xref: /netbsd-src/sys/kern/kern_fork.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: kern_fork.c,v 1.125 2006/06/07 22:33:39 kardel Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2001, 2004 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.
10  * This code is derived from software contributed to The NetBSD Foundation
11  * by Charles M. Hannum.
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. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by the NetBSD
24  *	Foundation, Inc. and its contributors.
25  * 4. Neither the name of The NetBSD Foundation nor the names of its
26  *    contributors may be used to endorse or promote products derived
27  *    from this software without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
30  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
31  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
32  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
33  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
34  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
35  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
36  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
37  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
38  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
39  * POSSIBILITY OF SUCH DAMAGE.
40  */
41 
42 /*
43  * Copyright (c) 1982, 1986, 1989, 1991, 1993
44  *	The Regents of the University of California.  All rights reserved.
45  * (c) UNIX System Laboratories, Inc.
46  * All or some portions of this file are derived from material licensed
47  * to the University of California by American Telephone and Telegraph
48  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
49  * the permission of UNIX System Laboratories, Inc.
50  *
51  * Redistribution and use in source and binary forms, with or without
52  * modification, are permitted provided that the following conditions
53  * are met:
54  * 1. Redistributions of source code must retain the above copyright
55  *    notice, this list of conditions and the following disclaimer.
56  * 2. Redistributions in binary form must reproduce the above copyright
57  *    notice, this list of conditions and the following disclaimer in the
58  *    documentation and/or other materials provided with the distribution.
59  * 3. Neither the name of the University nor the names of its contributors
60  *    may be used to endorse or promote products derived from this software
61  *    without specific prior written permission.
62  *
63  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
64  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
65  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
66  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
67  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
68  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
69  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
70  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
71  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
72  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
73  * SUCH DAMAGE.
74  *
75  *	@(#)kern_fork.c	8.8 (Berkeley) 2/14/95
76  */
77 
78 #include <sys/cdefs.h>
79 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.125 2006/06/07 22:33:39 kardel Exp $");
80 
81 #include "opt_ktrace.h"
82 #include "opt_systrace.h"
83 #include "opt_multiprocessor.h"
84 
85 #include <sys/param.h>
86 #include <sys/systm.h>
87 #include <sys/filedesc.h>
88 #include <sys/kernel.h>
89 #include <sys/malloc.h>
90 #include <sys/pool.h>
91 #include <sys/mount.h>
92 #include <sys/proc.h>
93 #include <sys/ras.h>
94 #include <sys/resourcevar.h>
95 #include <sys/vnode.h>
96 #include <sys/file.h>
97 #include <sys/acct.h>
98 #include <sys/ktrace.h>
99 #include <sys/vmmeter.h>
100 #include <sys/sched.h>
101 #include <sys/signalvar.h>
102 #include <sys/systrace.h>
103 #include <sys/kauth.h>
104 
105 #include <sys/sa.h>
106 #include <sys/syscallargs.h>
107 
108 #include <uvm/uvm_extern.h>
109 
110 
111 int	nprocs = 1;		/* process 0 */
112 
113 /*
114  * Number of ticks to sleep if fork() would fail due to process hitting
115  * limits. Exported in miliseconds to userland via sysctl.
116  */
117 int	forkfsleep = 0;
118 
119 /*ARGSUSED*/
120 int
121 sys_fork(struct lwp *l, void *v, register_t *retval)
122 {
123 
124 	return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
125 }
126 
127 /*
128  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
129  * Address space is not shared, but parent is blocked until child exit.
130  */
131 /*ARGSUSED*/
132 int
133 sys_vfork(struct lwp *l, void *v, register_t *retval)
134 {
135 
136 	return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
137 	    retval, NULL));
138 }
139 
140 /*
141  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
142  * semantics.  Address space is shared, and parent is blocked until child exit.
143  */
144 /*ARGSUSED*/
145 int
146 sys___vfork14(struct lwp *l, void *v, register_t *retval)
147 {
148 
149 	return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
150 	    NULL, NULL, retval, NULL));
151 }
152 
153 /*
154  * Linux-compatible __clone(2) system call.
