xref: /netbsd-src/sys/kern/kern_fork.c (revision e4d7c2e329d54c97e0c0bd3016bbe74f550c3d5e)
1 /*	$NetBSD: kern_fork.c,v 1.61 1999/07/22 21:08:31 thorpej Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)kern_fork.c	8.8 (Berkeley) 2/14/95
41  */
42 
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/map.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
51 #include <sys/pool.h>
52 #include <sys/mount.h>
53 #include <sys/proc.h>
54 #include <sys/resourcevar.h>
55 #include <sys/vnode.h>
56 #include <sys/file.h>
57 #include <sys/acct.h>
58 #include <sys/ktrace.h>
59 #include <sys/vmmeter.h>
60 #include <sys/sched.h>
61 #include <sys/signalvar.h>
62 
63 #include <sys/syscallargs.h>
64 
65 #include <vm/vm.h>
66 #include <vm/vm_kern.h>
67 
68 #include <uvm/uvm_extern.h>
69 
70 int	nprocs = 1;		/* process 0 */
71 
72 /*ARGSUSED*/
73 int
74 sys_fork(p, v, retval)
75 	struct proc *p;
76 	void *v;
77 	register_t *retval;
78 {
79 
80 	return (fork1(p, 0, SIGCHLD, NULL, 0, retval, NULL));
81 }
82 
83 /*
84  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
85  * Address space is not shared, but parent is blocked until child exit.
86  */
87 /*ARGSUSED*/
88 int
89 sys_vfork(p, v, retval)
90 	struct proc *p;
91 	void *v;
92 	register_t *retval;
93 {
94 
95 	return (fork1(p, FORK_PPWAIT, SIGCHLD, NULL, 0, retval, NULL));
96 }
97 
98 /*
99  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
100  * semantics.  Address space is shared, and parent is blocked until child exit.
101  */
102 /*ARGSUSED*/
103 int
104 sys___vfork14(p, v, retval)
105 	struct proc *p;
106 	void *v;
107 	register_t *retval;
108 {
109 
110 	return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
111 	    retval, NULL));
112 }
113 
114 int
115 fork1(p1, flags, exitsig, stack, stacksize, retval, rnewprocp)
116 	register struct proc *p1;
117 	int flags;
118 	int exitsig;
119 	void *stack;
120 	size_t stacksize;
121 	register_t *retval;
122 	struct proc **rnewprocp;
123 {
124 	register struct proc *p2;
125 	register uid_t uid;
126 	struct proc *newproc;
127 	int count, s;
128 	vaddr_t uaddr;
129 	static int nextpid, pidchecked = 0;
130 
131 	/*
132 	 * Although process entries are dynamically created, we still keep
133 	 * a global limit on the maximum number we will create.  Don't allow
134 	 * a nonprivileged user to use the last process; don't let root
135 	 * exceed the limit. The variable nprocs is the current number of
136 	 * processes, maxproc is the limit.
137 	 */
138 	uid = p1->p_cred->p_ruid;
139 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
140 		tablefull("proc");
141 		return (EAGAIN);
142 	}
143 
144 	/*
145 	 * Increment the count of procs running with this uid. Don't allow
146 	 * a nonprivileged user to exceed their current limit.
147 	 */
148 	count = chgproccnt(uid, 1);
149 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
150 		(void)chgproccnt(uid, -1);
151 		return (EAGAIN);
152 	}
153 
154 	/*
155 	 * Allocate virtual address space for the U-area now, while it
156 	 * is still easy to abort the fork operation if we're out of
157 	 * kernel virtual address space.  The actual U-area pages will
158 	 * be allocated and wired in vm_fork().
159 	 */
160 	uaddr = uvm_km_valloc(kernel_map, USPACE);
161 	if (uaddr == 0) {
162 		(void)chgproccnt(uid, -1);
163 		return (ENOMEM);
164 	}
165 
166 	/*
167 	 * We are now committed to the fork.  From here on, we may
168 	 * block on resources, but resource allocation may NOT fail.
