xref: /netbsd-src/sys/kern/kern_fork.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: kern_fork.c,v 1.178 2010/07/07 01:30:37 chs 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 1.178 2010/07/07 01:30:37 chs 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/pool.h>
79 #include <sys/mount.h>
80 #include <sys/proc.h>
81 #include <sys/ras.h>
82 #include <sys/resourcevar.h>
83 #include <sys/vnode.h>
84 #include <sys/file.h>
85 #include <sys/acct.h>
86 #include <sys/ktrace.h>
87 #include <sys/vmmeter.h>
88 #include <sys/sched.h>
89 #include <sys/signalvar.h>
90 #include <sys/kauth.h>
91 #include <sys/atomic.h>
92 #include <sys/syscallargs.h>
93 #include <sys/uidinfo.h>
94 #include <sys/sdt.h>
95 
96 #include <uvm/uvm_extern.h>
97 
98 /*
99  * DTrace SDT provider definitions
100  */
101 SDT_PROBE_DEFINE(proc,,,create,
102 	    "struct proc *", NULL,	/* new process */
103 	    "struct proc *", NULL,	/* parent process */
104 	    "int", NULL,		/* flags */
105 	    NULL, NULL, NULL, NULL);
106 
107 u_int	nprocs = 1;		/* process 0 */
108 
109 /*
110  * Number of ticks to sleep if fork() would fail due to process hitting
111  * limits. Exported in miliseconds to userland via sysctl.
112  */
113 int	forkfsleep = 0;
114 
115 /*ARGSUSED*/
116 int
117 sys_fork(struct lwp *l, const void *v, register_t *retval)
118 {
119 
120 	return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
121 }
122 
123 /*
124  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
125  * Address space is not shared, but parent is blocked until child exit.
126  */
127 /*ARGSUSED*/
128 int
129 sys_vfork(struct lwp *l, const void *v, register_t *retval)
130 {
131 
132 	return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
133 	    retval, NULL));
134 }
135 
136 /*
137  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
138  * semantics.  Address space is shared, and parent is blocked until child exit.
139  */
140 /*ARGSUSED*/
141 int
142 sys___vfork14(struct lwp *l, const void *v, register_t *retval)
143 {
144 
145 	return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
146 	    NULL, NULL, retval, NULL));
147 }
148 
149 /*
150  * Linux-compatible __clone(2) system call.
151  */
152 int
153 sys___clone(struct lwp *l, const struct sys___clone_args *uap, register_t *retval)
154 {
155 	/* {
156 		syscallarg(int) flags;
157 		syscallarg(void *) stack;
158 	} */
159 	int flags, sig;
160 
161 	/*
162 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
163 	 */
164 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
165 		return (EINVAL);
166 
167 	/*
168 	 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
169 	 */
170 	if (SCARG(uap, flags) & CLONE_SIGHAND
171 	    && (SCARG(uap, flags) & CLONE_VM) == 0)
172 		return (EINVAL);
173 
174 	flags = 0;
175 
176 	if (SCARG(uap, flags) & CLONE_VM)
177 		flags |= FORK_SHAREVM;
178 	if (SCARG(uap, flags) & CLONE_FS)
179 		flags |= FORK_SHARECWD;
180 	if (SCARG(uap, flags) & CLONE_FILES)
181 		flags |= FORK_SHAREFILES;
182 	if (SCARG(uap, flags) & CLONE_SIGHAND)
183 		flags |= FORK_SHARESIGS;
184 	if (SCARG(uap, flags) & CLONE_VFORK)
185 		flags |= FORK_PPWAIT;
186 
187 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
188 	if (sig < 0 || sig >= _NSIG)
189 		return (EINVAL);
190 
191 	/*
192 	 * Note that the Linux API does not provide a portable way of
193 	 * specifying the stack area; the caller must know if the stack
194 	 * grows up or down.  So, we pass a stack size of 0, so that the
195 	 * code that makes this adjustment is a noop.
