xref: /netbsd-src/sys/kern/kern_fork.c (revision 3c618b8f0665ca5acf2910b7200a6f7d6be3506e)
1 /*	$NetBSD: kern_fork.c,v 1.54 1999/03/24 05:51:23 mrg 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 
62 #include <sys/syscallargs.h>
63 
64 #include <vm/vm.h>
65 #include <vm/vm_kern.h>
66 
67 #include <uvm/uvm_extern.h>
68 
69 int	nprocs = 1;		/* process 0 */
70 
71 /*ARGSUSED*/
72 int
73 sys_fork(p, v, retval)
74 	struct proc *p;
75 	void *v;
76 	register_t *retval;
77 {
78 
79 	return (fork1(p, 0, retval, NULL));
80 }
81 
82 /*
83  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
84  * Address space is not shared, but parent is blocked until child exit.
85  */
86 /*ARGSUSED*/
87 int
88 sys_vfork(p, v, retval)
89 	struct proc *p;
90 	void *v;
91 	register_t *retval;
92 {
93 
94 	return (fork1(p, FORK_PPWAIT, retval, NULL));
95 }
96 
97 /*
98  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
99  * semantics.  Address space is shared, and parent is blocked until child exit.
100  */
101 /*ARGSUSED*/
102 int
103 sys___vfork14(p, v, retval)
104 	struct proc *p;
105 	void *v;
106 	register_t *retval;
107 {
108 
109 	return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, retval, NULL));
110 }
111 
112 int
113 fork1(p1, flags, retval, rnewprocp)
114 	register struct proc *p1;
115 	int flags;
116 	register_t *retval;
117 	struct proc **rnewprocp;
118 {
119 	register struct proc *p2;
120 	register uid_t uid;
121 	struct proc *newproc;
122 	int count, s;
123 	vaddr_t uaddr;
124 	static int nextpid, pidchecked = 0;
125 
126 	/*
127 	 * Although process entries are dynamically created, we still keep
128 	 * a global limit on the maximum number we will create.  Don't allow
129 	 * a nonprivileged user to use the last process; don't let root
130 	 * exceed the limit. The variable nprocs is the current number of
131 	 * processes, maxproc is the limit.
132 	 */
133 	uid = p1->p_cred->p_ruid;
134 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
135 		tablefull("proc");
136 		return (EAGAIN);
137 	}
138 
139 	/*
140 	 * Increment the count of procs running with this uid. Don't allow
141 	 * a nonprivileged user to exceed their current limit.
142 	 */
143 	count = chgproccnt(uid, 1);
144 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
145 		(void)chgproccnt(uid, -1);
146 		return (EAGAIN);
147 	}
148 
149 	/*
150 	 * Allocate virtual address space for the U-area now, while it
151 	 * is still easy to abort the fork operation if we're out of
152 	 * kernel virtual address space.  The actual U-area pages will
153 	 * be allocated and wired in vm_fork().
154 	 */
155 	uaddr = uvm_km_valloc(kernel_map, USPACE);
156 	if (uaddr == 0) {
157 		(void)chgproccnt(uid, -1);
158 		return (ENOMEM);
159 	}
160 
161 	/*
162 	 * We are now committed to the fork.  From here on, we may
163 	 * block on resources, but resource allocation may NOT fail.
164 	 */
165 
166 	/* Allocate new proc. */
167 	newproc = pool_get(&proc_pool, PR_WAITOK);
168 
169 	/*
170 	 * BEGIN PID ALLOCATION.  (Lock PID allocation variables eventually).
171 	 */
172 
173 	/*
174 	 * Find an unused process ID.  We remember a range of unused IDs
175 	 * ready to use (from nextpid+1 through pidchecked-1).
176 	 */
177 	nextpid++;
178 retry:
179 	/*
180 	 * If the process ID prototype has wrapped around,
181 	 * restart somewhat above 0, as the low-numbered procs
182 	 * tend to include daemons that don't exit.
183 	 */
184 	if (nextpid >= PID_MAX) {
185 		nextpid = 100;
186 		pidchecked = 0;
187 	}
188 	if (nextpid >= pidchecked) {
189 		const struct proclist_desc *pd;
190 
191 		pidchecked = PID_MAX;
192 		/*
193 		 * Scan the process lists to check whether this pid
194 		 * is in use.  Remember the lowest pid that's greater
195 		 * than nextpid, so we can avoid checking for a while.
196 		 */
197 		pd = proclists;
198 again:
199 		for (p2 = LIST_FIRST(pd->pd_list); p2 != 0;
200 		     p2 = LIST_NEXT(p2, p_list)) {
201 			while (p2->p_pid == nextpid ||
202 			    p2->p_pgrp->pg_id == nextpid ||
203 			    p2->p_session->s_sid == nextpid) {
204 				nextpid++;
205 				if (nextpid >= pidchecked)
206 					goto retry;
207 			}
208 			if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
209 				pidchecked = p2->p_pid;
210 
211 			if (p2->p_pgrp->pg_id > nextpid &&
212 			    pidchecked > p2->p_pgrp->pg_id)
213 				pidchecked = p2->p_pgrp->pg_id;
214 
215 			if (p2->p_session->s_sid > nextpid &&
216 			    pidchecked > p2->p_session->s_sid)
217 				pidchecked = p2->p_session->s_sid;
218 		}
219 
220 		/*
221 		 * If there's another list, scan it.  If we have checked
222 		 * them all, we've found one!