169 	 */
170 
171 	/* Allocate new proc. */
172 	newproc = pool_get(&proc_pool, PR_WAITOK);
173 
174 	/*
175 	 * BEGIN PID ALLOCATION.
176 	 */
177 	s = proclist_lock_write();
178 
179 	/*
180 	 * Find an unused process ID.  We remember a range of unused IDs
181 	 * ready to use (from nextpid+1 through pidchecked-1).
182 	 */
183 	nextpid++;
184 retry:
185 	/*
186 	 * If the process ID prototype has wrapped around,
187 	 * restart somewhat above 0, as the low-numbered procs
188 	 * tend to include daemons that don't exit.
189 	 */
190 	if (nextpid >= PID_MAX) {
191 		nextpid = 100;
192 		pidchecked = 0;
193 	}
194 	if (nextpid >= pidchecked) {
195 		const struct proclist_desc *pd;
196 
197 		pidchecked = PID_MAX;
198 		/*
199 		 * Scan the process lists to check whether this pid
200 		 * is in use.  Remember the lowest pid that's greater
201 		 * than nextpid, so we can avoid checking for a while.
202 		 */
203 		pd = proclists;
204 again:
205 		for (p2 = LIST_FIRST(pd->pd_list); p2 != 0;
206 		     p2 = LIST_NEXT(p2, p_list)) {
207 			while (p2->p_pid == nextpid ||
208 			    p2->p_pgrp->pg_id == nextpid ||
209 			    p2->p_session->s_sid == nextpid) {
210 				nextpid++;
211 				if (nextpid >= pidchecked)
212 					goto retry;
213 			}
214 			if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
215 				pidchecked = p2->p_pid;
216 
217 			if (p2->p_pgrp->pg_id > nextpid &&
218 			    pidchecked > p2->p_pgrp->pg_id)
219 				pidchecked = p2->p_pgrp->pg_id;
220 
221 			if (p2->p_session->s_sid > nextpid &&
222 			    pidchecked > p2->p_session->s_sid)
223 				pidchecked = p2->p_session->s_sid;
224 		}
225 
226 		/*
227 		 * If there's another list, scan it.  If we have checked
228 		 * them all, we've found one!
229 		 */
230 		pd++;
231 		if (pd->pd_list != NULL)
232 			goto again;
233 	}
234 
235 	nprocs++;
236 	p2 = newproc;
237 
238 	/* Record the pid we've allocated. */
239 	p2->p_pid = nextpid;
240 
241 	/* Record the signal to be delivered to the parent on exit. */
242 	p2->p_exitsig = exitsig;
243 
244 	/*
245 	 * Put the proc on allproc before unlocking PID allocation
246 	 * so that waiters won't grab it as soon as we unlock.
247 	 */
248 
249 	p2->p_stat = SIDL;			/* protect against others */
250 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
251 
252 	LIST_INSERT_HEAD(&allproc, p2, p_list);
253 
254 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
255 
256 	/*
257 	 * END PID ALLOCATION.
258 	 */
259 	proclist_unlock_write(s);
260 
261 	/*
262 	 * Make a proc table entry for the new process.
263 	 * Start by zeroing the section of proc that is zero-initialized,
264 	 * then copy the section that is copied directly from the parent.
265 	 */
266 	memset(&p2->p_startzero, 0,
267 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
268 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
269 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
270 
271 	/*
272 	 * Duplicate sub-structures as needed.
273 	 * Increase reference counts on shared objects.
274 	 * The p_stats and p_sigacts substructs are set in vm_fork.
