xref: /netbsd-src/sys/kern/kern_fork.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: kern_fork.c,v 1.119 2004/09/17 23:20:21 enami Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2001 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  *
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 1.119 2004/09/17 23:20:21 enami Exp $");
78 
79 #include "opt_ktrace.h"
80 #include "opt_systrace.h"
81 #include "opt_multiprocessor.h"
82 
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/filedesc.h>
86 #include <sys/kernel.h>
87 #include <sys/malloc.h>
88 #include <sys/pool.h>
89 #include <sys/mount.h>
90 #include <sys/proc.h>
91 #include <sys/ras.h>
92 #include <sys/resourcevar.h>
93 #include <sys/vnode.h>
94 #include <sys/file.h>
95 #include <sys/acct.h>
96 #include <sys/ktrace.h>
97 #include <sys/vmmeter.h>
98 #include <sys/sched.h>
99 #include <sys/signalvar.h>
100 #include <sys/systrace.h>
101 
102 #include <sys/sa.h>
103 #include <sys/syscallargs.h>
104 
105 #include <uvm/uvm_extern.h>
106 
107 
108 int	nprocs = 1;		/* process 0 */
109 
110 /*
111  * Number of ticks to sleep if fork() would fail due to process hitting
112  * limits. Exported in miliseconds to userland via sysctl.
113  */
114 int	forkfsleep = 0;
115 
116 /*ARGSUSED*/
117 int
118 sys_fork(struct lwp *l, void *v, register_t *retval)
119 {
120 
121 	return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
122 }
123 
124 /*
125  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
126  * Address space is not shared, but parent is blocked until child exit.
127  */
128 /*ARGSUSED*/
129 int
130 sys_vfork(struct lwp *l, void *v, register_t *retval)
131 {
132 
133 	return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
134 	    retval, NULL));
135 }
136 
137 /*
138  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
139  * semantics.  Address space is shared, and parent is blocked until child exit.
140  */
141 /*ARGSUSED*/
142 int
143 sys___vfork14(struct lwp *l, void *v, register_t *retval)
144 {
145 
146 	return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
147 	    NULL, NULL, retval, NULL));
148 }
149 
150 /*
151  * Linux-compatible __clone(2) system call.
152  */
153 int
154 sys___clone(struct lwp *l, void *v, register_t *retval)
155 {
156 	struct sys___clone_args /* {
157 		syscallarg(int) flags;
158 		syscallarg(void *) stack;
159 	} */ *uap = v;
160 	int flags, sig;
161 
162 	/*
163 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
164 	 */
165 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
166 		return (EINVAL);
167 
168 	/*
169 	 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
170 	 */
171 	if (SCARG(uap, flags) & CLONE_SIGHAND
172 	    && (SCARG(uap, flags) & CLONE_VM) == 0)
173 		return (EINVAL);
174 
175 	flags = 0;
176 
177 	if (SCARG(uap, flags) & CLONE_VM)
178 		flags |= FORK_SHAREVM;
179 	if (SCARG(uap, flags) & CLONE_FS)
180 		flags |= FORK_SHARECWD;
181 	if (SCARG(uap, flags) & CLONE_FILES)
182 		flags |= FORK_SHAREFILES;
183 	if (SCARG(uap, flags) & CLONE_SIGHAND)
184 		flags |= FORK_SHARESIGS;
185 	if (SCARG(uap, flags) & CLONE_VFORK)
186 		flags |= FORK_PPWAIT;
187 
188 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
189 	if (sig < 0 || sig >= _NSIG)
190 		return (EINVAL);
191 
192 	/*
193 	 * Note that the Linux API does not provide a portable way of
194 	 * specifying the stack area; the caller must know if the stack
195 	 * grows up or down.  So, we pass a stack size of 0, so that the
196 	 * code that makes this adjustment is a noop.
197 	 */
198 	return (fork1(l, flags, sig, SCARG(uap, stack), 0,
199 	    NULL, NULL, retval, NULL));
200 }
201 
202 /* print the 'table full' message once per 10 seconds */
203 struct timeval fork_tfmrate = { 10, 0 };
204 
205 int
206 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
207     void (*func)(void *), void *arg, register_t *retval,
208     struct proc **rnewprocp)
209 {
210 	struct proc	*p1, *p2, *parent;
211 	uid_t		uid;
212 	struct lwp	*l2;
213 	int		count, s;
214 	vaddr_t		uaddr;
215 	boolean_t	inmem;
216 
217 	/*
218 	 * Although process entries are dynamically created, we still keep
219 	 * a global limit on the maximum number we will create.  Don't allow
220 	 * a nonprivileged user to use the last few processes; don't let root
221 	 * exceed the limit. The variable nprocs is the current number of
222 	 * processes, maxproc is the limit.