155  */
156 int
157 sys___clone(struct lwp *l, void *v, register_t *retval)
158 {
159 	struct sys___clone_args /* {
160 		syscallarg(int) flags;
161 		syscallarg(void *) stack;
162 	} */ *uap = v;
163 	int flags, sig;
164 
165 	/*
166 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
167 	 */
168 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
169 		return (EINVAL);
170 
171 	/*
172 	 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
173 	 */
174 	if (SCARG(uap, flags) & CLONE_SIGHAND
175 	    && (SCARG(uap, flags) & CLONE_VM) == 0)
176 		return (EINVAL);
177 
178 	flags = 0;
179 
180 	if (SCARG(uap, flags) & CLONE_VM)
181 		flags |= FORK_SHAREVM;
182 	if (SCARG(uap, flags) & CLONE_FS)
183 		flags |= FORK_SHARECWD;
184 	if (SCARG(uap, flags) & CLONE_FILES)
185 		flags |= FORK_SHAREFILES;
186 	if (SCARG(uap, flags) & CLONE_SIGHAND)
187 		flags |= FORK_SHARESIGS;
188 	if (SCARG(uap, flags) & CLONE_VFORK)
189 		flags |= FORK_PPWAIT;
190 
191 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
192 	if (sig < 0 || sig >= _NSIG)
193 		return (EINVAL);
194 
195 	/*
196 	 * Note that the Linux API does not provide a portable way of
197 	 * specifying the stack area; the caller must know if the stack
198 	 * grows up or down.  So, we pass a stack size of 0, so that the
199 	 * code that makes this adjustment is a noop.
200 	 */
201 	return (fork1(l, flags, sig, SCARG(uap, stack), 0,
202 	    NULL, NULL, retval, NULL));
203 }
204 
205 /* print the 'table full' message once per 10 seconds */
206 struct timeval fork_tfmrate = { 10, 0 };
207 
208 int
209 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
210     void (*func)(void *), void *arg, register_t *retval,
211     struct proc **rnewprocp)
212 {
213 	struct proc	*p1, *p2, *parent;
214 	uid_t		uid;
215 	struct lwp	*l2;
216 	int		count, s;
217 	vaddr_t		uaddr;
218 	boolean_t	inmem;
219 
220 	/*
221 	 * Although process entries are dynamically created, we still keep
222 	 * a global limit on the maximum number we will create.  Don't allow
223 	 * a nonprivileged user to use the last few processes; don't let root
224 	 * exceed the limit. The variable nprocs is the current number of
225 	 * processes, maxproc is the limit.
226 	 */
227 	p1 = l1->l_proc;
228 	uid = kauth_cred_getuid(p1->p_cred);
229 	if (__predict_false((nprocs >= maxproc - 5 && uid != 0) ||
230 			    nprocs >= maxproc)) {
231 		static struct timeval lasttfm;
232 
233 		if (ratecheck(&lasttfm, &fork_tfmrate))
234 			tablefull("proc", "increase kern.maxproc or NPROC");
235 		if (forkfsleep)
236 			(void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep);
237 		return (EAGAIN);
238 	}
239 	nprocs++;
240 
241 	/*
242 	 * Increment the count of procs running with this uid. Don't allow
243 	 * a nonprivileged user to exceed their current limit.
244 	 */
245 	count = chgproccnt(uid, 1);
246 	if (__predict_false(uid != 0 && count >
247 			    p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
248 		(void)chgproccnt(uid, -1);
249 		nprocs--;
250 		if (forkfsleep)
251 			(void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep);
252 		return (EAGAIN);
253 	}
254 
255 	/*
256 	 * Allocate virtual address space for the U-area now, while it
257 	 * is still easy to abort the fork operation if we're out of
258 	 * kernel virtual address space.  The actual U-area pages will
259 	 * be allocated and wired in uvm_fork() if needed.
260 	 */
261 
262 	inmem = uvm_uarea_alloc(&uaddr);
263 	if (__predict_false(uaddr == 0)) {
264 		(void)chgproccnt(uid, -1);
265 		nprocs--;
266 		return (ENOMEM);
267 	}
268 
269 	/*
270 	 * We are now committed to the fork.  From here on, we may
271 	 * block on resources, but resource allocation may NOT fail.
