xref: /netbsd-src/sys/kern/kern_fork.c (revision 28c37e673e4d9b6cbdc7483062b915cc61d1ccf5)
1 /*	$NetBSD: kern_fork.c,v 1.94 2002/09/25 22:21:42 thorpej 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 1.94 2002/09/25 22:21:42 thorpej 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/syscallargs.h>
107 
108 #include <uvm/uvm_extern.h>
109 
110 
111 int	nprocs = 1;		/* process 0 */
112 
113 /*ARGSUSED*/
114 int
115 sys_fork(struct proc *p, void *v, register_t *retval)
116 {
117 
118 	return (fork1(p, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
119 }
120 
121 /*
122  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
123  * Address space is not shared, but parent is blocked until child exit.
124  */
125 /*ARGSUSED*/
126 int
127 sys_vfork(struct proc *p, void *v, register_t *retval)
128 {
129 
130 	return (fork1(p, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
131 	    retval, NULL));
132 }
133 
134 /*
135  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
136  * semantics.  Address space is shared, and parent is blocked until child exit.
137  */
138 /*ARGSUSED*/
139 int
140 sys___vfork14(struct proc *p, void *v, register_t *retval)
141 {
142 
143 	return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
144 	    NULL, NULL, retval, NULL));
145 }
146 
147 /*
148  * Linux-compatible __clone(2) system call.
149  */
150 int
151 sys___clone(struct proc *p, void *v, register_t *retval)
152 {
153 	struct sys___clone_args /* {
154 		syscallarg(int) flags;
155 		syscallarg(void *) stack;
156 	} */ *uap = v;
157 	int flags, sig;
158 
159 	/*
160 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
161 	 */
162 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
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(p, flags, sig, SCARG(uap, stack), 0,
189 	    NULL, NULL, retval, NULL));
190 }
191 
192 int
193 fork1(struct proc *p1, int flags, int exitsig, void *stack, size_t stacksize,
194     void (*func)(void *), void *arg, register_t *retval,
195     struct proc **rnewprocp)
196 {
197 	struct proc	*p2, *tp;
198 	uid_t		uid;
199 	int		count, s;
200 	vaddr_t		uaddr;
201 	static int	nextpid, pidchecked;
202 
203 	/*
204 	 * Although process entries are dynamically created, we still keep
205 	 * a global limit on the maximum number we will create.  Don't allow
206 	 * a nonprivileged user to use the last process; don't let root
207 	 * exceed the limit. The variable nprocs is the current number of
208 	 * processes, maxproc is the limit.
209 	 */
210 	uid = p1->p_cred->p_ruid;
211 	if (__predict_false((nprocs >= maxproc - 1 && uid != 0) ||
212 			    nprocs >= maxproc)) {
213 		tablefull("proc", "increase kern.maxproc or NPROC");
214 		return (EAGAIN);
215 	}
216 	nprocs++;
217 
218 	/*
219 	 * Increment the count of procs running with this uid. Don't allow
220 	 * a nonprivileged user to exceed their current limit.
221 	 */
222 	count = chgproccnt(uid, 1);
223 	if (__predict_false(uid != 0 && count >
224 			    p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
225 		(void)chgproccnt(uid, -1);
226 		nprocs--;
227 		return (EAGAIN);
228 	}
229 
230 	/*
231 	 * Allocate virtual address space for the U-area now, while it
232 	 * is still easy to abort the fork operation if we're out of
233 	 * kernel virtual address space.  The actual U-area pages will
234 	 * be allocated and wired in uvm_fork().
235 	 */
236 
237 	uaddr = uvm_uarea_alloc();
238 	if (__predict_false(uaddr == 0)) {
239 		(void)chgproccnt(uid, -1);
240 		nprocs--;
241 		return (ENOMEM);
242 	}
243 
244 	/*
245 	 * We are now committed to the fork.  From here on, we may
246 	 * block on resources, but resource allocation may NOT fail.
247 	 */
248 
249 	/* Allocate new proc. */
250 	p2 = pool_get(&proc_pool, PR_WAITOK);
251 
252 	/*
253 	 * BEGIN PID ALLOCATION.
254 	 */
255 	s = proclist_lock_write();
256 
257 	/*
258 	 * Find an unused process ID.  We remember a range of unused IDs
259 	 * ready to use (from nextpid+1 through pidchecked-1).
260 	 */
261 	nextpid++;
262  retry:
263 	/*
264 	 * If the process ID prototype has wrapped around,
265 	 * restart somewhat above 0, as the low-numbered procs
266 	 * tend to include daemons that don't exit.