223 	 */
224 	p1 = l1->l_proc;
225 	uid = p1->p_cred->p_ruid;
226 	if (__predict_false((nprocs >= maxproc - 5 && uid != 0) ||
227 			    nprocs >= maxproc)) {
228 		static struct timeval lasttfm;
229 
230 		if (ratecheck(&lasttfm, &fork_tfmrate))
231 			tablefull("proc", "increase kern.maxproc or NPROC");
232 		if (forkfsleep)
233 			(void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep);
234 		return (EAGAIN);
235 	}
236 	nprocs++;
237 
238 	/*
239 	 * Increment the count of procs running with this uid. Don't allow
240 	 * a nonprivileged user to exceed their current limit.
241 	 */
242 	count = chgproccnt(uid, 1);
243 	if (__predict_false(uid != 0 && count >
244 			    p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
245 		(void)chgproccnt(uid, -1);
246 		nprocs--;
247 		if (forkfsleep)
248 			(void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep);
249 		return (EAGAIN);
250 	}
251 
252 	/*
253 	 * Allocate virtual address space for the U-area now, while it
254 	 * is still easy to abort the fork operation if we're out of
255 	 * kernel virtual address space.  The actual U-area pages will
256 	 * be allocated and wired in uvm_fork() if needed.
257 	 */
258 
259 	inmem = uvm_uarea_alloc(&uaddr);
260 	if (__predict_false(uaddr == 0)) {
261 		(void)chgproccnt(uid, -1);
262 		nprocs--;
263 		return (ENOMEM);
264 	}
265 
266 	/*
267 	 * We are now committed to the fork.  From here on, we may
268 	 * block on resources, but resource allocation may NOT fail.
269 	 */
270 
271 	/* Allocate new proc. */
272 	p2 = proc_alloc();
273 
274 	/*
275 	 * Make a proc table entry for the new process.
276 	 * Start by zeroing the section of proc that is zero-initialized,
277 	 * then copy the section that is copied directly from the parent.
278 	 */
279 	memset(&p2->p_startzero, 0,
280 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
281 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
282 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
283 
284 	simple_lock_init(&p2->p_sigctx.ps_silock);
285 	CIRCLEQ_INIT(&p2->p_sigctx.ps_siginfo);
286 	simple_lock_init(&p2->p_lock);
287 	LIST_INIT(&p2->p_lwps);
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 uvm_fork().
293 	 * Inherit SUGID, STOPFORK and STOPEXEC flags.
294 	 */
295 	p2->p_flag = p1->p_flag & (P_SUGID | P_STOPFORK | P_STOPEXEC);
296 	p2->p_emul = p1->p_emul;
297 	p2->p_execsw = p1->p_execsw;
298 
299 	if (p1->p_flag & P_PROFIL)
300 		startprofclock(p2);
301 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
302 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
303 	p2->p_cred->p_refcnt = 1;
304 	crhold(p1->p_ucred);
305 
306 	LIST_INIT(&p2->p_raslist);
307 #if defined(__HAVE_RAS)
308 	ras_fork(p1, p2);
309 #endif
310 
311 	/* bump references to the text vnode (for procfs) */
312 	p2->p_textvp = p1->p_textvp;
313 	if (p2->p_textvp)
314 		VREF(p2->p_textvp);
315 
316 	if (flags & FORK_SHAREFILES)
317 		fdshare(p1, p2);
318 	else if (flags & FORK_CLEANFILES)
319 		p2->p_fd = fdinit(p1);
320 	else
321 		p2->p_fd = fdcopy(p1);
322 
323 	if (flags & FORK_SHARECWD)
324 		cwdshare(p1, p2);
325 	else
326 		p2->p_cwdi = cwdinit(p1);
327 
328 	/*
329 	 * If p_limit is still copy-on-write, bump refcnt,
330 	 * otherwise get a copy that won't be modified.
331 	 * (If PL_SHAREMOD is clear, the structure is shared
332 	 * copy-on-write.)
333 	 */
334 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
335 		p2->p_limit = limcopy(p1->p_limit);
336 	else {
337 		simple_lock(&p1->p_limit->p_slock);
338 		p1->p_limit->p_refcnt++;
339 		simple_unlock(&p1->p_limit->p_slock);
340 		p2->p_limit = p1->p_limit;
341 	}
342 
343 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
344 		p2->p_flag |= P_CONTROLT;
345 	if (flags & FORK_PPWAIT)
346 		p2->p_flag |= P_PPWAIT;
347 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
348 	p2->p_pptr = parent;
349 	LIST_INIT(&p2->p_children);
350 
351 	s = proclist_lock_write();
352 	LIST_INSERT_AFTER(p1, p2, p_pglist);
353 	LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
354 	proclist_unlock_write(s);
355 
356 #ifdef KTRACE
357 	/*
358 	 * Copy traceflag and tracefile if enabled.
