they exit.
314 		 * init(8) can easily wait them out for us.
315 		 */
316 		p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT);
317 	}
318 
319 	/* XXX p_smutex can be IPL_VM except for audio drivers */
320 	mutex_init(&p2->p_smutex, MUTEX_DEFAULT, IPL_SCHED);
321 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
322 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
323 	mutex_init(&p2->p_mutex, MUTEX_DEFAULT, IPL_NONE);
324 	rw_init(&p2->p_reflock);
325 	cv_init(&p2->p_waitcv, "wait");
326 	cv_init(&p2->p_lwpcv, "lwpwait");
327 
328 	kauth_proc_fork(p1, p2);
329 
330 	p2->p_raslist = NULL;
331 #if defined(__HAVE_RAS)
332 	ras_fork(p1, p2);
333 #endif
334 
335 	/* bump references to the text vnode (for procfs) */
336 	p2->p_textvp = p1->p_textvp;
337 	if (p2->p_textvp)
338 		VREF(p2->p_textvp);
339 
340 	if (flags & FORK_SHAREFILES)
341 		fdshare(p1, p2);
342 	else if (flags & FORK_CLEANFILES)
343 		p2->p_fd = fdinit(p1);
344 	else
345 		p2->p_fd = fdcopy(p1);
346 
347 	if (flags & FORK_SHARECWD)
348 		cwdshare(p1, p2);
349 	else
350 		p2->p_cwdi = cwdinit(p1);
351 
352 	/*
353 	 * p_limit (rlimit stuff) is usually copy-on-write, so we just need
354 	 * to bump pl_refcnt.
355 	 * However in some cases (see compat irix, and plausibly from clone)
356 	 * the parent and child share limits - in which case nothing else
357 	 * must have a copy of the limits (PL_SHAREMOD is set).
358 	 */
359 	if (__predict_false(flags & FORK_SHARELIMIT))
360 		lim_privatise(p1, 1);
361 	p1_lim = p1->p_limit;
362 	if (p1_lim->pl_flags & PL_WRITEABLE && !(flags & FORK_SHARELIMIT))
363 		p2->p_limit = lim_copy(p1_lim);
364 	else {
365 		lim_addref(p1_lim);
366 		p2->p_limit = p1_lim;
367 	}
368 
369 	p2->p_sflag = ((flags & FORK_PPWAIT) ? PS_PPWAIT : 0);
370 	p2->p_lflag = 0;
371 	p2->p_slflag = 0;
372 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
373 	p2->p_pptr = parent;
374 	LIST_INIT(&p2->p_children);
375 
376 	p2->p_aio = NULL;
377 
378 #ifdef KTRACE
379 	/*
380 	 * Copy traceflag and tracefile if enabled.
381 	 * If not inherited, these were zeroed above.
382 	 */
383 	if (p1->p_traceflag & KTRFAC_INHERIT) {
384 		mutex_enter(&ktrace_lock);
385 		p2->p_traceflag = p1->p_traceflag;
386 		if ((p2->p_tracep = p1->p_tracep) != NULL)
387 			ktradref(p2);
388 		mutex_exit(&ktrace_lock);
389 	}
390 #endif
391 
392 	/*
393 	 * Create signal actions for the child process.
394 	 */
395 	p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS);
396 	mutex_enter(&p1->p_smutex);
397 	p2->p_sflag |=
398 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
399 	sched_proc_fork(p1, p2);
400 	mutex_exit(&p1->p_smutex);
401 
402 	p2->p_stflag = p1->p_stflag;
403 
404 	/*
405 	 * p_stats.
406 	 * Copy parts of p_stats, and zero out the rest.
407 	 */
408 	p2->p_stats = pstatscopy(p1->p_stats);
409 
410 	/*
411 	 * If emulation has process fork hook, call it now.
412 	 */
413 	if (p2->p_emul->e_proc_fork)
414 		(*p2->p_emul->e_proc_fork)(p2, p1, flags);
415 
416 	/*
417 	 * ...and finally, any other random fork hooks that subsystems
418 	 * might have registered.
419 	 */
420 	doforkhooks(p2, p1);
421 
422 	/*
423 	 * This begins the section where we must prevent the parent
424 	 * from being swapped.