196 	 */
197 	return (fork1(l, flags, sig, SCARG(uap, stack), 0,
198 	    NULL, NULL, retval, NULL));
199 }
200 
201 /* print the 'table full' message once per 10 seconds */
202 struct timeval fork_tfmrate = { 10, 0 };
203 
204 /*
205  * General fork call.  Note that another LWP in the process may call exec()
206  * or exit() while we are forking.  It's safe to continue here, because
207  * neither operation will complete until all LWPs have exited the process.
208  */
209 int
210 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
211     void (*func)(void *), void *arg, register_t *retval,
212     struct proc **rnewprocp)
213 {
214 	struct proc	*p1, *p2, *parent;
215 	struct plimit   *p1_lim;
216 	uid_t		uid;
217 	struct lwp	*l2;
218 	int		count;
219 	vaddr_t		uaddr;
220 	int		tnprocs;
221 	int		error = 0;
222 
223 	p1 = l1->l_proc;
224 	uid = kauth_cred_getuid(l1->l_cred);
225 	tnprocs = atomic_inc_uint_nv(&nprocs);
226 
227 	/*
228 	 * Although process entries are dynamically created, we still keep
229 	 * a global limit on the maximum number we will create.
230 	 */
231 	if (__predict_false(tnprocs >= maxproc))
232 		error = -1;
233 	else
234 		error = kauth_authorize_process(l1->l_cred,
235 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
236 
237 	if (error) {
238 		static struct timeval lasttfm;
239 		atomic_dec_uint(&nprocs);
240 		if (ratecheck(&lasttfm, &fork_tfmrate))
241 			tablefull("proc", "increase kern.maxproc or NPROC");
242 		if (forkfsleep)
243 			kpause("forkmx", false, forkfsleep, NULL);
244 		return (EAGAIN);
245 	}
246 
247 	/*
248 	 * Enforce limits.
249 	 */
250 	count = chgproccnt(uid, 1);
251 	if (kauth_authorize_generic(l1->l_cred, KAUTH_GENERIC_ISSUSER, NULL) !=
252 	    0 && __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
253 		(void)chgproccnt(uid, -1);
254 		atomic_dec_uint(&nprocs);
255 		if (forkfsleep)
256 			kpause("forkulim", false, forkfsleep, NULL);
257 		return (EAGAIN);
258 	}
259 
260 	/*
261 	 * Allocate virtual address space for the U-area now, while it
262 	 * is still easy to abort the fork operation if we're out of
263 	 * kernel virtual address space.
264 	 */
265 	uaddr = uvm_uarea_alloc();
266 	if (__predict_false(uaddr == 0)) {
267 		(void)chgproccnt(uid, -1);
268 		atomic_dec_uint(&nprocs);
269 		return (ENOMEM);
270 	}
271 
272 	/*
273 	 * We are now committed to the fork.  From here on, we may
274 	 * block on resources, but resource allocation may NOT fail.
275 	 */
276 
277 	/* Allocate new proc. */
278 	p2 = proc_alloc();
279 
280 	/*
281 	 * Make a proc table entry for the new process.
282 	 * Start by zeroing the section of proc that is zero-initialized,
283 	 * then copy the section that is copied directly from the parent.
284 	 */
285 	memset(&p2->p_startzero, 0,
286 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
287 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
288 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
289 
290 	CIRCLEQ_INIT(&p2->p_sigpend.sp_info);
291 
292 	LIST_INIT(&p2->p_lwps);
293 	LIST_INIT(&p2->p_sigwaiters);
294 
295 	/*
296 	 * Duplicate sub-structures as needed.
297 	 * Increase reference counts on shared objects.
298 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
299 	 * handling are important in order to keep a consistent behaviour
300 	 * for the child after the fork.
301 	 */
302 	p2->p_flag = p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN);
303 	p2->p_emul = p1->p_emul;
304 	p2->p_execsw = p1->p_execsw;
305 
306 	if (flags & FORK_SYSTEM) {
307 		/*
308 		 * Mark it as a system process.  Set P_NOCLDWAIT so that
309 		 * children are reparented to init(8) when they exit.
310 		 * init(8) can easily wait them out for us.