267 	 */
268 	if (nextpid >= PID_MAX) {
269 		nextpid = 500;
270 		pidchecked = 0;
271 	}
272 	if (nextpid >= pidchecked) {
273 		const struct proclist_desc *pd;
274 
275 		pidchecked = PID_MAX;
276 		/*
277 		 * Scan the process lists to check whether this pid
278 		 * is in use.  Remember the lowest pid that's greater
279 		 * than nextpid, so we can avoid checking for a while.
280 		 */
281 		pd = proclists;
282  again:
283 		LIST_FOREACH(tp, pd->pd_list, p_list) {
284 			while (tp->p_pid == nextpid ||
285 			    tp->p_pgrp->pg_id == nextpid ||
286 			    tp->p_session->s_sid == nextpid) {
287 				nextpid++;
288 				if (nextpid >= pidchecked)
289 					goto retry;
290 			}
291 			if (tp->p_pid > nextpid && pidchecked > tp->p_pid)
292 				pidchecked = tp->p_pid;
293 
294 			if (tp->p_pgrp->pg_id > nextpid &&
295 			    pidchecked > tp->p_pgrp->pg_id)
296 				pidchecked = tp->p_pgrp->pg_id;
297 
298 			if (tp->p_session->s_sid > nextpid &&
299 			    pidchecked > tp->p_session->s_sid)
300 				pidchecked = tp->p_session->s_sid;
301 		}
302 
303 		/*
304 		 * If there's another list, scan it.  If we have checked
305 		 * them all, we've found one!
306 		 */
307 		pd++;
308 		if (pd->pd_list != NULL)
309 			goto again;
310 	}
311 
312 	/*
313 	 * Put the proc on allproc before unlocking PID allocation
314 	 * so that waiters won't grab it as soon as we unlock.
315 	 */
316 
317 	p2->p_stat = SIDL;			/* protect against others */
318 	p2->p_pid = nextpid;
319 	p2->p_exitsig = exitsig;		/* signal for parent on exit */
320 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
321 
322 	LIST_INSERT_HEAD(&allproc, p2, p_list);
323 
324 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
325 
326 	/*
327 	 * END PID ALLOCATION.
328 	 */
329 	proclist_unlock_write(s);
330 
331 	/*
332 	 * Make a proc table entry for the new process.
333 	 * Start by zeroing the section of proc that is zero-initialized,
334 	 * then copy the section that is copied directly from the parent.
335 	 */
336 	memset(&p2->p_startzero, 0,
337 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
338 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
339 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
340 
341 #if !defined(MULTIPROCESSOR)
342 	/*
343 	 * In the single-processor case, all processes will always run
344 	 * on the same CPU.  So, initialize the child's CPU to the parent's
345 	 * now.  In the multiprocessor case, the child's CPU will be
346 	 * initialized in the low-level context switch code when the
347 	 * process runs.
348 	 */
349 	p2->p_cpu = p1->p_cpu;
350 #else
351 	/*
352 	 * zero child's cpu pointer so we don't get trash.
353 	 */
354 	p2->p_cpu = NULL;
355 #endif /* ! MULTIPROCESSOR */
356 
357 	/*
358 	 * Duplicate sub-structures as needed.
359 	 * Increase reference counts on shared objects.
360 	 * The p_stats and p_sigacts substructs are set in uvm_fork().
361 	 */
362 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
363 	p2->p_emul = p1->p_emul;
364 	p2->p_execsw = p1->p_execsw;
365 
366 	if (p1->p_flag & P_PROFIL)
367 		startprofclock(p2);
368 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
369 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
370 	p2->p_cred->p_refcnt = 1;
371 	crhold(p1->p_ucred);
372 
373 	LIST_INIT(&p2->p_raslist);
374 	p2->p_nras = 0;
375 	simple_lock_init(&p2->p_raslock);
376 #if defined(__HAVE_RAS)
377 	ras_fork(p1, p2);
378 #endif
379 
380 	/* bump references to the text vnode (for procfs) */
381 	p2->p_textvp = p1->p_textvp;
382 	if (p2->p_textvp)
383 		VREF(p2->p_textvp);
384 
385 	if (flags & FORK_SHAREFILES)
386 		fdshare(p1, p2);
387 	else if (flags & FORK_CLEANFILES)
388 		p2->p_fd = fdinit(p1);
389 	else
390 		p2->p_fd = fdcopy(p1);
391 
392 	if (flags & FORK_SHARECWD)
393 		cwdshare(p1, p2);
394 	else
395 		p2->p_cwdi = cwdinit(p1);
396 
397 	/*
398 	 * If p_limit is still copy-on-write, bump refcnt,
399 	 * otherwise get a copy that won't be modified.
400 	 * (If PL_SHAREMOD is clear, the structure is shared
401 	 * copy-on-write.)