275 	 */
276 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
277 	p2->p_emul = p1->p_emul;
278 	if (p1->p_flag & P_PROFIL)
279 		startprofclock(p2);
280 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
281 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
282 	p2->p_cred->p_refcnt = 1;
283 	crhold(p1->p_ucred);
284 
285 	/* bump references to the text vnode (for procfs) */
286 	p2->p_textvp = p1->p_textvp;
287 	if (p2->p_textvp)
288 		VREF(p2->p_textvp);
289 
290 	if (flags & FORK_SHAREFILES)
291 		fdshare(p1, p2);
292 	else
293 		p2->p_fd = fdcopy(p1);
294 
295 	if (flags & FORK_SHARECWD)
296 		cwdshare(p1, p2);
297 	else
298 		p2->p_cwdi = cwdinit(p1);
299 
300 	/*
301 	 * If p_limit is still copy-on-write, bump refcnt,
302 	 * otherwise get a copy that won't be modified.
303 	 * (If PL_SHAREMOD is clear, the structure is shared
304 	 * copy-on-write.)
305 	 */
306 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
307 		p2->p_limit = limcopy(p1->p_limit);
308 	else {
309 		p2->p_limit = p1->p_limit;
310 		p2->p_limit->p_refcnt++;
311 	}
312 
313 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
314 		p2->p_flag |= P_CONTROLT;
315 	if (flags & FORK_PPWAIT)
316 		p2->p_flag |= P_PPWAIT;
317 	LIST_INSERT_AFTER(p1, p2, p_pglist);
318 	p2->p_pptr = p1;
319 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
320 	LIST_INIT(&p2->p_children);
321 
322 #ifdef KTRACE
323 	/*
324 	 * Copy traceflag and tracefile if enabled.
325 	 * If not inherited, these were zeroed above.
326 	 */
327 	if (p1->p_traceflag&KTRFAC_INHERIT) {
328 		p2->p_traceflag = p1->p_traceflag;
329 		if ((p2->p_tracep = p1->p_tracep) != NULL)
330 			ktradref(p2);
331 	}
332 #endif
333 	scheduler_fork_hook(p1, p2);
334 
335 	/*
336 	 * Create signal actions for the child process.
337 	 */
338 	if (flags & FORK_SHARESIGS)
339 		sigactsshare(p1, p2);
340 	else
341 		p2->p_sigacts = sigactsinit(p1);
342 
343 	/*
344 	 * This begins the section where we must prevent the parent
345 	 * from being swapped.
346 	 */
347 	PHOLD(p1);
348 
349 	/*
350 	 * Finish creating the child process.  It will return through a
351 	 * different path later.
352 	 */
353 	p2->p_addr = (struct user *)uaddr;
354 	uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE,
355 	    stack, stacksize);
356 
357 	/*
358 	 * Make child runnable, set start time, and add to run queue.
359 	 */
360 	s = splstatclock();
361 	p2->p_stats->p_start = time;
362 	p2->p_acflag = AFORK;
363 	p2->p_stat = SRUN;
364 	setrunqueue(p2);
365 	splx(s);
366 
367 	/*
368 	 * Now can be swapped.
369 	 */
370 	PRELE(p1);
371 
372 	/*
373 	 * Update stats now that we know the fork was successful.
374 	 */
375 	uvmexp.forks++;
376 	if (flags & FORK_PPWAIT)
377 		uvmexp.forks_ppwait++;
378 	if (flags & FORK_SHAREVM)
379 		uvmexp.forks_sharevm++;
380 
381 	/*
382 	 * Pass a pointer to the new process to the caller.
383 	 */
384 	if (rnewprocp != NULL)
385 		*rnewprocp = p2;
386 
387 	/*
388 	 * Preserve synchronization semantics of vfork.  If waiting for
389 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
390 	 * proc (in case of exit).
391 	 */
392 	if (flags & FORK_PPWAIT)
393 		while (p2->p_flag & P_PPWAIT)
394 			tsleep(p1, PWAIT, "ppwait", 0);
395 
396 	/*
397 	 * Return child pid to parent process,
398 	 * marking us as parent via retval[1].
399 	 */
400 	if (retval != NULL) {
401 		retval[0] = p2->p_pid;
402 		retval[1] = 0;
403 	}
404 	return (0);
405 }
406