311 		 */
312 		p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT);
313 	}
314 
315 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
316 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
317 	rw_init(&p2->p_reflock);
318 	cv_init(&p2->p_waitcv, "wait");
319 	cv_init(&p2->p_lwpcv, "lwpwait");
320 
321 	/*
322 	 * Share a lock between the processes if they are to share signal
323 	 * state: we must synchronize access to it.
324 	 */
325 	if (flags & FORK_SHARESIGS) {
326 		p2->p_lock = p1->p_lock;
327 		mutex_obj_hold(p1->p_lock);
328 	} else
329 		p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
330 
331 	kauth_proc_fork(p1, p2);
332 
333 	p2->p_raslist = NULL;
334 #if defined(__HAVE_RAS)
335 	ras_fork(p1, p2);
336 #endif
337 
338 	/* bump references to the text vnode (for procfs) */
339 	p2->p_textvp = p1->p_textvp;
340 	if (p2->p_textvp)
341 		vref(p2->p_textvp);
342 
343 	if (flags & FORK_SHAREFILES)
344 		fd_share(p2);
345 	else if (flags & FORK_CLEANFILES)
346 		p2->p_fd = fd_init(NULL);
347 	else
348 		p2->p_fd = fd_copy();
349 
350 	if (flags & FORK_SHARECWD)
351 		cwdshare(p2);
352 	else
353 		p2->p_cwdi = cwdinit();
354 
355 	/*
356 	 * p_limit (rlimit stuff) is usually copy-on-write, so we just need
357 	 * to bump pl_refcnt.
358 	 * However in some cases (see compat irix, and plausibly from clone)
359 	 * the parent and child share limits - in which case nothing else
360 	 * must have a copy of the limits (PL_SHAREMOD is set).
361 	 */
362 	if (__predict_false(flags & FORK_SHARELIMIT))
363 		lim_privatise(p1, 1);
364 	p1_lim = p1->p_limit;
365 	if (p1_lim->pl_flags & PL_WRITEABLE && !(flags & FORK_SHARELIMIT))
366 		p2->p_limit = lim_copy(p1_lim);
367 	else {
368 		lim_addref(p1_lim);
369 		p2->p_limit = p1_lim;
370 	}
371 
372 	p2->p_lflag = ((flags & FORK_PPWAIT) ? PL_PPWAIT : 0);
373 	p2->p_sflag = 0;
374 	p2->p_slflag = 0;
375 	parent = (flags & FORK_NOWAIT) ? initproc : p1;
376 	p2->p_pptr = parent;
377 	p2->p_ppid = parent->p_pid;
378 	LIST_INIT(&p2->p_children);
379 
380 	p2->p_aio = NULL;
381 
382 #ifdef KTRACE
383 	/*
384 	 * Copy traceflag and tracefile if enabled.
385 	 * If not inherited, these were zeroed above.
386 	 */
387 	if (p1->p_traceflag & KTRFAC_INHERIT) {
388 		mutex_enter(&ktrace_lock);
389 		p2->p_traceflag = p1->p_traceflag;
390 		if ((p2->p_tracep = p1->p_tracep) != NULL)
391 			ktradref(p2);
392 		mutex_exit(&ktrace_lock);
393 	}
394 #endif
395 
396 	/*
397 	 * Create signal actions for the child process.
398 	 */
399 	p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS);
400 	mutex_enter(p1->p_lock);
401 	p2->p_sflag |=
402 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
403 	sched_proc_fork(p1, p2);
404 	mutex_exit(p1->p_lock);
405 
406 	p2->p_stflag = p1->p_stflag;
407 
408 	/*
409 	 * p_stats.
410 	 * Copy parts of p_stats, and zero out the rest.
411 	 */
412 	p2->p_stats = pstatscopy(p1->p_stats);
413 
414 	/*
415 	 * Set up the new process address space.
416 	 */
417 	uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false);
418 
419 	/*
420 	 * Finish creating the child process.
421 	 * It will return through a different path later.
422 	 */
423 	lwp_create(l1, p2, uaddr, (flags & FORK_PPWAIT) ? LWP_VFORK : 0,
424 	    stack, stacksize, (func != NULL) ? func : child_return, arg, &l2,
425 	    l1->l_class);
426 
427 	/*
428 	 * If emulation has a process fork hook, call it now.
