xref: /netbsd-src/sys/kern/kern_fork.c (revision 27578b9aac214cc7796ead81dcc5427e79d5f2a0)
1 /*	$NetBSD: kern_fork.c,v 1.86 2001/07/07 23:33:54 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 "opt_ktrace.h"
81 #include "opt_multiprocessor.h"
82 
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/map.h>
86 #include <sys/filedesc.h>
87 #include <sys/kernel.h>
88 #include <sys/malloc.h>
89 #include <sys/pool.h>
90 #include <sys/mount.h>
91 #include <sys/proc.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 
101 #include <sys/syscallargs.h>
102 
103 #include <uvm/uvm_extern.h>
104 
105 int	nprocs = 1;		/* process 0 */
106 
107 /*ARGSUSED*/
108 int
109 sys_fork(struct proc *p, void *v, register_t *retval)
110 {
111 
112 	return (fork1(p, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
113 }
114 
115 /*
116  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
117  * Address space is not shared, but parent is blocked until child exit.
118  */
119 /*ARGSUSED*/
120 int
121 sys_vfork(struct proc *p, void *v, register_t *retval)
122 {
123 
124 	return (fork1(p, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
125 	    retval, NULL));
126 }
127 
128 /*
129  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
130  * semantics.  Address space is shared, and parent is blocked until child exit.
131  */
132 /*ARGSUSED*/
133 int
134 sys___vfork14(struct proc *p, void *v, register_t *retval)
135 {
136 
137 	return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
138 	    NULL, NULL, retval, NULL));
139 }
140 
141 /*
142  * Linux-compatible __clone(2) system call.
143  */
144 int
145 sys___clone(struct proc *p, void *v, register_t *retval)
146 {
147 	struct sys___clone_args /* {
148 		syscallarg(int) flags;
149 		syscallarg(void *) stack;
150 	} */ *uap = v;
151 	int flags, sig;
152 
153 	/*
154 	 * We don't support the CLONE_PID or CLONE_PTRACE flags.
155 	 */
156 	if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
157 		return (EINVAL);
158 
159 	flags = 0;
160 
161 	if (SCARG(uap, flags) & CLONE_VM)
162 		flags |= FORK_SHAREVM;
163 	if (SCARG(uap, flags) & CLONE_FS)
164 		flags |= FORK_SHARECWD;
165 	if (SCARG(uap, flags) & CLONE_FILES)
166 		flags |= FORK_SHAREFILES;
167 	if (SCARG(uap, flags) & CLONE_SIGHAND)
168 		flags |= FORK_SHARESIGS;
169 	if (SCARG(uap, flags) & CLONE_VFORK)
170 		flags |= FORK_PPWAIT;
171 
172 	sig = SCARG(uap, flags) & CLONE_CSIGNAL;
173 	if (sig < 0 || sig >= _NSIG)
174 		return (EINVAL);
175 
176 	/*
177 	 * Note that the Linux API does not provide a portable way of
178 	 * specifying the stack area; the caller must know if the stack
179 	 * grows up or down.  So, we pass a stack size of 0, so that the
180 	 * code that makes this adjustment is a noop.
181 	 */
182 	return (fork1(p, flags, sig, SCARG(uap, stack), 0,
183 	    NULL, NULL, retval, NULL));
184 }
185 
186 int
187 fork1(struct proc *p1, int flags, int exitsig, void *stack, size_t stacksize,
188     void (*func)(void *), void *arg, register_t *retval,
189     struct proc **rnewprocp)
190 {
191 	struct proc	*p2, *tp;
192 	uid_t		uid;
193 	int		count, s;
194 	vaddr_t		uaddr;
195 	static int	nextpid, pidchecked;
196 
197 	/*
198 	 * Although process entries are dynamically created, we still keep
199 	 * a global limit on the maximum number we will create.  Don't allow
200 	 * a nonprivileged user to use the last process; don't let root
201 	 * exceed the limit. The variable nprocs is the current number of
202 	 * processes, maxproc is the limit.
203 	 */
204 	uid = p1->p_cred->p_ruid;
205 	if (__predict_false((nprocs >= maxproc - 1 && uid != 0) ||
206 			    nprocs >= maxproc)) {
207 		tablefull("proc", "increase kern.maxproc or NPROC");
208 		return (EAGAIN);
209 	}
210 	nprocs++;
211 
212 	/*
213 	 * Increment the count of procs running with this uid. Don't allow
214 	 * a nonprivileged user to exceed their current limit.
215 	 */
216 	count = chgproccnt(uid, 1);
217 	if (__predict_false(uid != 0 && count >
218 			    p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
219 		(void)chgproccnt(uid, -1);
220 		nprocs--;
221 		return (EAGAIN);
222 	}
223 
224 	/*
225 	 * Allocate virtual address space for the U-area now, while it
226 	 * is still easy to abort the fork operation if we're out of
227 	 * kernel virtual address space.  The actual U-area pages will
228 	 * be allocated and wired in vm_fork().
