xref: /netbsd-src/sys/kern/kern_fork.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: kern_fork.c,v 1.109 2003/06/29 22:31:20 fvdl 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. All advertising materials mentioning features or use of this software
58  *    must display the following acknowledgement:
59  *	This product includes software developed by the University of
60  *	California, Berkeley and its contributors.
61  * 4. Neither the name of the University nor the names of its contributors
62  *    may be used to endorse or promote products derived from this software
63  *    without specific prior written permission.
64  *
65  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
66  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
67  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
68  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
69  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
70  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
71  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
72  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
73  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
74  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
75  * SUCH DAMAGE.
76  *
77  *	@(#)kern_fork.c	8.8 (Berkeley) 2/14/95
78  */
79 
80 #include <sys/cdefs.h>
81 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.109 2003/06/29 22:31:20 fvdl Exp $");
82 
83 #include "opt_ktrace.h"
84 #include "opt_systrace.h"
85 #include "opt_multiprocessor.h"
86 
87 #include <sys/param.h>
88 #include <sys/systm.h>
89 #include <sys/filedesc.h>
90 #include <sys/kernel.h>
91 #include <sys/malloc.h>
92 #include <sys/pool.h>
93 #include <sys/mount.h>
94 #include <sys/proc.h>
95 #include <sys/ras.h>
96 #include <sys/resourcevar.h>
97 #include <sys/vnode.h>
98 #include <sys/file.h>
99 #include <sys/acct.h>
100 #include <sys/ktrace.h>
101 #include <sys/vmmeter.h>
102 #include <sys/sched.h>
103 #include <sys/signalvar.h>
104 #include <sys/systrace.h>
105 
106 #include <sys/sa.h>
107 #include <sys/syscallargs.h>
108 
109 #include <uvm/uvm_extern.h>
110 
111 
112 int	nprocs = 1;		/* process 0 */
113 
114 /*
115  * Number of ticks to sleep if fork() would fail due to process hitting
116  * limits. Exported in miliseconds to userland via sysctl.
117  */
118 int	forkfsleep = 0;
119 
120 /*ARGSUSED*/
121 int
122 sys_fork(struct lwp *l, void *v, register_t *retval)
123 {
124 
125 	return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
126 }
127 
128 /*
129  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
130  * Address space is not shared, but parent is blocked until child exit.
131  */
132 /*ARGSUSED*/
133 int
134 sys_vfork(struct lwp *l, void *v, register_t *retval)
135 {
136 
137 	return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
138 	    retval, NULL));
139 }
140 
141 /*
142  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
143  * semantics.  Address space is shared, and parent is blocked until child exit.
144  */
145 /*ARGSUSED*/
146 int
147 sys___vfork14(struct lwp *l, void *v, register_t *retval)
148 {
149 
150 	return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
151 	    NULL, NULL, retval, NULL));
152 }
153 
154 /*
155  * Linux-compatible __clone(2) system call.
156  */
157 int
158 sys___clone(struct lwp *l, void *v, register_t *retval)
159 {
160 	struct sys___clone_args /* {
161 		syscallarg(int) flags;
162 		syscallarg(void *) stack;
163 	} */ *uap = v;
164 	int flags, sig;
165 
166 	/*
167 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
168 	 */
169 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
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 int
203 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
204     void (*func)(void *), void *arg, register_t *retval,
205     struct proc **rnewprocp)
206 {
207 	struct proc	*p1, *p2;
208 	uid_t		uid;
209 	struct lwp	*l2;
210 	int		count, s;
211 	vaddr_t		uaddr;
212 	boolean_t	inmem;
213 
214 	/*
215 	 * Although process entries are dynamically created, we still keep
216 	 * a global limit on the maximum number we will create.  Don't allow
217 	 * a nonprivileged user to use the last few processes; don't let root
218 	 * exceed the limit. The variable nprocs is the current number of
219 	 * processes, maxproc is the limit.
220 	 */
221 	p1 = l1->l_proc;
222 	uid = p1->p_cred->p_ruid;
223 	if (__predict_false((nprocs >= maxproc - 5 && uid != 0) ||
224 			    nprocs >= maxproc)) {
225 		static struct timeval lasttfm;
226 
227 		if (ratecheck(&lasttfm, &fork_tfmrate))
228 			tablefull("proc", "increase kern.maxproc or NPROC");
229 		if (forkfsleep)
230 			(void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep);
231 		return (EAGAIN);
232 	}
233 	nprocs++;
234 
235 	/*
236 	 * Increment the count of procs running with this uid. Don't allow
237 	 * a nonprivileged user to exceed their current limit.
