xref: /netbsd-src/sys/kern/kern_fork.c (revision 466a16a118933bd295a8a104f095714fadf9cf68)
1 /*	$NetBSD: kern_fork.c,v 1.171 2008/10/11 13:40:57 pooka Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2001, 2004, 2006, 2007, 2008 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  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1989, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  * (c) UNIX System Laboratories, Inc.
37  * All or some portions of this file are derived from material licensed
38  * to the University of California by American Telephone and Telegraph
39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40  * the permission of UNIX System Laboratories, Inc.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)kern_fork.c	8.8 (Berkeley) 2/14/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.171 2008/10/11 13:40:57 pooka Exp $");
71 
72 #include "opt_ktrace.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/filedesc.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/pool.h>
80 #include <sys/mount.h>
81 #include <sys/proc.h>
82 #include <sys/ras.h>
83 #include <sys/resourcevar.h>
84 #include <sys/vnode.h>
85 #include <sys/file.h>
86 #include <sys/acct.h>
87 #include <sys/ktrace.h>
88 #include <sys/vmmeter.h>
89 #include <sys/sched.h>
90 #include <sys/signalvar.h>
91 #include <sys/kauth.h>
92 #include <sys/atomic.h>
93 #include <sys/syscallargs.h>
94 #include <sys/uidinfo.h>
95 
96 #include <uvm/uvm_extern.h>
97 
98 u_int	nprocs = 1;		/* process 0 */
99 
100 /*
101  * Number of ticks to sleep if fork() would fail due to process hitting
102  * limits. Exported in miliseconds to userland via sysctl.
103  */
104 int	forkfsleep = 0;
105 
106 /*ARGSUSED*/
107 int
108 sys_fork(struct lwp *l, const void *v, register_t *retval)
109 {
110 
111 	return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
112 }
113 
114 /*
115  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
116  * Address space is not shared, but parent is blocked until child exit.
117  */
118 /*ARGSUSED*/
119 int
120 sys_vfork(struct lwp *l, const void *v, register_t *retval)
121 {
122 
123 	return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
124 	    retval, NULL));
125 }
126 
127 /*
128  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
129  * semantics.  Address space is shared, and parent is blocked until child exit.
130  */
131 /*ARGSUSED*/
132 int
133 sys___vfork14(struct lwp *l, const void *v, register_t *retval)
134 {
135 
136 	return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
137 	    NULL, NULL, retval, NULL));
138 }
139 
140 /*
141  * Linux-compatible __clone(2) system call.
142  */
143 int
144 sys___clone(struct lwp *l, const struct sys___clone_args *uap, register_t *retval)
145 {
146 	/* {
147 		syscallarg(int) flags;
148 		syscallarg(void *) stack;
149 	} */
150 	int flags, sig;
151 
152 	/*
153 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
154 	 */
155 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
156 		return (EINVAL);
157 
158 	/*
159 	 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
160 	 */
161 	if (SCARG(uap, flags) & CLONE_SIGHAND
162 	    && (SCARG(uap, flags) & CLONE_VM) == 0)
163 		return (EINVAL);
164 
165 	flags = 0;
166 
167 	if (SCARG(uap, flags) & CLONE_VM)
168 		flags |= FORK_SHAREVM;
169 	if (SCARG(uap, flags) & CLONE_FS)
170 		flags |= FORK_SHARECWD;
171 	if (SCARG(uap, flags) & CLONE_FILES)
172 		flags |= FORK_SHAREFILES;
173 	if (SCARG(uap, flags) & CLONE_SIGHAND)
174 		flags |= FORK_SHARESIGS;
175 	if (SCARG(uap, flags) & CLONE_VFORK)
176 		flags |= FORK_PPWAIT;
177 
178 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
179 	if (sig < 0 || sig >= _NSIG)
180 		return (EINVAL);
181 
182 	/*
183 	 * Note that the Linux API does not provide a portable way of
184 	 * specifying the stack area; the caller must know if the stack
185 	 * grows up or down.  So, we pass a stack size of 0, so that the
186 	 * code that makes this adjustment is a noop.
187 	 */
188 	return (fork1(l, flags, sig, SCARG(uap, stack), 0,
189 	    NULL, NULL, retval, NULL));
190 }
191 
192 /* print the 'table full' message once per 10 seconds */
193 struct timeval fork_tfmrate = { 10, 0 };
194 
195 /*
196  * General fork call.  Note that another LWP in the process may call exec()
197  * or exit() while we are forking.  It's safe to continue here, because
198  * neither operation will complete until all LWPs have exited the process.