359 	 * If not inherited, these were zeroed above.
360 	 */
361 	if (p1->p_traceflag & KTRFAC_INHERIT) {
362 		p2->p_traceflag = p1->p_traceflag;
363 		if ((p2->p_tracep = p1->p_tracep) != NULL)
364 			ktradref(p2);
365 	}
366 #endif
367 
368 	scheduler_fork_hook(p1, p2);
369 
370 	/*
371 	 * Create signal actions for the child process.
372 	 */
373 	sigactsinit(p2, p1, flags & FORK_SHARESIGS);
374 
375 	/*
376 	 * p_stats.
377 	 * Copy parts of p_stats, and zero out the rest.
378 	 */
379 	p2->p_stats = pstatscopy(p1->p_stats);
380 
381 	/*
382 	 * If emulation has process fork hook, call it now.
383 	 */
384 	if (p2->p_emul->e_proc_fork)
385 		(*p2->p_emul->e_proc_fork)(p2, p1, flags);
386 
387 	/*
388 	 * ...and finally, any other random fork hooks that subsystems
389 	 * might have registered.
390 	 */
391 	doforkhooks(p2, p1);
392 
393 	/*
394 	 * This begins the section where we must prevent the parent
395 	 * from being swapped.
396 	 */
397 	PHOLD(l1);
398 
399 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
400 
401 	/*
402 	 * Finish creating the child process.
403 	 * It will return through a different path later.
404 	 */
405 	newlwp(l1, p2, uaddr, inmem, 0, stack, stacksize,
406 	    (func != NULL) ? func : child_return,
407 	    arg, &l2);
408 
409 	/* Now safe for scheduler to see child process */
410 	s = proclist_lock_write();
411 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
412 	LIST_INSERT_HEAD(&allproc, p2, p_list);
413 	proclist_unlock_write(s);
414 
415 #ifdef SYSTRACE
416 	/* Tell systrace what's happening. */
417 	if (ISSET(p1->p_flag, P_SYSTRACE))
418 		systrace_sys_fork(p1, p2);
419 #endif
420 
421 #ifdef __HAVE_SYSCALL_INTERN
422 	(*p2->p_emul->e_syscall_intern)(p2);
423 #endif
424 
425 	/*
426 	 * Make child runnable, set start time, and add to run queue
427 	 * except if the parent requested the child to start in SSTOP state.
428 	 */
429 	SCHED_LOCK(s);
430 	p2->p_stats->p_start = time;
431 	p2->p_acflag = AFORK;
432 	if (p1->p_flag & P_STOPFORK) {
433 		p2->p_nrlwps = 0;
434 		p1->p_nstopchild++;
435 		p2->p_stat = SSTOP;
436 		l2->l_stat = LSSTOP;
437 	} else {
438 		p2->p_nrlwps = 1;
439 		p2->p_stat = SACTIVE;
440 		l2->l_stat = LSRUN;
441 		setrunqueue(l2);
442 	}
443 	SCHED_UNLOCK(s);
444 
445 	/*
446 	 * Now can be swapped.
447 	 */
448 	PRELE(l1);
449 
450 	/*
451 	 * Notify any interested parties about the new process.
452 	 */
453 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
454 
455 	/*
456 	 * Update stats now that we know the fork was successful.
457 	 */
458 	uvmexp.forks++;
459 	if (flags & FORK_PPWAIT)
460 		uvmexp.forks_ppwait++;
461 	if (flags & FORK_SHAREVM)
462 		uvmexp.forks_sharevm++;
463 
464 	/*
465 	 * Pass a pointer to the new process to the caller.
466 	 */
467 	if (rnewprocp != NULL)
468 		*rnewprocp = p2;
469 
470 #ifdef KTRACE
471 	if (KTRPOINT(p2, KTR_EMUL))
472 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
473 #endif
474 
475 	/*
476 	 * Preserve synchronization semantics of vfork.  If waiting for
477 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
478 	 * proc (in case of exit).
479 	 */
480 	if (flags & FORK_PPWAIT)
481 		while (p2->p_flag & P_PPWAIT)
482 			tsleep(p1, PWAIT, "ppwait", 0);
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 
493 	return (0);
494 }
495 
496 #if defined(MULTIPROCESSOR)
497 /*
498  * XXX This is a slight hack to get newly-formed processes to
499  * XXX acquire the kernel lock as soon as they run.
500  */
501 void
502 proc_trampoline_mp(void)
503 {
504 	struct lwp *l;
505 
506 	l = curlwp;
507 
508 	SCHED_ASSERT_UNLOCKED();
509 	KERNEL_PROC_LOCK(l);
510 }
511 #endif
512