238 	 */
239 	count = chgproccnt(uid, 1);
240 	if (__predict_false(uid != 0 && count >
241 			    p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
242 		(void)chgproccnt(uid, -1);
243 		nprocs--;
244 		if (forkfsleep)
245 			(void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep);
246 		return (EAGAIN);
247 	}
248 
249 	/*
250 	 * Allocate virtual address space for the U-area now, while it
251 	 * is still easy to abort the fork operation if we're out of
252 	 * kernel virtual address space.  The actual U-area pages will
253 	 * be allocated and wired in uvm_fork() if needed.
254 	 */
255 
256 	inmem = uvm_uarea_alloc(&uaddr);
257 	if (__predict_false(uaddr == 0)) {
258 		(void)chgproccnt(uid, -1);
259 		nprocs--;
260 		return (ENOMEM);
261 	}
262 
263 	/*
264 	 * We are now committed to the fork.  From here on, we may
265 	 * block on resources, but resource allocation may NOT fail.
266 	 */
267 
268 	/* Allocate new proc. */
269 	p2 = proc_alloc();
270 
271 	/*
272 	 * Make a proc table entry for the new process.
273 	 * Start by zeroing the section of proc that is zero-initialized,
274 	 * then copy the section that is copied directly from the parent.
275 	 */
276 	memset(&p2->p_startzero, 0,
277 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
278 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
279 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
280 
281 	simple_lock_init(&p2->p_lwplock);
282 	LIST_INIT(&p2->p_lwps);
283 
284 	/*
285 	 * Duplicate sub-structures as needed.
286 	 * Increase reference counts on shared objects.
287 	 * The p_stats and p_sigacts substructs are set in uvm_fork().
288 	 */
289 	p2->p_flag = (p1->p_flag & P_SUGID);
290 	p2->p_emul = p1->p_emul;
291 	p2->p_execsw = p1->p_execsw;
292 
293 	if (p1->p_flag & P_PROFIL)
294 		startprofclock(p2);
295 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
296 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
297 	p2->p_cred->p_refcnt = 1;
298 	crhold(p1->p_ucred);
299 
300 	LIST_INIT(&p2->p_raslist);
301 	p2->p_nras = 0;
302 	simple_lock_init(&p2->p_raslock);
303 #if defined(__HAVE_RAS)
304 	ras_fork(p1, p2);
305 #endif
306 
307 	/* bump references to the text vnode (for procfs) */
308 	p2->p_textvp = p1->p_textvp;
309 	if (p2->p_textvp)
310 		VREF(p2->p_textvp);
311 
312 	if (flags & FORK_SHAREFILES)
313 		fdshare(p1, p2);
314 	else if (flags & FORK_CLEANFILES)
315 		p2->p_fd = fdinit(p1);
316 	else
317 		p2->p_fd = fdcopy(p1);
318 
319 	if (flags & FORK_SHARECWD)
320 		cwdshare(p1, p2);
321 	else
322 		p2->p_cwdi = cwdinit(p1);
323 
324 	/*
325 	 * If p_limit is still copy-on-write, bump refcnt,
326 	 * otherwise get a copy that won't be modified.
327 	 * (If PL_SHAREMOD is clear, the structure is shared
328 	 * copy-on-write.)
329 	 */
330 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
331 		p2->p_limit = limcopy(p1->p_limit);
332 	else {
333 		p2->p_limit = p1->p_limit;
334 		p2->p_limit->p_refcnt++;
335 	}
336 
337 	/* Inherit STOPFORK and STOPEXEC flags */
338 	p2->p_flag |= p1->p_flag & (P_STOPFORK | P_STOPEXEC);
339 
340 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
341 		p2->p_flag |= P_CONTROLT;
342 	if (flags & FORK_PPWAIT)
343 		p2->p_flag |= P_PPWAIT;
344 	p2->p_pptr = (flags & FORK_NOWAIT) ? initproc : p1;
345 	LIST_INIT(&p2->p_children);
346 
347 	s = proclist_lock_write();
348 	LIST_INSERT_AFTER(p1, p2, p_pglist);
349 	LIST_INSERT_HEAD(&p2->p_pptr->p_children, p2, p_sibling);
350 	proclist_unlock_write(s);
351 
352 #ifdef KTRACE
353 	/*
354 	 * Copy traceflag and tracefile if enabled.