272 	 */
273 
274 	/* Allocate new proc. */
275 	p2 = proc_alloc();
276 
277 	/*
278 	 * Make a proc table entry for the new process.
279 	 * Start by zeroing the section of proc that is zero-initialized,
280 	 * then copy the section that is copied directly from the parent.
281 	 */
282 	memset(&p2->p_startzero, 0,
283 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
284 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
285 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
286 
287 	simple_lock_init(&p2->p_sigctx.ps_silock);
288 	CIRCLEQ_INIT(&p2->p_sigctx.ps_siginfo);
289 	simple_lock_init(&p2->p_lock);
290 	LIST_INIT(&p2->p_lwps);
291 
292 	/*
293 	 * Duplicate sub-structures as needed.
294 	 * Increase reference counts on shared objects.
295 	 * The p_stats and p_sigacts substructs are set in uvm_fork().
296 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
297 	 * handling are important in order to keep a consistent behaviour
298 	 * for the child after the fork.
299 	 */
300 	p2->p_flag = p1->p_flag & (P_SUGID | P_STOPFORK | P_STOPEXEC |
301 	    P_NOCLDSTOP | P_NOCLDWAIT | P_CLDSIGIGN);
302 	p2->p_emul = p1->p_emul;
303 	p2->p_execsw = p1->p_execsw;
304 
305 	if (p1->p_flag & P_PROFIL)
306 		startprofclock(p2);
307 
308 	p2->p_cred = kauth_cred_alloc();
309 	kauth_cred_clone(p1->p_cred, p2->p_cred);
310 
311 	LIST_INIT(&p2->p_raslist);
312 #if defined(__HAVE_RAS)
313 	ras_fork(p1, p2);
314 #endif
315 
316 	/* bump references to the text vnode (for procfs) */
317 	p2->p_textvp = p1->p_textvp;
318 	if (p2->p_textvp)
319 		VREF(p2->p_textvp);
320 
321 	if (flags & FORK_SHAREFILES)
322 		fdshare(p1, p2);
323 	else if (flags & FORK_CLEANFILES)
324 		p2->p_fd = fdinit(p1);
325 	else
326 		p2->p_fd = fdcopy(p1);
327 
328 	if (flags & FORK_SHARECWD)
329 		cwdshare(p1, p2);
330 	else
331 		p2->p_cwdi = cwdinit(p1);
332 
333 	/*
334 	 * If p_limit is still copy-on-write, bump refcnt,
335 	 * otherwise get a copy that won't be modified.
336 	 * (If PL_SHAREMOD is clear, the structure is shared
337 	 * copy-on-write.)
338 	 */
339 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
340 		p2->p_limit = limcopy(p1->p_limit);
341 	else {
342 		simple_lock(&p1->p_limit->p_slock);
343 		p1->p_limit->p_refcnt++;
344 		simple_unlock(&p1->p_limit->p_slock);
345 		p2->p_limit = p1->p_limit;
346 	}
347 
348 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
349 		p2->p_flag |= P_CONTROLT;
350 	if (flags & FORK_PPWAIT)
351 		p2->p_flag |= P_PPWAIT;
352 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
353 	p2->p_pptr = parent;
354 	LIST_INIT(&p2->p_children);
355 
356 	s = proclist_lock_write();
357 	LIST_INSERT_AFTER(p1, p2, p_pglist);
358 	LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
359 	proclist_unlock_write(s);
360 
361 #ifdef KTRACE
362 	/*
363 	 * Copy traceflag and tracefile if enabled.
364 	 * If not inherited, these were zeroed above.
365 	 */
366 	if (p1->p_traceflag & KTRFAC_INHERIT) {
367 		p2->p_traceflag = p1->p_traceflag;
368 		if ((p2->p_tracep = p1->p_tracep) != NULL)
369 			ktradref(p2);
370 	}
371 #endif
372 
373 	scheduler_fork_hook(p1, p2);
374 
375 	/*
376 	 * Create signal actions for the child process.
377 	 */
378 	sigactsinit(p2, p1, flags & FORK_SHARESIGS);
379 
380 	/*
381 	 * p_stats.
382 	 * Copy parts of p_stats, and zero out the rest.
