xref: /netbsd-src/sys/kern/kern_fork.c (revision 21e37cc72a480a47828990a439cde7ac9ffaf0c6)
1 /*	$NetBSD: kern_fork.c,v 1.115 2004/05/06 22:20:30 pk 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 1.115 2004/05/06 22:20:30 pk 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 	flags = 0;
169 
170 	if (SCARG(uap, flags) & CLONE_VM)
171 		flags |= FORK_SHAREVM;
172 	if (SCARG(uap, flags) & CLONE_FS)
173 		flags |= FORK_SHARECWD;
174 	if (SCARG(uap, flags) & CLONE_FILES)
175 		flags |= FORK_SHAREFILES;
176 	if (SCARG(uap, flags) & CLONE_SIGHAND)
177 		flags |= FORK_SHARESIGS;
178 	if (SCARG(uap, flags) & CLONE_VFORK)
179 		flags |= FORK_PPWAIT;
180 
181 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
182 	if (sig < 0 || sig >= _NSIG)
183 		return (EINVAL);
184 
185 	/*
186 	 * Note that the Linux API does not provide a portable way of
187 	 * specifying the stack area; the caller must know if the stack
188 	 * grows up or down.  So, we pass a stack size of 0, so that the
189 	 * code that makes this adjustment is a noop.
190 	 */
191 	return (fork1(l, flags, sig, SCARG(uap, stack), 0,
192 	    NULL, NULL, retval, NULL));
193 }
194 
195 /* print the 'table full' message once per 10 seconds */
196 struct timeval fork_tfmrate = { 10, 0 };
197 
198 int
199 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
200     void (*func)(void *), void *arg, register_t *retval,
201     struct proc **rnewprocp)
202 {
203 	struct proc	*p1, *p2, *parent;
204 	uid_t		uid;
205 	struct lwp	*l2;
206 	int		count, s;
207 	vaddr_t		uaddr;
208 	boolean_t	inmem;
209 
210 	/*
211 	 * Although process entries are dynamically created, we still keep
212 	 * a global limit on the maximum number we will create.  Don't allow
213 	 * a nonprivileged user to use the last few processes; don't let root
214 	 * exceed the limit. The variable nprocs is the current number of
215 	 * processes, maxproc is the limit.
216 	 */
217 	p1 = l1->l_proc;
218 	uid = p1->p_cred->p_ruid;
219 	if (__predict_false((nprocs >= maxproc - 5 && uid != 0) ||
220 			    nprocs >= maxproc)) {
221 		static struct timeval lasttfm;
222 
223 		if (ratecheck(&lasttfm, &fork_tfmrate))
224 			tablefull("proc", "increase kern.maxproc or NPROC");
225 		if (forkfsleep)
226 			(void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep);
227 		return (EAGAIN);
228 	}
229 	nprocs++;
230 
231 	/*
232 	 * Increment the count of procs running with this uid. Don't allow
233 	 * a nonprivileged user to exceed their current limit.
234 	 */
235 	count = chgproccnt(uid, 1);
236 	if (__predict_false(uid != 0 && count >
237 			    p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
238 		(void)chgproccnt(uid, -1);
239 		nprocs--;
240 		if (forkfsleep)
241 			(void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep);
242 		return (EAGAIN);
243 	}
244 
245 	/*
246 	 * Allocate virtual address space for the U-area now, while it
247 	 * is still easy to abort the fork operation if we're out of
248 	 * kernel virtual address space.  The actual U-area pages will
249 	 * be allocated and wired in uvm_fork() if needed.
250 	 */
251 
252 	inmem = uvm_uarea_alloc(&uaddr);
253 	if (__predict_false(uaddr == 0)) {
254 		(void)chgproccnt(uid, -1);
255 		nprocs--;
256 		return (ENOMEM);
257 	}
258 
259 	/*
260 	 * We are now committed to the fork.  From here on, we may
261 	 * block on resources, but resource allocation may NOT fail.
262 	 */
263 
264 	/* Allocate new proc. */
265 	p2 = proc_alloc();
266 
267 	/*
268 	 * Make a proc table entry for the new process.
269 	 * Start by zeroing the section of proc that is zero-initialized,
270 	 * then copy the section that is copied directly from the parent.
271 	 */
272 	memset(&p2->p_startzero, 0,
273 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
274 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
275 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
276 
277 	simple_lock_init(&p2->p_sigctx.ps_silock);
278 	CIRCLEQ_INIT(&p2->p_sigctx.ps_siginfo);
279 	simple_lock_init(&p2->p_lock);
280 	LIST_INIT(&p2->p_lwps);
281 
282 	/*
283 	 * Duplicate sub-structures as needed.
284 	 * Increase reference counts on shared objects.
285 	 * The p_stats and p_sigacts substructs are set in uvm_fork().
