xref: /netbsd-src/sys/kern/kern_fork.c (revision 93f9db1b75d415b78f73ed629beeb86235153473)
1 /*	$NetBSD: kern_fork.c,v 1.50 1998/11/11 22:44:25 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 #include "opt_uvm.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/map.h>
49 #include <sys/filedesc.h>
50 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/pool.h>
53 #include <sys/mount.h>
54 #include <sys/proc.h>
55 #include <sys/resourcevar.h>
56 #include <sys/vnode.h>
57 #include <sys/file.h>
58 #include <sys/acct.h>
59 #include <sys/ktrace.h>
60 #include <sys/vmmeter.h>
61 
62 #include <sys/syscallargs.h>
63 
64 #include <vm/vm.h>
65 #include <vm/vm_kern.h>
66 
67 #if defined(UVM)
68 #include <uvm/uvm_extern.h>
69 #endif
70 
71 int	nprocs = 1;		/* process 0 */
72 
73 /*ARGSUSED*/
74 int
75 sys_fork(p, v, retval)
76 	struct proc *p;
77 	void *v;
78 	register_t *retval;
79 {
80 
81 	return (fork1(p, 0, retval, NULL));
82 }
83 
84 /*
85  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
86  * Address space is not shared, but parent is blocked until child exit.
87  */
88 /*ARGSUSED*/
89 int
90 sys_vfork(p, v, retval)
91 	struct proc *p;
92 	void *v;
93 	register_t *retval;
94 {
95 
96 	return (fork1(p, FORK_PPWAIT, retval, NULL));
97 }
98 
99 /*
100  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
101  * semantics.  Address space is shared, and parent is blocked until child exit.
102  */
103 /*ARGSUSED*/
104 int
105 sys___vfork14(p, v, retval)
106 	struct proc *p;
107 	void *v;
108 	register_t *retval;
109 {
110 
111 	return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, retval, NULL));
112 }
113 
114 int
115 fork1(p1, flags, retval, rnewprocp)
116 	register struct proc *p1;
117 	int flags;
118 	register_t *retval;
119 	struct proc **rnewprocp;
120 {
121 	register struct proc *p2;
122 	register uid_t uid;
123 	struct proc *newproc;
124 	int count, s;
125 	vaddr_t uaddr;
126 	static int nextpid, pidchecked = 0;
127 
128 	/*
129 	 * Although process entries are dynamically created, we still keep
130 	 * a global limit on the maximum number we will create.  Don't allow
131 	 * a nonprivileged user to use the last process; don't let root
132 	 * exceed the limit. The variable nprocs is the current number of
133 	 * processes, maxproc is the limit.
134 	 */
135 	uid = p1->p_cred->p_ruid;
136 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
137 		tablefull("proc");
138 		return (EAGAIN);
139 	}
140 
141 	/*
142 	 * Increment the count of procs running with this uid. Don't allow
143 	 * a nonprivileged user to exceed their current limit.
144 	 */
145 	count = chgproccnt(uid, 1);
146 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
147 		(void)chgproccnt(uid, -1);
148 		return (EAGAIN);
149 	}
150 
151 	/*
152 	 * Allocate virtual address space for the U-area now, while it
153 	 * is still easy to abort the fork operation if we're out of
154 	 * kernel virtual address space.  The actual U-area pages will
155 	 * be allocated and wired in vm_fork().
156 	 */
157 #if defined(UVM)
158 	uaddr = uvm_km_valloc(kernel_map, USPACE);
159 #else
160 	uaddr = kmem_alloc_pageable(kernel_map, USPACE);
161 #endif
162 	if (uaddr == 0) {
163 		(void)chgproccnt(uid, -1);
164 		return (ENOMEM);
165 	}
166 
167 	/*
168 	 * We are now committed to the fork.  From here on, we may
169 	 * block on resources, but resource allocation may NOT fail.
170 	 */
171 
172 	/* Allocate new proc. */
173 	newproc = pool_get(&proc_pool, PR_WAITOK);
174 
175 	/*
176 	 * BEGIN PID ALLOCATION.  (Lock PID allocation variables eventually).
177 	 */
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 	/*
242 	 * Put the proc on allproc before unlocking PID allocation
243 	 * so that waiters won't grab it as soon as we unlock.
244 	 */
245 	LIST_INSERT_HEAD(&allproc, p2, p_list);
246 
247 	/*
248 	 * END PID ALLOCATION.  (Unlock PID allocation variables).
249 	 */
250 
251 	p2->p_stat = SIDL;			/* protect against others */
252 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
253 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
254 
255 	/*
256 	 * Make a proc table entry for the new process.