199  */
200 int
201 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
202     void (*func)(void *), void *arg, register_t *retval,
203     struct proc **rnewprocp)
204 {
205 	struct proc	*p1, *p2, *parent;
206 	struct plimit   *p1_lim;
207 	uid_t		uid;
208 	struct lwp	*l2;
209 	int		count;
210 	vaddr_t		uaddr;
211 	bool		inmem;
212 	int		tmp;
213 	int		tnprocs;
214 	int		error = 0;
215 
216 	p1 = l1->l_proc;
217 	uid = kauth_cred_getuid(l1->l_cred);
218 	tnprocs = atomic_inc_uint_nv(&nprocs);
219 
220 	/*
221 	 * Although process entries are dynamically created, we still keep
222 	 * a global limit on the maximum number we will create.
223 	 */
224 	if (__predict_false(tnprocs >= maxproc))
225 		error = -1;
226 	else
227 		error = kauth_authorize_process(l1->l_cred,
228 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
229 
230 	if (error) {
231 		static struct timeval lasttfm;
232 		atomic_dec_uint(&nprocs);
233 		if (ratecheck(&lasttfm, &fork_tfmrate))
234 			tablefull("proc", "increase kern.maxproc or NPROC");
235 		if (forkfsleep)
236 			kpause("forkmx", false, forkfsleep, NULL);
237 		return (EAGAIN);
238 	}
239 
240 	/*
241 	 * Enforce limits.
242 	 */
243 	count = chgproccnt(uid, 1);
244 	if (uid != 0 &&
245 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
246 		(void)chgproccnt(uid, -1);
247 		atomic_dec_uint(&nprocs);
248 		if (forkfsleep)
249 			kpause("forkulim", false, forkfsleep, NULL);
250 		return (EAGAIN);
251 	}
252 
253 	/*
254 	 * Allocate virtual address space for the U-area now, while it
255 	 * is still easy to abort the fork operation if we're out of
256 	 * kernel virtual address space.  The actual U-area pages will
257 	 * be allocated and wired in uvm_fork() if needed.
258 	 */
259 
260 	inmem = uvm_uarea_alloc(&uaddr);
261 	if (__predict_false(uaddr == 0)) {
262 		(void)chgproccnt(uid, -1);
263 		atomic_dec_uint(&nprocs);
264 		return (ENOMEM);
265 	}
266 
267 	/*
268 	 * We are now committed to the fork.  From here on, we may
269 	 * block on resources, but resource allocation may NOT fail.
270 	 */
271 
272 	/* Allocate new proc. */
273 	p2 = proc_alloc();
274 
275 	/*
276 	 * Make a proc table entry for the new process.
277 	 * Start by zeroing the section of proc that is zero-initialized,
278 	 * then copy the section that is copied directly from the parent.
279 	 */
280 	memset(&p2->p_startzero, 0,
281 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
282 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
283 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
284 
285 	CIRCLEQ_INIT(&p2->p_sigpend.sp_info);
286 
287 	LIST_INIT(&p2->p_lwps);
288 	LIST_INIT(&p2->p_sigwaiters);
289 
290 	/*
291 	 * Duplicate sub-structures as needed.
292 	 * Increase reference counts on shared objects.
293 	 * The p_stats and p_sigacts substructs are set in uvm_fork().
294 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
295 	 * handling are important in order to keep a consistent behaviour
296 	 * for the child after the fork.
297 	 */
298 	p2->p_flag = p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN);
299 	p2->p_emul = p1->p_emul;
300 	p2->p_execsw = p1->p_execsw;
301 
302 	if (flags & FORK_SYSTEM) {
303 		/*
304 		 * Mark it as a system process.  Set P_NOCLDWAIT so that
305 		 * children are reparented to init(8) when they exit.
306 		 * init(8) can easily wait them out for us.
307 		 */
308 		p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT);
309 	}
310 
311 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
312 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
313 	rw_init(&p2->p_reflock);
314 	cv_init(&p2->p_waitcv, "wait");
315 	cv_init(&p2->p_lwpcv, "lwpwait");
316 
317 	/*
318 	 * Share a lock between the processes if they are to share signal
319 	 * state: we must synchronize access to it.
320 	 */
321 	if (flags & FORK_SHARESIGS) {
322 		p2->p_lock = p1->p_lock;
323 		mutex_obj_hold(p1->p_lock);
324 	} else
325 		p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
326 
327 	kauth_proc_fork(p1, p2);
328 
329 	p2->p_raslist = NULL;
330 #if defined(__HAVE_RAS)
331 	ras_fork(p1, p2);
332 #endif
333 
334 	/* bump references to the text vnode (for procfs) */
335 	p2->p_textvp = p1->p_textvp;
336 	if (p2->p_textvp)
337 		VREF(p2->p_textvp);
338 
339 	if (flags & FORK_SHAREFILES)
340 		fd_share(p2);
341 	else if (flags & FORK_CLEANFILES)
342 		p2->p_fd = fd_init(NULL);
343 	else
344 		p2->p_fd = fd_copy();
345 
346 	if (flags & FORK_SHARECWD)
347 		cwdshare(p2);
348 	else
349 		p2->p_cwdi = cwdinit();
350 
351 	/*
352 	 * p_limit (rlimit stuff) is usually copy-on-write, so we just need
353 	 * to bump pl_refcnt.