402 	 */
403 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
404 		p2->p_limit = limcopy(p1->p_limit);
405 	else {
406 		p2->p_limit = p1->p_limit;
407 		p2->p_limit->p_refcnt++;
408 	}
409 
410 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
411 		p2->p_flag |= P_CONTROLT;
412 	if (flags & FORK_PPWAIT)
413 		p2->p_flag |= P_PPWAIT;
414 	LIST_INSERT_AFTER(p1, p2, p_pglist);
415 	p2->p_pptr = (flags & FORK_NOWAIT) ? initproc : p1;
416 	LIST_INSERT_HEAD(&p2->p_pptr->p_children, p2, p_sibling);
417 	LIST_INIT(&p2->p_children);
418 
419 	callout_init(&p2->p_realit_ch);
420 	callout_init(&p2->p_tsleep_ch);
421 
422 #ifdef KTRACE
423 	/*
424 	 * Copy traceflag and tracefile if enabled.
425 	 * If not inherited, these were zeroed above.
426 	 */
427 	if (p1->p_traceflag & KTRFAC_INHERIT) {
428 		p2->p_traceflag = p1->p_traceflag;
429 		if ((p2->p_tracep = p1->p_tracep) != NULL)
430 			ktradref(p2);
431 	}
432 #endif
433 #ifdef SYSTRACE
434 	/* Tell systrace what's happening. */
435 	if (ISSET(p1->p_flag, P_SYSTRACE))
436 		systrace_sys_fork(p1, p2);
437 #endif
438 
439 
440 #ifdef __HAVE_SYSCALL_INTERN
441 	(*p2->p_emul->e_syscall_intern)(p2);
442 #endif
443 
444 	scheduler_fork_hook(p1, p2);
445 
446 	/*
447 	 * Create signal actions for the child process.
448 	 */
449 	sigactsinit(p2, p1, flags & FORK_SHARESIGS);
450 
451 	/*
452 	 * If emulation has process fork hook, call it now.
453 	 */
454 	if (p2->p_emul->e_proc_fork)
455 		(*p2->p_emul->e_proc_fork)(p2, p1);
456 
457 	/*
458 	 * This begins the section where we must prevent the parent
459 	 * from being swapped.
460 	 */
461 	PHOLD(p1);
462 
463 	/*
464 	 * Finish creating the child process.  It will return through a
465 	 * different path later.
466 	 */
467 	p2->p_addr = (struct user *)uaddr;
468 	uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE,
469 	    stack, stacksize,
470 	    (func != NULL) ? func : child_return,
471 	    (arg != NULL) ? arg : p2);
472 
473 	/*
474 	 * Make child runnable, set start time, and add to run queue.
475 	 */
476 	SCHED_LOCK(s);
477 	p2->p_stats->p_start = time;
478 	p2->p_acflag = AFORK;
479 	p2->p_stat = SRUN;
480 	setrunqueue(p2);
481 	SCHED_UNLOCK(s);
482 
483 	/*
484 	 * Now can be swapped.
485 	 */
486 	PRELE(p1);
487 
488 	/*
489 	 * Update stats now that we know the fork was successful.
490 	 */
491 	uvmexp.forks++;
492 	if (flags & FORK_PPWAIT)
493 		uvmexp.forks_ppwait++;
494 	if (flags & FORK_SHAREVM)
495 		uvmexp.forks_sharevm++;
496 
497 	/*
498 	 * Pass a pointer to the new process to the caller.
499 	 */
500 	if (rnewprocp != NULL)
501 		*rnewprocp = p2;
502 
503 #ifdef KTRACE
504 	if (KTRPOINT(p2, KTR_EMUL))
505 		ktremul(p2);
506 #endif
507 
508 	/*
509 	 * Preserve synchronization semantics of vfork.  If waiting for
510 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
511 	 * proc (in case of exit).
512 	 */
513 	if (flags & FORK_PPWAIT)
514 		while (p2->p_flag & P_PPWAIT)
515 			tsleep(p1, PWAIT, "ppwait", 0);
516 
517 	/*
518 	 * Return child pid to parent process,
519 	 * marking us as parent via retval[1].
520 	 */
521 	if (retval != NULL) {
522 		retval[0] = p2->p_pid;
523 		retval[1] = 0;
524 	}
525 
526 	return (0);
527 }
528 
529 #if defined(MULTIPROCESSOR)
530 /*
531  * XXX This is a slight hack to get newly-formed processes to
532  * XXX acquire the kernel lock as soon as they run.
533  */
534 void
535 proc_trampoline_mp(void)
536 {
537 	struct proc *p;
538 
539 	p = curproc;
540 
541 	SCHED_ASSERT_UNLOCKED();
542 	KERNEL_PROC_LOCK(p);
543 }
544 #endif
545