383 	 */
384 	p2->p_stats = pstatscopy(p1->p_stats);
385 
386 	/*
387 	 * If emulation has process fork hook, call it now.
388 	 */
389 	if (p2->p_emul->e_proc_fork)
390 		(*p2->p_emul->e_proc_fork)(p2, p1, flags);
391 
392 	/*
393 	 * ...and finally, any other random fork hooks that subsystems
394 	 * might have registered.
395 	 */
396 	doforkhooks(p2, p1);
397 
398 	/*
399 	 * This begins the section where we must prevent the parent
400 	 * from being swapped.
401 	 */
402 	PHOLD(l1);
403 
404 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
405 
406 	/*
407 	 * Finish creating the child process.
408 	 * It will return through a different path later.
409 	 */
410 	newlwp(l1, p2, uaddr, inmem, 0, stack, stacksize,
411 	    (func != NULL) ? func : child_return,
412 	    arg, &l2);
413 
414 	/* Now safe for scheduler to see child process */
415 	s = proclist_lock_write();
416 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
417 	LIST_INSERT_HEAD(&allproc, p2, p_list);
418 	proclist_unlock_write(s);
419 
420 #ifdef SYSTRACE
421 	/* Tell systrace what's happening. */
422 	if (ISSET(p1->p_flag, P_SYSTRACE))
423 		systrace_sys_fork(p1, p2);
424 #endif
425 
426 #ifdef __HAVE_SYSCALL_INTERN
427 	(*p2->p_emul->e_syscall_intern)(p2);
428 #endif
429 
430 	/*
431 	 * Make child runnable, set start time, and add to run queue
432 	 * except if the parent requested the child to start in SSTOP state.
433 	 */
434 	SCHED_LOCK(s);
435 	getmicrotime(&p2->p_stats->p_start);
436 	p2->p_acflag = AFORK;
437 	if (p1->p_flag & P_STOPFORK) {
438 		p2->p_nrlwps = 0;
439 		p1->p_nstopchild++;
440 		p2->p_stat = SSTOP;
441 		l2->l_stat = LSSTOP;
442 	} else {
443 		p2->p_nrlwps = 1;
444 		p2->p_stat = SACTIVE;
445 		l2->l_stat = LSRUN;
446 		setrunqueue(l2);
447 	}
448 	SCHED_UNLOCK(s);
449 
450 	/*
451 	 * Now can be swapped.
452 	 */
453 	PRELE(l1);
454 
455 	/*
456 	 * Notify any interested parties about the new process.
457 	 */
458 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
459 
460 	/*
461 	 * Update stats now that we know the fork was successful.
462 	 */
463 	uvmexp.forks++;
464 	if (flags & FORK_PPWAIT)
465 		uvmexp.forks_ppwait++;
466 	if (flags & FORK_SHAREVM)
467 		uvmexp.forks_sharevm++;
468 
469 	/*
470 	 * Pass a pointer to the new process to the caller.
471 	 */
472 	if (rnewprocp != NULL)
473 		*rnewprocp = p2;
474 
475 #ifdef KTRACE
476 	if (KTRPOINT(p2, KTR_EMUL))
477 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
478 #endif
479 
480 	/*
481 	 * Preserve synchronization semantics of vfork.  If waiting for
482 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
483 	 * proc (in case of exit).
484 	 */
485 	if (flags & FORK_PPWAIT)
486 		while (p2->p_flag & P_PPWAIT)
487 			tsleep(p1, PWAIT, "ppwait", 0);
488 
489 	/*
490 	 * Return child pid to parent process,
491 	 * marking us as parent via retval[1].
492 	 */
493 	if (retval != NULL) {
494 		retval[0] = p2->p_pid;
495 		retval[1] = 0;
496 	}
497 
498 	return (0);
499 }
500 
501 #if defined(MULTIPROCESSOR)
502 /*
503  * XXX This is a slight hack to get newly-formed processes to
504  * XXX acquire the kernel lock as soon as they run.
505  */
506 void
507 proc_trampoline_mp(void)
508 {
509 	struct lwp *l;
510 
511 	l = curlwp;
512 
513 	SCHED_ASSERT_UNLOCKED();
514 	KERNEL_PROC_LOCK(l);
515 }
516 #endif
517