229 	 */
230 
231 #ifndef USPACE_ALIGN
232 #define	USPACE_ALIGN	0
233 #endif
234 
235 	uaddr = uvm_km_valloc_align(kernel_map, USPACE, USPACE_ALIGN);
236 	if (__predict_false(uaddr == 0)) {
237 		(void)chgproccnt(uid, -1);
238 		nprocs--;
239 		return (ENOMEM);
240 	}
241 
242 	/*
243 	 * We are now committed to the fork.  From here on, we may
244 	 * block on resources, but resource allocation may NOT fail.
245 	 */
246 
247 	/* Allocate new proc. */
248 	p2 = pool_get(&proc_pool, PR_WAITOK);
249 
250 	/*
251 	 * Make a proc table entry for the new process.
252 	 * Start by zeroing the section of proc that is zero-initialized,
253 	 * then copy the section that is copied directly from the parent.
254 	 */
255 	memset(&p2->p_startzero, 0,
256 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
257 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
258 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
259 
260 #if !defined(MULTIPROCESSOR)
261 	/*
262 	 * In the single-processor case, all processes will always run
263 	 * on the same CPU.  So, initialize the child's CPU to the parent's
264 	 * now.  In the multiprocessor case, the child's CPU will be
265 	 * initialized in the low-level context switch code when the
266 	 * process runs.
267 	 */
268 	p2->p_cpu = p1->p_cpu;
269 #else
270 	/*
271 	 * zero child's cpu pointer so we don't get trash.
272 	 */
273 	p2->p_cpu = NULL;
274 #endif /* ! MULTIPROCESSOR */
275 
276 	/*
277 	 * Duplicate sub-structures as needed.
278 	 * Increase reference counts on shared objects.
279 	 * The p_stats and p_sigacts substructs are set in uvm_fork().
280 	 */
281 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
282 	p2->p_emul = p1->p_emul;
283 
284 	if (p1->p_flag & P_PROFIL)
285 		startprofclock(p2);
286 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
287 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
288 	p2->p_cred->p_refcnt = 1;
289 	crhold(p1->p_ucred);
290 
291 	/* bump references to the text vnode (for procfs) */
292 	p2->p_textvp = p1->p_textvp;
293 	if (p2->p_textvp)
294 		VREF(p2->p_textvp);
295 
296 	if (flags & FORK_SHAREFILES)
297 		fdshare(p1, p2);
298 	else
299 		p2->p_fd = fdcopy(p1);
300 
301 	if (flags & FORK_SHARECWD)
302 		cwdshare(p1, p2);
303 	else
304 		p2->p_cwdi = cwdinit(p1);
305 
306 	/*
307 	 * If p_limit is still copy-on-write, bump refcnt,
308 	 * otherwise get a copy that won't be modified.
309 	 * (If PL_SHAREMOD is clear, the structure is shared
310 	 * copy-on-write.)
311 	 */
312 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
313 		p2->p_limit = limcopy(p1->p_limit);
314 	else {
315 		p2->p_limit = p1->p_limit;
316 		p2->p_limit->p_refcnt++;
317 	}
318 
319 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
320 		p2->p_flag |= P_CONTROLT;
321 	if (flags & FORK_PPWAIT)
322 		p2->p_flag |= P_PPWAIT;
323 	LIST_INSERT_AFTER(p1, p2, p_pglist);
324 	p2->p_pptr = p1;
325 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
326 	LIST_INIT(&p2->p_children);
327 
328 	callout_init(&p2->p_realit_ch);
329 	callout_init(&p2->p_tsleep_ch);
330 
331 #ifdef KTRACE
332 	/*
333 	 * Copy traceflag and tracefile if enabled.
334 	 * If not inherited, these were zeroed above.
335 	 */
336 	if (p1->p_traceflag & KTRFAC_INHERIT) {
337 		p2->p_traceflag = p1->p_traceflag;
338 		if ((p2->p_tracep = p1->p_tracep) != NULL)
339 			ktradref(p2);
340 	}
341 #endif
342 
343 #ifdef __HAVE_SYSCALL_INTERN
344 	(*p2->p_emul->e_syscall_intern)(p2);
345 #endif
346 
347 	scheduler_fork_hook(p1, p2);
348 
349 	/*
350 	 * Create signal actions for the child process.
351 	 */
352 	sigactsinit(p2, p1, flags & FORK_SHARESIGS);
353 
354 	/*
355 	 * If emulation has process fork hook, call it now.
356 	 */
357 	if (p2->p_emul->e_proc_fork)
358 		(*p2->p_emul->e_proc_fork)(p2, p1);
359 
360 	/*
361 	 * This begins the section where we must prevent the parent
362 	 * from being swapped.
363 	 */
364 	PHOLD(p1);
365 
366 	/*
367 	 * Finish creating the child process.  It will return through a
368 	 * different path later.
369 	 */
370 	p2->p_addr = (struct user *)uaddr;
371 	uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE,
372 	    stack, stacksize,
373 	    (func != NULL) ? func : child_return,
374 	    (arg != NULL) ? arg : p2);
375 
376 	/*
377 	 * BEGIN PID ALLOCATION.