354 	 * However in some cases (see compat irix, and plausibly from clone)
355 	 * the parent and child share limits - in which case nothing else
356 	 * must have a copy of the limits (PL_SHAREMOD is set).
357 	 */
358 	if (__predict_false(flags & FORK_SHARELIMIT))
359 		lim_privatise(p1, 1);
360 	p1_lim = p1->p_limit;
361 	if (p1_lim->pl_flags & PL_WRITEABLE && !(flags & FORK_SHARELIMIT))
362 		p2->p_limit = lim_copy(p1_lim);
363 	else {
364 		lim_addref(p1_lim);
365 		p2->p_limit = p1_lim;
366 	}
367 
368 	p2->p_lflag = ((flags & FORK_PPWAIT) ? PL_PPWAIT : 0);
369 	p2->p_sflag = 0;
370 	p2->p_slflag = 0;
371 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
372 	p2->p_pptr = parent;
373 	p2->p_ppid = parent->p_pid;
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_lock);
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_lock);
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, (flags & FORK_PPWAIT) ? LWP_VFORK : 0,
434 	    stack, stacksize, (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(proc_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 	LIST_INSERT_AFTER(p1, p2, p_pglist);
450 	LIST_INSERT_HEAD(&allproc, p2, p_list);
451 
452 	p2->p_trace_enabled = trace_is_enabled(p2);
453 #ifdef __HAVE_SYSCALL_INTERN
454 	(*p2->p_emul->e_syscall_intern)(p2);
455 #endif
456 
457 	/*
458 	 * Update stats now that we know the fork was successful.
459 	 */
460 	uvmexp.forks++;
461 	if (flags & FORK_PPWAIT)
462 		uvmexp.forks_ppwait++;
463 	if (flags & FORK_SHAREVM)
464 		uvmexp.forks_sharevm++;
465 
466 	/*
467 	 * Pass a pointer to the new process to the caller.
468 	 */
469 	if (rnewprocp != NULL)
470 		*rnewprocp = p2;
471 
472 	if (ktrpoint(KTR_EMUL))
473 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
474 
475 	/*
476 	 * Now can be swapped.
477 	 */
478 	uvm_lwp_rele(l1);
479 
480 	/*
481 	 * Notify any interested parties about the new process.
482 	 */
483 	if (!SLIST_EMPTY(&p1->p_klist)) {
484 		mutex_exit(proc_lock);
485 		KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
486 		mutex_enter(proc_lock);
487 	}
488 
489 	/*
490 	 * Make child runnable, set start time, and add to run queue except
491 	 * if the parent requested the child to start in SSTOP state.
492 	 */
493 	tmp = (p2->p_userret != NULL ? LW_WUSERRET : 0);
494 	mutex_enter(p2->p_lock);
495 
496 	/*
497 	 * Start profiling.
498 	 */
499 	if ((p2->p_stflag & PST_PROFIL) != 0) {
500 		mutex_spin_enter(&p2->p_stmutex);
501 		startprofclock(p2);
502 		mutex_spin_exit(&p2->p_stmutex);
503 	}
504 
505 	getmicrotime(&p2->p_stats->p_start);
506 	p2->p_acflag = AFORK;
507 	lwp_lock(l2);
508 	if (p2->p_sflag & PS_STOPFORK) {
509 		p2->p_nrlwps = 0;
510 		p2->p_stat = SSTOP;
511 		p2->p_waited = 0;
512 		p1->p_nstopchild++;
513 		l2->l_stat = LSSTOP;
514 		l2->l_flag |= tmp;
515 		lwp_unlock(l2);
516 	} else {
517 		p2->p_nrlwps = 1;
518 		p2->p_stat = SACTIVE;
519 		l2->l_stat = LSRUN;
520 		l2->l_flag |= tmp;
521 		sched_enqueue(l2, false);
522 		lwp_unlock(l2);
523 	}
524 
525 	mutex_exit(p2->p_lock);
526 
527 	/*
528 	 * Preserve synchronization semantics of vfork.  If waiting for
529 	 * child to exec or exit, set PL_PPWAIT on child, and sleep on our
530 	 * proc (in case of exit).
531 	 */
532 	while (p2->p_lflag & PL_PPWAIT)
533 		cv_wait(&p1->p_waitcv, proc_lock);
534 
535 	mutex_exit(proc_lock);
536 
537 	/*
538 	 * Return child pid to parent process,
539 	 * marking us as parent via retval[1].
540 	 */
541 	if (retval != NULL) {
542 		retval[0] = p2->p_pid;
543 		retval[1] = 0;
544 	}
545 
546 	return (0);
547 }
548