286 	 */
287 	p2->p_flag = (p1->p_flag & P_SUGID);
288 	p2->p_emul = p1->p_emul;
289 	p2->p_execsw = p1->p_execsw;
290 
291 	if (p1->p_flag & P_PROFIL)
292 		startprofclock(p2);
293 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
294 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
295 	p2->p_cred->p_refcnt = 1;
296 	crhold(p1->p_ucred);
297 
298 	LIST_INIT(&p2->p_raslist);
299 #if defined(__HAVE_RAS)
300 	ras_fork(p1, p2);
301 #endif
302 
303 	/* bump references to the text vnode (for procfs) */
304 	p2->p_textvp = p1->p_textvp;
305 	if (p2->p_textvp)
306 		VREF(p2->p_textvp);
307 
308 	if (flags & FORK_SHAREFILES)
309 		fdshare(p1, p2);
310 	else if (flags & FORK_CLEANFILES)
311 		p2->p_fd = fdinit(p1);
312 	else
313 		p2->p_fd = fdcopy(p1);
314 
315 	if (flags & FORK_SHARECWD)
316 		cwdshare(p1, p2);
317 	else
318 		p2->p_cwdi = cwdinit(p1);
319 
320 	/*
321 	 * If p_limit is still copy-on-write, bump refcnt,
322 	 * otherwise get a copy that won't be modified.
323 	 * (If PL_SHAREMOD is clear, the structure is shared
324 	 * copy-on-write.)
325 	 */
326 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
327 		p2->p_limit = limcopy(p1->p_limit);
328 	else {
329 		simple_lock(&p1->p_limit->p_slock);
330 		p1->p_limit->p_refcnt++;
331 		simple_unlock(&p1->p_limit->p_slock);
332 		p2->p_limit = p1->p_limit;
333 	}
334 
335 	/* Inherit STOPFORK and STOPEXEC flags */
336 	p2->p_flag |= p1->p_flag & (P_STOPFORK | P_STOPEXEC);
337 
338 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
339 		p2->p_flag |= P_CONTROLT;
340 	if (flags & FORK_PPWAIT)
341 		p2->p_flag |= P_PPWAIT;
342 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
343 	p2->p_pptr = parent;
344 	LIST_INIT(&p2->p_children);
345 
346 	s = proclist_lock_write();
347 	LIST_INSERT_AFTER(p1, p2, p_pglist);
348 	LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
349 	proclist_unlock_write(s);
350 
351 #ifdef KTRACE
352 	/*
353 	 * Copy traceflag and tracefile if enabled.
354 	 * If not inherited, these were zeroed above.
355 	 */
356 	if (p1->p_traceflag & KTRFAC_INHERIT) {
357 		p2->p_traceflag = p1->p_traceflag;
358 		if ((p2->p_tracep = p1->p_tracep) != NULL)
359 			ktradref(p2);
360 	}
361 #endif
362 
363 	scheduler_fork_hook(p1, p2);
364 
365 	/*
366 	 * Create signal actions for the child process.
367 	 */
368 	sigactsinit(p2, p1, flags & FORK_SHARESIGS);
369 
370 	/*
371 	 * p_stats.
372 	 * Copy parts of p_stats, and zero out the rest.
373 	 */
374 	p2->p_stats = pstatscopy(p1->p_stats);
375 
376 	/*
377 	 * If emulation has process fork hook, call it now.
378 	 */
379 	if (p2->p_emul->e_proc_fork)
380 		(*p2->p_emul->e_proc_fork)(p2, p1);
381 
382 	/*
383 	 * ...and finally, any other random fork hooks that subsystems
384 	 * might have registered.
385 	 */
386 	doforkhooks(p2, p1);
387 
388 	/*
389 	 * This begins the section where we must prevent the parent
390 	 * from being swapped.
391 	 */
392 	PHOLD(l1);
393 
394 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
395 
396 	/*
397 	 * Finish creating the child process.
398 	 * It will return through a different path later.
399 	 */
400 	newlwp(l1, p2, uaddr, inmem, 0, stack, stacksize,
401 	    (func != NULL) ? func : child_return,
402 	    arg, &l2);
403 
404 	/* Now safe for scheduler to see child process */
405 	s = proclist_lock_write();
406 	p2->p_stat = SIDL;			/* protect against others */
407 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
408 	LIST_INSERT_HEAD(&allproc, p2, p_list);
409 	proclist_unlock_write(s);
410 
411 #ifdef SYSTRACE
412 	/* Tell systrace what's happening. */
413 	if (ISSET(p1->p_flag, P_SYSTRACE))
414 		systrace_sys_fork(p1, p2);
415 #endif
416 
417 #ifdef __HAVE_SYSCALL_INTERN
418 	(*p2->p_emul->e_syscall_intern)(p2);
419 #endif
420 
421 	/*
422 	 * Make child runnable, set start time, and add to run queue
423 	 * except if the parent requested the child to start in SSTOP state.
424 	 */
425 	SCHED_LOCK(s);
426 	p2->p_stats->p_start = time;
427 	p2->p_acflag = AFORK;
428 	p2->p_nrlwps = 1;
429 	if (p1->p_flag & P_STOPFORK) {
430 		p1->p_nstopchild++;
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 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
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