378 	 */
379 	s = proclist_lock_write();
380 
381 	/*
382 	 * Find an unused process ID.  We remember a range of unused IDs
383 	 * ready to use (from nextpid+1 through pidchecked-1).
384 	 */
385 	nextpid++;
386  retry:
387 	/*
388 	 * If the process ID prototype has wrapped around,
389 	 * restart somewhat above 0, as the low-numbered procs
390 	 * tend to include daemons that don't exit.
391 	 */
392 	if (nextpid >= PID_MAX) {
393 		nextpid = 500;
394 		pidchecked = 0;
395 	}
396 	if (nextpid >= pidchecked) {
397 		const struct proclist_desc *pd;
398 
399 		pidchecked = PID_MAX;
400 		/*
401 		 * Scan the process lists to check whether this pid
402 		 * is in use.  Remember the lowest pid that's greater
403 		 * than nextpid, so we can avoid checking for a while.
404 		 */
405 		pd = proclists;
406  again:
407 		LIST_FOREACH(tp, pd->pd_list, p_list) {
408 			while (tp->p_pid == nextpid ||
409 			    tp->p_pgrp->pg_id == nextpid ||
410 			    tp->p_session->s_sid == nextpid) {
411 				nextpid++;
412 				if (nextpid >= pidchecked)
413 					goto retry;
414 			}
415 			if (tp->p_pid > nextpid && pidchecked > tp->p_pid)
416 				pidchecked = tp->p_pid;
417 
418 			if (tp->p_pgrp->pg_id > nextpid &&
419 			    pidchecked > tp->p_pgrp->pg_id)
420 				pidchecked = tp->p_pgrp->pg_id;
421 
422 			if (tp->p_session->s_sid > nextpid &&
423 			    pidchecked > tp->p_session->s_sid)
424 				pidchecked = tp->p_session->s_sid;
425 		}
426 
427 		/*
428 		 * If there's another list, scan it.  If we have checked
429 		 * them all, we've found one!
430 		 */
431 		pd++;
432 		if (pd->pd_list != NULL)
433 			goto again;
434 	}
435 
436 	/* Record the pid we've allocated. */
437 	p2->p_pid = nextpid;
438 
439 	/* Record the signal to be delivered to the parent on exit. */
440 	p2->p_exitsig = exitsig;
441 
442 	/*
443 	 * Put the proc on allproc before unlocking PID allocation
444 	 * so that waiters won't grab it as soon as we unlock.
445 	 */
446 
447 	p2->p_stat = SIDL;			/* protect against others */
448 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
449 
450 	LIST_INSERT_HEAD(&allproc, p2, p_list);
451 
452 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
453 
454 	/*
455 	 * END PID ALLOCATION.
456 	 */
457 	proclist_unlock_write(s);
458 
459 	/*
460 	 * Make child runnable, set start time, and add to run queue.
461 	 */
462 	SCHED_LOCK(s);
463 	p2->p_stats->p_start = time;
464 	p2->p_acflag = AFORK;
465 	p2->p_stat = SRUN;
466 	setrunqueue(p2);
467 	SCHED_UNLOCK(s);
468 
469 	/*
470 	 * Now can be swapped.
471 	 */
472 	PRELE(p1);
473 
474 	/*
475 	 * Update stats now that we know the fork was successful.
476 	 */
477 	uvmexp.forks++;
478 	if (flags & FORK_PPWAIT)
479 		uvmexp.forks_ppwait++;
480 	if (flags & FORK_SHAREVM)
481 		uvmexp.forks_sharevm++;
482 
483 	/*
484 	 * Pass a pointer to the new process to the caller.
485 	 */
486 	if (rnewprocp != NULL)
487 		*rnewprocp = p2;
488 
489 #ifdef KTRACE
490 	if (KTRPOINT(p2, KTR_EMUL))
491 		ktremul(p2);
492 #endif
493 
494 	/*
495 	 * Preserve synchronization semantics of vfork.  If waiting for
496 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
497 	 * proc (in case of exit).
498 	 */
499 	if (flags & FORK_PPWAIT)
500 		while (p2->p_flag & P_PPWAIT)
501 			tsleep(p1, PWAIT, "ppwait", 0);
502 
503 	/*
504 	 * Return child pid to parent process,
505 	 * marking us as parent via retval[1].
506 	 */
507 	if (retval != NULL) {
508 		retval[0] = p2->p_pid;
509 		retval[1] = 0;
510 	}
511 
512 	return (0);
513 }
514 
515 #if defined(MULTIPROCESSOR)
516 /*
517  * XXX This is a slight hack to get newly-formed processes to
518  * XXX acquire the kernel lock as soon as they run.
519  */
520 void
521 proc_trampoline_mp(void)
522 {
523 	struct proc *p;
524 
525 	p = curproc;
526 
527 	SCHED_ASSERT_UNLOCKED();
528 	KERNEL_PROC_LOCK(p);
529 }
530 #endif
531