223 		 */
224 		pd++;
225 		if (pd->pd_list != NULL)
226 			goto again;
227 	}
228 
229 	nprocs++;
230 	p2 = newproc;
231 
232 	/* Record the pid we've allocated. */
233 	p2->p_pid = nextpid;
234 
235 	/*
236 	 * Put the proc on allproc before unlocking PID allocation
237 	 * so that waiters won't grab it as soon as we unlock.
238 	 */
239 	LIST_INSERT_HEAD(&allproc, p2, p_list);
240 
241 	/*
242 	 * END PID ALLOCATION.  (Unlock PID allocation variables).
243 	 */
244 
245 	p2->p_stat = SIDL;			/* protect against others */
246 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
247 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
248 
249 	/*
250 	 * Make a proc table entry for the new process.
251 	 * Start by zeroing the section of proc that is zero-initialized,
252 	 * then copy the section that is copied directly from the parent.
253 	 */
254 	memset(&p2->p_startzero, 0,
255 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
256 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
257 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
258 
259 	/*
260 	 * Duplicate sub-structures as needed.
261 	 * Increase reference counts on shared objects.
262 	 * The p_stats and p_sigacts substructs are set in vm_fork.
263 	 */
264 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
265 	p2->p_emul = p1->p_emul;
266 	if (p1->p_flag & P_PROFIL)
267 		startprofclock(p2);
268 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
269 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
270 	p2->p_cred->p_refcnt = 1;
271 	crhold(p1->p_ucred);
272 
273 	/* bump references to the text vnode (for procfs) */
274 	p2->p_textvp = p1->p_textvp;
275 	if (p2->p_textvp)
276 		VREF(p2->p_textvp);
277 
278 	p2->p_fd = fdcopy(p1);
279 	/*
280 	 * If p_limit is still copy-on-write, bump refcnt,
281 	 * otherwise get a copy that won't be modified.
282 	 * (If PL_SHAREMOD is clear, the structure is shared
283 	 * copy-on-write.)
284 	 */
285 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
286 		p2->p_limit = limcopy(p1->p_limit);
287 	else {
288 		p2->p_limit = p1->p_limit;
289 		p2->p_limit->p_refcnt++;
290 	}
291 
292 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
293 		p2->p_flag |= P_CONTROLT;
294 	if (flags & FORK_PPWAIT)
295 		p2->p_flag |= P_PPWAIT;
296 	LIST_INSERT_AFTER(p1, p2, p_pglist);
297 	p2->p_pptr = p1;
298 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
299 	LIST_INIT(&p2->p_children);
300 
301 #ifdef KTRACE
302 	/*
303 	 * Copy traceflag and tracefile if enabled.
304 	 * If not inherited, these were zeroed above.
305 	 */
306 	if (p1->p_traceflag&KTRFAC_INHERIT) {
307 		p2->p_traceflag = p1->p_traceflag;
308 		if ((p2->p_tracep = p1->p_tracep) != NULL)
309 			ktradref(p2);
310 	}
311 #endif
312 	scheduler_fork_hook(p1, p2);
313 
314 	/*
315 	 * This begins the section where we must prevent the parent
316 	 * from being swapped.
317 	 */
318 	PHOLD(p1);
319 
320 	/*
321 	 * Finish creating the child process.  It will return through a
322 	 * different path later.
323 	 */
324 	p2->p_addr = (struct user *)uaddr;
325 	uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
326 
327 	/*
328 	 * Make child runnable, set start time, and add to run queue.
329 	 */
330 	s = splstatclock();
331 	p2->p_stats->p_start = time;
332 	p2->p_acflag = AFORK;
333 	p2->p_stat = SRUN;
334 	setrunqueue(p2);
335 	splx(s);
336 
337 	/*
338 	 * Now can be swapped.
339 	 */
340 	PRELE(p1);
341 
342 	/*
343 	 * Update stats now that we know the fork was successful.
344 	 */
345 	uvmexp.forks++;
346 	if (flags & FORK_PPWAIT)
347 		uvmexp.forks_ppwait++;
348 	if (flags & FORK_SHAREVM)
349 		uvmexp.forks_sharevm++;
350 
351 	/*
352 	 * Pass a pointer to the new process to the caller.
353 	 */
354 	if (rnewprocp != NULL)
355 		*rnewprocp = p2;
356 
357 	/*
358 	 * Preserve synchronization semantics of vfork.  If waiting for
359 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
360 	 * proc (in case of exit).
361 	 */
362 	if (flags & FORK_PPWAIT)
363 		while (p2->p_flag & P_PPWAIT)
364 			tsleep(p1, PWAIT, "ppwait", 0);
365 
366 	/*
367 	 * Return child pid to parent process,
368 	 * marking us as parent via retval[1].
369 	 */
370 	if (retval != NULL) {
371 		retval[0] = p2->p_pid;
372 		retval[1] = 0;
373 	}
374 	return (0);
375 }
376