355 	 * If not inherited, these were zeroed above.
356 	 */
357 	if (p1->p_traceflag & KTRFAC_INHERIT) {
358 		p2->p_traceflag = p1->p_traceflag;
359 		if ((p2->p_tracep = p1->p_tracep) != NULL)
360 			ktradref(p2);
361 	}
362 #endif
363 
364 	scheduler_fork_hook(p1, p2);
365 
366 	/*
367 	 * Create signal actions for the child process.
368 	 */
369 	sigactsinit(p2, p1, flags & FORK_SHARESIGS);
370 
371 	/*
372 	 * p_stats.
373 	 * Copy parts of p_stats, and zero out the rest.
374 	 */
375 	p2->p_stats = pstatscopy(p1->p_stats);
376 
377 	/*
378 	 * If emulation has process fork hook, call it now.
379 	 */
380 	if (p2->p_emul->e_proc_fork)
381 		(*p2->p_emul->e_proc_fork)(p2, p1);
382 
383 	/*
384 	 * ...and finally, any other random fork hooks that subsystems
385 	 * might have registered.
386 	 */
387 	doforkhooks(p2, p1);
388 
389 	/*
390 	 * This begins the section where we must prevent the parent
391 	 * from being swapped.
392 	 */
393 	PHOLD(l1);
394 
395 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
396 
397 	/*
398 	 * Finish creating the child process.
399 	 * It will return through a different path later.
400 	 */
401 	newlwp(l1, p2, uaddr, inmem, 0, stack, stacksize,
402 	    (func != NULL) ? func : child_return,
403 	    arg, &l2);
404 
405 	/* Now safe for scheduler to see child process */
406 	s = proclist_lock_write();
407 	p2->p_stat = SIDL;			/* protect against others */
408 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
409 	LIST_INSERT_HEAD(&allproc, p2, p_list);
410 	proclist_unlock_write(s);
411 
412 #ifdef SYSTRACE
413 	/* Tell systrace what's happening. */
414 	if (ISSET(p1->p_flag, P_SYSTRACE))
415 		systrace_sys_fork(p1, p2);
416 #endif
417 
418 #ifdef __HAVE_SYSCALL_INTERN
419 	(*p2->p_emul->e_syscall_intern)(p2);
420 #endif
421 
422 	/*
423 	 * Make child runnable, set start time, and add to run queue
424 	 * except if the parent requested the child to start in SSTOP state.
425 	 */
426 	SCHED_LOCK(s);
427 	p2->p_stats->p_start = time;
428 	p2->p_acflag = AFORK;
429 	p2->p_nrlwps = 1;
430 	if (p1->p_flag & P_STOPFORK) {
431 		p2->p_stat = SSTOP;
432 		l2->l_stat = LSSTOP;
433 	} else {
434 		p2->p_stat = SACTIVE;
435 		l2->l_stat = LSRUN;
436 		setrunqueue(l2);
437 	}
438 	SCHED_UNLOCK(s);
439 
440 	/*
441 	 * Now can be swapped.
442 	 */
443 	PRELE(l1);
444 
445 	/*
446 	 * Notify any interested parties about the new process.
447 	 */
448 	KNOTE(&p1->p_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 #ifdef KTRACE
466 	if (KTRPOINT(p2, KTR_EMUL))
467 		ktremul(p2);
468 #endif
469 
470 	/*
471 	 * Preserve synchronization semantics of vfork.  If waiting for
472 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
473 	 * proc (in case of exit).
474 	 */
475 	if (flags & FORK_PPWAIT)
476 		while (p2->p_flag & P_PPWAIT)
477 			tsleep(p1, PWAIT, "ppwait", 0);
478 
479 	/*
480 	 * Return child pid to parent process,
481 	 * marking us as parent via retval[1].
482 	 */
483 	if (retval != NULL) {
484 		retval[0] = p2->p_pid;
485 		retval[1] = 0;
486 	}
487 
488 	return (0);
489 }
490 
491 #if defined(MULTIPROCESSOR)
492 /*
493  * XXX This is a slight hack to get newly-formed processes to
494  * XXX acquire the kernel lock as soon as they run.
495  */
496 void
497 proc_trampoline_mp(void)
498 {
499 	struct lwp *l;
500 
501 	l = curlwp;
502 
503 	SCHED_ASSERT_UNLOCKED();
504 	KERNEL_PROC_LOCK(l);
505 }
506 #endif
507