257 	 * Start by zeroing the section of proc that is zero-initialized,
258 	 * then copy the section that is copied directly from the parent.
259 	 */
260 	memset(&p2->p_startzero, 0,
261 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
262 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
263 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
264 
265 	/*
266 	 * Duplicate sub-structures as needed.
267 	 * Increase reference counts on shared objects.
268 	 * The p_stats and p_sigacts substructs are set in vm_fork.
269 	 */
270 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
271 	p2->p_emul = p1->p_emul;
272 	if (p1->p_flag & P_PROFIL)
273 		startprofclock(p2);
274 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
275 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
276 	p2->p_cred->p_refcnt = 1;
277 	crhold(p1->p_ucred);
278 
279 	/* bump references to the text vnode (for procfs) */
280 	p2->p_textvp = p1->p_textvp;
281 	if (p2->p_textvp)
282 		VREF(p2->p_textvp);
283 
284 	p2->p_fd = fdcopy(p1);
285 	/*
286 	 * If p_limit is still copy-on-write, bump refcnt,
287 	 * otherwise get a copy that won't be modified.
288 	 * (If PL_SHAREMOD is clear, the structure is shared
289 	 * copy-on-write.)
290 	 */
291 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
292 		p2->p_limit = limcopy(p1->p_limit);
293 	else {
294 		p2->p_limit = p1->p_limit;
295 		p2->p_limit->p_refcnt++;
296 	}
297 
298 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
299 		p2->p_flag |= P_CONTROLT;
300 	if (flags & FORK_PPWAIT)
301 		p2->p_flag |= P_PPWAIT;
302 	LIST_INSERT_AFTER(p1, p2, p_pglist);
303 	p2->p_pptr = p1;
304 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
305 	LIST_INIT(&p2->p_children);
306 
307 #ifdef KTRACE
308 	/*
309 	 * Copy traceflag and tracefile if enabled.
310 	 * If not inherited, these were zeroed above.
311 	 */
312 	if (p1->p_traceflag&KTRFAC_INHERIT) {
313 		p2->p_traceflag = p1->p_traceflag;
314 		if ((p2->p_tracep = p1->p_tracep) != NULL)
315 			ktradref(p2);
316 	}
317 #endif
318 
319 	/*
320 	 * This begins the section where we must prevent the parent
321 	 * from being swapped.
322 	 */
323 	PHOLD(p1);
324 
325 	/*
326 	 * Finish creating the child process.  It will return through a
327 	 * different path later.
328 	 */
329 	p2->p_addr = (struct user *)uaddr;
330 #if defined(UVM)
331 	uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
332 #else
333 	vm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
334 #endif
335 
336 	/*
337 	 * Make child runnable, set start time, and add to run queue.
338 	 */
339 	s = splstatclock();
340 	p2->p_stats->p_start = time;
341 	p2->p_acflag = AFORK;
342 	p2->p_stat = SRUN;
343 	setrunqueue(p2);
344 	splx(s);
345 
346 	/*
347 	 * Now can be swapped.
348 	 */
349 	PRELE(p1);
350 
351 	/*
352 	 * Update stats now that we know the fork was successful.
353 	 */
354 #if defined(UVM)
355 	uvmexp.forks++;
356 	if (flags & FORK_PPWAIT)
357 		uvmexp.forks_ppwait++;
358 	if (flags & FORK_SHAREVM)
359 		uvmexp.forks_sharevm++;
360 #else
361 	cnt.v_forks++;
362 	if (flags & FORK_PPWAIT)
363 		cnt.v_forks_ppwait++;
364 	if (flags & FORK_SHAREVM)
365 		cnt.v_forks_sharevm++;
366 #endif
367 
368 	/*
369 	 * Pass a pointer to the new process to the caller.
370 	 */
371 	if (rnewprocp != NULL)
372 		*rnewprocp = p2;
373 
374 	/*
375 	 * Preserve synchronization semantics of vfork.  If waiting for
376 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
377 	 * proc (in case of exit).
378 	 */
379 	if (flags & FORK_PPWAIT)
380 		while (p2->p_flag & P_PPWAIT)
381 			tsleep(p1, PWAIT, "ppwait", 0);
382 
383 	/*
384 	 * Return child pid to parent process,
385 	 * marking us as parent via retval[1].
386 	 */
387 	if (retval != NULL) {
388 		retval[0] = p2->p_pid;
389 		retval[1] = 0;
390 	}
391 	return (0);
392 }
393