429 	 */
430 	if (p2->p_emul->e_proc_fork)
431 		(*p2->p_emul->e_proc_fork)(p2, l1, flags);
432 
433 	/*
434 	 * ...and finally, any other random fork hooks that subsystems
435 	 * might have registered.
436 	 */
437 	doforkhooks(p2, p1);
438 
439 	SDT_PROBE(proc,,,create, p2, p1, flags, 0, 0);
440 
441 	/*
442 	 * It's now safe for the scheduler and other processes to see the
443 	 * child process.
444 	 */
445 	mutex_enter(proc_lock);
446 
447 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
448 		p2->p_lflag |= PL_CONTROLT;
449 
450 	LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
451 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
452 
453 	LIST_INSERT_AFTER(p1, p2, p_pglist);
454 	LIST_INSERT_HEAD(&allproc, p2, p_list);
455 
456 	p2->p_trace_enabled = trace_is_enabled(p2);
457 #ifdef __HAVE_SYSCALL_INTERN
458 	(*p2->p_emul->e_syscall_intern)(p2);
459 #endif
460 
461 	/*
462 	 * Update stats now that we know the fork was successful.
463 	 */
464 	uvmexp.forks++;
465 	if (flags & FORK_PPWAIT)
466 		uvmexp.forks_ppwait++;
467 	if (flags & FORK_SHAREVM)
468 		uvmexp.forks_sharevm++;
469 
470 	/*
471 	 * Pass a pointer to the new process to the caller.
472 	 */
473 	if (rnewprocp != NULL)
474 		*rnewprocp = p2;
475 
476 	if (ktrpoint(KTR_EMUL))
477 		p2->p_traceflag |= KTRFAC_TRC_EMUL;
478 
479 	/*
480 	 * Notify any interested parties about the new process.
481 	 */
482 	if (!SLIST_EMPTY(&p1->p_klist)) {
483 		mutex_exit(proc_lock);
484 		KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
485 		mutex_enter(proc_lock);
486 	}
487 
488 	/*
489 	 * Make child runnable, set start time, and add to run queue except
490 	 * if the parent requested the child to start in SSTOP state.
491 	 */
492 	mutex_enter(p2->p_lock);
493 
494 	/*
495 	 * Start profiling.
496 	 */
497 	if ((p2->p_stflag & PST_PROFIL) != 0) {
498 		mutex_spin_enter(&p2->p_stmutex);
499 		startprofclock(p2);
500 		mutex_spin_exit(&p2->p_stmutex);
501 	}
502 
503 	getmicrotime(&p2->p_stats->p_start);
504 	p2->p_acflag = AFORK;
505 	lwp_lock(l2);
506 	KASSERT(p2->p_nrlwps == 1);
507 	if (p2->p_sflag & PS_STOPFORK) {
508 		p2->p_nrlwps = 0;
509 		p2->p_stat = SSTOP;
510 		p2->p_waited = 0;
511 		p1->p_nstopchild++;
512 		l2->l_stat = LSSTOP;
513 		lwp_unlock(l2);
514 	} else {
515 		p2->p_nrlwps = 1;
516 		p2->p_stat = SACTIVE;
517 		l2->l_stat = LSRUN;
518 		sched_enqueue(l2, false);
519 		lwp_unlock(l2);
520 	}
521 
522 	mutex_exit(p2->p_lock);
523 
524 	/*
525 	 * Preserve synchronization semantics of vfork.  If waiting for
526 	 * child to exec or exit, set PL_PPWAIT on child, and sleep on our
527 	 * proc (in case of exit).
528 	 */
529 	while (p2->p_lflag & PL_PPWAIT)
530 		cv_wait(&p1->p_waitcv, proc_lock);
531 
532 	mutex_exit(proc_lock);
533 
534 	/*
535 	 * Return child pid to parent process,
536 	 * marking us as parent via retval[1].
537 	 */
538 	if (retval != NULL) {
539 		retval[0] = p2->p_pid;
540 		retval[1] = 0;
541 	}
542 
543 	return (0);
544 }
545