425 	 */
426 	uvm_lwp_hold(l1);
427 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false);
428 
429 	/*
430 	 * Finish creating the child process.
431 	 * It will return through a different path later.
432 	 */
433 	lwp_create(l1, p2, uaddr, inmem, 0, stack, stacksize,
434 	    (func != NULL) ? func : child_return, arg, &l2,
435 	    l1->l_class);
436 
437 	/*
438 	 * It's now safe for the scheduler and other processes to see the
439 	 * child process.
440 	 */
441 	mutex_enter(&proclist_lock);
442 
443 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
444 		p2->p_lflag |= PL_CONTROLT;
445 
446 	LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
447 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
448 
449 	mutex_enter(&proclist_mutex);
450 	LIST_INSERT_AFTER(p1, p2, p_pglist);
451 	mutex_exit(&proclist_mutex);
452 	LIST_INSERT_HEAD(&allproc, p2, p_list);
453 
454 	mutex_exit(&proclist_lock);
455 
456 #ifdef __HAVE_SYSCALL_INTERN
457 	(*p2->p_emul->e_syscall_intern)(p2);
458 #endif
459 
460 	/*
461 	 * Now can be swapped.
462 	 */
463 	uvm_lwp_rele(l1);
464 
465 	/*
466 	 * Notify any interested parties about the new process.
467 	 */
468 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
469 
470 	/*
471 	 * Update stats now that we know the fork was successful.
472 	 */
473 	uvmexp.forks++;
474 	if (flags & FORK_PPWAIT)
475 		uvmexp.forks_ppwait++;
476 	if (flags & FORK_SHAREVM)
477 		uvmexp.forks_sharevm++;
478 
479 	/*
480 	 * Pass a pointer to the new process to the caller.
481 	 */
482 	if (rnewprocp != NULL)
483 		*rnewprocp = p2;
484 
485 	if (ktrpoint(KTR_EMUL))
486 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
487 
488 	/*
489 	 * Make child runnable, set start time, and add to run queue except
490 	 * if the parent requested the child to start in SSTOP state.
491 	 */
492 	tmp = (p2->p_userret != NULL ? LW_WUSERRET : 0);
493 	mutex_enter(&proclist_mutex);
494 	mutex_enter(&p2->p_smutex);
495 
496 	getmicrotime(&p2->p_stats->p_start);
497 	p2->p_acflag = AFORK;
498 	if (p2->p_sflag & PS_STOPFORK) {
499 		lwp_lock(l2);
500 		p2->p_nrlwps = 0;
501 		p2->p_stat = SSTOP;
502 		p2->p_waited = 0;
503 		p1->p_nstopchild++;
504 		l2->l_stat = LSSTOP;
505 		l2->l_flag |= tmp;
506 		lwp_unlock(l2);
507 	} else {
508 		p2->p_nrlwps = 1;
509 		p2->p_stat = SACTIVE;
510 		lwp_lock(l2);
511 		l2->l_stat = LSRUN;
512 		l2->l_flag |= tmp;
513 		sched_enqueue(l2, false);
514 		lwp_unlock(l2);
515 	}
516 
517 	mutex_exit(&proclist_mutex);
518 
519 	/*
520 	 * Start profiling.
521 	 */
522 	if ((p2->p_stflag & PST_PROFIL) != 0) {
523 		mutex_spin_enter(&p2->p_stmutex);
524 		startprofclock(p2);
525 		mutex_spin_exit(&p2->p_stmutex);
526 	}
527 
528 	/*
529 	 * Preserve synchronization semantics of vfork.  If waiting for
530 	 * child to exec or exit, set PS_PPWAIT on child, and sleep on our
531 	 * proc (in case of exit).
532 	 */
533 	if (flags & FORK_PPWAIT)
534 		while (p2->p_sflag & PS_PPWAIT)
535 			cv_wait(&p1->p_waitcv, &p2->p_smutex);
536 
537 	mutex_exit(&p2->p_smutex);
538 
539 	/*
540 	 * Return child pid to parent process,
541 	 * marking us as parent via retval[1].
542 	 */
543 	if (retval != NULL) {
544 		retval[0] = p2->p_pid;
545 		retval[1] = 0;
546 	}
547 
548 	return (0);
549 }
550